ext3: Remove outdated comment about lock_super()
[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                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_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         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_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 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
566                 seq_puts(seq, ",usrquota");
567
568         if (sbi->s_mount_opt & EXT3_MOUNT_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(sbi->s_mount_opt &
660                                                      EXT3_MOUNT_DATA_FLAGS));
661         if (test_opt(sb, DATA_ERR_ABORT))
662                 seq_puts(seq, ",data_err=abort");
663
664         if (test_opt(sb, NOLOAD))
665                 seq_puts(seq, ",norecovery");
666
667         ext3_show_quota_options(seq, sb);
668
669         return 0;
670 }
671
672
673 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
674                 u64 ino, u32 generation)
675 {
676         struct inode *inode;
677
678         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
679                 return ERR_PTR(-ESTALE);
680         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
681                 return ERR_PTR(-ESTALE);
682
683         /* iget isn't really right if the inode is currently unallocated!!
684          *
685          * ext3_read_inode will return a bad_inode if the inode had been
686          * deleted, so we should be safe.
687          *
688          * Currently we don't know the generation for parent directory, so
689          * a generation of 0 means "accept any"
690          */
691         inode = ext3_iget(sb, ino);
692         if (IS_ERR(inode))
693                 return ERR_CAST(inode);
694         if (generation && inode->i_generation != generation) {
695                 iput(inode);
696                 return ERR_PTR(-ESTALE);
697         }
698
699         return inode;
700 }
701
702 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
703                 int fh_len, int fh_type)
704 {
705         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
706                                     ext3_nfs_get_inode);
707 }
708
709 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
710                 int fh_len, int fh_type)
711 {
712         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
713                                     ext3_nfs_get_inode);
714 }
715
716 /*
717  * Try to release metadata pages (indirect blocks, directories) which are
718  * mapped via the block device.  Since these pages could have journal heads
719  * which would prevent try_to_free_buffers() from freeing them, we must use
720  * jbd layer's try_to_free_buffers() function to release them.
721  */
722 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
723                                  gfp_t wait)
724 {
725         journal_t *journal = EXT3_SB(sb)->s_journal;
726
727         WARN_ON(PageChecked(page));
728         if (!page_has_buffers(page))
729                 return 0;
730         if (journal)
731                 return journal_try_to_free_buffers(journal, page, 
732                                                    wait & ~__GFP_WAIT);
733         return try_to_free_buffers(page);
734 }
735
736 #ifdef CONFIG_QUOTA
737 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
738 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
739
740 static int ext3_write_dquot(struct dquot *dquot);
741 static int ext3_acquire_dquot(struct dquot *dquot);
742 static int ext3_release_dquot(struct dquot *dquot);
743 static int ext3_mark_dquot_dirty(struct dquot *dquot);
744 static int ext3_write_info(struct super_block *sb, int type);
745 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
746                                 char *path, int remount);
747 static int ext3_quota_on_mount(struct super_block *sb, int type);
748 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
749                                size_t len, loff_t off);
750 static ssize_t ext3_quota_write(struct super_block *sb, int type,
751                                 const char *data, size_t len, loff_t off);
752
753 static const struct dquot_operations ext3_quota_operations = {
754         .initialize     = dquot_initialize,
755         .drop           = dquot_drop,
756         .alloc_space    = dquot_alloc_space,
757         .alloc_inode    = dquot_alloc_inode,
758         .free_space     = dquot_free_space,
759         .free_inode     = dquot_free_inode,
760         .transfer       = dquot_transfer,
761         .write_dquot    = ext3_write_dquot,
762         .acquire_dquot  = ext3_acquire_dquot,
763         .release_dquot  = ext3_release_dquot,
764         .mark_dirty     = ext3_mark_dquot_dirty,
765         .write_info     = ext3_write_info,
766         .alloc_dquot    = dquot_alloc,
767         .destroy_dquot  = dquot_destroy,
768 };
769
770 static const struct quotactl_ops ext3_qctl_operations = {
771         .quota_on       = ext3_quota_on,
772         .quota_off      = vfs_quota_off,
773         .quota_sync     = vfs_quota_sync,
774         .get_info       = vfs_get_dqinfo,
775         .set_info       = vfs_set_dqinfo,
776         .get_dqblk      = vfs_get_dqblk,
777         .set_dqblk      = vfs_set_dqblk
778 };
779 #endif
780
781 static const struct super_operations ext3_sops = {
782         .alloc_inode    = ext3_alloc_inode,
783         .destroy_inode  = ext3_destroy_inode,
784         .write_inode    = ext3_write_inode,
785         .dirty_inode    = ext3_dirty_inode,
786         .delete_inode   = ext3_delete_inode,
787         .put_super      = ext3_put_super,
788         .sync_fs        = ext3_sync_fs,
789         .freeze_fs      = ext3_freeze,
790         .unfreeze_fs    = ext3_unfreeze,
791         .statfs         = ext3_statfs,
792         .remount_fs     = ext3_remount,
793         .clear_inode    = ext3_clear_inode,
794         .show_options   = ext3_show_options,
795 #ifdef CONFIG_QUOTA
796         .quota_read     = ext3_quota_read,
797         .quota_write    = ext3_quota_write,
798 #endif
799         .bdev_try_to_free_page = bdev_try_to_free_page,
800 };
801
802 static const struct export_operations ext3_export_ops = {
803         .fh_to_dentry = ext3_fh_to_dentry,
804         .fh_to_parent = ext3_fh_to_parent,
805         .get_parent = ext3_get_parent,
806 };
807
808 enum {
809         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
810         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
811         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
812         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
813         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
814         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
815         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
816         Opt_data_err_abort, Opt_data_err_ignore,
817         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
818         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
819         Opt_noquota, Opt_ignore, Opt_barrier, Opt_err, Opt_resize,
820         Opt_usrquota, Opt_grpquota
821 };
822
823 static const match_table_t tokens = {
824         {Opt_bsd_df, "bsddf"},
825         {Opt_minix_df, "minixdf"},
826         {Opt_grpid, "grpid"},
827         {Opt_grpid, "bsdgroups"},
828         {Opt_nogrpid, "nogrpid"},
829         {Opt_nogrpid, "sysvgroups"},
830         {Opt_resgid, "resgid=%u"},
831         {Opt_resuid, "resuid=%u"},
832         {Opt_sb, "sb=%u"},
833         {Opt_err_cont, "errors=continue"},
834         {Opt_err_panic, "errors=panic"},
835         {Opt_err_ro, "errors=remount-ro"},
836         {Opt_nouid32, "nouid32"},
837         {Opt_nocheck, "nocheck"},
838         {Opt_nocheck, "check=none"},
839         {Opt_debug, "debug"},
840         {Opt_oldalloc, "oldalloc"},
841         {Opt_orlov, "orlov"},
842         {Opt_user_xattr, "user_xattr"},
843         {Opt_nouser_xattr, "nouser_xattr"},
844         {Opt_acl, "acl"},
845         {Opt_noacl, "noacl"},
846         {Opt_reservation, "reservation"},
847         {Opt_noreservation, "noreservation"},
848         {Opt_noload, "noload"},
849         {Opt_noload, "norecovery"},
850         {Opt_nobh, "nobh"},
851         {Opt_bh, "bh"},
852         {Opt_commit, "commit=%u"},
853         {Opt_journal_update, "journal=update"},
854         {Opt_journal_inum, "journal=%u"},
855         {Opt_journal_dev, "journal_dev=%u"},
856         {Opt_abort, "abort"},
857         {Opt_data_journal, "data=journal"},
858         {Opt_data_ordered, "data=ordered"},
859         {Opt_data_writeback, "data=writeback"},
860         {Opt_data_err_abort, "data_err=abort"},
861         {Opt_data_err_ignore, "data_err=ignore"},
862         {Opt_offusrjquota, "usrjquota="},
863         {Opt_usrjquota, "usrjquota=%s"},
864         {Opt_offgrpjquota, "grpjquota="},
865         {Opt_grpjquota, "grpjquota=%s"},
866         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
867         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
868         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
869         {Opt_grpquota, "grpquota"},
870         {Opt_noquota, "noquota"},
871         {Opt_quota, "quota"},
872         {Opt_usrquota, "usrquota"},
873         {Opt_barrier, "barrier=%u"},
874         {Opt_resize, "resize"},
875         {Opt_err, NULL},
876 };
877
878 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
879 {
880         ext3_fsblk_t    sb_block;
881         char            *options = (char *) *data;
882
883         if (!options || strncmp(options, "sb=", 3) != 0)
884                 return 1;       /* Default location */
885         options += 3;
886         /*todo: use simple_strtoll with >32bit ext3 */
887         sb_block = simple_strtoul(options, &options, 0);
888         if (*options && *options != ',') {
889                 ext3_msg(sb, "error: invalid sb specification: %s",
890                        (char *) *data);
891                 return 1;
892         }
893         if (*options == ',')
894                 options++;
895         *data = (void *) options;
896         return sb_block;
897 }
898
899 static int parse_options (char *options, struct super_block *sb,
900                           unsigned int *inum, unsigned long *journal_devnum,
901                           ext3_fsblk_t *n_blocks_count, int is_remount)
902 {
903         struct ext3_sb_info *sbi = EXT3_SB(sb);
904         char * p;
905         substring_t args[MAX_OPT_ARGS];
906         int data_opt = 0;
907         int option;
908 #ifdef CONFIG_QUOTA
909         int qtype, qfmt;
910         char *qname;
911 #endif
912
913         if (!options)
914                 return 1;
915
916         while ((p = strsep (&options, ",")) != NULL) {
917                 int token;
918                 if (!*p)
919                         continue;
920
921                 token = match_token(p, tokens, args);
922                 switch (token) {
923                 case Opt_bsd_df:
924                         clear_opt (sbi->s_mount_opt, MINIX_DF);
925                         break;
926                 case Opt_minix_df:
927                         set_opt (sbi->s_mount_opt, MINIX_DF);
928                         break;
929                 case Opt_grpid:
930                         set_opt (sbi->s_mount_opt, GRPID);
931                         break;
932                 case Opt_nogrpid:
933                         clear_opt (sbi->s_mount_opt, GRPID);
934                         break;
935                 case Opt_resuid:
936                         if (match_int(&args[0], &option))
937                                 return 0;
938                         sbi->s_resuid = option;
939                         break;
940                 case Opt_resgid:
941                         if (match_int(&args[0], &option))
942                                 return 0;
943                         sbi->s_resgid = option;
944                         break;
945                 case Opt_sb:
946                         /* handled by get_sb_block() instead of here */
947                         /* *sb_block = match_int(&args[0]); */
948                         break;
949                 case Opt_err_panic:
950                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
951                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
952                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
953                         break;
954                 case Opt_err_ro:
955                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
956                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
957                         set_opt (sbi->s_mount_opt, ERRORS_RO);
958                         break;
959                 case Opt_err_cont:
960                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
961                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
962                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
963                         break;
964                 case Opt_nouid32:
965                         set_opt (sbi->s_mount_opt, NO_UID32);
966                         break;
967                 case Opt_nocheck:
968                         clear_opt (sbi->s_mount_opt, CHECK);
969                         break;
970                 case Opt_debug:
971                         set_opt (sbi->s_mount_opt, DEBUG);
972                         break;
973                 case Opt_oldalloc:
974                         set_opt (sbi->s_mount_opt, OLDALLOC);
975                         break;
976                 case Opt_orlov:
977                         clear_opt (sbi->s_mount_opt, OLDALLOC);
978                         break;
979 #ifdef CONFIG_EXT3_FS_XATTR
980                 case Opt_user_xattr:
981                         set_opt (sbi->s_mount_opt, XATTR_USER);
982                         break;
983                 case Opt_nouser_xattr:
984                         clear_opt (sbi->s_mount_opt, XATTR_USER);
985                         break;
986 #else
987                 case Opt_user_xattr:
988                 case Opt_nouser_xattr:
989                         ext3_msg(sb, KERN_INFO,
990                                 "(no)user_xattr options not supported");
991                         break;
992 #endif
993 #ifdef CONFIG_EXT3_FS_POSIX_ACL
994                 case Opt_acl:
995                         set_opt(sbi->s_mount_opt, POSIX_ACL);
996                         break;
997                 case Opt_noacl:
998                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
999                         break;
1000 #else
1001                 case Opt_acl:
1002                 case Opt_noacl:
1003                         ext3_msg(sb, KERN_INFO,
1004                                 "(no)acl options not supported");
1005                         break;
1006 #endif
1007                 case Opt_reservation:
1008                         set_opt(sbi->s_mount_opt, RESERVATION);
1009                         break;
1010                 case Opt_noreservation:
1011                         clear_opt(sbi->s_mount_opt, RESERVATION);
1012                         break;
1013                 case Opt_journal_update:
1014                         /* @@@ FIXME */
1015                         /* Eventually we will want to be able to create
1016                            a journal file here.  For now, only allow the
1017                            user to specify an existing inode to be the
1018                            journal file. */
1019                         if (is_remount) {
1020                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1021                                         "journal on remount");
1022                                 return 0;
1023                         }
1024                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1025                         break;
1026                 case Opt_journal_inum:
1027                         if (is_remount) {
1028                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1029                                        "journal on remount");
1030                                 return 0;
1031                         }
1032                         if (match_int(&args[0], &option))
1033                                 return 0;
1034                         *inum = option;
1035                         break;
1036                 case Opt_journal_dev:
1037                         if (is_remount) {
1038                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1039                                        "journal on remount");
1040                                 return 0;
1041                         }
1042                         if (match_int(&args[0], &option))
1043                                 return 0;
1044                         *journal_devnum = option;
1045                         break;
1046                 case Opt_noload:
1047                         set_opt (sbi->s_mount_opt, NOLOAD);
1048                         break;
1049                 case Opt_commit:
1050                         if (match_int(&args[0], &option))
1051                                 return 0;
1052                         if (option < 0)
1053                                 return 0;
1054                         if (option == 0)
1055                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1056                         sbi->s_commit_interval = HZ * option;
1057                         break;
1058                 case Opt_data_journal:
1059                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1060                         goto datacheck;
1061                 case Opt_data_ordered:
1062                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1063                         goto datacheck;
1064                 case Opt_data_writeback:
1065                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1066                 datacheck:
1067                         if (is_remount) {
1068                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
1069                                                 == data_opt)
1070                                         break;
1071                                 ext3_msg(sb, KERN_ERR,
1072                                         "error: cannot change "
1073                                         "data mode on remount. The filesystem "
1074                                         "is mounted in data=%s mode and you "
1075                                         "try to remount it in data=%s mode.",
1076                                         data_mode_string(sbi->s_mount_opt &
1077                                                         EXT3_MOUNT_DATA_FLAGS),
1078                                         data_mode_string(data_opt));
1079                                 return 0;
1080                         } else {
1081                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
1082                                 sbi->s_mount_opt |= data_opt;
1083                         }
1084                         break;
1085                 case Opt_data_err_abort:
1086                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1087                         break;
1088                 case Opt_data_err_ignore:
1089                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1090                         break;
1091 #ifdef CONFIG_QUOTA
1092                 case Opt_usrjquota:
1093                         qtype = USRQUOTA;
1094                         goto set_qf_name;
1095                 case Opt_grpjquota:
1096                         qtype = GRPQUOTA;
1097 set_qf_name:
1098                         if (sb_any_quota_loaded(sb) &&
1099                             !sbi->s_qf_names[qtype]) {
1100                                 ext3_msg(sb, KERN_ERR,
1101                                         "error: cannot change journaled "
1102                                         "quota options when quota turned on.");
1103                                 return 0;
1104                         }
1105                         qname = match_strdup(&args[0]);
1106                         if (!qname) {
1107                                 ext3_msg(sb, KERN_ERR,
1108                                         "error: not enough memory for "
1109                                         "storing quotafile name.");
1110                                 return 0;
1111                         }
1112                         if (sbi->s_qf_names[qtype] &&
1113                             strcmp(sbi->s_qf_names[qtype], qname)) {
1114                                 ext3_msg(sb, KERN_ERR,
1115                                         "error: %s quota file already "
1116                                         "specified.", QTYPE2NAME(qtype));
1117                                 kfree(qname);
1118                                 return 0;
1119                         }
1120                         sbi->s_qf_names[qtype] = qname;
1121                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1122                                 ext3_msg(sb, KERN_ERR,
1123                                         "error: quotafile must be on "
1124                                         "filesystem root.");
1125                                 kfree(sbi->s_qf_names[qtype]);
1126                                 sbi->s_qf_names[qtype] = NULL;
1127                                 return 0;
1128                         }
1129                         set_opt(sbi->s_mount_opt, QUOTA);
1130                         break;
1131                 case Opt_offusrjquota:
1132                         qtype = USRQUOTA;
1133                         goto clear_qf_name;
1134                 case Opt_offgrpjquota:
1135                         qtype = GRPQUOTA;
1136 clear_qf_name:
1137                         if (sb_any_quota_loaded(sb) &&
1138                             sbi->s_qf_names[qtype]) {
1139                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1140                                         "journaled quota options when "
1141                                         "quota turned on.");
1142                                 return 0;
1143                         }
1144                         /*
1145                          * The space will be released later when all options
1146                          * are confirmed to be correct
1147                          */
1148                         sbi->s_qf_names[qtype] = NULL;
1149                         break;
1150                 case Opt_jqfmt_vfsold:
1151                         qfmt = QFMT_VFS_OLD;
1152                         goto set_qf_format;
1153                 case Opt_jqfmt_vfsv0:
1154                         qfmt = QFMT_VFS_V0;
1155                         goto set_qf_format;
1156                 case Opt_jqfmt_vfsv1:
1157                         qfmt = QFMT_VFS_V1;
1158 set_qf_format:
1159                         if (sb_any_quota_loaded(sb) &&
1160                             sbi->s_jquota_fmt != qfmt) {
1161                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1162                                         "journaled quota options when "
1163                                         "quota turned on.");
1164                                 return 0;
1165                         }
1166                         sbi->s_jquota_fmt = qfmt;
1167                         break;
1168                 case Opt_quota:
1169                 case Opt_usrquota:
1170                         set_opt(sbi->s_mount_opt, QUOTA);
1171                         set_opt(sbi->s_mount_opt, USRQUOTA);
1172                         break;
1173                 case Opt_grpquota:
1174                         set_opt(sbi->s_mount_opt, QUOTA);
1175                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1176                         break;
1177                 case Opt_noquota:
1178                         if (sb_any_quota_loaded(sb)) {
1179                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1180                                         "quota options when quota turned on.");
1181                                 return 0;
1182                         }
1183                         clear_opt(sbi->s_mount_opt, QUOTA);
1184                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1185                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1186                         break;
1187 #else
1188                 case Opt_quota:
1189                 case Opt_usrquota:
1190                 case Opt_grpquota:
1191                         ext3_msg(sb, KERN_ERR,
1192                                 "error: quota options not supported.");
1193                         break;
1194                 case Opt_usrjquota:
1195                 case Opt_grpjquota:
1196                 case Opt_offusrjquota:
1197                 case Opt_offgrpjquota:
1198                 case Opt_jqfmt_vfsold:
1199                 case Opt_jqfmt_vfsv0:
1200                 case Opt_jqfmt_vfsv1:
1201                         ext3_msg(sb, KERN_ERR,
1202                                 "error: journaled quota options not "
1203                                 "supported.");
1204                         break;
1205                 case Opt_noquota:
1206                         break;
1207 #endif
1208                 case Opt_abort:
1209                         set_opt(sbi->s_mount_opt, ABORT);
1210                         break;
1211                 case Opt_barrier:
1212                         if (match_int(&args[0], &option))
1213                                 return 0;
1214                         if (option)
1215                                 set_opt(sbi->s_mount_opt, BARRIER);
1216                         else
1217                                 clear_opt(sbi->s_mount_opt, BARRIER);
1218                         break;
1219                 case Opt_ignore:
1220                         break;
1221                 case Opt_resize:
1222                         if (!is_remount) {
1223                                 ext3_msg(sb, KERN_ERR,
1224                                         "error: resize option only available "
1225                                         "for remount");
1226                                 return 0;
1227                         }
1228                         if (match_int(&args[0], &option) != 0)
1229                                 return 0;
1230                         *n_blocks_count = option;
1231                         break;
1232                 case Opt_nobh:
1233                         set_opt(sbi->s_mount_opt, NOBH);
1234                         break;
1235                 case Opt_bh:
1236                         clear_opt(sbi->s_mount_opt, NOBH);
1237                         break;
1238                 default:
1239                         ext3_msg(sb, KERN_ERR,
1240                                 "error: unrecognized mount option \"%s\" "
1241                                 "or missing value", p);
1242                         return 0;
1243                 }
1244         }
1245 #ifdef CONFIG_QUOTA
1246         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1247                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1248                      sbi->s_qf_names[USRQUOTA])
1249                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1250
1251                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1252                      sbi->s_qf_names[GRPQUOTA])
1253                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1254
1255                 if ((sbi->s_qf_names[USRQUOTA] &&
1256                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1257                     (sbi->s_qf_names[GRPQUOTA] &&
1258                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1259                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1260                                         "format mixing.");
1261                         return 0;
1262                 }
1263
1264                 if (!sbi->s_jquota_fmt) {
1265                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1266                                         "not specified.");
1267                         return 0;
1268                 }
1269         } else {
1270                 if (sbi->s_jquota_fmt) {
1271                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1272                                         "specified with no journaling "
1273                                         "enabled.");
1274                         return 0;
1275                 }
1276         }
1277 #endif
1278         return 1;
1279 }
1280
1281 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1282                             int read_only)
1283 {
1284         struct ext3_sb_info *sbi = EXT3_SB(sb);
1285         int res = 0;
1286
1287         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1288                 ext3_msg(sb, KERN_ERR,
1289                         "error: revision level too high, "
1290                         "forcing read-only mode");
1291                 res = MS_RDONLY;
1292         }
1293         if (read_only)
1294                 return res;
1295         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1296                 ext3_msg(sb, KERN_WARNING,
1297                         "warning: mounting unchecked fs, "
1298                         "running e2fsck is recommended");
1299         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1300                 ext3_msg(sb, KERN_WARNING,
1301                         "warning: mounting fs with errors, "
1302                         "running e2fsck is recommended");
1303         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1304                  le16_to_cpu(es->s_mnt_count) >=
1305                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1306                 ext3_msg(sb, KERN_WARNING,
1307                         "warning: maximal mount count reached, "
1308                         "running e2fsck is recommended");
1309         else if (le32_to_cpu(es->s_checkinterval) &&
1310                 (le32_to_cpu(es->s_lastcheck) +
1311                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1312                 ext3_msg(sb, KERN_WARNING,
1313                         "warning: checktime reached, "
1314                         "running e2fsck is recommended");
1315 #if 0
1316                 /* @@@ We _will_ want to clear the valid bit if we find
1317                    inconsistencies, to force a fsck at reboot.  But for
1318                    a plain journaled filesystem we can keep it set as
1319                    valid forever! :) */
1320         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1321 #endif
1322         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1323                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1324         le16_add_cpu(&es->s_mnt_count, 1);
1325         es->s_mtime = cpu_to_le32(get_seconds());
1326         ext3_update_dynamic_rev(sb);
1327         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1328
1329         ext3_commit_super(sb, es, 1);
1330         if (test_opt(sb, DEBUG))
1331                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1332                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1333                         sb->s_blocksize,
1334                         sbi->s_groups_count,
1335                         EXT3_BLOCKS_PER_GROUP(sb),
1336                         EXT3_INODES_PER_GROUP(sb),
1337                         sbi->s_mount_opt);
1338
1339         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1340                 char b[BDEVNAME_SIZE];
1341                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1342                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1343         } else {
1344                 ext3_msg(sb, KERN_INFO, "using internal journal");
1345         }
1346         return res;
1347 }
1348
1349 /* Called at mount-time, super-block is locked */
1350 static int ext3_check_descriptors(struct super_block *sb)
1351 {
1352         struct ext3_sb_info *sbi = EXT3_SB(sb);
1353         int i;
1354
1355         ext3_debug ("Checking group descriptors");
1356
1357         for (i = 0; i < sbi->s_groups_count; i++) {
1358                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1359                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1360                 ext3_fsblk_t last_block;
1361
1362                 if (i == sbi->s_groups_count - 1)
1363                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1364                 else
1365                         last_block = first_block +
1366                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1367
1368                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1369                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1370                 {
1371                         ext3_error (sb, "ext3_check_descriptors",
1372                                     "Block bitmap for group %d"
1373                                     " not in group (block %lu)!",
1374                                     i, (unsigned long)
1375                                         le32_to_cpu(gdp->bg_block_bitmap));
1376                         return 0;
1377                 }
1378                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1379                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1380                 {
1381                         ext3_error (sb, "ext3_check_descriptors",
1382                                     "Inode bitmap for group %d"
1383                                     " not in group (block %lu)!",
1384                                     i, (unsigned long)
1385                                         le32_to_cpu(gdp->bg_inode_bitmap));
1386                         return 0;
1387                 }
1388                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1389                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1390                     last_block)
1391                 {
1392                         ext3_error (sb, "ext3_check_descriptors",
1393                                     "Inode table for group %d"
1394                                     " not in group (block %lu)!",
1395                                     i, (unsigned long)
1396                                         le32_to_cpu(gdp->bg_inode_table));
1397                         return 0;
1398                 }
1399         }
1400
1401         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1402         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1403         return 1;
1404 }
1405
1406
1407 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1408  * the superblock) which were deleted from all directories, but held open by
1409  * a process at the time of a crash.  We walk the list and try to delete these
1410  * inodes at recovery time (only with a read-write filesystem).
1411  *
1412  * In order to keep the orphan inode chain consistent during traversal (in
1413  * case of crash during recovery), we link each inode into the superblock
1414  * orphan list_head and handle it the same way as an inode deletion during
1415  * normal operation (which journals the operations for us).
1416  *
1417  * We only do an iget() and an iput() on each inode, which is very safe if we
1418  * accidentally point at an in-use or already deleted inode.  The worst that
1419  * can happen in this case is that we get a "bit already cleared" message from
1420  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1421  * e2fsck was run on this filesystem, and it must have already done the orphan
1422  * inode cleanup for us, so we can safely abort without any further action.
1423  */
1424 static void ext3_orphan_cleanup (struct super_block * sb,
1425                                  struct ext3_super_block * es)
1426 {
1427         unsigned int s_flags = sb->s_flags;
1428         int nr_orphans = 0, nr_truncates = 0;
1429 #ifdef CONFIG_QUOTA
1430         int i;
1431 #endif
1432         if (!es->s_last_orphan) {
1433                 jbd_debug(4, "no orphan inodes to clean up\n");
1434                 return;
1435         }
1436
1437         if (bdev_read_only(sb->s_bdev)) {
1438                 ext3_msg(sb, KERN_ERR, "error: write access "
1439                         "unavailable, skipping orphan cleanup.");
1440                 return;
1441         }
1442
1443         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1444                 if (es->s_last_orphan)
1445                         jbd_debug(1, "Errors on filesystem, "
1446                                   "clearing orphan list.\n");
1447                 es->s_last_orphan = 0;
1448                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1449                 return;
1450         }
1451
1452         if (s_flags & MS_RDONLY) {
1453                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1454                 sb->s_flags &= ~MS_RDONLY;
1455         }
1456 #ifdef CONFIG_QUOTA
1457         /* Needed for iput() to work correctly and not trash data */
1458         sb->s_flags |= MS_ACTIVE;
1459         /* Turn on quotas so that they are updated correctly */
1460         for (i = 0; i < MAXQUOTAS; i++) {
1461                 if (EXT3_SB(sb)->s_qf_names[i]) {
1462                         int ret = ext3_quota_on_mount(sb, i);
1463                         if (ret < 0)
1464                                 ext3_msg(sb, KERN_ERR,
1465                                         "error: cannot turn on journaled "
1466                                         "quota: %d", ret);
1467                 }
1468         }
1469 #endif
1470
1471         while (es->s_last_orphan) {
1472                 struct inode *inode;
1473
1474                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1475                 if (IS_ERR(inode)) {
1476                         es->s_last_orphan = 0;
1477                         break;
1478                 }
1479
1480                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1481                 vfs_dq_init(inode);
1482                 if (inode->i_nlink) {
1483                         printk(KERN_DEBUG
1484                                 "%s: truncating inode %lu to %Ld bytes\n",
1485                                 __func__, inode->i_ino, inode->i_size);
1486                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1487                                   inode->i_ino, inode->i_size);
1488                         ext3_truncate(inode);
1489                         nr_truncates++;
1490                 } else {
1491                         printk(KERN_DEBUG
1492                                 "%s: deleting unreferenced inode %lu\n",
1493                                 __func__, inode->i_ino);
1494                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1495                                   inode->i_ino);
1496                         nr_orphans++;
1497                 }
1498                 iput(inode);  /* The delete magic happens here! */
1499         }
1500
1501 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1502
1503         if (nr_orphans)
1504                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1505                        PLURAL(nr_orphans));
1506         if (nr_truncates)
1507                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1508                        PLURAL(nr_truncates));
1509 #ifdef CONFIG_QUOTA
1510         /* Turn quotas off */
1511         for (i = 0; i < MAXQUOTAS; i++) {
1512                 if (sb_dqopt(sb)->files[i])
1513                         vfs_quota_off(sb, i, 0);
1514         }
1515 #endif
1516         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1517 }
1518
1519 /*
1520  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1521  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1522  * We need to be 1 filesystem block less than the 2^32 sector limit.
1523  */
1524 static loff_t ext3_max_size(int bits)
1525 {
1526         loff_t res = EXT3_NDIR_BLOCKS;
1527         int meta_blocks;
1528         loff_t upper_limit;
1529
1530         /* This is calculated to be the largest file size for a
1531          * dense, file such that the total number of
1532          * sectors in the file, including data and all indirect blocks,
1533          * does not exceed 2^32 -1
1534          * __u32 i_blocks representing the total number of
1535          * 512 bytes blocks of the file
1536          */
1537         upper_limit = (1LL << 32) - 1;
1538
1539         /* total blocks in file system block size */
1540         upper_limit >>= (bits - 9);
1541
1542
1543         /* indirect blocks */
1544         meta_blocks = 1;
1545         /* double indirect blocks */
1546         meta_blocks += 1 + (1LL << (bits-2));
1547         /* tripple indirect blocks */
1548         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1549
1550         upper_limit -= meta_blocks;
1551         upper_limit <<= bits;
1552
1553         res += 1LL << (bits-2);
1554         res += 1LL << (2*(bits-2));
1555         res += 1LL << (3*(bits-2));
1556         res <<= bits;
1557         if (res > upper_limit)
1558                 res = upper_limit;
1559
1560         if (res > MAX_LFS_FILESIZE)
1561                 res = MAX_LFS_FILESIZE;
1562
1563         return res;
1564 }
1565
1566 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1567                                     ext3_fsblk_t logic_sb_block,
1568                                     int nr)
1569 {
1570         struct ext3_sb_info *sbi = EXT3_SB(sb);
1571         unsigned long bg, first_meta_bg;
1572         int has_super = 0;
1573
1574         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1575
1576         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1577             nr < first_meta_bg)
1578                 return (logic_sb_block + nr + 1);
1579         bg = sbi->s_desc_per_block * nr;
1580         if (ext3_bg_has_super(sb, bg))
1581                 has_super = 1;
1582         return (has_super + ext3_group_first_block_no(sb, bg));
1583 }
1584
1585
1586 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1587 {
1588         struct buffer_head * bh;
1589         struct ext3_super_block *es = NULL;
1590         struct ext3_sb_info *sbi;
1591         ext3_fsblk_t block;
1592         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1593         ext3_fsblk_t logic_sb_block;
1594         unsigned long offset = 0;
1595         unsigned int journal_inum = 0;
1596         unsigned long journal_devnum = 0;
1597         unsigned long def_mount_opts;
1598         struct inode *root;
1599         int blocksize;
1600         int hblock;
1601         int db_count;
1602         int i;
1603         int needs_recovery;
1604         int ret = -EINVAL;
1605         __le32 features;
1606         int err;
1607
1608         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1609         if (!sbi)
1610                 return -ENOMEM;
1611
1612         sbi->s_blockgroup_lock =
1613                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1614         if (!sbi->s_blockgroup_lock) {
1615                 kfree(sbi);
1616                 return -ENOMEM;
1617         }
1618         sb->s_fs_info = sbi;
1619         sbi->s_mount_opt = 0;
1620         sbi->s_resuid = EXT3_DEF_RESUID;
1621         sbi->s_resgid = EXT3_DEF_RESGID;
1622         sbi->s_sb_block = sb_block;
1623
1624         unlock_kernel();
1625
1626         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1627         if (!blocksize) {
1628                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1629                 goto out_fail;
1630         }
1631
1632         /*
1633          * The ext3 superblock will not be buffer aligned for other than 1kB
1634          * block sizes.  We need to calculate the offset from buffer start.
1635          */
1636         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1637                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1638                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1639         } else {
1640                 logic_sb_block = sb_block;
1641         }
1642
1643         if (!(bh = sb_bread(sb, logic_sb_block))) {
1644                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1645                 goto out_fail;
1646         }
1647         /*
1648          * Note: s_es must be initialized as soon as possible because
1649          *       some ext3 macro-instructions depend on its value
1650          */
1651         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1652         sbi->s_es = es;
1653         sb->s_magic = le16_to_cpu(es->s_magic);
1654         if (sb->s_magic != EXT3_SUPER_MAGIC)
1655                 goto cantfind_ext3;
1656
1657         /* Set defaults before we parse the mount options */
1658         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1659         if (def_mount_opts & EXT3_DEFM_DEBUG)
1660                 set_opt(sbi->s_mount_opt, DEBUG);
1661         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1662                 set_opt(sbi->s_mount_opt, GRPID);
1663         if (def_mount_opts & EXT3_DEFM_UID16)
1664                 set_opt(sbi->s_mount_opt, NO_UID32);
1665 #ifdef CONFIG_EXT3_FS_XATTR
1666         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1667                 set_opt(sbi->s_mount_opt, XATTR_USER);
1668 #endif
1669 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1670         if (def_mount_opts & EXT3_DEFM_ACL)
1671                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1672 #endif
1673         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1674                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1675         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1676                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1677         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1678                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1679
1680         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1681                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1682         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1683                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1684         else
1685                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1686
1687         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1688         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1689
1690         set_opt(sbi->s_mount_opt, RESERVATION);
1691
1692         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1693                             NULL, 0))
1694                 goto failed_mount;
1695
1696         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1697                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1698
1699         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1700             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1701              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1702              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1703                 ext3_msg(sb, KERN_WARNING,
1704                         "warning: feature flags set on rev 0 fs, "
1705                         "running e2fsck is recommended");
1706         /*
1707          * Check feature flags regardless of the revision level, since we
1708          * previously didn't change the revision level when setting the flags,
1709          * so there is a chance incompat flags are set on a rev 0 filesystem.
1710          */
1711         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1712         if (features) {
1713                 ext3_msg(sb, KERN_ERR,
1714                         "error: couldn't mount because of unsupported "
1715                         "optional features (%x)", le32_to_cpu(features));
1716                 goto failed_mount;
1717         }
1718         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1719         if (!(sb->s_flags & MS_RDONLY) && features) {
1720                 ext3_msg(sb, KERN_ERR,
1721                         "error: couldn't mount RDWR because of unsupported "
1722                         "optional features (%x)", le32_to_cpu(features));
1723                 goto failed_mount;
1724         }
1725         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1726
1727         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1728             blocksize > EXT3_MAX_BLOCK_SIZE) {
1729                 ext3_msg(sb, KERN_ERR,
1730                         "error: couldn't mount because of unsupported "
1731                         "filesystem blocksize %d", blocksize);
1732                 goto failed_mount;
1733         }
1734
1735         hblock = bdev_logical_block_size(sb->s_bdev);
1736         if (sb->s_blocksize != blocksize) {
1737                 /*
1738                  * Make sure the blocksize for the filesystem is larger
1739                  * than the hardware sectorsize for the machine.
1740                  */
1741                 if (blocksize < hblock) {
1742                         ext3_msg(sb, KERN_ERR,
1743                                 "error: fsblocksize %d too small for "
1744                                 "hardware sectorsize %d", blocksize, hblock);
1745                         goto failed_mount;
1746                 }
1747
1748                 brelse (bh);
1749                 if (!sb_set_blocksize(sb, blocksize)) {
1750                         ext3_msg(sb, KERN_ERR,
1751                                 "error: bad blocksize %d", blocksize);
1752                         goto out_fail;
1753                 }
1754                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1755                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1756                 bh = sb_bread(sb, logic_sb_block);
1757                 if (!bh) {
1758                         ext3_msg(sb, KERN_ERR,
1759                                "error: can't read superblock on 2nd try");
1760                         goto failed_mount;
1761                 }
1762                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1763                 sbi->s_es = es;
1764                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1765                         ext3_msg(sb, KERN_ERR,
1766                                 "error: magic mismatch");
1767                         goto failed_mount;
1768                 }
1769         }
1770
1771         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1772
1773         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1774                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1775                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1776         } else {
1777                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1778                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1779                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1780                     (!is_power_of_2(sbi->s_inode_size)) ||
1781                     (sbi->s_inode_size > blocksize)) {
1782                         ext3_msg(sb, KERN_ERR,
1783                                 "error: unsupported inode size: %d",
1784                                 sbi->s_inode_size);
1785                         goto failed_mount;
1786                 }
1787         }
1788         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1789                                    le32_to_cpu(es->s_log_frag_size);
1790         if (blocksize != sbi->s_frag_size) {
1791                 ext3_msg(sb, KERN_ERR,
1792                        "error: fragsize %lu != blocksize %u (unsupported)",
1793                        sbi->s_frag_size, blocksize);
1794                 goto failed_mount;
1795         }
1796         sbi->s_frags_per_block = 1;
1797         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1798         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1799         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1800         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1801                 goto cantfind_ext3;
1802         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1803         if (sbi->s_inodes_per_block == 0)
1804                 goto cantfind_ext3;
1805         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1806                                         sbi->s_inodes_per_block;
1807         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1808         sbi->s_sbh = bh;
1809         sbi->s_mount_state = le16_to_cpu(es->s_state);
1810         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1811         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1812         for (i=0; i < 4; i++)
1813                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1814         sbi->s_def_hash_version = es->s_def_hash_version;
1815         i = le32_to_cpu(es->s_flags);
1816         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1817                 sbi->s_hash_unsigned = 3;
1818         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1819 #ifdef __CHAR_UNSIGNED__
1820                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1821                 sbi->s_hash_unsigned = 3;
1822 #else
1823                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1824 #endif
1825         }
1826
1827         if (sbi->s_blocks_per_group > blocksize * 8) {
1828                 ext3_msg(sb, KERN_ERR,
1829                         "#blocks per group too big: %lu",
1830                         sbi->s_blocks_per_group);
1831                 goto failed_mount;
1832         }
1833         if (sbi->s_frags_per_group > blocksize * 8) {
1834                 ext3_msg(sb, KERN_ERR,
1835                         "error: #fragments per group too big: %lu",
1836                         sbi->s_frags_per_group);
1837                 goto failed_mount;
1838         }
1839         if (sbi->s_inodes_per_group > blocksize * 8) {
1840                 ext3_msg(sb, KERN_ERR,
1841                         "error: #inodes per group too big: %lu",
1842                         sbi->s_inodes_per_group);
1843                 goto failed_mount;
1844         }
1845
1846         if (le32_to_cpu(es->s_blocks_count) >
1847                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1848                 ext3_msg(sb, KERN_ERR,
1849                         "error: filesystem is too large to mount safely");
1850                 if (sizeof(sector_t) < 8)
1851                         ext3_msg(sb, KERN_ERR,
1852                                 "error: CONFIG_LBDAF not enabled");
1853                 goto failed_mount;
1854         }
1855
1856         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1857                 goto cantfind_ext3;
1858         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1859                                le32_to_cpu(es->s_first_data_block) - 1)
1860                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1861         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1862                    EXT3_DESC_PER_BLOCK(sb);
1863         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1864                                     GFP_KERNEL);
1865         if (sbi->s_group_desc == NULL) {
1866                 ext3_msg(sb, KERN_ERR,
1867                         "error: not enough memory");
1868                 goto failed_mount;
1869         }
1870
1871         bgl_lock_init(sbi->s_blockgroup_lock);
1872
1873         for (i = 0; i < db_count; i++) {
1874                 block = descriptor_loc(sb, logic_sb_block, i);
1875                 sbi->s_group_desc[i] = sb_bread(sb, block);
1876                 if (!sbi->s_group_desc[i]) {
1877                         ext3_msg(sb, KERN_ERR,
1878                                 "error: can't read group descriptor %d", i);
1879                         db_count = i;
1880                         goto failed_mount2;
1881                 }
1882         }
1883         if (!ext3_check_descriptors (sb)) {
1884                 ext3_msg(sb, KERN_ERR,
1885                         "error: group descriptors corrupted");
1886                 goto failed_mount2;
1887         }
1888         sbi->s_gdb_count = db_count;
1889         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1890         spin_lock_init(&sbi->s_next_gen_lock);
1891
1892         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1893                         ext3_count_free_blocks(sb));
1894         if (!err) {
1895                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1896                                 ext3_count_free_inodes(sb));
1897         }
1898         if (!err) {
1899                 err = percpu_counter_init(&sbi->s_dirs_counter,
1900                                 ext3_count_dirs(sb));
1901         }
1902         if (err) {
1903                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
1904                 goto failed_mount3;
1905         }
1906
1907         /* per fileystem reservation list head & lock */
1908         spin_lock_init(&sbi->s_rsv_window_lock);
1909         sbi->s_rsv_window_root = RB_ROOT;
1910         /* Add a single, static dummy reservation to the start of the
1911          * reservation window list --- it gives us a placeholder for
1912          * append-at-start-of-list which makes the allocation logic
1913          * _much_ simpler. */
1914         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1915         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1916         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1917         sbi->s_rsv_window_head.rsv_goal_size = 0;
1918         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1919
1920         /*
1921          * set up enough so that it can read an inode
1922          */
1923         sb->s_op = &ext3_sops;
1924         sb->s_export_op = &ext3_export_ops;
1925         sb->s_xattr = ext3_xattr_handlers;
1926 #ifdef CONFIG_QUOTA
1927         sb->s_qcop = &ext3_qctl_operations;
1928         sb->dq_op = &ext3_quota_operations;
1929 #endif
1930         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1931
1932         sb->s_root = NULL;
1933
1934         needs_recovery = (es->s_last_orphan != 0 ||
1935                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1936                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1937
1938         /*
1939          * The first inode we look at is the journal inode.  Don't try
1940          * root first: it may be modified in the journal!
1941          */
1942         if (!test_opt(sb, NOLOAD) &&
1943             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1944                 if (ext3_load_journal(sb, es, journal_devnum))
1945                         goto failed_mount3;
1946         } else if (journal_inum) {
1947                 if (ext3_create_journal(sb, es, journal_inum))
1948                         goto failed_mount3;
1949         } else {
1950                 if (!silent)
1951                         ext3_msg(sb, KERN_ERR,
1952                                 "error: no journal found. "
1953                                 "mounting ext3 over ext2?");
1954                 goto failed_mount3;
1955         }
1956
1957         /* We have now updated the journal if required, so we can
1958          * validate the data journaling mode. */
1959         switch (test_opt(sb, DATA_FLAGS)) {
1960         case 0:
1961                 /* No mode set, assume a default based on the journal
1962                    capabilities: ORDERED_DATA if the journal can
1963                    cope, else JOURNAL_DATA */
1964                 if (journal_check_available_features
1965                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1966                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
1967                 else
1968                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1969                 break;
1970
1971         case EXT3_MOUNT_ORDERED_DATA:
1972         case EXT3_MOUNT_WRITEBACK_DATA:
1973                 if (!journal_check_available_features
1974                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1975                         ext3_msg(sb, KERN_ERR,
1976                                 "error: journal does not support "
1977                                 "requested data journaling mode");
1978                         goto failed_mount4;
1979                 }
1980         default:
1981                 break;
1982         }
1983
1984         if (test_opt(sb, NOBH)) {
1985                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1986                         ext3_msg(sb, KERN_WARNING,
1987                                 "warning: ignoring nobh option - "
1988                                 "it is supported only with writeback mode");
1989                         clear_opt(sbi->s_mount_opt, NOBH);
1990                 }
1991         }
1992         /*
1993          * The journal_load will have done any necessary log recovery,
1994          * so we can safely mount the rest of the filesystem now.
1995          */
1996
1997         root = ext3_iget(sb, EXT3_ROOT_INO);
1998         if (IS_ERR(root)) {
1999                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
2000                 ret = PTR_ERR(root);
2001                 goto failed_mount4;
2002         }
2003         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2004                 iput(root);
2005                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2006                 goto failed_mount4;
2007         }
2008         sb->s_root = d_alloc_root(root);
2009         if (!sb->s_root) {
2010                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2011                 iput(root);
2012                 ret = -ENOMEM;
2013                 goto failed_mount4;
2014         }
2015
2016         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2017
2018         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2019         ext3_orphan_cleanup(sb, es);
2020         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2021         if (needs_recovery)
2022                 ext3_msg(sb, KERN_INFO, "recovery complete");
2023         ext3_mark_recovery_complete(sb, es);
2024         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2025                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2026                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2027                 "writeback");
2028
2029         lock_kernel();
2030         return 0;
2031
2032 cantfind_ext3:
2033         if (!silent)
2034                 ext3_msg(sb, KERN_INFO,
2035                         "error: can't find ext3 filesystem on dev %s.",
2036                        sb->s_id);
2037         goto failed_mount;
2038
2039 failed_mount4:
2040         journal_destroy(sbi->s_journal);
2041 failed_mount3:
2042         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2043         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2044         percpu_counter_destroy(&sbi->s_dirs_counter);
2045 failed_mount2:
2046         for (i = 0; i < db_count; i++)
2047                 brelse(sbi->s_group_desc[i]);
2048         kfree(sbi->s_group_desc);
2049 failed_mount:
2050 #ifdef CONFIG_QUOTA
2051         for (i = 0; i < MAXQUOTAS; i++)
2052                 kfree(sbi->s_qf_names[i]);
2053 #endif
2054         ext3_blkdev_remove(sbi);
2055         brelse(bh);
2056 out_fail:
2057         sb->s_fs_info = NULL;
2058         kfree(sbi->s_blockgroup_lock);
2059         kfree(sbi);
2060         lock_kernel();
2061         return ret;
2062 }
2063
2064 /*
2065  * Setup any per-fs journal parameters now.  We'll do this both on
2066  * initial mount, once the journal has been initialised but before we've
2067  * done any recovery; and again on any subsequent remount.
2068  */
2069 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2070 {
2071         struct ext3_sb_info *sbi = EXT3_SB(sb);
2072
2073         if (sbi->s_commit_interval)
2074                 journal->j_commit_interval = sbi->s_commit_interval;
2075         /* We could also set up an ext3-specific default for the commit
2076          * interval here, but for now we'll just fall back to the jbd
2077          * default. */
2078
2079         spin_lock(&journal->j_state_lock);
2080         if (test_opt(sb, BARRIER))
2081                 journal->j_flags |= JFS_BARRIER;
2082         else
2083                 journal->j_flags &= ~JFS_BARRIER;
2084         if (test_opt(sb, DATA_ERR_ABORT))
2085                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2086         else
2087                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2088         spin_unlock(&journal->j_state_lock);
2089 }
2090
2091 static journal_t *ext3_get_journal(struct super_block *sb,
2092                                    unsigned int journal_inum)
2093 {
2094         struct inode *journal_inode;
2095         journal_t *journal;
2096
2097         /* First, test for the existence of a valid inode on disk.  Bad
2098          * things happen if we iget() an unused inode, as the subsequent
2099          * iput() will try to delete it. */
2100
2101         journal_inode = ext3_iget(sb, journal_inum);
2102         if (IS_ERR(journal_inode)) {
2103                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2104                 return NULL;
2105         }
2106         if (!journal_inode->i_nlink) {
2107                 make_bad_inode(journal_inode);
2108                 iput(journal_inode);
2109                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2110                 return NULL;
2111         }
2112
2113         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2114                   journal_inode, journal_inode->i_size);
2115         if (!S_ISREG(journal_inode->i_mode)) {
2116                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2117                 iput(journal_inode);
2118                 return NULL;
2119         }
2120
2121         journal = journal_init_inode(journal_inode);
2122         if (!journal) {
2123                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2124                 iput(journal_inode);
2125                 return NULL;
2126         }
2127         journal->j_private = sb;
2128         ext3_init_journal_params(sb, journal);
2129         return journal;
2130 }
2131
2132 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2133                                        dev_t j_dev)
2134 {
2135         struct buffer_head * bh;
2136         journal_t *journal;
2137         ext3_fsblk_t start;
2138         ext3_fsblk_t len;
2139         int hblock, blocksize;
2140         ext3_fsblk_t sb_block;
2141         unsigned long offset;
2142         struct ext3_super_block * es;
2143         struct block_device *bdev;
2144
2145         bdev = ext3_blkdev_get(j_dev, sb);
2146         if (bdev == NULL)
2147                 return NULL;
2148
2149         if (bd_claim(bdev, sb)) {
2150                 ext3_msg(sb, KERN_ERR,
2151                         "error: failed to claim external journal device");
2152                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2153                 return NULL;
2154         }
2155
2156         blocksize = sb->s_blocksize;
2157         hblock = bdev_logical_block_size(bdev);
2158         if (blocksize < hblock) {
2159                 ext3_msg(sb, KERN_ERR,
2160                         "error: blocksize too small for journal device");
2161                 goto out_bdev;
2162         }
2163
2164         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2165         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2166         set_blocksize(bdev, blocksize);
2167         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2168                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2169                         "external journal");
2170                 goto out_bdev;
2171         }
2172
2173         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2174         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2175             !(le32_to_cpu(es->s_feature_incompat) &
2176               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2177                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2178                         "bad superblock");
2179                 brelse(bh);
2180                 goto out_bdev;
2181         }
2182
2183         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2184                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2185                 brelse(bh);
2186                 goto out_bdev;
2187         }
2188
2189         len = le32_to_cpu(es->s_blocks_count);
2190         start = sb_block + 1;
2191         brelse(bh);     /* we're done with the superblock */
2192
2193         journal = journal_init_dev(bdev, sb->s_bdev,
2194                                         start, len, blocksize);
2195         if (!journal) {
2196                 ext3_msg(sb, KERN_ERR,
2197                         "error: failed to create device journal");
2198                 goto out_bdev;
2199         }
2200         journal->j_private = sb;
2201         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2202         wait_on_buffer(journal->j_sb_buffer);
2203         if (!buffer_uptodate(journal->j_sb_buffer)) {
2204                 ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2205                 goto out_journal;
2206         }
2207         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2208                 ext3_msg(sb, KERN_ERR,
2209                         "error: external journal has more than one "
2210                         "user (unsupported) - %d",
2211                         be32_to_cpu(journal->j_superblock->s_nr_users));
2212                 goto out_journal;
2213         }
2214         EXT3_SB(sb)->journal_bdev = bdev;
2215         ext3_init_journal_params(sb, journal);
2216         return journal;
2217 out_journal:
2218         journal_destroy(journal);
2219 out_bdev:
2220         ext3_blkdev_put(bdev);
2221         return NULL;
2222 }
2223
2224 static int ext3_load_journal(struct super_block *sb,
2225                              struct ext3_super_block *es,
2226                              unsigned long journal_devnum)
2227 {
2228         journal_t *journal;
2229         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2230         dev_t journal_dev;
2231         int err = 0;
2232         int really_read_only;
2233
2234         if (journal_devnum &&
2235             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2236                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2237                         "numbers have changed");
2238                 journal_dev = new_decode_dev(journal_devnum);
2239         } else
2240                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2241
2242         really_read_only = bdev_read_only(sb->s_bdev);
2243
2244         /*
2245          * Are we loading a blank journal or performing recovery after a
2246          * crash?  For recovery, we need to check in advance whether we
2247          * can get read-write access to the device.
2248          */
2249
2250         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2251                 if (sb->s_flags & MS_RDONLY) {
2252                         ext3_msg(sb, KERN_INFO,
2253                                 "recovery required on readonly filesystem");
2254                         if (really_read_only) {
2255                                 ext3_msg(sb, KERN_ERR, "error: write access "
2256                                         "unavailable, cannot proceed");
2257                                 return -EROFS;
2258                         }
2259                         ext3_msg(sb, KERN_INFO,
2260                                 "write access will be enabled during recovery");
2261                 }
2262         }
2263
2264         if (journal_inum && journal_dev) {
2265                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2266                        "and inode journals");
2267                 return -EINVAL;
2268         }
2269
2270         if (journal_inum) {
2271                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2272                         return -EINVAL;
2273         } else {
2274                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2275                         return -EINVAL;
2276         }
2277
2278         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2279                 err = journal_update_format(journal);
2280                 if (err)  {
2281                         ext3_msg(sb, KERN_ERR, "error updating journal");
2282                         journal_destroy(journal);
2283                         return err;
2284                 }
2285         }
2286
2287         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2288                 err = journal_wipe(journal, !really_read_only);
2289         if (!err)
2290                 err = journal_load(journal);
2291
2292         if (err) {
2293                 ext3_msg(sb, KERN_ERR, "error loading journal");
2294                 journal_destroy(journal);
2295                 return err;
2296         }
2297
2298         EXT3_SB(sb)->s_journal = journal;
2299         ext3_clear_journal_err(sb, es);
2300
2301         if (journal_devnum &&
2302             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2303                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2304
2305                 /* Make sure we flush the recovery flag to disk. */
2306                 ext3_commit_super(sb, es, 1);
2307         }
2308
2309         return 0;
2310 }
2311
2312 static int ext3_create_journal(struct super_block *sb,
2313                                struct ext3_super_block *es,
2314                                unsigned int journal_inum)
2315 {
2316         journal_t *journal;
2317         int err;
2318
2319         if (sb->s_flags & MS_RDONLY) {
2320                 ext3_msg(sb, KERN_ERR,
2321                         "error: readonly filesystem when trying to "
2322                         "create journal");
2323                 return -EROFS;
2324         }
2325
2326         journal = ext3_get_journal(sb, journal_inum);
2327         if (!journal)
2328                 return -EINVAL;
2329
2330         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2331                journal_inum);
2332
2333         err = journal_create(journal);
2334         if (err) {
2335                 ext3_msg(sb, KERN_ERR, "error creating journal");
2336                 journal_destroy(journal);
2337                 return -EIO;
2338         }
2339
2340         EXT3_SB(sb)->s_journal = journal;
2341
2342         ext3_update_dynamic_rev(sb);
2343         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2344         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2345
2346         es->s_journal_inum = cpu_to_le32(journal_inum);
2347
2348         /* Make sure we flush the recovery flag to disk. */
2349         ext3_commit_super(sb, es, 1);
2350
2351         return 0;
2352 }
2353
2354 static int ext3_commit_super(struct super_block *sb,
2355                                struct ext3_super_block *es,
2356                                int sync)
2357 {
2358         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2359         int error = 0;
2360
2361         if (!sbh)
2362                 return error;
2363         /*
2364          * If the file system is mounted read-only, don't update the
2365          * superblock write time.  This avoids updating the superblock
2366          * write time when we are mounting the root file system
2367          * read/only but we need to replay the journal; at that point,
2368          * for people who are east of GMT and who make their clock
2369          * tick in localtime for Windows bug-for-bug compatibility,
2370          * the clock is set in the future, and this will cause e2fsck
2371          * to complain and force a full file system check.
2372          */
2373         if (!(sb->s_flags & MS_RDONLY))
2374                 es->s_wtime = cpu_to_le32(get_seconds());
2375         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2376         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2377         BUFFER_TRACE(sbh, "marking dirty");
2378         mark_buffer_dirty(sbh);
2379         if (sync)
2380                 error = sync_dirty_buffer(sbh);
2381         return error;
2382 }
2383
2384
2385 /*
2386  * Have we just finished recovery?  If so, and if we are mounting (or
2387  * remounting) the filesystem readonly, then we will end up with a
2388  * consistent fs on disk.  Record that fact.
2389  */
2390 static void ext3_mark_recovery_complete(struct super_block * sb,
2391                                         struct ext3_super_block * es)
2392 {
2393         journal_t *journal = EXT3_SB(sb)->s_journal;
2394
2395         journal_lock_updates(journal);
2396         if (journal_flush(journal) < 0)
2397                 goto out;
2398
2399         lock_super(sb);
2400         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2401             sb->s_flags & MS_RDONLY) {
2402                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2403                 ext3_commit_super(sb, es, 1);
2404         }
2405         unlock_super(sb);
2406
2407 out:
2408         journal_unlock_updates(journal);
2409 }
2410
2411 /*
2412  * If we are mounting (or read-write remounting) a filesystem whose journal
2413  * has recorded an error from a previous lifetime, move that error to the
2414  * main filesystem now.
2415  */
2416 static void ext3_clear_journal_err(struct super_block *sb,
2417                                    struct ext3_super_block *es)
2418 {
2419         journal_t *journal;
2420         int j_errno;
2421         const char *errstr;
2422
2423         journal = EXT3_SB(sb)->s_journal;
2424
2425         /*
2426          * Now check for any error status which may have been recorded in the
2427          * journal by a prior ext3_error() or ext3_abort()
2428          */
2429
2430         j_errno = journal_errno(journal);
2431         if (j_errno) {
2432                 char nbuf[16];
2433
2434                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2435                 ext3_warning(sb, __func__, "Filesystem error recorded "
2436                              "from previous mount: %s", errstr);
2437                 ext3_warning(sb, __func__, "Marking fs in need of "
2438                              "filesystem check.");
2439
2440                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2441                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2442                 ext3_commit_super (sb, es, 1);
2443
2444                 journal_clear_err(journal);
2445         }
2446 }
2447
2448 /*
2449  * Force the running and committing transactions to commit,
2450  * and wait on the commit.
2451  */
2452 int ext3_force_commit(struct super_block *sb)
2453 {
2454         journal_t *journal;
2455         int ret;
2456
2457         if (sb->s_flags & MS_RDONLY)
2458                 return 0;
2459
2460         journal = EXT3_SB(sb)->s_journal;
2461         ret = ext3_journal_force_commit(journal);
2462         return ret;
2463 }
2464
2465 static int ext3_sync_fs(struct super_block *sb, int wait)
2466 {
2467         tid_t target;
2468
2469         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2470                 if (wait)
2471                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2472         }
2473         return 0;
2474 }
2475
2476 /*
2477  * LVM calls this function before a (read-only) snapshot is created.  This
2478  * gives us a chance to flush the journal completely and mark the fs clean.
2479  */
2480 static int ext3_freeze(struct super_block *sb)
2481 {
2482         int error = 0;
2483         journal_t *journal;
2484
2485         if (!(sb->s_flags & MS_RDONLY)) {
2486                 journal = EXT3_SB(sb)->s_journal;
2487
2488                 /* Now we set up the journal barrier. */
2489                 journal_lock_updates(journal);
2490
2491                 /*
2492                  * We don't want to clear needs_recovery flag when we failed
2493                  * to flush the journal.
2494                  */
2495                 error = journal_flush(journal);
2496                 if (error < 0)
2497                         goto out;
2498
2499                 /* Journal blocked and flushed, clear needs_recovery flag. */
2500                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2501                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2502                 if (error)
2503                         goto out;
2504         }
2505         return 0;
2506
2507 out:
2508         journal_unlock_updates(journal);
2509         return error;
2510 }
2511
2512 /*
2513  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2514  * flag here, even though the filesystem is not technically dirty yet.
2515  */
2516 static int ext3_unfreeze(struct super_block *sb)
2517 {
2518         if (!(sb->s_flags & MS_RDONLY)) {
2519                 lock_super(sb);
2520                 /* Reser the needs_recovery flag before the fs is unlocked. */
2521                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2522                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2523                 unlock_super(sb);
2524                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2525         }
2526         return 0;
2527 }
2528
2529 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2530 {
2531         struct ext3_super_block * es;
2532         struct ext3_sb_info *sbi = EXT3_SB(sb);
2533         ext3_fsblk_t n_blocks_count = 0;
2534         unsigned long old_sb_flags;
2535         struct ext3_mount_options old_opts;
2536         int err;
2537 #ifdef CONFIG_QUOTA
2538         int i;
2539 #endif
2540
2541         lock_kernel();
2542
2543         /* Store the original options */
2544         lock_super(sb);
2545         old_sb_flags = sb->s_flags;
2546         old_opts.s_mount_opt = sbi->s_mount_opt;
2547         old_opts.s_resuid = sbi->s_resuid;
2548         old_opts.s_resgid = sbi->s_resgid;
2549         old_opts.s_commit_interval = sbi->s_commit_interval;
2550 #ifdef CONFIG_QUOTA
2551         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2552         for (i = 0; i < MAXQUOTAS; i++)
2553                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2554 #endif
2555
2556         /*
2557          * Allow the "check" option to be passed as a remount option.
2558          */
2559         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2560                 err = -EINVAL;
2561                 goto restore_opts;
2562         }
2563
2564         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2565                 ext3_abort(sb, __func__, "Abort forced by user");
2566
2567         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2568                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2569
2570         es = sbi->s_es;
2571
2572         ext3_init_journal_params(sb, sbi->s_journal);
2573
2574         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2575                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2576                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2577                         err = -EROFS;
2578                         goto restore_opts;
2579                 }
2580
2581                 if (*flags & MS_RDONLY) {
2582                         /*
2583                          * First of all, the unconditional stuff we have to do
2584                          * to disable replay of the journal when we next remount
2585                          */
2586                         sb->s_flags |= MS_RDONLY;
2587
2588                         /*
2589                          * OK, test if we are remounting a valid rw partition
2590                          * readonly, and if so set the rdonly flag and then
2591                          * mark the partition as valid again.
2592                          */
2593                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2594                             (sbi->s_mount_state & EXT3_VALID_FS))
2595                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2596
2597                         /*
2598                          * We have to unlock super so that we can wait for
2599                          * transactions.
2600                          */
2601                         unlock_super(sb);
2602                         ext3_mark_recovery_complete(sb, es);
2603                         lock_super(sb);
2604                 } else {
2605                         __le32 ret;
2606                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2607                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2608                                 ext3_msg(sb, KERN_WARNING,
2609                                         "warning: couldn't remount RDWR "
2610                                         "because of unsupported optional "
2611                                         "features (%x)", le32_to_cpu(ret));
2612                                 err = -EROFS;
2613                                 goto restore_opts;
2614                         }
2615
2616                         /*
2617                          * If we have an unprocessed orphan list hanging
2618                          * around from a previously readonly bdev mount,
2619                          * require a full umount/remount for now.
2620                          */
2621                         if (es->s_last_orphan) {
2622                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2623                                        "remount RDWR because of unprocessed "
2624                                        "orphan inode list.  Please "
2625                                        "umount/remount instead.");
2626                                 err = -EINVAL;
2627                                 goto restore_opts;
2628                         }
2629
2630                         /*
2631                          * Mounting a RDONLY partition read-write, so reread
2632                          * and store the current valid flag.  (It may have
2633                          * been changed by e2fsck since we originally mounted
2634                          * the partition.)
2635                          */
2636                         ext3_clear_journal_err(sb, es);
2637                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2638                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2639                                 goto restore_opts;
2640                         if (!ext3_setup_super (sb, es, 0))
2641                                 sb->s_flags &= ~MS_RDONLY;
2642                 }
2643         }
2644 #ifdef CONFIG_QUOTA
2645         /* Release old quota file names */
2646         for (i = 0; i < MAXQUOTAS; i++)
2647                 if (old_opts.s_qf_names[i] &&
2648                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2649                         kfree(old_opts.s_qf_names[i]);
2650 #endif
2651         unlock_super(sb);
2652         unlock_kernel();
2653         return 0;
2654 restore_opts:
2655         sb->s_flags = old_sb_flags;
2656         sbi->s_mount_opt = old_opts.s_mount_opt;
2657         sbi->s_resuid = old_opts.s_resuid;
2658         sbi->s_resgid = old_opts.s_resgid;
2659         sbi->s_commit_interval = old_opts.s_commit_interval;
2660 #ifdef CONFIG_QUOTA
2661         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2662         for (i = 0; i < MAXQUOTAS; i++) {
2663                 if (sbi->s_qf_names[i] &&
2664                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2665                         kfree(sbi->s_qf_names[i]);
2666                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2667         }
2668 #endif
2669         unlock_super(sb);
2670         unlock_kernel();
2671         return err;
2672 }
2673
2674 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2675 {
2676         struct super_block *sb = dentry->d_sb;
2677         struct ext3_sb_info *sbi = EXT3_SB(sb);
2678         struct ext3_super_block *es = sbi->s_es;
2679         u64 fsid;
2680
2681         if (test_opt(sb, MINIX_DF)) {
2682                 sbi->s_overhead_last = 0;
2683         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2684                 unsigned long ngroups = sbi->s_groups_count, i;
2685                 ext3_fsblk_t overhead = 0;
2686                 smp_rmb();
2687
2688                 /*
2689                  * Compute the overhead (FS structures).  This is constant
2690                  * for a given filesystem unless the number of block groups
2691                  * changes so we cache the previous value until it does.
2692                  */
2693
2694                 /*
2695                  * All of the blocks before first_data_block are
2696                  * overhead
2697                  */
2698                 overhead = le32_to_cpu(es->s_first_data_block);
2699
2700                 /*
2701                  * Add the overhead attributed to the superblock and
2702                  * block group descriptors.  If the sparse superblocks
2703                  * feature is turned on, then not all groups have this.
2704                  */
2705                 for (i = 0; i < ngroups; i++) {
2706                         overhead += ext3_bg_has_super(sb, i) +
2707                                 ext3_bg_num_gdb(sb, i);
2708                         cond_resched();
2709                 }
2710
2711                 /*
2712                  * Every block group has an inode bitmap, a block
2713                  * bitmap, and an inode table.
2714                  */
2715                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2716                 sbi->s_overhead_last = overhead;
2717                 smp_wmb();
2718                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2719         }
2720
2721         buf->f_type = EXT3_SUPER_MAGIC;
2722         buf->f_bsize = sb->s_blocksize;
2723         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2724         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2725         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2726         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2727                 buf->f_bavail = 0;
2728         buf->f_files = le32_to_cpu(es->s_inodes_count);
2729         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2730         buf->f_namelen = EXT3_NAME_LEN;
2731         fsid = le64_to_cpup((void *)es->s_uuid) ^
2732                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2733         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2734         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2735         return 0;
2736 }
2737
2738 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2739  * is locked for write. Otherwise the are possible deadlocks:
2740  * Process 1                         Process 2
2741  * ext3_create()                     quota_sync()
2742  *   journal_start()                   write_dquot()
2743  *   vfs_dq_init()                       down(dqio_mutex)
2744  *     down(dqio_mutex)                    journal_start()
2745  *
2746  */
2747
2748 #ifdef CONFIG_QUOTA
2749
2750 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2751 {
2752         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2753 }
2754
2755 static int ext3_write_dquot(struct dquot *dquot)
2756 {
2757         int ret, err;
2758         handle_t *handle;
2759         struct inode *inode;
2760
2761         inode = dquot_to_inode(dquot);
2762         handle = ext3_journal_start(inode,
2763                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2764         if (IS_ERR(handle))
2765                 return PTR_ERR(handle);
2766         ret = dquot_commit(dquot);
2767         err = ext3_journal_stop(handle);
2768         if (!ret)
2769                 ret = err;
2770         return ret;
2771 }
2772
2773 static int ext3_acquire_dquot(struct dquot *dquot)
2774 {
2775         int ret, err;
2776         handle_t *handle;
2777
2778         handle = ext3_journal_start(dquot_to_inode(dquot),
2779                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2780         if (IS_ERR(handle))
2781                 return PTR_ERR(handle);
2782         ret = dquot_acquire(dquot);
2783         err = ext3_journal_stop(handle);
2784         if (!ret)
2785                 ret = err;
2786         return ret;
2787 }
2788
2789 static int ext3_release_dquot(struct dquot *dquot)
2790 {
2791         int ret, err;
2792         handle_t *handle;
2793
2794         handle = ext3_journal_start(dquot_to_inode(dquot),
2795                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2796         if (IS_ERR(handle)) {
2797                 /* Release dquot anyway to avoid endless cycle in dqput() */
2798                 dquot_release(dquot);
2799                 return PTR_ERR(handle);
2800         }
2801         ret = dquot_release(dquot);
2802         err = ext3_journal_stop(handle);
2803         if (!ret)
2804                 ret = err;
2805         return ret;
2806 }
2807
2808 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2809 {
2810         /* Are we journaling quotas? */
2811         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2812             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2813                 dquot_mark_dquot_dirty(dquot);
2814                 return ext3_write_dquot(dquot);
2815         } else {
2816                 return dquot_mark_dquot_dirty(dquot);
2817         }
2818 }
2819
2820 static int ext3_write_info(struct super_block *sb, int type)
2821 {
2822         int ret, err;
2823         handle_t *handle;
2824
2825         /* Data block + inode block */
2826         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2827         if (IS_ERR(handle))
2828                 return PTR_ERR(handle);
2829         ret = dquot_commit_info(sb, type);
2830         err = ext3_journal_stop(handle);
2831         if (!ret)
2832                 ret = err;
2833         return ret;
2834 }
2835
2836 /*
2837  * Turn on quotas during mount time - we need to find
2838  * the quota file and such...
2839  */
2840 static int ext3_quota_on_mount(struct super_block *sb, int type)
2841 {
2842         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2843                         EXT3_SB(sb)->s_jquota_fmt, type);
2844 }
2845
2846 /*
2847  * Standard function to be called on quota_on
2848  */
2849 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2850                          char *name, int remount)
2851 {
2852         int err;
2853         struct path path;
2854
2855         if (!test_opt(sb, QUOTA))
2856                 return -EINVAL;
2857         /* When remounting, no checks are needed and in fact, name is NULL */
2858         if (remount)
2859                 return vfs_quota_on(sb, type, format_id, name, remount);
2860
2861         err = kern_path(name, LOOKUP_FOLLOW, &path);
2862         if (err)
2863                 return err;
2864
2865         /* Quotafile not on the same filesystem? */
2866         if (path.mnt->mnt_sb != sb) {
2867                 path_put(&path);
2868                 return -EXDEV;
2869         }
2870         /* Journaling quota? */
2871         if (EXT3_SB(sb)->s_qf_names[type]) {
2872                 /* Quotafile not of fs root? */
2873                 if (path.dentry->d_parent != sb->s_root)
2874                         ext3_msg(sb, KERN_WARNING,
2875                                 "warning: Quota file not on filesystem root. "
2876                                 "Journaled quota will not work.");
2877         }
2878
2879         /*
2880          * When we journal data on quota file, we have to flush journal to see
2881          * all updates to the file when we bypass pagecache...
2882          */
2883         if (ext3_should_journal_data(path.dentry->d_inode)) {
2884                 /*
2885                  * We don't need to lock updates but journal_flush() could
2886                  * otherwise be livelocked...
2887                  */
2888                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2889                 err = journal_flush(EXT3_SB(sb)->s_journal);
2890                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2891                 if (err) {
2892                         path_put(&path);
2893                         return err;
2894                 }
2895         }
2896
2897         err = vfs_quota_on_path(sb, type, format_id, &path);
2898         path_put(&path);
2899         return err;
2900 }
2901
2902 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2903  * acquiring the locks... As quota files are never truncated and quota code
2904  * itself serializes the operations (and noone else should touch the files)
2905  * we don't have to be afraid of races */
2906 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2907                                size_t len, loff_t off)
2908 {
2909         struct inode *inode = sb_dqopt(sb)->files[type];
2910         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2911         int err = 0;
2912         int offset = off & (sb->s_blocksize - 1);
2913         int tocopy;
2914         size_t toread;
2915         struct buffer_head *bh;
2916         loff_t i_size = i_size_read(inode);
2917
2918         if (off > i_size)
2919                 return 0;
2920         if (off+len > i_size)
2921                 len = i_size-off;
2922         toread = len;
2923         while (toread > 0) {
2924                 tocopy = sb->s_blocksize - offset < toread ?
2925                                 sb->s_blocksize - offset : toread;
2926                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2927                 if (err)
2928                         return err;
2929                 if (!bh)        /* A hole? */
2930                         memset(data, 0, tocopy);
2931                 else
2932                         memcpy(data, bh->b_data+offset, tocopy);
2933                 brelse(bh);
2934                 offset = 0;
2935                 toread -= tocopy;
2936                 data += tocopy;
2937                 blk++;
2938         }
2939         return len;
2940 }
2941
2942 /* Write to quotafile (we know the transaction is already started and has
2943  * enough credits) */
2944 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2945                                 const char *data, size_t len, loff_t off)
2946 {
2947         struct inode *inode = sb_dqopt(sb)->files[type];
2948         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2949         int err = 0;
2950         int offset = off & (sb->s_blocksize - 1);
2951         int tocopy;
2952         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2953         size_t towrite = len;
2954         struct buffer_head *bh;
2955         handle_t *handle = journal_current_handle();
2956
2957         if (!handle) {
2958                 ext3_msg(sb, KERN_WARNING,
2959                         "warning: quota write (off=%llu, len=%llu)"
2960                         " cancelled because transaction is not started.",
2961                         (unsigned long long)off, (unsigned long long)len);
2962                 return -EIO;
2963         }
2964         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2965         while (towrite > 0) {
2966                 tocopy = sb->s_blocksize - offset < towrite ?
2967                                 sb->s_blocksize - offset : towrite;
2968                 bh = ext3_bread(handle, inode, blk, 1, &err);
2969                 if (!bh)
2970                         goto out;
2971                 if (journal_quota) {
2972                         err = ext3_journal_get_write_access(handle, bh);
2973                         if (err) {
2974                                 brelse(bh);
2975                                 goto out;
2976                         }
2977                 }
2978                 lock_buffer(bh);
2979                 memcpy(bh->b_data+offset, data, tocopy);
2980                 flush_dcache_page(bh->b_page);
2981                 unlock_buffer(bh);
2982                 if (journal_quota)
2983                         err = ext3_journal_dirty_metadata(handle, bh);
2984                 else {
2985                         /* Always do at least ordered writes for quotas */
2986                         err = ext3_journal_dirty_data(handle, bh);
2987                         mark_buffer_dirty(bh);
2988                 }
2989                 brelse(bh);
2990                 if (err)
2991                         goto out;
2992                 offset = 0;
2993                 towrite -= tocopy;
2994                 data += tocopy;
2995                 blk++;
2996         }
2997 out:
2998         if (len == towrite) {
2999                 mutex_unlock(&inode->i_mutex);
3000                 return err;
3001         }
3002         if (inode->i_size < off+len-towrite) {
3003                 i_size_write(inode, off+len-towrite);
3004                 EXT3_I(inode)->i_disksize = inode->i_size;
3005         }
3006         inode->i_version++;
3007         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3008         ext3_mark_inode_dirty(handle, inode);
3009         mutex_unlock(&inode->i_mutex);
3010         return len - towrite;
3011 }
3012
3013 #endif
3014
3015 static int ext3_get_sb(struct file_system_type *fs_type,
3016         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3017 {
3018         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
3019 }
3020
3021 static struct file_system_type ext3_fs_type = {
3022         .owner          = THIS_MODULE,
3023         .name           = "ext3",
3024         .get_sb         = ext3_get_sb,
3025         .kill_sb        = kill_block_super,
3026         .fs_flags       = FS_REQUIRES_DEV,
3027 };
3028
3029 static int __init init_ext3_fs(void)
3030 {
3031         int err = init_ext3_xattr();
3032         if (err)
3033                 return err;
3034         err = init_inodecache();
3035         if (err)
3036                 goto out1;
3037         err = register_filesystem(&ext3_fs_type);
3038         if (err)
3039                 goto out;
3040         return 0;
3041 out:
3042         destroy_inodecache();
3043 out1:
3044         exit_ext3_xattr();
3045         return err;
3046 }
3047
3048 static void __exit exit_ext3_fs(void)
3049 {
3050         unregister_filesystem(&ext3_fs_type);
3051         destroy_inodecache();
3052         exit_ext3_xattr();
3053 }
3054
3055 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3056 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3057 MODULE_LICENSE("GPL");
3058 module_init(init_ext3_fs)
3059 module_exit(exit_ext3_fs)