ext4: Fix spelling of CONTIG_FS_EXT3 to CONFIG_FS_EXT3
[linux-2.6.git] / fs / ext4 / super.c
1 /*
2  *  linux/fs/ext4/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/module.h>
20 #include <linux/string.h>
21 #include <linux/fs.h>
22 #include <linux/time.h>
23 #include <linux/vmalloc.h>
24 #include <linux/jbd2.h>
25 #include <linux/slab.h>
26 #include <linux/init.h>
27 #include <linux/blkdev.h>
28 #include <linux/parser.h>
29 #include <linux/smp_lock.h>
30 #include <linux/buffer_head.h>
31 #include <linux/exportfs.h>
32 #include <linux/vfs.h>
33 #include <linux/random.h>
34 #include <linux/mount.h>
35 #include <linux/namei.h>
36 #include <linux/quotaops.h>
37 #include <linux/seq_file.h>
38 #include <linux/proc_fs.h>
39 #include <linux/ctype.h>
40 #include <linux/log2.h>
41 #include <linux/crc16.h>
42 #include <asm/uaccess.h>
43
44 #include "ext4.h"
45 #include "ext4_jbd2.h"
46 #include "xattr.h"
47 #include "acl.h"
48 #include "mballoc.h"
49
50 #define CREATE_TRACE_POINTS
51 #include <trace/events/ext4.h>
52
53 struct proc_dir_entry *ext4_proc_root;
54 static struct kset *ext4_kset;
55
56 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
57                              unsigned long journal_devnum);
58 static int ext4_commit_super(struct super_block *sb, int sync);
59 static void ext4_mark_recovery_complete(struct super_block *sb,
60                                         struct ext4_super_block *es);
61 static void ext4_clear_journal_err(struct super_block *sb,
62                                    struct ext4_super_block *es);
63 static int ext4_sync_fs(struct super_block *sb, int wait);
64 static const char *ext4_decode_error(struct super_block *sb, int errno,
65                                      char nbuf[16]);
66 static int ext4_remount(struct super_block *sb, int *flags, char *data);
67 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
68 static int ext4_unfreeze(struct super_block *sb);
69 static void ext4_write_super(struct super_block *sb);
70 static int ext4_freeze(struct super_block *sb);
71
72
73 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
74                                struct ext4_group_desc *bg)
75 {
76         return le32_to_cpu(bg->bg_block_bitmap_lo) |
77                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
78                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
79 }
80
81 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
82                                struct ext4_group_desc *bg)
83 {
84         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
85                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
86                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
87 }
88
89 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
90                               struct ext4_group_desc *bg)
91 {
92         return le32_to_cpu(bg->bg_inode_table_lo) |
93                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
94                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
95 }
96
97 __u32 ext4_free_blks_count(struct super_block *sb,
98                               struct ext4_group_desc *bg)
99 {
100         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
101                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
102                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
103 }
104
105 __u32 ext4_free_inodes_count(struct super_block *sb,
106                               struct ext4_group_desc *bg)
107 {
108         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
109                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
110                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
111 }
112
113 __u32 ext4_used_dirs_count(struct super_block *sb,
114                               struct ext4_group_desc *bg)
115 {
116         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
117                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
118                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
119 }
120
121 __u32 ext4_itable_unused_count(struct super_block *sb,
122                               struct ext4_group_desc *bg)
123 {
124         return le16_to_cpu(bg->bg_itable_unused_lo) |
125                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
126                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
127 }
128
129 void ext4_block_bitmap_set(struct super_block *sb,
130                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
131 {
132         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
133         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
134                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
135 }
136
137 void ext4_inode_bitmap_set(struct super_block *sb,
138                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
139 {
140         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
141         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
142                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
143 }
144
145 void ext4_inode_table_set(struct super_block *sb,
146                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
147 {
148         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
149         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
150                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
151 }
152
153 void ext4_free_blks_set(struct super_block *sb,
154                           struct ext4_group_desc *bg, __u32 count)
155 {
156         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
157         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
158                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
159 }
160
161 void ext4_free_inodes_set(struct super_block *sb,
162                           struct ext4_group_desc *bg, __u32 count)
163 {
164         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
165         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
166                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
167 }
168
169 void ext4_used_dirs_set(struct super_block *sb,
170                           struct ext4_group_desc *bg, __u32 count)
171 {
172         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
173         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
174                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
175 }
176
177 void ext4_itable_unused_set(struct super_block *sb,
178                           struct ext4_group_desc *bg, __u32 count)
179 {
180         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
181         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
182                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
183 }
184
185
186 /* Just increment the non-pointer handle value */
187 static handle_t *ext4_get_nojournal(void)
188 {
189         handle_t *handle = current->journal_info;
190         unsigned long ref_cnt = (unsigned long)handle;
191
192         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
193
194         ref_cnt++;
195         handle = (handle_t *)ref_cnt;
196
197         current->journal_info = handle;
198         return handle;
199 }
200
201
202 /* Decrement the non-pointer handle value */
203 static void ext4_put_nojournal(handle_t *handle)
204 {
205         unsigned long ref_cnt = (unsigned long)handle;
206
207         BUG_ON(ref_cnt == 0);
208
209         ref_cnt--;
210         handle = (handle_t *)ref_cnt;
211
212         current->journal_info = handle;
213 }
214
215 /*
216  * Wrappers for jbd2_journal_start/end.
217  *
218  * The only special thing we need to do here is to make sure that all
219  * journal_end calls result in the superblock being marked dirty, so
220  * that sync() will call the filesystem's write_super callback if
221  * appropriate.
222  */
223 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
224 {
225         journal_t *journal;
226
227         if (sb->s_flags & MS_RDONLY)
228                 return ERR_PTR(-EROFS);
229
230         /* Special case here: if the journal has aborted behind our
231          * backs (eg. EIO in the commit thread), then we still need to
232          * take the FS itself readonly cleanly. */
233         journal = EXT4_SB(sb)->s_journal;
234         if (journal) {
235                 if (is_journal_aborted(journal)) {
236                         ext4_abort(sb, __func__, "Detected aborted journal");
237                         return ERR_PTR(-EROFS);
238                 }
239                 return jbd2_journal_start(journal, nblocks);
240         }
241         return ext4_get_nojournal();
242 }
243
244 /*
245  * The only special thing we need to do here is to make sure that all
246  * jbd2_journal_stop calls result in the superblock being marked dirty, so
247  * that sync() will call the filesystem's write_super callback if
248  * appropriate.
249  */
250 int __ext4_journal_stop(const char *where, handle_t *handle)
251 {
252         struct super_block *sb;
253         int err;
254         int rc;
255
256         if (!ext4_handle_valid(handle)) {
257                 ext4_put_nojournal(handle);
258                 return 0;
259         }
260         sb = handle->h_transaction->t_journal->j_private;
261         err = handle->h_err;
262         rc = jbd2_journal_stop(handle);
263
264         if (!err)
265                 err = rc;
266         if (err)
267                 __ext4_std_error(sb, where, err);
268         return err;
269 }
270
271 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
272                 struct buffer_head *bh, handle_t *handle, int err)
273 {
274         char nbuf[16];
275         const char *errstr = ext4_decode_error(NULL, err, nbuf);
276
277         BUG_ON(!ext4_handle_valid(handle));
278
279         if (bh)
280                 BUFFER_TRACE(bh, "abort");
281
282         if (!handle->h_err)
283                 handle->h_err = err;
284
285         if (is_handle_aborted(handle))
286                 return;
287
288         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
289                caller, errstr, err_fn);
290
291         jbd2_journal_abort_handle(handle);
292 }
293
294 /* Deal with the reporting of failure conditions on a filesystem such as
295  * inconsistencies detected or read IO failures.
296  *
297  * On ext2, we can store the error state of the filesystem in the
298  * superblock.  That is not possible on ext4, because we may have other
299  * write ordering constraints on the superblock which prevent us from
300  * writing it out straight away; and given that the journal is about to
301  * be aborted, we can't rely on the current, or future, transactions to
302  * write out the superblock safely.
303  *
304  * We'll just use the jbd2_journal_abort() error code to record an error in
305  * the journal instead.  On recovery, the journal will compain about
306  * that error until we've noted it down and cleared it.
307  */
308
309 static void ext4_handle_error(struct super_block *sb)
310 {
311         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
312
313         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
314         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
315
316         if (sb->s_flags & MS_RDONLY)
317                 return;
318
319         if (!test_opt(sb, ERRORS_CONT)) {
320                 journal_t *journal = EXT4_SB(sb)->s_journal;
321
322                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
323                 if (journal)
324                         jbd2_journal_abort(journal, -EIO);
325         }
326         if (test_opt(sb, ERRORS_RO)) {
327                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
328                 sb->s_flags |= MS_RDONLY;
329         }
330         ext4_commit_super(sb, 1);
331         if (test_opt(sb, ERRORS_PANIC))
332                 panic("EXT4-fs (device %s): panic forced after error\n",
333                         sb->s_id);
334 }
335
336 void __ext4_error(struct super_block *sb, const char *function,
337                 const char *fmt, ...)
338 {
339         va_list args;
340
341         va_start(args, fmt);
342         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
343         vprintk(fmt, args);
344         printk("\n");
345         va_end(args);
346
347         ext4_handle_error(sb);
348 }
349
350 void ext4_error_inode(const char *function, struct inode *inode,
351                       const char *fmt, ...)
352 {
353         va_list args;
354
355         va_start(args, fmt);
356         printk(KERN_CRIT "EXT4-fs error (device %s): %s: inode #%lu: (comm %s) ",
357                inode->i_sb->s_id, function, inode->i_ino, current->comm);
358         vprintk(fmt, args);
359         printk("\n");
360         va_end(args);
361
362         ext4_handle_error(inode->i_sb);
363 }
364
365 void ext4_error_file(const char *function, struct file *file,
366                      const char *fmt, ...)
367 {
368         va_list args;
369         struct inode *inode = file->f_dentry->d_inode;
370         char pathname[80], *path;
371
372         va_start(args, fmt);
373         path = d_path(&(file->f_path), pathname, sizeof(pathname));
374         if (!path)
375                 path = "(unknown)";
376         printk(KERN_CRIT
377                "EXT4-fs error (device %s): %s: inode #%lu (comm %s path %s): ",
378                inode->i_sb->s_id, function, inode->i_ino, current->comm, path);
379         vprintk(fmt, args);
380         printk("\n");
381         va_end(args);
382
383         ext4_handle_error(inode->i_sb);
384 }
385
386 static const char *ext4_decode_error(struct super_block *sb, int errno,
387                                      char nbuf[16])
388 {
389         char *errstr = NULL;
390
391         switch (errno) {
392         case -EIO:
393                 errstr = "IO failure";
394                 break;
395         case -ENOMEM:
396                 errstr = "Out of memory";
397                 break;
398         case -EROFS:
399                 if (!sb || (EXT4_SB(sb)->s_journal &&
400                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
401                         errstr = "Journal has aborted";
402                 else
403                         errstr = "Readonly filesystem";
404                 break;
405         default:
406                 /* If the caller passed in an extra buffer for unknown
407                  * errors, textualise them now.  Else we just return
408                  * NULL. */
409                 if (nbuf) {
410                         /* Check for truncated error codes... */
411                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
412                                 errstr = nbuf;
413                 }
414                 break;
415         }
416
417         return errstr;
418 }
419
420 /* __ext4_std_error decodes expected errors from journaling functions
421  * automatically and invokes the appropriate error response.  */
422
423 void __ext4_std_error(struct super_block *sb, const char *function, int errno)
424 {
425         char nbuf[16];
426         const char *errstr;
427
428         /* Special case: if the error is EROFS, and we're not already
429          * inside a transaction, then there's really no point in logging
430          * an error. */
431         if (errno == -EROFS && journal_current_handle() == NULL &&
432             (sb->s_flags & MS_RDONLY))
433                 return;
434
435         errstr = ext4_decode_error(sb, errno, nbuf);
436         printk(KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
437                sb->s_id, function, errstr);
438
439         ext4_handle_error(sb);
440 }
441
442 /*
443  * ext4_abort is a much stronger failure handler than ext4_error.  The
444  * abort function may be used to deal with unrecoverable failures such
445  * as journal IO errors or ENOMEM at a critical moment in log management.
446  *
447  * We unconditionally force the filesystem into an ABORT|READONLY state,
448  * unless the error response on the fs has been set to panic in which
449  * case we take the easy way out and panic immediately.
450  */
451
452 void ext4_abort(struct super_block *sb, const char *function,
453                 const char *fmt, ...)
454 {
455         va_list args;
456
457         va_start(args, fmt);
458         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
459         vprintk(fmt, args);
460         printk("\n");
461         va_end(args);
462
463         if (test_opt(sb, ERRORS_PANIC))
464                 panic("EXT4-fs panic from previous error\n");
465
466         if (sb->s_flags & MS_RDONLY)
467                 return;
468
469         ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
470         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
471         sb->s_flags |= MS_RDONLY;
472         EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
473         if (EXT4_SB(sb)->s_journal)
474                 jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
475 }
476
477 void ext4_msg (struct super_block * sb, const char *prefix,
478                    const char *fmt, ...)
479 {
480         va_list args;
481
482         va_start(args, fmt);
483         printk("%sEXT4-fs (%s): ", prefix, sb->s_id);
484         vprintk(fmt, args);
485         printk("\n");
486         va_end(args);
487 }
488
489 void __ext4_warning(struct super_block *sb, const char *function,
490                   const char *fmt, ...)
491 {
492         va_list args;
493
494         va_start(args, fmt);
495         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
496                sb->s_id, function);
497         vprintk(fmt, args);
498         printk("\n");
499         va_end(args);
500 }
501
502 void ext4_grp_locked_error(struct super_block *sb, ext4_group_t grp,
503                            const char *function, const char *fmt, ...)
504 __releases(bitlock)
505 __acquires(bitlock)
506 {
507         va_list args;
508         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
509
510         va_start(args, fmt);
511         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
512         vprintk(fmt, args);
513         printk("\n");
514         va_end(args);
515
516         if (test_opt(sb, ERRORS_CONT)) {
517                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
518                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
519                 ext4_commit_super(sb, 0);
520                 return;
521         }
522         ext4_unlock_group(sb, grp);
523         ext4_handle_error(sb);
524         /*
525          * We only get here in the ERRORS_RO case; relocking the group
526          * may be dangerous, but nothing bad will happen since the
527          * filesystem will have already been marked read/only and the
528          * journal has been aborted.  We return 1 as a hint to callers
529          * who might what to use the return value from
530          * ext4_grp_locked_error() to distinguish beween the
531          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
532          * aggressively from the ext4 function in question, with a
533          * more appropriate error code.
534          */
535         ext4_lock_group(sb, grp);
536         return;
537 }
538
539 void ext4_update_dynamic_rev(struct super_block *sb)
540 {
541         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
542
543         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
544                 return;
545
546         ext4_warning(sb,
547                      "updating to rev %d because of new feature flag, "
548                      "running e2fsck is recommended",
549                      EXT4_DYNAMIC_REV);
550
551         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
552         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
553         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
554         /* leave es->s_feature_*compat flags alone */
555         /* es->s_uuid will be set by e2fsck if empty */
556
557         /*
558          * The rest of the superblock fields should be zero, and if not it
559          * means they are likely already in use, so leave them alone.  We
560          * can leave it up to e2fsck to clean up any inconsistencies there.
561          */
562 }
563
564 /*
565  * Open the external journal device
566  */
567 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
568 {
569         struct block_device *bdev;
570         char b[BDEVNAME_SIZE];
571
572         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
573         if (IS_ERR(bdev))
574                 goto fail;
575         return bdev;
576
577 fail:
578         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
579                         __bdevname(dev, b), PTR_ERR(bdev));
580         return NULL;
581 }
582
583 /*
584  * Release the journal device
585  */
586 static int ext4_blkdev_put(struct block_device *bdev)
587 {
588         bd_release(bdev);
589         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
590 }
591
592 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
593 {
594         struct block_device *bdev;
595         int ret = -ENODEV;
596
597         bdev = sbi->journal_bdev;
598         if (bdev) {
599                 ret = ext4_blkdev_put(bdev);
600                 sbi->journal_bdev = NULL;
601         }
602         return ret;
603 }
604
605 static inline struct inode *orphan_list_entry(struct list_head *l)
606 {
607         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
608 }
609
610 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
611 {
612         struct list_head *l;
613
614         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
615                  le32_to_cpu(sbi->s_es->s_last_orphan));
616
617         printk(KERN_ERR "sb_info orphan list:\n");
618         list_for_each(l, &sbi->s_orphan) {
619                 struct inode *inode = orphan_list_entry(l);
620                 printk(KERN_ERR "  "
621                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
622                        inode->i_sb->s_id, inode->i_ino, inode,
623                        inode->i_mode, inode->i_nlink,
624                        NEXT_ORPHAN(inode));
625         }
626 }
627
628 static void ext4_put_super(struct super_block *sb)
629 {
630         struct ext4_sb_info *sbi = EXT4_SB(sb);
631         struct ext4_super_block *es = sbi->s_es;
632         int i, err;
633
634         flush_workqueue(sbi->dio_unwritten_wq);
635         destroy_workqueue(sbi->dio_unwritten_wq);
636
637         lock_super(sb);
638         lock_kernel();
639         if (sb->s_dirt)
640                 ext4_commit_super(sb, 1);
641
642         if (sbi->s_journal) {
643                 err = jbd2_journal_destroy(sbi->s_journal);
644                 sbi->s_journal = NULL;
645                 if (err < 0)
646                         ext4_abort(sb, __func__,
647                                    "Couldn't clean up the journal");
648         }
649
650         ext4_release_system_zone(sb);
651         ext4_mb_release(sb);
652         ext4_ext_release(sb);
653         ext4_xattr_put_super(sb);
654
655         if (!(sb->s_flags & MS_RDONLY)) {
656                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
657                 es->s_state = cpu_to_le16(sbi->s_mount_state);
658                 ext4_commit_super(sb, 1);
659         }
660         if (sbi->s_proc) {
661                 remove_proc_entry(sb->s_id, ext4_proc_root);
662         }
663         kobject_del(&sbi->s_kobj);
664
665         for (i = 0; i < sbi->s_gdb_count; i++)
666                 brelse(sbi->s_group_desc[i]);
667         kfree(sbi->s_group_desc);
668         if (is_vmalloc_addr(sbi->s_flex_groups))
669                 vfree(sbi->s_flex_groups);
670         else
671                 kfree(sbi->s_flex_groups);
672         percpu_counter_destroy(&sbi->s_freeblocks_counter);
673         percpu_counter_destroy(&sbi->s_freeinodes_counter);
674         percpu_counter_destroy(&sbi->s_dirs_counter);
675         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
676         brelse(sbi->s_sbh);
677 #ifdef CONFIG_QUOTA
678         for (i = 0; i < MAXQUOTAS; i++)
679                 kfree(sbi->s_qf_names[i]);
680 #endif
681
682         /* Debugging code just in case the in-memory inode orphan list
683          * isn't empty.  The on-disk one can be non-empty if we've
684          * detected an error and taken the fs readonly, but the
685          * in-memory list had better be clean by this point. */
686         if (!list_empty(&sbi->s_orphan))
687                 dump_orphan_list(sb, sbi);
688         J_ASSERT(list_empty(&sbi->s_orphan));
689
690         invalidate_bdev(sb->s_bdev);
691         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
692                 /*
693                  * Invalidate the journal device's buffers.  We don't want them
694                  * floating about in memory - the physical journal device may
695                  * hotswapped, and it breaks the `ro-after' testing code.
696                  */
697                 sync_blockdev(sbi->journal_bdev);
698                 invalidate_bdev(sbi->journal_bdev);
699                 ext4_blkdev_remove(sbi);
700         }
701         sb->s_fs_info = NULL;
702         /*
703          * Now that we are completely done shutting down the
704          * superblock, we need to actually destroy the kobject.
705          */
706         unlock_kernel();
707         unlock_super(sb);
708         kobject_put(&sbi->s_kobj);
709         wait_for_completion(&sbi->s_kobj_unregister);
710         kfree(sbi->s_blockgroup_lock);
711         kfree(sbi);
712 }
713
714 static struct kmem_cache *ext4_inode_cachep;
715
716 /*
717  * Called inside transaction, so use GFP_NOFS
718  */
719 static struct inode *ext4_alloc_inode(struct super_block *sb)
720 {
721         struct ext4_inode_info *ei;
722
723         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
724         if (!ei)
725                 return NULL;
726
727         ei->vfs_inode.i_version = 1;
728         ei->vfs_inode.i_data.writeback_index = 0;
729         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
730         INIT_LIST_HEAD(&ei->i_prealloc_list);
731         spin_lock_init(&ei->i_prealloc_lock);
732         /*
733          * Note:  We can be called before EXT4_SB(sb)->s_journal is set,
734          * therefore it can be null here.  Don't check it, just initialize
735          * jinode.
736          */
737         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
738         ei->i_reserved_data_blocks = 0;
739         ei->i_reserved_meta_blocks = 0;
740         ei->i_allocated_meta_blocks = 0;
741         ei->i_da_metadata_calc_len = 0;
742         ei->i_delalloc_reserved_flag = 0;
743         spin_lock_init(&(ei->i_block_reservation_lock));
744 #ifdef CONFIG_QUOTA
745         ei->i_reserved_quota = 0;
746 #endif
747         INIT_LIST_HEAD(&ei->i_completed_io_list);
748         spin_lock_init(&ei->i_completed_io_lock);
749         ei->cur_aio_dio = NULL;
750         ei->i_sync_tid = 0;
751         ei->i_datasync_tid = 0;
752
753         return &ei->vfs_inode;
754 }
755
756 static void ext4_destroy_inode(struct inode *inode)
757 {
758         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
759                 ext4_msg(inode->i_sb, KERN_ERR,
760                          "Inode %lu (%p): orphan list check failed!",
761                          inode->i_ino, EXT4_I(inode));
762                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
763                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
764                                 true);
765                 dump_stack();
766         }
767         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
768 }
769
770 static void init_once(void *foo)
771 {
772         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
773
774         INIT_LIST_HEAD(&ei->i_orphan);
775 #ifdef CONFIG_EXT4_FS_XATTR
776         init_rwsem(&ei->xattr_sem);
777 #endif
778         init_rwsem(&ei->i_data_sem);
779         inode_init_once(&ei->vfs_inode);
780 }
781
782 static int init_inodecache(void)
783 {
784         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
785                                              sizeof(struct ext4_inode_info),
786                                              0, (SLAB_RECLAIM_ACCOUNT|
787                                                 SLAB_MEM_SPREAD),
788                                              init_once);
789         if (ext4_inode_cachep == NULL)
790                 return -ENOMEM;
791         return 0;
792 }
793
794 static void destroy_inodecache(void)
795 {
796         kmem_cache_destroy(ext4_inode_cachep);
797 }
798
799 static void ext4_clear_inode(struct inode *inode)
800 {
801         dquot_drop(inode);
802         ext4_discard_preallocations(inode);
803         if (EXT4_JOURNAL(inode))
804                 jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
805                                        &EXT4_I(inode)->jinode);
806 }
807
808 static inline void ext4_show_quota_options(struct seq_file *seq,
809                                            struct super_block *sb)
810 {
811 #if defined(CONFIG_QUOTA)
812         struct ext4_sb_info *sbi = EXT4_SB(sb);
813
814         if (sbi->s_jquota_fmt) {
815                 char *fmtname = "";
816
817                 switch (sbi->s_jquota_fmt) {
818                 case QFMT_VFS_OLD:
819                         fmtname = "vfsold";
820                         break;
821                 case QFMT_VFS_V0:
822                         fmtname = "vfsv0";
823                         break;
824                 case QFMT_VFS_V1:
825                         fmtname = "vfsv1";
826                         break;
827                 }
828                 seq_printf(seq, ",jqfmt=%s", fmtname);
829         }
830
831         if (sbi->s_qf_names[USRQUOTA])
832                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
833
834         if (sbi->s_qf_names[GRPQUOTA])
835                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
836
837         if (test_opt(sb, USRQUOTA))
838                 seq_puts(seq, ",usrquota");
839
840         if (test_opt(sb, GRPQUOTA))
841                 seq_puts(seq, ",grpquota");
842 #endif
843 }
844
845 /*
846  * Show an option if
847  *  - it's set to a non-default value OR
848  *  - if the per-sb default is different from the global default
849  */
850 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
851 {
852         int def_errors;
853         unsigned long def_mount_opts;
854         struct super_block *sb = vfs->mnt_sb;
855         struct ext4_sb_info *sbi = EXT4_SB(sb);
856         struct ext4_super_block *es = sbi->s_es;
857
858         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
859         def_errors     = le16_to_cpu(es->s_errors);
860
861         if (sbi->s_sb_block != 1)
862                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
863         if (test_opt(sb, MINIX_DF))
864                 seq_puts(seq, ",minixdf");
865         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
866                 seq_puts(seq, ",grpid");
867         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
868                 seq_puts(seq, ",nogrpid");
869         if (sbi->s_resuid != EXT4_DEF_RESUID ||
870             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
871                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
872         }
873         if (sbi->s_resgid != EXT4_DEF_RESGID ||
874             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
875                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
876         }
877         if (test_opt(sb, ERRORS_RO)) {
878                 if (def_errors == EXT4_ERRORS_PANIC ||
879                     def_errors == EXT4_ERRORS_CONTINUE) {
880                         seq_puts(seq, ",errors=remount-ro");
881                 }
882         }
883         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
884                 seq_puts(seq, ",errors=continue");
885         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
886                 seq_puts(seq, ",errors=panic");
887         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
888                 seq_puts(seq, ",nouid32");
889         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
890                 seq_puts(seq, ",debug");
891         if (test_opt(sb, OLDALLOC))
892                 seq_puts(seq, ",oldalloc");
893 #ifdef CONFIG_EXT4_FS_XATTR
894         if (test_opt(sb, XATTR_USER) &&
895                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
896                 seq_puts(seq, ",user_xattr");
897         if (!test_opt(sb, XATTR_USER) &&
898             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
899                 seq_puts(seq, ",nouser_xattr");
900         }
901 #endif
902 #ifdef CONFIG_EXT4_FS_POSIX_ACL
903         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
904                 seq_puts(seq, ",acl");
905         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
906                 seq_puts(seq, ",noacl");
907 #endif
908         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
909                 seq_printf(seq, ",commit=%u",
910                            (unsigned) (sbi->s_commit_interval / HZ));
911         }
912         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
913                 seq_printf(seq, ",min_batch_time=%u",
914                            (unsigned) sbi->s_min_batch_time);
915         }
916         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
917                 seq_printf(seq, ",max_batch_time=%u",
918                            (unsigned) sbi->s_min_batch_time);
919         }
920
921         /*
922          * We're changing the default of barrier mount option, so
923          * let's always display its mount state so it's clear what its
924          * status is.
925          */
926         seq_puts(seq, ",barrier=");
927         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
928         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
929                 seq_puts(seq, ",journal_async_commit");
930         if (test_opt(sb, NOBH))
931                 seq_puts(seq, ",nobh");
932         if (test_opt(sb, I_VERSION))
933                 seq_puts(seq, ",i_version");
934         if (!test_opt(sb, DELALLOC))
935                 seq_puts(seq, ",nodelalloc");
936
937
938         if (sbi->s_stripe)
939                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
940         /*
941          * journal mode get enabled in different ways
942          * So just print the value even if we didn't specify it
943          */
944         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
945                 seq_puts(seq, ",data=journal");
946         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
947                 seq_puts(seq, ",data=ordered");
948         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
949                 seq_puts(seq, ",data=writeback");
950
951         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
952                 seq_printf(seq, ",inode_readahead_blks=%u",
953                            sbi->s_inode_readahead_blks);
954
955         if (test_opt(sb, DATA_ERR_ABORT))
956                 seq_puts(seq, ",data_err=abort");
957
958         if (test_opt(sb, NO_AUTO_DA_ALLOC))
959                 seq_puts(seq, ",noauto_da_alloc");
960
961         if (test_opt(sb, DISCARD))
962                 seq_puts(seq, ",discard");
963
964         if (test_opt(sb, NOLOAD))
965                 seq_puts(seq, ",norecovery");
966
967         if (test_opt(sb, DIOREAD_NOLOCK))
968                 seq_puts(seq, ",dioread_nolock");
969
970         ext4_show_quota_options(seq, sb);
971
972         return 0;
973 }
974
975 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
976                                         u64 ino, u32 generation)
977 {
978         struct inode *inode;
979
980         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
981                 return ERR_PTR(-ESTALE);
982         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
983                 return ERR_PTR(-ESTALE);
984
985         /* iget isn't really right if the inode is currently unallocated!!
986          *
987          * ext4_read_inode will return a bad_inode if the inode had been
988          * deleted, so we should be safe.
989          *
990          * Currently we don't know the generation for parent directory, so
991          * a generation of 0 means "accept any"
992          */
993         inode = ext4_iget(sb, ino);
994         if (IS_ERR(inode))
995                 return ERR_CAST(inode);
996         if (generation && inode->i_generation != generation) {
997                 iput(inode);
998                 return ERR_PTR(-ESTALE);
999         }
1000
1001         return inode;
1002 }
1003
1004 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1005                                         int fh_len, int fh_type)
1006 {
1007         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1008                                     ext4_nfs_get_inode);
1009 }
1010
1011 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1012                                         int fh_len, int fh_type)
1013 {
1014         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1015                                     ext4_nfs_get_inode);
1016 }
1017
1018 /*
1019  * Try to release metadata pages (indirect blocks, directories) which are
1020  * mapped via the block device.  Since these pages could have journal heads
1021  * which would prevent try_to_free_buffers() from freeing them, we must use
1022  * jbd2 layer's try_to_free_buffers() function to release them.
1023  */
1024 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1025                                  gfp_t wait)
1026 {
1027         journal_t *journal = EXT4_SB(sb)->s_journal;
1028
1029         WARN_ON(PageChecked(page));
1030         if (!page_has_buffers(page))
1031                 return 0;
1032         if (journal)
1033                 return jbd2_journal_try_to_free_buffers(journal, page,
1034                                                         wait & ~__GFP_WAIT);
1035         return try_to_free_buffers(page);
1036 }
1037
1038 #ifdef CONFIG_QUOTA
1039 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1040 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1041
1042 static int ext4_write_dquot(struct dquot *dquot);
1043 static int ext4_acquire_dquot(struct dquot *dquot);
1044 static int ext4_release_dquot(struct dquot *dquot);
1045 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1046 static int ext4_write_info(struct super_block *sb, int type);
1047 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1048                                 char *path, int remount);
1049 static int ext4_quota_on_mount(struct super_block *sb, int type);
1050 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1051                                size_t len, loff_t off);
1052 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1053                                 const char *data, size_t len, loff_t off);
1054
1055 static const struct dquot_operations ext4_quota_operations = {
1056 #ifdef CONFIG_QUOTA
1057         .get_reserved_space = ext4_get_reserved_space,
1058 #endif
1059         .write_dquot    = ext4_write_dquot,
1060         .acquire_dquot  = ext4_acquire_dquot,
1061         .release_dquot  = ext4_release_dquot,
1062         .mark_dirty     = ext4_mark_dquot_dirty,
1063         .write_info     = ext4_write_info,
1064         .alloc_dquot    = dquot_alloc,
1065         .destroy_dquot  = dquot_destroy,
1066 };
1067
1068 static const struct quotactl_ops ext4_qctl_operations = {
1069         .quota_on       = ext4_quota_on,
1070         .quota_off      = vfs_quota_off,
1071         .quota_sync     = vfs_quota_sync,
1072         .get_info       = vfs_get_dqinfo,
1073         .set_info       = vfs_set_dqinfo,
1074         .get_dqblk      = vfs_get_dqblk,
1075         .set_dqblk      = vfs_set_dqblk
1076 };
1077 #endif
1078
1079 static const struct super_operations ext4_sops = {
1080         .alloc_inode    = ext4_alloc_inode,
1081         .destroy_inode  = ext4_destroy_inode,
1082         .write_inode    = ext4_write_inode,
1083         .dirty_inode    = ext4_dirty_inode,
1084         .delete_inode   = ext4_delete_inode,
1085         .put_super      = ext4_put_super,
1086         .sync_fs        = ext4_sync_fs,
1087         .freeze_fs      = ext4_freeze,
1088         .unfreeze_fs    = ext4_unfreeze,
1089         .statfs         = ext4_statfs,
1090         .remount_fs     = ext4_remount,
1091         .clear_inode    = ext4_clear_inode,
1092         .show_options   = ext4_show_options,
1093 #ifdef CONFIG_QUOTA
1094         .quota_read     = ext4_quota_read,
1095         .quota_write    = ext4_quota_write,
1096 #endif
1097         .bdev_try_to_free_page = bdev_try_to_free_page,
1098 };
1099
1100 static const struct super_operations ext4_nojournal_sops = {
1101         .alloc_inode    = ext4_alloc_inode,
1102         .destroy_inode  = ext4_destroy_inode,
1103         .write_inode    = ext4_write_inode,
1104         .dirty_inode    = ext4_dirty_inode,
1105         .delete_inode   = ext4_delete_inode,
1106         .write_super    = ext4_write_super,
1107         .put_super      = ext4_put_super,
1108         .statfs         = ext4_statfs,
1109         .remount_fs     = ext4_remount,
1110         .clear_inode    = ext4_clear_inode,
1111         .show_options   = ext4_show_options,
1112 #ifdef CONFIG_QUOTA
1113         .quota_read     = ext4_quota_read,
1114         .quota_write    = ext4_quota_write,
1115 #endif
1116         .bdev_try_to_free_page = bdev_try_to_free_page,
1117 };
1118
1119 static const struct export_operations ext4_export_ops = {
1120         .fh_to_dentry = ext4_fh_to_dentry,
1121         .fh_to_parent = ext4_fh_to_parent,
1122         .get_parent = ext4_get_parent,
1123 };
1124
1125 enum {
1126         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1127         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1128         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1129         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1130         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1131         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1132         Opt_journal_update, Opt_journal_dev,
1133         Opt_journal_checksum, Opt_journal_async_commit,
1134         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1135         Opt_data_err_abort, Opt_data_err_ignore,
1136         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1137         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1138         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1139         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1140         Opt_stripe, Opt_delalloc, Opt_nodelalloc,
1141         Opt_block_validity, Opt_noblock_validity,
1142         Opt_inode_readahead_blks, Opt_journal_ioprio,
1143         Opt_dioread_nolock, Opt_dioread_lock,
1144         Opt_discard, Opt_nodiscard,
1145 };
1146
1147 static const match_table_t tokens = {
1148         {Opt_bsd_df, "bsddf"},
1149         {Opt_minix_df, "minixdf"},
1150         {Opt_grpid, "grpid"},
1151         {Opt_grpid, "bsdgroups"},
1152         {Opt_nogrpid, "nogrpid"},
1153         {Opt_nogrpid, "sysvgroups"},
1154         {Opt_resgid, "resgid=%u"},
1155         {Opt_resuid, "resuid=%u"},
1156         {Opt_sb, "sb=%u"},
1157         {Opt_err_cont, "errors=continue"},
1158         {Opt_err_panic, "errors=panic"},
1159         {Opt_err_ro, "errors=remount-ro"},
1160         {Opt_nouid32, "nouid32"},
1161         {Opt_debug, "debug"},
1162         {Opt_oldalloc, "oldalloc"},
1163         {Opt_orlov, "orlov"},
1164         {Opt_user_xattr, "user_xattr"},
1165         {Opt_nouser_xattr, "nouser_xattr"},
1166         {Opt_acl, "acl"},
1167         {Opt_noacl, "noacl"},
1168         {Opt_noload, "noload"},
1169         {Opt_noload, "norecovery"},
1170         {Opt_nobh, "nobh"},
1171         {Opt_bh, "bh"},
1172         {Opt_commit, "commit=%u"},
1173         {Opt_min_batch_time, "min_batch_time=%u"},
1174         {Opt_max_batch_time, "max_batch_time=%u"},
1175         {Opt_journal_update, "journal=update"},
1176         {Opt_journal_dev, "journal_dev=%u"},
1177         {Opt_journal_checksum, "journal_checksum"},
1178         {Opt_journal_async_commit, "journal_async_commit"},
1179         {Opt_abort, "abort"},
1180         {Opt_data_journal, "data=journal"},
1181         {Opt_data_ordered, "data=ordered"},
1182         {Opt_data_writeback, "data=writeback"},
1183         {Opt_data_err_abort, "data_err=abort"},
1184         {Opt_data_err_ignore, "data_err=ignore"},
1185         {Opt_offusrjquota, "usrjquota="},
1186         {Opt_usrjquota, "usrjquota=%s"},
1187         {Opt_offgrpjquota, "grpjquota="},
1188         {Opt_grpjquota, "grpjquota=%s"},
1189         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1190         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1191         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1192         {Opt_grpquota, "grpquota"},
1193         {Opt_noquota, "noquota"},
1194         {Opt_quota, "quota"},
1195         {Opt_usrquota, "usrquota"},
1196         {Opt_barrier, "barrier=%u"},
1197         {Opt_barrier, "barrier"},
1198         {Opt_nobarrier, "nobarrier"},
1199         {Opt_i_version, "i_version"},
1200         {Opt_stripe, "stripe=%u"},
1201         {Opt_resize, "resize"},
1202         {Opt_delalloc, "delalloc"},
1203         {Opt_nodelalloc, "nodelalloc"},
1204         {Opt_block_validity, "block_validity"},
1205         {Opt_noblock_validity, "noblock_validity"},
1206         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1207         {Opt_journal_ioprio, "journal_ioprio=%u"},
1208         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1209         {Opt_auto_da_alloc, "auto_da_alloc"},
1210         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1211         {Opt_dioread_nolock, "dioread_nolock"},
1212         {Opt_dioread_lock, "dioread_lock"},
1213         {Opt_discard, "discard"},
1214         {Opt_nodiscard, "nodiscard"},
1215         {Opt_err, NULL},
1216 };
1217
1218 static ext4_fsblk_t get_sb_block(void **data)
1219 {
1220         ext4_fsblk_t    sb_block;
1221         char            *options = (char *) *data;
1222
1223         if (!options || strncmp(options, "sb=", 3) != 0)
1224                 return 1;       /* Default location */
1225
1226         options += 3;
1227         /* TODO: use simple_strtoll with >32bit ext4 */
1228         sb_block = simple_strtoul(options, &options, 0);
1229         if (*options && *options != ',') {
1230                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1231                        (char *) *data);
1232                 return 1;
1233         }
1234         if (*options == ',')
1235                 options++;
1236         *data = (void *) options;
1237
1238         return sb_block;
1239 }
1240
1241 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1242 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1243         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1244
1245 #ifdef CONFIG_QUOTA
1246 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1247 {
1248         struct ext4_sb_info *sbi = EXT4_SB(sb);
1249         char *qname;
1250
1251         if (sb_any_quota_loaded(sb) &&
1252                 !sbi->s_qf_names[qtype]) {
1253                 ext4_msg(sb, KERN_ERR,
1254                         "Cannot change journaled "
1255                         "quota options when quota turned on");
1256                 return 0;
1257         }
1258         qname = match_strdup(args);
1259         if (!qname) {
1260                 ext4_msg(sb, KERN_ERR,
1261                         "Not enough memory for storing quotafile name");
1262                 return 0;
1263         }
1264         if (sbi->s_qf_names[qtype] &&
1265                 strcmp(sbi->s_qf_names[qtype], qname)) {
1266                 ext4_msg(sb, KERN_ERR,
1267                         "%s quota file already specified", QTYPE2NAME(qtype));
1268                 kfree(qname);
1269                 return 0;
1270         }
1271         sbi->s_qf_names[qtype] = qname;
1272         if (strchr(sbi->s_qf_names[qtype], '/')) {
1273                 ext4_msg(sb, KERN_ERR,
1274                         "quotafile must be on filesystem root");
1275                 kfree(sbi->s_qf_names[qtype]);
1276                 sbi->s_qf_names[qtype] = NULL;
1277                 return 0;
1278         }
1279         set_opt(sbi->s_mount_opt, QUOTA);
1280         return 1;
1281 }
1282
1283 static int clear_qf_name(struct super_block *sb, int qtype)
1284 {
1285
1286         struct ext4_sb_info *sbi = EXT4_SB(sb);
1287
1288         if (sb_any_quota_loaded(sb) &&
1289                 sbi->s_qf_names[qtype]) {
1290                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1291                         " when quota turned on");
1292                 return 0;
1293         }
1294         /*
1295          * The space will be released later when all options are confirmed
1296          * to be correct
1297          */
1298         sbi->s_qf_names[qtype] = NULL;
1299         return 1;
1300 }
1301 #endif
1302
1303 static int parse_options(char *options, struct super_block *sb,
1304                          unsigned long *journal_devnum,
1305                          unsigned int *journal_ioprio,
1306                          ext4_fsblk_t *n_blocks_count, int is_remount)
1307 {
1308         struct ext4_sb_info *sbi = EXT4_SB(sb);
1309         char *p;
1310         substring_t args[MAX_OPT_ARGS];
1311         int data_opt = 0;
1312         int option;
1313 #ifdef CONFIG_QUOTA
1314         int qfmt;
1315 #endif
1316
1317         if (!options)
1318                 return 1;
1319
1320         while ((p = strsep(&options, ",")) != NULL) {
1321                 int token;
1322                 if (!*p)
1323                         continue;
1324
1325                 /*
1326                  * Initialize args struct so we know whether arg was
1327                  * found; some options take optional arguments.
1328                  */
1329                 args[0].to = args[0].from = 0;
1330                 token = match_token(p, tokens, args);
1331                 switch (token) {
1332                 case Opt_bsd_df:
1333                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1334                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1335                         break;
1336                 case Opt_minix_df:
1337                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1338                         set_opt(sbi->s_mount_opt, MINIX_DF);
1339
1340                         break;
1341                 case Opt_grpid:
1342                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1343                         set_opt(sbi->s_mount_opt, GRPID);
1344
1345                         break;
1346                 case Opt_nogrpid:
1347                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1348                         clear_opt(sbi->s_mount_opt, GRPID);
1349
1350                         break;
1351                 case Opt_resuid:
1352                         if (match_int(&args[0], &option))
1353                                 return 0;
1354                         sbi->s_resuid = option;
1355                         break;
1356                 case Opt_resgid:
1357                         if (match_int(&args[0], &option))
1358                                 return 0;
1359                         sbi->s_resgid = option;
1360                         break;
1361                 case Opt_sb:
1362                         /* handled by get_sb_block() instead of here */
1363                         /* *sb_block = match_int(&args[0]); */
1364                         break;
1365                 case Opt_err_panic:
1366                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1367                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1368                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1369                         break;
1370                 case Opt_err_ro:
1371                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1372                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1373                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1374                         break;
1375                 case Opt_err_cont:
1376                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1377                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1378                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1379                         break;
1380                 case Opt_nouid32:
1381                         set_opt(sbi->s_mount_opt, NO_UID32);
1382                         break;
1383                 case Opt_debug:
1384                         set_opt(sbi->s_mount_opt, DEBUG);
1385                         break;
1386                 case Opt_oldalloc:
1387                         set_opt(sbi->s_mount_opt, OLDALLOC);
1388                         break;
1389                 case Opt_orlov:
1390                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1391                         break;
1392 #ifdef CONFIG_EXT4_FS_XATTR
1393                 case Opt_user_xattr:
1394                         set_opt(sbi->s_mount_opt, XATTR_USER);
1395                         break;
1396                 case Opt_nouser_xattr:
1397                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1398                         break;
1399 #else
1400                 case Opt_user_xattr:
1401                 case Opt_nouser_xattr:
1402                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1403                         break;
1404 #endif
1405 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1406                 case Opt_acl:
1407                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1408                         break;
1409                 case Opt_noacl:
1410                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1411                         break;
1412 #else
1413                 case Opt_acl:
1414                 case Opt_noacl:
1415                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1416                         break;
1417 #endif
1418                 case Opt_journal_update:
1419                         /* @@@ FIXME */
1420                         /* Eventually we will want to be able to create
1421                            a journal file here.  For now, only allow the
1422                            user to specify an existing inode to be the
1423                            journal file. */
1424                         if (is_remount) {
1425                                 ext4_msg(sb, KERN_ERR,
1426                                          "Cannot specify journal on remount");
1427                                 return 0;
1428                         }
1429                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1430                         break;
1431                 case Opt_journal_dev:
1432                         if (is_remount) {
1433                                 ext4_msg(sb, KERN_ERR,
1434                                         "Cannot specify journal on remount");
1435                                 return 0;
1436                         }
1437                         if (match_int(&args[0], &option))
1438                                 return 0;
1439                         *journal_devnum = option;
1440                         break;
1441                 case Opt_journal_checksum:
1442                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1443                         break;
1444                 case Opt_journal_async_commit:
1445                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1446                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1447                         break;
1448                 case Opt_noload:
1449                         set_opt(sbi->s_mount_opt, NOLOAD);
1450                         break;
1451                 case Opt_commit:
1452                         if (match_int(&args[0], &option))
1453                                 return 0;
1454                         if (option < 0)
1455                                 return 0;
1456                         if (option == 0)
1457                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1458                         sbi->s_commit_interval = HZ * option;
1459                         break;
1460                 case Opt_max_batch_time:
1461                         if (match_int(&args[0], &option))
1462                                 return 0;
1463                         if (option < 0)
1464                                 return 0;
1465                         if (option == 0)
1466                                 option = EXT4_DEF_MAX_BATCH_TIME;
1467                         sbi->s_max_batch_time = option;
1468                         break;
1469                 case Opt_min_batch_time:
1470                         if (match_int(&args[0], &option))
1471                                 return 0;
1472                         if (option < 0)
1473                                 return 0;
1474                         sbi->s_min_batch_time = option;
1475                         break;
1476                 case Opt_data_journal:
1477                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1478                         goto datacheck;
1479                 case Opt_data_ordered:
1480                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1481                         goto datacheck;
1482                 case Opt_data_writeback:
1483                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1484                 datacheck:
1485                         if (is_remount) {
1486                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1487                                         ext4_msg(sb, KERN_ERR,
1488                                                 "Cannot change data mode on remount");
1489                                         return 0;
1490                                 }
1491                         } else {
1492                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1493                                 sbi->s_mount_opt |= data_opt;
1494                         }
1495                         break;
1496                 case Opt_data_err_abort:
1497                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1498                         break;
1499                 case Opt_data_err_ignore:
1500                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1501                         break;
1502 #ifdef CONFIG_QUOTA
1503                 case Opt_usrjquota:
1504                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1505                                 return 0;
1506                         break;
1507                 case Opt_grpjquota:
1508                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1509                                 return 0;
1510                         break;
1511                 case Opt_offusrjquota:
1512                         if (!clear_qf_name(sb, USRQUOTA))
1513                                 return 0;
1514                         break;
1515                 case Opt_offgrpjquota:
1516                         if (!clear_qf_name(sb, GRPQUOTA))
1517                                 return 0;
1518                         break;
1519
1520                 case Opt_jqfmt_vfsold:
1521                         qfmt = QFMT_VFS_OLD;
1522                         goto set_qf_format;
1523                 case Opt_jqfmt_vfsv0:
1524                         qfmt = QFMT_VFS_V0;
1525                         goto set_qf_format;
1526                 case Opt_jqfmt_vfsv1:
1527                         qfmt = QFMT_VFS_V1;
1528 set_qf_format:
1529                         if (sb_any_quota_loaded(sb) &&
1530                             sbi->s_jquota_fmt != qfmt) {
1531                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1532                                         "journaled quota options when "
1533                                         "quota turned on");
1534                                 return 0;
1535                         }
1536                         sbi->s_jquota_fmt = qfmt;
1537                         break;
1538                 case Opt_quota:
1539                 case Opt_usrquota:
1540                         set_opt(sbi->s_mount_opt, QUOTA);
1541                         set_opt(sbi->s_mount_opt, USRQUOTA);
1542                         break;
1543                 case Opt_grpquota:
1544                         set_opt(sbi->s_mount_opt, QUOTA);
1545                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1546                         break;
1547                 case Opt_noquota:
1548                         if (sb_any_quota_loaded(sb)) {
1549                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1550                                         "options when quota turned on");
1551                                 return 0;
1552                         }
1553                         clear_opt(sbi->s_mount_opt, QUOTA);
1554                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1555                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1556                         break;
1557 #else
1558                 case Opt_quota:
1559                 case Opt_usrquota:
1560                 case Opt_grpquota:
1561                         ext4_msg(sb, KERN_ERR,
1562                                 "quota options not supported");
1563                         break;
1564                 case Opt_usrjquota:
1565                 case Opt_grpjquota:
1566                 case Opt_offusrjquota:
1567                 case Opt_offgrpjquota:
1568                 case Opt_jqfmt_vfsold:
1569                 case Opt_jqfmt_vfsv0:
1570                 case Opt_jqfmt_vfsv1:
1571                         ext4_msg(sb, KERN_ERR,
1572                                 "journaled quota options not supported");
1573                         break;
1574                 case Opt_noquota:
1575                         break;
1576 #endif
1577                 case Opt_abort:
1578                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1579                         break;
1580                 case Opt_nobarrier:
1581                         clear_opt(sbi->s_mount_opt, BARRIER);
1582                         break;
1583                 case Opt_barrier:
1584                         if (args[0].from) {
1585                                 if (match_int(&args[0], &option))
1586                                         return 0;
1587                         } else
1588                                 option = 1;     /* No argument, default to 1 */
1589                         if (option)
1590                                 set_opt(sbi->s_mount_opt, BARRIER);
1591                         else
1592                                 clear_opt(sbi->s_mount_opt, BARRIER);
1593                         break;
1594                 case Opt_ignore:
1595                         break;
1596                 case Opt_resize:
1597                         if (!is_remount) {
1598                                 ext4_msg(sb, KERN_ERR,
1599                                         "resize option only available "
1600                                         "for remount");
1601                                 return 0;
1602                         }
1603                         if (match_int(&args[0], &option) != 0)
1604                                 return 0;
1605                         *n_blocks_count = option;
1606                         break;
1607                 case Opt_nobh:
1608                         set_opt(sbi->s_mount_opt, NOBH);
1609                         break;
1610                 case Opt_bh:
1611                         clear_opt(sbi->s_mount_opt, NOBH);
1612                         break;
1613                 case Opt_i_version:
1614                         set_opt(sbi->s_mount_opt, I_VERSION);
1615                         sb->s_flags |= MS_I_VERSION;
1616                         break;
1617                 case Opt_nodelalloc:
1618                         clear_opt(sbi->s_mount_opt, DELALLOC);
1619                         break;
1620                 case Opt_stripe:
1621                         if (match_int(&args[0], &option))
1622                                 return 0;
1623                         if (option < 0)
1624                                 return 0;
1625                         sbi->s_stripe = option;
1626                         break;
1627                 case Opt_delalloc:
1628                         set_opt(sbi->s_mount_opt, DELALLOC);
1629                         break;
1630                 case Opt_block_validity:
1631                         set_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1632                         break;
1633                 case Opt_noblock_validity:
1634                         clear_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1635                         break;
1636                 case Opt_inode_readahead_blks:
1637                         if (match_int(&args[0], &option))
1638                                 return 0;
1639                         if (option < 0 || option > (1 << 30))
1640                                 return 0;
1641                         if (!is_power_of_2(option)) {
1642                                 ext4_msg(sb, KERN_ERR,
1643                                          "EXT4-fs: inode_readahead_blks"
1644                                          " must be a power of 2");
1645                                 return 0;
1646                         }
1647                         sbi->s_inode_readahead_blks = option;
1648                         break;
1649                 case Opt_journal_ioprio:
1650                         if (match_int(&args[0], &option))
1651                                 return 0;
1652                         if (option < 0 || option > 7)
1653                                 break;
1654                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1655                                                             option);
1656                         break;
1657                 case Opt_noauto_da_alloc:
1658                         set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1659                         break;
1660                 case Opt_auto_da_alloc:
1661                         if (args[0].from) {
1662                                 if (match_int(&args[0], &option))
1663                                         return 0;
1664                         } else
1665                                 option = 1;     /* No argument, default to 1 */
1666                         if (option)
1667                                 clear_opt(sbi->s_mount_opt, NO_AUTO_DA_ALLOC);
1668                         else
1669                                 set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1670                         break;
1671                 case Opt_discard:
1672                         set_opt(sbi->s_mount_opt, DISCARD);
1673                         break;
1674                 case Opt_nodiscard:
1675                         clear_opt(sbi->s_mount_opt, DISCARD);
1676                         break;
1677                 case Opt_dioread_nolock:
1678                         set_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1679                         break;
1680                 case Opt_dioread_lock:
1681                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1682                         break;
1683                 default:
1684                         ext4_msg(sb, KERN_ERR,
1685                                "Unrecognized mount option \"%s\" "
1686                                "or missing value", p);
1687                         return 0;
1688                 }
1689         }
1690 #ifdef CONFIG_QUOTA
1691         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1692                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1693                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1694
1695                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1696                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1697
1698                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1699                         ext4_msg(sb, KERN_ERR, "old and new quota "
1700                                         "format mixing");
1701                         return 0;
1702                 }
1703
1704                 if (!sbi->s_jquota_fmt) {
1705                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1706                                         "not specified");
1707                         return 0;
1708                 }
1709         } else {
1710                 if (sbi->s_jquota_fmt) {
1711                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1712                                         "specified with no journaling "
1713                                         "enabled");
1714                         return 0;
1715                 }
1716         }
1717 #endif
1718         return 1;
1719 }
1720
1721 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1722                             int read_only)
1723 {
1724         struct ext4_sb_info *sbi = EXT4_SB(sb);
1725         int res = 0;
1726
1727         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1728                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1729                          "forcing read-only mode");
1730                 res = MS_RDONLY;
1731         }
1732         if (read_only)
1733                 return res;
1734         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1735                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1736                          "running e2fsck is recommended");
1737         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1738                 ext4_msg(sb, KERN_WARNING,
1739                          "warning: mounting fs with errors, "
1740                          "running e2fsck is recommended");
1741         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1742                  le16_to_cpu(es->s_mnt_count) >=
1743                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1744                 ext4_msg(sb, KERN_WARNING,
1745                          "warning: maximal mount count reached, "
1746                          "running e2fsck is recommended");
1747         else if (le32_to_cpu(es->s_checkinterval) &&
1748                 (le32_to_cpu(es->s_lastcheck) +
1749                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1750                 ext4_msg(sb, KERN_WARNING,
1751                          "warning: checktime reached, "
1752                          "running e2fsck is recommended");
1753         if (!sbi->s_journal)
1754                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1755         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1756                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1757         le16_add_cpu(&es->s_mnt_count, 1);
1758         es->s_mtime = cpu_to_le32(get_seconds());
1759         ext4_update_dynamic_rev(sb);
1760         if (sbi->s_journal)
1761                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1762
1763         ext4_commit_super(sb, 1);
1764         if (test_opt(sb, DEBUG))
1765                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1766                                 "bpg=%lu, ipg=%lu, mo=%04x]\n",
1767                         sb->s_blocksize,
1768                         sbi->s_groups_count,
1769                         EXT4_BLOCKS_PER_GROUP(sb),
1770                         EXT4_INODES_PER_GROUP(sb),
1771                         sbi->s_mount_opt);
1772
1773         return res;
1774 }
1775
1776 static int ext4_fill_flex_info(struct super_block *sb)
1777 {
1778         struct ext4_sb_info *sbi = EXT4_SB(sb);
1779         struct ext4_group_desc *gdp = NULL;
1780         ext4_group_t flex_group_count;
1781         ext4_group_t flex_group;
1782         int groups_per_flex = 0;
1783         size_t size;
1784         int i;
1785
1786         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1787         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1788
1789         if (groups_per_flex < 2) {
1790                 sbi->s_log_groups_per_flex = 0;
1791                 return 1;
1792         }
1793
1794         /* We allocate both existing and potentially added groups */
1795         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1796                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1797                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1798         size = flex_group_count * sizeof(struct flex_groups);
1799         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1800         if (sbi->s_flex_groups == NULL) {
1801                 sbi->s_flex_groups = vmalloc(size);
1802                 if (sbi->s_flex_groups)
1803                         memset(sbi->s_flex_groups, 0, size);
1804         }
1805         if (sbi->s_flex_groups == NULL) {
1806                 ext4_msg(sb, KERN_ERR, "not enough memory for "
1807                                 "%u flex groups", flex_group_count);
1808                 goto failed;
1809         }
1810
1811         for (i = 0; i < sbi->s_groups_count; i++) {
1812                 gdp = ext4_get_group_desc(sb, i, NULL);
1813
1814                 flex_group = ext4_flex_group(sbi, i);
1815                 atomic_add(ext4_free_inodes_count(sb, gdp),
1816                            &sbi->s_flex_groups[flex_group].free_inodes);
1817                 atomic_add(ext4_free_blks_count(sb, gdp),
1818                            &sbi->s_flex_groups[flex_group].free_blocks);
1819                 atomic_add(ext4_used_dirs_count(sb, gdp),
1820                            &sbi->s_flex_groups[flex_group].used_dirs);
1821         }
1822
1823         return 1;
1824 failed:
1825         return 0;
1826 }
1827
1828 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1829                             struct ext4_group_desc *gdp)
1830 {
1831         __u16 crc = 0;
1832
1833         if (sbi->s_es->s_feature_ro_compat &
1834             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1835                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1836                 __le32 le_group = cpu_to_le32(block_group);
1837
1838                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1839                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1840                 crc = crc16(crc, (__u8 *)gdp, offset);
1841                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1842                 /* for checksum of struct ext4_group_desc do the rest...*/
1843                 if ((sbi->s_es->s_feature_incompat &
1844                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1845                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1846                         crc = crc16(crc, (__u8 *)gdp + offset,
1847                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1848                                         offset);
1849         }
1850
1851         return cpu_to_le16(crc);
1852 }
1853
1854 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1855                                 struct ext4_group_desc *gdp)
1856 {
1857         if ((sbi->s_es->s_feature_ro_compat &
1858              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1859             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1860                 return 0;
1861
1862         return 1;
1863 }
1864
1865 /* Called at mount-time, super-block is locked */
1866 static int ext4_check_descriptors(struct super_block *sb)
1867 {
1868         struct ext4_sb_info *sbi = EXT4_SB(sb);
1869         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1870         ext4_fsblk_t last_block;
1871         ext4_fsblk_t block_bitmap;
1872         ext4_fsblk_t inode_bitmap;
1873         ext4_fsblk_t inode_table;
1874         int flexbg_flag = 0;
1875         ext4_group_t i;
1876
1877         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1878                 flexbg_flag = 1;
1879
1880         ext4_debug("Checking group descriptors");
1881
1882         for (i = 0; i < sbi->s_groups_count; i++) {
1883                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1884
1885                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1886                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1887                 else
1888                         last_block = first_block +
1889                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1890
1891                 block_bitmap = ext4_block_bitmap(sb, gdp);
1892                 if (block_bitmap < first_block || block_bitmap > last_block) {
1893                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1894                                "Block bitmap for group %u not in group "
1895                                "(block %llu)!", i, block_bitmap);
1896                         return 0;
1897                 }
1898                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1899                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1900                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1901                                "Inode bitmap for group %u not in group "
1902                                "(block %llu)!", i, inode_bitmap);
1903                         return 0;
1904                 }
1905                 inode_table = ext4_inode_table(sb, gdp);
1906                 if (inode_table < first_block ||
1907                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1908                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1909                                "Inode table for group %u not in group "
1910                                "(block %llu)!", i, inode_table);
1911                         return 0;
1912                 }
1913                 ext4_lock_group(sb, i);
1914                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1915                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1916                                  "Checksum for group %u failed (%u!=%u)",
1917                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1918                                      gdp)), le16_to_cpu(gdp->bg_checksum));
1919                         if (!(sb->s_flags & MS_RDONLY)) {
1920                                 ext4_unlock_group(sb, i);
1921                                 return 0;
1922                         }
1923                 }
1924                 ext4_unlock_group(sb, i);
1925                 if (!flexbg_flag)
1926                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1927         }
1928
1929         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1930         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
1931         return 1;
1932 }
1933
1934 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1935  * the superblock) which were deleted from all directories, but held open by
1936  * a process at the time of a crash.  We walk the list and try to delete these
1937  * inodes at recovery time (only with a read-write filesystem).
1938  *
1939  * In order to keep the orphan inode chain consistent during traversal (in
1940  * case of crash during recovery), we link each inode into the superblock
1941  * orphan list_head and handle it the same way as an inode deletion during
1942  * normal operation (which journals the operations for us).
1943  *
1944  * We only do an iget() and an iput() on each inode, which is very safe if we
1945  * accidentally point at an in-use or already deleted inode.  The worst that
1946  * can happen in this case is that we get a "bit already cleared" message from
1947  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1948  * e2fsck was run on this filesystem, and it must have already done the orphan
1949  * inode cleanup for us, so we can safely abort without any further action.
1950  */
1951 static void ext4_orphan_cleanup(struct super_block *sb,
1952                                 struct ext4_super_block *es)
1953 {
1954         unsigned int s_flags = sb->s_flags;
1955         int nr_orphans = 0, nr_truncates = 0;
1956 #ifdef CONFIG_QUOTA
1957         int i;
1958 #endif
1959         if (!es->s_last_orphan) {
1960                 jbd_debug(4, "no orphan inodes to clean up\n");
1961                 return;
1962         }
1963
1964         if (bdev_read_only(sb->s_bdev)) {
1965                 ext4_msg(sb, KERN_ERR, "write access "
1966                         "unavailable, skipping orphan cleanup");
1967                 return;
1968         }
1969
1970         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1971                 if (es->s_last_orphan)
1972                         jbd_debug(1, "Errors on filesystem, "
1973                                   "clearing orphan list.\n");
1974                 es->s_last_orphan = 0;
1975                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1976                 return;
1977         }
1978
1979         if (s_flags & MS_RDONLY) {
1980                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1981                 sb->s_flags &= ~MS_RDONLY;
1982         }
1983 #ifdef CONFIG_QUOTA
1984         /* Needed for iput() to work correctly and not trash data */
1985         sb->s_flags |= MS_ACTIVE;
1986         /* Turn on quotas so that they are updated correctly */
1987         for (i = 0; i < MAXQUOTAS; i++) {
1988                 if (EXT4_SB(sb)->s_qf_names[i]) {
1989                         int ret = ext4_quota_on_mount(sb, i);
1990                         if (ret < 0)
1991                                 ext4_msg(sb, KERN_ERR,
1992                                         "Cannot turn on journaled "
1993                                         "quota: error %d", ret);
1994                 }
1995         }
1996 #endif
1997
1998         while (es->s_last_orphan) {
1999                 struct inode *inode;
2000
2001                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2002                 if (IS_ERR(inode)) {
2003                         es->s_last_orphan = 0;
2004                         break;
2005                 }
2006
2007                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2008                 dquot_initialize(inode);
2009                 if (inode->i_nlink) {
2010                         ext4_msg(sb, KERN_DEBUG,
2011                                 "%s: truncating inode %lu to %lld bytes",
2012                                 __func__, inode->i_ino, inode->i_size);
2013                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2014                                   inode->i_ino, inode->i_size);
2015                         ext4_truncate(inode);
2016                         nr_truncates++;
2017                 } else {
2018                         ext4_msg(sb, KERN_DEBUG,
2019                                 "%s: deleting unreferenced inode %lu",
2020                                 __func__, inode->i_ino);
2021                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2022                                   inode->i_ino);
2023                         nr_orphans++;
2024                 }
2025                 iput(inode);  /* The delete magic happens here! */
2026         }
2027
2028 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2029
2030         if (nr_orphans)
2031                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2032                        PLURAL(nr_orphans));
2033         if (nr_truncates)
2034                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2035                        PLURAL(nr_truncates));
2036 #ifdef CONFIG_QUOTA
2037         /* Turn quotas off */
2038         for (i = 0; i < MAXQUOTAS; i++) {
2039                 if (sb_dqopt(sb)->files[i])
2040                         vfs_quota_off(sb, i, 0);
2041         }
2042 #endif
2043         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2044 }
2045
2046 /*
2047  * Maximal extent format file size.
2048  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2049  * extent format containers, within a sector_t, and within i_blocks
2050  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2051  * so that won't be a limiting factor.
2052  *
2053  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2054  */
2055 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2056 {
2057         loff_t res;
2058         loff_t upper_limit = MAX_LFS_FILESIZE;
2059
2060         /* small i_blocks in vfs inode? */
2061         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2062                 /*
2063                  * CONFIG_LBDAF is not enabled implies the inode
2064                  * i_block represent total blocks in 512 bytes
2065                  * 32 == size of vfs inode i_blocks * 8
2066                  */
2067                 upper_limit = (1LL << 32) - 1;
2068
2069                 /* total blocks in file system block size */
2070                 upper_limit >>= (blkbits - 9);
2071                 upper_limit <<= blkbits;
2072         }
2073
2074         /* 32-bit extent-start container, ee_block */
2075         res = 1LL << 32;
2076         res <<= blkbits;
2077         res -= 1;
2078
2079         /* Sanity check against vm- & vfs- imposed limits */
2080         if (res > upper_limit)
2081                 res = upper_limit;
2082
2083         return res;
2084 }
2085
2086 /*
2087  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2088  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2089  * We need to be 1 filesystem block less than the 2^48 sector limit.
2090  */
2091 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2092 {
2093         loff_t res = EXT4_NDIR_BLOCKS;
2094         int meta_blocks;
2095         loff_t upper_limit;
2096         /* This is calculated to be the largest file size for a dense, block
2097          * mapped file such that the file's total number of 512-byte sectors,
2098          * including data and all indirect blocks, does not exceed (2^48 - 1).
2099          *
2100          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2101          * number of 512-byte sectors of the file.
2102          */
2103
2104         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2105                 /*
2106                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2107                  * the inode i_block field represents total file blocks in
2108                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2109                  */
2110                 upper_limit = (1LL << 32) - 1;
2111
2112                 /* total blocks in file system block size */
2113                 upper_limit >>= (bits - 9);
2114
2115         } else {
2116                 /*
2117                  * We use 48 bit ext4_inode i_blocks
2118                  * With EXT4_HUGE_FILE_FL set the i_blocks
2119                  * represent total number of blocks in
2120                  * file system block size
2121                  */
2122                 upper_limit = (1LL << 48) - 1;
2123
2124         }
2125
2126         /* indirect blocks */
2127         meta_blocks = 1;
2128         /* double indirect blocks */
2129         meta_blocks += 1 + (1LL << (bits-2));
2130         /* tripple indirect blocks */
2131         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2132
2133         upper_limit -= meta_blocks;
2134         upper_limit <<= bits;
2135
2136         res += 1LL << (bits-2);
2137         res += 1LL << (2*(bits-2));
2138         res += 1LL << (3*(bits-2));
2139         res <<= bits;
2140         if (res > upper_limit)
2141                 res = upper_limit;
2142
2143         if (res > MAX_LFS_FILESIZE)
2144                 res = MAX_LFS_FILESIZE;
2145
2146         return res;
2147 }
2148
2149 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2150                                    ext4_fsblk_t logical_sb_block, int nr)
2151 {
2152         struct ext4_sb_info *sbi = EXT4_SB(sb);
2153         ext4_group_t bg, first_meta_bg;
2154         int has_super = 0;
2155
2156         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2157
2158         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2159             nr < first_meta_bg)
2160                 return logical_sb_block + nr + 1;
2161         bg = sbi->s_desc_per_block * nr;
2162         if (ext4_bg_has_super(sb, bg))
2163                 has_super = 1;
2164
2165         return (has_super + ext4_group_first_block_no(sb, bg));
2166 }
2167
2168 /**
2169  * ext4_get_stripe_size: Get the stripe size.
2170  * @sbi: In memory super block info
2171  *
2172  * If we have specified it via mount option, then
2173  * use the mount option value. If the value specified at mount time is
2174  * greater than the blocks per group use the super block value.
2175  * If the super block value is greater than blocks per group return 0.
2176  * Allocator needs it be less than blocks per group.
2177  *
2178  */
2179 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2180 {
2181         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2182         unsigned long stripe_width =
2183                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2184
2185         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2186                 return sbi->s_stripe;
2187
2188         if (stripe_width <= sbi->s_blocks_per_group)
2189                 return stripe_width;
2190
2191         if (stride <= sbi->s_blocks_per_group)
2192                 return stride;
2193
2194         return 0;
2195 }
2196
2197 /* sysfs supprt */
2198
2199 struct ext4_attr {
2200         struct attribute attr;
2201         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2202         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *, 
2203                          const char *, size_t);
2204         int offset;
2205 };
2206
2207 static int parse_strtoul(const char *buf,
2208                 unsigned long max, unsigned long *value)
2209 {
2210         char *endp;
2211
2212         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2213         endp = skip_spaces(endp);
2214         if (*endp || *value > max)
2215                 return -EINVAL;
2216
2217         return 0;
2218 }
2219
2220 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2221                                               struct ext4_sb_info *sbi,
2222                                               char *buf)
2223 {
2224         return snprintf(buf, PAGE_SIZE, "%llu\n",
2225                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2226 }
2227
2228 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2229                                          struct ext4_sb_info *sbi, char *buf)
2230 {
2231         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2232
2233         return snprintf(buf, PAGE_SIZE, "%lu\n",
2234                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2235                          sbi->s_sectors_written_start) >> 1);
2236 }
2237
2238 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2239                                           struct ext4_sb_info *sbi, char *buf)
2240 {
2241         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2242
2243         return snprintf(buf, PAGE_SIZE, "%llu\n",
2244                         (unsigned long long)(sbi->s_kbytes_written +
2245                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2246                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2247 }
2248
2249 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2250                                           struct ext4_sb_info *sbi,
2251                                           const char *buf, size_t count)
2252 {
2253         unsigned long t;
2254
2255         if (parse_strtoul(buf, 0x40000000, &t))
2256                 return -EINVAL;
2257
2258         if (!is_power_of_2(t))
2259                 return -EINVAL;
2260
2261         sbi->s_inode_readahead_blks = t;
2262         return count;
2263 }
2264
2265 static ssize_t sbi_ui_show(struct ext4_attr *a,
2266                            struct ext4_sb_info *sbi, char *buf)
2267 {
2268         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2269
2270         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2271 }
2272
2273 static ssize_t sbi_ui_store(struct ext4_attr *a,
2274                             struct ext4_sb_info *sbi,
2275                             const char *buf, size_t count)
2276 {
2277         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2278         unsigned long t;
2279
2280         if (parse_strtoul(buf, 0xffffffff, &t))
2281                 return -EINVAL;
2282         *ui = t;
2283         return count;
2284 }
2285
2286 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2287 static struct ext4_attr ext4_attr_##_name = {                   \
2288         .attr = {.name = __stringify(_name), .mode = _mode },   \
2289         .show   = _show,                                        \
2290         .store  = _store,                                       \
2291         .offset = offsetof(struct ext4_sb_info, _elname),       \
2292 }
2293 #define EXT4_ATTR(name, mode, show, store) \
2294 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2295
2296 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2297 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2298 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2299         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2300 #define ATTR_LIST(name) &ext4_attr_##name.attr
2301
2302 EXT4_RO_ATTR(delayed_allocation_blocks);
2303 EXT4_RO_ATTR(session_write_kbytes);
2304 EXT4_RO_ATTR(lifetime_write_kbytes);
2305 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2306                  inode_readahead_blks_store, s_inode_readahead_blks);
2307 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2308 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2309 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2310 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2311 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2312 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2313 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2314 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2315
2316 static struct attribute *ext4_attrs[] = {
2317         ATTR_LIST(delayed_allocation_blocks),
2318         ATTR_LIST(session_write_kbytes),
2319         ATTR_LIST(lifetime_write_kbytes),
2320         ATTR_LIST(inode_readahead_blks),
2321         ATTR_LIST(inode_goal),
2322         ATTR_LIST(mb_stats),
2323         ATTR_LIST(mb_max_to_scan),
2324         ATTR_LIST(mb_min_to_scan),
2325         ATTR_LIST(mb_order2_req),
2326         ATTR_LIST(mb_stream_req),
2327         ATTR_LIST(mb_group_prealloc),
2328         ATTR_LIST(max_writeback_mb_bump),
2329         NULL,
2330 };
2331
2332 static ssize_t ext4_attr_show(struct kobject *kobj,
2333                               struct attribute *attr, char *buf)
2334 {
2335         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2336                                                 s_kobj);
2337         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2338
2339         return a->show ? a->show(a, sbi, buf) : 0;
2340 }
2341
2342 static ssize_t ext4_attr_store(struct kobject *kobj,
2343                                struct attribute *attr,
2344                                const char *buf, size_t len)
2345 {
2346         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2347                                                 s_kobj);
2348         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2349
2350         return a->store ? a->store(a, sbi, buf, len) : 0;
2351 }
2352
2353 static void ext4_sb_release(struct kobject *kobj)
2354 {
2355         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2356                                                 s_kobj);
2357         complete(&sbi->s_kobj_unregister);
2358 }
2359
2360
2361 static const struct sysfs_ops ext4_attr_ops = {
2362         .show   = ext4_attr_show,
2363         .store  = ext4_attr_store,
2364 };
2365
2366 static struct kobj_type ext4_ktype = {
2367         .default_attrs  = ext4_attrs,
2368         .sysfs_ops      = &ext4_attr_ops,
2369         .release        = ext4_sb_release,
2370 };
2371
2372 /*
2373  * Check whether this filesystem can be mounted based on
2374  * the features present and the RDONLY/RDWR mount requested.
2375  * Returns 1 if this filesystem can be mounted as requested,
2376  * 0 if it cannot be.
2377  */
2378 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2379 {
2380         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2381                 ext4_msg(sb, KERN_ERR,
2382                         "Couldn't mount because of "
2383                         "unsupported optional features (%x)",
2384                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2385                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2386                 return 0;
2387         }
2388
2389         if (readonly)
2390                 return 1;
2391
2392         /* Check that feature set is OK for a read-write mount */
2393         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2394                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2395                          "unsupported optional features (%x)",
2396                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2397                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2398                 return 0;
2399         }
2400         /*
2401          * Large file size enabled file system can only be mounted
2402          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2403          */
2404         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2405                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2406                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2407                                  "cannot be mounted RDWR without "
2408                                  "CONFIG_LBDAF");
2409                         return 0;
2410                 }
2411         }
2412         return 1;
2413 }
2414
2415 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2416                                 __releases(kernel_lock)
2417                                 __acquires(kernel_lock)
2418 {
2419         struct buffer_head *bh;
2420         struct ext4_super_block *es = NULL;
2421         struct ext4_sb_info *sbi;
2422         ext4_fsblk_t block;
2423         ext4_fsblk_t sb_block = get_sb_block(&data);
2424         ext4_fsblk_t logical_sb_block;
2425         unsigned long offset = 0;
2426         unsigned long journal_devnum = 0;
2427         unsigned long def_mount_opts;
2428         struct inode *root;
2429         char *cp;
2430         const char *descr;
2431         int ret = -EINVAL;
2432         int blocksize;
2433         unsigned int db_count;
2434         unsigned int i;
2435         int needs_recovery, has_huge_files;
2436         __u64 blocks_count;
2437         int err;
2438         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2439
2440         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2441         if (!sbi)
2442                 return -ENOMEM;
2443
2444         sbi->s_blockgroup_lock =
2445                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
2446         if (!sbi->s_blockgroup_lock) {
2447                 kfree(sbi);
2448                 return -ENOMEM;
2449         }
2450         sb->s_fs_info = sbi;
2451         sbi->s_mount_opt = 0;
2452         sbi->s_resuid = EXT4_DEF_RESUID;
2453         sbi->s_resgid = EXT4_DEF_RESGID;
2454         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
2455         sbi->s_sb_block = sb_block;
2456         sbi->s_sectors_written_start = part_stat_read(sb->s_bdev->bd_part,
2457                                                       sectors[1]);
2458
2459         unlock_kernel();
2460
2461         /* Cleanup superblock name */
2462         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
2463                 *cp = '!';
2464
2465         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
2466         if (!blocksize) {
2467                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
2468                 goto out_fail;
2469         }
2470
2471         /*
2472          * The ext4 superblock will not be buffer aligned for other than 1kB
2473          * block sizes.  We need to calculate the offset from buffer start.
2474          */
2475         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
2476                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2477                 offset = do_div(logical_sb_block, blocksize);
2478         } else {
2479                 logical_sb_block = sb_block;
2480         }
2481
2482         if (!(bh = sb_bread(sb, logical_sb_block))) {
2483                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
2484                 goto out_fail;
2485         }
2486         /*
2487          * Note: s_es must be initialized as soon as possible because
2488          *       some ext4 macro-instructions depend on its value
2489          */
2490         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2491         sbi->s_es = es;
2492         sb->s_magic = le16_to_cpu(es->s_magic);
2493         if (sb->s_magic != EXT4_SUPER_MAGIC)
2494                 goto cantfind_ext4;
2495         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
2496
2497         /* Set defaults before we parse the mount options */
2498         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
2499         if (def_mount_opts & EXT4_DEFM_DEBUG)
2500                 set_opt(sbi->s_mount_opt, DEBUG);
2501         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
2502                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
2503                         "2.6.38");
2504                 set_opt(sbi->s_mount_opt, GRPID);
2505         }
2506         if (def_mount_opts & EXT4_DEFM_UID16)
2507                 set_opt(sbi->s_mount_opt, NO_UID32);
2508 #ifdef CONFIG_EXT4_FS_XATTR
2509         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
2510                 set_opt(sbi->s_mount_opt, XATTR_USER);
2511 #endif
2512 #ifdef CONFIG_EXT4_FS_POSIX_ACL
2513         if (def_mount_opts & EXT4_DEFM_ACL)
2514                 set_opt(sbi->s_mount_opt, POSIX_ACL);
2515 #endif
2516         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
2517                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2518         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
2519                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
2520         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
2521                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
2522
2523         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
2524                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
2525         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
2526                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
2527         else
2528                 set_opt(sbi->s_mount_opt, ERRORS_RO);
2529
2530         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
2531         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
2532         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
2533         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
2534         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
2535
2536         set_opt(sbi->s_mount_opt, BARRIER);
2537
2538         /*
2539          * enable delayed allocation by default
2540          * Use -o nodelalloc to turn it off
2541          */
2542         set_opt(sbi->s_mount_opt, DELALLOC);
2543
2544         if (!parse_options((char *) data, sb, &journal_devnum,
2545                            &journal_ioprio, NULL, 0))
2546                 goto failed_mount;
2547
2548         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2549                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2550
2551         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
2552             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
2553              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
2554              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
2555                 ext4_msg(sb, KERN_WARNING,
2556                        "feature flags set on rev 0 fs, "
2557                        "running e2fsck is recommended");
2558
2559         /*
2560          * Check feature flags regardless of the revision level, since we
2561          * previously didn't change the revision level when setting the flags,
2562          * so there is a chance incompat flags are set on a rev 0 filesystem.
2563          */
2564         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
2565                 goto failed_mount;
2566
2567         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
2568
2569         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
2570             blocksize > EXT4_MAX_BLOCK_SIZE) {
2571                 ext4_msg(sb, KERN_ERR,
2572                        "Unsupported filesystem blocksize %d", blocksize);
2573                 goto failed_mount;
2574         }
2575
2576         if (sb->s_blocksize != blocksize) {
2577                 /* Validate the filesystem blocksize */
2578                 if (!sb_set_blocksize(sb, blocksize)) {
2579                         ext4_msg(sb, KERN_ERR, "bad block size %d",
2580                                         blocksize);
2581                         goto failed_mount;
2582                 }
2583
2584                 brelse(bh);
2585                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2586                 offset = do_div(logical_sb_block, blocksize);
2587                 bh = sb_bread(sb, logical_sb_block);
2588                 if (!bh) {
2589                         ext4_msg(sb, KERN_ERR,
2590                                "Can't read superblock on 2nd try");
2591                         goto failed_mount;
2592                 }
2593                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2594                 sbi->s_es = es;
2595                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2596                         ext4_msg(sb, KERN_ERR,
2597                                "Magic mismatch, very weird!");
2598                         goto failed_mount;
2599                 }
2600         }
2601
2602         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2603                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
2604         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
2605                                                       has_huge_files);
2606         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
2607
2608         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2609                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2610                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2611         } else {
2612                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2613                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2614                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2615                     (!is_power_of_2(sbi->s_inode_size)) ||
2616                     (sbi->s_inode_size > blocksize)) {
2617                         ext4_msg(sb, KERN_ERR,
2618                                "unsupported inode size: %d",
2619                                sbi->s_inode_size);
2620                         goto failed_mount;
2621                 }
2622                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2623                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2624         }
2625
2626         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2627         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2628                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2629                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2630                     !is_power_of_2(sbi->s_desc_size)) {
2631                         ext4_msg(sb, KERN_ERR,
2632                                "unsupported descriptor size %lu",
2633                                sbi->s_desc_size);
2634                         goto failed_mount;
2635                 }
2636         } else
2637                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2638
2639         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2640         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2641         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2642                 goto cantfind_ext4;
2643
2644         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2645         if (sbi->s_inodes_per_block == 0)
2646                 goto cantfind_ext4;
2647         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2648                                         sbi->s_inodes_per_block;
2649         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2650         sbi->s_sbh = bh;
2651         sbi->s_mount_state = le16_to_cpu(es->s_state);
2652         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2653         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2654
2655         for (i = 0; i < 4; i++)
2656                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2657         sbi->s_def_hash_version = es->s_def_hash_version;
2658         i = le32_to_cpu(es->s_flags);
2659         if (i & EXT2_FLAGS_UNSIGNED_HASH)
2660                 sbi->s_hash_unsigned = 3;
2661         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
2662 #ifdef __CHAR_UNSIGNED__
2663                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
2664                 sbi->s_hash_unsigned = 3;
2665 #else
2666                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
2667 #endif
2668                 sb->s_dirt = 1;
2669         }
2670
2671         if (sbi->s_blocks_per_group > blocksize * 8) {
2672                 ext4_msg(sb, KERN_ERR,
2673                        "#blocks per group too big: %lu",
2674                        sbi->s_blocks_per_group);
2675                 goto failed_mount;
2676         }
2677         if (sbi->s_inodes_per_group > blocksize * 8) {
2678                 ext4_msg(sb, KERN_ERR,
2679                        "#inodes per group too big: %lu",
2680                        sbi->s_inodes_per_group);
2681                 goto failed_mount;
2682         }
2683
2684         /*
2685          * Test whether we have more sectors than will fit in sector_t,
2686          * and whether the max offset is addressable by the page cache.
2687          */
2688         if ((ext4_blocks_count(es) >
2689              (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) ||
2690             (ext4_blocks_count(es) >
2691              (pgoff_t)(~0ULL) >> (PAGE_CACHE_SHIFT - sb->s_blocksize_bits))) {
2692                 ext4_msg(sb, KERN_ERR, "filesystem"
2693                          " too large to mount safely on this system");
2694                 if (sizeof(sector_t) < 8)
2695                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
2696                 ret = -EFBIG;
2697                 goto failed_mount;
2698         }
2699
2700         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2701                 goto cantfind_ext4;
2702
2703         /* check blocks count against device size */
2704         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
2705         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
2706                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
2707                        "exceeds size of device (%llu blocks)",
2708                        ext4_blocks_count(es), blocks_count);
2709                 goto failed_mount;
2710         }
2711
2712         /*
2713          * It makes no sense for the first data block to be beyond the end
2714          * of the filesystem.
2715          */
2716         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
2717                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
2718                          "block %u is beyond end of filesystem (%llu)",
2719                          le32_to_cpu(es->s_first_data_block),
2720                          ext4_blocks_count(es));
2721                 goto failed_mount;
2722         }
2723         blocks_count = (ext4_blocks_count(es) -
2724                         le32_to_cpu(es->s_first_data_block) +
2725                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2726         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2727         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
2728                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
2729                        "(block count %llu, first data block %u, "
2730                        "blocks per group %lu)", sbi->s_groups_count,
2731                        ext4_blocks_count(es),
2732                        le32_to_cpu(es->s_first_data_block),
2733                        EXT4_BLOCKS_PER_GROUP(sb));
2734                 goto failed_mount;
2735         }
2736         sbi->s_groups_count = blocks_count;
2737         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
2738                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
2739         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2740                    EXT4_DESC_PER_BLOCK(sb);
2741         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
2742                                     GFP_KERNEL);
2743         if (sbi->s_group_desc == NULL) {
2744                 ext4_msg(sb, KERN_ERR, "not enough memory");
2745                 goto failed_mount;
2746         }
2747
2748 #ifdef CONFIG_PROC_FS
2749         if (ext4_proc_root)
2750                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
2751 #endif
2752
2753         bgl_lock_init(sbi->s_blockgroup_lock);
2754
2755         for (i = 0; i < db_count; i++) {
2756                 block = descriptor_loc(sb, logical_sb_block, i);
2757                 sbi->s_group_desc[i] = sb_bread(sb, block);
2758                 if (!sbi->s_group_desc[i]) {
2759                         ext4_msg(sb, KERN_ERR,
2760                                "can't read group descriptor %d", i);
2761                         db_count = i;
2762                         goto failed_mount2;
2763                 }
2764         }
2765         if (!ext4_check_descriptors(sb)) {
2766                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
2767                 goto failed_mount2;
2768         }
2769         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2770                 if (!ext4_fill_flex_info(sb)) {
2771                         ext4_msg(sb, KERN_ERR,
2772                                "unable to initialize "
2773                                "flex_bg meta info!");
2774                         goto failed_mount2;
2775                 }
2776
2777         sbi->s_gdb_count = db_count;
2778         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2779         spin_lock_init(&sbi->s_next_gen_lock);
2780
2781         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2782                         ext4_count_free_blocks(sb));
2783         if (!err) {
2784                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2785                                 ext4_count_free_inodes(sb));
2786         }
2787         if (!err) {
2788                 err = percpu_counter_init(&sbi->s_dirs_counter,
2789                                 ext4_count_dirs(sb));
2790         }
2791         if (!err) {
2792                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
2793         }
2794         if (err) {
2795                 ext4_msg(sb, KERN_ERR, "insufficient memory");
2796                 goto failed_mount3;
2797         }
2798
2799         sbi->s_stripe = ext4_get_stripe_size(sbi);
2800         sbi->s_max_writeback_mb_bump = 128;
2801
2802         /*
2803          * set up enough so that it can read an inode
2804          */
2805         if (!test_opt(sb, NOLOAD) &&
2806             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
2807                 sb->s_op = &ext4_sops;
2808         else
2809                 sb->s_op = &ext4_nojournal_sops;
2810         sb->s_export_op = &ext4_export_ops;
2811         sb->s_xattr = ext4_xattr_handlers;
2812 #ifdef CONFIG_QUOTA
2813         sb->s_qcop = &ext4_qctl_operations;
2814         sb->dq_op = &ext4_quota_operations;
2815 #endif
2816         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2817         mutex_init(&sbi->s_orphan_lock);
2818         mutex_init(&sbi->s_resize_lock);
2819
2820         sb->s_root = NULL;
2821
2822         needs_recovery = (es->s_last_orphan != 0 ||
2823                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2824                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2825
2826         /*
2827          * The first inode we look at is the journal inode.  Don't try
2828          * root first: it may be modified in the journal!
2829          */
2830         if (!test_opt(sb, NOLOAD) &&
2831             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2832                 if (ext4_load_journal(sb, es, journal_devnum))
2833                         goto failed_mount3;
2834         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
2835               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2836                 ext4_msg(sb, KERN_ERR, "required journal recovery "
2837                        "suppressed and not mounted read-only");
2838                 goto failed_mount_wq;
2839         } else {
2840                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
2841                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
2842                 sbi->s_journal = NULL;
2843                 needs_recovery = 0;
2844                 goto no_journal;
2845         }
2846
2847         if (ext4_blocks_count(es) > 0xffffffffULL &&
2848             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2849                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2850                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
2851                 goto failed_mount_wq;
2852         }
2853
2854         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2855                 jbd2_journal_set_features(sbi->s_journal,
2856                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2857                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2858         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2859                 jbd2_journal_set_features(sbi->s_journal,
2860                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2861                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2862                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2863         } else {
2864                 jbd2_journal_clear_features(sbi->s_journal,
2865                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2866                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2867         }
2868
2869         /* We have now updated the journal if required, so we can
2870          * validate the data journaling mode. */
2871         switch (test_opt(sb, DATA_FLAGS)) {
2872         case 0:
2873                 /* No mode set, assume a default based on the journal
2874                  * capabilities: ORDERED_DATA if the journal can
2875                  * cope, else JOURNAL_DATA
2876                  */
2877                 if (jbd2_journal_check_available_features
2878                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2879                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2880                 else
2881                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2882                 break;
2883
2884         case EXT4_MOUNT_ORDERED_DATA:
2885         case EXT4_MOUNT_WRITEBACK_DATA:
2886                 if (!jbd2_journal_check_available_features
2887                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2888                         ext4_msg(sb, KERN_ERR, "Journal does not support "
2889                                "requested data journaling mode");
2890                         goto failed_mount_wq;
2891                 }
2892         default:
2893                 break;
2894         }
2895         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
2896
2897 no_journal:
2898         if (test_opt(sb, NOBH)) {
2899                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2900                         ext4_msg(sb, KERN_WARNING, "Ignoring nobh option - "
2901                                 "its supported only with writeback mode");
2902                         clear_opt(sbi->s_mount_opt, NOBH);
2903                 }
2904                 if (test_opt(sb, DIOREAD_NOLOCK)) {
2905                         ext4_msg(sb, KERN_WARNING, "dioread_nolock option is "
2906                                 "not supported with nobh mode");
2907                         goto failed_mount_wq;
2908                 }
2909         }
2910         EXT4_SB(sb)->dio_unwritten_wq = create_workqueue("ext4-dio-unwritten");
2911         if (!EXT4_SB(sb)->dio_unwritten_wq) {
2912                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
2913                 goto failed_mount_wq;
2914         }
2915
2916         /*
2917          * The jbd2_journal_load will have done any necessary log recovery,
2918          * so we can safely mount the rest of the filesystem now.
2919          */
2920
2921         root = ext4_iget(sb, EXT4_ROOT_INO);
2922         if (IS_ERR(root)) {
2923                 ext4_msg(sb, KERN_ERR, "get root inode failed");
2924                 ret = PTR_ERR(root);
2925                 goto failed_mount4;
2926         }
2927         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2928                 iput(root);
2929                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
2930                 goto failed_mount4;
2931         }
2932         sb->s_root = d_alloc_root(root);
2933         if (!sb->s_root) {
2934                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
2935                 iput(root);
2936                 ret = -ENOMEM;
2937                 goto failed_mount4;
2938         }
2939
2940         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
2941
2942         /* determine the minimum size of new large inodes, if present */
2943         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2944                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2945                                                      EXT4_GOOD_OLD_INODE_SIZE;
2946                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2947                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2948                         if (sbi->s_want_extra_isize <
2949                             le16_to_cpu(es->s_want_extra_isize))
2950                                 sbi->s_want_extra_isize =
2951                                         le16_to_cpu(es->s_want_extra_isize);
2952                         if (sbi->s_want_extra_isize <
2953                             le16_to_cpu(es->s_min_extra_isize))
2954                                 sbi->s_want_extra_isize =
2955                                         le16_to_cpu(es->s_min_extra_isize);
2956                 }
2957         }
2958         /* Check if enough inode space is available */
2959         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2960                                                         sbi->s_inode_size) {
2961                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2962                                                        EXT4_GOOD_OLD_INODE_SIZE;
2963                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
2964                          "available");
2965         }
2966
2967         if (test_opt(sb, DELALLOC) &&
2968             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
2969                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
2970                          "requested data journaling mode");
2971                 clear_opt(sbi->s_mount_opt, DELALLOC);
2972         }
2973         if (test_opt(sb, DIOREAD_NOLOCK)) {
2974                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
2975                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
2976                                 "option - requested data journaling mode");
2977                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
2978                 }
2979                 if (sb->s_blocksize < PAGE_SIZE) {
2980                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
2981                                 "option - block size is too small");
2982                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
2983                 }
2984         }
2985
2986         err = ext4_setup_system_zone(sb);
2987         if (err) {
2988                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
2989                          "zone (%d)\n", err);
2990                 goto failed_mount4;
2991         }
2992
2993         ext4_ext_init(sb);
2994         err = ext4_mb_init(sb, needs_recovery);
2995         if (err) {
2996                 ext4_msg(sb, KERN_ERR, "failed to initalize mballoc (%d)",
2997                          err);
2998                 goto failed_mount4;
2999         }
3000
3001         sbi->s_kobj.kset = ext4_kset;
3002         init_completion(&sbi->s_kobj_unregister);
3003         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3004                                    "%s", sb->s_id);
3005         if (err) {
3006                 ext4_mb_release(sb);
3007                 ext4_ext_release(sb);
3008                 goto failed_mount4;
3009         };
3010
3011         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3012         ext4_orphan_cleanup(sb, es);
3013         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3014         if (needs_recovery) {
3015                 ext4_msg(sb, KERN_INFO, "recovery complete");
3016                 ext4_mark_recovery_complete(sb, es);
3017         }
3018         if (EXT4_SB(sb)->s_journal) {
3019                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3020                         descr = " journalled data mode";
3021                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3022                         descr = " ordered data mode";
3023                 else
3024                         descr = " writeback data mode";
3025         } else
3026                 descr = "out journal";
3027
3028         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s", descr);
3029
3030         lock_kernel();
3031         return 0;
3032
3033 cantfind_ext4:
3034         if (!silent)
3035                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3036         goto failed_mount;
3037
3038 failed_mount4:
3039         ext4_msg(sb, KERN_ERR, "mount failed");
3040         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3041 failed_mount_wq:
3042         ext4_release_system_zone(sb);
3043         if (sbi->s_journal) {
3044                 jbd2_journal_destroy(sbi->s_journal);
3045                 sbi->s_journal = NULL;
3046         }
3047 failed_mount3:
3048         if (sbi->s_flex_groups) {
3049                 if (is_vmalloc_addr(sbi->s_flex_groups))
3050                         vfree(sbi->s_flex_groups);
3051                 else
3052                         kfree(sbi->s_flex_groups);
3053         }
3054         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3055         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3056         percpu_counter_destroy(&sbi->s_dirs_counter);
3057         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3058 failed_mount2:
3059         for (i = 0; i < db_count; i++)
3060                 brelse(sbi->s_group_desc[i]);
3061         kfree(sbi->s_group_desc);
3062 failed_mount:
3063         if (sbi->s_proc) {
3064                 remove_proc_entry(sb->s_id, ext4_proc_root);
3065         }
3066 #ifdef CONFIG_QUOTA
3067         for (i = 0; i < MAXQUOTAS; i++)
3068                 kfree(sbi->s_qf_names[i]);
3069 #endif
3070         ext4_blkdev_remove(sbi);
3071         brelse(bh);
3072 out_fail:
3073         sb->s_fs_info = NULL;
3074         kfree(sbi->s_blockgroup_lock);
3075         kfree(sbi);
3076         lock_kernel();
3077         return ret;
3078 }
3079
3080 /*
3081  * Setup any per-fs journal parameters now.  We'll do this both on
3082  * initial mount, once the journal has been initialised but before we've
3083  * done any recovery; and again on any subsequent remount.
3084  */
3085 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3086 {
3087         struct ext4_sb_info *sbi = EXT4_SB(sb);
3088
3089         journal->j_commit_interval = sbi->s_commit_interval;
3090         journal->j_min_batch_time = sbi->s_min_batch_time;
3091         journal->j_max_batch_time = sbi->s_max_batch_time;
3092
3093         spin_lock(&journal->j_state_lock);
3094         if (test_opt(sb, BARRIER))
3095                 journal->j_flags |= JBD2_BARRIER;
3096         else
3097                 journal->j_flags &= ~JBD2_BARRIER;
3098         if (test_opt(sb, DATA_ERR_ABORT))
3099                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3100         else
3101                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3102         spin_unlock(&journal->j_state_lock);
3103 }
3104
3105 static journal_t *ext4_get_journal(struct super_block *sb,
3106                                    unsigned int journal_inum)
3107 {
3108         struct inode *journal_inode;
3109         journal_t *journal;
3110
3111         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3112
3113         /* First, test for the existence of a valid inode on disk.  Bad
3114          * things happen if we iget() an unused inode, as the subsequent
3115          * iput() will try to delete it. */
3116
3117         journal_inode = ext4_iget(sb, journal_inum);
3118         if (IS_ERR(journal_inode)) {
3119                 ext4_msg(sb, KERN_ERR, "no journal found");
3120                 return NULL;
3121         }
3122         if (!journal_inode->i_nlink) {
3123                 make_bad_inode(journal_inode);
3124                 iput(journal_inode);
3125                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3126                 return NULL;
3127         }
3128
3129         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3130                   journal_inode, journal_inode->i_size);
3131         if (!S_ISREG(journal_inode->i_mode)) {
3132                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3133                 iput(journal_inode);
3134                 return NULL;
3135         }
3136
3137         journal = jbd2_journal_init_inode(journal_inode);
3138         if (!journal) {
3139                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3140                 iput(journal_inode);
3141                 return NULL;
3142         }
3143         journal->j_private = sb;
3144         ext4_init_journal_params(sb, journal);
3145         return journal;
3146 }
3147
3148 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3149                                        dev_t j_dev)
3150 {
3151         struct buffer_head *bh;
3152         journal_t *journal;
3153         ext4_fsblk_t start;
3154         ext4_fsblk_t len;
3155         int hblock, blocksize;
3156         ext4_fsblk_t sb_block;
3157         unsigned long offset;
3158         struct ext4_super_block *es;
3159         struct block_device *bdev;
3160
3161         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3162
3163         bdev = ext4_blkdev_get(j_dev, sb);
3164         if (bdev == NULL)
3165                 return NULL;
3166
3167         if (bd_claim(bdev, sb)) {
3168                 ext4_msg(sb, KERN_ERR,
3169                         "failed to claim external journal device");
3170                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
3171                 return NULL;
3172         }
3173
3174         blocksize = sb->s_blocksize;
3175         hblock = bdev_logical_block_size(bdev);
3176         if (blocksize < hblock) {
3177                 ext4_msg(sb, KERN_ERR,
3178                         "blocksize too small for journal device");
3179                 goto out_bdev;
3180         }
3181
3182         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3183         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3184         set_blocksize(bdev, blocksize);
3185         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3186                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3187                        "external journal");
3188                 goto out_bdev;
3189         }
3190
3191         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3192         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3193             !(le32_to_cpu(es->s_feature_incompat) &
3194               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3195                 ext4_msg(sb, KERN_ERR, "external journal has "
3196                                         "bad superblock");
3197                 brelse(bh);
3198                 goto out_bdev;
3199         }
3200
3201         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3202                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3203                 brelse(bh);
3204                 goto out_bdev;
3205         }
3206
3207         len = ext4_blocks_count(es);
3208         start = sb_block + 1;
3209         brelse(bh);     /* we're done with the superblock */
3210
3211         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3212                                         start, len, blocksize);
3213         if (!journal) {
3214                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3215                 goto out_bdev;
3216         }
3217         journal->j_private = sb;
3218         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3219         wait_on_buffer(journal->j_sb_buffer);
3220         if (!buffer_uptodate(journal->j_sb_buffer)) {
3221                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3222                 goto out_journal;
3223         }
3224         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3225                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3226                                         "user (unsupported) - %d",
3227                         be32_to_cpu(journal->j_superblock->s_nr_users));
3228                 goto out_journal;
3229         }
3230         EXT4_SB(sb)->journal_bdev = bdev;
3231         ext4_init_journal_params(sb, journal);
3232         return journal;
3233
3234 out_journal:
3235         jbd2_journal_destroy(journal);
3236 out_bdev:
3237         ext4_blkdev_put(bdev);
3238         return NULL;
3239 }
3240
3241 static int ext4_load_journal(struct super_block *sb,
3242                              struct ext4_super_block *es,
3243                              unsigned long journal_devnum)
3244 {
3245         journal_t *journal;
3246         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3247         dev_t journal_dev;
3248         int err = 0;
3249         int really_read_only;
3250
3251         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3252
3253         if (journal_devnum &&
3254             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3255                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3256                         "numbers have changed");
3257                 journal_dev = new_decode_dev(journal_devnum);
3258         } else
3259                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3260
3261         really_read_only = bdev_read_only(sb->s_bdev);
3262
3263         /*
3264          * Are we loading a blank journal or performing recovery after a
3265          * crash?  For recovery, we need to check in advance whether we
3266          * can get read-write access to the device.
3267          */
3268         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3269                 if (sb->s_flags & MS_RDONLY) {
3270                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3271                                         "required on readonly filesystem");
3272                         if (really_read_only) {
3273                                 ext4_msg(sb, KERN_ERR, "write access "
3274                                         "unavailable, cannot proceed");
3275                                 return -EROFS;
3276                         }
3277                         ext4_msg(sb, KERN_INFO, "write access will "
3278                                "be enabled during recovery");
3279                 }
3280         }
3281
3282         if (journal_inum && journal_dev) {
3283                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3284                        "and inode journals!");
3285                 return -EINVAL;
3286         }
3287
3288         if (journal_inum) {
3289                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3290                         return -EINVAL;
3291         } else {
3292                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3293                         return -EINVAL;
3294         }
3295
3296         if (!(journal->j_flags & JBD2_BARRIER))
3297                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3298
3299         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3300                 err = jbd2_journal_update_format(journal);
3301                 if (err)  {
3302                         ext4_msg(sb, KERN_ERR, "error updating journal");
3303                         jbd2_journal_destroy(journal);
3304                         return err;
3305                 }
3306         }
3307
3308         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3309                 err = jbd2_journal_wipe(journal, !really_read_only);
3310         if (!err)
3311                 err = jbd2_journal_load(journal);
3312
3313         if (err) {
3314                 ext4_msg(sb, KERN_ERR, "error loading journal");
3315                 jbd2_journal_destroy(journal);
3316                 return err;
3317         }
3318
3319         EXT4_SB(sb)->s_journal = journal;
3320         ext4_clear_journal_err(sb, es);
3321
3322         if (journal_devnum &&
3323             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3324                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3325
3326                 /* Make sure we flush the recovery flag to disk. */
3327                 ext4_commit_super(sb, 1);
3328         }
3329
3330         return 0;
3331 }
3332
3333 static int ext4_commit_super(struct super_block *sb, int sync)
3334 {
3335         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3336         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3337         int error = 0;
3338
3339         if (!sbh)
3340                 return error;
3341         if (buffer_write_io_error(sbh)) {
3342                 /*
3343                  * Oh, dear.  A previous attempt to write the
3344                  * superblock failed.  This could happen because the
3345                  * USB device was yanked out.  Or it could happen to
3346                  * be a transient write error and maybe the block will
3347                  * be remapped.  Nothing we can do but to retry the
3348                  * write and hope for the best.
3349                  */
3350                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3351                        "superblock detected");
3352                 clear_buffer_write_io_error(sbh);
3353                 set_buffer_uptodate(sbh);
3354         }
3355         /*
3356          * If the file system is mounted read-only, don't update the
3357          * superblock write time.  This avoids updating the superblock
3358          * write time when we are mounting the root file system
3359          * read/only but we need to replay the journal; at that point,
3360          * for people who are east of GMT and who make their clock
3361          * tick in localtime for Windows bug-for-bug compatibility,
3362          * the clock is set in the future, and this will cause e2fsck
3363          * to complain and force a full file system check.
3364          */
3365         if (!(sb->s_flags & MS_RDONLY))
3366                 es->s_wtime = cpu_to_le32(get_seconds());
3367         es->s_kbytes_written =
3368                 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written + 
3369                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
3370                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
3371         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
3372                                         &EXT4_SB(sb)->s_freeblocks_counter));
3373         es->s_free_inodes_count = cpu_to_le32(percpu_counter_sum_positive(
3374                                         &EXT4_SB(sb)->s_freeinodes_counter));
3375         sb->s_dirt = 0;
3376         BUFFER_TRACE(sbh, "marking dirty");
3377         mark_buffer_dirty(sbh);
3378         if (sync) {
3379                 error = sync_dirty_buffer(sbh);
3380                 if (error)
3381                         return error;
3382
3383                 error = buffer_write_io_error(sbh);
3384                 if (error) {
3385                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
3386                                "superblock");
3387                         clear_buffer_write_io_error(sbh);
3388                         set_buffer_uptodate(sbh);
3389                 }
3390         }
3391         return error;
3392 }
3393
3394 /*
3395  * Have we just finished recovery?  If so, and if we are mounting (or
3396  * remounting) the filesystem readonly, then we will end up with a
3397  * consistent fs on disk.  Record that fact.
3398  */
3399 static void ext4_mark_recovery_complete(struct super_block *sb,
3400                                         struct ext4_super_block *es)
3401 {
3402         journal_t *journal = EXT4_SB(sb)->s_journal;
3403
3404         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3405                 BUG_ON(journal != NULL);
3406                 return;
3407         }
3408         jbd2_journal_lock_updates(journal);
3409         if (jbd2_journal_flush(journal) < 0)
3410                 goto out;
3411
3412         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
3413             sb->s_flags & MS_RDONLY) {
3414                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3415                 ext4_commit_super(sb, 1);
3416         }
3417
3418 out:
3419         jbd2_journal_unlock_updates(journal);
3420 }
3421
3422 /*
3423  * If we are mounting (or read-write remounting) a filesystem whose journal
3424  * has recorded an error from a previous lifetime, move that error to the
3425  * main filesystem now.
3426  */
3427 static void ext4_clear_journal_err(struct super_block *sb,
3428                                    struct ext4_super_block *es)
3429 {
3430         journal_t *journal;
3431         int j_errno;
3432         const char *errstr;
3433
3434         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3435
3436         journal = EXT4_SB(sb)->s_journal;
3437
3438         /*
3439          * Now check for any error status which may have been recorded in the
3440          * journal by a prior ext4_error() or ext4_abort()
3441          */
3442
3443         j_errno = jbd2_journal_errno(journal);
3444         if (j_errno) {
3445                 char nbuf[16];
3446
3447                 errstr = ext4_decode_error(sb, j_errno, nbuf);
3448                 ext4_warning(sb, "Filesystem error recorded "
3449                              "from previous mount: %s", errstr);
3450                 ext4_warning(sb, "Marking fs in need of filesystem check.");
3451
3452                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
3453                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
3454                 ext4_commit_super(sb, 1);
3455
3456                 jbd2_journal_clear_err(journal);
3457         }
3458 }
3459
3460 /*
3461  * Force the running and committing transactions to commit,
3462  * and wait on the commit.
3463  */
3464 int ext4_force_commit(struct super_block *sb)
3465 {
3466         journal_t *journal;
3467         int ret = 0;
3468
3469         if (sb->s_flags & MS_RDONLY)
3470                 return 0;
3471
3472         journal = EXT4_SB(sb)->s_journal;
3473         if (journal)
3474                 ret = ext4_journal_force_commit(journal);
3475
3476         return ret;
3477 }
3478
3479 static void ext4_write_super(struct super_block *sb)
3480 {
3481         lock_super(sb);
3482         ext4_commit_super(sb, 1);
3483         unlock_super(sb);
3484 }
3485
3486 static int ext4_sync_fs(struct super_block *sb, int wait)
3487 {
3488         int ret = 0;
3489         tid_t target;
3490         struct ext4_sb_info *sbi = EXT4_SB(sb);
3491
3492         trace_ext4_sync_fs(sb, wait);
3493         flush_workqueue(sbi->dio_unwritten_wq);
3494         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
3495                 if (wait)
3496                         jbd2_log_wait_commit(sbi->s_journal, target);
3497         }
3498         return ret;
3499 }
3500
3501 /*
3502  * LVM calls this function before a (read-only) snapshot is created.  This
3503  * gives us a chance to flush the journal completely and mark the fs clean.
3504  */
3505 static int ext4_freeze(struct super_block *sb)
3506 {
3507         int error = 0;
3508         journal_t *journal;
3509
3510         if (sb->s_flags & MS_RDONLY)
3511                 return 0;
3512
3513         journal = EXT4_SB(sb)->s_journal;
3514
3515         /* Now we set up the journal barrier. */
3516         jbd2_journal_lock_updates(journal);
3517
3518         /*
3519          * Don't clear the needs_recovery flag if we failed to flush
3520          * the journal.
3521          */
3522         error = jbd2_journal_flush(journal);
3523         if (error < 0) {
3524         out:
3525                 jbd2_journal_unlock_updates(journal);
3526                 return error;
3527         }
3528
3529         /* Journal blocked and flushed, clear needs_recovery flag. */
3530         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3531         error = ext4_commit_super(sb, 1);
3532         if (error)
3533                 goto out;
3534         return 0;
3535 }
3536
3537 /*
3538  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
3539  * flag here, even though the filesystem is not technically dirty yet.
3540  */
3541 static int ext4_unfreeze(struct super_block *sb)
3542 {
3543         if (sb->s_flags & MS_RDONLY)
3544                 return 0;
3545
3546         lock_super(sb);
3547         /* Reset the needs_recovery flag before the fs is unlocked. */
3548         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3549         ext4_commit_super(sb, 1);
3550         unlock_super(sb);
3551         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3552         return 0;
3553 }
3554
3555 static int ext4_remount(struct super_block *sb, int *flags, char *data)
3556 {
3557         struct ext4_super_block *es;
3558         struct ext4_sb_info *sbi = EXT4_SB(sb);
3559         ext4_fsblk_t n_blocks_count = 0;
3560         unsigned long old_sb_flags;
3561         struct ext4_mount_options old_opts;
3562         ext4_group_t g;
3563         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3564         int err;
3565 #ifdef CONFIG_QUOTA
3566         int i;
3567 #endif
3568
3569         lock_kernel();
3570
3571         /* Store the original options */
3572         lock_super(sb);
3573         old_sb_flags = sb->s_flags;
3574         old_opts.s_mount_opt = sbi->s_mount_opt;
3575         old_opts.s_resuid = sbi->s_resuid;
3576         old_opts.s_resgid = sbi->s_resgid;
3577         old_opts.s_commit_interval = sbi->s_commit_interval;
3578         old_opts.s_min_batch_time = sbi->s_min_batch_time;
3579         old_opts.s_max_batch_time = sbi->s_max_batch_time;
3580 #ifdef CONFIG_QUOTA
3581         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
3582         for (i = 0; i < MAXQUOTAS; i++)
3583                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
3584 #endif
3585         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
3586                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
3587
3588         /*
3589          * Allow the "check" option to be passed as a remount option.
3590          */
3591         if (!parse_options(data, sb, NULL, &journal_ioprio,
3592                            &n_blocks_count, 1)) {
3593                 err = -EINVAL;
3594                 goto restore_opts;
3595         }
3596
3597         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
3598                 ext4_abort(sb, __func__, "Abort forced by user");
3599
3600         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3601                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3602
3603         es = sbi->s_es;
3604
3605         if (sbi->s_journal) {
3606                 ext4_init_journal_params(sb, sbi->s_journal);
3607                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3608         }
3609
3610         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
3611                 n_blocks_count > ext4_blocks_count(es)) {
3612                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
3613                         err = -EROFS;
3614                         goto restore_opts;
3615                 }
3616
3617                 if (*flags & MS_RDONLY) {
3618                         /*
3619                          * First of all, the unconditional stuff we have to do
3620                          * to disable replay of the journal when we next remount
3621                          */
3622                         sb->s_flags |= MS_RDONLY;
3623
3624                         /*
3625                          * OK, test if we are remounting a valid rw partition
3626                          * readonly, and if so set the rdonly flag and then
3627                          * mark the partition as valid again.
3628                          */
3629                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
3630                             (sbi->s_mount_state & EXT4_VALID_FS))
3631                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
3632
3633                         if (sbi->s_journal)
3634                                 ext4_mark_recovery_complete(sb, es);
3635                 } else {
3636                         /* Make sure we can mount this feature set readwrite */
3637                         if (!ext4_feature_set_ok(sb, 0)) {
3638                                 err = -EROFS;
3639                                 goto restore_opts;
3640                         }
3641                         /*
3642                          * Make sure the group descriptor checksums
3643                          * are sane.  If they aren't, refuse to remount r/w.
3644                          */
3645                         for (g = 0; g < sbi->s_groups_count; g++) {
3646                                 struct ext4_group_desc *gdp =
3647                                         ext4_get_group_desc(sb, g, NULL);
3648
3649                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
3650                                         ext4_msg(sb, KERN_ERR,
3651                "ext4_remount: Checksum for group %u failed (%u!=%u)",
3652                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
3653                                                le16_to_cpu(gdp->bg_checksum));
3654                                         err = -EINVAL;
3655                                         goto restore_opts;
3656                                 }
3657                         }
3658
3659                         /*
3660                          * If we have an unprocessed orphan list hanging
3661                          * around from a previously readonly bdev mount,
3662                          * require a full umount/remount for now.
3663                          */
3664                         if (es->s_last_orphan) {
3665                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
3666                                        "remount RDWR because of unprocessed "
3667                                        "orphan inode list.  Please "
3668                                        "umount/remount instead");
3669                                 err = -EINVAL;
3670                                 goto restore_opts;
3671                         }
3672
3673                         /*
3674                          * Mounting a RDONLY partition read-write, so reread
3675                          * and store the current valid flag.  (It may have
3676                          * been changed by e2fsck since we originally mounted
3677                          * the partition.)
3678                          */
3679                         if (sbi->s_journal)
3680                                 ext4_clear_journal_err(sb, es);
3681                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3682                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3683                                 goto restore_opts;
3684                         if (!ext4_setup_super(sb, es, 0))
3685                                 sb->s_flags &= ~MS_RDONLY;
3686                 }
3687         }
3688         ext4_setup_system_zone(sb);
3689         if (sbi->s_journal == NULL)
3690                 ext4_commit_super(sb, 1);
3691
3692 #ifdef CONFIG_QUOTA
3693         /* Release old quota file names */
3694         for (i = 0; i < MAXQUOTAS; i++)
3695                 if (old_opts.s_qf_names[i] &&
3696                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3697                         kfree(old_opts.s_qf_names[i]);
3698 #endif
3699         unlock_super(sb);
3700         unlock_kernel();
3701         return 0;
3702
3703 restore_opts:
3704         sb->s_flags = old_sb_flags;
3705         sbi->s_mount_opt = old_opts.s_mount_opt;
3706         sbi->s_resuid = old_opts.s_resuid;
3707         sbi->s_resgid = old_opts.s_resgid;
3708         sbi->s_commit_interval = old_opts.s_commit_interval;
3709         sbi->s_min_batch_time = old_opts.s_min_batch_time;
3710         sbi->s_max_batch_time = old_opts.s_max_batch_time;
3711 #ifdef CONFIG_QUOTA
3712         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3713         for (i = 0; i < MAXQUOTAS; i++) {
3714                 if (sbi->s_qf_names[i] &&
3715                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3716                         kfree(sbi->s_qf_names[i]);
3717                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3718         }
3719 #endif
3720         unlock_super(sb);
3721         unlock_kernel();
3722         return err;
3723 }
3724
3725 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
3726 {
3727         struct super_block *sb = dentry->d_sb;
3728         struct ext4_sb_info *sbi = EXT4_SB(sb);
3729         struct ext4_super_block *es = sbi->s_es;
3730         u64 fsid;
3731
3732         if (test_opt(sb, MINIX_DF)) {
3733                 sbi->s_overhead_last = 0;
3734         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3735                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3736                 ext4_fsblk_t overhead = 0;
3737
3738                 /*
3739                  * Compute the overhead (FS structures).  This is constant
3740                  * for a given filesystem unless the number of block groups
3741                  * changes so we cache the previous value until it does.
3742                  */
3743
3744                 /*
3745                  * All of the blocks before first_data_block are
3746                  * overhead
3747                  */
3748                 overhead = le32_to_cpu(es->s_first_data_block);
3749
3750                 /*
3751                  * Add the overhead attributed to the superblock and
3752                  * block group descriptors.  If the sparse superblocks
3753                  * feature is turned on, then not all groups have this.
3754                  */
3755                 for (i = 0; i < ngroups; i++) {
3756                         overhead += ext4_bg_has_super(sb, i) +
3757                                 ext4_bg_num_gdb(sb, i);
3758                         cond_resched();
3759                 }
3760
3761                 /*
3762                  * Every block group has an inode bitmap, a block
3763                  * bitmap, and an inode table.
3764                  */
3765                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3766                 sbi->s_overhead_last = overhead;
3767                 smp_wmb();
3768                 sbi->s_blocks_last = ext4_blocks_count(es);
3769         }
3770
3771         buf->f_type = EXT4_SUPER_MAGIC;
3772         buf->f_bsize = sb->s_blocksize;
3773         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3774         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
3775                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
3776         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3777         if (buf->f_bfree < ext4_r_blocks_count(es))
3778                 buf->f_bavail = 0;
3779         buf->f_files = le32_to_cpu(es->s_inodes_count);
3780         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3781         buf->f_namelen = EXT4_NAME_LEN;
3782         fsid = le64_to_cpup((void *)es->s_uuid) ^
3783                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3784         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3785         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3786
3787         return 0;
3788 }
3789
3790 /* Helper function for writing quotas on sync - we need to start transaction
3791  * before quota file is locked for write. Otherwise the are possible deadlocks:
3792  * Process 1                         Process 2
3793  * ext4_create()                     quota_sync()
3794  *   jbd2_journal_start()                  write_dquot()
3795  *   dquot_initialize()                         down(dqio_mutex)
3796  *     down(dqio_mutex)                    jbd2_journal_start()
3797  *
3798  */
3799
3800 #ifdef CONFIG_QUOTA
3801
3802 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3803 {
3804         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3805 }
3806
3807 static int ext4_write_dquot(struct dquot *dquot)
3808 {
3809         int ret, err;
3810         handle_t *handle;
3811         struct inode *inode;
3812
3813         inode = dquot_to_inode(dquot);
3814         handle = ext4_journal_start(inode,
3815                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3816         if (IS_ERR(handle))
3817                 return PTR_ERR(handle);
3818         ret = dquot_commit(dquot);
3819         err = ext4_journal_stop(handle);
3820         if (!ret)
3821                 ret = err;
3822         return ret;
3823 }
3824
3825 static int ext4_acquire_dquot(struct dquot *dquot)
3826 {
3827         int ret, err;
3828         handle_t *handle;
3829
3830         handle = ext4_journal_start(dquot_to_inode(dquot),
3831                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3832         if (IS_ERR(handle))
3833                 return PTR_ERR(handle);
3834         ret = dquot_acquire(dquot);
3835         err = ext4_journal_stop(handle);
3836         if (!ret)
3837                 ret = err;
3838         return ret;
3839 }
3840
3841 static int ext4_release_dquot(struct dquot *dquot)
3842 {
3843         int ret, err;
3844         handle_t *handle;
3845
3846         handle = ext4_journal_start(dquot_to_inode(dquot),
3847                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3848         if (IS_ERR(handle)) {
3849                 /* Release dquot anyway to avoid endless cycle in dqput() */
3850                 dquot_release(dquot);
3851                 return PTR_ERR(handle);
3852         }
3853         ret = dquot_release(dquot);
3854         err = ext4_journal_stop(handle);
3855         if (!ret)
3856                 ret = err;
3857         return ret;
3858 }
3859
3860 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3861 {
3862         /* Are we journaling quotas? */
3863         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3864             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3865                 dquot_mark_dquot_dirty(dquot);
3866                 return ext4_write_dquot(dquot);
3867         } else {
3868                 return dquot_mark_dquot_dirty(dquot);
3869         }
3870 }
3871
3872 static int ext4_write_info(struct super_block *sb, int type)
3873 {
3874         int ret, err;
3875         handle_t *handle;
3876
3877         /* Data block + inode block */
3878         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3879         if (IS_ERR(handle))
3880                 return PTR_ERR(handle);
3881         ret = dquot_commit_info(sb, type);
3882         err = ext4_journal_stop(handle);
3883         if (!ret)
3884                 ret = err;
3885         return ret;
3886 }
3887
3888 /*
3889  * Turn on quotas during mount time - we need to find
3890  * the quota file and such...
3891  */
3892 static int ext4_quota_on_mount(struct super_block *sb, int type)
3893 {
3894         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3895                                   EXT4_SB(sb)->s_jquota_fmt, type);
3896 }
3897
3898 /*
3899  * Standard function to be called on quota_on
3900  */
3901 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3902                          char *name, int remount)
3903 {
3904         int err;
3905         struct path path;
3906
3907         if (!test_opt(sb, QUOTA))
3908                 return -EINVAL;
3909         /* When remounting, no checks are needed and in fact, name is NULL */
3910         if (remount)
3911                 return vfs_quota_on(sb, type, format_id, name, remount);
3912
3913         err = kern_path(name, LOOKUP_FOLLOW, &path);
3914         if (err)
3915                 return err;
3916
3917         /* Quotafile not on the same filesystem? */
3918         if (path.mnt->mnt_sb != sb) {
3919                 path_put(&path);
3920                 return -EXDEV;
3921         }
3922         /* Journaling quota? */
3923         if (EXT4_SB(sb)->s_qf_names[type]) {
3924                 /* Quotafile not in fs root? */
3925                 if (path.dentry->d_parent != sb->s_root)
3926                         ext4_msg(sb, KERN_WARNING,
3927                                 "Quota file not on filesystem root. "
3928                                 "Journaled quota will not work");
3929         }
3930
3931         /*
3932          * When we journal data on quota file, we have to flush journal to see
3933          * all updates to the file when we bypass pagecache...
3934          */
3935         if (EXT4_SB(sb)->s_journal &&
3936             ext4_should_journal_data(path.dentry->d_inode)) {
3937                 /*
3938                  * We don't need to lock updates but journal_flush() could
3939                  * otherwise be livelocked...
3940                  */
3941                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3942                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3943                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3944                 if (err) {
3945                         path_put(&path);
3946                         return err;
3947                 }
3948         }
3949
3950         err = vfs_quota_on_path(sb, type, format_id, &path);
3951         path_put(&path);
3952         return err;
3953 }
3954
3955 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3956  * acquiring the locks... As quota files are never truncated and quota code
3957  * itself serializes the operations (and noone else should touch the files)
3958  * we don't have to be afraid of races */
3959 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3960                                size_t len, loff_t off)
3961 {
3962         struct inode *inode = sb_dqopt(sb)->files[type];
3963         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3964         int err = 0;
3965         int offset = off & (sb->s_blocksize - 1);
3966         int tocopy;
3967         size_t toread;
3968         struct buffer_head *bh;
3969         loff_t i_size = i_size_read(inode);
3970
3971         if (off > i_size)
3972                 return 0;
3973         if (off+len > i_size)
3974                 len = i_size-off;
3975         toread = len;
3976         while (toread > 0) {
3977                 tocopy = sb->s_blocksize - offset < toread ?
3978                                 sb->s_blocksize - offset : toread;
3979                 bh = ext4_bread(NULL, inode, blk, 0, &err);
3980                 if (err)
3981                         return err;
3982                 if (!bh)        /* A hole? */
3983                         memset(data, 0, tocopy);
3984                 else
3985                         memcpy(data, bh->b_data+offset, tocopy);
3986                 brelse(bh);
3987                 offset = 0;
3988                 toread -= tocopy;
3989                 data += tocopy;
3990                 blk++;
3991         }
3992         return len;
3993 }
3994
3995 /* Write to quotafile (we know the transaction is already started and has
3996  * enough credits) */
3997 static ssize_t ext4_quota_write(struct super_block *sb, int type,
3998                                 const char *data, size_t len, loff_t off)
3999 {
4000         struct inode *inode = sb_dqopt(sb)->files[type];
4001         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4002         int err = 0;
4003         int offset = off & (sb->s_blocksize - 1);
4004         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
4005         struct buffer_head *bh;
4006         handle_t *handle = journal_current_handle();
4007
4008         if (EXT4_SB(sb)->s_journal && !handle) {
4009                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4010                         " cancelled because transaction is not started",
4011                         (unsigned long long)off, (unsigned long long)len);
4012                 return -EIO;
4013         }
4014         /*
4015          * Since we account only one data block in transaction credits,
4016          * then it is impossible to cross a block boundary.
4017          */
4018         if (sb->s_blocksize - offset < len) {
4019                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4020                         " cancelled because not block aligned",
4021                         (unsigned long long)off, (unsigned long long)len);
4022                 return -EIO;
4023         }
4024
4025         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4026         bh = ext4_bread(handle, inode, blk, 1, &err);
4027         if (!bh)
4028                 goto out;
4029         if (journal_quota) {
4030                 err = ext4_journal_get_write_access(handle, bh);
4031                 if (err) {
4032                         brelse(bh);
4033                         goto out;
4034                 }
4035         }
4036         lock_buffer(bh);
4037         memcpy(bh->b_data+offset, data, len);
4038         flush_dcache_page(bh->b_page);
4039         unlock_buffer(bh);
4040         if (journal_quota)
4041                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
4042         else {
4043                 /* Always do at least ordered writes for quotas */
4044                 err = ext4_jbd2_file_inode(handle, inode);
4045                 mark_buffer_dirty(bh);
4046         }
4047         brelse(bh);
4048 out:
4049         if (err) {
4050                 mutex_unlock(&inode->i_mutex);
4051                 return err;
4052         }
4053         if (inode->i_size < off + len) {
4054                 i_size_write(inode, off + len);
4055                 EXT4_I(inode)->i_disksize = inode->i_size;
4056         }
4057         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4058         ext4_mark_inode_dirty(handle, inode);
4059         mutex_unlock(&inode->i_mutex);
4060         return len;
4061 }
4062
4063 #endif
4064
4065 static int ext4_get_sb(struct file_system_type *fs_type, int flags,
4066                        const char *dev_name, void *data, struct vfsmount *mnt)
4067 {
4068         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super,mnt);
4069 }
4070
4071 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4072 static struct file_system_type ext2_fs_type = {
4073         .owner          = THIS_MODULE,
4074         .name           = "ext2",
4075         .get_sb         = ext4_get_sb,
4076         .kill_sb        = kill_block_super,
4077         .fs_flags       = FS_REQUIRES_DEV,
4078 };
4079
4080 static inline void register_as_ext2(void)
4081 {
4082         int err = register_filesystem(&ext2_fs_type);
4083         if (err)
4084                 printk(KERN_WARNING
4085                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4086 }
4087
4088 static inline void unregister_as_ext2(void)
4089 {
4090         unregister_filesystem(&ext2_fs_type);
4091 }
4092 MODULE_ALIAS("ext2");
4093 #else
4094 static inline void register_as_ext2(void) { }
4095 static inline void unregister_as_ext2(void) { }
4096 #endif
4097
4098 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4099 static struct file_system_type ext3_fs_type = {
4100         .owner          = THIS_MODULE,
4101         .name           = "ext3",
4102         .get_sb         = ext4_get_sb,
4103         .kill_sb        = kill_block_super,
4104         .fs_flags       = FS_REQUIRES_DEV,
4105 };
4106
4107 static inline void register_as_ext3(void)
4108 {
4109         int err = register_filesystem(&ext3_fs_type);
4110         if (err)
4111                 printk(KERN_WARNING
4112                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4113 }
4114
4115 static inline void unregister_as_ext3(void)
4116 {
4117         unregister_filesystem(&ext3_fs_type);
4118 }
4119 MODULE_ALIAS("ext3");
4120 #else
4121 static inline void register_as_ext3(void) { }
4122 static inline void unregister_as_ext3(void) { }
4123 #endif
4124
4125 static struct file_system_type ext4_fs_type = {
4126         .owner          = THIS_MODULE,
4127         .name           = "ext4",
4128         .get_sb         = ext4_get_sb,
4129         .kill_sb        = kill_block_super,
4130         .fs_flags       = FS_REQUIRES_DEV,
4131 };
4132
4133 static int __init init_ext4_fs(void)
4134 {
4135         int err;
4136
4137         err = init_ext4_system_zone();
4138         if (err)
4139                 return err;
4140         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4141         if (!ext4_kset)
4142                 goto out4;
4143         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4144         err = init_ext4_mballoc();
4145         if (err)
4146                 goto out3;
4147
4148         err = init_ext4_xattr();
4149         if (err)
4150                 goto out2;
4151         err = init_inodecache();
4152         if (err)
4153                 goto out1;
4154         register_as_ext2();
4155         register_as_ext3();
4156         err = register_filesystem(&ext4_fs_type);
4157         if (err)
4158                 goto out;
4159         return 0;
4160 out:
4161         unregister_as_ext2();
4162         unregister_as_ext3();
4163         destroy_inodecache();
4164 out1:
4165         exit_ext4_xattr();
4166 out2:
4167         exit_ext4_mballoc();
4168 out3:
4169         remove_proc_entry("fs/ext4", NULL);
4170         kset_unregister(ext4_kset);
4171 out4:
4172         exit_ext4_system_zone();
4173         return err;
4174 }
4175
4176 static void __exit exit_ext4_fs(void)
4177 {
4178         unregister_as_ext2();
4179         unregister_as_ext3();
4180         unregister_filesystem(&ext4_fs_type);
4181         destroy_inodecache();
4182         exit_ext4_xattr();
4183         exit_ext4_mballoc();
4184         remove_proc_entry("fs/ext4", NULL);
4185         kset_unregister(ext4_kset);
4186         exit_ext4_system_zone();
4187 }
4188
4189 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4190 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4191 MODULE_LICENSE("GPL");
4192 module_init(init_ext4_fs)
4193 module_exit(exit_ext4_fs)