08d31101eb0597d1eeb04235b71f52f8ce5a3cac
[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 static int ext4_get_sb(struct file_system_type *fs_type, int flags,
72                        const char *dev_name, void *data, struct vfsmount *mnt);
73
74 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
75 static struct file_system_type ext3_fs_type = {
76         .owner          = THIS_MODULE,
77         .name           = "ext3",
78         .get_sb         = ext4_get_sb,
79         .kill_sb        = kill_block_super,
80         .fs_flags       = FS_REQUIRES_DEV,
81 };
82 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
83 #else
84 #define IS_EXT3_SB(sb) (0)
85 #endif
86
87 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
88                                struct ext4_group_desc *bg)
89 {
90         return le32_to_cpu(bg->bg_block_bitmap_lo) |
91                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
92                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
93 }
94
95 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
96                                struct ext4_group_desc *bg)
97 {
98         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
99                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
100                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
101 }
102
103 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
104                               struct ext4_group_desc *bg)
105 {
106         return le32_to_cpu(bg->bg_inode_table_lo) |
107                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
108                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
109 }
110
111 __u32 ext4_free_blks_count(struct super_block *sb,
112                               struct ext4_group_desc *bg)
113 {
114         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
115                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
116                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
117 }
118
119 __u32 ext4_free_inodes_count(struct super_block *sb,
120                               struct ext4_group_desc *bg)
121 {
122         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
123                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
124                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
125 }
126
127 __u32 ext4_used_dirs_count(struct super_block *sb,
128                               struct ext4_group_desc *bg)
129 {
130         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
131                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
132                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
133 }
134
135 __u32 ext4_itable_unused_count(struct super_block *sb,
136                               struct ext4_group_desc *bg)
137 {
138         return le16_to_cpu(bg->bg_itable_unused_lo) |
139                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
140                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
141 }
142
143 void ext4_block_bitmap_set(struct super_block *sb,
144                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
145 {
146         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
147         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
148                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
149 }
150
151 void ext4_inode_bitmap_set(struct super_block *sb,
152                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
153 {
154         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
155         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
156                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
157 }
158
159 void ext4_inode_table_set(struct super_block *sb,
160                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
161 {
162         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
163         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
164                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
165 }
166
167 void ext4_free_blks_set(struct super_block *sb,
168                           struct ext4_group_desc *bg, __u32 count)
169 {
170         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
171         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
172                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
173 }
174
175 void ext4_free_inodes_set(struct super_block *sb,
176                           struct ext4_group_desc *bg, __u32 count)
177 {
178         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
179         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
180                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
181 }
182
183 void ext4_used_dirs_set(struct super_block *sb,
184                           struct ext4_group_desc *bg, __u32 count)
185 {
186         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
187         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
188                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
189 }
190
191 void ext4_itable_unused_set(struct super_block *sb,
192                           struct ext4_group_desc *bg, __u32 count)
193 {
194         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
195         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
196                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
197 }
198
199
200 /* Just increment the non-pointer handle value */
201 static handle_t *ext4_get_nojournal(void)
202 {
203         handle_t *handle = current->journal_info;
204         unsigned long ref_cnt = (unsigned long)handle;
205
206         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
207
208         ref_cnt++;
209         handle = (handle_t *)ref_cnt;
210
211         current->journal_info = handle;
212         return handle;
213 }
214
215
216 /* Decrement the non-pointer handle value */
217 static void ext4_put_nojournal(handle_t *handle)
218 {
219         unsigned long ref_cnt = (unsigned long)handle;
220
221         BUG_ON(ref_cnt == 0);
222
223         ref_cnt--;
224         handle = (handle_t *)ref_cnt;
225
226         current->journal_info = handle;
227 }
228
229 /*
230  * Wrappers for jbd2_journal_start/end.
231  *
232  * The only special thing we need to do here is to make sure that all
233  * journal_end calls result in the superblock being marked dirty, so
234  * that sync() will call the filesystem's write_super callback if
235  * appropriate.
236  */
237 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
238 {
239         journal_t *journal;
240
241         if (sb->s_flags & MS_RDONLY)
242                 return ERR_PTR(-EROFS);
243
244         /* Special case here: if the journal has aborted behind our
245          * backs (eg. EIO in the commit thread), then we still need to
246          * take the FS itself readonly cleanly. */
247         journal = EXT4_SB(sb)->s_journal;
248         if (journal) {
249                 if (is_journal_aborted(journal)) {
250                         ext4_abort(sb, __func__, "Detected aborted journal");
251                         return ERR_PTR(-EROFS);
252                 }
253                 return jbd2_journal_start(journal, nblocks);
254         }
255         return ext4_get_nojournal();
256 }
257
258 /*
259  * The only special thing we need to do here is to make sure that all
260  * jbd2_journal_stop calls result in the superblock being marked dirty, so
261  * that sync() will call the filesystem's write_super callback if
262  * appropriate.
263  */
264 int __ext4_journal_stop(const char *where, handle_t *handle)
265 {
266         struct super_block *sb;
267         int err;
268         int rc;
269
270         if (!ext4_handle_valid(handle)) {
271                 ext4_put_nojournal(handle);
272                 return 0;
273         }
274         sb = handle->h_transaction->t_journal->j_private;
275         err = handle->h_err;
276         rc = jbd2_journal_stop(handle);
277
278         if (!err)
279                 err = rc;
280         if (err)
281                 __ext4_std_error(sb, where, err);
282         return err;
283 }
284
285 void ext4_journal_abort_handle(const char *caller, const char *err_fn,
286                 struct buffer_head *bh, handle_t *handle, int err)
287 {
288         char nbuf[16];
289         const char *errstr = ext4_decode_error(NULL, err, nbuf);
290
291         BUG_ON(!ext4_handle_valid(handle));
292
293         if (bh)
294                 BUFFER_TRACE(bh, "abort");
295
296         if (!handle->h_err)
297                 handle->h_err = err;
298
299         if (is_handle_aborted(handle))
300                 return;
301
302         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
303                caller, errstr, err_fn);
304
305         jbd2_journal_abort_handle(handle);
306 }
307
308 /* Deal with the reporting of failure conditions on a filesystem such as
309  * inconsistencies detected or read IO failures.
310  *
311  * On ext2, we can store the error state of the filesystem in the
312  * superblock.  That is not possible on ext4, because we may have other
313  * write ordering constraints on the superblock which prevent us from
314  * writing it out straight away; and given that the journal is about to
315  * be aborted, we can't rely on the current, or future, transactions to
316  * write out the superblock safely.
317  *
318  * We'll just use the jbd2_journal_abort() error code to record an error in
319  * the journal instead.  On recovery, the journal will complain about
320  * that error until we've noted it down and cleared it.
321  */
322
323 static void ext4_handle_error(struct super_block *sb)
324 {
325         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
326
327         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
328         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
329
330         if (sb->s_flags & MS_RDONLY)
331                 return;
332
333         if (!test_opt(sb, ERRORS_CONT)) {
334                 journal_t *journal = EXT4_SB(sb)->s_journal;
335
336                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
337                 if (journal)
338                         jbd2_journal_abort(journal, -EIO);
339         }
340         if (test_opt(sb, ERRORS_RO)) {
341                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
342                 sb->s_flags |= MS_RDONLY;
343         }
344         ext4_commit_super(sb, 1);
345         if (test_opt(sb, ERRORS_PANIC))
346                 panic("EXT4-fs (device %s): panic forced after error\n",
347                         sb->s_id);
348 }
349
350 void __ext4_error(struct super_block *sb, const char *function,
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: ", sb->s_id, function);
357         vprintk(fmt, args);
358         printk("\n");
359         va_end(args);
360
361         ext4_handle_error(sb);
362 }
363
364 void ext4_error_inode(const char *function, struct inode *inode,
365                       const char *fmt, ...)
366 {
367         va_list args;
368
369         va_start(args, fmt);
370         printk(KERN_CRIT "EXT4-fs error (device %s): %s: inode #%lu: (comm %s) ",
371                inode->i_sb->s_id, function, inode->i_ino, current->comm);
372         vprintk(fmt, args);
373         printk("\n");
374         va_end(args);
375
376         ext4_handle_error(inode->i_sb);
377 }
378
379 void ext4_error_file(const char *function, struct file *file,
380                      const char *fmt, ...)
381 {
382         va_list args;
383         struct inode *inode = file->f_dentry->d_inode;
384         char pathname[80], *path;
385
386         va_start(args, fmt);
387         path = d_path(&(file->f_path), pathname, sizeof(pathname));
388         if (!path)
389                 path = "(unknown)";
390         printk(KERN_CRIT
391                "EXT4-fs error (device %s): %s: inode #%lu (comm %s path %s): ",
392                inode->i_sb->s_id, function, inode->i_ino, current->comm, path);
393         vprintk(fmt, args);
394         printk("\n");
395         va_end(args);
396
397         ext4_handle_error(inode->i_sb);
398 }
399
400 static const char *ext4_decode_error(struct super_block *sb, int errno,
401                                      char nbuf[16])
402 {
403         char *errstr = NULL;
404
405         switch (errno) {
406         case -EIO:
407                 errstr = "IO failure";
408                 break;
409         case -ENOMEM:
410                 errstr = "Out of memory";
411                 break;
412         case -EROFS:
413                 if (!sb || (EXT4_SB(sb)->s_journal &&
414                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
415                         errstr = "Journal has aborted";
416                 else
417                         errstr = "Readonly filesystem";
418                 break;
419         default:
420                 /* If the caller passed in an extra buffer for unknown
421                  * errors, textualise them now.  Else we just return
422                  * NULL. */
423                 if (nbuf) {
424                         /* Check for truncated error codes... */
425                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
426                                 errstr = nbuf;
427                 }
428                 break;
429         }
430
431         return errstr;
432 }
433
434 /* __ext4_std_error decodes expected errors from journaling functions
435  * automatically and invokes the appropriate error response.  */
436
437 void __ext4_std_error(struct super_block *sb, const char *function, int errno)
438 {
439         char nbuf[16];
440         const char *errstr;
441
442         /* Special case: if the error is EROFS, and we're not already
443          * inside a transaction, then there's really no point in logging
444          * an error. */
445         if (errno == -EROFS && journal_current_handle() == NULL &&
446             (sb->s_flags & MS_RDONLY))
447                 return;
448
449         errstr = ext4_decode_error(sb, errno, nbuf);
450         printk(KERN_CRIT "EXT4-fs error (device %s) in %s: %s\n",
451                sb->s_id, function, errstr);
452
453         ext4_handle_error(sb);
454 }
455
456 /*
457  * ext4_abort is a much stronger failure handler than ext4_error.  The
458  * abort function may be used to deal with unrecoverable failures such
459  * as journal IO errors or ENOMEM at a critical moment in log management.
460  *
461  * We unconditionally force the filesystem into an ABORT|READONLY state,
462  * unless the error response on the fs has been set to panic in which
463  * case we take the easy way out and panic immediately.
464  */
465
466 void ext4_abort(struct super_block *sb, const char *function,
467                 const char *fmt, ...)
468 {
469         va_list args;
470
471         va_start(args, fmt);
472         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
473         vprintk(fmt, args);
474         printk("\n");
475         va_end(args);
476
477         if (test_opt(sb, ERRORS_PANIC))
478                 panic("EXT4-fs panic from previous error\n");
479
480         if (sb->s_flags & MS_RDONLY)
481                 return;
482
483         ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
484         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
485         sb->s_flags |= MS_RDONLY;
486         EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
487         if (EXT4_SB(sb)->s_journal)
488                 jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
489 }
490
491 void ext4_msg (struct super_block * sb, const char *prefix,
492                    const char *fmt, ...)
493 {
494         va_list args;
495
496         va_start(args, fmt);
497         printk("%sEXT4-fs (%s): ", prefix, sb->s_id);
498         vprintk(fmt, args);
499         printk("\n");
500         va_end(args);
501 }
502
503 void __ext4_warning(struct super_block *sb, const char *function,
504                   const char *fmt, ...)
505 {
506         va_list args;
507
508         va_start(args, fmt);
509         printk(KERN_WARNING "EXT4-fs warning (device %s): %s: ",
510                sb->s_id, function);
511         vprintk(fmt, args);
512         printk("\n");
513         va_end(args);
514 }
515
516 void ext4_grp_locked_error(struct super_block *sb, ext4_group_t grp,
517                            const char *function, const char *fmt, ...)
518 __releases(bitlock)
519 __acquires(bitlock)
520 {
521         va_list args;
522         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
523
524         va_start(args, fmt);
525         printk(KERN_CRIT "EXT4-fs error (device %s): %s: ", sb->s_id, function);
526         vprintk(fmt, args);
527         printk("\n");
528         va_end(args);
529
530         if (test_opt(sb, ERRORS_CONT)) {
531                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
532                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
533                 ext4_commit_super(sb, 0);
534                 return;
535         }
536         ext4_unlock_group(sb, grp);
537         ext4_handle_error(sb);
538         /*
539          * We only get here in the ERRORS_RO case; relocking the group
540          * may be dangerous, but nothing bad will happen since the
541          * filesystem will have already been marked read/only and the
542          * journal has been aborted.  We return 1 as a hint to callers
543          * who might what to use the return value from
544          * ext4_grp_locked_error() to distinguish beween the
545          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
546          * aggressively from the ext4 function in question, with a
547          * more appropriate error code.
548          */
549         ext4_lock_group(sb, grp);
550         return;
551 }
552
553 void ext4_update_dynamic_rev(struct super_block *sb)
554 {
555         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
556
557         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
558                 return;
559
560         ext4_warning(sb,
561                      "updating to rev %d because of new feature flag, "
562                      "running e2fsck is recommended",
563                      EXT4_DYNAMIC_REV);
564
565         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
566         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
567         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
568         /* leave es->s_feature_*compat flags alone */
569         /* es->s_uuid will be set by e2fsck if empty */
570
571         /*
572          * The rest of the superblock fields should be zero, and if not it
573          * means they are likely already in use, so leave them alone.  We
574          * can leave it up to e2fsck to clean up any inconsistencies there.
575          */
576 }
577
578 /*
579  * Open the external journal device
580  */
581 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
582 {
583         struct block_device *bdev;
584         char b[BDEVNAME_SIZE];
585
586         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
587         if (IS_ERR(bdev))
588                 goto fail;
589         return bdev;
590
591 fail:
592         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
593                         __bdevname(dev, b), PTR_ERR(bdev));
594         return NULL;
595 }
596
597 /*
598  * Release the journal device
599  */
600 static int ext4_blkdev_put(struct block_device *bdev)
601 {
602         bd_release(bdev);
603         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
604 }
605
606 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
607 {
608         struct block_device *bdev;
609         int ret = -ENODEV;
610
611         bdev = sbi->journal_bdev;
612         if (bdev) {
613                 ret = ext4_blkdev_put(bdev);
614                 sbi->journal_bdev = NULL;
615         }
616         return ret;
617 }
618
619 static inline struct inode *orphan_list_entry(struct list_head *l)
620 {
621         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
622 }
623
624 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
625 {
626         struct list_head *l;
627
628         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
629                  le32_to_cpu(sbi->s_es->s_last_orphan));
630
631         printk(KERN_ERR "sb_info orphan list:\n");
632         list_for_each(l, &sbi->s_orphan) {
633                 struct inode *inode = orphan_list_entry(l);
634                 printk(KERN_ERR "  "
635                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
636                        inode->i_sb->s_id, inode->i_ino, inode,
637                        inode->i_mode, inode->i_nlink,
638                        NEXT_ORPHAN(inode));
639         }
640 }
641
642 static void ext4_put_super(struct super_block *sb)
643 {
644         struct ext4_sb_info *sbi = EXT4_SB(sb);
645         struct ext4_super_block *es = sbi->s_es;
646         int i, err;
647
648         flush_workqueue(sbi->dio_unwritten_wq);
649         destroy_workqueue(sbi->dio_unwritten_wq);
650
651         lock_super(sb);
652         lock_kernel();
653         if (sb->s_dirt)
654                 ext4_commit_super(sb, 1);
655
656         if (sbi->s_journal) {
657                 err = jbd2_journal_destroy(sbi->s_journal);
658                 sbi->s_journal = NULL;
659                 if (err < 0)
660                         ext4_abort(sb, __func__,
661                                    "Couldn't clean up the journal");
662         }
663
664         ext4_release_system_zone(sb);
665         ext4_mb_release(sb);
666         ext4_ext_release(sb);
667         ext4_xattr_put_super(sb);
668
669         if (!(sb->s_flags & MS_RDONLY)) {
670                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
671                 es->s_state = cpu_to_le16(sbi->s_mount_state);
672                 ext4_commit_super(sb, 1);
673         }
674         if (sbi->s_proc) {
675                 remove_proc_entry(sb->s_id, ext4_proc_root);
676         }
677         kobject_del(&sbi->s_kobj);
678
679         for (i = 0; i < sbi->s_gdb_count; i++)
680                 brelse(sbi->s_group_desc[i]);
681         kfree(sbi->s_group_desc);
682         if (is_vmalloc_addr(sbi->s_flex_groups))
683                 vfree(sbi->s_flex_groups);
684         else
685                 kfree(sbi->s_flex_groups);
686         percpu_counter_destroy(&sbi->s_freeblocks_counter);
687         percpu_counter_destroy(&sbi->s_freeinodes_counter);
688         percpu_counter_destroy(&sbi->s_dirs_counter);
689         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
690         brelse(sbi->s_sbh);
691 #ifdef CONFIG_QUOTA
692         for (i = 0; i < MAXQUOTAS; i++)
693                 kfree(sbi->s_qf_names[i]);
694 #endif
695
696         /* Debugging code just in case the in-memory inode orphan list
697          * isn't empty.  The on-disk one can be non-empty if we've
698          * detected an error and taken the fs readonly, but the
699          * in-memory list had better be clean by this point. */
700         if (!list_empty(&sbi->s_orphan))
701                 dump_orphan_list(sb, sbi);
702         J_ASSERT(list_empty(&sbi->s_orphan));
703
704         invalidate_bdev(sb->s_bdev);
705         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
706                 /*
707                  * Invalidate the journal device's buffers.  We don't want them
708                  * floating about in memory - the physical journal device may
709                  * hotswapped, and it breaks the `ro-after' testing code.
710                  */
711                 sync_blockdev(sbi->journal_bdev);
712                 invalidate_bdev(sbi->journal_bdev);
713                 ext4_blkdev_remove(sbi);
714         }
715         sb->s_fs_info = NULL;
716         /*
717          * Now that we are completely done shutting down the
718          * superblock, we need to actually destroy the kobject.
719          */
720         unlock_kernel();
721         unlock_super(sb);
722         kobject_put(&sbi->s_kobj);
723         wait_for_completion(&sbi->s_kobj_unregister);
724         kfree(sbi->s_blockgroup_lock);
725         kfree(sbi);
726 }
727
728 static struct kmem_cache *ext4_inode_cachep;
729
730 /*
731  * Called inside transaction, so use GFP_NOFS
732  */
733 static struct inode *ext4_alloc_inode(struct super_block *sb)
734 {
735         struct ext4_inode_info *ei;
736
737         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
738         if (!ei)
739                 return NULL;
740
741         ei->vfs_inode.i_version = 1;
742         ei->vfs_inode.i_data.writeback_index = 0;
743         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
744         INIT_LIST_HEAD(&ei->i_prealloc_list);
745         spin_lock_init(&ei->i_prealloc_lock);
746         /*
747          * Note:  We can be called before EXT4_SB(sb)->s_journal is set,
748          * therefore it can be null here.  Don't check it, just initialize
749          * jinode.
750          */
751         jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
752         ei->i_reserved_data_blocks = 0;
753         ei->i_reserved_meta_blocks = 0;
754         ei->i_allocated_meta_blocks = 0;
755         ei->i_da_metadata_calc_len = 0;
756         ei->i_delalloc_reserved_flag = 0;
757         spin_lock_init(&(ei->i_block_reservation_lock));
758 #ifdef CONFIG_QUOTA
759         ei->i_reserved_quota = 0;
760 #endif
761         INIT_LIST_HEAD(&ei->i_completed_io_list);
762         spin_lock_init(&ei->i_completed_io_lock);
763         ei->cur_aio_dio = NULL;
764         ei->i_sync_tid = 0;
765         ei->i_datasync_tid = 0;
766
767         return &ei->vfs_inode;
768 }
769
770 static void ext4_destroy_inode(struct inode *inode)
771 {
772         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
773                 ext4_msg(inode->i_sb, KERN_ERR,
774                          "Inode %lu (%p): orphan list check failed!",
775                          inode->i_ino, EXT4_I(inode));
776                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
777                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
778                                 true);
779                 dump_stack();
780         }
781         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
782 }
783
784 static void init_once(void *foo)
785 {
786         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
787
788         INIT_LIST_HEAD(&ei->i_orphan);
789 #ifdef CONFIG_EXT4_FS_XATTR
790         init_rwsem(&ei->xattr_sem);
791 #endif
792         init_rwsem(&ei->i_data_sem);
793         inode_init_once(&ei->vfs_inode);
794 }
795
796 static int init_inodecache(void)
797 {
798         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
799                                              sizeof(struct ext4_inode_info),
800                                              0, (SLAB_RECLAIM_ACCOUNT|
801                                                 SLAB_MEM_SPREAD),
802                                              init_once);
803         if (ext4_inode_cachep == NULL)
804                 return -ENOMEM;
805         return 0;
806 }
807
808 static void destroy_inodecache(void)
809 {
810         kmem_cache_destroy(ext4_inode_cachep);
811 }
812
813 static void ext4_clear_inode(struct inode *inode)
814 {
815         dquot_drop(inode);
816         ext4_discard_preallocations(inode);
817         if (EXT4_JOURNAL(inode))
818                 jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
819                                        &EXT4_I(inode)->jinode);
820 }
821
822 static inline void ext4_show_quota_options(struct seq_file *seq,
823                                            struct super_block *sb)
824 {
825 #if defined(CONFIG_QUOTA)
826         struct ext4_sb_info *sbi = EXT4_SB(sb);
827
828         if (sbi->s_jquota_fmt) {
829                 char *fmtname = "";
830
831                 switch (sbi->s_jquota_fmt) {
832                 case QFMT_VFS_OLD:
833                         fmtname = "vfsold";
834                         break;
835                 case QFMT_VFS_V0:
836                         fmtname = "vfsv0";
837                         break;
838                 case QFMT_VFS_V1:
839                         fmtname = "vfsv1";
840                         break;
841                 }
842                 seq_printf(seq, ",jqfmt=%s", fmtname);
843         }
844
845         if (sbi->s_qf_names[USRQUOTA])
846                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
847
848         if (sbi->s_qf_names[GRPQUOTA])
849                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
850
851         if (test_opt(sb, USRQUOTA))
852                 seq_puts(seq, ",usrquota");
853
854         if (test_opt(sb, GRPQUOTA))
855                 seq_puts(seq, ",grpquota");
856 #endif
857 }
858
859 /*
860  * Show an option if
861  *  - it's set to a non-default value OR
862  *  - if the per-sb default is different from the global default
863  */
864 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
865 {
866         int def_errors;
867         unsigned long def_mount_opts;
868         struct super_block *sb = vfs->mnt_sb;
869         struct ext4_sb_info *sbi = EXT4_SB(sb);
870         struct ext4_super_block *es = sbi->s_es;
871
872         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
873         def_errors     = le16_to_cpu(es->s_errors);
874
875         if (sbi->s_sb_block != 1)
876                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
877         if (test_opt(sb, MINIX_DF))
878                 seq_puts(seq, ",minixdf");
879         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
880                 seq_puts(seq, ",grpid");
881         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
882                 seq_puts(seq, ",nogrpid");
883         if (sbi->s_resuid != EXT4_DEF_RESUID ||
884             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
885                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
886         }
887         if (sbi->s_resgid != EXT4_DEF_RESGID ||
888             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
889                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
890         }
891         if (test_opt(sb, ERRORS_RO)) {
892                 if (def_errors == EXT4_ERRORS_PANIC ||
893                     def_errors == EXT4_ERRORS_CONTINUE) {
894                         seq_puts(seq, ",errors=remount-ro");
895                 }
896         }
897         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
898                 seq_puts(seq, ",errors=continue");
899         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
900                 seq_puts(seq, ",errors=panic");
901         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
902                 seq_puts(seq, ",nouid32");
903         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
904                 seq_puts(seq, ",debug");
905         if (test_opt(sb, OLDALLOC))
906                 seq_puts(seq, ",oldalloc");
907 #ifdef CONFIG_EXT4_FS_XATTR
908         if (test_opt(sb, XATTR_USER) &&
909                 !(def_mount_opts & EXT4_DEFM_XATTR_USER))
910                 seq_puts(seq, ",user_xattr");
911         if (!test_opt(sb, XATTR_USER) &&
912             (def_mount_opts & EXT4_DEFM_XATTR_USER)) {
913                 seq_puts(seq, ",nouser_xattr");
914         }
915 #endif
916 #ifdef CONFIG_EXT4_FS_POSIX_ACL
917         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
918                 seq_puts(seq, ",acl");
919         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
920                 seq_puts(seq, ",noacl");
921 #endif
922         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
923                 seq_printf(seq, ",commit=%u",
924                            (unsigned) (sbi->s_commit_interval / HZ));
925         }
926         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
927                 seq_printf(seq, ",min_batch_time=%u",
928                            (unsigned) sbi->s_min_batch_time);
929         }
930         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
931                 seq_printf(seq, ",max_batch_time=%u",
932                            (unsigned) sbi->s_min_batch_time);
933         }
934
935         /*
936          * We're changing the default of barrier mount option, so
937          * let's always display its mount state so it's clear what its
938          * status is.
939          */
940         seq_puts(seq, ",barrier=");
941         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
942         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
943                 seq_puts(seq, ",journal_async_commit");
944         if (test_opt(sb, NOBH))
945                 seq_puts(seq, ",nobh");
946         if (test_opt(sb, I_VERSION))
947                 seq_puts(seq, ",i_version");
948         if (!test_opt(sb, DELALLOC))
949                 seq_puts(seq, ",nodelalloc");
950
951
952         if (sbi->s_stripe)
953                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
954         /*
955          * journal mode get enabled in different ways
956          * So just print the value even if we didn't specify it
957          */
958         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
959                 seq_puts(seq, ",data=journal");
960         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
961                 seq_puts(seq, ",data=ordered");
962         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
963                 seq_puts(seq, ",data=writeback");
964
965         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
966                 seq_printf(seq, ",inode_readahead_blks=%u",
967                            sbi->s_inode_readahead_blks);
968
969         if (test_opt(sb, DATA_ERR_ABORT))
970                 seq_puts(seq, ",data_err=abort");
971
972         if (test_opt(sb, NO_AUTO_DA_ALLOC))
973                 seq_puts(seq, ",noauto_da_alloc");
974
975         if (test_opt(sb, DISCARD))
976                 seq_puts(seq, ",discard");
977
978         if (test_opt(sb, NOLOAD))
979                 seq_puts(seq, ",norecovery");
980
981         if (test_opt(sb, DIOREAD_NOLOCK))
982                 seq_puts(seq, ",dioread_nolock");
983
984         ext4_show_quota_options(seq, sb);
985
986         return 0;
987 }
988
989 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
990                                         u64 ino, u32 generation)
991 {
992         struct inode *inode;
993
994         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
995                 return ERR_PTR(-ESTALE);
996         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
997                 return ERR_PTR(-ESTALE);
998
999         /* iget isn't really right if the inode is currently unallocated!!
1000          *
1001          * ext4_read_inode will return a bad_inode if the inode had been
1002          * deleted, so we should be safe.
1003          *
1004          * Currently we don't know the generation for parent directory, so
1005          * a generation of 0 means "accept any"
1006          */
1007         inode = ext4_iget(sb, ino);
1008         if (IS_ERR(inode))
1009                 return ERR_CAST(inode);
1010         if (generation && inode->i_generation != generation) {
1011                 iput(inode);
1012                 return ERR_PTR(-ESTALE);
1013         }
1014
1015         return inode;
1016 }
1017
1018 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1019                                         int fh_len, int fh_type)
1020 {
1021         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1022                                     ext4_nfs_get_inode);
1023 }
1024
1025 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1026                                         int fh_len, int fh_type)
1027 {
1028         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1029                                     ext4_nfs_get_inode);
1030 }
1031
1032 /*
1033  * Try to release metadata pages (indirect blocks, directories) which are
1034  * mapped via the block device.  Since these pages could have journal heads
1035  * which would prevent try_to_free_buffers() from freeing them, we must use
1036  * jbd2 layer's try_to_free_buffers() function to release them.
1037  */
1038 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1039                                  gfp_t wait)
1040 {
1041         journal_t *journal = EXT4_SB(sb)->s_journal;
1042
1043         WARN_ON(PageChecked(page));
1044         if (!page_has_buffers(page))
1045                 return 0;
1046         if (journal)
1047                 return jbd2_journal_try_to_free_buffers(journal, page,
1048                                                         wait & ~__GFP_WAIT);
1049         return try_to_free_buffers(page);
1050 }
1051
1052 #ifdef CONFIG_QUOTA
1053 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1054 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1055
1056 static int ext4_write_dquot(struct dquot *dquot);
1057 static int ext4_acquire_dquot(struct dquot *dquot);
1058 static int ext4_release_dquot(struct dquot *dquot);
1059 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1060 static int ext4_write_info(struct super_block *sb, int type);
1061 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1062                                 char *path, int remount);
1063 static int ext4_quota_on_mount(struct super_block *sb, int type);
1064 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1065                                size_t len, loff_t off);
1066 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1067                                 const char *data, size_t len, loff_t off);
1068
1069 static const struct dquot_operations ext4_quota_operations = {
1070 #ifdef CONFIG_QUOTA
1071         .get_reserved_space = ext4_get_reserved_space,
1072 #endif
1073         .write_dquot    = ext4_write_dquot,
1074         .acquire_dquot  = ext4_acquire_dquot,
1075         .release_dquot  = ext4_release_dquot,
1076         .mark_dirty     = ext4_mark_dquot_dirty,
1077         .write_info     = ext4_write_info,
1078         .alloc_dquot    = dquot_alloc,
1079         .destroy_dquot  = dquot_destroy,
1080 };
1081
1082 static const struct quotactl_ops ext4_qctl_operations = {
1083         .quota_on       = ext4_quota_on,
1084         .quota_off      = vfs_quota_off,
1085         .quota_sync     = vfs_quota_sync,
1086         .get_info       = vfs_get_dqinfo,
1087         .set_info       = vfs_set_dqinfo,
1088         .get_dqblk      = vfs_get_dqblk,
1089         .set_dqblk      = vfs_set_dqblk
1090 };
1091 #endif
1092
1093 static const struct super_operations ext4_sops = {
1094         .alloc_inode    = ext4_alloc_inode,
1095         .destroy_inode  = ext4_destroy_inode,
1096         .write_inode    = ext4_write_inode,
1097         .dirty_inode    = ext4_dirty_inode,
1098         .delete_inode   = ext4_delete_inode,
1099         .put_super      = ext4_put_super,
1100         .sync_fs        = ext4_sync_fs,
1101         .freeze_fs      = ext4_freeze,
1102         .unfreeze_fs    = ext4_unfreeze,
1103         .statfs         = ext4_statfs,
1104         .remount_fs     = ext4_remount,
1105         .clear_inode    = ext4_clear_inode,
1106         .show_options   = ext4_show_options,
1107 #ifdef CONFIG_QUOTA
1108         .quota_read     = ext4_quota_read,
1109         .quota_write    = ext4_quota_write,
1110 #endif
1111         .bdev_try_to_free_page = bdev_try_to_free_page,
1112 };
1113
1114 static const struct super_operations ext4_nojournal_sops = {
1115         .alloc_inode    = ext4_alloc_inode,
1116         .destroy_inode  = ext4_destroy_inode,
1117         .write_inode    = ext4_write_inode,
1118         .dirty_inode    = ext4_dirty_inode,
1119         .delete_inode   = ext4_delete_inode,
1120         .write_super    = ext4_write_super,
1121         .put_super      = ext4_put_super,
1122         .statfs         = ext4_statfs,
1123         .remount_fs     = ext4_remount,
1124         .clear_inode    = ext4_clear_inode,
1125         .show_options   = ext4_show_options,
1126 #ifdef CONFIG_QUOTA
1127         .quota_read     = ext4_quota_read,
1128         .quota_write    = ext4_quota_write,
1129 #endif
1130         .bdev_try_to_free_page = bdev_try_to_free_page,
1131 };
1132
1133 static const struct export_operations ext4_export_ops = {
1134         .fh_to_dentry = ext4_fh_to_dentry,
1135         .fh_to_parent = ext4_fh_to_parent,
1136         .get_parent = ext4_get_parent,
1137 };
1138
1139 enum {
1140         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1141         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1142         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1143         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1144         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1145         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1146         Opt_journal_update, Opt_journal_dev,
1147         Opt_journal_checksum, Opt_journal_async_commit,
1148         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1149         Opt_data_err_abort, Opt_data_err_ignore,
1150         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1151         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1152         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1153         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1154         Opt_stripe, Opt_delalloc, Opt_nodelalloc,
1155         Opt_block_validity, Opt_noblock_validity,
1156         Opt_inode_readahead_blks, Opt_journal_ioprio,
1157         Opt_dioread_nolock, Opt_dioread_lock,
1158         Opt_discard, Opt_nodiscard,
1159 };
1160
1161 static const match_table_t tokens = {
1162         {Opt_bsd_df, "bsddf"},
1163         {Opt_minix_df, "minixdf"},
1164         {Opt_grpid, "grpid"},
1165         {Opt_grpid, "bsdgroups"},
1166         {Opt_nogrpid, "nogrpid"},
1167         {Opt_nogrpid, "sysvgroups"},
1168         {Opt_resgid, "resgid=%u"},
1169         {Opt_resuid, "resuid=%u"},
1170         {Opt_sb, "sb=%u"},
1171         {Opt_err_cont, "errors=continue"},
1172         {Opt_err_panic, "errors=panic"},
1173         {Opt_err_ro, "errors=remount-ro"},
1174         {Opt_nouid32, "nouid32"},
1175         {Opt_debug, "debug"},
1176         {Opt_oldalloc, "oldalloc"},
1177         {Opt_orlov, "orlov"},
1178         {Opt_user_xattr, "user_xattr"},
1179         {Opt_nouser_xattr, "nouser_xattr"},
1180         {Opt_acl, "acl"},
1181         {Opt_noacl, "noacl"},
1182         {Opt_noload, "noload"},
1183         {Opt_noload, "norecovery"},
1184         {Opt_nobh, "nobh"},
1185         {Opt_bh, "bh"},
1186         {Opt_commit, "commit=%u"},
1187         {Opt_min_batch_time, "min_batch_time=%u"},
1188         {Opt_max_batch_time, "max_batch_time=%u"},
1189         {Opt_journal_update, "journal=update"},
1190         {Opt_journal_dev, "journal_dev=%u"},
1191         {Opt_journal_checksum, "journal_checksum"},
1192         {Opt_journal_async_commit, "journal_async_commit"},
1193         {Opt_abort, "abort"},
1194         {Opt_data_journal, "data=journal"},
1195         {Opt_data_ordered, "data=ordered"},
1196         {Opt_data_writeback, "data=writeback"},
1197         {Opt_data_err_abort, "data_err=abort"},
1198         {Opt_data_err_ignore, "data_err=ignore"},
1199         {Opt_offusrjquota, "usrjquota="},
1200         {Opt_usrjquota, "usrjquota=%s"},
1201         {Opt_offgrpjquota, "grpjquota="},
1202         {Opt_grpjquota, "grpjquota=%s"},
1203         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1204         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1205         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1206         {Opt_grpquota, "grpquota"},
1207         {Opt_noquota, "noquota"},
1208         {Opt_quota, "quota"},
1209         {Opt_usrquota, "usrquota"},
1210         {Opt_barrier, "barrier=%u"},
1211         {Opt_barrier, "barrier"},
1212         {Opt_nobarrier, "nobarrier"},
1213         {Opt_i_version, "i_version"},
1214         {Opt_stripe, "stripe=%u"},
1215         {Opt_resize, "resize"},
1216         {Opt_delalloc, "delalloc"},
1217         {Opt_nodelalloc, "nodelalloc"},
1218         {Opt_block_validity, "block_validity"},
1219         {Opt_noblock_validity, "noblock_validity"},
1220         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1221         {Opt_journal_ioprio, "journal_ioprio=%u"},
1222         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1223         {Opt_auto_da_alloc, "auto_da_alloc"},
1224         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1225         {Opt_dioread_nolock, "dioread_nolock"},
1226         {Opt_dioread_lock, "dioread_lock"},
1227         {Opt_discard, "discard"},
1228         {Opt_nodiscard, "nodiscard"},
1229         {Opt_err, NULL},
1230 };
1231
1232 static ext4_fsblk_t get_sb_block(void **data)
1233 {
1234         ext4_fsblk_t    sb_block;
1235         char            *options = (char *) *data;
1236
1237         if (!options || strncmp(options, "sb=", 3) != 0)
1238                 return 1;       /* Default location */
1239
1240         options += 3;
1241         /* TODO: use simple_strtoll with >32bit ext4 */
1242         sb_block = simple_strtoul(options, &options, 0);
1243         if (*options && *options != ',') {
1244                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1245                        (char *) *data);
1246                 return 1;
1247         }
1248         if (*options == ',')
1249                 options++;
1250         *data = (void *) options;
1251
1252         return sb_block;
1253 }
1254
1255 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1256 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1257         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1258
1259 #ifdef CONFIG_QUOTA
1260 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1261 {
1262         struct ext4_sb_info *sbi = EXT4_SB(sb);
1263         char *qname;
1264
1265         if (sb_any_quota_loaded(sb) &&
1266                 !sbi->s_qf_names[qtype]) {
1267                 ext4_msg(sb, KERN_ERR,
1268                         "Cannot change journaled "
1269                         "quota options when quota turned on");
1270                 return 0;
1271         }
1272         qname = match_strdup(args);
1273         if (!qname) {
1274                 ext4_msg(sb, KERN_ERR,
1275                         "Not enough memory for storing quotafile name");
1276                 return 0;
1277         }
1278         if (sbi->s_qf_names[qtype] &&
1279                 strcmp(sbi->s_qf_names[qtype], qname)) {
1280                 ext4_msg(sb, KERN_ERR,
1281                         "%s quota file already specified", QTYPE2NAME(qtype));
1282                 kfree(qname);
1283                 return 0;
1284         }
1285         sbi->s_qf_names[qtype] = qname;
1286         if (strchr(sbi->s_qf_names[qtype], '/')) {
1287                 ext4_msg(sb, KERN_ERR,
1288                         "quotafile must be on filesystem root");
1289                 kfree(sbi->s_qf_names[qtype]);
1290                 sbi->s_qf_names[qtype] = NULL;
1291                 return 0;
1292         }
1293         set_opt(sbi->s_mount_opt, QUOTA);
1294         return 1;
1295 }
1296
1297 static int clear_qf_name(struct super_block *sb, int qtype)
1298 {
1299
1300         struct ext4_sb_info *sbi = EXT4_SB(sb);
1301
1302         if (sb_any_quota_loaded(sb) &&
1303                 sbi->s_qf_names[qtype]) {
1304                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1305                         " when quota turned on");
1306                 return 0;
1307         }
1308         /*
1309          * The space will be released later when all options are confirmed
1310          * to be correct
1311          */
1312         sbi->s_qf_names[qtype] = NULL;
1313         return 1;
1314 }
1315 #endif
1316
1317 static int parse_options(char *options, struct super_block *sb,
1318                          unsigned long *journal_devnum,
1319                          unsigned int *journal_ioprio,
1320                          ext4_fsblk_t *n_blocks_count, int is_remount)
1321 {
1322         struct ext4_sb_info *sbi = EXT4_SB(sb);
1323         char *p;
1324         substring_t args[MAX_OPT_ARGS];
1325         int data_opt = 0;
1326         int option;
1327 #ifdef CONFIG_QUOTA
1328         int qfmt;
1329 #endif
1330
1331         if (!options)
1332                 return 1;
1333
1334         while ((p = strsep(&options, ",")) != NULL) {
1335                 int token;
1336                 if (!*p)
1337                         continue;
1338
1339                 /*
1340                  * Initialize args struct so we know whether arg was
1341                  * found; some options take optional arguments.
1342                  */
1343                 args[0].to = args[0].from = 0;
1344                 token = match_token(p, tokens, args);
1345                 switch (token) {
1346                 case Opt_bsd_df:
1347                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1348                         clear_opt(sbi->s_mount_opt, MINIX_DF);
1349                         break;
1350                 case Opt_minix_df:
1351                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1352                         set_opt(sbi->s_mount_opt, MINIX_DF);
1353
1354                         break;
1355                 case Opt_grpid:
1356                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1357                         set_opt(sbi->s_mount_opt, GRPID);
1358
1359                         break;
1360                 case Opt_nogrpid:
1361                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1362                         clear_opt(sbi->s_mount_opt, GRPID);
1363
1364                         break;
1365                 case Opt_resuid:
1366                         if (match_int(&args[0], &option))
1367                                 return 0;
1368                         sbi->s_resuid = option;
1369                         break;
1370                 case Opt_resgid:
1371                         if (match_int(&args[0], &option))
1372                                 return 0;
1373                         sbi->s_resgid = option;
1374                         break;
1375                 case Opt_sb:
1376                         /* handled by get_sb_block() instead of here */
1377                         /* *sb_block = match_int(&args[0]); */
1378                         break;
1379                 case Opt_err_panic:
1380                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1381                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1382                         set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1383                         break;
1384                 case Opt_err_ro:
1385                         clear_opt(sbi->s_mount_opt, ERRORS_CONT);
1386                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1387                         set_opt(sbi->s_mount_opt, ERRORS_RO);
1388                         break;
1389                 case Opt_err_cont:
1390                         clear_opt(sbi->s_mount_opt, ERRORS_RO);
1391                         clear_opt(sbi->s_mount_opt, ERRORS_PANIC);
1392                         set_opt(sbi->s_mount_opt, ERRORS_CONT);
1393                         break;
1394                 case Opt_nouid32:
1395                         set_opt(sbi->s_mount_opt, NO_UID32);
1396                         break;
1397                 case Opt_debug:
1398                         set_opt(sbi->s_mount_opt, DEBUG);
1399                         break;
1400                 case Opt_oldalloc:
1401                         set_opt(sbi->s_mount_opt, OLDALLOC);
1402                         break;
1403                 case Opt_orlov:
1404                         clear_opt(sbi->s_mount_opt, OLDALLOC);
1405                         break;
1406 #ifdef CONFIG_EXT4_FS_XATTR
1407                 case Opt_user_xattr:
1408                         set_opt(sbi->s_mount_opt, XATTR_USER);
1409                         break;
1410                 case Opt_nouser_xattr:
1411                         clear_opt(sbi->s_mount_opt, XATTR_USER);
1412                         break;
1413 #else
1414                 case Opt_user_xattr:
1415                 case Opt_nouser_xattr:
1416                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1417                         break;
1418 #endif
1419 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1420                 case Opt_acl:
1421                         set_opt(sbi->s_mount_opt, POSIX_ACL);
1422                         break;
1423                 case Opt_noacl:
1424                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
1425                         break;
1426 #else
1427                 case Opt_acl:
1428                 case Opt_noacl:
1429                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1430                         break;
1431 #endif
1432                 case Opt_journal_update:
1433                         /* @@@ FIXME */
1434                         /* Eventually we will want to be able to create
1435                            a journal file here.  For now, only allow the
1436                            user to specify an existing inode to be the
1437                            journal file. */
1438                         if (is_remount) {
1439                                 ext4_msg(sb, KERN_ERR,
1440                                          "Cannot specify journal on remount");
1441                                 return 0;
1442                         }
1443                         set_opt(sbi->s_mount_opt, UPDATE_JOURNAL);
1444                         break;
1445                 case Opt_journal_dev:
1446                         if (is_remount) {
1447                                 ext4_msg(sb, KERN_ERR,
1448                                         "Cannot specify journal on remount");
1449                                 return 0;
1450                         }
1451                         if (match_int(&args[0], &option))
1452                                 return 0;
1453                         *journal_devnum = option;
1454                         break;
1455                 case Opt_journal_checksum:
1456                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1457                         break;
1458                 case Opt_journal_async_commit:
1459                         set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
1460                         set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
1461                         break;
1462                 case Opt_noload:
1463                         set_opt(sbi->s_mount_opt, NOLOAD);
1464                         break;
1465                 case Opt_commit:
1466                         if (match_int(&args[0], &option))
1467                                 return 0;
1468                         if (option < 0)
1469                                 return 0;
1470                         if (option == 0)
1471                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1472                         sbi->s_commit_interval = HZ * option;
1473                         break;
1474                 case Opt_max_batch_time:
1475                         if (match_int(&args[0], &option))
1476                                 return 0;
1477                         if (option < 0)
1478                                 return 0;
1479                         if (option == 0)
1480                                 option = EXT4_DEF_MAX_BATCH_TIME;
1481                         sbi->s_max_batch_time = option;
1482                         break;
1483                 case Opt_min_batch_time:
1484                         if (match_int(&args[0], &option))
1485                                 return 0;
1486                         if (option < 0)
1487                                 return 0;
1488                         sbi->s_min_batch_time = option;
1489                         break;
1490                 case Opt_data_journal:
1491                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1492                         goto datacheck;
1493                 case Opt_data_ordered:
1494                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1495                         goto datacheck;
1496                 case Opt_data_writeback:
1497                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1498                 datacheck:
1499                         if (is_remount) {
1500                                 if (test_opt(sb, DATA_FLAGS) != data_opt) {
1501                                         ext4_msg(sb, KERN_ERR,
1502                                                 "Cannot change data mode on remount");
1503                                         return 0;
1504                                 }
1505                         } else {
1506                                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
1507                                 sbi->s_mount_opt |= data_opt;
1508                         }
1509                         break;
1510                 case Opt_data_err_abort:
1511                         set_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1512                         break;
1513                 case Opt_data_err_ignore:
1514                         clear_opt(sbi->s_mount_opt, DATA_ERR_ABORT);
1515                         break;
1516 #ifdef CONFIG_QUOTA
1517                 case Opt_usrjquota:
1518                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1519                                 return 0;
1520                         break;
1521                 case Opt_grpjquota:
1522                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1523                                 return 0;
1524                         break;
1525                 case Opt_offusrjquota:
1526                         if (!clear_qf_name(sb, USRQUOTA))
1527                                 return 0;
1528                         break;
1529                 case Opt_offgrpjquota:
1530                         if (!clear_qf_name(sb, GRPQUOTA))
1531                                 return 0;
1532                         break;
1533
1534                 case Opt_jqfmt_vfsold:
1535                         qfmt = QFMT_VFS_OLD;
1536                         goto set_qf_format;
1537                 case Opt_jqfmt_vfsv0:
1538                         qfmt = QFMT_VFS_V0;
1539                         goto set_qf_format;
1540                 case Opt_jqfmt_vfsv1:
1541                         qfmt = QFMT_VFS_V1;
1542 set_qf_format:
1543                         if (sb_any_quota_loaded(sb) &&
1544                             sbi->s_jquota_fmt != qfmt) {
1545                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1546                                         "journaled quota options when "
1547                                         "quota turned on");
1548                                 return 0;
1549                         }
1550                         sbi->s_jquota_fmt = qfmt;
1551                         break;
1552                 case Opt_quota:
1553                 case Opt_usrquota:
1554                         set_opt(sbi->s_mount_opt, QUOTA);
1555                         set_opt(sbi->s_mount_opt, USRQUOTA);
1556                         break;
1557                 case Opt_grpquota:
1558                         set_opt(sbi->s_mount_opt, QUOTA);
1559                         set_opt(sbi->s_mount_opt, GRPQUOTA);
1560                         break;
1561                 case Opt_noquota:
1562                         if (sb_any_quota_loaded(sb)) {
1563                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1564                                         "options when quota turned on");
1565                                 return 0;
1566                         }
1567                         clear_opt(sbi->s_mount_opt, QUOTA);
1568                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1569                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1570                         break;
1571 #else
1572                 case Opt_quota:
1573                 case Opt_usrquota:
1574                 case Opt_grpquota:
1575                         ext4_msg(sb, KERN_ERR,
1576                                 "quota options not supported");
1577                         break;
1578                 case Opt_usrjquota:
1579                 case Opt_grpjquota:
1580                 case Opt_offusrjquota:
1581                 case Opt_offgrpjquota:
1582                 case Opt_jqfmt_vfsold:
1583                 case Opt_jqfmt_vfsv0:
1584                 case Opt_jqfmt_vfsv1:
1585                         ext4_msg(sb, KERN_ERR,
1586                                 "journaled quota options not supported");
1587                         break;
1588                 case Opt_noquota:
1589                         break;
1590 #endif
1591                 case Opt_abort:
1592                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1593                         break;
1594                 case Opt_nobarrier:
1595                         clear_opt(sbi->s_mount_opt, BARRIER);
1596                         break;
1597                 case Opt_barrier:
1598                         if (args[0].from) {
1599                                 if (match_int(&args[0], &option))
1600                                         return 0;
1601                         } else
1602                                 option = 1;     /* No argument, default to 1 */
1603                         if (option)
1604                                 set_opt(sbi->s_mount_opt, BARRIER);
1605                         else
1606                                 clear_opt(sbi->s_mount_opt, BARRIER);
1607                         break;
1608                 case Opt_ignore:
1609                         break;
1610                 case Opt_resize:
1611                         if (!is_remount) {
1612                                 ext4_msg(sb, KERN_ERR,
1613                                         "resize option only available "
1614                                         "for remount");
1615                                 return 0;
1616                         }
1617                         if (match_int(&args[0], &option) != 0)
1618                                 return 0;
1619                         *n_blocks_count = option;
1620                         break;
1621                 case Opt_nobh:
1622                         set_opt(sbi->s_mount_opt, NOBH);
1623                         break;
1624                 case Opt_bh:
1625                         clear_opt(sbi->s_mount_opt, NOBH);
1626                         break;
1627                 case Opt_i_version:
1628                         set_opt(sbi->s_mount_opt, I_VERSION);
1629                         sb->s_flags |= MS_I_VERSION;
1630                         break;
1631                 case Opt_nodelalloc:
1632                         clear_opt(sbi->s_mount_opt, DELALLOC);
1633                         break;
1634                 case Opt_stripe:
1635                         if (match_int(&args[0], &option))
1636                                 return 0;
1637                         if (option < 0)
1638                                 return 0;
1639                         sbi->s_stripe = option;
1640                         break;
1641                 case Opt_delalloc:
1642                         set_opt(sbi->s_mount_opt, DELALLOC);
1643                         break;
1644                 case Opt_block_validity:
1645                         set_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1646                         break;
1647                 case Opt_noblock_validity:
1648                         clear_opt(sbi->s_mount_opt, BLOCK_VALIDITY);
1649                         break;
1650                 case Opt_inode_readahead_blks:
1651                         if (match_int(&args[0], &option))
1652                                 return 0;
1653                         if (option < 0 || option > (1 << 30))
1654                                 return 0;
1655                         if (!is_power_of_2(option)) {
1656                                 ext4_msg(sb, KERN_ERR,
1657                                          "EXT4-fs: inode_readahead_blks"
1658                                          " must be a power of 2");
1659                                 return 0;
1660                         }
1661                         sbi->s_inode_readahead_blks = option;
1662                         break;
1663                 case Opt_journal_ioprio:
1664                         if (match_int(&args[0], &option))
1665                                 return 0;
1666                         if (option < 0 || option > 7)
1667                                 break;
1668                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1669                                                             option);
1670                         break;
1671                 case Opt_noauto_da_alloc:
1672                         set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1673                         break;
1674                 case Opt_auto_da_alloc:
1675                         if (args[0].from) {
1676                                 if (match_int(&args[0], &option))
1677                                         return 0;
1678                         } else
1679                                 option = 1;     /* No argument, default to 1 */
1680                         if (option)
1681                                 clear_opt(sbi->s_mount_opt, NO_AUTO_DA_ALLOC);
1682                         else
1683                                 set_opt(sbi->s_mount_opt,NO_AUTO_DA_ALLOC);
1684                         break;
1685                 case Opt_discard:
1686                         set_opt(sbi->s_mount_opt, DISCARD);
1687                         break;
1688                 case Opt_nodiscard:
1689                         clear_opt(sbi->s_mount_opt, DISCARD);
1690                         break;
1691                 case Opt_dioread_nolock:
1692                         set_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1693                         break;
1694                 case Opt_dioread_lock:
1695                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
1696                         break;
1697                 default:
1698                         ext4_msg(sb, KERN_ERR,
1699                                "Unrecognized mount option \"%s\" "
1700                                "or missing value", p);
1701                         return 0;
1702                 }
1703         }
1704 #ifdef CONFIG_QUOTA
1705         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1706                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1707                         clear_opt(sbi->s_mount_opt, USRQUOTA);
1708
1709                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1710                         clear_opt(sbi->s_mount_opt, GRPQUOTA);
1711
1712                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1713                         ext4_msg(sb, KERN_ERR, "old and new quota "
1714                                         "format mixing");
1715                         return 0;
1716                 }
1717
1718                 if (!sbi->s_jquota_fmt) {
1719                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1720                                         "not specified");
1721                         return 0;
1722                 }
1723         } else {
1724                 if (sbi->s_jquota_fmt) {
1725                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1726                                         "specified with no journaling "
1727                                         "enabled");
1728                         return 0;
1729                 }
1730         }
1731 #endif
1732         return 1;
1733 }
1734
1735 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1736                             int read_only)
1737 {
1738         struct ext4_sb_info *sbi = EXT4_SB(sb);
1739         int res = 0;
1740
1741         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1742                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1743                          "forcing read-only mode");
1744                 res = MS_RDONLY;
1745         }
1746         if (read_only)
1747                 return res;
1748         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1749                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1750                          "running e2fsck is recommended");
1751         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1752                 ext4_msg(sb, KERN_WARNING,
1753                          "warning: mounting fs with errors, "
1754                          "running e2fsck is recommended");
1755         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
1756                  le16_to_cpu(es->s_mnt_count) >=
1757                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1758                 ext4_msg(sb, KERN_WARNING,
1759                          "warning: maximal mount count reached, "
1760                          "running e2fsck is recommended");
1761         else if (le32_to_cpu(es->s_checkinterval) &&
1762                 (le32_to_cpu(es->s_lastcheck) +
1763                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1764                 ext4_msg(sb, KERN_WARNING,
1765                          "warning: checktime reached, "
1766                          "running e2fsck is recommended");
1767         if (!sbi->s_journal)
1768                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1769         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1770                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1771         le16_add_cpu(&es->s_mnt_count, 1);
1772         es->s_mtime = cpu_to_le32(get_seconds());
1773         ext4_update_dynamic_rev(sb);
1774         if (sbi->s_journal)
1775                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1776
1777         ext4_commit_super(sb, 1);
1778         if (test_opt(sb, DEBUG))
1779                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1780                                 "bpg=%lu, ipg=%lu, mo=%04x]\n",
1781                         sb->s_blocksize,
1782                         sbi->s_groups_count,
1783                         EXT4_BLOCKS_PER_GROUP(sb),
1784                         EXT4_INODES_PER_GROUP(sb),
1785                         sbi->s_mount_opt);
1786
1787         return res;
1788 }
1789
1790 static int ext4_fill_flex_info(struct super_block *sb)
1791 {
1792         struct ext4_sb_info *sbi = EXT4_SB(sb);
1793         struct ext4_group_desc *gdp = NULL;
1794         ext4_group_t flex_group_count;
1795         ext4_group_t flex_group;
1796         int groups_per_flex = 0;
1797         size_t size;
1798         int i;
1799
1800         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1801         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1802
1803         if (groups_per_flex < 2) {
1804                 sbi->s_log_groups_per_flex = 0;
1805                 return 1;
1806         }
1807
1808         /* We allocate both existing and potentially added groups */
1809         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1810                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1811                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1812         size = flex_group_count * sizeof(struct flex_groups);
1813         sbi->s_flex_groups = kzalloc(size, GFP_KERNEL);
1814         if (sbi->s_flex_groups == NULL) {
1815                 sbi->s_flex_groups = vmalloc(size);
1816                 if (sbi->s_flex_groups)
1817                         memset(sbi->s_flex_groups, 0, size);
1818         }
1819         if (sbi->s_flex_groups == NULL) {
1820                 ext4_msg(sb, KERN_ERR, "not enough memory for "
1821                                 "%u flex groups", flex_group_count);
1822                 goto failed;
1823         }
1824
1825         for (i = 0; i < sbi->s_groups_count; i++) {
1826                 gdp = ext4_get_group_desc(sb, i, NULL);
1827
1828                 flex_group = ext4_flex_group(sbi, i);
1829                 atomic_add(ext4_free_inodes_count(sb, gdp),
1830                            &sbi->s_flex_groups[flex_group].free_inodes);
1831                 atomic_add(ext4_free_blks_count(sb, gdp),
1832                            &sbi->s_flex_groups[flex_group].free_blocks);
1833                 atomic_add(ext4_used_dirs_count(sb, gdp),
1834                            &sbi->s_flex_groups[flex_group].used_dirs);
1835         }
1836
1837         return 1;
1838 failed:
1839         return 0;
1840 }
1841
1842 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1843                             struct ext4_group_desc *gdp)
1844 {
1845         __u16 crc = 0;
1846
1847         if (sbi->s_es->s_feature_ro_compat &
1848             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1849                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1850                 __le32 le_group = cpu_to_le32(block_group);
1851
1852                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1853                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1854                 crc = crc16(crc, (__u8 *)gdp, offset);
1855                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1856                 /* for checksum of struct ext4_group_desc do the rest...*/
1857                 if ((sbi->s_es->s_feature_incompat &
1858                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1859                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1860                         crc = crc16(crc, (__u8 *)gdp + offset,
1861                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1862                                         offset);
1863         }
1864
1865         return cpu_to_le16(crc);
1866 }
1867
1868 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1869                                 struct ext4_group_desc *gdp)
1870 {
1871         if ((sbi->s_es->s_feature_ro_compat &
1872              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1873             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1874                 return 0;
1875
1876         return 1;
1877 }
1878
1879 /* Called at mount-time, super-block is locked */
1880 static int ext4_check_descriptors(struct super_block *sb)
1881 {
1882         struct ext4_sb_info *sbi = EXT4_SB(sb);
1883         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1884         ext4_fsblk_t last_block;
1885         ext4_fsblk_t block_bitmap;
1886         ext4_fsblk_t inode_bitmap;
1887         ext4_fsblk_t inode_table;
1888         int flexbg_flag = 0;
1889         ext4_group_t i;
1890
1891         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1892                 flexbg_flag = 1;
1893
1894         ext4_debug("Checking group descriptors");
1895
1896         for (i = 0; i < sbi->s_groups_count; i++) {
1897                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1898
1899                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1900                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1901                 else
1902                         last_block = first_block +
1903                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1904
1905                 block_bitmap = ext4_block_bitmap(sb, gdp);
1906                 if (block_bitmap < first_block || block_bitmap > last_block) {
1907                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1908                                "Block bitmap for group %u not in group "
1909                                "(block %llu)!", i, block_bitmap);
1910                         return 0;
1911                 }
1912                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1913                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1914                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1915                                "Inode bitmap for group %u not in group "
1916                                "(block %llu)!", i, inode_bitmap);
1917                         return 0;
1918                 }
1919                 inode_table = ext4_inode_table(sb, gdp);
1920                 if (inode_table < first_block ||
1921                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1922                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1923                                "Inode table for group %u not in group "
1924                                "(block %llu)!", i, inode_table);
1925                         return 0;
1926                 }
1927                 ext4_lock_group(sb, i);
1928                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1929                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1930                                  "Checksum for group %u failed (%u!=%u)",
1931                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1932                                      gdp)), le16_to_cpu(gdp->bg_checksum));
1933                         if (!(sb->s_flags & MS_RDONLY)) {
1934                                 ext4_unlock_group(sb, i);
1935                                 return 0;
1936                         }
1937                 }
1938                 ext4_unlock_group(sb, i);
1939                 if (!flexbg_flag)
1940                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1941         }
1942
1943         ext4_free_blocks_count_set(sbi->s_es, ext4_count_free_blocks(sb));
1944         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
1945         return 1;
1946 }
1947
1948 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
1949  * the superblock) which were deleted from all directories, but held open by
1950  * a process at the time of a crash.  We walk the list and try to delete these
1951  * inodes at recovery time (only with a read-write filesystem).
1952  *
1953  * In order to keep the orphan inode chain consistent during traversal (in
1954  * case of crash during recovery), we link each inode into the superblock
1955  * orphan list_head and handle it the same way as an inode deletion during
1956  * normal operation (which journals the operations for us).
1957  *
1958  * We only do an iget() and an iput() on each inode, which is very safe if we
1959  * accidentally point at an in-use or already deleted inode.  The worst that
1960  * can happen in this case is that we get a "bit already cleared" message from
1961  * ext4_free_inode().  The only reason we would point at a wrong inode is if
1962  * e2fsck was run on this filesystem, and it must have already done the orphan
1963  * inode cleanup for us, so we can safely abort without any further action.
1964  */
1965 static void ext4_orphan_cleanup(struct super_block *sb,
1966                                 struct ext4_super_block *es)
1967 {
1968         unsigned int s_flags = sb->s_flags;
1969         int nr_orphans = 0, nr_truncates = 0;
1970 #ifdef CONFIG_QUOTA
1971         int i;
1972 #endif
1973         if (!es->s_last_orphan) {
1974                 jbd_debug(4, "no orphan inodes to clean up\n");
1975                 return;
1976         }
1977
1978         if (bdev_read_only(sb->s_bdev)) {
1979                 ext4_msg(sb, KERN_ERR, "write access "
1980                         "unavailable, skipping orphan cleanup");
1981                 return;
1982         }
1983
1984         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
1985                 if (es->s_last_orphan)
1986                         jbd_debug(1, "Errors on filesystem, "
1987                                   "clearing orphan list.\n");
1988                 es->s_last_orphan = 0;
1989                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1990                 return;
1991         }
1992
1993         if (s_flags & MS_RDONLY) {
1994                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
1995                 sb->s_flags &= ~MS_RDONLY;
1996         }
1997 #ifdef CONFIG_QUOTA
1998         /* Needed for iput() to work correctly and not trash data */
1999         sb->s_flags |= MS_ACTIVE;
2000         /* Turn on quotas so that they are updated correctly */
2001         for (i = 0; i < MAXQUOTAS; i++) {
2002                 if (EXT4_SB(sb)->s_qf_names[i]) {
2003                         int ret = ext4_quota_on_mount(sb, i);
2004                         if (ret < 0)
2005                                 ext4_msg(sb, KERN_ERR,
2006                                         "Cannot turn on journaled "
2007                                         "quota: error %d", ret);
2008                 }
2009         }
2010 #endif
2011
2012         while (es->s_last_orphan) {
2013                 struct inode *inode;
2014
2015                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2016                 if (IS_ERR(inode)) {
2017                         es->s_last_orphan = 0;
2018                         break;
2019                 }
2020
2021                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2022                 dquot_initialize(inode);
2023                 if (inode->i_nlink) {
2024                         ext4_msg(sb, KERN_DEBUG,
2025                                 "%s: truncating inode %lu to %lld bytes",
2026                                 __func__, inode->i_ino, inode->i_size);
2027                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2028                                   inode->i_ino, inode->i_size);
2029                         ext4_truncate(inode);
2030                         nr_truncates++;
2031                 } else {
2032                         ext4_msg(sb, KERN_DEBUG,
2033                                 "%s: deleting unreferenced inode %lu",
2034                                 __func__, inode->i_ino);
2035                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2036                                   inode->i_ino);
2037                         nr_orphans++;
2038                 }
2039                 iput(inode);  /* The delete magic happens here! */
2040         }
2041
2042 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2043
2044         if (nr_orphans)
2045                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2046                        PLURAL(nr_orphans));
2047         if (nr_truncates)
2048                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2049                        PLURAL(nr_truncates));
2050 #ifdef CONFIG_QUOTA
2051         /* Turn quotas off */
2052         for (i = 0; i < MAXQUOTAS; i++) {
2053                 if (sb_dqopt(sb)->files[i])
2054                         vfs_quota_off(sb, i, 0);
2055         }
2056 #endif
2057         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2058 }
2059
2060 /*
2061  * Maximal extent format file size.
2062  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2063  * extent format containers, within a sector_t, and within i_blocks
2064  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2065  * so that won't be a limiting factor.
2066  *
2067  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2068  */
2069 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2070 {
2071         loff_t res;
2072         loff_t upper_limit = MAX_LFS_FILESIZE;
2073
2074         /* small i_blocks in vfs inode? */
2075         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2076                 /*
2077                  * CONFIG_LBDAF is not enabled implies the inode
2078                  * i_block represent total blocks in 512 bytes
2079                  * 32 == size of vfs inode i_blocks * 8
2080                  */
2081                 upper_limit = (1LL << 32) - 1;
2082
2083                 /* total blocks in file system block size */
2084                 upper_limit >>= (blkbits - 9);
2085                 upper_limit <<= blkbits;
2086         }
2087
2088         /* 32-bit extent-start container, ee_block */
2089         res = 1LL << 32;
2090         res <<= blkbits;
2091         res -= 1;
2092
2093         /* Sanity check against vm- & vfs- imposed limits */
2094         if (res > upper_limit)
2095                 res = upper_limit;
2096
2097         return res;
2098 }
2099
2100 /*
2101  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2102  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2103  * We need to be 1 filesystem block less than the 2^48 sector limit.
2104  */
2105 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2106 {
2107         loff_t res = EXT4_NDIR_BLOCKS;
2108         int meta_blocks;
2109         loff_t upper_limit;
2110         /* This is calculated to be the largest file size for a dense, block
2111          * mapped file such that the file's total number of 512-byte sectors,
2112          * including data and all indirect blocks, does not exceed (2^48 - 1).
2113          *
2114          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2115          * number of 512-byte sectors of the file.
2116          */
2117
2118         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2119                 /*
2120                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2121                  * the inode i_block field represents total file blocks in
2122                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2123                  */
2124                 upper_limit = (1LL << 32) - 1;
2125
2126                 /* total blocks in file system block size */
2127                 upper_limit >>= (bits - 9);
2128
2129         } else {
2130                 /*
2131                  * We use 48 bit ext4_inode i_blocks
2132                  * With EXT4_HUGE_FILE_FL set the i_blocks
2133                  * represent total number of blocks in
2134                  * file system block size
2135                  */
2136                 upper_limit = (1LL << 48) - 1;
2137
2138         }
2139
2140         /* indirect blocks */
2141         meta_blocks = 1;
2142         /* double indirect blocks */
2143         meta_blocks += 1 + (1LL << (bits-2));
2144         /* tripple indirect blocks */
2145         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2146
2147         upper_limit -= meta_blocks;
2148         upper_limit <<= bits;
2149
2150         res += 1LL << (bits-2);
2151         res += 1LL << (2*(bits-2));
2152         res += 1LL << (3*(bits-2));
2153         res <<= bits;
2154         if (res > upper_limit)
2155                 res = upper_limit;
2156
2157         if (res > MAX_LFS_FILESIZE)
2158                 res = MAX_LFS_FILESIZE;
2159
2160         return res;
2161 }
2162
2163 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2164                                    ext4_fsblk_t logical_sb_block, int nr)
2165 {
2166         struct ext4_sb_info *sbi = EXT4_SB(sb);
2167         ext4_group_t bg, first_meta_bg;
2168         int has_super = 0;
2169
2170         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2171
2172         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2173             nr < first_meta_bg)
2174                 return logical_sb_block + nr + 1;
2175         bg = sbi->s_desc_per_block * nr;
2176         if (ext4_bg_has_super(sb, bg))
2177                 has_super = 1;
2178
2179         return (has_super + ext4_group_first_block_no(sb, bg));
2180 }
2181
2182 /**
2183  * ext4_get_stripe_size: Get the stripe size.
2184  * @sbi: In memory super block info
2185  *
2186  * If we have specified it via mount option, then
2187  * use the mount option value. If the value specified at mount time is
2188  * greater than the blocks per group use the super block value.
2189  * If the super block value is greater than blocks per group return 0.
2190  * Allocator needs it be less than blocks per group.
2191  *
2192  */
2193 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2194 {
2195         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2196         unsigned long stripe_width =
2197                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2198
2199         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2200                 return sbi->s_stripe;
2201
2202         if (stripe_width <= sbi->s_blocks_per_group)
2203                 return stripe_width;
2204
2205         if (stride <= sbi->s_blocks_per_group)
2206                 return stride;
2207
2208         return 0;
2209 }
2210
2211 /* sysfs supprt */
2212
2213 struct ext4_attr {
2214         struct attribute attr;
2215         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2216         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *, 
2217                          const char *, size_t);
2218         int offset;
2219 };
2220
2221 static int parse_strtoul(const char *buf,
2222                 unsigned long max, unsigned long *value)
2223 {
2224         char *endp;
2225
2226         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2227         endp = skip_spaces(endp);
2228         if (*endp || *value > max)
2229                 return -EINVAL;
2230
2231         return 0;
2232 }
2233
2234 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2235                                               struct ext4_sb_info *sbi,
2236                                               char *buf)
2237 {
2238         return snprintf(buf, PAGE_SIZE, "%llu\n",
2239                         (s64) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2240 }
2241
2242 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2243                                          struct ext4_sb_info *sbi, char *buf)
2244 {
2245         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2246
2247         return snprintf(buf, PAGE_SIZE, "%lu\n",
2248                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2249                          sbi->s_sectors_written_start) >> 1);
2250 }
2251
2252 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2253                                           struct ext4_sb_info *sbi, char *buf)
2254 {
2255         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2256
2257         return snprintf(buf, PAGE_SIZE, "%llu\n",
2258                         (unsigned long long)(sbi->s_kbytes_written +
2259                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2260                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2261 }
2262
2263 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2264                                           struct ext4_sb_info *sbi,
2265                                           const char *buf, size_t count)
2266 {
2267         unsigned long t;
2268
2269         if (parse_strtoul(buf, 0x40000000, &t))
2270                 return -EINVAL;
2271
2272         if (!is_power_of_2(t))
2273                 return -EINVAL;
2274
2275         sbi->s_inode_readahead_blks = t;
2276         return count;
2277 }
2278
2279 static ssize_t sbi_ui_show(struct ext4_attr *a,
2280                            struct ext4_sb_info *sbi, char *buf)
2281 {
2282         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2283
2284         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2285 }
2286
2287 static ssize_t sbi_ui_store(struct ext4_attr *a,
2288                             struct ext4_sb_info *sbi,
2289                             const char *buf, size_t count)
2290 {
2291         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2292         unsigned long t;
2293
2294         if (parse_strtoul(buf, 0xffffffff, &t))
2295                 return -EINVAL;
2296         *ui = t;
2297         return count;
2298 }
2299
2300 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2301 static struct ext4_attr ext4_attr_##_name = {                   \
2302         .attr = {.name = __stringify(_name), .mode = _mode },   \
2303         .show   = _show,                                        \
2304         .store  = _store,                                       \
2305         .offset = offsetof(struct ext4_sb_info, _elname),       \
2306 }
2307 #define EXT4_ATTR(name, mode, show, store) \
2308 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2309
2310 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2311 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2312 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2313         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2314 #define ATTR_LIST(name) &ext4_attr_##name.attr
2315
2316 EXT4_RO_ATTR(delayed_allocation_blocks);
2317 EXT4_RO_ATTR(session_write_kbytes);
2318 EXT4_RO_ATTR(lifetime_write_kbytes);
2319 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2320                  inode_readahead_blks_store, s_inode_readahead_blks);
2321 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2322 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2323 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2324 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2325 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2326 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2327 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2328 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2329
2330 static struct attribute *ext4_attrs[] = {
2331         ATTR_LIST(delayed_allocation_blocks),
2332         ATTR_LIST(session_write_kbytes),
2333         ATTR_LIST(lifetime_write_kbytes),
2334         ATTR_LIST(inode_readahead_blks),
2335         ATTR_LIST(inode_goal),
2336         ATTR_LIST(mb_stats),
2337         ATTR_LIST(mb_max_to_scan),
2338         ATTR_LIST(mb_min_to_scan),
2339         ATTR_LIST(mb_order2_req),
2340         ATTR_LIST(mb_stream_req),
2341         ATTR_LIST(mb_group_prealloc),
2342         ATTR_LIST(max_writeback_mb_bump),
2343         NULL,
2344 };
2345
2346 static ssize_t ext4_attr_show(struct kobject *kobj,
2347                               struct attribute *attr, char *buf)
2348 {
2349         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2350                                                 s_kobj);
2351         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2352
2353         return a->show ? a->show(a, sbi, buf) : 0;
2354 }
2355
2356 static ssize_t ext4_attr_store(struct kobject *kobj,
2357                                struct attribute *attr,
2358                                const char *buf, size_t len)
2359 {
2360         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2361                                                 s_kobj);
2362         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2363
2364         return a->store ? a->store(a, sbi, buf, len) : 0;
2365 }
2366
2367 static void ext4_sb_release(struct kobject *kobj)
2368 {
2369         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2370                                                 s_kobj);
2371         complete(&sbi->s_kobj_unregister);
2372 }
2373
2374
2375 static const struct sysfs_ops ext4_attr_ops = {
2376         .show   = ext4_attr_show,
2377         .store  = ext4_attr_store,
2378 };
2379
2380 static struct kobj_type ext4_ktype = {
2381         .default_attrs  = ext4_attrs,
2382         .sysfs_ops      = &ext4_attr_ops,
2383         .release        = ext4_sb_release,
2384 };
2385
2386 /*
2387  * Check whether this filesystem can be mounted based on
2388  * the features present and the RDONLY/RDWR mount requested.
2389  * Returns 1 if this filesystem can be mounted as requested,
2390  * 0 if it cannot be.
2391  */
2392 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2393 {
2394         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2395                 ext4_msg(sb, KERN_ERR,
2396                         "Couldn't mount because of "
2397                         "unsupported optional features (%x)",
2398                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2399                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2400                 return 0;
2401         }
2402
2403         if (readonly)
2404                 return 1;
2405
2406         /* Check that feature set is OK for a read-write mount */
2407         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2408                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2409                          "unsupported optional features (%x)",
2410                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2411                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2412                 return 0;
2413         }
2414         /*
2415          * Large file size enabled file system can only be mounted
2416          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2417          */
2418         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2419                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2420                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2421                                  "cannot be mounted RDWR without "
2422                                  "CONFIG_LBDAF");
2423                         return 0;
2424                 }
2425         }
2426         return 1;
2427 }
2428
2429 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2430                                 __releases(kernel_lock)
2431                                 __acquires(kernel_lock)
2432 {
2433         struct buffer_head *bh;
2434         struct ext4_super_block *es = NULL;
2435         struct ext4_sb_info *sbi;
2436         ext4_fsblk_t block;
2437         ext4_fsblk_t sb_block = get_sb_block(&data);
2438         ext4_fsblk_t logical_sb_block;
2439         unsigned long offset = 0;
2440         unsigned long journal_devnum = 0;
2441         unsigned long def_mount_opts;
2442         struct inode *root;
2443         char *cp;
2444         const char *descr;
2445         int ret = -EINVAL;
2446         int blocksize;
2447         unsigned int db_count;
2448         unsigned int i;
2449         int needs_recovery, has_huge_files;
2450         __u64 blocks_count;
2451         int err;
2452         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2453
2454         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2455         if (!sbi)
2456                 return -ENOMEM;
2457
2458         sbi->s_blockgroup_lock =
2459                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
2460         if (!sbi->s_blockgroup_lock) {
2461                 kfree(sbi);
2462                 return -ENOMEM;
2463         }
2464         sb->s_fs_info = sbi;
2465         sbi->s_mount_opt = 0;
2466         sbi->s_resuid = EXT4_DEF_RESUID;
2467         sbi->s_resgid = EXT4_DEF_RESGID;
2468         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
2469         sbi->s_sb_block = sb_block;
2470         sbi->s_sectors_written_start = part_stat_read(sb->s_bdev->bd_part,
2471                                                       sectors[1]);
2472
2473         unlock_kernel();
2474
2475         /* Cleanup superblock name */
2476         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
2477                 *cp = '!';
2478
2479         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
2480         if (!blocksize) {
2481                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
2482                 goto out_fail;
2483         }
2484
2485         /*
2486          * The ext4 superblock will not be buffer aligned for other than 1kB
2487          * block sizes.  We need to calculate the offset from buffer start.
2488          */
2489         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
2490                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2491                 offset = do_div(logical_sb_block, blocksize);
2492         } else {
2493                 logical_sb_block = sb_block;
2494         }
2495
2496         if (!(bh = sb_bread(sb, logical_sb_block))) {
2497                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
2498                 goto out_fail;
2499         }
2500         /*
2501          * Note: s_es must be initialized as soon as possible because
2502          *       some ext4 macro-instructions depend on its value
2503          */
2504         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
2505         sbi->s_es = es;
2506         sb->s_magic = le16_to_cpu(es->s_magic);
2507         if (sb->s_magic != EXT4_SUPER_MAGIC)
2508                 goto cantfind_ext4;
2509         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
2510
2511         /* Set defaults before we parse the mount options */
2512         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
2513         if (def_mount_opts & EXT4_DEFM_DEBUG)
2514                 set_opt(sbi->s_mount_opt, DEBUG);
2515         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
2516                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
2517                         "2.6.38");
2518                 set_opt(sbi->s_mount_opt, GRPID);
2519         }
2520         if (def_mount_opts & EXT4_DEFM_UID16)
2521                 set_opt(sbi->s_mount_opt, NO_UID32);
2522 #ifdef CONFIG_EXT4_FS_XATTR
2523         if (def_mount_opts & EXT4_DEFM_XATTR_USER)
2524                 set_opt(sbi->s_mount_opt, XATTR_USER);
2525 #endif
2526 #ifdef CONFIG_EXT4_FS_POSIX_ACL
2527         if (def_mount_opts & EXT4_DEFM_ACL)
2528                 set_opt(sbi->s_mount_opt, POSIX_ACL);
2529 #endif
2530         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
2531                 set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2532         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
2533                 set_opt(sbi->s_mount_opt, ORDERED_DATA);
2534         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
2535                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
2536
2537         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
2538                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
2539         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
2540                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
2541         else
2542                 set_opt(sbi->s_mount_opt, ERRORS_RO);
2543
2544         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
2545         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
2546         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
2547         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
2548         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
2549
2550         set_opt(sbi->s_mount_opt, BARRIER);
2551
2552         /*
2553          * enable delayed allocation by default
2554          * Use -o nodelalloc to turn it off
2555          */
2556         if (!IS_EXT3_SB(sb))
2557                 set_opt(sbi->s_mount_opt, DELALLOC);
2558
2559         if (!parse_options((char *) data, sb, &journal_devnum,
2560                            &journal_ioprio, NULL, 0))
2561                 goto failed_mount;
2562
2563         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2564                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
2565
2566         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
2567             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
2568              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
2569              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
2570                 ext4_msg(sb, KERN_WARNING,
2571                        "feature flags set on rev 0 fs, "
2572                        "running e2fsck is recommended");
2573
2574         /*
2575          * Check feature flags regardless of the revision level, since we
2576          * previously didn't change the revision level when setting the flags,
2577          * so there is a chance incompat flags are set on a rev 0 filesystem.
2578          */
2579         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
2580                 goto failed_mount;
2581
2582         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
2583
2584         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
2585             blocksize > EXT4_MAX_BLOCK_SIZE) {
2586                 ext4_msg(sb, KERN_ERR,
2587                        "Unsupported filesystem blocksize %d", blocksize);
2588                 goto failed_mount;
2589         }
2590
2591         if (sb->s_blocksize != blocksize) {
2592                 /* Validate the filesystem blocksize */
2593                 if (!sb_set_blocksize(sb, blocksize)) {
2594                         ext4_msg(sb, KERN_ERR, "bad block size %d",
2595                                         blocksize);
2596                         goto failed_mount;
2597                 }
2598
2599                 brelse(bh);
2600                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
2601                 offset = do_div(logical_sb_block, blocksize);
2602                 bh = sb_bread(sb, logical_sb_block);
2603                 if (!bh) {
2604                         ext4_msg(sb, KERN_ERR,
2605                                "Can't read superblock on 2nd try");
2606                         goto failed_mount;
2607                 }
2608                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
2609                 sbi->s_es = es;
2610                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
2611                         ext4_msg(sb, KERN_ERR,
2612                                "Magic mismatch, very weird!");
2613                         goto failed_mount;
2614                 }
2615         }
2616
2617         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
2618                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
2619         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
2620                                                       has_huge_files);
2621         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
2622
2623         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
2624                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
2625                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
2626         } else {
2627                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
2628                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
2629                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
2630                     (!is_power_of_2(sbi->s_inode_size)) ||
2631                     (sbi->s_inode_size > blocksize)) {
2632                         ext4_msg(sb, KERN_ERR,
2633                                "unsupported inode size: %d",
2634                                sbi->s_inode_size);
2635                         goto failed_mount;
2636                 }
2637                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
2638                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
2639         }
2640
2641         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
2642         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
2643                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
2644                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
2645                     !is_power_of_2(sbi->s_desc_size)) {
2646                         ext4_msg(sb, KERN_ERR,
2647                                "unsupported descriptor size %lu",
2648                                sbi->s_desc_size);
2649                         goto failed_mount;
2650                 }
2651         } else
2652                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
2653
2654         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
2655         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
2656         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
2657                 goto cantfind_ext4;
2658
2659         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
2660         if (sbi->s_inodes_per_block == 0)
2661                 goto cantfind_ext4;
2662         sbi->s_itb_per_group = sbi->s_inodes_per_group /
2663                                         sbi->s_inodes_per_block;
2664         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
2665         sbi->s_sbh = bh;
2666         sbi->s_mount_state = le16_to_cpu(es->s_state);
2667         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
2668         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
2669
2670         for (i = 0; i < 4; i++)
2671                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
2672         sbi->s_def_hash_version = es->s_def_hash_version;
2673         i = le32_to_cpu(es->s_flags);
2674         if (i & EXT2_FLAGS_UNSIGNED_HASH)
2675                 sbi->s_hash_unsigned = 3;
2676         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
2677 #ifdef __CHAR_UNSIGNED__
2678                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
2679                 sbi->s_hash_unsigned = 3;
2680 #else
2681                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
2682 #endif
2683                 sb->s_dirt = 1;
2684         }
2685
2686         if (sbi->s_blocks_per_group > blocksize * 8) {
2687                 ext4_msg(sb, KERN_ERR,
2688                        "#blocks per group too big: %lu",
2689                        sbi->s_blocks_per_group);
2690                 goto failed_mount;
2691         }
2692         if (sbi->s_inodes_per_group > blocksize * 8) {
2693                 ext4_msg(sb, KERN_ERR,
2694                        "#inodes per group too big: %lu",
2695                        sbi->s_inodes_per_group);
2696                 goto failed_mount;
2697         }
2698
2699         /*
2700          * Test whether we have more sectors than will fit in sector_t,
2701          * and whether the max offset is addressable by the page cache.
2702          */
2703         if ((ext4_blocks_count(es) >
2704              (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) ||
2705             (ext4_blocks_count(es) >
2706              (pgoff_t)(~0ULL) >> (PAGE_CACHE_SHIFT - sb->s_blocksize_bits))) {
2707                 ext4_msg(sb, KERN_ERR, "filesystem"
2708                          " too large to mount safely on this system");
2709                 if (sizeof(sector_t) < 8)
2710                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
2711                 ret = -EFBIG;
2712                 goto failed_mount;
2713         }
2714
2715         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
2716                 goto cantfind_ext4;
2717
2718         /* check blocks count against device size */
2719         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
2720         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
2721                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
2722                        "exceeds size of device (%llu blocks)",
2723                        ext4_blocks_count(es), blocks_count);
2724                 goto failed_mount;
2725         }
2726
2727         /*
2728          * It makes no sense for the first data block to be beyond the end
2729          * of the filesystem.
2730          */
2731         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
2732                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
2733                          "block %u is beyond end of filesystem (%llu)",
2734                          le32_to_cpu(es->s_first_data_block),
2735                          ext4_blocks_count(es));
2736                 goto failed_mount;
2737         }
2738         blocks_count = (ext4_blocks_count(es) -
2739                         le32_to_cpu(es->s_first_data_block) +
2740                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
2741         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
2742         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
2743                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
2744                        "(block count %llu, first data block %u, "
2745                        "blocks per group %lu)", sbi->s_groups_count,
2746                        ext4_blocks_count(es),
2747                        le32_to_cpu(es->s_first_data_block),
2748                        EXT4_BLOCKS_PER_GROUP(sb));
2749                 goto failed_mount;
2750         }
2751         sbi->s_groups_count = blocks_count;
2752         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
2753                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
2754         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
2755                    EXT4_DESC_PER_BLOCK(sb);
2756         sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
2757                                     GFP_KERNEL);
2758         if (sbi->s_group_desc == NULL) {
2759                 ext4_msg(sb, KERN_ERR, "not enough memory");
2760                 goto failed_mount;
2761         }
2762
2763 #ifdef CONFIG_PROC_FS
2764         if (ext4_proc_root)
2765                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
2766 #endif
2767
2768         bgl_lock_init(sbi->s_blockgroup_lock);
2769
2770         for (i = 0; i < db_count; i++) {
2771                 block = descriptor_loc(sb, logical_sb_block, i);
2772                 sbi->s_group_desc[i] = sb_bread(sb, block);
2773                 if (!sbi->s_group_desc[i]) {
2774                         ext4_msg(sb, KERN_ERR,
2775                                "can't read group descriptor %d", i);
2776                         db_count = i;
2777                         goto failed_mount2;
2778                 }
2779         }
2780         if (!ext4_check_descriptors(sb)) {
2781                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
2782                 goto failed_mount2;
2783         }
2784         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2785                 if (!ext4_fill_flex_info(sb)) {
2786                         ext4_msg(sb, KERN_ERR,
2787                                "unable to initialize "
2788                                "flex_bg meta info!");
2789                         goto failed_mount2;
2790                 }
2791
2792         sbi->s_gdb_count = db_count;
2793         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2794         spin_lock_init(&sbi->s_next_gen_lock);
2795
2796         err = percpu_counter_init(&sbi->s_freeblocks_counter,
2797                         ext4_count_free_blocks(sb));
2798         if (!err) {
2799                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
2800                                 ext4_count_free_inodes(sb));
2801         }
2802         if (!err) {
2803                 err = percpu_counter_init(&sbi->s_dirs_counter,
2804                                 ext4_count_dirs(sb));
2805         }
2806         if (!err) {
2807                 err = percpu_counter_init(&sbi->s_dirtyblocks_counter, 0);
2808         }
2809         if (err) {
2810                 ext4_msg(sb, KERN_ERR, "insufficient memory");
2811                 goto failed_mount3;
2812         }
2813
2814         sbi->s_stripe = ext4_get_stripe_size(sbi);
2815         sbi->s_max_writeback_mb_bump = 128;
2816
2817         /*
2818          * set up enough so that it can read an inode
2819          */
2820         if (!test_opt(sb, NOLOAD) &&
2821             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
2822                 sb->s_op = &ext4_sops;
2823         else
2824                 sb->s_op = &ext4_nojournal_sops;
2825         sb->s_export_op = &ext4_export_ops;
2826         sb->s_xattr = ext4_xattr_handlers;
2827 #ifdef CONFIG_QUOTA
2828         sb->s_qcop = &ext4_qctl_operations;
2829         sb->dq_op = &ext4_quota_operations;
2830 #endif
2831         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
2832         mutex_init(&sbi->s_orphan_lock);
2833         mutex_init(&sbi->s_resize_lock);
2834
2835         sb->s_root = NULL;
2836
2837         needs_recovery = (es->s_last_orphan != 0 ||
2838                           EXT4_HAS_INCOMPAT_FEATURE(sb,
2839                                     EXT4_FEATURE_INCOMPAT_RECOVER));
2840
2841         /*
2842          * The first inode we look at is the journal inode.  Don't try
2843          * root first: it may be modified in the journal!
2844          */
2845         if (!test_opt(sb, NOLOAD) &&
2846             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
2847                 if (ext4_load_journal(sb, es, journal_devnum))
2848                         goto failed_mount3;
2849         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
2850               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
2851                 ext4_msg(sb, KERN_ERR, "required journal recovery "
2852                        "suppressed and not mounted read-only");
2853                 goto failed_mount_wq;
2854         } else {
2855                 clear_opt(sbi->s_mount_opt, DATA_FLAGS);
2856                 set_opt(sbi->s_mount_opt, WRITEBACK_DATA);
2857                 sbi->s_journal = NULL;
2858                 needs_recovery = 0;
2859                 goto no_journal;
2860         }
2861
2862         if (ext4_blocks_count(es) > 0xffffffffULL &&
2863             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
2864                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
2865                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
2866                 goto failed_mount_wq;
2867         }
2868
2869         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2870                 jbd2_journal_set_features(sbi->s_journal,
2871                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2872                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2873         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2874                 jbd2_journal_set_features(sbi->s_journal,
2875                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
2876                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2877                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2878         } else {
2879                 jbd2_journal_clear_features(sbi->s_journal,
2880                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2881                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2882         }
2883
2884         /* We have now updated the journal if required, so we can
2885          * validate the data journaling mode. */
2886         switch (test_opt(sb, DATA_FLAGS)) {
2887         case 0:
2888                 /* No mode set, assume a default based on the journal
2889                  * capabilities: ORDERED_DATA if the journal can
2890                  * cope, else JOURNAL_DATA
2891                  */
2892                 if (jbd2_journal_check_available_features
2893                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
2894                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
2895                 else
2896                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
2897                 break;
2898
2899         case EXT4_MOUNT_ORDERED_DATA:
2900         case EXT4_MOUNT_WRITEBACK_DATA:
2901                 if (!jbd2_journal_check_available_features
2902                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
2903                         ext4_msg(sb, KERN_ERR, "Journal does not support "
2904                                "requested data journaling mode");
2905                         goto failed_mount_wq;
2906                 }
2907         default:
2908                 break;
2909         }
2910         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
2911
2912 no_journal:
2913         if (test_opt(sb, NOBH)) {
2914                 if (!(test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)) {
2915                         ext4_msg(sb, KERN_WARNING, "Ignoring nobh option - "
2916                                 "its supported only with writeback mode");
2917                         clear_opt(sbi->s_mount_opt, NOBH);
2918                 }
2919                 if (test_opt(sb, DIOREAD_NOLOCK)) {
2920                         ext4_msg(sb, KERN_WARNING, "dioread_nolock option is "
2921                                 "not supported with nobh mode");
2922                         goto failed_mount_wq;
2923                 }
2924         }
2925         EXT4_SB(sb)->dio_unwritten_wq = create_workqueue("ext4-dio-unwritten");
2926         if (!EXT4_SB(sb)->dio_unwritten_wq) {
2927                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
2928                 goto failed_mount_wq;
2929         }
2930
2931         /*
2932          * The jbd2_journal_load will have done any necessary log recovery,
2933          * so we can safely mount the rest of the filesystem now.
2934          */
2935
2936         root = ext4_iget(sb, EXT4_ROOT_INO);
2937         if (IS_ERR(root)) {
2938                 ext4_msg(sb, KERN_ERR, "get root inode failed");
2939                 ret = PTR_ERR(root);
2940                 goto failed_mount4;
2941         }
2942         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
2943                 iput(root);
2944                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
2945                 goto failed_mount4;
2946         }
2947         sb->s_root = d_alloc_root(root);
2948         if (!sb->s_root) {
2949                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
2950                 iput(root);
2951                 ret = -ENOMEM;
2952                 goto failed_mount4;
2953         }
2954
2955         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
2956
2957         /* determine the minimum size of new large inodes, if present */
2958         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
2959                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2960                                                      EXT4_GOOD_OLD_INODE_SIZE;
2961                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2962                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
2963                         if (sbi->s_want_extra_isize <
2964                             le16_to_cpu(es->s_want_extra_isize))
2965                                 sbi->s_want_extra_isize =
2966                                         le16_to_cpu(es->s_want_extra_isize);
2967                         if (sbi->s_want_extra_isize <
2968                             le16_to_cpu(es->s_min_extra_isize))
2969                                 sbi->s_want_extra_isize =
2970                                         le16_to_cpu(es->s_min_extra_isize);
2971                 }
2972         }
2973         /* Check if enough inode space is available */
2974         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
2975                                                         sbi->s_inode_size) {
2976                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
2977                                                        EXT4_GOOD_OLD_INODE_SIZE;
2978                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
2979                          "available");
2980         }
2981
2982         if (test_opt(sb, DELALLOC) &&
2983             (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)) {
2984                 ext4_msg(sb, KERN_WARNING, "Ignoring delalloc option - "
2985                          "requested data journaling mode");
2986                 clear_opt(sbi->s_mount_opt, DELALLOC);
2987         }
2988         if (test_opt(sb, DIOREAD_NOLOCK)) {
2989                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
2990                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
2991                                 "option - requested data journaling mode");
2992                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
2993                 }
2994                 if (sb->s_blocksize < PAGE_SIZE) {
2995                         ext4_msg(sb, KERN_WARNING, "Ignoring dioread_nolock "
2996                                 "option - block size is too small");
2997                         clear_opt(sbi->s_mount_opt, DIOREAD_NOLOCK);
2998                 }
2999         }
3000
3001         err = ext4_setup_system_zone(sb);
3002         if (err) {
3003                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3004                          "zone (%d)\n", err);
3005                 goto failed_mount4;
3006         }
3007
3008         ext4_ext_init(sb);
3009         err = ext4_mb_init(sb, needs_recovery);
3010         if (err) {
3011                 ext4_msg(sb, KERN_ERR, "failed to initalize mballoc (%d)",
3012                          err);
3013                 goto failed_mount4;
3014         }
3015
3016         sbi->s_kobj.kset = ext4_kset;
3017         init_completion(&sbi->s_kobj_unregister);
3018         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3019                                    "%s", sb->s_id);
3020         if (err) {
3021                 ext4_mb_release(sb);
3022                 ext4_ext_release(sb);
3023                 goto failed_mount4;
3024         };
3025
3026         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3027         ext4_orphan_cleanup(sb, es);
3028         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3029         if (needs_recovery) {
3030                 ext4_msg(sb, KERN_INFO, "recovery complete");
3031                 ext4_mark_recovery_complete(sb, es);
3032         }
3033         if (EXT4_SB(sb)->s_journal) {
3034                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3035                         descr = " journalled data mode";
3036                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3037                         descr = " ordered data mode";
3038                 else
3039                         descr = " writeback data mode";
3040         } else
3041                 descr = "out journal";
3042
3043         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s", descr);
3044
3045         lock_kernel();
3046         return 0;
3047
3048 cantfind_ext4:
3049         if (!silent)
3050                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3051         goto failed_mount;
3052
3053 failed_mount4:
3054         ext4_msg(sb, KERN_ERR, "mount failed");
3055         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3056 failed_mount_wq:
3057         ext4_release_system_zone(sb);
3058         if (sbi->s_journal) {
3059                 jbd2_journal_destroy(sbi->s_journal);
3060                 sbi->s_journal = NULL;
3061         }
3062 failed_mount3:
3063         if (sbi->s_flex_groups) {
3064                 if (is_vmalloc_addr(sbi->s_flex_groups))
3065                         vfree(sbi->s_flex_groups);
3066                 else
3067                         kfree(sbi->s_flex_groups);
3068         }
3069         percpu_counter_destroy(&sbi->s_freeblocks_counter);
3070         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3071         percpu_counter_destroy(&sbi->s_dirs_counter);
3072         percpu_counter_destroy(&sbi->s_dirtyblocks_counter);
3073 failed_mount2:
3074         for (i = 0; i < db_count; i++)
3075                 brelse(sbi->s_group_desc[i]);
3076         kfree(sbi->s_group_desc);
3077 failed_mount:
3078         if (sbi->s_proc) {
3079                 remove_proc_entry(sb->s_id, ext4_proc_root);
3080         }
3081 #ifdef CONFIG_QUOTA
3082         for (i = 0; i < MAXQUOTAS; i++)
3083                 kfree(sbi->s_qf_names[i]);
3084 #endif
3085         ext4_blkdev_remove(sbi);
3086         brelse(bh);
3087 out_fail:
3088         sb->s_fs_info = NULL;
3089         kfree(sbi->s_blockgroup_lock);
3090         kfree(sbi);
3091         lock_kernel();
3092         return ret;
3093 }
3094
3095 /*
3096  * Setup any per-fs journal parameters now.  We'll do this both on
3097  * initial mount, once the journal has been initialised but before we've
3098  * done any recovery; and again on any subsequent remount.
3099  */
3100 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3101 {
3102         struct ext4_sb_info *sbi = EXT4_SB(sb);
3103
3104         journal->j_commit_interval = sbi->s_commit_interval;
3105         journal->j_min_batch_time = sbi->s_min_batch_time;
3106         journal->j_max_batch_time = sbi->s_max_batch_time;
3107
3108         spin_lock(&journal->j_state_lock);
3109         if (test_opt(sb, BARRIER))
3110                 journal->j_flags |= JBD2_BARRIER;
3111         else
3112                 journal->j_flags &= ~JBD2_BARRIER;
3113         if (test_opt(sb, DATA_ERR_ABORT))
3114                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3115         else
3116                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3117         spin_unlock(&journal->j_state_lock);
3118 }
3119
3120 static journal_t *ext4_get_journal(struct super_block *sb,
3121                                    unsigned int journal_inum)
3122 {
3123         struct inode *journal_inode;
3124         journal_t *journal;
3125
3126         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3127
3128         /* First, test for the existence of a valid inode on disk.  Bad
3129          * things happen if we iget() an unused inode, as the subsequent
3130          * iput() will try to delete it. */
3131
3132         journal_inode = ext4_iget(sb, journal_inum);
3133         if (IS_ERR(journal_inode)) {
3134                 ext4_msg(sb, KERN_ERR, "no journal found");
3135                 return NULL;
3136         }
3137         if (!journal_inode->i_nlink) {
3138                 make_bad_inode(journal_inode);
3139                 iput(journal_inode);
3140                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3141                 return NULL;
3142         }
3143
3144         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3145                   journal_inode, journal_inode->i_size);
3146         if (!S_ISREG(journal_inode->i_mode)) {
3147                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3148                 iput(journal_inode);
3149                 return NULL;
3150         }
3151
3152         journal = jbd2_journal_init_inode(journal_inode);
3153         if (!journal) {
3154                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3155                 iput(journal_inode);
3156                 return NULL;
3157         }
3158         journal->j_private = sb;
3159         ext4_init_journal_params(sb, journal);
3160         return journal;
3161 }
3162
3163 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3164                                        dev_t j_dev)
3165 {
3166         struct buffer_head *bh;
3167         journal_t *journal;
3168         ext4_fsblk_t start;
3169         ext4_fsblk_t len;
3170         int hblock, blocksize;
3171         ext4_fsblk_t sb_block;
3172         unsigned long offset;
3173         struct ext4_super_block *es;
3174         struct block_device *bdev;
3175
3176         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3177
3178         bdev = ext4_blkdev_get(j_dev, sb);
3179         if (bdev == NULL)
3180                 return NULL;
3181
3182         if (bd_claim(bdev, sb)) {
3183                 ext4_msg(sb, KERN_ERR,
3184                         "failed to claim external journal device");
3185                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
3186                 return NULL;
3187         }
3188
3189         blocksize = sb->s_blocksize;
3190         hblock = bdev_logical_block_size(bdev);
3191         if (blocksize < hblock) {
3192                 ext4_msg(sb, KERN_ERR,
3193                         "blocksize too small for journal device");
3194                 goto out_bdev;
3195         }
3196
3197         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3198         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3199         set_blocksize(bdev, blocksize);
3200         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3201                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3202                        "external journal");
3203                 goto out_bdev;
3204         }
3205
3206         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3207         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3208             !(le32_to_cpu(es->s_feature_incompat) &
3209               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3210                 ext4_msg(sb, KERN_ERR, "external journal has "
3211                                         "bad superblock");
3212                 brelse(bh);
3213                 goto out_bdev;
3214         }
3215
3216         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3217                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3218                 brelse(bh);
3219                 goto out_bdev;
3220         }
3221
3222         len = ext4_blocks_count(es);
3223         start = sb_block + 1;
3224         brelse(bh);     /* we're done with the superblock */
3225
3226         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3227                                         start, len, blocksize);
3228         if (!journal) {
3229                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3230                 goto out_bdev;
3231         }
3232         journal->j_private = sb;
3233         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3234         wait_on_buffer(journal->j_sb_buffer);
3235         if (!buffer_uptodate(journal->j_sb_buffer)) {
3236                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3237                 goto out_journal;
3238         }
3239         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3240                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3241                                         "user (unsupported) - %d",
3242                         be32_to_cpu(journal->j_superblock->s_nr_users));
3243                 goto out_journal;
3244         }
3245         EXT4_SB(sb)->journal_bdev = bdev;
3246         ext4_init_journal_params(sb, journal);
3247         return journal;
3248
3249 out_journal:
3250         jbd2_journal_destroy(journal);
3251 out_bdev:
3252         ext4_blkdev_put(bdev);
3253         return NULL;
3254 }
3255
3256 static int ext4_load_journal(struct super_block *sb,
3257                              struct ext4_super_block *es,
3258                              unsigned long journal_devnum)
3259 {
3260         journal_t *journal;
3261         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3262         dev_t journal_dev;
3263         int err = 0;
3264         int really_read_only;
3265
3266         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3267
3268         if (journal_devnum &&
3269             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3270                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3271                         "numbers have changed");
3272                 journal_dev = new_decode_dev(journal_devnum);
3273         } else
3274                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3275
3276         really_read_only = bdev_read_only(sb->s_bdev);
3277
3278         /*
3279          * Are we loading a blank journal or performing recovery after a
3280          * crash?  For recovery, we need to check in advance whether we
3281          * can get read-write access to the device.
3282          */
3283         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3284                 if (sb->s_flags & MS_RDONLY) {
3285                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3286                                         "required on readonly filesystem");
3287                         if (really_read_only) {
3288                                 ext4_msg(sb, KERN_ERR, "write access "
3289                                         "unavailable, cannot proceed");
3290                                 return -EROFS;
3291                         }
3292                         ext4_msg(sb, KERN_INFO, "write access will "
3293                                "be enabled during recovery");
3294                 }
3295         }
3296
3297         if (journal_inum && journal_dev) {
3298                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3299                        "and inode journals!");
3300                 return -EINVAL;
3301         }
3302
3303         if (journal_inum) {
3304                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3305                         return -EINVAL;
3306         } else {
3307                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3308                         return -EINVAL;
3309         }
3310
3311         if (!(journal->j_flags & JBD2_BARRIER))
3312                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3313
3314         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
3315                 err = jbd2_journal_update_format(journal);
3316                 if (err)  {
3317                         ext4_msg(sb, KERN_ERR, "error updating journal");
3318                         jbd2_journal_destroy(journal);
3319                         return err;
3320                 }
3321         }
3322
3323         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3324                 err = jbd2_journal_wipe(journal, !really_read_only);
3325         if (!err)
3326                 err = jbd2_journal_load(journal);
3327
3328         if (err) {
3329                 ext4_msg(sb, KERN_ERR, "error loading journal");
3330                 jbd2_journal_destroy(journal);
3331                 return err;
3332         }
3333
3334         EXT4_SB(sb)->s_journal = journal;
3335         ext4_clear_journal_err(sb, es);
3336
3337         if (journal_devnum &&
3338             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3339                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3340
3341                 /* Make sure we flush the recovery flag to disk. */
3342                 ext4_commit_super(sb, 1);
3343         }
3344
3345         return 0;
3346 }
3347
3348 static int ext4_commit_super(struct super_block *sb, int sync)
3349 {
3350         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3351         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3352         int error = 0;
3353
3354         if (!sbh)
3355                 return error;
3356         if (buffer_write_io_error(sbh)) {
3357                 /*
3358                  * Oh, dear.  A previous attempt to write the
3359                  * superblock failed.  This could happen because the
3360                  * USB device was yanked out.  Or it could happen to
3361                  * be a transient write error and maybe the block will
3362                  * be remapped.  Nothing we can do but to retry the
3363                  * write and hope for the best.
3364                  */
3365                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3366                        "superblock detected");
3367                 clear_buffer_write_io_error(sbh);
3368                 set_buffer_uptodate(sbh);
3369         }
3370         /*
3371          * If the file system is mounted read-only, don't update the
3372          * superblock write time.  This avoids updating the superblock
3373          * write time when we are mounting the root file system
3374          * read/only but we need to replay the journal; at that point,
3375          * for people who are east of GMT and who make their clock
3376          * tick in localtime for Windows bug-for-bug compatibility,
3377          * the clock is set in the future, and this will cause e2fsck
3378          * to complain and force a full file system check.
3379          */
3380         if (!(sb->s_flags & MS_RDONLY))
3381                 es->s_wtime = cpu_to_le32(get_seconds());
3382         es->s_kbytes_written =
3383                 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written + 
3384                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
3385                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
3386         ext4_free_blocks_count_set(es, percpu_counter_sum_positive(
3387                                         &EXT4_SB(sb)->s_freeblocks_counter));
3388         es->s_free_inodes_count = cpu_to_le32(percpu_counter_sum_positive(
3389                                         &EXT4_SB(sb)->s_freeinodes_counter));
3390         sb->s_dirt = 0;
3391         BUFFER_TRACE(sbh, "marking dirty");
3392         mark_buffer_dirty(sbh);
3393         if (sync) {
3394                 error = sync_dirty_buffer(sbh);
3395                 if (error)
3396                         return error;
3397
3398                 error = buffer_write_io_error(sbh);
3399                 if (error) {
3400                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
3401                                "superblock");
3402                         clear_buffer_write_io_error(sbh);
3403                         set_buffer_uptodate(sbh);
3404                 }
3405         }
3406         return error;
3407 }
3408
3409 /*
3410  * Have we just finished recovery?  If so, and if we are mounting (or
3411  * remounting) the filesystem readonly, then we will end up with a
3412  * consistent fs on disk.  Record that fact.
3413  */
3414 static void ext4_mark_recovery_complete(struct super_block *sb,
3415                                         struct ext4_super_block *es)
3416 {
3417         journal_t *journal = EXT4_SB(sb)->s_journal;
3418
3419         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3420                 BUG_ON(journal != NULL);
3421                 return;
3422         }
3423         jbd2_journal_lock_updates(journal);
3424         if (jbd2_journal_flush(journal) < 0)
3425                 goto out;
3426
3427         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
3428             sb->s_flags & MS_RDONLY) {
3429                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3430                 ext4_commit_super(sb, 1);
3431         }
3432
3433 out:
3434         jbd2_journal_unlock_updates(journal);
3435 }
3436
3437 /*
3438  * If we are mounting (or read-write remounting) a filesystem whose journal
3439  * has recorded an error from a previous lifetime, move that error to the
3440  * main filesystem now.
3441  */
3442 static void ext4_clear_journal_err(struct super_block *sb,
3443                                    struct ext4_super_block *es)
3444 {
3445         journal_t *journal;
3446         int j_errno;
3447         const char *errstr;
3448
3449         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3450
3451         journal = EXT4_SB(sb)->s_journal;
3452
3453         /*
3454          * Now check for any error status which may have been recorded in the
3455          * journal by a prior ext4_error() or ext4_abort()
3456          */
3457
3458         j_errno = jbd2_journal_errno(journal);
3459         if (j_errno) {
3460                 char nbuf[16];
3461
3462                 errstr = ext4_decode_error(sb, j_errno, nbuf);
3463                 ext4_warning(sb, "Filesystem error recorded "
3464                              "from previous mount: %s", errstr);
3465                 ext4_warning(sb, "Marking fs in need of filesystem check.");
3466
3467                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
3468                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
3469                 ext4_commit_super(sb, 1);
3470
3471                 jbd2_journal_clear_err(journal);
3472         }
3473 }
3474
3475 /*
3476  * Force the running and committing transactions to commit,
3477  * and wait on the commit.
3478  */
3479 int ext4_force_commit(struct super_block *sb)
3480 {
3481         journal_t *journal;
3482         int ret = 0;
3483
3484         if (sb->s_flags & MS_RDONLY)
3485                 return 0;
3486
3487         journal = EXT4_SB(sb)->s_journal;
3488         if (journal)
3489                 ret = ext4_journal_force_commit(journal);
3490
3491         return ret;
3492 }
3493
3494 static void ext4_write_super(struct super_block *sb)
3495 {
3496         lock_super(sb);
3497         ext4_commit_super(sb, 1);
3498         unlock_super(sb);
3499 }
3500
3501 static int ext4_sync_fs(struct super_block *sb, int wait)
3502 {
3503         int ret = 0;
3504         tid_t target;
3505         struct ext4_sb_info *sbi = EXT4_SB(sb);
3506
3507         trace_ext4_sync_fs(sb, wait);
3508         flush_workqueue(sbi->dio_unwritten_wq);
3509         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
3510                 if (wait)
3511                         jbd2_log_wait_commit(sbi->s_journal, target);
3512         }
3513         return ret;
3514 }
3515
3516 /*
3517  * LVM calls this function before a (read-only) snapshot is created.  This
3518  * gives us a chance to flush the journal completely and mark the fs clean.
3519  */
3520 static int ext4_freeze(struct super_block *sb)
3521 {
3522         int error = 0;
3523         journal_t *journal;
3524
3525         if (sb->s_flags & MS_RDONLY)
3526                 return 0;
3527
3528         journal = EXT4_SB(sb)->s_journal;
3529
3530         /* Now we set up the journal barrier. */
3531         jbd2_journal_lock_updates(journal);
3532
3533         /*
3534          * Don't clear the needs_recovery flag if we failed to flush
3535          * the journal.
3536          */
3537         error = jbd2_journal_flush(journal);
3538         if (error < 0) {
3539         out:
3540                 jbd2_journal_unlock_updates(journal);
3541                 return error;
3542         }
3543
3544         /* Journal blocked and flushed, clear needs_recovery flag. */
3545         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3546         error = ext4_commit_super(sb, 1);
3547         if (error)
3548                 goto out;
3549         return 0;
3550 }
3551
3552 /*
3553  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
3554  * flag here, even though the filesystem is not technically dirty yet.
3555  */
3556 static int ext4_unfreeze(struct super_block *sb)
3557 {
3558         if (sb->s_flags & MS_RDONLY)
3559                 return 0;
3560
3561         lock_super(sb);
3562         /* Reset the needs_recovery flag before the fs is unlocked. */
3563         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
3564         ext4_commit_super(sb, 1);
3565         unlock_super(sb);
3566         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3567         return 0;
3568 }
3569
3570 static int ext4_remount(struct super_block *sb, int *flags, char *data)
3571 {
3572         struct ext4_super_block *es;
3573         struct ext4_sb_info *sbi = EXT4_SB(sb);
3574         ext4_fsblk_t n_blocks_count = 0;
3575         unsigned long old_sb_flags;
3576         struct ext4_mount_options old_opts;
3577         int enable_quota = 0;
3578         ext4_group_t g;
3579         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3580         int err;
3581 #ifdef CONFIG_QUOTA
3582         int i;
3583 #endif
3584
3585         lock_kernel();
3586
3587         /* Store the original options */
3588         lock_super(sb);
3589         old_sb_flags = sb->s_flags;
3590         old_opts.s_mount_opt = sbi->s_mount_opt;
3591         old_opts.s_resuid = sbi->s_resuid;
3592         old_opts.s_resgid = sbi->s_resgid;
3593         old_opts.s_commit_interval = sbi->s_commit_interval;
3594         old_opts.s_min_batch_time = sbi->s_min_batch_time;
3595         old_opts.s_max_batch_time = sbi->s_max_batch_time;
3596 #ifdef CONFIG_QUOTA
3597         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
3598         for (i = 0; i < MAXQUOTAS; i++)
3599                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
3600 #endif
3601         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
3602                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
3603
3604         /*
3605          * Allow the "check" option to be passed as a remount option.
3606          */
3607         if (!parse_options(data, sb, NULL, &journal_ioprio,
3608                            &n_blocks_count, 1)) {
3609                 err = -EINVAL;
3610                 goto restore_opts;
3611         }
3612
3613         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
3614                 ext4_abort(sb, __func__, "Abort forced by user");
3615
3616         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3617                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3618
3619         es = sbi->s_es;
3620
3621         if (sbi->s_journal) {
3622                 ext4_init_journal_params(sb, sbi->s_journal);
3623                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3624         }
3625
3626         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
3627                 n_blocks_count > ext4_blocks_count(es)) {
3628                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
3629                         err = -EROFS;
3630                         goto restore_opts;
3631                 }
3632
3633                 if (*flags & MS_RDONLY) {
3634                         err = dquot_suspend(sb, -1);
3635                         if (err < 0)
3636                                 goto restore_opts;
3637
3638                         /*
3639                          * First of all, the unconditional stuff we have to do
3640                          * to disable replay of the journal when we next remount
3641                          */
3642                         sb->s_flags |= MS_RDONLY;
3643
3644                         /*
3645                          * OK, test if we are remounting a valid rw partition
3646                          * readonly, and if so set the rdonly flag and then
3647                          * mark the partition as valid again.
3648                          */
3649                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
3650                             (sbi->s_mount_state & EXT4_VALID_FS))
3651                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
3652
3653                         if (sbi->s_journal)
3654                                 ext4_mark_recovery_complete(sb, es);
3655                 } else {
3656                         /* Make sure we can mount this feature set readwrite */
3657                         if (!ext4_feature_set_ok(sb, 0)) {
3658                                 err = -EROFS;
3659                                 goto restore_opts;
3660                         }
3661                         /*
3662                          * Make sure the group descriptor checksums
3663                          * are sane.  If they aren't, refuse to remount r/w.
3664                          */
3665                         for (g = 0; g < sbi->s_groups_count; g++) {
3666                                 struct ext4_group_desc *gdp =
3667                                         ext4_get_group_desc(sb, g, NULL);
3668
3669                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
3670                                         ext4_msg(sb, KERN_ERR,
3671                "ext4_remount: Checksum for group %u failed (%u!=%u)",
3672                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
3673                                                le16_to_cpu(gdp->bg_checksum));
3674                                         err = -EINVAL;
3675                                         goto restore_opts;
3676                                 }
3677                         }
3678
3679                         /*
3680                          * If we have an unprocessed orphan list hanging
3681                          * around from a previously readonly bdev mount,
3682                          * require a full umount/remount for now.
3683                          */
3684                         if (es->s_last_orphan) {
3685                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
3686                                        "remount RDWR because of unprocessed "
3687                                        "orphan inode list.  Please "
3688                                        "umount/remount instead");
3689                                 err = -EINVAL;
3690                                 goto restore_opts;
3691                         }
3692
3693                         /*
3694                          * Mounting a RDONLY partition read-write, so reread
3695                          * and store the current valid flag.  (It may have
3696                          * been changed by e2fsck since we originally mounted
3697                          * the partition.)
3698                          */
3699                         if (sbi->s_journal)
3700                                 ext4_clear_journal_err(sb, es);
3701                         sbi->s_mount_state = le16_to_cpu(es->s_state);
3702                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
3703                                 goto restore_opts;
3704                         if (!ext4_setup_super(sb, es, 0))
3705                                 sb->s_flags &= ~MS_RDONLY;
3706                         enable_quota = 1;
3707                 }
3708         }
3709         ext4_setup_system_zone(sb);
3710         if (sbi->s_journal == NULL)
3711                 ext4_commit_super(sb, 1);
3712
3713 #ifdef CONFIG_QUOTA
3714         /* Release old quota file names */
3715         for (i = 0; i < MAXQUOTAS; i++)
3716                 if (old_opts.s_qf_names[i] &&
3717                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3718                         kfree(old_opts.s_qf_names[i]);
3719 #endif
3720         unlock_super(sb);
3721         unlock_kernel();
3722         if (enable_quota)
3723                 dquot_resume(sb, -1);
3724         return 0;
3725
3726 restore_opts:
3727         sb->s_flags = old_sb_flags;
3728         sbi->s_mount_opt = old_opts.s_mount_opt;
3729         sbi->s_resuid = old_opts.s_resuid;
3730         sbi->s_resgid = old_opts.s_resgid;
3731         sbi->s_commit_interval = old_opts.s_commit_interval;
3732         sbi->s_min_batch_time = old_opts.s_min_batch_time;
3733         sbi->s_max_batch_time = old_opts.s_max_batch_time;
3734 #ifdef CONFIG_QUOTA
3735         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
3736         for (i = 0; i < MAXQUOTAS; i++) {
3737                 if (sbi->s_qf_names[i] &&
3738                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
3739                         kfree(sbi->s_qf_names[i]);
3740                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
3741         }
3742 #endif
3743         unlock_super(sb);
3744         unlock_kernel();
3745         return err;
3746 }
3747
3748 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
3749 {
3750         struct super_block *sb = dentry->d_sb;
3751         struct ext4_sb_info *sbi = EXT4_SB(sb);
3752         struct ext4_super_block *es = sbi->s_es;
3753         u64 fsid;
3754
3755         if (test_opt(sb, MINIX_DF)) {
3756                 sbi->s_overhead_last = 0;
3757         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
3758                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3759                 ext4_fsblk_t overhead = 0;
3760
3761                 /*
3762                  * Compute the overhead (FS structures).  This is constant
3763                  * for a given filesystem unless the number of block groups
3764                  * changes so we cache the previous value until it does.
3765                  */
3766
3767                 /*
3768                  * All of the blocks before first_data_block are
3769                  * overhead
3770                  */
3771                 overhead = le32_to_cpu(es->s_first_data_block);
3772
3773                 /*
3774                  * Add the overhead attributed to the superblock and
3775                  * block group descriptors.  If the sparse superblocks
3776                  * feature is turned on, then not all groups have this.
3777                  */
3778                 for (i = 0; i < ngroups; i++) {
3779                         overhead += ext4_bg_has_super(sb, i) +
3780                                 ext4_bg_num_gdb(sb, i);
3781                         cond_resched();
3782                 }
3783
3784                 /*
3785                  * Every block group has an inode bitmap, a block
3786                  * bitmap, and an inode table.
3787                  */
3788                 overhead += ngroups * (2 + sbi->s_itb_per_group);
3789                 sbi->s_overhead_last = overhead;
3790                 smp_wmb();
3791                 sbi->s_blocks_last = ext4_blocks_count(es);
3792         }
3793
3794         buf->f_type = EXT4_SUPER_MAGIC;
3795         buf->f_bsize = sb->s_blocksize;
3796         buf->f_blocks = ext4_blocks_count(es) - sbi->s_overhead_last;
3797         buf->f_bfree = percpu_counter_sum_positive(&sbi->s_freeblocks_counter) -
3798                        percpu_counter_sum_positive(&sbi->s_dirtyblocks_counter);
3799         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
3800         if (buf->f_bfree < ext4_r_blocks_count(es))
3801                 buf->f_bavail = 0;
3802         buf->f_files = le32_to_cpu(es->s_inodes_count);
3803         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
3804         buf->f_namelen = EXT4_NAME_LEN;
3805         fsid = le64_to_cpup((void *)es->s_uuid) ^
3806                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
3807         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
3808         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
3809
3810         return 0;
3811 }
3812
3813 /* Helper function for writing quotas on sync - we need to start transaction
3814  * before quota file is locked for write. Otherwise the are possible deadlocks:
3815  * Process 1                         Process 2
3816  * ext4_create()                     quota_sync()
3817  *   jbd2_journal_start()                  write_dquot()
3818  *   dquot_initialize()                         down(dqio_mutex)
3819  *     down(dqio_mutex)                    jbd2_journal_start()
3820  *
3821  */
3822
3823 #ifdef CONFIG_QUOTA
3824
3825 static inline struct inode *dquot_to_inode(struct dquot *dquot)
3826 {
3827         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
3828 }
3829
3830 static int ext4_write_dquot(struct dquot *dquot)
3831 {
3832         int ret, err;
3833         handle_t *handle;
3834         struct inode *inode;
3835
3836         inode = dquot_to_inode(dquot);
3837         handle = ext4_journal_start(inode,
3838                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
3839         if (IS_ERR(handle))
3840                 return PTR_ERR(handle);
3841         ret = dquot_commit(dquot);
3842         err = ext4_journal_stop(handle);
3843         if (!ret)
3844                 ret = err;
3845         return ret;
3846 }
3847
3848 static int ext4_acquire_dquot(struct dquot *dquot)
3849 {
3850         int ret, err;
3851         handle_t *handle;
3852
3853         handle = ext4_journal_start(dquot_to_inode(dquot),
3854                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
3855         if (IS_ERR(handle))
3856                 return PTR_ERR(handle);
3857         ret = dquot_acquire(dquot);
3858         err = ext4_journal_stop(handle);
3859         if (!ret)
3860                 ret = err;
3861         return ret;
3862 }
3863
3864 static int ext4_release_dquot(struct dquot *dquot)
3865 {
3866         int ret, err;
3867         handle_t *handle;
3868
3869         handle = ext4_journal_start(dquot_to_inode(dquot),
3870                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
3871         if (IS_ERR(handle)) {
3872                 /* Release dquot anyway to avoid endless cycle in dqput() */
3873                 dquot_release(dquot);
3874                 return PTR_ERR(handle);
3875         }
3876         ret = dquot_release(dquot);
3877         err = ext4_journal_stop(handle);
3878         if (!ret)
3879                 ret = err;
3880         return ret;
3881 }
3882
3883 static int ext4_mark_dquot_dirty(struct dquot *dquot)
3884 {
3885         /* Are we journaling quotas? */
3886         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
3887             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
3888                 dquot_mark_dquot_dirty(dquot);
3889                 return ext4_write_dquot(dquot);
3890         } else {
3891                 return dquot_mark_dquot_dirty(dquot);
3892         }
3893 }
3894
3895 static int ext4_write_info(struct super_block *sb, int type)
3896 {
3897         int ret, err;
3898         handle_t *handle;
3899
3900         /* Data block + inode block */
3901         handle = ext4_journal_start(sb->s_root->d_inode, 2);
3902         if (IS_ERR(handle))
3903                 return PTR_ERR(handle);
3904         ret = dquot_commit_info(sb, type);
3905         err = ext4_journal_stop(handle);
3906         if (!ret)
3907                 ret = err;
3908         return ret;
3909 }
3910
3911 /*
3912  * Turn on quotas during mount time - we need to find
3913  * the quota file and such...
3914  */
3915 static int ext4_quota_on_mount(struct super_block *sb, int type)
3916 {
3917         return vfs_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
3918                                   EXT4_SB(sb)->s_jquota_fmt, type);
3919 }
3920
3921 /*
3922  * Standard function to be called on quota_on
3923  */
3924 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
3925                          char *name, int remount)
3926 {
3927         int err;
3928         struct path path;
3929
3930         if (!test_opt(sb, QUOTA))
3931                 return -EINVAL;
3932         /* When remounting, no checks are needed and in fact, name is NULL */
3933         if (remount)
3934                 return vfs_quota_on(sb, type, format_id, name, remount);
3935
3936         err = kern_path(name, LOOKUP_FOLLOW, &path);
3937         if (err)
3938                 return err;
3939
3940         /* Quotafile not on the same filesystem? */
3941         if (path.mnt->mnt_sb != sb) {
3942                 path_put(&path);
3943                 return -EXDEV;
3944         }
3945         /* Journaling quota? */
3946         if (EXT4_SB(sb)->s_qf_names[type]) {
3947                 /* Quotafile not in fs root? */
3948                 if (path.dentry->d_parent != sb->s_root)
3949                         ext4_msg(sb, KERN_WARNING,
3950                                 "Quota file not on filesystem root. "
3951                                 "Journaled quota will not work");
3952         }
3953
3954         /*
3955          * When we journal data on quota file, we have to flush journal to see
3956          * all updates to the file when we bypass pagecache...
3957          */
3958         if (EXT4_SB(sb)->s_journal &&
3959             ext4_should_journal_data(path.dentry->d_inode)) {
3960                 /*
3961                  * We don't need to lock updates but journal_flush() could
3962                  * otherwise be livelocked...
3963                  */
3964                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
3965                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
3966                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
3967                 if (err) {
3968                         path_put(&path);
3969                         return err;
3970                 }
3971         }
3972
3973         err = vfs_quota_on_path(sb, type, format_id, &path);
3974         path_put(&path);
3975         return err;
3976 }
3977
3978 /* Read data from quotafile - avoid pagecache and such because we cannot afford
3979  * acquiring the locks... As quota files are never truncated and quota code
3980  * itself serializes the operations (and noone else should touch the files)
3981  * we don't have to be afraid of races */
3982 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
3983                                size_t len, loff_t off)
3984 {
3985         struct inode *inode = sb_dqopt(sb)->files[type];
3986         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
3987         int err = 0;
3988         int offset = off & (sb->s_blocksize - 1);
3989         int tocopy;
3990         size_t toread;
3991         struct buffer_head *bh;
3992         loff_t i_size = i_size_read(inode);
3993
3994         if (off > i_size)
3995                 return 0;
3996         if (off+len > i_size)
3997                 len = i_size-off;
3998         toread = len;
3999         while (toread > 0) {
4000                 tocopy = sb->s_blocksize - offset < toread ?
4001                                 sb->s_blocksize - offset : toread;
4002                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4003                 if (err)
4004                         return err;
4005                 if (!bh)        /* A hole? */
4006                         memset(data, 0, tocopy);
4007                 else
4008                         memcpy(data, bh->b_data+offset, tocopy);
4009                 brelse(bh);
4010                 offset = 0;
4011                 toread -= tocopy;
4012                 data += tocopy;
4013                 blk++;
4014         }
4015         return len;
4016 }
4017
4018 /* Write to quotafile (we know the transaction is already started and has
4019  * enough credits) */
4020 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4021                                 const char *data, size_t len, loff_t off)
4022 {
4023         struct inode *inode = sb_dqopt(sb)->files[type];
4024         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4025         int err = 0;
4026         int offset = off & (sb->s_blocksize - 1);
4027         int journal_quota = EXT4_SB(sb)->s_qf_names[type] != NULL;
4028         struct buffer_head *bh;
4029         handle_t *handle = journal_current_handle();
4030
4031         if (EXT4_SB(sb)->s_journal && !handle) {
4032                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4033                         " cancelled because transaction is not started",
4034                         (unsigned long long)off, (unsigned long long)len);
4035                 return -EIO;
4036         }
4037         /*
4038          * Since we account only one data block in transaction credits,
4039          * then it is impossible to cross a block boundary.
4040          */
4041         if (sb->s_blocksize - offset < len) {
4042                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4043                         " cancelled because not block aligned",
4044                         (unsigned long long)off, (unsigned long long)len);
4045                 return -EIO;
4046         }
4047
4048         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4049         bh = ext4_bread(handle, inode, blk, 1, &err);
4050         if (!bh)
4051                 goto out;
4052         if (journal_quota) {
4053                 err = ext4_journal_get_write_access(handle, bh);
4054                 if (err) {
4055                         brelse(bh);
4056                         goto out;
4057                 }
4058         }
4059         lock_buffer(bh);
4060         memcpy(bh->b_data+offset, data, len);
4061         flush_dcache_page(bh->b_page);
4062         unlock_buffer(bh);
4063         if (journal_quota)
4064                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
4065         else {
4066                 /* Always do at least ordered writes for quotas */
4067                 err = ext4_jbd2_file_inode(handle, inode);
4068                 mark_buffer_dirty(bh);
4069         }
4070         brelse(bh);
4071 out:
4072         if (err) {
4073                 mutex_unlock(&inode->i_mutex);
4074                 return err;
4075         }
4076         if (inode->i_size < off + len) {
4077                 i_size_write(inode, off + len);
4078                 EXT4_I(inode)->i_disksize = inode->i_size;
4079         }
4080         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4081         ext4_mark_inode_dirty(handle, inode);
4082         mutex_unlock(&inode->i_mutex);
4083         return len;
4084 }
4085
4086 #endif
4087
4088 static int ext4_get_sb(struct file_system_type *fs_type, int flags,
4089                        const char *dev_name, void *data, struct vfsmount *mnt)
4090 {
4091         return get_sb_bdev(fs_type, flags, dev_name, data, ext4_fill_super,mnt);
4092 }
4093
4094 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4095 static struct file_system_type ext2_fs_type = {
4096         .owner          = THIS_MODULE,
4097         .name           = "ext2",
4098         .get_sb         = ext4_get_sb,
4099         .kill_sb        = kill_block_super,
4100         .fs_flags       = FS_REQUIRES_DEV,
4101 };
4102
4103 static inline void register_as_ext2(void)
4104 {
4105         int err = register_filesystem(&ext2_fs_type);
4106         if (err)
4107                 printk(KERN_WARNING
4108                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4109 }
4110
4111 static inline void unregister_as_ext2(void)
4112 {
4113         unregister_filesystem(&ext2_fs_type);
4114 }
4115 MODULE_ALIAS("ext2");
4116 #else
4117 static inline void register_as_ext2(void) { }
4118 static inline void unregister_as_ext2(void) { }
4119 #endif
4120
4121 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4122 static inline void register_as_ext3(void)
4123 {
4124         int err = register_filesystem(&ext3_fs_type);
4125         if (err)
4126                 printk(KERN_WARNING
4127                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4128 }
4129
4130 static inline void unregister_as_ext3(void)
4131 {
4132         unregister_filesystem(&ext3_fs_type);
4133 }
4134 MODULE_ALIAS("ext3");
4135 #else
4136 static inline void register_as_ext3(void) { }
4137 static inline void unregister_as_ext3(void) { }
4138 #endif
4139
4140 static struct file_system_type ext4_fs_type = {
4141         .owner          = THIS_MODULE,
4142         .name           = "ext4",
4143         .get_sb         = ext4_get_sb,
4144         .kill_sb        = kill_block_super,
4145         .fs_flags       = FS_REQUIRES_DEV,
4146 };
4147
4148 static int __init init_ext4_fs(void)
4149 {
4150         int err;
4151
4152         err = init_ext4_system_zone();
4153         if (err)
4154                 return err;
4155         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4156         if (!ext4_kset)
4157                 goto out4;
4158         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4159         err = init_ext4_mballoc();
4160         if (err)
4161                 goto out3;
4162
4163         err = init_ext4_xattr();
4164         if (err)
4165                 goto out2;
4166         err = init_inodecache();
4167         if (err)
4168                 goto out1;
4169         register_as_ext2();
4170         register_as_ext3();
4171         err = register_filesystem(&ext4_fs_type);
4172         if (err)
4173                 goto out;
4174         return 0;
4175 out:
4176         unregister_as_ext2();
4177         unregister_as_ext3();
4178         destroy_inodecache();
4179 out1:
4180         exit_ext4_xattr();
4181 out2:
4182         exit_ext4_mballoc();
4183 out3:
4184         remove_proc_entry("fs/ext4", NULL);
4185         kset_unregister(ext4_kset);
4186 out4:
4187         exit_ext4_system_zone();
4188         return err;
4189 }
4190
4191 static void __exit exit_ext4_fs(void)
4192 {
4193         unregister_as_ext2();
4194         unregister_as_ext3();
4195         unregister_filesystem(&ext4_fs_type);
4196         destroy_inodecache();
4197         exit_ext4_xattr();
4198         exit_ext4_mballoc();
4199         remove_proc_entry("fs/ext4", NULL);
4200         kset_unregister(ext4_kset);
4201         exit_ext4_system_zone();
4202 }
4203
4204 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4205 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4206 MODULE_LICENSE("GPL");
4207 module_init(init_ext4_fs)
4208 module_exit(exit_ext4_fs)