6a66f3d02b750084949a27fe3dd42d27e0d2815f
[linux-2.6.git] / fs / ext3 / super.c
1 /*
2  *  linux/fs/ext3/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/module.h>
20 #include <linux/string.h>
21 #include <linux/fs.h>
22 #include <linux/time.h>
23 #include <linux/jbd.h>
24 #include <linux/ext3_fs.h>
25 #include <linux/ext3_jbd.h>
26 #include <linux/slab.h>
27 #include <linux/init.h>
28 #include <linux/blkdev.h>
29 #include <linux/parser.h>
30 #include <linux/smp_lock.h>
31 #include <linux/buffer_head.h>
32 #include <linux/exportfs.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 #include <linux/log2.h>
40
41 #include <asm/uaccess.h>
42
43 #include "xattr.h"
44 #include "acl.h"
45 #include "namei.h"
46
47 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
48                              unsigned long journal_devnum);
49 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
50                                unsigned int);
51 static void ext3_commit_super (struct super_block * sb,
52                                struct ext3_super_block * es,
53                                int sync);
54 static void ext3_mark_recovery_complete(struct super_block * sb,
55                                         struct ext3_super_block * es);
56 static void ext3_clear_journal_err(struct super_block * sb,
57                                    struct ext3_super_block * es);
58 static int ext3_sync_fs(struct super_block *sb, int wait);
59 static const char *ext3_decode_error(struct super_block * sb, int errno,
60                                      char nbuf[16]);
61 static int ext3_remount (struct super_block * sb, int * flags, char * data);
62 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
63 static void ext3_unlockfs(struct super_block *sb);
64 static void ext3_write_super (struct super_block * sb);
65 static void ext3_write_super_lockfs(struct super_block *sb);
66
67 /*
68  * Wrappers for journal_start/end.
69  *
70  * The only special thing we need to do here is to make sure that all
71  * journal_end calls result in the superblock being marked dirty, so
72  * that sync() will call the filesystem's write_super callback if
73  * appropriate.
74  */
75 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
76 {
77         journal_t *journal;
78
79         if (sb->s_flags & MS_RDONLY)
80                 return ERR_PTR(-EROFS);
81
82         /* Special case here: if the journal has aborted behind our
83          * backs (eg. EIO in the commit thread), then we still need to
84          * take the FS itself readonly cleanly. */
85         journal = EXT3_SB(sb)->s_journal;
86         if (is_journal_aborted(journal)) {
87                 ext3_abort(sb, __FUNCTION__,
88                            "Detected aborted journal");
89                 return ERR_PTR(-EROFS);
90         }
91
92         return journal_start(journal, nblocks);
93 }
94
95 /*
96  * The only special thing we need to do here is to make sure that all
97  * journal_stop calls result in the superblock being marked dirty, so
98  * that sync() will call the filesystem's write_super callback if
99  * appropriate.
100  */
101 int __ext3_journal_stop(const char *where, handle_t *handle)
102 {
103         struct super_block *sb;
104         int err;
105         int rc;
106
107         sb = handle->h_transaction->t_journal->j_private;
108         err = handle->h_err;
109         rc = journal_stop(handle);
110
111         if (!err)
112                 err = rc;
113         if (err)
114                 __ext3_std_error(sb, where, err);
115         return err;
116 }
117
118 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
119                 struct buffer_head *bh, handle_t *handle, int err)
120 {
121         char nbuf[16];
122         const char *errstr = ext3_decode_error(NULL, err, nbuf);
123
124         if (bh)
125                 BUFFER_TRACE(bh, "abort");
126
127         if (!handle->h_err)
128                 handle->h_err = err;
129
130         if (is_handle_aborted(handle))
131                 return;
132
133         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
134                caller, errstr, err_fn);
135
136         journal_abort_handle(handle);
137 }
138
139 /* Deal with the reporting of failure conditions on a filesystem such as
140  * inconsistencies detected or read IO failures.
141  *
142  * On ext2, we can store the error state of the filesystem in the
143  * superblock.  That is not possible on ext3, because we may have other
144  * write ordering constraints on the superblock which prevent us from
145  * writing it out straight away; and given that the journal is about to
146  * be aborted, we can't rely on the current, or future, transactions to
147  * write out the superblock safely.
148  *
149  * We'll just use the journal_abort() error code to record an error in
150  * the journal instead.  On recovery, the journal will compain about
151  * that error until we've noted it down and cleared it.
152  */
153
154 static void ext3_handle_error(struct super_block *sb)
155 {
156         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
157
158         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
159         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
160
161         if (sb->s_flags & MS_RDONLY)
162                 return;
163
164         if (!test_opt (sb, ERRORS_CONT)) {
165                 journal_t *journal = EXT3_SB(sb)->s_journal;
166
167                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
168                 if (journal)
169                         journal_abort(journal, -EIO);
170         }
171         if (test_opt (sb, ERRORS_RO)) {
172                 printk (KERN_CRIT "Remounting filesystem read-only\n");
173                 sb->s_flags |= MS_RDONLY;
174         }
175         ext3_commit_super(sb, es, 1);
176         if (test_opt(sb, ERRORS_PANIC))
177                 panic("EXT3-fs (device %s): panic forced after error\n",
178                         sb->s_id);
179 }
180
181 void ext3_error (struct super_block * sb, const char * function,
182                  const char * fmt, ...)
183 {
184         va_list args;
185
186         va_start(args, fmt);
187         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
188         vprintk(fmt, args);
189         printk("\n");
190         va_end(args);
191
192         ext3_handle_error(sb);
193 }
194
195 static const char *ext3_decode_error(struct super_block * sb, int errno,
196                                      char nbuf[16])
197 {
198         char *errstr = NULL;
199
200         switch (errno) {
201         case -EIO:
202                 errstr = "IO failure";
203                 break;
204         case -ENOMEM:
205                 errstr = "Out of memory";
206                 break;
207         case -EROFS:
208                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
209                         errstr = "Journal has aborted";
210                 else
211                         errstr = "Readonly filesystem";
212                 break;
213         default:
214                 /* If the caller passed in an extra buffer for unknown
215                  * errors, textualise them now.  Else we just return
216                  * NULL. */
217                 if (nbuf) {
218                         /* Check for truncated error codes... */
219                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
220                                 errstr = nbuf;
221                 }
222                 break;
223         }
224
225         return errstr;
226 }
227
228 /* __ext3_std_error decodes expected errors from journaling functions
229  * automatically and invokes the appropriate error response.  */
230
231 void __ext3_std_error (struct super_block * sb, const char * function,
232                        int errno)
233 {
234         char nbuf[16];
235         const char *errstr;
236
237         /* Special case: if the error is EROFS, and we're not already
238          * inside a transaction, then there's really no point in logging
239          * an error. */
240         if (errno == -EROFS && journal_current_handle() == NULL &&
241             (sb->s_flags & MS_RDONLY))
242                 return;
243
244         errstr = ext3_decode_error(sb, errno, nbuf);
245         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
246                 sb->s_id, function, errstr);
247
248         ext3_handle_error(sb);
249 }
250
251 /*
252  * ext3_abort is a much stronger failure handler than ext3_error.  The
253  * abort function may be used to deal with unrecoverable failures such
254  * as journal IO errors or ENOMEM at a critical moment in log management.
255  *
256  * We unconditionally force the filesystem into an ABORT|READONLY state,
257  * unless the error response on the fs has been set to panic in which
258  * case we take the easy way out and panic immediately.
259  */
260
261 void ext3_abort (struct super_block * sb, const char * function,
262                  const char * fmt, ...)
263 {
264         va_list args;
265
266         printk (KERN_CRIT "ext3_abort called.\n");
267
268         va_start(args, fmt);
269         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
270         vprintk(fmt, args);
271         printk("\n");
272         va_end(args);
273
274         if (test_opt(sb, ERRORS_PANIC))
275                 panic("EXT3-fs panic from previous error\n");
276
277         if (sb->s_flags & MS_RDONLY)
278                 return;
279
280         printk(KERN_CRIT "Remounting filesystem read-only\n");
281         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
282         sb->s_flags |= MS_RDONLY;
283         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
284         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
285 }
286
287 void ext3_warning (struct super_block * sb, const char * function,
288                    const char * fmt, ...)
289 {
290         va_list args;
291
292         va_start(args, fmt);
293         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
294                sb->s_id, function);
295         vprintk(fmt, args);
296         printk("\n");
297         va_end(args);
298 }
299
300 void ext3_update_dynamic_rev(struct super_block *sb)
301 {
302         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
303
304         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
305                 return;
306
307         ext3_warning(sb, __FUNCTION__,
308                      "updating to rev %d because of new feature flag, "
309                      "running e2fsck is recommended",
310                      EXT3_DYNAMIC_REV);
311
312         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
313         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
314         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
315         /* leave es->s_feature_*compat flags alone */
316         /* es->s_uuid will be set by e2fsck if empty */
317
318         /*
319          * The rest of the superblock fields should be zero, and if not it
320          * means they are likely already in use, so leave them alone.  We
321          * can leave it up to e2fsck to clean up any inconsistencies there.
322          */
323 }
324
325 /*
326  * Open the external journal device
327  */
328 static struct block_device *ext3_blkdev_get(dev_t dev)
329 {
330         struct block_device *bdev;
331         char b[BDEVNAME_SIZE];
332
333         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
334         if (IS_ERR(bdev))
335                 goto fail;
336         return bdev;
337
338 fail:
339         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
340                         __bdevname(dev, b), PTR_ERR(bdev));
341         return NULL;
342 }
343
344 /*
345  * Release the journal device
346  */
347 static int ext3_blkdev_put(struct block_device *bdev)
348 {
349         bd_release(bdev);
350         return blkdev_put(bdev);
351 }
352
353 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
354 {
355         struct block_device *bdev;
356         int ret = -ENODEV;
357
358         bdev = sbi->journal_bdev;
359         if (bdev) {
360                 ret = ext3_blkdev_put(bdev);
361                 sbi->journal_bdev = NULL;
362         }
363         return ret;
364 }
365
366 static inline struct inode *orphan_list_entry(struct list_head *l)
367 {
368         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
369 }
370
371 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
372 {
373         struct list_head *l;
374
375         printk(KERN_ERR "sb orphan head is %d\n",
376                le32_to_cpu(sbi->s_es->s_last_orphan));
377
378         printk(KERN_ERR "sb_info orphan list:\n");
379         list_for_each(l, &sbi->s_orphan) {
380                 struct inode *inode = orphan_list_entry(l);
381                 printk(KERN_ERR "  "
382                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
383                        inode->i_sb->s_id, inode->i_ino, inode,
384                        inode->i_mode, inode->i_nlink,
385                        NEXT_ORPHAN(inode));
386         }
387 }
388
389 static void ext3_put_super (struct super_block * sb)
390 {
391         struct ext3_sb_info *sbi = EXT3_SB(sb);
392         struct ext3_super_block *es = sbi->s_es;
393         int i;
394
395         ext3_xattr_put_super(sb);
396         journal_destroy(sbi->s_journal);
397         if (!(sb->s_flags & MS_RDONLY)) {
398                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
399                 es->s_state = cpu_to_le16(sbi->s_mount_state);
400                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
401                 mark_buffer_dirty(sbi->s_sbh);
402                 ext3_commit_super(sb, es, 1);
403         }
404
405         for (i = 0; i < sbi->s_gdb_count; i++)
406                 brelse(sbi->s_group_desc[i]);
407         kfree(sbi->s_group_desc);
408         percpu_counter_destroy(&sbi->s_freeblocks_counter);
409         percpu_counter_destroy(&sbi->s_freeinodes_counter);
410         percpu_counter_destroy(&sbi->s_dirs_counter);
411         brelse(sbi->s_sbh);
412 #ifdef CONFIG_QUOTA
413         for (i = 0; i < MAXQUOTAS; i++)
414                 kfree(sbi->s_qf_names[i]);
415 #endif
416
417         /* Debugging code just in case the in-memory inode orphan list
418          * isn't empty.  The on-disk one can be non-empty if we've
419          * detected an error and taken the fs readonly, but the
420          * in-memory list had better be clean by this point. */
421         if (!list_empty(&sbi->s_orphan))
422                 dump_orphan_list(sb, sbi);
423         J_ASSERT(list_empty(&sbi->s_orphan));
424
425         invalidate_bdev(sb->s_bdev);
426         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
427                 /*
428                  * Invalidate the journal device's buffers.  We don't want them
429                  * floating about in memory - the physical journal device may
430                  * hotswapped, and it breaks the `ro-after' testing code.
431                  */
432                 sync_blockdev(sbi->journal_bdev);
433                 invalidate_bdev(sbi->journal_bdev);
434                 ext3_blkdev_remove(sbi);
435         }
436         sb->s_fs_info = NULL;
437         kfree(sbi);
438         return;
439 }
440
441 static struct kmem_cache *ext3_inode_cachep;
442
443 /*
444  * Called inside transaction, so use GFP_NOFS
445  */
446 static struct inode *ext3_alloc_inode(struct super_block *sb)
447 {
448         struct ext3_inode_info *ei;
449
450         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
451         if (!ei)
452                 return NULL;
453 #ifdef CONFIG_EXT3_FS_POSIX_ACL
454         ei->i_acl = EXT3_ACL_NOT_CACHED;
455         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
456 #endif
457         ei->i_block_alloc_info = NULL;
458         ei->vfs_inode.i_version = 1;
459         return &ei->vfs_inode;
460 }
461
462 static void ext3_destroy_inode(struct inode *inode)
463 {
464         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
465                 printk("EXT3 Inode %p: orphan list check failed!\n",
466                         EXT3_I(inode));
467                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
468                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
469                                 false);
470                 dump_stack();
471         }
472         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
473 }
474
475 static void init_once(struct kmem_cache * cachep, void *foo)
476 {
477         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
478
479         INIT_LIST_HEAD(&ei->i_orphan);
480 #ifdef CONFIG_EXT3_FS_XATTR
481         init_rwsem(&ei->xattr_sem);
482 #endif
483         mutex_init(&ei->truncate_mutex);
484         inode_init_once(&ei->vfs_inode);
485 }
486
487 static int init_inodecache(void)
488 {
489         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
490                                              sizeof(struct ext3_inode_info),
491                                              0, (SLAB_RECLAIM_ACCOUNT|
492                                                 SLAB_MEM_SPREAD),
493                                              init_once);
494         if (ext3_inode_cachep == NULL)
495                 return -ENOMEM;
496         return 0;
497 }
498
499 static void destroy_inodecache(void)
500 {
501         kmem_cache_destroy(ext3_inode_cachep);
502 }
503
504 static void ext3_clear_inode(struct inode *inode)
505 {
506         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
507 #ifdef CONFIG_EXT3_FS_POSIX_ACL
508         if (EXT3_I(inode)->i_acl &&
509                         EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
510                 posix_acl_release(EXT3_I(inode)->i_acl);
511                 EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
512         }
513         if (EXT3_I(inode)->i_default_acl &&
514                         EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
515                 posix_acl_release(EXT3_I(inode)->i_default_acl);
516                 EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
517         }
518 #endif
519         ext3_discard_reservation(inode);
520         EXT3_I(inode)->i_block_alloc_info = NULL;
521         if (unlikely(rsv))
522                 kfree(rsv);
523 }
524
525 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
526 {
527 #if defined(CONFIG_QUOTA)
528         struct ext3_sb_info *sbi = EXT3_SB(sb);
529
530         if (sbi->s_jquota_fmt)
531                 seq_printf(seq, ",jqfmt=%s",
532                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
533
534         if (sbi->s_qf_names[USRQUOTA])
535                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
536
537         if (sbi->s_qf_names[GRPQUOTA])
538                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
539
540         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
541                 seq_puts(seq, ",usrquota");
542
543         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
544                 seq_puts(seq, ",grpquota");
545 #endif
546 }
547
548 /*
549  * Show an option if
550  *  - it's set to a non-default value OR
551  *  - if the per-sb default is different from the global default
552  */
553 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
554 {
555         struct super_block *sb = vfs->mnt_sb;
556         struct ext3_sb_info *sbi = EXT3_SB(sb);
557         struct ext3_super_block *es = sbi->s_es;
558         unsigned long def_mount_opts;
559
560         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
561
562         if (sbi->s_sb_block != 1)
563                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
564         if (test_opt(sb, MINIX_DF))
565                 seq_puts(seq, ",minixdf");
566         if (test_opt(sb, GRPID))
567                 seq_puts(seq, ",grpid");
568         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
569                 seq_puts(seq, ",nogrpid");
570         if (sbi->s_resuid != EXT3_DEF_RESUID ||
571             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
572                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
573         }
574         if (sbi->s_resgid != EXT3_DEF_RESGID ||
575             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
576                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
577         }
578         if (test_opt(sb, ERRORS_CONT)) {
579                 int def_errors = le16_to_cpu(es->s_errors);
580
581                 if (def_errors == EXT3_ERRORS_PANIC ||
582                     def_errors == EXT3_ERRORS_RO) {
583                         seq_puts(seq, ",errors=continue");
584                 }
585         }
586         if (test_opt(sb, ERRORS_RO))
587                 seq_puts(seq, ",errors=remount-ro");
588         if (test_opt(sb, ERRORS_PANIC))
589                 seq_puts(seq, ",errors=panic");
590         if (test_opt(sb, NO_UID32))
591                 seq_puts(seq, ",nouid32");
592         if (test_opt(sb, DEBUG))
593                 seq_puts(seq, ",debug");
594         if (test_opt(sb, OLDALLOC))
595                 seq_puts(seq, ",oldalloc");
596 #ifdef CONFIG_EXT3_FS_XATTR
597         if (test_opt(sb, XATTR_USER))
598                 seq_puts(seq, ",user_xattr");
599         if (!test_opt(sb, XATTR_USER) &&
600             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
601                 seq_puts(seq, ",nouser_xattr");
602         }
603 #endif
604 #ifdef CONFIG_EXT3_FS_POSIX_ACL
605         if (test_opt(sb, POSIX_ACL))
606                 seq_puts(seq, ",acl");
607         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
608                 seq_puts(seq, ",noacl");
609 #endif
610         if (!test_opt(sb, RESERVATION))
611                 seq_puts(seq, ",noreservation");
612         if (sbi->s_commit_interval) {
613                 seq_printf(seq, ",commit=%u",
614                            (unsigned) (sbi->s_commit_interval / HZ));
615         }
616         if (test_opt(sb, BARRIER))
617                 seq_puts(seq, ",barrier=1");
618         if (test_opt(sb, NOBH))
619                 seq_puts(seq, ",nobh");
620
621         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
622                 seq_puts(seq, ",data=journal");
623         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
624                 seq_puts(seq, ",data=ordered");
625         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
626                 seq_puts(seq, ",data=writeback");
627
628         ext3_show_quota_options(seq, sb);
629
630         return 0;
631 }
632
633
634 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
635                 u64 ino, u32 generation)
636 {
637         struct inode *inode;
638
639         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
640                 return ERR_PTR(-ESTALE);
641         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
642                 return ERR_PTR(-ESTALE);
643
644         /* iget isn't really right if the inode is currently unallocated!!
645          *
646          * ext3_read_inode will return a bad_inode if the inode had been
647          * deleted, so we should be safe.
648          *
649          * Currently we don't know the generation for parent directory, so
650          * a generation of 0 means "accept any"
651          */
652         inode = iget(sb, ino);
653         if (inode == NULL)
654                 return ERR_PTR(-ENOMEM);
655         if (is_bad_inode(inode) ||
656             (generation && inode->i_generation != generation)) {
657                 iput(inode);
658                 return ERR_PTR(-ESTALE);
659         }
660
661         return inode;
662 }
663
664 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
665                 int fh_len, int fh_type)
666 {
667         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
668                                     ext3_nfs_get_inode);
669 }
670
671 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
672                 int fh_len, int fh_type)
673 {
674         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
675                                     ext3_nfs_get_inode);
676 }
677
678 #ifdef CONFIG_QUOTA
679 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
680 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
681
682 static int ext3_dquot_initialize(struct inode *inode, int type);
683 static int ext3_dquot_drop(struct inode *inode);
684 static int ext3_write_dquot(struct dquot *dquot);
685 static int ext3_acquire_dquot(struct dquot *dquot);
686 static int ext3_release_dquot(struct dquot *dquot);
687 static int ext3_mark_dquot_dirty(struct dquot *dquot);
688 static int ext3_write_info(struct super_block *sb, int type);
689 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
690 static int ext3_quota_on_mount(struct super_block *sb, int type);
691 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
692                                size_t len, loff_t off);
693 static ssize_t ext3_quota_write(struct super_block *sb, int type,
694                                 const char *data, size_t len, loff_t off);
695
696 static struct dquot_operations ext3_quota_operations = {
697         .initialize     = ext3_dquot_initialize,
698         .drop           = ext3_dquot_drop,
699         .alloc_space    = dquot_alloc_space,
700         .alloc_inode    = dquot_alloc_inode,
701         .free_space     = dquot_free_space,
702         .free_inode     = dquot_free_inode,
703         .transfer       = dquot_transfer,
704         .write_dquot    = ext3_write_dquot,
705         .acquire_dquot  = ext3_acquire_dquot,
706         .release_dquot  = ext3_release_dquot,
707         .mark_dirty     = ext3_mark_dquot_dirty,
708         .write_info     = ext3_write_info
709 };
710
711 static struct quotactl_ops ext3_qctl_operations = {
712         .quota_on       = ext3_quota_on,
713         .quota_off      = vfs_quota_off,
714         .quota_sync     = vfs_quota_sync,
715         .get_info       = vfs_get_dqinfo,
716         .set_info       = vfs_set_dqinfo,
717         .get_dqblk      = vfs_get_dqblk,
718         .set_dqblk      = vfs_set_dqblk
719 };
720 #endif
721
722 static const struct super_operations ext3_sops = {
723         .alloc_inode    = ext3_alloc_inode,
724         .destroy_inode  = ext3_destroy_inode,
725         .read_inode     = ext3_read_inode,
726         .write_inode    = ext3_write_inode,
727         .dirty_inode    = ext3_dirty_inode,
728         .delete_inode   = ext3_delete_inode,
729         .put_super      = ext3_put_super,
730         .write_super    = ext3_write_super,
731         .sync_fs        = ext3_sync_fs,
732         .write_super_lockfs = ext3_write_super_lockfs,
733         .unlockfs       = ext3_unlockfs,
734         .statfs         = ext3_statfs,
735         .remount_fs     = ext3_remount,
736         .clear_inode    = ext3_clear_inode,
737         .show_options   = ext3_show_options,
738 #ifdef CONFIG_QUOTA
739         .quota_read     = ext3_quota_read,
740         .quota_write    = ext3_quota_write,
741 #endif
742 };
743
744 static struct export_operations ext3_export_ops = {
745         .fh_to_dentry = ext3_fh_to_dentry,
746         .fh_to_parent = ext3_fh_to_parent,
747         .get_parent = ext3_get_parent,
748 };
749
750 enum {
751         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
752         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
753         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
754         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
755         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
756         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
757         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
758         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
759         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
760         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
761         Opt_grpquota
762 };
763
764 static match_table_t tokens = {
765         {Opt_bsd_df, "bsddf"},
766         {Opt_minix_df, "minixdf"},
767         {Opt_grpid, "grpid"},
768         {Opt_grpid, "bsdgroups"},
769         {Opt_nogrpid, "nogrpid"},
770         {Opt_nogrpid, "sysvgroups"},
771         {Opt_resgid, "resgid=%u"},
772         {Opt_resuid, "resuid=%u"},
773         {Opt_sb, "sb=%u"},
774         {Opt_err_cont, "errors=continue"},
775         {Opt_err_panic, "errors=panic"},
776         {Opt_err_ro, "errors=remount-ro"},
777         {Opt_nouid32, "nouid32"},
778         {Opt_nocheck, "nocheck"},
779         {Opt_nocheck, "check=none"},
780         {Opt_debug, "debug"},
781         {Opt_oldalloc, "oldalloc"},
782         {Opt_orlov, "orlov"},
783         {Opt_user_xattr, "user_xattr"},
784         {Opt_nouser_xattr, "nouser_xattr"},
785         {Opt_acl, "acl"},
786         {Opt_noacl, "noacl"},
787         {Opt_reservation, "reservation"},
788         {Opt_noreservation, "noreservation"},
789         {Opt_noload, "noload"},
790         {Opt_nobh, "nobh"},
791         {Opt_bh, "bh"},
792         {Opt_commit, "commit=%u"},
793         {Opt_journal_update, "journal=update"},
794         {Opt_journal_inum, "journal=%u"},
795         {Opt_journal_dev, "journal_dev=%u"},
796         {Opt_abort, "abort"},
797         {Opt_data_journal, "data=journal"},
798         {Opt_data_ordered, "data=ordered"},
799         {Opt_data_writeback, "data=writeback"},
800         {Opt_offusrjquota, "usrjquota="},
801         {Opt_usrjquota, "usrjquota=%s"},
802         {Opt_offgrpjquota, "grpjquota="},
803         {Opt_grpjquota, "grpjquota=%s"},
804         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
805         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
806         {Opt_grpquota, "grpquota"},
807         {Opt_noquota, "noquota"},
808         {Opt_quota, "quota"},
809         {Opt_usrquota, "usrquota"},
810         {Opt_barrier, "barrier=%u"},
811         {Opt_err, NULL},
812         {Opt_resize, "resize"},
813 };
814
815 static ext3_fsblk_t get_sb_block(void **data)
816 {
817         ext3_fsblk_t    sb_block;
818         char            *options = (char *) *data;
819
820         if (!options || strncmp(options, "sb=", 3) != 0)
821                 return 1;       /* Default location */
822         options += 3;
823         /*todo: use simple_strtoll with >32bit ext3 */
824         sb_block = simple_strtoul(options, &options, 0);
825         if (*options && *options != ',') {
826                 printk("EXT3-fs: Invalid sb specification: %s\n",
827                        (char *) *data);
828                 return 1;
829         }
830         if (*options == ',')
831                 options++;
832         *data = (void *) options;
833         return sb_block;
834 }
835
836 static int parse_options (char *options, struct super_block *sb,
837                           unsigned int *inum, unsigned long *journal_devnum,
838                           ext3_fsblk_t *n_blocks_count, int is_remount)
839 {
840         struct ext3_sb_info *sbi = EXT3_SB(sb);
841         char * p;
842         substring_t args[MAX_OPT_ARGS];
843         int data_opt = 0;
844         int option;
845 #ifdef CONFIG_QUOTA
846         int qtype;
847         char *qname;
848 #endif
849
850         if (!options)
851                 return 1;
852
853         while ((p = strsep (&options, ",")) != NULL) {
854                 int token;
855                 if (!*p)
856                         continue;
857
858                 token = match_token(p, tokens, args);
859                 switch (token) {
860                 case Opt_bsd_df:
861                         clear_opt (sbi->s_mount_opt, MINIX_DF);
862                         break;
863                 case Opt_minix_df:
864                         set_opt (sbi->s_mount_opt, MINIX_DF);
865                         break;
866                 case Opt_grpid:
867                         set_opt (sbi->s_mount_opt, GRPID);
868                         break;
869                 case Opt_nogrpid:
870                         clear_opt (sbi->s_mount_opt, GRPID);
871                         break;
872                 case Opt_resuid:
873                         if (match_int(&args[0], &option))
874                                 return 0;
875                         sbi->s_resuid = option;
876                         break;
877                 case Opt_resgid:
878                         if (match_int(&args[0], &option))
879                                 return 0;
880                         sbi->s_resgid = option;
881                         break;
882                 case Opt_sb:
883                         /* handled by get_sb_block() instead of here */
884                         /* *sb_block = match_int(&args[0]); */
885                         break;
886                 case Opt_err_panic:
887                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
888                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
889                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
890                         break;
891                 case Opt_err_ro:
892                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
893                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
894                         set_opt (sbi->s_mount_opt, ERRORS_RO);
895                         break;
896                 case Opt_err_cont:
897                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
898                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
899                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
900                         break;
901                 case Opt_nouid32:
902                         set_opt (sbi->s_mount_opt, NO_UID32);
903                         break;
904                 case Opt_nocheck:
905                         clear_opt (sbi->s_mount_opt, CHECK);
906                         break;
907                 case Opt_debug:
908                         set_opt (sbi->s_mount_opt, DEBUG);
909                         break;
910                 case Opt_oldalloc:
911                         set_opt (sbi->s_mount_opt, OLDALLOC);
912                         break;
913                 case Opt_orlov:
914                         clear_opt (sbi->s_mount_opt, OLDALLOC);
915                         break;
916 #ifdef CONFIG_EXT3_FS_XATTR
917                 case Opt_user_xattr:
918                         set_opt (sbi->s_mount_opt, XATTR_USER);
919                         break;
920                 case Opt_nouser_xattr:
921                         clear_opt (sbi->s_mount_opt, XATTR_USER);
922                         break;
923 #else
924                 case Opt_user_xattr:
925                 case Opt_nouser_xattr:
926                         printk("EXT3 (no)user_xattr options not supported\n");
927                         break;
928 #endif
929 #ifdef CONFIG_EXT3_FS_POSIX_ACL
930                 case Opt_acl:
931                         set_opt(sbi->s_mount_opt, POSIX_ACL);
932                         break;
933                 case Opt_noacl:
934                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
935                         break;
936 #else
937                 case Opt_acl:
938                 case Opt_noacl:
939                         printk("EXT3 (no)acl options not supported\n");
940                         break;
941 #endif
942                 case Opt_reservation:
943                         set_opt(sbi->s_mount_opt, RESERVATION);
944                         break;
945                 case Opt_noreservation:
946                         clear_opt(sbi->s_mount_opt, RESERVATION);
947                         break;
948                 case Opt_journal_update:
949                         /* @@@ FIXME */
950                         /* Eventually we will want to be able to create
951                            a journal file here.  For now, only allow the
952                            user to specify an existing inode to be the
953                            journal file. */
954                         if (is_remount) {
955                                 printk(KERN_ERR "EXT3-fs: cannot specify "
956                                        "journal on remount\n");
957                                 return 0;
958                         }
959                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
960                         break;
961                 case Opt_journal_inum:
962                         if (is_remount) {
963                                 printk(KERN_ERR "EXT3-fs: cannot specify "
964                                        "journal on remount\n");
965                                 return 0;
966                         }
967                         if (match_int(&args[0], &option))
968                                 return 0;
969                         *inum = option;
970                         break;
971                 case Opt_journal_dev:
972                         if (is_remount) {
973                                 printk(KERN_ERR "EXT3-fs: cannot specify "
974                                        "journal on remount\n");
975                                 return 0;
976                         }
977                         if (match_int(&args[0], &option))
978                                 return 0;
979                         *journal_devnum = option;
980                         break;
981                 case Opt_noload:
982                         set_opt (sbi->s_mount_opt, NOLOAD);
983                         break;
984                 case Opt_commit:
985                         if (match_int(&args[0], &option))
986                                 return 0;
987                         if (option < 0)
988                                 return 0;
989                         if (option == 0)
990                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
991                         sbi->s_commit_interval = HZ * option;
992                         break;
993                 case Opt_data_journal:
994                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
995                         goto datacheck;
996                 case Opt_data_ordered:
997                         data_opt = EXT3_MOUNT_ORDERED_DATA;
998                         goto datacheck;
999                 case Opt_data_writeback:
1000                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1001                 datacheck:
1002                         if (is_remount) {
1003                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
1004                                                 != data_opt) {
1005                                         printk(KERN_ERR
1006                                                 "EXT3-fs: cannot change data "
1007                                                 "mode on remount\n");
1008                                         return 0;
1009                                 }
1010                         } else {
1011                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
1012                                 sbi->s_mount_opt |= data_opt;
1013                         }
1014                         break;
1015 #ifdef CONFIG_QUOTA
1016                 case Opt_usrjquota:
1017                         qtype = USRQUOTA;
1018                         goto set_qf_name;
1019                 case Opt_grpjquota:
1020                         qtype = GRPQUOTA;
1021 set_qf_name:
1022                         if (sb_any_quota_enabled(sb)) {
1023                                 printk(KERN_ERR
1024                                         "EXT3-fs: Cannot change journalled "
1025                                         "quota options when quota turned on.\n");
1026                                 return 0;
1027                         }
1028                         qname = match_strdup(&args[0]);
1029                         if (!qname) {
1030                                 printk(KERN_ERR
1031                                         "EXT3-fs: not enough memory for "
1032                                         "storing quotafile name.\n");
1033                                 return 0;
1034                         }
1035                         if (sbi->s_qf_names[qtype] &&
1036                             strcmp(sbi->s_qf_names[qtype], qname)) {
1037                                 printk(KERN_ERR
1038                                         "EXT3-fs: %s quota file already "
1039                                         "specified.\n", QTYPE2NAME(qtype));
1040                                 kfree(qname);
1041                                 return 0;
1042                         }
1043                         sbi->s_qf_names[qtype] = qname;
1044                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1045                                 printk(KERN_ERR
1046                                         "EXT3-fs: quotafile must be on "
1047                                         "filesystem root.\n");
1048                                 kfree(sbi->s_qf_names[qtype]);
1049                                 sbi->s_qf_names[qtype] = NULL;
1050                                 return 0;
1051                         }
1052                         set_opt(sbi->s_mount_opt, QUOTA);
1053                         break;
1054                 case Opt_offusrjquota:
1055                         qtype = USRQUOTA;
1056                         goto clear_qf_name;
1057                 case Opt_offgrpjquota:
1058                         qtype = GRPQUOTA;
1059 clear_qf_name:
1060                         if (sb_any_quota_enabled(sb)) {
1061                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1062                                         "journalled quota options when "
1063                                         "quota turned on.\n");
1064                                 return 0;
1065                         }
1066                         /*
1067                          * The space will be released later when all options
1068                          * are confirmed to be correct
1069                          */
1070                         sbi->s_qf_names[qtype] = NULL;
1071                         break;
1072                 case Opt_jqfmt_vfsold:
1073                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
1074                         break;
1075                 case Opt_jqfmt_vfsv0:
1076                         sbi->s_jquota_fmt = QFMT_VFS_V0;
1077                         break;
1078                 case Opt_quota:
1079                 case Opt_usrquota:
1080                         set_opt(sbi->s_mount_opt, QUOTA);
1081                         set_opt(sbi->s_mount_opt, USRQUOTA);
1082                         break;
1083                 case Opt_grpquota:
1084                         set_opt(sbi->s_mount_opt, QUOTA);
1085                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1086                         break;
1087                 case Opt_noquota:
1088                         if (sb_any_quota_enabled(sb)) {
1089                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
1090                                         "options when quota turned on.\n");
1091                                 return 0;
1092                         }
1093                         clear_opt(sbi->s_mount_opt, QUOTA);
1094                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1095                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1096                         break;
1097 #else
1098                 case Opt_quota:
1099                 case Opt_usrquota:
1100                 case Opt_grpquota:
1101                 case Opt_usrjquota:
1102                 case Opt_grpjquota:
1103                 case Opt_offusrjquota:
1104                 case Opt_offgrpjquota:
1105                 case Opt_jqfmt_vfsold:
1106                 case Opt_jqfmt_vfsv0:
1107                         printk(KERN_ERR
1108                                 "EXT3-fs: journalled quota options not "
1109                                 "supported.\n");
1110                         break;
1111                 case Opt_noquota:
1112                         break;
1113 #endif
1114                 case Opt_abort:
1115                         set_opt(sbi->s_mount_opt, ABORT);
1116                         break;
1117                 case Opt_barrier:
1118                         if (match_int(&args[0], &option))
1119                                 return 0;
1120                         if (option)
1121                                 set_opt(sbi->s_mount_opt, BARRIER);
1122                         else
1123                                 clear_opt(sbi->s_mount_opt, BARRIER);
1124                         break;
1125                 case Opt_ignore:
1126                         break;
1127                 case Opt_resize:
1128                         if (!is_remount) {
1129                                 printk("EXT3-fs: resize option only available "
1130                                         "for remount\n");
1131                                 return 0;
1132                         }
1133                         if (match_int(&args[0], &option) != 0)
1134                                 return 0;
1135                         *n_blocks_count = option;
1136                         break;
1137                 case Opt_nobh:
1138                         set_opt(sbi->s_mount_opt, NOBH);
1139                         break;
1140                 case Opt_bh:
1141                         clear_opt(sbi->s_mount_opt, NOBH);
1142                         break;
1143                 default:
1144                         printk (KERN_ERR
1145                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1146                                 "or missing value\n", p);
1147                         return 0;
1148                 }
1149         }
1150 #ifdef CONFIG_QUOTA
1151         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1152                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1153                      sbi->s_qf_names[USRQUOTA])
1154                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1155
1156                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1157                      sbi->s_qf_names[GRPQUOTA])
1158                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1159
1160                 if ((sbi->s_qf_names[USRQUOTA] &&
1161                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1162                     (sbi->s_qf_names[GRPQUOTA] &&
1163                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1164                         printk(KERN_ERR "EXT3-fs: old and new quota "
1165                                         "format mixing.\n");
1166                         return 0;
1167                 }
1168
1169                 if (!sbi->s_jquota_fmt) {
1170                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1171                                         "not specified.\n");
1172                         return 0;
1173                 }
1174         } else {
1175                 if (sbi->s_jquota_fmt) {
1176                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1177                                         "specified with no journalling "
1178                                         "enabled.\n");
1179                         return 0;
1180                 }
1181         }
1182 #endif
1183         return 1;
1184 }
1185
1186 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1187                             int read_only)
1188 {
1189         struct ext3_sb_info *sbi = EXT3_SB(sb);
1190         int res = 0;
1191
1192         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1193                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1194                         "forcing read-only mode\n");
1195                 res = MS_RDONLY;
1196         }
1197         if (read_only)
1198                 return res;
1199         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1200                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1201                         "running e2fsck is recommended\n");
1202         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1203                 printk (KERN_WARNING
1204                         "EXT3-fs warning: mounting fs with errors, "
1205                         "running e2fsck is recommended\n");
1206         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1207                  le16_to_cpu(es->s_mnt_count) >=
1208                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1209                 printk (KERN_WARNING
1210                         "EXT3-fs warning: maximal mount count reached, "
1211                         "running e2fsck is recommended\n");
1212         else if (le32_to_cpu(es->s_checkinterval) &&
1213                 (le32_to_cpu(es->s_lastcheck) +
1214                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1215                 printk (KERN_WARNING
1216                         "EXT3-fs warning: checktime reached, "
1217                         "running e2fsck is recommended\n");
1218 #if 0
1219                 /* @@@ We _will_ want to clear the valid bit if we find
1220                    inconsistencies, to force a fsck at reboot.  But for
1221                    a plain journaled filesystem we can keep it set as
1222                    valid forever! :) */
1223         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1224 #endif
1225         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1226                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1227         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1228         es->s_mtime = cpu_to_le32(get_seconds());
1229         ext3_update_dynamic_rev(sb);
1230         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1231
1232         ext3_commit_super(sb, es, 1);
1233         if (test_opt(sb, DEBUG))
1234                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1235                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1236                         sb->s_blocksize,
1237                         sbi->s_groups_count,
1238                         EXT3_BLOCKS_PER_GROUP(sb),
1239                         EXT3_INODES_PER_GROUP(sb),
1240                         sbi->s_mount_opt);
1241
1242         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1243         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1244                 char b[BDEVNAME_SIZE];
1245
1246                 printk("external journal on %s\n",
1247                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1248         } else {
1249                 printk("internal journal\n");
1250         }
1251         return res;
1252 }
1253
1254 /* Called at mount-time, super-block is locked */
1255 static int ext3_check_descriptors (struct super_block * sb)
1256 {
1257         struct ext3_sb_info *sbi = EXT3_SB(sb);
1258         ext3_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1259         ext3_fsblk_t last_block;
1260         struct ext3_group_desc * gdp = NULL;
1261         int desc_block = 0;
1262         int i;
1263
1264         ext3_debug ("Checking group descriptors");
1265
1266         for (i = 0; i < sbi->s_groups_count; i++)
1267         {
1268                 if (i == sbi->s_groups_count - 1)
1269                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1270                 else
1271                         last_block = first_block +
1272                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1273
1274                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1275                         gdp = (struct ext3_group_desc *)
1276                                         sbi->s_group_desc[desc_block++]->b_data;
1277                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1278                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1279                 {
1280                         ext3_error (sb, "ext3_check_descriptors",
1281                                     "Block bitmap for group %d"
1282                                     " not in group (block %lu)!",
1283                                     i, (unsigned long)
1284                                         le32_to_cpu(gdp->bg_block_bitmap));
1285                         return 0;
1286                 }
1287                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1288                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1289                 {
1290                         ext3_error (sb, "ext3_check_descriptors",
1291                                     "Inode bitmap for group %d"
1292                                     " not in group (block %lu)!",
1293                                     i, (unsigned long)
1294                                         le32_to_cpu(gdp->bg_inode_bitmap));
1295                         return 0;
1296                 }
1297                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1298                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1299                     last_block)
1300                 {
1301                         ext3_error (sb, "ext3_check_descriptors",
1302                                     "Inode table for group %d"
1303                                     " not in group (block %lu)!",
1304                                     i, (unsigned long)
1305                                         le32_to_cpu(gdp->bg_inode_table));
1306                         return 0;
1307                 }
1308                 first_block += EXT3_BLOCKS_PER_GROUP(sb);
1309                 gdp++;
1310         }
1311
1312         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1313         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1314         return 1;
1315 }
1316
1317
1318 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1319  * the superblock) which were deleted from all directories, but held open by
1320  * a process at the time of a crash.  We walk the list and try to delete these
1321  * inodes at recovery time (only with a read-write filesystem).
1322  *
1323  * In order to keep the orphan inode chain consistent during traversal (in
1324  * case of crash during recovery), we link each inode into the superblock
1325  * orphan list_head and handle it the same way as an inode deletion during
1326  * normal operation (which journals the operations for us).
1327  *
1328  * We only do an iget() and an iput() on each inode, which is very safe if we
1329  * accidentally point at an in-use or already deleted inode.  The worst that
1330  * can happen in this case is that we get a "bit already cleared" message from
1331  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1332  * e2fsck was run on this filesystem, and it must have already done the orphan
1333  * inode cleanup for us, so we can safely abort without any further action.
1334  */
1335 static void ext3_orphan_cleanup (struct super_block * sb,
1336                                  struct ext3_super_block * es)
1337 {
1338         unsigned int s_flags = sb->s_flags;
1339         int nr_orphans = 0, nr_truncates = 0;
1340 #ifdef CONFIG_QUOTA
1341         int i;
1342 #endif
1343         if (!es->s_last_orphan) {
1344                 jbd_debug(4, "no orphan inodes to clean up\n");
1345                 return;
1346         }
1347
1348         if (bdev_read_only(sb->s_bdev)) {
1349                 printk(KERN_ERR "EXT3-fs: write access "
1350                         "unavailable, skipping orphan cleanup.\n");
1351                 return;
1352         }
1353
1354         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1355                 if (es->s_last_orphan)
1356                         jbd_debug(1, "Errors on filesystem, "
1357                                   "clearing orphan list.\n");
1358                 es->s_last_orphan = 0;
1359                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1360                 return;
1361         }
1362
1363         if (s_flags & MS_RDONLY) {
1364                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1365                        sb->s_id);
1366                 sb->s_flags &= ~MS_RDONLY;
1367         }
1368 #ifdef CONFIG_QUOTA
1369         /* Needed for iput() to work correctly and not trash data */
1370         sb->s_flags |= MS_ACTIVE;
1371         /* Turn on quotas so that they are updated correctly */
1372         for (i = 0; i < MAXQUOTAS; i++) {
1373                 if (EXT3_SB(sb)->s_qf_names[i]) {
1374                         int ret = ext3_quota_on_mount(sb, i);
1375                         if (ret < 0)
1376                                 printk(KERN_ERR
1377                                         "EXT3-fs: Cannot turn on journalled "
1378                                         "quota: error %d\n", ret);
1379                 }
1380         }
1381 #endif
1382
1383         while (es->s_last_orphan) {
1384                 struct inode *inode;
1385
1386                 if (!(inode =
1387                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1388                         es->s_last_orphan = 0;
1389                         break;
1390                 }
1391
1392                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1393                 DQUOT_INIT(inode);
1394                 if (inode->i_nlink) {
1395                         printk(KERN_DEBUG
1396                                 "%s: truncating inode %lu to %Ld bytes\n",
1397                                 __FUNCTION__, inode->i_ino, inode->i_size);
1398                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1399                                   inode->i_ino, inode->i_size);
1400                         ext3_truncate(inode);
1401                         nr_truncates++;
1402                 } else {
1403                         printk(KERN_DEBUG
1404                                 "%s: deleting unreferenced inode %lu\n",
1405                                 __FUNCTION__, inode->i_ino);
1406                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1407                                   inode->i_ino);
1408                         nr_orphans++;
1409                 }
1410                 iput(inode);  /* The delete magic happens here! */
1411         }
1412
1413 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1414
1415         if (nr_orphans)
1416                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1417                        sb->s_id, PLURAL(nr_orphans));
1418         if (nr_truncates)
1419                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1420                        sb->s_id, PLURAL(nr_truncates));
1421 #ifdef CONFIG_QUOTA
1422         /* Turn quotas off */
1423         for (i = 0; i < MAXQUOTAS; i++) {
1424                 if (sb_dqopt(sb)->files[i])
1425                         vfs_quota_off(sb, i);
1426         }
1427 #endif
1428         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1429 }
1430
1431 /*
1432  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1433  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1434  * We need to be 1 filesystem block less than the 2^32 sector limit.
1435  */
1436 static loff_t ext3_max_size(int bits)
1437 {
1438         loff_t res = EXT3_NDIR_BLOCKS;
1439         /* This constant is calculated to be the largest file size for a
1440          * dense, 4k-blocksize file such that the total number of
1441          * sectors in the file, including data and all indirect blocks,
1442          * does not exceed 2^32. */
1443         const loff_t upper_limit = 0x1ff7fffd000LL;
1444
1445         res += 1LL << (bits-2);
1446         res += 1LL << (2*(bits-2));
1447         res += 1LL << (3*(bits-2));
1448         res <<= bits;
1449         if (res > upper_limit)
1450                 res = upper_limit;
1451         return res;
1452 }
1453
1454 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1455                                     ext3_fsblk_t logic_sb_block,
1456                                     int nr)
1457 {
1458         struct ext3_sb_info *sbi = EXT3_SB(sb);
1459         unsigned long bg, first_meta_bg;
1460         int has_super = 0;
1461
1462         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1463
1464         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1465             nr < first_meta_bg)
1466                 return (logic_sb_block + nr + 1);
1467         bg = sbi->s_desc_per_block * nr;
1468         if (ext3_bg_has_super(sb, bg))
1469                 has_super = 1;
1470         return (has_super + ext3_group_first_block_no(sb, bg));
1471 }
1472
1473
1474 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1475 {
1476         struct buffer_head * bh;
1477         struct ext3_super_block *es = NULL;
1478         struct ext3_sb_info *sbi;
1479         ext3_fsblk_t block;
1480         ext3_fsblk_t sb_block = get_sb_block(&data);
1481         ext3_fsblk_t logic_sb_block;
1482         unsigned long offset = 0;
1483         unsigned int journal_inum = 0;
1484         unsigned long journal_devnum = 0;
1485         unsigned long def_mount_opts;
1486         struct inode *root;
1487         int blocksize;
1488         int hblock;
1489         int db_count;
1490         int i;
1491         int needs_recovery;
1492         __le32 features;
1493         int err;
1494
1495         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1496         if (!sbi)
1497                 return -ENOMEM;
1498         sb->s_fs_info = sbi;
1499         sbi->s_mount_opt = 0;
1500         sbi->s_resuid = EXT3_DEF_RESUID;
1501         sbi->s_resgid = EXT3_DEF_RESGID;
1502         sbi->s_sb_block = sb_block;
1503
1504         unlock_kernel();
1505
1506         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1507         if (!blocksize) {
1508                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1509                 goto out_fail;
1510         }
1511
1512         /*
1513          * The ext3 superblock will not be buffer aligned for other than 1kB
1514          * block sizes.  We need to calculate the offset from buffer start.
1515          */
1516         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1517                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1518                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1519         } else {
1520                 logic_sb_block = sb_block;
1521         }
1522
1523         if (!(bh = sb_bread(sb, logic_sb_block))) {
1524                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1525                 goto out_fail;
1526         }
1527         /*
1528          * Note: s_es must be initialized as soon as possible because
1529          *       some ext3 macro-instructions depend on its value
1530          */
1531         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1532         sbi->s_es = es;
1533         sb->s_magic = le16_to_cpu(es->s_magic);
1534         if (sb->s_magic != EXT3_SUPER_MAGIC)
1535                 goto cantfind_ext3;
1536
1537         /* Set defaults before we parse the mount options */
1538         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1539         if (def_mount_opts & EXT3_DEFM_DEBUG)
1540                 set_opt(sbi->s_mount_opt, DEBUG);
1541         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1542                 set_opt(sbi->s_mount_opt, GRPID);
1543         if (def_mount_opts & EXT3_DEFM_UID16)
1544                 set_opt(sbi->s_mount_opt, NO_UID32);
1545 #ifdef CONFIG_EXT3_FS_XATTR
1546         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1547                 set_opt(sbi->s_mount_opt, XATTR_USER);
1548 #endif
1549 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1550         if (def_mount_opts & EXT3_DEFM_ACL)
1551                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1552 #endif
1553         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1554                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1555         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1556                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1557         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1558                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1559
1560         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1561                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1562         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1563                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1564         else
1565                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1566
1567         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1568         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1569
1570         set_opt(sbi->s_mount_opt, RESERVATION);
1571
1572         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1573                             NULL, 0))
1574                 goto failed_mount;
1575
1576         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1577                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1578
1579         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1580             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1581              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1582              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1583                 printk(KERN_WARNING
1584                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1585                        "running e2fsck is recommended\n");
1586         /*
1587          * Check feature flags regardless of the revision level, since we
1588          * previously didn't change the revision level when setting the flags,
1589          * so there is a chance incompat flags are set on a rev 0 filesystem.
1590          */
1591         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1592         if (features) {
1593                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1594                        "unsupported optional features (%x).\n",
1595                        sb->s_id, le32_to_cpu(features));
1596                 goto failed_mount;
1597         }
1598         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1599         if (!(sb->s_flags & MS_RDONLY) && features) {
1600                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1601                        "unsupported optional features (%x).\n",
1602                        sb->s_id, le32_to_cpu(features));
1603                 goto failed_mount;
1604         }
1605         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1606
1607         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1608             blocksize > EXT3_MAX_BLOCK_SIZE) {
1609                 printk(KERN_ERR
1610                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1611                        blocksize, sb->s_id);
1612                 goto failed_mount;
1613         }
1614
1615         hblock = bdev_hardsect_size(sb->s_bdev);
1616         if (sb->s_blocksize != blocksize) {
1617                 /*
1618                  * Make sure the blocksize for the filesystem is larger
1619                  * than the hardware sectorsize for the machine.
1620                  */
1621                 if (blocksize < hblock) {
1622                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1623                                "device blocksize %d.\n", blocksize, hblock);
1624                         goto failed_mount;
1625                 }
1626
1627                 brelse (bh);
1628                 if (!sb_set_blocksize(sb, blocksize)) {
1629                         printk(KERN_ERR "EXT3-fs: bad blocksize %d.\n",
1630                                 blocksize);
1631                         goto out_fail;
1632                 }
1633                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1634                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1635                 bh = sb_bread(sb, logic_sb_block);
1636                 if (!bh) {
1637                         printk(KERN_ERR
1638                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1639                         goto failed_mount;
1640                 }
1641                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1642                 sbi->s_es = es;
1643                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1644                         printk (KERN_ERR
1645                                 "EXT3-fs: Magic mismatch, very weird !\n");
1646                         goto failed_mount;
1647                 }
1648         }
1649
1650         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1651
1652         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1653                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1654                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1655         } else {
1656                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1657                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1658                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1659                     (!is_power_of_2(sbi->s_inode_size)) ||
1660                     (sbi->s_inode_size > blocksize)) {
1661                         printk (KERN_ERR
1662                                 "EXT3-fs: unsupported inode size: %d\n",
1663                                 sbi->s_inode_size);
1664                         goto failed_mount;
1665                 }
1666         }
1667         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1668                                    le32_to_cpu(es->s_log_frag_size);
1669         if (blocksize != sbi->s_frag_size) {
1670                 printk(KERN_ERR
1671                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1672                        sbi->s_frag_size, blocksize);
1673                 goto failed_mount;
1674         }
1675         sbi->s_frags_per_block = 1;
1676         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1677         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1678         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1679         if (EXT3_INODE_SIZE(sb) == 0)
1680                 goto cantfind_ext3;
1681         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1682         if (sbi->s_inodes_per_block == 0)
1683                 goto cantfind_ext3;
1684         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1685                                         sbi->s_inodes_per_block;
1686         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1687         sbi->s_sbh = bh;
1688         sbi->s_mount_state = le16_to_cpu(es->s_state);
1689         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1690         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1691         for (i=0; i < 4; i++)
1692                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1693         sbi->s_def_hash_version = es->s_def_hash_version;
1694
1695         if (sbi->s_blocks_per_group > blocksize * 8) {
1696                 printk (KERN_ERR
1697                         "EXT3-fs: #blocks per group too big: %lu\n",
1698                         sbi->s_blocks_per_group);
1699                 goto failed_mount;
1700         }
1701         if (sbi->s_frags_per_group > blocksize * 8) {
1702                 printk (KERN_ERR
1703                         "EXT3-fs: #fragments per group too big: %lu\n",
1704                         sbi->s_frags_per_group);
1705                 goto failed_mount;
1706         }
1707         if (sbi->s_inodes_per_group > blocksize * 8) {
1708                 printk (KERN_ERR
1709                         "EXT3-fs: #inodes per group too big: %lu\n",
1710                         sbi->s_inodes_per_group);
1711                 goto failed_mount;
1712         }
1713
1714         if (le32_to_cpu(es->s_blocks_count) >
1715                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1716                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1717                         " too large to mount safely\n", sb->s_id);
1718                 if (sizeof(sector_t) < 8)
1719                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1720                                         "enabled\n");
1721                 goto failed_mount;
1722         }
1723
1724         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1725                 goto cantfind_ext3;
1726         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1727                                le32_to_cpu(es->s_first_data_block) - 1)
1728                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1729         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1730                    EXT3_DESC_PER_BLOCK(sb);
1731         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1732                                     GFP_KERNEL);
1733         if (sbi->s_group_desc == NULL) {
1734                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1735                 goto failed_mount;
1736         }
1737
1738         bgl_lock_init(&sbi->s_blockgroup_lock);
1739
1740         for (i = 0; i < db_count; i++) {
1741                 block = descriptor_loc(sb, logic_sb_block, i);
1742                 sbi->s_group_desc[i] = sb_bread(sb, block);
1743                 if (!sbi->s_group_desc[i]) {
1744                         printk (KERN_ERR "EXT3-fs: "
1745                                 "can't read group descriptor %d\n", i);
1746                         db_count = i;
1747                         goto failed_mount2;
1748                 }
1749         }
1750         if (!ext3_check_descriptors (sb)) {
1751                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1752                 goto failed_mount2;
1753         }
1754         sbi->s_gdb_count = db_count;
1755         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1756         spin_lock_init(&sbi->s_next_gen_lock);
1757
1758         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1759                         ext3_count_free_blocks(sb));
1760         if (!err) {
1761                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1762                                 ext3_count_free_inodes(sb));
1763         }
1764         if (!err) {
1765                 err = percpu_counter_init(&sbi->s_dirs_counter,
1766                                 ext3_count_dirs(sb));
1767         }
1768         if (err) {
1769                 printk(KERN_ERR "EXT3-fs: insufficient memory\n");
1770                 goto failed_mount3;
1771         }
1772
1773         /* per fileystem reservation list head & lock */
1774         spin_lock_init(&sbi->s_rsv_window_lock);
1775         sbi->s_rsv_window_root = RB_ROOT;
1776         /* Add a single, static dummy reservation to the start of the
1777          * reservation window list --- it gives us a placeholder for
1778          * append-at-start-of-list which makes the allocation logic
1779          * _much_ simpler. */
1780         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1781         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1782         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1783         sbi->s_rsv_window_head.rsv_goal_size = 0;
1784         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1785
1786         /*
1787          * set up enough so that it can read an inode
1788          */
1789         sb->s_op = &ext3_sops;
1790         sb->s_export_op = &ext3_export_ops;
1791         sb->s_xattr = ext3_xattr_handlers;
1792 #ifdef CONFIG_QUOTA
1793         sb->s_qcop = &ext3_qctl_operations;
1794         sb->dq_op = &ext3_quota_operations;
1795 #endif
1796         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1797
1798         sb->s_root = NULL;
1799
1800         needs_recovery = (es->s_last_orphan != 0 ||
1801                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1802                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1803
1804         /*
1805          * The first inode we look at is the journal inode.  Don't try
1806          * root first: it may be modified in the journal!
1807          */
1808         if (!test_opt(sb, NOLOAD) &&
1809             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1810                 if (ext3_load_journal(sb, es, journal_devnum))
1811                         goto failed_mount3;
1812         } else if (journal_inum) {
1813                 if (ext3_create_journal(sb, es, journal_inum))
1814                         goto failed_mount3;
1815         } else {
1816                 if (!silent)
1817                         printk (KERN_ERR
1818                                 "ext3: No journal on filesystem on %s\n",
1819                                 sb->s_id);
1820                 goto failed_mount3;
1821         }
1822
1823         /* We have now updated the journal if required, so we can
1824          * validate the data journaling mode. */
1825         switch (test_opt(sb, DATA_FLAGS)) {
1826         case 0:
1827                 /* No mode set, assume a default based on the journal
1828                    capabilities: ORDERED_DATA if the journal can
1829                    cope, else JOURNAL_DATA */
1830                 if (journal_check_available_features
1831                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1832                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1833                 else
1834                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1835                 break;
1836
1837         case EXT3_MOUNT_ORDERED_DATA:
1838         case EXT3_MOUNT_WRITEBACK_DATA:
1839                 if (!journal_check_available_features
1840                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1841                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1842                                "requested data journaling mode\n");
1843                         goto failed_mount4;
1844                 }
1845         default:
1846                 break;
1847         }
1848
1849         if (test_opt(sb, NOBH)) {
1850                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1851                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1852                                 "its supported only with writeback mode\n");
1853                         clear_opt(sbi->s_mount_opt, NOBH);
1854                 }
1855         }
1856         /*
1857          * The journal_load will have done any necessary log recovery,
1858          * so we can safely mount the rest of the filesystem now.
1859          */
1860
1861         root = iget(sb, EXT3_ROOT_INO);
1862         sb->s_root = d_alloc_root(root);
1863         if (!sb->s_root) {
1864                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1865                 iput(root);
1866                 goto failed_mount4;
1867         }
1868         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1869                 dput(sb->s_root);
1870                 sb->s_root = NULL;
1871                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1872                 goto failed_mount4;
1873         }
1874
1875         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1876         /*
1877          * akpm: core read_super() calls in here with the superblock locked.
1878          * That deadlocks, because orphan cleanup needs to lock the superblock
1879          * in numerous places.  Here we just pop the lock - it's relatively
1880          * harmless, because we are now ready to accept write_super() requests,
1881          * and aviro says that's the only reason for hanging onto the
1882          * superblock lock.
1883          */
1884         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1885         ext3_orphan_cleanup(sb, es);
1886         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1887         if (needs_recovery)
1888                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1889         ext3_mark_recovery_complete(sb, es);
1890         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1891                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1892                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1893                 "writeback");
1894
1895         lock_kernel();
1896         return 0;
1897
1898 cantfind_ext3:
1899         if (!silent)
1900                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1901                        sb->s_id);
1902         goto failed_mount;
1903
1904 failed_mount4:
1905         journal_destroy(sbi->s_journal);
1906 failed_mount3:
1907         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1908         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1909         percpu_counter_destroy(&sbi->s_dirs_counter);
1910 failed_mount2:
1911         for (i = 0; i < db_count; i++)
1912                 brelse(sbi->s_group_desc[i]);
1913         kfree(sbi->s_group_desc);
1914 failed_mount:
1915 #ifdef CONFIG_QUOTA
1916         for (i = 0; i < MAXQUOTAS; i++)
1917                 kfree(sbi->s_qf_names[i]);
1918 #endif
1919         ext3_blkdev_remove(sbi);
1920         brelse(bh);
1921 out_fail:
1922         sb->s_fs_info = NULL;
1923         kfree(sbi);
1924         lock_kernel();
1925         return -EINVAL;
1926 }
1927
1928 /*
1929  * Setup any per-fs journal parameters now.  We'll do this both on
1930  * initial mount, once the journal has been initialised but before we've
1931  * done any recovery; and again on any subsequent remount.
1932  */
1933 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1934 {
1935         struct ext3_sb_info *sbi = EXT3_SB(sb);
1936
1937         if (sbi->s_commit_interval)
1938                 journal->j_commit_interval = sbi->s_commit_interval;
1939         /* We could also set up an ext3-specific default for the commit
1940          * interval here, but for now we'll just fall back to the jbd
1941          * default. */
1942
1943         spin_lock(&journal->j_state_lock);
1944         if (test_opt(sb, BARRIER))
1945                 journal->j_flags |= JFS_BARRIER;
1946         else
1947                 journal->j_flags &= ~JFS_BARRIER;
1948         spin_unlock(&journal->j_state_lock);
1949 }
1950
1951 static journal_t *ext3_get_journal(struct super_block *sb,
1952                                    unsigned int journal_inum)
1953 {
1954         struct inode *journal_inode;
1955         journal_t *journal;
1956
1957         /* First, test for the existence of a valid inode on disk.  Bad
1958          * things happen if we iget() an unused inode, as the subsequent
1959          * iput() will try to delete it. */
1960
1961         journal_inode = iget(sb, journal_inum);
1962         if (!journal_inode) {
1963                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1964                 return NULL;
1965         }
1966         if (!journal_inode->i_nlink) {
1967                 make_bad_inode(journal_inode);
1968                 iput(journal_inode);
1969                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1970                 return NULL;
1971         }
1972
1973         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1974                   journal_inode, journal_inode->i_size);
1975         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1976                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1977                 iput(journal_inode);
1978                 return NULL;
1979         }
1980
1981         journal = journal_init_inode(journal_inode);
1982         if (!journal) {
1983                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1984                 iput(journal_inode);
1985                 return NULL;
1986         }
1987         journal->j_private = sb;
1988         ext3_init_journal_params(sb, journal);
1989         return journal;
1990 }
1991
1992 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1993                                        dev_t j_dev)
1994 {
1995         struct buffer_head * bh;
1996         journal_t *journal;
1997         ext3_fsblk_t start;
1998         ext3_fsblk_t len;
1999         int hblock, blocksize;
2000         ext3_fsblk_t sb_block;
2001         unsigned long offset;
2002         struct ext3_super_block * es;
2003         struct block_device *bdev;
2004
2005         bdev = ext3_blkdev_get(j_dev);
2006         if (bdev == NULL)
2007                 return NULL;
2008
2009         if (bd_claim(bdev, sb)) {
2010                 printk(KERN_ERR
2011                         "EXT3: failed to claim external journal device.\n");
2012                 blkdev_put(bdev);
2013                 return NULL;
2014         }
2015
2016         blocksize = sb->s_blocksize;
2017         hblock = bdev_hardsect_size(bdev);
2018         if (blocksize < hblock) {
2019                 printk(KERN_ERR
2020                         "EXT3-fs: blocksize too small for journal device.\n");
2021                 goto out_bdev;
2022         }
2023
2024         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2025         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2026         set_blocksize(bdev, blocksize);
2027         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2028                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
2029                        "external journal\n");
2030                 goto out_bdev;
2031         }
2032
2033         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2034         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2035             !(le32_to_cpu(es->s_feature_incompat) &
2036               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2037                 printk(KERN_ERR "EXT3-fs: external journal has "
2038                                         "bad superblock\n");
2039                 brelse(bh);
2040                 goto out_bdev;
2041         }
2042
2043         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2044                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
2045                 brelse(bh);
2046                 goto out_bdev;
2047         }
2048
2049         len = le32_to_cpu(es->s_blocks_count);
2050         start = sb_block + 1;
2051         brelse(bh);     /* we're done with the superblock */
2052
2053         journal = journal_init_dev(bdev, sb->s_bdev,
2054                                         start, len, blocksize);
2055         if (!journal) {
2056                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
2057                 goto out_bdev;
2058         }
2059         journal->j_private = sb;
2060         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2061         wait_on_buffer(journal->j_sb_buffer);
2062         if (!buffer_uptodate(journal->j_sb_buffer)) {
2063                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
2064                 goto out_journal;
2065         }
2066         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2067                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
2068                                         "user (unsupported) - %d\n",
2069                         be32_to_cpu(journal->j_superblock->s_nr_users));
2070                 goto out_journal;
2071         }
2072         EXT3_SB(sb)->journal_bdev = bdev;
2073         ext3_init_journal_params(sb, journal);
2074         return journal;
2075 out_journal:
2076         journal_destroy(journal);
2077 out_bdev:
2078         ext3_blkdev_put(bdev);
2079         return NULL;
2080 }
2081
2082 static int ext3_load_journal(struct super_block *sb,
2083                              struct ext3_super_block *es,
2084                              unsigned long journal_devnum)
2085 {
2086         journal_t *journal;
2087         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2088         dev_t journal_dev;
2089         int err = 0;
2090         int really_read_only;
2091
2092         if (journal_devnum &&
2093             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2094                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
2095                         "numbers have changed\n");
2096                 journal_dev = new_decode_dev(journal_devnum);
2097         } else
2098                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2099
2100         really_read_only = bdev_read_only(sb->s_bdev);
2101
2102         /*
2103          * Are we loading a blank journal or performing recovery after a
2104          * crash?  For recovery, we need to check in advance whether we
2105          * can get read-write access to the device.
2106          */
2107
2108         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2109                 if (sb->s_flags & MS_RDONLY) {
2110                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
2111                                         "required on readonly filesystem.\n");
2112                         if (really_read_only) {
2113                                 printk(KERN_ERR "EXT3-fs: write access "
2114                                         "unavailable, cannot proceed.\n");
2115                                 return -EROFS;
2116                         }
2117                         printk (KERN_INFO "EXT3-fs: write access will "
2118                                         "be enabled during recovery.\n");
2119                 }
2120         }
2121
2122         if (journal_inum && journal_dev) {
2123                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
2124                        "and inode journals!\n");
2125                 return -EINVAL;
2126         }
2127
2128         if (journal_inum) {
2129                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2130                         return -EINVAL;
2131         } else {
2132                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2133                         return -EINVAL;
2134         }
2135
2136         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2137                 err = journal_update_format(journal);
2138                 if (err)  {
2139                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
2140                         journal_destroy(journal);
2141                         return err;
2142                 }
2143         }
2144
2145         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2146                 err = journal_wipe(journal, !really_read_only);
2147         if (!err)
2148                 err = journal_load(journal);
2149
2150         if (err) {
2151                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2152                 journal_destroy(journal);
2153                 return err;
2154         }
2155
2156         EXT3_SB(sb)->s_journal = journal;
2157         ext3_clear_journal_err(sb, es);
2158
2159         if (journal_devnum &&
2160             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2161                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2162                 sb->s_dirt = 1;
2163
2164                 /* Make sure we flush the recovery flag to disk. */
2165                 ext3_commit_super(sb, es, 1);
2166         }
2167
2168         return 0;
2169 }
2170
2171 static int ext3_create_journal(struct super_block * sb,
2172                                struct ext3_super_block * es,
2173                                unsigned int journal_inum)
2174 {
2175         journal_t *journal;
2176         int err;
2177
2178         if (sb->s_flags & MS_RDONLY) {
2179                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2180                                 "create journal.\n");
2181                 return -EROFS;
2182         }
2183
2184         journal = ext3_get_journal(sb, journal_inum);
2185         if (!journal)
2186                 return -EINVAL;
2187
2188         printk(KERN_INFO "EXT3-fs: creating new journal on inode %u\n",
2189                journal_inum);
2190
2191         err = journal_create(journal);
2192         if (err) {
2193                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2194                 journal_destroy(journal);
2195                 return -EIO;
2196         }
2197
2198         EXT3_SB(sb)->s_journal = journal;
2199
2200         ext3_update_dynamic_rev(sb);
2201         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2202         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2203
2204         es->s_journal_inum = cpu_to_le32(journal_inum);
2205         sb->s_dirt = 1;
2206
2207         /* Make sure we flush the recovery flag to disk. */
2208         ext3_commit_super(sb, es, 1);
2209
2210         return 0;
2211 }
2212
2213 static void ext3_commit_super (struct super_block * sb,
2214                                struct ext3_super_block * es,
2215                                int sync)
2216 {
2217         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2218
2219         if (!sbh)
2220                 return;
2221         es->s_wtime = cpu_to_le32(get_seconds());
2222         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2223         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2224         BUFFER_TRACE(sbh, "marking dirty");
2225         mark_buffer_dirty(sbh);
2226         if (sync)
2227                 sync_dirty_buffer(sbh);
2228 }
2229
2230
2231 /*
2232  * Have we just finished recovery?  If so, and if we are mounting (or
2233  * remounting) the filesystem readonly, then we will end up with a
2234  * consistent fs on disk.  Record that fact.
2235  */
2236 static void ext3_mark_recovery_complete(struct super_block * sb,
2237                                         struct ext3_super_block * es)
2238 {
2239         journal_t *journal = EXT3_SB(sb)->s_journal;
2240
2241         journal_lock_updates(journal);
2242         journal_flush(journal);
2243         lock_super(sb);
2244         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2245             sb->s_flags & MS_RDONLY) {
2246                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2247                 sb->s_dirt = 0;
2248                 ext3_commit_super(sb, es, 1);
2249         }
2250         unlock_super(sb);
2251         journal_unlock_updates(journal);
2252 }
2253
2254 /*
2255  * If we are mounting (or read-write remounting) a filesystem whose journal
2256  * has recorded an error from a previous lifetime, move that error to the
2257  * main filesystem now.
2258  */
2259 static void ext3_clear_journal_err(struct super_block * sb,
2260                                    struct ext3_super_block * es)
2261 {
2262         journal_t *journal;
2263         int j_errno;
2264         const char *errstr;
2265
2266         journal = EXT3_SB(sb)->s_journal;
2267
2268         /*
2269          * Now check for any error status which may have been recorded in the
2270          * journal by a prior ext3_error() or ext3_abort()
2271          */
2272
2273         j_errno = journal_errno(journal);
2274         if (j_errno) {
2275                 char nbuf[16];
2276
2277                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2278                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2279                              "from previous mount: %s", errstr);
2280                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2281                              "filesystem check.");
2282
2283                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2284                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2285                 ext3_commit_super (sb, es, 1);
2286
2287                 journal_clear_err(journal);
2288         }
2289 }
2290
2291 /*
2292  * Force the running and committing transactions to commit,
2293  * and wait on the commit.
2294  */
2295 int ext3_force_commit(struct super_block *sb)
2296 {
2297         journal_t *journal;
2298         int ret;
2299
2300         if (sb->s_flags & MS_RDONLY)
2301                 return 0;
2302
2303         journal = EXT3_SB(sb)->s_journal;
2304         sb->s_dirt = 0;
2305         ret = ext3_journal_force_commit(journal);
2306         return ret;
2307 }
2308
2309 /*
2310  * Ext3 always journals updates to the superblock itself, so we don't
2311  * have to propagate any other updates to the superblock on disk at this
2312  * point.  Just start an async writeback to get the buffers on their way
2313  * to the disk.
2314  *
2315  * This implicitly triggers the writebehind on sync().
2316  */
2317
2318 static void ext3_write_super (struct super_block * sb)
2319 {
2320         if (mutex_trylock(&sb->s_lock) != 0)
2321                 BUG();
2322         sb->s_dirt = 0;
2323 }
2324
2325 static int ext3_sync_fs(struct super_block *sb, int wait)
2326 {
2327         tid_t target;
2328
2329         sb->s_dirt = 0;
2330         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2331                 if (wait)
2332                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2333         }
2334         return 0;
2335 }
2336
2337 /*
2338  * LVM calls this function before a (read-only) snapshot is created.  This
2339  * gives us a chance to flush the journal completely and mark the fs clean.
2340  */
2341 static void ext3_write_super_lockfs(struct super_block *sb)
2342 {
2343         sb->s_dirt = 0;
2344
2345         if (!(sb->s_flags & MS_RDONLY)) {
2346                 journal_t *journal = EXT3_SB(sb)->s_journal;
2347
2348                 /* Now we set up the journal barrier. */
2349                 journal_lock_updates(journal);
2350                 journal_flush(journal);
2351
2352                 /* Journal blocked and flushed, clear needs_recovery flag. */
2353                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2354                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2355         }
2356 }
2357
2358 /*
2359  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2360  * flag here, even though the filesystem is not technically dirty yet.
2361  */
2362 static void ext3_unlockfs(struct super_block *sb)
2363 {
2364         if (!(sb->s_flags & MS_RDONLY)) {
2365                 lock_super(sb);
2366                 /* Reser the needs_recovery flag before the fs is unlocked. */
2367                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2368                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2369                 unlock_super(sb);
2370                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2371         }
2372 }
2373
2374 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2375 {
2376         struct ext3_super_block * es;
2377         struct ext3_sb_info *sbi = EXT3_SB(sb);
2378         ext3_fsblk_t n_blocks_count = 0;
2379         unsigned long old_sb_flags;
2380         struct ext3_mount_options old_opts;
2381         int err;
2382 #ifdef CONFIG_QUOTA
2383         int i;
2384 #endif
2385
2386         /* Store the original options */
2387         old_sb_flags = sb->s_flags;
2388         old_opts.s_mount_opt = sbi->s_mount_opt;
2389         old_opts.s_resuid = sbi->s_resuid;
2390         old_opts.s_resgid = sbi->s_resgid;
2391         old_opts.s_commit_interval = sbi->s_commit_interval;
2392 #ifdef CONFIG_QUOTA
2393         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2394         for (i = 0; i < MAXQUOTAS; i++)
2395                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2396 #endif
2397
2398         /*
2399          * Allow the "check" option to be passed as a remount option.
2400          */
2401         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2402                 err = -EINVAL;
2403                 goto restore_opts;
2404         }
2405
2406         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2407                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2408
2409         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2410                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2411
2412         es = sbi->s_es;
2413
2414         ext3_init_journal_params(sb, sbi->s_journal);
2415
2416         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2417                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2418                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2419                         err = -EROFS;
2420                         goto restore_opts;
2421                 }
2422
2423                 if (*flags & MS_RDONLY) {
2424                         /*
2425                          * First of all, the unconditional stuff we have to do
2426                          * to disable replay of the journal when we next remount
2427                          */
2428                         sb->s_flags |= MS_RDONLY;
2429
2430                         /*
2431                          * OK, test if we are remounting a valid rw partition
2432                          * readonly, and if so set the rdonly flag and then
2433                          * mark the partition as valid again.
2434                          */
2435                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2436                             (sbi->s_mount_state & EXT3_VALID_FS))
2437                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2438
2439                         /*
2440                          * We have to unlock super so that we can wait for
2441                          * transactions.
2442                          */
2443                         unlock_super(sb);
2444                         ext3_mark_recovery_complete(sb, es);
2445                         lock_super(sb);
2446                 } else {
2447                         __le32 ret;
2448                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2449                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2450                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2451                                        "remount RDWR because of unsupported "
2452                                        "optional features (%x).\n",
2453                                        sb->s_id, le32_to_cpu(ret));
2454                                 err = -EROFS;
2455                                 goto restore_opts;
2456                         }
2457
2458                         /*
2459                          * If we have an unprocessed orphan list hanging
2460                          * around from a previously readonly bdev mount,
2461                          * require a full umount/remount for now.
2462                          */
2463                         if (es->s_last_orphan) {
2464                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2465                                        "remount RDWR because of unprocessed "
2466                                        "orphan inode list.  Please "
2467                                        "umount/remount instead.\n",
2468                                        sb->s_id);
2469                                 err = -EINVAL;
2470                                 goto restore_opts;
2471                         }
2472
2473                         /*
2474                          * Mounting a RDONLY partition read-write, so reread
2475                          * and store the current valid flag.  (It may have
2476                          * been changed by e2fsck since we originally mounted
2477                          * the partition.)
2478                          */
2479                         ext3_clear_journal_err(sb, es);
2480                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2481                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2482                                 goto restore_opts;
2483                         if (!ext3_setup_super (sb, es, 0))
2484                                 sb->s_flags &= ~MS_RDONLY;
2485                 }
2486         }
2487 #ifdef CONFIG_QUOTA
2488         /* Release old quota file names */
2489         for (i = 0; i < MAXQUOTAS; i++)
2490                 if (old_opts.s_qf_names[i] &&
2491                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2492                         kfree(old_opts.s_qf_names[i]);
2493 #endif
2494         return 0;
2495 restore_opts:
2496         sb->s_flags = old_sb_flags;
2497         sbi->s_mount_opt = old_opts.s_mount_opt;
2498         sbi->s_resuid = old_opts.s_resuid;
2499         sbi->s_resgid = old_opts.s_resgid;
2500         sbi->s_commit_interval = old_opts.s_commit_interval;
2501 #ifdef CONFIG_QUOTA
2502         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2503         for (i = 0; i < MAXQUOTAS; i++) {
2504                 if (sbi->s_qf_names[i] &&
2505                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2506                         kfree(sbi->s_qf_names[i]);
2507                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2508         }
2509 #endif
2510         return err;
2511 }
2512
2513 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2514 {
2515         struct super_block *sb = dentry->d_sb;
2516         struct ext3_sb_info *sbi = EXT3_SB(sb);
2517         struct ext3_super_block *es = sbi->s_es;
2518         u64 fsid;
2519
2520         if (test_opt(sb, MINIX_DF)) {
2521                 sbi->s_overhead_last = 0;
2522         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2523                 unsigned long ngroups = sbi->s_groups_count, i;
2524                 ext3_fsblk_t overhead = 0;
2525                 smp_rmb();
2526
2527                 /*
2528                  * Compute the overhead (FS structures).  This is constant
2529                  * for a given filesystem unless the number of block groups
2530                  * changes so we cache the previous value until it does.
2531                  */
2532
2533                 /*
2534                  * All of the blocks before first_data_block are
2535                  * overhead
2536                  */
2537                 overhead = le32_to_cpu(es->s_first_data_block);
2538
2539                 /*
2540                  * Add the overhead attributed to the superblock and
2541                  * block group descriptors.  If the sparse superblocks
2542                  * feature is turned on, then not all groups have this.
2543                  */
2544                 for (i = 0; i < ngroups; i++) {
2545                         overhead += ext3_bg_has_super(sb, i) +
2546                                 ext3_bg_num_gdb(sb, i);
2547                         cond_resched();
2548                 }
2549
2550                 /*
2551                  * Every block group has an inode bitmap, a block
2552                  * bitmap, and an inode table.
2553                  */
2554                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2555                 sbi->s_overhead_last = overhead;
2556                 smp_wmb();
2557                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2558         }
2559
2560         buf->f_type = EXT3_SUPER_MAGIC;
2561         buf->f_bsize = sb->s_blocksize;
2562         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2563         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2564         es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
2565         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2566         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2567                 buf->f_bavail = 0;
2568         buf->f_files = le32_to_cpu(es->s_inodes_count);
2569         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2570         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2571         buf->f_namelen = EXT3_NAME_LEN;
2572         fsid = le64_to_cpup((void *)es->s_uuid) ^
2573                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2574         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2575         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2576         return 0;
2577 }
2578
2579 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2580  * is locked for write. Otherwise the are possible deadlocks:
2581  * Process 1                         Process 2
2582  * ext3_create()                     quota_sync()
2583  *   journal_start()                   write_dquot()
2584  *   DQUOT_INIT()                        down(dqio_mutex)
2585  *     down(dqio_mutex)                    journal_start()
2586  *
2587  */
2588
2589 #ifdef CONFIG_QUOTA
2590
2591 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2592 {
2593         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2594 }
2595
2596 static int ext3_dquot_initialize(struct inode *inode, int type)
2597 {
2598         handle_t *handle;
2599         int ret, err;
2600
2601         /* We may create quota structure so we need to reserve enough blocks */
2602         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2603         if (IS_ERR(handle))
2604                 return PTR_ERR(handle);
2605         ret = dquot_initialize(inode, type);
2606         err = ext3_journal_stop(handle);
2607         if (!ret)
2608                 ret = err;
2609         return ret;
2610 }
2611
2612 static int ext3_dquot_drop(struct inode *inode)
2613 {
2614         handle_t *handle;
2615         int ret, err;
2616
2617         /* We may delete quota structure so we need to reserve enough blocks */
2618         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2619         if (IS_ERR(handle))
2620                 return PTR_ERR(handle);
2621         ret = dquot_drop(inode);
2622         err = ext3_journal_stop(handle);
2623         if (!ret)
2624                 ret = err;
2625         return ret;
2626 }
2627
2628 static int ext3_write_dquot(struct dquot *dquot)
2629 {
2630         int ret, err;
2631         handle_t *handle;
2632         struct inode *inode;
2633
2634         inode = dquot_to_inode(dquot);
2635         handle = ext3_journal_start(inode,
2636                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2637         if (IS_ERR(handle))
2638                 return PTR_ERR(handle);
2639         ret = dquot_commit(dquot);
2640         err = ext3_journal_stop(handle);
2641         if (!ret)
2642                 ret = err;
2643         return ret;
2644 }
2645
2646 static int ext3_acquire_dquot(struct dquot *dquot)
2647 {
2648         int ret, err;
2649         handle_t *handle;
2650
2651         handle = ext3_journal_start(dquot_to_inode(dquot),
2652                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2653         if (IS_ERR(handle))
2654                 return PTR_ERR(handle);
2655         ret = dquot_acquire(dquot);
2656         err = ext3_journal_stop(handle);
2657         if (!ret)
2658                 ret = err;
2659         return ret;
2660 }
2661
2662 static int ext3_release_dquot(struct dquot *dquot)
2663 {
2664         int ret, err;
2665         handle_t *handle;
2666
2667         handle = ext3_journal_start(dquot_to_inode(dquot),
2668                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2669         if (IS_ERR(handle)) {
2670                 /* Release dquot anyway to avoid endless cycle in dqput() */
2671                 dquot_release(dquot);
2672                 return PTR_ERR(handle);
2673         }
2674         ret = dquot_release(dquot);
2675         err = ext3_journal_stop(handle);
2676         if (!ret)
2677                 ret = err;
2678         return ret;
2679 }
2680
2681 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2682 {
2683         /* Are we journalling quotas? */
2684         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2685             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2686                 dquot_mark_dquot_dirty(dquot);
2687                 return ext3_write_dquot(dquot);
2688         } else {
2689                 return dquot_mark_dquot_dirty(dquot);
2690         }
2691 }
2692
2693 static int ext3_write_info(struct super_block *sb, int type)
2694 {
2695         int ret, err;
2696         handle_t *handle;
2697
2698         /* Data block + inode block */
2699         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2700         if (IS_ERR(handle))
2701                 return PTR_ERR(handle);
2702         ret = dquot_commit_info(sb, type);
2703         err = ext3_journal_stop(handle);
2704         if (!ret)
2705                 ret = err;
2706         return ret;
2707 }
2708
2709 /*
2710  * Turn on quotas during mount time - we need to find
2711  * the quota file and such...
2712  */
2713 static int ext3_quota_on_mount(struct super_block *sb, int type)
2714 {
2715         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2716                         EXT3_SB(sb)->s_jquota_fmt, type);
2717 }
2718
2719 /*
2720  * Standard function to be called on quota_on
2721  */
2722 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2723                          char *path)
2724 {
2725         int err;
2726         struct nameidata nd;
2727
2728         if (!test_opt(sb, QUOTA))
2729                 return -EINVAL;
2730         /* Not journalling quota? */
2731         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2732             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2733                 return vfs_quota_on(sb, type, format_id, path);
2734         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2735         if (err)
2736                 return err;
2737         /* Quotafile not on the same filesystem? */
2738         if (nd.mnt->mnt_sb != sb) {
2739                 path_release(&nd);
2740                 return -EXDEV;
2741         }
2742         /* Quotafile not of fs root? */
2743         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2744                 printk(KERN_WARNING
2745                         "EXT3-fs: Quota file not on filesystem root. "
2746                         "Journalled quota will not work.\n");
2747         path_release(&nd);
2748         return vfs_quota_on(sb, type, format_id, path);
2749 }
2750
2751 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2752  * acquiring the locks... As quota files are never truncated and quota code
2753  * itself serializes the operations (and noone else should touch the files)
2754  * we don't have to be afraid of races */
2755 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2756                                size_t len, loff_t off)
2757 {
2758         struct inode *inode = sb_dqopt(sb)->files[type];
2759         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2760         int err = 0;
2761         int offset = off & (sb->s_blocksize - 1);
2762         int tocopy;
2763         size_t toread;
2764         struct buffer_head *bh;
2765         loff_t i_size = i_size_read(inode);
2766
2767         if (off > i_size)
2768                 return 0;
2769         if (off+len > i_size)
2770                 len = i_size-off;
2771         toread = len;
2772         while (toread > 0) {
2773                 tocopy = sb->s_blocksize - offset < toread ?
2774                                 sb->s_blocksize - offset : toread;
2775                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2776                 if (err)
2777                         return err;
2778                 if (!bh)        /* A hole? */
2779                         memset(data, 0, tocopy);
2780                 else
2781                         memcpy(data, bh->b_data+offset, tocopy);
2782                 brelse(bh);
2783                 offset = 0;
2784                 toread -= tocopy;
2785                 data += tocopy;
2786                 blk++;
2787         }
2788         return len;
2789 }
2790
2791 /* Write to quotafile (we know the transaction is already started and has
2792  * enough credits) */
2793 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2794                                 const char *data, size_t len, loff_t off)
2795 {
2796         struct inode *inode = sb_dqopt(sb)->files[type];
2797         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2798         int err = 0;
2799         int offset = off & (sb->s_blocksize - 1);
2800         int tocopy;
2801         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2802         size_t towrite = len;
2803         struct buffer_head *bh;
2804         handle_t *handle = journal_current_handle();
2805
2806         if (!handle) {
2807                 printk(KERN_WARNING "EXT3-fs: Quota write (off=%Lu, len=%Lu)"
2808                         " cancelled because transaction is not started.\n",
2809                         (unsigned long long)off, (unsigned long long)len);
2810                 return -EIO;
2811         }
2812         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2813         while (towrite > 0) {
2814                 tocopy = sb->s_blocksize - offset < towrite ?
2815                                 sb->s_blocksize - offset : towrite;
2816                 bh = ext3_bread(handle, inode, blk, 1, &err);
2817                 if (!bh)
2818                         goto out;
2819                 if (journal_quota) {
2820                         err = ext3_journal_get_write_access(handle, bh);
2821                         if (err) {
2822                                 brelse(bh);
2823                                 goto out;
2824                         }
2825                 }
2826                 lock_buffer(bh);
2827                 memcpy(bh->b_data+offset, data, tocopy);
2828                 flush_dcache_page(bh->b_page);
2829                 unlock_buffer(bh);
2830                 if (journal_quota)
2831                         err = ext3_journal_dirty_metadata(handle, bh);
2832                 else {
2833                         /* Always do at least ordered writes for quotas */
2834                         err = ext3_journal_dirty_data(handle, bh);
2835                         mark_buffer_dirty(bh);
2836                 }
2837                 brelse(bh);
2838                 if (err)
2839                         goto out;
2840                 offset = 0;
2841                 towrite -= tocopy;
2842                 data += tocopy;
2843                 blk++;
2844         }
2845 out:
2846         if (len == towrite)
2847                 return err;
2848         if (inode->i_size < off+len-towrite) {
2849                 i_size_write(inode, off+len-towrite);
2850                 EXT3_I(inode)->i_disksize = inode->i_size;
2851         }
2852         inode->i_version++;
2853         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2854         ext3_mark_inode_dirty(handle, inode);
2855         mutex_unlock(&inode->i_mutex);
2856         return len - towrite;
2857 }
2858
2859 #endif
2860
2861 static int ext3_get_sb(struct file_system_type *fs_type,
2862         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2863 {
2864         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
2865 }
2866
2867 static struct file_system_type ext3_fs_type = {
2868         .owner          = THIS_MODULE,
2869         .name           = "ext3",
2870         .get_sb         = ext3_get_sb,
2871         .kill_sb        = kill_block_super,
2872         .fs_flags       = FS_REQUIRES_DEV,
2873 };
2874
2875 static int __init init_ext3_fs(void)
2876 {
2877         int err = init_ext3_xattr();
2878         if (err)
2879                 return err;
2880         err = init_inodecache();
2881         if (err)
2882                 goto out1;
2883         err = register_filesystem(&ext3_fs_type);
2884         if (err)
2885                 goto out;
2886         return 0;
2887 out:
2888         destroy_inodecache();
2889 out1:
2890         exit_ext3_xattr();
2891         return err;
2892 }
2893
2894 static void __exit exit_ext3_fs(void)
2895 {
2896         unregister_filesystem(&ext3_fs_type);
2897         destroy_inodecache();
2898         exit_ext3_xattr();
2899 }
2900
2901 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2902 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2903 MODULE_LICENSE("GPL");
2904 module_init(init_ext3_fs)
2905 module_exit(exit_ext3_fs)