[PATCH] sem2mutex: quota
[linux-2.6] / fs / ext3 / super.c
1 /*
2  *  linux/fs/ext3/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/config.h>
20 #include <linux/module.h>
21 #include <linux/string.h>
22 #include <linux/fs.h>
23 #include <linux/time.h>
24 #include <linux/jbd.h>
25 #include <linux/ext3_fs.h>
26 #include <linux/ext3_jbd.h>
27 #include <linux/slab.h>
28 #include <linux/init.h>
29 #include <linux/blkdev.h>
30 #include <linux/parser.h>
31 #include <linux/smp_lock.h>
32 #include <linux/buffer_head.h>
33 #include <linux/vfs.h>
34 #include <linux/random.h>
35 #include <linux/mount.h>
36 #include <linux/namei.h>
37 #include <linux/quotaops.h>
38 #include <linux/seq_file.h>
39
40 #include <asm/uaccess.h>
41
42 #include "xattr.h"
43 #include "acl.h"
44 #include "namei.h"
45
46 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
47                              unsigned long journal_devnum);
48 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
49                                int);
50 static void ext3_commit_super (struct super_block * sb,
51                                struct ext3_super_block * es,
52                                int sync);
53 static void ext3_mark_recovery_complete(struct super_block * sb,
54                                         struct ext3_super_block * es);
55 static void ext3_clear_journal_err(struct super_block * sb,
56                                    struct ext3_super_block * es);
57 static int ext3_sync_fs(struct super_block *sb, int wait);
58 static const char *ext3_decode_error(struct super_block * sb, int errno,
59                                      char nbuf[16]);
60 static int ext3_remount (struct super_block * sb, int * flags, char * data);
61 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf);
62 static void ext3_unlockfs(struct super_block *sb);
63 static void ext3_write_super (struct super_block * sb);
64 static void ext3_write_super_lockfs(struct super_block *sb);
65
66 /* 
67  * Wrappers for journal_start/end.
68  *
69  * The only special thing we need to do here is to make sure that all
70  * journal_end calls result in the superblock being marked dirty, so
71  * that sync() will call the filesystem's write_super callback if
72  * appropriate. 
73  */
74 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
75 {
76         journal_t *journal;
77
78         if (sb->s_flags & MS_RDONLY)
79                 return ERR_PTR(-EROFS);
80
81         /* Special case here: if the journal has aborted behind our
82          * backs (eg. EIO in the commit thread), then we still need to
83          * take the FS itself readonly cleanly. */
84         journal = EXT3_SB(sb)->s_journal;
85         if (is_journal_aborted(journal)) {
86                 ext3_abort(sb, __FUNCTION__,
87                            "Detected aborted journal");
88                 return ERR_PTR(-EROFS);
89         }
90
91         return journal_start(journal, nblocks);
92 }
93
94 /* 
95  * The only special thing we need to do here is to make sure that all
96  * journal_stop calls result in the superblock being marked dirty, so
97  * that sync() will call the filesystem's write_super callback if
98  * appropriate. 
99  */
100 int __ext3_journal_stop(const char *where, handle_t *handle)
101 {
102         struct super_block *sb;
103         int err;
104         int rc;
105
106         sb = handle->h_transaction->t_journal->j_private;
107         err = handle->h_err;
108         rc = journal_stop(handle);
109
110         if (!err)
111                 err = rc;
112         if (err)
113                 __ext3_std_error(sb, where, err);
114         return err;
115 }
116
117 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
118                 struct buffer_head *bh, handle_t *handle, int err)
119 {
120         char nbuf[16];
121         const char *errstr = ext3_decode_error(NULL, err, nbuf);
122
123         if (bh)
124                 BUFFER_TRACE(bh, "abort");
125
126         if (!handle->h_err)
127                 handle->h_err = err;
128
129         if (is_handle_aborted(handle))
130                 return;
131
132         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
133                caller, errstr, err_fn);
134
135         journal_abort_handle(handle);
136 }
137
138 /* Deal with the reporting of failure conditions on a filesystem such as
139  * inconsistencies detected or read IO failures.
140  *
141  * On ext2, we can store the error state of the filesystem in the
142  * superblock.  That is not possible on ext3, because we may have other
143  * write ordering constraints on the superblock which prevent us from
144  * writing it out straight away; and given that the journal is about to
145  * be aborted, we can't rely on the current, or future, transactions to
146  * write out the superblock safely.
147  *
148  * We'll just use the journal_abort() error code to record an error in
149  * the journal instead.  On recovery, the journal will compain about
150  * that error until we've noted it down and cleared it.
151  */
152
153 static void ext3_handle_error(struct super_block *sb)
154 {
155         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
156
157         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
158         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
159
160         if (sb->s_flags & MS_RDONLY)
161                 return;
162
163         if (test_opt (sb, ERRORS_RO)) {
164                 printk (KERN_CRIT "Remounting filesystem read-only\n");
165                 sb->s_flags |= MS_RDONLY;
166         } else {
167                 journal_t *journal = EXT3_SB(sb)->s_journal;
168
169                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
170                 if (journal)
171                         journal_abort(journal, -EIO);
172         }
173         if (test_opt(sb, ERRORS_PANIC))
174                 panic("EXT3-fs (device %s): panic forced after error\n",
175                         sb->s_id);
176         ext3_commit_super(sb, es, 1);
177 }
178
179 void ext3_error (struct super_block * sb, const char * function,
180                  const char * fmt, ...)
181 {
182         va_list args;
183
184         va_start(args, fmt);
185         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
186         vprintk(fmt, args);
187         printk("\n");
188         va_end(args);
189
190         ext3_handle_error(sb);
191 }
192
193 static const char *ext3_decode_error(struct super_block * sb, int errno,
194                                      char nbuf[16])
195 {
196         char *errstr = NULL;
197
198         switch (errno) {
199         case -EIO:
200                 errstr = "IO failure";
201                 break;
202         case -ENOMEM:
203                 errstr = "Out of memory";
204                 break;
205         case -EROFS:
206                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
207                         errstr = "Journal has aborted";
208                 else
209                         errstr = "Readonly filesystem";
210                 break;
211         default:
212                 /* If the caller passed in an extra buffer for unknown
213                  * errors, textualise them now.  Else we just return
214                  * NULL. */
215                 if (nbuf) {
216                         /* Check for truncated error codes... */
217                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
218                                 errstr = nbuf;
219                 }
220                 break;
221         }
222
223         return errstr;
224 }
225
226 /* __ext3_std_error decodes expected errors from journaling functions
227  * automatically and invokes the appropriate error response.  */
228
229 void __ext3_std_error (struct super_block * sb, const char * function,
230                        int errno)
231 {
232         char nbuf[16];
233         const char *errstr;
234
235         /* Special case: if the error is EROFS, and we're not already
236          * inside a transaction, then there's really no point in logging
237          * an error. */
238         if (errno == -EROFS && journal_current_handle() == NULL &&
239             (sb->s_flags & MS_RDONLY))
240                 return;
241
242         errstr = ext3_decode_error(sb, errno, nbuf);
243         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
244                 sb->s_id, function, errstr);
245
246         ext3_handle_error(sb);
247 }
248
249 /*
250  * ext3_abort is a much stronger failure handler than ext3_error.  The
251  * abort function may be used to deal with unrecoverable failures such
252  * as journal IO errors or ENOMEM at a critical moment in log management.
253  *
254  * We unconditionally force the filesystem into an ABORT|READONLY state,
255  * unless the error response on the fs has been set to panic in which
256  * case we take the easy way out and panic immediately.
257  */
258
259 void ext3_abort (struct super_block * sb, const char * function,
260                  const char * fmt, ...)
261 {
262         va_list args;
263
264         printk (KERN_CRIT "ext3_abort called.\n");
265
266         va_start(args, fmt);
267         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
268         vprintk(fmt, args);
269         printk("\n");
270         va_end(args);
271
272         if (test_opt(sb, ERRORS_PANIC))
273                 panic("EXT3-fs panic from previous error\n");
274
275         if (sb->s_flags & MS_RDONLY)
276                 return;
277
278         printk(KERN_CRIT "Remounting filesystem read-only\n");
279         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
280         sb->s_flags |= MS_RDONLY;
281         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
282         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
283 }
284
285 void ext3_warning (struct super_block * sb, const char * function,
286                    const char * fmt, ...)
287 {
288         va_list args;
289
290         va_start(args, fmt);
291         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
292                sb->s_id, function);
293         vprintk(fmt, args);
294         printk("\n");
295         va_end(args);
296 }
297
298 void ext3_update_dynamic_rev(struct super_block *sb)
299 {
300         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
301
302         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
303                 return;
304
305         ext3_warning(sb, __FUNCTION__,
306                      "updating to rev %d because of new feature flag, "
307                      "running e2fsck is recommended",
308                      EXT3_DYNAMIC_REV);
309
310         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
311         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
312         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
313         /* leave es->s_feature_*compat flags alone */
314         /* es->s_uuid will be set by e2fsck if empty */
315
316         /*
317          * The rest of the superblock fields should be zero, and if not it
318          * means they are likely already in use, so leave them alone.  We
319          * can leave it up to e2fsck to clean up any inconsistencies there.
320          */
321 }
322
323 /*
324  * Open the external journal device
325  */
326 static struct block_device *ext3_blkdev_get(dev_t dev)
327 {
328         struct block_device *bdev;
329         char b[BDEVNAME_SIZE];
330
331         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
332         if (IS_ERR(bdev))
333                 goto fail;
334         return bdev;
335
336 fail:
337         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
338                         __bdevname(dev, b), PTR_ERR(bdev));
339         return NULL;
340 }
341
342 /*
343  * Release the journal device
344  */
345 static int ext3_blkdev_put(struct block_device *bdev)
346 {
347         bd_release(bdev);
348         return blkdev_put(bdev);
349 }
350
351 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
352 {
353         struct block_device *bdev;
354         int ret = -ENODEV;
355
356         bdev = sbi->journal_bdev;
357         if (bdev) {
358                 ret = ext3_blkdev_put(bdev);
359                 sbi->journal_bdev = NULL;
360         }
361         return ret;
362 }
363
364 static inline struct inode *orphan_list_entry(struct list_head *l)
365 {
366         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
367 }
368
369 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
370 {
371         struct list_head *l;
372
373         printk(KERN_ERR "sb orphan head is %d\n", 
374                le32_to_cpu(sbi->s_es->s_last_orphan));
375
376         printk(KERN_ERR "sb_info orphan list:\n");
377         list_for_each(l, &sbi->s_orphan) {
378                 struct inode *inode = orphan_list_entry(l);
379                 printk(KERN_ERR "  "
380                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
381                        inode->i_sb->s_id, inode->i_ino, inode,
382                        inode->i_mode, inode->i_nlink, 
383                        NEXT_ORPHAN(inode));
384         }
385 }
386
387 static void ext3_put_super (struct super_block * sb)
388 {
389         struct ext3_sb_info *sbi = EXT3_SB(sb);
390         struct ext3_super_block *es = sbi->s_es;
391         int i;
392
393         ext3_xattr_put_super(sb);
394         journal_destroy(sbi->s_journal);
395         if (!(sb->s_flags & MS_RDONLY)) {
396                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
397                 es->s_state = cpu_to_le16(sbi->s_mount_state);
398                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
399                 mark_buffer_dirty(sbi->s_sbh);
400                 ext3_commit_super(sb, es, 1);
401         }
402
403         for (i = 0; i < sbi->s_gdb_count; i++)
404                 brelse(sbi->s_group_desc[i]);
405         kfree(sbi->s_group_desc);
406         percpu_counter_destroy(&sbi->s_freeblocks_counter);
407         percpu_counter_destroy(&sbi->s_freeinodes_counter);
408         percpu_counter_destroy(&sbi->s_dirs_counter);
409         brelse(sbi->s_sbh);
410 #ifdef CONFIG_QUOTA
411         for (i = 0; i < MAXQUOTAS; i++)
412                 kfree(sbi->s_qf_names[i]);
413 #endif
414
415         /* Debugging code just in case the in-memory inode orphan list
416          * isn't empty.  The on-disk one can be non-empty if we've
417          * detected an error and taken the fs readonly, but the
418          * in-memory list had better be clean by this point. */
419         if (!list_empty(&sbi->s_orphan))
420                 dump_orphan_list(sb, sbi);
421         J_ASSERT(list_empty(&sbi->s_orphan));
422
423         invalidate_bdev(sb->s_bdev, 0);
424         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
425                 /*
426                  * Invalidate the journal device's buffers.  We don't want them
427                  * floating about in memory - the physical journal device may
428                  * hotswapped, and it breaks the `ro-after' testing code.
429                  */
430                 sync_blockdev(sbi->journal_bdev);
431                 invalidate_bdev(sbi->journal_bdev, 0);
432                 ext3_blkdev_remove(sbi);
433         }
434         sb->s_fs_info = NULL;
435         kfree(sbi);
436         return;
437 }
438
439 static kmem_cache_t *ext3_inode_cachep;
440
441 /*
442  * Called inside transaction, so use GFP_NOFS
443  */
444 static struct inode *ext3_alloc_inode(struct super_block *sb)
445 {
446         struct ext3_inode_info *ei;
447
448         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
449         if (!ei)
450                 return NULL;
451 #ifdef CONFIG_EXT3_FS_POSIX_ACL
452         ei->i_acl = EXT3_ACL_NOT_CACHED;
453         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
454 #endif
455         ei->i_block_alloc_info = NULL;
456         ei->vfs_inode.i_version = 1;
457         return &ei->vfs_inode;
458 }
459
460 static void ext3_destroy_inode(struct inode *inode)
461 {
462         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
463 }
464
465 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
466 {
467         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
468
469         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
470             SLAB_CTOR_CONSTRUCTOR) {
471                 INIT_LIST_HEAD(&ei->i_orphan);
472 #ifdef CONFIG_EXT3_FS_XATTR
473                 init_rwsem(&ei->xattr_sem);
474 #endif
475                 init_MUTEX(&ei->truncate_sem);
476                 inode_init_once(&ei->vfs_inode);
477         }
478 }
479  
480 static int init_inodecache(void)
481 {
482         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
483                                              sizeof(struct ext3_inode_info),
484                                              0, SLAB_RECLAIM_ACCOUNT,
485                                              init_once, NULL);
486         if (ext3_inode_cachep == NULL)
487                 return -ENOMEM;
488         return 0;
489 }
490
491 static void destroy_inodecache(void)
492 {
493         if (kmem_cache_destroy(ext3_inode_cachep))
494                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
495 }
496
497 static void ext3_clear_inode(struct inode *inode)
498 {
499         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
500 #ifdef CONFIG_EXT3_FS_POSIX_ACL
501        if (EXT3_I(inode)->i_acl &&
502            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
503                posix_acl_release(EXT3_I(inode)->i_acl);
504                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
505        }
506        if (EXT3_I(inode)->i_default_acl &&
507            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
508                posix_acl_release(EXT3_I(inode)->i_default_acl);
509                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
510        }
511 #endif
512         ext3_discard_reservation(inode);
513         EXT3_I(inode)->i_block_alloc_info = NULL;
514         kfree(rsv);
515 }
516
517 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
518 {
519 #if defined(CONFIG_QUOTA)
520         struct ext3_sb_info *sbi = EXT3_SB(sb);
521
522         if (sbi->s_jquota_fmt)
523                 seq_printf(seq, ",jqfmt=%s",
524                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
525
526         if (sbi->s_qf_names[USRQUOTA])
527                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
528
529         if (sbi->s_qf_names[GRPQUOTA])
530                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
531
532         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
533                 seq_puts(seq, ",usrquota");
534
535         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
536                 seq_puts(seq, ",grpquota");
537 #endif
538 }
539
540 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
541 {
542         struct super_block *sb = vfs->mnt_sb;
543
544         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
545                 seq_puts(seq, ",data=journal");
546         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
547                 seq_puts(seq, ",data=ordered");
548         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
549                 seq_puts(seq, ",data=writeback");
550
551         ext3_show_quota_options(seq, sb);
552
553         return 0;
554 }
555
556 #ifdef CONFIG_QUOTA
557 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
558 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
559
560 static int ext3_dquot_initialize(struct inode *inode, int type);
561 static int ext3_dquot_drop(struct inode *inode);
562 static int ext3_write_dquot(struct dquot *dquot);
563 static int ext3_acquire_dquot(struct dquot *dquot);
564 static int ext3_release_dquot(struct dquot *dquot);
565 static int ext3_mark_dquot_dirty(struct dquot *dquot);
566 static int ext3_write_info(struct super_block *sb, int type);
567 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
568 static int ext3_quota_on_mount(struct super_block *sb, int type);
569 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
570                                size_t len, loff_t off);
571 static ssize_t ext3_quota_write(struct super_block *sb, int type,
572                                 const char *data, size_t len, loff_t off);
573
574 static struct dquot_operations ext3_quota_operations = {
575         .initialize     = ext3_dquot_initialize,
576         .drop           = ext3_dquot_drop,
577         .alloc_space    = dquot_alloc_space,
578         .alloc_inode    = dquot_alloc_inode,
579         .free_space     = dquot_free_space,
580         .free_inode     = dquot_free_inode,
581         .transfer       = dquot_transfer,
582         .write_dquot    = ext3_write_dquot,
583         .acquire_dquot  = ext3_acquire_dquot,
584         .release_dquot  = ext3_release_dquot,
585         .mark_dirty     = ext3_mark_dquot_dirty,
586         .write_info     = ext3_write_info
587 };
588
589 static struct quotactl_ops ext3_qctl_operations = {
590         .quota_on       = ext3_quota_on,
591         .quota_off      = vfs_quota_off,
592         .quota_sync     = vfs_quota_sync,
593         .get_info       = vfs_get_dqinfo,
594         .set_info       = vfs_set_dqinfo,
595         .get_dqblk      = vfs_get_dqblk,
596         .set_dqblk      = vfs_set_dqblk
597 };
598 #endif
599
600 static struct super_operations ext3_sops = {
601         .alloc_inode    = ext3_alloc_inode,
602         .destroy_inode  = ext3_destroy_inode,
603         .read_inode     = ext3_read_inode,
604         .write_inode    = ext3_write_inode,
605         .dirty_inode    = ext3_dirty_inode,
606         .delete_inode   = ext3_delete_inode,
607         .put_super      = ext3_put_super,
608         .write_super    = ext3_write_super,
609         .sync_fs        = ext3_sync_fs,
610         .write_super_lockfs = ext3_write_super_lockfs,
611         .unlockfs       = ext3_unlockfs,
612         .statfs         = ext3_statfs,
613         .remount_fs     = ext3_remount,
614         .clear_inode    = ext3_clear_inode,
615         .show_options   = ext3_show_options,
616 #ifdef CONFIG_QUOTA
617         .quota_read     = ext3_quota_read,
618         .quota_write    = ext3_quota_write,
619 #endif
620 };
621
622 static struct export_operations ext3_export_ops = {
623         .get_parent = ext3_get_parent,
624 };
625
626 enum {
627         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
628         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
629         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
630         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
631         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh,
632         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
633         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
634         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
635         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
636         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
637         Opt_grpquota
638 };
639
640 static match_table_t tokens = {
641         {Opt_bsd_df, "bsddf"},
642         {Opt_minix_df, "minixdf"},
643         {Opt_grpid, "grpid"},
644         {Opt_grpid, "bsdgroups"},
645         {Opt_nogrpid, "nogrpid"},
646         {Opt_nogrpid, "sysvgroups"},
647         {Opt_resgid, "resgid=%u"},
648         {Opt_resuid, "resuid=%u"},
649         {Opt_sb, "sb=%u"},
650         {Opt_err_cont, "errors=continue"},
651         {Opt_err_panic, "errors=panic"},
652         {Opt_err_ro, "errors=remount-ro"},
653         {Opt_nouid32, "nouid32"},
654         {Opt_nocheck, "nocheck"},
655         {Opt_nocheck, "check=none"},
656         {Opt_debug, "debug"},
657         {Opt_oldalloc, "oldalloc"},
658         {Opt_orlov, "orlov"},
659         {Opt_user_xattr, "user_xattr"},
660         {Opt_nouser_xattr, "nouser_xattr"},
661         {Opt_acl, "acl"},
662         {Opt_noacl, "noacl"},
663         {Opt_reservation, "reservation"},
664         {Opt_noreservation, "noreservation"},
665         {Opt_noload, "noload"},
666         {Opt_nobh, "nobh"},
667         {Opt_commit, "commit=%u"},
668         {Opt_journal_update, "journal=update"},
669         {Opt_journal_inum, "journal=%u"},
670         {Opt_journal_dev, "journal_dev=%u"},
671         {Opt_abort, "abort"},
672         {Opt_data_journal, "data=journal"},
673         {Opt_data_ordered, "data=ordered"},
674         {Opt_data_writeback, "data=writeback"},
675         {Opt_offusrjquota, "usrjquota="},
676         {Opt_usrjquota, "usrjquota=%s"},
677         {Opt_offgrpjquota, "grpjquota="},
678         {Opt_grpjquota, "grpjquota=%s"},
679         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
680         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
681         {Opt_grpquota, "grpquota"},
682         {Opt_noquota, "noquota"},
683         {Opt_quota, "quota"},
684         {Opt_usrquota, "usrquota"},
685         {Opt_barrier, "barrier=%u"},
686         {Opt_err, NULL},
687         {Opt_resize, "resize"},
688 };
689
690 static unsigned long get_sb_block(void **data)
691 {
692         unsigned long   sb_block;
693         char            *options = (char *) *data;
694
695         if (!options || strncmp(options, "sb=", 3) != 0)
696                 return 1;       /* Default location */
697         options += 3;
698         sb_block = simple_strtoul(options, &options, 0);
699         if (*options && *options != ',') {
700                 printk("EXT3-fs: Invalid sb specification: %s\n",
701                        (char *) *data);
702                 return 1;
703         }
704         if (*options == ',')
705                 options++;
706         *data = (void *) options;
707         return sb_block;
708 }
709
710 static int parse_options (char *options, struct super_block *sb,
711                           unsigned long *inum, unsigned long *journal_devnum,
712                           unsigned long *n_blocks_count, int is_remount)
713 {
714         struct ext3_sb_info *sbi = EXT3_SB(sb);
715         char * p;
716         substring_t args[MAX_OPT_ARGS];
717         int data_opt = 0;
718         int option;
719 #ifdef CONFIG_QUOTA
720         int qtype;
721         char *qname;
722 #endif
723
724         if (!options)
725                 return 1;
726
727         while ((p = strsep (&options, ",")) != NULL) {
728                 int token;
729                 if (!*p)
730                         continue;
731
732                 token = match_token(p, tokens, args);
733                 switch (token) {
734                 case Opt_bsd_df:
735                         clear_opt (sbi->s_mount_opt, MINIX_DF);
736                         break;
737                 case Opt_minix_df:
738                         set_opt (sbi->s_mount_opt, MINIX_DF);
739                         break;
740                 case Opt_grpid:
741                         set_opt (sbi->s_mount_opt, GRPID);
742                         break;
743                 case Opt_nogrpid:
744                         clear_opt (sbi->s_mount_opt, GRPID);
745                         break;
746                 case Opt_resuid:
747                         if (match_int(&args[0], &option))
748                                 return 0;
749                         sbi->s_resuid = option;
750                         break;
751                 case Opt_resgid:
752                         if (match_int(&args[0], &option))
753                                 return 0;
754                         sbi->s_resgid = option;
755                         break;
756                 case Opt_sb:
757                         /* handled by get_sb_block() instead of here */
758                         /* *sb_block = match_int(&args[0]); */
759                         break;
760                 case Opt_err_panic:
761                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
762                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
763                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
764                         break;
765                 case Opt_err_ro:
766                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
767                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
768                         set_opt (sbi->s_mount_opt, ERRORS_RO);
769                         break;
770                 case Opt_err_cont:
771                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
772                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
773                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
774                         break;
775                 case Opt_nouid32:
776                         set_opt (sbi->s_mount_opt, NO_UID32);
777                         break;
778                 case Opt_nocheck:
779                         clear_opt (sbi->s_mount_opt, CHECK);
780                         break;
781                 case Opt_debug:
782                         set_opt (sbi->s_mount_opt, DEBUG);
783                         break;
784                 case Opt_oldalloc:
785                         set_opt (sbi->s_mount_opt, OLDALLOC);
786                         break;
787                 case Opt_orlov:
788                         clear_opt (sbi->s_mount_opt, OLDALLOC);
789                         break;
790 #ifdef CONFIG_EXT3_FS_XATTR
791                 case Opt_user_xattr:
792                         set_opt (sbi->s_mount_opt, XATTR_USER);
793                         break;
794                 case Opt_nouser_xattr:
795                         clear_opt (sbi->s_mount_opt, XATTR_USER);
796                         break;
797 #else
798                 case Opt_user_xattr:
799                 case Opt_nouser_xattr:
800                         printk("EXT3 (no)user_xattr options not supported\n");
801                         break;
802 #endif
803 #ifdef CONFIG_EXT3_FS_POSIX_ACL
804                 case Opt_acl:
805                         set_opt(sbi->s_mount_opt, POSIX_ACL);
806                         break;
807                 case Opt_noacl:
808                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
809                         break;
810 #else
811                 case Opt_acl:
812                 case Opt_noacl:
813                         printk("EXT3 (no)acl options not supported\n");
814                         break;
815 #endif
816                 case Opt_reservation:
817                         set_opt(sbi->s_mount_opt, RESERVATION);
818                         break;
819                 case Opt_noreservation:
820                         clear_opt(sbi->s_mount_opt, RESERVATION);
821                         break;
822                 case Opt_journal_update:
823                         /* @@@ FIXME */
824                         /* Eventually we will want to be able to create
825                            a journal file here.  For now, only allow the
826                            user to specify an existing inode to be the
827                            journal file. */
828                         if (is_remount) {
829                                 printk(KERN_ERR "EXT3-fs: cannot specify "
830                                        "journal on remount\n");
831                                 return 0;
832                         }
833                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
834                         break;
835                 case Opt_journal_inum:
836                         if (is_remount) {
837                                 printk(KERN_ERR "EXT3-fs: cannot specify "
838                                        "journal on remount\n");
839                                 return 0;
840                         }
841                         if (match_int(&args[0], &option))
842                                 return 0;
843                         *inum = option;
844                         break;
845                 case Opt_journal_dev:
846                         if (is_remount) {
847                                 printk(KERN_ERR "EXT3-fs: cannot specify "
848                                        "journal on remount\n");
849                                 return 0;
850                         }
851                         if (match_int(&args[0], &option))
852                                 return 0;
853                         *journal_devnum = option;
854                         break;
855                 case Opt_noload:
856                         set_opt (sbi->s_mount_opt, NOLOAD);
857                         break;
858                 case Opt_commit:
859                         if (match_int(&args[0], &option))
860                                 return 0;
861                         if (option < 0)
862                                 return 0;
863                         if (option == 0)
864                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
865                         sbi->s_commit_interval = HZ * option;
866                         break;
867                 case Opt_data_journal:
868                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
869                         goto datacheck;
870                 case Opt_data_ordered:
871                         data_opt = EXT3_MOUNT_ORDERED_DATA;
872                         goto datacheck;
873                 case Opt_data_writeback:
874                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
875                 datacheck:
876                         if (is_remount) {
877                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
878                                                 != data_opt) {
879                                         printk(KERN_ERR
880                                                 "EXT3-fs: cannot change data "
881                                                 "mode on remount\n");
882                                         return 0;
883                                 }
884                         } else {
885                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
886                                 sbi->s_mount_opt |= data_opt;
887                         }
888                         break;
889 #ifdef CONFIG_QUOTA
890                 case Opt_usrjquota:
891                         qtype = USRQUOTA;
892                         goto set_qf_name;
893                 case Opt_grpjquota:
894                         qtype = GRPQUOTA;
895 set_qf_name:
896                         if (sb_any_quota_enabled(sb)) {
897                                 printk(KERN_ERR
898                                         "EXT3-fs: Cannot change journalled "
899                                         "quota options when quota turned on.\n");
900                                 return 0;
901                         }
902                         qname = match_strdup(&args[0]);
903                         if (!qname) {
904                                 printk(KERN_ERR
905                                         "EXT3-fs: not enough memory for "
906                                         "storing quotafile name.\n");
907                                 return 0;
908                         }
909                         if (sbi->s_qf_names[qtype] &&
910                             strcmp(sbi->s_qf_names[qtype], qname)) {
911                                 printk(KERN_ERR
912                                         "EXT3-fs: %s quota file already "
913                                         "specified.\n", QTYPE2NAME(qtype));
914                                 kfree(qname);
915                                 return 0;
916                         }
917                         sbi->s_qf_names[qtype] = qname;
918                         if (strchr(sbi->s_qf_names[qtype], '/')) {
919                                 printk(KERN_ERR
920                                         "EXT3-fs: quotafile must be on "
921                                         "filesystem root.\n");
922                                 kfree(sbi->s_qf_names[qtype]);
923                                 sbi->s_qf_names[qtype] = NULL;
924                                 return 0;
925                         }
926                         set_opt(sbi->s_mount_opt, QUOTA);
927                         break;
928                 case Opt_offusrjquota:
929                         qtype = USRQUOTA;
930                         goto clear_qf_name;
931                 case Opt_offgrpjquota:
932                         qtype = GRPQUOTA;
933 clear_qf_name:
934                         if (sb_any_quota_enabled(sb)) {
935                                 printk(KERN_ERR "EXT3-fs: Cannot change "
936                                         "journalled quota options when "
937                                         "quota turned on.\n");
938                                 return 0;
939                         }
940                         /*
941                          * The space will be released later when all options
942                          * are confirmed to be correct
943                          */
944                         sbi->s_qf_names[qtype] = NULL;
945                         break;
946                 case Opt_jqfmt_vfsold:
947                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
948                         break;
949                 case Opt_jqfmt_vfsv0:
950                         sbi->s_jquota_fmt = QFMT_VFS_V0;
951                         break;
952                 case Opt_quota:
953                 case Opt_usrquota:
954                         set_opt(sbi->s_mount_opt, QUOTA);
955                         set_opt(sbi->s_mount_opt, USRQUOTA);
956                         break;
957                 case Opt_grpquota:
958                         set_opt(sbi->s_mount_opt, QUOTA);
959                         set_opt(sbi->s_mount_opt, GRPQUOTA);
960                         break;
961                 case Opt_noquota:
962                         if (sb_any_quota_enabled(sb)) {
963                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
964                                         "options when quota turned on.\n");
965                                 return 0;
966                         }
967                         clear_opt(sbi->s_mount_opt, QUOTA);
968                         clear_opt(sbi->s_mount_opt, USRQUOTA);
969                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
970                         break;
971 #else
972                 case Opt_quota:
973                 case Opt_usrquota:
974                 case Opt_grpquota:
975                 case Opt_usrjquota:
976                 case Opt_grpjquota:
977                 case Opt_offusrjquota:
978                 case Opt_offgrpjquota:
979                 case Opt_jqfmt_vfsold:
980                 case Opt_jqfmt_vfsv0:
981                         printk(KERN_ERR
982                                 "EXT3-fs: journalled quota options not "
983                                 "supported.\n");
984                         break;
985                 case Opt_noquota:
986                         break;
987 #endif
988                 case Opt_abort:
989                         set_opt(sbi->s_mount_opt, ABORT);
990                         break;
991                 case Opt_barrier:
992                         if (match_int(&args[0], &option))
993                                 return 0;
994                         if (option)
995                                 set_opt(sbi->s_mount_opt, BARRIER);
996                         else
997                                 clear_opt(sbi->s_mount_opt, BARRIER);
998                         break;
999                 case Opt_ignore:
1000                         break;
1001                 case Opt_resize:
1002                         if (!is_remount) {
1003                                 printk("EXT3-fs: resize option only available "
1004                                         "for remount\n");
1005                                 return 0;
1006                         }
1007                         if (match_int(&args[0], &option) != 0)
1008                                 return 0;
1009                         *n_blocks_count = option;
1010                         break;
1011                 case Opt_nobh:
1012                         set_opt(sbi->s_mount_opt, NOBH);
1013                         break;
1014                 default:
1015                         printk (KERN_ERR
1016                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1017                                 "or missing value\n", p);
1018                         return 0;
1019                 }
1020         }
1021 #ifdef CONFIG_QUOTA
1022         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1023                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1024                      sbi->s_qf_names[USRQUOTA])
1025                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1026
1027                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1028                      sbi->s_qf_names[GRPQUOTA])
1029                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1030
1031                 if ((sbi->s_qf_names[USRQUOTA] &&
1032                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1033                     (sbi->s_qf_names[GRPQUOTA] &&
1034                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1035                         printk(KERN_ERR "EXT3-fs: old and new quota "
1036                                         "format mixing.\n");
1037                         return 0;
1038                 }
1039
1040                 if (!sbi->s_jquota_fmt) {
1041                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1042                                         "not specified.\n");
1043                         return 0;
1044                 }
1045         } else {
1046                 if (sbi->s_jquota_fmt) {
1047                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1048                                         "specified with no journalling "
1049                                         "enabled.\n");
1050                         return 0;
1051                 }
1052         }
1053 #endif
1054         return 1;
1055 }
1056
1057 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1058                             int read_only)
1059 {
1060         struct ext3_sb_info *sbi = EXT3_SB(sb);
1061         int res = 0;
1062
1063         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1064                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1065                         "forcing read-only mode\n");
1066                 res = MS_RDONLY;
1067         }
1068         if (read_only)
1069                 return res;
1070         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1071                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1072                         "running e2fsck is recommended\n");
1073         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1074                 printk (KERN_WARNING
1075                         "EXT3-fs warning: mounting fs with errors, "
1076                         "running e2fsck is recommended\n");
1077         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1078                  le16_to_cpu(es->s_mnt_count) >=
1079                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1080                 printk (KERN_WARNING
1081                         "EXT3-fs warning: maximal mount count reached, "
1082                         "running e2fsck is recommended\n");
1083         else if (le32_to_cpu(es->s_checkinterval) &&
1084                 (le32_to_cpu(es->s_lastcheck) +
1085                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1086                 printk (KERN_WARNING
1087                         "EXT3-fs warning: checktime reached, "
1088                         "running e2fsck is recommended\n");
1089 #if 0
1090                 /* @@@ We _will_ want to clear the valid bit if we find
1091                    inconsistencies, to force a fsck at reboot.  But for
1092                    a plain journaled filesystem we can keep it set as
1093                    valid forever! :) */
1094         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1095 #endif
1096         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1097                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1098         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1099         es->s_mtime = cpu_to_le32(get_seconds());
1100         ext3_update_dynamic_rev(sb);
1101         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1102
1103         ext3_commit_super(sb, es, 1);
1104         if (test_opt(sb, DEBUG))
1105                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1106                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1107                         sb->s_blocksize,
1108                         sbi->s_groups_count,
1109                         EXT3_BLOCKS_PER_GROUP(sb),
1110                         EXT3_INODES_PER_GROUP(sb),
1111                         sbi->s_mount_opt);
1112
1113         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1114         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1115                 char b[BDEVNAME_SIZE];
1116
1117                 printk("external journal on %s\n",
1118                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1119         } else {
1120                 printk("internal journal\n");
1121         }
1122         return res;
1123 }
1124
1125 /* Called at mount-time, super-block is locked */
1126 static int ext3_check_descriptors (struct super_block * sb)
1127 {
1128         struct ext3_sb_info *sbi = EXT3_SB(sb);
1129         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1130         struct ext3_group_desc * gdp = NULL;
1131         int desc_block = 0;
1132         int i;
1133
1134         ext3_debug ("Checking group descriptors");
1135
1136         for (i = 0; i < sbi->s_groups_count; i++)
1137         {
1138                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1139                         gdp = (struct ext3_group_desc *)
1140                                         sbi->s_group_desc[desc_block++]->b_data;
1141                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1142                     le32_to_cpu(gdp->bg_block_bitmap) >=
1143                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1144                 {
1145                         ext3_error (sb, "ext3_check_descriptors",
1146                                     "Block bitmap for group %d"
1147                                     " not in group (block %lu)!",
1148                                     i, (unsigned long)
1149                                         le32_to_cpu(gdp->bg_block_bitmap));
1150                         return 0;
1151                 }
1152                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1153                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1154                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1155                 {
1156                         ext3_error (sb, "ext3_check_descriptors",
1157                                     "Inode bitmap for group %d"
1158                                     " not in group (block %lu)!",
1159                                     i, (unsigned long)
1160                                         le32_to_cpu(gdp->bg_inode_bitmap));
1161                         return 0;
1162                 }
1163                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1164                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1165                     block + EXT3_BLOCKS_PER_GROUP(sb))
1166                 {
1167                         ext3_error (sb, "ext3_check_descriptors",
1168                                     "Inode table for group %d"
1169                                     " not in group (block %lu)!",
1170                                     i, (unsigned long)
1171                                         le32_to_cpu(gdp->bg_inode_table));
1172                         return 0;
1173                 }
1174                 block += EXT3_BLOCKS_PER_GROUP(sb);
1175                 gdp++;
1176         }
1177
1178         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1179         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1180         return 1;
1181 }
1182
1183
1184 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1185  * the superblock) which were deleted from all directories, but held open by
1186  * a process at the time of a crash.  We walk the list and try to delete these
1187  * inodes at recovery time (only with a read-write filesystem).
1188  *
1189  * In order to keep the orphan inode chain consistent during traversal (in
1190  * case of crash during recovery), we link each inode into the superblock
1191  * orphan list_head and handle it the same way as an inode deletion during
1192  * normal operation (which journals the operations for us).
1193  *
1194  * We only do an iget() and an iput() on each inode, which is very safe if we
1195  * accidentally point at an in-use or already deleted inode.  The worst that
1196  * can happen in this case is that we get a "bit already cleared" message from
1197  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1198  * e2fsck was run on this filesystem, and it must have already done the orphan
1199  * inode cleanup for us, so we can safely abort without any further action.
1200  */
1201 static void ext3_orphan_cleanup (struct super_block * sb,
1202                                  struct ext3_super_block * es)
1203 {
1204         unsigned int s_flags = sb->s_flags;
1205         int nr_orphans = 0, nr_truncates = 0;
1206 #ifdef CONFIG_QUOTA
1207         int i;
1208 #endif
1209         if (!es->s_last_orphan) {
1210                 jbd_debug(4, "no orphan inodes to clean up\n");
1211                 return;
1212         }
1213
1214         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1215                 if (es->s_last_orphan)
1216                         jbd_debug(1, "Errors on filesystem, "
1217                                   "clearing orphan list.\n");
1218                 es->s_last_orphan = 0;
1219                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1220                 return;
1221         }
1222
1223         if (s_flags & MS_RDONLY) {
1224                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1225                        sb->s_id);
1226                 sb->s_flags &= ~MS_RDONLY;
1227         }
1228 #ifdef CONFIG_QUOTA
1229         /* Needed for iput() to work correctly and not trash data */
1230         sb->s_flags |= MS_ACTIVE;
1231         /* Turn on quotas so that they are updated correctly */
1232         for (i = 0; i < MAXQUOTAS; i++) {
1233                 if (EXT3_SB(sb)->s_qf_names[i]) {
1234                         int ret = ext3_quota_on_mount(sb, i);
1235                         if (ret < 0)
1236                                 printk(KERN_ERR
1237                                         "EXT3-fs: Cannot turn on journalled "
1238                                         "quota: error %d\n", ret);
1239                 }
1240         }
1241 #endif
1242
1243         while (es->s_last_orphan) {
1244                 struct inode *inode;
1245
1246                 if (!(inode =
1247                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1248                         es->s_last_orphan = 0;
1249                         break;
1250                 }
1251
1252                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1253                 DQUOT_INIT(inode);
1254                 if (inode->i_nlink) {
1255                         printk(KERN_DEBUG
1256                                 "%s: truncating inode %ld to %Ld bytes\n",
1257                                 __FUNCTION__, inode->i_ino, inode->i_size);
1258                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1259                                   inode->i_ino, inode->i_size);
1260                         ext3_truncate(inode);
1261                         nr_truncates++;
1262                 } else {
1263                         printk(KERN_DEBUG
1264                                 "%s: deleting unreferenced inode %ld\n",
1265                                 __FUNCTION__, inode->i_ino);
1266                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1267                                   inode->i_ino);
1268                         nr_orphans++;
1269                 }
1270                 iput(inode);  /* The delete magic happens here! */
1271         }
1272
1273 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1274
1275         if (nr_orphans)
1276                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1277                        sb->s_id, PLURAL(nr_orphans));
1278         if (nr_truncates)
1279                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1280                        sb->s_id, PLURAL(nr_truncates));
1281 #ifdef CONFIG_QUOTA
1282         /* Turn quotas off */
1283         for (i = 0; i < MAXQUOTAS; i++) {
1284                 if (sb_dqopt(sb)->files[i])
1285                         vfs_quota_off(sb, i);
1286         }
1287 #endif
1288         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1289 }
1290
1291 #define log2(n) ffz(~(n))
1292
1293 /*
1294  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1295  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1296  * We need to be 1 filesystem block less than the 2^32 sector limit.
1297  */
1298 static loff_t ext3_max_size(int bits)
1299 {
1300         loff_t res = EXT3_NDIR_BLOCKS;
1301         /* This constant is calculated to be the largest file size for a
1302          * dense, 4k-blocksize file such that the total number of
1303          * sectors in the file, including data and all indirect blocks,
1304          * does not exceed 2^32. */
1305         const loff_t upper_limit = 0x1ff7fffd000LL;
1306
1307         res += 1LL << (bits-2);
1308         res += 1LL << (2*(bits-2));
1309         res += 1LL << (3*(bits-2));
1310         res <<= bits;
1311         if (res > upper_limit)
1312                 res = upper_limit;
1313         return res;
1314 }
1315
1316 static unsigned long descriptor_loc(struct super_block *sb,
1317                                     unsigned long logic_sb_block,
1318                                     int nr)
1319 {
1320         struct ext3_sb_info *sbi = EXT3_SB(sb);
1321         unsigned long bg, first_data_block, first_meta_bg;
1322         int has_super = 0;
1323
1324         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1325         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1326
1327         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1328             nr < first_meta_bg)
1329                 return (logic_sb_block + nr + 1);
1330         bg = sbi->s_desc_per_block * nr;
1331         if (ext3_bg_has_super(sb, bg))
1332                 has_super = 1;
1333         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1334 }
1335
1336
1337 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1338 {
1339         struct buffer_head * bh;
1340         struct ext3_super_block *es = NULL;
1341         struct ext3_sb_info *sbi;
1342         unsigned long block;
1343         unsigned long sb_block = get_sb_block(&data);
1344         unsigned long logic_sb_block;
1345         unsigned long offset = 0;
1346         unsigned long journal_inum = 0;
1347         unsigned long journal_devnum = 0;
1348         unsigned long def_mount_opts;
1349         struct inode *root;
1350         int blocksize;
1351         int hblock;
1352         int db_count;
1353         int i;
1354         int needs_recovery;
1355         __le32 features;
1356
1357         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1358         if (!sbi)
1359                 return -ENOMEM;
1360         sb->s_fs_info = sbi;
1361         memset(sbi, 0, sizeof(*sbi));
1362         sbi->s_mount_opt = 0;
1363         sbi->s_resuid = EXT3_DEF_RESUID;
1364         sbi->s_resgid = EXT3_DEF_RESGID;
1365
1366         unlock_kernel();
1367
1368         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1369         if (!blocksize) {
1370                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1371                 goto out_fail;
1372         }
1373
1374         /*
1375          * The ext3 superblock will not be buffer aligned for other than 1kB
1376          * block sizes.  We need to calculate the offset from buffer start.
1377          */
1378         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1379                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1380                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1381         } else {
1382                 logic_sb_block = sb_block;
1383         }
1384
1385         if (!(bh = sb_bread(sb, logic_sb_block))) {
1386                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1387                 goto out_fail;
1388         }
1389         /*
1390          * Note: s_es must be initialized as soon as possible because
1391          *       some ext3 macro-instructions depend on its value
1392          */
1393         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1394         sbi->s_es = es;
1395         sb->s_magic = le16_to_cpu(es->s_magic);
1396         if (sb->s_magic != EXT3_SUPER_MAGIC)
1397                 goto cantfind_ext3;
1398
1399         /* Set defaults before we parse the mount options */
1400         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1401         if (def_mount_opts & EXT3_DEFM_DEBUG)
1402                 set_opt(sbi->s_mount_opt, DEBUG);
1403         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1404                 set_opt(sbi->s_mount_opt, GRPID);
1405         if (def_mount_opts & EXT3_DEFM_UID16)
1406                 set_opt(sbi->s_mount_opt, NO_UID32);
1407         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1408                 set_opt(sbi->s_mount_opt, XATTR_USER);
1409         if (def_mount_opts & EXT3_DEFM_ACL)
1410                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1411         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1412                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1413         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1414                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1415         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1416                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1417
1418         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1419                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1420         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1421                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1422
1423         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1424         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1425
1426         set_opt(sbi->s_mount_opt, RESERVATION);
1427
1428         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1429                             NULL, 0))
1430                 goto failed_mount;
1431
1432         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1433                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1434
1435         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1436             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1437              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1438              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1439                 printk(KERN_WARNING 
1440                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1441                        "running e2fsck is recommended\n");
1442         /*
1443          * Check feature flags regardless of the revision level, since we
1444          * previously didn't change the revision level when setting the flags,
1445          * so there is a chance incompat flags are set on a rev 0 filesystem.
1446          */
1447         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1448         if (features) {
1449                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1450                        "unsupported optional features (%x).\n",
1451                        sb->s_id, le32_to_cpu(features));
1452                 goto failed_mount;
1453         }
1454         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1455         if (!(sb->s_flags & MS_RDONLY) && features) {
1456                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1457                        "unsupported optional features (%x).\n",
1458                        sb->s_id, le32_to_cpu(features));
1459                 goto failed_mount;
1460         }
1461         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1462
1463         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1464             blocksize > EXT3_MAX_BLOCK_SIZE) {
1465                 printk(KERN_ERR 
1466                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1467                        blocksize, sb->s_id);
1468                 goto failed_mount;
1469         }
1470
1471         hblock = bdev_hardsect_size(sb->s_bdev);
1472         if (sb->s_blocksize != blocksize) {
1473                 /*
1474                  * Make sure the blocksize for the filesystem is larger
1475                  * than the hardware sectorsize for the machine.
1476                  */
1477                 if (blocksize < hblock) {
1478                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1479                                "device blocksize %d.\n", blocksize, hblock);
1480                         goto failed_mount;
1481                 }
1482
1483                 brelse (bh);
1484                 sb_set_blocksize(sb, blocksize);
1485                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1486                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1487                 bh = sb_bread(sb, logic_sb_block);
1488                 if (!bh) {
1489                         printk(KERN_ERR 
1490                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1491                         goto failed_mount;
1492                 }
1493                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1494                 sbi->s_es = es;
1495                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1496                         printk (KERN_ERR 
1497                                 "EXT3-fs: Magic mismatch, very weird !\n");
1498                         goto failed_mount;
1499                 }
1500         }
1501
1502         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1503
1504         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1505                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1506                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1507         } else {
1508                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1509                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1510                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1511                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1512                     (sbi->s_inode_size > blocksize)) {
1513                         printk (KERN_ERR
1514                                 "EXT3-fs: unsupported inode size: %d\n",
1515                                 sbi->s_inode_size);
1516                         goto failed_mount;
1517                 }
1518         }
1519         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1520                                    le32_to_cpu(es->s_log_frag_size);
1521         if (blocksize != sbi->s_frag_size) {
1522                 printk(KERN_ERR
1523                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1524                        sbi->s_frag_size, blocksize);
1525                 goto failed_mount;
1526         }
1527         sbi->s_frags_per_block = 1;
1528         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1529         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1530         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1531         if (EXT3_INODE_SIZE(sb) == 0)
1532                 goto cantfind_ext3;
1533         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1534         if (sbi->s_inodes_per_block == 0)
1535                 goto cantfind_ext3;
1536         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1537                                         sbi->s_inodes_per_block;
1538         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1539         sbi->s_sbh = bh;
1540         sbi->s_mount_state = le16_to_cpu(es->s_state);
1541         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1542         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1543         for (i=0; i < 4; i++)
1544                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1545         sbi->s_def_hash_version = es->s_def_hash_version;
1546
1547         if (sbi->s_blocks_per_group > blocksize * 8) {
1548                 printk (KERN_ERR
1549                         "EXT3-fs: #blocks per group too big: %lu\n",
1550                         sbi->s_blocks_per_group);
1551                 goto failed_mount;
1552         }
1553         if (sbi->s_frags_per_group > blocksize * 8) {
1554                 printk (KERN_ERR
1555                         "EXT3-fs: #fragments per group too big: %lu\n",
1556                         sbi->s_frags_per_group);
1557                 goto failed_mount;
1558         }
1559         if (sbi->s_inodes_per_group > blocksize * 8) {
1560                 printk (KERN_ERR
1561                         "EXT3-fs: #inodes per group too big: %lu\n",
1562                         sbi->s_inodes_per_group);
1563                 goto failed_mount;
1564         }
1565
1566         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1567                 goto cantfind_ext3;
1568         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1569                                le32_to_cpu(es->s_first_data_block) +
1570                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1571                               EXT3_BLOCKS_PER_GROUP(sb);
1572         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1573                    EXT3_DESC_PER_BLOCK(sb);
1574         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1575                                     GFP_KERNEL);
1576         if (sbi->s_group_desc == NULL) {
1577                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1578                 goto failed_mount;
1579         }
1580
1581         percpu_counter_init(&sbi->s_freeblocks_counter);
1582         percpu_counter_init(&sbi->s_freeinodes_counter);
1583         percpu_counter_init(&sbi->s_dirs_counter);
1584         bgl_lock_init(&sbi->s_blockgroup_lock);
1585
1586         for (i = 0; i < db_count; i++) {
1587                 block = descriptor_loc(sb, logic_sb_block, i);
1588                 sbi->s_group_desc[i] = sb_bread(sb, block);
1589                 if (!sbi->s_group_desc[i]) {
1590                         printk (KERN_ERR "EXT3-fs: "
1591                                 "can't read group descriptor %d\n", i);
1592                         db_count = i;
1593                         goto failed_mount2;
1594                 }
1595         }
1596         if (!ext3_check_descriptors (sb)) {
1597                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1598                 goto failed_mount2;
1599         }
1600         sbi->s_gdb_count = db_count;
1601         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1602         spin_lock_init(&sbi->s_next_gen_lock);
1603         /* per fileystem reservation list head & lock */
1604         spin_lock_init(&sbi->s_rsv_window_lock);
1605         sbi->s_rsv_window_root = RB_ROOT;
1606         /* Add a single, static dummy reservation to the start of the
1607          * reservation window list --- it gives us a placeholder for
1608          * append-at-start-of-list which makes the allocation logic
1609          * _much_ simpler. */
1610         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1611         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1612         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1613         sbi->s_rsv_window_head.rsv_goal_size = 0;
1614         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1615
1616         /*
1617          * set up enough so that it can read an inode
1618          */
1619         sb->s_op = &ext3_sops;
1620         sb->s_export_op = &ext3_export_ops;
1621         sb->s_xattr = ext3_xattr_handlers;
1622 #ifdef CONFIG_QUOTA
1623         sb->s_qcop = &ext3_qctl_operations;
1624         sb->dq_op = &ext3_quota_operations;
1625 #endif
1626         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1627
1628         sb->s_root = NULL;
1629
1630         needs_recovery = (es->s_last_orphan != 0 ||
1631                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1632                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1633
1634         /*
1635          * The first inode we look at is the journal inode.  Don't try
1636          * root first: it may be modified in the journal!
1637          */
1638         if (!test_opt(sb, NOLOAD) &&
1639             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1640                 if (ext3_load_journal(sb, es, journal_devnum))
1641                         goto failed_mount2;
1642         } else if (journal_inum) {
1643                 if (ext3_create_journal(sb, es, journal_inum))
1644                         goto failed_mount2;
1645         } else {
1646                 if (!silent)
1647                         printk (KERN_ERR
1648                                 "ext3: No journal on filesystem on %s\n",
1649                                 sb->s_id);
1650                 goto failed_mount2;
1651         }
1652
1653         /* We have now updated the journal if required, so we can
1654          * validate the data journaling mode. */
1655         switch (test_opt(sb, DATA_FLAGS)) {
1656         case 0:
1657                 /* No mode set, assume a default based on the journal
1658                    capabilities: ORDERED_DATA if the journal can
1659                    cope, else JOURNAL_DATA */
1660                 if (journal_check_available_features
1661                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1662                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1663                 else
1664                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1665                 break;
1666
1667         case EXT3_MOUNT_ORDERED_DATA:
1668         case EXT3_MOUNT_WRITEBACK_DATA:
1669                 if (!journal_check_available_features
1670                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1671                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1672                                "requested data journaling mode\n");
1673                         goto failed_mount3;
1674                 }
1675         default:
1676                 break;
1677         }
1678
1679         if (test_opt(sb, NOBH)) {
1680                 if (sb->s_blocksize_bits != PAGE_CACHE_SHIFT) {
1681                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option "
1682                                 "since filesystem blocksize doesn't match "
1683                                 "pagesize\n");
1684                         clear_opt(sbi->s_mount_opt, NOBH);
1685                 }
1686                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1687                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1688                                 "its supported only with writeback mode\n");
1689                         clear_opt(sbi->s_mount_opt, NOBH);
1690                 }
1691         }
1692         /*
1693          * The journal_load will have done any necessary log recovery,
1694          * so we can safely mount the rest of the filesystem now.
1695          */
1696
1697         root = iget(sb, EXT3_ROOT_INO);
1698         sb->s_root = d_alloc_root(root);
1699         if (!sb->s_root) {
1700                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1701                 iput(root);
1702                 goto failed_mount3;
1703         }
1704         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1705                 dput(sb->s_root);
1706                 sb->s_root = NULL;
1707                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1708                 goto failed_mount3;
1709         }
1710
1711         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1712         /*
1713          * akpm: core read_super() calls in here with the superblock locked.
1714          * That deadlocks, because orphan cleanup needs to lock the superblock
1715          * in numerous places.  Here we just pop the lock - it's relatively
1716          * harmless, because we are now ready to accept write_super() requests,
1717          * and aviro says that's the only reason for hanging onto the
1718          * superblock lock.
1719          */
1720         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1721         ext3_orphan_cleanup(sb, es);
1722         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1723         if (needs_recovery)
1724                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1725         ext3_mark_recovery_complete(sb, es);
1726         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1727                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1728                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1729                 "writeback");
1730
1731         percpu_counter_mod(&sbi->s_freeblocks_counter,
1732                 ext3_count_free_blocks(sb));
1733         percpu_counter_mod(&sbi->s_freeinodes_counter,
1734                 ext3_count_free_inodes(sb));
1735         percpu_counter_mod(&sbi->s_dirs_counter,
1736                 ext3_count_dirs(sb));
1737
1738         lock_kernel();
1739         return 0;
1740
1741 cantfind_ext3:
1742         if (!silent)
1743                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1744                        sb->s_id);
1745         goto failed_mount;
1746
1747 failed_mount3:
1748         journal_destroy(sbi->s_journal);
1749 failed_mount2:
1750         for (i = 0; i < db_count; i++)
1751                 brelse(sbi->s_group_desc[i]);
1752         kfree(sbi->s_group_desc);
1753 failed_mount:
1754 #ifdef CONFIG_QUOTA
1755         for (i = 0; i < MAXQUOTAS; i++)
1756                 kfree(sbi->s_qf_names[i]);
1757 #endif
1758         ext3_blkdev_remove(sbi);
1759         brelse(bh);
1760 out_fail:
1761         sb->s_fs_info = NULL;
1762         kfree(sbi);
1763         lock_kernel();
1764         return -EINVAL;
1765 }
1766
1767 /*
1768  * Setup any per-fs journal parameters now.  We'll do this both on
1769  * initial mount, once the journal has been initialised but before we've
1770  * done any recovery; and again on any subsequent remount. 
1771  */
1772 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1773 {
1774         struct ext3_sb_info *sbi = EXT3_SB(sb);
1775
1776         if (sbi->s_commit_interval)
1777                 journal->j_commit_interval = sbi->s_commit_interval;
1778         /* We could also set up an ext3-specific default for the commit
1779          * interval here, but for now we'll just fall back to the jbd
1780          * default. */
1781
1782         spin_lock(&journal->j_state_lock);
1783         if (test_opt(sb, BARRIER))
1784                 journal->j_flags |= JFS_BARRIER;
1785         else
1786                 journal->j_flags &= ~JFS_BARRIER;
1787         spin_unlock(&journal->j_state_lock);
1788 }
1789
1790 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1791 {
1792         struct inode *journal_inode;
1793         journal_t *journal;
1794
1795         /* First, test for the existence of a valid inode on disk.  Bad
1796          * things happen if we iget() an unused inode, as the subsequent
1797          * iput() will try to delete it. */
1798
1799         journal_inode = iget(sb, journal_inum);
1800         if (!journal_inode) {
1801                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1802                 return NULL;
1803         }
1804         if (!journal_inode->i_nlink) {
1805                 make_bad_inode(journal_inode);
1806                 iput(journal_inode);
1807                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1808                 return NULL;
1809         }
1810
1811         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1812                   journal_inode, journal_inode->i_size);
1813         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1814                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1815                 iput(journal_inode);
1816                 return NULL;
1817         }
1818
1819         journal = journal_init_inode(journal_inode);
1820         if (!journal) {
1821                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1822                 iput(journal_inode);
1823                 return NULL;
1824         }
1825         journal->j_private = sb;
1826         ext3_init_journal_params(sb, journal);
1827         return journal;
1828 }
1829
1830 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1831                                        dev_t j_dev)
1832 {
1833         struct buffer_head * bh;
1834         journal_t *journal;
1835         int start;
1836         int len;
1837         int hblock, blocksize;
1838         unsigned long sb_block;
1839         unsigned long offset;
1840         struct ext3_super_block * es;
1841         struct block_device *bdev;
1842
1843         bdev = ext3_blkdev_get(j_dev);
1844         if (bdev == NULL)
1845                 return NULL;
1846
1847         if (bd_claim(bdev, sb)) {
1848                 printk(KERN_ERR
1849                         "EXT3: failed to claim external journal device.\n");
1850                 blkdev_put(bdev);
1851                 return NULL;
1852         }
1853
1854         blocksize = sb->s_blocksize;
1855         hblock = bdev_hardsect_size(bdev);
1856         if (blocksize < hblock) {
1857                 printk(KERN_ERR
1858                         "EXT3-fs: blocksize too small for journal device.\n");
1859                 goto out_bdev;
1860         }
1861
1862         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1863         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1864         set_blocksize(bdev, blocksize);
1865         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1866                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1867                        "external journal\n");
1868                 goto out_bdev;
1869         }
1870
1871         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1872         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1873             !(le32_to_cpu(es->s_feature_incompat) &
1874               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1875                 printk(KERN_ERR "EXT3-fs: external journal has "
1876                                         "bad superblock\n");
1877                 brelse(bh);
1878                 goto out_bdev;
1879         }
1880
1881         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1882                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1883                 brelse(bh);
1884                 goto out_bdev;
1885         }
1886
1887         len = le32_to_cpu(es->s_blocks_count);
1888         start = sb_block + 1;
1889         brelse(bh);     /* we're done with the superblock */
1890
1891         journal = journal_init_dev(bdev, sb->s_bdev,
1892                                         start, len, blocksize);
1893         if (!journal) {
1894                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1895                 goto out_bdev;
1896         }
1897         journal->j_private = sb;
1898         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1899         wait_on_buffer(journal->j_sb_buffer);
1900         if (!buffer_uptodate(journal->j_sb_buffer)) {
1901                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1902                 goto out_journal;
1903         }
1904         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1905                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1906                                         "user (unsupported) - %d\n",
1907                         be32_to_cpu(journal->j_superblock->s_nr_users));
1908                 goto out_journal;
1909         }
1910         EXT3_SB(sb)->journal_bdev = bdev;
1911         ext3_init_journal_params(sb, journal);
1912         return journal;
1913 out_journal:
1914         journal_destroy(journal);
1915 out_bdev:
1916         ext3_blkdev_put(bdev);
1917         return NULL;
1918 }
1919
1920 static int ext3_load_journal(struct super_block *sb,
1921                              struct ext3_super_block *es,
1922                              unsigned long journal_devnum)
1923 {
1924         journal_t *journal;
1925         int journal_inum = le32_to_cpu(es->s_journal_inum);
1926         dev_t journal_dev;
1927         int err = 0;
1928         int really_read_only;
1929
1930         if (journal_devnum &&
1931             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
1932                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
1933                         "numbers have changed\n");
1934                 journal_dev = new_decode_dev(journal_devnum);
1935         } else
1936                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1937
1938         really_read_only = bdev_read_only(sb->s_bdev);
1939
1940         /*
1941          * Are we loading a blank journal or performing recovery after a
1942          * crash?  For recovery, we need to check in advance whether we
1943          * can get read-write access to the device.
1944          */
1945
1946         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1947                 if (sb->s_flags & MS_RDONLY) {
1948                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1949                                         "required on readonly filesystem.\n");
1950                         if (really_read_only) {
1951                                 printk(KERN_ERR "EXT3-fs: write access "
1952                                         "unavailable, cannot proceed.\n");
1953                                 return -EROFS;
1954                         }
1955                         printk (KERN_INFO "EXT3-fs: write access will "
1956                                         "be enabled during recovery.\n");
1957                 }
1958         }
1959
1960         if (journal_inum && journal_dev) {
1961                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1962                        "and inode journals!\n");
1963                 return -EINVAL;
1964         }
1965
1966         if (journal_inum) {
1967                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1968                         return -EINVAL;
1969         } else {
1970                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1971                         return -EINVAL;
1972         }
1973
1974         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1975                 err = journal_update_format(journal);
1976                 if (err)  {
1977                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1978                         journal_destroy(journal);
1979                         return err;
1980                 }
1981         }
1982
1983         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1984                 err = journal_wipe(journal, !really_read_only);
1985         if (!err)
1986                 err = journal_load(journal);
1987
1988         if (err) {
1989                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1990                 journal_destroy(journal);
1991                 return err;
1992         }
1993
1994         EXT3_SB(sb)->s_journal = journal;
1995         ext3_clear_journal_err(sb, es);
1996
1997         if (journal_devnum &&
1998             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
1999                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2000                 sb->s_dirt = 1;
2001
2002                 /* Make sure we flush the recovery flag to disk. */
2003                 ext3_commit_super(sb, es, 1);
2004         }
2005
2006         return 0;
2007 }
2008
2009 static int ext3_create_journal(struct super_block * sb,
2010                                struct ext3_super_block * es,
2011                                int journal_inum)
2012 {
2013         journal_t *journal;
2014
2015         if (sb->s_flags & MS_RDONLY) {
2016                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2017                                 "create journal.\n");
2018                 return -EROFS;
2019         }
2020
2021         if (!(journal = ext3_get_journal(sb, journal_inum)))
2022                 return -EINVAL;
2023
2024         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
2025                journal_inum);
2026
2027         if (journal_create(journal)) {
2028                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2029                 journal_destroy(journal);
2030                 return -EIO;
2031         }
2032
2033         EXT3_SB(sb)->s_journal = journal;
2034
2035         ext3_update_dynamic_rev(sb);
2036         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2037         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2038
2039         es->s_journal_inum = cpu_to_le32(journal_inum);
2040         sb->s_dirt = 1;
2041
2042         /* Make sure we flush the recovery flag to disk. */
2043         ext3_commit_super(sb, es, 1);
2044
2045         return 0;
2046 }
2047
2048 static void ext3_commit_super (struct super_block * sb,
2049                                struct ext3_super_block * es,
2050                                int sync)
2051 {
2052         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2053
2054         if (!sbh)
2055                 return;
2056         es->s_wtime = cpu_to_le32(get_seconds());
2057         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2058         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2059         BUFFER_TRACE(sbh, "marking dirty");
2060         mark_buffer_dirty(sbh);
2061         if (sync)
2062                 sync_dirty_buffer(sbh);
2063 }
2064
2065
2066 /*
2067  * Have we just finished recovery?  If so, and if we are mounting (or
2068  * remounting) the filesystem readonly, then we will end up with a
2069  * consistent fs on disk.  Record that fact.
2070  */
2071 static void ext3_mark_recovery_complete(struct super_block * sb,
2072                                         struct ext3_super_block * es)
2073 {
2074         journal_t *journal = EXT3_SB(sb)->s_journal;
2075
2076         journal_lock_updates(journal);
2077         journal_flush(journal);
2078         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2079             sb->s_flags & MS_RDONLY) {
2080                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2081                 sb->s_dirt = 0;
2082                 ext3_commit_super(sb, es, 1);
2083         }
2084         journal_unlock_updates(journal);
2085 }
2086
2087 /*
2088  * If we are mounting (or read-write remounting) a filesystem whose journal
2089  * has recorded an error from a previous lifetime, move that error to the
2090  * main filesystem now.
2091  */
2092 static void ext3_clear_journal_err(struct super_block * sb,
2093                                    struct ext3_super_block * es)
2094 {
2095         journal_t *journal;
2096         int j_errno;
2097         const char *errstr;
2098
2099         journal = EXT3_SB(sb)->s_journal;
2100
2101         /*
2102          * Now check for any error status which may have been recorded in the
2103          * journal by a prior ext3_error() or ext3_abort()
2104          */
2105
2106         j_errno = journal_errno(journal);
2107         if (j_errno) {
2108                 char nbuf[16];
2109
2110                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2111                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2112                              "from previous mount: %s", errstr);
2113                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2114                              "filesystem check.");
2115
2116                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2117                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2118                 ext3_commit_super (sb, es, 1);
2119
2120                 journal_clear_err(journal);
2121         }
2122 }
2123
2124 /*
2125  * Force the running and committing transactions to commit,
2126  * and wait on the commit.
2127  */
2128 int ext3_force_commit(struct super_block *sb)
2129 {
2130         journal_t *journal;
2131         int ret;
2132
2133         if (sb->s_flags & MS_RDONLY)
2134                 return 0;
2135
2136         journal = EXT3_SB(sb)->s_journal;
2137         sb->s_dirt = 0;
2138         ret = ext3_journal_force_commit(journal);
2139         return ret;
2140 }
2141
2142 /*
2143  * Ext3 always journals updates to the superblock itself, so we don't
2144  * have to propagate any other updates to the superblock on disk at this
2145  * point.  Just start an async writeback to get the buffers on their way
2146  * to the disk.
2147  *
2148  * This implicitly triggers the writebehind on sync().
2149  */
2150
2151 static void ext3_write_super (struct super_block * sb)
2152 {
2153         if (mutex_trylock(&sb->s_lock) != 0)
2154                 BUG();
2155         sb->s_dirt = 0;
2156 }
2157
2158 static int ext3_sync_fs(struct super_block *sb, int wait)
2159 {
2160         tid_t target;
2161
2162         sb->s_dirt = 0;
2163         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2164                 if (wait)
2165                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2166         }
2167         return 0;
2168 }
2169
2170 /*
2171  * LVM calls this function before a (read-only) snapshot is created.  This
2172  * gives us a chance to flush the journal completely and mark the fs clean.
2173  */
2174 static void ext3_write_super_lockfs(struct super_block *sb)
2175 {
2176         sb->s_dirt = 0;
2177
2178         if (!(sb->s_flags & MS_RDONLY)) {
2179                 journal_t *journal = EXT3_SB(sb)->s_journal;
2180
2181                 /* Now we set up the journal barrier. */
2182                 journal_lock_updates(journal);
2183                 journal_flush(journal);
2184
2185                 /* Journal blocked and flushed, clear needs_recovery flag. */
2186                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2187                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2188         }
2189 }
2190
2191 /*
2192  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2193  * flag here, even though the filesystem is not technically dirty yet.
2194  */
2195 static void ext3_unlockfs(struct super_block *sb)
2196 {
2197         if (!(sb->s_flags & MS_RDONLY)) {
2198                 lock_super(sb);
2199                 /* Reser the needs_recovery flag before the fs is unlocked. */
2200                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2201                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2202                 unlock_super(sb);
2203                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2204         }
2205 }
2206
2207 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2208 {
2209         struct ext3_super_block * es;
2210         struct ext3_sb_info *sbi = EXT3_SB(sb);
2211         unsigned long n_blocks_count = 0;
2212         unsigned long old_sb_flags;
2213         struct ext3_mount_options old_opts;
2214         int err;
2215 #ifdef CONFIG_QUOTA
2216         int i;
2217 #endif
2218
2219         /* Store the original options */
2220         old_sb_flags = sb->s_flags;
2221         old_opts.s_mount_opt = sbi->s_mount_opt;
2222         old_opts.s_resuid = sbi->s_resuid;
2223         old_opts.s_resgid = sbi->s_resgid;
2224         old_opts.s_commit_interval = sbi->s_commit_interval;
2225 #ifdef CONFIG_QUOTA
2226         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2227         for (i = 0; i < MAXQUOTAS; i++)
2228                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2229 #endif
2230
2231         /*
2232          * Allow the "check" option to be passed as a remount option.
2233          */
2234         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2235                 err = -EINVAL;
2236                 goto restore_opts;
2237         }
2238
2239         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2240                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2241
2242         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2243                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2244
2245         es = sbi->s_es;
2246
2247         ext3_init_journal_params(sb, sbi->s_journal);
2248
2249         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2250                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2251                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2252                         err = -EROFS;
2253                         goto restore_opts;
2254                 }
2255
2256                 if (*flags & MS_RDONLY) {
2257                         /*
2258                          * First of all, the unconditional stuff we have to do
2259                          * to disable replay of the journal when we next remount
2260                          */
2261                         sb->s_flags |= MS_RDONLY;
2262
2263                         /*
2264                          * OK, test if we are remounting a valid rw partition
2265                          * readonly, and if so set the rdonly flag and then
2266                          * mark the partition as valid again.
2267                          */
2268                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2269                             (sbi->s_mount_state & EXT3_VALID_FS))
2270                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2271
2272                         ext3_mark_recovery_complete(sb, es);
2273                 } else {
2274                         __le32 ret;
2275                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2276                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2277                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2278                                        "remount RDWR because of unsupported "
2279                                        "optional features (%x).\n",
2280                                        sb->s_id, le32_to_cpu(ret));
2281                                 err = -EROFS;
2282                                 goto restore_opts;
2283                         }
2284                         /*
2285                          * Mounting a RDONLY partition read-write, so reread
2286                          * and store the current valid flag.  (It may have
2287                          * been changed by e2fsck since we originally mounted
2288                          * the partition.)
2289                          */
2290                         ext3_clear_journal_err(sb, es);
2291                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2292                         if ((ret = ext3_group_extend(sb, es, n_blocks_count))) {
2293                                 err = ret;
2294                                 goto restore_opts;
2295                         }
2296                         if (!ext3_setup_super (sb, es, 0))
2297                                 sb->s_flags &= ~MS_RDONLY;
2298                 }
2299         }
2300 #ifdef CONFIG_QUOTA
2301         /* Release old quota file names */
2302         for (i = 0; i < MAXQUOTAS; i++)
2303                 if (old_opts.s_qf_names[i] &&
2304                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2305                         kfree(old_opts.s_qf_names[i]);
2306 #endif
2307         return 0;
2308 restore_opts:
2309         sb->s_flags = old_sb_flags;
2310         sbi->s_mount_opt = old_opts.s_mount_opt;
2311         sbi->s_resuid = old_opts.s_resuid;
2312         sbi->s_resgid = old_opts.s_resgid;
2313         sbi->s_commit_interval = old_opts.s_commit_interval;
2314 #ifdef CONFIG_QUOTA
2315         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2316         for (i = 0; i < MAXQUOTAS; i++) {
2317                 if (sbi->s_qf_names[i] &&
2318                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2319                         kfree(sbi->s_qf_names[i]);
2320                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2321         }
2322 #endif
2323         return err;
2324 }
2325
2326 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2327 {
2328         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2329         unsigned long overhead;
2330         int i;
2331
2332         if (test_opt (sb, MINIX_DF))
2333                 overhead = 0;
2334         else {
2335                 unsigned long ngroups;
2336                 ngroups = EXT3_SB(sb)->s_groups_count;
2337                 smp_rmb();
2338
2339                 /*
2340                  * Compute the overhead (FS structures)
2341                  */
2342
2343                 /*
2344                  * All of the blocks before first_data_block are
2345                  * overhead
2346                  */
2347                 overhead = le32_to_cpu(es->s_first_data_block);
2348
2349                 /*
2350                  * Add the overhead attributed to the superblock and
2351                  * block group descriptors.  If the sparse superblocks
2352                  * feature is turned on, then not all groups have this.
2353                  */
2354                 for (i = 0; i < ngroups; i++) {
2355                         overhead += ext3_bg_has_super(sb, i) +
2356                                 ext3_bg_num_gdb(sb, i);
2357                         cond_resched();
2358                 }
2359
2360                 /*
2361                  * Every block group has an inode bitmap, a block
2362                  * bitmap, and an inode table.
2363                  */
2364                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2365         }
2366
2367         buf->f_type = EXT3_SUPER_MAGIC;
2368         buf->f_bsize = sb->s_blocksize;
2369         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2370         buf->f_bfree = ext3_count_free_blocks (sb);
2371         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2372         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2373                 buf->f_bavail = 0;
2374         buf->f_files = le32_to_cpu(es->s_inodes_count);
2375         buf->f_ffree = ext3_count_free_inodes (sb);
2376         buf->f_namelen = EXT3_NAME_LEN;
2377         return 0;
2378 }
2379
2380 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2381  * is locked for write. Otherwise the are possible deadlocks:
2382  * Process 1                         Process 2
2383  * ext3_create()                     quota_sync()
2384  *   journal_start()                   write_dquot()
2385  *   DQUOT_INIT()                        down(dqio_mutex)
2386  *     down(dqio_mutex)                    journal_start()
2387  *
2388  */
2389
2390 #ifdef CONFIG_QUOTA
2391
2392 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2393 {
2394         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2395 }
2396
2397 static int ext3_dquot_initialize(struct inode *inode, int type)
2398 {
2399         handle_t *handle;
2400         int ret, err;
2401
2402         /* We may create quota structure so we need to reserve enough blocks */
2403         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2404         if (IS_ERR(handle))
2405                 return PTR_ERR(handle);
2406         ret = dquot_initialize(inode, type);
2407         err = ext3_journal_stop(handle);
2408         if (!ret)
2409                 ret = err;
2410         return ret;
2411 }
2412
2413 static int ext3_dquot_drop(struct inode *inode)
2414 {
2415         handle_t *handle;
2416         int ret, err;
2417
2418         /* We may delete quota structure so we need to reserve enough blocks */
2419         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2420         if (IS_ERR(handle))
2421                 return PTR_ERR(handle);
2422         ret = dquot_drop(inode);
2423         err = ext3_journal_stop(handle);
2424         if (!ret)
2425                 ret = err;
2426         return ret;
2427 }
2428
2429 static int ext3_write_dquot(struct dquot *dquot)
2430 {
2431         int ret, err;
2432         handle_t *handle;
2433         struct inode *inode;
2434
2435         inode = dquot_to_inode(dquot);
2436         handle = ext3_journal_start(inode,
2437                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2438         if (IS_ERR(handle))
2439                 return PTR_ERR(handle);
2440         ret = dquot_commit(dquot);
2441         err = ext3_journal_stop(handle);
2442         if (!ret)
2443                 ret = err;
2444         return ret;
2445 }
2446
2447 static int ext3_acquire_dquot(struct dquot *dquot)
2448 {
2449         int ret, err;
2450         handle_t *handle;
2451
2452         handle = ext3_journal_start(dquot_to_inode(dquot),
2453                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2454         if (IS_ERR(handle))
2455                 return PTR_ERR(handle);
2456         ret = dquot_acquire(dquot);
2457         err = ext3_journal_stop(handle);
2458         if (!ret)
2459                 ret = err;
2460         return ret;
2461 }
2462
2463 static int ext3_release_dquot(struct dquot *dquot)
2464 {
2465         int ret, err;
2466         handle_t *handle;
2467
2468         handle = ext3_journal_start(dquot_to_inode(dquot),
2469                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2470         if (IS_ERR(handle))
2471                 return PTR_ERR(handle);
2472         ret = dquot_release(dquot);
2473         err = ext3_journal_stop(handle);
2474         if (!ret)
2475                 ret = err;
2476         return ret;
2477 }
2478
2479 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2480 {
2481         /* Are we journalling quotas? */
2482         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2483             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2484                 dquot_mark_dquot_dirty(dquot);
2485                 return ext3_write_dquot(dquot);
2486         } else {
2487                 return dquot_mark_dquot_dirty(dquot);
2488         }
2489 }
2490
2491 static int ext3_write_info(struct super_block *sb, int type)
2492 {
2493         int ret, err;
2494         handle_t *handle;
2495
2496         /* Data block + inode block */
2497         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2498         if (IS_ERR(handle))
2499                 return PTR_ERR(handle);
2500         ret = dquot_commit_info(sb, type);
2501         err = ext3_journal_stop(handle);
2502         if (!ret)
2503                 ret = err;
2504         return ret;
2505 }
2506
2507 /*
2508  * Turn on quotas during mount time - we need to find
2509  * the quota file and such...
2510  */
2511 static int ext3_quota_on_mount(struct super_block *sb, int type)
2512 {
2513         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2514                         EXT3_SB(sb)->s_jquota_fmt, type);
2515 }
2516
2517 /*
2518  * Standard function to be called on quota_on
2519  */
2520 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2521                          char *path)
2522 {
2523         int err;
2524         struct nameidata nd;
2525
2526         if (!test_opt(sb, QUOTA))
2527                 return -EINVAL;
2528         /* Not journalling quota? */
2529         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2530             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2531                 return vfs_quota_on(sb, type, format_id, path);
2532         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2533         if (err)
2534                 return err;
2535         /* Quotafile not on the same filesystem? */
2536         if (nd.mnt->mnt_sb != sb) {
2537                 path_release(&nd);
2538                 return -EXDEV;
2539         }
2540         /* Quotafile not of fs root? */
2541         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2542                 printk(KERN_WARNING
2543                         "EXT3-fs: Quota file not on filesystem root. "
2544                         "Journalled quota will not work.\n");
2545         path_release(&nd);
2546         return vfs_quota_on(sb, type, format_id, path);
2547 }
2548
2549 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2550  * acquiring the locks... As quota files are never truncated and quota code
2551  * itself serializes the operations (and noone else should touch the files)
2552  * we don't have to be afraid of races */
2553 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2554                                size_t len, loff_t off)
2555 {
2556         struct inode *inode = sb_dqopt(sb)->files[type];
2557         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2558         int err = 0;
2559         int offset = off & (sb->s_blocksize - 1);
2560         int tocopy;
2561         size_t toread;
2562         struct buffer_head *bh;
2563         loff_t i_size = i_size_read(inode);
2564
2565         if (off > i_size)
2566                 return 0;
2567         if (off+len > i_size)
2568                 len = i_size-off;
2569         toread = len;
2570         while (toread > 0) {
2571                 tocopy = sb->s_blocksize - offset < toread ?
2572                                 sb->s_blocksize - offset : toread;
2573                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2574                 if (err)
2575                         return err;
2576                 if (!bh)        /* A hole? */
2577                         memset(data, 0, tocopy);
2578                 else
2579                         memcpy(data, bh->b_data+offset, tocopy);
2580                 brelse(bh);
2581                 offset = 0;
2582                 toread -= tocopy;
2583                 data += tocopy;
2584                 blk++;
2585         }
2586         return len;
2587 }
2588
2589 /* Write to quotafile (we know the transaction is already started and has
2590  * enough credits) */
2591 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2592                                 const char *data, size_t len, loff_t off)
2593 {
2594         struct inode *inode = sb_dqopt(sb)->files[type];
2595         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2596         int err = 0;
2597         int offset = off & (sb->s_blocksize - 1);
2598         int tocopy;
2599         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2600         size_t towrite = len;
2601         struct buffer_head *bh;
2602         handle_t *handle = journal_current_handle();
2603
2604         mutex_lock(&inode->i_mutex);
2605         while (towrite > 0) {
2606                 tocopy = sb->s_blocksize - offset < towrite ?
2607                                 sb->s_blocksize - offset : towrite;
2608                 bh = ext3_bread(handle, inode, blk, 1, &err);
2609                 if (!bh)
2610                         goto out;
2611                 if (journal_quota) {
2612                         err = ext3_journal_get_write_access(handle, bh);
2613                         if (err) {
2614                                 brelse(bh);
2615                                 goto out;
2616                         }
2617                 }
2618                 lock_buffer(bh);
2619                 memcpy(bh->b_data+offset, data, tocopy);
2620                 flush_dcache_page(bh->b_page);
2621                 unlock_buffer(bh);
2622                 if (journal_quota)
2623                         err = ext3_journal_dirty_metadata(handle, bh);
2624                 else {
2625                         /* Always do at least ordered writes for quotas */
2626                         err = ext3_journal_dirty_data(handle, bh);
2627                         mark_buffer_dirty(bh);
2628                 }
2629                 brelse(bh);
2630                 if (err)
2631                         goto out;
2632                 offset = 0;
2633                 towrite -= tocopy;
2634                 data += tocopy;
2635                 blk++;
2636         }
2637 out:
2638         if (len == towrite)
2639                 return err;
2640         if (inode->i_size < off+len-towrite) {
2641                 i_size_write(inode, off+len-towrite);
2642                 EXT3_I(inode)->i_disksize = inode->i_size;
2643         }
2644         inode->i_version++;
2645         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2646         ext3_mark_inode_dirty(handle, inode);
2647         mutex_unlock(&inode->i_mutex);
2648         return len - towrite;
2649 }
2650
2651 #endif
2652
2653 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2654         int flags, const char *dev_name, void *data)
2655 {
2656         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2657 }
2658
2659 static struct file_system_type ext3_fs_type = {
2660         .owner          = THIS_MODULE,
2661         .name           = "ext3",
2662         .get_sb         = ext3_get_sb,
2663         .kill_sb        = kill_block_super,
2664         .fs_flags       = FS_REQUIRES_DEV,
2665 };
2666
2667 static int __init init_ext3_fs(void)
2668 {
2669         int err = init_ext3_xattr();
2670         if (err)
2671                 return err;
2672         err = init_inodecache();
2673         if (err)
2674                 goto out1;
2675         err = register_filesystem(&ext3_fs_type);
2676         if (err)
2677                 goto out;
2678         return 0;
2679 out:
2680         destroy_inodecache();
2681 out1:
2682         exit_ext3_xattr();
2683         return err;
2684 }
2685
2686 static void __exit exit_ext3_fs(void)
2687 {
2688         unregister_filesystem(&ext3_fs_type);
2689         destroy_inodecache();
2690         exit_ext3_xattr();
2691 }
2692
2693 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2694 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2695 MODULE_LICENSE("GPL");
2696 module_init(init_ext3_fs)
2697 module_exit(exit_ext3_fs)