Merge branch 'for_paulus' of master.kernel.org:/pub/scm/linux/kernel/git/galak/powerpc
[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 dentry * dentry, 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                 mutex_init(&ei->truncate_mutex);
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                                                 SLAB_MEM_SPREAD),
486                                              init_once, NULL);
487         if (ext3_inode_cachep == NULL)
488                 return -ENOMEM;
489         return 0;
490 }
491
492 static void destroy_inodecache(void)
493 {
494         if (kmem_cache_destroy(ext3_inode_cachep))
495                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
496 }
497
498 static void ext3_clear_inode(struct inode *inode)
499 {
500         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
501 #ifdef CONFIG_EXT3_FS_POSIX_ACL
502         if (EXT3_I(inode)->i_acl &&
503                         EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
504                 posix_acl_release(EXT3_I(inode)->i_acl);
505                 EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
506         }
507         if (EXT3_I(inode)->i_default_acl &&
508                         EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
509                 posix_acl_release(EXT3_I(inode)->i_default_acl);
510                 EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
511         }
512 #endif
513         ext3_discard_reservation(inode);
514         EXT3_I(inode)->i_block_alloc_info = NULL;
515         if (unlikely(rsv))
516                 kfree(rsv);
517 }
518
519 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
520 {
521 #if defined(CONFIG_QUOTA)
522         struct ext3_sb_info *sbi = EXT3_SB(sb);
523
524         if (sbi->s_jquota_fmt)
525                 seq_printf(seq, ",jqfmt=%s",
526                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
527
528         if (sbi->s_qf_names[USRQUOTA])
529                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
530
531         if (sbi->s_qf_names[GRPQUOTA])
532                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
533
534         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
535                 seq_puts(seq, ",usrquota");
536
537         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
538                 seq_puts(seq, ",grpquota");
539 #endif
540 }
541
542 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
543 {
544         struct super_block *sb = vfs->mnt_sb;
545
546         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
547                 seq_puts(seq, ",data=journal");
548         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
549                 seq_puts(seq, ",data=ordered");
550         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
551                 seq_puts(seq, ",data=writeback");
552
553         ext3_show_quota_options(seq, sb);
554
555         return 0;
556 }
557
558 #ifdef CONFIG_QUOTA
559 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
560 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
561
562 static int ext3_dquot_initialize(struct inode *inode, int type);
563 static int ext3_dquot_drop(struct inode *inode);
564 static int ext3_write_dquot(struct dquot *dquot);
565 static int ext3_acquire_dquot(struct dquot *dquot);
566 static int ext3_release_dquot(struct dquot *dquot);
567 static int ext3_mark_dquot_dirty(struct dquot *dquot);
568 static int ext3_write_info(struct super_block *sb, int type);
569 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
570 static int ext3_quota_on_mount(struct super_block *sb, int type);
571 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
572                                size_t len, loff_t off);
573 static ssize_t ext3_quota_write(struct super_block *sb, int type,
574                                 const char *data, size_t len, loff_t off);
575
576 static struct dquot_operations ext3_quota_operations = {
577         .initialize     = ext3_dquot_initialize,
578         .drop           = ext3_dquot_drop,
579         .alloc_space    = dquot_alloc_space,
580         .alloc_inode    = dquot_alloc_inode,
581         .free_space     = dquot_free_space,
582         .free_inode     = dquot_free_inode,
583         .transfer       = dquot_transfer,
584         .write_dquot    = ext3_write_dquot,
585         .acquire_dquot  = ext3_acquire_dquot,
586         .release_dquot  = ext3_release_dquot,
587         .mark_dirty     = ext3_mark_dquot_dirty,
588         .write_info     = ext3_write_info
589 };
590
591 static struct quotactl_ops ext3_qctl_operations = {
592         .quota_on       = ext3_quota_on,
593         .quota_off      = vfs_quota_off,
594         .quota_sync     = vfs_quota_sync,
595         .get_info       = vfs_get_dqinfo,
596         .set_info       = vfs_set_dqinfo,
597         .get_dqblk      = vfs_get_dqblk,
598         .set_dqblk      = vfs_set_dqblk
599 };
600 #endif
601
602 static struct super_operations ext3_sops = {
603         .alloc_inode    = ext3_alloc_inode,
604         .destroy_inode  = ext3_destroy_inode,
605         .read_inode     = ext3_read_inode,
606         .write_inode    = ext3_write_inode,
607         .dirty_inode    = ext3_dirty_inode,
608         .delete_inode   = ext3_delete_inode,
609         .put_super      = ext3_put_super,
610         .write_super    = ext3_write_super,
611         .sync_fs        = ext3_sync_fs,
612         .write_super_lockfs = ext3_write_super_lockfs,
613         .unlockfs       = ext3_unlockfs,
614         .statfs         = ext3_statfs,
615         .remount_fs     = ext3_remount,
616         .clear_inode    = ext3_clear_inode,
617         .show_options   = ext3_show_options,
618 #ifdef CONFIG_QUOTA
619         .quota_read     = ext3_quota_read,
620         .quota_write    = ext3_quota_write,
621 #endif
622 };
623
624 static struct export_operations ext3_export_ops = {
625         .get_parent = ext3_get_parent,
626 };
627
628 enum {
629         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
630         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
631         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
632         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
633         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
634         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
635         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
636         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
637         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
638         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
639         Opt_grpquota
640 };
641
642 static match_table_t tokens = {
643         {Opt_bsd_df, "bsddf"},
644         {Opt_minix_df, "minixdf"},
645         {Opt_grpid, "grpid"},
646         {Opt_grpid, "bsdgroups"},
647         {Opt_nogrpid, "nogrpid"},
648         {Opt_nogrpid, "sysvgroups"},
649         {Opt_resgid, "resgid=%u"},
650         {Opt_resuid, "resuid=%u"},
651         {Opt_sb, "sb=%u"},
652         {Opt_err_cont, "errors=continue"},
653         {Opt_err_panic, "errors=panic"},
654         {Opt_err_ro, "errors=remount-ro"},
655         {Opt_nouid32, "nouid32"},
656         {Opt_nocheck, "nocheck"},
657         {Opt_nocheck, "check=none"},
658         {Opt_debug, "debug"},
659         {Opt_oldalloc, "oldalloc"},
660         {Opt_orlov, "orlov"},
661         {Opt_user_xattr, "user_xattr"},
662         {Opt_nouser_xattr, "nouser_xattr"},
663         {Opt_acl, "acl"},
664         {Opt_noacl, "noacl"},
665         {Opt_reservation, "reservation"},
666         {Opt_noreservation, "noreservation"},
667         {Opt_noload, "noload"},
668         {Opt_nobh, "nobh"},
669         {Opt_bh, "bh"},
670         {Opt_commit, "commit=%u"},
671         {Opt_journal_update, "journal=update"},
672         {Opt_journal_inum, "journal=%u"},
673         {Opt_journal_dev, "journal_dev=%u"},
674         {Opt_abort, "abort"},
675         {Opt_data_journal, "data=journal"},
676         {Opt_data_ordered, "data=ordered"},
677         {Opt_data_writeback, "data=writeback"},
678         {Opt_offusrjquota, "usrjquota="},
679         {Opt_usrjquota, "usrjquota=%s"},
680         {Opt_offgrpjquota, "grpjquota="},
681         {Opt_grpjquota, "grpjquota=%s"},
682         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
683         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
684         {Opt_grpquota, "grpquota"},
685         {Opt_noquota, "noquota"},
686         {Opt_quota, "quota"},
687         {Opt_usrquota, "usrquota"},
688         {Opt_barrier, "barrier=%u"},
689         {Opt_err, NULL},
690         {Opt_resize, "resize"},
691 };
692
693 static ext3_fsblk_t get_sb_block(void **data)
694 {
695         ext3_fsblk_t    sb_block;
696         char            *options = (char *) *data;
697
698         if (!options || strncmp(options, "sb=", 3) != 0)
699                 return 1;       /* Default location */
700         options += 3;
701         /*todo: use simple_strtoll with >32bit ext3 */
702         sb_block = simple_strtoul(options, &options, 0);
703         if (*options && *options != ',') {
704                 printk("EXT3-fs: Invalid sb specification: %s\n",
705                        (char *) *data);
706                 return 1;
707         }
708         if (*options == ',')
709                 options++;
710         *data = (void *) options;
711         return sb_block;
712 }
713
714 static int parse_options (char *options, struct super_block *sb,
715                           unsigned long *inum, unsigned long *journal_devnum,
716                           ext3_fsblk_t *n_blocks_count, int is_remount)
717 {
718         struct ext3_sb_info *sbi = EXT3_SB(sb);
719         char * p;
720         substring_t args[MAX_OPT_ARGS];
721         int data_opt = 0;
722         int option;
723 #ifdef CONFIG_QUOTA
724         int qtype;
725         char *qname;
726 #endif
727
728         if (!options)
729                 return 1;
730
731         while ((p = strsep (&options, ",")) != NULL) {
732                 int token;
733                 if (!*p)
734                         continue;
735
736                 token = match_token(p, tokens, args);
737                 switch (token) {
738                 case Opt_bsd_df:
739                         clear_opt (sbi->s_mount_opt, MINIX_DF);
740                         break;
741                 case Opt_minix_df:
742                         set_opt (sbi->s_mount_opt, MINIX_DF);
743                         break;
744                 case Opt_grpid:
745                         set_opt (sbi->s_mount_opt, GRPID);
746                         break;
747                 case Opt_nogrpid:
748                         clear_opt (sbi->s_mount_opt, GRPID);
749                         break;
750                 case Opt_resuid:
751                         if (match_int(&args[0], &option))
752                                 return 0;
753                         sbi->s_resuid = option;
754                         break;
755                 case Opt_resgid:
756                         if (match_int(&args[0], &option))
757                                 return 0;
758                         sbi->s_resgid = option;
759                         break;
760                 case Opt_sb:
761                         /* handled by get_sb_block() instead of here */
762                         /* *sb_block = match_int(&args[0]); */
763                         break;
764                 case Opt_err_panic:
765                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
766                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
767                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
768                         break;
769                 case Opt_err_ro:
770                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
771                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
772                         set_opt (sbi->s_mount_opt, ERRORS_RO);
773                         break;
774                 case Opt_err_cont:
775                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
776                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
777                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
778                         break;
779                 case Opt_nouid32:
780                         set_opt (sbi->s_mount_opt, NO_UID32);
781                         break;
782                 case Opt_nocheck:
783                         clear_opt (sbi->s_mount_opt, CHECK);
784                         break;
785                 case Opt_debug:
786                         set_opt (sbi->s_mount_opt, DEBUG);
787                         break;
788                 case Opt_oldalloc:
789                         set_opt (sbi->s_mount_opt, OLDALLOC);
790                         break;
791                 case Opt_orlov:
792                         clear_opt (sbi->s_mount_opt, OLDALLOC);
793                         break;
794 #ifdef CONFIG_EXT3_FS_XATTR
795                 case Opt_user_xattr:
796                         set_opt (sbi->s_mount_opt, XATTR_USER);
797                         break;
798                 case Opt_nouser_xattr:
799                         clear_opt (sbi->s_mount_opt, XATTR_USER);
800                         break;
801 #else
802                 case Opt_user_xattr:
803                 case Opt_nouser_xattr:
804                         printk("EXT3 (no)user_xattr options not supported\n");
805                         break;
806 #endif
807 #ifdef CONFIG_EXT3_FS_POSIX_ACL
808                 case Opt_acl:
809                         set_opt(sbi->s_mount_opt, POSIX_ACL);
810                         break;
811                 case Opt_noacl:
812                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
813                         break;
814 #else
815                 case Opt_acl:
816                 case Opt_noacl:
817                         printk("EXT3 (no)acl options not supported\n");
818                         break;
819 #endif
820                 case Opt_reservation:
821                         set_opt(sbi->s_mount_opt, RESERVATION);
822                         break;
823                 case Opt_noreservation:
824                         clear_opt(sbi->s_mount_opt, RESERVATION);
825                         break;
826                 case Opt_journal_update:
827                         /* @@@ FIXME */
828                         /* Eventually we will want to be able to create
829                            a journal file here.  For now, only allow the
830                            user to specify an existing inode to be the
831                            journal file. */
832                         if (is_remount) {
833                                 printk(KERN_ERR "EXT3-fs: cannot specify "
834                                        "journal on remount\n");
835                                 return 0;
836                         }
837                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
838                         break;
839                 case Opt_journal_inum:
840                         if (is_remount) {
841                                 printk(KERN_ERR "EXT3-fs: cannot specify "
842                                        "journal on remount\n");
843                                 return 0;
844                         }
845                         if (match_int(&args[0], &option))
846                                 return 0;
847                         *inum = option;
848                         break;
849                 case Opt_journal_dev:
850                         if (is_remount) {
851                                 printk(KERN_ERR "EXT3-fs: cannot specify "
852                                        "journal on remount\n");
853                                 return 0;
854                         }
855                         if (match_int(&args[0], &option))
856                                 return 0;
857                         *journal_devnum = option;
858                         break;
859                 case Opt_noload:
860                         set_opt (sbi->s_mount_opt, NOLOAD);
861                         break;
862                 case Opt_commit:
863                         if (match_int(&args[0], &option))
864                                 return 0;
865                         if (option < 0)
866                                 return 0;
867                         if (option == 0)
868                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
869                         sbi->s_commit_interval = HZ * option;
870                         break;
871                 case Opt_data_journal:
872                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
873                         goto datacheck;
874                 case Opt_data_ordered:
875                         data_opt = EXT3_MOUNT_ORDERED_DATA;
876                         goto datacheck;
877                 case Opt_data_writeback:
878                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
879                 datacheck:
880                         if (is_remount) {
881                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
882                                                 != data_opt) {
883                                         printk(KERN_ERR
884                                                 "EXT3-fs: cannot change data "
885                                                 "mode on remount\n");
886                                         return 0;
887                                 }
888                         } else {
889                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
890                                 sbi->s_mount_opt |= data_opt;
891                         }
892                         break;
893 #ifdef CONFIG_QUOTA
894                 case Opt_usrjquota:
895                         qtype = USRQUOTA;
896                         goto set_qf_name;
897                 case Opt_grpjquota:
898                         qtype = GRPQUOTA;
899 set_qf_name:
900                         if (sb_any_quota_enabled(sb)) {
901                                 printk(KERN_ERR
902                                         "EXT3-fs: Cannot change journalled "
903                                         "quota options when quota turned on.\n");
904                                 return 0;
905                         }
906                         qname = match_strdup(&args[0]);
907                         if (!qname) {
908                                 printk(KERN_ERR
909                                         "EXT3-fs: not enough memory for "
910                                         "storing quotafile name.\n");
911                                 return 0;
912                         }
913                         if (sbi->s_qf_names[qtype] &&
914                             strcmp(sbi->s_qf_names[qtype], qname)) {
915                                 printk(KERN_ERR
916                                         "EXT3-fs: %s quota file already "
917                                         "specified.\n", QTYPE2NAME(qtype));
918                                 kfree(qname);
919                                 return 0;
920                         }
921                         sbi->s_qf_names[qtype] = qname;
922                         if (strchr(sbi->s_qf_names[qtype], '/')) {
923                                 printk(KERN_ERR
924                                         "EXT3-fs: quotafile must be on "
925                                         "filesystem root.\n");
926                                 kfree(sbi->s_qf_names[qtype]);
927                                 sbi->s_qf_names[qtype] = NULL;
928                                 return 0;
929                         }
930                         set_opt(sbi->s_mount_opt, QUOTA);
931                         break;
932                 case Opt_offusrjquota:
933                         qtype = USRQUOTA;
934                         goto clear_qf_name;
935                 case Opt_offgrpjquota:
936                         qtype = GRPQUOTA;
937 clear_qf_name:
938                         if (sb_any_quota_enabled(sb)) {
939                                 printk(KERN_ERR "EXT3-fs: Cannot change "
940                                         "journalled quota options when "
941                                         "quota turned on.\n");
942                                 return 0;
943                         }
944                         /*
945                          * The space will be released later when all options
946                          * are confirmed to be correct
947                          */
948                         sbi->s_qf_names[qtype] = NULL;
949                         break;
950                 case Opt_jqfmt_vfsold:
951                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
952                         break;
953                 case Opt_jqfmt_vfsv0:
954                         sbi->s_jquota_fmt = QFMT_VFS_V0;
955                         break;
956                 case Opt_quota:
957                 case Opt_usrquota:
958                         set_opt(sbi->s_mount_opt, QUOTA);
959                         set_opt(sbi->s_mount_opt, USRQUOTA);
960                         break;
961                 case Opt_grpquota:
962                         set_opt(sbi->s_mount_opt, QUOTA);
963                         set_opt(sbi->s_mount_opt, GRPQUOTA);
964                         break;
965                 case Opt_noquota:
966                         if (sb_any_quota_enabled(sb)) {
967                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
968                                         "options when quota turned on.\n");
969                                 return 0;
970                         }
971                         clear_opt(sbi->s_mount_opt, QUOTA);
972                         clear_opt(sbi->s_mount_opt, USRQUOTA);
973                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
974                         break;
975 #else
976                 case Opt_quota:
977                 case Opt_usrquota:
978                 case Opt_grpquota:
979                 case Opt_usrjquota:
980                 case Opt_grpjquota:
981                 case Opt_offusrjquota:
982                 case Opt_offgrpjquota:
983                 case Opt_jqfmt_vfsold:
984                 case Opt_jqfmt_vfsv0:
985                         printk(KERN_ERR
986                                 "EXT3-fs: journalled quota options not "
987                                 "supported.\n");
988                         break;
989                 case Opt_noquota:
990                         break;
991 #endif
992                 case Opt_abort:
993                         set_opt(sbi->s_mount_opt, ABORT);
994                         break;
995                 case Opt_barrier:
996                         if (match_int(&args[0], &option))
997                                 return 0;
998                         if (option)
999                                 set_opt(sbi->s_mount_opt, BARRIER);
1000                         else
1001                                 clear_opt(sbi->s_mount_opt, BARRIER);
1002                         break;
1003                 case Opt_ignore:
1004                         break;
1005                 case Opt_resize:
1006                         if (!is_remount) {
1007                                 printk("EXT3-fs: resize option only available "
1008                                         "for remount\n");
1009                                 return 0;
1010                         }
1011                         if (match_int(&args[0], &option) != 0)
1012                                 return 0;
1013                         *n_blocks_count = option;
1014                         break;
1015                 case Opt_nobh:
1016                         set_opt(sbi->s_mount_opt, NOBH);
1017                         break;
1018                 case Opt_bh:
1019                         clear_opt(sbi->s_mount_opt, NOBH);
1020                         break;
1021                 default:
1022                         printk (KERN_ERR
1023                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1024                                 "or missing value\n", p);
1025                         return 0;
1026                 }
1027         }
1028 #ifdef CONFIG_QUOTA
1029         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1030                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1031                      sbi->s_qf_names[USRQUOTA])
1032                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1033
1034                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1035                      sbi->s_qf_names[GRPQUOTA])
1036                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1037
1038                 if ((sbi->s_qf_names[USRQUOTA] &&
1039                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1040                     (sbi->s_qf_names[GRPQUOTA] &&
1041                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1042                         printk(KERN_ERR "EXT3-fs: old and new quota "
1043                                         "format mixing.\n");
1044                         return 0;
1045                 }
1046
1047                 if (!sbi->s_jquota_fmt) {
1048                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1049                                         "not specified.\n");
1050                         return 0;
1051                 }
1052         } else {
1053                 if (sbi->s_jquota_fmt) {
1054                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1055                                         "specified with no journalling "
1056                                         "enabled.\n");
1057                         return 0;
1058                 }
1059         }
1060 #endif
1061         return 1;
1062 }
1063
1064 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1065                             int read_only)
1066 {
1067         struct ext3_sb_info *sbi = EXT3_SB(sb);
1068         int res = 0;
1069
1070         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1071                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1072                         "forcing read-only mode\n");
1073                 res = MS_RDONLY;
1074         }
1075         if (read_only)
1076                 return res;
1077         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1078                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1079                         "running e2fsck is recommended\n");
1080         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1081                 printk (KERN_WARNING
1082                         "EXT3-fs warning: mounting fs with errors, "
1083                         "running e2fsck is recommended\n");
1084         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1085                  le16_to_cpu(es->s_mnt_count) >=
1086                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1087                 printk (KERN_WARNING
1088                         "EXT3-fs warning: maximal mount count reached, "
1089                         "running e2fsck is recommended\n");
1090         else if (le32_to_cpu(es->s_checkinterval) &&
1091                 (le32_to_cpu(es->s_lastcheck) +
1092                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1093                 printk (KERN_WARNING
1094                         "EXT3-fs warning: checktime reached, "
1095                         "running e2fsck is recommended\n");
1096 #if 0
1097                 /* @@@ We _will_ want to clear the valid bit if we find
1098                    inconsistencies, to force a fsck at reboot.  But for
1099                    a plain journaled filesystem we can keep it set as
1100                    valid forever! :) */
1101         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1102 #endif
1103         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1104                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1105         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1106         es->s_mtime = cpu_to_le32(get_seconds());
1107         ext3_update_dynamic_rev(sb);
1108         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1109
1110         ext3_commit_super(sb, es, 1);
1111         if (test_opt(sb, DEBUG))
1112                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1113                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1114                         sb->s_blocksize,
1115                         sbi->s_groups_count,
1116                         EXT3_BLOCKS_PER_GROUP(sb),
1117                         EXT3_INODES_PER_GROUP(sb),
1118                         sbi->s_mount_opt);
1119
1120         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1121         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1122                 char b[BDEVNAME_SIZE];
1123
1124                 printk("external journal on %s\n",
1125                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1126         } else {
1127                 printk("internal journal\n");
1128         }
1129         return res;
1130 }
1131
1132 /* Called at mount-time, super-block is locked */
1133 static int ext3_check_descriptors (struct super_block * sb)
1134 {
1135         struct ext3_sb_info *sbi = EXT3_SB(sb);
1136         ext3_fsblk_t block = le32_to_cpu(sbi->s_es->s_first_data_block);
1137         struct ext3_group_desc * gdp = NULL;
1138         int desc_block = 0;
1139         int i;
1140
1141         ext3_debug ("Checking group descriptors");
1142
1143         for (i = 0; i < sbi->s_groups_count; i++)
1144         {
1145                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1146                         gdp = (struct ext3_group_desc *)
1147                                         sbi->s_group_desc[desc_block++]->b_data;
1148                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1149                     le32_to_cpu(gdp->bg_block_bitmap) >=
1150                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1151                 {
1152                         ext3_error (sb, "ext3_check_descriptors",
1153                                     "Block bitmap for group %d"
1154                                     " not in group (block %lu)!",
1155                                     i, (unsigned long)
1156                                         le32_to_cpu(gdp->bg_block_bitmap));
1157                         return 0;
1158                 }
1159                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1160                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1161                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1162                 {
1163                         ext3_error (sb, "ext3_check_descriptors",
1164                                     "Inode bitmap for group %d"
1165                                     " not in group (block %lu)!",
1166                                     i, (unsigned long)
1167                                         le32_to_cpu(gdp->bg_inode_bitmap));
1168                         return 0;
1169                 }
1170                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1171                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1172                     block + EXT3_BLOCKS_PER_GROUP(sb))
1173                 {
1174                         ext3_error (sb, "ext3_check_descriptors",
1175                                     "Inode table for group %d"
1176                                     " not in group (block %lu)!",
1177                                     i, (unsigned long)
1178                                         le32_to_cpu(gdp->bg_inode_table));
1179                         return 0;
1180                 }
1181                 block += EXT3_BLOCKS_PER_GROUP(sb);
1182                 gdp++;
1183         }
1184
1185         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1186         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1187         return 1;
1188 }
1189
1190
1191 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1192  * the superblock) which were deleted from all directories, but held open by
1193  * a process at the time of a crash.  We walk the list and try to delete these
1194  * inodes at recovery time (only with a read-write filesystem).
1195  *
1196  * In order to keep the orphan inode chain consistent during traversal (in
1197  * case of crash during recovery), we link each inode into the superblock
1198  * orphan list_head and handle it the same way as an inode deletion during
1199  * normal operation (which journals the operations for us).
1200  *
1201  * We only do an iget() and an iput() on each inode, which is very safe if we
1202  * accidentally point at an in-use or already deleted inode.  The worst that
1203  * can happen in this case is that we get a "bit already cleared" message from
1204  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1205  * e2fsck was run on this filesystem, and it must have already done the orphan
1206  * inode cleanup for us, so we can safely abort without any further action.
1207  */
1208 static void ext3_orphan_cleanup (struct super_block * sb,
1209                                  struct ext3_super_block * es)
1210 {
1211         unsigned int s_flags = sb->s_flags;
1212         int nr_orphans = 0, nr_truncates = 0;
1213 #ifdef CONFIG_QUOTA
1214         int i;
1215 #endif
1216         if (!es->s_last_orphan) {
1217                 jbd_debug(4, "no orphan inodes to clean up\n");
1218                 return;
1219         }
1220
1221         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1222                 if (es->s_last_orphan)
1223                         jbd_debug(1, "Errors on filesystem, "
1224                                   "clearing orphan list.\n");
1225                 es->s_last_orphan = 0;
1226                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1227                 return;
1228         }
1229
1230         if (s_flags & MS_RDONLY) {
1231                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1232                        sb->s_id);
1233                 sb->s_flags &= ~MS_RDONLY;
1234         }
1235 #ifdef CONFIG_QUOTA
1236         /* Needed for iput() to work correctly and not trash data */
1237         sb->s_flags |= MS_ACTIVE;
1238         /* Turn on quotas so that they are updated correctly */
1239         for (i = 0; i < MAXQUOTAS; i++) {
1240                 if (EXT3_SB(sb)->s_qf_names[i]) {
1241                         int ret = ext3_quota_on_mount(sb, i);
1242                         if (ret < 0)
1243                                 printk(KERN_ERR
1244                                         "EXT3-fs: Cannot turn on journalled "
1245                                         "quota: error %d\n", ret);
1246                 }
1247         }
1248 #endif
1249
1250         while (es->s_last_orphan) {
1251                 struct inode *inode;
1252
1253                 if (!(inode =
1254                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1255                         es->s_last_orphan = 0;
1256                         break;
1257                 }
1258
1259                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1260                 DQUOT_INIT(inode);
1261                 if (inode->i_nlink) {
1262                         printk(KERN_DEBUG
1263                                 "%s: truncating inode %ld to %Ld bytes\n",
1264                                 __FUNCTION__, inode->i_ino, inode->i_size);
1265                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1266                                   inode->i_ino, inode->i_size);
1267                         ext3_truncate(inode);
1268                         nr_truncates++;
1269                 } else {
1270                         printk(KERN_DEBUG
1271                                 "%s: deleting unreferenced inode %ld\n",
1272                                 __FUNCTION__, inode->i_ino);
1273                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1274                                   inode->i_ino);
1275                         nr_orphans++;
1276                 }
1277                 iput(inode);  /* The delete magic happens here! */
1278         }
1279
1280 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1281
1282         if (nr_orphans)
1283                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1284                        sb->s_id, PLURAL(nr_orphans));
1285         if (nr_truncates)
1286                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1287                        sb->s_id, PLURAL(nr_truncates));
1288 #ifdef CONFIG_QUOTA
1289         /* Turn quotas off */
1290         for (i = 0; i < MAXQUOTAS; i++) {
1291                 if (sb_dqopt(sb)->files[i])
1292                         vfs_quota_off(sb, i);
1293         }
1294 #endif
1295         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1296 }
1297
1298 #define log2(n) ffz(~(n))
1299
1300 /*
1301  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1302  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1303  * We need to be 1 filesystem block less than the 2^32 sector limit.
1304  */
1305 static loff_t ext3_max_size(int bits)
1306 {
1307         loff_t res = EXT3_NDIR_BLOCKS;
1308         /* This constant is calculated to be the largest file size for a
1309          * dense, 4k-blocksize file such that the total number of
1310          * sectors in the file, including data and all indirect blocks,
1311          * does not exceed 2^32. */
1312         const loff_t upper_limit = 0x1ff7fffd000LL;
1313
1314         res += 1LL << (bits-2);
1315         res += 1LL << (2*(bits-2));
1316         res += 1LL << (3*(bits-2));
1317         res <<= bits;
1318         if (res > upper_limit)
1319                 res = upper_limit;
1320         return res;
1321 }
1322
1323 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1324                                     ext3_fsblk_t logic_sb_block,
1325                                     int nr)
1326 {
1327         struct ext3_sb_info *sbi = EXT3_SB(sb);
1328         unsigned long bg, first_meta_bg;
1329         int has_super = 0;
1330
1331         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1332
1333         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1334             nr < first_meta_bg)
1335                 return (logic_sb_block + nr + 1);
1336         bg = sbi->s_desc_per_block * nr;
1337         if (ext3_bg_has_super(sb, bg))
1338                 has_super = 1;
1339         return (has_super + ext3_group_first_block_no(sb, bg));
1340 }
1341
1342
1343 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1344 {
1345         struct buffer_head * bh;
1346         struct ext3_super_block *es = NULL;
1347         struct ext3_sb_info *sbi;
1348         ext3_fsblk_t block;
1349         ext3_fsblk_t sb_block = get_sb_block(&data);
1350         ext3_fsblk_t logic_sb_block;
1351         unsigned long offset = 0;
1352         unsigned long journal_inum = 0;
1353         unsigned long journal_devnum = 0;
1354         unsigned long def_mount_opts;
1355         struct inode *root;
1356         int blocksize;
1357         int hblock;
1358         int db_count;
1359         int i;
1360         int needs_recovery;
1361         __le32 features;
1362
1363         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1364         if (!sbi)
1365                 return -ENOMEM;
1366         sb->s_fs_info = sbi;
1367         memset(sbi, 0, sizeof(*sbi));
1368         sbi->s_mount_opt = 0;
1369         sbi->s_resuid = EXT3_DEF_RESUID;
1370         sbi->s_resgid = EXT3_DEF_RESGID;
1371
1372         unlock_kernel();
1373
1374         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1375         if (!blocksize) {
1376                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1377                 goto out_fail;
1378         }
1379
1380         /*
1381          * The ext3 superblock will not be buffer aligned for other than 1kB
1382          * block sizes.  We need to calculate the offset from buffer start.
1383          */
1384         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1385                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1386                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1387         } else {
1388                 logic_sb_block = sb_block;
1389         }
1390
1391         if (!(bh = sb_bread(sb, logic_sb_block))) {
1392                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1393                 goto out_fail;
1394         }
1395         /*
1396          * Note: s_es must be initialized as soon as possible because
1397          *       some ext3 macro-instructions depend on its value
1398          */
1399         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1400         sbi->s_es = es;
1401         sb->s_magic = le16_to_cpu(es->s_magic);
1402         if (sb->s_magic != EXT3_SUPER_MAGIC)
1403                 goto cantfind_ext3;
1404
1405         /* Set defaults before we parse the mount options */
1406         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1407         if (def_mount_opts & EXT3_DEFM_DEBUG)
1408                 set_opt(sbi->s_mount_opt, DEBUG);
1409         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1410                 set_opt(sbi->s_mount_opt, GRPID);
1411         if (def_mount_opts & EXT3_DEFM_UID16)
1412                 set_opt(sbi->s_mount_opt, NO_UID32);
1413         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1414                 set_opt(sbi->s_mount_opt, XATTR_USER);
1415         if (def_mount_opts & EXT3_DEFM_ACL)
1416                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1417         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1418                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1419         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1420                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1421         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1422                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1423
1424         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1425                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1426         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1427                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1428
1429         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1430         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1431
1432         set_opt(sbi->s_mount_opt, RESERVATION);
1433
1434         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1435                             NULL, 0))
1436                 goto failed_mount;
1437
1438         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1439                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1440
1441         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1442             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1443              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1444              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1445                 printk(KERN_WARNING 
1446                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1447                        "running e2fsck is recommended\n");
1448         /*
1449          * Check feature flags regardless of the revision level, since we
1450          * previously didn't change the revision level when setting the flags,
1451          * so there is a chance incompat flags are set on a rev 0 filesystem.
1452          */
1453         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1454         if (features) {
1455                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1456                        "unsupported optional features (%x).\n",
1457                        sb->s_id, le32_to_cpu(features));
1458                 goto failed_mount;
1459         }
1460         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1461         if (!(sb->s_flags & MS_RDONLY) && features) {
1462                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1463                        "unsupported optional features (%x).\n",
1464                        sb->s_id, le32_to_cpu(features));
1465                 goto failed_mount;
1466         }
1467         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1468
1469         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1470             blocksize > EXT3_MAX_BLOCK_SIZE) {
1471                 printk(KERN_ERR 
1472                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1473                        blocksize, sb->s_id);
1474                 goto failed_mount;
1475         }
1476
1477         hblock = bdev_hardsect_size(sb->s_bdev);
1478         if (sb->s_blocksize != blocksize) {
1479                 /*
1480                  * Make sure the blocksize for the filesystem is larger
1481                  * than the hardware sectorsize for the machine.
1482                  */
1483                 if (blocksize < hblock) {
1484                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1485                                "device blocksize %d.\n", blocksize, hblock);
1486                         goto failed_mount;
1487                 }
1488
1489                 brelse (bh);
1490                 sb_set_blocksize(sb, blocksize);
1491                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1492                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1493                 bh = sb_bread(sb, logic_sb_block);
1494                 if (!bh) {
1495                         printk(KERN_ERR 
1496                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1497                         goto failed_mount;
1498                 }
1499                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1500                 sbi->s_es = es;
1501                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1502                         printk (KERN_ERR 
1503                                 "EXT3-fs: Magic mismatch, very weird !\n");
1504                         goto failed_mount;
1505                 }
1506         }
1507
1508         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1509
1510         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1511                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1512                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1513         } else {
1514                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1515                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1516                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1517                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1518                     (sbi->s_inode_size > blocksize)) {
1519                         printk (KERN_ERR
1520                                 "EXT3-fs: unsupported inode size: %d\n",
1521                                 sbi->s_inode_size);
1522                         goto failed_mount;
1523                 }
1524         }
1525         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1526                                    le32_to_cpu(es->s_log_frag_size);
1527         if (blocksize != sbi->s_frag_size) {
1528                 printk(KERN_ERR
1529                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1530                        sbi->s_frag_size, blocksize);
1531                 goto failed_mount;
1532         }
1533         sbi->s_frags_per_block = 1;
1534         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1535         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1536         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1537         if (EXT3_INODE_SIZE(sb) == 0)
1538                 goto cantfind_ext3;
1539         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1540         if (sbi->s_inodes_per_block == 0)
1541                 goto cantfind_ext3;
1542         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1543                                         sbi->s_inodes_per_block;
1544         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1545         sbi->s_sbh = bh;
1546         sbi->s_mount_state = le16_to_cpu(es->s_state);
1547         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1548         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1549         for (i=0; i < 4; i++)
1550                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1551         sbi->s_def_hash_version = es->s_def_hash_version;
1552
1553         if (sbi->s_blocks_per_group > blocksize * 8) {
1554                 printk (KERN_ERR
1555                         "EXT3-fs: #blocks per group too big: %lu\n",
1556                         sbi->s_blocks_per_group);
1557                 goto failed_mount;
1558         }
1559         if (sbi->s_frags_per_group > blocksize * 8) {
1560                 printk (KERN_ERR
1561                         "EXT3-fs: #fragments per group too big: %lu\n",
1562                         sbi->s_frags_per_group);
1563                 goto failed_mount;
1564         }
1565         if (sbi->s_inodes_per_group > blocksize * 8) {
1566                 printk (KERN_ERR
1567                         "EXT3-fs: #inodes per group too big: %lu\n",
1568                         sbi->s_inodes_per_group);
1569                 goto failed_mount;
1570         }
1571
1572         if (le32_to_cpu(es->s_blocks_count) >
1573                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1574                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1575                         " too large to mount safely\n", sb->s_id);
1576                 if (sizeof(sector_t) < 8)
1577                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1578                                         "enabled\n");
1579                 goto failed_mount;
1580         }
1581
1582         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1583                 goto cantfind_ext3;
1584         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1585                                le32_to_cpu(es->s_first_data_block) +
1586                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1587                               EXT3_BLOCKS_PER_GROUP(sb);
1588         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1589                    EXT3_DESC_PER_BLOCK(sb);
1590         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1591                                     GFP_KERNEL);
1592         if (sbi->s_group_desc == NULL) {
1593                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1594                 goto failed_mount;
1595         }
1596
1597         bgl_lock_init(&sbi->s_blockgroup_lock);
1598
1599         for (i = 0; i < db_count; i++) {
1600                 block = descriptor_loc(sb, logic_sb_block, i);
1601                 sbi->s_group_desc[i] = sb_bread(sb, block);
1602                 if (!sbi->s_group_desc[i]) {
1603                         printk (KERN_ERR "EXT3-fs: "
1604                                 "can't read group descriptor %d\n", i);
1605                         db_count = i;
1606                         goto failed_mount2;
1607                 }
1608         }
1609         if (!ext3_check_descriptors (sb)) {
1610                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1611                 goto failed_mount2;
1612         }
1613         sbi->s_gdb_count = db_count;
1614         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1615         spin_lock_init(&sbi->s_next_gen_lock);
1616
1617         percpu_counter_init(&sbi->s_freeblocks_counter,
1618                 ext3_count_free_blocks(sb));
1619         percpu_counter_init(&sbi->s_freeinodes_counter,
1620                 ext3_count_free_inodes(sb));
1621         percpu_counter_init(&sbi->s_dirs_counter,
1622                 ext3_count_dirs(sb));
1623
1624         /* per fileystem reservation list head & lock */
1625         spin_lock_init(&sbi->s_rsv_window_lock);
1626         sbi->s_rsv_window_root = RB_ROOT;
1627         /* Add a single, static dummy reservation to the start of the
1628          * reservation window list --- it gives us a placeholder for
1629          * append-at-start-of-list which makes the allocation logic
1630          * _much_ simpler. */
1631         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1632         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1633         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1634         sbi->s_rsv_window_head.rsv_goal_size = 0;
1635         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1636
1637         /*
1638          * set up enough so that it can read an inode
1639          */
1640         sb->s_op = &ext3_sops;
1641         sb->s_export_op = &ext3_export_ops;
1642         sb->s_xattr = ext3_xattr_handlers;
1643 #ifdef CONFIG_QUOTA
1644         sb->s_qcop = &ext3_qctl_operations;
1645         sb->dq_op = &ext3_quota_operations;
1646 #endif
1647         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1648
1649         sb->s_root = NULL;
1650
1651         needs_recovery = (es->s_last_orphan != 0 ||
1652                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1653                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1654
1655         /*
1656          * The first inode we look at is the journal inode.  Don't try
1657          * root first: it may be modified in the journal!
1658          */
1659         if (!test_opt(sb, NOLOAD) &&
1660             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1661                 if (ext3_load_journal(sb, es, journal_devnum))
1662                         goto failed_mount3;
1663         } else if (journal_inum) {
1664                 if (ext3_create_journal(sb, es, journal_inum))
1665                         goto failed_mount3;
1666         } else {
1667                 if (!silent)
1668                         printk (KERN_ERR
1669                                 "ext3: No journal on filesystem on %s\n",
1670                                 sb->s_id);
1671                 goto failed_mount3;
1672         }
1673
1674         /* We have now updated the journal if required, so we can
1675          * validate the data journaling mode. */
1676         switch (test_opt(sb, DATA_FLAGS)) {
1677         case 0:
1678                 /* No mode set, assume a default based on the journal
1679                    capabilities: ORDERED_DATA if the journal can
1680                    cope, else JOURNAL_DATA */
1681                 if (journal_check_available_features
1682                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1683                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1684                 else
1685                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1686                 break;
1687
1688         case EXT3_MOUNT_ORDERED_DATA:
1689         case EXT3_MOUNT_WRITEBACK_DATA:
1690                 if (!journal_check_available_features
1691                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1692                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1693                                "requested data journaling mode\n");
1694                         goto failed_mount4;
1695                 }
1696         default:
1697                 break;
1698         }
1699
1700         if (test_opt(sb, NOBH)) {
1701                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1702                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1703                                 "its supported only with writeback mode\n");
1704                         clear_opt(sbi->s_mount_opt, NOBH);
1705                 }
1706         }
1707         /*
1708          * The journal_load will have done any necessary log recovery,
1709          * so we can safely mount the rest of the filesystem now.
1710          */
1711
1712         root = iget(sb, EXT3_ROOT_INO);
1713         sb->s_root = d_alloc_root(root);
1714         if (!sb->s_root) {
1715                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1716                 iput(root);
1717                 goto failed_mount4;
1718         }
1719         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1720                 dput(sb->s_root);
1721                 sb->s_root = NULL;
1722                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1723                 goto failed_mount4;
1724         }
1725
1726         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1727         /*
1728          * akpm: core read_super() calls in here with the superblock locked.
1729          * That deadlocks, because orphan cleanup needs to lock the superblock
1730          * in numerous places.  Here we just pop the lock - it's relatively
1731          * harmless, because we are now ready to accept write_super() requests,
1732          * and aviro says that's the only reason for hanging onto the
1733          * superblock lock.
1734          */
1735         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1736         ext3_orphan_cleanup(sb, es);
1737         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1738         if (needs_recovery)
1739                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1740         ext3_mark_recovery_complete(sb, es);
1741         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1742                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1743                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1744                 "writeback");
1745
1746         lock_kernel();
1747         return 0;
1748
1749 cantfind_ext3:
1750         if (!silent)
1751                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1752                        sb->s_id);
1753         goto failed_mount;
1754
1755 failed_mount4:
1756         journal_destroy(sbi->s_journal);
1757 failed_mount3:
1758         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1759         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1760         percpu_counter_destroy(&sbi->s_dirs_counter);
1761 failed_mount2:
1762         for (i = 0; i < db_count; i++)
1763                 brelse(sbi->s_group_desc[i]);
1764         kfree(sbi->s_group_desc);
1765 failed_mount:
1766 #ifdef CONFIG_QUOTA
1767         for (i = 0; i < MAXQUOTAS; i++)
1768                 kfree(sbi->s_qf_names[i]);
1769 #endif
1770         ext3_blkdev_remove(sbi);
1771         brelse(bh);
1772 out_fail:
1773         sb->s_fs_info = NULL;
1774         kfree(sbi);
1775         lock_kernel();
1776         return -EINVAL;
1777 }
1778
1779 /*
1780  * Setup any per-fs journal parameters now.  We'll do this both on
1781  * initial mount, once the journal has been initialised but before we've
1782  * done any recovery; and again on any subsequent remount. 
1783  */
1784 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1785 {
1786         struct ext3_sb_info *sbi = EXT3_SB(sb);
1787
1788         if (sbi->s_commit_interval)
1789                 journal->j_commit_interval = sbi->s_commit_interval;
1790         /* We could also set up an ext3-specific default for the commit
1791          * interval here, but for now we'll just fall back to the jbd
1792          * default. */
1793
1794         spin_lock(&journal->j_state_lock);
1795         if (test_opt(sb, BARRIER))
1796                 journal->j_flags |= JFS_BARRIER;
1797         else
1798                 journal->j_flags &= ~JFS_BARRIER;
1799         spin_unlock(&journal->j_state_lock);
1800 }
1801
1802 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1803 {
1804         struct inode *journal_inode;
1805         journal_t *journal;
1806
1807         /* First, test for the existence of a valid inode on disk.  Bad
1808          * things happen if we iget() an unused inode, as the subsequent
1809          * iput() will try to delete it. */
1810
1811         journal_inode = iget(sb, journal_inum);
1812         if (!journal_inode) {
1813                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1814                 return NULL;
1815         }
1816         if (!journal_inode->i_nlink) {
1817                 make_bad_inode(journal_inode);
1818                 iput(journal_inode);
1819                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1820                 return NULL;
1821         }
1822
1823         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1824                   journal_inode, journal_inode->i_size);
1825         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1826                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1827                 iput(journal_inode);
1828                 return NULL;
1829         }
1830
1831         journal = journal_init_inode(journal_inode);
1832         if (!journal) {
1833                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1834                 iput(journal_inode);
1835                 return NULL;
1836         }
1837         journal->j_private = sb;
1838         ext3_init_journal_params(sb, journal);
1839         return journal;
1840 }
1841
1842 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1843                                        dev_t j_dev)
1844 {
1845         struct buffer_head * bh;
1846         journal_t *journal;
1847         ext3_fsblk_t start;
1848         ext3_fsblk_t len;
1849         int hblock, blocksize;
1850         ext3_fsblk_t sb_block;
1851         unsigned long offset;
1852         struct ext3_super_block * es;
1853         struct block_device *bdev;
1854
1855         bdev = ext3_blkdev_get(j_dev);
1856         if (bdev == NULL)
1857                 return NULL;
1858
1859         if (bd_claim(bdev, sb)) {
1860                 printk(KERN_ERR
1861                         "EXT3: failed to claim external journal device.\n");
1862                 blkdev_put(bdev);
1863                 return NULL;
1864         }
1865
1866         blocksize = sb->s_blocksize;
1867         hblock = bdev_hardsect_size(bdev);
1868         if (blocksize < hblock) {
1869                 printk(KERN_ERR
1870                         "EXT3-fs: blocksize too small for journal device.\n");
1871                 goto out_bdev;
1872         }
1873
1874         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1875         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1876         set_blocksize(bdev, blocksize);
1877         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1878                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1879                        "external journal\n");
1880                 goto out_bdev;
1881         }
1882
1883         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1884         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1885             !(le32_to_cpu(es->s_feature_incompat) &
1886               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1887                 printk(KERN_ERR "EXT3-fs: external journal has "
1888                                         "bad superblock\n");
1889                 brelse(bh);
1890                 goto out_bdev;
1891         }
1892
1893         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1894                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1895                 brelse(bh);
1896                 goto out_bdev;
1897         }
1898
1899         len = le32_to_cpu(es->s_blocks_count);
1900         start = sb_block + 1;
1901         brelse(bh);     /* we're done with the superblock */
1902
1903         journal = journal_init_dev(bdev, sb->s_bdev,
1904                                         start, len, blocksize);
1905         if (!journal) {
1906                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1907                 goto out_bdev;
1908         }
1909         journal->j_private = sb;
1910         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1911         wait_on_buffer(journal->j_sb_buffer);
1912         if (!buffer_uptodate(journal->j_sb_buffer)) {
1913                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1914                 goto out_journal;
1915         }
1916         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1917                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1918                                         "user (unsupported) - %d\n",
1919                         be32_to_cpu(journal->j_superblock->s_nr_users));
1920                 goto out_journal;
1921         }
1922         EXT3_SB(sb)->journal_bdev = bdev;
1923         ext3_init_journal_params(sb, journal);
1924         return journal;
1925 out_journal:
1926         journal_destroy(journal);
1927 out_bdev:
1928         ext3_blkdev_put(bdev);
1929         return NULL;
1930 }
1931
1932 static int ext3_load_journal(struct super_block *sb,
1933                              struct ext3_super_block *es,
1934                              unsigned long journal_devnum)
1935 {
1936         journal_t *journal;
1937         int journal_inum = le32_to_cpu(es->s_journal_inum);
1938         dev_t journal_dev;
1939         int err = 0;
1940         int really_read_only;
1941
1942         if (journal_devnum &&
1943             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
1944                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
1945                         "numbers have changed\n");
1946                 journal_dev = new_decode_dev(journal_devnum);
1947         } else
1948                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1949
1950         really_read_only = bdev_read_only(sb->s_bdev);
1951
1952         /*
1953          * Are we loading a blank journal or performing recovery after a
1954          * crash?  For recovery, we need to check in advance whether we
1955          * can get read-write access to the device.
1956          */
1957
1958         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1959                 if (sb->s_flags & MS_RDONLY) {
1960                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1961                                         "required on readonly filesystem.\n");
1962                         if (really_read_only) {
1963                                 printk(KERN_ERR "EXT3-fs: write access "
1964                                         "unavailable, cannot proceed.\n");
1965                                 return -EROFS;
1966                         }
1967                         printk (KERN_INFO "EXT3-fs: write access will "
1968                                         "be enabled during recovery.\n");
1969                 }
1970         }
1971
1972         if (journal_inum && journal_dev) {
1973                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1974                        "and inode journals!\n");
1975                 return -EINVAL;
1976         }
1977
1978         if (journal_inum) {
1979                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1980                         return -EINVAL;
1981         } else {
1982                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1983                         return -EINVAL;
1984         }
1985
1986         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1987                 err = journal_update_format(journal);
1988                 if (err)  {
1989                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1990                         journal_destroy(journal);
1991                         return err;
1992                 }
1993         }
1994
1995         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1996                 err = journal_wipe(journal, !really_read_only);
1997         if (!err)
1998                 err = journal_load(journal);
1999
2000         if (err) {
2001                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2002                 journal_destroy(journal);
2003                 return err;
2004         }
2005
2006         EXT3_SB(sb)->s_journal = journal;
2007         ext3_clear_journal_err(sb, es);
2008
2009         if (journal_devnum &&
2010             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2011                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2012                 sb->s_dirt = 1;
2013
2014                 /* Make sure we flush the recovery flag to disk. */
2015                 ext3_commit_super(sb, es, 1);
2016         }
2017
2018         return 0;
2019 }
2020
2021 static int ext3_create_journal(struct super_block * sb,
2022                                struct ext3_super_block * es,
2023                                int journal_inum)
2024 {
2025         journal_t *journal;
2026
2027         if (sb->s_flags & MS_RDONLY) {
2028                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2029                                 "create journal.\n");
2030                 return -EROFS;
2031         }
2032
2033         if (!(journal = ext3_get_journal(sb, journal_inum)))
2034                 return -EINVAL;
2035
2036         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
2037                journal_inum);
2038
2039         if (journal_create(journal)) {
2040                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2041                 journal_destroy(journal);
2042                 return -EIO;
2043         }
2044
2045         EXT3_SB(sb)->s_journal = journal;
2046
2047         ext3_update_dynamic_rev(sb);
2048         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2049         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2050
2051         es->s_journal_inum = cpu_to_le32(journal_inum);
2052         sb->s_dirt = 1;
2053
2054         /* Make sure we flush the recovery flag to disk. */
2055         ext3_commit_super(sb, es, 1);
2056
2057         return 0;
2058 }
2059
2060 static void ext3_commit_super (struct super_block * sb,
2061                                struct ext3_super_block * es,
2062                                int sync)
2063 {
2064         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2065
2066         if (!sbh)
2067                 return;
2068         es->s_wtime = cpu_to_le32(get_seconds());
2069         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2070         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2071         BUFFER_TRACE(sbh, "marking dirty");
2072         mark_buffer_dirty(sbh);
2073         if (sync)
2074                 sync_dirty_buffer(sbh);
2075 }
2076
2077
2078 /*
2079  * Have we just finished recovery?  If so, and if we are mounting (or
2080  * remounting) the filesystem readonly, then we will end up with a
2081  * consistent fs on disk.  Record that fact.
2082  */
2083 static void ext3_mark_recovery_complete(struct super_block * sb,
2084                                         struct ext3_super_block * es)
2085 {
2086         journal_t *journal = EXT3_SB(sb)->s_journal;
2087
2088         journal_lock_updates(journal);
2089         journal_flush(journal);
2090         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2091             sb->s_flags & MS_RDONLY) {
2092                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2093                 sb->s_dirt = 0;
2094                 ext3_commit_super(sb, es, 1);
2095         }
2096         journal_unlock_updates(journal);
2097 }
2098
2099 /*
2100  * If we are mounting (or read-write remounting) a filesystem whose journal
2101  * has recorded an error from a previous lifetime, move that error to the
2102  * main filesystem now.
2103  */
2104 static void ext3_clear_journal_err(struct super_block * sb,
2105                                    struct ext3_super_block * es)
2106 {
2107         journal_t *journal;
2108         int j_errno;
2109         const char *errstr;
2110
2111         journal = EXT3_SB(sb)->s_journal;
2112
2113         /*
2114          * Now check for any error status which may have been recorded in the
2115          * journal by a prior ext3_error() or ext3_abort()
2116          */
2117
2118         j_errno = journal_errno(journal);
2119         if (j_errno) {
2120                 char nbuf[16];
2121
2122                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2123                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2124                              "from previous mount: %s", errstr);
2125                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2126                              "filesystem check.");
2127
2128                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2129                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2130                 ext3_commit_super (sb, es, 1);
2131
2132                 journal_clear_err(journal);
2133         }
2134 }
2135
2136 /*
2137  * Force the running and committing transactions to commit,
2138  * and wait on the commit.
2139  */
2140 int ext3_force_commit(struct super_block *sb)
2141 {
2142         journal_t *journal;
2143         int ret;
2144
2145         if (sb->s_flags & MS_RDONLY)
2146                 return 0;
2147
2148         journal = EXT3_SB(sb)->s_journal;
2149         sb->s_dirt = 0;
2150         ret = ext3_journal_force_commit(journal);
2151         return ret;
2152 }
2153
2154 /*
2155  * Ext3 always journals updates to the superblock itself, so we don't
2156  * have to propagate any other updates to the superblock on disk at this
2157  * point.  Just start an async writeback to get the buffers on their way
2158  * to the disk.
2159  *
2160  * This implicitly triggers the writebehind on sync().
2161  */
2162
2163 static void ext3_write_super (struct super_block * sb)
2164 {
2165         if (mutex_trylock(&sb->s_lock) != 0)
2166                 BUG();
2167         sb->s_dirt = 0;
2168 }
2169
2170 static int ext3_sync_fs(struct super_block *sb, int wait)
2171 {
2172         tid_t target;
2173
2174         sb->s_dirt = 0;
2175         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2176                 if (wait)
2177                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2178         }
2179         return 0;
2180 }
2181
2182 /*
2183  * LVM calls this function before a (read-only) snapshot is created.  This
2184  * gives us a chance to flush the journal completely and mark the fs clean.
2185  */
2186 static void ext3_write_super_lockfs(struct super_block *sb)
2187 {
2188         sb->s_dirt = 0;
2189
2190         if (!(sb->s_flags & MS_RDONLY)) {
2191                 journal_t *journal = EXT3_SB(sb)->s_journal;
2192
2193                 /* Now we set up the journal barrier. */
2194                 journal_lock_updates(journal);
2195                 journal_flush(journal);
2196
2197                 /* Journal blocked and flushed, clear needs_recovery flag. */
2198                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2199                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2200         }
2201 }
2202
2203 /*
2204  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2205  * flag here, even though the filesystem is not technically dirty yet.
2206  */
2207 static void ext3_unlockfs(struct super_block *sb)
2208 {
2209         if (!(sb->s_flags & MS_RDONLY)) {
2210                 lock_super(sb);
2211                 /* Reser the needs_recovery flag before the fs is unlocked. */
2212                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2213                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2214                 unlock_super(sb);
2215                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2216         }
2217 }
2218
2219 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2220 {
2221         struct ext3_super_block * es;
2222         struct ext3_sb_info *sbi = EXT3_SB(sb);
2223         ext3_fsblk_t n_blocks_count = 0;
2224         unsigned long old_sb_flags;
2225         struct ext3_mount_options old_opts;
2226         int err;
2227 #ifdef CONFIG_QUOTA
2228         int i;
2229 #endif
2230
2231         /* Store the original options */
2232         old_sb_flags = sb->s_flags;
2233         old_opts.s_mount_opt = sbi->s_mount_opt;
2234         old_opts.s_resuid = sbi->s_resuid;
2235         old_opts.s_resgid = sbi->s_resgid;
2236         old_opts.s_commit_interval = sbi->s_commit_interval;
2237 #ifdef CONFIG_QUOTA
2238         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2239         for (i = 0; i < MAXQUOTAS; i++)
2240                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2241 #endif
2242
2243         /*
2244          * Allow the "check" option to be passed as a remount option.
2245          */
2246         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2247                 err = -EINVAL;
2248                 goto restore_opts;
2249         }
2250
2251         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2252                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2253
2254         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2255                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2256
2257         es = sbi->s_es;
2258
2259         ext3_init_journal_params(sb, sbi->s_journal);
2260
2261         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2262                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2263                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2264                         err = -EROFS;
2265                         goto restore_opts;
2266                 }
2267
2268                 if (*flags & MS_RDONLY) {
2269                         /*
2270                          * First of all, the unconditional stuff we have to do
2271                          * to disable replay of the journal when we next remount
2272                          */
2273                         sb->s_flags |= MS_RDONLY;
2274
2275                         /*
2276                          * OK, test if we are remounting a valid rw partition
2277                          * readonly, and if so set the rdonly flag and then
2278                          * mark the partition as valid again.
2279                          */
2280                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2281                             (sbi->s_mount_state & EXT3_VALID_FS))
2282                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2283
2284                         ext3_mark_recovery_complete(sb, es);
2285                 } else {
2286                         __le32 ret;
2287                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2288                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2289                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2290                                        "remount RDWR because of unsupported "
2291                                        "optional features (%x).\n",
2292                                        sb->s_id, le32_to_cpu(ret));
2293                                 err = -EROFS;
2294                                 goto restore_opts;
2295                         }
2296                         /*
2297                          * Mounting a RDONLY partition read-write, so reread
2298                          * and store the current valid flag.  (It may have
2299                          * been changed by e2fsck since we originally mounted
2300                          * the partition.)
2301                          */
2302                         ext3_clear_journal_err(sb, es);
2303                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2304                         if ((ret = ext3_group_extend(sb, es, n_blocks_count))) {
2305                                 err = ret;
2306                                 goto restore_opts;
2307                         }
2308                         if (!ext3_setup_super (sb, es, 0))
2309                                 sb->s_flags &= ~MS_RDONLY;
2310                 }
2311         }
2312 #ifdef CONFIG_QUOTA
2313         /* Release old quota file names */
2314         for (i = 0; i < MAXQUOTAS; i++)
2315                 if (old_opts.s_qf_names[i] &&
2316                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2317                         kfree(old_opts.s_qf_names[i]);
2318 #endif
2319         return 0;
2320 restore_opts:
2321         sb->s_flags = old_sb_flags;
2322         sbi->s_mount_opt = old_opts.s_mount_opt;
2323         sbi->s_resuid = old_opts.s_resuid;
2324         sbi->s_resgid = old_opts.s_resgid;
2325         sbi->s_commit_interval = old_opts.s_commit_interval;
2326 #ifdef CONFIG_QUOTA
2327         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2328         for (i = 0; i < MAXQUOTAS; i++) {
2329                 if (sbi->s_qf_names[i] &&
2330                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2331                         kfree(sbi->s_qf_names[i]);
2332                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2333         }
2334 #endif
2335         return err;
2336 }
2337
2338 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2339 {
2340         struct super_block *sb = dentry->d_sb;
2341         struct ext3_sb_info *sbi = EXT3_SB(sb);
2342         struct ext3_super_block *es = sbi->s_es;
2343         ext3_fsblk_t overhead;
2344         int i;
2345
2346         if (test_opt (sb, MINIX_DF))
2347                 overhead = 0;
2348         else {
2349                 unsigned long ngroups;
2350                 ngroups = EXT3_SB(sb)->s_groups_count;
2351                 smp_rmb();
2352
2353                 /*
2354                  * Compute the overhead (FS structures)
2355                  */
2356
2357                 /*
2358                  * All of the blocks before first_data_block are
2359                  * overhead
2360                  */
2361                 overhead = le32_to_cpu(es->s_first_data_block);
2362
2363                 /*
2364                  * Add the overhead attributed to the superblock and
2365                  * block group descriptors.  If the sparse superblocks
2366                  * feature is turned on, then not all groups have this.
2367                  */
2368                 for (i = 0; i < ngroups; i++) {
2369                         overhead += ext3_bg_has_super(sb, i) +
2370                                 ext3_bg_num_gdb(sb, i);
2371                         cond_resched();
2372                 }
2373
2374                 /*
2375                  * Every block group has an inode bitmap, a block
2376                  * bitmap, and an inode table.
2377                  */
2378                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2379         }
2380
2381         buf->f_type = EXT3_SUPER_MAGIC;
2382         buf->f_bsize = sb->s_blocksize;
2383         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2384         buf->f_bfree = percpu_counter_sum(&sbi->s_freeblocks_counter);
2385         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2386         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2387                 buf->f_bavail = 0;
2388         buf->f_files = le32_to_cpu(es->s_inodes_count);
2389         buf->f_ffree = percpu_counter_sum(&sbi->s_freeinodes_counter);
2390         buf->f_namelen = EXT3_NAME_LEN;
2391         return 0;
2392 }
2393
2394 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2395  * is locked for write. Otherwise the are possible deadlocks:
2396  * Process 1                         Process 2
2397  * ext3_create()                     quota_sync()
2398  *   journal_start()                   write_dquot()
2399  *   DQUOT_INIT()                        down(dqio_mutex)
2400  *     down(dqio_mutex)                    journal_start()
2401  *
2402  */
2403
2404 #ifdef CONFIG_QUOTA
2405
2406 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2407 {
2408         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2409 }
2410
2411 static int ext3_dquot_initialize(struct inode *inode, int type)
2412 {
2413         handle_t *handle;
2414         int ret, err;
2415
2416         /* We may create quota structure so we need to reserve enough blocks */
2417         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2418         if (IS_ERR(handle))
2419                 return PTR_ERR(handle);
2420         ret = dquot_initialize(inode, type);
2421         err = ext3_journal_stop(handle);
2422         if (!ret)
2423                 ret = err;
2424         return ret;
2425 }
2426
2427 static int ext3_dquot_drop(struct inode *inode)
2428 {
2429         handle_t *handle;
2430         int ret, err;
2431
2432         /* We may delete quota structure so we need to reserve enough blocks */
2433         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2434         if (IS_ERR(handle))
2435                 return PTR_ERR(handle);
2436         ret = dquot_drop(inode);
2437         err = ext3_journal_stop(handle);
2438         if (!ret)
2439                 ret = err;
2440         return ret;
2441 }
2442
2443 static int ext3_write_dquot(struct dquot *dquot)
2444 {
2445         int ret, err;
2446         handle_t *handle;
2447         struct inode *inode;
2448
2449         inode = dquot_to_inode(dquot);
2450         handle = ext3_journal_start(inode,
2451                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2452         if (IS_ERR(handle))
2453                 return PTR_ERR(handle);
2454         ret = dquot_commit(dquot);
2455         err = ext3_journal_stop(handle);
2456         if (!ret)
2457                 ret = err;
2458         return ret;
2459 }
2460
2461 static int ext3_acquire_dquot(struct dquot *dquot)
2462 {
2463         int ret, err;
2464         handle_t *handle;
2465
2466         handle = ext3_journal_start(dquot_to_inode(dquot),
2467                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2468         if (IS_ERR(handle))
2469                 return PTR_ERR(handle);
2470         ret = dquot_acquire(dquot);
2471         err = ext3_journal_stop(handle);
2472         if (!ret)
2473                 ret = err;
2474         return ret;
2475 }
2476
2477 static int ext3_release_dquot(struct dquot *dquot)
2478 {
2479         int ret, err;
2480         handle_t *handle;
2481
2482         handle = ext3_journal_start(dquot_to_inode(dquot),
2483                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2484         if (IS_ERR(handle))
2485                 return PTR_ERR(handle);
2486         ret = dquot_release(dquot);
2487         err = ext3_journal_stop(handle);
2488         if (!ret)
2489                 ret = err;
2490         return ret;
2491 }
2492
2493 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2494 {
2495         /* Are we journalling quotas? */
2496         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2497             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2498                 dquot_mark_dquot_dirty(dquot);
2499                 return ext3_write_dquot(dquot);
2500         } else {
2501                 return dquot_mark_dquot_dirty(dquot);
2502         }
2503 }
2504
2505 static int ext3_write_info(struct super_block *sb, int type)
2506 {
2507         int ret, err;
2508         handle_t *handle;
2509
2510         /* Data block + inode block */
2511         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2512         if (IS_ERR(handle))
2513                 return PTR_ERR(handle);
2514         ret = dquot_commit_info(sb, type);
2515         err = ext3_journal_stop(handle);
2516         if (!ret)
2517                 ret = err;
2518         return ret;
2519 }
2520
2521 /*
2522  * Turn on quotas during mount time - we need to find
2523  * the quota file and such...
2524  */
2525 static int ext3_quota_on_mount(struct super_block *sb, int type)
2526 {
2527         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2528                         EXT3_SB(sb)->s_jquota_fmt, type);
2529 }
2530
2531 /*
2532  * Standard function to be called on quota_on
2533  */
2534 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2535                          char *path)
2536 {
2537         int err;
2538         struct nameidata nd;
2539
2540         if (!test_opt(sb, QUOTA))
2541                 return -EINVAL;
2542         /* Not journalling quota? */
2543         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2544             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2545                 return vfs_quota_on(sb, type, format_id, path);
2546         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2547         if (err)
2548                 return err;
2549         /* Quotafile not on the same filesystem? */
2550         if (nd.mnt->mnt_sb != sb) {
2551                 path_release(&nd);
2552                 return -EXDEV;
2553         }
2554         /* Quotafile not of fs root? */
2555         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2556                 printk(KERN_WARNING
2557                         "EXT3-fs: Quota file not on filesystem root. "
2558                         "Journalled quota will not work.\n");
2559         path_release(&nd);
2560         return vfs_quota_on(sb, type, format_id, path);
2561 }
2562
2563 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2564  * acquiring the locks... As quota files are never truncated and quota code
2565  * itself serializes the operations (and noone else should touch the files)
2566  * we don't have to be afraid of races */
2567 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2568                                size_t len, loff_t off)
2569 {
2570         struct inode *inode = sb_dqopt(sb)->files[type];
2571         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2572         int err = 0;
2573         int offset = off & (sb->s_blocksize - 1);
2574         int tocopy;
2575         size_t toread;
2576         struct buffer_head *bh;
2577         loff_t i_size = i_size_read(inode);
2578
2579         if (off > i_size)
2580                 return 0;
2581         if (off+len > i_size)
2582                 len = i_size-off;
2583         toread = len;
2584         while (toread > 0) {
2585                 tocopy = sb->s_blocksize - offset < toread ?
2586                                 sb->s_blocksize - offset : toread;
2587                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2588                 if (err)
2589                         return err;
2590                 if (!bh)        /* A hole? */
2591                         memset(data, 0, tocopy);
2592                 else
2593                         memcpy(data, bh->b_data+offset, tocopy);
2594                 brelse(bh);
2595                 offset = 0;
2596                 toread -= tocopy;
2597                 data += tocopy;
2598                 blk++;
2599         }
2600         return len;
2601 }
2602
2603 /* Write to quotafile (we know the transaction is already started and has
2604  * enough credits) */
2605 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2606                                 const char *data, size_t len, loff_t off)
2607 {
2608         struct inode *inode = sb_dqopt(sb)->files[type];
2609         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2610         int err = 0;
2611         int offset = off & (sb->s_blocksize - 1);
2612         int tocopy;
2613         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2614         size_t towrite = len;
2615         struct buffer_head *bh;
2616         handle_t *handle = journal_current_handle();
2617
2618         mutex_lock(&inode->i_mutex);
2619         while (towrite > 0) {
2620                 tocopy = sb->s_blocksize - offset < towrite ?
2621                                 sb->s_blocksize - offset : towrite;
2622                 bh = ext3_bread(handle, inode, blk, 1, &err);
2623                 if (!bh)
2624                         goto out;
2625                 if (journal_quota) {
2626                         err = ext3_journal_get_write_access(handle, bh);
2627                         if (err) {
2628                                 brelse(bh);
2629                                 goto out;
2630                         }
2631                 }
2632                 lock_buffer(bh);
2633                 memcpy(bh->b_data+offset, data, tocopy);
2634                 flush_dcache_page(bh->b_page);
2635                 unlock_buffer(bh);
2636                 if (journal_quota)
2637                         err = ext3_journal_dirty_metadata(handle, bh);
2638                 else {
2639                         /* Always do at least ordered writes for quotas */
2640                         err = ext3_journal_dirty_data(handle, bh);
2641                         mark_buffer_dirty(bh);
2642                 }
2643                 brelse(bh);
2644                 if (err)
2645                         goto out;
2646                 offset = 0;
2647                 towrite -= tocopy;
2648                 data += tocopy;
2649                 blk++;
2650         }
2651 out:
2652         if (len == towrite)
2653                 return err;
2654         if (inode->i_size < off+len-towrite) {
2655                 i_size_write(inode, off+len-towrite);
2656                 EXT3_I(inode)->i_disksize = inode->i_size;
2657         }
2658         inode->i_version++;
2659         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2660         ext3_mark_inode_dirty(handle, inode);
2661         mutex_unlock(&inode->i_mutex);
2662         return len - towrite;
2663 }
2664
2665 #endif
2666
2667 static int ext3_get_sb(struct file_system_type *fs_type,
2668         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2669 {
2670         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
2671 }
2672
2673 static struct file_system_type ext3_fs_type = {
2674         .owner          = THIS_MODULE,
2675         .name           = "ext3",
2676         .get_sb         = ext3_get_sb,
2677         .kill_sb        = kill_block_super,
2678         .fs_flags       = FS_REQUIRES_DEV,
2679 };
2680
2681 static int __init init_ext3_fs(void)
2682 {
2683         int err = init_ext3_xattr();
2684         if (err)
2685                 return err;
2686         err = init_inodecache();
2687         if (err)
2688                 goto out1;
2689         err = register_filesystem(&ext3_fs_type);
2690         if (err)
2691                 goto out;
2692         return 0;
2693 out:
2694         destroy_inodecache();
2695 out1:
2696         exit_ext3_xattr();
2697         return err;
2698 }
2699
2700 static void __exit exit_ext3_fs(void)
2701 {
2702         unregister_filesystem(&ext3_fs_type);
2703         destroy_inodecache();
2704         exit_ext3_xattr();
2705 }
2706
2707 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2708 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2709 MODULE_LICENSE("GPL");
2710 module_init(init_ext3_fs)
2711 module_exit(exit_ext3_fs)