Merge branches 'x86/cleanups', 'x86/mpparse', 'x86/numa' and 'x86/uv' into x86/urgent
[linux-2.6] / 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/slab.h>
25 #include <linux/init.h>
26 #include <linux/blkdev.h>
27 #include <linux/parser.h>
28 #include <linux/smp_lock.h>
29 #include <linux/buffer_head.h>
30 #include <linux/exportfs.h>
31 #include <linux/vfs.h>
32 #include <linux/random.h>
33 #include <linux/mount.h>
34 #include <linux/namei.h>
35 #include <linux/quotaops.h>
36 #include <linux/seq_file.h>
37 #include <linux/proc_fs.h>
38 #include <linux/marker.h>
39 #include <linux/log2.h>
40 #include <linux/crc16.h>
41 #include <asm/uaccess.h>
42
43 #include "ext4.h"
44 #include "ext4_jbd2.h"
45 #include "xattr.h"
46 #include "acl.h"
47 #include "namei.h"
48 #include "group.h"
49
50 struct proc_dir_entry *ext4_proc_root;
51
52 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
53                              unsigned long journal_devnum);
54 static int ext4_create_journal(struct super_block *, struct ext4_super_block *,
55                                unsigned int);
56 static void ext4_commit_super(struct super_block *sb,
57                               struct ext4_super_block *es, int sync);
58 static void ext4_mark_recovery_complete(struct super_block *sb,
59                                         struct ext4_super_block *es);
60 static void ext4_clear_journal_err(struct super_block *sb,
61                                    struct ext4_super_block *es);
62 static int ext4_sync_fs(struct super_block *sb, int wait);
63 static const char *ext4_decode_error(struct super_block *sb, int errno,
64                                      char nbuf[16]);
65 static int ext4_remount(struct super_block *sb, int *flags, char *data);
66 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
67 static void ext4_unlockfs(struct super_block *sb);
68 static void ext4_write_super(struct super_block *sb);
69 static void ext4_write_super_lockfs(struct super_block *sb);
70
71
72 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
73                                struct ext4_group_desc *bg)
74 {
75         return le32_to_cpu(bg->bg_block_bitmap_lo) |
76                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
77                 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
78 }
79
80 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
81                                struct ext4_group_desc *bg)
82 {
83         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
84                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
85                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
86 }
87
88 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
89                               struct ext4_group_desc *bg)
90 {
91         return le32_to_cpu(bg->bg_inode_table_lo) |
92                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
93                 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
94 }
95
96 void ext4_block_bitmap_set(struct super_block *sb,
97                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
98 {
99         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
100         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
101                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
102 }
103
104 void ext4_inode_bitmap_set(struct super_block *sb,
105                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
106 {
107         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
108         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
109                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
110 }
111
112 void ext4_inode_table_set(struct super_block *sb,
113                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
114 {
115         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
116         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
117                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
118 }
119
120 /*
121  * Wrappers for jbd2_journal_start/end.
122  *
123  * The only special thing we need to do here is to make sure that all
124  * journal_end calls result in the superblock being marked dirty, so
125  * that sync() will call the filesystem's write_super callback if
126  * appropriate.
127  */
128 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
129 {
130         journal_t *journal;
131
132         if (sb->s_flags & MS_RDONLY)
133                 return ERR_PTR(-EROFS);
134
135         /* Special case here: if the journal has aborted behind our
136          * backs (eg. EIO in the commit thread), then we still need to
137          * take the FS itself readonly cleanly. */
138         journal = EXT4_SB(sb)->s_journal;
139         if (is_journal_aborted(journal)) {
140                 ext4_abort(sb, __func__,
141                            "Detected aborted journal");
142                 return ERR_PTR(-EROFS);
143         }
144
145         return jbd2_journal_start(journal, nblocks);
146 }
147
148 /*
149  * The only special thing we need to do here is to make sure that all
150  * jbd2_journal_stop calls result in the superblock being marked dirty, so
151  * that sync() will call the filesystem's write_super callback if
152  * appropriate.
153  */
154 int __ext4_journal_stop(const char *where, handle_t *handle)
155 {
156         struct super_block *sb;
157         int err;
158         int rc;
159
160         sb = handle->h_transaction->t_journal->j_private;
161         err = handle->h_err;
162         rc = jbd2_journal_stop(handle);
163
164         if (!err)
165                 err = rc;
166         if (err)
167                 __ext4_std_error(sb, where, err);
168         return err;
169 }
170
171 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
172                 struct buffer_head *bh, handle_t *handle, int err)
173 {
174         char nbuf[16];
175         const char *errstr = ext4_decode_error(NULL, err, nbuf);
176
177         if (bh)
178                 BUFFER_TRACE(bh, "abort");
179
180         if (!handle->h_err)
181                 handle->h_err = err;
182
183         if (is_handle_aborted(handle))
184                 return;
185
186         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
187                caller, errstr, err_fn);
188
189         jbd2_journal_abort_handle(handle);
190 }
191
192 /* Deal with the reporting of failure conditions on a filesystem such as
193  * inconsistencies detected or read IO failures.
194  *
195  * On ext2, we can store the error state of the filesystem in the
196  * superblock.  That is not possible on ext4, because we may have other
197  * write ordering constraints on the superblock which prevent us from
198  * writing it out straight away; and given that the journal is about to
199  * be aborted, we can't rely on the current, or future, transactions to
200  * write out the superblock safely.
201  *
202  * We'll just use the jbd2_journal_abort() error code to record an error in
203  * the journal instead.  On recovery, the journal will compain about
204  * that error until we've noted it down and cleared it.
205  */
206
207 static void ext4_handle_error(struct super_block *sb)
208 {
209         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
210
211         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
212         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
213
214         if (sb->s_flags & MS_RDONLY)
215                 return;
216
217         if (!test_opt(sb, ERRORS_CONT)) {
218                 journal_t *journal = EXT4_SB(sb)->s_journal;
219
220                 EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
221                 if (journal)
222                         jbd2_journal_abort(journal, -EIO);
223         }
224         if (test_opt(sb, ERRORS_RO)) {
225                 printk(KERN_CRIT "Remounting filesystem read-only\n");
226                 sb->s_flags |= MS_RDONLY;
227         }
228         ext4_commit_super(sb, es, 1);
229         if (test_opt(sb, ERRORS_PANIC))
230                 panic("EXT4-fs (device %s): panic forced after error\n",
231                         sb->s_id);
232 }
233
234 void ext4_error(struct super_block *sb, const char *function,
235                 const char *fmt, ...)
236 {
237         va_list args;
238
239         va_start(args, fmt);
240         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
241         vprintk(fmt, args);
242         printk("\n");
243         va_end(args);
244
245         ext4_handle_error(sb);
246 }
247
248 static const char *ext4_decode_error(struct super_block *sb, int errno,
249                                      char nbuf[16])
250 {
251         char *errstr = NULL;
252
253         switch (errno) {
254         case -EIO:
255                 errstr = "IO failure";
256                 break;
257         case -ENOMEM:
258                 errstr = "Out of memory";
259                 break;
260         case -EROFS:
261                 if (!sb || EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT)
262                         errstr = "Journal has aborted";
263                 else
264                         errstr = "Readonly filesystem";
265                 break;
266         default:
267                 /* If the caller passed in an extra buffer for unknown
268                  * errors, textualise them now.  Else we just return
269                  * NULL. */
270                 if (nbuf) {
271                         /* Check for truncated error codes... */
272                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
273                                 errstr = nbuf;
274                 }
275                 break;
276         }
277
278         return errstr;
279 }
280
281 /* __ext4_std_error decodes expected errors from journaling functions
282  * automatically and invokes the appropriate error response.  */
283
284 void __ext4_std_error(struct super_block *sb, const char *function, int errno)
285 {
286         char nbuf[16];
287         const char *errstr;
288
289         /* Special case: if the error is EROFS, and we're not already
290          * inside a transaction, then there's really no point in logging
291          * an error. */
292         if (errno == -EROFS && journal_current_handle() == NULL &&
293             (sb->s_flags & MS_RDONLY))
294                 return;
295
296         errstr = ext4_decode_error(sb, errno, nbuf);
297         printk(KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
298                sb->s_id, function, errstr);
299
300         ext4_handle_error(sb);
301 }
302
303 /*
304  * ext4_abort is a much stronger failure handler than ext4_error.  The
305  * abort function may be used to deal with unrecoverable failures such
306  * as journal IO errors or ENOMEM at a critical moment in log management.
307  *
308  * We unconditionally force the filesystem into an ABORT|READONLY state,
309  * unless the error response on the fs has been set to panic in which
310  * case we take the easy way out and panic immediately.
311  */
312
313 void ext4_abort(struct super_block *sb, const char *function,
314                 const char *fmt, ...)
315 {
316         va_list args;
317
318         printk(KERN_CRIT "ext4_abort called.\n");
319
320         va_start(args, fmt);
321         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
322         vprintk(fmt, args);
323         printk("\n");
324         va_end(args);
325
326         if (test_opt(sb, ERRORS_PANIC))
327                 panic("EXT4-fs panic from previous error\n");
328
329         if (sb->s_flags & MS_RDONLY)
330                 return;
331
332         printk(KERN_CRIT "Remounting filesystem read-only\n");
333         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
334         sb->s_flags |= MS_RDONLY;
335         EXT4_SB(sb)->s_mount_opt |= EXT4_MOUNT_ABORT;
336         if (EXT4_SB(sb)->s_journal)
337                 jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
338 }
339
340 void ext4_warning(struct super_block *sb, const char *function,
341                   const char *fmt, ...)
342 {
343         va_list args;
344
345         va_start(args, fmt);
346         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
347                sb->s_id, function);
348         vprintk(fmt, args);
349         printk("\n");
350         va_end(args);
351 }
352
353 void ext4_update_dynamic_rev(struct super_block *sb)
354 {
355         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
356
357         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
358                 return;
359
360         ext4_warning(sb, __func__,
361                      "updating to rev %d because of new feature flag, "
362                      "running e2fsck is recommended",
363                      EXT4_DYNAMIC_REV);
364
365         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
366         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
367         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
368         /* leave es->s_feature_*compat flags alone */
369         /* es->s_uuid will be set by e2fsck if empty */
370
371         /*
372          * The rest of the superblock fields should be zero, and if not it
373          * means they are likely already in use, so leave them alone.  We
374          * can leave it up to e2fsck to clean up any inconsistencies there.
375          */
376 }
377
378 /*
379  * Open the external journal device
380  */
381 static struct block_device *ext4_blkdev_get(dev_t dev)
382 {
383         struct block_device *bdev;
384         char b[BDEVNAME_SIZE];
385
386         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
387         if (IS_ERR(bdev))
388                 goto fail;
389         return bdev;
390
391 fail:
392         printk(KERN_ERR "EXT4: failed to open journal device %s: %ld\n",
393                         __bdevname(dev, b), PTR_ERR(bdev));
394         return NULL;
395 }
396
397 /*
398  * Release the journal device
399  */
400 static int ext4_blkdev_put(struct block_device *bdev)
401 {
402         bd_release(bdev);
403         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
404 }
405
406 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
407 {
408         struct block_device *bdev;
409         int ret = -ENODEV;
410
411         bdev = sbi->journal_bdev;
412         if (bdev) {
413                 ret = ext4_blkdev_put(bdev);
414                 sbi->journal_bdev = NULL;
415         }
416         return ret;
417 }
418
419 static inline struct inode *orphan_list_entry(struct list_head *l)
420 {
421         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
422 }
423
424 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
425 {
426         struct list_head *l;
427
428         printk(KERN_ERR "sb orphan head is %d\n",
429                le32_to_cpu(sbi->s_es->s_last_orphan));
430
431         printk(KERN_ERR "sb_info orphan list:\n");
432         list_for_each(l, &sbi->s_orphan) {
433                 struct inode *inode = orphan_list_entry(l);
434                 printk(KERN_ERR "  "
435                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
436                        inode->i_sb->s_id, inode->i_ino, inode,
437                        inode->i_mode, inode->i_nlink,
438                        NEXT_ORPHAN(inode));
439         }
440 }
441
442 static void ext4_put_super(struct super_block *sb)
443 {
444         struct ext4_sb_info *sbi = EXT4_SB(sb);
445         struct ext4_super_block *es = sbi->s_es;
446         int i, err;
447
448         ext4_mb_release(sb);
449         ext4_ext_release(sb);
450         ext4_xattr_put_super(sb);
451         err = jbd2_journal_destroy(sbi->s_journal);
452         sbi->s_journal = NULL;
453         if (err < 0)
454                 ext4_abort(sb, __func__, "Couldn't clean up the journal");
455
456         if (!(sb->s_flags & MS_RDONLY)) {
457                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
458                 es->s_state = cpu_to_le16(sbi->s_mount_state);
459                 ext4_commit_super(sb, es, 1);
460         }
461         if (sbi->s_proc) {
462                 remove_proc_entry("inode_readahead_blks", sbi->s_proc);
463                 remove_proc_entry(sb->s_id, ext4_proc_root);
464         }
465
466         for (i = 0; i < sbi->s_gdb_count; i++)
467                 brelse(sbi->s_group_desc[i]);
468         kfree(sbi->s_group_desc);
469         kfree(sbi->s_flex_groups);
470         percpu_counter_destroy(&sbi->s_freeblocks_counter);
471         percpu_counter_destroy(&sbi->s_freeinodes_counter);
472         percpu_counter_destroy(&sbi->s_dirs_counter);
473         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
474         brelse(sbi->s_sbh);
475 #ifdef CONFIG_QUOTA
476         for (i = 0; i < MAXQUOTAS; i++)
477                 kfree(sbi->s_qf_names[i]);
478 #endif
479
480         /* Debugging code just in case the in-memory inode orphan list
481          * isn't empty.  The on-disk one can be non-empty if we've
482          * detected an error and taken the fs readonly, but the
483          * in-memory list had better be clean by this point. */
484         if (!list_empty(&sbi->s_orphan))
485                 dump_orphan_list(sb, sbi);
486         J_ASSERT(list_empty(&sbi->s_orphan));
487
488         invalidate_bdev(sb->s_bdev);
489         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
490                 /*
491                  * Invalidate the journal device's buffers.  We don't want them
492                  * floating about in memory - the physical journal device may
493                  * hotswapped, and it breaks the `ro-after' testing code.
494                  */
495                 sync_blockdev(sbi->journal_bdev);
496                 invalidate_bdev(sbi->journal_bdev);
497                 ext4_blkdev_remove(sbi);
498         }
499         sb->s_fs_info = NULL;
500         kfree(sbi);
501         return;
502 }
503
504 static struct kmem_cache *ext4_inode_cachep;
505
506 /*
507  * Called inside transaction, so use GFP_NOFS
508  */
509 static struct inode *ext4_alloc_inode(struct super_block *sb)
510 {
511         struct ext4_inode_info *ei;
512
513         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
514         if (!ei)
515                 return NULL;
516 #ifdef CONFIG_EXT4_FS_POSIX_ACL
517         ei->i_acl = EXT4_ACL_NOT_CACHED;
518         ei->i_default_acl = EXT4_ACL_NOT_CACHED;
519 #endif
520         ei->vfs_inode.i_version = 1;
521         ei->vfs_inode.i_data.writeback_index = 0;
522         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
523         INIT_LIST_HEAD(&ei->i_prealloc_list);
524         spin_lock_init(&ei->i_prealloc_lock);
525         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
526         ei->i_reserved_data_blocks = 0;
527         ei->i_reserved_meta_blocks = 0;
528         ei->i_allocated_meta_blocks = 0;
529         ei->i_delalloc_reserved_flag = 0;
530         spin_lock_init(&(ei->i_block_reservation_lock));
531         return &ei->vfs_inode;
532 }
533
534 static void ext4_destroy_inode(struct inode *inode)
535 {
536         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
537                 printk("EXT4 Inode %p: orphan list check failed!\n",
538                         EXT4_I(inode));
539                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
540                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
541                                 true);
542                 dump_stack();
543         }
544         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
545 }
546
547 static void init_once(void *foo)
548 {
549         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
550
551         INIT_LIST_HEAD(&ei->i_orphan);
552 #ifdef CONFIG_EXT4_FS_XATTR
553         init_rwsem(&ei->xattr_sem);
554 #endif
555         init_rwsem(&ei->i_data_sem);
556         inode_init_once(&ei->vfs_inode);
557 }
558
559 static int init_inodecache(void)
560 {
561         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
562                                              sizeof(struct ext4_inode_info),
563                                              0, (SLAB_RECLAIM_ACCOUNT|
564                                                 SLAB_MEM_SPREAD),
565                                              init_once);
566         if (ext4_inode_cachep == NULL)
567                 return -ENOMEM;
568         return 0;
569 }
570
571 static void destroy_inodecache(void)
572 {
573         kmem_cache_destroy(ext4_inode_cachep);
574 }
575
576 static void ext4_clear_inode(struct inode *inode)
577 {
578 #ifdef CONFIG_EXT4_FS_POSIX_ACL
579         if (EXT4_I(inode)->i_acl &&
580                         EXT4_I(inode)->i_acl != EXT4_ACL_NOT_CACHED) {
581                 posix_acl_release(EXT4_I(inode)->i_acl);
582                 EXT4_I(inode)->i_acl = EXT4_ACL_NOT_CACHED;
583         }
584         if (EXT4_I(inode)->i_default_acl &&
585                         EXT4_I(inode)->i_default_acl != EXT4_ACL_NOT_CACHED) {
586                 posix_acl_release(EXT4_I(inode)->i_default_acl);
587                 EXT4_I(inode)->i_default_acl = EXT4_ACL_NOT_CACHED;
588         }
589 #endif
590         ext4_discard_preallocations(inode);
591         jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
592                                        &EXT4_I(inode)->jinode);
593 }
594
595 static inline void ext4_show_quota_options(struct seq_file *seq,
596                                            struct super_block *sb)
597 {
598 #if defined(CONFIG_QUOTA)
599         struct ext4_sb_info *sbi = EXT4_SB(sb);
600
601         if (sbi->s_jquota_fmt)
602                 seq_printf(seq, ",jqfmt=%s",
603                 (sbi->s_jquota_fmt == QFMT_VFS_OLD) ? "vfsold" : "vfsv0");
604
605         if (sbi->s_qf_names[USRQUOTA])
606                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
607
608         if (sbi->s_qf_names[GRPQUOTA])
609                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
610
611         if (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA)
612                 seq_puts(seq, ",usrquota");
613
614         if (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)
615                 seq_puts(seq, ",grpquota");
616 #endif
617 }
618
619 /*
620  * Show an option if
621  *  - it's set to a non-default value OR
622  *  - if the per-sb default is different from the global default
623  */
624 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
625 {
626         int def_errors;
627         unsigned long def_mount_opts;
628         struct super_block *sb = vfs->mnt_sb;
629         struct ext4_sb_info *sbi = EXT4_SB(sb);
630         struct ext4_super_block *es = sbi->s_es;
631
632         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
633         def_errors     = le16_to_cpu(es->s_errors);
634
635         if (sbi->s_sb_block != 1)
636                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
637         if (test_opt(sb, MINIX_DF))
638                 seq_puts(seq, ",minixdf");
639         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
640                 seq_puts(seq, ",grpid");
641         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
642                 seq_puts(seq, ",nogrpid");
643         if (sbi->s_resuid != EXT4_DEF_RESUID ||
644             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
645                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
646         }
647         if (sbi->s_resgid != EXT4_DEF_RESGID ||
648             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
649                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
650         }
651         if (test_opt(sb, ERRORS_RO)) {
652                 if (def_errors == EXT4_ERRORS_PANIC ||
653                     def_errors == EXT4_ERRORS_CONTINUE) {
654                         seq_puts(seq, ",errors=remount-ro");
655                 }
656         }
657         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
658                 seq_puts(seq, ",errors=continue");
659         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
660                 seq_puts(seq, ",errors=panic");
661         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
662                 seq_puts(seq, ",nouid32");
663         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
664                 seq_puts(seq, ",debug");
665         if (test_opt(sb, OLDALLOC))
666                 seq_puts(seq, ",oldalloc");
667 #ifdef CONFIG_EXT4_FS_XATTR
668         if (test_opt(sb, XATTR_USER) &&
669                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
670                 seq_puts(seq, ",user_xattr");
671         if (!test_opt(sb, XATTR_USER) &&
672             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
673                 seq_puts(seq, ",nouser_xattr");
674         }
675 #endif
676 #ifdef CONFIG_EXT4_FS_POSIX_ACL
677         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
678                 seq_puts(seq, ",acl");
679         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
680                 seq_puts(seq, ",noacl");
681 #endif
682         if (!test_opt(sb, RESERVATION))
683                 seq_puts(seq, ",noreservation");
684         if (sbi->s_commit_interval) {
685                 seq_printf(seq, ",commit=%u",
686                            (unsigned) (sbi->s_commit_interval / HZ));
687         }
688         /*
689          * We're changing the default of barrier mount option, so
690          * let's always display its mount state so it's clear what its
691          * status is.
692          */
693         seq_puts(seq, ",barrier=");
694         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
695         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
696                 seq_puts(seq, ",journal_async_commit");
697         if (test_opt(sb, NOBH))
698                 seq_puts(seq, ",nobh");
699         if (!test_opt(sb, EXTENTS))
700                 seq_puts(seq, ",noextents");
701         if (test_opt(sb, I_VERSION))
702                 seq_puts(seq, ",i_version");
703         if (!test_opt(sb, DELALLOC))
704                 seq_puts(seq, ",nodelalloc");
705
706
707         if (sbi->s_stripe)
708                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
709         /*
710          * journal mode get enabled in different ways
711          * So just print the value even if we didn't specify it
712          */
713         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
714                 seq_puts(seq, ",data=journal");
715         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
716                 seq_puts(seq, ",data=ordered");
717         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
718                 seq_puts(seq, ",data=writeback");
719
720         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
721                 seq_printf(seq, ",inode_readahead_blks=%u",
722                            sbi->s_inode_readahead_blks);
723
724         if (test_opt(sb, DATA_ERR_ABORT))
725                 seq_puts(seq, ",data_err=abort");
726
727         ext4_show_quota_options(seq, sb);
728         return 0;
729 }
730
731
732 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
733                 u64 ino, u32 generation)
734 {
735         struct inode *inode;
736
737         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
738                 return ERR_PTR(-ESTALE);
739         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
740                 return ERR_PTR(-ESTALE);
741
742         /* iget isn't really right if the inode is currently unallocated!!
743          *
744          * ext4_read_inode will return a bad_inode if the inode had been
745          * deleted, so we should be safe.
746          *
747          * Currently we don't know the generation for parent directory, so
748          * a generation of 0 means "accept any"
749          */
750         inode = ext4_iget(sb, ino);
751         if (IS_ERR(inode))
752                 return ERR_CAST(inode);
753         if (generation && inode->i_generation != generation) {
754                 iput(inode);
755                 return ERR_PTR(-ESTALE);
756         }
757
758         return inode;
759 }
760
761 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
762                 int fh_len, int fh_type)
763 {
764         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
765                                     ext4_nfs_get_inode);
766 }
767
768 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
769                 int fh_len, int fh_type)
770 {
771         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
772                                     ext4_nfs_get_inode);
773 }
774
775 #ifdef CONFIG_QUOTA
776 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
777 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
778
779 static int ext4_dquot_initialize(struct inode *inode, int type);
780 static int ext4_dquot_drop(struct inode *inode);
781 static int ext4_write_dquot(struct dquot *dquot);
782 static int ext4_acquire_dquot(struct dquot *dquot);
783 static int ext4_release_dquot(struct dquot *dquot);
784 static int ext4_mark_dquot_dirty(struct dquot *dquot);
785 static int ext4_write_info(struct super_block *sb, int type);
786 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
787                                 char *path, int remount);
788 static int ext4_quota_on_mount(struct super_block *sb, int type);
789 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
790                                size_t len, loff_t off);
791 static ssize_t ext4_quota_write(struct super_block *sb, int type,
792                                 const char *data, size_t len, loff_t off);
793
794 static struct dquot_operations ext4_quota_operations = {
795         .initialize     = ext4_dquot_initialize,
796         .drop           = ext4_dquot_drop,
797         .alloc_space    = dquot_alloc_space,
798         .alloc_inode    = dquot_alloc_inode,
799         .free_space     = dquot_free_space,
800         .free_inode     = dquot_free_inode,
801         .transfer       = dquot_transfer,
802         .write_dquot    = ext4_write_dquot,
803         .acquire_dquot  = ext4_acquire_dquot,
804         .release_dquot  = ext4_release_dquot,
805         .mark_dirty     = ext4_mark_dquot_dirty,
806         .write_info     = ext4_write_info,
807         .alloc_dquot    = dquot_alloc,
808         .destroy_dquot  = dquot_destroy,
809 };
810
811 static struct quotactl_ops ext4_qctl_operations = {
812         .quota_on       = ext4_quota_on,
813         .quota_off      = vfs_quota_off,
814         .quota_sync     = vfs_quota_sync,
815         .get_info       = vfs_get_dqinfo,
816         .set_info       = vfs_set_dqinfo,
817         .get_dqblk      = vfs_get_dqblk,
818         .set_dqblk      = vfs_set_dqblk
819 };
820 #endif
821
822 static const struct super_operations ext4_sops = {
823         .alloc_inode    = ext4_alloc_inode,
824         .destroy_inode  = ext4_destroy_inode,
825         .write_inode    = ext4_write_inode,
826         .dirty_inode    = ext4_dirty_inode,
827         .delete_inode   = ext4_delete_inode,
828         .put_super      = ext4_put_super,
829         .write_super    = ext4_write_super,
830         .sync_fs        = ext4_sync_fs,
831         .write_super_lockfs = ext4_write_super_lockfs,
832         .unlockfs       = ext4_unlockfs,
833         .statfs         = ext4_statfs,
834         .remount_fs     = ext4_remount,
835         .clear_inode    = ext4_clear_inode,
836         .show_options   = ext4_show_options,
837 #ifdef CONFIG_QUOTA
838         .quota_read     = ext4_quota_read,
839         .quota_write    = ext4_quota_write,
840 #endif
841 };
842
843 static const struct export_operations ext4_export_ops = {
844         .fh_to_dentry = ext4_fh_to_dentry,
845         .fh_to_parent = ext4_fh_to_parent,
846         .get_parent = ext4_get_parent,
847 };
848
849 enum {
850         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
851         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
852         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
853         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
854         Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
855         Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
856         Opt_journal_checksum, Opt_journal_async_commit,
857         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
858         Opt_data_err_abort, Opt_data_err_ignore,
859         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
860         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
861         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
862         Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version,
863         Opt_stripe, Opt_delalloc, Opt_nodelalloc,
864         Opt_inode_readahead_blks
865 };
866
867 static const match_table_t tokens = {
868         {Opt_bsd_df, "bsddf"},
869         {Opt_minix_df, "minixdf"},
870         {Opt_grpid, "grpid"},
871         {Opt_grpid, "bsdgroups"},
872         {Opt_nogrpid, "nogrpid"},
873         {Opt_nogrpid, "sysvgroups"},
874         {Opt_resgid, "resgid=%u"},
875         {Opt_resuid, "resuid=%u"},
876         {Opt_sb, "sb=%u"},
877         {Opt_err_cont, "errors=continue"},
878         {Opt_err_panic, "errors=panic"},
879         {Opt_err_ro, "errors=remount-ro"},
880         {Opt_nouid32, "nouid32"},
881         {Opt_debug, "debug"},
882         {Opt_oldalloc, "oldalloc"},
883         {Opt_orlov, "orlov"},
884         {Opt_user_xattr, "user_xattr"},
885         {Opt_nouser_xattr, "nouser_xattr"},
886         {Opt_acl, "acl"},
887         {Opt_noacl, "noacl"},
888         {Opt_reservation, "reservation"},
889         {Opt_noreservation, "noreservation"},
890         {Opt_noload, "noload"},
891         {Opt_nobh, "nobh"},
892         {Opt_bh, "bh"},
893         {Opt_commit, "commit=%u"},
894         {Opt_journal_update, "journal=update"},
895         {Opt_journal_inum, "journal=%u"},
896         {Opt_journal_dev, "journal_dev=%u"},
897         {Opt_journal_checksum, "journal_checksum"},
898         {Opt_journal_async_commit, "journal_async_commit"},
899         {Opt_abort, "abort"},
900         {Opt_data_journal, "data=journal"},
901         {Opt_data_ordered, "data=ordered"},
902         {Opt_data_writeback, "data=writeback"},
903         {Opt_data_err_abort, "data_err=abort"},
904         {Opt_data_err_ignore, "data_err=ignore"},
905         {Opt_offusrjquota, "usrjquota="},
906         {Opt_usrjquota, "usrjquota=%s"},
907         {Opt_offgrpjquota, "grpjquota="},
908         {Opt_grpjquota, "grpjquota=%s"},
909         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
910         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
911         {Opt_grpquota, "grpquota"},
912         {Opt_noquota, "noquota"},
913         {Opt_quota, "quota"},
914         {Opt_usrquota, "usrquota"},
915         {Opt_barrier, "barrier=%u"},
916         {Opt_extents, "extents"},
917         {Opt_noextents, "noextents"},
918         {Opt_i_version, "i_version"},
919         {Opt_stripe, "stripe=%u"},
920         {Opt_resize, "resize"},
921         {Opt_delalloc, "delalloc"},
922         {Opt_nodelalloc, "nodelalloc"},
923         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
924         {Opt_err, NULL},
925 };
926
927 static ext4_fsblk_t get_sb_block(void **data)
928 {
929         ext4_fsblk_t    sb_block;
930         char            *options = (char *) *data;
931
932         if (!options || strncmp(options, "sb=", 3) != 0)
933                 return 1;       /* Default location */
934         options += 3;
935         /*todo: use simple_strtoll with >32bit ext4 */
936         sb_block = simple_strtoul(options, &options, 0);
937         if (*options && *options != ',') {
938                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
939                        (char *) *data);
940                 return 1;
941         }
942         if (*options == ',')
943                 options++;
944         *data = (void *) options;
945         return sb_block;
946 }
947
948 static int parse_options(char *options, struct super_block *sb,
949                          unsigned int *inum, unsigned long *journal_devnum,
950                          ext4_fsblk_t *n_blocks_count, int is_remount)
951 {
952         struct ext4_sb_info *sbi = EXT4_SB(sb);
953         char *p;
954         substring_t args[MAX_OPT_ARGS];
955         int data_opt = 0;
956         int option;
957 #ifdef CONFIG_QUOTA
958         int qtype, qfmt;
959         char *qname;
960 #endif
961         ext4_fsblk_t last_block;
962
963         if (!options)
964                 return 1;
965
966         while ((p = strsep(&options, ",")) != NULL) {
967                 int token;
968                 if (!*p)
969                         continue;
970
971                 token = match_token(p, tokens, args);
972                 switch (token) {
973                 case Opt_bsd_df:
974                         clear_opt(sbi->s_mount_opt, MINIX_DF);
975                         break;
976                 case Opt_minix_df:
977                         set_opt(sbi->s_mount_opt, MINIX_DF);
978                         break;
979                 case Opt_grpid:
980                         set_opt(sbi->s_mount_opt, GRPID);
981                         break;
982                 case Opt_nogrpid:
983                         clear_opt(sbi->s_mount_opt, GRPID);
984                         break;
985                 case Opt_resuid:
986                         if (match_int(&args[0], &option))
987                                 return 0;
988                         sbi->s_resuid = option;
989                         break;
990                 case Opt_resgid:
991                         if (match_int(&args[0], &option))
992                                 return 0;
993                         sbi->s_resgid = option;
994                         break;
995                 case Opt_sb:
996                         /* handled by get_sb_block() instead of here */
997                         /* *sb_block = match_int(&args[0]); */
998                         break;
999                 case Opt_err_panic:
1000                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1001                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1002                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1003                         break;
1004                 case Opt_err_ro:
1005                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1006                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1007                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1008                         break;
1009                 case Opt_err_cont:
1010                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1011                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1012                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1013                         break;
1014                 case Opt_nouid32:
1015                         set_opt(sbi->s_mount_opt, NO_UID32);
1016                         break;
1017                 case Opt_debug:
1018                         set_opt(sbi->s_mount_opt, DEBUG);
1019                         break;
1020                 case Opt_oldalloc:
1021                         set_opt(sbi->s_mount_opt, OLDALLOC);
1022                         break;
1023                 case Opt_orlov:
1024                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1025                         break;
1026 #ifdef CONFIG_EXT4_FS_XATTR
1027                 case Opt_user_xattr:
1028                         set_opt(sbi->s_mount_opt, XATTR_USER);
1029                         break;
1030                 case Opt_nouser_xattr:
1031                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1032                         break;
1033 #else
1034                 case Opt_user_xattr:
1035                 case Opt_nouser_xattr:
1036                         printk(KERN_ERR "EXT4 (no)user_xattr options "
1037                                "not supported\n");
1038                         break;
1039 #endif
1040 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1041                 case Opt_acl:
1042                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1043                         break;
1044                 case Opt_noacl:
1045                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1046                         break;
1047 #else
1048                 case Opt_acl:
1049                 case Opt_noacl:
1050                         printk(KERN_ERR "EXT4 (no)acl options "
1051                                "not supported\n");
1052                         break;
1053 #endif
1054                 case Opt_reservation:
1055                         set_opt(sbi->s_mount_opt, RESERVATION);
1056                         break;
1057                 case Opt_noreservation:
1058                         clear_opt(sbi->s_mount_opt, RESERVATION);
1059                         break;
1060                 case Opt_journal_update:
1061                         /* @@@ FIXME */
1062                         /* Eventually we will want to be able to create
1063                            a journal file here.  For now, only allow the
1064                            user to specify an existing inode to be the
1065                            journal file. */
1066                         if (is_remount) {
1067                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1068                                        "journal on remount\n");
1069                                 return 0;
1070                         }
1071                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1072                         break;
1073                 case Opt_journal_inum:
1074                         if (is_remount) {
1075                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1076                                        "journal on remount\n");
1077                                 return 0;
1078                         }
1079                         if (match_int(&args[0], &option))
1080                                 return 0;
1081                         *inum = option;
1082                         break;
1083                 case Opt_journal_dev:
1084                         if (is_remount) {
1085                                 printk(KERN_ERR "EXT4-fs: cannot specify "
1086                                        "journal on remount\n");
1087                                 return 0;
1088                         }
1089                         if (match_int(&args[0], &option))
1090                                 return 0;
1091                         *journal_devnum = option;
1092                         break;
1093                 case Opt_journal_checksum:
1094                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1095                         break;
1096                 case Opt_journal_async_commit:
1097                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1098                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1099                         break;
1100                 case Opt_noload:
1101                         set_opt(sbi->s_mount_opt, NOLOAD);
1102                         break;
1103                 case Opt_commit:
1104                         if (match_int(&args[0], &option))
1105                                 return 0;
1106                         if (option < 0)
1107                                 return 0;
1108                         if (option == 0)
1109                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1110                         sbi->s_commit_interval = HZ * option;
1111                         break;
1112                 case Opt_data_journal:
1113                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1114                         goto datacheck;
1115                 case Opt_data_ordered:
1116                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1117                         goto datacheck;
1118                 case Opt_data_writeback:
1119                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1120                 datacheck:
1121                         if (is_remount) {
1122                                 if ((sbi->s_mount_opt & EXT4_MOUNT_DATA_FLAGS)
1123                                                 != data_opt) {
1124                                         printk(KERN_ERR
1125                                                 "EXT4-fs: cannot change data "
1126                                                 "mode on remount\n");
1127                                         return 0;
1128                                 }
1129                         } else {
1130                                 sbi->s_mount_opt &= ~EXT4_MOUNT_DATA_FLAGS;
1131                                 sbi->s_mount_opt |= data_opt;
1132                         }
1133                         break;
1134                 case Opt_data_err_abort:
1135                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1136                         break;
1137                 case Opt_data_err_ignore:
1138                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1139                         break;
1140 #ifdef CONFIG_QUOTA
1141                 case Opt_usrjquota:
1142                         qtype = USRQUOTA;
1143                         goto set_qf_name;
1144                 case Opt_grpjquota:
1145                         qtype = GRPQUOTA;
1146 set_qf_name:
1147                         if (sb_any_quota_loaded(sb) &&
1148                             !sbi->s_qf_names[qtype]) {
1149                                 printk(KERN_ERR
1150                                        "EXT4-fs: Cannot change journaled "
1151                                        "quota options when quota turned on.\n");
1152                                 return 0;
1153                         }
1154                         qname = match_strdup(&args[0]);
1155                         if (!qname) {
1156                                 printk(KERN_ERR
1157                                         "EXT4-fs: not enough memory for "
1158                                         "storing quotafile name.\n");
1159                                 return 0;
1160                         }
1161                         if (sbi->s_qf_names[qtype] &&
1162                             strcmp(sbi->s_qf_names[qtype], qname)) {
1163                                 printk(KERN_ERR
1164                                         "EXT4-fs: %s quota file already "
1165                                         "specified.\n", QTYPE2NAME(qtype));
1166                                 kfree(qname);
1167                                 return 0;
1168                         }
1169                         sbi->s_qf_names[qtype] = qname;
1170                         if (strchr(sbi->s_qf_names[qtype], '/')) {
1171                                 printk(KERN_ERR
1172                                         "EXT4-fs: quotafile must be on "
1173                                         "filesystem root.\n");
1174                                 kfree(sbi->s_qf_names[qtype]);
1175                                 sbi->s_qf_names[qtype] = NULL;
1176                                 return 0;
1177                         }
1178                         set_opt(sbi->s_mount_opt, QUOTA);
1179                         break;
1180                 case Opt_offusrjquota:
1181                         qtype = USRQUOTA;
1182                         goto clear_qf_name;
1183                 case Opt_offgrpjquota:
1184                         qtype = GRPQUOTA;
1185 clear_qf_name:
1186                         if (sb_any_quota_loaded(sb) &&
1187                             sbi->s_qf_names[qtype]) {
1188                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1189                                         "journaled quota options when "
1190                                         "quota turned on.\n");
1191                                 return 0;
1192                         }
1193                         /*
1194                          * The space will be released later when all options
1195                          * are confirmed to be correct
1196                          */
1197                         sbi->s_qf_names[qtype] = NULL;
1198                         break;
1199                 case Opt_jqfmt_vfsold:
1200                         qfmt = QFMT_VFS_OLD;
1201                         goto set_qf_format;
1202                 case Opt_jqfmt_vfsv0:
1203                         qfmt = QFMT_VFS_V0;
1204 set_qf_format:
1205                         if (sb_any_quota_loaded(sb) &&
1206                             sbi->s_jquota_fmt != qfmt) {
1207                                 printk(KERN_ERR "EXT4-fs: Cannot change "
1208                                         "journaled quota options when "
1209                                         "quota turned on.\n");
1210                                 return 0;
1211                         }
1212                         sbi->s_jquota_fmt = qfmt;
1213                         break;
1214                 case Opt_quota:
1215                 case Opt_usrquota:
1216                         set_opt(sbi->s_mount_opt, QUOTA);
1217                         set_opt(sbi->s_mount_opt, USRQUOTA);
1218                         break;
1219                 case Opt_grpquota:
1220                         set_opt(sbi->s_mount_opt, QUOTA);
1221                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1222                         break;
1223                 case Opt_noquota:
1224                         if (sb_any_quota_loaded(sb)) {
1225                                 printk(KERN_ERR "EXT4-fs: Cannot change quota "
1226                                         "options when quota turned on.\n");
1227                                 return 0;
1228                         }
1229                         clear_opt(sbi->s_mount_opt, QUOTA);
1230                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1231                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1232                         break;
1233 #else
1234                 case Opt_quota:
1235                 case Opt_usrquota:
1236                 case Opt_grpquota:
1237                         printk(KERN_ERR
1238                                 "EXT4-fs: quota options not supported.\n");
1239                         break;
1240                 case Opt_usrjquota:
1241                 case Opt_grpjquota:
1242                 case Opt_offusrjquota:
1243                 case Opt_offgrpjquota:
1244                 case Opt_jqfmt_vfsold:
1245                 case Opt_jqfmt_vfsv0:
1246                         printk(KERN_ERR
1247                                 "EXT4-fs: journaled quota options not "
1248                                 "supported.\n");
1249                         break;
1250                 case Opt_noquota:
1251                         break;
1252 #endif
1253                 case Opt_abort:
1254                         set_opt(sbi->s_mount_opt, ABORT);
1255                         break;
1256                 case Opt_barrier:
1257                         if (match_int(&args[0], &option))
1258                                 return 0;
1259                         if (option)
1260                                 set_opt(sbi->s_mount_opt, BARRIER);
1261                         else
1262                                 clear_opt(sbi->s_mount_opt, BARRIER);
1263                         break;
1264                 case Opt_ignore:
1265                         break;
1266                 case Opt_resize:
1267                         if (!is_remount) {
1268                                 printk("EXT4-fs: resize option only available "
1269                                         "for remount\n");
1270                                 return 0;
1271                         }
1272                         if (match_int(&args[0], &option) != 0)
1273                                 return 0;
1274                         *n_blocks_count = option;
1275                         break;
1276                 case Opt_nobh:
1277                         set_opt(sbi->s_mount_opt, NOBH);
1278                         break;
1279                 case Opt_bh:
1280                         clear_opt(sbi->s_mount_opt, NOBH);
1281                         break;
1282                 case Opt_extents:
1283                         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,
1284                                         EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1285                                 ext4_warning(sb, __func__,
1286                                         "extents feature not enabled "
1287                                         "on this filesystem, use tune2fs\n");
1288                                 return 0;
1289                         }
1290                         set_opt(sbi->s_mount_opt, EXTENTS);
1291                         break;
1292                 case Opt_noextents:
1293                         /*
1294                          * When e2fsprogs support resizing an already existing
1295                          * ext3 file system to greater than 2**32 we need to
1296                          * add support to block allocator to handle growing
1297                          * already existing block  mapped inode so that blocks
1298                          * allocated for them fall within 2**32
1299                          */
1300                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1301                         if (last_block  > 0xffffffffULL) {
1302                                 printk(KERN_ERR "EXT4-fs: Filesystem too "
1303                                                 "large to mount with "
1304                                                 "-o noextents options\n");
1305                                 return 0;
1306                         }
1307                         clear_opt(sbi->s_mount_opt, EXTENTS);
1308                         break;
1309                 case Opt_i_version:
1310                         set_opt(sbi->s_mount_opt, I_VERSION);
1311                         sb->s_flags |= MS_I_VERSION;
1312                         break;
1313                 case Opt_nodelalloc:
1314                         clear_opt(sbi->s_mount_opt, DELALLOC);
1315                         break;
1316                 case Opt_stripe:
1317                         if (match_int(&args[0], &option))
1318                                 return 0;
1319                         if (option < 0)
1320                                 return 0;
1321                         sbi->s_stripe = option;
1322                         break;
1323                 case Opt_delalloc:
1324                         set_opt(sbi->s_mount_opt, DELALLOC);
1325                         break;
1326                 case Opt_inode_readahead_blks:
1327                         if (match_int(&args[0], &option))
1328                                 return 0;
1329                         if (option < 0 || option > (1 << 30))
1330                                 return 0;
1331                         sbi->s_inode_readahead_blks = option;
1332                         break;
1333                 default:
1334                         printk(KERN_ERR
1335                                "EXT4-fs: Unrecognized mount option \"%s\" "
1336                                "or missing value\n", p);
1337                         return 0;
1338                 }
1339         }
1340 #ifdef CONFIG_QUOTA
1341         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1342                 if ((sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA) &&
1343                      sbi->s_qf_names[USRQUOTA])
1344                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1345
1346                 if ((sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA) &&
1347                      sbi->s_qf_names[GRPQUOTA])
1348                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1349
1350                 if ((sbi->s_qf_names[USRQUOTA] &&
1351                                 (sbi->s_mount_opt & EXT4_MOUNT_GRPQUOTA)) ||
1352                     (sbi->s_qf_names[GRPQUOTA] &&
1353                                 (sbi->s_mount_opt & EXT4_MOUNT_USRQUOTA))) {
1354                         printk(KERN_ERR "EXT4-fs: old and new quota "
1355                                         "format mixing.\n");
1356                         return 0;
1357                 }
1358
1359                 if (!sbi->s_jquota_fmt) {
1360                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1361                                         "not specified.\n");
1362                         return 0;
1363                 }
1364         } else {
1365                 if (sbi->s_jquota_fmt) {
1366                         printk(KERN_ERR "EXT4-fs: journaled quota format "
1367                                         "specified with no journaling "
1368                                         "enabled.\n");
1369                         return 0;
1370                 }
1371         }
1372 #endif
1373         return 1;
1374 }
1375
1376 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1377                             int read_only)
1378 {
1379         struct ext4_sb_info *sbi = EXT4_SB(sb);
1380         int res = 0;
1381
1382         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1383                 printk(KERN_ERR "EXT4-fs warning: revision level too high, "
1384                        "forcing read-only mode\n");
1385                 res = MS_RDONLY;
1386         }
1387         if (read_only)
1388                 return res;
1389         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1390                 printk(KERN_WARNING "EXT4-fs warning: mounting unchecked fs, "
1391                        "running e2fsck is recommended\n");
1392         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1393                 printk(KERN_WARNING
1394                        "EXT4-fs warning: mounting fs with errors, "
1395                        "running e2fsck is recommended\n");
1396         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1397                  le16_to_cpu(es->s_mnt_count) >=
1398                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1399                 printk(KERN_WARNING
1400                        "EXT4-fs warning: maximal mount count reached, "
1401                        "running e2fsck is recommended\n");
1402         else if (le32_to_cpu(es->s_checkinterval) &&
1403                 (le32_to_cpu(es->s_lastcheck) +
1404                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1405                 printk(KERN_WARNING
1406                        "EXT4-fs warning: checktime reached, "
1407                        "running e2fsck is recommended\n");
1408 #if 0
1409                 /* @@@ We _will_ want to clear the valid bit if we find
1410                  * inconsistencies, to force a fsck at reboot.  But for
1411                  * a plain journaled filesystem we can keep it set as
1412                  * valid forever! :)
1413                  */
1414         es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1415 #endif
1416         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1417                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1418         le16_add_cpu(&es->s_mnt_count, 1);
1419         es->s_mtime = cpu_to_le32(get_seconds());
1420         ext4_update_dynamic_rev(sb);
1421         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1422
1423         ext4_commit_super(sb, es, 1);
1424         if (test_opt(sb, DEBUG))
1425                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%lu, "
1426                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1427                         sb->s_blocksize,
1428                         sbi->s_groups_count,
1429                         EXT4_BLOCKS_PER_GROUP(sb),
1430                         EXT4_INODES_PER_GROUP(sb),
1431                         sbi->s_mount_opt);
1432
1433         printk(KERN_INFO "EXT4 FS on %s, %s journal on %s\n",
1434                sb->s_id, EXT4_SB(sb)->s_journal->j_inode ? "internal" :
1435                "external", EXT4_SB(sb)->s_journal->j_devname);
1436         return res;
1437 }
1438
1439 static int ext4_fill_flex_info(struct super_block *sb)
1440 {
1441         struct ext4_sb_info *sbi = EXT4_SB(sb);
1442         struct ext4_group_desc *gdp = NULL;
1443         struct buffer_head *bh;
1444         ext4_group_t flex_group_count;
1445         ext4_group_t flex_group;
1446         int groups_per_flex = 0;
1447         __u64 block_bitmap = 0;
1448         int i;
1449
1450         if (!sbi->s_es->s_log_groups_per_flex) {
1451                 sbi->s_log_groups_per_flex = 0;
1452                 return 1;
1453         }
1454
1455         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1456         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1457
1458         /* We allocate both existing and potentially added groups */
1459         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1460                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1461                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1462         sbi->s_flex_groups = kzalloc(flex_group_count *
1463                                      sizeof(struct flex_groups), GFP_KERNEL);
1464         if (sbi->s_flex_groups == NULL) {
1465                 printk(KERN_ERR "EXT4-fs: not enough memory for "
1466                                 "%lu flex groups\n", flex_group_count);
1467                 goto failed;
1468         }
1469
1470         gdp = ext4_get_group_desc(sb, 1, &bh);
1471         block_bitmap = ext4_block_bitmap(sb, gdp) - 1;
1472
1473         for (i = 0; i < sbi->s_groups_count; i++) {
1474                 gdp = ext4_get_group_desc(sb, i, &bh);
1475
1476                 flex_group = ext4_flex_group(sbi, i);
1477                 sbi->s_flex_groups[flex_group].free_inodes +=
1478                         le16_to_cpu(gdp->bg_free_inodes_count);
1479                 sbi->s_flex_groups[flex_group].free_blocks +=
1480                         le16_to_cpu(gdp->bg_free_blocks_count);
1481         }
1482
1483         return 1;
1484 failed:
1485         return 0;
1486 }
1487
1488 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1489                             struct ext4_group_desc *gdp)
1490 {
1491         __u16 crc = 0;
1492
1493         if (sbi->s_es->s_feature_ro_compat &
1494             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1495                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1496                 __le32 le_group = cpu_to_le32(block_group);
1497
1498                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1499                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1500                 crc = crc16(crc, (__u8 *)gdp, offset);
1501                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1502                 /* for checksum of struct ext4_group_desc do the rest...*/
1503                 if ((sbi->s_es->s_feature_incompat &
1504                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1505                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1506                         crc = crc16(crc, (__u8 *)gdp + offset,
1507                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1508                                         offset);
1509         }
1510
1511         return cpu_to_le16(crc);
1512 }
1513
1514 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1515                                 struct ext4_group_desc *gdp)
1516 {
1517         if ((sbi->s_es->s_feature_ro_compat &
1518              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1519             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1520                 return 0;
1521
1522         return 1;
1523 }
1524
1525 /* Called at mount-time, super-block is locked */
1526 static int ext4_check_descriptors(struct super_block *sb)
1527 {
1528         struct ext4_sb_info *sbi = EXT4_SB(sb);
1529         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1530         ext4_fsblk_t last_block;
1531         ext4_fsblk_t block_bitmap;
1532         ext4_fsblk_t inode_bitmap;
1533         ext4_fsblk_t inode_table;
1534         int flexbg_flag = 0;
1535         ext4_group_t i;
1536
1537         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1538                 flexbg_flag = 1;
1539
1540         ext4_debug("Checking group descriptors");
1541
1542         for (i = 0; i < sbi->s_groups_count; i++) {
1543                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1544
1545                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1546                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1547                 else
1548                         last_block = first_block +
1549                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1550
1551                 block_bitmap = ext4_block_bitmap(sb, gdp);
1552                 if (block_bitmap < first_block || block_bitmap > last_block) {
1553                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1554                                "Block bitmap for group %lu not in group "
1555                                "(block %llu)!\n", i, block_bitmap);
1556                         return 0;
1557                 }
1558                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1559                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1560                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1561                                "Inode bitmap for group %lu not in group "
1562                                "(block %llu)!\n", i, inode_bitmap);
1563                         return 0;
1564                 }
1565                 inode_table = ext4_inode_table(sb, gdp);
1566                 if (inode_table < first_block ||
1567                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1568                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1569                                "Inode table for group %lu not in group "
1570                                "(block %llu)!\n", i, inode_table);
1571                         return 0;
1572                 }
1573                 spin_lock(sb_bgl_lock(sbi, i));
1574                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1575                         printk(KERN_ERR "EXT4-fs: ext4_check_descriptors: "
1576                                "Checksum for group %lu failed (%u!=%u)\n",
1577                                i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1578                                gdp)), le16_to_cpu(gdp->bg_checksum));
1579                         if (!(sb->s_flags & MS_RDONLY)) {
1580                                 spin_unlock(sb_bgl_lock(sbi, i));
1581                                 return 0;
1582                         }
1583                 }
1584                 spin_unlock(sb_bgl_lock(sbi, i));
1585                 if (!flexbg_flag)
1586                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1587         }
1588
1589         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1590         sbi->s_es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
1591         return 1;
1592 }
1593
1594 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1595  * the superblock) which were deleted from all directories, but held open by
1596  * a process at the time of a crash.  We walk the list and try to delete these
1597  * inodes at recovery time (only with a read-write filesystem).
1598  *
1599  * In order to keep the orphan inode chain consistent during traversal (in
1600  * case of crash during recovery), we link each inode into the superblock
1601  * orphan list_head and handle it the same way as an inode deletion during
1602  * normal operation (which journals the operations for us).
1603  *
1604  * We only do an iget() and an iput() on each inode, which is very safe if we
1605  * accidentally point at an in-use or already deleted inode.  The worst that
1606  * can happen in this case is that we get a "bit already cleared" message from
1607  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1608  * e2fsck was run on this filesystem, and it must have already done the orphan
1609  * inode cleanup for us, so we can safely abort without any further action.
1610  */
1611 static void ext4_orphan_cleanup(struct super_block *sb,
1612                                 struct ext4_super_block *es)
1613 {
1614         unsigned int s_flags = sb->s_flags;
1615         int nr_orphans = 0, nr_truncates = 0;
1616 #ifdef CONFIG_QUOTA
1617         int i;
1618 #endif
1619         if (!es->s_last_orphan) {
1620                 jbd_debug(4, "no orphan inodes to clean up\n");
1621                 return;
1622         }
1623
1624         if (bdev_read_only(sb->s_bdev)) {
1625                 printk(KERN_ERR "EXT4-fs: write access "
1626                         "unavailable, skipping orphan cleanup.\n");
1627                 return;
1628         }
1629
1630         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1631                 if (es->s_last_orphan)
1632                         jbd_debug(1, "Errors on filesystem, "
1633                                   "clearing orphan list.\n");
1634                 es->s_last_orphan = 0;
1635                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1636                 return;
1637         }
1638
1639         if (s_flags & MS_RDONLY) {
1640                 printk(KERN_INFO "EXT4-fs: %s: orphan cleanup on readonly fs\n",
1641                        sb->s_id);
1642                 sb->s_flags &= ~MS_RDONLY;
1643         }
1644 #ifdef CONFIG_QUOTA
1645         /* Needed for iput() to work correctly and not trash data */
1646         sb->s_flags |= MS_ACTIVE;
1647         /* Turn on quotas so that they are updated correctly */
1648         for (i = 0; i < MAXQUOTAS; i++) {
1649                 if (EXT4_SB(sb)->s_qf_names[i]) {
1650                         int ret = ext4_quota_on_mount(sb, i);
1651                         if (ret < 0)
1652                                 printk(KERN_ERR
1653                                         "EXT4-fs: Cannot turn on journaled "
1654                                         "quota: error %d\n", ret);
1655                 }
1656         }
1657 #endif
1658
1659         while (es->s_last_orphan) {
1660                 struct inode *inode;
1661
1662                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
1663                 if (IS_ERR(inode)) {
1664                         es->s_last_orphan = 0;
1665                         break;
1666                 }
1667
1668                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
1669                 DQUOT_INIT(inode);
1670                 if (inode->i_nlink) {
1671                         printk(KERN_DEBUG
1672                                 "%s: truncating inode %lu to %lld bytes\n",
1673                                 __func__, inode->i_ino, inode->i_size);
1674                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
1675                                   inode->i_ino, inode->i_size);
1676                         ext4_truncate(inode);
1677                         nr_truncates++;
1678                 } else {
1679                         printk(KERN_DEBUG
1680                                 "%s: deleting unreferenced inode %lu\n",
1681                                 __func__, inode->i_ino);
1682                         jbd_debug(2, "deleting unreferenced inode %lu\n",
1683                                   inode->i_ino);
1684                         nr_orphans++;
1685                 }
1686                 iput(inode);  /* The delete magic happens here! */
1687         }
1688
1689 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
1690
1691         if (nr_orphans)
1692                 printk(KERN_INFO "EXT4-fs: %s: %d orphan inode%s deleted\n",
1693                        sb->s_id, PLURAL(nr_orphans));
1694         if (nr_truncates)
1695                 printk(KERN_INFO "EXT4-fs: %s: %d truncate%s cleaned up\n",
1696                        sb->s_id, PLURAL(nr_truncates));
1697 #ifdef CONFIG_QUOTA
1698         /* Turn quotas off */
1699         for (i = 0; i < MAXQUOTAS; i++) {
1700                 if (sb_dqopt(sb)->files[i])
1701                         vfs_quota_off(sb, i, 0);
1702         }
1703 #endif
1704         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1705 }
1706 /*
1707  * Maximal extent format file size.
1708  * Resulting logical blkno at s_maxbytes must fit in our on-disk
1709  * extent format containers, within a sector_t, and within i_blocks
1710  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
1711  * so that won't be a limiting factor.
1712  *
1713  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
1714  */
1715 static loff_t ext4_max_size(int blkbits, int has_huge_files)
1716 {
1717         loff_t res;
1718         loff_t upper_limit = MAX_LFS_FILESIZE;
1719
1720         /* small i_blocks in vfs inode? */
1721         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
1722                 /*
1723                  * CONFIG_LBD is not enabled implies the inode
1724                  * i_block represent total blocks in 512 bytes
1725                  * 32 == size of vfs inode i_blocks * 8
1726                  */
1727                 upper_limit = (1LL << 32) - 1;
1728
1729                 /* total blocks in file system block size */
1730                 upper_limit >>= (blkbits - 9);
1731                 upper_limit <<= blkbits;
1732         }
1733
1734         /* 32-bit extent-start container, ee_block */
1735         res = 1LL << 32;
1736         res <<= blkbits;
1737         res -= 1;
1738
1739         /* Sanity check against vm- & vfs- imposed limits */
1740         if (res > upper_limit)
1741                 res = upper_limit;
1742
1743         return res;
1744 }
1745
1746 /*
1747  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
1748  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
1749  * We need to be 1 filesystem block less than the 2^48 sector limit.
1750  */
1751 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
1752 {
1753         loff_t res = EXT4_NDIR_BLOCKS;
1754         int meta_blocks;
1755         loff_t upper_limit;
1756         /* This is calculated to be the largest file size for a
1757          * dense, bitmapped file such that the total number of
1758          * sectors in the file, including data and all indirect blocks,
1759          * does not exceed 2^48 -1
1760          * __u32 i_blocks_lo and _u16 i_blocks_high representing the
1761          * total number of  512 bytes blocks of the file
1762          */
1763
1764         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
1765                 /*
1766                  * !has_huge_files or CONFIG_LBD is not enabled
1767                  * implies the inode i_block represent total blocks in
1768                  * 512 bytes 32 == size of vfs inode i_blocks * 8
1769                  */
1770                 upper_limit = (1LL << 32) - 1;
1771
1772                 /* total blocks in file system block size */
1773                 upper_limit >>= (bits - 9);
1774
1775         } else {
1776                 /*
1777                  * We use 48 bit ext4_inode i_blocks
1778                  * With EXT4_HUGE_FILE_FL set the i_blocks
1779                  * represent total number of blocks in
1780                  * file system block size
1781                  */
1782                 upper_limit = (1LL << 48) - 1;
1783
1784         }
1785
1786         /* indirect blocks */
1787         meta_blocks = 1;
1788         /* double indirect blocks */
1789         meta_blocks += 1 + (1LL << (bits-2));
1790         /* tripple indirect blocks */
1791         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
1792
1793         upper_limit -= meta_blocks;
1794         upper_limit <<= bits;
1795
1796         res += 1LL << (bits-2);
1797         res += 1LL << (2*(bits-2));
1798         res += 1LL << (3*(bits-2));
1799         res <<= bits;
1800         if (res > upper_limit)
1801                 res = upper_limit;
1802
1803         if (res > MAX_LFS_FILESIZE)
1804                 res = MAX_LFS_FILESIZE;
1805
1806         return res;
1807 }
1808
1809 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
1810                                 ext4_fsblk_t logical_sb_block, int nr)
1811 {
1812         struct ext4_sb_info *sbi = EXT4_SB(sb);
1813         ext4_group_t bg, first_meta_bg;
1814         int has_super = 0;
1815
1816         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1817
1818         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
1819             nr < first_meta_bg)
1820                 return logical_sb_block + nr + 1;
1821         bg = sbi->s_desc_per_block * nr;
1822         if (ext4_bg_has_super(sb, bg))
1823                 has_super = 1;
1824         return (has_super + ext4_group_first_block_no(sb, bg));
1825 }
1826
1827 /**
1828  * ext4_get_stripe_size: Get the stripe size.
1829  * @sbi: In memory super block info
1830  *
1831  * If we have specified it via mount option, then
1832  * use the mount option value. If the value specified at mount time is
1833  * greater than the blocks per group use the super block value.
1834  * If the super block value is greater than blocks per group return 0.
1835  * Allocator needs it be less than blocks per group.
1836  *
1837  */
1838 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
1839 {
1840         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
1841         unsigned long stripe_width =
1842                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
1843
1844         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
1845                 return sbi->s_stripe;
1846
1847         if (stripe_width <= sbi->s_blocks_per_group)
1848                 return stripe_width;
1849
1850         if (stride <= sbi->s_blocks_per_group)
1851                 return stride;
1852
1853         return 0;
1854 }
1855
1856 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
1857                                 __releases(kernel_lock)
1858                                 __acquires(kernel_lock)
1859
1860 {
1861         struct buffer_head *bh;
1862         struct ext4_super_block *es = NULL;
1863         struct ext4_sb_info *sbi;
1864         ext4_fsblk_t block;
1865         ext4_fsblk_t sb_block = get_sb_block(&data);
1866         ext4_fsblk_t logical_sb_block;
1867         unsigned long offset = 0;
1868         unsigned int journal_inum = 0;
1869         unsigned long journal_devnum = 0;
1870         unsigned long def_mount_opts;
1871         struct inode *root;
1872         char *cp;
1873         int ret = -EINVAL;
1874         int blocksize;
1875         int db_count;
1876         int i;
1877         int needs_recovery, has_huge_files;
1878         __le32 features;
1879         __u64 blocks_count;
1880         int err;
1881
1882         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1883         if (!sbi)
1884                 return -ENOMEM;
1885         sb->s_fs_info = sbi;
1886         sbi->s_mount_opt = 0;
1887         sbi->s_resuid = EXT4_DEF_RESUID;
1888         sbi->s_resgid = EXT4_DEF_RESGID;
1889         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
1890         sbi->s_sb_block = sb_block;
1891
1892         unlock_kernel();
1893
1894         /* Cleanup superblock name */
1895         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
1896                 *cp = '!';
1897
1898         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
1899         if (!blocksize) {
1900                 printk(KERN_ERR "EXT4-fs: unable to set blocksize\n");
1901                 goto out_fail;
1902         }
1903
1904         /*
1905          * The ext4 superblock will not be buffer aligned for other than 1kB
1906          * block sizes.  We need to calculate the offset from buffer start.
1907          */
1908         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
1909                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
1910                 offset = do_div(logical_sb_block, blocksize);
1911         } else {
1912                 logical_sb_block = sb_block;
1913         }
1914
1915         if (!(bh = sb_bread(sb, logical_sb_block))) {
1916                 printk(KERN_ERR "EXT4-fs: unable to read superblock\n");
1917                 goto out_fail;
1918         }
1919         /*
1920          * Note: s_es must be initialized as soon as possible because
1921          *       some ext4 macro-instructions depend on its value
1922          */
1923         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
1924         sbi->s_es = es;
1925         sb->s_magic = le16_to_cpu(es->s_magic);
1926         if (sb->s_magic != EXT4_SUPER_MAGIC)
1927                 goto cantfind_ext4;
1928
1929         /* Set defaults before we parse the mount options */
1930         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1931         if (def_mount_opts & EXT4_DEFM_DEBUG)
1932                 set_opt(sbi->s_mount_opt, DEBUG);
1933         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
1934                 set_opt(sbi->s_mount_opt, GRPID);
1935         if (def_mount_opts & EXT4_DEFM_UID16)
1936                 set_opt(sbi->s_mount_opt, NO_UID32);
1937 #ifdef CONFIG_EXT4_FS_XATTR
1938         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
1939                 set_opt(sbi->s_mount_opt, XATTR_USER);
1940 #endif
1941 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1942         if (def_mount_opts & EXT4_DEFM_ACL)
1943                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1944 #endif
1945         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
1946                 sbi->s_mount_opt |= EXT4_MOUNT_JOURNAL_DATA;
1947         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
1948                 sbi->s_mount_opt |= EXT4_MOUNT_ORDERED_DATA;
1949         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
1950                 sbi->s_mount_opt |= EXT4_MOUNT_WRITEBACK_DATA;
1951
1952         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
1953                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1954         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
1955                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
1956         else
1957                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1958
1959         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1960         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1961
1962         set_opt(sbi->s_mount_opt, RESERVATION);
1963         set_opt(sbi->s_mount_opt, BARRIER);
1964
1965         /*
1966          * turn on extents feature by default in ext4 filesystem
1967          * only if feature flag already set by mkfs or tune2fs.
1968          * Use -o noextents to turn it off
1969          */
1970         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
1971                 set_opt(sbi->s_mount_opt, EXTENTS);
1972         else
1973                 ext4_warning(sb, __func__,
1974                         "extents feature not enabled on this filesystem, "
1975                         "use tune2fs.\n");
1976
1977         /*
1978          * enable delayed allocation by default
1979          * Use -o nodelalloc to turn it off
1980          */
1981         set_opt(sbi->s_mount_opt, DELALLOC);
1982
1983
1984         if (!parse_options((char *) data, sb, &journal_inum, &journal_devnum,
1985                            NULL, 0))
1986                 goto failed_mount;
1987
1988         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1989                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1990
1991         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
1992             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
1993              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1994              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1995                 printk(KERN_WARNING
1996                        "EXT4-fs warning: feature flags set on rev 0 fs, "
1997                        "running e2fsck is recommended\n");
1998
1999         /*
2000          * Check feature flags regardless of the revision level, since we
2001          * previously didn't change the revision level when setting the flags,
2002          * so there is a chance incompat flags are set on a rev 0 filesystem.
2003          */
2004         features = EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP);
2005         if (features) {
2006                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount because of "
2007                        "unsupported optional features (%x).\n",
2008                        sb->s_id, le32_to_cpu(features));
2009                 goto failed_mount;
2010         }
2011         features = EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP);
2012         if (!(sb->s_flags & MS_RDONLY) && features) {
2013                 printk(KERN_ERR "EXT4-fs: %s: couldn't mount RDWR because of "
2014                        "unsupported optional features (%x).\n",
2015                        sb->s_id, le32_to_cpu(features));
2016                 goto failed_mount;
2017         }
2018         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2019                                     EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
2020         if (has_huge_files) {
2021                 /*
2022                  * Large file size enabled file system can only be
2023                  * mount if kernel is build with CONFIG_LBD
2024                  */
2025                 if (sizeof(root->i_blocks) < sizeof(u64) &&
2026                                 !(sb->s_flags & MS_RDONLY)) {
2027                         printk(KERN_ERR "EXT4-fs: %s: Filesystem with huge "
2028                                         "files cannot be mounted read-write "
2029                                         "without CONFIG_LBD.\n", sb->s_id);
2030                         goto failed_mount;
2031                 }
2032         }
2033         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
2034
2035         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
2036             blocksize > EXT4_MAX_BLOCK_SIZE) {
2037                 printk(KERN_ERR
2038                        "EXT4-fs: Unsupported filesystem blocksize %d on %s.\n",
2039                        blocksize, sb->s_id);
2040                 goto failed_mount;
2041         }
2042
2043         if (sb->s_blocksize != blocksize) {
2044
2045                 /* Validate the filesystem blocksize */
2046                 if (!sb_set_blocksize(sb, blocksize)) {
2047                         printk(KERN_ERR "EXT4-fs: bad block size %d.\n",
2048                                         blocksize);
2049                         goto failed_mount;
2050                 }
2051
2052                 brelse(bh);
2053                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2054                 offset = do_div(logical_sb_block, blocksize);
2055                 bh = sb_bread(sb, logical_sb_block);
2056                 if (!bh) {
2057                         printk(KERN_ERR
2058                                "EXT4-fs: Can't read superblock on 2nd try.\n");
2059                         goto failed_mount;
2060                 }
2061                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2062                 sbi->s_es = es;
2063                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2064                         printk(KERN_ERR
2065                                "EXT4-fs: Magic mismatch, very weird !\n");
2066                         goto failed_mount;
2067                 }
2068         }
2069
2070         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
2071                                                       has_huge_files);
2072         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
2073
2074         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2075                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2076                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2077         } else {
2078                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2079                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2080                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2081                     (!is_power_of_2(sbi->s_inode_size)) ||
2082                     (sbi->s_inode_size > blocksize)) {
2083                         printk(KERN_ERR
2084                                "EXT4-fs: unsupported inode size: %d\n",
2085                                sbi->s_inode_size);
2086                         goto failed_mount;
2087                 }
2088                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2089                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2090         }
2091         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2092         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2093                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2094                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2095                     !is_power_of_2(sbi->s_desc_size)) {
2096                         printk(KERN_ERR
2097                                "EXT4-fs: unsupported descriptor size %lu\n",
2098                                sbi->s_desc_size);
2099                         goto failed_mount;
2100                 }
2101         } else
2102                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2103         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2104         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2105         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2106                 goto cantfind_ext4;
2107         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2108         if (sbi->s_inodes_per_block == 0)
2109                 goto cantfind_ext4;
2110         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2111                                         sbi->s_inodes_per_block;
2112         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2113         sbi->s_sbh = bh;
2114         sbi->s_mount_state = le16_to_cpu(es->s_state);
2115         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2116         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2117         for (i = 0; i < 4; i++)
2118                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2119         sbi->s_def_hash_version = es->s_def_hash_version;
2120
2121         if (sbi->s_blocks_per_group > blocksize * 8) {
2122                 printk(KERN_ERR
2123                        "EXT4-fs: #blocks per group too big: %lu\n",
2124                        sbi->s_blocks_per_group);
2125                 goto failed_mount;
2126         }
2127         if (sbi->s_inodes_per_group > blocksize * 8) {
2128                 printk(KERN_ERR
2129                        "EXT4-fs: #inodes per group too big: %lu\n",
2130                        sbi->s_inodes_per_group);
2131                 goto failed_mount;
2132         }
2133
2134         if (ext4_blocks_count(es) >
2135                     (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
2136                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
2137                         " too large to mount safely\n", sb->s_id);
2138                 if (sizeof(sector_t) < 8)
2139                         printk(KERN_WARNING "EXT4-fs: CONFIG_LBD not "
2140                                         "enabled\n");
2141                 goto failed_mount;
2142         }
2143
2144         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2145                 goto cantfind_ext4;
2146
2147         /* ensure blocks_count calculation below doesn't sign-extend */
2148         if (ext4_blocks_count(es) + EXT4_BLOCKS_PER_GROUP(sb) <
2149             le32_to_cpu(es->s_first_data_block) + 1) {
2150                 printk(KERN_WARNING "EXT4-fs: bad geometry: block count %llu, "
2151                        "first data block %u, blocks per group %lu\n",
2152                         ext4_blocks_count(es),
2153                         le32_to_cpu(es->s_first_data_block),
2154                         EXT4_BLOCKS_PER_GROUP(sb));
2155                 goto failed_mount;
2156         }
2157         blocks_count = (ext4_blocks_count(es) -
2158                         le32_to_cpu(es->s_first_data_block) +
2159                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2160         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2161         sbi->s_groups_count = blocks_count;
2162         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2163                    EXT4_DESC_PER_BLOCK(sb);
2164         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
2165                                     GFP_KERNEL);
2166         if (sbi->s_group_desc == NULL) {
2167                 printk(KERN_ERR "EXT4-fs: not enough memory\n");
2168                 goto failed_mount;
2169         }
2170
2171 #ifdef CONFIG_PROC_FS
2172         if (ext4_proc_root)
2173                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
2174
2175         if (sbi->s_proc)
2176                 proc_create_data("inode_readahead_blks", 0644, sbi->s_proc,
2177                                  &ext4_ui_proc_fops,
2178                                  &sbi->s_inode_readahead_blks);
2179 #endif
2180
2181         bgl_lock_init(&sbi->s_blockgroup_lock);
2182
2183         for (i = 0; i < db_count; i++) {
2184                 block = descriptor_loc(sb, logical_sb_block, i);
2185                 sbi->s_group_desc[i] = sb_bread(sb, block);
2186                 if (!sbi->s_group_desc[i]) {
2187                         printk(KERN_ERR "EXT4-fs: "
2188                                "can't read group descriptor %d\n", i);
2189                         db_count = i;
2190                         goto failed_mount2;
2191                 }
2192         }
2193         if (!ext4_check_descriptors(sb)) {
2194                 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
2195                 goto failed_mount2;
2196         }
2197         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2198                 if (!ext4_fill_flex_info(sb)) {
2199                         printk(KERN_ERR
2200                                "EXT4-fs: unable to initialize "
2201                                "flex_bg meta info!\n");
2202                         goto failed_mount2;
2203                 }
2204
2205         sbi->s_gdb_count = db_count;
2206         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2207         spin_lock_init(&sbi->s_next_gen_lock);
2208
2209         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2210                         ext4_count_free_blocks(sb));
2211         if (!err) {
2212                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2213                                 ext4_count_free_inodes(sb));
2214         }
2215         if (!err) {
2216                 err = percpu_counter_init(&sbi->s_dirs_counter,
2217                                 ext4_count_dirs(sb));
2218         }
2219         if (!err) {
2220                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
2221         }
2222         if (err) {
2223                 printk(KERN_ERR "EXT4-fs: insufficient memory\n");
2224                 goto failed_mount3;
2225         }
2226
2227         sbi->s_stripe = ext4_get_stripe_size(sbi);
2228
2229         /*
2230          * set up enough so that it can read an inode
2231          */
2232         sb->s_op = &ext4_sops;
2233         sb->s_export_op = &ext4_export_ops;
2234         sb->s_xattr = ext4_xattr_handlers;
2235 #ifdef CONFIG_QUOTA
2236         sb->s_qcop = &ext4_qctl_operations;
2237         sb->dq_op = &ext4_quota_operations;
2238 #endif
2239         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2240
2241         sb->s_root = NULL;
2242
2243         needs_recovery = (es->s_last_orphan != 0 ||
2244                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2245                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2246
2247         /*
2248          * The first inode we look at is the journal inode.  Don't try
2249          * root first: it may be modified in the journal!
2250          */
2251         if (!test_opt(sb, NOLOAD) &&
2252             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2253                 if (ext4_load_journal(sb, es, journal_devnum))
2254                         goto failed_mount3;
2255                 if (!(sb->s_flags & MS_RDONLY) &&
2256                     EXT4_SB(sb)->s_journal->j_failed_commit) {
2257                         printk(KERN_CRIT "EXT4-fs error (device %s): "
2258                                "ext4_fill_super: Journal transaction "
2259                                "%u is corrupt\n", sb->s_id,
2260                                EXT4_SB(sb)->s_journal->j_failed_commit);
2261                         if (test_opt(sb, ERRORS_RO)) {
2262                                 printk(KERN_CRIT
2263                                        "Mounting filesystem read-only\n");
2264                                 sb->s_flags |= MS_RDONLY;
2265                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2266                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2267                         }
2268                         if (test_opt(sb, ERRORS_PANIC)) {
2269                                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2270                                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2271                                 ext4_commit_super(sb, es, 1);
2272                                 printk(KERN_CRIT
2273                                        "EXT4-fs (device %s): mount failed\n",
2274                                       sb->s_id);
2275                                 goto failed_mount4;
2276                         }
2277                 }
2278         } else if (journal_inum) {
2279                 if (ext4_create_journal(sb, es, journal_inum))
2280                         goto failed_mount3;
2281         } else {
2282                 if (!silent)
2283                         printk(KERN_ERR
2284                                "ext4: No journal on filesystem on %s\n",
2285                                sb->s_id);
2286                 goto failed_mount3;
2287         }
2288
2289         if (ext4_blocks_count(es) > 0xffffffffULL &&
2290             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2291                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2292                 printk(KERN_ERR "ext4: Failed to set 64-bit journal feature\n");
2293                 goto failed_mount4;
2294         }
2295
2296         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2297                 jbd2_journal_set_features(sbi->s_journal,
2298                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2299                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2300         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2301                 jbd2_journal_set_features(sbi->s_journal,
2302                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2303                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2304                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2305         } else {
2306                 jbd2_journal_clear_features(sbi->s_journal,
2307                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2308                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2309         }
2310
2311         /* We have now updated the journal if required, so we can
2312          * validate the data journaling mode. */
2313         switch (test_opt(sb, DATA_FLAGS)) {
2314         case 0:
2315                 /* No mode set, assume a default based on the journal
2316                  * capabilities: ORDERED_DATA if the journal can
2317                  * cope, else JOURNAL_DATA
2318                  */
2319                 if (jbd2_journal_check_available_features
2320                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2321                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2322                 else
2323                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2324                 break;
2325
2326         case EXT4_MOUNT_ORDERED_DATA:
2327         case EXT4_MOUNT_WRITEBACK_DATA:
2328                 if (!jbd2_journal_check_available_features
2329                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2330                         printk(KERN_ERR "EXT4-fs: Journal does not support "
2331                                "requested data journaling mode\n");
2332                         goto failed_mount4;
2333                 }
2334         default:
2335                 break;
2336         }
2337
2338         if (test_opt(sb, NOBH)) {
2339                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2340                         printk(KERN_WARNING "EXT4-fs: Ignoring nobh option - "
2341                                 "its supported only with writeback mode\n");
2342                         clear_opt(sbi->s_mount_opt, NOBH);
2343                 }
2344         }
2345         /*
2346          * The jbd2_journal_load will have done any necessary log recovery,
2347          * so we can safely mount the rest of the filesystem now.
2348          */
2349
2350         root = ext4_iget(sb, EXT4_ROOT_INO);
2351         if (IS_ERR(root)) {
2352                 printk(KERN_ERR "EXT4-fs: get root inode failed\n");
2353                 ret = PTR_ERR(root);
2354                 goto failed_mount4;
2355         }
2356         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2357                 iput(root);
2358                 printk(KERN_ERR "EXT4-fs: corrupt root inode, run e2fsck\n");
2359                 goto failed_mount4;
2360         }
2361         sb->s_root = d_alloc_root(root);
2362         if (!sb->s_root) {
2363                 printk(KERN_ERR "EXT4-fs: get root dentry failed\n");
2364                 iput(root);
2365                 ret = -ENOMEM;
2366                 goto failed_mount4;
2367         }
2368
2369         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
2370
2371         /* determine the minimum size of new large inodes, if present */
2372         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2373                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2374                                                      EXT4_GOOD_OLD_INODE_SIZE;
2375                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2376                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2377                         if (sbi->s_want_extra_isize <
2378                             le16_to_cpu(es->s_want_extra_isize))
2379                                 sbi->s_want_extra_isize =
2380                                         le16_to_cpu(es->s_want_extra_isize);
2381                         if (sbi->s_want_extra_isize <
2382                             le16_to_cpu(es->s_min_extra_isize))
2383                                 sbi->s_want_extra_isize =
2384                                         le16_to_cpu(es->s_min_extra_isize);
2385                 }
2386         }
2387         /* Check if enough inode space is available */
2388         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2389                                                         sbi->s_inode_size) {
2390                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2391                                                        EXT4_GOOD_OLD_INODE_SIZE;
2392                 printk(KERN_INFO "EXT4-fs: required extra inode space not"
2393                         "available.\n");
2394         }
2395
2396         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
2397                 printk(KERN_WARNING "EXT4-fs: Ignoring delalloc option - "
2398                                 "requested data journaling mode\n");
2399                 clear_opt(sbi->s_mount_opt, DELALLOC);
2400         } else if (test_opt(sb, DELALLOC))
2401                 printk(KERN_INFO "EXT4-fs: delayed allocation enabled\n");
2402
2403         ext4_ext_init(sb);
2404         err = ext4_mb_init(sb, needs_recovery);
2405         if (err) {
2406                 printk(KERN_ERR "EXT4-fs: failed to initalize mballoc (%d)\n",
2407                        err);
2408                 goto failed_mount4;
2409         }
2410
2411         /*
2412          * akpm: core read_super() calls in here with the superblock locked.
2413          * That deadlocks, because orphan cleanup needs to lock the superblock
2414          * in numerous places.  Here we just pop the lock - it's relatively
2415          * harmless, because we are now ready to accept write_super() requests,
2416          * and aviro says that's the only reason for hanging onto the
2417          * superblock lock.
2418          */
2419         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
2420         ext4_orphan_cleanup(sb, es);
2421         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
2422         if (needs_recovery)
2423                 printk(KERN_INFO "EXT4-fs: recovery complete.\n");
2424         ext4_mark_recovery_complete(sb, es);
2425         printk(KERN_INFO "EXT4-fs: mounted filesystem with %s data mode.\n",
2426                test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ? "journal":
2427                test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2428                "writeback");
2429
2430         lock_kernel();
2431         return 0;
2432
2433 cantfind_ext4:
2434         if (!silent)
2435                 printk(KERN_ERR "VFS: Can't find ext4 filesystem on dev %s.\n",
2436                        sb->s_id);
2437         goto failed_mount;
2438
2439 failed_mount4:
2440         jbd2_journal_destroy(sbi->s_journal);
2441         sbi->s_journal = NULL;
2442 failed_mount3:
2443         percpu_counter_destroy(&sbi->s_freeblocks_counter);
2444         percpu_counter_destroy(&sbi->s_freeinodes_counter);
2445         percpu_counter_destroy(&sbi->s_dirs_counter);
2446         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
2447 failed_mount2:
2448         for (i = 0; i < db_count; i++)
2449                 brelse(sbi->s_group_desc[i]);
2450         kfree(sbi->s_group_desc);
2451 failed_mount:
2452         if (sbi->s_proc) {
2453                 remove_proc_entry("inode_readahead_blks", sbi->s_proc);
2454                 remove_proc_entry(sb->s_id, ext4_proc_root);
2455         }
2456 #ifdef CONFIG_QUOTA
2457         for (i = 0; i < MAXQUOTAS; i++)
2458                 kfree(sbi->s_qf_names[i]);
2459 #endif
2460         ext4_blkdev_remove(sbi);
2461         brelse(bh);
2462 out_fail:
2463         sb->s_fs_info = NULL;
2464         kfree(sbi);
2465         lock_kernel();
2466         return ret;
2467 }
2468
2469 /*
2470  * Setup any per-fs journal parameters now.  We'll do this both on
2471  * initial mount, once the journal has been initialised but before we've
2472  * done any recovery; and again on any subsequent remount.
2473  */
2474 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
2475 {
2476         struct ext4_sb_info *sbi = EXT4_SB(sb);
2477
2478         if (sbi->s_commit_interval)
2479                 journal->j_commit_interval = sbi->s_commit_interval;
2480         /* We could also set up an ext4-specific default for the commit
2481          * interval here, but for now we'll just fall back to the jbd
2482          * default. */
2483
2484         spin_lock(&journal->j_state_lock);
2485         if (test_opt(sb, BARRIER))
2486                 journal->j_flags |= JBD2_BARRIER;
2487         else
2488                 journal->j_flags &= ~JBD2_BARRIER;
2489         if (test_opt(sb, DATA_ERR_ABORT))
2490                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
2491         else
2492                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
2493         spin_unlock(&journal->j_state_lock);
2494 }
2495
2496 static journal_t *ext4_get_journal(struct super_block *sb,
2497                                    unsigned int journal_inum)
2498 {
2499         struct inode *journal_inode;
2500         journal_t *journal;
2501
2502         /* First, test for the existence of a valid inode on disk.  Bad
2503          * things happen if we iget() an unused inode, as the subsequent
2504          * iput() will try to delete it. */
2505
2506         journal_inode = ext4_iget(sb, journal_inum);
2507         if (IS_ERR(journal_inode)) {
2508                 printk(KERN_ERR "EXT4-fs: no journal found.\n");
2509                 return NULL;
2510         }
2511         if (!journal_inode->i_nlink) {
2512                 make_bad_inode(journal_inode);
2513                 iput(journal_inode);
2514                 printk(KERN_ERR "EXT4-fs: journal inode is deleted.\n");
2515                 return NULL;
2516         }
2517
2518         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
2519                   journal_inode, journal_inode->i_size);
2520         if (!S_ISREG(journal_inode->i_mode)) {
2521                 printk(KERN_ERR "EXT4-fs: invalid journal inode.\n");
2522                 iput(journal_inode);
2523                 return NULL;
2524         }
2525
2526         journal = jbd2_journal_init_inode(journal_inode);
2527         if (!journal) {
2528                 printk(KERN_ERR "EXT4-fs: Could not load journal inode\n");
2529                 iput(journal_inode);
2530                 return NULL;
2531         }
2532         journal->j_private = sb;
2533         ext4_init_journal_params(sb, journal);
2534         return journal;
2535 }
2536
2537 static journal_t *ext4_get_dev_journal(struct super_block *sb,
2538                                        dev_t j_dev)
2539 {
2540         struct buffer_head *bh;
2541         journal_t *journal;
2542         ext4_fsblk_t start;
2543         ext4_fsblk_t len;
2544         int hblock, blocksize;
2545         ext4_fsblk_t sb_block;
2546         unsigned long offset;
2547         struct ext4_super_block *es;
2548         struct block_device *bdev;
2549
2550         bdev = ext4_blkdev_get(j_dev);
2551         if (bdev == NULL)
2552                 return NULL;
2553
2554         if (bd_claim(bdev, sb)) {
2555                 printk(KERN_ERR
2556                         "EXT4: failed to claim external journal device.\n");
2557                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
2558                 return NULL;
2559         }
2560
2561         blocksize = sb->s_blocksize;
2562         hblock = bdev_hardsect_size(bdev);
2563         if (blocksize < hblock) {
2564                 printk(KERN_ERR
2565                         "EXT4-fs: blocksize too small for journal device.\n");
2566                 goto out_bdev;
2567         }
2568
2569         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
2570         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
2571         set_blocksize(bdev, blocksize);
2572         if (!(bh = __bread(bdev, sb_block, blocksize))) {
2573                 printk(KERN_ERR "EXT4-fs: couldn't read superblock of "
2574                        "external journal\n");
2575                 goto out_bdev;
2576         }
2577
2578         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2579         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
2580             !(le32_to_cpu(es->s_feature_incompat) &
2581               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
2582                 printk(KERN_ERR "EXT4-fs: external journal has "
2583                                         "bad superblock\n");
2584                 brelse(bh);
2585                 goto out_bdev;
2586         }
2587
2588         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
2589                 printk(KERN_ERR "EXT4-fs: journal UUID does not match\n");
2590                 brelse(bh);
2591                 goto out_bdev;
2592         }
2593
2594         len = ext4_blocks_count(es);
2595         start = sb_block + 1;
2596         brelse(bh);     /* we're done with the superblock */
2597
2598         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
2599                                         start, len, blocksize);
2600         if (!journal) {
2601                 printk(KERN_ERR "EXT4-fs: failed to create device journal\n");
2602                 goto out_bdev;
2603         }
2604         journal->j_private = sb;
2605         ll_rw_block(READ, 1, &journal->j_sb_buffer);
2606         wait_on_buffer(journal->j_sb_buffer);
2607         if (!buffer_uptodate(journal->j_sb_buffer)) {
2608                 printk(KERN_ERR "EXT4-fs: I/O error on journal device\n");
2609                 goto out_journal;
2610         }
2611         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
2612                 printk(KERN_ERR "EXT4-fs: External journal has more than one "
2613                                         "user (unsupported) - %d\n",
2614                         be32_to_cpu(journal->j_superblock->s_nr_users));
2615                 goto out_journal;
2616         }
2617         EXT4_SB(sb)->journal_bdev = bdev;
2618         ext4_init_journal_params(sb, journal);
2619         return journal;
2620 out_journal:
2621         jbd2_journal_destroy(journal);
2622 out_bdev:
2623         ext4_blkdev_put(bdev);
2624         return NULL;
2625 }
2626
2627 static int ext4_load_journal(struct super_block *sb,
2628                              struct ext4_super_block *es,
2629                              unsigned long journal_devnum)
2630 {
2631         journal_t *journal;
2632         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
2633         dev_t journal_dev;
2634         int err = 0;
2635         int really_read_only;
2636
2637         if (journal_devnum &&
2638             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2639                 printk(KERN_INFO "EXT4-fs: external journal device major/minor "
2640                         "numbers have changed\n");
2641                 journal_dev = new_decode_dev(journal_devnum);
2642         } else
2643                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
2644
2645         really_read_only = bdev_read_only(sb->s_bdev);
2646
2647         /*
2648          * Are we loading a blank journal or performing recovery after a
2649          * crash?  For recovery, we need to check in advance whether we
2650          * can get read-write access to the device.
2651          */
2652
2653         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2654                 if (sb->s_flags & MS_RDONLY) {
2655                         printk(KERN_INFO "EXT4-fs: INFO: recovery "
2656                                         "required on readonly filesystem.\n");
2657                         if (really_read_only) {
2658                                 printk(KERN_ERR "EXT4-fs: write access "
2659                                         "unavailable, cannot proceed.\n");
2660                                 return -EROFS;
2661                         }
2662                         printk(KERN_INFO "EXT4-fs: write access will "
2663                                "be enabled during recovery.\n");
2664                 }
2665         }
2666
2667         if (journal_inum && journal_dev) {
2668                 printk(KERN_ERR "EXT4-fs: filesystem has both journal "
2669                        "and inode journals!\n");
2670                 return -EINVAL;
2671         }
2672
2673         if (journal_inum) {
2674                 if (!(journal = ext4_get_journal(sb, journal_inum)))
2675                         return -EINVAL;
2676         } else {
2677                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
2678                         return -EINVAL;
2679         }
2680
2681         if (journal->j_flags & JBD2_BARRIER)
2682                 printk(KERN_INFO "EXT4-fs: barriers enabled\n");
2683         else
2684                 printk(KERN_INFO "EXT4-fs: barriers disabled\n");
2685
2686         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
2687                 err = jbd2_journal_update_format(journal);
2688                 if (err)  {
2689                         printk(KERN_ERR "EXT4-fs: error updating journal.\n");
2690                         jbd2_journal_destroy(journal);
2691                         return err;
2692                 }
2693         }
2694
2695         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
2696                 err = jbd2_journal_wipe(journal, !really_read_only);
2697         if (!err)
2698                 err = jbd2_journal_load(journal);
2699
2700         if (err) {
2701                 printk(KERN_ERR "EXT4-fs: error loading journal.\n");
2702                 jbd2_journal_destroy(journal);
2703                 return err;
2704         }
2705
2706         EXT4_SB(sb)->s_journal = journal;
2707         ext4_clear_journal_err(sb, es);
2708
2709         if (journal_devnum &&
2710             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
2711                 es->s_journal_dev = cpu_to_le32(journal_devnum);
2712                 sb->s_dirt = 1;
2713
2714                 /* Make sure we flush the recovery flag to disk. */
2715                 ext4_commit_super(sb, es, 1);
2716         }
2717
2718         return 0;
2719 }
2720
2721 static int ext4_create_journal(struct super_block *sb,
2722                                struct ext4_super_block *es,
2723                                unsigned int journal_inum)
2724 {
2725         journal_t *journal;
2726         int err;
2727
2728         if (sb->s_flags & MS_RDONLY) {
2729                 printk(KERN_ERR "EXT4-fs: readonly filesystem when trying to "
2730                                 "create journal.\n");
2731                 return -EROFS;
2732         }
2733
2734         journal = ext4_get_journal(sb, journal_inum);
2735         if (!journal)
2736                 return -EINVAL;
2737
2738         printk(KERN_INFO "EXT4-fs: creating new journal on inode %u\n",
2739                journal_inum);
2740
2741         err = jbd2_journal_create(journal);
2742         if (err) {
2743                 printk(KERN_ERR "EXT4-fs: error creating journal.\n");
2744                 jbd2_journal_destroy(journal);
2745                 return -EIO;
2746         }
2747
2748         EXT4_SB(sb)->s_journal = journal;
2749
2750         ext4_update_dynamic_rev(sb);
2751         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2752         EXT4_SET_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL);
2753
2754         es->s_journal_inum = cpu_to_le32(journal_inum);
2755         sb->s_dirt = 1;
2756
2757         /* Make sure we flush the recovery flag to disk. */
2758         ext4_commit_super(sb, es, 1);
2759
2760         return 0;
2761 }
2762
2763 static void ext4_commit_super(struct super_block *sb,
2764                               struct ext4_super_block *es, int sync)
2765 {
2766         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
2767
2768         if (!sbh)
2769                 return;
2770         if (buffer_write_io_error(sbh)) {
2771                 /*
2772                  * Oh, dear.  A previous attempt to write the
2773                  * superblock failed.  This could happen because the
2774                  * USB device was yanked out.  Or it could happen to
2775                  * be a transient write error and maybe the block will
2776                  * be remapped.  Nothing we can do but to retry the
2777                  * write and hope for the best.
2778                  */
2779                 printk(KERN_ERR "ext4: previous I/O error to "
2780                        "superblock detected for %s.\n", sb->s_id);
2781                 clear_buffer_write_io_error(sbh);
2782                 set_buffer_uptodate(sbh);
2783         }
2784         es->s_wtime = cpu_to_le32(get_seconds());
2785         ext4_free_blocks_count_set(es, ext4_count_free_blocks(sb));
2786         es->s_free_inodes_count = cpu_to_le32(ext4_count_free_inodes(sb));
2787         BUFFER_TRACE(sbh, "marking dirty");
2788         mark_buffer_dirty(sbh);
2789         if (sync) {
2790                 sync_dirty_buffer(sbh);
2791                 if (buffer_write_io_error(sbh)) {
2792                         printk(KERN_ERR "ext4: I/O error while writing "
2793                                "superblock for %s.\n", sb->s_id);
2794                         clear_buffer_write_io_error(sbh);
2795                         set_buffer_uptodate(sbh);
2796                 }
2797         }
2798 }
2799
2800
2801 /*
2802  * Have we just finished recovery?  If so, and if we are mounting (or
2803  * remounting) the filesystem readonly, then we will end up with a
2804  * consistent fs on disk.  Record that fact.
2805  */
2806 static void ext4_mark_recovery_complete(struct super_block *sb,
2807                                         struct ext4_super_block *es)
2808 {
2809         journal_t *journal = EXT4_SB(sb)->s_journal;
2810
2811         jbd2_journal_lock_updates(journal);
2812         if (jbd2_journal_flush(journal) < 0)
2813                 goto out;
2814
2815         lock_super(sb);
2816         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
2817             sb->s_flags & MS_RDONLY) {
2818                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2819                 sb->s_dirt = 0;
2820                 ext4_commit_super(sb, es, 1);
2821         }
2822         unlock_super(sb);
2823
2824 out:
2825         jbd2_journal_unlock_updates(journal);
2826 }
2827
2828 /*
2829  * If we are mounting (or read-write remounting) a filesystem whose journal
2830  * has recorded an error from a previous lifetime, move that error to the
2831  * main filesystem now.
2832  */
2833 static void ext4_clear_journal_err(struct super_block *sb,
2834                                    struct ext4_super_block *es)
2835 {
2836         journal_t *journal;
2837         int j_errno;
2838         const char *errstr;
2839
2840         journal = EXT4_SB(sb)->s_journal;
2841
2842         /*
2843          * Now check for any error status which may have been recorded in the
2844          * journal by a prior ext4_error() or ext4_abort()
2845          */
2846
2847         j_errno = jbd2_journal_errno(journal);
2848         if (j_errno) {
2849                 char nbuf[16];
2850
2851                 errstr = ext4_decode_error(sb, j_errno, nbuf);
2852                 ext4_warning(sb, __func__, "Filesystem error recorded "
2853                              "from previous mount: %s", errstr);
2854                 ext4_warning(sb, __func__, "Marking fs in need of "
2855                              "filesystem check.");
2856
2857                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
2858                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
2859                 ext4_commit_super(sb, es, 1);
2860
2861                 jbd2_journal_clear_err(journal);
2862         }
2863 }
2864
2865 /*
2866  * Force the running and committing transactions to commit,
2867  * and wait on the commit.
2868  */
2869 int ext4_force_commit(struct super_block *sb)
2870 {
2871         journal_t *journal;
2872         int ret;
2873
2874         if (sb->s_flags & MS_RDONLY)
2875                 return 0;
2876
2877         journal = EXT4_SB(sb)->s_journal;
2878         sb->s_dirt = 0;
2879         ret = ext4_journal_force_commit(journal);
2880         return ret;
2881 }
2882
2883 /*
2884  * Ext4 always journals updates to the superblock itself, so we don't
2885  * have to propagate any other updates to the superblock on disk at this
2886  * point.  (We can probably nuke this function altogether, and remove
2887  * any mention to sb->s_dirt in all of fs/ext4; eventual cleanup...)
2888  */
2889 static void ext4_write_super(struct super_block *sb)
2890 {
2891         if (mutex_trylock(&sb->s_lock) != 0)
2892                 BUG();
2893         sb->s_dirt = 0;
2894 }
2895
2896 static int ext4_sync_fs(struct super_block *sb, int wait)
2897 {
2898         int ret = 0;
2899
2900         trace_mark(ext4_sync_fs, "dev %s wait %d", sb->s_id, wait);
2901         sb->s_dirt = 0;
2902         if (wait)
2903                 ret = ext4_force_commit(sb);
2904         else
2905                 jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, NULL);
2906         return ret;
2907 }
2908
2909 /*
2910  * LVM calls this function before a (read-only) snapshot is created.  This
2911  * gives us a chance to flush the journal completely and mark the fs clean.
2912  */
2913 static void ext4_write_super_lockfs(struct super_block *sb)
2914 {
2915         sb->s_dirt = 0;
2916
2917         if (!(sb->s_flags & MS_RDONLY)) {
2918                 journal_t *journal = EXT4_SB(sb)->s_journal;
2919
2920                 /* Now we set up the journal barrier. */
2921                 jbd2_journal_lock_updates(journal);
2922
2923                 /*
2924                  * We don't want to clear needs_recovery flag when we failed
2925                  * to flush the journal.
2926                  */
2927                 if (jbd2_journal_flush(journal) < 0)
2928                         return;
2929
2930                 /* Journal blocked and flushed, clear needs_recovery flag. */
2931                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2932                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2933         }
2934 }
2935
2936 /*
2937  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2938  * flag here, even though the filesystem is not technically dirty yet.
2939  */
2940 static void ext4_unlockfs(struct super_block *sb)
2941 {
2942         if (!(sb->s_flags & MS_RDONLY)) {
2943                 lock_super(sb);
2944                 /* Reser the needs_recovery flag before the fs is unlocked. */
2945                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
2946                 ext4_commit_super(sb, EXT4_SB(sb)->s_es, 1);
2947                 unlock_super(sb);
2948                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
2949         }
2950 }
2951
2952 static int ext4_remount(struct super_block *sb, int *flags, char *data)
2953 {
2954         struct ext4_super_block *es;
2955         struct ext4_sb_info *sbi = EXT4_SB(sb);
2956         ext4_fsblk_t n_blocks_count = 0;
2957         unsigned long old_sb_flags;
2958         struct ext4_mount_options old_opts;
2959         ext4_group_t g;
2960         int err;
2961 #ifdef CONFIG_QUOTA
2962         int i;
2963 #endif
2964
2965         /* Store the original options */
2966         old_sb_flags = sb->s_flags;
2967         old_opts.s_mount_opt = sbi->s_mount_opt;
2968         old_opts.s_resuid = sbi->s_resuid;
2969         old_opts.s_resgid = sbi->s_resgid;
2970         old_opts.s_commit_interval = sbi->s_commit_interval;
2971 #ifdef CONFIG_QUOTA
2972         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
2973         for (i = 0; i < MAXQUOTAS; i++)
2974                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
2975 #endif
2976
2977         /*
2978          * Allow the "check" option to be passed as a remount option.
2979          */
2980         if (!parse_options(data, sb, NULL, NULL, &n_blocks_count, 1)) {
2981                 err = -EINVAL;
2982                 goto restore_opts;
2983         }
2984
2985         if (sbi->s_mount_opt & EXT4_MOUNT_ABORT)
2986                 ext4_abort(sb, __func__, "Abort forced by user");
2987
2988         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2989                 ((sbi->s_mount_opt & EXT4_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2990
2991         es = sbi->s_es;
2992
2993         ext4_init_journal_params(sb, sbi->s_journal);
2994
2995         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2996                 n_blocks_count > ext4_blocks_count(es)) {
2997                 if (sbi->s_mount_opt & EXT4_MOUNT_ABORT) {
2998                         err = -EROFS;
2999                         goto restore_opts;
3000                 }
3001
3002                 if (*flags & MS_RDONLY) {
3003                         /*
3004                          * First of all, the unconditional stuff we have to do
3005                          * to disable replay of the journal when we next remount
3006                          */
3007                         sb->s_flags |= MS_RDONLY;
3008
3009                         /*
3010                          * OK, test if we are remounting a valid rw partition
3011                          * readonly, and if so set the rdonly flag and then
3012                          * mark the partition as valid again.
3013                          */
3014                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
3015                             (sbi->s_mount_state & EXT4_VALID_FS))
3016                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
3017
3018                         /*
3019                          * We have to unlock super so that we can wait for
3020                          * transactions.
3021                          */
3022                         unlock_super(sb);
3023                         ext4_mark_recovery_complete(sb, es);
3024                         lock_super(sb);
3025                 } else {
3026                         __le32 ret;
3027                         if ((ret = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3028                                         ~EXT4_FEATURE_RO_COMPAT_SUPP))) {
3029                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3030                                        "remount RDWR because of unsupported "
3031                                        "optional features (%x).\n",
3032                                        sb->s_id, le32_to_cpu(ret));
3033                                 err = -EROFS;
3034                                 goto restore_opts;
3035                         }
3036
3037                         /*
3038                          * Make sure the group descriptor checksums
3039                          * are sane.  If they aren't, refuse to
3040                          * remount r/w.
3041                          */
3042                         for (g = 0; g < sbi->s_groups_count; g++) {
3043                                 struct ext4_group_desc *gdp =
3044                                         ext4_get_group_desc(sb, g, NULL);
3045
3046                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
3047                                         printk(KERN_ERR
3048                "EXT4-fs: ext4_remount: "
3049                 "Checksum for group %lu failed (%u!=%u)\n",
3050                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
3051                                                le16_to_cpu(gdp->bg_checksum));
3052                                         err = -EINVAL;
3053                                         goto restore_opts;
3054                                 }
3055                         }
3056
3057                         /*
3058                          * If we have an unprocessed orphan list hanging
3059                          * around from a previously readonly bdev mount,
3060                          * require a full umount/remount for now.
3061                          */
3062                         if (es->s_last_orphan) {
3063                                 printk(KERN_WARNING "EXT4-fs: %s: couldn't "
3064                                        "remount RDWR because of unprocessed "
3065                                        "orphan inode list.  Please "
3066                                        "umount/remount instead.\n",
3067                                        sb->s_id);
3068                                 err = -EINVAL;
3069                                 goto restore_opts;
3070                         }
3071
3072                         /*
3073                          * Mounting a RDONLY partition read-write, so reread
3074                          * and store the current valid flag.  (It may have
3075                          * been changed by e2fsck since we originally mounted
3076                          * the partition.)
3077                          */
3078                         ext4_clear_journal_err(sb, es);
3079                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3080                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3081                                 goto restore_opts;
3082                         if (!ext4_setup_super(sb, es, 0))
3083                                 sb->s_flags &= ~MS_RDONLY;
3084                 }
3085         }
3086 #ifdef CONFIG_QUOTA
3087         /* Release old quota file names */
3088         for (i = 0; i < MAXQUOTAS; i++)
3089                 if (old_opts.s_qf_names[i] &&
3090                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3091                         kfree(old_opts.s_qf_names[i]);
3092 #endif
3093         return 0;
3094 restore_opts:
3095         sb->s_flags = old_sb_flags;
3096         sbi->s_mount_opt = old_opts.s_mount_opt;
3097         sbi->s_resuid = old_opts.s_resuid;
3098         sbi->s_resgid = old_opts.s_resgid;
3099         sbi->s_commit_interval = old_opts.s_commit_interval;
3100 #ifdef CONFIG_QUOTA
3101         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3102         for (i = 0; i < MAXQUOTAS; i++) {
3103                 if (sbi->s_qf_names[i] &&
3104                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3105                         kfree(sbi->s_qf_names[i]);
3106                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3107         }
3108 #endif
3109         return err;
3110 }
3111
3112 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
3113 {
3114         struct super_block *sb = dentry->d_sb;
3115         struct ext4_sb_info *sbi = EXT4_SB(sb);
3116         struct ext4_super_block *es = sbi->s_es;
3117         u64 fsid;
3118
3119         if (test_opt(sb, MINIX_DF)) {
3120                 sbi->s_overhead_last = 0;
3121         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3122                 ext4_group_t ngroups = sbi->s_groups_count, i;
3123                 ext4_fsblk_t overhead = 0;
3124                 smp_rmb();
3125
3126                 /*
3127                  * Compute the overhead (FS structures).  This is constant
3128                  * for a given filesystem unless the number of block groups
3129                  * changes so we cache the previous value until it does.
3130                  */
3131
3132                 /*
3133                  * All of the blocks before first_data_block are
3134                  * overhead
3135                  */
3136                 overhead = le32_to_cpu(es->s_first_data_block);
3137
3138                 /*
3139                  * Add the overhead attributed to the superblock and
3140                  * block group descriptors.  If the sparse superblocks
3141                  * feature is turned on, then not all groups have this.
3142                  */
3143                 for (i = 0; i < ngroups; i++) {
3144                         overhead += ext4_bg_has_super(sb, i) +
3145                                 ext4_bg_num_gdb(sb, i);
3146                         cond_resched();
3147                 }
3148
3149                 /*
3150                  * Every block group has an inode bitmap, a block
3151                  * bitmap, and an inode table.
3152                  */
3153                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3154                 sbi->s_overhead_last = overhead;
3155                 smp_wmb();
3156                 sbi->s_blocks_last = ext4_blocks_count(es);
3157         }
3158
3159         buf->f_type = EXT4_SUPER_MAGIC;
3160         buf->f_bsize = sb->s_blocksize;
3161         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3162         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
3163                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
3164         ext4_free_blocks_count_set(es, buf->f_bfree);
3165         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3166         if (buf->f_bfree < ext4_r_blocks_count(es))
3167                 buf->f_bavail = 0;
3168         buf->f_files = le32_to_cpu(es->s_inodes_count);
3169         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3170         es->s_free_inodes_count = cpu_to_le32(buf->f_ffree);
3171         buf->f_namelen = EXT4_NAME_LEN;
3172         fsid = le64_to_cpup((void *)es->s_uuid) ^
3173                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3174         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3175         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3176         return 0;
3177 }
3178
3179 /* Helper function for writing quotas on sync - we need to start transaction before quota file
3180  * is locked for write. Otherwise the are possible deadlocks:
3181  * Process 1                         Process 2
3182  * ext4_create()                     quota_sync()
3183  *   jbd2_journal_start()                   write_dquot()
3184  *   DQUOT_INIT()                        down(dqio_mutex)
3185  *     down(dqio_mutex)                    jbd2_journal_start()
3186  *
3187  */
3188
3189 #ifdef CONFIG_QUOTA
3190
3191 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3192 {
3193         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3194 }
3195
3196 static int ext4_dquot_initialize(struct inode *inode, int type)
3197 {
3198         handle_t *handle;
3199         int ret, err;
3200
3201         /* We may create quota structure so we need to reserve enough blocks */
3202         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_INIT_BLOCKS(inode->i_sb));
3203         if (IS_ERR(handle))
3204                 return PTR_ERR(handle);
3205         ret = dquot_initialize(inode, type);
3206         err = ext4_journal_stop(handle);
3207         if (!ret)
3208                 ret = err;
3209         return ret;
3210 }
3211
3212 static int ext4_dquot_drop(struct inode *inode)
3213 {
3214         handle_t *handle;
3215         int ret, err;
3216
3217         /* We may delete quota structure so we need to reserve enough blocks */
3218         handle = ext4_journal_start(inode, 2*EXT4_QUOTA_DEL_BLOCKS(inode->i_sb));
3219         if (IS_ERR(handle)) {
3220                 /*
3221                  * We call dquot_drop() anyway to at least release references
3222                  * to quota structures so that umount does not hang.
3223                  */
3224                 dquot_drop(inode);
3225                 return PTR_ERR(handle);
3226         }
3227         ret = dquot_drop(inode);
3228         err = ext4_journal_stop(handle);
3229         if (!ret)
3230                 ret = err;
3231         return ret;
3232 }
3233
3234 static int ext4_write_dquot(struct dquot *dquot)
3235 {
3236         int ret, err;
3237         handle_t *handle;
3238         struct inode *inode;
3239
3240         inode = dquot_to_inode(dquot);
3241         handle = ext4_journal_start(inode,
3242                                         EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3243         if (IS_ERR(handle))
3244                 return PTR_ERR(handle);
3245         ret = dquot_commit(dquot);
3246         err = ext4_journal_stop(handle);
3247         if (!ret)
3248                 ret = err;
3249         return ret;
3250 }
3251
3252 static int ext4_acquire_dquot(struct dquot *dquot)
3253 {
3254         int ret, err;
3255         handle_t *handle;
3256
3257         handle = ext4_journal_start(dquot_to_inode(dquot),
3258                                         EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3259         if (IS_ERR(handle))
3260                 return PTR_ERR(handle);
3261         ret = dquot_acquire(dquot);
3262         err = ext4_journal_stop(handle);
3263         if (!ret)
3264                 ret = err;
3265         return ret;
3266 }
3267
3268 static int ext4_release_dquot(struct dquot *dquot)
3269 {
3270         int ret, err;
3271         handle_t *handle;
3272
3273         handle = ext4_journal_start(dquot_to_inode(dquot),
3274                                         EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3275         if (IS_ERR(handle)) {
3276                 /* Release dquot anyway to avoid endless cycle in dqput() */
3277                 dquot_release(dquot);
3278                 return PTR_ERR(handle);
3279         }
3280         ret = dquot_release(dquot);
3281         err = ext4_journal_stop(handle);
3282         if (!ret)
3283                 ret = err;
3284         return ret;
3285 }
3286
3287 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3288 {
3289         /* Are we journaling quotas? */
3290         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3291             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3292                 dquot_mark_dquot_dirty(dquot);
3293                 return ext4_write_dquot(dquot);
3294         } else {
3295                 return dquot_mark_dquot_dirty(dquot);
3296         }
3297 }
3298
3299 static int ext4_write_info(struct super_block *sb, int type)
3300 {
3301         int ret, err;
3302         handle_t *handle;
3303
3304         /* Data block + inode block */
3305         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3306         if (IS_ERR(handle))
3307                 return PTR_ERR(handle);
3308         ret = dquot_commit_info(sb, type);
3309         err = ext4_journal_stop(handle);
3310         if (!ret)
3311                 ret = err;
3312         return ret;
3313 }
3314
3315 /*
3316  * Turn on quotas during mount time - we need to find
3317  * the quota file and such...
3318  */
3319 static int ext4_quota_on_mount(struct super_block *sb, int type)
3320 {
3321         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3322                         EXT4_SB(sb)->s_jquota_fmt, type);
3323 }
3324
3325 /*
3326  * Standard function to be called on quota_on
3327  */
3328 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3329                          char *name, int remount)
3330 {
3331         int err;
3332         struct path path;
3333
3334         if (!test_opt(sb, QUOTA))
3335                 return -EINVAL;
3336         /* When remounting, no checks are needed and in fact, name is NULL */
3337         if (remount)
3338                 return vfs_quota_on(sb, type, format_id, name, remount);
3339
3340         err = kern_path(name, LOOKUP_FOLLOW, &path);
3341         if (err)
3342                 return err;
3343
3344         /* Quotafile not on the same filesystem? */
3345         if (path.mnt->mnt_sb != sb) {
3346                 path_put(&path);
3347                 return -EXDEV;
3348         }
3349         /* Journaling quota? */
3350         if (EXT4_SB(sb)->s_qf_names[type]) {
3351                 /* Quotafile not in fs root? */
3352                 if (path.dentry->d_parent != sb->s_root)
3353                         printk(KERN_WARNING
3354                                 "EXT4-fs: Quota file not on filesystem root. "
3355                                 "Journaled quota will not work.\n");
3356         }
3357
3358         /*
3359          * When we journal data on quota file, we have to flush journal to see
3360          * all updates to the file when we bypass pagecache...
3361          */
3362         if (ext4_should_journal_data(path.dentry->d_inode)) {
3363                 /*
3364                  * We don't need to lock updates but journal_flush() could
3365                  * otherwise be livelocked...
3366                  */
3367                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3368                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3369                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3370                 if (err) {
3371                         path_put(&path);
3372                         return err;
3373                 }
3374         }
3375
3376         err = vfs_quota_on_path(sb, type, format_id, &path);
3377         path_put(&path);
3378         return err;
3379 }
3380
3381 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3382  * acquiring the locks... As quota files are never truncated and quota code
3383  * itself serializes the operations (and noone else should touch the files)
3384  * we don't have to be afraid of races */
3385 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3386                                size_t len, loff_t off)
3387 {
3388         struct inode *inode = sb_dqopt(sb)->files[type];
3389         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3390         int err = 0;
3391         int offset = off & (sb->s_blocksize - 1);
3392         int tocopy;
3393         size_t toread;
3394         struct buffer_head *bh;
3395         loff_t i_size = i_size_read(inode);
3396
3397         if (off > i_size)
3398                 return 0;
3399         if (off+len > i_size)
3400                 len = i_size-off;
3401         toread = len;
3402         while (toread > 0) {
3403                 tocopy = sb->s_blocksize - offset < toread ?
3404                                 sb->s_blocksize - offset : toread;
3405                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3406                 if (err)
3407                         return err;
3408                 if (!bh)        /* A hole? */
3409                         memset(data, 0, tocopy);
3410                 else
3411                         memcpy(data, bh->b_data+offset, tocopy);
3412                 brelse(bh);
3413                 offset = 0;
3414                 toread -= tocopy;
3415                 data += tocopy;
3416                 blk++;
3417         }
3418         return len;
3419 }
3420
3421 /* Write to quotafile (we know the transaction is already started and has
3422  * enough credits) */
3423 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3424                                 const char *data, size_t len, loff_t off)
3425 {
3426         struct inode *inode = sb_dqopt(sb)->files[type];
3427         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3428         int err = 0;
3429         int offset = off & (sb->s_blocksize - 1);
3430         int tocopy;
3431         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
3432         size_t towrite = len;
3433         struct buffer_head *bh;
3434         handle_t *handle = journal_current_handle();
3435
3436         if (!handle) {
3437                 printk(KERN_WARNING "EXT4-fs: Quota write (off=%llu, len=%llu)"
3438                         " cancelled because transaction is not started.\n",
3439                         (unsigned long long)off, (unsigned long long)len);
3440                 return -EIO;
3441         }
3442         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
3443         while (towrite > 0) {
3444                 tocopy = sb->s_blocksize - offset < towrite ?
3445                                 sb->s_blocksize - offset : towrite;
3446                 bh = ext4_bread(handle, inode, blk, 1, &err);
3447                 if (!bh)
3448                         goto out;
3449                 if (journal_quota) {
3450                         err = ext4_journal_get_write_access(handle, bh);
3451                         if (err) {
3452                                 brelse(bh);
3453                                 goto out;
3454                         }
3455                 }
3456                 lock_buffer(bh);
3457                 memcpy(bh->b_data+offset, data, tocopy);
3458                 flush_dcache_page(bh->b_page);
3459                 unlock_buffer(bh);
3460                 if (journal_quota)
3461                         err = ext4_journal_dirty_metadata(handle, bh);
3462                 else {
3463                         /* Always do at least ordered writes for quotas */
3464                         err = ext4_jbd2_file_inode(handle, inode);
3465                         mark_buffer_dirty(bh);
3466                 }
3467                 brelse(bh);
3468                 if (err)
3469                         goto out;
3470                 offset = 0;
3471                 towrite -= tocopy;
3472                 data += tocopy;
3473                 blk++;
3474         }
3475 out:
3476         if (len == towrite) {
3477                 mutex_unlock(&inode->i_mutex);
3478                 return err;
3479         }
3480         if (inode->i_size < off+len-towrite) {
3481                 i_size_write(inode, off+len-towrite);
3482                 EXT4_I(inode)->i_disksize = inode->i_size;
3483         }
3484         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
3485         ext4_mark_inode_dirty(handle, inode);
3486         mutex_unlock(&inode->i_mutex);
3487         return len - towrite;
3488 }
3489
3490 #endif
3491
3492 static int ext4_get_sb(struct file_system_type *fs_type,
3493         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3494 {
3495         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3496 }
3497
3498 #ifdef CONFIG_PROC_FS
3499 static int ext4_ui_proc_show(struct seq_file *m, void *v)
3500 {
3501         unsigned int *p = m->private;
3502
3503         seq_printf(m, "%u\n", *p);
3504         return 0;
3505 }
3506
3507 static int ext4_ui_proc_open(struct inode *inode, struct file *file)
3508 {
3509         return single_open(file, ext4_ui_proc_show, PDE(inode)->data);
3510 }
3511
3512 static ssize_t ext4_ui_proc_write(struct file *file, const char __user *buf,
3513                                size_t cnt, loff_t *ppos)
3514 {
3515         unsigned int *p = PDE(file->f_path.dentry->d_inode)->data;
3516         char str[32];
3517         unsigned long value;
3518
3519         if (cnt >= sizeof(str))
3520                 return -EINVAL;
3521         if (copy_from_user(str, buf, cnt))
3522                 return -EFAULT;
3523         value = simple_strtol(str, NULL, 0);
3524         if (value < 0)
3525                 return -ERANGE;
3526         *p = value;
3527         return cnt;
3528 }
3529
3530 const struct file_operations ext4_ui_proc_fops = {
3531         .owner          = THIS_MODULE,
3532         .open           = ext4_ui_proc_open,
3533         .read           = seq_read,
3534         .llseek         = seq_lseek,
3535         .release        = single_release,
3536         .write          = ext4_ui_proc_write,
3537 };
3538 #endif
3539
3540 static struct file_system_type ext4_fs_type = {
3541         .owner          = THIS_MODULE,
3542         .name           = "ext4",
3543         .get_sb         = ext4_get_sb,
3544         .kill_sb        = kill_block_super,
3545         .fs_flags       = FS_REQUIRES_DEV,
3546 };
3547
3548 #ifdef CONFIG_EXT4DEV_COMPAT
3549 static int ext4dev_get_sb(struct file_system_type *fs_type,
3550         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
3551 {
3552         printk(KERN_WARNING "EXT4-fs: Update your userspace programs "
3553                "to mount using ext4\n");
3554         printk(KERN_WARNING "EXT4-fs: ext4dev backwards compatibility "
3555                "will go away by 2.6.31\n");
3556         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super, mnt);
3557 }
3558
3559 static struct file_system_type ext4dev_fs_type = {
3560         .owner          = THIS_MODULE,
3561         .name           = "ext4dev",
3562         .get_sb         = ext4dev_get_sb,
3563         .kill_sb        = kill_block_super,
3564         .fs_flags       = FS_REQUIRES_DEV,
3565 };
3566 MODULE_ALIAS("ext4dev");
3567 #endif
3568
3569 static int __init init_ext4_fs(void)
3570 {
3571         int err;
3572
3573         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
3574         err = init_ext4_mballoc();
3575         if (err)
3576                 return err;
3577
3578         err = init_ext4_xattr();
3579         if (err)
3580                 goto out2;
3581         err = init_inodecache();
3582         if (err)
3583                 goto out1;
3584         err = register_filesystem(&ext4_fs_type);
3585         if (err)
3586                 goto out;
3587 #ifdef CONFIG_EXT4DEV_COMPAT
3588         err = register_filesystem(&ext4dev_fs_type);
3589         if (err) {
3590                 unregister_filesystem(&ext4_fs_type);
3591                 goto out;
3592         }
3593 #endif
3594         return 0;
3595 out:
3596         destroy_inodecache();
3597 out1:
3598         exit_ext4_xattr();
3599 out2:
3600         exit_ext4_mballoc();
3601         return err;
3602 }
3603
3604 static void __exit exit_ext4_fs(void)
3605 {
3606         unregister_filesystem(&ext4_fs_type);
3607 #ifdef CONFIG_EXT4DEV_COMPAT
3608         unregister_filesystem(&ext4dev_fs_type);
3609 #endif
3610         destroy_inodecache();
3611         exit_ext4_xattr();
3612         exit_ext4_mballoc();
3613         remove_proc_entry("fs/ext4", NULL);
3614 }
3615
3616 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
3617 MODULE_DESCRIPTION("Fourth Extended Filesystem with extents");
3618 MODULE_LICENSE("GPL");
3619 module_init(init_ext4_fs)
3620 module_exit(exit_ext4_fs)