ceebaf853beb74c7e139e63f46bf44975696566b
[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         end_writeback(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 0;
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 0;
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 0;
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 0;
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 0;
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         int arg = 0;
1452
1453         if (args->from && match_int(args, &arg))
1454                 return -1;
1455         switch (token) {
1456         case Opt_noacl:
1457         case Opt_nouser_xattr:
1458                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1459                 break;
1460         case Opt_sb:
1461                 return 1;       /* handled by get_sb_block() */
1462         case Opt_removed:
1463                 ext4_msg(sb, KERN_WARNING,
1464                          "Ignoring removed %s option", opt);
1465                 return 1;
1466         case Opt_resuid:
1467                 sbi->s_resuid = arg;
1468                 return 1;
1469         case Opt_resgid:
1470                 sbi->s_resgid = arg;
1471                 return 1;
1472         case Opt_abort:
1473                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1474                 return 1;
1475         case Opt_i_version:
1476                 sb->s_flags |= MS_I_VERSION;
1477                 return 1;
1478         case Opt_journal_dev:
1479                 if (is_remount) {
1480                         ext4_msg(sb, KERN_ERR,
1481                                  "Cannot specify journal on remount");
1482                         return -1;
1483                 }
1484                 *journal_devnum = arg;
1485                 return 1;
1486         case Opt_journal_ioprio:
1487                 if (arg < 0 || arg > 7)
1488                         return -1;
1489                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1490                 return 1;
1491         }
1492
1493         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1494                 if (token != m->token)
1495                         continue;
1496                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1497                         return -1;
1498                 if (m->flags & MOPT_EXPLICIT)
1499                         set_opt2(sb, EXPLICIT_DELALLOC);
1500                 if (m->flags & MOPT_CLEAR_ERR)
1501                         clear_opt(sb, ERRORS_MASK);
1502                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1503                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1504                                  "options when quota turned on");
1505                         return -1;
1506                 }
1507
1508                 if (m->flags & MOPT_NOSUPPORT) {
1509                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1510                 } else if (token == Opt_commit) {
1511                         if (arg == 0)
1512                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1513                         sbi->s_commit_interval = HZ * arg;
1514                 } else if (token == Opt_max_batch_time) {
1515                         if (arg == 0)
1516                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1517                         sbi->s_max_batch_time = arg;
1518                 } else if (token == Opt_min_batch_time) {
1519                         sbi->s_min_batch_time = arg;
1520                 } else if (token == Opt_inode_readahead_blks) {
1521                         if (arg > (1 << 30))
1522                                 return -1;
1523                         if (arg && !is_power_of_2(arg)) {
1524                                 ext4_msg(sb, KERN_ERR,
1525                                          "EXT4-fs: inode_readahead_blks"
1526                                          " must be a power of 2");
1527                                 return -1;
1528                         }
1529                         sbi->s_inode_readahead_blks = arg;
1530                 } else if (token == Opt_init_itable) {
1531                         set_opt(sb, INIT_INODE_TABLE);
1532                         if (!args->from)
1533                                 arg = EXT4_DEF_LI_WAIT_MULT;
1534                         sbi->s_li_wait_mult = arg;
1535                 } else if (token == Opt_stripe) {
1536                         sbi->s_stripe = arg;
1537                 } else if (m->flags & MOPT_DATAJ) {
1538                         if (is_remount) {
1539                                 if (!sbi->s_journal)
1540                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1541                                 else if (test_opt(sb, DATA_FLAGS) !=
1542                                          m->mount_opt) {
1543                                         ext4_msg(sb, KERN_ERR,
1544                                          "Cannot change data mode on remount");
1545                                         return -1;
1546                                 }
1547                         } else {
1548                                 clear_opt(sb, DATA_FLAGS);
1549                                 sbi->s_mount_opt |= m->mount_opt;
1550                         }
1551 #ifdef CONFIG_QUOTA
1552                 } else if (token == Opt_usrjquota) {
1553                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1554                                 return -1;
1555                 } else if (token == Opt_grpjquota) {
1556                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1557                                 return -1;
1558                 } else if (token == Opt_offusrjquota) {
1559                         if (!clear_qf_name(sb, USRQUOTA))
1560                                 return -1;
1561                 } else if (token == Opt_offgrpjquota) {
1562                         if (!clear_qf_name(sb, GRPQUOTA))
1563                                 return -1;
1564                 } else if (m->flags & MOPT_QFMT) {
1565                         if (sb_any_quota_loaded(sb) &&
1566                             sbi->s_jquota_fmt != m->mount_opt) {
1567                                 ext4_msg(sb, KERN_ERR, "Cannot "
1568                                          "change journaled quota options "
1569                                          "when quota turned on");
1570                                 return -1;
1571                         }
1572                         sbi->s_jquota_fmt = m->mount_opt;
1573 #endif
1574                 } else {
1575                         if (!args->from)
1576                                 arg = 1;
1577                         if (m->flags & MOPT_CLEAR)
1578                                 arg = !arg;
1579                         else if (unlikely(!(m->flags & MOPT_SET))) {
1580                                 ext4_msg(sb, KERN_WARNING,
1581                                          "buggy handling of option %s", opt);
1582                                 WARN_ON(1);
1583                                 return -1;
1584                         }
1585                         if (arg != 0)
1586                                 sbi->s_mount_opt |= m->mount_opt;
1587                         else
1588                                 sbi->s_mount_opt &= ~m->mount_opt;
1589                 }
1590                 return 1;
1591         }
1592         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1593                  "or missing value", opt);
1594         return -1;
1595 }
1596
1597 static int parse_options(char *options, struct super_block *sb,
1598                          unsigned long *journal_devnum,
1599                          unsigned int *journal_ioprio,
1600                          int is_remount)
1601 {
1602         struct ext4_sb_info *sbi = EXT4_SB(sb);
1603         char *p;
1604         substring_t args[MAX_OPT_ARGS];
1605         int token;
1606
1607         if (!options)
1608                 return 1;
1609
1610         while ((p = strsep(&options, ",")) != NULL) {
1611                 if (!*p)
1612                         continue;
1613                 /*
1614                  * Initialize args struct so we know whether arg was
1615                  * found; some options take optional arguments.
1616                  */
1617                 args[0].to = args[0].from = 0;
1618                 token = match_token(p, tokens, args);
1619                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1620                                      journal_ioprio, is_remount) < 0)
1621                         return 0;
1622         }
1623 #ifdef CONFIG_QUOTA
1624         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1625                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1626                         clear_opt(sb, USRQUOTA);
1627
1628                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1629                         clear_opt(sb, GRPQUOTA);
1630
1631                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1632                         ext4_msg(sb, KERN_ERR, "old and new quota "
1633                                         "format mixing");
1634                         return 0;
1635                 }
1636
1637                 if (!sbi->s_jquota_fmt) {
1638                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1639                                         "not specified");
1640                         return 0;
1641                 }
1642         } else {
1643                 if (sbi->s_jquota_fmt) {
1644                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1645                                         "specified with no journaling "
1646                                         "enabled");
1647                         return 0;
1648                 }
1649         }
1650 #endif
1651         return 1;
1652 }
1653
1654 static inline void ext4_show_quota_options(struct seq_file *seq,
1655                                            struct super_block *sb)
1656 {
1657 #if defined(CONFIG_QUOTA)
1658         struct ext4_sb_info *sbi = EXT4_SB(sb);
1659
1660         if (sbi->s_jquota_fmt) {
1661                 char *fmtname = "";
1662
1663                 switch (sbi->s_jquota_fmt) {
1664                 case QFMT_VFS_OLD:
1665                         fmtname = "vfsold";
1666                         break;
1667                 case QFMT_VFS_V0:
1668                         fmtname = "vfsv0";
1669                         break;
1670                 case QFMT_VFS_V1:
1671                         fmtname = "vfsv1";
1672                         break;
1673                 }
1674                 seq_printf(seq, ",jqfmt=%s", fmtname);
1675         }
1676
1677         if (sbi->s_qf_names[USRQUOTA])
1678                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1679
1680         if (sbi->s_qf_names[GRPQUOTA])
1681                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1682
1683         if (test_opt(sb, USRQUOTA))
1684                 seq_puts(seq, ",usrquota");
1685
1686         if (test_opt(sb, GRPQUOTA))
1687                 seq_puts(seq, ",grpquota");
1688 #endif
1689 }
1690
1691 static const char *token2str(int token)
1692 {
1693         static const struct match_token *t;
1694
1695         for (t = tokens; t->token != Opt_err; t++)
1696                 if (t->token == token && !strchr(t->pattern, '='))
1697                         break;
1698         return t->pattern;
1699 }
1700
1701 /*
1702  * Show an option if
1703  *  - it's set to a non-default value OR
1704  *  - if the per-sb default is different from the global default
1705  */
1706 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1707                               int nodefs)
1708 {
1709         struct ext4_sb_info *sbi = EXT4_SB(sb);
1710         struct ext4_super_block *es = sbi->s_es;
1711         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1712         const struct mount_opts *m;
1713         char sep = nodefs ? '\n' : ',';
1714
1715 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1716 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1717
1718         if (sbi->s_sb_block != 1)
1719                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1720
1721         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1722                 int want_set = m->flags & MOPT_SET;
1723                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1724                     (m->flags & MOPT_CLEAR_ERR))
1725                         continue;
1726                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1727                         continue; /* skip if same as the default */
1728                 if ((want_set &&
1729                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1730                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1731                         continue; /* select Opt_noFoo vs Opt_Foo */
1732                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1733         }
1734
1735         if (nodefs || sbi->s_resuid != EXT4_DEF_RESUID ||
1736             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1737                 SEQ_OPTS_PRINT("resuid=%u", sbi->s_resuid);
1738         if (nodefs || sbi->s_resgid != EXT4_DEF_RESGID ||
1739             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1740                 SEQ_OPTS_PRINT("resgid=%u", sbi->s_resgid);
1741         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1742         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1743                 SEQ_OPTS_PUTS("errors=remount-ro");
1744         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1745                 SEQ_OPTS_PUTS("errors=continue");
1746         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1747                 SEQ_OPTS_PUTS("errors=panic");
1748         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1749                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1750         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1751                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1752         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1753                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1754         if (sb->s_flags & MS_I_VERSION)
1755                 SEQ_OPTS_PUTS("i_version");
1756         if (nodefs || sbi->s_stripe)
1757                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1758         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1759                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1760                         SEQ_OPTS_PUTS("data=journal");
1761                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1762                         SEQ_OPTS_PUTS("data=ordered");
1763                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1764                         SEQ_OPTS_PUTS("data=writeback");
1765         }
1766         if (nodefs ||
1767             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1768                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1769                                sbi->s_inode_readahead_blks);
1770
1771         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1772                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1773                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1774
1775         ext4_show_quota_options(seq, sb);
1776         return 0;
1777 }
1778
1779 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1780 {
1781         return _ext4_show_options(seq, root->d_sb, 0);
1782 }
1783
1784 static int options_seq_show(struct seq_file *seq, void *offset)
1785 {
1786         struct super_block *sb = seq->private;
1787         int rc;
1788
1789         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1790         rc = _ext4_show_options(seq, sb, 1);
1791         seq_puts(seq, "\n");
1792         return rc;
1793 }
1794
1795 static int options_open_fs(struct inode *inode, struct file *file)
1796 {
1797         return single_open(file, options_seq_show, PDE(inode)->data);
1798 }
1799
1800 static const struct file_operations ext4_seq_options_fops = {
1801         .owner = THIS_MODULE,
1802         .open = options_open_fs,
1803         .read = seq_read,
1804         .llseek = seq_lseek,
1805         .release = single_release,
1806 };
1807
1808 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1809                             int read_only)
1810 {
1811         struct ext4_sb_info *sbi = EXT4_SB(sb);
1812         int res = 0;
1813
1814         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1815                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1816                          "forcing read-only mode");
1817                 res = MS_RDONLY;
1818         }
1819         if (read_only)
1820                 goto done;
1821         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1822                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1823                          "running e2fsck is recommended");
1824         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1825                 ext4_msg(sb, KERN_WARNING,
1826                          "warning: mounting fs with errors, "
1827                          "running e2fsck is recommended");
1828         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1829                  le16_to_cpu(es->s_mnt_count) >=
1830                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1831                 ext4_msg(sb, KERN_WARNING,
1832                          "warning: maximal mount count reached, "
1833                          "running e2fsck is recommended");
1834         else if (le32_to_cpu(es->s_checkinterval) &&
1835                 (le32_to_cpu(es->s_lastcheck) +
1836                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1837                 ext4_msg(sb, KERN_WARNING,
1838                          "warning: checktime reached, "
1839                          "running e2fsck is recommended");
1840         if (!sbi->s_journal)
1841                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1842         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1843                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1844         le16_add_cpu(&es->s_mnt_count, 1);
1845         es->s_mtime = cpu_to_le32(get_seconds());
1846         ext4_update_dynamic_rev(sb);
1847         if (sbi->s_journal)
1848                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1849
1850         ext4_commit_super(sb, 1);
1851 done:
1852         if (test_opt(sb, DEBUG))
1853                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1854                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1855                         sb->s_blocksize,
1856                         sbi->s_groups_count,
1857                         EXT4_BLOCKS_PER_GROUP(sb),
1858                         EXT4_INODES_PER_GROUP(sb),
1859                         sbi->s_mount_opt, sbi->s_mount_opt2);
1860
1861         cleancache_init_fs(sb);
1862         return res;
1863 }
1864
1865 static int ext4_fill_flex_info(struct super_block *sb)
1866 {
1867         struct ext4_sb_info *sbi = EXT4_SB(sb);
1868         struct ext4_group_desc *gdp = NULL;
1869         ext4_group_t flex_group_count;
1870         ext4_group_t flex_group;
1871         unsigned int groups_per_flex = 0;
1872         size_t size;
1873         int i;
1874
1875         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1876         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1877                 sbi->s_log_groups_per_flex = 0;
1878                 return 1;
1879         }
1880         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1881
1882         /* We allocate both existing and potentially added groups */
1883         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1884                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1885                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1886         size = flex_group_count * sizeof(struct flex_groups);
1887         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1888         if (sbi->s_flex_groups == NULL) {
1889                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1890                          flex_group_count);
1891                 goto failed;
1892         }
1893
1894         for (i = 0; i < sbi->s_groups_count; i++) {
1895                 gdp = ext4_get_group_desc(sb, i, NULL);
1896
1897                 flex_group = ext4_flex_group(sbi, i);
1898                 atomic_add(ext4_free_inodes_count(sb, gdp),
1899                            &sbi->s_flex_groups[flex_group].free_inodes);
1900                 atomic_add(ext4_free_group_clusters(sb, gdp),
1901                            &sbi->s_flex_groups[flex_group].free_clusters);
1902                 atomic_add(ext4_used_dirs_count(sb, gdp),
1903                            &sbi->s_flex_groups[flex_group].used_dirs);
1904         }
1905
1906         return 1;
1907 failed:
1908         return 0;
1909 }
1910
1911 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1912                             struct ext4_group_desc *gdp)
1913 {
1914         __u16 crc = 0;
1915
1916         if (sbi->s_es->s_feature_ro_compat &
1917             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1918                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1919                 __le32 le_group = cpu_to_le32(block_group);
1920
1921                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1922                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1923                 crc = crc16(crc, (__u8 *)gdp, offset);
1924                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1925                 /* for checksum of struct ext4_group_desc do the rest...*/
1926                 if ((sbi->s_es->s_feature_incompat &
1927                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1928                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1929                         crc = crc16(crc, (__u8 *)gdp + offset,
1930                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1931                                         offset);
1932         }
1933
1934         return cpu_to_le16(crc);
1935 }
1936
1937 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1938                                 struct ext4_group_desc *gdp)
1939 {
1940         if ((sbi->s_es->s_feature_ro_compat &
1941              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1942             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1943                 return 0;
1944
1945         return 1;
1946 }
1947
1948 /* Called at mount-time, super-block is locked */
1949 static int ext4_check_descriptors(struct super_block *sb,
1950                                   ext4_group_t *first_not_zeroed)
1951 {
1952         struct ext4_sb_info *sbi = EXT4_SB(sb);
1953         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1954         ext4_fsblk_t last_block;
1955         ext4_fsblk_t block_bitmap;
1956         ext4_fsblk_t inode_bitmap;
1957         ext4_fsblk_t inode_table;
1958         int flexbg_flag = 0;
1959         ext4_group_t i, grp = sbi->s_groups_count;
1960
1961         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1962                 flexbg_flag = 1;
1963
1964         ext4_debug("Checking group descriptors");
1965
1966         for (i = 0; i < sbi->s_groups_count; i++) {
1967                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1968
1969                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1970                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1971                 else
1972                         last_block = first_block +
1973                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1974
1975                 if ((grp == sbi->s_groups_count) &&
1976                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
1977                         grp = i;
1978
1979                 block_bitmap = ext4_block_bitmap(sb, gdp);
1980                 if (block_bitmap < first_block || block_bitmap > last_block) {
1981                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1982                                "Block bitmap for group %u not in group "
1983                                "(block %llu)!", i, block_bitmap);
1984                         return 0;
1985                 }
1986                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1987                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1988                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1989                                "Inode bitmap for group %u not in group "
1990                                "(block %llu)!", i, inode_bitmap);
1991                         return 0;
1992                 }
1993                 inode_table = ext4_inode_table(sb, gdp);
1994                 if (inode_table < first_block ||
1995                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1996                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1997                                "Inode table for group %u not in group "
1998                                "(block %llu)!", i, inode_table);
1999                         return 0;
2000                 }
2001                 ext4_lock_group(sb, i);
2002                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2003                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2004                                  "Checksum for group %u failed (%u!=%u)",
2005                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2006                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2007                         if (!(sb->s_flags & MS_RDONLY)) {
2008                                 ext4_unlock_group(sb, i);
2009                                 return 0;
2010                         }
2011                 }
2012                 ext4_unlock_group(sb, i);
2013                 if (!flexbg_flag)
2014                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2015         }
2016         if (NULL != first_not_zeroed)
2017                 *first_not_zeroed = grp;
2018
2019         ext4_free_blocks_count_set(sbi->s_es,
2020                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2021         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2022         return 1;
2023 }
2024
2025 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2026  * the superblock) which were deleted from all directories, but held open by
2027  * a process at the time of a crash.  We walk the list and try to delete these
2028  * inodes at recovery time (only with a read-write filesystem).
2029  *
2030  * In order to keep the orphan inode chain consistent during traversal (in
2031  * case of crash during recovery), we link each inode into the superblock
2032  * orphan list_head and handle it the same way as an inode deletion during
2033  * normal operation (which journals the operations for us).
2034  *
2035  * We only do an iget() and an iput() on each inode, which is very safe if we
2036  * accidentally point at an in-use or already deleted inode.  The worst that
2037  * can happen in this case is that we get a "bit already cleared" message from
2038  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2039  * e2fsck was run on this filesystem, and it must have already done the orphan
2040  * inode cleanup for us, so we can safely abort without any further action.
2041  */
2042 static void ext4_orphan_cleanup(struct super_block *sb,
2043                                 struct ext4_super_block *es)
2044 {
2045         unsigned int s_flags = sb->s_flags;
2046         int nr_orphans = 0, nr_truncates = 0;
2047 #ifdef CONFIG_QUOTA
2048         int i;
2049 #endif
2050         if (!es->s_last_orphan) {
2051                 jbd_debug(4, "no orphan inodes to clean up\n");
2052                 return;
2053         }
2054
2055         if (bdev_read_only(sb->s_bdev)) {
2056                 ext4_msg(sb, KERN_ERR, "write access "
2057                         "unavailable, skipping orphan cleanup");
2058                 return;
2059         }
2060
2061         /* Check if feature set would not allow a r/w mount */
2062         if (!ext4_feature_set_ok(sb, 0)) {
2063                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2064                          "unknown ROCOMPAT features");
2065                 return;
2066         }
2067
2068         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2069                 if (es->s_last_orphan)
2070                         jbd_debug(1, "Errors on filesystem, "
2071                                   "clearing orphan list.\n");
2072                 es->s_last_orphan = 0;
2073                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2074                 return;
2075         }
2076
2077         if (s_flags & MS_RDONLY) {
2078                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2079                 sb->s_flags &= ~MS_RDONLY;
2080         }
2081 #ifdef CONFIG_QUOTA
2082         /* Needed for iput() to work correctly and not trash data */
2083         sb->s_flags |= MS_ACTIVE;
2084         /* Turn on quotas so that they are updated correctly */
2085         for (i = 0; i < MAXQUOTAS; i++) {
2086                 if (EXT4_SB(sb)->s_qf_names[i]) {
2087                         int ret = ext4_quota_on_mount(sb, i);
2088                         if (ret < 0)
2089                                 ext4_msg(sb, KERN_ERR,
2090                                         "Cannot turn on journaled "
2091                                         "quota: error %d", ret);
2092                 }
2093         }
2094 #endif
2095
2096         while (es->s_last_orphan) {
2097                 struct inode *inode;
2098
2099                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2100                 if (IS_ERR(inode)) {
2101                         es->s_last_orphan = 0;
2102                         break;
2103                 }
2104
2105                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2106                 dquot_initialize(inode);
2107                 if (inode->i_nlink) {
2108                         ext4_msg(sb, KERN_DEBUG,
2109                                 "%s: truncating inode %lu to %lld bytes",
2110                                 __func__, inode->i_ino, inode->i_size);
2111                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2112                                   inode->i_ino, inode->i_size);
2113                         ext4_truncate(inode);
2114                         nr_truncates++;
2115                 } else {
2116                         ext4_msg(sb, KERN_DEBUG,
2117                                 "%s: deleting unreferenced inode %lu",
2118                                 __func__, inode->i_ino);
2119                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2120                                   inode->i_ino);
2121                         nr_orphans++;
2122                 }
2123                 iput(inode);  /* The delete magic happens here! */
2124         }
2125
2126 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2127
2128         if (nr_orphans)
2129                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2130                        PLURAL(nr_orphans));
2131         if (nr_truncates)
2132                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2133                        PLURAL(nr_truncates));
2134 #ifdef CONFIG_QUOTA
2135         /* Turn quotas off */
2136         for (i = 0; i < MAXQUOTAS; i++) {
2137                 if (sb_dqopt(sb)->files[i])
2138                         dquot_quota_off(sb, i);
2139         }
2140 #endif
2141         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2142 }
2143
2144 /*
2145  * Maximal extent format file size.
2146  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2147  * extent format containers, within a sector_t, and within i_blocks
2148  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2149  * so that won't be a limiting factor.
2150  *
2151  * However there is other limiting factor. We do store extents in the form
2152  * of starting block and length, hence the resulting length of the extent
2153  * covering maximum file size must fit into on-disk format containers as
2154  * well. Given that length is always by 1 unit bigger than max unit (because
2155  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2156  *
2157  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2158  */
2159 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2160 {
2161         loff_t res;
2162         loff_t upper_limit = MAX_LFS_FILESIZE;
2163
2164         /* small i_blocks in vfs inode? */
2165         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2166                 /*
2167                  * CONFIG_LBDAF is not enabled implies the inode
2168                  * i_block represent total blocks in 512 bytes
2169                  * 32 == size of vfs inode i_blocks * 8
2170                  */
2171                 upper_limit = (1LL << 32) - 1;
2172
2173                 /* total blocks in file system block size */
2174                 upper_limit >>= (blkbits - 9);
2175                 upper_limit <<= blkbits;
2176         }
2177
2178         /*
2179          * 32-bit extent-start container, ee_block. We lower the maxbytes
2180          * by one fs block, so ee_len can cover the extent of maximum file
2181          * size
2182          */
2183         res = (1LL << 32) - 1;
2184         res <<= blkbits;
2185
2186         /* Sanity check against vm- & vfs- imposed limits */
2187         if (res > upper_limit)
2188                 res = upper_limit;
2189
2190         return res;
2191 }
2192
2193 /*
2194  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2195  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2196  * We need to be 1 filesystem block less than the 2^48 sector limit.
2197  */
2198 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2199 {
2200         loff_t res = EXT4_NDIR_BLOCKS;
2201         int meta_blocks;
2202         loff_t upper_limit;
2203         /* This is calculated to be the largest file size for a dense, block
2204          * mapped file such that the file's total number of 512-byte sectors,
2205          * including data and all indirect blocks, does not exceed (2^48 - 1).
2206          *
2207          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2208          * number of 512-byte sectors of the file.
2209          */
2210
2211         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2212                 /*
2213                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2214                  * the inode i_block field represents total file blocks in
2215                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2216                  */
2217                 upper_limit = (1LL << 32) - 1;
2218
2219                 /* total blocks in file system block size */
2220                 upper_limit >>= (bits - 9);
2221
2222         } else {
2223                 /*
2224                  * We use 48 bit ext4_inode i_blocks
2225                  * With EXT4_HUGE_FILE_FL set the i_blocks
2226                  * represent total number of blocks in
2227                  * file system block size
2228                  */
2229                 upper_limit = (1LL << 48) - 1;
2230
2231         }
2232
2233         /* indirect blocks */
2234         meta_blocks = 1;
2235         /* double indirect blocks */
2236         meta_blocks += 1 + (1LL << (bits-2));
2237         /* tripple indirect blocks */
2238         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2239
2240         upper_limit -= meta_blocks;
2241         upper_limit <<= bits;
2242
2243         res += 1LL << (bits-2);
2244         res += 1LL << (2*(bits-2));
2245         res += 1LL << (3*(bits-2));
2246         res <<= bits;
2247         if (res > upper_limit)
2248                 res = upper_limit;
2249
2250         if (res > MAX_LFS_FILESIZE)
2251                 res = MAX_LFS_FILESIZE;
2252
2253         return res;
2254 }
2255
2256 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2257                                    ext4_fsblk_t logical_sb_block, int nr)
2258 {
2259         struct ext4_sb_info *sbi = EXT4_SB(sb);
2260         ext4_group_t bg, first_meta_bg;
2261         int has_super = 0;
2262
2263         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2264
2265         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2266             nr < first_meta_bg)
2267                 return logical_sb_block + nr + 1;
2268         bg = sbi->s_desc_per_block * nr;
2269         if (ext4_bg_has_super(sb, bg))
2270                 has_super = 1;
2271
2272         return (has_super + ext4_group_first_block_no(sb, bg));
2273 }
2274
2275 /**
2276  * ext4_get_stripe_size: Get the stripe size.
2277  * @sbi: In memory super block info
2278  *
2279  * If we have specified it via mount option, then
2280  * use the mount option value. If the value specified at mount time is
2281  * greater than the blocks per group use the super block value.
2282  * If the super block value is greater than blocks per group return 0.
2283  * Allocator needs it be less than blocks per group.
2284  *
2285  */
2286 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2287 {
2288         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2289         unsigned long stripe_width =
2290                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2291         int ret;
2292
2293         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2294                 ret = sbi->s_stripe;
2295         else if (stripe_width <= sbi->s_blocks_per_group)
2296                 ret = stripe_width;
2297         else if (stride <= sbi->s_blocks_per_group)
2298                 ret = stride;
2299         else
2300                 ret = 0;
2301
2302         /*
2303          * If the stripe width is 1, this makes no sense and
2304          * we set it to 0 to turn off stripe handling code.
2305          */
2306         if (ret <= 1)
2307                 ret = 0;
2308
2309         return ret;
2310 }
2311
2312 /* sysfs supprt */
2313
2314 struct ext4_attr {
2315         struct attribute attr;
2316         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2317         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2318                          const char *, size_t);
2319         int offset;
2320 };
2321
2322 static int parse_strtoul(const char *buf,
2323                 unsigned long max, unsigned long *value)
2324 {
2325         char *endp;
2326
2327         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2328         endp = skip_spaces(endp);
2329         if (*endp || *value > max)
2330                 return -EINVAL;
2331
2332         return 0;
2333 }
2334
2335 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2336                                               struct ext4_sb_info *sbi,
2337                                               char *buf)
2338 {
2339         return snprintf(buf, PAGE_SIZE, "%llu\n",
2340                 (s64) EXT4_C2B(sbi,
2341                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2342 }
2343
2344 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2345                                          struct ext4_sb_info *sbi, char *buf)
2346 {
2347         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2348
2349         if (!sb->s_bdev->bd_part)
2350                 return snprintf(buf, PAGE_SIZE, "0\n");
2351         return snprintf(buf, PAGE_SIZE, "%lu\n",
2352                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2353                          sbi->s_sectors_written_start) >> 1);
2354 }
2355
2356 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2357                                           struct ext4_sb_info *sbi, char *buf)
2358 {
2359         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2360
2361         if (!sb->s_bdev->bd_part)
2362                 return snprintf(buf, PAGE_SIZE, "0\n");
2363         return snprintf(buf, PAGE_SIZE, "%llu\n",
2364                         (unsigned long long)(sbi->s_kbytes_written +
2365                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2366                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2367 }
2368
2369 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2370                                       struct ext4_sb_info *sbi, char *buf)
2371 {
2372         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2373 }
2374
2375 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2376                                         struct ext4_sb_info *sbi, char *buf)
2377 {
2378         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2379 }
2380
2381 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2382                                           struct ext4_sb_info *sbi,
2383                                           const char *buf, size_t count)
2384 {
2385         unsigned long t;
2386
2387         if (parse_strtoul(buf, 0x40000000, &t))
2388                 return -EINVAL;
2389
2390         if (t && !is_power_of_2(t))
2391                 return -EINVAL;
2392
2393         sbi->s_inode_readahead_blks = t;
2394         return count;
2395 }
2396
2397 static ssize_t sbi_ui_show(struct ext4_attr *a,
2398                            struct ext4_sb_info *sbi, char *buf)
2399 {
2400         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2401
2402         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2403 }
2404
2405 static ssize_t sbi_ui_store(struct ext4_attr *a,
2406                             struct ext4_sb_info *sbi,
2407                             const char *buf, size_t count)
2408 {
2409         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2410         unsigned long t;
2411
2412         if (parse_strtoul(buf, 0xffffffff, &t))
2413                 return -EINVAL;
2414         *ui = t;
2415         return count;
2416 }
2417
2418 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2419 static struct ext4_attr ext4_attr_##_name = {                   \
2420         .attr = {.name = __stringify(_name), .mode = _mode },   \
2421         .show   = _show,                                        \
2422         .store  = _store,                                       \
2423         .offset = offsetof(struct ext4_sb_info, _elname),       \
2424 }
2425 #define EXT4_ATTR(name, mode, show, store) \
2426 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2427
2428 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2429 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2430 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2431 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2432         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2433 #define ATTR_LIST(name) &ext4_attr_##name.attr
2434
2435 EXT4_RO_ATTR(delayed_allocation_blocks);
2436 EXT4_RO_ATTR(session_write_kbytes);
2437 EXT4_RO_ATTR(lifetime_write_kbytes);
2438 EXT4_RO_ATTR(extent_cache_hits);
2439 EXT4_RO_ATTR(extent_cache_misses);
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(extent_cache_hits),
2456         ATTR_LIST(extent_cache_misses),
2457         ATTR_LIST(inode_readahead_blks),
2458         ATTR_LIST(inode_goal),
2459         ATTR_LIST(mb_stats),
2460         ATTR_LIST(mb_max_to_scan),
2461         ATTR_LIST(mb_min_to_scan),
2462         ATTR_LIST(mb_order2_req),
2463         ATTR_LIST(mb_stream_req),
2464         ATTR_LIST(mb_group_prealloc),
2465         ATTR_LIST(max_writeback_mb_bump),
2466         NULL,
2467 };
2468
2469 /* Features this copy of ext4 supports */
2470 EXT4_INFO_ATTR(lazy_itable_init);
2471 EXT4_INFO_ATTR(batched_discard);
2472
2473 static struct attribute *ext4_feat_attrs[] = {
2474         ATTR_LIST(lazy_itable_init),
2475         ATTR_LIST(batched_discard),
2476         NULL,
2477 };
2478
2479 static ssize_t ext4_attr_show(struct kobject *kobj,
2480                               struct attribute *attr, char *buf)
2481 {
2482         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2483                                                 s_kobj);
2484         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2485
2486         return a->show ? a->show(a, sbi, buf) : 0;
2487 }
2488
2489 static ssize_t ext4_attr_store(struct kobject *kobj,
2490                                struct attribute *attr,
2491                                const char *buf, size_t len)
2492 {
2493         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2494                                                 s_kobj);
2495         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2496
2497         return a->store ? a->store(a, sbi, buf, len) : 0;
2498 }
2499
2500 static void ext4_sb_release(struct kobject *kobj)
2501 {
2502         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2503                                                 s_kobj);
2504         complete(&sbi->s_kobj_unregister);
2505 }
2506
2507 static const struct sysfs_ops ext4_attr_ops = {
2508         .show   = ext4_attr_show,
2509         .store  = ext4_attr_store,
2510 };
2511
2512 static struct kobj_type ext4_ktype = {
2513         .default_attrs  = ext4_attrs,
2514         .sysfs_ops      = &ext4_attr_ops,
2515         .release        = ext4_sb_release,
2516 };
2517
2518 static void ext4_feat_release(struct kobject *kobj)
2519 {
2520         complete(&ext4_feat->f_kobj_unregister);
2521 }
2522
2523 static struct kobj_type ext4_feat_ktype = {
2524         .default_attrs  = ext4_feat_attrs,
2525         .sysfs_ops      = &ext4_attr_ops,
2526         .release        = ext4_feat_release,
2527 };
2528
2529 /*
2530  * Check whether this filesystem can be mounted based on
2531  * the features present and the RDONLY/RDWR mount requested.
2532  * Returns 1 if this filesystem can be mounted as requested,
2533  * 0 if it cannot be.
2534  */
2535 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2536 {
2537         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2538                 ext4_msg(sb, KERN_ERR,
2539                         "Couldn't mount because of "
2540                         "unsupported optional features (%x)",
2541                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2542                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2543                 return 0;
2544         }
2545
2546         if (readonly)
2547                 return 1;
2548
2549         /* Check that feature set is OK for a read-write mount */
2550         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2551                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2552                          "unsupported optional features (%x)",
2553                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2554                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2555                 return 0;
2556         }
2557         /*
2558          * Large file size enabled file system can only be mounted
2559          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2560          */
2561         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2562                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2563                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2564                                  "cannot be mounted RDWR without "
2565                                  "CONFIG_LBDAF");
2566                         return 0;
2567                 }
2568         }
2569         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2570             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2571                 ext4_msg(sb, KERN_ERR,
2572                          "Can't support bigalloc feature without "
2573                          "extents feature\n");
2574                 return 0;
2575         }
2576         return 1;
2577 }
2578
2579 /*
2580  * This function is called once a day if we have errors logged
2581  * on the file system
2582  */
2583 static void print_daily_error_info(unsigned long arg)
2584 {
2585         struct super_block *sb = (struct super_block *) arg;
2586         struct ext4_sb_info *sbi;
2587         struct ext4_super_block *es;
2588
2589         sbi = EXT4_SB(sb);
2590         es = sbi->s_es;
2591
2592         if (es->s_error_count)
2593                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2594                          le32_to_cpu(es->s_error_count));
2595         if (es->s_first_error_time) {
2596                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2597                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2598                        (int) sizeof(es->s_first_error_func),
2599                        es->s_first_error_func,
2600                        le32_to_cpu(es->s_first_error_line));
2601                 if (es->s_first_error_ino)
2602                         printk(": inode %u",
2603                                le32_to_cpu(es->s_first_error_ino));
2604                 if (es->s_first_error_block)
2605                         printk(": block %llu", (unsigned long long)
2606                                le64_to_cpu(es->s_first_error_block));
2607                 printk("\n");
2608         }
2609         if (es->s_last_error_time) {
2610                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2611                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2612                        (int) sizeof(es->s_last_error_func),
2613                        es->s_last_error_func,
2614                        le32_to_cpu(es->s_last_error_line));
2615                 if (es->s_last_error_ino)
2616                         printk(": inode %u",
2617                                le32_to_cpu(es->s_last_error_ino));
2618                 if (es->s_last_error_block)
2619                         printk(": block %llu", (unsigned long long)
2620                                le64_to_cpu(es->s_last_error_block));
2621                 printk("\n");
2622         }
2623         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2624 }
2625
2626 /* Find next suitable group and run ext4_init_inode_table */
2627 static int ext4_run_li_request(struct ext4_li_request *elr)
2628 {
2629         struct ext4_group_desc *gdp = NULL;
2630         ext4_group_t group, ngroups;
2631         struct super_block *sb;
2632         unsigned long timeout = 0;
2633         int ret = 0;
2634
2635         sb = elr->lr_super;
2636         ngroups = EXT4_SB(sb)->s_groups_count;
2637
2638         for (group = elr->lr_next_group; group < ngroups; group++) {
2639                 gdp = ext4_get_group_desc(sb, group, NULL);
2640                 if (!gdp) {
2641                         ret = 1;
2642                         break;
2643                 }
2644
2645                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2646                         break;
2647         }
2648
2649         if (group == ngroups)
2650                 ret = 1;
2651
2652         if (!ret) {
2653                 timeout = jiffies;
2654                 ret = ext4_init_inode_table(sb, group,
2655                                             elr->lr_timeout ? 0 : 1);
2656                 if (elr->lr_timeout == 0) {
2657                         timeout = (jiffies - timeout) *
2658                                   elr->lr_sbi->s_li_wait_mult;
2659                         elr->lr_timeout = timeout;
2660                 }
2661                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2662                 elr->lr_next_group = group + 1;
2663         }
2664
2665         return ret;
2666 }
2667
2668 /*
2669  * Remove lr_request from the list_request and free the
2670  * request structure. Should be called with li_list_mtx held
2671  */
2672 static void ext4_remove_li_request(struct ext4_li_request *elr)
2673 {
2674         struct ext4_sb_info *sbi;
2675
2676         if (!elr)
2677                 return;
2678
2679         sbi = elr->lr_sbi;
2680
2681         list_del(&elr->lr_request);
2682         sbi->s_li_request = NULL;
2683         kfree(elr);
2684 }
2685
2686 static void ext4_unregister_li_request(struct super_block *sb)
2687 {
2688         mutex_lock(&ext4_li_mtx);
2689         if (!ext4_li_info) {
2690                 mutex_unlock(&ext4_li_mtx);
2691                 return;
2692         }
2693
2694         mutex_lock(&ext4_li_info->li_list_mtx);
2695         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2696         mutex_unlock(&ext4_li_info->li_list_mtx);
2697         mutex_unlock(&ext4_li_mtx);
2698 }
2699
2700 static struct task_struct *ext4_lazyinit_task;
2701
2702 /*
2703  * This is the function where ext4lazyinit thread lives. It walks
2704  * through the request list searching for next scheduled filesystem.
2705  * When such a fs is found, run the lazy initialization request
2706  * (ext4_rn_li_request) and keep track of the time spend in this
2707  * function. Based on that time we compute next schedule time of
2708  * the request. When walking through the list is complete, compute
2709  * next waking time and put itself into sleep.
2710  */
2711 static int ext4_lazyinit_thread(void *arg)
2712 {
2713         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2714         struct list_head *pos, *n;
2715         struct ext4_li_request *elr;
2716         unsigned long next_wakeup, cur;
2717
2718         BUG_ON(NULL == eli);
2719
2720 cont_thread:
2721         while (true) {
2722                 next_wakeup = MAX_JIFFY_OFFSET;
2723
2724                 mutex_lock(&eli->li_list_mtx);
2725                 if (list_empty(&eli->li_request_list)) {
2726                         mutex_unlock(&eli->li_list_mtx);
2727                         goto exit_thread;
2728                 }
2729
2730                 list_for_each_safe(pos, n, &eli->li_request_list) {
2731                         elr = list_entry(pos, struct ext4_li_request,
2732                                          lr_request);
2733
2734                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2735                                 if (ext4_run_li_request(elr) != 0) {
2736                                         /* error, remove the lazy_init job */
2737                                         ext4_remove_li_request(elr);
2738                                         continue;
2739                                 }
2740                         }
2741
2742                         if (time_before(elr->lr_next_sched, next_wakeup))
2743                                 next_wakeup = elr->lr_next_sched;
2744                 }
2745                 mutex_unlock(&eli->li_list_mtx);
2746
2747                 try_to_freeze();
2748
2749                 cur = jiffies;
2750                 if ((time_after_eq(cur, next_wakeup)) ||
2751                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2752                         cond_resched();
2753                         continue;
2754                 }
2755
2756                 schedule_timeout_interruptible(next_wakeup - cur);
2757
2758                 if (kthread_should_stop()) {
2759                         ext4_clear_request_list();
2760                         goto exit_thread;
2761                 }
2762         }
2763
2764 exit_thread:
2765         /*
2766          * It looks like the request list is empty, but we need
2767          * to check it under the li_list_mtx lock, to prevent any
2768          * additions into it, and of course we should lock ext4_li_mtx
2769          * to atomically free the list and ext4_li_info, because at
2770          * this point another ext4 filesystem could be registering
2771          * new one.
2772          */
2773         mutex_lock(&ext4_li_mtx);
2774         mutex_lock(&eli->li_list_mtx);
2775         if (!list_empty(&eli->li_request_list)) {
2776                 mutex_unlock(&eli->li_list_mtx);
2777                 mutex_unlock(&ext4_li_mtx);
2778                 goto cont_thread;
2779         }
2780         mutex_unlock(&eli->li_list_mtx);
2781         kfree(ext4_li_info);
2782         ext4_li_info = NULL;
2783         mutex_unlock(&ext4_li_mtx);
2784
2785         return 0;
2786 }
2787
2788 static void ext4_clear_request_list(void)
2789 {
2790         struct list_head *pos, *n;
2791         struct ext4_li_request *elr;
2792
2793         mutex_lock(&ext4_li_info->li_list_mtx);
2794         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2795                 elr = list_entry(pos, struct ext4_li_request,
2796                                  lr_request);
2797                 ext4_remove_li_request(elr);
2798         }
2799         mutex_unlock(&ext4_li_info->li_list_mtx);
2800 }
2801
2802 static int ext4_run_lazyinit_thread(void)
2803 {
2804         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2805                                          ext4_li_info, "ext4lazyinit");
2806         if (IS_ERR(ext4_lazyinit_task)) {
2807                 int err = PTR_ERR(ext4_lazyinit_task);
2808                 ext4_clear_request_list();
2809                 kfree(ext4_li_info);
2810                 ext4_li_info = NULL;
2811                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2812                                  "initialization thread\n",
2813                                  err);
2814                 return err;
2815         }
2816         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2817         return 0;
2818 }
2819
2820 /*
2821  * Check whether it make sense to run itable init. thread or not.
2822  * If there is at least one uninitialized inode table, return
2823  * corresponding group number, else the loop goes through all
2824  * groups and return total number of groups.
2825  */
2826 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2827 {
2828         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2829         struct ext4_group_desc *gdp = NULL;
2830
2831         for (group = 0; group < ngroups; group++) {
2832                 gdp = ext4_get_group_desc(sb, group, NULL);
2833                 if (!gdp)
2834                         continue;
2835
2836                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2837                         break;
2838         }
2839
2840         return group;
2841 }
2842
2843 static int ext4_li_info_new(void)
2844 {
2845         struct ext4_lazy_init *eli = NULL;
2846
2847         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2848         if (!eli)
2849                 return -ENOMEM;
2850
2851         INIT_LIST_HEAD(&eli->li_request_list);
2852         mutex_init(&eli->li_list_mtx);
2853
2854         eli->li_state |= EXT4_LAZYINIT_QUIT;
2855
2856         ext4_li_info = eli;
2857
2858         return 0;
2859 }
2860
2861 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2862                                             ext4_group_t start)
2863 {
2864         struct ext4_sb_info *sbi = EXT4_SB(sb);
2865         struct ext4_li_request *elr;
2866         unsigned long rnd;
2867
2868         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2869         if (!elr)
2870                 return NULL;
2871
2872         elr->lr_super = sb;
2873         elr->lr_sbi = sbi;
2874         elr->lr_next_group = start;
2875
2876         /*
2877          * Randomize first schedule time of the request to
2878          * spread the inode table initialization requests
2879          * better.
2880          */
2881         get_random_bytes(&rnd, sizeof(rnd));
2882         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2883                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2884
2885         return elr;
2886 }
2887
2888 static int ext4_register_li_request(struct super_block *sb,
2889                                     ext4_group_t first_not_zeroed)
2890 {
2891         struct ext4_sb_info *sbi = EXT4_SB(sb);
2892         struct ext4_li_request *elr;
2893         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2894         int ret = 0;
2895
2896         if (sbi->s_li_request != NULL) {
2897                 /*
2898                  * Reset timeout so it can be computed again, because
2899                  * s_li_wait_mult might have changed.
2900                  */
2901                 sbi->s_li_request->lr_timeout = 0;
2902                 return 0;
2903         }
2904
2905         if (first_not_zeroed == ngroups ||
2906             (sb->s_flags & MS_RDONLY) ||
2907             !test_opt(sb, INIT_INODE_TABLE))
2908                 return 0;
2909
2910         elr = ext4_li_request_new(sb, first_not_zeroed);
2911         if (!elr)
2912                 return -ENOMEM;
2913
2914         mutex_lock(&ext4_li_mtx);
2915
2916         if (NULL == ext4_li_info) {
2917                 ret = ext4_li_info_new();
2918                 if (ret)
2919                         goto out;
2920         }
2921
2922         mutex_lock(&ext4_li_info->li_list_mtx);
2923         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2924         mutex_unlock(&ext4_li_info->li_list_mtx);
2925
2926         sbi->s_li_request = elr;
2927         /*
2928          * set elr to NULL here since it has been inserted to
2929          * the request_list and the removal and free of it is
2930          * handled by ext4_clear_request_list from now on.
2931          */
2932         elr = NULL;
2933
2934         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2935                 ret = ext4_run_lazyinit_thread();
2936                 if (ret)
2937                         goto out;
2938         }
2939 out:
2940         mutex_unlock(&ext4_li_mtx);
2941         if (ret)
2942                 kfree(elr);
2943         return ret;
2944 }
2945
2946 /*
2947  * We do not need to lock anything since this is called on
2948  * module unload.
2949  */
2950 static void ext4_destroy_lazyinit_thread(void)
2951 {
2952         /*
2953          * If thread exited earlier
2954          * there's nothing to be done.
2955          */
2956         if (!ext4_li_info || !ext4_lazyinit_task)
2957                 return;
2958
2959         kthread_stop(ext4_lazyinit_task);
2960 }
2961
2962 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2963 {
2964         char *orig_data = kstrdup(data, GFP_KERNEL);
2965         struct buffer_head *bh;
2966         struct ext4_super_block *es = NULL;
2967         struct ext4_sb_info *sbi;
2968         ext4_fsblk_t block;
2969         ext4_fsblk_t sb_block = get_sb_block(&data);
2970         ext4_fsblk_t logical_sb_block;
2971         unsigned long offset = 0;
2972         unsigned long journal_devnum = 0;
2973         unsigned long def_mount_opts;
2974         struct inode *root;
2975         char *cp;
2976         const char *descr;
2977         int ret = -ENOMEM;
2978         int blocksize, clustersize;
2979         unsigned int db_count;
2980         unsigned int i;
2981         int needs_recovery, has_huge_files, has_bigalloc;
2982         __u64 blocks_count;
2983         int err;
2984         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2985         ext4_group_t first_not_zeroed;
2986
2987         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2988         if (!sbi)
2989                 goto out_free_orig;
2990
2991         sbi->s_blockgroup_lock =
2992                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
2993         if (!sbi->s_blockgroup_lock) {
2994                 kfree(sbi);
2995                 goto out_free_orig;
2996         }
2997         sb->s_fs_info = sbi;
2998         sbi->s_mount_opt = 0;
2999         sbi->s_resuid = EXT4_DEF_RESUID;
3000         sbi->s_resgid = EXT4_DEF_RESGID;
3001         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3002         sbi->s_sb_block = sb_block;
3003         if (sb->s_bdev->bd_part)
3004                 sbi->s_sectors_written_start =
3005                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3006
3007         /* Cleanup superblock name */
3008         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3009                 *cp = '!';
3010
3011         ret = -EINVAL;
3012         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3013         if (!blocksize) {
3014                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3015                 goto out_fail;
3016         }
3017
3018         /*
3019          * The ext4 superblock will not be buffer aligned for other than 1kB
3020          * block sizes.  We need to calculate the offset from buffer start.
3021          */
3022         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3023                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3024                 offset = do_div(logical_sb_block, blocksize);
3025         } else {
3026                 logical_sb_block = sb_block;
3027         }
3028
3029         if (!(bh = sb_bread(sb, logical_sb_block))) {
3030                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3031                 goto out_fail;
3032         }
3033         /*
3034          * Note: s_es must be initialized as soon as possible because
3035          *       some ext4 macro-instructions depend on its value
3036          */
3037         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3038         sbi->s_es = es;
3039         sb->s_magic = le16_to_cpu(es->s_magic);
3040         if (sb->s_magic != EXT4_SUPER_MAGIC)
3041                 goto cantfind_ext4;
3042         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3043
3044         /* Set defaults before we parse the mount options */
3045         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3046         set_opt(sb, INIT_INODE_TABLE);
3047         if (def_mount_opts & EXT4_DEFM_DEBUG)
3048                 set_opt(sb, DEBUG);
3049         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3050                 set_opt(sb, GRPID);
3051         if (def_mount_opts & EXT4_DEFM_UID16)
3052                 set_opt(sb, NO_UID32);
3053         /* xattr user namespace & acls are now defaulted on */
3054 #ifdef CONFIG_EXT4_FS_XATTR
3055         set_opt(sb, XATTR_USER);
3056 #endif
3057 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3058         set_opt(sb, POSIX_ACL);
3059 #endif
3060         set_opt(sb, MBLK_IO_SUBMIT);
3061         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3062                 set_opt(sb, JOURNAL_DATA);
3063         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3064                 set_opt(sb, ORDERED_DATA);
3065         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3066                 set_opt(sb, WRITEBACK_DATA);
3067
3068         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3069                 set_opt(sb, ERRORS_PANIC);
3070         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3071                 set_opt(sb, ERRORS_CONT);
3072         else
3073                 set_opt(sb, ERRORS_RO);
3074         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3075                 set_opt(sb, BLOCK_VALIDITY);
3076         if (def_mount_opts & EXT4_DEFM_DISCARD)
3077                 set_opt(sb, DISCARD);
3078
3079         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3080         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3081         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3082         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3083         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3084
3085         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3086                 set_opt(sb, BARRIER);
3087
3088         /*
3089          * enable delayed allocation by default
3090          * Use -o nodelalloc to turn it off
3091          */
3092         if (!IS_EXT3_SB(sb) &&
3093             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3094                 set_opt(sb, DELALLOC);
3095
3096         /*
3097          * set default s_li_wait_mult for lazyinit, for the case there is
3098          * no mount option specified.
3099          */
3100         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3101
3102         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3103                            &journal_devnum, &journal_ioprio, 0)) {
3104                 ext4_msg(sb, KERN_WARNING,
3105                          "failed to parse options in superblock: %s",
3106                          sbi->s_es->s_mount_opts);
3107         }
3108         sbi->s_def_mount_opt = sbi->s_mount_opt;
3109         if (!parse_options((char *) data, sb, &journal_devnum,
3110                            &journal_ioprio, 0))
3111                 goto failed_mount;
3112
3113         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3114                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3115                             "with data=journal disables delayed "
3116                             "allocation and O_DIRECT support!\n");
3117                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3118                         ext4_msg(sb, KERN_ERR, "can't mount with "
3119                                  "both data=journal and delalloc");
3120                         goto failed_mount;
3121                 }
3122                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3123                         ext4_msg(sb, KERN_ERR, "can't mount with "
3124                                  "both data=journal and delalloc");
3125                         goto failed_mount;
3126                 }
3127                 if (test_opt(sb, DELALLOC))
3128                         clear_opt(sb, DELALLOC);
3129         }
3130
3131         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3132         if (test_opt(sb, DIOREAD_NOLOCK)) {
3133                 if (blocksize < PAGE_SIZE) {
3134                         ext4_msg(sb, KERN_ERR, "can't mount with "
3135                                  "dioread_nolock if block size != PAGE_SIZE");
3136                         goto failed_mount;
3137                 }
3138         }
3139
3140         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3141                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3142
3143         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3144             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3145              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3146              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3147                 ext4_msg(sb, KERN_WARNING,
3148                        "feature flags set on rev 0 fs, "
3149                        "running e2fsck is recommended");
3150
3151         if (IS_EXT2_SB(sb)) {
3152                 if (ext2_feature_set_ok(sb))
3153                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3154                                  "using the ext4 subsystem");
3155                 else {
3156                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3157                                  "to feature incompatibilities");
3158                         goto failed_mount;
3159                 }
3160         }
3161
3162         if (IS_EXT3_SB(sb)) {
3163                 if (ext3_feature_set_ok(sb))
3164                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3165                                  "using the ext4 subsystem");
3166                 else {
3167                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3168                                  "to feature incompatibilities");
3169                         goto failed_mount;
3170                 }
3171         }
3172
3173         /*
3174          * Check feature flags regardless of the revision level, since we
3175          * previously didn't change the revision level when setting the flags,
3176          * so there is a chance incompat flags are set on a rev 0 filesystem.
3177          */
3178         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3179                 goto failed_mount;
3180
3181         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3182             blocksize > EXT4_MAX_BLOCK_SIZE) {
3183                 ext4_msg(sb, KERN_ERR,
3184                        "Unsupported filesystem blocksize %d", blocksize);
3185                 goto failed_mount;
3186         }
3187
3188         if (sb->s_blocksize != blocksize) {
3189                 /* Validate the filesystem blocksize */
3190                 if (!sb_set_blocksize(sb, blocksize)) {
3191                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3192                                         blocksize);
3193                         goto failed_mount;
3194                 }
3195
3196                 brelse(bh);
3197                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3198                 offset = do_div(logical_sb_block, blocksize);
3199                 bh = sb_bread(sb, logical_sb_block);
3200                 if (!bh) {
3201                         ext4_msg(sb, KERN_ERR,
3202                                "Can't read superblock on 2nd try");
3203                         goto failed_mount;
3204                 }
3205                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3206                 sbi->s_es = es;
3207                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3208                         ext4_msg(sb, KERN_ERR,
3209                                "Magic mismatch, very weird!");
3210                         goto failed_mount;
3211                 }
3212         }
3213
3214         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3215                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3216         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3217                                                       has_huge_files);
3218         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3219
3220         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3221                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3222                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3223         } else {
3224                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3225                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3226                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3227                     (!is_power_of_2(sbi->s_inode_size)) ||
3228                     (sbi->s_inode_size > blocksize)) {
3229                         ext4_msg(sb, KERN_ERR,
3230                                "unsupported inode size: %d",
3231                                sbi->s_inode_size);
3232                         goto failed_mount;
3233                 }
3234                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3235                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3236         }
3237
3238         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3239         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3240                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3241                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3242                     !is_power_of_2(sbi->s_desc_size)) {
3243                         ext4_msg(sb, KERN_ERR,
3244                                "unsupported descriptor size %lu",
3245                                sbi->s_desc_size);
3246                         goto failed_mount;
3247                 }
3248         } else
3249                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3250
3251         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3252         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3253         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3254                 goto cantfind_ext4;
3255
3256         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3257         if (sbi->s_inodes_per_block == 0)
3258                 goto cantfind_ext4;
3259         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3260                                         sbi->s_inodes_per_block;
3261         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3262         sbi->s_sbh = bh;
3263         sbi->s_mount_state = le16_to_cpu(es->s_state);
3264         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3265         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3266
3267         for (i = 0; i < 4; i++)
3268                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3269         sbi->s_def_hash_version = es->s_def_hash_version;
3270         i = le32_to_cpu(es->s_flags);
3271         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3272                 sbi->s_hash_unsigned = 3;
3273         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3274 #ifdef __CHAR_UNSIGNED__
3275                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3276                 sbi->s_hash_unsigned = 3;
3277 #else
3278                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3279 #endif
3280         }
3281
3282         /* Handle clustersize */
3283         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3284         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3285                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3286         if (has_bigalloc) {
3287                 if (clustersize < blocksize) {
3288                         ext4_msg(sb, KERN_ERR,
3289                                  "cluster size (%d) smaller than "
3290                                  "block size (%d)", clustersize, blocksize);
3291                         goto failed_mount;
3292                 }
3293                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3294                         le32_to_cpu(es->s_log_block_size);
3295                 sbi->s_clusters_per_group =
3296                         le32_to_cpu(es->s_clusters_per_group);
3297                 if (sbi->s_clusters_per_group > blocksize * 8) {
3298                         ext4_msg(sb, KERN_ERR,
3299                                  "#clusters per group too big: %lu",
3300                                  sbi->s_clusters_per_group);
3301                         goto failed_mount;
3302                 }
3303                 if (sbi->s_blocks_per_group !=
3304                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3305                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3306                                  "clusters per group (%lu) inconsistent",
3307                                  sbi->s_blocks_per_group,
3308                                  sbi->s_clusters_per_group);
3309                         goto failed_mount;
3310                 }
3311         } else {
3312                 if (clustersize != blocksize) {
3313                         ext4_warning(sb, "fragment/cluster size (%d) != "
3314                                      "block size (%d)", clustersize,
3315                                      blocksize);
3316                         clustersize = blocksize;
3317                 }
3318                 if (sbi->s_blocks_per_group > blocksize * 8) {
3319                         ext4_msg(sb, KERN_ERR,
3320                                  "#blocks per group too big: %lu",
3321                                  sbi->s_blocks_per_group);
3322                         goto failed_mount;
3323                 }
3324                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3325                 sbi->s_cluster_bits = 0;
3326         }
3327         sbi->s_cluster_ratio = clustersize / blocksize;
3328
3329         if (sbi->s_inodes_per_group > blocksize * 8) {
3330                 ext4_msg(sb, KERN_ERR,
3331                        "#inodes per group too big: %lu",
3332                        sbi->s_inodes_per_group);
3333                 goto failed_mount;
3334         }
3335
3336         /*
3337          * Test whether we have more sectors than will fit in sector_t,
3338          * and whether the max offset is addressable by the page cache.
3339          */
3340         err = generic_check_addressable(sb->s_blocksize_bits,
3341                                         ext4_blocks_count(es));
3342         if (err) {
3343                 ext4_msg(sb, KERN_ERR, "filesystem"
3344                          " too large to mount safely on this system");
3345                 if (sizeof(sector_t) < 8)
3346                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3347                 ret = err;
3348                 goto failed_mount;
3349         }
3350
3351         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3352                 goto cantfind_ext4;
3353
3354         /* check blocks count against device size */
3355         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3356         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3357                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3358                        "exceeds size of device (%llu blocks)",
3359                        ext4_blocks_count(es), blocks_count);
3360                 goto failed_mount;
3361         }
3362
3363         /*
3364          * It makes no sense for the first data block to be beyond the end
3365          * of the filesystem.
3366          */
3367         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3368                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3369                          "block %u is beyond end of filesystem (%llu)",
3370                          le32_to_cpu(es->s_first_data_block),
3371                          ext4_blocks_count(es));
3372                 goto failed_mount;
3373         }
3374         blocks_count = (ext4_blocks_count(es) -
3375                         le32_to_cpu(es->s_first_data_block) +
3376                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3377         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3378         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3379                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3380                        "(block count %llu, first data block %u, "
3381                        "blocks per group %lu)", sbi->s_groups_count,
3382                        ext4_blocks_count(es),
3383                        le32_to_cpu(es->s_first_data_block),
3384                        EXT4_BLOCKS_PER_GROUP(sb));
3385                 goto failed_mount;
3386         }
3387         sbi->s_groups_count = blocks_count;
3388         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3389                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3390         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3391                    EXT4_DESC_PER_BLOCK(sb);
3392         sbi->s_group_desc = ext4_kvmalloc(db_count *
3393                                           sizeof(struct buffer_head *),
3394                                           GFP_KERNEL);
3395         if (sbi->s_group_desc == NULL) {
3396                 ext4_msg(sb, KERN_ERR, "not enough memory");
3397                 goto failed_mount;
3398         }
3399
3400         if (ext4_proc_root)
3401                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3402
3403         if (sbi->s_proc)
3404                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3405                                  &ext4_seq_options_fops, sb);
3406
3407         bgl_lock_init(sbi->s_blockgroup_lock);
3408
3409         for (i = 0; i < db_count; i++) {
3410                 block = descriptor_loc(sb, logical_sb_block, i);
3411                 sbi->s_group_desc[i] = sb_bread(sb, block);
3412                 if (!sbi->s_group_desc[i]) {
3413                         ext4_msg(sb, KERN_ERR,
3414                                "can't read group descriptor %d", i);
3415                         db_count = i;
3416                         goto failed_mount2;
3417                 }
3418         }
3419         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3420                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3421                 goto failed_mount2;
3422         }
3423         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3424                 if (!ext4_fill_flex_info(sb)) {
3425                         ext4_msg(sb, KERN_ERR,
3426                                "unable to initialize "
3427                                "flex_bg meta info!");
3428                         goto failed_mount2;
3429                 }
3430
3431         sbi->s_gdb_count = db_count;
3432         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3433         spin_lock_init(&sbi->s_next_gen_lock);
3434
3435         init_timer(&sbi->s_err_report);
3436         sbi->s_err_report.function = print_daily_error_info;
3437         sbi->s_err_report.data = (unsigned long) sb;
3438
3439         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3440                         ext4_count_free_clusters(sb));
3441         if (!err) {
3442                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3443                                 ext4_count_free_inodes(sb));
3444         }
3445         if (!err) {
3446                 err = percpu_counter_init(&sbi->s_dirs_counter,
3447                                 ext4_count_dirs(sb));
3448         }
3449         if (!err) {
3450                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3451         }
3452         if (err) {
3453                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3454                 goto failed_mount3;
3455         }
3456
3457         sbi->s_stripe = ext4_get_stripe_size(sbi);
3458         sbi->s_max_writeback_mb_bump = 128;
3459
3460         /*
3461          * set up enough so that it can read an inode
3462          */
3463         if (!test_opt(sb, NOLOAD) &&
3464             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3465                 sb->s_op = &ext4_sops;
3466         else
3467                 sb->s_op = &ext4_nojournal_sops;
3468         sb->s_export_op = &ext4_export_ops;
3469         sb->s_xattr = ext4_xattr_handlers;
3470 #ifdef CONFIG_QUOTA
3471         sb->s_qcop = &ext4_qctl_operations;
3472         sb->dq_op = &ext4_quota_operations;
3473 #endif
3474         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3475
3476         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3477         mutex_init(&sbi->s_orphan_lock);
3478         sbi->s_resize_flags = 0;
3479
3480         sb->s_root = NULL;
3481
3482         needs_recovery = (es->s_last_orphan != 0 ||
3483                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3484                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3485
3486         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3487             !(sb->s_flags & MS_RDONLY))
3488                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3489                         goto failed_mount3;
3490
3491         /*
3492          * The first inode we look at is the journal inode.  Don't try
3493          * root first: it may be modified in the journal!
3494          */
3495         if (!test_opt(sb, NOLOAD) &&
3496             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3497                 if (ext4_load_journal(sb, es, journal_devnum))
3498                         goto failed_mount3;
3499         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3500               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3501                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3502                        "suppressed and not mounted read-only");
3503                 goto failed_mount_wq;
3504         } else {
3505                 clear_opt(sb, DATA_FLAGS);
3506                 sbi->s_journal = NULL;
3507                 needs_recovery = 0;
3508                 goto no_journal;
3509         }
3510
3511         if (ext4_blocks_count(es) > 0xffffffffULL &&
3512             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3513                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3514                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3515                 goto failed_mount_wq;
3516         }
3517
3518         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3519                 jbd2_journal_set_features(sbi->s_journal,
3520                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3521                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3522         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3523                 jbd2_journal_set_features(sbi->s_journal,
3524                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3525                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3526                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3527         } else {
3528                 jbd2_journal_clear_features(sbi->s_journal,
3529                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3530                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3531         }
3532
3533         /* We have now updated the journal if required, so we can
3534          * validate the data journaling mode. */
3535         switch (test_opt(sb, DATA_FLAGS)) {
3536         case 0:
3537                 /* No mode set, assume a default based on the journal
3538                  * capabilities: ORDERED_DATA if the journal can
3539                  * cope, else JOURNAL_DATA
3540                  */
3541                 if (jbd2_journal_check_available_features
3542                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3543                         set_opt(sb, ORDERED_DATA);
3544                 else
3545                         set_opt(sb, JOURNAL_DATA);
3546                 break;
3547
3548         case EXT4_MOUNT_ORDERED_DATA:
3549         case EXT4_MOUNT_WRITEBACK_DATA:
3550                 if (!jbd2_journal_check_available_features
3551                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3552                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3553                                "requested data journaling mode");
3554                         goto failed_mount_wq;
3555                 }
3556         default:
3557                 break;
3558         }
3559         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3560
3561         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3562
3563         /*
3564          * The journal may have updated the bg summary counts, so we
3565          * need to update the global counters.
3566          */
3567         percpu_counter_set(&sbi->s_freeclusters_counter,
3568                            ext4_count_free_clusters(sb));
3569         percpu_counter_set(&sbi->s_freeinodes_counter,
3570                            ext4_count_free_inodes(sb));
3571         percpu_counter_set(&sbi->s_dirs_counter,
3572                            ext4_count_dirs(sb));
3573         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3574
3575 no_journal:
3576         /*
3577          * The maximum number of concurrent works can be high and
3578          * concurrency isn't really necessary.  Limit it to 1.
3579          */
3580         EXT4_SB(sb)->dio_unwritten_wq =
3581                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3582         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3583                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3584                 goto failed_mount_wq;
3585         }
3586
3587         /*
3588          * The jbd2_journal_load will have done any necessary log recovery,
3589          * so we can safely mount the rest of the filesystem now.
3590          */
3591
3592         root = ext4_iget(sb, EXT4_ROOT_INO);
3593         if (IS_ERR(root)) {
3594                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3595                 ret = PTR_ERR(root);
3596                 root = NULL;
3597                 goto failed_mount4;
3598         }
3599         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3600                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3601                 iput(root);
3602                 goto failed_mount4;
3603         }
3604         sb->s_root = d_make_root(root);
3605         if (!sb->s_root) {
3606                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3607                 ret = -ENOMEM;
3608                 goto failed_mount4;
3609         }
3610
3611         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3612
3613         /* determine the minimum size of new large inodes, if present */
3614         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3615                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3616                                                      EXT4_GOOD_OLD_INODE_SIZE;
3617                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3618                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3619                         if (sbi->s_want_extra_isize <
3620                             le16_to_cpu(es->s_want_extra_isize))
3621                                 sbi->s_want_extra_isize =
3622                                         le16_to_cpu(es->s_want_extra_isize);
3623                         if (sbi->s_want_extra_isize <
3624                             le16_to_cpu(es->s_min_extra_isize))
3625                                 sbi->s_want_extra_isize =
3626                                         le16_to_cpu(es->s_min_extra_isize);
3627                 }
3628         }
3629         /* Check if enough inode space is available */
3630         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3631                                                         sbi->s_inode_size) {
3632                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3633                                                        EXT4_GOOD_OLD_INODE_SIZE;
3634                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3635                          "available");
3636         }
3637
3638         err = ext4_setup_system_zone(sb);
3639         if (err) {
3640                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3641                          "zone (%d)", err);
3642                 goto failed_mount4a;
3643         }
3644
3645         ext4_ext_init(sb);
3646         err = ext4_mb_init(sb, needs_recovery);
3647         if (err) {
3648                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3649                          err);
3650                 goto failed_mount5;
3651         }
3652
3653         err = ext4_register_li_request(sb, first_not_zeroed);
3654         if (err)
3655                 goto failed_mount6;
3656
3657         sbi->s_kobj.kset = ext4_kset;
3658         init_completion(&sbi->s_kobj_unregister);
3659         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3660                                    "%s", sb->s_id);
3661         if (err)
3662                 goto failed_mount7;
3663
3664         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3665         ext4_orphan_cleanup(sb, es);
3666         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3667         if (needs_recovery) {
3668                 ext4_msg(sb, KERN_INFO, "recovery complete");
3669                 ext4_mark_recovery_complete(sb, es);
3670         }
3671         if (EXT4_SB(sb)->s_journal) {
3672                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3673                         descr = " journalled data mode";
3674                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3675                         descr = " ordered data mode";
3676                 else
3677                         descr = " writeback data mode";
3678         } else
3679                 descr = "out journal";
3680
3681         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3682                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3683                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3684
3685         if (es->s_error_count)
3686                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3687
3688         kfree(orig_data);
3689         return 0;
3690
3691 cantfind_ext4:
3692         if (!silent)
3693                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3694         goto failed_mount;
3695
3696 failed_mount7:
3697         ext4_unregister_li_request(sb);
3698 failed_mount6:
3699         ext4_mb_release(sb);
3700 failed_mount5:
3701         ext4_ext_release(sb);
3702         ext4_release_system_zone(sb);
3703 failed_mount4a:
3704         dput(sb->s_root);
3705         sb->s_root = NULL;
3706 failed_mount4:
3707         ext4_msg(sb, KERN_ERR, "mount failed");
3708         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3709 failed_mount_wq:
3710         if (sbi->s_journal) {
3711                 jbd2_journal_destroy(sbi->s_journal);
3712                 sbi->s_journal = NULL;
3713         }
3714 failed_mount3:
3715         del_timer(&sbi->s_err_report);
3716         if (sbi->s_flex_groups)
3717                 ext4_kvfree(sbi->s_flex_groups);
3718         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3719         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3720         percpu_counter_destroy(&sbi->s_dirs_counter);
3721         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3722         if (sbi->s_mmp_tsk)
3723                 kthread_stop(sbi->s_mmp_tsk);
3724 failed_mount2:
3725         for (i = 0; i < db_count; i++)
3726                 brelse(sbi->s_group_desc[i]);
3727         ext4_kvfree(sbi->s_group_desc);
3728 failed_mount:
3729         if (sbi->s_proc) {
3730                 remove_proc_entry("options", sbi->s_proc);
3731                 remove_proc_entry(sb->s_id, ext4_proc_root);
3732         }
3733 #ifdef CONFIG_QUOTA
3734         for (i = 0; i < MAXQUOTAS; i++)
3735                 kfree(sbi->s_qf_names[i]);
3736 #endif
3737         ext4_blkdev_remove(sbi);
3738         brelse(bh);
3739 out_fail:
3740         sb->s_fs_info = NULL;
3741         kfree(sbi->s_blockgroup_lock);
3742         kfree(sbi);
3743 out_free_orig:
3744         kfree(orig_data);
3745         return ret;
3746 }
3747
3748 /*
3749  * Setup any per-fs journal parameters now.  We'll do this both on
3750  * initial mount, once the journal has been initialised but before we've
3751  * done any recovery; and again on any subsequent remount.
3752  */
3753 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3754 {
3755         struct ext4_sb_info *sbi = EXT4_SB(sb);
3756
3757         journal->j_commit_interval = sbi->s_commit_interval;
3758         journal->j_min_batch_time = sbi->s_min_batch_time;
3759         journal->j_max_batch_time = sbi->s_max_batch_time;
3760
3761         write_lock(&journal->j_state_lock);
3762         if (test_opt(sb, BARRIER))
3763                 journal->j_flags |= JBD2_BARRIER;
3764         else
3765                 journal->j_flags &= ~JBD2_BARRIER;
3766         if (test_opt(sb, DATA_ERR_ABORT))
3767                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3768         else
3769                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3770         write_unlock(&journal->j_state_lock);
3771 }
3772
3773 static journal_t *ext4_get_journal(struct super_block *sb,
3774                                    unsigned int journal_inum)
3775 {
3776         struct inode *journal_inode;
3777         journal_t *journal;
3778
3779         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3780
3781         /* First, test for the existence of a valid inode on disk.  Bad
3782          * things happen if we iget() an unused inode, as the subsequent
3783          * iput() will try to delete it. */
3784
3785         journal_inode = ext4_iget(sb, journal_inum);
3786         if (IS_ERR(journal_inode)) {
3787                 ext4_msg(sb, KERN_ERR, "no journal found");
3788                 return NULL;
3789         }
3790         if (!journal_inode->i_nlink) {
3791                 make_bad_inode(journal_inode);
3792                 iput(journal_inode);
3793                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3794                 return NULL;
3795         }
3796
3797         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3798                   journal_inode, journal_inode->i_size);
3799         if (!S_ISREG(journal_inode->i_mode)) {
3800                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3801                 iput(journal_inode);
3802                 return NULL;
3803         }
3804
3805         journal = jbd2_journal_init_inode(journal_inode);
3806         if (!journal) {
3807                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3808                 iput(journal_inode);
3809                 return NULL;
3810         }
3811         journal->j_private = sb;
3812         ext4_init_journal_params(sb, journal);
3813         return journal;
3814 }
3815
3816 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3817                                        dev_t j_dev)
3818 {
3819         struct buffer_head *bh;
3820         journal_t *journal;
3821         ext4_fsblk_t start;
3822         ext4_fsblk_t len;
3823         int hblock, blocksize;
3824         ext4_fsblk_t sb_block;
3825         unsigned long offset;
3826         struct ext4_super_block *es;
3827         struct block_device *bdev;
3828
3829         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3830
3831         bdev = ext4_blkdev_get(j_dev, sb);
3832         if (bdev == NULL)
3833                 return NULL;
3834
3835         blocksize = sb->s_blocksize;
3836         hblock = bdev_logical_block_size(bdev);
3837         if (blocksize < hblock) {
3838                 ext4_msg(sb, KERN_ERR,
3839                         "blocksize too small for journal device");
3840                 goto out_bdev;
3841         }
3842
3843         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3844         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3845         set_blocksize(bdev, blocksize);
3846         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3847                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3848                        "external journal");
3849                 goto out_bdev;
3850         }
3851
3852         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3853         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3854             !(le32_to_cpu(es->s_feature_incompat) &
3855               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3856                 ext4_msg(sb, KERN_ERR, "external journal has "
3857                                         "bad superblock");
3858                 brelse(bh);
3859                 goto out_bdev;
3860         }
3861
3862         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3863                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3864                 brelse(bh);
3865                 goto out_bdev;
3866         }
3867
3868         len = ext4_blocks_count(es);
3869         start = sb_block + 1;
3870         brelse(bh);     /* we're done with the superblock */
3871
3872         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3873                                         start, len, blocksize);
3874         if (!journal) {
3875                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3876                 goto out_bdev;
3877         }
3878         journal->j_private = sb;
3879         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3880         wait_on_buffer(journal->j_sb_buffer);
3881         if (!buffer_uptodate(journal->j_sb_buffer)) {
3882                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3883                 goto out_journal;
3884         }
3885         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3886                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3887                                         "user (unsupported) - %d",
3888                         be32_to_cpu(journal->j_superblock->s_nr_users));
3889                 goto out_journal;
3890         }
3891         EXT4_SB(sb)->journal_bdev = bdev;
3892         ext4_init_journal_params(sb, journal);
3893         return journal;
3894
3895 out_journal:
3896         jbd2_journal_destroy(journal);
3897 out_bdev:
3898         ext4_blkdev_put(bdev);
3899         return NULL;
3900 }
3901
3902 static int ext4_load_journal(struct super_block *sb,
3903                              struct ext4_super_block *es,
3904                              unsigned long journal_devnum)
3905 {
3906         journal_t *journal;
3907         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3908         dev_t journal_dev;
3909         int err = 0;
3910         int really_read_only;
3911
3912         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3913
3914         if (journal_devnum &&
3915             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3916                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3917                         "numbers have changed");
3918                 journal_dev = new_decode_dev(journal_devnum);
3919         } else
3920                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3921
3922         really_read_only = bdev_read_only(sb->s_bdev);
3923
3924         /*
3925          * Are we loading a blank journal or performing recovery after a
3926          * crash?  For recovery, we need to check in advance whether we
3927          * can get read-write access to the device.
3928          */
3929         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3930                 if (sb->s_flags & MS_RDONLY) {
3931                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3932                                         "required on readonly filesystem");
3933                         if (really_read_only) {
3934                                 ext4_msg(sb, KERN_ERR, "write access "
3935                                         "unavailable, cannot proceed");
3936                                 return -EROFS;
3937                         }
3938                         ext4_msg(sb, KERN_INFO, "write access will "
3939                                "be enabled during recovery");
3940                 }
3941         }
3942
3943         if (journal_inum && journal_dev) {
3944                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3945                        "and inode journals!");
3946                 return -EINVAL;
3947         }
3948
3949         if (journal_inum) {
3950                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3951                         return -EINVAL;
3952         } else {
3953                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3954                         return -EINVAL;
3955         }
3956
3957         if (!(journal->j_flags & JBD2_BARRIER))
3958                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3959
3960         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3961                 err = jbd2_journal_wipe(journal, !really_read_only);
3962         if (!err) {
3963                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3964                 if (save)
3965                         memcpy(save, ((char *) es) +
3966                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3967                 err = jbd2_journal_load(journal);
3968                 if (save)
3969                         memcpy(((char *) es) + EXT4_S_ERR_START,
3970                                save, EXT4_S_ERR_LEN);
3971                 kfree(save);
3972         }
3973
3974         if (err) {
3975                 ext4_msg(sb, KERN_ERR, "error loading journal");
3976                 jbd2_journal_destroy(journal);
3977                 return err;
3978         }
3979
3980         EXT4_SB(sb)->s_journal = journal;
3981         ext4_clear_journal_err(sb, es);
3982
3983         if (!really_read_only && journal_devnum &&
3984             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3985                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3986
3987                 /* Make sure we flush the recovery flag to disk. */
3988                 ext4_commit_super(sb, 1);
3989         }
3990
3991         return 0;
3992 }
3993
3994 static int ext4_commit_super(struct super_block *sb, int sync)
3995 {
3996         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3997         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3998         int error = 0;
3999
4000         if (!sbh || block_device_ejected(sb))
4001                 return error;
4002         if (buffer_write_io_error(sbh)) {
4003                 /*
4004                  * Oh, dear.  A previous attempt to write the
4005                  * superblock failed.  This could happen because the
4006                  * USB device was yanked out.  Or it could happen to
4007                  * be a transient write error and maybe the block will
4008                  * be remapped.  Nothing we can do but to retry the
4009                  * write and hope for the best.
4010                  */
4011                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4012                        "superblock detected");
4013                 clear_buffer_write_io_error(sbh);
4014                 set_buffer_uptodate(sbh);
4015         }
4016         /*
4017          * If the file system is mounted read-only, don't update the
4018          * superblock write time.  This avoids updating the superblock
4019          * write time when we are mounting the root file system
4020          * read/only but we need to replay the journal; at that point,
4021          * for people who are east of GMT and who make their clock
4022          * tick in localtime for Windows bug-for-bug compatibility,
4023          * the clock is set in the future, and this will cause e2fsck
4024          * to complain and force a full file system check.
4025          */
4026         if (!(sb->s_flags & MS_RDONLY))
4027                 es->s_wtime = cpu_to_le32(get_seconds());
4028         if (sb->s_bdev->bd_part)
4029                 es->s_kbytes_written =
4030                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4031                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4032                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4033         else
4034                 es->s_kbytes_written =
4035                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4036         ext4_free_blocks_count_set(es,
4037                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4038                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4039         es->s_free_inodes_count =
4040                 cpu_to_le32(percpu_counter_sum_positive(
4041                                 &EXT4_SB(sb)->s_freeinodes_counter));
4042         sb->s_dirt = 0;
4043         BUFFER_TRACE(sbh, "marking dirty");
4044         mark_buffer_dirty(sbh);
4045         if (sync) {
4046                 error = sync_dirty_buffer(sbh);
4047                 if (error)
4048                         return error;
4049
4050                 error = buffer_write_io_error(sbh);
4051                 if (error) {
4052                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4053                                "superblock");
4054                         clear_buffer_write_io_error(sbh);
4055                         set_buffer_uptodate(sbh);
4056                 }
4057         }
4058         return error;
4059 }
4060
4061 /*
4062  * Have we just finished recovery?  If so, and if we are mounting (or
4063  * remounting) the filesystem readonly, then we will end up with a
4064  * consistent fs on disk.  Record that fact.
4065  */
4066 static void ext4_mark_recovery_complete(struct super_block *sb,
4067                                         struct ext4_super_block *es)
4068 {
4069         journal_t *journal = EXT4_SB(sb)->s_journal;
4070
4071         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4072                 BUG_ON(journal != NULL);
4073                 return;
4074         }
4075         jbd2_journal_lock_updates(journal);
4076         if (jbd2_journal_flush(journal) < 0)
4077                 goto out;
4078
4079         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4080             sb->s_flags & MS_RDONLY) {
4081                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4082                 ext4_commit_super(sb, 1);
4083         }
4084
4085 out:
4086         jbd2_journal_unlock_updates(journal);
4087 }
4088
4089 /*
4090  * If we are mounting (or read-write remounting) a filesystem whose journal
4091  * has recorded an error from a previous lifetime, move that error to the
4092  * main filesystem now.
4093  */
4094 static void ext4_clear_journal_err(struct super_block *sb,
4095                                    struct ext4_super_block *es)
4096 {
4097         journal_t *journal;
4098         int j_errno;
4099         const char *errstr;
4100
4101         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4102
4103         journal = EXT4_SB(sb)->s_journal;
4104
4105         /*
4106          * Now check for any error status which may have been recorded in the
4107          * journal by a prior ext4_error() or ext4_abort()
4108          */
4109
4110         j_errno = jbd2_journal_errno(journal);
4111         if (j_errno) {
4112                 char nbuf[16];
4113
4114                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4115                 ext4_warning(sb, "Filesystem error recorded "
4116                              "from previous mount: %s", errstr);
4117                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4118
4119                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4120                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4121                 ext4_commit_super(sb, 1);
4122
4123                 jbd2_journal_clear_err(journal);
4124         }
4125 }
4126
4127 /*
4128  * Force the running and committing transactions to commit,
4129  * and wait on the commit.
4130  */
4131 int ext4_force_commit(struct super_block *sb)
4132 {
4133         journal_t *journal;
4134         int ret = 0;
4135
4136         if (sb->s_flags & MS_RDONLY)
4137                 return 0;
4138
4139         journal = EXT4_SB(sb)->s_journal;
4140         if (journal) {
4141                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4142                 ret = ext4_journal_force_commit(journal);
4143         }
4144
4145         return ret;
4146 }
4147
4148 static void ext4_write_super(struct super_block *sb)
4149 {
4150         lock_super(sb);
4151         ext4_commit_super(sb, 1);
4152         unlock_super(sb);
4153 }
4154
4155 static int ext4_sync_fs(struct super_block *sb, int wait)
4156 {
4157         int ret = 0;
4158         tid_t target;
4159         struct ext4_sb_info *sbi = EXT4_SB(sb);
4160
4161         trace_ext4_sync_fs(sb, wait);
4162         flush_workqueue(sbi->dio_unwritten_wq);
4163         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4164                 if (wait)
4165                         jbd2_log_wait_commit(sbi->s_journal, target);
4166         }
4167         return ret;
4168 }
4169
4170 /*
4171  * LVM calls this function before a (read-only) snapshot is created.  This
4172  * gives us a chance to flush the journal completely and mark the fs clean.
4173  *
4174  * Note that only this function cannot bring a filesystem to be in a clean
4175  * state independently, because ext4 prevents a new handle from being started
4176  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4177  * the upper layer.
4178  */
4179 static int ext4_freeze(struct super_block *sb)
4180 {
4181         int error = 0;
4182         journal_t *journal;
4183
4184         if (sb->s_flags & MS_RDONLY)
4185                 return 0;
4186
4187         journal = EXT4_SB(sb)->s_journal;
4188
4189         /* Now we set up the journal barrier. */
4190         jbd2_journal_lock_updates(journal);
4191
4192         /*
4193          * Don't clear the needs_recovery flag if we failed to flush
4194          * the journal.
4195          */
4196         error = jbd2_journal_flush(journal);
4197         if (error < 0)
4198                 goto out;
4199
4200         /* Journal blocked and flushed, clear needs_recovery flag. */
4201         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4202         error = ext4_commit_super(sb, 1);
4203 out:
4204         /* we rely on s_frozen to stop further updates */
4205         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4206         return error;
4207 }
4208
4209 /*
4210  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4211  * flag here, even though the filesystem is not technically dirty yet.
4212  */
4213 static int ext4_unfreeze(struct super_block *sb)
4214 {
4215         if (sb->s_flags & MS_RDONLY)
4216                 return 0;
4217
4218         lock_super(sb);
4219         /* Reset the needs_recovery flag before the fs is unlocked. */
4220         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4221         ext4_commit_super(sb, 1);
4222         unlock_super(sb);
4223         return 0;
4224 }
4225
4226 /*
4227  * Structure to save mount options for ext4_remount's benefit
4228  */
4229 struct ext4_mount_options {
4230         unsigned long s_mount_opt;
4231         unsigned long s_mount_opt2;
4232         uid_t s_resuid;
4233         gid_t s_resgid;
4234         unsigned long s_commit_interval;
4235         u32 s_min_batch_time, s_max_batch_time;
4236 #ifdef CONFIG_QUOTA
4237         int s_jquota_fmt;
4238         char *s_qf_names[MAXQUOTAS];
4239 #endif
4240 };
4241
4242 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4243 {
4244         struct ext4_super_block *es;
4245         struct ext4_sb_info *sbi = EXT4_SB(sb);
4246         unsigned long old_sb_flags;
4247         struct ext4_mount_options old_opts;
4248         int enable_quota = 0;
4249         ext4_group_t g;
4250         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4251         int err = 0;
4252 #ifdef CONFIG_QUOTA
4253         int i;
4254 #endif
4255         char *orig_data = kstrdup(data, GFP_KERNEL);
4256
4257         /* Store the original options */
4258         lock_super(sb);
4259         old_sb_flags = sb->s_flags;
4260         old_opts.s_mount_opt = sbi->s_mount_opt;
4261         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4262         old_opts.s_resuid = sbi->s_resuid;
4263         old_opts.s_resgid = sbi->s_resgid;
4264         old_opts.s_commit_interval = sbi->s_commit_interval;
4265         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4266         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4267 #ifdef CONFIG_QUOTA
4268         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4269         for (i = 0; i < MAXQUOTAS; i++)
4270                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4271 #endif
4272         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4273                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4274
4275         /*
4276          * Allow the "check" option to be passed as a remount option.
4277          */
4278         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4279                 err = -EINVAL;
4280                 goto restore_opts;
4281         }
4282
4283         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4284                 ext4_abort(sb, "Abort forced by user");
4285
4286         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4287                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4288
4289         es = sbi->s_es;
4290
4291         if (sbi->s_journal) {
4292                 ext4_init_journal_params(sb, sbi->s_journal);
4293                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4294         }
4295
4296         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4297                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4298                         err = -EROFS;
4299                         goto restore_opts;
4300                 }
4301
4302                 if (*flags & MS_RDONLY) {
4303                         err = dquot_suspend(sb, -1);
4304                         if (err < 0)
4305                                 goto restore_opts;
4306
4307                         /*
4308                          * First of all, the unconditional stuff we have to do
4309                          * to disable replay of the journal when we next remount
4310                          */
4311                         sb->s_flags |= MS_RDONLY;
4312
4313                         /*
4314                          * OK, test if we are remounting a valid rw partition
4315                          * readonly, and if so set the rdonly flag and then
4316                          * mark the partition as valid again.
4317                          */
4318                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4319                             (sbi->s_mount_state & EXT4_VALID_FS))
4320                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4321
4322                         if (sbi->s_journal)
4323                                 ext4_mark_recovery_complete(sb, es);
4324                 } else {
4325                         /* Make sure we can mount this feature set readwrite */
4326                         if (!ext4_feature_set_ok(sb, 0)) {
4327                                 err = -EROFS;
4328                                 goto restore_opts;
4329                         }
4330                         /*
4331                          * Make sure the group descriptor checksums
4332                          * are sane.  If they aren't, refuse to remount r/w.
4333                          */
4334                         for (g = 0; g < sbi->s_groups_count; g++) {
4335                                 struct ext4_group_desc *gdp =
4336                                         ext4_get_group_desc(sb, g, NULL);
4337
4338                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4339                                         ext4_msg(sb, KERN_ERR,
4340                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4341                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4342                                                le16_to_cpu(gdp->bg_checksum));
4343                                         err = -EINVAL;
4344                                         goto restore_opts;
4345                                 }
4346                         }
4347
4348                         /*
4349                          * If we have an unprocessed orphan list hanging
4350                          * around from a previously readonly bdev mount,
4351                          * require a full umount/remount for now.
4352                          */
4353                         if (es->s_last_orphan) {
4354                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4355                                        "remount RDWR because of unprocessed "
4356                                        "orphan inode list.  Please "
4357                                        "umount/remount instead");
4358                                 err = -EINVAL;
4359                                 goto restore_opts;
4360                         }
4361
4362                         /*
4363                          * Mounting a RDONLY partition read-write, so reread
4364                          * and store the current valid flag.  (It may have
4365                          * been changed by e2fsck since we originally mounted
4366                          * the partition.)
4367                          */
4368                         if (sbi->s_journal)
4369                                 ext4_clear_journal_err(sb, es);
4370                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4371                         if (!ext4_setup_super(sb, es, 0))
4372                                 sb->s_flags &= ~MS_RDONLY;
4373                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4374                                                      EXT4_FEATURE_INCOMPAT_MMP))
4375                                 if (ext4_multi_mount_protect(sb,
4376                                                 le64_to_cpu(es->s_mmp_block))) {
4377                                         err = -EROFS;
4378                                         goto restore_opts;
4379                                 }
4380                         enable_quota = 1;
4381                 }
4382         }
4383
4384         /*
4385          * Reinitialize lazy itable initialization thread based on
4386          * current settings
4387          */
4388         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4389                 ext4_unregister_li_request(sb);
4390         else {
4391                 ext4_group_t first_not_zeroed;
4392                 first_not_zeroed = ext4_has_uninit_itable(sb);
4393                 ext4_register_li_request(sb, first_not_zeroed);
4394         }
4395
4396         ext4_setup_system_zone(sb);
4397         if (sbi->s_journal == NULL)
4398                 ext4_commit_super(sb, 1);
4399
4400 #ifdef CONFIG_QUOTA
4401         /* Release old quota file names */
4402         for (i = 0; i < MAXQUOTAS; i++)
4403                 if (old_opts.s_qf_names[i] &&
4404                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4405                         kfree(old_opts.s_qf_names[i]);
4406 #endif
4407         unlock_super(sb);
4408         if (enable_quota)
4409                 dquot_resume(sb, -1);
4410
4411         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4412         kfree(orig_data);
4413         return 0;
4414
4415 restore_opts:
4416         sb->s_flags = old_sb_flags;
4417         sbi->s_mount_opt = old_opts.s_mount_opt;
4418         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4419         sbi->s_resuid = old_opts.s_resuid;
4420         sbi->s_resgid = old_opts.s_resgid;
4421         sbi->s_commit_interval = old_opts.s_commit_interval;
4422         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4423         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4424 #ifdef CONFIG_QUOTA
4425         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4426         for (i = 0; i < MAXQUOTAS; i++) {
4427                 if (sbi->s_qf_names[i] &&
4428                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4429                         kfree(sbi->s_qf_names[i]);
4430                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4431         }
4432 #endif
4433         unlock_super(sb);
4434         kfree(orig_data);
4435         return err;
4436 }
4437
4438 /*
4439  * Note: calculating the overhead so we can be compatible with
4440  * historical BSD practice is quite difficult in the face of
4441  * clusters/bigalloc.  This is because multiple metadata blocks from
4442  * different block group can end up in the same allocation cluster.
4443  * Calculating the exact overhead in the face of clustered allocation
4444  * requires either O(all block bitmaps) in memory or O(number of block
4445  * groups**2) in time.  We will still calculate the superblock for
4446  * older file systems --- and if we come across with a bigalloc file
4447  * system with zero in s_overhead_clusters the estimate will be close to
4448  * correct especially for very large cluster sizes --- but for newer
4449  * file systems, it's better to calculate this figure once at mkfs
4450  * time, and store it in the superblock.  If the superblock value is
4451  * present (even for non-bigalloc file systems), we will use it.
4452  */
4453 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4454 {
4455         struct super_block *sb = dentry->d_sb;
4456         struct ext4_sb_info *sbi = EXT4_SB(sb);
4457         struct ext4_super_block *es = sbi->s_es;
4458         struct ext4_group_desc *gdp;
4459         u64 fsid;
4460         s64 bfree;
4461
4462         if (test_opt(sb, MINIX_DF)) {
4463                 sbi->s_overhead_last = 0;
4464         } else if (es->s_overhead_clusters) {
4465                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4466         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4467                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4468                 ext4_fsblk_t overhead = 0;
4469
4470                 /*
4471                  * Compute the overhead (FS structures).  This is constant
4472                  * for a given filesystem unless the number of block groups
4473                  * changes so we cache the previous value until it does.
4474                  */
4475
4476                 /*
4477                  * All of the blocks before first_data_block are
4478                  * overhead
4479                  */
4480                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4481
4482                 /*
4483                  * Add the overhead found in each block group
4484                  */
4485                 for (i = 0; i < ngroups; i++) {
4486                         gdp = ext4_get_group_desc(sb, i, NULL);
4487                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4488                         cond_resched();
4489                 }
4490                 sbi->s_overhead_last = overhead;
4491                 smp_wmb();
4492                 sbi->s_blocks_last = ext4_blocks_count(es);
4493         }
4494
4495         buf->f_type = EXT4_SUPER_MAGIC;
4496         buf->f_bsize = sb->s_blocksize;
4497         buf->f_blocks = (ext4_blocks_count(es) -
4498                          EXT4_C2B(sbi, sbi->s_overhead_last));
4499         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4500                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4501         /* prevent underflow in case that few free space is available */
4502         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4503         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4504         if (buf->f_bfree < ext4_r_blocks_count(es))
4505                 buf->f_bavail = 0;
4506         buf->f_files = le32_to_cpu(es->s_inodes_count);
4507         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4508         buf->f_namelen = EXT4_NAME_LEN;
4509         fsid = le64_to_cpup((void *)es->s_uuid) ^
4510                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4511         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4512         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4513
4514         return 0;
4515 }
4516
4517 /* Helper function for writing quotas on sync - we need to start transaction
4518  * before quota file is locked for write. Otherwise the are possible deadlocks:
4519  * Process 1                         Process 2
4520  * ext4_create()                     quota_sync()
4521  *   jbd2_journal_start()                  write_dquot()
4522  *   dquot_initialize()                         down(dqio_mutex)
4523  *     down(dqio_mutex)                    jbd2_journal_start()
4524  *
4525  */
4526
4527 #ifdef CONFIG_QUOTA
4528
4529 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4530 {
4531         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4532 }
4533
4534 static int ext4_write_dquot(struct dquot *dquot)
4535 {
4536         int ret, err;
4537         handle_t *handle;
4538         struct inode *inode;
4539
4540         inode = dquot_to_inode(dquot);
4541         handle = ext4_journal_start(inode,
4542                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4543         if (IS_ERR(handle))
4544                 return PTR_ERR(handle);
4545         ret = dquot_commit(dquot);
4546         err = ext4_journal_stop(handle);
4547         if (!ret)
4548                 ret = err;
4549         return ret;
4550 }
4551
4552 static int ext4_acquire_dquot(struct dquot *dquot)
4553 {
4554         int ret, err;
4555         handle_t *handle;
4556
4557         handle = ext4_journal_start(dquot_to_inode(dquot),
4558                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4559         if (IS_ERR(handle))
4560                 return PTR_ERR(handle);
4561         ret = dquot_acquire(dquot);
4562         err = ext4_journal_stop(handle);
4563         if (!ret)
4564                 ret = err;
4565         return ret;
4566 }
4567
4568 static int ext4_release_dquot(struct dquot *dquot)
4569 {
4570         int ret, err;
4571         handle_t *handle;
4572
4573         handle = ext4_journal_start(dquot_to_inode(dquot),
4574                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4575         if (IS_ERR(handle)) {
4576                 /* Release dquot anyway to avoid endless cycle in dqput() */
4577                 dquot_release(dquot);
4578                 return PTR_ERR(handle);
4579         }
4580         ret = dquot_release(dquot);
4581         err = ext4_journal_stop(handle);
4582         if (!ret)
4583                 ret = err;
4584         return ret;
4585 }
4586
4587 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4588 {
4589         /* Are we journaling quotas? */
4590         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4591             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4592                 dquot_mark_dquot_dirty(dquot);
4593                 return ext4_write_dquot(dquot);
4594         } else {
4595                 return dquot_mark_dquot_dirty(dquot);
4596         }
4597 }
4598
4599 static int ext4_write_info(struct super_block *sb, int type)
4600 {
4601         int ret, err;
4602         handle_t *handle;
4603
4604         /* Data block + inode block */
4605         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4606         if (IS_ERR(handle))
4607                 return PTR_ERR(handle);
4608         ret = dquot_commit_info(sb, type);
4609         err = ext4_journal_stop(handle);
4610         if (!ret)
4611                 ret = err;
4612         return ret;
4613 }
4614
4615 /*
4616  * Turn on quotas during mount time - we need to find
4617  * the quota file and such...
4618  */
4619 static int ext4_quota_on_mount(struct super_block *sb, int type)
4620 {
4621         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4622                                         EXT4_SB(sb)->s_jquota_fmt, type);
4623 }
4624
4625 /*
4626  * Standard function to be called on quota_on
4627  */
4628 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4629                          struct path *path)
4630 {
4631         int err;
4632
4633         if (!test_opt(sb, QUOTA))
4634                 return -EINVAL;
4635
4636         /* Quotafile not on the same filesystem? */
4637         if (path->dentry->d_sb != sb)
4638                 return -EXDEV;
4639         /* Journaling quota? */
4640         if (EXT4_SB(sb)->s_qf_names[type]) {
4641                 /* Quotafile not in fs root? */
4642                 if (path->dentry->d_parent != sb->s_root)
4643                         ext4_msg(sb, KERN_WARNING,
4644                                 "Quota file not on filesystem root. "
4645                                 "Journaled quota will not work");
4646         }
4647
4648         /*
4649          * When we journal data on quota file, we have to flush journal to see
4650          * all updates to the file when we bypass pagecache...
4651          */
4652         if (EXT4_SB(sb)->s_journal &&
4653             ext4_should_journal_data(path->dentry->d_inode)) {
4654                 /*
4655                  * We don't need to lock updates but journal_flush() could
4656                  * otherwise be livelocked...
4657                  */
4658                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4659                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4660                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4661                 if (err)
4662                         return err;
4663         }
4664
4665         return dquot_quota_on(sb, type, format_id, path);
4666 }
4667
4668 static int ext4_quota_off(struct super_block *sb, int type)
4669 {
4670         struct inode *inode = sb_dqopt(sb)->files[type];
4671         handle_t *handle;
4672
4673         /* Force all delayed allocation blocks to be allocated.
4674          * Caller already holds s_umount sem */
4675         if (test_opt(sb, DELALLOC))
4676                 sync_filesystem(sb);
4677
4678         if (!inode)
4679                 goto out;
4680
4681         /* Update modification times of quota files when userspace can
4682          * start looking at them */
4683         handle = ext4_journal_start(inode, 1);
4684         if (IS_ERR(handle))
4685                 goto out;
4686         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4687         ext4_mark_inode_dirty(handle, inode);
4688         ext4_journal_stop(handle);
4689
4690 out:
4691         return dquot_quota_off(sb, type);
4692 }
4693
4694 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4695  * acquiring the locks... As quota files are never truncated and quota code
4696  * itself serializes the operations (and no one else should touch the files)
4697  * we don't have to be afraid of races */
4698 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4699                                size_t len, loff_t off)
4700 {
4701         struct inode *inode = sb_dqopt(sb)->files[type];
4702         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4703         int err = 0;
4704         int offset = off & (sb->s_blocksize - 1);
4705         int tocopy;
4706         size_t toread;
4707         struct buffer_head *bh;
4708         loff_t i_size = i_size_read(inode);
4709
4710         if (off > i_size)
4711                 return 0;
4712         if (off+len > i_size)
4713                 len = i_size-off;
4714         toread = len;
4715         while (toread > 0) {
4716                 tocopy = sb->s_blocksize - offset < toread ?
4717                                 sb->s_blocksize - offset : toread;
4718                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4719                 if (err)
4720                         return err;
4721                 if (!bh)        /* A hole? */
4722                         memset(data, 0, tocopy);
4723                 else
4724                         memcpy(data, bh->b_data+offset, tocopy);
4725                 brelse(bh);
4726                 offset = 0;
4727                 toread -= tocopy;
4728                 data += tocopy;
4729                 blk++;
4730         }
4731         return len;
4732 }
4733
4734 /* Write to quotafile (we know the transaction is already started and has
4735  * enough credits) */
4736 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4737                                 const char *data, size_t len, loff_t off)
4738 {
4739         struct inode *inode = sb_dqopt(sb)->files[type];
4740         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4741         int err = 0;
4742         int offset = off & (sb->s_blocksize - 1);
4743         struct buffer_head *bh;
4744         handle_t *handle = journal_current_handle();
4745
4746         if (EXT4_SB(sb)->s_journal && !handle) {
4747                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4748                         " cancelled because transaction is not started",
4749                         (unsigned long long)off, (unsigned long long)len);
4750                 return -EIO;
4751         }
4752         /*
4753          * Since we account only one data block in transaction credits,
4754          * then it is impossible to cross a block boundary.
4755          */
4756         if (sb->s_blocksize - offset < len) {
4757                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4758                         " cancelled because not block aligned",
4759                         (unsigned long long)off, (unsigned long long)len);
4760                 return -EIO;
4761         }
4762
4763         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4764         bh = ext4_bread(handle, inode, blk, 1, &err);
4765         if (!bh)
4766                 goto out;
4767         err = ext4_journal_get_write_access(handle, bh);
4768         if (err) {
4769                 brelse(bh);
4770                 goto out;
4771         }
4772         lock_buffer(bh);
4773         memcpy(bh->b_data+offset, data, len);
4774         flush_dcache_page(bh->b_page);
4775         unlock_buffer(bh);
4776         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4777         brelse(bh);
4778 out:
4779         if (err) {
4780                 mutex_unlock(&inode->i_mutex);
4781                 return err;
4782         }
4783         if (inode->i_size < off + len) {
4784                 i_size_write(inode, off + len);
4785                 EXT4_I(inode)->i_disksize = inode->i_size;
4786                 ext4_mark_inode_dirty(handle, inode);
4787         }
4788         mutex_unlock(&inode->i_mutex);
4789         return len;
4790 }
4791
4792 #endif
4793
4794 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4795                        const char *dev_name, void *data)
4796 {
4797         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4798 }
4799
4800 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4801 static inline void register_as_ext2(void)
4802 {
4803         int err = register_filesystem(&ext2_fs_type);
4804         if (err)
4805                 printk(KERN_WARNING
4806                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4807 }
4808
4809 static inline void unregister_as_ext2(void)
4810 {
4811         unregister_filesystem(&ext2_fs_type);
4812 }
4813
4814 static inline int ext2_feature_set_ok(struct super_block *sb)
4815 {
4816         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4817                 return 0;
4818         if (sb->s_flags & MS_RDONLY)
4819                 return 1;
4820         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4821                 return 0;
4822         return 1;
4823 }
4824 MODULE_ALIAS("ext2");
4825 #else
4826 static inline void register_as_ext2(void) { }
4827 static inline void unregister_as_ext2(void) { }
4828 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4829 #endif
4830
4831 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4832 static inline void register_as_ext3(void)
4833 {
4834         int err = register_filesystem(&ext3_fs_type);
4835         if (err)
4836                 printk(KERN_WARNING
4837                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4838 }
4839
4840 static inline void unregister_as_ext3(void)
4841 {
4842         unregister_filesystem(&ext3_fs_type);
4843 }
4844
4845 static inline int ext3_feature_set_ok(struct super_block *sb)
4846 {
4847         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4848                 return 0;
4849         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4850                 return 0;
4851         if (sb->s_flags & MS_RDONLY)
4852                 return 1;
4853         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4854                 return 0;
4855         return 1;
4856 }
4857 MODULE_ALIAS("ext3");
4858 #else
4859 static inline void register_as_ext3(void) { }
4860 static inline void unregister_as_ext3(void) { }
4861 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4862 #endif
4863
4864 static struct file_system_type ext4_fs_type = {
4865         .owner          = THIS_MODULE,
4866         .name           = "ext4",
4867         .mount          = ext4_mount,
4868         .kill_sb        = kill_block_super,
4869         .fs_flags       = FS_REQUIRES_DEV,
4870 };
4871
4872 static int __init ext4_init_feat_adverts(void)
4873 {
4874         struct ext4_features *ef;
4875         int ret = -ENOMEM;
4876
4877         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4878         if (!ef)
4879                 goto out;
4880
4881         ef->f_kobj.kset = ext4_kset;
4882         init_completion(&ef->f_kobj_unregister);
4883         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4884                                    "features");
4885         if (ret) {
4886                 kfree(ef);
4887                 goto out;
4888         }
4889
4890         ext4_feat = ef;
4891         ret = 0;
4892 out:
4893         return ret;
4894 }
4895
4896 static void ext4_exit_feat_adverts(void)
4897 {
4898         kobject_put(&ext4_feat->f_kobj);
4899         wait_for_completion(&ext4_feat->f_kobj_unregister);
4900         kfree(ext4_feat);
4901 }
4902
4903 /* Shared across all ext4 file systems */
4904 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4905 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4906
4907 static int __init ext4_init_fs(void)
4908 {
4909         int i, err;
4910
4911         ext4_li_info = NULL;
4912         mutex_init(&ext4_li_mtx);
4913
4914         ext4_check_flag_values();
4915
4916         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4917                 mutex_init(&ext4__aio_mutex[i]);
4918                 init_waitqueue_head(&ext4__ioend_wq[i]);
4919         }
4920
4921         err = ext4_init_pageio();
4922         if (err)
4923                 return err;
4924         err = ext4_init_system_zone();
4925         if (err)
4926                 goto out6;
4927         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4928         if (!ext4_kset)
4929                 goto out5;
4930         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4931
4932         err = ext4_init_feat_adverts();
4933         if (err)
4934                 goto out4;
4935
4936         err = ext4_init_mballoc();
4937         if (err)
4938                 goto out3;
4939
4940         err = ext4_init_xattr();
4941         if (err)
4942                 goto out2;
4943         err = init_inodecache();
4944         if (err)
4945                 goto out1;
4946         register_as_ext3();
4947         register_as_ext2();
4948         err = register_filesystem(&ext4_fs_type);
4949         if (err)
4950                 goto out;
4951
4952         return 0;
4953 out:
4954         unregister_as_ext2();
4955         unregister_as_ext3();
4956         destroy_inodecache();
4957 out1:
4958         ext4_exit_xattr();
4959 out2:
4960         ext4_exit_mballoc();
4961 out3:
4962         ext4_exit_feat_adverts();
4963 out4:
4964         if (ext4_proc_root)
4965                 remove_proc_entry("fs/ext4", NULL);
4966         kset_unregister(ext4_kset);
4967 out5:
4968         ext4_exit_system_zone();
4969 out6:
4970         ext4_exit_pageio();
4971         return err;
4972 }
4973
4974 static void __exit ext4_exit_fs(void)
4975 {
4976         ext4_destroy_lazyinit_thread();
4977         unregister_as_ext2();
4978         unregister_as_ext3();
4979         unregister_filesystem(&ext4_fs_type);
4980         destroy_inodecache();
4981         ext4_exit_xattr();
4982         ext4_exit_mballoc();
4983         ext4_exit_feat_adverts();
4984         remove_proc_entry("fs/ext4", NULL);
4985         kset_unregister(ext4_kset);
4986         ext4_exit_system_zone();
4987         ext4_exit_pageio();
4988 }
4989
4990 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4991 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4992 MODULE_LICENSE("GPL");
4993 module_init(ext4_init_fs)
4994 module_exit(ext4_exit_fs)