ext3: mount flags manipulation cleanup
[linux-2.6.git] / 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 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
48   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
49 #else
50   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
51 #endif
52
53 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
54                              unsigned long journal_devnum);
55 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
56                                unsigned int);
57 static int ext3_commit_super(struct super_block *sb,
58                                struct ext3_super_block *es,
59                                int sync);
60 static void ext3_mark_recovery_complete(struct super_block * sb,
61                                         struct ext3_super_block * es);
62 static void ext3_clear_journal_err(struct super_block * sb,
63                                    struct ext3_super_block * es);
64 static int ext3_sync_fs(struct super_block *sb, int wait);
65 static const char *ext3_decode_error(struct super_block * sb, int errno,
66                                      char nbuf[16]);
67 static int ext3_remount (struct super_block * sb, int * flags, char * data);
68 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
69 static int ext3_unfreeze(struct super_block *sb);
70 static int ext3_freeze(struct super_block *sb);
71
72 /*
73  * Wrappers for journal_start/end.
74  *
75  * The only special thing we need to do here is to make sure that all
76  * journal_end calls result in the superblock being marked dirty, so
77  * that sync() will call the filesystem's write_super callback if
78  * appropriate.
79  */
80 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
81 {
82         journal_t *journal;
83
84         if (sb->s_flags & MS_RDONLY)
85                 return ERR_PTR(-EROFS);
86
87         /* Special case here: if the journal has aborted behind our
88          * backs (eg. EIO in the commit thread), then we still need to
89          * take the FS itself readonly cleanly. */
90         journal = EXT3_SB(sb)->s_journal;
91         if (is_journal_aborted(journal)) {
92                 ext3_abort(sb, __func__,
93                            "Detected aborted journal");
94                 return ERR_PTR(-EROFS);
95         }
96
97         return journal_start(journal, nblocks);
98 }
99
100 /*
101  * The only special thing we need to do here is to make sure that all
102  * journal_stop calls result in the superblock being marked dirty, so
103  * that sync() will call the filesystem's write_super callback if
104  * appropriate.
105  */
106 int __ext3_journal_stop(const char *where, handle_t *handle)
107 {
108         struct super_block *sb;
109         int err;
110         int rc;
111
112         sb = handle->h_transaction->t_journal->j_private;
113         err = handle->h_err;
114         rc = journal_stop(handle);
115
116         if (!err)
117                 err = rc;
118         if (err)
119                 __ext3_std_error(sb, where, err);
120         return err;
121 }
122
123 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
124                 struct buffer_head *bh, handle_t *handle, int err)
125 {
126         char nbuf[16];
127         const char *errstr = ext3_decode_error(NULL, err, nbuf);
128
129         if (bh)
130                 BUFFER_TRACE(bh, "abort");
131
132         if (!handle->h_err)
133                 handle->h_err = err;
134
135         if (is_handle_aborted(handle))
136                 return;
137
138         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
139                 caller, errstr, err_fn);
140
141         journal_abort_handle(handle);
142 }
143
144 void ext3_msg(struct super_block *sb, const char *prefix,
145                 const char *fmt, ...)
146 {
147         va_list args;
148
149         va_start(args, fmt);
150         printk("%sEXT3-fs (%s): ", prefix, sb->s_id);
151         vprintk(fmt, args);
152         printk("\n");
153         va_end(args);
154 }
155
156 /* Deal with the reporting of failure conditions on a filesystem such as
157  * inconsistencies detected or read IO failures.
158  *
159  * On ext2, we can store the error state of the filesystem in the
160  * superblock.  That is not possible on ext3, because we may have other
161  * write ordering constraints on the superblock which prevent us from
162  * writing it out straight away; and given that the journal is about to
163  * be aborted, we can't rely on the current, or future, transactions to
164  * write out the superblock safely.
165  *
166  * We'll just use the journal_abort() error code to record an error in
167  * the journal instead.  On recovery, the journal will compain about
168  * that error until we've noted it down and cleared it.
169  */
170
171 static void ext3_handle_error(struct super_block *sb)
172 {
173         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
174
175         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
176         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
177
178         if (sb->s_flags & MS_RDONLY)
179                 return;
180
181         if (!test_opt (sb, ERRORS_CONT)) {
182                 journal_t *journal = EXT3_SB(sb)->s_journal;
183
184                 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
185                 if (journal)
186                         journal_abort(journal, -EIO);
187         }
188         if (test_opt (sb, ERRORS_RO)) {
189                 ext3_msg(sb, KERN_CRIT,
190                         "error: remounting filesystem read-only");
191                 sb->s_flags |= MS_RDONLY;
192         }
193         ext3_commit_super(sb, es, 1);
194         if (test_opt(sb, ERRORS_PANIC))
195                 panic("EXT3-fs (%s): panic forced after error\n",
196                         sb->s_id);
197 }
198
199 void ext3_error (struct super_block * sb, const char * function,
200                  const char * fmt, ...)
201 {
202         va_list args;
203
204         va_start(args, fmt);
205         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
206         vprintk(fmt, args);
207         printk("\n");
208         va_end(args);
209
210         ext3_handle_error(sb);
211 }
212
213 static const char *ext3_decode_error(struct super_block * sb, int errno,
214                                      char nbuf[16])
215 {
216         char *errstr = NULL;
217
218         switch (errno) {
219         case -EIO:
220                 errstr = "IO failure";
221                 break;
222         case -ENOMEM:
223                 errstr = "Out of memory";
224                 break;
225         case -EROFS:
226                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
227                         errstr = "Journal has aborted";
228                 else
229                         errstr = "Readonly filesystem";
230                 break;
231         default:
232                 /* If the caller passed in an extra buffer for unknown
233                  * errors, textualise them now.  Else we just return
234                  * NULL. */
235                 if (nbuf) {
236                         /* Check for truncated error codes... */
237                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
238                                 errstr = nbuf;
239                 }
240                 break;
241         }
242
243         return errstr;
244 }
245
246 /* __ext3_std_error decodes expected errors from journaling functions
247  * automatically and invokes the appropriate error response.  */
248
249 void __ext3_std_error (struct super_block * sb, const char * function,
250                        int errno)
251 {
252         char nbuf[16];
253         const char *errstr;
254
255         /* Special case: if the error is EROFS, and we're not already
256          * inside a transaction, then there's really no point in logging
257          * an error. */
258         if (errno == -EROFS && journal_current_handle() == NULL &&
259             (sb->s_flags & MS_RDONLY))
260                 return;
261
262         errstr = ext3_decode_error(sb, errno, nbuf);
263         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
264
265         ext3_handle_error(sb);
266 }
267
268 /*
269  * ext3_abort is a much stronger failure handler than ext3_error.  The
270  * abort function may be used to deal with unrecoverable failures such
271  * as journal IO errors or ENOMEM at a critical moment in log management.
272  *
273  * We unconditionally force the filesystem into an ABORT|READONLY state,
274  * unless the error response on the fs has been set to panic in which
275  * case we take the easy way out and panic immediately.
276  */
277
278 void ext3_abort (struct super_block * sb, const char * function,
279                  const char * fmt, ...)
280 {
281         va_list args;
282
283         va_start(args, fmt);
284         printk(KERN_CRIT "EXT3-fs (%s): error: %s: ", sb->s_id, function);
285         vprintk(fmt, args);
286         printk("\n");
287         va_end(args);
288
289         if (test_opt(sb, ERRORS_PANIC))
290                 panic("EXT3-fs: panic from previous error\n");
291
292         if (sb->s_flags & MS_RDONLY)
293                 return;
294
295         ext3_msg(sb, KERN_CRIT,
296                 "error: remounting filesystem read-only");
297         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
298         sb->s_flags |= MS_RDONLY;
299         set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
300         if (EXT3_SB(sb)->s_journal)
301                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
302 }
303
304 void ext3_warning (struct super_block * sb, const char * function,
305                    const char * fmt, ...)
306 {
307         va_list args;
308
309         va_start(args, fmt);
310         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: ",
311                sb->s_id, function);
312         vprintk(fmt, args);
313         printk("\n");
314         va_end(args);
315 }
316
317 void ext3_update_dynamic_rev(struct super_block *sb)
318 {
319         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
320
321         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
322                 return;
323
324         ext3_msg(sb, KERN_WARNING,
325                 "warning: updating to rev %d because of "
326                 "new feature flag, running e2fsck is recommended",
327                 EXT3_DYNAMIC_REV);
328
329         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
330         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
331         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
332         /* leave es->s_feature_*compat flags alone */
333         /* es->s_uuid will be set by e2fsck if empty */
334
335         /*
336          * The rest of the superblock fields should be zero, and if not it
337          * means they are likely already in use, so leave them alone.  We
338          * can leave it up to e2fsck to clean up any inconsistencies there.
339          */
340 }
341
342 /*
343  * Open the external journal device
344  */
345 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
346 {
347         struct block_device *bdev;
348         char b[BDEVNAME_SIZE];
349
350         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
351         if (IS_ERR(bdev))
352                 goto fail;
353         return bdev;
354
355 fail:
356         ext3_msg(sb, "error: failed to open journal device %s: %ld",
357                 __bdevname(dev, b), PTR_ERR(bdev));
358
359         return NULL;
360 }
361
362 /*
363  * Release the journal device
364  */
365 static int ext3_blkdev_put(struct block_device *bdev)
366 {
367         bd_release(bdev);
368         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
369 }
370
371 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
372 {
373         struct block_device *bdev;
374         int ret = -ENODEV;
375
376         bdev = sbi->journal_bdev;
377         if (bdev) {
378                 ret = ext3_blkdev_put(bdev);
379                 sbi->journal_bdev = NULL;
380         }
381         return ret;
382 }
383
384 static inline struct inode *orphan_list_entry(struct list_head *l)
385 {
386         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
387 }
388
389 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
390 {
391         struct list_head *l;
392
393         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
394                le32_to_cpu(sbi->s_es->s_last_orphan));
395
396         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
397         list_for_each(l, &sbi->s_orphan) {
398                 struct inode *inode = orphan_list_entry(l);
399                 ext3_msg(sb, KERN_ERR, "  "
400                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
401                        inode->i_sb->s_id, inode->i_ino, inode,
402                        inode->i_mode, inode->i_nlink,
403                        NEXT_ORPHAN(inode));
404         }
405 }
406
407 static void ext3_put_super (struct super_block * sb)
408 {
409         struct ext3_sb_info *sbi = EXT3_SB(sb);
410         struct ext3_super_block *es = sbi->s_es;
411         int i, err;
412
413         lock_kernel();
414
415         ext3_xattr_put_super(sb);
416         err = journal_destroy(sbi->s_journal);
417         sbi->s_journal = NULL;
418         if (err < 0)
419                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
420
421         if (!(sb->s_flags & MS_RDONLY)) {
422                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
423                 es->s_state = cpu_to_le16(sbi->s_mount_state);
424                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
425                 mark_buffer_dirty(sbi->s_sbh);
426                 ext3_commit_super(sb, es, 1);
427         }
428
429         for (i = 0; i < sbi->s_gdb_count; i++)
430                 brelse(sbi->s_group_desc[i]);
431         kfree(sbi->s_group_desc);
432         percpu_counter_destroy(&sbi->s_freeblocks_counter);
433         percpu_counter_destroy(&sbi->s_freeinodes_counter);
434         percpu_counter_destroy(&sbi->s_dirs_counter);
435         brelse(sbi->s_sbh);
436 #ifdef CONFIG_QUOTA
437         for (i = 0; i < MAXQUOTAS; i++)
438                 kfree(sbi->s_qf_names[i]);
439 #endif
440
441         /* Debugging code just in case the in-memory inode orphan list
442          * isn't empty.  The on-disk one can be non-empty if we've
443          * detected an error and taken the fs readonly, but the
444          * in-memory list had better be clean by this point. */
445         if (!list_empty(&sbi->s_orphan))
446                 dump_orphan_list(sb, sbi);
447         J_ASSERT(list_empty(&sbi->s_orphan));
448
449         invalidate_bdev(sb->s_bdev);
450         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
451                 /*
452                  * Invalidate the journal device's buffers.  We don't want them
453                  * floating about in memory - the physical journal device may
454                  * hotswapped, and it breaks the `ro-after' testing code.
455                  */
456                 sync_blockdev(sbi->journal_bdev);
457                 invalidate_bdev(sbi->journal_bdev);
458                 ext3_blkdev_remove(sbi);
459         }
460         sb->s_fs_info = NULL;
461         kfree(sbi->s_blockgroup_lock);
462         kfree(sbi);
463
464         unlock_kernel();
465 }
466
467 static struct kmem_cache *ext3_inode_cachep;
468
469 /*
470  * Called inside transaction, so use GFP_NOFS
471  */
472 static struct inode *ext3_alloc_inode(struct super_block *sb)
473 {
474         struct ext3_inode_info *ei;
475
476         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
477         if (!ei)
478                 return NULL;
479         ei->i_block_alloc_info = NULL;
480         ei->vfs_inode.i_version = 1;
481         atomic_set(&ei->i_datasync_tid, 0);
482         atomic_set(&ei->i_sync_tid, 0);
483         return &ei->vfs_inode;
484 }
485
486 static void ext3_destroy_inode(struct inode *inode)
487 {
488         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
489                 printk("EXT3 Inode %p: orphan list check failed!\n",
490                         EXT3_I(inode));
491                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
492                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
493                                 false);
494                 dump_stack();
495         }
496         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
497 }
498
499 static void init_once(void *foo)
500 {
501         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
502
503         INIT_LIST_HEAD(&ei->i_orphan);
504 #ifdef CONFIG_EXT3_FS_XATTR
505         init_rwsem(&ei->xattr_sem);
506 #endif
507         mutex_init(&ei->truncate_mutex);
508         inode_init_once(&ei->vfs_inode);
509 }
510
511 static int init_inodecache(void)
512 {
513         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
514                                              sizeof(struct ext3_inode_info),
515                                              0, (SLAB_RECLAIM_ACCOUNT|
516                                                 SLAB_MEM_SPREAD),
517                                              init_once);
518         if (ext3_inode_cachep == NULL)
519                 return -ENOMEM;
520         return 0;
521 }
522
523 static void destroy_inodecache(void)
524 {
525         kmem_cache_destroy(ext3_inode_cachep);
526 }
527
528 static void ext3_clear_inode(struct inode *inode)
529 {
530         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
531         ext3_discard_reservation(inode);
532         EXT3_I(inode)->i_block_alloc_info = NULL;
533         if (unlikely(rsv))
534                 kfree(rsv);
535 }
536
537 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
538 {
539 #if defined(CONFIG_QUOTA)
540         struct ext3_sb_info *sbi = EXT3_SB(sb);
541
542         if (sbi->s_jquota_fmt) {
543                 char *fmtname = "";
544
545                 switch (sbi->s_jquota_fmt) {
546                 case QFMT_VFS_OLD:
547                         fmtname = "vfsold";
548                         break;
549                 case QFMT_VFS_V0:
550                         fmtname = "vfsv0";
551                         break;
552                 case QFMT_VFS_V1:
553                         fmtname = "vfsv1";
554                         break;
555                 }
556                 seq_printf(seq, ",jqfmt=%s", fmtname);
557         }
558
559         if (sbi->s_qf_names[USRQUOTA])
560                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
561
562         if (sbi->s_qf_names[GRPQUOTA])
563                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
564
565         if (test_opt(sb, USRQUOTA))
566                 seq_puts(seq, ",usrquota");
567
568         if (test_opt(sb, GRPQUOTA))
569                 seq_puts(seq, ",grpquota");
570 #endif
571 }
572
573 static char *data_mode_string(unsigned long mode)
574 {
575         switch (mode) {
576         case EXT3_MOUNT_JOURNAL_DATA:
577                 return "journal";
578         case EXT3_MOUNT_ORDERED_DATA:
579                 return "ordered";
580         case EXT3_MOUNT_WRITEBACK_DATA:
581                 return "writeback";
582         }
583         return "unknown";
584 }
585
586 /*
587  * Show an option if
588  *  - it's set to a non-default value OR
589  *  - if the per-sb default is different from the global default
590  */
591 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
592 {
593         struct super_block *sb = vfs->mnt_sb;
594         struct ext3_sb_info *sbi = EXT3_SB(sb);
595         struct ext3_super_block *es = sbi->s_es;
596         unsigned long def_mount_opts;
597
598         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
599
600         if (sbi->s_sb_block != 1)
601                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
602         if (test_opt(sb, MINIX_DF))
603                 seq_puts(seq, ",minixdf");
604         if (test_opt(sb, GRPID))
605                 seq_puts(seq, ",grpid");
606         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
607                 seq_puts(seq, ",nogrpid");
608         if (sbi->s_resuid != EXT3_DEF_RESUID ||
609             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
610                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
611         }
612         if (sbi->s_resgid != EXT3_DEF_RESGID ||
613             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
614                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
615         }
616         if (test_opt(sb, ERRORS_RO)) {
617                 int def_errors = le16_to_cpu(es->s_errors);
618
619                 if (def_errors == EXT3_ERRORS_PANIC ||
620                     def_errors == EXT3_ERRORS_CONTINUE) {
621                         seq_puts(seq, ",errors=remount-ro");
622                 }
623         }
624         if (test_opt(sb, ERRORS_CONT))
625                 seq_puts(seq, ",errors=continue");
626         if (test_opt(sb, ERRORS_PANIC))
627                 seq_puts(seq, ",errors=panic");
628         if (test_opt(sb, NO_UID32))
629                 seq_puts(seq, ",nouid32");
630         if (test_opt(sb, DEBUG))
631                 seq_puts(seq, ",debug");
632         if (test_opt(sb, OLDALLOC))
633                 seq_puts(seq, ",oldalloc");
634 #ifdef CONFIG_EXT3_FS_XATTR
635         if (test_opt(sb, XATTR_USER))
636                 seq_puts(seq, ",user_xattr");
637         if (!test_opt(sb, XATTR_USER) &&
638             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
639                 seq_puts(seq, ",nouser_xattr");
640         }
641 #endif
642 #ifdef CONFIG_EXT3_FS_POSIX_ACL
643         if (test_opt(sb, POSIX_ACL))
644                 seq_puts(seq, ",acl");
645         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
646                 seq_puts(seq, ",noacl");
647 #endif
648         if (!test_opt(sb, RESERVATION))
649                 seq_puts(seq, ",noreservation");
650         if (sbi->s_commit_interval) {
651                 seq_printf(seq, ",commit=%u",
652                            (unsigned) (sbi->s_commit_interval / HZ));
653         }
654         if (test_opt(sb, BARRIER))
655                 seq_puts(seq, ",barrier=1");
656         if (test_opt(sb, NOBH))
657                 seq_puts(seq, ",nobh");
658
659         seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
660         if (test_opt(sb, DATA_ERR_ABORT))
661                 seq_puts(seq, ",data_err=abort");
662
663         if (test_opt(sb, NOLOAD))
664                 seq_puts(seq, ",norecovery");
665
666         ext3_show_quota_options(seq, sb);
667
668         return 0;
669 }
670
671
672 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
673                 u64 ino, u32 generation)
674 {
675         struct inode *inode;
676
677         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
678                 return ERR_PTR(-ESTALE);
679         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
680                 return ERR_PTR(-ESTALE);
681
682         /* iget isn't really right if the inode is currently unallocated!!
683          *
684          * ext3_read_inode will return a bad_inode if the inode had been
685          * deleted, so we should be safe.
686          *
687          * Currently we don't know the generation for parent directory, so
688          * a generation of 0 means "accept any"
689          */
690         inode = ext3_iget(sb, ino);
691         if (IS_ERR(inode))
692                 return ERR_CAST(inode);
693         if (generation && inode->i_generation != generation) {
694                 iput(inode);
695                 return ERR_PTR(-ESTALE);
696         }
697
698         return inode;
699 }
700
701 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
702                 int fh_len, int fh_type)
703 {
704         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
705                                     ext3_nfs_get_inode);
706 }
707
708 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
709                 int fh_len, int fh_type)
710 {
711         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
712                                     ext3_nfs_get_inode);
713 }
714
715 /*
716  * Try to release metadata pages (indirect blocks, directories) which are
717  * mapped via the block device.  Since these pages could have journal heads
718  * which would prevent try_to_free_buffers() from freeing them, we must use
719  * jbd layer's try_to_free_buffers() function to release them.
720  */
721 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
722                                  gfp_t wait)
723 {
724         journal_t *journal = EXT3_SB(sb)->s_journal;
725
726         WARN_ON(PageChecked(page));
727         if (!page_has_buffers(page))
728                 return 0;
729         if (journal)
730                 return journal_try_to_free_buffers(journal, page, 
731                                                    wait & ~__GFP_WAIT);
732         return try_to_free_buffers(page);
733 }
734
735 #ifdef CONFIG_QUOTA
736 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
737 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
738
739 static int ext3_write_dquot(struct dquot *dquot);
740 static int ext3_acquire_dquot(struct dquot *dquot);
741 static int ext3_release_dquot(struct dquot *dquot);
742 static int ext3_mark_dquot_dirty(struct dquot *dquot);
743 static int ext3_write_info(struct super_block *sb, int type);
744 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
745                                 char *path, int remount);
746 static int ext3_quota_on_mount(struct super_block *sb, int type);
747 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
748                                size_t len, loff_t off);
749 static ssize_t ext3_quota_write(struct super_block *sb, int type,
750                                 const char *data, size_t len, loff_t off);
751
752 static const struct dquot_operations ext3_quota_operations = {
753         .initialize     = dquot_initialize,
754         .drop           = dquot_drop,
755         .alloc_space    = dquot_alloc_space,
756         .alloc_inode    = dquot_alloc_inode,
757         .free_space     = dquot_free_space,
758         .free_inode     = dquot_free_inode,
759         .transfer       = dquot_transfer,
760         .write_dquot    = ext3_write_dquot,
761         .acquire_dquot  = ext3_acquire_dquot,
762         .release_dquot  = ext3_release_dquot,
763         .mark_dirty     = ext3_mark_dquot_dirty,
764         .write_info     = ext3_write_info,
765         .alloc_dquot    = dquot_alloc,
766         .destroy_dquot  = dquot_destroy,
767 };
768
769 static const struct quotactl_ops ext3_qctl_operations = {
770         .quota_on       = ext3_quota_on,
771         .quota_off      = vfs_quota_off,
772         .quota_sync     = vfs_quota_sync,
773         .get_info       = vfs_get_dqinfo,
774         .set_info       = vfs_set_dqinfo,
775         .get_dqblk      = vfs_get_dqblk,
776         .set_dqblk      = vfs_set_dqblk
777 };
778 #endif
779
780 static const struct super_operations ext3_sops = {
781         .alloc_inode    = ext3_alloc_inode,
782         .destroy_inode  = ext3_destroy_inode,
783         .write_inode    = ext3_write_inode,
784         .dirty_inode    = ext3_dirty_inode,
785         .delete_inode   = ext3_delete_inode,
786         .put_super      = ext3_put_super,
787         .sync_fs        = ext3_sync_fs,
788         .freeze_fs      = ext3_freeze,
789         .unfreeze_fs    = ext3_unfreeze,
790         .statfs         = ext3_statfs,
791         .remount_fs     = ext3_remount,
792         .clear_inode    = ext3_clear_inode,
793         .show_options   = ext3_show_options,
794 #ifdef CONFIG_QUOTA
795         .quota_read     = ext3_quota_read,
796         .quota_write    = ext3_quota_write,
797 #endif
798         .bdev_try_to_free_page = bdev_try_to_free_page,
799 };
800
801 static const struct export_operations ext3_export_ops = {
802         .fh_to_dentry = ext3_fh_to_dentry,
803         .fh_to_parent = ext3_fh_to_parent,
804         .get_parent = ext3_get_parent,
805 };
806
807 enum {
808         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
809         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
810         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
811         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
812         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
813         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
814         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
815         Opt_data_err_abort, Opt_data_err_ignore,
816         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
817         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
818         Opt_noquota, Opt_ignore, Opt_barrier, Opt_err, Opt_resize,
819         Opt_usrquota, Opt_grpquota
820 };
821
822 static const match_table_t tokens = {
823         {Opt_bsd_df, "bsddf"},
824         {Opt_minix_df, "minixdf"},
825         {Opt_grpid, "grpid"},
826         {Opt_grpid, "bsdgroups"},
827         {Opt_nogrpid, "nogrpid"},
828         {Opt_nogrpid, "sysvgroups"},
829         {Opt_resgid, "resgid=%u"},
830         {Opt_resuid, "resuid=%u"},
831         {Opt_sb, "sb=%u"},
832         {Opt_err_cont, "errors=continue"},
833         {Opt_err_panic, "errors=panic"},
834         {Opt_err_ro, "errors=remount-ro"},
835         {Opt_nouid32, "nouid32"},
836         {Opt_nocheck, "nocheck"},
837         {Opt_nocheck, "check=none"},
838         {Opt_debug, "debug"},
839         {Opt_oldalloc, "oldalloc"},
840         {Opt_orlov, "orlov"},
841         {Opt_user_xattr, "user_xattr"},
842         {Opt_nouser_xattr, "nouser_xattr"},
843         {Opt_acl, "acl"},
844         {Opt_noacl, "noacl"},
845         {Opt_reservation, "reservation"},
846         {Opt_noreservation, "noreservation"},
847         {Opt_noload, "noload"},
848         {Opt_noload, "norecovery"},
849         {Opt_nobh, "nobh"},
850         {Opt_bh, "bh"},
851         {Opt_commit, "commit=%u"},
852         {Opt_journal_update, "journal=update"},
853         {Opt_journal_inum, "journal=%u"},
854         {Opt_journal_dev, "journal_dev=%u"},
855         {Opt_abort, "abort"},
856         {Opt_data_journal, "data=journal"},
857         {Opt_data_ordered, "data=ordered"},
858         {Opt_data_writeback, "data=writeback"},
859         {Opt_data_err_abort, "data_err=abort"},
860         {Opt_data_err_ignore, "data_err=ignore"},
861         {Opt_offusrjquota, "usrjquota="},
862         {Opt_usrjquota, "usrjquota=%s"},
863         {Opt_offgrpjquota, "grpjquota="},
864         {Opt_grpjquota, "grpjquota=%s"},
865         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
866         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
867         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
868         {Opt_grpquota, "grpquota"},
869         {Opt_noquota, "noquota"},
870         {Opt_quota, "quota"},
871         {Opt_usrquota, "usrquota"},
872         {Opt_barrier, "barrier=%u"},
873         {Opt_resize, "resize"},
874         {Opt_err, NULL},
875 };
876
877 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
878 {
879         ext3_fsblk_t    sb_block;
880         char            *options = (char *) *data;
881
882         if (!options || strncmp(options, "sb=", 3) != 0)
883                 return 1;       /* Default location */
884         options += 3;
885         /*todo: use simple_strtoll with >32bit ext3 */
886         sb_block = simple_strtoul(options, &options, 0);
887         if (*options && *options != ',') {
888                 ext3_msg(sb, "error: invalid sb specification: %s",
889                        (char *) *data);
890                 return 1;
891         }
892         if (*options == ',')
893                 options++;
894         *data = (void *) options;
895         return sb_block;
896 }
897
898 static int parse_options (char *options, struct super_block *sb,
899                           unsigned int *inum, unsigned long *journal_devnum,
900                           ext3_fsblk_t *n_blocks_count, int is_remount)
901 {
902         struct ext3_sb_info *sbi = EXT3_SB(sb);
903         char * p;
904         substring_t args[MAX_OPT_ARGS];
905         int data_opt = 0;
906         int option;
907 #ifdef CONFIG_QUOTA
908         int qtype, qfmt;
909         char *qname;
910 #endif
911
912         if (!options)
913                 return 1;
914
915         while ((p = strsep (&options, ",")) != NULL) {
916                 int token;
917                 if (!*p)
918                         continue;
919
920                 token = match_token(p, tokens, args);
921                 switch (token) {
922                 case Opt_bsd_df:
923                         clear_opt (sbi->s_mount_opt, MINIX_DF);
924                         break;
925                 case Opt_minix_df:
926                         set_opt (sbi->s_mount_opt, MINIX_DF);
927                         break;
928                 case Opt_grpid:
929                         set_opt (sbi->s_mount_opt, GRPID);
930                         break;
931                 case Opt_nogrpid:
932                         clear_opt (sbi->s_mount_opt, GRPID);
933                         break;
934                 case Opt_resuid:
935                         if (match_int(&args[0], &option))
936                                 return 0;
937                         sbi->s_resuid = option;
938                         break;
939                 case Opt_resgid:
940                         if (match_int(&args[0], &option))
941                                 return 0;
942                         sbi->s_resgid = option;
943                         break;
944                 case Opt_sb:
945                         /* handled by get_sb_block() instead of here */
946                         /* *sb_block = match_int(&args[0]); */
947                         break;
948                 case Opt_err_panic:
949                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
950                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
951                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
952                         break;
953                 case Opt_err_ro:
954                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
955                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
956                         set_opt (sbi->s_mount_opt, ERRORS_RO);
957                         break;
958                 case Opt_err_cont:
959                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
960                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
961                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
962                         break;
963                 case Opt_nouid32:
964                         set_opt (sbi->s_mount_opt, NO_UID32);
965                         break;
966                 case Opt_nocheck:
967                         clear_opt (sbi->s_mount_opt, CHECK);
968                         break;
969                 case Opt_debug:
970                         set_opt (sbi->s_mount_opt, DEBUG);
971                         break;
972                 case Opt_oldalloc:
973                         set_opt (sbi->s_mount_opt, OLDALLOC);
974                         break;
975                 case Opt_orlov:
976                         clear_opt (sbi->s_mount_opt, OLDALLOC);
977                         break;
978 #ifdef CONFIG_EXT3_FS_XATTR
979                 case Opt_user_xattr:
980                         set_opt (sbi->s_mount_opt, XATTR_USER);
981                         break;
982                 case Opt_nouser_xattr:
983                         clear_opt (sbi->s_mount_opt, XATTR_USER);
984                         break;
985 #else
986                 case Opt_user_xattr:
987                 case Opt_nouser_xattr:
988                         ext3_msg(sb, KERN_INFO,
989                                 "(no)user_xattr options not supported");
990                         break;
991 #endif
992 #ifdef CONFIG_EXT3_FS_POSIX_ACL
993                 case Opt_acl:
994                         set_opt(sbi->s_mount_opt, POSIX_ACL);
995                         break;
996                 case Opt_noacl:
997                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
998                         break;
999 #else
1000                 case Opt_acl:
1001                 case Opt_noacl:
1002                         ext3_msg(sb, KERN_INFO,
1003                                 "(no)acl options not supported");
1004                         break;
1005 #endif
1006                 case Opt_reservation:
1007                         set_opt(sbi->s_mount_opt, RESERVATION);
1008                         break;
1009                 case Opt_noreservation:
1010                         clear_opt(sbi->s_mount_opt, RESERVATION);
1011                         break;
1012                 case Opt_journal_update:
1013                         /* @@@ FIXME */
1014                         /* Eventually we will want to be able to create
1015                            a journal file here.  For now, only allow the
1016                            user to specify an existing inode to be the
1017                            journal file. */
1018                         if (is_remount) {
1019                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1020                                         "journal on remount");
1021                                 return 0;
1022                         }
1023                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1024                         break;
1025                 case Opt_journal_inum:
1026                         if (is_remount) {
1027                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1028                                        "journal on remount");
1029                                 return 0;
1030                         }
1031                         if (match_int(&args[0], &option))
1032                                 return 0;
1033                         *inum = option;
1034                         break;
1035                 case Opt_journal_dev:
1036                         if (is_remount) {
1037                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1038                                        "journal on remount");
1039                                 return 0;
1040                         }
1041                         if (match_int(&args[0], &option))
1042                                 return 0;
1043                         *journal_devnum = option;
1044                         break;
1045                 case Opt_noload:
1046                         set_opt (sbi->s_mount_opt, NOLOAD);
1047                         break;
1048                 case Opt_commit:
1049                         if (match_int(&args[0], &option))
1050                                 return 0;
1051                         if (option < 0)
1052                                 return 0;
1053                         if (option == 0)
1054                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1055                         sbi->s_commit_interval = HZ * option;
1056                         break;
1057                 case Opt_data_journal:
1058                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1059                         goto datacheck;
1060                 case Opt_data_ordered:
1061                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1062                         goto datacheck;
1063                 case Opt_data_writeback:
1064                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1065                 datacheck:
1066                         if (is_remount) {
1067                                 if (test_opt(sb, DATA_FLAGS) == data_opt)
1068                                         break;
1069                                 ext3_msg(sb, KERN_ERR,
1070                                         "error: cannot change "
1071                                         "data mode on remount. The filesystem "
1072                                         "is mounted in data=%s mode and you "
1073                                         "try to remount it in data=%s mode.",
1074                                         data_mode_string(test_opt(sb,
1075                                                         DATA_FLAGS)),
1076                                         data_mode_string(data_opt));
1077                                 return 0;
1078                         } else {
1079                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1080                                 sbi->s_mount_opt |= data_opt;
1081                         }
1082                         break;
1083                 case Opt_data_err_abort:
1084                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1085                         break;
1086                 case Opt_data_err_ignore:
1087                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1088                         break;
1089 #ifdef CONFIG_QUOTA
1090                 case Opt_usrjquota:
1091                         qtype = USRQUOTA;
1092                         goto set_qf_name;
1093                 case Opt_grpjquota:
1094                         qtype = GRPQUOTA;
1095 set_qf_name:
1096                         if (sb_any_quota_loaded(sb) &&
1097                             !sbi->s_qf_names[qtype]) {
1098                                 ext3_msg(sb, KERN_ERR,
1099                                         "error: cannot change journaled "
1100                                         "quota options when quota turned on.");
1101                                 return 0;
1102                         }
1103                         qname = match_strdup(&args[0]);
1104                         if (!qname) {
1105                                 ext3_msg(sb, KERN_ERR,
1106                                         "error: not enough memory for "
1107                                         "storing quotafile name.");
1108                                 return 0;
1109                         }
1110                         if (sbi->s_qf_names[qtype] &&
1111                             strcmp(sbi->s_qf_names[qtype], qname)) {
1112                                 ext3_msg(sb, KERN_ERR,
1113                                         "error: %s quota file already "
1114                                         "specified.", QTYPE2NAME(qtype));
1115                                 kfree(qname);
1116                                 return 0;
1117                         }
1118                         sbi->s_qf_names[qtype] = qname;
1119                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1120                                 ext3_msg(sb, KERN_ERR,
1121                                         "error: quotafile must be on "
1122                                         "filesystem root.");
1123                                 kfree(sbi->s_qf_names[qtype]);
1124                                 sbi->s_qf_names[qtype] = NULL;
1125                                 return 0;
1126                         }
1127                         set_opt(sbi->s_mount_opt, QUOTA);
1128                         break;
1129                 case Opt_offusrjquota:
1130                         qtype = USRQUOTA;
1131                         goto clear_qf_name;
1132                 case Opt_offgrpjquota:
1133                         qtype = GRPQUOTA;
1134 clear_qf_name:
1135                         if (sb_any_quota_loaded(sb) &&
1136                             sbi->s_qf_names[qtype]) {
1137                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1138                                         "journaled quota options when "
1139                                         "quota turned on.");
1140                                 return 0;
1141                         }
1142                         /*
1143                          * The space will be released later when all options
1144                          * are confirmed to be correct
1145                          */
1146                         sbi->s_qf_names[qtype] = NULL;
1147                         break;
1148                 case Opt_jqfmt_vfsold:
1149                         qfmt = QFMT_VFS_OLD;
1150                         goto set_qf_format;
1151                 case Opt_jqfmt_vfsv0:
1152                         qfmt = QFMT_VFS_V0;
1153                         goto set_qf_format;
1154                 case Opt_jqfmt_vfsv1:
1155                         qfmt = QFMT_VFS_V1;
1156 set_qf_format:
1157                         if (sb_any_quota_loaded(sb) &&
1158                             sbi->s_jquota_fmt != qfmt) {
1159                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1160                                         "journaled quota options when "
1161                                         "quota turned on.");
1162                                 return 0;
1163                         }
1164                         sbi->s_jquota_fmt = qfmt;
1165                         break;
1166                 case Opt_quota:
1167                 case Opt_usrquota:
1168                         set_opt(sbi->s_mount_opt, QUOTA);
1169                         set_opt(sbi->s_mount_opt, USRQUOTA);
1170                         break;
1171                 case Opt_grpquota:
1172                         set_opt(sbi->s_mount_opt, QUOTA);
1173                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1174                         break;
1175                 case Opt_noquota:
1176                         if (sb_any_quota_loaded(sb)) {
1177                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1178                                         "quota options when quota turned on.");
1179                                 return 0;
1180                         }
1181                         clear_opt(sbi->s_mount_opt, QUOTA);
1182                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1183                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1184                         break;
1185 #else
1186                 case Opt_quota:
1187                 case Opt_usrquota:
1188                 case Opt_grpquota:
1189                         ext3_msg(sb, KERN_ERR,
1190                                 "error: quota options not supported.");
1191                         break;
1192                 case Opt_usrjquota:
1193                 case Opt_grpjquota:
1194                 case Opt_offusrjquota:
1195                 case Opt_offgrpjquota:
1196                 case Opt_jqfmt_vfsold:
1197                 case Opt_jqfmt_vfsv0:
1198                 case Opt_jqfmt_vfsv1:
1199                         ext3_msg(sb, KERN_ERR,
1200                                 "error: journaled quota options not "
1201                                 "supported.");
1202                         break;
1203                 case Opt_noquota:
1204                         break;
1205 #endif
1206                 case Opt_abort:
1207                         set_opt(sbi->s_mount_opt, ABORT);
1208                         break;
1209                 case Opt_barrier:
1210                         if (match_int(&args[0], &option))
1211                                 return 0;
1212                         if (option)
1213                                 set_opt(sbi->s_mount_opt, BARRIER);
1214                         else
1215                                 clear_opt(sbi->s_mount_opt, BARRIER);
1216                         break;
1217                 case Opt_ignore:
1218                         break;
1219                 case Opt_resize:
1220                         if (!is_remount) {
1221                                 ext3_msg(sb, KERN_ERR,
1222                                         "error: resize option only available "
1223                                         "for remount");
1224                                 return 0;
1225                         }
1226                         if (match_int(&args[0], &option) != 0)
1227                                 return 0;
1228                         *n_blocks_count = option;
1229                         break;
1230                 case Opt_nobh:
1231                         set_opt(sbi->s_mount_opt, NOBH);
1232                         break;
1233                 case Opt_bh:
1234                         clear_opt(sbi->s_mount_opt, NOBH);
1235                         break;
1236                 default:
1237                         ext3_msg(sb, KERN_ERR,
1238                                 "error: unrecognized mount option \"%s\" "
1239                                 "or missing value", p);
1240                         return 0;
1241                 }
1242         }
1243 #ifdef CONFIG_QUOTA
1244         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1245                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1246                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1247                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1248                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1249
1250                 if ((sbi->s_qf_names[USRQUOTA] && test_opt(sb, GRPQUOTA)) ||
1251                         (sbi->s_qf_names[GRPQUOTA] && test_opt(sb, USRQUOTA))) {
1252                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1253                                         "format mixing.");
1254                         return 0;
1255                 }
1256
1257                 if (!sbi->s_jquota_fmt) {
1258                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1259                                         "not specified.");
1260                         return 0;
1261                 }
1262         } else {
1263                 if (sbi->s_jquota_fmt) {
1264                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1265                                         "specified with no journaling "
1266                                         "enabled.");
1267                         return 0;
1268                 }
1269         }
1270 #endif
1271         return 1;
1272 }
1273
1274 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1275                             int read_only)
1276 {
1277         struct ext3_sb_info *sbi = EXT3_SB(sb);
1278         int res = 0;
1279
1280         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1281                 ext3_msg(sb, KERN_ERR,
1282                         "error: revision level too high, "
1283                         "forcing read-only mode");
1284                 res = MS_RDONLY;
1285         }
1286         if (read_only)
1287                 return res;
1288         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1289                 ext3_msg(sb, KERN_WARNING,
1290                         "warning: mounting unchecked fs, "
1291                         "running e2fsck is recommended");
1292         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1293                 ext3_msg(sb, KERN_WARNING,
1294                         "warning: mounting fs with errors, "
1295                         "running e2fsck is recommended");
1296         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1297                  le16_to_cpu(es->s_mnt_count) >=
1298                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1299                 ext3_msg(sb, KERN_WARNING,
1300                         "warning: maximal mount count reached, "
1301                         "running e2fsck is recommended");
1302         else if (le32_to_cpu(es->s_checkinterval) &&
1303                 (le32_to_cpu(es->s_lastcheck) +
1304                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1305                 ext3_msg(sb, KERN_WARNING,
1306                         "warning: checktime reached, "
1307                         "running e2fsck is recommended");
1308 #if 0
1309                 /* @@@ We _will_ want to clear the valid bit if we find
1310                    inconsistencies, to force a fsck at reboot.  But for
1311                    a plain journaled filesystem we can keep it set as
1312                    valid forever! :) */
1313         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1314 #endif
1315         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1316                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1317         le16_add_cpu(&es->s_mnt_count, 1);
1318         es->s_mtime = cpu_to_le32(get_seconds());
1319         ext3_update_dynamic_rev(sb);
1320         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1321
1322         ext3_commit_super(sb, es, 1);
1323         if (test_opt(sb, DEBUG))
1324                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1325                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1326                         sb->s_blocksize,
1327                         sbi->s_groups_count,
1328                         EXT3_BLOCKS_PER_GROUP(sb),
1329                         EXT3_INODES_PER_GROUP(sb),
1330                         sbi->s_mount_opt);
1331
1332         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1333                 char b[BDEVNAME_SIZE];
1334                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1335                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1336         } else {
1337                 ext3_msg(sb, KERN_INFO, "using internal journal");
1338         }
1339         return res;
1340 }
1341
1342 /* Called at mount-time, super-block is locked */
1343 static int ext3_check_descriptors(struct super_block *sb)
1344 {
1345         struct ext3_sb_info *sbi = EXT3_SB(sb);
1346         int i;
1347
1348         ext3_debug ("Checking group descriptors");
1349
1350         for (i = 0; i < sbi->s_groups_count; i++) {
1351                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1352                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1353                 ext3_fsblk_t last_block;
1354
1355                 if (i == sbi->s_groups_count - 1)
1356                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1357                 else
1358                         last_block = first_block +
1359                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1360
1361                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1362                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1363                 {
1364                         ext3_error (sb, "ext3_check_descriptors",
1365                                     "Block bitmap for group %d"
1366                                     " not in group (block %lu)!",
1367                                     i, (unsigned long)
1368                                         le32_to_cpu(gdp->bg_block_bitmap));
1369                         return 0;
1370                 }
1371                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1372                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1373                 {
1374                         ext3_error (sb, "ext3_check_descriptors",
1375                                     "Inode bitmap for group %d"
1376                                     " not in group (block %lu)!",
1377                                     i, (unsigned long)
1378                                         le32_to_cpu(gdp->bg_inode_bitmap));
1379                         return 0;
1380                 }
1381                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1382                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1383                     last_block)
1384                 {
1385                         ext3_error (sb, "ext3_check_descriptors",
1386                                     "Inode table for group %d"
1387                                     " not in group (block %lu)!",
1388                                     i, (unsigned long)
1389                                         le32_to_cpu(gdp->bg_inode_table));
1390                         return 0;
1391                 }
1392         }
1393
1394         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1395         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1396         return 1;
1397 }
1398
1399
1400 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1401  * the superblock) which were deleted from all directories, but held open by
1402  * a process at the time of a crash.  We walk the list and try to delete these
1403  * inodes at recovery time (only with a read-write filesystem).
1404  *
1405  * In order to keep the orphan inode chain consistent during traversal (in
1406  * case of crash during recovery), we link each inode into the superblock
1407  * orphan list_head and handle it the same way as an inode deletion during
1408  * normal operation (which journals the operations for us).
1409  *
1410  * We only do an iget() and an iput() on each inode, which is very safe if we
1411  * accidentally point at an in-use or already deleted inode.  The worst that
1412  * can happen in this case is that we get a "bit already cleared" message from
1413  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1414  * e2fsck was run on this filesystem, and it must have already done the orphan
1415  * inode cleanup for us, so we can safely abort without any further action.
1416  */
1417 static void ext3_orphan_cleanup (struct super_block * sb,
1418                                  struct ext3_super_block * es)
1419 {
1420         unsigned int s_flags = sb->s_flags;
1421         int nr_orphans = 0, nr_truncates = 0;
1422 #ifdef CONFIG_QUOTA
1423         int i;
1424 #endif
1425         if (!es->s_last_orphan) {
1426                 jbd_debug(4, "no orphan inodes to clean up\n");
1427                 return;
1428         }
1429
1430         if (bdev_read_only(sb->s_bdev)) {
1431                 ext3_msg(sb, KERN_ERR, "error: write access "
1432                         "unavailable, skipping orphan cleanup.");
1433                 return;
1434         }
1435
1436         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1437                 if (es->s_last_orphan)
1438                         jbd_debug(1, "Errors on filesystem, "
1439                                   "clearing orphan list.\n");
1440                 es->s_last_orphan = 0;
1441                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1442                 return;
1443         }
1444
1445         if (s_flags & MS_RDONLY) {
1446                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1447                 sb->s_flags &= ~MS_RDONLY;
1448         }
1449 #ifdef CONFIG_QUOTA
1450         /* Needed for iput() to work correctly and not trash data */
1451         sb->s_flags |= MS_ACTIVE;
1452         /* Turn on quotas so that they are updated correctly */
1453         for (i = 0; i < MAXQUOTAS; i++) {
1454                 if (EXT3_SB(sb)->s_qf_names[i]) {
1455                         int ret = ext3_quota_on_mount(sb, i);
1456                         if (ret < 0)
1457                                 ext3_msg(sb, KERN_ERR,
1458                                         "error: cannot turn on journaled "
1459                                         "quota: %d", ret);
1460                 }
1461         }
1462 #endif
1463
1464         while (es->s_last_orphan) {
1465                 struct inode *inode;
1466
1467                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1468                 if (IS_ERR(inode)) {
1469                         es->s_last_orphan = 0;
1470                         break;
1471                 }
1472
1473                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1474                 vfs_dq_init(inode);
1475                 if (inode->i_nlink) {
1476                         printk(KERN_DEBUG
1477                                 "%s: truncating inode %lu to %Ld bytes\n",
1478                                 __func__, inode->i_ino, inode->i_size);
1479                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1480                                   inode->i_ino, inode->i_size);
1481                         ext3_truncate(inode);
1482                         nr_truncates++;
1483                 } else {
1484                         printk(KERN_DEBUG
1485                                 "%s: deleting unreferenced inode %lu\n",
1486                                 __func__, inode->i_ino);
1487                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1488                                   inode->i_ino);
1489                         nr_orphans++;
1490                 }
1491                 iput(inode);  /* The delete magic happens here! */
1492         }
1493
1494 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1495
1496         if (nr_orphans)
1497                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1498                        PLURAL(nr_orphans));
1499         if (nr_truncates)
1500                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1501                        PLURAL(nr_truncates));
1502 #ifdef CONFIG_QUOTA
1503         /* Turn quotas off */
1504         for (i = 0; i < MAXQUOTAS; i++) {
1505                 if (sb_dqopt(sb)->files[i])
1506                         vfs_quota_off(sb, i, 0);
1507         }
1508 #endif
1509         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1510 }
1511
1512 /*
1513  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1514  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1515  * We need to be 1 filesystem block less than the 2^32 sector limit.
1516  */
1517 static loff_t ext3_max_size(int bits)
1518 {
1519         loff_t res = EXT3_NDIR_BLOCKS;
1520         int meta_blocks;
1521         loff_t upper_limit;
1522
1523         /* This is calculated to be the largest file size for a
1524          * dense, file such that the total number of
1525          * sectors in the file, including data and all indirect blocks,
1526          * does not exceed 2^32 -1
1527          * __u32 i_blocks representing the total number of
1528          * 512 bytes blocks of the file
1529          */
1530         upper_limit = (1LL << 32) - 1;
1531
1532         /* total blocks in file system block size */
1533         upper_limit >>= (bits - 9);
1534
1535
1536         /* indirect blocks */
1537         meta_blocks = 1;
1538         /* double indirect blocks */
1539         meta_blocks += 1 + (1LL << (bits-2));
1540         /* tripple indirect blocks */
1541         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1542
1543         upper_limit -= meta_blocks;
1544         upper_limit <<= bits;
1545
1546         res += 1LL << (bits-2);
1547         res += 1LL << (2*(bits-2));
1548         res += 1LL << (3*(bits-2));
1549         res <<= bits;
1550         if (res > upper_limit)
1551                 res = upper_limit;
1552
1553         if (res > MAX_LFS_FILESIZE)
1554                 res = MAX_LFS_FILESIZE;
1555
1556         return res;
1557 }
1558
1559 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1560                                     ext3_fsblk_t logic_sb_block,
1561                                     int nr)
1562 {
1563         struct ext3_sb_info *sbi = EXT3_SB(sb);
1564         unsigned long bg, first_meta_bg;
1565         int has_super = 0;
1566
1567         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1568
1569         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1570             nr < first_meta_bg)
1571                 return (logic_sb_block + nr + 1);
1572         bg = sbi->s_desc_per_block * nr;
1573         if (ext3_bg_has_super(sb, bg))
1574                 has_super = 1;
1575         return (has_super + ext3_group_first_block_no(sb, bg));
1576 }
1577
1578
1579 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1580 {
1581         struct buffer_head * bh;
1582         struct ext3_super_block *es = NULL;
1583         struct ext3_sb_info *sbi;
1584         ext3_fsblk_t block;
1585         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1586         ext3_fsblk_t logic_sb_block;
1587         unsigned long offset = 0;
1588         unsigned int journal_inum = 0;
1589         unsigned long journal_devnum = 0;
1590         unsigned long def_mount_opts;
1591         struct inode *root;
1592         int blocksize;
1593         int hblock;
1594         int db_count;
1595         int i;
1596         int needs_recovery;
1597         int ret = -EINVAL;
1598         __le32 features;
1599         int err;
1600
1601         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1602         if (!sbi)
1603                 return -ENOMEM;
1604
1605         sbi->s_blockgroup_lock =
1606                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1607         if (!sbi->s_blockgroup_lock) {
1608                 kfree(sbi);
1609                 return -ENOMEM;
1610         }
1611         sb->s_fs_info = sbi;
1612         sbi->s_mount_opt = 0;
1613         sbi->s_resuid = EXT3_DEF_RESUID;
1614         sbi->s_resgid = EXT3_DEF_RESGID;
1615         sbi->s_sb_block = sb_block;
1616
1617         unlock_kernel();
1618
1619         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1620         if (!blocksize) {
1621                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1622                 goto out_fail;
1623         }
1624
1625         /*
1626          * The ext3 superblock will not be buffer aligned for other than 1kB
1627          * block sizes.  We need to calculate the offset from buffer start.
1628          */
1629         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1630                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1631                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1632         } else {
1633                 logic_sb_block = sb_block;
1634         }
1635
1636         if (!(bh = sb_bread(sb, logic_sb_block))) {
1637                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1638                 goto out_fail;
1639         }
1640         /*
1641          * Note: s_es must be initialized as soon as possible because
1642          *       some ext3 macro-instructions depend on its value
1643          */
1644         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1645         sbi->s_es = es;
1646         sb->s_magic = le16_to_cpu(es->s_magic);
1647         if (sb->s_magic != EXT3_SUPER_MAGIC)
1648                 goto cantfind_ext3;
1649
1650         /* Set defaults before we parse the mount options */
1651         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1652         if (def_mount_opts & EXT3_DEFM_DEBUG)
1653                 set_opt(sbi->s_mount_opt, DEBUG);
1654         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1655                 set_opt(sbi->s_mount_opt, GRPID);
1656         if (def_mount_opts & EXT3_DEFM_UID16)
1657                 set_opt(sbi->s_mount_opt, NO_UID32);
1658 #ifdef CONFIG_EXT3_FS_XATTR
1659         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1660                 set_opt(sbi->s_mount_opt, XATTR_USER);
1661 #endif
1662 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1663         if (def_mount_opts & EXT3_DEFM_ACL)
1664                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1665 #endif
1666         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1667                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1668         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1669                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1670         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1671                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1672
1673         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1674                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1675         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1676                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1677         else
1678                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1679
1680         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1681         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1682
1683         set_opt(sbi->s_mount_opt, RESERVATION);
1684
1685         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1686                             NULL, 0))
1687                 goto failed_mount;
1688
1689         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1690                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1691
1692         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1693             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1694              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1695              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1696                 ext3_msg(sb, KERN_WARNING,
1697                         "warning: feature flags set on rev 0 fs, "
1698                         "running e2fsck is recommended");
1699         /*
1700          * Check feature flags regardless of the revision level, since we
1701          * previously didn't change the revision level when setting the flags,
1702          * so there is a chance incompat flags are set on a rev 0 filesystem.
1703          */
1704         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1705         if (features) {
1706                 ext3_msg(sb, KERN_ERR,
1707                         "error: couldn't mount because of unsupported "
1708                         "optional features (%x)", le32_to_cpu(features));
1709                 goto failed_mount;
1710         }
1711         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1712         if (!(sb->s_flags & MS_RDONLY) && features) {
1713                 ext3_msg(sb, KERN_ERR,
1714                         "error: couldn't mount RDWR because of unsupported "
1715                         "optional features (%x)", le32_to_cpu(features));
1716                 goto failed_mount;
1717         }
1718         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1719
1720         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1721             blocksize > EXT3_MAX_BLOCK_SIZE) {
1722                 ext3_msg(sb, KERN_ERR,
1723                         "error: couldn't mount because of unsupported "
1724                         "filesystem blocksize %d", blocksize);
1725                 goto failed_mount;
1726         }
1727
1728         hblock = bdev_logical_block_size(sb->s_bdev);
1729         if (sb->s_blocksize != blocksize) {
1730                 /*
1731                  * Make sure the blocksize for the filesystem is larger
1732                  * than the hardware sectorsize for the machine.
1733                  */
1734                 if (blocksize < hblock) {
1735                         ext3_msg(sb, KERN_ERR,
1736                                 "error: fsblocksize %d too small for "
1737                                 "hardware sectorsize %d", blocksize, hblock);
1738                         goto failed_mount;
1739                 }
1740
1741                 brelse (bh);
1742                 if (!sb_set_blocksize(sb, blocksize)) {
1743                         ext3_msg(sb, KERN_ERR,
1744                                 "error: bad blocksize %d", blocksize);
1745                         goto out_fail;
1746                 }
1747                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1748                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1749                 bh = sb_bread(sb, logic_sb_block);
1750                 if (!bh) {
1751                         ext3_msg(sb, KERN_ERR,
1752                                "error: can't read superblock on 2nd try");
1753                         goto failed_mount;
1754                 }
1755                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1756                 sbi->s_es = es;
1757                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1758                         ext3_msg(sb, KERN_ERR,
1759                                 "error: magic mismatch");
1760                         goto failed_mount;
1761                 }
1762         }
1763
1764         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1765
1766         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1767                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1768                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1769         } else {
1770                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1771                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1772                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1773                     (!is_power_of_2(sbi->s_inode_size)) ||
1774                     (sbi->s_inode_size > blocksize)) {
1775                         ext3_msg(sb, KERN_ERR,
1776                                 "error: unsupported inode size: %d",
1777                                 sbi->s_inode_size);
1778                         goto failed_mount;
1779                 }
1780         }
1781         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1782                                    le32_to_cpu(es->s_log_frag_size);
1783         if (blocksize != sbi->s_frag_size) {
1784                 ext3_msg(sb, KERN_ERR,
1785                        "error: fragsize %lu != blocksize %u (unsupported)",
1786                        sbi->s_frag_size, blocksize);
1787                 goto failed_mount;
1788         }
1789         sbi->s_frags_per_block = 1;
1790         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1791         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1792         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1793         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1794                 goto cantfind_ext3;
1795         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1796         if (sbi->s_inodes_per_block == 0)
1797                 goto cantfind_ext3;
1798         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1799                                         sbi->s_inodes_per_block;
1800         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1801         sbi->s_sbh = bh;
1802         sbi->s_mount_state = le16_to_cpu(es->s_state);
1803         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1804         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1805         for (i=0; i < 4; i++)
1806                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1807         sbi->s_def_hash_version = es->s_def_hash_version;
1808         i = le32_to_cpu(es->s_flags);
1809         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1810                 sbi->s_hash_unsigned = 3;
1811         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1812 #ifdef __CHAR_UNSIGNED__
1813                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1814                 sbi->s_hash_unsigned = 3;
1815 #else
1816                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1817 #endif
1818         }
1819
1820         if (sbi->s_blocks_per_group > blocksize * 8) {
1821                 ext3_msg(sb, KERN_ERR,
1822                         "#blocks per group too big: %lu",
1823                         sbi->s_blocks_per_group);
1824                 goto failed_mount;
1825         }
1826         if (sbi->s_frags_per_group > blocksize * 8) {
1827                 ext3_msg(sb, KERN_ERR,
1828                         "error: #fragments per group too big: %lu",
1829                         sbi->s_frags_per_group);
1830                 goto failed_mount;
1831         }
1832         if (sbi->s_inodes_per_group > blocksize * 8) {
1833                 ext3_msg(sb, KERN_ERR,
1834                         "error: #inodes per group too big: %lu",
1835                         sbi->s_inodes_per_group);
1836                 goto failed_mount;
1837         }
1838
1839         if (le32_to_cpu(es->s_blocks_count) >
1840                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1841                 ext3_msg(sb, KERN_ERR,
1842                         "error: filesystem is too large to mount safely");
1843                 if (sizeof(sector_t) < 8)
1844                         ext3_msg(sb, KERN_ERR,
1845                                 "error: CONFIG_LBDAF not enabled");
1846                 goto failed_mount;
1847         }
1848
1849         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1850                 goto cantfind_ext3;
1851         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1852                                le32_to_cpu(es->s_first_data_block) - 1)
1853                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1854         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1855                    EXT3_DESC_PER_BLOCK(sb);
1856         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1857                                     GFP_KERNEL);
1858         if (sbi->s_group_desc == NULL) {
1859                 ext3_msg(sb, KERN_ERR,
1860                         "error: not enough memory");
1861                 goto failed_mount;
1862         }
1863
1864         bgl_lock_init(sbi->s_blockgroup_lock);
1865
1866         for (i = 0; i < db_count; i++) {
1867                 block = descriptor_loc(sb, logic_sb_block, i);
1868                 sbi->s_group_desc[i] = sb_bread(sb, block);
1869                 if (!sbi->s_group_desc[i]) {
1870                         ext3_msg(sb, KERN_ERR,
1871                                 "error: can't read group descriptor %d", i);
1872                         db_count = i;
1873                         goto failed_mount2;
1874                 }
1875         }
1876         if (!ext3_check_descriptors (sb)) {
1877                 ext3_msg(sb, KERN_ERR,
1878                         "error: group descriptors corrupted");
1879                 goto failed_mount2;
1880         }
1881         sbi->s_gdb_count = db_count;
1882         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1883         spin_lock_init(&sbi->s_next_gen_lock);
1884
1885         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1886                         ext3_count_free_blocks(sb));
1887         if (!err) {
1888                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1889                                 ext3_count_free_inodes(sb));
1890         }
1891         if (!err) {
1892                 err = percpu_counter_init(&sbi->s_dirs_counter,
1893                                 ext3_count_dirs(sb));
1894         }
1895         if (err) {
1896                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
1897                 goto failed_mount3;
1898         }
1899
1900         /* per fileystem reservation list head & lock */
1901         spin_lock_init(&sbi->s_rsv_window_lock);
1902         sbi->s_rsv_window_root = RB_ROOT;
1903         /* Add a single, static dummy reservation to the start of the
1904          * reservation window list --- it gives us a placeholder for
1905          * append-at-start-of-list which makes the allocation logic
1906          * _much_ simpler. */
1907         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1908         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1909         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1910         sbi->s_rsv_window_head.rsv_goal_size = 0;
1911         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1912
1913         /*
1914          * set up enough so that it can read an inode
1915          */
1916         sb->s_op = &ext3_sops;
1917         sb->s_export_op = &ext3_export_ops;
1918         sb->s_xattr = ext3_xattr_handlers;
1919 #ifdef CONFIG_QUOTA
1920         sb->s_qcop = &ext3_qctl_operations;
1921         sb->dq_op = &ext3_quota_operations;
1922 #endif
1923         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1924         mutex_init(&sbi->s_orphan_lock);
1925         mutex_init(&sbi->s_resize_lock);
1926
1927         sb->s_root = NULL;
1928
1929         needs_recovery = (es->s_last_orphan != 0 ||
1930                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1931                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1932
1933         /*
1934          * The first inode we look at is the journal inode.  Don't try
1935          * root first: it may be modified in the journal!
1936          */
1937         if (!test_opt(sb, NOLOAD) &&
1938             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1939                 if (ext3_load_journal(sb, es, journal_devnum))
1940                         goto failed_mount3;
1941         } else if (journal_inum) {
1942                 if (ext3_create_journal(sb, es, journal_inum))
1943                         goto failed_mount3;
1944         } else {
1945                 if (!silent)
1946                         ext3_msg(sb, KERN_ERR,
1947                                 "error: no journal found. "
1948                                 "mounting ext3 over ext2?");
1949                 goto failed_mount3;
1950         }
1951
1952         /* We have now updated the journal if required, so we can
1953          * validate the data journaling mode. */
1954         switch (test_opt(sb, DATA_FLAGS)) {
1955         case 0:
1956                 /* No mode set, assume a default based on the journal
1957                    capabilities: ORDERED_DATA if the journal can
1958                    cope, else JOURNAL_DATA */
1959                 if (journal_check_available_features
1960                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1961                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
1962                 else
1963                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1964                 break;
1965
1966         case EXT3_MOUNT_ORDERED_DATA:
1967         case EXT3_MOUNT_WRITEBACK_DATA:
1968                 if (!journal_check_available_features
1969                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1970                         ext3_msg(sb, KERN_ERR,
1971                                 "error: journal does not support "
1972                                 "requested data journaling mode");
1973                         goto failed_mount4;
1974                 }
1975         default:
1976                 break;
1977         }
1978
1979         if (test_opt(sb, NOBH)) {
1980                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1981                         ext3_msg(sb, KERN_WARNING,
1982                                 "warning: ignoring nobh option - "
1983                                 "it is supported only with writeback mode");
1984                         clear_opt(sbi->s_mount_opt, NOBH);
1985                 }
1986         }
1987         /*
1988          * The journal_load will have done any necessary log recovery,
1989          * so we can safely mount the rest of the filesystem now.
1990          */
1991
1992         root = ext3_iget(sb, EXT3_ROOT_INO);
1993         if (IS_ERR(root)) {
1994                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
1995                 ret = PTR_ERR(root);
1996                 goto failed_mount4;
1997         }
1998         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1999                 iput(root);
2000                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2001                 goto failed_mount4;
2002         }
2003         sb->s_root = d_alloc_root(root);
2004         if (!sb->s_root) {
2005                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2006                 iput(root);
2007                 ret = -ENOMEM;
2008                 goto failed_mount4;
2009         }
2010
2011         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2012
2013         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2014         ext3_orphan_cleanup(sb, es);
2015         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2016         if (needs_recovery)
2017                 ext3_msg(sb, KERN_INFO, "recovery complete");
2018         ext3_mark_recovery_complete(sb, es);
2019         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2020                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2021                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2022                 "writeback");
2023
2024         lock_kernel();
2025         return 0;
2026
2027 cantfind_ext3:
2028         if (!silent)
2029                 ext3_msg(sb, KERN_INFO,
2030                         "error: can't find ext3 filesystem on dev %s.",
2031                        sb->s_id);
2032         goto failed_mount;
2033
2034 failed_mount4:
2035         journal_destroy(sbi->s_journal);
2036 failed_mount3:
2037         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2038         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2039         percpu_counter_destroy(&sbi->s_dirs_counter);
2040 failed_mount2:
2041         for (i = 0; i < db_count; i++)
2042                 brelse(sbi->s_group_desc[i]);
2043         kfree(sbi->s_group_desc);
2044 failed_mount:
2045 #ifdef CONFIG_QUOTA
2046         for (i = 0; i < MAXQUOTAS; i++)
2047                 kfree(sbi->s_qf_names[i]);
2048 #endif
2049         ext3_blkdev_remove(sbi);
2050         brelse(bh);
2051 out_fail:
2052         sb->s_fs_info = NULL;
2053         kfree(sbi->s_blockgroup_lock);
2054         kfree(sbi);
2055         lock_kernel();
2056         return ret;
2057 }
2058
2059 /*
2060  * Setup any per-fs journal parameters now.  We'll do this both on
2061  * initial mount, once the journal has been initialised but before we've
2062  * done any recovery; and again on any subsequent remount.
2063  */
2064 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2065 {
2066         struct ext3_sb_info *sbi = EXT3_SB(sb);
2067
2068         if (sbi->s_commit_interval)
2069                 journal->j_commit_interval = sbi->s_commit_interval;
2070         /* We could also set up an ext3-specific default for the commit
2071          * interval here, but for now we'll just fall back to the jbd
2072          * default. */
2073
2074         spin_lock(&journal->j_state_lock);
2075         if (test_opt(sb, BARRIER))
2076                 journal->j_flags |= JFS_BARRIER;
2077         else
2078                 journal->j_flags &= ~JFS_BARRIER;
2079         if (test_opt(sb, DATA_ERR_ABORT))
2080                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2081         else
2082                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2083         spin_unlock(&journal->j_state_lock);
2084 }
2085
2086 static journal_t *ext3_get_journal(struct super_block *sb,
2087                                    unsigned int journal_inum)
2088 {
2089         struct inode *journal_inode;
2090         journal_t *journal;
2091
2092         /* First, test for the existence of a valid inode on disk.  Bad
2093          * things happen if we iget() an unused inode, as the subsequent
2094          * iput() will try to delete it. */
2095
2096         journal_inode = ext3_iget(sb, journal_inum);
2097         if (IS_ERR(journal_inode)) {
2098                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2099                 return NULL;
2100         }
2101         if (!journal_inode->i_nlink) {
2102                 make_bad_inode(journal_inode);
2103                 iput(journal_inode);
2104                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2105                 return NULL;
2106         }
2107
2108         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2109                   journal_inode, journal_inode->i_size);
2110         if (!S_ISREG(journal_inode->i_mode)) {
2111                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2112                 iput(journal_inode);
2113                 return NULL;
2114         }
2115
2116         journal = journal_init_inode(journal_inode);
2117         if (!journal) {
2118                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2119                 iput(journal_inode);
2120                 return NULL;
2121         }
2122         journal->j_private = sb;
2123         ext3_init_journal_params(sb, journal);
2124         return journal;
2125 }
2126
2127 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2128                                        dev_t j_dev)
2129 {
2130         struct buffer_head * bh;
2131         journal_t *journal;
2132         ext3_fsblk_t start;
2133         ext3_fsblk_t len;
2134         int hblock, blocksize;
2135         ext3_fsblk_t sb_block;
2136         unsigned long offset;
2137         struct ext3_super_block * es;
2138         struct block_device *bdev;
2139
2140         bdev = ext3_blkdev_get(j_dev, sb);
2141         if (bdev == NULL)
2142                 return NULL;
2143
2144         if (bd_claim(bdev, sb)) {
2145                 ext3_msg(sb, KERN_ERR,
2146                         "error: failed to claim external journal device");
2147                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2148                 return NULL;
2149         }
2150
2151         blocksize = sb->s_blocksize;
2152         hblock = bdev_logical_block_size(bdev);
2153         if (blocksize < hblock) {
2154                 ext3_msg(sb, KERN_ERR,
2155                         "error: blocksize too small for journal device");
2156                 goto out_bdev;
2157         }
2158
2159         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2160         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2161         set_blocksize(bdev, blocksize);
2162         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2163                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2164                         "external journal");
2165                 goto out_bdev;
2166         }
2167
2168         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2169         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2170             !(le32_to_cpu(es->s_feature_incompat) &
2171               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2172                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2173                         "bad superblock");
2174                 brelse(bh);
2175                 goto out_bdev;
2176         }
2177
2178         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2179                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2180                 brelse(bh);
2181                 goto out_bdev;
2182         }
2183
2184         len = le32_to_cpu(es->s_blocks_count);
2185         start = sb_block + 1;
2186         brelse(bh);     /* we're done with the superblock */
2187
2188         journal = journal_init_dev(bdev, sb->s_bdev,
2189                                         start, len, blocksize);
2190         if (!journal) {
2191                 ext3_msg(sb, KERN_ERR,
2192                         "error: failed to create device journal");
2193                 goto out_bdev;
2194         }
2195         journal->j_private = sb;
2196         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2197         wait_on_buffer(journal->j_sb_buffer);
2198         if (!buffer_uptodate(journal->j_sb_buffer)) {
2199                 ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2200                 goto out_journal;
2201         }
2202         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2203                 ext3_msg(sb, KERN_ERR,
2204                         "error: external journal has more than one "
2205                         "user (unsupported) - %d",
2206                         be32_to_cpu(journal->j_superblock->s_nr_users));
2207                 goto out_journal;
2208         }
2209         EXT3_SB(sb)->journal_bdev = bdev;
2210         ext3_init_journal_params(sb, journal);
2211         return journal;
2212 out_journal:
2213         journal_destroy(journal);
2214 out_bdev:
2215         ext3_blkdev_put(bdev);
2216         return NULL;
2217 }
2218
2219 static int ext3_load_journal(struct super_block *sb,
2220                              struct ext3_super_block *es,
2221                              unsigned long journal_devnum)
2222 {
2223         journal_t *journal;
2224         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2225         dev_t journal_dev;
2226         int err = 0;
2227         int really_read_only;
2228
2229         if (journal_devnum &&
2230             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2231                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2232                         "numbers have changed");
2233                 journal_dev = new_decode_dev(journal_devnum);
2234         } else
2235                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2236
2237         really_read_only = bdev_read_only(sb->s_bdev);
2238
2239         /*
2240          * Are we loading a blank journal or performing recovery after a
2241          * crash?  For recovery, we need to check in advance whether we
2242          * can get read-write access to the device.
2243          */
2244
2245         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2246                 if (sb->s_flags & MS_RDONLY) {
2247                         ext3_msg(sb, KERN_INFO,
2248                                 "recovery required on readonly filesystem");
2249                         if (really_read_only) {
2250                                 ext3_msg(sb, KERN_ERR, "error: write access "
2251                                         "unavailable, cannot proceed");
2252                                 return -EROFS;
2253                         }
2254                         ext3_msg(sb, KERN_INFO,
2255                                 "write access will be enabled during recovery");
2256                 }
2257         }
2258
2259         if (journal_inum && journal_dev) {
2260                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2261                        "and inode journals");
2262                 return -EINVAL;
2263         }
2264
2265         if (journal_inum) {
2266                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2267                         return -EINVAL;
2268         } else {
2269                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2270                         return -EINVAL;
2271         }
2272
2273         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2274                 err = journal_update_format(journal);
2275                 if (err)  {
2276                         ext3_msg(sb, KERN_ERR, "error updating journal");
2277                         journal_destroy(journal);
2278                         return err;
2279                 }
2280         }
2281
2282         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2283                 err = journal_wipe(journal, !really_read_only);
2284         if (!err)
2285                 err = journal_load(journal);
2286
2287         if (err) {
2288                 ext3_msg(sb, KERN_ERR, "error loading journal");
2289                 journal_destroy(journal);
2290                 return err;
2291         }
2292
2293         EXT3_SB(sb)->s_journal = journal;
2294         ext3_clear_journal_err(sb, es);
2295
2296         if (journal_devnum &&
2297             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2298                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2299
2300                 /* Make sure we flush the recovery flag to disk. */
2301                 ext3_commit_super(sb, es, 1);
2302         }
2303
2304         return 0;
2305 }
2306
2307 static int ext3_create_journal(struct super_block *sb,
2308                                struct ext3_super_block *es,
2309                                unsigned int journal_inum)
2310 {
2311         journal_t *journal;
2312         int err;
2313
2314         if (sb->s_flags & MS_RDONLY) {
2315                 ext3_msg(sb, KERN_ERR,
2316                         "error: readonly filesystem when trying to "
2317                         "create journal");
2318                 return -EROFS;
2319         }
2320
2321         journal = ext3_get_journal(sb, journal_inum);
2322         if (!journal)
2323                 return -EINVAL;
2324
2325         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2326                journal_inum);
2327
2328         err = journal_create(journal);
2329         if (err) {
2330                 ext3_msg(sb, KERN_ERR, "error creating journal");
2331                 journal_destroy(journal);
2332                 return -EIO;
2333         }
2334
2335         EXT3_SB(sb)->s_journal = journal;
2336
2337         ext3_update_dynamic_rev(sb);
2338         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2339         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2340
2341         es->s_journal_inum = cpu_to_le32(journal_inum);
2342
2343         /* Make sure we flush the recovery flag to disk. */
2344         ext3_commit_super(sb, es, 1);
2345
2346         return 0;
2347 }
2348
2349 static int ext3_commit_super(struct super_block *sb,
2350                                struct ext3_super_block *es,
2351                                int sync)
2352 {
2353         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2354         int error = 0;
2355
2356         if (!sbh)
2357                 return error;
2358         /*
2359          * If the file system is mounted read-only, don't update the
2360          * superblock write time.  This avoids updating the superblock
2361          * write time when we are mounting the root file system
2362          * read/only but we need to replay the journal; at that point,
2363          * for people who are east of GMT and who make their clock
2364          * tick in localtime for Windows bug-for-bug compatibility,
2365          * the clock is set in the future, and this will cause e2fsck
2366          * to complain and force a full file system check.
2367          */
2368         if (!(sb->s_flags & MS_RDONLY))
2369                 es->s_wtime = cpu_to_le32(get_seconds());
2370         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2371         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2372         BUFFER_TRACE(sbh, "marking dirty");
2373         mark_buffer_dirty(sbh);
2374         if (sync)
2375                 error = sync_dirty_buffer(sbh);
2376         return error;
2377 }
2378
2379
2380 /*
2381  * Have we just finished recovery?  If so, and if we are mounting (or
2382  * remounting) the filesystem readonly, then we will end up with a
2383  * consistent fs on disk.  Record that fact.
2384  */
2385 static void ext3_mark_recovery_complete(struct super_block * sb,
2386                                         struct ext3_super_block * es)
2387 {
2388         journal_t *journal = EXT3_SB(sb)->s_journal;
2389
2390         journal_lock_updates(journal);
2391         if (journal_flush(journal) < 0)
2392                 goto out;
2393
2394         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2395             sb->s_flags & MS_RDONLY) {
2396                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2397                 ext3_commit_super(sb, es, 1);
2398         }
2399
2400 out:
2401         journal_unlock_updates(journal);
2402 }
2403
2404 /*
2405  * If we are mounting (or read-write remounting) a filesystem whose journal
2406  * has recorded an error from a previous lifetime, move that error to the
2407  * main filesystem now.
2408  */
2409 static void ext3_clear_journal_err(struct super_block *sb,
2410                                    struct ext3_super_block *es)
2411 {
2412         journal_t *journal;
2413         int j_errno;
2414         const char *errstr;
2415
2416         journal = EXT3_SB(sb)->s_journal;
2417
2418         /*
2419          * Now check for any error status which may have been recorded in the
2420          * journal by a prior ext3_error() or ext3_abort()
2421          */
2422
2423         j_errno = journal_errno(journal);
2424         if (j_errno) {
2425                 char nbuf[16];
2426
2427                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2428                 ext3_warning(sb, __func__, "Filesystem error recorded "
2429                              "from previous mount: %s", errstr);
2430                 ext3_warning(sb, __func__, "Marking fs in need of "
2431                              "filesystem check.");
2432
2433                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2434                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2435                 ext3_commit_super (sb, es, 1);
2436
2437                 journal_clear_err(journal);
2438         }
2439 }
2440
2441 /*
2442  * Force the running and committing transactions to commit,
2443  * and wait on the commit.
2444  */
2445 int ext3_force_commit(struct super_block *sb)
2446 {
2447         journal_t *journal;
2448         int ret;
2449
2450         if (sb->s_flags & MS_RDONLY)
2451                 return 0;
2452
2453         journal = EXT3_SB(sb)->s_journal;
2454         ret = ext3_journal_force_commit(journal);
2455         return ret;
2456 }
2457
2458 static int ext3_sync_fs(struct super_block *sb, int wait)
2459 {
2460         tid_t target;
2461
2462         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2463                 if (wait)
2464                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2465         }
2466         return 0;
2467 }
2468
2469 /*
2470  * LVM calls this function before a (read-only) snapshot is created.  This
2471  * gives us a chance to flush the journal completely and mark the fs clean.
2472  */
2473 static int ext3_freeze(struct super_block *sb)
2474 {
2475         int error = 0;
2476         journal_t *journal;
2477
2478         if (!(sb->s_flags & MS_RDONLY)) {
2479                 journal = EXT3_SB(sb)->s_journal;
2480
2481                 /* Now we set up the journal barrier. */
2482                 journal_lock_updates(journal);
2483
2484                 /*
2485                  * We don't want to clear needs_recovery flag when we failed
2486                  * to flush the journal.
2487                  */
2488                 error = journal_flush(journal);
2489                 if (error < 0)
2490                         goto out;
2491
2492                 /* Journal blocked and flushed, clear needs_recovery flag. */
2493                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2494                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2495                 if (error)
2496                         goto out;
2497         }
2498         return 0;
2499
2500 out:
2501         journal_unlock_updates(journal);
2502         return error;
2503 }
2504
2505 /*
2506  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2507  * flag here, even though the filesystem is not technically dirty yet.
2508  */
2509 static int ext3_unfreeze(struct super_block *sb)
2510 {
2511         if (!(sb->s_flags & MS_RDONLY)) {
2512                 lock_super(sb);
2513                 /* Reser the needs_recovery flag before the fs is unlocked. */
2514                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2515                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2516                 unlock_super(sb);
2517                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2518         }
2519         return 0;
2520 }
2521
2522 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2523 {
2524         struct ext3_super_block * es;
2525         struct ext3_sb_info *sbi = EXT3_SB(sb);
2526         ext3_fsblk_t n_blocks_count = 0;
2527         unsigned long old_sb_flags;
2528         struct ext3_mount_options old_opts;
2529         int err;
2530 #ifdef CONFIG_QUOTA
2531         int i;
2532 #endif
2533
2534         lock_kernel();
2535
2536         /* Store the original options */
2537         lock_super(sb);
2538         old_sb_flags = sb->s_flags;
2539         old_opts.s_mount_opt = sbi->s_mount_opt;
2540         old_opts.s_resuid = sbi->s_resuid;
2541         old_opts.s_resgid = sbi->s_resgid;
2542         old_opts.s_commit_interval = sbi->s_commit_interval;
2543 #ifdef CONFIG_QUOTA
2544         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2545         for (i = 0; i < MAXQUOTAS; i++)
2546                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2547 #endif
2548
2549         /*
2550          * Allow the "check" option to be passed as a remount option.
2551          */
2552         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2553                 err = -EINVAL;
2554                 goto restore_opts;
2555         }
2556
2557         if (test_opt(sb, ABORT))
2558                 ext3_abort(sb, __func__, "Abort forced by user");
2559
2560         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2561                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2562
2563         es = sbi->s_es;
2564
2565         ext3_init_journal_params(sb, sbi->s_journal);
2566
2567         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2568                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2569                 if (test_opt(sb, ABORT)) {
2570                         err = -EROFS;
2571                         goto restore_opts;
2572                 }
2573
2574                 if (*flags & MS_RDONLY) {
2575                         /*
2576                          * First of all, the unconditional stuff we have to do
2577                          * to disable replay of the journal when we next remount
2578                          */
2579                         sb->s_flags |= MS_RDONLY;
2580
2581                         /*
2582                          * OK, test if we are remounting a valid rw partition
2583                          * readonly, and if so set the rdonly flag and then
2584                          * mark the partition as valid again.
2585                          */
2586                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2587                             (sbi->s_mount_state & EXT3_VALID_FS))
2588                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2589
2590                         ext3_mark_recovery_complete(sb, es);
2591                 } else {
2592                         __le32 ret;
2593                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2594                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2595                                 ext3_msg(sb, KERN_WARNING,
2596                                         "warning: couldn't remount RDWR "
2597                                         "because of unsupported optional "
2598                                         "features (%x)", le32_to_cpu(ret));
2599                                 err = -EROFS;
2600                                 goto restore_opts;
2601                         }
2602
2603                         /*
2604                          * If we have an unprocessed orphan list hanging
2605                          * around from a previously readonly bdev mount,
2606                          * require a full umount/remount for now.
2607                          */
2608                         if (es->s_last_orphan) {
2609                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2610                                        "remount RDWR because of unprocessed "
2611                                        "orphan inode list.  Please "
2612                                        "umount/remount instead.");
2613                                 err = -EINVAL;
2614                                 goto restore_opts;
2615                         }
2616
2617                         /*
2618                          * Mounting a RDONLY partition read-write, so reread
2619                          * and store the current valid flag.  (It may have
2620                          * been changed by e2fsck since we originally mounted
2621                          * the partition.)
2622                          */
2623                         ext3_clear_journal_err(sb, es);
2624                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2625                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2626                                 goto restore_opts;
2627                         if (!ext3_setup_super (sb, es, 0))
2628                                 sb->s_flags &= ~MS_RDONLY;
2629                 }
2630         }
2631 #ifdef CONFIG_QUOTA
2632         /* Release old quota file names */
2633         for (i = 0; i < MAXQUOTAS; i++)
2634                 if (old_opts.s_qf_names[i] &&
2635                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2636                         kfree(old_opts.s_qf_names[i]);
2637 #endif
2638         unlock_super(sb);
2639         unlock_kernel();
2640         return 0;
2641 restore_opts:
2642         sb->s_flags = old_sb_flags;
2643         sbi->s_mount_opt = old_opts.s_mount_opt;
2644         sbi->s_resuid = old_opts.s_resuid;
2645         sbi->s_resgid = old_opts.s_resgid;
2646         sbi->s_commit_interval = old_opts.s_commit_interval;
2647 #ifdef CONFIG_QUOTA
2648         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2649         for (i = 0; i < MAXQUOTAS; i++) {
2650                 if (sbi->s_qf_names[i] &&
2651                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2652                         kfree(sbi->s_qf_names[i]);
2653                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2654         }
2655 #endif
2656         unlock_super(sb);
2657         unlock_kernel();
2658         return err;
2659 }
2660
2661 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2662 {
2663         struct super_block *sb = dentry->d_sb;
2664         struct ext3_sb_info *sbi = EXT3_SB(sb);
2665         struct ext3_super_block *es = sbi->s_es;
2666         u64 fsid;
2667
2668         if (test_opt(sb, MINIX_DF)) {
2669                 sbi->s_overhead_last = 0;
2670         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2671                 unsigned long ngroups = sbi->s_groups_count, i;
2672                 ext3_fsblk_t overhead = 0;
2673                 smp_rmb();
2674
2675                 /*
2676                  * Compute the overhead (FS structures).  This is constant
2677                  * for a given filesystem unless the number of block groups
2678                  * changes so we cache the previous value until it does.
2679                  */
2680
2681                 /*
2682                  * All of the blocks before first_data_block are
2683                  * overhead
2684                  */
2685                 overhead = le32_to_cpu(es->s_first_data_block);
2686
2687                 /*
2688                  * Add the overhead attributed to the superblock and
2689                  * block group descriptors.  If the sparse superblocks
2690                  * feature is turned on, then not all groups have this.
2691                  */
2692                 for (i = 0; i < ngroups; i++) {
2693                         overhead += ext3_bg_has_super(sb, i) +
2694                                 ext3_bg_num_gdb(sb, i);
2695                         cond_resched();
2696                 }
2697
2698                 /*
2699                  * Every block group has an inode bitmap, a block
2700                  * bitmap, and an inode table.
2701                  */
2702                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2703                 sbi->s_overhead_last = overhead;
2704                 smp_wmb();
2705                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2706         }
2707
2708         buf->f_type = EXT3_SUPER_MAGIC;
2709         buf->f_bsize = sb->s_blocksize;
2710         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2711         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2712         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2713         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2714                 buf->f_bavail = 0;
2715         buf->f_files = le32_to_cpu(es->s_inodes_count);
2716         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2717         buf->f_namelen = EXT3_NAME_LEN;
2718         fsid = le64_to_cpup((void *)es->s_uuid) ^
2719                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2720         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2721         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2722         return 0;
2723 }
2724
2725 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2726  * is locked for write. Otherwise the are possible deadlocks:
2727  * Process 1                         Process 2
2728  * ext3_create()                     quota_sync()
2729  *   journal_start()                   write_dquot()
2730  *   vfs_dq_init()                       down(dqio_mutex)
2731  *     down(dqio_mutex)                    journal_start()
2732  *
2733  */
2734
2735 #ifdef CONFIG_QUOTA
2736
2737 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2738 {
2739         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2740 }
2741
2742 static int ext3_write_dquot(struct dquot *dquot)
2743 {
2744         int ret, err;
2745         handle_t *handle;
2746         struct inode *inode;
2747
2748         inode = dquot_to_inode(dquot);
2749         handle = ext3_journal_start(inode,
2750                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2751         if (IS_ERR(handle))
2752                 return PTR_ERR(handle);
2753         ret = dquot_commit(dquot);
2754         err = ext3_journal_stop(handle);
2755         if (!ret)
2756                 ret = err;
2757         return ret;
2758 }
2759
2760 static int ext3_acquire_dquot(struct dquot *dquot)
2761 {
2762         int ret, err;
2763         handle_t *handle;
2764
2765         handle = ext3_journal_start(dquot_to_inode(dquot),
2766                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2767         if (IS_ERR(handle))
2768                 return PTR_ERR(handle);
2769         ret = dquot_acquire(dquot);
2770         err = ext3_journal_stop(handle);
2771         if (!ret)
2772                 ret = err;
2773         return ret;
2774 }
2775
2776 static int ext3_release_dquot(struct dquot *dquot)
2777 {
2778         int ret, err;
2779         handle_t *handle;
2780
2781         handle = ext3_journal_start(dquot_to_inode(dquot),
2782                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2783         if (IS_ERR(handle)) {
2784                 /* Release dquot anyway to avoid endless cycle in dqput() */
2785                 dquot_release(dquot);
2786                 return PTR_ERR(handle);
2787         }
2788         ret = dquot_release(dquot);
2789         err = ext3_journal_stop(handle);
2790         if (!ret)
2791                 ret = err;
2792         return ret;
2793 }
2794
2795 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2796 {
2797         /* Are we journaling quotas? */
2798         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2799             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2800                 dquot_mark_dquot_dirty(dquot);
2801                 return ext3_write_dquot(dquot);
2802         } else {
2803                 return dquot_mark_dquot_dirty(dquot);
2804         }
2805 }
2806
2807 static int ext3_write_info(struct super_block *sb, int type)
2808 {
2809         int ret, err;
2810         handle_t *handle;
2811
2812         /* Data block + inode block */
2813         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2814         if (IS_ERR(handle))
2815                 return PTR_ERR(handle);
2816         ret = dquot_commit_info(sb, type);
2817         err = ext3_journal_stop(handle);
2818         if (!ret)
2819                 ret = err;
2820         return ret;
2821 }
2822
2823 /*
2824  * Turn on quotas during mount time - we need to find
2825  * the quota file and such...
2826  */
2827 static int ext3_quota_on_mount(struct super_block *sb, int type)
2828 {
2829         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2830                         EXT3_SB(sb)->s_jquota_fmt, type);
2831 }
2832
2833 /*
2834  * Standard function to be called on quota_on
2835  */
2836 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2837                          char *name, int remount)
2838 {
2839         int err;
2840         struct path path;
2841
2842         if (!test_opt(sb, QUOTA))
2843                 return -EINVAL;
2844         /* When remounting, no checks are needed and in fact, name is NULL */
2845         if (remount)
2846                 return vfs_quota_on(sb, type, format_id, name, remount);
2847
2848         err = kern_path(name, LOOKUP_FOLLOW, &path);
2849         if (err)
2850                 return err;
2851
2852         /* Quotafile not on the same filesystem? */
2853         if (path.mnt->mnt_sb != sb) {
2854                 path_put(&path);
2855                 return -EXDEV;
2856         }
2857         /* Journaling quota? */
2858         if (EXT3_SB(sb)->s_qf_names[type]) {
2859                 /* Quotafile not of fs root? */
2860                 if (path.dentry->d_parent != sb->s_root)
2861                         ext3_msg(sb, KERN_WARNING,
2862                                 "warning: Quota file not on filesystem root. "
2863                                 "Journaled quota will not work.");
2864         }
2865
2866         /*
2867          * When we journal data on quota file, we have to flush journal to see
2868          * all updates to the file when we bypass pagecache...
2869          */
2870         if (ext3_should_journal_data(path.dentry->d_inode)) {
2871                 /*
2872                  * We don't need to lock updates but journal_flush() could
2873                  * otherwise be livelocked...
2874                  */
2875                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2876                 err = journal_flush(EXT3_SB(sb)->s_journal);
2877                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2878                 if (err) {
2879                         path_put(&path);
2880                         return err;
2881                 }
2882         }
2883
2884         err = vfs_quota_on_path(sb, type, format_id, &path);
2885         path_put(&path);
2886         return err;
2887 }
2888
2889 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2890  * acquiring the locks... As quota files are never truncated and quota code
2891  * itself serializes the operations (and noone else should touch the files)
2892  * we don't have to be afraid of races */
2893 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2894                                size_t len, loff_t off)
2895 {
2896         struct inode *inode = sb_dqopt(sb)->files[type];
2897         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2898         int err = 0;
2899         int offset = off & (sb->s_blocksize - 1);
2900         int tocopy;
2901         size_t toread;
2902         struct buffer_head *bh;
2903         loff_t i_size = i_size_read(inode);
2904
2905         if (off > i_size)
2906                 return 0;
2907         if (off+len > i_size)
2908                 len = i_size-off;
2909         toread = len;
2910         while (toread > 0) {
2911                 tocopy = sb->s_blocksize - offset < toread ?
2912                                 sb->s_blocksize - offset : toread;
2913                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2914                 if (err)
2915                         return err;
2916                 if (!bh)        /* A hole? */
2917                         memset(data, 0, tocopy);
2918                 else
2919                         memcpy(data, bh->b_data+offset, tocopy);
2920                 brelse(bh);
2921                 offset = 0;
2922                 toread -= tocopy;
2923                 data += tocopy;
2924                 blk++;
2925         }
2926         return len;
2927 }
2928
2929 /* Write to quotafile (we know the transaction is already started and has
2930  * enough credits) */
2931 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2932                                 const char *data, size_t len, loff_t off)
2933 {
2934         struct inode *inode = sb_dqopt(sb)->files[type];
2935         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2936         int err = 0;
2937         int offset = off & (sb->s_blocksize - 1);
2938         int tocopy;
2939         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2940         size_t towrite = len;
2941         struct buffer_head *bh;
2942         handle_t *handle = journal_current_handle();
2943
2944         if (!handle) {
2945                 ext3_msg(sb, KERN_WARNING,
2946                         "warning: quota write (off=%llu, len=%llu)"
2947                         " cancelled because transaction is not started.",
2948                         (unsigned long long)off, (unsigned long long)len);
2949                 return -EIO;
2950         }
2951         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2952         while (towrite > 0) {
2953                 tocopy = sb->s_blocksize - offset < towrite ?
2954                                 sb->s_blocksize - offset : towrite;
2955                 bh = ext3_bread(handle, inode, blk, 1, &err);
2956                 if (!bh)
2957                         goto out;
2958                 if (journal_quota) {
2959                         err = ext3_journal_get_write_access(handle, bh);
2960                         if (err) {
2961                                 brelse(bh);
2962                                 goto out;
2963                         }
2964                 }
2965                 lock_buffer(bh);
2966                 memcpy(bh->b_data+offset, data, tocopy);
2967                 flush_dcache_page(bh->b_page);
2968                 unlock_buffer(bh);
2969                 if (journal_quota)
2970                         err = ext3_journal_dirty_metadata(handle, bh);
2971                 else {
2972                         /* Always do at least ordered writes for quotas */
2973                         err = ext3_journal_dirty_data(handle, bh);
2974                         mark_buffer_dirty(bh);
2975                 }
2976                 brelse(bh);
2977                 if (err)
2978                         goto out;
2979                 offset = 0;
2980                 towrite -= tocopy;
2981                 data += tocopy;
2982                 blk++;
2983         }
2984 out:
2985         if (len == towrite) {
2986                 mutex_unlock(&inode->i_mutex);
2987                 return err;
2988         }
2989         if (inode->i_size < off+len-towrite) {
2990                 i_size_write(inode, off+len-towrite);
2991                 EXT3_I(inode)->i_disksize = inode->i_size;
2992         }
2993         inode->i_version++;
2994         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2995         ext3_mark_inode_dirty(handle, inode);
2996         mutex_unlock(&inode->i_mutex);
2997         return len - towrite;
2998 }
2999
3000 #endif
3001
3002 static int ext3_get_sb(struct file_system_type *fs_type,
3003         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3004 {
3005         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
3006 }
3007
3008 static struct file_system_type ext3_fs_type = {
3009         .owner          = THIS_MODULE,
3010         .name           = "ext3",
3011         .get_sb         = ext3_get_sb,
3012         .kill_sb        = kill_block_super,
3013         .fs_flags       = FS_REQUIRES_DEV,
3014 };
3015
3016 static int __init init_ext3_fs(void)
3017 {
3018         int err = init_ext3_xattr();
3019         if (err)
3020                 return err;
3021         err = init_inodecache();
3022         if (err)
3023                 goto out1;
3024         err = register_filesystem(&ext3_fs_type);
3025         if (err)
3026                 goto out;
3027         return 0;
3028 out:
3029         destroy_inodecache();
3030 out1:
3031         exit_ext3_xattr();
3032         return err;
3033 }
3034
3035 static void __exit exit_ext3_fs(void)
3036 {
3037         unregister_filesystem(&ext3_fs_type);
3038         destroy_inodecache();
3039         exit_ext3_xattr();
3040 }
3041
3042 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3043 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3044 MODULE_LICENSE("GPL");
3045 module_init(init_ext3_fs)
3046 module_exit(exit_ext3_fs)