[PATCH] quota: consolidate code surrounding vfs_quota_on_mount
[linux-2.6.git] / fs / ext3 / super.c
1 /*
2  *  linux/fs/ext3/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/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,
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_ignore, "grpquota"},
638         {Opt_ignore, "noquota"},
639         {Opt_ignore, "quota"},
640         {Opt_ignore, "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                         break;
880                 case Opt_offusrjquota:
881                         qtype = USRQUOTA;
882                         goto clear_qf_name;
883                 case Opt_offgrpjquota:
884                         qtype = GRPQUOTA;
885 clear_qf_name:
886                         if (sb_any_quota_enabled(sb)) {
887                                 printk(KERN_ERR "EXT3-fs: Cannot change "
888                                         "journalled quota options when "
889                                         "quota turned on.\n");
890                                 return 0;
891                         }
892                         kfree(sbi->s_qf_names[qtype]);
893                         sbi->s_qf_names[qtype] = NULL;
894                         break;
895                 case Opt_jqfmt_vfsold:
896                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
897                         break;
898                 case Opt_jqfmt_vfsv0:
899                         sbi->s_jquota_fmt = QFMT_VFS_V0;
900                         break;
901 #else
902                 case Opt_usrjquota:
903                 case Opt_grpjquota:
904                 case Opt_offusrjquota:
905                 case Opt_offgrpjquota:
906                 case Opt_jqfmt_vfsold:
907                 case Opt_jqfmt_vfsv0:
908                         printk(KERN_ERR
909                                 "EXT3-fs: journalled quota options not "
910                                 "supported.\n");
911                         break;
912 #endif
913                 case Opt_abort:
914                         set_opt(sbi->s_mount_opt, ABORT);
915                         break;
916                 case Opt_barrier:
917                         if (match_int(&args[0], &option))
918                                 return 0;
919                         if (option)
920                                 set_opt(sbi->s_mount_opt, BARRIER);
921                         else
922                                 clear_opt(sbi->s_mount_opt, BARRIER);
923                         break;
924                 case Opt_ignore:
925                         break;
926                 case Opt_resize:
927                         if (!n_blocks_count) {
928                                 printk("EXT3-fs: resize option only available "
929                                         "for remount\n");
930                                 return 0;
931                         }
932                         match_int(&args[0], &option);
933                         *n_blocks_count = option;
934                         break;
935                 case Opt_nobh:
936                         set_opt(sbi->s_mount_opt, NOBH);
937                         break;
938                 default:
939                         printk (KERN_ERR
940                                 "EXT3-fs: Unrecognized mount option \"%s\" "
941                                 "or missing value\n", p);
942                         return 0;
943                 }
944         }
945 #ifdef CONFIG_QUOTA
946         if (!sbi->s_jquota_fmt && (sbi->s_qf_names[USRQUOTA] ||
947             sbi->s_qf_names[GRPQUOTA])) {
948                 printk(KERN_ERR
949                         "EXT3-fs: journalled quota format not specified.\n");
950                 return 0;
951         }
952 #endif
953
954         return 1;
955 }
956
957 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
958                             int read_only)
959 {
960         struct ext3_sb_info *sbi = EXT3_SB(sb);
961         int res = 0;
962
963         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
964                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
965                         "forcing read-only mode\n");
966                 res = MS_RDONLY;
967         }
968         if (read_only)
969                 return res;
970         if (!(sbi->s_mount_state & EXT3_VALID_FS))
971                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
972                         "running e2fsck is recommended\n");
973         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
974                 printk (KERN_WARNING
975                         "EXT3-fs warning: mounting fs with errors, "
976                         "running e2fsck is recommended\n");
977         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
978                  le16_to_cpu(es->s_mnt_count) >=
979                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
980                 printk (KERN_WARNING
981                         "EXT3-fs warning: maximal mount count reached, "
982                         "running e2fsck is recommended\n");
983         else if (le32_to_cpu(es->s_checkinterval) &&
984                 (le32_to_cpu(es->s_lastcheck) +
985                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
986                 printk (KERN_WARNING
987                         "EXT3-fs warning: checktime reached, "
988                         "running e2fsck is recommended\n");
989 #if 0
990                 /* @@@ We _will_ want to clear the valid bit if we find
991                    inconsistencies, to force a fsck at reboot.  But for
992                    a plain journaled filesystem we can keep it set as
993                    valid forever! :) */
994         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
995 #endif
996         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
997                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
998         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
999         es->s_mtime = cpu_to_le32(get_seconds());
1000         ext3_update_dynamic_rev(sb);
1001         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1002
1003         ext3_commit_super(sb, es, 1);
1004         if (test_opt(sb, DEBUG))
1005                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1006                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1007                         sb->s_blocksize,
1008                         sbi->s_groups_count,
1009                         EXT3_BLOCKS_PER_GROUP(sb),
1010                         EXT3_INODES_PER_GROUP(sb),
1011                         sbi->s_mount_opt);
1012
1013         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1014         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1015                 char b[BDEVNAME_SIZE];
1016
1017                 printk("external journal on %s\n",
1018                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1019         } else {
1020                 printk("internal journal\n");
1021         }
1022 #ifdef CONFIG_EXT3_CHECK
1023         if (test_opt (sb, CHECK)) {
1024                 ext3_check_blocks_bitmap (sb);
1025                 ext3_check_inodes_bitmap (sb);
1026         }
1027 #endif
1028         return res;
1029 }
1030
1031 /* Called at mount-time, super-block is locked */
1032 static int ext3_check_descriptors (struct super_block * sb)
1033 {
1034         struct ext3_sb_info *sbi = EXT3_SB(sb);
1035         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1036         struct ext3_group_desc * gdp = NULL;
1037         int desc_block = 0;
1038         int i;
1039
1040         ext3_debug ("Checking group descriptors");
1041
1042         for (i = 0; i < sbi->s_groups_count; i++)
1043         {
1044                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1045                         gdp = (struct ext3_group_desc *)
1046                                         sbi->s_group_desc[desc_block++]->b_data;
1047                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1048                     le32_to_cpu(gdp->bg_block_bitmap) >=
1049                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1050                 {
1051                         ext3_error (sb, "ext3_check_descriptors",
1052                                     "Block bitmap for group %d"
1053                                     " not in group (block %lu)!",
1054                                     i, (unsigned long)
1055                                         le32_to_cpu(gdp->bg_block_bitmap));
1056                         return 0;
1057                 }
1058                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1059                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1060                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1061                 {
1062                         ext3_error (sb, "ext3_check_descriptors",
1063                                     "Inode bitmap for group %d"
1064                                     " not in group (block %lu)!",
1065                                     i, (unsigned long)
1066                                         le32_to_cpu(gdp->bg_inode_bitmap));
1067                         return 0;
1068                 }
1069                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1070                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1071                     block + EXT3_BLOCKS_PER_GROUP(sb))
1072                 {
1073                         ext3_error (sb, "ext3_check_descriptors",
1074                                     "Inode table for group %d"
1075                                     " not in group (block %lu)!",
1076                                     i, (unsigned long)
1077                                         le32_to_cpu(gdp->bg_inode_table));
1078                         return 0;
1079                 }
1080                 block += EXT3_BLOCKS_PER_GROUP(sb);
1081                 gdp++;
1082         }
1083
1084         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1085         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1086         return 1;
1087 }
1088
1089
1090 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1091  * the superblock) which were deleted from all directories, but held open by
1092  * a process at the time of a crash.  We walk the list and try to delete these
1093  * inodes at recovery time (only with a read-write filesystem).
1094  *
1095  * In order to keep the orphan inode chain consistent during traversal (in
1096  * case of crash during recovery), we link each inode into the superblock
1097  * orphan list_head and handle it the same way as an inode deletion during
1098  * normal operation (which journals the operations for us).
1099  *
1100  * We only do an iget() and an iput() on each inode, which is very safe if we
1101  * accidentally point at an in-use or already deleted inode.  The worst that
1102  * can happen in this case is that we get a "bit already cleared" message from
1103  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1104  * e2fsck was run on this filesystem, and it must have already done the orphan
1105  * inode cleanup for us, so we can safely abort without any further action.
1106  */
1107 static void ext3_orphan_cleanup (struct super_block * sb,
1108                                  struct ext3_super_block * es)
1109 {
1110         unsigned int s_flags = sb->s_flags;
1111         int nr_orphans = 0, nr_truncates = 0;
1112 #ifdef CONFIG_QUOTA
1113         int i;
1114 #endif
1115         if (!es->s_last_orphan) {
1116                 jbd_debug(4, "no orphan inodes to clean up\n");
1117                 return;
1118         }
1119
1120         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1121                 if (es->s_last_orphan)
1122                         jbd_debug(1, "Errors on filesystem, "
1123                                   "clearing orphan list.\n");
1124                 es->s_last_orphan = 0;
1125                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1126                 return;
1127         }
1128
1129         if (s_flags & MS_RDONLY) {
1130                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1131                        sb->s_id);
1132                 sb->s_flags &= ~MS_RDONLY;
1133         }
1134 #ifdef CONFIG_QUOTA
1135         /* Needed for iput() to work correctly and not trash data */
1136         sb->s_flags |= MS_ACTIVE;
1137         /* Turn on quotas so that they are updated correctly */
1138         for (i = 0; i < MAXQUOTAS; i++) {
1139                 if (EXT3_SB(sb)->s_qf_names[i]) {
1140                         int ret = ext3_quota_on_mount(sb, i);
1141                         if (ret < 0)
1142                                 printk(KERN_ERR
1143                                         "EXT3-fs: Cannot turn on journalled "
1144                                         "quota: error %d\n", ret);
1145                 }
1146         }
1147 #endif
1148
1149         while (es->s_last_orphan) {
1150                 struct inode *inode;
1151
1152                 if (!(inode =
1153                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1154                         es->s_last_orphan = 0;
1155                         break;
1156                 }
1157
1158                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1159                 DQUOT_INIT(inode);
1160                 if (inode->i_nlink) {
1161                         printk(KERN_DEBUG
1162                                 "%s: truncating inode %ld to %Ld bytes\n",
1163                                 __FUNCTION__, inode->i_ino, inode->i_size);
1164                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1165                                   inode->i_ino, inode->i_size);
1166                         ext3_truncate(inode);
1167                         nr_truncates++;
1168                 } else {
1169                         printk(KERN_DEBUG
1170                                 "%s: deleting unreferenced inode %ld\n",
1171                                 __FUNCTION__, inode->i_ino);
1172                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1173                                   inode->i_ino);
1174                         nr_orphans++;
1175                 }
1176                 iput(inode);  /* The delete magic happens here! */
1177         }
1178
1179 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1180
1181         if (nr_orphans)
1182                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1183                        sb->s_id, PLURAL(nr_orphans));
1184         if (nr_truncates)
1185                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1186                        sb->s_id, PLURAL(nr_truncates));
1187 #ifdef CONFIG_QUOTA
1188         /* Turn quotas off */
1189         for (i = 0; i < MAXQUOTAS; i++) {
1190                 if (sb_dqopt(sb)->files[i])
1191                         vfs_quota_off(sb, i);
1192         }
1193 #endif
1194         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1195 }
1196
1197 #define log2(n) ffz(~(n))
1198
1199 /*
1200  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1201  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1202  * We need to be 1 filesystem block less than the 2^32 sector limit.
1203  */
1204 static loff_t ext3_max_size(int bits)
1205 {
1206         loff_t res = EXT3_NDIR_BLOCKS;
1207         /* This constant is calculated to be the largest file size for a
1208          * dense, 4k-blocksize file such that the total number of
1209          * sectors in the file, including data and all indirect blocks,
1210          * does not exceed 2^32. */
1211         const loff_t upper_limit = 0x1ff7fffd000LL;
1212
1213         res += 1LL << (bits-2);
1214         res += 1LL << (2*(bits-2));
1215         res += 1LL << (3*(bits-2));
1216         res <<= bits;
1217         if (res > upper_limit)
1218                 res = upper_limit;
1219         return res;
1220 }
1221
1222 static unsigned long descriptor_loc(struct super_block *sb,
1223                                     unsigned long logic_sb_block,
1224                                     int nr)
1225 {
1226         struct ext3_sb_info *sbi = EXT3_SB(sb);
1227         unsigned long bg, first_data_block, first_meta_bg;
1228         int has_super = 0;
1229
1230         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1231         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1232
1233         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1234             nr < first_meta_bg)
1235                 return (logic_sb_block + nr + 1);
1236         bg = sbi->s_desc_per_block * nr;
1237         if (ext3_bg_has_super(sb, bg))
1238                 has_super = 1;
1239         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1240 }
1241
1242
1243 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1244 {
1245         struct buffer_head * bh;
1246         struct ext3_super_block *es = NULL;
1247         struct ext3_sb_info *sbi;
1248         unsigned long block;
1249         unsigned long sb_block = get_sb_block(&data);
1250         unsigned long logic_sb_block;
1251         unsigned long offset = 0;
1252         unsigned long journal_inum = 0;
1253         unsigned long def_mount_opts;
1254         struct inode *root;
1255         int blocksize;
1256         int hblock;
1257         int db_count;
1258         int i;
1259         int needs_recovery;
1260         __le32 features;
1261
1262         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1263         if (!sbi)
1264                 return -ENOMEM;
1265         sb->s_fs_info = sbi;
1266         memset(sbi, 0, sizeof(*sbi));
1267         sbi->s_mount_opt = 0;
1268         sbi->s_resuid = EXT3_DEF_RESUID;
1269         sbi->s_resgid = EXT3_DEF_RESGID;
1270
1271         unlock_kernel();
1272
1273         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1274         if (!blocksize) {
1275                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1276                 goto out_fail;
1277         }
1278
1279         /*
1280          * The ext3 superblock will not be buffer aligned for other than 1kB
1281          * block sizes.  We need to calculate the offset from buffer start.
1282          */
1283         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1284                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1285                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1286         } else {
1287                 logic_sb_block = sb_block;
1288         }
1289
1290         if (!(bh = sb_bread(sb, logic_sb_block))) {
1291                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1292                 goto out_fail;
1293         }
1294         /*
1295          * Note: s_es must be initialized as soon as possible because
1296          *       some ext3 macro-instructions depend on its value
1297          */
1298         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1299         sbi->s_es = es;
1300         sb->s_magic = le16_to_cpu(es->s_magic);
1301         if (sb->s_magic != EXT3_SUPER_MAGIC)
1302                 goto cantfind_ext3;
1303
1304         /* Set defaults before we parse the mount options */
1305         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1306         if (def_mount_opts & EXT3_DEFM_DEBUG)
1307                 set_opt(sbi->s_mount_opt, DEBUG);
1308         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1309                 set_opt(sbi->s_mount_opt, GRPID);
1310         if (def_mount_opts & EXT3_DEFM_UID16)
1311                 set_opt(sbi->s_mount_opt, NO_UID32);
1312         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1313                 set_opt(sbi->s_mount_opt, XATTR_USER);
1314         if (def_mount_opts & EXT3_DEFM_ACL)
1315                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1316         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1317                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1318         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1319                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1320         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1321                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1322
1323         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1324                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1325         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1326                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1327
1328         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1329         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1330
1331         set_opt(sbi->s_mount_opt, RESERVATION);
1332
1333         if (!parse_options ((char *) data, sb, &journal_inum, NULL, 0))
1334                 goto failed_mount;
1335
1336         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1337                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1338
1339         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1340             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1341              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1342              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1343                 printk(KERN_WARNING 
1344                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1345                        "running e2fsck is recommended\n");
1346         /*
1347          * Check feature flags regardless of the revision level, since we
1348          * previously didn't change the revision level when setting the flags,
1349          * so there is a chance incompat flags are set on a rev 0 filesystem.
1350          */
1351         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1352         if (features) {
1353                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1354                        "unsupported optional features (%x).\n",
1355                        sb->s_id, le32_to_cpu(features));
1356                 goto failed_mount;
1357         }
1358         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1359         if (!(sb->s_flags & MS_RDONLY) && features) {
1360                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1361                        "unsupported optional features (%x).\n",
1362                        sb->s_id, le32_to_cpu(features));
1363                 goto failed_mount;
1364         }
1365         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1366
1367         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1368             blocksize > EXT3_MAX_BLOCK_SIZE) {
1369                 printk(KERN_ERR 
1370                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1371                        blocksize, sb->s_id);
1372                 goto failed_mount;
1373         }
1374
1375         hblock = bdev_hardsect_size(sb->s_bdev);
1376         if (sb->s_blocksize != blocksize) {
1377                 /*
1378                  * Make sure the blocksize for the filesystem is larger
1379                  * than the hardware sectorsize for the machine.
1380                  */
1381                 if (blocksize < hblock) {
1382                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1383                                "device blocksize %d.\n", blocksize, hblock);
1384                         goto failed_mount;
1385                 }
1386
1387                 brelse (bh);
1388                 sb_set_blocksize(sb, blocksize);
1389                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1390                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1391                 bh = sb_bread(sb, logic_sb_block);
1392                 if (!bh) {
1393                         printk(KERN_ERR 
1394                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1395                         goto failed_mount;
1396                 }
1397                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1398                 sbi->s_es = es;
1399                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1400                         printk (KERN_ERR 
1401                                 "EXT3-fs: Magic mismatch, very weird !\n");
1402                         goto failed_mount;
1403                 }
1404         }
1405
1406         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1407
1408         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1409                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1410                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1411         } else {
1412                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1413                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1414                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1415                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1416                     (sbi->s_inode_size > blocksize)) {
1417                         printk (KERN_ERR
1418                                 "EXT3-fs: unsupported inode size: %d\n",
1419                                 sbi->s_inode_size);
1420                         goto failed_mount;
1421                 }
1422         }
1423         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1424                                    le32_to_cpu(es->s_log_frag_size);
1425         if (blocksize != sbi->s_frag_size) {
1426                 printk(KERN_ERR
1427                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1428                        sbi->s_frag_size, blocksize);
1429                 goto failed_mount;
1430         }
1431         sbi->s_frags_per_block = 1;
1432         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1433         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1434         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1435         if (EXT3_INODE_SIZE(sb) == 0)
1436                 goto cantfind_ext3;
1437         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1438         if (sbi->s_inodes_per_block == 0)
1439                 goto cantfind_ext3;
1440         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1441                                         sbi->s_inodes_per_block;
1442         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1443         sbi->s_sbh = bh;
1444         sbi->s_mount_state = le16_to_cpu(es->s_state);
1445         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1446         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1447         for (i=0; i < 4; i++)
1448                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1449         sbi->s_def_hash_version = es->s_def_hash_version;
1450
1451         if (sbi->s_blocks_per_group > blocksize * 8) {
1452                 printk (KERN_ERR
1453                         "EXT3-fs: #blocks per group too big: %lu\n",
1454                         sbi->s_blocks_per_group);
1455                 goto failed_mount;
1456         }
1457         if (sbi->s_frags_per_group > blocksize * 8) {
1458                 printk (KERN_ERR
1459                         "EXT3-fs: #fragments per group too big: %lu\n",
1460                         sbi->s_frags_per_group);
1461                 goto failed_mount;
1462         }
1463         if (sbi->s_inodes_per_group > blocksize * 8) {
1464                 printk (KERN_ERR
1465                         "EXT3-fs: #inodes per group too big: %lu\n",
1466                         sbi->s_inodes_per_group);
1467                 goto failed_mount;
1468         }
1469
1470         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1471                 goto cantfind_ext3;
1472         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1473                                le32_to_cpu(es->s_first_data_block) +
1474                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1475                               EXT3_BLOCKS_PER_GROUP(sb);
1476         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1477                    EXT3_DESC_PER_BLOCK(sb);
1478         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1479                                     GFP_KERNEL);
1480         if (sbi->s_group_desc == NULL) {
1481                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1482                 goto failed_mount;
1483         }
1484
1485         percpu_counter_init(&sbi->s_freeblocks_counter);
1486         percpu_counter_init(&sbi->s_freeinodes_counter);
1487         percpu_counter_init(&sbi->s_dirs_counter);
1488         bgl_lock_init(&sbi->s_blockgroup_lock);
1489
1490         for (i = 0; i < db_count; i++) {
1491                 block = descriptor_loc(sb, logic_sb_block, i);
1492                 sbi->s_group_desc[i] = sb_bread(sb, block);
1493                 if (!sbi->s_group_desc[i]) {
1494                         printk (KERN_ERR "EXT3-fs: "
1495                                 "can't read group descriptor %d\n", i);
1496                         db_count = i;
1497                         goto failed_mount2;
1498                 }
1499         }
1500         if (!ext3_check_descriptors (sb)) {
1501                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1502                 goto failed_mount2;
1503         }
1504         sbi->s_gdb_count = db_count;
1505         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1506         spin_lock_init(&sbi->s_next_gen_lock);
1507         /* per fileystem reservation list head & lock */
1508         spin_lock_init(&sbi->s_rsv_window_lock);
1509         sbi->s_rsv_window_root = RB_ROOT;
1510         /* Add a single, static dummy reservation to the start of the
1511          * reservation window list --- it gives us a placeholder for
1512          * append-at-start-of-list which makes the allocation logic
1513          * _much_ simpler. */
1514         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1515         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1516         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1517         sbi->s_rsv_window_head.rsv_goal_size = 0;
1518         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1519
1520         /*
1521          * set up enough so that it can read an inode
1522          */
1523         sb->s_op = &ext3_sops;
1524         sb->s_export_op = &ext3_export_ops;
1525         sb->s_xattr = ext3_xattr_handlers;
1526 #ifdef CONFIG_QUOTA
1527         sb->s_qcop = &ext3_qctl_operations;
1528         sb->dq_op = &ext3_quota_operations;
1529 #endif
1530         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1531
1532         sb->s_root = NULL;
1533
1534         needs_recovery = (es->s_last_orphan != 0 ||
1535                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1536                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1537
1538         /*
1539          * The first inode we look at is the journal inode.  Don't try
1540          * root first: it may be modified in the journal!
1541          */
1542         if (!test_opt(sb, NOLOAD) &&
1543             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1544                 if (ext3_load_journal(sb, es))
1545                         goto failed_mount2;
1546         } else if (journal_inum) {
1547                 if (ext3_create_journal(sb, es, journal_inum))
1548                         goto failed_mount2;
1549         } else {
1550                 if (!silent)
1551                         printk (KERN_ERR
1552                                 "ext3: No journal on filesystem on %s\n",
1553                                 sb->s_id);
1554                 goto failed_mount2;
1555         }
1556
1557         /* We have now updated the journal if required, so we can
1558          * validate the data journaling mode. */
1559         switch (test_opt(sb, DATA_FLAGS)) {
1560         case 0:
1561                 /* No mode set, assume a default based on the journal
1562                    capabilities: ORDERED_DATA if the journal can
1563                    cope, else JOURNAL_DATA */
1564                 if (journal_check_available_features
1565                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1566                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1567                 else
1568                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1569                 break;
1570
1571         case EXT3_MOUNT_ORDERED_DATA:
1572         case EXT3_MOUNT_WRITEBACK_DATA:
1573                 if (!journal_check_available_features
1574                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1575                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1576                                "requested data journaling mode\n");
1577                         goto failed_mount3;
1578                 }
1579         default:
1580                 break;
1581         }
1582
1583         if (test_opt(sb, NOBH)) {
1584                 if (sb->s_blocksize_bits != PAGE_CACHE_SHIFT) {
1585                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option "
1586                                 "since filesystem blocksize doesn't match "
1587                                 "pagesize\n");
1588                         clear_opt(sbi->s_mount_opt, NOBH);
1589                 }
1590                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1591                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1592                                 "its supported only with writeback mode\n");
1593                         clear_opt(sbi->s_mount_opt, NOBH);
1594                 }
1595         }
1596         /*
1597          * The journal_load will have done any necessary log recovery,
1598          * so we can safely mount the rest of the filesystem now.
1599          */
1600
1601         root = iget(sb, EXT3_ROOT_INO);
1602         sb->s_root = d_alloc_root(root);
1603         if (!sb->s_root) {
1604                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1605                 iput(root);
1606                 goto failed_mount3;
1607         }
1608         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1609                 dput(sb->s_root);
1610                 sb->s_root = NULL;
1611                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1612                 goto failed_mount3;
1613         }
1614
1615         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1616         /*
1617          * akpm: core read_super() calls in here with the superblock locked.
1618          * That deadlocks, because orphan cleanup needs to lock the superblock
1619          * in numerous places.  Here we just pop the lock - it's relatively
1620          * harmless, because we are now ready to accept write_super() requests,
1621          * and aviro says that's the only reason for hanging onto the
1622          * superblock lock.
1623          */
1624         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1625         ext3_orphan_cleanup(sb, es);
1626         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1627         if (needs_recovery)
1628                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1629         ext3_mark_recovery_complete(sb, es);
1630         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1631                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1632                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1633                 "writeback");
1634
1635         percpu_counter_mod(&sbi->s_freeblocks_counter,
1636                 ext3_count_free_blocks(sb));
1637         percpu_counter_mod(&sbi->s_freeinodes_counter,
1638                 ext3_count_free_inodes(sb));
1639         percpu_counter_mod(&sbi->s_dirs_counter,
1640                 ext3_count_dirs(sb));
1641
1642         lock_kernel();
1643         return 0;
1644
1645 cantfind_ext3:
1646         if (!silent)
1647                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1648                        sb->s_id);
1649         goto failed_mount;
1650
1651 failed_mount3:
1652         journal_destroy(sbi->s_journal);
1653 failed_mount2:
1654         for (i = 0; i < db_count; i++)
1655                 brelse(sbi->s_group_desc[i]);
1656         kfree(sbi->s_group_desc);
1657 failed_mount:
1658 #ifdef CONFIG_QUOTA
1659         for (i = 0; i < MAXQUOTAS; i++)
1660                 kfree(sbi->s_qf_names[i]);
1661 #endif
1662         ext3_blkdev_remove(sbi);
1663         brelse(bh);
1664 out_fail:
1665         sb->s_fs_info = NULL;
1666         kfree(sbi);
1667         lock_kernel();
1668         return -EINVAL;
1669 }
1670
1671 /*
1672  * Setup any per-fs journal parameters now.  We'll do this both on
1673  * initial mount, once the journal has been initialised but before we've
1674  * done any recovery; and again on any subsequent remount. 
1675  */
1676 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1677 {
1678         struct ext3_sb_info *sbi = EXT3_SB(sb);
1679
1680         if (sbi->s_commit_interval)
1681                 journal->j_commit_interval = sbi->s_commit_interval;
1682         /* We could also set up an ext3-specific default for the commit
1683          * interval here, but for now we'll just fall back to the jbd
1684          * default. */
1685
1686         spin_lock(&journal->j_state_lock);
1687         if (test_opt(sb, BARRIER))
1688                 journal->j_flags |= JFS_BARRIER;
1689         else
1690                 journal->j_flags &= ~JFS_BARRIER;
1691         spin_unlock(&journal->j_state_lock);
1692 }
1693
1694 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1695 {
1696         struct inode *journal_inode;
1697         journal_t *journal;
1698
1699         /* First, test for the existence of a valid inode on disk.  Bad
1700          * things happen if we iget() an unused inode, as the subsequent
1701          * iput() will try to delete it. */
1702
1703         journal_inode = iget(sb, journal_inum);
1704         if (!journal_inode) {
1705                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1706                 return NULL;
1707         }
1708         if (!journal_inode->i_nlink) {
1709                 make_bad_inode(journal_inode);
1710                 iput(journal_inode);
1711                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1712                 return NULL;
1713         }
1714
1715         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1716                   journal_inode, journal_inode->i_size);
1717         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1718                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1719                 iput(journal_inode);
1720                 return NULL;
1721         }
1722
1723         journal = journal_init_inode(journal_inode);
1724         if (!journal) {
1725                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1726                 iput(journal_inode);
1727                 return NULL;
1728         }
1729         journal->j_private = sb;
1730         ext3_init_journal_params(sb, journal);
1731         return journal;
1732 }
1733
1734 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1735                                        dev_t j_dev)
1736 {
1737         struct buffer_head * bh;
1738         journal_t *journal;
1739         int start;
1740         int len;
1741         int hblock, blocksize;
1742         unsigned long sb_block;
1743         unsigned long offset;
1744         struct ext3_super_block * es;
1745         struct block_device *bdev;
1746
1747         bdev = ext3_blkdev_get(j_dev);
1748         if (bdev == NULL)
1749                 return NULL;
1750
1751         if (bd_claim(bdev, sb)) {
1752                 printk(KERN_ERR
1753                         "EXT3: failed to claim external journal device.\n");
1754                 blkdev_put(bdev);
1755                 return NULL;
1756         }
1757
1758         blocksize = sb->s_blocksize;
1759         hblock = bdev_hardsect_size(bdev);
1760         if (blocksize < hblock) {
1761                 printk(KERN_ERR
1762                         "EXT3-fs: blocksize too small for journal device.\n");
1763                 goto out_bdev;
1764         }
1765
1766         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1767         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1768         set_blocksize(bdev, blocksize);
1769         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1770                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1771                        "external journal\n");
1772                 goto out_bdev;
1773         }
1774
1775         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1776         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1777             !(le32_to_cpu(es->s_feature_incompat) &
1778               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1779                 printk(KERN_ERR "EXT3-fs: external journal has "
1780                                         "bad superblock\n");
1781                 brelse(bh);
1782                 goto out_bdev;
1783         }
1784
1785         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1786                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1787                 brelse(bh);
1788                 goto out_bdev;
1789         }
1790
1791         len = le32_to_cpu(es->s_blocks_count);
1792         start = sb_block + 1;
1793         brelse(bh);     /* we're done with the superblock */
1794
1795         journal = journal_init_dev(bdev, sb->s_bdev,
1796                                         start, len, blocksize);
1797         if (!journal) {
1798                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1799                 goto out_bdev;
1800         }
1801         journal->j_private = sb;
1802         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1803         wait_on_buffer(journal->j_sb_buffer);
1804         if (!buffer_uptodate(journal->j_sb_buffer)) {
1805                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1806                 goto out_journal;
1807         }
1808         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1809                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1810                                         "user (unsupported) - %d\n",
1811                         be32_to_cpu(journal->j_superblock->s_nr_users));
1812                 goto out_journal;
1813         }
1814         EXT3_SB(sb)->journal_bdev = bdev;
1815         ext3_init_journal_params(sb, journal);
1816         return journal;
1817 out_journal:
1818         journal_destroy(journal);
1819 out_bdev:
1820         ext3_blkdev_put(bdev);
1821         return NULL;
1822 }
1823
1824 static int ext3_load_journal(struct super_block * sb,
1825                              struct ext3_super_block * es)
1826 {
1827         journal_t *journal;
1828         int journal_inum = le32_to_cpu(es->s_journal_inum);
1829         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1830         int err = 0;
1831         int really_read_only;
1832
1833         really_read_only = bdev_read_only(sb->s_bdev);
1834
1835         /*
1836          * Are we loading a blank journal or performing recovery after a
1837          * crash?  For recovery, we need to check in advance whether we
1838          * can get read-write access to the device.
1839          */
1840
1841         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1842                 if (sb->s_flags & MS_RDONLY) {
1843                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1844                                         "required on readonly filesystem.\n");
1845                         if (really_read_only) {
1846                                 printk(KERN_ERR "EXT3-fs: write access "
1847                                         "unavailable, cannot proceed.\n");
1848                                 return -EROFS;
1849                         }
1850                         printk (KERN_INFO "EXT3-fs: write access will "
1851                                         "be enabled during recovery.\n");
1852                 }
1853         }
1854
1855         if (journal_inum && journal_dev) {
1856                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1857                        "and inode journals!\n");
1858                 return -EINVAL;
1859         }
1860
1861         if (journal_inum) {
1862                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1863                         return -EINVAL;
1864         } else {
1865                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1866                         return -EINVAL;
1867         }
1868
1869         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1870                 err = journal_update_format(journal);
1871                 if (err)  {
1872                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1873                         journal_destroy(journal);
1874                         return err;
1875                 }
1876         }
1877
1878         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1879                 err = journal_wipe(journal, !really_read_only);
1880         if (!err)
1881                 err = journal_load(journal);
1882
1883         if (err) {
1884                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1885                 journal_destroy(journal);
1886                 return err;
1887         }
1888
1889         EXT3_SB(sb)->s_journal = journal;
1890         ext3_clear_journal_err(sb, es);
1891         return 0;
1892 }
1893
1894 static int ext3_create_journal(struct super_block * sb,
1895                                struct ext3_super_block * es,
1896                                int journal_inum)
1897 {
1898         journal_t *journal;
1899
1900         if (sb->s_flags & MS_RDONLY) {
1901                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1902                                 "create journal.\n");
1903                 return -EROFS;
1904         }
1905
1906         if (!(journal = ext3_get_journal(sb, journal_inum)))
1907                 return -EINVAL;
1908
1909         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1910                journal_inum);
1911
1912         if (journal_create(journal)) {
1913                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1914                 journal_destroy(journal);
1915                 return -EIO;
1916         }
1917
1918         EXT3_SB(sb)->s_journal = journal;
1919
1920         ext3_update_dynamic_rev(sb);
1921         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1922         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1923
1924         es->s_journal_inum = cpu_to_le32(journal_inum);
1925         sb->s_dirt = 1;
1926
1927         /* Make sure we flush the recovery flag to disk. */
1928         ext3_commit_super(sb, es, 1);
1929
1930         return 0;
1931 }
1932
1933 static void ext3_commit_super (struct super_block * sb,
1934                                struct ext3_super_block * es,
1935                                int sync)
1936 {
1937         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1938
1939         if (!sbh)
1940                 return;
1941         es->s_wtime = cpu_to_le32(get_seconds());
1942         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1943         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1944         BUFFER_TRACE(sbh, "marking dirty");
1945         mark_buffer_dirty(sbh);
1946         if (sync)
1947                 sync_dirty_buffer(sbh);
1948 }
1949
1950
1951 /*
1952  * Have we just finished recovery?  If so, and if we are mounting (or
1953  * remounting) the filesystem readonly, then we will end up with a
1954  * consistent fs on disk.  Record that fact.
1955  */
1956 static void ext3_mark_recovery_complete(struct super_block * sb,
1957                                         struct ext3_super_block * es)
1958 {
1959         journal_t *journal = EXT3_SB(sb)->s_journal;
1960
1961         journal_lock_updates(journal);
1962         journal_flush(journal);
1963         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
1964             sb->s_flags & MS_RDONLY) {
1965                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1966                 sb->s_dirt = 0;
1967                 ext3_commit_super(sb, es, 1);
1968         }
1969         journal_unlock_updates(journal);
1970 }
1971
1972 /*
1973  * If we are mounting (or read-write remounting) a filesystem whose journal
1974  * has recorded an error from a previous lifetime, move that error to the
1975  * main filesystem now.
1976  */
1977 static void ext3_clear_journal_err(struct super_block * sb,
1978                                    struct ext3_super_block * es)
1979 {
1980         journal_t *journal;
1981         int j_errno;
1982         const char *errstr;
1983
1984         journal = EXT3_SB(sb)->s_journal;
1985
1986         /*
1987          * Now check for any error status which may have been recorded in the
1988          * journal by a prior ext3_error() or ext3_abort()
1989          */
1990
1991         j_errno = journal_errno(journal);
1992         if (j_errno) {
1993                 char nbuf[16];
1994
1995                 errstr = ext3_decode_error(sb, j_errno, nbuf);
1996                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
1997                              "from previous mount: %s", errstr);
1998                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
1999                              "filesystem check.");
2000
2001                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2002                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2003                 ext3_commit_super (sb, es, 1);
2004
2005                 journal_clear_err(journal);
2006         }
2007 }
2008
2009 /*
2010  * Force the running and committing transactions to commit,
2011  * and wait on the commit.
2012  */
2013 int ext3_force_commit(struct super_block *sb)
2014 {
2015         journal_t *journal;
2016         int ret;
2017
2018         if (sb->s_flags & MS_RDONLY)
2019                 return 0;
2020
2021         journal = EXT3_SB(sb)->s_journal;
2022         sb->s_dirt = 0;
2023         ret = ext3_journal_force_commit(journal);
2024         return ret;
2025 }
2026
2027 /*
2028  * Ext3 always journals updates to the superblock itself, so we don't
2029  * have to propagate any other updates to the superblock on disk at this
2030  * point.  Just start an async writeback to get the buffers on their way
2031  * to the disk.
2032  *
2033  * This implicitly triggers the writebehind on sync().
2034  */
2035
2036 static void ext3_write_super (struct super_block * sb)
2037 {
2038         if (down_trylock(&sb->s_lock) == 0)
2039                 BUG();
2040         sb->s_dirt = 0;
2041 }
2042
2043 static int ext3_sync_fs(struct super_block *sb, int wait)
2044 {
2045         tid_t target;
2046
2047         sb->s_dirt = 0;
2048         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2049                 if (wait)
2050                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2051         }
2052         return 0;
2053 }
2054
2055 /*
2056  * LVM calls this function before a (read-only) snapshot is created.  This
2057  * gives us a chance to flush the journal completely and mark the fs clean.
2058  */
2059 static void ext3_write_super_lockfs(struct super_block *sb)
2060 {
2061         sb->s_dirt = 0;
2062
2063         if (!(sb->s_flags & MS_RDONLY)) {
2064                 journal_t *journal = EXT3_SB(sb)->s_journal;
2065
2066                 /* Now we set up the journal barrier. */
2067                 journal_lock_updates(journal);
2068                 journal_flush(journal);
2069
2070                 /* Journal blocked and flushed, clear needs_recovery flag. */
2071                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2072                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2073         }
2074 }
2075
2076 /*
2077  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2078  * flag here, even though the filesystem is not technically dirty yet.
2079  */
2080 static void ext3_unlockfs(struct super_block *sb)
2081 {
2082         if (!(sb->s_flags & MS_RDONLY)) {
2083                 lock_super(sb);
2084                 /* Reser the needs_recovery flag before the fs is unlocked. */
2085                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2086                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2087                 unlock_super(sb);
2088                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2089         }
2090 }
2091
2092 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2093 {
2094         struct ext3_super_block * es;
2095         struct ext3_sb_info *sbi = EXT3_SB(sb);
2096         unsigned long tmp;
2097         unsigned long n_blocks_count = 0;
2098
2099         /*
2100          * Allow the "check" option to be passed as a remount option.
2101          */
2102         if (!parse_options(data, sb, &tmp, &n_blocks_count, 1))
2103                 return -EINVAL;
2104
2105         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2106                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2107
2108         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2109                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2110
2111         es = sbi->s_es;
2112
2113         ext3_init_journal_params(sb, sbi->s_journal);
2114
2115         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2116                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2117                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2118                         return -EROFS;
2119
2120                 if (*flags & MS_RDONLY) {
2121                         /*
2122                          * First of all, the unconditional stuff we have to do
2123                          * to disable replay of the journal when we next remount
2124                          */
2125                         sb->s_flags |= MS_RDONLY;
2126
2127                         /*
2128                          * OK, test if we are remounting a valid rw partition
2129                          * readonly, and if so set the rdonly flag and then
2130                          * mark the partition as valid again.
2131                          */
2132                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2133                             (sbi->s_mount_state & EXT3_VALID_FS))
2134                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2135
2136                         ext3_mark_recovery_complete(sb, es);
2137                 } else {
2138                         __le32 ret;
2139                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2140                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2141                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2142                                        "remount RDWR because of unsupported "
2143                                        "optional features (%x).\n",
2144                                        sb->s_id, le32_to_cpu(ret));
2145                                 return -EROFS;
2146                         }
2147                         /*
2148                          * Mounting a RDONLY partition read-write, so reread
2149                          * and store the current valid flag.  (It may have
2150                          * been changed by e2fsck since we originally mounted
2151                          * the partition.)
2152                          */
2153                         ext3_clear_journal_err(sb, es);
2154                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2155                         if ((ret = ext3_group_extend(sb, es, n_blocks_count)))
2156                                 return ret;
2157                         if (!ext3_setup_super (sb, es, 0))
2158                                 sb->s_flags &= ~MS_RDONLY;
2159                 }
2160         }
2161         return 0;
2162 }
2163
2164 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2165 {
2166         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2167         unsigned long overhead;
2168         int i;
2169
2170         if (test_opt (sb, MINIX_DF))
2171                 overhead = 0;
2172         else {
2173                 unsigned long ngroups;
2174                 ngroups = EXT3_SB(sb)->s_groups_count;
2175                 smp_rmb();
2176
2177                 /*
2178                  * Compute the overhead (FS structures)
2179                  */
2180
2181                 /*
2182                  * All of the blocks before first_data_block are
2183                  * overhead
2184                  */
2185                 overhead = le32_to_cpu(es->s_first_data_block);
2186
2187                 /*
2188                  * Add the overhead attributed to the superblock and
2189                  * block group descriptors.  If the sparse superblocks
2190                  * feature is turned on, then not all groups have this.
2191                  */
2192                 for (i = 0; i < ngroups; i++) {
2193                         overhead += ext3_bg_has_super(sb, i) +
2194                                 ext3_bg_num_gdb(sb, i);
2195                         cond_resched();
2196                 }
2197
2198                 /*
2199                  * Every block group has an inode bitmap, a block
2200                  * bitmap, and an inode table.
2201                  */
2202                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2203         }
2204
2205         buf->f_type = EXT3_SUPER_MAGIC;
2206         buf->f_bsize = sb->s_blocksize;
2207         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2208         buf->f_bfree = ext3_count_free_blocks (sb);
2209         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2210         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2211                 buf->f_bavail = 0;
2212         buf->f_files = le32_to_cpu(es->s_inodes_count);
2213         buf->f_ffree = ext3_count_free_inodes (sb);
2214         buf->f_namelen = EXT3_NAME_LEN;
2215         return 0;
2216 }
2217
2218 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2219  * is locked for write. Otherwise the are possible deadlocks:
2220  * Process 1                         Process 2
2221  * ext3_create()                     quota_sync()
2222  *   journal_start()                   write_dquot()
2223  *   DQUOT_INIT()                        down(dqio_sem)
2224  *     down(dqio_sem)                    journal_start()
2225  *
2226  */
2227
2228 #ifdef CONFIG_QUOTA
2229
2230 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2231 {
2232         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2233 }
2234
2235 static int ext3_dquot_initialize(struct inode *inode, int type)
2236 {
2237         handle_t *handle;
2238         int ret, err;
2239
2240         /* We may create quota structure so we need to reserve enough blocks */
2241         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2242         if (IS_ERR(handle))
2243                 return PTR_ERR(handle);
2244         ret = dquot_initialize(inode, type);
2245         err = ext3_journal_stop(handle);
2246         if (!ret)
2247                 ret = err;
2248         return ret;
2249 }
2250
2251 static int ext3_dquot_drop(struct inode *inode)
2252 {
2253         handle_t *handle;
2254         int ret, err;
2255
2256         /* We may delete quota structure so we need to reserve enough blocks */
2257         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2258         if (IS_ERR(handle))
2259                 return PTR_ERR(handle);
2260         ret = dquot_drop(inode);
2261         err = ext3_journal_stop(handle);
2262         if (!ret)
2263                 ret = err;
2264         return ret;
2265 }
2266
2267 static int ext3_write_dquot(struct dquot *dquot)
2268 {
2269         int ret, err;
2270         handle_t *handle;
2271         struct inode *inode;
2272
2273         inode = dquot_to_inode(dquot);
2274         handle = ext3_journal_start(inode,
2275                                         EXT3_QUOTA_TRANS_BLOCKS);
2276         if (IS_ERR(handle))
2277                 return PTR_ERR(handle);
2278         ret = dquot_commit(dquot);
2279         err = ext3_journal_stop(handle);
2280         if (!ret)
2281                 ret = err;
2282         return ret;
2283 }
2284
2285 static int ext3_acquire_dquot(struct dquot *dquot)
2286 {
2287         int ret, err;
2288         handle_t *handle;
2289
2290         handle = ext3_journal_start(dquot_to_inode(dquot),
2291                                         EXT3_QUOTA_INIT_BLOCKS);
2292         if (IS_ERR(handle))
2293                 return PTR_ERR(handle);
2294         ret = dquot_acquire(dquot);
2295         err = ext3_journal_stop(handle);
2296         if (!ret)
2297                 ret = err;
2298         return ret;
2299 }
2300
2301 static int ext3_release_dquot(struct dquot *dquot)
2302 {
2303         int ret, err;
2304         handle_t *handle;
2305
2306         handle = ext3_journal_start(dquot_to_inode(dquot),
2307                                         EXT3_QUOTA_INIT_BLOCKS);
2308         if (IS_ERR(handle))
2309                 return PTR_ERR(handle);
2310         ret = dquot_release(dquot);
2311         err = ext3_journal_stop(handle);
2312         if (!ret)
2313                 ret = err;
2314         return ret;
2315 }
2316
2317 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2318 {
2319         /* Are we journalling quotas? */
2320         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2321             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2322                 dquot_mark_dquot_dirty(dquot);
2323                 return ext3_write_dquot(dquot);
2324         } else {
2325                 return dquot_mark_dquot_dirty(dquot);
2326         }
2327 }
2328
2329 static int ext3_write_info(struct super_block *sb, int type)
2330 {
2331         int ret, err;
2332         handle_t *handle;
2333
2334         /* Data block + inode block */
2335         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2336         if (IS_ERR(handle))
2337                 return PTR_ERR(handle);
2338         ret = dquot_commit_info(sb, type);
2339         err = ext3_journal_stop(handle);
2340         if (!ret)
2341                 ret = err;
2342         return ret;
2343 }
2344
2345 /*
2346  * Turn on quotas during mount time - we need to find
2347  * the quota file and such...
2348  */
2349 static int ext3_quota_on_mount(struct super_block *sb, int type)
2350 {
2351         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2352                         EXT3_SB(sb)->s_jquota_fmt, type);
2353 }
2354
2355 /*
2356  * Standard function to be called on quota_on
2357  */
2358 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2359                          char *path)
2360 {
2361         int err;
2362         struct nameidata nd;
2363
2364         /* Not journalling quota? */
2365         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2366             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2367                 return vfs_quota_on(sb, type, format_id, path);
2368         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2369         if (err)
2370                 return err;
2371         /* Quotafile not on the same filesystem? */
2372         if (nd.mnt->mnt_sb != sb) {
2373                 path_release(&nd);
2374                 return -EXDEV;
2375         }
2376         /* Quotafile not of fs root? */
2377         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2378                 printk(KERN_WARNING
2379                         "EXT3-fs: Quota file not on filesystem root. "
2380                         "Journalled quota will not work.\n");
2381         path_release(&nd);
2382         return vfs_quota_on(sb, type, format_id, path);
2383 }
2384
2385 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2386  * acquiring the locks... As quota files are never truncated and quota code
2387  * itself serializes the operations (and noone else should touch the files)
2388  * we don't have to be afraid of races */
2389 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2390                                size_t len, loff_t off)
2391 {
2392         struct inode *inode = sb_dqopt(sb)->files[type];
2393         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2394         int err = 0;
2395         int offset = off & (sb->s_blocksize - 1);
2396         int tocopy;
2397         size_t toread;
2398         struct buffer_head *bh;
2399         loff_t i_size = i_size_read(inode);
2400
2401         if (off > i_size)
2402                 return 0;
2403         if (off+len > i_size)
2404                 len = i_size-off;
2405         toread = len;
2406         while (toread > 0) {
2407                 tocopy = sb->s_blocksize - offset < toread ?
2408                                 sb->s_blocksize - offset : toread;
2409                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2410                 if (err)
2411                         return err;
2412                 if (!bh)        /* A hole? */
2413                         memset(data, 0, tocopy);
2414                 else
2415                         memcpy(data, bh->b_data+offset, tocopy);
2416                 brelse(bh);
2417                 offset = 0;
2418                 toread -= tocopy;
2419                 data += tocopy;
2420                 blk++;
2421         }
2422         return len;
2423 }
2424
2425 /* Write to quotafile (we know the transaction is already started and has
2426  * enough credits) */
2427 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2428                                 const char *data, size_t len, loff_t off)
2429 {
2430         struct inode *inode = sb_dqopt(sb)->files[type];
2431         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2432         int err = 0;
2433         int offset = off & (sb->s_blocksize - 1);
2434         int tocopy;
2435         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2436         size_t towrite = len;
2437         struct buffer_head *bh;
2438         handle_t *handle = journal_current_handle();
2439
2440         down(&inode->i_sem);
2441         while (towrite > 0) {
2442                 tocopy = sb->s_blocksize - offset < towrite ?
2443                                 sb->s_blocksize - offset : towrite;
2444                 bh = ext3_bread(handle, inode, blk, 1, &err);
2445                 if (!bh)
2446                         goto out;
2447                 if (journal_quota) {
2448                         err = ext3_journal_get_write_access(handle, bh);
2449                         if (err) {
2450                                 brelse(bh);
2451                                 goto out;
2452                         }
2453                 }
2454                 lock_buffer(bh);
2455                 memcpy(bh->b_data+offset, data, tocopy);
2456                 flush_dcache_page(bh->b_page);
2457                 unlock_buffer(bh);
2458                 if (journal_quota)
2459                         err = ext3_journal_dirty_metadata(handle, bh);
2460                 else {
2461                         /* Always do at least ordered writes for quotas */
2462                         err = ext3_journal_dirty_data(handle, bh);
2463                         mark_buffer_dirty(bh);
2464                 }
2465                 brelse(bh);
2466                 if (err)
2467                         goto out;
2468                 offset = 0;
2469                 towrite -= tocopy;
2470                 data += tocopy;
2471                 blk++;
2472         }
2473 out:
2474         if (len == towrite)
2475                 return err;
2476         if (inode->i_size < off+len-towrite) {
2477                 i_size_write(inode, off+len-towrite);
2478                 EXT3_I(inode)->i_disksize = inode->i_size;
2479         }
2480         inode->i_version++;
2481         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2482         ext3_mark_inode_dirty(handle, inode);
2483         up(&inode->i_sem);
2484         return len - towrite;
2485 }
2486
2487 #endif
2488
2489 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2490         int flags, const char *dev_name, void *data)
2491 {
2492         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2493 }
2494
2495 static struct file_system_type ext3_fs_type = {
2496         .owner          = THIS_MODULE,
2497         .name           = "ext3",
2498         .get_sb         = ext3_get_sb,
2499         .kill_sb        = kill_block_super,
2500         .fs_flags       = FS_REQUIRES_DEV,
2501 };
2502
2503 static int __init init_ext3_fs(void)
2504 {
2505         int err = init_ext3_xattr();
2506         if (err)
2507                 return err;
2508         err = init_inodecache();
2509         if (err)
2510                 goto out1;
2511         err = register_filesystem(&ext3_fs_type);
2512         if (err)
2513                 goto out;
2514         return 0;
2515 out:
2516         destroy_inodecache();
2517 out1:
2518         exit_ext3_xattr();
2519         return err;
2520 }
2521
2522 static void __exit exit_ext3_fs(void)
2523 {
2524         unregister_filesystem(&ext3_fs_type);
2525         destroy_inodecache();
2526         exit_ext3_xattr();
2527 }
2528
2529 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2530 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2531 MODULE_LICENSE("GPL");
2532 module_init(init_ext3_fs)
2533 module_exit(exit_ext3_fs)