Merge branch 'master' of git://git.infradead.org/~dwmw2/solos-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_dquot_initialize(struct inode *inode, int type);
711 static int ext3_dquot_drop(struct inode *inode);
712 static int ext3_write_dquot(struct dquot *dquot);
713 static int ext3_acquire_dquot(struct dquot *dquot);
714 static int ext3_release_dquot(struct dquot *dquot);
715 static int ext3_mark_dquot_dirty(struct dquot *dquot);
716 static int ext3_write_info(struct super_block *sb, int type);
717 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
718                                 char *path, int remount);
719 static int ext3_quota_on_mount(struct super_block *sb, int type);
720 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
721                                size_t len, loff_t off);
722 static ssize_t ext3_quota_write(struct super_block *sb, int type,
723                                 const char *data, size_t len, loff_t off);
724
725 static struct dquot_operations ext3_quota_operations = {
726         .initialize     = ext3_dquot_initialize,
727         .drop           = ext3_dquot_drop,
728         .alloc_space    = dquot_alloc_space,
729         .alloc_inode    = dquot_alloc_inode,
730         .free_space     = dquot_free_space,
731         .free_inode     = dquot_free_inode,
732         .transfer       = dquot_transfer,
733         .write_dquot    = ext3_write_dquot,
734         .acquire_dquot  = ext3_acquire_dquot,
735         .release_dquot  = ext3_release_dquot,
736         .mark_dirty     = ext3_mark_dquot_dirty,
737         .write_info     = ext3_write_info,
738         .alloc_dquot    = dquot_alloc,
739         .destroy_dquot  = dquot_destroy,
740 };
741
742 static struct quotactl_ops ext3_qctl_operations = {
743         .quota_on       = ext3_quota_on,
744         .quota_off      = vfs_quota_off,
745         .quota_sync     = vfs_quota_sync,
746         .get_info       = vfs_get_dqinfo,
747         .set_info       = vfs_set_dqinfo,
748         .get_dqblk      = vfs_get_dqblk,
749         .set_dqblk      = vfs_set_dqblk
750 };
751 #endif
752
753 static const struct super_operations ext3_sops = {
754         .alloc_inode    = ext3_alloc_inode,
755         .destroy_inode  = ext3_destroy_inode,
756         .write_inode    = ext3_write_inode,
757         .dirty_inode    = ext3_dirty_inode,
758         .delete_inode   = ext3_delete_inode,
759         .put_super      = ext3_put_super,
760         .write_super    = ext3_write_super,
761         .sync_fs        = ext3_sync_fs,
762         .freeze_fs      = ext3_freeze,
763         .unfreeze_fs    = ext3_unfreeze,
764         .statfs         = ext3_statfs,
765         .remount_fs     = ext3_remount,
766         .clear_inode    = ext3_clear_inode,
767         .show_options   = ext3_show_options,
768 #ifdef CONFIG_QUOTA
769         .quota_read     = ext3_quota_read,
770         .quota_write    = ext3_quota_write,
771 #endif
772         .bdev_try_to_free_page = bdev_try_to_free_page,
773 };
774
775 static const struct export_operations ext3_export_ops = {
776         .fh_to_dentry = ext3_fh_to_dentry,
777         .fh_to_parent = ext3_fh_to_parent,
778         .get_parent = ext3_get_parent,
779 };
780
781 enum {
782         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
783         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
784         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
785         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
786         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
787         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
788         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
789         Opt_data_err_abort, Opt_data_err_ignore,
790         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
791         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
792         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
793         Opt_grpquota
794 };
795
796 static const match_table_t tokens = {
797         {Opt_bsd_df, "bsddf"},
798         {Opt_minix_df, "minixdf"},
799         {Opt_grpid, "grpid"},
800         {Opt_grpid, "bsdgroups"},
801         {Opt_nogrpid, "nogrpid"},
802         {Opt_nogrpid, "sysvgroups"},
803         {Opt_resgid, "resgid=%u"},
804         {Opt_resuid, "resuid=%u"},
805         {Opt_sb, "sb=%u"},
806         {Opt_err_cont, "errors=continue"},
807         {Opt_err_panic, "errors=panic"},
808         {Opt_err_ro, "errors=remount-ro"},
809         {Opt_nouid32, "nouid32"},
810         {Opt_nocheck, "nocheck"},
811         {Opt_nocheck, "check=none"},
812         {Opt_debug, "debug"},
813         {Opt_oldalloc, "oldalloc"},
814         {Opt_orlov, "orlov"},
815         {Opt_user_xattr, "user_xattr"},
816         {Opt_nouser_xattr, "nouser_xattr"},
817         {Opt_acl, "acl"},
818         {Opt_noacl, "noacl"},
819         {Opt_reservation, "reservation"},
820         {Opt_noreservation, "noreservation"},
821         {Opt_noload, "noload"},
822         {Opt_nobh, "nobh"},
823         {Opt_bh, "bh"},
824         {Opt_commit, "commit=%u"},
825         {Opt_journal_update, "journal=update"},
826         {Opt_journal_inum, "journal=%u"},
827         {Opt_journal_dev, "journal_dev=%u"},
828         {Opt_abort, "abort"},
829         {Opt_data_journal, "data=journal"},
830         {Opt_data_ordered, "data=ordered"},
831         {Opt_data_writeback, "data=writeback"},
832         {Opt_data_err_abort, "data_err=abort"},
833         {Opt_data_err_ignore, "data_err=ignore"},
834         {Opt_offusrjquota, "usrjquota="},
835         {Opt_usrjquota, "usrjquota=%s"},
836         {Opt_offgrpjquota, "grpjquota="},
837         {Opt_grpjquota, "grpjquota=%s"},
838         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
839         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
840         {Opt_grpquota, "grpquota"},
841         {Opt_noquota, "noquota"},
842         {Opt_quota, "quota"},
843         {Opt_usrquota, "usrquota"},
844         {Opt_barrier, "barrier=%u"},
845         {Opt_resize, "resize"},
846         {Opt_err, NULL},
847 };
848
849 static ext3_fsblk_t get_sb_block(void **data)
850 {
851         ext3_fsblk_t    sb_block;
852         char            *options = (char *) *data;
853
854         if (!options || strncmp(options, "sb=", 3) != 0)
855                 return 1;       /* Default location */
856         options += 3;
857         /*todo: use simple_strtoll with >32bit ext3 */
858         sb_block = simple_strtoul(options, &options, 0);
859         if (*options && *options != ',') {
860                 printk("EXT3-fs: Invalid sb specification: %s\n",
861                        (char *) *data);
862                 return 1;
863         }
864         if (*options == ',')
865                 options++;
866         *data = (void *) options;
867         return sb_block;
868 }
869
870 static int parse_options (char *options, struct super_block *sb,
871                           unsigned int *inum, unsigned long *journal_devnum,
872                           ext3_fsblk_t *n_blocks_count, int is_remount)
873 {
874         struct ext3_sb_info *sbi = EXT3_SB(sb);
875         char * p;
876         substring_t args[MAX_OPT_ARGS];
877         int data_opt = 0;
878         int option;
879 #ifdef CONFIG_QUOTA
880         int qtype, qfmt;
881         char *qname;
882 #endif
883
884         if (!options)
885                 return 1;
886
887         while ((p = strsep (&options, ",")) != NULL) {
888                 int token;
889                 if (!*p)
890                         continue;
891
892                 token = match_token(p, tokens, args);
893                 switch (token) {
894                 case Opt_bsd_df:
895                         clear_opt (sbi->s_mount_opt, MINIX_DF);
896                         break;
897                 case Opt_minix_df:
898                         set_opt (sbi->s_mount_opt, MINIX_DF);
899                         break;
900                 case Opt_grpid:
901                         set_opt (sbi->s_mount_opt, GRPID);
902                         break;
903                 case Opt_nogrpid:
904                         clear_opt (sbi->s_mount_opt, GRPID);
905                         break;
906                 case Opt_resuid:
907                         if (match_int(&args[0], &option))
908                                 return 0;
909                         sbi->s_resuid = option;
910                         break;
911                 case Opt_resgid:
912                         if (match_int(&args[0], &option))
913                                 return 0;
914                         sbi->s_resgid = option;
915                         break;
916                 case Opt_sb:
917                         /* handled by get_sb_block() instead of here */
918                         /* *sb_block = match_int(&args[0]); */
919                         break;
920                 case Opt_err_panic:
921                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
922                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
923                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
924                         break;
925                 case Opt_err_ro:
926                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
927                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
928                         set_opt (sbi->s_mount_opt, ERRORS_RO);
929                         break;
930                 case Opt_err_cont:
931                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
932                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
933                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
934                         break;
935                 case Opt_nouid32:
936                         set_opt (sbi->s_mount_opt, NO_UID32);
937                         break;
938                 case Opt_nocheck:
939                         clear_opt (sbi->s_mount_opt, CHECK);
940                         break;
941                 case Opt_debug:
942                         set_opt (sbi->s_mount_opt, DEBUG);
943                         break;
944                 case Opt_oldalloc:
945                         set_opt (sbi->s_mount_opt, OLDALLOC);
946                         break;
947                 case Opt_orlov:
948                         clear_opt (sbi->s_mount_opt, OLDALLOC);
949                         break;
950 #ifdef CONFIG_EXT3_FS_XATTR
951                 case Opt_user_xattr:
952                         set_opt (sbi->s_mount_opt, XATTR_USER);
953                         break;
954                 case Opt_nouser_xattr:
955                         clear_opt (sbi->s_mount_opt, XATTR_USER);
956                         break;
957 #else
958                 case Opt_user_xattr:
959                 case Opt_nouser_xattr:
960                         printk("EXT3 (no)user_xattr options not supported\n");
961                         break;
962 #endif
963 #ifdef CONFIG_EXT3_FS_POSIX_ACL
964                 case Opt_acl:
965                         set_opt(sbi->s_mount_opt, POSIX_ACL);
966                         break;
967                 case Opt_noacl:
968                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
969                         break;
970 #else
971                 case Opt_acl:
972                 case Opt_noacl:
973                         printk("EXT3 (no)acl options not supported\n");
974                         break;
975 #endif
976                 case Opt_reservation:
977                         set_opt(sbi->s_mount_opt, RESERVATION);
978                         break;
979                 case Opt_noreservation:
980                         clear_opt(sbi->s_mount_opt, RESERVATION);
981                         break;
982                 case Opt_journal_update:
983                         /* @@@ FIXME */
984                         /* Eventually we will want to be able to create
985                            a journal file here.  For now, only allow the
986                            user to specify an existing inode to be the
987                            journal file. */
988                         if (is_remount) {
989                                 printk(KERN_ERR "EXT3-fs: cannot specify "
990                                        "journal on remount\n");
991                                 return 0;
992                         }
993                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
994                         break;
995                 case Opt_journal_inum:
996                         if (is_remount) {
997                                 printk(KERN_ERR "EXT3-fs: cannot specify "
998                                        "journal on remount\n");
999                                 return 0;
1000                         }
1001                         if (match_int(&args[0], &option))
1002                                 return 0;
1003                         *inum = option;
1004                         break;
1005                 case Opt_journal_dev:
1006                         if (is_remount) {
1007                                 printk(KERN_ERR "EXT3-fs: cannot specify "
1008                                        "journal on remount\n");
1009                                 return 0;
1010                         }
1011                         if (match_int(&args[0], &option))
1012                                 return 0;
1013                         *journal_devnum = option;
1014                         break;
1015                 case Opt_noload:
1016                         set_opt (sbi->s_mount_opt, NOLOAD);
1017                         break;
1018                 case Opt_commit:
1019                         if (match_int(&args[0], &option))
1020                                 return 0;
1021                         if (option < 0)
1022                                 return 0;
1023                         if (option == 0)
1024                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1025                         sbi->s_commit_interval = HZ * option;
1026                         break;
1027                 case Opt_data_journal:
1028                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1029                         goto datacheck;
1030                 case Opt_data_ordered:
1031                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1032                         goto datacheck;
1033                 case Opt_data_writeback:
1034                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1035                 datacheck:
1036                         if (is_remount) {
1037                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
1038                                                 != data_opt) {
1039                                         printk(KERN_ERR
1040                                                 "EXT3-fs: cannot change data "
1041                                                 "mode on remount\n");
1042                                         return 0;
1043                                 }
1044                         } else {
1045                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
1046                                 sbi->s_mount_opt |= data_opt;
1047                         }
1048                         break;
1049                 case Opt_data_err_abort:
1050                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1051                         break;
1052                 case Opt_data_err_ignore:
1053                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1054                         break;
1055 #ifdef CONFIG_QUOTA
1056                 case Opt_usrjquota:
1057                         qtype = USRQUOTA;
1058                         goto set_qf_name;
1059                 case Opt_grpjquota:
1060                         qtype = GRPQUOTA;
1061 set_qf_name:
1062                         if (sb_any_quota_loaded(sb) &&
1063                             !sbi->s_qf_names[qtype]) {
1064                                 printk(KERN_ERR
1065                                         "EXT3-fs: Cannot change journaled "
1066                                         "quota options when quota turned on.\n");
1067                                 return 0;
1068                         }
1069                         qname = match_strdup(&args[0]);
1070                         if (!qname) {
1071                                 printk(KERN_ERR
1072                                         "EXT3-fs: not enough memory for "
1073                                         "storing quotafile name.\n");
1074                                 return 0;
1075                         }
1076                         if (sbi->s_qf_names[qtype] &&
1077                             strcmp(sbi->s_qf_names[qtype], qname)) {
1078                                 printk(KERN_ERR
1079                                         "EXT3-fs: %s quota file already "
1080                                         "specified.\n", QTYPE2NAME(qtype));
1081                                 kfree(qname);
1082                                 return 0;
1083                         }
1084                         sbi->s_qf_names[qtype] = qname;
1085                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1086                                 printk(KERN_ERR
1087                                         "EXT3-fs: quotafile must be on "
1088                                         "filesystem root.\n");
1089                                 kfree(sbi->s_qf_names[qtype]);
1090                                 sbi->s_qf_names[qtype] = NULL;
1091                                 return 0;
1092                         }
1093                         set_opt(sbi->s_mount_opt, QUOTA);
1094                         break;
1095                 case Opt_offusrjquota:
1096                         qtype = USRQUOTA;
1097                         goto clear_qf_name;
1098                 case Opt_offgrpjquota:
1099                         qtype = GRPQUOTA;
1100 clear_qf_name:
1101                         if (sb_any_quota_loaded(sb) &&
1102                             sbi->s_qf_names[qtype]) {
1103                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1104                                         "journaled quota options when "
1105                                         "quota turned on.\n");
1106                                 return 0;
1107                         }
1108                         /*
1109                          * The space will be released later when all options
1110                          * are confirmed to be correct
1111                          */
1112                         sbi->s_qf_names[qtype] = NULL;
1113                         break;
1114                 case Opt_jqfmt_vfsold:
1115                         qfmt = QFMT_VFS_OLD;
1116                         goto set_qf_format;
1117                 case Opt_jqfmt_vfsv0:
1118                         qfmt = QFMT_VFS_V0;
1119 set_qf_format:
1120                         if (sb_any_quota_loaded(sb) &&
1121                             sbi->s_jquota_fmt != qfmt) {
1122                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1123                                         "journaled quota options when "
1124                                         "quota turned on.\n");
1125                                 return 0;
1126                         }
1127                         sbi->s_jquota_fmt = qfmt;
1128                         break;
1129                 case Opt_quota:
1130                 case Opt_usrquota:
1131                         set_opt(sbi->s_mount_opt, QUOTA);
1132                         set_opt(sbi->s_mount_opt, USRQUOTA);
1133                         break;
1134                 case Opt_grpquota:
1135                         set_opt(sbi->s_mount_opt, QUOTA);
1136                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1137                         break;
1138                 case Opt_noquota:
1139                         if (sb_any_quota_loaded(sb)) {
1140                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
1141                                         "options when quota turned on.\n");
1142                                 return 0;
1143                         }
1144                         clear_opt(sbi->s_mount_opt, QUOTA);
1145                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1146                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1147                         break;
1148 #else
1149                 case Opt_quota:
1150                 case Opt_usrquota:
1151                 case Opt_grpquota:
1152                         printk(KERN_ERR
1153                                 "EXT3-fs: quota options not supported.\n");
1154                         break;
1155                 case Opt_usrjquota:
1156                 case Opt_grpjquota:
1157                 case Opt_offusrjquota:
1158                 case Opt_offgrpjquota:
1159                 case Opt_jqfmt_vfsold:
1160                 case Opt_jqfmt_vfsv0:
1161                         printk(KERN_ERR
1162                                 "EXT3-fs: journaled quota options not "
1163                                 "supported.\n");
1164                         break;
1165                 case Opt_noquota:
1166                         break;
1167 #endif
1168                 case Opt_abort:
1169                         set_opt(sbi->s_mount_opt, ABORT);
1170                         break;
1171                 case Opt_barrier:
1172                         if (match_int(&args[0], &option))
1173                                 return 0;
1174                         if (option)
1175                                 set_opt(sbi->s_mount_opt, BARRIER);
1176                         else
1177                                 clear_opt(sbi->s_mount_opt, BARRIER);
1178                         break;
1179                 case Opt_ignore:
1180                         break;
1181                 case Opt_resize:
1182                         if (!is_remount) {
1183                                 printk("EXT3-fs: resize option only available "
1184                                         "for remount\n");
1185                                 return 0;
1186                         }
1187                         if (match_int(&args[0], &option) != 0)
1188                                 return 0;
1189                         *n_blocks_count = option;
1190                         break;
1191                 case Opt_nobh:
1192                         set_opt(sbi->s_mount_opt, NOBH);
1193                         break;
1194                 case Opt_bh:
1195                         clear_opt(sbi->s_mount_opt, NOBH);
1196                         break;
1197                 default:
1198                         printk (KERN_ERR
1199                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1200                                 "or missing value\n", p);
1201                         return 0;
1202                 }
1203         }
1204 #ifdef CONFIG_QUOTA
1205         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1206                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1207                      sbi->s_qf_names[USRQUOTA])
1208                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1209
1210                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1211                      sbi->s_qf_names[GRPQUOTA])
1212                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1213
1214                 if ((sbi->s_qf_names[USRQUOTA] &&
1215                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1216                     (sbi->s_qf_names[GRPQUOTA] &&
1217                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1218                         printk(KERN_ERR "EXT3-fs: old and new quota "
1219                                         "format mixing.\n");
1220                         return 0;
1221                 }
1222
1223                 if (!sbi->s_jquota_fmt) {
1224                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1225                                         "not specified.\n");
1226                         return 0;
1227                 }
1228         } else {
1229                 if (sbi->s_jquota_fmt) {
1230                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1231                                         "specified with no journaling "
1232                                         "enabled.\n");
1233                         return 0;
1234                 }
1235         }
1236 #endif
1237         return 1;
1238 }
1239
1240 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1241                             int read_only)
1242 {
1243         struct ext3_sb_info *sbi = EXT3_SB(sb);
1244         int res = 0;
1245
1246         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1247                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1248                         "forcing read-only mode\n");
1249                 res = MS_RDONLY;
1250         }
1251         if (read_only)
1252                 return res;
1253         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1254                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1255                         "running e2fsck is recommended\n");
1256         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1257                 printk (KERN_WARNING
1258                         "EXT3-fs warning: mounting fs with errors, "
1259                         "running e2fsck is recommended\n");
1260         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1261                  le16_to_cpu(es->s_mnt_count) >=
1262                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1263                 printk (KERN_WARNING
1264                         "EXT3-fs warning: maximal mount count reached, "
1265                         "running e2fsck is recommended\n");
1266         else if (le32_to_cpu(es->s_checkinterval) &&
1267                 (le32_to_cpu(es->s_lastcheck) +
1268                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1269                 printk (KERN_WARNING
1270                         "EXT3-fs warning: checktime reached, "
1271                         "running e2fsck is recommended\n");
1272 #if 0
1273                 /* @@@ We _will_ want to clear the valid bit if we find
1274                    inconsistencies, to force a fsck at reboot.  But for
1275                    a plain journaled filesystem we can keep it set as
1276                    valid forever! :) */
1277         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1278 #endif
1279         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1280                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1281         le16_add_cpu(&es->s_mnt_count, 1);
1282         es->s_mtime = cpu_to_le32(get_seconds());
1283         ext3_update_dynamic_rev(sb);
1284         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1285
1286         ext3_commit_super(sb, es, 1);
1287         if (test_opt(sb, DEBUG))
1288                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1289                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1290                         sb->s_blocksize,
1291                         sbi->s_groups_count,
1292                         EXT3_BLOCKS_PER_GROUP(sb),
1293                         EXT3_INODES_PER_GROUP(sb),
1294                         sbi->s_mount_opt);
1295
1296         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1297         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1298                 char b[BDEVNAME_SIZE];
1299
1300                 printk("external journal on %s\n",
1301                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1302         } else {
1303                 printk("internal journal\n");
1304         }
1305         return res;
1306 }
1307
1308 /* Called at mount-time, super-block is locked */
1309 static int ext3_check_descriptors(struct super_block *sb)
1310 {
1311         struct ext3_sb_info *sbi = EXT3_SB(sb);
1312         int i;
1313
1314         ext3_debug ("Checking group descriptors");
1315
1316         for (i = 0; i < sbi->s_groups_count; i++) {
1317                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1318                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1319                 ext3_fsblk_t last_block;
1320
1321                 if (i == sbi->s_groups_count - 1)
1322                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1323                 else
1324                         last_block = first_block +
1325                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1326
1327                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1328                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1329                 {
1330                         ext3_error (sb, "ext3_check_descriptors",
1331                                     "Block bitmap for group %d"
1332                                     " not in group (block %lu)!",
1333                                     i, (unsigned long)
1334                                         le32_to_cpu(gdp->bg_block_bitmap));
1335                         return 0;
1336                 }
1337                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1338                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1339                 {
1340                         ext3_error (sb, "ext3_check_descriptors",
1341                                     "Inode bitmap for group %d"
1342                                     " not in group (block %lu)!",
1343                                     i, (unsigned long)
1344                                         le32_to_cpu(gdp->bg_inode_bitmap));
1345                         return 0;
1346                 }
1347                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1348                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1349                     last_block)
1350                 {
1351                         ext3_error (sb, "ext3_check_descriptors",
1352                                     "Inode table for group %d"
1353                                     " not in group (block %lu)!",
1354                                     i, (unsigned long)
1355                                         le32_to_cpu(gdp->bg_inode_table));
1356                         return 0;
1357                 }
1358         }
1359
1360         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1361         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1362         return 1;
1363 }
1364
1365
1366 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1367  * the superblock) which were deleted from all directories, but held open by
1368  * a process at the time of a crash.  We walk the list and try to delete these
1369  * inodes at recovery time (only with a read-write filesystem).
1370  *
1371  * In order to keep the orphan inode chain consistent during traversal (in
1372  * case of crash during recovery), we link each inode into the superblock
1373  * orphan list_head and handle it the same way as an inode deletion during
1374  * normal operation (which journals the operations for us).
1375  *
1376  * We only do an iget() and an iput() on each inode, which is very safe if we
1377  * accidentally point at an in-use or already deleted inode.  The worst that
1378  * can happen in this case is that we get a "bit already cleared" message from
1379  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1380  * e2fsck was run on this filesystem, and it must have already done the orphan
1381  * inode cleanup for us, so we can safely abort without any further action.
1382  */
1383 static void ext3_orphan_cleanup (struct super_block * sb,
1384                                  struct ext3_super_block * es)
1385 {
1386         unsigned int s_flags = sb->s_flags;
1387         int nr_orphans = 0, nr_truncates = 0;
1388 #ifdef CONFIG_QUOTA
1389         int i;
1390 #endif
1391         if (!es->s_last_orphan) {
1392                 jbd_debug(4, "no orphan inodes to clean up\n");
1393                 return;
1394         }
1395
1396         if (bdev_read_only(sb->s_bdev)) {
1397                 printk(KERN_ERR "EXT3-fs: write access "
1398                         "unavailable, skipping orphan cleanup.\n");
1399                 return;
1400         }
1401
1402         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1403                 if (es->s_last_orphan)
1404                         jbd_debug(1, "Errors on filesystem, "
1405                                   "clearing orphan list.\n");
1406                 es->s_last_orphan = 0;
1407                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1408                 return;
1409         }
1410
1411         if (s_flags & MS_RDONLY) {
1412                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1413                        sb->s_id);
1414                 sb->s_flags &= ~MS_RDONLY;
1415         }
1416 #ifdef CONFIG_QUOTA
1417         /* Needed for iput() to work correctly and not trash data */
1418         sb->s_flags |= MS_ACTIVE;
1419         /* Turn on quotas so that they are updated correctly */
1420         for (i = 0; i < MAXQUOTAS; i++) {
1421                 if (EXT3_SB(sb)->s_qf_names[i]) {
1422                         int ret = ext3_quota_on_mount(sb, i);
1423                         if (ret < 0)
1424                                 printk(KERN_ERR
1425                                         "EXT3-fs: Cannot turn on journaled "
1426                                         "quota: error %d\n", ret);
1427                 }
1428         }
1429 #endif
1430
1431         while (es->s_last_orphan) {
1432                 struct inode *inode;
1433
1434                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1435                 if (IS_ERR(inode)) {
1436                         es->s_last_orphan = 0;
1437                         break;
1438                 }
1439
1440                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1441                 DQUOT_INIT(inode);
1442                 if (inode->i_nlink) {
1443                         printk(KERN_DEBUG
1444                                 "%s: truncating inode %lu to %Ld bytes\n",
1445                                 __func__, inode->i_ino, inode->i_size);
1446                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1447                                   inode->i_ino, inode->i_size);
1448                         ext3_truncate(inode);
1449                         nr_truncates++;
1450                 } else {
1451                         printk(KERN_DEBUG
1452                                 "%s: deleting unreferenced inode %lu\n",
1453                                 __func__, inode->i_ino);
1454                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1455                                   inode->i_ino);
1456                         nr_orphans++;
1457                 }
1458                 iput(inode);  /* The delete magic happens here! */
1459         }
1460
1461 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1462
1463         if (nr_orphans)
1464                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1465                        sb->s_id, PLURAL(nr_orphans));
1466         if (nr_truncates)
1467                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1468                        sb->s_id, PLURAL(nr_truncates));
1469 #ifdef CONFIG_QUOTA
1470         /* Turn quotas off */
1471         for (i = 0; i < MAXQUOTAS; i++) {
1472                 if (sb_dqopt(sb)->files[i])
1473                         vfs_quota_off(sb, i, 0);
1474         }
1475 #endif
1476         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1477 }
1478
1479 /*
1480  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1481  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1482  * We need to be 1 filesystem block less than the 2^32 sector limit.
1483  */
1484 static loff_t ext3_max_size(int bits)
1485 {
1486         loff_t res = EXT3_NDIR_BLOCKS;
1487         int meta_blocks;
1488         loff_t upper_limit;
1489
1490         /* This is calculated to be the largest file size for a
1491          * dense, file such that the total number of
1492          * sectors in the file, including data and all indirect blocks,
1493          * does not exceed 2^32 -1
1494          * __u32 i_blocks representing the total number of
1495          * 512 bytes blocks of the file
1496          */
1497         upper_limit = (1LL << 32) - 1;
1498
1499         /* total blocks in file system block size */
1500         upper_limit >>= (bits - 9);
1501
1502
1503         /* indirect blocks */
1504         meta_blocks = 1;
1505         /* double indirect blocks */
1506         meta_blocks += 1 + (1LL << (bits-2));
1507         /* tripple indirect blocks */
1508         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1509
1510         upper_limit -= meta_blocks;
1511         upper_limit <<= bits;
1512
1513         res += 1LL << (bits-2);
1514         res += 1LL << (2*(bits-2));
1515         res += 1LL << (3*(bits-2));
1516         res <<= bits;
1517         if (res > upper_limit)
1518                 res = upper_limit;
1519
1520         if (res > MAX_LFS_FILESIZE)
1521                 res = MAX_LFS_FILESIZE;
1522
1523         return res;
1524 }
1525
1526 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1527                                     ext3_fsblk_t logic_sb_block,
1528                                     int nr)
1529 {
1530         struct ext3_sb_info *sbi = EXT3_SB(sb);
1531         unsigned long bg, first_meta_bg;
1532         int has_super = 0;
1533
1534         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1535
1536         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1537             nr < first_meta_bg)
1538                 return (logic_sb_block + nr + 1);
1539         bg = sbi->s_desc_per_block * nr;
1540         if (ext3_bg_has_super(sb, bg))
1541                 has_super = 1;
1542         return (has_super + ext3_group_first_block_no(sb, bg));
1543 }
1544
1545
1546 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1547 {
1548         struct buffer_head * bh;
1549         struct ext3_super_block *es = NULL;
1550         struct ext3_sb_info *sbi;
1551         ext3_fsblk_t block;
1552         ext3_fsblk_t sb_block = get_sb_block(&data);
1553         ext3_fsblk_t logic_sb_block;
1554         unsigned long offset = 0;
1555         unsigned int journal_inum = 0;
1556         unsigned long journal_devnum = 0;
1557         unsigned long def_mount_opts;
1558         struct inode *root;
1559         int blocksize;
1560         int hblock;
1561         int db_count;
1562         int i;
1563         int needs_recovery;
1564         int ret = -EINVAL;
1565         __le32 features;
1566         int err;
1567
1568         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1569         if (!sbi)
1570                 return -ENOMEM;
1571
1572         sbi->s_blockgroup_lock =
1573                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1574         if (!sbi->s_blockgroup_lock) {
1575                 kfree(sbi);
1576                 return -ENOMEM;
1577         }
1578         sb->s_fs_info = sbi;
1579         sbi->s_mount_opt = 0;
1580         sbi->s_resuid = EXT3_DEF_RESUID;
1581         sbi->s_resgid = EXT3_DEF_RESGID;
1582         sbi->s_sb_block = sb_block;
1583
1584         unlock_kernel();
1585
1586         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1587         if (!blocksize) {
1588                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1589                 goto out_fail;
1590         }
1591
1592         /*
1593          * The ext3 superblock will not be buffer aligned for other than 1kB
1594          * block sizes.  We need to calculate the offset from buffer start.
1595          */
1596         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1597                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1598                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1599         } else {
1600                 logic_sb_block = sb_block;
1601         }
1602
1603         if (!(bh = sb_bread(sb, logic_sb_block))) {
1604                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1605                 goto out_fail;
1606         }
1607         /*
1608          * Note: s_es must be initialized as soon as possible because
1609          *       some ext3 macro-instructions depend on its value
1610          */
1611         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1612         sbi->s_es = es;
1613         sb->s_magic = le16_to_cpu(es->s_magic);
1614         if (sb->s_magic != EXT3_SUPER_MAGIC)
1615                 goto cantfind_ext3;
1616
1617         /* Set defaults before we parse the mount options */
1618         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1619         if (def_mount_opts & EXT3_DEFM_DEBUG)
1620                 set_opt(sbi->s_mount_opt, DEBUG);
1621         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1622                 set_opt(sbi->s_mount_opt, GRPID);
1623         if (def_mount_opts & EXT3_DEFM_UID16)
1624                 set_opt(sbi->s_mount_opt, NO_UID32);
1625 #ifdef CONFIG_EXT3_FS_XATTR
1626         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1627                 set_opt(sbi->s_mount_opt, XATTR_USER);
1628 #endif
1629 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1630         if (def_mount_opts & EXT3_DEFM_ACL)
1631                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1632 #endif
1633         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1634                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1635         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1636                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1637         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1638                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1639
1640         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1641                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1642         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1643                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1644         else
1645                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1646
1647         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1648         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1649
1650         set_opt(sbi->s_mount_opt, RESERVATION);
1651
1652         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1653                             NULL, 0))
1654                 goto failed_mount;
1655
1656         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1657                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1658
1659         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1660             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1661              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1662              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1663                 printk(KERN_WARNING
1664                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1665                        "running e2fsck is recommended\n");
1666         /*
1667          * Check feature flags regardless of the revision level, since we
1668          * previously didn't change the revision level when setting the flags,
1669          * so there is a chance incompat flags are set on a rev 0 filesystem.
1670          */
1671         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1672         if (features) {
1673                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1674                        "unsupported optional features (%x).\n",
1675                        sb->s_id, le32_to_cpu(features));
1676                 goto failed_mount;
1677         }
1678         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1679         if (!(sb->s_flags & MS_RDONLY) && features) {
1680                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1681                        "unsupported optional features (%x).\n",
1682                        sb->s_id, le32_to_cpu(features));
1683                 goto failed_mount;
1684         }
1685         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1686
1687         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1688             blocksize > EXT3_MAX_BLOCK_SIZE) {
1689                 printk(KERN_ERR
1690                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1691                        blocksize, sb->s_id);
1692                 goto failed_mount;
1693         }
1694
1695         hblock = bdev_hardsect_size(sb->s_bdev);
1696         if (sb->s_blocksize != blocksize) {
1697                 /*
1698                  * Make sure the blocksize for the filesystem is larger
1699                  * than the hardware sectorsize for the machine.
1700                  */
1701                 if (blocksize < hblock) {
1702                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1703                                "device blocksize %d.\n", blocksize, hblock);
1704                         goto failed_mount;
1705                 }
1706
1707                 brelse (bh);
1708                 if (!sb_set_blocksize(sb, blocksize)) {
1709                         printk(KERN_ERR "EXT3-fs: bad blocksize %d.\n",
1710                                 blocksize);
1711                         goto out_fail;
1712                 }
1713                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1714                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1715                 bh = sb_bread(sb, logic_sb_block);
1716                 if (!bh) {
1717                         printk(KERN_ERR
1718                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1719                         goto failed_mount;
1720                 }
1721                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1722                 sbi->s_es = es;
1723                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1724                         printk (KERN_ERR
1725                                 "EXT3-fs: Magic mismatch, very weird !\n");
1726                         goto failed_mount;
1727                 }
1728         }
1729
1730         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1731
1732         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1733                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1734                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1735         } else {
1736                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1737                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1738                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1739                     (!is_power_of_2(sbi->s_inode_size)) ||
1740                     (sbi->s_inode_size > blocksize)) {
1741                         printk (KERN_ERR
1742                                 "EXT3-fs: unsupported inode size: %d\n",
1743                                 sbi->s_inode_size);
1744                         goto failed_mount;
1745                 }
1746         }
1747         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1748                                    le32_to_cpu(es->s_log_frag_size);
1749         if (blocksize != sbi->s_frag_size) {
1750                 printk(KERN_ERR
1751                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1752                        sbi->s_frag_size, blocksize);
1753                 goto failed_mount;
1754         }
1755         sbi->s_frags_per_block = 1;
1756         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1757         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1758         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1759         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1760                 goto cantfind_ext3;
1761         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1762         if (sbi->s_inodes_per_block == 0)
1763                 goto cantfind_ext3;
1764         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1765                                         sbi->s_inodes_per_block;
1766         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1767         sbi->s_sbh = bh;
1768         sbi->s_mount_state = le16_to_cpu(es->s_state);
1769         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1770         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1771         for (i=0; i < 4; i++)
1772                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1773         sbi->s_def_hash_version = es->s_def_hash_version;
1774         i = le32_to_cpu(es->s_flags);
1775         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1776                 sbi->s_hash_unsigned = 3;
1777         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1778 #ifdef __CHAR_UNSIGNED__
1779                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1780                 sbi->s_hash_unsigned = 3;
1781 #else
1782                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1783 #endif
1784                 sb->s_dirt = 1;
1785         }
1786
1787         if (sbi->s_blocks_per_group > blocksize * 8) {
1788                 printk (KERN_ERR
1789                         "EXT3-fs: #blocks per group too big: %lu\n",
1790                         sbi->s_blocks_per_group);
1791                 goto failed_mount;
1792         }
1793         if (sbi->s_frags_per_group > blocksize * 8) {
1794                 printk (KERN_ERR
1795                         "EXT3-fs: #fragments per group too big: %lu\n",
1796                         sbi->s_frags_per_group);
1797                 goto failed_mount;
1798         }
1799         if (sbi->s_inodes_per_group > blocksize * 8) {
1800                 printk (KERN_ERR
1801                         "EXT3-fs: #inodes per group too big: %lu\n",
1802                         sbi->s_inodes_per_group);
1803                 goto failed_mount;
1804         }
1805
1806         if (le32_to_cpu(es->s_blocks_count) >
1807                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1808                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1809                         " too large to mount safely\n", sb->s_id);
1810                 if (sizeof(sector_t) < 8)
1811                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1812                                         "enabled\n");
1813                 goto failed_mount;
1814         }
1815
1816         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1817                 goto cantfind_ext3;
1818         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1819                                le32_to_cpu(es->s_first_data_block) - 1)
1820                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1821         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1822                    EXT3_DESC_PER_BLOCK(sb);
1823         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1824                                     GFP_KERNEL);
1825         if (sbi->s_group_desc == NULL) {
1826                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1827                 goto failed_mount;
1828         }
1829
1830         bgl_lock_init(sbi->s_blockgroup_lock);
1831
1832         for (i = 0; i < db_count; i++) {
1833                 block = descriptor_loc(sb, logic_sb_block, i);
1834                 sbi->s_group_desc[i] = sb_bread(sb, block);
1835                 if (!sbi->s_group_desc[i]) {
1836                         printk (KERN_ERR "EXT3-fs: "
1837                                 "can't read group descriptor %d\n", i);
1838                         db_count = i;
1839                         goto failed_mount2;
1840                 }
1841         }
1842         if (!ext3_check_descriptors (sb)) {
1843                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1844                 goto failed_mount2;
1845         }
1846         sbi->s_gdb_count = db_count;
1847         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1848         spin_lock_init(&sbi->s_next_gen_lock);
1849
1850         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1851                         ext3_count_free_blocks(sb));
1852         if (!err) {
1853                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1854                                 ext3_count_free_inodes(sb));
1855         }
1856         if (!err) {
1857                 err = percpu_counter_init(&sbi->s_dirs_counter,
1858                                 ext3_count_dirs(sb));
1859         }
1860         if (err) {
1861                 printk(KERN_ERR "EXT3-fs: insufficient memory\n");
1862                 goto failed_mount3;
1863         }
1864
1865         /* per fileystem reservation list head & lock */
1866         spin_lock_init(&sbi->s_rsv_window_lock);
1867         sbi->s_rsv_window_root = RB_ROOT;
1868         /* Add a single, static dummy reservation to the start of the
1869          * reservation window list --- it gives us a placeholder for
1870          * append-at-start-of-list which makes the allocation logic
1871          * _much_ simpler. */
1872         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1873         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1874         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1875         sbi->s_rsv_window_head.rsv_goal_size = 0;
1876         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1877
1878         /*
1879          * set up enough so that it can read an inode
1880          */
1881         sb->s_op = &ext3_sops;
1882         sb->s_export_op = &ext3_export_ops;
1883         sb->s_xattr = ext3_xattr_handlers;
1884 #ifdef CONFIG_QUOTA
1885         sb->s_qcop = &ext3_qctl_operations;
1886         sb->dq_op = &ext3_quota_operations;
1887 #endif
1888         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1889
1890         sb->s_root = NULL;
1891
1892         needs_recovery = (es->s_last_orphan != 0 ||
1893                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1894                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1895
1896         /*
1897          * The first inode we look at is the journal inode.  Don't try
1898          * root first: it may be modified in the journal!
1899          */
1900         if (!test_opt(sb, NOLOAD) &&
1901             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1902                 if (ext3_load_journal(sb, es, journal_devnum))
1903                         goto failed_mount3;
1904         } else if (journal_inum) {
1905                 if (ext3_create_journal(sb, es, journal_inum))
1906                         goto failed_mount3;
1907         } else {
1908                 if (!silent)
1909                         printk (KERN_ERR
1910                                 "ext3: No journal on filesystem on %s\n",
1911                                 sb->s_id);
1912                 goto failed_mount3;
1913         }
1914
1915         /* We have now updated the journal if required, so we can
1916          * validate the data journaling mode. */
1917         switch (test_opt(sb, DATA_FLAGS)) {
1918         case 0:
1919                 /* No mode set, assume a default based on the journal
1920                    capabilities: ORDERED_DATA if the journal can
1921                    cope, else JOURNAL_DATA */
1922                 if (journal_check_available_features
1923                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1924                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1925                 else
1926                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1927                 break;
1928
1929         case EXT3_MOUNT_ORDERED_DATA:
1930         case EXT3_MOUNT_WRITEBACK_DATA:
1931                 if (!journal_check_available_features
1932                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1933                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1934                                "requested data journaling mode\n");
1935                         goto failed_mount4;
1936                 }
1937         default:
1938                 break;
1939         }
1940
1941         if (test_opt(sb, NOBH)) {
1942                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1943                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1944                                 "its supported only with writeback mode\n");
1945                         clear_opt(sbi->s_mount_opt, NOBH);
1946                 }
1947         }
1948         /*
1949          * The journal_load will have done any necessary log recovery,
1950          * so we can safely mount the rest of the filesystem now.
1951          */
1952
1953         root = ext3_iget(sb, EXT3_ROOT_INO);
1954         if (IS_ERR(root)) {
1955                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1956                 ret = PTR_ERR(root);
1957                 goto failed_mount4;
1958         }
1959         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1960                 iput(root);
1961                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1962                 goto failed_mount4;
1963         }
1964         sb->s_root = d_alloc_root(root);
1965         if (!sb->s_root) {
1966                 printk(KERN_ERR "EXT3-fs: get root dentry failed\n");
1967                 iput(root);
1968                 ret = -ENOMEM;
1969                 goto failed_mount4;
1970         }
1971
1972         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1973         /*
1974          * akpm: core read_super() calls in here with the superblock locked.
1975          * That deadlocks, because orphan cleanup needs to lock the superblock
1976          * in numerous places.  Here we just pop the lock - it's relatively
1977          * harmless, because we are now ready to accept write_super() requests,
1978          * and aviro says that's the only reason for hanging onto the
1979          * superblock lock.
1980          */
1981         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1982         ext3_orphan_cleanup(sb, es);
1983         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1984         if (needs_recovery)
1985                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1986         ext3_mark_recovery_complete(sb, es);
1987         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1988                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1989                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1990                 "writeback");
1991
1992         lock_kernel();
1993         return 0;
1994
1995 cantfind_ext3:
1996         if (!silent)
1997                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1998                        sb->s_id);
1999         goto failed_mount;
2000
2001 failed_mount4:
2002         journal_destroy(sbi->s_journal);
2003 failed_mount3:
2004         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2005         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2006         percpu_counter_destroy(&sbi->s_dirs_counter);
2007 failed_mount2:
2008         for (i = 0; i < db_count; i++)
2009                 brelse(sbi->s_group_desc[i]);
2010         kfree(sbi->s_group_desc);
2011 failed_mount:
2012 #ifdef CONFIG_QUOTA
2013         for (i = 0; i < MAXQUOTAS; i++)
2014                 kfree(sbi->s_qf_names[i]);
2015 #endif
2016         ext3_blkdev_remove(sbi);
2017         brelse(bh);
2018 out_fail:
2019         sb->s_fs_info = NULL;
2020         kfree(sbi);
2021         lock_kernel();
2022         return ret;
2023 }
2024
2025 /*
2026  * Setup any per-fs journal parameters now.  We'll do this both on
2027  * initial mount, once the journal has been initialised but before we've
2028  * done any recovery; and again on any subsequent remount.
2029  */
2030 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2031 {
2032         struct ext3_sb_info *sbi = EXT3_SB(sb);
2033
2034         if (sbi->s_commit_interval)
2035                 journal->j_commit_interval = sbi->s_commit_interval;
2036         /* We could also set up an ext3-specific default for the commit
2037          * interval here, but for now we'll just fall back to the jbd
2038          * default. */
2039
2040         spin_lock(&journal->j_state_lock);
2041         if (test_opt(sb, BARRIER))
2042                 journal->j_flags |= JFS_BARRIER;
2043         else
2044                 journal->j_flags &= ~JFS_BARRIER;
2045         if (test_opt(sb, DATA_ERR_ABORT))
2046                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2047         else
2048                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2049         spin_unlock(&journal->j_state_lock);
2050 }
2051
2052 static journal_t *ext3_get_journal(struct super_block *sb,
2053                                    unsigned int journal_inum)
2054 {
2055         struct inode *journal_inode;
2056         journal_t *journal;
2057
2058         /* First, test for the existence of a valid inode on disk.  Bad
2059          * things happen if we iget() an unused inode, as the subsequent
2060          * iput() will try to delete it. */
2061
2062         journal_inode = ext3_iget(sb, journal_inum);
2063         if (IS_ERR(journal_inode)) {
2064                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
2065                 return NULL;
2066         }
2067         if (!journal_inode->i_nlink) {
2068                 make_bad_inode(journal_inode);
2069                 iput(journal_inode);
2070                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
2071                 return NULL;
2072         }
2073
2074         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2075                   journal_inode, journal_inode->i_size);
2076         if (!S_ISREG(journal_inode->i_mode)) {
2077                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
2078                 iput(journal_inode);
2079                 return NULL;
2080         }
2081
2082         journal = journal_init_inode(journal_inode);
2083         if (!journal) {
2084                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
2085                 iput(journal_inode);
2086                 return NULL;
2087         }
2088         journal->j_private = sb;
2089         ext3_init_journal_params(sb, journal);
2090         return journal;
2091 }
2092
2093 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2094                                        dev_t j_dev)
2095 {
2096         struct buffer_head * bh;
2097         journal_t *journal;
2098         ext3_fsblk_t start;
2099         ext3_fsblk_t len;
2100         int hblock, blocksize;
2101         ext3_fsblk_t sb_block;
2102         unsigned long offset;
2103         struct ext3_super_block * es;
2104         struct block_device *bdev;
2105
2106         bdev = ext3_blkdev_get(j_dev);
2107         if (bdev == NULL)
2108                 return NULL;
2109
2110         if (bd_claim(bdev, sb)) {
2111                 printk(KERN_ERR
2112                         "EXT3: failed to claim external journal device.\n");
2113                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2114                 return NULL;
2115         }
2116
2117         blocksize = sb->s_blocksize;
2118         hblock = bdev_hardsect_size(bdev);
2119         if (blocksize < hblock) {
2120                 printk(KERN_ERR
2121                         "EXT3-fs: blocksize too small for journal device.\n");
2122                 goto out_bdev;
2123         }
2124
2125         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2126         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2127         set_blocksize(bdev, blocksize);
2128         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2129                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
2130                        "external journal\n");
2131                 goto out_bdev;
2132         }
2133
2134         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2135         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2136             !(le32_to_cpu(es->s_feature_incompat) &
2137               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2138                 printk(KERN_ERR "EXT3-fs: external journal has "
2139                                         "bad superblock\n");
2140                 brelse(bh);
2141                 goto out_bdev;
2142         }
2143
2144         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2145                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
2146                 brelse(bh);
2147                 goto out_bdev;
2148         }
2149
2150         len = le32_to_cpu(es->s_blocks_count);
2151         start = sb_block + 1;
2152         brelse(bh);     /* we're done with the superblock */
2153
2154         journal = journal_init_dev(bdev, sb->s_bdev,
2155                                         start, len, blocksize);
2156         if (!journal) {
2157                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
2158                 goto out_bdev;
2159         }
2160         journal->j_private = sb;
2161         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2162         wait_on_buffer(journal->j_sb_buffer);
2163         if (!buffer_uptodate(journal->j_sb_buffer)) {
2164                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
2165                 goto out_journal;
2166         }
2167         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2168                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
2169                                         "user (unsupported) - %d\n",
2170                         be32_to_cpu(journal->j_superblock->s_nr_users));
2171                 goto out_journal;
2172         }
2173         EXT3_SB(sb)->journal_bdev = bdev;
2174         ext3_init_journal_params(sb, journal);
2175         return journal;
2176 out_journal:
2177         journal_destroy(journal);
2178 out_bdev:
2179         ext3_blkdev_put(bdev);
2180         return NULL;
2181 }
2182
2183 static int ext3_load_journal(struct super_block *sb,
2184                              struct ext3_super_block *es,
2185                              unsigned long journal_devnum)
2186 {
2187         journal_t *journal;
2188         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2189         dev_t journal_dev;
2190         int err = 0;
2191         int really_read_only;
2192
2193         if (journal_devnum &&
2194             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2195                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
2196                         "numbers have changed\n");
2197                 journal_dev = new_decode_dev(journal_devnum);
2198         } else
2199                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2200
2201         really_read_only = bdev_read_only(sb->s_bdev);
2202
2203         /*
2204          * Are we loading a blank journal or performing recovery after a
2205          * crash?  For recovery, we need to check in advance whether we
2206          * can get read-write access to the device.
2207          */
2208
2209         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2210                 if (sb->s_flags & MS_RDONLY) {
2211                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
2212                                         "required on readonly filesystem.\n");
2213                         if (really_read_only) {
2214                                 printk(KERN_ERR "EXT3-fs: write access "
2215                                         "unavailable, cannot proceed.\n");
2216                                 return -EROFS;
2217                         }
2218                         printk (KERN_INFO "EXT3-fs: write access will "
2219                                         "be enabled during recovery.\n");
2220                 }
2221         }
2222
2223         if (journal_inum && journal_dev) {
2224                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
2225                        "and inode journals!\n");
2226                 return -EINVAL;
2227         }
2228
2229         if (journal_inum) {
2230                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2231                         return -EINVAL;
2232         } else {
2233                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2234                         return -EINVAL;
2235         }
2236
2237         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2238                 err = journal_update_format(journal);
2239                 if (err)  {
2240                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
2241                         journal_destroy(journal);
2242                         return err;
2243                 }
2244         }
2245
2246         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2247                 err = journal_wipe(journal, !really_read_only);
2248         if (!err)
2249                 err = journal_load(journal);
2250
2251         if (err) {
2252                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2253                 journal_destroy(journal);
2254                 return err;
2255         }
2256
2257         EXT3_SB(sb)->s_journal = journal;
2258         ext3_clear_journal_err(sb, es);
2259
2260         if (journal_devnum &&
2261             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2262                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2263                 sb->s_dirt = 1;
2264
2265                 /* Make sure we flush the recovery flag to disk. */
2266                 ext3_commit_super(sb, es, 1);
2267         }
2268
2269         return 0;
2270 }
2271
2272 static int ext3_create_journal(struct super_block * sb,
2273                                struct ext3_super_block * es,
2274                                unsigned int journal_inum)
2275 {
2276         journal_t *journal;
2277         int err;
2278
2279         if (sb->s_flags & MS_RDONLY) {
2280                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2281                                 "create journal.\n");
2282                 return -EROFS;
2283         }
2284
2285         journal = ext3_get_journal(sb, journal_inum);
2286         if (!journal)
2287                 return -EINVAL;
2288
2289         printk(KERN_INFO "EXT3-fs: creating new journal on inode %u\n",
2290                journal_inum);
2291
2292         err = journal_create(journal);
2293         if (err) {
2294                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2295                 journal_destroy(journal);
2296                 return -EIO;
2297         }
2298
2299         EXT3_SB(sb)->s_journal = journal;
2300
2301         ext3_update_dynamic_rev(sb);
2302         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2303         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2304
2305         es->s_journal_inum = cpu_to_le32(journal_inum);
2306         sb->s_dirt = 1;
2307
2308         /* Make sure we flush the recovery flag to disk. */
2309         ext3_commit_super(sb, es, 1);
2310
2311         return 0;
2312 }
2313
2314 static int ext3_commit_super(struct super_block *sb,
2315                                struct ext3_super_block *es,
2316                                int sync)
2317 {
2318         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2319         int error = 0;
2320
2321         if (!sbh)
2322                 return error;
2323         es->s_wtime = cpu_to_le32(get_seconds());
2324         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2325         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2326         BUFFER_TRACE(sbh, "marking dirty");
2327         mark_buffer_dirty(sbh);
2328         if (sync)
2329                 error = sync_dirty_buffer(sbh);
2330         return error;
2331 }
2332
2333
2334 /*
2335  * Have we just finished recovery?  If so, and if we are mounting (or
2336  * remounting) the filesystem readonly, then we will end up with a
2337  * consistent fs on disk.  Record that fact.
2338  */
2339 static void ext3_mark_recovery_complete(struct super_block * sb,
2340                                         struct ext3_super_block * es)
2341 {
2342         journal_t *journal = EXT3_SB(sb)->s_journal;
2343
2344         journal_lock_updates(journal);
2345         if (journal_flush(journal) < 0)
2346                 goto out;
2347
2348         lock_super(sb);
2349         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2350             sb->s_flags & MS_RDONLY) {
2351                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2352                 sb->s_dirt = 0;
2353                 ext3_commit_super(sb, es, 1);
2354         }
2355         unlock_super(sb);
2356
2357 out:
2358         journal_unlock_updates(journal);
2359 }
2360
2361 /*
2362  * If we are mounting (or read-write remounting) a filesystem whose journal
2363  * has recorded an error from a previous lifetime, move that error to the
2364  * main filesystem now.
2365  */
2366 static void ext3_clear_journal_err(struct super_block * sb,
2367                                    struct ext3_super_block * es)
2368 {
2369         journal_t *journal;
2370         int j_errno;
2371         const char *errstr;
2372
2373         journal = EXT3_SB(sb)->s_journal;
2374
2375         /*
2376          * Now check for any error status which may have been recorded in the
2377          * journal by a prior ext3_error() or ext3_abort()
2378          */
2379
2380         j_errno = journal_errno(journal);
2381         if (j_errno) {
2382                 char nbuf[16];
2383
2384                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2385                 ext3_warning(sb, __func__, "Filesystem error recorded "
2386                              "from previous mount: %s", errstr);
2387                 ext3_warning(sb, __func__, "Marking fs in need of "
2388                              "filesystem check.");
2389
2390                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2391                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2392                 ext3_commit_super (sb, es, 1);
2393
2394                 journal_clear_err(journal);
2395         }
2396 }
2397
2398 /*
2399  * Force the running and committing transactions to commit,
2400  * and wait on the commit.
2401  */
2402 int ext3_force_commit(struct super_block *sb)
2403 {
2404         journal_t *journal;
2405         int ret;
2406
2407         if (sb->s_flags & MS_RDONLY)
2408                 return 0;
2409
2410         journal = EXT3_SB(sb)->s_journal;
2411         sb->s_dirt = 0;
2412         ret = ext3_journal_force_commit(journal);
2413         return ret;
2414 }
2415
2416 /*
2417  * Ext3 always journals updates to the superblock itself, so we don't
2418  * have to propagate any other updates to the superblock on disk at this
2419  * point.  (We can probably nuke this function altogether, and remove
2420  * any mention to sb->s_dirt in all of fs/ext3; eventual cleanup...)
2421  */
2422 static void ext3_write_super (struct super_block * sb)
2423 {
2424         if (mutex_trylock(&sb->s_lock) != 0)
2425                 BUG();
2426         sb->s_dirt = 0;
2427 }
2428
2429 static int ext3_sync_fs(struct super_block *sb, int wait)
2430 {
2431         tid_t target;
2432
2433         sb->s_dirt = 0;
2434         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2435                 if (wait)
2436                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2437         }
2438         return 0;
2439 }
2440
2441 /*
2442  * LVM calls this function before a (read-only) snapshot is created.  This
2443  * gives us a chance to flush the journal completely and mark the fs clean.
2444  */
2445 static int ext3_freeze(struct super_block *sb)
2446 {
2447         int error = 0;
2448         journal_t *journal;
2449         sb->s_dirt = 0;
2450
2451         if (!(sb->s_flags & MS_RDONLY)) {
2452                 journal = EXT3_SB(sb)->s_journal;
2453
2454                 /* Now we set up the journal barrier. */
2455                 journal_lock_updates(journal);
2456
2457                 /*
2458                  * We don't want to clear needs_recovery flag when we failed
2459                  * to flush the journal.
2460                  */
2461                 error = journal_flush(journal);
2462                 if (error < 0)
2463                         goto out;
2464
2465                 /* Journal blocked and flushed, clear needs_recovery flag. */
2466                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2467                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2468                 if (error)
2469                         goto out;
2470         }
2471         return 0;
2472
2473 out:
2474         journal_unlock_updates(journal);
2475         return error;
2476 }
2477
2478 /*
2479  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2480  * flag here, even though the filesystem is not technically dirty yet.
2481  */
2482 static int ext3_unfreeze(struct super_block *sb)
2483 {
2484         if (!(sb->s_flags & MS_RDONLY)) {
2485                 lock_super(sb);
2486                 /* Reser the needs_recovery flag before the fs is unlocked. */
2487                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2488                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2489                 unlock_super(sb);
2490                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2491         }
2492         return 0;
2493 }
2494
2495 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2496 {
2497         struct ext3_super_block * es;
2498         struct ext3_sb_info *sbi = EXT3_SB(sb);
2499         ext3_fsblk_t n_blocks_count = 0;
2500         unsigned long old_sb_flags;
2501         struct ext3_mount_options old_opts;
2502         int err;
2503 #ifdef CONFIG_QUOTA
2504         int i;
2505 #endif
2506
2507         /* Store the original options */
2508         old_sb_flags = sb->s_flags;
2509         old_opts.s_mount_opt = sbi->s_mount_opt;
2510         old_opts.s_resuid = sbi->s_resuid;
2511         old_opts.s_resgid = sbi->s_resgid;
2512         old_opts.s_commit_interval = sbi->s_commit_interval;
2513 #ifdef CONFIG_QUOTA
2514         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2515         for (i = 0; i < MAXQUOTAS; i++)
2516                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2517 #endif
2518
2519         /*
2520          * Allow the "check" option to be passed as a remount option.
2521          */
2522         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2523                 err = -EINVAL;
2524                 goto restore_opts;
2525         }
2526
2527         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2528                 ext3_abort(sb, __func__, "Abort forced by user");
2529
2530         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2531                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2532
2533         es = sbi->s_es;
2534
2535         ext3_init_journal_params(sb, sbi->s_journal);
2536
2537         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2538                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2539                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2540                         err = -EROFS;
2541                         goto restore_opts;
2542                 }
2543
2544                 if (*flags & MS_RDONLY) {
2545                         /*
2546                          * First of all, the unconditional stuff we have to do
2547                          * to disable replay of the journal when we next remount
2548                          */
2549                         sb->s_flags |= MS_RDONLY;
2550
2551                         /*
2552                          * OK, test if we are remounting a valid rw partition
2553                          * readonly, and if so set the rdonly flag and then
2554                          * mark the partition as valid again.
2555                          */
2556                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2557                             (sbi->s_mount_state & EXT3_VALID_FS))
2558                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2559
2560                         /*
2561                          * We have to unlock super so that we can wait for
2562                          * transactions.
2563                          */
2564                         unlock_super(sb);
2565                         ext3_mark_recovery_complete(sb, es);
2566                         lock_super(sb);
2567                 } else {
2568                         __le32 ret;
2569                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2570                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2571                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2572                                        "remount RDWR because of unsupported "
2573                                        "optional features (%x).\n",
2574                                        sb->s_id, le32_to_cpu(ret));
2575                                 err = -EROFS;
2576                                 goto restore_opts;
2577                         }
2578
2579                         /*
2580                          * If we have an unprocessed orphan list hanging
2581                          * around from a previously readonly bdev mount,
2582                          * require a full umount/remount for now.
2583                          */
2584                         if (es->s_last_orphan) {
2585                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2586                                        "remount RDWR because of unprocessed "
2587                                        "orphan inode list.  Please "
2588                                        "umount/remount instead.\n",
2589                                        sb->s_id);
2590                                 err = -EINVAL;
2591                                 goto restore_opts;
2592                         }
2593
2594                         /*
2595                          * Mounting a RDONLY partition read-write, so reread
2596                          * and store the current valid flag.  (It may have
2597                          * been changed by e2fsck since we originally mounted
2598                          * the partition.)
2599                          */
2600                         ext3_clear_journal_err(sb, es);
2601                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2602                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2603                                 goto restore_opts;
2604                         if (!ext3_setup_super (sb, es, 0))
2605                                 sb->s_flags &= ~MS_RDONLY;
2606                 }
2607         }
2608 #ifdef CONFIG_QUOTA
2609         /* Release old quota file names */
2610         for (i = 0; i < MAXQUOTAS; i++)
2611                 if (old_opts.s_qf_names[i] &&
2612                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2613                         kfree(old_opts.s_qf_names[i]);
2614 #endif
2615         return 0;
2616 restore_opts:
2617         sb->s_flags = old_sb_flags;
2618         sbi->s_mount_opt = old_opts.s_mount_opt;
2619         sbi->s_resuid = old_opts.s_resuid;
2620         sbi->s_resgid = old_opts.s_resgid;
2621         sbi->s_commit_interval = old_opts.s_commit_interval;
2622 #ifdef CONFIG_QUOTA
2623         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2624         for (i = 0; i < MAXQUOTAS; i++) {
2625                 if (sbi->s_qf_names[i] &&
2626                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2627                         kfree(sbi->s_qf_names[i]);
2628                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2629         }
2630 #endif
2631         return err;
2632 }
2633
2634 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2635 {
2636         struct super_block *sb = dentry->d_sb;
2637         struct ext3_sb_info *sbi = EXT3_SB(sb);
2638         struct ext3_super_block *es = sbi->s_es;
2639         u64 fsid;
2640
2641         if (test_opt(sb, MINIX_DF)) {
2642                 sbi->s_overhead_last = 0;
2643         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2644                 unsigned long ngroups = sbi->s_groups_count, i;
2645                 ext3_fsblk_t overhead = 0;
2646                 smp_rmb();
2647
2648                 /*
2649                  * Compute the overhead (FS structures).  This is constant
2650                  * for a given filesystem unless the number of block groups
2651                  * changes so we cache the previous value until it does.
2652                  */
2653
2654                 /*
2655                  * All of the blocks before first_data_block are
2656                  * overhead
2657                  */
2658                 overhead = le32_to_cpu(es->s_first_data_block);
2659
2660                 /*
2661                  * Add the overhead attributed to the superblock and
2662                  * block group descriptors.  If the sparse superblocks
2663                  * feature is turned on, then not all groups have this.
2664                  */
2665                 for (i = 0; i < ngroups; i++) {
2666                         overhead += ext3_bg_has_super(sb, i) +
2667                                 ext3_bg_num_gdb(sb, i);
2668                         cond_resched();
2669                 }
2670
2671                 /*
2672                  * Every block group has an inode bitmap, a block
2673                  * bitmap, and an inode table.
2674                  */
2675                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2676                 sbi->s_overhead_last = overhead;
2677                 smp_wmb();
2678                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2679         }
2680
2681         buf->f_type = EXT3_SUPER_MAGIC;
2682         buf->f_bsize = sb->s_blocksize;
2683         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2684         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2685         es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
2686         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2687         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2688                 buf->f_bavail = 0;
2689         buf->f_files = le32_to_cpu(es->s_inodes_count);
2690         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2691         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2692         buf->f_namelen = EXT3_NAME_LEN;
2693         fsid = le64_to_cpup((void *)es->s_uuid) ^
2694                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2695         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2696         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2697         return 0;
2698 }
2699
2700 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2701  * is locked for write. Otherwise the are possible deadlocks:
2702  * Process 1                         Process 2
2703  * ext3_create()                     quota_sync()
2704  *   journal_start()                   write_dquot()
2705  *   DQUOT_INIT()                        down(dqio_mutex)
2706  *     down(dqio_mutex)                    journal_start()
2707  *
2708  */
2709
2710 #ifdef CONFIG_QUOTA
2711
2712 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2713 {
2714         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2715 }
2716
2717 static int ext3_dquot_initialize(struct inode *inode, int type)
2718 {
2719         handle_t *handle;
2720         int ret, err;
2721
2722         /* We may create quota structure so we need to reserve enough blocks */
2723         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2724         if (IS_ERR(handle))
2725                 return PTR_ERR(handle);
2726         ret = dquot_initialize(inode, type);
2727         err = ext3_journal_stop(handle);
2728         if (!ret)
2729                 ret = err;
2730         return ret;
2731 }
2732
2733 static int ext3_dquot_drop(struct inode *inode)
2734 {
2735         handle_t *handle;
2736         int ret, err;
2737
2738         /* We may delete quota structure so we need to reserve enough blocks */
2739         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2740         if (IS_ERR(handle)) {
2741                 /*
2742                  * We call dquot_drop() anyway to at least release references
2743                  * to quota structures so that umount does not hang.
2744                  */
2745                 dquot_drop(inode);
2746                 return PTR_ERR(handle);
2747         }
2748         ret = dquot_drop(inode);
2749         err = ext3_journal_stop(handle);
2750         if (!ret)
2751                 ret = err;
2752         return ret;
2753 }
2754
2755 static int ext3_write_dquot(struct dquot *dquot)
2756 {
2757         int ret, err;
2758         handle_t *handle;
2759         struct inode *inode;
2760
2761         inode = dquot_to_inode(dquot);
2762         handle = ext3_journal_start(inode,
2763                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2764         if (IS_ERR(handle))
2765                 return PTR_ERR(handle);
2766         ret = dquot_commit(dquot);
2767         err = ext3_journal_stop(handle);
2768         if (!ret)
2769                 ret = err;
2770         return ret;
2771 }
2772
2773 static int ext3_acquire_dquot(struct dquot *dquot)
2774 {
2775         int ret, err;
2776         handle_t *handle;
2777
2778         handle = ext3_journal_start(dquot_to_inode(dquot),
2779                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2780         if (IS_ERR(handle))
2781                 return PTR_ERR(handle);
2782         ret = dquot_acquire(dquot);
2783         err = ext3_journal_stop(handle);
2784         if (!ret)
2785                 ret = err;
2786         return ret;
2787 }
2788
2789 static int ext3_release_dquot(struct dquot *dquot)
2790 {
2791         int ret, err;
2792         handle_t *handle;
2793
2794         handle = ext3_journal_start(dquot_to_inode(dquot),
2795                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2796         if (IS_ERR(handle)) {
2797                 /* Release dquot anyway to avoid endless cycle in dqput() */
2798                 dquot_release(dquot);
2799                 return PTR_ERR(handle);
2800         }
2801         ret = dquot_release(dquot);
2802         err = ext3_journal_stop(handle);
2803         if (!ret)
2804                 ret = err;
2805         return ret;
2806 }
2807
2808 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2809 {
2810         /* Are we journaling quotas? */
2811         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2812             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2813                 dquot_mark_dquot_dirty(dquot);
2814                 return ext3_write_dquot(dquot);
2815         } else {
2816                 return dquot_mark_dquot_dirty(dquot);
2817         }
2818 }
2819
2820 static int ext3_write_info(struct super_block *sb, int type)
2821 {
2822         int ret, err;
2823         handle_t *handle;
2824
2825         /* Data block + inode block */
2826         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2827         if (IS_ERR(handle))
2828                 return PTR_ERR(handle);
2829         ret = dquot_commit_info(sb, type);
2830         err = ext3_journal_stop(handle);
2831         if (!ret)
2832                 ret = err;
2833         return ret;
2834 }
2835
2836 /*
2837  * Turn on quotas during mount time - we need to find
2838  * the quota file and such...
2839  */
2840 static int ext3_quota_on_mount(struct super_block *sb, int type)
2841 {
2842         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2843                         EXT3_SB(sb)->s_jquota_fmt, type);
2844 }
2845
2846 /*
2847  * Standard function to be called on quota_on
2848  */
2849 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2850                          char *name, int remount)
2851 {
2852         int err;
2853         struct path path;
2854
2855         if (!test_opt(sb, QUOTA))
2856                 return -EINVAL;
2857         /* When remounting, no checks are needed and in fact, name is NULL */
2858         if (remount)
2859                 return vfs_quota_on(sb, type, format_id, name, remount);
2860
2861         err = kern_path(name, LOOKUP_FOLLOW, &path);
2862         if (err)
2863                 return err;
2864
2865         /* Quotafile not on the same filesystem? */
2866         if (path.mnt->mnt_sb != sb) {
2867                 path_put(&path);
2868                 return -EXDEV;
2869         }
2870         /* Journaling quota? */
2871         if (EXT3_SB(sb)->s_qf_names[type]) {
2872                 /* Quotafile not of fs root? */
2873                 if (path.dentry->d_parent != sb->s_root)
2874                         printk(KERN_WARNING
2875                                 "EXT3-fs: Quota file not on filesystem root. "
2876                                 "Journaled quota will not work.\n");
2877         }
2878
2879         /*
2880          * When we journal data on quota file, we have to flush journal to see
2881          * all updates to the file when we bypass pagecache...
2882          */
2883         if (ext3_should_journal_data(path.dentry->d_inode)) {
2884                 /*
2885                  * We don't need to lock updates but journal_flush() could
2886                  * otherwise be livelocked...
2887                  */
2888                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2889                 err = journal_flush(EXT3_SB(sb)->s_journal);
2890                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2891                 if (err) {
2892                         path_put(&path);
2893                         return err;
2894                 }
2895         }
2896
2897         err = vfs_quota_on_path(sb, type, format_id, &path);
2898         path_put(&path);
2899         return err;
2900 }
2901
2902 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2903  * acquiring the locks... As quota files are never truncated and quota code
2904  * itself serializes the operations (and noone else should touch the files)
2905  * we don't have to be afraid of races */
2906 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2907                                size_t len, loff_t off)
2908 {
2909         struct inode *inode = sb_dqopt(sb)->files[type];
2910         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2911         int err = 0;
2912         int offset = off & (sb->s_blocksize - 1);
2913         int tocopy;
2914         size_t toread;
2915         struct buffer_head *bh;
2916         loff_t i_size = i_size_read(inode);
2917
2918         if (off > i_size)
2919                 return 0;
2920         if (off+len > i_size)
2921                 len = i_size-off;
2922         toread = len;
2923         while (toread > 0) {
2924                 tocopy = sb->s_blocksize - offset < toread ?
2925                                 sb->s_blocksize - offset : toread;
2926                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2927                 if (err)
2928                         return err;
2929                 if (!bh)        /* A hole? */
2930                         memset(data, 0, tocopy);
2931                 else
2932                         memcpy(data, bh->b_data+offset, tocopy);
2933                 brelse(bh);
2934                 offset = 0;
2935                 toread -= tocopy;
2936                 data += tocopy;
2937                 blk++;
2938         }
2939         return len;
2940 }
2941
2942 /* Write to quotafile (we know the transaction is already started and has
2943  * enough credits) */
2944 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2945                                 const char *data, size_t len, loff_t off)
2946 {
2947         struct inode *inode = sb_dqopt(sb)->files[type];
2948         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2949         int err = 0;
2950         int offset = off & (sb->s_blocksize - 1);
2951         int tocopy;
2952         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2953         size_t towrite = len;
2954         struct buffer_head *bh;
2955         handle_t *handle = journal_current_handle();
2956
2957         if (!handle) {
2958                 printk(KERN_WARNING "EXT3-fs: Quota write (off=%Lu, len=%Lu)"
2959                         " cancelled because transaction is not started.\n",
2960                         (unsigned long long)off, (unsigned long long)len);
2961                 return -EIO;
2962         }
2963         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2964         while (towrite > 0) {
2965                 tocopy = sb->s_blocksize - offset < towrite ?
2966                                 sb->s_blocksize - offset : towrite;
2967                 bh = ext3_bread(handle, inode, blk, 1, &err);
2968                 if (!bh)
2969                         goto out;
2970                 if (journal_quota) {
2971                         err = ext3_journal_get_write_access(handle, bh);
2972                         if (err) {
2973                                 brelse(bh);
2974                                 goto out;
2975                         }
2976                 }
2977                 lock_buffer(bh);
2978                 memcpy(bh->b_data+offset, data, tocopy);
2979                 flush_dcache_page(bh->b_page);
2980                 unlock_buffer(bh);
2981                 if (journal_quota)
2982                         err = ext3_journal_dirty_metadata(handle, bh);
2983                 else {
2984                         /* Always do at least ordered writes for quotas */
2985                         err = ext3_journal_dirty_data(handle, bh);
2986                         mark_buffer_dirty(bh);
2987                 }
2988                 brelse(bh);
2989                 if (err)
2990                         goto out;
2991                 offset = 0;
2992                 towrite -= tocopy;
2993                 data += tocopy;
2994                 blk++;
2995         }
2996 out:
2997         if (len == towrite) {
2998                 mutex_unlock(&inode->i_mutex);
2999                 return err;
3000         }
3001         if (inode->i_size < off+len-towrite) {
3002                 i_size_write(inode, off+len-towrite);
3003                 EXT3_I(inode)->i_disksize = inode->i_size;
3004         }
3005         inode->i_version++;
3006         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3007         ext3_mark_inode_dirty(handle, inode);
3008         mutex_unlock(&inode->i_mutex);
3009         return len - towrite;
3010 }
3011
3012 #endif
3013
3014 static int ext3_get_sb(struct file_system_type *fs_type,
3015         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3016 {
3017         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
3018 }
3019
3020 static struct file_system_type ext3_fs_type = {
3021         .owner          = THIS_MODULE,
3022         .name           = "ext3",
3023         .get_sb         = ext3_get_sb,
3024         .kill_sb        = kill_block_super,
3025         .fs_flags       = FS_REQUIRES_DEV,
3026 };
3027
3028 static int __init init_ext3_fs(void)
3029 {
3030         int err = init_ext3_xattr();
3031         if (err)
3032                 return err;
3033         err = init_inodecache();
3034         if (err)
3035                 goto out1;
3036         err = register_filesystem(&ext3_fs_type);
3037         if (err)
3038                 goto out;
3039         return 0;
3040 out:
3041         destroy_inodecache();
3042 out1:
3043         exit_ext3_xattr();
3044         return err;
3045 }
3046
3047 static void __exit exit_ext3_fs(void)
3048 {
3049         unregister_filesystem(&ext3_fs_type);
3050         destroy_inodecache();
3051         exit_ext3_xattr();
3052 }
3053
3054 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3055 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3056 MODULE_LICENSE("GPL");
3057 module_init(init_ext3_fs)
3058 module_exit(exit_ext3_fs)