[PATCH] ext3: ext3_show_options fix
[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 #include <asm/uaccess.h>
40 #include "xattr.h"
41 #include "acl.h"
42
43 static int ext3_load_journal(struct super_block *, struct ext3_super_block *);
44 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
45                                int);
46 static void ext3_commit_super (struct super_block * sb,
47                                struct ext3_super_block * es,
48                                int sync);
49 static void ext3_mark_recovery_complete(struct super_block * sb,
50                                         struct ext3_super_block * es);
51 static void ext3_clear_journal_err(struct super_block * sb,
52                                    struct ext3_super_block * es);
53 static int ext3_sync_fs(struct super_block *sb, int wait);
54 static const char *ext3_decode_error(struct super_block * sb, int errno,
55                                      char nbuf[16]);
56 static int ext3_remount (struct super_block * sb, int * flags, char * data);
57 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf);
58 static void ext3_unlockfs(struct super_block *sb);
59 static void ext3_write_super (struct super_block * sb);
60 static void ext3_write_super_lockfs(struct super_block *sb);
61
62 /* 
63  * Wrappers for journal_start/end.
64  *
65  * The only special thing we need to do here is to make sure that all
66  * journal_end calls result in the superblock being marked dirty, so
67  * that sync() will call the filesystem's write_super callback if
68  * appropriate. 
69  */
70 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
71 {
72         journal_t *journal;
73
74         if (sb->s_flags & MS_RDONLY)
75                 return ERR_PTR(-EROFS);
76
77         /* Special case here: if the journal has aborted behind our
78          * backs (eg. EIO in the commit thread), then we still need to
79          * take the FS itself readonly cleanly. */
80         journal = EXT3_SB(sb)->s_journal;
81         if (is_journal_aborted(journal)) {
82                 ext3_abort(sb, __FUNCTION__,
83                            "Detected aborted journal");
84                 return ERR_PTR(-EROFS);
85         }
86
87         return journal_start(journal, nblocks);
88 }
89
90 /* 
91  * The only special thing we need to do here is to make sure that all
92  * journal_stop calls result in the superblock being marked dirty, so
93  * that sync() will call the filesystem's write_super callback if
94  * appropriate. 
95  */
96 int __ext3_journal_stop(const char *where, handle_t *handle)
97 {
98         struct super_block *sb;
99         int err;
100         int rc;
101
102         sb = handle->h_transaction->t_journal->j_private;
103         err = handle->h_err;
104         rc = journal_stop(handle);
105
106         if (!err)
107                 err = rc;
108         if (err)
109                 __ext3_std_error(sb, where, err);
110         return err;
111 }
112
113 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
114                 struct buffer_head *bh, handle_t *handle, int err)
115 {
116         char nbuf[16];
117         const char *errstr = ext3_decode_error(NULL, err, nbuf);
118
119         if (bh)
120                 BUFFER_TRACE(bh, "abort");
121
122         if (!handle->h_err)
123                 handle->h_err = err;
124
125         if (is_handle_aborted(handle))
126                 return;
127
128         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
129                caller, errstr, err_fn);
130
131         journal_abort_handle(handle);
132 }
133
134 /* Deal with the reporting of failure conditions on a filesystem such as
135  * inconsistencies detected or read IO failures.
136  *
137  * On ext2, we can store the error state of the filesystem in the
138  * superblock.  That is not possible on ext3, because we may have other
139  * write ordering constraints on the superblock which prevent us from
140  * writing it out straight away; and given that the journal is about to
141  * be aborted, we can't rely on the current, or future, transactions to
142  * write out the superblock safely.
143  *
144  * We'll just use the journal_abort() error code to record an error in
145  * the journal instead.  On recovery, the journal will compain about
146  * that error until we've noted it down and cleared it.
147  */
148
149 static void ext3_handle_error(struct super_block *sb)
150 {
151         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
152
153         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
154         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
155
156         if (sb->s_flags & MS_RDONLY)
157                 return;
158
159         if (test_opt (sb, ERRORS_RO)) {
160                 printk (KERN_CRIT "Remounting filesystem read-only\n");
161                 sb->s_flags |= MS_RDONLY;
162         } else {
163                 journal_t *journal = EXT3_SB(sb)->s_journal;
164
165                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
166                 if (journal)
167                         journal_abort(journal, -EIO);
168         }
169         if (test_opt(sb, ERRORS_PANIC))
170                 panic("EXT3-fs (device %s): panic forced after error\n",
171                         sb->s_id);
172         ext3_commit_super(sb, es, 1);
173 }
174
175 void ext3_error (struct super_block * sb, const char * function,
176                  const char * fmt, ...)
177 {
178         va_list args;
179
180         va_start(args, fmt);
181         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
182         vprintk(fmt, args);
183         printk("\n");
184         va_end(args);
185
186         ext3_handle_error(sb);
187 }
188
189 static const char *ext3_decode_error(struct super_block * sb, int errno,
190                                      char nbuf[16])
191 {
192         char *errstr = NULL;
193
194         switch (errno) {
195         case -EIO:
196                 errstr = "IO failure";
197                 break;
198         case -ENOMEM:
199                 errstr = "Out of memory";
200                 break;
201         case -EROFS:
202                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
203                         errstr = "Journal has aborted";
204                 else
205                         errstr = "Readonly filesystem";
206                 break;
207         default:
208                 /* If the caller passed in an extra buffer for unknown
209                  * errors, textualise them now.  Else we just return
210                  * NULL. */
211                 if (nbuf) {
212                         /* Check for truncated error codes... */
213                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
214                                 errstr = nbuf;
215                 }
216                 break;
217         }
218
219         return errstr;
220 }
221
222 /* __ext3_std_error decodes expected errors from journaling functions
223  * automatically and invokes the appropriate error response.  */
224
225 void __ext3_std_error (struct super_block * sb, const char * function,
226                        int errno)
227 {
228         char nbuf[16];
229         const char *errstr;
230
231         /* Special case: if the error is EROFS, and we're not already
232          * inside a transaction, then there's really no point in logging
233          * an error. */
234         if (errno == -EROFS && journal_current_handle() == NULL &&
235             (sb->s_flags & MS_RDONLY))
236                 return;
237
238         errstr = ext3_decode_error(sb, errno, nbuf);
239         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
240                 sb->s_id, function, errstr);
241
242         ext3_handle_error(sb);
243 }
244
245 /*
246  * ext3_abort is a much stronger failure handler than ext3_error.  The
247  * abort function may be used to deal with unrecoverable failures such
248  * as journal IO errors or ENOMEM at a critical moment in log management.
249  *
250  * We unconditionally force the filesystem into an ABORT|READONLY state,
251  * unless the error response on the fs has been set to panic in which
252  * case we take the easy way out and panic immediately.
253  */
254
255 void ext3_abort (struct super_block * sb, const char * function,
256                  const char * fmt, ...)
257 {
258         va_list args;
259
260         printk (KERN_CRIT "ext3_abort called.\n");
261
262         va_start(args, fmt);
263         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
264         vprintk(fmt, args);
265         printk("\n");
266         va_end(args);
267
268         if (test_opt(sb, ERRORS_PANIC))
269                 panic("EXT3-fs panic from previous error\n");
270
271         if (sb->s_flags & MS_RDONLY)
272                 return;
273
274         printk(KERN_CRIT "Remounting filesystem read-only\n");
275         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
276         sb->s_flags |= MS_RDONLY;
277         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
278         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
279 }
280
281 void ext3_warning (struct super_block * sb, const char * function,
282                    const char * fmt, ...)
283 {
284         va_list args;
285
286         va_start(args, fmt);
287         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
288                sb->s_id, function);
289         vprintk(fmt, args);
290         printk("\n");
291         va_end(args);
292 }
293
294 void ext3_update_dynamic_rev(struct super_block *sb)
295 {
296         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
297
298         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
299                 return;
300
301         ext3_warning(sb, __FUNCTION__,
302                      "updating to rev %d because of new feature flag, "
303                      "running e2fsck is recommended",
304                      EXT3_DYNAMIC_REV);
305
306         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
307         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
308         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
309         /* leave es->s_feature_*compat flags alone */
310         /* es->s_uuid will be set by e2fsck if empty */
311
312         /*
313          * The rest of the superblock fields should be zero, and if not it
314          * means they are likely already in use, so leave them alone.  We
315          * can leave it up to e2fsck to clean up any inconsistencies there.
316          */
317 }
318
319 /*
320  * Open the external journal device
321  */
322 static struct block_device *ext3_blkdev_get(dev_t dev)
323 {
324         struct block_device *bdev;
325         char b[BDEVNAME_SIZE];
326
327         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
328         if (IS_ERR(bdev))
329                 goto fail;
330         return bdev;
331
332 fail:
333         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
334                         __bdevname(dev, b), PTR_ERR(bdev));
335         return NULL;
336 }
337
338 /*
339  * Release the journal device
340  */
341 static int ext3_blkdev_put(struct block_device *bdev)
342 {
343         bd_release(bdev);
344         return blkdev_put(bdev);
345 }
346
347 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
348 {
349         struct block_device *bdev;
350         int ret = -ENODEV;
351
352         bdev = sbi->journal_bdev;
353         if (bdev) {
354                 ret = ext3_blkdev_put(bdev);
355                 sbi->journal_bdev = NULL;
356         }
357         return ret;
358 }
359
360 static inline struct inode *orphan_list_entry(struct list_head *l)
361 {
362         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
363 }
364
365 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
366 {
367         struct list_head *l;
368
369         printk(KERN_ERR "sb orphan head is %d\n", 
370                le32_to_cpu(sbi->s_es->s_last_orphan));
371
372         printk(KERN_ERR "sb_info orphan list:\n");
373         list_for_each(l, &sbi->s_orphan) {
374                 struct inode *inode = orphan_list_entry(l);
375                 printk(KERN_ERR "  "
376                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
377                        inode->i_sb->s_id, inode->i_ino, inode,
378                        inode->i_mode, inode->i_nlink, 
379                        NEXT_ORPHAN(inode));
380         }
381 }
382
383 static void ext3_put_super (struct super_block * sb)
384 {
385         struct ext3_sb_info *sbi = EXT3_SB(sb);
386         struct ext3_super_block *es = sbi->s_es;
387         int i;
388
389         ext3_xattr_put_super(sb);
390         journal_destroy(sbi->s_journal);
391         if (!(sb->s_flags & MS_RDONLY)) {
392                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
393                 es->s_state = cpu_to_le16(sbi->s_mount_state);
394                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
395                 mark_buffer_dirty(sbi->s_sbh);
396                 ext3_commit_super(sb, es, 1);
397         }
398
399         for (i = 0; i < sbi->s_gdb_count; i++)
400                 brelse(sbi->s_group_desc[i]);
401         kfree(sbi->s_group_desc);
402         percpu_counter_destroy(&sbi->s_freeblocks_counter);
403         percpu_counter_destroy(&sbi->s_freeinodes_counter);
404         percpu_counter_destroy(&sbi->s_dirs_counter);
405         brelse(sbi->s_sbh);
406 #ifdef CONFIG_QUOTA
407         for (i = 0; i < MAXQUOTAS; i++)
408                 kfree(sbi->s_qf_names[i]);
409 #endif
410
411         /* Debugging code just in case the in-memory inode orphan list
412          * isn't empty.  The on-disk one can be non-empty if we've
413          * detected an error and taken the fs readonly, but the
414          * in-memory list had better be clean by this point. */
415         if (!list_empty(&sbi->s_orphan))
416                 dump_orphan_list(sb, sbi);
417         J_ASSERT(list_empty(&sbi->s_orphan));
418
419         invalidate_bdev(sb->s_bdev, 0);
420         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
421                 /*
422                  * Invalidate the journal device's buffers.  We don't want them
423                  * floating about in memory - the physical journal device may
424                  * hotswapped, and it breaks the `ro-after' testing code.
425                  */
426                 sync_blockdev(sbi->journal_bdev);
427                 invalidate_bdev(sbi->journal_bdev, 0);
428                 ext3_blkdev_remove(sbi);
429         }
430         sb->s_fs_info = NULL;
431         kfree(sbi);
432         return;
433 }
434
435 static kmem_cache_t *ext3_inode_cachep;
436
437 /*
438  * Called inside transaction, so use GFP_NOFS
439  */
440 static struct inode *ext3_alloc_inode(struct super_block *sb)
441 {
442         struct ext3_inode_info *ei;
443
444         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
445         if (!ei)
446                 return NULL;
447 #ifdef CONFIG_EXT3_FS_POSIX_ACL
448         ei->i_acl = EXT3_ACL_NOT_CACHED;
449         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
450 #endif
451         ei->i_block_alloc_info = NULL;
452         ei->vfs_inode.i_version = 1;
453         return &ei->vfs_inode;
454 }
455
456 static void ext3_destroy_inode(struct inode *inode)
457 {
458         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
459 }
460
461 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
462 {
463         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
464
465         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
466             SLAB_CTOR_CONSTRUCTOR) {
467                 INIT_LIST_HEAD(&ei->i_orphan);
468 #ifdef CONFIG_EXT3_FS_XATTR
469                 init_rwsem(&ei->xattr_sem);
470 #endif
471                 init_MUTEX(&ei->truncate_sem);
472                 inode_init_once(&ei->vfs_inode);
473         }
474 }
475  
476 static int init_inodecache(void)
477 {
478         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
479                                              sizeof(struct ext3_inode_info),
480                                              0, SLAB_RECLAIM_ACCOUNT,
481                                              init_once, NULL);
482         if (ext3_inode_cachep == NULL)
483                 return -ENOMEM;
484         return 0;
485 }
486
487 static void destroy_inodecache(void)
488 {
489         if (kmem_cache_destroy(ext3_inode_cachep))
490                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
491 }
492
493 static void ext3_clear_inode(struct inode *inode)
494 {
495         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
496 #ifdef CONFIG_EXT3_FS_POSIX_ACL
497        if (EXT3_I(inode)->i_acl &&
498            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
499                posix_acl_release(EXT3_I(inode)->i_acl);
500                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
501        }
502        if (EXT3_I(inode)->i_default_acl &&
503            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
504                posix_acl_release(EXT3_I(inode)->i_default_acl);
505                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
506        }
507 #endif
508         ext3_discard_reservation(inode);
509         EXT3_I(inode)->i_block_alloc_info = NULL;
510         kfree(rsv);
511 }
512
513 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
514 {
515         struct super_block *sb = vfs->mnt_sb;
516         struct ext3_sb_info *sbi = EXT3_SB(sb);
517
518         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
519                 seq_puts(seq, ",data=journal");
520         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
521                 seq_puts(seq, ",data=ordered");
522         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
523                 seq_puts(seq, ",data=writeback");
524
525 #if defined(CONFIG_QUOTA)
526         if (sbi->s_jquota_fmt)
527                 seq_printf(seq, ",jqfmt=%s",
528                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
529
530         if (sbi->s_qf_names[USRQUOTA])
531                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
532
533         if (sbi->s_qf_names[GRPQUOTA])
534                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
535
536         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
537                 seq_puts(seq, ",usrquota");
538
539         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
540                 seq_puts(seq, ",grpquota");
541 #endif
542
543         return 0;
544 }
545
546 #ifdef CONFIG_QUOTA
547 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
548 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
549
550 static int ext3_dquot_initialize(struct inode *inode, int type);
551 static int ext3_dquot_drop(struct inode *inode);
552 static int ext3_write_dquot(struct dquot *dquot);
553 static int ext3_acquire_dquot(struct dquot *dquot);
554 static int ext3_release_dquot(struct dquot *dquot);
555 static int ext3_mark_dquot_dirty(struct dquot *dquot);
556 static int ext3_write_info(struct super_block *sb, int type);
557 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
558 static int ext3_quota_on_mount(struct super_block *sb, int type);
559 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
560                                size_t len, loff_t off);
561 static ssize_t ext3_quota_write(struct super_block *sb, int type,
562                                 const char *data, size_t len, loff_t off);
563
564 static struct dquot_operations ext3_quota_operations = {
565         .initialize     = ext3_dquot_initialize,
566         .drop           = ext3_dquot_drop,
567         .alloc_space    = dquot_alloc_space,
568         .alloc_inode    = dquot_alloc_inode,
569         .free_space     = dquot_free_space,
570         .free_inode     = dquot_free_inode,
571         .transfer       = dquot_transfer,
572         .write_dquot    = ext3_write_dquot,
573         .acquire_dquot  = ext3_acquire_dquot,
574         .release_dquot  = ext3_release_dquot,
575         .mark_dirty     = ext3_mark_dquot_dirty,
576         .write_info     = ext3_write_info
577 };
578
579 static struct quotactl_ops ext3_qctl_operations = {
580         .quota_on       = ext3_quota_on,
581         .quota_off      = vfs_quota_off,
582         .quota_sync     = vfs_quota_sync,
583         .get_info       = vfs_get_dqinfo,
584         .set_info       = vfs_set_dqinfo,
585         .get_dqblk      = vfs_get_dqblk,
586         .set_dqblk      = vfs_set_dqblk
587 };
588 #endif
589
590 static struct super_operations ext3_sops = {
591         .alloc_inode    = ext3_alloc_inode,
592         .destroy_inode  = ext3_destroy_inode,
593         .read_inode     = ext3_read_inode,
594         .write_inode    = ext3_write_inode,
595         .dirty_inode    = ext3_dirty_inode,
596         .delete_inode   = ext3_delete_inode,
597         .put_super      = ext3_put_super,
598         .write_super    = ext3_write_super,
599         .sync_fs        = ext3_sync_fs,
600         .write_super_lockfs = ext3_write_super_lockfs,
601         .unlockfs       = ext3_unlockfs,
602         .statfs         = ext3_statfs,
603         .remount_fs     = ext3_remount,
604         .clear_inode    = ext3_clear_inode,
605         .show_options   = ext3_show_options,
606 #ifdef CONFIG_QUOTA
607         .quota_read     = ext3_quota_read,
608         .quota_write    = ext3_quota_write,
609 #endif
610 };
611
612 struct dentry *ext3_get_parent(struct dentry *child);
613 static struct export_operations ext3_export_ops = {
614         .get_parent = ext3_get_parent,
615 };
616
617 enum {
618         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
619         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
620         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
621         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
622         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh,
623         Opt_commit, Opt_journal_update, Opt_journal_inum,
624         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
625         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
626         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
627         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
628         Opt_grpquota
629 };
630
631 static match_table_t tokens = {
632         {Opt_bsd_df, "bsddf"},
633         {Opt_minix_df, "minixdf"},
634         {Opt_grpid, "grpid"},
635         {Opt_grpid, "bsdgroups"},
636         {Opt_nogrpid, "nogrpid"},
637         {Opt_nogrpid, "sysvgroups"},
638         {Opt_resgid, "resgid=%u"},
639         {Opt_resuid, "resuid=%u"},
640         {Opt_sb, "sb=%u"},
641         {Opt_err_cont, "errors=continue"},
642         {Opt_err_panic, "errors=panic"},
643         {Opt_err_ro, "errors=remount-ro"},
644         {Opt_nouid32, "nouid32"},
645         {Opt_nocheck, "nocheck"},
646         {Opt_nocheck, "check=none"},
647         {Opt_check, "check"},
648         {Opt_debug, "debug"},
649         {Opt_oldalloc, "oldalloc"},
650         {Opt_orlov, "orlov"},
651         {Opt_user_xattr, "user_xattr"},
652         {Opt_nouser_xattr, "nouser_xattr"},
653         {Opt_acl, "acl"},
654         {Opt_noacl, "noacl"},
655         {Opt_reservation, "reservation"},
656         {Opt_noreservation, "noreservation"},
657         {Opt_noload, "noload"},
658         {Opt_nobh, "nobh"},
659         {Opt_commit, "commit=%u"},
660         {Opt_journal_update, "journal=update"},
661         {Opt_journal_inum, "journal=%u"},
662         {Opt_abort, "abort"},
663         {Opt_data_journal, "data=journal"},
664         {Opt_data_ordered, "data=ordered"},
665         {Opt_data_writeback, "data=writeback"},
666         {Opt_offusrjquota, "usrjquota="},
667         {Opt_usrjquota, "usrjquota=%s"},
668         {Opt_offgrpjquota, "grpjquota="},
669         {Opt_grpjquota, "grpjquota=%s"},
670         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
671         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
672         {Opt_grpquota, "grpquota"},
673         {Opt_noquota, "noquota"},
674         {Opt_quota, "quota"},
675         {Opt_usrquota, "usrquota"},
676         {Opt_barrier, "barrier=%u"},
677         {Opt_err, NULL},
678         {Opt_resize, "resize"},
679 };
680
681 static unsigned long get_sb_block(void **data)
682 {
683         unsigned long   sb_block;
684         char            *options = (char *) *data;
685
686         if (!options || strncmp(options, "sb=", 3) != 0)
687                 return 1;       /* Default location */
688         options += 3;
689         sb_block = simple_strtoul(options, &options, 0);
690         if (*options && *options != ',') {
691                 printk("EXT3-fs: Invalid sb specification: %s\n",
692                        (char *) *data);
693                 return 1;
694         }
695         if (*options == ',')
696                 options++;
697         *data = (void *) options;
698         return sb_block;
699 }
700
701 static int parse_options (char * options, struct super_block *sb,
702                           unsigned long * inum, unsigned long *n_blocks_count, int is_remount)
703 {
704         struct ext3_sb_info *sbi = EXT3_SB(sb);
705         char * p;
706         substring_t args[MAX_OPT_ARGS];
707         int data_opt = 0;
708         int option;
709 #ifdef CONFIG_QUOTA
710         int qtype;
711         char *qname;
712 #endif
713
714         if (!options)
715                 return 1;
716
717         while ((p = strsep (&options, ",")) != NULL) {
718                 int token;
719                 if (!*p)
720                         continue;
721
722                 token = match_token(p, tokens, args);
723                 switch (token) {
724                 case Opt_bsd_df:
725                         clear_opt (sbi->s_mount_opt, MINIX_DF);
726                         break;
727                 case Opt_minix_df:
728                         set_opt (sbi->s_mount_opt, MINIX_DF);
729                         break;
730                 case Opt_grpid:
731                         set_opt (sbi->s_mount_opt, GRPID);
732                         break;
733                 case Opt_nogrpid:
734                         clear_opt (sbi->s_mount_opt, GRPID);
735                         break;
736                 case Opt_resuid:
737                         if (match_int(&args[0], &option))
738                                 return 0;
739                         sbi->s_resuid = option;
740                         break;
741                 case Opt_resgid:
742                         if (match_int(&args[0], &option))
743                                 return 0;
744                         sbi->s_resgid = option;
745                         break;
746                 case Opt_sb:
747                         /* handled by get_sb_block() instead of here */
748                         /* *sb_block = match_int(&args[0]); */
749                         break;
750                 case Opt_err_panic:
751                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
752                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
753                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
754                         break;
755                 case Opt_err_ro:
756                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
757                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
758                         set_opt (sbi->s_mount_opt, ERRORS_RO);
759                         break;
760                 case Opt_err_cont:
761                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
762                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
763                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
764                         break;
765                 case Opt_nouid32:
766                         set_opt (sbi->s_mount_opt, NO_UID32);
767                         break;
768                 case Opt_check:
769 #ifdef CONFIG_EXT3_CHECK
770                         set_opt (sbi->s_mount_opt, CHECK);
771 #else
772                         printk(KERN_ERR
773                                "EXT3 Check option not supported\n");
774 #endif
775                         break;
776                 case Opt_nocheck:
777                         clear_opt (sbi->s_mount_opt, CHECK);
778                         break;
779                 case Opt_debug:
780                         set_opt (sbi->s_mount_opt, DEBUG);
781                         break;
782                 case Opt_oldalloc:
783                         set_opt (sbi->s_mount_opt, OLDALLOC);
784                         break;
785                 case Opt_orlov:
786                         clear_opt (sbi->s_mount_opt, OLDALLOC);
787                         break;
788 #ifdef CONFIG_EXT3_FS_XATTR
789                 case Opt_user_xattr:
790                         set_opt (sbi->s_mount_opt, XATTR_USER);
791                         break;
792                 case Opt_nouser_xattr:
793                         clear_opt (sbi->s_mount_opt, XATTR_USER);
794                         break;
795 #else
796                 case Opt_user_xattr:
797                 case Opt_nouser_xattr:
798                         printk("EXT3 (no)user_xattr options not supported\n");
799                         break;
800 #endif
801 #ifdef CONFIG_EXT3_FS_POSIX_ACL
802                 case Opt_acl:
803                         set_opt(sbi->s_mount_opt, POSIX_ACL);
804                         break;
805                 case Opt_noacl:
806                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
807                         break;
808 #else
809                 case Opt_acl:
810                 case Opt_noacl:
811                         printk("EXT3 (no)acl options not supported\n");
812                         break;
813 #endif
814                 case Opt_reservation:
815                         set_opt(sbi->s_mount_opt, RESERVATION);
816                         break;
817                 case Opt_noreservation:
818                         clear_opt(sbi->s_mount_opt, RESERVATION);
819                         break;
820                 case Opt_journal_update:
821                         /* @@@ FIXME */
822                         /* Eventually we will want to be able to create
823                            a journal file here.  For now, only allow the
824                            user to specify an existing inode to be the
825                            journal file. */
826                         if (is_remount) {
827                                 printk(KERN_ERR "EXT3-fs: cannot specify "
828                                        "journal on remount\n");
829                                 return 0;
830                         }
831                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
832                         break;
833                 case Opt_journal_inum:
834                         if (is_remount) {
835                                 printk(KERN_ERR "EXT3-fs: cannot specify "
836                                        "journal on remount\n");
837                                 return 0;
838                         }
839                         if (match_int(&args[0], &option))
840                                 return 0;
841                         *inum = option;
842                         break;
843                 case Opt_noload:
844                         set_opt (sbi->s_mount_opt, NOLOAD);
845                         break;
846                 case Opt_commit:
847                         if (match_int(&args[0], &option))
848                                 return 0;
849                         if (option < 0)
850                                 return 0;
851                         if (option == 0)
852                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
853                         sbi->s_commit_interval = HZ * option;
854                         break;
855                 case Opt_data_journal:
856                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
857                         goto datacheck;
858                 case Opt_data_ordered:
859                         data_opt = EXT3_MOUNT_ORDERED_DATA;
860                         goto datacheck;
861                 case Opt_data_writeback:
862                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
863                 datacheck:
864                         if (is_remount) {
865                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
866                                                 != data_opt) {
867                                         printk(KERN_ERR
868                                                 "EXT3-fs: cannot change data "
869                                                 "mode on remount\n");
870                                         return 0;
871                                 }
872                         } else {
873                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
874                                 sbi->s_mount_opt |= data_opt;
875                         }
876                         break;
877 #ifdef CONFIG_QUOTA
878                 case Opt_usrjquota:
879                         qtype = USRQUOTA;
880                         goto set_qf_name;
881                 case Opt_grpjquota:
882                         qtype = GRPQUOTA;
883 set_qf_name:
884                         if (sb_any_quota_enabled(sb)) {
885                                 printk(KERN_ERR
886                                         "EXT3-fs: Cannot change journalled "
887                                         "quota options when quota turned on.\n");
888                                 return 0;
889                         }
890                         qname = match_strdup(&args[0]);
891                         if (!qname) {
892                                 printk(KERN_ERR
893                                         "EXT3-fs: not enough memory for "
894                                         "storing quotafile name.\n");
895                                 return 0;
896                         }
897                         if (sbi->s_qf_names[qtype] &&
898                             strcmp(sbi->s_qf_names[qtype], qname)) {
899                                 printk(KERN_ERR
900                                         "EXT3-fs: %s quota file already "
901                                         "specified.\n", QTYPE2NAME(qtype));
902                                 kfree(qname);
903                                 return 0;
904                         }
905                         sbi->s_qf_names[qtype] = qname;
906                         if (strchr(sbi->s_qf_names[qtype], '/')) {
907                                 printk(KERN_ERR
908                                         "EXT3-fs: quotafile must be on "
909                                         "filesystem root.\n");
910                                 kfree(sbi->s_qf_names[qtype]);
911                                 sbi->s_qf_names[qtype] = NULL;
912                                 return 0;
913                         }
914                         set_opt(sbi->s_mount_opt, QUOTA);
915                         break;
916                 case Opt_offusrjquota:
917                         qtype = USRQUOTA;
918                         goto clear_qf_name;
919                 case Opt_offgrpjquota:
920                         qtype = GRPQUOTA;
921 clear_qf_name:
922                         if (sb_any_quota_enabled(sb)) {
923                                 printk(KERN_ERR "EXT3-fs: Cannot change "
924                                         "journalled quota options when "
925                                         "quota turned on.\n");
926                                 return 0;
927                         }
928                         /*
929                          * The space will be released later when all options
930                          * are confirmed to be correct
931                          */
932                         sbi->s_qf_names[qtype] = NULL;
933                         break;
934                 case Opt_jqfmt_vfsold:
935                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
936                         break;
937                 case Opt_jqfmt_vfsv0:
938                         sbi->s_jquota_fmt = QFMT_VFS_V0;
939                         break;
940                 case Opt_quota:
941                 case Opt_usrquota:
942                         set_opt(sbi->s_mount_opt, QUOTA);
943                         set_opt(sbi->s_mount_opt, USRQUOTA);
944                         break;
945                 case Opt_grpquota:
946                         set_opt(sbi->s_mount_opt, QUOTA);
947                         set_opt(sbi->s_mount_opt, GRPQUOTA);
948                         break;
949                 case Opt_noquota:
950                         if (sb_any_quota_enabled(sb)) {
951                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
952                                         "options when quota turned on.\n");
953                                 return 0;
954                         }
955                         clear_opt(sbi->s_mount_opt, QUOTA);
956                         clear_opt(sbi->s_mount_opt, USRQUOTA);
957                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
958                         break;
959 #else
960                 case Opt_quota:
961                 case Opt_usrquota:
962                 case Opt_grpquota:
963                 case Opt_usrjquota:
964                 case Opt_grpjquota:
965                 case Opt_offusrjquota:
966                 case Opt_offgrpjquota:
967                 case Opt_jqfmt_vfsold:
968                 case Opt_jqfmt_vfsv0:
969                         printk(KERN_ERR
970                                 "EXT3-fs: journalled quota options not "
971                                 "supported.\n");
972                         break;
973                 case Opt_noquota:
974                         break;
975 #endif
976                 case Opt_abort:
977                         set_opt(sbi->s_mount_opt, ABORT);
978                         break;
979                 case Opt_barrier:
980                         if (match_int(&args[0], &option))
981                                 return 0;
982                         if (option)
983                                 set_opt(sbi->s_mount_opt, BARRIER);
984                         else
985                                 clear_opt(sbi->s_mount_opt, BARRIER);
986                         break;
987                 case Opt_ignore:
988                         break;
989                 case Opt_resize:
990                         if (!is_remount) {
991                                 printk("EXT3-fs: resize option only available "
992                                         "for remount\n");
993                                 return 0;
994                         }
995                         if (match_int(&args[0], &option) != 0)
996                                 return 0;
997                         *n_blocks_count = option;
998                         break;
999                 case Opt_nobh:
1000                         set_opt(sbi->s_mount_opt, NOBH);
1001                         break;
1002                 default:
1003                         printk (KERN_ERR
1004                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1005                                 "or missing value\n", p);
1006                         return 0;
1007                 }
1008         }
1009 #ifdef CONFIG_QUOTA
1010         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1011                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1012                      sbi->s_qf_names[USRQUOTA])
1013                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1014
1015                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1016                      sbi->s_qf_names[GRPQUOTA])
1017                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1018
1019                 if ((sbi->s_qf_names[USRQUOTA] &&
1020                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1021                     (sbi->s_qf_names[GRPQUOTA] &&
1022                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1023                         printk(KERN_ERR "EXT3-fs: old and new quota "
1024                                         "format mixing.\n");
1025                         return 0;
1026                 }
1027
1028                 if (!sbi->s_jquota_fmt) {
1029                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1030                                         "not specified.\n");
1031                         return 0;
1032                 }
1033         } else {
1034                 if (sbi->s_jquota_fmt) {
1035                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1036                                         "specified with no journalling "
1037                                         "enabled.\n");
1038                         return 0;
1039                 }
1040         }
1041 #endif
1042         return 1;
1043 }
1044
1045 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1046                             int read_only)
1047 {
1048         struct ext3_sb_info *sbi = EXT3_SB(sb);
1049         int res = 0;
1050
1051         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1052                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1053                         "forcing read-only mode\n");
1054                 res = MS_RDONLY;
1055         }
1056         if (read_only)
1057                 return res;
1058         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1059                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1060                         "running e2fsck is recommended\n");
1061         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1062                 printk (KERN_WARNING
1063                         "EXT3-fs warning: mounting fs with errors, "
1064                         "running e2fsck is recommended\n");
1065         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1066                  le16_to_cpu(es->s_mnt_count) >=
1067                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1068                 printk (KERN_WARNING
1069                         "EXT3-fs warning: maximal mount count reached, "
1070                         "running e2fsck is recommended\n");
1071         else if (le32_to_cpu(es->s_checkinterval) &&
1072                 (le32_to_cpu(es->s_lastcheck) +
1073                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1074                 printk (KERN_WARNING
1075                         "EXT3-fs warning: checktime reached, "
1076                         "running e2fsck is recommended\n");
1077 #if 0
1078                 /* @@@ We _will_ want to clear the valid bit if we find
1079                    inconsistencies, to force a fsck at reboot.  But for
1080                    a plain journaled filesystem we can keep it set as
1081                    valid forever! :) */
1082         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1083 #endif
1084         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1085                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1086         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1087         es->s_mtime = cpu_to_le32(get_seconds());
1088         ext3_update_dynamic_rev(sb);
1089         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1090
1091         ext3_commit_super(sb, es, 1);
1092         if (test_opt(sb, DEBUG))
1093                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1094                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1095                         sb->s_blocksize,
1096                         sbi->s_groups_count,
1097                         EXT3_BLOCKS_PER_GROUP(sb),
1098                         EXT3_INODES_PER_GROUP(sb),
1099                         sbi->s_mount_opt);
1100
1101         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1102         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1103                 char b[BDEVNAME_SIZE];
1104
1105                 printk("external journal on %s\n",
1106                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1107         } else {
1108                 printk("internal journal\n");
1109         }
1110 #ifdef CONFIG_EXT3_CHECK
1111         if (test_opt (sb, CHECK)) {
1112                 ext3_check_blocks_bitmap (sb);
1113                 ext3_check_inodes_bitmap (sb);
1114         }
1115 #endif
1116         return res;
1117 }
1118
1119 /* Called at mount-time, super-block is locked */
1120 static int ext3_check_descriptors (struct super_block * sb)
1121 {
1122         struct ext3_sb_info *sbi = EXT3_SB(sb);
1123         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1124         struct ext3_group_desc * gdp = NULL;
1125         int desc_block = 0;
1126         int i;
1127
1128         ext3_debug ("Checking group descriptors");
1129
1130         for (i = 0; i < sbi->s_groups_count; i++)
1131         {
1132                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1133                         gdp = (struct ext3_group_desc *)
1134                                         sbi->s_group_desc[desc_block++]->b_data;
1135                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1136                     le32_to_cpu(gdp->bg_block_bitmap) >=
1137                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1138                 {
1139                         ext3_error (sb, "ext3_check_descriptors",
1140                                     "Block bitmap for group %d"
1141                                     " not in group (block %lu)!",
1142                                     i, (unsigned long)
1143                                         le32_to_cpu(gdp->bg_block_bitmap));
1144                         return 0;
1145                 }
1146                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1147                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1148                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1149                 {
1150                         ext3_error (sb, "ext3_check_descriptors",
1151                                     "Inode bitmap for group %d"
1152                                     " not in group (block %lu)!",
1153                                     i, (unsigned long)
1154                                         le32_to_cpu(gdp->bg_inode_bitmap));
1155                         return 0;
1156                 }
1157                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1158                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1159                     block + EXT3_BLOCKS_PER_GROUP(sb))
1160                 {
1161                         ext3_error (sb, "ext3_check_descriptors",
1162                                     "Inode table for group %d"
1163                                     " not in group (block %lu)!",
1164                                     i, (unsigned long)
1165                                         le32_to_cpu(gdp->bg_inode_table));
1166                         return 0;
1167                 }
1168                 block += EXT3_BLOCKS_PER_GROUP(sb);
1169                 gdp++;
1170         }
1171
1172         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1173         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1174         return 1;
1175 }
1176
1177
1178 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1179  * the superblock) which were deleted from all directories, but held open by
1180  * a process at the time of a crash.  We walk the list and try to delete these
1181  * inodes at recovery time (only with a read-write filesystem).
1182  *
1183  * In order to keep the orphan inode chain consistent during traversal (in
1184  * case of crash during recovery), we link each inode into the superblock
1185  * orphan list_head and handle it the same way as an inode deletion during
1186  * normal operation (which journals the operations for us).
1187  *
1188  * We only do an iget() and an iput() on each inode, which is very safe if we
1189  * accidentally point at an in-use or already deleted inode.  The worst that
1190  * can happen in this case is that we get a "bit already cleared" message from
1191  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1192  * e2fsck was run on this filesystem, and it must have already done the orphan
1193  * inode cleanup for us, so we can safely abort without any further action.
1194  */
1195 static void ext3_orphan_cleanup (struct super_block * sb,
1196                                  struct ext3_super_block * es)
1197 {
1198         unsigned int s_flags = sb->s_flags;
1199         int nr_orphans = 0, nr_truncates = 0;
1200 #ifdef CONFIG_QUOTA
1201         int i;
1202 #endif
1203         if (!es->s_last_orphan) {
1204                 jbd_debug(4, "no orphan inodes to clean up\n");
1205                 return;
1206         }
1207
1208         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1209                 if (es->s_last_orphan)
1210                         jbd_debug(1, "Errors on filesystem, "
1211                                   "clearing orphan list.\n");
1212                 es->s_last_orphan = 0;
1213                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1214                 return;
1215         }
1216
1217         if (s_flags & MS_RDONLY) {
1218                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1219                        sb->s_id);
1220                 sb->s_flags &= ~MS_RDONLY;
1221         }
1222 #ifdef CONFIG_QUOTA
1223         /* Needed for iput() to work correctly and not trash data */
1224         sb->s_flags |= MS_ACTIVE;
1225         /* Turn on quotas so that they are updated correctly */
1226         for (i = 0; i < MAXQUOTAS; i++) {
1227                 if (EXT3_SB(sb)->s_qf_names[i]) {
1228                         int ret = ext3_quota_on_mount(sb, i);
1229                         if (ret < 0)
1230                                 printk(KERN_ERR
1231                                         "EXT3-fs: Cannot turn on journalled "
1232                                         "quota: error %d\n", ret);
1233                 }
1234         }
1235 #endif
1236
1237         while (es->s_last_orphan) {
1238                 struct inode *inode;
1239
1240                 if (!(inode =
1241                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1242                         es->s_last_orphan = 0;
1243                         break;
1244                 }
1245
1246                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1247                 DQUOT_INIT(inode);
1248                 if (inode->i_nlink) {
1249                         printk(KERN_DEBUG
1250                                 "%s: truncating inode %ld to %Ld bytes\n",
1251                                 __FUNCTION__, inode->i_ino, inode->i_size);
1252                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1253                                   inode->i_ino, inode->i_size);
1254                         ext3_truncate(inode);
1255                         nr_truncates++;
1256                 } else {
1257                         printk(KERN_DEBUG
1258                                 "%s: deleting unreferenced inode %ld\n",
1259                                 __FUNCTION__, inode->i_ino);
1260                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1261                                   inode->i_ino);
1262                         nr_orphans++;
1263                 }
1264                 iput(inode);  /* The delete magic happens here! */
1265         }
1266
1267 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1268
1269         if (nr_orphans)
1270                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1271                        sb->s_id, PLURAL(nr_orphans));
1272         if (nr_truncates)
1273                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1274                        sb->s_id, PLURAL(nr_truncates));
1275 #ifdef CONFIG_QUOTA
1276         /* Turn quotas off */
1277         for (i = 0; i < MAXQUOTAS; i++) {
1278                 if (sb_dqopt(sb)->files[i])
1279                         vfs_quota_off(sb, i);
1280         }
1281 #endif
1282         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1283 }
1284
1285 #define log2(n) ffz(~(n))
1286
1287 /*
1288  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1289  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1290  * We need to be 1 filesystem block less than the 2^32 sector limit.
1291  */
1292 static loff_t ext3_max_size(int bits)
1293 {
1294         loff_t res = EXT3_NDIR_BLOCKS;
1295         /* This constant is calculated to be the largest file size for a
1296          * dense, 4k-blocksize file such that the total number of
1297          * sectors in the file, including data and all indirect blocks,
1298          * does not exceed 2^32. */
1299         const loff_t upper_limit = 0x1ff7fffd000LL;
1300
1301         res += 1LL << (bits-2);
1302         res += 1LL << (2*(bits-2));
1303         res += 1LL << (3*(bits-2));
1304         res <<= bits;
1305         if (res > upper_limit)
1306                 res = upper_limit;
1307         return res;
1308 }
1309
1310 static unsigned long descriptor_loc(struct super_block *sb,
1311                                     unsigned long logic_sb_block,
1312                                     int nr)
1313 {
1314         struct ext3_sb_info *sbi = EXT3_SB(sb);
1315         unsigned long bg, first_data_block, first_meta_bg;
1316         int has_super = 0;
1317
1318         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1319         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1320
1321         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1322             nr < first_meta_bg)
1323                 return (logic_sb_block + nr + 1);
1324         bg = sbi->s_desc_per_block * nr;
1325         if (ext3_bg_has_super(sb, bg))
1326                 has_super = 1;
1327         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1328 }
1329
1330
1331 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1332 {
1333         struct buffer_head * bh;
1334         struct ext3_super_block *es = NULL;
1335         struct ext3_sb_info *sbi;
1336         unsigned long block;
1337         unsigned long sb_block = get_sb_block(&data);
1338         unsigned long logic_sb_block;
1339         unsigned long offset = 0;
1340         unsigned long journal_inum = 0;
1341         unsigned long def_mount_opts;
1342         struct inode *root;
1343         int blocksize;
1344         int hblock;
1345         int db_count;
1346         int i;
1347         int needs_recovery;
1348         __le32 features;
1349
1350         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1351         if (!sbi)
1352                 return -ENOMEM;
1353         sb->s_fs_info = sbi;
1354         memset(sbi, 0, sizeof(*sbi));
1355         sbi->s_mount_opt = 0;
1356         sbi->s_resuid = EXT3_DEF_RESUID;
1357         sbi->s_resgid = EXT3_DEF_RESGID;
1358
1359         unlock_kernel();
1360
1361         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1362         if (!blocksize) {
1363                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1364                 goto out_fail;
1365         }
1366
1367         /*
1368          * The ext3 superblock will not be buffer aligned for other than 1kB
1369          * block sizes.  We need to calculate the offset from buffer start.
1370          */
1371         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1372                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1373                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1374         } else {
1375                 logic_sb_block = sb_block;
1376         }
1377
1378         if (!(bh = sb_bread(sb, logic_sb_block))) {
1379                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1380                 goto out_fail;
1381         }
1382         /*
1383          * Note: s_es must be initialized as soon as possible because
1384          *       some ext3 macro-instructions depend on its value
1385          */
1386         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1387         sbi->s_es = es;
1388         sb->s_magic = le16_to_cpu(es->s_magic);
1389         if (sb->s_magic != EXT3_SUPER_MAGIC)
1390                 goto cantfind_ext3;
1391
1392         /* Set defaults before we parse the mount options */
1393         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1394         if (def_mount_opts & EXT3_DEFM_DEBUG)
1395                 set_opt(sbi->s_mount_opt, DEBUG);
1396         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1397                 set_opt(sbi->s_mount_opt, GRPID);
1398         if (def_mount_opts & EXT3_DEFM_UID16)
1399                 set_opt(sbi->s_mount_opt, NO_UID32);
1400         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1401                 set_opt(sbi->s_mount_opt, XATTR_USER);
1402         if (def_mount_opts & EXT3_DEFM_ACL)
1403                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1404         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1405                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1406         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1407                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1408         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1409                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1410
1411         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1412                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1413         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1414                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1415
1416         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1417         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1418
1419         set_opt(sbi->s_mount_opt, RESERVATION);
1420
1421         if (!parse_options ((char *) data, sb, &journal_inum, NULL, 0))
1422                 goto failed_mount;
1423
1424         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1425                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1426
1427         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1428             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1429              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1430              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1431                 printk(KERN_WARNING 
1432                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1433                        "running e2fsck is recommended\n");
1434         /*
1435          * Check feature flags regardless of the revision level, since we
1436          * previously didn't change the revision level when setting the flags,
1437          * so there is a chance incompat flags are set on a rev 0 filesystem.
1438          */
1439         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1440         if (features) {
1441                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1442                        "unsupported optional features (%x).\n",
1443                        sb->s_id, le32_to_cpu(features));
1444                 goto failed_mount;
1445         }
1446         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1447         if (!(sb->s_flags & MS_RDONLY) && features) {
1448                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1449                        "unsupported optional features (%x).\n",
1450                        sb->s_id, le32_to_cpu(features));
1451                 goto failed_mount;
1452         }
1453         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1454
1455         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1456             blocksize > EXT3_MAX_BLOCK_SIZE) {
1457                 printk(KERN_ERR 
1458                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1459                        blocksize, sb->s_id);
1460                 goto failed_mount;
1461         }
1462
1463         hblock = bdev_hardsect_size(sb->s_bdev);
1464         if (sb->s_blocksize != blocksize) {
1465                 /*
1466                  * Make sure the blocksize for the filesystem is larger
1467                  * than the hardware sectorsize for the machine.
1468                  */
1469                 if (blocksize < hblock) {
1470                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1471                                "device blocksize %d.\n", blocksize, hblock);
1472                         goto failed_mount;
1473                 }
1474
1475                 brelse (bh);
1476                 sb_set_blocksize(sb, blocksize);
1477                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1478                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1479                 bh = sb_bread(sb, logic_sb_block);
1480                 if (!bh) {
1481                         printk(KERN_ERR 
1482                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1483                         goto failed_mount;
1484                 }
1485                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1486                 sbi->s_es = es;
1487                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1488                         printk (KERN_ERR 
1489                                 "EXT3-fs: Magic mismatch, very weird !\n");
1490                         goto failed_mount;
1491                 }
1492         }
1493
1494         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1495
1496         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1497                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1498                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1499         } else {
1500                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1501                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1502                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1503                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1504                     (sbi->s_inode_size > blocksize)) {
1505                         printk (KERN_ERR
1506                                 "EXT3-fs: unsupported inode size: %d\n",
1507                                 sbi->s_inode_size);
1508                         goto failed_mount;
1509                 }
1510         }
1511         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1512                                    le32_to_cpu(es->s_log_frag_size);
1513         if (blocksize != sbi->s_frag_size) {
1514                 printk(KERN_ERR
1515                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1516                        sbi->s_frag_size, blocksize);
1517                 goto failed_mount;
1518         }
1519         sbi->s_frags_per_block = 1;
1520         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1521         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1522         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1523         if (EXT3_INODE_SIZE(sb) == 0)
1524                 goto cantfind_ext3;
1525         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1526         if (sbi->s_inodes_per_block == 0)
1527                 goto cantfind_ext3;
1528         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1529                                         sbi->s_inodes_per_block;
1530         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1531         sbi->s_sbh = bh;
1532         sbi->s_mount_state = le16_to_cpu(es->s_state);
1533         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1534         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1535         for (i=0; i < 4; i++)
1536                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1537         sbi->s_def_hash_version = es->s_def_hash_version;
1538
1539         if (sbi->s_blocks_per_group > blocksize * 8) {
1540                 printk (KERN_ERR
1541                         "EXT3-fs: #blocks per group too big: %lu\n",
1542                         sbi->s_blocks_per_group);
1543                 goto failed_mount;
1544         }
1545         if (sbi->s_frags_per_group > blocksize * 8) {
1546                 printk (KERN_ERR
1547                         "EXT3-fs: #fragments per group too big: %lu\n",
1548                         sbi->s_frags_per_group);
1549                 goto failed_mount;
1550         }
1551         if (sbi->s_inodes_per_group > blocksize * 8) {
1552                 printk (KERN_ERR
1553                         "EXT3-fs: #inodes per group too big: %lu\n",
1554                         sbi->s_inodes_per_group);
1555                 goto failed_mount;
1556         }
1557
1558         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1559                 goto cantfind_ext3;
1560         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1561                                le32_to_cpu(es->s_first_data_block) +
1562                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1563                               EXT3_BLOCKS_PER_GROUP(sb);
1564         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1565                    EXT3_DESC_PER_BLOCK(sb);
1566         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1567                                     GFP_KERNEL);
1568         if (sbi->s_group_desc == NULL) {
1569                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1570                 goto failed_mount;
1571         }
1572
1573         percpu_counter_init(&sbi->s_freeblocks_counter);
1574         percpu_counter_init(&sbi->s_freeinodes_counter);
1575         percpu_counter_init(&sbi->s_dirs_counter);
1576         bgl_lock_init(&sbi->s_blockgroup_lock);
1577
1578         for (i = 0; i < db_count; i++) {
1579                 block = descriptor_loc(sb, logic_sb_block, i);
1580                 sbi->s_group_desc[i] = sb_bread(sb, block);
1581                 if (!sbi->s_group_desc[i]) {
1582                         printk (KERN_ERR "EXT3-fs: "
1583                                 "can't read group descriptor %d\n", i);
1584                         db_count = i;
1585                         goto failed_mount2;
1586                 }
1587         }
1588         if (!ext3_check_descriptors (sb)) {
1589                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1590                 goto failed_mount2;
1591         }
1592         sbi->s_gdb_count = db_count;
1593         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1594         spin_lock_init(&sbi->s_next_gen_lock);
1595         /* per fileystem reservation list head & lock */
1596         spin_lock_init(&sbi->s_rsv_window_lock);
1597         sbi->s_rsv_window_root = RB_ROOT;
1598         /* Add a single, static dummy reservation to the start of the
1599          * reservation window list --- it gives us a placeholder for
1600          * append-at-start-of-list which makes the allocation logic
1601          * _much_ simpler. */
1602         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1603         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1604         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1605         sbi->s_rsv_window_head.rsv_goal_size = 0;
1606         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1607
1608         /*
1609          * set up enough so that it can read an inode
1610          */
1611         sb->s_op = &ext3_sops;
1612         sb->s_export_op = &ext3_export_ops;
1613         sb->s_xattr = ext3_xattr_handlers;
1614 #ifdef CONFIG_QUOTA
1615         sb->s_qcop = &ext3_qctl_operations;
1616         sb->dq_op = &ext3_quota_operations;
1617 #endif
1618         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1619
1620         sb->s_root = NULL;
1621
1622         needs_recovery = (es->s_last_orphan != 0 ||
1623                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1624                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1625
1626         /*
1627          * The first inode we look at is the journal inode.  Don't try
1628          * root first: it may be modified in the journal!
1629          */
1630         if (!test_opt(sb, NOLOAD) &&
1631             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1632                 if (ext3_load_journal(sb, es))
1633                         goto failed_mount2;
1634         } else if (journal_inum) {
1635                 if (ext3_create_journal(sb, es, journal_inum))
1636                         goto failed_mount2;
1637         } else {
1638                 if (!silent)
1639                         printk (KERN_ERR
1640                                 "ext3: No journal on filesystem on %s\n",
1641                                 sb->s_id);
1642                 goto failed_mount2;
1643         }
1644
1645         /* We have now updated the journal if required, so we can
1646          * validate the data journaling mode. */
1647         switch (test_opt(sb, DATA_FLAGS)) {
1648         case 0:
1649                 /* No mode set, assume a default based on the journal
1650                    capabilities: ORDERED_DATA if the journal can
1651                    cope, else JOURNAL_DATA */
1652                 if (journal_check_available_features
1653                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1654                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1655                 else
1656                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1657                 break;
1658
1659         case EXT3_MOUNT_ORDERED_DATA:
1660         case EXT3_MOUNT_WRITEBACK_DATA:
1661                 if (!journal_check_available_features
1662                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1663                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1664                                "requested data journaling mode\n");
1665                         goto failed_mount3;
1666                 }
1667         default:
1668                 break;
1669         }
1670
1671         if (test_opt(sb, NOBH)) {
1672                 if (sb->s_blocksize_bits != PAGE_CACHE_SHIFT) {
1673                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option "
1674                                 "since filesystem blocksize doesn't match "
1675                                 "pagesize\n");
1676                         clear_opt(sbi->s_mount_opt, NOBH);
1677                 }
1678                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1679                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1680                                 "its supported only with writeback mode\n");
1681                         clear_opt(sbi->s_mount_opt, NOBH);
1682                 }
1683         }
1684         /*
1685          * The journal_load will have done any necessary log recovery,
1686          * so we can safely mount the rest of the filesystem now.
1687          */
1688
1689         root = iget(sb, EXT3_ROOT_INO);
1690         sb->s_root = d_alloc_root(root);
1691         if (!sb->s_root) {
1692                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1693                 iput(root);
1694                 goto failed_mount3;
1695         }
1696         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1697                 dput(sb->s_root);
1698                 sb->s_root = NULL;
1699                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1700                 goto failed_mount3;
1701         }
1702
1703         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1704         /*
1705          * akpm: core read_super() calls in here with the superblock locked.
1706          * That deadlocks, because orphan cleanup needs to lock the superblock
1707          * in numerous places.  Here we just pop the lock - it's relatively
1708          * harmless, because we are now ready to accept write_super() requests,
1709          * and aviro says that's the only reason for hanging onto the
1710          * superblock lock.
1711          */
1712         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1713         ext3_orphan_cleanup(sb, es);
1714         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1715         if (needs_recovery)
1716                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1717         ext3_mark_recovery_complete(sb, es);
1718         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1719                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1720                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1721                 "writeback");
1722
1723         percpu_counter_mod(&sbi->s_freeblocks_counter,
1724                 ext3_count_free_blocks(sb));
1725         percpu_counter_mod(&sbi->s_freeinodes_counter,
1726                 ext3_count_free_inodes(sb));
1727         percpu_counter_mod(&sbi->s_dirs_counter,
1728                 ext3_count_dirs(sb));
1729
1730         lock_kernel();
1731         return 0;
1732
1733 cantfind_ext3:
1734         if (!silent)
1735                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1736                        sb->s_id);
1737         goto failed_mount;
1738
1739 failed_mount3:
1740         journal_destroy(sbi->s_journal);
1741 failed_mount2:
1742         for (i = 0; i < db_count; i++)
1743                 brelse(sbi->s_group_desc[i]);
1744         kfree(sbi->s_group_desc);
1745 failed_mount:
1746 #ifdef CONFIG_QUOTA
1747         for (i = 0; i < MAXQUOTAS; i++)
1748                 kfree(sbi->s_qf_names[i]);
1749 #endif
1750         ext3_blkdev_remove(sbi);
1751         brelse(bh);
1752 out_fail:
1753         sb->s_fs_info = NULL;
1754         kfree(sbi);
1755         lock_kernel();
1756         return -EINVAL;
1757 }
1758
1759 /*
1760  * Setup any per-fs journal parameters now.  We'll do this both on
1761  * initial mount, once the journal has been initialised but before we've
1762  * done any recovery; and again on any subsequent remount. 
1763  */
1764 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1765 {
1766         struct ext3_sb_info *sbi = EXT3_SB(sb);
1767
1768         if (sbi->s_commit_interval)
1769                 journal->j_commit_interval = sbi->s_commit_interval;
1770         /* We could also set up an ext3-specific default for the commit
1771          * interval here, but for now we'll just fall back to the jbd
1772          * default. */
1773
1774         spin_lock(&journal->j_state_lock);
1775         if (test_opt(sb, BARRIER))
1776                 journal->j_flags |= JFS_BARRIER;
1777         else
1778                 journal->j_flags &= ~JFS_BARRIER;
1779         spin_unlock(&journal->j_state_lock);
1780 }
1781
1782 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1783 {
1784         struct inode *journal_inode;
1785         journal_t *journal;
1786
1787         /* First, test for the existence of a valid inode on disk.  Bad
1788          * things happen if we iget() an unused inode, as the subsequent
1789          * iput() will try to delete it. */
1790
1791         journal_inode = iget(sb, journal_inum);
1792         if (!journal_inode) {
1793                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1794                 return NULL;
1795         }
1796         if (!journal_inode->i_nlink) {
1797                 make_bad_inode(journal_inode);
1798                 iput(journal_inode);
1799                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1800                 return NULL;
1801         }
1802
1803         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1804                   journal_inode, journal_inode->i_size);
1805         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1806                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1807                 iput(journal_inode);
1808                 return NULL;
1809         }
1810
1811         journal = journal_init_inode(journal_inode);
1812         if (!journal) {
1813                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1814                 iput(journal_inode);
1815                 return NULL;
1816         }
1817         journal->j_private = sb;
1818         ext3_init_journal_params(sb, journal);
1819         return journal;
1820 }
1821
1822 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1823                                        dev_t j_dev)
1824 {
1825         struct buffer_head * bh;
1826         journal_t *journal;
1827         int start;
1828         int len;
1829         int hblock, blocksize;
1830         unsigned long sb_block;
1831         unsigned long offset;
1832         struct ext3_super_block * es;
1833         struct block_device *bdev;
1834
1835         bdev = ext3_blkdev_get(j_dev);
1836         if (bdev == NULL)
1837                 return NULL;
1838
1839         if (bd_claim(bdev, sb)) {
1840                 printk(KERN_ERR
1841                         "EXT3: failed to claim external journal device.\n");
1842                 blkdev_put(bdev);
1843                 return NULL;
1844         }
1845
1846         blocksize = sb->s_blocksize;
1847         hblock = bdev_hardsect_size(bdev);
1848         if (blocksize < hblock) {
1849                 printk(KERN_ERR
1850                         "EXT3-fs: blocksize too small for journal device.\n");
1851                 goto out_bdev;
1852         }
1853
1854         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1855         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1856         set_blocksize(bdev, blocksize);
1857         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1858                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1859                        "external journal\n");
1860                 goto out_bdev;
1861         }
1862
1863         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1864         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1865             !(le32_to_cpu(es->s_feature_incompat) &
1866               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1867                 printk(KERN_ERR "EXT3-fs: external journal has "
1868                                         "bad superblock\n");
1869                 brelse(bh);
1870                 goto out_bdev;
1871         }
1872
1873         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1874                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1875                 brelse(bh);
1876                 goto out_bdev;
1877         }
1878
1879         len = le32_to_cpu(es->s_blocks_count);
1880         start = sb_block + 1;
1881         brelse(bh);     /* we're done with the superblock */
1882
1883         journal = journal_init_dev(bdev, sb->s_bdev,
1884                                         start, len, blocksize);
1885         if (!journal) {
1886                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1887                 goto out_bdev;
1888         }
1889         journal->j_private = sb;
1890         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1891         wait_on_buffer(journal->j_sb_buffer);
1892         if (!buffer_uptodate(journal->j_sb_buffer)) {
1893                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1894                 goto out_journal;
1895         }
1896         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1897                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1898                                         "user (unsupported) - %d\n",
1899                         be32_to_cpu(journal->j_superblock->s_nr_users));
1900                 goto out_journal;
1901         }
1902         EXT3_SB(sb)->journal_bdev = bdev;
1903         ext3_init_journal_params(sb, journal);
1904         return journal;
1905 out_journal:
1906         journal_destroy(journal);
1907 out_bdev:
1908         ext3_blkdev_put(bdev);
1909         return NULL;
1910 }
1911
1912 static int ext3_load_journal(struct super_block * sb,
1913                              struct ext3_super_block * es)
1914 {
1915         journal_t *journal;
1916         int journal_inum = le32_to_cpu(es->s_journal_inum);
1917         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1918         int err = 0;
1919         int really_read_only;
1920
1921         really_read_only = bdev_read_only(sb->s_bdev);
1922
1923         /*
1924          * Are we loading a blank journal or performing recovery after a
1925          * crash?  For recovery, we need to check in advance whether we
1926          * can get read-write access to the device.
1927          */
1928
1929         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1930                 if (sb->s_flags & MS_RDONLY) {
1931                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1932                                         "required on readonly filesystem.\n");
1933                         if (really_read_only) {
1934                                 printk(KERN_ERR "EXT3-fs: write access "
1935                                         "unavailable, cannot proceed.\n");
1936                                 return -EROFS;
1937                         }
1938                         printk (KERN_INFO "EXT3-fs: write access will "
1939                                         "be enabled during recovery.\n");
1940                 }
1941         }
1942
1943         if (journal_inum && journal_dev) {
1944                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1945                        "and inode journals!\n");
1946                 return -EINVAL;
1947         }
1948
1949         if (journal_inum) {
1950                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1951                         return -EINVAL;
1952         } else {
1953                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1954                         return -EINVAL;
1955         }
1956
1957         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1958                 err = journal_update_format(journal);
1959                 if (err)  {
1960                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1961                         journal_destroy(journal);
1962                         return err;
1963                 }
1964         }
1965
1966         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1967                 err = journal_wipe(journal, !really_read_only);
1968         if (!err)
1969                 err = journal_load(journal);
1970
1971         if (err) {
1972                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1973                 journal_destroy(journal);
1974                 return err;
1975         }
1976
1977         EXT3_SB(sb)->s_journal = journal;
1978         ext3_clear_journal_err(sb, es);
1979         return 0;
1980 }
1981
1982 static int ext3_create_journal(struct super_block * sb,
1983                                struct ext3_super_block * es,
1984                                int journal_inum)
1985 {
1986         journal_t *journal;
1987
1988         if (sb->s_flags & MS_RDONLY) {
1989                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1990                                 "create journal.\n");
1991                 return -EROFS;
1992         }
1993
1994         if (!(journal = ext3_get_journal(sb, journal_inum)))
1995                 return -EINVAL;
1996
1997         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1998                journal_inum);
1999
2000         if (journal_create(journal)) {
2001                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2002                 journal_destroy(journal);
2003                 return -EIO;
2004         }
2005
2006         EXT3_SB(sb)->s_journal = journal;
2007
2008         ext3_update_dynamic_rev(sb);
2009         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2010         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2011
2012         es->s_journal_inum = cpu_to_le32(journal_inum);
2013         sb->s_dirt = 1;
2014
2015         /* Make sure we flush the recovery flag to disk. */
2016         ext3_commit_super(sb, es, 1);
2017
2018         return 0;
2019 }
2020
2021 static void ext3_commit_super (struct super_block * sb,
2022                                struct ext3_super_block * es,
2023                                int sync)
2024 {
2025         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2026
2027         if (!sbh)
2028                 return;
2029         es->s_wtime = cpu_to_le32(get_seconds());
2030         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2031         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2032         BUFFER_TRACE(sbh, "marking dirty");
2033         mark_buffer_dirty(sbh);
2034         if (sync)
2035                 sync_dirty_buffer(sbh);
2036 }
2037
2038
2039 /*
2040  * Have we just finished recovery?  If so, and if we are mounting (or
2041  * remounting) the filesystem readonly, then we will end up with a
2042  * consistent fs on disk.  Record that fact.
2043  */
2044 static void ext3_mark_recovery_complete(struct super_block * sb,
2045                                         struct ext3_super_block * es)
2046 {
2047         journal_t *journal = EXT3_SB(sb)->s_journal;
2048
2049         journal_lock_updates(journal);
2050         journal_flush(journal);
2051         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2052             sb->s_flags & MS_RDONLY) {
2053                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2054                 sb->s_dirt = 0;
2055                 ext3_commit_super(sb, es, 1);
2056         }
2057         journal_unlock_updates(journal);
2058 }
2059
2060 /*
2061  * If we are mounting (or read-write remounting) a filesystem whose journal
2062  * has recorded an error from a previous lifetime, move that error to the
2063  * main filesystem now.
2064  */
2065 static void ext3_clear_journal_err(struct super_block * sb,
2066                                    struct ext3_super_block * es)
2067 {
2068         journal_t *journal;
2069         int j_errno;
2070         const char *errstr;
2071
2072         journal = EXT3_SB(sb)->s_journal;
2073
2074         /*
2075          * Now check for any error status which may have been recorded in the
2076          * journal by a prior ext3_error() or ext3_abort()
2077          */
2078
2079         j_errno = journal_errno(journal);
2080         if (j_errno) {
2081                 char nbuf[16];
2082
2083                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2084                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2085                              "from previous mount: %s", errstr);
2086                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2087                              "filesystem check.");
2088
2089                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2090                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2091                 ext3_commit_super (sb, es, 1);
2092
2093                 journal_clear_err(journal);
2094         }
2095 }
2096
2097 /*
2098  * Force the running and committing transactions to commit,
2099  * and wait on the commit.
2100  */
2101 int ext3_force_commit(struct super_block *sb)
2102 {
2103         journal_t *journal;
2104         int ret;
2105
2106         if (sb->s_flags & MS_RDONLY)
2107                 return 0;
2108
2109         journal = EXT3_SB(sb)->s_journal;
2110         sb->s_dirt = 0;
2111         ret = ext3_journal_force_commit(journal);
2112         return ret;
2113 }
2114
2115 /*
2116  * Ext3 always journals updates to the superblock itself, so we don't
2117  * have to propagate any other updates to the superblock on disk at this
2118  * point.  Just start an async writeback to get the buffers on their way
2119  * to the disk.
2120  *
2121  * This implicitly triggers the writebehind on sync().
2122  */
2123
2124 static void ext3_write_super (struct super_block * sb)
2125 {
2126         if (down_trylock(&sb->s_lock) == 0)
2127                 BUG();
2128         sb->s_dirt = 0;
2129 }
2130
2131 static int ext3_sync_fs(struct super_block *sb, int wait)
2132 {
2133         tid_t target;
2134
2135         sb->s_dirt = 0;
2136         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2137                 if (wait)
2138                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2139         }
2140         return 0;
2141 }
2142
2143 /*
2144  * LVM calls this function before a (read-only) snapshot is created.  This
2145  * gives us a chance to flush the journal completely and mark the fs clean.
2146  */
2147 static void ext3_write_super_lockfs(struct super_block *sb)
2148 {
2149         sb->s_dirt = 0;
2150
2151         if (!(sb->s_flags & MS_RDONLY)) {
2152                 journal_t *journal = EXT3_SB(sb)->s_journal;
2153
2154                 /* Now we set up the journal barrier. */
2155                 journal_lock_updates(journal);
2156                 journal_flush(journal);
2157
2158                 /* Journal blocked and flushed, clear needs_recovery flag. */
2159                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2160                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2161         }
2162 }
2163
2164 /*
2165  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2166  * flag here, even though the filesystem is not technically dirty yet.
2167  */
2168 static void ext3_unlockfs(struct super_block *sb)
2169 {
2170         if (!(sb->s_flags & MS_RDONLY)) {
2171                 lock_super(sb);
2172                 /* Reser the needs_recovery flag before the fs is unlocked. */
2173                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2174                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2175                 unlock_super(sb);
2176                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2177         }
2178 }
2179
2180 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2181 {
2182         struct ext3_super_block * es;
2183         struct ext3_sb_info *sbi = EXT3_SB(sb);
2184         unsigned long n_blocks_count = 0;
2185         unsigned long old_sb_flags;
2186         struct ext3_mount_options old_opts;
2187         int err;
2188 #ifdef CONFIG_QUOTA
2189         int i;
2190 #endif
2191
2192         /* Store the original options */
2193         old_sb_flags = sb->s_flags;
2194         old_opts.s_mount_opt = sbi->s_mount_opt;
2195         old_opts.s_resuid = sbi->s_resuid;
2196         old_opts.s_resgid = sbi->s_resgid;
2197         old_opts.s_commit_interval = sbi->s_commit_interval;
2198 #ifdef CONFIG_QUOTA
2199         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2200         for (i = 0; i < MAXQUOTAS; i++)
2201                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2202 #endif
2203
2204         /*
2205          * Allow the "check" option to be passed as a remount option.
2206          */
2207         if (!parse_options(data, sb, NULL, &n_blocks_count, 1)) {
2208                 err = -EINVAL;
2209                 goto restore_opts;
2210         }
2211
2212         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2213                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2214
2215         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2216                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2217
2218         es = sbi->s_es;
2219
2220         ext3_init_journal_params(sb, sbi->s_journal);
2221
2222         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2223                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2224                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2225                         err = -EROFS;
2226                         goto restore_opts;
2227                 }
2228
2229                 if (*flags & MS_RDONLY) {
2230                         /*
2231                          * First of all, the unconditional stuff we have to do
2232                          * to disable replay of the journal when we next remount
2233                          */
2234                         sb->s_flags |= MS_RDONLY;
2235
2236                         /*
2237                          * OK, test if we are remounting a valid rw partition
2238                          * readonly, and if so set the rdonly flag and then
2239                          * mark the partition as valid again.
2240                          */
2241                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2242                             (sbi->s_mount_state & EXT3_VALID_FS))
2243                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2244
2245                         ext3_mark_recovery_complete(sb, es);
2246                 } else {
2247                         __le32 ret;
2248                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2249                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2250                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2251                                        "remount RDWR because of unsupported "
2252                                        "optional features (%x).\n",
2253                                        sb->s_id, le32_to_cpu(ret));
2254                                 err = -EROFS;
2255                                 goto restore_opts;
2256                         }
2257                         /*
2258                          * Mounting a RDONLY partition read-write, so reread
2259                          * and store the current valid flag.  (It may have
2260                          * been changed by e2fsck since we originally mounted
2261                          * the partition.)
2262                          */
2263                         ext3_clear_journal_err(sb, es);
2264                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2265                         if ((ret = ext3_group_extend(sb, es, n_blocks_count))) {
2266                                 err = ret;
2267                                 goto restore_opts;
2268                         }
2269                         if (!ext3_setup_super (sb, es, 0))
2270                                 sb->s_flags &= ~MS_RDONLY;
2271                 }
2272         }
2273 #ifdef CONFIG_QUOTA
2274         /* Release old quota file names */
2275         for (i = 0; i < MAXQUOTAS; i++)
2276                 if (old_opts.s_qf_names[i] &&
2277                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2278                         kfree(old_opts.s_qf_names[i]);
2279 #endif
2280         return 0;
2281 restore_opts:
2282         sb->s_flags = old_sb_flags;
2283         sbi->s_mount_opt = old_opts.s_mount_opt;
2284         sbi->s_resuid = old_opts.s_resuid;
2285         sbi->s_resgid = old_opts.s_resgid;
2286         sbi->s_commit_interval = old_opts.s_commit_interval;
2287 #ifdef CONFIG_QUOTA
2288         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2289         for (i = 0; i < MAXQUOTAS; i++) {
2290                 if (sbi->s_qf_names[i] &&
2291                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2292                         kfree(sbi->s_qf_names[i]);
2293                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2294         }
2295 #endif
2296         return err;
2297 }
2298
2299 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2300 {
2301         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2302         unsigned long overhead;
2303         int i;
2304
2305         if (test_opt (sb, MINIX_DF))
2306                 overhead = 0;
2307         else {
2308                 unsigned long ngroups;
2309                 ngroups = EXT3_SB(sb)->s_groups_count;
2310                 smp_rmb();
2311
2312                 /*
2313                  * Compute the overhead (FS structures)
2314                  */
2315
2316                 /*
2317                  * All of the blocks before first_data_block are
2318                  * overhead
2319                  */
2320                 overhead = le32_to_cpu(es->s_first_data_block);
2321
2322                 /*
2323                  * Add the overhead attributed to the superblock and
2324                  * block group descriptors.  If the sparse superblocks
2325                  * feature is turned on, then not all groups have this.
2326                  */
2327                 for (i = 0; i < ngroups; i++) {
2328                         overhead += ext3_bg_has_super(sb, i) +
2329                                 ext3_bg_num_gdb(sb, i);
2330                         cond_resched();
2331                 }
2332
2333                 /*
2334                  * Every block group has an inode bitmap, a block
2335                  * bitmap, and an inode table.
2336                  */
2337                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2338         }
2339
2340         buf->f_type = EXT3_SUPER_MAGIC;
2341         buf->f_bsize = sb->s_blocksize;
2342         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2343         buf->f_bfree = ext3_count_free_blocks (sb);
2344         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2345         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2346                 buf->f_bavail = 0;
2347         buf->f_files = le32_to_cpu(es->s_inodes_count);
2348         buf->f_ffree = ext3_count_free_inodes (sb);
2349         buf->f_namelen = EXT3_NAME_LEN;
2350         return 0;
2351 }
2352
2353 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2354  * is locked for write. Otherwise the are possible deadlocks:
2355  * Process 1                         Process 2
2356  * ext3_create()                     quota_sync()
2357  *   journal_start()                   write_dquot()
2358  *   DQUOT_INIT()                        down(dqio_sem)
2359  *     down(dqio_sem)                    journal_start()
2360  *
2361  */
2362
2363 #ifdef CONFIG_QUOTA
2364
2365 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2366 {
2367         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2368 }
2369
2370 static int ext3_dquot_initialize(struct inode *inode, int type)
2371 {
2372         handle_t *handle;
2373         int ret, err;
2374
2375         /* We may create quota structure so we need to reserve enough blocks */
2376         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2377         if (IS_ERR(handle))
2378                 return PTR_ERR(handle);
2379         ret = dquot_initialize(inode, type);
2380         err = ext3_journal_stop(handle);
2381         if (!ret)
2382                 ret = err;
2383         return ret;
2384 }
2385
2386 static int ext3_dquot_drop(struct inode *inode)
2387 {
2388         handle_t *handle;
2389         int ret, err;
2390
2391         /* We may delete quota structure so we need to reserve enough blocks */
2392         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2393         if (IS_ERR(handle))
2394                 return PTR_ERR(handle);
2395         ret = dquot_drop(inode);
2396         err = ext3_journal_stop(handle);
2397         if (!ret)
2398                 ret = err;
2399         return ret;
2400 }
2401
2402 static int ext3_write_dquot(struct dquot *dquot)
2403 {
2404         int ret, err;
2405         handle_t *handle;
2406         struct inode *inode;
2407
2408         inode = dquot_to_inode(dquot);
2409         handle = ext3_journal_start(inode,
2410                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2411         if (IS_ERR(handle))
2412                 return PTR_ERR(handle);
2413         ret = dquot_commit(dquot);
2414         err = ext3_journal_stop(handle);
2415         if (!ret)
2416                 ret = err;
2417         return ret;
2418 }
2419
2420 static int ext3_acquire_dquot(struct dquot *dquot)
2421 {
2422         int ret, err;
2423         handle_t *handle;
2424
2425         handle = ext3_journal_start(dquot_to_inode(dquot),
2426                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2427         if (IS_ERR(handle))
2428                 return PTR_ERR(handle);
2429         ret = dquot_acquire(dquot);
2430         err = ext3_journal_stop(handle);
2431         if (!ret)
2432                 ret = err;
2433         return ret;
2434 }
2435
2436 static int ext3_release_dquot(struct dquot *dquot)
2437 {
2438         int ret, err;
2439         handle_t *handle;
2440
2441         handle = ext3_journal_start(dquot_to_inode(dquot),
2442                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2443         if (IS_ERR(handle))
2444                 return PTR_ERR(handle);
2445         ret = dquot_release(dquot);
2446         err = ext3_journal_stop(handle);
2447         if (!ret)
2448                 ret = err;
2449         return ret;
2450 }
2451
2452 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2453 {
2454         /* Are we journalling quotas? */
2455         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2456             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2457                 dquot_mark_dquot_dirty(dquot);
2458                 return ext3_write_dquot(dquot);
2459         } else {
2460                 return dquot_mark_dquot_dirty(dquot);
2461         }
2462 }
2463
2464 static int ext3_write_info(struct super_block *sb, int type)
2465 {
2466         int ret, err;
2467         handle_t *handle;
2468
2469         /* Data block + inode block */
2470         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2471         if (IS_ERR(handle))
2472                 return PTR_ERR(handle);
2473         ret = dquot_commit_info(sb, type);
2474         err = ext3_journal_stop(handle);
2475         if (!ret)
2476                 ret = err;
2477         return ret;
2478 }
2479
2480 /*
2481  * Turn on quotas during mount time - we need to find
2482  * the quota file and such...
2483  */
2484 static int ext3_quota_on_mount(struct super_block *sb, int type)
2485 {
2486         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2487                         EXT3_SB(sb)->s_jquota_fmt, type);
2488 }
2489
2490 /*
2491  * Standard function to be called on quota_on
2492  */
2493 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2494                          char *path)
2495 {
2496         int err;
2497         struct nameidata nd;
2498
2499         if (!test_opt(sb, QUOTA))
2500                 return -EINVAL;
2501         /* Not journalling quota? */
2502         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2503             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2504                 return vfs_quota_on(sb, type, format_id, path);
2505         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2506         if (err)
2507                 return err;
2508         /* Quotafile not on the same filesystem? */
2509         if (nd.mnt->mnt_sb != sb) {
2510                 path_release(&nd);
2511                 return -EXDEV;
2512         }
2513         /* Quotafile not of fs root? */
2514         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2515                 printk(KERN_WARNING
2516                         "EXT3-fs: Quota file not on filesystem root. "
2517                         "Journalled quota will not work.\n");
2518         path_release(&nd);
2519         return vfs_quota_on(sb, type, format_id, path);
2520 }
2521
2522 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2523  * acquiring the locks... As quota files are never truncated and quota code
2524  * itself serializes the operations (and noone else should touch the files)
2525  * we don't have to be afraid of races */
2526 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2527                                size_t len, loff_t off)
2528 {
2529         struct inode *inode = sb_dqopt(sb)->files[type];
2530         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2531         int err = 0;
2532         int offset = off & (sb->s_blocksize - 1);
2533         int tocopy;
2534         size_t toread;
2535         struct buffer_head *bh;
2536         loff_t i_size = i_size_read(inode);
2537
2538         if (off > i_size)
2539                 return 0;
2540         if (off+len > i_size)
2541                 len = i_size-off;
2542         toread = len;
2543         while (toread > 0) {
2544                 tocopy = sb->s_blocksize - offset < toread ?
2545                                 sb->s_blocksize - offset : toread;
2546                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2547                 if (err)
2548                         return err;
2549                 if (!bh)        /* A hole? */
2550                         memset(data, 0, tocopy);
2551                 else
2552                         memcpy(data, bh->b_data+offset, tocopy);
2553                 brelse(bh);
2554                 offset = 0;
2555                 toread -= tocopy;
2556                 data += tocopy;
2557                 blk++;
2558         }
2559         return len;
2560 }
2561
2562 /* Write to quotafile (we know the transaction is already started and has
2563  * enough credits) */
2564 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2565                                 const char *data, size_t len, loff_t off)
2566 {
2567         struct inode *inode = sb_dqopt(sb)->files[type];
2568         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2569         int err = 0;
2570         int offset = off & (sb->s_blocksize - 1);
2571         int tocopy;
2572         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2573         size_t towrite = len;
2574         struct buffer_head *bh;
2575         handle_t *handle = journal_current_handle();
2576
2577         down(&inode->i_sem);
2578         while (towrite > 0) {
2579                 tocopy = sb->s_blocksize - offset < towrite ?
2580                                 sb->s_blocksize - offset : towrite;
2581                 bh = ext3_bread(handle, inode, blk, 1, &err);
2582                 if (!bh)
2583                         goto out;
2584                 if (journal_quota) {
2585                         err = ext3_journal_get_write_access(handle, bh);
2586                         if (err) {
2587                                 brelse(bh);
2588                                 goto out;
2589                         }
2590                 }
2591                 lock_buffer(bh);
2592                 memcpy(bh->b_data+offset, data, tocopy);
2593                 flush_dcache_page(bh->b_page);
2594                 unlock_buffer(bh);
2595                 if (journal_quota)
2596                         err = ext3_journal_dirty_metadata(handle, bh);
2597                 else {
2598                         /* Always do at least ordered writes for quotas */
2599                         err = ext3_journal_dirty_data(handle, bh);
2600                         mark_buffer_dirty(bh);
2601                 }
2602                 brelse(bh);
2603                 if (err)
2604                         goto out;
2605                 offset = 0;
2606                 towrite -= tocopy;
2607                 data += tocopy;
2608                 blk++;
2609         }
2610 out:
2611         if (len == towrite)
2612                 return err;
2613         if (inode->i_size < off+len-towrite) {
2614                 i_size_write(inode, off+len-towrite);
2615                 EXT3_I(inode)->i_disksize = inode->i_size;
2616         }
2617         inode->i_version++;
2618         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2619         ext3_mark_inode_dirty(handle, inode);
2620         up(&inode->i_sem);
2621         return len - towrite;
2622 }
2623
2624 #endif
2625
2626 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2627         int flags, const char *dev_name, void *data)
2628 {
2629         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2630 }
2631
2632 static struct file_system_type ext3_fs_type = {
2633         .owner          = THIS_MODULE,
2634         .name           = "ext3",
2635         .get_sb         = ext3_get_sb,
2636         .kill_sb        = kill_block_super,
2637         .fs_flags       = FS_REQUIRES_DEV,
2638 };
2639
2640 static int __init init_ext3_fs(void)
2641 {
2642         int err = init_ext3_xattr();
2643         if (err)
2644                 return err;
2645         err = init_inodecache();
2646         if (err)
2647                 goto out1;
2648         err = register_filesystem(&ext3_fs_type);
2649         if (err)
2650                 goto out;
2651         return 0;
2652 out:
2653         destroy_inodecache();
2654 out1:
2655         exit_ext3_xattr();
2656         return err;
2657 }
2658
2659 static void __exit exit_ext3_fs(void)
2660 {
2661         unregister_filesystem(&ext3_fs_type);
2662         destroy_inodecache();
2663         exit_ext3_xattr();
2664 }
2665
2666 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2667 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2668 MODULE_LICENSE("GPL");
2669 module_init(init_ext3_fs)
2670 module_exit(exit_ext3_fs)