Merge remote-tracking branch 'origin/dev/sumit-linux-3.10.96' into TOT-merge
[linux-3.10.git] / fs / ext4 / super.c
1 /*
2  *  linux/fs/ext4/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/module.h>
20 #include <linux/string.h>
21 #include <linux/fs.h>
22 #include <linux/time.h>
23 #include <linux/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/buffer_head.h>
30 #include <linux/exportfs.h>
31 #include <linux/vfs.h>
32 #include <linux/random.h>
33 #include <linux/mount.h>
34 #include <linux/namei.h>
35 #include <linux/quotaops.h>
36 #include <linux/seq_file.h>
37 #include <linux/proc_fs.h>
38 #include <linux/ctype.h>
39 #include <linux/log2.h>
40 #include <linux/crc16.h>
41 #include <linux/cleancache.h>
42 #include <asm/uaccess.h>
43
44 #include <linux/kthread.h>
45 #include <linux/freezer.h>
46
47 #include "ext4.h"
48 #include "ext4_extents.h"       /* Needed for trace points definition */
49 #include "ext4_jbd2.h"
50 #include "xattr.h"
51 #include "acl.h"
52 #include "mballoc.h"
53
54 #define CREATE_TRACE_POINTS
55 #include <trace/events/ext4.h>
56
57 static struct proc_dir_entry *ext4_proc_root;
58 static struct kset *ext4_kset;
59 static struct ext4_lazy_init *ext4_li_info;
60 static struct mutex ext4_li_mtx;
61 static struct ext4_features *ext4_feat;
62
63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
64                              unsigned long journal_devnum);
65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
66 static int ext4_commit_super(struct super_block *sb, int sync);
67 static void ext4_mark_recovery_complete(struct super_block *sb,
68                                         struct ext4_super_block *es);
69 static void ext4_clear_journal_err(struct super_block *sb,
70                                    struct ext4_super_block *es);
71 static int ext4_sync_fs(struct super_block *sb, int wait);
72 static int ext4_remount(struct super_block *sb, int *flags, char *data);
73 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
74 static int ext4_unfreeze(struct super_block *sb);
75 static int ext4_freeze(struct super_block *sb);
76 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
77                        const char *dev_name, void *data);
78 static inline int ext2_feature_set_ok(struct super_block *sb);
79 static inline int ext3_feature_set_ok(struct super_block *sb);
80 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
81 static void ext4_destroy_lazyinit_thread(void);
82 static void ext4_unregister_li_request(struct super_block *sb);
83 static void ext4_clear_request_list(void);
84 static int ext4_reserve_clusters(struct ext4_sb_info *, ext4_fsblk_t);
85
86 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
87 static struct file_system_type ext2_fs_type = {
88         .owner          = THIS_MODULE,
89         .name           = "ext2",
90         .mount          = ext4_mount,
91         .kill_sb        = kill_block_super,
92         .fs_flags       = FS_REQUIRES_DEV,
93 };
94 MODULE_ALIAS_FS("ext2");
95 MODULE_ALIAS("ext2");
96 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
97 #else
98 #define IS_EXT2_SB(sb) (0)
99 #endif
100
101
102 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
103 static struct file_system_type ext3_fs_type = {
104         .owner          = THIS_MODULE,
105         .name           = "ext3",
106         .mount          = ext4_mount,
107         .kill_sb        = kill_block_super,
108         .fs_flags       = FS_REQUIRES_DEV,
109 };
110 MODULE_ALIAS_FS("ext3");
111 MODULE_ALIAS("ext3");
112 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
113 #else
114 #define IS_EXT3_SB(sb) (0)
115 #endif
116
117 static int ext4_verify_csum_type(struct super_block *sb,
118                                  struct ext4_super_block *es)
119 {
120         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
121                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
122                 return 1;
123
124         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
125 }
126
127 static __le32 ext4_superblock_csum(struct super_block *sb,
128                                    struct ext4_super_block *es)
129 {
130         struct ext4_sb_info *sbi = EXT4_SB(sb);
131         int offset = offsetof(struct ext4_super_block, s_checksum);
132         __u32 csum;
133
134         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
135
136         return cpu_to_le32(csum);
137 }
138
139 int ext4_superblock_csum_verify(struct super_block *sb,
140                                 struct ext4_super_block *es)
141 {
142         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
143                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
144                 return 1;
145
146         return es->s_checksum == ext4_superblock_csum(sb, es);
147 }
148
149 void ext4_superblock_csum_set(struct super_block *sb)
150 {
151         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
152
153         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
154                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
155                 return;
156
157         es->s_checksum = ext4_superblock_csum(sb, es);
158 }
159
160 void *ext4_kvmalloc(size_t size, gfp_t flags)
161 {
162         void *ret;
163
164         ret = kmalloc(size, flags);
165         if (!ret)
166                 ret = __vmalloc(size, flags, PAGE_KERNEL);
167         return ret;
168 }
169
170 void *ext4_kvzalloc(size_t size, gfp_t flags)
171 {
172         void *ret;
173
174         ret = kzalloc(size, flags);
175         if (!ret)
176                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
177         return ret;
178 }
179
180 void ext4_kvfree(void *ptr)
181 {
182         if (is_vmalloc_addr(ptr))
183                 vfree(ptr);
184         else
185                 kfree(ptr);
186
187 }
188
189 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
190                                struct ext4_group_desc *bg)
191 {
192         return le32_to_cpu(bg->bg_block_bitmap_lo) |
193                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
194                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
195 }
196
197 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
198                                struct ext4_group_desc *bg)
199 {
200         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
201                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
202                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
203 }
204
205 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
206                               struct ext4_group_desc *bg)
207 {
208         return le32_to_cpu(bg->bg_inode_table_lo) |
209                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
210                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
211 }
212
213 __u32 ext4_free_group_clusters(struct super_block *sb,
214                                struct ext4_group_desc *bg)
215 {
216         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
217                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
218                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
219 }
220
221 __u32 ext4_free_inodes_count(struct super_block *sb,
222                               struct ext4_group_desc *bg)
223 {
224         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
225                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
226                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
227 }
228
229 __u32 ext4_used_dirs_count(struct super_block *sb,
230                               struct ext4_group_desc *bg)
231 {
232         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
233                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
234                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
235 }
236
237 __u32 ext4_itable_unused_count(struct super_block *sb,
238                               struct ext4_group_desc *bg)
239 {
240         return le16_to_cpu(bg->bg_itable_unused_lo) |
241                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
242                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
243 }
244
245 void ext4_block_bitmap_set(struct super_block *sb,
246                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
247 {
248         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
249         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
250                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
251 }
252
253 void ext4_inode_bitmap_set(struct super_block *sb,
254                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
255 {
256         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
257         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
258                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
259 }
260
261 void ext4_inode_table_set(struct super_block *sb,
262                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
263 {
264         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
265         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
266                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
267 }
268
269 void ext4_free_group_clusters_set(struct super_block *sb,
270                                   struct ext4_group_desc *bg, __u32 count)
271 {
272         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
273         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
274                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
275 }
276
277 void ext4_free_inodes_set(struct super_block *sb,
278                           struct ext4_group_desc *bg, __u32 count)
279 {
280         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
281         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
282                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
283 }
284
285 void ext4_used_dirs_set(struct super_block *sb,
286                           struct ext4_group_desc *bg, __u32 count)
287 {
288         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
289         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
290                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
291 }
292
293 void ext4_itable_unused_set(struct super_block *sb,
294                           struct ext4_group_desc *bg, __u32 count)
295 {
296         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
297         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
298                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
299 }
300
301
302 static void __save_error_info(struct super_block *sb, const char *func,
303                             unsigned int line)
304 {
305         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
306
307         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
308         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
309         es->s_last_error_time = cpu_to_le32(get_seconds());
310         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
311         es->s_last_error_line = cpu_to_le32(line);
312         if (!es->s_first_error_time) {
313                 es->s_first_error_time = es->s_last_error_time;
314                 strncpy(es->s_first_error_func, func,
315                         sizeof(es->s_first_error_func));
316                 es->s_first_error_line = cpu_to_le32(line);
317                 es->s_first_error_ino = es->s_last_error_ino;
318                 es->s_first_error_block = es->s_last_error_block;
319         }
320         /*
321          * Start the daily error reporting function if it hasn't been
322          * started already
323          */
324         if (!es->s_error_count)
325                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
326         le32_add_cpu(&es->s_error_count, 1);
327 }
328
329 static void save_error_info(struct super_block *sb, const char *func,
330                             unsigned int line)
331 {
332         __save_error_info(sb, func, line);
333         ext4_commit_super(sb, 1);
334 }
335
336 /*
337  * The del_gendisk() function uninitializes the disk-specific data
338  * structures, including the bdi structure, without telling anyone
339  * else.  Once this happens, any attempt to call mark_buffer_dirty()
340  * (for example, by ext4_commit_super), will cause a kernel OOPS.
341  * This is a kludge to prevent these oops until we can put in a proper
342  * hook in del_gendisk() to inform the VFS and file system layers.
343  */
344 static int block_device_ejected(struct super_block *sb)
345 {
346         struct inode *bd_inode = sb->s_bdev->bd_inode;
347         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
348
349         return bdi->dev == NULL;
350 }
351
352 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
353 {
354         struct super_block              *sb = journal->j_private;
355         struct ext4_sb_info             *sbi = EXT4_SB(sb);
356         int                             error = is_journal_aborted(journal);
357         struct ext4_journal_cb_entry    *jce;
358
359         BUG_ON(txn->t_state == T_FINISHED);
360         spin_lock(&sbi->s_md_lock);
361         while (!list_empty(&txn->t_private_list)) {
362                 jce = list_entry(txn->t_private_list.next,
363                                  struct ext4_journal_cb_entry, jce_list);
364                 list_del_init(&jce->jce_list);
365                 spin_unlock(&sbi->s_md_lock);
366                 jce->jce_func(sb, jce, error);
367                 spin_lock(&sbi->s_md_lock);
368         }
369         spin_unlock(&sbi->s_md_lock);
370 }
371
372 /* Deal with the reporting of failure conditions on a filesystem such as
373  * inconsistencies detected or read IO failures.
374  *
375  * On ext2, we can store the error state of the filesystem in the
376  * superblock.  That is not possible on ext4, because we may have other
377  * write ordering constraints on the superblock which prevent us from
378  * writing it out straight away; and given that the journal is about to
379  * be aborted, we can't rely on the current, or future, transactions to
380  * write out the superblock safely.
381  *
382  * We'll just use the jbd2_journal_abort() error code to record an error in
383  * the journal instead.  On recovery, the journal will complain about
384  * that error until we've noted it down and cleared it.
385  */
386
387 static void ext4_handle_error(struct super_block *sb)
388 {
389         if (sb->s_flags & MS_RDONLY)
390                 return;
391
392         if (!test_opt(sb, ERRORS_CONT)) {
393                 journal_t *journal = EXT4_SB(sb)->s_journal;
394
395                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
396                 if (journal)
397                         jbd2_journal_abort(journal, -EIO);
398         }
399         if (test_opt(sb, ERRORS_RO)) {
400                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
401                 sb->s_flags |= MS_RDONLY;
402         }
403         if (test_opt(sb, ERRORS_PANIC)) {
404                 if (EXT4_SB(sb)->s_journal &&
405                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
406                         return;
407                 panic("EXT4-fs (device %s): panic forced after error\n",
408                         sb->s_id);
409         }
410 }
411
412 void __ext4_error(struct super_block *sb, const char *function,
413                   unsigned int line, const char *fmt, ...)
414 {
415         struct va_format vaf;
416         va_list args;
417
418         va_start(args, fmt);
419         vaf.fmt = fmt;
420         vaf.va = &args;
421         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
422                sb->s_id, function, line, current->comm, &vaf);
423         va_end(args);
424         save_error_info(sb, function, line);
425
426         ext4_handle_error(sb);
427 }
428
429 void ext4_error_inode(struct inode *inode, const char *function,
430                       unsigned int line, ext4_fsblk_t block,
431                       const char *fmt, ...)
432 {
433         va_list args;
434         struct va_format vaf;
435         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
436
437         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
438         es->s_last_error_block = cpu_to_le64(block);
439         save_error_info(inode->i_sb, function, line);
440         va_start(args, fmt);
441         vaf.fmt = fmt;
442         vaf.va = &args;
443         if (block)
444                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
445                        "inode #%lu: block %llu: comm %s: %pV\n",
446                        inode->i_sb->s_id, function, line, inode->i_ino,
447                        block, current->comm, &vaf);
448         else
449                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
450                        "inode #%lu: comm %s: %pV\n",
451                        inode->i_sb->s_id, function, line, inode->i_ino,
452                        current->comm, &vaf);
453         va_end(args);
454
455         ext4_handle_error(inode->i_sb);
456 }
457
458 void ext4_error_file(struct file *file, const char *function,
459                      unsigned int line, ext4_fsblk_t block,
460                      const char *fmt, ...)
461 {
462         va_list args;
463         struct va_format vaf;
464         struct ext4_super_block *es;
465         struct inode *inode = file_inode(file);
466         char pathname[80], *path;
467
468         es = EXT4_SB(inode->i_sb)->s_es;
469         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
470         save_error_info(inode->i_sb, function, line);
471         path = d_path(&(file->f_path), pathname, sizeof(pathname));
472         if (IS_ERR(path))
473                 path = "(unknown)";
474         va_start(args, fmt);
475         vaf.fmt = fmt;
476         vaf.va = &args;
477         if (block)
478                 printk(KERN_CRIT
479                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
480                        "block %llu: comm %s: path %s: %pV\n",
481                        inode->i_sb->s_id, function, line, inode->i_ino,
482                        block, current->comm, path, &vaf);
483         else
484                 printk(KERN_CRIT
485                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
486                        "comm %s: path %s: %pV\n",
487                        inode->i_sb->s_id, function, line, inode->i_ino,
488                        current->comm, path, &vaf);
489         va_end(args);
490
491         ext4_handle_error(inode->i_sb);
492 }
493
494 const char *ext4_decode_error(struct super_block *sb, int errno,
495                               char nbuf[16])
496 {
497         char *errstr = NULL;
498
499         switch (errno) {
500         case -EIO:
501                 errstr = "IO failure";
502                 break;
503         case -ENOMEM:
504                 errstr = "Out of memory";
505                 break;
506         case -EROFS:
507                 if (!sb || (EXT4_SB(sb)->s_journal &&
508                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
509                         errstr = "Journal has aborted";
510                 else
511                         errstr = "Readonly filesystem";
512                 break;
513         default:
514                 /* If the caller passed in an extra buffer for unknown
515                  * errors, textualise them now.  Else we just return
516                  * NULL. */
517                 if (nbuf) {
518                         /* Check for truncated error codes... */
519                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
520                                 errstr = nbuf;
521                 }
522                 break;
523         }
524
525         return errstr;
526 }
527
528 /* __ext4_std_error decodes expected errors from journaling functions
529  * automatically and invokes the appropriate error response.  */
530
531 void __ext4_std_error(struct super_block *sb, const char *function,
532                       unsigned int line, int errno)
533 {
534         char nbuf[16];
535         const char *errstr;
536
537         /* Special case: if the error is EROFS, and we're not already
538          * inside a transaction, then there's really no point in logging
539          * an error. */
540         if (errno == -EROFS && journal_current_handle() == NULL &&
541             (sb->s_flags & MS_RDONLY))
542                 return;
543
544         errstr = ext4_decode_error(sb, errno, nbuf);
545         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
546                sb->s_id, function, line, errstr);
547         save_error_info(sb, function, line);
548
549         ext4_handle_error(sb);
550 }
551
552 /*
553  * ext4_abort is a much stronger failure handler than ext4_error.  The
554  * abort function may be used to deal with unrecoverable failures such
555  * as journal IO errors or ENOMEM at a critical moment in log management.
556  *
557  * We unconditionally force the filesystem into an ABORT|READONLY state,
558  * unless the error response on the fs has been set to panic in which
559  * case we take the easy way out and panic immediately.
560  */
561
562 void __ext4_abort(struct super_block *sb, const char *function,
563                 unsigned int line, const char *fmt, ...)
564 {
565         va_list args;
566
567         save_error_info(sb, function, line);
568         va_start(args, fmt);
569         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
570                function, line);
571         vprintk(fmt, args);
572         printk("\n");
573         va_end(args);
574
575         if ((sb->s_flags & MS_RDONLY) == 0) {
576                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
577                 sb->s_flags |= MS_RDONLY;
578                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
579                 if (EXT4_SB(sb)->s_journal)
580                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
581                 save_error_info(sb, function, line);
582         }
583         if (test_opt(sb, ERRORS_PANIC)) {
584                 if (EXT4_SB(sb)->s_journal &&
585                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
586                         return;
587                 panic("EXT4-fs panic from previous error\n");
588         }
589 }
590
591 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
592 {
593         struct va_format vaf;
594         va_list args;
595
596         va_start(args, fmt);
597         vaf.fmt = fmt;
598         vaf.va = &args;
599         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
600         va_end(args);
601 }
602
603 void __ext4_warning(struct super_block *sb, const char *function,
604                     unsigned int line, const char *fmt, ...)
605 {
606         struct va_format vaf;
607         va_list args;
608
609         va_start(args, fmt);
610         vaf.fmt = fmt;
611         vaf.va = &args;
612         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
613                sb->s_id, function, line, &vaf);
614         va_end(args);
615 }
616
617 void __ext4_grp_locked_error(const char *function, unsigned int line,
618                              struct super_block *sb, ext4_group_t grp,
619                              unsigned long ino, ext4_fsblk_t block,
620                              const char *fmt, ...)
621 __releases(bitlock)
622 __acquires(bitlock)
623 {
624         struct va_format vaf;
625         va_list args;
626         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
627
628         es->s_last_error_ino = cpu_to_le32(ino);
629         es->s_last_error_block = cpu_to_le64(block);
630         __save_error_info(sb, function, line);
631
632         va_start(args, fmt);
633
634         vaf.fmt = fmt;
635         vaf.va = &args;
636         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
637                sb->s_id, function, line, grp);
638         if (ino)
639                 printk(KERN_CONT "inode %lu: ", ino);
640         if (block)
641                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
642         printk(KERN_CONT "%pV\n", &vaf);
643         va_end(args);
644
645         if (test_opt(sb, ERRORS_CONT)) {
646                 ext4_commit_super(sb, 0);
647                 return;
648         }
649
650         ext4_unlock_group(sb, grp);
651         ext4_handle_error(sb);
652         /*
653          * We only get here in the ERRORS_RO case; relocking the group
654          * may be dangerous, but nothing bad will happen since the
655          * filesystem will have already been marked read/only and the
656          * journal has been aborted.  We return 1 as a hint to callers
657          * who might what to use the return value from
658          * ext4_grp_locked_error() to distinguish between the
659          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
660          * aggressively from the ext4 function in question, with a
661          * more appropriate error code.
662          */
663         ext4_lock_group(sb, grp);
664         return;
665 }
666
667 void ext4_update_dynamic_rev(struct super_block *sb)
668 {
669         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
670
671         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
672                 return;
673
674         ext4_warning(sb,
675                      "updating to rev %d because of new feature flag, "
676                      "running e2fsck is recommended",
677                      EXT4_DYNAMIC_REV);
678
679         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
680         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
681         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
682         /* leave es->s_feature_*compat flags alone */
683         /* es->s_uuid will be set by e2fsck if empty */
684
685         /*
686          * The rest of the superblock fields should be zero, and if not it
687          * means they are likely already in use, so leave them alone.  We
688          * can leave it up to e2fsck to clean up any inconsistencies there.
689          */
690 }
691
692 /*
693  * Open the external journal device
694  */
695 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
696 {
697         struct block_device *bdev;
698         char b[BDEVNAME_SIZE];
699
700         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
701         if (IS_ERR(bdev))
702                 goto fail;
703         return bdev;
704
705 fail:
706         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
707                         __bdevname(dev, b), PTR_ERR(bdev));
708         return NULL;
709 }
710
711 /*
712  * Release the journal device
713  */
714 static void ext4_blkdev_put(struct block_device *bdev)
715 {
716         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
717 }
718
719 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
720 {
721         struct block_device *bdev;
722         bdev = sbi->journal_bdev;
723         if (bdev) {
724                 ext4_blkdev_put(bdev);
725                 sbi->journal_bdev = NULL;
726         }
727 }
728
729 static inline struct inode *orphan_list_entry(struct list_head *l)
730 {
731         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
732 }
733
734 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
735 {
736         struct list_head *l;
737
738         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
739                  le32_to_cpu(sbi->s_es->s_last_orphan));
740
741         printk(KERN_ERR "sb_info orphan list:\n");
742         list_for_each(l, &sbi->s_orphan) {
743                 struct inode *inode = orphan_list_entry(l);
744                 printk(KERN_ERR "  "
745                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
746                        inode->i_sb->s_id, inode->i_ino, inode,
747                        inode->i_mode, inode->i_nlink,
748                        NEXT_ORPHAN(inode));
749         }
750 }
751
752 static void ext4_put_super(struct super_block *sb)
753 {
754         struct ext4_sb_info *sbi = EXT4_SB(sb);
755         struct ext4_super_block *es = sbi->s_es;
756         int i, err;
757
758         ext4_unregister_li_request(sb);
759         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
760
761         flush_workqueue(sbi->dio_unwritten_wq);
762         destroy_workqueue(sbi->dio_unwritten_wq);
763
764         if (sbi->s_journal) {
765                 err = jbd2_journal_destroy(sbi->s_journal);
766                 sbi->s_journal = NULL;
767                 if (err < 0)
768                         ext4_abort(sb, "Couldn't clean up the journal");
769         }
770
771         ext4_es_unregister_shrinker(sb);
772         del_timer(&sbi->s_err_report);
773         ext4_release_system_zone(sb);
774         ext4_mb_release(sb);
775         ext4_ext_release(sb);
776         ext4_xattr_put_super(sb);
777
778         if (!(sb->s_flags & MS_RDONLY)) {
779                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
780                 es->s_state = cpu_to_le16(sbi->s_mount_state);
781         }
782         if (!(sb->s_flags & MS_RDONLY))
783                 ext4_commit_super(sb, 1);
784
785         if (sbi->s_proc) {
786                 remove_proc_entry("options", sbi->s_proc);
787                 remove_proc_entry(sb->s_id, ext4_proc_root);
788         }
789         kobject_del(&sbi->s_kobj);
790
791         for (i = 0; i < sbi->s_gdb_count; i++)
792                 brelse(sbi->s_group_desc[i]);
793         ext4_kvfree(sbi->s_group_desc);
794         ext4_kvfree(sbi->s_flex_groups);
795         percpu_counter_destroy(&sbi->s_freeclusters_counter);
796         percpu_counter_destroy(&sbi->s_freeinodes_counter);
797         percpu_counter_destroy(&sbi->s_dirs_counter);
798         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
799         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
800         brelse(sbi->s_sbh);
801 #ifdef CONFIG_QUOTA
802         for (i = 0; i < MAXQUOTAS; i++)
803                 kfree(sbi->s_qf_names[i]);
804 #endif
805
806         /* Debugging code just in case the in-memory inode orphan list
807          * isn't empty.  The on-disk one can be non-empty if we've
808          * detected an error and taken the fs readonly, but the
809          * in-memory list had better be clean by this point. */
810         if (!list_empty(&sbi->s_orphan))
811                 dump_orphan_list(sb, sbi);
812         J_ASSERT(list_empty(&sbi->s_orphan));
813
814         sync_blockdev(sb->s_bdev);
815         invalidate_bdev(sb->s_bdev);
816         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
817                 /*
818                  * Invalidate the journal device's buffers.  We don't want them
819                  * floating about in memory - the physical journal device may
820                  * hotswapped, and it breaks the `ro-after' testing code.
821                  */
822                 sync_blockdev(sbi->journal_bdev);
823                 invalidate_bdev(sbi->journal_bdev);
824                 ext4_blkdev_remove(sbi);
825         }
826         if (sbi->s_mmp_tsk)
827                 kthread_stop(sbi->s_mmp_tsk);
828         sb->s_fs_info = NULL;
829         /*
830          * Now that we are completely done shutting down the
831          * superblock, we need to actually destroy the kobject.
832          */
833         kobject_put(&sbi->s_kobj);
834         wait_for_completion(&sbi->s_kobj_unregister);
835         if (sbi->s_chksum_driver)
836                 crypto_free_shash(sbi->s_chksum_driver);
837         kfree(sbi->s_blockgroup_lock);
838         kfree(sbi);
839 }
840
841 static struct kmem_cache *ext4_inode_cachep;
842
843 /*
844  * Called inside transaction, so use GFP_NOFS
845  */
846 static struct inode *ext4_alloc_inode(struct super_block *sb)
847 {
848         struct ext4_inode_info *ei;
849
850         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
851         if (!ei)
852                 return NULL;
853
854         ei->vfs_inode.i_version = 1;
855         INIT_LIST_HEAD(&ei->i_prealloc_list);
856         spin_lock_init(&ei->i_prealloc_lock);
857         ext4_es_init_tree(&ei->i_es_tree);
858         rwlock_init(&ei->i_es_lock);
859         INIT_LIST_HEAD(&ei->i_es_lru);
860         ei->i_es_lru_nr = 0;
861         ei->i_reserved_data_blocks = 0;
862         ei->i_reserved_meta_blocks = 0;
863         ei->i_allocated_meta_blocks = 0;
864         ei->i_da_metadata_calc_len = 0;
865         ei->i_da_metadata_calc_last_lblock = 0;
866         spin_lock_init(&(ei->i_block_reservation_lock));
867 #ifdef CONFIG_QUOTA
868         ei->i_reserved_quota = 0;
869 #endif
870         ei->jinode = NULL;
871         INIT_LIST_HEAD(&ei->i_completed_io_list);
872         spin_lock_init(&ei->i_completed_io_lock);
873         ei->i_sync_tid = 0;
874         ei->i_datasync_tid = 0;
875         atomic_set(&ei->i_ioend_count, 0);
876         atomic_set(&ei->i_unwritten, 0);
877         INIT_WORK(&ei->i_unwritten_work, ext4_end_io_work);
878
879         return &ei->vfs_inode;
880 }
881
882 static int ext4_drop_inode(struct inode *inode)
883 {
884         int drop = generic_drop_inode(inode);
885
886         trace_ext4_drop_inode(inode, drop);
887         return drop;
888 }
889
890 static void ext4_i_callback(struct rcu_head *head)
891 {
892         struct inode *inode = container_of(head, struct inode, i_rcu);
893         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
894 }
895
896 static void ext4_destroy_inode(struct inode *inode)
897 {
898         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
899                 ext4_msg(inode->i_sb, KERN_ERR,
900                          "Inode %lu (%p): orphan list check failed!",
901                          inode->i_ino, EXT4_I(inode));
902                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
903                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
904                                 true);
905                 dump_stack();
906         }
907         call_rcu(&inode->i_rcu, ext4_i_callback);
908 }
909
910 static void init_once(void *foo)
911 {
912         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
913
914         INIT_LIST_HEAD(&ei->i_orphan);
915         init_rwsem(&ei->xattr_sem);
916         init_rwsem(&ei->i_data_sem);
917         inode_init_once(&ei->vfs_inode);
918 }
919
920 static int init_inodecache(void)
921 {
922         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
923                                              sizeof(struct ext4_inode_info),
924                                              0, (SLAB_RECLAIM_ACCOUNT|
925                                                 SLAB_MEM_SPREAD),
926                                              init_once);
927         if (ext4_inode_cachep == NULL)
928                 return -ENOMEM;
929         return 0;
930 }
931
932 static void destroy_inodecache(void)
933 {
934         /*
935          * Make sure all delayed rcu free inodes are flushed before we
936          * destroy cache.
937          */
938         rcu_barrier();
939         kmem_cache_destroy(ext4_inode_cachep);
940 }
941
942 void ext4_clear_inode(struct inode *inode)
943 {
944         invalidate_inode_buffers(inode);
945         clear_inode(inode);
946         dquot_drop(inode);
947         ext4_discard_preallocations(inode);
948         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
949         ext4_es_lru_del(inode);
950         if (EXT4_I(inode)->jinode) {
951                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
952                                                EXT4_I(inode)->jinode);
953                 jbd2_free_inode(EXT4_I(inode)->jinode);
954                 EXT4_I(inode)->jinode = NULL;
955         }
956 }
957
958 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
959                                         u64 ino, u32 generation)
960 {
961         struct inode *inode;
962
963         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
964                 return ERR_PTR(-ESTALE);
965         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
966                 return ERR_PTR(-ESTALE);
967
968         /* iget isn't really right if the inode is currently unallocated!!
969          *
970          * ext4_read_inode will return a bad_inode if the inode had been
971          * deleted, so we should be safe.
972          *
973          * Currently we don't know the generation for parent directory, so
974          * a generation of 0 means "accept any"
975          */
976         inode = ext4_iget_normal(sb, ino);
977         if (IS_ERR(inode))
978                 return ERR_CAST(inode);
979         if (generation && inode->i_generation != generation) {
980                 iput(inode);
981                 return ERR_PTR(-ESTALE);
982         }
983
984         return inode;
985 }
986
987 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
988                                         int fh_len, int fh_type)
989 {
990         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
991                                     ext4_nfs_get_inode);
992 }
993
994 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
995                                         int fh_len, int fh_type)
996 {
997         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
998                                     ext4_nfs_get_inode);
999 }
1000
1001 /*
1002  * Try to release metadata pages (indirect blocks, directories) which are
1003  * mapped via the block device.  Since these pages could have journal heads
1004  * which would prevent try_to_free_buffers() from freeing them, we must use
1005  * jbd2 layer's try_to_free_buffers() function to release them.
1006  */
1007 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1008                                  gfp_t wait)
1009 {
1010         journal_t *journal = EXT4_SB(sb)->s_journal;
1011
1012         WARN_ON(PageChecked(page));
1013         if (!page_has_buffers(page))
1014                 return 0;
1015         if (journal)
1016                 return jbd2_journal_try_to_free_buffers(journal, page,
1017                                                         wait & ~__GFP_WAIT);
1018         return try_to_free_buffers(page);
1019 }
1020
1021 #ifdef CONFIG_QUOTA
1022 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1023 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1024
1025 static int ext4_write_dquot(struct dquot *dquot);
1026 static int ext4_acquire_dquot(struct dquot *dquot);
1027 static int ext4_release_dquot(struct dquot *dquot);
1028 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1029 static int ext4_write_info(struct super_block *sb, int type);
1030 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1031                          struct path *path);
1032 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1033                                  int format_id);
1034 static int ext4_quota_off(struct super_block *sb, int type);
1035 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1036 static int ext4_quota_on_mount(struct super_block *sb, int type);
1037 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1038                                size_t len, loff_t off);
1039 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1040                                 const char *data, size_t len, loff_t off);
1041 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1042                              unsigned int flags);
1043 static int ext4_enable_quotas(struct super_block *sb);
1044
1045 static const struct dquot_operations ext4_quota_operations = {
1046         .get_reserved_space = ext4_get_reserved_space,
1047         .write_dquot    = ext4_write_dquot,
1048         .acquire_dquot  = ext4_acquire_dquot,
1049         .release_dquot  = ext4_release_dquot,
1050         .mark_dirty     = ext4_mark_dquot_dirty,
1051         .write_info     = ext4_write_info,
1052         .alloc_dquot    = dquot_alloc,
1053         .destroy_dquot  = dquot_destroy,
1054 };
1055
1056 static const struct quotactl_ops ext4_qctl_operations = {
1057         .quota_on       = ext4_quota_on,
1058         .quota_off      = ext4_quota_off,
1059         .quota_sync     = dquot_quota_sync,
1060         .get_info       = dquot_get_dqinfo,
1061         .set_info       = dquot_set_dqinfo,
1062         .get_dqblk      = dquot_get_dqblk,
1063         .set_dqblk      = dquot_set_dqblk
1064 };
1065
1066 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1067         .quota_on_meta  = ext4_quota_on_sysfile,
1068         .quota_off      = ext4_quota_off_sysfile,
1069         .quota_sync     = dquot_quota_sync,
1070         .get_info       = dquot_get_dqinfo,
1071         .set_info       = dquot_set_dqinfo,
1072         .get_dqblk      = dquot_get_dqblk,
1073         .set_dqblk      = dquot_set_dqblk
1074 };
1075 #endif
1076
1077 static const struct super_operations ext4_sops = {
1078         .alloc_inode    = ext4_alloc_inode,
1079         .destroy_inode  = ext4_destroy_inode,
1080         .write_inode    = ext4_write_inode,
1081         .dirty_inode    = ext4_dirty_inode,
1082         .drop_inode     = ext4_drop_inode,
1083         .evict_inode    = ext4_evict_inode,
1084         .put_super      = ext4_put_super,
1085         .sync_fs        = ext4_sync_fs,
1086         .freeze_fs      = ext4_freeze,
1087         .unfreeze_fs    = ext4_unfreeze,
1088         .statfs         = ext4_statfs,
1089         .remount_fs     = ext4_remount,
1090         .show_options   = ext4_show_options,
1091 #ifdef CONFIG_QUOTA
1092         .quota_read     = ext4_quota_read,
1093         .quota_write    = ext4_quota_write,
1094 #endif
1095         .bdev_try_to_free_page = bdev_try_to_free_page,
1096 };
1097
1098 static const struct super_operations ext4_nojournal_sops = {
1099         .alloc_inode    = ext4_alloc_inode,
1100         .destroy_inode  = ext4_destroy_inode,
1101         .write_inode    = ext4_write_inode,
1102         .dirty_inode    = ext4_dirty_inode,
1103         .drop_inode     = ext4_drop_inode,
1104         .evict_inode    = ext4_evict_inode,
1105         .put_super      = ext4_put_super,
1106         .statfs         = ext4_statfs,
1107         .remount_fs     = ext4_remount,
1108         .show_options   = ext4_show_options,
1109 #ifdef CONFIG_QUOTA
1110         .quota_read     = ext4_quota_read,
1111         .quota_write    = ext4_quota_write,
1112 #endif
1113         .bdev_try_to_free_page = bdev_try_to_free_page,
1114 };
1115
1116 static const struct export_operations ext4_export_ops = {
1117         .fh_to_dentry = ext4_fh_to_dentry,
1118         .fh_to_parent = ext4_fh_to_parent,
1119         .get_parent = ext4_get_parent,
1120 };
1121
1122 enum {
1123         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1124         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1125         Opt_nouid32, Opt_debug, Opt_removed,
1126         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1127         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1128         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1129         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1130         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1131         Opt_data_err_abort, Opt_data_err_ignore,
1132         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1133         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1134         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1135         Opt_usrquota, Opt_grpquota, Opt_i_version,
1136         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1137         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1138         Opt_inode_readahead_blks, Opt_journal_ioprio,
1139         Opt_dioread_nolock, Opt_dioread_lock,
1140         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1141         Opt_max_dir_size_kb,
1142 };
1143
1144 static const match_table_t tokens = {
1145         {Opt_bsd_df, "bsddf"},
1146         {Opt_minix_df, "minixdf"},
1147         {Opt_grpid, "grpid"},
1148         {Opt_grpid, "bsdgroups"},
1149         {Opt_nogrpid, "nogrpid"},
1150         {Opt_nogrpid, "sysvgroups"},
1151         {Opt_resgid, "resgid=%u"},
1152         {Opt_resuid, "resuid=%u"},
1153         {Opt_sb, "sb=%u"},
1154         {Opt_err_cont, "errors=continue"},
1155         {Opt_err_panic, "errors=panic"},
1156         {Opt_err_ro, "errors=remount-ro"},
1157         {Opt_nouid32, "nouid32"},
1158         {Opt_debug, "debug"},
1159         {Opt_removed, "oldalloc"},
1160         {Opt_removed, "orlov"},
1161         {Opt_user_xattr, "user_xattr"},
1162         {Opt_nouser_xattr, "nouser_xattr"},
1163         {Opt_acl, "acl"},
1164         {Opt_noacl, "noacl"},
1165         {Opt_noload, "norecovery"},
1166         {Opt_noload, "noload"},
1167         {Opt_removed, "nobh"},
1168         {Opt_removed, "bh"},
1169         {Opt_commit, "commit=%u"},
1170         {Opt_min_batch_time, "min_batch_time=%u"},
1171         {Opt_max_batch_time, "max_batch_time=%u"},
1172         {Opt_journal_dev, "journal_dev=%u"},
1173         {Opt_journal_checksum, "journal_checksum"},
1174         {Opt_journal_async_commit, "journal_async_commit"},
1175         {Opt_abort, "abort"},
1176         {Opt_data_journal, "data=journal"},
1177         {Opt_data_ordered, "data=ordered"},
1178         {Opt_data_writeback, "data=writeback"},
1179         {Opt_data_err_abort, "data_err=abort"},
1180         {Opt_data_err_ignore, "data_err=ignore"},
1181         {Opt_offusrjquota, "usrjquota="},
1182         {Opt_usrjquota, "usrjquota=%s"},
1183         {Opt_offgrpjquota, "grpjquota="},
1184         {Opt_grpjquota, "grpjquota=%s"},
1185         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1186         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1187         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1188         {Opt_grpquota, "grpquota"},
1189         {Opt_noquota, "noquota"},
1190         {Opt_quota, "quota"},
1191         {Opt_usrquota, "usrquota"},
1192         {Opt_barrier, "barrier=%u"},
1193         {Opt_barrier, "barrier"},
1194         {Opt_nobarrier, "nobarrier"},
1195         {Opt_i_version, "i_version"},
1196         {Opt_stripe, "stripe=%u"},
1197         {Opt_delalloc, "delalloc"},
1198         {Opt_nodelalloc, "nodelalloc"},
1199         {Opt_removed, "mblk_io_submit"},
1200         {Opt_removed, "nomblk_io_submit"},
1201         {Opt_block_validity, "block_validity"},
1202         {Opt_noblock_validity, "noblock_validity"},
1203         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1204         {Opt_journal_ioprio, "journal_ioprio=%u"},
1205         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1206         {Opt_auto_da_alloc, "auto_da_alloc"},
1207         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1208         {Opt_dioread_nolock, "dioread_nolock"},
1209         {Opt_dioread_lock, "dioread_lock"},
1210         {Opt_discard, "discard"},
1211         {Opt_nodiscard, "nodiscard"},
1212         {Opt_init_itable, "init_itable=%u"},
1213         {Opt_init_itable, "init_itable"},
1214         {Opt_noinit_itable, "noinit_itable"},
1215         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1216         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1217         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1218         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1219         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1220         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1221         {Opt_err, NULL},
1222 };
1223
1224 static ext4_fsblk_t get_sb_block(void **data)
1225 {
1226         ext4_fsblk_t    sb_block;
1227         char            *options = (char *) *data;
1228
1229         if (!options || strncmp(options, "sb=", 3) != 0)
1230                 return 1;       /* Default location */
1231
1232         options += 3;
1233         /* TODO: use simple_strtoll with >32bit ext4 */
1234         sb_block = simple_strtoul(options, &options, 0);
1235         if (*options && *options != ',') {
1236                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1237                        (char *) *data);
1238                 return 1;
1239         }
1240         if (*options == ',')
1241                 options++;
1242         *data = (void *) options;
1243
1244         return sb_block;
1245 }
1246
1247 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1248 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1249         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1250
1251 #ifdef CONFIG_QUOTA
1252 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1253 {
1254         struct ext4_sb_info *sbi = EXT4_SB(sb);
1255         char *qname;
1256         int ret = -1;
1257
1258         if (sb_any_quota_loaded(sb) &&
1259                 !sbi->s_qf_names[qtype]) {
1260                 ext4_msg(sb, KERN_ERR,
1261                         "Cannot change journaled "
1262                         "quota options when quota turned on");
1263                 return -1;
1264         }
1265         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1266                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1267                          "when QUOTA feature is enabled");
1268                 return -1;
1269         }
1270         qname = match_strdup(args);
1271         if (!qname) {
1272                 ext4_msg(sb, KERN_ERR,
1273                         "Not enough memory for storing quotafile name");
1274                 return -1;
1275         }
1276         if (sbi->s_qf_names[qtype]) {
1277                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1278                         ret = 1;
1279                 else
1280                         ext4_msg(sb, KERN_ERR,
1281                                  "%s quota file already specified",
1282                                  QTYPE2NAME(qtype));
1283                 goto errout;
1284         }
1285         if (strchr(qname, '/')) {
1286                 ext4_msg(sb, KERN_ERR,
1287                         "quotafile must be on filesystem root");
1288                 goto errout;
1289         }
1290         sbi->s_qf_names[qtype] = qname;
1291         set_opt(sb, QUOTA);
1292         return 1;
1293 errout:
1294         kfree(qname);
1295         return ret;
1296 }
1297
1298 static int clear_qf_name(struct super_block *sb, int qtype)
1299 {
1300
1301         struct ext4_sb_info *sbi = EXT4_SB(sb);
1302
1303         if (sb_any_quota_loaded(sb) &&
1304                 sbi->s_qf_names[qtype]) {
1305                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1306                         " when quota turned on");
1307                 return -1;
1308         }
1309         kfree(sbi->s_qf_names[qtype]);
1310         sbi->s_qf_names[qtype] = NULL;
1311         return 1;
1312 }
1313 #endif
1314
1315 #define MOPT_SET        0x0001
1316 #define MOPT_CLEAR      0x0002
1317 #define MOPT_NOSUPPORT  0x0004
1318 #define MOPT_EXPLICIT   0x0008
1319 #define MOPT_CLEAR_ERR  0x0010
1320 #define MOPT_GTE0       0x0020
1321 #ifdef CONFIG_QUOTA
1322 #define MOPT_Q          0
1323 #define MOPT_QFMT       0x0040
1324 #else
1325 #define MOPT_Q          MOPT_NOSUPPORT
1326 #define MOPT_QFMT       MOPT_NOSUPPORT
1327 #endif
1328 #define MOPT_DATAJ      0x0080
1329 #define MOPT_NO_EXT2    0x0100
1330 #define MOPT_NO_EXT3    0x0200
1331 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1332
1333 static const struct mount_opts {
1334         int     token;
1335         int     mount_opt;
1336         int     flags;
1337 } ext4_mount_opts[] = {
1338         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1339         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1340         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1341         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1342         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1343         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1344         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1345          MOPT_EXT4_ONLY | MOPT_SET},
1346         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1347          MOPT_EXT4_ONLY | MOPT_CLEAR},
1348         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1349         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1350         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1351          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1352         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1353          MOPT_EXT4_ONLY | MOPT_CLEAR},
1354         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1355          MOPT_EXT4_ONLY | MOPT_SET},
1356         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1357                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1358          MOPT_EXT4_ONLY | MOPT_SET},
1359         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1360         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1361         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1362         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1363         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1364          MOPT_NO_EXT2 | MOPT_SET},
1365         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1366          MOPT_NO_EXT2 | MOPT_CLEAR},
1367         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1368         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1369         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1370         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1371         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1372         {Opt_commit, 0, MOPT_GTE0},
1373         {Opt_max_batch_time, 0, MOPT_GTE0},
1374         {Opt_min_batch_time, 0, MOPT_GTE0},
1375         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1376         {Opt_init_itable, 0, MOPT_GTE0},
1377         {Opt_stripe, 0, MOPT_GTE0},
1378         {Opt_resuid, 0, MOPT_GTE0},
1379         {Opt_resgid, 0, MOPT_GTE0},
1380         {Opt_journal_dev, 0, MOPT_GTE0},
1381         {Opt_journal_ioprio, 0, MOPT_GTE0},
1382         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1383         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1384         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1385          MOPT_NO_EXT2 | MOPT_DATAJ},
1386         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1387         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1388 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1389         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1390         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1391 #else
1392         {Opt_acl, 0, MOPT_NOSUPPORT},
1393         {Opt_noacl, 0, MOPT_NOSUPPORT},
1394 #endif
1395         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1396         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1397         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1398         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1399                                                         MOPT_SET | MOPT_Q},
1400         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1401                                                         MOPT_SET | MOPT_Q},
1402         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1403                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1404         {Opt_usrjquota, 0, MOPT_Q},
1405         {Opt_grpjquota, 0, MOPT_Q},
1406         {Opt_offusrjquota, 0, MOPT_Q},
1407         {Opt_offgrpjquota, 0, MOPT_Q},
1408         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1409         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1410         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1411         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1412         {Opt_err, 0, 0}
1413 };
1414
1415 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1416                             substring_t *args, unsigned long *journal_devnum,
1417                             unsigned int *journal_ioprio, int is_remount)
1418 {
1419         struct ext4_sb_info *sbi = EXT4_SB(sb);
1420         const struct mount_opts *m;
1421         kuid_t uid;
1422         kgid_t gid;
1423         int arg = 0;
1424
1425 #ifdef CONFIG_QUOTA
1426         if (token == Opt_usrjquota)
1427                 return set_qf_name(sb, USRQUOTA, &args[0]);
1428         else if (token == Opt_grpjquota)
1429                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1430         else if (token == Opt_offusrjquota)
1431                 return clear_qf_name(sb, USRQUOTA);
1432         else if (token == Opt_offgrpjquota)
1433                 return clear_qf_name(sb, GRPQUOTA);
1434 #endif
1435         switch (token) {
1436         case Opt_noacl:
1437         case Opt_nouser_xattr:
1438                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1439                 break;
1440         case Opt_sb:
1441                 return 1;       /* handled by get_sb_block() */
1442         case Opt_removed:
1443                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1444                 return 1;
1445         case Opt_abort:
1446                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1447                 return 1;
1448         case Opt_i_version:
1449                 sb->s_flags |= MS_I_VERSION;
1450                 return 1;
1451         }
1452
1453         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1454                 if (token == m->token)
1455                         break;
1456
1457         if (m->token == Opt_err) {
1458                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1459                          "or missing value", opt);
1460                 return -1;
1461         }
1462
1463         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1464                 ext4_msg(sb, KERN_ERR,
1465                          "Mount option \"%s\" incompatible with ext2", opt);
1466                 return -1;
1467         }
1468         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1469                 ext4_msg(sb, KERN_ERR,
1470                          "Mount option \"%s\" incompatible with ext3", opt);
1471                 return -1;
1472         }
1473
1474         if (args->from && match_int(args, &arg))
1475                 return -1;
1476         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1477                 return -1;
1478         if (m->flags & MOPT_EXPLICIT)
1479                 set_opt2(sb, EXPLICIT_DELALLOC);
1480         if (m->flags & MOPT_CLEAR_ERR)
1481                 clear_opt(sb, ERRORS_MASK);
1482         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1483                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1484                          "options when quota turned on");
1485                 return -1;
1486         }
1487
1488         if (m->flags & MOPT_NOSUPPORT) {
1489                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1490         } else if (token == Opt_commit) {
1491                 if (arg == 0)
1492                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1493                 sbi->s_commit_interval = HZ * arg;
1494         } else if (token == Opt_max_batch_time) {
1495                 sbi->s_max_batch_time = arg;
1496         } else if (token == Opt_min_batch_time) {
1497                 sbi->s_min_batch_time = arg;
1498         } else if (token == Opt_inode_readahead_blks) {
1499                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1500                         ext4_msg(sb, KERN_ERR,
1501                                  "EXT4-fs: inode_readahead_blks must be "
1502                                  "0 or a power of 2 smaller than 2^31");
1503                         return -1;
1504                 }
1505                 sbi->s_inode_readahead_blks = arg;
1506         } else if (token == Opt_init_itable) {
1507                 set_opt(sb, INIT_INODE_TABLE);
1508                 if (!args->from)
1509                         arg = EXT4_DEF_LI_WAIT_MULT;
1510                 sbi->s_li_wait_mult = arg;
1511         } else if (token == Opt_max_dir_size_kb) {
1512                 sbi->s_max_dir_size_kb = arg;
1513         } else if (token == Opt_stripe) {
1514                 sbi->s_stripe = arg;
1515         } else if (token == Opt_resuid) {
1516                 uid = make_kuid(current_user_ns(), arg);
1517                 if (!uid_valid(uid)) {
1518                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1519                         return -1;
1520                 }
1521                 sbi->s_resuid = uid;
1522         } else if (token == Opt_resgid) {
1523                 gid = make_kgid(current_user_ns(), arg);
1524                 if (!gid_valid(gid)) {
1525                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1526                         return -1;
1527                 }
1528                 sbi->s_resgid = gid;
1529         } else if (token == Opt_journal_dev) {
1530                 if (is_remount) {
1531                         ext4_msg(sb, KERN_ERR,
1532                                  "Cannot specify journal on remount");
1533                         return -1;
1534                 }
1535                 *journal_devnum = arg;
1536         } else if (token == Opt_journal_ioprio) {
1537                 if (arg > 7) {
1538                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1539                                  " (must be 0-7)");
1540                         return -1;
1541                 }
1542                 *journal_ioprio =
1543                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1544         } else if (m->flags & MOPT_DATAJ) {
1545                 if (is_remount) {
1546                         if (!sbi->s_journal)
1547                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1548                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1549                                 ext4_msg(sb, KERN_ERR,
1550                                          "Cannot change data mode on remount");
1551                                 return -1;
1552                         }
1553                 } else {
1554                         clear_opt(sb, DATA_FLAGS);
1555                         sbi->s_mount_opt |= m->mount_opt;
1556                 }
1557 #ifdef CONFIG_QUOTA
1558         } else if (m->flags & MOPT_QFMT) {
1559                 if (sb_any_quota_loaded(sb) &&
1560                     sbi->s_jquota_fmt != m->mount_opt) {
1561                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1562                                  "quota options when quota turned on");
1563                         return -1;
1564                 }
1565                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1566                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1567                         ext4_msg(sb, KERN_ERR,
1568                                  "Cannot set journaled quota options "
1569                                  "when QUOTA feature is enabled");
1570                         return -1;
1571                 }
1572                 sbi->s_jquota_fmt = m->mount_opt;
1573 #endif
1574         } else {
1575                 if (!args->from)
1576                         arg = 1;
1577                 if (m->flags & MOPT_CLEAR)
1578                         arg = !arg;
1579                 else if (unlikely(!(m->flags & MOPT_SET))) {
1580                         ext4_msg(sb, KERN_WARNING,
1581                                  "buggy handling of option %s", opt);
1582                         WARN_ON(1);
1583                         return -1;
1584                 }
1585                 if (arg != 0)
1586                         sbi->s_mount_opt |= m->mount_opt;
1587                 else
1588                         sbi->s_mount_opt &= ~m->mount_opt;
1589         }
1590         return 1;
1591 }
1592
1593 static int parse_options(char *options, struct super_block *sb,
1594                          unsigned long *journal_devnum,
1595                          unsigned int *journal_ioprio,
1596                          int is_remount)
1597 {
1598         struct ext4_sb_info *sbi = EXT4_SB(sb);
1599         char *p;
1600         substring_t args[MAX_OPT_ARGS];
1601         int token;
1602
1603         if (!options)
1604                 return 1;
1605
1606         while ((p = strsep(&options, ",")) != NULL) {
1607                 if (!*p)
1608                         continue;
1609                 /*
1610                  * Initialize args struct so we know whether arg was
1611                  * found; some options take optional arguments.
1612                  */
1613                 args[0].to = args[0].from = NULL;
1614                 token = match_token(p, tokens, args);
1615                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1616                                      journal_ioprio, is_remount) < 0)
1617                         return 0;
1618         }
1619 #ifdef CONFIG_QUOTA
1620         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1621             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1622                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1623                          "feature is enabled");
1624                 return 0;
1625         }
1626         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1627                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1628                         clear_opt(sb, USRQUOTA);
1629
1630                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1631                         clear_opt(sb, GRPQUOTA);
1632
1633                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1634                         ext4_msg(sb, KERN_ERR, "old and new quota "
1635                                         "format mixing");
1636                         return 0;
1637                 }
1638
1639                 if (!sbi->s_jquota_fmt) {
1640                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1641                                         "not specified");
1642                         return 0;
1643                 }
1644         }
1645 #endif
1646         if (test_opt(sb, DIOREAD_NOLOCK)) {
1647                 int blocksize =
1648                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1649
1650                 if (blocksize < PAGE_CACHE_SIZE) {
1651                         ext4_msg(sb, KERN_ERR, "can't mount with "
1652                                  "dioread_nolock if block size != PAGE_SIZE");
1653                         return 0;
1654                 }
1655         }
1656         return 1;
1657 }
1658
1659 static inline void ext4_show_quota_options(struct seq_file *seq,
1660                                            struct super_block *sb)
1661 {
1662 #if defined(CONFIG_QUOTA)
1663         struct ext4_sb_info *sbi = EXT4_SB(sb);
1664
1665         if (sbi->s_jquota_fmt) {
1666                 char *fmtname = "";
1667
1668                 switch (sbi->s_jquota_fmt) {
1669                 case QFMT_VFS_OLD:
1670                         fmtname = "vfsold";
1671                         break;
1672                 case QFMT_VFS_V0:
1673                         fmtname = "vfsv0";
1674                         break;
1675                 case QFMT_VFS_V1:
1676                         fmtname = "vfsv1";
1677                         break;
1678                 }
1679                 seq_printf(seq, ",jqfmt=%s", fmtname);
1680         }
1681
1682         if (sbi->s_qf_names[USRQUOTA])
1683                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1684
1685         if (sbi->s_qf_names[GRPQUOTA])
1686                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1687 #endif
1688 }
1689
1690 static const char *token2str(int token)
1691 {
1692         const struct match_token *t;
1693
1694         for (t = tokens; t->token != Opt_err; t++)
1695                 if (t->token == token && !strchr(t->pattern, '='))
1696                         break;
1697         return t->pattern;
1698 }
1699
1700 /*
1701  * Show an option if
1702  *  - it's set to a non-default value OR
1703  *  - if the per-sb default is different from the global default
1704  */
1705 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1706                               int nodefs)
1707 {
1708         struct ext4_sb_info *sbi = EXT4_SB(sb);
1709         struct ext4_super_block *es = sbi->s_es;
1710         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1711         const struct mount_opts *m;
1712         char sep = nodefs ? '\n' : ',';
1713
1714 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1715 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1716
1717         if (sbi->s_sb_block != 1)
1718                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1719
1720         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1721                 int want_set = m->flags & MOPT_SET;
1722                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1723                     (m->flags & MOPT_CLEAR_ERR))
1724                         continue;
1725                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1726                         continue; /* skip if same as the default */
1727                 if ((want_set &&
1728                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1729                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1730                         continue; /* select Opt_noFoo vs Opt_Foo */
1731                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1732         }
1733
1734         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1735             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1736                 SEQ_OPTS_PRINT("resuid=%u",
1737                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1738         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1739             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1740                 SEQ_OPTS_PRINT("resgid=%u",
1741                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1742         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1743         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1744                 SEQ_OPTS_PUTS("errors=remount-ro");
1745         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1746                 SEQ_OPTS_PUTS("errors=continue");
1747         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1748                 SEQ_OPTS_PUTS("errors=panic");
1749         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1750                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1751         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1752                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1753         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1754                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1755         if (sb->s_flags & MS_I_VERSION)
1756                 SEQ_OPTS_PUTS("i_version");
1757         if (nodefs || sbi->s_stripe)
1758                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1759         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1760                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1761                         SEQ_OPTS_PUTS("data=journal");
1762                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1763                         SEQ_OPTS_PUTS("data=ordered");
1764                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1765                         SEQ_OPTS_PUTS("data=writeback");
1766         }
1767         if (nodefs ||
1768             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1769                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1770                                sbi->s_inode_readahead_blks);
1771
1772         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1773                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1774                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1775         if (nodefs || sbi->s_max_dir_size_kb)
1776                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1777
1778         ext4_show_quota_options(seq, sb);
1779         return 0;
1780 }
1781
1782 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1783 {
1784         return _ext4_show_options(seq, root->d_sb, 0);
1785 }
1786
1787 static int options_seq_show(struct seq_file *seq, void *offset)
1788 {
1789         struct super_block *sb = seq->private;
1790         int rc;
1791
1792         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1793         rc = _ext4_show_options(seq, sb, 1);
1794         seq_puts(seq, "\n");
1795         return rc;
1796 }
1797
1798 static int options_open_fs(struct inode *inode, struct file *file)
1799 {
1800         return single_open(file, options_seq_show, PDE_DATA(inode));
1801 }
1802
1803 static const struct file_operations ext4_seq_options_fops = {
1804         .owner = THIS_MODULE,
1805         .open = options_open_fs,
1806         .read = seq_read,
1807         .llseek = seq_lseek,
1808         .release = single_release,
1809 };
1810
1811 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1812                             int read_only)
1813 {
1814         struct ext4_sb_info *sbi = EXT4_SB(sb);
1815         int res = 0;
1816
1817         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1818                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1819                          "forcing read-only mode");
1820                 res = MS_RDONLY;
1821         }
1822         if (read_only)
1823                 goto done;
1824         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1825                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1826                          "running e2fsck is recommended");
1827         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1828                 ext4_msg(sb, KERN_WARNING,
1829                          "warning: mounting fs with errors, "
1830                          "running e2fsck is recommended");
1831         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1832                  le16_to_cpu(es->s_mnt_count) >=
1833                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1834                 ext4_msg(sb, KERN_WARNING,
1835                          "warning: maximal mount count reached, "
1836                          "running e2fsck is recommended");
1837         else if (le32_to_cpu(es->s_checkinterval) &&
1838                 (le32_to_cpu(es->s_lastcheck) +
1839                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1840                 ext4_msg(sb, KERN_WARNING,
1841                          "warning: checktime reached, "
1842                          "running e2fsck is recommended");
1843         if (!sbi->s_journal)
1844                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1845         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1846                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1847         le16_add_cpu(&es->s_mnt_count, 1);
1848         es->s_mtime = cpu_to_le32(get_seconds());
1849         ext4_update_dynamic_rev(sb);
1850         if (sbi->s_journal)
1851                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1852
1853         ext4_commit_super(sb, 1);
1854 done:
1855         if (test_opt(sb, DEBUG))
1856                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1857                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1858                         sb->s_blocksize,
1859                         sbi->s_groups_count,
1860                         EXT4_BLOCKS_PER_GROUP(sb),
1861                         EXT4_INODES_PER_GROUP(sb),
1862                         sbi->s_mount_opt, sbi->s_mount_opt2);
1863
1864         cleancache_init_fs(sb);
1865         return res;
1866 }
1867
1868 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1869 {
1870         struct ext4_sb_info *sbi = EXT4_SB(sb);
1871         struct flex_groups *new_groups;
1872         int size;
1873
1874         if (!sbi->s_log_groups_per_flex)
1875                 return 0;
1876
1877         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1878         if (size <= sbi->s_flex_groups_allocated)
1879                 return 0;
1880
1881         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1882         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1883         if (!new_groups) {
1884                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1885                          size / (int) sizeof(struct flex_groups));
1886                 return -ENOMEM;
1887         }
1888
1889         if (sbi->s_flex_groups) {
1890                 memcpy(new_groups, sbi->s_flex_groups,
1891                        (sbi->s_flex_groups_allocated *
1892                         sizeof(struct flex_groups)));
1893                 ext4_kvfree(sbi->s_flex_groups);
1894         }
1895         sbi->s_flex_groups = new_groups;
1896         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1897         return 0;
1898 }
1899
1900 static int ext4_fill_flex_info(struct super_block *sb)
1901 {
1902         struct ext4_sb_info *sbi = EXT4_SB(sb);
1903         struct ext4_group_desc *gdp = NULL;
1904         ext4_group_t flex_group;
1905         unsigned int groups_per_flex = 0;
1906         int i, err;
1907
1908         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1909         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1910                 sbi->s_log_groups_per_flex = 0;
1911                 return 1;
1912         }
1913         groups_per_flex = 1U << sbi->s_log_groups_per_flex;
1914
1915         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1916         if (err)
1917                 goto failed;
1918
1919         for (i = 0; i < sbi->s_groups_count; i++) {
1920                 gdp = ext4_get_group_desc(sb, i, NULL);
1921
1922                 flex_group = ext4_flex_group(sbi, i);
1923                 atomic_add(ext4_free_inodes_count(sb, gdp),
1924                            &sbi->s_flex_groups[flex_group].free_inodes);
1925                 atomic64_add(ext4_free_group_clusters(sb, gdp),
1926                              &sbi->s_flex_groups[flex_group].free_clusters);
1927                 atomic_add(ext4_used_dirs_count(sb, gdp),
1928                            &sbi->s_flex_groups[flex_group].used_dirs);
1929         }
1930
1931         return 1;
1932 failed:
1933         return 0;
1934 }
1935
1936 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1937                                    struct ext4_group_desc *gdp)
1938 {
1939         int offset;
1940         __u16 crc = 0;
1941         __le32 le_group = cpu_to_le32(block_group);
1942
1943         if ((sbi->s_es->s_feature_ro_compat &
1944              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1945                 /* Use new metadata_csum algorithm */
1946                 __le16 save_csum;
1947                 __u32 csum32;
1948
1949                 save_csum = gdp->bg_checksum;
1950                 gdp->bg_checksum = 0;
1951                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1952                                      sizeof(le_group));
1953                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
1954                                      sbi->s_desc_size);
1955                 gdp->bg_checksum = save_csum;
1956
1957                 crc = csum32 & 0xFFFF;
1958                 goto out;
1959         }
1960
1961         /* old crc16 code */
1962         if (!(sbi->s_es->s_feature_ro_compat &
1963               cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)))
1964                 return 0;
1965
1966         offset = offsetof(struct ext4_group_desc, bg_checksum);
1967
1968         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1969         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1970         crc = crc16(crc, (__u8 *)gdp, offset);
1971         offset += sizeof(gdp->bg_checksum); /* skip checksum */
1972         /* for checksum of struct ext4_group_desc do the rest...*/
1973         if ((sbi->s_es->s_feature_incompat &
1974              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1975             offset < le16_to_cpu(sbi->s_es->s_desc_size))
1976                 crc = crc16(crc, (__u8 *)gdp + offset,
1977                             le16_to_cpu(sbi->s_es->s_desc_size) -
1978                                 offset);
1979
1980 out:
1981         return cpu_to_le16(crc);
1982 }
1983
1984 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
1985                                 struct ext4_group_desc *gdp)
1986 {
1987         if (ext4_has_group_desc_csum(sb) &&
1988             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
1989                                                       block_group, gdp)))
1990                 return 0;
1991
1992         return 1;
1993 }
1994
1995 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
1996                               struct ext4_group_desc *gdp)
1997 {
1998         if (!ext4_has_group_desc_csum(sb))
1999                 return;
2000         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2001 }
2002
2003 /* Called at mount-time, super-block is locked */
2004 static int ext4_check_descriptors(struct super_block *sb,
2005                                   ext4_group_t *first_not_zeroed)
2006 {
2007         struct ext4_sb_info *sbi = EXT4_SB(sb);
2008         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2009         ext4_fsblk_t last_block;
2010         ext4_fsblk_t block_bitmap;
2011         ext4_fsblk_t inode_bitmap;
2012         ext4_fsblk_t inode_table;
2013         int flexbg_flag = 0;
2014         ext4_group_t i, grp = sbi->s_groups_count;
2015
2016         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2017                 flexbg_flag = 1;
2018
2019         ext4_debug("Checking group descriptors");
2020
2021         for (i = 0; i < sbi->s_groups_count; i++) {
2022                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2023
2024                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2025                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2026                 else
2027                         last_block = first_block +
2028                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2029
2030                 if ((grp == sbi->s_groups_count) &&
2031                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2032                         grp = i;
2033
2034                 block_bitmap = ext4_block_bitmap(sb, gdp);
2035                 if (block_bitmap < first_block || block_bitmap > last_block) {
2036                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2037                                "Block bitmap for group %u not in group "
2038                                "(block %llu)!", i, block_bitmap);
2039                         return 0;
2040                 }
2041                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2042                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2043                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2044                                "Inode bitmap for group %u not in group "
2045                                "(block %llu)!", i, inode_bitmap);
2046                         return 0;
2047                 }
2048                 inode_table = ext4_inode_table(sb, gdp);
2049                 if (inode_table < first_block ||
2050                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2051                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2052                                "Inode table for group %u not in group "
2053                                "(block %llu)!", i, inode_table);
2054                         return 0;
2055                 }
2056                 ext4_lock_group(sb, i);
2057                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2058                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2059                                  "Checksum for group %u failed (%u!=%u)",
2060                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2061                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2062                         if (!(sb->s_flags & MS_RDONLY)) {
2063                                 ext4_unlock_group(sb, i);
2064                                 return 0;
2065                         }
2066                 }
2067                 ext4_unlock_group(sb, i);
2068                 if (!flexbg_flag)
2069                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2070         }
2071         if (NULL != first_not_zeroed)
2072                 *first_not_zeroed = grp;
2073
2074         ext4_free_blocks_count_set(sbi->s_es,
2075                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2076         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2077         return 1;
2078 }
2079
2080 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2081  * the superblock) which were deleted from all directories, but held open by
2082  * a process at the time of a crash.  We walk the list and try to delete these
2083  * inodes at recovery time (only with a read-write filesystem).
2084  *
2085  * In order to keep the orphan inode chain consistent during traversal (in
2086  * case of crash during recovery), we link each inode into the superblock
2087  * orphan list_head and handle it the same way as an inode deletion during
2088  * normal operation (which journals the operations for us).
2089  *
2090  * We only do an iget() and an iput() on each inode, which is very safe if we
2091  * accidentally point at an in-use or already deleted inode.  The worst that
2092  * can happen in this case is that we get a "bit already cleared" message from
2093  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2094  * e2fsck was run on this filesystem, and it must have already done the orphan
2095  * inode cleanup for us, so we can safely abort without any further action.
2096  */
2097 static void ext4_orphan_cleanup(struct super_block *sb,
2098                                 struct ext4_super_block *es)
2099 {
2100         unsigned int s_flags = sb->s_flags;
2101         int nr_orphans = 0, nr_truncates = 0;
2102 #ifdef CONFIG_QUOTA
2103         int i;
2104 #endif
2105         if (!es->s_last_orphan) {
2106                 jbd_debug(4, "no orphan inodes to clean up\n");
2107                 return;
2108         }
2109
2110         if (bdev_read_only(sb->s_bdev)) {
2111                 ext4_msg(sb, KERN_ERR, "write access "
2112                         "unavailable, skipping orphan cleanup");
2113                 return;
2114         }
2115
2116         /* Check if feature set would not allow a r/w mount */
2117         if (!ext4_feature_set_ok(sb, 0)) {
2118                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2119                          "unknown ROCOMPAT features");
2120                 return;
2121         }
2122
2123         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2124                 /* don't clear list on RO mount w/ errors */
2125                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2126                         jbd_debug(1, "Errors on filesystem, "
2127                                   "clearing orphan list.\n");
2128                         es->s_last_orphan = 0;
2129                 }
2130                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2131                 return;
2132         }
2133
2134         if (s_flags & MS_RDONLY) {
2135                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2136                 sb->s_flags &= ~MS_RDONLY;
2137         }
2138 #ifdef CONFIG_QUOTA
2139         /* Needed for iput() to work correctly and not trash data */
2140         sb->s_flags |= MS_ACTIVE;
2141         /* Turn on quotas so that they are updated correctly */
2142         for (i = 0; i < MAXQUOTAS; i++) {
2143                 if (EXT4_SB(sb)->s_qf_names[i]) {
2144                         int ret = ext4_quota_on_mount(sb, i);
2145                         if (ret < 0)
2146                                 ext4_msg(sb, KERN_ERR,
2147                                         "Cannot turn on journaled "
2148                                         "quota: error %d", ret);
2149                 }
2150         }
2151 #endif
2152
2153         while (es->s_last_orphan) {
2154                 struct inode *inode;
2155
2156                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2157                 if (IS_ERR(inode)) {
2158                         es->s_last_orphan = 0;
2159                         break;
2160                 }
2161
2162                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2163                 dquot_initialize(inode);
2164                 if (inode->i_nlink) {
2165                         ext4_msg(sb, KERN_DEBUG,
2166                                 "%s: truncating inode %lu to %lld bytes",
2167                                 __func__, inode->i_ino, inode->i_size);
2168                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2169                                   inode->i_ino, inode->i_size);
2170                         mutex_lock(&inode->i_mutex);
2171                         ext4_truncate(inode);
2172                         mutex_unlock(&inode->i_mutex);
2173                         nr_truncates++;
2174                 } else {
2175                         ext4_msg(sb, KERN_DEBUG,
2176                                 "%s: deleting unreferenced inode %lu",
2177                                 __func__, inode->i_ino);
2178                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2179                                   inode->i_ino);
2180                         nr_orphans++;
2181                 }
2182                 iput(inode);  /* The delete magic happens here! */
2183         }
2184
2185 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2186
2187         if (nr_orphans)
2188                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2189                        PLURAL(nr_orphans));
2190         if (nr_truncates)
2191                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2192                        PLURAL(nr_truncates));
2193 #ifdef CONFIG_QUOTA
2194         /* Turn quotas off */
2195         for (i = 0; i < MAXQUOTAS; i++) {
2196                 if (sb_dqopt(sb)->files[i])
2197                         dquot_quota_off(sb, i);
2198         }
2199 #endif
2200         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2201 }
2202
2203 /*
2204  * Maximal extent format file size.
2205  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2206  * extent format containers, within a sector_t, and within i_blocks
2207  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2208  * so that won't be a limiting factor.
2209  *
2210  * However there is other limiting factor. We do store extents in the form
2211  * of starting block and length, hence the resulting length of the extent
2212  * covering maximum file size must fit into on-disk format containers as
2213  * well. Given that length is always by 1 unit bigger than max unit (because
2214  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2215  *
2216  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2217  */
2218 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2219 {
2220         loff_t res;
2221         loff_t upper_limit = MAX_LFS_FILESIZE;
2222
2223         /* small i_blocks in vfs inode? */
2224         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2225                 /*
2226                  * CONFIG_LBDAF is not enabled implies the inode
2227                  * i_block represent total blocks in 512 bytes
2228                  * 32 == size of vfs inode i_blocks * 8
2229                  */
2230                 upper_limit = (1LL << 32) - 1;
2231
2232                 /* total blocks in file system block size */
2233                 upper_limit >>= (blkbits - 9);
2234                 upper_limit <<= blkbits;
2235         }
2236
2237         /*
2238          * 32-bit extent-start container, ee_block. We lower the maxbytes
2239          * by one fs block, so ee_len can cover the extent of maximum file
2240          * size
2241          */
2242         res = (1LL << 32) - 1;
2243         res <<= blkbits;
2244
2245         /* Sanity check against vm- & vfs- imposed limits */
2246         if (res > upper_limit)
2247                 res = upper_limit;
2248
2249         return res;
2250 }
2251
2252 /*
2253  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2254  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2255  * We need to be 1 filesystem block less than the 2^48 sector limit.
2256  */
2257 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2258 {
2259         loff_t res = EXT4_NDIR_BLOCKS;
2260         int meta_blocks;
2261         loff_t upper_limit;
2262         /* This is calculated to be the largest file size for a dense, block
2263          * mapped file such that the file's total number of 512-byte sectors,
2264          * including data and all indirect blocks, does not exceed (2^48 - 1).
2265          *
2266          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2267          * number of 512-byte sectors of the file.
2268          */
2269
2270         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2271                 /*
2272                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2273                  * the inode i_block field represents total file blocks in
2274                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2275                  */
2276                 upper_limit = (1LL << 32) - 1;
2277
2278                 /* total blocks in file system block size */
2279                 upper_limit >>= (bits - 9);
2280
2281         } else {
2282                 /*
2283                  * We use 48 bit ext4_inode i_blocks
2284                  * With EXT4_HUGE_FILE_FL set the i_blocks
2285                  * represent total number of blocks in
2286                  * file system block size
2287                  */
2288                 upper_limit = (1LL << 48) - 1;
2289
2290         }
2291
2292         /* indirect blocks */
2293         meta_blocks = 1;
2294         /* double indirect blocks */
2295         meta_blocks += 1 + (1LL << (bits-2));
2296         /* tripple indirect blocks */
2297         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2298
2299         upper_limit -= meta_blocks;
2300         upper_limit <<= bits;
2301
2302         res += 1LL << (bits-2);
2303         res += 1LL << (2*(bits-2));
2304         res += 1LL << (3*(bits-2));
2305         res <<= bits;
2306         if (res > upper_limit)
2307                 res = upper_limit;
2308
2309         if (res > MAX_LFS_FILESIZE)
2310                 res = MAX_LFS_FILESIZE;
2311
2312         return res;
2313 }
2314
2315 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2316                                    ext4_fsblk_t logical_sb_block, int nr)
2317 {
2318         struct ext4_sb_info *sbi = EXT4_SB(sb);
2319         ext4_group_t bg, first_meta_bg;
2320         int has_super = 0;
2321
2322         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2323
2324         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2325             nr < first_meta_bg)
2326                 return logical_sb_block + nr + 1;
2327         bg = sbi->s_desc_per_block * nr;
2328         if (ext4_bg_has_super(sb, bg))
2329                 has_super = 1;
2330
2331         return (has_super + ext4_group_first_block_no(sb, bg));
2332 }
2333
2334 /**
2335  * ext4_get_stripe_size: Get the stripe size.
2336  * @sbi: In memory super block info
2337  *
2338  * If we have specified it via mount option, then
2339  * use the mount option value. If the value specified at mount time is
2340  * greater than the blocks per group use the super block value.
2341  * If the super block value is greater than blocks per group return 0.
2342  * Allocator needs it be less than blocks per group.
2343  *
2344  */
2345 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2346 {
2347         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2348         unsigned long stripe_width =
2349                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2350         int ret;
2351
2352         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2353                 ret = sbi->s_stripe;
2354         else if (stripe_width <= sbi->s_blocks_per_group)
2355                 ret = stripe_width;
2356         else if (stride <= sbi->s_blocks_per_group)
2357                 ret = stride;
2358         else
2359                 ret = 0;
2360
2361         /*
2362          * If the stripe width is 1, this makes no sense and
2363          * we set it to 0 to turn off stripe handling code.
2364          */
2365         if (ret <= 1)
2366                 ret = 0;
2367
2368         return ret;
2369 }
2370
2371 /* sysfs supprt */
2372
2373 struct ext4_attr {
2374         struct attribute attr;
2375         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2376         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2377                          const char *, size_t);
2378         int offset;
2379 };
2380
2381 static int parse_strtoull(const char *buf,
2382                 unsigned long long max, unsigned long long *value)
2383 {
2384         int ret;
2385
2386         ret = kstrtoull(skip_spaces(buf), 0, value);
2387         if (!ret && *value > max)
2388                 ret = -EINVAL;
2389         return ret;
2390 }
2391
2392 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2393                                               struct ext4_sb_info *sbi,
2394                                               char *buf)
2395 {
2396         return snprintf(buf, PAGE_SIZE, "%llu\n",
2397                 (s64) EXT4_C2B(sbi,
2398                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2399 }
2400
2401 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2402                                          struct ext4_sb_info *sbi, char *buf)
2403 {
2404         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2405
2406         if (!sb->s_bdev->bd_part)
2407                 return snprintf(buf, PAGE_SIZE, "0\n");
2408         return snprintf(buf, PAGE_SIZE, "%lu\n",
2409                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2410                          sbi->s_sectors_written_start) >> 1);
2411 }
2412
2413 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2414                                           struct ext4_sb_info *sbi, char *buf)
2415 {
2416         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2417
2418         if (!sb->s_bdev->bd_part)
2419                 return snprintf(buf, PAGE_SIZE, "0\n");
2420         return snprintf(buf, PAGE_SIZE, "%llu\n",
2421                         (unsigned long long)(sbi->s_kbytes_written +
2422                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2423                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2424 }
2425
2426 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2427                                           struct ext4_sb_info *sbi,
2428                                           const char *buf, size_t count)
2429 {
2430         unsigned long t;
2431         int ret;
2432
2433         ret = kstrtoul(skip_spaces(buf), 0, &t);
2434         if (ret)
2435                 return ret;
2436
2437         if (t && (!is_power_of_2(t) || t > 0x40000000))
2438                 return -EINVAL;
2439
2440         sbi->s_inode_readahead_blks = t;
2441         return count;
2442 }
2443
2444 static ssize_t sbi_ui_show(struct ext4_attr *a,
2445                            struct ext4_sb_info *sbi, char *buf)
2446 {
2447         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2448
2449         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2450 }
2451
2452 static ssize_t sbi_ui_store(struct ext4_attr *a,
2453                             struct ext4_sb_info *sbi,
2454                             const char *buf, size_t count)
2455 {
2456         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2457         unsigned long t;
2458         int ret;
2459
2460         ret = kstrtoul(skip_spaces(buf), 0, &t);
2461         if (ret)
2462                 return ret;
2463         *ui = t;
2464         return count;
2465 }
2466
2467 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2468                                   struct ext4_sb_info *sbi, char *buf)
2469 {
2470         return snprintf(buf, PAGE_SIZE, "%llu\n",
2471                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2472 }
2473
2474 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2475                                    struct ext4_sb_info *sbi,
2476                                    const char *buf, size_t count)
2477 {
2478         unsigned long long val;
2479         int ret;
2480
2481         if (parse_strtoull(buf, -1ULL, &val))
2482                 return -EINVAL;
2483         ret = ext4_reserve_clusters(sbi, val);
2484
2485         return ret ? ret : count;
2486 }
2487
2488 static ssize_t trigger_test_error(struct ext4_attr *a,
2489                                   struct ext4_sb_info *sbi,
2490                                   const char *buf, size_t count)
2491 {
2492         int len = count;
2493
2494         if (!capable(CAP_SYS_ADMIN))
2495                 return -EPERM;
2496
2497         if (len && buf[len-1] == '\n')
2498                 len--;
2499
2500         if (len)
2501                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2502         return count;
2503 }
2504
2505 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2506 static struct ext4_attr ext4_attr_##_name = {                   \
2507         .attr = {.name = __stringify(_name), .mode = _mode },   \
2508         .show   = _show,                                        \
2509         .store  = _store,                                       \
2510         .offset = offsetof(struct ext4_sb_info, _elname),       \
2511 }
2512 #define EXT4_ATTR(name, mode, show, store) \
2513 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2514
2515 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2516 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2517 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2518 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2519         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2520 #define ATTR_LIST(name) &ext4_attr_##name.attr
2521
2522 EXT4_RO_ATTR(delayed_allocation_blocks);
2523 EXT4_RO_ATTR(session_write_kbytes);
2524 EXT4_RO_ATTR(lifetime_write_kbytes);
2525 EXT4_RW_ATTR(reserved_clusters);
2526 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2527                  inode_readahead_blks_store, s_inode_readahead_blks);
2528 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2529 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2530 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2531 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2532 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2533 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2534 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2535 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2536 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2537 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2538
2539 static struct attribute *ext4_attrs[] = {
2540         ATTR_LIST(delayed_allocation_blocks),
2541         ATTR_LIST(session_write_kbytes),
2542         ATTR_LIST(lifetime_write_kbytes),
2543         ATTR_LIST(reserved_clusters),
2544         ATTR_LIST(inode_readahead_blks),
2545         ATTR_LIST(inode_goal),
2546         ATTR_LIST(mb_stats),
2547         ATTR_LIST(mb_max_to_scan),
2548         ATTR_LIST(mb_min_to_scan),
2549         ATTR_LIST(mb_order2_req),
2550         ATTR_LIST(mb_stream_req),
2551         ATTR_LIST(mb_group_prealloc),
2552         ATTR_LIST(max_writeback_mb_bump),
2553         ATTR_LIST(extent_max_zeroout_kb),
2554         ATTR_LIST(trigger_fs_error),
2555         NULL,
2556 };
2557
2558 /* Features this copy of ext4 supports */
2559 EXT4_INFO_ATTR(lazy_itable_init);
2560 EXT4_INFO_ATTR(batched_discard);
2561 EXT4_INFO_ATTR(meta_bg_resize);
2562
2563 static struct attribute *ext4_feat_attrs[] = {
2564         ATTR_LIST(lazy_itable_init),
2565         ATTR_LIST(batched_discard),
2566         ATTR_LIST(meta_bg_resize),
2567         NULL,
2568 };
2569
2570 static ssize_t ext4_attr_show(struct kobject *kobj,
2571                               struct attribute *attr, char *buf)
2572 {
2573         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2574                                                 s_kobj);
2575         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2576
2577         return a->show ? a->show(a, sbi, buf) : 0;
2578 }
2579
2580 static ssize_t ext4_attr_store(struct kobject *kobj,
2581                                struct attribute *attr,
2582                                const char *buf, size_t len)
2583 {
2584         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2585                                                 s_kobj);
2586         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2587
2588         return a->store ? a->store(a, sbi, buf, len) : 0;
2589 }
2590
2591 static void ext4_sb_release(struct kobject *kobj)
2592 {
2593         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2594                                                 s_kobj);
2595         complete(&sbi->s_kobj_unregister);
2596 }
2597
2598 static const struct sysfs_ops ext4_attr_ops = {
2599         .show   = ext4_attr_show,
2600         .store  = ext4_attr_store,
2601 };
2602
2603 static struct kobj_type ext4_ktype = {
2604         .default_attrs  = ext4_attrs,
2605         .sysfs_ops      = &ext4_attr_ops,
2606         .release        = ext4_sb_release,
2607 };
2608
2609 static void ext4_feat_release(struct kobject *kobj)
2610 {
2611         complete(&ext4_feat->f_kobj_unregister);
2612 }
2613
2614 static struct kobj_type ext4_feat_ktype = {
2615         .default_attrs  = ext4_feat_attrs,
2616         .sysfs_ops      = &ext4_attr_ops,
2617         .release        = ext4_feat_release,
2618 };
2619
2620 /*
2621  * Check whether this filesystem can be mounted based on
2622  * the features present and the RDONLY/RDWR mount requested.
2623  * Returns 1 if this filesystem can be mounted as requested,
2624  * 0 if it cannot be.
2625  */
2626 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2627 {
2628         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2629                 ext4_msg(sb, KERN_ERR,
2630                         "Couldn't mount because of "
2631                         "unsupported optional features (%x)",
2632                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2633                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2634                 return 0;
2635         }
2636
2637         if (readonly)
2638                 return 1;
2639
2640         /* Check that feature set is OK for a read-write mount */
2641         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2642                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2643                          "unsupported optional features (%x)",
2644                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2645                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2646                 return 0;
2647         }
2648         /*
2649          * Large file size enabled file system can only be mounted
2650          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2651          */
2652         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2653                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2654                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2655                                  "cannot be mounted RDWR without "
2656                                  "CONFIG_LBDAF");
2657                         return 0;
2658                 }
2659         }
2660         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2661             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2662                 ext4_msg(sb, KERN_ERR,
2663                          "Can't support bigalloc feature without "
2664                          "extents feature\n");
2665                 return 0;
2666         }
2667
2668 #ifndef CONFIG_QUOTA
2669         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2670             !readonly) {
2671                 ext4_msg(sb, KERN_ERR,
2672                          "Filesystem with quota feature cannot be mounted RDWR "
2673                          "without CONFIG_QUOTA");
2674                 return 0;
2675         }
2676 #endif  /* CONFIG_QUOTA */
2677         return 1;
2678 }
2679
2680 /*
2681  * This function is called once a day if we have errors logged
2682  * on the file system
2683  */
2684 static void print_daily_error_info(unsigned long arg)
2685 {
2686         struct super_block *sb = (struct super_block *) arg;
2687         struct ext4_sb_info *sbi;
2688         struct ext4_super_block *es;
2689
2690         sbi = EXT4_SB(sb);
2691         es = sbi->s_es;
2692
2693         if (es->s_error_count)
2694                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2695                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2696                          le32_to_cpu(es->s_error_count));
2697         if (es->s_first_error_time) {
2698                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2699                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2700                        (int) sizeof(es->s_first_error_func),
2701                        es->s_first_error_func,
2702                        le32_to_cpu(es->s_first_error_line));
2703                 if (es->s_first_error_ino)
2704                         printk(": inode %u",
2705                                le32_to_cpu(es->s_first_error_ino));
2706                 if (es->s_first_error_block)
2707                         printk(": block %llu", (unsigned long long)
2708                                le64_to_cpu(es->s_first_error_block));
2709                 printk("\n");
2710         }
2711         if (es->s_last_error_time) {
2712                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2713                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2714                        (int) sizeof(es->s_last_error_func),
2715                        es->s_last_error_func,
2716                        le32_to_cpu(es->s_last_error_line));
2717                 if (es->s_last_error_ino)
2718                         printk(": inode %u",
2719                                le32_to_cpu(es->s_last_error_ino));
2720                 if (es->s_last_error_block)
2721                         printk(": block %llu", (unsigned long long)
2722                                le64_to_cpu(es->s_last_error_block));
2723                 printk("\n");
2724         }
2725         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2726 }
2727
2728 /* Find next suitable group and run ext4_init_inode_table */
2729 static int ext4_run_li_request(struct ext4_li_request *elr)
2730 {
2731         struct ext4_group_desc *gdp = NULL;
2732         ext4_group_t group, ngroups;
2733         struct super_block *sb;
2734         unsigned long timeout = 0;
2735         int ret = 0;
2736
2737         sb = elr->lr_super;
2738         ngroups = EXT4_SB(sb)->s_groups_count;
2739
2740         sb_start_write(sb);
2741         for (group = elr->lr_next_group; group < ngroups; group++) {
2742                 gdp = ext4_get_group_desc(sb, group, NULL);
2743                 if (!gdp) {
2744                         ret = 1;
2745                         break;
2746                 }
2747
2748                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2749                         break;
2750         }
2751
2752         if (group >= ngroups)
2753                 ret = 1;
2754
2755         if (!ret) {
2756                 timeout = jiffies;
2757                 ret = ext4_init_inode_table(sb, group,
2758                                             elr->lr_timeout ? 0 : 1);
2759                 if (elr->lr_timeout == 0) {
2760                         timeout = (jiffies - timeout) *
2761                                   elr->lr_sbi->s_li_wait_mult;
2762                         elr->lr_timeout = timeout;
2763                 }
2764                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2765                 elr->lr_next_group = group + 1;
2766         }
2767         sb_end_write(sb);
2768
2769         return ret;
2770 }
2771
2772 /*
2773  * Remove lr_request from the list_request and free the
2774  * request structure. Should be called with li_list_mtx held
2775  */
2776 static void ext4_remove_li_request(struct ext4_li_request *elr)
2777 {
2778         struct ext4_sb_info *sbi;
2779
2780         if (!elr)
2781                 return;
2782
2783         sbi = elr->lr_sbi;
2784
2785         list_del(&elr->lr_request);
2786         sbi->s_li_request = NULL;
2787         kfree(elr);
2788 }
2789
2790 static void ext4_unregister_li_request(struct super_block *sb)
2791 {
2792         mutex_lock(&ext4_li_mtx);
2793         if (!ext4_li_info) {
2794                 mutex_unlock(&ext4_li_mtx);
2795                 return;
2796         }
2797
2798         mutex_lock(&ext4_li_info->li_list_mtx);
2799         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2800         mutex_unlock(&ext4_li_info->li_list_mtx);
2801         mutex_unlock(&ext4_li_mtx);
2802 }
2803
2804 static struct task_struct *ext4_lazyinit_task;
2805
2806 /*
2807  * This is the function where ext4lazyinit thread lives. It walks
2808  * through the request list searching for next scheduled filesystem.
2809  * When such a fs is found, run the lazy initialization request
2810  * (ext4_rn_li_request) and keep track of the time spend in this
2811  * function. Based on that time we compute next schedule time of
2812  * the request. When walking through the list is complete, compute
2813  * next waking time and put itself into sleep.
2814  */
2815 static int ext4_lazyinit_thread(void *arg)
2816 {
2817         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2818         struct list_head *pos, *n;
2819         struct ext4_li_request *elr;
2820         unsigned long next_wakeup, cur;
2821
2822         BUG_ON(NULL == eli);
2823
2824 cont_thread:
2825         while (true) {
2826                 next_wakeup = MAX_JIFFY_OFFSET;
2827
2828                 mutex_lock(&eli->li_list_mtx);
2829                 if (list_empty(&eli->li_request_list)) {
2830                         mutex_unlock(&eli->li_list_mtx);
2831                         goto exit_thread;
2832                 }
2833
2834                 list_for_each_safe(pos, n, &eli->li_request_list) {
2835                         elr = list_entry(pos, struct ext4_li_request,
2836                                          lr_request);
2837
2838                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2839                                 if (ext4_run_li_request(elr) != 0) {
2840                                         /* error, remove the lazy_init job */
2841                                         ext4_remove_li_request(elr);
2842                                         continue;
2843                                 }
2844                         }
2845
2846                         if (time_before(elr->lr_next_sched, next_wakeup))
2847                                 next_wakeup = elr->lr_next_sched;
2848                 }
2849                 mutex_unlock(&eli->li_list_mtx);
2850
2851                 try_to_freeze();
2852
2853                 cur = jiffies;
2854                 if ((time_after_eq(cur, next_wakeup)) ||
2855                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2856                         cond_resched();
2857                         continue;
2858                 }
2859
2860                 schedule_timeout_interruptible(next_wakeup - cur);
2861
2862                 if (kthread_should_stop()) {
2863                         ext4_clear_request_list();
2864                         goto exit_thread;
2865                 }
2866         }
2867
2868 exit_thread:
2869         /*
2870          * It looks like the request list is empty, but we need
2871          * to check it under the li_list_mtx lock, to prevent any
2872          * additions into it, and of course we should lock ext4_li_mtx
2873          * to atomically free the list and ext4_li_info, because at
2874          * this point another ext4 filesystem could be registering
2875          * new one.
2876          */
2877         mutex_lock(&ext4_li_mtx);
2878         mutex_lock(&eli->li_list_mtx);
2879         if (!list_empty(&eli->li_request_list)) {
2880                 mutex_unlock(&eli->li_list_mtx);
2881                 mutex_unlock(&ext4_li_mtx);
2882                 goto cont_thread;
2883         }
2884         mutex_unlock(&eli->li_list_mtx);
2885         kfree(ext4_li_info);
2886         ext4_li_info = NULL;
2887         mutex_unlock(&ext4_li_mtx);
2888
2889         return 0;
2890 }
2891
2892 static void ext4_clear_request_list(void)
2893 {
2894         struct list_head *pos, *n;
2895         struct ext4_li_request *elr;
2896
2897         mutex_lock(&ext4_li_info->li_list_mtx);
2898         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2899                 elr = list_entry(pos, struct ext4_li_request,
2900                                  lr_request);
2901                 ext4_remove_li_request(elr);
2902         }
2903         mutex_unlock(&ext4_li_info->li_list_mtx);
2904 }
2905
2906 static int ext4_run_lazyinit_thread(void)
2907 {
2908         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2909                                          ext4_li_info, "ext4lazyinit");
2910         if (IS_ERR(ext4_lazyinit_task)) {
2911                 int err = PTR_ERR(ext4_lazyinit_task);
2912                 ext4_clear_request_list();
2913                 kfree(ext4_li_info);
2914                 ext4_li_info = NULL;
2915                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2916                                  "initialization thread\n",
2917                                  err);
2918                 return err;
2919         }
2920         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2921         return 0;
2922 }
2923
2924 /*
2925  * Check whether it make sense to run itable init. thread or not.
2926  * If there is at least one uninitialized inode table, return
2927  * corresponding group number, else the loop goes through all
2928  * groups and return total number of groups.
2929  */
2930 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2931 {
2932         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2933         struct ext4_group_desc *gdp = NULL;
2934
2935         for (group = 0; group < ngroups; group++) {
2936                 gdp = ext4_get_group_desc(sb, group, NULL);
2937                 if (!gdp)
2938                         continue;
2939
2940                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2941                         break;
2942         }
2943
2944         return group;
2945 }
2946
2947 static int ext4_li_info_new(void)
2948 {
2949         struct ext4_lazy_init *eli = NULL;
2950
2951         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2952         if (!eli)
2953                 return -ENOMEM;
2954
2955         INIT_LIST_HEAD(&eli->li_request_list);
2956         mutex_init(&eli->li_list_mtx);
2957
2958         eli->li_state |= EXT4_LAZYINIT_QUIT;
2959
2960         ext4_li_info = eli;
2961
2962         return 0;
2963 }
2964
2965 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2966                                             ext4_group_t start)
2967 {
2968         struct ext4_sb_info *sbi = EXT4_SB(sb);
2969         struct ext4_li_request *elr;
2970         unsigned long rnd;
2971
2972         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2973         if (!elr)
2974                 return NULL;
2975
2976         elr->lr_super = sb;
2977         elr->lr_sbi = sbi;
2978         elr->lr_next_group = start;
2979
2980         /*
2981          * Randomize first schedule time of the request to
2982          * spread the inode table initialization requests
2983          * better.
2984          */
2985         get_random_bytes(&rnd, sizeof(rnd));
2986         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2987                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2988
2989         return elr;
2990 }
2991
2992 int ext4_register_li_request(struct super_block *sb,
2993                              ext4_group_t first_not_zeroed)
2994 {
2995         struct ext4_sb_info *sbi = EXT4_SB(sb);
2996         struct ext4_li_request *elr = NULL;
2997         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2998         int ret = 0;
2999
3000         mutex_lock(&ext4_li_mtx);
3001         if (sbi->s_li_request != NULL) {
3002                 /*
3003                  * Reset timeout so it can be computed again, because
3004                  * s_li_wait_mult might have changed.
3005                  */
3006                 sbi->s_li_request->lr_timeout = 0;
3007                 goto out;
3008         }
3009
3010         if (first_not_zeroed == ngroups ||
3011             (sb->s_flags & MS_RDONLY) ||
3012             !test_opt(sb, INIT_INODE_TABLE))
3013                 goto out;
3014
3015         elr = ext4_li_request_new(sb, first_not_zeroed);
3016         if (!elr) {
3017                 ret = -ENOMEM;
3018                 goto out;
3019         }
3020
3021         if (NULL == ext4_li_info) {
3022                 ret = ext4_li_info_new();
3023                 if (ret)
3024                         goto out;
3025         }
3026
3027         mutex_lock(&ext4_li_info->li_list_mtx);
3028         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3029         mutex_unlock(&ext4_li_info->li_list_mtx);
3030
3031         sbi->s_li_request = elr;
3032         /*
3033          * set elr to NULL here since it has been inserted to
3034          * the request_list and the removal and free of it is
3035          * handled by ext4_clear_request_list from now on.
3036          */
3037         elr = NULL;
3038
3039         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3040                 ret = ext4_run_lazyinit_thread();
3041                 if (ret)
3042                         goto out;
3043         }
3044 out:
3045         mutex_unlock(&ext4_li_mtx);
3046         if (ret)
3047                 kfree(elr);
3048         return ret;
3049 }
3050
3051 /*
3052  * We do not need to lock anything since this is called on
3053  * module unload.
3054  */
3055 static void ext4_destroy_lazyinit_thread(void)
3056 {
3057         /*
3058          * If thread exited earlier
3059          * there's nothing to be done.
3060          */
3061         if (!ext4_li_info || !ext4_lazyinit_task)
3062                 return;
3063
3064         kthread_stop(ext4_lazyinit_task);
3065 }
3066
3067 static int set_journal_csum_feature_set(struct super_block *sb)
3068 {
3069         int ret = 1;
3070         int compat, incompat;
3071         struct ext4_sb_info *sbi = EXT4_SB(sb);
3072
3073         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3074                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3075                 /* journal checksum v2 */
3076                 compat = 0;
3077                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3078         } else {
3079                 /* journal checksum v1 */
3080                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3081                 incompat = 0;
3082         }
3083
3084         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3085                 ret = jbd2_journal_set_features(sbi->s_journal,
3086                                 compat, 0,
3087                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3088                                 incompat);
3089         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3090                 ret = jbd2_journal_set_features(sbi->s_journal,
3091                                 compat, 0,
3092                                 incompat);
3093                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3094                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3095         } else {
3096                 jbd2_journal_clear_features(sbi->s_journal,
3097                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3098                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3099                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3100         }
3101
3102         return ret;
3103 }
3104
3105 /*
3106  * Note: calculating the overhead so we can be compatible with
3107  * historical BSD practice is quite difficult in the face of
3108  * clusters/bigalloc.  This is because multiple metadata blocks from
3109  * different block group can end up in the same allocation cluster.
3110  * Calculating the exact overhead in the face of clustered allocation
3111  * requires either O(all block bitmaps) in memory or O(number of block
3112  * groups**2) in time.  We will still calculate the superblock for
3113  * older file systems --- and if we come across with a bigalloc file
3114  * system with zero in s_overhead_clusters the estimate will be close to
3115  * correct especially for very large cluster sizes --- but for newer
3116  * file systems, it's better to calculate this figure once at mkfs
3117  * time, and store it in the superblock.  If the superblock value is
3118  * present (even for non-bigalloc file systems), we will use it.
3119  */
3120 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3121                           char *buf)
3122 {
3123         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3124         struct ext4_group_desc  *gdp;
3125         ext4_fsblk_t            first_block, last_block, b;
3126         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3127         int                     s, j, count = 0;
3128
3129         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3130                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3131                         sbi->s_itb_per_group + 2);
3132
3133         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3134                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3135         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3136         for (i = 0; i < ngroups; i++) {
3137                 gdp = ext4_get_group_desc(sb, i, NULL);
3138                 b = ext4_block_bitmap(sb, gdp);
3139                 if (b >= first_block && b <= last_block) {
3140                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3141                         count++;
3142                 }
3143                 b = ext4_inode_bitmap(sb, gdp);
3144                 if (b >= first_block && b <= last_block) {
3145                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3146                         count++;
3147                 }
3148                 b = ext4_inode_table(sb, gdp);
3149                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3150                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3151                                 int c = EXT4_B2C(sbi, b - first_block);
3152                                 ext4_set_bit(c, buf);
3153                                 count++;
3154                         }
3155                 if (i != grp)
3156                         continue;
3157                 s = 0;
3158                 if (ext4_bg_has_super(sb, grp)) {
3159                         ext4_set_bit(s++, buf);
3160                         count++;
3161                 }
3162                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3163                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3164                         count++;
3165                 }
3166         }
3167         if (!count)
3168                 return 0;
3169         return EXT4_CLUSTERS_PER_GROUP(sb) -
3170                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3171 }
3172
3173 /*
3174  * Compute the overhead and stash it in sbi->s_overhead
3175  */
3176 int ext4_calculate_overhead(struct super_block *sb)
3177 {
3178         struct ext4_sb_info *sbi = EXT4_SB(sb);
3179         struct ext4_super_block *es = sbi->s_es;
3180         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3181         ext4_fsblk_t overhead = 0;
3182         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3183
3184         if (!buf)
3185                 return -ENOMEM;
3186
3187         /*
3188          * Compute the overhead (FS structures).  This is constant
3189          * for a given filesystem unless the number of block groups
3190          * changes so we cache the previous value until it does.
3191          */
3192
3193         /*
3194          * All of the blocks before first_data_block are overhead
3195          */
3196         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3197
3198         /*
3199          * Add the overhead found in each block group
3200          */
3201         for (i = 0; i < ngroups; i++) {
3202                 int blks;
3203
3204                 blks = count_overhead(sb, i, buf);
3205                 overhead += blks;
3206                 if (blks)
3207                         memset(buf, 0, PAGE_SIZE);
3208                 cond_resched();
3209         }
3210         /* Add the journal blocks as well */
3211         if (sbi->s_journal)
3212                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3213
3214         sbi->s_overhead = overhead;
3215         smp_wmb();
3216         free_page((unsigned long) buf);
3217         return 0;
3218 }
3219
3220
3221 static ext4_fsblk_t ext4_calculate_resv_clusters(struct super_block *sb)
3222 {
3223         ext4_fsblk_t resv_clusters;
3224
3225         /*
3226          * There's no need to reserve anything when we aren't using extents.
3227          * The space estimates are exact, there are no unwritten extents,
3228          * hole punching doesn't need new metadata... This is needed especially
3229          * to keep ext2/3 backward compatibility.
3230          */
3231         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3232                 return 0;
3233         /*
3234          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3235          * This should cover the situations where we can not afford to run
3236          * out of space like for example punch hole, or converting
3237          * uninitialized extents in delalloc path. In most cases such
3238          * allocation would require 1, or 2 blocks, higher numbers are
3239          * very rare.
3240          */
3241         resv_clusters = ext4_blocks_count(EXT4_SB(sb)->s_es) >>
3242                         EXT4_SB(sb)->s_cluster_bits;
3243
3244         do_div(resv_clusters, 50);
3245         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3246
3247         return resv_clusters;
3248 }
3249
3250
3251 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3252 {
3253         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3254                                 sbi->s_cluster_bits;
3255
3256         if (count >= clusters)
3257                 return -EINVAL;
3258
3259         atomic64_set(&sbi->s_resv_clusters, count);
3260         return 0;
3261 }
3262
3263 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3264 {
3265         char *orig_data = kstrdup(data, GFP_KERNEL);
3266         struct buffer_head *bh;
3267         struct ext4_super_block *es = NULL;
3268         struct ext4_sb_info *sbi;
3269         ext4_fsblk_t block;
3270         ext4_fsblk_t sb_block = get_sb_block(&data);
3271         ext4_fsblk_t logical_sb_block;
3272         unsigned long offset = 0;
3273         unsigned long journal_devnum = 0;
3274         unsigned long def_mount_opts;
3275         struct inode *root;
3276         char *cp;
3277         const char *descr;
3278         int ret = -ENOMEM;
3279         int blocksize, clustersize;
3280         unsigned int db_count;
3281         unsigned int i;
3282         int needs_recovery, has_huge_files, has_bigalloc;
3283         __u64 blocks_count;
3284         int err = 0;
3285         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3286         ext4_group_t first_not_zeroed;
3287
3288         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3289         if (!sbi)
3290                 goto out_free_orig;
3291
3292         sbi->s_blockgroup_lock =
3293                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3294         if (!sbi->s_blockgroup_lock) {
3295                 kfree(sbi);
3296                 goto out_free_orig;
3297         }
3298         sb->s_fs_info = sbi;
3299         sbi->s_sb = sb;
3300         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3301         sbi->s_sb_block = sb_block;
3302         if (sb->s_bdev->bd_part)
3303                 sbi->s_sectors_written_start =
3304                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3305
3306         /* Cleanup superblock name */
3307         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3308                 *cp = '!';
3309
3310         /* -EINVAL is default */
3311         ret = -EINVAL;
3312         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3313         if (!blocksize) {
3314                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3315                 goto out_fail;
3316         }
3317
3318         /*
3319          * The ext4 superblock will not be buffer aligned for other than 1kB
3320          * block sizes.  We need to calculate the offset from buffer start.
3321          */
3322         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3323                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3324                 offset = do_div(logical_sb_block, blocksize);
3325         } else {
3326                 logical_sb_block = sb_block;
3327         }
3328
3329         if (!(bh = sb_bread(sb, logical_sb_block))) {
3330                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3331                 goto out_fail;
3332         }
3333         /*
3334          * Note: s_es must be initialized as soon as possible because
3335          *       some ext4 macro-instructions depend on its value
3336          */
3337         es = (struct ext4_super_block *) (bh->b_data + offset);
3338         sbi->s_es = es;
3339         sb->s_magic = le16_to_cpu(es->s_magic);
3340         if (sb->s_magic != EXT4_SUPER_MAGIC)
3341                 goto cantfind_ext4;
3342         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3343
3344         /* Warn if metadata_csum and gdt_csum are both set. */
3345         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3346                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3347             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3348                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3349                              "redundant flags; please run fsck.");
3350
3351         /* Check for a known checksum algorithm */
3352         if (!ext4_verify_csum_type(sb, es)) {
3353                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3354                          "unknown checksum algorithm.");
3355                 silent = 1;
3356                 goto cantfind_ext4;
3357         }
3358
3359         /* Load the checksum driver */
3360         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3361                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3362                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3363                 if (IS_ERR(sbi->s_chksum_driver)) {
3364                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3365                         ret = PTR_ERR(sbi->s_chksum_driver);
3366                         sbi->s_chksum_driver = NULL;
3367                         goto failed_mount;
3368                 }
3369         }
3370
3371         /* Check superblock checksum */
3372         if (!ext4_superblock_csum_verify(sb, es)) {
3373                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3374                          "invalid superblock checksum.  Run e2fsck?");
3375                 silent = 1;
3376                 goto cantfind_ext4;
3377         }
3378
3379         /* Precompute checksum seed for all metadata */
3380         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3381                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3382                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3383                                                sizeof(es->s_uuid));
3384
3385         /* Set defaults before we parse the mount options */
3386         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3387         set_opt(sb, INIT_INODE_TABLE);
3388         if (def_mount_opts & EXT4_DEFM_DEBUG)
3389                 set_opt(sb, DEBUG);
3390         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3391                 set_opt(sb, GRPID);
3392         if (def_mount_opts & EXT4_DEFM_UID16)
3393                 set_opt(sb, NO_UID32);
3394         /* xattr user namespace & acls are now defaulted on */
3395         set_opt(sb, XATTR_USER);
3396 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3397         set_opt(sb, POSIX_ACL);
3398 #endif
3399         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3400                 set_opt(sb, JOURNAL_DATA);
3401         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3402                 set_opt(sb, ORDERED_DATA);
3403         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3404                 set_opt(sb, WRITEBACK_DATA);
3405
3406         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3407                 set_opt(sb, ERRORS_PANIC);
3408         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3409                 set_opt(sb, ERRORS_CONT);
3410         else
3411                 set_opt(sb, ERRORS_RO);
3412         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3413                 set_opt(sb, BLOCK_VALIDITY);
3414         if (def_mount_opts & EXT4_DEFM_DISCARD)
3415                 set_opt(sb, DISCARD);
3416
3417         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3418         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3419         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3420         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3421         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3422
3423         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3424                 set_opt(sb, BARRIER);
3425
3426         /*
3427          * enable delayed allocation by default
3428          * Use -o nodelalloc to turn it off
3429          */
3430         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3431             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3432                 set_opt(sb, DELALLOC);
3433
3434         /*
3435          * set default s_li_wait_mult for lazyinit, for the case there is
3436          * no mount option specified.
3437          */
3438         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3439
3440         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3441                            &journal_devnum, &journal_ioprio, 0)) {
3442                 ext4_msg(sb, KERN_WARNING,
3443                          "failed to parse options in superblock: %s",
3444                          sbi->s_es->s_mount_opts);
3445         }
3446         sbi->s_def_mount_opt = sbi->s_mount_opt;
3447         if (!parse_options((char *) data, sb, &journal_devnum,
3448                            &journal_ioprio, 0))
3449                 goto failed_mount;
3450
3451         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3452                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3453                             "with data=journal disables delayed "
3454                             "allocation and O_DIRECT support!\n");
3455                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3456                         ext4_msg(sb, KERN_ERR, "can't mount with "
3457                                  "both data=journal and delalloc");
3458                         goto failed_mount;
3459                 }
3460                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3461                         ext4_msg(sb, KERN_ERR, "can't mount with "
3462                                  "both data=journal and dioread_nolock");
3463                         goto failed_mount;
3464                 }
3465                 if (test_opt(sb, DELALLOC))
3466                         clear_opt(sb, DELALLOC);
3467         }
3468
3469         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3470                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3471
3472         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3473             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3474              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3475              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3476                 ext4_msg(sb, KERN_WARNING,
3477                        "feature flags set on rev 0 fs, "
3478                        "running e2fsck is recommended");
3479
3480         if (IS_EXT2_SB(sb)) {
3481                 if (ext2_feature_set_ok(sb))
3482                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3483                                  "using the ext4 subsystem");
3484                 else {
3485                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3486                                  "to feature incompatibilities");
3487                         goto failed_mount;
3488                 }
3489         }
3490
3491         if (IS_EXT3_SB(sb)) {
3492                 if (ext3_feature_set_ok(sb))
3493                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3494                                  "using the ext4 subsystem");
3495                 else {
3496                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3497                                  "to feature incompatibilities");
3498                         goto failed_mount;
3499                 }
3500         }
3501
3502         /*
3503          * Check feature flags regardless of the revision level, since we
3504          * previously didn't change the revision level when setting the flags,
3505          * so there is a chance incompat flags are set on a rev 0 filesystem.
3506          */
3507         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3508                 goto failed_mount;
3509
3510         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3511         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3512             blocksize > EXT4_MAX_BLOCK_SIZE) {
3513                 ext4_msg(sb, KERN_ERR,
3514                        "Unsupported filesystem blocksize %d", blocksize);
3515                 goto failed_mount;
3516         }
3517
3518         if (sb->s_blocksize != blocksize) {
3519                 /* Validate the filesystem blocksize */
3520                 if (!sb_set_blocksize(sb, blocksize)) {
3521                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3522                                         blocksize);
3523                         goto failed_mount;
3524                 }
3525
3526                 brelse(bh);
3527                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3528                 offset = do_div(logical_sb_block, blocksize);
3529                 bh = sb_bread(sb, logical_sb_block);
3530                 if (!bh) {
3531                         ext4_msg(sb, KERN_ERR,
3532                                "Can't read superblock on 2nd try");
3533                         goto failed_mount;
3534                 }
3535                 es = (struct ext4_super_block *)(bh->b_data + offset);
3536                 sbi->s_es = es;
3537                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3538                         ext4_msg(sb, KERN_ERR,
3539                                "Magic mismatch, very weird!");
3540                         goto failed_mount;
3541                 }
3542         }
3543
3544         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3545                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3546         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3547                                                       has_huge_files);
3548         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3549
3550         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3551                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3552                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3553         } else {
3554                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3555                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3556                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3557                     (!is_power_of_2(sbi->s_inode_size)) ||
3558                     (sbi->s_inode_size > blocksize)) {
3559                         ext4_msg(sb, KERN_ERR,
3560                                "unsupported inode size: %d",
3561                                sbi->s_inode_size);
3562                         goto failed_mount;
3563                 }
3564                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3565                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3566         }
3567
3568         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3569         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3570                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3571                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3572                     !is_power_of_2(sbi->s_desc_size)) {
3573                         ext4_msg(sb, KERN_ERR,
3574                                "unsupported descriptor size %lu",
3575                                sbi->s_desc_size);
3576                         goto failed_mount;
3577                 }
3578         } else
3579                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3580
3581         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3582         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3583         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3584                 goto cantfind_ext4;
3585
3586         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3587         if (sbi->s_inodes_per_block == 0)
3588                 goto cantfind_ext4;
3589         sbi->s_itb_per_group = DIV_ROUND_UP(sbi->s_inodes_per_group,
3590                                         sbi->s_inodes_per_block);
3591         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3592         sbi->s_sbh = bh;
3593         sbi->s_mount_state = le16_to_cpu(es->s_state);
3594         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3595         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3596
3597         for (i = 0; i < 4; i++)
3598                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3599         sbi->s_def_hash_version = es->s_def_hash_version;
3600         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
3601                 i = le32_to_cpu(es->s_flags);
3602                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3603                         sbi->s_hash_unsigned = 3;
3604                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3605 #ifdef __CHAR_UNSIGNED__
3606                         if (!(sb->s_flags & MS_RDONLY))
3607                                 es->s_flags |=
3608                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3609                         sbi->s_hash_unsigned = 3;
3610 #else
3611                         if (!(sb->s_flags & MS_RDONLY))
3612                                 es->s_flags |=
3613                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3614 #endif
3615                 }
3616         }
3617
3618         /* Handle clustersize */
3619         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3620         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3621                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3622         if (has_bigalloc) {
3623                 if (clustersize < blocksize) {
3624                         ext4_msg(sb, KERN_ERR,
3625                                  "cluster size (%d) smaller than "
3626                                  "block size (%d)", clustersize, blocksize);
3627                         goto failed_mount;
3628                 }
3629                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3630                         le32_to_cpu(es->s_log_block_size);
3631                 sbi->s_clusters_per_group =
3632                         le32_to_cpu(es->s_clusters_per_group);
3633                 if (sbi->s_clusters_per_group > blocksize * 8) {
3634                         ext4_msg(sb, KERN_ERR,
3635                                  "#clusters per group too big: %lu",
3636                                  sbi->s_clusters_per_group);
3637                         goto failed_mount;
3638                 }
3639                 if (sbi->s_blocks_per_group !=
3640                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3641                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3642                                  "clusters per group (%lu) inconsistent",
3643                                  sbi->s_blocks_per_group,
3644                                  sbi->s_clusters_per_group);
3645                         goto failed_mount;
3646                 }
3647         } else {
3648                 if (clustersize != blocksize) {
3649                         ext4_warning(sb, "fragment/cluster size (%d) != "
3650                                      "block size (%d)", clustersize,
3651                                      blocksize);
3652                         clustersize = blocksize;
3653                 }
3654                 if (sbi->s_blocks_per_group > blocksize * 8) {
3655                         ext4_msg(sb, KERN_ERR,
3656                                  "#blocks per group too big: %lu",
3657                                  sbi->s_blocks_per_group);
3658                         goto failed_mount;
3659                 }
3660                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3661                 sbi->s_cluster_bits = 0;
3662         }
3663         sbi->s_cluster_ratio = clustersize / blocksize;
3664
3665         if (sbi->s_inodes_per_group > blocksize * 8) {
3666                 ext4_msg(sb, KERN_ERR,
3667                        "#inodes per group too big: %lu",
3668                        sbi->s_inodes_per_group);
3669                 goto failed_mount;
3670         }
3671
3672         /* Do we have standard group size of clustersize * 8 blocks ? */
3673         if (sbi->s_blocks_per_group == clustersize << 3)
3674                 set_opt2(sb, STD_GROUP_SIZE);
3675
3676         /*
3677          * Test whether we have more sectors than will fit in sector_t,
3678          * and whether the max offset is addressable by the page cache.
3679          */
3680         err = generic_check_addressable(sb->s_blocksize_bits,
3681                                         ext4_blocks_count(es));
3682         if (err) {
3683                 ext4_msg(sb, KERN_ERR, "filesystem"
3684                          " too large to mount safely on this system");
3685                 if (sizeof(sector_t) < 8)
3686                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3687                 goto failed_mount;
3688         }
3689
3690         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3691                 goto cantfind_ext4;
3692
3693         /* check blocks count against device size */
3694         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3695         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3696                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3697                        "exceeds size of device (%llu blocks)",
3698                        ext4_blocks_count(es), blocks_count);
3699                 goto failed_mount;
3700         }
3701
3702         /*
3703          * It makes no sense for the first data block to be beyond the end
3704          * of the filesystem.
3705          */
3706         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3707                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3708                          "block %u is beyond end of filesystem (%llu)",
3709                          le32_to_cpu(es->s_first_data_block),
3710                          ext4_blocks_count(es));
3711                 goto failed_mount;
3712         }
3713         blocks_count = (ext4_blocks_count(es) -
3714                         le32_to_cpu(es->s_first_data_block) +
3715                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3716         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3717         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3718                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3719                        "(block count %llu, first data block %u, "
3720                        "blocks per group %lu)", sbi->s_groups_count,
3721                        ext4_blocks_count(es),
3722                        le32_to_cpu(es->s_first_data_block),
3723                        EXT4_BLOCKS_PER_GROUP(sb));
3724                 goto failed_mount;
3725         }
3726         sbi->s_groups_count = blocks_count;
3727         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3728                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3729         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3730                    EXT4_DESC_PER_BLOCK(sb);
3731         sbi->s_group_desc = ext4_kvmalloc(db_count *
3732                                           sizeof(struct buffer_head *),
3733                                           GFP_KERNEL);
3734         if (sbi->s_group_desc == NULL) {
3735                 ext4_msg(sb, KERN_ERR, "not enough memory");
3736                 ret = -ENOMEM;
3737                 goto failed_mount;
3738         }
3739
3740         if (ext4_proc_root)
3741                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3742
3743         if (sbi->s_proc)
3744                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3745                                  &ext4_seq_options_fops, sb);
3746
3747         bgl_lock_init(sbi->s_blockgroup_lock);
3748
3749         for (i = 0; i < db_count; i++) {
3750                 block = descriptor_loc(sb, logical_sb_block, i);
3751                 sbi->s_group_desc[i] = sb_bread(sb, block);
3752                 if (!sbi->s_group_desc[i]) {
3753                         ext4_msg(sb, KERN_ERR,
3754                                "can't read group descriptor %d", i);
3755                         db_count = i;
3756                         goto failed_mount2;
3757                 }
3758         }
3759         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3760                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3761                 goto failed_mount2;
3762         }
3763         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3764                 if (!ext4_fill_flex_info(sb)) {
3765                         ext4_msg(sb, KERN_ERR,
3766                                "unable to initialize "
3767                                "flex_bg meta info!");
3768                         goto failed_mount2;
3769                 }
3770
3771         sbi->s_gdb_count = db_count;
3772         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3773         spin_lock_init(&sbi->s_next_gen_lock);
3774
3775         init_timer(&sbi->s_err_report);
3776         sbi->s_err_report.function = print_daily_error_info;
3777         sbi->s_err_report.data = (unsigned long) sb;
3778
3779         /* Register extent status tree shrinker */
3780         ext4_es_register_shrinker(sb);
3781
3782         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3783                         ext4_count_free_clusters(sb));
3784         if (!err) {
3785                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3786                                 ext4_count_free_inodes(sb));
3787         }
3788         if (!err) {
3789                 err = percpu_counter_init(&sbi->s_dirs_counter,
3790                                 ext4_count_dirs(sb));
3791         }
3792         if (!err) {
3793                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3794         }
3795         if (!err) {
3796                 err = percpu_counter_init(&sbi->s_extent_cache_cnt, 0);
3797         }
3798         if (err) {
3799                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3800                 goto failed_mount3;
3801         }
3802
3803         sbi->s_stripe = ext4_get_stripe_size(sbi);
3804         sbi->s_max_writeback_mb_bump = 128;
3805         sbi->s_extent_max_zeroout_kb = 32;
3806
3807         /*
3808          * set up enough so that it can read an inode
3809          */
3810         if (!test_opt(sb, NOLOAD) &&
3811             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3812                 sb->s_op = &ext4_sops;
3813         else
3814                 sb->s_op = &ext4_nojournal_sops;
3815         sb->s_export_op = &ext4_export_ops;
3816         sb->s_xattr = ext4_xattr_handlers;
3817 #ifdef CONFIG_QUOTA
3818         sb->dq_op = &ext4_quota_operations;
3819         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3820                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3821         else
3822                 sb->s_qcop = &ext4_qctl_operations;
3823 #endif
3824         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3825
3826         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3827         mutex_init(&sbi->s_orphan_lock);
3828
3829         sb->s_root = NULL;
3830
3831         needs_recovery = (es->s_last_orphan != 0 ||
3832                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3833                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3834
3835         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3836             !(sb->s_flags & MS_RDONLY))
3837                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3838                         goto failed_mount3;
3839
3840         /*
3841          * The first inode we look at is the journal inode.  Don't try
3842          * root first: it may be modified in the journal!
3843          */
3844         if (!test_opt(sb, NOLOAD) &&
3845             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3846                 if (ext4_load_journal(sb, es, journal_devnum))
3847                         goto failed_mount3;
3848         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3849               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3850                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3851                        "suppressed and not mounted read-only");
3852                 goto failed_mount_wq;
3853         } else {
3854                 clear_opt(sb, DATA_FLAGS);
3855                 sbi->s_journal = NULL;
3856                 needs_recovery = 0;
3857                 goto no_journal;
3858         }
3859
3860         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3861             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3862                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3863                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3864                 goto failed_mount_wq;
3865         }
3866
3867         if (!set_journal_csum_feature_set(sb)) {
3868                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3869                          "feature set");
3870                 goto failed_mount_wq;
3871         }
3872
3873         /* We have now updated the journal if required, so we can
3874          * validate the data journaling mode. */
3875         switch (test_opt(sb, DATA_FLAGS)) {
3876         case 0:
3877                 /* No mode set, assume a default based on the journal
3878                  * capabilities: ORDERED_DATA if the journal can
3879                  * cope, else JOURNAL_DATA
3880                  */
3881                 if (jbd2_journal_check_available_features
3882                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3883                         set_opt(sb, ORDERED_DATA);
3884                 else
3885                         set_opt(sb, JOURNAL_DATA);
3886                 break;
3887
3888         case EXT4_MOUNT_ORDERED_DATA:
3889         case EXT4_MOUNT_WRITEBACK_DATA:
3890                 if (!jbd2_journal_check_available_features
3891                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3892                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3893                                "requested data journaling mode");
3894                         goto failed_mount_wq;
3895                 }
3896         default:
3897                 break;
3898         }
3899         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3900
3901         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3902
3903         /*
3904          * The journal may have updated the bg summary counts, so we
3905          * need to update the global counters.
3906          */
3907         percpu_counter_set(&sbi->s_freeclusters_counter,
3908                            ext4_count_free_clusters(sb));
3909         percpu_counter_set(&sbi->s_freeinodes_counter,
3910                            ext4_count_free_inodes(sb));
3911         percpu_counter_set(&sbi->s_dirs_counter,
3912                            ext4_count_dirs(sb));
3913         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3914
3915 no_journal:
3916         /*
3917          * Get the # of file system overhead blocks from the
3918          * superblock if present.
3919          */
3920         if (es->s_overhead_clusters)
3921                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3922         else {
3923                 err = ext4_calculate_overhead(sb);
3924                 if (err)
3925                         goto failed_mount_wq;
3926         }
3927
3928         /*
3929          * The maximum number of concurrent works can be high and
3930          * concurrency isn't really necessary.  Limit it to 1.
3931          */
3932         EXT4_SB(sb)->dio_unwritten_wq =
3933                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3934         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3935                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3936                 ret = -ENOMEM;
3937                 goto failed_mount_wq;
3938         }
3939
3940         /*
3941          * The jbd2_journal_load will have done any necessary log recovery,
3942          * so we can safely mount the rest of the filesystem now.
3943          */
3944
3945         root = ext4_iget(sb, EXT4_ROOT_INO);
3946         if (IS_ERR(root)) {
3947                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3948                 ret = PTR_ERR(root);
3949                 root = NULL;
3950                 goto failed_mount4;
3951         }
3952         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3953                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3954                 iput(root);
3955                 goto failed_mount4;
3956         }
3957         sb->s_root = d_make_root(root);
3958         if (!sb->s_root) {
3959                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3960                 ret = -ENOMEM;
3961                 goto failed_mount4;
3962         }
3963
3964         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3965                 sb->s_flags |= MS_RDONLY;
3966
3967         /* determine the minimum size of new large inodes, if present */
3968         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3969                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3970                                                      EXT4_GOOD_OLD_INODE_SIZE;
3971                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3972                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3973                         if (sbi->s_want_extra_isize <
3974                             le16_to_cpu(es->s_want_extra_isize))
3975                                 sbi->s_want_extra_isize =
3976                                         le16_to_cpu(es->s_want_extra_isize);
3977                         if (sbi->s_want_extra_isize <
3978                             le16_to_cpu(es->s_min_extra_isize))
3979                                 sbi->s_want_extra_isize =
3980                                         le16_to_cpu(es->s_min_extra_isize);
3981                 }
3982         }
3983         /* Check if enough inode space is available */
3984         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3985                                                         sbi->s_inode_size) {
3986                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3987                                                        EXT4_GOOD_OLD_INODE_SIZE;
3988                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3989                          "available");
3990         }
3991
3992         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sb));
3993         if (err) {
3994                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
3995                          "reserved pool", ext4_calculate_resv_clusters(sb));
3996                 goto failed_mount4a;
3997         }
3998
3999         err = ext4_setup_system_zone(sb);
4000         if (err) {
4001                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4002                          "zone (%d)", err);
4003                 goto failed_mount4a;
4004         }
4005
4006         ext4_ext_init(sb);
4007         err = ext4_mb_init(sb);
4008         if (err) {
4009                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4010                          err);
4011                 goto failed_mount5;
4012         }
4013
4014         err = ext4_register_li_request(sb, first_not_zeroed);
4015         if (err)
4016                 goto failed_mount6;
4017
4018         sbi->s_kobj.kset = ext4_kset;
4019         init_completion(&sbi->s_kobj_unregister);
4020         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4021                                    "%s", sb->s_id);
4022         if (err)
4023                 goto failed_mount7;
4024
4025 #ifdef CONFIG_QUOTA
4026         /* Enable quota usage during mount. */
4027         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4028             !(sb->s_flags & MS_RDONLY)) {
4029                 err = ext4_enable_quotas(sb);
4030                 if (err)
4031                         goto failed_mount8;
4032         }
4033 #endif  /* CONFIG_QUOTA */
4034
4035         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4036         ext4_orphan_cleanup(sb, es);
4037         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4038         if (needs_recovery) {
4039                 ext4_msg(sb, KERN_INFO, "recovery complete");
4040                 ext4_mark_recovery_complete(sb, es);
4041         }
4042         if (EXT4_SB(sb)->s_journal) {
4043                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4044                         descr = " journalled data mode";
4045                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4046                         descr = " ordered data mode";
4047                 else
4048                         descr = " writeback data mode";
4049         } else
4050                 descr = "out journal";
4051
4052         if (test_opt(sb, DISCARD)) {
4053                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4054                 if (!blk_queue_discard(q))
4055                         ext4_msg(sb, KERN_WARNING,
4056                                  "mounting with \"discard\" option, but "
4057                                  "the device does not support discard");
4058         }
4059
4060         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4061                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4062                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4063
4064         if (es->s_error_count)
4065                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4066
4067         kfree(orig_data);
4068         return 0;
4069
4070 cantfind_ext4:
4071         if (!silent)
4072                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4073         goto failed_mount;
4074
4075 #ifdef CONFIG_QUOTA
4076 failed_mount8:
4077         kobject_del(&sbi->s_kobj);
4078 #endif
4079 failed_mount7:
4080         ext4_unregister_li_request(sb);
4081 failed_mount6:
4082         ext4_mb_release(sb);
4083 failed_mount5:
4084         ext4_ext_release(sb);
4085         ext4_release_system_zone(sb);
4086 failed_mount4a:
4087         dput(sb->s_root);
4088         sb->s_root = NULL;
4089 failed_mount4:
4090         ext4_msg(sb, KERN_ERR, "mount failed");
4091         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4092 failed_mount_wq:
4093         if (sbi->s_journal) {
4094                 jbd2_journal_destroy(sbi->s_journal);
4095                 sbi->s_journal = NULL;
4096         }
4097 failed_mount3:
4098         ext4_es_unregister_shrinker(sb);
4099         del_timer(&sbi->s_err_report);
4100         if (sbi->s_flex_groups)
4101                 ext4_kvfree(sbi->s_flex_groups);
4102         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4103         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4104         percpu_counter_destroy(&sbi->s_dirs_counter);
4105         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4106         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
4107         if (sbi->s_mmp_tsk)
4108                 kthread_stop(sbi->s_mmp_tsk);
4109 failed_mount2:
4110         for (i = 0; i < db_count; i++)
4111                 brelse(sbi->s_group_desc[i]);
4112         ext4_kvfree(sbi->s_group_desc);
4113 failed_mount:
4114         if (sbi->s_chksum_driver)
4115                 crypto_free_shash(sbi->s_chksum_driver);
4116         if (sbi->s_proc) {
4117                 remove_proc_entry("options", sbi->s_proc);
4118                 remove_proc_entry(sb->s_id, ext4_proc_root);
4119         }
4120 #ifdef CONFIG_QUOTA
4121         for (i = 0; i < MAXQUOTAS; i++)
4122                 kfree(sbi->s_qf_names[i]);
4123 #endif
4124         ext4_blkdev_remove(sbi);
4125         brelse(bh);
4126 out_fail:
4127         sb->s_fs_info = NULL;
4128         kfree(sbi->s_blockgroup_lock);
4129         kfree(sbi);
4130 out_free_orig:
4131         kfree(orig_data);
4132         return err ? err : ret;
4133 }
4134
4135 /*
4136  * Setup any per-fs journal parameters now.  We'll do this both on
4137  * initial mount, once the journal has been initialised but before we've
4138  * done any recovery; and again on any subsequent remount.
4139  */
4140 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4141 {
4142         struct ext4_sb_info *sbi = EXT4_SB(sb);
4143
4144         journal->j_commit_interval = sbi->s_commit_interval;
4145         journal->j_min_batch_time = sbi->s_min_batch_time;
4146         journal->j_max_batch_time = sbi->s_max_batch_time;
4147
4148         write_lock(&journal->j_state_lock);
4149         if (test_opt(sb, BARRIER))
4150                 journal->j_flags |= JBD2_BARRIER;
4151         else
4152                 journal->j_flags &= ~JBD2_BARRIER;
4153         if (test_opt(sb, DATA_ERR_ABORT))
4154                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4155         else
4156                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4157         write_unlock(&journal->j_state_lock);
4158 }
4159
4160 static journal_t *ext4_get_journal(struct super_block *sb,
4161                                    unsigned int journal_inum)
4162 {
4163         struct inode *journal_inode;
4164         journal_t *journal;
4165
4166         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4167
4168         /* First, test for the existence of a valid inode on disk.  Bad
4169          * things happen if we iget() an unused inode, as the subsequent
4170          * iput() will try to delete it. */
4171
4172         journal_inode = ext4_iget(sb, journal_inum);
4173         if (IS_ERR(journal_inode)) {
4174                 ext4_msg(sb, KERN_ERR, "no journal found");
4175                 return NULL;
4176         }
4177         if (!journal_inode->i_nlink) {
4178                 make_bad_inode(journal_inode);
4179                 iput(journal_inode);
4180                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4181                 return NULL;
4182         }
4183
4184         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4185                   journal_inode, journal_inode->i_size);
4186         if (!S_ISREG(journal_inode->i_mode)) {
4187                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4188                 iput(journal_inode);
4189                 return NULL;
4190         }
4191
4192         journal = jbd2_journal_init_inode(journal_inode);
4193         if (!journal) {
4194                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4195                 iput(journal_inode);
4196                 return NULL;
4197         }
4198         journal->j_private = sb;
4199         ext4_init_journal_params(sb, journal);
4200         return journal;
4201 }
4202
4203 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4204                                        dev_t j_dev)
4205 {
4206         struct buffer_head *bh;
4207         journal_t *journal;
4208         ext4_fsblk_t start;
4209         ext4_fsblk_t len;
4210         int hblock, blocksize;
4211         ext4_fsblk_t sb_block;
4212         unsigned long offset;
4213         struct ext4_super_block *es;
4214         struct block_device *bdev;
4215
4216         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4217
4218         bdev = ext4_blkdev_get(j_dev, sb);
4219         if (bdev == NULL)
4220                 return NULL;
4221
4222         blocksize = sb->s_blocksize;
4223         hblock = bdev_logical_block_size(bdev);
4224         if (blocksize < hblock) {
4225                 ext4_msg(sb, KERN_ERR,
4226                         "blocksize too small for journal device");
4227                 goto out_bdev;
4228         }
4229
4230         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4231         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4232         set_blocksize(bdev, blocksize);
4233         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4234                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4235                        "external journal");
4236                 goto out_bdev;
4237         }
4238
4239         es = (struct ext4_super_block *) (bh->b_data + offset);
4240         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4241             !(le32_to_cpu(es->s_feature_incompat) &
4242               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4243                 ext4_msg(sb, KERN_ERR, "external journal has "
4244                                         "bad superblock");
4245                 brelse(bh);
4246                 goto out_bdev;
4247         }
4248
4249         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4250                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4251                 brelse(bh);
4252                 goto out_bdev;
4253         }
4254
4255         len = ext4_blocks_count(es);
4256         start = sb_block + 1;
4257         brelse(bh);     /* we're done with the superblock */
4258
4259         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4260                                         start, len, blocksize);
4261         if (!journal) {
4262                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4263                 goto out_bdev;
4264         }
4265         journal->j_private = sb;
4266         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4267         wait_on_buffer(journal->j_sb_buffer);
4268         if (!buffer_uptodate(journal->j_sb_buffer)) {
4269                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4270                 goto out_journal;
4271         }
4272         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4273                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4274                                         "user (unsupported) - %d",
4275                         be32_to_cpu(journal->j_superblock->s_nr_users));
4276                 goto out_journal;
4277         }
4278         EXT4_SB(sb)->journal_bdev = bdev;
4279         ext4_init_journal_params(sb, journal);
4280         return journal;
4281
4282 out_journal:
4283         jbd2_journal_destroy(journal);
4284 out_bdev:
4285         ext4_blkdev_put(bdev);
4286         return NULL;
4287 }
4288
4289 static int ext4_load_journal(struct super_block *sb,
4290                              struct ext4_super_block *es,
4291                              unsigned long journal_devnum)
4292 {
4293         journal_t *journal;
4294         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4295         dev_t journal_dev;
4296         int err = 0;
4297         int really_read_only;
4298
4299         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4300
4301         if (journal_devnum &&
4302             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4303                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4304                         "numbers have changed");
4305                 journal_dev = new_decode_dev(journal_devnum);
4306         } else
4307                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4308
4309         really_read_only = bdev_read_only(sb->s_bdev);
4310
4311         /*
4312          * Are we loading a blank journal or performing recovery after a
4313          * crash?  For recovery, we need to check in advance whether we
4314          * can get read-write access to the device.
4315          */
4316         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4317                 if (sb->s_flags & MS_RDONLY) {
4318                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4319                                         "required on readonly filesystem");
4320                         if (really_read_only) {
4321                                 ext4_msg(sb, KERN_ERR, "write access "
4322                                         "unavailable, cannot proceed");
4323                                 return -EROFS;
4324                         }
4325                         ext4_msg(sb, KERN_INFO, "write access will "
4326                                "be enabled during recovery");
4327                 }
4328         }
4329
4330         if (journal_inum && journal_dev) {
4331                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4332                        "and inode journals!");
4333                 return -EINVAL;
4334         }
4335
4336         if (journal_inum) {
4337                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4338                         return -EINVAL;
4339         } else {
4340                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4341                         return -EINVAL;
4342         }
4343
4344         if (!(journal->j_flags & JBD2_BARRIER))
4345                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4346
4347         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4348                 err = jbd2_journal_wipe(journal, !really_read_only);
4349         if (!err) {
4350                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4351                 if (save)
4352                         memcpy(save, ((char *) es) +
4353                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4354                 err = jbd2_journal_load(journal);
4355                 if (save)
4356                         memcpy(((char *) es) + EXT4_S_ERR_START,
4357                                save, EXT4_S_ERR_LEN);
4358                 kfree(save);
4359         }
4360
4361         if (err) {
4362                 ext4_msg(sb, KERN_ERR, "error loading journal");
4363                 jbd2_journal_destroy(journal);
4364                 return err;
4365         }
4366
4367         EXT4_SB(sb)->s_journal = journal;
4368         ext4_clear_journal_err(sb, es);
4369
4370         if (!really_read_only && journal_devnum &&
4371             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4372                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4373
4374                 /* Make sure we flush the recovery flag to disk. */
4375                 ext4_commit_super(sb, 1);
4376         }
4377
4378         return 0;
4379 }
4380
4381 static int ext4_commit_super(struct super_block *sb, int sync)
4382 {
4383         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4384         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4385         int error = 0;
4386
4387         if (!sbh || block_device_ejected(sb))
4388                 return error;
4389         if (buffer_write_io_error(sbh)) {
4390                 /*
4391                  * Oh, dear.  A previous attempt to write the
4392                  * superblock failed.  This could happen because the
4393                  * USB device was yanked out.  Or it could happen to
4394                  * be a transient write error and maybe the block will
4395                  * be remapped.  Nothing we can do but to retry the
4396                  * write and hope for the best.
4397                  */
4398                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4399                        "superblock detected");
4400                 clear_buffer_write_io_error(sbh);
4401                 set_buffer_uptodate(sbh);
4402         }
4403         /*
4404          * If the file system is mounted read-only, don't update the
4405          * superblock write time.  This avoids updating the superblock
4406          * write time when we are mounting the root file system
4407          * read/only but we need to replay the journal; at that point,
4408          * for people who are east of GMT and who make their clock
4409          * tick in localtime for Windows bug-for-bug compatibility,
4410          * the clock is set in the future, and this will cause e2fsck
4411          * to complain and force a full file system check.
4412          */
4413         if (!(sb->s_flags & MS_RDONLY))
4414                 es->s_wtime = cpu_to_le32(get_seconds());
4415         if (sb->s_bdev->bd_part)
4416                 es->s_kbytes_written =
4417                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4418                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4419                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4420         else
4421                 es->s_kbytes_written =
4422                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4423         ext4_free_blocks_count_set(es,
4424                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4425                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4426         es->s_free_inodes_count =
4427                 cpu_to_le32(percpu_counter_sum_positive(
4428                                 &EXT4_SB(sb)->s_freeinodes_counter));
4429         BUFFER_TRACE(sbh, "marking dirty");
4430         ext4_superblock_csum_set(sb);
4431         mark_buffer_dirty(sbh);
4432         if (sync) {
4433                 error = sync_dirty_buffer(sbh);
4434                 if (error)
4435                         return error;
4436
4437                 error = buffer_write_io_error(sbh);
4438                 if (error) {
4439                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4440                                "superblock");
4441                         clear_buffer_write_io_error(sbh);
4442                         set_buffer_uptodate(sbh);
4443                 }
4444         }
4445         return error;
4446 }
4447
4448 /*
4449  * Have we just finished recovery?  If so, and if we are mounting (or
4450  * remounting) the filesystem readonly, then we will end up with a
4451  * consistent fs on disk.  Record that fact.
4452  */
4453 static void ext4_mark_recovery_complete(struct super_block *sb,
4454                                         struct ext4_super_block *es)
4455 {
4456         journal_t *journal = EXT4_SB(sb)->s_journal;
4457
4458         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4459                 BUG_ON(journal != NULL);
4460                 return;
4461         }
4462         jbd2_journal_lock_updates(journal);
4463         if (jbd2_journal_flush(journal) < 0)
4464                 goto out;
4465
4466         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4467             sb->s_flags & MS_RDONLY) {
4468                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4469                 ext4_commit_super(sb, 1);
4470         }
4471
4472 out:
4473         jbd2_journal_unlock_updates(journal);
4474 }
4475
4476 /*
4477  * If we are mounting (or read-write remounting) a filesystem whose journal
4478  * has recorded an error from a previous lifetime, move that error to the
4479  * main filesystem now.
4480  */
4481 static void ext4_clear_journal_err(struct super_block *sb,
4482                                    struct ext4_super_block *es)
4483 {
4484         journal_t *journal;
4485         int j_errno;
4486         const char *errstr;
4487
4488         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4489
4490         journal = EXT4_SB(sb)->s_journal;
4491
4492         /*
4493          * Now check for any error status which may have been recorded in the
4494          * journal by a prior ext4_error() or ext4_abort()
4495          */
4496
4497         j_errno = jbd2_journal_errno(journal);
4498         if (j_errno) {
4499                 char nbuf[16];
4500
4501                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4502                 ext4_warning(sb, "Filesystem error recorded "
4503                              "from previous mount: %s", errstr);
4504                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4505
4506                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4507                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4508                 ext4_commit_super(sb, 1);
4509
4510                 jbd2_journal_clear_err(journal);
4511                 jbd2_journal_update_sb_errno(journal);
4512         }
4513 }
4514
4515 /*
4516  * Force the running and committing transactions to commit,
4517  * and wait on the commit.
4518  */
4519 int ext4_force_commit(struct super_block *sb)
4520 {
4521         journal_t *journal;
4522
4523         if (sb->s_flags & MS_RDONLY)
4524                 return 0;
4525
4526         journal = EXT4_SB(sb)->s_journal;
4527         return ext4_journal_force_commit(journal);
4528 }
4529
4530 static int ext4_sync_fs(struct super_block *sb, int wait)
4531 {
4532         int ret = 0;
4533         tid_t target;
4534         struct ext4_sb_info *sbi = EXT4_SB(sb);
4535
4536         trace_ext4_sync_fs(sb, wait);
4537         flush_workqueue(sbi->dio_unwritten_wq);
4538         /*
4539          * Writeback quota in non-journalled quota case - journalled quota has
4540          * no dirty dquots
4541          */
4542         dquot_writeback_dquots(sb, -1);
4543         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4544                 if (wait)
4545                         jbd2_log_wait_commit(sbi->s_journal, target);
4546         }
4547         return ret;
4548 }
4549
4550 /*
4551  * LVM calls this function before a (read-only) snapshot is created.  This
4552  * gives us a chance to flush the journal completely and mark the fs clean.
4553  *
4554  * Note that only this function cannot bring a filesystem to be in a clean
4555  * state independently. It relies on upper layer to stop all data & metadata
4556  * modifications.
4557  */
4558 static int ext4_freeze(struct super_block *sb)
4559 {
4560         int error = 0;
4561         journal_t *journal;
4562
4563         if (sb->s_flags & MS_RDONLY)
4564                 return 0;
4565
4566         journal = EXT4_SB(sb)->s_journal;
4567
4568         /* Now we set up the journal barrier. */
4569         jbd2_journal_lock_updates(journal);
4570
4571         /*
4572          * Don't clear the needs_recovery flag if we failed to flush
4573          * the journal.
4574          */
4575         error = jbd2_journal_flush(journal);
4576         if (error < 0)
4577                 goto out;
4578
4579         /* Journal blocked and flushed, clear needs_recovery flag. */
4580         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4581         error = ext4_commit_super(sb, 1);
4582 out:
4583         /* we rely on upper layer to stop further updates */
4584         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4585         return error;
4586 }
4587
4588 /*
4589  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4590  * flag here, even though the filesystem is not technically dirty yet.
4591  */
4592 static int ext4_unfreeze(struct super_block *sb)
4593 {
4594         if (sb->s_flags & MS_RDONLY)
4595                 return 0;
4596
4597         /* Reset the needs_recovery flag before the fs is unlocked. */
4598         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4599         ext4_commit_super(sb, 1);
4600         return 0;
4601 }
4602
4603 /*
4604  * Structure to save mount options for ext4_remount's benefit
4605  */
4606 struct ext4_mount_options {
4607         unsigned long s_mount_opt;
4608         unsigned long s_mount_opt2;
4609         kuid_t s_resuid;
4610         kgid_t s_resgid;
4611         unsigned long s_commit_interval;
4612         u32 s_min_batch_time, s_max_batch_time;
4613 #ifdef CONFIG_QUOTA
4614         int s_jquota_fmt;
4615         char *s_qf_names[MAXQUOTAS];
4616 #endif
4617 };
4618
4619 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4620 {
4621         struct ext4_super_block *es;
4622         struct ext4_sb_info *sbi = EXT4_SB(sb);
4623         unsigned long old_sb_flags;
4624         struct ext4_mount_options old_opts;
4625         int enable_quota = 0;
4626         ext4_group_t g;
4627         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4628         int err = 0;
4629 #ifdef CONFIG_QUOTA
4630         int i, j;
4631 #endif
4632         char *orig_data = kstrdup(data, GFP_KERNEL);
4633
4634         sync_filesystem(sb);
4635
4636         /* Store the original options */
4637         old_sb_flags = sb->s_flags;
4638         old_opts.s_mount_opt = sbi->s_mount_opt;
4639         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4640         old_opts.s_resuid = sbi->s_resuid;
4641         old_opts.s_resgid = sbi->s_resgid;
4642         old_opts.s_commit_interval = sbi->s_commit_interval;
4643         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4644         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4645 #ifdef CONFIG_QUOTA
4646         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4647         for (i = 0; i < MAXQUOTAS; i++)
4648                 if (sbi->s_qf_names[i]) {
4649                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4650                                                          GFP_KERNEL);
4651                         if (!old_opts.s_qf_names[i]) {
4652                                 for (j = 0; j < i; j++)
4653                                         kfree(old_opts.s_qf_names[j]);
4654                                 kfree(orig_data);
4655                                 return -ENOMEM;
4656                         }
4657                 } else
4658                         old_opts.s_qf_names[i] = NULL;
4659 #endif
4660         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4661                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4662
4663         /*
4664          * Allow the "check" option to be passed as a remount option.
4665          */
4666         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4667                 err = -EINVAL;
4668                 goto restore_opts;
4669         }
4670
4671         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4672                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4673                         ext4_msg(sb, KERN_ERR, "can't mount with "
4674                                  "both data=journal and delalloc");
4675                         err = -EINVAL;
4676                         goto restore_opts;
4677                 }
4678                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4679                         ext4_msg(sb, KERN_ERR, "can't mount with "
4680                                  "both data=journal and dioread_nolock");
4681                         err = -EINVAL;
4682                         goto restore_opts;
4683                 }
4684         }
4685
4686         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4687                 ext4_abort(sb, "Abort forced by user");
4688
4689         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4690                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4691
4692         es = sbi->s_es;
4693
4694         if (sbi->s_journal) {
4695                 ext4_init_journal_params(sb, sbi->s_journal);
4696                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4697         }
4698
4699         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4700                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4701                         err = -EROFS;
4702                         goto restore_opts;
4703                 }
4704
4705                 if (*flags & MS_RDONLY) {
4706                         err = dquot_suspend(sb, -1);
4707                         if (err < 0)
4708                                 goto restore_opts;
4709
4710                         /*
4711                          * First of all, the unconditional stuff we have to do
4712                          * to disable replay of the journal when we next remount
4713                          */
4714                         sb->s_flags |= MS_RDONLY;
4715
4716                         /*
4717                          * OK, test if we are remounting a valid rw partition
4718                          * readonly, and if so set the rdonly flag and then
4719                          * mark the partition as valid again.
4720                          */
4721                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4722                             (sbi->s_mount_state & EXT4_VALID_FS))
4723                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4724
4725                         if (sbi->s_journal)
4726                                 ext4_mark_recovery_complete(sb, es);
4727                 } else {
4728                         /* Make sure we can mount this feature set readwrite */
4729                         if (!ext4_feature_set_ok(sb, 0)) {
4730                                 err = -EROFS;
4731                                 goto restore_opts;
4732                         }
4733                         /*
4734                          * Make sure the group descriptor checksums
4735                          * are sane.  If they aren't, refuse to remount r/w.
4736                          */
4737                         for (g = 0; g < sbi->s_groups_count; g++) {
4738                                 struct ext4_group_desc *gdp =
4739                                         ext4_get_group_desc(sb, g, NULL);
4740
4741                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4742                                         ext4_msg(sb, KERN_ERR,
4743                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4744                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4745                                                le16_to_cpu(gdp->bg_checksum));
4746                                         err = -EINVAL;
4747                                         goto restore_opts;
4748                                 }
4749                         }
4750
4751                         /*
4752                          * If we have an unprocessed orphan list hanging
4753                          * around from a previously readonly bdev mount,
4754                          * require a full umount/remount for now.
4755                          */
4756                         if (es->s_last_orphan) {
4757                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4758                                        "remount RDWR because of unprocessed "
4759                                        "orphan inode list.  Please "
4760                                        "umount/remount instead");
4761                                 err = -EINVAL;
4762                                 goto restore_opts;
4763                         }
4764
4765                         /*
4766                          * Mounting a RDONLY partition read-write, so reread
4767                          * and store the current valid flag.  (It may have
4768                          * been changed by e2fsck since we originally mounted
4769                          * the partition.)
4770                          */
4771                         if (sbi->s_journal)
4772                                 ext4_clear_journal_err(sb, es);
4773                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4774                         if (!ext4_setup_super(sb, es, 0))
4775                                 sb->s_flags &= ~MS_RDONLY;
4776                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4777                                                      EXT4_FEATURE_INCOMPAT_MMP))
4778                                 if (ext4_multi_mount_protect(sb,
4779                                                 le64_to_cpu(es->s_mmp_block))) {
4780                                         err = -EROFS;
4781                                         goto restore_opts;
4782                                 }
4783                         enable_quota = 1;
4784                 }
4785         }
4786
4787         /*
4788          * Reinitialize lazy itable initialization thread based on
4789          * current settings
4790          */
4791         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4792                 ext4_unregister_li_request(sb);
4793         else {
4794                 ext4_group_t first_not_zeroed;
4795                 first_not_zeroed = ext4_has_uninit_itable(sb);
4796                 ext4_register_li_request(sb, first_not_zeroed);
4797         }
4798
4799         ext4_setup_system_zone(sb);
4800         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4801                 ext4_commit_super(sb, 1);
4802
4803 #ifdef CONFIG_QUOTA
4804         /* Release old quota file names */
4805         for (i = 0; i < MAXQUOTAS; i++)
4806                 kfree(old_opts.s_qf_names[i]);
4807         if (enable_quota) {
4808                 if (sb_any_quota_suspended(sb))
4809                         dquot_resume(sb, -1);
4810                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4811                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4812                         err = ext4_enable_quotas(sb);
4813                         if (err)
4814                                 goto restore_opts;
4815                 }
4816         }
4817 #endif
4818
4819         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4820         kfree(orig_data);
4821         return 0;
4822
4823 restore_opts:
4824         sb->s_flags = old_sb_flags;
4825         sbi->s_mount_opt = old_opts.s_mount_opt;
4826         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4827         sbi->s_resuid = old_opts.s_resuid;
4828         sbi->s_resgid = old_opts.s_resgid;
4829         sbi->s_commit_interval = old_opts.s_commit_interval;
4830         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4831         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4832 #ifdef CONFIG_QUOTA
4833         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4834         for (i = 0; i < MAXQUOTAS; i++) {
4835                 kfree(sbi->s_qf_names[i]);
4836                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4837         }
4838 #endif
4839         kfree(orig_data);
4840         return err;
4841 }
4842
4843 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4844 {
4845         struct super_block *sb = dentry->d_sb;
4846         struct ext4_sb_info *sbi = EXT4_SB(sb);
4847         struct ext4_super_block *es = sbi->s_es;
4848         ext4_fsblk_t overhead = 0, resv_blocks;
4849         u64 fsid;
4850         s64 bfree;
4851         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4852
4853         if (!test_opt(sb, MINIX_DF))
4854                 overhead = sbi->s_overhead;
4855
4856         buf->f_type = EXT4_SUPER_MAGIC;
4857         buf->f_bsize = sb->s_blocksize;
4858         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4859         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4860                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4861         /* prevent underflow in case that few free space is available */
4862         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4863         buf->f_bavail = buf->f_bfree -
4864                         (ext4_r_blocks_count(es) + resv_blocks);
4865         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4866                 buf->f_bavail = 0;
4867         buf->f_files = le32_to_cpu(es->s_inodes_count);
4868         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4869         buf->f_namelen = EXT4_NAME_LEN;
4870         fsid = le64_to_cpup((void *)es->s_uuid) ^
4871                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4872         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4873         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4874
4875         return 0;
4876 }
4877
4878 /* Helper function for writing quotas on sync - we need to start transaction
4879  * before quota file is locked for write. Otherwise the are possible deadlocks:
4880  * Process 1                         Process 2
4881  * ext4_create()                     quota_sync()
4882  *   jbd2_journal_start()                  write_dquot()
4883  *   dquot_initialize()                         down(dqio_mutex)
4884  *     down(dqio_mutex)                    jbd2_journal_start()
4885  *
4886  */
4887
4888 #ifdef CONFIG_QUOTA
4889
4890 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4891 {
4892         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4893 }
4894
4895 static int ext4_write_dquot(struct dquot *dquot)
4896 {
4897         int ret, err;
4898         handle_t *handle;
4899         struct inode *inode;
4900
4901         inode = dquot_to_inode(dquot);
4902         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4903                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4904         if (IS_ERR(handle))
4905                 return PTR_ERR(handle);
4906         ret = dquot_commit(dquot);
4907         err = ext4_journal_stop(handle);
4908         if (!ret)
4909                 ret = err;
4910         return ret;
4911 }
4912
4913 static int ext4_acquire_dquot(struct dquot *dquot)
4914 {
4915         int ret, err;
4916         handle_t *handle;
4917
4918         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4919                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4920         if (IS_ERR(handle))
4921                 return PTR_ERR(handle);
4922         ret = dquot_acquire(dquot);
4923         err = ext4_journal_stop(handle);
4924         if (!ret)
4925                 ret = err;
4926         return ret;
4927 }
4928
4929 static int ext4_release_dquot(struct dquot *dquot)
4930 {
4931         int ret, err;
4932         handle_t *handle;
4933
4934         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4935                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4936         if (IS_ERR(handle)) {
4937                 /* Release dquot anyway to avoid endless cycle in dqput() */
4938                 dquot_release(dquot);
4939                 return PTR_ERR(handle);
4940         }
4941         ret = dquot_release(dquot);
4942         err = ext4_journal_stop(handle);
4943         if (!ret)
4944                 ret = err;
4945         return ret;
4946 }
4947
4948 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4949 {
4950         struct super_block *sb = dquot->dq_sb;
4951         struct ext4_sb_info *sbi = EXT4_SB(sb);
4952
4953         /* Are we journaling quotas? */
4954         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
4955             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
4956                 dquot_mark_dquot_dirty(dquot);
4957                 return ext4_write_dquot(dquot);
4958         } else {
4959                 return dquot_mark_dquot_dirty(dquot);
4960         }
4961 }
4962
4963 static int ext4_write_info(struct super_block *sb, int type)
4964 {
4965         int ret, err;
4966         handle_t *handle;
4967
4968         /* Data block + inode block */
4969         handle = ext4_journal_start(sb->s_root->d_inode, EXT4_HT_QUOTA, 2);
4970         if (IS_ERR(handle))
4971                 return PTR_ERR(handle);
4972         ret = dquot_commit_info(sb, type);
4973         err = ext4_journal_stop(handle);
4974         if (!ret)
4975                 ret = err;
4976         return ret;
4977 }
4978
4979 /*
4980  * Turn on quotas during mount time - we need to find
4981  * the quota file and such...
4982  */
4983 static int ext4_quota_on_mount(struct super_block *sb, int type)
4984 {
4985         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4986                                         EXT4_SB(sb)->s_jquota_fmt, type);
4987 }
4988
4989 /*
4990  * Standard function to be called on quota_on
4991  */
4992 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4993                          struct path *path)
4994 {
4995         int err;
4996
4997         if (!test_opt(sb, QUOTA))
4998                 return -EINVAL;
4999
5000         /* Quotafile not on the same filesystem? */
5001         if (path->dentry->d_sb != sb)
5002                 return -EXDEV;
5003         /* Journaling quota? */
5004         if (EXT4_SB(sb)->s_qf_names[type]) {
5005                 /* Quotafile not in fs root? */
5006                 if (path->dentry->d_parent != sb->s_root)
5007                         ext4_msg(sb, KERN_WARNING,
5008                                 "Quota file not on filesystem root. "
5009                                 "Journaled quota will not work");
5010         }
5011
5012         /*
5013          * When we journal data on quota file, we have to flush journal to see
5014          * all updates to the file when we bypass pagecache...
5015          */
5016         if (EXT4_SB(sb)->s_journal &&
5017             ext4_should_journal_data(path->dentry->d_inode)) {
5018                 /*
5019                  * We don't need to lock updates but journal_flush() could
5020                  * otherwise be livelocked...
5021                  */
5022                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5023                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5024                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5025                 if (err)
5026                         return err;
5027         }
5028
5029         return dquot_quota_on(sb, type, format_id, path);
5030 }
5031
5032 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5033                              unsigned int flags)
5034 {
5035         int err;
5036         struct inode *qf_inode;
5037         unsigned long qf_inums[MAXQUOTAS] = {
5038                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5039                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5040         };
5041
5042         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5043
5044         if (!qf_inums[type])
5045                 return -EPERM;
5046
5047         qf_inode = ext4_iget(sb, qf_inums[type]);
5048         if (IS_ERR(qf_inode)) {
5049                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5050                 return PTR_ERR(qf_inode);
5051         }
5052
5053         /* Don't account quota for quota files to avoid recursion */
5054         qf_inode->i_flags |= S_NOQUOTA;
5055         err = dquot_enable(qf_inode, type, format_id, flags);
5056         iput(qf_inode);
5057
5058         return err;
5059 }
5060
5061 /* Enable usage tracking for all quota types. */
5062 static int ext4_enable_quotas(struct super_block *sb)
5063 {
5064         int type, err = 0;
5065         unsigned long qf_inums[MAXQUOTAS] = {
5066                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5067                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5068         };
5069
5070         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5071         for (type = 0; type < MAXQUOTAS; type++) {
5072                 if (qf_inums[type]) {
5073                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5074                                                 DQUOT_USAGE_ENABLED);
5075                         if (err) {
5076                                 ext4_warning(sb,
5077                                         "Failed to enable quota tracking "
5078                                         "(type=%d, err=%d). Please run "
5079                                         "e2fsck to fix.", type, err);
5080                                 return err;
5081                         }
5082                 }
5083         }
5084         return 0;
5085 }
5086
5087 /*
5088  * quota_on function that is used when QUOTA feature is set.
5089  */
5090 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5091                                  int format_id)
5092 {
5093         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5094                 return -EINVAL;
5095
5096         /*
5097          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5098          */
5099         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5100 }
5101
5102 static int ext4_quota_off(struct super_block *sb, int type)
5103 {
5104         struct inode *inode = sb_dqopt(sb)->files[type];
5105         handle_t *handle;
5106
5107         /* Force all delayed allocation blocks to be allocated.
5108          * Caller already holds s_umount sem */
5109         if (test_opt(sb, DELALLOC))
5110                 sync_filesystem(sb);
5111
5112         if (!inode)
5113                 goto out;
5114
5115         /* Update modification times of quota files when userspace can
5116          * start looking at them */
5117         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5118         if (IS_ERR(handle))
5119                 goto out;
5120         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5121         ext4_mark_inode_dirty(handle, inode);
5122         ext4_journal_stop(handle);
5123
5124 out:
5125         return dquot_quota_off(sb, type);
5126 }
5127
5128 /*
5129  * quota_off function that is used when QUOTA feature is set.
5130  */
5131 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5132 {
5133         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5134                 return -EINVAL;
5135
5136         /* Disable only the limits. */
5137         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5138 }
5139
5140 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5141  * acquiring the locks... As quota files are never truncated and quota code
5142  * itself serializes the operations (and no one else should touch the files)
5143  * we don't have to be afraid of races */
5144 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5145                                size_t len, loff_t off)
5146 {
5147         struct inode *inode = sb_dqopt(sb)->files[type];
5148         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5149         int err = 0;
5150         int offset = off & (sb->s_blocksize - 1);
5151         int tocopy;
5152         size_t toread;
5153         struct buffer_head *bh;
5154         loff_t i_size = i_size_read(inode);
5155
5156         if (off > i_size)
5157                 return 0;
5158         if (off+len > i_size)
5159                 len = i_size-off;
5160         toread = len;
5161         while (toread > 0) {
5162                 tocopy = sb->s_blocksize - offset < toread ?
5163                                 sb->s_blocksize - offset : toread;
5164                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5165                 if (err)
5166                         return err;
5167                 if (!bh)        /* A hole? */
5168                         memset(data, 0, tocopy);
5169                 else
5170                         memcpy(data, bh->b_data+offset, tocopy);
5171                 brelse(bh);
5172                 offset = 0;
5173                 toread -= tocopy;
5174                 data += tocopy;
5175                 blk++;
5176         }
5177         return len;
5178 }
5179
5180 /* Write to quotafile (we know the transaction is already started and has
5181  * enough credits) */
5182 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5183                                 const char *data, size_t len, loff_t off)
5184 {
5185         struct inode *inode = sb_dqopt(sb)->files[type];
5186         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5187         int err = 0;
5188         int offset = off & (sb->s_blocksize - 1);
5189         struct buffer_head *bh;
5190         handle_t *handle = journal_current_handle();
5191
5192         if (EXT4_SB(sb)->s_journal && !handle) {
5193                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5194                         " cancelled because transaction is not started",
5195                         (unsigned long long)off, (unsigned long long)len);
5196                 return -EIO;
5197         }
5198         /*
5199          * Since we account only one data block in transaction credits,
5200          * then it is impossible to cross a block boundary.
5201          */
5202         if (sb->s_blocksize - offset < len) {
5203                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5204                         " cancelled because not block aligned",
5205                         (unsigned long long)off, (unsigned long long)len);
5206                 return -EIO;
5207         }
5208
5209         bh = ext4_bread(handle, inode, blk, 1, &err);
5210         if (!bh)
5211                 goto out;
5212         err = ext4_journal_get_write_access(handle, bh);
5213         if (err) {
5214                 brelse(bh);
5215                 goto out;
5216         }
5217         lock_buffer(bh);
5218         memcpy(bh->b_data+offset, data, len);
5219         flush_dcache_page(bh->b_page);
5220         unlock_buffer(bh);
5221         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5222         brelse(bh);
5223 out:
5224         if (err)
5225                 return err;
5226         if (inode->i_size < off + len) {
5227                 i_size_write(inode, off + len);
5228                 EXT4_I(inode)->i_disksize = inode->i_size;
5229                 ext4_mark_inode_dirty(handle, inode);
5230         }
5231         return len;
5232 }
5233
5234 #endif
5235
5236 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5237                        const char *dev_name, void *data)
5238 {
5239         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5240 }
5241
5242 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5243 static inline void register_as_ext2(void)
5244 {
5245         int err = register_filesystem(&ext2_fs_type);
5246         if (err)
5247                 printk(KERN_WARNING
5248                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5249 }
5250
5251 static inline void unregister_as_ext2(void)
5252 {
5253         unregister_filesystem(&ext2_fs_type);
5254 }
5255
5256 static inline int ext2_feature_set_ok(struct super_block *sb)
5257 {
5258         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5259                 return 0;
5260         if (sb->s_flags & MS_RDONLY)
5261                 return 1;
5262         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5263                 return 0;
5264         return 1;
5265 }
5266 #else
5267 static inline void register_as_ext2(void) { }
5268 static inline void unregister_as_ext2(void) { }
5269 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5270 #endif
5271
5272 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5273 static inline void register_as_ext3(void)
5274 {
5275         int err = register_filesystem(&ext3_fs_type);
5276         if (err)
5277                 printk(KERN_WARNING
5278                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5279 }
5280
5281 static inline void unregister_as_ext3(void)
5282 {
5283         unregister_filesystem(&ext3_fs_type);
5284 }
5285
5286 static inline int ext3_feature_set_ok(struct super_block *sb)
5287 {
5288         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5289                 return 0;
5290         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5291                 return 0;
5292         if (sb->s_flags & MS_RDONLY)
5293                 return 1;
5294         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5295                 return 0;
5296         return 1;
5297 }
5298 #else
5299 static inline void register_as_ext3(void) { }
5300 static inline void unregister_as_ext3(void) { }
5301 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5302 #endif
5303
5304 static struct file_system_type ext4_fs_type = {
5305         .owner          = THIS_MODULE,
5306         .name           = "ext4",
5307         .mount          = ext4_mount,
5308         .kill_sb        = kill_block_super,
5309         .fs_flags       = FS_REQUIRES_DEV,
5310 };
5311 MODULE_ALIAS_FS("ext4");
5312
5313 static int __init ext4_init_feat_adverts(void)
5314 {
5315         struct ext4_features *ef;
5316         int ret = -ENOMEM;
5317
5318         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5319         if (!ef)
5320                 goto out;
5321
5322         ef->f_kobj.kset = ext4_kset;
5323         init_completion(&ef->f_kobj_unregister);
5324         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5325                                    "features");
5326         if (ret) {
5327                 kfree(ef);
5328                 goto out;
5329         }
5330
5331         ext4_feat = ef;
5332         ret = 0;
5333 out:
5334         return ret;
5335 }
5336
5337 static void ext4_exit_feat_adverts(void)
5338 {
5339         kobject_put(&ext4_feat->f_kobj);
5340         wait_for_completion(&ext4_feat->f_kobj_unregister);
5341         kfree(ext4_feat);
5342 }
5343
5344 /* Shared across all ext4 file systems */
5345 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5346 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5347
5348 static int __init ext4_init_fs(void)
5349 {
5350         int i, err;
5351
5352         ext4_li_info = NULL;
5353         mutex_init(&ext4_li_mtx);
5354
5355         /* Build-time check for flags consistency */
5356         ext4_check_flag_values();
5357
5358         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5359                 mutex_init(&ext4__aio_mutex[i]);
5360                 init_waitqueue_head(&ext4__ioend_wq[i]);
5361         }
5362
5363         err = ext4_init_es();
5364         if (err)
5365                 return err;
5366
5367         err = ext4_init_pageio();
5368         if (err)
5369                 goto out7;
5370
5371         err = ext4_init_system_zone();
5372         if (err)
5373                 goto out6;
5374         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5375         if (!ext4_kset) {
5376                 err = -ENOMEM;
5377                 goto out5;
5378         }
5379         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5380
5381         err = ext4_init_feat_adverts();
5382         if (err)
5383                 goto out4;
5384
5385         err = ext4_init_mballoc();
5386         if (err)
5387                 goto out3;
5388
5389         err = ext4_init_xattr();
5390         if (err)
5391                 goto out2;
5392         err = init_inodecache();
5393         if (err)
5394                 goto out1;
5395         register_as_ext3();
5396         register_as_ext2();
5397         err = register_filesystem(&ext4_fs_type);
5398         if (err)
5399                 goto out;
5400
5401         return 0;
5402 out:
5403         unregister_as_ext2();
5404         unregister_as_ext3();
5405         destroy_inodecache();
5406 out1:
5407         ext4_exit_xattr();
5408 out2:
5409         ext4_exit_mballoc();
5410 out3:
5411         ext4_exit_feat_adverts();
5412 out4:
5413         if (ext4_proc_root)
5414                 remove_proc_entry("fs/ext4", NULL);
5415         kset_unregister(ext4_kset);
5416 out5:
5417         ext4_exit_system_zone();
5418 out6:
5419         ext4_exit_pageio();
5420 out7:
5421         ext4_exit_es();
5422
5423         return err;
5424 }
5425
5426 static void __exit ext4_exit_fs(void)
5427 {
5428         ext4_destroy_lazyinit_thread();
5429         unregister_as_ext2();
5430         unregister_as_ext3();
5431         unregister_filesystem(&ext4_fs_type);
5432         destroy_inodecache();
5433         ext4_exit_xattr();
5434         ext4_exit_mballoc();
5435         ext4_exit_feat_adverts();
5436         remove_proc_entry("fs/ext4", NULL);
5437         kset_unregister(ext4_kset);
5438         ext4_exit_system_zone();
5439         ext4_exit_pageio();
5440         ext4_exit_es();
5441 }
5442
5443 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5444 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5445 MODULE_LICENSE("GPL");
5446 module_init(ext4_init_fs)
5447 module_exit(ext4_exit_fs)