0b5057e0dc1e2c72df39807c10857e86142700da
[linux-3.10.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_RO)) {
579                 int def_errors = le16_to_cpu(es->s_errors);
580
581                 if (def_errors == EXT3_ERRORS_PANIC ||
582                     def_errors == EXT3_ERRORS_CONTINUE) {
583                         seq_puts(seq, ",errors=remount-ro");
584                 }
585         }
586         if (test_opt(sb, ERRORS_CONT))
587                 seq_puts(seq, ",errors=continue");
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 = ext3_iget(sb, ino);
653         if (IS_ERR(inode))
654                 return ERR_CAST(inode);
655         if (generation && inode->i_generation != generation) {
656                 iput(inode);
657                 return ERR_PTR(-ESTALE);
658         }
659
660         return inode;
661 }
662
663 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
664                 int fh_len, int fh_type)
665 {
666         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
667                                     ext3_nfs_get_inode);
668 }
669
670 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
671                 int fh_len, int fh_type)
672 {
673         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
674                                     ext3_nfs_get_inode);
675 }
676
677 #ifdef CONFIG_QUOTA
678 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
679 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
680
681 static int ext3_dquot_initialize(struct inode *inode, int type);
682 static int ext3_dquot_drop(struct inode *inode);
683 static int ext3_write_dquot(struct dquot *dquot);
684 static int ext3_acquire_dquot(struct dquot *dquot);
685 static int ext3_release_dquot(struct dquot *dquot);
686 static int ext3_mark_dquot_dirty(struct dquot *dquot);
687 static int ext3_write_info(struct super_block *sb, int type);
688 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
689 static int ext3_quota_on_mount(struct super_block *sb, int type);
690 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
691                                size_t len, loff_t off);
692 static ssize_t ext3_quota_write(struct super_block *sb, int type,
693                                 const char *data, size_t len, loff_t off);
694
695 static struct dquot_operations ext3_quota_operations = {
696         .initialize     = ext3_dquot_initialize,
697         .drop           = ext3_dquot_drop,
698         .alloc_space    = dquot_alloc_space,
699         .alloc_inode    = dquot_alloc_inode,
700         .free_space     = dquot_free_space,
701         .free_inode     = dquot_free_inode,
702         .transfer       = dquot_transfer,
703         .write_dquot    = ext3_write_dquot,
704         .acquire_dquot  = ext3_acquire_dquot,
705         .release_dquot  = ext3_release_dquot,
706         .mark_dirty     = ext3_mark_dquot_dirty,
707         .write_info     = ext3_write_info
708 };
709
710 static struct quotactl_ops ext3_qctl_operations = {
711         .quota_on       = ext3_quota_on,
712         .quota_off      = vfs_quota_off,
713         .quota_sync     = vfs_quota_sync,
714         .get_info       = vfs_get_dqinfo,
715         .set_info       = vfs_set_dqinfo,
716         .get_dqblk      = vfs_get_dqblk,
717         .set_dqblk      = vfs_set_dqblk
718 };
719 #endif
720
721 static const struct super_operations ext3_sops = {
722         .alloc_inode    = ext3_alloc_inode,
723         .destroy_inode  = ext3_destroy_inode,
724         .write_inode    = ext3_write_inode,
725         .dirty_inode    = ext3_dirty_inode,
726         .delete_inode   = ext3_delete_inode,
727         .put_super      = ext3_put_super,
728         .write_super    = ext3_write_super,
729         .sync_fs        = ext3_sync_fs,
730         .write_super_lockfs = ext3_write_super_lockfs,
731         .unlockfs       = ext3_unlockfs,
732         .statfs         = ext3_statfs,
733         .remount_fs     = ext3_remount,
734         .clear_inode    = ext3_clear_inode,
735         .show_options   = ext3_show_options,
736 #ifdef CONFIG_QUOTA
737         .quota_read     = ext3_quota_read,
738         .quota_write    = ext3_quota_write,
739 #endif
740 };
741
742 static const struct export_operations ext3_export_ops = {
743         .fh_to_dentry = ext3_fh_to_dentry,
744         .fh_to_parent = ext3_fh_to_parent,
745         .get_parent = ext3_get_parent,
746 };
747
748 enum {
749         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
750         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
751         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
752         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
753         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
754         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
755         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
756         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
757         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
758         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
759         Opt_grpquota
760 };
761
762 static match_table_t tokens = {
763         {Opt_bsd_df, "bsddf"},
764         {Opt_minix_df, "minixdf"},
765         {Opt_grpid, "grpid"},
766         {Opt_grpid, "bsdgroups"},
767         {Opt_nogrpid, "nogrpid"},
768         {Opt_nogrpid, "sysvgroups"},
769         {Opt_resgid, "resgid=%u"},
770         {Opt_resuid, "resuid=%u"},
771         {Opt_sb, "sb=%u"},
772         {Opt_err_cont, "errors=continue"},
773         {Opt_err_panic, "errors=panic"},
774         {Opt_err_ro, "errors=remount-ro"},
775         {Opt_nouid32, "nouid32"},
776         {Opt_nocheck, "nocheck"},
777         {Opt_nocheck, "check=none"},
778         {Opt_debug, "debug"},
779         {Opt_oldalloc, "oldalloc"},
780         {Opt_orlov, "orlov"},
781         {Opt_user_xattr, "user_xattr"},
782         {Opt_nouser_xattr, "nouser_xattr"},
783         {Opt_acl, "acl"},
784         {Opt_noacl, "noacl"},
785         {Opt_reservation, "reservation"},
786         {Opt_noreservation, "noreservation"},
787         {Opt_noload, "noload"},
788         {Opt_nobh, "nobh"},
789         {Opt_bh, "bh"},
790         {Opt_commit, "commit=%u"},
791         {Opt_journal_update, "journal=update"},
792         {Opt_journal_inum, "journal=%u"},
793         {Opt_journal_dev, "journal_dev=%u"},
794         {Opt_abort, "abort"},
795         {Opt_data_journal, "data=journal"},
796         {Opt_data_ordered, "data=ordered"},
797         {Opt_data_writeback, "data=writeback"},
798         {Opt_offusrjquota, "usrjquota="},
799         {Opt_usrjquota, "usrjquota=%s"},
800         {Opt_offgrpjquota, "grpjquota="},
801         {Opt_grpjquota, "grpjquota=%s"},
802         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
803         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
804         {Opt_grpquota, "grpquota"},
805         {Opt_noquota, "noquota"},
806         {Opt_quota, "quota"},
807         {Opt_usrquota, "usrquota"},
808         {Opt_barrier, "barrier=%u"},
809         {Opt_err, NULL},
810         {Opt_resize, "resize"},
811 };
812
813 static ext3_fsblk_t get_sb_block(void **data)
814 {
815         ext3_fsblk_t    sb_block;
816         char            *options = (char *) *data;
817
818         if (!options || strncmp(options, "sb=", 3) != 0)
819                 return 1;       /* Default location */
820         options += 3;
821         /*todo: use simple_strtoll with >32bit ext3 */
822         sb_block = simple_strtoul(options, &options, 0);
823         if (*options && *options != ',') {
824                 printk("EXT3-fs: Invalid sb specification: %s\n",
825                        (char *) *data);
826                 return 1;
827         }
828         if (*options == ',')
829                 options++;
830         *data = (void *) options;
831         return sb_block;
832 }
833
834 static int parse_options (char *options, struct super_block *sb,
835                           unsigned int *inum, unsigned long *journal_devnum,
836                           ext3_fsblk_t *n_blocks_count, int is_remount)
837 {
838         struct ext3_sb_info *sbi = EXT3_SB(sb);
839         char * p;
840         substring_t args[MAX_OPT_ARGS];
841         int data_opt = 0;
842         int option;
843 #ifdef CONFIG_QUOTA
844         int qtype;
845         char *qname;
846 #endif
847
848         if (!options)
849                 return 1;
850
851         while ((p = strsep (&options, ",")) != NULL) {
852                 int token;
853                 if (!*p)
854                         continue;
855
856                 token = match_token(p, tokens, args);
857                 switch (token) {
858                 case Opt_bsd_df:
859                         clear_opt (sbi->s_mount_opt, MINIX_DF);
860                         break;
861                 case Opt_minix_df:
862                         set_opt (sbi->s_mount_opt, MINIX_DF);
863                         break;
864                 case Opt_grpid:
865                         set_opt (sbi->s_mount_opt, GRPID);
866                         break;
867                 case Opt_nogrpid:
868                         clear_opt (sbi->s_mount_opt, GRPID);
869                         break;
870                 case Opt_resuid:
871                         if (match_int(&args[0], &option))
872                                 return 0;
873                         sbi->s_resuid = option;
874                         break;
875                 case Opt_resgid:
876                         if (match_int(&args[0], &option))
877                                 return 0;
878                         sbi->s_resgid = option;
879                         break;
880                 case Opt_sb:
881                         /* handled by get_sb_block() instead of here */
882                         /* *sb_block = match_int(&args[0]); */
883                         break;
884                 case Opt_err_panic:
885                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
886                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
887                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
888                         break;
889                 case Opt_err_ro:
890                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
891                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
892                         set_opt (sbi->s_mount_opt, ERRORS_RO);
893                         break;
894                 case Opt_err_cont:
895                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
896                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
897                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
898                         break;
899                 case Opt_nouid32:
900                         set_opt (sbi->s_mount_opt, NO_UID32);
901                         break;
902                 case Opt_nocheck:
903                         clear_opt (sbi->s_mount_opt, CHECK);
904                         break;
905                 case Opt_debug:
906                         set_opt (sbi->s_mount_opt, DEBUG);
907                         break;
908                 case Opt_oldalloc:
909                         set_opt (sbi->s_mount_opt, OLDALLOC);
910                         break;
911                 case Opt_orlov:
912                         clear_opt (sbi->s_mount_opt, OLDALLOC);
913                         break;
914 #ifdef CONFIG_EXT3_FS_XATTR
915                 case Opt_user_xattr:
916                         set_opt (sbi->s_mount_opt, XATTR_USER);
917                         break;
918                 case Opt_nouser_xattr:
919                         clear_opt (sbi->s_mount_opt, XATTR_USER);
920                         break;
921 #else
922                 case Opt_user_xattr:
923                 case Opt_nouser_xattr:
924                         printk("EXT3 (no)user_xattr options not supported\n");
925                         break;
926 #endif
927 #ifdef CONFIG_EXT3_FS_POSIX_ACL
928                 case Opt_acl:
929                         set_opt(sbi->s_mount_opt, POSIX_ACL);
930                         break;
931                 case Opt_noacl:
932                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
933                         break;
934 #else
935                 case Opt_acl:
936                 case Opt_noacl:
937                         printk("EXT3 (no)acl options not supported\n");
938                         break;
939 #endif
940                 case Opt_reservation:
941                         set_opt(sbi->s_mount_opt, RESERVATION);
942                         break;
943                 case Opt_noreservation:
944                         clear_opt(sbi->s_mount_opt, RESERVATION);
945                         break;
946                 case Opt_journal_update:
947                         /* @@@ FIXME */
948                         /* Eventually we will want to be able to create
949                            a journal file here.  For now, only allow the
950                            user to specify an existing inode to be the
951                            journal file. */
952                         if (is_remount) {
953                                 printk(KERN_ERR "EXT3-fs: cannot specify "
954                                        "journal on remount\n");
955                                 return 0;
956                         }
957                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
958                         break;
959                 case Opt_journal_inum:
960                         if (is_remount) {
961                                 printk(KERN_ERR "EXT3-fs: cannot specify "
962                                        "journal on remount\n");
963                                 return 0;
964                         }
965                         if (match_int(&args[0], &option))
966                                 return 0;
967                         *inum = option;
968                         break;
969                 case Opt_journal_dev:
970                         if (is_remount) {
971                                 printk(KERN_ERR "EXT3-fs: cannot specify "
972                                        "journal on remount\n");
973                                 return 0;
974                         }
975                         if (match_int(&args[0], &option))
976                                 return 0;
977                         *journal_devnum = option;
978                         break;
979                 case Opt_noload:
980                         set_opt (sbi->s_mount_opt, NOLOAD);
981                         break;
982                 case Opt_commit:
983                         if (match_int(&args[0], &option))
984                                 return 0;
985                         if (option < 0)
986                                 return 0;
987                         if (option == 0)
988                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
989                         sbi->s_commit_interval = HZ * option;
990                         break;
991                 case Opt_data_journal:
992                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
993                         goto datacheck;
994                 case Opt_data_ordered:
995                         data_opt = EXT3_MOUNT_ORDERED_DATA;
996                         goto datacheck;
997                 case Opt_data_writeback:
998                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
999                 datacheck:
1000                         if (is_remount) {
1001                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
1002                                                 != data_opt) {
1003                                         printk(KERN_ERR
1004                                                 "EXT3-fs: cannot change data "
1005                                                 "mode on remount\n");
1006                                         return 0;
1007                                 }
1008                         } else {
1009                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
1010                                 sbi->s_mount_opt |= data_opt;
1011                         }
1012                         break;
1013 #ifdef CONFIG_QUOTA
1014                 case Opt_usrjquota:
1015                         qtype = USRQUOTA;
1016                         goto set_qf_name;
1017                 case Opt_grpjquota:
1018                         qtype = GRPQUOTA;
1019 set_qf_name:
1020                         if (sb_any_quota_enabled(sb)) {
1021                                 printk(KERN_ERR
1022                                         "EXT3-fs: Cannot change journalled "
1023                                         "quota options when quota turned on.\n");
1024                                 return 0;
1025                         }
1026                         qname = match_strdup(&args[0]);
1027                         if (!qname) {
1028                                 printk(KERN_ERR
1029                                         "EXT3-fs: not enough memory for "
1030                                         "storing quotafile name.\n");
1031                                 return 0;
1032                         }
1033                         if (sbi->s_qf_names[qtype] &&
1034                             strcmp(sbi->s_qf_names[qtype], qname)) {
1035                                 printk(KERN_ERR
1036                                         "EXT3-fs: %s quota file already "
1037                                         "specified.\n", QTYPE2NAME(qtype));
1038                                 kfree(qname);
1039                                 return 0;
1040                         }
1041                         sbi->s_qf_names[qtype] = qname;
1042                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1043                                 printk(KERN_ERR
1044                                         "EXT3-fs: quotafile must be on "
1045                                         "filesystem root.\n");
1046                                 kfree(sbi->s_qf_names[qtype]);
1047                                 sbi->s_qf_names[qtype] = NULL;
1048                                 return 0;
1049                         }
1050                         set_opt(sbi->s_mount_opt, QUOTA);
1051                         break;
1052                 case Opt_offusrjquota:
1053                         qtype = USRQUOTA;
1054                         goto clear_qf_name;
1055                 case Opt_offgrpjquota:
1056                         qtype = GRPQUOTA;
1057 clear_qf_name:
1058                         if (sb_any_quota_enabled(sb)) {
1059                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1060                                         "journalled quota options when "
1061                                         "quota turned on.\n");
1062                                 return 0;
1063                         }
1064                         /*
1065                          * The space will be released later when all options
1066                          * are confirmed to be correct
1067                          */
1068                         sbi->s_qf_names[qtype] = NULL;
1069                         break;
1070                 case Opt_jqfmt_vfsold:
1071                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
1072                         break;
1073                 case Opt_jqfmt_vfsv0:
1074                         sbi->s_jquota_fmt = QFMT_VFS_V0;
1075                         break;
1076                 case Opt_quota:
1077                 case Opt_usrquota:
1078                         set_opt(sbi->s_mount_opt, QUOTA);
1079                         set_opt(sbi->s_mount_opt, USRQUOTA);
1080                         break;
1081                 case Opt_grpquota:
1082                         set_opt(sbi->s_mount_opt, QUOTA);
1083                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1084                         break;
1085                 case Opt_noquota:
1086                         if (sb_any_quota_enabled(sb)) {
1087                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
1088                                         "options when quota turned on.\n");
1089                                 return 0;
1090                         }
1091                         clear_opt(sbi->s_mount_opt, QUOTA);
1092                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1093                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1094                         break;
1095 #else
1096                 case Opt_quota:
1097                 case Opt_usrquota:
1098                 case Opt_grpquota:
1099                 case Opt_usrjquota:
1100                 case Opt_grpjquota:
1101                 case Opt_offusrjquota:
1102                 case Opt_offgrpjquota:
1103                 case Opt_jqfmt_vfsold:
1104                 case Opt_jqfmt_vfsv0:
1105                         printk(KERN_ERR
1106                                 "EXT3-fs: journalled quota options not "
1107                                 "supported.\n");
1108                         break;
1109                 case Opt_noquota:
1110                         break;
1111 #endif
1112                 case Opt_abort:
1113                         set_opt(sbi->s_mount_opt, ABORT);
1114                         break;
1115                 case Opt_barrier:
1116                         if (match_int(&args[0], &option))
1117                                 return 0;
1118                         if (option)
1119                                 set_opt(sbi->s_mount_opt, BARRIER);
1120                         else
1121                                 clear_opt(sbi->s_mount_opt, BARRIER);
1122                         break;
1123                 case Opt_ignore:
1124                         break;
1125                 case Opt_resize:
1126                         if (!is_remount) {
1127                                 printk("EXT3-fs: resize option only available "
1128                                         "for remount\n");
1129                                 return 0;
1130                         }
1131                         if (match_int(&args[0], &option) != 0)
1132                                 return 0;
1133                         *n_blocks_count = option;
1134                         break;
1135                 case Opt_nobh:
1136                         set_opt(sbi->s_mount_opt, NOBH);
1137                         break;
1138                 case Opt_bh:
1139                         clear_opt(sbi->s_mount_opt, NOBH);
1140                         break;
1141                 default:
1142                         printk (KERN_ERR
1143                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1144                                 "or missing value\n", p);
1145                         return 0;
1146                 }
1147         }
1148 #ifdef CONFIG_QUOTA
1149         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1150                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1151                      sbi->s_qf_names[USRQUOTA])
1152                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1153
1154                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1155                      sbi->s_qf_names[GRPQUOTA])
1156                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1157
1158                 if ((sbi->s_qf_names[USRQUOTA] &&
1159                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1160                     (sbi->s_qf_names[GRPQUOTA] &&
1161                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1162                         printk(KERN_ERR "EXT3-fs: old and new quota "
1163                                         "format mixing.\n");
1164                         return 0;
1165                 }
1166
1167                 if (!sbi->s_jquota_fmt) {
1168                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1169                                         "not specified.\n");
1170                         return 0;
1171                 }
1172         } else {
1173                 if (sbi->s_jquota_fmt) {
1174                         printk(KERN_ERR "EXT3-fs: journalled quota format "
1175                                         "specified with no journalling "
1176                                         "enabled.\n");
1177                         return 0;
1178                 }
1179         }
1180 #endif
1181         return 1;
1182 }
1183
1184 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1185                             int read_only)
1186 {
1187         struct ext3_sb_info *sbi = EXT3_SB(sb);
1188         int res = 0;
1189
1190         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1191                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1192                         "forcing read-only mode\n");
1193                 res = MS_RDONLY;
1194         }
1195         if (read_only)
1196                 return res;
1197         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1198                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1199                         "running e2fsck is recommended\n");
1200         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1201                 printk (KERN_WARNING
1202                         "EXT3-fs warning: mounting fs with errors, "
1203                         "running e2fsck is recommended\n");
1204         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1205                  le16_to_cpu(es->s_mnt_count) >=
1206                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1207                 printk (KERN_WARNING
1208                         "EXT3-fs warning: maximal mount count reached, "
1209                         "running e2fsck is recommended\n");
1210         else if (le32_to_cpu(es->s_checkinterval) &&
1211                 (le32_to_cpu(es->s_lastcheck) +
1212                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1213                 printk (KERN_WARNING
1214                         "EXT3-fs warning: checktime reached, "
1215                         "running e2fsck is recommended\n");
1216 #if 0
1217                 /* @@@ We _will_ want to clear the valid bit if we find
1218                    inconsistencies, to force a fsck at reboot.  But for
1219                    a plain journaled filesystem we can keep it set as
1220                    valid forever! :) */
1221         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
1222 #endif
1223         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1224                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1225         le16_add_cpu(&es->s_mnt_count, 1);
1226         es->s_mtime = cpu_to_le32(get_seconds());
1227         ext3_update_dynamic_rev(sb);
1228         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1229
1230         ext3_commit_super(sb, es, 1);
1231         if (test_opt(sb, DEBUG))
1232                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1233                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1234                         sb->s_blocksize,
1235                         sbi->s_groups_count,
1236                         EXT3_BLOCKS_PER_GROUP(sb),
1237                         EXT3_INODES_PER_GROUP(sb),
1238                         sbi->s_mount_opt);
1239
1240         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1241         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1242                 char b[BDEVNAME_SIZE];
1243
1244                 printk("external journal on %s\n",
1245                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1246         } else {
1247                 printk("internal journal\n");
1248         }
1249         return res;
1250 }
1251
1252 /* Called at mount-time, super-block is locked */
1253 static int ext3_check_descriptors(struct super_block *sb)
1254 {
1255         struct ext3_sb_info *sbi = EXT3_SB(sb);
1256         ext3_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1257         ext3_fsblk_t last_block;
1258         int i;
1259
1260         ext3_debug ("Checking group descriptors");
1261
1262         for (i = 0; i < sbi->s_groups_count; i++) {
1263                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1264
1265                 if (i == sbi->s_groups_count - 1)
1266                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1267                 else
1268                         last_block = first_block +
1269                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1270
1271                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1272                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1273                 {
1274                         ext3_error (sb, "ext3_check_descriptors",
1275                                     "Block bitmap for group %d"
1276                                     " not in group (block %lu)!",
1277                                     i, (unsigned long)
1278                                         le32_to_cpu(gdp->bg_block_bitmap));
1279                         return 0;
1280                 }
1281                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1282                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1283                 {
1284                         ext3_error (sb, "ext3_check_descriptors",
1285                                     "Inode bitmap for group %d"
1286                                     " not in group (block %lu)!",
1287                                     i, (unsigned long)
1288                                         le32_to_cpu(gdp->bg_inode_bitmap));
1289                         return 0;
1290                 }
1291                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1292                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1293                     last_block)
1294                 {
1295                         ext3_error (sb, "ext3_check_descriptors",
1296                                     "Inode table for group %d"
1297                                     " not in group (block %lu)!",
1298                                     i, (unsigned long)
1299                                         le32_to_cpu(gdp->bg_inode_table));
1300                         return 0;
1301                 }
1302                 first_block += EXT3_BLOCKS_PER_GROUP(sb);
1303         }
1304
1305         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1306         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1307         return 1;
1308 }
1309
1310
1311 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1312  * the superblock) which were deleted from all directories, but held open by
1313  * a process at the time of a crash.  We walk the list and try to delete these
1314  * inodes at recovery time (only with a read-write filesystem).
1315  *
1316  * In order to keep the orphan inode chain consistent during traversal (in
1317  * case of crash during recovery), we link each inode into the superblock
1318  * orphan list_head and handle it the same way as an inode deletion during
1319  * normal operation (which journals the operations for us).
1320  *
1321  * We only do an iget() and an iput() on each inode, which is very safe if we
1322  * accidentally point at an in-use or already deleted inode.  The worst that
1323  * can happen in this case is that we get a "bit already cleared" message from
1324  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1325  * e2fsck was run on this filesystem, and it must have already done the orphan
1326  * inode cleanup for us, so we can safely abort without any further action.
1327  */
1328 static void ext3_orphan_cleanup (struct super_block * sb,
1329                                  struct ext3_super_block * es)
1330 {
1331         unsigned int s_flags = sb->s_flags;
1332         int nr_orphans = 0, nr_truncates = 0;
1333 #ifdef CONFIG_QUOTA
1334         int i;
1335 #endif
1336         if (!es->s_last_orphan) {
1337                 jbd_debug(4, "no orphan inodes to clean up\n");
1338                 return;
1339         }
1340
1341         if (bdev_read_only(sb->s_bdev)) {
1342                 printk(KERN_ERR "EXT3-fs: write access "
1343                         "unavailable, skipping orphan cleanup.\n");
1344                 return;
1345         }
1346
1347         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1348                 if (es->s_last_orphan)
1349                         jbd_debug(1, "Errors on filesystem, "
1350                                   "clearing orphan list.\n");
1351                 es->s_last_orphan = 0;
1352                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1353                 return;
1354         }
1355
1356         if (s_flags & MS_RDONLY) {
1357                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1358                        sb->s_id);
1359                 sb->s_flags &= ~MS_RDONLY;
1360         }
1361 #ifdef CONFIG_QUOTA
1362         /* Needed for iput() to work correctly and not trash data */
1363         sb->s_flags |= MS_ACTIVE;
1364         /* Turn on quotas so that they are updated correctly */
1365         for (i = 0; i < MAXQUOTAS; i++) {
1366                 if (EXT3_SB(sb)->s_qf_names[i]) {
1367                         int ret = ext3_quota_on_mount(sb, i);
1368                         if (ret < 0)
1369                                 printk(KERN_ERR
1370                                         "EXT3-fs: Cannot turn on journalled "
1371                                         "quota: error %d\n", ret);
1372                 }
1373         }
1374 #endif
1375
1376         while (es->s_last_orphan) {
1377                 struct inode *inode;
1378
1379                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1380                 if (IS_ERR(inode)) {
1381                         es->s_last_orphan = 0;
1382                         break;
1383                 }
1384
1385                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1386                 DQUOT_INIT(inode);
1387                 if (inode->i_nlink) {
1388                         printk(KERN_DEBUG
1389                                 "%s: truncating inode %lu to %Ld bytes\n",
1390                                 __FUNCTION__, inode->i_ino, inode->i_size);
1391                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1392                                   inode->i_ino, inode->i_size);
1393                         ext3_truncate(inode);
1394                         nr_truncates++;
1395                 } else {
1396                         printk(KERN_DEBUG
1397                                 "%s: deleting unreferenced inode %lu\n",
1398                                 __FUNCTION__, inode->i_ino);
1399                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1400                                   inode->i_ino);
1401                         nr_orphans++;
1402                 }
1403                 iput(inode);  /* The delete magic happens here! */
1404         }
1405
1406 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1407
1408         if (nr_orphans)
1409                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1410                        sb->s_id, PLURAL(nr_orphans));
1411         if (nr_truncates)
1412                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1413                        sb->s_id, PLURAL(nr_truncates));
1414 #ifdef CONFIG_QUOTA
1415         /* Turn quotas off */
1416         for (i = 0; i < MAXQUOTAS; i++) {
1417                 if (sb_dqopt(sb)->files[i])
1418                         vfs_quota_off(sb, i);
1419         }
1420 #endif
1421         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1422 }
1423
1424 /*
1425  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1426  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1427  * We need to be 1 filesystem block less than the 2^32 sector limit.
1428  */
1429 static loff_t ext3_max_size(int bits)
1430 {
1431         loff_t res = EXT3_NDIR_BLOCKS;
1432         int meta_blocks;
1433         loff_t upper_limit;
1434
1435         /* This is calculated to be the largest file size for a
1436          * dense, file such that the total number of
1437          * sectors in the file, including data and all indirect blocks,
1438          * does not exceed 2^32 -1
1439          * __u32 i_blocks representing the total number of
1440          * 512 bytes blocks of the file
1441          */
1442         upper_limit = (1LL << 32) - 1;
1443
1444         /* total blocks in file system block size */
1445         upper_limit >>= (bits - 9);
1446
1447
1448         /* indirect blocks */
1449         meta_blocks = 1;
1450         /* double indirect blocks */
1451         meta_blocks += 1 + (1LL << (bits-2));
1452         /* tripple indirect blocks */
1453         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1454
1455         upper_limit -= meta_blocks;
1456         upper_limit <<= bits;
1457
1458         res += 1LL << (bits-2);
1459         res += 1LL << (2*(bits-2));
1460         res += 1LL << (3*(bits-2));
1461         res <<= bits;
1462         if (res > upper_limit)
1463                 res = upper_limit;
1464
1465         if (res > MAX_LFS_FILESIZE)
1466                 res = MAX_LFS_FILESIZE;
1467
1468         return res;
1469 }
1470
1471 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1472                                     ext3_fsblk_t logic_sb_block,
1473                                     int nr)
1474 {
1475         struct ext3_sb_info *sbi = EXT3_SB(sb);
1476         unsigned long bg, first_meta_bg;
1477         int has_super = 0;
1478
1479         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1480
1481         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1482             nr < first_meta_bg)
1483                 return (logic_sb_block + nr + 1);
1484         bg = sbi->s_desc_per_block * nr;
1485         if (ext3_bg_has_super(sb, bg))
1486                 has_super = 1;
1487         return (has_super + ext3_group_first_block_no(sb, bg));
1488 }
1489
1490
1491 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1492 {
1493         struct buffer_head * bh;
1494         struct ext3_super_block *es = NULL;
1495         struct ext3_sb_info *sbi;
1496         ext3_fsblk_t block;
1497         ext3_fsblk_t sb_block = get_sb_block(&data);
1498         ext3_fsblk_t logic_sb_block;
1499         unsigned long offset = 0;
1500         unsigned int journal_inum = 0;
1501         unsigned long journal_devnum = 0;
1502         unsigned long def_mount_opts;
1503         struct inode *root;
1504         int blocksize;
1505         int hblock;
1506         int db_count;
1507         int i;
1508         int needs_recovery;
1509         int ret = -EINVAL;
1510         __le32 features;
1511         int err;
1512
1513         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1514         if (!sbi)
1515                 return -ENOMEM;
1516         sb->s_fs_info = sbi;
1517         sbi->s_mount_opt = 0;
1518         sbi->s_resuid = EXT3_DEF_RESUID;
1519         sbi->s_resgid = EXT3_DEF_RESGID;
1520         sbi->s_sb_block = sb_block;
1521
1522         unlock_kernel();
1523
1524         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1525         if (!blocksize) {
1526                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1527                 goto out_fail;
1528         }
1529
1530         /*
1531          * The ext3 superblock will not be buffer aligned for other than 1kB
1532          * block sizes.  We need to calculate the offset from buffer start.
1533          */
1534         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1535                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1536                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1537         } else {
1538                 logic_sb_block = sb_block;
1539         }
1540
1541         if (!(bh = sb_bread(sb, logic_sb_block))) {
1542                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1543                 goto out_fail;
1544         }
1545         /*
1546          * Note: s_es must be initialized as soon as possible because
1547          *       some ext3 macro-instructions depend on its value
1548          */
1549         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1550         sbi->s_es = es;
1551         sb->s_magic = le16_to_cpu(es->s_magic);
1552         if (sb->s_magic != EXT3_SUPER_MAGIC)
1553                 goto cantfind_ext3;
1554
1555         /* Set defaults before we parse the mount options */
1556         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1557         if (def_mount_opts & EXT3_DEFM_DEBUG)
1558                 set_opt(sbi->s_mount_opt, DEBUG);
1559         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1560                 set_opt(sbi->s_mount_opt, GRPID);
1561         if (def_mount_opts & EXT3_DEFM_UID16)
1562                 set_opt(sbi->s_mount_opt, NO_UID32);
1563 #ifdef CONFIG_EXT3_FS_XATTR
1564         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1565                 set_opt(sbi->s_mount_opt, XATTR_USER);
1566 #endif
1567 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1568         if (def_mount_opts & EXT3_DEFM_ACL)
1569                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1570 #endif
1571         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1572                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1573         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1574                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1575         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1576                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1577
1578         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1579                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1580         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1581                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1582         else
1583                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1584
1585         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1586         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1587
1588         set_opt(sbi->s_mount_opt, RESERVATION);
1589
1590         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1591                             NULL, 0))
1592                 goto failed_mount;
1593
1594         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1595                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1596
1597         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1598             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1599              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1600              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1601                 printk(KERN_WARNING
1602                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1603                        "running e2fsck is recommended\n");
1604         /*
1605          * Check feature flags regardless of the revision level, since we
1606          * previously didn't change the revision level when setting the flags,
1607          * so there is a chance incompat flags are set on a rev 0 filesystem.
1608          */
1609         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1610         if (features) {
1611                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1612                        "unsupported optional features (%x).\n",
1613                        sb->s_id, le32_to_cpu(features));
1614                 goto failed_mount;
1615         }
1616         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1617         if (!(sb->s_flags & MS_RDONLY) && features) {
1618                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1619                        "unsupported optional features (%x).\n",
1620                        sb->s_id, le32_to_cpu(features));
1621                 goto failed_mount;
1622         }
1623         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1624
1625         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1626             blocksize > EXT3_MAX_BLOCK_SIZE) {
1627                 printk(KERN_ERR
1628                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1629                        blocksize, sb->s_id);
1630                 goto failed_mount;
1631         }
1632
1633         hblock = bdev_hardsect_size(sb->s_bdev);
1634         if (sb->s_blocksize != blocksize) {
1635                 /*
1636                  * Make sure the blocksize for the filesystem is larger
1637                  * than the hardware sectorsize for the machine.
1638                  */
1639                 if (blocksize < hblock) {
1640                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1641                                "device blocksize %d.\n", blocksize, hblock);
1642                         goto failed_mount;
1643                 }
1644
1645                 brelse (bh);
1646                 if (!sb_set_blocksize(sb, blocksize)) {
1647                         printk(KERN_ERR "EXT3-fs: bad blocksize %d.\n",
1648                                 blocksize);
1649                         goto out_fail;
1650                 }
1651                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1652                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1653                 bh = sb_bread(sb, logic_sb_block);
1654                 if (!bh) {
1655                         printk(KERN_ERR
1656                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1657                         goto failed_mount;
1658                 }
1659                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1660                 sbi->s_es = es;
1661                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1662                         printk (KERN_ERR
1663                                 "EXT3-fs: Magic mismatch, very weird !\n");
1664                         goto failed_mount;
1665                 }
1666         }
1667
1668         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1669
1670         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1671                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1672                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1673         } else {
1674                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1675                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1676                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1677                     (!is_power_of_2(sbi->s_inode_size)) ||
1678                     (sbi->s_inode_size > blocksize)) {
1679                         printk (KERN_ERR
1680                                 "EXT3-fs: unsupported inode size: %d\n",
1681                                 sbi->s_inode_size);
1682                         goto failed_mount;
1683                 }
1684         }
1685         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1686                                    le32_to_cpu(es->s_log_frag_size);
1687         if (blocksize != sbi->s_frag_size) {
1688                 printk(KERN_ERR
1689                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1690                        sbi->s_frag_size, blocksize);
1691                 goto failed_mount;
1692         }
1693         sbi->s_frags_per_block = 1;
1694         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1695         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1696         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1697         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1698                 goto cantfind_ext3;
1699         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1700         if (sbi->s_inodes_per_block == 0)
1701                 goto cantfind_ext3;
1702         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1703                                         sbi->s_inodes_per_block;
1704         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1705         sbi->s_sbh = bh;
1706         sbi->s_mount_state = le16_to_cpu(es->s_state);
1707         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1708         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1709         for (i=0; i < 4; i++)
1710                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1711         sbi->s_def_hash_version = es->s_def_hash_version;
1712
1713         if (sbi->s_blocks_per_group > blocksize * 8) {
1714                 printk (KERN_ERR
1715                         "EXT3-fs: #blocks per group too big: %lu\n",
1716                         sbi->s_blocks_per_group);
1717                 goto failed_mount;
1718         }
1719         if (sbi->s_frags_per_group > blocksize * 8) {
1720                 printk (KERN_ERR
1721                         "EXT3-fs: #fragments per group too big: %lu\n",
1722                         sbi->s_frags_per_group);
1723                 goto failed_mount;
1724         }
1725         if (sbi->s_inodes_per_group > blocksize * 8) {
1726                 printk (KERN_ERR
1727                         "EXT3-fs: #inodes per group too big: %lu\n",
1728                         sbi->s_inodes_per_group);
1729                 goto failed_mount;
1730         }
1731
1732         if (le32_to_cpu(es->s_blocks_count) >
1733                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1734                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1735                         " too large to mount safely\n", sb->s_id);
1736                 if (sizeof(sector_t) < 8)
1737                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1738                                         "enabled\n");
1739                 goto failed_mount;
1740         }
1741
1742         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1743                 goto cantfind_ext3;
1744         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1745                                le32_to_cpu(es->s_first_data_block) - 1)
1746                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1747         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1748                    EXT3_DESC_PER_BLOCK(sb);
1749         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1750                                     GFP_KERNEL);
1751         if (sbi->s_group_desc == NULL) {
1752                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1753                 goto failed_mount;
1754         }
1755
1756         bgl_lock_init(&sbi->s_blockgroup_lock);
1757
1758         for (i = 0; i < db_count; i++) {
1759                 block = descriptor_loc(sb, logic_sb_block, i);
1760                 sbi->s_group_desc[i] = sb_bread(sb, block);
1761                 if (!sbi->s_group_desc[i]) {
1762                         printk (KERN_ERR "EXT3-fs: "
1763                                 "can't read group descriptor %d\n", i);
1764                         db_count = i;
1765                         goto failed_mount2;
1766                 }
1767         }
1768         if (!ext3_check_descriptors (sb)) {
1769                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1770                 goto failed_mount2;
1771         }
1772         sbi->s_gdb_count = db_count;
1773         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1774         spin_lock_init(&sbi->s_next_gen_lock);
1775
1776         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1777                         ext3_count_free_blocks(sb));
1778         if (!err) {
1779                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1780                                 ext3_count_free_inodes(sb));
1781         }
1782         if (!err) {
1783                 err = percpu_counter_init(&sbi->s_dirs_counter,
1784                                 ext3_count_dirs(sb));
1785         }
1786         if (err) {
1787                 printk(KERN_ERR "EXT3-fs: insufficient memory\n");
1788                 goto failed_mount3;
1789         }
1790
1791         /* per fileystem reservation list head & lock */
1792         spin_lock_init(&sbi->s_rsv_window_lock);
1793         sbi->s_rsv_window_root = RB_ROOT;
1794         /* Add a single, static dummy reservation to the start of the
1795          * reservation window list --- it gives us a placeholder for
1796          * append-at-start-of-list which makes the allocation logic
1797          * _much_ simpler. */
1798         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1799         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1800         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1801         sbi->s_rsv_window_head.rsv_goal_size = 0;
1802         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1803
1804         /*
1805          * set up enough so that it can read an inode
1806          */
1807         sb->s_op = &ext3_sops;
1808         sb->s_export_op = &ext3_export_ops;
1809         sb->s_xattr = ext3_xattr_handlers;
1810 #ifdef CONFIG_QUOTA
1811         sb->s_qcop = &ext3_qctl_operations;
1812         sb->dq_op = &ext3_quota_operations;
1813 #endif
1814         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1815
1816         sb->s_root = NULL;
1817
1818         needs_recovery = (es->s_last_orphan != 0 ||
1819                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1820                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1821
1822         /*
1823          * The first inode we look at is the journal inode.  Don't try
1824          * root first: it may be modified in the journal!
1825          */
1826         if (!test_opt(sb, NOLOAD) &&
1827             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1828                 if (ext3_load_journal(sb, es, journal_devnum))
1829                         goto failed_mount3;
1830         } else if (journal_inum) {
1831                 if (ext3_create_journal(sb, es, journal_inum))
1832                         goto failed_mount3;
1833         } else {
1834                 if (!silent)
1835                         printk (KERN_ERR
1836                                 "ext3: No journal on filesystem on %s\n",
1837                                 sb->s_id);
1838                 goto failed_mount3;
1839         }
1840
1841         /* We have now updated the journal if required, so we can
1842          * validate the data journaling mode. */
1843         switch (test_opt(sb, DATA_FLAGS)) {
1844         case 0:
1845                 /* No mode set, assume a default based on the journal
1846                    capabilities: ORDERED_DATA if the journal can
1847                    cope, else JOURNAL_DATA */
1848                 if (journal_check_available_features
1849                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1850                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1851                 else
1852                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1853                 break;
1854
1855         case EXT3_MOUNT_ORDERED_DATA:
1856         case EXT3_MOUNT_WRITEBACK_DATA:
1857                 if (!journal_check_available_features
1858                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1859                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1860                                "requested data journaling mode\n");
1861                         goto failed_mount4;
1862                 }
1863         default:
1864                 break;
1865         }
1866
1867         if (test_opt(sb, NOBH)) {
1868                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1869                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1870                                 "its supported only with writeback mode\n");
1871                         clear_opt(sbi->s_mount_opt, NOBH);
1872                 }
1873         }
1874         /*
1875          * The journal_load will have done any necessary log recovery,
1876          * so we can safely mount the rest of the filesystem now.
1877          */
1878
1879         root = ext3_iget(sb, EXT3_ROOT_INO);
1880         if (IS_ERR(root)) {
1881                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1882                 ret = PTR_ERR(root);
1883                 goto failed_mount4;
1884         }
1885         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1886                 iput(root);
1887                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1888                 goto failed_mount4;
1889         }
1890         sb->s_root = d_alloc_root(root);
1891         if (!sb->s_root) {
1892                 printk(KERN_ERR "EXT3-fs: get root dentry failed\n");
1893                 iput(root);
1894                 ret = -ENOMEM;
1895                 goto failed_mount4;
1896         }
1897
1898         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1899         /*
1900          * akpm: core read_super() calls in here with the superblock locked.
1901          * That deadlocks, because orphan cleanup needs to lock the superblock
1902          * in numerous places.  Here we just pop the lock - it's relatively
1903          * harmless, because we are now ready to accept write_super() requests,
1904          * and aviro says that's the only reason for hanging onto the
1905          * superblock lock.
1906          */
1907         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1908         ext3_orphan_cleanup(sb, es);
1909         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1910         if (needs_recovery)
1911                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1912         ext3_mark_recovery_complete(sb, es);
1913         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1914                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1915                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1916                 "writeback");
1917
1918         lock_kernel();
1919         return 0;
1920
1921 cantfind_ext3:
1922         if (!silent)
1923                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1924                        sb->s_id);
1925         goto failed_mount;
1926
1927 failed_mount4:
1928         journal_destroy(sbi->s_journal);
1929 failed_mount3:
1930         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1931         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1932         percpu_counter_destroy(&sbi->s_dirs_counter);
1933 failed_mount2:
1934         for (i = 0; i < db_count; i++)
1935                 brelse(sbi->s_group_desc[i]);
1936         kfree(sbi->s_group_desc);
1937 failed_mount:
1938 #ifdef CONFIG_QUOTA
1939         for (i = 0; i < MAXQUOTAS; i++)
1940                 kfree(sbi->s_qf_names[i]);
1941 #endif
1942         ext3_blkdev_remove(sbi);
1943         brelse(bh);
1944 out_fail:
1945         sb->s_fs_info = NULL;
1946         kfree(sbi);
1947         lock_kernel();
1948         return ret;
1949 }
1950
1951 /*
1952  * Setup any per-fs journal parameters now.  We'll do this both on
1953  * initial mount, once the journal has been initialised but before we've
1954  * done any recovery; and again on any subsequent remount.
1955  */
1956 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1957 {
1958         struct ext3_sb_info *sbi = EXT3_SB(sb);
1959
1960         if (sbi->s_commit_interval)
1961                 journal->j_commit_interval = sbi->s_commit_interval;
1962         /* We could also set up an ext3-specific default for the commit
1963          * interval here, but for now we'll just fall back to the jbd
1964          * default. */
1965
1966         spin_lock(&journal->j_state_lock);
1967         if (test_opt(sb, BARRIER))
1968                 journal->j_flags |= JFS_BARRIER;
1969         else
1970                 journal->j_flags &= ~JFS_BARRIER;
1971         spin_unlock(&journal->j_state_lock);
1972 }
1973
1974 static journal_t *ext3_get_journal(struct super_block *sb,
1975                                    unsigned int journal_inum)
1976 {
1977         struct inode *journal_inode;
1978         journal_t *journal;
1979
1980         /* First, test for the existence of a valid inode on disk.  Bad
1981          * things happen if we iget() an unused inode, as the subsequent
1982          * iput() will try to delete it. */
1983
1984         journal_inode = ext3_iget(sb, journal_inum);
1985         if (IS_ERR(journal_inode)) {
1986                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1987                 return NULL;
1988         }
1989         if (!journal_inode->i_nlink) {
1990                 make_bad_inode(journal_inode);
1991                 iput(journal_inode);
1992                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1993                 return NULL;
1994         }
1995
1996         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1997                   journal_inode, journal_inode->i_size);
1998         if (!S_ISREG(journal_inode->i_mode)) {
1999                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
2000                 iput(journal_inode);
2001                 return NULL;
2002         }
2003
2004         journal = journal_init_inode(journal_inode);
2005         if (!journal) {
2006                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
2007                 iput(journal_inode);
2008                 return NULL;
2009         }
2010         journal->j_private = sb;
2011         ext3_init_journal_params(sb, journal);
2012         return journal;
2013 }
2014
2015 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2016                                        dev_t j_dev)
2017 {
2018         struct buffer_head * bh;
2019         journal_t *journal;
2020         ext3_fsblk_t start;
2021         ext3_fsblk_t len;
2022         int hblock, blocksize;
2023         ext3_fsblk_t sb_block;
2024         unsigned long offset;
2025         struct ext3_super_block * es;
2026         struct block_device *bdev;
2027
2028         bdev = ext3_blkdev_get(j_dev);
2029         if (bdev == NULL)
2030                 return NULL;
2031
2032         if (bd_claim(bdev, sb)) {
2033                 printk(KERN_ERR
2034                         "EXT3: failed to claim external journal device.\n");
2035                 blkdev_put(bdev);
2036                 return NULL;
2037         }
2038
2039         blocksize = sb->s_blocksize;
2040         hblock = bdev_hardsect_size(bdev);
2041         if (blocksize < hblock) {
2042                 printk(KERN_ERR
2043                         "EXT3-fs: blocksize too small for journal device.\n");
2044                 goto out_bdev;
2045         }
2046
2047         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2048         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2049         set_blocksize(bdev, blocksize);
2050         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2051                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
2052                        "external journal\n");
2053                 goto out_bdev;
2054         }
2055
2056         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2057         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2058             !(le32_to_cpu(es->s_feature_incompat) &
2059               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2060                 printk(KERN_ERR "EXT3-fs: external journal has "
2061                                         "bad superblock\n");
2062                 brelse(bh);
2063                 goto out_bdev;
2064         }
2065
2066         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2067                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
2068                 brelse(bh);
2069                 goto out_bdev;
2070         }
2071
2072         len = le32_to_cpu(es->s_blocks_count);
2073         start = sb_block + 1;
2074         brelse(bh);     /* we're done with the superblock */
2075
2076         journal = journal_init_dev(bdev, sb->s_bdev,
2077                                         start, len, blocksize);
2078         if (!journal) {
2079                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
2080                 goto out_bdev;
2081         }
2082         journal->j_private = sb;
2083         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2084         wait_on_buffer(journal->j_sb_buffer);
2085         if (!buffer_uptodate(journal->j_sb_buffer)) {
2086                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
2087                 goto out_journal;
2088         }
2089         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2090                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
2091                                         "user (unsupported) - %d\n",
2092                         be32_to_cpu(journal->j_superblock->s_nr_users));
2093                 goto out_journal;
2094         }
2095         EXT3_SB(sb)->journal_bdev = bdev;
2096         ext3_init_journal_params(sb, journal);
2097         return journal;
2098 out_journal:
2099         journal_destroy(journal);
2100 out_bdev:
2101         ext3_blkdev_put(bdev);
2102         return NULL;
2103 }
2104
2105 static int ext3_load_journal(struct super_block *sb,
2106                              struct ext3_super_block *es,
2107                              unsigned long journal_devnum)
2108 {
2109         journal_t *journal;
2110         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2111         dev_t journal_dev;
2112         int err = 0;
2113         int really_read_only;
2114
2115         if (journal_devnum &&
2116             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2117                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
2118                         "numbers have changed\n");
2119                 journal_dev = new_decode_dev(journal_devnum);
2120         } else
2121                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2122
2123         really_read_only = bdev_read_only(sb->s_bdev);
2124
2125         /*
2126          * Are we loading a blank journal or performing recovery after a
2127          * crash?  For recovery, we need to check in advance whether we
2128          * can get read-write access to the device.
2129          */
2130
2131         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2132                 if (sb->s_flags & MS_RDONLY) {
2133                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
2134                                         "required on readonly filesystem.\n");
2135                         if (really_read_only) {
2136                                 printk(KERN_ERR "EXT3-fs: write access "
2137                                         "unavailable, cannot proceed.\n");
2138                                 return -EROFS;
2139                         }
2140                         printk (KERN_INFO "EXT3-fs: write access will "
2141                                         "be enabled during recovery.\n");
2142                 }
2143         }
2144
2145         if (journal_inum && journal_dev) {
2146                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
2147                        "and inode journals!\n");
2148                 return -EINVAL;
2149         }
2150
2151         if (journal_inum) {
2152                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2153                         return -EINVAL;
2154         } else {
2155                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2156                         return -EINVAL;
2157         }
2158
2159         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2160                 err = journal_update_format(journal);
2161                 if (err)  {
2162                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
2163                         journal_destroy(journal);
2164                         return err;
2165                 }
2166         }
2167
2168         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2169                 err = journal_wipe(journal, !really_read_only);
2170         if (!err)
2171                 err = journal_load(journal);
2172
2173         if (err) {
2174                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2175                 journal_destroy(journal);
2176                 return err;
2177         }
2178
2179         EXT3_SB(sb)->s_journal = journal;
2180         ext3_clear_journal_err(sb, es);
2181
2182         if (journal_devnum &&
2183             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2184                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2185                 sb->s_dirt = 1;
2186
2187                 /* Make sure we flush the recovery flag to disk. */
2188                 ext3_commit_super(sb, es, 1);
2189         }
2190
2191         return 0;
2192 }
2193
2194 static int ext3_create_journal(struct super_block * sb,
2195                                struct ext3_super_block * es,
2196                                unsigned int journal_inum)
2197 {
2198         journal_t *journal;
2199         int err;
2200
2201         if (sb->s_flags & MS_RDONLY) {
2202                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2203                                 "create journal.\n");
2204                 return -EROFS;
2205         }
2206
2207         journal = ext3_get_journal(sb, journal_inum);
2208         if (!journal)
2209                 return -EINVAL;
2210
2211         printk(KERN_INFO "EXT3-fs: creating new journal on inode %u\n",
2212                journal_inum);
2213
2214         err = journal_create(journal);
2215         if (err) {
2216                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2217                 journal_destroy(journal);
2218                 return -EIO;
2219         }
2220
2221         EXT3_SB(sb)->s_journal = journal;
2222
2223         ext3_update_dynamic_rev(sb);
2224         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2225         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2226
2227         es->s_journal_inum = cpu_to_le32(journal_inum);
2228         sb->s_dirt = 1;
2229
2230         /* Make sure we flush the recovery flag to disk. */
2231         ext3_commit_super(sb, es, 1);
2232
2233         return 0;
2234 }
2235
2236 static void ext3_commit_super (struct super_block * sb,
2237                                struct ext3_super_block * es,
2238                                int sync)
2239 {
2240         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2241
2242         if (!sbh)
2243                 return;
2244         es->s_wtime = cpu_to_le32(get_seconds());
2245         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2246         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2247         BUFFER_TRACE(sbh, "marking dirty");
2248         mark_buffer_dirty(sbh);
2249         if (sync)
2250                 sync_dirty_buffer(sbh);
2251 }
2252
2253
2254 /*
2255  * Have we just finished recovery?  If so, and if we are mounting (or
2256  * remounting) the filesystem readonly, then we will end up with a
2257  * consistent fs on disk.  Record that fact.
2258  */
2259 static void ext3_mark_recovery_complete(struct super_block * sb,
2260                                         struct ext3_super_block * es)
2261 {
2262         journal_t *journal = EXT3_SB(sb)->s_journal;
2263
2264         journal_lock_updates(journal);
2265         journal_flush(journal);
2266         lock_super(sb);
2267         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2268             sb->s_flags & MS_RDONLY) {
2269                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2270                 sb->s_dirt = 0;
2271                 ext3_commit_super(sb, es, 1);
2272         }
2273         unlock_super(sb);
2274         journal_unlock_updates(journal);
2275 }
2276
2277 /*
2278  * If we are mounting (or read-write remounting) a filesystem whose journal
2279  * has recorded an error from a previous lifetime, move that error to the
2280  * main filesystem now.
2281  */
2282 static void ext3_clear_journal_err(struct super_block * sb,
2283                                    struct ext3_super_block * es)
2284 {
2285         journal_t *journal;
2286         int j_errno;
2287         const char *errstr;
2288
2289         journal = EXT3_SB(sb)->s_journal;
2290
2291         /*
2292          * Now check for any error status which may have been recorded in the
2293          * journal by a prior ext3_error() or ext3_abort()
2294          */
2295
2296         j_errno = journal_errno(journal);
2297         if (j_errno) {
2298                 char nbuf[16];
2299
2300                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2301                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
2302                              "from previous mount: %s", errstr);
2303                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
2304                              "filesystem check.");
2305
2306                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2307                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2308                 ext3_commit_super (sb, es, 1);
2309
2310                 journal_clear_err(journal);
2311         }
2312 }
2313
2314 /*
2315  * Force the running and committing transactions to commit,
2316  * and wait on the commit.
2317  */
2318 int ext3_force_commit(struct super_block *sb)
2319 {
2320         journal_t *journal;
2321         int ret;
2322
2323         if (sb->s_flags & MS_RDONLY)
2324                 return 0;
2325
2326         journal = EXT3_SB(sb)->s_journal;
2327         sb->s_dirt = 0;
2328         ret = ext3_journal_force_commit(journal);
2329         return ret;
2330 }
2331
2332 /*
2333  * Ext3 always journals updates to the superblock itself, so we don't
2334  * have to propagate any other updates to the superblock on disk at this
2335  * point.  Just start an async writeback to get the buffers on their way
2336  * to the disk.
2337  *
2338  * This implicitly triggers the writebehind on sync().
2339  */
2340
2341 static void ext3_write_super (struct super_block * sb)
2342 {
2343         if (mutex_trylock(&sb->s_lock) != 0)
2344                 BUG();
2345         sb->s_dirt = 0;
2346 }
2347
2348 static int ext3_sync_fs(struct super_block *sb, int wait)
2349 {
2350         tid_t target;
2351
2352         sb->s_dirt = 0;
2353         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2354                 if (wait)
2355                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2356         }
2357         return 0;
2358 }
2359
2360 /*
2361  * LVM calls this function before a (read-only) snapshot is created.  This
2362  * gives us a chance to flush the journal completely and mark the fs clean.
2363  */
2364 static void ext3_write_super_lockfs(struct super_block *sb)
2365 {
2366         sb->s_dirt = 0;
2367
2368         if (!(sb->s_flags & MS_RDONLY)) {
2369                 journal_t *journal = EXT3_SB(sb)->s_journal;
2370
2371                 /* Now we set up the journal barrier. */
2372                 journal_lock_updates(journal);
2373                 journal_flush(journal);
2374
2375                 /* Journal blocked and flushed, clear needs_recovery flag. */
2376                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2377                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2378         }
2379 }
2380
2381 /*
2382  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2383  * flag here, even though the filesystem is not technically dirty yet.
2384  */
2385 static void ext3_unlockfs(struct super_block *sb)
2386 {
2387         if (!(sb->s_flags & MS_RDONLY)) {
2388                 lock_super(sb);
2389                 /* Reser the needs_recovery flag before the fs is unlocked. */
2390                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2391                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2392                 unlock_super(sb);
2393                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2394         }
2395 }
2396
2397 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2398 {
2399         struct ext3_super_block * es;
2400         struct ext3_sb_info *sbi = EXT3_SB(sb);
2401         ext3_fsblk_t n_blocks_count = 0;
2402         unsigned long old_sb_flags;
2403         struct ext3_mount_options old_opts;
2404         int err;
2405 #ifdef CONFIG_QUOTA
2406         int i;
2407 #endif
2408
2409         /* Store the original options */
2410         old_sb_flags = sb->s_flags;
2411         old_opts.s_mount_opt = sbi->s_mount_opt;
2412         old_opts.s_resuid = sbi->s_resuid;
2413         old_opts.s_resgid = sbi->s_resgid;
2414         old_opts.s_commit_interval = sbi->s_commit_interval;
2415 #ifdef CONFIG_QUOTA
2416         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2417         for (i = 0; i < MAXQUOTAS; i++)
2418                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2419 #endif
2420
2421         /*
2422          * Allow the "check" option to be passed as a remount option.
2423          */
2424         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2425                 err = -EINVAL;
2426                 goto restore_opts;
2427         }
2428
2429         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2430                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2431
2432         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2433                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2434
2435         es = sbi->s_es;
2436
2437         ext3_init_journal_params(sb, sbi->s_journal);
2438
2439         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2440                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2441                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2442                         err = -EROFS;
2443                         goto restore_opts;
2444                 }
2445
2446                 if (*flags & MS_RDONLY) {
2447                         /*
2448                          * First of all, the unconditional stuff we have to do
2449                          * to disable replay of the journal when we next remount
2450                          */
2451                         sb->s_flags |= MS_RDONLY;
2452
2453                         /*
2454                          * OK, test if we are remounting a valid rw partition
2455                          * readonly, and if so set the rdonly flag and then
2456                          * mark the partition as valid again.
2457                          */
2458                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2459                             (sbi->s_mount_state & EXT3_VALID_FS))
2460                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2461
2462                         /*
2463                          * We have to unlock super so that we can wait for
2464                          * transactions.
2465                          */
2466                         unlock_super(sb);
2467                         ext3_mark_recovery_complete(sb, es);
2468                         lock_super(sb);
2469                 } else {
2470                         __le32 ret;
2471                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2472                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2473                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2474                                        "remount RDWR because of unsupported "
2475                                        "optional features (%x).\n",
2476                                        sb->s_id, le32_to_cpu(ret));
2477                                 err = -EROFS;
2478                                 goto restore_opts;
2479                         }
2480
2481                         /*
2482                          * If we have an unprocessed orphan list hanging
2483                          * around from a previously readonly bdev mount,
2484                          * require a full umount/remount for now.
2485                          */
2486                         if (es->s_last_orphan) {
2487                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2488                                        "remount RDWR because of unprocessed "
2489                                        "orphan inode list.  Please "
2490                                        "umount/remount instead.\n",
2491                                        sb->s_id);
2492                                 err = -EINVAL;
2493                                 goto restore_opts;
2494                         }
2495
2496                         /*
2497                          * Mounting a RDONLY partition read-write, so reread
2498                          * and store the current valid flag.  (It may have
2499                          * been changed by e2fsck since we originally mounted
2500                          * the partition.)
2501                          */
2502                         ext3_clear_journal_err(sb, es);
2503                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2504                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2505                                 goto restore_opts;
2506                         if (!ext3_setup_super (sb, es, 0))
2507                                 sb->s_flags &= ~MS_RDONLY;
2508                 }
2509         }
2510 #ifdef CONFIG_QUOTA
2511         /* Release old quota file names */
2512         for (i = 0; i < MAXQUOTAS; i++)
2513                 if (old_opts.s_qf_names[i] &&
2514                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2515                         kfree(old_opts.s_qf_names[i]);
2516 #endif
2517         return 0;
2518 restore_opts:
2519         sb->s_flags = old_sb_flags;
2520         sbi->s_mount_opt = old_opts.s_mount_opt;
2521         sbi->s_resuid = old_opts.s_resuid;
2522         sbi->s_resgid = old_opts.s_resgid;
2523         sbi->s_commit_interval = old_opts.s_commit_interval;
2524 #ifdef CONFIG_QUOTA
2525         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2526         for (i = 0; i < MAXQUOTAS; i++) {
2527                 if (sbi->s_qf_names[i] &&
2528                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2529                         kfree(sbi->s_qf_names[i]);
2530                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2531         }
2532 #endif
2533         return err;
2534 }
2535
2536 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2537 {
2538         struct super_block *sb = dentry->d_sb;
2539         struct ext3_sb_info *sbi = EXT3_SB(sb);
2540         struct ext3_super_block *es = sbi->s_es;
2541         u64 fsid;
2542
2543         if (test_opt(sb, MINIX_DF)) {
2544                 sbi->s_overhead_last = 0;
2545         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2546                 unsigned long ngroups = sbi->s_groups_count, i;
2547                 ext3_fsblk_t overhead = 0;
2548                 smp_rmb();
2549
2550                 /*
2551                  * Compute the overhead (FS structures).  This is constant
2552                  * for a given filesystem unless the number of block groups
2553                  * changes so we cache the previous value until it does.
2554                  */
2555
2556                 /*
2557                  * All of the blocks before first_data_block are
2558                  * overhead
2559                  */
2560                 overhead = le32_to_cpu(es->s_first_data_block);
2561
2562                 /*
2563                  * Add the overhead attributed to the superblock and
2564                  * block group descriptors.  If the sparse superblocks
2565                  * feature is turned on, then not all groups have this.
2566                  */
2567                 for (i = 0; i < ngroups; i++) {
2568                         overhead += ext3_bg_has_super(sb, i) +
2569                                 ext3_bg_num_gdb(sb, i);
2570                         cond_resched();
2571                 }
2572
2573                 /*
2574                  * Every block group has an inode bitmap, a block
2575                  * bitmap, and an inode table.
2576                  */
2577                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2578                 sbi->s_overhead_last = overhead;
2579                 smp_wmb();
2580                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2581         }
2582
2583         buf->f_type = EXT3_SUPER_MAGIC;
2584         buf->f_bsize = sb->s_blocksize;
2585         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2586         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2587         es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
2588         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2589         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2590                 buf->f_bavail = 0;
2591         buf->f_files = le32_to_cpu(es->s_inodes_count);
2592         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2593         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2594         buf->f_namelen = EXT3_NAME_LEN;
2595         fsid = le64_to_cpup((void *)es->s_uuid) ^
2596                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2597         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2598         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2599         return 0;
2600 }
2601
2602 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2603  * is locked for write. Otherwise the are possible deadlocks:
2604  * Process 1                         Process 2
2605  * ext3_create()                     quota_sync()
2606  *   journal_start()                   write_dquot()
2607  *   DQUOT_INIT()                        down(dqio_mutex)
2608  *     down(dqio_mutex)                    journal_start()
2609  *
2610  */
2611
2612 #ifdef CONFIG_QUOTA
2613
2614 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2615 {
2616         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2617 }
2618
2619 static int ext3_dquot_initialize(struct inode *inode, int type)
2620 {
2621         handle_t *handle;
2622         int ret, err;
2623
2624         /* We may create quota structure so we need to reserve enough blocks */
2625         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2626         if (IS_ERR(handle))
2627                 return PTR_ERR(handle);
2628         ret = dquot_initialize(inode, type);
2629         err = ext3_journal_stop(handle);
2630         if (!ret)
2631                 ret = err;
2632         return ret;
2633 }
2634
2635 static int ext3_dquot_drop(struct inode *inode)
2636 {
2637         handle_t *handle;
2638         int ret, err;
2639
2640         /* We may delete quota structure so we need to reserve enough blocks */
2641         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2642         if (IS_ERR(handle))
2643                 return PTR_ERR(handle);
2644         ret = dquot_drop(inode);
2645         err = ext3_journal_stop(handle);
2646         if (!ret)
2647                 ret = err;
2648         return ret;
2649 }
2650
2651 static int ext3_write_dquot(struct dquot *dquot)
2652 {
2653         int ret, err;
2654         handle_t *handle;
2655         struct inode *inode;
2656
2657         inode = dquot_to_inode(dquot);
2658         handle = ext3_journal_start(inode,
2659                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2660         if (IS_ERR(handle))
2661                 return PTR_ERR(handle);
2662         ret = dquot_commit(dquot);
2663         err = ext3_journal_stop(handle);
2664         if (!ret)
2665                 ret = err;
2666         return ret;
2667 }
2668
2669 static int ext3_acquire_dquot(struct dquot *dquot)
2670 {
2671         int ret, err;
2672         handle_t *handle;
2673
2674         handle = ext3_journal_start(dquot_to_inode(dquot),
2675                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2676         if (IS_ERR(handle))
2677                 return PTR_ERR(handle);
2678         ret = dquot_acquire(dquot);
2679         err = ext3_journal_stop(handle);
2680         if (!ret)
2681                 ret = err;
2682         return ret;
2683 }
2684
2685 static int ext3_release_dquot(struct dquot *dquot)
2686 {
2687         int ret, err;
2688         handle_t *handle;
2689
2690         handle = ext3_journal_start(dquot_to_inode(dquot),
2691                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2692         if (IS_ERR(handle)) {
2693                 /* Release dquot anyway to avoid endless cycle in dqput() */
2694                 dquot_release(dquot);
2695                 return PTR_ERR(handle);
2696         }
2697         ret = dquot_release(dquot);
2698         err = ext3_journal_stop(handle);
2699         if (!ret)
2700                 ret = err;
2701         return ret;
2702 }
2703
2704 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2705 {
2706         /* Are we journalling quotas? */
2707         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2708             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2709                 dquot_mark_dquot_dirty(dquot);
2710                 return ext3_write_dquot(dquot);
2711         } else {
2712                 return dquot_mark_dquot_dirty(dquot);
2713         }
2714 }
2715
2716 static int ext3_write_info(struct super_block *sb, int type)
2717 {
2718         int ret, err;
2719         handle_t *handle;
2720
2721         /* Data block + inode block */
2722         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2723         if (IS_ERR(handle))
2724                 return PTR_ERR(handle);
2725         ret = dquot_commit_info(sb, type);
2726         err = ext3_journal_stop(handle);
2727         if (!ret)
2728                 ret = err;
2729         return ret;
2730 }
2731
2732 /*
2733  * Turn on quotas during mount time - we need to find
2734  * the quota file and such...
2735  */
2736 static int ext3_quota_on_mount(struct super_block *sb, int type)
2737 {
2738         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2739                         EXT3_SB(sb)->s_jquota_fmt, type);
2740 }
2741
2742 /*
2743  * Standard function to be called on quota_on
2744  */
2745 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2746                          char *path)
2747 {
2748         int err;
2749         struct nameidata nd;
2750
2751         if (!test_opt(sb, QUOTA))
2752                 return -EINVAL;
2753         /* Not journalling quota? */
2754         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2755             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2756                 return vfs_quota_on(sb, type, format_id, path);
2757         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2758         if (err)
2759                 return err;
2760         /* Quotafile not on the same filesystem? */
2761         if (nd.path.mnt->mnt_sb != sb) {
2762                 path_release(&nd);
2763                 return -EXDEV;
2764         }
2765         /* Quotafile not of fs root? */
2766         if (nd.path.dentry->d_parent->d_inode != sb->s_root->d_inode)
2767                 printk(KERN_WARNING
2768                         "EXT3-fs: Quota file not on filesystem root. "
2769                         "Journalled quota will not work.\n");
2770         path_release(&nd);
2771         return vfs_quota_on(sb, type, format_id, path);
2772 }
2773
2774 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2775  * acquiring the locks... As quota files are never truncated and quota code
2776  * itself serializes the operations (and noone else should touch the files)
2777  * we don't have to be afraid of races */
2778 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2779                                size_t len, loff_t off)
2780 {
2781         struct inode *inode = sb_dqopt(sb)->files[type];
2782         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2783         int err = 0;
2784         int offset = off & (sb->s_blocksize - 1);
2785         int tocopy;
2786         size_t toread;
2787         struct buffer_head *bh;
2788         loff_t i_size = i_size_read(inode);
2789
2790         if (off > i_size)
2791                 return 0;
2792         if (off+len > i_size)
2793                 len = i_size-off;
2794         toread = len;
2795         while (toread > 0) {
2796                 tocopy = sb->s_blocksize - offset < toread ?
2797                                 sb->s_blocksize - offset : toread;
2798                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2799                 if (err)
2800                         return err;
2801                 if (!bh)        /* A hole? */
2802                         memset(data, 0, tocopy);
2803                 else
2804                         memcpy(data, bh->b_data+offset, tocopy);
2805                 brelse(bh);
2806                 offset = 0;
2807                 toread -= tocopy;
2808                 data += tocopy;
2809                 blk++;
2810         }
2811         return len;
2812 }
2813
2814 /* Write to quotafile (we know the transaction is already started and has
2815  * enough credits) */
2816 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2817                                 const char *data, size_t len, loff_t off)
2818 {
2819         struct inode *inode = sb_dqopt(sb)->files[type];
2820         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2821         int err = 0;
2822         int offset = off & (sb->s_blocksize - 1);
2823         int tocopy;
2824         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2825         size_t towrite = len;
2826         struct buffer_head *bh;
2827         handle_t *handle = journal_current_handle();
2828
2829         if (!handle) {
2830                 printk(KERN_WARNING "EXT3-fs: Quota write (off=%Lu, len=%Lu)"
2831                         " cancelled because transaction is not started.\n",
2832                         (unsigned long long)off, (unsigned long long)len);
2833                 return -EIO;
2834         }
2835         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2836         while (towrite > 0) {
2837                 tocopy = sb->s_blocksize - offset < towrite ?
2838                                 sb->s_blocksize - offset : towrite;
2839                 bh = ext3_bread(handle, inode, blk, 1, &err);
2840                 if (!bh)
2841                         goto out;
2842                 if (journal_quota) {
2843                         err = ext3_journal_get_write_access(handle, bh);
2844                         if (err) {
2845                                 brelse(bh);
2846                                 goto out;
2847                         }
2848                 }
2849                 lock_buffer(bh);
2850                 memcpy(bh->b_data+offset, data, tocopy);
2851                 flush_dcache_page(bh->b_page);
2852                 unlock_buffer(bh);
2853                 if (journal_quota)
2854                         err = ext3_journal_dirty_metadata(handle, bh);
2855                 else {
2856                         /* Always do at least ordered writes for quotas */
2857                         err = ext3_journal_dirty_data(handle, bh);
2858                         mark_buffer_dirty(bh);
2859                 }
2860                 brelse(bh);
2861                 if (err)
2862                         goto out;
2863                 offset = 0;
2864                 towrite -= tocopy;
2865                 data += tocopy;
2866                 blk++;
2867         }
2868 out:
2869         if (len == towrite)
2870                 return err;
2871         if (inode->i_size < off+len-towrite) {
2872                 i_size_write(inode, off+len-towrite);
2873                 EXT3_I(inode)->i_disksize = inode->i_size;
2874         }
2875         inode->i_version++;
2876         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2877         ext3_mark_inode_dirty(handle, inode);
2878         mutex_unlock(&inode->i_mutex);
2879         return len - towrite;
2880 }
2881
2882 #endif
2883
2884 static int ext3_get_sb(struct file_system_type *fs_type,
2885         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2886 {
2887         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
2888 }
2889
2890 static struct file_system_type ext3_fs_type = {
2891         .owner          = THIS_MODULE,
2892         .name           = "ext3",
2893         .get_sb         = ext3_get_sb,
2894         .kill_sb        = kill_block_super,
2895         .fs_flags       = FS_REQUIRES_DEV,
2896 };
2897
2898 static int __init init_ext3_fs(void)
2899 {
2900         int err = init_ext3_xattr();
2901         if (err)
2902                 return err;
2903         err = init_inodecache();
2904         if (err)
2905                 goto out1;
2906         err = register_filesystem(&ext3_fs_type);
2907         if (err)
2908                 goto out;
2909         return 0;
2910 out:
2911         destroy_inodecache();
2912 out1:
2913         exit_ext3_xattr();
2914         return err;
2915 }
2916
2917 static void __exit exit_ext3_fs(void)
2918 {
2919         unregister_filesystem(&ext3_fs_type);
2920         destroy_inodecache();
2921         exit_ext3_xattr();
2922 }
2923
2924 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2925 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2926 MODULE_LICENSE("GPL");
2927 module_init(init_ext3_fs)
2928 module_exit(exit_ext3_fs)