Merge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs
[~shefty/rdma-dev.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/blkdev.h>
21 #include <linux/parser.h>
22 #include <linux/exportfs.h>
23 #include <linux/statfs.h>
24 #include <linux/random.h>
25 #include <linux/mount.h>
26 #include <linux/quotaops.h>
27 #include <linux/seq_file.h>
28 #include <linux/log2.h>
29 #include <linux/cleancache.h>
30
31 #include <asm/uaccess.h>
32
33 #define CREATE_TRACE_POINTS
34
35 #include "ext3.h"
36 #include "xattr.h"
37 #include "acl.h"
38 #include "namei.h"
39
40 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
41   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
42 #else
43   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
44 #endif
45
46 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
47                              unsigned long journal_devnum);
48 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
49                                unsigned int);
50 static int ext3_commit_super(struct super_block *sb,
51                                struct ext3_super_block *es,
52                                int sync);
53 static void ext3_mark_recovery_complete(struct super_block * sb,
54                                         struct ext3_super_block * es);
55 static void ext3_clear_journal_err(struct super_block * sb,
56                                    struct ext3_super_block * es);
57 static int ext3_sync_fs(struct super_block *sb, int wait);
58 static const char *ext3_decode_error(struct super_block * sb, int errno,
59                                      char nbuf[16]);
60 static int ext3_remount (struct super_block * sb, int * flags, char * data);
61 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
62 static int ext3_unfreeze(struct super_block *sb);
63 static int ext3_freeze(struct super_block *sb);
64
65 /*
66  * Wrappers for journal_start/end.
67  */
68 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
69 {
70         journal_t *journal;
71
72         if (sb->s_flags & MS_RDONLY)
73                 return ERR_PTR(-EROFS);
74
75         /* Special case here: if the journal has aborted behind our
76          * backs (eg. EIO in the commit thread), then we still need to
77          * take the FS itself readonly cleanly. */
78         journal = EXT3_SB(sb)->s_journal;
79         if (is_journal_aborted(journal)) {
80                 ext3_abort(sb, __func__,
81                            "Detected aborted journal");
82                 return ERR_PTR(-EROFS);
83         }
84
85         return journal_start(journal, nblocks);
86 }
87
88 int __ext3_journal_stop(const char *where, handle_t *handle)
89 {
90         struct super_block *sb;
91         int err;
92         int rc;
93
94         sb = handle->h_transaction->t_journal->j_private;
95         err = handle->h_err;
96         rc = journal_stop(handle);
97
98         if (!err)
99                 err = rc;
100         if (err)
101                 __ext3_std_error(sb, where, err);
102         return err;
103 }
104
105 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
106                 struct buffer_head *bh, handle_t *handle, int err)
107 {
108         char nbuf[16];
109         const char *errstr = ext3_decode_error(NULL, err, nbuf);
110
111         if (bh)
112                 BUFFER_TRACE(bh, "abort");
113
114         if (!handle->h_err)
115                 handle->h_err = err;
116
117         if (is_handle_aborted(handle))
118                 return;
119
120         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
121                 caller, errstr, err_fn);
122
123         journal_abort_handle(handle);
124 }
125
126 void ext3_msg(struct super_block *sb, const char *prefix,
127                 const char *fmt, ...)
128 {
129         struct va_format vaf;
130         va_list args;
131
132         va_start(args, fmt);
133
134         vaf.fmt = fmt;
135         vaf.va = &args;
136
137         printk("%sEXT3-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
138
139         va_end(args);
140 }
141
142 /* Deal with the reporting of failure conditions on a filesystem such as
143  * inconsistencies detected or read IO failures.
144  *
145  * On ext2, we can store the error state of the filesystem in the
146  * superblock.  That is not possible on ext3, because we may have other
147  * write ordering constraints on the superblock which prevent us from
148  * writing it out straight away; and given that the journal is about to
149  * be aborted, we can't rely on the current, or future, transactions to
150  * write out the superblock safely.
151  *
152  * We'll just use the journal_abort() error code to record an error in
153  * the journal instead.  On recovery, the journal will complain about
154  * that error until we've noted it down and cleared it.
155  */
156
157 static void ext3_handle_error(struct super_block *sb)
158 {
159         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
160
161         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
162         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
163
164         if (sb->s_flags & MS_RDONLY)
165                 return;
166
167         if (!test_opt (sb, ERRORS_CONT)) {
168                 journal_t *journal = EXT3_SB(sb)->s_journal;
169
170                 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
171                 if (journal)
172                         journal_abort(journal, -EIO);
173         }
174         if (test_opt (sb, ERRORS_RO)) {
175                 ext3_msg(sb, KERN_CRIT,
176                         "error: remounting filesystem read-only");
177                 sb->s_flags |= MS_RDONLY;
178         }
179         ext3_commit_super(sb, es, 1);
180         if (test_opt(sb, ERRORS_PANIC))
181                 panic("EXT3-fs (%s): panic forced after error\n",
182                         sb->s_id);
183 }
184
185 void ext3_error(struct super_block *sb, const char *function,
186                 const char *fmt, ...)
187 {
188         struct va_format vaf;
189         va_list args;
190
191         va_start(args, fmt);
192
193         vaf.fmt = fmt;
194         vaf.va = &args;
195
196         printk(KERN_CRIT "EXT3-fs error (device %s): %s: %pV\n",
197                sb->s_id, function, &vaf);
198
199         va_end(args);
200
201         ext3_handle_error(sb);
202 }
203
204 static const char *ext3_decode_error(struct super_block * sb, int errno,
205                                      char nbuf[16])
206 {
207         char *errstr = NULL;
208
209         switch (errno) {
210         case -EIO:
211                 errstr = "IO failure";
212                 break;
213         case -ENOMEM:
214                 errstr = "Out of memory";
215                 break;
216         case -EROFS:
217                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
218                         errstr = "Journal has aborted";
219                 else
220                         errstr = "Readonly filesystem";
221                 break;
222         default:
223                 /* If the caller passed in an extra buffer for unknown
224                  * errors, textualise them now.  Else we just return
225                  * NULL. */
226                 if (nbuf) {
227                         /* Check for truncated error codes... */
228                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
229                                 errstr = nbuf;
230                 }
231                 break;
232         }
233
234         return errstr;
235 }
236
237 /* __ext3_std_error decodes expected errors from journaling functions
238  * automatically and invokes the appropriate error response.  */
239
240 void __ext3_std_error (struct super_block * sb, const char * function,
241                        int errno)
242 {
243         char nbuf[16];
244         const char *errstr;
245
246         /* Special case: if the error is EROFS, and we're not already
247          * inside a transaction, then there's really no point in logging
248          * an error. */
249         if (errno == -EROFS && journal_current_handle() == NULL &&
250             (sb->s_flags & MS_RDONLY))
251                 return;
252
253         errstr = ext3_decode_error(sb, errno, nbuf);
254         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
255
256         ext3_handle_error(sb);
257 }
258
259 /*
260  * ext3_abort is a much stronger failure handler than ext3_error.  The
261  * abort function may be used to deal with unrecoverable failures such
262  * as journal IO errors or ENOMEM at a critical moment in log management.
263  *
264  * We unconditionally force the filesystem into an ABORT|READONLY state,
265  * unless the error response on the fs has been set to panic in which
266  * case we take the easy way out and panic immediately.
267  */
268
269 void ext3_abort(struct super_block *sb, const char *function,
270                  const char *fmt, ...)
271 {
272         struct va_format vaf;
273         va_list args;
274
275         va_start(args, fmt);
276
277         vaf.fmt = fmt;
278         vaf.va = &args;
279
280         printk(KERN_CRIT "EXT3-fs (%s): error: %s: %pV\n",
281                sb->s_id, function, &vaf);
282
283         va_end(args);
284
285         if (test_opt(sb, ERRORS_PANIC))
286                 panic("EXT3-fs: panic from previous error\n");
287
288         if (sb->s_flags & MS_RDONLY)
289                 return;
290
291         ext3_msg(sb, KERN_CRIT,
292                 "error: remounting filesystem read-only");
293         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
294         sb->s_flags |= MS_RDONLY;
295         set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
296         if (EXT3_SB(sb)->s_journal)
297                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
298 }
299
300 void ext3_warning(struct super_block *sb, const char *function,
301                   const char *fmt, ...)
302 {
303         struct va_format vaf;
304         va_list args;
305
306         va_start(args, fmt);
307
308         vaf.fmt = fmt;
309         vaf.va = &args;
310
311         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: %pV\n",
312                sb->s_id, function, &vaf);
313
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 = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
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         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
368 }
369
370 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
371 {
372         struct block_device *bdev;
373         int ret = -ENODEV;
374
375         bdev = sbi->journal_bdev;
376         if (bdev) {
377                 ret = ext3_blkdev_put(bdev);
378                 sbi->journal_bdev = NULL;
379         }
380         return ret;
381 }
382
383 static inline struct inode *orphan_list_entry(struct list_head *l)
384 {
385         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
386 }
387
388 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
389 {
390         struct list_head *l;
391
392         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
393                le32_to_cpu(sbi->s_es->s_last_orphan));
394
395         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
396         list_for_each(l, &sbi->s_orphan) {
397                 struct inode *inode = orphan_list_entry(l);
398                 ext3_msg(sb, KERN_ERR, "  "
399                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
400                        inode->i_sb->s_id, inode->i_ino, inode,
401                        inode->i_mode, inode->i_nlink,
402                        NEXT_ORPHAN(inode));
403         }
404 }
405
406 static void ext3_put_super (struct super_block * sb)
407 {
408         struct ext3_sb_info *sbi = EXT3_SB(sb);
409         struct ext3_super_block *es = sbi->s_es;
410         int i, err;
411
412         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
413         ext3_xattr_put_super(sb);
414         err = journal_destroy(sbi->s_journal);
415         sbi->s_journal = NULL;
416         if (err < 0)
417                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
418
419         if (!(sb->s_flags & MS_RDONLY)) {
420                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
421                 es->s_state = cpu_to_le16(sbi->s_mount_state);
422                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
423                 mark_buffer_dirty(sbi->s_sbh);
424                 ext3_commit_super(sb, es, 1);
425         }
426
427         for (i = 0; i < sbi->s_gdb_count; i++)
428                 brelse(sbi->s_group_desc[i]);
429         kfree(sbi->s_group_desc);
430         percpu_counter_destroy(&sbi->s_freeblocks_counter);
431         percpu_counter_destroy(&sbi->s_freeinodes_counter);
432         percpu_counter_destroy(&sbi->s_dirs_counter);
433         brelse(sbi->s_sbh);
434 #ifdef CONFIG_QUOTA
435         for (i = 0; i < MAXQUOTAS; i++)
436                 kfree(sbi->s_qf_names[i]);
437 #endif
438
439         /* Debugging code just in case the in-memory inode orphan list
440          * isn't empty.  The on-disk one can be non-empty if we've
441          * detected an error and taken the fs readonly, but the
442          * in-memory list had better be clean by this point. */
443         if (!list_empty(&sbi->s_orphan))
444                 dump_orphan_list(sb, sbi);
445         J_ASSERT(list_empty(&sbi->s_orphan));
446
447         invalidate_bdev(sb->s_bdev);
448         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
449                 /*
450                  * Invalidate the journal device's buffers.  We don't want them
451                  * floating about in memory - the physical journal device may
452                  * hotswapped, and it breaks the `ro-after' testing code.
453                  */
454                 sync_blockdev(sbi->journal_bdev);
455                 invalidate_bdev(sbi->journal_bdev);
456                 ext3_blkdev_remove(sbi);
457         }
458         sb->s_fs_info = NULL;
459         kfree(sbi->s_blockgroup_lock);
460         kfree(sbi);
461 }
462
463 static struct kmem_cache *ext3_inode_cachep;
464
465 /*
466  * Called inside transaction, so use GFP_NOFS
467  */
468 static struct inode *ext3_alloc_inode(struct super_block *sb)
469 {
470         struct ext3_inode_info *ei;
471
472         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
473         if (!ei)
474                 return NULL;
475         ei->i_block_alloc_info = NULL;
476         ei->vfs_inode.i_version = 1;
477         atomic_set(&ei->i_datasync_tid, 0);
478         atomic_set(&ei->i_sync_tid, 0);
479         return &ei->vfs_inode;
480 }
481
482 static int ext3_drop_inode(struct inode *inode)
483 {
484         int drop = generic_drop_inode(inode);
485
486         trace_ext3_drop_inode(inode, drop);
487         return drop;
488 }
489
490 static void ext3_i_callback(struct rcu_head *head)
491 {
492         struct inode *inode = container_of(head, struct inode, i_rcu);
493         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
494 }
495
496 static void ext3_destroy_inode(struct inode *inode)
497 {
498         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
499                 printk("EXT3 Inode %p: orphan list check failed!\n",
500                         EXT3_I(inode));
501                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
502                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
503                                 false);
504                 dump_stack();
505         }
506         call_rcu(&inode->i_rcu, ext3_i_callback);
507 }
508
509 static void init_once(void *foo)
510 {
511         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
512
513         INIT_LIST_HEAD(&ei->i_orphan);
514 #ifdef CONFIG_EXT3_FS_XATTR
515         init_rwsem(&ei->xattr_sem);
516 #endif
517         mutex_init(&ei->truncate_mutex);
518         inode_init_once(&ei->vfs_inode);
519 }
520
521 static int init_inodecache(void)
522 {
523         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
524                                              sizeof(struct ext3_inode_info),
525                                              0, (SLAB_RECLAIM_ACCOUNT|
526                                                 SLAB_MEM_SPREAD),
527                                              init_once);
528         if (ext3_inode_cachep == NULL)
529                 return -ENOMEM;
530         return 0;
531 }
532
533 static void destroy_inodecache(void)
534 {
535         /*
536          * Make sure all delayed rcu free inodes are flushed before we
537          * destroy cache.
538          */
539         rcu_barrier();
540         kmem_cache_destroy(ext3_inode_cachep);
541 }
542
543 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
544 {
545 #if defined(CONFIG_QUOTA)
546         struct ext3_sb_info *sbi = EXT3_SB(sb);
547
548         if (sbi->s_jquota_fmt) {
549                 char *fmtname = "";
550
551                 switch (sbi->s_jquota_fmt) {
552                 case QFMT_VFS_OLD:
553                         fmtname = "vfsold";
554                         break;
555                 case QFMT_VFS_V0:
556                         fmtname = "vfsv0";
557                         break;
558                 case QFMT_VFS_V1:
559                         fmtname = "vfsv1";
560                         break;
561                 }
562                 seq_printf(seq, ",jqfmt=%s", fmtname);
563         }
564
565         if (sbi->s_qf_names[USRQUOTA])
566                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
567
568         if (sbi->s_qf_names[GRPQUOTA])
569                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
570
571         if (test_opt(sb, USRQUOTA))
572                 seq_puts(seq, ",usrquota");
573
574         if (test_opt(sb, GRPQUOTA))
575                 seq_puts(seq, ",grpquota");
576 #endif
577 }
578
579 static char *data_mode_string(unsigned long mode)
580 {
581         switch (mode) {
582         case EXT3_MOUNT_JOURNAL_DATA:
583                 return "journal";
584         case EXT3_MOUNT_ORDERED_DATA:
585                 return "ordered";
586         case EXT3_MOUNT_WRITEBACK_DATA:
587                 return "writeback";
588         }
589         return "unknown";
590 }
591
592 /*
593  * Show an option if
594  *  - it's set to a non-default value OR
595  *  - if the per-sb default is different from the global default
596  */
597 static int ext3_show_options(struct seq_file *seq, struct dentry *root)
598 {
599         struct super_block *sb = root->d_sb;
600         struct ext3_sb_info *sbi = EXT3_SB(sb);
601         struct ext3_super_block *es = sbi->s_es;
602         unsigned long def_mount_opts;
603
604         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
605
606         if (sbi->s_sb_block != 1)
607                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
608         if (test_opt(sb, MINIX_DF))
609                 seq_puts(seq, ",minixdf");
610         if (test_opt(sb, GRPID))
611                 seq_puts(seq, ",grpid");
612         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
613                 seq_puts(seq, ",nogrpid");
614         if (!uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT3_DEF_RESUID)) ||
615             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
616                 seq_printf(seq, ",resuid=%u",
617                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
618         }
619         if (!gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT3_DEF_RESGID)) ||
620             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
621                 seq_printf(seq, ",resgid=%u",
622                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
623         }
624         if (test_opt(sb, ERRORS_RO)) {
625                 int def_errors = le16_to_cpu(es->s_errors);
626
627                 if (def_errors == EXT3_ERRORS_PANIC ||
628                     def_errors == EXT3_ERRORS_CONTINUE) {
629                         seq_puts(seq, ",errors=remount-ro");
630                 }
631         }
632         if (test_opt(sb, ERRORS_CONT))
633                 seq_puts(seq, ",errors=continue");
634         if (test_opt(sb, ERRORS_PANIC))
635                 seq_puts(seq, ",errors=panic");
636         if (test_opt(sb, NO_UID32))
637                 seq_puts(seq, ",nouid32");
638         if (test_opt(sb, DEBUG))
639                 seq_puts(seq, ",debug");
640 #ifdef CONFIG_EXT3_FS_XATTR
641         if (test_opt(sb, XATTR_USER))
642                 seq_puts(seq, ",user_xattr");
643         if (!test_opt(sb, XATTR_USER) &&
644             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
645                 seq_puts(seq, ",nouser_xattr");
646         }
647 #endif
648 #ifdef CONFIG_EXT3_FS_POSIX_ACL
649         if (test_opt(sb, POSIX_ACL))
650                 seq_puts(seq, ",acl");
651         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
652                 seq_puts(seq, ",noacl");
653 #endif
654         if (!test_opt(sb, RESERVATION))
655                 seq_puts(seq, ",noreservation");
656         if (sbi->s_commit_interval) {
657                 seq_printf(seq, ",commit=%u",
658                            (unsigned) (sbi->s_commit_interval / HZ));
659         }
660
661         /*
662          * Always display barrier state so it's clear what the status is.
663          */
664         seq_puts(seq, ",barrier=");
665         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
666         seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
667         if (test_opt(sb, DATA_ERR_ABORT))
668                 seq_puts(seq, ",data_err=abort");
669
670         if (test_opt(sb, NOLOAD))
671                 seq_puts(seq, ",norecovery");
672
673         ext3_show_quota_options(seq, sb);
674
675         return 0;
676 }
677
678
679 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
680                 u64 ino, u32 generation)
681 {
682         struct inode *inode;
683
684         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
685                 return ERR_PTR(-ESTALE);
686         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
687                 return ERR_PTR(-ESTALE);
688
689         /* iget isn't really right if the inode is currently unallocated!!
690          *
691          * ext3_read_inode will return a bad_inode if the inode had been
692          * deleted, so we should be safe.
693          *
694          * Currently we don't know the generation for parent directory, so
695          * a generation of 0 means "accept any"
696          */
697         inode = ext3_iget(sb, ino);
698         if (IS_ERR(inode))
699                 return ERR_CAST(inode);
700         if (generation && inode->i_generation != generation) {
701                 iput(inode);
702                 return ERR_PTR(-ESTALE);
703         }
704
705         return inode;
706 }
707
708 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
709                 int fh_len, int fh_type)
710 {
711         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
712                                     ext3_nfs_get_inode);
713 }
714
715 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
716                 int fh_len, int fh_type)
717 {
718         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
719                                     ext3_nfs_get_inode);
720 }
721
722 /*
723  * Try to release metadata pages (indirect blocks, directories) which are
724  * mapped via the block device.  Since these pages could have journal heads
725  * which would prevent try_to_free_buffers() from freeing them, we must use
726  * jbd layer's try_to_free_buffers() function to release them.
727  */
728 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
729                                  gfp_t wait)
730 {
731         journal_t *journal = EXT3_SB(sb)->s_journal;
732
733         WARN_ON(PageChecked(page));
734         if (!page_has_buffers(page))
735                 return 0;
736         if (journal)
737                 return journal_try_to_free_buffers(journal, page, 
738                                                    wait & ~__GFP_WAIT);
739         return try_to_free_buffers(page);
740 }
741
742 #ifdef CONFIG_QUOTA
743 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
744 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
745
746 static int ext3_write_dquot(struct dquot *dquot);
747 static int ext3_acquire_dquot(struct dquot *dquot);
748 static int ext3_release_dquot(struct dquot *dquot);
749 static int ext3_mark_dquot_dirty(struct dquot *dquot);
750 static int ext3_write_info(struct super_block *sb, int type);
751 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
752                          struct path *path);
753 static int ext3_quota_on_mount(struct super_block *sb, int type);
754 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
755                                size_t len, loff_t off);
756 static ssize_t ext3_quota_write(struct super_block *sb, int type,
757                                 const char *data, size_t len, loff_t off);
758
759 static const struct dquot_operations ext3_quota_operations = {
760         .write_dquot    = ext3_write_dquot,
761         .acquire_dquot  = ext3_acquire_dquot,
762         .release_dquot  = ext3_release_dquot,
763         .mark_dirty     = ext3_mark_dquot_dirty,
764         .write_info     = ext3_write_info,
765         .alloc_dquot    = dquot_alloc,
766         .destroy_dquot  = dquot_destroy,
767 };
768
769 static const struct quotactl_ops ext3_qctl_operations = {
770         .quota_on       = ext3_quota_on,
771         .quota_off      = dquot_quota_off,
772         .quota_sync     = dquot_quota_sync,
773         .get_info       = dquot_get_dqinfo,
774         .set_info       = dquot_set_dqinfo,
775         .get_dqblk      = dquot_get_dqblk,
776         .set_dqblk      = dquot_set_dqblk
777 };
778 #endif
779
780 static const struct super_operations ext3_sops = {
781         .alloc_inode    = ext3_alloc_inode,
782         .destroy_inode  = ext3_destroy_inode,
783         .write_inode    = ext3_write_inode,
784         .dirty_inode    = ext3_dirty_inode,
785         .drop_inode     = ext3_drop_inode,
786         .evict_inode    = ext3_evict_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         .show_options   = ext3_show_options,
794 #ifdef CONFIG_QUOTA
795         .quota_read     = ext3_quota_read,
796         .quota_write    = ext3_quota_write,
797 #endif
798         .bdev_try_to_free_page = bdev_try_to_free_page,
799 };
800
801 static const struct export_operations ext3_export_ops = {
802         .fh_to_dentry = ext3_fh_to_dentry,
803         .fh_to_parent = ext3_fh_to_parent,
804         .get_parent = ext3_get_parent,
805 };
806
807 enum {
808         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
809         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
810         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
811         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
812         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
813         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
814         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
815         Opt_data_err_abort, Opt_data_err_ignore,
816         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
817         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
818         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
819         Opt_resize, Opt_usrquota, Opt_grpquota
820 };
821
822 static const match_table_t tokens = {
823         {Opt_bsd_df, "bsddf"},
824         {Opt_minix_df, "minixdf"},
825         {Opt_grpid, "grpid"},
826         {Opt_grpid, "bsdgroups"},
827         {Opt_nogrpid, "nogrpid"},
828         {Opt_nogrpid, "sysvgroups"},
829         {Opt_resgid, "resgid=%u"},
830         {Opt_resuid, "resuid=%u"},
831         {Opt_sb, "sb=%u"},
832         {Opt_err_cont, "errors=continue"},
833         {Opt_err_panic, "errors=panic"},
834         {Opt_err_ro, "errors=remount-ro"},
835         {Opt_nouid32, "nouid32"},
836         {Opt_nocheck, "nocheck"},
837         {Opt_nocheck, "check=none"},
838         {Opt_debug, "debug"},
839         {Opt_oldalloc, "oldalloc"},
840         {Opt_orlov, "orlov"},
841         {Opt_user_xattr, "user_xattr"},
842         {Opt_nouser_xattr, "nouser_xattr"},
843         {Opt_acl, "acl"},
844         {Opt_noacl, "noacl"},
845         {Opt_reservation, "reservation"},
846         {Opt_noreservation, "noreservation"},
847         {Opt_noload, "noload"},
848         {Opt_noload, "norecovery"},
849         {Opt_nobh, "nobh"},
850         {Opt_bh, "bh"},
851         {Opt_commit, "commit=%u"},
852         {Opt_journal_update, "journal=update"},
853         {Opt_journal_inum, "journal=%u"},
854         {Opt_journal_dev, "journal_dev=%u"},
855         {Opt_abort, "abort"},
856         {Opt_data_journal, "data=journal"},
857         {Opt_data_ordered, "data=ordered"},
858         {Opt_data_writeback, "data=writeback"},
859         {Opt_data_err_abort, "data_err=abort"},
860         {Opt_data_err_ignore, "data_err=ignore"},
861         {Opt_offusrjquota, "usrjquota="},
862         {Opt_usrjquota, "usrjquota=%s"},
863         {Opt_offgrpjquota, "grpjquota="},
864         {Opt_grpjquota, "grpjquota=%s"},
865         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
866         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
867         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
868         {Opt_grpquota, "grpquota"},
869         {Opt_noquota, "noquota"},
870         {Opt_quota, "quota"},
871         {Opt_usrquota, "usrquota"},
872         {Opt_barrier, "barrier=%u"},
873         {Opt_barrier, "barrier"},
874         {Opt_nobarrier, "nobarrier"},
875         {Opt_resize, "resize"},
876         {Opt_err, NULL},
877 };
878
879 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
880 {
881         ext3_fsblk_t    sb_block;
882         char            *options = (char *) *data;
883
884         if (!options || strncmp(options, "sb=", 3) != 0)
885                 return 1;       /* Default location */
886         options += 3;
887         /*todo: use simple_strtoll with >32bit ext3 */
888         sb_block = simple_strtoul(options, &options, 0);
889         if (*options && *options != ',') {
890                 ext3_msg(sb, "error: invalid sb specification: %s",
891                        (char *) *data);
892                 return 1;
893         }
894         if (*options == ',')
895                 options++;
896         *data = (void *) options;
897         return sb_block;
898 }
899
900 #ifdef CONFIG_QUOTA
901 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
902 {
903         struct ext3_sb_info *sbi = EXT3_SB(sb);
904         char *qname;
905
906         if (sb_any_quota_loaded(sb) &&
907                 !sbi->s_qf_names[qtype]) {
908                 ext3_msg(sb, KERN_ERR,
909                         "Cannot change journaled "
910                         "quota options when quota turned on");
911                 return 0;
912         }
913         qname = match_strdup(args);
914         if (!qname) {
915                 ext3_msg(sb, KERN_ERR,
916                         "Not enough memory for storing quotafile name");
917                 return 0;
918         }
919         if (sbi->s_qf_names[qtype]) {
920                 int same = !strcmp(sbi->s_qf_names[qtype], qname);
921
922                 kfree(qname);
923                 if (!same) {
924                         ext3_msg(sb, KERN_ERR,
925                                  "%s quota file already specified",
926                                  QTYPE2NAME(qtype));
927                 }
928                 return same;
929         }
930         if (strchr(qname, '/')) {
931                 ext3_msg(sb, KERN_ERR,
932                         "quotafile must be on filesystem root");
933                 kfree(qname);
934                 return 0;
935         }
936         sbi->s_qf_names[qtype] = qname;
937         set_opt(sbi->s_mount_opt, QUOTA);
938         return 1;
939 }
940
941 static int clear_qf_name(struct super_block *sb, int qtype) {
942
943         struct ext3_sb_info *sbi = EXT3_SB(sb);
944
945         if (sb_any_quota_loaded(sb) &&
946                 sbi->s_qf_names[qtype]) {
947                 ext3_msg(sb, KERN_ERR, "Cannot change journaled quota options"
948                         " when quota turned on");
949                 return 0;
950         }
951         if (sbi->s_qf_names[qtype]) {
952                 kfree(sbi->s_qf_names[qtype]);
953                 sbi->s_qf_names[qtype] = NULL;
954         }
955         return 1;
956 }
957 #endif
958
959 static int parse_options (char *options, struct super_block *sb,
960                           unsigned int *inum, unsigned long *journal_devnum,
961                           ext3_fsblk_t *n_blocks_count, int is_remount)
962 {
963         struct ext3_sb_info *sbi = EXT3_SB(sb);
964         char * p;
965         substring_t args[MAX_OPT_ARGS];
966         int data_opt = 0;
967         int option;
968         kuid_t uid;
969         kgid_t gid;
970 #ifdef CONFIG_QUOTA
971         int qfmt;
972 #endif
973
974         if (!options)
975                 return 1;
976
977         while ((p = strsep (&options, ",")) != NULL) {
978                 int token;
979                 if (!*p)
980                         continue;
981                 /*
982                  * Initialize args struct so we know whether arg was
983                  * found; some options take optional arguments.
984                  */
985                 args[0].to = args[0].from = NULL;
986                 token = match_token(p, tokens, args);
987                 switch (token) {
988                 case Opt_bsd_df:
989                         clear_opt (sbi->s_mount_opt, MINIX_DF);
990                         break;
991                 case Opt_minix_df:
992                         set_opt (sbi->s_mount_opt, MINIX_DF);
993                         break;
994                 case Opt_grpid:
995                         set_opt (sbi->s_mount_opt, GRPID);
996                         break;
997                 case Opt_nogrpid:
998                         clear_opt (sbi->s_mount_opt, GRPID);
999                         break;
1000                 case Opt_resuid:
1001                         if (match_int(&args[0], &option))
1002                                 return 0;
1003                         uid = make_kuid(current_user_ns(), option);
1004                         if (!uid_valid(uid)) {
1005                                 ext3_msg(sb, KERN_ERR, "Invalid uid value %d", option);
1006                                 return 0;
1007
1008                         }
1009                         sbi->s_resuid = uid;
1010                         break;
1011                 case Opt_resgid:
1012                         if (match_int(&args[0], &option))
1013                                 return 0;
1014                         gid = make_kgid(current_user_ns(), option);
1015                         if (!gid_valid(gid)) {
1016                                 ext3_msg(sb, KERN_ERR, "Invalid gid value %d", option);
1017                                 return 0;
1018                         }
1019                         sbi->s_resgid = gid;
1020                         break;
1021                 case Opt_sb:
1022                         /* handled by get_sb_block() instead of here */
1023                         /* *sb_block = match_int(&args[0]); */
1024                         break;
1025                 case Opt_err_panic:
1026                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1027                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1028                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1029                         break;
1030                 case Opt_err_ro:
1031                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1032                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1033                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1034                         break;
1035                 case Opt_err_cont:
1036                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1037                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1038                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1039                         break;
1040                 case Opt_nouid32:
1041                         set_opt (sbi->s_mount_opt, NO_UID32);
1042                         break;
1043                 case Opt_nocheck:
1044                         clear_opt (sbi->s_mount_opt, CHECK);
1045                         break;
1046                 case Opt_debug:
1047                         set_opt (sbi->s_mount_opt, DEBUG);
1048                         break;
1049                 case Opt_oldalloc:
1050                         ext3_msg(sb, KERN_WARNING,
1051                                 "Ignoring deprecated oldalloc option");
1052                         break;
1053                 case Opt_orlov:
1054                         ext3_msg(sb, KERN_WARNING,
1055                                 "Ignoring deprecated orlov option");
1056                         break;
1057 #ifdef CONFIG_EXT3_FS_XATTR
1058                 case Opt_user_xattr:
1059                         set_opt (sbi->s_mount_opt, XATTR_USER);
1060                         break;
1061                 case Opt_nouser_xattr:
1062                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1063                         break;
1064 #else
1065                 case Opt_user_xattr:
1066                 case Opt_nouser_xattr:
1067                         ext3_msg(sb, KERN_INFO,
1068                                 "(no)user_xattr options not supported");
1069                         break;
1070 #endif
1071 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1072                 case Opt_acl:
1073                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1074                         break;
1075                 case Opt_noacl:
1076                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1077                         break;
1078 #else
1079                 case Opt_acl:
1080                 case Opt_noacl:
1081                         ext3_msg(sb, KERN_INFO,
1082                                 "(no)acl options not supported");
1083                         break;
1084 #endif
1085                 case Opt_reservation:
1086                         set_opt(sbi->s_mount_opt, RESERVATION);
1087                         break;
1088                 case Opt_noreservation:
1089                         clear_opt(sbi->s_mount_opt, RESERVATION);
1090                         break;
1091                 case Opt_journal_update:
1092                         /* @@@ FIXME */
1093                         /* Eventually we will want to be able to create
1094                            a journal file here.  For now, only allow the
1095                            user to specify an existing inode to be the
1096                            journal file. */
1097                         if (is_remount) {
1098                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1099                                         "journal on remount");
1100                                 return 0;
1101                         }
1102                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1103                         break;
1104                 case Opt_journal_inum:
1105                         if (is_remount) {
1106                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1107                                        "journal on remount");
1108                                 return 0;
1109                         }
1110                         if (match_int(&args[0], &option))
1111                                 return 0;
1112                         *inum = option;
1113                         break;
1114                 case Opt_journal_dev:
1115                         if (is_remount) {
1116                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1117                                        "journal on remount");
1118                                 return 0;
1119                         }
1120                         if (match_int(&args[0], &option))
1121                                 return 0;
1122                         *journal_devnum = option;
1123                         break;
1124                 case Opt_noload:
1125                         set_opt (sbi->s_mount_opt, NOLOAD);
1126                         break;
1127                 case Opt_commit:
1128                         if (match_int(&args[0], &option))
1129                                 return 0;
1130                         if (option < 0)
1131                                 return 0;
1132                         if (option == 0)
1133                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1134                         sbi->s_commit_interval = HZ * option;
1135                         break;
1136                 case Opt_data_journal:
1137                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1138                         goto datacheck;
1139                 case Opt_data_ordered:
1140                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1141                         goto datacheck;
1142                 case Opt_data_writeback:
1143                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1144                 datacheck:
1145                         if (is_remount) {
1146                                 if (test_opt(sb, DATA_FLAGS) == data_opt)
1147                                         break;
1148                                 ext3_msg(sb, KERN_ERR,
1149                                         "error: cannot change "
1150                                         "data mode on remount. The filesystem "
1151                                         "is mounted in data=%s mode and you "
1152                                         "try to remount it in data=%s mode.",
1153                                         data_mode_string(test_opt(sb,
1154                                                         DATA_FLAGS)),
1155                                         data_mode_string(data_opt));
1156                                 return 0;
1157                         } else {
1158                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1159                                 sbi->s_mount_opt |= data_opt;
1160                         }
1161                         break;
1162                 case Opt_data_err_abort:
1163                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1164                         break;
1165                 case Opt_data_err_ignore:
1166                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1167                         break;
1168 #ifdef CONFIG_QUOTA
1169                 case Opt_usrjquota:
1170                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1171                                 return 0;
1172                         break;
1173                 case Opt_grpjquota:
1174                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1175                                 return 0;
1176                         break;
1177                 case Opt_offusrjquota:
1178                         if (!clear_qf_name(sb, USRQUOTA))
1179                                 return 0;
1180                         break;
1181                 case Opt_offgrpjquota:
1182                         if (!clear_qf_name(sb, GRPQUOTA))
1183                                 return 0;
1184                         break;
1185                 case Opt_jqfmt_vfsold:
1186                         qfmt = QFMT_VFS_OLD;
1187                         goto set_qf_format;
1188                 case Opt_jqfmt_vfsv0:
1189                         qfmt = QFMT_VFS_V0;
1190                         goto set_qf_format;
1191                 case Opt_jqfmt_vfsv1:
1192                         qfmt = QFMT_VFS_V1;
1193 set_qf_format:
1194                         if (sb_any_quota_loaded(sb) &&
1195                             sbi->s_jquota_fmt != qfmt) {
1196                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1197                                         "journaled quota options when "
1198                                         "quota turned on.");
1199                                 return 0;
1200                         }
1201                         sbi->s_jquota_fmt = qfmt;
1202                         break;
1203                 case Opt_quota:
1204                 case Opt_usrquota:
1205                         set_opt(sbi->s_mount_opt, QUOTA);
1206                         set_opt(sbi->s_mount_opt, USRQUOTA);
1207                         break;
1208                 case Opt_grpquota:
1209                         set_opt(sbi->s_mount_opt, QUOTA);
1210                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1211                         break;
1212                 case Opt_noquota:
1213                         if (sb_any_quota_loaded(sb)) {
1214                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1215                                         "quota options when quota turned on.");
1216                                 return 0;
1217                         }
1218                         clear_opt(sbi->s_mount_opt, QUOTA);
1219                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1220                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1221                         break;
1222 #else
1223                 case Opt_quota:
1224                 case Opt_usrquota:
1225                 case Opt_grpquota:
1226                         ext3_msg(sb, KERN_ERR,
1227                                 "error: quota options not supported.");
1228                         break;
1229                 case Opt_usrjquota:
1230                 case Opt_grpjquota:
1231                 case Opt_offusrjquota:
1232                 case Opt_offgrpjquota:
1233                 case Opt_jqfmt_vfsold:
1234                 case Opt_jqfmt_vfsv0:
1235                 case Opt_jqfmt_vfsv1:
1236                         ext3_msg(sb, KERN_ERR,
1237                                 "error: journaled quota options not "
1238                                 "supported.");
1239                         break;
1240                 case Opt_noquota:
1241                         break;
1242 #endif
1243                 case Opt_abort:
1244                         set_opt(sbi->s_mount_opt, ABORT);
1245                         break;
1246                 case Opt_nobarrier:
1247                         clear_opt(sbi->s_mount_opt, BARRIER);
1248                         break;
1249                 case Opt_barrier:
1250                         if (args[0].from) {
1251                                 if (match_int(&args[0], &option))
1252                                         return 0;
1253                         } else
1254                                 option = 1;     /* No argument, default to 1 */
1255                         if (option)
1256                                 set_opt(sbi->s_mount_opt, BARRIER);
1257                         else
1258                                 clear_opt(sbi->s_mount_opt, BARRIER);
1259                         break;
1260                 case Opt_ignore:
1261                         break;
1262                 case Opt_resize:
1263                         if (!is_remount) {
1264                                 ext3_msg(sb, KERN_ERR,
1265                                         "error: resize option only available "
1266                                         "for remount");
1267                                 return 0;
1268                         }
1269                         if (match_int(&args[0], &option) != 0)
1270                                 return 0;
1271                         *n_blocks_count = option;
1272                         break;
1273                 case Opt_nobh:
1274                         ext3_msg(sb, KERN_WARNING,
1275                                 "warning: ignoring deprecated nobh option");
1276                         break;
1277                 case Opt_bh:
1278                         ext3_msg(sb, KERN_WARNING,
1279                                 "warning: ignoring deprecated bh option");
1280                         break;
1281                 default:
1282                         ext3_msg(sb, KERN_ERR,
1283                                 "error: unrecognized mount option \"%s\" "
1284                                 "or missing value", p);
1285                         return 0;
1286                 }
1287         }
1288 #ifdef CONFIG_QUOTA
1289         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1290                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1291                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1292                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1293                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1294
1295                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1296                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1297                                         "format mixing.");
1298                         return 0;
1299                 }
1300
1301                 if (!sbi->s_jquota_fmt) {
1302                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1303                                         "not specified.");
1304                         return 0;
1305                 }
1306         } else {
1307                 if (sbi->s_jquota_fmt) {
1308                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1309                                         "specified with no journaling "
1310                                         "enabled.");
1311                         return 0;
1312                 }
1313         }
1314 #endif
1315         return 1;
1316 }
1317
1318 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1319                             int read_only)
1320 {
1321         struct ext3_sb_info *sbi = EXT3_SB(sb);
1322         int res = 0;
1323
1324         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1325                 ext3_msg(sb, KERN_ERR,
1326                         "error: revision level too high, "
1327                         "forcing read-only mode");
1328                 res = MS_RDONLY;
1329         }
1330         if (read_only)
1331                 return res;
1332         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1333                 ext3_msg(sb, KERN_WARNING,
1334                         "warning: mounting unchecked fs, "
1335                         "running e2fsck is recommended");
1336         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1337                 ext3_msg(sb, KERN_WARNING,
1338                         "warning: mounting fs with errors, "
1339                         "running e2fsck is recommended");
1340         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1341                  le16_to_cpu(es->s_mnt_count) >=
1342                         le16_to_cpu(es->s_max_mnt_count))
1343                 ext3_msg(sb, KERN_WARNING,
1344                         "warning: maximal mount count reached, "
1345                         "running e2fsck is recommended");
1346         else if (le32_to_cpu(es->s_checkinterval) &&
1347                 (le32_to_cpu(es->s_lastcheck) +
1348                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1349                 ext3_msg(sb, KERN_WARNING,
1350                         "warning: checktime reached, "
1351                         "running e2fsck is recommended");
1352 #if 0
1353                 /* @@@ We _will_ want to clear the valid bit if we find
1354                    inconsistencies, to force a fsck at reboot.  But for
1355                    a plain journaled filesystem we can keep it set as
1356                    valid forever! :) */
1357         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1358 #endif
1359         if (!le16_to_cpu(es->s_max_mnt_count))
1360                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1361         le16_add_cpu(&es->s_mnt_count, 1);
1362         es->s_mtime = cpu_to_le32(get_seconds());
1363         ext3_update_dynamic_rev(sb);
1364         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1365
1366         ext3_commit_super(sb, es, 1);
1367         if (test_opt(sb, DEBUG))
1368                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1369                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1370                         sb->s_blocksize,
1371                         sbi->s_groups_count,
1372                         EXT3_BLOCKS_PER_GROUP(sb),
1373                         EXT3_INODES_PER_GROUP(sb),
1374                         sbi->s_mount_opt);
1375
1376         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1377                 char b[BDEVNAME_SIZE];
1378                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1379                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1380         } else {
1381                 ext3_msg(sb, KERN_INFO, "using internal journal");
1382         }
1383         cleancache_init_fs(sb);
1384         return res;
1385 }
1386
1387 /* Called at mount-time, super-block is locked */
1388 static int ext3_check_descriptors(struct super_block *sb)
1389 {
1390         struct ext3_sb_info *sbi = EXT3_SB(sb);
1391         int i;
1392
1393         ext3_debug ("Checking group descriptors");
1394
1395         for (i = 0; i < sbi->s_groups_count; i++) {
1396                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1397                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1398                 ext3_fsblk_t last_block;
1399
1400                 if (i == sbi->s_groups_count - 1)
1401                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1402                 else
1403                         last_block = first_block +
1404                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1405
1406                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1407                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1408                 {
1409                         ext3_error (sb, "ext3_check_descriptors",
1410                                     "Block bitmap for group %d"
1411                                     " not in group (block %lu)!",
1412                                     i, (unsigned long)
1413                                         le32_to_cpu(gdp->bg_block_bitmap));
1414                         return 0;
1415                 }
1416                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1417                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1418                 {
1419                         ext3_error (sb, "ext3_check_descriptors",
1420                                     "Inode bitmap for group %d"
1421                                     " not in group (block %lu)!",
1422                                     i, (unsigned long)
1423                                         le32_to_cpu(gdp->bg_inode_bitmap));
1424                         return 0;
1425                 }
1426                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1427                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1428                     last_block)
1429                 {
1430                         ext3_error (sb, "ext3_check_descriptors",
1431                                     "Inode table for group %d"
1432                                     " not in group (block %lu)!",
1433                                     i, (unsigned long)
1434                                         le32_to_cpu(gdp->bg_inode_table));
1435                         return 0;
1436                 }
1437         }
1438
1439         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1440         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1441         return 1;
1442 }
1443
1444
1445 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1446  * the superblock) which were deleted from all directories, but held open by
1447  * a process at the time of a crash.  We walk the list and try to delete these
1448  * inodes at recovery time (only with a read-write filesystem).
1449  *
1450  * In order to keep the orphan inode chain consistent during traversal (in
1451  * case of crash during recovery), we link each inode into the superblock
1452  * orphan list_head and handle it the same way as an inode deletion during
1453  * normal operation (which journals the operations for us).
1454  *
1455  * We only do an iget() and an iput() on each inode, which is very safe if we
1456  * accidentally point at an in-use or already deleted inode.  The worst that
1457  * can happen in this case is that we get a "bit already cleared" message from
1458  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1459  * e2fsck was run on this filesystem, and it must have already done the orphan
1460  * inode cleanup for us, so we can safely abort without any further action.
1461  */
1462 static void ext3_orphan_cleanup (struct super_block * sb,
1463                                  struct ext3_super_block * es)
1464 {
1465         unsigned int s_flags = sb->s_flags;
1466         int nr_orphans = 0, nr_truncates = 0;
1467 #ifdef CONFIG_QUOTA
1468         int i;
1469 #endif
1470         if (!es->s_last_orphan) {
1471                 jbd_debug(4, "no orphan inodes to clean up\n");
1472                 return;
1473         }
1474
1475         if (bdev_read_only(sb->s_bdev)) {
1476                 ext3_msg(sb, KERN_ERR, "error: write access "
1477                         "unavailable, skipping orphan cleanup.");
1478                 return;
1479         }
1480
1481         /* Check if feature set allows readwrite operations */
1482         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP)) {
1483                 ext3_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
1484                          "unknown ROCOMPAT features");
1485                 return;
1486         }
1487
1488         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1489                 /* don't clear list on RO mount w/ errors */
1490                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
1491                         jbd_debug(1, "Errors on filesystem, "
1492                                   "clearing orphan list.\n");
1493                         es->s_last_orphan = 0;
1494                 }
1495                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1496                 return;
1497         }
1498
1499         if (s_flags & MS_RDONLY) {
1500                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1501                 sb->s_flags &= ~MS_RDONLY;
1502         }
1503 #ifdef CONFIG_QUOTA
1504         /* Needed for iput() to work correctly and not trash data */
1505         sb->s_flags |= MS_ACTIVE;
1506         /* Turn on quotas so that they are updated correctly */
1507         for (i = 0; i < MAXQUOTAS; i++) {
1508                 if (EXT3_SB(sb)->s_qf_names[i]) {
1509                         int ret = ext3_quota_on_mount(sb, i);
1510                         if (ret < 0)
1511                                 ext3_msg(sb, KERN_ERR,
1512                                         "error: cannot turn on journaled "
1513                                         "quota: %d", ret);
1514                 }
1515         }
1516 #endif
1517
1518         while (es->s_last_orphan) {
1519                 struct inode *inode;
1520
1521                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1522                 if (IS_ERR(inode)) {
1523                         es->s_last_orphan = 0;
1524                         break;
1525                 }
1526
1527                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1528                 dquot_initialize(inode);
1529                 if (inode->i_nlink) {
1530                         printk(KERN_DEBUG
1531                                 "%s: truncating inode %lu to %Ld bytes\n",
1532                                 __func__, inode->i_ino, inode->i_size);
1533                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1534                                   inode->i_ino, inode->i_size);
1535                         ext3_truncate(inode);
1536                         nr_truncates++;
1537                 } else {
1538                         printk(KERN_DEBUG
1539                                 "%s: deleting unreferenced inode %lu\n",
1540                                 __func__, inode->i_ino);
1541                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1542                                   inode->i_ino);
1543                         nr_orphans++;
1544                 }
1545                 iput(inode);  /* The delete magic happens here! */
1546         }
1547
1548 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1549
1550         if (nr_orphans)
1551                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1552                        PLURAL(nr_orphans));
1553         if (nr_truncates)
1554                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1555                        PLURAL(nr_truncates));
1556 #ifdef CONFIG_QUOTA
1557         /* Turn quotas off */
1558         for (i = 0; i < MAXQUOTAS; i++) {
1559                 if (sb_dqopt(sb)->files[i])
1560                         dquot_quota_off(sb, i);
1561         }
1562 #endif
1563         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1564 }
1565
1566 /*
1567  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1568  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1569  * We need to be 1 filesystem block less than the 2^32 sector limit.
1570  */
1571 static loff_t ext3_max_size(int bits)
1572 {
1573         loff_t res = EXT3_NDIR_BLOCKS;
1574         int meta_blocks;
1575         loff_t upper_limit;
1576
1577         /* This is calculated to be the largest file size for a
1578          * dense, file such that the total number of
1579          * sectors in the file, including data and all indirect blocks,
1580          * does not exceed 2^32 -1
1581          * __u32 i_blocks representing the total number of
1582          * 512 bytes blocks of the file
1583          */
1584         upper_limit = (1LL << 32) - 1;
1585
1586         /* total blocks in file system block size */
1587         upper_limit >>= (bits - 9);
1588
1589
1590         /* indirect blocks */
1591         meta_blocks = 1;
1592         /* double indirect blocks */
1593         meta_blocks += 1 + (1LL << (bits-2));
1594         /* tripple indirect blocks */
1595         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1596
1597         upper_limit -= meta_blocks;
1598         upper_limit <<= bits;
1599
1600         res += 1LL << (bits-2);
1601         res += 1LL << (2*(bits-2));
1602         res += 1LL << (3*(bits-2));
1603         res <<= bits;
1604         if (res > upper_limit)
1605                 res = upper_limit;
1606
1607         if (res > MAX_LFS_FILESIZE)
1608                 res = MAX_LFS_FILESIZE;
1609
1610         return res;
1611 }
1612
1613 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1614                                     ext3_fsblk_t logic_sb_block,
1615                                     int nr)
1616 {
1617         struct ext3_sb_info *sbi = EXT3_SB(sb);
1618         unsigned long bg, first_meta_bg;
1619         int has_super = 0;
1620
1621         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1622
1623         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1624             nr < first_meta_bg)
1625                 return (logic_sb_block + nr + 1);
1626         bg = sbi->s_desc_per_block * nr;
1627         if (ext3_bg_has_super(sb, bg))
1628                 has_super = 1;
1629         return (has_super + ext3_group_first_block_no(sb, bg));
1630 }
1631
1632
1633 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1634 {
1635         struct buffer_head * bh;
1636         struct ext3_super_block *es = NULL;
1637         struct ext3_sb_info *sbi;
1638         ext3_fsblk_t block;
1639         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1640         ext3_fsblk_t logic_sb_block;
1641         unsigned long offset = 0;
1642         unsigned int journal_inum = 0;
1643         unsigned long journal_devnum = 0;
1644         unsigned long def_mount_opts;
1645         struct inode *root;
1646         int blocksize;
1647         int hblock;
1648         int db_count;
1649         int i;
1650         int needs_recovery;
1651         int ret = -EINVAL;
1652         __le32 features;
1653         int err;
1654
1655         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1656         if (!sbi)
1657                 return -ENOMEM;
1658
1659         sbi->s_blockgroup_lock =
1660                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1661         if (!sbi->s_blockgroup_lock) {
1662                 kfree(sbi);
1663                 return -ENOMEM;
1664         }
1665         sb->s_fs_info = sbi;
1666         sbi->s_sb_block = sb_block;
1667
1668         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1669         if (!blocksize) {
1670                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1671                 goto out_fail;
1672         }
1673
1674         /*
1675          * The ext3 superblock will not be buffer aligned for other than 1kB
1676          * block sizes.  We need to calculate the offset from buffer start.
1677          */
1678         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1679                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1680                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1681         } else {
1682                 logic_sb_block = sb_block;
1683         }
1684
1685         if (!(bh = sb_bread(sb, logic_sb_block))) {
1686                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1687                 goto out_fail;
1688         }
1689         /*
1690          * Note: s_es must be initialized as soon as possible because
1691          *       some ext3 macro-instructions depend on its value
1692          */
1693         es = (struct ext3_super_block *) (bh->b_data + offset);
1694         sbi->s_es = es;
1695         sb->s_magic = le16_to_cpu(es->s_magic);
1696         if (sb->s_magic != EXT3_SUPER_MAGIC)
1697                 goto cantfind_ext3;
1698
1699         /* Set defaults before we parse the mount options */
1700         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1701         if (def_mount_opts & EXT3_DEFM_DEBUG)
1702                 set_opt(sbi->s_mount_opt, DEBUG);
1703         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1704                 set_opt(sbi->s_mount_opt, GRPID);
1705         if (def_mount_opts & EXT3_DEFM_UID16)
1706                 set_opt(sbi->s_mount_opt, NO_UID32);
1707 #ifdef CONFIG_EXT3_FS_XATTR
1708         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1709                 set_opt(sbi->s_mount_opt, XATTR_USER);
1710 #endif
1711 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1712         if (def_mount_opts & EXT3_DEFM_ACL)
1713                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1714 #endif
1715         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1716                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1717         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1718                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1719         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1720                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1721
1722         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1723                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1724         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1725                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1726         else
1727                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1728
1729         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
1730         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
1731
1732         /* enable barriers by default */
1733         set_opt(sbi->s_mount_opt, BARRIER);
1734         set_opt(sbi->s_mount_opt, RESERVATION);
1735
1736         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1737                             NULL, 0))
1738                 goto failed_mount;
1739
1740         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1741                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1742
1743         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1744             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1745              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1746              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1747                 ext3_msg(sb, KERN_WARNING,
1748                         "warning: feature flags set on rev 0 fs, "
1749                         "running e2fsck is recommended");
1750         /*
1751          * Check feature flags regardless of the revision level, since we
1752          * previously didn't change the revision level when setting the flags,
1753          * so there is a chance incompat flags are set on a rev 0 filesystem.
1754          */
1755         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1756         if (features) {
1757                 ext3_msg(sb, KERN_ERR,
1758                         "error: couldn't mount because of unsupported "
1759                         "optional features (%x)", le32_to_cpu(features));
1760                 goto failed_mount;
1761         }
1762         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1763         if (!(sb->s_flags & MS_RDONLY) && features) {
1764                 ext3_msg(sb, KERN_ERR,
1765                         "error: couldn't mount RDWR because of unsupported "
1766                         "optional features (%x)", le32_to_cpu(features));
1767                 goto failed_mount;
1768         }
1769         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1770
1771         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1772             blocksize > EXT3_MAX_BLOCK_SIZE) {
1773                 ext3_msg(sb, KERN_ERR,
1774                         "error: couldn't mount because of unsupported "
1775                         "filesystem blocksize %d", blocksize);
1776                 goto failed_mount;
1777         }
1778
1779         hblock = bdev_logical_block_size(sb->s_bdev);
1780         if (sb->s_blocksize != blocksize) {
1781                 /*
1782                  * Make sure the blocksize for the filesystem is larger
1783                  * than the hardware sectorsize for the machine.
1784                  */
1785                 if (blocksize < hblock) {
1786                         ext3_msg(sb, KERN_ERR,
1787                                 "error: fsblocksize %d too small for "
1788                                 "hardware sectorsize %d", blocksize, hblock);
1789                         goto failed_mount;
1790                 }
1791
1792                 brelse (bh);
1793                 if (!sb_set_blocksize(sb, blocksize)) {
1794                         ext3_msg(sb, KERN_ERR,
1795                                 "error: bad blocksize %d", blocksize);
1796                         goto out_fail;
1797                 }
1798                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1799                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1800                 bh = sb_bread(sb, logic_sb_block);
1801                 if (!bh) {
1802                         ext3_msg(sb, KERN_ERR,
1803                                "error: can't read superblock on 2nd try");
1804                         goto failed_mount;
1805                 }
1806                 es = (struct ext3_super_block *)(bh->b_data + offset);
1807                 sbi->s_es = es;
1808                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1809                         ext3_msg(sb, KERN_ERR,
1810                                 "error: magic mismatch");
1811                         goto failed_mount;
1812                 }
1813         }
1814
1815         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1816
1817         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1818                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1819                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1820         } else {
1821                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1822                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1823                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1824                     (!is_power_of_2(sbi->s_inode_size)) ||
1825                     (sbi->s_inode_size > blocksize)) {
1826                         ext3_msg(sb, KERN_ERR,
1827                                 "error: unsupported inode size: %d",
1828                                 sbi->s_inode_size);
1829                         goto failed_mount;
1830                 }
1831         }
1832         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1833                                    le32_to_cpu(es->s_log_frag_size);
1834         if (blocksize != sbi->s_frag_size) {
1835                 ext3_msg(sb, KERN_ERR,
1836                        "error: fragsize %lu != blocksize %u (unsupported)",
1837                        sbi->s_frag_size, blocksize);
1838                 goto failed_mount;
1839         }
1840         sbi->s_frags_per_block = 1;
1841         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1842         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1843         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1844         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1845                 goto cantfind_ext3;
1846         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1847         if (sbi->s_inodes_per_block == 0)
1848                 goto cantfind_ext3;
1849         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1850                                         sbi->s_inodes_per_block;
1851         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1852         sbi->s_sbh = bh;
1853         sbi->s_mount_state = le16_to_cpu(es->s_state);
1854         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1855         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1856         for (i=0; i < 4; i++)
1857                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1858         sbi->s_def_hash_version = es->s_def_hash_version;
1859         i = le32_to_cpu(es->s_flags);
1860         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1861                 sbi->s_hash_unsigned = 3;
1862         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1863 #ifdef __CHAR_UNSIGNED__
1864                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1865                 sbi->s_hash_unsigned = 3;
1866 #else
1867                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1868 #endif
1869         }
1870
1871         if (sbi->s_blocks_per_group > blocksize * 8) {
1872                 ext3_msg(sb, KERN_ERR,
1873                         "#blocks per group too big: %lu",
1874                         sbi->s_blocks_per_group);
1875                 goto failed_mount;
1876         }
1877         if (sbi->s_frags_per_group > blocksize * 8) {
1878                 ext3_msg(sb, KERN_ERR,
1879                         "error: #fragments per group too big: %lu",
1880                         sbi->s_frags_per_group);
1881                 goto failed_mount;
1882         }
1883         if (sbi->s_inodes_per_group > blocksize * 8) {
1884                 ext3_msg(sb, KERN_ERR,
1885                         "error: #inodes per group too big: %lu",
1886                         sbi->s_inodes_per_group);
1887                 goto failed_mount;
1888         }
1889
1890         err = generic_check_addressable(sb->s_blocksize_bits,
1891                                         le32_to_cpu(es->s_blocks_count));
1892         if (err) {
1893                 ext3_msg(sb, KERN_ERR,
1894                         "error: filesystem is too large to mount safely");
1895                 if (sizeof(sector_t) < 8)
1896                         ext3_msg(sb, KERN_ERR,
1897                                 "error: CONFIG_LBDAF not enabled");
1898                 ret = err;
1899                 goto failed_mount;
1900         }
1901
1902         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1903                 goto cantfind_ext3;
1904         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1905                                le32_to_cpu(es->s_first_data_block) - 1)
1906                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1907         db_count = DIV_ROUND_UP(sbi->s_groups_count, EXT3_DESC_PER_BLOCK(sb));
1908         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1909                                     GFP_KERNEL);
1910         if (sbi->s_group_desc == NULL) {
1911                 ext3_msg(sb, KERN_ERR,
1912                         "error: not enough memory");
1913                 ret = -ENOMEM;
1914                 goto failed_mount;
1915         }
1916
1917         bgl_lock_init(sbi->s_blockgroup_lock);
1918
1919         for (i = 0; i < db_count; i++) {
1920                 block = descriptor_loc(sb, logic_sb_block, i);
1921                 sbi->s_group_desc[i] = sb_bread(sb, block);
1922                 if (!sbi->s_group_desc[i]) {
1923                         ext3_msg(sb, KERN_ERR,
1924                                 "error: can't read group descriptor %d", i);
1925                         db_count = i;
1926                         goto failed_mount2;
1927                 }
1928         }
1929         if (!ext3_check_descriptors (sb)) {
1930                 ext3_msg(sb, KERN_ERR,
1931                         "error: group descriptors corrupted");
1932                 goto failed_mount2;
1933         }
1934         sbi->s_gdb_count = db_count;
1935         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1936         spin_lock_init(&sbi->s_next_gen_lock);
1937
1938         /* per fileystem reservation list head & lock */
1939         spin_lock_init(&sbi->s_rsv_window_lock);
1940         sbi->s_rsv_window_root = RB_ROOT;
1941         /* Add a single, static dummy reservation to the start of the
1942          * reservation window list --- it gives us a placeholder for
1943          * append-at-start-of-list which makes the allocation logic
1944          * _much_ simpler. */
1945         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1946         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1947         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1948         sbi->s_rsv_window_head.rsv_goal_size = 0;
1949         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1950
1951         /*
1952          * set up enough so that it can read an inode
1953          */
1954         sb->s_op = &ext3_sops;
1955         sb->s_export_op = &ext3_export_ops;
1956         sb->s_xattr = ext3_xattr_handlers;
1957 #ifdef CONFIG_QUOTA
1958         sb->s_qcop = &ext3_qctl_operations;
1959         sb->dq_op = &ext3_quota_operations;
1960 #endif
1961         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
1962         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1963         mutex_init(&sbi->s_orphan_lock);
1964         mutex_init(&sbi->s_resize_lock);
1965
1966         sb->s_root = NULL;
1967
1968         needs_recovery = (es->s_last_orphan != 0 ||
1969                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1970                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1971
1972         /*
1973          * The first inode we look at is the journal inode.  Don't try
1974          * root first: it may be modified in the journal!
1975          */
1976         if (!test_opt(sb, NOLOAD) &&
1977             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1978                 if (ext3_load_journal(sb, es, journal_devnum))
1979                         goto failed_mount2;
1980         } else if (journal_inum) {
1981                 if (ext3_create_journal(sb, es, journal_inum))
1982                         goto failed_mount2;
1983         } else {
1984                 if (!silent)
1985                         ext3_msg(sb, KERN_ERR,
1986                                 "error: no journal found. "
1987                                 "mounting ext3 over ext2?");
1988                 goto failed_mount2;
1989         }
1990         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1991                         ext3_count_free_blocks(sb));
1992         if (!err) {
1993                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1994                                 ext3_count_free_inodes(sb));
1995         }
1996         if (!err) {
1997                 err = percpu_counter_init(&sbi->s_dirs_counter,
1998                                 ext3_count_dirs(sb));
1999         }
2000         if (err) {
2001                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
2002                 ret = err;
2003                 goto failed_mount3;
2004         }
2005
2006         /* We have now updated the journal if required, so we can
2007          * validate the data journaling mode. */
2008         switch (test_opt(sb, DATA_FLAGS)) {
2009         case 0:
2010                 /* No mode set, assume a default based on the journal
2011                    capabilities: ORDERED_DATA if the journal can
2012                    cope, else JOURNAL_DATA */
2013                 if (journal_check_available_features
2014                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
2015                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
2016                 else
2017                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2018                 break;
2019
2020         case EXT3_MOUNT_ORDERED_DATA:
2021         case EXT3_MOUNT_WRITEBACK_DATA:
2022                 if (!journal_check_available_features
2023                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
2024                         ext3_msg(sb, KERN_ERR,
2025                                 "error: journal does not support "
2026                                 "requested data journaling mode");
2027                         goto failed_mount3;
2028                 }
2029         default:
2030                 break;
2031         }
2032
2033         /*
2034          * The journal_load will have done any necessary log recovery,
2035          * so we can safely mount the rest of the filesystem now.
2036          */
2037
2038         root = ext3_iget(sb, EXT3_ROOT_INO);
2039         if (IS_ERR(root)) {
2040                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
2041                 ret = PTR_ERR(root);
2042                 goto failed_mount3;
2043         }
2044         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2045                 iput(root);
2046                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2047                 goto failed_mount3;
2048         }
2049         sb->s_root = d_make_root(root);
2050         if (!sb->s_root) {
2051                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2052                 ret = -ENOMEM;
2053                 goto failed_mount3;
2054         }
2055
2056         if (ext3_setup_super(sb, es, sb->s_flags & MS_RDONLY))
2057                 sb->s_flags |= MS_RDONLY;
2058
2059         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2060         ext3_orphan_cleanup(sb, es);
2061         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2062         if (needs_recovery) {
2063                 ext3_mark_recovery_complete(sb, es);
2064                 ext3_msg(sb, KERN_INFO, "recovery complete");
2065         }
2066         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2067                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2068                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2069                 "writeback");
2070         sb->s_flags |= MS_SNAP_STABLE;
2071
2072         return 0;
2073
2074 cantfind_ext3:
2075         if (!silent)
2076                 ext3_msg(sb, KERN_INFO,
2077                         "error: can't find ext3 filesystem on dev %s.",
2078                        sb->s_id);
2079         goto failed_mount;
2080
2081 failed_mount3:
2082         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2083         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2084         percpu_counter_destroy(&sbi->s_dirs_counter);
2085         journal_destroy(sbi->s_journal);
2086 failed_mount2:
2087         for (i = 0; i < db_count; i++)
2088                 brelse(sbi->s_group_desc[i]);
2089         kfree(sbi->s_group_desc);
2090 failed_mount:
2091 #ifdef CONFIG_QUOTA
2092         for (i = 0; i < MAXQUOTAS; i++)
2093                 kfree(sbi->s_qf_names[i]);
2094 #endif
2095         ext3_blkdev_remove(sbi);
2096         brelse(bh);
2097 out_fail:
2098         sb->s_fs_info = NULL;
2099         kfree(sbi->s_blockgroup_lock);
2100         kfree(sbi);
2101         return ret;
2102 }
2103
2104 /*
2105  * Setup any per-fs journal parameters now.  We'll do this both on
2106  * initial mount, once the journal has been initialised but before we've
2107  * done any recovery; and again on any subsequent remount.
2108  */
2109 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2110 {
2111         struct ext3_sb_info *sbi = EXT3_SB(sb);
2112
2113         if (sbi->s_commit_interval)
2114                 journal->j_commit_interval = sbi->s_commit_interval;
2115         /* We could also set up an ext3-specific default for the commit
2116          * interval here, but for now we'll just fall back to the jbd
2117          * default. */
2118
2119         spin_lock(&journal->j_state_lock);
2120         if (test_opt(sb, BARRIER))
2121                 journal->j_flags |= JFS_BARRIER;
2122         else
2123                 journal->j_flags &= ~JFS_BARRIER;
2124         if (test_opt(sb, DATA_ERR_ABORT))
2125                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2126         else
2127                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2128         spin_unlock(&journal->j_state_lock);
2129 }
2130
2131 static journal_t *ext3_get_journal(struct super_block *sb,
2132                                    unsigned int journal_inum)
2133 {
2134         struct inode *journal_inode;
2135         journal_t *journal;
2136
2137         /* First, test for the existence of a valid inode on disk.  Bad
2138          * things happen if we iget() an unused inode, as the subsequent
2139          * iput() will try to delete it. */
2140
2141         journal_inode = ext3_iget(sb, journal_inum);
2142         if (IS_ERR(journal_inode)) {
2143                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2144                 return NULL;
2145         }
2146         if (!journal_inode->i_nlink) {
2147                 make_bad_inode(journal_inode);
2148                 iput(journal_inode);
2149                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2150                 return NULL;
2151         }
2152
2153         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2154                   journal_inode, journal_inode->i_size);
2155         if (!S_ISREG(journal_inode->i_mode)) {
2156                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2157                 iput(journal_inode);
2158                 return NULL;
2159         }
2160
2161         journal = journal_init_inode(journal_inode);
2162         if (!journal) {
2163                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2164                 iput(journal_inode);
2165                 return NULL;
2166         }
2167         journal->j_private = sb;
2168         ext3_init_journal_params(sb, journal);
2169         return journal;
2170 }
2171
2172 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2173                                        dev_t j_dev)
2174 {
2175         struct buffer_head * bh;
2176         journal_t *journal;
2177         ext3_fsblk_t start;
2178         ext3_fsblk_t len;
2179         int hblock, blocksize;
2180         ext3_fsblk_t sb_block;
2181         unsigned long offset;
2182         struct ext3_super_block * es;
2183         struct block_device *bdev;
2184
2185         bdev = ext3_blkdev_get(j_dev, sb);
2186         if (bdev == NULL)
2187                 return NULL;
2188
2189         blocksize = sb->s_blocksize;
2190         hblock = bdev_logical_block_size(bdev);
2191         if (blocksize < hblock) {
2192                 ext3_msg(sb, KERN_ERR,
2193                         "error: blocksize too small for journal device");
2194                 goto out_bdev;
2195         }
2196
2197         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2198         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2199         set_blocksize(bdev, blocksize);
2200         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2201                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2202                         "external journal");
2203                 goto out_bdev;
2204         }
2205
2206         es = (struct ext3_super_block *) (bh->b_data + offset);
2207         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2208             !(le32_to_cpu(es->s_feature_incompat) &
2209               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2210                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2211                         "bad superblock");
2212                 brelse(bh);
2213                 goto out_bdev;
2214         }
2215
2216         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2217                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2218                 brelse(bh);
2219                 goto out_bdev;
2220         }
2221
2222         len = le32_to_cpu(es->s_blocks_count);
2223         start = sb_block + 1;
2224         brelse(bh);     /* we're done with the superblock */
2225
2226         journal = journal_init_dev(bdev, sb->s_bdev,
2227                                         start, len, blocksize);
2228         if (!journal) {
2229                 ext3_msg(sb, KERN_ERR,
2230                         "error: failed to create device journal");
2231                 goto out_bdev;
2232         }
2233         journal->j_private = sb;
2234         if (!bh_uptodate_or_lock(journal->j_sb_buffer)) {
2235                 if (bh_submit_read(journal->j_sb_buffer)) {
2236                         ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2237                         goto out_journal;
2238                 }
2239         }
2240         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2241                 ext3_msg(sb, KERN_ERR,
2242                         "error: external journal has more than one "
2243                         "user (unsupported) - %d",
2244                         be32_to_cpu(journal->j_superblock->s_nr_users));
2245                 goto out_journal;
2246         }
2247         EXT3_SB(sb)->journal_bdev = bdev;
2248         ext3_init_journal_params(sb, journal);
2249         return journal;
2250 out_journal:
2251         journal_destroy(journal);
2252 out_bdev:
2253         ext3_blkdev_put(bdev);
2254         return NULL;
2255 }
2256
2257 static int ext3_load_journal(struct super_block *sb,
2258                              struct ext3_super_block *es,
2259                              unsigned long journal_devnum)
2260 {
2261         journal_t *journal;
2262         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2263         dev_t journal_dev;
2264         int err = 0;
2265         int really_read_only;
2266
2267         if (journal_devnum &&
2268             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2269                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2270                         "numbers have changed");
2271                 journal_dev = new_decode_dev(journal_devnum);
2272         } else
2273                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2274
2275         really_read_only = bdev_read_only(sb->s_bdev);
2276
2277         /*
2278          * Are we loading a blank journal or performing recovery after a
2279          * crash?  For recovery, we need to check in advance whether we
2280          * can get read-write access to the device.
2281          */
2282
2283         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2284                 if (sb->s_flags & MS_RDONLY) {
2285                         ext3_msg(sb, KERN_INFO,
2286                                 "recovery required on readonly filesystem");
2287                         if (really_read_only) {
2288                                 ext3_msg(sb, KERN_ERR, "error: write access "
2289                                         "unavailable, cannot proceed");
2290                                 return -EROFS;
2291                         }
2292                         ext3_msg(sb, KERN_INFO,
2293                                 "write access will be enabled during recovery");
2294                 }
2295         }
2296
2297         if (journal_inum && journal_dev) {
2298                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2299                        "and inode journals");
2300                 return -EINVAL;
2301         }
2302
2303         if (journal_inum) {
2304                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2305                         return -EINVAL;
2306         } else {
2307                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2308                         return -EINVAL;
2309         }
2310
2311         if (!(journal->j_flags & JFS_BARRIER))
2312                 printk(KERN_INFO "EXT3-fs: barriers not enabled\n");
2313
2314         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2315                 err = journal_update_format(journal);
2316                 if (err)  {
2317                         ext3_msg(sb, KERN_ERR, "error updating journal");
2318                         journal_destroy(journal);
2319                         return err;
2320                 }
2321         }
2322
2323         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2324                 err = journal_wipe(journal, !really_read_only);
2325         if (!err)
2326                 err = journal_load(journal);
2327
2328         if (err) {
2329                 ext3_msg(sb, KERN_ERR, "error loading journal");
2330                 journal_destroy(journal);
2331                 return err;
2332         }
2333
2334         EXT3_SB(sb)->s_journal = journal;
2335         ext3_clear_journal_err(sb, es);
2336
2337         if (!really_read_only && journal_devnum &&
2338             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2339                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2340
2341                 /* Make sure we flush the recovery flag to disk. */
2342                 ext3_commit_super(sb, es, 1);
2343         }
2344
2345         return 0;
2346 }
2347
2348 static int ext3_create_journal(struct super_block *sb,
2349                                struct ext3_super_block *es,
2350                                unsigned int journal_inum)
2351 {
2352         journal_t *journal;
2353         int err;
2354
2355         if (sb->s_flags & MS_RDONLY) {
2356                 ext3_msg(sb, KERN_ERR,
2357                         "error: readonly filesystem when trying to "
2358                         "create journal");
2359                 return -EROFS;
2360         }
2361
2362         journal = ext3_get_journal(sb, journal_inum);
2363         if (!journal)
2364                 return -EINVAL;
2365
2366         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2367                journal_inum);
2368
2369         err = journal_create(journal);
2370         if (err) {
2371                 ext3_msg(sb, KERN_ERR, "error creating journal");
2372                 journal_destroy(journal);
2373                 return -EIO;
2374         }
2375
2376         EXT3_SB(sb)->s_journal = journal;
2377
2378         ext3_update_dynamic_rev(sb);
2379         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2380         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2381
2382         es->s_journal_inum = cpu_to_le32(journal_inum);
2383
2384         /* Make sure we flush the recovery flag to disk. */
2385         ext3_commit_super(sb, es, 1);
2386
2387         return 0;
2388 }
2389
2390 static int ext3_commit_super(struct super_block *sb,
2391                                struct ext3_super_block *es,
2392                                int sync)
2393 {
2394         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2395         int error = 0;
2396
2397         if (!sbh)
2398                 return error;
2399
2400         if (buffer_write_io_error(sbh)) {
2401                 /*
2402                  * Oh, dear.  A previous attempt to write the
2403                  * superblock failed.  This could happen because the
2404                  * USB device was yanked out.  Or it could happen to
2405                  * be a transient write error and maybe the block will
2406                  * be remapped.  Nothing we can do but to retry the
2407                  * write and hope for the best.
2408                  */
2409                 ext3_msg(sb, KERN_ERR, "previous I/O error to "
2410                        "superblock detected");
2411                 clear_buffer_write_io_error(sbh);
2412                 set_buffer_uptodate(sbh);
2413         }
2414         /*
2415          * If the file system is mounted read-only, don't update the
2416          * superblock write time.  This avoids updating the superblock
2417          * write time when we are mounting the root file system
2418          * read/only but we need to replay the journal; at that point,
2419          * for people who are east of GMT and who make their clock
2420          * tick in localtime for Windows bug-for-bug compatibility,
2421          * the clock is set in the future, and this will cause e2fsck
2422          * to complain and force a full file system check.
2423          */
2424         if (!(sb->s_flags & MS_RDONLY))
2425                 es->s_wtime = cpu_to_le32(get_seconds());
2426         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2427         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2428         BUFFER_TRACE(sbh, "marking dirty");
2429         mark_buffer_dirty(sbh);
2430         if (sync) {
2431                 error = sync_dirty_buffer(sbh);
2432                 if (buffer_write_io_error(sbh)) {
2433                         ext3_msg(sb, KERN_ERR, "I/O error while writing "
2434                                "superblock");
2435                         clear_buffer_write_io_error(sbh);
2436                         set_buffer_uptodate(sbh);
2437                 }
2438         }
2439         return error;
2440 }
2441
2442
2443 /*
2444  * Have we just finished recovery?  If so, and if we are mounting (or
2445  * remounting) the filesystem readonly, then we will end up with a
2446  * consistent fs on disk.  Record that fact.
2447  */
2448 static void ext3_mark_recovery_complete(struct super_block * sb,
2449                                         struct ext3_super_block * es)
2450 {
2451         journal_t *journal = EXT3_SB(sb)->s_journal;
2452
2453         journal_lock_updates(journal);
2454         if (journal_flush(journal) < 0)
2455                 goto out;
2456
2457         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2458             sb->s_flags & MS_RDONLY) {
2459                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2460                 ext3_commit_super(sb, es, 1);
2461         }
2462
2463 out:
2464         journal_unlock_updates(journal);
2465 }
2466
2467 /*
2468  * If we are mounting (or read-write remounting) a filesystem whose journal
2469  * has recorded an error from a previous lifetime, move that error to the
2470  * main filesystem now.
2471  */
2472 static void ext3_clear_journal_err(struct super_block *sb,
2473                                    struct ext3_super_block *es)
2474 {
2475         journal_t *journal;
2476         int j_errno;
2477         const char *errstr;
2478
2479         journal = EXT3_SB(sb)->s_journal;
2480
2481         /*
2482          * Now check for any error status which may have been recorded in the
2483          * journal by a prior ext3_error() or ext3_abort()
2484          */
2485
2486         j_errno = journal_errno(journal);
2487         if (j_errno) {
2488                 char nbuf[16];
2489
2490                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2491                 ext3_warning(sb, __func__, "Filesystem error recorded "
2492                              "from previous mount: %s", errstr);
2493                 ext3_warning(sb, __func__, "Marking fs in need of "
2494                              "filesystem check.");
2495
2496                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2497                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2498                 ext3_commit_super (sb, es, 1);
2499
2500                 journal_clear_err(journal);
2501         }
2502 }
2503
2504 /*
2505  * Force the running and committing transactions to commit,
2506  * and wait on the commit.
2507  */
2508 int ext3_force_commit(struct super_block *sb)
2509 {
2510         journal_t *journal;
2511         int ret;
2512
2513         if (sb->s_flags & MS_RDONLY)
2514                 return 0;
2515
2516         journal = EXT3_SB(sb)->s_journal;
2517         ret = ext3_journal_force_commit(journal);
2518         return ret;
2519 }
2520
2521 static int ext3_sync_fs(struct super_block *sb, int wait)
2522 {
2523         tid_t target;
2524
2525         trace_ext3_sync_fs(sb, wait);
2526         /*
2527          * Writeback quota in non-journalled quota case - journalled quota has
2528          * no dirty dquots
2529          */
2530         dquot_writeback_dquots(sb, -1);
2531         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2532                 if (wait)
2533                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2534         }
2535         return 0;
2536 }
2537
2538 /*
2539  * LVM calls this function before a (read-only) snapshot is created.  This
2540  * gives us a chance to flush the journal completely and mark the fs clean.
2541  */
2542 static int ext3_freeze(struct super_block *sb)
2543 {
2544         int error = 0;
2545         journal_t *journal;
2546
2547         if (!(sb->s_flags & MS_RDONLY)) {
2548                 journal = EXT3_SB(sb)->s_journal;
2549
2550                 /* Now we set up the journal barrier. */
2551                 journal_lock_updates(journal);
2552
2553                 /*
2554                  * We don't want to clear needs_recovery flag when we failed
2555                  * to flush the journal.
2556                  */
2557                 error = journal_flush(journal);
2558                 if (error < 0)
2559                         goto out;
2560
2561                 /* Journal blocked and flushed, clear needs_recovery flag. */
2562                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2563                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2564                 if (error)
2565                         goto out;
2566         }
2567         return 0;
2568
2569 out:
2570         journal_unlock_updates(journal);
2571         return error;
2572 }
2573
2574 /*
2575  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2576  * flag here, even though the filesystem is not technically dirty yet.
2577  */
2578 static int ext3_unfreeze(struct super_block *sb)
2579 {
2580         if (!(sb->s_flags & MS_RDONLY)) {
2581                 /* Reser the needs_recovery flag before the fs is unlocked. */
2582                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2583                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2584                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2585         }
2586         return 0;
2587 }
2588
2589 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2590 {
2591         struct ext3_super_block * es;
2592         struct ext3_sb_info *sbi = EXT3_SB(sb);
2593         ext3_fsblk_t n_blocks_count = 0;
2594         unsigned long old_sb_flags;
2595         struct ext3_mount_options old_opts;
2596         int enable_quota = 0;
2597         int err;
2598 #ifdef CONFIG_QUOTA
2599         int i;
2600 #endif
2601
2602         /* Store the original options */
2603         old_sb_flags = sb->s_flags;
2604         old_opts.s_mount_opt = sbi->s_mount_opt;
2605         old_opts.s_resuid = sbi->s_resuid;
2606         old_opts.s_resgid = sbi->s_resgid;
2607         old_opts.s_commit_interval = sbi->s_commit_interval;
2608 #ifdef CONFIG_QUOTA
2609         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2610         for (i = 0; i < MAXQUOTAS; i++)
2611                 if (sbi->s_qf_names[i]) {
2612                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
2613                                                          GFP_KERNEL);
2614                         if (!old_opts.s_qf_names[i]) {
2615                                 int j;
2616
2617                                 for (j = 0; j < i; j++)
2618                                         kfree(old_opts.s_qf_names[j]);
2619                                 return -ENOMEM;
2620                         }
2621                 } else
2622                         old_opts.s_qf_names[i] = NULL;
2623 #endif
2624
2625         /*
2626          * Allow the "check" option to be passed as a remount option.
2627          */
2628         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2629                 err = -EINVAL;
2630                 goto restore_opts;
2631         }
2632
2633         if (test_opt(sb, ABORT))
2634                 ext3_abort(sb, __func__, "Abort forced by user");
2635
2636         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2637                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2638
2639         es = sbi->s_es;
2640
2641         ext3_init_journal_params(sb, sbi->s_journal);
2642
2643         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2644                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2645                 if (test_opt(sb, ABORT)) {
2646                         err = -EROFS;
2647                         goto restore_opts;
2648                 }
2649
2650                 if (*flags & MS_RDONLY) {
2651                         err = dquot_suspend(sb, -1);
2652                         if (err < 0)
2653                                 goto restore_opts;
2654
2655                         /*
2656                          * First of all, the unconditional stuff we have to do
2657                          * to disable replay of the journal when we next remount
2658                          */
2659                         sb->s_flags |= MS_RDONLY;
2660
2661                         /*
2662                          * OK, test if we are remounting a valid rw partition
2663                          * readonly, and if so set the rdonly flag and then
2664                          * mark the partition as valid again.
2665                          */
2666                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2667                             (sbi->s_mount_state & EXT3_VALID_FS))
2668                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2669
2670                         ext3_mark_recovery_complete(sb, es);
2671                 } else {
2672                         __le32 ret;
2673                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2674                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2675                                 ext3_msg(sb, KERN_WARNING,
2676                                         "warning: couldn't remount RDWR "
2677                                         "because of unsupported optional "
2678                                         "features (%x)", le32_to_cpu(ret));
2679                                 err = -EROFS;
2680                                 goto restore_opts;
2681                         }
2682
2683                         /*
2684                          * If we have an unprocessed orphan list hanging
2685                          * around from a previously readonly bdev mount,
2686                          * require a full umount & mount for now.
2687                          */
2688                         if (es->s_last_orphan) {
2689                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2690                                        "remount RDWR because of unprocessed "
2691                                        "orphan inode list.  Please "
2692                                        "umount & mount instead.");
2693                                 err = -EINVAL;
2694                                 goto restore_opts;
2695                         }
2696
2697                         /*
2698                          * Mounting a RDONLY partition read-write, so reread
2699                          * and store the current valid flag.  (It may have
2700                          * been changed by e2fsck since we originally mounted
2701                          * the partition.)
2702                          */
2703                         ext3_clear_journal_err(sb, es);
2704                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2705                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2706                                 goto restore_opts;
2707                         if (!ext3_setup_super (sb, es, 0))
2708                                 sb->s_flags &= ~MS_RDONLY;
2709                         enable_quota = 1;
2710                 }
2711         }
2712 #ifdef CONFIG_QUOTA
2713         /* Release old quota file names */
2714         for (i = 0; i < MAXQUOTAS; i++)
2715                 kfree(old_opts.s_qf_names[i]);
2716 #endif
2717         if (enable_quota)
2718                 dquot_resume(sb, -1);
2719         return 0;
2720 restore_opts:
2721         sb->s_flags = old_sb_flags;
2722         sbi->s_mount_opt = old_opts.s_mount_opt;
2723         sbi->s_resuid = old_opts.s_resuid;
2724         sbi->s_resgid = old_opts.s_resgid;
2725         sbi->s_commit_interval = old_opts.s_commit_interval;
2726 #ifdef CONFIG_QUOTA
2727         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2728         for (i = 0; i < MAXQUOTAS; i++) {
2729                 kfree(sbi->s_qf_names[i]);
2730                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2731         }
2732 #endif
2733         return err;
2734 }
2735
2736 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2737 {
2738         struct super_block *sb = dentry->d_sb;
2739         struct ext3_sb_info *sbi = EXT3_SB(sb);
2740         struct ext3_super_block *es = sbi->s_es;
2741         u64 fsid;
2742
2743         if (test_opt(sb, MINIX_DF)) {
2744                 sbi->s_overhead_last = 0;
2745         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2746                 unsigned long ngroups = sbi->s_groups_count, i;
2747                 ext3_fsblk_t overhead = 0;
2748                 smp_rmb();
2749
2750                 /*
2751                  * Compute the overhead (FS structures).  This is constant
2752                  * for a given filesystem unless the number of block groups
2753                  * changes so we cache the previous value until it does.
2754                  */
2755
2756                 /*
2757                  * All of the blocks before first_data_block are
2758                  * overhead
2759                  */
2760                 overhead = le32_to_cpu(es->s_first_data_block);
2761
2762                 /*
2763                  * Add the overhead attributed to the superblock and
2764                  * block group descriptors.  If the sparse superblocks
2765                  * feature is turned on, then not all groups have this.
2766                  */
2767                 for (i = 0; i < ngroups; i++) {
2768                         overhead += ext3_bg_has_super(sb, i) +
2769                                 ext3_bg_num_gdb(sb, i);
2770                         cond_resched();
2771                 }
2772
2773                 /*
2774                  * Every block group has an inode bitmap, a block
2775                  * bitmap, and an inode table.
2776                  */
2777                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2778                 sbi->s_overhead_last = overhead;
2779                 smp_wmb();
2780                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2781         }
2782
2783         buf->f_type = EXT3_SUPER_MAGIC;
2784         buf->f_bsize = sb->s_blocksize;
2785         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2786         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2787         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2788         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2789                 buf->f_bavail = 0;
2790         buf->f_files = le32_to_cpu(es->s_inodes_count);
2791         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2792         buf->f_namelen = EXT3_NAME_LEN;
2793         fsid = le64_to_cpup((void *)es->s_uuid) ^
2794                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2795         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2796         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2797         return 0;
2798 }
2799
2800 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2801  * is locked for write. Otherwise the are possible deadlocks:
2802  * Process 1                         Process 2
2803  * ext3_create()                     quota_sync()
2804  *   journal_start()                   write_dquot()
2805  *   dquot_initialize()                       down(dqio_mutex)
2806  *     down(dqio_mutex)                    journal_start()
2807  *
2808  */
2809
2810 #ifdef CONFIG_QUOTA
2811
2812 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2813 {
2814         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
2815 }
2816
2817 static int ext3_write_dquot(struct dquot *dquot)
2818 {
2819         int ret, err;
2820         handle_t *handle;
2821         struct inode *inode;
2822
2823         inode = dquot_to_inode(dquot);
2824         handle = ext3_journal_start(inode,
2825                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2826         if (IS_ERR(handle))
2827                 return PTR_ERR(handle);
2828         ret = dquot_commit(dquot);
2829         err = ext3_journal_stop(handle);
2830         if (!ret)
2831                 ret = err;
2832         return ret;
2833 }
2834
2835 static int ext3_acquire_dquot(struct dquot *dquot)
2836 {
2837         int ret, err;
2838         handle_t *handle;
2839
2840         handle = ext3_journal_start(dquot_to_inode(dquot),
2841                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2842         if (IS_ERR(handle))
2843                 return PTR_ERR(handle);
2844         ret = dquot_acquire(dquot);
2845         err = ext3_journal_stop(handle);
2846         if (!ret)
2847                 ret = err;
2848         return ret;
2849 }
2850
2851 static int ext3_release_dquot(struct dquot *dquot)
2852 {
2853         int ret, err;
2854         handle_t *handle;
2855
2856         handle = ext3_journal_start(dquot_to_inode(dquot),
2857                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2858         if (IS_ERR(handle)) {
2859                 /* Release dquot anyway to avoid endless cycle in dqput() */
2860                 dquot_release(dquot);
2861                 return PTR_ERR(handle);
2862         }
2863         ret = dquot_release(dquot);
2864         err = ext3_journal_stop(handle);
2865         if (!ret)
2866                 ret = err;
2867         return ret;
2868 }
2869
2870 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2871 {
2872         /* Are we journaling quotas? */
2873         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2874             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2875                 dquot_mark_dquot_dirty(dquot);
2876                 return ext3_write_dquot(dquot);
2877         } else {
2878                 return dquot_mark_dquot_dirty(dquot);
2879         }
2880 }
2881
2882 static int ext3_write_info(struct super_block *sb, int type)
2883 {
2884         int ret, err;
2885         handle_t *handle;
2886
2887         /* Data block + inode block */
2888         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2889         if (IS_ERR(handle))
2890                 return PTR_ERR(handle);
2891         ret = dquot_commit_info(sb, type);
2892         err = ext3_journal_stop(handle);
2893         if (!ret)
2894                 ret = err;
2895         return ret;
2896 }
2897
2898 /*
2899  * Turn on quotas during mount time - we need to find
2900  * the quota file and such...
2901  */
2902 static int ext3_quota_on_mount(struct super_block *sb, int type)
2903 {
2904         return dquot_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2905                                         EXT3_SB(sb)->s_jquota_fmt, type);
2906 }
2907
2908 /*
2909  * Standard function to be called on quota_on
2910  */
2911 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2912                          struct path *path)
2913 {
2914         int err;
2915
2916         if (!test_opt(sb, QUOTA))
2917                 return -EINVAL;
2918
2919         /* Quotafile not on the same filesystem? */
2920         if (path->dentry->d_sb != sb)
2921                 return -EXDEV;
2922         /* Journaling quota? */
2923         if (EXT3_SB(sb)->s_qf_names[type]) {
2924                 /* Quotafile not of fs root? */
2925                 if (path->dentry->d_parent != sb->s_root)
2926                         ext3_msg(sb, KERN_WARNING,
2927                                 "warning: Quota file not on filesystem root. "
2928                                 "Journaled quota will not work.");
2929         }
2930
2931         /*
2932          * When we journal data on quota file, we have to flush journal to see
2933          * all updates to the file when we bypass pagecache...
2934          */
2935         if (ext3_should_journal_data(path->dentry->d_inode)) {
2936                 /*
2937                  * We don't need to lock updates but journal_flush() could
2938                  * otherwise be livelocked...
2939                  */
2940                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2941                 err = journal_flush(EXT3_SB(sb)->s_journal);
2942                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2943                 if (err)
2944                         return err;
2945         }
2946
2947         return dquot_quota_on(sb, type, format_id, path);
2948 }
2949
2950 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2951  * acquiring the locks... As quota files are never truncated and quota code
2952  * itself serializes the operations (and no one else should touch the files)
2953  * we don't have to be afraid of races */
2954 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2955                                size_t len, loff_t off)
2956 {
2957         struct inode *inode = sb_dqopt(sb)->files[type];
2958         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2959         int err = 0;
2960         int offset = off & (sb->s_blocksize - 1);
2961         int tocopy;
2962         size_t toread;
2963         struct buffer_head *bh;
2964         loff_t i_size = i_size_read(inode);
2965
2966         if (off > i_size)
2967                 return 0;
2968         if (off+len > i_size)
2969                 len = i_size-off;
2970         toread = len;
2971         while (toread > 0) {
2972                 tocopy = sb->s_blocksize - offset < toread ?
2973                                 sb->s_blocksize - offset : toread;
2974                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2975                 if (err)
2976                         return err;
2977                 if (!bh)        /* A hole? */
2978                         memset(data, 0, tocopy);
2979                 else
2980                         memcpy(data, bh->b_data+offset, tocopy);
2981                 brelse(bh);
2982                 offset = 0;
2983                 toread -= tocopy;
2984                 data += tocopy;
2985                 blk++;
2986         }
2987         return len;
2988 }
2989
2990 /* Write to quotafile (we know the transaction is already started and has
2991  * enough credits) */
2992 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2993                                 const char *data, size_t len, loff_t off)
2994 {
2995         struct inode *inode = sb_dqopt(sb)->files[type];
2996         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2997         int err = 0;
2998         int offset = off & (sb->s_blocksize - 1);
2999         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
3000         struct buffer_head *bh;
3001         handle_t *handle = journal_current_handle();
3002
3003         if (!handle) {
3004                 ext3_msg(sb, KERN_WARNING,
3005                         "warning: quota write (off=%llu, len=%llu)"
3006                         " cancelled because transaction is not started.",
3007                         (unsigned long long)off, (unsigned long long)len);
3008                 return -EIO;
3009         }
3010
3011         /*
3012          * Since we account only one data block in transaction credits,
3013          * then it is impossible to cross a block boundary.
3014          */
3015         if (sb->s_blocksize - offset < len) {
3016                 ext3_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
3017                         " cancelled because not block aligned",
3018                         (unsigned long long)off, (unsigned long long)len);
3019                 return -EIO;
3020         }
3021         bh = ext3_bread(handle, inode, blk, 1, &err);
3022         if (!bh)
3023                 goto out;
3024         if (journal_quota) {
3025                 err = ext3_journal_get_write_access(handle, bh);
3026                 if (err) {
3027                         brelse(bh);
3028                         goto out;
3029                 }
3030         }
3031         lock_buffer(bh);
3032         memcpy(bh->b_data+offset, data, len);
3033         flush_dcache_page(bh->b_page);
3034         unlock_buffer(bh);
3035         if (journal_quota)
3036                 err = ext3_journal_dirty_metadata(handle, bh);
3037         else {
3038                 /* Always do at least ordered writes for quotas */
3039                 err = ext3_journal_dirty_data(handle, bh);
3040                 mark_buffer_dirty(bh);
3041         }
3042         brelse(bh);
3043 out:
3044         if (err)
3045                 return err;
3046         if (inode->i_size < off + len) {
3047                 i_size_write(inode, off + len);
3048                 EXT3_I(inode)->i_disksize = inode->i_size;
3049         }
3050         inode->i_version++;
3051         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3052         ext3_mark_inode_dirty(handle, inode);
3053         return len;
3054 }
3055
3056 #endif
3057
3058 static struct dentry *ext3_mount(struct file_system_type *fs_type,
3059         int flags, const char *dev_name, void *data)
3060 {
3061         return mount_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
3062 }
3063
3064 static struct file_system_type ext3_fs_type = {
3065         .owner          = THIS_MODULE,
3066         .name           = "ext3",
3067         .mount          = ext3_mount,
3068         .kill_sb        = kill_block_super,
3069         .fs_flags       = FS_REQUIRES_DEV,
3070 };
3071
3072 static int __init init_ext3_fs(void)
3073 {
3074         int err = init_ext3_xattr();
3075         if (err)
3076                 return err;
3077         err = init_inodecache();
3078         if (err)
3079                 goto out1;
3080         err = register_filesystem(&ext3_fs_type);
3081         if (err)
3082                 goto out;
3083         return 0;
3084 out:
3085         destroy_inodecache();
3086 out1:
3087         exit_ext3_xattr();
3088         return err;
3089 }
3090
3091 static void __exit exit_ext3_fs(void)
3092 {
3093         unregister_filesystem(&ext3_fs_type);
3094         destroy_inodecache();
3095         exit_ext3_xattr();
3096 }
3097
3098 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3099 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3100 MODULE_LICENSE("GPL");
3101 module_init(init_ext3_fs)
3102 module_exit(exit_ext3_fs)