Merge branch 'upstream' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik/misc-2.6
[linux-2.6] / fs / ext3 / super.c
1 /*
2  *  linux/fs/ext3/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/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 ext3_sb_info *sbi = EXT3_SB(vfs->mnt_sb);
516
517         if (sbi->s_mount_opt & EXT3_MOUNT_JOURNAL_DATA)
518                 seq_puts(seq, ",data=journal");
519
520         if (sbi->s_mount_opt & EXT3_MOUNT_ORDERED_DATA)
521                 seq_puts(seq, ",data=ordered");
522
523         if (sbi->s_mount_opt & EXT3_MOUNT_WRITEBACK_DATA)
524                 seq_puts(seq, ",data=writeback");
525
526 #if defined(CONFIG_QUOTA)
527         if (sbi->s_jquota_fmt)
528                 seq_printf(seq, ",jqfmt=%s",
529                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
530
531         if (sbi->s_qf_names[USRQUOTA])
532                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
533
534         if (sbi->s_qf_names[GRPQUOTA])
535                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
536
537         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
538                 seq_puts(seq, ",usrquota");
539
540         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
541                 seq_puts(seq, ",grpquota");
542 #endif
543
544         return 0;
545 }
546
547 #ifdef CONFIG_QUOTA
548 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
549 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
550
551 static int ext3_dquot_initialize(struct inode *inode, int type);
552 static int ext3_dquot_drop(struct inode *inode);
553 static int ext3_write_dquot(struct dquot *dquot);
554 static int ext3_acquire_dquot(struct dquot *dquot);
555 static int ext3_release_dquot(struct dquot *dquot);
556 static int ext3_mark_dquot_dirty(struct dquot *dquot);
557 static int ext3_write_info(struct super_block *sb, int type);
558 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
559 static int ext3_quota_on_mount(struct super_block *sb, int type);
560 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
561                                size_t len, loff_t off);
562 static ssize_t ext3_quota_write(struct super_block *sb, int type,
563                                 const char *data, size_t len, loff_t off);
564
565 static struct dquot_operations ext3_quota_operations = {
566         .initialize     = ext3_dquot_initialize,
567         .drop           = ext3_dquot_drop,
568         .alloc_space    = dquot_alloc_space,
569         .alloc_inode    = dquot_alloc_inode,
570         .free_space     = dquot_free_space,
571         .free_inode     = dquot_free_inode,
572         .transfer       = dquot_transfer,
573         .write_dquot    = ext3_write_dquot,
574         .acquire_dquot  = ext3_acquire_dquot,
575         .release_dquot  = ext3_release_dquot,
576         .mark_dirty     = ext3_mark_dquot_dirty,
577         .write_info     = ext3_write_info
578 };
579
580 static struct quotactl_ops ext3_qctl_operations = {
581         .quota_on       = ext3_quota_on,
582         .quota_off      = vfs_quota_off,
583         .quota_sync     = vfs_quota_sync,
584         .get_info       = vfs_get_dqinfo,
585         .set_info       = vfs_set_dqinfo,
586         .get_dqblk      = vfs_get_dqblk,
587         .set_dqblk      = vfs_set_dqblk
588 };
589 #endif
590
591 static struct super_operations ext3_sops = {
592         .alloc_inode    = ext3_alloc_inode,
593         .destroy_inode  = ext3_destroy_inode,
594         .read_inode     = ext3_read_inode,
595         .write_inode    = ext3_write_inode,
596         .dirty_inode    = ext3_dirty_inode,
597         .delete_inode   = ext3_delete_inode,
598         .put_super      = ext3_put_super,
599         .write_super    = ext3_write_super,
600         .sync_fs        = ext3_sync_fs,
601         .write_super_lockfs = ext3_write_super_lockfs,
602         .unlockfs       = ext3_unlockfs,
603         .statfs         = ext3_statfs,
604         .remount_fs     = ext3_remount,
605         .clear_inode    = ext3_clear_inode,
606         .show_options   = ext3_show_options,
607 #ifdef CONFIG_QUOTA
608         .quota_read     = ext3_quota_read,
609         .quota_write    = ext3_quota_write,
610 #endif
611 };
612
613 struct dentry *ext3_get_parent(struct dentry *child);
614 static struct export_operations ext3_export_ops = {
615         .get_parent = ext3_get_parent,
616 };
617
618 enum {
619         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
620         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
621         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
622         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
623         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh,
624         Opt_commit, Opt_journal_update, Opt_journal_inum,
625         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
626         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
627         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
628         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
629         Opt_grpquota
630 };
631
632 static match_table_t tokens = {
633         {Opt_bsd_df, "bsddf"},
634         {Opt_minix_df, "minixdf"},
635         {Opt_grpid, "grpid"},
636         {Opt_grpid, "bsdgroups"},
637         {Opt_nogrpid, "nogrpid"},
638         {Opt_nogrpid, "sysvgroups"},
639         {Opt_resgid, "resgid=%u"},
640         {Opt_resuid, "resuid=%u"},
641         {Opt_sb, "sb=%u"},
642         {Opt_err_cont, "errors=continue"},
643         {Opt_err_panic, "errors=panic"},
644         {Opt_err_ro, "errors=remount-ro"},
645         {Opt_nouid32, "nouid32"},
646         {Opt_nocheck, "nocheck"},
647         {Opt_nocheck, "check=none"},
648         {Opt_check, "check"},
649         {Opt_debug, "debug"},
650         {Opt_oldalloc, "oldalloc"},
651         {Opt_orlov, "orlov"},
652         {Opt_user_xattr, "user_xattr"},
653         {Opt_nouser_xattr, "nouser_xattr"},
654         {Opt_acl, "acl"},
655         {Opt_noacl, "noacl"},
656         {Opt_reservation, "reservation"},
657         {Opt_noreservation, "noreservation"},
658         {Opt_noload, "noload"},
659         {Opt_nobh, "nobh"},
660         {Opt_commit, "commit=%u"},
661         {Opt_journal_update, "journal=update"},
662         {Opt_journal_inum, "journal=%u"},
663         {Opt_abort, "abort"},
664         {Opt_data_journal, "data=journal"},
665         {Opt_data_ordered, "data=ordered"},
666         {Opt_data_writeback, "data=writeback"},
667         {Opt_offusrjquota, "usrjquota="},
668         {Opt_usrjquota, "usrjquota=%s"},
669         {Opt_offgrpjquota, "grpjquota="},
670         {Opt_grpjquota, "grpjquota=%s"},
671         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
672         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
673         {Opt_grpquota, "grpquota"},
674         {Opt_noquota, "noquota"},
675         {Opt_quota, "quota"},
676         {Opt_usrquota, "usrquota"},
677         {Opt_barrier, "barrier=%u"},
678         {Opt_err, NULL},
679         {Opt_resize, "resize"},
680 };
681
682 static unsigned long get_sb_block(void **data)
683 {
684         unsigned long   sb_block;
685         char            *options = (char *) *data;
686
687         if (!options || strncmp(options, "sb=", 3) != 0)
688                 return 1;       /* Default location */
689         options += 3;
690         sb_block = simple_strtoul(options, &options, 0);
691         if (*options && *options != ',') {
692                 printk("EXT3-fs: Invalid sb specification: %s\n",
693                        (char *) *data);
694                 return 1;
695         }
696         if (*options == ',')
697                 options++;
698         *data = (void *) options;
699         return sb_block;
700 }
701
702 static int parse_options (char * options, struct super_block *sb,
703                           unsigned long * inum, unsigned long *n_blocks_count, int is_remount)
704 {
705         struct ext3_sb_info *sbi = EXT3_SB(sb);
706         char * p;
707         substring_t args[MAX_OPT_ARGS];
708         int data_opt = 0;
709         int option;
710 #ifdef CONFIG_QUOTA
711         int qtype;
712         char *qname;
713 #endif
714
715         if (!options)
716                 return 1;
717
718         while ((p = strsep (&options, ",")) != NULL) {
719                 int token;
720                 if (!*p)
721                         continue;
722
723                 token = match_token(p, tokens, args);
724                 switch (token) {
725                 case Opt_bsd_df:
726                         clear_opt (sbi->s_mount_opt, MINIX_DF);
727                         break;
728                 case Opt_minix_df:
729                         set_opt (sbi->s_mount_opt, MINIX_DF);
730                         break;
731                 case Opt_grpid:
732                         set_opt (sbi->s_mount_opt, GRPID);
733                         break;
734                 case Opt_nogrpid:
735                         clear_opt (sbi->s_mount_opt, GRPID);
736                         break;
737                 case Opt_resuid:
738                         if (match_int(&args[0], &option))
739                                 return 0;
740                         sbi->s_resuid = option;
741                         break;
742                 case Opt_resgid:
743                         if (match_int(&args[0], &option))
744                                 return 0;
745                         sbi->s_resgid = option;
746                         break;
747                 case Opt_sb:
748                         /* handled by get_sb_block() instead of here */
749                         /* *sb_block = match_int(&args[0]); */
750                         break;
751                 case Opt_err_panic:
752                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
753                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
754                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
755                         break;
756                 case Opt_err_ro:
757                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
758                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
759                         set_opt (sbi->s_mount_opt, ERRORS_RO);
760                         break;
761                 case Opt_err_cont:
762                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
763                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
764                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
765                         break;
766                 case Opt_nouid32:
767                         set_opt (sbi->s_mount_opt, NO_UID32);
768                         break;
769                 case Opt_check:
770 #ifdef CONFIG_EXT3_CHECK
771                         set_opt (sbi->s_mount_opt, CHECK);
772 #else
773                         printk(KERN_ERR
774                                "EXT3 Check option not supported\n");
775 #endif
776                         break;
777                 case Opt_nocheck:
778                         clear_opt (sbi->s_mount_opt, CHECK);
779                         break;
780                 case Opt_debug:
781                         set_opt (sbi->s_mount_opt, DEBUG);
782                         break;
783                 case Opt_oldalloc:
784                         set_opt (sbi->s_mount_opt, OLDALLOC);
785                         break;
786                 case Opt_orlov:
787                         clear_opt (sbi->s_mount_opt, OLDALLOC);
788                         break;
789 #ifdef CONFIG_EXT3_FS_XATTR
790                 case Opt_user_xattr:
791                         set_opt (sbi->s_mount_opt, XATTR_USER);
792                         break;
793                 case Opt_nouser_xattr:
794                         clear_opt (sbi->s_mount_opt, XATTR_USER);
795                         break;
796 #else
797                 case Opt_user_xattr:
798                 case Opt_nouser_xattr:
799                         printk("EXT3 (no)user_xattr options not supported\n");
800                         break;
801 #endif
802 #ifdef CONFIG_EXT3_FS_POSIX_ACL
803                 case Opt_acl:
804                         set_opt(sbi->s_mount_opt, POSIX_ACL);
805                         break;
806                 case Opt_noacl:
807                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
808                         break;
809 #else
810                 case Opt_acl:
811                 case Opt_noacl:
812                         printk("EXT3 (no)acl options not supported\n");
813                         break;
814 #endif
815                 case Opt_reservation:
816                         set_opt(sbi->s_mount_opt, RESERVATION);
817                         break;
818                 case Opt_noreservation:
819                         clear_opt(sbi->s_mount_opt, RESERVATION);
820                         break;
821                 case Opt_journal_update:
822                         /* @@@ FIXME */
823                         /* Eventually we will want to be able to create
824                            a journal file here.  For now, only allow the
825                            user to specify an existing inode to be the
826                            journal file. */
827                         if (is_remount) {
828                                 printk(KERN_ERR "EXT3-fs: cannot specify "
829                                        "journal on remount\n");
830                                 return 0;
831                         }
832                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
833                         break;
834                 case Opt_journal_inum:
835                         if (is_remount) {
836                                 printk(KERN_ERR "EXT3-fs: cannot specify "
837                                        "journal on remount\n");
838                                 return 0;
839                         }
840                         if (match_int(&args[0], &option))
841                                 return 0;
842                         *inum = option;
843                         break;
844                 case Opt_noload:
845                         set_opt (sbi->s_mount_opt, NOLOAD);
846                         break;
847                 case Opt_commit:
848                         if (match_int(&args[0], &option))
849                                 return 0;
850                         if (option < 0)
851                                 return 0;
852                         if (option == 0)
853                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
854                         sbi->s_commit_interval = HZ * option;
855                         break;
856                 case Opt_data_journal:
857                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
858                         goto datacheck;
859                 case Opt_data_ordered:
860                         data_opt = EXT3_MOUNT_ORDERED_DATA;
861                         goto datacheck;
862                 case Opt_data_writeback:
863                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
864                 datacheck:
865                         if (is_remount) {
866                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
867                                                 != data_opt) {
868                                         printk(KERN_ERR
869                                                 "EXT3-fs: cannot change data "
870                                                 "mode on remount\n");
871                                         return 0;
872                                 }
873                         } else {
874                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
875                                 sbi->s_mount_opt |= data_opt;
876                         }
877                         break;
878 #ifdef CONFIG_QUOTA
879                 case Opt_usrjquota:
880                         qtype = USRQUOTA;
881                         goto set_qf_name;
882                 case Opt_grpjquota:
883                         qtype = GRPQUOTA;
884 set_qf_name:
885                         if (sb_any_quota_enabled(sb)) {
886                                 printk(KERN_ERR
887                                         "EXT3-fs: Cannot change journalled "
888                                         "quota options when quota turned on.\n");
889                                 return 0;
890                         }
891                         qname = match_strdup(&args[0]);
892                         if (!qname) {
893                                 printk(KERN_ERR
894                                         "EXT3-fs: not enough memory for "
895                                         "storing quotafile name.\n");
896                                 return 0;
897                         }
898                         if (sbi->s_qf_names[qtype] &&
899                             strcmp(sbi->s_qf_names[qtype], qname)) {
900                                 printk(KERN_ERR
901                                         "EXT3-fs: %s quota file already "
902                                         "specified.\n", QTYPE2NAME(qtype));
903                                 kfree(qname);
904                                 return 0;
905                         }
906                         sbi->s_qf_names[qtype] = qname;
907                         if (strchr(sbi->s_qf_names[qtype], '/')) {
908                                 printk(KERN_ERR
909                                         "EXT3-fs: quotafile must be on "
910                                         "filesystem root.\n");
911                                 kfree(sbi->s_qf_names[qtype]);
912                                 sbi->s_qf_names[qtype] = NULL;
913                                 return 0;
914                         }
915                         set_opt(sbi->s_mount_opt, QUOTA);
916                         break;
917                 case Opt_offusrjquota:
918                         qtype = USRQUOTA;
919                         goto clear_qf_name;
920                 case Opt_offgrpjquota:
921                         qtype = GRPQUOTA;
922 clear_qf_name:
923                         if (sb_any_quota_enabled(sb)) {
924                                 printk(KERN_ERR "EXT3-fs: Cannot change "
925                                         "journalled quota options when "
926                                         "quota turned on.\n");
927                                 return 0;
928                         }
929                         /*
930                          * The space will be released later when all options
931                          * are confirmed to be correct
932                          */
933                         sbi->s_qf_names[qtype] = NULL;
934                         break;
935                 case Opt_jqfmt_vfsold:
936                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
937                         break;
938                 case Opt_jqfmt_vfsv0:
939                         sbi->s_jquota_fmt = QFMT_VFS_V0;
940                         break;
941                 case Opt_quota:
942                 case Opt_usrquota:
943                         set_opt(sbi->s_mount_opt, QUOTA);
944                         set_opt(sbi->s_mount_opt, USRQUOTA);
945                         break;
946                 case Opt_grpquota:
947                         set_opt(sbi->s_mount_opt, QUOTA);
948                         set_opt(sbi->s_mount_opt, GRPQUOTA);
949                         break;
950                 case Opt_noquota:
951                         if (sb_any_quota_enabled(sb)) {
952                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
953                                         "options when quota turned on.\n");
954                                 return 0;
955                         }
956                         clear_opt(sbi->s_mount_opt, QUOTA);
957                         clear_opt(sbi->s_mount_opt, USRQUOTA);
958                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
959                         break;
960 #else
961                 case Opt_quota:
962                 case Opt_usrquota:
963                 case Opt_grpquota:
964                 case Opt_usrjquota:
965                 case Opt_grpjquota:
966                 case Opt_offusrjquota:
967                 case Opt_offgrpjquota:
968                 case Opt_jqfmt_vfsold:
969                 case Opt_jqfmt_vfsv0:
970                         printk(KERN_ERR
971                                 "EXT3-fs: journalled quota options not "
972                                 "supported.\n");
973                         break;
974                 case Opt_noquota:
975                         break;
976 #endif
977                 case Opt_abort:
978                         set_opt(sbi->s_mount_opt, ABORT);
979                         break;
980                 case Opt_barrier:
981                         if (match_int(&args[0], &option))
982                                 return 0;
983                         if (option)
984                                 set_opt(sbi->s_mount_opt, BARRIER);
985                         else
986                                 clear_opt(sbi->s_mount_opt, BARRIER);
987                         break;
988                 case Opt_ignore:
989                         break;
990                 case Opt_resize:
991                         if (!is_remount) {
992                                 printk("EXT3-fs: resize option only available "
993                                         "for remount\n");
994                                 return 0;
995                         }
996                         if (match_int(&args[0], &option) != 0)
997                                 return 0;
998                         *n_blocks_count = option;
999                         break;
1000                 case Opt_nobh:
1001                         set_opt(sbi->s_mount_opt, NOBH);
1002                         break;
1003                 default:
1004                         printk (KERN_ERR
1005                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1006                                 "or missing value\n", p);
1007                         return 0;
1008                 }
1009         }
1010 #ifdef CONFIG_QUOTA
1011         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1012                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1013                      sbi->s_qf_names[USRQUOTA])
1014                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1015
1016                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1017                      sbi->s_qf_names[GRPQUOTA])
1018                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1019
1020                 if ((sbi->s_qf_names[USRQUOTA] &&
1021                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1022                     (sbi->s_qf_names[GRPQUOTA] &&
1023                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1024                         printk(KERN_ERR "EXT3-fs: old and new quota "
1025                                         "format mixing.\n");
1026                         return 0;
1027                 }
1028
1029                 if (!sbi->s_jquota_fmt) {
1030                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1031                                         "not specified.\n");
1032                         return 0;
1033                 }
1034         } else {
1035                 if (sbi->s_jquota_fmt) {
1036                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1037                                         "specified with no journalling "
1038                                         "enabled.\n");
1039                         return 0;
1040                 }
1041         }
1042 #endif
1043         return 1;
1044 }
1045
1046 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1047                             int read_only)
1048 {
1049         struct ext3_sb_info *sbi = EXT3_SB(sb);
1050         int res = 0;
1051
1052         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1053                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1054                         "forcing read-only mode\n");
1055                 res = MS_RDONLY;
1056         }
1057         if (read_only)
1058                 return res;
1059         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1060                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1061                         "running e2fsck is recommended\n");
1062         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1063                 printk (KERN_WARNING
1064                         "EXT3-fs warning: mounting fs with errors, "
1065                         "running e2fsck is recommended\n");
1066         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1067                  le16_to_cpu(es->s_mnt_count) >=
1068                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1069                 printk (KERN_WARNING
1070                         "EXT3-fs warning: maximal mount count reached, "
1071                         "running e2fsck is recommended\n");
1072         else if (le32_to_cpu(es->s_checkinterval) &&
1073                 (le32_to_cpu(es->s_lastcheck) +
1074                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1075                 printk (KERN_WARNING
1076                         "EXT3-fs warning: checktime reached, "
1077                         "running e2fsck is recommended\n");
1078 #if 0
1079                 /* @@@ We _will_ want to clear the valid bit if we find
1080                    inconsistencies, to force a fsck at reboot.  But for
1081                    a plain journaled filesystem we can keep it set as
1082                    valid forever! :) */
1083         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1084 #endif
1085         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1086                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1087         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1088         es->s_mtime = cpu_to_le32(get_seconds());
1089         ext3_update_dynamic_rev(sb);
1090         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1091
1092         ext3_commit_super(sb, es, 1);
1093         if (test_opt(sb, DEBUG))
1094                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1095                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1096                         sb->s_blocksize,
1097                         sbi->s_groups_count,
1098                         EXT3_BLOCKS_PER_GROUP(sb),
1099                         EXT3_INODES_PER_GROUP(sb),
1100                         sbi->s_mount_opt);
1101
1102         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1103         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1104                 char b[BDEVNAME_SIZE];
1105
1106                 printk("external journal on %s\n",
1107                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1108         } else {
1109                 printk("internal journal\n");
1110         }
1111 #ifdef CONFIG_EXT3_CHECK
1112         if (test_opt (sb, CHECK)) {
1113                 ext3_check_blocks_bitmap (sb);
1114                 ext3_check_inodes_bitmap (sb);
1115         }
1116 #endif
1117         return res;
1118 }
1119
1120 /* Called at mount-time, super-block is locked */
1121 static int ext3_check_descriptors (struct super_block * sb)
1122 {
1123         struct ext3_sb_info *sbi = EXT3_SB(sb);
1124         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1125         struct ext3_group_desc * gdp = NULL;
1126         int desc_block = 0;
1127         int i;
1128
1129         ext3_debug ("Checking group descriptors");
1130
1131         for (i = 0; i < sbi->s_groups_count; i++)
1132         {
1133                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1134                         gdp = (struct ext3_group_desc *)
1135                                         sbi->s_group_desc[desc_block++]->b_data;
1136                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1137                     le32_to_cpu(gdp->bg_block_bitmap) >=
1138                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1139                 {
1140                         ext3_error (sb, "ext3_check_descriptors",
1141                                     "Block bitmap for group %d"
1142                                     " not in group (block %lu)!",
1143                                     i, (unsigned long)
1144                                         le32_to_cpu(gdp->bg_block_bitmap));
1145                         return 0;
1146                 }
1147                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1148                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1149                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1150                 {
1151                         ext3_error (sb, "ext3_check_descriptors",
1152                                     "Inode bitmap for group %d"
1153                                     " not in group (block %lu)!",
1154                                     i, (unsigned long)
1155                                         le32_to_cpu(gdp->bg_inode_bitmap));
1156                         return 0;
1157                 }
1158                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1159                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1160                     block + EXT3_BLOCKS_PER_GROUP(sb))
1161                 {
1162                         ext3_error (sb, "ext3_check_descriptors",
1163                                     "Inode table for group %d"
1164                                     " not in group (block %lu)!",
1165                                     i, (unsigned long)
1166                                         le32_to_cpu(gdp->bg_inode_table));
1167                         return 0;
1168                 }
1169                 block += EXT3_BLOCKS_PER_GROUP(sb);
1170                 gdp++;
1171         }
1172
1173         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1174         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1175         return 1;
1176 }
1177
1178
1179 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1180  * the superblock) which were deleted from all directories, but held open by
1181  * a process at the time of a crash.  We walk the list and try to delete these
1182  * inodes at recovery time (only with a read-write filesystem).
1183  *
1184  * In order to keep the orphan inode chain consistent during traversal (in
1185  * case of crash during recovery), we link each inode into the superblock
1186  * orphan list_head and handle it the same way as an inode deletion during
1187  * normal operation (which journals the operations for us).
1188  *
1189  * We only do an iget() and an iput() on each inode, which is very safe if we
1190  * accidentally point at an in-use or already deleted inode.  The worst that
1191  * can happen in this case is that we get a "bit already cleared" message from
1192  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1193  * e2fsck was run on this filesystem, and it must have already done the orphan
1194  * inode cleanup for us, so we can safely abort without any further action.
1195  */
1196 static void ext3_orphan_cleanup (struct super_block * sb,
1197                                  struct ext3_super_block * es)
1198 {
1199         unsigned int s_flags = sb->s_flags;
1200         int nr_orphans = 0, nr_truncates = 0;
1201 #ifdef CONFIG_QUOTA
1202         int i;
1203 #endif
1204         if (!es->s_last_orphan) {
1205                 jbd_debug(4, "no orphan inodes to clean up\n");
1206                 return;
1207         }
1208
1209         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1210                 if (es->s_last_orphan)
1211                         jbd_debug(1, "Errors on filesystem, "
1212                                   "clearing orphan list.\n");
1213                 es->s_last_orphan = 0;
1214                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1215                 return;
1216         }
1217
1218         if (s_flags & MS_RDONLY) {
1219                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1220                        sb->s_id);
1221                 sb->s_flags &= ~MS_RDONLY;
1222         }
1223 #ifdef CONFIG_QUOTA
1224         /* Needed for iput() to work correctly and not trash data */
1225         sb->s_flags |= MS_ACTIVE;
1226         /* Turn on quotas so that they are updated correctly */
1227         for (i = 0; i < MAXQUOTAS; i++) {
1228                 if (EXT3_SB(sb)->s_qf_names[i]) {
1229                         int ret = ext3_quota_on_mount(sb, i);
1230                         if (ret < 0)
1231                                 printk(KERN_ERR
1232                                         "EXT3-fs: Cannot turn on journalled "
1233                                         "quota: error %d\n", ret);
1234                 }
1235         }
1236 #endif
1237
1238         while (es->s_last_orphan) {
1239                 struct inode *inode;
1240
1241                 if (!(inode =
1242                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1243                         es->s_last_orphan = 0;
1244                         break;
1245                 }
1246
1247                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1248                 DQUOT_INIT(inode);
1249                 if (inode->i_nlink) {
1250                         printk(KERN_DEBUG
1251                                 "%s: truncating inode %ld to %Ld bytes\n",
1252                                 __FUNCTION__, inode->i_ino, inode->i_size);
1253                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1254                                   inode->i_ino, inode->i_size);
1255                         ext3_truncate(inode);
1256                         nr_truncates++;
1257                 } else {
1258                         printk(KERN_DEBUG
1259                                 "%s: deleting unreferenced inode %ld\n",
1260                                 __FUNCTION__, inode->i_ino);
1261                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1262                                   inode->i_ino);
1263                         nr_orphans++;
1264                 }
1265                 iput(inode);  /* The delete magic happens here! */
1266         }
1267
1268 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1269
1270         if (nr_orphans)
1271                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1272                        sb->s_id, PLURAL(nr_orphans));
1273         if (nr_truncates)
1274                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1275                        sb->s_id, PLURAL(nr_truncates));
1276 #ifdef CONFIG_QUOTA
1277         /* Turn quotas off */
1278         for (i = 0; i < MAXQUOTAS; i++) {
1279                 if (sb_dqopt(sb)->files[i])
1280                         vfs_quota_off(sb, i);
1281         }
1282 #endif
1283         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1284 }
1285
1286 #define log2(n) ffz(~(n))
1287
1288 /*
1289  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1290  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1291  * We need to be 1 filesystem block less than the 2^32 sector limit.
1292  */
1293 static loff_t ext3_max_size(int bits)
1294 {
1295         loff_t res = EXT3_NDIR_BLOCKS;
1296         /* This constant is calculated to be the largest file size for a
1297          * dense, 4k-blocksize file such that the total number of
1298          * sectors in the file, including data and all indirect blocks,
1299          * does not exceed 2^32. */
1300         const loff_t upper_limit = 0x1ff7fffd000LL;
1301
1302         res += 1LL << (bits-2);
1303         res += 1LL << (2*(bits-2));
1304         res += 1LL << (3*(bits-2));
1305         res <<= bits;
1306         if (res > upper_limit)
1307                 res = upper_limit;
1308         return res;
1309 }
1310
1311 static unsigned long descriptor_loc(struct super_block *sb,
1312                                     unsigned long logic_sb_block,
1313                                     int nr)
1314 {
1315         struct ext3_sb_info *sbi = EXT3_SB(sb);
1316         unsigned long bg, first_data_block, first_meta_bg;
1317         int has_super = 0;
1318
1319         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1320         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1321
1322         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1323             nr < first_meta_bg)
1324                 return (logic_sb_block + nr + 1);
1325         bg = sbi->s_desc_per_block * nr;
1326         if (ext3_bg_has_super(sb, bg))
1327                 has_super = 1;
1328         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1329 }
1330
1331
1332 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1333 {
1334         struct buffer_head * bh;
1335         struct ext3_super_block *es = NULL;
1336         struct ext3_sb_info *sbi;
1337         unsigned long block;
1338         unsigned long sb_block = get_sb_block(&data);
1339         unsigned long logic_sb_block;
1340         unsigned long offset = 0;
1341         unsigned long journal_inum = 0;
1342         unsigned long def_mount_opts;
1343         struct inode *root;
1344         int blocksize;
1345         int hblock;
1346         int db_count;
1347         int i;
1348         int needs_recovery;
1349         __le32 features;
1350
1351         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1352         if (!sbi)
1353                 return -ENOMEM;
1354         sb->s_fs_info = sbi;
1355         memset(sbi, 0, sizeof(*sbi));
1356         sbi->s_mount_opt = 0;
1357         sbi->s_resuid = EXT3_DEF_RESUID;
1358         sbi->s_resgid = EXT3_DEF_RESGID;
1359
1360         unlock_kernel();
1361
1362         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1363         if (!blocksize) {
1364                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1365                 goto out_fail;
1366         }
1367
1368         /*
1369          * The ext3 superblock will not be buffer aligned for other than 1kB
1370          * block sizes.  We need to calculate the offset from buffer start.
1371          */
1372         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1373                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1374                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1375         } else {
1376                 logic_sb_block = sb_block;
1377         }
1378
1379         if (!(bh = sb_bread(sb, logic_sb_block))) {
1380                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1381                 goto out_fail;
1382         }
1383         /*
1384          * Note: s_es must be initialized as soon as possible because
1385          *       some ext3 macro-instructions depend on its value
1386          */
1387         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1388         sbi->s_es = es;
1389         sb->s_magic = le16_to_cpu(es->s_magic);
1390         if (sb->s_magic != EXT3_SUPER_MAGIC)
1391                 goto cantfind_ext3;
1392
1393         /* Set defaults before we parse the mount options */
1394         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1395         if (def_mount_opts & EXT3_DEFM_DEBUG)
1396                 set_opt(sbi->s_mount_opt, DEBUG);
1397         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1398                 set_opt(sbi->s_mount_opt, GRPID);
1399         if (def_mount_opts & EXT3_DEFM_UID16)
1400                 set_opt(sbi->s_mount_opt, NO_UID32);
1401         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1402                 set_opt(sbi->s_mount_opt, XATTR_USER);
1403         if (def_mount_opts & EXT3_DEFM_ACL)
1404                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1405         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1406                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1407         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1408                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1409         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1410                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1411
1412         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1413                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1414         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1415                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1416
1417         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1418         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1419
1420         set_opt(sbi->s_mount_opt, RESERVATION);
1421
1422         if (!parse_options ((char *) data, sb, &journal_inum, NULL, 0))
1423                 goto failed_mount;
1424
1425         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1426                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1427
1428         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1429             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1430              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1431              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1432                 printk(KERN_WARNING 
1433                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1434                        "running e2fsck is recommended\n");
1435         /*
1436          * Check feature flags regardless of the revision level, since we
1437          * previously didn't change the revision level when setting the flags,
1438          * so there is a chance incompat flags are set on a rev 0 filesystem.
1439          */
1440         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1441         if (features) {
1442                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1443                        "unsupported optional features (%x).\n",
1444                        sb->s_id, le32_to_cpu(features));
1445                 goto failed_mount;
1446         }
1447         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1448         if (!(sb->s_flags & MS_RDONLY) && features) {
1449                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1450                        "unsupported optional features (%x).\n",
1451                        sb->s_id, le32_to_cpu(features));
1452                 goto failed_mount;
1453         }
1454         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1455
1456         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1457             blocksize > EXT3_MAX_BLOCK_SIZE) {
1458                 printk(KERN_ERR 
1459                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1460                        blocksize, sb->s_id);
1461                 goto failed_mount;
1462         }
1463
1464         hblock = bdev_hardsect_size(sb->s_bdev);
1465         if (sb->s_blocksize != blocksize) {
1466                 /*
1467                  * Make sure the blocksize for the filesystem is larger
1468                  * than the hardware sectorsize for the machine.
1469                  */
1470                 if (blocksize < hblock) {
1471                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1472                                "device blocksize %d.\n", blocksize, hblock);
1473                         goto failed_mount;
1474                 }
1475
1476                 brelse (bh);
1477                 sb_set_blocksize(sb, blocksize);
1478                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1479                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1480                 bh = sb_bread(sb, logic_sb_block);
1481                 if (!bh) {
1482                         printk(KERN_ERR 
1483                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1484                         goto failed_mount;
1485                 }
1486                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1487                 sbi->s_es = es;
1488                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1489                         printk (KERN_ERR 
1490                                 "EXT3-fs: Magic mismatch, very weird !\n");
1491                         goto failed_mount;
1492                 }
1493         }
1494
1495         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1496
1497         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1498                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1499                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1500         } else {
1501                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1502                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1503                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1504                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1505                     (sbi->s_inode_size > blocksize)) {
1506                         printk (KERN_ERR
1507                                 "EXT3-fs: unsupported inode size: %d\n",
1508                                 sbi->s_inode_size);
1509                         goto failed_mount;
1510                 }
1511         }
1512         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1513                                    le32_to_cpu(es->s_log_frag_size);
1514         if (blocksize != sbi->s_frag_size) {
1515                 printk(KERN_ERR
1516                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1517                        sbi->s_frag_size, blocksize);
1518                 goto failed_mount;
1519         }
1520         sbi->s_frags_per_block = 1;
1521         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1522         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1523         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1524         if (EXT3_INODE_SIZE(sb) == 0)
1525                 goto cantfind_ext3;
1526         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1527         if (sbi->s_inodes_per_block == 0)
1528                 goto cantfind_ext3;
1529         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1530                                         sbi->s_inodes_per_block;
1531         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1532         sbi->s_sbh = bh;
1533         sbi->s_mount_state = le16_to_cpu(es->s_state);
1534         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1535         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1536         for (i=0; i < 4; i++)
1537                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1538         sbi->s_def_hash_version = es->s_def_hash_version;
1539
1540         if (sbi->s_blocks_per_group > blocksize * 8) {
1541                 printk (KERN_ERR
1542                         "EXT3-fs: #blocks per group too big: %lu\n",
1543                         sbi->s_blocks_per_group);
1544                 goto failed_mount;
1545         }
1546         if (sbi->s_frags_per_group > blocksize * 8) {
1547                 printk (KERN_ERR
1548                         "EXT3-fs: #fragments per group too big: %lu\n",
1549                         sbi->s_frags_per_group);
1550                 goto failed_mount;
1551         }
1552         if (sbi->s_inodes_per_group > blocksize * 8) {
1553                 printk (KERN_ERR
1554                         "EXT3-fs: #inodes per group too big: %lu\n",
1555                         sbi->s_inodes_per_group);
1556                 goto failed_mount;
1557         }
1558
1559         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1560                 goto cantfind_ext3;
1561         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1562                                le32_to_cpu(es->s_first_data_block) +
1563                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1564                               EXT3_BLOCKS_PER_GROUP(sb);
1565         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1566                    EXT3_DESC_PER_BLOCK(sb);
1567         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1568                                     GFP_KERNEL);
1569         if (sbi->s_group_desc == NULL) {
1570                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1571                 goto failed_mount;
1572         }
1573
1574         percpu_counter_init(&sbi->s_freeblocks_counter);
1575         percpu_counter_init(&sbi->s_freeinodes_counter);
1576         percpu_counter_init(&sbi->s_dirs_counter);
1577         bgl_lock_init(&sbi->s_blockgroup_lock);
1578
1579         for (i = 0; i < db_count; i++) {
1580                 block = descriptor_loc(sb, logic_sb_block, i);
1581                 sbi->s_group_desc[i] = sb_bread(sb, block);
1582                 if (!sbi->s_group_desc[i]) {
1583                         printk (KERN_ERR "EXT3-fs: "
1584                                 "can't read group descriptor %d\n", i);
1585                         db_count = i;
1586                         goto failed_mount2;
1587                 }
1588         }
1589         if (!ext3_check_descriptors (sb)) {
1590                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1591                 goto failed_mount2;
1592         }
1593         sbi->s_gdb_count = db_count;
1594         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1595         spin_lock_init(&sbi->s_next_gen_lock);
1596         /* per fileystem reservation list head & lock */
1597         spin_lock_init(&sbi->s_rsv_window_lock);
1598         sbi->s_rsv_window_root = RB_ROOT;
1599         /* Add a single, static dummy reservation to the start of the
1600          * reservation window list --- it gives us a placeholder for
1601          * append-at-start-of-list which makes the allocation logic
1602          * _much_ simpler. */
1603         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1604         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1605         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1606         sbi->s_rsv_window_head.rsv_goal_size = 0;
1607         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1608
1609         /*
1610          * set up enough so that it can read an inode
1611          */
1612         sb->s_op = &ext3_sops;
1613         sb->s_export_op = &ext3_export_ops;
1614         sb->s_xattr = ext3_xattr_handlers;
1615 #ifdef CONFIG_QUOTA
1616         sb->s_qcop = &ext3_qctl_operations;
1617         sb->dq_op = &ext3_quota_operations;
1618 #endif
1619         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1620
1621         sb->s_root = NULL;
1622
1623         needs_recovery = (es->s_last_orphan != 0 ||
1624                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1625                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1626
1627         /*
1628          * The first inode we look at is the journal inode.  Don't try
1629          * root first: it may be modified in the journal!
1630          */
1631         if (!test_opt(sb, NOLOAD) &&
1632             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1633                 if (ext3_load_journal(sb, es))
1634                         goto failed_mount2;
1635         } else if (journal_inum) {
1636                 if (ext3_create_journal(sb, es, journal_inum))
1637                         goto failed_mount2;
1638         } else {
1639                 if (!silent)
1640                         printk (KERN_ERR
1641                                 "ext3: No journal on filesystem on %s\n",
1642                                 sb->s_id);
1643                 goto failed_mount2;
1644         }
1645
1646         /* We have now updated the journal if required, so we can
1647          * validate the data journaling mode. */
1648         switch (test_opt(sb, DATA_FLAGS)) {
1649         case 0:
1650                 /* No mode set, assume a default based on the journal
1651                    capabilities: ORDERED_DATA if the journal can
1652                    cope, else JOURNAL_DATA */
1653                 if (journal_check_available_features
1654                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1655                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1656                 else
1657                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1658                 break;
1659
1660         case EXT3_MOUNT_ORDERED_DATA:
1661         case EXT3_MOUNT_WRITEBACK_DATA:
1662                 if (!journal_check_available_features
1663                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1664                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1665                                "requested data journaling mode\n");
1666                         goto failed_mount3;
1667                 }
1668         default:
1669                 break;
1670         }
1671
1672         if (test_opt(sb, NOBH)) {
1673                 if (sb->s_blocksize_bits != PAGE_CACHE_SHIFT) {
1674                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option "
1675                                 "since filesystem blocksize doesn't match "
1676                                 "pagesize\n");
1677                         clear_opt(sbi->s_mount_opt, NOBH);
1678                 }
1679                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1680                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1681                                 "its supported only with writeback mode\n");
1682                         clear_opt(sbi->s_mount_opt, NOBH);
1683                 }
1684         }
1685         /*
1686          * The journal_load will have done any necessary log recovery,
1687          * so we can safely mount the rest of the filesystem now.
1688          */
1689
1690         root = iget(sb, EXT3_ROOT_INO);
1691         sb->s_root = d_alloc_root(root);
1692         if (!sb->s_root) {
1693                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1694                 iput(root);
1695                 goto failed_mount3;
1696         }
1697         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1698                 dput(sb->s_root);
1699                 sb->s_root = NULL;
1700                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1701                 goto failed_mount3;
1702         }
1703
1704         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1705         /*
1706          * akpm: core read_super() calls in here with the superblock locked.
1707          * That deadlocks, because orphan cleanup needs to lock the superblock
1708          * in numerous places.  Here we just pop the lock - it's relatively
1709          * harmless, because we are now ready to accept write_super() requests,
1710          * and aviro says that's the only reason for hanging onto the
1711          * superblock lock.
1712          */
1713         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1714         ext3_orphan_cleanup(sb, es);
1715         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1716         if (needs_recovery)
1717                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1718         ext3_mark_recovery_complete(sb, es);
1719         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1720                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1721                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1722                 "writeback");
1723
1724         percpu_counter_mod(&sbi->s_freeblocks_counter,
1725                 ext3_count_free_blocks(sb));
1726         percpu_counter_mod(&sbi->s_freeinodes_counter,
1727                 ext3_count_free_inodes(sb));
1728         percpu_counter_mod(&sbi->s_dirs_counter,
1729                 ext3_count_dirs(sb));
1730
1731         lock_kernel();
1732         return 0;
1733
1734 cantfind_ext3:
1735         if (!silent)
1736                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1737                        sb->s_id);
1738         goto failed_mount;
1739
1740 failed_mount3:
1741         journal_destroy(sbi->s_journal);
1742 failed_mount2:
1743         for (i = 0; i < db_count; i++)
1744                 brelse(sbi->s_group_desc[i]);
1745         kfree(sbi->s_group_desc);
1746 failed_mount:
1747 #ifdef CONFIG_QUOTA
1748         for (i = 0; i < MAXQUOTAS; i++)
1749                 kfree(sbi->s_qf_names[i]);
1750 #endif
1751         ext3_blkdev_remove(sbi);
1752         brelse(bh);
1753 out_fail:
1754         sb->s_fs_info = NULL;
1755         kfree(sbi);
1756         lock_kernel();
1757         return -EINVAL;
1758 }
1759
1760 /*
1761  * Setup any per-fs journal parameters now.  We'll do this both on
1762  * initial mount, once the journal has been initialised but before we've
1763  * done any recovery; and again on any subsequent remount. 
1764  */
1765 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1766 {
1767         struct ext3_sb_info *sbi = EXT3_SB(sb);
1768
1769         if (sbi->s_commit_interval)
1770                 journal->j_commit_interval = sbi->s_commit_interval;
1771         /* We could also set up an ext3-specific default for the commit
1772          * interval here, but for now we'll just fall back to the jbd
1773          * default. */
1774
1775         spin_lock(&journal->j_state_lock);
1776         if (test_opt(sb, BARRIER))
1777                 journal->j_flags |= JFS_BARRIER;
1778         else
1779                 journal->j_flags &= ~JFS_BARRIER;
1780         spin_unlock(&journal->j_state_lock);
1781 }
1782
1783 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1784 {
1785         struct inode *journal_inode;
1786         journal_t *journal;
1787
1788         /* First, test for the existence of a valid inode on disk.  Bad
1789          * things happen if we iget() an unused inode, as the subsequent
1790          * iput() will try to delete it. */
1791
1792         journal_inode = iget(sb, journal_inum);
1793         if (!journal_inode) {
1794                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1795                 return NULL;
1796         }
1797         if (!journal_inode->i_nlink) {
1798                 make_bad_inode(journal_inode);
1799                 iput(journal_inode);
1800                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1801                 return NULL;
1802         }
1803
1804         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1805                   journal_inode, journal_inode->i_size);
1806         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1807                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1808                 iput(journal_inode);
1809                 return NULL;
1810         }
1811
1812         journal = journal_init_inode(journal_inode);
1813         if (!journal) {
1814                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1815                 iput(journal_inode);
1816                 return NULL;
1817         }
1818         journal->j_private = sb;
1819         ext3_init_journal_params(sb, journal);
1820         return journal;
1821 }
1822
1823 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1824                                        dev_t j_dev)
1825 {
1826         struct buffer_head * bh;
1827         journal_t *journal;
1828         int start;
1829         int len;
1830         int hblock, blocksize;
1831         unsigned long sb_block;
1832         unsigned long offset;
1833         struct ext3_super_block * es;
1834         struct block_device *bdev;
1835
1836         bdev = ext3_blkdev_get(j_dev);
1837         if (bdev == NULL)
1838                 return NULL;
1839
1840         if (bd_claim(bdev, sb)) {
1841                 printk(KERN_ERR
1842                         "EXT3: failed to claim external journal device.\n");
1843                 blkdev_put(bdev);
1844                 return NULL;
1845         }
1846
1847         blocksize = sb->s_blocksize;
1848         hblock = bdev_hardsect_size(bdev);
1849         if (blocksize < hblock) {
1850                 printk(KERN_ERR
1851                         "EXT3-fs: blocksize too small for journal device.\n");
1852                 goto out_bdev;
1853         }
1854
1855         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1856         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1857         set_blocksize(bdev, blocksize);
1858         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1859                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1860                        "external journal\n");
1861                 goto out_bdev;
1862         }
1863
1864         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1865         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1866             !(le32_to_cpu(es->s_feature_incompat) &
1867               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1868                 printk(KERN_ERR "EXT3-fs: external journal has "
1869                                         "bad superblock\n");
1870                 brelse(bh);
1871                 goto out_bdev;
1872         }
1873
1874         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1875                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1876                 brelse(bh);
1877                 goto out_bdev;
1878         }
1879
1880         len = le32_to_cpu(es->s_blocks_count);
1881         start = sb_block + 1;
1882         brelse(bh);     /* we're done with the superblock */
1883
1884         journal = journal_init_dev(bdev, sb->s_bdev,
1885                                         start, len, blocksize);
1886         if (!journal) {
1887                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1888                 goto out_bdev;
1889         }
1890         journal->j_private = sb;
1891         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1892         wait_on_buffer(journal->j_sb_buffer);
1893         if (!buffer_uptodate(journal->j_sb_buffer)) {
1894                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1895                 goto out_journal;
1896         }
1897         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1898                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1899                                         "user (unsupported) - %d\n",
1900                         be32_to_cpu(journal->j_superblock->s_nr_users));
1901                 goto out_journal;
1902         }
1903         EXT3_SB(sb)->journal_bdev = bdev;
1904         ext3_init_journal_params(sb, journal);
1905         return journal;
1906 out_journal:
1907         journal_destroy(journal);
1908 out_bdev:
1909         ext3_blkdev_put(bdev);
1910         return NULL;
1911 }
1912
1913 static int ext3_load_journal(struct super_block * sb,
1914                              struct ext3_super_block * es)
1915 {
1916         journal_t *journal;
1917         int journal_inum = le32_to_cpu(es->s_journal_inum);
1918         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1919         int err = 0;
1920         int really_read_only;
1921
1922         really_read_only = bdev_read_only(sb->s_bdev);
1923
1924         /*
1925          * Are we loading a blank journal or performing recovery after a
1926          * crash?  For recovery, we need to check in advance whether we
1927          * can get read-write access to the device.
1928          */
1929
1930         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1931                 if (sb->s_flags & MS_RDONLY) {
1932                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1933                                         "required on readonly filesystem.\n");
1934                         if (really_read_only) {
1935                                 printk(KERN_ERR "EXT3-fs: write access "
1936                                         "unavailable, cannot proceed.\n");
1937                                 return -EROFS;
1938                         }
1939                         printk (KERN_INFO "EXT3-fs: write access will "
1940                                         "be enabled during recovery.\n");
1941                 }
1942         }
1943
1944         if (journal_inum && journal_dev) {
1945                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1946                        "and inode journals!\n");
1947                 return -EINVAL;
1948         }
1949
1950         if (journal_inum) {
1951                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1952                         return -EINVAL;
1953         } else {
1954                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1955                         return -EINVAL;
1956         }
1957
1958         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1959                 err = journal_update_format(journal);
1960                 if (err)  {
1961                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1962                         journal_destroy(journal);
1963                         return err;
1964                 }
1965         }
1966
1967         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1968                 err = journal_wipe(journal, !really_read_only);
1969         if (!err)
1970                 err = journal_load(journal);
1971
1972         if (err) {
1973                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1974                 journal_destroy(journal);
1975                 return err;
1976         }
1977
1978         EXT3_SB(sb)->s_journal = journal;
1979         ext3_clear_journal_err(sb, es);
1980         return 0;
1981 }
1982
1983 static int ext3_create_journal(struct super_block * sb,
1984                                struct ext3_super_block * es,
1985                                int journal_inum)
1986 {
1987         journal_t *journal;
1988
1989         if (sb->s_flags & MS_RDONLY) {
1990                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1991                                 "create journal.\n");
1992                 return -EROFS;
1993         }
1994
1995         if (!(journal = ext3_get_journal(sb, journal_inum)))
1996                 return -EINVAL;
1997
1998         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1999                journal_inum);
2000
2001         if (journal_create(journal)) {
2002                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2003                 journal_destroy(journal);
2004                 return -EIO;
2005         }
2006
2007         EXT3_SB(sb)->s_journal = journal;
2008
2009         ext3_update_dynamic_rev(sb);
2010         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2011         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2012
2013         es->s_journal_inum = cpu_to_le32(journal_inum);
2014         sb->s_dirt = 1;
2015
2016         /* Make sure we flush the recovery flag to disk. */
2017         ext3_commit_super(sb, es, 1);
2018
2019         return 0;
2020 }
2021
2022 static void ext3_commit_super (struct super_block * sb,
2023                                struct ext3_super_block * es,
2024                                int sync)
2025 {
2026         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2027
2028         if (!sbh)
2029                 return;
2030         es->s_wtime = cpu_to_le32(get_seconds());
2031         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2032         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2033         BUFFER_TRACE(sbh, "marking dirty");
2034         mark_buffer_dirty(sbh);
2035         if (sync)
2036                 sync_dirty_buffer(sbh);
2037 }
2038
2039
2040 /*
2041  * Have we just finished recovery?  If so, and if we are mounting (or
2042  * remounting) the filesystem readonly, then we will end up with a
2043  * consistent fs on disk.  Record that fact.
2044  */
2045 static void ext3_mark_recovery_complete(struct super_block * sb,
2046                                         struct ext3_super_block * es)
2047 {
2048         journal_t *journal = EXT3_SB(sb)->s_journal;
2049
2050         journal_lock_updates(journal);
2051         journal_flush(journal);
2052         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2053             sb->s_flags & MS_RDONLY) {
2054                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2055                 sb->s_dirt = 0;
2056                 ext3_commit_super(sb, es, 1);
2057         }
2058         journal_unlock_updates(journal);
2059 }
2060
2061 /*
2062  * If we are mounting (or read-write remounting) a filesystem whose journal
2063  * has recorded an error from a previous lifetime, move that error to the
2064  * main filesystem now.
2065  */
2066 static void ext3_clear_journal_err(struct super_block * sb,
2067                                    struct ext3_super_block * es)
2068 {
2069         journal_t *journal;
2070         int j_errno;
2071         const char *errstr;
2072
2073         journal = EXT3_SB(sb)->s_journal;
2074
2075         /*
2076          * Now check for any error status which may have been recorded in the
2077          * journal by a prior ext3_error() or ext3_abort()
2078          */
2079
2080         j_errno = journal_errno(journal);
2081         if (j_errno) {
2082                 char nbuf[16];
2083
2084                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2085                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2086                              "from previous mount: %s", errstr);
2087                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2088                              "filesystem check.");
2089
2090                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2091                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2092                 ext3_commit_super (sb, es, 1);
2093
2094                 journal_clear_err(journal);
2095         }
2096 }
2097
2098 /*
2099  * Force the running and committing transactions to commit,
2100  * and wait on the commit.
2101  */
2102 int ext3_force_commit(struct super_block *sb)
2103 {
2104         journal_t *journal;
2105         int ret;
2106
2107         if (sb->s_flags & MS_RDONLY)
2108                 return 0;
2109
2110         journal = EXT3_SB(sb)->s_journal;
2111         sb->s_dirt = 0;
2112         ret = ext3_journal_force_commit(journal);
2113         return ret;
2114 }
2115
2116 /*
2117  * Ext3 always journals updates to the superblock itself, so we don't
2118  * have to propagate any other updates to the superblock on disk at this
2119  * point.  Just start an async writeback to get the buffers on their way
2120  * to the disk.
2121  *
2122  * This implicitly triggers the writebehind on sync().
2123  */
2124
2125 static void ext3_write_super (struct super_block * sb)
2126 {
2127         if (down_trylock(&sb->s_lock) == 0)
2128                 BUG();
2129         sb->s_dirt = 0;
2130 }
2131
2132 static int ext3_sync_fs(struct super_block *sb, int wait)
2133 {
2134         tid_t target;
2135
2136         sb->s_dirt = 0;
2137         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2138                 if (wait)
2139                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2140         }
2141         return 0;
2142 }
2143
2144 /*
2145  * LVM calls this function before a (read-only) snapshot is created.  This
2146  * gives us a chance to flush the journal completely and mark the fs clean.
2147  */
2148 static void ext3_write_super_lockfs(struct super_block *sb)
2149 {
2150         sb->s_dirt = 0;
2151
2152         if (!(sb->s_flags & MS_RDONLY)) {
2153                 journal_t *journal = EXT3_SB(sb)->s_journal;
2154
2155                 /* Now we set up the journal barrier. */
2156                 journal_lock_updates(journal);
2157                 journal_flush(journal);
2158
2159                 /* Journal blocked and flushed, clear needs_recovery flag. */
2160                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2161                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2162         }
2163 }
2164
2165 /*
2166  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2167  * flag here, even though the filesystem is not technically dirty yet.
2168  */
2169 static void ext3_unlockfs(struct super_block *sb)
2170 {
2171         if (!(sb->s_flags & MS_RDONLY)) {
2172                 lock_super(sb);
2173                 /* Reser the needs_recovery flag before the fs is unlocked. */
2174                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2175                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2176                 unlock_super(sb);
2177                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2178         }
2179 }
2180
2181 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2182 {
2183         struct ext3_super_block * es;
2184         struct ext3_sb_info *sbi = EXT3_SB(sb);
2185         unsigned long n_blocks_count = 0;
2186         unsigned long old_sb_flags;
2187         struct ext3_mount_options old_opts;
2188         int err;
2189 #ifdef CONFIG_QUOTA
2190         int i;
2191 #endif
2192
2193         /* Store the original options */
2194         old_sb_flags = sb->s_flags;
2195         old_opts.s_mount_opt = sbi->s_mount_opt;
2196         old_opts.s_resuid = sbi->s_resuid;
2197         old_opts.s_resgid = sbi->s_resgid;
2198         old_opts.s_commit_interval = sbi->s_commit_interval;
2199 #ifdef CONFIG_QUOTA
2200         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2201         for (i = 0; i < MAXQUOTAS; i++)
2202                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2203 #endif
2204
2205         /*
2206          * Allow the "check" option to be passed as a remount option.
2207          */
2208         if (!parse_options(data, sb, NULL, &n_blocks_count, 1)) {
2209                 err = -EINVAL;
2210                 goto restore_opts;
2211         }
2212
2213         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2214                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2215
2216         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2217                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2218
2219         es = sbi->s_es;
2220
2221         ext3_init_journal_params(sb, sbi->s_journal);
2222
2223         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2224                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2225                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2226                         err = -EROFS;
2227                         goto restore_opts;
2228                 }
2229
2230                 if (*flags & MS_RDONLY) {
2231                         /*
2232                          * First of all, the unconditional stuff we have to do
2233                          * to disable replay of the journal when we next remount
2234                          */
2235                         sb->s_flags |= MS_RDONLY;
2236
2237                         /*
2238                          * OK, test if we are remounting a valid rw partition
2239                          * readonly, and if so set the rdonly flag and then
2240                          * mark the partition as valid again.
2241                          */
2242                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2243                             (sbi->s_mount_state & EXT3_VALID_FS))
2244                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2245
2246                         ext3_mark_recovery_complete(sb, es);
2247                 } else {
2248                         __le32 ret;
2249                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2250                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2251                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2252                                        "remount RDWR because of unsupported "
2253                                        "optional features (%x).\n",
2254                                        sb->s_id, le32_to_cpu(ret));
2255                                 err = -EROFS;
2256                                 goto restore_opts;
2257                         }
2258                         /*
2259                          * Mounting a RDONLY partition read-write, so reread
2260                          * and store the current valid flag.  (It may have
2261                          * been changed by e2fsck since we originally mounted
2262                          * the partition.)
2263                          */
2264                         ext3_clear_journal_err(sb, es);
2265                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2266                         if ((ret = ext3_group_extend(sb, es, n_blocks_count))) {
2267                                 err = ret;
2268                                 goto restore_opts;
2269                         }
2270                         if (!ext3_setup_super (sb, es, 0))
2271                                 sb->s_flags &= ~MS_RDONLY;
2272                 }
2273         }
2274 #ifdef CONFIG_QUOTA
2275         /* Release old quota file names */
2276         for (i = 0; i < MAXQUOTAS; i++)
2277                 if (old_opts.s_qf_names[i] &&
2278                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2279                         kfree(old_opts.s_qf_names[i]);
2280 #endif
2281         return 0;
2282 restore_opts:
2283         sb->s_flags = old_sb_flags;
2284         sbi->s_mount_opt = old_opts.s_mount_opt;
2285         sbi->s_resuid = old_opts.s_resuid;
2286         sbi->s_resgid = old_opts.s_resgid;
2287         sbi->s_commit_interval = old_opts.s_commit_interval;
2288 #ifdef CONFIG_QUOTA
2289         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2290         for (i = 0; i < MAXQUOTAS; i++) {
2291                 if (sbi->s_qf_names[i] &&
2292                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2293                         kfree(sbi->s_qf_names[i]);
2294                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2295         }
2296 #endif
2297         return err;
2298 }
2299
2300 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2301 {
2302         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2303         unsigned long overhead;
2304         int i;
2305
2306         if (test_opt (sb, MINIX_DF))
2307                 overhead = 0;
2308         else {
2309                 unsigned long ngroups;
2310                 ngroups = EXT3_SB(sb)->s_groups_count;
2311                 smp_rmb();
2312
2313                 /*
2314                  * Compute the overhead (FS structures)
2315                  */
2316
2317                 /*
2318                  * All of the blocks before first_data_block are
2319                  * overhead
2320                  */
2321                 overhead = le32_to_cpu(es->s_first_data_block);
2322
2323                 /*
2324                  * Add the overhead attributed to the superblock and
2325                  * block group descriptors.  If the sparse superblocks
2326                  * feature is turned on, then not all groups have this.
2327                  */
2328                 for (i = 0; i < ngroups; i++) {
2329                         overhead += ext3_bg_has_super(sb, i) +
2330                                 ext3_bg_num_gdb(sb, i);
2331                         cond_resched();
2332                 }
2333
2334                 /*
2335                  * Every block group has an inode bitmap, a block
2336                  * bitmap, and an inode table.
2337                  */
2338                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2339         }
2340
2341         buf->f_type = EXT3_SUPER_MAGIC;
2342         buf->f_bsize = sb->s_blocksize;
2343         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2344         buf->f_bfree = ext3_count_free_blocks (sb);
2345         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2346         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2347                 buf->f_bavail = 0;
2348         buf->f_files = le32_to_cpu(es->s_inodes_count);
2349         buf->f_ffree = ext3_count_free_inodes (sb);
2350         buf->f_namelen = EXT3_NAME_LEN;
2351         return 0;
2352 }
2353
2354 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2355  * is locked for write. Otherwise the are possible deadlocks:
2356  * Process 1                         Process 2
2357  * ext3_create()                     quota_sync()
2358  *   journal_start()                   write_dquot()
2359  *   DQUOT_INIT()                        down(dqio_sem)
2360  *     down(dqio_sem)                    journal_start()
2361  *
2362  */
2363
2364 #ifdef CONFIG_QUOTA
2365
2366 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2367 {
2368         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2369 }
2370
2371 static int ext3_dquot_initialize(struct inode *inode, int type)
2372 {
2373         handle_t *handle;
2374         int ret, err;
2375
2376         /* We may create quota structure so we need to reserve enough blocks */
2377         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2378         if (IS_ERR(handle))
2379                 return PTR_ERR(handle);
2380         ret = dquot_initialize(inode, type);
2381         err = ext3_journal_stop(handle);
2382         if (!ret)
2383                 ret = err;
2384         return ret;
2385 }
2386
2387 static int ext3_dquot_drop(struct inode *inode)
2388 {
2389         handle_t *handle;
2390         int ret, err;
2391
2392         /* We may delete quota structure so we need to reserve enough blocks */
2393         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2394         if (IS_ERR(handle))
2395                 return PTR_ERR(handle);
2396         ret = dquot_drop(inode);
2397         err = ext3_journal_stop(handle);
2398         if (!ret)
2399                 ret = err;
2400         return ret;
2401 }
2402
2403 static int ext3_write_dquot(struct dquot *dquot)
2404 {
2405         int ret, err;
2406         handle_t *handle;
2407         struct inode *inode;
2408
2409         inode = dquot_to_inode(dquot);
2410         handle = ext3_journal_start(inode,
2411                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2412         if (IS_ERR(handle))
2413                 return PTR_ERR(handle);
2414         ret = dquot_commit(dquot);
2415         err = ext3_journal_stop(handle);
2416         if (!ret)
2417                 ret = err;
2418         return ret;
2419 }
2420
2421 static int ext3_acquire_dquot(struct dquot *dquot)
2422 {
2423         int ret, err;
2424         handle_t *handle;
2425
2426         handle = ext3_journal_start(dquot_to_inode(dquot),
2427                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2428         if (IS_ERR(handle))
2429                 return PTR_ERR(handle);
2430         ret = dquot_acquire(dquot);
2431         err = ext3_journal_stop(handle);
2432         if (!ret)
2433                 ret = err;
2434         return ret;
2435 }
2436
2437 static int ext3_release_dquot(struct dquot *dquot)
2438 {
2439         int ret, err;
2440         handle_t *handle;
2441
2442         handle = ext3_journal_start(dquot_to_inode(dquot),
2443                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2444         if (IS_ERR(handle))
2445                 return PTR_ERR(handle);
2446         ret = dquot_release(dquot);
2447         err = ext3_journal_stop(handle);
2448         if (!ret)
2449                 ret = err;
2450         return ret;
2451 }
2452
2453 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2454 {
2455         /* Are we journalling quotas? */
2456         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2457             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2458                 dquot_mark_dquot_dirty(dquot);
2459                 return ext3_write_dquot(dquot);
2460         } else {
2461                 return dquot_mark_dquot_dirty(dquot);
2462         }
2463 }
2464
2465 static int ext3_write_info(struct super_block *sb, int type)
2466 {
2467         int ret, err;
2468         handle_t *handle;
2469
2470         /* Data block + inode block */
2471         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2472         if (IS_ERR(handle))
2473                 return PTR_ERR(handle);
2474         ret = dquot_commit_info(sb, type);
2475         err = ext3_journal_stop(handle);
2476         if (!ret)
2477                 ret = err;
2478         return ret;
2479 }
2480
2481 /*
2482  * Turn on quotas during mount time - we need to find
2483  * the quota file and such...
2484  */
2485 static int ext3_quota_on_mount(struct super_block *sb, int type)
2486 {
2487         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2488                         EXT3_SB(sb)->s_jquota_fmt, type);
2489 }
2490
2491 /*
2492  * Standard function to be called on quota_on
2493  */
2494 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2495                          char *path)
2496 {
2497         int err;
2498         struct nameidata nd;
2499
2500         if (!test_opt(sb, QUOTA))
2501                 return -EINVAL;
2502         /* Not journalling quota? */
2503         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2504             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2505                 return vfs_quota_on(sb, type, format_id, path);
2506         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2507         if (err)
2508                 return err;
2509         /* Quotafile not on the same filesystem? */
2510         if (nd.mnt->mnt_sb != sb) {
2511                 path_release(&nd);
2512                 return -EXDEV;
2513         }
2514         /* Quotafile not of fs root? */
2515         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2516                 printk(KERN_WARNING
2517                         "EXT3-fs: Quota file not on filesystem root. "
2518                         "Journalled quota will not work.\n");
2519         path_release(&nd);
2520         return vfs_quota_on(sb, type, format_id, path);
2521 }
2522
2523 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2524  * acquiring the locks... As quota files are never truncated and quota code
2525  * itself serializes the operations (and noone else should touch the files)
2526  * we don't have to be afraid of races */
2527 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2528                                size_t len, loff_t off)
2529 {
2530         struct inode *inode = sb_dqopt(sb)->files[type];
2531         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2532         int err = 0;
2533         int offset = off & (sb->s_blocksize - 1);
2534         int tocopy;
2535         size_t toread;
2536         struct buffer_head *bh;
2537         loff_t i_size = i_size_read(inode);
2538
2539         if (off > i_size)
2540                 return 0;
2541         if (off+len > i_size)
2542                 len = i_size-off;
2543         toread = len;
2544         while (toread > 0) {
2545                 tocopy = sb->s_blocksize - offset < toread ?
2546                                 sb->s_blocksize - offset : toread;
2547                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2548                 if (err)
2549                         return err;
2550                 if (!bh)        /* A hole? */
2551                         memset(data, 0, tocopy);
2552                 else
2553                         memcpy(data, bh->b_data+offset, tocopy);
2554                 brelse(bh);
2555                 offset = 0;
2556                 toread -= tocopy;
2557                 data += tocopy;
2558                 blk++;
2559         }
2560         return len;
2561 }
2562
2563 /* Write to quotafile (we know the transaction is already started and has
2564  * enough credits) */
2565 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2566                                 const char *data, size_t len, loff_t off)
2567 {
2568         struct inode *inode = sb_dqopt(sb)->files[type];
2569         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2570         int err = 0;
2571         int offset = off & (sb->s_blocksize - 1);
2572         int tocopy;
2573         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2574         size_t towrite = len;
2575         struct buffer_head *bh;
2576         handle_t *handle = journal_current_handle();
2577
2578         down(&inode->i_sem);
2579         while (towrite > 0) {
2580                 tocopy = sb->s_blocksize - offset < towrite ?
2581                                 sb->s_blocksize - offset : towrite;
2582                 bh = ext3_bread(handle, inode, blk, 1, &err);
2583                 if (!bh)
2584                         goto out;
2585                 if (journal_quota) {
2586                         err = ext3_journal_get_write_access(handle, bh);
2587                         if (err) {
2588                                 brelse(bh);
2589                                 goto out;
2590                         }
2591                 }
2592                 lock_buffer(bh);
2593                 memcpy(bh->b_data+offset, data, tocopy);
2594                 flush_dcache_page(bh->b_page);
2595                 unlock_buffer(bh);
2596                 if (journal_quota)
2597                         err = ext3_journal_dirty_metadata(handle, bh);
2598                 else {
2599                         /* Always do at least ordered writes for quotas */
2600                         err = ext3_journal_dirty_data(handle, bh);
2601                         mark_buffer_dirty(bh);
2602                 }
2603                 brelse(bh);
2604                 if (err)
2605                         goto out;
2606                 offset = 0;
2607                 towrite -= tocopy;
2608                 data += tocopy;
2609                 blk++;
2610         }
2611 out:
2612         if (len == towrite)
2613                 return err;
2614         if (inode->i_size < off+len-towrite) {
2615                 i_size_write(inode, off+len-towrite);
2616                 EXT3_I(inode)->i_disksize = inode->i_size;
2617         }
2618         inode->i_version++;
2619         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2620         ext3_mark_inode_dirty(handle, inode);
2621         up(&inode->i_sem);
2622         return len - towrite;
2623 }
2624
2625 #endif
2626
2627 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2628         int flags, const char *dev_name, void *data)
2629 {
2630         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2631 }
2632
2633 static struct file_system_type ext3_fs_type = {
2634         .owner          = THIS_MODULE,
2635         .name           = "ext3",
2636         .get_sb         = ext3_get_sb,
2637         .kill_sb        = kill_block_super,
2638         .fs_flags       = FS_REQUIRES_DEV,
2639 };
2640
2641 static int __init init_ext3_fs(void)
2642 {
2643         int err = init_ext3_xattr();
2644         if (err)
2645                 return err;
2646         err = init_inodecache();
2647         if (err)
2648                 goto out1;
2649         err = register_filesystem(&ext3_fs_type);
2650         if (err)
2651                 goto out;
2652         return 0;
2653 out:
2654         destroy_inodecache();
2655 out1:
2656         exit_ext3_xattr();
2657         return err;
2658 }
2659
2660 static void __exit exit_ext3_fs(void)
2661 {
2662         unregister_filesystem(&ext3_fs_type);
2663         destroy_inodecache();
2664         exit_ext3_xattr();
2665 }
2666
2667 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2668 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2669 MODULE_LICENSE("GPL");
2670 module_init(init_ext3_fs)
2671 module_exit(exit_ext3_fs)