Merge tag 'writeback' of git://git.kernel.org/pub/scm/linux/kernel/git/wfg/linux
[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"
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 const char *ext4_decode_error(struct super_block *sb, int errno,
73                                      char nbuf[16]);
74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
76 static int ext4_unfreeze(struct super_block *sb);
77 static void ext4_write_super(struct super_block *sb);
78 static int ext4_freeze(struct super_block *sb);
79 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
80                        const char *dev_name, void *data);
81 static inline int ext2_feature_set_ok(struct super_block *sb);
82 static inline int ext3_feature_set_ok(struct super_block *sb);
83 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
84 static void ext4_destroy_lazyinit_thread(void);
85 static void ext4_unregister_li_request(struct super_block *sb);
86 static void ext4_clear_request_list(void);
87
88 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
89 static struct file_system_type ext2_fs_type = {
90         .owner          = THIS_MODULE,
91         .name           = "ext2",
92         .mount          = ext4_mount,
93         .kill_sb        = kill_block_super,
94         .fs_flags       = FS_REQUIRES_DEV,
95 };
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 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
111 #else
112 #define IS_EXT3_SB(sb) (0)
113 #endif
114
115 void *ext4_kvmalloc(size_t size, gfp_t flags)
116 {
117         void *ret;
118
119         ret = kmalloc(size, flags);
120         if (!ret)
121                 ret = __vmalloc(size, flags, PAGE_KERNEL);
122         return ret;
123 }
124
125 void *ext4_kvzalloc(size_t size, gfp_t flags)
126 {
127         void *ret;
128
129         ret = kzalloc(size, flags);
130         if (!ret)
131                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
132         return ret;
133 }
134
135 void ext4_kvfree(void *ptr)
136 {
137         if (is_vmalloc_addr(ptr))
138                 vfree(ptr);
139         else
140                 kfree(ptr);
141
142 }
143
144 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
145                                struct ext4_group_desc *bg)
146 {
147         return le32_to_cpu(bg->bg_block_bitmap_lo) |
148                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
149                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
150 }
151
152 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
153                                struct ext4_group_desc *bg)
154 {
155         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
156                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
157                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
158 }
159
160 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
161                               struct ext4_group_desc *bg)
162 {
163         return le32_to_cpu(bg->bg_inode_table_lo) |
164                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
165                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
166 }
167
168 __u32 ext4_free_group_clusters(struct super_block *sb,
169                                struct ext4_group_desc *bg)
170 {
171         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
172                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
173                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
174 }
175
176 __u32 ext4_free_inodes_count(struct super_block *sb,
177                               struct ext4_group_desc *bg)
178 {
179         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
180                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
181                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
182 }
183
184 __u32 ext4_used_dirs_count(struct super_block *sb,
185                               struct ext4_group_desc *bg)
186 {
187         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
188                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
189                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
190 }
191
192 __u32 ext4_itable_unused_count(struct super_block *sb,
193                               struct ext4_group_desc *bg)
194 {
195         return le16_to_cpu(bg->bg_itable_unused_lo) |
196                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
197                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
198 }
199
200 void ext4_block_bitmap_set(struct super_block *sb,
201                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
202 {
203         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
204         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
205                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
206 }
207
208 void ext4_inode_bitmap_set(struct super_block *sb,
209                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
210 {
211         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
212         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
213                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
214 }
215
216 void ext4_inode_table_set(struct super_block *sb,
217                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
218 {
219         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
220         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
221                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
222 }
223
224 void ext4_free_group_clusters_set(struct super_block *sb,
225                                   struct ext4_group_desc *bg, __u32 count)
226 {
227         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
228         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
229                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
230 }
231
232 void ext4_free_inodes_set(struct super_block *sb,
233                           struct ext4_group_desc *bg, __u32 count)
234 {
235         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
236         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
237                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
238 }
239
240 void ext4_used_dirs_set(struct super_block *sb,
241                           struct ext4_group_desc *bg, __u32 count)
242 {
243         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
244         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
245                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
246 }
247
248 void ext4_itable_unused_set(struct super_block *sb,
249                           struct ext4_group_desc *bg, __u32 count)
250 {
251         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
252         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
253                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
254 }
255
256
257 /* Just increment the non-pointer handle value */
258 static handle_t *ext4_get_nojournal(void)
259 {
260         handle_t *handle = current->journal_info;
261         unsigned long ref_cnt = (unsigned long)handle;
262
263         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
264
265         ref_cnt++;
266         handle = (handle_t *)ref_cnt;
267
268         current->journal_info = handle;
269         return handle;
270 }
271
272
273 /* Decrement the non-pointer handle value */
274 static void ext4_put_nojournal(handle_t *handle)
275 {
276         unsigned long ref_cnt = (unsigned long)handle;
277
278         BUG_ON(ref_cnt == 0);
279
280         ref_cnt--;
281         handle = (handle_t *)ref_cnt;
282
283         current->journal_info = handle;
284 }
285
286 /*
287  * Wrappers for jbd2_journal_start/end.
288  *
289  * The only special thing we need to do here is to make sure that all
290  * journal_end calls result in the superblock being marked dirty, so
291  * that sync() will call the filesystem's write_super callback if
292  * appropriate.
293  *
294  * To avoid j_barrier hold in userspace when a user calls freeze(),
295  * ext4 prevents a new handle from being started by s_frozen, which
296  * is in an upper layer.
297  */
298 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
299 {
300         journal_t *journal;
301         handle_t  *handle;
302
303         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
304         if (sb->s_flags & MS_RDONLY)
305                 return ERR_PTR(-EROFS);
306
307         journal = EXT4_SB(sb)->s_journal;
308         handle = ext4_journal_current_handle();
309
310         /*
311          * If a handle has been started, it should be allowed to
312          * finish, otherwise deadlock could happen between freeze
313          * and others(e.g. truncate) due to the restart of the
314          * journal handle if the filesystem is forzen and active
315          * handles are not stopped.
316          */
317         if (!handle)
318                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
319
320         if (!journal)
321                 return ext4_get_nojournal();
322         /*
323          * Special case here: if the journal has aborted behind our
324          * backs (eg. EIO in the commit thread), then we still need to
325          * take the FS itself readonly cleanly.
326          */
327         if (is_journal_aborted(journal)) {
328                 ext4_abort(sb, "Detected aborted journal");
329                 return ERR_PTR(-EROFS);
330         }
331         return jbd2_journal_start(journal, nblocks);
332 }
333
334 /*
335  * The only special thing we need to do here is to make sure that all
336  * jbd2_journal_stop calls result in the superblock being marked dirty, so
337  * that sync() will call the filesystem's write_super callback if
338  * appropriate.
339  */
340 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
341 {
342         struct super_block *sb;
343         int err;
344         int rc;
345
346         if (!ext4_handle_valid(handle)) {
347                 ext4_put_nojournal(handle);
348                 return 0;
349         }
350         sb = handle->h_transaction->t_journal->j_private;
351         err = handle->h_err;
352         rc = jbd2_journal_stop(handle);
353
354         if (!err)
355                 err = rc;
356         if (err)
357                 __ext4_std_error(sb, where, line, err);
358         return err;
359 }
360
361 void ext4_journal_abort_handle(const char *caller, unsigned int line,
362                                const char *err_fn, struct buffer_head *bh,
363                                handle_t *handle, int err)
364 {
365         char nbuf[16];
366         const char *errstr = ext4_decode_error(NULL, err, nbuf);
367
368         BUG_ON(!ext4_handle_valid(handle));
369
370         if (bh)
371                 BUFFER_TRACE(bh, "abort");
372
373         if (!handle->h_err)
374                 handle->h_err = err;
375
376         if (is_handle_aborted(handle))
377                 return;
378
379         printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
380                caller, line, errstr, err_fn);
381
382         jbd2_journal_abort_handle(handle);
383 }
384
385 static void __save_error_info(struct super_block *sb, const char *func,
386                             unsigned int line)
387 {
388         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
389
390         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
391         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
392         es->s_last_error_time = cpu_to_le32(get_seconds());
393         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
394         es->s_last_error_line = cpu_to_le32(line);
395         if (!es->s_first_error_time) {
396                 es->s_first_error_time = es->s_last_error_time;
397                 strncpy(es->s_first_error_func, func,
398                         sizeof(es->s_first_error_func));
399                 es->s_first_error_line = cpu_to_le32(line);
400                 es->s_first_error_ino = es->s_last_error_ino;
401                 es->s_first_error_block = es->s_last_error_block;
402         }
403         /*
404          * Start the daily error reporting function if it hasn't been
405          * started already
406          */
407         if (!es->s_error_count)
408                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
409         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
410 }
411
412 static void save_error_info(struct super_block *sb, const char *func,
413                             unsigned int line)
414 {
415         __save_error_info(sb, func, line);
416         ext4_commit_super(sb, 1);
417 }
418
419 /*
420  * The del_gendisk() function uninitializes the disk-specific data
421  * structures, including the bdi structure, without telling anyone
422  * else.  Once this happens, any attempt to call mark_buffer_dirty()
423  * (for example, by ext4_commit_super), will cause a kernel OOPS.
424  * This is a kludge to prevent these oops until we can put in a proper
425  * hook in del_gendisk() to inform the VFS and file system layers.
426  */
427 static int block_device_ejected(struct super_block *sb)
428 {
429         struct inode *bd_inode = sb->s_bdev->bd_inode;
430         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
431
432         return bdi->dev == NULL;
433 }
434
435 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
436 {
437         struct super_block              *sb = journal->j_private;
438         struct ext4_sb_info             *sbi = EXT4_SB(sb);
439         int                             error = is_journal_aborted(journal);
440         struct ext4_journal_cb_entry    *jce, *tmp;
441
442         spin_lock(&sbi->s_md_lock);
443         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
444                 list_del_init(&jce->jce_list);
445                 spin_unlock(&sbi->s_md_lock);
446                 jce->jce_func(sb, jce, error);
447                 spin_lock(&sbi->s_md_lock);
448         }
449         spin_unlock(&sbi->s_md_lock);
450 }
451
452 /* Deal with the reporting of failure conditions on a filesystem such as
453  * inconsistencies detected or read IO failures.
454  *
455  * On ext2, we can store the error state of the filesystem in the
456  * superblock.  That is not possible on ext4, because we may have other
457  * write ordering constraints on the superblock which prevent us from
458  * writing it out straight away; and given that the journal is about to
459  * be aborted, we can't rely on the current, or future, transactions to
460  * write out the superblock safely.
461  *
462  * We'll just use the jbd2_journal_abort() error code to record an error in
463  * the journal instead.  On recovery, the journal will complain about
464  * that error until we've noted it down and cleared it.
465  */
466
467 static void ext4_handle_error(struct super_block *sb)
468 {
469         if (sb->s_flags & MS_RDONLY)
470                 return;
471
472         if (!test_opt(sb, ERRORS_CONT)) {
473                 journal_t *journal = EXT4_SB(sb)->s_journal;
474
475                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
476                 if (journal)
477                         jbd2_journal_abort(journal, -EIO);
478         }
479         if (test_opt(sb, ERRORS_RO)) {
480                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
481                 sb->s_flags |= MS_RDONLY;
482         }
483         if (test_opt(sb, ERRORS_PANIC))
484                 panic("EXT4-fs (device %s): panic forced after error\n",
485                         sb->s_id);
486 }
487
488 void __ext4_error(struct super_block *sb, const char *function,
489                   unsigned int line, const char *fmt, ...)
490 {
491         struct va_format vaf;
492         va_list args;
493
494         va_start(args, fmt);
495         vaf.fmt = fmt;
496         vaf.va = &args;
497         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
498                sb->s_id, function, line, current->comm, &vaf);
499         va_end(args);
500
501         ext4_handle_error(sb);
502 }
503
504 void ext4_error_inode(struct inode *inode, const char *function,
505                       unsigned int line, ext4_fsblk_t block,
506                       const char *fmt, ...)
507 {
508         va_list args;
509         struct va_format vaf;
510         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
511
512         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
513         es->s_last_error_block = cpu_to_le64(block);
514         save_error_info(inode->i_sb, function, line);
515         va_start(args, fmt);
516         vaf.fmt = fmt;
517         vaf.va = &args;
518         if (block)
519                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
520                        "inode #%lu: block %llu: comm %s: %pV\n",
521                        inode->i_sb->s_id, function, line, inode->i_ino,
522                        block, current->comm, &vaf);
523         else
524                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
525                        "inode #%lu: comm %s: %pV\n",
526                        inode->i_sb->s_id, function, line, inode->i_ino,
527                        current->comm, &vaf);
528         va_end(args);
529
530         ext4_handle_error(inode->i_sb);
531 }
532
533 void ext4_error_file(struct file *file, const char *function,
534                      unsigned int line, ext4_fsblk_t block,
535                      const char *fmt, ...)
536 {
537         va_list args;
538         struct va_format vaf;
539         struct ext4_super_block *es;
540         struct inode *inode = file->f_dentry->d_inode;
541         char pathname[80], *path;
542
543         es = EXT4_SB(inode->i_sb)->s_es;
544         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
545         save_error_info(inode->i_sb, function, line);
546         path = d_path(&(file->f_path), pathname, sizeof(pathname));
547         if (IS_ERR(path))
548                 path = "(unknown)";
549         va_start(args, fmt);
550         vaf.fmt = fmt;
551         vaf.va = &args;
552         if (block)
553                 printk(KERN_CRIT
554                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
555                        "block %llu: comm %s: path %s: %pV\n",
556                        inode->i_sb->s_id, function, line, inode->i_ino,
557                        block, current->comm, path, &vaf);
558         else
559                 printk(KERN_CRIT
560                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
561                        "comm %s: path %s: %pV\n",
562                        inode->i_sb->s_id, function, line, inode->i_ino,
563                        current->comm, path, &vaf);
564         va_end(args);
565
566         ext4_handle_error(inode->i_sb);
567 }
568
569 static const char *ext4_decode_error(struct super_block *sb, int errno,
570                                      char nbuf[16])
571 {
572         char *errstr = NULL;
573
574         switch (errno) {
575         case -EIO:
576                 errstr = "IO failure";
577                 break;
578         case -ENOMEM:
579                 errstr = "Out of memory";
580                 break;
581         case -EROFS:
582                 if (!sb || (EXT4_SB(sb)->s_journal &&
583                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
584                         errstr = "Journal has aborted";
585                 else
586                         errstr = "Readonly filesystem";
587                 break;
588         default:
589                 /* If the caller passed in an extra buffer for unknown
590                  * errors, textualise them now.  Else we just return
591                  * NULL. */
592                 if (nbuf) {
593                         /* Check for truncated error codes... */
594                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
595                                 errstr = nbuf;
596                 }
597                 break;
598         }
599
600         return errstr;
601 }
602
603 /* __ext4_std_error decodes expected errors from journaling functions
604  * automatically and invokes the appropriate error response.  */
605
606 void __ext4_std_error(struct super_block *sb, const char *function,
607                       unsigned int line, int errno)
608 {
609         char nbuf[16];
610         const char *errstr;
611
612         /* Special case: if the error is EROFS, and we're not already
613          * inside a transaction, then there's really no point in logging
614          * an error. */
615         if (errno == -EROFS && journal_current_handle() == NULL &&
616             (sb->s_flags & MS_RDONLY))
617                 return;
618
619         errstr = ext4_decode_error(sb, errno, nbuf);
620         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
621                sb->s_id, function, line, errstr);
622         save_error_info(sb, function, line);
623
624         ext4_handle_error(sb);
625 }
626
627 /*
628  * ext4_abort is a much stronger failure handler than ext4_error.  The
629  * abort function may be used to deal with unrecoverable failures such
630  * as journal IO errors or ENOMEM at a critical moment in log management.
631  *
632  * We unconditionally force the filesystem into an ABORT|READONLY state,
633  * unless the error response on the fs has been set to panic in which
634  * case we take the easy way out and panic immediately.
635  */
636
637 void __ext4_abort(struct super_block *sb, const char *function,
638                 unsigned int line, const char *fmt, ...)
639 {
640         va_list args;
641
642         save_error_info(sb, function, line);
643         va_start(args, fmt);
644         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
645                function, line);
646         vprintk(fmt, args);
647         printk("\n");
648         va_end(args);
649
650         if ((sb->s_flags & MS_RDONLY) == 0) {
651                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
652                 sb->s_flags |= MS_RDONLY;
653                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
654                 if (EXT4_SB(sb)->s_journal)
655                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
656                 save_error_info(sb, function, line);
657         }
658         if (test_opt(sb, ERRORS_PANIC))
659                 panic("EXT4-fs panic from previous error\n");
660 }
661
662 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
663 {
664         struct va_format vaf;
665         va_list args;
666
667         va_start(args, fmt);
668         vaf.fmt = fmt;
669         vaf.va = &args;
670         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
671         va_end(args);
672 }
673
674 void __ext4_warning(struct super_block *sb, const char *function,
675                     unsigned int line, const char *fmt, ...)
676 {
677         struct va_format vaf;
678         va_list args;
679
680         va_start(args, fmt);
681         vaf.fmt = fmt;
682         vaf.va = &args;
683         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
684                sb->s_id, function, line, &vaf);
685         va_end(args);
686 }
687
688 void __ext4_grp_locked_error(const char *function, unsigned int line,
689                              struct super_block *sb, ext4_group_t grp,
690                              unsigned long ino, ext4_fsblk_t block,
691                              const char *fmt, ...)
692 __releases(bitlock)
693 __acquires(bitlock)
694 {
695         struct va_format vaf;
696         va_list args;
697         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
698
699         es->s_last_error_ino = cpu_to_le32(ino);
700         es->s_last_error_block = cpu_to_le64(block);
701         __save_error_info(sb, function, line);
702
703         va_start(args, fmt);
704
705         vaf.fmt = fmt;
706         vaf.va = &args;
707         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
708                sb->s_id, function, line, grp);
709         if (ino)
710                 printk(KERN_CONT "inode %lu: ", ino);
711         if (block)
712                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
713         printk(KERN_CONT "%pV\n", &vaf);
714         va_end(args);
715
716         if (test_opt(sb, ERRORS_CONT)) {
717                 ext4_commit_super(sb, 0);
718                 return;
719         }
720
721         ext4_unlock_group(sb, grp);
722         ext4_handle_error(sb);
723         /*
724          * We only get here in the ERRORS_RO case; relocking the group
725          * may be dangerous, but nothing bad will happen since the
726          * filesystem will have already been marked read/only and the
727          * journal has been aborted.  We return 1 as a hint to callers
728          * who might what to use the return value from
729          * ext4_grp_locked_error() to distinguish between the
730          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
731          * aggressively from the ext4 function in question, with a
732          * more appropriate error code.
733          */
734         ext4_lock_group(sb, grp);
735         return;
736 }
737
738 void ext4_update_dynamic_rev(struct super_block *sb)
739 {
740         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
741
742         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
743                 return;
744
745         ext4_warning(sb,
746                      "updating to rev %d because of new feature flag, "
747                      "running e2fsck is recommended",
748                      EXT4_DYNAMIC_REV);
749
750         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
751         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
752         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
753         /* leave es->s_feature_*compat flags alone */
754         /* es->s_uuid will be set by e2fsck if empty */
755
756         /*
757          * The rest of the superblock fields should be zero, and if not it
758          * means they are likely already in use, so leave them alone.  We
759          * can leave it up to e2fsck to clean up any inconsistencies there.
760          */
761 }
762
763 /*
764  * Open the external journal device
765  */
766 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
767 {
768         struct block_device *bdev;
769         char b[BDEVNAME_SIZE];
770
771         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
772         if (IS_ERR(bdev))
773                 goto fail;
774         return bdev;
775
776 fail:
777         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
778                         __bdevname(dev, b), PTR_ERR(bdev));
779         return NULL;
780 }
781
782 /*
783  * Release the journal device
784  */
785 static int ext4_blkdev_put(struct block_device *bdev)
786 {
787         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
788 }
789
790 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
791 {
792         struct block_device *bdev;
793         int ret = -ENODEV;
794
795         bdev = sbi->journal_bdev;
796         if (bdev) {
797                 ret = ext4_blkdev_put(bdev);
798                 sbi->journal_bdev = NULL;
799         }
800         return ret;
801 }
802
803 static inline struct inode *orphan_list_entry(struct list_head *l)
804 {
805         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
806 }
807
808 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
809 {
810         struct list_head *l;
811
812         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
813                  le32_to_cpu(sbi->s_es->s_last_orphan));
814
815         printk(KERN_ERR "sb_info orphan list:\n");
816         list_for_each(l, &sbi->s_orphan) {
817                 struct inode *inode = orphan_list_entry(l);
818                 printk(KERN_ERR "  "
819                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
820                        inode->i_sb->s_id, inode->i_ino, inode,
821                        inode->i_mode, inode->i_nlink,
822                        NEXT_ORPHAN(inode));
823         }
824 }
825
826 static void ext4_put_super(struct super_block *sb)
827 {
828         struct ext4_sb_info *sbi = EXT4_SB(sb);
829         struct ext4_super_block *es = sbi->s_es;
830         int i, err;
831
832         ext4_unregister_li_request(sb);
833         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
834
835         flush_workqueue(sbi->dio_unwritten_wq);
836         destroy_workqueue(sbi->dio_unwritten_wq);
837
838         lock_super(sb);
839         if (sbi->s_journal) {
840                 err = jbd2_journal_destroy(sbi->s_journal);
841                 sbi->s_journal = NULL;
842                 if (err < 0)
843                         ext4_abort(sb, "Couldn't clean up the journal");
844         }
845
846         del_timer(&sbi->s_err_report);
847         ext4_release_system_zone(sb);
848         ext4_mb_release(sb);
849         ext4_ext_release(sb);
850         ext4_xattr_put_super(sb);
851
852         if (!(sb->s_flags & MS_RDONLY)) {
853                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
854                 es->s_state = cpu_to_le16(sbi->s_mount_state);
855         }
856         if (sb->s_dirt || !(sb->s_flags & MS_RDONLY))
857                 ext4_commit_super(sb, 1);
858
859         if (sbi->s_proc) {
860                 remove_proc_entry("options", sbi->s_proc);
861                 remove_proc_entry(sb->s_id, ext4_proc_root);
862         }
863         kobject_del(&sbi->s_kobj);
864
865         for (i = 0; i < sbi->s_gdb_count; i++)
866                 brelse(sbi->s_group_desc[i]);
867         ext4_kvfree(sbi->s_group_desc);
868         ext4_kvfree(sbi->s_flex_groups);
869         percpu_counter_destroy(&sbi->s_freeclusters_counter);
870         percpu_counter_destroy(&sbi->s_freeinodes_counter);
871         percpu_counter_destroy(&sbi->s_dirs_counter);
872         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
873         brelse(sbi->s_sbh);
874 #ifdef CONFIG_QUOTA
875         for (i = 0; i < MAXQUOTAS; i++)
876                 kfree(sbi->s_qf_names[i]);
877 #endif
878
879         /* Debugging code just in case the in-memory inode orphan list
880          * isn't empty.  The on-disk one can be non-empty if we've
881          * detected an error and taken the fs readonly, but the
882          * in-memory list had better be clean by this point. */
883         if (!list_empty(&sbi->s_orphan))
884                 dump_orphan_list(sb, sbi);
885         J_ASSERT(list_empty(&sbi->s_orphan));
886
887         invalidate_bdev(sb->s_bdev);
888         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
889                 /*
890                  * Invalidate the journal device's buffers.  We don't want them
891                  * floating about in memory - the physical journal device may
892                  * hotswapped, and it breaks the `ro-after' testing code.
893                  */
894                 sync_blockdev(sbi->journal_bdev);
895                 invalidate_bdev(sbi->journal_bdev);
896                 ext4_blkdev_remove(sbi);
897         }
898         if (sbi->s_mmp_tsk)
899                 kthread_stop(sbi->s_mmp_tsk);
900         sb->s_fs_info = NULL;
901         /*
902          * Now that we are completely done shutting down the
903          * superblock, we need to actually destroy the kobject.
904          */
905         unlock_super(sb);
906         kobject_put(&sbi->s_kobj);
907         wait_for_completion(&sbi->s_kobj_unregister);
908         kfree(sbi->s_blockgroup_lock);
909         kfree(sbi);
910 }
911
912 static struct kmem_cache *ext4_inode_cachep;
913
914 /*
915  * Called inside transaction, so use GFP_NOFS
916  */
917 static struct inode *ext4_alloc_inode(struct super_block *sb)
918 {
919         struct ext4_inode_info *ei;
920
921         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
922         if (!ei)
923                 return NULL;
924
925         ei->vfs_inode.i_version = 1;
926         ei->vfs_inode.i_data.writeback_index = 0;
927         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
928         INIT_LIST_HEAD(&ei->i_prealloc_list);
929         spin_lock_init(&ei->i_prealloc_lock);
930         ei->i_reserved_data_blocks = 0;
931         ei->i_reserved_meta_blocks = 0;
932         ei->i_allocated_meta_blocks = 0;
933         ei->i_da_metadata_calc_len = 0;
934         spin_lock_init(&(ei->i_block_reservation_lock));
935 #ifdef CONFIG_QUOTA
936         ei->i_reserved_quota = 0;
937 #endif
938         ei->jinode = NULL;
939         INIT_LIST_HEAD(&ei->i_completed_io_list);
940         spin_lock_init(&ei->i_completed_io_lock);
941         ei->cur_aio_dio = NULL;
942         ei->i_sync_tid = 0;
943         ei->i_datasync_tid = 0;
944         atomic_set(&ei->i_ioend_count, 0);
945         atomic_set(&ei->i_aiodio_unwritten, 0);
946
947         return &ei->vfs_inode;
948 }
949
950 static int ext4_drop_inode(struct inode *inode)
951 {
952         int drop = generic_drop_inode(inode);
953
954         trace_ext4_drop_inode(inode, drop);
955         return drop;
956 }
957
958 static void ext4_i_callback(struct rcu_head *head)
959 {
960         struct inode *inode = container_of(head, struct inode, i_rcu);
961         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
962 }
963
964 static void ext4_destroy_inode(struct inode *inode)
965 {
966         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
967                 ext4_msg(inode->i_sb, KERN_ERR,
968                          "Inode %lu (%p): orphan list check failed!",
969                          inode->i_ino, EXT4_I(inode));
970                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
971                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
972                                 true);
973                 dump_stack();
974         }
975         call_rcu(&inode->i_rcu, ext4_i_callback);
976 }
977
978 static void init_once(void *foo)
979 {
980         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
981
982         INIT_LIST_HEAD(&ei->i_orphan);
983 #ifdef CONFIG_EXT4_FS_XATTR
984         init_rwsem(&ei->xattr_sem);
985 #endif
986         init_rwsem(&ei->i_data_sem);
987         inode_init_once(&ei->vfs_inode);
988 }
989
990 static int init_inodecache(void)
991 {
992         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
993                                              sizeof(struct ext4_inode_info),
994                                              0, (SLAB_RECLAIM_ACCOUNT|
995                                                 SLAB_MEM_SPREAD),
996                                              init_once);
997         if (ext4_inode_cachep == NULL)
998                 return -ENOMEM;
999         return 0;
1000 }
1001
1002 static void destroy_inodecache(void)
1003 {
1004         kmem_cache_destroy(ext4_inode_cachep);
1005 }
1006
1007 void ext4_clear_inode(struct inode *inode)
1008 {
1009         invalidate_inode_buffers(inode);
1010         clear_inode(inode);
1011         dquot_drop(inode);
1012         ext4_discard_preallocations(inode);
1013         if (EXT4_I(inode)->jinode) {
1014                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1015                                                EXT4_I(inode)->jinode);
1016                 jbd2_free_inode(EXT4_I(inode)->jinode);
1017                 EXT4_I(inode)->jinode = NULL;
1018         }
1019 }
1020
1021 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1022                                         u64 ino, u32 generation)
1023 {
1024         struct inode *inode;
1025
1026         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1027                 return ERR_PTR(-ESTALE);
1028         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1029                 return ERR_PTR(-ESTALE);
1030
1031         /* iget isn't really right if the inode is currently unallocated!!
1032          *
1033          * ext4_read_inode will return a bad_inode if the inode had been
1034          * deleted, so we should be safe.
1035          *
1036          * Currently we don't know the generation for parent directory, so
1037          * a generation of 0 means "accept any"
1038          */
1039         inode = ext4_iget(sb, ino);
1040         if (IS_ERR(inode))
1041                 return ERR_CAST(inode);
1042         if (generation && inode->i_generation != generation) {
1043                 iput(inode);
1044                 return ERR_PTR(-ESTALE);
1045         }
1046
1047         return inode;
1048 }
1049
1050 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1051                                         int fh_len, int fh_type)
1052 {
1053         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1054                                     ext4_nfs_get_inode);
1055 }
1056
1057 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1058                                         int fh_len, int fh_type)
1059 {
1060         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1061                                     ext4_nfs_get_inode);
1062 }
1063
1064 /*
1065  * Try to release metadata pages (indirect blocks, directories) which are
1066  * mapped via the block device.  Since these pages could have journal heads
1067  * which would prevent try_to_free_buffers() from freeing them, we must use
1068  * jbd2 layer's try_to_free_buffers() function to release them.
1069  */
1070 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1071                                  gfp_t wait)
1072 {
1073         journal_t *journal = EXT4_SB(sb)->s_journal;
1074
1075         WARN_ON(PageChecked(page));
1076         if (!page_has_buffers(page))
1077                 return 0;
1078         if (journal)
1079                 return jbd2_journal_try_to_free_buffers(journal, page,
1080                                                         wait & ~__GFP_WAIT);
1081         return try_to_free_buffers(page);
1082 }
1083
1084 #ifdef CONFIG_QUOTA
1085 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1086 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1087
1088 static int ext4_write_dquot(struct dquot *dquot);
1089 static int ext4_acquire_dquot(struct dquot *dquot);
1090 static int ext4_release_dquot(struct dquot *dquot);
1091 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1092 static int ext4_write_info(struct super_block *sb, int type);
1093 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1094                          struct path *path);
1095 static int ext4_quota_off(struct super_block *sb, int type);
1096 static int ext4_quota_on_mount(struct super_block *sb, int type);
1097 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1098                                size_t len, loff_t off);
1099 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1100                                 const char *data, size_t len, loff_t off);
1101
1102 static const struct dquot_operations ext4_quota_operations = {
1103         .get_reserved_space = ext4_get_reserved_space,
1104         .write_dquot    = ext4_write_dquot,
1105         .acquire_dquot  = ext4_acquire_dquot,
1106         .release_dquot  = ext4_release_dquot,
1107         .mark_dirty     = ext4_mark_dquot_dirty,
1108         .write_info     = ext4_write_info,
1109         .alloc_dquot    = dquot_alloc,
1110         .destroy_dquot  = dquot_destroy,
1111 };
1112
1113 static const struct quotactl_ops ext4_qctl_operations = {
1114         .quota_on       = ext4_quota_on,
1115         .quota_off      = ext4_quota_off,
1116         .quota_sync     = dquot_quota_sync,
1117         .get_info       = dquot_get_dqinfo,
1118         .set_info       = dquot_set_dqinfo,
1119         .get_dqblk      = dquot_get_dqblk,
1120         .set_dqblk      = dquot_set_dqblk
1121 };
1122 #endif
1123
1124 static const struct super_operations ext4_sops = {
1125         .alloc_inode    = ext4_alloc_inode,
1126         .destroy_inode  = ext4_destroy_inode,
1127         .write_inode    = ext4_write_inode,
1128         .dirty_inode    = ext4_dirty_inode,
1129         .drop_inode     = ext4_drop_inode,
1130         .evict_inode    = ext4_evict_inode,
1131         .put_super      = ext4_put_super,
1132         .sync_fs        = ext4_sync_fs,
1133         .freeze_fs      = ext4_freeze,
1134         .unfreeze_fs    = ext4_unfreeze,
1135         .statfs         = ext4_statfs,
1136         .remount_fs     = ext4_remount,
1137         .show_options   = ext4_show_options,
1138 #ifdef CONFIG_QUOTA
1139         .quota_read     = ext4_quota_read,
1140         .quota_write    = ext4_quota_write,
1141 #endif
1142         .bdev_try_to_free_page = bdev_try_to_free_page,
1143 };
1144
1145 static const struct super_operations ext4_nojournal_sops = {
1146         .alloc_inode    = ext4_alloc_inode,
1147         .destroy_inode  = ext4_destroy_inode,
1148         .write_inode    = ext4_write_inode,
1149         .dirty_inode    = ext4_dirty_inode,
1150         .drop_inode     = ext4_drop_inode,
1151         .evict_inode    = ext4_evict_inode,
1152         .write_super    = ext4_write_super,
1153         .put_super      = ext4_put_super,
1154         .statfs         = ext4_statfs,
1155         .remount_fs     = ext4_remount,
1156         .show_options   = ext4_show_options,
1157 #ifdef CONFIG_QUOTA
1158         .quota_read     = ext4_quota_read,
1159         .quota_write    = ext4_quota_write,
1160 #endif
1161         .bdev_try_to_free_page = bdev_try_to_free_page,
1162 };
1163
1164 static const struct export_operations ext4_export_ops = {
1165         .fh_to_dentry = ext4_fh_to_dentry,
1166         .fh_to_parent = ext4_fh_to_parent,
1167         .get_parent = ext4_get_parent,
1168 };
1169
1170 enum {
1171         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1172         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1173         Opt_nouid32, Opt_debug, Opt_removed,
1174         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1175         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1176         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1177         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1178         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1179         Opt_data_err_abort, Opt_data_err_ignore,
1180         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1181         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1182         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1183         Opt_usrquota, Opt_grpquota, Opt_i_version,
1184         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1185         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1186         Opt_inode_readahead_blks, Opt_journal_ioprio,
1187         Opt_dioread_nolock, Opt_dioread_lock,
1188         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1189 };
1190
1191 static const match_table_t tokens = {
1192         {Opt_bsd_df, "bsddf"},
1193         {Opt_minix_df, "minixdf"},
1194         {Opt_grpid, "grpid"},
1195         {Opt_grpid, "bsdgroups"},
1196         {Opt_nogrpid, "nogrpid"},
1197         {Opt_nogrpid, "sysvgroups"},
1198         {Opt_resgid, "resgid=%u"},
1199         {Opt_resuid, "resuid=%u"},
1200         {Opt_sb, "sb=%u"},
1201         {Opt_err_cont, "errors=continue"},
1202         {Opt_err_panic, "errors=panic"},
1203         {Opt_err_ro, "errors=remount-ro"},
1204         {Opt_nouid32, "nouid32"},
1205         {Opt_debug, "debug"},
1206         {Opt_removed, "oldalloc"},
1207         {Opt_removed, "orlov"},
1208         {Opt_user_xattr, "user_xattr"},
1209         {Opt_nouser_xattr, "nouser_xattr"},
1210         {Opt_acl, "acl"},
1211         {Opt_noacl, "noacl"},
1212         {Opt_noload, "norecovery"},
1213         {Opt_noload, "noload"},
1214         {Opt_removed, "nobh"},
1215         {Opt_removed, "bh"},
1216         {Opt_commit, "commit=%u"},
1217         {Opt_min_batch_time, "min_batch_time=%u"},
1218         {Opt_max_batch_time, "max_batch_time=%u"},
1219         {Opt_journal_dev, "journal_dev=%u"},
1220         {Opt_journal_checksum, "journal_checksum"},
1221         {Opt_journal_async_commit, "journal_async_commit"},
1222         {Opt_abort, "abort"},
1223         {Opt_data_journal, "data=journal"},
1224         {Opt_data_ordered, "data=ordered"},
1225         {Opt_data_writeback, "data=writeback"},
1226         {Opt_data_err_abort, "data_err=abort"},
1227         {Opt_data_err_ignore, "data_err=ignore"},
1228         {Opt_offusrjquota, "usrjquota="},
1229         {Opt_usrjquota, "usrjquota=%s"},
1230         {Opt_offgrpjquota, "grpjquota="},
1231         {Opt_grpjquota, "grpjquota=%s"},
1232         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1233         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1234         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1235         {Opt_grpquota, "grpquota"},
1236         {Opt_noquota, "noquota"},
1237         {Opt_quota, "quota"},
1238         {Opt_usrquota, "usrquota"},
1239         {Opt_barrier, "barrier=%u"},
1240         {Opt_barrier, "barrier"},
1241         {Opt_nobarrier, "nobarrier"},
1242         {Opt_i_version, "i_version"},
1243         {Opt_stripe, "stripe=%u"},
1244         {Opt_delalloc, "delalloc"},
1245         {Opt_nodelalloc, "nodelalloc"},
1246         {Opt_mblk_io_submit, "mblk_io_submit"},
1247         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1248         {Opt_block_validity, "block_validity"},
1249         {Opt_noblock_validity, "noblock_validity"},
1250         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1251         {Opt_journal_ioprio, "journal_ioprio=%u"},
1252         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1253         {Opt_auto_da_alloc, "auto_da_alloc"},
1254         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1255         {Opt_dioread_nolock, "dioread_nolock"},
1256         {Opt_dioread_lock, "dioread_lock"},
1257         {Opt_discard, "discard"},
1258         {Opt_nodiscard, "nodiscard"},
1259         {Opt_init_itable, "init_itable=%u"},
1260         {Opt_init_itable, "init_itable"},
1261         {Opt_noinit_itable, "noinit_itable"},
1262         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1263         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1264         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1265         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1266         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1267         {Opt_err, NULL},
1268 };
1269
1270 static ext4_fsblk_t get_sb_block(void **data)
1271 {
1272         ext4_fsblk_t    sb_block;
1273         char            *options = (char *) *data;
1274
1275         if (!options || strncmp(options, "sb=", 3) != 0)
1276                 return 1;       /* Default location */
1277
1278         options += 3;
1279         /* TODO: use simple_strtoll with >32bit ext4 */
1280         sb_block = simple_strtoul(options, &options, 0);
1281         if (*options && *options != ',') {
1282                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1283                        (char *) *data);
1284                 return 1;
1285         }
1286         if (*options == ',')
1287                 options++;
1288         *data = (void *) options;
1289
1290         return sb_block;
1291 }
1292
1293 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1294 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1295         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1296
1297 #ifdef CONFIG_QUOTA
1298 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1299 {
1300         struct ext4_sb_info *sbi = EXT4_SB(sb);
1301         char *qname;
1302
1303         if (sb_any_quota_loaded(sb) &&
1304                 !sbi->s_qf_names[qtype]) {
1305                 ext4_msg(sb, KERN_ERR,
1306                         "Cannot change journaled "
1307                         "quota options when quota turned on");
1308                 return -1;
1309         }
1310         qname = match_strdup(args);
1311         if (!qname) {
1312                 ext4_msg(sb, KERN_ERR,
1313                         "Not enough memory for storing quotafile name");
1314                 return -1;
1315         }
1316         if (sbi->s_qf_names[qtype] &&
1317                 strcmp(sbi->s_qf_names[qtype], qname)) {
1318                 ext4_msg(sb, KERN_ERR,
1319                         "%s quota file already specified", QTYPE2NAME(qtype));
1320                 kfree(qname);
1321                 return -1;
1322         }
1323         sbi->s_qf_names[qtype] = qname;
1324         if (strchr(sbi->s_qf_names[qtype], '/')) {
1325                 ext4_msg(sb, KERN_ERR,
1326                         "quotafile must be on filesystem root");
1327                 kfree(sbi->s_qf_names[qtype]);
1328                 sbi->s_qf_names[qtype] = NULL;
1329                 return -1;
1330         }
1331         set_opt(sb, QUOTA);
1332         return 1;
1333 }
1334
1335 static int clear_qf_name(struct super_block *sb, int qtype)
1336 {
1337
1338         struct ext4_sb_info *sbi = EXT4_SB(sb);
1339
1340         if (sb_any_quota_loaded(sb) &&
1341                 sbi->s_qf_names[qtype]) {
1342                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1343                         " when quota turned on");
1344                 return -1;
1345         }
1346         /*
1347          * The space will be released later when all options are confirmed
1348          * to be correct
1349          */
1350         sbi->s_qf_names[qtype] = NULL;
1351         return 1;
1352 }
1353 #endif
1354
1355 #define MOPT_SET        0x0001
1356 #define MOPT_CLEAR      0x0002
1357 #define MOPT_NOSUPPORT  0x0004
1358 #define MOPT_EXPLICIT   0x0008
1359 #define MOPT_CLEAR_ERR  0x0010
1360 #define MOPT_GTE0       0x0020
1361 #ifdef CONFIG_QUOTA
1362 #define MOPT_Q          0
1363 #define MOPT_QFMT       0x0040
1364 #else
1365 #define MOPT_Q          MOPT_NOSUPPORT
1366 #define MOPT_QFMT       MOPT_NOSUPPORT
1367 #endif
1368 #define MOPT_DATAJ      0x0080
1369
1370 static const struct mount_opts {
1371         int     token;
1372         int     mount_opt;
1373         int     flags;
1374 } ext4_mount_opts[] = {
1375         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1376         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1377         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1378         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1379         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1380         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1381         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1382         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1383         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1384         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1385         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1386         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1387         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1388         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1389         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1390         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1391                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1392         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1393         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1394         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1395         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1396         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1397         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1398         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1399         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1400         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1401         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1402         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1403         {Opt_commit, 0, MOPT_GTE0},
1404         {Opt_max_batch_time, 0, MOPT_GTE0},
1405         {Opt_min_batch_time, 0, MOPT_GTE0},
1406         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1407         {Opt_init_itable, 0, MOPT_GTE0},
1408         {Opt_stripe, 0, MOPT_GTE0},
1409         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1410         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1411         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1412 #ifdef CONFIG_EXT4_FS_XATTR
1413         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1414         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1415 #else
1416         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1417         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1418 #endif
1419 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1420         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1421         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1422 #else
1423         {Opt_acl, 0, MOPT_NOSUPPORT},
1424         {Opt_noacl, 0, MOPT_NOSUPPORT},
1425 #endif
1426         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1427         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1428         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1429         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1430                                                         MOPT_SET | MOPT_Q},
1431         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1432                                                         MOPT_SET | MOPT_Q},
1433         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1434                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1435         {Opt_usrjquota, 0, MOPT_Q},
1436         {Opt_grpjquota, 0, MOPT_Q},
1437         {Opt_offusrjquota, 0, MOPT_Q},
1438         {Opt_offgrpjquota, 0, MOPT_Q},
1439         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1440         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1441         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1442         {Opt_err, 0, 0}
1443 };
1444
1445 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1446                             substring_t *args, unsigned long *journal_devnum,
1447                             unsigned int *journal_ioprio, int is_remount)
1448 {
1449         struct ext4_sb_info *sbi = EXT4_SB(sb);
1450         const struct mount_opts *m;
1451         kuid_t uid;
1452         kgid_t gid;
1453         int arg = 0;
1454
1455 #ifdef CONFIG_QUOTA
1456         if (token == Opt_usrjquota)
1457                 return set_qf_name(sb, USRQUOTA, &args[0]);
1458         else if (token == Opt_grpjquota)
1459                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1460         else if (token == Opt_offusrjquota)
1461                 return clear_qf_name(sb, USRQUOTA);
1462         else if (token == Opt_offgrpjquota)
1463                 return clear_qf_name(sb, GRPQUOTA);
1464 #endif
1465         if (args->from && match_int(args, &arg))
1466                 return -1;
1467         switch (token) {
1468         case Opt_noacl:
1469         case Opt_nouser_xattr:
1470                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1471                 break;
1472         case Opt_sb:
1473                 return 1;       /* handled by get_sb_block() */
1474         case Opt_removed:
1475                 ext4_msg(sb, KERN_WARNING,
1476                          "Ignoring removed %s option", opt);
1477                 return 1;
1478         case Opt_resuid:
1479                 uid = make_kuid(current_user_ns(), arg);
1480                 if (!uid_valid(uid)) {
1481                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1482                         return -1;
1483                 }
1484                 sbi->s_resuid = uid;
1485                 return 1;
1486         case Opt_resgid:
1487                 gid = make_kgid(current_user_ns(), arg);
1488                 if (!gid_valid(gid)) {
1489                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1490                         return -1;
1491                 }
1492                 sbi->s_resgid = gid;
1493                 return 1;
1494         case Opt_abort:
1495                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1496                 return 1;
1497         case Opt_i_version:
1498                 sb->s_flags |= MS_I_VERSION;
1499                 return 1;
1500         case Opt_journal_dev:
1501                 if (is_remount) {
1502                         ext4_msg(sb, KERN_ERR,
1503                                  "Cannot specify journal on remount");
1504                         return -1;
1505                 }
1506                 *journal_devnum = arg;
1507                 return 1;
1508         case Opt_journal_ioprio:
1509                 if (arg < 0 || arg > 7)
1510                         return -1;
1511                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1512                 return 1;
1513         }
1514
1515         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1516                 if (token != m->token)
1517                         continue;
1518                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1519                         return -1;
1520                 if (m->flags & MOPT_EXPLICIT)
1521                         set_opt2(sb, EXPLICIT_DELALLOC);
1522                 if (m->flags & MOPT_CLEAR_ERR)
1523                         clear_opt(sb, ERRORS_MASK);
1524                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1525                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1526                                  "options when quota turned on");
1527                         return -1;
1528                 }
1529
1530                 if (m->flags & MOPT_NOSUPPORT) {
1531                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1532                 } else if (token == Opt_commit) {
1533                         if (arg == 0)
1534                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1535                         sbi->s_commit_interval = HZ * arg;
1536                 } else if (token == Opt_max_batch_time) {
1537                         if (arg == 0)
1538                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1539                         sbi->s_max_batch_time = arg;
1540                 } else if (token == Opt_min_batch_time) {
1541                         sbi->s_min_batch_time = arg;
1542                 } else if (token == Opt_inode_readahead_blks) {
1543                         if (arg > (1 << 30))
1544                                 return -1;
1545                         if (arg && !is_power_of_2(arg)) {
1546                                 ext4_msg(sb, KERN_ERR,
1547                                          "EXT4-fs: inode_readahead_blks"
1548                                          " must be a power of 2");
1549                                 return -1;
1550                         }
1551                         sbi->s_inode_readahead_blks = arg;
1552                 } else if (token == Opt_init_itable) {
1553                         set_opt(sb, INIT_INODE_TABLE);
1554                         if (!args->from)
1555                                 arg = EXT4_DEF_LI_WAIT_MULT;
1556                         sbi->s_li_wait_mult = arg;
1557                 } else if (token == Opt_stripe) {
1558                         sbi->s_stripe = arg;
1559                 } else if (m->flags & MOPT_DATAJ) {
1560                         if (is_remount) {
1561                                 if (!sbi->s_journal)
1562                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1563                                 else if (test_opt(sb, DATA_FLAGS) !=
1564                                          m->mount_opt) {
1565                                         ext4_msg(sb, KERN_ERR,
1566                                          "Cannot change data mode on remount");
1567                                         return -1;
1568                                 }
1569                         } else {
1570                                 clear_opt(sb, DATA_FLAGS);
1571                                 sbi->s_mount_opt |= m->mount_opt;
1572                         }
1573 #ifdef CONFIG_QUOTA
1574                 } else if (m->flags & MOPT_QFMT) {
1575                         if (sb_any_quota_loaded(sb) &&
1576                             sbi->s_jquota_fmt != m->mount_opt) {
1577                                 ext4_msg(sb, KERN_ERR, "Cannot "
1578                                          "change journaled quota options "
1579                                          "when quota turned on");
1580                                 return -1;
1581                         }
1582                         sbi->s_jquota_fmt = m->mount_opt;
1583 #endif
1584                 } else {
1585                         if (!args->from)
1586                                 arg = 1;
1587                         if (m->flags & MOPT_CLEAR)
1588                                 arg = !arg;
1589                         else if (unlikely(!(m->flags & MOPT_SET))) {
1590                                 ext4_msg(sb, KERN_WARNING,
1591                                          "buggy handling of option %s", opt);
1592                                 WARN_ON(1);
1593                                 return -1;
1594                         }
1595                         if (arg != 0)
1596                                 sbi->s_mount_opt |= m->mount_opt;
1597                         else
1598                                 sbi->s_mount_opt &= ~m->mount_opt;
1599                 }
1600                 return 1;
1601         }
1602         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1603                  "or missing value", opt);
1604         return -1;
1605 }
1606
1607 static int parse_options(char *options, struct super_block *sb,
1608                          unsigned long *journal_devnum,
1609                          unsigned int *journal_ioprio,
1610                          int is_remount)
1611 {
1612 #ifdef CONFIG_QUOTA
1613         struct ext4_sb_info *sbi = EXT4_SB(sb);
1614 #endif
1615         char *p;
1616         substring_t args[MAX_OPT_ARGS];
1617         int token;
1618
1619         if (!options)
1620                 return 1;
1621
1622         while ((p = strsep(&options, ",")) != NULL) {
1623                 if (!*p)
1624                         continue;
1625                 /*
1626                  * Initialize args struct so we know whether arg was
1627                  * found; some options take optional arguments.
1628                  */
1629                 args[0].to = args[0].from = 0;
1630                 token = match_token(p, tokens, args);
1631                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1632                                      journal_ioprio, is_remount) < 0)
1633                         return 0;
1634         }
1635 #ifdef CONFIG_QUOTA
1636         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1637                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1638                         clear_opt(sb, USRQUOTA);
1639
1640                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1641                         clear_opt(sb, GRPQUOTA);
1642
1643                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1644                         ext4_msg(sb, KERN_ERR, "old and new quota "
1645                                         "format mixing");
1646                         return 0;
1647                 }
1648
1649                 if (!sbi->s_jquota_fmt) {
1650                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1651                                         "not specified");
1652                         return 0;
1653                 }
1654         } else {
1655                 if (sbi->s_jquota_fmt) {
1656                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1657                                         "specified with no journaling "
1658                                         "enabled");
1659                         return 0;
1660                 }
1661         }
1662 #endif
1663         return 1;
1664 }
1665
1666 static inline void ext4_show_quota_options(struct seq_file *seq,
1667                                            struct super_block *sb)
1668 {
1669 #if defined(CONFIG_QUOTA)
1670         struct ext4_sb_info *sbi = EXT4_SB(sb);
1671
1672         if (sbi->s_jquota_fmt) {
1673                 char *fmtname = "";
1674
1675                 switch (sbi->s_jquota_fmt) {
1676                 case QFMT_VFS_OLD:
1677                         fmtname = "vfsold";
1678                         break;
1679                 case QFMT_VFS_V0:
1680                         fmtname = "vfsv0";
1681                         break;
1682                 case QFMT_VFS_V1:
1683                         fmtname = "vfsv1";
1684                         break;
1685                 }
1686                 seq_printf(seq, ",jqfmt=%s", fmtname);
1687         }
1688
1689         if (sbi->s_qf_names[USRQUOTA])
1690                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1691
1692         if (sbi->s_qf_names[GRPQUOTA])
1693                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1694
1695         if (test_opt(sb, USRQUOTA))
1696                 seq_puts(seq, ",usrquota");
1697
1698         if (test_opt(sb, GRPQUOTA))
1699                 seq_puts(seq, ",grpquota");
1700 #endif
1701 }
1702
1703 static const char *token2str(int token)
1704 {
1705         static const struct match_token *t;
1706
1707         for (t = tokens; t->token != Opt_err; t++)
1708                 if (t->token == token && !strchr(t->pattern, '='))
1709                         break;
1710         return t->pattern;
1711 }
1712
1713 /*
1714  * Show an option if
1715  *  - it's set to a non-default value OR
1716  *  - if the per-sb default is different from the global default
1717  */
1718 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1719                               int nodefs)
1720 {
1721         struct ext4_sb_info *sbi = EXT4_SB(sb);
1722         struct ext4_super_block *es = sbi->s_es;
1723         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1724         const struct mount_opts *m;
1725         char sep = nodefs ? '\n' : ',';
1726
1727 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1728 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1729
1730         if (sbi->s_sb_block != 1)
1731                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1732
1733         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1734                 int want_set = m->flags & MOPT_SET;
1735                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1736                     (m->flags & MOPT_CLEAR_ERR))
1737                         continue;
1738                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1739                         continue; /* skip if same as the default */
1740                 if ((want_set &&
1741                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1742                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1743                         continue; /* select Opt_noFoo vs Opt_Foo */
1744                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1745         }
1746
1747         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1748             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1749                 SEQ_OPTS_PRINT("resuid=%u",
1750                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1751         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1752             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1753                 SEQ_OPTS_PRINT("resgid=%u",
1754                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1755         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1756         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1757                 SEQ_OPTS_PUTS("errors=remount-ro");
1758         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1759                 SEQ_OPTS_PUTS("errors=continue");
1760         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1761                 SEQ_OPTS_PUTS("errors=panic");
1762         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1763                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1764         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1765                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1766         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1767                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1768         if (sb->s_flags & MS_I_VERSION)
1769                 SEQ_OPTS_PUTS("i_version");
1770         if (nodefs || sbi->s_stripe)
1771                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1772         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1773                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1774                         SEQ_OPTS_PUTS("data=journal");
1775                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1776                         SEQ_OPTS_PUTS("data=ordered");
1777                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1778                         SEQ_OPTS_PUTS("data=writeback");
1779         }
1780         if (nodefs ||
1781             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1782                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1783                                sbi->s_inode_readahead_blks);
1784
1785         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1786                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1787                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1788
1789         ext4_show_quota_options(seq, sb);
1790         return 0;
1791 }
1792
1793 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1794 {
1795         return _ext4_show_options(seq, root->d_sb, 0);
1796 }
1797
1798 static int options_seq_show(struct seq_file *seq, void *offset)
1799 {
1800         struct super_block *sb = seq->private;
1801         int rc;
1802
1803         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1804         rc = _ext4_show_options(seq, sb, 1);
1805         seq_puts(seq, "\n");
1806         return rc;
1807 }
1808
1809 static int options_open_fs(struct inode *inode, struct file *file)
1810 {
1811         return single_open(file, options_seq_show, PDE(inode)->data);
1812 }
1813
1814 static const struct file_operations ext4_seq_options_fops = {
1815         .owner = THIS_MODULE,
1816         .open = options_open_fs,
1817         .read = seq_read,
1818         .llseek = seq_lseek,
1819         .release = single_release,
1820 };
1821
1822 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1823                             int read_only)
1824 {
1825         struct ext4_sb_info *sbi = EXT4_SB(sb);
1826         int res = 0;
1827
1828         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1829                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1830                          "forcing read-only mode");
1831                 res = MS_RDONLY;
1832         }
1833         if (read_only)
1834                 goto done;
1835         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1836                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1837                          "running e2fsck is recommended");
1838         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1839                 ext4_msg(sb, KERN_WARNING,
1840                          "warning: mounting fs with errors, "
1841                          "running e2fsck is recommended");
1842         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1843                  le16_to_cpu(es->s_mnt_count) >=
1844                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1845                 ext4_msg(sb, KERN_WARNING,
1846                          "warning: maximal mount count reached, "
1847                          "running e2fsck is recommended");
1848         else if (le32_to_cpu(es->s_checkinterval) &&
1849                 (le32_to_cpu(es->s_lastcheck) +
1850                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1851                 ext4_msg(sb, KERN_WARNING,
1852                          "warning: checktime reached, "
1853                          "running e2fsck is recommended");
1854         if (!sbi->s_journal)
1855                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1856         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1857                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1858         le16_add_cpu(&es->s_mnt_count, 1);
1859         es->s_mtime = cpu_to_le32(get_seconds());
1860         ext4_update_dynamic_rev(sb);
1861         if (sbi->s_journal)
1862                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1863
1864         ext4_commit_super(sb, 1);
1865 done:
1866         if (test_opt(sb, DEBUG))
1867                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1868                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1869                         sb->s_blocksize,
1870                         sbi->s_groups_count,
1871                         EXT4_BLOCKS_PER_GROUP(sb),
1872                         EXT4_INODES_PER_GROUP(sb),
1873                         sbi->s_mount_opt, sbi->s_mount_opt2);
1874
1875         cleancache_init_fs(sb);
1876         return res;
1877 }
1878
1879 static int ext4_fill_flex_info(struct super_block *sb)
1880 {
1881         struct ext4_sb_info *sbi = EXT4_SB(sb);
1882         struct ext4_group_desc *gdp = NULL;
1883         ext4_group_t flex_group_count;
1884         ext4_group_t flex_group;
1885         unsigned int groups_per_flex = 0;
1886         size_t size;
1887         int i;
1888
1889         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1890         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1891                 sbi->s_log_groups_per_flex = 0;
1892                 return 1;
1893         }
1894         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1895
1896         /* We allocate both existing and potentially added groups */
1897         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1898                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1899                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1900         size = flex_group_count * sizeof(struct flex_groups);
1901         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1902         if (sbi->s_flex_groups == NULL) {
1903                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1904                          flex_group_count);
1905                 goto failed;
1906         }
1907
1908         for (i = 0; i < sbi->s_groups_count; i++) {
1909                 gdp = ext4_get_group_desc(sb, i, NULL);
1910
1911                 flex_group = ext4_flex_group(sbi, i);
1912                 atomic_add(ext4_free_inodes_count(sb, gdp),
1913                            &sbi->s_flex_groups[flex_group].free_inodes);
1914                 atomic_add(ext4_free_group_clusters(sb, gdp),
1915                            &sbi->s_flex_groups[flex_group].free_clusters);
1916                 atomic_add(ext4_used_dirs_count(sb, gdp),
1917                            &sbi->s_flex_groups[flex_group].used_dirs);
1918         }
1919
1920         return 1;
1921 failed:
1922         return 0;
1923 }
1924
1925 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1926                             struct ext4_group_desc *gdp)
1927 {
1928         __u16 crc = 0;
1929
1930         if (sbi->s_es->s_feature_ro_compat &
1931             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1932                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1933                 __le32 le_group = cpu_to_le32(block_group);
1934
1935                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1936                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1937                 crc = crc16(crc, (__u8 *)gdp, offset);
1938                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1939                 /* for checksum of struct ext4_group_desc do the rest...*/
1940                 if ((sbi->s_es->s_feature_incompat &
1941                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1942                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1943                         crc = crc16(crc, (__u8 *)gdp + offset,
1944                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1945                                         offset);
1946         }
1947
1948         return cpu_to_le16(crc);
1949 }
1950
1951 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1952                                 struct ext4_group_desc *gdp)
1953 {
1954         if ((sbi->s_es->s_feature_ro_compat &
1955              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1956             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1957                 return 0;
1958
1959         return 1;
1960 }
1961
1962 /* Called at mount-time, super-block is locked */
1963 static int ext4_check_descriptors(struct super_block *sb,
1964                                   ext4_group_t *first_not_zeroed)
1965 {
1966         struct ext4_sb_info *sbi = EXT4_SB(sb);
1967         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1968         ext4_fsblk_t last_block;
1969         ext4_fsblk_t block_bitmap;
1970         ext4_fsblk_t inode_bitmap;
1971         ext4_fsblk_t inode_table;
1972         int flexbg_flag = 0;
1973         ext4_group_t i, grp = sbi->s_groups_count;
1974
1975         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1976                 flexbg_flag = 1;
1977
1978         ext4_debug("Checking group descriptors");
1979
1980         for (i = 0; i < sbi->s_groups_count; i++) {
1981                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1982
1983                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1984                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1985                 else
1986                         last_block = first_block +
1987                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1988
1989                 if ((grp == sbi->s_groups_count) &&
1990                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
1991                         grp = i;
1992
1993                 block_bitmap = ext4_block_bitmap(sb, gdp);
1994                 if (block_bitmap < first_block || block_bitmap > last_block) {
1995                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1996                                "Block bitmap for group %u not in group "
1997                                "(block %llu)!", i, block_bitmap);
1998                         return 0;
1999                 }
2000                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2001                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2002                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2003                                "Inode bitmap for group %u not in group "
2004                                "(block %llu)!", i, inode_bitmap);
2005                         return 0;
2006                 }
2007                 inode_table = ext4_inode_table(sb, gdp);
2008                 if (inode_table < first_block ||
2009                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2010                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2011                                "Inode table for group %u not in group "
2012                                "(block %llu)!", i, inode_table);
2013                         return 0;
2014                 }
2015                 ext4_lock_group(sb, i);
2016                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2017                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2018                                  "Checksum for group %u failed (%u!=%u)",
2019                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2020                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2021                         if (!(sb->s_flags & MS_RDONLY)) {
2022                                 ext4_unlock_group(sb, i);
2023                                 return 0;
2024                         }
2025                 }
2026                 ext4_unlock_group(sb, i);
2027                 if (!flexbg_flag)
2028                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2029         }
2030         if (NULL != first_not_zeroed)
2031                 *first_not_zeroed = grp;
2032
2033         ext4_free_blocks_count_set(sbi->s_es,
2034                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2035         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2036         return 1;
2037 }
2038
2039 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2040  * the superblock) which were deleted from all directories, but held open by
2041  * a process at the time of a crash.  We walk the list and try to delete these
2042  * inodes at recovery time (only with a read-write filesystem).
2043  *
2044  * In order to keep the orphan inode chain consistent during traversal (in
2045  * case of crash during recovery), we link each inode into the superblock
2046  * orphan list_head and handle it the same way as an inode deletion during
2047  * normal operation (which journals the operations for us).
2048  *
2049  * We only do an iget() and an iput() on each inode, which is very safe if we
2050  * accidentally point at an in-use or already deleted inode.  The worst that
2051  * can happen in this case is that we get a "bit already cleared" message from
2052  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2053  * e2fsck was run on this filesystem, and it must have already done the orphan
2054  * inode cleanup for us, so we can safely abort without any further action.
2055  */
2056 static void ext4_orphan_cleanup(struct super_block *sb,
2057                                 struct ext4_super_block *es)
2058 {
2059         unsigned int s_flags = sb->s_flags;
2060         int nr_orphans = 0, nr_truncates = 0;
2061 #ifdef CONFIG_QUOTA
2062         int i;
2063 #endif
2064         if (!es->s_last_orphan) {
2065                 jbd_debug(4, "no orphan inodes to clean up\n");
2066                 return;
2067         }
2068
2069         if (bdev_read_only(sb->s_bdev)) {
2070                 ext4_msg(sb, KERN_ERR, "write access "
2071                         "unavailable, skipping orphan cleanup");
2072                 return;
2073         }
2074
2075         /* Check if feature set would not allow a r/w mount */
2076         if (!ext4_feature_set_ok(sb, 0)) {
2077                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2078                          "unknown ROCOMPAT features");
2079                 return;
2080         }
2081
2082         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2083                 if (es->s_last_orphan)
2084                         jbd_debug(1, "Errors on filesystem, "
2085                                   "clearing orphan list.\n");
2086                 es->s_last_orphan = 0;
2087                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2088                 return;
2089         }
2090
2091         if (s_flags & MS_RDONLY) {
2092                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2093                 sb->s_flags &= ~MS_RDONLY;
2094         }
2095 #ifdef CONFIG_QUOTA
2096         /* Needed for iput() to work correctly and not trash data */
2097         sb->s_flags |= MS_ACTIVE;
2098         /* Turn on quotas so that they are updated correctly */
2099         for (i = 0; i < MAXQUOTAS; i++) {
2100                 if (EXT4_SB(sb)->s_qf_names[i]) {
2101                         int ret = ext4_quota_on_mount(sb, i);
2102                         if (ret < 0)
2103                                 ext4_msg(sb, KERN_ERR,
2104                                         "Cannot turn on journaled "
2105                                         "quota: error %d", ret);
2106                 }
2107         }
2108 #endif
2109
2110         while (es->s_last_orphan) {
2111                 struct inode *inode;
2112
2113                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2114                 if (IS_ERR(inode)) {
2115                         es->s_last_orphan = 0;
2116                         break;
2117                 }
2118
2119                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2120                 dquot_initialize(inode);
2121                 if (inode->i_nlink) {
2122                         ext4_msg(sb, KERN_DEBUG,
2123                                 "%s: truncating inode %lu to %lld bytes",
2124                                 __func__, inode->i_ino, inode->i_size);
2125                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2126                                   inode->i_ino, inode->i_size);
2127                         ext4_truncate(inode);
2128                         nr_truncates++;
2129                 } else {
2130                         ext4_msg(sb, KERN_DEBUG,
2131                                 "%s: deleting unreferenced inode %lu",
2132                                 __func__, inode->i_ino);
2133                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2134                                   inode->i_ino);
2135                         nr_orphans++;
2136                 }
2137                 iput(inode);  /* The delete magic happens here! */
2138         }
2139
2140 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2141
2142         if (nr_orphans)
2143                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2144                        PLURAL(nr_orphans));
2145         if (nr_truncates)
2146                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2147                        PLURAL(nr_truncates));
2148 #ifdef CONFIG_QUOTA
2149         /* Turn quotas off */
2150         for (i = 0; i < MAXQUOTAS; i++) {
2151                 if (sb_dqopt(sb)->files[i])
2152                         dquot_quota_off(sb, i);
2153         }
2154 #endif
2155         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2156 }
2157
2158 /*
2159  * Maximal extent format file size.
2160  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2161  * extent format containers, within a sector_t, and within i_blocks
2162  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2163  * so that won't be a limiting factor.
2164  *
2165  * However there is other limiting factor. We do store extents in the form
2166  * of starting block and length, hence the resulting length of the extent
2167  * covering maximum file size must fit into on-disk format containers as
2168  * well. Given that length is always by 1 unit bigger than max unit (because
2169  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2170  *
2171  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2172  */
2173 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2174 {
2175         loff_t res;
2176         loff_t upper_limit = MAX_LFS_FILESIZE;
2177
2178         /* small i_blocks in vfs inode? */
2179         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2180                 /*
2181                  * CONFIG_LBDAF is not enabled implies the inode
2182                  * i_block represent total blocks in 512 bytes
2183                  * 32 == size of vfs inode i_blocks * 8
2184                  */
2185                 upper_limit = (1LL << 32) - 1;
2186
2187                 /* total blocks in file system block size */
2188                 upper_limit >>= (blkbits - 9);
2189                 upper_limit <<= blkbits;
2190         }
2191
2192         /*
2193          * 32-bit extent-start container, ee_block. We lower the maxbytes
2194          * by one fs block, so ee_len can cover the extent of maximum file
2195          * size
2196          */
2197         res = (1LL << 32) - 1;
2198         res <<= blkbits;
2199
2200         /* Sanity check against vm- & vfs- imposed limits */
2201         if (res > upper_limit)
2202                 res = upper_limit;
2203
2204         return res;
2205 }
2206
2207 /*
2208  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2209  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2210  * We need to be 1 filesystem block less than the 2^48 sector limit.
2211  */
2212 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2213 {
2214         loff_t res = EXT4_NDIR_BLOCKS;
2215         int meta_blocks;
2216         loff_t upper_limit;
2217         /* This is calculated to be the largest file size for a dense, block
2218          * mapped file such that the file's total number of 512-byte sectors,
2219          * including data and all indirect blocks, does not exceed (2^48 - 1).
2220          *
2221          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2222          * number of 512-byte sectors of the file.
2223          */
2224
2225         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2226                 /*
2227                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2228                  * the inode i_block field represents total file blocks in
2229                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2230                  */
2231                 upper_limit = (1LL << 32) - 1;
2232
2233                 /* total blocks in file system block size */
2234                 upper_limit >>= (bits - 9);
2235
2236         } else {
2237                 /*
2238                  * We use 48 bit ext4_inode i_blocks
2239                  * With EXT4_HUGE_FILE_FL set the i_blocks
2240                  * represent total number of blocks in
2241                  * file system block size
2242                  */
2243                 upper_limit = (1LL << 48) - 1;
2244
2245         }
2246
2247         /* indirect blocks */
2248         meta_blocks = 1;
2249         /* double indirect blocks */
2250         meta_blocks += 1 + (1LL << (bits-2));
2251         /* tripple indirect blocks */
2252         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2253
2254         upper_limit -= meta_blocks;
2255         upper_limit <<= bits;
2256
2257         res += 1LL << (bits-2);
2258         res += 1LL << (2*(bits-2));
2259         res += 1LL << (3*(bits-2));
2260         res <<= bits;
2261         if (res > upper_limit)
2262                 res = upper_limit;
2263
2264         if (res > MAX_LFS_FILESIZE)
2265                 res = MAX_LFS_FILESIZE;
2266
2267         return res;
2268 }
2269
2270 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2271                                    ext4_fsblk_t logical_sb_block, int nr)
2272 {
2273         struct ext4_sb_info *sbi = EXT4_SB(sb);
2274         ext4_group_t bg, first_meta_bg;
2275         int has_super = 0;
2276
2277         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2278
2279         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2280             nr < first_meta_bg)
2281                 return logical_sb_block + nr + 1;
2282         bg = sbi->s_desc_per_block * nr;
2283         if (ext4_bg_has_super(sb, bg))
2284                 has_super = 1;
2285
2286         return (has_super + ext4_group_first_block_no(sb, bg));
2287 }
2288
2289 /**
2290  * ext4_get_stripe_size: Get the stripe size.
2291  * @sbi: In memory super block info
2292  *
2293  * If we have specified it via mount option, then
2294  * use the mount option value. If the value specified at mount time is
2295  * greater than the blocks per group use the super block value.
2296  * If the super block value is greater than blocks per group return 0.
2297  * Allocator needs it be less than blocks per group.
2298  *
2299  */
2300 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2301 {
2302         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2303         unsigned long stripe_width =
2304                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2305         int ret;
2306
2307         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2308                 ret = sbi->s_stripe;
2309         else if (stripe_width <= sbi->s_blocks_per_group)
2310                 ret = stripe_width;
2311         else if (stride <= sbi->s_blocks_per_group)
2312                 ret = stride;
2313         else
2314                 ret = 0;
2315
2316         /*
2317          * If the stripe width is 1, this makes no sense and
2318          * we set it to 0 to turn off stripe handling code.
2319          */
2320         if (ret <= 1)
2321                 ret = 0;
2322
2323         return ret;
2324 }
2325
2326 /* sysfs supprt */
2327
2328 struct ext4_attr {
2329         struct attribute attr;
2330         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2331         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2332                          const char *, size_t);
2333         int offset;
2334 };
2335
2336 static int parse_strtoul(const char *buf,
2337                 unsigned long max, unsigned long *value)
2338 {
2339         char *endp;
2340
2341         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2342         endp = skip_spaces(endp);
2343         if (*endp || *value > max)
2344                 return -EINVAL;
2345
2346         return 0;
2347 }
2348
2349 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2350                                               struct ext4_sb_info *sbi,
2351                                               char *buf)
2352 {
2353         return snprintf(buf, PAGE_SIZE, "%llu\n",
2354                 (s64) EXT4_C2B(sbi,
2355                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2356 }
2357
2358 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2359                                          struct ext4_sb_info *sbi, char *buf)
2360 {
2361         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2362
2363         if (!sb->s_bdev->bd_part)
2364                 return snprintf(buf, PAGE_SIZE, "0\n");
2365         return snprintf(buf, PAGE_SIZE, "%lu\n",
2366                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2367                          sbi->s_sectors_written_start) >> 1);
2368 }
2369
2370 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2371                                           struct ext4_sb_info *sbi, char *buf)
2372 {
2373         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2374
2375         if (!sb->s_bdev->bd_part)
2376                 return snprintf(buf, PAGE_SIZE, "0\n");
2377         return snprintf(buf, PAGE_SIZE, "%llu\n",
2378                         (unsigned long long)(sbi->s_kbytes_written +
2379                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2380                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2381 }
2382
2383 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2384                                           struct ext4_sb_info *sbi,
2385                                           const char *buf, size_t count)
2386 {
2387         unsigned long t;
2388
2389         if (parse_strtoul(buf, 0x40000000, &t))
2390                 return -EINVAL;
2391
2392         if (t && !is_power_of_2(t))
2393                 return -EINVAL;
2394
2395         sbi->s_inode_readahead_blks = t;
2396         return count;
2397 }
2398
2399 static ssize_t sbi_ui_show(struct ext4_attr *a,
2400                            struct ext4_sb_info *sbi, char *buf)
2401 {
2402         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2403
2404         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2405 }
2406
2407 static ssize_t sbi_ui_store(struct ext4_attr *a,
2408                             struct ext4_sb_info *sbi,
2409                             const char *buf, size_t count)
2410 {
2411         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2412         unsigned long t;
2413
2414         if (parse_strtoul(buf, 0xffffffff, &t))
2415                 return -EINVAL;
2416         *ui = t;
2417         return count;
2418 }
2419
2420 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2421 static struct ext4_attr ext4_attr_##_name = {                   \
2422         .attr = {.name = __stringify(_name), .mode = _mode },   \
2423         .show   = _show,                                        \
2424         .store  = _store,                                       \
2425         .offset = offsetof(struct ext4_sb_info, _elname),       \
2426 }
2427 #define EXT4_ATTR(name, mode, show, store) \
2428 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2429
2430 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2431 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2432 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2433 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2434         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2435 #define ATTR_LIST(name) &ext4_attr_##name.attr
2436
2437 EXT4_RO_ATTR(delayed_allocation_blocks);
2438 EXT4_RO_ATTR(session_write_kbytes);
2439 EXT4_RO_ATTR(lifetime_write_kbytes);
2440 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2441                  inode_readahead_blks_store, s_inode_readahead_blks);
2442 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2443 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2444 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2445 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2446 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2447 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2448 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2449 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2450
2451 static struct attribute *ext4_attrs[] = {
2452         ATTR_LIST(delayed_allocation_blocks),
2453         ATTR_LIST(session_write_kbytes),
2454         ATTR_LIST(lifetime_write_kbytes),
2455         ATTR_LIST(inode_readahead_blks),
2456         ATTR_LIST(inode_goal),
2457         ATTR_LIST(mb_stats),
2458         ATTR_LIST(mb_max_to_scan),
2459         ATTR_LIST(mb_min_to_scan),
2460         ATTR_LIST(mb_order2_req),
2461         ATTR_LIST(mb_stream_req),
2462         ATTR_LIST(mb_group_prealloc),
2463         ATTR_LIST(max_writeback_mb_bump),
2464         NULL,
2465 };
2466
2467 /* Features this copy of ext4 supports */
2468 EXT4_INFO_ATTR(lazy_itable_init);
2469 EXT4_INFO_ATTR(batched_discard);
2470
2471 static struct attribute *ext4_feat_attrs[] = {
2472         ATTR_LIST(lazy_itable_init),
2473         ATTR_LIST(batched_discard),
2474         NULL,
2475 };
2476
2477 static ssize_t ext4_attr_show(struct kobject *kobj,
2478                               struct attribute *attr, char *buf)
2479 {
2480         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2481                                                 s_kobj);
2482         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2483
2484         return a->show ? a->show(a, sbi, buf) : 0;
2485 }
2486
2487 static ssize_t ext4_attr_store(struct kobject *kobj,
2488                                struct attribute *attr,
2489                                const char *buf, size_t len)
2490 {
2491         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2492                                                 s_kobj);
2493         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2494
2495         return a->store ? a->store(a, sbi, buf, len) : 0;
2496 }
2497
2498 static void ext4_sb_release(struct kobject *kobj)
2499 {
2500         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2501                                                 s_kobj);
2502         complete(&sbi->s_kobj_unregister);
2503 }
2504
2505 static const struct sysfs_ops ext4_attr_ops = {
2506         .show   = ext4_attr_show,
2507         .store  = ext4_attr_store,
2508 };
2509
2510 static struct kobj_type ext4_ktype = {
2511         .default_attrs  = ext4_attrs,
2512         .sysfs_ops      = &ext4_attr_ops,
2513         .release        = ext4_sb_release,
2514 };
2515
2516 static void ext4_feat_release(struct kobject *kobj)
2517 {
2518         complete(&ext4_feat->f_kobj_unregister);
2519 }
2520
2521 static struct kobj_type ext4_feat_ktype = {
2522         .default_attrs  = ext4_feat_attrs,
2523         .sysfs_ops      = &ext4_attr_ops,
2524         .release        = ext4_feat_release,
2525 };
2526
2527 /*
2528  * Check whether this filesystem can be mounted based on
2529  * the features present and the RDONLY/RDWR mount requested.
2530  * Returns 1 if this filesystem can be mounted as requested,
2531  * 0 if it cannot be.
2532  */
2533 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2534 {
2535         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2536                 ext4_msg(sb, KERN_ERR,
2537                         "Couldn't mount because of "
2538                         "unsupported optional features (%x)",
2539                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2540                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2541                 return 0;
2542         }
2543
2544         if (readonly)
2545                 return 1;
2546
2547         /* Check that feature set is OK for a read-write mount */
2548         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2549                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2550                          "unsupported optional features (%x)",
2551                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2552                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2553                 return 0;
2554         }
2555         /*
2556          * Large file size enabled file system can only be mounted
2557          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2558          */
2559         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2560                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2561                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2562                                  "cannot be mounted RDWR without "
2563                                  "CONFIG_LBDAF");
2564                         return 0;
2565                 }
2566         }
2567         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2568             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2569                 ext4_msg(sb, KERN_ERR,
2570                          "Can't support bigalloc feature without "
2571                          "extents feature\n");
2572                 return 0;
2573         }
2574         return 1;
2575 }
2576
2577 /*
2578  * This function is called once a day if we have errors logged
2579  * on the file system
2580  */
2581 static void print_daily_error_info(unsigned long arg)
2582 {
2583         struct super_block *sb = (struct super_block *) arg;
2584         struct ext4_sb_info *sbi;
2585         struct ext4_super_block *es;
2586
2587         sbi = EXT4_SB(sb);
2588         es = sbi->s_es;
2589
2590         if (es->s_error_count)
2591                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2592                          le32_to_cpu(es->s_error_count));
2593         if (es->s_first_error_time) {
2594                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2595                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2596                        (int) sizeof(es->s_first_error_func),
2597                        es->s_first_error_func,
2598                        le32_to_cpu(es->s_first_error_line));
2599                 if (es->s_first_error_ino)
2600                         printk(": inode %u",
2601                                le32_to_cpu(es->s_first_error_ino));
2602                 if (es->s_first_error_block)
2603                         printk(": block %llu", (unsigned long long)
2604                                le64_to_cpu(es->s_first_error_block));
2605                 printk("\n");
2606         }
2607         if (es->s_last_error_time) {
2608                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2609                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2610                        (int) sizeof(es->s_last_error_func),
2611                        es->s_last_error_func,
2612                        le32_to_cpu(es->s_last_error_line));
2613                 if (es->s_last_error_ino)
2614                         printk(": inode %u",
2615                                le32_to_cpu(es->s_last_error_ino));
2616                 if (es->s_last_error_block)
2617                         printk(": block %llu", (unsigned long long)
2618                                le64_to_cpu(es->s_last_error_block));
2619                 printk("\n");
2620         }
2621         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2622 }
2623
2624 /* Find next suitable group and run ext4_init_inode_table */
2625 static int ext4_run_li_request(struct ext4_li_request *elr)
2626 {
2627         struct ext4_group_desc *gdp = NULL;
2628         ext4_group_t group, ngroups;
2629         struct super_block *sb;
2630         unsigned long timeout = 0;
2631         int ret = 0;
2632
2633         sb = elr->lr_super;
2634         ngroups = EXT4_SB(sb)->s_groups_count;
2635
2636         for (group = elr->lr_next_group; group < ngroups; group++) {
2637                 gdp = ext4_get_group_desc(sb, group, NULL);
2638                 if (!gdp) {
2639                         ret = 1;
2640                         break;
2641                 }
2642
2643                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2644                         break;
2645         }
2646
2647         if (group == ngroups)
2648                 ret = 1;
2649
2650         if (!ret) {
2651                 timeout = jiffies;
2652                 ret = ext4_init_inode_table(sb, group,
2653                                             elr->lr_timeout ? 0 : 1);
2654                 if (elr->lr_timeout == 0) {
2655                         timeout = (jiffies - timeout) *
2656                                   elr->lr_sbi->s_li_wait_mult;
2657                         elr->lr_timeout = timeout;
2658                 }
2659                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2660                 elr->lr_next_group = group + 1;
2661         }
2662
2663         return ret;
2664 }
2665
2666 /*
2667  * Remove lr_request from the list_request and free the
2668  * request structure. Should be called with li_list_mtx held
2669  */
2670 static void ext4_remove_li_request(struct ext4_li_request *elr)
2671 {
2672         struct ext4_sb_info *sbi;
2673
2674         if (!elr)
2675                 return;
2676
2677         sbi = elr->lr_sbi;
2678
2679         list_del(&elr->lr_request);
2680         sbi->s_li_request = NULL;
2681         kfree(elr);
2682 }
2683
2684 static void ext4_unregister_li_request(struct super_block *sb)
2685 {
2686         mutex_lock(&ext4_li_mtx);
2687         if (!ext4_li_info) {
2688                 mutex_unlock(&ext4_li_mtx);
2689                 return;
2690         }
2691
2692         mutex_lock(&ext4_li_info->li_list_mtx);
2693         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2694         mutex_unlock(&ext4_li_info->li_list_mtx);
2695         mutex_unlock(&ext4_li_mtx);
2696 }
2697
2698 static struct task_struct *ext4_lazyinit_task;
2699
2700 /*
2701  * This is the function where ext4lazyinit thread lives. It walks
2702  * through the request list searching for next scheduled filesystem.
2703  * When such a fs is found, run the lazy initialization request
2704  * (ext4_rn_li_request) and keep track of the time spend in this
2705  * function. Based on that time we compute next schedule time of
2706  * the request. When walking through the list is complete, compute
2707  * next waking time and put itself into sleep.
2708  */
2709 static int ext4_lazyinit_thread(void *arg)
2710 {
2711         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2712         struct list_head *pos, *n;
2713         struct ext4_li_request *elr;
2714         unsigned long next_wakeup, cur;
2715
2716         BUG_ON(NULL == eli);
2717
2718 cont_thread:
2719         while (true) {
2720                 next_wakeup = MAX_JIFFY_OFFSET;
2721
2722                 mutex_lock(&eli->li_list_mtx);
2723                 if (list_empty(&eli->li_request_list)) {
2724                         mutex_unlock(&eli->li_list_mtx);
2725                         goto exit_thread;
2726                 }
2727
2728                 list_for_each_safe(pos, n, &eli->li_request_list) {
2729                         elr = list_entry(pos, struct ext4_li_request,
2730                                          lr_request);
2731
2732                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2733                                 if (ext4_run_li_request(elr) != 0) {
2734                                         /* error, remove the lazy_init job */
2735                                         ext4_remove_li_request(elr);
2736                                         continue;
2737                                 }
2738                         }
2739
2740                         if (time_before(elr->lr_next_sched, next_wakeup))
2741                                 next_wakeup = elr->lr_next_sched;
2742                 }
2743                 mutex_unlock(&eli->li_list_mtx);
2744
2745                 try_to_freeze();
2746
2747                 cur = jiffies;
2748                 if ((time_after_eq(cur, next_wakeup)) ||
2749                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2750                         cond_resched();
2751                         continue;
2752                 }
2753
2754                 schedule_timeout_interruptible(next_wakeup - cur);
2755
2756                 if (kthread_should_stop()) {
2757                         ext4_clear_request_list();
2758                         goto exit_thread;
2759                 }
2760         }
2761
2762 exit_thread:
2763         /*
2764          * It looks like the request list is empty, but we need
2765          * to check it under the li_list_mtx lock, to prevent any
2766          * additions into it, and of course we should lock ext4_li_mtx
2767          * to atomically free the list and ext4_li_info, because at
2768          * this point another ext4 filesystem could be registering
2769          * new one.
2770          */
2771         mutex_lock(&ext4_li_mtx);
2772         mutex_lock(&eli->li_list_mtx);
2773         if (!list_empty(&eli->li_request_list)) {
2774                 mutex_unlock(&eli->li_list_mtx);
2775                 mutex_unlock(&ext4_li_mtx);
2776                 goto cont_thread;
2777         }
2778         mutex_unlock(&eli->li_list_mtx);
2779         kfree(ext4_li_info);
2780         ext4_li_info = NULL;
2781         mutex_unlock(&ext4_li_mtx);
2782
2783         return 0;
2784 }
2785
2786 static void ext4_clear_request_list(void)
2787 {
2788         struct list_head *pos, *n;
2789         struct ext4_li_request *elr;
2790
2791         mutex_lock(&ext4_li_info->li_list_mtx);
2792         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2793                 elr = list_entry(pos, struct ext4_li_request,
2794                                  lr_request);
2795                 ext4_remove_li_request(elr);
2796         }
2797         mutex_unlock(&ext4_li_info->li_list_mtx);
2798 }
2799
2800 static int ext4_run_lazyinit_thread(void)
2801 {
2802         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2803                                          ext4_li_info, "ext4lazyinit");
2804         if (IS_ERR(ext4_lazyinit_task)) {
2805                 int err = PTR_ERR(ext4_lazyinit_task);
2806                 ext4_clear_request_list();
2807                 kfree(ext4_li_info);
2808                 ext4_li_info = NULL;
2809                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2810                                  "initialization thread\n",
2811                                  err);
2812                 return err;
2813         }
2814         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2815         return 0;
2816 }
2817
2818 /*
2819  * Check whether it make sense to run itable init. thread or not.
2820  * If there is at least one uninitialized inode table, return
2821  * corresponding group number, else the loop goes through all
2822  * groups and return total number of groups.
2823  */
2824 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2825 {
2826         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2827         struct ext4_group_desc *gdp = NULL;
2828
2829         for (group = 0; group < ngroups; group++) {
2830                 gdp = ext4_get_group_desc(sb, group, NULL);
2831                 if (!gdp)
2832                         continue;
2833
2834                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2835                         break;
2836         }
2837
2838         return group;
2839 }
2840
2841 static int ext4_li_info_new(void)
2842 {
2843         struct ext4_lazy_init *eli = NULL;
2844
2845         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2846         if (!eli)
2847                 return -ENOMEM;
2848
2849         INIT_LIST_HEAD(&eli->li_request_list);
2850         mutex_init(&eli->li_list_mtx);
2851
2852         eli->li_state |= EXT4_LAZYINIT_QUIT;
2853
2854         ext4_li_info = eli;
2855
2856         return 0;
2857 }
2858
2859 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2860                                             ext4_group_t start)
2861 {
2862         struct ext4_sb_info *sbi = EXT4_SB(sb);
2863         struct ext4_li_request *elr;
2864         unsigned long rnd;
2865
2866         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2867         if (!elr)
2868                 return NULL;
2869
2870         elr->lr_super = sb;
2871         elr->lr_sbi = sbi;
2872         elr->lr_next_group = start;
2873
2874         /*
2875          * Randomize first schedule time of the request to
2876          * spread the inode table initialization requests
2877          * better.
2878          */
2879         get_random_bytes(&rnd, sizeof(rnd));
2880         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2881                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2882
2883         return elr;
2884 }
2885
2886 static int ext4_register_li_request(struct super_block *sb,
2887                                     ext4_group_t first_not_zeroed)
2888 {
2889         struct ext4_sb_info *sbi = EXT4_SB(sb);
2890         struct ext4_li_request *elr;
2891         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2892         int ret = 0;
2893
2894         if (sbi->s_li_request != NULL) {
2895                 /*
2896                  * Reset timeout so it can be computed again, because
2897                  * s_li_wait_mult might have changed.
2898                  */
2899                 sbi->s_li_request->lr_timeout = 0;
2900                 return 0;
2901         }
2902
2903         if (first_not_zeroed == ngroups ||
2904             (sb->s_flags & MS_RDONLY) ||
2905             !test_opt(sb, INIT_INODE_TABLE))
2906                 return 0;
2907
2908         elr = ext4_li_request_new(sb, first_not_zeroed);
2909         if (!elr)
2910                 return -ENOMEM;
2911
2912         mutex_lock(&ext4_li_mtx);
2913
2914         if (NULL == ext4_li_info) {
2915                 ret = ext4_li_info_new();
2916                 if (ret)
2917                         goto out;
2918         }
2919
2920         mutex_lock(&ext4_li_info->li_list_mtx);
2921         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2922         mutex_unlock(&ext4_li_info->li_list_mtx);
2923
2924         sbi->s_li_request = elr;
2925         /*
2926          * set elr to NULL here since it has been inserted to
2927          * the request_list and the removal and free of it is
2928          * handled by ext4_clear_request_list from now on.
2929          */
2930         elr = NULL;
2931
2932         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2933                 ret = ext4_run_lazyinit_thread();
2934                 if (ret)
2935                         goto out;
2936         }
2937 out:
2938         mutex_unlock(&ext4_li_mtx);
2939         if (ret)
2940                 kfree(elr);
2941         return ret;
2942 }
2943
2944 /*
2945  * We do not need to lock anything since this is called on
2946  * module unload.
2947  */
2948 static void ext4_destroy_lazyinit_thread(void)
2949 {
2950         /*
2951          * If thread exited earlier
2952          * there's nothing to be done.
2953          */
2954         if (!ext4_li_info || !ext4_lazyinit_task)
2955                 return;
2956
2957         kthread_stop(ext4_lazyinit_task);
2958 }
2959
2960 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2961 {
2962         char *orig_data = kstrdup(data, GFP_KERNEL);
2963         struct buffer_head *bh;
2964         struct ext4_super_block *es = NULL;
2965         struct ext4_sb_info *sbi;
2966         ext4_fsblk_t block;
2967         ext4_fsblk_t sb_block = get_sb_block(&data);
2968         ext4_fsblk_t logical_sb_block;
2969         unsigned long offset = 0;
2970         unsigned long journal_devnum = 0;
2971         unsigned long def_mount_opts;
2972         struct inode *root;
2973         char *cp;
2974         const char *descr;
2975         int ret = -ENOMEM;
2976         int blocksize, clustersize;
2977         unsigned int db_count;
2978         unsigned int i;
2979         int needs_recovery, has_huge_files, has_bigalloc;
2980         __u64 blocks_count;
2981         int err;
2982         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2983         ext4_group_t first_not_zeroed;
2984
2985         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2986         if (!sbi)
2987                 goto out_free_orig;
2988
2989         sbi->s_blockgroup_lock =
2990                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
2991         if (!sbi->s_blockgroup_lock) {
2992                 kfree(sbi);
2993                 goto out_free_orig;
2994         }
2995         sb->s_fs_info = sbi;
2996         sbi->s_mount_opt = 0;
2997         sbi->s_resuid = make_kuid(&init_user_ns, EXT4_DEF_RESUID);
2998         sbi->s_resgid = make_kgid(&init_user_ns, EXT4_DEF_RESGID);
2999         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3000         sbi->s_sb_block = sb_block;
3001         if (sb->s_bdev->bd_part)
3002                 sbi->s_sectors_written_start =
3003                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3004
3005         /* Cleanup superblock name */
3006         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3007                 *cp = '!';
3008
3009         ret = -EINVAL;
3010         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3011         if (!blocksize) {
3012                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3013                 goto out_fail;
3014         }
3015
3016         /*
3017          * The ext4 superblock will not be buffer aligned for other than 1kB
3018          * block sizes.  We need to calculate the offset from buffer start.
3019          */
3020         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3021                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3022                 offset = do_div(logical_sb_block, blocksize);
3023         } else {
3024                 logical_sb_block = sb_block;
3025         }
3026
3027         if (!(bh = sb_bread(sb, logical_sb_block))) {
3028                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3029                 goto out_fail;
3030         }
3031         /*
3032          * Note: s_es must be initialized as soon as possible because
3033          *       some ext4 macro-instructions depend on its value
3034          */
3035         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3036         sbi->s_es = es;
3037         sb->s_magic = le16_to_cpu(es->s_magic);
3038         if (sb->s_magic != EXT4_SUPER_MAGIC)
3039                 goto cantfind_ext4;
3040         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3041
3042         /* Set defaults before we parse the mount options */
3043         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3044         set_opt(sb, INIT_INODE_TABLE);
3045         if (def_mount_opts & EXT4_DEFM_DEBUG)
3046                 set_opt(sb, DEBUG);
3047         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3048                 set_opt(sb, GRPID);
3049         if (def_mount_opts & EXT4_DEFM_UID16)
3050                 set_opt(sb, NO_UID32);
3051         /* xattr user namespace & acls are now defaulted on */
3052 #ifdef CONFIG_EXT4_FS_XATTR
3053         set_opt(sb, XATTR_USER);
3054 #endif
3055 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3056         set_opt(sb, POSIX_ACL);
3057 #endif
3058         set_opt(sb, MBLK_IO_SUBMIT);
3059         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3060                 set_opt(sb, JOURNAL_DATA);
3061         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3062                 set_opt(sb, ORDERED_DATA);
3063         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3064                 set_opt(sb, WRITEBACK_DATA);
3065
3066         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3067                 set_opt(sb, ERRORS_PANIC);
3068         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3069                 set_opt(sb, ERRORS_CONT);
3070         else
3071                 set_opt(sb, ERRORS_RO);
3072         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3073                 set_opt(sb, BLOCK_VALIDITY);
3074         if (def_mount_opts & EXT4_DEFM_DISCARD)
3075                 set_opt(sb, DISCARD);
3076
3077         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3078         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3079         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3080         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3081         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3082
3083         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3084                 set_opt(sb, BARRIER);
3085
3086         /*
3087          * enable delayed allocation by default
3088          * Use -o nodelalloc to turn it off
3089          */
3090         if (!IS_EXT3_SB(sb) &&
3091             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3092                 set_opt(sb, DELALLOC);
3093
3094         /*
3095          * set default s_li_wait_mult for lazyinit, for the case there is
3096          * no mount option specified.
3097          */
3098         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3099
3100         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3101                            &journal_devnum, &journal_ioprio, 0)) {
3102                 ext4_msg(sb, KERN_WARNING,
3103                          "failed to parse options in superblock: %s",
3104                          sbi->s_es->s_mount_opts);
3105         }
3106         sbi->s_def_mount_opt = sbi->s_mount_opt;
3107         if (!parse_options((char *) data, sb, &journal_devnum,
3108                            &journal_ioprio, 0))
3109                 goto failed_mount;
3110
3111         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3112                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3113                             "with data=journal disables delayed "
3114                             "allocation and O_DIRECT support!\n");
3115                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3116                         ext4_msg(sb, KERN_ERR, "can't mount with "
3117                                  "both data=journal and delalloc");
3118                         goto failed_mount;
3119                 }
3120                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3121                         ext4_msg(sb, KERN_ERR, "can't mount with "
3122                                  "both data=journal and delalloc");
3123                         goto failed_mount;
3124                 }
3125                 if (test_opt(sb, DELALLOC))
3126                         clear_opt(sb, DELALLOC);
3127         }
3128
3129         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3130         if (test_opt(sb, DIOREAD_NOLOCK)) {
3131                 if (blocksize < PAGE_SIZE) {
3132                         ext4_msg(sb, KERN_ERR, "can't mount with "
3133                                  "dioread_nolock if block size != PAGE_SIZE");
3134                         goto failed_mount;
3135                 }
3136         }
3137
3138         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3139                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3140
3141         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3142             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3143              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3144              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3145                 ext4_msg(sb, KERN_WARNING,
3146                        "feature flags set on rev 0 fs, "
3147                        "running e2fsck is recommended");
3148
3149         if (IS_EXT2_SB(sb)) {
3150                 if (ext2_feature_set_ok(sb))
3151                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3152                                  "using the ext4 subsystem");
3153                 else {
3154                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3155                                  "to feature incompatibilities");
3156                         goto failed_mount;
3157                 }
3158         }
3159
3160         if (IS_EXT3_SB(sb)) {
3161                 if (ext3_feature_set_ok(sb))
3162                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3163                                  "using the ext4 subsystem");
3164                 else {
3165                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3166                                  "to feature incompatibilities");
3167                         goto failed_mount;
3168                 }
3169         }
3170
3171         /*
3172          * Check feature flags regardless of the revision level, since we
3173          * previously didn't change the revision level when setting the flags,
3174          * so there is a chance incompat flags are set on a rev 0 filesystem.
3175          */
3176         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3177                 goto failed_mount;
3178
3179         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3180             blocksize > EXT4_MAX_BLOCK_SIZE) {
3181                 ext4_msg(sb, KERN_ERR,
3182                        "Unsupported filesystem blocksize %d", blocksize);
3183                 goto failed_mount;
3184         }
3185
3186         if (sb->s_blocksize != blocksize) {
3187                 /* Validate the filesystem blocksize */
3188                 if (!sb_set_blocksize(sb, blocksize)) {
3189                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3190                                         blocksize);
3191                         goto failed_mount;
3192                 }
3193
3194                 brelse(bh);
3195                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3196                 offset = do_div(logical_sb_block, blocksize);
3197                 bh = sb_bread(sb, logical_sb_block);
3198                 if (!bh) {
3199                         ext4_msg(sb, KERN_ERR,
3200                                "Can't read superblock on 2nd try");
3201                         goto failed_mount;
3202                 }
3203                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3204                 sbi->s_es = es;
3205                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3206                         ext4_msg(sb, KERN_ERR,
3207                                "Magic mismatch, very weird!");
3208                         goto failed_mount;
3209                 }
3210         }
3211
3212         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3213                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3214         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3215                                                       has_huge_files);
3216         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3217
3218         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3219                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3220                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3221         } else {
3222                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3223                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3224                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3225                     (!is_power_of_2(sbi->s_inode_size)) ||
3226                     (sbi->s_inode_size > blocksize)) {
3227                         ext4_msg(sb, KERN_ERR,
3228                                "unsupported inode size: %d",
3229                                sbi->s_inode_size);
3230                         goto failed_mount;
3231                 }
3232                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3233                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3234         }
3235
3236         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3237         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3238                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3239                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3240                     !is_power_of_2(sbi->s_desc_size)) {
3241                         ext4_msg(sb, KERN_ERR,
3242                                "unsupported descriptor size %lu",
3243                                sbi->s_desc_size);
3244                         goto failed_mount;
3245                 }
3246         } else
3247                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3248
3249         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3250         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3251         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3252                 goto cantfind_ext4;
3253
3254         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3255         if (sbi->s_inodes_per_block == 0)
3256                 goto cantfind_ext4;
3257         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3258                                         sbi->s_inodes_per_block;
3259         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3260         sbi->s_sbh = bh;
3261         sbi->s_mount_state = le16_to_cpu(es->s_state);
3262         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3263         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3264
3265         for (i = 0; i < 4; i++)
3266                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3267         sbi->s_def_hash_version = es->s_def_hash_version;
3268         i = le32_to_cpu(es->s_flags);
3269         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3270                 sbi->s_hash_unsigned = 3;
3271         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3272 #ifdef __CHAR_UNSIGNED__
3273                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3274                 sbi->s_hash_unsigned = 3;
3275 #else
3276                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3277 #endif
3278         }
3279
3280         /* Handle clustersize */
3281         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3282         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3283                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3284         if (has_bigalloc) {
3285                 if (clustersize < blocksize) {
3286                         ext4_msg(sb, KERN_ERR,
3287                                  "cluster size (%d) smaller than "
3288                                  "block size (%d)", clustersize, blocksize);
3289                         goto failed_mount;
3290                 }
3291                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3292                         le32_to_cpu(es->s_log_block_size);
3293                 sbi->s_clusters_per_group =
3294                         le32_to_cpu(es->s_clusters_per_group);
3295                 if (sbi->s_clusters_per_group > blocksize * 8) {
3296                         ext4_msg(sb, KERN_ERR,
3297                                  "#clusters per group too big: %lu",
3298                                  sbi->s_clusters_per_group);
3299                         goto failed_mount;
3300                 }
3301                 if (sbi->s_blocks_per_group !=
3302                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3303                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3304                                  "clusters per group (%lu) inconsistent",
3305                                  sbi->s_blocks_per_group,
3306                                  sbi->s_clusters_per_group);
3307                         goto failed_mount;
3308                 }
3309         } else {
3310                 if (clustersize != blocksize) {
3311                         ext4_warning(sb, "fragment/cluster size (%d) != "
3312                                      "block size (%d)", clustersize,
3313                                      blocksize);
3314                         clustersize = blocksize;
3315                 }
3316                 if (sbi->s_blocks_per_group > blocksize * 8) {
3317                         ext4_msg(sb, KERN_ERR,
3318                                  "#blocks per group too big: %lu",
3319                                  sbi->s_blocks_per_group);
3320                         goto failed_mount;
3321                 }
3322                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3323                 sbi->s_cluster_bits = 0;
3324         }
3325         sbi->s_cluster_ratio = clustersize / blocksize;
3326
3327         if (sbi->s_inodes_per_group > blocksize * 8) {
3328                 ext4_msg(sb, KERN_ERR,
3329                        "#inodes per group too big: %lu",
3330                        sbi->s_inodes_per_group);
3331                 goto failed_mount;
3332         }
3333
3334         /*
3335          * Test whether we have more sectors than will fit in sector_t,
3336          * and whether the max offset is addressable by the page cache.
3337          */
3338         err = generic_check_addressable(sb->s_blocksize_bits,
3339                                         ext4_blocks_count(es));
3340         if (err) {
3341                 ext4_msg(sb, KERN_ERR, "filesystem"
3342                          " too large to mount safely on this system");
3343                 if (sizeof(sector_t) < 8)
3344                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3345                 ret = err;
3346                 goto failed_mount;
3347         }
3348
3349         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3350                 goto cantfind_ext4;
3351
3352         /* check blocks count against device size */
3353         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3354         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3355                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3356                        "exceeds size of device (%llu blocks)",
3357                        ext4_blocks_count(es), blocks_count);
3358                 goto failed_mount;
3359         }
3360
3361         /*
3362          * It makes no sense for the first data block to be beyond the end
3363          * of the filesystem.
3364          */
3365         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3366                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3367                          "block %u is beyond end of filesystem (%llu)",
3368                          le32_to_cpu(es->s_first_data_block),
3369                          ext4_blocks_count(es));
3370                 goto failed_mount;
3371         }
3372         blocks_count = (ext4_blocks_count(es) -
3373                         le32_to_cpu(es->s_first_data_block) +
3374                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3375         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3376         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3377                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3378                        "(block count %llu, first data block %u, "
3379                        "blocks per group %lu)", sbi->s_groups_count,
3380                        ext4_blocks_count(es),
3381                        le32_to_cpu(es->s_first_data_block),
3382                        EXT4_BLOCKS_PER_GROUP(sb));
3383                 goto failed_mount;
3384         }
3385         sbi->s_groups_count = blocks_count;
3386         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3387                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3388         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3389                    EXT4_DESC_PER_BLOCK(sb);
3390         sbi->s_group_desc = ext4_kvmalloc(db_count *
3391                                           sizeof(struct buffer_head *),
3392                                           GFP_KERNEL);
3393         if (sbi->s_group_desc == NULL) {
3394                 ext4_msg(sb, KERN_ERR, "not enough memory");
3395                 goto failed_mount;
3396         }
3397
3398         if (ext4_proc_root)
3399                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3400
3401         if (sbi->s_proc)
3402                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3403                                  &ext4_seq_options_fops, sb);
3404
3405         bgl_lock_init(sbi->s_blockgroup_lock);
3406
3407         for (i = 0; i < db_count; i++) {
3408                 block = descriptor_loc(sb, logical_sb_block, i);
3409                 sbi->s_group_desc[i] = sb_bread(sb, block);
3410                 if (!sbi->s_group_desc[i]) {
3411                         ext4_msg(sb, KERN_ERR,
3412                                "can't read group descriptor %d", i);
3413                         db_count = i;
3414                         goto failed_mount2;
3415                 }
3416         }
3417         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3418                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3419                 goto failed_mount2;
3420         }
3421         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3422                 if (!ext4_fill_flex_info(sb)) {
3423                         ext4_msg(sb, KERN_ERR,
3424                                "unable to initialize "
3425                                "flex_bg meta info!");
3426                         goto failed_mount2;
3427                 }
3428
3429         sbi->s_gdb_count = db_count;
3430         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3431         spin_lock_init(&sbi->s_next_gen_lock);
3432
3433         init_timer(&sbi->s_err_report);
3434         sbi->s_err_report.function = print_daily_error_info;
3435         sbi->s_err_report.data = (unsigned long) sb;
3436
3437         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3438                         ext4_count_free_clusters(sb));
3439         if (!err) {
3440                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3441                                 ext4_count_free_inodes(sb));
3442         }
3443         if (!err) {
3444                 err = percpu_counter_init(&sbi->s_dirs_counter,
3445                                 ext4_count_dirs(sb));
3446         }
3447         if (!err) {
3448                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3449         }
3450         if (err) {
3451                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3452                 goto failed_mount3;
3453         }
3454
3455         sbi->s_stripe = ext4_get_stripe_size(sbi);
3456         sbi->s_max_writeback_mb_bump = 128;
3457
3458         /*
3459          * set up enough so that it can read an inode
3460          */
3461         if (!test_opt(sb, NOLOAD) &&
3462             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3463                 sb->s_op = &ext4_sops;
3464         else
3465                 sb->s_op = &ext4_nojournal_sops;
3466         sb->s_export_op = &ext4_export_ops;
3467         sb->s_xattr = ext4_xattr_handlers;
3468 #ifdef CONFIG_QUOTA
3469         sb->s_qcop = &ext4_qctl_operations;
3470         sb->dq_op = &ext4_quota_operations;
3471 #endif
3472         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3473
3474         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3475         mutex_init(&sbi->s_orphan_lock);
3476         sbi->s_resize_flags = 0;
3477
3478         sb->s_root = NULL;
3479
3480         needs_recovery = (es->s_last_orphan != 0 ||
3481                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3482                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3483
3484         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3485             !(sb->s_flags & MS_RDONLY))
3486                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3487                         goto failed_mount3;
3488
3489         /*
3490          * The first inode we look at is the journal inode.  Don't try
3491          * root first: it may be modified in the journal!
3492          */
3493         if (!test_opt(sb, NOLOAD) &&
3494             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3495                 if (ext4_load_journal(sb, es, journal_devnum))
3496                         goto failed_mount3;
3497         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3498               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3499                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3500                        "suppressed and not mounted read-only");
3501                 goto failed_mount_wq;
3502         } else {
3503                 clear_opt(sb, DATA_FLAGS);
3504                 sbi->s_journal = NULL;
3505                 needs_recovery = 0;
3506                 goto no_journal;
3507         }
3508
3509         if (ext4_blocks_count(es) > 0xffffffffULL &&
3510             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3511                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3512                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3513                 goto failed_mount_wq;
3514         }
3515
3516         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3517                 jbd2_journal_set_features(sbi->s_journal,
3518                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3519                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3520         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3521                 jbd2_journal_set_features(sbi->s_journal,
3522                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3523                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3524                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3525         } else {
3526                 jbd2_journal_clear_features(sbi->s_journal,
3527                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3528                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3529         }
3530
3531         /* We have now updated the journal if required, so we can
3532          * validate the data journaling mode. */
3533         switch (test_opt(sb, DATA_FLAGS)) {
3534         case 0:
3535                 /* No mode set, assume a default based on the journal
3536                  * capabilities: ORDERED_DATA if the journal can
3537                  * cope, else JOURNAL_DATA
3538                  */
3539                 if (jbd2_journal_check_available_features
3540                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3541                         set_opt(sb, ORDERED_DATA);
3542                 else
3543                         set_opt(sb, JOURNAL_DATA);
3544                 break;
3545
3546         case EXT4_MOUNT_ORDERED_DATA:
3547         case EXT4_MOUNT_WRITEBACK_DATA:
3548                 if (!jbd2_journal_check_available_features
3549                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3550                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3551                                "requested data journaling mode");
3552                         goto failed_mount_wq;
3553                 }
3554         default:
3555                 break;
3556         }
3557         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3558
3559         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3560
3561         /*
3562          * The journal may have updated the bg summary counts, so we
3563          * need to update the global counters.
3564          */
3565         percpu_counter_set(&sbi->s_freeclusters_counter,
3566                            ext4_count_free_clusters(sb));
3567         percpu_counter_set(&sbi->s_freeinodes_counter,
3568                            ext4_count_free_inodes(sb));
3569         percpu_counter_set(&sbi->s_dirs_counter,
3570                            ext4_count_dirs(sb));
3571         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3572
3573 no_journal:
3574         /*
3575          * The maximum number of concurrent works can be high and
3576          * concurrency isn't really necessary.  Limit it to 1.
3577          */
3578         EXT4_SB(sb)->dio_unwritten_wq =
3579                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3580         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3581                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3582                 goto failed_mount_wq;
3583         }
3584
3585         /*
3586          * The jbd2_journal_load will have done any necessary log recovery,
3587          * so we can safely mount the rest of the filesystem now.
3588          */
3589
3590         root = ext4_iget(sb, EXT4_ROOT_INO);
3591         if (IS_ERR(root)) {
3592                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3593                 ret = PTR_ERR(root);
3594                 root = NULL;
3595                 goto failed_mount4;
3596         }
3597         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3598                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3599                 iput(root);
3600                 goto failed_mount4;
3601         }
3602         sb->s_root = d_make_root(root);
3603         if (!sb->s_root) {
3604                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3605                 ret = -ENOMEM;
3606                 goto failed_mount4;
3607         }
3608
3609         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3610
3611         /* determine the minimum size of new large inodes, if present */
3612         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3613                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3614                                                      EXT4_GOOD_OLD_INODE_SIZE;
3615                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3616                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3617                         if (sbi->s_want_extra_isize <
3618                             le16_to_cpu(es->s_want_extra_isize))
3619                                 sbi->s_want_extra_isize =
3620                                         le16_to_cpu(es->s_want_extra_isize);
3621                         if (sbi->s_want_extra_isize <
3622                             le16_to_cpu(es->s_min_extra_isize))
3623                                 sbi->s_want_extra_isize =
3624                                         le16_to_cpu(es->s_min_extra_isize);
3625                 }
3626         }
3627         /* Check if enough inode space is available */
3628         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3629                                                         sbi->s_inode_size) {
3630                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3631                                                        EXT4_GOOD_OLD_INODE_SIZE;
3632                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3633                          "available");
3634         }
3635
3636         err = ext4_setup_system_zone(sb);
3637         if (err) {
3638                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3639                          "zone (%d)", err);
3640                 goto failed_mount4a;
3641         }
3642
3643         ext4_ext_init(sb);
3644         err = ext4_mb_init(sb, needs_recovery);
3645         if (err) {
3646                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3647                          err);
3648                 goto failed_mount5;
3649         }
3650
3651         err = ext4_register_li_request(sb, first_not_zeroed);
3652         if (err)
3653                 goto failed_mount6;
3654
3655         sbi->s_kobj.kset = ext4_kset;
3656         init_completion(&sbi->s_kobj_unregister);
3657         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3658                                    "%s", sb->s_id);
3659         if (err)
3660                 goto failed_mount7;
3661
3662         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3663         ext4_orphan_cleanup(sb, es);
3664         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3665         if (needs_recovery) {
3666                 ext4_msg(sb, KERN_INFO, "recovery complete");
3667                 ext4_mark_recovery_complete(sb, es);
3668         }
3669         if (EXT4_SB(sb)->s_journal) {
3670                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3671                         descr = " journalled data mode";
3672                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3673                         descr = " ordered data mode";
3674                 else
3675                         descr = " writeback data mode";
3676         } else
3677                 descr = "out journal";
3678
3679         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3680                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3681                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3682
3683         if (es->s_error_count)
3684                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3685
3686         kfree(orig_data);
3687         return 0;
3688
3689 cantfind_ext4:
3690         if (!silent)
3691                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3692         goto failed_mount;
3693
3694 failed_mount7:
3695         ext4_unregister_li_request(sb);
3696 failed_mount6:
3697         ext4_mb_release(sb);
3698 failed_mount5:
3699         ext4_ext_release(sb);
3700         ext4_release_system_zone(sb);
3701 failed_mount4a:
3702         dput(sb->s_root);
3703         sb->s_root = NULL;
3704 failed_mount4:
3705         ext4_msg(sb, KERN_ERR, "mount failed");
3706         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3707 failed_mount_wq:
3708         if (sbi->s_journal) {
3709                 jbd2_journal_destroy(sbi->s_journal);
3710                 sbi->s_journal = NULL;
3711         }
3712 failed_mount3:
3713         del_timer(&sbi->s_err_report);
3714         if (sbi->s_flex_groups)
3715                 ext4_kvfree(sbi->s_flex_groups);
3716         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3717         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3718         percpu_counter_destroy(&sbi->s_dirs_counter);
3719         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3720         if (sbi->s_mmp_tsk)
3721                 kthread_stop(sbi->s_mmp_tsk);
3722 failed_mount2:
3723         for (i = 0; i < db_count; i++)
3724                 brelse(sbi->s_group_desc[i]);
3725         ext4_kvfree(sbi->s_group_desc);
3726 failed_mount:
3727         if (sbi->s_proc) {
3728                 remove_proc_entry("options", sbi->s_proc);
3729                 remove_proc_entry(sb->s_id, ext4_proc_root);
3730         }
3731 #ifdef CONFIG_QUOTA
3732         for (i = 0; i < MAXQUOTAS; i++)
3733                 kfree(sbi->s_qf_names[i]);
3734 #endif
3735         ext4_blkdev_remove(sbi);
3736         brelse(bh);
3737 out_fail:
3738         sb->s_fs_info = NULL;
3739         kfree(sbi->s_blockgroup_lock);
3740         kfree(sbi);
3741 out_free_orig:
3742         kfree(orig_data);
3743         return ret;
3744 }
3745
3746 /*
3747  * Setup any per-fs journal parameters now.  We'll do this both on
3748  * initial mount, once the journal has been initialised but before we've
3749  * done any recovery; and again on any subsequent remount.
3750  */
3751 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3752 {
3753         struct ext4_sb_info *sbi = EXT4_SB(sb);
3754
3755         journal->j_commit_interval = sbi->s_commit_interval;
3756         journal->j_min_batch_time = sbi->s_min_batch_time;
3757         journal->j_max_batch_time = sbi->s_max_batch_time;
3758
3759         write_lock(&journal->j_state_lock);
3760         if (test_opt(sb, BARRIER))
3761                 journal->j_flags |= JBD2_BARRIER;
3762         else
3763                 journal->j_flags &= ~JBD2_BARRIER;
3764         if (test_opt(sb, DATA_ERR_ABORT))
3765                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3766         else
3767                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3768         write_unlock(&journal->j_state_lock);
3769 }
3770
3771 static journal_t *ext4_get_journal(struct super_block *sb,
3772                                    unsigned int journal_inum)
3773 {
3774         struct inode *journal_inode;
3775         journal_t *journal;
3776
3777         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3778
3779         /* First, test for the existence of a valid inode on disk.  Bad
3780          * things happen if we iget() an unused inode, as the subsequent
3781          * iput() will try to delete it. */
3782
3783         journal_inode = ext4_iget(sb, journal_inum);
3784         if (IS_ERR(journal_inode)) {
3785                 ext4_msg(sb, KERN_ERR, "no journal found");
3786                 return NULL;
3787         }
3788         if (!journal_inode->i_nlink) {
3789                 make_bad_inode(journal_inode);
3790                 iput(journal_inode);
3791                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3792                 return NULL;
3793         }
3794
3795         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3796                   journal_inode, journal_inode->i_size);
3797         if (!S_ISREG(journal_inode->i_mode)) {
3798                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3799                 iput(journal_inode);
3800                 return NULL;
3801         }
3802
3803         journal = jbd2_journal_init_inode(journal_inode);
3804         if (!journal) {
3805                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3806                 iput(journal_inode);
3807                 return NULL;
3808         }
3809         journal->j_private = sb;
3810         ext4_init_journal_params(sb, journal);
3811         return journal;
3812 }
3813
3814 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3815                                        dev_t j_dev)
3816 {
3817         struct buffer_head *bh;
3818         journal_t *journal;
3819         ext4_fsblk_t start;
3820         ext4_fsblk_t len;
3821         int hblock, blocksize;
3822         ext4_fsblk_t sb_block;
3823         unsigned long offset;
3824         struct ext4_super_block *es;
3825         struct block_device *bdev;
3826
3827         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3828
3829         bdev = ext4_blkdev_get(j_dev, sb);
3830         if (bdev == NULL)
3831                 return NULL;
3832
3833         blocksize = sb->s_blocksize;
3834         hblock = bdev_logical_block_size(bdev);
3835         if (blocksize < hblock) {
3836                 ext4_msg(sb, KERN_ERR,
3837                         "blocksize too small for journal device");
3838                 goto out_bdev;
3839         }
3840
3841         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3842         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3843         set_blocksize(bdev, blocksize);
3844         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3845                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3846                        "external journal");
3847                 goto out_bdev;
3848         }
3849
3850         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3851         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3852             !(le32_to_cpu(es->s_feature_incompat) &
3853               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3854                 ext4_msg(sb, KERN_ERR, "external journal has "
3855                                         "bad superblock");
3856                 brelse(bh);
3857                 goto out_bdev;
3858         }
3859
3860         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3861                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3862                 brelse(bh);
3863                 goto out_bdev;
3864         }
3865
3866         len = ext4_blocks_count(es);
3867         start = sb_block + 1;
3868         brelse(bh);     /* we're done with the superblock */
3869
3870         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3871                                         start, len, blocksize);
3872         if (!journal) {
3873                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3874                 goto out_bdev;
3875         }
3876         journal->j_private = sb;
3877         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3878         wait_on_buffer(journal->j_sb_buffer);
3879         if (!buffer_uptodate(journal->j_sb_buffer)) {
3880                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3881                 goto out_journal;
3882         }
3883         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3884                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3885                                         "user (unsupported) - %d",
3886                         be32_to_cpu(journal->j_superblock->s_nr_users));
3887                 goto out_journal;
3888         }
3889         EXT4_SB(sb)->journal_bdev = bdev;
3890         ext4_init_journal_params(sb, journal);
3891         return journal;
3892
3893 out_journal:
3894         jbd2_journal_destroy(journal);
3895 out_bdev:
3896         ext4_blkdev_put(bdev);
3897         return NULL;
3898 }
3899
3900 static int ext4_load_journal(struct super_block *sb,
3901                              struct ext4_super_block *es,
3902                              unsigned long journal_devnum)
3903 {
3904         journal_t *journal;
3905         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3906         dev_t journal_dev;
3907         int err = 0;
3908         int really_read_only;
3909
3910         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3911
3912         if (journal_devnum &&
3913             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3914                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3915                         "numbers have changed");
3916                 journal_dev = new_decode_dev(journal_devnum);
3917         } else
3918                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3919
3920         really_read_only = bdev_read_only(sb->s_bdev);
3921
3922         /*
3923          * Are we loading a blank journal or performing recovery after a
3924          * crash?  For recovery, we need to check in advance whether we
3925          * can get read-write access to the device.
3926          */
3927         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3928                 if (sb->s_flags & MS_RDONLY) {
3929                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3930                                         "required on readonly filesystem");
3931                         if (really_read_only) {
3932                                 ext4_msg(sb, KERN_ERR, "write access "
3933                                         "unavailable, cannot proceed");
3934                                 return -EROFS;
3935                         }
3936                         ext4_msg(sb, KERN_INFO, "write access will "
3937                                "be enabled during recovery");
3938                 }
3939         }
3940
3941         if (journal_inum && journal_dev) {
3942                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3943                        "and inode journals!");
3944                 return -EINVAL;
3945         }
3946
3947         if (journal_inum) {
3948                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3949                         return -EINVAL;
3950         } else {
3951                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3952                         return -EINVAL;
3953         }
3954
3955         if (!(journal->j_flags & JBD2_BARRIER))
3956                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3957
3958         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3959                 err = jbd2_journal_wipe(journal, !really_read_only);
3960         if (!err) {
3961                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3962                 if (save)
3963                         memcpy(save, ((char *) es) +
3964                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3965                 err = jbd2_journal_load(journal);
3966                 if (save)
3967                         memcpy(((char *) es) + EXT4_S_ERR_START,
3968                                save, EXT4_S_ERR_LEN);
3969                 kfree(save);
3970         }
3971
3972         if (err) {
3973                 ext4_msg(sb, KERN_ERR, "error loading journal");
3974                 jbd2_journal_destroy(journal);
3975                 return err;
3976         }
3977
3978         EXT4_SB(sb)->s_journal = journal;
3979         ext4_clear_journal_err(sb, es);
3980
3981         if (!really_read_only && journal_devnum &&
3982             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3983                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3984
3985                 /* Make sure we flush the recovery flag to disk. */
3986                 ext4_commit_super(sb, 1);
3987         }
3988
3989         return 0;
3990 }
3991
3992 static int ext4_commit_super(struct super_block *sb, int sync)
3993 {
3994         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3995         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3996         int error = 0;
3997
3998         if (!sbh || block_device_ejected(sb))
3999                 return error;
4000         if (buffer_write_io_error(sbh)) {
4001                 /*
4002                  * Oh, dear.  A previous attempt to write the
4003                  * superblock failed.  This could happen because the
4004                  * USB device was yanked out.  Or it could happen to
4005                  * be a transient write error and maybe the block will
4006                  * be remapped.  Nothing we can do but to retry the
4007                  * write and hope for the best.
4008                  */
4009                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4010                        "superblock detected");
4011                 clear_buffer_write_io_error(sbh);
4012                 set_buffer_uptodate(sbh);
4013         }
4014         /*
4015          * If the file system is mounted read-only, don't update the
4016          * superblock write time.  This avoids updating the superblock
4017          * write time when we are mounting the root file system
4018          * read/only but we need to replay the journal; at that point,
4019          * for people who are east of GMT and who make their clock
4020          * tick in localtime for Windows bug-for-bug compatibility,
4021          * the clock is set in the future, and this will cause e2fsck
4022          * to complain and force a full file system check.
4023          */
4024         if (!(sb->s_flags & MS_RDONLY))
4025                 es->s_wtime = cpu_to_le32(get_seconds());
4026         if (sb->s_bdev->bd_part)
4027                 es->s_kbytes_written =
4028                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4029                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4030                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4031         else
4032                 es->s_kbytes_written =
4033                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4034         ext4_free_blocks_count_set(es,
4035                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4036                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4037         es->s_free_inodes_count =
4038                 cpu_to_le32(percpu_counter_sum_positive(
4039                                 &EXT4_SB(sb)->s_freeinodes_counter));
4040         sb->s_dirt = 0;
4041         BUFFER_TRACE(sbh, "marking dirty");
4042         mark_buffer_dirty(sbh);
4043         if (sync) {
4044                 error = sync_dirty_buffer(sbh);
4045                 if (error)
4046                         return error;
4047
4048                 error = buffer_write_io_error(sbh);
4049                 if (error) {
4050                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4051                                "superblock");
4052                         clear_buffer_write_io_error(sbh);
4053                         set_buffer_uptodate(sbh);
4054                 }
4055         }
4056         return error;
4057 }
4058
4059 /*
4060  * Have we just finished recovery?  If so, and if we are mounting (or
4061  * remounting) the filesystem readonly, then we will end up with a
4062  * consistent fs on disk.  Record that fact.
4063  */
4064 static void ext4_mark_recovery_complete(struct super_block *sb,
4065                                         struct ext4_super_block *es)
4066 {
4067         journal_t *journal = EXT4_SB(sb)->s_journal;
4068
4069         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4070                 BUG_ON(journal != NULL);
4071                 return;
4072         }
4073         jbd2_journal_lock_updates(journal);
4074         if (jbd2_journal_flush(journal) < 0)
4075                 goto out;
4076
4077         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4078             sb->s_flags & MS_RDONLY) {
4079                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4080                 ext4_commit_super(sb, 1);
4081         }
4082
4083 out:
4084         jbd2_journal_unlock_updates(journal);
4085 }
4086
4087 /*
4088  * If we are mounting (or read-write remounting) a filesystem whose journal
4089  * has recorded an error from a previous lifetime, move that error to the
4090  * main filesystem now.
4091  */
4092 static void ext4_clear_journal_err(struct super_block *sb,
4093                                    struct ext4_super_block *es)
4094 {
4095         journal_t *journal;
4096         int j_errno;
4097         const char *errstr;
4098
4099         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4100
4101         journal = EXT4_SB(sb)->s_journal;
4102
4103         /*
4104          * Now check for any error status which may have been recorded in the
4105          * journal by a prior ext4_error() or ext4_abort()
4106          */
4107
4108         j_errno = jbd2_journal_errno(journal);
4109         if (j_errno) {
4110                 char nbuf[16];
4111
4112                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4113                 ext4_warning(sb, "Filesystem error recorded "
4114                              "from previous mount: %s", errstr);
4115                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4116
4117                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4118                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4119                 ext4_commit_super(sb, 1);
4120
4121                 jbd2_journal_clear_err(journal);
4122         }
4123 }
4124
4125 /*
4126  * Force the running and committing transactions to commit,
4127  * and wait on the commit.
4128  */
4129 int ext4_force_commit(struct super_block *sb)
4130 {
4131         journal_t *journal;
4132         int ret = 0;
4133
4134         if (sb->s_flags & MS_RDONLY)
4135                 return 0;
4136
4137         journal = EXT4_SB(sb)->s_journal;
4138         if (journal) {
4139                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4140                 ret = ext4_journal_force_commit(journal);
4141         }
4142
4143         return ret;
4144 }
4145
4146 static void ext4_write_super(struct super_block *sb)
4147 {
4148         lock_super(sb);
4149         ext4_commit_super(sb, 1);
4150         unlock_super(sb);
4151 }
4152
4153 static int ext4_sync_fs(struct super_block *sb, int wait)
4154 {
4155         int ret = 0;
4156         tid_t target;
4157         struct ext4_sb_info *sbi = EXT4_SB(sb);
4158
4159         trace_ext4_sync_fs(sb, wait);
4160         flush_workqueue(sbi->dio_unwritten_wq);
4161         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4162                 if (wait)
4163                         jbd2_log_wait_commit(sbi->s_journal, target);
4164         }
4165         return ret;
4166 }
4167
4168 /*
4169  * LVM calls this function before a (read-only) snapshot is created.  This
4170  * gives us a chance to flush the journal completely and mark the fs clean.
4171  *
4172  * Note that only this function cannot bring a filesystem to be in a clean
4173  * state independently, because ext4 prevents a new handle from being started
4174  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4175  * the upper layer.
4176  */
4177 static int ext4_freeze(struct super_block *sb)
4178 {
4179         int error = 0;
4180         journal_t *journal;
4181
4182         if (sb->s_flags & MS_RDONLY)
4183                 return 0;
4184
4185         journal = EXT4_SB(sb)->s_journal;
4186
4187         /* Now we set up the journal barrier. */
4188         jbd2_journal_lock_updates(journal);
4189
4190         /*
4191          * Don't clear the needs_recovery flag if we failed to flush
4192          * the journal.
4193          */
4194         error = jbd2_journal_flush(journal);
4195         if (error < 0)
4196                 goto out;
4197
4198         /* Journal blocked and flushed, clear needs_recovery flag. */
4199         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4200         error = ext4_commit_super(sb, 1);
4201 out:
4202         /* we rely on s_frozen to stop further updates */
4203         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4204         return error;
4205 }
4206
4207 /*
4208  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4209  * flag here, even though the filesystem is not technically dirty yet.
4210  */
4211 static int ext4_unfreeze(struct super_block *sb)
4212 {
4213         if (sb->s_flags & MS_RDONLY)
4214                 return 0;
4215
4216         lock_super(sb);
4217         /* Reset the needs_recovery flag before the fs is unlocked. */
4218         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4219         ext4_commit_super(sb, 1);
4220         unlock_super(sb);
4221         return 0;
4222 }
4223
4224 /*
4225  * Structure to save mount options for ext4_remount's benefit
4226  */
4227 struct ext4_mount_options {
4228         unsigned long s_mount_opt;
4229         unsigned long s_mount_opt2;
4230         kuid_t s_resuid;
4231         kgid_t s_resgid;
4232         unsigned long s_commit_interval;
4233         u32 s_min_batch_time, s_max_batch_time;
4234 #ifdef CONFIG_QUOTA
4235         int s_jquota_fmt;
4236         char *s_qf_names[MAXQUOTAS];
4237 #endif
4238 };
4239
4240 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4241 {
4242         struct ext4_super_block *es;
4243         struct ext4_sb_info *sbi = EXT4_SB(sb);
4244         unsigned long old_sb_flags;
4245         struct ext4_mount_options old_opts;
4246         int enable_quota = 0;
4247         ext4_group_t g;
4248         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4249         int err = 0;
4250 #ifdef CONFIG_QUOTA
4251         int i;
4252 #endif
4253         char *orig_data = kstrdup(data, GFP_KERNEL);
4254
4255         /* Store the original options */
4256         lock_super(sb);
4257         old_sb_flags = sb->s_flags;
4258         old_opts.s_mount_opt = sbi->s_mount_opt;
4259         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4260         old_opts.s_resuid = sbi->s_resuid;
4261         old_opts.s_resgid = sbi->s_resgid;
4262         old_opts.s_commit_interval = sbi->s_commit_interval;
4263         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4264         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4265 #ifdef CONFIG_QUOTA
4266         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4267         for (i = 0; i < MAXQUOTAS; i++)
4268                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4269 #endif
4270         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4271                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4272
4273         /*
4274          * Allow the "check" option to be passed as a remount option.
4275          */
4276         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4277                 err = -EINVAL;
4278                 goto restore_opts;
4279         }
4280
4281         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4282                 ext4_abort(sb, "Abort forced by user");
4283
4284         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4285                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4286
4287         es = sbi->s_es;
4288
4289         if (sbi->s_journal) {
4290                 ext4_init_journal_params(sb, sbi->s_journal);
4291                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4292         }
4293
4294         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4295                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4296                         err = -EROFS;
4297                         goto restore_opts;
4298                 }
4299
4300                 if (*flags & MS_RDONLY) {
4301                         err = dquot_suspend(sb, -1);
4302                         if (err < 0)
4303                                 goto restore_opts;
4304
4305                         /*
4306                          * First of all, the unconditional stuff we have to do
4307                          * to disable replay of the journal when we next remount
4308                          */
4309                         sb->s_flags |= MS_RDONLY;
4310
4311                         /*
4312                          * OK, test if we are remounting a valid rw partition
4313                          * readonly, and if so set the rdonly flag and then
4314                          * mark the partition as valid again.
4315                          */
4316                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4317                             (sbi->s_mount_state & EXT4_VALID_FS))
4318                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4319
4320                         if (sbi->s_journal)
4321                                 ext4_mark_recovery_complete(sb, es);
4322                 } else {
4323                         /* Make sure we can mount this feature set readwrite */
4324                         if (!ext4_feature_set_ok(sb, 0)) {
4325                                 err = -EROFS;
4326                                 goto restore_opts;
4327                         }
4328                         /*
4329                          * Make sure the group descriptor checksums
4330                          * are sane.  If they aren't, refuse to remount r/w.
4331                          */
4332                         for (g = 0; g < sbi->s_groups_count; g++) {
4333                                 struct ext4_group_desc *gdp =
4334                                         ext4_get_group_desc(sb, g, NULL);
4335
4336                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4337                                         ext4_msg(sb, KERN_ERR,
4338                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4339                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4340                                                le16_to_cpu(gdp->bg_checksum));
4341                                         err = -EINVAL;
4342                                         goto restore_opts;
4343                                 }
4344                         }
4345
4346                         /*
4347                          * If we have an unprocessed orphan list hanging
4348                          * around from a previously readonly bdev mount,
4349                          * require a full umount/remount for now.
4350                          */
4351                         if (es->s_last_orphan) {
4352                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4353                                        "remount RDWR because of unprocessed "
4354                                        "orphan inode list.  Please "
4355                                        "umount/remount instead");
4356                                 err = -EINVAL;
4357                                 goto restore_opts;
4358                         }
4359
4360                         /*
4361                          * Mounting a RDONLY partition read-write, so reread
4362                          * and store the current valid flag.  (It may have
4363                          * been changed by e2fsck since we originally mounted
4364                          * the partition.)
4365                          */
4366                         if (sbi->s_journal)
4367                                 ext4_clear_journal_err(sb, es);
4368                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4369                         if (!ext4_setup_super(sb, es, 0))
4370                                 sb->s_flags &= ~MS_RDONLY;
4371                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4372                                                      EXT4_FEATURE_INCOMPAT_MMP))
4373                                 if (ext4_multi_mount_protect(sb,
4374                                                 le64_to_cpu(es->s_mmp_block))) {
4375                                         err = -EROFS;
4376                                         goto restore_opts;
4377                                 }
4378                         enable_quota = 1;
4379                 }
4380         }
4381
4382         /*
4383          * Reinitialize lazy itable initialization thread based on
4384          * current settings
4385          */
4386         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4387                 ext4_unregister_li_request(sb);
4388         else {
4389                 ext4_group_t first_not_zeroed;
4390                 first_not_zeroed = ext4_has_uninit_itable(sb);
4391                 ext4_register_li_request(sb, first_not_zeroed);
4392         }
4393
4394         ext4_setup_system_zone(sb);
4395         if (sbi->s_journal == NULL)
4396                 ext4_commit_super(sb, 1);
4397
4398 #ifdef CONFIG_QUOTA
4399         /* Release old quota file names */
4400         for (i = 0; i < MAXQUOTAS; i++)
4401                 if (old_opts.s_qf_names[i] &&
4402                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4403                         kfree(old_opts.s_qf_names[i]);
4404 #endif
4405         unlock_super(sb);
4406         if (enable_quota)
4407                 dquot_resume(sb, -1);
4408
4409         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4410         kfree(orig_data);
4411         return 0;
4412
4413 restore_opts:
4414         sb->s_flags = old_sb_flags;
4415         sbi->s_mount_opt = old_opts.s_mount_opt;
4416         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4417         sbi->s_resuid = old_opts.s_resuid;
4418         sbi->s_resgid = old_opts.s_resgid;
4419         sbi->s_commit_interval = old_opts.s_commit_interval;
4420         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4421         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4422 #ifdef CONFIG_QUOTA
4423         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4424         for (i = 0; i < MAXQUOTAS; i++) {
4425                 if (sbi->s_qf_names[i] &&
4426                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4427                         kfree(sbi->s_qf_names[i]);
4428                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4429         }
4430 #endif
4431         unlock_super(sb);
4432         kfree(orig_data);
4433         return err;
4434 }
4435
4436 /*
4437  * Note: calculating the overhead so we can be compatible with
4438  * historical BSD practice is quite difficult in the face of
4439  * clusters/bigalloc.  This is because multiple metadata blocks from
4440  * different block group can end up in the same allocation cluster.
4441  * Calculating the exact overhead in the face of clustered allocation
4442  * requires either O(all block bitmaps) in memory or O(number of block
4443  * groups**2) in time.  We will still calculate the superblock for
4444  * older file systems --- and if we come across with a bigalloc file
4445  * system with zero in s_overhead_clusters the estimate will be close to
4446  * correct especially for very large cluster sizes --- but for newer
4447  * file systems, it's better to calculate this figure once at mkfs
4448  * time, and store it in the superblock.  If the superblock value is
4449  * present (even for non-bigalloc file systems), we will use it.
4450  */
4451 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4452 {
4453         struct super_block *sb = dentry->d_sb;
4454         struct ext4_sb_info *sbi = EXT4_SB(sb);
4455         struct ext4_super_block *es = sbi->s_es;
4456         struct ext4_group_desc *gdp;
4457         u64 fsid;
4458         s64 bfree;
4459
4460         if (test_opt(sb, MINIX_DF)) {
4461                 sbi->s_overhead_last = 0;
4462         } else if (es->s_overhead_clusters) {
4463                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4464         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4465                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4466                 ext4_fsblk_t overhead = 0;
4467
4468                 /*
4469                  * Compute the overhead (FS structures).  This is constant
4470                  * for a given filesystem unless the number of block groups
4471                  * changes so we cache the previous value until it does.
4472                  */
4473
4474                 /*
4475                  * All of the blocks before first_data_block are
4476                  * overhead
4477                  */
4478                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4479
4480                 /*
4481                  * Add the overhead found in each block group
4482                  */
4483                 for (i = 0; i < ngroups; i++) {
4484                         gdp = ext4_get_group_desc(sb, i, NULL);
4485                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4486                         cond_resched();
4487                 }
4488                 sbi->s_overhead_last = overhead;
4489                 smp_wmb();
4490                 sbi->s_blocks_last = ext4_blocks_count(es);
4491         }
4492
4493         buf->f_type = EXT4_SUPER_MAGIC;
4494         buf->f_bsize = sb->s_blocksize;
4495         buf->f_blocks = (ext4_blocks_count(es) -
4496                          EXT4_C2B(sbi, sbi->s_overhead_last));
4497         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4498                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4499         /* prevent underflow in case that few free space is available */
4500         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4501         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4502         if (buf->f_bfree < ext4_r_blocks_count(es))
4503                 buf->f_bavail = 0;
4504         buf->f_files = le32_to_cpu(es->s_inodes_count);
4505         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4506         buf->f_namelen = EXT4_NAME_LEN;
4507         fsid = le64_to_cpup((void *)es->s_uuid) ^
4508                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4509         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4510         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4511
4512         return 0;
4513 }
4514
4515 /* Helper function for writing quotas on sync - we need to start transaction
4516  * before quota file is locked for write. Otherwise the are possible deadlocks:
4517  * Process 1                         Process 2
4518  * ext4_create()                     quota_sync()
4519  *   jbd2_journal_start()                  write_dquot()
4520  *   dquot_initialize()                         down(dqio_mutex)
4521  *     down(dqio_mutex)                    jbd2_journal_start()
4522  *
4523  */
4524
4525 #ifdef CONFIG_QUOTA
4526
4527 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4528 {
4529         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4530 }
4531
4532 static int ext4_write_dquot(struct dquot *dquot)
4533 {
4534         int ret, err;
4535         handle_t *handle;
4536         struct inode *inode;
4537
4538         inode = dquot_to_inode(dquot);
4539         handle = ext4_journal_start(inode,
4540                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4541         if (IS_ERR(handle))
4542                 return PTR_ERR(handle);
4543         ret = dquot_commit(dquot);
4544         err = ext4_journal_stop(handle);
4545         if (!ret)
4546                 ret = err;
4547         return ret;
4548 }
4549
4550 static int ext4_acquire_dquot(struct dquot *dquot)
4551 {
4552         int ret, err;
4553         handle_t *handle;
4554
4555         handle = ext4_journal_start(dquot_to_inode(dquot),
4556                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4557         if (IS_ERR(handle))
4558                 return PTR_ERR(handle);
4559         ret = dquot_acquire(dquot);
4560         err = ext4_journal_stop(handle);
4561         if (!ret)
4562                 ret = err;
4563         return ret;
4564 }
4565
4566 static int ext4_release_dquot(struct dquot *dquot)
4567 {
4568         int ret, err;
4569         handle_t *handle;
4570
4571         handle = ext4_journal_start(dquot_to_inode(dquot),
4572                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4573         if (IS_ERR(handle)) {
4574                 /* Release dquot anyway to avoid endless cycle in dqput() */
4575                 dquot_release(dquot);
4576                 return PTR_ERR(handle);
4577         }
4578         ret = dquot_release(dquot);
4579         err = ext4_journal_stop(handle);
4580         if (!ret)
4581                 ret = err;
4582         return ret;
4583 }
4584
4585 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4586 {
4587         /* Are we journaling quotas? */
4588         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4589             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4590                 dquot_mark_dquot_dirty(dquot);
4591                 return ext4_write_dquot(dquot);
4592         } else {
4593                 return dquot_mark_dquot_dirty(dquot);
4594         }
4595 }
4596
4597 static int ext4_write_info(struct super_block *sb, int type)
4598 {
4599         int ret, err;
4600         handle_t *handle;
4601
4602         /* Data block + inode block */
4603         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4604         if (IS_ERR(handle))
4605                 return PTR_ERR(handle);
4606         ret = dquot_commit_info(sb, type);
4607         err = ext4_journal_stop(handle);
4608         if (!ret)
4609                 ret = err;
4610         return ret;
4611 }
4612
4613 /*
4614  * Turn on quotas during mount time - we need to find
4615  * the quota file and such...
4616  */
4617 static int ext4_quota_on_mount(struct super_block *sb, int type)
4618 {
4619         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4620                                         EXT4_SB(sb)->s_jquota_fmt, type);
4621 }
4622
4623 /*
4624  * Standard function to be called on quota_on
4625  */
4626 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4627                          struct path *path)
4628 {
4629         int err;
4630
4631         if (!test_opt(sb, QUOTA))
4632                 return -EINVAL;
4633
4634         /* Quotafile not on the same filesystem? */
4635         if (path->dentry->d_sb != sb)
4636                 return -EXDEV;
4637         /* Journaling quota? */
4638         if (EXT4_SB(sb)->s_qf_names[type]) {
4639                 /* Quotafile not in fs root? */
4640                 if (path->dentry->d_parent != sb->s_root)
4641                         ext4_msg(sb, KERN_WARNING,
4642                                 "Quota file not on filesystem root. "
4643                                 "Journaled quota will not work");
4644         }
4645
4646         /*
4647          * When we journal data on quota file, we have to flush journal to see
4648          * all updates to the file when we bypass pagecache...
4649          */
4650         if (EXT4_SB(sb)->s_journal &&
4651             ext4_should_journal_data(path->dentry->d_inode)) {
4652                 /*
4653                  * We don't need to lock updates but journal_flush() could
4654                  * otherwise be livelocked...
4655                  */
4656                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4657                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4658                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4659                 if (err)
4660                         return err;
4661         }
4662
4663         return dquot_quota_on(sb, type, format_id, path);
4664 }
4665
4666 static int ext4_quota_off(struct super_block *sb, int type)
4667 {
4668         struct inode *inode = sb_dqopt(sb)->files[type];
4669         handle_t *handle;
4670
4671         /* Force all delayed allocation blocks to be allocated.
4672          * Caller already holds s_umount sem */
4673         if (test_opt(sb, DELALLOC))
4674                 sync_filesystem(sb);
4675
4676         if (!inode)
4677                 goto out;
4678
4679         /* Update modification times of quota files when userspace can
4680          * start looking at them */
4681         handle = ext4_journal_start(inode, 1);
4682         if (IS_ERR(handle))
4683                 goto out;
4684         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4685         ext4_mark_inode_dirty(handle, inode);
4686         ext4_journal_stop(handle);
4687
4688 out:
4689         return dquot_quota_off(sb, type);
4690 }
4691
4692 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4693  * acquiring the locks... As quota files are never truncated and quota code
4694  * itself serializes the operations (and no one else should touch the files)
4695  * we don't have to be afraid of races */
4696 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4697                                size_t len, loff_t off)
4698 {
4699         struct inode *inode = sb_dqopt(sb)->files[type];
4700         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4701         int err = 0;
4702         int offset = off & (sb->s_blocksize - 1);
4703         int tocopy;
4704         size_t toread;
4705         struct buffer_head *bh;
4706         loff_t i_size = i_size_read(inode);
4707
4708         if (off > i_size)
4709                 return 0;
4710         if (off+len > i_size)
4711                 len = i_size-off;
4712         toread = len;
4713         while (toread > 0) {
4714                 tocopy = sb->s_blocksize - offset < toread ?
4715                                 sb->s_blocksize - offset : toread;
4716                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4717                 if (err)
4718                         return err;
4719                 if (!bh)        /* A hole? */
4720                         memset(data, 0, tocopy);
4721                 else
4722                         memcpy(data, bh->b_data+offset, tocopy);
4723                 brelse(bh);
4724                 offset = 0;
4725                 toread -= tocopy;
4726                 data += tocopy;
4727                 blk++;
4728         }
4729         return len;
4730 }
4731
4732 /* Write to quotafile (we know the transaction is already started and has
4733  * enough credits) */
4734 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4735                                 const char *data, size_t len, loff_t off)
4736 {
4737         struct inode *inode = sb_dqopt(sb)->files[type];
4738         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4739         int err = 0;
4740         int offset = off & (sb->s_blocksize - 1);
4741         struct buffer_head *bh;
4742         handle_t *handle = journal_current_handle();
4743
4744         if (EXT4_SB(sb)->s_journal && !handle) {
4745                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4746                         " cancelled because transaction is not started",
4747                         (unsigned long long)off, (unsigned long long)len);
4748                 return -EIO;
4749         }
4750         /*
4751          * Since we account only one data block in transaction credits,
4752          * then it is impossible to cross a block boundary.
4753          */
4754         if (sb->s_blocksize - offset < len) {
4755                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4756                         " cancelled because not block aligned",
4757                         (unsigned long long)off, (unsigned long long)len);
4758                 return -EIO;
4759         }
4760
4761         bh = ext4_bread(handle, inode, blk, 1, &err);
4762         if (!bh)
4763                 goto out;
4764         err = ext4_journal_get_write_access(handle, bh);
4765         if (err) {
4766                 brelse(bh);
4767                 goto out;
4768         }
4769         lock_buffer(bh);
4770         memcpy(bh->b_data+offset, data, len);
4771         flush_dcache_page(bh->b_page);
4772         unlock_buffer(bh);
4773         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4774         brelse(bh);
4775 out:
4776         if (err)
4777                 return err;
4778         if (inode->i_size < off + len) {
4779                 i_size_write(inode, off + len);
4780                 EXT4_I(inode)->i_disksize = inode->i_size;
4781                 ext4_mark_inode_dirty(handle, inode);
4782         }
4783         return len;
4784 }
4785
4786 #endif
4787
4788 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4789                        const char *dev_name, void *data)
4790 {
4791         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4792 }
4793
4794 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4795 static inline void register_as_ext2(void)
4796 {
4797         int err = register_filesystem(&ext2_fs_type);
4798         if (err)
4799                 printk(KERN_WARNING
4800                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4801 }
4802
4803 static inline void unregister_as_ext2(void)
4804 {
4805         unregister_filesystem(&ext2_fs_type);
4806 }
4807
4808 static inline int ext2_feature_set_ok(struct super_block *sb)
4809 {
4810         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4811                 return 0;
4812         if (sb->s_flags & MS_RDONLY)
4813                 return 1;
4814         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4815                 return 0;
4816         return 1;
4817 }
4818 MODULE_ALIAS("ext2");
4819 #else
4820 static inline void register_as_ext2(void) { }
4821 static inline void unregister_as_ext2(void) { }
4822 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4823 #endif
4824
4825 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4826 static inline void register_as_ext3(void)
4827 {
4828         int err = register_filesystem(&ext3_fs_type);
4829         if (err)
4830                 printk(KERN_WARNING
4831                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4832 }
4833
4834 static inline void unregister_as_ext3(void)
4835 {
4836         unregister_filesystem(&ext3_fs_type);
4837 }
4838
4839 static inline int ext3_feature_set_ok(struct super_block *sb)
4840 {
4841         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4842                 return 0;
4843         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4844                 return 0;
4845         if (sb->s_flags & MS_RDONLY)
4846                 return 1;
4847         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4848                 return 0;
4849         return 1;
4850 }
4851 MODULE_ALIAS("ext3");
4852 #else
4853 static inline void register_as_ext3(void) { }
4854 static inline void unregister_as_ext3(void) { }
4855 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4856 #endif
4857
4858 static struct file_system_type ext4_fs_type = {
4859         .owner          = THIS_MODULE,
4860         .name           = "ext4",
4861         .mount          = ext4_mount,
4862         .kill_sb        = kill_block_super,
4863         .fs_flags       = FS_REQUIRES_DEV,
4864 };
4865
4866 static int __init ext4_init_feat_adverts(void)
4867 {
4868         struct ext4_features *ef;
4869         int ret = -ENOMEM;
4870
4871         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4872         if (!ef)
4873                 goto out;
4874
4875         ef->f_kobj.kset = ext4_kset;
4876         init_completion(&ef->f_kobj_unregister);
4877         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4878                                    "features");
4879         if (ret) {
4880                 kfree(ef);
4881                 goto out;
4882         }
4883
4884         ext4_feat = ef;
4885         ret = 0;
4886 out:
4887         return ret;
4888 }
4889
4890 static void ext4_exit_feat_adverts(void)
4891 {
4892         kobject_put(&ext4_feat->f_kobj);
4893         wait_for_completion(&ext4_feat->f_kobj_unregister);
4894         kfree(ext4_feat);
4895 }
4896
4897 /* Shared across all ext4 file systems */
4898 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4899 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4900
4901 static int __init ext4_init_fs(void)
4902 {
4903         int i, err;
4904
4905         ext4_li_info = NULL;
4906         mutex_init(&ext4_li_mtx);
4907
4908         ext4_check_flag_values();
4909
4910         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4911                 mutex_init(&ext4__aio_mutex[i]);
4912                 init_waitqueue_head(&ext4__ioend_wq[i]);
4913         }
4914
4915         err = ext4_init_pageio();
4916         if (err)
4917                 return err;
4918         err = ext4_init_system_zone();
4919         if (err)
4920                 goto out6;
4921         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4922         if (!ext4_kset)
4923                 goto out5;
4924         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4925
4926         err = ext4_init_feat_adverts();
4927         if (err)
4928                 goto out4;
4929
4930         err = ext4_init_mballoc();
4931         if (err)
4932                 goto out3;
4933
4934         err = ext4_init_xattr();
4935         if (err)
4936                 goto out2;
4937         err = init_inodecache();
4938         if (err)
4939                 goto out1;
4940         register_as_ext3();
4941         register_as_ext2();
4942         err = register_filesystem(&ext4_fs_type);
4943         if (err)
4944                 goto out;
4945
4946         return 0;
4947 out:
4948         unregister_as_ext2();
4949         unregister_as_ext3();
4950         destroy_inodecache();
4951 out1:
4952         ext4_exit_xattr();
4953 out2:
4954         ext4_exit_mballoc();
4955 out3:
4956         ext4_exit_feat_adverts();
4957 out4:
4958         if (ext4_proc_root)
4959                 remove_proc_entry("fs/ext4", NULL);
4960         kset_unregister(ext4_kset);
4961 out5:
4962         ext4_exit_system_zone();
4963 out6:
4964         ext4_exit_pageio();
4965         return err;
4966 }
4967
4968 static void __exit ext4_exit_fs(void)
4969 {
4970         ext4_destroy_lazyinit_thread();
4971         unregister_as_ext2();
4972         unregister_as_ext3();
4973         unregister_filesystem(&ext4_fs_type);
4974         destroy_inodecache();
4975         ext4_exit_xattr();
4976         ext4_exit_mballoc();
4977         ext4_exit_feat_adverts();
4978         remove_proc_entry("fs/ext4", NULL);
4979         kset_unregister(ext4_kset);
4980         ext4_exit_system_zone();
4981         ext4_exit_pageio();
4982 }
4983
4984 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4985 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4986 MODULE_LICENSE("GPL");
4987 module_init(ext4_init_fs)
4988 module_exit(ext4_exit_fs)