ext3: add cleancache support
[linux-2.6.git] / fs / ext3 / super.c
1 /*
2  *  linux/fs/ext3/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/module.h>
20 #include <linux/string.h>
21 #include <linux/fs.h>
22 #include <linux/time.h>
23 #include <linux/jbd.h>
24 #include <linux/ext3_fs.h>
25 #include <linux/ext3_jbd.h>
26 #include <linux/slab.h>
27 #include <linux/init.h>
28 #include <linux/blkdev.h>
29 #include <linux/parser.h>
30 #include <linux/buffer_head.h>
31 #include <linux/exportfs.h>
32 #include <linux/vfs.h>
33 #include <linux/random.h>
34 #include <linux/mount.h>
35 #include <linux/namei.h>
36 #include <linux/quotaops.h>
37 #include <linux/seq_file.h>
38 #include <linux/log2.h>
39 #include <linux/cleancache.h>
40
41 #include <asm/uaccess.h>
42
43 #include "xattr.h"
44 #include "acl.h"
45 #include "namei.h"
46
47 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
48   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
49 #else
50   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
51 #endif
52
53 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
54                              unsigned long journal_devnum);
55 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
56                                unsigned int);
57 static int ext3_commit_super(struct super_block *sb,
58                                struct ext3_super_block *es,
59                                int sync);
60 static void ext3_mark_recovery_complete(struct super_block * sb,
61                                         struct ext3_super_block * es);
62 static void ext3_clear_journal_err(struct super_block * sb,
63                                    struct ext3_super_block * es);
64 static int ext3_sync_fs(struct super_block *sb, int wait);
65 static const char *ext3_decode_error(struct super_block * sb, int errno,
66                                      char nbuf[16]);
67 static int ext3_remount (struct super_block * sb, int * flags, char * data);
68 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
69 static int ext3_unfreeze(struct super_block *sb);
70 static int ext3_freeze(struct super_block *sb);
71
72 /*
73  * Wrappers for journal_start/end.
74  *
75  * The only special thing we need to do here is to make sure that all
76  * journal_end calls result in the superblock being marked dirty, so
77  * that sync() will call the filesystem's write_super callback if
78  * appropriate.
79  */
80 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
81 {
82         journal_t *journal;
83
84         if (sb->s_flags & MS_RDONLY)
85                 return ERR_PTR(-EROFS);
86
87         /* Special case here: if the journal has aborted behind our
88          * backs (eg. EIO in the commit thread), then we still need to
89          * take the FS itself readonly cleanly. */
90         journal = EXT3_SB(sb)->s_journal;
91         if (is_journal_aborted(journal)) {
92                 ext3_abort(sb, __func__,
93                            "Detected aborted journal");
94                 return ERR_PTR(-EROFS);
95         }
96
97         return journal_start(journal, nblocks);
98 }
99
100 /*
101  * The only special thing we need to do here is to make sure that all
102  * journal_stop calls result in the superblock being marked dirty, so
103  * that sync() will call the filesystem's write_super callback if
104  * appropriate.
105  */
106 int __ext3_journal_stop(const char *where, handle_t *handle)
107 {
108         struct super_block *sb;
109         int err;
110         int rc;
111
112         sb = handle->h_transaction->t_journal->j_private;
113         err = handle->h_err;
114         rc = journal_stop(handle);
115
116         if (!err)
117                 err = rc;
118         if (err)
119                 __ext3_std_error(sb, where, err);
120         return err;
121 }
122
123 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
124                 struct buffer_head *bh, handle_t *handle, int err)
125 {
126         char nbuf[16];
127         const char *errstr = ext3_decode_error(NULL, err, nbuf);
128
129         if (bh)
130                 BUFFER_TRACE(bh, "abort");
131
132         if (!handle->h_err)
133                 handle->h_err = err;
134
135         if (is_handle_aborted(handle))
136                 return;
137
138         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
139                 caller, errstr, err_fn);
140
141         journal_abort_handle(handle);
142 }
143
144 void ext3_msg(struct super_block *sb, const char *prefix,
145                 const char *fmt, ...)
146 {
147         struct va_format vaf;
148         va_list args;
149
150         va_start(args, fmt);
151
152         vaf.fmt = fmt;
153         vaf.va = &args;
154
155         printk("%sEXT3-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
156
157         va_end(args);
158 }
159
160 /* Deal with the reporting of failure conditions on a filesystem such as
161  * inconsistencies detected or read IO failures.
162  *
163  * On ext2, we can store the error state of the filesystem in the
164  * superblock.  That is not possible on ext3, because we may have other
165  * write ordering constraints on the superblock which prevent us from
166  * writing it out straight away; and given that the journal is about to
167  * be aborted, we can't rely on the current, or future, transactions to
168  * write out the superblock safely.
169  *
170  * We'll just use the journal_abort() error code to record an error in
171  * the journal instead.  On recovery, the journal will complain about
172  * that error until we've noted it down and cleared it.
173  */
174
175 static void ext3_handle_error(struct super_block *sb)
176 {
177         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
178
179         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
180         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
181
182         if (sb->s_flags & MS_RDONLY)
183                 return;
184
185         if (!test_opt (sb, ERRORS_CONT)) {
186                 journal_t *journal = EXT3_SB(sb)->s_journal;
187
188                 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
189                 if (journal)
190                         journal_abort(journal, -EIO);
191         }
192         if (test_opt (sb, ERRORS_RO)) {
193                 ext3_msg(sb, KERN_CRIT,
194                         "error: remounting filesystem read-only");
195                 sb->s_flags |= MS_RDONLY;
196         }
197         ext3_commit_super(sb, es, 1);
198         if (test_opt(sb, ERRORS_PANIC))
199                 panic("EXT3-fs (%s): panic forced after error\n",
200                         sb->s_id);
201 }
202
203 void ext3_error(struct super_block *sb, const char *function,
204                 const char *fmt, ...)
205 {
206         struct va_format vaf;
207         va_list args;
208
209         va_start(args, fmt);
210
211         vaf.fmt = fmt;
212         vaf.va = &args;
213
214         printk(KERN_CRIT "EXT3-fs error (device %s): %s: %pV\n",
215                sb->s_id, function, &vaf);
216
217         va_end(args);
218
219         ext3_handle_error(sb);
220 }
221
222 static const char *ext3_decode_error(struct super_block * sb, int errno,
223                                      char nbuf[16])
224 {
225         char *errstr = NULL;
226
227         switch (errno) {
228         case -EIO:
229                 errstr = "IO failure";
230                 break;
231         case -ENOMEM:
232                 errstr = "Out of memory";
233                 break;
234         case -EROFS:
235                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
236                         errstr = "Journal has aborted";
237                 else
238                         errstr = "Readonly filesystem";
239                 break;
240         default:
241                 /* If the caller passed in an extra buffer for unknown
242                  * errors, textualise them now.  Else we just return
243                  * NULL. */
244                 if (nbuf) {
245                         /* Check for truncated error codes... */
246                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
247                                 errstr = nbuf;
248                 }
249                 break;
250         }
251
252         return errstr;
253 }
254
255 /* __ext3_std_error decodes expected errors from journaling functions
256  * automatically and invokes the appropriate error response.  */
257
258 void __ext3_std_error (struct super_block * sb, const char * function,
259                        int errno)
260 {
261         char nbuf[16];
262         const char *errstr;
263
264         /* Special case: if the error is EROFS, and we're not already
265          * inside a transaction, then there's really no point in logging
266          * an error. */
267         if (errno == -EROFS && journal_current_handle() == NULL &&
268             (sb->s_flags & MS_RDONLY))
269                 return;
270
271         errstr = ext3_decode_error(sb, errno, nbuf);
272         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
273
274         ext3_handle_error(sb);
275 }
276
277 /*
278  * ext3_abort is a much stronger failure handler than ext3_error.  The
279  * abort function may be used to deal with unrecoverable failures such
280  * as journal IO errors or ENOMEM at a critical moment in log management.
281  *
282  * We unconditionally force the filesystem into an ABORT|READONLY state,
283  * unless the error response on the fs has been set to panic in which
284  * case we take the easy way out and panic immediately.
285  */
286
287 void ext3_abort(struct super_block *sb, const char *function,
288                  const char *fmt, ...)
289 {
290         struct va_format vaf;
291         va_list args;
292
293         va_start(args, fmt);
294
295         vaf.fmt = fmt;
296         vaf.va = &args;
297
298         printk(KERN_CRIT "EXT3-fs (%s): error: %s: %pV\n",
299                sb->s_id, function, &vaf);
300
301         va_end(args);
302
303         if (test_opt(sb, ERRORS_PANIC))
304                 panic("EXT3-fs: panic from previous error\n");
305
306         if (sb->s_flags & MS_RDONLY)
307                 return;
308
309         ext3_msg(sb, KERN_CRIT,
310                 "error: remounting filesystem read-only");
311         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
312         sb->s_flags |= MS_RDONLY;
313         set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
314         if (EXT3_SB(sb)->s_journal)
315                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
316 }
317
318 void ext3_warning(struct super_block *sb, const char *function,
319                   const char *fmt, ...)
320 {
321         struct va_format vaf;
322         va_list args;
323
324         va_start(args, fmt);
325
326         vaf.fmt = fmt;
327         vaf.va = &args;
328
329         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: %pV\n",
330                sb->s_id, function, &vaf);
331
332         va_end(args);
333 }
334
335 void ext3_update_dynamic_rev(struct super_block *sb)
336 {
337         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
338
339         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
340                 return;
341
342         ext3_msg(sb, KERN_WARNING,
343                 "warning: updating to rev %d because of "
344                 "new feature flag, running e2fsck is recommended",
345                 EXT3_DYNAMIC_REV);
346
347         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
348         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
349         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
350         /* leave es->s_feature_*compat flags alone */
351         /* es->s_uuid will be set by e2fsck if empty */
352
353         /*
354          * The rest of the superblock fields should be zero, and if not it
355          * means they are likely already in use, so leave them alone.  We
356          * can leave it up to e2fsck to clean up any inconsistencies there.
357          */
358 }
359
360 /*
361  * Open the external journal device
362  */
363 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
364 {
365         struct block_device *bdev;
366         char b[BDEVNAME_SIZE];
367
368         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
369         if (IS_ERR(bdev))
370                 goto fail;
371         return bdev;
372
373 fail:
374         ext3_msg(sb, "error: failed to open journal device %s: %ld",
375                 __bdevname(dev, b), PTR_ERR(bdev));
376
377         return NULL;
378 }
379
380 /*
381  * Release the journal device
382  */
383 static int ext3_blkdev_put(struct block_device *bdev)
384 {
385         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
386 }
387
388 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
389 {
390         struct block_device *bdev;
391         int ret = -ENODEV;
392
393         bdev = sbi->journal_bdev;
394         if (bdev) {
395                 ret = ext3_blkdev_put(bdev);
396                 sbi->journal_bdev = NULL;
397         }
398         return ret;
399 }
400
401 static inline struct inode *orphan_list_entry(struct list_head *l)
402 {
403         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
404 }
405
406 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
407 {
408         struct list_head *l;
409
410         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
411                le32_to_cpu(sbi->s_es->s_last_orphan));
412
413         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
414         list_for_each(l, &sbi->s_orphan) {
415                 struct inode *inode = orphan_list_entry(l);
416                 ext3_msg(sb, KERN_ERR, "  "
417                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
418                        inode->i_sb->s_id, inode->i_ino, inode,
419                        inode->i_mode, inode->i_nlink,
420                        NEXT_ORPHAN(inode));
421         }
422 }
423
424 static void ext3_put_super (struct super_block * sb)
425 {
426         struct ext3_sb_info *sbi = EXT3_SB(sb);
427         struct ext3_super_block *es = sbi->s_es;
428         int i, err;
429
430         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
431         ext3_xattr_put_super(sb);
432         err = journal_destroy(sbi->s_journal);
433         sbi->s_journal = NULL;
434         if (err < 0)
435                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
436
437         if (!(sb->s_flags & MS_RDONLY)) {
438                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
439                 es->s_state = cpu_to_le16(sbi->s_mount_state);
440                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
441                 mark_buffer_dirty(sbi->s_sbh);
442                 ext3_commit_super(sb, es, 1);
443         }
444
445         for (i = 0; i < sbi->s_gdb_count; i++)
446                 brelse(sbi->s_group_desc[i]);
447         kfree(sbi->s_group_desc);
448         percpu_counter_destroy(&sbi->s_freeblocks_counter);
449         percpu_counter_destroy(&sbi->s_freeinodes_counter);
450         percpu_counter_destroy(&sbi->s_dirs_counter);
451         brelse(sbi->s_sbh);
452 #ifdef CONFIG_QUOTA
453         for (i = 0; i < MAXQUOTAS; i++)
454                 kfree(sbi->s_qf_names[i]);
455 #endif
456
457         /* Debugging code just in case the in-memory inode orphan list
458          * isn't empty.  The on-disk one can be non-empty if we've
459          * detected an error and taken the fs readonly, but the
460          * in-memory list had better be clean by this point. */
461         if (!list_empty(&sbi->s_orphan))
462                 dump_orphan_list(sb, sbi);
463         J_ASSERT(list_empty(&sbi->s_orphan));
464
465         invalidate_bdev(sb->s_bdev);
466         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
467                 /*
468                  * Invalidate the journal device's buffers.  We don't want them
469                  * floating about in memory - the physical journal device may
470                  * hotswapped, and it breaks the `ro-after' testing code.
471                  */
472                 sync_blockdev(sbi->journal_bdev);
473                 invalidate_bdev(sbi->journal_bdev);
474                 ext3_blkdev_remove(sbi);
475         }
476         sb->s_fs_info = NULL;
477         kfree(sbi->s_blockgroup_lock);
478         kfree(sbi);
479 }
480
481 static struct kmem_cache *ext3_inode_cachep;
482
483 /*
484  * Called inside transaction, so use GFP_NOFS
485  */
486 static struct inode *ext3_alloc_inode(struct super_block *sb)
487 {
488         struct ext3_inode_info *ei;
489
490         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
491         if (!ei)
492                 return NULL;
493         ei->i_block_alloc_info = NULL;
494         ei->vfs_inode.i_version = 1;
495         atomic_set(&ei->i_datasync_tid, 0);
496         atomic_set(&ei->i_sync_tid, 0);
497         return &ei->vfs_inode;
498 }
499
500 static void ext3_i_callback(struct rcu_head *head)
501 {
502         struct inode *inode = container_of(head, struct inode, i_rcu);
503         INIT_LIST_HEAD(&inode->i_dentry);
504         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
505 }
506
507 static void ext3_destroy_inode(struct inode *inode)
508 {
509         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
510                 printk("EXT3 Inode %p: orphan list check failed!\n",
511                         EXT3_I(inode));
512                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
513                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
514                                 false);
515                 dump_stack();
516         }
517         call_rcu(&inode->i_rcu, ext3_i_callback);
518 }
519
520 static void init_once(void *foo)
521 {
522         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
523
524         INIT_LIST_HEAD(&ei->i_orphan);
525 #ifdef CONFIG_EXT3_FS_XATTR
526         init_rwsem(&ei->xattr_sem);
527 #endif
528         mutex_init(&ei->truncate_mutex);
529         inode_init_once(&ei->vfs_inode);
530 }
531
532 static int init_inodecache(void)
533 {
534         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
535                                              sizeof(struct ext3_inode_info),
536                                              0, (SLAB_RECLAIM_ACCOUNT|
537                                                 SLAB_MEM_SPREAD),
538                                              init_once);
539         if (ext3_inode_cachep == NULL)
540                 return -ENOMEM;
541         return 0;
542 }
543
544 static void destroy_inodecache(void)
545 {
546         kmem_cache_destroy(ext3_inode_cachep);
547 }
548
549 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
550 {
551 #if defined(CONFIG_QUOTA)
552         struct ext3_sb_info *sbi = EXT3_SB(sb);
553
554         if (sbi->s_jquota_fmt) {
555                 char *fmtname = "";
556
557                 switch (sbi->s_jquota_fmt) {
558                 case QFMT_VFS_OLD:
559                         fmtname = "vfsold";
560                         break;
561                 case QFMT_VFS_V0:
562                         fmtname = "vfsv0";
563                         break;
564                 case QFMT_VFS_V1:
565                         fmtname = "vfsv1";
566                         break;
567                 }
568                 seq_printf(seq, ",jqfmt=%s", fmtname);
569         }
570
571         if (sbi->s_qf_names[USRQUOTA])
572                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
573
574         if (sbi->s_qf_names[GRPQUOTA])
575                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
576
577         if (test_opt(sb, USRQUOTA))
578                 seq_puts(seq, ",usrquota");
579
580         if (test_opt(sb, GRPQUOTA))
581                 seq_puts(seq, ",grpquota");
582 #endif
583 }
584
585 static char *data_mode_string(unsigned long mode)
586 {
587         switch (mode) {
588         case EXT3_MOUNT_JOURNAL_DATA:
589                 return "journal";
590         case EXT3_MOUNT_ORDERED_DATA:
591                 return "ordered";
592         case EXT3_MOUNT_WRITEBACK_DATA:
593                 return "writeback";
594         }
595         return "unknown";
596 }
597
598 /*
599  * Show an option if
600  *  - it's set to a non-default value OR
601  *  - if the per-sb default is different from the global default
602  */
603 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
604 {
605         struct super_block *sb = vfs->mnt_sb;
606         struct ext3_sb_info *sbi = EXT3_SB(sb);
607         struct ext3_super_block *es = sbi->s_es;
608         unsigned long def_mount_opts;
609
610         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
611
612         if (sbi->s_sb_block != 1)
613                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
614         if (test_opt(sb, MINIX_DF))
615                 seq_puts(seq, ",minixdf");
616         if (test_opt(sb, GRPID))
617                 seq_puts(seq, ",grpid");
618         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
619                 seq_puts(seq, ",nogrpid");
620         if (sbi->s_resuid != EXT3_DEF_RESUID ||
621             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
622                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
623         }
624         if (sbi->s_resgid != EXT3_DEF_RESGID ||
625             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
626                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
627         }
628         if (test_opt(sb, ERRORS_RO)) {
629                 int def_errors = le16_to_cpu(es->s_errors);
630
631                 if (def_errors == EXT3_ERRORS_PANIC ||
632                     def_errors == EXT3_ERRORS_CONTINUE) {
633                         seq_puts(seq, ",errors=remount-ro");
634                 }
635         }
636         if (test_opt(sb, ERRORS_CONT))
637                 seq_puts(seq, ",errors=continue");
638         if (test_opt(sb, ERRORS_PANIC))
639                 seq_puts(seq, ",errors=panic");
640         if (test_opt(sb, NO_UID32))
641                 seq_puts(seq, ",nouid32");
642         if (test_opt(sb, DEBUG))
643                 seq_puts(seq, ",debug");
644         if (test_opt(sb, OLDALLOC))
645                 seq_puts(seq, ",oldalloc");
646 #ifdef CONFIG_EXT3_FS_XATTR
647         if (test_opt(sb, XATTR_USER))
648                 seq_puts(seq, ",user_xattr");
649         if (!test_opt(sb, XATTR_USER) &&
650             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
651                 seq_puts(seq, ",nouser_xattr");
652         }
653 #endif
654 #ifdef CONFIG_EXT3_FS_POSIX_ACL
655         if (test_opt(sb, POSIX_ACL))
656                 seq_puts(seq, ",acl");
657         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
658                 seq_puts(seq, ",noacl");
659 #endif
660         if (!test_opt(sb, RESERVATION))
661                 seq_puts(seq, ",noreservation");
662         if (sbi->s_commit_interval) {
663                 seq_printf(seq, ",commit=%u",
664                            (unsigned) (sbi->s_commit_interval / HZ));
665         }
666
667         /*
668          * Always display barrier state so it's clear what the status is.
669          */
670         seq_puts(seq, ",barrier=");
671         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
672         seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
673         if (test_opt(sb, DATA_ERR_ABORT))
674                 seq_puts(seq, ",data_err=abort");
675
676         if (test_opt(sb, NOLOAD))
677                 seq_puts(seq, ",norecovery");
678
679         ext3_show_quota_options(seq, sb);
680
681         return 0;
682 }
683
684
685 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
686                 u64 ino, u32 generation)
687 {
688         struct inode *inode;
689
690         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
691                 return ERR_PTR(-ESTALE);
692         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
693                 return ERR_PTR(-ESTALE);
694
695         /* iget isn't really right if the inode is currently unallocated!!
696          *
697          * ext3_read_inode will return a bad_inode if the inode had been
698          * deleted, so we should be safe.
699          *
700          * Currently we don't know the generation for parent directory, so
701          * a generation of 0 means "accept any"
702          */
703         inode = ext3_iget(sb, ino);
704         if (IS_ERR(inode))
705                 return ERR_CAST(inode);
706         if (generation && inode->i_generation != generation) {
707                 iput(inode);
708                 return ERR_PTR(-ESTALE);
709         }
710
711         return inode;
712 }
713
714 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
715                 int fh_len, int fh_type)
716 {
717         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
718                                     ext3_nfs_get_inode);
719 }
720
721 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
722                 int fh_len, int fh_type)
723 {
724         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
725                                     ext3_nfs_get_inode);
726 }
727
728 /*
729  * Try to release metadata pages (indirect blocks, directories) which are
730  * mapped via the block device.  Since these pages could have journal heads
731  * which would prevent try_to_free_buffers() from freeing them, we must use
732  * jbd layer's try_to_free_buffers() function to release them.
733  */
734 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
735                                  gfp_t wait)
736 {
737         journal_t *journal = EXT3_SB(sb)->s_journal;
738
739         WARN_ON(PageChecked(page));
740         if (!page_has_buffers(page))
741                 return 0;
742         if (journal)
743                 return journal_try_to_free_buffers(journal, page, 
744                                                    wait & ~__GFP_WAIT);
745         return try_to_free_buffers(page);
746 }
747
748 #ifdef CONFIG_QUOTA
749 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
750 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
751
752 static int ext3_write_dquot(struct dquot *dquot);
753 static int ext3_acquire_dquot(struct dquot *dquot);
754 static int ext3_release_dquot(struct dquot *dquot);
755 static int ext3_mark_dquot_dirty(struct dquot *dquot);
756 static int ext3_write_info(struct super_block *sb, int type);
757 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
758                          struct path *path);
759 static int ext3_quota_on_mount(struct super_block *sb, int type);
760 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
761                                size_t len, loff_t off);
762 static ssize_t ext3_quota_write(struct super_block *sb, int type,
763                                 const char *data, size_t len, loff_t off);
764
765 static const struct dquot_operations ext3_quota_operations = {
766         .write_dquot    = ext3_write_dquot,
767         .acquire_dquot  = ext3_acquire_dquot,
768         .release_dquot  = ext3_release_dquot,
769         .mark_dirty     = ext3_mark_dquot_dirty,
770         .write_info     = ext3_write_info,
771         .alloc_dquot    = dquot_alloc,
772         .destroy_dquot  = dquot_destroy,
773 };
774
775 static const struct quotactl_ops ext3_qctl_operations = {
776         .quota_on       = ext3_quota_on,
777         .quota_off      = dquot_quota_off,
778         .quota_sync     = dquot_quota_sync,
779         .get_info       = dquot_get_dqinfo,
780         .set_info       = dquot_set_dqinfo,
781         .get_dqblk      = dquot_get_dqblk,
782         .set_dqblk      = dquot_set_dqblk
783 };
784 #endif
785
786 static const struct super_operations ext3_sops = {
787         .alloc_inode    = ext3_alloc_inode,
788         .destroy_inode  = ext3_destroy_inode,
789         .write_inode    = ext3_write_inode,
790         .dirty_inode    = ext3_dirty_inode,
791         .evict_inode    = ext3_evict_inode,
792         .put_super      = ext3_put_super,
793         .sync_fs        = ext3_sync_fs,
794         .freeze_fs      = ext3_freeze,
795         .unfreeze_fs    = ext3_unfreeze,
796         .statfs         = ext3_statfs,
797         .remount_fs     = ext3_remount,
798         .show_options   = ext3_show_options,
799 #ifdef CONFIG_QUOTA
800         .quota_read     = ext3_quota_read,
801         .quota_write    = ext3_quota_write,
802 #endif
803         .bdev_try_to_free_page = bdev_try_to_free_page,
804 };
805
806 static const struct export_operations ext3_export_ops = {
807         .fh_to_dentry = ext3_fh_to_dentry,
808         .fh_to_parent = ext3_fh_to_parent,
809         .get_parent = ext3_get_parent,
810 };
811
812 enum {
813         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
814         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
815         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
816         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
817         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
818         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
819         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
820         Opt_data_err_abort, Opt_data_err_ignore,
821         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
822         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
823         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
824         Opt_resize, Opt_usrquota, Opt_grpquota
825 };
826
827 static const match_table_t tokens = {
828         {Opt_bsd_df, "bsddf"},
829         {Opt_minix_df, "minixdf"},
830         {Opt_grpid, "grpid"},
831         {Opt_grpid, "bsdgroups"},
832         {Opt_nogrpid, "nogrpid"},
833         {Opt_nogrpid, "sysvgroups"},
834         {Opt_resgid, "resgid=%u"},
835         {Opt_resuid, "resuid=%u"},
836         {Opt_sb, "sb=%u"},
837         {Opt_err_cont, "errors=continue"},
838         {Opt_err_panic, "errors=panic"},
839         {Opt_err_ro, "errors=remount-ro"},
840         {Opt_nouid32, "nouid32"},
841         {Opt_nocheck, "nocheck"},
842         {Opt_nocheck, "check=none"},
843         {Opt_debug, "debug"},
844         {Opt_oldalloc, "oldalloc"},
845         {Opt_orlov, "orlov"},
846         {Opt_user_xattr, "user_xattr"},
847         {Opt_nouser_xattr, "nouser_xattr"},
848         {Opt_acl, "acl"},
849         {Opt_noacl, "noacl"},
850         {Opt_reservation, "reservation"},
851         {Opt_noreservation, "noreservation"},
852         {Opt_noload, "noload"},
853         {Opt_noload, "norecovery"},
854         {Opt_nobh, "nobh"},
855         {Opt_bh, "bh"},
856         {Opt_commit, "commit=%u"},
857         {Opt_journal_update, "journal=update"},
858         {Opt_journal_inum, "journal=%u"},
859         {Opt_journal_dev, "journal_dev=%u"},
860         {Opt_abort, "abort"},
861         {Opt_data_journal, "data=journal"},
862         {Opt_data_ordered, "data=ordered"},
863         {Opt_data_writeback, "data=writeback"},
864         {Opt_data_err_abort, "data_err=abort"},
865         {Opt_data_err_ignore, "data_err=ignore"},
866         {Opt_offusrjquota, "usrjquota="},
867         {Opt_usrjquota, "usrjquota=%s"},
868         {Opt_offgrpjquota, "grpjquota="},
869         {Opt_grpjquota, "grpjquota=%s"},
870         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
871         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
872         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
873         {Opt_grpquota, "grpquota"},
874         {Opt_noquota, "noquota"},
875         {Opt_quota, "quota"},
876         {Opt_usrquota, "usrquota"},
877         {Opt_barrier, "barrier=%u"},
878         {Opt_barrier, "barrier"},
879         {Opt_nobarrier, "nobarrier"},
880         {Opt_resize, "resize"},
881         {Opt_err, NULL},
882 };
883
884 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
885 {
886         ext3_fsblk_t    sb_block;
887         char            *options = (char *) *data;
888
889         if (!options || strncmp(options, "sb=", 3) != 0)
890                 return 1;       /* Default location */
891         options += 3;
892         /*todo: use simple_strtoll with >32bit ext3 */
893         sb_block = simple_strtoul(options, &options, 0);
894         if (*options && *options != ',') {
895                 ext3_msg(sb, "error: invalid sb specification: %s",
896                        (char *) *data);
897                 return 1;
898         }
899         if (*options == ',')
900                 options++;
901         *data = (void *) options;
902         return sb_block;
903 }
904
905 #ifdef CONFIG_QUOTA
906 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
907 {
908         struct ext3_sb_info *sbi = EXT3_SB(sb);
909         char *qname;
910
911         if (sb_any_quota_loaded(sb) &&
912                 !sbi->s_qf_names[qtype]) {
913                 ext3_msg(sb, KERN_ERR,
914                         "Cannot change journaled "
915                         "quota options when quota turned on");
916                 return 0;
917         }
918         qname = match_strdup(args);
919         if (!qname) {
920                 ext3_msg(sb, KERN_ERR,
921                         "Not enough memory for storing quotafile name");
922                 return 0;
923         }
924         if (sbi->s_qf_names[qtype] &&
925                 strcmp(sbi->s_qf_names[qtype], qname)) {
926                 ext3_msg(sb, KERN_ERR,
927                         "%s quota file already specified", QTYPE2NAME(qtype));
928                 kfree(qname);
929                 return 0;
930         }
931         sbi->s_qf_names[qtype] = qname;
932         if (strchr(sbi->s_qf_names[qtype], '/')) {
933                 ext3_msg(sb, KERN_ERR,
934                         "quotafile must be on filesystem root");
935                 kfree(sbi->s_qf_names[qtype]);
936                 sbi->s_qf_names[qtype] = NULL;
937                 return 0;
938         }
939         set_opt(sbi->s_mount_opt, QUOTA);
940         return 1;
941 }
942
943 static int clear_qf_name(struct super_block *sb, int qtype) {
944
945         struct ext3_sb_info *sbi = EXT3_SB(sb);
946
947         if (sb_any_quota_loaded(sb) &&
948                 sbi->s_qf_names[qtype]) {
949                 ext3_msg(sb, KERN_ERR, "Cannot change journaled quota options"
950                         " when quota turned on");
951                 return 0;
952         }
953         /*
954          * The space will be released later when all options are confirmed
955          * to be correct
956          */
957         sbi->s_qf_names[qtype] = NULL;
958         return 1;
959 }
960 #endif
961
962 static int parse_options (char *options, struct super_block *sb,
963                           unsigned int *inum, unsigned long *journal_devnum,
964                           ext3_fsblk_t *n_blocks_count, int is_remount)
965 {
966         struct ext3_sb_info *sbi = EXT3_SB(sb);
967         char * p;
968         substring_t args[MAX_OPT_ARGS];
969         int data_opt = 0;
970         int option;
971 #ifdef CONFIG_QUOTA
972         int qfmt;
973 #endif
974
975         if (!options)
976                 return 1;
977
978         while ((p = strsep (&options, ",")) != NULL) {
979                 int token;
980                 if (!*p)
981                         continue;
982                 /*
983                  * Initialize args struct so we know whether arg was
984                  * found; some options take optional arguments.
985                  */
986                 args[0].to = args[0].from = 0;
987                 token = match_token(p, tokens, args);
988                 switch (token) {
989                 case Opt_bsd_df:
990                         clear_opt (sbi->s_mount_opt, MINIX_DF);
991                         break;
992                 case Opt_minix_df:
993                         set_opt (sbi->s_mount_opt, MINIX_DF);
994                         break;
995                 case Opt_grpid:
996                         set_opt (sbi->s_mount_opt, GRPID);
997                         break;
998                 case Opt_nogrpid:
999                         clear_opt (sbi->s_mount_opt, GRPID);
1000                         break;
1001                 case Opt_resuid:
1002                         if (match_int(&args[0], &option))
1003                                 return 0;
1004                         sbi->s_resuid = option;
1005                         break;
1006                 case Opt_resgid:
1007                         if (match_int(&args[0], &option))
1008                                 return 0;
1009                         sbi->s_resgid = option;
1010                         break;
1011                 case Opt_sb:
1012                         /* handled by get_sb_block() instead of here */
1013                         /* *sb_block = match_int(&args[0]); */
1014                         break;
1015                 case Opt_err_panic:
1016                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1017                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1018                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1019                         break;
1020                 case Opt_err_ro:
1021                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1022                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1023                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1024                         break;
1025                 case Opt_err_cont:
1026                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1027                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1028                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1029                         break;
1030                 case Opt_nouid32:
1031                         set_opt (sbi->s_mount_opt, NO_UID32);
1032                         break;
1033                 case Opt_nocheck:
1034                         clear_opt (sbi->s_mount_opt, CHECK);
1035                         break;
1036                 case Opt_debug:
1037                         set_opt (sbi->s_mount_opt, DEBUG);
1038                         break;
1039                 case Opt_oldalloc:
1040                         set_opt (sbi->s_mount_opt, OLDALLOC);
1041                         break;
1042                 case Opt_orlov:
1043                         clear_opt (sbi->s_mount_opt, OLDALLOC);
1044                         break;
1045 #ifdef CONFIG_EXT3_FS_XATTR
1046                 case Opt_user_xattr:
1047                         set_opt (sbi->s_mount_opt, XATTR_USER);
1048                         break;
1049                 case Opt_nouser_xattr:
1050                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1051                         break;
1052 #else
1053                 case Opt_user_xattr:
1054                 case Opt_nouser_xattr:
1055                         ext3_msg(sb, KERN_INFO,
1056                                 "(no)user_xattr options not supported");
1057                         break;
1058 #endif
1059 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1060                 case Opt_acl:
1061                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1062                         break;
1063                 case Opt_noacl:
1064                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1065                         break;
1066 #else
1067                 case Opt_acl:
1068                 case Opt_noacl:
1069                         ext3_msg(sb, KERN_INFO,
1070                                 "(no)acl options not supported");
1071                         break;
1072 #endif
1073                 case Opt_reservation:
1074                         set_opt(sbi->s_mount_opt, RESERVATION);
1075                         break;
1076                 case Opt_noreservation:
1077                         clear_opt(sbi->s_mount_opt, RESERVATION);
1078                         break;
1079                 case Opt_journal_update:
1080                         /* @@@ FIXME */
1081                         /* Eventually we will want to be able to create
1082                            a journal file here.  For now, only allow the
1083                            user to specify an existing inode to be the
1084                            journal file. */
1085                         if (is_remount) {
1086                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1087                                         "journal on remount");
1088                                 return 0;
1089                         }
1090                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1091                         break;
1092                 case Opt_journal_inum:
1093                         if (is_remount) {
1094                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1095                                        "journal on remount");
1096                                 return 0;
1097                         }
1098                         if (match_int(&args[0], &option))
1099                                 return 0;
1100                         *inum = option;
1101                         break;
1102                 case Opt_journal_dev:
1103                         if (is_remount) {
1104                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1105                                        "journal on remount");
1106                                 return 0;
1107                         }
1108                         if (match_int(&args[0], &option))
1109                                 return 0;
1110                         *journal_devnum = option;
1111                         break;
1112                 case Opt_noload:
1113                         set_opt (sbi->s_mount_opt, NOLOAD);
1114                         break;
1115                 case Opt_commit:
1116                         if (match_int(&args[0], &option))
1117                                 return 0;
1118                         if (option < 0)
1119                                 return 0;
1120                         if (option == 0)
1121                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1122                         sbi->s_commit_interval = HZ * option;
1123                         break;
1124                 case Opt_data_journal:
1125                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1126                         goto datacheck;
1127                 case Opt_data_ordered:
1128                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1129                         goto datacheck;
1130                 case Opt_data_writeback:
1131                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1132                 datacheck:
1133                         if (is_remount) {
1134                                 if (test_opt(sb, DATA_FLAGS) == data_opt)
1135                                         break;
1136                                 ext3_msg(sb, KERN_ERR,
1137                                         "error: cannot change "
1138                                         "data mode on remount. The filesystem "
1139                                         "is mounted in data=%s mode and you "
1140                                         "try to remount it in data=%s mode.",
1141                                         data_mode_string(test_opt(sb,
1142                                                         DATA_FLAGS)),
1143                                         data_mode_string(data_opt));
1144                                 return 0;
1145                         } else {
1146                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1147                                 sbi->s_mount_opt |= data_opt;
1148                         }
1149                         break;
1150                 case Opt_data_err_abort:
1151                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1152                         break;
1153                 case Opt_data_err_ignore:
1154                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1155                         break;
1156 #ifdef CONFIG_QUOTA
1157                 case Opt_usrjquota:
1158                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1159                                 return 0;
1160                         break;
1161                 case Opt_grpjquota:
1162                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1163                                 return 0;
1164                         break;
1165                 case Opt_offusrjquota:
1166                         if (!clear_qf_name(sb, USRQUOTA))
1167                                 return 0;
1168                         break;
1169                 case Opt_offgrpjquota:
1170                         if (!clear_qf_name(sb, GRPQUOTA))
1171                                 return 0;
1172                         break;
1173                 case Opt_jqfmt_vfsold:
1174                         qfmt = QFMT_VFS_OLD;
1175                         goto set_qf_format;
1176                 case Opt_jqfmt_vfsv0:
1177                         qfmt = QFMT_VFS_V0;
1178                         goto set_qf_format;
1179                 case Opt_jqfmt_vfsv1:
1180                         qfmt = QFMT_VFS_V1;
1181 set_qf_format:
1182                         if (sb_any_quota_loaded(sb) &&
1183                             sbi->s_jquota_fmt != qfmt) {
1184                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1185                                         "journaled quota options when "
1186                                         "quota turned on.");
1187                                 return 0;
1188                         }
1189                         sbi->s_jquota_fmt = qfmt;
1190                         break;
1191                 case Opt_quota:
1192                 case Opt_usrquota:
1193                         set_opt(sbi->s_mount_opt, QUOTA);
1194                         set_opt(sbi->s_mount_opt, USRQUOTA);
1195                         break;
1196                 case Opt_grpquota:
1197                         set_opt(sbi->s_mount_opt, QUOTA);
1198                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1199                         break;
1200                 case Opt_noquota:
1201                         if (sb_any_quota_loaded(sb)) {
1202                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1203                                         "quota options when quota turned on.");
1204                                 return 0;
1205                         }
1206                         clear_opt(sbi->s_mount_opt, QUOTA);
1207                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1208                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1209                         break;
1210 #else
1211                 case Opt_quota:
1212                 case Opt_usrquota:
1213                 case Opt_grpquota:
1214                         ext3_msg(sb, KERN_ERR,
1215                                 "error: quota options not supported.");
1216                         break;
1217                 case Opt_usrjquota:
1218                 case Opt_grpjquota:
1219                 case Opt_offusrjquota:
1220                 case Opt_offgrpjquota:
1221                 case Opt_jqfmt_vfsold:
1222                 case Opt_jqfmt_vfsv0:
1223                 case Opt_jqfmt_vfsv1:
1224                         ext3_msg(sb, KERN_ERR,
1225                                 "error: journaled quota options not "
1226                                 "supported.");
1227                         break;
1228                 case Opt_noquota:
1229                         break;
1230 #endif
1231                 case Opt_abort:
1232                         set_opt(sbi->s_mount_opt, ABORT);
1233                         break;
1234                 case Opt_nobarrier:
1235                         clear_opt(sbi->s_mount_opt, BARRIER);
1236                         break;
1237                 case Opt_barrier:
1238                         if (args[0].from) {
1239                                 if (match_int(&args[0], &option))
1240                                         return 0;
1241                         } else
1242                                 option = 1;     /* No argument, default to 1 */
1243                         if (option)
1244                                 set_opt(sbi->s_mount_opt, BARRIER);
1245                         else
1246                                 clear_opt(sbi->s_mount_opt, BARRIER);
1247                         break;
1248                 case Opt_ignore:
1249                         break;
1250                 case Opt_resize:
1251                         if (!is_remount) {
1252                                 ext3_msg(sb, KERN_ERR,
1253                                         "error: resize option only available "
1254                                         "for remount");
1255                                 return 0;
1256                         }
1257                         if (match_int(&args[0], &option) != 0)
1258                                 return 0;
1259                         *n_blocks_count = option;
1260                         break;
1261                 case Opt_nobh:
1262                         ext3_msg(sb, KERN_WARNING,
1263                                 "warning: ignoring deprecated nobh option");
1264                         break;
1265                 case Opt_bh:
1266                         ext3_msg(sb, KERN_WARNING,
1267                                 "warning: ignoring deprecated bh option");
1268                         break;
1269                 default:
1270                         ext3_msg(sb, KERN_ERR,
1271                                 "error: unrecognized mount option \"%s\" "
1272                                 "or missing value", p);
1273                         return 0;
1274                 }
1275         }
1276 #ifdef CONFIG_QUOTA
1277         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1278                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1279                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1280                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1281                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1282
1283                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1284                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1285                                         "format mixing.");
1286                         return 0;
1287                 }
1288
1289                 if (!sbi->s_jquota_fmt) {
1290                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1291                                         "not specified.");
1292                         return 0;
1293                 }
1294         } else {
1295                 if (sbi->s_jquota_fmt) {
1296                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1297                                         "specified with no journaling "
1298                                         "enabled.");
1299                         return 0;
1300                 }
1301         }
1302 #endif
1303         return 1;
1304 }
1305
1306 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1307                             int read_only)
1308 {
1309         struct ext3_sb_info *sbi = EXT3_SB(sb);
1310         int res = 0;
1311
1312         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1313                 ext3_msg(sb, KERN_ERR,
1314                         "error: revision level too high, "
1315                         "forcing read-only mode");
1316                 res = MS_RDONLY;
1317         }
1318         if (read_only)
1319                 return res;
1320         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1321                 ext3_msg(sb, KERN_WARNING,
1322                         "warning: mounting unchecked fs, "
1323                         "running e2fsck is recommended");
1324         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1325                 ext3_msg(sb, KERN_WARNING,
1326                         "warning: mounting fs with errors, "
1327                         "running e2fsck is recommended");
1328         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1329                  le16_to_cpu(es->s_mnt_count) >=
1330                         le16_to_cpu(es->s_max_mnt_count))
1331                 ext3_msg(sb, KERN_WARNING,
1332                         "warning: maximal mount count reached, "
1333                         "running e2fsck is recommended");
1334         else if (le32_to_cpu(es->s_checkinterval) &&
1335                 (le32_to_cpu(es->s_lastcheck) +
1336                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1337                 ext3_msg(sb, KERN_WARNING,
1338                         "warning: checktime reached, "
1339                         "running e2fsck is recommended");
1340 #if 0
1341                 /* @@@ We _will_ want to clear the valid bit if we find
1342                    inconsistencies, to force a fsck at reboot.  But for
1343                    a plain journaled filesystem we can keep it set as
1344                    valid forever! :) */
1345         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1346 #endif
1347         if (!le16_to_cpu(es->s_max_mnt_count))
1348                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1349         le16_add_cpu(&es->s_mnt_count, 1);
1350         es->s_mtime = cpu_to_le32(get_seconds());
1351         ext3_update_dynamic_rev(sb);
1352         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1353
1354         ext3_commit_super(sb, es, 1);
1355         if (test_opt(sb, DEBUG))
1356                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1357                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1358                         sb->s_blocksize,
1359                         sbi->s_groups_count,
1360                         EXT3_BLOCKS_PER_GROUP(sb),
1361                         EXT3_INODES_PER_GROUP(sb),
1362                         sbi->s_mount_opt);
1363
1364         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1365                 char b[BDEVNAME_SIZE];
1366                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1367                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1368         } else {
1369                 ext3_msg(sb, KERN_INFO, "using internal journal");
1370         }
1371         cleancache_init_fs(sb);
1372         return res;
1373 }
1374
1375 /* Called at mount-time, super-block is locked */
1376 static int ext3_check_descriptors(struct super_block *sb)
1377 {
1378         struct ext3_sb_info *sbi = EXT3_SB(sb);
1379         int i;
1380
1381         ext3_debug ("Checking group descriptors");
1382
1383         for (i = 0; i < sbi->s_groups_count; i++) {
1384                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1385                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1386                 ext3_fsblk_t last_block;
1387
1388                 if (i == sbi->s_groups_count - 1)
1389                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1390                 else
1391                         last_block = first_block +
1392                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1393
1394                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1395                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1396                 {
1397                         ext3_error (sb, "ext3_check_descriptors",
1398                                     "Block bitmap for group %d"
1399                                     " not in group (block %lu)!",
1400                                     i, (unsigned long)
1401                                         le32_to_cpu(gdp->bg_block_bitmap));
1402                         return 0;
1403                 }
1404                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1405                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1406                 {
1407                         ext3_error (sb, "ext3_check_descriptors",
1408                                     "Inode bitmap for group %d"
1409                                     " not in group (block %lu)!",
1410                                     i, (unsigned long)
1411                                         le32_to_cpu(gdp->bg_inode_bitmap));
1412                         return 0;
1413                 }
1414                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1415                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1416                     last_block)
1417                 {
1418                         ext3_error (sb, "ext3_check_descriptors",
1419                                     "Inode table for group %d"
1420                                     " not in group (block %lu)!",
1421                                     i, (unsigned long)
1422                                         le32_to_cpu(gdp->bg_inode_table));
1423                         return 0;
1424                 }
1425         }
1426
1427         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1428         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1429         return 1;
1430 }
1431
1432
1433 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1434  * the superblock) which were deleted from all directories, but held open by
1435  * a process at the time of a crash.  We walk the list and try to delete these
1436  * inodes at recovery time (only with a read-write filesystem).
1437  *
1438  * In order to keep the orphan inode chain consistent during traversal (in
1439  * case of crash during recovery), we link each inode into the superblock
1440  * orphan list_head and handle it the same way as an inode deletion during
1441  * normal operation (which journals the operations for us).
1442  *
1443  * We only do an iget() and an iput() on each inode, which is very safe if we
1444  * accidentally point at an in-use or already deleted inode.  The worst that
1445  * can happen in this case is that we get a "bit already cleared" message from
1446  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1447  * e2fsck was run on this filesystem, and it must have already done the orphan
1448  * inode cleanup for us, so we can safely abort without any further action.
1449  */
1450 static void ext3_orphan_cleanup (struct super_block * sb,
1451                                  struct ext3_super_block * es)
1452 {
1453         unsigned int s_flags = sb->s_flags;
1454         int nr_orphans = 0, nr_truncates = 0;
1455 #ifdef CONFIG_QUOTA
1456         int i;
1457 #endif
1458         if (!es->s_last_orphan) {
1459                 jbd_debug(4, "no orphan inodes to clean up\n");
1460                 return;
1461         }
1462
1463         if (bdev_read_only(sb->s_bdev)) {
1464                 ext3_msg(sb, KERN_ERR, "error: write access "
1465                         "unavailable, skipping orphan cleanup.");
1466                 return;
1467         }
1468
1469         /* Check if feature set allows readwrite operations */
1470         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP)) {
1471                 ext3_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
1472                          "unknown ROCOMPAT features");
1473                 return;
1474         }
1475
1476         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1477                 if (es->s_last_orphan)
1478                         jbd_debug(1, "Errors on filesystem, "
1479                                   "clearing orphan list.\n");
1480                 es->s_last_orphan = 0;
1481                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1482                 return;
1483         }
1484
1485         if (s_flags & MS_RDONLY) {
1486                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1487                 sb->s_flags &= ~MS_RDONLY;
1488         }
1489 #ifdef CONFIG_QUOTA
1490         /* Needed for iput() to work correctly and not trash data */
1491         sb->s_flags |= MS_ACTIVE;
1492         /* Turn on quotas so that they are updated correctly */
1493         for (i = 0; i < MAXQUOTAS; i++) {
1494                 if (EXT3_SB(sb)->s_qf_names[i]) {
1495                         int ret = ext3_quota_on_mount(sb, i);
1496                         if (ret < 0)
1497                                 ext3_msg(sb, KERN_ERR,
1498                                         "error: cannot turn on journaled "
1499                                         "quota: %d", ret);
1500                 }
1501         }
1502 #endif
1503
1504         while (es->s_last_orphan) {
1505                 struct inode *inode;
1506
1507                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1508                 if (IS_ERR(inode)) {
1509                         es->s_last_orphan = 0;
1510                         break;
1511                 }
1512
1513                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1514                 dquot_initialize(inode);
1515                 if (inode->i_nlink) {
1516                         printk(KERN_DEBUG
1517                                 "%s: truncating inode %lu to %Ld bytes\n",
1518                                 __func__, inode->i_ino, inode->i_size);
1519                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1520                                   inode->i_ino, inode->i_size);
1521                         ext3_truncate(inode);
1522                         nr_truncates++;
1523                 } else {
1524                         printk(KERN_DEBUG
1525                                 "%s: deleting unreferenced inode %lu\n",
1526                                 __func__, inode->i_ino);
1527                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1528                                   inode->i_ino);
1529                         nr_orphans++;
1530                 }
1531                 iput(inode);  /* The delete magic happens here! */
1532         }
1533
1534 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1535
1536         if (nr_orphans)
1537                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1538                        PLURAL(nr_orphans));
1539         if (nr_truncates)
1540                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1541                        PLURAL(nr_truncates));
1542 #ifdef CONFIG_QUOTA
1543         /* Turn quotas off */
1544         for (i = 0; i < MAXQUOTAS; i++) {
1545                 if (sb_dqopt(sb)->files[i])
1546                         dquot_quota_off(sb, i);
1547         }
1548 #endif
1549         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1550 }
1551
1552 /*
1553  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1554  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1555  * We need to be 1 filesystem block less than the 2^32 sector limit.
1556  */
1557 static loff_t ext3_max_size(int bits)
1558 {
1559         loff_t res = EXT3_NDIR_BLOCKS;
1560         int meta_blocks;
1561         loff_t upper_limit;
1562
1563         /* This is calculated to be the largest file size for a
1564          * dense, file such that the total number of
1565          * sectors in the file, including data and all indirect blocks,
1566          * does not exceed 2^32 -1
1567          * __u32 i_blocks representing the total number of
1568          * 512 bytes blocks of the file
1569          */
1570         upper_limit = (1LL << 32) - 1;
1571
1572         /* total blocks in file system block size */
1573         upper_limit >>= (bits - 9);
1574
1575
1576         /* indirect blocks */
1577         meta_blocks = 1;
1578         /* double indirect blocks */
1579         meta_blocks += 1 + (1LL << (bits-2));
1580         /* tripple indirect blocks */
1581         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1582
1583         upper_limit -= meta_blocks;
1584         upper_limit <<= bits;
1585
1586         res += 1LL << (bits-2);
1587         res += 1LL << (2*(bits-2));
1588         res += 1LL << (3*(bits-2));
1589         res <<= bits;
1590         if (res > upper_limit)
1591                 res = upper_limit;
1592
1593         if (res > MAX_LFS_FILESIZE)
1594                 res = MAX_LFS_FILESIZE;
1595
1596         return res;
1597 }
1598
1599 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1600                                     ext3_fsblk_t logic_sb_block,
1601                                     int nr)
1602 {
1603         struct ext3_sb_info *sbi = EXT3_SB(sb);
1604         unsigned long bg, first_meta_bg;
1605         int has_super = 0;
1606
1607         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1608
1609         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1610             nr < first_meta_bg)
1611                 return (logic_sb_block + nr + 1);
1612         bg = sbi->s_desc_per_block * nr;
1613         if (ext3_bg_has_super(sb, bg))
1614                 has_super = 1;
1615         return (has_super + ext3_group_first_block_no(sb, bg));
1616 }
1617
1618
1619 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1620 {
1621         struct buffer_head * bh;
1622         struct ext3_super_block *es = NULL;
1623         struct ext3_sb_info *sbi;
1624         ext3_fsblk_t block;
1625         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1626         ext3_fsblk_t logic_sb_block;
1627         unsigned long offset = 0;
1628         unsigned int journal_inum = 0;
1629         unsigned long journal_devnum = 0;
1630         unsigned long def_mount_opts;
1631         struct inode *root;
1632         int blocksize;
1633         int hblock;
1634         int db_count;
1635         int i;
1636         int needs_recovery;
1637         int ret = -EINVAL;
1638         __le32 features;
1639         int err;
1640
1641         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1642         if (!sbi)
1643                 return -ENOMEM;
1644
1645         sbi->s_blockgroup_lock =
1646                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1647         if (!sbi->s_blockgroup_lock) {
1648                 kfree(sbi);
1649                 return -ENOMEM;
1650         }
1651         sb->s_fs_info = sbi;
1652         sbi->s_mount_opt = 0;
1653         sbi->s_resuid = EXT3_DEF_RESUID;
1654         sbi->s_resgid = EXT3_DEF_RESGID;
1655         sbi->s_sb_block = sb_block;
1656
1657         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1658         if (!blocksize) {
1659                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1660                 goto out_fail;
1661         }
1662
1663         /*
1664          * The ext3 superblock will not be buffer aligned for other than 1kB
1665          * block sizes.  We need to calculate the offset from buffer start.
1666          */
1667         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1668                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1669                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1670         } else {
1671                 logic_sb_block = sb_block;
1672         }
1673
1674         if (!(bh = sb_bread(sb, logic_sb_block))) {
1675                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1676                 goto out_fail;
1677         }
1678         /*
1679          * Note: s_es must be initialized as soon as possible because
1680          *       some ext3 macro-instructions depend on its value
1681          */
1682         es = (struct ext3_super_block *) (bh->b_data + offset);
1683         sbi->s_es = es;
1684         sb->s_magic = le16_to_cpu(es->s_magic);
1685         if (sb->s_magic != EXT3_SUPER_MAGIC)
1686                 goto cantfind_ext3;
1687
1688         /* Set defaults before we parse the mount options */
1689         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1690         if (def_mount_opts & EXT3_DEFM_DEBUG)
1691                 set_opt(sbi->s_mount_opt, DEBUG);
1692         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1693                 set_opt(sbi->s_mount_opt, GRPID);
1694         if (def_mount_opts & EXT3_DEFM_UID16)
1695                 set_opt(sbi->s_mount_opt, NO_UID32);
1696 #ifdef CONFIG_EXT3_FS_XATTR
1697         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1698                 set_opt(sbi->s_mount_opt, XATTR_USER);
1699 #endif
1700 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1701         if (def_mount_opts & EXT3_DEFM_ACL)
1702                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1703 #endif
1704         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1705                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1706         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1707                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1708         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1709                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1710
1711         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1712                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1713         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1714                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1715         else
1716                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1717
1718         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1719         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1720
1721         set_opt(sbi->s_mount_opt, RESERVATION);
1722
1723         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1724                             NULL, 0))
1725                 goto failed_mount;
1726
1727         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1728                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1729
1730         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1731             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1732              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1733              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1734                 ext3_msg(sb, KERN_WARNING,
1735                         "warning: feature flags set on rev 0 fs, "
1736                         "running e2fsck is recommended");
1737         /*
1738          * Check feature flags regardless of the revision level, since we
1739          * previously didn't change the revision level when setting the flags,
1740          * so there is a chance incompat flags are set on a rev 0 filesystem.
1741          */
1742         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1743         if (features) {
1744                 ext3_msg(sb, KERN_ERR,
1745                         "error: couldn't mount because of unsupported "
1746                         "optional features (%x)", le32_to_cpu(features));
1747                 goto failed_mount;
1748         }
1749         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1750         if (!(sb->s_flags & MS_RDONLY) && features) {
1751                 ext3_msg(sb, KERN_ERR,
1752                         "error: couldn't mount RDWR because of unsupported "
1753                         "optional features (%x)", le32_to_cpu(features));
1754                 goto failed_mount;
1755         }
1756         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1757
1758         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1759             blocksize > EXT3_MAX_BLOCK_SIZE) {
1760                 ext3_msg(sb, KERN_ERR,
1761                         "error: couldn't mount because of unsupported "
1762                         "filesystem blocksize %d", blocksize);
1763                 goto failed_mount;
1764         }
1765
1766         hblock = bdev_logical_block_size(sb->s_bdev);
1767         if (sb->s_blocksize != blocksize) {
1768                 /*
1769                  * Make sure the blocksize for the filesystem is larger
1770                  * than the hardware sectorsize for the machine.
1771                  */
1772                 if (blocksize < hblock) {
1773                         ext3_msg(sb, KERN_ERR,
1774                                 "error: fsblocksize %d too small for "
1775                                 "hardware sectorsize %d", blocksize, hblock);
1776                         goto failed_mount;
1777                 }
1778
1779                 brelse (bh);
1780                 if (!sb_set_blocksize(sb, blocksize)) {
1781                         ext3_msg(sb, KERN_ERR,
1782                                 "error: bad blocksize %d", blocksize);
1783                         goto out_fail;
1784                 }
1785                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1786                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1787                 bh = sb_bread(sb, logic_sb_block);
1788                 if (!bh) {
1789                         ext3_msg(sb, KERN_ERR,
1790                                "error: can't read superblock on 2nd try");
1791                         goto failed_mount;
1792                 }
1793                 es = (struct ext3_super_block *)(bh->b_data + offset);
1794                 sbi->s_es = es;
1795                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1796                         ext3_msg(sb, KERN_ERR,
1797                                 "error: magic mismatch");
1798                         goto failed_mount;
1799                 }
1800         }
1801
1802         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1803
1804         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1805                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1806                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1807         } else {
1808                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1809                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1810                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1811                     (!is_power_of_2(sbi->s_inode_size)) ||
1812                     (sbi->s_inode_size > blocksize)) {
1813                         ext3_msg(sb, KERN_ERR,
1814                                 "error: unsupported inode size: %d",
1815                                 sbi->s_inode_size);
1816                         goto failed_mount;
1817                 }
1818         }
1819         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1820                                    le32_to_cpu(es->s_log_frag_size);
1821         if (blocksize != sbi->s_frag_size) {
1822                 ext3_msg(sb, KERN_ERR,
1823                        "error: fragsize %lu != blocksize %u (unsupported)",
1824                        sbi->s_frag_size, blocksize);
1825                 goto failed_mount;
1826         }
1827         sbi->s_frags_per_block = 1;
1828         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1829         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1830         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1831         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1832                 goto cantfind_ext3;
1833         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1834         if (sbi->s_inodes_per_block == 0)
1835                 goto cantfind_ext3;
1836         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1837                                         sbi->s_inodes_per_block;
1838         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1839         sbi->s_sbh = bh;
1840         sbi->s_mount_state = le16_to_cpu(es->s_state);
1841         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1842         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1843         for (i=0; i < 4; i++)
1844                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1845         sbi->s_def_hash_version = es->s_def_hash_version;
1846         i = le32_to_cpu(es->s_flags);
1847         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1848                 sbi->s_hash_unsigned = 3;
1849         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1850 #ifdef __CHAR_UNSIGNED__
1851                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1852                 sbi->s_hash_unsigned = 3;
1853 #else
1854                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1855 #endif
1856         }
1857
1858         if (sbi->s_blocks_per_group > blocksize * 8) {
1859                 ext3_msg(sb, KERN_ERR,
1860                         "#blocks per group too big: %lu",
1861                         sbi->s_blocks_per_group);
1862                 goto failed_mount;
1863         }
1864         if (sbi->s_frags_per_group > blocksize * 8) {
1865                 ext3_msg(sb, KERN_ERR,
1866                         "error: #fragments per group too big: %lu",
1867                         sbi->s_frags_per_group);
1868                 goto failed_mount;
1869         }
1870         if (sbi->s_inodes_per_group > blocksize * 8) {
1871                 ext3_msg(sb, KERN_ERR,
1872                         "error: #inodes per group too big: %lu",
1873                         sbi->s_inodes_per_group);
1874                 goto failed_mount;
1875         }
1876
1877         err = generic_check_addressable(sb->s_blocksize_bits,
1878                                         le32_to_cpu(es->s_blocks_count));
1879         if (err) {
1880                 ext3_msg(sb, KERN_ERR,
1881                         "error: filesystem is too large to mount safely");
1882                 if (sizeof(sector_t) < 8)
1883                         ext3_msg(sb, KERN_ERR,
1884                                 "error: CONFIG_LBDAF not enabled");
1885                 ret = err;
1886                 goto failed_mount;
1887         }
1888
1889         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1890                 goto cantfind_ext3;
1891         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1892                                le32_to_cpu(es->s_first_data_block) - 1)
1893                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1894         db_count = DIV_ROUND_UP(sbi->s_groups_count, EXT3_DESC_PER_BLOCK(sb));
1895         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1896                                     GFP_KERNEL);
1897         if (sbi->s_group_desc == NULL) {
1898                 ext3_msg(sb, KERN_ERR,
1899                         "error: not enough memory");
1900                 ret = -ENOMEM;
1901                 goto failed_mount;
1902         }
1903
1904         bgl_lock_init(sbi->s_blockgroup_lock);
1905
1906         for (i = 0; i < db_count; i++) {
1907                 block = descriptor_loc(sb, logic_sb_block, i);
1908                 sbi->s_group_desc[i] = sb_bread(sb, block);
1909                 if (!sbi->s_group_desc[i]) {
1910                         ext3_msg(sb, KERN_ERR,
1911                                 "error: can't read group descriptor %d", i);
1912                         db_count = i;
1913                         goto failed_mount2;
1914                 }
1915         }
1916         if (!ext3_check_descriptors (sb)) {
1917                 ext3_msg(sb, KERN_ERR,
1918                         "error: group descriptors corrupted");
1919                 goto failed_mount2;
1920         }
1921         sbi->s_gdb_count = db_count;
1922         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1923         spin_lock_init(&sbi->s_next_gen_lock);
1924
1925         /* per fileystem reservation list head & lock */
1926         spin_lock_init(&sbi->s_rsv_window_lock);
1927         sbi->s_rsv_window_root = RB_ROOT;
1928         /* Add a single, static dummy reservation to the start of the
1929          * reservation window list --- it gives us a placeholder for
1930          * append-at-start-of-list which makes the allocation logic
1931          * _much_ simpler. */
1932         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1933         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1934         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1935         sbi->s_rsv_window_head.rsv_goal_size = 0;
1936         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1937
1938         /*
1939          * set up enough so that it can read an inode
1940          */
1941         sb->s_op = &ext3_sops;
1942         sb->s_export_op = &ext3_export_ops;
1943         sb->s_xattr = ext3_xattr_handlers;
1944 #ifdef CONFIG_QUOTA
1945         sb->s_qcop = &ext3_qctl_operations;
1946         sb->dq_op = &ext3_quota_operations;
1947 #endif
1948         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
1949         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1950         mutex_init(&sbi->s_orphan_lock);
1951         mutex_init(&sbi->s_resize_lock);
1952
1953         sb->s_root = NULL;
1954
1955         needs_recovery = (es->s_last_orphan != 0 ||
1956                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1957                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1958
1959         /*
1960          * The first inode we look at is the journal inode.  Don't try
1961          * root first: it may be modified in the journal!
1962          */
1963         if (!test_opt(sb, NOLOAD) &&
1964             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1965                 if (ext3_load_journal(sb, es, journal_devnum))
1966                         goto failed_mount2;
1967         } else if (journal_inum) {
1968                 if (ext3_create_journal(sb, es, journal_inum))
1969                         goto failed_mount2;
1970         } else {
1971                 if (!silent)
1972                         ext3_msg(sb, KERN_ERR,
1973                                 "error: no journal found. "
1974                                 "mounting ext3 over ext2?");
1975                 goto failed_mount2;
1976         }
1977         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1978                         ext3_count_free_blocks(sb));
1979         if (!err) {
1980                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1981                                 ext3_count_free_inodes(sb));
1982         }
1983         if (!err) {
1984                 err = percpu_counter_init(&sbi->s_dirs_counter,
1985                                 ext3_count_dirs(sb));
1986         }
1987         if (err) {
1988                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
1989                 ret = err;
1990                 goto failed_mount3;
1991         }
1992
1993         /* We have now updated the journal if required, so we can
1994          * validate the data journaling mode. */
1995         switch (test_opt(sb, DATA_FLAGS)) {
1996         case 0:
1997                 /* No mode set, assume a default based on the journal
1998                    capabilities: ORDERED_DATA if the journal can
1999                    cope, else JOURNAL_DATA */
2000                 if (journal_check_available_features
2001                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
2002                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
2003                 else
2004                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2005                 break;
2006
2007         case EXT3_MOUNT_ORDERED_DATA:
2008         case EXT3_MOUNT_WRITEBACK_DATA:
2009                 if (!journal_check_available_features
2010                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
2011                         ext3_msg(sb, KERN_ERR,
2012                                 "error: journal does not support "
2013                                 "requested data journaling mode");
2014                         goto failed_mount3;
2015                 }
2016         default:
2017                 break;
2018         }
2019
2020         /*
2021          * The journal_load will have done any necessary log recovery,
2022          * so we can safely mount the rest of the filesystem now.
2023          */
2024
2025         root = ext3_iget(sb, EXT3_ROOT_INO);
2026         if (IS_ERR(root)) {
2027                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
2028                 ret = PTR_ERR(root);
2029                 goto failed_mount3;
2030         }
2031         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2032                 iput(root);
2033                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2034                 goto failed_mount3;
2035         }
2036         sb->s_root = d_alloc_root(root);
2037         if (!sb->s_root) {
2038                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2039                 iput(root);
2040                 ret = -ENOMEM;
2041                 goto failed_mount3;
2042         }
2043
2044         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2045
2046         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2047         ext3_orphan_cleanup(sb, es);
2048         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2049         if (needs_recovery)
2050                 ext3_msg(sb, KERN_INFO, "recovery complete");
2051         ext3_mark_recovery_complete(sb, es);
2052         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2053                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2054                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2055                 "writeback");
2056
2057         return 0;
2058
2059 cantfind_ext3:
2060         if (!silent)
2061                 ext3_msg(sb, KERN_INFO,
2062                         "error: can't find ext3 filesystem on dev %s.",
2063                        sb->s_id);
2064         goto failed_mount;
2065
2066 failed_mount3:
2067         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2068         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2069         percpu_counter_destroy(&sbi->s_dirs_counter);
2070         journal_destroy(sbi->s_journal);
2071 failed_mount2:
2072         for (i = 0; i < db_count; i++)
2073                 brelse(sbi->s_group_desc[i]);
2074         kfree(sbi->s_group_desc);
2075 failed_mount:
2076 #ifdef CONFIG_QUOTA
2077         for (i = 0; i < MAXQUOTAS; i++)
2078                 kfree(sbi->s_qf_names[i]);
2079 #endif
2080         ext3_blkdev_remove(sbi);
2081         brelse(bh);
2082 out_fail:
2083         sb->s_fs_info = NULL;
2084         kfree(sbi->s_blockgroup_lock);
2085         kfree(sbi);
2086         return ret;
2087 }
2088
2089 /*
2090  * Setup any per-fs journal parameters now.  We'll do this both on
2091  * initial mount, once the journal has been initialised but before we've
2092  * done any recovery; and again on any subsequent remount.
2093  */
2094 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2095 {
2096         struct ext3_sb_info *sbi = EXT3_SB(sb);
2097
2098         if (sbi->s_commit_interval)
2099                 journal->j_commit_interval = sbi->s_commit_interval;
2100         /* We could also set up an ext3-specific default for the commit
2101          * interval here, but for now we'll just fall back to the jbd
2102          * default. */
2103
2104         spin_lock(&journal->j_state_lock);
2105         if (test_opt(sb, BARRIER))
2106                 journal->j_flags |= JFS_BARRIER;
2107         else
2108                 journal->j_flags &= ~JFS_BARRIER;
2109         if (test_opt(sb, DATA_ERR_ABORT))
2110                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2111         else
2112                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2113         spin_unlock(&journal->j_state_lock);
2114 }
2115
2116 static journal_t *ext3_get_journal(struct super_block *sb,
2117                                    unsigned int journal_inum)
2118 {
2119         struct inode *journal_inode;
2120         journal_t *journal;
2121
2122         /* First, test for the existence of a valid inode on disk.  Bad
2123          * things happen if we iget() an unused inode, as the subsequent
2124          * iput() will try to delete it. */
2125
2126         journal_inode = ext3_iget(sb, journal_inum);
2127         if (IS_ERR(journal_inode)) {
2128                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2129                 return NULL;
2130         }
2131         if (!journal_inode->i_nlink) {
2132                 make_bad_inode(journal_inode);
2133                 iput(journal_inode);
2134                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2135                 return NULL;
2136         }
2137
2138         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2139                   journal_inode, journal_inode->i_size);
2140         if (!S_ISREG(journal_inode->i_mode)) {
2141                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2142                 iput(journal_inode);
2143                 return NULL;
2144         }
2145
2146         journal = journal_init_inode(journal_inode);
2147         if (!journal) {
2148                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2149                 iput(journal_inode);
2150                 return NULL;
2151         }
2152         journal->j_private = sb;
2153         ext3_init_journal_params(sb, journal);
2154         return journal;
2155 }
2156
2157 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2158                                        dev_t j_dev)
2159 {
2160         struct buffer_head * bh;
2161         journal_t *journal;
2162         ext3_fsblk_t start;
2163         ext3_fsblk_t len;
2164         int hblock, blocksize;
2165         ext3_fsblk_t sb_block;
2166         unsigned long offset;
2167         struct ext3_super_block * es;
2168         struct block_device *bdev;
2169
2170         bdev = ext3_blkdev_get(j_dev, sb);
2171         if (bdev == NULL)
2172                 return NULL;
2173
2174         blocksize = sb->s_blocksize;
2175         hblock = bdev_logical_block_size(bdev);
2176         if (blocksize < hblock) {
2177                 ext3_msg(sb, KERN_ERR,
2178                         "error: blocksize too small for journal device");
2179                 goto out_bdev;
2180         }
2181
2182         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2183         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2184         set_blocksize(bdev, blocksize);
2185         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2186                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2187                         "external journal");
2188                 goto out_bdev;
2189         }
2190
2191         es = (struct ext3_super_block *) (bh->b_data + offset);
2192         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2193             !(le32_to_cpu(es->s_feature_incompat) &
2194               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2195                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2196                         "bad superblock");
2197                 brelse(bh);
2198                 goto out_bdev;
2199         }
2200
2201         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2202                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2203                 brelse(bh);
2204                 goto out_bdev;
2205         }
2206
2207         len = le32_to_cpu(es->s_blocks_count);
2208         start = sb_block + 1;
2209         brelse(bh);     /* we're done with the superblock */
2210
2211         journal = journal_init_dev(bdev, sb->s_bdev,
2212                                         start, len, blocksize);
2213         if (!journal) {
2214                 ext3_msg(sb, KERN_ERR,
2215                         "error: failed to create device journal");
2216                 goto out_bdev;
2217         }
2218         journal->j_private = sb;
2219         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2220         wait_on_buffer(journal->j_sb_buffer);
2221         if (!buffer_uptodate(journal->j_sb_buffer)) {
2222                 ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2223                 goto out_journal;
2224         }
2225         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2226                 ext3_msg(sb, KERN_ERR,
2227                         "error: external journal has more than one "
2228                         "user (unsupported) - %d",
2229                         be32_to_cpu(journal->j_superblock->s_nr_users));
2230                 goto out_journal;
2231         }
2232         EXT3_SB(sb)->journal_bdev = bdev;
2233         ext3_init_journal_params(sb, journal);
2234         return journal;
2235 out_journal:
2236         journal_destroy(journal);
2237 out_bdev:
2238         ext3_blkdev_put(bdev);
2239         return NULL;
2240 }
2241
2242 static int ext3_load_journal(struct super_block *sb,
2243                              struct ext3_super_block *es,
2244                              unsigned long journal_devnum)
2245 {
2246         journal_t *journal;
2247         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2248         dev_t journal_dev;
2249         int err = 0;
2250         int really_read_only;
2251
2252         if (journal_devnum &&
2253             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2254                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2255                         "numbers have changed");
2256                 journal_dev = new_decode_dev(journal_devnum);
2257         } else
2258                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2259
2260         really_read_only = bdev_read_only(sb->s_bdev);
2261
2262         /*
2263          * Are we loading a blank journal or performing recovery after a
2264          * crash?  For recovery, we need to check in advance whether we
2265          * can get read-write access to the device.
2266          */
2267
2268         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2269                 if (sb->s_flags & MS_RDONLY) {
2270                         ext3_msg(sb, KERN_INFO,
2271                                 "recovery required on readonly filesystem");
2272                         if (really_read_only) {
2273                                 ext3_msg(sb, KERN_ERR, "error: write access "
2274                                         "unavailable, cannot proceed");
2275                                 return -EROFS;
2276                         }
2277                         ext3_msg(sb, KERN_INFO,
2278                                 "write access will be enabled during recovery");
2279                 }
2280         }
2281
2282         if (journal_inum && journal_dev) {
2283                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2284                        "and inode journals");
2285                 return -EINVAL;
2286         }
2287
2288         if (journal_inum) {
2289                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2290                         return -EINVAL;
2291         } else {
2292                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2293                         return -EINVAL;
2294         }
2295
2296         if (!(journal->j_flags & JFS_BARRIER))
2297                 printk(KERN_INFO "EXT3-fs: barriers not enabled\n");
2298
2299         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2300                 err = journal_update_format(journal);
2301                 if (err)  {
2302                         ext3_msg(sb, KERN_ERR, "error updating journal");
2303                         journal_destroy(journal);
2304                         return err;
2305                 }
2306         }
2307
2308         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2309                 err = journal_wipe(journal, !really_read_only);
2310         if (!err)
2311                 err = journal_load(journal);
2312
2313         if (err) {
2314                 ext3_msg(sb, KERN_ERR, "error loading journal");
2315                 journal_destroy(journal);
2316                 return err;
2317         }
2318
2319         EXT3_SB(sb)->s_journal = journal;
2320         ext3_clear_journal_err(sb, es);
2321
2322         if (!really_read_only && journal_devnum &&
2323             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2324                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2325
2326                 /* Make sure we flush the recovery flag to disk. */
2327                 ext3_commit_super(sb, es, 1);
2328         }
2329
2330         return 0;
2331 }
2332
2333 static int ext3_create_journal(struct super_block *sb,
2334                                struct ext3_super_block *es,
2335                                unsigned int journal_inum)
2336 {
2337         journal_t *journal;
2338         int err;
2339
2340         if (sb->s_flags & MS_RDONLY) {
2341                 ext3_msg(sb, KERN_ERR,
2342                         "error: readonly filesystem when trying to "
2343                         "create journal");
2344                 return -EROFS;
2345         }
2346
2347         journal = ext3_get_journal(sb, journal_inum);
2348         if (!journal)
2349                 return -EINVAL;
2350
2351         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2352                journal_inum);
2353
2354         err = journal_create(journal);
2355         if (err) {
2356                 ext3_msg(sb, KERN_ERR, "error creating journal");
2357                 journal_destroy(journal);
2358                 return -EIO;
2359         }
2360
2361         EXT3_SB(sb)->s_journal = journal;
2362
2363         ext3_update_dynamic_rev(sb);
2364         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2365         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2366
2367         es->s_journal_inum = cpu_to_le32(journal_inum);
2368
2369         /* Make sure we flush the recovery flag to disk. */
2370         ext3_commit_super(sb, es, 1);
2371
2372         return 0;
2373 }
2374
2375 static int ext3_commit_super(struct super_block *sb,
2376                                struct ext3_super_block *es,
2377                                int sync)
2378 {
2379         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2380         int error = 0;
2381
2382         if (!sbh)
2383                 return error;
2384
2385         if (buffer_write_io_error(sbh)) {
2386                 /*
2387                  * Oh, dear.  A previous attempt to write the
2388                  * superblock failed.  This could happen because the
2389                  * USB device was yanked out.  Or it could happen to
2390                  * be a transient write error and maybe the block will
2391                  * be remapped.  Nothing we can do but to retry the
2392                  * write and hope for the best.
2393                  */
2394                 ext3_msg(sb, KERN_ERR, "previous I/O error to "
2395                        "superblock detected");
2396                 clear_buffer_write_io_error(sbh);
2397                 set_buffer_uptodate(sbh);
2398         }
2399         /*
2400          * If the file system is mounted read-only, don't update the
2401          * superblock write time.  This avoids updating the superblock
2402          * write time when we are mounting the root file system
2403          * read/only but we need to replay the journal; at that point,
2404          * for people who are east of GMT and who make their clock
2405          * tick in localtime for Windows bug-for-bug compatibility,
2406          * the clock is set in the future, and this will cause e2fsck
2407          * to complain and force a full file system check.
2408          */
2409         if (!(sb->s_flags & MS_RDONLY))
2410                 es->s_wtime = cpu_to_le32(get_seconds());
2411         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2412         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2413         BUFFER_TRACE(sbh, "marking dirty");
2414         mark_buffer_dirty(sbh);
2415         if (sync) {
2416                 error = sync_dirty_buffer(sbh);
2417                 if (buffer_write_io_error(sbh)) {
2418                         ext3_msg(sb, KERN_ERR, "I/O error while writing "
2419                                "superblock");
2420                         clear_buffer_write_io_error(sbh);
2421                         set_buffer_uptodate(sbh);
2422                 }
2423         }
2424         return error;
2425 }
2426
2427
2428 /*
2429  * Have we just finished recovery?  If so, and if we are mounting (or
2430  * remounting) the filesystem readonly, then we will end up with a
2431  * consistent fs on disk.  Record that fact.
2432  */
2433 static void ext3_mark_recovery_complete(struct super_block * sb,
2434                                         struct ext3_super_block * es)
2435 {
2436         journal_t *journal = EXT3_SB(sb)->s_journal;
2437
2438         journal_lock_updates(journal);
2439         if (journal_flush(journal) < 0)
2440                 goto out;
2441
2442         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2443             sb->s_flags & MS_RDONLY) {
2444                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2445                 ext3_commit_super(sb, es, 1);
2446         }
2447
2448 out:
2449         journal_unlock_updates(journal);
2450 }
2451
2452 /*
2453  * If we are mounting (or read-write remounting) a filesystem whose journal
2454  * has recorded an error from a previous lifetime, move that error to the
2455  * main filesystem now.
2456  */
2457 static void ext3_clear_journal_err(struct super_block *sb,
2458                                    struct ext3_super_block *es)
2459 {
2460         journal_t *journal;
2461         int j_errno;
2462         const char *errstr;
2463
2464         journal = EXT3_SB(sb)->s_journal;
2465
2466         /*
2467          * Now check for any error status which may have been recorded in the
2468          * journal by a prior ext3_error() or ext3_abort()
2469          */
2470
2471         j_errno = journal_errno(journal);
2472         if (j_errno) {
2473                 char nbuf[16];
2474
2475                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2476                 ext3_warning(sb, __func__, "Filesystem error recorded "
2477                              "from previous mount: %s", errstr);
2478                 ext3_warning(sb, __func__, "Marking fs in need of "
2479                              "filesystem check.");
2480
2481                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2482                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2483                 ext3_commit_super (sb, es, 1);
2484
2485                 journal_clear_err(journal);
2486         }
2487 }
2488
2489 /*
2490  * Force the running and committing transactions to commit,
2491  * and wait on the commit.
2492  */
2493 int ext3_force_commit(struct super_block *sb)
2494 {
2495         journal_t *journal;
2496         int ret;
2497
2498         if (sb->s_flags & MS_RDONLY)
2499                 return 0;
2500
2501         journal = EXT3_SB(sb)->s_journal;
2502         ret = ext3_journal_force_commit(journal);
2503         return ret;
2504 }
2505
2506 static int ext3_sync_fs(struct super_block *sb, int wait)
2507 {
2508         tid_t target;
2509
2510         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2511                 if (wait)
2512                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2513         }
2514         return 0;
2515 }
2516
2517 /*
2518  * LVM calls this function before a (read-only) snapshot is created.  This
2519  * gives us a chance to flush the journal completely and mark the fs clean.
2520  */
2521 static int ext3_freeze(struct super_block *sb)
2522 {
2523         int error = 0;
2524         journal_t *journal;
2525
2526         if (!(sb->s_flags & MS_RDONLY)) {
2527                 journal = EXT3_SB(sb)->s_journal;
2528
2529                 /* Now we set up the journal barrier. */
2530                 journal_lock_updates(journal);
2531
2532                 /*
2533                  * We don't want to clear needs_recovery flag when we failed
2534                  * to flush the journal.
2535                  */
2536                 error = journal_flush(journal);
2537                 if (error < 0)
2538                         goto out;
2539
2540                 /* Journal blocked and flushed, clear needs_recovery flag. */
2541                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2542                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2543                 if (error)
2544                         goto out;
2545         }
2546         return 0;
2547
2548 out:
2549         journal_unlock_updates(journal);
2550         return error;
2551 }
2552
2553 /*
2554  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2555  * flag here, even though the filesystem is not technically dirty yet.
2556  */
2557 static int ext3_unfreeze(struct super_block *sb)
2558 {
2559         if (!(sb->s_flags & MS_RDONLY)) {
2560                 lock_super(sb);
2561                 /* Reser the needs_recovery flag before the fs is unlocked. */
2562                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2563                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2564                 unlock_super(sb);
2565                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2566         }
2567         return 0;
2568 }
2569
2570 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2571 {
2572         struct ext3_super_block * es;
2573         struct ext3_sb_info *sbi = EXT3_SB(sb);
2574         ext3_fsblk_t n_blocks_count = 0;
2575         unsigned long old_sb_flags;
2576         struct ext3_mount_options old_opts;
2577         int enable_quota = 0;
2578         int err;
2579 #ifdef CONFIG_QUOTA
2580         int i;
2581 #endif
2582
2583         /* Store the original options */
2584         lock_super(sb);
2585         old_sb_flags = sb->s_flags;
2586         old_opts.s_mount_opt = sbi->s_mount_opt;
2587         old_opts.s_resuid = sbi->s_resuid;
2588         old_opts.s_resgid = sbi->s_resgid;
2589         old_opts.s_commit_interval = sbi->s_commit_interval;
2590 #ifdef CONFIG_QUOTA
2591         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2592         for (i = 0; i < MAXQUOTAS; i++)
2593                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2594 #endif
2595
2596         /*
2597          * Allow the "check" option to be passed as a remount option.
2598          */
2599         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2600                 err = -EINVAL;
2601                 goto restore_opts;
2602         }
2603
2604         if (test_opt(sb, ABORT))
2605                 ext3_abort(sb, __func__, "Abort forced by user");
2606
2607         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2608                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2609
2610         es = sbi->s_es;
2611
2612         ext3_init_journal_params(sb, sbi->s_journal);
2613
2614         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2615                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2616                 if (test_opt(sb, ABORT)) {
2617                         err = -EROFS;
2618                         goto restore_opts;
2619                 }
2620
2621                 if (*flags & MS_RDONLY) {
2622                         err = dquot_suspend(sb, -1);
2623                         if (err < 0)
2624                                 goto restore_opts;
2625
2626                         /*
2627                          * First of all, the unconditional stuff we have to do
2628                          * to disable replay of the journal when we next remount
2629                          */
2630                         sb->s_flags |= MS_RDONLY;
2631
2632                         /*
2633                          * OK, test if we are remounting a valid rw partition
2634                          * readonly, and if so set the rdonly flag and then
2635                          * mark the partition as valid again.
2636                          */
2637                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2638                             (sbi->s_mount_state & EXT3_VALID_FS))
2639                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2640
2641                         ext3_mark_recovery_complete(sb, es);
2642                 } else {
2643                         __le32 ret;
2644                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2645                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2646                                 ext3_msg(sb, KERN_WARNING,
2647                                         "warning: couldn't remount RDWR "
2648                                         "because of unsupported optional "
2649                                         "features (%x)", le32_to_cpu(ret));
2650                                 err = -EROFS;
2651                                 goto restore_opts;
2652                         }
2653
2654                         /*
2655                          * If we have an unprocessed orphan list hanging
2656                          * around from a previously readonly bdev mount,
2657                          * require a full umount/remount for now.
2658                          */
2659                         if (es->s_last_orphan) {
2660                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2661                                        "remount RDWR because of unprocessed "
2662                                        "orphan inode list.  Please "
2663                                        "umount/remount instead.");
2664                                 err = -EINVAL;
2665                                 goto restore_opts;
2666                         }
2667
2668                         /*
2669                          * Mounting a RDONLY partition read-write, so reread
2670                          * and store the current valid flag.  (It may have
2671                          * been changed by e2fsck since we originally mounted
2672                          * the partition.)
2673                          */
2674                         ext3_clear_journal_err(sb, es);
2675                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2676                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2677                                 goto restore_opts;
2678                         if (!ext3_setup_super (sb, es, 0))
2679                                 sb->s_flags &= ~MS_RDONLY;
2680                         enable_quota = 1;
2681                 }
2682         }
2683 #ifdef CONFIG_QUOTA
2684         /* Release old quota file names */
2685         for (i = 0; i < MAXQUOTAS; i++)
2686                 if (old_opts.s_qf_names[i] &&
2687                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2688                         kfree(old_opts.s_qf_names[i]);
2689 #endif
2690         unlock_super(sb);
2691
2692         if (enable_quota)
2693                 dquot_resume(sb, -1);
2694         return 0;
2695 restore_opts:
2696         sb->s_flags = old_sb_flags;
2697         sbi->s_mount_opt = old_opts.s_mount_opt;
2698         sbi->s_resuid = old_opts.s_resuid;
2699         sbi->s_resgid = old_opts.s_resgid;
2700         sbi->s_commit_interval = old_opts.s_commit_interval;
2701 #ifdef CONFIG_QUOTA
2702         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2703         for (i = 0; i < MAXQUOTAS; i++) {
2704                 if (sbi->s_qf_names[i] &&
2705                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2706                         kfree(sbi->s_qf_names[i]);
2707                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2708         }
2709 #endif
2710         unlock_super(sb);
2711         return err;
2712 }
2713
2714 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2715 {
2716         struct super_block *sb = dentry->d_sb;
2717         struct ext3_sb_info *sbi = EXT3_SB(sb);
2718         struct ext3_super_block *es = sbi->s_es;
2719         u64 fsid;
2720
2721         if (test_opt(sb, MINIX_DF)) {
2722                 sbi->s_overhead_last = 0;
2723         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2724                 unsigned long ngroups = sbi->s_groups_count, i;
2725                 ext3_fsblk_t overhead = 0;
2726                 smp_rmb();
2727
2728                 /*
2729                  * Compute the overhead (FS structures).  This is constant
2730                  * for a given filesystem unless the number of block groups
2731                  * changes so we cache the previous value until it does.
2732                  */
2733
2734                 /*
2735                  * All of the blocks before first_data_block are
2736                  * overhead
2737                  */
2738                 overhead = le32_to_cpu(es->s_first_data_block);
2739
2740                 /*
2741                  * Add the overhead attributed to the superblock and
2742                  * block group descriptors.  If the sparse superblocks
2743                  * feature is turned on, then not all groups have this.
2744                  */
2745                 for (i = 0; i < ngroups; i++) {
2746                         overhead += ext3_bg_has_super(sb, i) +
2747                                 ext3_bg_num_gdb(sb, i);
2748                         cond_resched();
2749                 }
2750
2751                 /*
2752                  * Every block group has an inode bitmap, a block
2753                  * bitmap, and an inode table.
2754                  */
2755                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2756                 sbi->s_overhead_last = overhead;
2757                 smp_wmb();
2758                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2759         }
2760
2761         buf->f_type = EXT3_SUPER_MAGIC;
2762         buf->f_bsize = sb->s_blocksize;
2763         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2764         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2765         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2766         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2767                 buf->f_bavail = 0;
2768         buf->f_files = le32_to_cpu(es->s_inodes_count);
2769         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2770         buf->f_namelen = EXT3_NAME_LEN;
2771         fsid = le64_to_cpup((void *)es->s_uuid) ^
2772                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2773         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2774         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2775         return 0;
2776 }
2777
2778 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2779  * is locked for write. Otherwise the are possible deadlocks:
2780  * Process 1                         Process 2
2781  * ext3_create()                     quota_sync()
2782  *   journal_start()                   write_dquot()
2783  *   dquot_initialize()                       down(dqio_mutex)
2784  *     down(dqio_mutex)                    journal_start()
2785  *
2786  */
2787
2788 #ifdef CONFIG_QUOTA
2789
2790 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2791 {
2792         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2793 }
2794
2795 static int ext3_write_dquot(struct dquot *dquot)
2796 {
2797         int ret, err;
2798         handle_t *handle;
2799         struct inode *inode;
2800
2801         inode = dquot_to_inode(dquot);
2802         handle = ext3_journal_start(inode,
2803                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2804         if (IS_ERR(handle))
2805                 return PTR_ERR(handle);
2806         ret = dquot_commit(dquot);
2807         err = ext3_journal_stop(handle);
2808         if (!ret)
2809                 ret = err;
2810         return ret;
2811 }
2812
2813 static int ext3_acquire_dquot(struct dquot *dquot)
2814 {
2815         int ret, err;
2816         handle_t *handle;
2817
2818         handle = ext3_journal_start(dquot_to_inode(dquot),
2819                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2820         if (IS_ERR(handle))
2821                 return PTR_ERR(handle);
2822         ret = dquot_acquire(dquot);
2823         err = ext3_journal_stop(handle);
2824         if (!ret)
2825                 ret = err;
2826         return ret;
2827 }
2828
2829 static int ext3_release_dquot(struct dquot *dquot)
2830 {
2831         int ret, err;
2832         handle_t *handle;
2833
2834         handle = ext3_journal_start(dquot_to_inode(dquot),
2835                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2836         if (IS_ERR(handle)) {
2837                 /* Release dquot anyway to avoid endless cycle in dqput() */
2838                 dquot_release(dquot);
2839                 return PTR_ERR(handle);
2840         }
2841         ret = dquot_release(dquot);
2842         err = ext3_journal_stop(handle);
2843         if (!ret)
2844                 ret = err;
2845         return ret;
2846 }
2847
2848 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2849 {
2850         /* Are we journaling quotas? */
2851         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2852             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2853                 dquot_mark_dquot_dirty(dquot);
2854                 return ext3_write_dquot(dquot);
2855         } else {
2856                 return dquot_mark_dquot_dirty(dquot);
2857         }
2858 }
2859
2860 static int ext3_write_info(struct super_block *sb, int type)
2861 {
2862         int ret, err;
2863         handle_t *handle;
2864
2865         /* Data block + inode block */
2866         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2867         if (IS_ERR(handle))
2868                 return PTR_ERR(handle);
2869         ret = dquot_commit_info(sb, type);
2870         err = ext3_journal_stop(handle);
2871         if (!ret)
2872                 ret = err;
2873         return ret;
2874 }
2875
2876 /*
2877  * Turn on quotas during mount time - we need to find
2878  * the quota file and such...
2879  */
2880 static int ext3_quota_on_mount(struct super_block *sb, int type)
2881 {
2882         return dquot_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2883                                         EXT3_SB(sb)->s_jquota_fmt, type);
2884 }
2885
2886 /*
2887  * Standard function to be called on quota_on
2888  */
2889 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2890                          struct path *path)
2891 {
2892         int err;
2893
2894         if (!test_opt(sb, QUOTA))
2895                 return -EINVAL;
2896
2897         /* Quotafile not on the same filesystem? */
2898         if (path->mnt->mnt_sb != sb)
2899                 return -EXDEV;
2900         /* Journaling quota? */
2901         if (EXT3_SB(sb)->s_qf_names[type]) {
2902                 /* Quotafile not of fs root? */
2903                 if (path->dentry->d_parent != sb->s_root)
2904                         ext3_msg(sb, KERN_WARNING,
2905                                 "warning: Quota file not on filesystem root. "
2906                                 "Journaled quota will not work.");
2907         }
2908
2909         /*
2910          * When we journal data on quota file, we have to flush journal to see
2911          * all updates to the file when we bypass pagecache...
2912          */
2913         if (ext3_should_journal_data(path->dentry->d_inode)) {
2914                 /*
2915                  * We don't need to lock updates but journal_flush() could
2916                  * otherwise be livelocked...
2917                  */
2918                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2919                 err = journal_flush(EXT3_SB(sb)->s_journal);
2920                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2921                 if (err)
2922                         return err;
2923         }
2924
2925         return dquot_quota_on(sb, type, format_id, path);
2926 }
2927
2928 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2929  * acquiring the locks... As quota files are never truncated and quota code
2930  * itself serializes the operations (and no one else should touch the files)
2931  * we don't have to be afraid of races */
2932 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2933                                size_t len, loff_t off)
2934 {
2935         struct inode *inode = sb_dqopt(sb)->files[type];
2936         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2937         int err = 0;
2938         int offset = off & (sb->s_blocksize - 1);
2939         int tocopy;
2940         size_t toread;
2941         struct buffer_head *bh;
2942         loff_t i_size = i_size_read(inode);
2943
2944         if (off > i_size)
2945                 return 0;
2946         if (off+len > i_size)
2947                 len = i_size-off;
2948         toread = len;
2949         while (toread > 0) {
2950                 tocopy = sb->s_blocksize - offset < toread ?
2951                                 sb->s_blocksize - offset : toread;
2952                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2953                 if (err)
2954                         return err;
2955                 if (!bh)        /* A hole? */
2956                         memset(data, 0, tocopy);
2957                 else
2958                         memcpy(data, bh->b_data+offset, tocopy);
2959                 brelse(bh);
2960                 offset = 0;
2961                 toread -= tocopy;
2962                 data += tocopy;
2963                 blk++;
2964         }
2965         return len;
2966 }
2967
2968 /* Write to quotafile (we know the transaction is already started and has
2969  * enough credits) */
2970 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2971                                 const char *data, size_t len, loff_t off)
2972 {
2973         struct inode *inode = sb_dqopt(sb)->files[type];
2974         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2975         int err = 0;
2976         int offset = off & (sb->s_blocksize - 1);
2977         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2978         struct buffer_head *bh;
2979         handle_t *handle = journal_current_handle();
2980
2981         if (!handle) {
2982                 ext3_msg(sb, KERN_WARNING,
2983                         "warning: quota write (off=%llu, len=%llu)"
2984                         " cancelled because transaction is not started.",
2985                         (unsigned long long)off, (unsigned long long)len);
2986                 return -EIO;
2987         }
2988
2989         /*
2990          * Since we account only one data block in transaction credits,
2991          * then it is impossible to cross a block boundary.
2992          */
2993         if (sb->s_blocksize - offset < len) {
2994                 ext3_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
2995                         " cancelled because not block aligned",
2996                         (unsigned long long)off, (unsigned long long)len);
2997                 return -EIO;
2998         }
2999         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3000         bh = ext3_bread(handle, inode, blk, 1, &err);
3001         if (!bh)
3002                 goto out;
3003         if (journal_quota) {
3004                 err = ext3_journal_get_write_access(handle, bh);
3005                 if (err) {
3006                         brelse(bh);
3007                         goto out;
3008                 }
3009         }
3010         lock_buffer(bh);
3011         memcpy(bh->b_data+offset, data, len);
3012         flush_dcache_page(bh->b_page);
3013         unlock_buffer(bh);
3014         if (journal_quota)
3015                 err = ext3_journal_dirty_metadata(handle, bh);
3016         else {
3017                 /* Always do at least ordered writes for quotas */
3018                 err = ext3_journal_dirty_data(handle, bh);
3019                 mark_buffer_dirty(bh);
3020         }
3021         brelse(bh);
3022 out:
3023         if (err) {
3024                 mutex_unlock(&inode->i_mutex);
3025                 return err;
3026         }
3027         if (inode->i_size < off + len) {
3028                 i_size_write(inode, off + len);
3029                 EXT3_I(inode)->i_disksize = inode->i_size;
3030         }
3031         inode->i_version++;
3032         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3033         ext3_mark_inode_dirty(handle, inode);
3034         mutex_unlock(&inode->i_mutex);
3035         return len;
3036 }
3037
3038 #endif
3039
3040 static struct dentry *ext3_mount(struct file_system_type *fs_type,
3041         int flags, const char *dev_name, void *data)
3042 {
3043         return mount_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
3044 }
3045
3046 static struct file_system_type ext3_fs_type = {
3047         .owner          = THIS_MODULE,
3048         .name           = "ext3",
3049         .mount          = ext3_mount,
3050         .kill_sb        = kill_block_super,
3051         .fs_flags       = FS_REQUIRES_DEV,
3052 };
3053
3054 static int __init init_ext3_fs(void)
3055 {
3056         int err = init_ext3_xattr();
3057         if (err)
3058                 return err;
3059         err = init_inodecache();
3060         if (err)
3061                 goto out1;
3062         err = register_filesystem(&ext3_fs_type);
3063         if (err)
3064                 goto out;
3065         return 0;
3066 out:
3067         destroy_inodecache();
3068 out1:
3069         exit_ext3_xattr();
3070         return err;
3071 }
3072
3073 static void __exit exit_ext3_fs(void)
3074 {
3075         unregister_filesystem(&ext3_fs_type);
3076         destroy_inodecache();
3077         exit_ext3_xattr();
3078 }
3079
3080 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3081 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3082 MODULE_LICENSE("GPL");
3083 module_init(init_ext3_fs)
3084 module_exit(exit_ext3_fs)