[PATCH] fix race in __block_prepare_write
[linux-2.6] / 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/config.h>
20 #include <linux/module.h>
21 #include <linux/string.h>
22 #include <linux/fs.h>
23 #include <linux/time.h>
24 #include <linux/jbd.h>
25 #include <linux/ext3_fs.h>
26 #include <linux/ext3_jbd.h>
27 #include <linux/slab.h>
28 #include <linux/init.h>
29 #include <linux/blkdev.h>
30 #include <linux/parser.h>
31 #include <linux/smp_lock.h>
32 #include <linux/buffer_head.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 <asm/uaccess.h>
39 #include "xattr.h"
40 #include "acl.h"
41
42 static int ext3_load_journal(struct super_block *, struct ext3_super_block *);
43 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
44                                int);
45 static void ext3_commit_super (struct super_block * sb,
46                                struct ext3_super_block * es,
47                                int sync);
48 static void ext3_mark_recovery_complete(struct super_block * sb,
49                                         struct ext3_super_block * es);
50 static void ext3_clear_journal_err(struct super_block * sb,
51                                    struct ext3_super_block * es);
52 static int ext3_sync_fs(struct super_block *sb, int wait);
53 static const char *ext3_decode_error(struct super_block * sb, int errno,
54                                      char nbuf[16]);
55 static int ext3_remount (struct super_block * sb, int * flags, char * data);
56 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf);
57 static void ext3_unlockfs(struct super_block *sb);
58 static void ext3_write_super (struct super_block * sb);
59 static void ext3_write_super_lockfs(struct super_block *sb);
60
61 /* 
62  * Wrappers for journal_start/end.
63  *
64  * The only special thing we need to do here is to make sure that all
65  * journal_end calls result in the superblock being marked dirty, so
66  * that sync() will call the filesystem's write_super callback if
67  * appropriate. 
68  */
69 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
70 {
71         journal_t *journal;
72
73         if (sb->s_flags & MS_RDONLY)
74                 return ERR_PTR(-EROFS);
75
76         /* Special case here: if the journal has aborted behind our
77          * backs (eg. EIO in the commit thread), then we still need to
78          * take the FS itself readonly cleanly. */
79         journal = EXT3_SB(sb)->s_journal;
80         if (is_journal_aborted(journal)) {
81                 ext3_abort(sb, __FUNCTION__,
82                            "Detected aborted journal");
83                 return ERR_PTR(-EROFS);
84         }
85
86         return journal_start(journal, nblocks);
87 }
88
89 /* 
90  * The only special thing we need to do here is to make sure that all
91  * journal_stop calls result in the superblock being marked dirty, so
92  * that sync() will call the filesystem's write_super callback if
93  * appropriate. 
94  */
95 int __ext3_journal_stop(const char *where, handle_t *handle)
96 {
97         struct super_block *sb;
98         int err;
99         int rc;
100
101         sb = handle->h_transaction->t_journal->j_private;
102         err = handle->h_err;
103         rc = journal_stop(handle);
104
105         if (!err)
106                 err = rc;
107         if (err)
108                 __ext3_std_error(sb, where, err);
109         return err;
110 }
111
112 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
113                 struct buffer_head *bh, handle_t *handle, int err)
114 {
115         char nbuf[16];
116         const char *errstr = ext3_decode_error(NULL, err, nbuf);
117
118         if (bh)
119                 BUFFER_TRACE(bh, "abort");
120
121         if (!handle->h_err)
122                 handle->h_err = err;
123
124         if (is_handle_aborted(handle))
125                 return;
126
127         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
128                caller, errstr, err_fn);
129
130         journal_abort_handle(handle);
131 }
132
133 /* Deal with the reporting of failure conditions on a filesystem such as
134  * inconsistencies detected or read IO failures.
135  *
136  * On ext2, we can store the error state of the filesystem in the
137  * superblock.  That is not possible on ext3, because we may have other
138  * write ordering constraints on the superblock which prevent us from
139  * writing it out straight away; and given that the journal is about to
140  * be aborted, we can't rely on the current, or future, transactions to
141  * write out the superblock safely.
142  *
143  * We'll just use the journal_abort() error code to record an error in
144  * the journal instead.  On recovery, the journal will compain about
145  * that error until we've noted it down and cleared it.
146  */
147
148 static void ext3_handle_error(struct super_block *sb)
149 {
150         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
151
152         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
153         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
154
155         if (sb->s_flags & MS_RDONLY)
156                 return;
157
158         if (test_opt (sb, ERRORS_RO)) {
159                 printk (KERN_CRIT "Remounting filesystem read-only\n");
160                 sb->s_flags |= MS_RDONLY;
161         } else {
162                 journal_t *journal = EXT3_SB(sb)->s_journal;
163
164                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
165                 if (journal)
166                         journal_abort(journal, -EIO);
167         }
168         if (test_opt(sb, ERRORS_PANIC))
169                 panic("EXT3-fs (device %s): panic forced after error\n",
170                         sb->s_id);
171         ext3_commit_super(sb, es, 1);
172 }
173
174 void ext3_error (struct super_block * sb, const char * function,
175                  const char * fmt, ...)
176 {
177         va_list args;
178
179         va_start(args, fmt);
180         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
181         vprintk(fmt, args);
182         printk("\n");
183         va_end(args);
184
185         ext3_handle_error(sb);
186 }
187
188 static const char *ext3_decode_error(struct super_block * sb, int errno,
189                                      char nbuf[16])
190 {
191         char *errstr = NULL;
192
193         switch (errno) {
194         case -EIO:
195                 errstr = "IO failure";
196                 break;
197         case -ENOMEM:
198                 errstr = "Out of memory";
199                 break;
200         case -EROFS:
201                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
202                         errstr = "Journal has aborted";
203                 else
204                         errstr = "Readonly filesystem";
205                 break;
206         default:
207                 /* If the caller passed in an extra buffer for unknown
208                  * errors, textualise them now.  Else we just return
209                  * NULL. */
210                 if (nbuf) {
211                         /* Check for truncated error codes... */
212                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
213                                 errstr = nbuf;
214                 }
215                 break;
216         }
217
218         return errstr;
219 }
220
221 /* __ext3_std_error decodes expected errors from journaling functions
222  * automatically and invokes the appropriate error response.  */
223
224 void __ext3_std_error (struct super_block * sb, const char * function,
225                        int errno)
226 {
227         char nbuf[16];
228         const char *errstr = ext3_decode_error(sb, errno, nbuf);
229
230         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
231                 sb->s_id, function, errstr);
232
233         ext3_handle_error(sb);
234 }
235
236 /*
237  * ext3_abort is a much stronger failure handler than ext3_error.  The
238  * abort function may be used to deal with unrecoverable failures such
239  * as journal IO errors or ENOMEM at a critical moment in log management.
240  *
241  * We unconditionally force the filesystem into an ABORT|READONLY state,
242  * unless the error response on the fs has been set to panic in which
243  * case we take the easy way out and panic immediately.
244  */
245
246 void ext3_abort (struct super_block * sb, const char * function,
247                  const char * fmt, ...)
248 {
249         va_list args;
250
251         printk (KERN_CRIT "ext3_abort called.\n");
252
253         va_start(args, fmt);
254         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
255         vprintk(fmt, args);
256         printk("\n");
257         va_end(args);
258
259         if (test_opt(sb, ERRORS_PANIC))
260                 panic("EXT3-fs panic from previous error\n");
261
262         if (sb->s_flags & MS_RDONLY)
263                 return;
264
265         printk(KERN_CRIT "Remounting filesystem read-only\n");
266         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
267         sb->s_flags |= MS_RDONLY;
268         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
269         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
270 }
271
272 void ext3_warning (struct super_block * sb, const char * function,
273                    const char * fmt, ...)
274 {
275         va_list args;
276
277         va_start(args, fmt);
278         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
279                sb->s_id, function);
280         vprintk(fmt, args);
281         printk("\n");
282         va_end(args);
283 }
284
285 void ext3_update_dynamic_rev(struct super_block *sb)
286 {
287         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
288
289         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
290                 return;
291
292         ext3_warning(sb, __FUNCTION__,
293                      "updating to rev %d because of new feature flag, "
294                      "running e2fsck is recommended",
295                      EXT3_DYNAMIC_REV);
296
297         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
298         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
299         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
300         /* leave es->s_feature_*compat flags alone */
301         /* es->s_uuid will be set by e2fsck if empty */
302
303         /*
304          * The rest of the superblock fields should be zero, and if not it
305          * means they are likely already in use, so leave them alone.  We
306          * can leave it up to e2fsck to clean up any inconsistencies there.
307          */
308 }
309
310 /*
311  * Open the external journal device
312  */
313 static struct block_device *ext3_blkdev_get(dev_t dev)
314 {
315         struct block_device *bdev;
316         char b[BDEVNAME_SIZE];
317
318         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
319         if (IS_ERR(bdev))
320                 goto fail;
321         return bdev;
322
323 fail:
324         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
325                         __bdevname(dev, b), PTR_ERR(bdev));
326         return NULL;
327 }
328
329 /*
330  * Release the journal device
331  */
332 static int ext3_blkdev_put(struct block_device *bdev)
333 {
334         bd_release(bdev);
335         return blkdev_put(bdev);
336 }
337
338 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
339 {
340         struct block_device *bdev;
341         int ret = -ENODEV;
342
343         bdev = sbi->journal_bdev;
344         if (bdev) {
345                 ret = ext3_blkdev_put(bdev);
346                 sbi->journal_bdev = NULL;
347         }
348         return ret;
349 }
350
351 static inline struct inode *orphan_list_entry(struct list_head *l)
352 {
353         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
354 }
355
356 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
357 {
358         struct list_head *l;
359
360         printk(KERN_ERR "sb orphan head is %d\n", 
361                le32_to_cpu(sbi->s_es->s_last_orphan));
362
363         printk(KERN_ERR "sb_info orphan list:\n");
364         list_for_each(l, &sbi->s_orphan) {
365                 struct inode *inode = orphan_list_entry(l);
366                 printk(KERN_ERR "  "
367                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
368                        inode->i_sb->s_id, inode->i_ino, inode,
369                        inode->i_mode, inode->i_nlink, 
370                        NEXT_ORPHAN(inode));
371         }
372 }
373
374 static void ext3_put_super (struct super_block * sb)
375 {
376         struct ext3_sb_info *sbi = EXT3_SB(sb);
377         struct ext3_super_block *es = sbi->s_es;
378         int i;
379
380         ext3_xattr_put_super(sb);
381         journal_destroy(sbi->s_journal);
382         if (!(sb->s_flags & MS_RDONLY)) {
383                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
384                 es->s_state = cpu_to_le16(sbi->s_mount_state);
385                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
386                 mark_buffer_dirty(sbi->s_sbh);
387                 ext3_commit_super(sb, es, 1);
388         }
389
390         for (i = 0; i < sbi->s_gdb_count; i++)
391                 brelse(sbi->s_group_desc[i]);
392         kfree(sbi->s_group_desc);
393         percpu_counter_destroy(&sbi->s_freeblocks_counter);
394         percpu_counter_destroy(&sbi->s_freeinodes_counter);
395         percpu_counter_destroy(&sbi->s_dirs_counter);
396         brelse(sbi->s_sbh);
397 #ifdef CONFIG_QUOTA
398         for (i = 0; i < MAXQUOTAS; i++)
399                 kfree(sbi->s_qf_names[i]);
400 #endif
401
402         /* Debugging code just in case the in-memory inode orphan list
403          * isn't empty.  The on-disk one can be non-empty if we've
404          * detected an error and taken the fs readonly, but the
405          * in-memory list had better be clean by this point. */
406         if (!list_empty(&sbi->s_orphan))
407                 dump_orphan_list(sb, sbi);
408         J_ASSERT(list_empty(&sbi->s_orphan));
409
410         invalidate_bdev(sb->s_bdev, 0);
411         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
412                 /*
413                  * Invalidate the journal device's buffers.  We don't want them
414                  * floating about in memory - the physical journal device may
415                  * hotswapped, and it breaks the `ro-after' testing code.
416                  */
417                 sync_blockdev(sbi->journal_bdev);
418                 invalidate_bdev(sbi->journal_bdev, 0);
419                 ext3_blkdev_remove(sbi);
420         }
421         sb->s_fs_info = NULL;
422         kfree(sbi);
423         return;
424 }
425
426 static kmem_cache_t *ext3_inode_cachep;
427
428 /*
429  * Called inside transaction, so use GFP_NOFS
430  */
431 static struct inode *ext3_alloc_inode(struct super_block *sb)
432 {
433         struct ext3_inode_info *ei;
434
435         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
436         if (!ei)
437                 return NULL;
438 #ifdef CONFIG_EXT3_FS_POSIX_ACL
439         ei->i_acl = EXT3_ACL_NOT_CACHED;
440         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
441 #endif
442         ei->i_block_alloc_info = NULL;
443         ei->vfs_inode.i_version = 1;
444         return &ei->vfs_inode;
445 }
446
447 static void ext3_destroy_inode(struct inode *inode)
448 {
449         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
450 }
451
452 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
453 {
454         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
455
456         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
457             SLAB_CTOR_CONSTRUCTOR) {
458                 INIT_LIST_HEAD(&ei->i_orphan);
459 #ifdef CONFIG_EXT3_FS_XATTR
460                 init_rwsem(&ei->xattr_sem);
461 #endif
462                 init_MUTEX(&ei->truncate_sem);
463                 inode_init_once(&ei->vfs_inode);
464         }
465 }
466  
467 static int init_inodecache(void)
468 {
469         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
470                                              sizeof(struct ext3_inode_info),
471                                              0, SLAB_RECLAIM_ACCOUNT,
472                                              init_once, NULL);
473         if (ext3_inode_cachep == NULL)
474                 return -ENOMEM;
475         return 0;
476 }
477
478 static void destroy_inodecache(void)
479 {
480         if (kmem_cache_destroy(ext3_inode_cachep))
481                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
482 }
483
484 static void ext3_clear_inode(struct inode *inode)
485 {
486         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
487 #ifdef CONFIG_EXT3_FS_POSIX_ACL
488        if (EXT3_I(inode)->i_acl &&
489            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
490                posix_acl_release(EXT3_I(inode)->i_acl);
491                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
492        }
493        if (EXT3_I(inode)->i_default_acl &&
494            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
495                posix_acl_release(EXT3_I(inode)->i_default_acl);
496                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
497        }
498 #endif
499         ext3_discard_reservation(inode);
500         EXT3_I(inode)->i_block_alloc_info = NULL;
501         kfree(rsv);
502 }
503
504 #ifdef CONFIG_QUOTA
505
506 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
507 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
508
509 static int ext3_dquot_initialize(struct inode *inode, int type);
510 static int ext3_dquot_drop(struct inode *inode);
511 static int ext3_write_dquot(struct dquot *dquot);
512 static int ext3_acquire_dquot(struct dquot *dquot);
513 static int ext3_release_dquot(struct dquot *dquot);
514 static int ext3_mark_dquot_dirty(struct dquot *dquot);
515 static int ext3_write_info(struct super_block *sb, int type);
516 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
517 static int ext3_quota_on_mount(struct super_block *sb, int type);
518 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
519                                size_t len, loff_t off);
520 static ssize_t ext3_quota_write(struct super_block *sb, int type,
521                                 const char *data, size_t len, loff_t off);
522
523 static struct dquot_operations ext3_quota_operations = {
524         .initialize     = ext3_dquot_initialize,
525         .drop           = ext3_dquot_drop,
526         .alloc_space    = dquot_alloc_space,
527         .alloc_inode    = dquot_alloc_inode,
528         .free_space     = dquot_free_space,
529         .free_inode     = dquot_free_inode,
530         .transfer       = dquot_transfer,
531         .write_dquot    = ext3_write_dquot,
532         .acquire_dquot  = ext3_acquire_dquot,
533         .release_dquot  = ext3_release_dquot,
534         .mark_dirty     = ext3_mark_dquot_dirty,
535         .write_info     = ext3_write_info
536 };
537
538 static struct quotactl_ops ext3_qctl_operations = {
539         .quota_on       = ext3_quota_on,
540         .quota_off      = vfs_quota_off,
541         .quota_sync     = vfs_quota_sync,
542         .get_info       = vfs_get_dqinfo,
543         .set_info       = vfs_set_dqinfo,
544         .get_dqblk      = vfs_get_dqblk,
545         .set_dqblk      = vfs_set_dqblk
546 };
547 #endif
548
549 static struct super_operations ext3_sops = {
550         .alloc_inode    = ext3_alloc_inode,
551         .destroy_inode  = ext3_destroy_inode,
552         .read_inode     = ext3_read_inode,
553         .write_inode    = ext3_write_inode,
554         .dirty_inode    = ext3_dirty_inode,
555         .delete_inode   = ext3_delete_inode,
556         .put_super      = ext3_put_super,
557         .write_super    = ext3_write_super,
558         .sync_fs        = ext3_sync_fs,
559         .write_super_lockfs = ext3_write_super_lockfs,
560         .unlockfs       = ext3_unlockfs,
561         .statfs         = ext3_statfs,
562         .remount_fs     = ext3_remount,
563         .clear_inode    = ext3_clear_inode,
564 #ifdef CONFIG_QUOTA
565         .quota_read     = ext3_quota_read,
566         .quota_write    = ext3_quota_write,
567 #endif
568 };
569
570 struct dentry *ext3_get_parent(struct dentry *child);
571 static struct export_operations ext3_export_ops = {
572         .get_parent = ext3_get_parent,
573 };
574
575 enum {
576         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
577         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
578         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
579         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
580         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh,
581         Opt_commit, Opt_journal_update, Opt_journal_inum,
582         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
583         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
584         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0,
585         Opt_ignore, Opt_barrier, Opt_err, Opt_resize,
586 };
587
588 static match_table_t tokens = {
589         {Opt_bsd_df, "bsddf"},
590         {Opt_minix_df, "minixdf"},
591         {Opt_grpid, "grpid"},
592         {Opt_grpid, "bsdgroups"},
593         {Opt_nogrpid, "nogrpid"},
594         {Opt_nogrpid, "sysvgroups"},
595         {Opt_resgid, "resgid=%u"},
596         {Opt_resuid, "resuid=%u"},
597         {Opt_sb, "sb=%u"},
598         {Opt_err_cont, "errors=continue"},
599         {Opt_err_panic, "errors=panic"},
600         {Opt_err_ro, "errors=remount-ro"},
601         {Opt_nouid32, "nouid32"},
602         {Opt_nocheck, "nocheck"},
603         {Opt_nocheck, "check=none"},
604         {Opt_check, "check"},
605         {Opt_debug, "debug"},
606         {Opt_oldalloc, "oldalloc"},
607         {Opt_orlov, "orlov"},
608         {Opt_user_xattr, "user_xattr"},
609         {Opt_nouser_xattr, "nouser_xattr"},
610         {Opt_acl, "acl"},
611         {Opt_noacl, "noacl"},
612         {Opt_reservation, "reservation"},
613         {Opt_noreservation, "noreservation"},
614         {Opt_noload, "noload"},
615         {Opt_nobh, "nobh"},
616         {Opt_commit, "commit=%u"},
617         {Opt_journal_update, "journal=update"},
618         {Opt_journal_inum, "journal=%u"},
619         {Opt_abort, "abort"},
620         {Opt_data_journal, "data=journal"},
621         {Opt_data_ordered, "data=ordered"},
622         {Opt_data_writeback, "data=writeback"},
623         {Opt_offusrjquota, "usrjquota="},
624         {Opt_usrjquota, "usrjquota=%s"},
625         {Opt_offgrpjquota, "grpjquota="},
626         {Opt_grpjquota, "grpjquota=%s"},
627         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
628         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
629         {Opt_ignore, "grpquota"},
630         {Opt_ignore, "noquota"},
631         {Opt_ignore, "quota"},
632         {Opt_ignore, "usrquota"},
633         {Opt_barrier, "barrier=%u"},
634         {Opt_err, NULL},
635         {Opt_resize, "resize"},
636 };
637
638 static unsigned long get_sb_block(void **data)
639 {
640         unsigned long   sb_block;
641         char            *options = (char *) *data;
642
643         if (!options || strncmp(options, "sb=", 3) != 0)
644                 return 1;       /* Default location */
645         options += 3;
646         sb_block = simple_strtoul(options, &options, 0);
647         if (*options && *options != ',') {
648                 printk("EXT3-fs: Invalid sb specification: %s\n",
649                        (char *) *data);
650                 return 1;
651         }
652         if (*options == ',')
653                 options++;
654         *data = (void *) options;
655         return sb_block;
656 }
657
658 static int parse_options (char * options, struct super_block *sb,
659                           unsigned long * inum, unsigned long *n_blocks_count, int is_remount)
660 {
661         struct ext3_sb_info *sbi = EXT3_SB(sb);
662         char * p;
663         substring_t args[MAX_OPT_ARGS];
664         int data_opt = 0;
665         int option;
666 #ifdef CONFIG_QUOTA
667         int qtype;
668         char *qname;
669 #endif
670
671         if (!options)
672                 return 1;
673
674         while ((p = strsep (&options, ",")) != NULL) {
675                 int token;
676                 if (!*p)
677                         continue;
678
679                 token = match_token(p, tokens, args);
680                 switch (token) {
681                 case Opt_bsd_df:
682                         clear_opt (sbi->s_mount_opt, MINIX_DF);
683                         break;
684                 case Opt_minix_df:
685                         set_opt (sbi->s_mount_opt, MINIX_DF);
686                         break;
687                 case Opt_grpid:
688                         set_opt (sbi->s_mount_opt, GRPID);
689                         break;
690                 case Opt_nogrpid:
691                         clear_opt (sbi->s_mount_opt, GRPID);
692                         break;
693                 case Opt_resuid:
694                         if (match_int(&args[0], &option))
695                                 return 0;
696                         sbi->s_resuid = option;
697                         break;
698                 case Opt_resgid:
699                         if (match_int(&args[0], &option))
700                                 return 0;
701                         sbi->s_resgid = option;
702                         break;
703                 case Opt_sb:
704                         /* handled by get_sb_block() instead of here */
705                         /* *sb_block = match_int(&args[0]); */
706                         break;
707                 case Opt_err_panic:
708                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
709                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
710                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
711                         break;
712                 case Opt_err_ro:
713                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
714                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
715                         set_opt (sbi->s_mount_opt, ERRORS_RO);
716                         break;
717                 case Opt_err_cont:
718                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
719                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
720                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
721                         break;
722                 case Opt_nouid32:
723                         set_opt (sbi->s_mount_opt, NO_UID32);
724                         break;
725                 case Opt_check:
726 #ifdef CONFIG_EXT3_CHECK
727                         set_opt (sbi->s_mount_opt, CHECK);
728 #else
729                         printk(KERN_ERR
730                                "EXT3 Check option not supported\n");
731 #endif
732                         break;
733                 case Opt_nocheck:
734                         clear_opt (sbi->s_mount_opt, CHECK);
735                         break;
736                 case Opt_debug:
737                         set_opt (sbi->s_mount_opt, DEBUG);
738                         break;
739                 case Opt_oldalloc:
740                         set_opt (sbi->s_mount_opt, OLDALLOC);
741                         break;
742                 case Opt_orlov:
743                         clear_opt (sbi->s_mount_opt, OLDALLOC);
744                         break;
745 #ifdef CONFIG_EXT3_FS_XATTR
746                 case Opt_user_xattr:
747                         set_opt (sbi->s_mount_opt, XATTR_USER);
748                         break;
749                 case Opt_nouser_xattr:
750                         clear_opt (sbi->s_mount_opt, XATTR_USER);
751                         break;
752 #else
753                 case Opt_user_xattr:
754                 case Opt_nouser_xattr:
755                         printk("EXT3 (no)user_xattr options not supported\n");
756                         break;
757 #endif
758 #ifdef CONFIG_EXT3_FS_POSIX_ACL
759                 case Opt_acl:
760                         set_opt(sbi->s_mount_opt, POSIX_ACL);
761                         break;
762                 case Opt_noacl:
763                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
764                         break;
765 #else
766                 case Opt_acl:
767                 case Opt_noacl:
768                         printk("EXT3 (no)acl options not supported\n");
769                         break;
770 #endif
771                 case Opt_reservation:
772                         set_opt(sbi->s_mount_opt, RESERVATION);
773                         break;
774                 case Opt_noreservation:
775                         clear_opt(sbi->s_mount_opt, RESERVATION);
776                         break;
777                 case Opt_journal_update:
778                         /* @@@ FIXME */
779                         /* Eventually we will want to be able to create
780                            a journal file here.  For now, only allow the
781                            user to specify an existing inode to be the
782                            journal file. */
783                         if (is_remount) {
784                                 printk(KERN_ERR "EXT3-fs: cannot specify "
785                                        "journal on remount\n");
786                                 return 0;
787                         }
788                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
789                         break;
790                 case Opt_journal_inum:
791                         if (is_remount) {
792                                 printk(KERN_ERR "EXT3-fs: cannot specify "
793                                        "journal on remount\n");
794                                 return 0;
795                         }
796                         if (match_int(&args[0], &option))
797                                 return 0;
798                         *inum = option;
799                         break;
800                 case Opt_noload:
801                         set_opt (sbi->s_mount_opt, NOLOAD);
802                         break;
803                 case Opt_commit:
804                         if (match_int(&args[0], &option))
805                                 return 0;
806                         if (option < 0)
807                                 return 0;
808                         if (option == 0)
809                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
810                         sbi->s_commit_interval = HZ * option;
811                         break;
812                 case Opt_data_journal:
813                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
814                         goto datacheck;
815                 case Opt_data_ordered:
816                         data_opt = EXT3_MOUNT_ORDERED_DATA;
817                         goto datacheck;
818                 case Opt_data_writeback:
819                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
820                 datacheck:
821                         if (is_remount) {
822                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
823                                                 != data_opt) {
824                                         printk(KERN_ERR
825                                                 "EXT3-fs: cannot change data "
826                                                 "mode on remount\n");
827                                         return 0;
828                                 }
829                         } else {
830                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
831                                 sbi->s_mount_opt |= data_opt;
832                         }
833                         break;
834 #ifdef CONFIG_QUOTA
835                 case Opt_usrjquota:
836                         qtype = USRQUOTA;
837                         goto set_qf_name;
838                 case Opt_grpjquota:
839                         qtype = GRPQUOTA;
840 set_qf_name:
841                         if (sb_any_quota_enabled(sb)) {
842                                 printk(KERN_ERR
843                                         "EXT3-fs: Cannot change journalled "
844                                         "quota options when quota turned on.\n");
845                                 return 0;
846                         }
847                         qname = match_strdup(&args[0]);
848                         if (!qname) {
849                                 printk(KERN_ERR
850                                         "EXT3-fs: not enough memory for "
851                                         "storing quotafile name.\n");
852                                 return 0;
853                         }
854                         if (sbi->s_qf_names[qtype] &&
855                             strcmp(sbi->s_qf_names[qtype], qname)) {
856                                 printk(KERN_ERR
857                                         "EXT3-fs: %s quota file already "
858                                         "specified.\n", QTYPE2NAME(qtype));
859                                 kfree(qname);
860                                 return 0;
861                         }
862                         sbi->s_qf_names[qtype] = qname;
863                         if (strchr(sbi->s_qf_names[qtype], '/')) {
864                                 printk(KERN_ERR
865                                         "EXT3-fs: quotafile must be on "
866                                         "filesystem root.\n");
867                                 kfree(sbi->s_qf_names[qtype]);
868                                 sbi->s_qf_names[qtype] = NULL;
869                                 return 0;
870                         }
871                         break;
872                 case Opt_offusrjquota:
873                         qtype = USRQUOTA;
874                         goto clear_qf_name;
875                 case Opt_offgrpjquota:
876                         qtype = GRPQUOTA;
877 clear_qf_name:
878                         if (sb_any_quota_enabled(sb)) {
879                                 printk(KERN_ERR "EXT3-fs: Cannot change "
880                                         "journalled quota options when "
881                                         "quota turned on.\n");
882                                 return 0;
883                         }
884                         kfree(sbi->s_qf_names[qtype]);
885                         sbi->s_qf_names[qtype] = NULL;
886                         break;
887                 case Opt_jqfmt_vfsold:
888                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
889                         break;
890                 case Opt_jqfmt_vfsv0:
891                         sbi->s_jquota_fmt = QFMT_VFS_V0;
892                         break;
893 #else
894                 case Opt_usrjquota:
895                 case Opt_grpjquota:
896                 case Opt_offusrjquota:
897                 case Opt_offgrpjquota:
898                 case Opt_jqfmt_vfsold:
899                 case Opt_jqfmt_vfsv0:
900                         printk(KERN_ERR
901                                 "EXT3-fs: journalled quota options not "
902                                 "supported.\n");
903                         break;
904 #endif
905                 case Opt_abort:
906                         set_opt(sbi->s_mount_opt, ABORT);
907                         break;
908                 case Opt_barrier:
909                         if (match_int(&args[0], &option))
910                                 return 0;
911                         if (option)
912                                 set_opt(sbi->s_mount_opt, BARRIER);
913                         else
914                                 clear_opt(sbi->s_mount_opt, BARRIER);
915                         break;
916                 case Opt_ignore:
917                         break;
918                 case Opt_resize:
919                         if (!n_blocks_count) {
920                                 printk("EXT3-fs: resize option only available "
921                                         "for remount\n");
922                                 return 0;
923                         }
924                         match_int(&args[0], &option);
925                         *n_blocks_count = option;
926                         break;
927                 case Opt_nobh:
928                         set_opt(sbi->s_mount_opt, NOBH);
929                         break;
930                 default:
931                         printk (KERN_ERR
932                                 "EXT3-fs: Unrecognized mount option \"%s\" "
933                                 "or missing value\n", p);
934                         return 0;
935                 }
936         }
937 #ifdef CONFIG_QUOTA
938         if (!sbi->s_jquota_fmt && (sbi->s_qf_names[USRQUOTA] ||
939             sbi->s_qf_names[GRPQUOTA])) {
940                 printk(KERN_ERR
941                         "EXT3-fs: journalled quota format not specified.\n");
942                 return 0;
943         }
944 #endif
945
946         return 1;
947 }
948
949 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
950                             int read_only)
951 {
952         struct ext3_sb_info *sbi = EXT3_SB(sb);
953         int res = 0;
954
955         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
956                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
957                         "forcing read-only mode\n");
958                 res = MS_RDONLY;
959         }
960         if (read_only)
961                 return res;
962         if (!(sbi->s_mount_state & EXT3_VALID_FS))
963                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
964                         "running e2fsck is recommended\n");
965         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
966                 printk (KERN_WARNING
967                         "EXT3-fs warning: mounting fs with errors, "
968                         "running e2fsck is recommended\n");
969         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
970                  le16_to_cpu(es->s_mnt_count) >=
971                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
972                 printk (KERN_WARNING
973                         "EXT3-fs warning: maximal mount count reached, "
974                         "running e2fsck is recommended\n");
975         else if (le32_to_cpu(es->s_checkinterval) &&
976                 (le32_to_cpu(es->s_lastcheck) +
977                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
978                 printk (KERN_WARNING
979                         "EXT3-fs warning: checktime reached, "
980                         "running e2fsck is recommended\n");
981 #if 0
982                 /* @@@ We _will_ want to clear the valid bit if we find
983                    inconsistencies, to force a fsck at reboot.  But for
984                    a plain journaled filesystem we can keep it set as
985                    valid forever! :) */
986         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
987 #endif
988         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
989                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
990         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
991         es->s_mtime = cpu_to_le32(get_seconds());
992         ext3_update_dynamic_rev(sb);
993         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
994
995         ext3_commit_super(sb, es, 1);
996         if (test_opt(sb, DEBUG))
997                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
998                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
999                         sb->s_blocksize,
1000                         sbi->s_groups_count,
1001                         EXT3_BLOCKS_PER_GROUP(sb),
1002                         EXT3_INODES_PER_GROUP(sb),
1003                         sbi->s_mount_opt);
1004
1005         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1006         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1007                 char b[BDEVNAME_SIZE];
1008
1009                 printk("external journal on %s\n",
1010                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1011         } else {
1012                 printk("internal journal\n");
1013         }
1014 #ifdef CONFIG_EXT3_CHECK
1015         if (test_opt (sb, CHECK)) {
1016                 ext3_check_blocks_bitmap (sb);
1017                 ext3_check_inodes_bitmap (sb);
1018         }
1019 #endif
1020         return res;
1021 }
1022
1023 /* Called at mount-time, super-block is locked */
1024 static int ext3_check_descriptors (struct super_block * sb)
1025 {
1026         struct ext3_sb_info *sbi = EXT3_SB(sb);
1027         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1028         struct ext3_group_desc * gdp = NULL;
1029         int desc_block = 0;
1030         int i;
1031
1032         ext3_debug ("Checking group descriptors");
1033
1034         for (i = 0; i < sbi->s_groups_count; i++)
1035         {
1036                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1037                         gdp = (struct ext3_group_desc *)
1038                                         sbi->s_group_desc[desc_block++]->b_data;
1039                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1040                     le32_to_cpu(gdp->bg_block_bitmap) >=
1041                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1042                 {
1043                         ext3_error (sb, "ext3_check_descriptors",
1044                                     "Block bitmap for group %d"
1045                                     " not in group (block %lu)!",
1046                                     i, (unsigned long)
1047                                         le32_to_cpu(gdp->bg_block_bitmap));
1048                         return 0;
1049                 }
1050                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1051                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1052                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1053                 {
1054                         ext3_error (sb, "ext3_check_descriptors",
1055                                     "Inode bitmap for group %d"
1056                                     " not in group (block %lu)!",
1057                                     i, (unsigned long)
1058                                         le32_to_cpu(gdp->bg_inode_bitmap));
1059                         return 0;
1060                 }
1061                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1062                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1063                     block + EXT3_BLOCKS_PER_GROUP(sb))
1064                 {
1065                         ext3_error (sb, "ext3_check_descriptors",
1066                                     "Inode table for group %d"
1067                                     " not in group (block %lu)!",
1068                                     i, (unsigned long)
1069                                         le32_to_cpu(gdp->bg_inode_table));
1070                         return 0;
1071                 }
1072                 block += EXT3_BLOCKS_PER_GROUP(sb);
1073                 gdp++;
1074         }
1075
1076         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1077         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1078         return 1;
1079 }
1080
1081
1082 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1083  * the superblock) which were deleted from all directories, but held open by
1084  * a process at the time of a crash.  We walk the list and try to delete these
1085  * inodes at recovery time (only with a read-write filesystem).
1086  *
1087  * In order to keep the orphan inode chain consistent during traversal (in
1088  * case of crash during recovery), we link each inode into the superblock
1089  * orphan list_head and handle it the same way as an inode deletion during
1090  * normal operation (which journals the operations for us).
1091  *
1092  * We only do an iget() and an iput() on each inode, which is very safe if we
1093  * accidentally point at an in-use or already deleted inode.  The worst that
1094  * can happen in this case is that we get a "bit already cleared" message from
1095  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1096  * e2fsck was run on this filesystem, and it must have already done the orphan
1097  * inode cleanup for us, so we can safely abort without any further action.
1098  */
1099 static void ext3_orphan_cleanup (struct super_block * sb,
1100                                  struct ext3_super_block * es)
1101 {
1102         unsigned int s_flags = sb->s_flags;
1103         int nr_orphans = 0, nr_truncates = 0;
1104 #ifdef CONFIG_QUOTA
1105         int i;
1106 #endif
1107         if (!es->s_last_orphan) {
1108                 jbd_debug(4, "no orphan inodes to clean up\n");
1109                 return;
1110         }
1111
1112         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1113                 if (es->s_last_orphan)
1114                         jbd_debug(1, "Errors on filesystem, "
1115                                   "clearing orphan list.\n");
1116                 es->s_last_orphan = 0;
1117                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1118                 return;
1119         }
1120
1121         if (s_flags & MS_RDONLY) {
1122                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1123                        sb->s_id);
1124                 sb->s_flags &= ~MS_RDONLY;
1125         }
1126 #ifdef CONFIG_QUOTA
1127         /* Needed for iput() to work correctly and not trash data */
1128         sb->s_flags |= MS_ACTIVE;
1129         /* Turn on quotas so that they are updated correctly */
1130         for (i = 0; i < MAXQUOTAS; i++) {
1131                 if (EXT3_SB(sb)->s_qf_names[i]) {
1132                         int ret = ext3_quota_on_mount(sb, i);
1133                         if (ret < 0)
1134                                 printk(KERN_ERR
1135                                         "EXT3-fs: Cannot turn on journalled "
1136                                         "quota: error %d\n", ret);
1137                 }
1138         }
1139 #endif
1140
1141         while (es->s_last_orphan) {
1142                 struct inode *inode;
1143
1144                 if (!(inode =
1145                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1146                         es->s_last_orphan = 0;
1147                         break;
1148                 }
1149
1150                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1151                 DQUOT_INIT(inode);
1152                 if (inode->i_nlink) {
1153                         printk(KERN_DEBUG
1154                                 "%s: truncating inode %ld to %Ld bytes\n",
1155                                 __FUNCTION__, inode->i_ino, inode->i_size);
1156                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1157                                   inode->i_ino, inode->i_size);
1158                         ext3_truncate(inode);
1159                         nr_truncates++;
1160                 } else {
1161                         printk(KERN_DEBUG
1162                                 "%s: deleting unreferenced inode %ld\n",
1163                                 __FUNCTION__, inode->i_ino);
1164                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1165                                   inode->i_ino);
1166                         nr_orphans++;
1167                 }
1168                 iput(inode);  /* The delete magic happens here! */
1169         }
1170
1171 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1172
1173         if (nr_orphans)
1174                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1175                        sb->s_id, PLURAL(nr_orphans));
1176         if (nr_truncates)
1177                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1178                        sb->s_id, PLURAL(nr_truncates));
1179 #ifdef CONFIG_QUOTA
1180         /* Turn quotas off */
1181         for (i = 0; i < MAXQUOTAS; i++) {
1182                 if (sb_dqopt(sb)->files[i])
1183                         vfs_quota_off(sb, i);
1184         }
1185 #endif
1186         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1187 }
1188
1189 #define log2(n) ffz(~(n))
1190
1191 /*
1192  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1193  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1194  * We need to be 1 filesystem block less than the 2^32 sector limit.
1195  */
1196 static loff_t ext3_max_size(int bits)
1197 {
1198         loff_t res = EXT3_NDIR_BLOCKS;
1199         /* This constant is calculated to be the largest file size for a
1200          * dense, 4k-blocksize file such that the total number of
1201          * sectors in the file, including data and all indirect blocks,
1202          * does not exceed 2^32. */
1203         const loff_t upper_limit = 0x1ff7fffd000LL;
1204
1205         res += 1LL << (bits-2);
1206         res += 1LL << (2*(bits-2));
1207         res += 1LL << (3*(bits-2));
1208         res <<= bits;
1209         if (res > upper_limit)
1210                 res = upper_limit;
1211         return res;
1212 }
1213
1214 static unsigned long descriptor_loc(struct super_block *sb,
1215                                     unsigned long logic_sb_block,
1216                                     int nr)
1217 {
1218         struct ext3_sb_info *sbi = EXT3_SB(sb);
1219         unsigned long bg, first_data_block, first_meta_bg;
1220         int has_super = 0;
1221
1222         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1223         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1224
1225         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1226             nr < first_meta_bg)
1227                 return (logic_sb_block + nr + 1);
1228         bg = sbi->s_desc_per_block * nr;
1229         if (ext3_bg_has_super(sb, bg))
1230                 has_super = 1;
1231         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1232 }
1233
1234
1235 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1236 {
1237         struct buffer_head * bh;
1238         struct ext3_super_block *es = NULL;
1239         struct ext3_sb_info *sbi;
1240         unsigned long block;
1241         unsigned long sb_block = get_sb_block(&data);
1242         unsigned long logic_sb_block;
1243         unsigned long offset = 0;
1244         unsigned long journal_inum = 0;
1245         unsigned long def_mount_opts;
1246         struct inode *root;
1247         int blocksize;
1248         int hblock;
1249         int db_count;
1250         int i;
1251         int needs_recovery;
1252         __le32 features;
1253
1254         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1255         if (!sbi)
1256                 return -ENOMEM;
1257         sb->s_fs_info = sbi;
1258         memset(sbi, 0, sizeof(*sbi));
1259         sbi->s_mount_opt = 0;
1260         sbi->s_resuid = EXT3_DEF_RESUID;
1261         sbi->s_resgid = EXT3_DEF_RESGID;
1262
1263         unlock_kernel();
1264
1265         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1266         if (!blocksize) {
1267                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1268                 goto out_fail;
1269         }
1270
1271         /*
1272          * The ext3 superblock will not be buffer aligned for other than 1kB
1273          * block sizes.  We need to calculate the offset from buffer start.
1274          */
1275         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1276                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1277                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1278         } else {
1279                 logic_sb_block = sb_block;
1280         }
1281
1282         if (!(bh = sb_bread(sb, logic_sb_block))) {
1283                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1284                 goto out_fail;
1285         }
1286         /*
1287          * Note: s_es must be initialized as soon as possible because
1288          *       some ext3 macro-instructions depend on its value
1289          */
1290         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1291         sbi->s_es = es;
1292         sb->s_magic = le16_to_cpu(es->s_magic);
1293         if (sb->s_magic != EXT3_SUPER_MAGIC)
1294                 goto cantfind_ext3;
1295
1296         /* Set defaults before we parse the mount options */
1297         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1298         if (def_mount_opts & EXT3_DEFM_DEBUG)
1299                 set_opt(sbi->s_mount_opt, DEBUG);
1300         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1301                 set_opt(sbi->s_mount_opt, GRPID);
1302         if (def_mount_opts & EXT3_DEFM_UID16)
1303                 set_opt(sbi->s_mount_opt, NO_UID32);
1304         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1305                 set_opt(sbi->s_mount_opt, XATTR_USER);
1306         if (def_mount_opts & EXT3_DEFM_ACL)
1307                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1308         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1309                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1310         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1311                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1312         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1313                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1314
1315         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1316                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1317         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1318                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1319
1320         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1321         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1322
1323         set_opt(sbi->s_mount_opt, RESERVATION);
1324
1325         if (!parse_options ((char *) data, sb, &journal_inum, NULL, 0))
1326                 goto failed_mount;
1327
1328         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1329                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1330
1331         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1332             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1333              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1334              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1335                 printk(KERN_WARNING 
1336                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1337                        "running e2fsck is recommended\n");
1338         /*
1339          * Check feature flags regardless of the revision level, since we
1340          * previously didn't change the revision level when setting the flags,
1341          * so there is a chance incompat flags are set on a rev 0 filesystem.
1342          */
1343         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1344         if (features) {
1345                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1346                        "unsupported optional features (%x).\n",
1347                        sb->s_id, le32_to_cpu(features));
1348                 goto failed_mount;
1349         }
1350         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1351         if (!(sb->s_flags & MS_RDONLY) && features) {
1352                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1353                        "unsupported optional features (%x).\n",
1354                        sb->s_id, le32_to_cpu(features));
1355                 goto failed_mount;
1356         }
1357         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1358
1359         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1360             blocksize > EXT3_MAX_BLOCK_SIZE) {
1361                 printk(KERN_ERR 
1362                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1363                        blocksize, sb->s_id);
1364                 goto failed_mount;
1365         }
1366
1367         hblock = bdev_hardsect_size(sb->s_bdev);
1368         if (sb->s_blocksize != blocksize) {
1369                 /*
1370                  * Make sure the blocksize for the filesystem is larger
1371                  * than the hardware sectorsize for the machine.
1372                  */
1373                 if (blocksize < hblock) {
1374                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1375                                "device blocksize %d.\n", blocksize, hblock);
1376                         goto failed_mount;
1377                 }
1378
1379                 brelse (bh);
1380                 sb_set_blocksize(sb, blocksize);
1381                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1382                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1383                 bh = sb_bread(sb, logic_sb_block);
1384                 if (!bh) {
1385                         printk(KERN_ERR 
1386                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1387                         goto failed_mount;
1388                 }
1389                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1390                 sbi->s_es = es;
1391                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1392                         printk (KERN_ERR 
1393                                 "EXT3-fs: Magic mismatch, very weird !\n");
1394                         goto failed_mount;
1395                 }
1396         }
1397
1398         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1399
1400         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1401                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1402                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1403         } else {
1404                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1405                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1406                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1407                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1408                     (sbi->s_inode_size > blocksize)) {
1409                         printk (KERN_ERR
1410                                 "EXT3-fs: unsupported inode size: %d\n",
1411                                 sbi->s_inode_size);
1412                         goto failed_mount;
1413                 }
1414         }
1415         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1416                                    le32_to_cpu(es->s_log_frag_size);
1417         if (blocksize != sbi->s_frag_size) {
1418                 printk(KERN_ERR
1419                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1420                        sbi->s_frag_size, blocksize);
1421                 goto failed_mount;
1422         }
1423         sbi->s_frags_per_block = 1;
1424         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1425         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1426         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1427         if (EXT3_INODE_SIZE(sb) == 0)
1428                 goto cantfind_ext3;
1429         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1430         if (sbi->s_inodes_per_block == 0)
1431                 goto cantfind_ext3;
1432         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1433                                         sbi->s_inodes_per_block;
1434         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1435         sbi->s_sbh = bh;
1436         sbi->s_mount_state = le16_to_cpu(es->s_state);
1437         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1438         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1439         for (i=0; i < 4; i++)
1440                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1441         sbi->s_def_hash_version = es->s_def_hash_version;
1442
1443         if (sbi->s_blocks_per_group > blocksize * 8) {
1444                 printk (KERN_ERR
1445                         "EXT3-fs: #blocks per group too big: %lu\n",
1446                         sbi->s_blocks_per_group);
1447                 goto failed_mount;
1448         }
1449         if (sbi->s_frags_per_group > blocksize * 8) {
1450                 printk (KERN_ERR
1451                         "EXT3-fs: #fragments per group too big: %lu\n",
1452                         sbi->s_frags_per_group);
1453                 goto failed_mount;
1454         }
1455         if (sbi->s_inodes_per_group > blocksize * 8) {
1456                 printk (KERN_ERR
1457                         "EXT3-fs: #inodes per group too big: %lu\n",
1458                         sbi->s_inodes_per_group);
1459                 goto failed_mount;
1460         }
1461
1462         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1463                 goto cantfind_ext3;
1464         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1465                                le32_to_cpu(es->s_first_data_block) +
1466                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1467                               EXT3_BLOCKS_PER_GROUP(sb);
1468         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1469                    EXT3_DESC_PER_BLOCK(sb);
1470         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1471                                     GFP_KERNEL);
1472         if (sbi->s_group_desc == NULL) {
1473                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1474                 goto failed_mount;
1475         }
1476
1477         percpu_counter_init(&sbi->s_freeblocks_counter);
1478         percpu_counter_init(&sbi->s_freeinodes_counter);
1479         percpu_counter_init(&sbi->s_dirs_counter);
1480         bgl_lock_init(&sbi->s_blockgroup_lock);
1481
1482         for (i = 0; i < db_count; i++) {
1483                 block = descriptor_loc(sb, logic_sb_block, i);
1484                 sbi->s_group_desc[i] = sb_bread(sb, block);
1485                 if (!sbi->s_group_desc[i]) {
1486                         printk (KERN_ERR "EXT3-fs: "
1487                                 "can't read group descriptor %d\n", i);
1488                         db_count = i;
1489                         goto failed_mount2;
1490                 }
1491         }
1492         if (!ext3_check_descriptors (sb)) {
1493                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1494                 goto failed_mount2;
1495         }
1496         sbi->s_gdb_count = db_count;
1497         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1498         spin_lock_init(&sbi->s_next_gen_lock);
1499         /* per fileystem reservation list head & lock */
1500         spin_lock_init(&sbi->s_rsv_window_lock);
1501         sbi->s_rsv_window_root = RB_ROOT;
1502         /* Add a single, static dummy reservation to the start of the
1503          * reservation window list --- it gives us a placeholder for
1504          * append-at-start-of-list which makes the allocation logic
1505          * _much_ simpler. */
1506         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1507         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1508         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1509         sbi->s_rsv_window_head.rsv_goal_size = 0;
1510         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1511
1512         /*
1513          * set up enough so that it can read an inode
1514          */
1515         sb->s_op = &ext3_sops;
1516         sb->s_export_op = &ext3_export_ops;
1517         sb->s_xattr = ext3_xattr_handlers;
1518 #ifdef CONFIG_QUOTA
1519         sb->s_qcop = &ext3_qctl_operations;
1520         sb->dq_op = &ext3_quota_operations;
1521 #endif
1522         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1523
1524         sb->s_root = NULL;
1525
1526         needs_recovery = (es->s_last_orphan != 0 ||
1527                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1528                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1529
1530         /*
1531          * The first inode we look at is the journal inode.  Don't try
1532          * root first: it may be modified in the journal!
1533          */
1534         if (!test_opt(sb, NOLOAD) &&
1535             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1536                 if (ext3_load_journal(sb, es))
1537                         goto failed_mount2;
1538         } else if (journal_inum) {
1539                 if (ext3_create_journal(sb, es, journal_inum))
1540                         goto failed_mount2;
1541         } else {
1542                 if (!silent)
1543                         printk (KERN_ERR
1544                                 "ext3: No journal on filesystem on %s\n",
1545                                 sb->s_id);
1546                 goto failed_mount2;
1547         }
1548
1549         /* We have now updated the journal if required, so we can
1550          * validate the data journaling mode. */
1551         switch (test_opt(sb, DATA_FLAGS)) {
1552         case 0:
1553                 /* No mode set, assume a default based on the journal
1554                    capabilities: ORDERED_DATA if the journal can
1555                    cope, else JOURNAL_DATA */
1556                 if (journal_check_available_features
1557                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1558                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1559                 else
1560                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1561                 break;
1562
1563         case EXT3_MOUNT_ORDERED_DATA:
1564         case EXT3_MOUNT_WRITEBACK_DATA:
1565                 if (!journal_check_available_features
1566                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1567                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1568                                "requested data journaling mode\n");
1569                         goto failed_mount3;
1570                 }
1571         default:
1572                 break;
1573         }
1574
1575         if (test_opt(sb, NOBH)) {
1576                 if (sb->s_blocksize_bits != PAGE_CACHE_SHIFT) {
1577                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option "
1578                                 "since filesystem blocksize doesn't match "
1579                                 "pagesize\n");
1580                         clear_opt(sbi->s_mount_opt, NOBH);
1581                 }
1582                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1583                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1584                                 "its supported only with writeback mode\n");
1585                         clear_opt(sbi->s_mount_opt, NOBH);
1586                 }
1587         }
1588         /*
1589          * The journal_load will have done any necessary log recovery,
1590          * so we can safely mount the rest of the filesystem now.
1591          */
1592
1593         root = iget(sb, EXT3_ROOT_INO);
1594         sb->s_root = d_alloc_root(root);
1595         if (!sb->s_root) {
1596                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1597                 iput(root);
1598                 goto failed_mount3;
1599         }
1600         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1601                 dput(sb->s_root);
1602                 sb->s_root = NULL;
1603                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1604                 goto failed_mount3;
1605         }
1606
1607         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1608         /*
1609          * akpm: core read_super() calls in here with the superblock locked.
1610          * That deadlocks, because orphan cleanup needs to lock the superblock
1611          * in numerous places.  Here we just pop the lock - it's relatively
1612          * harmless, because we are now ready to accept write_super() requests,
1613          * and aviro says that's the only reason for hanging onto the
1614          * superblock lock.
1615          */
1616         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1617         ext3_orphan_cleanup(sb, es);
1618         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1619         if (needs_recovery)
1620                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1621         ext3_mark_recovery_complete(sb, es);
1622         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1623                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1624                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1625                 "writeback");
1626
1627         percpu_counter_mod(&sbi->s_freeblocks_counter,
1628                 ext3_count_free_blocks(sb));
1629         percpu_counter_mod(&sbi->s_freeinodes_counter,
1630                 ext3_count_free_inodes(sb));
1631         percpu_counter_mod(&sbi->s_dirs_counter,
1632                 ext3_count_dirs(sb));
1633
1634         lock_kernel();
1635         return 0;
1636
1637 cantfind_ext3:
1638         if (!silent)
1639                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1640                        sb->s_id);
1641         goto failed_mount;
1642
1643 failed_mount3:
1644         journal_destroy(sbi->s_journal);
1645 failed_mount2:
1646         for (i = 0; i < db_count; i++)
1647                 brelse(sbi->s_group_desc[i]);
1648         kfree(sbi->s_group_desc);
1649 failed_mount:
1650 #ifdef CONFIG_QUOTA
1651         for (i = 0; i < MAXQUOTAS; i++)
1652                 kfree(sbi->s_qf_names[i]);
1653 #endif
1654         ext3_blkdev_remove(sbi);
1655         brelse(bh);
1656 out_fail:
1657         sb->s_fs_info = NULL;
1658         kfree(sbi);
1659         lock_kernel();
1660         return -EINVAL;
1661 }
1662
1663 /*
1664  * Setup any per-fs journal parameters now.  We'll do this both on
1665  * initial mount, once the journal has been initialised but before we've
1666  * done any recovery; and again on any subsequent remount. 
1667  */
1668 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1669 {
1670         struct ext3_sb_info *sbi = EXT3_SB(sb);
1671
1672         if (sbi->s_commit_interval)
1673                 journal->j_commit_interval = sbi->s_commit_interval;
1674         /* We could also set up an ext3-specific default for the commit
1675          * interval here, but for now we'll just fall back to the jbd
1676          * default. */
1677
1678         spin_lock(&journal->j_state_lock);
1679         if (test_opt(sb, BARRIER))
1680                 journal->j_flags |= JFS_BARRIER;
1681         else
1682                 journal->j_flags &= ~JFS_BARRIER;
1683         spin_unlock(&journal->j_state_lock);
1684 }
1685
1686 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1687 {
1688         struct inode *journal_inode;
1689         journal_t *journal;
1690
1691         /* First, test for the existence of a valid inode on disk.  Bad
1692          * things happen if we iget() an unused inode, as the subsequent
1693          * iput() will try to delete it. */
1694
1695         journal_inode = iget(sb, journal_inum);
1696         if (!journal_inode) {
1697                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1698                 return NULL;
1699         }
1700         if (!journal_inode->i_nlink) {
1701                 make_bad_inode(journal_inode);
1702                 iput(journal_inode);
1703                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1704                 return NULL;
1705         }
1706
1707         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1708                   journal_inode, journal_inode->i_size);
1709         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1710                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1711                 iput(journal_inode);
1712                 return NULL;
1713         }
1714
1715         journal = journal_init_inode(journal_inode);
1716         if (!journal) {
1717                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1718                 iput(journal_inode);
1719                 return NULL;
1720         }
1721         journal->j_private = sb;
1722         ext3_init_journal_params(sb, journal);
1723         return journal;
1724 }
1725
1726 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1727                                        dev_t j_dev)
1728 {
1729         struct buffer_head * bh;
1730         journal_t *journal;
1731         int start;
1732         int len;
1733         int hblock, blocksize;
1734         unsigned long sb_block;
1735         unsigned long offset;
1736         struct ext3_super_block * es;
1737         struct block_device *bdev;
1738
1739         bdev = ext3_blkdev_get(j_dev);
1740         if (bdev == NULL)
1741                 return NULL;
1742
1743         if (bd_claim(bdev, sb)) {
1744                 printk(KERN_ERR
1745                         "EXT3: failed to claim external journal device.\n");
1746                 blkdev_put(bdev);
1747                 return NULL;
1748         }
1749
1750         blocksize = sb->s_blocksize;
1751         hblock = bdev_hardsect_size(bdev);
1752         if (blocksize < hblock) {
1753                 printk(KERN_ERR
1754                         "EXT3-fs: blocksize too small for journal device.\n");
1755                 goto out_bdev;
1756         }
1757
1758         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1759         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1760         set_blocksize(bdev, blocksize);
1761         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1762                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1763                        "external journal\n");
1764                 goto out_bdev;
1765         }
1766
1767         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1768         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1769             !(le32_to_cpu(es->s_feature_incompat) &
1770               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1771                 printk(KERN_ERR "EXT3-fs: external journal has "
1772                                         "bad superblock\n");
1773                 brelse(bh);
1774                 goto out_bdev;
1775         }
1776
1777         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1778                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1779                 brelse(bh);
1780                 goto out_bdev;
1781         }
1782
1783         len = le32_to_cpu(es->s_blocks_count);
1784         start = sb_block + 1;
1785         brelse(bh);     /* we're done with the superblock */
1786
1787         journal = journal_init_dev(bdev, sb->s_bdev,
1788                                         start, len, blocksize);
1789         if (!journal) {
1790                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1791                 goto out_bdev;
1792         }
1793         journal->j_private = sb;
1794         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1795         wait_on_buffer(journal->j_sb_buffer);
1796         if (!buffer_uptodate(journal->j_sb_buffer)) {
1797                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1798                 goto out_journal;
1799         }
1800         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1801                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1802                                         "user (unsupported) - %d\n",
1803                         be32_to_cpu(journal->j_superblock->s_nr_users));
1804                 goto out_journal;
1805         }
1806         EXT3_SB(sb)->journal_bdev = bdev;
1807         ext3_init_journal_params(sb, journal);
1808         return journal;
1809 out_journal:
1810         journal_destroy(journal);
1811 out_bdev:
1812         ext3_blkdev_put(bdev);
1813         return NULL;
1814 }
1815
1816 static int ext3_load_journal(struct super_block * sb,
1817                              struct ext3_super_block * es)
1818 {
1819         journal_t *journal;
1820         int journal_inum = le32_to_cpu(es->s_journal_inum);
1821         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1822         int err = 0;
1823         int really_read_only;
1824
1825         really_read_only = bdev_read_only(sb->s_bdev);
1826
1827         /*
1828          * Are we loading a blank journal or performing recovery after a
1829          * crash?  For recovery, we need to check in advance whether we
1830          * can get read-write access to the device.
1831          */
1832
1833         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1834                 if (sb->s_flags & MS_RDONLY) {
1835                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1836                                         "required on readonly filesystem.\n");
1837                         if (really_read_only) {
1838                                 printk(KERN_ERR "EXT3-fs: write access "
1839                                         "unavailable, cannot proceed.\n");
1840                                 return -EROFS;
1841                         }
1842                         printk (KERN_INFO "EXT3-fs: write access will "
1843                                         "be enabled during recovery.\n");
1844                 }
1845         }
1846
1847         if (journal_inum && journal_dev) {
1848                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1849                        "and inode journals!\n");
1850                 return -EINVAL;
1851         }
1852
1853         if (journal_inum) {
1854                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1855                         return -EINVAL;
1856         } else {
1857                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1858                         return -EINVAL;
1859         }
1860
1861         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1862                 err = journal_update_format(journal);
1863                 if (err)  {
1864                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1865                         journal_destroy(journal);
1866                         return err;
1867                 }
1868         }
1869
1870         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1871                 err = journal_wipe(journal, !really_read_only);
1872         if (!err)
1873                 err = journal_load(journal);
1874
1875         if (err) {
1876                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1877                 journal_destroy(journal);
1878                 return err;
1879         }
1880
1881         EXT3_SB(sb)->s_journal = journal;
1882         ext3_clear_journal_err(sb, es);
1883         return 0;
1884 }
1885
1886 static int ext3_create_journal(struct super_block * sb,
1887                                struct ext3_super_block * es,
1888                                int journal_inum)
1889 {
1890         journal_t *journal;
1891
1892         if (sb->s_flags & MS_RDONLY) {
1893                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1894                                 "create journal.\n");
1895                 return -EROFS;
1896         }
1897
1898         if (!(journal = ext3_get_journal(sb, journal_inum)))
1899                 return -EINVAL;
1900
1901         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1902                journal_inum);
1903
1904         if (journal_create(journal)) {
1905                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1906                 journal_destroy(journal);
1907                 return -EIO;
1908         }
1909
1910         EXT3_SB(sb)->s_journal = journal;
1911
1912         ext3_update_dynamic_rev(sb);
1913         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1914         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1915
1916         es->s_journal_inum = cpu_to_le32(journal_inum);
1917         sb->s_dirt = 1;
1918
1919         /* Make sure we flush the recovery flag to disk. */
1920         ext3_commit_super(sb, es, 1);
1921
1922         return 0;
1923 }
1924
1925 static void ext3_commit_super (struct super_block * sb,
1926                                struct ext3_super_block * es,
1927                                int sync)
1928 {
1929         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1930
1931         if (!sbh)
1932                 return;
1933         es->s_wtime = cpu_to_le32(get_seconds());
1934         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1935         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1936         BUFFER_TRACE(sbh, "marking dirty");
1937         mark_buffer_dirty(sbh);
1938         if (sync)
1939                 sync_dirty_buffer(sbh);
1940 }
1941
1942
1943 /*
1944  * Have we just finished recovery?  If so, and if we are mounting (or
1945  * remounting) the filesystem readonly, then we will end up with a
1946  * consistent fs on disk.  Record that fact.
1947  */
1948 static void ext3_mark_recovery_complete(struct super_block * sb,
1949                                         struct ext3_super_block * es)
1950 {
1951         journal_t *journal = EXT3_SB(sb)->s_journal;
1952
1953         journal_lock_updates(journal);
1954         journal_flush(journal);
1955         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
1956             sb->s_flags & MS_RDONLY) {
1957                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1958                 sb->s_dirt = 0;
1959                 ext3_commit_super(sb, es, 1);
1960         }
1961         journal_unlock_updates(journal);
1962 }
1963
1964 /*
1965  * If we are mounting (or read-write remounting) a filesystem whose journal
1966  * has recorded an error from a previous lifetime, move that error to the
1967  * main filesystem now.
1968  */
1969 static void ext3_clear_journal_err(struct super_block * sb,
1970                                    struct ext3_super_block * es)
1971 {
1972         journal_t *journal;
1973         int j_errno;
1974         const char *errstr;
1975
1976         journal = EXT3_SB(sb)->s_journal;
1977
1978         /*
1979          * Now check for any error status which may have been recorded in the
1980          * journal by a prior ext3_error() or ext3_abort()
1981          */
1982
1983         j_errno = journal_errno(journal);
1984         if (j_errno) {
1985                 char nbuf[16];
1986
1987                 errstr = ext3_decode_error(sb, j_errno, nbuf);
1988                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
1989                              "from previous mount: %s", errstr);
1990                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
1991                              "filesystem check.");
1992
1993                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
1994                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
1995                 ext3_commit_super (sb, es, 1);
1996
1997                 journal_clear_err(journal);
1998         }
1999 }
2000
2001 /*
2002  * Force the running and committing transactions to commit,
2003  * and wait on the commit.
2004  */
2005 int ext3_force_commit(struct super_block *sb)
2006 {
2007         journal_t *journal;
2008         int ret;
2009
2010         if (sb->s_flags & MS_RDONLY)
2011                 return 0;
2012
2013         journal = EXT3_SB(sb)->s_journal;
2014         sb->s_dirt = 0;
2015         ret = ext3_journal_force_commit(journal);
2016         return ret;
2017 }
2018
2019 /*
2020  * Ext3 always journals updates to the superblock itself, so we don't
2021  * have to propagate any other updates to the superblock on disk at this
2022  * point.  Just start an async writeback to get the buffers on their way
2023  * to the disk.
2024  *
2025  * This implicitly triggers the writebehind on sync().
2026  */
2027
2028 static void ext3_write_super (struct super_block * sb)
2029 {
2030         if (down_trylock(&sb->s_lock) == 0)
2031                 BUG();
2032         sb->s_dirt = 0;
2033 }
2034
2035 static int ext3_sync_fs(struct super_block *sb, int wait)
2036 {
2037         tid_t target;
2038
2039         sb->s_dirt = 0;
2040         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2041                 if (wait)
2042                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2043         }
2044         return 0;
2045 }
2046
2047 /*
2048  * LVM calls this function before a (read-only) snapshot is created.  This
2049  * gives us a chance to flush the journal completely and mark the fs clean.
2050  */
2051 static void ext3_write_super_lockfs(struct super_block *sb)
2052 {
2053         sb->s_dirt = 0;
2054
2055         if (!(sb->s_flags & MS_RDONLY)) {
2056                 journal_t *journal = EXT3_SB(sb)->s_journal;
2057
2058                 /* Now we set up the journal barrier. */
2059                 journal_lock_updates(journal);
2060                 journal_flush(journal);
2061
2062                 /* Journal blocked and flushed, clear needs_recovery flag. */
2063                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2064                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2065         }
2066 }
2067
2068 /*
2069  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2070  * flag here, even though the filesystem is not technically dirty yet.
2071  */
2072 static void ext3_unlockfs(struct super_block *sb)
2073 {
2074         if (!(sb->s_flags & MS_RDONLY)) {
2075                 lock_super(sb);
2076                 /* Reser the needs_recovery flag before the fs is unlocked. */
2077                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2078                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2079                 unlock_super(sb);
2080                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2081         }
2082 }
2083
2084 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2085 {
2086         struct ext3_super_block * es;
2087         struct ext3_sb_info *sbi = EXT3_SB(sb);
2088         unsigned long tmp;
2089         unsigned long n_blocks_count = 0;
2090
2091         /*
2092          * Allow the "check" option to be passed as a remount option.
2093          */
2094         if (!parse_options(data, sb, &tmp, &n_blocks_count, 1))
2095                 return -EINVAL;
2096
2097         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2098                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2099
2100         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2101                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2102
2103         es = sbi->s_es;
2104
2105         ext3_init_journal_params(sb, sbi->s_journal);
2106
2107         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2108                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2109                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2110                         return -EROFS;
2111
2112                 if (*flags & MS_RDONLY) {
2113                         /*
2114                          * First of all, the unconditional stuff we have to do
2115                          * to disable replay of the journal when we next remount
2116                          */
2117                         sb->s_flags |= MS_RDONLY;
2118
2119                         /*
2120                          * OK, test if we are remounting a valid rw partition
2121                          * readonly, and if so set the rdonly flag and then
2122                          * mark the partition as valid again.
2123                          */
2124                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2125                             (sbi->s_mount_state & EXT3_VALID_FS))
2126                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2127
2128                         ext3_mark_recovery_complete(sb, es);
2129                 } else {
2130                         __le32 ret;
2131                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2132                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2133                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2134                                        "remount RDWR because of unsupported "
2135                                        "optional features (%x).\n",
2136                                        sb->s_id, le32_to_cpu(ret));
2137                                 return -EROFS;
2138                         }
2139                         /*
2140                          * Mounting a RDONLY partition read-write, so reread
2141                          * and store the current valid flag.  (It may have
2142                          * been changed by e2fsck since we originally mounted
2143                          * the partition.)
2144                          */
2145                         ext3_clear_journal_err(sb, es);
2146                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2147                         if ((ret = ext3_group_extend(sb, es, n_blocks_count)))
2148                                 return ret;
2149                         if (!ext3_setup_super (sb, es, 0))
2150                                 sb->s_flags &= ~MS_RDONLY;
2151                 }
2152         }
2153         return 0;
2154 }
2155
2156 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2157 {
2158         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2159         unsigned long overhead;
2160         int i;
2161
2162         if (test_opt (sb, MINIX_DF))
2163                 overhead = 0;
2164         else {
2165                 unsigned long ngroups;
2166                 ngroups = EXT3_SB(sb)->s_groups_count;
2167                 smp_rmb();
2168
2169                 /*
2170                  * Compute the overhead (FS structures)
2171                  */
2172
2173                 /*
2174                  * All of the blocks before first_data_block are
2175                  * overhead
2176                  */
2177                 overhead = le32_to_cpu(es->s_first_data_block);
2178
2179                 /*
2180                  * Add the overhead attributed to the superblock and
2181                  * block group descriptors.  If the sparse superblocks
2182                  * feature is turned on, then not all groups have this.
2183                  */
2184                 for (i = 0; i < ngroups; i++) {
2185                         overhead += ext3_bg_has_super(sb, i) +
2186                                 ext3_bg_num_gdb(sb, i);
2187                         cond_resched();
2188                 }
2189
2190                 /*
2191                  * Every block group has an inode bitmap, a block
2192                  * bitmap, and an inode table.
2193                  */
2194                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2195         }
2196
2197         buf->f_type = EXT3_SUPER_MAGIC;
2198         buf->f_bsize = sb->s_blocksize;
2199         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2200         buf->f_bfree = ext3_count_free_blocks (sb);
2201         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2202         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2203                 buf->f_bavail = 0;
2204         buf->f_files = le32_to_cpu(es->s_inodes_count);
2205         buf->f_ffree = ext3_count_free_inodes (sb);
2206         buf->f_namelen = EXT3_NAME_LEN;
2207         return 0;
2208 }
2209
2210 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2211  * is locked for write. Otherwise the are possible deadlocks:
2212  * Process 1                         Process 2
2213  * ext3_create()                     quota_sync()
2214  *   journal_start()                   write_dquot()
2215  *   DQUOT_INIT()                        down(dqio_sem)
2216  *     down(dqio_sem)                    journal_start()
2217  *
2218  */
2219
2220 #ifdef CONFIG_QUOTA
2221
2222 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2223 {
2224         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2225 }
2226
2227 static int ext3_dquot_initialize(struct inode *inode, int type)
2228 {
2229         handle_t *handle;
2230         int ret, err;
2231
2232         /* We may create quota structure so we need to reserve enough blocks */
2233         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2234         if (IS_ERR(handle))
2235                 return PTR_ERR(handle);
2236         ret = dquot_initialize(inode, type);
2237         err = ext3_journal_stop(handle);
2238         if (!ret)
2239                 ret = err;
2240         return ret;
2241 }
2242
2243 static int ext3_dquot_drop(struct inode *inode)
2244 {
2245         handle_t *handle;
2246         int ret, err;
2247
2248         /* We may delete quota structure so we need to reserve enough blocks */
2249         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2250         if (IS_ERR(handle))
2251                 return PTR_ERR(handle);
2252         ret = dquot_drop(inode);
2253         err = ext3_journal_stop(handle);
2254         if (!ret)
2255                 ret = err;
2256         return ret;
2257 }
2258
2259 static int ext3_write_dquot(struct dquot *dquot)
2260 {
2261         int ret, err;
2262         handle_t *handle;
2263         struct inode *inode;
2264
2265         inode = dquot_to_inode(dquot);
2266         handle = ext3_journal_start(inode,
2267                                         EXT3_QUOTA_TRANS_BLOCKS);
2268         if (IS_ERR(handle))
2269                 return PTR_ERR(handle);
2270         ret = dquot_commit(dquot);
2271         err = ext3_journal_stop(handle);
2272         if (!ret)
2273                 ret = err;
2274         return ret;
2275 }
2276
2277 static int ext3_acquire_dquot(struct dquot *dquot)
2278 {
2279         int ret, err;
2280         handle_t *handle;
2281
2282         handle = ext3_journal_start(dquot_to_inode(dquot),
2283                                         EXT3_QUOTA_INIT_BLOCKS);
2284         if (IS_ERR(handle))
2285                 return PTR_ERR(handle);
2286         ret = dquot_acquire(dquot);
2287         err = ext3_journal_stop(handle);
2288         if (!ret)
2289                 ret = err;
2290         return ret;
2291 }
2292
2293 static int ext3_release_dquot(struct dquot *dquot)
2294 {
2295         int ret, err;
2296         handle_t *handle;
2297
2298         handle = ext3_journal_start(dquot_to_inode(dquot),
2299                                         EXT3_QUOTA_INIT_BLOCKS);
2300         if (IS_ERR(handle))
2301                 return PTR_ERR(handle);
2302         ret = dquot_release(dquot);
2303         err = ext3_journal_stop(handle);
2304         if (!ret)
2305                 ret = err;
2306         return ret;
2307 }
2308
2309 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2310 {
2311         /* Are we journalling quotas? */
2312         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2313             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2314                 dquot_mark_dquot_dirty(dquot);
2315                 return ext3_write_dquot(dquot);
2316         } else {
2317                 return dquot_mark_dquot_dirty(dquot);
2318         }
2319 }
2320
2321 static int ext3_write_info(struct super_block *sb, int type)
2322 {
2323         int ret, err;
2324         handle_t *handle;
2325
2326         /* Data block + inode block */
2327         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2328         if (IS_ERR(handle))
2329                 return PTR_ERR(handle);
2330         ret = dquot_commit_info(sb, type);
2331         err = ext3_journal_stop(handle);
2332         if (!ret)
2333                 ret = err;
2334         return ret;
2335 }
2336
2337 /*
2338  * Turn on quotas during mount time - we need to find
2339  * the quota file and such...
2340  */
2341 static int ext3_quota_on_mount(struct super_block *sb, int type)
2342 {
2343         int err;
2344         struct dentry *dentry;
2345         struct qstr name = { .name = EXT3_SB(sb)->s_qf_names[type],
2346                              .hash = 0,
2347                              .len = strlen(EXT3_SB(sb)->s_qf_names[type])};
2348
2349         dentry = lookup_hash(&name, sb->s_root);
2350         if (IS_ERR(dentry))
2351                 return PTR_ERR(dentry);
2352         err = vfs_quota_on_mount(type, EXT3_SB(sb)->s_jquota_fmt, dentry);
2353         /* Now invalidate and put the dentry - quota got its own reference
2354          * to inode and dentry has at least wrong hash so we had better
2355          * throw it away */
2356         d_invalidate(dentry);
2357         dput(dentry);
2358         return err;
2359 }
2360
2361 /*
2362  * Standard function to be called on quota_on
2363  */
2364 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2365                          char *path)
2366 {
2367         int err;
2368         struct nameidata nd;
2369
2370         /* Not journalling quota? */
2371         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2372             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2373                 return vfs_quota_on(sb, type, format_id, path);
2374         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2375         if (err)
2376                 return err;
2377         /* Quotafile not on the same filesystem? */
2378         if (nd.mnt->mnt_sb != sb) {
2379                 path_release(&nd);
2380                 return -EXDEV;
2381         }
2382         /* Quotafile not of fs root? */
2383         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2384                 printk(KERN_WARNING
2385                         "EXT3-fs: Quota file not on filesystem root. "
2386                         "Journalled quota will not work.\n");
2387         path_release(&nd);
2388         return vfs_quota_on(sb, type, format_id, path);
2389 }
2390
2391 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2392  * acquiring the locks... As quota files are never truncated and quota code
2393  * itself serializes the operations (and noone else should touch the files)
2394  * we don't have to be afraid of races */
2395 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2396                                size_t len, loff_t off)
2397 {
2398         struct inode *inode = sb_dqopt(sb)->files[type];
2399         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2400         int err = 0;
2401         int offset = off & (sb->s_blocksize - 1);
2402         int tocopy;
2403         size_t toread;
2404         struct buffer_head *bh;
2405         loff_t i_size = i_size_read(inode);
2406
2407         if (off > i_size)
2408                 return 0;
2409         if (off+len > i_size)
2410                 len = i_size-off;
2411         toread = len;
2412         while (toread > 0) {
2413                 tocopy = sb->s_blocksize - offset < toread ?
2414                                 sb->s_blocksize - offset : toread;
2415                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2416                 if (err)
2417                         return err;
2418                 if (!bh)        /* A hole? */
2419                         memset(data, 0, tocopy);
2420                 else
2421                         memcpy(data, bh->b_data+offset, tocopy);
2422                 brelse(bh);
2423                 offset = 0;
2424                 toread -= tocopy;
2425                 data += tocopy;
2426                 blk++;
2427         }
2428         return len;
2429 }
2430
2431 /* Write to quotafile (we know the transaction is already started and has
2432  * enough credits) */
2433 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2434                                 const char *data, size_t len, loff_t off)
2435 {
2436         struct inode *inode = sb_dqopt(sb)->files[type];
2437         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2438         int err = 0;
2439         int offset = off & (sb->s_blocksize - 1);
2440         int tocopy;
2441         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2442         size_t towrite = len;
2443         struct buffer_head *bh;
2444         handle_t *handle = journal_current_handle();
2445
2446         down(&inode->i_sem);
2447         while (towrite > 0) {
2448                 tocopy = sb->s_blocksize - offset < towrite ?
2449                                 sb->s_blocksize - offset : towrite;
2450                 bh = ext3_bread(handle, inode, blk, 1, &err);
2451                 if (!bh)
2452                         goto out;
2453                 if (journal_quota) {
2454                         err = ext3_journal_get_write_access(handle, bh);
2455                         if (err) {
2456                                 brelse(bh);
2457                                 goto out;
2458                         }
2459                 }
2460                 lock_buffer(bh);
2461                 memcpy(bh->b_data+offset, data, tocopy);
2462                 flush_dcache_page(bh->b_page);
2463                 unlock_buffer(bh);
2464                 if (journal_quota)
2465                         err = ext3_journal_dirty_metadata(handle, bh);
2466                 else {
2467                         /* Always do at least ordered writes for quotas */
2468                         err = ext3_journal_dirty_data(handle, bh);
2469                         mark_buffer_dirty(bh);
2470                 }
2471                 brelse(bh);
2472                 if (err)
2473                         goto out;
2474                 offset = 0;
2475                 towrite -= tocopy;
2476                 data += tocopy;
2477                 blk++;
2478         }
2479 out:
2480         if (len == towrite)
2481                 return err;
2482         if (inode->i_size < off+len-towrite) {
2483                 i_size_write(inode, off+len-towrite);
2484                 EXT3_I(inode)->i_disksize = inode->i_size;
2485         }
2486         inode->i_version++;
2487         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2488         ext3_mark_inode_dirty(handle, inode);
2489         up(&inode->i_sem);
2490         return len - towrite;
2491 }
2492
2493 #endif
2494
2495 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2496         int flags, const char *dev_name, void *data)
2497 {
2498         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2499 }
2500
2501 static struct file_system_type ext3_fs_type = {
2502         .owner          = THIS_MODULE,
2503         .name           = "ext3",
2504         .get_sb         = ext3_get_sb,
2505         .kill_sb        = kill_block_super,
2506         .fs_flags       = FS_REQUIRES_DEV,
2507 };
2508
2509 static int __init init_ext3_fs(void)
2510 {
2511         int err = init_ext3_xattr();
2512         if (err)
2513                 return err;
2514         err = init_inodecache();
2515         if (err)
2516                 goto out1;
2517         err = register_filesystem(&ext3_fs_type);
2518         if (err)
2519                 goto out;
2520         return 0;
2521 out:
2522         destroy_inodecache();
2523 out1:
2524         exit_ext3_xattr();
2525         return err;
2526 }
2527
2528 static void __exit exit_ext3_fs(void)
2529 {
2530         unregister_filesystem(&ext3_fs_type);
2531         destroy_inodecache();
2532         exit_ext3_xattr();
2533 }
2534
2535 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2536 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2537 MODULE_LICENSE("GPL");
2538 module_init(init_ext3_fs)
2539 module_exit(exit_ext3_fs)