ext3: provide function to release metadata pages under memory pressure
[linux-2.6.git] / fs / ext3 / super.c
1 /*
2  *  linux/fs/ext3/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/module.h>
20 #include <linux/string.h>
21 #include <linux/fs.h>
22 #include <linux/time.h>
23 #include <linux/jbd.h>
24 #include <linux/ext3_fs.h>
25 #include <linux/ext3_jbd.h>
26 #include <linux/slab.h>
27 #include <linux/init.h>
28 #include <linux/blkdev.h>
29 #include <linux/parser.h>
30 #include <linux/smp_lock.h>
31 #include <linux/buffer_head.h>
32 #include <linux/exportfs.h>
33 #include <linux/vfs.h>
34 #include <linux/random.h>
35 #include <linux/mount.h>
36 #include <linux/namei.h>
37 #include <linux/quotaops.h>
38 #include <linux/seq_file.h>
39 #include <linux/log2.h>
40
41 #include <asm/uaccess.h>
42
43 #include "xattr.h"
44 #include "acl.h"
45 #include "namei.h"
46
47 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
48                              unsigned long journal_devnum);
49 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
50                                unsigned int);
51 static void ext3_commit_super (struct super_block * sb,
52                                struct ext3_super_block * es,
53                                int sync);
54 static void ext3_mark_recovery_complete(struct super_block * sb,
55                                         struct ext3_super_block * es);
56 static void ext3_clear_journal_err(struct super_block * sb,
57                                    struct ext3_super_block * es);
58 static int ext3_sync_fs(struct super_block *sb, int wait);
59 static const char *ext3_decode_error(struct super_block * sb, int errno,
60                                      char nbuf[16]);
61 static int ext3_remount (struct super_block * sb, int * flags, char * data);
62 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
63 static void ext3_unlockfs(struct super_block *sb);
64 static void ext3_write_super (struct super_block * sb);
65 static void ext3_write_super_lockfs(struct super_block *sb);
66
67 /*
68  * Wrappers for journal_start/end.
69  *
70  * The only special thing we need to do here is to make sure that all
71  * journal_end calls result in the superblock being marked dirty, so
72  * that sync() will call the filesystem's write_super callback if
73  * appropriate.
74  */
75 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
76 {
77         journal_t *journal;
78
79         if (sb->s_flags & MS_RDONLY)
80                 return ERR_PTR(-EROFS);
81
82         /* Special case here: if the journal has aborted behind our
83          * backs (eg. EIO in the commit thread), then we still need to
84          * take the FS itself readonly cleanly. */
85         journal = EXT3_SB(sb)->s_journal;
86         if (is_journal_aborted(journal)) {
87                 ext3_abort(sb, __func__,
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         if (EXT3_SB(sb)->s_journal)
285                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
286 }
287
288 void ext3_warning (struct super_block * sb, const char * function,
289                    const char * fmt, ...)
290 {
291         va_list args;
292
293         va_start(args, fmt);
294         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
295                sb->s_id, function);
296         vprintk(fmt, args);
297         printk("\n");
298         va_end(args);
299 }
300
301 void ext3_update_dynamic_rev(struct super_block *sb)
302 {
303         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
304
305         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
306                 return;
307
308         ext3_warning(sb, __func__,
309                      "updating to rev %d because of new feature flag, "
310                      "running e2fsck is recommended",
311                      EXT3_DYNAMIC_REV);
312
313         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
314         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
315         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
316         /* leave es->s_feature_*compat flags alone */
317         /* es->s_uuid will be set by e2fsck if empty */
318
319         /*
320          * The rest of the superblock fields should be zero, and if not it
321          * means they are likely already in use, so leave them alone.  We
322          * can leave it up to e2fsck to clean up any inconsistencies there.
323          */
324 }
325
326 /*
327  * Open the external journal device
328  */
329 static struct block_device *ext3_blkdev_get(dev_t dev)
330 {
331         struct block_device *bdev;
332         char b[BDEVNAME_SIZE];
333
334         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
335         if (IS_ERR(bdev))
336                 goto fail;
337         return bdev;
338
339 fail:
340         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
341                         __bdevname(dev, b), PTR_ERR(bdev));
342         return NULL;
343 }
344
345 /*
346  * Release the journal device
347  */
348 static int ext3_blkdev_put(struct block_device *bdev)
349 {
350         bd_release(bdev);
351         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
352 }
353
354 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
355 {
356         struct block_device *bdev;
357         int ret = -ENODEV;
358
359         bdev = sbi->journal_bdev;
360         if (bdev) {
361                 ret = ext3_blkdev_put(bdev);
362                 sbi->journal_bdev = NULL;
363         }
364         return ret;
365 }
366
367 static inline struct inode *orphan_list_entry(struct list_head *l)
368 {
369         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
370 }
371
372 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
373 {
374         struct list_head *l;
375
376         printk(KERN_ERR "sb orphan head is %d\n",
377                le32_to_cpu(sbi->s_es->s_last_orphan));
378
379         printk(KERN_ERR "sb_info orphan list:\n");
380         list_for_each(l, &sbi->s_orphan) {
381                 struct inode *inode = orphan_list_entry(l);
382                 printk(KERN_ERR "  "
383                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
384                        inode->i_sb->s_id, inode->i_ino, inode,
385                        inode->i_mode, inode->i_nlink,
386                        NEXT_ORPHAN(inode));
387         }
388 }
389
390 static void ext3_put_super (struct super_block * sb)
391 {
392         struct ext3_sb_info *sbi = EXT3_SB(sb);
393         struct ext3_super_block *es = sbi->s_es;
394         int i, err;
395
396         ext3_xattr_put_super(sb);
397         err = journal_destroy(sbi->s_journal);
398         sbi->s_journal = NULL;
399         if (err < 0)
400                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
401
402         if (!(sb->s_flags & MS_RDONLY)) {
403                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
404                 es->s_state = cpu_to_le16(sbi->s_mount_state);
405                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
406                 mark_buffer_dirty(sbi->s_sbh);
407                 ext3_commit_super(sb, es, 1);
408         }
409
410         for (i = 0; i < sbi->s_gdb_count; i++)
411                 brelse(sbi->s_group_desc[i]);
412         kfree(sbi->s_group_desc);
413         percpu_counter_destroy(&sbi->s_freeblocks_counter);
414         percpu_counter_destroy(&sbi->s_freeinodes_counter);
415         percpu_counter_destroy(&sbi->s_dirs_counter);
416         brelse(sbi->s_sbh);
417 #ifdef CONFIG_QUOTA
418         for (i = 0; i < MAXQUOTAS; i++)
419                 kfree(sbi->s_qf_names[i]);
420 #endif
421
422         /* Debugging code just in case the in-memory inode orphan list
423          * isn't empty.  The on-disk one can be non-empty if we've
424          * detected an error and taken the fs readonly, but the
425          * in-memory list had better be clean by this point. */
426         if (!list_empty(&sbi->s_orphan))
427                 dump_orphan_list(sb, sbi);
428         J_ASSERT(list_empty(&sbi->s_orphan));
429
430         invalidate_bdev(sb->s_bdev);
431         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
432                 /*
433                  * Invalidate the journal device's buffers.  We don't want them
434                  * floating about in memory - the physical journal device may
435                  * hotswapped, and it breaks the `ro-after' testing code.
436                  */
437                 sync_blockdev(sbi->journal_bdev);
438                 invalidate_bdev(sbi->journal_bdev);
439                 ext3_blkdev_remove(sbi);
440         }
441         sb->s_fs_info = NULL;
442         kfree(sbi);
443         return;
444 }
445
446 static struct kmem_cache *ext3_inode_cachep;
447
448 /*
449  * Called inside transaction, so use GFP_NOFS
450  */
451 static struct inode *ext3_alloc_inode(struct super_block *sb)
452 {
453         struct ext3_inode_info *ei;
454
455         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
456         if (!ei)
457                 return NULL;
458 #ifdef CONFIG_EXT3_FS_POSIX_ACL
459         ei->i_acl = EXT3_ACL_NOT_CACHED;
460         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
461 #endif
462         ei->i_block_alloc_info = NULL;
463         ei->vfs_inode.i_version = 1;
464         return &ei->vfs_inode;
465 }
466
467 static void ext3_destroy_inode(struct inode *inode)
468 {
469         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
470                 printk("EXT3 Inode %p: orphan list check failed!\n",
471                         EXT3_I(inode));
472                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
473                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
474                                 false);
475                 dump_stack();
476         }
477         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
478 }
479
480 static void init_once(void *foo)
481 {
482         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
483
484         INIT_LIST_HEAD(&ei->i_orphan);
485 #ifdef CONFIG_EXT3_FS_XATTR
486         init_rwsem(&ei->xattr_sem);
487 #endif
488         mutex_init(&ei->truncate_mutex);
489         inode_init_once(&ei->vfs_inode);
490 }
491
492 static int init_inodecache(void)
493 {
494         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
495                                              sizeof(struct ext3_inode_info),
496                                              0, (SLAB_RECLAIM_ACCOUNT|
497                                                 SLAB_MEM_SPREAD),
498                                              init_once);
499         if (ext3_inode_cachep == NULL)
500                 return -ENOMEM;
501         return 0;
502 }
503
504 static void destroy_inodecache(void)
505 {
506         kmem_cache_destroy(ext3_inode_cachep);
507 }
508
509 static void ext3_clear_inode(struct inode *inode)
510 {
511         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
512 #ifdef CONFIG_EXT3_FS_POSIX_ACL
513         if (EXT3_I(inode)->i_acl &&
514                         EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
515                 posix_acl_release(EXT3_I(inode)->i_acl);
516                 EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
517         }
518         if (EXT3_I(inode)->i_default_acl &&
519                         EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
520                 posix_acl_release(EXT3_I(inode)->i_default_acl);
521                 EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
522         }
523 #endif
524         ext3_discard_reservation(inode);
525         EXT3_I(inode)->i_block_alloc_info = NULL;
526         if (unlikely(rsv))
527                 kfree(rsv);
528 }
529
530 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
531 {
532 #if defined(CONFIG_QUOTA)
533         struct ext3_sb_info *sbi = EXT3_SB(sb);
534
535         if (sbi->s_jquota_fmt)
536                 seq_printf(seq, ",jqfmt=%s",
537                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
538
539         if (sbi->s_qf_names[USRQUOTA])
540                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
541
542         if (sbi->s_qf_names[GRPQUOTA])
543                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
544
545         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
546                 seq_puts(seq, ",usrquota");
547
548         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
549                 seq_puts(seq, ",grpquota");
550 #endif
551 }
552
553 /*
554  * Show an option if
555  *  - it's set to a non-default value OR
556  *  - if the per-sb default is different from the global default
557  */
558 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
559 {
560         struct super_block *sb = vfs->mnt_sb;
561         struct ext3_sb_info *sbi = EXT3_SB(sb);
562         struct ext3_super_block *es = sbi->s_es;
563         unsigned long def_mount_opts;
564
565         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
566
567         if (sbi->s_sb_block != 1)
568                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
569         if (test_opt(sb, MINIX_DF))
570                 seq_puts(seq, ",minixdf");
571         if (test_opt(sb, GRPID))
572                 seq_puts(seq, ",grpid");
573         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
574                 seq_puts(seq, ",nogrpid");
575         if (sbi->s_resuid != EXT3_DEF_RESUID ||
576             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
577                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
578         }
579         if (sbi->s_resgid != EXT3_DEF_RESGID ||
580             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
581                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
582         }
583         if (test_opt(sb, ERRORS_RO)) {
584                 int def_errors = le16_to_cpu(es->s_errors);
585
586                 if (def_errors == EXT3_ERRORS_PANIC ||
587                     def_errors == EXT3_ERRORS_CONTINUE) {
588                         seq_puts(seq, ",errors=remount-ro");
589                 }
590         }
591         if (test_opt(sb, ERRORS_CONT))
592                 seq_puts(seq, ",errors=continue");
593         if (test_opt(sb, ERRORS_PANIC))
594                 seq_puts(seq, ",errors=panic");
595         if (test_opt(sb, NO_UID32))
596                 seq_puts(seq, ",nouid32");
597         if (test_opt(sb, DEBUG))
598                 seq_puts(seq, ",debug");
599         if (test_opt(sb, OLDALLOC))
600                 seq_puts(seq, ",oldalloc");
601 #ifdef CONFIG_EXT3_FS_XATTR
602         if (test_opt(sb, XATTR_USER))
603                 seq_puts(seq, ",user_xattr");
604         if (!test_opt(sb, XATTR_USER) &&
605             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
606                 seq_puts(seq, ",nouser_xattr");
607         }
608 #endif
609 #ifdef CONFIG_EXT3_FS_POSIX_ACL
610         if (test_opt(sb, POSIX_ACL))
611                 seq_puts(seq, ",acl");
612         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
613                 seq_puts(seq, ",noacl");
614 #endif
615         if (!test_opt(sb, RESERVATION))
616                 seq_puts(seq, ",noreservation");
617         if (sbi->s_commit_interval) {
618                 seq_printf(seq, ",commit=%u",
619                            (unsigned) (sbi->s_commit_interval / HZ));
620         }
621         if (test_opt(sb, BARRIER))
622                 seq_puts(seq, ",barrier=1");
623         if (test_opt(sb, NOBH))
624                 seq_puts(seq, ",nobh");
625
626         if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA)
627                 seq_puts(seq, ",data=journal");
628         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA)
629                 seq_puts(seq, ",data=ordered");
630         else if (test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)
631                 seq_puts(seq, ",data=writeback");
632
633         if (test_opt(sb, DATA_ERR_ABORT))
634                 seq_puts(seq, ",data_err=abort");
635
636         ext3_show_quota_options(seq, sb);
637
638         return 0;
639 }
640
641
642 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
643                 u64 ino, u32 generation)
644 {
645         struct inode *inode;
646
647         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
648                 return ERR_PTR(-ESTALE);
649         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
650                 return ERR_PTR(-ESTALE);
651
652         /* iget isn't really right if the inode is currently unallocated!!
653          *
654          * ext3_read_inode will return a bad_inode if the inode had been
655          * deleted, so we should be safe.
656          *
657          * Currently we don't know the generation for parent directory, so
658          * a generation of 0 means "accept any"
659          */
660         inode = ext3_iget(sb, ino);
661         if (IS_ERR(inode))
662                 return ERR_CAST(inode);
663         if (generation && inode->i_generation != generation) {
664                 iput(inode);
665                 return ERR_PTR(-ESTALE);
666         }
667
668         return inode;
669 }
670
671 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
672                 int fh_len, int fh_type)
673 {
674         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
675                                     ext3_nfs_get_inode);
676 }
677
678 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
679                 int fh_len, int fh_type)
680 {
681         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
682                                     ext3_nfs_get_inode);
683 }
684
685 /*
686  * Try to release metadata pages (indirect blocks, directories) which are
687  * mapped via the block device.  Since these pages could have journal heads
688  * which would prevent try_to_free_buffers() from freeing them, we must use
689  * jbd layer's try_to_free_buffers() function to release them.
690  */
691 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
692                                  gfp_t wait)
693 {
694         journal_t *journal = EXT3_SB(sb)->s_journal;
695
696         WARN_ON(PageChecked(page));
697         if (!page_has_buffers(page))
698                 return 0;
699         if (journal)
700                 return journal_try_to_free_buffers(journal, page, 
701                                                    wait & ~__GFP_WAIT);
702         return try_to_free_buffers(page);
703 }
704
705 #ifdef CONFIG_QUOTA
706 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
707 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
708
709 static int ext3_dquot_initialize(struct inode *inode, int type);
710 static int ext3_dquot_drop(struct inode *inode);
711 static int ext3_write_dquot(struct dquot *dquot);
712 static int ext3_acquire_dquot(struct dquot *dquot);
713 static int ext3_release_dquot(struct dquot *dquot);
714 static int ext3_mark_dquot_dirty(struct dquot *dquot);
715 static int ext3_write_info(struct super_block *sb, int type);
716 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
717                                 char *path, int remount);
718 static int ext3_quota_on_mount(struct super_block *sb, int type);
719 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
720                                size_t len, loff_t off);
721 static ssize_t ext3_quota_write(struct super_block *sb, int type,
722                                 const char *data, size_t len, loff_t off);
723
724 static struct dquot_operations ext3_quota_operations = {
725         .initialize     = ext3_dquot_initialize,
726         .drop           = ext3_dquot_drop,
727         .alloc_space    = dquot_alloc_space,
728         .alloc_inode    = dquot_alloc_inode,
729         .free_space     = dquot_free_space,
730         .free_inode     = dquot_free_inode,
731         .transfer       = dquot_transfer,
732         .write_dquot    = ext3_write_dquot,
733         .acquire_dquot  = ext3_acquire_dquot,
734         .release_dquot  = ext3_release_dquot,
735         .mark_dirty     = ext3_mark_dquot_dirty,
736         .write_info     = ext3_write_info
737 };
738
739 static struct quotactl_ops ext3_qctl_operations = {
740         .quota_on       = ext3_quota_on,
741         .quota_off      = vfs_quota_off,
742         .quota_sync     = vfs_quota_sync,
743         .get_info       = vfs_get_dqinfo,
744         .set_info       = vfs_set_dqinfo,
745         .get_dqblk      = vfs_get_dqblk,
746         .set_dqblk      = vfs_set_dqblk
747 };
748 #endif
749
750 static const struct super_operations ext3_sops = {
751         .alloc_inode    = ext3_alloc_inode,
752         .destroy_inode  = ext3_destroy_inode,
753         .write_inode    = ext3_write_inode,
754         .dirty_inode    = ext3_dirty_inode,
755         .delete_inode   = ext3_delete_inode,
756         .put_super      = ext3_put_super,
757         .write_super    = ext3_write_super,
758         .sync_fs        = ext3_sync_fs,
759         .write_super_lockfs = ext3_write_super_lockfs,
760         .unlockfs       = ext3_unlockfs,
761         .statfs         = ext3_statfs,
762         .remount_fs     = ext3_remount,
763         .clear_inode    = ext3_clear_inode,
764         .show_options   = ext3_show_options,
765 #ifdef CONFIG_QUOTA
766         .quota_read     = ext3_quota_read,
767         .quota_write    = ext3_quota_write,
768 #endif
769         .bdev_try_to_free_page = bdev_try_to_free_page,
770 };
771
772 static const struct export_operations ext3_export_ops = {
773         .fh_to_dentry = ext3_fh_to_dentry,
774         .fh_to_parent = ext3_fh_to_parent,
775         .get_parent = ext3_get_parent,
776 };
777
778 enum {
779         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
780         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
781         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
782         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
783         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
784         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
785         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
786         Opt_data_err_abort, Opt_data_err_ignore,
787         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
788         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
789         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
790         Opt_grpquota
791 };
792
793 static const match_table_t tokens = {
794         {Opt_bsd_df, "bsddf"},
795         {Opt_minix_df, "minixdf"},
796         {Opt_grpid, "grpid"},
797         {Opt_grpid, "bsdgroups"},
798         {Opt_nogrpid, "nogrpid"},
799         {Opt_nogrpid, "sysvgroups"},
800         {Opt_resgid, "resgid=%u"},
801         {Opt_resuid, "resuid=%u"},
802         {Opt_sb, "sb=%u"},
803         {Opt_err_cont, "errors=continue"},
804         {Opt_err_panic, "errors=panic"},
805         {Opt_err_ro, "errors=remount-ro"},
806         {Opt_nouid32, "nouid32"},
807         {Opt_nocheck, "nocheck"},
808         {Opt_nocheck, "check=none"},
809         {Opt_debug, "debug"},
810         {Opt_oldalloc, "oldalloc"},
811         {Opt_orlov, "orlov"},
812         {Opt_user_xattr, "user_xattr"},
813         {Opt_nouser_xattr, "nouser_xattr"},
814         {Opt_acl, "acl"},
815         {Opt_noacl, "noacl"},
816         {Opt_reservation, "reservation"},
817         {Opt_noreservation, "noreservation"},
818         {Opt_noload, "noload"},
819         {Opt_nobh, "nobh"},
820         {Opt_bh, "bh"},
821         {Opt_commit, "commit=%u"},
822         {Opt_journal_update, "journal=update"},
823         {Opt_journal_inum, "journal=%u"},
824         {Opt_journal_dev, "journal_dev=%u"},
825         {Opt_abort, "abort"},
826         {Opt_data_journal, "data=journal"},
827         {Opt_data_ordered, "data=ordered"},
828         {Opt_data_writeback, "data=writeback"},
829         {Opt_data_err_abort, "data_err=abort"},
830         {Opt_data_err_ignore, "data_err=ignore"},
831         {Opt_offusrjquota, "usrjquota="},
832         {Opt_usrjquota, "usrjquota=%s"},
833         {Opt_offgrpjquota, "grpjquota="},
834         {Opt_grpjquota, "grpjquota=%s"},
835         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
836         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
837         {Opt_grpquota, "grpquota"},
838         {Opt_noquota, "noquota"},
839         {Opt_quota, "quota"},
840         {Opt_usrquota, "usrquota"},
841         {Opt_barrier, "barrier=%u"},
842         {Opt_resize, "resize"},
843         {Opt_err, NULL},
844 };
845
846 static ext3_fsblk_t get_sb_block(void **data)
847 {
848         ext3_fsblk_t    sb_block;
849         char            *options = (char *) *data;
850
851         if (!options || strncmp(options, "sb=", 3) != 0)
852                 return 1;       /* Default location */
853         options += 3;
854         /*todo: use simple_strtoll with >32bit ext3 */
855         sb_block = simple_strtoul(options, &options, 0);
856         if (*options && *options != ',') {
857                 printk("EXT3-fs: Invalid sb specification: %s\n",
858                        (char *) *data);
859                 return 1;
860         }
861         if (*options == ',')
862                 options++;
863         *data = (void *) options;
864         return sb_block;
865 }
866
867 static int parse_options (char *options, struct super_block *sb,
868                           unsigned int *inum, unsigned long *journal_devnum,
869                           ext3_fsblk_t *n_blocks_count, int is_remount)
870 {
871         struct ext3_sb_info *sbi = EXT3_SB(sb);
872         char * p;
873         substring_t args[MAX_OPT_ARGS];
874         int data_opt = 0;
875         int option;
876 #ifdef CONFIG_QUOTA
877         int qtype, qfmt;
878         char *qname;
879 #endif
880
881         if (!options)
882                 return 1;
883
884         while ((p = strsep (&options, ",")) != NULL) {
885                 int token;
886                 if (!*p)
887                         continue;
888
889                 token = match_token(p, tokens, args);
890                 switch (token) {
891                 case Opt_bsd_df:
892                         clear_opt (sbi->s_mount_opt, MINIX_DF);
893                         break;
894                 case Opt_minix_df:
895                         set_opt (sbi->s_mount_opt, MINIX_DF);
896                         break;
897                 case Opt_grpid:
898                         set_opt (sbi->s_mount_opt, GRPID);
899                         break;
900                 case Opt_nogrpid:
901                         clear_opt (sbi->s_mount_opt, GRPID);
902                         break;
903                 case Opt_resuid:
904                         if (match_int(&args[0], &option))
905                                 return 0;
906                         sbi->s_resuid = option;
907                         break;
908                 case Opt_resgid:
909                         if (match_int(&args[0], &option))
910                                 return 0;
911                         sbi->s_resgid = option;
912                         break;
913                 case Opt_sb:
914                         /* handled by get_sb_block() instead of here */
915                         /* *sb_block = match_int(&args[0]); */
916                         break;
917                 case Opt_err_panic:
918                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
919                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
920                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
921                         break;
922                 case Opt_err_ro:
923                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
924                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
925                         set_opt (sbi->s_mount_opt, ERRORS_RO);
926                         break;
927                 case Opt_err_cont:
928                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
929                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
930                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
931                         break;
932                 case Opt_nouid32:
933                         set_opt (sbi->s_mount_opt, NO_UID32);
934                         break;
935                 case Opt_nocheck:
936                         clear_opt (sbi->s_mount_opt, CHECK);
937                         break;
938                 case Opt_debug:
939                         set_opt (sbi->s_mount_opt, DEBUG);
940                         break;
941                 case Opt_oldalloc:
942                         set_opt (sbi->s_mount_opt, OLDALLOC);
943                         break;
944                 case Opt_orlov:
945                         clear_opt (sbi->s_mount_opt, OLDALLOC);
946                         break;
947 #ifdef CONFIG_EXT3_FS_XATTR
948                 case Opt_user_xattr:
949                         set_opt (sbi->s_mount_opt, XATTR_USER);
950                         break;
951                 case Opt_nouser_xattr:
952                         clear_opt (sbi->s_mount_opt, XATTR_USER);
953                         break;
954 #else
955                 case Opt_user_xattr:
956                 case Opt_nouser_xattr:
957                         printk("EXT3 (no)user_xattr options not supported\n");
958                         break;
959 #endif
960 #ifdef CONFIG_EXT3_FS_POSIX_ACL
961                 case Opt_acl:
962                         set_opt(sbi->s_mount_opt, POSIX_ACL);
963                         break;
964                 case Opt_noacl:
965                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
966                         break;
967 #else
968                 case Opt_acl:
969                 case Opt_noacl:
970                         printk("EXT3 (no)acl options not supported\n");
971                         break;
972 #endif
973                 case Opt_reservation:
974                         set_opt(sbi->s_mount_opt, RESERVATION);
975                         break;
976                 case Opt_noreservation:
977                         clear_opt(sbi->s_mount_opt, RESERVATION);
978                         break;
979                 case Opt_journal_update:
980                         /* @@@ FIXME */
981                         /* Eventually we will want to be able to create
982                            a journal file here.  For now, only allow the
983                            user to specify an existing inode to be the
984                            journal file. */
985                         if (is_remount) {
986                                 printk(KERN_ERR "EXT3-fs: cannot specify "
987                                        "journal on remount\n");
988                                 return 0;
989                         }
990                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
991                         break;
992                 case Opt_journal_inum:
993                         if (is_remount) {
994                                 printk(KERN_ERR "EXT3-fs: cannot specify "
995                                        "journal on remount\n");
996                                 return 0;
997                         }
998                         if (match_int(&args[0], &option))
999                                 return 0;
1000                         *inum = option;
1001                         break;
1002                 case Opt_journal_dev:
1003                         if (is_remount) {
1004                                 printk(KERN_ERR "EXT3-fs: cannot specify "
1005                                        "journal on remount\n");
1006                                 return 0;
1007                         }
1008                         if (match_int(&args[0], &option))
1009                                 return 0;
1010                         *journal_devnum = option;
1011                         break;
1012                 case Opt_noload:
1013                         set_opt (sbi->s_mount_opt, NOLOAD);
1014                         break;
1015                 case Opt_commit:
1016                         if (match_int(&args[0], &option))
1017                                 return 0;
1018                         if (option < 0)
1019                                 return 0;
1020                         if (option == 0)
1021                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1022                         sbi->s_commit_interval = HZ * option;
1023                         break;
1024                 case Opt_data_journal:
1025                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1026                         goto datacheck;
1027                 case Opt_data_ordered:
1028                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1029                         goto datacheck;
1030                 case Opt_data_writeback:
1031                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1032                 datacheck:
1033                         if (is_remount) {
1034                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
1035                                                 != data_opt) {
1036                                         printk(KERN_ERR
1037                                                 "EXT3-fs: cannot change data "
1038                                                 "mode on remount\n");
1039                                         return 0;
1040                                 }
1041                         } else {
1042                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
1043                                 sbi->s_mount_opt |= data_opt;
1044                         }
1045                         break;
1046                 case Opt_data_err_abort:
1047                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1048                         break;
1049                 case Opt_data_err_ignore:
1050                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1051                         break;
1052 #ifdef CONFIG_QUOTA
1053                 case Opt_usrjquota:
1054                         qtype = USRQUOTA;
1055                         goto set_qf_name;
1056                 case Opt_grpjquota:
1057                         qtype = GRPQUOTA;
1058 set_qf_name:
1059                         if ((sb_any_quota_enabled(sb) ||
1060                              sb_any_quota_suspended(sb)) &&
1061                             !sbi->s_qf_names[qtype]) {
1062                                 printk(KERN_ERR
1063                                         "EXT3-fs: Cannot change journaled "
1064                                         "quota options when quota turned on.\n");
1065                                 return 0;
1066                         }
1067                         qname = match_strdup(&args[0]);
1068                         if (!qname) {
1069                                 printk(KERN_ERR
1070                                         "EXT3-fs: not enough memory for "
1071                                         "storing quotafile name.\n");
1072                                 return 0;
1073                         }
1074                         if (sbi->s_qf_names[qtype] &&
1075                             strcmp(sbi->s_qf_names[qtype], qname)) {
1076                                 printk(KERN_ERR
1077                                         "EXT3-fs: %s quota file already "
1078                                         "specified.\n", QTYPE2NAME(qtype));
1079                                 kfree(qname);
1080                                 return 0;
1081                         }
1082                         sbi->s_qf_names[qtype] = qname;
1083                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1084                                 printk(KERN_ERR
1085                                         "EXT3-fs: quotafile must be on "
1086                                         "filesystem root.\n");
1087                                 kfree(sbi->s_qf_names[qtype]);
1088                                 sbi->s_qf_names[qtype] = NULL;
1089                                 return 0;
1090                         }
1091                         set_opt(sbi->s_mount_opt, QUOTA);
1092                         break;
1093                 case Opt_offusrjquota:
1094                         qtype = USRQUOTA;
1095                         goto clear_qf_name;
1096                 case Opt_offgrpjquota:
1097                         qtype = GRPQUOTA;
1098 clear_qf_name:
1099                         if ((sb_any_quota_enabled(sb) ||
1100                              sb_any_quota_suspended(sb)) &&
1101                             sbi->s_qf_names[qtype]) {
1102                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1103                                         "journaled quota options when "
1104                                         "quota turned on.\n");
1105                                 return 0;
1106                         }
1107                         /*
1108                          * The space will be released later when all options
1109                          * are confirmed to be correct
1110                          */
1111                         sbi->s_qf_names[qtype] = NULL;
1112                         break;
1113                 case Opt_jqfmt_vfsold:
1114                         qfmt = QFMT_VFS_OLD;
1115                         goto set_qf_format;
1116                 case Opt_jqfmt_vfsv0:
1117                         qfmt = QFMT_VFS_V0;
1118 set_qf_format:
1119                         if ((sb_any_quota_enabled(sb) ||
1120                              sb_any_quota_suspended(sb)) &&
1121                             sbi->s_jquota_fmt != qfmt) {
1122                                 printk(KERN_ERR "EXT3-fs: Cannot change "
1123                                         "journaled quota options when "
1124                                         "quota turned on.\n");
1125                                 return 0;
1126                         }
1127                         sbi->s_jquota_fmt = qfmt;
1128                         break;
1129                 case Opt_quota:
1130                 case Opt_usrquota:
1131                         set_opt(sbi->s_mount_opt, QUOTA);
1132                         set_opt(sbi->s_mount_opt, USRQUOTA);
1133                         break;
1134                 case Opt_grpquota:
1135                         set_opt(sbi->s_mount_opt, QUOTA);
1136                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1137                         break;
1138                 case Opt_noquota:
1139                         if (sb_any_quota_enabled(sb) ||
1140                             sb_any_quota_suspended(sb)) {
1141                                 printk(KERN_ERR "EXT3-fs: Cannot change quota "
1142                                         "options when quota turned on.\n");
1143                                 return 0;
1144                         }
1145                         clear_opt(sbi->s_mount_opt, QUOTA);
1146                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1147                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1148                         break;
1149 #else
1150                 case Opt_quota:
1151                 case Opt_usrquota:
1152                 case Opt_grpquota:
1153                         printk(KERN_ERR
1154                                 "EXT3-fs: quota options not supported.\n");
1155                         break;
1156                 case Opt_usrjquota:
1157                 case Opt_grpjquota:
1158                 case Opt_offusrjquota:
1159                 case Opt_offgrpjquota:
1160                 case Opt_jqfmt_vfsold:
1161                 case Opt_jqfmt_vfsv0:
1162                         printk(KERN_ERR
1163                                 "EXT3-fs: journaled quota options not "
1164                                 "supported.\n");
1165                         break;
1166                 case Opt_noquota:
1167                         break;
1168 #endif
1169                 case Opt_abort:
1170                         set_opt(sbi->s_mount_opt, ABORT);
1171                         break;
1172                 case Opt_barrier:
1173                         if (match_int(&args[0], &option))
1174                                 return 0;
1175                         if (option)
1176                                 set_opt(sbi->s_mount_opt, BARRIER);
1177                         else
1178                                 clear_opt(sbi->s_mount_opt, BARRIER);
1179                         break;
1180                 case Opt_ignore:
1181                         break;
1182                 case Opt_resize:
1183                         if (!is_remount) {
1184                                 printk("EXT3-fs: resize option only available "
1185                                         "for remount\n");
1186                                 return 0;
1187                         }
1188                         if (match_int(&args[0], &option) != 0)
1189                                 return 0;
1190                         *n_blocks_count = option;
1191                         break;
1192                 case Opt_nobh:
1193                         set_opt(sbi->s_mount_opt, NOBH);
1194                         break;
1195                 case Opt_bh:
1196                         clear_opt(sbi->s_mount_opt, NOBH);
1197                         break;
1198                 default:
1199                         printk (KERN_ERR
1200                                 "EXT3-fs: Unrecognized mount option \"%s\" "
1201                                 "or missing value\n", p);
1202                         return 0;
1203                 }
1204         }
1205 #ifdef CONFIG_QUOTA
1206         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1207                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1208                      sbi->s_qf_names[USRQUOTA])
1209                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1210
1211                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1212                      sbi->s_qf_names[GRPQUOTA])
1213                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1214
1215                 if ((sbi->s_qf_names[USRQUOTA] &&
1216                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1217                     (sbi->s_qf_names[GRPQUOTA] &&
1218                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1219                         printk(KERN_ERR "EXT3-fs: old and new quota "
1220                                         "format mixing.\n");
1221                         return 0;
1222                 }
1223
1224                 if (!sbi->s_jquota_fmt) {
1225                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1226                                         "not specified.\n");
1227                         return 0;
1228                 }
1229         } else {
1230                 if (sbi->s_jquota_fmt) {
1231                         printk(KERN_ERR "EXT3-fs: journaled quota format "
1232                                         "specified with no journaling "
1233                                         "enabled.\n");
1234                         return 0;
1235                 }
1236         }
1237 #endif
1238         return 1;
1239 }
1240
1241 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1242                             int read_only)
1243 {
1244         struct ext3_sb_info *sbi = EXT3_SB(sb);
1245         int res = 0;
1246
1247         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1248                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
1249                         "forcing read-only mode\n");
1250                 res = MS_RDONLY;
1251         }
1252         if (read_only)
1253                 return res;
1254         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1255                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
1256                         "running e2fsck is recommended\n");
1257         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1258                 printk (KERN_WARNING
1259                         "EXT3-fs warning: mounting fs with errors, "
1260                         "running e2fsck is recommended\n");
1261         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1262                  le16_to_cpu(es->s_mnt_count) >=
1263                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1264                 printk (KERN_WARNING
1265                         "EXT3-fs warning: maximal mount count reached, "
1266                         "running e2fsck is recommended\n");
1267         else if (le32_to_cpu(es->s_checkinterval) &&
1268                 (le32_to_cpu(es->s_lastcheck) +
1269                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1270                 printk (KERN_WARNING
1271                         "EXT3-fs warning: checktime reached, "
1272                         "running e2fsck is recommended\n");
1273 #if 0
1274                 /* @@@ We _will_ want to clear the valid bit if we find
1275                    inconsistencies, to force a fsck at reboot.  But for
1276                    a plain journaled filesystem we can keep it set as
1277                    valid forever! :) */
1278         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1279 #endif
1280         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1281                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1282         le16_add_cpu(&es->s_mnt_count, 1);
1283         es->s_mtime = cpu_to_le32(get_seconds());
1284         ext3_update_dynamic_rev(sb);
1285         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1286
1287         ext3_commit_super(sb, es, 1);
1288         if (test_opt(sb, DEBUG))
1289                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1290                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1291                         sb->s_blocksize,
1292                         sbi->s_groups_count,
1293                         EXT3_BLOCKS_PER_GROUP(sb),
1294                         EXT3_INODES_PER_GROUP(sb),
1295                         sbi->s_mount_opt);
1296
1297         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1298         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1299                 char b[BDEVNAME_SIZE];
1300
1301                 printk("external journal on %s\n",
1302                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1303         } else {
1304                 printk("internal journal\n");
1305         }
1306         return res;
1307 }
1308
1309 /* Called at mount-time, super-block is locked */
1310 static int ext3_check_descriptors(struct super_block *sb)
1311 {
1312         struct ext3_sb_info *sbi = EXT3_SB(sb);
1313         int i;
1314
1315         ext3_debug ("Checking group descriptors");
1316
1317         for (i = 0; i < sbi->s_groups_count; i++) {
1318                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1319                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1320                 ext3_fsblk_t last_block;
1321
1322                 if (i == sbi->s_groups_count - 1)
1323                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1324                 else
1325                         last_block = first_block +
1326                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1327
1328                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1329                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1330                 {
1331                         ext3_error (sb, "ext3_check_descriptors",
1332                                     "Block bitmap for group %d"
1333                                     " not in group (block %lu)!",
1334                                     i, (unsigned long)
1335                                         le32_to_cpu(gdp->bg_block_bitmap));
1336                         return 0;
1337                 }
1338                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1339                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1340                 {
1341                         ext3_error (sb, "ext3_check_descriptors",
1342                                     "Inode bitmap for group %d"
1343                                     " not in group (block %lu)!",
1344                                     i, (unsigned long)
1345                                         le32_to_cpu(gdp->bg_inode_bitmap));
1346                         return 0;
1347                 }
1348                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1349                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1350                     last_block)
1351                 {
1352                         ext3_error (sb, "ext3_check_descriptors",
1353                                     "Inode table for group %d"
1354                                     " not in group (block %lu)!",
1355                                     i, (unsigned long)
1356                                         le32_to_cpu(gdp->bg_inode_table));
1357                         return 0;
1358                 }
1359         }
1360
1361         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1362         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1363         return 1;
1364 }
1365
1366
1367 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1368  * the superblock) which were deleted from all directories, but held open by
1369  * a process at the time of a crash.  We walk the list and try to delete these
1370  * inodes at recovery time (only with a read-write filesystem).
1371  *
1372  * In order to keep the orphan inode chain consistent during traversal (in
1373  * case of crash during recovery), we link each inode into the superblock
1374  * orphan list_head and handle it the same way as an inode deletion during
1375  * normal operation (which journals the operations for us).
1376  *
1377  * We only do an iget() and an iput() on each inode, which is very safe if we
1378  * accidentally point at an in-use or already deleted inode.  The worst that
1379  * can happen in this case is that we get a "bit already cleared" message from
1380  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1381  * e2fsck was run on this filesystem, and it must have already done the orphan
1382  * inode cleanup for us, so we can safely abort without any further action.
1383  */
1384 static void ext3_orphan_cleanup (struct super_block * sb,
1385                                  struct ext3_super_block * es)
1386 {
1387         unsigned int s_flags = sb->s_flags;
1388         int nr_orphans = 0, nr_truncates = 0;
1389 #ifdef CONFIG_QUOTA
1390         int i;
1391 #endif
1392         if (!es->s_last_orphan) {
1393                 jbd_debug(4, "no orphan inodes to clean up\n");
1394                 return;
1395         }
1396
1397         if (bdev_read_only(sb->s_bdev)) {
1398                 printk(KERN_ERR "EXT3-fs: write access "
1399                         "unavailable, skipping orphan cleanup.\n");
1400                 return;
1401         }
1402
1403         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1404                 if (es->s_last_orphan)
1405                         jbd_debug(1, "Errors on filesystem, "
1406                                   "clearing orphan list.\n");
1407                 es->s_last_orphan = 0;
1408                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1409                 return;
1410         }
1411
1412         if (s_flags & MS_RDONLY) {
1413                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1414                        sb->s_id);
1415                 sb->s_flags &= ~MS_RDONLY;
1416         }
1417 #ifdef CONFIG_QUOTA
1418         /* Needed for iput() to work correctly and not trash data */
1419         sb->s_flags |= MS_ACTIVE;
1420         /* Turn on quotas so that they are updated correctly */
1421         for (i = 0; i < MAXQUOTAS; i++) {
1422                 if (EXT3_SB(sb)->s_qf_names[i]) {
1423                         int ret = ext3_quota_on_mount(sb, i);
1424                         if (ret < 0)
1425                                 printk(KERN_ERR
1426                                         "EXT3-fs: Cannot turn on journaled "
1427                                         "quota: error %d\n", ret);
1428                 }
1429         }
1430 #endif
1431
1432         while (es->s_last_orphan) {
1433                 struct inode *inode;
1434
1435                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1436                 if (IS_ERR(inode)) {
1437                         es->s_last_orphan = 0;
1438                         break;
1439                 }
1440
1441                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1442                 DQUOT_INIT(inode);
1443                 if (inode->i_nlink) {
1444                         printk(KERN_DEBUG
1445                                 "%s: truncating inode %lu to %Ld bytes\n",
1446                                 __func__, inode->i_ino, inode->i_size);
1447                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1448                                   inode->i_ino, inode->i_size);
1449                         ext3_truncate(inode);
1450                         nr_truncates++;
1451                 } else {
1452                         printk(KERN_DEBUG
1453                                 "%s: deleting unreferenced inode %lu\n",
1454                                 __func__, inode->i_ino);
1455                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1456                                   inode->i_ino);
1457                         nr_orphans++;
1458                 }
1459                 iput(inode);  /* The delete magic happens here! */
1460         }
1461
1462 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1463
1464         if (nr_orphans)
1465                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1466                        sb->s_id, PLURAL(nr_orphans));
1467         if (nr_truncates)
1468                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1469                        sb->s_id, PLURAL(nr_truncates));
1470 #ifdef CONFIG_QUOTA
1471         /* Turn quotas off */
1472         for (i = 0; i < MAXQUOTAS; i++) {
1473                 if (sb_dqopt(sb)->files[i])
1474                         vfs_quota_off(sb, i, 0);
1475         }
1476 #endif
1477         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1478 }
1479
1480 /*
1481  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1482  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1483  * We need to be 1 filesystem block less than the 2^32 sector limit.
1484  */
1485 static loff_t ext3_max_size(int bits)
1486 {
1487         loff_t res = EXT3_NDIR_BLOCKS;
1488         int meta_blocks;
1489         loff_t upper_limit;
1490
1491         /* This is calculated to be the largest file size for a
1492          * dense, file such that the total number of
1493          * sectors in the file, including data and all indirect blocks,
1494          * does not exceed 2^32 -1
1495          * __u32 i_blocks representing the total number of
1496          * 512 bytes blocks of the file
1497          */
1498         upper_limit = (1LL << 32) - 1;
1499
1500         /* total blocks in file system block size */
1501         upper_limit >>= (bits - 9);
1502
1503
1504         /* indirect blocks */
1505         meta_blocks = 1;
1506         /* double indirect blocks */
1507         meta_blocks += 1 + (1LL << (bits-2));
1508         /* tripple indirect blocks */
1509         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1510
1511         upper_limit -= meta_blocks;
1512         upper_limit <<= bits;
1513
1514         res += 1LL << (bits-2);
1515         res += 1LL << (2*(bits-2));
1516         res += 1LL << (3*(bits-2));
1517         res <<= bits;
1518         if (res > upper_limit)
1519                 res = upper_limit;
1520
1521         if (res > MAX_LFS_FILESIZE)
1522                 res = MAX_LFS_FILESIZE;
1523
1524         return res;
1525 }
1526
1527 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1528                                     ext3_fsblk_t logic_sb_block,
1529                                     int nr)
1530 {
1531         struct ext3_sb_info *sbi = EXT3_SB(sb);
1532         unsigned long bg, first_meta_bg;
1533         int has_super = 0;
1534
1535         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1536
1537         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1538             nr < first_meta_bg)
1539                 return (logic_sb_block + nr + 1);
1540         bg = sbi->s_desc_per_block * nr;
1541         if (ext3_bg_has_super(sb, bg))
1542                 has_super = 1;
1543         return (has_super + ext3_group_first_block_no(sb, bg));
1544 }
1545
1546
1547 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1548 {
1549         struct buffer_head * bh;
1550         struct ext3_super_block *es = NULL;
1551         struct ext3_sb_info *sbi;
1552         ext3_fsblk_t block;
1553         ext3_fsblk_t sb_block = get_sb_block(&data);
1554         ext3_fsblk_t logic_sb_block;
1555         unsigned long offset = 0;
1556         unsigned int journal_inum = 0;
1557         unsigned long journal_devnum = 0;
1558         unsigned long def_mount_opts;
1559         struct inode *root;
1560         int blocksize;
1561         int hblock;
1562         int db_count;
1563         int i;
1564         int needs_recovery;
1565         int ret = -EINVAL;
1566         __le32 features;
1567         int err;
1568
1569         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1570         if (!sbi)
1571                 return -ENOMEM;
1572         sb->s_fs_info = sbi;
1573         sbi->s_mount_opt = 0;
1574         sbi->s_resuid = EXT3_DEF_RESUID;
1575         sbi->s_resgid = EXT3_DEF_RESGID;
1576         sbi->s_sb_block = sb_block;
1577
1578         unlock_kernel();
1579
1580         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1581         if (!blocksize) {
1582                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1583                 goto out_fail;
1584         }
1585
1586         /*
1587          * The ext3 superblock will not be buffer aligned for other than 1kB
1588          * block sizes.  We need to calculate the offset from buffer start.
1589          */
1590         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1591                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1592                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1593         } else {
1594                 logic_sb_block = sb_block;
1595         }
1596
1597         if (!(bh = sb_bread(sb, logic_sb_block))) {
1598                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1599                 goto out_fail;
1600         }
1601         /*
1602          * Note: s_es must be initialized as soon as possible because
1603          *       some ext3 macro-instructions depend on its value
1604          */
1605         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1606         sbi->s_es = es;
1607         sb->s_magic = le16_to_cpu(es->s_magic);
1608         if (sb->s_magic != EXT3_SUPER_MAGIC)
1609                 goto cantfind_ext3;
1610
1611         /* Set defaults before we parse the mount options */
1612         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1613         if (def_mount_opts & EXT3_DEFM_DEBUG)
1614                 set_opt(sbi->s_mount_opt, DEBUG);
1615         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1616                 set_opt(sbi->s_mount_opt, GRPID);
1617         if (def_mount_opts & EXT3_DEFM_UID16)
1618                 set_opt(sbi->s_mount_opt, NO_UID32);
1619 #ifdef CONFIG_EXT3_FS_XATTR
1620         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1621                 set_opt(sbi->s_mount_opt, XATTR_USER);
1622 #endif
1623 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1624         if (def_mount_opts & EXT3_DEFM_ACL)
1625                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1626 #endif
1627         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1628                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1629         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1630                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1631         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1632                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1633
1634         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1635                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1636         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1637                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1638         else
1639                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1640
1641         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1642         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1643
1644         set_opt(sbi->s_mount_opt, RESERVATION);
1645
1646         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1647                             NULL, 0))
1648                 goto failed_mount;
1649
1650         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1651                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1652
1653         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1654             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1655              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1656              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1657                 printk(KERN_WARNING
1658                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1659                        "running e2fsck is recommended\n");
1660         /*
1661          * Check feature flags regardless of the revision level, since we
1662          * previously didn't change the revision level when setting the flags,
1663          * so there is a chance incompat flags are set on a rev 0 filesystem.
1664          */
1665         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1666         if (features) {
1667                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1668                        "unsupported optional features (%x).\n",
1669                        sb->s_id, le32_to_cpu(features));
1670                 goto failed_mount;
1671         }
1672         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1673         if (!(sb->s_flags & MS_RDONLY) && features) {
1674                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1675                        "unsupported optional features (%x).\n",
1676                        sb->s_id, le32_to_cpu(features));
1677                 goto failed_mount;
1678         }
1679         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1680
1681         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1682             blocksize > EXT3_MAX_BLOCK_SIZE) {
1683                 printk(KERN_ERR
1684                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1685                        blocksize, sb->s_id);
1686                 goto failed_mount;
1687         }
1688
1689         hblock = bdev_hardsect_size(sb->s_bdev);
1690         if (sb->s_blocksize != blocksize) {
1691                 /*
1692                  * Make sure the blocksize for the filesystem is larger
1693                  * than the hardware sectorsize for the machine.
1694                  */
1695                 if (blocksize < hblock) {
1696                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1697                                "device blocksize %d.\n", blocksize, hblock);
1698                         goto failed_mount;
1699                 }
1700
1701                 brelse (bh);
1702                 if (!sb_set_blocksize(sb, blocksize)) {
1703                         printk(KERN_ERR "EXT3-fs: bad blocksize %d.\n",
1704                                 blocksize);
1705                         goto out_fail;
1706                 }
1707                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1708                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1709                 bh = sb_bread(sb, logic_sb_block);
1710                 if (!bh) {
1711                         printk(KERN_ERR
1712                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1713                         goto failed_mount;
1714                 }
1715                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1716                 sbi->s_es = es;
1717                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1718                         printk (KERN_ERR
1719                                 "EXT3-fs: Magic mismatch, very weird !\n");
1720                         goto failed_mount;
1721                 }
1722         }
1723
1724         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1725
1726         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1727                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1728                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1729         } else {
1730                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1731                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1732                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1733                     (!is_power_of_2(sbi->s_inode_size)) ||
1734                     (sbi->s_inode_size > blocksize)) {
1735                         printk (KERN_ERR
1736                                 "EXT3-fs: unsupported inode size: %d\n",
1737                                 sbi->s_inode_size);
1738                         goto failed_mount;
1739                 }
1740         }
1741         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1742                                    le32_to_cpu(es->s_log_frag_size);
1743         if (blocksize != sbi->s_frag_size) {
1744                 printk(KERN_ERR
1745                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1746                        sbi->s_frag_size, blocksize);
1747                 goto failed_mount;
1748         }
1749         sbi->s_frags_per_block = 1;
1750         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1751         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1752         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1753         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1754                 goto cantfind_ext3;
1755         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1756         if (sbi->s_inodes_per_block == 0)
1757                 goto cantfind_ext3;
1758         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1759                                         sbi->s_inodes_per_block;
1760         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1761         sbi->s_sbh = bh;
1762         sbi->s_mount_state = le16_to_cpu(es->s_state);
1763         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1764         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1765         for (i=0; i < 4; i++)
1766                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1767         sbi->s_def_hash_version = es->s_def_hash_version;
1768         i = le32_to_cpu(es->s_flags);
1769         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1770                 sbi->s_hash_unsigned = 3;
1771         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1772 #ifdef __CHAR_UNSIGNED__
1773                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1774                 sbi->s_hash_unsigned = 3;
1775 #else
1776                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1777 #endif
1778                 sb->s_dirt = 1;
1779         }
1780
1781         if (sbi->s_blocks_per_group > blocksize * 8) {
1782                 printk (KERN_ERR
1783                         "EXT3-fs: #blocks per group too big: %lu\n",
1784                         sbi->s_blocks_per_group);
1785                 goto failed_mount;
1786         }
1787         if (sbi->s_frags_per_group > blocksize * 8) {
1788                 printk (KERN_ERR
1789                         "EXT3-fs: #fragments per group too big: %lu\n",
1790                         sbi->s_frags_per_group);
1791                 goto failed_mount;
1792         }
1793         if (sbi->s_inodes_per_group > blocksize * 8) {
1794                 printk (KERN_ERR
1795                         "EXT3-fs: #inodes per group too big: %lu\n",
1796                         sbi->s_inodes_per_group);
1797                 goto failed_mount;
1798         }
1799
1800         if (le32_to_cpu(es->s_blocks_count) >
1801                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1802                 printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1803                         " too large to mount safely\n", sb->s_id);
1804                 if (sizeof(sector_t) < 8)
1805                         printk(KERN_WARNING "EXT3-fs: CONFIG_LBD not "
1806                                         "enabled\n");
1807                 goto failed_mount;
1808         }
1809
1810         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1811                 goto cantfind_ext3;
1812         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1813                                le32_to_cpu(es->s_first_data_block) - 1)
1814                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1815         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1816                    EXT3_DESC_PER_BLOCK(sb);
1817         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1818                                     GFP_KERNEL);
1819         if (sbi->s_group_desc == NULL) {
1820                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1821                 goto failed_mount;
1822         }
1823
1824         bgl_lock_init(&sbi->s_blockgroup_lock);
1825
1826         for (i = 0; i < db_count; i++) {
1827                 block = descriptor_loc(sb, logic_sb_block, i);
1828                 sbi->s_group_desc[i] = sb_bread(sb, block);
1829                 if (!sbi->s_group_desc[i]) {
1830                         printk (KERN_ERR "EXT3-fs: "
1831                                 "can't read group descriptor %d\n", i);
1832                         db_count = i;
1833                         goto failed_mount2;
1834                 }
1835         }
1836         if (!ext3_check_descriptors (sb)) {
1837                 printk(KERN_ERR "EXT3-fs: group descriptors corrupted!\n");
1838                 goto failed_mount2;
1839         }
1840         sbi->s_gdb_count = db_count;
1841         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1842         spin_lock_init(&sbi->s_next_gen_lock);
1843
1844         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1845                         ext3_count_free_blocks(sb));
1846         if (!err) {
1847                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1848                                 ext3_count_free_inodes(sb));
1849         }
1850         if (!err) {
1851                 err = percpu_counter_init(&sbi->s_dirs_counter,
1852                                 ext3_count_dirs(sb));
1853         }
1854         if (err) {
1855                 printk(KERN_ERR "EXT3-fs: insufficient memory\n");
1856                 goto failed_mount3;
1857         }
1858
1859         /* per fileystem reservation list head & lock */
1860         spin_lock_init(&sbi->s_rsv_window_lock);
1861         sbi->s_rsv_window_root = RB_ROOT;
1862         /* Add a single, static dummy reservation to the start of the
1863          * reservation window list --- it gives us a placeholder for
1864          * append-at-start-of-list which makes the allocation logic
1865          * _much_ simpler. */
1866         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1867         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1868         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1869         sbi->s_rsv_window_head.rsv_goal_size = 0;
1870         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1871
1872         /*
1873          * set up enough so that it can read an inode
1874          */
1875         sb->s_op = &ext3_sops;
1876         sb->s_export_op = &ext3_export_ops;
1877         sb->s_xattr = ext3_xattr_handlers;
1878 #ifdef CONFIG_QUOTA
1879         sb->s_qcop = &ext3_qctl_operations;
1880         sb->dq_op = &ext3_quota_operations;
1881 #endif
1882         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1883
1884         sb->s_root = NULL;
1885
1886         needs_recovery = (es->s_last_orphan != 0 ||
1887                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1888                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1889
1890         /*
1891          * The first inode we look at is the journal inode.  Don't try
1892          * root first: it may be modified in the journal!
1893          */
1894         if (!test_opt(sb, NOLOAD) &&
1895             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1896                 if (ext3_load_journal(sb, es, journal_devnum))
1897                         goto failed_mount3;
1898         } else if (journal_inum) {
1899                 if (ext3_create_journal(sb, es, journal_inum))
1900                         goto failed_mount3;
1901         } else {
1902                 if (!silent)
1903                         printk (KERN_ERR
1904                                 "ext3: No journal on filesystem on %s\n",
1905                                 sb->s_id);
1906                 goto failed_mount3;
1907         }
1908
1909         /* We have now updated the journal if required, so we can
1910          * validate the data journaling mode. */
1911         switch (test_opt(sb, DATA_FLAGS)) {
1912         case 0:
1913                 /* No mode set, assume a default based on the journal
1914                    capabilities: ORDERED_DATA if the journal can
1915                    cope, else JOURNAL_DATA */
1916                 if (journal_check_available_features
1917                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1918                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1919                 else
1920                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1921                 break;
1922
1923         case EXT3_MOUNT_ORDERED_DATA:
1924         case EXT3_MOUNT_WRITEBACK_DATA:
1925                 if (!journal_check_available_features
1926                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1927                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1928                                "requested data journaling mode\n");
1929                         goto failed_mount4;
1930                 }
1931         default:
1932                 break;
1933         }
1934
1935         if (test_opt(sb, NOBH)) {
1936                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1937                         printk(KERN_WARNING "EXT3-fs: Ignoring nobh option - "
1938                                 "its supported only with writeback mode\n");
1939                         clear_opt(sbi->s_mount_opt, NOBH);
1940                 }
1941         }
1942         /*
1943          * The journal_load will have done any necessary log recovery,
1944          * so we can safely mount the rest of the filesystem now.
1945          */
1946
1947         root = ext3_iget(sb, EXT3_ROOT_INO);
1948         if (IS_ERR(root)) {
1949                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1950                 ret = PTR_ERR(root);
1951                 goto failed_mount4;
1952         }
1953         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1954                 iput(root);
1955                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1956                 goto failed_mount4;
1957         }
1958         sb->s_root = d_alloc_root(root);
1959         if (!sb->s_root) {
1960                 printk(KERN_ERR "EXT3-fs: get root dentry failed\n");
1961                 iput(root);
1962                 ret = -ENOMEM;
1963                 goto failed_mount4;
1964         }
1965
1966         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1967         /*
1968          * akpm: core read_super() calls in here with the superblock locked.
1969          * That deadlocks, because orphan cleanup needs to lock the superblock
1970          * in numerous places.  Here we just pop the lock - it's relatively
1971          * harmless, because we are now ready to accept write_super() requests,
1972          * and aviro says that's the only reason for hanging onto the
1973          * superblock lock.
1974          */
1975         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1976         ext3_orphan_cleanup(sb, es);
1977         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1978         if (needs_recovery)
1979                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1980         ext3_mark_recovery_complete(sb, es);
1981         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1982                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1983                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1984                 "writeback");
1985
1986         lock_kernel();
1987         return 0;
1988
1989 cantfind_ext3:
1990         if (!silent)
1991                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1992                        sb->s_id);
1993         goto failed_mount;
1994
1995 failed_mount4:
1996         journal_destroy(sbi->s_journal);
1997 failed_mount3:
1998         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1999         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2000         percpu_counter_destroy(&sbi->s_dirs_counter);
2001 failed_mount2:
2002         for (i = 0; i < db_count; i++)
2003                 brelse(sbi->s_group_desc[i]);
2004         kfree(sbi->s_group_desc);
2005 failed_mount:
2006 #ifdef CONFIG_QUOTA
2007         for (i = 0; i < MAXQUOTAS; i++)
2008                 kfree(sbi->s_qf_names[i]);
2009 #endif
2010         ext3_blkdev_remove(sbi);
2011         brelse(bh);
2012 out_fail:
2013         sb->s_fs_info = NULL;
2014         kfree(sbi);
2015         lock_kernel();
2016         return ret;
2017 }
2018
2019 /*
2020  * Setup any per-fs journal parameters now.  We'll do this both on
2021  * initial mount, once the journal has been initialised but before we've
2022  * done any recovery; and again on any subsequent remount.
2023  */
2024 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2025 {
2026         struct ext3_sb_info *sbi = EXT3_SB(sb);
2027
2028         if (sbi->s_commit_interval)
2029                 journal->j_commit_interval = sbi->s_commit_interval;
2030         /* We could also set up an ext3-specific default for the commit
2031          * interval here, but for now we'll just fall back to the jbd
2032          * default. */
2033
2034         spin_lock(&journal->j_state_lock);
2035         if (test_opt(sb, BARRIER))
2036                 journal->j_flags |= JFS_BARRIER;
2037         else
2038                 journal->j_flags &= ~JFS_BARRIER;
2039         if (test_opt(sb, DATA_ERR_ABORT))
2040                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2041         else
2042                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2043         spin_unlock(&journal->j_state_lock);
2044 }
2045
2046 static journal_t *ext3_get_journal(struct super_block *sb,
2047                                    unsigned int journal_inum)
2048 {
2049         struct inode *journal_inode;
2050         journal_t *journal;
2051
2052         /* First, test for the existence of a valid inode on disk.  Bad
2053          * things happen if we iget() an unused inode, as the subsequent
2054          * iput() will try to delete it. */
2055
2056         journal_inode = ext3_iget(sb, journal_inum);
2057         if (IS_ERR(journal_inode)) {
2058                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
2059                 return NULL;
2060         }
2061         if (!journal_inode->i_nlink) {
2062                 make_bad_inode(journal_inode);
2063                 iput(journal_inode);
2064                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
2065                 return NULL;
2066         }
2067
2068         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2069                   journal_inode, journal_inode->i_size);
2070         if (!S_ISREG(journal_inode->i_mode)) {
2071                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
2072                 iput(journal_inode);
2073                 return NULL;
2074         }
2075
2076         journal = journal_init_inode(journal_inode);
2077         if (!journal) {
2078                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
2079                 iput(journal_inode);
2080                 return NULL;
2081         }
2082         journal->j_private = sb;
2083         ext3_init_journal_params(sb, journal);
2084         return journal;
2085 }
2086
2087 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2088                                        dev_t j_dev)
2089 {
2090         struct buffer_head * bh;
2091         journal_t *journal;
2092         ext3_fsblk_t start;
2093         ext3_fsblk_t len;
2094         int hblock, blocksize;
2095         ext3_fsblk_t sb_block;
2096         unsigned long offset;
2097         struct ext3_super_block * es;
2098         struct block_device *bdev;
2099
2100         bdev = ext3_blkdev_get(j_dev);
2101         if (bdev == NULL)
2102                 return NULL;
2103
2104         if (bd_claim(bdev, sb)) {
2105                 printk(KERN_ERR
2106                         "EXT3: failed to claim external journal device.\n");
2107                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2108                 return NULL;
2109         }
2110
2111         blocksize = sb->s_blocksize;
2112         hblock = bdev_hardsect_size(bdev);
2113         if (blocksize < hblock) {
2114                 printk(KERN_ERR
2115                         "EXT3-fs: blocksize too small for journal device.\n");
2116                 goto out_bdev;
2117         }
2118
2119         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2120         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2121         set_blocksize(bdev, blocksize);
2122         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2123                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
2124                        "external journal\n");
2125                 goto out_bdev;
2126         }
2127
2128         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2129         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2130             !(le32_to_cpu(es->s_feature_incompat) &
2131               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2132                 printk(KERN_ERR "EXT3-fs: external journal has "
2133                                         "bad superblock\n");
2134                 brelse(bh);
2135                 goto out_bdev;
2136         }
2137
2138         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2139                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
2140                 brelse(bh);
2141                 goto out_bdev;
2142         }
2143
2144         len = le32_to_cpu(es->s_blocks_count);
2145         start = sb_block + 1;
2146         brelse(bh);     /* we're done with the superblock */
2147
2148         journal = journal_init_dev(bdev, sb->s_bdev,
2149                                         start, len, blocksize);
2150         if (!journal) {
2151                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
2152                 goto out_bdev;
2153         }
2154         journal->j_private = sb;
2155         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2156         wait_on_buffer(journal->j_sb_buffer);
2157         if (!buffer_uptodate(journal->j_sb_buffer)) {
2158                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
2159                 goto out_journal;
2160         }
2161         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2162                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
2163                                         "user (unsupported) - %d\n",
2164                         be32_to_cpu(journal->j_superblock->s_nr_users));
2165                 goto out_journal;
2166         }
2167         EXT3_SB(sb)->journal_bdev = bdev;
2168         ext3_init_journal_params(sb, journal);
2169         return journal;
2170 out_journal:
2171         journal_destroy(journal);
2172 out_bdev:
2173         ext3_blkdev_put(bdev);
2174         return NULL;
2175 }
2176
2177 static int ext3_load_journal(struct super_block *sb,
2178                              struct ext3_super_block *es,
2179                              unsigned long journal_devnum)
2180 {
2181         journal_t *journal;
2182         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2183         dev_t journal_dev;
2184         int err = 0;
2185         int really_read_only;
2186
2187         if (journal_devnum &&
2188             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2189                 printk(KERN_INFO "EXT3-fs: external journal device major/minor "
2190                         "numbers have changed\n");
2191                 journal_dev = new_decode_dev(journal_devnum);
2192         } else
2193                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2194
2195         really_read_only = bdev_read_only(sb->s_bdev);
2196
2197         /*
2198          * Are we loading a blank journal or performing recovery after a
2199          * crash?  For recovery, we need to check in advance whether we
2200          * can get read-write access to the device.
2201          */
2202
2203         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2204                 if (sb->s_flags & MS_RDONLY) {
2205                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
2206                                         "required on readonly filesystem.\n");
2207                         if (really_read_only) {
2208                                 printk(KERN_ERR "EXT3-fs: write access "
2209                                         "unavailable, cannot proceed.\n");
2210                                 return -EROFS;
2211                         }
2212                         printk (KERN_INFO "EXT3-fs: write access will "
2213                                         "be enabled during recovery.\n");
2214                 }
2215         }
2216
2217         if (journal_inum && journal_dev) {
2218                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
2219                        "and inode journals!\n");
2220                 return -EINVAL;
2221         }
2222
2223         if (journal_inum) {
2224                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2225                         return -EINVAL;
2226         } else {
2227                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2228                         return -EINVAL;
2229         }
2230
2231         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2232                 err = journal_update_format(journal);
2233                 if (err)  {
2234                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
2235                         journal_destroy(journal);
2236                         return err;
2237                 }
2238         }
2239
2240         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2241                 err = journal_wipe(journal, !really_read_only);
2242         if (!err)
2243                 err = journal_load(journal);
2244
2245         if (err) {
2246                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
2247                 journal_destroy(journal);
2248                 return err;
2249         }
2250
2251         EXT3_SB(sb)->s_journal = journal;
2252         ext3_clear_journal_err(sb, es);
2253
2254         if (journal_devnum &&
2255             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2256                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2257                 sb->s_dirt = 1;
2258
2259                 /* Make sure we flush the recovery flag to disk. */
2260                 ext3_commit_super(sb, es, 1);
2261         }
2262
2263         return 0;
2264 }
2265
2266 static int ext3_create_journal(struct super_block * sb,
2267                                struct ext3_super_block * es,
2268                                unsigned int journal_inum)
2269 {
2270         journal_t *journal;
2271         int err;
2272
2273         if (sb->s_flags & MS_RDONLY) {
2274                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
2275                                 "create journal.\n");
2276                 return -EROFS;
2277         }
2278
2279         journal = ext3_get_journal(sb, journal_inum);
2280         if (!journal)
2281                 return -EINVAL;
2282
2283         printk(KERN_INFO "EXT3-fs: creating new journal on inode %u\n",
2284                journal_inum);
2285
2286         err = journal_create(journal);
2287         if (err) {
2288                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
2289                 journal_destroy(journal);
2290                 return -EIO;
2291         }
2292
2293         EXT3_SB(sb)->s_journal = journal;
2294
2295         ext3_update_dynamic_rev(sb);
2296         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2297         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2298
2299         es->s_journal_inum = cpu_to_le32(journal_inum);
2300         sb->s_dirt = 1;
2301
2302         /* Make sure we flush the recovery flag to disk. */
2303         ext3_commit_super(sb, es, 1);
2304
2305         return 0;
2306 }
2307
2308 static void ext3_commit_super (struct super_block * sb,
2309                                struct ext3_super_block * es,
2310                                int sync)
2311 {
2312         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2313
2314         if (!sbh)
2315                 return;
2316         es->s_wtime = cpu_to_le32(get_seconds());
2317         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2318         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2319         BUFFER_TRACE(sbh, "marking dirty");
2320         mark_buffer_dirty(sbh);
2321         if (sync)
2322                 sync_dirty_buffer(sbh);
2323 }
2324
2325
2326 /*
2327  * Have we just finished recovery?  If so, and if we are mounting (or
2328  * remounting) the filesystem readonly, then we will end up with a
2329  * consistent fs on disk.  Record that fact.
2330  */
2331 static void ext3_mark_recovery_complete(struct super_block * sb,
2332                                         struct ext3_super_block * es)
2333 {
2334         journal_t *journal = EXT3_SB(sb)->s_journal;
2335
2336         journal_lock_updates(journal);
2337         if (journal_flush(journal) < 0)
2338                 goto out;
2339
2340         lock_super(sb);
2341         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2342             sb->s_flags & MS_RDONLY) {
2343                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2344                 sb->s_dirt = 0;
2345                 ext3_commit_super(sb, es, 1);
2346         }
2347         unlock_super(sb);
2348
2349 out:
2350         journal_unlock_updates(journal);
2351 }
2352
2353 /*
2354  * If we are mounting (or read-write remounting) a filesystem whose journal
2355  * has recorded an error from a previous lifetime, move that error to the
2356  * main filesystem now.
2357  */
2358 static void ext3_clear_journal_err(struct super_block * sb,
2359                                    struct ext3_super_block * es)
2360 {
2361         journal_t *journal;
2362         int j_errno;
2363         const char *errstr;
2364
2365         journal = EXT3_SB(sb)->s_journal;
2366
2367         /*
2368          * Now check for any error status which may have been recorded in the
2369          * journal by a prior ext3_error() or ext3_abort()
2370          */
2371
2372         j_errno = journal_errno(journal);
2373         if (j_errno) {
2374                 char nbuf[16];
2375
2376                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2377                 ext3_warning(sb, __func__, "Filesystem error recorded "
2378                              "from previous mount: %s", errstr);
2379                 ext3_warning(sb, __func__, "Marking fs in need of "
2380                              "filesystem check.");
2381
2382                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2383                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2384                 ext3_commit_super (sb, es, 1);
2385
2386                 journal_clear_err(journal);
2387         }
2388 }
2389
2390 /*
2391  * Force the running and committing transactions to commit,
2392  * and wait on the commit.
2393  */
2394 int ext3_force_commit(struct super_block *sb)
2395 {
2396         journal_t *journal;
2397         int ret;
2398
2399         if (sb->s_flags & MS_RDONLY)
2400                 return 0;
2401
2402         journal = EXT3_SB(sb)->s_journal;
2403         sb->s_dirt = 0;
2404         ret = ext3_journal_force_commit(journal);
2405         return ret;
2406 }
2407
2408 /*
2409  * Ext3 always journals updates to the superblock itself, so we don't
2410  * have to propagate any other updates to the superblock on disk at this
2411  * point.  (We can probably nuke this function altogether, and remove
2412  * any mention to sb->s_dirt in all of fs/ext3; eventual cleanup...)
2413  */
2414 static void ext3_write_super (struct super_block * sb)
2415 {
2416         if (mutex_trylock(&sb->s_lock) != 0)
2417                 BUG();
2418         sb->s_dirt = 0;
2419 }
2420
2421 static int ext3_sync_fs(struct super_block *sb, int wait)
2422 {
2423         sb->s_dirt = 0;
2424         if (wait)
2425                 ext3_force_commit(sb);
2426         else
2427                 journal_start_commit(EXT3_SB(sb)->s_journal, NULL);
2428
2429         return 0;
2430 }
2431
2432 /*
2433  * LVM calls this function before a (read-only) snapshot is created.  This
2434  * gives us a chance to flush the journal completely and mark the fs clean.
2435  */
2436 static void ext3_write_super_lockfs(struct super_block *sb)
2437 {
2438         sb->s_dirt = 0;
2439
2440         if (!(sb->s_flags & MS_RDONLY)) {
2441                 journal_t *journal = EXT3_SB(sb)->s_journal;
2442
2443                 /* Now we set up the journal barrier. */
2444                 journal_lock_updates(journal);
2445
2446                 /*
2447                  * We don't want to clear needs_recovery flag when we failed
2448                  * to flush the journal.
2449                  */
2450                 if (journal_flush(journal) < 0)
2451                         return;
2452
2453                 /* Journal blocked and flushed, clear needs_recovery flag. */
2454                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2455                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2456         }
2457 }
2458
2459 /*
2460  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2461  * flag here, even though the filesystem is not technically dirty yet.
2462  */
2463 static void ext3_unlockfs(struct super_block *sb)
2464 {
2465         if (!(sb->s_flags & MS_RDONLY)) {
2466                 lock_super(sb);
2467                 /* Reser the needs_recovery flag before the fs is unlocked. */
2468                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2469                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2470                 unlock_super(sb);
2471                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2472         }
2473 }
2474
2475 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2476 {
2477         struct ext3_super_block * es;
2478         struct ext3_sb_info *sbi = EXT3_SB(sb);
2479         ext3_fsblk_t n_blocks_count = 0;
2480         unsigned long old_sb_flags;
2481         struct ext3_mount_options old_opts;
2482         int err;
2483 #ifdef CONFIG_QUOTA
2484         int i;
2485 #endif
2486
2487         /* Store the original options */
2488         old_sb_flags = sb->s_flags;
2489         old_opts.s_mount_opt = sbi->s_mount_opt;
2490         old_opts.s_resuid = sbi->s_resuid;
2491         old_opts.s_resgid = sbi->s_resgid;
2492         old_opts.s_commit_interval = sbi->s_commit_interval;
2493 #ifdef CONFIG_QUOTA
2494         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2495         for (i = 0; i < MAXQUOTAS; i++)
2496                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2497 #endif
2498
2499         /*
2500          * Allow the "check" option to be passed as a remount option.
2501          */
2502         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2503                 err = -EINVAL;
2504                 goto restore_opts;
2505         }
2506
2507         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2508                 ext3_abort(sb, __func__, "Abort forced by user");
2509
2510         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2511                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2512
2513         es = sbi->s_es;
2514
2515         ext3_init_journal_params(sb, sbi->s_journal);
2516
2517         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2518                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2519                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2520                         err = -EROFS;
2521                         goto restore_opts;
2522                 }
2523
2524                 if (*flags & MS_RDONLY) {
2525                         /*
2526                          * First of all, the unconditional stuff we have to do
2527                          * to disable replay of the journal when we next remount
2528                          */
2529                         sb->s_flags |= MS_RDONLY;
2530
2531                         /*
2532                          * OK, test if we are remounting a valid rw partition
2533                          * readonly, and if so set the rdonly flag and then
2534                          * mark the partition as valid again.
2535                          */
2536                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2537                             (sbi->s_mount_state & EXT3_VALID_FS))
2538                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2539
2540                         /*
2541                          * We have to unlock super so that we can wait for
2542                          * transactions.
2543                          */
2544                         unlock_super(sb);
2545                         ext3_mark_recovery_complete(sb, es);
2546                         lock_super(sb);
2547                 } else {
2548                         __le32 ret;
2549                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2550                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2551                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2552                                        "remount RDWR because of unsupported "
2553                                        "optional features (%x).\n",
2554                                        sb->s_id, le32_to_cpu(ret));
2555                                 err = -EROFS;
2556                                 goto restore_opts;
2557                         }
2558
2559                         /*
2560                          * If we have an unprocessed orphan list hanging
2561                          * around from a previously readonly bdev mount,
2562                          * require a full umount/remount for now.
2563                          */
2564                         if (es->s_last_orphan) {
2565                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2566                                        "remount RDWR because of unprocessed "
2567                                        "orphan inode list.  Please "
2568                                        "umount/remount instead.\n",
2569                                        sb->s_id);
2570                                 err = -EINVAL;
2571                                 goto restore_opts;
2572                         }
2573
2574                         /*
2575                          * Mounting a RDONLY partition read-write, so reread
2576                          * and store the current valid flag.  (It may have
2577                          * been changed by e2fsck since we originally mounted
2578                          * the partition.)
2579                          */
2580                         ext3_clear_journal_err(sb, es);
2581                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2582                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2583                                 goto restore_opts;
2584                         if (!ext3_setup_super (sb, es, 0))
2585                                 sb->s_flags &= ~MS_RDONLY;
2586                 }
2587         }
2588 #ifdef CONFIG_QUOTA
2589         /* Release old quota file names */
2590         for (i = 0; i < MAXQUOTAS; i++)
2591                 if (old_opts.s_qf_names[i] &&
2592                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2593                         kfree(old_opts.s_qf_names[i]);
2594 #endif
2595         return 0;
2596 restore_opts:
2597         sb->s_flags = old_sb_flags;
2598         sbi->s_mount_opt = old_opts.s_mount_opt;
2599         sbi->s_resuid = old_opts.s_resuid;
2600         sbi->s_resgid = old_opts.s_resgid;
2601         sbi->s_commit_interval = old_opts.s_commit_interval;
2602 #ifdef CONFIG_QUOTA
2603         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2604         for (i = 0; i < MAXQUOTAS; i++) {
2605                 if (sbi->s_qf_names[i] &&
2606                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2607                         kfree(sbi->s_qf_names[i]);
2608                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2609         }
2610 #endif
2611         return err;
2612 }
2613
2614 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2615 {
2616         struct super_block *sb = dentry->d_sb;
2617         struct ext3_sb_info *sbi = EXT3_SB(sb);
2618         struct ext3_super_block *es = sbi->s_es;
2619         u64 fsid;
2620
2621         if (test_opt(sb, MINIX_DF)) {
2622                 sbi->s_overhead_last = 0;
2623         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2624                 unsigned long ngroups = sbi->s_groups_count, i;
2625                 ext3_fsblk_t overhead = 0;
2626                 smp_rmb();
2627
2628                 /*
2629                  * Compute the overhead (FS structures).  This is constant
2630                  * for a given filesystem unless the number of block groups
2631                  * changes so we cache the previous value until it does.
2632                  */
2633
2634                 /*
2635                  * All of the blocks before first_data_block are
2636                  * overhead
2637                  */
2638                 overhead = le32_to_cpu(es->s_first_data_block);
2639
2640                 /*
2641                  * Add the overhead attributed to the superblock and
2642                  * block group descriptors.  If the sparse superblocks
2643                  * feature is turned on, then not all groups have this.
2644                  */
2645                 for (i = 0; i < ngroups; i++) {
2646                         overhead += ext3_bg_has_super(sb, i) +
2647                                 ext3_bg_num_gdb(sb, i);
2648                         cond_resched();
2649                 }
2650
2651                 /*
2652                  * Every block group has an inode bitmap, a block
2653                  * bitmap, and an inode table.
2654                  */
2655                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2656                 sbi->s_overhead_last = overhead;
2657                 smp_wmb();
2658                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2659         }
2660
2661         buf->f_type = EXT3_SUPER_MAGIC;
2662         buf->f_bsize = sb->s_blocksize;
2663         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2664         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2665         es->s_free_blocks_count = cpu_to_le32(buf->f_bfree);
2666         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2667         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2668                 buf->f_bavail = 0;
2669         buf->f_files = le32_to_cpu(es->s_inodes_count);
2670         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2671         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
2672         buf->f_namelen = EXT3_NAME_LEN;
2673         fsid = le64_to_cpup((void *)es->s_uuid) ^
2674                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2675         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2676         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2677         return 0;
2678 }
2679
2680 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2681  * is locked for write. Otherwise the are possible deadlocks:
2682  * Process 1                         Process 2
2683  * ext3_create()                     quota_sync()
2684  *   journal_start()                   write_dquot()
2685  *   DQUOT_INIT()                        down(dqio_mutex)
2686  *     down(dqio_mutex)                    journal_start()
2687  *
2688  */
2689
2690 #ifdef CONFIG_QUOTA
2691
2692 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2693 {
2694         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2695 }
2696
2697 static int ext3_dquot_initialize(struct inode *inode, int type)
2698 {
2699         handle_t *handle;
2700         int ret, err;
2701
2702         /* We may create quota structure so we need to reserve enough blocks */
2703         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS(inode->i_sb));
2704         if (IS_ERR(handle))
2705                 return PTR_ERR(handle);
2706         ret = dquot_initialize(inode, type);
2707         err = ext3_journal_stop(handle);
2708         if (!ret)
2709                 ret = err;
2710         return ret;
2711 }
2712
2713 static int ext3_dquot_drop(struct inode *inode)
2714 {
2715         handle_t *handle;
2716         int ret, err;
2717
2718         /* We may delete quota structure so we need to reserve enough blocks */
2719         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_DEL_BLOCKS(inode->i_sb));
2720         if (IS_ERR(handle)) {
2721                 /*
2722                  * We call dquot_drop() anyway to at least release references
2723                  * to quota structures so that umount does not hang.
2724                  */
2725                 dquot_drop(inode);
2726                 return PTR_ERR(handle);
2727         }
2728         ret = dquot_drop(inode);
2729         err = ext3_journal_stop(handle);
2730         if (!ret)
2731                 ret = err;
2732         return ret;
2733 }
2734
2735 static int ext3_write_dquot(struct dquot *dquot)
2736 {
2737         int ret, err;
2738         handle_t *handle;
2739         struct inode *inode;
2740
2741         inode = dquot_to_inode(dquot);
2742         handle = ext3_journal_start(inode,
2743                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2744         if (IS_ERR(handle))
2745                 return PTR_ERR(handle);
2746         ret = dquot_commit(dquot);
2747         err = ext3_journal_stop(handle);
2748         if (!ret)
2749                 ret = err;
2750         return ret;
2751 }
2752
2753 static int ext3_acquire_dquot(struct dquot *dquot)
2754 {
2755         int ret, err;
2756         handle_t *handle;
2757
2758         handle = ext3_journal_start(dquot_to_inode(dquot),
2759                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2760         if (IS_ERR(handle))
2761                 return PTR_ERR(handle);
2762         ret = dquot_acquire(dquot);
2763         err = ext3_journal_stop(handle);
2764         if (!ret)
2765                 ret = err;
2766         return ret;
2767 }
2768
2769 static int ext3_release_dquot(struct dquot *dquot)
2770 {
2771         int ret, err;
2772         handle_t *handle;
2773
2774         handle = ext3_journal_start(dquot_to_inode(dquot),
2775                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2776         if (IS_ERR(handle)) {
2777                 /* Release dquot anyway to avoid endless cycle in dqput() */
2778                 dquot_release(dquot);
2779                 return PTR_ERR(handle);
2780         }
2781         ret = dquot_release(dquot);
2782         err = ext3_journal_stop(handle);
2783         if (!ret)
2784                 ret = err;
2785         return ret;
2786 }
2787
2788 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2789 {
2790         /* Are we journaling quotas? */
2791         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2792             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2793                 dquot_mark_dquot_dirty(dquot);
2794                 return ext3_write_dquot(dquot);
2795         } else {
2796                 return dquot_mark_dquot_dirty(dquot);
2797         }
2798 }
2799
2800 static int ext3_write_info(struct super_block *sb, int type)
2801 {
2802         int ret, err;
2803         handle_t *handle;
2804
2805         /* Data block + inode block */
2806         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2807         if (IS_ERR(handle))
2808                 return PTR_ERR(handle);
2809         ret = dquot_commit_info(sb, type);
2810         err = ext3_journal_stop(handle);
2811         if (!ret)
2812                 ret = err;
2813         return ret;
2814 }
2815
2816 /*
2817  * Turn on quotas during mount time - we need to find
2818  * the quota file and such...
2819  */
2820 static int ext3_quota_on_mount(struct super_block *sb, int type)
2821 {
2822         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2823                         EXT3_SB(sb)->s_jquota_fmt, type);
2824 }
2825
2826 /*
2827  * Standard function to be called on quota_on
2828  */
2829 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2830                          char *name, int remount)
2831 {
2832         int err;
2833         struct path path;
2834
2835         if (!test_opt(sb, QUOTA))
2836                 return -EINVAL;
2837         /* When remounting, no checks are needed and in fact, name is NULL */
2838         if (remount)
2839                 return vfs_quota_on(sb, type, format_id, name, remount);
2840
2841         err = kern_path(name, LOOKUP_FOLLOW, &path);
2842         if (err)
2843                 return err;
2844
2845         /* Quotafile not on the same filesystem? */
2846         if (path.mnt->mnt_sb != sb) {
2847                 path_put(&path);
2848                 return -EXDEV;
2849         }
2850         /* Journaling quota? */
2851         if (EXT3_SB(sb)->s_qf_names[type]) {
2852                 /* Quotafile not of fs root? */
2853                 if (path.dentry->d_parent != sb->s_root)
2854                         printk(KERN_WARNING
2855                                 "EXT3-fs: Quota file not on filesystem root. "
2856                                 "Journaled quota will not work.\n");
2857         }
2858
2859         /*
2860          * When we journal data on quota file, we have to flush journal to see
2861          * all updates to the file when we bypass pagecache...
2862          */
2863         if (ext3_should_journal_data(path.dentry->d_inode)) {
2864                 /*
2865                  * We don't need to lock updates but journal_flush() could
2866                  * otherwise be livelocked...
2867                  */
2868                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2869                 err = journal_flush(EXT3_SB(sb)->s_journal);
2870                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2871                 if (err) {
2872                         path_put(&path);
2873                         return err;
2874                 }
2875         }
2876
2877         err = vfs_quota_on_path(sb, type, format_id, &path);
2878         path_put(&path);
2879         return err;
2880 }
2881
2882 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2883  * acquiring the locks... As quota files are never truncated and quota code
2884  * itself serializes the operations (and noone else should touch the files)
2885  * we don't have to be afraid of races */
2886 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2887                                size_t len, loff_t off)
2888 {
2889         struct inode *inode = sb_dqopt(sb)->files[type];
2890         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2891         int err = 0;
2892         int offset = off & (sb->s_blocksize - 1);
2893         int tocopy;
2894         size_t toread;
2895         struct buffer_head *bh;
2896         loff_t i_size = i_size_read(inode);
2897
2898         if (off > i_size)
2899                 return 0;
2900         if (off+len > i_size)
2901                 len = i_size-off;
2902         toread = len;
2903         while (toread > 0) {
2904                 tocopy = sb->s_blocksize - offset < toread ?
2905                                 sb->s_blocksize - offset : toread;
2906                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2907                 if (err)
2908                         return err;
2909                 if (!bh)        /* A hole? */
2910                         memset(data, 0, tocopy);
2911                 else
2912                         memcpy(data, bh->b_data+offset, tocopy);
2913                 brelse(bh);
2914                 offset = 0;
2915                 toread -= tocopy;
2916                 data += tocopy;
2917                 blk++;
2918         }
2919         return len;
2920 }
2921
2922 /* Write to quotafile (we know the transaction is already started and has
2923  * enough credits) */
2924 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2925                                 const char *data, size_t len, loff_t off)
2926 {
2927         struct inode *inode = sb_dqopt(sb)->files[type];
2928         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2929         int err = 0;
2930         int offset = off & (sb->s_blocksize - 1);
2931         int tocopy;
2932         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2933         size_t towrite = len;
2934         struct buffer_head *bh;
2935         handle_t *handle = journal_current_handle();
2936
2937         if (!handle) {
2938                 printk(KERN_WARNING "EXT3-fs: Quota write (off=%Lu, len=%Lu)"
2939                         " cancelled because transaction is not started.\n",
2940                         (unsigned long long)off, (unsigned long long)len);
2941                 return -EIO;
2942         }
2943         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2944         while (towrite > 0) {
2945                 tocopy = sb->s_blocksize - offset < towrite ?
2946                                 sb->s_blocksize - offset : towrite;
2947                 bh = ext3_bread(handle, inode, blk, 1, &err);
2948                 if (!bh)
2949                         goto out;
2950                 if (journal_quota) {
2951                         err = ext3_journal_get_write_access(handle, bh);
2952                         if (err) {
2953                                 brelse(bh);
2954                                 goto out;
2955                         }
2956                 }
2957                 lock_buffer(bh);
2958                 memcpy(bh->b_data+offset, data, tocopy);
2959                 flush_dcache_page(bh->b_page);
2960                 unlock_buffer(bh);
2961                 if (journal_quota)
2962                         err = ext3_journal_dirty_metadata(handle, bh);
2963                 else {
2964                         /* Always do at least ordered writes for quotas */
2965                         err = ext3_journal_dirty_data(handle, bh);
2966                         mark_buffer_dirty(bh);
2967                 }
2968                 brelse(bh);
2969                 if (err)
2970                         goto out;
2971                 offset = 0;
2972                 towrite -= tocopy;
2973                 data += tocopy;
2974                 blk++;
2975         }
2976 out:
2977         if (len == towrite) {
2978                 mutex_unlock(&inode->i_mutex);
2979                 return err;
2980         }
2981         if (inode->i_size < off+len-towrite) {
2982                 i_size_write(inode, off+len-towrite);
2983                 EXT3_I(inode)->i_disksize = inode->i_size;
2984         }
2985         inode->i_version++;
2986         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2987         ext3_mark_inode_dirty(handle, inode);
2988         mutex_unlock(&inode->i_mutex);
2989         return len - towrite;
2990 }
2991
2992 #endif
2993
2994 static int ext3_get_sb(struct file_system_type *fs_type,
2995         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2996 {
2997         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
2998 }
2999
3000 static struct file_system_type ext3_fs_type = {
3001         .owner          = THIS_MODULE,
3002         .name           = "ext3",
3003         .get_sb         = ext3_get_sb,
3004         .kill_sb        = kill_block_super,
3005         .fs_flags       = FS_REQUIRES_DEV,
3006 };
3007
3008 static int __init init_ext3_fs(void)
3009 {
3010         int err = init_ext3_xattr();
3011         if (err)
3012                 return err;
3013         err = init_inodecache();
3014         if (err)
3015                 goto out1;
3016         err = register_filesystem(&ext3_fs_type);
3017         if (err)
3018                 goto out;
3019         return 0;
3020 out:
3021         destroy_inodecache();
3022 out1:
3023         exit_ext3_xattr();
3024         return err;
3025 }
3026
3027 static void __exit exit_ext3_fs(void)
3028 {
3029         unregister_filesystem(&ext3_fs_type);
3030         destroy_inodecache();
3031         exit_ext3_xattr();
3032 }
3033
3034 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3035 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3036 MODULE_LICENSE("GPL");
3037 module_init(init_ext3_fs)
3038 module_exit(exit_ext3_fs)