ext3: trivial quota cleanup
[linux-2.6.git] / fs / ext3 / super.c
1 /*
2  *  linux/fs/ext3/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/module.h>
20 #include <linux/string.h>
21 #include <linux/fs.h>
22 #include <linux/time.h>
23 #include <linux/jbd.h>
24 #include <linux/ext3_fs.h>
25 #include <linux/ext3_jbd.h>
26 #include <linux/slab.h>
27 #include <linux/init.h>
28 #include <linux/blkdev.h>
29 #include <linux/parser.h>
30 #include <linux/smp_lock.h>
31 #include <linux/buffer_head.h>
32 #include <linux/exportfs.h>
33 #include <linux/vfs.h>
34 #include <linux/random.h>
35 #include <linux/mount.h>
36 #include <linux/namei.h>
37 #include <linux/quotaops.h>
38 #include <linux/seq_file.h>
39 #include <linux/log2.h>
40
41 #include <asm/uaccess.h>
42
43 #include "xattr.h"
44 #include "acl.h"
45 #include "namei.h"
46
47 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
48   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
49 #else
50   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
51 #endif
52
53 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
54                              unsigned long journal_devnum);
55 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
56                                unsigned int);
57 static int ext3_commit_super(struct super_block *sb,
58                                struct ext3_super_block *es,
59                                int sync);
60 static void ext3_mark_recovery_complete(struct super_block * sb,
61                                         struct ext3_super_block * es);
62 static void ext3_clear_journal_err(struct super_block * sb,
63                                    struct ext3_super_block * es);
64 static int ext3_sync_fs(struct super_block *sb, int wait);
65 static const char *ext3_decode_error(struct super_block * sb, int errno,
66                                      char nbuf[16]);
67 static int ext3_remount (struct super_block * sb, int * flags, char * data);
68 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
69 static int ext3_unfreeze(struct super_block *sb);
70 static int ext3_freeze(struct super_block *sb);
71
72 /*
73  * Wrappers for journal_start/end.
74  *
75  * The only special thing we need to do here is to make sure that all
76  * journal_end calls result in the superblock being marked dirty, so
77  * that sync() will call the filesystem's write_super callback if
78  * appropriate.
79  */
80 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
81 {
82         journal_t *journal;
83
84         if (sb->s_flags & MS_RDONLY)
85                 return ERR_PTR(-EROFS);
86
87         /* Special case here: if the journal has aborted behind our
88          * backs (eg. EIO in the commit thread), then we still need to
89          * take the FS itself readonly cleanly. */
90         journal = EXT3_SB(sb)->s_journal;
91         if (is_journal_aborted(journal)) {
92                 ext3_abort(sb, __func__,
93                            "Detected aborted journal");
94                 return ERR_PTR(-EROFS);
95         }
96
97         return journal_start(journal, nblocks);
98 }
99
100 /*
101  * The only special thing we need to do here is to make sure that all
102  * journal_stop calls result in the superblock being marked dirty, so
103  * that sync() will call the filesystem's write_super callback if
104  * appropriate.
105  */
106 int __ext3_journal_stop(const char *where, handle_t *handle)
107 {
108         struct super_block *sb;
109         int err;
110         int rc;
111
112         sb = handle->h_transaction->t_journal->j_private;
113         err = handle->h_err;
114         rc = journal_stop(handle);
115
116         if (!err)
117                 err = rc;
118         if (err)
119                 __ext3_std_error(sb, where, err);
120         return err;
121 }
122
123 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
124                 struct buffer_head *bh, handle_t *handle, int err)
125 {
126         char nbuf[16];
127         const char *errstr = ext3_decode_error(NULL, err, nbuf);
128
129         if (bh)
130                 BUFFER_TRACE(bh, "abort");
131
132         if (!handle->h_err)
133                 handle->h_err = err;
134
135         if (is_handle_aborted(handle))
136                 return;
137
138         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
139                 caller, errstr, err_fn);
140
141         journal_abort_handle(handle);
142 }
143
144 void ext3_msg(struct super_block *sb, const char *prefix,
145                 const char *fmt, ...)
146 {
147         va_list args;
148
149         va_start(args, fmt);
150         printk("%sEXT3-fs (%s): ", prefix, sb->s_id);
151         vprintk(fmt, args);
152         printk("\n");
153         va_end(args);
154 }
155
156 /* Deal with the reporting of failure conditions on a filesystem such as
157  * inconsistencies detected or read IO failures.
158  *
159  * On ext2, we can store the error state of the filesystem in the
160  * superblock.  That is not possible on ext3, because we may have other
161  * write ordering constraints on the superblock which prevent us from
162  * writing it out straight away; and given that the journal is about to
163  * be aborted, we can't rely on the current, or future, transactions to
164  * write out the superblock safely.
165  *
166  * We'll just use the journal_abort() error code to record an error in
167  * the journal instead.  On recovery, the journal will compain about
168  * that error until we've noted it down and cleared it.
169  */
170
171 static void ext3_handle_error(struct super_block *sb)
172 {
173         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
174
175         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
176         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
177
178         if (sb->s_flags & MS_RDONLY)
179                 return;
180
181         if (!test_opt (sb, ERRORS_CONT)) {
182                 journal_t *journal = EXT3_SB(sb)->s_journal;
183
184                 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
185                 if (journal)
186                         journal_abort(journal, -EIO);
187         }
188         if (test_opt (sb, ERRORS_RO)) {
189                 ext3_msg(sb, KERN_CRIT,
190                         "error: remounting filesystem read-only");
191                 sb->s_flags |= MS_RDONLY;
192         }
193         ext3_commit_super(sb, es, 1);
194         if (test_opt(sb, ERRORS_PANIC))
195                 panic("EXT3-fs (%s): panic forced after error\n",
196                         sb->s_id);
197 }
198
199 void ext3_error (struct super_block * sb, const char * function,
200                  const char * fmt, ...)
201 {
202         va_list args;
203
204         va_start(args, fmt);
205         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
206         vprintk(fmt, args);
207         printk("\n");
208         va_end(args);
209
210         ext3_handle_error(sb);
211 }
212
213 static const char *ext3_decode_error(struct super_block * sb, int errno,
214                                      char nbuf[16])
215 {
216         char *errstr = NULL;
217
218         switch (errno) {
219         case -EIO:
220                 errstr = "IO failure";
221                 break;
222         case -ENOMEM:
223                 errstr = "Out of memory";
224                 break;
225         case -EROFS:
226                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
227                         errstr = "Journal has aborted";
228                 else
229                         errstr = "Readonly filesystem";
230                 break;
231         default:
232                 /* If the caller passed in an extra buffer for unknown
233                  * errors, textualise them now.  Else we just return
234                  * NULL. */
235                 if (nbuf) {
236                         /* Check for truncated error codes... */
237                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
238                                 errstr = nbuf;
239                 }
240                 break;
241         }
242
243         return errstr;
244 }
245
246 /* __ext3_std_error decodes expected errors from journaling functions
247  * automatically and invokes the appropriate error response.  */
248
249 void __ext3_std_error (struct super_block * sb, const char * function,
250                        int errno)
251 {
252         char nbuf[16];
253         const char *errstr;
254
255         /* Special case: if the error is EROFS, and we're not already
256          * inside a transaction, then there's really no point in logging
257          * an error. */
258         if (errno == -EROFS && journal_current_handle() == NULL &&
259             (sb->s_flags & MS_RDONLY))
260                 return;
261
262         errstr = ext3_decode_error(sb, errno, nbuf);
263         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
264
265         ext3_handle_error(sb);
266 }
267
268 /*
269  * ext3_abort is a much stronger failure handler than ext3_error.  The
270  * abort function may be used to deal with unrecoverable failures such
271  * as journal IO errors or ENOMEM at a critical moment in log management.
272  *
273  * We unconditionally force the filesystem into an ABORT|READONLY state,
274  * unless the error response on the fs has been set to panic in which
275  * case we take the easy way out and panic immediately.
276  */
277
278 void ext3_abort (struct super_block * sb, const char * function,
279                  const char * fmt, ...)
280 {
281         va_list args;
282
283         va_start(args, fmt);
284         printk(KERN_CRIT "EXT3-fs (%s): error: %s: ", sb->s_id, function);
285         vprintk(fmt, args);
286         printk("\n");
287         va_end(args);
288
289         if (test_opt(sb, ERRORS_PANIC))
290                 panic("EXT3-fs: panic from previous error\n");
291
292         if (sb->s_flags & MS_RDONLY)
293                 return;
294
295         ext3_msg(sb, KERN_CRIT,
296                 "error: remounting filesystem read-only");
297         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
298         sb->s_flags |= MS_RDONLY;
299         set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
300         if (EXT3_SB(sb)->s_journal)
301                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
302 }
303
304 void ext3_warning (struct super_block * sb, const char * function,
305                    const char * fmt, ...)
306 {
307         va_list args;
308
309         va_start(args, fmt);
310         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: ",
311                sb->s_id, function);
312         vprintk(fmt, args);
313         printk("\n");
314         va_end(args);
315 }
316
317 void ext3_update_dynamic_rev(struct super_block *sb)
318 {
319         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
320
321         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
322                 return;
323
324         ext3_msg(sb, KERN_WARNING,
325                 "warning: updating to rev %d because of "
326                 "new feature flag, running e2fsck is recommended",
327                 EXT3_DYNAMIC_REV);
328
329         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
330         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
331         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
332         /* leave es->s_feature_*compat flags alone */
333         /* es->s_uuid will be set by e2fsck if empty */
334
335         /*
336          * The rest of the superblock fields should be zero, and if not it
337          * means they are likely already in use, so leave them alone.  We
338          * can leave it up to e2fsck to clean up any inconsistencies there.
339          */
340 }
341
342 /*
343  * Open the external journal device
344  */
345 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
346 {
347         struct block_device *bdev;
348         char b[BDEVNAME_SIZE];
349
350         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
351         if (IS_ERR(bdev))
352                 goto fail;
353         return bdev;
354
355 fail:
356         ext3_msg(sb, "error: failed to open journal device %s: %ld",
357                 __bdevname(dev, b), PTR_ERR(bdev));
358
359         return NULL;
360 }
361
362 /*
363  * Release the journal device
364  */
365 static int ext3_blkdev_put(struct block_device *bdev)
366 {
367         bd_release(bdev);
368         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
369 }
370
371 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
372 {
373         struct block_device *bdev;
374         int ret = -ENODEV;
375
376         bdev = sbi->journal_bdev;
377         if (bdev) {
378                 ret = ext3_blkdev_put(bdev);
379                 sbi->journal_bdev = NULL;
380         }
381         return ret;
382 }
383
384 static inline struct inode *orphan_list_entry(struct list_head *l)
385 {
386         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
387 }
388
389 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
390 {
391         struct list_head *l;
392
393         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
394                le32_to_cpu(sbi->s_es->s_last_orphan));
395
396         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
397         list_for_each(l, &sbi->s_orphan) {
398                 struct inode *inode = orphan_list_entry(l);
399                 ext3_msg(sb, KERN_ERR, "  "
400                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
401                        inode->i_sb->s_id, inode->i_ino, inode,
402                        inode->i_mode, inode->i_nlink,
403                        NEXT_ORPHAN(inode));
404         }
405 }
406
407 static void ext3_put_super (struct super_block * sb)
408 {
409         struct ext3_sb_info *sbi = EXT3_SB(sb);
410         struct ext3_super_block *es = sbi->s_es;
411         int i, err;
412
413         lock_kernel();
414
415         ext3_xattr_put_super(sb);
416         err = journal_destroy(sbi->s_journal);
417         sbi->s_journal = NULL;
418         if (err < 0)
419                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
420
421         if (!(sb->s_flags & MS_RDONLY)) {
422                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
423                 es->s_state = cpu_to_le16(sbi->s_mount_state);
424                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
425                 mark_buffer_dirty(sbi->s_sbh);
426                 ext3_commit_super(sb, es, 1);
427         }
428
429         for (i = 0; i < sbi->s_gdb_count; i++)
430                 brelse(sbi->s_group_desc[i]);
431         kfree(sbi->s_group_desc);
432         percpu_counter_destroy(&sbi->s_freeblocks_counter);
433         percpu_counter_destroy(&sbi->s_freeinodes_counter);
434         percpu_counter_destroy(&sbi->s_dirs_counter);
435         brelse(sbi->s_sbh);
436 #ifdef CONFIG_QUOTA
437         for (i = 0; i < MAXQUOTAS; i++)
438                 kfree(sbi->s_qf_names[i]);
439 #endif
440
441         /* Debugging code just in case the in-memory inode orphan list
442          * isn't empty.  The on-disk one can be non-empty if we've
443          * detected an error and taken the fs readonly, but the
444          * in-memory list had better be clean by this point. */
445         if (!list_empty(&sbi->s_orphan))
446                 dump_orphan_list(sb, sbi);
447         J_ASSERT(list_empty(&sbi->s_orphan));
448
449         invalidate_bdev(sb->s_bdev);
450         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
451                 /*
452                  * Invalidate the journal device's buffers.  We don't want them
453                  * floating about in memory - the physical journal device may
454                  * hotswapped, and it breaks the `ro-after' testing code.
455                  */
456                 sync_blockdev(sbi->journal_bdev);
457                 invalidate_bdev(sbi->journal_bdev);
458                 ext3_blkdev_remove(sbi);
459         }
460         sb->s_fs_info = NULL;
461         kfree(sbi->s_blockgroup_lock);
462         kfree(sbi);
463
464         unlock_kernel();
465 }
466
467 static struct kmem_cache *ext3_inode_cachep;
468
469 /*
470  * Called inside transaction, so use GFP_NOFS
471  */
472 static struct inode *ext3_alloc_inode(struct super_block *sb)
473 {
474         struct ext3_inode_info *ei;
475
476         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
477         if (!ei)
478                 return NULL;
479         ei->i_block_alloc_info = NULL;
480         ei->vfs_inode.i_version = 1;
481         atomic_set(&ei->i_datasync_tid, 0);
482         atomic_set(&ei->i_sync_tid, 0);
483         return &ei->vfs_inode;
484 }
485
486 static void ext3_destroy_inode(struct inode *inode)
487 {
488         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
489                 printk("EXT3 Inode %p: orphan list check failed!\n",
490                         EXT3_I(inode));
491                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
492                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
493                                 false);
494                 dump_stack();
495         }
496         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
497 }
498
499 static void init_once(void *foo)
500 {
501         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
502
503         INIT_LIST_HEAD(&ei->i_orphan);
504 #ifdef CONFIG_EXT3_FS_XATTR
505         init_rwsem(&ei->xattr_sem);
506 #endif
507         mutex_init(&ei->truncate_mutex);
508         inode_init_once(&ei->vfs_inode);
509 }
510
511 static int init_inodecache(void)
512 {
513         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
514                                              sizeof(struct ext3_inode_info),
515                                              0, (SLAB_RECLAIM_ACCOUNT|
516                                                 SLAB_MEM_SPREAD),
517                                              init_once);
518         if (ext3_inode_cachep == NULL)
519                 return -ENOMEM;
520         return 0;
521 }
522
523 static void destroy_inodecache(void)
524 {
525         kmem_cache_destroy(ext3_inode_cachep);
526 }
527
528 static void ext3_clear_inode(struct inode *inode)
529 {
530         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
531         ext3_discard_reservation(inode);
532         EXT3_I(inode)->i_block_alloc_info = NULL;
533         if (unlikely(rsv))
534                 kfree(rsv);
535 }
536
537 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
538 {
539 #if defined(CONFIG_QUOTA)
540         struct ext3_sb_info *sbi = EXT3_SB(sb);
541
542         if (sbi->s_jquota_fmt) {
543                 char *fmtname = "";
544
545                 switch (sbi->s_jquota_fmt) {
546                 case QFMT_VFS_OLD:
547                         fmtname = "vfsold";
548                         break;
549                 case QFMT_VFS_V0:
550                         fmtname = "vfsv0";
551                         break;
552                 case QFMT_VFS_V1:
553                         fmtname = "vfsv1";
554                         break;
555                 }
556                 seq_printf(seq, ",jqfmt=%s", fmtname);
557         }
558
559         if (sbi->s_qf_names[USRQUOTA])
560                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
561
562         if (sbi->s_qf_names[GRPQUOTA])
563                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
564
565         if (test_opt(sb, USRQUOTA))
566                 seq_puts(seq, ",usrquota");
567
568         if (test_opt(sb, GRPQUOTA))
569                 seq_puts(seq, ",grpquota");
570 #endif
571 }
572
573 static char *data_mode_string(unsigned long mode)
574 {
575         switch (mode) {
576         case EXT3_MOUNT_JOURNAL_DATA:
577                 return "journal";
578         case EXT3_MOUNT_ORDERED_DATA:
579                 return "ordered";
580         case EXT3_MOUNT_WRITEBACK_DATA:
581                 return "writeback";
582         }
583         return "unknown";
584 }
585
586 /*
587  * Show an option if
588  *  - it's set to a non-default value OR
589  *  - if the per-sb default is different from the global default
590  */
591 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
592 {
593         struct super_block *sb = vfs->mnt_sb;
594         struct ext3_sb_info *sbi = EXT3_SB(sb);
595         struct ext3_super_block *es = sbi->s_es;
596         unsigned long def_mount_opts;
597
598         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
599
600         if (sbi->s_sb_block != 1)
601                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
602         if (test_opt(sb, MINIX_DF))
603                 seq_puts(seq, ",minixdf");
604         if (test_opt(sb, GRPID))
605                 seq_puts(seq, ",grpid");
606         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
607                 seq_puts(seq, ",nogrpid");
608         if (sbi->s_resuid != EXT3_DEF_RESUID ||
609             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
610                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
611         }
612         if (sbi->s_resgid != EXT3_DEF_RESGID ||
613             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
614                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
615         }
616         if (test_opt(sb, ERRORS_RO)) {
617                 int def_errors = le16_to_cpu(es->s_errors);
618
619                 if (def_errors == EXT3_ERRORS_PANIC ||
620                     def_errors == EXT3_ERRORS_CONTINUE) {
621                         seq_puts(seq, ",errors=remount-ro");
622                 }
623         }
624         if (test_opt(sb, ERRORS_CONT))
625                 seq_puts(seq, ",errors=continue");
626         if (test_opt(sb, ERRORS_PANIC))
627                 seq_puts(seq, ",errors=panic");
628         if (test_opt(sb, NO_UID32))
629                 seq_puts(seq, ",nouid32");
630         if (test_opt(sb, DEBUG))
631                 seq_puts(seq, ",debug");
632         if (test_opt(sb, OLDALLOC))
633                 seq_puts(seq, ",oldalloc");
634 #ifdef CONFIG_EXT3_FS_XATTR
635         if (test_opt(sb, XATTR_USER))
636                 seq_puts(seq, ",user_xattr");
637         if (!test_opt(sb, XATTR_USER) &&
638             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
639                 seq_puts(seq, ",nouser_xattr");
640         }
641 #endif
642 #ifdef CONFIG_EXT3_FS_POSIX_ACL
643         if (test_opt(sb, POSIX_ACL))
644                 seq_puts(seq, ",acl");
645         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
646                 seq_puts(seq, ",noacl");
647 #endif
648         if (!test_opt(sb, RESERVATION))
649                 seq_puts(seq, ",noreservation");
650         if (sbi->s_commit_interval) {
651                 seq_printf(seq, ",commit=%u",
652                            (unsigned) (sbi->s_commit_interval / HZ));
653         }
654         if (test_opt(sb, BARRIER))
655                 seq_puts(seq, ",barrier=1");
656         if (test_opt(sb, NOBH))
657                 seq_puts(seq, ",nobh");
658
659         seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
660         if (test_opt(sb, DATA_ERR_ABORT))
661                 seq_puts(seq, ",data_err=abort");
662
663         if (test_opt(sb, NOLOAD))
664                 seq_puts(seq, ",norecovery");
665
666         ext3_show_quota_options(seq, sb);
667
668         return 0;
669 }
670
671
672 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
673                 u64 ino, u32 generation)
674 {
675         struct inode *inode;
676
677         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
678                 return ERR_PTR(-ESTALE);
679         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
680                 return ERR_PTR(-ESTALE);
681
682         /* iget isn't really right if the inode is currently unallocated!!
683          *
684          * ext3_read_inode will return a bad_inode if the inode had been
685          * deleted, so we should be safe.
686          *
687          * Currently we don't know the generation for parent directory, so
688          * a generation of 0 means "accept any"
689          */
690         inode = ext3_iget(sb, ino);
691         if (IS_ERR(inode))
692                 return ERR_CAST(inode);
693         if (generation && inode->i_generation != generation) {
694                 iput(inode);
695                 return ERR_PTR(-ESTALE);
696         }
697
698         return inode;
699 }
700
701 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
702                 int fh_len, int fh_type)
703 {
704         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
705                                     ext3_nfs_get_inode);
706 }
707
708 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
709                 int fh_len, int fh_type)
710 {
711         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
712                                     ext3_nfs_get_inode);
713 }
714
715 /*
716  * Try to release metadata pages (indirect blocks, directories) which are
717  * mapped via the block device.  Since these pages could have journal heads
718  * which would prevent try_to_free_buffers() from freeing them, we must use
719  * jbd layer's try_to_free_buffers() function to release them.
720  */
721 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
722                                  gfp_t wait)
723 {
724         journal_t *journal = EXT3_SB(sb)->s_journal;
725
726         WARN_ON(PageChecked(page));
727         if (!page_has_buffers(page))
728                 return 0;
729         if (journal)
730                 return journal_try_to_free_buffers(journal, page, 
731                                                    wait & ~__GFP_WAIT);
732         return try_to_free_buffers(page);
733 }
734
735 #ifdef CONFIG_QUOTA
736 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
737 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
738
739 static int ext3_write_dquot(struct dquot *dquot);
740 static int ext3_acquire_dquot(struct dquot *dquot);
741 static int ext3_release_dquot(struct dquot *dquot);
742 static int ext3_mark_dquot_dirty(struct dquot *dquot);
743 static int ext3_write_info(struct super_block *sb, int type);
744 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
745                                 char *path, int remount);
746 static int ext3_quota_on_mount(struct super_block *sb, int type);
747 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
748                                size_t len, loff_t off);
749 static ssize_t ext3_quota_write(struct super_block *sb, int type,
750                                 const char *data, size_t len, loff_t off);
751
752 static const struct dquot_operations ext3_quota_operations = {
753         .initialize     = dquot_initialize,
754         .drop           = dquot_drop,
755         .alloc_space    = dquot_alloc_space,
756         .alloc_inode    = dquot_alloc_inode,
757         .free_space     = dquot_free_space,
758         .free_inode     = dquot_free_inode,
759         .transfer       = dquot_transfer,
760         .write_dquot    = ext3_write_dquot,
761         .acquire_dquot  = ext3_acquire_dquot,
762         .release_dquot  = ext3_release_dquot,
763         .mark_dirty     = ext3_mark_dquot_dirty,
764         .write_info     = ext3_write_info,
765         .alloc_dquot    = dquot_alloc,
766         .destroy_dquot  = dquot_destroy,
767 };
768
769 static const struct quotactl_ops ext3_qctl_operations = {
770         .quota_on       = ext3_quota_on,
771         .quota_off      = vfs_quota_off,
772         .quota_sync     = vfs_quota_sync,
773         .get_info       = vfs_get_dqinfo,
774         .set_info       = vfs_set_dqinfo,
775         .get_dqblk      = vfs_get_dqblk,
776         .set_dqblk      = vfs_set_dqblk
777 };
778 #endif
779
780 static const struct super_operations ext3_sops = {
781         .alloc_inode    = ext3_alloc_inode,
782         .destroy_inode  = ext3_destroy_inode,
783         .write_inode    = ext3_write_inode,
784         .dirty_inode    = ext3_dirty_inode,
785         .delete_inode   = ext3_delete_inode,
786         .put_super      = ext3_put_super,
787         .sync_fs        = ext3_sync_fs,
788         .freeze_fs      = ext3_freeze,
789         .unfreeze_fs    = ext3_unfreeze,
790         .statfs         = ext3_statfs,
791         .remount_fs     = ext3_remount,
792         .clear_inode    = ext3_clear_inode,
793         .show_options   = ext3_show_options,
794 #ifdef CONFIG_QUOTA
795         .quota_read     = ext3_quota_read,
796         .quota_write    = ext3_quota_write,
797 #endif
798         .bdev_try_to_free_page = bdev_try_to_free_page,
799 };
800
801 static const struct export_operations ext3_export_ops = {
802         .fh_to_dentry = ext3_fh_to_dentry,
803         .fh_to_parent = ext3_fh_to_parent,
804         .get_parent = ext3_get_parent,
805 };
806
807 enum {
808         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
809         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
810         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
811         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
812         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
813         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
814         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
815         Opt_data_err_abort, Opt_data_err_ignore,
816         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
817         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
818         Opt_noquota, Opt_ignore, Opt_barrier, Opt_err, Opt_resize,
819         Opt_usrquota, Opt_grpquota
820 };
821
822 static const match_table_t tokens = {
823         {Opt_bsd_df, "bsddf"},
824         {Opt_minix_df, "minixdf"},
825         {Opt_grpid, "grpid"},
826         {Opt_grpid, "bsdgroups"},
827         {Opt_nogrpid, "nogrpid"},
828         {Opt_nogrpid, "sysvgroups"},
829         {Opt_resgid, "resgid=%u"},
830         {Opt_resuid, "resuid=%u"},
831         {Opt_sb, "sb=%u"},
832         {Opt_err_cont, "errors=continue"},
833         {Opt_err_panic, "errors=panic"},
834         {Opt_err_ro, "errors=remount-ro"},
835         {Opt_nouid32, "nouid32"},
836         {Opt_nocheck, "nocheck"},
837         {Opt_nocheck, "check=none"},
838         {Opt_debug, "debug"},
839         {Opt_oldalloc, "oldalloc"},
840         {Opt_orlov, "orlov"},
841         {Opt_user_xattr, "user_xattr"},
842         {Opt_nouser_xattr, "nouser_xattr"},
843         {Opt_acl, "acl"},
844         {Opt_noacl, "noacl"},
845         {Opt_reservation, "reservation"},
846         {Opt_noreservation, "noreservation"},
847         {Opt_noload, "noload"},
848         {Opt_noload, "norecovery"},
849         {Opt_nobh, "nobh"},
850         {Opt_bh, "bh"},
851         {Opt_commit, "commit=%u"},
852         {Opt_journal_update, "journal=update"},
853         {Opt_journal_inum, "journal=%u"},
854         {Opt_journal_dev, "journal_dev=%u"},
855         {Opt_abort, "abort"},
856         {Opt_data_journal, "data=journal"},
857         {Opt_data_ordered, "data=ordered"},
858         {Opt_data_writeback, "data=writeback"},
859         {Opt_data_err_abort, "data_err=abort"},
860         {Opt_data_err_ignore, "data_err=ignore"},
861         {Opt_offusrjquota, "usrjquota="},
862         {Opt_usrjquota, "usrjquota=%s"},
863         {Opt_offgrpjquota, "grpjquota="},
864         {Opt_grpjquota, "grpjquota=%s"},
865         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
866         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
867         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
868         {Opt_grpquota, "grpquota"},
869         {Opt_noquota, "noquota"},
870         {Opt_quota, "quota"},
871         {Opt_usrquota, "usrquota"},
872         {Opt_barrier, "barrier=%u"},
873         {Opt_resize, "resize"},
874         {Opt_err, NULL},
875 };
876
877 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
878 {
879         ext3_fsblk_t    sb_block;
880         char            *options = (char *) *data;
881
882         if (!options || strncmp(options, "sb=", 3) != 0)
883                 return 1;       /* Default location */
884         options += 3;
885         /*todo: use simple_strtoll with >32bit ext3 */
886         sb_block = simple_strtoul(options, &options, 0);
887         if (*options && *options != ',') {
888                 ext3_msg(sb, "error: invalid sb specification: %s",
889                        (char *) *data);
890                 return 1;
891         }
892         if (*options == ',')
893                 options++;
894         *data = (void *) options;
895         return sb_block;
896 }
897
898 #ifdef CONFIG_QUOTA
899 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
900 {
901         struct ext3_sb_info *sbi = EXT3_SB(sb);
902         char *qname;
903
904         if (sb_any_quota_loaded(sb) &&
905                 !sbi->s_qf_names[qtype]) {
906                 ext3_msg(sb, KERN_ERR,
907                         "Cannot change journaled "
908                         "quota options when quota turned on");
909                 return 0;
910         }
911         qname = match_strdup(args);
912         if (!qname) {
913                 ext3_msg(sb, KERN_ERR,
914                         "Not enough memory for storing quotafile name");
915                 return 0;
916         }
917         if (sbi->s_qf_names[qtype] &&
918                 strcmp(sbi->s_qf_names[qtype], qname)) {
919                 ext3_msg(sb, KERN_ERR,
920                         "%s quota file already specified", QTYPE2NAME(qtype));
921                 kfree(qname);
922                 return 0;
923         }
924         sbi->s_qf_names[qtype] = qname;
925         if (strchr(sbi->s_qf_names[qtype], '/')) {
926                 ext3_msg(sb, KERN_ERR,
927                         "quotafile must be on filesystem root");
928                 kfree(sbi->s_qf_names[qtype]);
929                 sbi->s_qf_names[qtype] = NULL;
930                 return 0;
931         }
932         set_opt(sbi->s_mount_opt, QUOTA);
933         return 1;
934 }
935
936 static int clear_qf_name(struct super_block *sb, int qtype) {
937
938         struct ext3_sb_info *sbi = EXT3_SB(sb);
939
940         if (sb_any_quota_loaded(sb) &&
941                 sbi->s_qf_names[qtype]) {
942                 ext3_msg(sb, KERN_ERR, "Cannot change journaled quota options"
943                         " when quota turned on");
944                 return 0;
945         }
946         /*
947          * The space will be released later when all options are confirmed
948          * to be correct
949          */
950         sbi->s_qf_names[qtype] = NULL;
951         return 1;
952 }
953 #endif
954
955 static int parse_options (char *options, struct super_block *sb,
956                           unsigned int *inum, unsigned long *journal_devnum,
957                           ext3_fsblk_t *n_blocks_count, int is_remount)
958 {
959         struct ext3_sb_info *sbi = EXT3_SB(sb);
960         char * p;
961         substring_t args[MAX_OPT_ARGS];
962         int data_opt = 0;
963         int option;
964 #ifdef CONFIG_QUOTA
965         int qfmt;
966 #endif
967
968         if (!options)
969                 return 1;
970
971         while ((p = strsep (&options, ",")) != NULL) {
972                 int token;
973                 if (!*p)
974                         continue;
975
976                 token = match_token(p, tokens, args);
977                 switch (token) {
978                 case Opt_bsd_df:
979                         clear_opt (sbi->s_mount_opt, MINIX_DF);
980                         break;
981                 case Opt_minix_df:
982                         set_opt (sbi->s_mount_opt, MINIX_DF);
983                         break;
984                 case Opt_grpid:
985                         set_opt (sbi->s_mount_opt, GRPID);
986                         break;
987                 case Opt_nogrpid:
988                         clear_opt (sbi->s_mount_opt, GRPID);
989                         break;
990                 case Opt_resuid:
991                         if (match_int(&args[0], &option))
992                                 return 0;
993                         sbi->s_resuid = option;
994                         break;
995                 case Opt_resgid:
996                         if (match_int(&args[0], &option))
997                                 return 0;
998                         sbi->s_resgid = option;
999                         break;
1000                 case Opt_sb:
1001                         /* handled by get_sb_block() instead of here */
1002                         /* *sb_block = match_int(&args[0]); */
1003                         break;
1004                 case Opt_err_panic:
1005                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1006                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1007                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1008                         break;
1009                 case Opt_err_ro:
1010                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1011                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1012                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1013                         break;
1014                 case Opt_err_cont:
1015                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1016                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1017                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1018                         break;
1019                 case Opt_nouid32:
1020                         set_opt (sbi->s_mount_opt, NO_UID32);
1021                         break;
1022                 case Opt_nocheck:
1023                         clear_opt (sbi->s_mount_opt, CHECK);
1024                         break;
1025                 case Opt_debug:
1026                         set_opt (sbi->s_mount_opt, DEBUG);
1027                         break;
1028                 case Opt_oldalloc:
1029                         set_opt (sbi->s_mount_opt, OLDALLOC);
1030                         break;
1031                 case Opt_orlov:
1032                         clear_opt (sbi->s_mount_opt, OLDALLOC);
1033                         break;
1034 #ifdef CONFIG_EXT3_FS_XATTR
1035                 case Opt_user_xattr:
1036                         set_opt (sbi->s_mount_opt, XATTR_USER);
1037                         break;
1038                 case Opt_nouser_xattr:
1039                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1040                         break;
1041 #else
1042                 case Opt_user_xattr:
1043                 case Opt_nouser_xattr:
1044                         ext3_msg(sb, KERN_INFO,
1045                                 "(no)user_xattr options not supported");
1046                         break;
1047 #endif
1048 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1049                 case Opt_acl:
1050                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1051                         break;
1052                 case Opt_noacl:
1053                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1054                         break;
1055 #else
1056                 case Opt_acl:
1057                 case Opt_noacl:
1058                         ext3_msg(sb, KERN_INFO,
1059                                 "(no)acl options not supported");
1060                         break;
1061 #endif
1062                 case Opt_reservation:
1063                         set_opt(sbi->s_mount_opt, RESERVATION);
1064                         break;
1065                 case Opt_noreservation:
1066                         clear_opt(sbi->s_mount_opt, RESERVATION);
1067                         break;
1068                 case Opt_journal_update:
1069                         /* @@@ FIXME */
1070                         /* Eventually we will want to be able to create
1071                            a journal file here.  For now, only allow the
1072                            user to specify an existing inode to be the
1073                            journal file. */
1074                         if (is_remount) {
1075                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1076                                         "journal on remount");
1077                                 return 0;
1078                         }
1079                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1080                         break;
1081                 case Opt_journal_inum:
1082                         if (is_remount) {
1083                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1084                                        "journal on remount");
1085                                 return 0;
1086                         }
1087                         if (match_int(&args[0], &option))
1088                                 return 0;
1089                         *inum = option;
1090                         break;
1091                 case Opt_journal_dev:
1092                         if (is_remount) {
1093                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1094                                        "journal on remount");
1095                                 return 0;
1096                         }
1097                         if (match_int(&args[0], &option))
1098                                 return 0;
1099                         *journal_devnum = option;
1100                         break;
1101                 case Opt_noload:
1102                         set_opt (sbi->s_mount_opt, NOLOAD);
1103                         break;
1104                 case Opt_commit:
1105                         if (match_int(&args[0], &option))
1106                                 return 0;
1107                         if (option < 0)
1108                                 return 0;
1109                         if (option == 0)
1110                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1111                         sbi->s_commit_interval = HZ * option;
1112                         break;
1113                 case Opt_data_journal:
1114                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1115                         goto datacheck;
1116                 case Opt_data_ordered:
1117                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1118                         goto datacheck;
1119                 case Opt_data_writeback:
1120                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1121                 datacheck:
1122                         if (is_remount) {
1123                                 if (test_opt(sb, DATA_FLAGS) == data_opt)
1124                                         break;
1125                                 ext3_msg(sb, KERN_ERR,
1126                                         "error: cannot change "
1127                                         "data mode on remount. The filesystem "
1128                                         "is mounted in data=%s mode and you "
1129                                         "try to remount it in data=%s mode.",
1130                                         data_mode_string(test_opt(sb,
1131                                                         DATA_FLAGS)),
1132                                         data_mode_string(data_opt));
1133                                 return 0;
1134                         } else {
1135                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1136                                 sbi->s_mount_opt |= data_opt;
1137                         }
1138                         break;
1139                 case Opt_data_err_abort:
1140                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1141                         break;
1142                 case Opt_data_err_ignore:
1143                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1144                         break;
1145 #ifdef CONFIG_QUOTA
1146                 case Opt_usrjquota:
1147                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1148                                 return 0;
1149                         break;
1150                 case Opt_grpjquota:
1151                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1152                                 return 0;
1153                         break;
1154                 case Opt_offusrjquota:
1155                         if (!clear_qf_name(sb, USRQUOTA))
1156                                 return 0;
1157                         break;
1158                 case Opt_offgrpjquota:
1159                         if (!clear_qf_name(sb, GRPQUOTA))
1160                                 return 0;
1161                         break;
1162                 case Opt_jqfmt_vfsold:
1163                         qfmt = QFMT_VFS_OLD;
1164                         goto set_qf_format;
1165                 case Opt_jqfmt_vfsv0:
1166                         qfmt = QFMT_VFS_V0;
1167                         goto set_qf_format;
1168                 case Opt_jqfmt_vfsv1:
1169                         qfmt = QFMT_VFS_V1;
1170 set_qf_format:
1171                         if (sb_any_quota_loaded(sb) &&
1172                             sbi->s_jquota_fmt != qfmt) {
1173                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1174                                         "journaled quota options when "
1175                                         "quota turned on.");
1176                                 return 0;
1177                         }
1178                         sbi->s_jquota_fmt = qfmt;
1179                         break;
1180                 case Opt_quota:
1181                 case Opt_usrquota:
1182                         set_opt(sbi->s_mount_opt, QUOTA);
1183                         set_opt(sbi->s_mount_opt, USRQUOTA);
1184                         break;
1185                 case Opt_grpquota:
1186                         set_opt(sbi->s_mount_opt, QUOTA);
1187                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1188                         break;
1189                 case Opt_noquota:
1190                         if (sb_any_quota_loaded(sb)) {
1191                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1192                                         "quota options when quota turned on.");
1193                                 return 0;
1194                         }
1195                         clear_opt(sbi->s_mount_opt, QUOTA);
1196                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1197                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1198                         break;
1199 #else
1200                 case Opt_quota:
1201                 case Opt_usrquota:
1202                 case Opt_grpquota:
1203                         ext3_msg(sb, KERN_ERR,
1204                                 "error: quota options not supported.");
1205                         break;
1206                 case Opt_usrjquota:
1207                 case Opt_grpjquota:
1208                 case Opt_offusrjquota:
1209                 case Opt_offgrpjquota:
1210                 case Opt_jqfmt_vfsold:
1211                 case Opt_jqfmt_vfsv0:
1212                 case Opt_jqfmt_vfsv1:
1213                         ext3_msg(sb, KERN_ERR,
1214                                 "error: journaled quota options not "
1215                                 "supported.");
1216                         break;
1217                 case Opt_noquota:
1218                         break;
1219 #endif
1220                 case Opt_abort:
1221                         set_opt(sbi->s_mount_opt, ABORT);
1222                         break;
1223                 case Opt_barrier:
1224                         if (match_int(&args[0], &option))
1225                                 return 0;
1226                         if (option)
1227                                 set_opt(sbi->s_mount_opt, BARRIER);
1228                         else
1229                                 clear_opt(sbi->s_mount_opt, BARRIER);
1230                         break;
1231                 case Opt_ignore:
1232                         break;
1233                 case Opt_resize:
1234                         if (!is_remount) {
1235                                 ext3_msg(sb, KERN_ERR,
1236                                         "error: resize option only available "
1237                                         "for remount");
1238                                 return 0;
1239                         }
1240                         if (match_int(&args[0], &option) != 0)
1241                                 return 0;
1242                         *n_blocks_count = option;
1243                         break;
1244                 case Opt_nobh:
1245                         set_opt(sbi->s_mount_opt, NOBH);
1246                         break;
1247                 case Opt_bh:
1248                         clear_opt(sbi->s_mount_opt, NOBH);
1249                         break;
1250                 default:
1251                         ext3_msg(sb, KERN_ERR,
1252                                 "error: unrecognized mount option \"%s\" "
1253                                 "or missing value", p);
1254                         return 0;
1255                 }
1256         }
1257 #ifdef CONFIG_QUOTA
1258         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1259                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1260                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1261                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1262                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1263
1264                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1265                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1266                                         "format mixing.");
1267                         return 0;
1268                 }
1269
1270                 if (!sbi->s_jquota_fmt) {
1271                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1272                                         "not specified.");
1273                         return 0;
1274                 }
1275         } else {
1276                 if (sbi->s_jquota_fmt) {
1277                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1278                                         "specified with no journaling "
1279                                         "enabled.");
1280                         return 0;
1281                 }
1282         }
1283 #endif
1284         return 1;
1285 }
1286
1287 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1288                             int read_only)
1289 {
1290         struct ext3_sb_info *sbi = EXT3_SB(sb);
1291         int res = 0;
1292
1293         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1294                 ext3_msg(sb, KERN_ERR,
1295                         "error: revision level too high, "
1296                         "forcing read-only mode");
1297                 res = MS_RDONLY;
1298         }
1299         if (read_only)
1300                 return res;
1301         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1302                 ext3_msg(sb, KERN_WARNING,
1303                         "warning: mounting unchecked fs, "
1304                         "running e2fsck is recommended");
1305         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1306                 ext3_msg(sb, KERN_WARNING,
1307                         "warning: mounting fs with errors, "
1308                         "running e2fsck is recommended");
1309         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1310                  le16_to_cpu(es->s_mnt_count) >=
1311                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1312                 ext3_msg(sb, KERN_WARNING,
1313                         "warning: maximal mount count reached, "
1314                         "running e2fsck is recommended");
1315         else if (le32_to_cpu(es->s_checkinterval) &&
1316                 (le32_to_cpu(es->s_lastcheck) +
1317                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1318                 ext3_msg(sb, KERN_WARNING,
1319                         "warning: checktime reached, "
1320                         "running e2fsck is recommended");
1321 #if 0
1322                 /* @@@ We _will_ want to clear the valid bit if we find
1323                    inconsistencies, to force a fsck at reboot.  But for
1324                    a plain journaled filesystem we can keep it set as
1325                    valid forever! :) */
1326         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1327 #endif
1328         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1329                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1330         le16_add_cpu(&es->s_mnt_count, 1);
1331         es->s_mtime = cpu_to_le32(get_seconds());
1332         ext3_update_dynamic_rev(sb);
1333         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1334
1335         ext3_commit_super(sb, es, 1);
1336         if (test_opt(sb, DEBUG))
1337                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1338                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1339                         sb->s_blocksize,
1340                         sbi->s_groups_count,
1341                         EXT3_BLOCKS_PER_GROUP(sb),
1342                         EXT3_INODES_PER_GROUP(sb),
1343                         sbi->s_mount_opt);
1344
1345         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1346                 char b[BDEVNAME_SIZE];
1347                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1348                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1349         } else {
1350                 ext3_msg(sb, KERN_INFO, "using internal journal");
1351         }
1352         return res;
1353 }
1354
1355 /* Called at mount-time, super-block is locked */
1356 static int ext3_check_descriptors(struct super_block *sb)
1357 {
1358         struct ext3_sb_info *sbi = EXT3_SB(sb);
1359         int i;
1360
1361         ext3_debug ("Checking group descriptors");
1362
1363         for (i = 0; i < sbi->s_groups_count; i++) {
1364                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1365                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1366                 ext3_fsblk_t last_block;
1367
1368                 if (i == sbi->s_groups_count - 1)
1369                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1370                 else
1371                         last_block = first_block +
1372                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1373
1374                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1375                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1376                 {
1377                         ext3_error (sb, "ext3_check_descriptors",
1378                                     "Block bitmap for group %d"
1379                                     " not in group (block %lu)!",
1380                                     i, (unsigned long)
1381                                         le32_to_cpu(gdp->bg_block_bitmap));
1382                         return 0;
1383                 }
1384                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1385                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1386                 {
1387                         ext3_error (sb, "ext3_check_descriptors",
1388                                     "Inode bitmap for group %d"
1389                                     " not in group (block %lu)!",
1390                                     i, (unsigned long)
1391                                         le32_to_cpu(gdp->bg_inode_bitmap));
1392                         return 0;
1393                 }
1394                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1395                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1396                     last_block)
1397                 {
1398                         ext3_error (sb, "ext3_check_descriptors",
1399                                     "Inode table for group %d"
1400                                     " not in group (block %lu)!",
1401                                     i, (unsigned long)
1402                                         le32_to_cpu(gdp->bg_inode_table));
1403                         return 0;
1404                 }
1405         }
1406
1407         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1408         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1409         return 1;
1410 }
1411
1412
1413 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1414  * the superblock) which were deleted from all directories, but held open by
1415  * a process at the time of a crash.  We walk the list and try to delete these
1416  * inodes at recovery time (only with a read-write filesystem).
1417  *
1418  * In order to keep the orphan inode chain consistent during traversal (in
1419  * case of crash during recovery), we link each inode into the superblock
1420  * orphan list_head and handle it the same way as an inode deletion during
1421  * normal operation (which journals the operations for us).
1422  *
1423  * We only do an iget() and an iput() on each inode, which is very safe if we
1424  * accidentally point at an in-use or already deleted inode.  The worst that
1425  * can happen in this case is that we get a "bit already cleared" message from
1426  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1427  * e2fsck was run on this filesystem, and it must have already done the orphan
1428  * inode cleanup for us, so we can safely abort without any further action.
1429  */
1430 static void ext3_orphan_cleanup (struct super_block * sb,
1431                                  struct ext3_super_block * es)
1432 {
1433         unsigned int s_flags = sb->s_flags;
1434         int nr_orphans = 0, nr_truncates = 0;
1435 #ifdef CONFIG_QUOTA
1436         int i;
1437 #endif
1438         if (!es->s_last_orphan) {
1439                 jbd_debug(4, "no orphan inodes to clean up\n");
1440                 return;
1441         }
1442
1443         if (bdev_read_only(sb->s_bdev)) {
1444                 ext3_msg(sb, KERN_ERR, "error: write access "
1445                         "unavailable, skipping orphan cleanup.");
1446                 return;
1447         }
1448
1449         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1450                 if (es->s_last_orphan)
1451                         jbd_debug(1, "Errors on filesystem, "
1452                                   "clearing orphan list.\n");
1453                 es->s_last_orphan = 0;
1454                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1455                 return;
1456         }
1457
1458         if (s_flags & MS_RDONLY) {
1459                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1460                 sb->s_flags &= ~MS_RDONLY;
1461         }
1462 #ifdef CONFIG_QUOTA
1463         /* Needed for iput() to work correctly and not trash data */
1464         sb->s_flags |= MS_ACTIVE;
1465         /* Turn on quotas so that they are updated correctly */
1466         for (i = 0; i < MAXQUOTAS; i++) {
1467                 if (EXT3_SB(sb)->s_qf_names[i]) {
1468                         int ret = ext3_quota_on_mount(sb, i);
1469                         if (ret < 0)
1470                                 ext3_msg(sb, KERN_ERR,
1471                                         "error: cannot turn on journaled "
1472                                         "quota: %d", ret);
1473                 }
1474         }
1475 #endif
1476
1477         while (es->s_last_orphan) {
1478                 struct inode *inode;
1479
1480                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1481                 if (IS_ERR(inode)) {
1482                         es->s_last_orphan = 0;
1483                         break;
1484                 }
1485
1486                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1487                 vfs_dq_init(inode);
1488                 if (inode->i_nlink) {
1489                         printk(KERN_DEBUG
1490                                 "%s: truncating inode %lu to %Ld bytes\n",
1491                                 __func__, inode->i_ino, inode->i_size);
1492                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1493                                   inode->i_ino, inode->i_size);
1494                         ext3_truncate(inode);
1495                         nr_truncates++;
1496                 } else {
1497                         printk(KERN_DEBUG
1498                                 "%s: deleting unreferenced inode %lu\n",
1499                                 __func__, inode->i_ino);
1500                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1501                                   inode->i_ino);
1502                         nr_orphans++;
1503                 }
1504                 iput(inode);  /* The delete magic happens here! */
1505         }
1506
1507 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1508
1509         if (nr_orphans)
1510                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1511                        PLURAL(nr_orphans));
1512         if (nr_truncates)
1513                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1514                        PLURAL(nr_truncates));
1515 #ifdef CONFIG_QUOTA
1516         /* Turn quotas off */
1517         for (i = 0; i < MAXQUOTAS; i++) {
1518                 if (sb_dqopt(sb)->files[i])
1519                         vfs_quota_off(sb, i, 0);
1520         }
1521 #endif
1522         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1523 }
1524
1525 /*
1526  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1527  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1528  * We need to be 1 filesystem block less than the 2^32 sector limit.
1529  */
1530 static loff_t ext3_max_size(int bits)
1531 {
1532         loff_t res = EXT3_NDIR_BLOCKS;
1533         int meta_blocks;
1534         loff_t upper_limit;
1535
1536         /* This is calculated to be the largest file size for a
1537          * dense, file such that the total number of
1538          * sectors in the file, including data and all indirect blocks,
1539          * does not exceed 2^32 -1
1540          * __u32 i_blocks representing the total number of
1541          * 512 bytes blocks of the file
1542          */
1543         upper_limit = (1LL << 32) - 1;
1544
1545         /* total blocks in file system block size */
1546         upper_limit >>= (bits - 9);
1547
1548
1549         /* indirect blocks */
1550         meta_blocks = 1;
1551         /* double indirect blocks */
1552         meta_blocks += 1 + (1LL << (bits-2));
1553         /* tripple indirect blocks */
1554         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1555
1556         upper_limit -= meta_blocks;
1557         upper_limit <<= bits;
1558
1559         res += 1LL << (bits-2);
1560         res += 1LL << (2*(bits-2));
1561         res += 1LL << (3*(bits-2));
1562         res <<= bits;
1563         if (res > upper_limit)
1564                 res = upper_limit;
1565
1566         if (res > MAX_LFS_FILESIZE)
1567                 res = MAX_LFS_FILESIZE;
1568
1569         return res;
1570 }
1571
1572 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1573                                     ext3_fsblk_t logic_sb_block,
1574                                     int nr)
1575 {
1576         struct ext3_sb_info *sbi = EXT3_SB(sb);
1577         unsigned long bg, first_meta_bg;
1578         int has_super = 0;
1579
1580         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1581
1582         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1583             nr < first_meta_bg)
1584                 return (logic_sb_block + nr + 1);
1585         bg = sbi->s_desc_per_block * nr;
1586         if (ext3_bg_has_super(sb, bg))
1587                 has_super = 1;
1588         return (has_super + ext3_group_first_block_no(sb, bg));
1589 }
1590
1591
1592 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1593 {
1594         struct buffer_head * bh;
1595         struct ext3_super_block *es = NULL;
1596         struct ext3_sb_info *sbi;
1597         ext3_fsblk_t block;
1598         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1599         ext3_fsblk_t logic_sb_block;
1600         unsigned long offset = 0;
1601         unsigned int journal_inum = 0;
1602         unsigned long journal_devnum = 0;
1603         unsigned long def_mount_opts;
1604         struct inode *root;
1605         int blocksize;
1606         int hblock;
1607         int db_count;
1608         int i;
1609         int needs_recovery;
1610         int ret = -EINVAL;
1611         __le32 features;
1612         int err;
1613
1614         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1615         if (!sbi)
1616                 return -ENOMEM;
1617
1618         sbi->s_blockgroup_lock =
1619                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1620         if (!sbi->s_blockgroup_lock) {
1621                 kfree(sbi);
1622                 return -ENOMEM;
1623         }
1624         sb->s_fs_info = sbi;
1625         sbi->s_mount_opt = 0;
1626         sbi->s_resuid = EXT3_DEF_RESUID;
1627         sbi->s_resgid = EXT3_DEF_RESGID;
1628         sbi->s_sb_block = sb_block;
1629
1630         unlock_kernel();
1631
1632         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1633         if (!blocksize) {
1634                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1635                 goto out_fail;
1636         }
1637
1638         /*
1639          * The ext3 superblock will not be buffer aligned for other than 1kB
1640          * block sizes.  We need to calculate the offset from buffer start.
1641          */
1642         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1643                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1644                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1645         } else {
1646                 logic_sb_block = sb_block;
1647         }
1648
1649         if (!(bh = sb_bread(sb, logic_sb_block))) {
1650                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1651                 goto out_fail;
1652         }
1653         /*
1654          * Note: s_es must be initialized as soon as possible because
1655          *       some ext3 macro-instructions depend on its value
1656          */
1657         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1658         sbi->s_es = es;
1659         sb->s_magic = le16_to_cpu(es->s_magic);
1660         if (sb->s_magic != EXT3_SUPER_MAGIC)
1661                 goto cantfind_ext3;
1662
1663         /* Set defaults before we parse the mount options */
1664         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1665         if (def_mount_opts & EXT3_DEFM_DEBUG)
1666                 set_opt(sbi->s_mount_opt, DEBUG);
1667         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1668                 set_opt(sbi->s_mount_opt, GRPID);
1669         if (def_mount_opts & EXT3_DEFM_UID16)
1670                 set_opt(sbi->s_mount_opt, NO_UID32);
1671 #ifdef CONFIG_EXT3_FS_XATTR
1672         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1673                 set_opt(sbi->s_mount_opt, XATTR_USER);
1674 #endif
1675 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1676         if (def_mount_opts & EXT3_DEFM_ACL)
1677                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1678 #endif
1679         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1680                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1681         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1682                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1683         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1684                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1685
1686         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1687                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1688         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1689                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1690         else
1691                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1692
1693         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1694         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1695
1696         set_opt(sbi->s_mount_opt, RESERVATION);
1697
1698         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1699                             NULL, 0))
1700                 goto failed_mount;
1701
1702         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1703                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1704
1705         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1706             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1707              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1708              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1709                 ext3_msg(sb, KERN_WARNING,
1710                         "warning: feature flags set on rev 0 fs, "
1711                         "running e2fsck is recommended");
1712         /*
1713          * Check feature flags regardless of the revision level, since we
1714          * previously didn't change the revision level when setting the flags,
1715          * so there is a chance incompat flags are set on a rev 0 filesystem.
1716          */
1717         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1718         if (features) {
1719                 ext3_msg(sb, KERN_ERR,
1720                         "error: couldn't mount because of unsupported "
1721                         "optional features (%x)", le32_to_cpu(features));
1722                 goto failed_mount;
1723         }
1724         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1725         if (!(sb->s_flags & MS_RDONLY) && features) {
1726                 ext3_msg(sb, KERN_ERR,
1727                         "error: couldn't mount RDWR because of unsupported "
1728                         "optional features (%x)", le32_to_cpu(features));
1729                 goto failed_mount;
1730         }
1731         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1732
1733         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1734             blocksize > EXT3_MAX_BLOCK_SIZE) {
1735                 ext3_msg(sb, KERN_ERR,
1736                         "error: couldn't mount because of unsupported "
1737                         "filesystem blocksize %d", blocksize);
1738                 goto failed_mount;
1739         }
1740
1741         hblock = bdev_logical_block_size(sb->s_bdev);
1742         if (sb->s_blocksize != blocksize) {
1743                 /*
1744                  * Make sure the blocksize for the filesystem is larger
1745                  * than the hardware sectorsize for the machine.
1746                  */
1747                 if (blocksize < hblock) {
1748                         ext3_msg(sb, KERN_ERR,
1749                                 "error: fsblocksize %d too small for "
1750                                 "hardware sectorsize %d", blocksize, hblock);
1751                         goto failed_mount;
1752                 }
1753
1754                 brelse (bh);
1755                 if (!sb_set_blocksize(sb, blocksize)) {
1756                         ext3_msg(sb, KERN_ERR,
1757                                 "error: bad blocksize %d", blocksize);
1758                         goto out_fail;
1759                 }
1760                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1761                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1762                 bh = sb_bread(sb, logic_sb_block);
1763                 if (!bh) {
1764                         ext3_msg(sb, KERN_ERR,
1765                                "error: can't read superblock on 2nd try");
1766                         goto failed_mount;
1767                 }
1768                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1769                 sbi->s_es = es;
1770                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1771                         ext3_msg(sb, KERN_ERR,
1772                                 "error: magic mismatch");
1773                         goto failed_mount;
1774                 }
1775         }
1776
1777         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1778
1779         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1780                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1781                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1782         } else {
1783                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1784                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1785                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1786                     (!is_power_of_2(sbi->s_inode_size)) ||
1787                     (sbi->s_inode_size > blocksize)) {
1788                         ext3_msg(sb, KERN_ERR,
1789                                 "error: unsupported inode size: %d",
1790                                 sbi->s_inode_size);
1791                         goto failed_mount;
1792                 }
1793         }
1794         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1795                                    le32_to_cpu(es->s_log_frag_size);
1796         if (blocksize != sbi->s_frag_size) {
1797                 ext3_msg(sb, KERN_ERR,
1798                        "error: fragsize %lu != blocksize %u (unsupported)",
1799                        sbi->s_frag_size, blocksize);
1800                 goto failed_mount;
1801         }
1802         sbi->s_frags_per_block = 1;
1803         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1804         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1805         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1806         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1807                 goto cantfind_ext3;
1808         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1809         if (sbi->s_inodes_per_block == 0)
1810                 goto cantfind_ext3;
1811         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1812                                         sbi->s_inodes_per_block;
1813         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1814         sbi->s_sbh = bh;
1815         sbi->s_mount_state = le16_to_cpu(es->s_state);
1816         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1817         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1818         for (i=0; i < 4; i++)
1819                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1820         sbi->s_def_hash_version = es->s_def_hash_version;
1821         i = le32_to_cpu(es->s_flags);
1822         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1823                 sbi->s_hash_unsigned = 3;
1824         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1825 #ifdef __CHAR_UNSIGNED__
1826                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1827                 sbi->s_hash_unsigned = 3;
1828 #else
1829                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1830 #endif
1831         }
1832
1833         if (sbi->s_blocks_per_group > blocksize * 8) {
1834                 ext3_msg(sb, KERN_ERR,
1835                         "#blocks per group too big: %lu",
1836                         sbi->s_blocks_per_group);
1837                 goto failed_mount;
1838         }
1839         if (sbi->s_frags_per_group > blocksize * 8) {
1840                 ext3_msg(sb, KERN_ERR,
1841                         "error: #fragments per group too big: %lu",
1842                         sbi->s_frags_per_group);
1843                 goto failed_mount;
1844         }
1845         if (sbi->s_inodes_per_group > blocksize * 8) {
1846                 ext3_msg(sb, KERN_ERR,
1847                         "error: #inodes per group too big: %lu",
1848                         sbi->s_inodes_per_group);
1849                 goto failed_mount;
1850         }
1851
1852         if (le32_to_cpu(es->s_blocks_count) >
1853                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1854                 ext3_msg(sb, KERN_ERR,
1855                         "error: filesystem is too large to mount safely");
1856                 if (sizeof(sector_t) < 8)
1857                         ext3_msg(sb, KERN_ERR,
1858                                 "error: CONFIG_LBDAF not enabled");
1859                 goto failed_mount;
1860         }
1861
1862         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1863                 goto cantfind_ext3;
1864         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1865                                le32_to_cpu(es->s_first_data_block) - 1)
1866                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1867         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1868                    EXT3_DESC_PER_BLOCK(sb);
1869         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1870                                     GFP_KERNEL);
1871         if (sbi->s_group_desc == NULL) {
1872                 ext3_msg(sb, KERN_ERR,
1873                         "error: not enough memory");
1874                 goto failed_mount;
1875         }
1876
1877         bgl_lock_init(sbi->s_blockgroup_lock);
1878
1879         for (i = 0; i < db_count; i++) {
1880                 block = descriptor_loc(sb, logic_sb_block, i);
1881                 sbi->s_group_desc[i] = sb_bread(sb, block);
1882                 if (!sbi->s_group_desc[i]) {
1883                         ext3_msg(sb, KERN_ERR,
1884                                 "error: can't read group descriptor %d", i);
1885                         db_count = i;
1886                         goto failed_mount2;
1887                 }
1888         }
1889         if (!ext3_check_descriptors (sb)) {
1890                 ext3_msg(sb, KERN_ERR,
1891                         "error: group descriptors corrupted");
1892                 goto failed_mount2;
1893         }
1894         sbi->s_gdb_count = db_count;
1895         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1896         spin_lock_init(&sbi->s_next_gen_lock);
1897
1898         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1899                         ext3_count_free_blocks(sb));
1900         if (!err) {
1901                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1902                                 ext3_count_free_inodes(sb));
1903         }
1904         if (!err) {
1905                 err = percpu_counter_init(&sbi->s_dirs_counter,
1906                                 ext3_count_dirs(sb));
1907         }
1908         if (err) {
1909                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
1910                 goto failed_mount3;
1911         }
1912
1913         /* per fileystem reservation list head & lock */
1914         spin_lock_init(&sbi->s_rsv_window_lock);
1915         sbi->s_rsv_window_root = RB_ROOT;
1916         /* Add a single, static dummy reservation to the start of the
1917          * reservation window list --- it gives us a placeholder for
1918          * append-at-start-of-list which makes the allocation logic
1919          * _much_ simpler. */
1920         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1921         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1922         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1923         sbi->s_rsv_window_head.rsv_goal_size = 0;
1924         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1925
1926         /*
1927          * set up enough so that it can read an inode
1928          */
1929         sb->s_op = &ext3_sops;
1930         sb->s_export_op = &ext3_export_ops;
1931         sb->s_xattr = ext3_xattr_handlers;
1932 #ifdef CONFIG_QUOTA
1933         sb->s_qcop = &ext3_qctl_operations;
1934         sb->dq_op = &ext3_quota_operations;
1935 #endif
1936         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1937         mutex_init(&sbi->s_orphan_lock);
1938         mutex_init(&sbi->s_resize_lock);
1939
1940         sb->s_root = NULL;
1941
1942         needs_recovery = (es->s_last_orphan != 0 ||
1943                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1944                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1945
1946         /*
1947          * The first inode we look at is the journal inode.  Don't try
1948          * root first: it may be modified in the journal!
1949          */
1950         if (!test_opt(sb, NOLOAD) &&
1951             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1952                 if (ext3_load_journal(sb, es, journal_devnum))
1953                         goto failed_mount3;
1954         } else if (journal_inum) {
1955                 if (ext3_create_journal(sb, es, journal_inum))
1956                         goto failed_mount3;
1957         } else {
1958                 if (!silent)
1959                         ext3_msg(sb, KERN_ERR,
1960                                 "error: no journal found. "
1961                                 "mounting ext3 over ext2?");
1962                 goto failed_mount3;
1963         }
1964
1965         /* We have now updated the journal if required, so we can
1966          * validate the data journaling mode. */
1967         switch (test_opt(sb, DATA_FLAGS)) {
1968         case 0:
1969                 /* No mode set, assume a default based on the journal
1970                    capabilities: ORDERED_DATA if the journal can
1971                    cope, else JOURNAL_DATA */
1972                 if (journal_check_available_features
1973                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1974                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
1975                 else
1976                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1977                 break;
1978
1979         case EXT3_MOUNT_ORDERED_DATA:
1980         case EXT3_MOUNT_WRITEBACK_DATA:
1981                 if (!journal_check_available_features
1982                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1983                         ext3_msg(sb, KERN_ERR,
1984                                 "error: journal does not support "
1985                                 "requested data journaling mode");
1986                         goto failed_mount4;
1987                 }
1988         default:
1989                 break;
1990         }
1991
1992         if (test_opt(sb, NOBH)) {
1993                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1994                         ext3_msg(sb, KERN_WARNING,
1995                                 "warning: ignoring nobh option - "
1996                                 "it is supported only with writeback mode");
1997                         clear_opt(sbi->s_mount_opt, NOBH);
1998                 }
1999         }
2000         /*
2001          * The journal_load will have done any necessary log recovery,
2002          * so we can safely mount the rest of the filesystem now.
2003          */
2004
2005         root = ext3_iget(sb, EXT3_ROOT_INO);
2006         if (IS_ERR(root)) {
2007                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
2008                 ret = PTR_ERR(root);
2009                 goto failed_mount4;
2010         }
2011         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2012                 iput(root);
2013                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2014                 goto failed_mount4;
2015         }
2016         sb->s_root = d_alloc_root(root);
2017         if (!sb->s_root) {
2018                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2019                 iput(root);
2020                 ret = -ENOMEM;
2021                 goto failed_mount4;
2022         }
2023
2024         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2025
2026         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2027         ext3_orphan_cleanup(sb, es);
2028         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2029         if (needs_recovery)
2030                 ext3_msg(sb, KERN_INFO, "recovery complete");
2031         ext3_mark_recovery_complete(sb, es);
2032         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2033                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2034                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2035                 "writeback");
2036
2037         lock_kernel();
2038         return 0;
2039
2040 cantfind_ext3:
2041         if (!silent)
2042                 ext3_msg(sb, KERN_INFO,
2043                         "error: can't find ext3 filesystem on dev %s.",
2044                        sb->s_id);
2045         goto failed_mount;
2046
2047 failed_mount4:
2048         journal_destroy(sbi->s_journal);
2049 failed_mount3:
2050         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2051         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2052         percpu_counter_destroy(&sbi->s_dirs_counter);
2053 failed_mount2:
2054         for (i = 0; i < db_count; i++)
2055                 brelse(sbi->s_group_desc[i]);
2056         kfree(sbi->s_group_desc);
2057 failed_mount:
2058 #ifdef CONFIG_QUOTA
2059         for (i = 0; i < MAXQUOTAS; i++)
2060                 kfree(sbi->s_qf_names[i]);
2061 #endif
2062         ext3_blkdev_remove(sbi);
2063         brelse(bh);
2064 out_fail:
2065         sb->s_fs_info = NULL;
2066         kfree(sbi->s_blockgroup_lock);
2067         kfree(sbi);
2068         lock_kernel();
2069         return ret;
2070 }
2071
2072 /*
2073  * Setup any per-fs journal parameters now.  We'll do this both on
2074  * initial mount, once the journal has been initialised but before we've
2075  * done any recovery; and again on any subsequent remount.
2076  */
2077 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2078 {
2079         struct ext3_sb_info *sbi = EXT3_SB(sb);
2080
2081         if (sbi->s_commit_interval)
2082                 journal->j_commit_interval = sbi->s_commit_interval;
2083         /* We could also set up an ext3-specific default for the commit
2084          * interval here, but for now we'll just fall back to the jbd
2085          * default. */
2086
2087         spin_lock(&journal->j_state_lock);
2088         if (test_opt(sb, BARRIER))
2089                 journal->j_flags |= JFS_BARRIER;
2090         else
2091                 journal->j_flags &= ~JFS_BARRIER;
2092         if (test_opt(sb, DATA_ERR_ABORT))
2093                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2094         else
2095                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2096         spin_unlock(&journal->j_state_lock);
2097 }
2098
2099 static journal_t *ext3_get_journal(struct super_block *sb,
2100                                    unsigned int journal_inum)
2101 {
2102         struct inode *journal_inode;
2103         journal_t *journal;
2104
2105         /* First, test for the existence of a valid inode on disk.  Bad
2106          * things happen if we iget() an unused inode, as the subsequent
2107          * iput() will try to delete it. */
2108
2109         journal_inode = ext3_iget(sb, journal_inum);
2110         if (IS_ERR(journal_inode)) {
2111                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2112                 return NULL;
2113         }
2114         if (!journal_inode->i_nlink) {
2115                 make_bad_inode(journal_inode);
2116                 iput(journal_inode);
2117                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2118                 return NULL;
2119         }
2120
2121         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2122                   journal_inode, journal_inode->i_size);
2123         if (!S_ISREG(journal_inode->i_mode)) {
2124                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2125                 iput(journal_inode);
2126                 return NULL;
2127         }
2128
2129         journal = journal_init_inode(journal_inode);
2130         if (!journal) {
2131                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2132                 iput(journal_inode);
2133                 return NULL;
2134         }
2135         journal->j_private = sb;
2136         ext3_init_journal_params(sb, journal);
2137         return journal;
2138 }
2139
2140 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2141                                        dev_t j_dev)
2142 {
2143         struct buffer_head * bh;
2144         journal_t *journal;
2145         ext3_fsblk_t start;
2146         ext3_fsblk_t len;
2147         int hblock, blocksize;
2148         ext3_fsblk_t sb_block;
2149         unsigned long offset;
2150         struct ext3_super_block * es;
2151         struct block_device *bdev;
2152
2153         bdev = ext3_blkdev_get(j_dev, sb);
2154         if (bdev == NULL)
2155                 return NULL;
2156
2157         if (bd_claim(bdev, sb)) {
2158                 ext3_msg(sb, KERN_ERR,
2159                         "error: failed to claim external journal device");
2160                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2161                 return NULL;
2162         }
2163
2164         blocksize = sb->s_blocksize;
2165         hblock = bdev_logical_block_size(bdev);
2166         if (blocksize < hblock) {
2167                 ext3_msg(sb, KERN_ERR,
2168                         "error: blocksize too small for journal device");
2169                 goto out_bdev;
2170         }
2171
2172         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2173         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2174         set_blocksize(bdev, blocksize);
2175         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2176                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2177                         "external journal");
2178                 goto out_bdev;
2179         }
2180
2181         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2182         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2183             !(le32_to_cpu(es->s_feature_incompat) &
2184               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2185                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2186                         "bad superblock");
2187                 brelse(bh);
2188                 goto out_bdev;
2189         }
2190
2191         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2192                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2193                 brelse(bh);
2194                 goto out_bdev;
2195         }
2196
2197         len = le32_to_cpu(es->s_blocks_count);
2198         start = sb_block + 1;
2199         brelse(bh);     /* we're done with the superblock */
2200
2201         journal = journal_init_dev(bdev, sb->s_bdev,
2202                                         start, len, blocksize);
2203         if (!journal) {
2204                 ext3_msg(sb, KERN_ERR,
2205                         "error: failed to create device journal");
2206                 goto out_bdev;
2207         }
2208         journal->j_private = sb;
2209         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2210         wait_on_buffer(journal->j_sb_buffer);
2211         if (!buffer_uptodate(journal->j_sb_buffer)) {
2212                 ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2213                 goto out_journal;
2214         }
2215         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2216                 ext3_msg(sb, KERN_ERR,
2217                         "error: external journal has more than one "
2218                         "user (unsupported) - %d",
2219                         be32_to_cpu(journal->j_superblock->s_nr_users));
2220                 goto out_journal;
2221         }
2222         EXT3_SB(sb)->journal_bdev = bdev;
2223         ext3_init_journal_params(sb, journal);
2224         return journal;
2225 out_journal:
2226         journal_destroy(journal);
2227 out_bdev:
2228         ext3_blkdev_put(bdev);
2229         return NULL;
2230 }
2231
2232 static int ext3_load_journal(struct super_block *sb,
2233                              struct ext3_super_block *es,
2234                              unsigned long journal_devnum)
2235 {
2236         journal_t *journal;
2237         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2238         dev_t journal_dev;
2239         int err = 0;
2240         int really_read_only;
2241
2242         if (journal_devnum &&
2243             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2244                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2245                         "numbers have changed");
2246                 journal_dev = new_decode_dev(journal_devnum);
2247         } else
2248                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2249
2250         really_read_only = bdev_read_only(sb->s_bdev);
2251
2252         /*
2253          * Are we loading a blank journal or performing recovery after a
2254          * crash?  For recovery, we need to check in advance whether we
2255          * can get read-write access to the device.
2256          */
2257
2258         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2259                 if (sb->s_flags & MS_RDONLY) {
2260                         ext3_msg(sb, KERN_INFO,
2261                                 "recovery required on readonly filesystem");
2262                         if (really_read_only) {
2263                                 ext3_msg(sb, KERN_ERR, "error: write access "
2264                                         "unavailable, cannot proceed");
2265                                 return -EROFS;
2266                         }
2267                         ext3_msg(sb, KERN_INFO,
2268                                 "write access will be enabled during recovery");
2269                 }
2270         }
2271
2272         if (journal_inum && journal_dev) {
2273                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2274                        "and inode journals");
2275                 return -EINVAL;
2276         }
2277
2278         if (journal_inum) {
2279                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2280                         return -EINVAL;
2281         } else {
2282                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2283                         return -EINVAL;
2284         }
2285
2286         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2287                 err = journal_update_format(journal);
2288                 if (err)  {
2289                         ext3_msg(sb, KERN_ERR, "error updating journal");
2290                         journal_destroy(journal);
2291                         return err;
2292                 }
2293         }
2294
2295         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2296                 err = journal_wipe(journal, !really_read_only);
2297         if (!err)
2298                 err = journal_load(journal);
2299
2300         if (err) {
2301                 ext3_msg(sb, KERN_ERR, "error loading journal");
2302                 journal_destroy(journal);
2303                 return err;
2304         }
2305
2306         EXT3_SB(sb)->s_journal = journal;
2307         ext3_clear_journal_err(sb, es);
2308
2309         if (journal_devnum &&
2310             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2311                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2312
2313                 /* Make sure we flush the recovery flag to disk. */
2314                 ext3_commit_super(sb, es, 1);
2315         }
2316
2317         return 0;
2318 }
2319
2320 static int ext3_create_journal(struct super_block *sb,
2321                                struct ext3_super_block *es,
2322                                unsigned int journal_inum)
2323 {
2324         journal_t *journal;
2325         int err;
2326
2327         if (sb->s_flags & MS_RDONLY) {
2328                 ext3_msg(sb, KERN_ERR,
2329                         "error: readonly filesystem when trying to "
2330                         "create journal");
2331                 return -EROFS;
2332         }
2333
2334         journal = ext3_get_journal(sb, journal_inum);
2335         if (!journal)
2336                 return -EINVAL;
2337
2338         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2339                journal_inum);
2340
2341         err = journal_create(journal);
2342         if (err) {
2343                 ext3_msg(sb, KERN_ERR, "error creating journal");
2344                 journal_destroy(journal);
2345                 return -EIO;
2346         }
2347
2348         EXT3_SB(sb)->s_journal = journal;
2349
2350         ext3_update_dynamic_rev(sb);
2351         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2352         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2353
2354         es->s_journal_inum = cpu_to_le32(journal_inum);
2355
2356         /* Make sure we flush the recovery flag to disk. */
2357         ext3_commit_super(sb, es, 1);
2358
2359         return 0;
2360 }
2361
2362 static int ext3_commit_super(struct super_block *sb,
2363                                struct ext3_super_block *es,
2364                                int sync)
2365 {
2366         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2367         int error = 0;
2368
2369         if (!sbh)
2370                 return error;
2371         /*
2372          * If the file system is mounted read-only, don't update the
2373          * superblock write time.  This avoids updating the superblock
2374          * write time when we are mounting the root file system
2375          * read/only but we need to replay the journal; at that point,
2376          * for people who are east of GMT and who make their clock
2377          * tick in localtime for Windows bug-for-bug compatibility,
2378          * the clock is set in the future, and this will cause e2fsck
2379          * to complain and force a full file system check.
2380          */
2381         if (!(sb->s_flags & MS_RDONLY))
2382                 es->s_wtime = cpu_to_le32(get_seconds());
2383         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2384         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2385         BUFFER_TRACE(sbh, "marking dirty");
2386         mark_buffer_dirty(sbh);
2387         if (sync)
2388                 error = sync_dirty_buffer(sbh);
2389         return error;
2390 }
2391
2392
2393 /*
2394  * Have we just finished recovery?  If so, and if we are mounting (or
2395  * remounting) the filesystem readonly, then we will end up with a
2396  * consistent fs on disk.  Record that fact.
2397  */
2398 static void ext3_mark_recovery_complete(struct super_block * sb,
2399                                         struct ext3_super_block * es)
2400 {
2401         journal_t *journal = EXT3_SB(sb)->s_journal;
2402
2403         journal_lock_updates(journal);
2404         if (journal_flush(journal) < 0)
2405                 goto out;
2406
2407         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2408             sb->s_flags & MS_RDONLY) {
2409                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2410                 ext3_commit_super(sb, es, 1);
2411         }
2412
2413 out:
2414         journal_unlock_updates(journal);
2415 }
2416
2417 /*
2418  * If we are mounting (or read-write remounting) a filesystem whose journal
2419  * has recorded an error from a previous lifetime, move that error to the
2420  * main filesystem now.
2421  */
2422 static void ext3_clear_journal_err(struct super_block *sb,
2423                                    struct ext3_super_block *es)
2424 {
2425         journal_t *journal;
2426         int j_errno;
2427         const char *errstr;
2428
2429         journal = EXT3_SB(sb)->s_journal;
2430
2431         /*
2432          * Now check for any error status which may have been recorded in the
2433          * journal by a prior ext3_error() or ext3_abort()
2434          */
2435
2436         j_errno = journal_errno(journal);
2437         if (j_errno) {
2438                 char nbuf[16];
2439
2440                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2441                 ext3_warning(sb, __func__, "Filesystem error recorded "
2442                              "from previous mount: %s", errstr);
2443                 ext3_warning(sb, __func__, "Marking fs in need of "
2444                              "filesystem check.");
2445
2446                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2447                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2448                 ext3_commit_super (sb, es, 1);
2449
2450                 journal_clear_err(journal);
2451         }
2452 }
2453
2454 /*
2455  * Force the running and committing transactions to commit,
2456  * and wait on the commit.
2457  */
2458 int ext3_force_commit(struct super_block *sb)
2459 {
2460         journal_t *journal;
2461         int ret;
2462
2463         if (sb->s_flags & MS_RDONLY)
2464                 return 0;
2465
2466         journal = EXT3_SB(sb)->s_journal;
2467         ret = ext3_journal_force_commit(journal);
2468         return ret;
2469 }
2470
2471 static int ext3_sync_fs(struct super_block *sb, int wait)
2472 {
2473         tid_t target;
2474
2475         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2476                 if (wait)
2477                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2478         }
2479         return 0;
2480 }
2481
2482 /*
2483  * LVM calls this function before a (read-only) snapshot is created.  This
2484  * gives us a chance to flush the journal completely and mark the fs clean.
2485  */
2486 static int ext3_freeze(struct super_block *sb)
2487 {
2488         int error = 0;
2489         journal_t *journal;
2490
2491         if (!(sb->s_flags & MS_RDONLY)) {
2492                 journal = EXT3_SB(sb)->s_journal;
2493
2494                 /* Now we set up the journal barrier. */
2495                 journal_lock_updates(journal);
2496
2497                 /*
2498                  * We don't want to clear needs_recovery flag when we failed
2499                  * to flush the journal.
2500                  */
2501                 error = journal_flush(journal);
2502                 if (error < 0)
2503                         goto out;
2504
2505                 /* Journal blocked and flushed, clear needs_recovery flag. */
2506                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2507                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2508                 if (error)
2509                         goto out;
2510         }
2511         return 0;
2512
2513 out:
2514         journal_unlock_updates(journal);
2515         return error;
2516 }
2517
2518 /*
2519  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2520  * flag here, even though the filesystem is not technically dirty yet.
2521  */
2522 static int ext3_unfreeze(struct super_block *sb)
2523 {
2524         if (!(sb->s_flags & MS_RDONLY)) {
2525                 lock_super(sb);
2526                 /* Reser the needs_recovery flag before the fs is unlocked. */
2527                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2528                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2529                 unlock_super(sb);
2530                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2531         }
2532         return 0;
2533 }
2534
2535 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2536 {
2537         struct ext3_super_block * es;
2538         struct ext3_sb_info *sbi = EXT3_SB(sb);
2539         ext3_fsblk_t n_blocks_count = 0;
2540         unsigned long old_sb_flags;
2541         struct ext3_mount_options old_opts;
2542         int err;
2543 #ifdef CONFIG_QUOTA
2544         int i;
2545 #endif
2546
2547         lock_kernel();
2548
2549         /* Store the original options */
2550         lock_super(sb);
2551         old_sb_flags = sb->s_flags;
2552         old_opts.s_mount_opt = sbi->s_mount_opt;
2553         old_opts.s_resuid = sbi->s_resuid;
2554         old_opts.s_resgid = sbi->s_resgid;
2555         old_opts.s_commit_interval = sbi->s_commit_interval;
2556 #ifdef CONFIG_QUOTA
2557         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2558         for (i = 0; i < MAXQUOTAS; i++)
2559                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2560 #endif
2561
2562         /*
2563          * Allow the "check" option to be passed as a remount option.
2564          */
2565         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2566                 err = -EINVAL;
2567                 goto restore_opts;
2568         }
2569
2570         if (test_opt(sb, ABORT))
2571                 ext3_abort(sb, __func__, "Abort forced by user");
2572
2573         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2574                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2575
2576         es = sbi->s_es;
2577
2578         ext3_init_journal_params(sb, sbi->s_journal);
2579
2580         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2581                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2582                 if (test_opt(sb, ABORT)) {
2583                         err = -EROFS;
2584                         goto restore_opts;
2585                 }
2586
2587                 if (*flags & MS_RDONLY) {
2588                         /*
2589                          * First of all, the unconditional stuff we have to do
2590                          * to disable replay of the journal when we next remount
2591                          */
2592                         sb->s_flags |= MS_RDONLY;
2593
2594                         /*
2595                          * OK, test if we are remounting a valid rw partition
2596                          * readonly, and if so set the rdonly flag and then
2597                          * mark the partition as valid again.
2598                          */
2599                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2600                             (sbi->s_mount_state & EXT3_VALID_FS))
2601                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2602
2603                         ext3_mark_recovery_complete(sb, es);
2604                 } else {
2605                         __le32 ret;
2606                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2607                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2608                                 ext3_msg(sb, KERN_WARNING,
2609                                         "warning: couldn't remount RDWR "
2610                                         "because of unsupported optional "
2611                                         "features (%x)", le32_to_cpu(ret));
2612                                 err = -EROFS;
2613                                 goto restore_opts;
2614                         }
2615
2616                         /*
2617                          * If we have an unprocessed orphan list hanging
2618                          * around from a previously readonly bdev mount,
2619                          * require a full umount/remount for now.
2620                          */
2621                         if (es->s_last_orphan) {
2622                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2623                                        "remount RDWR because of unprocessed "
2624                                        "orphan inode list.  Please "
2625                                        "umount/remount instead.");
2626                                 err = -EINVAL;
2627                                 goto restore_opts;
2628                         }
2629
2630                         /*
2631                          * Mounting a RDONLY partition read-write, so reread
2632                          * and store the current valid flag.  (It may have
2633                          * been changed by e2fsck since we originally mounted
2634                          * the partition.)
2635                          */
2636                         ext3_clear_journal_err(sb, es);
2637                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2638                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2639                                 goto restore_opts;
2640                         if (!ext3_setup_super (sb, es, 0))
2641                                 sb->s_flags &= ~MS_RDONLY;
2642                 }
2643         }
2644 #ifdef CONFIG_QUOTA
2645         /* Release old quota file names */
2646         for (i = 0; i < MAXQUOTAS; i++)
2647                 if (old_opts.s_qf_names[i] &&
2648                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2649                         kfree(old_opts.s_qf_names[i]);
2650 #endif
2651         unlock_super(sb);
2652         unlock_kernel();
2653         return 0;
2654 restore_opts:
2655         sb->s_flags = old_sb_flags;
2656         sbi->s_mount_opt = old_opts.s_mount_opt;
2657         sbi->s_resuid = old_opts.s_resuid;
2658         sbi->s_resgid = old_opts.s_resgid;
2659         sbi->s_commit_interval = old_opts.s_commit_interval;
2660 #ifdef CONFIG_QUOTA
2661         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2662         for (i = 0; i < MAXQUOTAS; i++) {
2663                 if (sbi->s_qf_names[i] &&
2664                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2665                         kfree(sbi->s_qf_names[i]);
2666                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2667         }
2668 #endif
2669         unlock_super(sb);
2670         unlock_kernel();
2671         return err;
2672 }
2673
2674 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2675 {
2676         struct super_block *sb = dentry->d_sb;
2677         struct ext3_sb_info *sbi = EXT3_SB(sb);
2678         struct ext3_super_block *es = sbi->s_es;
2679         u64 fsid;
2680
2681         if (test_opt(sb, MINIX_DF)) {
2682                 sbi->s_overhead_last = 0;
2683         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2684                 unsigned long ngroups = sbi->s_groups_count, i;
2685                 ext3_fsblk_t overhead = 0;
2686                 smp_rmb();
2687
2688                 /*
2689                  * Compute the overhead (FS structures).  This is constant
2690                  * for a given filesystem unless the number of block groups
2691                  * changes so we cache the previous value until it does.
2692                  */
2693
2694                 /*
2695                  * All of the blocks before first_data_block are
2696                  * overhead
2697                  */
2698                 overhead = le32_to_cpu(es->s_first_data_block);
2699
2700                 /*
2701                  * Add the overhead attributed to the superblock and
2702                  * block group descriptors.  If the sparse superblocks
2703                  * feature is turned on, then not all groups have this.
2704                  */
2705                 for (i = 0; i < ngroups; i++) {
2706                         overhead += ext3_bg_has_super(sb, i) +
2707                                 ext3_bg_num_gdb(sb, i);
2708                         cond_resched();
2709                 }
2710
2711                 /*
2712                  * Every block group has an inode bitmap, a block
2713                  * bitmap, and an inode table.
2714                  */
2715                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2716                 sbi->s_overhead_last = overhead;
2717                 smp_wmb();
2718                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2719         }
2720
2721         buf->f_type = EXT3_SUPER_MAGIC;
2722         buf->f_bsize = sb->s_blocksize;
2723         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2724         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2725         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2726         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2727                 buf->f_bavail = 0;
2728         buf->f_files = le32_to_cpu(es->s_inodes_count);
2729         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2730         buf->f_namelen = EXT3_NAME_LEN;
2731         fsid = le64_to_cpup((void *)es->s_uuid) ^
2732                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2733         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2734         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2735         return 0;
2736 }
2737
2738 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2739  * is locked for write. Otherwise the are possible deadlocks:
2740  * Process 1                         Process 2
2741  * ext3_create()                     quota_sync()
2742  *   journal_start()                   write_dquot()
2743  *   vfs_dq_init()                       down(dqio_mutex)
2744  *     down(dqio_mutex)                    journal_start()
2745  *
2746  */
2747
2748 #ifdef CONFIG_QUOTA
2749
2750 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2751 {
2752         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2753 }
2754
2755 static int ext3_write_dquot(struct dquot *dquot)
2756 {
2757         int ret, err;
2758         handle_t *handle;
2759         struct inode *inode;
2760
2761         inode = dquot_to_inode(dquot);
2762         handle = ext3_journal_start(inode,
2763                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2764         if (IS_ERR(handle))
2765                 return PTR_ERR(handle);
2766         ret = dquot_commit(dquot);
2767         err = ext3_journal_stop(handle);
2768         if (!ret)
2769                 ret = err;
2770         return ret;
2771 }
2772
2773 static int ext3_acquire_dquot(struct dquot *dquot)
2774 {
2775         int ret, err;
2776         handle_t *handle;
2777
2778         handle = ext3_journal_start(dquot_to_inode(dquot),
2779                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2780         if (IS_ERR(handle))
2781                 return PTR_ERR(handle);
2782         ret = dquot_acquire(dquot);
2783         err = ext3_journal_stop(handle);
2784         if (!ret)
2785                 ret = err;
2786         return ret;
2787 }
2788
2789 static int ext3_release_dquot(struct dquot *dquot)
2790 {
2791         int ret, err;
2792         handle_t *handle;
2793
2794         handle = ext3_journal_start(dquot_to_inode(dquot),
2795                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2796         if (IS_ERR(handle)) {
2797                 /* Release dquot anyway to avoid endless cycle in dqput() */
2798                 dquot_release(dquot);
2799                 return PTR_ERR(handle);
2800         }
2801         ret = dquot_release(dquot);
2802         err = ext3_journal_stop(handle);
2803         if (!ret)
2804                 ret = err;
2805         return ret;
2806 }
2807
2808 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2809 {
2810         /* Are we journaling quotas? */
2811         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2812             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2813                 dquot_mark_dquot_dirty(dquot);
2814                 return ext3_write_dquot(dquot);
2815         } else {
2816                 return dquot_mark_dquot_dirty(dquot);
2817         }
2818 }
2819
2820 static int ext3_write_info(struct super_block *sb, int type)
2821 {
2822         int ret, err;
2823         handle_t *handle;
2824
2825         /* Data block + inode block */
2826         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2827         if (IS_ERR(handle))
2828                 return PTR_ERR(handle);
2829         ret = dquot_commit_info(sb, type);
2830         err = ext3_journal_stop(handle);
2831         if (!ret)
2832                 ret = err;
2833         return ret;
2834 }
2835
2836 /*
2837  * Turn on quotas during mount time - we need to find
2838  * the quota file and such...
2839  */
2840 static int ext3_quota_on_mount(struct super_block *sb, int type)
2841 {
2842         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2843                         EXT3_SB(sb)->s_jquota_fmt, type);
2844 }
2845
2846 /*
2847  * Standard function to be called on quota_on
2848  */
2849 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2850                          char *name, int remount)
2851 {
2852         int err;
2853         struct path path;
2854
2855         if (!test_opt(sb, QUOTA))
2856                 return -EINVAL;
2857         /* When remounting, no checks are needed and in fact, name is NULL */
2858         if (remount)
2859                 return vfs_quota_on(sb, type, format_id, name, remount);
2860
2861         err = kern_path(name, LOOKUP_FOLLOW, &path);
2862         if (err)
2863                 return err;
2864
2865         /* Quotafile not on the same filesystem? */
2866         if (path.mnt->mnt_sb != sb) {
2867                 path_put(&path);
2868                 return -EXDEV;
2869         }
2870         /* Journaling quota? */
2871         if (EXT3_SB(sb)->s_qf_names[type]) {
2872                 /* Quotafile not of fs root? */
2873                 if (path.dentry->d_parent != sb->s_root)
2874                         ext3_msg(sb, KERN_WARNING,
2875                                 "warning: Quota file not on filesystem root. "
2876                                 "Journaled quota will not work.");
2877         }
2878
2879         /*
2880          * When we journal data on quota file, we have to flush journal to see
2881          * all updates to the file when we bypass pagecache...
2882          */
2883         if (ext3_should_journal_data(path.dentry->d_inode)) {
2884                 /*
2885                  * We don't need to lock updates but journal_flush() could
2886                  * otherwise be livelocked...
2887                  */
2888                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2889                 err = journal_flush(EXT3_SB(sb)->s_journal);
2890                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2891                 if (err) {
2892                         path_put(&path);
2893                         return err;
2894                 }
2895         }
2896
2897         err = vfs_quota_on_path(sb, type, format_id, &path);
2898         path_put(&path);
2899         return err;
2900 }
2901
2902 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2903  * acquiring the locks... As quota files are never truncated and quota code
2904  * itself serializes the operations (and noone else should touch the files)
2905  * we don't have to be afraid of races */
2906 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2907                                size_t len, loff_t off)
2908 {
2909         struct inode *inode = sb_dqopt(sb)->files[type];
2910         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2911         int err = 0;
2912         int offset = off & (sb->s_blocksize - 1);
2913         int tocopy;
2914         size_t toread;
2915         struct buffer_head *bh;
2916         loff_t i_size = i_size_read(inode);
2917
2918         if (off > i_size)
2919                 return 0;
2920         if (off+len > i_size)
2921                 len = i_size-off;
2922         toread = len;
2923         while (toread > 0) {
2924                 tocopy = sb->s_blocksize - offset < toread ?
2925                                 sb->s_blocksize - offset : toread;
2926                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2927                 if (err)
2928                         return err;
2929                 if (!bh)        /* A hole? */
2930                         memset(data, 0, tocopy);
2931                 else
2932                         memcpy(data, bh->b_data+offset, tocopy);
2933                 brelse(bh);
2934                 offset = 0;
2935                 toread -= tocopy;
2936                 data += tocopy;
2937                 blk++;
2938         }
2939         return len;
2940 }
2941
2942 /* Write to quotafile (we know the transaction is already started and has
2943  * enough credits) */
2944 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2945                                 const char *data, size_t len, loff_t off)
2946 {
2947         struct inode *inode = sb_dqopt(sb)->files[type];
2948         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2949         int err = 0;
2950         int offset = off & (sb->s_blocksize - 1);
2951         int tocopy;
2952         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2953         size_t towrite = len;
2954         struct buffer_head *bh;
2955         handle_t *handle = journal_current_handle();
2956
2957         if (!handle) {
2958                 ext3_msg(sb, KERN_WARNING,
2959                         "warning: quota write (off=%llu, len=%llu)"
2960                         " cancelled because transaction is not started.",
2961                         (unsigned long long)off, (unsigned long long)len);
2962                 return -EIO;
2963         }
2964         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2965         while (towrite > 0) {
2966                 tocopy = sb->s_blocksize - offset < towrite ?
2967                                 sb->s_blocksize - offset : towrite;
2968                 bh = ext3_bread(handle, inode, blk, 1, &err);
2969                 if (!bh)
2970                         goto out;
2971                 if (journal_quota) {
2972                         err = ext3_journal_get_write_access(handle, bh);
2973                         if (err) {
2974                                 brelse(bh);
2975                                 goto out;
2976                         }
2977                 }
2978                 lock_buffer(bh);
2979                 memcpy(bh->b_data+offset, data, tocopy);
2980                 flush_dcache_page(bh->b_page);
2981                 unlock_buffer(bh);
2982                 if (journal_quota)
2983                         err = ext3_journal_dirty_metadata(handle, bh);
2984                 else {
2985                         /* Always do at least ordered writes for quotas */
2986                         err = ext3_journal_dirty_data(handle, bh);
2987                         mark_buffer_dirty(bh);
2988                 }
2989                 brelse(bh);
2990                 if (err)
2991                         goto out;
2992                 offset = 0;
2993                 towrite -= tocopy;
2994                 data += tocopy;
2995                 blk++;
2996         }
2997 out:
2998         if (len == towrite) {
2999                 mutex_unlock(&inode->i_mutex);
3000                 return err;
3001         }
3002         if (inode->i_size < off+len-towrite) {
3003                 i_size_write(inode, off+len-towrite);
3004                 EXT3_I(inode)->i_disksize = inode->i_size;
3005         }
3006         inode->i_version++;
3007         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3008         ext3_mark_inode_dirty(handle, inode);
3009         mutex_unlock(&inode->i_mutex);
3010         return len - towrite;
3011 }
3012
3013 #endif
3014
3015 static int ext3_get_sb(struct file_system_type *fs_type,
3016         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3017 {
3018         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
3019 }
3020
3021 static struct file_system_type ext3_fs_type = {
3022         .owner          = THIS_MODULE,
3023         .name           = "ext3",
3024         .get_sb         = ext3_get_sb,
3025         .kill_sb        = kill_block_super,
3026         .fs_flags       = FS_REQUIRES_DEV,
3027 };
3028
3029 static int __init init_ext3_fs(void)
3030 {
3031         int err = init_ext3_xattr();
3032         if (err)
3033                 return err;
3034         err = init_inodecache();
3035         if (err)
3036                 goto out1;
3037         err = register_filesystem(&ext3_fs_type);
3038         if (err)
3039                 goto out;
3040         return 0;
3041 out:
3042         destroy_inodecache();
3043 out1:
3044         exit_ext3_xattr();
3045         return err;
3046 }
3047
3048 static void __exit exit_ext3_fs(void)
3049 {
3050         unregister_filesystem(&ext3_fs_type);
3051         destroy_inodecache();
3052         exit_ext3_xattr();
3053 }
3054
3055 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3056 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3057 MODULE_LICENSE("GPL");
3058 module_init(init_ext3_fs)
3059 module_exit(exit_ext3_fs)