quota: Reduce contention on dq_data_lock
[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/dax.h>
41 #include <linux/cleancache.h>
42 #include <linux/uaccess.h>
43
44 #include <linux/kthread.h>
45 #include <linux/freezer.h>
46
47 #include "ext4.h"
48 #include "ext4_extents.h"       /* Needed for trace points definition */
49 #include "ext4_jbd2.h"
50 #include "xattr.h"
51 #include "acl.h"
52 #include "mballoc.h"
53 #include "fsmap.h"
54
55 #define CREATE_TRACE_POINTS
56 #include <trace/events/ext4.h>
57
58 static struct ext4_lazy_init *ext4_li_info;
59 static struct mutex ext4_li_mtx;
60 static struct ratelimit_state ext4_mount_msg_ratelimit;
61
62 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
63                              unsigned long journal_devnum);
64 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
65 static int ext4_commit_super(struct super_block *sb, int sync);
66 static void ext4_mark_recovery_complete(struct super_block *sb,
67                                         struct ext4_super_block *es);
68 static void ext4_clear_journal_err(struct super_block *sb,
69                                    struct ext4_super_block *es);
70 static int ext4_sync_fs(struct super_block *sb, int wait);
71 static int ext4_remount(struct super_block *sb, int *flags, char *data);
72 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
73 static int ext4_unfreeze(struct super_block *sb);
74 static int ext4_freeze(struct super_block *sb);
75 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
76                        const char *dev_name, void *data);
77 static inline int ext2_feature_set_ok(struct super_block *sb);
78 static inline int ext3_feature_set_ok(struct super_block *sb);
79 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
80 static void ext4_destroy_lazyinit_thread(void);
81 static void ext4_unregister_li_request(struct super_block *sb);
82 static void ext4_clear_request_list(void);
83 static struct inode *ext4_get_journal_inode(struct super_block *sb,
84                                             unsigned int journal_inum);
85
86 /*
87  * Lock ordering
88  *
89  * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
90  * i_mmap_rwsem (inode->i_mmap_rwsem)!
91  *
92  * page fault path:
93  * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
94  *   page lock -> i_data_sem (rw)
95  *
96  * buffered write path:
97  * sb_start_write -> i_mutex -> mmap_sem
98  * sb_start_write -> i_mutex -> transaction start -> page lock ->
99  *   i_data_sem (rw)
100  *
101  * truncate:
102  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
103  *   i_mmap_rwsem (w) -> page lock
104  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
105  *   transaction start -> i_data_sem (rw)
106  *
107  * direct IO:
108  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) -> mmap_sem
109  * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) ->
110  *   transaction start -> i_data_sem (rw)
111  *
112  * writepages:
113  * transaction start -> page lock(s) -> i_data_sem (rw)
114  */
115
116 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
117 static struct file_system_type ext2_fs_type = {
118         .owner          = THIS_MODULE,
119         .name           = "ext2",
120         .mount          = ext4_mount,
121         .kill_sb        = kill_block_super,
122         .fs_flags       = FS_REQUIRES_DEV,
123 };
124 MODULE_ALIAS_FS("ext2");
125 MODULE_ALIAS("ext2");
126 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
127 #else
128 #define IS_EXT2_SB(sb) (0)
129 #endif
130
131
132 static struct file_system_type ext3_fs_type = {
133         .owner          = THIS_MODULE,
134         .name           = "ext3",
135         .mount          = ext4_mount,
136         .kill_sb        = kill_block_super,
137         .fs_flags       = FS_REQUIRES_DEV,
138 };
139 MODULE_ALIAS_FS("ext3");
140 MODULE_ALIAS("ext3");
141 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
142
143 static int ext4_verify_csum_type(struct super_block *sb,
144                                  struct ext4_super_block *es)
145 {
146         if (!ext4_has_feature_metadata_csum(sb))
147                 return 1;
148
149         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
150 }
151
152 static __le32 ext4_superblock_csum(struct super_block *sb,
153                                    struct ext4_super_block *es)
154 {
155         struct ext4_sb_info *sbi = EXT4_SB(sb);
156         int offset = offsetof(struct ext4_super_block, s_checksum);
157         __u32 csum;
158
159         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
160
161         return cpu_to_le32(csum);
162 }
163
164 static int ext4_superblock_csum_verify(struct super_block *sb,
165                                        struct ext4_super_block *es)
166 {
167         if (!ext4_has_metadata_csum(sb))
168                 return 1;
169
170         return es->s_checksum == ext4_superblock_csum(sb, es);
171 }
172
173 void ext4_superblock_csum_set(struct super_block *sb)
174 {
175         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
176
177         if (!ext4_has_metadata_csum(sb))
178                 return;
179
180         es->s_checksum = ext4_superblock_csum(sb, es);
181 }
182
183 void *ext4_kvmalloc(size_t size, gfp_t flags)
184 {
185         void *ret;
186
187         ret = kmalloc(size, flags | __GFP_NOWARN);
188         if (!ret)
189                 ret = __vmalloc(size, flags, PAGE_KERNEL);
190         return ret;
191 }
192
193 void *ext4_kvzalloc(size_t size, gfp_t flags)
194 {
195         void *ret;
196
197         ret = kzalloc(size, flags | __GFP_NOWARN);
198         if (!ret)
199                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
200         return ret;
201 }
202
203 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
204                                struct ext4_group_desc *bg)
205 {
206         return le32_to_cpu(bg->bg_block_bitmap_lo) |
207                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
208                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
209 }
210
211 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
212                                struct ext4_group_desc *bg)
213 {
214         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
215                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
216                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
217 }
218
219 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
220                               struct ext4_group_desc *bg)
221 {
222         return le32_to_cpu(bg->bg_inode_table_lo) |
223                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
224                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
225 }
226
227 __u32 ext4_free_group_clusters(struct super_block *sb,
228                                struct ext4_group_desc *bg)
229 {
230         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
231                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
232                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
233 }
234
235 __u32 ext4_free_inodes_count(struct super_block *sb,
236                               struct ext4_group_desc *bg)
237 {
238         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
239                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
240                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
241 }
242
243 __u32 ext4_used_dirs_count(struct super_block *sb,
244                               struct ext4_group_desc *bg)
245 {
246         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
247                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
248                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
249 }
250
251 __u32 ext4_itable_unused_count(struct super_block *sb,
252                               struct ext4_group_desc *bg)
253 {
254         return le16_to_cpu(bg->bg_itable_unused_lo) |
255                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
256                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
257 }
258
259 void ext4_block_bitmap_set(struct super_block *sb,
260                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
261 {
262         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
263         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
264                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
265 }
266
267 void ext4_inode_bitmap_set(struct super_block *sb,
268                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
269 {
270         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
271         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
272                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
273 }
274
275 void ext4_inode_table_set(struct super_block *sb,
276                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
277 {
278         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
279         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
280                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
281 }
282
283 void ext4_free_group_clusters_set(struct super_block *sb,
284                                   struct ext4_group_desc *bg, __u32 count)
285 {
286         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
287         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
288                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
289 }
290
291 void ext4_free_inodes_set(struct super_block *sb,
292                           struct ext4_group_desc *bg, __u32 count)
293 {
294         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
295         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
296                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
297 }
298
299 void ext4_used_dirs_set(struct super_block *sb,
300                           struct ext4_group_desc *bg, __u32 count)
301 {
302         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
303         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
304                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
305 }
306
307 void ext4_itable_unused_set(struct super_block *sb,
308                           struct ext4_group_desc *bg, __u32 count)
309 {
310         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
311         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
312                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
313 }
314
315
316 static void __save_error_info(struct super_block *sb, const char *func,
317                             unsigned int line)
318 {
319         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
320
321         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
322         if (bdev_read_only(sb->s_bdev))
323                 return;
324         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
325         es->s_last_error_time = cpu_to_le32(get_seconds());
326         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
327         es->s_last_error_line = cpu_to_le32(line);
328         if (!es->s_first_error_time) {
329                 es->s_first_error_time = es->s_last_error_time;
330                 strncpy(es->s_first_error_func, func,
331                         sizeof(es->s_first_error_func));
332                 es->s_first_error_line = cpu_to_le32(line);
333                 es->s_first_error_ino = es->s_last_error_ino;
334                 es->s_first_error_block = es->s_last_error_block;
335         }
336         /*
337          * Start the daily error reporting function if it hasn't been
338          * started already
339          */
340         if (!es->s_error_count)
341                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
342         le32_add_cpu(&es->s_error_count, 1);
343 }
344
345 static void save_error_info(struct super_block *sb, const char *func,
346                             unsigned int line)
347 {
348         __save_error_info(sb, func, line);
349         ext4_commit_super(sb, 1);
350 }
351
352 /*
353  * The del_gendisk() function uninitializes the disk-specific data
354  * structures, including the bdi structure, without telling anyone
355  * else.  Once this happens, any attempt to call mark_buffer_dirty()
356  * (for example, by ext4_commit_super), will cause a kernel OOPS.
357  * This is a kludge to prevent these oops until we can put in a proper
358  * hook in del_gendisk() to inform the VFS and file system layers.
359  */
360 static int block_device_ejected(struct super_block *sb)
361 {
362         struct inode *bd_inode = sb->s_bdev->bd_inode;
363         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
364
365         return bdi->dev == NULL;
366 }
367
368 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
369 {
370         struct super_block              *sb = journal->j_private;
371         struct ext4_sb_info             *sbi = EXT4_SB(sb);
372         int                             error = is_journal_aborted(journal);
373         struct ext4_journal_cb_entry    *jce;
374
375         BUG_ON(txn->t_state == T_FINISHED);
376
377         ext4_process_freed_data(sb, txn->t_tid);
378
379         spin_lock(&sbi->s_md_lock);
380         while (!list_empty(&txn->t_private_list)) {
381                 jce = list_entry(txn->t_private_list.next,
382                                  struct ext4_journal_cb_entry, jce_list);
383                 list_del_init(&jce->jce_list);
384                 spin_unlock(&sbi->s_md_lock);
385                 jce->jce_func(sb, jce, error);
386                 spin_lock(&sbi->s_md_lock);
387         }
388         spin_unlock(&sbi->s_md_lock);
389 }
390
391 /* Deal with the reporting of failure conditions on a filesystem such as
392  * inconsistencies detected or read IO failures.
393  *
394  * On ext2, we can store the error state of the filesystem in the
395  * superblock.  That is not possible on ext4, because we may have other
396  * write ordering constraints on the superblock which prevent us from
397  * writing it out straight away; and given that the journal is about to
398  * be aborted, we can't rely on the current, or future, transactions to
399  * write out the superblock safely.
400  *
401  * We'll just use the jbd2_journal_abort() error code to record an error in
402  * the journal instead.  On recovery, the journal will complain about
403  * that error until we've noted it down and cleared it.
404  */
405
406 static void ext4_handle_error(struct super_block *sb)
407 {
408         if (sb->s_flags & MS_RDONLY)
409                 return;
410
411         if (!test_opt(sb, ERRORS_CONT)) {
412                 journal_t *journal = EXT4_SB(sb)->s_journal;
413
414                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
415                 if (journal)
416                         jbd2_journal_abort(journal, -EIO);
417         }
418         if (test_opt(sb, ERRORS_RO)) {
419                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
420                 /*
421                  * Make sure updated value of ->s_mount_flags will be visible
422                  * before ->s_flags update
423                  */
424                 smp_wmb();
425                 sb->s_flags |= MS_RDONLY;
426         }
427         if (test_opt(sb, ERRORS_PANIC)) {
428                 if (EXT4_SB(sb)->s_journal &&
429                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
430                         return;
431                 panic("EXT4-fs (device %s): panic forced after error\n",
432                         sb->s_id);
433         }
434 }
435
436 #define ext4_error_ratelimit(sb)                                        \
437                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
438                              "EXT4-fs error")
439
440 void __ext4_error(struct super_block *sb, const char *function,
441                   unsigned int line, const char *fmt, ...)
442 {
443         struct va_format vaf;
444         va_list args;
445
446         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
447                 return;
448
449         if (ext4_error_ratelimit(sb)) {
450                 va_start(args, fmt);
451                 vaf.fmt = fmt;
452                 vaf.va = &args;
453                 printk(KERN_CRIT
454                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
455                        sb->s_id, function, line, current->comm, &vaf);
456                 va_end(args);
457         }
458         save_error_info(sb, function, line);
459         ext4_handle_error(sb);
460 }
461
462 void __ext4_error_inode(struct inode *inode, const char *function,
463                         unsigned int line, ext4_fsblk_t block,
464                         const char *fmt, ...)
465 {
466         va_list args;
467         struct va_format vaf;
468         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
469
470         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
471                 return;
472
473         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
474         es->s_last_error_block = cpu_to_le64(block);
475         if (ext4_error_ratelimit(inode->i_sb)) {
476                 va_start(args, fmt);
477                 vaf.fmt = fmt;
478                 vaf.va = &args;
479                 if (block)
480                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
481                                "inode #%lu: block %llu: comm %s: %pV\n",
482                                inode->i_sb->s_id, function, line, inode->i_ino,
483                                block, current->comm, &vaf);
484                 else
485                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
486                                "inode #%lu: comm %s: %pV\n",
487                                inode->i_sb->s_id, function, line, inode->i_ino,
488                                current->comm, &vaf);
489                 va_end(args);
490         }
491         save_error_info(inode->i_sb, function, line);
492         ext4_handle_error(inode->i_sb);
493 }
494
495 void __ext4_error_file(struct file *file, const char *function,
496                        unsigned int line, ext4_fsblk_t block,
497                        const char *fmt, ...)
498 {
499         va_list args;
500         struct va_format vaf;
501         struct ext4_super_block *es;
502         struct inode *inode = file_inode(file);
503         char pathname[80], *path;
504
505         if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
506                 return;
507
508         es = EXT4_SB(inode->i_sb)->s_es;
509         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
510         if (ext4_error_ratelimit(inode->i_sb)) {
511                 path = file_path(file, pathname, sizeof(pathname));
512                 if (IS_ERR(path))
513                         path = "(unknown)";
514                 va_start(args, fmt);
515                 vaf.fmt = fmt;
516                 vaf.va = &args;
517                 if (block)
518                         printk(KERN_CRIT
519                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
520                                "block %llu: comm %s: path %s: %pV\n",
521                                inode->i_sb->s_id, function, line, inode->i_ino,
522                                block, current->comm, path, &vaf);
523                 else
524                         printk(KERN_CRIT
525                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
526                                "comm %s: path %s: %pV\n",
527                                inode->i_sb->s_id, function, line, inode->i_ino,
528                                current->comm, path, &vaf);
529                 va_end(args);
530         }
531         save_error_info(inode->i_sb, function, line);
532         ext4_handle_error(inode->i_sb);
533 }
534
535 const char *ext4_decode_error(struct super_block *sb, int errno,
536                               char nbuf[16])
537 {
538         char *errstr = NULL;
539
540         switch (errno) {
541         case -EFSCORRUPTED:
542                 errstr = "Corrupt filesystem";
543                 break;
544         case -EFSBADCRC:
545                 errstr = "Filesystem failed CRC";
546                 break;
547         case -EIO:
548                 errstr = "IO failure";
549                 break;
550         case -ENOMEM:
551                 errstr = "Out of memory";
552                 break;
553         case -EROFS:
554                 if (!sb || (EXT4_SB(sb)->s_journal &&
555                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
556                         errstr = "Journal has aborted";
557                 else
558                         errstr = "Readonly filesystem";
559                 break;
560         default:
561                 /* If the caller passed in an extra buffer for unknown
562                  * errors, textualise them now.  Else we just return
563                  * NULL. */
564                 if (nbuf) {
565                         /* Check for truncated error codes... */
566                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
567                                 errstr = nbuf;
568                 }
569                 break;
570         }
571
572         return errstr;
573 }
574
575 /* __ext4_std_error decodes expected errors from journaling functions
576  * automatically and invokes the appropriate error response.  */
577
578 void __ext4_std_error(struct super_block *sb, const char *function,
579                       unsigned int line, int errno)
580 {
581         char nbuf[16];
582         const char *errstr;
583
584         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
585                 return;
586
587         /* Special case: if the error is EROFS, and we're not already
588          * inside a transaction, then there's really no point in logging
589          * an error. */
590         if (errno == -EROFS && journal_current_handle() == NULL &&
591             (sb->s_flags & MS_RDONLY))
592                 return;
593
594         if (ext4_error_ratelimit(sb)) {
595                 errstr = ext4_decode_error(sb, errno, nbuf);
596                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
597                        sb->s_id, function, line, errstr);
598         }
599
600         save_error_info(sb, function, line);
601         ext4_handle_error(sb);
602 }
603
604 /*
605  * ext4_abort is a much stronger failure handler than ext4_error.  The
606  * abort function may be used to deal with unrecoverable failures such
607  * as journal IO errors or ENOMEM at a critical moment in log management.
608  *
609  * We unconditionally force the filesystem into an ABORT|READONLY state,
610  * unless the error response on the fs has been set to panic in which
611  * case we take the easy way out and panic immediately.
612  */
613
614 void __ext4_abort(struct super_block *sb, const char *function,
615                 unsigned int line, const char *fmt, ...)
616 {
617         struct va_format vaf;
618         va_list args;
619
620         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
621                 return;
622
623         save_error_info(sb, function, line);
624         va_start(args, fmt);
625         vaf.fmt = fmt;
626         vaf.va = &args;
627         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: %pV\n",
628                sb->s_id, function, line, &vaf);
629         va_end(args);
630
631         if ((sb->s_flags & MS_RDONLY) == 0) {
632                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
633                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
634                 /*
635                  * Make sure updated value of ->s_mount_flags will be visible
636                  * before ->s_flags update
637                  */
638                 smp_wmb();
639                 sb->s_flags |= MS_RDONLY;
640                 if (EXT4_SB(sb)->s_journal)
641                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
642                 save_error_info(sb, function, line);
643         }
644         if (test_opt(sb, ERRORS_PANIC)) {
645                 if (EXT4_SB(sb)->s_journal &&
646                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
647                         return;
648                 panic("EXT4-fs panic from previous error\n");
649         }
650 }
651
652 void __ext4_msg(struct super_block *sb,
653                 const char *prefix, const char *fmt, ...)
654 {
655         struct va_format vaf;
656         va_list args;
657
658         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
659                 return;
660
661         va_start(args, fmt);
662         vaf.fmt = fmt;
663         vaf.va = &args;
664         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
665         va_end(args);
666 }
667
668 #define ext4_warning_ratelimit(sb)                                      \
669                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
670                              "EXT4-fs warning")
671
672 void __ext4_warning(struct super_block *sb, const char *function,
673                     unsigned int line, const char *fmt, ...)
674 {
675         struct va_format vaf;
676         va_list args;
677
678         if (!ext4_warning_ratelimit(sb))
679                 return;
680
681         va_start(args, fmt);
682         vaf.fmt = fmt;
683         vaf.va = &args;
684         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
685                sb->s_id, function, line, &vaf);
686         va_end(args);
687 }
688
689 void __ext4_warning_inode(const struct inode *inode, const char *function,
690                           unsigned int line, const char *fmt, ...)
691 {
692         struct va_format vaf;
693         va_list args;
694
695         if (!ext4_warning_ratelimit(inode->i_sb))
696                 return;
697
698         va_start(args, fmt);
699         vaf.fmt = fmt;
700         vaf.va = &args;
701         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
702                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
703                function, line, inode->i_ino, current->comm, &vaf);
704         va_end(args);
705 }
706
707 void __ext4_grp_locked_error(const char *function, unsigned int line,
708                              struct super_block *sb, ext4_group_t grp,
709                              unsigned long ino, ext4_fsblk_t block,
710                              const char *fmt, ...)
711 __releases(bitlock)
712 __acquires(bitlock)
713 {
714         struct va_format vaf;
715         va_list args;
716         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
717
718         if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
719                 return;
720
721         es->s_last_error_ino = cpu_to_le32(ino);
722         es->s_last_error_block = cpu_to_le64(block);
723         __save_error_info(sb, function, line);
724
725         if (ext4_error_ratelimit(sb)) {
726                 va_start(args, fmt);
727                 vaf.fmt = fmt;
728                 vaf.va = &args;
729                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
730                        sb->s_id, function, line, grp);
731                 if (ino)
732                         printk(KERN_CONT "inode %lu: ", ino);
733                 if (block)
734                         printk(KERN_CONT "block %llu:",
735                                (unsigned long long) block);
736                 printk(KERN_CONT "%pV\n", &vaf);
737                 va_end(args);
738         }
739
740         if (test_opt(sb, ERRORS_CONT)) {
741                 ext4_commit_super(sb, 0);
742                 return;
743         }
744
745         ext4_unlock_group(sb, grp);
746         ext4_handle_error(sb);
747         /*
748          * We only get here in the ERRORS_RO case; relocking the group
749          * may be dangerous, but nothing bad will happen since the
750          * filesystem will have already been marked read/only and the
751          * journal has been aborted.  We return 1 as a hint to callers
752          * who might what to use the return value from
753          * ext4_grp_locked_error() to distinguish between the
754          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
755          * aggressively from the ext4 function in question, with a
756          * more appropriate error code.
757          */
758         ext4_lock_group(sb, grp);
759         return;
760 }
761
762 void ext4_update_dynamic_rev(struct super_block *sb)
763 {
764         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
765
766         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
767                 return;
768
769         ext4_warning(sb,
770                      "updating to rev %d because of new feature flag, "
771                      "running e2fsck is recommended",
772                      EXT4_DYNAMIC_REV);
773
774         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
775         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
776         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
777         /* leave es->s_feature_*compat flags alone */
778         /* es->s_uuid will be set by e2fsck if empty */
779
780         /*
781          * The rest of the superblock fields should be zero, and if not it
782          * means they are likely already in use, so leave them alone.  We
783          * can leave it up to e2fsck to clean up any inconsistencies there.
784          */
785 }
786
787 /*
788  * Open the external journal device
789  */
790 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
791 {
792         struct block_device *bdev;
793         char b[BDEVNAME_SIZE];
794
795         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
796         if (IS_ERR(bdev))
797                 goto fail;
798         return bdev;
799
800 fail:
801         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
802                         __bdevname(dev, b), PTR_ERR(bdev));
803         return NULL;
804 }
805
806 /*
807  * Release the journal device
808  */
809 static void ext4_blkdev_put(struct block_device *bdev)
810 {
811         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
812 }
813
814 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
815 {
816         struct block_device *bdev;
817         bdev = sbi->journal_bdev;
818         if (bdev) {
819                 ext4_blkdev_put(bdev);
820                 sbi->journal_bdev = NULL;
821         }
822 }
823
824 static inline struct inode *orphan_list_entry(struct list_head *l)
825 {
826         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
827 }
828
829 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
830 {
831         struct list_head *l;
832
833         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
834                  le32_to_cpu(sbi->s_es->s_last_orphan));
835
836         printk(KERN_ERR "sb_info orphan list:\n");
837         list_for_each(l, &sbi->s_orphan) {
838                 struct inode *inode = orphan_list_entry(l);
839                 printk(KERN_ERR "  "
840                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
841                        inode->i_sb->s_id, inode->i_ino, inode,
842                        inode->i_mode, inode->i_nlink,
843                        NEXT_ORPHAN(inode));
844         }
845 }
846
847 #ifdef CONFIG_QUOTA
848 static int ext4_quota_off(struct super_block *sb, int type);
849
850 static inline void ext4_quota_off_umount(struct super_block *sb)
851 {
852         int type;
853
854         /* Use our quota_off function to clear inode flags etc. */
855         for (type = 0; type < EXT4_MAXQUOTAS; type++)
856                 ext4_quota_off(sb, type);
857 }
858 #else
859 static inline void ext4_quota_off_umount(struct super_block *sb)
860 {
861 }
862 #endif
863
864 static void ext4_put_super(struct super_block *sb)
865 {
866         struct ext4_sb_info *sbi = EXT4_SB(sb);
867         struct ext4_super_block *es = sbi->s_es;
868         int aborted = 0;
869         int i, err;
870
871         ext4_unregister_li_request(sb);
872         ext4_quota_off_umount(sb);
873
874         flush_workqueue(sbi->rsv_conversion_wq);
875         destroy_workqueue(sbi->rsv_conversion_wq);
876
877         if (sbi->s_journal) {
878                 aborted = is_journal_aborted(sbi->s_journal);
879                 err = jbd2_journal_destroy(sbi->s_journal);
880                 sbi->s_journal = NULL;
881                 if ((err < 0) && !aborted)
882                         ext4_abort(sb, "Couldn't clean up the journal");
883         }
884
885         ext4_unregister_sysfs(sb);
886         ext4_es_unregister_shrinker(sbi);
887         del_timer_sync(&sbi->s_err_report);
888         ext4_release_system_zone(sb);
889         ext4_mb_release(sb);
890         ext4_ext_release(sb);
891
892         if (!(sb->s_flags & MS_RDONLY) && !aborted) {
893                 ext4_clear_feature_journal_needs_recovery(sb);
894                 es->s_state = cpu_to_le16(sbi->s_mount_state);
895         }
896         if (!(sb->s_flags & MS_RDONLY))
897                 ext4_commit_super(sb, 1);
898
899         for (i = 0; i < sbi->s_gdb_count; i++)
900                 brelse(sbi->s_group_desc[i]);
901         kvfree(sbi->s_group_desc);
902         kvfree(sbi->s_flex_groups);
903         percpu_counter_destroy(&sbi->s_freeclusters_counter);
904         percpu_counter_destroy(&sbi->s_freeinodes_counter);
905         percpu_counter_destroy(&sbi->s_dirs_counter);
906         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
907         percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
908 #ifdef CONFIG_QUOTA
909         for (i = 0; i < EXT4_MAXQUOTAS; i++)
910                 kfree(sbi->s_qf_names[i]);
911 #endif
912
913         /* Debugging code just in case the in-memory inode orphan list
914          * isn't empty.  The on-disk one can be non-empty if we've
915          * detected an error and taken the fs readonly, but the
916          * in-memory list had better be clean by this point. */
917         if (!list_empty(&sbi->s_orphan))
918                 dump_orphan_list(sb, sbi);
919         J_ASSERT(list_empty(&sbi->s_orphan));
920
921         sync_blockdev(sb->s_bdev);
922         invalidate_bdev(sb->s_bdev);
923         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
924                 /*
925                  * Invalidate the journal device's buffers.  We don't want them
926                  * floating about in memory - the physical journal device may
927                  * hotswapped, and it breaks the `ro-after' testing code.
928                  */
929                 sync_blockdev(sbi->journal_bdev);
930                 invalidate_bdev(sbi->journal_bdev);
931                 ext4_blkdev_remove(sbi);
932         }
933         if (sbi->s_ea_inode_cache) {
934                 ext4_xattr_destroy_cache(sbi->s_ea_inode_cache);
935                 sbi->s_ea_inode_cache = NULL;
936         }
937         if (sbi->s_ea_block_cache) {
938                 ext4_xattr_destroy_cache(sbi->s_ea_block_cache);
939                 sbi->s_ea_block_cache = NULL;
940         }
941         if (sbi->s_mmp_tsk)
942                 kthread_stop(sbi->s_mmp_tsk);
943         brelse(sbi->s_sbh);
944         sb->s_fs_info = NULL;
945         /*
946          * Now that we are completely done shutting down the
947          * superblock, we need to actually destroy the kobject.
948          */
949         kobject_put(&sbi->s_kobj);
950         wait_for_completion(&sbi->s_kobj_unregister);
951         if (sbi->s_chksum_driver)
952                 crypto_free_shash(sbi->s_chksum_driver);
953         kfree(sbi->s_blockgroup_lock);
954         kfree(sbi);
955 }
956
957 static struct kmem_cache *ext4_inode_cachep;
958
959 /*
960  * Called inside transaction, so use GFP_NOFS
961  */
962 static struct inode *ext4_alloc_inode(struct super_block *sb)
963 {
964         struct ext4_inode_info *ei;
965
966         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
967         if (!ei)
968                 return NULL;
969
970         ei->vfs_inode.i_version = 1;
971         spin_lock_init(&ei->i_raw_lock);
972         INIT_LIST_HEAD(&ei->i_prealloc_list);
973         spin_lock_init(&ei->i_prealloc_lock);
974         ext4_es_init_tree(&ei->i_es_tree);
975         rwlock_init(&ei->i_es_lock);
976         INIT_LIST_HEAD(&ei->i_es_list);
977         ei->i_es_all_nr = 0;
978         ei->i_es_shk_nr = 0;
979         ei->i_es_shrink_lblk = 0;
980         ei->i_reserved_data_blocks = 0;
981         ei->i_da_metadata_calc_len = 0;
982         ei->i_da_metadata_calc_last_lblock = 0;
983         spin_lock_init(&(ei->i_block_reservation_lock));
984 #ifdef CONFIG_QUOTA
985         ei->i_reserved_quota = 0;
986         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
987 #endif
988         ei->jinode = NULL;
989         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
990         spin_lock_init(&ei->i_completed_io_lock);
991         ei->i_sync_tid = 0;
992         ei->i_datasync_tid = 0;
993         atomic_set(&ei->i_unwritten, 0);
994         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
995         return &ei->vfs_inode;
996 }
997
998 static int ext4_drop_inode(struct inode *inode)
999 {
1000         int drop = generic_drop_inode(inode);
1001
1002         trace_ext4_drop_inode(inode, drop);
1003         return drop;
1004 }
1005
1006 static void ext4_i_callback(struct rcu_head *head)
1007 {
1008         struct inode *inode = container_of(head, struct inode, i_rcu);
1009         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1010 }
1011
1012 static void ext4_destroy_inode(struct inode *inode)
1013 {
1014         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1015                 ext4_msg(inode->i_sb, KERN_ERR,
1016                          "Inode %lu (%p): orphan list check failed!",
1017                          inode->i_ino, EXT4_I(inode));
1018                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1019                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1020                                 true);
1021                 dump_stack();
1022         }
1023         call_rcu(&inode->i_rcu, ext4_i_callback);
1024 }
1025
1026 static void init_once(void *foo)
1027 {
1028         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1029
1030         INIT_LIST_HEAD(&ei->i_orphan);
1031         init_rwsem(&ei->xattr_sem);
1032         init_rwsem(&ei->i_data_sem);
1033         init_rwsem(&ei->i_mmap_sem);
1034         inode_init_once(&ei->vfs_inode);
1035 }
1036
1037 static int __init init_inodecache(void)
1038 {
1039         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1040                                              sizeof(struct ext4_inode_info),
1041                                              0, (SLAB_RECLAIM_ACCOUNT|
1042                                                 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1043                                              init_once);
1044         if (ext4_inode_cachep == NULL)
1045                 return -ENOMEM;
1046         return 0;
1047 }
1048
1049 static void destroy_inodecache(void)
1050 {
1051         /*
1052          * Make sure all delayed rcu free inodes are flushed before we
1053          * destroy cache.
1054          */
1055         rcu_barrier();
1056         kmem_cache_destroy(ext4_inode_cachep);
1057 }
1058
1059 void ext4_clear_inode(struct inode *inode)
1060 {
1061         invalidate_inode_buffers(inode);
1062         clear_inode(inode);
1063         dquot_drop(inode);
1064         ext4_discard_preallocations(inode);
1065         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1066         if (EXT4_I(inode)->jinode) {
1067                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1068                                                EXT4_I(inode)->jinode);
1069                 jbd2_free_inode(EXT4_I(inode)->jinode);
1070                 EXT4_I(inode)->jinode = NULL;
1071         }
1072 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1073         fscrypt_put_encryption_info(inode, NULL);
1074 #endif
1075 }
1076
1077 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1078                                         u64 ino, u32 generation)
1079 {
1080         struct inode *inode;
1081
1082         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1083                 return ERR_PTR(-ESTALE);
1084         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1085                 return ERR_PTR(-ESTALE);
1086
1087         /* iget isn't really right if the inode is currently unallocated!!
1088          *
1089          * ext4_read_inode will return a bad_inode if the inode had been
1090          * deleted, so we should be safe.
1091          *
1092          * Currently we don't know the generation for parent directory, so
1093          * a generation of 0 means "accept any"
1094          */
1095         inode = ext4_iget_normal(sb, ino);
1096         if (IS_ERR(inode))
1097                 return ERR_CAST(inode);
1098         if (generation && inode->i_generation != generation) {
1099                 iput(inode);
1100                 return ERR_PTR(-ESTALE);
1101         }
1102
1103         return inode;
1104 }
1105
1106 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1107                                         int fh_len, int fh_type)
1108 {
1109         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1110                                     ext4_nfs_get_inode);
1111 }
1112
1113 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1114                                         int fh_len, int fh_type)
1115 {
1116         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1117                                     ext4_nfs_get_inode);
1118 }
1119
1120 /*
1121  * Try to release metadata pages (indirect blocks, directories) which are
1122  * mapped via the block device.  Since these pages could have journal heads
1123  * which would prevent try_to_free_buffers() from freeing them, we must use
1124  * jbd2 layer's try_to_free_buffers() function to release them.
1125  */
1126 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1127                                  gfp_t wait)
1128 {
1129         journal_t *journal = EXT4_SB(sb)->s_journal;
1130
1131         WARN_ON(PageChecked(page));
1132         if (!page_has_buffers(page))
1133                 return 0;
1134         if (journal)
1135                 return jbd2_journal_try_to_free_buffers(journal, page,
1136                                                 wait & ~__GFP_DIRECT_RECLAIM);
1137         return try_to_free_buffers(page);
1138 }
1139
1140 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1141 static int ext4_get_context(struct inode *inode, void *ctx, size_t len)
1142 {
1143         return ext4_xattr_get(inode, EXT4_XATTR_INDEX_ENCRYPTION,
1144                                  EXT4_XATTR_NAME_ENCRYPTION_CONTEXT, ctx, len);
1145 }
1146
1147 static int ext4_set_context(struct inode *inode, const void *ctx, size_t len,
1148                                                         void *fs_data)
1149 {
1150         handle_t *handle = fs_data;
1151         int res, res2, credits, retries = 0;
1152
1153         /*
1154          * Encrypting the root directory is not allowed because e2fsck expects
1155          * lost+found to exist and be unencrypted, and encrypting the root
1156          * directory would imply encrypting the lost+found directory as well as
1157          * the filename "lost+found" itself.
1158          */
1159         if (inode->i_ino == EXT4_ROOT_INO)
1160                 return -EPERM;
1161
1162         res = ext4_convert_inline_data(inode);
1163         if (res)
1164                 return res;
1165
1166         /*
1167          * If a journal handle was specified, then the encryption context is
1168          * being set on a new inode via inheritance and is part of a larger
1169          * transaction to create the inode.  Otherwise the encryption context is
1170          * being set on an existing inode in its own transaction.  Only in the
1171          * latter case should the "retry on ENOSPC" logic be used.
1172          */
1173
1174         if (handle) {
1175                 res = ext4_xattr_set_handle(handle, inode,
1176                                             EXT4_XATTR_INDEX_ENCRYPTION,
1177                                             EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1178                                             ctx, len, 0);
1179                 if (!res) {
1180                         ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1181                         ext4_clear_inode_state(inode,
1182                                         EXT4_STATE_MAY_INLINE_DATA);
1183                         /*
1184                          * Update inode->i_flags - e.g. S_DAX may get disabled
1185                          */
1186                         ext4_set_inode_flags(inode);
1187                 }
1188                 return res;
1189         }
1190
1191         res = dquot_initialize(inode);
1192         if (res)
1193                 return res;
1194 retry:
1195         res = ext4_xattr_set_credits(inode, len, false /* is_create */,
1196                                      &credits);
1197         if (res)
1198                 return res;
1199
1200         handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
1201         if (IS_ERR(handle))
1202                 return PTR_ERR(handle);
1203
1204         res = ext4_xattr_set_handle(handle, inode, EXT4_XATTR_INDEX_ENCRYPTION,
1205                                     EXT4_XATTR_NAME_ENCRYPTION_CONTEXT,
1206                                     ctx, len, 0);
1207         if (!res) {
1208                 ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT);
1209                 /* Update inode->i_flags - e.g. S_DAX may get disabled */
1210                 ext4_set_inode_flags(inode);
1211                 res = ext4_mark_inode_dirty(handle, inode);
1212                 if (res)
1213                         EXT4_ERROR_INODE(inode, "Failed to mark inode dirty");
1214         }
1215         res2 = ext4_journal_stop(handle);
1216
1217         if (res == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1218                 goto retry;
1219         if (!res)
1220                 res = res2;
1221         return res;
1222 }
1223
1224 static bool ext4_dummy_context(struct inode *inode)
1225 {
1226         return DUMMY_ENCRYPTION_ENABLED(EXT4_SB(inode->i_sb));
1227 }
1228
1229 static unsigned ext4_max_namelen(struct inode *inode)
1230 {
1231         return S_ISLNK(inode->i_mode) ? inode->i_sb->s_blocksize :
1232                 EXT4_NAME_LEN;
1233 }
1234
1235 static const struct fscrypt_operations ext4_cryptops = {
1236         .key_prefix             = "ext4:",
1237         .get_context            = ext4_get_context,
1238         .set_context            = ext4_set_context,
1239         .dummy_context          = ext4_dummy_context,
1240         .is_encrypted           = ext4_encrypted_inode,
1241         .empty_dir              = ext4_empty_dir,
1242         .max_namelen            = ext4_max_namelen,
1243 };
1244 #else
1245 static const struct fscrypt_operations ext4_cryptops = {
1246         .is_encrypted           = ext4_encrypted_inode,
1247 };
1248 #endif
1249
1250 #ifdef CONFIG_QUOTA
1251 static const char * const quotatypes[] = INITQFNAMES;
1252 #define QTYPE2NAME(t) (quotatypes[t])
1253
1254 static int ext4_write_dquot(struct dquot *dquot);
1255 static int ext4_acquire_dquot(struct dquot *dquot);
1256 static int ext4_release_dquot(struct dquot *dquot);
1257 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1258 static int ext4_write_info(struct super_block *sb, int type);
1259 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1260                          const struct path *path);
1261 static int ext4_quota_on_mount(struct super_block *sb, int type);
1262 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1263                                size_t len, loff_t off);
1264 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1265                                 const char *data, size_t len, loff_t off);
1266 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1267                              unsigned int flags);
1268 static int ext4_enable_quotas(struct super_block *sb);
1269 static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1270
1271 static struct dquot **ext4_get_dquots(struct inode *inode)
1272 {
1273         return EXT4_I(inode)->i_dquot;
1274 }
1275
1276 static const struct dquot_operations ext4_quota_operations = {
1277         .get_reserved_space     = ext4_get_reserved_space,
1278         .write_dquot            = ext4_write_dquot,
1279         .acquire_dquot          = ext4_acquire_dquot,
1280         .release_dquot          = ext4_release_dquot,
1281         .mark_dirty             = ext4_mark_dquot_dirty,
1282         .write_info             = ext4_write_info,
1283         .alloc_dquot            = dquot_alloc,
1284         .destroy_dquot          = dquot_destroy,
1285         .get_projid             = ext4_get_projid,
1286         .get_inode_usage        = ext4_get_inode_usage,
1287         .get_next_id            = ext4_get_next_id,
1288 };
1289
1290 static const struct quotactl_ops ext4_qctl_operations = {
1291         .quota_on       = ext4_quota_on,
1292         .quota_off      = ext4_quota_off,
1293         .quota_sync     = dquot_quota_sync,
1294         .get_state      = dquot_get_state,
1295         .set_info       = dquot_set_dqinfo,
1296         .get_dqblk      = dquot_get_dqblk,
1297         .set_dqblk      = dquot_set_dqblk,
1298         .get_nextdqblk  = dquot_get_next_dqblk,
1299 };
1300 #endif
1301
1302 static const struct super_operations ext4_sops = {
1303         .alloc_inode    = ext4_alloc_inode,
1304         .destroy_inode  = ext4_destroy_inode,
1305         .write_inode    = ext4_write_inode,
1306         .dirty_inode    = ext4_dirty_inode,
1307         .drop_inode     = ext4_drop_inode,
1308         .evict_inode    = ext4_evict_inode,
1309         .put_super      = ext4_put_super,
1310         .sync_fs        = ext4_sync_fs,
1311         .freeze_fs      = ext4_freeze,
1312         .unfreeze_fs    = ext4_unfreeze,
1313         .statfs         = ext4_statfs,
1314         .remount_fs     = ext4_remount,
1315         .show_options   = ext4_show_options,
1316 #ifdef CONFIG_QUOTA
1317         .quota_read     = ext4_quota_read,
1318         .quota_write    = ext4_quota_write,
1319         .get_dquots     = ext4_get_dquots,
1320 #endif
1321         .bdev_try_to_free_page = bdev_try_to_free_page,
1322 };
1323
1324 static const struct export_operations ext4_export_ops = {
1325         .fh_to_dentry = ext4_fh_to_dentry,
1326         .fh_to_parent = ext4_fh_to_parent,
1327         .get_parent = ext4_get_parent,
1328 };
1329
1330 enum {
1331         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1332         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1333         Opt_nouid32, Opt_debug, Opt_removed,
1334         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1335         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1336         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1337         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1338         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1339         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1340         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1341         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1342         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1343         Opt_usrquota, Opt_grpquota, Opt_prjquota, Opt_i_version, Opt_dax,
1344         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1345         Opt_lazytime, Opt_nolazytime, Opt_debug_want_extra_isize,
1346         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1347         Opt_inode_readahead_blks, Opt_journal_ioprio,
1348         Opt_dioread_nolock, Opt_dioread_lock,
1349         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1350         Opt_max_dir_size_kb, Opt_nojournal_checksum, Opt_nombcache,
1351 };
1352
1353 static const match_table_t tokens = {
1354         {Opt_bsd_df, "bsddf"},
1355         {Opt_minix_df, "minixdf"},
1356         {Opt_grpid, "grpid"},
1357         {Opt_grpid, "bsdgroups"},
1358         {Opt_nogrpid, "nogrpid"},
1359         {Opt_nogrpid, "sysvgroups"},
1360         {Opt_resgid, "resgid=%u"},
1361         {Opt_resuid, "resuid=%u"},
1362         {Opt_sb, "sb=%u"},
1363         {Opt_err_cont, "errors=continue"},
1364         {Opt_err_panic, "errors=panic"},
1365         {Opt_err_ro, "errors=remount-ro"},
1366         {Opt_nouid32, "nouid32"},
1367         {Opt_debug, "debug"},
1368         {Opt_removed, "oldalloc"},
1369         {Opt_removed, "orlov"},
1370         {Opt_user_xattr, "user_xattr"},
1371         {Opt_nouser_xattr, "nouser_xattr"},
1372         {Opt_acl, "acl"},
1373         {Opt_noacl, "noacl"},
1374         {Opt_noload, "norecovery"},
1375         {Opt_noload, "noload"},
1376         {Opt_removed, "nobh"},
1377         {Opt_removed, "bh"},
1378         {Opt_commit, "commit=%u"},
1379         {Opt_min_batch_time, "min_batch_time=%u"},
1380         {Opt_max_batch_time, "max_batch_time=%u"},
1381         {Opt_journal_dev, "journal_dev=%u"},
1382         {Opt_journal_path, "journal_path=%s"},
1383         {Opt_journal_checksum, "journal_checksum"},
1384         {Opt_nojournal_checksum, "nojournal_checksum"},
1385         {Opt_journal_async_commit, "journal_async_commit"},
1386         {Opt_abort, "abort"},
1387         {Opt_data_journal, "data=journal"},
1388         {Opt_data_ordered, "data=ordered"},
1389         {Opt_data_writeback, "data=writeback"},
1390         {Opt_data_err_abort, "data_err=abort"},
1391         {Opt_data_err_ignore, "data_err=ignore"},
1392         {Opt_offusrjquota, "usrjquota="},
1393         {Opt_usrjquota, "usrjquota=%s"},
1394         {Opt_offgrpjquota, "grpjquota="},
1395         {Opt_grpjquota, "grpjquota=%s"},
1396         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1397         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1398         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1399         {Opt_grpquota, "grpquota"},
1400         {Opt_noquota, "noquota"},
1401         {Opt_quota, "quota"},
1402         {Opt_usrquota, "usrquota"},
1403         {Opt_prjquota, "prjquota"},
1404         {Opt_barrier, "barrier=%u"},
1405         {Opt_barrier, "barrier"},
1406         {Opt_nobarrier, "nobarrier"},
1407         {Opt_i_version, "i_version"},
1408         {Opt_dax, "dax"},
1409         {Opt_stripe, "stripe=%u"},
1410         {Opt_delalloc, "delalloc"},
1411         {Opt_lazytime, "lazytime"},
1412         {Opt_nolazytime, "nolazytime"},
1413         {Opt_debug_want_extra_isize, "debug_want_extra_isize=%u"},
1414         {Opt_nodelalloc, "nodelalloc"},
1415         {Opt_removed, "mblk_io_submit"},
1416         {Opt_removed, "nomblk_io_submit"},
1417         {Opt_block_validity, "block_validity"},
1418         {Opt_noblock_validity, "noblock_validity"},
1419         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1420         {Opt_journal_ioprio, "journal_ioprio=%u"},
1421         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1422         {Opt_auto_da_alloc, "auto_da_alloc"},
1423         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1424         {Opt_dioread_nolock, "dioread_nolock"},
1425         {Opt_dioread_lock, "dioread_lock"},
1426         {Opt_discard, "discard"},
1427         {Opt_nodiscard, "nodiscard"},
1428         {Opt_init_itable, "init_itable=%u"},
1429         {Opt_init_itable, "init_itable"},
1430         {Opt_noinit_itable, "noinit_itable"},
1431         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1432         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1433         {Opt_nombcache, "nombcache"},
1434         {Opt_nombcache, "no_mbcache"},  /* for backward compatibility */
1435         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1436         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1437         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1438         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1439         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1440         {Opt_err, NULL},
1441 };
1442
1443 static ext4_fsblk_t get_sb_block(void **data)
1444 {
1445         ext4_fsblk_t    sb_block;
1446         char            *options = (char *) *data;
1447
1448         if (!options || strncmp(options, "sb=", 3) != 0)
1449                 return 1;       /* Default location */
1450
1451         options += 3;
1452         /* TODO: use simple_strtoll with >32bit ext4 */
1453         sb_block = simple_strtoul(options, &options, 0);
1454         if (*options && *options != ',') {
1455                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1456                        (char *) *data);
1457                 return 1;
1458         }
1459         if (*options == ',')
1460                 options++;
1461         *data = (void *) options;
1462
1463         return sb_block;
1464 }
1465
1466 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1467 static const char deprecated_msg[] =
1468         "Mount option \"%s\" will be removed by %s\n"
1469         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1470
1471 #ifdef CONFIG_QUOTA
1472 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1473 {
1474         struct ext4_sb_info *sbi = EXT4_SB(sb);
1475         char *qname;
1476         int ret = -1;
1477
1478         if (sb_any_quota_loaded(sb) &&
1479                 !sbi->s_qf_names[qtype]) {
1480                 ext4_msg(sb, KERN_ERR,
1481                         "Cannot change journaled "
1482                         "quota options when quota turned on");
1483                 return -1;
1484         }
1485         if (ext4_has_feature_quota(sb)) {
1486                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1487                          "ignored when QUOTA feature is enabled");
1488                 return 1;
1489         }
1490         qname = match_strdup(args);
1491         if (!qname) {
1492                 ext4_msg(sb, KERN_ERR,
1493                         "Not enough memory for storing quotafile name");
1494                 return -1;
1495         }
1496         if (sbi->s_qf_names[qtype]) {
1497                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1498                         ret = 1;
1499                 else
1500                         ext4_msg(sb, KERN_ERR,
1501                                  "%s quota file already specified",
1502                                  QTYPE2NAME(qtype));
1503                 goto errout;
1504         }
1505         if (strchr(qname, '/')) {
1506                 ext4_msg(sb, KERN_ERR,
1507                         "quotafile must be on filesystem root");
1508                 goto errout;
1509         }
1510         sbi->s_qf_names[qtype] = qname;
1511         set_opt(sb, QUOTA);
1512         return 1;
1513 errout:
1514         kfree(qname);
1515         return ret;
1516 }
1517
1518 static int clear_qf_name(struct super_block *sb, int qtype)
1519 {
1520
1521         struct ext4_sb_info *sbi = EXT4_SB(sb);
1522
1523         if (sb_any_quota_loaded(sb) &&
1524                 sbi->s_qf_names[qtype]) {
1525                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1526                         " when quota turned on");
1527                 return -1;
1528         }
1529         kfree(sbi->s_qf_names[qtype]);
1530         sbi->s_qf_names[qtype] = NULL;
1531         return 1;
1532 }
1533 #endif
1534
1535 #define MOPT_SET        0x0001
1536 #define MOPT_CLEAR      0x0002
1537 #define MOPT_NOSUPPORT  0x0004
1538 #define MOPT_EXPLICIT   0x0008
1539 #define MOPT_CLEAR_ERR  0x0010
1540 #define MOPT_GTE0       0x0020
1541 #ifdef CONFIG_QUOTA
1542 #define MOPT_Q          0
1543 #define MOPT_QFMT       0x0040
1544 #else
1545 #define MOPT_Q          MOPT_NOSUPPORT
1546 #define MOPT_QFMT       MOPT_NOSUPPORT
1547 #endif
1548 #define MOPT_DATAJ      0x0080
1549 #define MOPT_NO_EXT2    0x0100
1550 #define MOPT_NO_EXT3    0x0200
1551 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1552 #define MOPT_STRING     0x0400
1553
1554 static const struct mount_opts {
1555         int     token;
1556         int     mount_opt;
1557         int     flags;
1558 } ext4_mount_opts[] = {
1559         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1560         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1561         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1562         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1563         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1564         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1565         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1566          MOPT_EXT4_ONLY | MOPT_SET},
1567         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1568          MOPT_EXT4_ONLY | MOPT_CLEAR},
1569         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1570         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1571         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1572          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1573         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1574          MOPT_EXT4_ONLY | MOPT_CLEAR},
1575         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1576          MOPT_EXT4_ONLY | MOPT_CLEAR},
1577         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1578          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1579         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1580                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1581          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1582         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1583         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1584         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1585         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1586         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1587          MOPT_NO_EXT2},
1588         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1589          MOPT_NO_EXT2},
1590         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1591         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1592         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1593         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1594         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1595         {Opt_commit, 0, MOPT_GTE0},
1596         {Opt_max_batch_time, 0, MOPT_GTE0},
1597         {Opt_min_batch_time, 0, MOPT_GTE0},
1598         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1599         {Opt_init_itable, 0, MOPT_GTE0},
1600         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1601         {Opt_stripe, 0, MOPT_GTE0},
1602         {Opt_resuid, 0, MOPT_GTE0},
1603         {Opt_resgid, 0, MOPT_GTE0},
1604         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1605         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1606         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1607         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1608         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1609         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1610          MOPT_NO_EXT2 | MOPT_DATAJ},
1611         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1612         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1613 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1614         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1615         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1616 #else
1617         {Opt_acl, 0, MOPT_NOSUPPORT},
1618         {Opt_noacl, 0, MOPT_NOSUPPORT},
1619 #endif
1620         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1621         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1622         {Opt_debug_want_extra_isize, 0, MOPT_GTE0},
1623         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1624         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1625                                                         MOPT_SET | MOPT_Q},
1626         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1627                                                         MOPT_SET | MOPT_Q},
1628         {Opt_prjquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_PRJQUOTA,
1629                                                         MOPT_SET | MOPT_Q},
1630         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1631                        EXT4_MOUNT_GRPQUOTA | EXT4_MOUNT_PRJQUOTA),
1632                                                         MOPT_CLEAR | MOPT_Q},
1633         {Opt_usrjquota, 0, MOPT_Q},
1634         {Opt_grpjquota, 0, MOPT_Q},
1635         {Opt_offusrjquota, 0, MOPT_Q},
1636         {Opt_offgrpjquota, 0, MOPT_Q},
1637         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1638         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1639         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1640         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1641         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1642         {Opt_nombcache, EXT4_MOUNT_NO_MBCACHE, MOPT_SET},
1643         {Opt_err, 0, 0}
1644 };
1645
1646 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1647                             substring_t *args, unsigned long *journal_devnum,
1648                             unsigned int *journal_ioprio, int is_remount)
1649 {
1650         struct ext4_sb_info *sbi = EXT4_SB(sb);
1651         const struct mount_opts *m;
1652         kuid_t uid;
1653         kgid_t gid;
1654         int arg = 0;
1655
1656 #ifdef CONFIG_QUOTA
1657         if (token == Opt_usrjquota)
1658                 return set_qf_name(sb, USRQUOTA, &args[0]);
1659         else if (token == Opt_grpjquota)
1660                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1661         else if (token == Opt_offusrjquota)
1662                 return clear_qf_name(sb, USRQUOTA);
1663         else if (token == Opt_offgrpjquota)
1664                 return clear_qf_name(sb, GRPQUOTA);
1665 #endif
1666         switch (token) {
1667         case Opt_noacl:
1668         case Opt_nouser_xattr:
1669                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1670                 break;
1671         case Opt_sb:
1672                 return 1;       /* handled by get_sb_block() */
1673         case Opt_removed:
1674                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1675                 return 1;
1676         case Opt_abort:
1677                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1678                 return 1;
1679         case Opt_i_version:
1680                 sb->s_flags |= MS_I_VERSION;
1681                 return 1;
1682         case Opt_lazytime:
1683                 sb->s_flags |= MS_LAZYTIME;
1684                 return 1;
1685         case Opt_nolazytime:
1686                 sb->s_flags &= ~MS_LAZYTIME;
1687                 return 1;
1688         }
1689
1690         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1691                 if (token == m->token)
1692                         break;
1693
1694         if (m->token == Opt_err) {
1695                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1696                          "or missing value", opt);
1697                 return -1;
1698         }
1699
1700         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1701                 ext4_msg(sb, KERN_ERR,
1702                          "Mount option \"%s\" incompatible with ext2", opt);
1703                 return -1;
1704         }
1705         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1706                 ext4_msg(sb, KERN_ERR,
1707                          "Mount option \"%s\" incompatible with ext3", opt);
1708                 return -1;
1709         }
1710
1711         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1712                 return -1;
1713         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1714                 return -1;
1715         if (m->flags & MOPT_EXPLICIT) {
1716                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1717                         set_opt2(sb, EXPLICIT_DELALLOC);
1718                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1719                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1720                 } else
1721                         return -1;
1722         }
1723         if (m->flags & MOPT_CLEAR_ERR)
1724                 clear_opt(sb, ERRORS_MASK);
1725         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1726                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1727                          "options when quota turned on");
1728                 return -1;
1729         }
1730
1731         if (m->flags & MOPT_NOSUPPORT) {
1732                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1733         } else if (token == Opt_commit) {
1734                 if (arg == 0)
1735                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1736                 sbi->s_commit_interval = HZ * arg;
1737         } else if (token == Opt_debug_want_extra_isize) {
1738                 sbi->s_want_extra_isize = arg;
1739         } else if (token == Opt_max_batch_time) {
1740                 sbi->s_max_batch_time = arg;
1741         } else if (token == Opt_min_batch_time) {
1742                 sbi->s_min_batch_time = arg;
1743         } else if (token == Opt_inode_readahead_blks) {
1744                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1745                         ext4_msg(sb, KERN_ERR,
1746                                  "EXT4-fs: inode_readahead_blks must be "
1747                                  "0 or a power of 2 smaller than 2^31");
1748                         return -1;
1749                 }
1750                 sbi->s_inode_readahead_blks = arg;
1751         } else if (token == Opt_init_itable) {
1752                 set_opt(sb, INIT_INODE_TABLE);
1753                 if (!args->from)
1754                         arg = EXT4_DEF_LI_WAIT_MULT;
1755                 sbi->s_li_wait_mult = arg;
1756         } else if (token == Opt_max_dir_size_kb) {
1757                 sbi->s_max_dir_size_kb = arg;
1758         } else if (token == Opt_stripe) {
1759                 sbi->s_stripe = arg;
1760         } else if (token == Opt_resuid) {
1761                 uid = make_kuid(current_user_ns(), arg);
1762                 if (!uid_valid(uid)) {
1763                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1764                         return -1;
1765                 }
1766                 sbi->s_resuid = uid;
1767         } else if (token == Opt_resgid) {
1768                 gid = make_kgid(current_user_ns(), arg);
1769                 if (!gid_valid(gid)) {
1770                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1771                         return -1;
1772                 }
1773                 sbi->s_resgid = gid;
1774         } else if (token == Opt_journal_dev) {
1775                 if (is_remount) {
1776                         ext4_msg(sb, KERN_ERR,
1777                                  "Cannot specify journal on remount");
1778                         return -1;
1779                 }
1780                 *journal_devnum = arg;
1781         } else if (token == Opt_journal_path) {
1782                 char *journal_path;
1783                 struct inode *journal_inode;
1784                 struct path path;
1785                 int error;
1786
1787                 if (is_remount) {
1788                         ext4_msg(sb, KERN_ERR,
1789                                  "Cannot specify journal on remount");
1790                         return -1;
1791                 }
1792                 journal_path = match_strdup(&args[0]);
1793                 if (!journal_path) {
1794                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1795                                 "journal device string");
1796                         return -1;
1797                 }
1798
1799                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1800                 if (error) {
1801                         ext4_msg(sb, KERN_ERR, "error: could not find "
1802                                 "journal device path: error %d", error);
1803                         kfree(journal_path);
1804                         return -1;
1805                 }
1806
1807                 journal_inode = d_inode(path.dentry);
1808                 if (!S_ISBLK(journal_inode->i_mode)) {
1809                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1810                                 "is not a block device", journal_path);
1811                         path_put(&path);
1812                         kfree(journal_path);
1813                         return -1;
1814                 }
1815
1816                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1817                 path_put(&path);
1818                 kfree(journal_path);
1819         } else if (token == Opt_journal_ioprio) {
1820                 if (arg > 7) {
1821                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1822                                  " (must be 0-7)");
1823                         return -1;
1824                 }
1825                 *journal_ioprio =
1826                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1827         } else if (token == Opt_test_dummy_encryption) {
1828 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1829                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1830                 ext4_msg(sb, KERN_WARNING,
1831                          "Test dummy encryption mode enabled");
1832 #else
1833                 ext4_msg(sb, KERN_WARNING,
1834                          "Test dummy encryption mount option ignored");
1835 #endif
1836         } else if (m->flags & MOPT_DATAJ) {
1837                 if (is_remount) {
1838                         if (!sbi->s_journal)
1839                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1840                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1841                                 ext4_msg(sb, KERN_ERR,
1842                                          "Cannot change data mode on remount");
1843                                 return -1;
1844                         }
1845                 } else {
1846                         clear_opt(sb, DATA_FLAGS);
1847                         sbi->s_mount_opt |= m->mount_opt;
1848                 }
1849 #ifdef CONFIG_QUOTA
1850         } else if (m->flags & MOPT_QFMT) {
1851                 if (sb_any_quota_loaded(sb) &&
1852                     sbi->s_jquota_fmt != m->mount_opt) {
1853                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1854                                  "quota options when quota turned on");
1855                         return -1;
1856                 }
1857                 if (ext4_has_feature_quota(sb)) {
1858                         ext4_msg(sb, KERN_INFO,
1859                                  "Quota format mount options ignored "
1860                                  "when QUOTA feature is enabled");
1861                         return 1;
1862                 }
1863                 sbi->s_jquota_fmt = m->mount_opt;
1864 #endif
1865         } else if (token == Opt_dax) {
1866 #ifdef CONFIG_FS_DAX
1867                 ext4_msg(sb, KERN_WARNING,
1868                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1869                         sbi->s_mount_opt |= m->mount_opt;
1870 #else
1871                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1872                 return -1;
1873 #endif
1874         } else if (token == Opt_data_err_abort) {
1875                 sbi->s_mount_opt |= m->mount_opt;
1876         } else if (token == Opt_data_err_ignore) {
1877                 sbi->s_mount_opt &= ~m->mount_opt;
1878         } else {
1879                 if (!args->from)
1880                         arg = 1;
1881                 if (m->flags & MOPT_CLEAR)
1882                         arg = !arg;
1883                 else if (unlikely(!(m->flags & MOPT_SET))) {
1884                         ext4_msg(sb, KERN_WARNING,
1885                                  "buggy handling of option %s", opt);
1886                         WARN_ON(1);
1887                         return -1;
1888                 }
1889                 if (arg != 0)
1890                         sbi->s_mount_opt |= m->mount_opt;
1891                 else
1892                         sbi->s_mount_opt &= ~m->mount_opt;
1893         }
1894         return 1;
1895 }
1896
1897 static int parse_options(char *options, struct super_block *sb,
1898                          unsigned long *journal_devnum,
1899                          unsigned int *journal_ioprio,
1900                          int is_remount)
1901 {
1902         struct ext4_sb_info *sbi = EXT4_SB(sb);
1903         char *p;
1904         substring_t args[MAX_OPT_ARGS];
1905         int token;
1906
1907         if (!options)
1908                 return 1;
1909
1910         while ((p = strsep(&options, ",")) != NULL) {
1911                 if (!*p)
1912                         continue;
1913                 /*
1914                  * Initialize args struct so we know whether arg was
1915                  * found; some options take optional arguments.
1916                  */
1917                 args[0].to = args[0].from = NULL;
1918                 token = match_token(p, tokens, args);
1919                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1920                                      journal_ioprio, is_remount) < 0)
1921                         return 0;
1922         }
1923 #ifdef CONFIG_QUOTA
1924         /*
1925          * We do the test below only for project quotas. 'usrquota' and
1926          * 'grpquota' mount options are allowed even without quota feature
1927          * to support legacy quotas in quota files.
1928          */
1929         if (test_opt(sb, PRJQUOTA) && !ext4_has_feature_project(sb)) {
1930                 ext4_msg(sb, KERN_ERR, "Project quota feature not enabled. "
1931                          "Cannot enable project quota enforcement.");
1932                 return 0;
1933         }
1934         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1935                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1936                         clear_opt(sb, USRQUOTA);
1937
1938                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1939                         clear_opt(sb, GRPQUOTA);
1940
1941                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1942                         ext4_msg(sb, KERN_ERR, "old and new quota "
1943                                         "format mixing");
1944                         return 0;
1945                 }
1946
1947                 if (!sbi->s_jquota_fmt) {
1948                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1949                                         "not specified");
1950                         return 0;
1951                 }
1952         }
1953 #endif
1954         if (test_opt(sb, DIOREAD_NOLOCK)) {
1955                 int blocksize =
1956                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1957
1958                 if (blocksize < PAGE_SIZE) {
1959                         ext4_msg(sb, KERN_ERR, "can't mount with "
1960                                  "dioread_nolock if block size != PAGE_SIZE");
1961                         return 0;
1962                 }
1963         }
1964         return 1;
1965 }
1966
1967 static inline void ext4_show_quota_options(struct seq_file *seq,
1968                                            struct super_block *sb)
1969 {
1970 #if defined(CONFIG_QUOTA)
1971         struct ext4_sb_info *sbi = EXT4_SB(sb);
1972
1973         if (sbi->s_jquota_fmt) {
1974                 char *fmtname = "";
1975
1976                 switch (sbi->s_jquota_fmt) {
1977                 case QFMT_VFS_OLD:
1978                         fmtname = "vfsold";
1979                         break;
1980                 case QFMT_VFS_V0:
1981                         fmtname = "vfsv0";
1982                         break;
1983                 case QFMT_VFS_V1:
1984                         fmtname = "vfsv1";
1985                         break;
1986                 }
1987                 seq_printf(seq, ",jqfmt=%s", fmtname);
1988         }
1989
1990         if (sbi->s_qf_names[USRQUOTA])
1991                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1992
1993         if (sbi->s_qf_names[GRPQUOTA])
1994                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1995 #endif
1996 }
1997
1998 static const char *token2str(int token)
1999 {
2000         const struct match_token *t;
2001
2002         for (t = tokens; t->token != Opt_err; t++)
2003                 if (t->token == token && !strchr(t->pattern, '='))
2004                         break;
2005         return t->pattern;
2006 }
2007
2008 /*
2009  * Show an option if
2010  *  - it's set to a non-default value OR
2011  *  - if the per-sb default is different from the global default
2012  */
2013 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
2014                               int nodefs)
2015 {
2016         struct ext4_sb_info *sbi = EXT4_SB(sb);
2017         struct ext4_super_block *es = sbi->s_es;
2018         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
2019         const struct mount_opts *m;
2020         char sep = nodefs ? '\n' : ',';
2021
2022 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
2023 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
2024
2025         if (sbi->s_sb_block != 1)
2026                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
2027
2028         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
2029                 int want_set = m->flags & MOPT_SET;
2030                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
2031                     (m->flags & MOPT_CLEAR_ERR))
2032                         continue;
2033                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
2034                         continue; /* skip if same as the default */
2035                 if ((want_set &&
2036                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
2037                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
2038                         continue; /* select Opt_noFoo vs Opt_Foo */
2039                 SEQ_OPTS_PRINT("%s", token2str(m->token));
2040         }
2041
2042         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
2043             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
2044                 SEQ_OPTS_PRINT("resuid=%u",
2045                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
2046         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
2047             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
2048                 SEQ_OPTS_PRINT("resgid=%u",
2049                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
2050         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
2051         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
2052                 SEQ_OPTS_PUTS("errors=remount-ro");
2053         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
2054                 SEQ_OPTS_PUTS("errors=continue");
2055         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
2056                 SEQ_OPTS_PUTS("errors=panic");
2057         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
2058                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
2059         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
2060                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
2061         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
2062                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
2063         if (sb->s_flags & MS_I_VERSION)
2064                 SEQ_OPTS_PUTS("i_version");
2065         if (nodefs || sbi->s_stripe)
2066                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
2067         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
2068                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
2069                         SEQ_OPTS_PUTS("data=journal");
2070                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
2071                         SEQ_OPTS_PUTS("data=ordered");
2072                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
2073                         SEQ_OPTS_PUTS("data=writeback");
2074         }
2075         if (nodefs ||
2076             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
2077                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
2078                                sbi->s_inode_readahead_blks);
2079
2080         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
2081                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
2082                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
2083         if (nodefs || sbi->s_max_dir_size_kb)
2084                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
2085         if (test_opt(sb, DATA_ERR_ABORT))
2086                 SEQ_OPTS_PUTS("data_err=abort");
2087
2088         ext4_show_quota_options(seq, sb);
2089         return 0;
2090 }
2091
2092 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
2093 {
2094         return _ext4_show_options(seq, root->d_sb, 0);
2095 }
2096
2097 int ext4_seq_options_show(struct seq_file *seq, void *offset)
2098 {
2099         struct super_block *sb = seq->private;
2100         int rc;
2101
2102         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
2103         rc = _ext4_show_options(seq, sb, 1);
2104         seq_puts(seq, "\n");
2105         return rc;
2106 }
2107
2108 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
2109                             int read_only)
2110 {
2111         struct ext4_sb_info *sbi = EXT4_SB(sb);
2112         int res = 0;
2113
2114         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
2115                 ext4_msg(sb, KERN_ERR, "revision level too high, "
2116                          "forcing read-only mode");
2117                 res = MS_RDONLY;
2118         }
2119         if (read_only)
2120                 goto done;
2121         if (!(sbi->s_mount_state & EXT4_VALID_FS))
2122                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
2123                          "running e2fsck is recommended");
2124         else if (sbi->s_mount_state & EXT4_ERROR_FS)
2125                 ext4_msg(sb, KERN_WARNING,
2126                          "warning: mounting fs with errors, "
2127                          "running e2fsck is recommended");
2128         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
2129                  le16_to_cpu(es->s_mnt_count) >=
2130                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
2131                 ext4_msg(sb, KERN_WARNING,
2132                          "warning: maximal mount count reached, "
2133                          "running e2fsck is recommended");
2134         else if (le32_to_cpu(es->s_checkinterval) &&
2135                 (le32_to_cpu(es->s_lastcheck) +
2136                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
2137                 ext4_msg(sb, KERN_WARNING,
2138                          "warning: checktime reached, "
2139                          "running e2fsck is recommended");
2140         if (!sbi->s_journal)
2141                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
2142         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
2143                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
2144         le16_add_cpu(&es->s_mnt_count, 1);
2145         es->s_mtime = cpu_to_le32(get_seconds());
2146         ext4_update_dynamic_rev(sb);
2147         if (sbi->s_journal)
2148                 ext4_set_feature_journal_needs_recovery(sb);
2149
2150         ext4_commit_super(sb, 1);
2151 done:
2152         if (test_opt(sb, DEBUG))
2153                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
2154                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
2155                         sb->s_blocksize,
2156                         sbi->s_groups_count,
2157                         EXT4_BLOCKS_PER_GROUP(sb),
2158                         EXT4_INODES_PER_GROUP(sb),
2159                         sbi->s_mount_opt, sbi->s_mount_opt2);
2160
2161         cleancache_init_fs(sb);
2162         return res;
2163 }
2164
2165 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2166 {
2167         struct ext4_sb_info *sbi = EXT4_SB(sb);
2168         struct flex_groups *new_groups;
2169         int size;
2170
2171         if (!sbi->s_log_groups_per_flex)
2172                 return 0;
2173
2174         size = ext4_flex_group(sbi, ngroup - 1) + 1;
2175         if (size <= sbi->s_flex_groups_allocated)
2176                 return 0;
2177
2178         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2179         new_groups = kvzalloc(size, GFP_KERNEL);
2180         if (!new_groups) {
2181                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2182                          size / (int) sizeof(struct flex_groups));
2183                 return -ENOMEM;
2184         }
2185
2186         if (sbi->s_flex_groups) {
2187                 memcpy(new_groups, sbi->s_flex_groups,
2188                        (sbi->s_flex_groups_allocated *
2189                         sizeof(struct flex_groups)));
2190                 kvfree(sbi->s_flex_groups);
2191         }
2192         sbi->s_flex_groups = new_groups;
2193         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2194         return 0;
2195 }
2196
2197 static int ext4_fill_flex_info(struct super_block *sb)
2198 {
2199         struct ext4_sb_info *sbi = EXT4_SB(sb);
2200         struct ext4_group_desc *gdp = NULL;
2201         ext4_group_t flex_group;
2202         int i, err;
2203
2204         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2205         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2206                 sbi->s_log_groups_per_flex = 0;
2207                 return 1;
2208         }
2209
2210         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2211         if (err)
2212                 goto failed;
2213
2214         for (i = 0; i < sbi->s_groups_count; i++) {
2215                 gdp = ext4_get_group_desc(sb, i, NULL);
2216
2217                 flex_group = ext4_flex_group(sbi, i);
2218                 atomic_add(ext4_free_inodes_count(sb, gdp),
2219                            &sbi->s_flex_groups[flex_group].free_inodes);
2220                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2221                              &sbi->s_flex_groups[flex_group].free_clusters);
2222                 atomic_add(ext4_used_dirs_count(sb, gdp),
2223                            &sbi->s_flex_groups[flex_group].used_dirs);
2224         }
2225
2226         return 1;
2227 failed:
2228         return 0;
2229 }
2230
2231 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2232                                    struct ext4_group_desc *gdp)
2233 {
2234         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2235         __u16 crc = 0;
2236         __le32 le_group = cpu_to_le32(block_group);
2237         struct ext4_sb_info *sbi = EXT4_SB(sb);
2238
2239         if (ext4_has_metadata_csum(sbi->s_sb)) {
2240                 /* Use new metadata_csum algorithm */
2241                 __u32 csum32;
2242                 __u16 dummy_csum = 0;
2243
2244                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2245                                      sizeof(le_group));
2246                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2247                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2248                                      sizeof(dummy_csum));
2249                 offset += sizeof(dummy_csum);
2250                 if (offset < sbi->s_desc_size)
2251                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2252                                              sbi->s_desc_size - offset);
2253
2254                 crc = csum32 & 0xFFFF;
2255                 goto out;
2256         }
2257
2258         /* old crc16 code */
2259         if (!ext4_has_feature_gdt_csum(sb))
2260                 return 0;
2261
2262         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2263         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2264         crc = crc16(crc, (__u8 *)gdp, offset);
2265         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2266         /* for checksum of struct ext4_group_desc do the rest...*/
2267         if (ext4_has_feature_64bit(sb) &&
2268             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2269                 crc = crc16(crc, (__u8 *)gdp + offset,
2270                             le16_to_cpu(sbi->s_es->s_desc_size) -
2271                                 offset);
2272
2273 out:
2274         return cpu_to_le16(crc);
2275 }
2276
2277 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2278                                 struct ext4_group_desc *gdp)
2279 {
2280         if (ext4_has_group_desc_csum(sb) &&
2281             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2282                 return 0;
2283
2284         return 1;
2285 }
2286
2287 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2288                               struct ext4_group_desc *gdp)
2289 {
2290         if (!ext4_has_group_desc_csum(sb))
2291                 return;
2292         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2293 }
2294
2295 /* Called at mount-time, super-block is locked */
2296 static int ext4_check_descriptors(struct super_block *sb,
2297                                   ext4_fsblk_t sb_block,
2298                                   ext4_group_t *first_not_zeroed)
2299 {
2300         struct ext4_sb_info *sbi = EXT4_SB(sb);
2301         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2302         ext4_fsblk_t last_block;
2303         ext4_fsblk_t block_bitmap;
2304         ext4_fsblk_t inode_bitmap;
2305         ext4_fsblk_t inode_table;
2306         int flexbg_flag = 0;
2307         ext4_group_t i, grp = sbi->s_groups_count;
2308
2309         if (ext4_has_feature_flex_bg(sb))
2310                 flexbg_flag = 1;
2311
2312         ext4_debug("Checking group descriptors");
2313
2314         for (i = 0; i < sbi->s_groups_count; i++) {
2315                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2316
2317                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2318                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2319                 else
2320                         last_block = first_block +
2321                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2322
2323                 if ((grp == sbi->s_groups_count) &&
2324                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2325                         grp = i;
2326
2327                 block_bitmap = ext4_block_bitmap(sb, gdp);
2328                 if (block_bitmap == sb_block) {
2329                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2330                                  "Block bitmap for group %u overlaps "
2331                                  "superblock", i);
2332                 }
2333                 if (block_bitmap < first_block || block_bitmap > last_block) {
2334                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2335                                "Block bitmap for group %u not in group "
2336                                "(block %llu)!", i, block_bitmap);
2337                         return 0;
2338                 }
2339                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2340                 if (inode_bitmap == sb_block) {
2341                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2342                                  "Inode bitmap for group %u overlaps "
2343                                  "superblock", i);
2344                 }
2345                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2346                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2347                                "Inode bitmap for group %u not in group "
2348                                "(block %llu)!", i, inode_bitmap);
2349                         return 0;
2350                 }
2351                 inode_table = ext4_inode_table(sb, gdp);
2352                 if (inode_table == sb_block) {
2353                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2354                                  "Inode table for group %u overlaps "
2355                                  "superblock", i);
2356                 }
2357                 if (inode_table < first_block ||
2358                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2359                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2360                                "Inode table for group %u not in group "
2361                                "(block %llu)!", i, inode_table);
2362                         return 0;
2363                 }
2364                 ext4_lock_group(sb, i);
2365                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2366                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2367                                  "Checksum for group %u failed (%u!=%u)",
2368                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2369                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2370                         if (!(sb->s_flags & MS_RDONLY)) {
2371                                 ext4_unlock_group(sb, i);
2372                                 return 0;
2373                         }
2374                 }
2375                 ext4_unlock_group(sb, i);
2376                 if (!flexbg_flag)
2377                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2378         }
2379         if (NULL != first_not_zeroed)
2380                 *first_not_zeroed = grp;
2381         return 1;
2382 }
2383
2384 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2385  * the superblock) which were deleted from all directories, but held open by
2386  * a process at the time of a crash.  We walk the list and try to delete these
2387  * inodes at recovery time (only with a read-write filesystem).
2388  *
2389  * In order to keep the orphan inode chain consistent during traversal (in
2390  * case of crash during recovery), we link each inode into the superblock
2391  * orphan list_head and handle it the same way as an inode deletion during
2392  * normal operation (which journals the operations for us).
2393  *
2394  * We only do an iget() and an iput() on each inode, which is very safe if we
2395  * accidentally point at an in-use or already deleted inode.  The worst that
2396  * can happen in this case is that we get a "bit already cleared" message from
2397  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2398  * e2fsck was run on this filesystem, and it must have already done the orphan
2399  * inode cleanup for us, so we can safely abort without any further action.
2400  */
2401 static void ext4_orphan_cleanup(struct super_block *sb,
2402                                 struct ext4_super_block *es)
2403 {
2404         unsigned int s_flags = sb->s_flags;
2405         int ret, nr_orphans = 0, nr_truncates = 0;
2406 #ifdef CONFIG_QUOTA
2407         int i;
2408 #endif
2409         if (!es->s_last_orphan) {
2410                 jbd_debug(4, "no orphan inodes to clean up\n");
2411                 return;
2412         }
2413
2414         if (bdev_read_only(sb->s_bdev)) {
2415                 ext4_msg(sb, KERN_ERR, "write access "
2416                         "unavailable, skipping orphan cleanup");
2417                 return;
2418         }
2419
2420         /* Check if feature set would not allow a r/w mount */
2421         if (!ext4_feature_set_ok(sb, 0)) {
2422                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2423                          "unknown ROCOMPAT features");
2424                 return;
2425         }
2426
2427         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2428                 /* don't clear list on RO mount w/ errors */
2429                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2430                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2431                                   "clearing orphan list.\n");
2432                         es->s_last_orphan = 0;
2433                 }
2434                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2435                 return;
2436         }
2437
2438         if (s_flags & MS_RDONLY) {
2439                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2440                 sb->s_flags &= ~MS_RDONLY;
2441         }
2442 #ifdef CONFIG_QUOTA
2443         /* Needed for iput() to work correctly and not trash data */
2444         sb->s_flags |= MS_ACTIVE;
2445         /* Turn on quotas so that they are updated correctly */
2446         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2447                 if (EXT4_SB(sb)->s_qf_names[i]) {
2448                         int ret = ext4_quota_on_mount(sb, i);
2449                         if (ret < 0)
2450                                 ext4_msg(sb, KERN_ERR,
2451                                         "Cannot turn on journaled "
2452                                         "quota: error %d", ret);
2453                 }
2454         }
2455 #endif
2456
2457         while (es->s_last_orphan) {
2458                 struct inode *inode;
2459
2460                 /*
2461                  * We may have encountered an error during cleanup; if
2462                  * so, skip the rest.
2463                  */
2464                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2465                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2466                         es->s_last_orphan = 0;
2467                         break;
2468                 }
2469
2470                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2471                 if (IS_ERR(inode)) {
2472                         es->s_last_orphan = 0;
2473                         break;
2474                 }
2475
2476                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2477                 dquot_initialize(inode);
2478                 if (inode->i_nlink) {
2479                         if (test_opt(sb, DEBUG))
2480                                 ext4_msg(sb, KERN_DEBUG,
2481                                         "%s: truncating inode %lu to %lld bytes",
2482                                         __func__, inode->i_ino, inode->i_size);
2483                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2484                                   inode->i_ino, inode->i_size);
2485                         inode_lock(inode);
2486                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2487                         ret = ext4_truncate(inode);
2488                         if (ret)
2489                                 ext4_std_error(inode->i_sb, ret);
2490                         inode_unlock(inode);
2491                         nr_truncates++;
2492                 } else {
2493                         if (test_opt(sb, DEBUG))
2494                                 ext4_msg(sb, KERN_DEBUG,
2495                                         "%s: deleting unreferenced inode %lu",
2496                                         __func__, inode->i_ino);
2497                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2498                                   inode->i_ino);
2499                         nr_orphans++;
2500                 }
2501                 iput(inode);  /* The delete magic happens here! */
2502         }
2503
2504 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2505
2506         if (nr_orphans)
2507                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2508                        PLURAL(nr_orphans));
2509         if (nr_truncates)
2510                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2511                        PLURAL(nr_truncates));
2512 #ifdef CONFIG_QUOTA
2513         /* Turn quotas off */
2514         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2515                 if (sb_dqopt(sb)->files[i])
2516                         dquot_quota_off(sb, i);
2517         }
2518 #endif
2519         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2520 }
2521
2522 /*
2523  * Maximal extent format file size.
2524  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2525  * extent format containers, within a sector_t, and within i_blocks
2526  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2527  * so that won't be a limiting factor.
2528  *
2529  * However there is other limiting factor. We do store extents in the form
2530  * of starting block and length, hence the resulting length of the extent
2531  * covering maximum file size must fit into on-disk format containers as
2532  * well. Given that length is always by 1 unit bigger than max unit (because
2533  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2534  *
2535  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2536  */
2537 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2538 {
2539         loff_t res;
2540         loff_t upper_limit = MAX_LFS_FILESIZE;
2541
2542         /* small i_blocks in vfs inode? */
2543         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2544                 /*
2545                  * CONFIG_LBDAF is not enabled implies the inode
2546                  * i_block represent total blocks in 512 bytes
2547                  * 32 == size of vfs inode i_blocks * 8
2548                  */
2549                 upper_limit = (1LL << 32) - 1;
2550
2551                 /* total blocks in file system block size */
2552                 upper_limit >>= (blkbits - 9);
2553                 upper_limit <<= blkbits;
2554         }
2555
2556         /*
2557          * 32-bit extent-start container, ee_block. We lower the maxbytes
2558          * by one fs block, so ee_len can cover the extent of maximum file
2559          * size
2560          */
2561         res = (1LL << 32) - 1;
2562         res <<= blkbits;
2563
2564         /* Sanity check against vm- & vfs- imposed limits */
2565         if (res > upper_limit)
2566                 res = upper_limit;
2567
2568         return res;
2569 }
2570
2571 /*
2572  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2573  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2574  * We need to be 1 filesystem block less than the 2^48 sector limit.
2575  */
2576 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2577 {
2578         loff_t res = EXT4_NDIR_BLOCKS;
2579         int meta_blocks;
2580         loff_t upper_limit;
2581         /* This is calculated to be the largest file size for a dense, block
2582          * mapped file such that the file's total number of 512-byte sectors,
2583          * including data and all indirect blocks, does not exceed (2^48 - 1).
2584          *
2585          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2586          * number of 512-byte sectors of the file.
2587          */
2588
2589         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2590                 /*
2591                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2592                  * the inode i_block field represents total file blocks in
2593                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2594                  */
2595                 upper_limit = (1LL << 32) - 1;
2596
2597                 /* total blocks in file system block size */
2598                 upper_limit >>= (bits - 9);
2599
2600         } else {
2601                 /*
2602                  * We use 48 bit ext4_inode i_blocks
2603                  * With EXT4_HUGE_FILE_FL set the i_blocks
2604                  * represent total number of blocks in
2605                  * file system block size
2606                  */
2607                 upper_limit = (1LL << 48) - 1;
2608
2609         }
2610
2611         /* indirect blocks */
2612         meta_blocks = 1;
2613         /* double indirect blocks */
2614         meta_blocks += 1 + (1LL << (bits-2));
2615         /* tripple indirect blocks */
2616         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2617
2618         upper_limit -= meta_blocks;
2619         upper_limit <<= bits;
2620
2621         res += 1LL << (bits-2);
2622         res += 1LL << (2*(bits-2));
2623         res += 1LL << (3*(bits-2));
2624         res <<= bits;
2625         if (res > upper_limit)
2626                 res = upper_limit;
2627
2628         if (res > MAX_LFS_FILESIZE)
2629                 res = MAX_LFS_FILESIZE;
2630
2631         return res;
2632 }
2633
2634 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2635                                    ext4_fsblk_t logical_sb_block, int nr)
2636 {
2637         struct ext4_sb_info *sbi = EXT4_SB(sb);
2638         ext4_group_t bg, first_meta_bg;
2639         int has_super = 0;
2640
2641         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2642
2643         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2644                 return logical_sb_block + nr + 1;
2645         bg = sbi->s_desc_per_block * nr;
2646         if (ext4_bg_has_super(sb, bg))
2647                 has_super = 1;
2648
2649         /*
2650          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2651          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2652          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2653          * compensate.
2654          */
2655         if (sb->s_blocksize == 1024 && nr == 0 &&
2656             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2657                 has_super++;
2658
2659         return (has_super + ext4_group_first_block_no(sb, bg));
2660 }
2661
2662 /**
2663  * ext4_get_stripe_size: Get the stripe size.
2664  * @sbi: In memory super block info
2665  *
2666  * If we have specified it via mount option, then
2667  * use the mount option value. If the value specified at mount time is
2668  * greater than the blocks per group use the super block value.
2669  * If the super block value is greater than blocks per group return 0.
2670  * Allocator needs it be less than blocks per group.
2671  *
2672  */
2673 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2674 {
2675         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2676         unsigned long stripe_width =
2677                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2678         int ret;
2679
2680         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2681                 ret = sbi->s_stripe;
2682         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2683                 ret = stripe_width;
2684         else if (stride && stride <= sbi->s_blocks_per_group)
2685                 ret = stride;
2686         else
2687                 ret = 0;
2688
2689         /*
2690          * If the stripe width is 1, this makes no sense and
2691          * we set it to 0 to turn off stripe handling code.
2692          */
2693         if (ret <= 1)
2694                 ret = 0;
2695
2696         return ret;
2697 }
2698
2699 /*
2700  * Check whether this filesystem can be mounted based on
2701  * the features present and the RDONLY/RDWR mount requested.
2702  * Returns 1 if this filesystem can be mounted as requested,
2703  * 0 if it cannot be.
2704  */
2705 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2706 {
2707         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2708                 ext4_msg(sb, KERN_ERR,
2709                         "Couldn't mount because of "
2710                         "unsupported optional features (%x)",
2711                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2712                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2713                 return 0;
2714         }
2715
2716         if (readonly)
2717                 return 1;
2718
2719         if (ext4_has_feature_readonly(sb)) {
2720                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2721                 sb->s_flags |= MS_RDONLY;
2722                 return 1;
2723         }
2724
2725         /* Check that feature set is OK for a read-write mount */
2726         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2727                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2728                          "unsupported optional features (%x)",
2729                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2730                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2731                 return 0;
2732         }
2733         /*
2734          * Large file size enabled file system can only be mounted
2735          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2736          */
2737         if (ext4_has_feature_huge_file(sb)) {
2738                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2739                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2740                                  "cannot be mounted RDWR without "
2741                                  "CONFIG_LBDAF");
2742                         return 0;
2743                 }
2744         }
2745         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2746                 ext4_msg(sb, KERN_ERR,
2747                          "Can't support bigalloc feature without "
2748                          "extents feature\n");
2749                 return 0;
2750         }
2751
2752 #ifndef CONFIG_QUOTA
2753         if (ext4_has_feature_quota(sb) && !readonly) {
2754                 ext4_msg(sb, KERN_ERR,
2755                          "Filesystem with quota feature cannot be mounted RDWR "
2756                          "without CONFIG_QUOTA");
2757                 return 0;
2758         }
2759         if (ext4_has_feature_project(sb) && !readonly) {
2760                 ext4_msg(sb, KERN_ERR,
2761                          "Filesystem with project quota feature cannot be mounted RDWR "
2762                          "without CONFIG_QUOTA");
2763                 return 0;
2764         }
2765 #endif  /* CONFIG_QUOTA */
2766         return 1;
2767 }
2768
2769 /*
2770  * This function is called once a day if we have errors logged
2771  * on the file system
2772  */
2773 static void print_daily_error_info(unsigned long arg)
2774 {
2775         struct super_block *sb = (struct super_block *) arg;
2776         struct ext4_sb_info *sbi;
2777         struct ext4_super_block *es;
2778
2779         sbi = EXT4_SB(sb);
2780         es = sbi->s_es;
2781
2782         if (es->s_error_count)
2783                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2784                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2785                          le32_to_cpu(es->s_error_count));
2786         if (es->s_first_error_time) {
2787                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2788                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2789                        (int) sizeof(es->s_first_error_func),
2790                        es->s_first_error_func,
2791                        le32_to_cpu(es->s_first_error_line));
2792                 if (es->s_first_error_ino)
2793                         printk(KERN_CONT ": inode %u",
2794                                le32_to_cpu(es->s_first_error_ino));
2795                 if (es->s_first_error_block)
2796                         printk(KERN_CONT ": block %llu", (unsigned long long)
2797                                le64_to_cpu(es->s_first_error_block));
2798                 printk(KERN_CONT "\n");
2799         }
2800         if (es->s_last_error_time) {
2801                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2802                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2803                        (int) sizeof(es->s_last_error_func),
2804                        es->s_last_error_func,
2805                        le32_to_cpu(es->s_last_error_line));
2806                 if (es->s_last_error_ino)
2807                         printk(KERN_CONT ": inode %u",
2808                                le32_to_cpu(es->s_last_error_ino));
2809                 if (es->s_last_error_block)
2810                         printk(KERN_CONT ": block %llu", (unsigned long long)
2811                                le64_to_cpu(es->s_last_error_block));
2812                 printk(KERN_CONT "\n");
2813         }
2814         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2815 }
2816
2817 /* Find next suitable group and run ext4_init_inode_table */
2818 static int ext4_run_li_request(struct ext4_li_request *elr)
2819 {
2820         struct ext4_group_desc *gdp = NULL;
2821         ext4_group_t group, ngroups;
2822         struct super_block *sb;
2823         unsigned long timeout = 0;
2824         int ret = 0;
2825
2826         sb = elr->lr_super;
2827         ngroups = EXT4_SB(sb)->s_groups_count;
2828
2829         for (group = elr->lr_next_group; group < ngroups; group++) {
2830                 gdp = ext4_get_group_desc(sb, group, NULL);
2831                 if (!gdp) {
2832                         ret = 1;
2833                         break;
2834                 }
2835
2836                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2837                         break;
2838         }
2839
2840         if (group >= ngroups)
2841                 ret = 1;
2842
2843         if (!ret) {
2844                 timeout = jiffies;
2845                 ret = ext4_init_inode_table(sb, group,
2846                                             elr->lr_timeout ? 0 : 1);
2847                 if (elr->lr_timeout == 0) {
2848                         timeout = (jiffies - timeout) *
2849                                   elr->lr_sbi->s_li_wait_mult;
2850                         elr->lr_timeout = timeout;
2851                 }
2852                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2853                 elr->lr_next_group = group + 1;
2854         }
2855         return ret;
2856 }
2857
2858 /*
2859  * Remove lr_request from the list_request and free the
2860  * request structure. Should be called with li_list_mtx held
2861  */
2862 static void ext4_remove_li_request(struct ext4_li_request *elr)
2863 {
2864         struct ext4_sb_info *sbi;
2865
2866         if (!elr)
2867                 return;
2868
2869         sbi = elr->lr_sbi;
2870
2871         list_del(&elr->lr_request);
2872         sbi->s_li_request = NULL;
2873         kfree(elr);
2874 }
2875
2876 static void ext4_unregister_li_request(struct super_block *sb)
2877 {
2878         mutex_lock(&ext4_li_mtx);
2879         if (!ext4_li_info) {
2880                 mutex_unlock(&ext4_li_mtx);
2881                 return;
2882         }
2883
2884         mutex_lock(&ext4_li_info->li_list_mtx);
2885         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2886         mutex_unlock(&ext4_li_info->li_list_mtx);
2887         mutex_unlock(&ext4_li_mtx);
2888 }
2889
2890 static struct task_struct *ext4_lazyinit_task;
2891
2892 /*
2893  * This is the function where ext4lazyinit thread lives. It walks
2894  * through the request list searching for next scheduled filesystem.
2895  * When such a fs is found, run the lazy initialization request
2896  * (ext4_rn_li_request) and keep track of the time spend in this
2897  * function. Based on that time we compute next schedule time of
2898  * the request. When walking through the list is complete, compute
2899  * next waking time and put itself into sleep.
2900  */
2901 static int ext4_lazyinit_thread(void *arg)
2902 {
2903         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2904         struct list_head *pos, *n;
2905         struct ext4_li_request *elr;
2906         unsigned long next_wakeup, cur;
2907
2908         BUG_ON(NULL == eli);
2909
2910 cont_thread:
2911         while (true) {
2912                 next_wakeup = MAX_JIFFY_OFFSET;
2913
2914                 mutex_lock(&eli->li_list_mtx);
2915                 if (list_empty(&eli->li_request_list)) {
2916                         mutex_unlock(&eli->li_list_mtx);
2917                         goto exit_thread;
2918                 }
2919                 list_for_each_safe(pos, n, &eli->li_request_list) {
2920                         int err = 0;
2921                         int progress = 0;
2922                         elr = list_entry(pos, struct ext4_li_request,
2923                                          lr_request);
2924
2925                         if (time_before(jiffies, elr->lr_next_sched)) {
2926                                 if (time_before(elr->lr_next_sched, next_wakeup))
2927                                         next_wakeup = elr->lr_next_sched;
2928                                 continue;
2929                         }
2930                         if (down_read_trylock(&elr->lr_super->s_umount)) {
2931                                 if (sb_start_write_trylock(elr->lr_super)) {
2932                                         progress = 1;
2933                                         /*
2934                                          * We hold sb->s_umount, sb can not
2935                                          * be removed from the list, it is
2936                                          * now safe to drop li_list_mtx
2937                                          */
2938                                         mutex_unlock(&eli->li_list_mtx);
2939                                         err = ext4_run_li_request(elr);
2940                                         sb_end_write(elr->lr_super);
2941                                         mutex_lock(&eli->li_list_mtx);
2942                                         n = pos->next;
2943                                 }
2944                                 up_read((&elr->lr_super->s_umount));
2945                         }
2946                         /* error, remove the lazy_init job */
2947                         if (err) {
2948                                 ext4_remove_li_request(elr);
2949                                 continue;
2950                         }
2951                         if (!progress) {
2952                                 elr->lr_next_sched = jiffies +
2953                                         (prandom_u32()
2954                                          % (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2955                         }
2956                         if (time_before(elr->lr_next_sched, next_wakeup))
2957                                 next_wakeup = elr->lr_next_sched;
2958                 }
2959                 mutex_unlock(&eli->li_list_mtx);
2960
2961                 try_to_freeze();
2962
2963                 cur = jiffies;
2964                 if ((time_after_eq(cur, next_wakeup)) ||
2965                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2966                         cond_resched();
2967                         continue;
2968                 }
2969
2970                 schedule_timeout_interruptible(next_wakeup - cur);
2971
2972                 if (kthread_should_stop()) {
2973                         ext4_clear_request_list();
2974                         goto exit_thread;
2975                 }
2976         }
2977
2978 exit_thread:
2979         /*
2980          * It looks like the request list is empty, but we need
2981          * to check it under the li_list_mtx lock, to prevent any
2982          * additions into it, and of course we should lock ext4_li_mtx
2983          * to atomically free the list and ext4_li_info, because at
2984          * this point another ext4 filesystem could be registering
2985          * new one.
2986          */
2987         mutex_lock(&ext4_li_mtx);
2988         mutex_lock(&eli->li_list_mtx);
2989         if (!list_empty(&eli->li_request_list)) {
2990                 mutex_unlock(&eli->li_list_mtx);
2991                 mutex_unlock(&ext4_li_mtx);
2992                 goto cont_thread;
2993         }
2994         mutex_unlock(&eli->li_list_mtx);
2995         kfree(ext4_li_info);
2996         ext4_li_info = NULL;
2997         mutex_unlock(&ext4_li_mtx);
2998
2999         return 0;
3000 }
3001
3002 static void ext4_clear_request_list(void)
3003 {
3004         struct list_head *pos, *n;
3005         struct ext4_li_request *elr;
3006
3007         mutex_lock(&ext4_li_info->li_list_mtx);
3008         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3009                 elr = list_entry(pos, struct ext4_li_request,
3010                                  lr_request);
3011                 ext4_remove_li_request(elr);
3012         }
3013         mutex_unlock(&ext4_li_info->li_list_mtx);
3014 }
3015
3016 static int ext4_run_lazyinit_thread(void)
3017 {
3018         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3019                                          ext4_li_info, "ext4lazyinit");
3020         if (IS_ERR(ext4_lazyinit_task)) {
3021                 int err = PTR_ERR(ext4_lazyinit_task);
3022                 ext4_clear_request_list();
3023                 kfree(ext4_li_info);
3024                 ext4_li_info = NULL;
3025                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3026                                  "initialization thread\n",
3027                                  err);
3028                 return err;
3029         }
3030         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3031         return 0;
3032 }
3033
3034 /*
3035  * Check whether it make sense to run itable init. thread or not.
3036  * If there is at least one uninitialized inode table, return
3037  * corresponding group number, else the loop goes through all
3038  * groups and return total number of groups.
3039  */
3040 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3041 {
3042         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3043         struct ext4_group_desc *gdp = NULL;
3044
3045         for (group = 0; group < ngroups; group++) {
3046                 gdp = ext4_get_group_desc(sb, group, NULL);
3047                 if (!gdp)
3048                         continue;
3049
3050                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3051                         break;
3052         }
3053
3054         return group;
3055 }
3056
3057 static int ext4_li_info_new(void)
3058 {
3059         struct ext4_lazy_init *eli = NULL;
3060
3061         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3062         if (!eli)
3063                 return -ENOMEM;
3064
3065         INIT_LIST_HEAD(&eli->li_request_list);
3066         mutex_init(&eli->li_list_mtx);
3067
3068         eli->li_state |= EXT4_LAZYINIT_QUIT;
3069
3070         ext4_li_info = eli;
3071
3072         return 0;
3073 }
3074
3075 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3076                                             ext4_group_t start)
3077 {
3078         struct ext4_sb_info *sbi = EXT4_SB(sb);
3079         struct ext4_li_request *elr;
3080
3081         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3082         if (!elr)
3083                 return NULL;
3084
3085         elr->lr_super = sb;
3086         elr->lr_sbi = sbi;
3087         elr->lr_next_group = start;
3088
3089         /*
3090          * Randomize first schedule time of the request to
3091          * spread the inode table initialization requests
3092          * better.
3093          */
3094         elr->lr_next_sched = jiffies + (prandom_u32() %
3095                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3096         return elr;
3097 }
3098
3099 int ext4_register_li_request(struct super_block *sb,
3100                              ext4_group_t first_not_zeroed)
3101 {
3102         struct ext4_sb_info *sbi = EXT4_SB(sb);
3103         struct ext4_li_request *elr = NULL;
3104         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3105         int ret = 0;
3106
3107         mutex_lock(&ext4_li_mtx);
3108         if (sbi->s_li_request != NULL) {
3109                 /*
3110                  * Reset timeout so it can be computed again, because
3111                  * s_li_wait_mult might have changed.
3112                  */
3113                 sbi->s_li_request->lr_timeout = 0;
3114                 goto out;
3115         }
3116
3117         if (first_not_zeroed == ngroups ||
3118             (sb->s_flags & MS_RDONLY) ||
3119             !test_opt(sb, INIT_INODE_TABLE))
3120                 goto out;
3121
3122         elr = ext4_li_request_new(sb, first_not_zeroed);
3123         if (!elr) {
3124                 ret = -ENOMEM;
3125                 goto out;
3126         }
3127
3128         if (NULL == ext4_li_info) {
3129                 ret = ext4_li_info_new();
3130                 if (ret)
3131                         goto out;
3132         }
3133
3134         mutex_lock(&ext4_li_info->li_list_mtx);
3135         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3136         mutex_unlock(&ext4_li_info->li_list_mtx);
3137
3138         sbi->s_li_request = elr;
3139         /*
3140          * set elr to NULL here since it has been inserted to
3141          * the request_list and the removal and free of it is
3142          * handled by ext4_clear_request_list from now on.
3143          */
3144         elr = NULL;
3145
3146         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3147                 ret = ext4_run_lazyinit_thread();
3148                 if (ret)
3149                         goto out;
3150         }
3151 out:
3152         mutex_unlock(&ext4_li_mtx);
3153         if (ret)
3154                 kfree(elr);
3155         return ret;
3156 }
3157
3158 /*
3159  * We do not need to lock anything since this is called on
3160  * module unload.
3161  */
3162 static void ext4_destroy_lazyinit_thread(void)
3163 {
3164         /*
3165          * If thread exited earlier
3166          * there's nothing to be done.
3167          */
3168         if (!ext4_li_info || !ext4_lazyinit_task)
3169                 return;
3170
3171         kthread_stop(ext4_lazyinit_task);
3172 }
3173
3174 static int set_journal_csum_feature_set(struct super_block *sb)
3175 {
3176         int ret = 1;
3177         int compat, incompat;
3178         struct ext4_sb_info *sbi = EXT4_SB(sb);
3179
3180         if (ext4_has_metadata_csum(sb)) {
3181                 /* journal checksum v3 */
3182                 compat = 0;
3183                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3184         } else {
3185                 /* journal checksum v1 */
3186                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3187                 incompat = 0;
3188         }
3189
3190         jbd2_journal_clear_features(sbi->s_journal,
3191                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3192                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3193                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3194         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3195                 ret = jbd2_journal_set_features(sbi->s_journal,
3196                                 compat, 0,
3197                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3198                                 incompat);
3199         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3200                 ret = jbd2_journal_set_features(sbi->s_journal,
3201                                 compat, 0,
3202                                 incompat);
3203                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3204                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3205         } else {
3206                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3207                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3208         }
3209
3210         return ret;
3211 }
3212
3213 /*
3214  * Note: calculating the overhead so we can be compatible with
3215  * historical BSD practice is quite difficult in the face of
3216  * clusters/bigalloc.  This is because multiple metadata blocks from
3217  * different block group can end up in the same allocation cluster.
3218  * Calculating the exact overhead in the face of clustered allocation
3219  * requires either O(all block bitmaps) in memory or O(number of block
3220  * groups**2) in time.  We will still calculate the superblock for
3221  * older file systems --- and if we come across with a bigalloc file
3222  * system with zero in s_overhead_clusters the estimate will be close to
3223  * correct especially for very large cluster sizes --- but for newer
3224  * file systems, it's better to calculate this figure once at mkfs
3225  * time, and store it in the superblock.  If the superblock value is
3226  * present (even for non-bigalloc file systems), we will use it.
3227  */
3228 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3229                           char *buf)