ext3: Unify log messages in ext3
[linux-2.6.git] / fs / ext3 / super.c
1 /*
2  *  linux/fs/ext3/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/module.h>
20 #include <linux/string.h>
21 #include <linux/fs.h>
22 #include <linux/time.h>
23 #include <linux/jbd.h>
24 #include <linux/ext3_fs.h>
25 #include <linux/ext3_jbd.h>
26 #include <linux/slab.h>
27 #include <linux/init.h>
28 #include <linux/blkdev.h>
29 #include <linux/parser.h>
30 #include <linux/smp_lock.h>
31 #include <linux/buffer_head.h>
32 #include <linux/exportfs.h>
33 #include <linux/vfs.h>
34 #include <linux/random.h>
35 #include <linux/mount.h>
36 #include <linux/namei.h>
37 #include <linux/quotaops.h>
38 #include <linux/seq_file.h>
39 #include <linux/log2.h>
40
41 #include <asm/uaccess.h>
42
43 #include "xattr.h"
44 #include "acl.h"
45 #include "namei.h"
46
47 #ifdef CONFIG_EXT3_DEFAULTS_TO_ORDERED
48   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_ORDERED_DATA
49 #else
50   #define EXT3_MOUNT_DEFAULT_DATA_MODE EXT3_MOUNT_WRITEBACK_DATA
51 #endif
52
53 static int ext3_load_journal(struct super_block *, struct ext3_super_block *,
54                              unsigned long journal_devnum);
55 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
56                                unsigned int);
57 static int ext3_commit_super(struct super_block *sb,
58                                struct ext3_super_block *es,
59                                int sync);
60 static void ext3_mark_recovery_complete(struct super_block * sb,
61                                         struct ext3_super_block * es);
62 static void ext3_clear_journal_err(struct super_block * sb,
63                                    struct ext3_super_block * es);
64 static int ext3_sync_fs(struct super_block *sb, int wait);
65 static const char *ext3_decode_error(struct super_block * sb, int errno,
66                                      char nbuf[16]);
67 static int ext3_remount (struct super_block * sb, int * flags, char * data);
68 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf);
69 static int ext3_unfreeze(struct super_block *sb);
70 static int ext3_freeze(struct super_block *sb);
71
72 /*
73  * Wrappers for journal_start/end.
74  *
75  * The only special thing we need to do here is to make sure that all
76  * journal_end calls result in the superblock being marked dirty, so
77  * that sync() will call the filesystem's write_super callback if
78  * appropriate.
79  */
80 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
81 {
82         journal_t *journal;
83
84         if (sb->s_flags & MS_RDONLY)
85                 return ERR_PTR(-EROFS);
86
87         /* Special case here: if the journal has aborted behind our
88          * backs (eg. EIO in the commit thread), then we still need to
89          * take the FS itself readonly cleanly. */
90         journal = EXT3_SB(sb)->s_journal;
91         if (is_journal_aborted(journal)) {
92                 ext3_abort(sb, __func__,
93                            "Detected aborted journal");
94                 return ERR_PTR(-EROFS);
95         }
96
97         return journal_start(journal, nblocks);
98 }
99
100 /*
101  * The only special thing we need to do here is to make sure that all
102  * journal_stop calls result in the superblock being marked dirty, so
103  * that sync() will call the filesystem's write_super callback if
104  * appropriate.
105  */
106 int __ext3_journal_stop(const char *where, handle_t *handle)
107 {
108         struct super_block *sb;
109         int err;
110         int rc;
111
112         sb = handle->h_transaction->t_journal->j_private;
113         err = handle->h_err;
114         rc = journal_stop(handle);
115
116         if (!err)
117                 err = rc;
118         if (err)
119                 __ext3_std_error(sb, where, err);
120         return err;
121 }
122
123 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
124                 struct buffer_head *bh, handle_t *handle, int err)
125 {
126         char nbuf[16];
127         const char *errstr = ext3_decode_error(NULL, err, nbuf);
128
129         if (bh)
130                 BUFFER_TRACE(bh, "abort");
131
132         if (!handle->h_err)
133                 handle->h_err = err;
134
135         if (is_handle_aborted(handle))
136                 return;
137
138         printk(KERN_ERR "EXT3-fs: %s: aborting transaction: %s in %s\n",
139                 caller, errstr, err_fn);
140
141         journal_abort_handle(handle);
142 }
143
144 void ext3_msg(struct super_block *sb, const char *prefix,
145                 const char *fmt, ...)
146 {
147         va_list args;
148
149         va_start(args, fmt);
150         printk("%sEXT3-fs (%s): ", prefix, sb->s_id);
151         vprintk(fmt, args);
152         printk("\n");
153         va_end(args);
154 }
155
156 /* Deal with the reporting of failure conditions on a filesystem such as
157  * inconsistencies detected or read IO failures.
158  *
159  * On ext2, we can store the error state of the filesystem in the
160  * superblock.  That is not possible on ext3, because we may have other
161  * write ordering constraints on the superblock which prevent us from
162  * writing it out straight away; and given that the journal is about to
163  * be aborted, we can't rely on the current, or future, transactions to
164  * write out the superblock safely.
165  *
166  * We'll just use the journal_abort() error code to record an error in
167  * the journal instead.  On recovery, the journal will compain about
168  * that error until we've noted it down and cleared it.
169  */
170
171 static void ext3_handle_error(struct super_block *sb)
172 {
173         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
174
175         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
176         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
177
178         if (sb->s_flags & MS_RDONLY)
179                 return;
180
181         if (!test_opt (sb, ERRORS_CONT)) {
182                 journal_t *journal = EXT3_SB(sb)->s_journal;
183
184                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
185                 if (journal)
186                         journal_abort(journal, -EIO);
187         }
188         if (test_opt (sb, ERRORS_RO)) {
189                 ext3_msg(sb, KERN_CRIT,
190                         "error: remounting filesystem read-only");
191                 sb->s_flags |= MS_RDONLY;
192         }
193         ext3_commit_super(sb, es, 1);
194         if (test_opt(sb, ERRORS_PANIC))
195                 panic("EXT3-fs (%s): panic forced after error\n",
196                         sb->s_id);
197 }
198
199 void ext3_error (struct super_block * sb, const char * function,
200                  const char * fmt, ...)
201 {
202         va_list args;
203
204         va_start(args, fmt);
205         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
206         vprintk(fmt, args);
207         printk("\n");
208         va_end(args);
209
210         ext3_handle_error(sb);
211 }
212
213 static const char *ext3_decode_error(struct super_block * sb, int errno,
214                                      char nbuf[16])
215 {
216         char *errstr = NULL;
217
218         switch (errno) {
219         case -EIO:
220                 errstr = "IO failure";
221                 break;
222         case -ENOMEM:
223                 errstr = "Out of memory";
224                 break;
225         case -EROFS:
226                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
227                         errstr = "Journal has aborted";
228                 else
229                         errstr = "Readonly filesystem";
230                 break;
231         default:
232                 /* If the caller passed in an extra buffer for unknown
233                  * errors, textualise them now.  Else we just return
234                  * NULL. */
235                 if (nbuf) {
236                         /* Check for truncated error codes... */
237                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
238                                 errstr = nbuf;
239                 }
240                 break;
241         }
242
243         return errstr;
244 }
245
246 /* __ext3_std_error decodes expected errors from journaling functions
247  * automatically and invokes the appropriate error response.  */
248
249 void __ext3_std_error (struct super_block * sb, const char * function,
250                        int errno)
251 {
252         char nbuf[16];
253         const char *errstr;
254
255         /* Special case: if the error is EROFS, and we're not already
256          * inside a transaction, then there's really no point in logging
257          * an error. */
258         if (errno == -EROFS && journal_current_handle() == NULL &&
259             (sb->s_flags & MS_RDONLY))
260                 return;
261
262         errstr = ext3_decode_error(sb, errno, nbuf);
263         ext3_msg(sb, KERN_CRIT, "error in %s: %s", function, errstr);
264
265         ext3_handle_error(sb);
266 }
267
268 /*
269  * ext3_abort is a much stronger failure handler than ext3_error.  The
270  * abort function may be used to deal with unrecoverable failures such
271  * as journal IO errors or ENOMEM at a critical moment in log management.
272  *
273  * We unconditionally force the filesystem into an ABORT|READONLY state,
274  * unless the error response on the fs has been set to panic in which
275  * case we take the easy way out and panic immediately.
276  */
277
278 void ext3_abort (struct super_block * sb, const char * function,
279                  const char * fmt, ...)
280 {
281         va_list args;
282
283         va_start(args, fmt);
284         printk(KERN_CRIT "EXT3-fs (%s): error: %s: ", sb->s_id, function);
285         vprintk(fmt, args);
286         printk("\n");
287         va_end(args);
288
289         if (test_opt(sb, ERRORS_PANIC))
290                 panic("EXT3-fs: panic from previous error\n");
291
292         if (sb->s_flags & MS_RDONLY)
293                 return;
294
295         ext3_msg(sb, KERN_CRIT,
296                 "error: remounting filesystem read-only");
297         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
298         sb->s_flags |= MS_RDONLY;
299         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
300         if (EXT3_SB(sb)->s_journal)
301                 journal_abort(EXT3_SB(sb)->s_journal, -EIO);
302 }
303
304 void ext3_warning (struct super_block * sb, const char * function,
305                    const char * fmt, ...)
306 {
307         va_list args;
308
309         va_start(args, fmt);
310         printk(KERN_WARNING "EXT3-fs (%s): warning: %s: ",
311                sb->s_id, function);
312         vprintk(fmt, args);
313         printk("\n");
314         va_end(args);
315 }
316
317 void ext3_update_dynamic_rev(struct super_block *sb)
318 {
319         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
320
321         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
322                 return;
323
324         ext3_msg(sb, KERN_WARNING,
325                 "warning: updating to rev %d because of "
326                 "new feature flag, running e2fsck is recommended",
327                 EXT3_DYNAMIC_REV);
328
329         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
330         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
331         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
332         /* leave es->s_feature_*compat flags alone */
333         /* es->s_uuid will be set by e2fsck if empty */
334
335         /*
336          * The rest of the superblock fields should be zero, and if not it
337          * means they are likely already in use, so leave them alone.  We
338          * can leave it up to e2fsck to clean up any inconsistencies there.
339          */
340 }
341
342 /*
343  * Open the external journal device
344  */
345 static struct block_device *ext3_blkdev_get(dev_t dev, struct super_block *sb)
346 {
347         struct block_device *bdev;
348         char b[BDEVNAME_SIZE];
349
350         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
351         if (IS_ERR(bdev))
352                 goto fail;
353         return bdev;
354
355 fail:
356         ext3_msg(sb, "error: failed to open journal device %s: %ld",
357                 __bdevname(dev, b), PTR_ERR(bdev));
358
359         return NULL;
360 }
361
362 /*
363  * Release the journal device
364  */
365 static int ext3_blkdev_put(struct block_device *bdev)
366 {
367         bd_release(bdev);
368         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
369 }
370
371 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
372 {
373         struct block_device *bdev;
374         int ret = -ENODEV;
375
376         bdev = sbi->journal_bdev;
377         if (bdev) {
378                 ret = ext3_blkdev_put(bdev);
379                 sbi->journal_bdev = NULL;
380         }
381         return ret;
382 }
383
384 static inline struct inode *orphan_list_entry(struct list_head *l)
385 {
386         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
387 }
388
389 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
390 {
391         struct list_head *l;
392
393         ext3_msg(sb, KERN_ERR, "error: sb orphan head is %d",
394                le32_to_cpu(sbi->s_es->s_last_orphan));
395
396         ext3_msg(sb, KERN_ERR, "sb_info orphan list:");
397         list_for_each(l, &sbi->s_orphan) {
398                 struct inode *inode = orphan_list_entry(l);
399                 ext3_msg(sb, KERN_ERR, "  "
400                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
401                        inode->i_sb->s_id, inode->i_ino, inode,
402                        inode->i_mode, inode->i_nlink,
403                        NEXT_ORPHAN(inode));
404         }
405 }
406
407 static void ext3_put_super (struct super_block * sb)
408 {
409         struct ext3_sb_info *sbi = EXT3_SB(sb);
410         struct ext3_super_block *es = sbi->s_es;
411         int i, err;
412
413         lock_kernel();
414
415         ext3_xattr_put_super(sb);
416         err = journal_destroy(sbi->s_journal);
417         sbi->s_journal = NULL;
418         if (err < 0)
419                 ext3_abort(sb, __func__, "Couldn't clean up the journal");
420
421         if (!(sb->s_flags & MS_RDONLY)) {
422                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
423                 es->s_state = cpu_to_le16(sbi->s_mount_state);
424                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
425                 mark_buffer_dirty(sbi->s_sbh);
426                 ext3_commit_super(sb, es, 1);
427         }
428
429         for (i = 0; i < sbi->s_gdb_count; i++)
430                 brelse(sbi->s_group_desc[i]);
431         kfree(sbi->s_group_desc);
432         percpu_counter_destroy(&sbi->s_freeblocks_counter);
433         percpu_counter_destroy(&sbi->s_freeinodes_counter);
434         percpu_counter_destroy(&sbi->s_dirs_counter);
435         brelse(sbi->s_sbh);
436 #ifdef CONFIG_QUOTA
437         for (i = 0; i < MAXQUOTAS; i++)
438                 kfree(sbi->s_qf_names[i]);
439 #endif
440
441         /* Debugging code just in case the in-memory inode orphan list
442          * isn't empty.  The on-disk one can be non-empty if we've
443          * detected an error and taken the fs readonly, but the
444          * in-memory list had better be clean by this point. */
445         if (!list_empty(&sbi->s_orphan))
446                 dump_orphan_list(sb, sbi);
447         J_ASSERT(list_empty(&sbi->s_orphan));
448
449         invalidate_bdev(sb->s_bdev);
450         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
451                 /*
452                  * Invalidate the journal device's buffers.  We don't want them
453                  * floating about in memory - the physical journal device may
454                  * hotswapped, and it breaks the `ro-after' testing code.
455                  */
456                 sync_blockdev(sbi->journal_bdev);
457                 invalidate_bdev(sbi->journal_bdev);
458                 ext3_blkdev_remove(sbi);
459         }
460         sb->s_fs_info = NULL;
461         kfree(sbi->s_blockgroup_lock);
462         kfree(sbi);
463
464         unlock_kernel();
465 }
466
467 static struct kmem_cache *ext3_inode_cachep;
468
469 /*
470  * Called inside transaction, so use GFP_NOFS
471  */
472 static struct inode *ext3_alloc_inode(struct super_block *sb)
473 {
474         struct ext3_inode_info *ei;
475
476         ei = kmem_cache_alloc(ext3_inode_cachep, GFP_NOFS);
477         if (!ei)
478                 return NULL;
479         ei->i_block_alloc_info = NULL;
480         ei->vfs_inode.i_version = 1;
481         atomic_set(&ei->i_datasync_tid, 0);
482         atomic_set(&ei->i_sync_tid, 0);
483         return &ei->vfs_inode;
484 }
485
486 static void ext3_destroy_inode(struct inode *inode)
487 {
488         if (!list_empty(&(EXT3_I(inode)->i_orphan))) {
489                 printk("EXT3 Inode %p: orphan list check failed!\n",
490                         EXT3_I(inode));
491                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
492                                 EXT3_I(inode), sizeof(struct ext3_inode_info),
493                                 false);
494                 dump_stack();
495         }
496         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
497 }
498
499 static void init_once(void *foo)
500 {
501         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
502
503         INIT_LIST_HEAD(&ei->i_orphan);
504 #ifdef CONFIG_EXT3_FS_XATTR
505         init_rwsem(&ei->xattr_sem);
506 #endif
507         mutex_init(&ei->truncate_mutex);
508         inode_init_once(&ei->vfs_inode);
509 }
510
511 static int init_inodecache(void)
512 {
513         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
514                                              sizeof(struct ext3_inode_info),
515                                              0, (SLAB_RECLAIM_ACCOUNT|
516                                                 SLAB_MEM_SPREAD),
517                                              init_once);
518         if (ext3_inode_cachep == NULL)
519                 return -ENOMEM;
520         return 0;
521 }
522
523 static void destroy_inodecache(void)
524 {
525         kmem_cache_destroy(ext3_inode_cachep);
526 }
527
528 static void ext3_clear_inode(struct inode *inode)
529 {
530         struct ext3_block_alloc_info *rsv = EXT3_I(inode)->i_block_alloc_info;
531         ext3_discard_reservation(inode);
532         EXT3_I(inode)->i_block_alloc_info = NULL;
533         if (unlikely(rsv))
534                 kfree(rsv);
535 }
536
537 static inline void ext3_show_quota_options(struct seq_file *seq, struct super_block *sb)
538 {
539 #if defined(CONFIG_QUOTA)
540         struct ext3_sb_info *sbi = EXT3_SB(sb);
541
542         if (sbi->s_jquota_fmt)
543                 seq_printf(seq, ",jqfmt=%s",
544                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
545
546         if (sbi->s_qf_names[USRQUOTA])
547                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
548
549         if (sbi->s_qf_names[GRPQUOTA])
550                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
551
552         if (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA)
553                 seq_puts(seq, ",usrquota");
554
555         if (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)
556                 seq_puts(seq, ",grpquota");
557 #endif
558 }
559
560 static char *data_mode_string(unsigned long mode)
561 {
562         switch (mode) {
563         case EXT3_MOUNT_JOURNAL_DATA:
564                 return "journal";
565         case EXT3_MOUNT_ORDERED_DATA:
566                 return "ordered";
567         case EXT3_MOUNT_WRITEBACK_DATA:
568                 return "writeback";
569         }
570         return "unknown";
571 }
572
573 /*
574  * Show an option if
575  *  - it's set to a non-default value OR
576  *  - if the per-sb default is different from the global default
577  */
578 static int ext3_show_options(struct seq_file *seq, struct vfsmount *vfs)
579 {
580         struct super_block *sb = vfs->mnt_sb;
581         struct ext3_sb_info *sbi = EXT3_SB(sb);
582         struct ext3_super_block *es = sbi->s_es;
583         unsigned long def_mount_opts;
584
585         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
586
587         if (sbi->s_sb_block != 1)
588                 seq_printf(seq, ",sb=%lu", sbi->s_sb_block);
589         if (test_opt(sb, MINIX_DF))
590                 seq_puts(seq, ",minixdf");
591         if (test_opt(sb, GRPID))
592                 seq_puts(seq, ",grpid");
593         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT3_DEFM_BSDGROUPS))
594                 seq_puts(seq, ",nogrpid");
595         if (sbi->s_resuid != EXT3_DEF_RESUID ||
596             le16_to_cpu(es->s_def_resuid) != EXT3_DEF_RESUID) {
597                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
598         }
599         if (sbi->s_resgid != EXT3_DEF_RESGID ||
600             le16_to_cpu(es->s_def_resgid) != EXT3_DEF_RESGID) {
601                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
602         }
603         if (test_opt(sb, ERRORS_RO)) {
604                 int def_errors = le16_to_cpu(es->s_errors);
605
606                 if (def_errors == EXT3_ERRORS_PANIC ||
607                     def_errors == EXT3_ERRORS_CONTINUE) {
608                         seq_puts(seq, ",errors=remount-ro");
609                 }
610         }
611         if (test_opt(sb, ERRORS_CONT))
612                 seq_puts(seq, ",errors=continue");
613         if (test_opt(sb, ERRORS_PANIC))
614                 seq_puts(seq, ",errors=panic");
615         if (test_opt(sb, NO_UID32))
616                 seq_puts(seq, ",nouid32");
617         if (test_opt(sb, DEBUG))
618                 seq_puts(seq, ",debug");
619         if (test_opt(sb, OLDALLOC))
620                 seq_puts(seq, ",oldalloc");
621 #ifdef CONFIG_EXT3_FS_XATTR
622         if (test_opt(sb, XATTR_USER))
623                 seq_puts(seq, ",user_xattr");
624         if (!test_opt(sb, XATTR_USER) &&
625             (def_mount_opts & EXT3_DEFM_XATTR_USER)) {
626                 seq_puts(seq, ",nouser_xattr");
627         }
628 #endif
629 #ifdef CONFIG_EXT3_FS_POSIX_ACL
630         if (test_opt(sb, POSIX_ACL))
631                 seq_puts(seq, ",acl");
632         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT3_DEFM_ACL))
633                 seq_puts(seq, ",noacl");
634 #endif
635         if (!test_opt(sb, RESERVATION))
636                 seq_puts(seq, ",noreservation");
637         if (sbi->s_commit_interval) {
638                 seq_printf(seq, ",commit=%u",
639                            (unsigned) (sbi->s_commit_interval / HZ));
640         }
641         if (test_opt(sb, BARRIER))
642                 seq_puts(seq, ",barrier=1");
643         if (test_opt(sb, NOBH))
644                 seq_puts(seq, ",nobh");
645
646         seq_printf(seq, ",data=%s", data_mode_string(sbi->s_mount_opt &
647                                                      EXT3_MOUNT_DATA_FLAGS));
648         if (test_opt(sb, DATA_ERR_ABORT))
649                 seq_puts(seq, ",data_err=abort");
650
651         if (test_opt(sb, NOLOAD))
652                 seq_puts(seq, ",norecovery");
653
654         ext3_show_quota_options(seq, sb);
655
656         return 0;
657 }
658
659
660 static struct inode *ext3_nfs_get_inode(struct super_block *sb,
661                 u64 ino, u32 generation)
662 {
663         struct inode *inode;
664
665         if (ino < EXT3_FIRST_INO(sb) && ino != EXT3_ROOT_INO)
666                 return ERR_PTR(-ESTALE);
667         if (ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count))
668                 return ERR_PTR(-ESTALE);
669
670         /* iget isn't really right if the inode is currently unallocated!!
671          *
672          * ext3_read_inode will return a bad_inode if the inode had been
673          * deleted, so we should be safe.
674          *
675          * Currently we don't know the generation for parent directory, so
676          * a generation of 0 means "accept any"
677          */
678         inode = ext3_iget(sb, ino);
679         if (IS_ERR(inode))
680                 return ERR_CAST(inode);
681         if (generation && inode->i_generation != generation) {
682                 iput(inode);
683                 return ERR_PTR(-ESTALE);
684         }
685
686         return inode;
687 }
688
689 static struct dentry *ext3_fh_to_dentry(struct super_block *sb, struct fid *fid,
690                 int fh_len, int fh_type)
691 {
692         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
693                                     ext3_nfs_get_inode);
694 }
695
696 static struct dentry *ext3_fh_to_parent(struct super_block *sb, struct fid *fid,
697                 int fh_len, int fh_type)
698 {
699         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
700                                     ext3_nfs_get_inode);
701 }
702
703 /*
704  * Try to release metadata pages (indirect blocks, directories) which are
705  * mapped via the block device.  Since these pages could have journal heads
706  * which would prevent try_to_free_buffers() from freeing them, we must use
707  * jbd layer's try_to_free_buffers() function to release them.
708  */
709 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
710                                  gfp_t wait)
711 {
712         journal_t *journal = EXT3_SB(sb)->s_journal;
713
714         WARN_ON(PageChecked(page));
715         if (!page_has_buffers(page))
716                 return 0;
717         if (journal)
718                 return journal_try_to_free_buffers(journal, page, 
719                                                    wait & ~__GFP_WAIT);
720         return try_to_free_buffers(page);
721 }
722
723 #ifdef CONFIG_QUOTA
724 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
725 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
726
727 static int ext3_write_dquot(struct dquot *dquot);
728 static int ext3_acquire_dquot(struct dquot *dquot);
729 static int ext3_release_dquot(struct dquot *dquot);
730 static int ext3_mark_dquot_dirty(struct dquot *dquot);
731 static int ext3_write_info(struct super_block *sb, int type);
732 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
733                                 char *path, int remount);
734 static int ext3_quota_on_mount(struct super_block *sb, int type);
735 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
736                                size_t len, loff_t off);
737 static ssize_t ext3_quota_write(struct super_block *sb, int type,
738                                 const char *data, size_t len, loff_t off);
739
740 static const struct dquot_operations ext3_quota_operations = {
741         .initialize     = dquot_initialize,
742         .drop           = dquot_drop,
743         .alloc_space    = dquot_alloc_space,
744         .alloc_inode    = dquot_alloc_inode,
745         .free_space     = dquot_free_space,
746         .free_inode     = dquot_free_inode,
747         .transfer       = dquot_transfer,
748         .write_dquot    = ext3_write_dquot,
749         .acquire_dquot  = ext3_acquire_dquot,
750         .release_dquot  = ext3_release_dquot,
751         .mark_dirty     = ext3_mark_dquot_dirty,
752         .write_info     = ext3_write_info,
753         .alloc_dquot    = dquot_alloc,
754         .destroy_dquot  = dquot_destroy,
755 };
756
757 static const struct quotactl_ops ext3_qctl_operations = {
758         .quota_on       = ext3_quota_on,
759         .quota_off      = vfs_quota_off,
760         .quota_sync     = vfs_quota_sync,
761         .get_info       = vfs_get_dqinfo,
762         .set_info       = vfs_set_dqinfo,
763         .get_dqblk      = vfs_get_dqblk,
764         .set_dqblk      = vfs_set_dqblk
765 };
766 #endif
767
768 static const struct super_operations ext3_sops = {
769         .alloc_inode    = ext3_alloc_inode,
770         .destroy_inode  = ext3_destroy_inode,
771         .write_inode    = ext3_write_inode,
772         .dirty_inode    = ext3_dirty_inode,
773         .delete_inode   = ext3_delete_inode,
774         .put_super      = ext3_put_super,
775         .sync_fs        = ext3_sync_fs,
776         .freeze_fs      = ext3_freeze,
777         .unfreeze_fs    = ext3_unfreeze,
778         .statfs         = ext3_statfs,
779         .remount_fs     = ext3_remount,
780         .clear_inode    = ext3_clear_inode,
781         .show_options   = ext3_show_options,
782 #ifdef CONFIG_QUOTA
783         .quota_read     = ext3_quota_read,
784         .quota_write    = ext3_quota_write,
785 #endif
786         .bdev_try_to_free_page = bdev_try_to_free_page,
787 };
788
789 static const struct export_operations ext3_export_ops = {
790         .fh_to_dentry = ext3_fh_to_dentry,
791         .fh_to_parent = ext3_fh_to_parent,
792         .get_parent = ext3_get_parent,
793 };
794
795 enum {
796         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
797         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
798         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
799         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
800         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
801         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
802         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
803         Opt_data_err_abort, Opt_data_err_ignore,
804         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
805         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
806         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
807         Opt_grpquota
808 };
809
810 static const match_table_t tokens = {
811         {Opt_bsd_df, "bsddf"},
812         {Opt_minix_df, "minixdf"},
813         {Opt_grpid, "grpid"},
814         {Opt_grpid, "bsdgroups"},
815         {Opt_nogrpid, "nogrpid"},
816         {Opt_nogrpid, "sysvgroups"},
817         {Opt_resgid, "resgid=%u"},
818         {Opt_resuid, "resuid=%u"},
819         {Opt_sb, "sb=%u"},
820         {Opt_err_cont, "errors=continue"},
821         {Opt_err_panic, "errors=panic"},
822         {Opt_err_ro, "errors=remount-ro"},
823         {Opt_nouid32, "nouid32"},
824         {Opt_nocheck, "nocheck"},
825         {Opt_nocheck, "check=none"},
826         {Opt_debug, "debug"},
827         {Opt_oldalloc, "oldalloc"},
828         {Opt_orlov, "orlov"},
829         {Opt_user_xattr, "user_xattr"},
830         {Opt_nouser_xattr, "nouser_xattr"},
831         {Opt_acl, "acl"},
832         {Opt_noacl, "noacl"},
833         {Opt_reservation, "reservation"},
834         {Opt_noreservation, "noreservation"},
835         {Opt_noload, "noload"},
836         {Opt_noload, "norecovery"},
837         {Opt_nobh, "nobh"},
838         {Opt_bh, "bh"},
839         {Opt_commit, "commit=%u"},
840         {Opt_journal_update, "journal=update"},
841         {Opt_journal_inum, "journal=%u"},
842         {Opt_journal_dev, "journal_dev=%u"},
843         {Opt_abort, "abort"},
844         {Opt_data_journal, "data=journal"},
845         {Opt_data_ordered, "data=ordered"},
846         {Opt_data_writeback, "data=writeback"},
847         {Opt_data_err_abort, "data_err=abort"},
848         {Opt_data_err_ignore, "data_err=ignore"},
849         {Opt_offusrjquota, "usrjquota="},
850         {Opt_usrjquota, "usrjquota=%s"},
851         {Opt_offgrpjquota, "grpjquota="},
852         {Opt_grpjquota, "grpjquota=%s"},
853         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
854         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
855         {Opt_grpquota, "grpquota"},
856         {Opt_noquota, "noquota"},
857         {Opt_quota, "quota"},
858         {Opt_usrquota, "usrquota"},
859         {Opt_barrier, "barrier=%u"},
860         {Opt_resize, "resize"},
861         {Opt_err, NULL},
862 };
863
864 static ext3_fsblk_t get_sb_block(void **data, struct super_block *sb)
865 {
866         ext3_fsblk_t    sb_block;
867         char            *options = (char *) *data;
868
869         if (!options || strncmp(options, "sb=", 3) != 0)
870                 return 1;       /* Default location */
871         options += 3;
872         /*todo: use simple_strtoll with >32bit ext3 */
873         sb_block = simple_strtoul(options, &options, 0);
874         if (*options && *options != ',') {
875                 ext3_msg(sb, "error: invalid sb specification: %s",
876                        (char *) *data);
877                 return 1;
878         }
879         if (*options == ',')
880                 options++;
881         *data = (void *) options;
882         return sb_block;
883 }
884
885 static int parse_options (char *options, struct super_block *sb,
886                           unsigned int *inum, unsigned long *journal_devnum,
887                           ext3_fsblk_t *n_blocks_count, int is_remount)
888 {
889         struct ext3_sb_info *sbi = EXT3_SB(sb);
890         char * p;
891         substring_t args[MAX_OPT_ARGS];
892         int data_opt = 0;
893         int option;
894 #ifdef CONFIG_QUOTA
895         int qtype, qfmt;
896         char *qname;
897 #endif
898
899         if (!options)
900                 return 1;
901
902         while ((p = strsep (&options, ",")) != NULL) {
903                 int token;
904                 if (!*p)
905                         continue;
906
907                 token = match_token(p, tokens, args);
908                 switch (token) {
909                 case Opt_bsd_df:
910                         clear_opt (sbi->s_mount_opt, MINIX_DF);
911                         break;
912                 case Opt_minix_df:
913                         set_opt (sbi->s_mount_opt, MINIX_DF);
914                         break;
915                 case Opt_grpid:
916                         set_opt (sbi->s_mount_opt, GRPID);
917                         break;
918                 case Opt_nogrpid:
919                         clear_opt (sbi->s_mount_opt, GRPID);
920                         break;
921                 case Opt_resuid:
922                         if (match_int(&args[0], &option))
923                                 return 0;
924                         sbi->s_resuid = option;
925                         break;
926                 case Opt_resgid:
927                         if (match_int(&args[0], &option))
928                                 return 0;
929                         sbi->s_resgid = option;
930                         break;
931                 case Opt_sb:
932                         /* handled by get_sb_block() instead of here */
933                         /* *sb_block = match_int(&args[0]); */
934                         break;
935                 case Opt_err_panic:
936                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
937                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
938                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
939                         break;
940                 case Opt_err_ro:
941                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
942                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
943                         set_opt (sbi->s_mount_opt, ERRORS_RO);
944                         break;
945                 case Opt_err_cont:
946                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
947                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
948                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
949                         break;
950                 case Opt_nouid32:
951                         set_opt (sbi->s_mount_opt, NO_UID32);
952                         break;
953                 case Opt_nocheck:
954                         clear_opt (sbi->s_mount_opt, CHECK);
955                         break;
956                 case Opt_debug:
957                         set_opt (sbi->s_mount_opt, DEBUG);
958                         break;
959                 case Opt_oldalloc:
960                         set_opt (sbi->s_mount_opt, OLDALLOC);
961                         break;
962                 case Opt_orlov:
963                         clear_opt (sbi->s_mount_opt, OLDALLOC);
964                         break;
965 #ifdef CONFIG_EXT3_FS_XATTR
966                 case Opt_user_xattr:
967                         set_opt (sbi->s_mount_opt, XATTR_USER);
968                         break;
969                 case Opt_nouser_xattr:
970                         clear_opt (sbi->s_mount_opt, XATTR_USER);
971                         break;
972 #else
973                 case Opt_user_xattr:
974                 case Opt_nouser_xattr:
975                         ext3_msg(sb, KERN_INFO,
976                                 "(no)user_xattr options not supported");
977                         break;
978 #endif
979 #ifdef CONFIG_EXT3_FS_POSIX_ACL
980                 case Opt_acl:
981                         set_opt(sbi->s_mount_opt, POSIX_ACL);
982                         break;
983                 case Opt_noacl:
984                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
985                         break;
986 #else
987                 case Opt_acl:
988                 case Opt_noacl:
989                         ext3_msg(sb, KERN_INFO,
990                                 "(no)acl options not supported");
991                         break;
992 #endif
993                 case Opt_reservation:
994                         set_opt(sbi->s_mount_opt, RESERVATION);
995                         break;
996                 case Opt_noreservation:
997                         clear_opt(sbi->s_mount_opt, RESERVATION);
998                         break;
999                 case Opt_journal_update:
1000                         /* @@@ FIXME */
1001                         /* Eventually we will want to be able to create
1002                            a journal file here.  For now, only allow the
1003                            user to specify an existing inode to be the
1004                            journal file. */
1005                         if (is_remount) {
1006                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1007                                         "journal on remount");
1008                                 return 0;
1009                         }
1010                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
1011                         break;
1012                 case Opt_journal_inum:
1013                         if (is_remount) {
1014                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1015                                        "journal on remount");
1016                                 return 0;
1017                         }
1018                         if (match_int(&args[0], &option))
1019                                 return 0;
1020                         *inum = option;
1021                         break;
1022                 case Opt_journal_dev:
1023                         if (is_remount) {
1024                                 ext3_msg(sb, KERN_ERR, "error: cannot specify "
1025                                        "journal on remount");
1026                                 return 0;
1027                         }
1028                         if (match_int(&args[0], &option))
1029                                 return 0;
1030                         *journal_devnum = option;
1031                         break;
1032                 case Opt_noload:
1033                         set_opt (sbi->s_mount_opt, NOLOAD);
1034                         break;
1035                 case Opt_commit:
1036                         if (match_int(&args[0], &option))
1037                                 return 0;
1038                         if (option < 0)
1039                                 return 0;
1040                         if (option == 0)
1041                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
1042                         sbi->s_commit_interval = HZ * option;
1043                         break;
1044                 case Opt_data_journal:
1045                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
1046                         goto datacheck;
1047                 case Opt_data_ordered:
1048                         data_opt = EXT3_MOUNT_ORDERED_DATA;
1049                         goto datacheck;
1050                 case Opt_data_writeback:
1051                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
1052                 datacheck:
1053                         if (is_remount) {
1054                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
1055                                                 == data_opt)
1056                                         break;
1057                                 ext3_msg(sb, KERN_ERR,
1058                                         "error: cannot change "
1059                                         "data mode on remount. The filesystem "
1060                                         "is mounted in data=%s mode and you "
1061                                         "try to remount it in data=%s mode.",
1062                                         data_mode_string(sbi->s_mount_opt &
1063                                                         EXT3_MOUNT_DATA_FLAGS),
1064                                         data_mode_string(data_opt));
1065                                 return 0;
1066                         } else {
1067                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
1068                                 sbi->s_mount_opt |= data_opt;
1069                         }
1070                         break;
1071                 case Opt_data_err_abort:
1072                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1073                         break;
1074                 case Opt_data_err_ignore:
1075                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1076                         break;
1077 #ifdef CONFIG_QUOTA
1078                 case Opt_usrjquota:
1079                         qtype = USRQUOTA;
1080                         goto set_qf_name;
1081                 case Opt_grpjquota:
1082                         qtype = GRPQUOTA;
1083 set_qf_name:
1084                         if (sb_any_quota_loaded(sb) &&
1085                             !sbi->s_qf_names[qtype]) {
1086                                 ext3_msg(sb, KERN_ERR,
1087                                         "error: cannot change journaled "
1088                                         "quota options when quota turned on.");
1089                                 return 0;
1090                         }
1091                         qname = match_strdup(&args[0]);
1092                         if (!qname) {
1093                                 ext3_msg(sb, KERN_ERR,
1094                                         "error: not enough memory for "
1095                                         "storing quotafile name.");
1096                                 return 0;
1097                         }
1098                         if (sbi->s_qf_names[qtype] &&
1099                             strcmp(sbi->s_qf_names[qtype], qname)) {
1100                                 ext3_msg(sb, KERN_ERR,
1101                                         "error: %s quota file already "
1102                                         "specified.", QTYPE2NAME(qtype));
1103                                 kfree(qname);
1104                                 return 0;
1105                         }
1106                         sbi->s_qf_names[qtype] = qname;
1107                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1108                                 ext3_msg(sb, KERN_ERR,
1109                                         "error: quotafile must be on "
1110                                         "filesystem root.");
1111                                 kfree(sbi->s_qf_names[qtype]);
1112                                 sbi->s_qf_names[qtype] = NULL;
1113                                 return 0;
1114                         }
1115                         set_opt(sbi->s_mount_opt, QUOTA);
1116                         break;
1117                 case Opt_offusrjquota:
1118                         qtype = USRQUOTA;
1119                         goto clear_qf_name;
1120                 case Opt_offgrpjquota:
1121                         qtype = GRPQUOTA;
1122 clear_qf_name:
1123                         if (sb_any_quota_loaded(sb) &&
1124                             sbi->s_qf_names[qtype]) {
1125                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1126                                         "journaled quota options when "
1127                                         "quota turned on.");
1128                                 return 0;
1129                         }
1130                         /*
1131                          * The space will be released later when all options
1132                          * are confirmed to be correct
1133                          */
1134                         sbi->s_qf_names[qtype] = NULL;
1135                         break;
1136                 case Opt_jqfmt_vfsold:
1137                         qfmt = QFMT_VFS_OLD;
1138                         goto set_qf_format;
1139                 case Opt_jqfmt_vfsv0:
1140                         qfmt = QFMT_VFS_V0;
1141 set_qf_format:
1142                         if (sb_any_quota_loaded(sb) &&
1143                             sbi->s_jquota_fmt != qfmt) {
1144                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1145                                         "journaled quota options when "
1146                                         "quota turned on.");
1147                                 return 0;
1148                         }
1149                         sbi->s_jquota_fmt = qfmt;
1150                         break;
1151                 case Opt_quota:
1152                 case Opt_usrquota:
1153                         set_opt(sbi->s_mount_opt, QUOTA);
1154                         set_opt(sbi->s_mount_opt, USRQUOTA);
1155                         break;
1156                 case Opt_grpquota:
1157                         set_opt(sbi->s_mount_opt, QUOTA);
1158                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1159                         break;
1160                 case Opt_noquota:
1161                         if (sb_any_quota_loaded(sb)) {
1162                                 ext3_msg(sb, KERN_ERR, "error: cannot change "
1163                                         "quota options when quota turned on.");
1164                                 return 0;
1165                         }
1166                         clear_opt(sbi->s_mount_opt, QUOTA);
1167                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1168                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1169                         break;
1170 #else
1171                 case Opt_quota:
1172                 case Opt_usrquota:
1173                 case Opt_grpquota:
1174                         ext3_msg(sb, KERN_ERR,
1175                                 "error: quota options not supported.");
1176                         break;
1177                 case Opt_usrjquota:
1178                 case Opt_grpjquota:
1179                 case Opt_offusrjquota:
1180                 case Opt_offgrpjquota:
1181                 case Opt_jqfmt_vfsold:
1182                 case Opt_jqfmt_vfsv0:
1183                         ext3_msg(sb, KERN_ERR,
1184                                 "error: journaled quota options not "
1185                                 "supported.");
1186                         break;
1187                 case Opt_noquota:
1188                         break;
1189 #endif
1190                 case Opt_abort:
1191                         set_opt(sbi->s_mount_opt, ABORT);
1192                         break;
1193                 case Opt_barrier:
1194                         if (match_int(&args[0], &option))
1195                                 return 0;
1196                         if (option)
1197                                 set_opt(sbi->s_mount_opt, BARRIER);
1198                         else
1199                                 clear_opt(sbi->s_mount_opt, BARRIER);
1200                         break;
1201                 case Opt_ignore:
1202                         break;
1203                 case Opt_resize:
1204                         if (!is_remount) {
1205                                 ext3_msg(sb, KERN_ERR,
1206                                         "error: resize option only available "
1207                                         "for remount");
1208                                 return 0;
1209                         }
1210                         if (match_int(&args[0], &option) != 0)
1211                                 return 0;
1212                         *n_blocks_count = option;
1213                         break;
1214                 case Opt_nobh:
1215                         set_opt(sbi->s_mount_opt, NOBH);
1216                         break;
1217                 case Opt_bh:
1218                         clear_opt(sbi->s_mount_opt, NOBH);
1219                         break;
1220                 default:
1221                         ext3_msg(sb, KERN_ERR,
1222                                 "error: unrecognized mount option \"%s\" "
1223                                 "or missing value", p);
1224                         return 0;
1225                 }
1226         }
1227 #ifdef CONFIG_QUOTA
1228         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1229                 if ((sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA) &&
1230                      sbi->s_qf_names[USRQUOTA])
1231                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1232
1233                 if ((sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA) &&
1234                      sbi->s_qf_names[GRPQUOTA])
1235                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1236
1237                 if ((sbi->s_qf_names[USRQUOTA] &&
1238                                 (sbi->s_mount_opt & EXT3_MOUNT_GRPQUOTA)) ||
1239                     (sbi->s_qf_names[GRPQUOTA] &&
1240                                 (sbi->s_mount_opt & EXT3_MOUNT_USRQUOTA))) {
1241                         ext3_msg(sb, KERN_ERR, "error: old and new quota "
1242                                         "format mixing.");
1243                         return 0;
1244                 }
1245
1246                 if (!sbi->s_jquota_fmt) {
1247                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1248                                         "not specified.");
1249                         return 0;
1250                 }
1251         } else {
1252                 if (sbi->s_jquota_fmt) {
1253                         ext3_msg(sb, KERN_ERR, "error: journaled quota format "
1254                                         "specified with no journaling "
1255                                         "enabled.");
1256                         return 0;
1257                 }
1258         }
1259 #endif
1260         return 1;
1261 }
1262
1263 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
1264                             int read_only)
1265 {
1266         struct ext3_sb_info *sbi = EXT3_SB(sb);
1267         int res = 0;
1268
1269         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
1270                 ext3_msg(sb, KERN_ERR,
1271                         "error: revision level too high, "
1272                         "forcing read-only mode");
1273                 res = MS_RDONLY;
1274         }
1275         if (read_only)
1276                 return res;
1277         if (!(sbi->s_mount_state & EXT3_VALID_FS))
1278                 ext3_msg(sb, KERN_WARNING,
1279                         "warning: mounting unchecked fs, "
1280                         "running e2fsck is recommended");
1281         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
1282                 ext3_msg(sb, KERN_WARNING,
1283                         "warning: mounting fs with errors, "
1284                         "running e2fsck is recommended");
1285         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1286                  le16_to_cpu(es->s_mnt_count) >=
1287                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1288                 ext3_msg(sb, KERN_WARNING,
1289                         "warning: maximal mount count reached, "
1290                         "running e2fsck is recommended");
1291         else if (le32_to_cpu(es->s_checkinterval) &&
1292                 (le32_to_cpu(es->s_lastcheck) +
1293                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1294                 ext3_msg(sb, KERN_WARNING,
1295                         "warning: checktime reached, "
1296                         "running e2fsck is recommended");
1297 #if 0
1298                 /* @@@ We _will_ want to clear the valid bit if we find
1299                    inconsistencies, to force a fsck at reboot.  But for
1300                    a plain journaled filesystem we can keep it set as
1301                    valid forever! :) */
1302         es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
1303 #endif
1304         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1305                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
1306         le16_add_cpu(&es->s_mnt_count, 1);
1307         es->s_mtime = cpu_to_le32(get_seconds());
1308         ext3_update_dynamic_rev(sb);
1309         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1310
1311         ext3_commit_super(sb, es, 1);
1312         if (test_opt(sb, DEBUG))
1313                 ext3_msg(sb, KERN_INFO, "[bs=%lu, gc=%lu, "
1314                                 "bpg=%lu, ipg=%lu, mo=%04lx]",
1315                         sb->s_blocksize,
1316                         sbi->s_groups_count,
1317                         EXT3_BLOCKS_PER_GROUP(sb),
1318                         EXT3_INODES_PER_GROUP(sb),
1319                         sbi->s_mount_opt);
1320
1321         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1322                 char b[BDEVNAME_SIZE];
1323                 ext3_msg(sb, KERN_INFO, "using external journal on %s",
1324                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1325         } else {
1326                 ext3_msg(sb, KERN_INFO, "using internal journal");
1327         }
1328         return res;
1329 }
1330
1331 /* Called at mount-time, super-block is locked */
1332 static int ext3_check_descriptors(struct super_block *sb)
1333 {
1334         struct ext3_sb_info *sbi = EXT3_SB(sb);
1335         int i;
1336
1337         ext3_debug ("Checking group descriptors");
1338
1339         for (i = 0; i < sbi->s_groups_count; i++) {
1340                 struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL);
1341                 ext3_fsblk_t first_block = ext3_group_first_block_no(sb, i);
1342                 ext3_fsblk_t last_block;
1343
1344                 if (i == sbi->s_groups_count - 1)
1345                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
1346                 else
1347                         last_block = first_block +
1348                                 (EXT3_BLOCKS_PER_GROUP(sb) - 1);
1349
1350                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
1351                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
1352                 {
1353                         ext3_error (sb, "ext3_check_descriptors",
1354                                     "Block bitmap for group %d"
1355                                     " not in group (block %lu)!",
1356                                     i, (unsigned long)
1357                                         le32_to_cpu(gdp->bg_block_bitmap));
1358                         return 0;
1359                 }
1360                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
1361                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
1362                 {
1363                         ext3_error (sb, "ext3_check_descriptors",
1364                                     "Inode bitmap for group %d"
1365                                     " not in group (block %lu)!",
1366                                     i, (unsigned long)
1367                                         le32_to_cpu(gdp->bg_inode_bitmap));
1368                         return 0;
1369                 }
1370                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
1371                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group - 1 >
1372                     last_block)
1373                 {
1374                         ext3_error (sb, "ext3_check_descriptors",
1375                                     "Inode table for group %d"
1376                                     " not in group (block %lu)!",
1377                                     i, (unsigned long)
1378                                         le32_to_cpu(gdp->bg_inode_table));
1379                         return 0;
1380                 }
1381         }
1382
1383         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1384         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1385         return 1;
1386 }
1387
1388
1389 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1390  * the superblock) which were deleted from all directories, but held open by
1391  * a process at the time of a crash.  We walk the list and try to delete these
1392  * inodes at recovery time (only with a read-write filesystem).
1393  *
1394  * In order to keep the orphan inode chain consistent during traversal (in
1395  * case of crash during recovery), we link each inode into the superblock
1396  * orphan list_head and handle it the same way as an inode deletion during
1397  * normal operation (which journals the operations for us).
1398  *
1399  * We only do an iget() and an iput() on each inode, which is very safe if we
1400  * accidentally point at an in-use or already deleted inode.  The worst that
1401  * can happen in this case is that we get a "bit already cleared" message from
1402  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1403  * e2fsck was run on this filesystem, and it must have already done the orphan
1404  * inode cleanup for us, so we can safely abort without any further action.
1405  */
1406 static void ext3_orphan_cleanup (struct super_block * sb,
1407                                  struct ext3_super_block * es)
1408 {
1409         unsigned int s_flags = sb->s_flags;
1410         int nr_orphans = 0, nr_truncates = 0;
1411 #ifdef CONFIG_QUOTA
1412         int i;
1413 #endif
1414         if (!es->s_last_orphan) {
1415                 jbd_debug(4, "no orphan inodes to clean up\n");
1416                 return;
1417         }
1418
1419         if (bdev_read_only(sb->s_bdev)) {
1420                 ext3_msg(sb, KERN_ERR, "error: write access "
1421                         "unavailable, skipping orphan cleanup.");
1422                 return;
1423         }
1424
1425         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1426                 if (es->s_last_orphan)
1427                         jbd_debug(1, "Errors on filesystem, "
1428                                   "clearing orphan list.\n");
1429                 es->s_last_orphan = 0;
1430                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1431                 return;
1432         }
1433
1434         if (s_flags & MS_RDONLY) {
1435                 ext3_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1436                 sb->s_flags &= ~MS_RDONLY;
1437         }
1438 #ifdef CONFIG_QUOTA
1439         /* Needed for iput() to work correctly and not trash data */
1440         sb->s_flags |= MS_ACTIVE;
1441         /* Turn on quotas so that they are updated correctly */
1442         for (i = 0; i < MAXQUOTAS; i++) {
1443                 if (EXT3_SB(sb)->s_qf_names[i]) {
1444                         int ret = ext3_quota_on_mount(sb, i);
1445                         if (ret < 0)
1446                                 ext3_msg(sb, KERN_ERR,
1447                                         "error: cannot turn on journaled "
1448                                         "quota: %d", ret);
1449                 }
1450         }
1451 #endif
1452
1453         while (es->s_last_orphan) {
1454                 struct inode *inode;
1455
1456                 inode = ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1457                 if (IS_ERR(inode)) {
1458                         es->s_last_orphan = 0;
1459                         break;
1460                 }
1461
1462                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1463                 vfs_dq_init(inode);
1464                 if (inode->i_nlink) {
1465                         printk(KERN_DEBUG
1466                                 "%s: truncating inode %lu to %Ld bytes\n",
1467                                 __func__, inode->i_ino, inode->i_size);
1468                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1469                                   inode->i_ino, inode->i_size);
1470                         ext3_truncate(inode);
1471                         nr_truncates++;
1472                 } else {
1473                         printk(KERN_DEBUG
1474                                 "%s: deleting unreferenced inode %lu\n",
1475                                 __func__, inode->i_ino);
1476                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1477                                   inode->i_ino);
1478                         nr_orphans++;
1479                 }
1480                 iput(inode);  /* The delete magic happens here! */
1481         }
1482
1483 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1484
1485         if (nr_orphans)
1486                 ext3_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
1487                        PLURAL(nr_orphans));
1488         if (nr_truncates)
1489                 ext3_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
1490                        PLURAL(nr_truncates));
1491 #ifdef CONFIG_QUOTA
1492         /* Turn quotas off */
1493         for (i = 0; i < MAXQUOTAS; i++) {
1494                 if (sb_dqopt(sb)->files[i])
1495                         vfs_quota_off(sb, i, 0);
1496         }
1497 #endif
1498         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1499 }
1500
1501 /*
1502  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1503  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1504  * We need to be 1 filesystem block less than the 2^32 sector limit.
1505  */
1506 static loff_t ext3_max_size(int bits)
1507 {
1508         loff_t res = EXT3_NDIR_BLOCKS;
1509         int meta_blocks;
1510         loff_t upper_limit;
1511
1512         /* This is calculated to be the largest file size for a
1513          * dense, file such that the total number of
1514          * sectors in the file, including data and all indirect blocks,
1515          * does not exceed 2^32 -1
1516          * __u32 i_blocks representing the total number of
1517          * 512 bytes blocks of the file
1518          */
1519         upper_limit = (1LL << 32) - 1;
1520
1521         /* total blocks in file system block size */
1522         upper_limit >>= (bits - 9);
1523
1524
1525         /* indirect blocks */
1526         meta_blocks = 1;
1527         /* double indirect blocks */
1528         meta_blocks += 1 + (1LL << (bits-2));
1529         /* tripple indirect blocks */
1530         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1531
1532         upper_limit -= meta_blocks;
1533         upper_limit <<= bits;
1534
1535         res += 1LL << (bits-2);
1536         res += 1LL << (2*(bits-2));
1537         res += 1LL << (3*(bits-2));
1538         res <<= bits;
1539         if (res > upper_limit)
1540                 res = upper_limit;
1541
1542         if (res > MAX_LFS_FILESIZE)
1543                 res = MAX_LFS_FILESIZE;
1544
1545         return res;
1546 }
1547
1548 static ext3_fsblk_t descriptor_loc(struct super_block *sb,
1549                                     ext3_fsblk_t logic_sb_block,
1550                                     int nr)
1551 {
1552         struct ext3_sb_info *sbi = EXT3_SB(sb);
1553         unsigned long bg, first_meta_bg;
1554         int has_super = 0;
1555
1556         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1557
1558         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1559             nr < first_meta_bg)
1560                 return (logic_sb_block + nr + 1);
1561         bg = sbi->s_desc_per_block * nr;
1562         if (ext3_bg_has_super(sb, bg))
1563                 has_super = 1;
1564         return (has_super + ext3_group_first_block_no(sb, bg));
1565 }
1566
1567
1568 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1569 {
1570         struct buffer_head * bh;
1571         struct ext3_super_block *es = NULL;
1572         struct ext3_sb_info *sbi;
1573         ext3_fsblk_t block;
1574         ext3_fsblk_t sb_block = get_sb_block(&data, sb);
1575         ext3_fsblk_t logic_sb_block;
1576         unsigned long offset = 0;
1577         unsigned int journal_inum = 0;
1578         unsigned long journal_devnum = 0;
1579         unsigned long def_mount_opts;
1580         struct inode *root;
1581         int blocksize;
1582         int hblock;
1583         int db_count;
1584         int i;
1585         int needs_recovery;
1586         int ret = -EINVAL;
1587         __le32 features;
1588         int err;
1589
1590         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1591         if (!sbi)
1592                 return -ENOMEM;
1593
1594         sbi->s_blockgroup_lock =
1595                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
1596         if (!sbi->s_blockgroup_lock) {
1597                 kfree(sbi);
1598                 return -ENOMEM;
1599         }
1600         sb->s_fs_info = sbi;
1601         sbi->s_mount_opt = 0;
1602         sbi->s_resuid = EXT3_DEF_RESUID;
1603         sbi->s_resgid = EXT3_DEF_RESGID;
1604         sbi->s_sb_block = sb_block;
1605
1606         unlock_kernel();
1607
1608         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1609         if (!blocksize) {
1610                 ext3_msg(sb, KERN_ERR, "error: unable to set blocksize");
1611                 goto out_fail;
1612         }
1613
1614         /*
1615          * The ext3 superblock will not be buffer aligned for other than 1kB
1616          * block sizes.  We need to calculate the offset from buffer start.
1617          */
1618         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1619                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1620                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1621         } else {
1622                 logic_sb_block = sb_block;
1623         }
1624
1625         if (!(bh = sb_bread(sb, logic_sb_block))) {
1626                 ext3_msg(sb, KERN_ERR, "error: unable to read superblock");
1627                 goto out_fail;
1628         }
1629         /*
1630          * Note: s_es must be initialized as soon as possible because
1631          *       some ext3 macro-instructions depend on its value
1632          */
1633         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1634         sbi->s_es = es;
1635         sb->s_magic = le16_to_cpu(es->s_magic);
1636         if (sb->s_magic != EXT3_SUPER_MAGIC)
1637                 goto cantfind_ext3;
1638
1639         /* Set defaults before we parse the mount options */
1640         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1641         if (def_mount_opts & EXT3_DEFM_DEBUG)
1642                 set_opt(sbi->s_mount_opt, DEBUG);
1643         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1644                 set_opt(sbi->s_mount_opt, GRPID);
1645         if (def_mount_opts & EXT3_DEFM_UID16)
1646                 set_opt(sbi->s_mount_opt, NO_UID32);
1647 #ifdef CONFIG_EXT3_FS_XATTR
1648         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1649                 set_opt(sbi->s_mount_opt, XATTR_USER);
1650 #endif
1651 #ifdef CONFIG_EXT3_FS_POSIX_ACL
1652         if (def_mount_opts & EXT3_DEFM_ACL)
1653                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1654 #endif
1655         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1656                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1657         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1658                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1659         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1660                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1661
1662         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1663                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1664         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_CONTINUE)
1665                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1666         else
1667                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1668
1669         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1670         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1671
1672         set_opt(sbi->s_mount_opt, RESERVATION);
1673
1674         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1675                             NULL, 0))
1676                 goto failed_mount;
1677
1678         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1679                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1680
1681         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1682             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1683              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1684              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1685                 ext3_msg(sb, KERN_WARNING,
1686                         "warning: feature flags set on rev 0 fs, "
1687                         "running e2fsck is recommended");
1688         /*
1689          * Check feature flags regardless of the revision level, since we
1690          * previously didn't change the revision level when setting the flags,
1691          * so there is a chance incompat flags are set on a rev 0 filesystem.
1692          */
1693         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1694         if (features) {
1695                 ext3_msg(sb, KERN_ERR,
1696                         "error: couldn't mount because of unsupported "
1697                         "optional features (%x)", le32_to_cpu(features));
1698                 goto failed_mount;
1699         }
1700         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1701         if (!(sb->s_flags & MS_RDONLY) && features) {
1702                 ext3_msg(sb, KERN_ERR,
1703                         "error: couldn't mount RDWR because of unsupported "
1704                         "optional features (%x)", le32_to_cpu(features));
1705                 goto failed_mount;
1706         }
1707         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1708
1709         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1710             blocksize > EXT3_MAX_BLOCK_SIZE) {
1711                 ext3_msg(sb, KERN_ERR,
1712                         "error: couldn't mount because of unsupported "
1713                         "filesystem blocksize %d", blocksize);
1714                 goto failed_mount;
1715         }
1716
1717         hblock = bdev_logical_block_size(sb->s_bdev);
1718         if (sb->s_blocksize != blocksize) {
1719                 /*
1720                  * Make sure the blocksize for the filesystem is larger
1721                  * than the hardware sectorsize for the machine.
1722                  */
1723                 if (blocksize < hblock) {
1724                         ext3_msg(sb, KERN_ERR,
1725                                 "error: fsblocksize %d too small for "
1726                                 "hardware sectorsize %d", blocksize, hblock);
1727                         goto failed_mount;
1728                 }
1729
1730                 brelse (bh);
1731                 if (!sb_set_blocksize(sb, blocksize)) {
1732                         ext3_msg(sb, KERN_ERR,
1733                                 "error: bad blocksize %d", blocksize);
1734                         goto out_fail;
1735                 }
1736                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1737                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1738                 bh = sb_bread(sb, logic_sb_block);
1739                 if (!bh) {
1740                         ext3_msg(sb, KERN_ERR,
1741                                "error: can't read superblock on 2nd try");
1742                         goto failed_mount;
1743                 }
1744                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1745                 sbi->s_es = es;
1746                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1747                         ext3_msg(sb, KERN_ERR,
1748                                 "error: magic mismatch");
1749                         goto failed_mount;
1750                 }
1751         }
1752
1753         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1754
1755         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1756                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1757                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1758         } else {
1759                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1760                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1761                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1762                     (!is_power_of_2(sbi->s_inode_size)) ||
1763                     (sbi->s_inode_size > blocksize)) {
1764                         ext3_msg(sb, KERN_ERR,
1765                                 "error: unsupported inode size: %d",
1766                                 sbi->s_inode_size);
1767                         goto failed_mount;
1768                 }
1769         }
1770         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1771                                    le32_to_cpu(es->s_log_frag_size);
1772         if (blocksize != sbi->s_frag_size) {
1773                 ext3_msg(sb, KERN_ERR,
1774                        "error: fragsize %lu != blocksize %u (unsupported)",
1775                        sbi->s_frag_size, blocksize);
1776                 goto failed_mount;
1777         }
1778         sbi->s_frags_per_block = 1;
1779         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1780         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1781         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1782         if (EXT3_INODE_SIZE(sb) == 0 || EXT3_INODES_PER_GROUP(sb) == 0)
1783                 goto cantfind_ext3;
1784         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1785         if (sbi->s_inodes_per_block == 0)
1786                 goto cantfind_ext3;
1787         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1788                                         sbi->s_inodes_per_block;
1789         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1790         sbi->s_sbh = bh;
1791         sbi->s_mount_state = le16_to_cpu(es->s_state);
1792         sbi->s_addr_per_block_bits = ilog2(EXT3_ADDR_PER_BLOCK(sb));
1793         sbi->s_desc_per_block_bits = ilog2(EXT3_DESC_PER_BLOCK(sb));
1794         for (i=0; i < 4; i++)
1795                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1796         sbi->s_def_hash_version = es->s_def_hash_version;
1797         i = le32_to_cpu(es->s_flags);
1798         if (i & EXT2_FLAGS_UNSIGNED_HASH)
1799                 sbi->s_hash_unsigned = 3;
1800         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
1801 #ifdef __CHAR_UNSIGNED__
1802                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
1803                 sbi->s_hash_unsigned = 3;
1804 #else
1805                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
1806 #endif
1807         }
1808
1809         if (sbi->s_blocks_per_group > blocksize * 8) {
1810                 ext3_msg(sb, KERN_ERR,
1811                         "#blocks per group too big: %lu",
1812                         sbi->s_blocks_per_group);
1813                 goto failed_mount;
1814         }
1815         if (sbi->s_frags_per_group > blocksize * 8) {
1816                 ext3_msg(sb, KERN_ERR,
1817                         "error: #fragments per group too big: %lu",
1818                         sbi->s_frags_per_group);
1819                 goto failed_mount;
1820         }
1821         if (sbi->s_inodes_per_group > blocksize * 8) {
1822                 ext3_msg(sb, KERN_ERR,
1823                         "error: #inodes per group too big: %lu",
1824                         sbi->s_inodes_per_group);
1825                 goto failed_mount;
1826         }
1827
1828         if (le32_to_cpu(es->s_blocks_count) >
1829                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1830                 ext3_msg(sb, KERN_ERR,
1831                         "error: filesystem is too large to mount safely");
1832                 if (sizeof(sector_t) < 8)
1833                         ext3_msg(sb, KERN_ERR,
1834                                 "error: CONFIG_LBDAF not enabled");
1835                 goto failed_mount;
1836         }
1837
1838         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1839                 goto cantfind_ext3;
1840         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
1841                                le32_to_cpu(es->s_first_data_block) - 1)
1842                                        / EXT3_BLOCKS_PER_GROUP(sb)) + 1;
1843         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1844                    EXT3_DESC_PER_BLOCK(sb);
1845         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1846                                     GFP_KERNEL);
1847         if (sbi->s_group_desc == NULL) {
1848                 ext3_msg(sb, KERN_ERR,
1849                         "error: not enough memory");
1850                 goto failed_mount;
1851         }
1852
1853         bgl_lock_init(sbi->s_blockgroup_lock);
1854
1855         for (i = 0; i < db_count; i++) {
1856                 block = descriptor_loc(sb, logic_sb_block, i);
1857                 sbi->s_group_desc[i] = sb_bread(sb, block);
1858                 if (!sbi->s_group_desc[i]) {
1859                         ext3_msg(sb, KERN_ERR,
1860                                 "error: can't read group descriptor %d", i);
1861                         db_count = i;
1862                         goto failed_mount2;
1863                 }
1864         }
1865         if (!ext3_check_descriptors (sb)) {
1866                 ext3_msg(sb, KERN_ERR,
1867                         "error: group descriptors corrupted");
1868                 goto failed_mount2;
1869         }
1870         sbi->s_gdb_count = db_count;
1871         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1872         spin_lock_init(&sbi->s_next_gen_lock);
1873
1874         err = percpu_counter_init(&sbi->s_freeblocks_counter,
1875                         ext3_count_free_blocks(sb));
1876         if (!err) {
1877                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
1878                                 ext3_count_free_inodes(sb));
1879         }
1880         if (!err) {
1881                 err = percpu_counter_init(&sbi->s_dirs_counter,
1882                                 ext3_count_dirs(sb));
1883         }
1884         if (err) {
1885                 ext3_msg(sb, KERN_ERR, "error: insufficient memory");
1886                 goto failed_mount3;
1887         }
1888
1889         /* per fileystem reservation list head & lock */
1890         spin_lock_init(&sbi->s_rsv_window_lock);
1891         sbi->s_rsv_window_root = RB_ROOT;
1892         /* Add a single, static dummy reservation to the start of the
1893          * reservation window list --- it gives us a placeholder for
1894          * append-at-start-of-list which makes the allocation logic
1895          * _much_ simpler. */
1896         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1897         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1898         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1899         sbi->s_rsv_window_head.rsv_goal_size = 0;
1900         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1901
1902         /*
1903          * set up enough so that it can read an inode
1904          */
1905         sb->s_op = &ext3_sops;
1906         sb->s_export_op = &ext3_export_ops;
1907         sb->s_xattr = ext3_xattr_handlers;
1908 #ifdef CONFIG_QUOTA
1909         sb->s_qcop = &ext3_qctl_operations;
1910         sb->dq_op = &ext3_quota_operations;
1911 #endif
1912         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1913
1914         sb->s_root = NULL;
1915
1916         needs_recovery = (es->s_last_orphan != 0 ||
1917                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1918                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1919
1920         /*
1921          * The first inode we look at is the journal inode.  Don't try
1922          * root first: it may be modified in the journal!
1923          */
1924         if (!test_opt(sb, NOLOAD) &&
1925             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1926                 if (ext3_load_journal(sb, es, journal_devnum))
1927                         goto failed_mount3;
1928         } else if (journal_inum) {
1929                 if (ext3_create_journal(sb, es, journal_inum))
1930                         goto failed_mount3;
1931         } else {
1932                 if (!silent)
1933                         ext3_msg(sb, KERN_ERR,
1934                                 "error: no journal found. "
1935                                 "mounting ext3 over ext2?");
1936                 goto failed_mount3;
1937         }
1938
1939         /* We have now updated the journal if required, so we can
1940          * validate the data journaling mode. */
1941         switch (test_opt(sb, DATA_FLAGS)) {
1942         case 0:
1943                 /* No mode set, assume a default based on the journal
1944                    capabilities: ORDERED_DATA if the journal can
1945                    cope, else JOURNAL_DATA */
1946                 if (journal_check_available_features
1947                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1948                         set_opt(sbi->s_mount_opt, DEFAULT_DATA_MODE);
1949                 else
1950                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1951                 break;
1952
1953         case EXT3_MOUNT_ORDERED_DATA:
1954         case EXT3_MOUNT_WRITEBACK_DATA:
1955                 if (!journal_check_available_features
1956                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1957                         ext3_msg(sb, KERN_ERR,
1958                                 "error: journal does not support "
1959                                 "requested data journaling mode");
1960                         goto failed_mount4;
1961                 }
1962         default:
1963                 break;
1964         }
1965
1966         if (test_opt(sb, NOBH)) {
1967                 if (!(test_opt(sb, DATA_FLAGS) == EXT3_MOUNT_WRITEBACK_DATA)) {
1968                         ext3_msg(sb, KERN_WARNING,
1969                                 "warning: ignoring nobh option - "
1970                                 "it is supported only with writeback mode");
1971                         clear_opt(sbi->s_mount_opt, NOBH);
1972                 }
1973         }
1974         /*
1975          * The journal_load will have done any necessary log recovery,
1976          * so we can safely mount the rest of the filesystem now.
1977          */
1978
1979         root = ext3_iget(sb, EXT3_ROOT_INO);
1980         if (IS_ERR(root)) {
1981                 ext3_msg(sb, KERN_ERR, "error: get root inode failed");
1982                 ret = PTR_ERR(root);
1983                 goto failed_mount4;
1984         }
1985         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1986                 iput(root);
1987                 ext3_msg(sb, KERN_ERR, "error: corrupt root inode, run e2fsck");
1988                 goto failed_mount4;
1989         }
1990         sb->s_root = d_alloc_root(root);
1991         if (!sb->s_root) {
1992                 ext3_msg(sb, KERN_ERR, "error: get root dentry failed");
1993                 iput(root);
1994                 ret = -ENOMEM;
1995                 goto failed_mount4;
1996         }
1997
1998         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1999         /*
2000          * akpm: core read_super() calls in here with the superblock locked.
2001          * That deadlocks, because orphan cleanup needs to lock the superblock
2002          * in numerous places.  Here we just pop the lock - it's relatively
2003          * harmless, because we are now ready to accept write_super() requests,
2004          * and aviro says that's the only reason for hanging onto the
2005          * superblock lock.
2006          */
2007         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
2008         ext3_orphan_cleanup(sb, es);
2009         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
2010         if (needs_recovery)
2011                 ext3_msg(sb, KERN_INFO, "recovery complete");
2012         ext3_mark_recovery_complete(sb, es);
2013         ext3_msg(sb, KERN_INFO, "mounted filesystem with %s data mode",
2014                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
2015                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
2016                 "writeback");
2017
2018         lock_kernel();
2019         return 0;
2020
2021 cantfind_ext3:
2022         if (!silent)
2023                 ext3_msg(sb, KERN_INFO,
2024                         "error: can't find ext3 filesystem on dev %s.",
2025                        sb->s_id);
2026         goto failed_mount;
2027
2028 failed_mount4:
2029         journal_destroy(sbi->s_journal);
2030 failed_mount3:
2031         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2032         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2033         percpu_counter_destroy(&sbi->s_dirs_counter);
2034 failed_mount2:
2035         for (i = 0; i < db_count; i++)
2036                 brelse(sbi->s_group_desc[i]);
2037         kfree(sbi->s_group_desc);
2038 failed_mount:
2039 #ifdef CONFIG_QUOTA
2040         for (i = 0; i < MAXQUOTAS; i++)
2041                 kfree(sbi->s_qf_names[i]);
2042 #endif
2043         ext3_blkdev_remove(sbi);
2044         brelse(bh);
2045 out_fail:
2046         sb->s_fs_info = NULL;
2047         kfree(sbi->s_blockgroup_lock);
2048         kfree(sbi);
2049         lock_kernel();
2050         return ret;
2051 }
2052
2053 /*
2054  * Setup any per-fs journal parameters now.  We'll do this both on
2055  * initial mount, once the journal has been initialised but before we've
2056  * done any recovery; and again on any subsequent remount.
2057  */
2058 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
2059 {
2060         struct ext3_sb_info *sbi = EXT3_SB(sb);
2061
2062         if (sbi->s_commit_interval)
2063                 journal->j_commit_interval = sbi->s_commit_interval;
2064         /* We could also set up an ext3-specific default for the commit
2065          * interval here, but for now we'll just fall back to the jbd
2066          * default. */
2067
2068         spin_lock(&journal->j_state_lock);
2069         if (test_opt(sb, BARRIER))
2070                 journal->j_flags |= JFS_BARRIER;
2071         else
2072                 journal->j_flags &= ~JFS_BARRIER;
2073         if (test_opt(sb, DATA_ERR_ABORT))
2074                 journal->j_flags |= JFS_ABORT_ON_SYNCDATA_ERR;
2075         else
2076                 journal->j_flags &= ~JFS_ABORT_ON_SYNCDATA_ERR;
2077         spin_unlock(&journal->j_state_lock);
2078 }
2079
2080 static journal_t *ext3_get_journal(struct super_block *sb,
2081                                    unsigned int journal_inum)
2082 {
2083         struct inode *journal_inode;
2084         journal_t *journal;
2085
2086         /* First, test for the existence of a valid inode on disk.  Bad
2087          * things happen if we iget() an unused inode, as the subsequent
2088          * iput() will try to delete it. */
2089
2090         journal_inode = ext3_iget(sb, journal_inum);
2091         if (IS_ERR(journal_inode)) {
2092                 ext3_msg(sb, KERN_ERR, "error: no journal found");
2093                 return NULL;
2094         }
2095         if (!journal_inode->i_nlink) {
2096                 make_bad_inode(journal_inode);
2097                 iput(journal_inode);
2098                 ext3_msg(sb, KERN_ERR, "error: journal inode is deleted");
2099                 return NULL;
2100         }
2101
2102         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
2103                   journal_inode, journal_inode->i_size);
2104         if (!S_ISREG(journal_inode->i_mode)) {
2105                 ext3_msg(sb, KERN_ERR, "error: invalid journal inode");
2106                 iput(journal_inode);
2107                 return NULL;
2108         }
2109
2110         journal = journal_init_inode(journal_inode);
2111         if (!journal) {
2112                 ext3_msg(sb, KERN_ERR, "error: could not load journal inode");
2113                 iput(journal_inode);
2114                 return NULL;
2115         }
2116         journal->j_private = sb;
2117         ext3_init_journal_params(sb, journal);
2118         return journal;
2119 }
2120
2121 static journal_t *ext3_get_dev_journal(struct super_block *sb,
2122                                        dev_t j_dev)
2123 {
2124         struct buffer_head * bh;
2125         journal_t *journal;
2126         ext3_fsblk_t start;
2127         ext3_fsblk_t len;
2128         int hblock, blocksize;
2129         ext3_fsblk_t sb_block;
2130         unsigned long offset;
2131         struct ext3_super_block * es;
2132         struct block_device *bdev;
2133
2134         bdev = ext3_blkdev_get(j_dev, sb);
2135         if (bdev == NULL)
2136                 return NULL;
2137
2138         if (bd_claim(bdev, sb)) {
2139                 ext3_msg(sb, KERN_ERR,
2140                         "error: failed to claim external journal device");
2141                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2142                 return NULL;
2143         }
2144
2145         blocksize = sb->s_blocksize;
2146         hblock = bdev_logical_block_size(bdev);
2147         if (blocksize < hblock) {
2148                 ext3_msg(sb, KERN_ERR,
2149                         "error: blocksize too small for journal device");
2150                 goto out_bdev;
2151         }
2152
2153         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
2154         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
2155         set_blocksize(bdev, blocksize);
2156         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2157                 ext3_msg(sb, KERN_ERR, "error: couldn't read superblock of "
2158                         "external journal");
2159                 goto out_bdev;
2160         }
2161
2162         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
2163         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
2164             !(le32_to_cpu(es->s_feature_incompat) &
2165               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2166                 ext3_msg(sb, KERN_ERR, "error: external journal has "
2167                         "bad superblock");
2168                 brelse(bh);
2169                 goto out_bdev;
2170         }
2171
2172         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2173                 ext3_msg(sb, KERN_ERR, "error: journal UUID does not match");
2174                 brelse(bh);
2175                 goto out_bdev;
2176         }
2177
2178         len = le32_to_cpu(es->s_blocks_count);
2179         start = sb_block + 1;
2180         brelse(bh);     /* we're done with the superblock */
2181
2182         journal = journal_init_dev(bdev, sb->s_bdev,
2183                                         start, len, blocksize);
2184         if (!journal) {
2185                 ext3_msg(sb, KERN_ERR,
2186                         "error: failed to create device journal");
2187                 goto out_bdev;
2188         }
2189         journal->j_private = sb;
2190         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2191         wait_on_buffer(journal->j_sb_buffer);
2192         if (!buffer_uptodate(journal->j_sb_buffer)) {
2193                 ext3_msg(sb, KERN_ERR, "I/O error on journal device");
2194                 goto out_journal;
2195         }
2196         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2197                 ext3_msg(sb, KERN_ERR,
2198                         "error: external journal has more than one "
2199                         "user (unsupported) - %d",
2200                         be32_to_cpu(journal->j_superblock->s_nr_users));
2201                 goto out_journal;
2202         }
2203         EXT3_SB(sb)->journal_bdev = bdev;
2204         ext3_init_journal_params(sb, journal);
2205         return journal;
2206 out_journal:
2207         journal_destroy(journal);
2208 out_bdev:
2209         ext3_blkdev_put(bdev);
2210         return NULL;
2211 }
2212
2213 static int ext3_load_journal(struct super_block *sb,
2214                              struct ext3_super_block *es,
2215                              unsigned long journal_devnum)
2216 {
2217         journal_t *journal;
2218         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2219         dev_t journal_dev;
2220         int err = 0;
2221         int really_read_only;
2222
2223         if (journal_devnum &&
2224             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2225                 ext3_msg(sb, KERN_INFO, "external journal device major/minor "
2226                         "numbers have changed");
2227                 journal_dev = new_decode_dev(journal_devnum);
2228         } else
2229                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2230
2231         really_read_only = bdev_read_only(sb->s_bdev);
2232
2233         /*
2234          * Are we loading a blank journal or performing recovery after a
2235          * crash?  For recovery, we need to check in advance whether we
2236          * can get read-write access to the device.
2237          */
2238
2239         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
2240                 if (sb->s_flags & MS_RDONLY) {
2241                         ext3_msg(sb, KERN_INFO,
2242                                 "recovery required on readonly filesystem");
2243                         if (really_read_only) {
2244                                 ext3_msg(sb, KERN_ERR, "error: write access "
2245                                         "unavailable, cannot proceed");
2246                                 return -EROFS;
2247                         }
2248                         ext3_msg(sb, KERN_INFO,
2249                                 "write access will be enabled during recovery");
2250                 }
2251         }
2252
2253         if (journal_inum && journal_dev) {
2254                 ext3_msg(sb, KERN_ERR, "error: filesystem has both journal "
2255                        "and inode journals");
2256                 return -EINVAL;
2257         }
2258
2259         if (journal_inum) {
2260                 if (!(journal = ext3_get_journal(sb, journal_inum)))
2261                         return -EINVAL;
2262         } else {
2263                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
2264                         return -EINVAL;
2265         }
2266
2267         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2268                 err = journal_update_format(journal);
2269                 if (err)  {
2270                         ext3_msg(sb, KERN_ERR, "error updating journal");
2271                         journal_destroy(journal);
2272                         return err;
2273                 }
2274         }
2275
2276         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
2277                 err = journal_wipe(journal, !really_read_only);
2278         if (!err)
2279                 err = journal_load(journal);
2280
2281         if (err) {
2282                 ext3_msg(sb, KERN_ERR, "error loading journal");
2283                 journal_destroy(journal);
2284                 return err;
2285         }
2286
2287         EXT3_SB(sb)->s_journal = journal;
2288         ext3_clear_journal_err(sb, es);
2289
2290         if (journal_devnum &&
2291             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2292                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2293
2294                 /* Make sure we flush the recovery flag to disk. */
2295                 ext3_commit_super(sb, es, 1);
2296         }
2297
2298         return 0;
2299 }
2300
2301 static int ext3_create_journal(struct super_block *sb,
2302                                struct ext3_super_block *es,
2303                                unsigned int journal_inum)
2304 {
2305         journal_t *journal;
2306         int err;
2307
2308         if (sb->s_flags & MS_RDONLY) {
2309                 ext3_msg(sb, KERN_ERR,
2310                         "error: readonly filesystem when trying to "
2311                         "create journal");
2312                 return -EROFS;
2313         }
2314
2315         journal = ext3_get_journal(sb, journal_inum);
2316         if (!journal)
2317                 return -EINVAL;
2318
2319         ext3_msg(sb, KERN_INFO, "creating new journal on inode %u",
2320                journal_inum);
2321
2322         err = journal_create(journal);
2323         if (err) {
2324                 ext3_msg(sb, KERN_ERR, "error creating journal");
2325                 journal_destroy(journal);
2326                 return -EIO;
2327         }
2328
2329         EXT3_SB(sb)->s_journal = journal;
2330
2331         ext3_update_dynamic_rev(sb);
2332         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2333         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
2334
2335         es->s_journal_inum = cpu_to_le32(journal_inum);
2336
2337         /* Make sure we flush the recovery flag to disk. */
2338         ext3_commit_super(sb, es, 1);
2339
2340         return 0;
2341 }
2342
2343 static int ext3_commit_super(struct super_block *sb,
2344                                struct ext3_super_block *es,
2345                                int sync)
2346 {
2347         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
2348         int error = 0;
2349
2350         if (!sbh)
2351                 return error;
2352         /*
2353          * If the file system is mounted read-only, don't update the
2354          * superblock write time.  This avoids updating the superblock
2355          * write time when we are mounting the root file system
2356          * read/only but we need to replay the journal; at that point,
2357          * for people who are east of GMT and who make their clock
2358          * tick in localtime for Windows bug-for-bug compatibility,
2359          * the clock is set in the future, and this will cause e2fsck
2360          * to complain and force a full file system check.
2361          */
2362         if (!(sb->s_flags & MS_RDONLY))
2363                 es->s_wtime = cpu_to_le32(get_seconds());
2364         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
2365         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
2366         BUFFER_TRACE(sbh, "marking dirty");
2367         mark_buffer_dirty(sbh);
2368         if (sync)
2369                 error = sync_dirty_buffer(sbh);
2370         return error;
2371 }
2372
2373
2374 /*
2375  * Have we just finished recovery?  If so, and if we are mounting (or
2376  * remounting) the filesystem readonly, then we will end up with a
2377  * consistent fs on disk.  Record that fact.
2378  */
2379 static void ext3_mark_recovery_complete(struct super_block * sb,
2380                                         struct ext3_super_block * es)
2381 {
2382         journal_t *journal = EXT3_SB(sb)->s_journal;
2383
2384         journal_lock_updates(journal);
2385         if (journal_flush(journal) < 0)
2386                 goto out;
2387
2388         lock_super(sb);
2389         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
2390             sb->s_flags & MS_RDONLY) {
2391                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2392                 ext3_commit_super(sb, es, 1);
2393         }
2394         unlock_super(sb);
2395
2396 out:
2397         journal_unlock_updates(journal);
2398 }
2399
2400 /*
2401  * If we are mounting (or read-write remounting) a filesystem whose journal
2402  * has recorded an error from a previous lifetime, move that error to the
2403  * main filesystem now.
2404  */
2405 static void ext3_clear_journal_err(struct super_block *sb,
2406                                    struct ext3_super_block *es)
2407 {
2408         journal_t *journal;
2409         int j_errno;
2410         const char *errstr;
2411
2412         journal = EXT3_SB(sb)->s_journal;
2413
2414         /*
2415          * Now check for any error status which may have been recorded in the
2416          * journal by a prior ext3_error() or ext3_abort()
2417          */
2418
2419         j_errno = journal_errno(journal);
2420         if (j_errno) {
2421                 char nbuf[16];
2422
2423                 errstr = ext3_decode_error(sb, j_errno, nbuf);
2424                 ext3_warning(sb, __func__, "Filesystem error recorded "
2425                              "from previous mount: %s", errstr);
2426                 ext3_warning(sb, __func__, "Marking fs in need of "
2427                              "filesystem check.");
2428
2429                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
2430                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
2431                 ext3_commit_super (sb, es, 1);
2432
2433                 journal_clear_err(journal);
2434         }
2435 }
2436
2437 /*
2438  * Force the running and committing transactions to commit,
2439  * and wait on the commit.
2440  */
2441 int ext3_force_commit(struct super_block *sb)
2442 {
2443         journal_t *journal;
2444         int ret;
2445
2446         if (sb->s_flags & MS_RDONLY)
2447                 return 0;
2448
2449         journal = EXT3_SB(sb)->s_journal;
2450         ret = ext3_journal_force_commit(journal);
2451         return ret;
2452 }
2453
2454 static int ext3_sync_fs(struct super_block *sb, int wait)
2455 {
2456         tid_t target;
2457
2458         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2459                 if (wait)
2460                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2461         }
2462         return 0;
2463 }
2464
2465 /*
2466  * LVM calls this function before a (read-only) snapshot is created.  This
2467  * gives us a chance to flush the journal completely and mark the fs clean.
2468  */
2469 static int ext3_freeze(struct super_block *sb)
2470 {
2471         int error = 0;
2472         journal_t *journal;
2473
2474         if (!(sb->s_flags & MS_RDONLY)) {
2475                 journal = EXT3_SB(sb)->s_journal;
2476
2477                 /* Now we set up the journal barrier. */
2478                 journal_lock_updates(journal);
2479
2480                 /*
2481                  * We don't want to clear needs_recovery flag when we failed
2482                  * to flush the journal.
2483                  */
2484                 error = journal_flush(journal);
2485                 if (error < 0)
2486                         goto out;
2487
2488                 /* Journal blocked and flushed, clear needs_recovery flag. */
2489                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2490                 error = ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2491                 if (error)
2492                         goto out;
2493         }
2494         return 0;
2495
2496 out:
2497         journal_unlock_updates(journal);
2498         return error;
2499 }
2500
2501 /*
2502  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2503  * flag here, even though the filesystem is not technically dirty yet.
2504  */
2505 static int ext3_unfreeze(struct super_block *sb)
2506 {
2507         if (!(sb->s_flags & MS_RDONLY)) {
2508                 lock_super(sb);
2509                 /* Reser the needs_recovery flag before the fs is unlocked. */
2510                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2511                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2512                 unlock_super(sb);
2513                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2514         }
2515         return 0;
2516 }
2517
2518 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2519 {
2520         struct ext3_super_block * es;
2521         struct ext3_sb_info *sbi = EXT3_SB(sb);
2522         ext3_fsblk_t n_blocks_count = 0;
2523         unsigned long old_sb_flags;
2524         struct ext3_mount_options old_opts;
2525         int err;
2526 #ifdef CONFIG_QUOTA
2527         int i;
2528 #endif
2529
2530         lock_kernel();
2531
2532         /* Store the original options */
2533         lock_super(sb);
2534         old_sb_flags = sb->s_flags;
2535         old_opts.s_mount_opt = sbi->s_mount_opt;
2536         old_opts.s_resuid = sbi->s_resuid;
2537         old_opts.s_resgid = sbi->s_resgid;
2538         old_opts.s_commit_interval = sbi->s_commit_interval;
2539 #ifdef CONFIG_QUOTA
2540         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2541         for (i = 0; i < MAXQUOTAS; i++)
2542                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2543 #endif
2544
2545         /*
2546          * Allow the "check" option to be passed as a remount option.
2547          */
2548         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2549                 err = -EINVAL;
2550                 goto restore_opts;
2551         }
2552
2553         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2554                 ext3_abort(sb, __func__, "Abort forced by user");
2555
2556         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2557                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2558
2559         es = sbi->s_es;
2560
2561         ext3_init_journal_params(sb, sbi->s_journal);
2562
2563         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2564                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2565                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT) {
2566                         err = -EROFS;
2567                         goto restore_opts;
2568                 }
2569
2570                 if (*flags & MS_RDONLY) {
2571                         /*
2572                          * First of all, the unconditional stuff we have to do
2573                          * to disable replay of the journal when we next remount
2574                          */
2575                         sb->s_flags |= MS_RDONLY;
2576
2577                         /*
2578                          * OK, test if we are remounting a valid rw partition
2579                          * readonly, and if so set the rdonly flag and then
2580                          * mark the partition as valid again.
2581                          */
2582                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2583                             (sbi->s_mount_state & EXT3_VALID_FS))
2584                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2585
2586                         /*
2587                          * We have to unlock super so that we can wait for
2588                          * transactions.
2589                          */
2590                         unlock_super(sb);
2591                         ext3_mark_recovery_complete(sb, es);
2592                         lock_super(sb);
2593                 } else {
2594                         __le32 ret;
2595                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2596                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2597                                 ext3_msg(sb, KERN_WARNING,
2598                                         "warning: couldn't remount RDWR "
2599                                         "because of unsupported optional "
2600                                         "features (%x)", le32_to_cpu(ret));
2601                                 err = -EROFS;
2602                                 goto restore_opts;
2603                         }
2604
2605                         /*
2606                          * If we have an unprocessed orphan list hanging
2607                          * around from a previously readonly bdev mount,
2608                          * require a full umount/remount for now.
2609                          */
2610                         if (es->s_last_orphan) {
2611                                 ext3_msg(sb, KERN_WARNING, "warning: couldn't "
2612                                        "remount RDWR because of unprocessed "
2613                                        "orphan inode list.  Please "
2614                                        "umount/remount instead.");
2615                                 err = -EINVAL;
2616                                 goto restore_opts;
2617                         }
2618
2619                         /*
2620                          * Mounting a RDONLY partition read-write, so reread
2621                          * and store the current valid flag.  (It may have
2622                          * been changed by e2fsck since we originally mounted
2623                          * the partition.)
2624                          */
2625                         ext3_clear_journal_err(sb, es);
2626                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2627                         if ((err = ext3_group_extend(sb, es, n_blocks_count)))
2628                                 goto restore_opts;
2629                         if (!ext3_setup_super (sb, es, 0))
2630                                 sb->s_flags &= ~MS_RDONLY;
2631                 }
2632         }
2633 #ifdef CONFIG_QUOTA
2634         /* Release old quota file names */
2635         for (i = 0; i < MAXQUOTAS; i++)
2636                 if (old_opts.s_qf_names[i] &&
2637                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2638                         kfree(old_opts.s_qf_names[i]);
2639 #endif
2640         unlock_super(sb);
2641         unlock_kernel();
2642         return 0;
2643 restore_opts:
2644         sb->s_flags = old_sb_flags;
2645         sbi->s_mount_opt = old_opts.s_mount_opt;
2646         sbi->s_resuid = old_opts.s_resuid;
2647         sbi->s_resgid = old_opts.s_resgid;
2648         sbi->s_commit_interval = old_opts.s_commit_interval;
2649 #ifdef CONFIG_QUOTA
2650         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2651         for (i = 0; i < MAXQUOTAS; i++) {
2652                 if (sbi->s_qf_names[i] &&
2653                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2654                         kfree(sbi->s_qf_names[i]);
2655                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2656         }
2657 #endif
2658         unlock_super(sb);
2659         unlock_kernel();
2660         return err;
2661 }
2662
2663 static int ext3_statfs (struct dentry * dentry, struct kstatfs * buf)
2664 {
2665         struct super_block *sb = dentry->d_sb;
2666         struct ext3_sb_info *sbi = EXT3_SB(sb);
2667         struct ext3_super_block *es = sbi->s_es;
2668         u64 fsid;
2669
2670         if (test_opt(sb, MINIX_DF)) {
2671                 sbi->s_overhead_last = 0;
2672         } else if (sbi->s_blocks_last != le32_to_cpu(es->s_blocks_count)) {
2673                 unsigned long ngroups = sbi->s_groups_count, i;
2674                 ext3_fsblk_t overhead = 0;
2675                 smp_rmb();
2676
2677                 /*
2678                  * Compute the overhead (FS structures).  This is constant
2679                  * for a given filesystem unless the number of block groups
2680                  * changes so we cache the previous value until it does.
2681                  */
2682
2683                 /*
2684                  * All of the blocks before first_data_block are
2685                  * overhead
2686                  */
2687                 overhead = le32_to_cpu(es->s_first_data_block);
2688
2689                 /*
2690                  * Add the overhead attributed to the superblock and
2691                  * block group descriptors.  If the sparse superblocks
2692                  * feature is turned on, then not all groups have this.
2693                  */
2694                 for (i = 0; i < ngroups; i++) {
2695                         overhead += ext3_bg_has_super(sb, i) +
2696                                 ext3_bg_num_gdb(sb, i);
2697                         cond_resched();
2698                 }
2699
2700                 /*
2701                  * Every block group has an inode bitmap, a block
2702                  * bitmap, and an inode table.
2703                  */
2704                 overhead += ngroups * (2 + sbi->s_itb_per_group);
2705                 sbi->s_overhead_last = overhead;
2706                 smp_wmb();
2707                 sbi->s_blocks_last = le32_to_cpu(es->s_blocks_count);
2708         }
2709
2710         buf->f_type = EXT3_SUPER_MAGIC;
2711         buf->f_bsize = sb->s_blocksize;
2712         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - sbi->s_overhead_last;
2713         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter);
2714         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2715         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2716                 buf->f_bavail = 0;
2717         buf->f_files = le32_to_cpu(es->s_inodes_count);
2718         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
2719         buf->f_namelen = EXT3_NAME_LEN;
2720         fsid = le64_to_cpup((void *)es->s_uuid) ^
2721                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
2722         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
2723         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
2724         return 0;
2725 }
2726
2727 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2728  * is locked for write. Otherwise the are possible deadlocks:
2729  * Process 1                         Process 2
2730  * ext3_create()                     quota_sync()
2731  *   journal_start()                   write_dquot()
2732  *   vfs_dq_init()                       down(dqio_mutex)
2733  *     down(dqio_mutex)                    journal_start()
2734  *
2735  */
2736
2737 #ifdef CONFIG_QUOTA
2738
2739 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2740 {
2741         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2742 }
2743
2744 static int ext3_write_dquot(struct dquot *dquot)
2745 {
2746         int ret, err;
2747         handle_t *handle;
2748         struct inode *inode;
2749
2750         inode = dquot_to_inode(dquot);
2751         handle = ext3_journal_start(inode,
2752                                         EXT3_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2753         if (IS_ERR(handle))
2754                 return PTR_ERR(handle);
2755         ret = dquot_commit(dquot);
2756         err = ext3_journal_stop(handle);
2757         if (!ret)
2758                 ret = err;
2759         return ret;
2760 }
2761
2762 static int ext3_acquire_dquot(struct dquot *dquot)
2763 {
2764         int ret, err;
2765         handle_t *handle;
2766
2767         handle = ext3_journal_start(dquot_to_inode(dquot),
2768                                         EXT3_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2769         if (IS_ERR(handle))
2770                 return PTR_ERR(handle);
2771         ret = dquot_acquire(dquot);
2772         err = ext3_journal_stop(handle);
2773         if (!ret)
2774                 ret = err;
2775         return ret;
2776 }
2777
2778 static int ext3_release_dquot(struct dquot *dquot)
2779 {
2780         int ret, err;
2781         handle_t *handle;
2782
2783         handle = ext3_journal_start(dquot_to_inode(dquot),
2784                                         EXT3_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2785         if (IS_ERR(handle)) {
2786                 /* Release dquot anyway to avoid endless cycle in dqput() */
2787                 dquot_release(dquot);
2788                 return PTR_ERR(handle);
2789         }
2790         ret = dquot_release(dquot);
2791         err = ext3_journal_stop(handle);
2792         if (!ret)
2793                 ret = err;
2794         return ret;
2795 }
2796
2797 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2798 {
2799         /* Are we journaling quotas? */
2800         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2801             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2802                 dquot_mark_dquot_dirty(dquot);
2803                 return ext3_write_dquot(dquot);
2804         } else {
2805                 return dquot_mark_dquot_dirty(dquot);
2806         }
2807 }
2808
2809 static int ext3_write_info(struct super_block *sb, int type)
2810 {
2811         int ret, err;
2812         handle_t *handle;
2813
2814         /* Data block + inode block */
2815         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2816         if (IS_ERR(handle))
2817                 return PTR_ERR(handle);
2818         ret = dquot_commit_info(sb, type);
2819         err = ext3_journal_stop(handle);
2820         if (!ret)
2821                 ret = err;
2822         return ret;
2823 }
2824
2825 /*
2826  * Turn on quotas during mount time - we need to find
2827  * the quota file and such...
2828  */
2829 static int ext3_quota_on_mount(struct super_block *sb, int type)
2830 {
2831         return vfs_quota_on_mount(sb, EXT3_SB(sb)->s_qf_names[type],
2832                         EXT3_SB(sb)->s_jquota_fmt, type);
2833 }
2834
2835 /*
2836  * Standard function to be called on quota_on
2837  */
2838 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2839                          char *name, int remount)
2840 {
2841         int err;
2842         struct path path;
2843
2844         if (!test_opt(sb, QUOTA))
2845                 return -EINVAL;
2846         /* When remounting, no checks are needed and in fact, name is NULL */
2847         if (remount)
2848                 return vfs_quota_on(sb, type, format_id, name, remount);
2849
2850         err = kern_path(name, LOOKUP_FOLLOW, &path);
2851         if (err)
2852                 return err;
2853
2854         /* Quotafile not on the same filesystem? */
2855         if (path.mnt->mnt_sb != sb) {
2856                 path_put(&path);
2857                 return -EXDEV;
2858         }
2859         /* Journaling quota? */
2860         if (EXT3_SB(sb)->s_qf_names[type]) {
2861                 /* Quotafile not of fs root? */
2862                 if (path.dentry->d_parent != sb->s_root)
2863                         ext3_msg(sb, KERN_WARNING,
2864                                 "warning: Quota file not on filesystem root. "
2865                                 "Journaled quota will not work.");
2866         }
2867
2868         /*
2869          * When we journal data on quota file, we have to flush journal to see
2870          * all updates to the file when we bypass pagecache...
2871          */
2872         if (ext3_should_journal_data(path.dentry->d_inode)) {
2873                 /*
2874                  * We don't need to lock updates but journal_flush() could
2875                  * otherwise be livelocked...
2876                  */
2877                 journal_lock_updates(EXT3_SB(sb)->s_journal);
2878                 err = journal_flush(EXT3_SB(sb)->s_journal);
2879                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2880                 if (err) {
2881                         path_put(&path);
2882                         return err;
2883                 }
2884         }
2885
2886         err = vfs_quota_on_path(sb, type, format_id, &path);
2887         path_put(&path);
2888         return err;
2889 }
2890
2891 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2892  * acquiring the locks... As quota files are never truncated and quota code
2893  * itself serializes the operations (and noone else should touch the files)
2894  * we don't have to be afraid of races */
2895 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2896                                size_t len, loff_t off)
2897 {
2898         struct inode *inode = sb_dqopt(sb)->files[type];
2899         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2900         int err = 0;
2901         int offset = off & (sb->s_blocksize - 1);
2902         int tocopy;
2903         size_t toread;
2904         struct buffer_head *bh;
2905         loff_t i_size = i_size_read(inode);
2906
2907         if (off > i_size)
2908                 return 0;
2909         if (off+len > i_size)
2910                 len = i_size-off;
2911         toread = len;
2912         while (toread > 0) {
2913                 tocopy = sb->s_blocksize - offset < toread ?
2914                                 sb->s_blocksize - offset : toread;
2915                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2916                 if (err)
2917                         return err;
2918                 if (!bh)        /* A hole? */
2919                         memset(data, 0, tocopy);
2920                 else
2921                         memcpy(data, bh->b_data+offset, tocopy);
2922                 brelse(bh);
2923                 offset = 0;
2924                 toread -= tocopy;
2925                 data += tocopy;
2926                 blk++;
2927         }
2928         return len;
2929 }
2930
2931 /* Write to quotafile (we know the transaction is already started and has
2932  * enough credits) */
2933 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2934                                 const char *data, size_t len, loff_t off)
2935 {
2936         struct inode *inode = sb_dqopt(sb)->files[type];
2937         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2938         int err = 0;
2939         int offset = off & (sb->s_blocksize - 1);
2940         int tocopy;
2941         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2942         size_t towrite = len;
2943         struct buffer_head *bh;
2944         handle_t *handle = journal_current_handle();
2945
2946         if (!handle) {
2947                 ext3_msg(sb, KERN_WARNING,
2948                         "warning: quota write (off=%llu, len=%llu)"
2949                         " cancelled because transaction is not started.",
2950                         (unsigned long long)off, (unsigned long long)len);
2951                 return -EIO;
2952         }
2953         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2954         while (towrite > 0) {
2955                 tocopy = sb->s_blocksize - offset < towrite ?
2956                                 sb->s_blocksize - offset : towrite;
2957                 bh = ext3_bread(handle, inode, blk, 1, &err);
2958                 if (!bh)
2959                         goto out;
2960                 if (journal_quota) {
2961                         err = ext3_journal_get_write_access(handle, bh);
2962                         if (err) {
2963                                 brelse(bh);
2964                                 goto out;
2965                         }
2966                 }
2967                 lock_buffer(bh);
2968                 memcpy(bh->b_data+offset, data, tocopy);
2969                 flush_dcache_page(bh->b_page);
2970                 unlock_buffer(bh);
2971                 if (journal_quota)
2972                         err = ext3_journal_dirty_metadata(handle, bh);
2973                 else {
2974                         /* Always do at least ordered writes for quotas */
2975                         err = ext3_journal_dirty_data(handle, bh);
2976                         mark_buffer_dirty(bh);
2977                 }
2978                 brelse(bh);
2979                 if (err)
2980                         goto out;
2981                 offset = 0;
2982                 towrite -= tocopy;
2983                 data += tocopy;
2984                 blk++;
2985         }
2986 out:
2987         if (len == towrite) {
2988                 mutex_unlock(&inode->i_mutex);
2989                 return err;
2990         }
2991         if (inode->i_size < off+len-towrite) {
2992                 i_size_write(inode, off+len-towrite);
2993                 EXT3_I(inode)->i_disksize = inode->i_size;
2994         }
2995         inode->i_version++;
2996         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2997         ext3_mark_inode_dirty(handle, inode);
2998         mutex_unlock(&inode->i_mutex);
2999         return len - towrite;
3000 }
3001
3002 #endif
3003
3004 static int ext3_get_sb(struct file_system_type *fs_type,
3005         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3006 {
3007         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super, mnt);
3008 }
3009
3010 static struct file_system_type ext3_fs_type = {
3011         .owner          = THIS_MODULE,
3012         .name           = "ext3",
3013         .get_sb         = ext3_get_sb,
3014         .kill_sb        = kill_block_super,
3015         .fs_flags       = FS_REQUIRES_DEV,
3016 };
3017
3018 static int __init init_ext3_fs(void)
3019 {
3020         int err = init_ext3_xattr();
3021         if (err)
3022                 return err;
3023         err = init_inodecache();
3024         if (err)
3025                 goto out1;
3026         err = register_filesystem(&ext3_fs_type);
3027         if (err)
3028                 goto out;
3029         return 0;
3030 out:
3031         destroy_inodecache();
3032 out1:
3033         exit_ext3_xattr();
3034         return err;
3035 }
3036
3037 static void __exit exit_ext3_fs(void)
3038 {
3039         unregister_filesystem(&ext3_fs_type);
3040         destroy_inodecache();
3041         exit_ext3_xattr();
3042 }
3043
3044 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3045 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
3046 MODULE_LICENSE("GPL");
3047 module_init(init_ext3_fs)
3048 module_exit(exit_ext3_fs)