Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux-drm-fsl-dcu.git] / fs / ext4 / super.c
1 /*
2  *  linux/fs/ext4/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/module.h>
20 #include <linux/string.h>
21 #include <linux/fs.h>
22 #include <linux/time.h>
23 #include <linux/vmalloc.h>
24 #include <linux/jbd2.h>
25 #include <linux/slab.h>
26 #include <linux/init.h>
27 #include <linux/blkdev.h>
28 #include <linux/parser.h>
29 #include <linux/buffer_head.h>
30 #include <linux/exportfs.h>
31 #include <linux/vfs.h>
32 #include <linux/random.h>
33 #include <linux/mount.h>
34 #include <linux/namei.h>
35 #include <linux/quotaops.h>
36 #include <linux/seq_file.h>
37 #include <linux/proc_fs.h>
38 #include <linux/ctype.h>
39 #include <linux/log2.h>
40 #include <linux/crc16.h>
41 #include <linux/cleancache.h>
42 #include <asm/uaccess.h>
43
44 #include <linux/kthread.h>
45 #include <linux/freezer.h>
46
47 #include "ext4.h"
48 #include "ext4_extents.h"
49 #include "ext4_jbd2.h"
50 #include "xattr.h"
51 #include "acl.h"
52 #include "mballoc.h"
53
54 #define CREATE_TRACE_POINTS
55 #include <trace/events/ext4.h>
56
57 static struct proc_dir_entry *ext4_proc_root;
58 static struct kset *ext4_kset;
59 static struct ext4_lazy_init *ext4_li_info;
60 static struct mutex ext4_li_mtx;
61 static struct ext4_features *ext4_feat;
62
63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
64                              unsigned long journal_devnum);
65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
66 static int ext4_commit_super(struct super_block *sb, int sync);
67 static void ext4_mark_recovery_complete(struct super_block *sb,
68                                         struct ext4_super_block *es);
69 static void ext4_clear_journal_err(struct super_block *sb,
70                                    struct ext4_super_block *es);
71 static int ext4_sync_fs(struct super_block *sb, int wait);
72 static const char *ext4_decode_error(struct super_block *sb, int errno,
73                                      char nbuf[16]);
74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
76 static int ext4_unfreeze(struct super_block *sb);
77 static int ext4_freeze(struct super_block *sb);
78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
79                        const char *dev_name, void *data);
80 static inline int ext2_feature_set_ok(struct super_block *sb);
81 static inline int ext3_feature_set_ok(struct super_block *sb);
82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
83 static void ext4_destroy_lazyinit_thread(void);
84 static void ext4_unregister_li_request(struct super_block *sb);
85 static void ext4_clear_request_list(void);
86
87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
88 static struct file_system_type ext2_fs_type = {
89         .owner          = THIS_MODULE,
90         .name           = "ext2",
91         .mount          = ext4_mount,
92         .kill_sb        = kill_block_super,
93         .fs_flags       = FS_REQUIRES_DEV,
94 };
95 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
96 #else
97 #define IS_EXT2_SB(sb) (0)
98 #endif
99
100
101 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
102 static struct file_system_type ext3_fs_type = {
103         .owner          = THIS_MODULE,
104         .name           = "ext3",
105         .mount          = ext4_mount,
106         .kill_sb        = kill_block_super,
107         .fs_flags       = FS_REQUIRES_DEV,
108 };
109 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
110 #else
111 #define IS_EXT3_SB(sb) (0)
112 #endif
113
114 static int ext4_verify_csum_type(struct super_block *sb,
115                                  struct ext4_super_block *es)
116 {
117         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
118                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
119                 return 1;
120
121         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
122 }
123
124 static __le32 ext4_superblock_csum(struct super_block *sb,
125                                    struct ext4_super_block *es)
126 {
127         struct ext4_sb_info *sbi = EXT4_SB(sb);
128         int offset = offsetof(struct ext4_super_block, s_checksum);
129         __u32 csum;
130
131         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
132
133         return cpu_to_le32(csum);
134 }
135
136 int ext4_superblock_csum_verify(struct super_block *sb,
137                                 struct ext4_super_block *es)
138 {
139         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
140                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
141                 return 1;
142
143         return es->s_checksum == ext4_superblock_csum(sb, es);
144 }
145
146 void ext4_superblock_csum_set(struct super_block *sb,
147                               struct ext4_super_block *es)
148 {
149         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
150                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
151                 return;
152
153         es->s_checksum = ext4_superblock_csum(sb, es);
154 }
155
156 void *ext4_kvmalloc(size_t size, gfp_t flags)
157 {
158         void *ret;
159
160         ret = kmalloc(size, flags);
161         if (!ret)
162                 ret = __vmalloc(size, flags, PAGE_KERNEL);
163         return ret;
164 }
165
166 void *ext4_kvzalloc(size_t size, gfp_t flags)
167 {
168         void *ret;
169
170         ret = kzalloc(size, flags);
171         if (!ret)
172                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
173         return ret;
174 }
175
176 void ext4_kvfree(void *ptr)
177 {
178         if (is_vmalloc_addr(ptr))
179                 vfree(ptr);
180         else
181                 kfree(ptr);
182
183 }
184
185 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
186                                struct ext4_group_desc *bg)
187 {
188         return le32_to_cpu(bg->bg_block_bitmap_lo) |
189                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
190                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
191 }
192
193 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
194                                struct ext4_group_desc *bg)
195 {
196         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
197                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
198                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
199 }
200
201 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
202                               struct ext4_group_desc *bg)
203 {
204         return le32_to_cpu(bg->bg_inode_table_lo) |
205                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
206                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
207 }
208
209 __u32 ext4_free_group_clusters(struct super_block *sb,
210                                struct ext4_group_desc *bg)
211 {
212         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
213                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
214                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
215 }
216
217 __u32 ext4_free_inodes_count(struct super_block *sb,
218                               struct ext4_group_desc *bg)
219 {
220         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
221                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
222                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
223 }
224
225 __u32 ext4_used_dirs_count(struct super_block *sb,
226                               struct ext4_group_desc *bg)
227 {
228         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
229                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
230                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
231 }
232
233 __u32 ext4_itable_unused_count(struct super_block *sb,
234                               struct ext4_group_desc *bg)
235 {
236         return le16_to_cpu(bg->bg_itable_unused_lo) |
237                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
238                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
239 }
240
241 void ext4_block_bitmap_set(struct super_block *sb,
242                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
243 {
244         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
245         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
246                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
247 }
248
249 void ext4_inode_bitmap_set(struct super_block *sb,
250                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
251 {
252         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
253         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
254                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
255 }
256
257 void ext4_inode_table_set(struct super_block *sb,
258                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
259 {
260         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
261         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
262                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
263 }
264
265 void ext4_free_group_clusters_set(struct super_block *sb,
266                                   struct ext4_group_desc *bg, __u32 count)
267 {
268         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
269         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
270                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
271 }
272
273 void ext4_free_inodes_set(struct super_block *sb,
274                           struct ext4_group_desc *bg, __u32 count)
275 {
276         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
277         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
278                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
279 }
280
281 void ext4_used_dirs_set(struct super_block *sb,
282                           struct ext4_group_desc *bg, __u32 count)
283 {
284         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
285         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
286                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
287 }
288
289 void ext4_itable_unused_set(struct super_block *sb,
290                           struct ext4_group_desc *bg, __u32 count)
291 {
292         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
293         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
294                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
295 }
296
297
298 /* Just increment the non-pointer handle value */
299 static handle_t *ext4_get_nojournal(void)
300 {
301         handle_t *handle = current->journal_info;
302         unsigned long ref_cnt = (unsigned long)handle;
303
304         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
305
306         ref_cnt++;
307         handle = (handle_t *)ref_cnt;
308
309         current->journal_info = handle;
310         return handle;
311 }
312
313
314 /* Decrement the non-pointer handle value */
315 static void ext4_put_nojournal(handle_t *handle)
316 {
317         unsigned long ref_cnt = (unsigned long)handle;
318
319         BUG_ON(ref_cnt == 0);
320
321         ref_cnt--;
322         handle = (handle_t *)ref_cnt;
323
324         current->journal_info = handle;
325 }
326
327 /*
328  * Wrappers for jbd2_journal_start/end.
329  *
330  * The only special thing we need to do here is to make sure that all
331  * journal_end calls result in the superblock being marked dirty, so
332  * that sync() will call the filesystem's write_super callback if
333  * appropriate.
334  */
335 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
336 {
337         journal_t *journal;
338
339         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
340         if (sb->s_flags & MS_RDONLY)
341                 return ERR_PTR(-EROFS);
342
343         WARN_ON(sb->s_writers.frozen == SB_FREEZE_COMPLETE);
344         journal = EXT4_SB(sb)->s_journal;
345         if (!journal)
346                 return ext4_get_nojournal();
347         /*
348          * Special case here: if the journal has aborted behind our
349          * backs (eg. EIO in the commit thread), then we still need to
350          * take the FS itself readonly cleanly.
351          */
352         if (is_journal_aborted(journal)) {
353                 ext4_abort(sb, "Detected aborted journal");
354                 return ERR_PTR(-EROFS);
355         }
356         return jbd2_journal_start(journal, nblocks);
357 }
358
359 /*
360  * The only special thing we need to do here is to make sure that all
361  * jbd2_journal_stop calls result in the superblock being marked dirty, so
362  * that sync() will call the filesystem's write_super callback if
363  * appropriate.
364  */
365 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
366 {
367         struct super_block *sb;
368         int err;
369         int rc;
370
371         if (!ext4_handle_valid(handle)) {
372                 ext4_put_nojournal(handle);
373                 return 0;
374         }
375         sb = handle->h_transaction->t_journal->j_private;
376         err = handle->h_err;
377         rc = jbd2_journal_stop(handle);
378
379         if (!err)
380                 err = rc;
381         if (err)
382                 __ext4_std_error(sb, where, line, err);
383         return err;
384 }
385
386 void ext4_journal_abort_handle(const char *caller, unsigned int line,
387                                const char *err_fn, struct buffer_head *bh,
388                                handle_t *handle, int err)
389 {
390         char nbuf[16];
391         const char *errstr = ext4_decode_error(NULL, err, nbuf);
392
393         BUG_ON(!ext4_handle_valid(handle));
394
395         if (bh)
396                 BUFFER_TRACE(bh, "abort");
397
398         if (!handle->h_err)
399                 handle->h_err = err;
400
401         if (is_handle_aborted(handle))
402                 return;
403
404         printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
405                caller, line, errstr, err_fn);
406
407         jbd2_journal_abort_handle(handle);
408 }
409
410 static void __save_error_info(struct super_block *sb, const char *func,
411                             unsigned int line)
412 {
413         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
414
415         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
416         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
417         es->s_last_error_time = cpu_to_le32(get_seconds());
418         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
419         es->s_last_error_line = cpu_to_le32(line);
420         if (!es->s_first_error_time) {
421                 es->s_first_error_time = es->s_last_error_time;
422                 strncpy(es->s_first_error_func, func,
423                         sizeof(es->s_first_error_func));
424                 es->s_first_error_line = cpu_to_le32(line);
425                 es->s_first_error_ino = es->s_last_error_ino;
426                 es->s_first_error_block = es->s_last_error_block;
427         }
428         /*
429          * Start the daily error reporting function if it hasn't been
430          * started already
431          */
432         if (!es->s_error_count)
433                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
434         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
435 }
436
437 static void save_error_info(struct super_block *sb, const char *func,
438                             unsigned int line)
439 {
440         __save_error_info(sb, func, line);
441         ext4_commit_super(sb, 1);
442 }
443
444 /*
445  * The del_gendisk() function uninitializes the disk-specific data
446  * structures, including the bdi structure, without telling anyone
447  * else.  Once this happens, any attempt to call mark_buffer_dirty()
448  * (for example, by ext4_commit_super), will cause a kernel OOPS.
449  * This is a kludge to prevent these oops until we can put in a proper
450  * hook in del_gendisk() to inform the VFS and file system layers.
451  */
452 static int block_device_ejected(struct super_block *sb)
453 {
454         struct inode *bd_inode = sb->s_bdev->bd_inode;
455         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
456
457         return bdi->dev == NULL;
458 }
459
460 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
461 {
462         struct super_block              *sb = journal->j_private;
463         struct ext4_sb_info             *sbi = EXT4_SB(sb);
464         int                             error = is_journal_aborted(journal);
465         struct ext4_journal_cb_entry    *jce, *tmp;
466
467         spin_lock(&sbi->s_md_lock);
468         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
469                 list_del_init(&jce->jce_list);
470                 spin_unlock(&sbi->s_md_lock);
471                 jce->jce_func(sb, jce, error);
472                 spin_lock(&sbi->s_md_lock);
473         }
474         spin_unlock(&sbi->s_md_lock);
475 }
476
477 /* Deal with the reporting of failure conditions on a filesystem such as
478  * inconsistencies detected or read IO failures.
479  *
480  * On ext2, we can store the error state of the filesystem in the
481  * superblock.  That is not possible on ext4, because we may have other
482  * write ordering constraints on the superblock which prevent us from
483  * writing it out straight away; and given that the journal is about to
484  * be aborted, we can't rely on the current, or future, transactions to
485  * write out the superblock safely.
486  *
487  * We'll just use the jbd2_journal_abort() error code to record an error in
488  * the journal instead.  On recovery, the journal will complain about
489  * that error until we've noted it down and cleared it.
490  */
491
492 static void ext4_handle_error(struct super_block *sb)
493 {
494         if (sb->s_flags & MS_RDONLY)
495                 return;
496
497         if (!test_opt(sb, ERRORS_CONT)) {
498                 journal_t *journal = EXT4_SB(sb)->s_journal;
499
500                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
501                 if (journal)
502                         jbd2_journal_abort(journal, -EIO);
503         }
504         if (test_opt(sb, ERRORS_RO)) {
505                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
506                 sb->s_flags |= MS_RDONLY;
507         }
508         if (test_opt(sb, ERRORS_PANIC))
509                 panic("EXT4-fs (device %s): panic forced after error\n",
510                         sb->s_id);
511 }
512
513 void __ext4_error(struct super_block *sb, const char *function,
514                   unsigned int line, const char *fmt, ...)
515 {
516         struct va_format vaf;
517         va_list args;
518
519         va_start(args, fmt);
520         vaf.fmt = fmt;
521         vaf.va = &args;
522         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
523                sb->s_id, function, line, current->comm, &vaf);
524         va_end(args);
525         save_error_info(sb, function, line);
526
527         ext4_handle_error(sb);
528 }
529
530 void ext4_error_inode(struct inode *inode, const char *function,
531                       unsigned int line, ext4_fsblk_t block,
532                       const char *fmt, ...)
533 {
534         va_list args;
535         struct va_format vaf;
536         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
537
538         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
539         es->s_last_error_block = cpu_to_le64(block);
540         save_error_info(inode->i_sb, function, line);
541         va_start(args, fmt);
542         vaf.fmt = fmt;
543         vaf.va = &args;
544         if (block)
545                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
546                        "inode #%lu: block %llu: comm %s: %pV\n",
547                        inode->i_sb->s_id, function, line, inode->i_ino,
548                        block, current->comm, &vaf);
549         else
550                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
551                        "inode #%lu: comm %s: %pV\n",
552                        inode->i_sb->s_id, function, line, inode->i_ino,
553                        current->comm, &vaf);
554         va_end(args);
555
556         ext4_handle_error(inode->i_sb);
557 }
558
559 void ext4_error_file(struct file *file, const char *function,
560                      unsigned int line, ext4_fsblk_t block,
561                      const char *fmt, ...)
562 {
563         va_list args;
564         struct va_format vaf;
565         struct ext4_super_block *es;
566         struct inode *inode = file->f_dentry->d_inode;
567         char pathname[80], *path;
568
569         es = EXT4_SB(inode->i_sb)->s_es;
570         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
571         save_error_info(inode->i_sb, function, line);
572         path = d_path(&(file->f_path), pathname, sizeof(pathname));
573         if (IS_ERR(path))
574                 path = "(unknown)";
575         va_start(args, fmt);
576         vaf.fmt = fmt;
577         vaf.va = &args;
578         if (block)
579                 printk(KERN_CRIT
580                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
581                        "block %llu: comm %s: path %s: %pV\n",
582                        inode->i_sb->s_id, function, line, inode->i_ino,
583                        block, current->comm, path, &vaf);
584         else
585                 printk(KERN_CRIT
586                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
587                        "comm %s: path %s: %pV\n",
588                        inode->i_sb->s_id, function, line, inode->i_ino,
589                        current->comm, path, &vaf);
590         va_end(args);
591
592         ext4_handle_error(inode->i_sb);
593 }
594
595 static const char *ext4_decode_error(struct super_block *sb, int errno,
596                                      char nbuf[16])
597 {
598         char *errstr = NULL;
599
600         switch (errno) {
601         case -EIO:
602                 errstr = "IO failure";
603                 break;
604         case -ENOMEM:
605                 errstr = "Out of memory";
606                 break;
607         case -EROFS:
608                 if (!sb || (EXT4_SB(sb)->s_journal &&
609                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
610                         errstr = "Journal has aborted";
611                 else
612                         errstr = "Readonly filesystem";
613                 break;
614         default:
615                 /* If the caller passed in an extra buffer for unknown
616                  * errors, textualise them now.  Else we just return
617                  * NULL. */
618                 if (nbuf) {
619                         /* Check for truncated error codes... */
620                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
621                                 errstr = nbuf;
622                 }
623                 break;
624         }
625
626         return errstr;
627 }
628
629 /* __ext4_std_error decodes expected errors from journaling functions
630  * automatically and invokes the appropriate error response.  */
631
632 void __ext4_std_error(struct super_block *sb, const char *function,
633                       unsigned int line, int errno)
634 {
635         char nbuf[16];
636         const char *errstr;
637
638         /* Special case: if the error is EROFS, and we're not already
639          * inside a transaction, then there's really no point in logging
640          * an error. */
641         if (errno == -EROFS && journal_current_handle() == NULL &&
642             (sb->s_flags & MS_RDONLY))
643                 return;
644
645         errstr = ext4_decode_error(sb, errno, nbuf);
646         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
647                sb->s_id, function, line, errstr);
648         save_error_info(sb, function, line);
649
650         ext4_handle_error(sb);
651 }
652
653 /*
654  * ext4_abort is a much stronger failure handler than ext4_error.  The
655  * abort function may be used to deal with unrecoverable failures such
656  * as journal IO errors or ENOMEM at a critical moment in log management.
657  *
658  * We unconditionally force the filesystem into an ABORT|READONLY state,
659  * unless the error response on the fs has been set to panic in which
660  * case we take the easy way out and panic immediately.
661  */
662
663 void __ext4_abort(struct super_block *sb, const char *function,
664                 unsigned int line, const char *fmt, ...)
665 {
666         va_list args;
667
668         save_error_info(sb, function, line);
669         va_start(args, fmt);
670         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
671                function, line);
672         vprintk(fmt, args);
673         printk("\n");
674         va_end(args);
675
676         if ((sb->s_flags & MS_RDONLY) == 0) {
677                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
678                 sb->s_flags |= MS_RDONLY;
679                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
680                 if (EXT4_SB(sb)->s_journal)
681                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
682                 save_error_info(sb, function, line);
683         }
684         if (test_opt(sb, ERRORS_PANIC))
685                 panic("EXT4-fs panic from previous error\n");
686 }
687
688 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
689 {
690         struct va_format vaf;
691         va_list args;
692
693         va_start(args, fmt);
694         vaf.fmt = fmt;
695         vaf.va = &args;
696         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
697         va_end(args);
698 }
699
700 void __ext4_warning(struct super_block *sb, const char *function,
701                     unsigned int line, const char *fmt, ...)
702 {
703         struct va_format vaf;
704         va_list args;
705
706         va_start(args, fmt);
707         vaf.fmt = fmt;
708         vaf.va = &args;
709         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
710                sb->s_id, function, line, &vaf);
711         va_end(args);
712 }
713
714 void __ext4_grp_locked_error(const char *function, unsigned int line,
715                              struct super_block *sb, ext4_group_t grp,
716                              unsigned long ino, ext4_fsblk_t block,
717                              const char *fmt, ...)
718 __releases(bitlock)
719 __acquires(bitlock)
720 {
721         struct va_format vaf;
722         va_list args;
723         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
724
725         es->s_last_error_ino = cpu_to_le32(ino);
726         es->s_last_error_block = cpu_to_le64(block);
727         __save_error_info(sb, function, line);
728
729         va_start(args, fmt);
730
731         vaf.fmt = fmt;
732         vaf.va = &args;
733         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
734                sb->s_id, function, line, grp);
735         if (ino)
736                 printk(KERN_CONT "inode %lu: ", ino);
737         if (block)
738                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
739         printk(KERN_CONT "%pV\n", &vaf);
740         va_end(args);
741
742         if (test_opt(sb, ERRORS_CONT)) {
743                 ext4_commit_super(sb, 0);
744                 return;
745         }
746
747         ext4_unlock_group(sb, grp);
748         ext4_handle_error(sb);
749         /*
750          * We only get here in the ERRORS_RO case; relocking the group
751          * may be dangerous, but nothing bad will happen since the
752          * filesystem will have already been marked read/only and the
753          * journal has been aborted.  We return 1 as a hint to callers
754          * who might what to use the return value from
755          * ext4_grp_locked_error() to distinguish between the
756          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
757          * aggressively from the ext4 function in question, with a
758          * more appropriate error code.
759          */
760         ext4_lock_group(sb, grp);
761         return;
762 }
763
764 void ext4_update_dynamic_rev(struct super_block *sb)
765 {
766         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
767
768         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
769                 return;
770
771         ext4_warning(sb,
772                      "updating to rev %d because of new feature flag, "
773                      "running e2fsck is recommended",
774                      EXT4_DYNAMIC_REV);
775
776         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
777         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
778         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
779         /* leave es->s_feature_*compat flags alone */
780         /* es->s_uuid will be set by e2fsck if empty */
781
782         /*
783          * The rest of the superblock fields should be zero, and if not it
784          * means they are likely already in use, so leave them alone.  We
785          * can leave it up to e2fsck to clean up any inconsistencies there.
786          */
787 }
788
789 /*
790  * Open the external journal device
791  */
792 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
793 {
794         struct block_device *bdev;
795         char b[BDEVNAME_SIZE];
796
797         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
798         if (IS_ERR(bdev))
799                 goto fail;
800         return bdev;
801
802 fail:
803         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
804                         __bdevname(dev, b), PTR_ERR(bdev));
805         return NULL;
806 }
807
808 /*
809  * Release the journal device
810  */
811 static int ext4_blkdev_put(struct block_device *bdev)
812 {
813         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
814 }
815
816 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
817 {
818         struct block_device *bdev;
819         int ret = -ENODEV;
820
821         bdev = sbi->journal_bdev;
822         if (bdev) {
823                 ret = ext4_blkdev_put(bdev);
824                 sbi->journal_bdev = NULL;
825         }
826         return ret;
827 }
828
829 static inline struct inode *orphan_list_entry(struct list_head *l)
830 {
831         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
832 }
833
834 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
835 {
836         struct list_head *l;
837
838         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
839                  le32_to_cpu(sbi->s_es->s_last_orphan));
840
841         printk(KERN_ERR "sb_info orphan list:\n");
842         list_for_each(l, &sbi->s_orphan) {
843                 struct inode *inode = orphan_list_entry(l);
844                 printk(KERN_ERR "  "
845                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
846                        inode->i_sb->s_id, inode->i_ino, inode,
847                        inode->i_mode, inode->i_nlink,
848                        NEXT_ORPHAN(inode));
849         }
850 }
851
852 static void ext4_put_super(struct super_block *sb)
853 {
854         struct ext4_sb_info *sbi = EXT4_SB(sb);
855         struct ext4_super_block *es = sbi->s_es;
856         int i, err;
857
858         ext4_unregister_li_request(sb);
859         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
860
861         flush_workqueue(sbi->dio_unwritten_wq);
862         destroy_workqueue(sbi->dio_unwritten_wq);
863
864         lock_super(sb);
865         if (sbi->s_journal) {
866                 err = jbd2_journal_destroy(sbi->s_journal);
867                 sbi->s_journal = NULL;
868                 if (err < 0)
869                         ext4_abort(sb, "Couldn't clean up the journal");
870         }
871
872         del_timer(&sbi->s_err_report);
873         ext4_release_system_zone(sb);
874         ext4_mb_release(sb);
875         ext4_ext_release(sb);
876         ext4_xattr_put_super(sb);
877
878         if (!(sb->s_flags & MS_RDONLY)) {
879                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
880                 es->s_state = cpu_to_le16(sbi->s_mount_state);
881         }
882         if (!(sb->s_flags & MS_RDONLY))
883                 ext4_commit_super(sb, 1);
884
885         if (sbi->s_proc) {
886                 remove_proc_entry("options", sbi->s_proc);
887                 remove_proc_entry(sb->s_id, ext4_proc_root);
888         }
889         kobject_del(&sbi->s_kobj);
890
891         for (i = 0; i < sbi->s_gdb_count; i++)
892                 brelse(sbi->s_group_desc[i]);
893         ext4_kvfree(sbi->s_group_desc);
894         ext4_kvfree(sbi->s_flex_groups);
895         percpu_counter_destroy(&sbi->s_freeclusters_counter);
896         percpu_counter_destroy(&sbi->s_freeinodes_counter);
897         percpu_counter_destroy(&sbi->s_dirs_counter);
898         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
899         brelse(sbi->s_sbh);
900 #ifdef CONFIG_QUOTA
901         for (i = 0; i < MAXQUOTAS; i++)
902                 kfree(sbi->s_qf_names[i]);
903 #endif
904
905         /* Debugging code just in case the in-memory inode orphan list
906          * isn't empty.  The on-disk one can be non-empty if we've
907          * detected an error and taken the fs readonly, but the
908          * in-memory list had better be clean by this point. */
909         if (!list_empty(&sbi->s_orphan))
910                 dump_orphan_list(sb, sbi);
911         J_ASSERT(list_empty(&sbi->s_orphan));
912
913         invalidate_bdev(sb->s_bdev);
914         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
915                 /*
916                  * Invalidate the journal device's buffers.  We don't want them
917                  * floating about in memory - the physical journal device may
918                  * hotswapped, and it breaks the `ro-after' testing code.
919                  */
920                 sync_blockdev(sbi->journal_bdev);
921                 invalidate_bdev(sbi->journal_bdev);
922                 ext4_blkdev_remove(sbi);
923         }
924         if (sbi->s_mmp_tsk)
925                 kthread_stop(sbi->s_mmp_tsk);
926         sb->s_fs_info = NULL;
927         /*
928          * Now that we are completely done shutting down the
929          * superblock, we need to actually destroy the kobject.
930          */
931         unlock_super(sb);
932         kobject_put(&sbi->s_kobj);
933         wait_for_completion(&sbi->s_kobj_unregister);
934         if (sbi->s_chksum_driver)
935                 crypto_free_shash(sbi->s_chksum_driver);
936         kfree(sbi->s_blockgroup_lock);
937         kfree(sbi);
938 }
939
940 static struct kmem_cache *ext4_inode_cachep;
941
942 /*
943  * Called inside transaction, so use GFP_NOFS
944  */
945 static struct inode *ext4_alloc_inode(struct super_block *sb)
946 {
947         struct ext4_inode_info *ei;
948
949         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
950         if (!ei)
951                 return NULL;
952
953         ei->vfs_inode.i_version = 1;
954         ei->vfs_inode.i_data.writeback_index = 0;
955         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
956         INIT_LIST_HEAD(&ei->i_prealloc_list);
957         spin_lock_init(&ei->i_prealloc_lock);
958         ei->i_reserved_data_blocks = 0;
959         ei->i_reserved_meta_blocks = 0;
960         ei->i_allocated_meta_blocks = 0;
961         ei->i_da_metadata_calc_len = 0;
962         spin_lock_init(&(ei->i_block_reservation_lock));
963 #ifdef CONFIG_QUOTA
964         ei->i_reserved_quota = 0;
965 #endif
966         ei->jinode = NULL;
967         INIT_LIST_HEAD(&ei->i_completed_io_list);
968         spin_lock_init(&ei->i_completed_io_lock);
969         ei->cur_aio_dio = NULL;
970         ei->i_sync_tid = 0;
971         ei->i_datasync_tid = 0;
972         atomic_set(&ei->i_ioend_count, 0);
973         atomic_set(&ei->i_aiodio_unwritten, 0);
974
975         return &ei->vfs_inode;
976 }
977
978 static int ext4_drop_inode(struct inode *inode)
979 {
980         int drop = generic_drop_inode(inode);
981
982         trace_ext4_drop_inode(inode, drop);
983         return drop;
984 }
985
986 static void ext4_i_callback(struct rcu_head *head)
987 {
988         struct inode *inode = container_of(head, struct inode, i_rcu);
989         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
990 }
991
992 static void ext4_destroy_inode(struct inode *inode)
993 {
994         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
995                 ext4_msg(inode->i_sb, KERN_ERR,
996                          "Inode %lu (%p): orphan list check failed!",
997                          inode->i_ino, EXT4_I(inode));
998                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
999                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1000                                 true);
1001                 dump_stack();
1002         }
1003         call_rcu(&inode->i_rcu, ext4_i_callback);
1004 }
1005
1006 static void init_once(void *foo)
1007 {
1008         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1009
1010         INIT_LIST_HEAD(&ei->i_orphan);
1011 #ifdef CONFIG_EXT4_FS_XATTR
1012         init_rwsem(&ei->xattr_sem);
1013 #endif
1014         init_rwsem(&ei->i_data_sem);
1015         inode_init_once(&ei->vfs_inode);
1016 }
1017
1018 static int init_inodecache(void)
1019 {
1020         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1021                                              sizeof(struct ext4_inode_info),
1022                                              0, (SLAB_RECLAIM_ACCOUNT|
1023                                                 SLAB_MEM_SPREAD),
1024                                              init_once);
1025         if (ext4_inode_cachep == NULL)
1026                 return -ENOMEM;
1027         return 0;
1028 }
1029
1030 static void destroy_inodecache(void)
1031 {
1032         kmem_cache_destroy(ext4_inode_cachep);
1033 }
1034
1035 void ext4_clear_inode(struct inode *inode)
1036 {
1037         invalidate_inode_buffers(inode);
1038         clear_inode(inode);
1039         dquot_drop(inode);
1040         ext4_discard_preallocations(inode);
1041         if (EXT4_I(inode)->jinode) {
1042                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1043                                                EXT4_I(inode)->jinode);
1044                 jbd2_free_inode(EXT4_I(inode)->jinode);
1045                 EXT4_I(inode)->jinode = NULL;
1046         }
1047 }
1048
1049 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1050                                         u64 ino, u32 generation)
1051 {
1052         struct inode *inode;
1053
1054         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1055                 return ERR_PTR(-ESTALE);
1056         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1057                 return ERR_PTR(-ESTALE);
1058
1059         /* iget isn't really right if the inode is currently unallocated!!
1060          *
1061          * ext4_read_inode will return a bad_inode if the inode had been
1062          * deleted, so we should be safe.
1063          *
1064          * Currently we don't know the generation for parent directory, so
1065          * a generation of 0 means "accept any"
1066          */
1067         inode = ext4_iget(sb, ino);
1068         if (IS_ERR(inode))
1069                 return ERR_CAST(inode);
1070         if (generation && inode->i_generation != generation) {
1071                 iput(inode);
1072                 return ERR_PTR(-ESTALE);
1073         }
1074
1075         return inode;
1076 }
1077
1078 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1079                                         int fh_len, int fh_type)
1080 {
1081         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1082                                     ext4_nfs_get_inode);
1083 }
1084
1085 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1086                                         int fh_len, int fh_type)
1087 {
1088         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1089                                     ext4_nfs_get_inode);
1090 }
1091
1092 /*
1093  * Try to release metadata pages (indirect blocks, directories) which are
1094  * mapped via the block device.  Since these pages could have journal heads
1095  * which would prevent try_to_free_buffers() from freeing them, we must use
1096  * jbd2 layer's try_to_free_buffers() function to release them.
1097  */
1098 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1099                                  gfp_t wait)
1100 {
1101         journal_t *journal = EXT4_SB(sb)->s_journal;
1102
1103         WARN_ON(PageChecked(page));
1104         if (!page_has_buffers(page))
1105                 return 0;
1106         if (journal)
1107                 return jbd2_journal_try_to_free_buffers(journal, page,
1108                                                         wait & ~__GFP_WAIT);
1109         return try_to_free_buffers(page);
1110 }
1111
1112 #ifdef CONFIG_QUOTA
1113 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1114 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1115
1116 static int ext4_write_dquot(struct dquot *dquot);
1117 static int ext4_acquire_dquot(struct dquot *dquot);
1118 static int ext4_release_dquot(struct dquot *dquot);
1119 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1120 static int ext4_write_info(struct super_block *sb, int type);
1121 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1122                          struct path *path);
1123 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1124                                  int format_id);
1125 static int ext4_quota_off(struct super_block *sb, int type);
1126 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1127 static int ext4_quota_on_mount(struct super_block *sb, int type);
1128 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1129                                size_t len, loff_t off);
1130 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1131                                 const char *data, size_t len, loff_t off);
1132 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1133                              unsigned int flags);
1134 static int ext4_enable_quotas(struct super_block *sb);
1135
1136 static const struct dquot_operations ext4_quota_operations = {
1137         .get_reserved_space = ext4_get_reserved_space,
1138         .write_dquot    = ext4_write_dquot,
1139         .acquire_dquot  = ext4_acquire_dquot,
1140         .release_dquot  = ext4_release_dquot,
1141         .mark_dirty     = ext4_mark_dquot_dirty,
1142         .write_info     = ext4_write_info,
1143         .alloc_dquot    = dquot_alloc,
1144         .destroy_dquot  = dquot_destroy,
1145 };
1146
1147 static const struct quotactl_ops ext4_qctl_operations = {
1148         .quota_on       = ext4_quota_on,
1149         .quota_off      = ext4_quota_off,
1150         .quota_sync     = dquot_quota_sync,
1151         .get_info       = dquot_get_dqinfo,
1152         .set_info       = dquot_set_dqinfo,
1153         .get_dqblk      = dquot_get_dqblk,
1154         .set_dqblk      = dquot_set_dqblk
1155 };
1156
1157 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1158         .quota_on_meta  = ext4_quota_on_sysfile,
1159         .quota_off      = ext4_quota_off_sysfile,
1160         .quota_sync     = dquot_quota_sync,
1161         .get_info       = dquot_get_dqinfo,
1162         .set_info       = dquot_set_dqinfo,
1163         .get_dqblk      = dquot_get_dqblk,
1164         .set_dqblk      = dquot_set_dqblk
1165 };
1166 #endif
1167
1168 static const struct super_operations ext4_sops = {
1169         .alloc_inode    = ext4_alloc_inode,
1170         .destroy_inode  = ext4_destroy_inode,
1171         .write_inode    = ext4_write_inode,
1172         .dirty_inode    = ext4_dirty_inode,
1173         .drop_inode     = ext4_drop_inode,
1174         .evict_inode    = ext4_evict_inode,
1175         .put_super      = ext4_put_super,
1176         .sync_fs        = ext4_sync_fs,
1177         .freeze_fs      = ext4_freeze,
1178         .unfreeze_fs    = ext4_unfreeze,
1179         .statfs         = ext4_statfs,
1180         .remount_fs     = ext4_remount,
1181         .show_options   = ext4_show_options,
1182 #ifdef CONFIG_QUOTA
1183         .quota_read     = ext4_quota_read,
1184         .quota_write    = ext4_quota_write,
1185 #endif
1186         .bdev_try_to_free_page = bdev_try_to_free_page,
1187 };
1188
1189 static const struct super_operations ext4_nojournal_sops = {
1190         .alloc_inode    = ext4_alloc_inode,
1191         .destroy_inode  = ext4_destroy_inode,
1192         .write_inode    = ext4_write_inode,
1193         .dirty_inode    = ext4_dirty_inode,
1194         .drop_inode     = ext4_drop_inode,
1195         .evict_inode    = ext4_evict_inode,
1196         .put_super      = ext4_put_super,
1197         .statfs         = ext4_statfs,
1198         .remount_fs     = ext4_remount,
1199         .show_options   = ext4_show_options,
1200 #ifdef CONFIG_QUOTA
1201         .quota_read     = ext4_quota_read,
1202         .quota_write    = ext4_quota_write,
1203 #endif
1204         .bdev_try_to_free_page = bdev_try_to_free_page,
1205 };
1206
1207 static const struct export_operations ext4_export_ops = {
1208         .fh_to_dentry = ext4_fh_to_dentry,
1209         .fh_to_parent = ext4_fh_to_parent,
1210         .get_parent = ext4_get_parent,
1211 };
1212
1213 enum {
1214         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1215         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1216         Opt_nouid32, Opt_debug, Opt_removed,
1217         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1218         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1219         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1220         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1221         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1222         Opt_data_err_abort, Opt_data_err_ignore,
1223         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1224         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1225         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1226         Opt_usrquota, Opt_grpquota, Opt_i_version,
1227         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1228         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1229         Opt_inode_readahead_blks, Opt_journal_ioprio,
1230         Opt_dioread_nolock, Opt_dioread_lock,
1231         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1232 };
1233
1234 static const match_table_t tokens = {
1235         {Opt_bsd_df, "bsddf"},
1236         {Opt_minix_df, "minixdf"},
1237         {Opt_grpid, "grpid"},
1238         {Opt_grpid, "bsdgroups"},
1239         {Opt_nogrpid, "nogrpid"},
1240         {Opt_nogrpid, "sysvgroups"},
1241         {Opt_resgid, "resgid=%u"},
1242         {Opt_resuid, "resuid=%u"},
1243         {Opt_sb, "sb=%u"},
1244         {Opt_err_cont, "errors=continue"},
1245         {Opt_err_panic, "errors=panic"},
1246         {Opt_err_ro, "errors=remount-ro"},
1247         {Opt_nouid32, "nouid32"},
1248         {Opt_debug, "debug"},
1249         {Opt_removed, "oldalloc"},
1250         {Opt_removed, "orlov"},
1251         {Opt_user_xattr, "user_xattr"},
1252         {Opt_nouser_xattr, "nouser_xattr"},
1253         {Opt_acl, "acl"},
1254         {Opt_noacl, "noacl"},
1255         {Opt_noload, "norecovery"},
1256         {Opt_noload, "noload"},
1257         {Opt_removed, "nobh"},
1258         {Opt_removed, "bh"},
1259         {Opt_commit, "commit=%u"},
1260         {Opt_min_batch_time, "min_batch_time=%u"},
1261         {Opt_max_batch_time, "max_batch_time=%u"},
1262         {Opt_journal_dev, "journal_dev=%u"},
1263         {Opt_journal_checksum, "journal_checksum"},
1264         {Opt_journal_async_commit, "journal_async_commit"},
1265         {Opt_abort, "abort"},
1266         {Opt_data_journal, "data=journal"},
1267         {Opt_data_ordered, "data=ordered"},
1268         {Opt_data_writeback, "data=writeback"},
1269         {Opt_data_err_abort, "data_err=abort"},
1270         {Opt_data_err_ignore, "data_err=ignore"},
1271         {Opt_offusrjquota, "usrjquota="},
1272         {Opt_usrjquota, "usrjquota=%s"},
1273         {Opt_offgrpjquota, "grpjquota="},
1274         {Opt_grpjquota, "grpjquota=%s"},
1275         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1276         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1277         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1278         {Opt_grpquota, "grpquota"},
1279         {Opt_noquota, "noquota"},
1280         {Opt_quota, "quota"},
1281         {Opt_usrquota, "usrquota"},
1282         {Opt_barrier, "barrier=%u"},
1283         {Opt_barrier, "barrier"},
1284         {Opt_nobarrier, "nobarrier"},
1285         {Opt_i_version, "i_version"},
1286         {Opt_stripe, "stripe=%u"},
1287         {Opt_delalloc, "delalloc"},
1288         {Opt_nodelalloc, "nodelalloc"},
1289         {Opt_mblk_io_submit, "mblk_io_submit"},
1290         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1291         {Opt_block_validity, "block_validity"},
1292         {Opt_noblock_validity, "noblock_validity"},
1293         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1294         {Opt_journal_ioprio, "journal_ioprio=%u"},
1295         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1296         {Opt_auto_da_alloc, "auto_da_alloc"},
1297         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1298         {Opt_dioread_nolock, "dioread_nolock"},
1299         {Opt_dioread_lock, "dioread_lock"},
1300         {Opt_discard, "discard"},
1301         {Opt_nodiscard, "nodiscard"},
1302         {Opt_init_itable, "init_itable=%u"},
1303         {Opt_init_itable, "init_itable"},
1304         {Opt_noinit_itable, "noinit_itable"},
1305         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1306         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1307         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1308         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1309         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1310         {Opt_err, NULL},
1311 };
1312
1313 static ext4_fsblk_t get_sb_block(void **data)
1314 {
1315         ext4_fsblk_t    sb_block;
1316         char            *options = (char *) *data;
1317
1318         if (!options || strncmp(options, "sb=", 3) != 0)
1319                 return 1;       /* Default location */
1320
1321         options += 3;
1322         /* TODO: use simple_strtoll with >32bit ext4 */
1323         sb_block = simple_strtoul(options, &options, 0);
1324         if (*options && *options != ',') {
1325                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1326                        (char *) *data);
1327                 return 1;
1328         }
1329         if (*options == ',')
1330                 options++;
1331         *data = (void *) options;
1332
1333         return sb_block;
1334 }
1335
1336 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1337 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1338         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1339
1340 #ifdef CONFIG_QUOTA
1341 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1342 {
1343         struct ext4_sb_info *sbi = EXT4_SB(sb);
1344         char *qname;
1345
1346         if (sb_any_quota_loaded(sb) &&
1347                 !sbi->s_qf_names[qtype]) {
1348                 ext4_msg(sb, KERN_ERR,
1349                         "Cannot change journaled "
1350                         "quota options when quota turned on");
1351                 return -1;
1352         }
1353         qname = match_strdup(args);
1354         if (!qname) {
1355                 ext4_msg(sb, KERN_ERR,
1356                         "Not enough memory for storing quotafile name");
1357                 return -1;
1358         }
1359         if (sbi->s_qf_names[qtype] &&
1360                 strcmp(sbi->s_qf_names[qtype], qname)) {
1361                 ext4_msg(sb, KERN_ERR,
1362                         "%s quota file already specified", QTYPE2NAME(qtype));
1363                 kfree(qname);
1364                 return -1;
1365         }
1366         sbi->s_qf_names[qtype] = qname;
1367         if (strchr(sbi->s_qf_names[qtype], '/')) {
1368                 ext4_msg(sb, KERN_ERR,
1369                         "quotafile must be on filesystem root");
1370                 kfree(sbi->s_qf_names[qtype]);
1371                 sbi->s_qf_names[qtype] = NULL;
1372                 return -1;
1373         }
1374         set_opt(sb, QUOTA);
1375         return 1;
1376 }
1377
1378 static int clear_qf_name(struct super_block *sb, int qtype)
1379 {
1380
1381         struct ext4_sb_info *sbi = EXT4_SB(sb);
1382
1383         if (sb_any_quota_loaded(sb) &&
1384                 sbi->s_qf_names[qtype]) {
1385                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1386                         " when quota turned on");
1387                 return -1;
1388         }
1389         /*
1390          * The space will be released later when all options are confirmed
1391          * to be correct
1392          */
1393         sbi->s_qf_names[qtype] = NULL;
1394         return 1;
1395 }
1396 #endif
1397
1398 #define MOPT_SET        0x0001
1399 #define MOPT_CLEAR      0x0002
1400 #define MOPT_NOSUPPORT  0x0004
1401 #define MOPT_EXPLICIT   0x0008
1402 #define MOPT_CLEAR_ERR  0x0010
1403 #define MOPT_GTE0       0x0020
1404 #ifdef CONFIG_QUOTA
1405 #define MOPT_Q          0
1406 #define MOPT_QFMT       0x0040
1407 #else
1408 #define MOPT_Q          MOPT_NOSUPPORT
1409 #define MOPT_QFMT       MOPT_NOSUPPORT
1410 #endif
1411 #define MOPT_DATAJ      0x0080
1412
1413 static const struct mount_opts {
1414         int     token;
1415         int     mount_opt;
1416         int     flags;
1417 } ext4_mount_opts[] = {
1418         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1419         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1420         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1421         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1422         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1423         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1424         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1425         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1426         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1427         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1428         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1429         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1430         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1431         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1432         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1433         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1434                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1435         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1436         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1437         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1438         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1439         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1440         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1441         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1442         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1443         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1444         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1445         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1446         {Opt_commit, 0, MOPT_GTE0},
1447         {Opt_max_batch_time, 0, MOPT_GTE0},
1448         {Opt_min_batch_time, 0, MOPT_GTE0},
1449         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1450         {Opt_init_itable, 0, MOPT_GTE0},
1451         {Opt_stripe, 0, MOPT_GTE0},
1452         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1453         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1454         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1455 #ifdef CONFIG_EXT4_FS_XATTR
1456         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1457         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1458 #else
1459         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1460         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1461 #endif
1462 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1463         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1464         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1465 #else
1466         {Opt_acl, 0, MOPT_NOSUPPORT},
1467         {Opt_noacl, 0, MOPT_NOSUPPORT},
1468 #endif
1469         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1470         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1471         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1472         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1473                                                         MOPT_SET | MOPT_Q},
1474         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1475                                                         MOPT_SET | MOPT_Q},
1476         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1477                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1478         {Opt_usrjquota, 0, MOPT_Q},
1479         {Opt_grpjquota, 0, MOPT_Q},
1480         {Opt_offusrjquota, 0, MOPT_Q},
1481         {Opt_offgrpjquota, 0, MOPT_Q},
1482         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1483         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1484         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1485         {Opt_err, 0, 0}
1486 };
1487
1488 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1489                             substring_t *args, unsigned long *journal_devnum,
1490                             unsigned int *journal_ioprio, int is_remount)
1491 {
1492         struct ext4_sb_info *sbi = EXT4_SB(sb);
1493         const struct mount_opts *m;
1494         kuid_t uid;
1495         kgid_t gid;
1496         int arg = 0;
1497
1498 #ifdef CONFIG_QUOTA
1499         if (token == Opt_usrjquota)
1500                 return set_qf_name(sb, USRQUOTA, &args[0]);
1501         else if (token == Opt_grpjquota)
1502                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1503         else if (token == Opt_offusrjquota)
1504                 return clear_qf_name(sb, USRQUOTA);
1505         else if (token == Opt_offgrpjquota)
1506                 return clear_qf_name(sb, GRPQUOTA);
1507 #endif
1508         if (args->from && match_int(args, &arg))
1509                 return -1;
1510         switch (token) {
1511         case Opt_noacl:
1512         case Opt_nouser_xattr:
1513                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1514                 break;
1515         case Opt_sb:
1516                 return 1;       /* handled by get_sb_block() */
1517         case Opt_removed:
1518                 ext4_msg(sb, KERN_WARNING,
1519                          "Ignoring removed %s option", opt);
1520                 return 1;
1521         case Opt_resuid:
1522                 uid = make_kuid(current_user_ns(), arg);
1523                 if (!uid_valid(uid)) {
1524                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1525                         return -1;
1526                 }
1527                 sbi->s_resuid = uid;
1528                 return 1;
1529         case Opt_resgid:
1530                 gid = make_kgid(current_user_ns(), arg);
1531                 if (!gid_valid(gid)) {
1532                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1533                         return -1;
1534                 }
1535                 sbi->s_resgid = gid;
1536                 return 1;
1537         case Opt_abort:
1538                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1539                 return 1;
1540         case Opt_i_version:
1541                 sb->s_flags |= MS_I_VERSION;
1542                 return 1;
1543         case Opt_journal_dev:
1544                 if (is_remount) {
1545                         ext4_msg(sb, KERN_ERR,
1546                                  "Cannot specify journal on remount");
1547                         return -1;
1548                 }
1549                 *journal_devnum = arg;
1550                 return 1;
1551         case Opt_journal_ioprio:
1552                 if (arg < 0 || arg > 7)
1553                         return -1;
1554                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1555                 return 1;
1556         }
1557
1558         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1559                 if (token != m->token)
1560                         continue;
1561                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1562                         return -1;
1563                 if (m->flags & MOPT_EXPLICIT)
1564                         set_opt2(sb, EXPLICIT_DELALLOC);
1565                 if (m->flags & MOPT_CLEAR_ERR)
1566                         clear_opt(sb, ERRORS_MASK);
1567                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1568                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1569                                  "options when quota turned on");
1570                         return -1;
1571                 }
1572
1573                 if (m->flags & MOPT_NOSUPPORT) {
1574                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1575                 } else if (token == Opt_commit) {
1576                         if (arg == 0)
1577                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1578                         sbi->s_commit_interval = HZ * arg;
1579                 } else if (token == Opt_max_batch_time) {
1580                         if (arg == 0)
1581                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1582                         sbi->s_max_batch_time = arg;
1583                 } else if (token == Opt_min_batch_time) {
1584                         sbi->s_min_batch_time = arg;
1585                 } else if (token == Opt_inode_readahead_blks) {
1586                         if (arg > (1 << 30))
1587                                 return -1;
1588                         if (arg && !is_power_of_2(arg)) {
1589                                 ext4_msg(sb, KERN_ERR,
1590                                          "EXT4-fs: inode_readahead_blks"
1591                                          " must be a power of 2");
1592                                 return -1;
1593                         }
1594                         sbi->s_inode_readahead_blks = arg;
1595                 } else if (token == Opt_init_itable) {
1596                         set_opt(sb, INIT_INODE_TABLE);
1597                         if (!args->from)
1598                                 arg = EXT4_DEF_LI_WAIT_MULT;
1599                         sbi->s_li_wait_mult = arg;
1600                 } else if (token == Opt_stripe) {
1601                         sbi->s_stripe = arg;
1602                 } else if (m->flags & MOPT_DATAJ) {
1603                         if (is_remount) {
1604                                 if (!sbi->s_journal)
1605                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1606                                 else if (test_opt(sb, DATA_FLAGS) !=
1607                                          m->mount_opt) {
1608                                         ext4_msg(sb, KERN_ERR,
1609                                          "Cannot change data mode on remount");
1610                                         return -1;
1611                                 }
1612                         } else {
1613                                 clear_opt(sb, DATA_FLAGS);
1614                                 sbi->s_mount_opt |= m->mount_opt;
1615                         }
1616 #ifdef CONFIG_QUOTA
1617                 } else if (m->flags & MOPT_QFMT) {
1618                         if (sb_any_quota_loaded(sb) &&
1619                             sbi->s_jquota_fmt != m->mount_opt) {
1620                                 ext4_msg(sb, KERN_ERR, "Cannot "
1621                                          "change journaled quota options "
1622                                          "when quota turned on");
1623                                 return -1;
1624                         }
1625                         sbi->s_jquota_fmt = m->mount_opt;
1626 #endif
1627                 } else {
1628                         if (!args->from)
1629                                 arg = 1;
1630                         if (m->flags & MOPT_CLEAR)
1631                                 arg = !arg;
1632                         else if (unlikely(!(m->flags & MOPT_SET))) {
1633                                 ext4_msg(sb, KERN_WARNING,
1634                                          "buggy handling of option %s", opt);
1635                                 WARN_ON(1);
1636                                 return -1;
1637                         }
1638                         if (arg != 0)
1639                                 sbi->s_mount_opt |= m->mount_opt;
1640                         else
1641                                 sbi->s_mount_opt &= ~m->mount_opt;
1642                 }
1643                 return 1;
1644         }
1645         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1646                  "or missing value", opt);
1647         return -1;
1648 }
1649
1650 static int parse_options(char *options, struct super_block *sb,
1651                          unsigned long *journal_devnum,
1652                          unsigned int *journal_ioprio,
1653                          int is_remount)
1654 {
1655 #ifdef CONFIG_QUOTA
1656         struct ext4_sb_info *sbi = EXT4_SB(sb);
1657 #endif
1658         char *p;
1659         substring_t args[MAX_OPT_ARGS];
1660         int token;
1661
1662         if (!options)
1663                 return 1;
1664
1665         while ((p = strsep(&options, ",")) != NULL) {
1666                 if (!*p)
1667                         continue;
1668                 /*
1669                  * Initialize args struct so we know whether arg was
1670                  * found; some options take optional arguments.
1671                  */
1672                 args[0].to = args[0].from = 0;
1673                 token = match_token(p, tokens, args);
1674                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1675                                      journal_ioprio, is_remount) < 0)
1676                         return 0;
1677         }
1678 #ifdef CONFIG_QUOTA
1679         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1680                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1681                         clear_opt(sb, USRQUOTA);
1682
1683                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1684                         clear_opt(sb, GRPQUOTA);
1685
1686                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1687                         ext4_msg(sb, KERN_ERR, "old and new quota "
1688                                         "format mixing");
1689                         return 0;
1690                 }
1691
1692                 if (!sbi->s_jquota_fmt) {
1693                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1694                                         "not specified");
1695                         return 0;
1696                 }
1697         } else {
1698                 if (sbi->s_jquota_fmt) {
1699                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1700                                         "specified with no journaling "
1701                                         "enabled");
1702                         return 0;
1703                 }
1704         }
1705 #endif
1706         return 1;
1707 }
1708
1709 static inline void ext4_show_quota_options(struct seq_file *seq,
1710                                            struct super_block *sb)
1711 {
1712 #if defined(CONFIG_QUOTA)
1713         struct ext4_sb_info *sbi = EXT4_SB(sb);
1714
1715         if (sbi->s_jquota_fmt) {
1716                 char *fmtname = "";
1717
1718                 switch (sbi->s_jquota_fmt) {
1719                 case QFMT_VFS_OLD:
1720                         fmtname = "vfsold";
1721                         break;
1722                 case QFMT_VFS_V0:
1723                         fmtname = "vfsv0";
1724                         break;
1725                 case QFMT_VFS_V1:
1726                         fmtname = "vfsv1";
1727                         break;
1728                 }
1729                 seq_printf(seq, ",jqfmt=%s", fmtname);
1730         }
1731
1732         if (sbi->s_qf_names[USRQUOTA])
1733                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1734
1735         if (sbi->s_qf_names[GRPQUOTA])
1736                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1737
1738         if (test_opt(sb, USRQUOTA))
1739                 seq_puts(seq, ",usrquota");
1740
1741         if (test_opt(sb, GRPQUOTA))
1742                 seq_puts(seq, ",grpquota");
1743 #endif
1744 }
1745
1746 static const char *token2str(int token)
1747 {
1748         static const struct match_token *t;
1749
1750         for (t = tokens; t->token != Opt_err; t++)
1751                 if (t->token == token && !strchr(t->pattern, '='))
1752                         break;
1753         return t->pattern;
1754 }
1755
1756 /*
1757  * Show an option if
1758  *  - it's set to a non-default value OR
1759  *  - if the per-sb default is different from the global default
1760  */
1761 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1762                               int nodefs)
1763 {
1764         struct ext4_sb_info *sbi = EXT4_SB(sb);
1765         struct ext4_super_block *es = sbi->s_es;
1766         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1767         const struct mount_opts *m;
1768         char sep = nodefs ? '\n' : ',';
1769
1770 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1771 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1772
1773         if (sbi->s_sb_block != 1)
1774                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1775
1776         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1777                 int want_set = m->flags & MOPT_SET;
1778                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1779                     (m->flags & MOPT_CLEAR_ERR))
1780                         continue;
1781                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1782                         continue; /* skip if same as the default */
1783                 if ((want_set &&
1784                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1785                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1786                         continue; /* select Opt_noFoo vs Opt_Foo */
1787                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1788         }
1789
1790         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1791             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1792                 SEQ_OPTS_PRINT("resuid=%u",
1793                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1794         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1795             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1796                 SEQ_OPTS_PRINT("resgid=%u",
1797                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1798         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1799         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1800                 SEQ_OPTS_PUTS("errors=remount-ro");
1801         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1802                 SEQ_OPTS_PUTS("errors=continue");
1803         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1804                 SEQ_OPTS_PUTS("errors=panic");
1805         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1806                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1807         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1808                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1809         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1810                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1811         if (sb->s_flags & MS_I_VERSION)
1812                 SEQ_OPTS_PUTS("i_version");
1813         if (nodefs || sbi->s_stripe)
1814                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1815         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1816                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1817                         SEQ_OPTS_PUTS("data=journal");
1818                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1819                         SEQ_OPTS_PUTS("data=ordered");
1820                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1821                         SEQ_OPTS_PUTS("data=writeback");
1822         }
1823         if (nodefs ||
1824             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1825                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1826                                sbi->s_inode_readahead_blks);
1827
1828         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1829                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1830                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1831
1832         ext4_show_quota_options(seq, sb);
1833         return 0;
1834 }
1835
1836 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1837 {
1838         return _ext4_show_options(seq, root->d_sb, 0);
1839 }
1840
1841 static int options_seq_show(struct seq_file *seq, void *offset)
1842 {
1843         struct super_block *sb = seq->private;
1844         int rc;
1845
1846         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1847         rc = _ext4_show_options(seq, sb, 1);
1848         seq_puts(seq, "\n");
1849         return rc;
1850 }
1851
1852 static int options_open_fs(struct inode *inode, struct file *file)
1853 {
1854         return single_open(file, options_seq_show, PDE(inode)->data);
1855 }
1856
1857 static const struct file_operations ext4_seq_options_fops = {
1858         .owner = THIS_MODULE,
1859         .open = options_open_fs,
1860         .read = seq_read,
1861         .llseek = seq_lseek,
1862         .release = single_release,
1863 };
1864
1865 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1866                             int read_only)
1867 {
1868         struct ext4_sb_info *sbi = EXT4_SB(sb);
1869         int res = 0;
1870
1871         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1872                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1873                          "forcing read-only mode");
1874                 res = MS_RDONLY;
1875         }
1876         if (read_only)
1877                 goto done;
1878         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1879                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1880                          "running e2fsck is recommended");
1881         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1882                 ext4_msg(sb, KERN_WARNING,
1883                          "warning: mounting fs with errors, "
1884                          "running e2fsck is recommended");
1885         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1886                  le16_to_cpu(es->s_mnt_count) >=
1887                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1888                 ext4_msg(sb, KERN_WARNING,
1889                          "warning: maximal mount count reached, "
1890                          "running e2fsck is recommended");
1891         else if (le32_to_cpu(es->s_checkinterval) &&
1892                 (le32_to_cpu(es->s_lastcheck) +
1893                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1894                 ext4_msg(sb, KERN_WARNING,
1895                          "warning: checktime reached, "
1896                          "running e2fsck is recommended");
1897         if (!sbi->s_journal)
1898                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1899         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1900                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1901         le16_add_cpu(&es->s_mnt_count, 1);
1902         es->s_mtime = cpu_to_le32(get_seconds());
1903         ext4_update_dynamic_rev(sb);
1904         if (sbi->s_journal)
1905                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1906
1907         ext4_commit_super(sb, 1);
1908 done:
1909         if (test_opt(sb, DEBUG))
1910                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1911                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1912                         sb->s_blocksize,
1913                         sbi->s_groups_count,
1914                         EXT4_BLOCKS_PER_GROUP(sb),
1915                         EXT4_INODES_PER_GROUP(sb),
1916                         sbi->s_mount_opt, sbi->s_mount_opt2);
1917
1918         cleancache_init_fs(sb);
1919         return res;
1920 }
1921
1922 static int ext4_fill_flex_info(struct super_block *sb)
1923 {
1924         struct ext4_sb_info *sbi = EXT4_SB(sb);
1925         struct ext4_group_desc *gdp = NULL;
1926         ext4_group_t flex_group_count;
1927         ext4_group_t flex_group;
1928         unsigned int groups_per_flex = 0;
1929         size_t size;
1930         int i;
1931
1932         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1933         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1934                 sbi->s_log_groups_per_flex = 0;
1935                 return 1;
1936         }
1937         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1938
1939         /* We allocate both existing and potentially added groups */
1940         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1941                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1942                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1943         size = flex_group_count * sizeof(struct flex_groups);
1944         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1945         if (sbi->s_flex_groups == NULL) {
1946                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1947                          flex_group_count);
1948                 goto failed;
1949         }
1950
1951         for (i = 0; i < sbi->s_groups_count; i++) {
1952                 gdp = ext4_get_group_desc(sb, i, NULL);
1953
1954                 flex_group = ext4_flex_group(sbi, i);
1955                 atomic_add(ext4_free_inodes_count(sb, gdp),
1956                            &sbi->s_flex_groups[flex_group].free_inodes);
1957                 atomic_add(ext4_free_group_clusters(sb, gdp),
1958                            &sbi->s_flex_groups[flex_group].free_clusters);
1959                 atomic_add(ext4_used_dirs_count(sb, gdp),
1960                            &sbi->s_flex_groups[flex_group].used_dirs);
1961         }
1962
1963         return 1;
1964 failed:
1965         return 0;
1966 }
1967
1968 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1969                                    struct ext4_group_desc *gdp)
1970 {
1971         int offset;
1972         __u16 crc = 0;
1973         __le32 le_group = cpu_to_le32(block_group);
1974
1975         if ((sbi->s_es->s_feature_ro_compat &
1976              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1977                 /* Use new metadata_csum algorithm */
1978                 __u16 old_csum;
1979                 __u32 csum32;
1980
1981                 old_csum = gdp->bg_checksum;
1982                 gdp->bg_checksum = 0;
1983                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1984                                      sizeof(le_group));
1985                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
1986                                      sbi->s_desc_size);
1987                 gdp->bg_checksum = old_csum;
1988
1989                 crc = csum32 & 0xFFFF;
1990                 goto out;
1991         }
1992
1993         /* old crc16 code */
1994         offset = offsetof(struct ext4_group_desc, bg_checksum);
1995
1996         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1997         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1998         crc = crc16(crc, (__u8 *)gdp, offset);
1999         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2000         /* for checksum of struct ext4_group_desc do the rest...*/
2001         if ((sbi->s_es->s_feature_incompat &
2002              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2003             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2004                 crc = crc16(crc, (__u8 *)gdp + offset,
2005                             le16_to_cpu(sbi->s_es->s_desc_size) -
2006                                 offset);
2007
2008 out:
2009         return cpu_to_le16(crc);
2010 }
2011
2012 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2013                                 struct ext4_group_desc *gdp)
2014 {
2015         if (ext4_has_group_desc_csum(sb) &&
2016             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2017                                                       block_group, gdp)))
2018                 return 0;
2019
2020         return 1;
2021 }
2022
2023 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2024                               struct ext4_group_desc *gdp)
2025 {
2026         if (!ext4_has_group_desc_csum(sb))
2027                 return;
2028         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2029 }
2030
2031 /* Called at mount-time, super-block is locked */
2032 static int ext4_check_descriptors(struct super_block *sb,
2033                                   ext4_group_t *first_not_zeroed)
2034 {
2035         struct ext4_sb_info *sbi = EXT4_SB(sb);
2036         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2037         ext4_fsblk_t last_block;
2038         ext4_fsblk_t block_bitmap;
2039         ext4_fsblk_t inode_bitmap;
2040         ext4_fsblk_t inode_table;
2041         int flexbg_flag = 0;
2042         ext4_group_t i, grp = sbi->s_groups_count;
2043
2044         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2045                 flexbg_flag = 1;
2046
2047         ext4_debug("Checking group descriptors");
2048
2049         for (i = 0; i < sbi->s_groups_count; i++) {
2050                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2051
2052                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2053                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2054                 else
2055                         last_block = first_block +
2056                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2057
2058                 if ((grp == sbi->s_groups_count) &&
2059                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2060                         grp = i;
2061
2062                 block_bitmap = ext4_block_bitmap(sb, gdp);
2063                 if (block_bitmap < first_block || block_bitmap > last_block) {
2064                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2065                                "Block bitmap for group %u not in group "
2066                                "(block %llu)!", i, block_bitmap);
2067                         return 0;
2068                 }
2069                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2070                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2071                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2072                                "Inode bitmap for group %u not in group "
2073                                "(block %llu)!", i, inode_bitmap);
2074                         return 0;
2075                 }
2076                 inode_table = ext4_inode_table(sb, gdp);
2077                 if (inode_table < first_block ||
2078                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2079                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2080                                "Inode table for group %u not in group "
2081                                "(block %llu)!", i, inode_table);
2082                         return 0;
2083                 }
2084                 ext4_lock_group(sb, i);
2085                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2086                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2087                                  "Checksum for group %u failed (%u!=%u)",
2088                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2089                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2090                         if (!(sb->s_flags & MS_RDONLY)) {
2091                                 ext4_unlock_group(sb, i);
2092                                 return 0;
2093                         }
2094                 }
2095                 ext4_unlock_group(sb, i);
2096                 if (!flexbg_flag)
2097                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2098         }
2099         if (NULL != first_not_zeroed)
2100                 *first_not_zeroed = grp;
2101
2102         ext4_free_blocks_count_set(sbi->s_es,
2103                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2104         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2105         return 1;
2106 }
2107
2108 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2109  * the superblock) which were deleted from all directories, but held open by
2110  * a process at the time of a crash.  We walk the list and try to delete these
2111  * inodes at recovery time (only with a read-write filesystem).
2112  *
2113  * In order to keep the orphan inode chain consistent during traversal (in
2114  * case of crash during recovery), we link each inode into the superblock
2115  * orphan list_head and handle it the same way as an inode deletion during
2116  * normal operation (which journals the operations for us).
2117  *
2118  * We only do an iget() and an iput() on each inode, which is very safe if we
2119  * accidentally point at an in-use or already deleted inode.  The worst that
2120  * can happen in this case is that we get a "bit already cleared" message from
2121  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2122  * e2fsck was run on this filesystem, and it must have already done the orphan
2123  * inode cleanup for us, so we can safely abort without any further action.
2124  */
2125 static void ext4_orphan_cleanup(struct super_block *sb,
2126                                 struct ext4_super_block *es)
2127 {
2128         unsigned int s_flags = sb->s_flags;
2129         int nr_orphans = 0, nr_truncates = 0;
2130 #ifdef CONFIG_QUOTA
2131         int i;
2132 #endif
2133         if (!es->s_last_orphan) {
2134                 jbd_debug(4, "no orphan inodes to clean up\n");
2135                 return;
2136         }
2137
2138         if (bdev_read_only(sb->s_bdev)) {
2139                 ext4_msg(sb, KERN_ERR, "write access "
2140                         "unavailable, skipping orphan cleanup");
2141                 return;
2142         }
2143
2144         /* Check if feature set would not allow a r/w mount */
2145         if (!ext4_feature_set_ok(sb, 0)) {
2146                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2147                          "unknown ROCOMPAT features");
2148                 return;
2149         }
2150
2151         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2152                 if (es->s_last_orphan)
2153                         jbd_debug(1, "Errors on filesystem, "
2154                                   "clearing orphan list.\n");
2155                 es->s_last_orphan = 0;
2156                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2157                 return;
2158         }
2159
2160         if (s_flags & MS_RDONLY) {
2161                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2162                 sb->s_flags &= ~MS_RDONLY;
2163         }
2164 #ifdef CONFIG_QUOTA
2165         /* Needed for iput() to work correctly and not trash data */
2166         sb->s_flags |= MS_ACTIVE;
2167         /* Turn on quotas so that they are updated correctly */
2168         for (i = 0; i < MAXQUOTAS; i++) {
2169                 if (EXT4_SB(sb)->s_qf_names[i]) {
2170                         int ret = ext4_quota_on_mount(sb, i);
2171                         if (ret < 0)
2172                                 ext4_msg(sb, KERN_ERR,
2173                                         "Cannot turn on journaled "
2174                                         "quota: error %d", ret);
2175                 }
2176         }
2177 #endif
2178
2179         while (es->s_last_orphan) {
2180                 struct inode *inode;
2181
2182                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2183                 if (IS_ERR(inode)) {
2184                         es->s_last_orphan = 0;
2185                         break;
2186                 }
2187
2188                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2189                 dquot_initialize(inode);
2190                 if (inode->i_nlink) {
2191                         ext4_msg(sb, KERN_DEBUG,
2192                                 "%s: truncating inode %lu to %lld bytes",
2193                                 __func__, inode->i_ino, inode->i_size);
2194                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2195                                   inode->i_ino, inode->i_size);
2196                         ext4_truncate(inode);
2197                         nr_truncates++;
2198                 } else {
2199                         ext4_msg(sb, KERN_DEBUG,
2200                                 "%s: deleting unreferenced inode %lu",
2201                                 __func__, inode->i_ino);
2202                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2203                                   inode->i_ino);
2204                         nr_orphans++;
2205                 }
2206                 iput(inode);  /* The delete magic happens here! */
2207         }
2208
2209 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2210
2211         if (nr_orphans)
2212                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2213                        PLURAL(nr_orphans));
2214         if (nr_truncates)
2215                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2216                        PLURAL(nr_truncates));
2217 #ifdef CONFIG_QUOTA
2218         /* Turn quotas off */
2219         for (i = 0; i < MAXQUOTAS; i++) {
2220                 if (sb_dqopt(sb)->files[i])
2221                         dquot_quota_off(sb, i);
2222         }
2223 #endif
2224         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2225 }
2226
2227 /*
2228  * Maximal extent format file size.
2229  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2230  * extent format containers, within a sector_t, and within i_blocks
2231  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2232  * so that won't be a limiting factor.
2233  *
2234  * However there is other limiting factor. We do store extents in the form
2235  * of starting block and length, hence the resulting length of the extent
2236  * covering maximum file size must fit into on-disk format containers as
2237  * well. Given that length is always by 1 unit bigger than max unit (because
2238  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2239  *
2240  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2241  */
2242 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2243 {
2244         loff_t res;
2245         loff_t upper_limit = MAX_LFS_FILESIZE;
2246
2247         /* small i_blocks in vfs inode? */
2248         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2249                 /*
2250                  * CONFIG_LBDAF is not enabled implies the inode
2251                  * i_block represent total blocks in 512 bytes
2252                  * 32 == size of vfs inode i_blocks * 8
2253                  */
2254                 upper_limit = (1LL << 32) - 1;
2255
2256                 /* total blocks in file system block size */
2257                 upper_limit >>= (blkbits - 9);
2258                 upper_limit <<= blkbits;
2259         }
2260
2261         /*
2262          * 32-bit extent-start container, ee_block. We lower the maxbytes
2263          * by one fs block, so ee_len can cover the extent of maximum file
2264          * size
2265          */
2266         res = (1LL << 32) - 1;
2267         res <<= blkbits;
2268
2269         /* Sanity check against vm- & vfs- imposed limits */
2270         if (res > upper_limit)
2271                 res = upper_limit;
2272
2273         return res;
2274 }
2275
2276 /*
2277  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2278  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2279  * We need to be 1 filesystem block less than the 2^48 sector limit.
2280  */
2281 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2282 {
2283         loff_t res = EXT4_NDIR_BLOCKS;
2284         int meta_blocks;
2285         loff_t upper_limit;
2286         /* This is calculated to be the largest file size for a dense, block
2287          * mapped file such that the file's total number of 512-byte sectors,
2288          * including data and all indirect blocks, does not exceed (2^48 - 1).
2289          *
2290          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2291          * number of 512-byte sectors of the file.
2292          */
2293
2294         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2295                 /*
2296                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2297                  * the inode i_block field represents total file blocks in
2298                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2299                  */
2300                 upper_limit = (1LL << 32) - 1;
2301
2302                 /* total blocks in file system block size */
2303                 upper_limit >>= (bits - 9);
2304
2305         } else {
2306                 /*
2307                  * We use 48 bit ext4_inode i_blocks
2308                  * With EXT4_HUGE_FILE_FL set the i_blocks
2309                  * represent total number of blocks in
2310                  * file system block size
2311                  */
2312                 upper_limit = (1LL << 48) - 1;
2313
2314         }
2315
2316         /* indirect blocks */
2317         meta_blocks = 1;
2318         /* double indirect blocks */
2319         meta_blocks += 1 + (1LL << (bits-2));
2320         /* tripple indirect blocks */
2321         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2322
2323         upper_limit -= meta_blocks;
2324         upper_limit <<= bits;
2325
2326         res += 1LL << (bits-2);
2327         res += 1LL << (2*(bits-2));
2328         res += 1LL << (3*(bits-2));
2329         res <<= bits;
2330         if (res > upper_limit)
2331                 res = upper_limit;
2332
2333         if (res > MAX_LFS_FILESIZE)
2334                 res = MAX_LFS_FILESIZE;
2335
2336         return res;
2337 }
2338
2339 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2340                                    ext4_fsblk_t logical_sb_block, int nr)
2341 {
2342         struct ext4_sb_info *sbi = EXT4_SB(sb);
2343         ext4_group_t bg, first_meta_bg;
2344         int has_super = 0;
2345
2346         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2347
2348         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2349             nr < first_meta_bg)
2350                 return logical_sb_block + nr + 1;
2351         bg = sbi->s_desc_per_block * nr;
2352         if (ext4_bg_has_super(sb, bg))
2353                 has_super = 1;
2354
2355         return (has_super + ext4_group_first_block_no(sb, bg));
2356 }
2357
2358 /**
2359  * ext4_get_stripe_size: Get the stripe size.
2360  * @sbi: In memory super block info
2361  *
2362  * If we have specified it via mount option, then
2363  * use the mount option value. If the value specified at mount time is
2364  * greater than the blocks per group use the super block value.
2365  * If the super block value is greater than blocks per group return 0.
2366  * Allocator needs it be less than blocks per group.
2367  *
2368  */
2369 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2370 {
2371         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2372         unsigned long stripe_width =
2373                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2374         int ret;
2375
2376         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2377                 ret = sbi->s_stripe;
2378         else if (stripe_width <= sbi->s_blocks_per_group)
2379                 ret = stripe_width;
2380         else if (stride <= sbi->s_blocks_per_group)
2381                 ret = stride;
2382         else
2383                 ret = 0;
2384
2385         /*
2386          * If the stripe width is 1, this makes no sense and
2387          * we set it to 0 to turn off stripe handling code.
2388          */
2389         if (ret <= 1)
2390                 ret = 0;
2391
2392         return ret;
2393 }
2394
2395 /* sysfs supprt */
2396
2397 struct ext4_attr {
2398         struct attribute attr;
2399         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2400         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2401                          const char *, size_t);
2402         int offset;
2403 };
2404
2405 static int parse_strtoul(const char *buf,
2406                 unsigned long max, unsigned long *value)
2407 {
2408         char *endp;
2409
2410         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2411         endp = skip_spaces(endp);
2412         if (*endp || *value > max)
2413                 return -EINVAL;
2414
2415         return 0;
2416 }
2417
2418 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2419                                               struct ext4_sb_info *sbi,
2420                                               char *buf)
2421 {
2422         return snprintf(buf, PAGE_SIZE, "%llu\n",
2423                 (s64) EXT4_C2B(sbi,
2424                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2425 }
2426
2427 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2428                                          struct ext4_sb_info *sbi, char *buf)
2429 {
2430         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2431
2432         if (!sb->s_bdev->bd_part)
2433                 return snprintf(buf, PAGE_SIZE, "0\n");
2434         return snprintf(buf, PAGE_SIZE, "%lu\n",
2435                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2436                          sbi->s_sectors_written_start) >> 1);
2437 }
2438
2439 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2440                                           struct ext4_sb_info *sbi, char *buf)
2441 {
2442         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2443
2444         if (!sb->s_bdev->bd_part)
2445                 return snprintf(buf, PAGE_SIZE, "0\n");
2446         return snprintf(buf, PAGE_SIZE, "%llu\n",
2447                         (unsigned long long)(sbi->s_kbytes_written +
2448                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2449                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2450 }
2451
2452 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2453                                           struct ext4_sb_info *sbi,
2454                                           const char *buf, size_t count)
2455 {
2456         unsigned long t;
2457
2458         if (parse_strtoul(buf, 0x40000000, &t))
2459                 return -EINVAL;
2460
2461         if (t && !is_power_of_2(t))
2462                 return -EINVAL;
2463
2464         sbi->s_inode_readahead_blks = t;
2465         return count;
2466 }
2467
2468 static ssize_t sbi_ui_show(struct ext4_attr *a,
2469                            struct ext4_sb_info *sbi, char *buf)
2470 {
2471         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2472
2473         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2474 }
2475
2476 static ssize_t sbi_ui_store(struct ext4_attr *a,
2477                             struct ext4_sb_info *sbi,
2478                             const char *buf, size_t count)
2479 {
2480         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2481         unsigned long t;
2482
2483         if (parse_strtoul(buf, 0xffffffff, &t))
2484                 return -EINVAL;
2485         *ui = t;
2486         return count;
2487 }
2488
2489 static ssize_t trigger_test_error(struct ext4_attr *a,
2490                                   struct ext4_sb_info *sbi,
2491                                   const char *buf, size_t count)
2492 {
2493         int len = count;
2494
2495         if (!capable(CAP_SYS_ADMIN))
2496                 return -EPERM;
2497
2498         if (len && buf[len-1] == '\n')
2499                 len--;
2500
2501         if (len)
2502                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2503         return count;
2504 }
2505
2506 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2507 static struct ext4_attr ext4_attr_##_name = {                   \
2508         .attr = {.name = __stringify(_name), .mode = _mode },   \
2509         .show   = _show,                                        \
2510         .store  = _store,                                       \
2511         .offset = offsetof(struct ext4_sb_info, _elname),       \
2512 }
2513 #define EXT4_ATTR(name, mode, show, store) \
2514 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2515
2516 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2517 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2518 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2519 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2520         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2521 #define ATTR_LIST(name) &ext4_attr_##name.attr
2522
2523 EXT4_RO_ATTR(delayed_allocation_blocks);
2524 EXT4_RO_ATTR(session_write_kbytes);
2525 EXT4_RO_ATTR(lifetime_write_kbytes);
2526 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2527                  inode_readahead_blks_store, s_inode_readahead_blks);
2528 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2529 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2530 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2531 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2532 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2533 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2534 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2535 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2536 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2537
2538 static struct attribute *ext4_attrs[] = {
2539         ATTR_LIST(delayed_allocation_blocks),
2540         ATTR_LIST(session_write_kbytes),
2541         ATTR_LIST(lifetime_write_kbytes),
2542         ATTR_LIST(inode_readahead_blks),
2543         ATTR_LIST(inode_goal),
2544         ATTR_LIST(mb_stats),
2545         ATTR_LIST(mb_max_to_scan),
2546         ATTR_LIST(mb_min_to_scan),
2547         ATTR_LIST(mb_order2_req),
2548         ATTR_LIST(mb_stream_req),
2549         ATTR_LIST(mb_group_prealloc),
2550         ATTR_LIST(max_writeback_mb_bump),
2551         ATTR_LIST(trigger_fs_error),
2552         NULL,
2553 };
2554
2555 /* Features this copy of ext4 supports */
2556 EXT4_INFO_ATTR(lazy_itable_init);
2557 EXT4_INFO_ATTR(batched_discard);
2558
2559 static struct attribute *ext4_feat_attrs[] = {
2560         ATTR_LIST(lazy_itable_init),
2561         ATTR_LIST(batched_discard),
2562         NULL,
2563 };
2564
2565 static ssize_t ext4_attr_show(struct kobject *kobj,
2566                               struct attribute *attr, char *buf)
2567 {
2568         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2569                                                 s_kobj);
2570         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2571
2572         return a->show ? a->show(a, sbi, buf) : 0;
2573 }
2574
2575 static ssize_t ext4_attr_store(struct kobject *kobj,
2576                                struct attribute *attr,
2577                                const char *buf, size_t len)
2578 {
2579         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2580                                                 s_kobj);
2581         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2582
2583         return a->store ? a->store(a, sbi, buf, len) : 0;
2584 }
2585
2586 static void ext4_sb_release(struct kobject *kobj)
2587 {
2588         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2589                                                 s_kobj);
2590         complete(&sbi->s_kobj_unregister);
2591 }
2592
2593 static const struct sysfs_ops ext4_attr_ops = {
2594         .show   = ext4_attr_show,
2595         .store  = ext4_attr_store,
2596 };
2597
2598 static struct kobj_type ext4_ktype = {
2599         .default_attrs  = ext4_attrs,
2600         .sysfs_ops      = &ext4_attr_ops,
2601         .release        = ext4_sb_release,
2602 };
2603
2604 static void ext4_feat_release(struct kobject *kobj)
2605 {
2606         complete(&ext4_feat->f_kobj_unregister);
2607 }
2608
2609 static struct kobj_type ext4_feat_ktype = {
2610         .default_attrs  = ext4_feat_attrs,
2611         .sysfs_ops      = &ext4_attr_ops,
2612         .release        = ext4_feat_release,
2613 };
2614
2615 /*
2616  * Check whether this filesystem can be mounted based on
2617  * the features present and the RDONLY/RDWR mount requested.
2618  * Returns 1 if this filesystem can be mounted as requested,
2619  * 0 if it cannot be.
2620  */
2621 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2622 {
2623         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2624                 ext4_msg(sb, KERN_ERR,
2625                         "Couldn't mount because of "
2626                         "unsupported optional features (%x)",
2627                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2628                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2629                 return 0;
2630         }
2631
2632         if (readonly)
2633                 return 1;
2634
2635         /* Check that feature set is OK for a read-write mount */
2636         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2637                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2638                          "unsupported optional features (%x)",
2639                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2640                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2641                 return 0;
2642         }
2643         /*
2644          * Large file size enabled file system can only be mounted
2645          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2646          */
2647         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2648                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2649                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2650                                  "cannot be mounted RDWR without "
2651                                  "CONFIG_LBDAF");
2652                         return 0;
2653                 }
2654         }
2655         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2656             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2657                 ext4_msg(sb, KERN_ERR,
2658                          "Can't support bigalloc feature without "
2659                          "extents feature\n");
2660                 return 0;
2661         }
2662
2663 #ifndef CONFIG_QUOTA
2664         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2665             !readonly) {
2666                 ext4_msg(sb, KERN_ERR,
2667                          "Filesystem with quota feature cannot be mounted RDWR "
2668                          "without CONFIG_QUOTA");
2669                 return 0;
2670         }
2671 #endif  /* CONFIG_QUOTA */
2672         return 1;
2673 }
2674
2675 /*
2676  * This function is called once a day if we have errors logged
2677  * on the file system
2678  */
2679 static void print_daily_error_info(unsigned long arg)
2680 {
2681         struct super_block *sb = (struct super_block *) arg;
2682         struct ext4_sb_info *sbi;
2683         struct ext4_super_block *es;
2684
2685         sbi = EXT4_SB(sb);
2686         es = sbi->s_es;
2687
2688         if (es->s_error_count)
2689                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2690                          le32_to_cpu(es->s_error_count));
2691         if (es->s_first_error_time) {
2692                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2693                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2694                        (int) sizeof(es->s_first_error_func),
2695                        es->s_first_error_func,
2696                        le32_to_cpu(es->s_first_error_line));
2697                 if (es->s_first_error_ino)
2698                         printk(": inode %u",
2699                                le32_to_cpu(es->s_first_error_ino));
2700                 if (es->s_first_error_block)
2701                         printk(": block %llu", (unsigned long long)
2702                                le64_to_cpu(es->s_first_error_block));
2703                 printk("\n");
2704         }
2705         if (es->s_last_error_time) {
2706                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2707                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2708                        (int) sizeof(es->s_last_error_func),
2709                        es->s_last_error_func,
2710                        le32_to_cpu(es->s_last_error_line));
2711                 if (es->s_last_error_ino)
2712                         printk(": inode %u",
2713                                le32_to_cpu(es->s_last_error_ino));
2714                 if (es->s_last_error_block)
2715                         printk(": block %llu", (unsigned long long)
2716                                le64_to_cpu(es->s_last_error_block));
2717                 printk("\n");
2718         }
2719         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2720 }
2721
2722 /* Find next suitable group and run ext4_init_inode_table */
2723 static int ext4_run_li_request(struct ext4_li_request *elr)
2724 {
2725         struct ext4_group_desc *gdp = NULL;
2726         ext4_group_t group, ngroups;
2727         struct super_block *sb;
2728         unsigned long timeout = 0;
2729         int ret = 0;
2730
2731         sb = elr->lr_super;
2732         ngroups = EXT4_SB(sb)->s_groups_count;
2733
2734         sb_start_write(sb);
2735         for (group = elr->lr_next_group; group < ngroups; group++) {
2736                 gdp = ext4_get_group_desc(sb, group, NULL);
2737                 if (!gdp) {
2738                         ret = 1;
2739                         break;
2740                 }
2741
2742                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2743                         break;
2744         }
2745
2746         if (group == ngroups)
2747                 ret = 1;
2748
2749         if (!ret) {
2750                 timeout = jiffies;
2751                 ret = ext4_init_inode_table(sb, group,
2752                                             elr->lr_timeout ? 0 : 1);
2753                 if (elr->lr_timeout == 0) {
2754                         timeout = (jiffies - timeout) *
2755                                   elr->lr_sbi->s_li_wait_mult;
2756                         elr->lr_timeout = timeout;
2757                 }
2758                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2759                 elr->lr_next_group = group + 1;
2760         }
2761         sb_end_write(sb);
2762
2763         return ret;
2764 }
2765
2766 /*
2767  * Remove lr_request from the list_request and free the
2768  * request structure. Should be called with li_list_mtx held
2769  */
2770 static void ext4_remove_li_request(struct ext4_li_request *elr)
2771 {
2772         struct ext4_sb_info *sbi;
2773
2774         if (!elr)
2775                 return;
2776
2777         sbi = elr->lr_sbi;
2778
2779         list_del(&elr->lr_request);
2780         sbi->s_li_request = NULL;
2781         kfree(elr);
2782 }
2783
2784 static void ext4_unregister_li_request(struct super_block *sb)
2785 {
2786         mutex_lock(&ext4_li_mtx);
2787         if (!ext4_li_info) {
2788                 mutex_unlock(&ext4_li_mtx);
2789                 return;
2790         }
2791
2792         mutex_lock(&ext4_li_info->li_list_mtx);
2793         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2794         mutex_unlock(&ext4_li_info->li_list_mtx);
2795         mutex_unlock(&ext4_li_mtx);
2796 }
2797
2798 static struct task_struct *ext4_lazyinit_task;
2799
2800 /*
2801  * This is the function where ext4lazyinit thread lives. It walks
2802  * through the request list searching for next scheduled filesystem.
2803  * When such a fs is found, run the lazy initialization request
2804  * (ext4_rn_li_request) and keep track of the time spend in this
2805  * function. Based on that time we compute next schedule time of
2806  * the request. When walking through the list is complete, compute
2807  * next waking time and put itself into sleep.
2808  */
2809 static int ext4_lazyinit_thread(void *arg)
2810 {
2811         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2812         struct list_head *pos, *n;
2813         struct ext4_li_request *elr;
2814         unsigned long next_wakeup, cur;
2815
2816         BUG_ON(NULL == eli);
2817
2818 cont_thread:
2819         while (true) {
2820                 next_wakeup = MAX_JIFFY_OFFSET;
2821
2822                 mutex_lock(&eli->li_list_mtx);
2823                 if (list_empty(&eli->li_request_list)) {
2824                         mutex_unlock(&eli->li_list_mtx);
2825                         goto exit_thread;
2826                 }
2827
2828                 list_for_each_safe(pos, n, &eli->li_request_list) {
2829                         elr = list_entry(pos, struct ext4_li_request,
2830                                          lr_request);
2831
2832                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2833                                 if (ext4_run_li_request(elr) != 0) {
2834                                         /* error, remove the lazy_init job */
2835                                         ext4_remove_li_request(elr);
2836                                         continue;
2837                                 }
2838                         }
2839
2840                         if (time_before(elr->lr_next_sched, next_wakeup))
2841                                 next_wakeup = elr->lr_next_sched;
2842                 }
2843                 mutex_unlock(&eli->li_list_mtx);
2844
2845                 try_to_freeze();
2846
2847                 cur = jiffies;
2848                 if ((time_after_eq(cur, next_wakeup)) ||
2849                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2850                         cond_resched();
2851                         continue;
2852                 }
2853
2854                 schedule_timeout_interruptible(next_wakeup - cur);
2855
2856                 if (kthread_should_stop()) {
2857                         ext4_clear_request_list();
2858                         goto exit_thread;
2859                 }
2860         }
2861
2862 exit_thread:
2863         /*
2864          * It looks like the request list is empty, but we need
2865          * to check it under the li_list_mtx lock, to prevent any
2866          * additions into it, and of course we should lock ext4_li_mtx
2867          * to atomically free the list and ext4_li_info, because at
2868          * this point another ext4 filesystem could be registering
2869          * new one.
2870          */
2871         mutex_lock(&ext4_li_mtx);
2872         mutex_lock(&eli->li_list_mtx);
2873         if (!list_empty(&eli->li_request_list)) {
2874                 mutex_unlock(&eli->li_list_mtx);
2875                 mutex_unlock(&ext4_li_mtx);
2876                 goto cont_thread;
2877         }
2878         mutex_unlock(&eli->li_list_mtx);
2879         kfree(ext4_li_info);
2880         ext4_li_info = NULL;
2881         mutex_unlock(&ext4_li_mtx);
2882
2883         return 0;
2884 }
2885
2886 static void ext4_clear_request_list(void)
2887 {
2888         struct list_head *pos, *n;
2889         struct ext4_li_request *elr;
2890
2891         mutex_lock(&ext4_li_info->li_list_mtx);
2892         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2893                 elr = list_entry(pos, struct ext4_li_request,
2894                                  lr_request);
2895                 ext4_remove_li_request(elr);
2896         }
2897         mutex_unlock(&ext4_li_info->li_list_mtx);
2898 }
2899
2900 static int ext4_run_lazyinit_thread(void)
2901 {
2902         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2903                                          ext4_li_info, "ext4lazyinit");
2904         if (IS_ERR(ext4_lazyinit_task)) {
2905                 int err = PTR_ERR(ext4_lazyinit_task);
2906                 ext4_clear_request_list();
2907                 kfree(ext4_li_info);
2908                 ext4_li_info = NULL;
2909                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2910                                  "initialization thread\n",
2911                                  err);
2912                 return err;
2913         }
2914         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2915         return 0;
2916 }
2917
2918 /*
2919  * Check whether it make sense to run itable init. thread or not.
2920  * If there is at least one uninitialized inode table, return
2921  * corresponding group number, else the loop goes through all
2922  * groups and return total number of groups.
2923  */
2924 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2925 {
2926         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2927         struct ext4_group_desc *gdp = NULL;
2928
2929         for (group = 0; group < ngroups; group++) {
2930                 gdp = ext4_get_group_desc(sb, group, NULL);
2931                 if (!gdp)
2932                         continue;
2933
2934                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2935                         break;
2936         }
2937
2938         return group;
2939 }
2940
2941 static int ext4_li_info_new(void)
2942 {
2943         struct ext4_lazy_init *eli = NULL;
2944
2945         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2946         if (!eli)
2947                 return -ENOMEM;
2948
2949         INIT_LIST_HEAD(&eli->li_request_list);
2950         mutex_init(&eli->li_list_mtx);
2951
2952         eli->li_state |= EXT4_LAZYINIT_QUIT;
2953
2954         ext4_li_info = eli;
2955
2956         return 0;
2957 }
2958
2959 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2960                                             ext4_group_t start)
2961 {
2962         struct ext4_sb_info *sbi = EXT4_SB(sb);
2963         struct ext4_li_request *elr;
2964         unsigned long rnd;
2965
2966         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2967         if (!elr)
2968                 return NULL;
2969
2970         elr->lr_super = sb;
2971         elr->lr_sbi = sbi;
2972         elr->lr_next_group = start;
2973
2974         /*
2975          * Randomize first schedule time of the request to
2976          * spread the inode table initialization requests
2977          * better.
2978          */
2979         get_random_bytes(&rnd, sizeof(rnd));
2980         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2981                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2982
2983         return elr;
2984 }
2985
2986 static int ext4_register_li_request(struct super_block *sb,
2987                                     ext4_group_t first_not_zeroed)
2988 {
2989         struct ext4_sb_info *sbi = EXT4_SB(sb);
2990         struct ext4_li_request *elr;
2991         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2992         int ret = 0;
2993
2994         if (sbi->s_li_request != NULL) {
2995                 /*
2996                  * Reset timeout so it can be computed again, because
2997                  * s_li_wait_mult might have changed.
2998                  */
2999                 sbi->s_li_request->lr_timeout = 0;
3000                 return 0;
3001         }
3002
3003         if (first_not_zeroed == ngroups ||
3004             (sb->s_flags & MS_RDONLY) ||
3005             !test_opt(sb, INIT_INODE_TABLE))
3006                 return 0;
3007
3008         elr = ext4_li_request_new(sb, first_not_zeroed);
3009         if (!elr)
3010                 return -ENOMEM;
3011
3012         mutex_lock(&ext4_li_mtx);
3013
3014         if (NULL == ext4_li_info) {
3015                 ret = ext4_li_info_new();
3016                 if (ret)
3017                         goto out;
3018         }
3019
3020         mutex_lock(&ext4_li_info->li_list_mtx);
3021         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3022         mutex_unlock(&ext4_li_info->li_list_mtx);
3023
3024         sbi->s_li_request = elr;
3025         /*
3026          * set elr to NULL here since it has been inserted to
3027          * the request_list and the removal and free of it is
3028          * handled by ext4_clear_request_list from now on.
3029          */
3030         elr = NULL;
3031
3032         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3033                 ret = ext4_run_lazyinit_thread();
3034                 if (ret)
3035                         goto out;
3036         }
3037 out:
3038         mutex_unlock(&ext4_li_mtx);
3039         if (ret)
3040                 kfree(elr);
3041         return ret;
3042 }
3043
3044 /*
3045  * We do not need to lock anything since this is called on
3046  * module unload.
3047  */
3048 static void ext4_destroy_lazyinit_thread(void)
3049 {
3050         /*
3051          * If thread exited earlier
3052          * there's nothing to be done.
3053          */
3054         if (!ext4_li_info || !ext4_lazyinit_task)
3055                 return;
3056
3057         kthread_stop(ext4_lazyinit_task);
3058 }
3059
3060 static int set_journal_csum_feature_set(struct super_block *sb)
3061 {
3062         int ret = 1;
3063         int compat, incompat;
3064         struct ext4_sb_info *sbi = EXT4_SB(sb);
3065
3066         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3067                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3068                 /* journal checksum v2 */
3069                 compat = 0;
3070                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3071         } else {
3072                 /* journal checksum v1 */
3073                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3074                 incompat = 0;
3075         }
3076
3077         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3078                 ret = jbd2_journal_set_features(sbi->s_journal,
3079                                 compat, 0,
3080                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3081                                 incompat);
3082         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3083                 ret = jbd2_journal_set_features(sbi->s_journal,
3084                                 compat, 0,
3085                                 incompat);
3086                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3087                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3088         } else {
3089                 jbd2_journal_clear_features(sbi->s_journal,
3090                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3091                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3092                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3093         }
3094
3095         return ret;
3096 }
3097
3098 /*
3099  * Note: calculating the overhead so we can be compatible with
3100  * historical BSD practice is quite difficult in the face of
3101  * clusters/bigalloc.  This is because multiple metadata blocks from
3102  * different block group can end up in the same allocation cluster.
3103  * Calculating the exact overhead in the face of clustered allocation
3104  * requires either O(all block bitmaps) in memory or O(number of block
3105  * groups**2) in time.  We will still calculate the superblock for
3106  * older file systems --- and if we come across with a bigalloc file
3107  * system with zero in s_overhead_clusters the estimate will be close to
3108  * correct especially for very large cluster sizes --- but for newer
3109  * file systems, it's better to calculate this figure once at mkfs
3110  * time, and store it in the superblock.  If the superblock value is
3111  * present (even for non-bigalloc file systems), we will use it.
3112  */
3113 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3114                           char *buf)
3115 {
3116         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3117         struct ext4_group_desc  *gdp;
3118         ext4_fsblk_t            first_block, last_block, b;
3119         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3120         int                     s, j, count = 0;
3121
3122         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3123                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3124         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3125         for (i = 0; i < ngroups; i++) {
3126                 gdp = ext4_get_group_desc(sb, i, NULL);
3127                 b = ext4_block_bitmap(sb, gdp);
3128                 if (b >= first_block && b <= last_block) {
3129                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3130                         count++;
3131                 }
3132                 b = ext4_inode_bitmap(sb, gdp);
3133                 if (b >= first_block && b <= last_block) {
3134                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3135                         count++;
3136                 }
3137                 b = ext4_inode_table(sb, gdp);
3138                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3139                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3140                                 int c = EXT4_B2C(sbi, b - first_block);
3141                                 ext4_set_bit(c, buf);
3142                                 count++;
3143                         }
3144                 if (i != grp)
3145                         continue;
3146                 s = 0;
3147                 if (ext4_bg_has_super(sb, grp)) {
3148                         ext4_set_bit(s++, buf);
3149                         count++;
3150                 }
3151                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3152                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3153                         count++;
3154                 }
3155         }
3156         if (!count)
3157                 return 0;
3158         return EXT4_CLUSTERS_PER_GROUP(sb) -
3159                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3160 }
3161
3162 /*
3163  * Compute the overhead and stash it in sbi->s_overhead
3164  */
3165 int ext4_calculate_overhead(struct super_block *sb)
3166 {
3167         struct ext4_sb_info *sbi = EXT4_SB(sb);
3168         struct ext4_super_block *es = sbi->s_es;
3169         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3170         ext4_fsblk_t overhead = 0;
3171         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3172
3173         memset(buf, 0, PAGE_SIZE);
3174         if (!buf)
3175                 return -ENOMEM;
3176
3177         /*
3178          * Compute the overhead (FS structures).  This is constant
3179          * for a given filesystem unless the number of block groups
3180          * changes so we cache the previous value until it does.
3181          */
3182
3183         /*
3184          * All of the blocks before first_data_block are overhead
3185          */
3186         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3187
3188         /*
3189          * Add the overhead found in each block group
3190          */
3191         for (i = 0; i < ngroups; i++) {
3192                 int blks;
3193
3194                 blks = count_overhead(sb, i, buf);
3195                 overhead += blks;
3196                 if (blks)
3197                         memset(buf, 0, PAGE_SIZE);
3198                 cond_resched();
3199         }
3200         sbi->s_overhead = overhead;
3201         smp_wmb();
3202         free_page((unsigned long) buf);
3203         return 0;
3204 }
3205
3206 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3207 {
3208         char *orig_data = kstrdup(data, GFP_KERNEL);
3209         struct buffer_head *bh;
3210         struct ext4_super_block *es = NULL;
3211         struct ext4_sb_info *sbi;
3212         ext4_fsblk_t block;
3213         ext4_fsblk_t sb_block = get_sb_block(&data);
3214         ext4_fsblk_t logical_sb_block;
3215         unsigned long offset = 0;
3216         unsigned long journal_devnum = 0;
3217         unsigned long def_mount_opts;
3218         struct inode *root;
3219         char *cp;
3220         const char *descr;
3221         int ret = -ENOMEM;
3222         int blocksize, clustersize;
3223         unsigned int db_count;
3224         unsigned int i;
3225         int needs_recovery, has_huge_files, has_bigalloc;
3226         __u64 blocks_count;
3227         int err;
3228         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3229         ext4_group_t first_not_zeroed;
3230
3231         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3232         if (!sbi)
3233                 goto out_free_orig;
3234
3235         sbi->s_blockgroup_lock =
3236                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3237         if (!sbi->s_blockgroup_lock) {
3238                 kfree(sbi);
3239                 goto out_free_orig;
3240         }
3241         sb->s_fs_info = sbi;
3242         sbi->s_sb = sb;
3243         sbi->s_mount_opt = 0;
3244         sbi->s_resuid = make_kuid(&init_user_ns, EXT4_DEF_RESUID);
3245         sbi->s_resgid = make_kgid(&init_user_ns, EXT4_DEF_RESGID);
3246         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3247         sbi->s_sb_block = sb_block;
3248         if (sb->s_bdev->bd_part)
3249                 sbi->s_sectors_written_start =
3250                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3251
3252         /* Cleanup superblock name */
3253         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3254                 *cp = '!';
3255
3256         ret = -EINVAL;
3257         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3258         if (!blocksize) {
3259                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3260                 goto out_fail;
3261         }
3262
3263         /*
3264          * The ext4 superblock will not be buffer aligned for other than 1kB
3265          * block sizes.  We need to calculate the offset from buffer start.
3266          */
3267         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3268                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3269                 offset = do_div(logical_sb_block, blocksize);
3270         } else {
3271                 logical_sb_block = sb_block;
3272         }
3273
3274         if (!(bh = sb_bread(sb, logical_sb_block))) {
3275                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3276                 goto out_fail;
3277         }
3278         /*
3279          * Note: s_es must be initialized as soon as possible because
3280          *       some ext4 macro-instructions depend on its value
3281          */
3282         es = (struct ext4_super_block *) (bh->b_data + offset);
3283         sbi->s_es = es;
3284         sb->s_magic = le16_to_cpu(es->s_magic);
3285         if (sb->s_magic != EXT4_SUPER_MAGIC)
3286                 goto cantfind_ext4;
3287         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3288
3289         /* Warn if metadata_csum and gdt_csum are both set. */
3290         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3291                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3292             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3293                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3294                              "redundant flags; please run fsck.");
3295
3296         /* Check for a known checksum algorithm */
3297         if (!ext4_verify_csum_type(sb, es)) {
3298                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3299                          "unknown checksum algorithm.");
3300                 silent = 1;