c730cbc84030e6466522130f01639cde3652f808
[linux-3.10.git] / fs / ext4 / super.c
1 /*
2  *  linux/fs/ext4/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/jbd2.h>
24 #include <linux/ext4_fs.h>
25 #include <linux/ext4_jbd2.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/vfs.h>
33 #include <linux/random.h>
34 #include <linux/mount.h>
35 #include <linux/namei.h>
36 #include <linux/quotaops.h>
37 #include <linux/seq_file.h>
38
39 #include <asm/uaccess.h>
40
41 #include "xattr.h"
42 #include "acl.h"
43 #include "namei.h"
44
45 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
46                              unsigned long journal_devnum);
47 static int ext4_create_journal(struct super_block *, struct ext4_super_block *,
48                                unsigned int);
49 static void ext4_commit_super (struct super_block * sb,
50                                struct ext4_super_block * es,
51                                int sync);
52 static void ext4_mark_recovery_complete(struct super_block * sb,
53                                         struct ext4_super_block * es);
54 static void ext4_clear_journal_err(struct super_block * sb,
55                                    struct ext4_super_block * es);
56 static int ext4_sync_fs(struct super_block *sb, int wait);
57 static const char *ext4_decode_error(struct super_block * sb, int errno,
58                                      char nbuf[16]);
59 static int ext4_remount (struct super_block * sb, int * flags, char * data);
60 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf);
61 static void ext4_unlockfs(struct super_block *sb);
62 static void ext4_write_super (struct super_block * sb);
63 static void ext4_write_super_lockfs(struct super_block *sb);
64
65
66 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
67                                struct ext4_group_desc *bg)
68 {
69         return le32_to_cpu(bg->bg_block_bitmap) |
70                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
71                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
72 }
73
74 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
75                                struct ext4_group_desc *bg)
76 {
77         return le32_to_cpu(bg->bg_inode_bitmap) |
78                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
79                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
80 }
81
82 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
83                               struct ext4_group_desc *bg)
84 {
85         return le32_to_cpu(bg->bg_inode_table) |
86                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
87                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
88 }
89
90 void ext4_block_bitmap_set(struct super_block *sb,
91                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
92 {
93         bg->bg_block_bitmap = cpu_to_le32((u32)blk);
94         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
95                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
96 }
97
98 void ext4_inode_bitmap_set(struct super_block *sb,
99                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
100 {
101         bg->bg_inode_bitmap  = cpu_to_le32((u32)blk);
102         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
103                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
104 }
105
106 void ext4_inode_table_set(struct super_block *sb,
107                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
108 {
109         bg->bg_inode_table = cpu_to_le32((u32)blk);
110         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
111                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
112 }
113
114 /*
115  * Wrappers for jbd2_journal_start/end.
116  *
117  * The only special thing we need to do here is to make sure that all
118  * journal_end calls result in the superblock being marked dirty, so
119  * that sync() will call the filesystem's write_super callback if
120  * appropriate.
121  */
122 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
123 {
124         journal_t *journal;
125
126         if (sb->s_flags & MS_RDONLY)
127                 return ERR_PTR(-EROFS);
128
129         /* Special case here: if the journal has aborted behind our
130          * backs (eg. EIO in the commit thread), then we still need to
131          * take the FS itself readonly cleanly. */
132         journal = EXT4_SB(sb)->s_journal;
133         if (is_journal_aborted(journal)) {
134                 ext4_abort(sb, __FUNCTION__,
135                            "Detected aborted journal");
136                 return ERR_PTR(-EROFS);
137         }
138
139         return jbd2_journal_start(journal, nblocks);
140 }
141
142 /*
143  * The only special thing we need to do here is to make sure that all
144  * jbd2_journal_stop calls result in the superblock being marked dirty, so
145  * that sync() will call the filesystem's write_super callback if
146  * appropriate.
147  */
148 int __ext4_journal_stop(const char *where, handle_t *handle)
149 {
150         struct super_block *sb;
151         int err;
152         int rc;
153
154         sb = handle->h_transaction->t_journal->j_private;
155         err = handle->h_err;
156         rc = jbd2_journal_stop(handle);
157
158         if (!err)
159                 err = rc;
160         if (err)
161                 __ext4_std_error(sb, where, err);
162         return err;
163 }
164
165 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
166                 struct buffer_head *bh, handle_t *handle, int err)
167 {
168         char nbuf[16];
169         const char *errstr = ext4_decode_error(NULL, err, nbuf);
170
171         if (bh)
172                 BUFFER_TRACE(bh, "abort");
173
174         if (!handle->h_err)
175                 handle->h_err = err;
176
177         if (is_handle_aborted(handle))
178                 return;
179
180         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
181                caller, errstr, err_fn);
182
183         jbd2_journal_abort_handle(handle);
184 }
185
186 /* Deal with the reporting of failure conditions on a filesystem such as
187  * inconsistencies detected or read IO failures.
188  *
189  * On ext2, we can store the error state of the filesystem in the
190  * superblock.  That is not possible on ext4, because we may have other
191  * write ordering constraints on the superblock which prevent us from
192  * writing it out straight away; and given that the journal is about to
193  * be aborted, we can't rely on the current, or future, transactions to
194  * write out the superblock safely.
195  *
196  * We'll just use the jbd2_journal_abort() error code to record an error in
197  * the journal instead.  On recovery, the journal will compain about
198  * that error until we've noted it down and cleared it.
199  */
200
201 static void ext4_handle_error(struct super_block *sb)
202 {
203         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
204
205         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
206         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
207
208         if (sb->s_flags & MS_RDONLY)
209                 return;
210
211         if (!test_opt (sb, ERRORS_CONT)) {
212                 journal_t *journal = EXT4_SB(sb)->s_journal;
213
214                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
215                 if (journal)
216                         jbd2_journal_abort(journal, -EIO);
217         }
218         if (test_opt (sb, ERRORS_RO)) {
219                 printk (KERN_CRIT "Remounting filesystem read-only\n");
220                 sb->s_flags |= MS_RDONLY;
221         }
222         ext4_commit_super(sb, es, 1);
223         if (test_opt(sb, ERRORS_PANIC))
224                 panic("EXT4-fs (device %s): panic forced after error\n",
225                         sb->s_id);
226 }
227
228 void ext4_error (struct super_block * sb, const char * function,
229                  const char * fmt, ...)
230 {
231         va_list args;
232
233         va_start(args, fmt);
234         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
235         vprintk(fmt, args);
236         printk("\n");
237         va_end(args);
238
239         ext4_handle_error(sb);
240 }
241
242 static const char *ext4_decode_error(struct super_block * sb, int errno,
243                                      char nbuf[16])
244 {
245         char *errstr = NULL;
246
247         switch (errno) {
248         case -EIO:
249                 errstr = "IO failure";
250                 break;
251         case -ENOMEM:
252                 errstr = "Out of memory";
253                 break;
254         case -EROFS:
255                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
256                         errstr = "Journal has aborted";
257                 else
258                         errstr = "Readonly filesystem";
259                 break;
260         default:
261                 /* If the caller passed in an extra buffer for unknown
262                  * errors, textualise them now.  Else we just return
263                  * NULL. */
264                 if (nbuf) {
265                         /* Check for truncated error codes... */
266                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
267                                 errstr = nbuf;
268                 }
269                 break;
270         }
271
272         return errstr;
273 }
274
275 /* __ext4_std_error decodes expected errors from journaling functions
276  * automatically and invokes the appropriate error response.  */
277
278 void __ext4_std_error (struct super_block * sb, const char * function,
279                        int errno)
280 {
281         char nbuf[16];
282         const char *errstr;
283
284         /* Special case: if the error is EROFS, and we're not already
285          * inside a transaction, then there's really no point in logging
286          * an error. */
287         if (errno == -EROFS && journal_current_handle() == NULL &&
288             (sb->s_flags & MS_RDONLY))
289                 return;
290
291         errstr = ext4_decode_error(sb, errno, nbuf);
292         printk (KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
293                 sb->s_id, function, errstr);
294
295         ext4_handle_error(sb);
296 }
297
298 /*
299  * ext4_abort is a much stronger failure handler than ext4_error.  The
300  * abort function may be used to deal with unrecoverable failures such
301  * as journal IO errors or ENOMEM at a critical moment in log management.
302  *
303  * We unconditionally force the filesystem into an ABORT|READONLY state,
304  * unless the error response on the fs has been set to panic in which
305  * case we take the easy way out and panic immediately.
306  */
307
308 void ext4_abort (struct super_block * sb, const char * function,
309                  const char * fmt, ...)
310 {
311         va_list args;
312
313         printk (KERN_CRIT "ext4_abort called.\n");
314
315         va_start(args, fmt);
316         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ",sb->s_id, function);
317         vprintk(fmt, args);
318         printk("\n");
319         va_end(args);
320
321         if (test_opt(sb, ERRORS_PANIC))
322                 panic("EXT4-fs panic from previous error\n");
323
324         if (sb->s_flags & MS_RDONLY)
325                 return;
326
327         printk(KERN_CRIT "Remounting filesystem read-only\n");
328         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
329         sb->s_flags |= MS_RDONLY;
330         EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
331         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
332 }
333
334 void ext4_warning (struct super_block * sb, const char * function,
335                    const char * fmt, ...)
336 {
337         va_list args;
338
339         va_start(args, fmt);
340         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
341                sb->s_id, function);
342         vprintk(fmt, args);
343         printk("\n");
344         va_end(args);
345 }
346
347 void ext4_update_dynamic_rev(struct super_block *sb)
348 {
349         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
350
351         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
352                 return;
353
354         ext4_warning(sb, __FUNCTION__,
355                      "updating to rev %d because of new feature flag, "
356                      "running e2fsck is recommended",
357                      EXT4_DYNAMIC_REV);
358
359         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
360         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
361         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
362         /* leave es->s_feature_*compat flags alone */
363         /* es->s_uuid will be set by e2fsck if empty */
364
365         /*
366          * The rest of the superblock fields should be zero, and if not it
367          * means they are likely already in use, so leave them alone.  We
368          * can leave it up to e2fsck to clean up any inconsistencies there.
369          */
370 }
371
372 /*
373  * Open the external journal device
374  */
375 static struct block_device *ext4_blkdev_get(dev_t dev)
376 {
377         struct block_device *bdev;
378         char b[BDEVNAME_SIZE];
379
380         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
381         if (IS_ERR(bdev))
382                 goto fail;
383         return bdev;
384
385 fail:
386         printk(KERN_ERR "EXT4: failed to open journal device %s: %ld\n",
387                         __bdevname(dev, b), PTR_ERR(bdev));
388         return NULL;
389 }
390
391 /*
392  * Release the journal device
393  */
394 static int ext4_blkdev_put(struct block_device *bdev)
395 {
396         bd_release(bdev);
397         return blkdev_put(bdev);
398 }
399
400 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
401 {
402         struct block_device *bdev;
403         int ret = -ENODEV;
404
405         bdev = sbi->journal_bdev;
406         if (bdev) {
407                 ret = ext4_blkdev_put(bdev);
408                 sbi->journal_bdev = NULL;
409         }
410         return ret;
411 }
412
413 static inline struct inode *orphan_list_entry(struct list_head *l)
414 {
415         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
416 }
417
418 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
419 {
420         struct list_head *l;
421
422         printk(KERN_ERR "sb orphan head is %d\n",
423                le32_to_cpu(sbi->s_es->s_last_orphan));
424
425         printk(KERN_ERR "sb_info orphan list:\n");
426         list_for_each(l, &sbi->s_orphan) {
427                 struct inode *inode = orphan_list_entry(l);
428                 printk(KERN_ERR "  "
429                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
430                        inode->i_sb->s_id, inode->i_ino, inode,
431                        inode->i_mode, inode->i_nlink,
432                        NEXT_ORPHAN(inode));
433         }
434 }
435
436 static void ext4_put_super (struct super_block * sb)
437 {
438         struct ext4_sb_info *sbi = EXT4_SB(sb);
439         struct ext4_super_block *es = sbi->s_es;
440         int i;
441
442         ext4_ext_release(sb);
443         ext4_xattr_put_super(sb);
444         jbd2_journal_destroy(sbi->s_journal);
445         if (!(sb->s_flags & MS_RDONLY)) {
446                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
447                 es->s_state = cpu_to_le16(sbi->s_mount_state);
448                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
449                 mark_buffer_dirty(sbi->s_sbh);
450                 ext4_commit_super(sb, es, 1);
451         }
452
453         for (i = 0; i < sbi->s_gdb_count; i++)
454                 brelse(sbi->s_group_desc[i]);
455         kfree(sbi->s_group_desc);
456         percpu_counter_destroy(&sbi->s_freeblocks_counter);
457         percpu_counter_destroy(&sbi->s_freeinodes_counter);
458         percpu_counter_destroy(&sbi->s_dirs_counter);
459         brelse(sbi->s_sbh);
460 #ifdef CONFIG_QUOTA
461         for (i = 0; i < MAXQUOTAS; i++)
462                 kfree(sbi->s_qf_names[i]);
463 #endif
464
465         /* Debugging code just in case the in-memory inode orphan list
466          * isn't empty.  The on-disk one can be non-empty if we've
467          * detected an error and taken the fs readonly, but the
468          * in-memory list had better be clean by this point. */
469         if (!list_empty(&sbi->s_orphan))
470                 dump_orphan_list(sb, sbi);
471         J_ASSERT(list_empty(&sbi->s_orphan));
472
473         invalidate_bdev(sb->s_bdev, 0);
474         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
475                 /*
476                  * Invalidate the journal device's buffers.  We don't want them
477                  * floating about in memory - the physical journal device may
478                  * hotswapped, and it breaks the `ro-after' testing code.
479                  */
480                 sync_blockdev(sbi->journal_bdev);
481                 invalidate_bdev(sbi->journal_bdev, 0);
482                 ext4_blkdev_remove(sbi);
483         }
484         sb->s_fs_info = NULL;
485         kfree(sbi);
486         return;
487 }
488
489 static kmem_cache_t *ext4_inode_cachep;
490
491 /*
492  * Called inside transaction, so use GFP_NOFS
493  */
494 static struct inode *ext4_alloc_inode(struct super_block *sb)
495 {
496         struct ext4_inode_info *ei;
497
498         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
499         if (!ei)
500                 return NULL;
501 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
502         ei->i_acl = EXT4_ACL_NOT_CACHED;
503         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
504 #endif
505         ei->i_block_alloc_info = NULL;
506         ei->vfs_inode.i_version = 1;
507         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
508         return &ei->vfs_inode;
509 }
510
511 static void ext4_destroy_inode(struct inode *inode)
512 {
513         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
514 }
515
516 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
517 {
518         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
519
520         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
521             SLAB_CTOR_CONSTRUCTOR) {
522                 INIT_LIST_HEAD(&ei->i_orphan);
523 #ifdef CONFIG_EXT4DEV_FS_XATTR
524                 init_rwsem(&ei->xattr_sem);
525 #endif
526                 mutex_init(&ei->truncate_mutex);
527                 inode_init_once(&ei->vfs_inode);
528         }
529 }
530
531 static int init_inodecache(void)
532 {
533         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
534                                              sizeof(struct ext4_inode_info),
535                                              0, (SLAB_RECLAIM_ACCOUNT|
536                                                 SLAB_MEM_SPREAD),
537                                              init_once, NULL);
538         if (ext4_inode_cachep == NULL)
539                 return -ENOMEM;
540         return 0;
541 }
542
543 static void destroy_inodecache(void)
544 {
545         kmem_cache_destroy(ext4_inode_cachep);
546 }
547
548 static void ext4_clear_inode(struct inode *inode)
549 {
550         struct ext4_block_alloc_info *rsv = EXT4_I(inode)->i_block_alloc_info;
551 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
552         if (EXT4_I(inode)->i_acl &&
553                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
554                 posix_acl_release(EXT4_I(inode)->i_acl);
555                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
556         }
557         if (EXT4_I(inode)->i_default_acl &&
558                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
559                 posix_acl_release(EXT4_I(inode)->i_default_acl);
560                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
561         }
562 #endif
563         ext4_discard_reservation(inode);
564         EXT4_I(inode)->i_block_alloc_info = NULL;
565         if (unlikely(rsv))
566                 kfree(rsv);
567 }
568
569 static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb)
570 {
571 #if defined(CONFIG_QUOTA)
572         struct ext4_sb_info *sbi = EXT4_SB(sb);
573
574         if (sbi->s_jquota_fmt)
575                 seq_printf(seq, ",jqfmt=%s",
576                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold": "vfsv0");
577
578         if (sbi->s_qf_names[USRQUOTA])
579                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
580
581         if (sbi->s_qf_names[GRPQUOTA])
582                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
583
584         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
585                 seq_puts(seq, ",usrquota");
586
587         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
588                 seq_puts(seq, ",grpquota");
589 #endif
590 }
591
592 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
593 {
594         struct super_block *sb = vfs->mnt_sb;
595
596         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
597                 seq_puts(seq, ",data=journal");
598         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
599                 seq_puts(seq, ",data=ordered");
600         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
601                 seq_puts(seq, ",data=writeback");
602
603         ext4_show_quota_options(seq, sb);
604
605         return 0;
606 }
607
608
609 static struct dentry *ext4_get_dentry(struct super_block *sb, void *vobjp)
610 {
611         __u32 *objp = vobjp;
612         unsigned long ino = objp[0];
613         __u32 generation = objp[1];
614         struct inode *inode;
615         struct dentry *result;
616
617         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
618                 return ERR_PTR(-ESTALE);
619         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
620                 return ERR_PTR(-ESTALE);
621
622         /* iget isn't really right if the inode is currently unallocated!!
623          *
624          * ext4_read_inode will return a bad_inode if the inode had been
625          * deleted, so we should be safe.
626          *
627          * Currently we don't know the generation for parent directory, so
628          * a generation of 0 means "accept any"
629          */
630         inode = iget(sb, ino);
631         if (inode == NULL)
632                 return ERR_PTR(-ENOMEM);
633         if (is_bad_inode(inode) ||
634             (generation && inode->i_generation != generation)) {
635                 iput(inode);
636                 return ERR_PTR(-ESTALE);
637         }
638         /* now to find a dentry.
639          * If possible, get a well-connected one
640          */
641         result = d_alloc_anon(inode);
642         if (!result) {
643                 iput(inode);
644                 return ERR_PTR(-ENOMEM);
645         }
646         return result;
647 }
648
649 #ifdef CONFIG_QUOTA
650 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
651 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
652
653 static int ext4_dquot_initialize(struct inode *inode, int type);
654 static int ext4_dquot_drop(struct inode *inode);
655 static int ext4_write_dquot(struct dquot *dquot);
656 static int ext4_acquire_dquot(struct dquot *dquot);
657 static int ext4_release_dquot(struct dquot *dquot);
658 static int ext4_mark_dquot_dirty(struct dquot *dquot);
659 static int ext4_write_info(struct super_block *sb, int type);
660 static int ext4_quota_on(struct super_block *sb, int type, int format_id, char *path);
661 static int ext4_quota_on_mount(struct super_block *sb, int type);
662 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
663                                size_t len, loff_t off);
664 static ssize_t ext4_quota_write(struct super_block *sb, int type,
665                                 const char *data, size_t len, loff_t off);
666
667 static struct dquot_operations ext4_quota_operations = {
668         .initialize     = ext4_dquot_initialize,
669         .drop           = ext4_dquot_drop,
670         .alloc_space    = dquot_alloc_space,
671         .alloc_inode    = dquot_alloc_inode,
672         .free_space     = dquot_free_space,
673         .free_inode     = dquot_free_inode,
674         .transfer       = dquot_transfer,
675         .write_dquot    = ext4_write_dquot,
676         .acquire_dquot  = ext4_acquire_dquot,
677         .release_dquot  = ext4_release_dquot,
678         .mark_dirty     = ext4_mark_dquot_dirty,
679         .write_info     = ext4_write_info
680 };
681
682 static struct quotactl_ops ext4_qctl_operations = {
683         .quota_on       = ext4_quota_on,
684         .quota_off      = vfs_quota_off,
685         .quota_sync     = vfs_quota_sync,
686         .get_info       = vfs_get_dqinfo,
687         .set_info       = vfs_set_dqinfo,
688         .get_dqblk      = vfs_get_dqblk,
689         .set_dqblk      = vfs_set_dqblk
690 };
691 #endif
692
693 static struct super_operations ext4_sops = {
694         .alloc_inode    = ext4_alloc_inode,
695         .destroy_inode  = ext4_destroy_inode,
696         .read_inode     = ext4_read_inode,
697         .write_inode    = ext4_write_inode,
698         .dirty_inode    = ext4_dirty_inode,
699         .delete_inode   = ext4_delete_inode,
700         .put_super      = ext4_put_super,
701         .write_super    = ext4_write_super,
702         .sync_fs        = ext4_sync_fs,
703         .write_super_lockfs = ext4_write_super_lockfs,
704         .unlockfs       = ext4_unlockfs,
705         .statfs         = ext4_statfs,
706         .remount_fs     = ext4_remount,
707         .clear_inode    = ext4_clear_inode,
708         .show_options   = ext4_show_options,
709 #ifdef CONFIG_QUOTA
710         .quota_read     = ext4_quota_read,
711         .quota_write    = ext4_quota_write,
712 #endif
713 };
714
715 static struct export_operations ext4_export_ops = {
716         .get_parent = ext4_get_parent,
717         .get_dentry = ext4_get_dentry,
718 };
719
720 enum {
721         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
722         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
723         Opt_nouid32, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
724         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
725         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
726         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
727         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
728         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
729         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
730         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
731         Opt_grpquota, Opt_extents,
732 };
733
734 static match_table_t tokens = {
735         {Opt_bsd_df, "bsddf"},
736         {Opt_minix_df, "minixdf"},
737         {Opt_grpid, "grpid"},
738         {Opt_grpid, "bsdgroups"},
739         {Opt_nogrpid, "nogrpid"},
740         {Opt_nogrpid, "sysvgroups"},
741         {Opt_resgid, "resgid=%u"},
742         {Opt_resuid, "resuid=%u"},
743         {Opt_sb, "sb=%u"},
744         {Opt_err_cont, "errors=continue"},
745         {Opt_err_panic, "errors=panic"},
746         {Opt_err_ro, "errors=remount-ro"},
747         {Opt_nouid32, "nouid32"},
748         {Opt_nocheck, "nocheck"},
749         {Opt_nocheck, "check=none"},
750         {Opt_debug, "debug"},
751         {Opt_oldalloc, "oldalloc"},
752         {Opt_orlov, "orlov"},
753         {Opt_user_xattr, "user_xattr"},
754         {Opt_nouser_xattr, "nouser_xattr"},
755         {Opt_acl, "acl"},
756         {Opt_noacl, "noacl"},
757         {Opt_reservation, "reservation"},
758         {Opt_noreservation, "noreservation"},
759         {Opt_noload, "noload"},
760         {Opt_nobh, "nobh"},
761         {Opt_bh, "bh"},
762         {Opt_commit, "commit=%u"},
763         {Opt_journal_update, "journal=update"},
764         {Opt_journal_inum, "journal=%u"},
765         {Opt_journal_dev, "journal_dev=%u"},
766         {Opt_abort, "abort"},
767         {Opt_data_journal, "data=journal"},
768         {Opt_data_ordered, "data=ordered"},
769         {Opt_data_writeback, "data=writeback"},
770         {Opt_offusrjquota, "usrjquota="},
771         {Opt_usrjquota, "usrjquota=%s"},
772         {Opt_offgrpjquota, "grpjquota="},
773         {Opt_grpjquota, "grpjquota=%s"},
774         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
775         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
776         {Opt_grpquota, "grpquota"},
777         {Opt_noquota, "noquota"},
778         {Opt_quota, "quota"},
779         {Opt_usrquota, "usrquota"},
780         {Opt_barrier, "barrier=%u"},
781         {Opt_extents, "extents"},
782         {Opt_err, NULL},
783         {Opt_resize, "resize"},
784 };
785
786 static ext4_fsblk_t get_sb_block(void **data)
787 {
788         ext4_fsblk_t    sb_block;
789         char            *options = (char *) *data;
790
791         if (!options || strncmp(options, "sb=", 3) != 0)
792                 return 1;       /* Default location */
793         options += 3;
794         /*todo: use simple_strtoll with >32bit ext4 */
795         sb_block = simple_strtoul(options, &options, 0);
796         if (*options && *options != ',') {
797                 printk("EXT4-fs: Invalid sb specification: %s\n",
798                        (char *) *data);
799                 return 1;
800         }
801         if (*options == ',')
802                 options++;
803         *data = (void *) options;
804         return sb_block;
805 }
806
807 static int parse_options (char *options, struct super_block *sb,
808                           unsigned int *inum, unsigned long *journal_devnum,
809                           ext4_fsblk_t *n_blocks_count, int is_remount)
810 {
811         struct ext4_sb_info *sbi = EXT4_SB(sb);
812         char * p;
813         substring_t args[MAX_OPT_ARGS];
814         int data_opt = 0;
815         int option;
816 #ifdef CONFIG_QUOTA
817         int qtype;
818         char *qname;
819 #endif
820
821         if (!options)
822                 return 1;
823
824         while ((p = strsep (&options, ",")) != NULL) {
825                 int token;
826                 if (!*p)
827                         continue;
828
829                 token = match_token(p, tokens, args);
830                 switch (token) {
831                 case Opt_bsd_df:
832                         clear_opt (sbi->s_mount_opt, MINIX_DF);
833                         break;
834                 case Opt_minix_df:
835                         set_opt (sbi->s_mount_opt, MINIX_DF);
836                         break;
837                 case Opt_grpid:
838                         set_opt (sbi->s_mount_opt, GRPID);
839                         break;
840                 case Opt_nogrpid:
841                         clear_opt (sbi->s_mount_opt, GRPID);
842                         break;
843                 case Opt_resuid:
844                         if (match_int(&args[0], &option))
845                                 return 0;
846                         sbi->s_resuid = option;
847                         break;
848                 case Opt_resgid:
849                         if (match_int(&args[0], &option))
850                                 return 0;
851                         sbi->s_resgid = option;
852                         break;
853                 case Opt_sb:
854                         /* handled by get_sb_block() instead of here */
855                         /* *sb_block = match_int(&args[0]); */
856                         break;
857                 case Opt_err_panic:
858                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
859                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
860                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
861                         break;
862                 case Opt_err_ro:
863                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
864                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
865                         set_opt (sbi->s_mount_opt, ERRORS_RO);
866                         break;
867                 case Opt_err_cont:
868                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
869                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
870                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
871                         break;
872                 case Opt_nouid32:
873                         set_opt (sbi->s_mount_opt, NO_UID32);
874                         break;
875                 case Opt_nocheck:
876                         clear_opt (sbi->s_mount_opt, CHECK);
877                         break;
878                 case Opt_debug:
879                         set_opt (sbi->s_mount_opt, DEBUG);
880                         break;
881                 case Opt_oldalloc:
882                         set_opt (sbi->s_mount_opt, OLDALLOC);
883                         break;
884                 case Opt_orlov:
885                         clear_opt (sbi->s_mount_opt, OLDALLOC);
886                         break;
887 #ifdef CONFIG_EXT4DEV_FS_XATTR
888                 case Opt_user_xattr:
889                         set_opt (sbi->s_mount_opt, XATTR_USER);
890                         break;
891                 case Opt_nouser_xattr:
892                         clear_opt (sbi->s_mount_opt, XATTR_USER);
893                         break;
894 #else
895                 case Opt_user_xattr:
896                 case Opt_nouser_xattr:
897                         printk("EXT4 (no)user_xattr options not supported\n");
898                         break;
899 #endif
900 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
901                 case Opt_acl:
902                         set_opt(sbi->s_mount_opt, POSIX_ACL);
903                         break;
904                 case Opt_noacl:
905                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
906                         break;
907 #else
908                 case Opt_acl:
909                 case Opt_noacl:
910                         printk("EXT4 (no)acl options not supported\n");
911                         break;
912 #endif
913                 case Opt_reservation:
914                         set_opt(sbi->s_mount_opt, RESERVATION);
915                         break;
916                 case Opt_noreservation:
917                         clear_opt(sbi->s_mount_opt, RESERVATION);
918                         break;
919                 case Opt_journal_update:
920                         /* @@@ FIXME */
921                         /* Eventually we will want to be able to create
922                            a journal file here.  For now, only allow the
923                            user to specify an existing inode to be the
924                            journal file. */
925                         if (is_remount) {
926                                 printk(KERN_ERR "EXT4-fs: cannot specify "
927                                        "journal on remount\n");
928                                 return 0;
929                         }
930                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
931                         break;
932                 case Opt_journal_inum:
933                         if (is_remount) {
934                                 printk(KERN_ERR "EXT4-fs: cannot specify "
935                                        "journal on remount\n");
936                                 return 0;
937                         }
938                         if (match_int(&args[0], &option))
939                                 return 0;
940                         *inum = option;
941                         break;
942                 case Opt_journal_dev:
943                         if (is_remount) {
944                                 printk(KERN_ERR "EXT4-fs: cannot specify "
945                                        "journal on remount\n");
946                                 return 0;
947                         }
948                         if (match_int(&args[0], &option))
949                                 return 0;
950                         *journal_devnum = option;
951                         break;
952                 case Opt_noload:
953                         set_opt (sbi->s_mount_opt, NOLOAD);
954                         break;
955                 case Opt_commit:
956                         if (match_int(&args[0], &option))
957                                 return 0;
958                         if (option < 0)
959                                 return 0;
960                         if (option == 0)
961                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
962                         sbi->s_commit_interval = HZ * option;
963                         break;
964                 case Opt_data_journal:
965                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
966                         goto datacheck;
967                 case Opt_data_ordered:
968                         data_opt = EXT4_MOUNT_ORDERED_DATA;
969                         goto datacheck;
970                 case Opt_data_writeback:
971                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
972                 datacheck:
973                         if (is_remount) {
974                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
975                                                 != data_opt) {
976                                         printk(KERN_ERR
977                                                 "EXT4-fs: cannot change data "
978                                                 "mode on remount\n");
979                                         return 0;
980                                 }
981                         } else {
982                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
983                                 sbi->s_mount_opt |= data_opt;
984                         }
985                         break;
986 #ifdef CONFIG_QUOTA
987                 case Opt_usrjquota:
988                         qtype = USRQUOTA;
989                         goto set_qf_name;
990                 case Opt_grpjquota:
991                         qtype = GRPQUOTA;
992 set_qf_name:
993                         if (sb_any_quota_enabled(sb)) {
994                                 printk(KERN_ERR
995                                         "EXT4-fs: Cannot change journalled "
996                                         "quota options when quota turned on.\n");
997                                 return 0;
998                         }
999                         qname = match_strdup(&args[0]);
1000                         if (!qname) {
1001                                 printk(KERN_ERR
1002                                         "EXT4-fs: not enough memory for "
1003                                         "storing quotafile name.\n");
1004                                 return 0;
1005                         }
1006                         if (sbi->s_qf_names[qtype] &&
1007                             strcmp(sbi->s_qf_names[qtype], qname)) {
1008                                 printk(KERN_ERR
1009                                         "EXT4-fs: %s quota file already "
1010                                         "specified.\n", QTYPE2NAME(qtype));
1011                                 kfree(qname);
1012                                 return 0;
1013                         }
1014                         sbi->s_qf_names[qtype] = qname;
1015                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1016                                 printk(KERN_ERR
1017                                         "EXT4-fs: quotafile must be on "
1018                                         "filesystem root.\n");
1019                                 kfree(sbi->s_qf_names[qtype]);
1020                                 sbi->s_qf_names[qtype] = NULL;
1021                                 return 0;
1022                         }
1023                         set_opt(sbi->s_mount_opt, QUOTA);
1024                         break;
1025                 case Opt_offusrjquota:
1026                         qtype = USRQUOTA;
1027                         goto clear_qf_name;
1028                 case Opt_offgrpjquota:
1029                         qtype = GRPQUOTA;
1030 clear_qf_name:
1031                         if (sb_any_quota_enabled(sb)) {
1032                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1033                                         "journalled quota options when "
1034                                         "quota turned on.\n");
1035                                 return 0;
1036                         }
1037                         /*
1038                          * The space will be released later when all options
1039                          * are confirmed to be correct
1040                          */
1041                         sbi->s_qf_names[qtype] = NULL;
1042                         break;
1043                 case Opt_jqfmt_vfsold:
1044                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
1045                         break;
1046                 case Opt_jqfmt_vfsv0:
1047                         sbi->s_jquota_fmt = QFMT_VFS_V0;
1048                         break;
1049                 case Opt_quota:
1050                 case Opt_usrquota:
1051                         set_opt(sbi->s_mount_opt, QUOTA);
1052                         set_opt(sbi->s_mount_opt, USRQUOTA);
1053                         break;
1054                 case Opt_grpquota:
1055                         set_opt(sbi->s_mount_opt, QUOTA);
1056                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1057                         break;
1058                 case Opt_noquota:
1059                         if (sb_any_quota_enabled(sb)) {
1060                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1061                                         "options when quota turned on.\n");
1062                                 return 0;
1063                         }
1064                         clear_opt(sbi->s_mount_opt, QUOTA);
1065                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1066                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1067                         break;
1068 #else
1069                 case Opt_quota:
1070                 case Opt_usrquota:
1071                 case Opt_grpquota:
1072                 case Opt_usrjquota:
1073                 case Opt_grpjquota:
1074                 case Opt_offusrjquota:
1075                 case Opt_offgrpjquota:
1076                 case Opt_jqfmt_vfsold:
1077                 case Opt_jqfmt_vfsv0:
1078                         printk(KERN_ERR
1079                                 "EXT4-fs: journalled quota options not "
1080                                 "supported.\n");
1081                         break;
1082                 case Opt_noquota:
1083                         break;
1084 #endif
1085                 case Opt_abort:
1086                         set_opt(sbi->s_mount_opt, ABORT);
1087                         break;
1088                 case Opt_barrier:
1089                         if (match_int(&args[0], &option))
1090                                 return 0;
1091                         if (option)
1092                                 set_opt(sbi->s_mount_opt, BARRIER);
1093                         else
1094                                 clear_opt(sbi->s_mount_opt, BARRIER);
1095                         break;
1096                 case Opt_ignore:
1097                         break;
1098                 case Opt_resize:
1099                         if (!is_remount) {
1100                                 printk("EXT4-fs: resize option only available "
1101                                         "for remount\n");
1102                                 return 0;
1103                         }
1104                         if (match_int(&args[0], &option) != 0)
1105                                 return 0;
1106                         *n_blocks_count = option;
1107                         break;
1108                 case Opt_nobh:
1109                         set_opt(sbi->s_mount_opt, NOBH);
1110                         break;
1111                 case Opt_bh:
1112                         clear_opt(sbi->s_mount_opt, NOBH);
1113                         break;
1114                 case Opt_extents:
1115                         set_opt (sbi->s_mount_opt, EXTENTS);
1116                         break;
1117                 default:
1118                         printk (KERN_ERR
1119                                 "EXT4-fs: Unrecognized mount option \"%s\" "
1120                                 "or missing value\n", p);
1121                         return 0;
1122                 }
1123         }
1124 #ifdef CONFIG_QUOTA
1125         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1126                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1127                      sbi->s_qf_names[USRQUOTA])
1128                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1129
1130                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1131                      sbi->s_qf_names[GRPQUOTA])
1132                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1133
1134                 if ((sbi->s_qf_names[USRQUOTA] &&
1135                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1136                     (sbi->s_qf_names[GRPQUOTA] &&
1137                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1138                         printk(KERN_ERR "EXT4-fs: old and new quota "
1139                                         "format mixing.\n");
1140                         return 0;
1141                 }
1142
1143                 if (!sbi->s_jquota_fmt) {
1144                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1145                                         "not specified.\n");
1146                         return 0;
1147                 }
1148         } else {
1149                 if (sbi->s_jquota_fmt) {
1150                         printk(KERN_ERR "EXT4-fs: journalled quota format "
1151                                         "specified with no journalling "
1152                                         "enabled.\n");
1153                         return 0;
1154                 }
1155         }
1156 #endif
1157         return 1;
1158 }
1159
1160 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1161                             int read_only)
1162 {
1163         struct ext4_sb_info *sbi = EXT4_SB(sb);
1164         int res = 0;
1165
1166         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1167                 printk (KERN_ERR "EXT4-fs warning: revision level too high, "
1168                         "forcing read-only mode\n");
1169                 res = MS_RDONLY;
1170         }
1171         if (read_only)
1172                 return res;
1173         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1174                 printk (KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1175                         "running e2fsck is recommended\n");
1176         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1177                 printk (KERN_WARNING
1178                         "EXT4-fs warning: mounting fs with errors, "
1179                         "running e2fsck is recommended\n");
1180         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1181                  le16_to_cpu(es->s_mnt_count) >=
1182                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1183                 printk (KERN_WARNING
1184                         "EXT4-fs warning: maximal mount count reached, "
1185                         "running e2fsck is recommended\n");
1186         else if (le32_to_cpu(es->s_checkinterval) &&
1187                 (le32_to_cpu(es->s_lastcheck) +
1188                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1189                 printk (KERN_WARNING
1190                         "EXT4-fs warning: checktime reached, "
1191                         "running e2fsck is recommended\n");
1192 #if 0
1193                 /* @@@ We _will_ want to clear the valid bit if we find
1194                  * inconsistencies, to force a fsck at reboot.  But for
1195                  * a plain journaled filesystem we can keep it set as
1196                  * valid forever! :)
1197                  */
1198         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT4_VALID_FS);
1199 #endif
1200         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1201                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1202         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
1203         es->s_mtime = cpu_to_le32(get_seconds());
1204         ext4_update_dynamic_rev(sb);
1205         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1206
1207         ext4_commit_super(sb, es, 1);
1208         if (test_opt(sb, DEBUG))
1209                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1210                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1211                         sb->s_blocksize,
1212                         sbi->s_groups_count,
1213                         EXT4_BLOCKS_PER_GROUP(sb),
1214                         EXT4_INODES_PER_GROUP(sb),
1215                         sbi->s_mount_opt);
1216
1217         printk(KERN_INFO "EXT4 FS on %s, ", sb->s_id);
1218         if (EXT4_SB(sb)->s_journal->j_inode == NULL) {
1219                 char b[BDEVNAME_SIZE];
1220
1221                 printk("external journal on %s\n",
1222                         bdevname(EXT4_SB(sb)->s_journal->j_dev, b));
1223         } else {
1224                 printk("internal journal\n");
1225         }
1226         return res;
1227 }
1228
1229 /* Called at mount-time, super-block is locked */
1230 static int ext4_check_descriptors (struct super_block * sb)
1231 {
1232         struct ext4_sb_info *sbi = EXT4_SB(sb);
1233         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1234         ext4_fsblk_t last_block;
1235         ext4_fsblk_t block_bitmap;
1236         ext4_fsblk_t inode_bitmap;
1237         ext4_fsblk_t inode_table;
1238         struct ext4_group_desc * gdp = NULL;
1239         int desc_block = 0;
1240         int i;
1241
1242         ext4_debug ("Checking group descriptors");
1243
1244         for (i = 0; i < sbi->s_groups_count; i++)
1245         {
1246                 if (i == sbi->s_groups_count - 1)
1247                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1248                 else
1249                         last_block = first_block +
1250                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1251
1252                 if ((i % EXT4_DESC_PER_BLOCK(sb)) == 0)
1253                         gdp = (struct ext4_group_desc *)
1254                                         sbi->s_group_desc[desc_block++]->b_data;
1255                 block_bitmap = ext4_block_bitmap(sb, gdp);
1256                 if (block_bitmap < first_block || block_bitmap > last_block)
1257                 {
1258                         ext4_error (sb, "ext4_check_descriptors",
1259                                     "Block bitmap for group %d"
1260                                     " not in group (block %llu)!",
1261                                     i, block_bitmap);
1262                         return 0;
1263                 }
1264                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1265                 if (inode_bitmap < first_block || inode_bitmap > last_block)
1266                 {
1267                         ext4_error (sb, "ext4_check_descriptors",
1268                                     "Inode bitmap for group %d"
1269                                     " not in group (block %llu)!",
1270                                     i, inode_bitmap);
1271                         return 0;
1272                 }
1273                 inode_table = ext4_inode_table(sb, gdp);
1274                 if (inode_table < first_block ||
1275                     inode_table + sbi->s_itb_per_group > last_block)
1276                 {
1277                         ext4_error (sb, "ext4_check_descriptors",
1278                                     "Inode table for group %d"
1279                                     " not in group (block %llu)!",
1280                                     i, inode_table);
1281                         return 0;
1282                 }
1283                 first_block += EXT4_BLOCKS_PER_GROUP(sb);
1284                 gdp = (struct ext4_group_desc *)
1285                         ((__u8 *)gdp + EXT4_DESC_SIZE(sb));
1286         }
1287
1288         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1289         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext4_count_free_inodes(sb));
1290         return 1;
1291 }
1292
1293
1294 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1295  * the superblock) which were deleted from all directories, but held open by
1296  * a process at the time of a crash.  We walk the list and try to delete these
1297  * inodes at recovery time (only with a read-write filesystem).
1298  *
1299  * In order to keep the orphan inode chain consistent during traversal (in
1300  * case of crash during recovery), we link each inode into the superblock
1301  * orphan list_head and handle it the same way as an inode deletion during
1302  * normal operation (which journals the operations for us).
1303  *
1304  * We only do an iget() and an iput() on each inode, which is very safe if we
1305  * accidentally point at an in-use or already deleted inode.  The worst that
1306  * can happen in this case is that we get a "bit already cleared" message from
1307  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1308  * e2fsck was run on this filesystem, and it must have already done the orphan
1309  * inode cleanup for us, so we can safely abort without any further action.
1310  */
1311 static void ext4_orphan_cleanup (struct super_block * sb,
1312                                  struct ext4_super_block * es)
1313 {
1314         unsigned int s_flags = sb->s_flags;
1315         int nr_orphans = 0, nr_truncates = 0;
1316 #ifdef CONFIG_QUOTA
1317         int i;
1318 #endif
1319         if (!es->s_last_orphan) {
1320                 jbd_debug(4, "no orphan inodes to clean up\n");
1321                 return;
1322         }
1323
1324         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1325                 if (es->s_last_orphan)
1326                         jbd_debug(1, "Errors on filesystem, "
1327                                   "clearing orphan list.\n");
1328                 es->s_last_orphan = 0;
1329                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1330                 return;
1331         }
1332
1333         if (s_flags & MS_RDONLY) {
1334                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1335                        sb->s_id);
1336                 sb->s_flags &= ~MS_RDONLY;
1337         }
1338 #ifdef CONFIG_QUOTA
1339         /* Needed for iput() to work correctly and not trash data */
1340         sb->s_flags |= MS_ACTIVE;
1341         /* Turn on quotas so that they are updated correctly */
1342         for (i = 0; i < MAXQUOTAS; i++) {
1343                 if (EXT4_SB(sb)->s_qf_names[i]) {
1344                         int ret = ext4_quota_on_mount(sb, i);
1345                         if (ret < 0)
1346                                 printk(KERN_ERR
1347                                         "EXT4-fs: Cannot turn on journalled "
1348                                         "quota: error %d\n", ret);
1349                 }
1350         }
1351 #endif
1352
1353         while (es->s_last_orphan) {
1354                 struct inode *inode;
1355
1356                 if (!(inode =
1357                       ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1358                         es->s_last_orphan = 0;
1359                         break;
1360                 }
1361
1362                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1363                 DQUOT_INIT(inode);
1364                 if (inode->i_nlink) {
1365                         printk(KERN_DEBUG
1366                                 "%s: truncating inode %lu to %Ld bytes\n",
1367                                 __FUNCTION__, inode->i_ino, inode->i_size);
1368                         jbd_debug(2, "truncating inode %lu to %Ld bytes\n",
1369                                   inode->i_ino, inode->i_size);
1370                         ext4_truncate(inode);
1371                         nr_truncates++;
1372                 } else {
1373                         printk(KERN_DEBUG
1374                                 "%s: deleting unreferenced inode %lu\n",
1375                                 __FUNCTION__, inode->i_ino);
1376                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1377                                   inode->i_ino);
1378                         nr_orphans++;
1379                 }
1380                 iput(inode);  /* The delete magic happens here! */
1381         }
1382
1383 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1384
1385         if (nr_orphans)
1386                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1387                        sb->s_id, PLURAL(nr_orphans));
1388         if (nr_truncates)
1389                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1390                        sb->s_id, PLURAL(nr_truncates));
1391 #ifdef CONFIG_QUOTA
1392         /* Turn quotas off */
1393         for (i = 0; i < MAXQUOTAS; i++) {
1394                 if (sb_dqopt(sb)->files[i])
1395                         vfs_quota_off(sb, i);
1396         }
1397 #endif
1398         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1399 }
1400
1401 #define log2(n) ffz(~(n))
1402
1403 /*
1404  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1405  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1406  * We need to be 1 filesystem block less than the 2^32 sector limit.
1407  */
1408 static loff_t ext4_max_size(int bits)
1409 {
1410         loff_t res = EXT4_NDIR_BLOCKS;
1411         /* This constant is calculated to be the largest file size for a
1412          * dense, 4k-blocksize file such that the total number of
1413          * sectors in the file, including data and all indirect blocks,
1414          * does not exceed 2^32. */
1415         const loff_t upper_limit = 0x1ff7fffd000LL;
1416
1417         res += 1LL << (bits-2);
1418         res += 1LL << (2*(bits-2));
1419         res += 1LL << (3*(bits-2));
1420         res <<= bits;
1421         if (res > upper_limit)
1422                 res = upper_limit;
1423         return res;
1424 }
1425
1426 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1427                                 ext4_fsblk_t logical_sb_block, int nr)
1428 {
1429         struct ext4_sb_info *sbi = EXT4_SB(sb);
1430         unsigned long bg, first_meta_bg;
1431         int has_super = 0;
1432
1433         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1434
1435         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1436             nr < first_meta_bg)
1437                 return logical_sb_block + nr + 1;
1438         bg = sbi->s_desc_per_block * nr;
1439         if (ext4_bg_has_super(sb, bg))
1440                 has_super = 1;
1441         return (has_super + ext4_group_first_block_no(sb, bg));
1442 }
1443
1444
1445 static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1446 {
1447         struct buffer_head * bh;
1448         struct ext4_super_block *es = NULL;
1449         struct ext4_sb_info *sbi;
1450         ext4_fsblk_t block;
1451         ext4_fsblk_t sb_block = get_sb_block(&data);
1452         ext4_fsblk_t logical_sb_block;
1453         unsigned long offset = 0;
1454         unsigned int journal_inum = 0;
1455         unsigned long journal_devnum = 0;
1456         unsigned long def_mount_opts;
1457         struct inode *root;
1458         int blocksize;
1459         int hblock;
1460         int db_count;
1461         int i;
1462         int needs_recovery;
1463         __le32 features;
1464         __u64 blocks_count;
1465
1466         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1467         if (!sbi)
1468                 return -ENOMEM;
1469         sb->s_fs_info = sbi;
1470         sbi->s_mount_opt = 0;
1471         sbi->s_resuid = EXT4_DEF_RESUID;
1472         sbi->s_resgid = EXT4_DEF_RESGID;
1473
1474         unlock_kernel();
1475
1476         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1477         if (!blocksize) {
1478                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1479                 goto out_fail;
1480         }
1481
1482         /*
1483          * The ext4 superblock will not be buffer aligned for other than 1kB
1484          * block sizes.  We need to calculate the offset from buffer start.
1485          */
1486         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1487                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1488                 offset = do_div(logical_sb_block, blocksize);
1489         } else {
1490                 logical_sb_block = sb_block;
1491         }
1492
1493         if (!(bh = sb_bread(sb, logical_sb_block))) {
1494                 printk (KERN_ERR "EXT4-fs: unable to read superblock\n");
1495                 goto out_fail;
1496         }
1497         /*
1498          * Note: s_es must be initialized as soon as possible because
1499          *       some ext4 macro-instructions depend on its value
1500          */
1501         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1502         sbi->s_es = es;
1503         sb->s_magic = le16_to_cpu(es->s_magic);
1504         if (sb->s_magic != EXT4_SUPER_MAGIC)
1505                 goto cantfind_ext4;
1506
1507         /* Set defaults before we parse the mount options */
1508         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1509         if (def_mount_opts & EXT4_DEFM_DEBUG)
1510                 set_opt(sbi->s_mount_opt, DEBUG);
1511         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1512                 set_opt(sbi->s_mount_opt, GRPID);
1513         if (def_mount_opts & EXT4_DEFM_UID16)
1514                 set_opt(sbi->s_mount_opt, NO_UID32);
1515         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1516                 set_opt(sbi->s_mount_opt, XATTR_USER);
1517         if (def_mount_opts & EXT4_DEFM_ACL)
1518                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1519         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1520                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1521         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1522                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1523         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1524                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1525
1526         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1527                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1528         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_RO)
1529                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1530         else
1531                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1532
1533         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1534         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1535
1536         set_opt(sbi->s_mount_opt, RESERVATION);
1537
1538         if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1539                             NULL, 0))
1540                 goto failed_mount;
1541
1542         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1543                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1544
1545         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
1546             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
1547              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1548              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1549                 printk(KERN_WARNING
1550                        "EXT4-fs warning: feature flags set on rev 0 fs, "
1551                        "running e2fsck is recommended\n");
1552         /*
1553          * Check feature flags regardless of the revision level, since we
1554          * previously didn't change the revision level when setting the flags,
1555          * so there is a chance incompat flags are set on a rev 0 filesystem.
1556          */
1557         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
1558         if (features) {
1559                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
1560                        "unsupported optional features (%x).\n",
1561                        sb->s_id, le32_to_cpu(features));
1562                 goto failed_mount;
1563         }
1564         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
1565         if (!(sb->s_flags & MS_RDONLY) && features) {
1566                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
1567                        "unsupported optional features (%x).\n",
1568                        sb->s_id, le32_to_cpu(features));
1569                 goto failed_mount;
1570         }
1571         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1572
1573         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
1574             blocksize > EXT4_MAX_BLOCK_SIZE) {
1575                 printk(KERN_ERR
1576                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
1577                        blocksize, sb->s_id);
1578                 goto failed_mount;
1579         }
1580
1581         hblock = bdev_hardsect_size(sb->s_bdev);
1582         if (sb->s_blocksize != blocksize) {
1583                 /*
1584                  * Make sure the blocksize for the filesystem is larger
1585                  * than the hardware sectorsize for the machine.
1586                  */
1587                 if (blocksize < hblock) {
1588                         printk(KERN_ERR "EXT4-fs: blocksize %d too small for "
1589                                "device blocksize %d.\n", blocksize, hblock);
1590                         goto failed_mount;
1591                 }
1592
1593                 brelse (bh);
1594                 sb_set_blocksize(sb, blocksize);
1595                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1596                 offset = do_div(logical_sb_block, blocksize);
1597                 bh = sb_bread(sb, logical_sb_block);
1598                 if (!bh) {
1599                         printk(KERN_ERR
1600                                "EXT4-fs: Can't read superblock on 2nd try.\n");
1601                         goto failed_mount;
1602                 }
1603                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
1604                 sbi->s_es = es;
1605                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
1606                         printk (KERN_ERR
1607                                 "EXT4-fs: Magic mismatch, very weird !\n");
1608                         goto failed_mount;
1609                 }
1610         }
1611
1612         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits);
1613
1614         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
1615                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
1616                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
1617         } else {
1618                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1619                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1620                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
1621                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1622                     (sbi->s_inode_size > blocksize)) {
1623                         printk (KERN_ERR
1624                                 "EXT4-fs: unsupported inode size: %d\n",
1625                                 sbi->s_inode_size);
1626                         goto failed_mount;
1627                 }
1628         }
1629         sbi->s_frag_size = EXT4_MIN_FRAG_SIZE <<
1630                                    le32_to_cpu(es->s_log_frag_size);
1631         if (blocksize != sbi->s_frag_size) {
1632                 printk(KERN_ERR
1633                        "EXT4-fs: fragsize %lu != blocksize %u (unsupported)\n",
1634                        sbi->s_frag_size, blocksize);
1635                 goto failed_mount;
1636         }
1637         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
1638         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
1639                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
1640                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
1641                     sbi->s_desc_size & (sbi->s_desc_size - 1)) {
1642                         printk(KERN_ERR
1643                                "EXT4-fs: unsupported descriptor size %lu\n",
1644                                sbi->s_desc_size);
1645                         goto failed_mount;
1646                 }
1647         } else
1648                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
1649         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1650         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1651         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1652         if (EXT4_INODE_SIZE(sb) == 0)
1653                 goto cantfind_ext4;
1654         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
1655         if (sbi->s_inodes_per_block == 0)
1656                 goto cantfind_ext4;
1657         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1658                                         sbi->s_inodes_per_block;
1659         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
1660         sbi->s_sbh = bh;
1661         sbi->s_mount_state = le16_to_cpu(es->s_state);
1662         sbi->s_addr_per_block_bits = log2(EXT4_ADDR_PER_BLOCK(sb));
1663         sbi->s_desc_per_block_bits = log2(EXT4_DESC_PER_BLOCK(sb));
1664         for (i=0; i < 4; i++)
1665                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1666         sbi->s_def_hash_version = es->s_def_hash_version;
1667
1668         if (sbi->s_blocks_per_group > blocksize * 8) {
1669                 printk (KERN_ERR
1670                         "EXT4-fs: #blocks per group too big: %lu\n",
1671                         sbi->s_blocks_per_group);
1672                 goto failed_mount;
1673         }
1674         if (sbi->s_frags_per_group > blocksize * 8) {
1675                 printk (KERN_ERR
1676                         "EXT4-fs: #fragments per group too big: %lu\n",
1677                         sbi->s_frags_per_group);
1678                 goto failed_mount;
1679         }
1680         if (sbi->s_inodes_per_group > blocksize * 8) {
1681                 printk (KERN_ERR
1682                         "EXT4-fs: #inodes per group too big: %lu\n",
1683                         sbi->s_inodes_per_group);
1684                 goto failed_mount;
1685         }
1686
1687         if (ext4_blocks_count(es) >
1688                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1689                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
1690                         " too large to mount safely\n", sb->s_id);
1691                 if (sizeof(sector_t) < 8)
1692                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
1693                                         "enabled\n");
1694                 goto failed_mount;
1695         }
1696
1697         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
1698                 goto cantfind_ext4;
1699         blocks_count = (ext4_blocks_count(es) -
1700                         le32_to_cpu(es->s_first_data_block) +
1701                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
1702         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
1703         sbi->s_groups_count = blocks_count;
1704         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
1705                    EXT4_DESC_PER_BLOCK(sb);
1706         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1707                                     GFP_KERNEL);
1708         if (sbi->s_group_desc == NULL) {
1709                 printk (KERN_ERR "EXT4-fs: not enough memory\n");
1710                 goto failed_mount;
1711         }
1712
1713         bgl_lock_init(&sbi->s_blockgroup_lock);
1714
1715         for (i = 0; i < db_count; i++) {
1716                 block = descriptor_loc(sb, logical_sb_block, i);
1717                 sbi->s_group_desc[i] = sb_bread(sb, block);
1718                 if (!sbi->s_group_desc[i]) {
1719                         printk (KERN_ERR "EXT4-fs: "
1720                                 "can't read group descriptor %d\n", i);
1721                         db_count = i;
1722                         goto failed_mount2;
1723                 }
1724         }
1725         if (!ext4_check_descriptors (sb)) {
1726                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
1727                 goto failed_mount2;
1728         }
1729         sbi->s_gdb_count = db_count;
1730         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1731         spin_lock_init(&sbi->s_next_gen_lock);
1732
1733         percpu_counter_init(&sbi->s_freeblocks_counter,
1734                 ext4_count_free_blocks(sb));
1735         percpu_counter_init(&sbi->s_freeinodes_counter,
1736                 ext4_count_free_inodes(sb));
1737         percpu_counter_init(&sbi->s_dirs_counter,
1738                 ext4_count_dirs(sb));
1739
1740         /* per fileystem reservation list head & lock */
1741         spin_lock_init(&sbi->s_rsv_window_lock);
1742         sbi->s_rsv_window_root = RB_ROOT;
1743         /* Add a single, static dummy reservation to the start of the
1744          * reservation window list --- it gives us a placeholder for
1745          * append-at-start-of-list which makes the allocation logic
1746          * _much_ simpler. */
1747         sbi->s_rsv_window_head.rsv_start = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
1748         sbi->s_rsv_window_head.rsv_end = EXT4_RESERVE_WINDOW_NOT_ALLOCATED;
1749         sbi->s_rsv_window_head.rsv_alloc_hit = 0;
1750         sbi->s_rsv_window_head.rsv_goal_size = 0;
1751         ext4_rsv_window_add(sb, &sbi->s_rsv_window_head);
1752
1753         /*
1754          * set up enough so that it can read an inode
1755          */
1756         sb->s_op = &ext4_sops;
1757         sb->s_export_op = &ext4_export_ops;
1758         sb->s_xattr = ext4_xattr_handlers;
1759 #ifdef CONFIG_QUOTA
1760         sb->s_qcop = &ext4_qctl_operations;
1761         sb->dq_op = &ext4_quota_operations;
1762 #endif
1763         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1764
1765         sb->s_root = NULL;
1766
1767         needs_recovery = (es->s_last_orphan != 0 ||
1768                           EXT4_HAS_INCOMPAT_FEATURE(sb,
1769                                     EXT4_FEATURE_INCOMPAT_RECOVER));
1770
1771         /*
1772          * The first inode we look at is the journal inode.  Don't try
1773          * root first: it may be modified in the journal!
1774          */
1775         if (!test_opt(sb, NOLOAD) &&
1776             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
1777                 if (ext4_load_journal(sb, es, journal_devnum))
1778                         goto failed_mount3;
1779         } else if (journal_inum) {
1780                 if (ext4_create_journal(sb, es, journal_inum))
1781                         goto failed_mount3;
1782         } else {
1783                 if (!silent)
1784                         printk (KERN_ERR
1785                                 "ext4: No journal on filesystem on %s\n",
1786                                 sb->s_id);
1787                 goto failed_mount3;
1788         }
1789
1790         /* We have now updated the journal if required, so we can
1791          * validate the data journaling mode. */
1792         switch (test_opt(sb, DATA_FLAGS)) {
1793         case 0:
1794                 /* No mode set, assume a default based on the journal
1795                  * capabilities: ORDERED_DATA if the journal can
1796                  * cope, else JOURNAL_DATA
1797                  */
1798                 if (jbd2_journal_check_available_features
1799                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
1800                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1801                 else
1802                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1803                 break;
1804
1805         case EXT4_MOUNT_ORDERED_DATA:
1806         case EXT4_MOUNT_WRITEBACK_DATA:
1807                 if (!jbd2_journal_check_available_features
1808                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
1809                         printk(KERN_ERR "EXT4-fs: Journal does not support "
1810                                "requested data journaling mode\n");
1811                         goto failed_mount4;
1812                 }
1813         default:
1814                 break;
1815         }
1816
1817         if (test_opt(sb, NOBH)) {
1818                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
1819                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
1820                                 "its supported only with writeback mode\n");
1821                         clear_opt(sbi->s_mount_opt, NOBH);
1822                 }
1823         }
1824         /*
1825          * The jbd2_journal_load will have done any necessary log recovery,
1826          * so we can safely mount the rest of the filesystem now.
1827          */
1828
1829         root = iget(sb, EXT4_ROOT_INO);
1830         sb->s_root = d_alloc_root(root);
1831         if (!sb->s_root) {
1832                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
1833                 iput(root);
1834                 goto failed_mount4;
1835         }
1836         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1837                 dput(sb->s_root);
1838                 sb->s_root = NULL;
1839                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
1840                 goto failed_mount4;
1841         }
1842
1843         ext4_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1844         /*
1845          * akpm: core read_super() calls in here with the superblock locked.
1846          * That deadlocks, because orphan cleanup needs to lock the superblock
1847          * in numerous places.  Here we just pop the lock - it's relatively
1848          * harmless, because we are now ready to accept write_super() requests,
1849          * and aviro says that's the only reason for hanging onto the
1850          * superblock lock.
1851          */
1852         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
1853         ext4_orphan_cleanup(sb, es);
1854         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
1855         if (needs_recovery)
1856                 printk (KERN_INFO "EXT4-fs: recovery complete.\n");
1857         ext4_mark_recovery_complete(sb, es);
1858         printk (KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
1859                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
1860                 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
1861                 "writeback");
1862
1863         ext4_ext_init(sb);
1864
1865         lock_kernel();
1866         return 0;
1867
1868 cantfind_ext4:
1869         if (!silent)
1870                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
1871                        sb->s_id);
1872         goto failed_mount;
1873
1874 failed_mount4:
1875         jbd2_journal_destroy(sbi->s_journal);
1876 failed_mount3:
1877         percpu_counter_destroy(&sbi->s_freeblocks_counter);
1878         percpu_counter_destroy(&sbi->s_freeinodes_counter);
1879         percpu_counter_destroy(&sbi->s_dirs_counter);
1880 failed_mount2:
1881         for (i = 0; i < db_count; i++)
1882                 brelse(sbi->s_group_desc[i]);
1883         kfree(sbi->s_group_desc);
1884 failed_mount:
1885 #ifdef CONFIG_QUOTA
1886         for (i = 0; i < MAXQUOTAS; i++)
1887                 kfree(sbi->s_qf_names[i]);
1888 #endif
1889         ext4_blkdev_remove(sbi);
1890         brelse(bh);
1891 out_fail:
1892         sb->s_fs_info = NULL;
1893         kfree(sbi);
1894         lock_kernel();
1895         return -EINVAL;
1896 }
1897
1898 /*
1899  * Setup any per-fs journal parameters now.  We'll do this both on
1900  * initial mount, once the journal has been initialised but before we've
1901  * done any recovery; and again on any subsequent remount.
1902  */
1903 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
1904 {
1905         struct ext4_sb_info *sbi = EXT4_SB(sb);
1906
1907         if (sbi->s_commit_interval)
1908                 journal->j_commit_interval = sbi->s_commit_interval;
1909         /* We could also set up an ext4-specific default for the commit
1910          * interval here, but for now we'll just fall back to the jbd
1911          * default. */
1912
1913         spin_lock(&journal->j_state_lock);
1914         if (test_opt(sb, BARRIER))
1915                 journal->j_flags |= JBD2_BARRIER;
1916         else
1917                 journal->j_flags &= ~JBD2_BARRIER;
1918         spin_unlock(&journal->j_state_lock);
1919 }
1920
1921 static journal_t *ext4_get_journal(struct super_block *sb,
1922                                    unsigned int journal_inum)
1923 {
1924         struct inode *journal_inode;
1925         journal_t *journal;
1926
1927         /* First, test for the existence of a valid inode on disk.  Bad
1928          * things happen if we iget() an unused inode, as the subsequent
1929          * iput() will try to delete it. */
1930
1931         journal_inode = iget(sb, journal_inum);
1932         if (!journal_inode) {
1933                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
1934                 return NULL;
1935         }
1936         if (!journal_inode->i_nlink) {
1937                 make_bad_inode(journal_inode);
1938                 iput(journal_inode);
1939                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
1940                 return NULL;
1941         }
1942
1943         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1944                   journal_inode, journal_inode->i_size);
1945         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1946                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
1947                 iput(journal_inode);
1948                 return NULL;
1949         }
1950
1951         journal = jbd2_journal_init_inode(journal_inode);
1952         if (!journal) {
1953                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
1954                 iput(journal_inode);
1955                 return NULL;
1956         }
1957         journal->j_private = sb;
1958         ext4_init_journal_params(sb, journal);
1959         return journal;
1960 }
1961
1962 static journal_t *ext4_get_dev_journal(struct super_block *sb,
1963                                        dev_t j_dev)
1964 {
1965         struct buffer_head * bh;
1966         journal_t *journal;
1967         ext4_fsblk_t start;
1968         ext4_fsblk_t len;
1969         int hblock, blocksize;
1970         ext4_fsblk_t sb_block;
1971         unsigned long offset;
1972         struct ext4_super_block * es;
1973         struct block_device *bdev;
1974
1975         bdev = ext4_blkdev_get(j_dev);
1976         if (bdev == NULL)
1977                 return NULL;
1978
1979         if (bd_claim(bdev, sb)) {
1980                 printk(KERN_ERR
1981                         "EXT4: failed to claim external journal device.\n");
1982                 blkdev_put(bdev);
1983                 return NULL;
1984         }
1985
1986         blocksize = sb->s_blocksize;
1987         hblock = bdev_hardsect_size(bdev);
1988         if (blocksize < hblock) {
1989                 printk(KERN_ERR
1990                         "EXT4-fs: blocksize too small for journal device.\n");
1991                 goto out_bdev;
1992         }
1993
1994         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
1995         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
1996         set_blocksize(bdev, blocksize);
1997         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1998                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
1999                        "external journal\n");
2000                 goto out_bdev;
2001         }
2002
2003         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2004         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2005             !(le32_to_cpu(es->s_feature_incompat) &
2006               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2007                 printk(KERN_ERR "EXT4-fs: external journal has "
2008                                         "bad superblock\n");
2009                 brelse(bh);
2010                 goto out_bdev;
2011         }
2012
2013         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2014                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2015                 brelse(bh);
2016                 goto out_bdev;
2017         }
2018
2019         len = ext4_blocks_count(es);
2020         start = sb_block + 1;
2021         brelse(bh);     /* we're done with the superblock */
2022
2023         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2024                                         start, len, blocksize);
2025         if (!journal) {
2026                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2027                 goto out_bdev;
2028         }
2029         journal->j_private = sb;
2030         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2031         wait_on_buffer(journal->j_sb_buffer);
2032         if (!buffer_uptodate(journal->j_sb_buffer)) {
2033                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2034                 goto out_journal;
2035         }
2036         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2037                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2038                                         "user (unsupported) - %d\n",
2039                         be32_to_cpu(journal->j_superblock->s_nr_users));
2040                 goto out_journal;
2041         }
2042         EXT4_SB(sb)->journal_bdev = bdev;
2043         ext4_init_journal_params(sb, journal);
2044         return journal;
2045 out_journal:
2046         jbd2_journal_destroy(journal);
2047 out_bdev:
2048         ext4_blkdev_put(bdev);
2049         return NULL;
2050 }
2051
2052 static int ext4_load_journal(struct super_block *sb,
2053                              struct ext4_super_block *es,
2054                              unsigned long journal_devnum)
2055 {
2056         journal_t *journal;
2057         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2058         dev_t journal_dev;
2059         int err = 0;
2060         int really_read_only;
2061
2062         if (journal_devnum &&
2063             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2064                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2065                         "numbers have changed\n");
2066                 journal_dev = new_decode_dev(journal_devnum);
2067         } else
2068                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2069
2070         really_read_only = bdev_read_only(sb->s_bdev);
2071
2072         /*
2073          * Are we loading a blank journal or performing recovery after a
2074          * crash?  For recovery, we need to check in advance whether we
2075          * can get read-write access to the device.
2076          */
2077
2078         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2079                 if (sb->s_flags & MS_RDONLY) {
2080                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2081                                         "required on readonly filesystem.\n");
2082                         if (really_read_only) {
2083                                 printk(KERN_ERR "EXT4-fs: write access "
2084                                         "unavailable, cannot proceed.\n");
2085                                 return -EROFS;
2086                         }
2087                         printk (KERN_INFO "EXT4-fs: write access will "
2088                                         "be enabled during recovery.\n");
2089                 }
2090         }
2091
2092         if (journal_inum && journal_dev) {
2093                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2094                        "and inode journals!\n");
2095                 return -EINVAL;
2096         }
2097
2098         if (journal_inum) {
2099                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2100                         return -EINVAL;
2101         } else {
2102                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2103                         return -EINVAL;
2104         }
2105
2106         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2107                 err = jbd2_journal_update_format(journal);
2108                 if (err)  {
2109                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2110                         jbd2_journal_destroy(journal);
2111                         return err;
2112                 }
2113         }
2114
2115         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2116                 err = jbd2_journal_wipe(journal, !really_read_only);
2117         if (!err)
2118                 err = jbd2_journal_load(journal);
2119
2120         if (err) {
2121                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2122                 jbd2_journal_destroy(journal);
2123                 return err;
2124         }
2125
2126         EXT4_SB(sb)->s_journal = journal;
2127         ext4_clear_journal_err(sb, es);
2128
2129         if (journal_devnum &&
2130             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2131                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2132                 sb->s_dirt = 1;
2133
2134                 /* Make sure we flush the recovery flag to disk. */
2135                 ext4_commit_super(sb, es, 1);
2136         }
2137
2138         return 0;
2139 }
2140
2141 static int ext4_create_journal(struct super_block * sb,
2142                                struct ext4_super_block * es,
2143                                unsigned int journal_inum)
2144 {
2145         journal_t *journal;
2146
2147         if (sb->s_flags & MS_RDONLY) {
2148                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2149                                 "create journal.\n");
2150                 return -EROFS;
2151         }
2152
2153         if (!(journal = ext4_get_journal(sb, journal_inum)))
2154                 return -EINVAL;
2155
2156         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2157                journal_inum);
2158
2159         if (jbd2_journal_create(journal)) {
2160                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2161                 jbd2_journal_destroy(journal);
2162                 return -EIO;
2163         }
2164
2165         EXT4_SB(sb)->s_journal = journal;
2166
2167         ext4_update_dynamic_rev(sb);
2168         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2169         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2170
2171         es->s_journal_inum = cpu_to_le32(journal_inum);
2172         sb->s_dirt = 1;
2173
2174         /* Make sure we flush the recovery flag to disk. */
2175         ext4_commit_super(sb, es, 1);
2176
2177         return 0;
2178 }
2179
2180 static void ext4_commit_super (struct super_block * sb,
2181                                struct ext4_super_block * es,
2182                                int sync)
2183 {
2184         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2185
2186         if (!sbh)
2187                 return;
2188         es->s_wtime = cpu_to_le32(get_seconds());
2189         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2190         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2191         BUFFER_TRACE(sbh, "marking dirty");
2192         mark_buffer_dirty(sbh);
2193         if (sync)
2194                 sync_dirty_buffer(sbh);
2195 }
2196
2197
2198 /*
2199  * Have we just finished recovery?  If so, and if we are mounting (or
2200  * remounting) the filesystem readonly, then we will end up with a
2201  * consistent fs on disk.  Record that fact.
2202  */
2203 static void ext4_mark_recovery_complete(struct super_block * sb,
2204                                         struct ext4_super_block * es)
2205 {
2206         journal_t *journal = EXT4_SB(sb)->s_journal;
2207
2208         jbd2_journal_lock_updates(journal);
2209         jbd2_journal_flush(journal);
2210         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2211             sb->s_flags & MS_RDONLY) {
2212                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2213                 sb->s_dirt = 0;
2214                 ext4_commit_super(sb, es, 1);
2215         }
2216         jbd2_journal_unlock_updates(journal);
2217 }
2218
2219 /*
2220  * If we are mounting (or read-write remounting) a filesystem whose journal
2221  * has recorded an error from a previous lifetime, move that error to the
2222  * main filesystem now.
2223  */
2224 static void ext4_clear_journal_err(struct super_block * sb,
2225                                    struct ext4_super_block * es)
2226 {
2227         journal_t *journal;
2228         int j_errno;
2229         const char *errstr;
2230
2231         journal = EXT4_SB(sb)->s_journal;
2232
2233         /*
2234          * Now check for any error status which may have been recorded in the
2235          * journal by a prior ext4_error() or ext4_abort()
2236          */
2237
2238         j_errno = jbd2_journal_errno(journal);
2239         if (j_errno) {
2240                 char nbuf[16];
2241
2242                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2243                 ext4_warning(sb, __FUNCTION__, "Filesystem error recorded "
2244                              "from previous mount: %s", errstr);
2245                 ext4_warning(sb, __FUNCTION__, "Marking fs in need of "
2246                              "filesystem check.");
2247
2248                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2249                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2250                 ext4_commit_super (sb, es, 1);
2251
2252                 jbd2_journal_clear_err(journal);
2253         }
2254 }
2255
2256 /*
2257  * Force the running and committing transactions to commit,
2258  * and wait on the commit.
2259  */
2260 int ext4_force_commit(struct super_block *sb)
2261 {
2262         journal_t *journal;
2263         int ret;
2264
2265         if (sb->s_flags & MS_RDONLY)
2266                 return 0;
2267
2268         journal = EXT4_SB(sb)->s_journal;
2269         sb->s_dirt = 0;
2270         ret = ext4_journal_force_commit(journal);
2271         return ret;
2272 }
2273
2274 /*
2275  * Ext4 always journals updates to the superblock itself, so we don't
2276  * have to propagate any other updates to the superblock on disk at this
2277  * point.  Just start an async writeback to get the buffers on their way
2278  * to the disk.
2279  *
2280  * This implicitly triggers the writebehind on sync().
2281  */
2282
2283 static void ext4_write_super (struct super_block * sb)
2284 {
2285         if (mutex_trylock(&sb->s_lock) != 0)
2286                 BUG();
2287         sb->s_dirt = 0;
2288 }
2289
2290 static int ext4_sync_fs(struct super_block *sb, int wait)
2291 {
2292         tid_t target;
2293
2294         sb->s_dirt = 0;
2295         if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
2296                 if (wait)
2297                         jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
2298         }
2299         return 0;
2300 }
2301
2302 /*
2303  * LVM calls this function before a (read-only) snapshot is created.  This
2304  * gives us a chance to flush the journal completely and mark the fs clean.
2305  */
2306 static void ext4_write_super_lockfs(struct super_block *sb)
2307 {
2308         sb->s_dirt = 0;
2309
2310         if (!(sb->s_flags & MS_RDONLY)) {
2311                 journal_t *journal = EXT4_SB(sb)->s_journal;
2312
2313                 /* Now we set up the journal barrier. */
2314                 jbd2_journal_lock_updates(journal);
2315                 jbd2_journal_flush(journal);
2316
2317                 /* Journal blocked and flushed, clear needs_recovery flag. */
2318                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2319                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2320         }
2321 }
2322
2323 /*
2324  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2325  * flag here, even though the filesystem is not technically dirty yet.
2326  */
2327 static void ext4_unlockfs(struct super_block *sb)
2328 {
2329         if (!(sb->s_flags & MS_RDONLY)) {
2330                 lock_super(sb);
2331                 /* Reser the needs_recovery flag before the fs is unlocked. */
2332                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2333                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2334                 unlock_super(sb);
2335                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2336         }
2337 }
2338
2339 static int ext4_remount (struct super_block * sb, int * flags, char * data)
2340 {
2341         struct ext4_super_block * es;
2342         struct ext4_sb_info *sbi = EXT4_SB(sb);
2343         ext4_fsblk_t n_blocks_count = 0;
2344         unsigned long old_sb_flags;
2345         struct ext4_mount_options old_opts;
2346         int err;
2347 #ifdef CONFIG_QUOTA
2348         int i;
2349 #endif
2350
2351         /* Store the original options */
2352         old_sb_flags = sb->s_flags;
2353         old_opts.s_mount_opt = sbi->s_mount_opt;
2354         old_opts.s_resuid = sbi->s_resuid;
2355         old_opts.s_resgid = sbi->s_resgid;
2356         old_opts.s_commit_interval = sbi->s_commit_interval;
2357 #ifdef CONFIG_QUOTA
2358         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2359         for (i = 0; i < MAXQUOTAS; i++)
2360                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2361 #endif
2362
2363         /*
2364          * Allow the "check" option to be passed as a remount option.
2365          */
2366         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2367                 err = -EINVAL;
2368                 goto restore_opts;
2369         }
2370
2371         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2372                 ext4_abort(sb, __FUNCTION__, "Abort forced by user");
2373
2374         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2375                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2376
2377         es = sbi->s_es;
2378
2379         ext4_init_journal_params(sb, sbi->s_journal);
2380
2381         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2382                 n_blocks_count > ext4_blocks_count(es)) {
2383                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
2384                         err = -EROFS;
2385                         goto restore_opts;
2386                 }
2387
2388                 if (*flags & MS_RDONLY) {
2389                         /*
2390                          * First of all, the unconditional stuff we have to do
2391                          * to disable replay of the journal when we next remount
2392                          */
2393                         sb->s_flags |= MS_RDONLY;
2394
2395                         /*
2396                          * OK, test if we are remounting a valid rw partition
2397                          * readonly, and if so set the rdonly flag and then
2398                          * mark the partition as valid again.
2399                          */
2400                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
2401                             (sbi->s_mount_state & EXT4_VALID_FS))
2402                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2403
2404                         ext4_mark_recovery_complete(sb, es);
2405                 } else {
2406                         __le32 ret;
2407                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2408                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
2409                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
2410                                        "remount RDWR because of unsupported "
2411                                        "optional features (%x).\n",
2412                                        sb->s_id, le32_to_cpu(ret));
2413                                 err = -EROFS;
2414                                 goto restore_opts;
2415                         }
2416                         /*
2417                          * Mounting a RDONLY partition read-write, so reread
2418                          * and store the current valid flag.  (It may have
2419                          * been changed by e2fsck since we originally mounted
2420                          * the partition.)
2421                          */
2422                         ext4_clear_journal_err(sb, es);
2423                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2424                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
2425                                 goto restore_opts;
2426                         if (!ext4_setup_super (sb, es, 0))
2427                                 sb->s_flags &= ~MS_RDONLY;
2428                 }
2429         }
2430 #ifdef CONFIG_QUOTA
2431         /* Release old quota file names */
2432         for (i = 0; i < MAXQUOTAS; i++)
2433                 if (old_opts.s_qf_names[i] &&
2434                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2435                         kfree(old_opts.s_qf_names[i]);
2436 #endif
2437         return 0;
2438 restore_opts:
2439         sb->s_flags = old_sb_flags;
2440         sbi->s_mount_opt = old_opts.s_mount_opt;
2441         sbi->s_resuid = old_opts.s_resuid;
2442         sbi->s_resgid = old_opts.s_resgid;
2443         sbi->s_commit_interval = old_opts.s_commit_interval;
2444 #ifdef CONFIG_QUOTA
2445         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
2446         for (i = 0; i < MAXQUOTAS; i++) {
2447                 if (sbi->s_qf_names[i] &&
2448                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
2449                         kfree(sbi->s_qf_names[i]);
2450                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
2451         }
2452 #endif
2453         return err;
2454 }
2455
2456 static int ext4_statfs (struct dentry * dentry, struct kstatfs * buf)
2457 {
2458         struct super_block *sb = dentry->d_sb;
2459         struct ext4_sb_info *sbi = EXT4_SB(sb);
2460         struct ext4_super_block *es = sbi->s_es;
2461         ext4_fsblk_t overhead;
2462         int i;
2463
2464         if (test_opt (sb, MINIX_DF))
2465                 overhead = 0;
2466         else {
2467                 unsigned long ngroups;
2468                 ngroups = EXT4_SB(sb)->s_groups_count;
2469                 smp_rmb();
2470
2471                 /*
2472                  * Compute the overhead (FS structures)
2473                  */
2474
2475                 /*
2476                  * All of the blocks before first_data_block are
2477                  * overhead
2478                  */
2479                 overhead = le32_to_cpu(es->s_first_data_block);
2480
2481                 /*
2482                  * Add the overhead attributed to the superblock and
2483                  * block group descriptors.  If the sparse superblocks
2484                  * feature is turned on, then not all groups have this.
2485                  */
2486                 for (i = 0; i < ngroups; i++) {
2487                         overhead += ext4_bg_has_super(sb, i) +
2488                                 ext4_bg_num_gdb(sb, i);
2489                         cond_resched();
2490                 }
2491
2492                 /*
2493                  * Every block group has an inode bitmap, a block
2494                  * bitmap, and an inode table.
2495                  */
2496                 overhead += (ngroups * (2 + EXT4_SB(sb)->s_itb_per_group));
2497         }
2498
2499         buf->f_type = EXT4_SUPER_MAGIC;
2500         buf->f_bsize = sb->s_blocksize;
2501         buf->f_blocks = ext4_blocks_count(es) - overhead;
2502         buf->f_bfree = percpu_counter_sum(&sbi->s_freeblocks_counter);
2503         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
2504         if (buf->f_bfree < ext4_r_blocks_count(es))
2505                 buf->f_bavail = 0;
2506         buf->f_files = le32_to_cpu(es->s_inodes_count);
2507         buf->f_ffree = percpu_counter_sum(&sbi->s_freeinodes_counter);
2508         buf->f_namelen = EXT4_NAME_LEN;
2509         return 0;
2510 }
2511
2512 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2513  * is locked for write. Otherwise the are possible deadlocks:
2514  * Process 1                         Process 2
2515  * ext4_create()                     quota_sync()
2516  *   jbd2_journal_start()                   write_dquot()
2517  *   DQUOT_INIT()                        down(dqio_mutex)
2518  *     down(dqio_mutex)                    jbd2_journal_start()
2519  *
2520  */
2521
2522 #ifdef CONFIG_QUOTA
2523
2524 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2525 {
2526         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2527 }
2528
2529 static int ext4_dquot_initialize(struct inode *inode, int type)
2530 {
2531         handle_t *handle;
2532         int ret, err;
2533
2534         /* We may create quota structure so we need to reserve enough blocks */
2535         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
2536         if (IS_ERR(handle))
2537                 return PTR_ERR(handle);
2538         ret = dquot_initialize(inode, type);
2539         err = ext4_journal_stop(handle);
2540         if (!ret)
2541                 ret = err;
2542         return ret;
2543 }
2544
2545 static int ext4_dquot_drop(struct inode *inode)
2546 {
2547         handle_t *handle;
2548         int ret, err;
2549
2550         /* We may delete quota structure so we need to reserve enough blocks */
2551         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
2552         if (IS_ERR(handle))
2553                 return PTR_ERR(handle);
2554         ret = dquot_drop(inode);
2555         err = ext4_journal_stop(handle);
2556         if (!ret)
2557                 ret = err;
2558         return ret;
2559 }
2560
2561 static int ext4_write_dquot(struct dquot *dquot)
2562 {
2563         int ret, err;
2564         handle_t *handle;
2565         struct inode *inode;
2566
2567         inode = dquot_to_inode(dquot);
2568         handle = ext4_journal_start(inode,
2569                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
2570         if (IS_ERR(handle))
2571                 return PTR_ERR(handle);
2572         ret = dquot_commit(dquot);
2573         err = ext4_journal_stop(handle);
2574         if (!ret)
2575                 ret = err;
2576         return ret;
2577 }
2578
2579 static int ext4_acquire_dquot(struct dquot *dquot)
2580 {
2581         int ret, err;
2582         handle_t *handle;
2583
2584         handle = ext4_journal_start(dquot_to_inode(dquot),
2585                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
2586         if (IS_ERR(handle))
2587                 return PTR_ERR(handle);
2588         ret = dquot_acquire(dquot);
2589         err = ext4_journal_stop(handle);
2590         if (!ret)
2591                 ret = err;
2592         return ret;
2593 }
2594
2595 static int ext4_release_dquot(struct dquot *dquot)
2596 {
2597         int ret, err;
2598         handle_t *handle;
2599
2600         handle = ext4_journal_start(dquot_to_inode(dquot),
2601                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
2602         if (IS_ERR(handle))
2603                 return PTR_ERR(handle);
2604         ret = dquot_release(dquot);
2605         err = ext4_journal_stop(handle);
2606         if (!ret)
2607                 ret = err;
2608         return ret;
2609 }
2610
2611 static int ext4_mark_dquot_dirty(struct dquot *dquot)
2612 {
2613         /* Are we journalling quotas? */
2614         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2615             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2616                 dquot_mark_dquot_dirty(dquot);
2617                 return ext4_write_dquot(dquot);
2618         } else {
2619                 return dquot_mark_dquot_dirty(dquot);
2620         }
2621 }
2622
2623 static int ext4_write_info(struct super_block *sb, int type)
2624 {
2625         int ret, err;
2626         handle_t *handle;
2627
2628         /* Data block + inode block */
2629         handle = ext4_journal_start(sb->s_root->d_inode, 2);
2630         if (IS_ERR(handle))
2631                 return PTR_ERR(handle);
2632         ret = dquot_commit_info(sb, type);
2633         err = ext4_journal_stop(handle);
2634         if (!ret)
2635                 ret = err;
2636         return ret;
2637 }
2638
2639 /*
2640  * Turn on quotas during mount time - we need to find
2641  * the quota file and such...
2642  */
2643 static int ext4_quota_on_mount(struct super_block *sb, int type)
2644 {
2645         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
2646                         EXT4_SB(sb)->s_jquota_fmt, type);
2647 }
2648
2649 /*
2650  * Standard function to be called on quota_on
2651  */
2652 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
2653                          char *path)
2654 {
2655         int err;
2656         struct nameidata nd;
2657
2658         if (!test_opt(sb, QUOTA))
2659                 return -EINVAL;
2660         /* Not journalling quota? */
2661         if (!EXT4_SB(sb)->s_qf_names[USRQUOTA] &&
2662             !EXT4_SB(sb)->s_qf_names[GRPQUOTA])
2663                 return vfs_quota_on(sb, type, format_id, path);
2664         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2665         if (err)
2666                 return err;
2667         /* Quotafile not on the same filesystem? */
2668         if (nd.mnt->mnt_sb != sb) {
2669                 path_release(&nd);
2670                 return -EXDEV;
2671         }
2672         /* Quotafile not of fs root? */
2673         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2674                 printk(KERN_WARNING
2675                         "EXT4-fs: Quota file not on filesystem root. "
2676                         "Journalled quota will not work.\n");
2677         path_release(&nd);
2678         return vfs_quota_on(sb, type, format_id, path);
2679 }
2680
2681 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2682  * acquiring the locks... As quota files are never truncated and quota code
2683  * itself serializes the operations (and noone else should touch the files)
2684  * we don't have to be afraid of races */
2685 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
2686                                size_t len, loff_t off)
2687 {
2688         struct inode *inode = sb_dqopt(sb)->files[type];
2689         sector_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
2690         int err = 0;
2691         int offset = off & (sb->s_blocksize - 1);
2692         int tocopy;
2693         size_t toread;
2694         struct buffer_head *bh;
2695         loff_t i_size = i_size_read(inode);
2696
2697         if (off > i_size)
2698                 return 0;
2699         if (off+len > i_size)
2700                 len = i_size-off;
2701         toread = len;
2702         while (toread > 0) {
2703                 tocopy = sb->s_blocksize - offset < toread ?
2704                                 sb->s_blocksize - offset : toread;
2705                 bh = ext4_bread(NULL, inode, blk, 0, &err);
2706                 if (err)
2707                         return err;
2708                 if (!bh)        /* A hole? */
2709                         memset(data, 0, tocopy);
2710                 else
2711                         memcpy(data, bh->b_data+offset, tocopy);
2712                 brelse(bh);
2713                 offset = 0;
2714                 toread -= tocopy;
2715                 data += tocopy;
2716                 blk++;
2717         }
2718         return len;
2719 }
2720
2721 /* Write to quotafile (we know the transaction is already started and has
2722  * enough credits) */
2723 static ssize_t ext4_quota_write(struct super_block *sb, int type,
2724                                 const char *data, size_t len, loff_t off)
2725 {
2726         struct inode *inode = sb_dqopt(sb)->files[type];
2727         sector_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
2728         int err = 0;
2729         int offset = off & (sb->s_blocksize - 1);
2730         int tocopy;
2731         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
2732         size_t towrite = len;
2733         struct buffer_head *bh;
2734         handle_t *handle = journal_current_handle();
2735
2736         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
2737         while (towrite > 0) {
2738                 tocopy = sb->s_blocksize - offset < towrite ?
2739                                 sb->s_blocksize - offset : towrite;
2740                 bh = ext4_bread(handle, inode, blk, 1, &err);
2741                 if (!bh)
2742                         goto out;
2743                 if (journal_quota) {
2744                         err = ext4_journal_get_write_access(handle, bh);
2745                         if (err) {
2746                                 brelse(bh);
2747                                 goto out;
2748                         }
2749                 }
2750                 lock_buffer(bh);
2751                 memcpy(bh->b_data+offset, data, tocopy);
2752                 flush_dcache_page(bh->b_page);
2753                 unlock_buffer(bh);
2754                 if (journal_quota)
2755                         err = ext4_journal_dirty_metadata(handle, bh);
2756                 else {
2757                         /* Always do at least ordered writes for quotas */
2758                         err = ext4_journal_dirty_data(handle, bh);
2759                         mark_buffer_dirty(bh);
2760                 }
2761                 brelse(bh);
2762                 if (err)
2763                         goto out;
2764                 offset = 0;
2765                 towrite -= tocopy;
2766                 data += tocopy;
2767                 blk++;
2768         }
2769 out:
2770         if (len == towrite)
2771                 return err;
2772         if (inode->i_size < off+len-towrite) {
2773                 i_size_write(inode, off+len-towrite);
2774                 EXT4_I(inode)->i_disksize = inode->i_size;
2775         }
2776         inode->i_version++;
2777         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2778         ext4_mark_inode_dirty(handle, inode);
2779         mutex_unlock(&inode->i_mutex);
2780         return len - towrite;
2781 }
2782
2783 #endif
2784
2785 static int ext4_get_sb(struct file_system_type *fs_type,
2786         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2787 {
2788         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
2789 }
2790
2791 static struct file_system_type ext4dev_fs_type = {
2792         .owner          = THIS_MODULE,
2793         .name           = "ext4dev",
2794         .get_sb         = ext4_get_sb,
2795         .kill_sb        = kill_block_super,
2796         .fs_flags       = FS_REQUIRES_DEV,
2797 };
2798
2799 static int __init init_ext4_fs(void)
2800 {
2801         int err = init_ext4_xattr();
2802         if (err)
2803                 return err;
2804         err = init_inodecache();
2805         if (err)
2806                 goto out1;
2807         err = register_filesystem(&ext4dev_fs_type);
2808         if (err)
2809                 goto out;
2810         return 0;
2811 out:
2812         destroy_inodecache();
2813 out1:
2814         exit_ext4_xattr();
2815         return err;
2816 }
2817
2818 static void __exit exit_ext4_fs(void)
2819 {
2820         unregister_filesystem(&ext4dev_fs_type);
2821         destroy_inodecache();
2822         exit_ext4_xattr();
2823 }
2824
2825 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2826 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
2827 MODULE_LICENSE("GPL");
2828 module_init(init_ext4_fs)
2829 module_exit(exit_ext4_fs)