ext3: Return error code from generic_check_addressable
[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/buffer_head.h>
31 #include <linux/exportfs.h>
32 #include <linux/vfs.h>
33 #include <linux/random.h>
34 #include <linux/mount.h>
35 #include <linux/namei.h>
36 #include <linux/quotaops.h>
37 #include <linux/seq_file.h>
38 #include <linux/log2.h>
39
40 #include <asm/uaccess.h>
41
42 #include "xattr.h"
43 #include "acl.h"
44 #include "namei.h"
45
46 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
47   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
48 #else
49   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
50 #endif
51
52 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
53                              unsigned long journal_devnum);
54 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
55                                unsigned int);
56 static int ext3_commit_super(struct super_block *sb,
57                                struct ext3_super_block *es,
58                                int sync);
59 static void ext3_mark_recovery_complete(struct super_block * sb,
60                                         struct ext3_super_block * es);
61 static void ext3_clear_journal_err(struct super_block * sb,
62                                    struct ext3_super_block * es);
63 static int ext3_sync_fs(struct super_block *sb, int wait);
64 static const char *ext3_decode_error(struct super_block * sb, int errno,
65                                      char nbuf[16]);
66 static int ext3_remount (struct super_block * sb, int * flags, char * data);
67 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
68 static int ext3_unfreeze(struct super_block *sb);
69 static int ext3_freeze(struct super_block *sb);
70
71 /*
72  * Wrappers for journal_start/end.
73  *
74  * The only special thing we need to do here is to make sure that all
75  * journal_end calls result in the superblock being marked dirty, so
76  * that sync() will call the filesystem's write_super callback if
77  * appropriate.
78  */
79 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
80 {
81         journal_t *journal;
82
83         if (sb->s_flags & MS_RDONLY)
84                 return ERR_PTR(-EROFS);
85
86         /* Special case here: if the journal has aborted behind our
87          * backs (eg. EIO in the commit thread), then we still need to
88          * take the FS itself readonly cleanly. */
89         journal = EXT3_SB(sb)->s_journal;
90         if (is_journal_aborted(journal)) {
91                 ext3_abort(sb, __func__,
92                            "Detected aborted journal");
93                 return ERR_PTR(-EROFS);
94         }
95
96         return journal_start(journal, nblocks);
97 }
98
99 /*
100  * The only special thing we need to do here is to make sure that all
101  * journal_stop calls result in the superblock being marked dirty, so
102  * that sync() will call the filesystem's write_super callback if
103  * appropriate.
104  */
105 int __ext3_journal_stop(const char *where, handle_t *handle)
106 {
107         struct super_block *sb;
108         int err;
109         int rc;
110
111         sb = handle->h_transaction->t_journal->j_private;
112         err = handle->h_err;
113         rc = journal_stop(handle);
114
115         if (!err)
116                 err = rc;
117         if (err)
118                 __ext3_std_error(sb, where, err);
119         return err;
120 }
121
122 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
123                 struct buffer_head *bh, handle_t *handle, int err)
124 {
125         char nbuf[16];
126         const char *errstr = ext3_decode_error(NULL, err, nbuf);
127
128         if (bh)
129                 BUFFER_TRACE(bh, "abort");
130
131         if (!handle->h_err)
132                 handle->h_err = err;
133
134         if (is_handle_aborted(handle))
135                 return;
136
137         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
138                 caller, errstr, err_fn);
139
140         journal_abort_handle(handle);
141 }
142
143 void ext3_msg(struct super_block *sb, const char *prefix,
144                 const char *fmt, ...)
145 {
146         struct va_format vaf;
147         va_list args;
148
149         va_start(args, fmt);
150
151         vaf.fmt = fmt;
152         vaf.va = &args;
153
154         printk("%sEXT3-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
155
156         va_end(args);
157 }
158
159 /* Deal with the reporting of failure conditions on a filesystem such as
160  * inconsistencies detected or read IO failures.
161  *
162  * On ext2, we can store the error state of the filesystem in the
163  * superblock.  That is not possible on ext3, because we may have other
164  * write ordering constraints on the superblock which prevent us from
165  * writing it out straight away; and given that the journal is about to
166  * be aborted, we can't rely on the current, or future, transactions to
167  * write out the superblock safely.
168  *
169  * We'll just use the journal_abort() error code to record an error in
170  * the journal instead.  On recovery, the journal will complain about
171  * that error until we've noted it down and cleared it.
172  */
173
174 static void ext3_handle_error(struct super_block *sb)
175 {
176         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
177
178         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
179         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
180
181         if (sb->s_flags & MS_RDONLY)
182                 return;
183
184         if (!test_opt (sb, ERRORS_CONT)) {
185                 journal_t *journal = EXT3_SB(sb)->s_journal;
186
187                 set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
188                 if (journal)
189                         journal_abort(journal, -EIO);
190         }
191         if (test_opt (sb, ERRORS_RO)) {
192                 ext3_msg(sb, KERN_CRIT,
193                         "error: remounting filesystem read-only");
194                 sb->s_flags |= MS_RDONLY;
195         }
196         ext3_commit_super(sb, es, 1);
197         if (test_opt(sb, ERRORS_PANIC))
198                 panic("EXT3-fs (%s): panic forced after error\n",
199                         sb->s_id);
200 }
201
202 void ext3_error(struct super_block *sb, const char *function,
203                 const char *fmt, ...)
204 {
205         struct va_format vaf;
206         va_list args;
207
208         va_start(args, fmt);
209
210         vaf.fmt = fmt;
211         vaf.va = &args;
212
213         printk(KERN_CRIT "EXT3-fs error (device %s): %s: %pV\n",
214                sb->s_id, function, &vaf);
215
216         va_end(args);
217
218         ext3_handle_error(sb);
219 }
220
221 static const char *ext3_decode_error(struct super_block * sb, int errno,
222                                      char nbuf[16])
223 {
224         char *errstr = NULL;
225
226         switch (errno) {
227         case -EIO:
228                 errstr = "IO failure";
229                 break;
230         case -ENOMEM:
231                 errstr = "Out of memory";
232                 break;
233         case -EROFS:
234                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
235                         errstr = "Journal has aborted";
236                 else
237                         errstr = "Readonly filesystem";
238                 break;
239         default:
240                 /* If the caller passed in an extra buffer for unknown
241                  * errors, textualise them now.  Else we just return
242                  * NULL. */
243                 if (nbuf) {
244                         /* Check for truncated error codes... */
245                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
246                                 errstr = nbuf;
247                 }
248                 break;
249         }
250
251         return errstr;
252 }
253
254 /* __ext3_std_error decodes expected errors from journaling functions
255  * automatically and invokes the appropriate error response.  */
256
257 void __ext3_std_error (struct super_block * sb, const char * function,
258                        int errno)
259 {
260         char nbuf[16];
261         const char *errstr;
262
263         /* Special case: if the error is EROFS, and we're not already
264          * inside a transaction, then there's really no point in logging
265          * an error. */
266         if (errno == -EROFS && journal_current_handle() == NULL &&
267             (sb->s_flags & MS_RDONLY))
268                 return;
269
270         errstr = ext3_decode_error(sb, errno, nbuf);
271         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
272
273         ext3_handle_error(sb);
274 }
275
276 /*
277  * ext3_abort is a much stronger failure handler than ext3_error.  The
278  * abort function may be used to deal with unrecoverable failures such
279  * as journal IO errors or ENOMEM at a critical moment in log management.
280  *
281  * We unconditionally force the filesystem into an ABORT|READONLY state,
282  * unless the error response on the fs has been set to panic in which
283  * case we take the easy way out and panic immediately.
284  */
285
286 void ext3_abort(struct super_block *sb, const char *function,
287                  const char *fmt, ...)
288 {
289         struct va_format vaf;
290         va_list args;
291
292         va_start(args, fmt);
293
294         vaf.fmt = fmt;
295         vaf.va = &args;
296
297         printk(KERN_CRIT "EXT3-fs (%s): error: %s: %pV\n",
298                sb->s_id, function, &vaf);
299
300         va_end(args);
301
302         if (test_opt(sb, ERRORS_PANIC))
303                 panic("EXT3-fs: panic from previous error\n");
304
305         if (sb->s_flags & MS_RDONLY)
306                 return;
307
308         ext3_msg(sb, KERN_CRIT,
309                 "error: remounting filesystem read-only");
310         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
311         sb->s_flags |= MS_RDONLY;
312         set_opt(EXT3_SB(sb)->s_mount_opt, ABORT);
313         if (EXT3_SB(sb)->s_journal)
314                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
315 }
316
317 void ext3_warning(struct super_block *sb, const char *function,
318                   const char *fmt, ...)
319 {
320         struct va_format vaf;
321         va_list args;
322
323         va_start(args, fmt);
324
325         vaf.fmt = fmt;
326         vaf.va = &args;
327
328         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: %pV\n",
329                sb->s_id, function, &vaf);
330
331         va_end(args);
332 }
333
334 void ext3_update_dynamic_rev(struct super_block *sb)
335 {
336         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
337
338         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
339                 return;
340
341         ext3_msg(sb, KERN_WARNING,
342                 "warning: updating to rev %d because of "
343                 "new feature flag, running e2fsck is recommended",
344                 EXT3_DYNAMIC_REV);
345
346         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
347         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
348         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
349         /* leave es->s_feature_*compat flags alone */
350         /* es->s_uuid will be set by e2fsck if empty */
351
352         /*
353          * The rest of the superblock fields should be zero, and if not it
354          * means they are likely already in use, so leave them alone.  We
355          * can leave it up to e2fsck to clean up any inconsistencies there.
356          */
357 }
358
359 /*
360  * Open the external journal device
361  */
362 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
363 {
364         struct block_device *bdev;
365         char b[BDEVNAME_SIZE];
366
367         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
368         if (IS_ERR(bdev))
369                 goto fail;
370         return bdev;
371
372 fail:
373         ext3_msg(sb, "error: failed to open journal device %s: %ld",
374                 __bdevname(dev, b), PTR_ERR(bdev));
375
376         return NULL;
377 }
378
379 /*
380  * Release the journal device
381  */
382 static int ext3_blkdev_put(struct block_device *bdev)
383 {
384         bd_release(bdev);
385         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
386 }
387
388 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
389 {
390         struct block_device *bdev;
391         int ret = -ENODEV;
392
393         bdev = sbi->journal_bdev;
394         if (bdev) {
395                 ret = ext3_blkdev_put(bdev);
396                 sbi->journal_bdev = NULL;
397         }
398         return ret;
399 }
400
401 static inline struct inode *orphan_list_entry(struct list_head *l)
402 {
403         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
404 }
405
406 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
407 {
408         struct list_head *l;
409
410         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
411                le32_to_cpu(sbi->s_es->s_last_orphan));
412
413         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
414         list_for_each(l, &sbi->s_orphan) {
415                 struct inode *inode = orphan_list_entry(l);
416                 ext3_msg(sb, KERN_ERR, "  "
417                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
418                        inode->i_sb->s_id, inode->i_ino, inode,
419                        inode->i_mode, inode->i_nlink,
420                        NEXT_ORPHAN(inode));
421         }
422 }
423
424 static void ext3_put_super (struct super_block * sb)
425 {
426         struct ext3_sb_info *sbi = EXT3_SB(sb);
427         struct ext3_super_block *es = sbi->s_es;
428         int i, err;
429
430         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
431         ext3_xattr_put_super(sb);
432         err = journal_destroy(sbi->s_journal);
433         sbi->s_journal = NULL;
434         if (err < 0)
435                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
436
437         if (!(sb->s_flags & MS_RDONLY)) {
438                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
439                 es->s_state = cpu_to_le16(sbi->s_mount_state);
440                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
441                 mark_buffer_dirty(sbi->s_sbh);
442                 ext3_commit_super(sb, es, 1);
443         }
444
445         for (i = 0; i < sbi->s_gdb_count; i++)
446                 brelse(sbi->s_group_desc[i]);
447         kfree(sbi->s_group_desc);
448         percpu_counter_destroy(&sbi->s_freeblocks_counter);
449         percpu_counter_destroy(&sbi->s_freeinodes_counter);
450         percpu_counter_destroy(&sbi->s_dirs_counter);
451         brelse(sbi->s_sbh);
452 #ifdef CONFIG_QUOTA
453         for (i = 0; i < MAXQUOTAS; i++)
454                 kfree(sbi->s_qf_names[i]);
455 #endif
456
457         /* Debugging code just in case the in-memory inode orphan list
458          * isn't empty.  The on-disk one can be non-empty if we've
459          * detected an error and taken the fs readonly, but the
460          * in-memory list had better be clean by this point. */
461         if (!list_empty(&sbi->s_orphan))
462                 dump_orphan_list(sb, sbi);
463         J_ASSERT(list_empty(&sbi->s_orphan));
464
465         invalidate_bdev(sb->s_bdev);
466         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
467                 /*
468                  * Invalidate the journal device's buffers.  We don't want them
469                  * floating about in memory - the physical journal device may
470                  * hotswapped, and it breaks the `ro-after' testing code.
471                  */
472                 sync_blockdev(sbi->journal_bdev);
473                 invalidate_bdev(sbi->journal_bdev);
474                 ext3_blkdev_remove(sbi);
475         }
476         sb->s_fs_info = NULL;
477         kfree(sbi->s_blockgroup_lock);
478         kfree(sbi);
479 }
480
481 static struct kmem_cache *ext3_inode_cachep;
482
483 /*
484  * Called inside transaction, so use GFP_NOFS
485  */
486 static struct inode *ext3_alloc_inode(struct super_block *sb)
487 {
488         struct ext3_inode_info *ei;
489
490         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
491         if (!ei)
492                 return NULL;
493         ei->i_block_alloc_info = NULL;
494         ei->vfs_inode.i_version = 1;
495         atomic_set(&ei->i_datasync_tid, 0);
496         atomic_set(&ei->i_sync_tid, 0);
497         return &ei->vfs_inode;
498 }
499
500 static void ext3_destroy_inode(struct inode *inode)
501 {
502         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
503                 printk("EXT3 Inode %p: orphan list check failed!\n",
504                         EXT3_I(inode));
505                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
506                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
507                                 false);
508                 dump_stack();
509         }
510         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
511 }
512
513 static void init_once(void *foo)
514 {
515         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
516
517         INIT_LIST_HEAD(&ei->i_orphan);
518 #ifdef CONFIG_EXT3_FS_XATTR
519         init_rwsem(&ei->xattr_sem);
520 #endif
521         mutex_init(&ei->truncate_mutex);
522         inode_init_once(&ei->vfs_inode);
523 }
524
525 static int init_inodecache(void)
526 {
527         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
528                                              sizeof(struct ext3_inode_info),
529                                              0, (SLAB_RECLAIM_ACCOUNT|
530                                                 SLAB_MEM_SPREAD),
531                                              init_once);
532         if (ext3_inode_cachep == NULL)
533                 return -ENOMEM;
534         return 0;
535 }
536
537 static void destroy_inodecache(void)
538 {
539         kmem_cache_destroy(ext3_inode_cachep);
540 }
541
542 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
543 {
544 #if defined(CONFIG_QUOTA)
545         struct ext3_sb_info *sbi = EXT3_SB(sb);
546
547         if (sbi->s_jquota_fmt) {
548                 char *fmtname = "";
549
550                 switch (sbi->s_jquota_fmt) {
551                 case QFMT_VFS_OLD:
552                         fmtname = "vfsold";
553                         break;
554                 case QFMT_VFS_V0:
555                         fmtname = "vfsv0";
556                         break;
557                 case QFMT_VFS_V1:
558                         fmtname = "vfsv1";
559                         break;
560                 }
561                 seq_printf(seq, ",jqfmt=%s", fmtname);
562         }
563
564         if (sbi->s_qf_names[USRQUOTA])
565                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
566
567         if (sbi->s_qf_names[GRPQUOTA])
568                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
569
570         if (test_opt(sb, USRQUOTA))
571                 seq_puts(seq, ",usrquota");
572
573         if (test_opt(sb, GRPQUOTA))
574                 seq_puts(seq, ",grpquota");
575 #endif
576 }
577
578 static char *data_mode_string(unsigned long mode)
579 {
580         switch (mode) {
581         case EXT3_MOUNT_JOURNAL_DATA:
582                 return "journal";
583         case EXT3_MOUNT_ORDERED_DATA:
584                 return "ordered";
585         case EXT3_MOUNT_WRITEBACK_DATA:
586                 return "writeback";
587         }
588         return "unknown";
589 }
590
591 /*
592  * Show an option if
593  *  - it's set to a non-default value OR
594  *  - if the per-sb default is different from the global default
595  */
596 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
597 {
598         struct super_block *sb = vfs->mnt_sb;
599         struct ext3_sb_info *sbi = EXT3_SB(sb);
600         struct ext3_super_block *es = sbi->s_es;
601         unsigned long def_mount_opts;
602
603         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
604
605         if (sbi->s_sb_block != 1)
606                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
607         if (test_opt(sb, MINIX_DF))
608                 seq_puts(seq, ",minixdf");
609         if (test_opt(sb, GRPID))
610                 seq_puts(seq, ",grpid");
611         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
612                 seq_puts(seq, ",nogrpid");
613         if (sbi->s_resuid != EXT3_DEF_RESUID ||
614             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
615                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
616         }
617         if (sbi->s_resgid != EXT3_DEF_RESGID ||
618             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
619                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
620         }
621         if (test_opt(sb, ERRORS_RO)) {
622                 int def_errors = le16_to_cpu(es->s_errors);
623
624                 if (def_errors == EXT3_ERRORS_PANIC ||
625                     def_errors == EXT3_ERRORS_CONTINUE) {
626                         seq_puts(seq, ",errors=remount-ro");
627                 }
628         }
629         if (test_opt(sb, ERRORS_CONT))
630                 seq_puts(seq, ",errors=continue");
631         if (test_opt(sb, ERRORS_PANIC))
632                 seq_puts(seq, ",errors=panic");
633         if (test_opt(sb, NO_UID32))
634                 seq_puts(seq, ",nouid32");
635         if (test_opt(sb, DEBUG))
636                 seq_puts(seq, ",debug");
637         if (test_opt(sb, OLDALLOC))
638                 seq_puts(seq, ",oldalloc");
639 #ifdef CONFIG_EXT3_FS_XATTR
640         if (test_opt(sb, XATTR_USER))
641                 seq_puts(seq, ",user_xattr");
642         if (!test_opt(sb, XATTR_USER) &&
643             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
644                 seq_puts(seq, ",nouser_xattr");
645         }
646 #endif
647 #ifdef CONFIG_EXT3_FS_POSIX_ACL
648         if (test_opt(sb, POSIX_ACL))
649                 seq_puts(seq, ",acl");
650         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
651                 seq_puts(seq, ",noacl");
652 #endif
653         if (!test_opt(sb, RESERVATION))
654                 seq_puts(seq, ",noreservation");
655         if (sbi->s_commit_interval) {
656                 seq_printf(seq, ",commit=%u",
657                            (unsigned) (sbi->s_commit_interval / HZ));
658         }
659
660         /*
661          * Always display barrier state so it's clear what the status is.
662          */
663         seq_puts(seq, ",barrier=");
664         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
665         seq_printf(seq, ",data=%s", data_mode_string(test_opt(sb, DATA_FLAGS)));
666         if (test_opt(sb, DATA_ERR_ABORT))
667                 seq_puts(seq, ",data_err=abort");
668
669         if (test_opt(sb, NOLOAD))
670                 seq_puts(seq, ",norecovery");
671
672         ext3_show_quota_options(seq, sb);
673
674         return 0;
675 }
676
677
678 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
679                 u64 ino, u32 generation)
680 {
681         struct inode *inode;
682
683         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
684                 return ERR_PTR(-ESTALE);
685         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
686                 return ERR_PTR(-ESTALE);
687
688         /* iget isn't really right if the inode is currently unallocated!!
689          *
690          * ext3_read_inode will return a bad_inode if the inode had been
691          * deleted, so we should be safe.
692          *
693          * Currently we don't know the generation for parent directory, so
694          * a generation of 0 means "accept any"
695          */
696         inode = ext3_iget(sb, ino);
697         if (IS_ERR(inode))
698                 return ERR_CAST(inode);
699         if (generation && inode->i_generation != generation) {
700                 iput(inode);
701                 return ERR_PTR(-ESTALE);
702         }
703
704         return inode;
705 }
706
707 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
708                 int fh_len, int fh_type)
709 {
710         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
711                                     ext3_nfs_get_inode);
712 }
713
714 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
715                 int fh_len, int fh_type)
716 {
717         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
718                                     ext3_nfs_get_inode);
719 }
720
721 /*
722  * Try to release metadata pages (indirect blocks, directories) which are
723  * mapped via the block device.  Since these pages could have journal heads
724  * which would prevent try_to_free_buffers() from freeing them, we must use
725  * jbd layer's try_to_free_buffers() function to release them.
726  */
727 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
728                                  gfp_t wait)
729 {
730         journal_t *journal = EXT3_SB(sb)->s_journal;
731
732         WARN_ON(PageChecked(page));
733         if (!page_has_buffers(page))
734                 return 0;
735         if (journal)
736                 return journal_try_to_free_buffers(journal, page, 
737                                                    wait & ~__GFP_WAIT);
738         return try_to_free_buffers(page);
739 }
740
741 #ifdef CONFIG_QUOTA
742 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
743 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
744
745 static int ext3_write_dquot(struct dquot *dquot);
746 static int ext3_acquire_dquot(struct dquot *dquot);
747 static int ext3_release_dquot(struct dquot *dquot);
748 static int ext3_mark_dquot_dirty(struct dquot *dquot);
749 static int ext3_write_info(struct super_block *sb, int type);
750 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
751                                 char *path);
752 static int ext3_quota_on_mount(struct super_block *sb, int type);
753 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
754                                size_t len, loff_t off);
755 static ssize_t ext3_quota_write(struct super_block *sb, int type,
756                                 const char *data, size_t len, loff_t off);
757
758 static const struct dquot_operations ext3_quota_operations = {
759         .write_dquot    = ext3_write_dquot,
760         .acquire_dquot  = ext3_acquire_dquot,
761         .release_dquot  = ext3_release_dquot,
762         .mark_dirty     = ext3_mark_dquot_dirty,
763         .write_info     = ext3_write_info,
764         .alloc_dquot    = dquot_alloc,
765         .destroy_dquot  = dquot_destroy,
766 };
767
768 static const struct quotactl_ops ext3_qctl_operations = {
769         .quota_on       = ext3_quota_on,
770         .quota_off      = dquot_quota_off,
771         .quota_sync     = dquot_quota_sync,
772         .get_info       = dquot_get_dqinfo,
773         .set_info       = dquot_set_dqinfo,
774         .get_dqblk      = dquot_get_dqblk,
775         .set_dqblk      = dquot_set_dqblk
776 };
777 #endif
778
779 static const struct super_operations ext3_sops = {
780         .alloc_inode    = ext3_alloc_inode,
781         .destroy_inode  = ext3_destroy_inode,
782         .write_inode    = ext3_write_inode,
783         .dirty_inode    = ext3_dirty_inode,
784         .evict_inode    = ext3_evict_inode,
785         .put_super      = ext3_put_super,
786         .sync_fs        = ext3_sync_fs,
787         .freeze_fs      = ext3_freeze,
788         .unfreeze_fs    = ext3_unfreeze,
789         .statfs         = ext3_statfs,
790         .remount_fs     = ext3_remount,
791         .show_options   = ext3_show_options,
792 #ifdef CONFIG_QUOTA
793         .quota_read     = ext3_quota_read,
794         .quota_write    = ext3_quota_write,
795 #endif
796         .bdev_try_to_free_page = bdev_try_to_free_page,
797 };
798
799 static const struct export_operations ext3_export_ops = {
800         .fh_to_dentry = ext3_fh_to_dentry,
801         .fh_to_parent = ext3_fh_to_parent,
802         .get_parent = ext3_get_parent,
803 };
804
805 enum {
806         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
807         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
808         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
809         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
810         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
811         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
812         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
813         Opt_data_err_abort, Opt_data_err_ignore,
814         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
815         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
816         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
817         Opt_resize, Opt_usrquota, Opt_grpquota
818 };
819
820 static const match_table_t tokens = {
821         {Opt_bsd_df, "bsddf"},
822         {Opt_minix_df, "minixdf"},
823         {Opt_grpid, "grpid"},
824         {Opt_grpid, "bsdgroups"},
825         {Opt_nogrpid, "nogrpid"},
826         {Opt_nogrpid, "sysvgroups"},
827         {Opt_resgid, "resgid=%u"},
828         {Opt_resuid, "resuid=%u"},
829         {Opt_sb, "sb=%u"},
830         {Opt_err_cont, "errors=continue"},
831         {Opt_err_panic, "errors=panic"},
832         {Opt_err_ro, "errors=remount-ro"},
833         {Opt_nouid32, "nouid32"},
834         {Opt_nocheck, "nocheck"},
835         {Opt_nocheck, "check=none"},
836         {Opt_debug, "debug"},
837         {Opt_oldalloc, "oldalloc"},
838         {Opt_orlov, "orlov"},
839         {Opt_user_xattr, "user_xattr"},
840         {Opt_nouser_xattr, "nouser_xattr"},
841         {Opt_acl, "acl"},
842         {Opt_noacl, "noacl"},
843         {Opt_reservation, "reservation"},
844         {Opt_noreservation, "noreservation"},
845         {Opt_noload, "noload"},
846         {Opt_noload, "norecovery"},
847         {Opt_nobh, "nobh"},
848         {Opt_bh, "bh"},
849         {Opt_commit, "commit=%u"},
850         {Opt_journal_update, "journal=update"},
851         {Opt_journal_inum, "journal=%u"},
852         {Opt_journal_dev, "journal_dev=%u"},
853         {Opt_abort, "abort"},
854         {Opt_data_journal, "data=journal"},
855         {Opt_data_ordered, "data=ordered"},
856         {Opt_data_writeback, "data=writeback"},
857         {Opt_data_err_abort, "data_err=abort"},
858         {Opt_data_err_ignore, "data_err=ignore"},
859         {Opt_offusrjquota, "usrjquota="},
860         {Opt_usrjquota, "usrjquota=%s"},
861         {Opt_offgrpjquota, "grpjquota="},
862         {Opt_grpjquota, "grpjquota=%s"},
863         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
864         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
865         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
866         {Opt_grpquota, "grpquota"},
867         {Opt_noquota, "noquota"},
868         {Opt_quota, "quota"},
869         {Opt_usrquota, "usrquota"},
870         {Opt_barrier, "barrier=%u"},
871         {Opt_barrier, "barrier"},
872         {Opt_nobarrier, "nobarrier"},
873         {Opt_resize, "resize"},
874         {Opt_err, NULL},
875 };
876
877 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
878 {
879         ext3_fsblk_t    sb_block;
880         char            *options = (char *) *data;
881
882         if (!options || strncmp(options, "sb=", 3) != 0)
883                 return 1;       /* Default location */
884         options += 3;
885         /*todo: use simple_strtoll with >32bit ext3 */
886         sb_block = simple_strtoul(options, &options, 0);
887         if (*options && *options != ',') {
888                 ext3_msg(sb, "error: invalid sb specification: %s",
889                        (char *) *data);
890                 return 1;
891         }
892         if (*options == ',')
893                 options++;
894         *data = (void *) options;
895         return sb_block;
896 }
897
898 #ifdef CONFIG_QUOTA
899 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
900 {
901         struct ext3_sb_info *sbi = EXT3_SB(sb);
902         char *qname;
903
904         if (sb_any_quota_loaded(sb) &&
905                 !sbi->s_qf_names[qtype]) {
906                 ext3_msg(sb, KERN_ERR,
907                         "Cannot change journaled "
908                         "quota options when quota turned on");
909                 return 0;
910         }
911         qname = match_strdup(args);
912         if (!qname) {
913                 ext3_msg(sb, KERN_ERR,
914                         "Not enough memory for storing quotafile name");
915                 return 0;
916         }
917         if (sbi->s_qf_names[qtype] &&
918                 strcmp(sbi->s_qf_names[qtype], qname)) {
919                 ext3_msg(sb, KERN_ERR,
920                         "%s quota file already specified", QTYPE2NAME(qtype));
921                 kfree(qname);
922                 return 0;
923         }
924         sbi->s_qf_names[qtype] = qname;
925         if (strchr(sbi->s_qf_names[qtype], '/')) {
926                 ext3_msg(sb, KERN_ERR,
927                         "quotafile must be on filesystem root");
928                 kfree(sbi->s_qf_names[qtype]);
929                 sbi->s_qf_names[qtype] = NULL;
930                 return 0;
931         }
932         set_opt(sbi->s_mount_opt, QUOTA);
933         return 1;
934 }
935
936 static int clear_qf_name(struct super_block *sb, int qtype) {
937
938         struct ext3_sb_info *sbi = EXT3_SB(sb);
939
940         if (sb_any_quota_loaded(sb) &&
941                 sbi->s_qf_names[qtype]) {
942                 ext3_msg(sb, KERN_ERR, "Cannot change journaled quota options"
943                         " when quota turned on");
944                 return 0;
945         }
946         /*
947          * The space will be released later when all options are confirmed
948          * to be correct
949          */
950         sbi->s_qf_names[qtype] = NULL;
951         return 1;
952 }
953 #endif
954
955 static int parse_options (char *options, struct super_block *sb,
956                           unsigned int *inum, unsigned long *journal_devnum,
957                           ext3_fsblk_t *n_blocks_count, int is_remount)
958 {
959         struct ext3_sb_info *sbi = EXT3_SB(sb);
960         char * p;
961         substring_t args[MAX_OPT_ARGS];
962         int data_opt = 0;
963         int option;
964 #ifdef CONFIG_QUOTA
965         int qfmt;
966 #endif
967
968         if (!options)
969                 return 1;
970
971         while ((p = strsep (&options, ",")) != NULL) {
972                 int token;
973                 if (!*p)
974                         continue;
975                 /*
976                  * Initialize args struct so we know whether arg was
977                  * found; some options take optional arguments.
978                  */
979                 args[0].to = args[0].from = 0;
980                 token = match_token(p, tokens, args);
981                 switch (token) {
982                 case Opt_bsd_df:
983                         clear_opt (sbi->s_mount_opt, MINIX_DF);
984                         break;
985                 case Opt_minix_df:
986                         set_opt (sbi->s_mount_opt, MINIX_DF);
987                         break;
988                 case Opt_grpid:
989                         set_opt (sbi->s_mount_opt, GRPID);
990                         break;
991                 case Opt_nogrpid:
992                         clear_opt (sbi->s_mount_opt, GRPID);
993                         break;
994                 case Opt_resuid:
995                         if (match_int(&args[0], &option))
996                                 return 0;
997                         sbi->s_resuid = option;
998                         break;
999                 case Opt_resgid:
1000                         if (match_int(&args[0], &option))
1001                                 return 0;
1002                         sbi->s_resgid = option;
1003                         break;
1004                 case Opt_sb:
1005                         /* handled by get_sb_block() instead of here */
1006                         /* *sb_block = match_int(&args[0]); */
1007                         break;
1008                 case Opt_err_panic:
1009                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1010                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1011                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
1012                         break;
1013                 case Opt_err_ro:
1014                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
1015                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1016                         set_opt (sbi->s_mount_opt, ERRORS_RO);
1017                         break;
1018                 case Opt_err_cont:
1019                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
1020                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
1021                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
1022                         break;
1023                 case Opt_nouid32:
1024                         set_opt (sbi->s_mount_opt, NO_UID32);
1025                         break;
1026                 case Opt_nocheck:
1027                         clear_opt (sbi->s_mount_opt, CHECK);
1028                         break;
1029                 case Opt_debug:
1030                         set_opt (sbi->s_mount_opt, DEBUG);
1031                         break;
1032                 case Opt_oldalloc:
1033                         set_opt (sbi->s_mount_opt, OLDALLOC);
1034                         break;
1035                 case Opt_orlov:
1036                         clear_opt (sbi->s_mount_opt, OLDALLOC);
1037                         break;
1038 #ifdef CONFIG_EXT3_FS_XATTR
1039                 case Opt_user_xattr:
1040                         set_opt (sbi->s_mount_opt, XATTR_USER);
1041                         break;
1042                 case Opt_nouser_xattr:
1043                         clear_opt (sbi->s_mount_opt, XATTR_USER);
1044                         break;
1045 #else
1046                 case Opt_user_xattr:
1047                 case Opt_nouser_xattr:
1048                         ext3_msg(sb, KERN_INFO,
1049                                 "(no)user_xattr options not supported");
1050                         break;
1051 #endif
1052 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1053                 case Opt_acl:
1054                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1055                         break;
1056                 case Opt_noacl:
1057                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1058                         break;
1059 #else
1060                 case Opt_acl:
1061                 case Opt_noacl:
1062                         ext3_msg(sb, KERN_INFO,
1063                                 "(no)acl options not supported");
1064                         break;
1065 #endif
1066                 case Opt_reservation:
1067                         set_opt(sbi->s_mount_opt, RESERVATION);
1068                         break;
1069                 case Opt_noreservation:
1070                         clear_opt(sbi->s_mount_opt, RESERVATION);
1071                         break;
1072                 case Opt_journal_update:
1073                         /* @@@ FIXME */
1074                         /* Eventually we will want to be able to create
1075                            a journal file here.  For now, only allow the
1076                            user to specify an existing inode to be the
1077                            journal file. */
1078                         if (is_remount) {
1079                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1080                                         "journal on remount");
1081                                 return 0;
1082                         }
1083                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1084                         break;
1085                 case Opt_journal_inum:
1086                         if (is_remount) {
1087                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1088                                        "journal on remount");
1089                                 return 0;
1090                         }
1091                         if (match_int(&args[0], &option))
1092                                 return 0;
1093                         *inum = option;
1094                         break;
1095                 case Opt_journal_dev:
1096                         if (is_remount) {
1097                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1098                                        "journal on remount");
1099                                 return 0;
1100                         }
1101                         if (match_int(&args[0], &option))
1102                                 return 0;
1103                         *journal_devnum = option;
1104                         break;
1105                 case Opt_noload:
1106                         set_opt (sbi->s_mount_opt, NOLOAD);
1107                         break;
1108                 case Opt_commit:
1109                         if (match_int(&args[0], &option))
1110                                 return 0;
1111                         if (option < 0)
1112                                 return 0;
1113                         if (option == 0)
1114                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1115                         sbi->s_commit_interval = HZ * option;
1116                         break;
1117                 case Opt_data_journal:
1118                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1119                         goto datacheck;
1120                 case Opt_data_ordered:
1121                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1122                         goto datacheck;
1123                 case Opt_data_writeback:
1124                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1125                 datacheck:
1126                         if (is_remount) {
1127                                 if (test_opt(sb, DATA_FLAGS) == data_opt)
1128                                         break;
1129                                 ext3_msg(sb, KERN_ERR,
1130                                         "error: cannot change "
1131                                         "data mode on remount. The filesystem "
1132                                         "is mounted in data=%s mode and you "
1133                                         "try to remount it in data=%s mode.",
1134                                         data_mode_string(test_opt(sb,
1135                                                         DATA_FLAGS)),
1136                                         data_mode_string(data_opt));
1137                                 return 0;
1138                         } else {
1139                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1140                                 sbi->s_mount_opt |= data_opt;
1141                         }
1142                         break;
1143                 case Opt_data_err_abort:
1144                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1145                         break;
1146                 case Opt_data_err_ignore:
1147                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1148                         break;
1149 #ifdef CONFIG_QUOTA
1150                 case Opt_usrjquota:
1151                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1152                                 return 0;
1153                         break;
1154                 case Opt_grpjquota:
1155                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1156                                 return 0;
1157                         break;
1158                 case Opt_offusrjquota:
1159                         if (!clear_qf_name(sb, USRQUOTA))
1160                                 return 0;
1161                         break;
1162                 case Opt_offgrpjquota:
1163                         if (!clear_qf_name(sb, GRPQUOTA))
1164                                 return 0;
1165                         break;
1166                 case Opt_jqfmt_vfsold:
1167                         qfmt = QFMT_VFS_OLD;
1168                         goto set_qf_format;
1169                 case Opt_jqfmt_vfsv0:
1170                         qfmt = QFMT_VFS_V0;
1171                         goto set_qf_format;
1172                 case Opt_jqfmt_vfsv1:
1173                         qfmt = QFMT_VFS_V1;
1174 set_qf_format:
1175                         if (sb_any_quota_loaded(sb) &&
1176                             sbi->s_jquota_fmt != qfmt) {
1177                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1178                                         "journaled quota options when "
1179                                         "quota turned on.");
1180                                 return 0;
1181                         }
1182                         sbi->s_jquota_fmt = qfmt;
1183                         break;
1184                 case Opt_quota:
1185                 case Opt_usrquota:
1186                         set_opt(sbi->s_mount_opt, QUOTA);
1187                         set_opt(sbi->s_mount_opt, USRQUOTA);
1188                         break;
1189                 case Opt_grpquota:
1190                         set_opt(sbi->s_mount_opt, QUOTA);
1191                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1192                         break;
1193                 case Opt_noquota:
1194                         if (sb_any_quota_loaded(sb)) {
1195                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1196                                         "quota options when quota turned on.");
1197                                 return 0;
1198                         }
1199                         clear_opt(sbi->s_mount_opt, QUOTA);
1200                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1201                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1202                         break;
1203 #else
1204                 case Opt_quota:
1205                 case Opt_usrquota:
1206                 case Opt_grpquota:
1207                         ext3_msg(sb, KERN_ERR,
1208                                 "error: quota options not supported.");
1209                         break;
1210                 case Opt_usrjquota:
1211                 case Opt_grpjquota:
1212                 case Opt_offusrjquota:
1213                 case Opt_offgrpjquota:
1214                 case Opt_jqfmt_vfsold:
1215                 case Opt_jqfmt_vfsv0:
1216                 case Opt_jqfmt_vfsv1:
1217                         ext3_msg(sb, KERN_ERR,
1218                                 "error: journaled quota options not "
1219                                 "supported.");
1220                         break;
1221                 case Opt_noquota:
1222                         break;
1223 #endif
1224                 case Opt_abort:
1225                         set_opt(sbi->s_mount_opt, ABORT);
1226                         break;
1227                 case Opt_nobarrier:
1228                         clear_opt(sbi->s_mount_opt, BARRIER);
1229                         break;
1230                 case Opt_barrier:
1231                         if (args[0].from) {
1232                                 if (match_int(&args[0], &option))
1233                                         return 0;
1234                         } else
1235                                 option = 1;     /* No argument, default to 1 */
1236                         if (option)
1237                                 set_opt(sbi->s_mount_opt, BARRIER);
1238                         else
1239                                 clear_opt(sbi->s_mount_opt, BARRIER);
1240                         break;
1241                 case Opt_ignore:
1242                         break;
1243                 case Opt_resize:
1244                         if (!is_remount) {
1245                                 ext3_msg(sb, KERN_ERR,
1246                                         "error: resize option only available "
1247                                         "for remount");
1248                                 return 0;
1249                         }
1250                         if (match_int(&args[0], &option) != 0)
1251                                 return 0;
1252                         *n_blocks_count = option;
1253                         break;
1254                 case Opt_nobh:
1255                         ext3_msg(sb, KERN_WARNING,
1256                                 "warning: ignoring deprecated nobh option");
1257                         break;
1258                 case Opt_bh:
1259                         ext3_msg(sb, KERN_WARNING,
1260                                 "warning: ignoring deprecated bh option");
1261                         break;
1262                 default:
1263                         ext3_msg(sb, KERN_ERR,
1264                                 "error: unrecognized mount option \"%s\" "
1265                                 "or missing value", p);
1266                         return 0;
1267                 }
1268         }
1269 #ifdef CONFIG_QUOTA
1270         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1271                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1272                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1273                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1274                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1275
1276                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1277                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1278                                         "format mixing.");
1279                         return 0;
1280                 }
1281
1282                 if (!sbi->s_jquota_fmt) {
1283                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1284                                         "not specified.");
1285                         return 0;
1286                 }
1287         } else {
1288                 if (sbi->s_jquota_fmt) {
1289                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1290                                         "specified with no journaling "
1291                                         "enabled.");
1292                         return 0;
1293                 }
1294         }
1295 #endif
1296         return 1;
1297 }
1298
1299 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1300                             int read_only)
1301 {
1302         struct ext3_sb_info *sbi = EXT3_SB(sb);
1303         int res = 0;
1304
1305         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1306                 ext3_msg(sb, KERN_ERR,
1307                         "error: revision level too high, "
1308                         "forcing read-only mode");
1309                 res = MS_RDONLY;
1310         }
1311         if (read_only)
1312                 return res;
1313         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1314                 ext3_msg(sb, KERN_WARNING,
1315                         "warning: mounting unchecked fs, "
1316                         "running e2fsck is recommended");
1317         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1318                 ext3_msg(sb, KERN_WARNING,
1319                         "warning: mounting fs with errors, "
1320                         "running e2fsck is recommended");
1321         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1322                  le16_to_cpu(es->s_mnt_count) >=
1323                         le16_to_cpu(es->s_max_mnt_count))
1324                 ext3_msg(sb, KERN_WARNING,
1325                         "warning: maximal mount count reached, "
1326                         "running e2fsck is recommended");
1327         else if (le32_to_cpu(es->s_checkinterval) &&
1328                 (le32_to_cpu(es->s_lastcheck) +
1329                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1330                 ext3_msg(sb, KERN_WARNING,
1331                         "warning: checktime reached, "
1332                         "running e2fsck is recommended");
1333 #if 0
1334                 /* @@@ We _will_ want to clear the valid bit if we find
1335                    inconsistencies, to force a fsck at reboot.  But for
1336                    a plain journaled filesystem we can keep it set as
1337                    valid forever! :) */
1338         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1339 #endif
1340         if (!le16_to_cpu(es->s_max_mnt_count))
1341                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1342         le16_add_cpu(&es->s_mnt_count, 1);
1343         es->s_mtime = cpu_to_le32(get_seconds());
1344         ext3_update_dynamic_rev(sb);
1345         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1346
1347         ext3_commit_super(sb, es, 1);
1348         if (test_opt(sb, DEBUG))
1349                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1350                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1351                         sb->s_blocksize,
1352                         sbi->s_groups_count,
1353                         EXT3_BLOCKS_PER_GROUP(sb),
1354                         EXT3_INODES_PER_GROUP(sb),
1355                         sbi->s_mount_opt);
1356
1357         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1358                 char b[BDEVNAME_SIZE];
1359                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1360                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1361         } else {
1362                 ext3_msg(sb, KERN_INFO, "using internal journal");
1363         }
1364         return res;
1365 }
1366
1367 /* Called at mount-time, super-block is locked */
1368 static int ext3_check_descriptors(struct super_block *sb)
1369 {
1370         struct ext3_sb_info *sbi = EXT3_SB(sb);
1371         int i;
1372
1373         ext3_debug ("Checking group descriptors");
1374
1375         for (i = 0; i < sbi->s_groups_count; i++) {
1376                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1377                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1378                 ext3_fsblk_t last_block;
1379
1380                 if (i == sbi->s_groups_count - 1)
1381                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1382                 else
1383                         last_block = first_block +
1384                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1385
1386                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1387                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1388                 {
1389                         ext3_error (sb, "ext3_check_descriptors",
1390                                     "Block bitmap for group %d"
1391                                     " not in group (block %lu)!",
1392                                     i, (unsigned long)
1393                                         le32_to_cpu(gdp->bg_block_bitmap));
1394                         return 0;
1395                 }
1396                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1397                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1398                 {
1399                         ext3_error (sb, "ext3_check_descriptors",
1400                                     "Inode bitmap for group %d"
1401                                     " not in group (block %lu)!",
1402                                     i, (unsigned long)
1403                                         le32_to_cpu(gdp->bg_inode_bitmap));
1404                         return 0;
1405                 }
1406                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1407                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1408                     last_block)
1409                 {
1410                         ext3_error (sb, "ext3_check_descriptors",
1411                                     "Inode table for group %d"
1412                                     " not in group (block %lu)!",
1413                                     i, (unsigned long)
1414                                         le32_to_cpu(gdp->bg_inode_table));
1415                         return 0;
1416                 }
1417         }
1418
1419         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1420         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1421         return 1;
1422 }
1423
1424
1425 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1426  * the superblock) which were deleted from all directories, but held open by
1427  * a process at the time of a crash.  We walk the list and try to delete these
1428  * inodes at recovery time (only with a read-write filesystem).
1429  *
1430  * In order to keep the orphan inode chain consistent during traversal (in
1431  * case of crash during recovery), we link each inode into the superblock
1432  * orphan list_head and handle it the same way as an inode deletion during
1433  * normal operation (which journals the operations for us).
1434  *
1435  * We only do an iget() and an iput() on each inode, which is very safe if we
1436  * accidentally point at an in-use or already deleted inode.  The worst that
1437  * can happen in this case is that we get a "bit already cleared" message from
1438  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1439  * e2fsck was run on this filesystem, and it must have already done the orphan
1440  * inode cleanup for us, so we can safely abort without any further action.
1441  */
1442 static void ext3_orphan_cleanup (struct super_block * sb,
1443                                  struct ext3_super_block * es)
1444 {
1445         unsigned int s_flags = sb->s_flags;
1446         int nr_orphans = 0, nr_truncates = 0;
1447 #ifdef CONFIG_QUOTA
1448         int i;
1449 #endif
1450         if (!es->s_last_orphan) {
1451                 jbd_debug(4, "no orphan inodes to clean up\n");
1452                 return;
1453         }
1454
1455         if (bdev_read_only(sb->s_bdev)) {
1456                 ext3_msg(sb, KERN_ERR, "error: write access "
1457                         "unavailable, skipping orphan cleanup.");
1458                 return;
1459         }
1460
1461         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1462                 if (es->s_last_orphan)
1463                         jbd_debug(1, "Errors on filesystem, "
1464                                   "clearing orphan list.\n");
1465                 es->s_last_orphan = 0;
1466                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1467                 return;
1468         }
1469
1470         if (s_flags & MS_RDONLY) {
1471                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1472                 sb->s_flags &= ~MS_RDONLY;
1473         }
1474 #ifdef CONFIG_QUOTA
1475         /* Needed for iput() to work correctly and not trash data */
1476         sb->s_flags |= MS_ACTIVE;
1477         /* Turn on quotas so that they are updated correctly */
1478         for (i = 0; i < MAXQUOTAS; i++) {
1479                 if (EXT3_SB(sb)->s_qf_names[i]) {
1480                         int ret = ext3_quota_on_mount(sb, i);
1481                         if (ret < 0)
1482                                 ext3_msg(sb, KERN_ERR,
1483                                         "error: cannot turn on journaled "
1484                                         "quota: %d", ret);
1485                 }
1486         }
1487 #endif
1488
1489         while (es->s_last_orphan) {
1490                 struct inode *inode;
1491
1492                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1493                 if (IS_ERR(inode)) {
1494                         es->s_last_orphan = 0;
1495                         break;
1496                 }
1497
1498                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1499                 dquot_initialize(inode);
1500                 if (inode->i_nlink) {
1501                         printk(KERN_DEBUG
1502                                 "%s: truncating inode %lu to %Ld bytes\n",
1503                                 __func__, inode->i_ino, inode->i_size);
1504                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1505                                   inode->i_ino, inode->i_size);
1506                         ext3_truncate(inode);
1507                         nr_truncates++;
1508                 } else {
1509                         printk(KERN_DEBUG
1510                                 "%s: deleting unreferenced inode %lu\n",
1511                                 __func__, inode->i_ino);
1512                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1513                                   inode->i_ino);
1514                         nr_orphans++;
1515                 }
1516                 iput(inode);  /* The delete magic happens here! */
1517         }
1518
1519 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1520
1521         if (nr_orphans)
1522                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1523                        PLURAL(nr_orphans));
1524         if (nr_truncates)
1525                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1526                        PLURAL(nr_truncates));
1527 #ifdef CONFIG_QUOTA
1528         /* Turn quotas off */
1529         for (i = 0; i < MAXQUOTAS; i++) {
1530                 if (sb_dqopt(sb)->files[i])
1531                         dquot_quota_off(sb, i);
1532         }
1533 #endif
1534         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1535 }
1536
1537 /*
1538  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1539  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1540  * We need to be 1 filesystem block less than the 2^32 sector limit.
1541  */
1542 static loff_t ext3_max_size(int bits)
1543 {
1544         loff_t res = EXT3_NDIR_BLOCKS;
1545         int meta_blocks;
1546         loff_t upper_limit;
1547
1548         /* This is calculated to be the largest file size for a
1549          * dense, file such that the total number of
1550          * sectors in the file, including data and all indirect blocks,
1551          * does not exceed 2^32 -1
1552          * __u32 i_blocks representing the total number of
1553          * 512 bytes blocks of the file
1554          */
1555         upper_limit = (1LL << 32) - 1;
1556
1557         /* total blocks in file system block size */
1558         upper_limit >>= (bits - 9);
1559
1560
1561         /* indirect blocks */
1562         meta_blocks = 1;
1563         /* double indirect blocks */
1564         meta_blocks += 1 + (1LL << (bits-2));
1565         /* tripple indirect blocks */
1566         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1567
1568         upper_limit -= meta_blocks;
1569         upper_limit <<= bits;
1570
1571         res += 1LL << (bits-2);
1572         res += 1LL << (2*(bits-2));
1573         res += 1LL << (3*(bits-2));
1574         res <<= bits;
1575         if (res > upper_limit)
1576                 res = upper_limit;
1577
1578         if (res > MAX_LFS_FILESIZE)
1579                 res = MAX_LFS_FILESIZE;
1580
1581         return res;
1582 }
1583
1584 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1585                                     ext3_fsblk_t logic_sb_block,
1586                                     int nr)
1587 {
1588         struct ext3_sb_info *sbi = EXT3_SB(sb);
1589         unsigned long bg, first_meta_bg;
1590         int has_super = 0;
1591
1592         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1593
1594         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1595             nr < first_meta_bg)
1596                 return (logic_sb_block + nr + 1);
1597         bg = sbi->s_desc_per_block * nr;
1598         if (ext3_bg_has_super(sb, bg))
1599                 has_super = 1;
1600         return (has_super + ext3_group_first_block_no(sb, bg));
1601 }
1602
1603
1604 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1605 {
1606         struct buffer_head * bh;
1607         struct ext3_super_block *es = NULL;
1608         struct ext3_sb_info *sbi;
1609         ext3_fsblk_t block;
1610         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1611         ext3_fsblk_t logic_sb_block;
1612         unsigned long offset = 0;
1613         unsigned int journal_inum = 0;
1614         unsigned long journal_devnum = 0;
1615         unsigned long def_mount_opts;
1616         struct inode *root;
1617         int blocksize;
1618         int hblock;
1619         int db_count;
1620         int i;
1621         int needs_recovery;
1622         int ret = -EINVAL;
1623         __le32 features;
1624         int err;
1625
1626         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1627         if (!sbi)
1628                 return -ENOMEM;
1629
1630         sbi->s_blockgroup_lock =
1631                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1632         if (!sbi->s_blockgroup_lock) {
1633                 kfree(sbi);
1634                 return -ENOMEM;
1635         }
1636         sb->s_fs_info = sbi;
1637         sbi->s_mount_opt = 0;
1638         sbi->s_resuid = EXT3_DEF_RESUID;
1639         sbi->s_resgid = EXT3_DEF_RESGID;
1640         sbi->s_sb_block = sb_block;
1641
1642         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1643         if (!blocksize) {
1644                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1645                 goto out_fail;
1646         }
1647
1648         /*
1649          * The ext3 superblock will not be buffer aligned for other than 1kB
1650          * block sizes.  We need to calculate the offset from buffer start.
1651          */
1652         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1653                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1654                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1655         } else {
1656                 logic_sb_block = sb_block;
1657         }
1658
1659         if (!(bh = sb_bread(sb, logic_sb_block))) {
1660                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1661                 goto out_fail;
1662         }
1663         /*
1664          * Note: s_es must be initialized as soon as possible because
1665          *       some ext3 macro-instructions depend on its value
1666          */
1667         es = (struct ext3_super_block *) (bh->b_data + offset);
1668         sbi->s_es = es;
1669         sb->s_magic = le16_to_cpu(es->s_magic);
1670         if (sb->s_magic != EXT3_SUPER_MAGIC)
1671                 goto cantfind_ext3;
1672
1673         /* Set defaults before we parse the mount options */
1674         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1675         if (def_mount_opts & EXT3_DEFM_DEBUG)
1676                 set_opt(sbi->s_mount_opt, DEBUG);
1677         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1678                 set_opt(sbi->s_mount_opt, GRPID);
1679         if (def_mount_opts & EXT3_DEFM_UID16)
1680                 set_opt(sbi->s_mount_opt, NO_UID32);
1681 #ifdef CONFIG_EXT3_FS_XATTR
1682         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1683                 set_opt(sbi->s_mount_opt, XATTR_USER);
1684 #endif
1685 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1686         if (def_mount_opts & EXT3_DEFM_ACL)
1687                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1688 #endif
1689         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1690                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1691         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1692                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
1693         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1694                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
1695
1696         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1697                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1698         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1699                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1700         else
1701                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1702
1703         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1704         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1705
1706         set_opt(sbi->s_mount_opt, RESERVATION);
1707
1708         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1709                             NULL, 0))
1710                 goto failed_mount;
1711
1712         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1713                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
1714
1715         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1716             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1717              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1718              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1719                 ext3_msg(sb, KERN_WARNING,
1720                         "warning: feature flags set on rev 0 fs, "
1721                         "running e2fsck is recommended");
1722         /*
1723          * Check feature flags regardless of the revision level, since we
1724          * previously didn't change the revision level when setting the flags,
1725          * so there is a chance incompat flags are set on a rev 0 filesystem.
1726          */
1727         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1728         if (features) {
1729                 ext3_msg(sb, KERN_ERR,
1730                         "error: couldn't mount because of unsupported "
1731                         "optional features (%x)", le32_to_cpu(features));
1732                 goto failed_mount;
1733         }
1734         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1735         if (!(sb->s_flags & MS_RDONLY) && features) {
1736                 ext3_msg(sb, KERN_ERR,
1737                         "error: couldn't mount RDWR because of unsupported "
1738                         "optional features (%x)", le32_to_cpu(features));
1739                 goto failed_mount;
1740         }
1741         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1742
1743         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1744             blocksize > EXT3_MAX_BLOCK_SIZE) {
1745                 ext3_msg(sb, KERN_ERR,
1746                         "error: couldn't mount because of unsupported "
1747                         "filesystem blocksize %d", blocksize);
1748                 goto failed_mount;
1749         }
1750
1751         hblock = bdev_logical_block_size(sb->s_bdev);
1752         if (sb->s_blocksize != blocksize) {
1753                 /*
1754                  * Make sure the blocksize for the filesystem is larger
1755                  * than the hardware sectorsize for the machine.
1756                  */
1757                 if (blocksize < hblock) {
1758                         ext3_msg(sb, KERN_ERR,
1759                                 "error: fsblocksize %d too small for "
1760                                 "hardware sectorsize %d", blocksize, hblock);
1761                         goto failed_mount;
1762                 }
1763
1764                 brelse (bh);
1765                 if (!sb_set_blocksize(sb, blocksize)) {
1766                         ext3_msg(sb, KERN_ERR,
1767                                 "error: bad blocksize %d", blocksize);
1768                         goto out_fail;
1769                 }
1770                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1771                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1772                 bh = sb_bread(sb, logic_sb_block);
1773                 if (!bh) {
1774                         ext3_msg(sb, KERN_ERR,
1775                                "error: can't read superblock on 2nd try");
1776                         goto failed_mount;
1777                 }
1778                 es = (struct ext3_super_block *)(bh->b_data + offset);
1779                 sbi->s_es = es;
1780                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1781                         ext3_msg(sb, KERN_ERR,
1782                                 "error: magic mismatch");
1783                         goto failed_mount;
1784                 }
1785         }
1786
1787         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1788
1789         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1790                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1791                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1792         } else {
1793                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1794                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1795                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1796                     (!is_power_of_2(sbi->s_inode_size)) ||
1797                     (sbi->s_inode_size > blocksize)) {
1798                         ext3_msg(sb, KERN_ERR,
1799                                 "error: unsupported inode size: %d",
1800                                 sbi->s_inode_size);
1801                         goto failed_mount;
1802                 }
1803         }
1804         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1805                                    le32_to_cpu(es->s_log_frag_size);
1806         if (blocksize != sbi->s_frag_size) {
1807                 ext3_msg(sb, KERN_ERR,
1808                        "error: fragsize %lu != blocksize %u (unsupported)",
1809                        sbi->s_frag_size, blocksize);
1810                 goto failed_mount;
1811         }
1812         sbi->s_frags_per_block = 1;
1813         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1814         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1815         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1816         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1817                 goto cantfind_ext3;
1818         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1819         if (sbi->s_inodes_per_block == 0)
1820                 goto cantfind_ext3;
1821         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1822                                         sbi->s_inodes_per_block;
1823         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1824         sbi->s_sbh = bh;
1825         sbi->s_mount_state = le16_to_cpu(es->s_state);
1826         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1827         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1828         for (i=0; i < 4; i++)
1829                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1830         sbi->s_def_hash_version = es->s_def_hash_version;
1831         i = le32_to_cpu(es->s_flags);
1832         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1833                 sbi->s_hash_unsigned = 3;
1834         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1835 #ifdef __CHAR_UNSIGNED__
1836                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1837                 sbi->s_hash_unsigned = 3;
1838 #else
1839                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1840 #endif
1841         }
1842
1843         if (sbi->s_blocks_per_group > blocksize * 8) {
1844                 ext3_msg(sb, KERN_ERR,
1845                         "#blocks per group too big: %lu",
1846                         sbi->s_blocks_per_group);
1847                 goto failed_mount;
1848         }
1849         if (sbi->s_frags_per_group > blocksize * 8) {
1850                 ext3_msg(sb, KERN_ERR,
1851                         "error: #fragments per group too big: %lu",
1852                         sbi->s_frags_per_group);
1853                 goto failed_mount;
1854         }
1855         if (sbi->s_inodes_per_group > blocksize * 8) {
1856                 ext3_msg(sb, KERN_ERR,
1857                         "error: #inodes per group too big: %lu",
1858                         sbi->s_inodes_per_group);
1859                 goto failed_mount;
1860         }
1861
1862         err = generic_check_addressable(sb->s_blocksize_bits,
1863                                         le32_to_cpu(es->s_blocks_count));
1864         if (err) {
1865                 ext3_msg(sb, KERN_ERR,
1866                         "error: filesystem is too large to mount safely");
1867                 if (sizeof(sector_t) < 8)
1868                         ext3_msg(sb, KERN_ERR,
1869                                 "error: CONFIG_LBDAF not enabled");
1870                 ret = err;
1871                 goto failed_mount;
1872         }
1873
1874         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1875                 goto cantfind_ext3;
1876         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1877                                le32_to_cpu(es->s_first_data_block) - 1)
1878                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1879         db_count = DIV_ROUND_UP(sbi->s_groups_count, EXT3_DESC_PER_BLOCK(sb));
1880         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1881                                     GFP_KERNEL);
1882         if (sbi->s_group_desc == NULL) {
1883                 ext3_msg(sb, KERN_ERR,
1884                         "error: not enough memory");
1885                 ret = -ENOMEM;
1886                 goto failed_mount;
1887         }
1888
1889         bgl_lock_init(sbi->s_blockgroup_lock);
1890
1891         for (i = 0; i < db_count; i++) {
1892                 block = descriptor_loc(sb, logic_sb_block, i);
1893                 sbi->s_group_desc[i] = sb_bread(sb, block);
1894                 if (!sbi->s_group_desc[i]) {
1895                         ext3_msg(sb, KERN_ERR,
1896                                 "error: can't read group descriptor %d", i);
1897                         db_count = i;
1898                         goto failed_mount2;
1899                 }
1900         }
1901         if (!ext3_check_descriptors (sb)) {
1902                 ext3_msg(sb, KERN_ERR,
1903                         "error: group descriptors corrupted");
1904                 goto failed_mount2;
1905         }
1906         sbi->s_gdb_count = db_count;
1907         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1908         spin_lock_init(&sbi->s_next_gen_lock);
1909
1910         /* per fileystem reservation list head & lock */
1911         spin_lock_init(&sbi->s_rsv_window_lock);
1912         sbi->s_rsv_window_root = RB_ROOT;
1913         /* Add a single, static dummy reservation to the start of the
1914          * reservation window list --- it gives us a placeholder for
1915          * append-at-start-of-list which makes the allocation logic
1916          * _much_ simpler. */
1917         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1918         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1919         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1920         sbi->s_rsv_window_head.rsv_goal_size = 0;
1921         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1922
1923         /*
1924          * set up enough so that it can read an inode
1925          */
1926         sb->s_op = &ext3_sops;
1927         sb->s_export_op = &ext3_export_ops;
1928         sb->s_xattr = ext3_xattr_handlers;
1929 #ifdef CONFIG_QUOTA
1930         sb->s_qcop = &ext3_qctl_operations;
1931         sb->dq_op = &ext3_quota_operations;
1932 #endif
1933         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1934         mutex_init(&sbi->s_orphan_lock);
1935         mutex_init(&sbi->s_resize_lock);
1936
1937         sb->s_root = NULL;
1938
1939         needs_recovery = (es->s_last_orphan != 0 ||
1940                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1941                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1942
1943         /*
1944          * The first inode we look at is the journal inode.  Don't try
1945          * root first: it may be modified in the journal!
1946          */
1947         if (!test_opt(sb, NOLOAD) &&
1948             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1949                 if (ext3_load_journal(sb, es, journal_devnum))
1950                         goto failed_mount2;
1951         } else if (journal_inum) {
1952                 if (ext3_create_journal(sb, es, journal_inum))
1953                         goto failed_mount2;
1954         } else {
1955                 if (!silent)
1956                         ext3_msg(sb, KERN_ERR,
1957                                 "error: no journal found. "
1958                                 "mounting ext3 over ext2?");
1959                 goto failed_mount2;
1960         }
1961         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1962                         ext3_count_free_blocks(sb));
1963         if (!err) {
1964                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1965                                 ext3_count_free_inodes(sb));
1966         }
1967         if (!err) {
1968                 err = percpu_counter_init(&sbi->s_dirs_counter,
1969                                 ext3_count_dirs(sb));
1970         }
1971         if (err) {
1972                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
1973                 ret = err;
1974                 goto failed_mount3;
1975         }
1976
1977         /* We have now updated the journal if required, so we can
1978          * validate the data journaling mode. */
1979         switch (test_opt(sb, DATA_FLAGS)) {
1980         case 0:
1981                 /* No mode set, assume a default based on the journal
1982                    capabilities: ORDERED_DATA if the journal can
1983                    cope, else JOURNAL_DATA */
1984                 if (journal_check_available_features
1985                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1986                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
1987                 else
1988                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1989                 break;
1990
1991         case EXT3_MOUNT_ORDERED_DATA:
1992         case EXT3_MOUNT_WRITEBACK_DATA:
1993                 if (!journal_check_available_features
1994                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1995                         ext3_msg(sb, KERN_ERR,
1996                                 "error: journal does not support "
1997                                 "requested data journaling mode");
1998                         goto failed_mount3;
1999                 }
2000         default:
2001                 break;
2002         }
2003
2004         /*
2005          * The journal_load will have done any necessary log recovery,
2006          * so we can safely mount the rest of the filesystem now.
2007          */
2008
2009         root = ext3_iget(sb, EXT3_ROOT_INO);
2010         if (IS_ERR(root)) {
2011                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
2012                 ret = PTR_ERR(root);
2013                 goto failed_mount3;
2014         }
2015         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2016                 iput(root);
2017                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
2018                 goto failed_mount3;
2019         }
2020         sb->s_root = d_alloc_root(root);
2021         if (!sb->s_root) {
2022                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
2023                 iput(root);
2024                 ret = -ENOMEM;
2025                 goto failed_mount3;
2026         }
2027
2028         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
2029
2030         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2031         ext3_orphan_cleanup(sb, es);
2032         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2033         if (needs_recovery)
2034                 ext3_msg(sb, KERN_INFO, "recovery complete");
2035         ext3_mark_recovery_complete(sb, es);
2036         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2037                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2038                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2039                 "writeback");
2040
2041         return 0;
2042
2043 cantfind_ext3:
2044         if (!silent)
2045                 ext3_msg(sb, KERN_INFO,
2046                         "error: can't find ext3 filesystem on dev %s.",
2047                        sb->s_id);
2048         goto failed_mount;
2049
2050 failed_mount3:
2051         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2052         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2053         percpu_counter_destroy(&sbi->s_dirs_counter);
2054         journal_destroy(sbi->s_journal);
2055 failed_mount2:
2056         for (i = 0; i < db_count; i++)
2057                 brelse(sbi->s_group_desc[i]);
2058         kfree(sbi->s_group_desc);
2059 failed_mount:
2060 #ifdef CONFIG_QUOTA
2061         for (i = 0; i < MAXQUOTAS; i++)
2062                 kfree(sbi->s_qf_names[i]);
2063 #endif
2064         ext3_blkdev_remove(sbi);
2065         brelse(bh);
2066 out_fail:
2067         sb->s_fs_info = NULL;
2068         kfree(sbi->s_blockgroup_lock);
2069         kfree(sbi);
2070         return ret;
2071 }
2072
2073 /*
2074  * Setup any per-fs journal parameters now.  We'll do this both on
2075  * initial mount, once the journal has been initialised but before we've
2076  * done any recovery; and again on any subsequent remount.
2077  */
2078 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2079 {
2080         struct ext3_sb_info *sbi = EXT3_SB(sb);
2081
2082         if (sbi->s_commit_interval)
2083                 journal->j_commit_interval = sbi->s_commit_interval;
2084         /* We could also set up an ext3-specific default for the commit
2085          * interval here, but for now we'll just fall back to the jbd
2086          * default. */
2087
2088         spin_lock(&journal->j_state_lock);
2089         if (test_opt(sb, BARRIER))
2090                 journal->j_flags |= JFS_BARRIER;
2091         else
2092                 journal->j_flags &= ~JFS_BARRIER;
2093         if (test_opt(sb, DATA_ERR_ABORT))
2094                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2095         else
2096                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2097         spin_unlock(&journal->j_state_lock);
2098 }
2099
2100 static journal_t *ext3_get_journal(struct super_block *sb,
2101                                    unsigned int journal_inum)
2102 {
2103         struct inode *journal_inode;
2104         journal_t *journal;
2105
2106         /* First, test for the existence of a valid inode on disk.  Bad
2107          * things happen if we iget() an unused inode, as the subsequent
2108          * iput() will try to delete it. */
2109
2110         journal_inode = ext3_iget(sb, journal_inum);
2111         if (IS_ERR(journal_inode)) {
2112                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2113                 return NULL;
2114         }
2115         if (!journal_inode->i_nlink) {
2116                 make_bad_inode(journal_inode);
2117                 iput(journal_inode);
2118                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2119                 return NULL;
2120         }
2121
2122         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2123                   journal_inode, journal_inode->i_size);
2124         if (!S_ISREG(journal_inode->i_mode)) {
2125                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2126                 iput(journal_inode);
2127                 return NULL;
2128         }
2129
2130         journal = journal_init_inode(journal_inode);
2131         if (!journal) {
2132                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2133                 iput(journal_inode);
2134                 return NULL;
2135         }
2136         journal->j_private = sb;
2137         ext3_init_journal_params(sb, journal);
2138         return journal;
2139 }
2140
2141 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2142                                        dev_t j_dev)
2143 {
2144         struct buffer_head * bh;
2145         journal_t *journal;
2146         ext3_fsblk_t start;
2147         ext3_fsblk_t len;
2148         int hblock, blocksize;
2149         ext3_fsblk_t sb_block;
2150         unsigned long offset;
2151         struct ext3_super_block * es;
2152         struct block_device *bdev;
2153
2154         bdev = ext3_blkdev_get(j_dev, sb);
2155         if (bdev == NULL)
2156                 return NULL;
2157
2158         if (bd_claim(bdev, sb)) {
2159                 ext3_msg(sb, KERN_ERR,
2160                         "error: failed to claim external journal device");
2161                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2162                 return NULL;
2163         }
2164
2165         blocksize = sb->s_blocksize;
2166         hblock = bdev_logical_block_size(bdev);
2167         if (blocksize < hblock) {
2168                 ext3_msg(sb, KERN_ERR,
2169                         "error: blocksize too small for journal device");
2170                 goto out_bdev;
2171         }
2172
2173         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2174         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2175         set_blocksize(bdev, blocksize);
2176         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2177                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2178                         "external journal");
2179                 goto out_bdev;
2180         }
2181
2182         es = (struct ext3_super_block *) (bh->b_data + offset);
2183         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2184             !(le32_to_cpu(es->s_feature_incompat) &
2185               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2186                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2187                         "bad superblock");
2188                 brelse(bh);
2189                 goto out_bdev;
2190         }
2191
2192         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2193                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2194                 brelse(bh);
2195                 goto out_bdev;
2196         }
2197
2198         len = le32_to_cpu(es->s_blocks_count);
2199         start = sb_block + 1;
2200         brelse(bh);     /* we're done with the superblock */
2201
2202         journal = journal_init_dev(bdev, sb->s_bdev,
2203                                         start, len, blocksize);
2204         if (!journal) {
2205                 ext3_msg(sb, KERN_ERR,
2206                         "error: failed to create device journal");
2207                 goto out_bdev;
2208         }
2209         journal->j_private = sb;
2210         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2211         wait_on_buffer(journal->j_sb_buffer);
2212         if (!buffer_uptodate(journal->j_sb_buffer)) {
2213                 ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2214                 goto out_journal;
2215         }
2216         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2217                 ext3_msg(sb, KERN_ERR,
2218                         "error: external journal has more than one "
2219                         "user (unsupported) - %d",
2220                         be32_to_cpu(journal->j_superblock->s_nr_users));
2221                 goto out_journal;
2222         }
2223         EXT3_SB(sb)->journal_bdev = bdev;
2224         ext3_init_journal_params(sb, journal);
2225         return journal;
2226 out_journal:
2227         journal_destroy(journal);
2228 out_bdev:
2229         ext3_blkdev_put(bdev);
2230         return NULL;
2231 }
2232
2233 static int ext3_load_journal(struct super_block *sb,
2234                              struct ext3_super_block *es,
2235                              unsigned long journal_devnum)
2236 {
2237         journal_t *journal;
2238         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2239         dev_t journal_dev;
2240         int err = 0;
2241         int really_read_only;
2242
2243         if (journal_devnum &&
2244             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2245                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2246                         "numbers have changed");
2247                 journal_dev = new_decode_dev(journal_devnum);
2248         } else
2249                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2250
2251         really_read_only = bdev_read_only(sb->s_bdev);
2252
2253         /*
2254          * Are we loading a blank journal or performing recovery after a
2255          * crash?  For recovery, we need to check in advance whether we
2256          * can get read-write access to the device.
2257          */
2258
2259         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2260                 if (sb->s_flags & MS_RDONLY) {
2261                         ext3_msg(sb, KERN_INFO,
2262                                 "recovery required on readonly filesystem");
2263                         if (really_read_only) {
2264                                 ext3_msg(sb, KERN_ERR, "error: write access "
2265                                         "unavailable, cannot proceed");
2266                                 return -EROFS;
2267                         }
2268                         ext3_msg(sb, KERN_INFO,
2269                                 "write access will be enabled during recovery");
2270                 }
2271         }
2272
2273         if (journal_inum && journal_dev) {
2274                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2275                        "and inode journals");
2276                 return -EINVAL;
2277         }
2278
2279         if (journal_inum) {
2280                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2281                         return -EINVAL;
2282         } else {
2283                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2284                         return -EINVAL;
2285         }
2286
2287         if (!(journal->j_flags & JFS_BARRIER))
2288                 printk(KERN_INFO "EXT3-fs: barriers not enabled\n");
2289
2290         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2291                 err = journal_update_format(journal);
2292                 if (err)  {
2293                         ext3_msg(sb, KERN_ERR, "error updating journal");
2294                         journal_destroy(journal);
2295                         return err;
2296                 }
2297         }
2298
2299         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2300                 err = journal_wipe(journal, !really_read_only);
2301         if (!err)
2302                 err = journal_load(journal);
2303
2304         if (err) {
2305                 ext3_msg(sb, KERN_ERR, "error loading journal");
2306                 journal_destroy(journal);
2307                 return err;
2308         }
2309
2310         EXT3_SB(sb)->s_journal = journal;
2311         ext3_clear_journal_err(sb, es);
2312
2313         if (!really_read_only && journal_devnum &&
2314             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2315                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2316
2317                 /* Make sure we flush the recovery flag to disk. */
2318                 ext3_commit_super(sb, es, 1);
2319         }
2320
2321         return 0;
2322 }
2323
2324 static int ext3_create_journal(struct super_block *sb,
2325                                struct ext3_super_block *es,
2326                                unsigned int journal_inum)
2327 {
2328         journal_t *journal;
2329         int err;
2330
2331         if (sb->s_flags & MS_RDONLY) {
2332                 ext3_msg(sb, KERN_ERR,
2333                         "error: readonly filesystem when trying to "
2334                         "create journal");
2335                 return -EROFS;
2336         }
2337
2338         journal = ext3_get_journal(sb, journal_inum);
2339         if (!journal)
2340                 return -EINVAL;
2341
2342         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2343                journal_inum);
2344
2345         err = journal_create(journal);
2346         if (err) {
2347                 ext3_msg(sb, KERN_ERR, "error creating journal");
2348                 journal_destroy(journal);
2349                 return -EIO;
2350         }
2351
2352         EXT3_SB(sb)->s_journal = journal;
2353
2354         ext3_update_dynamic_rev(sb);
2355         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2356         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2357
2358         es->s_journal_inum = cpu_to_le32(journal_inum);
2359
2360         /* Make sure we flush the recovery flag to disk. */
2361         ext3_commit_super(sb, es, 1);
2362
2363         return 0;
2364 }
2365
2366 static int ext3_commit_super(struct super_block *sb,
2367                                struct ext3_super_block *es,
2368                                int sync)
2369 {
2370         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2371         int error = 0;
2372
2373         if (!sbh)
2374                 return error;
2375
2376         if (buffer_write_io_error(sbh)) {
2377                 /*
2378                  * Oh, dear.  A previous attempt to write the
2379                  * superblock failed.  This could happen because the
2380                  * USB device was yanked out.  Or it could happen to
2381                  * be a transient write error and maybe the block will
2382                  * be remapped.  Nothing we can do but to retry the
2383                  * write and hope for the best.
2384                  */
2385                 ext3_msg(sb, KERN_ERR, "previous I/O error to "
2386                        "superblock detected");
2387                 clear_buffer_write_io_error(sbh);
2388                 set_buffer_uptodate(sbh);
2389         }
2390         /*
2391          * If the file system is mounted read-only, don't update the
2392          * superblock write time.  This avoids updating the superblock
2393          * write time when we are mounting the root file system
2394          * read/only but we need to replay the journal; at that point,
2395          * for people who are east of GMT and who make their clock
2396          * tick in localtime for Windows bug-for-bug compatibility,
2397          * the clock is set in the future, and this will cause e2fsck
2398          * to complain and force a full file system check.
2399          */
2400         if (!(sb->s_flags & MS_RDONLY))
2401                 es->s_wtime = cpu_to_le32(get_seconds());
2402         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2403         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2404         BUFFER_TRACE(sbh, "marking dirty");
2405         mark_buffer_dirty(sbh);
2406         if (sync) {
2407                 error = sync_dirty_buffer(sbh);
2408                 if (buffer_write_io_error(sbh)) {
2409                         ext3_msg(sb, KERN_ERR, "I/O error while writing "
2410                                "superblock");
2411                         clear_buffer_write_io_error(sbh);
2412                         set_buffer_uptodate(sbh);
2413                 }
2414         }
2415         return error;
2416 }
2417
2418
2419 /*
2420  * Have we just finished recovery?  If so, and if we are mounting (or
2421  * remounting) the filesystem readonly, then we will end up with a
2422  * consistent fs on disk.  Record that fact.
2423  */
2424 static void ext3_mark_recovery_complete(struct super_block * sb,
2425                                         struct ext3_super_block * es)
2426 {
2427         journal_t *journal = EXT3_SB(sb)->s_journal;
2428
2429         journal_lock_updates(journal);
2430         if (journal_flush(journal) < 0)
2431                 goto out;
2432
2433         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2434             sb->s_flags & MS_RDONLY) {
2435                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2436                 ext3_commit_super(sb, es, 1);
2437         }
2438
2439 out:
2440         journal_unlock_updates(journal);
2441 }
2442
2443 /*
2444  * If we are mounting (or read-write remounting) a filesystem whose journal
2445  * has recorded an error from a previous lifetime, move that error to the
2446  * main filesystem now.
2447  */
2448 static void ext3_clear_journal_err(struct super_block *sb,
2449                                    struct ext3_super_block *es)
2450 {
2451         journal_t *journal;
2452         int j_errno;
2453         const char *errstr;
2454
2455         journal = EXT3_SB(sb)->s_journal;
2456
2457         /*
2458          * Now check for any error status which may have been recorded in the
2459          * journal by a prior ext3_error() or ext3_abort()
2460          */
2461
2462         j_errno = journal_errno(journal);
2463         if (j_errno) {
2464                 char nbuf[16];
2465
2466                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2467                 ext3_warning(sb, __func__, "Filesystem error recorded "
2468                              "from previous mount: %s", errstr);
2469                 ext3_warning(sb, __func__, "Marking fs in need of "
2470                              "filesystem check.");
2471
2472                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2473                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2474                 ext3_commit_super (sb, es, 1);
2475
2476                 journal_clear_err(journal);
2477         }
2478 }
2479
2480 /*
2481  * Force the running and committing transactions to commit,
2482  * and wait on the commit.
2483  */
2484 int ext3_force_commit(struct super_block *sb)
2485 {
2486         journal_t *journal;
2487         int ret;
2488
2489         if (sb->s_flags & MS_RDONLY)
2490                 return 0;
2491
2492         journal = EXT3_SB(sb)->s_journal;
2493         ret = ext3_journal_force_commit(journal);
2494         return ret;
2495 }
2496
2497 static int ext3_sync_fs(struct super_block *sb, int wait)
2498 {
2499         tid_t target;
2500
2501         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2502                 if (wait)
2503                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2504         }
2505         return 0;
2506 }
2507
2508 /*
2509  * LVM calls this function before a (read-only) snapshot is created.  This
2510  * gives us a chance to flush the journal completely and mark the fs clean.
2511  */
2512 static int ext3_freeze(struct super_block *sb)
2513 {
2514         int error = 0;
2515         journal_t *journal;
2516
2517         if (!(sb->s_flags & MS_RDONLY)) {
2518                 journal = EXT3_SB(sb)->s_journal;
2519
2520                 /* Now we set up the journal barrier. */
2521                 journal_lock_updates(journal);
2522
2523                 /*
2524                  * We don't want to clear needs_recovery flag when we failed
2525                  * to flush the journal.
2526                  */
2527                 error = journal_flush(journal);
2528                 if (error < 0)
2529                         goto out;
2530
2531                 /* Journal blocked and flushed, clear needs_recovery flag. */
2532                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2533                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2534                 if (error)
2535                         goto out;
2536         }
2537         return 0;
2538
2539 out:
2540         journal_unlock_updates(journal);
2541         return error;
2542 }
2543
2544 /*
2545  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2546  * flag here, even though the filesystem is not technically dirty yet.
2547  */
2548 static int ext3_unfreeze(struct super_block *sb)
2549 {
2550         if (!(sb->s_flags & MS_RDONLY)) {
2551                 lock_super(sb);
2552                 /* Reser the needs_recovery flag before the fs is unlocked. */
2553                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2554                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2555                 unlock_super(sb);
2556                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2557         }
2558         return 0;
2559 }
2560
2561 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2562 {
2563         struct ext3_super_block * es;
2564         struct ext3_sb_info *sbi = EXT3_SB(sb);
2565         ext3_fsblk_t n_blocks_count = 0;
2566         unsigned long old_sb_flags;
2567         struct ext3_mount_options old_opts;
2568         int enable_quota = 0;
2569         int err;
2570 #ifdef CONFIG_QUOTA
2571         int i;
2572 #endif
2573
2574         /* Store the original options */
2575         lock_super(sb);
2576         old_sb_flags = sb->s_flags;
2577         old_opts.s_mount_opt = sbi->s_mount_opt;
2578         old_opts.s_resuid = sbi->s_resuid;
2579         old_opts.s_resgid = sbi->s_resgid;
2580         old_opts.s_commit_interval = sbi->s_commit_interval;
2581 #ifdef CONFIG_QUOTA
2582         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2583         for (i = 0; i < MAXQUOTAS; i++)
2584                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2585 #endif
2586
2587         /*
2588          * Allow the "check" option to be passed as a remount option.
2589          */
2590         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2591                 err = -EINVAL;
2592                 goto restore_opts;
2593         }
2594
2595         if (test_opt(sb, ABORT))
2596                 ext3_abort(sb, __func__, "Abort forced by user");
2597
2598         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2599                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2600
2601         es = sbi->s_es;
2602
2603         ext3_init_journal_params(sb, sbi->s_journal);
2604
2605         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2606                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2607                 if (test_opt(sb, ABORT)) {
2608                         err = -EROFS;
2609                         goto restore_opts;
2610                 }
2611
2612                 if (*flags & MS_RDONLY) {
2613                         err = dquot_suspend(sb, -1);
2614                         if (err < 0)
2615                                 goto restore_opts;
2616
2617                         /*
2618                          * First of all, the unconditional stuff we have to do
2619                          * to disable replay of the journal when we next remount
2620                          */
2621                         sb->s_flags |= MS_RDONLY;
2622
2623                         /*
2624                          * OK, test if we are remounting a valid rw partition
2625                          * readonly, and if so set the rdonly flag and then
2626                          * mark the partition as valid again.
2627                          */
2628                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2629                             (sbi->s_mount_state & EXT3_VALID_FS))
2630                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2631
2632                         ext3_mark_recovery_complete(sb, es);
2633                 } else {
2634                         __le32 ret;
2635                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2636                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2637                                 ext3_msg(sb, KERN_WARNING,
2638                                         "warning: couldn't remount RDWR "
2639                                         "because of unsupported optional "
2640                                         "features (%x)", le32_to_cpu(ret));
2641                                 err = -EROFS;
2642                                 goto restore_opts;
2643                         }
2644
2645                         /*
2646                          * If we have an unprocessed orphan list hanging
2647                          * around from a previously readonly bdev mount,
2648                          * require a full umount/remount for now.
2649                          */
2650                         if (es->s_last_orphan) {
2651                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2652                                        "remount RDWR because of unprocessed "
2653                                        "orphan inode list.  Please "
2654                                        "umount/remount instead.");
2655                                 err = -EINVAL;
2656                                 goto restore_opts;
2657                         }
2658
2659                         /*
2660                          * Mounting a RDONLY partition read-write, so reread
2661                          * and store the current valid flag.  (It may have
2662                          * been changed by e2fsck since we originally mounted
2663                          * the partition.)
2664                          */
2665                         ext3_clear_journal_err(sb, es);
2666                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2667                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2668                                 goto restore_opts;
2669                         if (!ext3_setup_super (sb, es, 0))
2670                                 sb->s_flags &= ~MS_RDONLY;
2671                         enable_quota = 1;
2672                 }
2673         }
2674 #ifdef CONFIG_QUOTA
2675         /* Release old quota file names */
2676         for (i = 0; i < MAXQUOTAS; i++)
2677                 if (old_opts.s_qf_names[i] &&
2678                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2679                         kfree(old_opts.s_qf_names[i]);
2680 #endif
2681         unlock_super(sb);
2682
2683         if (enable_quota)
2684                 dquot_resume(sb, -1);
2685         return 0;
2686 restore_opts:
2687         sb->s_flags = old_sb_flags;
2688         sbi->s_mount_opt = old_opts.s_mount_opt;
2689         sbi->s_resuid = old_opts.s_resuid;
2690         sbi->s_resgid = old_opts.s_resgid;
2691         sbi->s_commit_interval = old_opts.s_commit_interval;
2692 #ifdef CONFIG_QUOTA
2693         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2694         for (i = 0; i < MAXQUOTAS; i++) {
2695                 if (sbi->s_qf_names[i] &&
2696                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2697                         kfree(sbi->s_qf_names[i]);
2698                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2699         }
2700 #endif
2701         unlock_super(sb);
2702         return err;
2703 }
2704
2705 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2706 {
2707         struct super_block *sb = dentry->d_sb;
2708         struct ext3_sb_info *sbi = EXT3_SB(sb);
2709         struct ext3_super_block *es = sbi->s_es;
2710         u64 fsid;
2711
2712         if (test_opt(sb, MINIX_DF)) {
2713                 sbi->s_overhead_last = 0;
2714         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2715                 unsigned long ngroups = sbi->s_groups_count, i;
2716                 ext3_fsblk_t overhead = 0;
2717                 smp_rmb();
2718
2719                 /*
2720                  * Compute the overhead (FS structures).  This is constant
2721                  * for a given filesystem unless the number of block groups
2722                  * changes so we cache the previous value until it does.
2723                  */
2724
2725                 /*
2726                  * All of the blocks before first_data_block are
2727                  * overhead
2728                  */
2729                 overhead = le32_to_cpu(es->s_first_data_block);
2730
2731                 /*
2732                  * Add the overhead attributed to the superblock and
2733                  * block group descriptors.  If the sparse superblocks
2734                  * feature is turned on, then not all groups have this.
2735                  */
2736                 for (i = 0; i < ngroups; i++) {
2737                         overhead += ext3_bg_has_super(sb, i) +
2738                                 ext3_bg_num_gdb(sb, i);
2739                         cond_resched();
2740                 }
2741
2742                 /*
2743                  * Every block group has an inode bitmap, a block
2744                  * bitmap, and an inode table.
2745                  */
2746                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2747                 sbi->s_overhead_last = overhead;
2748                 smp_wmb();
2749                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2750         }
2751
2752         buf->f_type = EXT3_SUPER_MAGIC;
2753         buf->f_bsize = sb->s_blocksize;
2754         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2755         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2756         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2757         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2758                 buf->f_bavail = 0;
2759         buf->f_files = le32_to_cpu(es->s_inodes_count);
2760         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2761         buf->f_namelen = EXT3_NAME_LEN;
2762         fsid = le64_to_cpup((void *)es->s_uuid) ^
2763                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2764         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2765         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2766         return 0;
2767 }
2768
2769 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2770  * is locked for write. Otherwise the are possible deadlocks:
2771  * Process 1                         Process 2
2772  * ext3_create()                     quota_sync()
2773  *   journal_start()                   write_dquot()
2774  *   dquot_initialize()                       down(dqio_mutex)
2775  *     down(dqio_mutex)                    journal_start()
2776  *
2777  */
2778
2779 #ifdef CONFIG_QUOTA
2780
2781 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2782 {
2783         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2784 }
2785
2786 static int ext3_write_dquot(struct dquot *dquot)
2787 {
2788         int ret, err;
2789         handle_t *handle;
2790         struct inode *inode;
2791
2792         inode = dquot_to_inode(dquot);
2793         handle = ext3_journal_start(inode,
2794                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2795         if (IS_ERR(handle))
2796                 return PTR_ERR(handle);
2797         ret = dquot_commit(dquot);
2798         err = ext3_journal_stop(handle);
2799         if (!ret)
2800                 ret = err;
2801         return ret;
2802 }
2803
2804 static int ext3_acquire_dquot(struct dquot *dquot)
2805 {
2806         int ret, err;
2807         handle_t *handle;
2808
2809         handle = ext3_journal_start(dquot_to_inode(dquot),
2810                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2811         if (IS_ERR(handle))
2812                 return PTR_ERR(handle);
2813         ret = dquot_acquire(dquot);
2814         err = ext3_journal_stop(handle);
2815         if (!ret)
2816                 ret = err;
2817         return ret;
2818 }
2819
2820 static int ext3_release_dquot(struct dquot *dquot)
2821 {
2822         int ret, err;
2823         handle_t *handle;
2824
2825         handle = ext3_journal_start(dquot_to_inode(dquot),
2826                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2827         if (IS_ERR(handle)) {
2828                 /* Release dquot anyway to avoid endless cycle in dqput() */
2829                 dquot_release(dquot);
2830                 return PTR_ERR(handle);
2831         }
2832         ret = dquot_release(dquot);
2833         err = ext3_journal_stop(handle);
2834         if (!ret)
2835                 ret = err;
2836         return ret;
2837 }
2838
2839 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2840 {
2841         /* Are we journaling quotas? */
2842         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2843             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2844                 dquot_mark_dquot_dirty(dquot);
2845                 return ext3_write_dquot(dquot);
2846         } else {
2847                 return dquot_mark_dquot_dirty(dquot);
2848         }
2849 }
2850
2851 static int ext3_write_info(struct super_block *sb, int type)
2852 {
2853         int ret, err;
2854         handle_t *handle;
2855
2856         /* Data block + inode block */
2857         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2858         if (IS_ERR(handle))
2859                 return PTR_ERR(handle);
2860         ret = dquot_commit_info(sb, type);
2861         err = ext3_journal_stop(handle);
2862         if (!ret)
2863                 ret = err;
2864         return ret;
2865 }
2866
2867 /*
2868  * Turn on quotas during mount time - we need to find
2869  * the quota file and such...
2870  */
2871 static int ext3_quota_on_mount(struct super_block *sb, int type)
2872 {
2873         return dquot_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2874                                         EXT3_SB(sb)->s_jquota_fmt, type);
2875 }
2876
2877 /*
2878  * Standard function to be called on quota_on
2879  */
2880 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2881                          char *name)
2882 {
2883         int err;
2884         struct path path;
2885
2886         if (!test_opt(sb, QUOTA))
2887                 return -EINVAL;
2888
2889         err = kern_path(name, LOOKUP_FOLLOW, &path);
2890         if (err)
2891                 return err;
2892
2893         /* Quotafile not on the same filesystem? */
2894         if (path.mnt->mnt_sb != sb) {
2895                 path_put(&path);
2896                 return -EXDEV;
2897         }
2898         /* Journaling quota? */
2899         if (EXT3_SB(sb)->s_qf_names[type]) {
2900                 /* Quotafile not of fs root? */
2901                 if (path.dentry->d_parent != sb->s_root)
2902                         ext3_msg(sb, KERN_WARNING,
2903                                 "warning: Quota file not on filesystem root. "
2904                                 "Journaled quota will not work.");
2905         }
2906
2907         /*
2908          * When we journal data on quota file, we have to flush journal to see
2909          * all updates to the file when we bypass pagecache...
2910          */
2911         if (ext3_should_journal_data(path.dentry->d_inode)) {
2912                 /*
2913                  * We don't need to lock updates but journal_flush() could
2914                  * otherwise be livelocked...
2915                  */
2916                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2917                 err = journal_flush(EXT3_SB(sb)->s_journal);
2918                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2919                 if (err) {
2920                         path_put(&path);
2921                         return err;
2922                 }
2923         }
2924
2925         err = dquot_quota_on_path(sb, type, format_id, &path);
2926         path_put(&path);
2927         return err;
2928 }
2929
2930 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2931  * acquiring the locks... As quota files are never truncated and quota code
2932  * itself serializes the operations (and noone else should touch the files)
2933  * we don't have to be afraid of races */
2934 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2935                                size_t len, loff_t off)
2936 {
2937         struct inode *inode = sb_dqopt(sb)->files[type];
2938         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2939         int err = 0;
2940         int offset = off & (sb->s_blocksize - 1);
2941         int tocopy;
2942         size_t toread;
2943         struct buffer_head *bh;
2944         loff_t i_size = i_size_read(inode);
2945
2946         if (off > i_size)
2947                 return 0;
2948         if (off+len > i_size)
2949                 len = i_size-off;
2950         toread = len;
2951         while (toread > 0) {
2952                 tocopy = sb->s_blocksize - offset < toread ?
2953                                 sb->s_blocksize - offset : toread;
2954                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2955                 if (err)
2956                         return err;
2957                 if (!bh)        /* A hole? */
2958                         memset(data, 0, tocopy);
2959                 else
2960                         memcpy(data, bh->b_data+offset, tocopy);
2961                 brelse(bh);
2962                 offset = 0;
2963                 toread -= tocopy;
2964                 data += tocopy;
2965                 blk++;
2966         }
2967         return len;
2968 }
2969
2970 /* Write to quotafile (we know the transaction is already started and has
2971  * enough credits) */
2972 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2973                                 const char *data, size_t len, loff_t off)
2974 {
2975         struct inode *inode = sb_dqopt(sb)->files[type];
2976         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2977         int err = 0;
2978         int offset = off & (sb->s_blocksize - 1);
2979         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2980         struct buffer_head *bh;
2981         handle_t *handle = journal_current_handle();
2982
2983         if (!handle) {
2984                 ext3_msg(sb, KERN_WARNING,
2985                         "warning: quota write (off=%llu, len=%llu)"
2986                         " cancelled because transaction is not started.",
2987                         (unsigned long long)off, (unsigned long long)len);
2988                 return -EIO;
2989         }
2990
2991         /*
2992          * Since we account only one data block in transaction credits,
2993          * then it is impossible to cross a block boundary.
2994          */
2995         if (sb->s_blocksize - offset < len) {
2996                 ext3_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
2997                         " cancelled because not block aligned",
2998                         (unsigned long long)off, (unsigned long long)len);
2999                 return -EIO;
3000         }
3001         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3002         bh = ext3_bread(handle, inode, blk, 1, &err);
3003         if (!bh)
3004                 goto out;
3005         if (journal_quota) {
3006                 err = ext3_journal_get_write_access(handle, bh);
3007                 if (err) {
3008                         brelse(bh);
3009                         goto out;
3010                 }
3011         }
3012         lock_buffer(bh);
3013         memcpy(bh->b_data+offset, data, len);
3014         flush_dcache_page(bh->b_page);
3015         unlock_buffer(bh);
3016         if (journal_quota)
3017                 err = ext3_journal_dirty_metadata(handle, bh);
3018         else {
3019                 /* Always do at least ordered writes for quotas */
3020                 err = ext3_journal_dirty_data(handle, bh);
3021                 mark_buffer_dirty(bh);
3022         }
3023         brelse(bh);
3024 out:
3025         if (err) {
3026                 mutex_unlock(&inode->i_mutex);
3027                 return err;
3028         }
3029         if (inode->i_size < off + len) {
3030                 i_size_write(inode, off + len);
3031                 EXT3_I(inode)->i_disksize = inode->i_size;
3032         }
3033         inode->i_version++;
3034         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3035         ext3_mark_inode_dirty(handle, inode);
3036         mutex_unlock(&inode->i_mutex);
3037         return len;
3038 }
3039
3040 #endif
3041
3042 static struct dentry *ext3_mount(struct file_system_type *fs_type,
3043         int flags, const char *dev_name, void *data)
3044 {
3045         return mount_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
3046 }
3047
3048 static struct file_system_type ext3_fs_type = {
3049         .owner          = THIS_MODULE,
3050         .name           = "ext3",
3051         .mount          = ext3_mount,
3052         .kill_sb        = kill_block_super,
3053         .fs_flags       = FS_REQUIRES_DEV,
3054 };
3055
3056 static int __init init_ext3_fs(void)
3057 {
3058         int err = init_ext3_xattr();
3059         if (err)
3060                 return err;
3061         err = init_inodecache();
3062         if (err)
3063                 goto out1;
3064         err = register_filesystem(&ext3_fs_type);
3065         if (err)
3066                 goto out;
3067         return 0;
3068 out:
3069         destroy_inodecache();
3070 out1:
3071         exit_ext3_xattr();
3072         return err;
3073 }
3074
3075 static void __exit exit_ext3_fs(void)
3076 {
3077         unregister_filesystem(&ext3_fs_type);
3078         destroy_inodecache();
3079         exit_ext3_xattr();
3080 }
3081
3082 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3083 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3084 MODULE_LICENSE("GPL");
3085 module_init(init_ext3_fs)
3086 module_exit(exit_ext3_fs)