ext4: mark superblock writes synchronous for nobarrier mounts
[sfrench/cifs-2.6.git] / fs / ext4 / super.c
1 /*
2  *  linux/fs/ext4/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/module.h>
20 #include <linux/string.h>
21 #include <linux/fs.h>
22 #include <linux/time.h>
23 #include <linux/vmalloc.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.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/ctype.h>
38 #include <linux/log2.h>
39 #include <linux/crc16.h>
40 #include <linux/cleancache.h>
41 #include <linux/uaccess.h>
42
43 #include <linux/kthread.h>
44 #include <linux/freezer.h>
45
46 #include "ext4.h"
47 #include "ext4_extents.h"       /* Needed for trace points definition */
48 #include "ext4_jbd2.h"
49 #include "xattr.h"
50 #include "acl.h"
51 #include "mballoc.h"
52 #include "fsmap.h"
53
54 #define CREATE_TRACE_POINTS
55 #include <trace/events/ext4.h>
56
57 static struct ext4_lazy_init *ext4_li_info;
58 static struct mutex ext4_li_mtx;
59 static struct ratelimit_state ext4_mount_msg_ratelimit;
60
61 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
62                              unsigned long journal_devnum);
63 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
64 static int ext4_commit_super(struct super_block *sb, int sync);
65 static void ext4_mark_recovery_complete(struct super_block *sb,
66                                         struct ext4_super_block *es);
67 static void ext4_clear_journal_err(struct super_block *sb,
68                                    struct ext4_super_block *es);
69 static int ext4_sync_fs(struct super_block *sb, int wait);
70 static int ext4_remount(struct super_block *sb, int *flags, char *data);
71 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
72 static int ext4_unfreeze(struct super_block *sb);
73 static int ext4_freeze(struct super_block *sb);
74 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
75                        const char *dev_name, void *data);
76 static inline int ext2_feature_set_ok(struct super_block *sb);
77 static inline int ext3_feature_set_ok(struct super_block *sb);
78 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
79 static void ext4_destroy_lazyinit_thread(void);
80 static void ext4_unregister_li_request(struct super_block *sb);
81 static void ext4_clear_request_list(void);
82 static struct inode *ext4_get_journal_inode(struct super_block *sb,
83                                             unsigned int journal_inum);
84
85 /*
86  * Lock ordering
87  *
88  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
89  * i_mmap_rwsem (inode->i_mmap_rwsem)!
90  *
91  * page fault path:
92  * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
93  *   page lock -> i_data_sem (rw)
94  *
95  * buffered write path:
96  * sb_start_write -> i_mutex -> mmap_sem
97  * sb_start_write -> i_mutex -> transaction start -> page lock ->
98  *   i_data_sem (rw)
99  *
100  * truncate:
101  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
102  *   i_mmap_rwsem (w) -> page lock
103  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
104  *   transaction start -> i_data_sem (rw)
105  *
106  * direct IO:
107  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) -> mmap_sem
108  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) ->
109  *   transaction start -> i_data_sem (rw)
110  *
111  * writepages:
112  * transaction start -> page lock(s) -> i_data_sem (rw)
113  */
114
115 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
116 static struct file_system_type ext2_fs_type = {
117         .owner          = THIS_MODULE,
118         .name           = "ext2",
119         .mount          = ext4_mount,
120         .kill_sb        = kill_block_super,
121         .fs_flags       = FS_REQUIRES_DEV,
122 };
123 MODULE_ALIAS_FS("ext2");
124 MODULE_ALIAS("ext2");
125 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
126 #else
127 #define IS_EXT2_SB(sb) (0)
128 #endif
129
130
131 static struct file_system_type ext3_fs_type = {
132         .owner          = THIS_MODULE,
133         .name           = "ext3",
134         .mount          = ext4_mount,
135         .kill_sb        = kill_block_super,
136         .fs_flags       = FS_REQUIRES_DEV,
137 };
138 MODULE_ALIAS_FS("ext3");
139 MODULE_ALIAS("ext3");
140 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
141
142 static int ext4_verify_csum_type(struct super_block *sb,
143                                  struct ext4_super_block *es)
144 {
145         if (!ext4_has_feature_metadata_csum(sb))
146                 return 1;
147
148         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
149 }
150
151 static __le32 ext4_superblock_csum(struct super_block *sb,
152                                    struct ext4_super_block *es)
153 {
154         struct ext4_sb_info *sbi = EXT4_SB(sb);
155         int offset = offsetof(struct ext4_super_block, s_checksum);
156         __u32 csum;
157
158         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
159
160         return cpu_to_le32(csum);
161 }
162
163 static int ext4_superblock_csum_verify(struct super_block *sb,
164                                        struct ext4_super_block *es)
165 {
166         if (!ext4_has_metadata_csum(sb))
167                 return 1;
168
169         return es->s_checksum == ext4_superblock_csum(sb, es);
170 }
171
172 void ext4_superblock_csum_set(struct super_block *sb)
173 {
174         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
175
176         if (!ext4_has_metadata_csum(sb))
177                 return;
178
179         es->s_checksum = ext4_superblock_csum(sb, es);
180 }
181
182 void *ext4_kvmalloc(size_t size, gfp_t flags)
183 {
184         void *ret;
185
186         ret = kmalloc(size, flags | __GFP_NOWARN);
187         if (!ret)
188                 ret = __vmalloc(size, flags, PAGE_KERNEL);
189         return ret;
190 }
191
192 void *ext4_kvzalloc(size_t size, gfp_t flags)
193 {
194         void *ret;
195
196         ret = kzalloc(size, flags | __GFP_NOWARN);
197         if (!ret)
198                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
199         return ret;
200 }
201
202 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
203                                struct ext4_group_desc *bg)
204 {
205         return le32_to_cpu(bg->bg_block_bitmap_lo) |
206                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
207                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
208 }
209
210 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
211                                struct ext4_group_desc *bg)
212 {
213         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
214                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
215                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
216 }
217
218 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
219                               struct ext4_group_desc *bg)
220 {
221         return le32_to_cpu(bg->bg_inode_table_lo) |
222                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
223                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
224 }
225
226 __u32 ext4_free_group_clusters(struct super_block *sb,
227                                struct ext4_group_desc *bg)
228 {
229         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
230                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
231                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
232 }
233
234 __u32 ext4_free_inodes_count(struct super_block *sb,
235                               struct ext4_group_desc *bg)
236 {
237         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
238                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
239                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
240 }
241
242 __u32 ext4_used_dirs_count(struct super_block *sb,
243                               struct ext4_group_desc *bg)
244 {
245         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
246                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
247                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
248 }
249
250 __u32 ext4_itable_unused_count(struct super_block *sb,
251                               struct ext4_group_desc *bg)
252 {
253         return le16_to_cpu(bg->bg_itable_unused_lo) |
254                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
255                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
256 }
257
258 void ext4_block_bitmap_set(struct super_block *sb,
259                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
260 {
261         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
262         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
263                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
264 }
265
266 void ext4_inode_bitmap_set(struct super_block *sb,
267                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
268 {
269         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
270         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
271                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
272 }
273
274 void ext4_inode_table_set(struct super_block *sb,
275                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
276 {
277         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
278         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
279                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
280 }
281
282 void ext4_free_group_clusters_set(struct super_block *sb,
283                                   struct ext4_group_desc *bg, __u32 count)
284 {
285         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
286         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
287                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
288 }
289
290 void ext4_free_inodes_set(struct super_block *sb,
291                           struct ext4_group_desc *bg, __u32 count)
292 {
293         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
294         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
295                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
296 }
297
298 void ext4_used_dirs_set(struct super_block *sb,
299                           struct ext4_group_desc *bg, __u32 count)
300 {
301         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
302         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
303                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
304 }
305
306 void ext4_itable_unused_set(struct super_block *sb,
307                           struct ext4_group_desc *bg, __u32 count)
308 {
309         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
310         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
311                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
312 }
313
314
315 static void __save_error_info(struct super_block *sb, const char *func,
316                             unsigned int line)
317 {
318         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
319
320         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
321         if (bdev_read_only(sb->s_bdev))
322                 return;
323         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
324         es->s_last_error_time = cpu_to_le32(get_seconds());
325         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
326         es->s_last_error_line = cpu_to_le32(line);
327         if (!es->s_first_error_time) {
328                 es->s_first_error_time = es->s_last_error_time;
329                 strncpy(es->s_first_error_func, func,
330                         sizeof(es->s_first_error_func));
331                 es->s_first_error_line = cpu_to_le32(line);
332                 es->s_first_error_ino = es->s_last_error_ino;
333                 es->s_first_error_block = es->s_last_error_block;
334         }
335         /*
336          * Start the daily error reporting function if it hasn't been
337          * started already
338          */
339         if (!es->s_error_count)
340                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
341         le32_add_cpu(&es->s_error_count, 1);
342 }
343
344 static void save_error_info(struct super_block *sb, const char *func,
345                             unsigned int line)
346 {
347         __save_error_info(sb, func, line);
348         ext4_commit_super(sb, 1);
349 }
350
351 /*
352  * The del_gendisk() function uninitializes the disk-specific data
353  * structures, including the bdi structure, without telling anyone
354  * else.  Once this happens, any attempt to call mark_buffer_dirty()
355  * (for example, by ext4_commit_super), will cause a kernel OOPS.
356  * This is a kludge to prevent these oops until we can put in a proper
357  * hook in del_gendisk() to inform the VFS and file system layers.
358  */
359 static int block_device_ejected(struct super_block *sb)
360 {
361         struct inode *bd_inode = sb->s_bdev->bd_inode;
362         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
363
364         return bdi->dev == NULL;
365 }
366
367 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
368 {
369         struct super_block              *sb = journal->j_private;
370         struct ext4_sb_info             *sbi = EXT4_SB(sb);
371         int                             error = is_journal_aborted(journal);
372         struct ext4_journal_cb_entry    *jce;
373
374         BUG_ON(txn->t_state == T_FINISHED);
375         spin_lock(&sbi->s_md_lock);
376         while (!list_empty(&txn->t_private_list)) {
377                 jce = list_entry(txn->t_private_list.next,
378                                  struct ext4_journal_cb_entry, jce_list);
379                 list_del_init(&jce->jce_list);
380                 spin_unlock(&sbi->s_md_lock);
381                 jce->jce_func(sb, jce, error);
382                 spin_lock(&sbi->s_md_lock);
383         }
384         spin_unlock(&sbi->s_md_lock);
385 }
386
387 /* Deal with the reporting of failure conditions on a filesystem such as
388  * inconsistencies detected or read IO failures.
389  *
390  * On ext2, we can store the error state of the filesystem in the
391  * superblock.  That is not possible on ext4, because we may have other
392  * write ordering constraints on the superblock which prevent us from
393  * writing it out straight away; and given that the journal is about to
394  * be aborted, we can't rely on the current, or future, transactions to
395  * write out the superblock safely.
396  *
397  * We'll just use the jbd2_journal_abort() error code to record an error in
398  * the journal instead.  On recovery, the journal will complain about
399  * that error until we've noted it down and cleared it.
400  */
401
402 static void ext4_handle_error(struct super_block *sb)
403 {
404         if (sb->s_flags & MS_RDONLY)
405                 return;
406
407         if (!test_opt(sb, ERRORS_CONT)) {
408                 journal_t *journal = EXT4_SB(sb)->s_journal;
409
410                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
411                 if (journal)
412                         jbd2_journal_abort(journal, -EIO);
413         }
414         if (test_opt(sb, ERRORS_RO)) {
415                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
416                 /*
417                  * Make sure updated value of ->s_mount_flags will be visible
418                  * before ->s_flags update
419                  */
420                 smp_wmb();
421                 sb->s_flags |= MS_RDONLY;
422         }
423         if (test_opt(sb, ERRORS_PANIC)) {
424                 if (EXT4_SB(sb)->s_journal &&
425                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
426                         return;
427                 panic("EXT4-fs (device %s): panic forced after error\n",
428                         sb->s_id);
429         }
430 }
431
432 #define ext4_error_ratelimit(sb)                                        \
433                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
434                              "EXT4-fs error")
435
436 void __ext4_error(struct super_block *sb, const char *function,
437                   unsigned int line, const char *fmt, ...)
438 {
439         struct va_format vaf;
440         va_list args;
441
442         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
443                 return;
444
445         if (ext4_error_ratelimit(sb)) {
446                 va_start(args, fmt);
447                 vaf.fmt = fmt;
448                 vaf.va = &args;
449                 printk(KERN_CRIT
450                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
451                        sb->s_id, function, line, current->comm, &vaf);
452                 va_end(args);
453         }
454         save_error_info(sb, function, line);
455         ext4_handle_error(sb);
456 }
457
458 void __ext4_error_inode(struct inode *inode, const char *function,
459                         unsigned int line, ext4_fsblk_t block,
460                         const char *fmt, ...)
461 {
462         va_list args;
463         struct va_format vaf;
464         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
465
466         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
467                 return;
468
469         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
470         es->s_last_error_block = cpu_to_le64(block);
471         if (ext4_error_ratelimit(inode->i_sb)) {
472                 va_start(args, fmt);
473                 vaf.fmt = fmt;
474                 vaf.va = &args;
475                 if (block)
476                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
477                                "inode #%lu: block %llu: comm %s: %pV\n",
478                                inode->i_sb->s_id, function, line, inode->i_ino,
479                                block, current->comm, &vaf);
480                 else
481                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
482                                "inode #%lu: comm %s: %pV\n",
483                                inode->i_sb->s_id, function, line, inode->i_ino,
484                                current->comm, &vaf);
485                 va_end(args);
486         }
487         save_error_info(inode->i_sb, function, line);
488         ext4_handle_error(inode->i_sb);
489 }
490
491 void __ext4_error_file(struct file *file, const char *function,
492                        unsigned int line, ext4_fsblk_t block,
493                        const char *fmt, ...)
494 {
495         va_list args;
496         struct va_format vaf;
497         struct ext4_super_block *es;
498         struct inode *inode = file_inode(file);
499         char pathname[80], *path;
500
501         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
502                 return;
503
504         es = EXT4_SB(inode->i_sb)->s_es;
505         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
506         if (ext4_error_ratelimit(inode->i_sb)) {
507                 path = file_path(file, pathname, sizeof(pathname));
508                 if (IS_ERR(path))
509                         path = "(unknown)";
510                 va_start(args, fmt);
511                 vaf.fmt = fmt;
512                 vaf.va = &args;
513                 if (block)
514                         printk(KERN_CRIT
515                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
516                                "block %llu: comm %s: path %s: %pV\n",
517                                inode->i_sb->s_id, function, line, inode->i_ino,
518                                block, current->comm, path, &vaf);
519                 else
520                         printk(KERN_CRIT
521                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
522                                "comm %s: path %s: %pV\n",
523                                inode->i_sb->s_id, function, line, inode->i_ino,
524                                current->comm, path, &vaf);
525                 va_end(args);
526         }
527         save_error_info(inode->i_sb, function, line);
528         ext4_handle_error(inode->i_sb);
529 }
530
531 const char *ext4_decode_error(struct super_block *sb, int errno,
532                               char nbuf[16])
533 {
534         char *errstr = NULL;
535
536         switch (errno) {
537         case -EFSCORRUPTED:
538                 errstr = "Corrupt filesystem";
539                 break;
540         case -EFSBADCRC:
541                 errstr = "Filesystem failed CRC";
542                 break;
543         case -EIO:
544                 errstr = "IO failure";
545                 break;
546         case -ENOMEM:
547                 errstr = "Out of memory";
548                 break;
549         case -EROFS:
550                 if (!sb || (EXT4_SB(sb)->s_journal &&
551                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
552                         errstr = "Journal has aborted";
553                 else
554                         errstr = "Readonly filesystem";
555                 break;
556         default:
557                 /* If the caller passed in an extra buffer for unknown
558                  * errors, textualise them now.  Else we just return
559                  * NULL. */
560                 if (nbuf) {
561                         /* Check for truncated error codes... */
562                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
563                                 errstr = nbuf;
564                 }
565                 break;
566         }
567
568         return errstr;
569 }
570
571 /* __ext4_std_error decodes expected errors from journaling functions
572  * automatically and invokes the appropriate error response.  */
573
574 void __ext4_std_error(struct super_block *sb, const char *function,
575                       unsigned int line, int errno)
576 {
577         char nbuf[16];
578         const char *errstr;
579
580         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
581                 return;
582
583         /* Special case: if the error is EROFS, and we're not already
584          * inside a transaction, then there's really no point in logging
585          * an error. */
586         if (errno == -EROFS && journal_current_handle() == NULL &&
587             (sb->s_flags & MS_RDONLY))
588                 return;
589
590         if (ext4_error_ratelimit(sb)) {
591                 errstr = ext4_decode_error(sb, errno, nbuf);
592                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
593                        sb->s_id, function, line, errstr);
594         }
595
596         save_error_info(sb, function, line);
597         ext4_handle_error(sb);
598 }
599
600 /*
601  * ext4_abort is a much stronger failure handler than ext4_error.  The
602  * abort function may be used to deal with unrecoverable failures such
603  * as journal IO errors or ENOMEM at a critical moment in log management.
604  *
605  * We unconditionally force the filesystem into an ABORT|READONLY state,
606  * unless the error response on the fs has been set to panic in which
607  * case we take the easy way out and panic immediately.
608  */
609
610 void __ext4_abort(struct super_block *sb, const char *function,
611                 unsigned int line, const char *fmt, ...)
612 {
613         struct va_format vaf;
614         va_list args;
615
616         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
617                 return;
618
619         save_error_info(sb, function, line);
620         va_start(args, fmt);
621         vaf.fmt = fmt;
622         vaf.va = &args;
623         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
624                sb->s_id, function, line, &vaf);
625         va_end(args);
626
627         if ((sb->s_flags & MS_RDONLY) == 0) {
628                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
629                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
630                 /*
631                  * Make sure updated value of ->s_mount_flags will be visible
632                  * before ->s_flags update
633                  */
634                 smp_wmb();
635                 sb->s_flags |= MS_RDONLY;
636                 if (EXT4_SB(sb)->s_journal)
637                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
638                 save_error_info(sb, function, line);
639         }
640         if (test_opt(sb, ERRORS_PANIC)) {
641                 if (EXT4_SB(sb)->s_journal &&
642                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
643                         return;
644                 panic("EXT4-fs panic from previous error\n");
645         }
646 }
647
648 void __ext4_msg(struct super_block *sb,
649                 const char *prefix, const char *fmt, ...)
650 {
651         struct va_format vaf;
652         va_list args;
653
654         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
655                 return;
656
657         va_start(args, fmt);
658         vaf.fmt = fmt;
659         vaf.va = &args;
660         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
661         va_end(args);
662 }
663
664 #define ext4_warning_ratelimit(sb)                                      \
665                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
666                              "EXT4-fs warning")
667
668 void __ext4_warning(struct super_block *sb, const char *function,
669                     unsigned int line, const char *fmt, ...)
670 {
671         struct va_format vaf;
672         va_list args;
673
674         if (!ext4_warning_ratelimit(sb))
675                 return;
676
677         va_start(args, fmt);
678         vaf.fmt = fmt;
679         vaf.va = &args;
680         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
681                sb->s_id, function, line, &vaf);
682         va_end(args);
683 }
684
685 void __ext4_warning_inode(const struct inode *inode, const char *function,
686                           unsigned int line, const char *fmt, ...)
687 {
688         struct va_format vaf;
689         va_list args;
690
691         if (!ext4_warning_ratelimit(inode->i_sb))
692                 return;
693
694         va_start(args, fmt);
695         vaf.fmt = fmt;
696         vaf.va = &args;
697         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
698                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
699                function, line, inode->i_ino, current->comm, &vaf);
700         va_end(args);
701 }
702
703 void __ext4_grp_locked_error(const char *function, unsigned int line,
704                              struct super_block *sb, ext4_group_t grp,
705                              unsigned long ino, ext4_fsblk_t block,
706                              const char *fmt, ...)
707 __releases(bitlock)
708 __acquires(bitlock)
709 {
710         struct va_format vaf;
711         va_list args;
712         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
713
714         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
715                 return;
716
717         es->s_last_error_ino = cpu_to_le32(ino);
718         es->s_last_error_block = cpu_to_le64(block);
719         __save_error_info(sb, function, line);
720
721         if (ext4_error_ratelimit(sb)) {
722                 va_start(args, fmt);
723                 vaf.fmt = fmt;
724                 vaf.va = &args;
725                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
726                        sb->s_id, function, line, grp);
727                 if (ino)
728                         printk(KERN_CONT "inode %lu: ", ino);
729                 if (block)
730                         printk(KERN_CONT "block %llu:",
731                                (unsigned long long) block);
732                 printk(KERN_CONT "%pV\n", &vaf);
733                 va_end(args);
734         }
735
736         if (test_opt(sb, ERRORS_CONT)) {
737                 ext4_commit_super(sb, 0);
738                 return;
739         }
740
741         ext4_unlock_group(sb, grp);
742         ext4_handle_error(sb);
743         /*
744          * We only get here in the ERRORS_RO case; relocking the group
745          * may be dangerous, but nothing bad will happen since the
746          * filesystem will have already been marked read/only and the
747          * journal has been aborted.  We return 1 as a hint to callers
748          * who might what to use the return value from
749          * ext4_grp_locked_error() to distinguish between the
750          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
751          * aggressively from the ext4 function in question, with a
752          * more appropriate error code.
753          */
754         ext4_lock_group(sb, grp);
755         return;
756 }
757
758 void ext4_update_dynamic_rev(struct super_block *sb)
759 {
760         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
761
762         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
763                 return;
764
765         ext4_warning(sb,
766                      "updating to rev %d because of new feature flag, "
767                      "running e2fsck is recommended",
768                      EXT4_DYNAMIC_REV);
769
770         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
771         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
772         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
773         /* leave es->s_feature_*compat flags alone */
774         /* es->s_uuid will be set by e2fsck if empty */
775
776         /*
777          * The rest of the superblock fields should be zero, and if not it
778          * means they are likely already in use, so leave them alone.  We
779          * can leave it up to e2fsck to clean up any inconsistencies there.
780          */
781 }
782
783 /*
784  * Open the external journal device
785  */
786 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
787 {
788         struct block_device *bdev;
789         char b[BDEVNAME_SIZE];
790
791         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
792         if (IS_ERR(bdev))
793                 goto fail;
794         return bdev;
795
796 fail:
797         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
798                         __bdevname(dev, b), PTR_ERR(bdev));
799         return NULL;
800 }
801
802 /*
803  * Release the journal device
804  */
805 static void ext4_blkdev_put(struct block_device *bdev)
806 {
807         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
808 }
809
810 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
811 {
812         struct block_device *bdev;
813         bdev = sbi->journal_bdev;
814         if (bdev) {
815                 ext4_blkdev_put(bdev);
816                 sbi->journal_bdev = NULL;
817         }
818 }
819
820 static inline struct inode *orphan_list_entry(struct list_head *l)
821 {
822         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
823 }
824
825 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
826 {
827         struct list_head *l;
828
829         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
830                  le32_to_cpu(sbi->s_es->s_last_orphan));
831
832         printk(KERN_ERR "sb_info orphan list:\n");
833         list_for_each(l, &sbi->s_orphan) {
834                 struct inode *inode = orphan_list_entry(l);
835                 printk(KERN_ERR "  "
836                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
837                        inode->i_sb->s_id, inode->i_ino, inode,
838                        inode->i_mode, inode->i_nlink,
839                        NEXT_ORPHAN(inode));
840         }
841 }
842
843 static void ext4_put_super(struct super_block *sb)
844 {
845         struct ext4_sb_info *sbi = EXT4_SB(sb);
846         struct ext4_super_block *es = sbi->s_es;
847         int aborted = 0;
848         int i, err;
849
850         ext4_unregister_li_request(sb);
851         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
852
853         flush_workqueue(sbi->rsv_conversion_wq);
854         destroy_workqueue(sbi->rsv_conversion_wq);
855
856         if (sbi->s_journal) {
857                 aborted = is_journal_aborted(sbi->s_journal);
858                 err = jbd2_journal_destroy(sbi->s_journal);
859                 sbi->s_journal = NULL;
860                 if ((err < 0) && !aborted)
861                         ext4_abort(sb, "Couldn't clean up the journal");
862         }
863
864         ext4_unregister_sysfs(sb);
865         ext4_es_unregister_shrinker(sbi);
866         del_timer_sync(&sbi->s_err_report);
867         ext4_release_system_zone(sb);
868         ext4_mb_release(sb);
869         ext4_ext_release(sb);
870
871         if (!(sb->s_flags & MS_RDONLY) && !aborted) {
872                 ext4_clear_feature_journal_needs_recovery(sb);
873                 es->s_state = cpu_to_le16(sbi->s_mount_state);
874         }
875         if (!(sb->s_flags & MS_RDONLY))
876                 ext4_commit_super(sb, 1);
877
878         for (i = 0; i < sbi->s_gdb_count; i++)
879                 brelse(sbi->s_group_desc[i]);
880         kvfree(sbi->s_group_desc);
881         kvfree(sbi->s_flex_groups);
882         percpu_counter_destroy(&sbi->s_freeclusters_counter);
883         percpu_counter_destroy(&sbi->s_freeinodes_counter);
884         percpu_counter_destroy(&sbi->s_dirs_counter);
885         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
886         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
887 #ifdef CONFIG_QUOTA
888         for (i = 0; i < EXT4_MAXQUOTAS; i++)
889                 kfree(sbi->s_qf_names[i]);
890 #endif
891
892         /* Debugging code just in case the in-memory inode orphan list
893          * isn't empty.  The on-disk one can be non-empty if we've
894          * detected an error and taken the fs readonly, but the
895          * in-memory list had better be clean by this point. */
896         if (!list_empty(&sbi->s_orphan))
897                 dump_orphan_list(sb, sbi);
898         J_ASSERT(list_empty(&sbi->s_orphan));
899
900         sync_blockdev(sb->s_bdev);
901         invalidate_bdev(sb->s_bdev);
902         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
903                 /*
904                  * Invalidate the journal device's buffers.  We don't want them
905                  * floating about in memory - the physical journal device may
906                  * hotswapped, and it breaks the `ro-after' testing code.
907                  */
908                 sync_blockdev(sbi->journal_bdev);
909                 invalidate_bdev(sbi->journal_bdev);
910                 ext4_blkdev_remove(sbi);
911         }
912         if (sbi->s_mb_cache) {
913                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
914                 sbi->s_mb_cache = NULL;
915         }
916         if (sbi->s_mmp_tsk)
917                 kthread_stop(sbi->s_mmp_tsk);
918         brelse(sbi->s_sbh);
919         sb->s_fs_info = NULL;
920         /*
921          * Now that we are completely done shutting down the
922          * superblock, we need to actually destroy the kobject.
923          */
924         kobject_put(&sbi->s_kobj);
925         wait_for_completion(&sbi->s_kobj_unregister);
926         if (sbi->s_chksum_driver)
927                 crypto_free_shash(sbi->s_chksum_driver);
928         kfree(sbi->s_blockgroup_lock);
929         kfree(sbi);
930 }
931
932 static struct kmem_cache *ext4_inode_cachep;
933
934 /*
935  * Called inside transaction, so use GFP_NOFS
936  */
937 static struct inode *ext4_alloc_inode(struct super_block *sb)
938 {
939         struct ext4_inode_info *ei;
940
941         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
942         if (!ei)
943                 return NULL;
944
945         ei->vfs_inode.i_version = 1;
946         spin_lock_init(&ei->i_raw_lock);
947         INIT_LIST_HEAD(&ei->i_prealloc_list);
948         spin_lock_init(&ei->i_prealloc_lock);
949         ext4_es_init_tree(&ei->i_es_tree);
950         rwlock_init(&ei->i_es_lock);
951         INIT_LIST_HEAD(&ei->i_es_list);
952         ei->i_es_all_nr = 0;
953         ei->i_es_shk_nr = 0;
954         ei->i_es_shrink_lblk = 0;
955         ei->i_reserved_data_blocks = 0;
956         ei->i_reserved_meta_blocks = 0;
957         ei->i_allocated_meta_blocks = 0;
958         ei->i_da_metadata_calc_len = 0;
959         ei->i_da_metadata_calc_last_lblock = 0;
960         spin_lock_init(&(ei->i_block_reservation_lock));
961 #ifdef CONFIG_QUOTA
962         ei->i_reserved_quota = 0;
963         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
964 #endif
965         ei->jinode = NULL;
966         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
967         spin_lock_init(&ei->i_completed_io_lock);
968         ei->i_sync_tid = 0;
969         ei->i_datasync_tid = 0;
970         atomic_set(&ei->i_unwritten, 0);
971         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
972         return &ei->vfs_inode;
973 }
974
975 static int ext4_drop_inode(struct inode *inode)
976 {
977         int drop = generic_drop_inode(inode);
978
979         trace_ext4_drop_inode(inode, drop);
980         return drop;
981 }
982
983 static void ext4_i_callback(struct rcu_head *head)
984 {
985         struct inode *inode = container_of(head, struct inode, i_rcu);
986         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
987 }
988
989 static void ext4_destroy_inode(struct inode *inode)
990 {
991         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
992                 ext4_msg(inode->i_sb, KERN_ERR,
993                          "Inode %lu (%p): orphan list check failed!",
994                          inode->i_ino, EXT4_I(inode));
995                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
996                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
997                                 true);
998                 dump_stack();
999         }
1000         call_rcu(&inode->i_rcu, ext4_i_callback);
1001 }
1002
1003 static void init_once(void *foo)
1004 {
1005         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1006
1007         INIT_LIST_HEAD(&ei->i_orphan);
1008         init_rwsem(&ei->xattr_sem);
1009         init_rwsem(&ei->i_data_sem);
1010         init_rwsem(&ei->i_mmap_sem);
1011         inode_init_once(&ei->vfs_inode);
1012 }
1013
1014 static int __init init_inodecache(void)
1015 {
1016         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1017                                              sizeof(struct ext4_inode_info),
1018                                              0, (SLAB_RECLAIM_ACCOUNT|
1019                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1020                                              init_once);
1021         if (ext4_inode_cachep == NULL)
1022                 return -ENOMEM;
1023         return 0;
1024 }
1025
1026 static void destroy_inodecache(void)
1027 {
1028         /*
1029          * Make sure all delayed rcu free inodes are flushed before we
1030          * destroy cache.
1031          */
1032         rcu_barrier();
1033         kmem_cache_destroy(ext4_inode_cachep);
1034 }
1035
1036 void ext4_clear_inode(struct inode *inode)
1037 {
1038         invalidate_inode_buffers(inode);
1039         clear_inode(inode);
1040         dquot_drop(inode);
1041         ext4_discard_preallocations(inode);
1042         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1043         if (EXT4_I(inode)->jinode) {
1044                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1045                                                EXT4_I(inode)->jinode);
1046                 jbd2_free_inode(EXT4_I(inode)->jinode);
1047                 EXT4_I(inode)->jinode = NULL;
1048         }
1049 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1050         fscrypt_put_encryption_info(inode, NULL);
1051 #endif
1052 }
1053
1054 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1055                                         u64 ino, u32 generation)
1056 {
1057         struct inode *inode;
1058
1059         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1060                 return ERR_PTR(-ESTALE);
1061         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1062                 return ERR_PTR(-ESTALE);
1063
1064         /* iget isn't really right if the inode is currently unallocated!!
1065          *
1066          * ext4_read_inode will return a bad_inode if the inode had been
1067          * deleted, so we should be safe.
1068          *
1069          * Currently we don't know the generation for parent directory, so
1070          * a generation of 0 means "accept any"
1071          */
1072         inode = ext4_iget_normal(sb, ino);
1073         if (IS_ERR(inode))
1074                 return ERR_CAST(inode);
1075         if (generation && inode->i_generation != generation) {
1076                 iput(inode);
1077                 return ERR_PTR(-ESTALE);
1078         }
1079
1080         return inode;
1081 }
1082
1083 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1084                                         int fh_len, int fh_type)
1085 {
1086         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1087                                     ext4_nfs_get_inode);
1088 }
1089
1090 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1091                                         int fh_len, int fh_type)
1092 {
1093         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1094                                     ext4_nfs_get_inode);
1095 }
1096
1097 /*
1098  * Try to release metadata pages (indirect blocks, directories) which are
1099  * mapped via the block device.  Since these pages could have journal heads
1100  * which would prevent try_to_free_buffers() from freeing them, we must use
1101  * jbd2 layer's try_to_free_buffers() function to release them.
1102  */
1103 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1104                                  gfp_t wait)
1105 {
1106         journal_t *journal = EXT4_SB(sb)->s_journal;
1107
1108         WARN_ON(PageChecked(page));
1109         if (!page_has_buffers(page))
1110                 return 0;
1111         if (journal)
1112                 return jbd2_journal_try_to_free_buffers(journal, page,
1113                                                 wait & ~__GFP_DIRECT_RECLAIM);
1114         return try_to_free_buffers(page);
1115 }
1116
1117 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1118 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1119 {
1120         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1121                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1122 }
1123
1124 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1125                                                         void *fs_data)
1126 {
1127         handle_t *handle = fs_data;
1128         int res, res2, retries = 0;
1129
1130         res = ext4_convert_inline_data(inode);
1131         if (res)
1132                 return res;
1133
1134         /*
1135          * If a journal handle was specified, then the encryption context is
1136          * being set on a new inode via inheritance and is part of a larger
1137          * transaction to create the inode.  Otherwise the encryption context is
1138          * being set on an existing inode in its own transaction.  Only in the
1139          * latter case should the "retry on ENOSPC" logic be used.
1140          */
1141
1142         if (handle) {
1143                 res = ext4_xattr_set_handle(handle, inode,
1144                                             EXT4_XATTR_INDEX_ENCRYPTION,
1145                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1146                                             ctx, len, 0);
1147                 if (!res) {
1148                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1149                         ext4_clear_inode_state(inode,
1150                                         EXT4_STATE_MAY_INLINE_DATA);
1151                         /*
1152                          * Update inode->i_flags - e.g. S_DAX may get disabled
1153                          */
1154                         ext4_set_inode_flags(inode);
1155                 }
1156                 return res;
1157         }
1158
1159 retry:
1160         handle = ext4_journal_start(inode, EXT4_HT_MISC,
1161                         ext4_jbd2_credits_xattr(inode));
1162         if (IS_ERR(handle))
1163                 return PTR_ERR(handle);
1164
1165         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1166                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1167                                     ctx, len, 0);
1168         if (!res) {
1169                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1170                 /* Update inode->i_flags - e.g. S_DAX may get disabled */
1171                 ext4_set_inode_flags(inode);
1172                 res = ext4_mark_inode_dirty(handle, inode);
1173                 if (res)
1174                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1175         }
1176         res2 = ext4_journal_stop(handle);
1177
1178         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1179                 goto retry;
1180         if (!res)
1181                 res = res2;
1182         return res;
1183 }
1184
1185 static int ext4_dummy_context(struct inode *inode)
1186 {
1187         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1188 }
1189
1190 static unsigned ext4_max_namelen(struct inode *inode)
1191 {
1192         return S_ISLNK(inode->i_mode) ? inode->i_sb->s_blocksize :
1193                 EXT4_NAME_LEN;
1194 }
1195
1196 static const struct fscrypt_operations ext4_cryptops = {
1197         .key_prefix             = "ext4:",
1198         .get_context            = ext4_get_context,
1199         .set_context            = ext4_set_context,
1200         .dummy_context          = ext4_dummy_context,
1201         .is_encrypted           = ext4_encrypted_inode,
1202         .empty_dir              = ext4_empty_dir,
1203         .max_namelen            = ext4_max_namelen,
1204 };
1205 #else
1206 static const struct fscrypt_operations ext4_cryptops = {
1207         .is_encrypted           = ext4_encrypted_inode,
1208 };
1209 #endif
1210
1211 #ifdef CONFIG_QUOTA
1212 static const char * const quotatypes[] = INITQFNAMES;
1213 #define QTYPE2NAME(t) (quotatypes[t])
1214
1215 static int ext4_write_dquot(struct dquot *dquot);
1216 static int ext4_acquire_dquot(struct dquot *dquot);
1217 static int ext4_release_dquot(struct dquot *dquot);
1218 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1219 static int ext4_write_info(struct super_block *sb, int type);
1220 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1221                          const struct path *path);
1222 static int ext4_quota_off(struct super_block *sb, int type);
1223 static int ext4_quota_on_mount(struct super_block *sb, int type);
1224 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1225                                size_t len, loff_t off);
1226 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1227                                 const char *data, size_t len, loff_t off);
1228 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1229                              unsigned int flags);
1230 static int ext4_enable_quotas(struct super_block *sb);
1231 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1232
1233 static struct dquot **ext4_get_dquots(struct inode *inode)
1234 {
1235         return EXT4_I(inode)->i_dquot;
1236 }
1237
1238 static const struct dquot_operations ext4_quota_operations = {
1239         .get_reserved_space = ext4_get_reserved_space,
1240         .write_dquot    = ext4_write_dquot,
1241         .acquire_dquot  = ext4_acquire_dquot,
1242         .release_dquot  = ext4_release_dquot,
1243         .mark_dirty     = ext4_mark_dquot_dirty,
1244         .write_info     = ext4_write_info,
1245         .alloc_dquot    = dquot_alloc,
1246         .destroy_dquot  = dquot_destroy,
1247         .get_projid     = ext4_get_projid,
1248         .get_next_id    = ext4_get_next_id,
1249 };
1250
1251 static const struct quotactl_ops ext4_qctl_operations = {
1252         .quota_on       = ext4_quota_on,
1253         .quota_off      = ext4_quota_off,
1254         .quota_sync     = dquot_quota_sync,
1255         .get_state      = dquot_get_state,
1256         .set_info       = dquot_set_dqinfo,
1257         .get_dqblk      = dquot_get_dqblk,
1258         .set_dqblk      = dquot_set_dqblk,
1259         .get_nextdqblk  = dquot_get_next_dqblk,
1260 };
1261 #endif
1262
1263 static const struct super_operations ext4_sops = {
1264         .alloc_inode    = ext4_alloc_inode,
1265         .destroy_inode  = ext4_destroy_inode,
1266         .write_inode    = ext4_write_inode,
1267         .dirty_inode    = ext4_dirty_inode,
1268         .drop_inode     = ext4_drop_inode,
1269         .evict_inode    = ext4_evict_inode,
1270         .put_super      = ext4_put_super,
1271         .sync_fs        = ext4_sync_fs,
1272         .freeze_fs      = ext4_freeze,
1273         .unfreeze_fs    = ext4_unfreeze,
1274         .statfs         = ext4_statfs,
1275         .remount_fs     = ext4_remount,
1276         .show_options   = ext4_show_options,
1277 #ifdef CONFIG_QUOTA
1278         .quota_read     = ext4_quota_read,
1279         .quota_write    = ext4_quota_write,
1280         .get_dquots     = ext4_get_dquots,
1281 #endif
1282         .bdev_try_to_free_page = bdev_try_to_free_page,
1283 };
1284
1285 static const struct export_operations ext4_export_ops = {
1286         .fh_to_dentry = ext4_fh_to_dentry,
1287         .fh_to_parent = ext4_fh_to_parent,
1288         .get_parent = ext4_get_parent,
1289 };
1290
1291 enum {
1292         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1293         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1294         Opt_nouid32, Opt_debug, Opt_removed,
1295         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1296         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1297         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1298         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1299         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1300         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1301         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1302         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1303         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1304         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1305         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1306         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1307         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1308         Opt_inode_readahead_blks, Opt_journal_ioprio,
1309         Opt_dioread_nolock, Opt_dioread_lock,
1310         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1311         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1312 };
1313
1314 static const match_table_t tokens = {
1315         {Opt_bsd_df, "bsddf"},
1316         {Opt_minix_df, "minixdf"},
1317         {Opt_grpid, "grpid"},
1318         {Opt_grpid, "bsdgroups"},
1319         {Opt_nogrpid, "nogrpid"},
1320         {Opt_nogrpid, "sysvgroups"},
1321         {Opt_resgid, "resgid=%u"},
1322         {Opt_resuid, "resuid=%u"},
1323         {Opt_sb, "sb=%u"},
1324         {Opt_err_cont, "errors=continue"},
1325         {Opt_err_panic, "errors=panic"},
1326         {Opt_err_ro, "errors=remount-ro"},
1327         {Opt_nouid32, "nouid32"},
1328         {Opt_debug, "debug"},
1329         {Opt_removed, "oldalloc"},
1330         {Opt_removed, "orlov"},
1331         {Opt_user_xattr, "user_xattr"},
1332         {Opt_nouser_xattr, "nouser_xattr"},
1333         {Opt_acl, "acl"},
1334         {Opt_noacl, "noacl"},
1335         {Opt_noload, "norecovery"},
1336         {Opt_noload, "noload"},
1337         {Opt_removed, "nobh"},
1338         {Opt_removed, "bh"},
1339         {Opt_commit, "commit=%u"},
1340         {Opt_min_batch_time, "min_batch_time=%u"},
1341         {Opt_max_batch_time, "max_batch_time=%u"},
1342         {Opt_journal_dev, "journal_dev=%u"},
1343         {Opt_journal_path, "journal_path=%s"},
1344         {Opt_journal_checksum, "journal_checksum"},
1345         {Opt_nojournal_checksum, "nojournal_checksum"},
1346         {Opt_journal_async_commit, "journal_async_commit"},
1347         {Opt_abort, "abort"},
1348         {Opt_data_journal, "data=journal"},
1349         {Opt_data_ordered, "data=ordered"},
1350         {Opt_data_writeback, "data=writeback"},
1351         {Opt_data_err_abort, "data_err=abort"},
1352         {Opt_data_err_ignore, "data_err=ignore"},
1353         {Opt_offusrjquota, "usrjquota="},
1354         {Opt_usrjquota, "usrjquota=%s"},
1355         {Opt_offgrpjquota, "grpjquota="},
1356         {Opt_grpjquota, "grpjquota=%s"},
1357         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1358         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1359         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1360         {Opt_grpquota, "grpquota"},
1361         {Opt_noquota, "noquota"},
1362         {Opt_quota, "quota"},
1363         {Opt_usrquota, "usrquota"},
1364         {Opt_prjquota, "prjquota"},
1365         {Opt_barrier, "barrier=%u"},
1366         {Opt_barrier, "barrier"},
1367         {Opt_nobarrier, "nobarrier"},
1368         {Opt_i_version, "i_version"},
1369         {Opt_dax, "dax"},
1370         {Opt_stripe, "stripe=%u"},
1371         {Opt_delalloc, "delalloc"},
1372         {Opt_lazytime, "lazytime"},
1373         {Opt_nolazytime, "nolazytime"},
1374         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1375         {Opt_nodelalloc, "nodelalloc"},
1376         {Opt_removed, "mblk_io_submit"},
1377         {Opt_removed, "nomblk_io_submit"},
1378         {Opt_block_validity, "block_validity"},
1379         {Opt_noblock_validity, "noblock_validity"},
1380         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1381         {Opt_journal_ioprio, "journal_ioprio=%u"},
1382         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1383         {Opt_auto_da_alloc, "auto_da_alloc"},
1384         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1385         {Opt_dioread_nolock, "dioread_nolock"},
1386         {Opt_dioread_lock, "dioread_lock"},
1387         {Opt_discard, "discard"},
1388         {Opt_nodiscard, "nodiscard"},
1389         {Opt_init_itable, "init_itable=%u"},
1390         {Opt_init_itable, "init_itable"},
1391         {Opt_noinit_itable, "noinit_itable"},
1392         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1393         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1394         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1395         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1396         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1397         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1398         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1399         {Opt_err, NULL},
1400 };
1401
1402 static ext4_fsblk_t get_sb_block(void **data)
1403 {
1404         ext4_fsblk_t    sb_block;
1405         char            *options = (char *) *data;
1406
1407         if (!options || strncmp(options, "sb=", 3) != 0)
1408                 return 1;       /* Default location */
1409
1410         options += 3;
1411         /* TODO: use simple_strtoll with >32bit ext4 */
1412         sb_block = simple_strtoul(options, &options, 0);
1413         if (*options && *options != ',') {
1414                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1415                        (char *) *data);
1416                 return 1;
1417         }
1418         if (*options == ',')
1419                 options++;
1420         *data = (void *) options;
1421
1422         return sb_block;
1423 }
1424
1425 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1426 static const char deprecated_msg[] =
1427         "Mount option \"%s\" will be removed by %s\n"
1428         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1429
1430 #ifdef CONFIG_QUOTA
1431 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1432 {
1433         struct ext4_sb_info *sbi = EXT4_SB(sb);
1434         char *qname;
1435         int ret = -1;
1436
1437         if (sb_any_quota_loaded(sb) &&
1438                 !sbi->s_qf_names[qtype]) {
1439                 ext4_msg(sb, KERN_ERR,
1440                         "Cannot change journaled "
1441                         "quota options when quota turned on");
1442                 return -1;
1443         }
1444         if (ext4_has_feature_quota(sb)) {
1445                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1446                          "ignored when QUOTA feature is enabled");
1447                 return 1;
1448         }
1449         qname = match_strdup(args);
1450         if (!qname) {
1451                 ext4_msg(sb, KERN_ERR,
1452                         "Not enough memory for storing quotafile name");
1453                 return -1;
1454         }
1455         if (sbi->s_qf_names[qtype]) {
1456                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1457                         ret = 1;
1458                 else
1459                         ext4_msg(sb, KERN_ERR,
1460                                  "%s quota file already specified",
1461                                  QTYPE2NAME(qtype));
1462                 goto errout;
1463         }
1464         if (strchr(qname, '/')) {
1465                 ext4_msg(sb, KERN_ERR,
1466                         "quotafile must be on filesystem root");
1467                 goto errout;
1468         }
1469         sbi->s_qf_names[qtype] = qname;
1470         set_opt(sb, QUOTA);
1471         return 1;
1472 errout:
1473         kfree(qname);
1474         return ret;
1475 }
1476
1477 static int clear_qf_name(struct super_block *sb, int qtype)
1478 {
1479
1480         struct ext4_sb_info *sbi = EXT4_SB(sb);
1481
1482         if (sb_any_quota_loaded(sb) &&
1483                 sbi->s_qf_names[qtype]) {
1484                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1485                         " when quota turned on");
1486                 return -1;
1487         }
1488         kfree(sbi->s_qf_names[qtype]);
1489         sbi->s_qf_names[qtype] = NULL;
1490         return 1;
1491 }
1492 #endif
1493
1494 #define MOPT_SET        0x0001
1495 #define MOPT_CLEAR      0x0002
1496 #define MOPT_NOSUPPORT  0x0004
1497 #define MOPT_EXPLICIT   0x0008
1498 #define MOPT_CLEAR_ERR  0x0010
1499 #define MOPT_GTE0       0x0020
1500 #ifdef CONFIG_QUOTA
1501 #define MOPT_Q          0
1502 #define MOPT_QFMT       0x0040
1503 #else
1504 #define MOPT_Q          MOPT_NOSUPPORT
1505 #define MOPT_QFMT       MOPT_NOSUPPORT
1506 #endif
1507 #define MOPT_DATAJ      0x0080
1508 #define MOPT_NO_EXT2    0x0100
1509 #define MOPT_NO_EXT3    0x0200
1510 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1511 #define MOPT_STRING     0x0400
1512
1513 static const struct mount_opts {
1514         int     token;
1515         int     mount_opt;
1516         int     flags;
1517 } ext4_mount_opts[] = {
1518         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1519         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1520         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1521         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1522         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1523         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1524         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1525          MOPT_EXT4_ONLY | MOPT_SET},
1526         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1527          MOPT_EXT4_ONLY | MOPT_CLEAR},
1528         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1529         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1530         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1531          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1532         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1533          MOPT_EXT4_ONLY | MOPT_CLEAR},
1534         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1535          MOPT_EXT4_ONLY | MOPT_CLEAR},
1536         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1537          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1538         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1539                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1540          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1541         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1542         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1543         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1544         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1545         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1546          MOPT_NO_EXT2},
1547         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1548          MOPT_NO_EXT2},
1549         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1550         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1551         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1552         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1553         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1554         {Opt_commit, 0, MOPT_GTE0},
1555         {Opt_max_batch_time, 0, MOPT_GTE0},
1556         {Opt_min_batch_time, 0, MOPT_GTE0},
1557         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1558         {Opt_init_itable, 0, MOPT_GTE0},
1559         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1560         {Opt_stripe, 0, MOPT_GTE0},
1561         {Opt_resuid, 0, MOPT_GTE0},
1562         {Opt_resgid, 0, MOPT_GTE0},
1563         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1564         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1565         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1566         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1567         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1568         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1569          MOPT_NO_EXT2 | MOPT_DATAJ},
1570         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1571         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1572 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1573         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1574         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1575 #else
1576         {Opt_acl, 0, MOPT_NOSUPPORT},
1577         {Opt_noacl, 0, MOPT_NOSUPPORT},
1578 #endif
1579         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1580         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1581         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1582         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1583         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1584                                                         MOPT_SET | MOPT_Q},
1585         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1586                                                         MOPT_SET | MOPT_Q},
1587         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1588                                                         MOPT_SET | MOPT_Q},
1589         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1590                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1591                                                         MOPT_CLEAR | MOPT_Q},
1592         {Opt_usrjquota, 0, MOPT_Q},
1593         {Opt_grpjquota, 0, MOPT_Q},
1594         {Opt_offusrjquota, 0, MOPT_Q},
1595         {Opt_offgrpjquota, 0, MOPT_Q},
1596         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1597         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1598         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1599         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1600         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1601         {Opt_err, 0, 0}
1602 };
1603
1604 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1605                             substring_t *args, unsigned long *journal_devnum,
1606                             unsigned int *journal_ioprio, int is_remount)
1607 {
1608         struct ext4_sb_info *sbi = EXT4_SB(sb);
1609         const struct mount_opts *m;
1610         kuid_t uid;
1611         kgid_t gid;
1612         int arg = 0;
1613
1614 #ifdef CONFIG_QUOTA
1615         if (token == Opt_usrjquota)
1616                 return set_qf_name(sb, USRQUOTA, &args[0]);
1617         else if (token == Opt_grpjquota)
1618                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1619         else if (token == Opt_offusrjquota)
1620                 return clear_qf_name(sb, USRQUOTA);
1621         else if (token == Opt_offgrpjquota)
1622                 return clear_qf_name(sb, GRPQUOTA);
1623 #endif
1624         switch (token) {
1625         case Opt_noacl:
1626         case Opt_nouser_xattr:
1627                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1628                 break;
1629         case Opt_sb:
1630                 return 1;       /* handled by get_sb_block() */
1631         case Opt_removed:
1632                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1633                 return 1;
1634         case Opt_abort:
1635                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1636                 return 1;
1637         case Opt_i_version:
1638                 sb->s_flags |= MS_I_VERSION;
1639                 return 1;
1640         case Opt_lazytime:
1641                 sb->s_flags |= MS_LAZYTIME;
1642                 return 1;
1643         case Opt_nolazytime:
1644                 sb->s_flags &= ~MS_LAZYTIME;
1645                 return 1;
1646         }
1647
1648         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1649                 if (token == m->token)
1650                         break;
1651
1652         if (m->token == Opt_err) {
1653                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1654                          "or missing value", opt);
1655                 return -1;
1656         }
1657
1658         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1659                 ext4_msg(sb, KERN_ERR,
1660                          "Mount option \"%s\" incompatible with ext2", opt);
1661                 return -1;
1662         }
1663         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1664                 ext4_msg(sb, KERN_ERR,
1665                          "Mount option \"%s\" incompatible with ext3", opt);
1666                 return -1;
1667         }
1668
1669         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1670                 return -1;
1671         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1672                 return -1;
1673         if (m->flags & MOPT_EXPLICIT) {
1674                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1675                         set_opt2(sb, EXPLICIT_DELALLOC);
1676                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1677                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1678                 } else
1679                         return -1;
1680         }
1681         if (m->flags & MOPT_CLEAR_ERR)
1682                 clear_opt(sb, ERRORS_MASK);
1683         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1684                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1685                          "options when quota turned on");
1686                 return -1;
1687         }
1688
1689         if (m->flags & MOPT_NOSUPPORT) {
1690                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1691         } else if (token == Opt_commit) {
1692                 if (arg == 0)
1693                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1694                 sbi->s_commit_interval = HZ * arg;
1695         } else if (token == Opt_debug_want_extra_isize) {
1696                 sbi->s_want_extra_isize = arg;
1697         } else if (token == Opt_max_batch_time) {
1698                 sbi->s_max_batch_time = arg;
1699         } else if (token == Opt_min_batch_time) {
1700                 sbi->s_min_batch_time = arg;
1701         } else if (token == Opt_inode_readahead_blks) {
1702                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1703                         ext4_msg(sb, KERN_ERR,
1704                                  "EXT4-fs: inode_readahead_blks must be "
1705                                  "0 or a power of 2 smaller than 2^31");
1706                         return -1;
1707                 }
1708                 sbi->s_inode_readahead_blks = arg;
1709         } else if (token == Opt_init_itable) {
1710                 set_opt(sb, INIT_INODE_TABLE);
1711                 if (!args->from)
1712                         arg = EXT4_DEF_LI_WAIT_MULT;
1713                 sbi->s_li_wait_mult = arg;
1714         } else if (token == Opt_max_dir_size_kb) {
1715                 sbi->s_max_dir_size_kb = arg;
1716         } else if (token == Opt_stripe) {
1717                 sbi->s_stripe = arg;
1718         } else if (token == Opt_resuid) {
1719                 uid = make_kuid(current_user_ns(), arg);
1720                 if (!uid_valid(uid)) {
1721                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1722                         return -1;
1723                 }
1724                 sbi->s_resuid = uid;
1725         } else if (token == Opt_resgid) {
1726                 gid = make_kgid(current_user_ns(), arg);
1727                 if (!gid_valid(gid)) {
1728                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1729                         return -1;
1730                 }
1731                 sbi->s_resgid = gid;
1732         } else if (token == Opt_journal_dev) {
1733                 if (is_remount) {
1734                         ext4_msg(sb, KERN_ERR,
1735                                  "Cannot specify journal on remount");
1736                         return -1;
1737                 }
1738                 *journal_devnum = arg;
1739         } else if (token == Opt_journal_path) {
1740                 char *journal_path;
1741                 struct inode *journal_inode;
1742                 struct path path;
1743                 int error;
1744
1745                 if (is_remount) {
1746                         ext4_msg(sb, KERN_ERR,
1747                                  "Cannot specify journal on remount");
1748                         return -1;
1749                 }
1750                 journal_path = match_strdup(&args[0]);
1751                 if (!journal_path) {
1752                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1753                                 "journal device string");
1754                         return -1;
1755                 }
1756
1757                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1758                 if (error) {
1759                         ext4_msg(sb, KERN_ERR, "error: could not find "
1760                                 "journal device path: error %d", error);
1761                         kfree(journal_path);
1762                         return -1;
1763                 }
1764
1765                 journal_inode = d_inode(path.dentry);
1766                 if (!S_ISBLK(journal_inode->i_mode)) {
1767                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1768                                 "is not a block device", journal_path);
1769                         path_put(&path);
1770                         kfree(journal_path);
1771                         return -1;
1772                 }
1773
1774                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1775                 path_put(&path);
1776                 kfree(journal_path);
1777         } else if (token == Opt_journal_ioprio) {
1778                 if (arg > 7) {
1779                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1780                                  " (must be 0-7)");
1781                         return -1;
1782                 }
1783                 *journal_ioprio =
1784                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1785         } else if (token == Opt_test_dummy_encryption) {
1786 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1787                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1788                 ext4_msg(sb, KERN_WARNING,
1789                          "Test dummy encryption mode enabled");
1790 #else
1791                 ext4_msg(sb, KERN_WARNING,
1792                          "Test dummy encryption mount option ignored");
1793 #endif
1794         } else if (m->flags & MOPT_DATAJ) {
1795                 if (is_remount) {
1796                         if (!sbi->s_journal)
1797                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1798                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1799                                 ext4_msg(sb, KERN_ERR,
1800                                          "Cannot change data mode on remount");
1801                                 return -1;
1802                         }
1803                 } else {
1804                         clear_opt(sb, DATA_FLAGS);
1805                         sbi->s_mount_opt |= m->mount_opt;
1806                 }
1807 #ifdef CONFIG_QUOTA
1808         } else if (m->flags & MOPT_QFMT) {
1809                 if (sb_any_quota_loaded(sb) &&
1810                     sbi->s_jquota_fmt != m->mount_opt) {
1811                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1812                                  "quota options when quota turned on");
1813                         return -1;
1814                 }
1815                 if (ext4_has_feature_quota(sb)) {
1816                         ext4_msg(sb, KERN_INFO,
1817                                  "Quota format mount options ignored "
1818                                  "when QUOTA feature is enabled");
1819                         return 1;
1820                 }
1821                 sbi->s_jquota_fmt = m->mount_opt;
1822 #endif
1823         } else if (token == Opt_dax) {
1824 #ifdef CONFIG_FS_DAX
1825                 ext4_msg(sb, KERN_WARNING,
1826                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1827                         sbi->s_mount_opt |= m->mount_opt;
1828 #else
1829                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1830                 return -1;
1831 #endif
1832         } else if (token == Opt_data_err_abort) {
1833                 sbi->s_mount_opt |= m->mount_opt;
1834         } else if (token == Opt_data_err_ignore) {
1835                 sbi->s_mount_opt &= ~m->mount_opt;
1836         } else {
1837                 if (!args->from)
1838                         arg = 1;
1839                 if (m->flags & MOPT_CLEAR)
1840                         arg = !arg;
1841                 else if (unlikely(!(m->flags & MOPT_SET))) {
1842                         ext4_msg(sb, KERN_WARNING,
1843                                  "buggy handling of option %s", opt);
1844                         WARN_ON(1);
1845                         return -1;
1846                 }
1847                 if (arg != 0)
1848                         sbi->s_mount_opt |= m->mount_opt;
1849                 else
1850                         sbi->s_mount_opt &= ~m->mount_opt;
1851         }
1852         return 1;
1853 }
1854
1855 static int parse_options(char *options, struct super_block *sb,
1856                          unsigned long *journal_devnum,
1857                          unsigned int *journal_ioprio,
1858                          int is_remount)
1859 {
1860         struct ext4_sb_info *sbi = EXT4_SB(sb);
1861         char *p;
1862         substring_t args[MAX_OPT_ARGS];
1863         int token;
1864
1865         if (!options)
1866                 return 1;
1867
1868         while ((p = strsep(&options, ",")) != NULL) {
1869                 if (!*p)
1870                         continue;
1871                 /*
1872                  * Initialize args struct so we know whether arg was
1873                  * found; some options take optional arguments.
1874                  */
1875                 args[0].to = args[0].from = NULL;
1876                 token = match_token(p, tokens, args);
1877                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1878                                      journal_ioprio, is_remount) < 0)
1879                         return 0;
1880         }
1881 #ifdef CONFIG_QUOTA
1882         /*
1883          * We do the test below only for project quotas. 'usrquota' and
1884          * 'grpquota' mount options are allowed even without quota feature
1885          * to support legacy quotas in quota files.
1886          */
1887         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
1888                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
1889                          "Cannot enable project quota enforcement.");
1890                 return 0;
1891         }
1892         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1893                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1894                         clear_opt(sb, USRQUOTA);
1895
1896                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1897                         clear_opt(sb, GRPQUOTA);
1898
1899                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1900                         ext4_msg(sb, KERN_ERR, "old and new quota "
1901                                         "format mixing");
1902                         return 0;
1903                 }
1904
1905                 if (!sbi->s_jquota_fmt) {
1906                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1907                                         "not specified");
1908                         return 0;
1909                 }
1910         }
1911 #endif
1912         if (test_opt(sb, DIOREAD_NOLOCK)) {
1913                 int blocksize =
1914                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1915
1916                 if (blocksize < PAGE_SIZE) {
1917                         ext4_msg(sb, KERN_ERR, "can't mount with "
1918                                  "dioread_nolock if block size != PAGE_SIZE");
1919                         return 0;
1920                 }
1921         }
1922         return 1;
1923 }
1924
1925 static inline void ext4_show_quota_options(struct seq_file *seq,
1926                                            struct super_block *sb)
1927 {
1928 #if defined(CONFIG_QUOTA)
1929         struct ext4_sb_info *sbi = EXT4_SB(sb);
1930
1931         if (sbi->s_jquota_fmt) {
1932                 char *fmtname = "";
1933
1934                 switch (sbi->s_jquota_fmt) {
1935                 case QFMT_VFS_OLD:
1936                         fmtname = "vfsold";
1937                         break;
1938                 case QFMT_VFS_V0:
1939                         fmtname = "vfsv0";
1940                         break;
1941                 case QFMT_VFS_V1:
1942                         fmtname = "vfsv1";
1943                         break;
1944                 }
1945                 seq_printf(seq, ",jqfmt=%s", fmtname);
1946         }
1947
1948         if (sbi->s_qf_names[USRQUOTA])
1949                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1950
1951         if (sbi->s_qf_names[GRPQUOTA])
1952                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1953 #endif
1954 }
1955
1956 static const char *token2str(int token)
1957 {
1958         const struct match_token *t;
1959
1960         for (t = tokens; t->token != Opt_err; t++)
1961                 if (t->token == token && !strchr(t->pattern, '='))
1962                         break;
1963         return t->pattern;
1964 }
1965
1966 /*
1967  * Show an option if
1968  *  - it's set to a non-default value OR
1969  *  - if the per-sb default is different from the global default
1970  */
1971 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1972                               int nodefs)
1973 {
1974         struct ext4_sb_info *sbi = EXT4_SB(sb);
1975         struct ext4_super_block *es = sbi->s_es;
1976         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1977         const struct mount_opts *m;
1978         char sep = nodefs ? '\n' : ',';
1979
1980 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1981 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1982
1983         if (sbi->s_sb_block != 1)
1984                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1985
1986         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1987                 int want_set = m->flags & MOPT_SET;
1988                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1989                     (m->flags & MOPT_CLEAR_ERR))
1990                         continue;
1991                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1992                         continue; /* skip if same as the default */
1993                 if ((want_set &&
1994                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1995                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1996                         continue; /* select Opt_noFoo vs Opt_Foo */
1997                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1998         }
1999
2000         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2001             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2002                 SEQ_OPTS_PRINT("resuid=%u",
2003                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2004         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2005             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2006                 SEQ_OPTS_PRINT("resgid=%u",
2007                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2008         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2009         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2010                 SEQ_OPTS_PUTS("errors=remount-ro");
2011         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2012                 SEQ_OPTS_PUTS("errors=continue");
2013         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2014                 SEQ_OPTS_PUTS("errors=panic");
2015         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2016                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2017         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2018                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2019         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2020                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2021         if (sb->s_flags & MS_I_VERSION)
2022                 SEQ_OPTS_PUTS("i_version");
2023         if (nodefs || sbi->s_stripe)
2024                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2025         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
2026                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2027                         SEQ_OPTS_PUTS("data=journal");
2028                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2029                         SEQ_OPTS_PUTS("data=ordered");
2030                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2031                         SEQ_OPTS_PUTS("data=writeback");
2032         }
2033         if (nodefs ||
2034             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2035                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2036                                sbi->s_inode_readahead_blks);
2037
2038         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
2039                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2040                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2041         if (nodefs || sbi->s_max_dir_size_kb)
2042                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2043         if (test_opt(sb, DATA_ERR_ABORT))
2044                 SEQ_OPTS_PUTS("data_err=abort");
2045
2046         ext4_show_quota_options(seq, sb);
2047         return 0;
2048 }
2049
2050 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2051 {
2052         return _ext4_show_options(seq, root->d_sb, 0);
2053 }
2054
2055 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2056 {
2057         struct super_block *sb = seq->private;
2058         int rc;
2059
2060         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
2061         rc = _ext4_show_options(seq, sb, 1);
2062         seq_puts(seq, "\n");
2063         return rc;
2064 }
2065
2066 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2067                             int read_only)
2068 {
2069         struct ext4_sb_info *sbi = EXT4_SB(sb);
2070         int res = 0;
2071
2072         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2073                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2074                          "forcing read-only mode");
2075                 res = MS_RDONLY;
2076         }
2077         if (read_only)
2078                 goto done;
2079         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2080                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2081                          "running e2fsck is recommended");
2082         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2083                 ext4_msg(sb, KERN_WARNING,
2084                          "warning: mounting fs with errors, "
2085                          "running e2fsck is recommended");
2086         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2087                  le16_to_cpu(es->s_mnt_count) >=
2088                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2089                 ext4_msg(sb, KERN_WARNING,
2090                          "warning: maximal mount count reached, "
2091                          "running e2fsck is recommended");
2092         else if (le32_to_cpu(es->s_checkinterval) &&
2093                 (le32_to_cpu(es->s_lastcheck) +
2094                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
2095                 ext4_msg(sb, KERN_WARNING,
2096                          "warning: checktime reached, "
2097                          "running e2fsck is recommended");
2098         if (!sbi->s_journal)
2099                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2100         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2101                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2102         le16_add_cpu(&es->s_mnt_count, 1);
2103         es->s_mtime = cpu_to_le32(get_seconds());
2104         ext4_update_dynamic_rev(sb);
2105         if (sbi->s_journal)
2106                 ext4_set_feature_journal_needs_recovery(sb);
2107
2108         ext4_commit_super(sb, 1);
2109 done:
2110         if (test_opt(sb, DEBUG))
2111                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2112                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2113                         sb->s_blocksize,
2114                         sbi->s_groups_count,
2115                         EXT4_BLOCKS_PER_GROUP(sb),
2116                         EXT4_INODES_PER_GROUP(sb),
2117                         sbi->s_mount_opt, sbi->s_mount_opt2);
2118
2119         cleancache_init_fs(sb);
2120         return res;
2121 }
2122
2123 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2124 {
2125         struct ext4_sb_info *sbi = EXT4_SB(sb);
2126         struct flex_groups *new_groups;
2127         int size;
2128
2129         if (!sbi->s_log_groups_per_flex)
2130                 return 0;
2131
2132         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2133         if (size <= sbi->s_flex_groups_allocated)
2134                 return 0;
2135
2136         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2137         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
2138         if (!new_groups) {
2139                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2140                          size / (int) sizeof(struct flex_groups));
2141                 return -ENOMEM;
2142         }
2143
2144         if (sbi->s_flex_groups) {
2145                 memcpy(new_groups, sbi->s_flex_groups,
2146                        (sbi->s_flex_groups_allocated *
2147                         sizeof(struct flex_groups)));
2148                 kvfree(sbi->s_flex_groups);
2149         }
2150         sbi->s_flex_groups = new_groups;
2151         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2152         return 0;
2153 }
2154
2155 static int ext4_fill_flex_info(struct super_block *sb)
2156 {
2157         struct ext4_sb_info *sbi = EXT4_SB(sb);
2158         struct ext4_group_desc *gdp = NULL;
2159         ext4_group_t flex_group;
2160         int i, err;
2161
2162         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2163         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2164                 sbi->s_log_groups_per_flex = 0;
2165                 return 1;
2166         }
2167
2168         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2169         if (err)
2170                 goto failed;
2171
2172         for (i = 0; i < sbi->s_groups_count; i++) {
2173                 gdp = ext4_get_group_desc(sb, i, NULL);
2174
2175                 flex_group = ext4_flex_group(sbi, i);
2176                 atomic_add(ext4_free_inodes_count(sb, gdp),
2177                            &sbi->s_flex_groups[flex_group].free_inodes);
2178                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2179                              &sbi->s_flex_groups[flex_group].free_clusters);
2180                 atomic_add(ext4_used_dirs_count(sb, gdp),
2181                            &sbi->s_flex_groups[flex_group].used_dirs);
2182         }
2183
2184         return 1;
2185 failed:
2186         return 0;
2187 }
2188
2189 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2190                                    struct ext4_group_desc *gdp)
2191 {
2192         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2193         __u16 crc = 0;
2194         __le32 le_group = cpu_to_le32(block_group);
2195         struct ext4_sb_info *sbi = EXT4_SB(sb);
2196
2197         if (ext4_has_metadata_csum(sbi->s_sb)) {
2198                 /* Use new metadata_csum algorithm */
2199                 __u32 csum32;
2200                 __u16 dummy_csum = 0;
2201
2202                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2203                                      sizeof(le_group));
2204                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2205                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2206                                      sizeof(dummy_csum));
2207                 offset += sizeof(dummy_csum);
2208                 if (offset < sbi->s_desc_size)
2209                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2210                                              sbi->s_desc_size - offset);
2211
2212                 crc = csum32 & 0xFFFF;
2213                 goto out;
2214         }
2215
2216         /* old crc16 code */
2217         if (!ext4_has_feature_gdt_csum(sb))
2218                 return 0;
2219
2220         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2221         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2222         crc = crc16(crc, (__u8 *)gdp, offset);
2223         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2224         /* for checksum of struct ext4_group_desc do the rest...*/
2225         if (ext4_has_feature_64bit(sb) &&
2226             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2227                 crc = crc16(crc, (__u8 *)gdp + offset,
2228                             le16_to_cpu(sbi->s_es->s_desc_size) -
2229                                 offset);
2230
2231 out:
2232         return cpu_to_le16(crc);
2233 }
2234
2235 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2236                                 struct ext4_group_desc *gdp)
2237 {
2238         if (ext4_has_group_desc_csum(sb) &&
2239             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2240                 return 0;
2241
2242         return 1;
2243 }
2244
2245 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2246                               struct ext4_group_desc *gdp)
2247 {
2248         if (!ext4_has_group_desc_csum(sb))
2249                 return;
2250         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2251 }
2252
2253 /* Called at mount-time, super-block is locked */
2254 static int ext4_check_descriptors(struct super_block *sb,
2255                                   ext4_fsblk_t sb_block,
2256                                   ext4_group_t *first_not_zeroed)
2257 {
2258         struct ext4_sb_info *sbi = EXT4_SB(sb);
2259         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2260         ext4_fsblk_t last_block;
2261         ext4_fsblk_t block_bitmap;
2262         ext4_fsblk_t inode_bitmap;
2263         ext4_fsblk_t inode_table;
2264         int flexbg_flag = 0;
2265         ext4_group_t i, grp = sbi->s_groups_count;
2266
2267         if (ext4_has_feature_flex_bg(sb))
2268                 flexbg_flag = 1;
2269
2270         ext4_debug("Checking group descriptors");
2271
2272         for (i = 0; i < sbi->s_groups_count; i++) {
2273                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2274
2275                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2276                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2277                 else
2278                         last_block = first_block +
2279                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2280
2281                 if ((grp == sbi->s_groups_count) &&
2282                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2283                         grp = i;
2284
2285                 block_bitmap = ext4_block_bitmap(sb, gdp);
2286                 if (block_bitmap == sb_block) {
2287                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2288                                  "Block bitmap for group %u overlaps "
2289                                  "superblock", i);
2290                 }
2291                 if (block_bitmap < first_block || block_bitmap > last_block) {
2292                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2293                                "Block bitmap for group %u not in group "
2294                                "(block %llu)!", i, block_bitmap);
2295                         return 0;
2296                 }
2297                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2298                 if (inode_bitmap == sb_block) {
2299                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2300                                  "Inode bitmap for group %u overlaps "
2301                                  "superblock", i);
2302                 }
2303                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2304                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2305                                "Inode bitmap for group %u not in group "
2306                                "(block %llu)!", i, inode_bitmap);
2307                         return 0;
2308                 }
2309                 inode_table = ext4_inode_table(sb, gdp);
2310                 if (inode_table == sb_block) {
2311                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2312                                  "Inode table for group %u overlaps "
2313                                  "superblock", i);
2314                 }
2315                 if (inode_table < first_block ||
2316                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2317                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2318                                "Inode table for group %u not in group "
2319                                "(block %llu)!", i, inode_table);
2320                         return 0;
2321                 }
2322                 ext4_lock_group(sb, i);
2323                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2324                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2325                                  "Checksum for group %u failed (%u!=%u)",
2326                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2327                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2328                         if (!(sb->s_flags & MS_RDONLY)) {
2329                                 ext4_unlock_group(sb, i);
2330                                 return 0;
2331                         }
2332                 }
2333                 ext4_unlock_group(sb, i);
2334                 if (!flexbg_flag)
2335                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2336         }
2337         if (NULL != first_not_zeroed)
2338                 *first_not_zeroed = grp;
2339         return 1;
2340 }
2341
2342 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2343  * the superblock) which were deleted from all directories, but held open by
2344  * a process at the time of a crash.  We walk the list and try to delete these
2345  * inodes at recovery time (only with a read-write filesystem).
2346  *
2347  * In order to keep the orphan inode chain consistent during traversal (in
2348  * case of crash during recovery), we link each inode into the superblock
2349  * orphan list_head and handle it the same way as an inode deletion during
2350  * normal operation (which journals the operations for us).
2351  *
2352  * We only do an iget() and an iput() on each inode, which is very safe if we
2353  * accidentally point at an in-use or already deleted inode.  The worst that
2354  * can happen in this case is that we get a "bit already cleared" message from
2355  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2356  * e2fsck was run on this filesystem, and it must have already done the orphan
2357  * inode cleanup for us, so we can safely abort without any further action.
2358  */
2359 static void ext4_orphan_cleanup(struct super_block *sb,
2360                                 struct ext4_super_block *es)
2361 {
2362         unsigned int s_flags = sb->s_flags;
2363         int ret, nr_orphans = 0, nr_truncates = 0;
2364 #ifdef CONFIG_QUOTA
2365         int i;
2366 #endif
2367         if (!es->s_last_orphan) {
2368                 jbd_debug(4, "no orphan inodes to clean up\n");
2369                 return;
2370         }
2371
2372         if (bdev_read_only(sb->s_bdev)) {
2373                 ext4_msg(sb, KERN_ERR, "write access "
2374                         "unavailable, skipping orphan cleanup");
2375                 return;
2376         }
2377
2378         /* Check if feature set would not allow a r/w mount */
2379         if (!ext4_feature_set_ok(sb, 0)) {
2380                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2381                          "unknown ROCOMPAT features");
2382                 return;
2383         }
2384
2385         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2386                 /* don't clear list on RO mount w/ errors */
2387                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2388                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2389                                   "clearing orphan list.\n");
2390                         es->s_last_orphan = 0;
2391                 }
2392                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2393                 return;
2394         }
2395
2396         if (s_flags & MS_RDONLY) {
2397                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2398                 sb->s_flags &= ~MS_RDONLY;
2399         }
2400 #ifdef CONFIG_QUOTA
2401         /* Needed for iput() to work correctly and not trash data */
2402         sb->s_flags |= MS_ACTIVE;
2403         /* Turn on quotas so that they are updated correctly */
2404         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2405                 if (EXT4_SB(sb)->s_qf_names[i]) {
2406                         int ret = ext4_quota_on_mount(sb, i);
2407                         if (ret < 0)
2408                                 ext4_msg(sb, KERN_ERR,
2409                                         "Cannot turn on journaled "
2410                                         "quota: error %d", ret);
2411                 }
2412         }
2413 #endif
2414
2415         while (es->s_last_orphan) {
2416                 struct inode *inode;
2417
2418                 /*
2419                  * We may have encountered an error during cleanup; if
2420                  * so, skip the rest.
2421                  */
2422                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2423                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2424                         es->s_last_orphan = 0;
2425                         break;
2426                 }
2427
2428                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2429                 if (IS_ERR(inode)) {
2430                         es->s_last_orphan = 0;
2431                         break;
2432                 }
2433
2434                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2435                 dquot_initialize(inode);
2436                 if (inode->i_nlink) {
2437                         if (test_opt(sb, DEBUG))
2438                                 ext4_msg(sb, KERN_DEBUG,
2439                                         "%s: truncating inode %lu to %lld bytes",
2440                                         __func__, inode->i_ino, inode->i_size);
2441                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2442                                   inode->i_ino, inode->i_size);
2443                         inode_lock(inode);
2444                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2445                         ret = ext4_truncate(inode);
2446                         if (ret)
2447                                 ext4_std_error(inode->i_sb, ret);
2448                         inode_unlock(inode);
2449                         nr_truncates++;
2450                 } else {
2451                         if (test_opt(sb, DEBUG))
2452                                 ext4_msg(sb, KERN_DEBUG,
2453                                         "%s: deleting unreferenced inode %lu",
2454                                         __func__, inode->i_ino);
2455                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2456                                   inode->i_ino);
2457                         nr_orphans++;
2458                 }
2459                 iput(inode);  /* The delete magic happens here! */
2460         }
2461
2462 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2463
2464         if (nr_orphans)
2465                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2466                        PLURAL(nr_orphans));
2467         if (nr_truncates)
2468                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2469                        PLURAL(nr_truncates));
2470 #ifdef CONFIG_QUOTA
2471         /* Turn quotas off */
2472         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2473                 if (sb_dqopt(sb)->files[i])
2474                         dquot_quota_off(sb, i);
2475         }
2476 #endif
2477         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2478 }
2479
2480 /*
2481  * Maximal extent format file size.
2482  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2483  * extent format containers, within a sector_t, and within i_blocks
2484  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2485  * so that won't be a limiting factor.
2486  *
2487  * However there is other limiting factor. We do store extents in the form
2488  * of starting block and length, hence the resulting length of the extent
2489  * covering maximum file size must fit into on-disk format containers as
2490  * well. Given that length is always by 1 unit bigger than max unit (because
2491  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2492  *
2493  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2494  */
2495 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2496 {
2497         loff_t res;
2498         loff_t upper_limit = MAX_LFS_FILESIZE;
2499
2500         /* small i_blocks in vfs inode? */
2501         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2502                 /*
2503                  * CONFIG_LBDAF is not enabled implies the inode
2504                  * i_block represent total blocks in 512 bytes
2505                  * 32 == size of vfs inode i_blocks * 8
2506                  */
2507                 upper_limit = (1LL << 32) - 1;
2508
2509                 /* total blocks in file system block size */
2510                 upper_limit >>= (blkbits - 9);
2511                 upper_limit <<= blkbits;
2512         }
2513
2514         /*
2515          * 32-bit extent-start container, ee_block. We lower the maxbytes
2516          * by one fs block, so ee_len can cover the extent of maximum file
2517          * size
2518          */
2519         res = (1LL << 32) - 1;
2520         res <<= blkbits;
2521
2522         /* Sanity check against vm- & vfs- imposed limits */
2523         if (res > upper_limit)
2524                 res = upper_limit;
2525
2526         return res;
2527 }
2528
2529 /*
2530  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2531  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2532  * We need to be 1 filesystem block less than the 2^48 sector limit.
2533  */
2534 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2535 {
2536         loff_t res = EXT4_NDIR_BLOCKS;
2537         int meta_blocks;
2538         loff_t upper_limit;
2539         /* This is calculated to be the largest file size for a dense, block
2540          * mapped file such that the file's total number of 512-byte sectors,
2541          * including data and all indirect blocks, does not exceed (2^48 - 1).
2542          *
2543          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2544          * number of 512-byte sectors of the file.
2545          */
2546
2547         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2548                 /*
2549                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2550                  * the inode i_block field represents total file blocks in
2551                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2552                  */
2553                 upper_limit = (1LL << 32) - 1;
2554
2555                 /* total blocks in file system block size */
2556                 upper_limit >>= (bits - 9);
2557
2558         } else {
2559                 /*
2560                  * We use 48 bit ext4_inode i_blocks
2561                  * With EXT4_HUGE_FILE_FL set the i_blocks
2562                  * represent total number of blocks in
2563                  * file system block size
2564                  */
2565                 upper_limit = (1LL << 48) - 1;
2566
2567         }
2568
2569         /* indirect blocks */
2570         meta_blocks = 1;
2571         /* double indirect blocks */
2572         meta_blocks += 1 + (1LL << (bits-2));
2573         /* tripple indirect blocks */
2574         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2575
2576         upper_limit -= meta_blocks;
2577         upper_limit <<= bits;
2578
2579         res += 1LL << (bits-2);
2580         res += 1LL << (2*(bits-2));
2581         res += 1LL << (3*(bits-2));
2582         res <<= bits;
2583         if (res > upper_limit)
2584                 res = upper_limit;
2585
2586         if (res > MAX_LFS_FILESIZE)
2587                 res = MAX_LFS_FILESIZE;
2588
2589         return res;
2590 }
2591
2592 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2593                                    ext4_fsblk_t logical_sb_block, int nr)
2594 {
2595         struct ext4_sb_info *sbi = EXT4_SB(sb);
2596         ext4_group_t bg, first_meta_bg;
2597         int has_super = 0;
2598
2599         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2600
2601         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2602                 return logical_sb_block + nr + 1;
2603         bg = sbi->s_desc_per_block * nr;
2604         if (ext4_bg_has_super(sb, bg))
2605                 has_super = 1;
2606
2607         /*
2608          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2609          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2610          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2611          * compensate.
2612          */
2613         if (sb->s_blocksize == 1024 && nr == 0 &&
2614             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2615                 has_super++;
2616
2617         return (has_super + ext4_group_first_block_no(sb, bg));
2618 }
2619
2620 /**
2621  * ext4_get_stripe_size: Get the stripe size.
2622  * @sbi: In memory super block info
2623  *
2624  * If we have specified it via mount option, then
2625  * use the mount option value. If the value specified at mount time is
2626  * greater than the blocks per group use the super block value.
2627  * If the super block value is greater than blocks per group return 0.
2628  * Allocator needs it be less than blocks per group.
2629  *
2630  */
2631 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2632 {
2633         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2634         unsigned long stripe_width =
2635                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2636         int ret;
2637
2638         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2639                 ret = sbi->s_stripe;
2640         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2641                 ret = stripe_width;
2642         else if (stride && stride <= sbi->s_blocks_per_group)
2643                 ret = stride;
2644         else
2645                 ret = 0;
2646
2647         /*
2648          * If the stripe width is 1, this makes no sense and
2649          * we set it to 0 to turn off stripe handling code.
2650          */
2651         if (ret <= 1)
2652                 ret = 0;
2653
2654         return ret;
2655 }
2656
2657 /*
2658  * Check whether this filesystem can be mounted based on
2659  * the features present and the RDONLY/RDWR mount requested.
2660  * Returns 1 if this filesystem can be mounted as requested,
2661  * 0 if it cannot be.
2662  */
2663 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2664 {
2665         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2666                 ext4_msg(sb, KERN_ERR,
2667                         "Couldn't mount because of "
2668                         "unsupported optional features (%x)",
2669                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2670                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2671                 return 0;
2672         }
2673
2674         if (readonly)
2675                 return 1;
2676
2677         if (ext4_has_feature_readonly(sb)) {
2678                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2679                 sb->s_flags |= MS_RDONLY;
2680                 return 1;
2681         }
2682
2683         /* Check that feature set is OK for a read-write mount */
2684         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2685                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2686                          "unsupported optional features (%x)",
2687                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2688                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2689                 return 0;
2690         }
2691         /*
2692          * Large file size enabled file system can only be mounted
2693          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2694          */
2695         if (ext4_has_feature_huge_file(sb)) {
2696                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2697                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2698                                  "cannot be mounted RDWR without "
2699                                  "CONFIG_LBDAF");
2700                         return 0;
2701                 }
2702         }
2703         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2704                 ext4_msg(sb, KERN_ERR,
2705                          "Can't support bigalloc feature without "
2706                          "extents feature\n");
2707                 return 0;
2708         }
2709
2710 #ifndef CONFIG_QUOTA
2711         if (ext4_has_feature_quota(sb) && !readonly) {
2712                 ext4_msg(sb, KERN_ERR,
2713                          "Filesystem with quota feature cannot be mounted RDWR "
2714                          "without CONFIG_QUOTA");
2715                 return 0;
2716         }
2717         if (ext4_has_feature_project(sb) && !readonly) {
2718                 ext4_msg(sb, KERN_ERR,
2719                          "Filesystem with project quota feature cannot be mounted RDWR "
2720                          "without CONFIG_QUOTA");
2721                 return 0;
2722         }
2723 #endif  /* CONFIG_QUOTA */
2724         return 1;
2725 }
2726
2727 /*
2728  * This function is called once a day if we have errors logged
2729  * on the file system
2730  */
2731 static void print_daily_error_info(unsigned long arg)
2732 {
2733         struct super_block *sb = (struct super_block *) arg;
2734         struct ext4_sb_info *sbi;
2735         struct ext4_super_block *es;
2736
2737         sbi = EXT4_SB(sb);
2738         es = sbi->s_es;
2739
2740         if (es->s_error_count)
2741                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2742                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2743                          le32_to_cpu(es->s_error_count));
2744         if (es->s_first_error_time) {
2745                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2746                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2747                        (int) sizeof(es->s_first_error_func),
2748                        es->s_first_error_func,
2749                        le32_to_cpu(es->s_first_error_line));
2750                 if (es->s_first_error_ino)
2751                         printk(KERN_CONT ": inode %u",
2752                                le32_to_cpu(es->s_first_error_ino));
2753                 if (es->s_first_error_block)
2754                         printk(KERN_CONT ": block %llu", (unsigned long long)
2755                                le64_to_cpu(es->s_first_error_block));
2756                 printk(KERN_CONT "\n");
2757         }
2758         if (es->s_last_error_time) {
2759                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2760                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2761                        (int) sizeof(es->s_last_error_func),
2762                        es->s_last_error_func,
2763                        le32_to_cpu(es->s_last_error_line));
2764                 if (es->s_last_error_ino)
2765                         printk(KERN_CONT ": inode %u",
2766                                le32_to_cpu(es->s_last_error_ino));
2767                 if (es->s_last_error_block)
2768                         printk(KERN_CONT ": block %llu", (unsigned long long)
2769                                le64_to_cpu(es->s_last_error_block));
2770                 printk(KERN_CONT "\n");
2771         }
2772         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2773 }
2774
2775 /* Find next suitable group and run ext4_init_inode_table */
2776 static int ext4_run_li_request(struct ext4_li_request *elr)
2777 {
2778         struct ext4_group_desc *gdp = NULL;
2779         ext4_group_t group, ngroups;
2780         struct super_block *sb;
2781         unsigned long timeout = 0;
2782         int ret = 0;
2783
2784         sb = elr->lr_super;
2785         ngroups = EXT4_SB(sb)->s_groups_count;
2786
2787         for (group = elr->lr_next_group; group < ngroups; group++) {
2788                 gdp = ext4_get_group_desc(sb, group, NULL);
2789                 if (!gdp) {
2790                         ret = 1;
2791                         break;
2792                 }
2793
2794                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2795                         break;
2796         }
2797
2798         if (group >= ngroups)
2799                 ret = 1;
2800
2801         if (!ret) {
2802                 timeout = jiffies;
2803                 ret = ext4_init_inode_table(sb, group,
2804                                             elr->lr_timeout ? 0 : 1);
2805                 if (elr->lr_timeout == 0) {
2806                         timeout = (jiffies - timeout) *
2807                                   elr->lr_sbi->s_li_wait_mult;
2808                         elr->lr_timeout = timeout;
2809                 }
2810                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2811                 elr->lr_next_group = group + 1;
2812         }
2813         return ret;
2814 }
2815
2816 /*
2817  * Remove lr_request from the list_request and free the
2818  * request structure. Should be called with li_list_mtx held
2819  */
2820 static void ext4_remove_li_request(struct ext4_li_request *elr)
2821 {
2822         struct ext4_sb_info *sbi;
2823
2824         if (!elr)
2825                 return;
2826
2827         sbi = elr->lr_sbi;
2828
2829         list_del(&elr->lr_request);
2830         sbi->s_li_request = NULL;
2831         kfree(elr);
2832 }
2833
2834 static void ext4_unregister_li_request(struct super_block *sb)
2835 {
2836         mutex_lock(&ext4_li_mtx);
2837         if (!ext4_li_info) {
2838                 mutex_unlock(&ext4_li_mtx);
2839                 return;
2840         }
2841
2842         mutex_lock(&ext4_li_info->li_list_mtx);
2843         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2844         mutex_unlock(&ext4_li_info->li_list_mtx);
2845         mutex_unlock(&ext4_li_mtx);
2846 }
2847
2848 static struct task_struct *ext4_lazyinit_task;
2849
2850 /*
2851  * This is the function where ext4lazyinit thread lives. It walks
2852  * through the request list searching for next scheduled filesystem.
2853  * When such a fs is found, run the lazy initialization request
2854  * (ext4_rn_li_request) and keep track of the time spend in this
2855  * function. Based on that time we compute next schedule time of
2856  * the request. When walking through the list is complete, compute
2857  * next waking time and put itself into sleep.
2858  */
2859 static int ext4_lazyinit_thread(void *arg)
2860 {
2861         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2862         struct list_head *pos, *n;
2863         struct ext4_li_request *elr;
2864         unsigned long next_wakeup, cur;
2865
2866         BUG_ON(NULL == eli);
2867
2868 cont_thread:
2869         while (true) {
2870                 next_wakeup = MAX_JIFFY_OFFSET;
2871
2872                 mutex_lock(&eli->li_list_mtx);
2873                 if (list_empty(&eli->li_request_list)) {
2874                         mutex_unlock(&eli->li_list_mtx);
2875                         goto exit_thread;
2876                 }
2877                 list_for_each_safe(pos, n, &eli->li_request_list) {
2878                         int err = 0;
2879                         int progress = 0;
2880                         elr = list_entry(pos, struct ext4_li_request,
2881                                          lr_request);
2882
2883                         if (time_before(jiffies, elr->lr_next_sched)) {
2884                                 if (time_before(elr->lr_next_sched, next_wakeup))
2885                                         next_wakeup = elr->lr_next_sched;
2886                                 continue;
2887                         }
2888                         if (down_read_trylock(&elr->lr_super->s_umount)) {
2889                                 if (sb_start_write_trylock(elr->lr_super)) {
2890                                         progress = 1;
2891                                         /*
2892                                          * We hold sb->s_umount, sb can not
2893                                          * be removed from the list, it is
2894                                          * now safe to drop li_list_mtx
2895                                          */
2896                                         mutex_unlock(&eli->li_list_mtx);
2897                                         err = ext4_run_li_request(elr);
2898                                         sb_end_write(elr->lr_super);
2899                                         mutex_lock(&eli->li_list_mtx);
2900                                         n = pos->next;
2901                                 }
2902                                 up_read((&elr->lr_super->s_umount));
2903                         }
2904                         /* error, remove the lazy_init job */
2905                         if (err) {
2906                                 ext4_remove_li_request(elr);
2907                                 continue;
2908                         }
2909                         if (!progress) {
2910                                 elr->lr_next_sched = jiffies +
2911                                         (prandom_u32()
2912                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2913                         }
2914                         if (time_before(elr->lr_next_sched, next_wakeup))
2915                                 next_wakeup = elr->lr_next_sched;
2916                 }
2917                 mutex_unlock(&eli->li_list_mtx);
2918
2919                 try_to_freeze();
2920
2921                 cur = jiffies;
2922                 if ((time_after_eq(cur, next_wakeup)) ||
2923                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2924                         cond_resched();
2925                         continue;
2926                 }
2927
2928                 schedule_timeout_interruptible(next_wakeup - cur);
2929
2930                 if (kthread_should_stop()) {
2931                         ext4_clear_request_list();
2932                         goto exit_thread;
2933                 }
2934         }
2935
2936 exit_thread:
2937         /*
2938          * It looks like the request list is empty, but we need
2939          * to check it under the li_list_mtx lock, to prevent any
2940          * additions into it, and of course we should lock ext4_li_mtx
2941          * to atomically free the list and ext4_li_info, because at
2942          * this point another ext4 filesystem could be registering
2943          * new one.
2944          */
2945         mutex_lock(&ext4_li_mtx);
2946         mutex_lock(&eli->li_list_mtx);
2947         if (!list_empty(&eli->li_request_list)) {
2948                 mutex_unlock(&eli->li_list_mtx);
2949                 mutex_unlock(&ext4_li_mtx);
2950                 goto cont_thread;
2951         }
2952         mutex_unlock(&eli->li_list_mtx);
2953         kfree(ext4_li_info);
2954         ext4_li_info = NULL;
2955         mutex_unlock(&ext4_li_mtx);
2956
2957         return 0;
2958 }
2959
2960 static void ext4_clear_request_list(void)
2961 {
2962         struct list_head *pos, *n;
2963         struct ext4_li_request *elr;
2964
2965         mutex_lock(&ext4_li_info->li_list_mtx);
2966         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2967                 elr = list_entry(pos, struct ext4_li_request,
2968                                  lr_request);
2969                 ext4_remove_li_request(elr);
2970         }
2971         mutex_unlock(&ext4_li_info->li_list_mtx);
2972 }
2973
2974 static int ext4_run_lazyinit_thread(void)
2975 {
2976         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2977                                          ext4_li_info, "ext4lazyinit");
2978         if (IS_ERR(ext4_lazyinit_task)) {
2979                 int err = PTR_ERR(ext4_lazyinit_task);
2980                 ext4_clear_request_list();
2981                 kfree(ext4_li_info);
2982                 ext4_li_info = NULL;
2983                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2984                                  "initialization thread\n",
2985                                  err);
2986                 return err;
2987         }
2988         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2989         return 0;
2990 }
2991
2992 /*
2993  * Check whether it make sense to run itable init. thread or not.
2994  * If there is at least one uninitialized inode table, return
2995  * corresponding group number, else the loop goes through all
2996  * groups and return total number of groups.
2997  */
2998 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2999 {
3000         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3001         struct ext4_group_desc *gdp = NULL;
3002
3003         for (group = 0; group < ngroups; group++) {
3004                 gdp = ext4_get_group_desc(sb, group, NULL);
3005                 if (!gdp)
3006                         continue;
3007
3008                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3009                         break;
3010         }
3011
3012         return group;
3013 }
3014
3015 static int ext4_li_info_new(void)
3016 {
3017         struct ext4_lazy_init *eli = NULL;
3018
3019         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3020         if (!eli)
3021                 return -ENOMEM;
3022
3023         INIT_LIST_HEAD(&eli->li_request_list);
3024         mutex_init(&eli->li_list_mtx);
3025
3026         eli->li_state |= EXT4_LAZYINIT_QUIT;
3027
3028         ext4_li_info = eli;
3029
3030         return 0;
3031 }
3032
3033 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3034                                             ext4_group_t start)
3035 {
3036         struct ext4_sb_info *sbi = EXT4_SB(sb);
3037         struct ext4_li_request *elr;
3038
3039         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3040         if (!elr)
3041                 return NULL;
3042
3043         elr->lr_super = sb;
3044         elr->lr_sbi = sbi;
3045         elr->lr_next_group = start;
3046
3047         /*
3048          * Randomize first schedule time of the request to
3049          * spread the inode table initialization requests
3050          * better.
3051          */
3052         elr->lr_next_sched = jiffies + (prandom_u32() %
3053                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3054         return elr;
3055 }
3056
3057 int ext4_register_li_request(struct super_block *sb,
3058                              ext4_group_t first_not_zeroed)
3059 {
3060         struct ext4_sb_info *sbi = EXT4_SB(sb);
3061         struct ext4_li_request *elr = NULL;
3062         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3063         int ret = 0;
3064
3065         mutex_lock(&ext4_li_mtx);
3066         if (sbi->s_li_request != NULL) {
3067                 /*
3068                  * Reset timeout so it can be computed again, because
3069                  * s_li_wait_mult might have changed.
3070                  */
3071                 sbi->s_li_request->lr_timeout = 0;
3072                 goto out;
3073         }
3074
3075         if (first_not_zeroed == ngroups ||
3076             (sb->s_flags & MS_RDONLY) ||
3077             !test_opt(sb, INIT_INODE_TABLE))
3078                 goto out;
3079
3080         elr = ext4_li_request_new(sb, first_not_zeroed);
3081         if (!elr) {
3082                 ret = -ENOMEM;
3083                 goto out;
3084         }
3085
3086         if (NULL == ext4_li_info) {
3087                 ret = ext4_li_info_new();
3088                 if (ret)
3089                         goto out;
3090         }
3091
3092         mutex_lock(&ext4_li_info->li_list_mtx);
3093         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3094         mutex_unlock(&ext4_li_info->li_list_mtx);
3095
3096         sbi->s_li_request = elr;
3097         /*
3098          * set elr to NULL here since it has been inserted to
3099          * the request_list and the removal and free of it is
3100          * handled by ext4_clear_request_list from now on.
3101          */
3102         elr = NULL;
3103
3104         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3105                 ret = ext4_run_lazyinit_thread();
3106                 if (ret)
3107                         goto out;
3108         }
3109 out:
3110         mutex_unlock(&ext4_li_mtx);
3111         if (ret)
3112                 kfree(elr);
3113         return ret;
3114 }
3115
3116 /*
3117  * We do not need to lock anything since this is called on
3118  * module unload.
3119  */
3120 static void ext4_destroy_lazyinit_thread(void)
3121 {
3122         /*
3123          * If thread exited earlier
3124          * there's nothing to be done.
3125          */
3126         if (!ext4_li_info || !ext4_lazyinit_task)
3127                 return;
3128
3129         kthread_stop(ext4_lazyinit_task);
3130 }
3131
3132 static int set_journal_csum_feature_set(struct super_block *sb)
3133 {
3134         int ret = 1;
3135         int compat, incompat;
3136         struct ext4_sb_info *sbi = EXT4_SB(sb);
3137
3138         if (ext4_has_metadata_csum(sb)) {
3139                 /* journal checksum v3 */
3140                 compat = 0;
3141                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3142         } else {
3143                 /* journal checksum v1 */
3144                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3145                 incompat = 0;
3146         }
3147
3148         jbd2_journal_clear_features(sbi->s_journal,
3149                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3150                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3151                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3152         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3153                 ret = jbd2_journal_set_features(sbi->s_journal,
3154                                 compat, 0,
3155                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3156                                 incompat);
3157         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3158                 ret = jbd2_journal_set_features(sbi->s_journal,
3159                                 compat, 0,
3160                                 incompat);
3161                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3162                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3163         } else {
3164                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3165                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3166         }
3167
3168         return ret;
3169 }
3170
3171 /*
3172  * Note: calculating the overhead so we can be compatible with
3173  * historical BSD practice is quite difficult in the face of
3174  * clusters/bigalloc.  This is because multiple metadata blocks from
3175  * different block group can end up in the same allocation cluster.
3176  * Calculating the exact overhead in the face of clustered allocation
3177  * requires either O(all block bitmaps) in memory or O(number of block
3178  * groups**2) in time.  We will still calculate the superblock for
3179  * older file systems --- and if we come across with a bigalloc file
3180  * system with zero in s_overhead_clusters the estimate will be close to
3181  * correct especially for very large cluster sizes --- but for newer
3182  * file systems, it's better to calculate this figure once at mkfs
3183  * time, and store it in the superblock.  If the superblock value is
3184  * present (even for non-bigalloc file systems), we will use it.
3185  */
3186 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3187                           char *buf)
3188 {
3189         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3190         struct ext4_group_desc  *gdp;
3191         ext4_fsblk_t            first_block, last_block, b;
3192         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3193         int                     s, j, count = 0;
3194
3195         if (!ext4_has_feature_bigalloc(sb))
3196                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3197                         sbi->s_itb_per_group + 2);
3198
3199         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3200                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3201         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3202         for (i = 0; i < ngroups; i++) {
3203                 gdp = ext4_get_group_desc(sb, i, NULL);
3204                 b = ext4_block_bitmap(sb, gdp);
3205                 if (b >= first_block && b <= last_block) {
3206                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3207                         count++;
3208                 }
3209                 b = ext4_inode_bitmap(sb, gdp);
3210                 if (b >= first_block && b <= last_block) {
3211                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3212                         count++;
3213                 }
3214                 b = ext4_inode_table(sb, gdp);
3215                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3216                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3217                                 int c = EXT4_B2C(sbi, b - first_block);
3218                                 ext4_set_bit(c, buf);
3219                                 count++;
3220                         }
3221                 if (i != grp)
3222                         continue;