Merge git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-misc-2.6
[linux-2.6] / fs / ext3 / super.c
1 /*
2  *  linux/fs/ext3/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/module.h>
20 #include <linux/string.h>
21 #include <linux/fs.h>
22 #include <linux/time.h>
23 #include <linux/jbd.h>
24 #include <linux/ext3_fs.h>
25 #include <linux/ext3_jbd.h>
26 #include <linux/slab.h>
27 #include <linux/init.h>
28 #include <linux/blkdev.h>
29 #include <linux/parser.h>
30 #include <linux/smp_lock.h>
31 #include <linux/buffer_head.h>
32 #include <linux/exportfs.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 <linux/log2.h>
40
41 #include <asm/uaccess.h>
42
43 #include "xattr.h"
44 #include "acl.h"
45 #include "namei.h"
46
47 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
48                              unsigned long journal_devnum);
49 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
50                                unsigned int);
51 static int ext3_commit_super(struct super_block *sb,
52                                struct ext3_super_block *es,
53                                int sync);
54 static void ext3_mark_recovery_complete(struct super_block * sb,
55                                         struct ext3_super_block * es);
56 static void ext3_clear_journal_err(struct super_block * sb,
57                                    struct ext3_super_block * es);
58 static int ext3_sync_fs(struct super_block *sb, int wait);
59 static const char *ext3_decode_error(struct super_block * sb, int errno,
60                                      char nbuf[16]);
61 static int ext3_remount (struct super_block * sb, int * flags, char * data);
62 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
63 static int ext3_unfreeze(struct super_block *sb);
64 static void ext3_write_super (struct super_block * sb);
65 static int ext3_freeze(struct super_block *sb);
66
67 /*
68  * Wrappers for journal_start/end.
69  *
70  * The only special thing we need to do here is to make sure that all
71  * journal_end calls result in the superblock being marked dirty, so
72  * that sync() will call the filesystem's write_super callback if
73  * appropriate.
74  */
75 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
76 {
77         journal_t *journal;
78
79         if (sb->s_flags & MS_RDONLY)
80                 return ERR_PTR(-EROFS);
81
82         /* Special case here: if the journal has aborted behind our
83          * backs (eg. EIO in the commit thread), then we still need to
84          * take the FS itself readonly cleanly. */
85         journal = EXT3_SB(sb)->s_journal;
86         if (is_journal_aborted(journal)) {
87                 ext3_abort(sb, __func__,
88                            "Detected aborted journal");
89                 return ERR_PTR(-EROFS);
90         }
91
92         return journal_start(journal, nblocks);
93 }
94
95 /*
96  * The only special thing we need to do here is to make sure that all
97  * journal_stop calls result in the superblock being marked dirty, so
98  * that sync() will call the filesystem's write_super callback if
99  * appropriate.
100  */
101 int __ext3_journal_stop(const char *where, handle_t *handle)
102 {
103         struct super_block *sb;
104         int err;
105         int rc;
106
107         sb = handle->h_transaction->t_journal->j_private;
108         err = handle->h_err;
109         rc = journal_stop(handle);
110
111         if (!err)
112                 err = rc;
113         if (err)
114                 __ext3_std_error(sb, where, err);
115         return err;
116 }
117
118 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
119                 struct buffer_head *bh, handle_t *handle, int err)
120 {
121         char nbuf[16];
122         const char *errstr = ext3_decode_error(NULL, err, nbuf);
123
124         if (bh)
125                 BUFFER_TRACE(bh, "abort");
126
127         if (!handle->h_err)
128                 handle->h_err = err;
129
130         if (is_handle_aborted(handle))
131                 return;
132
133         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
134                caller, errstr, err_fn);
135
136         journal_abort_handle(handle);
137 }
138
139 /* Deal with the reporting of failure conditions on a filesystem such as
140  * inconsistencies detected or read IO failures.
141  *
142  * On ext2, we can store the error state of the filesystem in the
143  * superblock.  That is not possible on ext3, because we may have other
144  * write ordering constraints on the superblock which prevent us from
145  * writing it out straight away; and given that the journal is about to
146  * be aborted, we can't rely on the current, or future, transactions to
147  * write out the superblock safely.
148  *
149  * We'll just use the journal_abort() error code to record an error in
150  * the journal instead.  On recovery, the journal will compain about
151  * that error until we've noted it down and cleared it.
152  */
153
154 static void ext3_handle_error(struct super_block *sb)
155 {
156         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
157
158         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
159         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
160
161         if (sb->s_flags & MS_RDONLY)
162                 return;
163
164         if (!test_opt (sb, ERRORS_CONT)) {
165                 journal_t *journal = EXT3_SB(sb)->s_journal;
166
167                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
168                 if (journal)
169                         journal_abort(journal, -EIO);
170         }
171         if (test_opt (sb, ERRORS_RO)) {
172                 printk (KERN_CRIT "Remounting filesystem read-only\n");
173                 sb->s_flags |= MS_RDONLY;
174         }
175         ext3_commit_super(sb, es, 1);
176         if (test_opt(sb, ERRORS_PANIC))
177                 panic("EXT3-fs (device %s): panic forced after error\n",
178                         sb->s_id);
179 }
180
181 void ext3_error (struct super_block * sb, const char * function,
182                  const char * fmt, ...)
183 {
184         va_list args;
185
186         va_start(args, fmt);
187         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
188         vprintk(fmt, args);
189         printk("\n");
190         va_end(args);
191
192         ext3_handle_error(sb);
193 }
194
195 static const char *ext3_decode_error(struct super_block * sb, int errno,
196                                      char nbuf[16])
197 {
198         char *errstr = NULL;
199
200         switch (errno) {
201         case -EIO:
202                 errstr = "IO failure";
203                 break;
204         case -ENOMEM:
205                 errstr = "Out of memory";
206                 break;
207         case -EROFS:
208                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
209                         errstr = "Journal has aborted";
210                 else
211                         errstr = "Readonly filesystem";
212                 break;
213         default:
214                 /* If the caller passed in an extra buffer for unknown
215                  * errors, textualise them now.  Else we just return
216                  * NULL. */
217                 if (nbuf) {
218                         /* Check for truncated error codes... */
219                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
220                                 errstr = nbuf;
221                 }
222                 break;
223         }
224
225         return errstr;
226 }
227
228 /* __ext3_std_error decodes expected errors from journaling functions
229  * automatically and invokes the appropriate error response.  */
230
231 void __ext3_std_error (struct super_block * sb, const char * function,
232                        int errno)
233 {
234         char nbuf[16];
235         const char *errstr;
236
237         /* Special case: if the error is EROFS, and we're not already
238          * inside a transaction, then there's really no point in logging
239          * an error. */
240         if (errno == -EROFS && journal_current_handle() == NULL &&
241             (sb->s_flags & MS_RDONLY))
242                 return;
243
244         errstr = ext3_decode_error(sb, errno, nbuf);
245         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
246                 sb->s_id, function, errstr);
247
248         ext3_handle_error(sb);
249 }
250
251 /*
252  * ext3_abort is a much stronger failure handler than ext3_error.  The
253  * abort function may be used to deal with unrecoverable failures such
254  * as journal IO errors or ENOMEM at a critical moment in log management.
255  *
256  * We unconditionally force the filesystem into an ABORT|READONLY state,
257  * unless the error response on the fs has been set to panic in which
258  * case we take the easy way out and panic immediately.
259  */
260
261 void ext3_abort (struct super_block * sb, const char * function,
262                  const char * fmt, ...)
263 {
264         va_list args;
265
266         printk (KERN_CRIT "ext3_abort called.\n");
267
268         va_start(args, fmt);
269         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
270         vprintk(fmt, args);
271         printk("\n");
272         va_end(args);
273
274         if (test_opt(sb, ERRORS_PANIC))
275                 panic("EXT3-fs panic from previous error\n");
276
277         if (sb->s_flags & MS_RDONLY)
278                 return;
279
280         printk(KERN_CRIT "Remounting filesystem read-only\n");
281         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
282         sb->s_flags |= MS_RDONLY;
283         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
284         if (EXT3_SB(sb)->s_journal)
285                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
286 }
287
288 void ext3_warning (struct super_block * sb, const char * function,
289                    const char * fmt, ...)
290 {
291         va_list args;
292
293         va_start(args, fmt);
294         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
295                sb->s_id, function);
296         vprintk(fmt, args);
297         printk("\n");
298         va_end(args);
299 }
300
301 void ext3_update_dynamic_rev(struct super_block *sb)
302 {
303         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
304
305         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
306                 return;
307
308         ext3_warning(sb, __func__,
309                      "updating to rev %d because of new feature flag, "
310                      "running e2fsck is recommended",
311                      EXT3_DYNAMIC_REV);
312
313         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
314         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
315         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
316         /* leave es->s_feature_*compat flags alone */
317         /* es->s_uuid will be set by e2fsck if empty */
318
319         /*
320          * The rest of the superblock fields should be zero, and if not it
321          * means they are likely already in use, so leave them alone.  We
322          * can leave it up to e2fsck to clean up any inconsistencies there.
323          */
324 }
325
326 /*
327  * Open the external journal device
328  */
329 static struct block_device *ext3_blkdev_get(dev_t dev)
330 {
331         struct block_device *bdev;
332         char b[BDEVNAME_SIZE];
333
334         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
335         if (IS_ERR(bdev))
336                 goto fail;
337         return bdev;
338
339 fail:
340         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
341                         __bdevname(dev, b), PTR_ERR(bdev));
342         return NULL;
343 }
344
345 /*
346  * Release the journal device
347  */
348 static int ext3_blkdev_put(struct block_device *bdev)
349 {
350         bd_release(bdev);
351         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
352 }
353
354 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
355 {
356         struct block_device *bdev;
357         int ret = -ENODEV;
358
359         bdev = sbi->journal_bdev;
360         if (bdev) {
361                 ret = ext3_blkdev_put(bdev);
362                 sbi->journal_bdev = NULL;
363         }
364         return ret;
365 }
366
367 static inline struct inode *orphan_list_entry(struct list_head *l)
368 {
369         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
370 }
371
372 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
373 {
374         struct list_head *l;
375
376         printk(KERN_ERR "sb orphan head is %d\n",
377                le32_to_cpu(sbi->s_es->s_last_orphan));
378
379         printk(KERN_ERR "sb_info orphan list:\n");
380         list_for_each(l, &sbi->s_orphan) {
381                 struct inode *inode = orphan_list_entry(l);
382                 printk(KERN_ERR "  "
383                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
384                        inode->i_sb->s_id, inode->i_ino, inode,
385                        inode->i_mode, inode->i_nlink,
386                        NEXT_ORPHAN(inode));
387         }
388 }
389
390 static void ext3_put_super (struct super_block * sb)
391 {
392         struct ext3_sb_info *sbi = EXT3_SB(sb);
393         struct ext3_super_block *es = sbi->s_es;
394         int i, err;
395
396         ext3_xattr_put_super(sb);
397         err = journal_destroy(sbi->s_journal);
398         sbi->s_journal = NULL;
399         if (err < 0)
400                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
401
402         if (!(sb->s_flags & MS_RDONLY)) {
403                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
404                 es->s_state = cpu_to_le16(sbi->s_mount_state);
405                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
406                 mark_buffer_dirty(sbi->s_sbh);
407                 ext3_commit_super(sb, es, 1);
408         }
409
410         for (i = 0; i < sbi->s_gdb_count; i++)
411                 brelse(sbi->s_group_desc[i]);
412         kfree(sbi->s_group_desc);
413         percpu_counter_destroy(&sbi->s_freeblocks_counter);
414         percpu_counter_destroy(&sbi->s_freeinodes_counter);
415         percpu_counter_destroy(&sbi->s_dirs_counter);
416         brelse(sbi->s_sbh);
417 #ifdef CONFIG_QUOTA
418         for (i = 0; i < MAXQUOTAS; i++)
419                 kfree(sbi->s_qf_names[i]);
420 #endif
421
422         /* Debugging code just in case the in-memory inode orphan list
423          * isn't empty.  The on-disk one can be non-empty if we've
424          * detected an error and taken the fs readonly, but the
425          * in-memory list had better be clean by this point. */
426         if (!list_empty(&sbi->s_orphan))
427                 dump_orphan_list(sb, sbi);
428         J_ASSERT(list_empty(&sbi->s_orphan));
429
430         invalidate_bdev(sb->s_bdev);
431         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
432                 /*
433                  * Invalidate the journal device's buffers.  We don't want them
434                  * floating about in memory - the physical journal device may
435                  * hotswapped, and it breaks the `ro-after' testing code.
436                  */
437                 sync_blockdev(sbi->journal_bdev);
438                 invalidate_bdev(sbi->journal_bdev);
439                 ext3_blkdev_remove(sbi);
440         }
441         sb->s_fs_info = NULL;
442         kfree(sbi->s_blockgroup_lock);
443         kfree(sbi);
444         return;
445 }
446
447 static struct kmem_cache *ext3_inode_cachep;
448
449 /*
450  * Called inside transaction, so use GFP_NOFS
451  */
452 static struct inode *ext3_alloc_inode(struct super_block *sb)
453 {
454         struct ext3_inode_info *ei;
455
456         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
457         if (!ei)
458                 return NULL;
459 #ifdef CONFIG_EXT3_FS_POSIX_ACL
460         ei->i_acl = EXT3_ACL_NOT_CACHED;
461         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
462 #endif
463         ei->i_block_alloc_info = NULL;
464         ei->vfs_inode.i_version = 1;
465         return &ei->vfs_inode;
466 }
467
468 static void ext3_destroy_inode(struct inode *inode)
469 {
470         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
471                 printk("EXT3 Inode %p: orphan list check failed!\n",
472                         EXT3_I(inode));
473                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
474                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
475                                 false);
476                 dump_stack();
477         }
478         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
479 }
480
481 static void init_once(void *foo)
482 {
483         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
484
485         INIT_LIST_HEAD(&ei->i_orphan);
486 #ifdef CONFIG_EXT3_FS_XATTR
487         init_rwsem(&ei->xattr_sem);
488 #endif
489         mutex_init(&ei->truncate_mutex);
490         inode_init_once(&ei->vfs_inode);
491 }
492
493 static int init_inodecache(void)
494 {
495         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
496                                              sizeof(struct ext3_inode_info),
497                                              0, (SLAB_RECLAIM_ACCOUNT|
498                                                 SLAB_MEM_SPREAD),
499                                              init_once);
500         if (ext3_inode_cachep == NULL)
501                 return -ENOMEM;
502         return 0;
503 }
504
505 static void destroy_inodecache(void)
506 {
507         kmem_cache_destroy(ext3_inode_cachep);
508 }
509
510 static void ext3_clear_inode(struct inode *inode)
511 {
512         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
513 #ifdef CONFIG_EXT3_FS_POSIX_ACL
514         if (EXT3_I(inode)->i_acl &&
515                         EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
516                 posix_acl_release(EXT3_I(inode)->i_acl);
517                 EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
518         }
519         if (EXT3_I(inode)->i_default_acl &&
520                         EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
521                 posix_acl_release(EXT3_I(inode)->i_default_acl);
522                 EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
523         }
524 #endif
525         ext3_discard_reservation(inode);
526         EXT3_I(inode)->i_block_alloc_info = NULL;
527         if (unlikely(rsv))
528                 kfree(rsv);
529 }
530
531 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
532 {
533 #if defined(CONFIG_QUOTA)
534         struct ext3_sb_info *sbi = EXT3_SB(sb);
535
536         if (sbi->s_jquota_fmt)
537                 seq_printf(seq, ",jqfmt=%s",
538                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
539
540         if (sbi->s_qf_names[USRQUOTA])
541                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
542
543         if (sbi->s_qf_names[GRPQUOTA])
544                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
545
546         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
547                 seq_puts(seq, ",usrquota");
548
549         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
550                 seq_puts(seq, ",grpquota");
551 #endif
552 }
553
554 /*
555  * Show an option if
556  *  - it's set to a non-default value OR
557  *  - if the per-sb default is different from the global default
558  */
559 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
560 {
561         struct super_block *sb = vfs->mnt_sb;
562         struct ext3_sb_info *sbi = EXT3_SB(sb);
563         struct ext3_super_block *es = sbi->s_es;
564         unsigned long def_mount_opts;
565
566         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
567
568         if (sbi->s_sb_block != 1)
569                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
570         if (test_opt(sb, MINIX_DF))
571                 seq_puts(seq, ",minixdf");
572         if (test_opt(sb, GRPID))
573                 seq_puts(seq, ",grpid");
574         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
575                 seq_puts(seq, ",nogrpid");
576         if (sbi->s_resuid != EXT3_DEF_RESUID ||
577             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
578                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
579         }
580         if (sbi->s_resgid != EXT3_DEF_RESGID ||
581             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
582                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
583         }
584         if (test_opt(sb, ERRORS_RO)) {
585                 int def_errors = le16_to_cpu(es->s_errors);
586
587                 if (def_errors == EXT3_ERRORS_PANIC ||
588                     def_errors == EXT3_ERRORS_CONTINUE) {
589                         seq_puts(seq, ",errors=remount-ro");
590                 }
591         }
592         if (test_opt(sb, ERRORS_CONT))
593                 seq_puts(seq, ",errors=continue");
594         if (test_opt(sb, ERRORS_PANIC))
595                 seq_puts(seq, ",errors=panic");
596         if (test_opt(sb, NO_UID32))
597                 seq_puts(seq, ",nouid32");
598         if (test_opt(sb, DEBUG))
599                 seq_puts(seq, ",debug");
600         if (test_opt(sb, OLDALLOC))
601                 seq_puts(seq, ",oldalloc");
602 #ifdef CONFIG_EXT3_FS_XATTR
603         if (test_opt(sb, XATTR_USER))
604                 seq_puts(seq, ",user_xattr");
605         if (!test_opt(sb, XATTR_USER) &&
606             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
607                 seq_puts(seq, ",nouser_xattr");
608         }
609 #endif
610 #ifdef CONFIG_EXT3_FS_POSIX_ACL
611         if (test_opt(sb, POSIX_ACL))
612                 seq_puts(seq, ",acl");
613         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
614                 seq_puts(seq, ",noacl");
615 #endif
616         if (!test_opt(sb, RESERVATION))
617                 seq_puts(seq, ",noreservation");
618         if (sbi->s_commit_interval) {
619                 seq_printf(seq, ",commit=%u",
620                            (unsigned) (sbi->s_commit_interval / HZ));
621         }
622         if (test_opt(sb, BARRIER))
623                 seq_puts(seq, ",barrier=1");
624         if (test_opt(sb, NOBH))
625                 seq_puts(seq, ",nobh");
626
627         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
628                 seq_puts(seq, ",data=journal");
629         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
630                 seq_puts(seq, ",data=ordered");
631         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
632                 seq_puts(seq, ",data=writeback");
633
634         if (test_opt(sb, DATA_ERR_ABORT))
635                 seq_puts(seq, ",data_err=abort");
636
637         ext3_show_quota_options(seq, sb);
638
639         return 0;
640 }
641
642
643 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
644                 u64 ino, u32 generation)
645 {
646         struct inode *inode;
647
648         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
649                 return ERR_PTR(-ESTALE);
650         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
651                 return ERR_PTR(-ESTALE);
652
653         /* iget isn't really right if the inode is currently unallocated!!
654          *
655          * ext3_read_inode will return a bad_inode if the inode had been
656          * deleted, so we should be safe.
657          *
658          * Currently we don't know the generation for parent directory, so
659          * a generation of 0 means "accept any"
660          */
661         inode = ext3_iget(sb, ino);
662         if (IS_ERR(inode))
663                 return ERR_CAST(inode);
664         if (generation && inode->i_generation != generation) {
665                 iput(inode);
666                 return ERR_PTR(-ESTALE);
667         }
668
669         return inode;
670 }
671
672 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
673                 int fh_len, int fh_type)
674 {
675         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
676                                     ext3_nfs_get_inode);
677 }
678
679 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
680                 int fh_len, int fh_type)
681 {
682         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
683                                     ext3_nfs_get_inode);
684 }
685
686 /*
687  * Try to release metadata pages (indirect blocks, directories) which are
688  * mapped via the block device.  Since these pages could have journal heads
689  * which would prevent try_to_free_buffers() from freeing them, we must use
690  * jbd layer's try_to_free_buffers() function to release them.
691  */
692 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
693                                  gfp_t wait)
694 {
695         journal_t *journal = EXT3_SB(sb)->s_journal;
696
697         WARN_ON(PageChecked(page));
698         if (!page_has_buffers(page))
699                 return 0;
700         if (journal)
701                 return journal_try_to_free_buffers(journal, page, 
702                                                    wait & ~__GFP_WAIT);
703         return try_to_free_buffers(page);
704 }
705
706 #ifdef CONFIG_QUOTA
707 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
708 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
709
710 static int ext3_write_dquot(struct dquot *dquot);
711 static int ext3_acquire_dquot(struct dquot *dquot);
712 static int ext3_release_dquot(struct dquot *dquot);
713 static int ext3_mark_dquot_dirty(struct dquot *dquot);
714 static int ext3_write_info(struct super_block *sb, int type);
715 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
716                                 char *path, int remount);
717 static int ext3_quota_on_mount(struct super_block *sb, int type);
718 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
719                                size_t len, loff_t off);
720 static ssize_t ext3_quota_write(struct super_block *sb, int type,
721                                 const char *data, size_t len, loff_t off);
722
723 static struct dquot_operations ext3_quota_operations = {
724         .initialize     = dquot_initialize,
725         .drop           = dquot_drop,
726         .alloc_space    = dquot_alloc_space,
727         .alloc_inode    = dquot_alloc_inode,
728         .free_space     = dquot_free_space,
729         .free_inode     = dquot_free_inode,
730         .transfer       = dquot_transfer,
731         .write_dquot    = ext3_write_dquot,
732         .acquire_dquot  = ext3_acquire_dquot,
733         .release_dquot  = ext3_release_dquot,
734         .mark_dirty     = ext3_mark_dquot_dirty,
735         .write_info     = ext3_write_info,
736         .alloc_dquot    = dquot_alloc,
737         .destroy_dquot  = dquot_destroy,
738 };
739
740 static struct quotactl_ops ext3_qctl_operations = {
741         .quota_on       = ext3_quota_on,
742         .quota_off      = vfs_quota_off,
743         .quota_sync     = vfs_quota_sync,
744         .get_info       = vfs_get_dqinfo,
745         .set_info       = vfs_set_dqinfo,
746         .get_dqblk      = vfs_get_dqblk,
747         .set_dqblk      = vfs_set_dqblk
748 };
749 #endif
750
751 static const struct super_operations ext3_sops = {
752         .alloc_inode    = ext3_alloc_inode,
753         .destroy_inode  = ext3_destroy_inode,
754         .write_inode    = ext3_write_inode,
755         .dirty_inode    = ext3_dirty_inode,
756         .delete_inode   = ext3_delete_inode,
757         .put_super      = ext3_put_super,
758         .write_super    = ext3_write_super,
759         .sync_fs        = ext3_sync_fs,
760         .freeze_fs      = ext3_freeze,
761         .unfreeze_fs    = ext3_unfreeze,
762         .statfs         = ext3_statfs,
763         .remount_fs     = ext3_remount,
764         .clear_inode    = ext3_clear_inode,
765         .show_options   = ext3_show_options,
766 #ifdef CONFIG_QUOTA
767         .quota_read     = ext3_quota_read,
768         .quota_write    = ext3_quota_write,
769 #endif
770         .bdev_try_to_free_page = bdev_try_to_free_page,
771 };
772
773 static const struct export_operations ext3_export_ops = {
774         .fh_to_dentry = ext3_fh_to_dentry,
775         .fh_to_parent = ext3_fh_to_parent,
776         .get_parent = ext3_get_parent,
777 };
778
779 enum {
780         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
781         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
782         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
783         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
784         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
785         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
786         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
787         Opt_data_err_abort, Opt_data_err_ignore,
788         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
789         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
790         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
791         Opt_grpquota
792 };
793
794 static const match_table_t tokens = {
795         {Opt_bsd_df, "bsddf"},
796         {Opt_minix_df, "minixdf"},
797         {Opt_grpid, "grpid"},
798         {Opt_grpid, "bsdgroups"},
799         {Opt_nogrpid, "nogrpid"},
800         {Opt_nogrpid, "sysvgroups"},
801         {Opt_resgid, "resgid=%u"},
802         {Opt_resuid, "resuid=%u"},
803         {Opt_sb, "sb=%u"},
804         {Opt_err_cont, "errors=continue"},
805         {Opt_err_panic, "errors=panic"},
806         {Opt_err_ro, "errors=remount-ro"},
807         {Opt_nouid32, "nouid32"},
808         {Opt_nocheck, "nocheck"},
809         {Opt_nocheck, "check=none"},
810         {Opt_debug, "debug"},
811         {Opt_oldalloc, "oldalloc"},
812         {Opt_orlov, "orlov"},
813         {Opt_user_xattr, "user_xattr"},
814         {Opt_nouser_xattr, "nouser_xattr"},
815         {Opt_acl, "acl"},
816         {Opt_noacl, "noacl"},
817         {Opt_reservation, "reservation"},
818         {Opt_noreservation, "noreservation"},
819         {Opt_noload, "noload"},
820         {Opt_nobh, "nobh"},
821         {Opt_bh, "bh"},
822         {Opt_commit, "commit=%u"},
823         {Opt_journal_update, "journal=update"},
824         {Opt_journal_inum, "journal=%u"},
825         {Opt_journal_dev, "journal_dev=%u"},
826         {Opt_abort, "abort"},
827         {Opt_data_journal, "data=journal"},
828         {Opt_data_ordered, "data=ordered"},
829         {Opt_data_writeback, "data=writeback"},
830         {Opt_data_err_abort, "data_err=abort"},
831         {Opt_data_err_ignore, "data_err=ignore"},
832         {Opt_offusrjquota, "usrjquota="},
833         {Opt_usrjquota, "usrjquota=%s"},
834         {Opt_offgrpjquota, "grpjquota="},
835         {Opt_grpjquota, "grpjquota=%s"},
836         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
837         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
838         {Opt_grpquota, "grpquota"},
839         {Opt_noquota, "noquota"},
840         {Opt_quota, "quota"},
841         {Opt_usrquota, "usrquota"},
842         {Opt_barrier, "barrier=%u"},
843         {Opt_resize, "resize"},
844         {Opt_err, NULL},
845 };
846
847 static ext3_fsblk_t get_sb_block(void **data)
848 {
849         ext3_fsblk_t    sb_block;
850         char            *options = (char *) *data;
851
852         if (!options || strncmp(options, "sb=", 3) != 0)
853                 return 1;       /* Default location */
854         options += 3;
855         /*todo: use simple_strtoll with >32bit ext3 */
856         sb_block = simple_strtoul(options, &options, 0);
857         if (*options && *options != ',') {
858                 printk("EXT3-fs: Invalid sb specification: %s\n",
859                        (char *) *data);
860                 return 1;
861         }
862         if (*options == ',')
863                 options++;
864         *data = (void *) options;
865         return sb_block;
866 }
867
868 static int parse_options (char *options, struct super_block *sb,
869                           unsigned int *inum, unsigned long *journal_devnum,
870                           ext3_fsblk_t *n_blocks_count, int is_remount)
871 {
872         struct ext3_sb_info *sbi = EXT3_SB(sb);
873         char * p;
874         substring_t args[MAX_OPT_ARGS];
875         int data_opt = 0;
876         int option;
877 #ifdef CONFIG_QUOTA
878         int qtype, qfmt;
879         char *qname;
880 #endif
881
882         if (!options)
883                 return 1;
884
885         while ((p = strsep (&options, ",")) != NULL) {
886                 int token;
887                 if (!*p)
888                         continue;
889
890                 token = match_token(p, tokens, args);
891                 switch (token) {
892                 case Opt_bsd_df:
893                         clear_opt (sbi->s_mount_opt, MINIX_DF);
894                         break;
895                 case Opt_minix_df:
896                         set_opt (sbi->s_mount_opt, MINIX_DF);
897                         break;
898                 case Opt_grpid:
899                         set_opt (sbi->s_mount_opt, GRPID);
900                         break;
901                 case Opt_nogrpid:
902                         clear_opt (sbi->s_mount_opt, GRPID);
903                         break;
904                 case Opt_resuid:
905                         if (match_int(&args[0], &option))
906                                 return 0;
907                         sbi->s_resuid = option;
908                         break;
909                 case Opt_resgid:
910                         if (match_int(&args[0], &option))
911                                 return 0;
912                         sbi->s_resgid = option;
913                         break;
914                 case Opt_sb:
915                         /* handled by get_sb_block() instead of here */
916                         /* *sb_block = match_int(&args[0]); */
917                         break;
918                 case Opt_err_panic:
919                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
920                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
921                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
922                         break;
923                 case Opt_err_ro:
924                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
925                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
926                         set_opt (sbi->s_mount_opt, ERRORS_RO);
927                         break;
928                 case Opt_err_cont:
929                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
930                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
931                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
932                         break;
933                 case Opt_nouid32:
934                         set_opt (sbi->s_mount_opt, NO_UID32);
935                         break;
936                 case Opt_nocheck:
937                         clear_opt (sbi->s_mount_opt, CHECK);
938                         break;
939                 case Opt_debug:
940                         set_opt (sbi->s_mount_opt, DEBUG);
941                         break;
942                 case Opt_oldalloc:
943                         set_opt (sbi->s_mount_opt, OLDALLOC);
944                         break;
945                 case Opt_orlov:
946                         clear_opt (sbi->s_mount_opt, OLDALLOC);
947                         break;
948 #ifdef CONFIG_EXT3_FS_XATTR
949                 case Opt_user_xattr:
950                         set_opt (sbi->s_mount_opt, XATTR_USER);
951                         break;
952                 case Opt_nouser_xattr:
953                         clear_opt (sbi->s_mount_opt, XATTR_USER);
954                         break;
955 #else
956                 case Opt_user_xattr:
957                 case Opt_nouser_xattr:
958                         printk("EXT3 (no)user_xattr options not supported\n");
959                         break;
960 #endif
961 #ifdef CONFIG_EXT3_FS_POSIX_ACL
962                 case Opt_acl:
963                         set_opt(sbi->s_mount_opt, POSIX_ACL);
964                         break;
965                 case Opt_noacl:
966                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
967                         break;
968 #else
969                 case Opt_acl:
970                 case Opt_noacl:
971                         printk("EXT3 (no)acl options not supported\n");
972                         break;
973 #endif
974                 case Opt_reservation:
975                         set_opt(sbi->s_mount_opt, RESERVATION);
976                         break;
977                 case Opt_noreservation:
978                         clear_opt(sbi->s_mount_opt, RESERVATION);
979                         break;
980                 case Opt_journal_update:
981                         /* @@@ FIXME */
982                         /* Eventually we will want to be able to create
983                            a journal file here.  For now, only allow the
984                            user to specify an existing inode to be the
985                            journal file. */
986                         if (is_remount) {
987                                 printk(KERN_ERR "EXT3-fs: cannot specify "
988                                        "journal on remount\n");
989                                 return 0;
990                         }
991                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
992                         break;
993                 case Opt_journal_inum:
994                         if (is_remount) {
995                                 printk(KERN_ERR "EXT3-fs: cannot specify "
996                                        "journal on remount\n");
997                                 return 0;
998                         }
999                         if (match_int(&args[0], &option))
1000                                 return 0;
1001                         *inum = option;
1002                         break;
1003                 case Opt_journal_dev:
1004                         if (is_remount) {
1005                                 printk(KERN_ERR "EXT3-fs: cannot specify "
1006                                        "journal on remount\n");
1007                                 return 0;
1008                         }
1009                         if (match_int(&args[0], &option))
1010                                 return 0;
1011                         *journal_devnum = option;
1012                         break;
1013                 case Opt_noload:
1014                         set_opt (sbi->s_mount_opt, NOLOAD);
1015                         break;
1016                 case Opt_commit:
1017                         if (match_int(&args[0], &option))
1018                                 return 0;
1019                         if (option < 0)
1020                                 return 0;
1021                         if (option == 0)
1022                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1023                         sbi->s_commit_interval = HZ * option;
1024                         break;
1025                 case Opt_data_journal:
1026                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1027                         goto datacheck;
1028                 case Opt_data_ordered:
1029                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1030                         goto datacheck;
1031                 case Opt_data_writeback:
1032                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1033                 datacheck:
1034                         if (is_remount) {
1035                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
1036                                                 != data_opt) {
1037                                         printk(KERN_ERR
1038                                                 "EXT3-fs: cannot change data "
1039                                                 "mode on remount\n");
1040                                         return 0;
1041                                 }
1042                         } else {
1043                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
1044                                 sbi->s_mount_opt |= data_opt;
1045                         }
1046                         break;
1047                 case Opt_data_err_abort:
1048                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1049                         break;
1050                 case Opt_data_err_ignore:
1051                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1052                         break;
1053 #ifdef CONFIG_QUOTA
1054                 case Opt_usrjquota:
1055                         qtype = USRQUOTA;
1056                         goto set_qf_name;
1057                 case Opt_grpjquota:
1058                         qtype = GRPQUOTA;
1059 set_qf_name:
1060                         if (sb_any_quota_loaded(sb) &&
1061                             !sbi->s_qf_names[qtype]) {
1062                                 printk(KERN_ERR
1063                                         "EXT3-fs: Cannot change journaled "
1064                                         "quota options when quota turned on.\n");
1065                                 return 0;
1066                         }
1067                         qname = match_strdup(&args[0]);
1068                         if (!qname) {
1069                                 printk(KERN_ERR
1070                                         "EXT3-fs: not enough memory for "
1071                                         "storing quotafile name.\n");
1072                                 return 0;
1073                         }
1074                         if (sbi->s_qf_names[qtype] &&
1075                             strcmp(sbi->s_qf_names[qtype], qname)) {
1076                                 printk(KERN_ERR
1077                                         "EXT3-fs: %s quota file already "
1078                                         "specified.\n", QTYPE2NAME(qtype));
1079                                 kfree(qname);
1080                                 return 0;
1081                         }
1082                         sbi->s_qf_names[qtype] = qname;
1083                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1084                                 printk(KERN_ERR
1085                                         "EXT3-fs: quotafile must be on "
1086                                         "filesystem root.\n");
1087                                 kfree(sbi->s_qf_names[qtype]);
1088                                 sbi->s_qf_names[qtype] = NULL;
1089                                 return 0;
1090                         }
1091                         set_opt(sbi->s_mount_opt, QUOTA);
1092                         break;
1093                 case Opt_offusrjquota:
1094                         qtype = USRQUOTA;
1095                         goto clear_qf_name;
1096                 case Opt_offgrpjquota:
1097                         qtype = GRPQUOTA;
1098 clear_qf_name:
1099                         if (sb_any_quota_loaded(sb) &&
1100                             sbi->s_qf_names[qtype]) {
1101                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1102                                         "journaled quota options when "
1103                                         "quota turned on.\n");
1104                                 return 0;
1105                         }
1106                         /*
1107                          * The space will be released later when all options
1108                          * are confirmed to be correct
1109                          */
1110                         sbi->s_qf_names[qtype] = NULL;
1111                         break;
1112                 case Opt_jqfmt_vfsold:
1113                         qfmt = QFMT_VFS_OLD;
1114                         goto set_qf_format;
1115                 case Opt_jqfmt_vfsv0:
1116                         qfmt = QFMT_VFS_V0;
1117 set_qf_format:
1118                         if (sb_any_quota_loaded(sb) &&
1119                             sbi->s_jquota_fmt != qfmt) {
1120                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1121                                         "journaled quota options when "
1122                                         "quota turned on.\n");
1123                                 return 0;
1124                         }
1125                         sbi->s_jquota_fmt = qfmt;
1126                         break;
1127                 case Opt_quota:
1128                 case Opt_usrquota:
1129                         set_opt(sbi->s_mount_opt, QUOTA);
1130                         set_opt(sbi->s_mount_opt, USRQUOTA);
1131                         break;
1132                 case Opt_grpquota:
1133                         set_opt(sbi->s_mount_opt, QUOTA);
1134                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1135                         break;
1136                 case Opt_noquota:
1137                         if (sb_any_quota_loaded(sb)) {
1138                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
1139                                         "options when quota turned on.\n");
1140                                 return 0;
1141                         }
1142                         clear_opt(sbi->s_mount_opt, QUOTA);
1143                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1144                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1145                         break;
1146 #else
1147                 case Opt_quota:
1148                 case Opt_usrquota:
1149                 case Opt_grpquota:
1150                         printk(KERN_ERR
1151                                 "EXT3-fs: quota options not supported.\n");
1152                         break;
1153                 case Opt_usrjquota:
1154                 case Opt_grpjquota:
1155                 case Opt_offusrjquota:
1156                 case Opt_offgrpjquota:
1157                 case Opt_jqfmt_vfsold:
1158                 case Opt_jqfmt_vfsv0:
1159                         printk(KERN_ERR
1160                                 "EXT3-fs: journaled quota options not "
1161                                 "supported.\n");
1162                         break;
1163                 case Opt_noquota:
1164                         break;
1165 #endif
1166                 case Opt_abort:
1167                         set_opt(sbi->s_mount_opt, ABORT);
1168                         break;
1169                 case Opt_barrier:
1170                         if (match_int(&args[0], &option))
1171                                 return 0;
1172                         if (option)
1173                                 set_opt(sbi->s_mount_opt, BARRIER);
1174                         else
1175                                 clear_opt(sbi->s_mount_opt, BARRIER);
1176                         break;
1177                 case Opt_ignore:
1178                         break;
1179                 case Opt_resize:
1180                         if (!is_remount) {
1181                                 printk("EXT3-fs: resize option only available "
1182                                         "for remount\n");
1183                                 return 0;
1184                         }
1185                         if (match_int(&args[0], &option) != 0)
1186                                 return 0;
1187                         *n_blocks_count = option;
1188                         break;
1189                 case Opt_nobh:
1190                         set_opt(sbi->s_mount_opt, NOBH);
1191                         break;
1192                 case Opt_bh:
1193                         clear_opt(sbi->s_mount_opt, NOBH);
1194                         break;
1195                 default:
1196                         printk (KERN_ERR
1197                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1198                                 "or missing value\n", p);
1199                         return 0;
1200                 }
1201         }
1202 #ifdef CONFIG_QUOTA
1203         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1204                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1205                      sbi->s_qf_names[USRQUOTA])
1206                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1207
1208                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1209                      sbi->s_qf_names[GRPQUOTA])
1210                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1211
1212                 if ((sbi->s_qf_names[USRQUOTA] &&
1213                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1214                     (sbi->s_qf_names[GRPQUOTA] &&
1215                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1216                         printk(KERN_ERR "EXT3-fs: old and new quota "
1217                                         "format mixing.\n");
1218                         return 0;
1219                 }
1220
1221                 if (!sbi->s_jquota_fmt) {
1222                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1223                                         "not specified.\n");
1224                         return 0;
1225                 }
1226         } else {
1227                 if (sbi->s_jquota_fmt) {
1228                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1229                                         "specified with no journaling "
1230                                         "enabled.\n");
1231                         return 0;
1232                 }
1233         }
1234 #endif
1235         return 1;
1236 }
1237
1238 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1239                             int read_only)
1240 {
1241         struct ext3_sb_info *sbi = EXT3_SB(sb);
1242         int res = 0;
1243
1244         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1245                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1246                         "forcing read-only mode\n");
1247                 res = MS_RDONLY;
1248         }
1249         if (read_only)
1250                 return res;
1251         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1252                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1253                         "running e2fsck is recommended\n");
1254         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1255                 printk (KERN_WARNING
1256                         "EXT3-fs warning: mounting fs with errors, "
1257                         "running e2fsck is recommended\n");
1258         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1259                  le16_to_cpu(es->s_mnt_count) >=
1260                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1261                 printk (KERN_WARNING
1262                         "EXT3-fs warning: maximal mount count reached, "
1263                         "running e2fsck is recommended\n");
1264         else if (le32_to_cpu(es->s_checkinterval) &&
1265                 (le32_to_cpu(es->s_lastcheck) +
1266                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1267                 printk (KERN_WARNING
1268                         "EXT3-fs warning: checktime reached, "
1269                         "running e2fsck is recommended\n");
1270 #if 0
1271                 /* @@@ We _will_ want to clear the valid bit if we find
1272                    inconsistencies, to force a fsck at reboot.  But for
1273                    a plain journaled filesystem we can keep it set as
1274                    valid forever! :) */
1275         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1276 #endif
1277         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1278                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1279         le16_add_cpu(&es->s_mnt_count, 1);
1280         es->s_mtime = cpu_to_le32(get_seconds());
1281         ext3_update_dynamic_rev(sb);
1282         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1283
1284         ext3_commit_super(sb, es, 1);
1285         if (test_opt(sb, DEBUG))
1286                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1287                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1288                         sb->s_blocksize,
1289                         sbi->s_groups_count,
1290                         EXT3_BLOCKS_PER_GROUP(sb),
1291                         EXT3_INODES_PER_GROUP(sb),
1292                         sbi->s_mount_opt);
1293
1294         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1295         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1296                 char b[BDEVNAME_SIZE];
1297
1298                 printk("external journal on %s\n",
1299                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1300         } else {
1301                 printk("internal journal\n");
1302         }
1303         return res;
1304 }
1305
1306 /* Called at mount-time, super-block is locked */
1307 static int ext3_check_descriptors(struct super_block *sb)
1308 {
1309         struct ext3_sb_info *sbi = EXT3_SB(sb);
1310         int i;
1311
1312         ext3_debug ("Checking group descriptors");
1313
1314         for (i = 0; i < sbi->s_groups_count; i++) {
1315                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1316                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1317                 ext3_fsblk_t last_block;
1318
1319                 if (i == sbi->s_groups_count - 1)
1320                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1321                 else
1322                         last_block = first_block +
1323                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1324
1325                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1326                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1327                 {
1328                         ext3_error (sb, "ext3_check_descriptors",
1329                                     "Block bitmap for group %d"
1330                                     " not in group (block %lu)!",
1331                                     i, (unsigned long)
1332                                         le32_to_cpu(gdp->bg_block_bitmap));
1333                         return 0;
1334                 }
1335                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1336                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1337                 {
1338                         ext3_error (sb, "ext3_check_descriptors",
1339                                     "Inode bitmap for group %d"
1340                                     " not in group (block %lu)!",
1341                                     i, (unsigned long)
1342                                         le32_to_cpu(gdp->bg_inode_bitmap));
1343                         return 0;
1344                 }
1345                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1346                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1347                     last_block)
1348                 {
1349                         ext3_error (sb, "ext3_check_descriptors",
1350                                     "Inode table for group %d"
1351                                     " not in group (block %lu)!",
1352                                     i, (unsigned long)
1353                                         le32_to_cpu(gdp->bg_inode_table));
1354                         return 0;
1355                 }
1356         }
1357
1358         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1359         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1360         return 1;
1361 }
1362
1363
1364 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1365  * the superblock) which were deleted from all directories, but held open by
1366  * a process at the time of a crash.  We walk the list and try to delete these
1367  * inodes at recovery time (only with a read-write filesystem).
1368  *
1369  * In order to keep the orphan inode chain consistent during traversal (in
1370  * case of crash during recovery), we link each inode into the superblock
1371  * orphan list_head and handle it the same way as an inode deletion during
1372  * normal operation (which journals the operations for us).
1373  *
1374  * We only do an iget() and an iput() on each inode, which is very safe if we
1375  * accidentally point at an in-use or already deleted inode.  The worst that
1376  * can happen in this case is that we get a "bit already cleared" message from
1377  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1378  * e2fsck was run on this filesystem, and it must have already done the orphan
1379  * inode cleanup for us, so we can safely abort without any further action.
1380  */
1381 static void ext3_orphan_cleanup (struct super_block * sb,
1382                                  struct ext3_super_block * es)
1383 {
1384         unsigned int s_flags = sb->s_flags;
1385         int nr_orphans = 0, nr_truncates = 0;
1386 #ifdef CONFIG_QUOTA
1387         int i;
1388 #endif
1389         if (!es->s_last_orphan) {
1390                 jbd_debug(4, "no orphan inodes to clean up\n");
1391                 return;
1392         }
1393
1394         if (bdev_read_only(sb->s_bdev)) {
1395                 printk(KERN_ERR "EXT3-fs: write access "
1396                         "unavailable, skipping orphan cleanup.\n");
1397                 return;
1398         }
1399
1400         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1401                 if (es->s_last_orphan)
1402                         jbd_debug(1, "Errors on filesystem, "
1403                                   "clearing orphan list.\n");
1404                 es->s_last_orphan = 0;
1405                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1406                 return;
1407         }
1408
1409         if (s_flags & MS_RDONLY) {
1410                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1411                        sb->s_id);
1412                 sb->s_flags &= ~MS_RDONLY;
1413         }
1414 #ifdef CONFIG_QUOTA
1415         /* Needed for iput() to work correctly and not trash data */
1416         sb->s_flags |= MS_ACTIVE;
1417         /* Turn on quotas so that they are updated correctly */
1418         for (i = 0; i < MAXQUOTAS; i++) {
1419                 if (EXT3_SB(sb)->s_qf_names[i]) {
1420                         int ret = ext3_quota_on_mount(sb, i);
1421                         if (ret < 0)
1422                                 printk(KERN_ERR
1423                                         "EXT3-fs: Cannot turn on journaled "
1424                                         "quota: error %d\n", ret);
1425                 }
1426         }
1427 #endif
1428
1429         while (es->s_last_orphan) {
1430                 struct inode *inode;
1431
1432                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1433                 if (IS_ERR(inode)) {
1434                         es->s_last_orphan = 0;
1435                         break;
1436                 }
1437
1438                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1439                 vfs_dq_init(inode);
1440                 if (inode->i_nlink) {
1441                         printk(KERN_DEBUG
1442                                 "%s: truncating inode %lu to %Ld bytes\n",
1443                                 __func__, inode->i_ino, inode->i_size);
1444                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1445                                   inode->i_ino, inode->i_size);
1446                         ext3_truncate(inode);
1447                         nr_truncates++;
1448                 } else {
1449                         printk(KERN_DEBUG
1450                                 "%s: deleting unreferenced inode %lu\n",
1451                                 __func__, inode->i_ino);
1452                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1453                                   inode->i_ino);
1454                         nr_orphans++;
1455                 }
1456                 iput(inode);  /* The delete magic happens here! */
1457         }
1458
1459 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1460
1461         if (nr_orphans)
1462                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1463                        sb->s_id, PLURAL(nr_orphans));
1464         if (nr_truncates)
1465                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1466                        sb->s_id, PLURAL(nr_truncates));
1467 #ifdef CONFIG_QUOTA
1468         /* Turn quotas off */
1469         for (i = 0; i < MAXQUOTAS; i++) {
1470                 if (sb_dqopt(sb)->files[i])
1471                         vfs_quota_off(sb, i, 0);
1472         }
1473 #endif
1474         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1475 }
1476
1477 /*
1478  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1479  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1480  * We need to be 1 filesystem block less than the 2^32 sector limit.
1481  */
1482 static loff_t ext3_max_size(int bits)
1483 {
1484         loff_t res = EXT3_NDIR_BLOCKS;
1485         int meta_blocks;
1486         loff_t upper_limit;
1487
1488         /* This is calculated to be the largest file size for a
1489          * dense, file such that the total number of
1490          * sectors in the file, including data and all indirect blocks,
1491          * does not exceed 2^32 -1
1492          * __u32 i_blocks representing the total number of
1493          * 512 bytes blocks of the file
1494          */
1495         upper_limit = (1LL << 32) - 1;
1496
1497         /* total blocks in file system block size */
1498         upper_limit >>= (bits - 9);
1499
1500
1501         /* indirect blocks */
1502         meta_blocks = 1;
1503         /* double indirect blocks */
1504         meta_blocks += 1 + (1LL << (bits-2));
1505         /* tripple indirect blocks */
1506         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1507
1508         upper_limit -= meta_blocks;
1509         upper_limit <<= bits;
1510
1511         res += 1LL << (bits-2);
1512         res += 1LL << (2*(bits-2));
1513         res += 1LL << (3*(bits-2));
1514         res <<= bits;
1515         if (res > upper_limit)
1516                 res = upper_limit;
1517
1518         if (res > MAX_LFS_FILESIZE)
1519                 res = MAX_LFS_FILESIZE;
1520
1521         return res;
1522 }
1523
1524 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1525                                     ext3_fsblk_t logic_sb_block,
1526                                     int nr)
1527 {
1528         struct ext3_sb_info *sbi = EXT3_SB(sb);
1529         unsigned long bg, first_meta_bg;
1530         int has_super = 0;
1531
1532         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1533
1534         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1535             nr < first_meta_bg)
1536                 return (logic_sb_block + nr + 1);
1537         bg = sbi->s_desc_per_block * nr;
1538         if (ext3_bg_has_super(sb, bg))
1539                 has_super = 1;
1540         return (has_super + ext3_group_first_block_no(sb, bg));
1541 }
1542
1543
1544 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1545 {
1546         struct buffer_head * bh;
1547         struct ext3_super_block *es = NULL;
1548         struct ext3_sb_info *sbi;
1549         ext3_fsblk_t block;
1550         ext3_fsblk_t sb_block = get_sb_block(&data);
1551         ext3_fsblk_t logic_sb_block;
1552         unsigned long offset = 0;
1553         unsigned int journal_inum = 0;
1554         unsigned long journal_devnum = 0;
1555         unsigned long def_mount_opts;
1556         struct inode *root;
1557         int blocksize;
1558         int hblock;
1559         int db_count;
1560         int i;
1561         int needs_recovery;
1562         int ret = -EINVAL;
1563         __le32 features;
1564         int err;
1565
1566         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1567         if (!sbi)
1568                 return -ENOMEM;
1569
1570         sbi->s_blockgroup_lock =
1571                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1572         if (!sbi->s_blockgroup_lock) {
1573                 kfree(sbi);
1574                 return -ENOMEM;
1575         }
1576         sb->s_fs_info = sbi;
1577         sbi->s_mount_opt = 0;
1578         sbi->s_resuid = EXT3_DEF_RESUID;
1579         sbi->s_resgid = EXT3_DEF_RESGID;
1580         sbi->s_sb_block = sb_block;
1581
1582         unlock_kernel();
1583
1584         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1585         if (!blocksize) {
1586                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1587                 goto out_fail;
1588         }
1589
1590         /*
1591          * The ext3 superblock will not be buffer aligned for other than 1kB
1592          * block sizes.  We need to calculate the offset from buffer start.
1593          */
1594         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1595                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1596                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1597         } else {
1598                 logic_sb_block = sb_block;
1599         }
1600
1601         if (!(bh = sb_bread(sb, logic_sb_block))) {
1602                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1603                 goto out_fail;
1604         }
1605         /*
1606          * Note: s_es must be initialized as soon as possible because
1607          *       some ext3 macro-instructions depend on its value
1608          */
1609         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1610         sbi->s_es = es;
1611         sb->s_magic = le16_to_cpu(es->s_magic);
1612         if (sb->s_magic != EXT3_SUPER_MAGIC)
1613                 goto cantfind_ext3;
1614
1615         /* Set defaults before we parse the mount options */
1616         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1617         if (def_mount_opts & EXT3_DEFM_DEBUG)
1618                 set_opt(sbi->s_mount_opt, DEBUG);
1619         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1620                 set_opt(sbi->s_mount_opt, GRPID);
1621         if (def_mount_opts & EXT3_DEFM_UID16)
1622                 set_opt(sbi->s_mount_opt, NO_UID32);
1623 #ifdef CONFIG_EXT3_FS_XATTR
1624         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1625                 set_opt(sbi->s_mount_opt, XATTR_USER);
1626 #endif
1627 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1628         if (def_mount_opts & EXT3_DEFM_ACL)
1629                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1630 #endif
1631         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1632                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1633         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1634                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1635         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1636                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1637
1638         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1639                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1640         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1641                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1642         else
1643                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1644
1645         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1646         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1647
1648         set_opt(sbi->s_mount_opt, RESERVATION);
1649
1650         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1651                             NULL, 0))
1652                 goto failed_mount;
1653
1654         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1655                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1656
1657         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1658             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1659              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1660              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1661                 printk(KERN_WARNING
1662                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1663                        "running e2fsck is recommended\n");
1664         /*
1665          * Check feature flags regardless of the revision level, since we
1666          * previously didn't change the revision level when setting the flags,
1667          * so there is a chance incompat flags are set on a rev 0 filesystem.
1668          */
1669         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1670         if (features) {
1671                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1672                        "unsupported optional features (%x).\n",
1673                        sb->s_id, le32_to_cpu(features));
1674                 goto failed_mount;
1675         }
1676         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1677         if (!(sb->s_flags & MS_RDONLY) && features) {
1678                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1679                        "unsupported optional features (%x).\n",
1680                        sb->s_id, le32_to_cpu(features));
1681                 goto failed_mount;
1682         }
1683         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1684
1685         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1686             blocksize > EXT3_MAX_BLOCK_SIZE) {
1687                 printk(KERN_ERR
1688                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1689                        blocksize, sb->s_id);
1690                 goto failed_mount;
1691         }
1692
1693         hblock = bdev_hardsect_size(sb->s_bdev);
1694         if (sb->s_blocksize != blocksize) {
1695                 /*
1696                  * Make sure the blocksize for the filesystem is larger
1697                  * than the hardware sectorsize for the machine.
1698                  */
1699                 if (blocksize < hblock) {
1700                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1701                                "device blocksize %d.\n", blocksize, hblock);
1702                         goto failed_mount;
1703                 }
1704
1705                 brelse (bh);
1706                 if (!sb_set_blocksize(sb, blocksize)) {
1707                         printk(KERN_ERR "EXT3-fs: bad blocksize %d.\n",
1708                                 blocksize);
1709                         goto out_fail;
1710                 }
1711                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1712                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1713                 bh = sb_bread(sb, logic_sb_block);
1714                 if (!bh) {
1715                         printk(KERN_ERR
1716                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1717                         goto failed_mount;
1718                 }
1719                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1720                 sbi->s_es = es;
1721                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1722                         printk (KERN_ERR
1723                                 "EXT3-fs: Magic mismatch, very weird !\n");
1724                         goto failed_mount;
1725                 }
1726         }
1727
1728         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1729
1730         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1731                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1732                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1733         } else {
1734                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1735                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1736                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1737                     (!is_power_of_2(sbi->s_inode_size)) ||
1738                     (sbi->s_inode_size > blocksize)) {
1739                         printk (KERN_ERR
1740                                 "EXT3-fs: unsupported inode size: %d\n",
1741                                 sbi->s_inode_size);
1742                         goto failed_mount;
1743                 }
1744         }
1745         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1746                                    le32_to_cpu(es->s_log_frag_size);
1747         if (blocksize != sbi->s_frag_size) {
1748                 printk(KERN_ERR
1749                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1750                        sbi->s_frag_size, blocksize);
1751                 goto failed_mount;
1752         }
1753         sbi->s_frags_per_block = 1;
1754         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1755         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1756         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1757         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1758                 goto cantfind_ext3;
1759         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1760         if (sbi->s_inodes_per_block == 0)
1761                 goto cantfind_ext3;
1762         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1763                                         sbi->s_inodes_per_block;
1764         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1765         sbi->s_sbh = bh;
1766         sbi->s_mount_state = le16_to_cpu(es->s_state);
1767         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1768         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1769         for (i=0; i < 4; i++)
1770                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1771         sbi->s_def_hash_version = es->s_def_hash_version;
1772         i = le32_to_cpu(es->s_flags);
1773         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1774                 sbi->s_hash_unsigned = 3;
1775         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1776 #ifdef __CHAR_UNSIGNED__
1777                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1778                 sbi->s_hash_unsigned = 3;
1779 #else
1780                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1781 #endif
1782                 sb->s_dirt = 1;
1783         }
1784
1785         if (sbi->s_blocks_per_group > blocksize * 8) {
1786                 printk (KERN_ERR
1787                         "EXT3-fs: #blocks per group too big: %lu\n",
1788                         sbi->s_blocks_per_group);
1789                 goto failed_mount;
1790         }
1791         if (sbi->s_frags_per_group > blocksize * 8) {
1792                 printk (KERN_ERR
1793                         "EXT3-fs: #fragments per group too big: %lu\n",
1794                         sbi->s_frags_per_group);
1795                 goto failed_mount;
1796         }
1797         if (sbi->s_inodes_per_group > blocksize * 8) {
1798                 printk (KERN_ERR
1799                         "EXT3-fs: #inodes per group too big: %lu\n",
1800                         sbi->s_inodes_per_group);
1801                 goto failed_mount;
1802         }
1803
1804         if (le32_to_cpu(es->s_blocks_count) >
1805                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1806                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1807                         " too large to mount safely\n", sb->s_id);
1808                 if (sizeof(sector_t) < 8)
1809                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1810                                         "enabled\n");
1811                 goto failed_mount;
1812         }
1813
1814         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1815                 goto cantfind_ext3;
1816         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1817                                le32_to_cpu(es->s_first_data_block) - 1)
1818                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1819         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1820                    EXT3_DESC_PER_BLOCK(sb);
1821         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1822                                     GFP_KERNEL);
1823         if (sbi->s_group_desc == NULL) {
1824                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1825                 goto failed_mount;
1826         }
1827
1828         bgl_lock_init(sbi->s_blockgroup_lock);
1829
1830         for (i = 0; i < db_count; i++) {
1831                 block = descriptor_loc(sb, logic_sb_block, i);
1832                 sbi->s_group_desc[i] = sb_bread(sb, block);
1833                 if (!sbi->s_group_desc[i]) {
1834                         printk (KERN_ERR "EXT3-fs: "
1835                                 "can't read group descriptor %d\n", i);
1836                         db_count = i;
1837                         goto failed_mount2;
1838                 }
1839         }
1840         if (!ext3_check_descriptors (sb)) {
1841                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1842                 goto failed_mount2;
1843         }
1844         sbi->s_gdb_count = db_count;
1845         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1846         spin_lock_init(&sbi->s_next_gen_lock);
1847
1848         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1849                         ext3_count_free_blocks(sb));
1850         if (!err) {
1851                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1852                                 ext3_count_free_inodes(sb));
1853         }
1854         if (!err) {
1855                 err = percpu_counter_init(&sbi->s_dirs_counter,
1856                                 ext3_count_dirs(sb));
1857         }
1858         if (err) {
1859                 printk(KERN_ERR "EXT3-fs: insufficient memory\n");
1860                 goto failed_mount3;
1861         }
1862
1863         /* per fileystem reservation list head & lock */
1864         spin_lock_init(&sbi->s_rsv_window_lock);
1865         sbi->s_rsv_window_root = RB_ROOT;
1866         /* Add a single, static dummy reservation to the start of the
1867          * reservation window list --- it gives us a placeholder for
1868          * append-at-start-of-list which makes the allocation logic
1869          * _much_ simpler. */
1870         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1871         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1872         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1873         sbi->s_rsv_window_head.rsv_goal_size = 0;
1874         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1875
1876         /*
1877          * set up enough so that it can read an inode
1878          */
1879         sb->s_op = &ext3_sops;
1880         sb->s_export_op = &ext3_export_ops;
1881         sb->s_xattr = ext3_xattr_handlers;
1882 #ifdef CONFIG_QUOTA
1883         sb->s_qcop = &ext3_qctl_operations;
1884         sb->dq_op = &ext3_quota_operations;
1885 #endif
1886         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1887
1888         sb->s_root = NULL;
1889
1890         needs_recovery = (es->s_last_orphan != 0 ||
1891                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1892                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1893
1894         /*
1895          * The first inode we look at is the journal inode.  Don't try
1896          * root first: it may be modified in the journal!
1897          */
1898         if (!test_opt(sb, NOLOAD) &&
1899             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1900                 if (ext3_load_journal(sb, es, journal_devnum))
1901                         goto failed_mount3;
1902         } else if (journal_inum) {
1903                 if (ext3_create_journal(sb, es, journal_inum))
1904                         goto failed_mount3;
1905         } else {
1906                 if (!silent)
1907                         printk (KERN_ERR
1908                                 "ext3: No journal on filesystem on %s\n",
1909                                 sb->s_id);
1910                 goto failed_mount3;
1911         }
1912
1913         /* We have now updated the journal if required, so we can
1914          * validate the data journaling mode. */
1915         switch (test_opt(sb, DATA_FLAGS)) {
1916         case 0:
1917                 /* No mode set, assume a default based on the journal
1918                    capabilities: ORDERED_DATA if the journal can
1919                    cope, else JOURNAL_DATA */
1920                 if (journal_check_available_features
1921                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1922                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1923                 else
1924                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1925                 break;
1926
1927         case EXT3_MOUNT_ORDERED_DATA:
1928         case EXT3_MOUNT_WRITEBACK_DATA:
1929                 if (!journal_check_available_features
1930                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1931                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1932                                "requested data journaling mode\n");
1933                         goto failed_mount4;
1934                 }
1935         default:
1936                 break;
1937         }
1938
1939         if (test_opt(sb, NOBH)) {
1940                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1941                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1942                                 "its supported only with writeback mode\n");
1943                         clear_opt(sbi->s_mount_opt, NOBH);
1944                 }
1945         }
1946         /*
1947          * The journal_load will have done any necessary log recovery,
1948          * so we can safely mount the rest of the filesystem now.
1949          */
1950
1951         root = ext3_iget(sb, EXT3_ROOT_INO);
1952         if (IS_ERR(root)) {
1953                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1954                 ret = PTR_ERR(root);
1955                 goto failed_mount4;
1956         }
1957         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1958                 iput(root);
1959                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1960                 goto failed_mount4;
1961         }
1962         sb->s_root = d_alloc_root(root);
1963         if (!sb->s_root) {
1964                 printk(KERN_ERR "EXT3-fs: get root dentry failed\n");
1965                 iput(root);
1966                 ret = -ENOMEM;
1967                 goto failed_mount4;
1968         }
1969
1970         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1971         /*
1972          * akpm: core read_super() calls in here with the superblock locked.
1973          * That deadlocks, because orphan cleanup needs to lock the superblock
1974          * in numerous places.  Here we just pop the lock - it's relatively
1975          * harmless, because we are now ready to accept write_super() requests,
1976          * and aviro says that's the only reason for hanging onto the
1977          * superblock lock.
1978          */
1979         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1980         ext3_orphan_cleanup(sb, es);
1981         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1982         if (needs_recovery)
1983                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1984         ext3_mark_recovery_complete(sb, es);
1985         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1986                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1987                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1988                 "writeback");
1989
1990         lock_kernel();
1991         return 0;
1992
1993 cantfind_ext3:
1994         if (!silent)
1995                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1996                        sb->s_id);
1997         goto failed_mount;
1998
1999 failed_mount4:
2000         journal_destroy(sbi->s_journal);
2001 failed_mount3:
2002         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2003         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2004         percpu_counter_destroy(&sbi->s_dirs_counter);
2005 failed_mount2:
2006         for (i = 0; i < db_count; i++)
2007                 brelse(sbi->s_group_desc[i]);
2008         kfree(sbi->s_group_desc);
2009 failed_mount:
2010 #ifdef CONFIG_QUOTA
2011         for (i = 0; i < MAXQUOTAS; i++)
2012                 kfree(sbi->s_qf_names[i]);
2013 #endif
2014         ext3_blkdev_remove(sbi);
2015         brelse(bh);
2016 out_fail:
2017         sb->s_fs_info = NULL;
2018         kfree(sbi);
2019         lock_kernel();
2020         return ret;
2021 }
2022
2023 /*
2024  * Setup any per-fs journal parameters now.  We'll do this both on
2025  * initial mount, once the journal has been initialised but before we've
2026  * done any recovery; and again on any subsequent remount.
2027  */
2028 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2029 {
2030         struct ext3_sb_info *sbi = EXT3_SB(sb);
2031
2032         if (sbi->s_commit_interval)
2033                 journal->j_commit_interval = sbi->s_commit_interval;
2034         /* We could also set up an ext3-specific default for the commit
2035          * interval here, but for now we'll just fall back to the jbd
2036          * default. */
2037
2038         spin_lock(&journal->j_state_lock);
2039         if (test_opt(sb, BARRIER))
2040                 journal->j_flags |= JFS_BARRIER;
2041         else
2042                 journal->j_flags &= ~JFS_BARRIER;
2043         if (test_opt(sb, DATA_ERR_ABORT))
2044                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2045         else
2046                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2047         spin_unlock(&journal->j_state_lock);
2048 }
2049
2050 static journal_t *ext3_get_journal(struct super_block *sb,
2051                                    unsigned int journal_inum)
2052 {
2053         struct inode *journal_inode;
2054         journal_t *journal;
2055
2056         /* First, test for the existence of a valid inode on disk.  Bad
2057          * things happen if we iget() an unused inode, as the subsequent
2058          * iput() will try to delete it. */
2059
2060         journal_inode = ext3_iget(sb, journal_inum);
2061         if (IS_ERR(journal_inode)) {
2062                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
2063                 return NULL;
2064         }
2065         if (!journal_inode->i_nlink) {
2066                 make_bad_inode(journal_inode);
2067                 iput(journal_inode);
2068                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
2069                 return NULL;
2070         }
2071
2072         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2073                   journal_inode, journal_inode->i_size);
2074         if (!S_ISREG(journal_inode->i_mode)) {
2075                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
2076                 iput(journal_inode);
2077                 return NULL;
2078         }
2079
2080         journal = journal_init_inode(journal_inode);
2081         if (!journal) {
2082                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
2083                 iput(journal_inode);
2084                 return NULL;
2085         }
2086         journal->j_private = sb;
2087         ext3_init_journal_params(sb, journal);
2088         return journal;
2089 }
2090
2091 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2092                                        dev_t j_dev)
2093 {
2094         struct buffer_head * bh;
2095         journal_t *journal;
2096         ext3_fsblk_t start;
2097         ext3_fsblk_t len;
2098         int hblock, blocksize;
2099         ext3_fsblk_t sb_block;
2100         unsigned long offset;
2101         struct ext3_super_block * es;
2102         struct block_device *bdev;
2103
2104         bdev = ext3_blkdev_get(j_dev);
2105         if (bdev == NULL)
2106                 return NULL;
2107
2108         if (bd_claim(bdev, sb)) {
2109                 printk(KERN_ERR
2110                         "EXT3: failed to claim external journal device.\n");
2111                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2112                 return NULL;
2113         }
2114
2115         blocksize = sb->s_blocksize;
2116         hblock = bdev_hardsect_size(bdev);
2117         if (blocksize < hblock) {
2118                 printk(KERN_ERR
2119                         "EXT3-fs: blocksize too small for journal device.\n");
2120                 goto out_bdev;
2121         }
2122
2123         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2124         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2125         set_blocksize(bdev, blocksize);
2126         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2127                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
2128                        "external journal\n");
2129                 goto out_bdev;
2130         }
2131
2132         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2133         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2134             !(le32_to_cpu(es->s_feature_incompat) &
2135               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2136                 printk(KERN_ERR "EXT3-fs: external journal has "
2137                                         "bad superblock\n");
2138                 brelse(bh);
2139                 goto out_bdev;
2140         }
2141
2142         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2143                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
2144                 brelse(bh);
2145                 goto out_bdev;
2146         }
2147
2148         len = le32_to_cpu(es->s_blocks_count);
2149         start = sb_block + 1;
2150         brelse(bh);     /* we're done with the superblock */
2151
2152         journal = journal_init_dev(bdev, sb->s_bdev,
2153                                         start, len, blocksize);
2154         if (!journal) {
2155                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
2156                 goto out_bdev;
2157         }
2158         journal->j_private = sb;
2159         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2160         wait_on_buffer(journal->j_sb_buffer);
2161         if (!buffer_uptodate(journal->j_sb_buffer)) {
2162                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
2163                 goto out_journal;
2164         }
2165         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2166                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
2167                                         "user (unsupported) - %d\n",
2168                         be32_to_cpu(journal->j_superblock->s_nr_users));
2169                 goto out_journal;
2170         }
2171         EXT3_SB(sb)->journal_bdev = bdev;
2172         ext3_init_journal_params(sb, journal);
2173         return journal;
2174 out_journal:
2175         journal_destroy(journal);
2176 out_bdev:
2177         ext3_blkdev_put(bdev);
2178         return NULL;
2179 }
2180
2181 static int ext3_load_journal(struct super_block *sb,
2182                              struct ext3_super_block *es,
2183                              unsigned long journal_devnum)
2184 {
2185         journal_t *journal;
2186         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2187         dev_t journal_dev;
2188         int err = 0;
2189         int really_read_only;
2190
2191         if (journal_devnum &&
2192             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2193                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
2194                         "numbers have changed\n");
2195                 journal_dev = new_decode_dev(journal_devnum);
2196         } else
2197                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2198
2199         really_read_only = bdev_read_only(sb->s_bdev);
2200
2201         /*
2202          * Are we loading a blank journal or performing recovery after a
2203          * crash?  For recovery, we need to check in advance whether we
2204          * can get read-write access to the device.
2205          */
2206
2207         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2208                 if (sb->s_flags & MS_RDONLY) {
2209                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
2210                                         "required on readonly filesystem.\n");
2211                         if (really_read_only) {
2212                                 printk(KERN_ERR "EXT3-fs: write access "
2213                                         "unavailable, cannot proceed.\n");
2214                                 return -EROFS;
2215                         }
2216                         printk (KERN_INFO "EXT3-fs: write access will "
2217                                         "be enabled during recovery.\n");
2218                 }
2219         }
2220
2221         if (journal_inum && journal_dev) {
2222                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
2223                        "and inode journals!\n");
2224                 return -EINVAL;
2225         }
2226
2227         if (journal_inum) {
2228                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2229                         return -EINVAL;
2230         } else {
2231                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2232                         return -EINVAL;
2233         }
2234
2235         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2236                 err = journal_update_format(journal);
2237                 if (err)  {
2238                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
2239                         journal_destroy(journal);
2240                         return err;
2241                 }
2242         }
2243
2244         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2245                 err = journal_wipe(journal, !really_read_only);
2246         if (!err)
2247                 err = journal_load(journal);
2248
2249         if (err) {
2250                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2251                 journal_destroy(journal);
2252                 return err;
2253         }
2254
2255         EXT3_SB(sb)->s_journal = journal;
2256         ext3_clear_journal_err(sb, es);
2257
2258         if (journal_devnum &&
2259             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2260                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2261                 sb->s_dirt = 1;
2262
2263                 /* Make sure we flush the recovery flag to disk. */
2264                 ext3_commit_super(sb, es, 1);
2265         }
2266
2267         return 0;
2268 }
2269
2270 static int ext3_create_journal(struct super_block * sb,
2271                                struct ext3_super_block * es,
2272                                unsigned int journal_inum)
2273 {
2274         journal_t *journal;
2275         int err;
2276
2277         if (sb->s_flags & MS_RDONLY) {
2278                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2279                                 "create journal.\n");
2280                 return -EROFS;
2281         }
2282
2283         journal = ext3_get_journal(sb, journal_inum);
2284         if (!journal)
2285                 return -EINVAL;
2286
2287         printk(KERN_INFO "EXT3-fs: creating new journal on inode %u\n",
2288                journal_inum);
2289
2290         err = journal_create(journal);
2291         if (err) {
2292                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2293                 journal_destroy(journal);
2294                 return -EIO;
2295         }
2296
2297         EXT3_SB(sb)->s_journal = journal;
2298
2299         ext3_update_dynamic_rev(sb);
2300         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2301         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2302
2303         es->s_journal_inum = cpu_to_le32(journal_inum);
2304         sb->s_dirt = 1;
2305
2306         /* Make sure we flush the recovery flag to disk. */
2307         ext3_commit_super(sb, es, 1);
2308
2309         return 0;
2310 }
2311
2312 static int ext3_commit_super(struct super_block *sb,
2313                                struct ext3_super_block *es,
2314                                int sync)
2315 {
2316         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2317         int error = 0;
2318
2319         if (!sbh)
2320                 return error;
2321         es->s_wtime = cpu_to_le32(get_seconds());
2322         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2323         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2324         BUFFER_TRACE(sbh, "marking dirty");
2325         mark_buffer_dirty(sbh);
2326         if (sync)
2327                 error = sync_dirty_buffer(sbh);
2328         return error;
2329 }
2330
2331
2332 /*
2333  * Have we just finished recovery?  If so, and if we are mounting (or
2334  * remounting) the filesystem readonly, then we will end up with a
2335  * consistent fs on disk.  Record that fact.
2336  */
2337 static void ext3_mark_recovery_complete(struct super_block * sb,
2338                                         struct ext3_super_block * es)
2339 {
2340         journal_t *journal = EXT3_SB(sb)->s_journal;
2341
2342         journal_lock_updates(journal);
2343         if (journal_flush(journal) < 0)
2344                 goto out;
2345
2346         lock_super(sb);
2347         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2348             sb->s_flags & MS_RDONLY) {
2349                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2350                 sb->s_dirt = 0;
2351                 ext3_commit_super(sb, es, 1);
2352         }
2353         unlock_super(sb);
2354
2355 out:
2356         journal_unlock_updates(journal);
2357 }
2358
2359 /*
2360  * If we are mounting (or read-write remounting) a filesystem whose journal
2361  * has recorded an error from a previous lifetime, move that error to the
2362  * main filesystem now.
2363  */
2364 static void ext3_clear_journal_err(struct super_block * sb,
2365                                    struct ext3_super_block * es)
2366 {
2367         journal_t *journal;
2368         int j_errno;
2369         const char *errstr;
2370
2371         journal = EXT3_SB(sb)->s_journal;
2372
2373         /*
2374          * Now check for any error status which may have been recorded in the
2375          * journal by a prior ext3_error() or ext3_abort()
2376          */
2377
2378         j_errno = journal_errno(journal);
2379         if (j_errno) {
2380                 char nbuf[16];
2381
2382                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2383                 ext3_warning(sb, __func__, "Filesystem error recorded "
2384                              "from previous mount: %s", errstr);
2385                 ext3_warning(sb, __func__, "Marking fs in need of "
2386                              "filesystem check.");
2387
2388                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2389                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2390                 ext3_commit_super (sb, es, 1);
2391
2392                 journal_clear_err(journal);
2393         }
2394 }
2395
2396 /*
2397  * Force the running and committing transactions to commit,
2398  * and wait on the commit.
2399  */
2400 int ext3_force_commit(struct super_block *sb)
2401 {
2402         journal_t *journal;
2403         int ret;
2404
2405         if (sb->s_flags & MS_RDONLY)
2406                 return 0;
2407
2408         journal = EXT3_SB(sb)->s_journal;
2409         sb->s_dirt = 0;
2410         ret = ext3_journal_force_commit(journal);
2411         return ret;
2412 }
2413
2414 /*
2415  * Ext3 always journals updates to the superblock itself, so we don't
2416  * have to propagate any other updates to the superblock on disk at this
2417  * point.  (We can probably nuke this function altogether, and remove
2418  * any mention to sb->s_dirt in all of fs/ext3; eventual cleanup...)
2419  */
2420 static void ext3_write_super (struct super_block * sb)
2421 {
2422         if (mutex_trylock(&sb->s_lock) != 0)
2423                 BUG();
2424         sb->s_dirt = 0;
2425 }
2426
2427 static int ext3_sync_fs(struct super_block *sb, int wait)
2428 {
2429         tid_t target;
2430
2431         sb->s_dirt = 0;
2432         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2433                 if (wait)
2434                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2435         }
2436         return 0;
2437 }
2438
2439 /*
2440  * LVM calls this function before a (read-only) snapshot is created.  This
2441  * gives us a chance to flush the journal completely and mark the fs clean.
2442  */
2443 static int ext3_freeze(struct super_block *sb)
2444 {
2445         int error = 0;
2446         journal_t *journal;
2447         sb->s_dirt = 0;
2448
2449         if (!(sb->s_flags & MS_RDONLY)) {
2450                 journal = EXT3_SB(sb)->s_journal;
2451
2452                 /* Now we set up the journal barrier. */
2453                 journal_lock_updates(journal);
2454
2455                 /*
2456                  * We don't want to clear needs_recovery flag when we failed
2457                  * to flush the journal.
2458                  */
2459                 error = journal_flush(journal);
2460                 if (error < 0)
2461                         goto out;
2462
2463                 /* Journal blocked and flushed, clear needs_recovery flag. */
2464                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2465                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2466                 if (error)
2467                         goto out;
2468         }
2469         return 0;
2470
2471 out:
2472         journal_unlock_updates(journal);
2473         return error;
2474 }
2475
2476 /*
2477  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2478  * flag here, even though the filesystem is not technically dirty yet.
2479  */
2480 static int ext3_unfreeze(struct super_block *sb)
2481 {
2482         if (!(sb->s_flags & MS_RDONLY)) {
2483                 lock_super(sb);
2484                 /* Reser the needs_recovery flag before the fs is unlocked. */
2485                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2486                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2487                 unlock_super(sb);
2488                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2489         }
2490         return 0;
2491 }
2492
2493 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2494 {
2495         struct ext3_super_block * es;
2496         struct ext3_sb_info *sbi = EXT3_SB(sb);
2497         ext3_fsblk_t n_blocks_count = 0;
2498         unsigned long old_sb_flags;
2499         struct ext3_mount_options old_opts;
2500         int err;
2501 #ifdef CONFIG_QUOTA
2502         int i;
2503 #endif
2504
2505         /* Store the original options */
2506         old_sb_flags = sb->s_flags;
2507         old_opts.s_mount_opt = sbi->s_mount_opt;
2508         old_opts.s_resuid = sbi->s_resuid;
2509         old_opts.s_resgid = sbi->s_resgid;
2510         old_opts.s_commit_interval = sbi->s_commit_interval;
2511 #ifdef CONFIG_QUOTA
2512         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2513         for (i = 0; i < MAXQUOTAS; i++)
2514                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2515 #endif
2516
2517         /*
2518          * Allow the "check" option to be passed as a remount option.
2519          */
2520         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2521                 err = -EINVAL;
2522                 goto restore_opts;
2523         }
2524
2525         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2526                 ext3_abort(sb, __func__, "Abort forced by user");
2527
2528         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2529                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2530
2531         es = sbi->s_es;
2532
2533         ext3_init_journal_params(sb, sbi->s_journal);
2534
2535         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2536                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2537                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2538                         err = -EROFS;
2539                         goto restore_opts;
2540                 }
2541
2542                 if (*flags & MS_RDONLY) {
2543                         /*
2544                          * First of all, the unconditional stuff we have to do
2545                          * to disable replay of the journal when we next remount
2546                          */
2547                         sb->s_flags |= MS_RDONLY;
2548
2549                         /*
2550                          * OK, test if we are remounting a valid rw partition
2551                          * readonly, and if so set the rdonly flag and then
2552                          * mark the partition as valid again.
2553                          */
2554                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2555                             (sbi->s_mount_state & EXT3_VALID_FS))
2556                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2557
2558                         /*
2559                          * We have to unlock super so that we can wait for
2560                          * transactions.
2561                          */
2562                         unlock_super(sb);
2563                         ext3_mark_recovery_complete(sb, es);
2564                         lock_super(sb);
2565                 } else {
2566                         __le32 ret;
2567                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2568                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2569                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2570                                        "remount RDWR because of unsupported "
2571                                        "optional features (%x).\n",
2572                                        sb->s_id, le32_to_cpu(ret));
2573                                 err = -EROFS;
2574                                 goto restore_opts;
2575                         }
2576
2577                         /*
2578                          * If we have an unprocessed orphan list hanging
2579                          * around from a previously readonly bdev mount,
2580                          * require a full umount/remount for now.
2581                          */
2582                         if (es->s_last_orphan) {
2583                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2584                                        "remount RDWR because of unprocessed "
2585                                        "orphan inode list.  Please "
2586                                        "umount/remount instead.\n",
2587                                        sb->s_id);
2588                                 err = -EINVAL;
2589                                 goto restore_opts;
2590                         }
2591
2592                         /*
2593                          * Mounting a RDONLY partition read-write, so reread
2594                          * and store the current valid flag.  (It may have
2595                          * been changed by e2fsck since we originally mounted
2596                          * the partition.)
2597                          */
2598                         ext3_clear_journal_err(sb, es);
2599                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2600                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2601                                 goto restore_opts;
2602                         if (!ext3_setup_super (sb, es, 0))
2603                                 sb->s_flags &= ~MS_RDONLY;
2604                 }
2605         }
2606 #ifdef CONFIG_QUOTA
2607         /* Release old quota file names */
2608         for (i = 0; i < MAXQUOTAS; i++)
2609                 if (old_opts.s_qf_names[i] &&
2610                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2611                         kfree(old_opts.s_qf_names[i]);
2612 #endif
2613         return 0;
2614 restore_opts:
2615         sb->s_flags = old_sb_flags;
2616         sbi->s_mount_opt = old_opts.s_mount_opt;
2617         sbi->s_resuid = old_opts.s_resuid;
2618         sbi->s_resgid = old_opts.s_resgid;
2619         sbi->s_commit_interval = old_opts.s_commit_interval;
2620 #ifdef CONFIG_QUOTA
2621         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2622         for (i = 0; i < MAXQUOTAS; i++) {
2623                 if (sbi->s_qf_names[i] &&
2624                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2625                         kfree(sbi->s_qf_names[i]);
2626                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2627         }
2628 #endif
2629         return err;
2630 }
2631
2632 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2633 {
2634         struct super_block *sb = dentry->d_sb;
2635         struct ext3_sb_info *sbi = EXT3_SB(sb);
2636         struct ext3_super_block *es = sbi->s_es;
2637         u64 fsid;
2638
2639         if (test_opt(sb, MINIX_DF)) {
2640                 sbi->s_overhead_last = 0;
2641         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2642                 unsigned long ngroups = sbi->s_groups_count, i;
2643                 ext3_fsblk_t overhead = 0;
2644                 smp_rmb();
2645
2646                 /*
2647                  * Compute the overhead (FS structures).  This is constant
2648                  * for a given filesystem unless the number of block groups
2649                  * changes so we cache the previous value until it does.
2650                  */
2651
2652                 /*
2653                  * All of the blocks before first_data_block are
2654                  * overhead
2655                  */
2656                 overhead = le32_to_cpu(es->s_first_data_block);
2657
2658                 /*
2659                  * Add the overhead attributed to the superblock and
2660                  * block group descriptors.  If the sparse superblocks
2661                  * feature is turned on, then not all groups have this.
2662                  */
2663                 for (i = 0; i < ngroups; i++) {
2664                         overhead += ext3_bg_has_super(sb, i) +
2665                                 ext3_bg_num_gdb(sb, i);
2666                         cond_resched();
2667                 }
2668
2669                 /*
2670                  * Every block group has an inode bitmap, a block
2671                  * bitmap, and an inode table.
2672                  */
2673                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2674                 sbi->s_overhead_last = overhead;
2675                 smp_wmb();
2676                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2677         }
2678
2679         buf->f_type = EXT3_SUPER_MAGIC;
2680         buf->f_bsize = sb->s_blocksize;
2681         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2682         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2683         es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
2684         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2685         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2686                 buf->f_bavail = 0;
2687         buf->f_files = le32_to_cpu(es->s_inodes_count);
2688         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2689         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2690         buf->f_namelen = EXT3_NAME_LEN;
2691         fsid = le64_to_cpup((void *)es->s_uuid) ^
2692                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2693         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2694         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2695         return 0;
2696 }
2697
2698 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2699  * is locked for write. Otherwise the are possible deadlocks:
2700  * Process 1                         Process 2
2701  * ext3_create()                     quota_sync()
2702  *   journal_start()                   write_dquot()
2703  *   vfs_dq_init()                       down(dqio_mutex)
2704  *     down(dqio_mutex)                    journal_start()
2705  *
2706  */
2707
2708 #ifdef CONFIG_QUOTA
2709
2710 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2711 {
2712         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2713 }
2714
2715 static int ext3_write_dquot(struct dquot *dquot)
2716 {
2717         int ret, err;
2718         handle_t *handle;
2719         struct inode *inode;
2720
2721         inode = dquot_to_inode(dquot);
2722         handle = ext3_journal_start(inode,
2723                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2724         if (IS_ERR(handle))
2725                 return PTR_ERR(handle);
2726         ret = dquot_commit(dquot);
2727         err = ext3_journal_stop(handle);
2728         if (!ret)
2729                 ret = err;
2730         return ret;
2731 }
2732
2733 static int ext3_acquire_dquot(struct dquot *dquot)
2734 {
2735         int ret, err;
2736         handle_t *handle;
2737
2738         handle = ext3_journal_start(dquot_to_inode(dquot),
2739                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2740         if (IS_ERR(handle))
2741                 return PTR_ERR(handle);
2742         ret = dquot_acquire(dquot);
2743         err = ext3_journal_stop(handle);
2744         if (!ret)
2745                 ret = err;
2746         return ret;
2747 }
2748
2749 static int ext3_release_dquot(struct dquot *dquot)
2750 {
2751         int ret, err;
2752         handle_t *handle;
2753
2754         handle = ext3_journal_start(dquot_to_inode(dquot),
2755                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2756         if (IS_ERR(handle)) {
2757                 /* Release dquot anyway to avoid endless cycle in dqput() */
2758                 dquot_release(dquot);
2759                 return PTR_ERR(handle);
2760         }
2761         ret = dquot_release(dquot);
2762         err = ext3_journal_stop(handle);
2763         if (!ret)
2764                 ret = err;
2765         return ret;
2766 }
2767
2768 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2769 {
2770         /* Are we journaling quotas? */
2771         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2772             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2773                 dquot_mark_dquot_dirty(dquot);
2774                 return ext3_write_dquot(dquot);
2775         } else {
2776                 return dquot_mark_dquot_dirty(dquot);
2777         }
2778 }
2779
2780 static int ext3_write_info(struct super_block *sb, int type)
2781 {
2782         int ret, err;
2783         handle_t *handle;
2784
2785         /* Data block + inode block */
2786         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2787         if (IS_ERR(handle))
2788                 return PTR_ERR(handle);
2789         ret = dquot_commit_info(sb, type);
2790         err = ext3_journal_stop(handle);
2791         if (!ret)
2792                 ret = err;
2793         return ret;
2794 }
2795
2796 /*
2797  * Turn on quotas during mount time - we need to find
2798  * the quota file and such...
2799  */
2800 static int ext3_quota_on_mount(struct super_block *sb, int type)
2801 {
2802         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2803                         EXT3_SB(sb)->s_jquota_fmt, type);
2804 }
2805
2806 /*
2807  * Standard function to be called on quota_on
2808  */
2809 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2810                          char *name, int remount)
2811 {
2812         int err;
2813         struct path path;
2814
2815         if (!test_opt(sb, QUOTA))
2816                 return -EINVAL;
2817         /* When remounting, no checks are needed and in fact, name is NULL */
2818         if (remount)
2819                 return vfs_quota_on(sb, type, format_id, name, remount);
2820
2821         err = kern_path(name, LOOKUP_FOLLOW, &path);
2822         if (err)
2823                 return err;
2824
2825         /* Quotafile not on the same filesystem? */
2826         if (path.mnt->mnt_sb != sb) {
2827                 path_put(&path);
2828                 return -EXDEV;
2829         }
2830         /* Journaling quota? */
2831         if (EXT3_SB(sb)->s_qf_names[type]) {
2832                 /* Quotafile not of fs root? */
2833                 if (path.dentry->d_parent != sb->s_root)
2834                         printk(KERN_WARNING
2835                                 "EXT3-fs: Quota file not on filesystem root. "
2836                                 "Journaled quota will not work.\n");
2837         }
2838
2839         /*
2840          * When we journal data on quota file, we have to flush journal to see
2841          * all updates to the file when we bypass pagecache...
2842          */
2843         if (ext3_should_journal_data(path.dentry->d_inode)) {
2844                 /*
2845                  * We don't need to lock updates but journal_flush() could
2846                  * otherwise be livelocked...
2847                  */
2848                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2849                 err = journal_flush(EXT3_SB(sb)->s_journal);
2850                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2851                 if (err) {
2852                         path_put(&path);
2853                         return err;
2854                 }
2855         }
2856
2857         err = vfs_quota_on_path(sb, type, format_id, &path);
2858         path_put(&path);
2859         return err;
2860 }
2861
2862 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2863  * acquiring the locks... As quota files are never truncated and quota code
2864  * itself serializes the operations (and noone else should touch the files)
2865  * we don't have to be afraid of races */
2866 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2867                                size_t len, loff_t off)
2868 {
2869         struct inode *inode = sb_dqopt(sb)->files[type];
2870         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2871         int err = 0;
2872         int offset = off & (sb->s_blocksize - 1);
2873         int tocopy;
2874         size_t toread;
2875         struct buffer_head *bh;
2876         loff_t i_size = i_size_read(inode);
2877
2878         if (off > i_size)
2879                 return 0;
2880         if (off+len > i_size)
2881                 len = i_size-off;
2882         toread = len;
2883         while (toread > 0) {
2884                 tocopy = sb->s_blocksize - offset < toread ?
2885                                 sb->s_blocksize - offset : toread;
2886                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2887                 if (err)
2888                         return err;
2889                 if (!bh)        /* A hole? */
2890                         memset(data, 0, tocopy);
2891                 else
2892                         memcpy(data, bh->b_data+offset, tocopy);
2893                 brelse(bh);
2894                 offset = 0;
2895                 toread -= tocopy;
2896                 data += tocopy;
2897                 blk++;
2898         }
2899         return len;
2900 }
2901
2902 /* Write to quotafile (we know the transaction is already started and has
2903  * enough credits) */
2904 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2905                                 const char *data, size_t len, loff_t off)
2906 {
2907         struct inode *inode = sb_dqopt(sb)->files[type];
2908         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2909         int err = 0;
2910         int offset = off & (sb->s_blocksize - 1);
2911         int tocopy;
2912         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2913         size_t towrite = len;
2914         struct buffer_head *bh;
2915         handle_t *handle = journal_current_handle();
2916
2917         if (!handle) {
2918                 printk(KERN_WARNING "EXT3-fs: Quota write (off=%Lu, len=%Lu)"
2919                         " cancelled because transaction is not started.\n",
2920                         (unsigned long long)off, (unsigned long long)len);
2921                 return -EIO;
2922         }
2923         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2924         while (towrite > 0) {
2925                 tocopy = sb->s_blocksize - offset < towrite ?
2926                                 sb->s_blocksize - offset : towrite;
2927                 bh = ext3_bread(handle, inode, blk, 1, &err);
2928                 if (!bh)
2929                         goto out;
2930                 if (journal_quota) {
2931                         err = ext3_journal_get_write_access(handle, bh);
2932                         if (err) {
2933                                 brelse(bh);
2934                                 goto out;
2935                         }
2936                 }
2937                 lock_buffer(bh);
2938                 memcpy(bh->b_data+offset, data, tocopy);
2939                 flush_dcache_page(bh->b_page);
2940                 unlock_buffer(bh);
2941                 if (journal_quota)
2942                         err = ext3_journal_dirty_metadata(handle, bh);
2943                 else {
2944                         /* Always do at least ordered writes for quotas */
2945                         err = ext3_journal_dirty_data(handle, bh);
2946                         mark_buffer_dirty(bh);
2947                 }
2948                 brelse(bh);
2949                 if (err)
2950                         goto out;
2951                 offset = 0;
2952                 towrite -= tocopy;
2953                 data += tocopy;
2954                 blk++;
2955         }
2956 out:
2957         if (len == towrite) {
2958                 mutex_unlock(&inode->i_mutex);
2959                 return err;
2960         }
2961         if (inode->i_size < off+len-towrite) {
2962                 i_size_write(inode, off+len-towrite);
2963                 EXT3_I(inode)->i_disksize = inode->i_size;
2964         }
2965         inode->i_version++;
2966         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2967         ext3_mark_inode_dirty(handle, inode);
2968         mutex_unlock(&inode->i_mutex);
2969         return len - towrite;
2970 }
2971
2972 #endif
2973
2974 static int ext3_get_sb(struct file_system_type *fs_type,
2975         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2976 {
2977         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
2978 }
2979
2980 static struct file_system_type ext3_fs_type = {
2981         .owner          = THIS_MODULE,
2982         .name           = "ext3",
2983         .get_sb         = ext3_get_sb,
2984         .kill_sb        = kill_block_super,
2985         .fs_flags       = FS_REQUIRES_DEV,
2986 };
2987
2988 static int __init init_ext3_fs(void)
2989 {
2990         int err = init_ext3_xattr();
2991         if (err)
2992                 return err;
2993         err = init_inodecache();
2994         if (err)
2995                 goto out1;
2996         err = register_filesystem(&ext3_fs_type);
2997         if (err)
2998                 goto out;
2999         return 0;
3000 out:
3001         destroy_inodecache();
3002 out1:
3003         exit_ext3_xattr();
3004         return err;
3005 }
3006
3007 static void __exit exit_ext3_fs(void)
3008 {
3009         unregister_filesystem(&ext3_fs_type);
3010         destroy_inodecache();
3011         exit_ext3_xattr();
3012 }
3013
3014 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3015 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3016 MODULE_LICENSE("GPL");
3017 module_init(init_ext3_fs)
3018 module_exit(exit_ext3_fs)