Merge acpi-2.6.12 to-akpm
[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;
229
230         /* Special case: if the error is EROFS, and we're not already
231          * inside a transaction, then there's really no point in logging
232          * an error. */
233         if (errno == -EROFS && journal_current_handle() == NULL &&
234             (sb->s_flags & MS_RDONLY))
235                 return;
236
237         errstr = ext3_decode_error(sb, errno, nbuf);
238         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
239                 sb->s_id, function, errstr);
240
241         ext3_handle_error(sb);
242 }
243
244 /*
245  * ext3_abort is a much stronger failure handler than ext3_error.  The
246  * abort function may be used to deal with unrecoverable failures such
247  * as journal IO errors or ENOMEM at a critical moment in log management.
248  *
249  * We unconditionally force the filesystem into an ABORT|READONLY state,
250  * unless the error response on the fs has been set to panic in which
251  * case we take the easy way out and panic immediately.
252  */
253
254 void ext3_abort (struct super_block * sb, const char * function,
255                  const char * fmt, ...)
256 {
257         va_list args;
258
259         printk (KERN_CRIT "ext3_abort called.\n");
260
261         va_start(args, fmt);
262         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
263         vprintk(fmt, args);
264         printk("\n");
265         va_end(args);
266
267         if (test_opt(sb, ERRORS_PANIC))
268                 panic("EXT3-fs panic from previous error\n");
269
270         if (sb->s_flags & MS_RDONLY)
271                 return;
272
273         printk(KERN_CRIT "Remounting filesystem read-only\n");
274         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
275         sb->s_flags |= MS_RDONLY;
276         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
277         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
278 }
279
280 void ext3_warning (struct super_block * sb, const char * function,
281                    const char * fmt, ...)
282 {
283         va_list args;
284
285         va_start(args, fmt);
286         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
287                sb->s_id, function);
288         vprintk(fmt, args);
289         printk("\n");
290         va_end(args);
291 }
292
293 void ext3_update_dynamic_rev(struct super_block *sb)
294 {
295         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
296
297         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
298                 return;
299
300         ext3_warning(sb, __FUNCTION__,
301                      "updating to rev %d because of new feature flag, "
302                      "running e2fsck is recommended",
303                      EXT3_DYNAMIC_REV);
304
305         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
306         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
307         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
308         /* leave es->s_feature_*compat flags alone */
309         /* es->s_uuid will be set by e2fsck if empty */
310
311         /*
312          * The rest of the superblock fields should be zero, and if not it
313          * means they are likely already in use, so leave them alone.  We
314          * can leave it up to e2fsck to clean up any inconsistencies there.
315          */
316 }
317
318 /*
319  * Open the external journal device
320  */
321 static struct block_device *ext3_blkdev_get(dev_t dev)
322 {
323         struct block_device *bdev;
324         char b[BDEVNAME_SIZE];
325
326         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
327         if (IS_ERR(bdev))
328                 goto fail;
329         return bdev;
330
331 fail:
332         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
333                         __bdevname(dev, b), PTR_ERR(bdev));
334         return NULL;
335 }
336
337 /*
338  * Release the journal device
339  */
340 static int ext3_blkdev_put(struct block_device *bdev)
341 {
342         bd_release(bdev);
343         return blkdev_put(bdev);
344 }
345
346 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
347 {
348         struct block_device *bdev;
349         int ret = -ENODEV;
350
351         bdev = sbi->journal_bdev;
352         if (bdev) {
353                 ret = ext3_blkdev_put(bdev);
354                 sbi->journal_bdev = NULL;
355         }
356         return ret;
357 }
358
359 static inline struct inode *orphan_list_entry(struct list_head *l)
360 {
361         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
362 }
363
364 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
365 {
366         struct list_head *l;
367
368         printk(KERN_ERR "sb orphan head is %d\n", 
369                le32_to_cpu(sbi->s_es->s_last_orphan));
370
371         printk(KERN_ERR "sb_info orphan list:\n");
372         list_for_each(l, &sbi->s_orphan) {
373                 struct inode *inode = orphan_list_entry(l);
374                 printk(KERN_ERR "  "
375                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
376                        inode->i_sb->s_id, inode->i_ino, inode,
377                        inode->i_mode, inode->i_nlink, 
378                        NEXT_ORPHAN(inode));
379         }
380 }
381
382 static void ext3_put_super (struct super_block * sb)
383 {
384         struct ext3_sb_info *sbi = EXT3_SB(sb);
385         struct ext3_super_block *es = sbi->s_es;
386         int i;
387
388         ext3_xattr_put_super(sb);
389         journal_destroy(sbi->s_journal);
390         if (!(sb->s_flags & MS_RDONLY)) {
391                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
392                 es->s_state = cpu_to_le16(sbi->s_mount_state);
393                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
394                 mark_buffer_dirty(sbi->s_sbh);
395                 ext3_commit_super(sb, es, 1);
396         }
397
398         for (i = 0; i < sbi->s_gdb_count; i++)
399                 brelse(sbi->s_group_desc[i]);
400         kfree(sbi->s_group_desc);
401         percpu_counter_destroy(&sbi->s_freeblocks_counter);
402         percpu_counter_destroy(&sbi->s_freeinodes_counter);
403         percpu_counter_destroy(&sbi->s_dirs_counter);
404         brelse(sbi->s_sbh);
405 #ifdef CONFIG_QUOTA
406         for (i = 0; i < MAXQUOTAS; i++)
407                 kfree(sbi->s_qf_names[i]);
408 #endif
409
410         /* Debugging code just in case the in-memory inode orphan list
411          * isn't empty.  The on-disk one can be non-empty if we've
412          * detected an error and taken the fs readonly, but the
413          * in-memory list had better be clean by this point. */
414         if (!list_empty(&sbi->s_orphan))
415                 dump_orphan_list(sb, sbi);
416         J_ASSERT(list_empty(&sbi->s_orphan));
417
418         invalidate_bdev(sb->s_bdev, 0);
419         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
420                 /*
421                  * Invalidate the journal device's buffers.  We don't want them
422                  * floating about in memory - the physical journal device may
423                  * hotswapped, and it breaks the `ro-after' testing code.
424                  */
425                 sync_blockdev(sbi->journal_bdev);
426                 invalidate_bdev(sbi->journal_bdev, 0);
427                 ext3_blkdev_remove(sbi);
428         }
429         sb->s_fs_info = NULL;
430         kfree(sbi);
431         return;
432 }
433
434 static kmem_cache_t *ext3_inode_cachep;
435
436 /*
437  * Called inside transaction, so use GFP_NOFS
438  */
439 static struct inode *ext3_alloc_inode(struct super_block *sb)
440 {
441         struct ext3_inode_info *ei;
442
443         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
444         if (!ei)
445                 return NULL;
446 #ifdef CONFIG_EXT3_FS_POSIX_ACL
447         ei->i_acl = EXT3_ACL_NOT_CACHED;
448         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
449 #endif
450         ei->i_block_alloc_info = NULL;
451         ei->vfs_inode.i_version = 1;
452         return &ei->vfs_inode;
453 }
454
455 static void ext3_destroy_inode(struct inode *inode)
456 {
457         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
458 }
459
460 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
461 {
462         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
463
464         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
465             SLAB_CTOR_CONSTRUCTOR) {
466                 INIT_LIST_HEAD(&ei->i_orphan);
467 #ifdef CONFIG_EXT3_FS_XATTR
468                 init_rwsem(&ei->xattr_sem);
469 #endif
470                 init_MUTEX(&ei->truncate_sem);
471                 inode_init_once(&ei->vfs_inode);
472         }
473 }
474  
475 static int init_inodecache(void)
476 {
477         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
478                                              sizeof(struct ext3_inode_info),
479                                              0, SLAB_RECLAIM_ACCOUNT,
480                                              init_once, NULL);
481         if (ext3_inode_cachep == NULL)
482                 return -ENOMEM;
483         return 0;
484 }
485
486 static void destroy_inodecache(void)
487 {
488         if (kmem_cache_destroy(ext3_inode_cachep))
489                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
490 }
491
492 static void ext3_clear_inode(struct inode *inode)
493 {
494         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
495 #ifdef CONFIG_EXT3_FS_POSIX_ACL
496        if (EXT3_I(inode)->i_acl &&
497            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
498                posix_acl_release(EXT3_I(inode)->i_acl);
499                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
500        }
501        if (EXT3_I(inode)->i_default_acl &&
502            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
503                posix_acl_release(EXT3_I(inode)->i_default_acl);
504                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
505        }
506 #endif
507         ext3_discard_reservation(inode);
508         EXT3_I(inode)->i_block_alloc_info = NULL;
509         kfree(rsv);
510 }
511
512 #ifdef CONFIG_QUOTA
513
514 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
515 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
516
517 static int ext3_dquot_initialize(struct inode *inode, int type);
518 static int ext3_dquot_drop(struct inode *inode);
519 static int ext3_write_dquot(struct dquot *dquot);
520 static int ext3_acquire_dquot(struct dquot *dquot);
521 static int ext3_release_dquot(struct dquot *dquot);
522 static int ext3_mark_dquot_dirty(struct dquot *dquot);
523 static int ext3_write_info(struct super_block *sb, int type);
524 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
525 static int ext3_quota_on_mount(struct super_block *sb, int type);
526 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
527                                size_t len, loff_t off);
528 static ssize_t ext3_quota_write(struct super_block *sb, int type,
529                                 const char *data, size_t len, loff_t off);
530
531 static struct dquot_operations ext3_quota_operations = {
532         .initialize     = ext3_dquot_initialize,
533         .drop           = ext3_dquot_drop,
534         .alloc_space    = dquot_alloc_space,
535         .alloc_inode    = dquot_alloc_inode,
536         .free_space     = dquot_free_space,
537         .free_inode     = dquot_free_inode,
538         .transfer       = dquot_transfer,
539         .write_dquot    = ext3_write_dquot,
540         .acquire_dquot  = ext3_acquire_dquot,
541         .release_dquot  = ext3_release_dquot,
542         .mark_dirty     = ext3_mark_dquot_dirty,
543         .write_info     = ext3_write_info
544 };
545
546 static struct quotactl_ops ext3_qctl_operations = {
547         .quota_on       = ext3_quota_on,
548         .quota_off      = vfs_quota_off,
549         .quota_sync     = vfs_quota_sync,
550         .get_info       = vfs_get_dqinfo,
551         .set_info       = vfs_set_dqinfo,
552         .get_dqblk      = vfs_get_dqblk,
553         .set_dqblk      = vfs_set_dqblk
554 };
555 #endif
556
557 static struct super_operations ext3_sops = {
558         .alloc_inode    = ext3_alloc_inode,
559         .destroy_inode  = ext3_destroy_inode,
560         .read_inode     = ext3_read_inode,
561         .write_inode    = ext3_write_inode,
562         .dirty_inode    = ext3_dirty_inode,
563         .delete_inode   = ext3_delete_inode,
564         .put_super      = ext3_put_super,
565         .write_super    = ext3_write_super,
566         .sync_fs        = ext3_sync_fs,
567         .write_super_lockfs = ext3_write_super_lockfs,
568         .unlockfs       = ext3_unlockfs,
569         .statfs         = ext3_statfs,
570         .remount_fs     = ext3_remount,
571         .clear_inode    = ext3_clear_inode,
572 #ifdef CONFIG_QUOTA
573         .quota_read     = ext3_quota_read,
574         .quota_write    = ext3_quota_write,
575 #endif
576 };
577
578 struct dentry *ext3_get_parent(struct dentry *child);
579 static struct export_operations ext3_export_ops = {
580         .get_parent = ext3_get_parent,
581 };
582
583 enum {
584         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
585         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
586         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
587         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
588         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh,
589         Opt_commit, Opt_journal_update, Opt_journal_inum,
590         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
591         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
592         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
593         Opt_ignore, Opt_barrier, Opt_err, Opt_resize,
594 };
595
596 static match_table_t tokens = {
597         {Opt_bsd_df, "bsddf"},
598         {Opt_minix_df, "minixdf"},
599         {Opt_grpid, "grpid"},
600         {Opt_grpid, "bsdgroups"},
601         {Opt_nogrpid, "nogrpid"},
602         {Opt_nogrpid, "sysvgroups"},
603         {Opt_resgid, "resgid=%u"},
604         {Opt_resuid, "resuid=%u"},
605         {Opt_sb, "sb=%u"},
606         {Opt_err_cont, "errors=continue"},
607         {Opt_err_panic, "errors=panic"},
608         {Opt_err_ro, "errors=remount-ro"},
609         {Opt_nouid32, "nouid32"},
610         {Opt_nocheck, "nocheck"},
611         {Opt_nocheck, "check=none"},
612         {Opt_check, "check"},
613         {Opt_debug, "debug"},
614         {Opt_oldalloc, "oldalloc"},
615         {Opt_orlov, "orlov"},
616         {Opt_user_xattr, "user_xattr"},
617         {Opt_nouser_xattr, "nouser_xattr"},
618         {Opt_acl, "acl"},
619         {Opt_noacl, "noacl"},
620         {Opt_reservation, "reservation"},
621         {Opt_noreservation, "noreservation"},
622         {Opt_noload, "noload"},
623         {Opt_nobh, "nobh"},
624         {Opt_commit, "commit=%u"},
625         {Opt_journal_update, "journal=update"},
626         {Opt_journal_inum, "journal=%u"},
627         {Opt_abort, "abort"},
628         {Opt_data_journal, "data=journal"},
629         {Opt_data_ordered, "data=ordered"},
630         {Opt_data_writeback, "data=writeback"},
631         {Opt_offusrjquota, "usrjquota="},
632         {Opt_usrjquota, "usrjquota=%s"},
633         {Opt_offgrpjquota, "grpjquota="},
634         {Opt_grpjquota, "grpjquota=%s"},
635         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
636         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
637         {Opt_quota, "grpquota"},
638         {Opt_noquota, "noquota"},
639         {Opt_quota, "quota"},
640         {Opt_quota, "usrquota"},
641         {Opt_barrier, "barrier=%u"},
642         {Opt_err, NULL},
643         {Opt_resize, "resize"},
644 };
645
646 static unsigned long get_sb_block(void **data)
647 {
648         unsigned long   sb_block;
649         char            *options = (char *) *data;
650
651         if (!options || strncmp(options, "sb=", 3) != 0)
652                 return 1;       /* Default location */
653         options += 3;
654         sb_block = simple_strtoul(options, &options, 0);
655         if (*options && *options != ',') {
656                 printk("EXT3-fs: Invalid sb specification: %s\n",
657                        (char *) *data);
658                 return 1;
659         }
660         if (*options == ',')
661                 options++;
662         *data = (void *) options;
663         return sb_block;
664 }
665
666 static int parse_options (char * options, struct super_block *sb,
667                           unsigned long * inum, unsigned long *n_blocks_count, int is_remount)
668 {
669         struct ext3_sb_info *sbi = EXT3_SB(sb);
670         char * p;
671         substring_t args[MAX_OPT_ARGS];
672         int data_opt = 0;
673         int option;
674 #ifdef CONFIG_QUOTA
675         int qtype;
676         char *qname;
677 #endif
678
679         if (!options)
680                 return 1;
681
682         while ((p = strsep (&options, ",")) != NULL) {
683                 int token;
684                 if (!*p)
685                         continue;
686
687                 token = match_token(p, tokens, args);
688                 switch (token) {
689                 case Opt_bsd_df:
690                         clear_opt (sbi->s_mount_opt, MINIX_DF);
691                         break;
692                 case Opt_minix_df:
693                         set_opt (sbi->s_mount_opt, MINIX_DF);
694                         break;
695                 case Opt_grpid:
696                         set_opt (sbi->s_mount_opt, GRPID);
697                         break;
698                 case Opt_nogrpid:
699                         clear_opt (sbi->s_mount_opt, GRPID);
700                         break;
701                 case Opt_resuid:
702                         if (match_int(&args[0], &option))
703                                 return 0;
704                         sbi->s_resuid = option;
705                         break;
706                 case Opt_resgid:
707                         if (match_int(&args[0], &option))
708                                 return 0;
709                         sbi->s_resgid = option;
710                         break;
711                 case Opt_sb:
712                         /* handled by get_sb_block() instead of here */
713                         /* *sb_block = match_int(&args[0]); */
714                         break;
715                 case Opt_err_panic:
716                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
717                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
718                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
719                         break;
720                 case Opt_err_ro:
721                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
722                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
723                         set_opt (sbi->s_mount_opt, ERRORS_RO);
724                         break;
725                 case Opt_err_cont:
726                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
727                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
728                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
729                         break;
730                 case Opt_nouid32:
731                         set_opt (sbi->s_mount_opt, NO_UID32);
732                         break;
733                 case Opt_check:
734 #ifdef CONFIG_EXT3_CHECK
735                         set_opt (sbi->s_mount_opt, CHECK);
736 #else
737                         printk(KERN_ERR
738                                "EXT3 Check option not supported\n");
739 #endif
740                         break;
741                 case Opt_nocheck:
742                         clear_opt (sbi->s_mount_opt, CHECK);
743                         break;
744                 case Opt_debug:
745                         set_opt (sbi->s_mount_opt, DEBUG);
746                         break;
747                 case Opt_oldalloc:
748                         set_opt (sbi->s_mount_opt, OLDALLOC);
749                         break;
750                 case Opt_orlov:
751                         clear_opt (sbi->s_mount_opt, OLDALLOC);
752                         break;
753 #ifdef CONFIG_EXT3_FS_XATTR
754                 case Opt_user_xattr:
755                         set_opt (sbi->s_mount_opt, XATTR_USER);
756                         break;
757                 case Opt_nouser_xattr:
758                         clear_opt (sbi->s_mount_opt, XATTR_USER);
759                         break;
760 #else
761                 case Opt_user_xattr:
762                 case Opt_nouser_xattr:
763                         printk("EXT3 (no)user_xattr options not supported\n");
764                         break;
765 #endif
766 #ifdef CONFIG_EXT3_FS_POSIX_ACL
767                 case Opt_acl:
768                         set_opt(sbi->s_mount_opt, POSIX_ACL);
769                         break;
770                 case Opt_noacl:
771                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
772                         break;
773 #else
774                 case Opt_acl:
775                 case Opt_noacl:
776                         printk("EXT3 (no)acl options not supported\n");
777                         break;
778 #endif
779                 case Opt_reservation:
780                         set_opt(sbi->s_mount_opt, RESERVATION);
781                         break;
782                 case Opt_noreservation:
783                         clear_opt(sbi->s_mount_opt, RESERVATION);
784                         break;
785                 case Opt_journal_update:
786                         /* @@@ FIXME */
787                         /* Eventually we will want to be able to create
788                            a journal file here.  For now, only allow the
789                            user to specify an existing inode to be the
790                            journal file. */
791                         if (is_remount) {
792                                 printk(KERN_ERR "EXT3-fs: cannot specify "
793                                        "journal on remount\n");
794                                 return 0;
795                         }
796                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
797                         break;
798                 case Opt_journal_inum:
799                         if (is_remount) {
800                                 printk(KERN_ERR "EXT3-fs: cannot specify "
801                                        "journal on remount\n");
802                                 return 0;
803                         }
804                         if (match_int(&args[0], &option))
805                                 return 0;
806                         *inum = option;
807                         break;
808                 case Opt_noload:
809                         set_opt (sbi->s_mount_opt, NOLOAD);
810                         break;
811                 case Opt_commit:
812                         if (match_int(&args[0], &option))
813                                 return 0;
814                         if (option < 0)
815                                 return 0;
816                         if (option == 0)
817                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
818                         sbi->s_commit_interval = HZ * option;
819                         break;
820                 case Opt_data_journal:
821                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
822                         goto datacheck;
823                 case Opt_data_ordered:
824                         data_opt = EXT3_MOUNT_ORDERED_DATA;
825                         goto datacheck;
826                 case Opt_data_writeback:
827                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
828                 datacheck:
829                         if (is_remount) {
830                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
831                                                 != data_opt) {
832                                         printk(KERN_ERR
833                                                 "EXT3-fs: cannot change data "
834                                                 "mode on remount\n");
835                                         return 0;
836                                 }
837                         } else {
838                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
839                                 sbi->s_mount_opt |= data_opt;
840                         }
841                         break;
842 #ifdef CONFIG_QUOTA
843                 case Opt_usrjquota:
844                         qtype = USRQUOTA;
845                         goto set_qf_name;
846                 case Opt_grpjquota:
847                         qtype = GRPQUOTA;
848 set_qf_name:
849                         if (sb_any_quota_enabled(sb)) {
850                                 printk(KERN_ERR
851                                         "EXT3-fs: Cannot change journalled "
852                                         "quota options when quota turned on.\n");
853                                 return 0;
854                         }
855                         qname = match_strdup(&args[0]);
856                         if (!qname) {
857                                 printk(KERN_ERR
858                                         "EXT3-fs: not enough memory for "
859                                         "storing quotafile name.\n");
860                                 return 0;
861                         }
862                         if (sbi->s_qf_names[qtype] &&
863                             strcmp(sbi->s_qf_names[qtype], qname)) {
864                                 printk(KERN_ERR
865                                         "EXT3-fs: %s quota file already "
866                                         "specified.\n", QTYPE2NAME(qtype));
867                                 kfree(qname);
868                                 return 0;
869                         }
870                         sbi->s_qf_names[qtype] = qname;
871                         if (strchr(sbi->s_qf_names[qtype], '/')) {
872                                 printk(KERN_ERR
873                                         "EXT3-fs: quotafile must be on "
874                                         "filesystem root.\n");
875                                 kfree(sbi->s_qf_names[qtype]);
876                                 sbi->s_qf_names[qtype] = NULL;
877                                 return 0;
878                         }
879                         set_opt(sbi->s_mount_opt, QUOTA);
880                         break;
881                 case Opt_offusrjquota:
882                         qtype = USRQUOTA;
883                         goto clear_qf_name;
884                 case Opt_offgrpjquota:
885                         qtype = GRPQUOTA;
886 clear_qf_name:
887                         if (sb_any_quota_enabled(sb)) {
888                                 printk(KERN_ERR "EXT3-fs: Cannot change "
889                                         "journalled quota options when "
890                                         "quota turned on.\n");
891                                 return 0;
892                         }
893                         /*
894                          * The space will be released later when all options
895                          * are confirmed to be correct
896                          */
897                         sbi->s_qf_names[qtype] = NULL;
898                         break;
899                 case Opt_jqfmt_vfsold:
900                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
901                         break;
902                 case Opt_jqfmt_vfsv0:
903                         sbi->s_jquota_fmt = QFMT_VFS_V0;
904                         break;
905                 case Opt_quota:
906                         set_opt(sbi->s_mount_opt, QUOTA);
907                         break;
908                 case Opt_noquota:
909                         if (sb_any_quota_enabled(sb)) {
910                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
911                                         "options when quota turned on.\n");
912                                 return 0;
913                         }
914                         clear_opt(sbi->s_mount_opt, QUOTA);
915                         break;
916 #else
917                 case Opt_usrjquota:
918                 case Opt_grpjquota:
919                 case Opt_offusrjquota:
920                 case Opt_offgrpjquota:
921                 case Opt_jqfmt_vfsold:
922                 case Opt_jqfmt_vfsv0:
923                         printk(KERN_ERR
924                                 "EXT3-fs: journalled quota options not "
925                                 "supported.\n");
926                         break;
927                 case Opt_quota:
928                 case Opt_noquota:
929                         break;
930 #endif
931                 case Opt_abort:
932                         set_opt(sbi->s_mount_opt, ABORT);
933                         break;
934                 case Opt_barrier:
935                         if (match_int(&args[0], &option))
936                                 return 0;
937                         if (option)
938                                 set_opt(sbi->s_mount_opt, BARRIER);
939                         else
940                                 clear_opt(sbi->s_mount_opt, BARRIER);
941                         break;
942                 case Opt_ignore:
943                         break;
944                 case Opt_resize:
945                         if (!is_remount) {
946                                 printk("EXT3-fs: resize option only available "
947                                         "for remount\n");
948                                 return 0;
949                         }
950                         if (match_int(&args[0], &option) != 0)
951                                 return 0;
952                         *n_blocks_count = option;
953                         break;
954                 case Opt_nobh:
955                         set_opt(sbi->s_mount_opt, NOBH);
956                         break;
957                 default:
958                         printk (KERN_ERR
959                                 "EXT3-fs: Unrecognized mount option \"%s\" "
960                                 "or missing value\n", p);
961                         return 0;
962                 }
963         }
964 #ifdef CONFIG_QUOTA
965         if (!sbi->s_jquota_fmt && (sbi->s_qf_names[USRQUOTA] ||
966             sbi->s_qf_names[GRPQUOTA])) {
967                 printk(KERN_ERR
968                         "EXT3-fs: journalled quota format not specified.\n");
969                 return 0;
970         }
971 #endif
972
973         return 1;
974 }
975
976 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
977                             int read_only)
978 {
979         struct ext3_sb_info *sbi = EXT3_SB(sb);
980         int res = 0;
981
982         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
983                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
984                         "forcing read-only mode\n");
985                 res = MS_RDONLY;
986         }
987         if (read_only)
988                 return res;
989         if (!(sbi->s_mount_state & EXT3_VALID_FS))
990                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
991                         "running e2fsck is recommended\n");
992         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
993                 printk (KERN_WARNING
994                         "EXT3-fs warning: mounting fs with errors, "
995                         "running e2fsck is recommended\n");
996         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
997                  le16_to_cpu(es->s_mnt_count) >=
998                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
999                 printk (KERN_WARNING
1000                         "EXT3-fs warning: maximal mount count reached, "
1001                         "running e2fsck is recommended\n");
1002         else if (le32_to_cpu(es->s_checkinterval) &&
1003                 (le32_to_cpu(es->s_lastcheck) +
1004                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1005                 printk (KERN_WARNING
1006                         "EXT3-fs warning: checktime reached, "
1007                         "running e2fsck is recommended\n");
1008 #if 0
1009                 /* @@@ We _will_ want to clear the valid bit if we find
1010                    inconsistencies, to force a fsck at reboot.  But for
1011                    a plain journaled filesystem we can keep it set as
1012                    valid forever! :) */
1013         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1014 #endif
1015         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1016                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1017         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1018         es->s_mtime = cpu_to_le32(get_seconds());
1019         ext3_update_dynamic_rev(sb);
1020         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1021
1022         ext3_commit_super(sb, es, 1);
1023         if (test_opt(sb, DEBUG))
1024                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1025                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1026                         sb->s_blocksize,
1027                         sbi->s_groups_count,
1028                         EXT3_BLOCKS_PER_GROUP(sb),
1029                         EXT3_INODES_PER_GROUP(sb),
1030                         sbi->s_mount_opt);
1031
1032         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1033         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1034                 char b[BDEVNAME_SIZE];
1035
1036                 printk("external journal on %s\n",
1037                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1038         } else {
1039                 printk("internal journal\n");
1040         }
1041 #ifdef CONFIG_EXT3_CHECK
1042         if (test_opt (sb, CHECK)) {
1043                 ext3_check_blocks_bitmap (sb);
1044                 ext3_check_inodes_bitmap (sb);
1045         }
1046 #endif
1047         return res;
1048 }
1049
1050 /* Called at mount-time, super-block is locked */
1051 static int ext3_check_descriptors (struct super_block * sb)
1052 {
1053         struct ext3_sb_info *sbi = EXT3_SB(sb);
1054         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1055         struct ext3_group_desc * gdp = NULL;
1056         int desc_block = 0;
1057         int i;
1058
1059         ext3_debug ("Checking group descriptors");
1060
1061         for (i = 0; i < sbi->s_groups_count; i++)
1062         {
1063                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1064                         gdp = (struct ext3_group_desc *)
1065                                         sbi->s_group_desc[desc_block++]->b_data;
1066                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1067                     le32_to_cpu(gdp->bg_block_bitmap) >=
1068                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1069                 {
1070                         ext3_error (sb, "ext3_check_descriptors",
1071                                     "Block bitmap for group %d"
1072                                     " not in group (block %lu)!",
1073                                     i, (unsigned long)
1074                                         le32_to_cpu(gdp->bg_block_bitmap));
1075                         return 0;
1076                 }
1077                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1078                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1079                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1080                 {
1081                         ext3_error (sb, "ext3_check_descriptors",
1082                                     "Inode bitmap for group %d"
1083                                     " not in group (block %lu)!",
1084                                     i, (unsigned long)
1085                                         le32_to_cpu(gdp->bg_inode_bitmap));
1086                         return 0;
1087                 }
1088                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1089                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1090                     block + EXT3_BLOCKS_PER_GROUP(sb))
1091                 {
1092                         ext3_error (sb, "ext3_check_descriptors",
1093                                     "Inode table for group %d"
1094                                     " not in group (block %lu)!",
1095                                     i, (unsigned long)
1096                                         le32_to_cpu(gdp->bg_inode_table));
1097                         return 0;
1098                 }
1099                 block += EXT3_BLOCKS_PER_GROUP(sb);
1100                 gdp++;
1101         }
1102
1103         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1104         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1105         return 1;
1106 }
1107
1108
1109 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1110  * the superblock) which were deleted from all directories, but held open by
1111  * a process at the time of a crash.  We walk the list and try to delete these
1112  * inodes at recovery time (only with a read-write filesystem).
1113  *
1114  * In order to keep the orphan inode chain consistent during traversal (in
1115  * case of crash during recovery), we link each inode into the superblock
1116  * orphan list_head and handle it the same way as an inode deletion during
1117  * normal operation (which journals the operations for us).
1118  *
1119  * We only do an iget() and an iput() on each inode, which is very safe if we
1120  * accidentally point at an in-use or already deleted inode.  The worst that
1121  * can happen in this case is that we get a "bit already cleared" message from
1122  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1123  * e2fsck was run on this filesystem, and it must have already done the orphan
1124  * inode cleanup for us, so we can safely abort without any further action.
1125  */
1126 static void ext3_orphan_cleanup (struct super_block * sb,
1127                                  struct ext3_super_block * es)
1128 {
1129         unsigned int s_flags = sb->s_flags;
1130         int nr_orphans = 0, nr_truncates = 0;
1131 #ifdef CONFIG_QUOTA
1132         int i;
1133 #endif
1134         if (!es->s_last_orphan) {
1135                 jbd_debug(4, "no orphan inodes to clean up\n");
1136                 return;
1137         }
1138
1139         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1140                 if (es->s_last_orphan)
1141                         jbd_debug(1, "Errors on filesystem, "
1142                                   "clearing orphan list.\n");
1143                 es->s_last_orphan = 0;
1144                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1145                 return;
1146         }
1147
1148         if (s_flags & MS_RDONLY) {
1149                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1150                        sb->s_id);
1151                 sb->s_flags &= ~MS_RDONLY;
1152         }
1153 #ifdef CONFIG_QUOTA
1154         /* Needed for iput() to work correctly and not trash data */
1155         sb->s_flags |= MS_ACTIVE;
1156         /* Turn on quotas so that they are updated correctly */
1157         for (i = 0; i < MAXQUOTAS; i++) {
1158                 if (EXT3_SB(sb)->s_qf_names[i]) {
1159                         int ret = ext3_quota_on_mount(sb, i);
1160                         if (ret < 0)
1161                                 printk(KERN_ERR
1162                                         "EXT3-fs: Cannot turn on journalled "
1163                                         "quota: error %d\n", ret);
1164                 }
1165         }
1166 #endif
1167
1168         while (es->s_last_orphan) {
1169                 struct inode *inode;
1170
1171                 if (!(inode =
1172                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1173                         es->s_last_orphan = 0;
1174                         break;
1175                 }
1176
1177                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1178                 DQUOT_INIT(inode);
1179                 if (inode->i_nlink) {
1180                         printk(KERN_DEBUG
1181                                 "%s: truncating inode %ld to %Ld bytes\n",
1182                                 __FUNCTION__, inode->i_ino, inode->i_size);
1183                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1184                                   inode->i_ino, inode->i_size);
1185                         ext3_truncate(inode);
1186                         nr_truncates++;
1187                 } else {
1188                         printk(KERN_DEBUG
1189                                 "%s: deleting unreferenced inode %ld\n",
1190                                 __FUNCTION__, inode->i_ino);
1191                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1192                                   inode->i_ino);
1193                         nr_orphans++;
1194                 }
1195                 iput(inode);  /* The delete magic happens here! */
1196         }
1197
1198 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1199
1200         if (nr_orphans)
1201                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1202                        sb->s_id, PLURAL(nr_orphans));
1203         if (nr_truncates)
1204                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1205                        sb->s_id, PLURAL(nr_truncates));
1206 #ifdef CONFIG_QUOTA
1207         /* Turn quotas off */
1208         for (i = 0; i < MAXQUOTAS; i++) {
1209                 if (sb_dqopt(sb)->files[i])
1210                         vfs_quota_off(sb, i);
1211         }
1212 #endif
1213         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1214 }
1215
1216 #define log2(n) ffz(~(n))
1217
1218 /*
1219  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1220  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1221  * We need to be 1 filesystem block less than the 2^32 sector limit.
1222  */
1223 static loff_t ext3_max_size(int bits)
1224 {
1225         loff_t res = EXT3_NDIR_BLOCKS;
1226         /* This constant is calculated to be the largest file size for a
1227          * dense, 4k-blocksize file such that the total number of
1228          * sectors in the file, including data and all indirect blocks,
1229          * does not exceed 2^32. */
1230         const loff_t upper_limit = 0x1ff7fffd000LL;
1231
1232         res += 1LL << (bits-2);
1233         res += 1LL << (2*(bits-2));
1234         res += 1LL << (3*(bits-2));
1235         res <<= bits;
1236         if (res > upper_limit)
1237                 res = upper_limit;
1238         return res;
1239 }
1240
1241 static unsigned long descriptor_loc(struct super_block *sb,
1242                                     unsigned long logic_sb_block,
1243                                     int nr)
1244 {
1245         struct ext3_sb_info *sbi = EXT3_SB(sb);
1246         unsigned long bg, first_data_block, first_meta_bg;
1247         int has_super = 0;
1248
1249         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1250         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1251
1252         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1253             nr < first_meta_bg)
1254                 return (logic_sb_block + nr + 1);
1255         bg = sbi->s_desc_per_block * nr;
1256         if (ext3_bg_has_super(sb, bg))
1257                 has_super = 1;
1258         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1259 }
1260
1261
1262 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1263 {
1264         struct buffer_head * bh;
1265         struct ext3_super_block *es = NULL;
1266         struct ext3_sb_info *sbi;
1267         unsigned long block;
1268         unsigned long sb_block = get_sb_block(&data);
1269         unsigned long logic_sb_block;
1270         unsigned long offset = 0;
1271         unsigned long journal_inum = 0;
1272         unsigned long def_mount_opts;
1273         struct inode *root;
1274         int blocksize;
1275         int hblock;
1276         int db_count;
1277         int i;
1278         int needs_recovery;
1279         __le32 features;
1280
1281         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1282         if (!sbi)
1283                 return -ENOMEM;
1284         sb->s_fs_info = sbi;
1285         memset(sbi, 0, sizeof(*sbi));
1286         sbi->s_mount_opt = 0;
1287         sbi->s_resuid = EXT3_DEF_RESUID;
1288         sbi->s_resgid = EXT3_DEF_RESGID;
1289
1290         unlock_kernel();
1291
1292         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1293         if (!blocksize) {
1294                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1295                 goto out_fail;
1296         }
1297
1298         /*
1299          * The ext3 superblock will not be buffer aligned for other than 1kB
1300          * block sizes.  We need to calculate the offset from buffer start.
1301          */
1302         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1303                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1304                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1305         } else {
1306                 logic_sb_block = sb_block;
1307         }
1308
1309         if (!(bh = sb_bread(sb, logic_sb_block))) {
1310                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1311                 goto out_fail;
1312         }
1313         /*
1314          * Note: s_es must be initialized as soon as possible because
1315          *       some ext3 macro-instructions depend on its value
1316          */
1317         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1318         sbi->s_es = es;
1319         sb->s_magic = le16_to_cpu(es->s_magic);
1320         if (sb->s_magic != EXT3_SUPER_MAGIC)
1321                 goto cantfind_ext3;
1322
1323         /* Set defaults before we parse the mount options */
1324         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1325         if (def_mount_opts & EXT3_DEFM_DEBUG)
1326                 set_opt(sbi->s_mount_opt, DEBUG);
1327         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1328                 set_opt(sbi->s_mount_opt, GRPID);
1329         if (def_mount_opts & EXT3_DEFM_UID16)
1330                 set_opt(sbi->s_mount_opt, NO_UID32);
1331         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1332                 set_opt(sbi->s_mount_opt, XATTR_USER);
1333         if (def_mount_opts & EXT3_DEFM_ACL)
1334                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1335         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1336                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1337         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1338                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1339         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1340                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1341
1342         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1343                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1344         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1345                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1346
1347         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1348         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1349
1350         set_opt(sbi->s_mount_opt, RESERVATION);
1351
1352         if (!parse_options ((char *) data, sb, &journal_inum, NULL, 0))
1353                 goto failed_mount;
1354
1355         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1356                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1357
1358         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1359             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1360              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1361              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1362                 printk(KERN_WARNING 
1363                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1364                        "running e2fsck is recommended\n");
1365         /*
1366          * Check feature flags regardless of the revision level, since we
1367          * previously didn't change the revision level when setting the flags,
1368          * so there is a chance incompat flags are set on a rev 0 filesystem.
1369          */
1370         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1371         if (features) {
1372                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1373                        "unsupported optional features (%x).\n",
1374                        sb->s_id, le32_to_cpu(features));
1375                 goto failed_mount;
1376         }
1377         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1378         if (!(sb->s_flags & MS_RDONLY) && features) {
1379                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1380                        "unsupported optional features (%x).\n",
1381                        sb->s_id, le32_to_cpu(features));
1382                 goto failed_mount;
1383         }
1384         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1385
1386         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1387             blocksize > EXT3_MAX_BLOCK_SIZE) {
1388                 printk(KERN_ERR 
1389                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1390                        blocksize, sb->s_id);
1391                 goto failed_mount;
1392         }
1393
1394         hblock = bdev_hardsect_size(sb->s_bdev);
1395         if (sb->s_blocksize != blocksize) {
1396                 /*
1397                  * Make sure the blocksize for the filesystem is larger
1398                  * than the hardware sectorsize for the machine.
1399                  */
1400                 if (blocksize < hblock) {
1401                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1402                                "device blocksize %d.\n", blocksize, hblock);
1403                         goto failed_mount;
1404                 }
1405
1406                 brelse (bh);
1407                 sb_set_blocksize(sb, blocksize);
1408                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1409                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1410                 bh = sb_bread(sb, logic_sb_block);
1411                 if (!bh) {
1412                         printk(KERN_ERR 
1413                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1414                         goto failed_mount;
1415                 }
1416                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1417                 sbi->s_es = es;
1418                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1419                         printk (KERN_ERR 
1420                                 "EXT3-fs: Magic mismatch, very weird !\n");
1421                         goto failed_mount;
1422                 }
1423         }
1424
1425         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1426
1427         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1428                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1429                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1430         } else {
1431                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1432                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1433                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1434                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1435                     (sbi->s_inode_size > blocksize)) {
1436                         printk (KERN_ERR
1437                                 "EXT3-fs: unsupported inode size: %d\n",
1438                                 sbi->s_inode_size);
1439                         goto failed_mount;
1440                 }
1441         }
1442         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1443                                    le32_to_cpu(es->s_log_frag_size);
1444         if (blocksize != sbi->s_frag_size) {
1445                 printk(KERN_ERR
1446                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1447                        sbi->s_frag_size, blocksize);
1448                 goto failed_mount;
1449         }
1450         sbi->s_frags_per_block = 1;
1451         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1452         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1453         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1454         if (EXT3_INODE_SIZE(sb) == 0)
1455                 goto cantfind_ext3;
1456         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1457         if (sbi->s_inodes_per_block == 0)
1458                 goto cantfind_ext3;
1459         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1460                                         sbi->s_inodes_per_block;
1461         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1462         sbi->s_sbh = bh;
1463         sbi->s_mount_state = le16_to_cpu(es->s_state);
1464         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1465         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1466         for (i=0; i < 4; i++)
1467                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1468         sbi->s_def_hash_version = es->s_def_hash_version;
1469
1470         if (sbi->s_blocks_per_group > blocksize * 8) {
1471                 printk (KERN_ERR
1472                         "EXT3-fs: #blocks per group too big: %lu\n",
1473                         sbi->s_blocks_per_group);
1474                 goto failed_mount;
1475         }
1476         if (sbi->s_frags_per_group > blocksize * 8) {
1477                 printk (KERN_ERR
1478                         "EXT3-fs: #fragments per group too big: %lu\n",
1479                         sbi->s_frags_per_group);
1480                 goto failed_mount;
1481         }
1482         if (sbi->s_inodes_per_group > blocksize * 8) {
1483                 printk (KERN_ERR
1484                         "EXT3-fs: #inodes per group too big: %lu\n",
1485                         sbi->s_inodes_per_group);
1486                 goto failed_mount;
1487         }
1488
1489         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1490                 goto cantfind_ext3;
1491         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1492                                le32_to_cpu(es->s_first_data_block) +
1493                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1494                               EXT3_BLOCKS_PER_GROUP(sb);
1495         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1496                    EXT3_DESC_PER_BLOCK(sb);
1497         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1498                                     GFP_KERNEL);
1499         if (sbi->s_group_desc == NULL) {
1500                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1501                 goto failed_mount;
1502         }
1503
1504         percpu_counter_init(&sbi->s_freeblocks_counter);
1505         percpu_counter_init(&sbi->s_freeinodes_counter);
1506         percpu_counter_init(&sbi->s_dirs_counter);
1507         bgl_lock_init(&sbi->s_blockgroup_lock);
1508
1509         for (i = 0; i < db_count; i++) {
1510                 block = descriptor_loc(sb, logic_sb_block, i);
1511                 sbi->s_group_desc[i] = sb_bread(sb, block);
1512                 if (!sbi->s_group_desc[i]) {
1513                         printk (KERN_ERR "EXT3-fs: "
1514                                 "can't read group descriptor %d\n", i);
1515                         db_count = i;
1516                         goto failed_mount2;
1517                 }
1518         }
1519         if (!ext3_check_descriptors (sb)) {
1520                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1521                 goto failed_mount2;
1522         }
1523         sbi->s_gdb_count = db_count;
1524         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1525         spin_lock_init(&sbi->s_next_gen_lock);
1526         /* per fileystem reservation list head & lock */
1527         spin_lock_init(&sbi->s_rsv_window_lock);
1528         sbi->s_rsv_window_root = RB_ROOT;
1529         /* Add a single, static dummy reservation to the start of the
1530          * reservation window list --- it gives us a placeholder for
1531          * append-at-start-of-list which makes the allocation logic
1532          * _much_ simpler. */
1533         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1534         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1535         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1536         sbi->s_rsv_window_head.rsv_goal_size = 0;
1537         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1538
1539         /*
1540          * set up enough so that it can read an inode
1541          */
1542         sb->s_op = &ext3_sops;
1543         sb->s_export_op = &ext3_export_ops;
1544         sb->s_xattr = ext3_xattr_handlers;
1545 #ifdef CONFIG_QUOTA
1546         sb->s_qcop = &ext3_qctl_operations;
1547         sb->dq_op = &ext3_quota_operations;
1548 #endif
1549         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1550
1551         sb->s_root = NULL;
1552
1553         needs_recovery = (es->s_last_orphan != 0 ||
1554                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1555                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1556
1557         /*
1558          * The first inode we look at is the journal inode.  Don't try
1559          * root first: it may be modified in the journal!
1560          */
1561         if (!test_opt(sb, NOLOAD) &&
1562             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1563                 if (ext3_load_journal(sb, es))
1564                         goto failed_mount2;
1565         } else if (journal_inum) {
1566                 if (ext3_create_journal(sb, es, journal_inum))
1567                         goto failed_mount2;
1568         } else {
1569                 if (!silent)
1570                         printk (KERN_ERR
1571                                 "ext3: No journal on filesystem on %s\n",
1572                                 sb->s_id);
1573                 goto failed_mount2;
1574         }
1575
1576         /* We have now updated the journal if required, so we can
1577          * validate the data journaling mode. */
1578         switch (test_opt(sb, DATA_FLAGS)) {
1579         case 0:
1580                 /* No mode set, assume a default based on the journal
1581                    capabilities: ORDERED_DATA if the journal can
1582                    cope, else JOURNAL_DATA */
1583                 if (journal_check_available_features
1584                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1585                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1586                 else
1587                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1588                 break;
1589
1590         case EXT3_MOUNT_ORDERED_DATA:
1591         case EXT3_MOUNT_WRITEBACK_DATA:
1592                 if (!journal_check_available_features
1593                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1594                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1595                                "requested data journaling mode\n");
1596                         goto failed_mount3;
1597                 }
1598         default:
1599                 break;
1600         }
1601
1602         if (test_opt(sb, NOBH)) {
1603                 if (sb->s_blocksize_bits != PAGE_CACHE_SHIFT) {
1604                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option "
1605                                 "since filesystem blocksize doesn't match "
1606                                 "pagesize\n");
1607                         clear_opt(sbi->s_mount_opt, NOBH);
1608                 }
1609                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1610                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1611                                 "its supported only with writeback mode\n");
1612                         clear_opt(sbi->s_mount_opt, NOBH);
1613                 }
1614         }
1615         /*
1616          * The journal_load will have done any necessary log recovery,
1617          * so we can safely mount the rest of the filesystem now.
1618          */
1619
1620         root = iget(sb, EXT3_ROOT_INO);
1621         sb->s_root = d_alloc_root(root);
1622         if (!sb->s_root) {
1623                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1624                 iput(root);
1625                 goto failed_mount3;
1626         }
1627         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1628                 dput(sb->s_root);
1629                 sb->s_root = NULL;
1630                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1631                 goto failed_mount3;
1632         }
1633
1634         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1635         /*
1636          * akpm: core read_super() calls in here with the superblock locked.
1637          * That deadlocks, because orphan cleanup needs to lock the superblock
1638          * in numerous places.  Here we just pop the lock - it's relatively
1639          * harmless, because we are now ready to accept write_super() requests,
1640          * and aviro says that's the only reason for hanging onto the
1641          * superblock lock.
1642          */
1643         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1644         ext3_orphan_cleanup(sb, es);
1645         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1646         if (needs_recovery)
1647                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1648         ext3_mark_recovery_complete(sb, es);
1649         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1650                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1651                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1652                 "writeback");
1653
1654         percpu_counter_mod(&sbi->s_freeblocks_counter,
1655                 ext3_count_free_blocks(sb));
1656         percpu_counter_mod(&sbi->s_freeinodes_counter,
1657                 ext3_count_free_inodes(sb));
1658         percpu_counter_mod(&sbi->s_dirs_counter,
1659                 ext3_count_dirs(sb));
1660
1661         lock_kernel();
1662         return 0;
1663
1664 cantfind_ext3:
1665         if (!silent)
1666                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1667                        sb->s_id);
1668         goto failed_mount;
1669
1670 failed_mount3:
1671         journal_destroy(sbi->s_journal);
1672 failed_mount2:
1673         for (i = 0; i < db_count; i++)
1674                 brelse(sbi->s_group_desc[i]);
1675         kfree(sbi->s_group_desc);
1676 failed_mount:
1677 #ifdef CONFIG_QUOTA
1678         for (i = 0; i < MAXQUOTAS; i++)
1679                 kfree(sbi->s_qf_names[i]);
1680 #endif
1681         ext3_blkdev_remove(sbi);
1682         brelse(bh);
1683 out_fail:
1684         sb->s_fs_info = NULL;
1685         kfree(sbi);
1686         lock_kernel();
1687         return -EINVAL;
1688 }
1689
1690 /*
1691  * Setup any per-fs journal parameters now.  We'll do this both on
1692  * initial mount, once the journal has been initialised but before we've
1693  * done any recovery; and again on any subsequent remount. 
1694  */
1695 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1696 {
1697         struct ext3_sb_info *sbi = EXT3_SB(sb);
1698
1699         if (sbi->s_commit_interval)
1700                 journal->j_commit_interval = sbi->s_commit_interval;
1701         /* We could also set up an ext3-specific default for the commit
1702          * interval here, but for now we'll just fall back to the jbd
1703          * default. */
1704
1705         spin_lock(&journal->j_state_lock);
1706         if (test_opt(sb, BARRIER))
1707                 journal->j_flags |= JFS_BARRIER;
1708         else
1709                 journal->j_flags &= ~JFS_BARRIER;
1710         spin_unlock(&journal->j_state_lock);
1711 }
1712
1713 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1714 {
1715         struct inode *journal_inode;
1716         journal_t *journal;
1717
1718         /* First, test for the existence of a valid inode on disk.  Bad
1719          * things happen if we iget() an unused inode, as the subsequent
1720          * iput() will try to delete it. */
1721
1722         journal_inode = iget(sb, journal_inum);
1723         if (!journal_inode) {
1724                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1725                 return NULL;
1726         }
1727         if (!journal_inode->i_nlink) {
1728                 make_bad_inode(journal_inode);
1729                 iput(journal_inode);
1730                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1731                 return NULL;
1732         }
1733
1734         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1735                   journal_inode, journal_inode->i_size);
1736         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1737                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1738                 iput(journal_inode);
1739                 return NULL;
1740         }
1741
1742         journal = journal_init_inode(journal_inode);
1743         if (!journal) {
1744                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1745                 iput(journal_inode);
1746                 return NULL;
1747         }
1748         journal->j_private = sb;
1749         ext3_init_journal_params(sb, journal);
1750         return journal;
1751 }
1752
1753 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1754                                        dev_t j_dev)
1755 {
1756         struct buffer_head * bh;
1757         journal_t *journal;
1758         int start;
1759         int len;
1760         int hblock, blocksize;
1761         unsigned long sb_block;
1762         unsigned long offset;
1763         struct ext3_super_block * es;
1764         struct block_device *bdev;
1765
1766         bdev = ext3_blkdev_get(j_dev);
1767         if (bdev == NULL)
1768                 return NULL;
1769
1770         if (bd_claim(bdev, sb)) {
1771                 printk(KERN_ERR
1772                         "EXT3: failed to claim external journal device.\n");
1773                 blkdev_put(bdev);
1774                 return NULL;
1775         }
1776
1777         blocksize = sb->s_blocksize;
1778         hblock = bdev_hardsect_size(bdev);
1779         if (blocksize < hblock) {
1780                 printk(KERN_ERR
1781                         "EXT3-fs: blocksize too small for journal device.\n");
1782                 goto out_bdev;
1783         }
1784
1785         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1786         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1787         set_blocksize(bdev, blocksize);
1788         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1789                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1790                        "external journal\n");
1791                 goto out_bdev;
1792         }
1793
1794         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1795         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1796             !(le32_to_cpu(es->s_feature_incompat) &
1797               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1798                 printk(KERN_ERR "EXT3-fs: external journal has "
1799                                         "bad superblock\n");
1800                 brelse(bh);
1801                 goto out_bdev;
1802         }
1803
1804         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1805                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1806                 brelse(bh);
1807                 goto out_bdev;
1808         }
1809
1810         len = le32_to_cpu(es->s_blocks_count);
1811         start = sb_block + 1;
1812         brelse(bh);     /* we're done with the superblock */
1813
1814         journal = journal_init_dev(bdev, sb->s_bdev,
1815                                         start, len, blocksize);
1816         if (!journal) {
1817                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1818                 goto out_bdev;
1819         }
1820         journal->j_private = sb;
1821         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1822         wait_on_buffer(journal->j_sb_buffer);
1823         if (!buffer_uptodate(journal->j_sb_buffer)) {
1824                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1825                 goto out_journal;
1826         }
1827         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1828                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1829                                         "user (unsupported) - %d\n",
1830                         be32_to_cpu(journal->j_superblock->s_nr_users));
1831                 goto out_journal;
1832         }
1833         EXT3_SB(sb)->journal_bdev = bdev;
1834         ext3_init_journal_params(sb, journal);
1835         return journal;
1836 out_journal:
1837         journal_destroy(journal);
1838 out_bdev:
1839         ext3_blkdev_put(bdev);
1840         return NULL;
1841 }
1842
1843 static int ext3_load_journal(struct super_block * sb,
1844                              struct ext3_super_block * es)
1845 {
1846         journal_t *journal;
1847         int journal_inum = le32_to_cpu(es->s_journal_inum);
1848         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1849         int err = 0;
1850         int really_read_only;
1851
1852         really_read_only = bdev_read_only(sb->s_bdev);
1853
1854         /*
1855          * Are we loading a blank journal or performing recovery after a
1856          * crash?  For recovery, we need to check in advance whether we
1857          * can get read-write access to the device.
1858          */
1859
1860         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1861                 if (sb->s_flags & MS_RDONLY) {
1862                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1863                                         "required on readonly filesystem.\n");
1864                         if (really_read_only) {
1865                                 printk(KERN_ERR "EXT3-fs: write access "
1866                                         "unavailable, cannot proceed.\n");
1867                                 return -EROFS;
1868                         }
1869                         printk (KERN_INFO "EXT3-fs: write access will "
1870                                         "be enabled during recovery.\n");
1871                 }
1872         }
1873
1874         if (journal_inum && journal_dev) {
1875                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1876                        "and inode journals!\n");
1877                 return -EINVAL;
1878         }
1879
1880         if (journal_inum) {
1881                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1882                         return -EINVAL;
1883         } else {
1884                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1885                         return -EINVAL;
1886         }
1887
1888         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1889                 err = journal_update_format(journal);
1890                 if (err)  {
1891                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1892                         journal_destroy(journal);
1893                         return err;
1894                 }
1895         }
1896
1897         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1898                 err = journal_wipe(journal, !really_read_only);
1899         if (!err)
1900                 err = journal_load(journal);
1901
1902         if (err) {
1903                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1904                 journal_destroy(journal);
1905                 return err;
1906         }
1907
1908         EXT3_SB(sb)->s_journal = journal;
1909         ext3_clear_journal_err(sb, es);
1910         return 0;
1911 }
1912
1913 static int ext3_create_journal(struct super_block * sb,
1914                                struct ext3_super_block * es,
1915                                int journal_inum)
1916 {
1917         journal_t *journal;
1918
1919         if (sb->s_flags & MS_RDONLY) {
1920                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1921                                 "create journal.\n");
1922                 return -EROFS;
1923         }
1924
1925         if (!(journal = ext3_get_journal(sb, journal_inum)))
1926                 return -EINVAL;
1927
1928         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1929                journal_inum);
1930
1931         if (journal_create(journal)) {
1932                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1933                 journal_destroy(journal);
1934                 return -EIO;
1935         }
1936
1937         EXT3_SB(sb)->s_journal = journal;
1938
1939         ext3_update_dynamic_rev(sb);
1940         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1941         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1942
1943         es->s_journal_inum = cpu_to_le32(journal_inum);
1944         sb->s_dirt = 1;
1945
1946         /* Make sure we flush the recovery flag to disk. */
1947         ext3_commit_super(sb, es, 1);
1948
1949         return 0;
1950 }
1951
1952 static void ext3_commit_super (struct super_block * sb,
1953                                struct ext3_super_block * es,
1954                                int sync)
1955 {
1956         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1957
1958         if (!sbh)
1959                 return;
1960         es->s_wtime = cpu_to_le32(get_seconds());
1961         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1962         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1963         BUFFER_TRACE(sbh, "marking dirty");
1964         mark_buffer_dirty(sbh);
1965         if (sync)
1966                 sync_dirty_buffer(sbh);
1967 }
1968
1969
1970 /*
1971  * Have we just finished recovery?  If so, and if we are mounting (or
1972  * remounting) the filesystem readonly, then we will end up with a
1973  * consistent fs on disk.  Record that fact.
1974  */
1975 static void ext3_mark_recovery_complete(struct super_block * sb,
1976                                         struct ext3_super_block * es)
1977 {
1978         journal_t *journal = EXT3_SB(sb)->s_journal;
1979
1980         journal_lock_updates(journal);
1981         journal_flush(journal);
1982         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
1983             sb->s_flags & MS_RDONLY) {
1984                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1985                 sb->s_dirt = 0;
1986                 ext3_commit_super(sb, es, 1);
1987         }
1988         journal_unlock_updates(journal);
1989 }
1990
1991 /*
1992  * If we are mounting (or read-write remounting) a filesystem whose journal
1993  * has recorded an error from a previous lifetime, move that error to the
1994  * main filesystem now.
1995  */
1996 static void ext3_clear_journal_err(struct super_block * sb,
1997                                    struct ext3_super_block * es)
1998 {
1999         journal_t *journal;
2000         int j_errno;
2001         const char *errstr;
2002
2003         journal = EXT3_SB(sb)->s_journal;
2004
2005         /*
2006          * Now check for any error status which may have been recorded in the
2007          * journal by a prior ext3_error() or ext3_abort()
2008          */
2009
2010         j_errno = journal_errno(journal);
2011         if (j_errno) {
2012                 char nbuf[16];
2013
2014                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2015                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2016                              "from previous mount: %s", errstr);
2017                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2018                              "filesystem check.");
2019
2020                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2021                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2022                 ext3_commit_super (sb, es, 1);
2023
2024                 journal_clear_err(journal);
2025         }
2026 }
2027
2028 /*
2029  * Force the running and committing transactions to commit,
2030  * and wait on the commit.
2031  */
2032 int ext3_force_commit(struct super_block *sb)
2033 {
2034         journal_t *journal;
2035         int ret;
2036
2037         if (sb->s_flags & MS_RDONLY)
2038                 return 0;
2039
2040         journal = EXT3_SB(sb)->s_journal;
2041         sb->s_dirt = 0;
2042         ret = ext3_journal_force_commit(journal);
2043         return ret;
2044 }
2045
2046 /*
2047  * Ext3 always journals updates to the superblock itself, so we don't
2048  * have to propagate any other updates to the superblock on disk at this
2049  * point.  Just start an async writeback to get the buffers on their way
2050  * to the disk.
2051  *
2052  * This implicitly triggers the writebehind on sync().
2053  */
2054
2055 static void ext3_write_super (struct super_block * sb)
2056 {
2057         if (down_trylock(&sb->s_lock) == 0)
2058                 BUG();
2059         sb->s_dirt = 0;
2060 }
2061
2062 static int ext3_sync_fs(struct super_block *sb, int wait)
2063 {
2064         tid_t target;
2065
2066         sb->s_dirt = 0;
2067         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2068                 if (wait)
2069                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2070         }
2071         return 0;
2072 }
2073
2074 /*
2075  * LVM calls this function before a (read-only) snapshot is created.  This
2076  * gives us a chance to flush the journal completely and mark the fs clean.
2077  */
2078 static void ext3_write_super_lockfs(struct super_block *sb)
2079 {
2080         sb->s_dirt = 0;
2081
2082         if (!(sb->s_flags & MS_RDONLY)) {
2083                 journal_t *journal = EXT3_SB(sb)->s_journal;
2084
2085                 /* Now we set up the journal barrier. */
2086                 journal_lock_updates(journal);
2087                 journal_flush(journal);
2088
2089                 /* Journal blocked and flushed, clear needs_recovery flag. */
2090                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2091                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2092         }
2093 }
2094
2095 /*
2096  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2097  * flag here, even though the filesystem is not technically dirty yet.
2098  */
2099 static void ext3_unlockfs(struct super_block *sb)
2100 {
2101         if (!(sb->s_flags & MS_RDONLY)) {
2102                 lock_super(sb);
2103                 /* Reser the needs_recovery flag before the fs is unlocked. */
2104                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2105                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2106                 unlock_super(sb);
2107                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2108         }
2109 }
2110
2111 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2112 {
2113         struct ext3_super_block * es;
2114         struct ext3_sb_info *sbi = EXT3_SB(sb);
2115         unsigned long n_blocks_count = 0;
2116         unsigned long old_sb_flags;
2117         struct ext3_mount_options old_opts;
2118         int err;
2119 #ifdef CONFIG_QUOTA
2120         int i;
2121 #endif
2122
2123         /* Store the original options */
2124         old_sb_flags = sb->s_flags;
2125         old_opts.s_mount_opt = sbi->s_mount_opt;
2126         old_opts.s_resuid = sbi->s_resuid;
2127         old_opts.s_resgid = sbi->s_resgid;
2128         old_opts.s_commit_interval = sbi->s_commit_interval;
2129 #ifdef CONFIG_QUOTA
2130         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2131         for (i = 0; i < MAXQUOTAS; i++)
2132                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2133 #endif
2134
2135         /*
2136          * Allow the "check" option to be passed as a remount option.
2137          */
2138         if (!parse_options(data, sb, NULL, &n_blocks_count, 1)) {
2139                 err = -EINVAL;
2140                 goto restore_opts;
2141         }
2142
2143         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2144                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2145
2146         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2147                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2148
2149         es = sbi->s_es;
2150
2151         ext3_init_journal_params(sb, sbi->s_journal);
2152
2153         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2154                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2155                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2156                         err = -EROFS;
2157                         goto restore_opts;
2158                 }
2159
2160                 if (*flags & MS_RDONLY) {
2161                         /*
2162                          * First of all, the unconditional stuff we have to do
2163                          * to disable replay of the journal when we next remount
2164                          */
2165                         sb->s_flags |= MS_RDONLY;
2166
2167                         /*
2168                          * OK, test if we are remounting a valid rw partition
2169                          * readonly, and if so set the rdonly flag and then
2170                          * mark the partition as valid again.
2171                          */
2172                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2173                             (sbi->s_mount_state & EXT3_VALID_FS))
2174                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2175
2176                         ext3_mark_recovery_complete(sb, es);
2177                 } else {
2178                         __le32 ret;
2179                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2180                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2181                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2182                                        "remount RDWR because of unsupported "
2183                                        "optional features (%x).\n",
2184                                        sb->s_id, le32_to_cpu(ret));
2185                                 err = -EROFS;
2186                                 goto restore_opts;
2187                         }
2188                         /*
2189                          * Mounting a RDONLY partition read-write, so reread
2190                          * and store the current valid flag.  (It may have
2191                          * been changed by e2fsck since we originally mounted
2192                          * the partition.)
2193                          */
2194                         ext3_clear_journal_err(sb, es);
2195                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2196                         if ((ret = ext3_group_extend(sb, es, n_blocks_count))) {
2197                                 err = ret;
2198                                 goto restore_opts;
2199                         }
2200                         if (!ext3_setup_super (sb, es, 0))
2201                                 sb->s_flags &= ~MS_RDONLY;
2202                 }
2203         }
2204 #ifdef CONFIG_QUOTA
2205         /* Release old quota file names */
2206         for (i = 0; i < MAXQUOTAS; i++)
2207                 if (old_opts.s_qf_names[i] &&
2208                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2209                         kfree(old_opts.s_qf_names[i]);
2210 #endif
2211         return 0;
2212 restore_opts:
2213         sb->s_flags = old_sb_flags;
2214         sbi->s_mount_opt = old_opts.s_mount_opt;
2215         sbi->s_resuid = old_opts.s_resuid;
2216         sbi->s_resgid = old_opts.s_resgid;
2217         sbi->s_commit_interval = old_opts.s_commit_interval;
2218 #ifdef CONFIG_QUOTA
2219         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2220         for (i = 0; i < MAXQUOTAS; i++) {
2221                 if (sbi->s_qf_names[i] &&
2222                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2223                         kfree(sbi->s_qf_names[i]);
2224                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2225         }
2226 #endif
2227         return err;
2228 }
2229
2230 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2231 {
2232         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2233         unsigned long overhead;
2234         int i;
2235
2236         if (test_opt (sb, MINIX_DF))
2237                 overhead = 0;
2238         else {
2239                 unsigned long ngroups;
2240                 ngroups = EXT3_SB(sb)->s_groups_count;
2241                 smp_rmb();
2242
2243                 /*
2244                  * Compute the overhead (FS structures)
2245                  */
2246
2247                 /*
2248                  * All of the blocks before first_data_block are
2249                  * overhead
2250                  */
2251                 overhead = le32_to_cpu(es->s_first_data_block);
2252
2253                 /*
2254                  * Add the overhead attributed to the superblock and
2255                  * block group descriptors.  If the sparse superblocks
2256                  * feature is turned on, then not all groups have this.
2257                  */
2258                 for (i = 0; i < ngroups; i++) {
2259                         overhead += ext3_bg_has_super(sb, i) +
2260                                 ext3_bg_num_gdb(sb, i);
2261                         cond_resched();
2262                 }
2263
2264                 /*
2265                  * Every block group has an inode bitmap, a block
2266                  * bitmap, and an inode table.
2267                  */
2268                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2269         }
2270
2271         buf->f_type = EXT3_SUPER_MAGIC;
2272         buf->f_bsize = sb->s_blocksize;
2273         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2274         buf->f_bfree = ext3_count_free_blocks (sb);
2275         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2276         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2277                 buf->f_bavail = 0;
2278         buf->f_files = le32_to_cpu(es->s_inodes_count);
2279         buf->f_ffree = ext3_count_free_inodes (sb);
2280         buf->f_namelen = EXT3_NAME_LEN;
2281         return 0;
2282 }
2283
2284 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2285  * is locked for write. Otherwise the are possible deadlocks:
2286  * Process 1                         Process 2
2287  * ext3_create()                     quota_sync()
2288  *   journal_start()                   write_dquot()
2289  *   DQUOT_INIT()                        down(dqio_sem)
2290  *     down(dqio_sem)                    journal_start()
2291  *
2292  */
2293
2294 #ifdef CONFIG_QUOTA
2295
2296 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2297 {
2298         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2299 }
2300
2301 static int ext3_dquot_initialize(struct inode *inode, int type)
2302 {
2303         handle_t *handle;
2304         int ret, err;
2305
2306         /* We may create quota structure so we need to reserve enough blocks */
2307         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2308         if (IS_ERR(handle))
2309                 return PTR_ERR(handle);
2310         ret = dquot_initialize(inode, type);
2311         err = ext3_journal_stop(handle);
2312         if (!ret)
2313                 ret = err;
2314         return ret;
2315 }
2316
2317 static int ext3_dquot_drop(struct inode *inode)
2318 {
2319         handle_t *handle;
2320         int ret, err;
2321
2322         /* We may delete quota structure so we need to reserve enough blocks */
2323         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2324         if (IS_ERR(handle))
2325                 return PTR_ERR(handle);
2326         ret = dquot_drop(inode);
2327         err = ext3_journal_stop(handle);
2328         if (!ret)
2329                 ret = err;
2330         return ret;
2331 }
2332
2333 static int ext3_write_dquot(struct dquot *dquot)
2334 {
2335         int ret, err;
2336         handle_t *handle;
2337         struct inode *inode;
2338
2339         inode = dquot_to_inode(dquot);
2340         handle = ext3_journal_start(inode,
2341                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2342         if (IS_ERR(handle))
2343                 return PTR_ERR(handle);
2344         ret = dquot_commit(dquot);
2345         err = ext3_journal_stop(handle);
2346         if (!ret)
2347                 ret = err;
2348         return ret;
2349 }
2350
2351 static int ext3_acquire_dquot(struct dquot *dquot)
2352 {
2353         int ret, err;
2354         handle_t *handle;
2355
2356         handle = ext3_journal_start(dquot_to_inode(dquot),
2357                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2358         if (IS_ERR(handle))
2359                 return PTR_ERR(handle);
2360         ret = dquot_acquire(dquot);
2361         err = ext3_journal_stop(handle);
2362         if (!ret)
2363                 ret = err;
2364         return ret;
2365 }
2366
2367 static int ext3_release_dquot(struct dquot *dquot)
2368 {
2369         int ret, err;
2370         handle_t *handle;
2371
2372         handle = ext3_journal_start(dquot_to_inode(dquot),
2373                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2374         if (IS_ERR(handle))
2375                 return PTR_ERR(handle);
2376         ret = dquot_release(dquot);
2377         err = ext3_journal_stop(handle);
2378         if (!ret)
2379                 ret = err;
2380         return ret;
2381 }
2382
2383 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2384 {
2385         /* Are we journalling quotas? */
2386         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2387             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2388                 dquot_mark_dquot_dirty(dquot);
2389                 return ext3_write_dquot(dquot);
2390         } else {
2391                 return dquot_mark_dquot_dirty(dquot);
2392         }
2393 }
2394
2395 static int ext3_write_info(struct super_block *sb, int type)
2396 {
2397         int ret, err;
2398         handle_t *handle;
2399
2400         /* Data block + inode block */
2401         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2402         if (IS_ERR(handle))
2403                 return PTR_ERR(handle);
2404         ret = dquot_commit_info(sb, type);
2405         err = ext3_journal_stop(handle);
2406         if (!ret)
2407                 ret = err;
2408         return ret;
2409 }
2410
2411 /*
2412  * Turn on quotas during mount time - we need to find
2413  * the quota file and such...
2414  */
2415 static int ext3_quota_on_mount(struct super_block *sb, int type)
2416 {
2417         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2418                         EXT3_SB(sb)->s_jquota_fmt, type);
2419 }
2420
2421 /*
2422  * Standard function to be called on quota_on
2423  */
2424 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2425                          char *path)
2426 {
2427         int err;
2428         struct nameidata nd;
2429
2430         if (!test_opt(sb, QUOTA))
2431                 return -EINVAL;
2432         /* Not journalling quota? */
2433         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2434             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2435                 return vfs_quota_on(sb, type, format_id, path);
2436         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2437         if (err)
2438                 return err;
2439         /* Quotafile not on the same filesystem? */
2440         if (nd.mnt->mnt_sb != sb) {
2441                 path_release(&nd);
2442                 return -EXDEV;
2443         }
2444         /* Quotafile not of fs root? */
2445         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2446                 printk(KERN_WARNING
2447                         "EXT3-fs: Quota file not on filesystem root. "
2448                         "Journalled quota will not work.\n");
2449         path_release(&nd);
2450         return vfs_quota_on(sb, type, format_id, path);
2451 }
2452
2453 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2454  * acquiring the locks... As quota files are never truncated and quota code
2455  * itself serializes the operations (and noone else should touch the files)
2456  * we don't have to be afraid of races */
2457 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2458                                size_t len, loff_t off)
2459 {
2460         struct inode *inode = sb_dqopt(sb)->files[type];
2461         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2462         int err = 0;
2463         int offset = off & (sb->s_blocksize - 1);
2464         int tocopy;
2465         size_t toread;
2466         struct buffer_head *bh;
2467         loff_t i_size = i_size_read(inode);
2468
2469         if (off > i_size)
2470                 return 0;
2471         if (off+len > i_size)
2472                 len = i_size-off;
2473         toread = len;
2474         while (toread > 0) {
2475                 tocopy = sb->s_blocksize - offset < toread ?
2476                                 sb->s_blocksize - offset : toread;
2477                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2478                 if (err)
2479                         return err;
2480                 if (!bh)        /* A hole? */
2481                         memset(data, 0, tocopy);
2482                 else
2483                         memcpy(data, bh->b_data+offset, tocopy);
2484                 brelse(bh);
2485                 offset = 0;
2486                 toread -= tocopy;
2487                 data += tocopy;
2488                 blk++;
2489         }
2490         return len;
2491 }
2492
2493 /* Write to quotafile (we know the transaction is already started and has
2494  * enough credits) */
2495 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2496                                 const char *data, size_t len, loff_t off)
2497 {
2498         struct inode *inode = sb_dqopt(sb)->files[type];
2499         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2500         int err = 0;
2501         int offset = off & (sb->s_blocksize - 1);
2502         int tocopy;
2503         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2504         size_t towrite = len;
2505         struct buffer_head *bh;
2506         handle_t *handle = journal_current_handle();
2507
2508         down(&inode->i_sem);
2509         while (towrite > 0) {
2510                 tocopy = sb->s_blocksize - offset < towrite ?
2511                                 sb->s_blocksize - offset : towrite;
2512                 bh = ext3_bread(handle, inode, blk, 1, &err);
2513                 if (!bh)
2514                         goto out;
2515                 if (journal_quota) {
2516                         err = ext3_journal_get_write_access(handle, bh);
2517                         if (err) {
2518                                 brelse(bh);
2519                                 goto out;
2520                         }
2521                 }
2522                 lock_buffer(bh);
2523                 memcpy(bh->b_data+offset, data, tocopy);
2524                 flush_dcache_page(bh->b_page);
2525                 unlock_buffer(bh);
2526                 if (journal_quota)
2527                         err = ext3_journal_dirty_metadata(handle, bh);
2528                 else {
2529                         /* Always do at least ordered writes for quotas */
2530                         err = ext3_journal_dirty_data(handle, bh);
2531                         mark_buffer_dirty(bh);
2532                 }
2533                 brelse(bh);
2534                 if (err)
2535                         goto out;
2536                 offset = 0;
2537                 towrite -= tocopy;
2538                 data += tocopy;
2539                 blk++;
2540         }
2541 out:
2542         if (len == towrite)
2543                 return err;
2544         if (inode->i_size < off+len-towrite) {
2545                 i_size_write(inode, off+len-towrite);
2546                 EXT3_I(inode)->i_disksize = inode->i_size;
2547         }
2548         inode->i_version++;
2549         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2550         ext3_mark_inode_dirty(handle, inode);
2551         up(&inode->i_sem);
2552         return len - towrite;
2553 }
2554
2555 #endif
2556
2557 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2558         int flags, const char *dev_name, void *data)
2559 {
2560         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2561 }
2562
2563 static struct file_system_type ext3_fs_type = {
2564         .owner          = THIS_MODULE,
2565         .name           = "ext3",
2566         .get_sb         = ext3_get_sb,
2567         .kill_sb        = kill_block_super,
2568         .fs_flags       = FS_REQUIRES_DEV,
2569 };
2570
2571 static int __init init_ext3_fs(void)
2572 {
2573         int err = init_ext3_xattr();
2574         if (err)
2575                 return err;
2576         err = init_inodecache();
2577         if (err)
2578                 goto out1;
2579         err = register_filesystem(&ext3_fs_type);
2580         if (err)
2581                 goto out;
2582         return 0;
2583 out:
2584         destroy_inodecache();
2585 out1:
2586         exit_ext3_xattr();
2587         return err;
2588 }
2589
2590 static void __exit exit_ext3_fs(void)
2591 {
2592         unregister_filesystem(&ext3_fs_type);
2593         destroy_inodecache();
2594         exit_ext3_xattr();
2595 }
2596
2597 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2598 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2599 MODULE_LICENSE("GPL");
2600 module_init(init_ext3_fs)
2601 module_exit(exit_ext3_fs)