502c61fd739262306f7665b355cbdeba4d1e1c43
[linux-2.6.git] / fs / ext4 / super.c
1 /*
2  *  linux/fs/ext4/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/module.h>
20 #include <linux/string.h>
21 #include <linux/fs.h>
22 #include <linux/time.h>
23 #include <linux/vmalloc.h>
24 #include <linux/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_commit_super(struct super_block *sb, int sync);
66 static void ext4_mark_recovery_complete(struct super_block *sb,
67                                         struct ext4_super_block *es);
68 static void ext4_clear_journal_err(struct super_block *sb,
69                                    struct ext4_super_block *es);
70 static int ext4_sync_fs(struct super_block *sb, int wait);
71 static const char *ext4_decode_error(struct super_block *sb, int errno,
72                                      char nbuf[16]);
73 static int ext4_remount(struct super_block *sb, int *flags, char *data);
74 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
75 static int ext4_unfreeze(struct super_block *sb);
76 static void ext4_write_super(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 void *ext4_kvmalloc(size_t size, gfp_t flags)
115 {
116         void *ret;
117
118         ret = kmalloc(size, flags);
119         if (!ret)
120                 ret = __vmalloc(size, flags, PAGE_KERNEL);
121         return ret;
122 }
123
124 void *ext4_kvzalloc(size_t size, gfp_t flags)
125 {
126         void *ret;
127
128         ret = kzalloc(size, flags);
129         if (!ret)
130                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
131         return ret;
132 }
133
134 void ext4_kvfree(void *ptr)
135 {
136         if (is_vmalloc_addr(ptr))
137                 vfree(ptr);
138         else
139                 kfree(ptr);
140
141 }
142
143 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
144                                struct ext4_group_desc *bg)
145 {
146         return le32_to_cpu(bg->bg_block_bitmap_lo) |
147                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
148                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
149 }
150
151 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
152                                struct ext4_group_desc *bg)
153 {
154         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
155                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
156                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
157 }
158
159 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
160                               struct ext4_group_desc *bg)
161 {
162         return le32_to_cpu(bg->bg_inode_table_lo) |
163                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
164                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
165 }
166
167 __u32 ext4_free_group_clusters(struct super_block *sb,
168                                struct ext4_group_desc *bg)
169 {
170         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
171                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
172                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
173 }
174
175 __u32 ext4_free_inodes_count(struct super_block *sb,
176                               struct ext4_group_desc *bg)
177 {
178         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
179                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
180                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
181 }
182
183 __u32 ext4_used_dirs_count(struct super_block *sb,
184                               struct ext4_group_desc *bg)
185 {
186         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
187                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
188                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
189 }
190
191 __u32 ext4_itable_unused_count(struct super_block *sb,
192                               struct ext4_group_desc *bg)
193 {
194         return le16_to_cpu(bg->bg_itable_unused_lo) |
195                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
196                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
197 }
198
199 void ext4_block_bitmap_set(struct super_block *sb,
200                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
201 {
202         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
203         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
204                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
205 }
206
207 void ext4_inode_bitmap_set(struct super_block *sb,
208                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
209 {
210         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
211         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
212                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
213 }
214
215 void ext4_inode_table_set(struct super_block *sb,
216                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
217 {
218         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
219         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
220                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
221 }
222
223 void ext4_free_group_clusters_set(struct super_block *sb,
224                                   struct ext4_group_desc *bg, __u32 count)
225 {
226         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
227         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
228                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
229 }
230
231 void ext4_free_inodes_set(struct super_block *sb,
232                           struct ext4_group_desc *bg, __u32 count)
233 {
234         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
235         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
236                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
237 }
238
239 void ext4_used_dirs_set(struct super_block *sb,
240                           struct ext4_group_desc *bg, __u32 count)
241 {
242         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
243         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
244                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
245 }
246
247 void ext4_itable_unused_set(struct super_block *sb,
248                           struct ext4_group_desc *bg, __u32 count)
249 {
250         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
251         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
252                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
253 }
254
255
256 /* Just increment the non-pointer handle value */
257 static handle_t *ext4_get_nojournal(void)
258 {
259         handle_t *handle = current->journal_info;
260         unsigned long ref_cnt = (unsigned long)handle;
261
262         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
263
264         ref_cnt++;
265         handle = (handle_t *)ref_cnt;
266
267         current->journal_info = handle;
268         return handle;
269 }
270
271
272 /* Decrement the non-pointer handle value */
273 static void ext4_put_nojournal(handle_t *handle)
274 {
275         unsigned long ref_cnt = (unsigned long)handle;
276
277         BUG_ON(ref_cnt == 0);
278
279         ref_cnt--;
280         handle = (handle_t *)ref_cnt;
281
282         current->journal_info = handle;
283 }
284
285 /*
286  * Wrappers for jbd2_journal_start/end.
287  *
288  * The only special thing we need to do here is to make sure that all
289  * journal_end calls result in the superblock being marked dirty, so
290  * that sync() will call the filesystem's write_super callback if
291  * appropriate.
292  *
293  * To avoid j_barrier hold in userspace when a user calls freeze(),
294  * ext4 prevents a new handle from being started by s_frozen, which
295  * is in an upper layer.
296  */
297 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
298 {
299         journal_t *journal;
300         handle_t  *handle;
301
302         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
303         if (sb->s_flags & MS_RDONLY)
304                 return ERR_PTR(-EROFS);
305
306         journal = EXT4_SB(sb)->s_journal;
307         handle = ext4_journal_current_handle();
308
309         /*
310          * If a handle has been started, it should be allowed to
311          * finish, otherwise deadlock could happen between freeze
312          * and others(e.g. truncate) due to the restart of the
313          * journal handle if the filesystem is forzen and active
314          * handles are not stopped.
315          */
316         if (!handle)
317                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
318
319         if (!journal)
320                 return ext4_get_nojournal();
321         /*
322          * Special case here: if the journal has aborted behind our
323          * backs (eg. EIO in the commit thread), then we still need to
324          * take the FS itself readonly cleanly.
325          */
326         if (is_journal_aborted(journal)) {
327                 ext4_abort(sb, "Detected aborted journal");
328                 return ERR_PTR(-EROFS);
329         }
330         return jbd2_journal_start(journal, nblocks);
331 }
332
333 /*
334  * The only special thing we need to do here is to make sure that all
335  * jbd2_journal_stop calls result in the superblock being marked dirty, so
336  * that sync() will call the filesystem's write_super callback if
337  * appropriate.
338  */
339 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
340 {
341         struct super_block *sb;
342         int err;
343         int rc;
344
345         if (!ext4_handle_valid(handle)) {
346                 ext4_put_nojournal(handle);
347                 return 0;
348         }
349         sb = handle->h_transaction->t_journal->j_private;
350         err = handle->h_err;
351         rc = jbd2_journal_stop(handle);
352
353         if (!err)
354                 err = rc;
355         if (err)
356                 __ext4_std_error(sb, where, line, err);
357         return err;
358 }
359
360 void ext4_journal_abort_handle(const char *caller, unsigned int line,
361                                const char *err_fn, struct buffer_head *bh,
362                                handle_t *handle, int err)
363 {
364         char nbuf[16];
365         const char *errstr = ext4_decode_error(NULL, err, nbuf);
366
367         BUG_ON(!ext4_handle_valid(handle));
368
369         if (bh)
370                 BUFFER_TRACE(bh, "abort");
371
372         if (!handle->h_err)
373                 handle->h_err = err;
374
375         if (is_handle_aborted(handle))
376                 return;
377
378         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
379                caller, line, errstr, err_fn);
380
381         jbd2_journal_abort_handle(handle);
382 }
383
384 static void __save_error_info(struct super_block *sb, const char *func,
385                             unsigned int line)
386 {
387         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
388
389         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
390         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
391         es->s_last_error_time = cpu_to_le32(get_seconds());
392         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
393         es->s_last_error_line = cpu_to_le32(line);
394         if (!es->s_first_error_time) {
395                 es->s_first_error_time = es->s_last_error_time;
396                 strncpy(es->s_first_error_func, func,
397                         sizeof(es->s_first_error_func));
398                 es->s_first_error_line = cpu_to_le32(line);
399                 es->s_first_error_ino = es->s_last_error_ino;
400                 es->s_first_error_block = es->s_last_error_block;
401         }
402         /*
403          * Start the daily error reporting function if it hasn't been
404          * started already
405          */
406         if (!es->s_error_count)
407                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
408         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
409 }
410
411 static void save_error_info(struct super_block *sb, const char *func,
412                             unsigned int line)
413 {
414         __save_error_info(sb, func, line);
415         ext4_commit_super(sb, 1);
416 }
417
418 /*
419  * The del_gendisk() function uninitializes the disk-specific data
420  * structures, including the bdi structure, without telling anyone
421  * else.  Once this happens, any attempt to call mark_buffer_dirty()
422  * (for example, by ext4_commit_super), will cause a kernel OOPS.
423  * This is a kludge to prevent these oops until we can put in a proper
424  * hook in del_gendisk() to inform the VFS and file system layers.
425  */
426 static int block_device_ejected(struct super_block *sb)
427 {
428         struct inode *bd_inode = sb->s_bdev->bd_inode;
429         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
430
431         return bdi->dev == NULL;
432 }
433
434
435 /* Deal with the reporting of failure conditions on a filesystem such as
436  * inconsistencies detected or read IO failures.
437  *
438  * On ext2, we can store the error state of the filesystem in the
439  * superblock.  That is not possible on ext4, because we may have other
440  * write ordering constraints on the superblock which prevent us from
441  * writing it out straight away; and given that the journal is about to
442  * be aborted, we can't rely on the current, or future, transactions to
443  * write out the superblock safely.
444  *
445  * We'll just use the jbd2_journal_abort() error code to record an error in
446  * the journal instead.  On recovery, the journal will complain about
447  * that error until we've noted it down and cleared it.
448  */
449
450 static void ext4_handle_error(struct super_block *sb)
451 {
452         if (sb->s_flags & MS_RDONLY)
453                 return;
454
455         if (!test_opt(sb, ERRORS_CONT)) {
456                 journal_t *journal = EXT4_SB(sb)->s_journal;
457
458                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
459                 if (journal)
460                         jbd2_journal_abort(journal, -EIO);
461         }
462         if (test_opt(sb, ERRORS_RO)) {
463                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
464                 sb->s_flags |= MS_RDONLY;
465         }
466         if (test_opt(sb, ERRORS_PANIC))
467                 panic("EXT4-fs (device %s): panic forced after error\n",
468                         sb->s_id);
469 }
470
471 void __ext4_error(struct super_block *sb, const char *function,
472                   unsigned int line, const char *fmt, ...)
473 {
474         struct va_format vaf;
475         va_list args;
476
477         va_start(args, fmt);
478         vaf.fmt = fmt;
479         vaf.va = &args;
480         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
481                sb->s_id, function, line, current->comm, &vaf);
482         va_end(args);
483
484         ext4_handle_error(sb);
485 }
486
487 void ext4_error_inode(struct inode *inode, const char *function,
488                       unsigned int line, ext4_fsblk_t block,
489                       const char *fmt, ...)
490 {
491         va_list args;
492         struct va_format vaf;
493         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
494
495         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
496         es->s_last_error_block = cpu_to_le64(block);
497         save_error_info(inode->i_sb, function, line);
498         va_start(args, fmt);
499         vaf.fmt = fmt;
500         vaf.va = &args;
501         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
502                inode->i_sb->s_id, function, line, inode->i_ino);
503         if (block)
504                 printk(KERN_CONT "block %llu: ", block);
505         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
506         va_end(args);
507
508         ext4_handle_error(inode->i_sb);
509 }
510
511 void ext4_error_file(struct file *file, const char *function,
512                      unsigned int line, ext4_fsblk_t block,
513                      const char *fmt, ...)
514 {
515         va_list args;
516         struct va_format vaf;
517         struct ext4_super_block *es;
518         struct inode *inode = file->f_dentry->d_inode;
519         char pathname[80], *path;
520
521         es = EXT4_SB(inode->i_sb)->s_es;
522         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
523         save_error_info(inode->i_sb, function, line);
524         path = d_path(&(file->f_path), pathname, sizeof(pathname));
525         if (IS_ERR(path))
526                 path = "(unknown)";
527         printk(KERN_CRIT
528                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
529                inode->i_sb->s_id, function, line, inode->i_ino);
530         if (block)
531                 printk(KERN_CONT "block %llu: ", block);
532         va_start(args, fmt);
533         vaf.fmt = fmt;
534         vaf.va = &args;
535         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
536         va_end(args);
537
538         ext4_handle_error(inode->i_sb);
539 }
540
541 static const char *ext4_decode_error(struct super_block *sb, int errno,
542                                      char nbuf[16])
543 {
544         char *errstr = NULL;
545
546         switch (errno) {
547         case -EIO:
548                 errstr = "IO failure";
549                 break;
550         case -ENOMEM:
551                 errstr = "Out of memory";
552                 break;
553         case -EROFS:
554                 if (!sb || (EXT4_SB(sb)->s_journal &&
555                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
556                         errstr = "Journal has aborted";
557                 else
558                         errstr = "Readonly filesystem";
559                 break;
560         default:
561                 /* If the caller passed in an extra buffer for unknown
562                  * errors, textualise them now.  Else we just return
563                  * NULL. */
564                 if (nbuf) {
565                         /* Check for truncated error codes... */
566                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
567                                 errstr = nbuf;
568                 }
569                 break;
570         }
571
572         return errstr;
573 }
574
575 /* __ext4_std_error decodes expected errors from journaling functions
576  * automatically and invokes the appropriate error response.  */
577
578 void __ext4_std_error(struct super_block *sb, const char *function,
579                       unsigned int line, int errno)
580 {
581         char nbuf[16];
582         const char *errstr;
583
584         /* Special case: if the error is EROFS, and we're not already
585          * inside a transaction, then there's really no point in logging
586          * an error. */
587         if (errno == -EROFS && journal_current_handle() == NULL &&
588             (sb->s_flags & MS_RDONLY))
589                 return;
590
591         errstr = ext4_decode_error(sb, errno, nbuf);
592         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
593                sb->s_id, function, line, errstr);
594         save_error_info(sb, function, line);
595
596         ext4_handle_error(sb);
597 }
598
599 /*
600  * ext4_abort is a much stronger failure handler than ext4_error.  The
601  * abort function may be used to deal with unrecoverable failures such
602  * as journal IO errors or ENOMEM at a critical moment in log management.
603  *
604  * We unconditionally force the filesystem into an ABORT|READONLY state,
605  * unless the error response on the fs has been set to panic in which
606  * case we take the easy way out and panic immediately.
607  */
608
609 void __ext4_abort(struct super_block *sb, const char *function,
610                 unsigned int line, const char *fmt, ...)
611 {
612         va_list args;
613
614         save_error_info(sb, function, line);
615         va_start(args, fmt);
616         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
617                function, line);
618         vprintk(fmt, args);
619         printk("\n");
620         va_end(args);
621
622         if ((sb->s_flags & MS_RDONLY) == 0) {
623                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
624                 sb->s_flags |= MS_RDONLY;
625                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
626                 if (EXT4_SB(sb)->s_journal)
627                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
628                 save_error_info(sb, function, line);
629         }
630         if (test_opt(sb, ERRORS_PANIC))
631                 panic("EXT4-fs panic from previous error\n");
632 }
633
634 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
635 {
636         struct va_format vaf;
637         va_list args;
638
639         va_start(args, fmt);
640         vaf.fmt = fmt;
641         vaf.va = &args;
642         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
643         va_end(args);
644 }
645
646 void __ext4_warning(struct super_block *sb, const char *function,
647                     unsigned int line, const char *fmt, ...)
648 {
649         struct va_format vaf;
650         va_list args;
651
652         va_start(args, fmt);
653         vaf.fmt = fmt;
654         vaf.va = &args;
655         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
656                sb->s_id, function, line, &vaf);
657         va_end(args);
658 }
659
660 void __ext4_grp_locked_error(const char *function, unsigned int line,
661                              struct super_block *sb, ext4_group_t grp,
662                              unsigned long ino, ext4_fsblk_t block,
663                              const char *fmt, ...)
664 __releases(bitlock)
665 __acquires(bitlock)
666 {
667         struct va_format vaf;
668         va_list args;
669         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
670
671         es->s_last_error_ino = cpu_to_le32(ino);
672         es->s_last_error_block = cpu_to_le64(block);
673         __save_error_info(sb, function, line);
674
675         va_start(args, fmt);
676
677         vaf.fmt = fmt;
678         vaf.va = &args;
679         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
680                sb->s_id, function, line, grp);
681         if (ino)
682                 printk(KERN_CONT "inode %lu: ", ino);
683         if (block)
684                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
685         printk(KERN_CONT "%pV\n", &vaf);
686         va_end(args);
687
688         if (test_opt(sb, ERRORS_CONT)) {
689                 ext4_commit_super(sb, 0);
690                 return;
691         }
692
693         ext4_unlock_group(sb, grp);
694         ext4_handle_error(sb);
695         /*
696          * We only get here in the ERRORS_RO case; relocking the group
697          * may be dangerous, but nothing bad will happen since the
698          * filesystem will have already been marked read/only and the
699          * journal has been aborted.  We return 1 as a hint to callers
700          * who might what to use the return value from
701          * ext4_grp_locked_error() to distinguish between the
702          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
703          * aggressively from the ext4 function in question, with a
704          * more appropriate error code.
705          */
706         ext4_lock_group(sb, grp);
707         return;
708 }
709
710 void ext4_update_dynamic_rev(struct super_block *sb)
711 {
712         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
713
714         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
715                 return;
716
717         ext4_warning(sb,
718                      "updating to rev %d because of new feature flag, "
719                      "running e2fsck is recommended",
720                      EXT4_DYNAMIC_REV);
721
722         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
723         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
724         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
725         /* leave es->s_feature_*compat flags alone */
726         /* es->s_uuid will be set by e2fsck if empty */
727
728         /*
729          * The rest of the superblock fields should be zero, and if not it
730          * means they are likely already in use, so leave them alone.  We
731          * can leave it up to e2fsck to clean up any inconsistencies there.
732          */
733 }
734
735 /*
736  * Open the external journal device
737  */
738 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
739 {
740         struct block_device *bdev;
741         char b[BDEVNAME_SIZE];
742
743         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
744         if (IS_ERR(bdev))
745                 goto fail;
746         return bdev;
747
748 fail:
749         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
750                         __bdevname(dev, b), PTR_ERR(bdev));
751         return NULL;
752 }
753
754 /*
755  * Release the journal device
756  */
757 static int ext4_blkdev_put(struct block_device *bdev)
758 {
759         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
760 }
761
762 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
763 {
764         struct block_device *bdev;
765         int ret = -ENODEV;
766
767         bdev = sbi->journal_bdev;
768         if (bdev) {
769                 ret = ext4_blkdev_put(bdev);
770                 sbi->journal_bdev = NULL;
771         }
772         return ret;
773 }
774
775 static inline struct inode *orphan_list_entry(struct list_head *l)
776 {
777         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
778 }
779
780 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
781 {
782         struct list_head *l;
783
784         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
785                  le32_to_cpu(sbi->s_es->s_last_orphan));
786
787         printk(KERN_ERR "sb_info orphan list:\n");
788         list_for_each(l, &sbi->s_orphan) {
789                 struct inode *inode = orphan_list_entry(l);
790                 printk(KERN_ERR "  "
791                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
792                        inode->i_sb->s_id, inode->i_ino, inode,
793                        inode->i_mode, inode->i_nlink,
794                        NEXT_ORPHAN(inode));
795         }
796 }
797
798 static void ext4_put_super(struct super_block *sb)
799 {
800         struct ext4_sb_info *sbi = EXT4_SB(sb);
801         struct ext4_super_block *es = sbi->s_es;
802         int i, err;
803
804         ext4_unregister_li_request(sb);
805         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
806
807         flush_workqueue(sbi->dio_unwritten_wq);
808         destroy_workqueue(sbi->dio_unwritten_wq);
809
810         lock_super(sb);
811         if (sb->s_dirt)
812                 ext4_commit_super(sb, 1);
813
814         if (sbi->s_journal) {
815                 err = jbd2_journal_destroy(sbi->s_journal);
816                 sbi->s_journal = NULL;
817                 if (err < 0)
818                         ext4_abort(sb, "Couldn't clean up the journal");
819         }
820
821         del_timer(&sbi->s_err_report);
822         ext4_release_system_zone(sb);
823         ext4_mb_release(sb);
824         ext4_ext_release(sb);
825         ext4_xattr_put_super(sb);
826
827         if (!(sb->s_flags & MS_RDONLY)) {
828                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
829                 es->s_state = cpu_to_le16(sbi->s_mount_state);
830                 ext4_commit_super(sb, 1);
831         }
832         if (sbi->s_proc) {
833                 remove_proc_entry(sb->s_id, ext4_proc_root);
834         }
835         kobject_del(&sbi->s_kobj);
836
837         for (i = 0; i < sbi->s_gdb_count; i++)
838                 brelse(sbi->s_group_desc[i]);
839         ext4_kvfree(sbi->s_group_desc);
840         ext4_kvfree(sbi->s_flex_groups);
841         percpu_counter_destroy(&sbi->s_freeclusters_counter);
842         percpu_counter_destroy(&sbi->s_freeinodes_counter);
843         percpu_counter_destroy(&sbi->s_dirs_counter);
844         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
845         brelse(sbi->s_sbh);
846 #ifdef CONFIG_QUOTA
847         for (i = 0; i < MAXQUOTAS; i++)
848                 kfree(sbi->s_qf_names[i]);
849 #endif
850
851         /* Debugging code just in case the in-memory inode orphan list
852          * isn't empty.  The on-disk one can be non-empty if we've
853          * detected an error and taken the fs readonly, but the
854          * in-memory list had better be clean by this point. */
855         if (!list_empty(&sbi->s_orphan))
856                 dump_orphan_list(sb, sbi);
857         J_ASSERT(list_empty(&sbi->s_orphan));
858
859         invalidate_bdev(sb->s_bdev);
860         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
861                 /*
862                  * Invalidate the journal device's buffers.  We don't want them
863                  * floating about in memory - the physical journal device may
864                  * hotswapped, and it breaks the `ro-after' testing code.
865                  */
866                 sync_blockdev(sbi->journal_bdev);
867                 invalidate_bdev(sbi->journal_bdev);
868                 ext4_blkdev_remove(sbi);
869         }
870         if (sbi->s_mmp_tsk)
871                 kthread_stop(sbi->s_mmp_tsk);
872         sb->s_fs_info = NULL;
873         /*
874          * Now that we are completely done shutting down the
875          * superblock, we need to actually destroy the kobject.
876          */
877         unlock_super(sb);
878         kobject_put(&sbi->s_kobj);
879         wait_for_completion(&sbi->s_kobj_unregister);
880         kfree(sbi->s_blockgroup_lock);
881         kfree(sbi);
882 }
883
884 static struct kmem_cache *ext4_inode_cachep;
885
886 /*
887  * Called inside transaction, so use GFP_NOFS
888  */
889 static struct inode *ext4_alloc_inode(struct super_block *sb)
890 {
891         struct ext4_inode_info *ei;
892
893         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
894         if (!ei)
895                 return NULL;
896
897         ei->vfs_inode.i_version = 1;
898         ei->vfs_inode.i_data.writeback_index = 0;
899         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
900         INIT_LIST_HEAD(&ei->i_prealloc_list);
901         spin_lock_init(&ei->i_prealloc_lock);
902         ei->i_reserved_data_blocks = 0;
903         ei->i_reserved_meta_blocks = 0;
904         ei->i_allocated_meta_blocks = 0;
905         ei->i_da_metadata_calc_len = 0;
906         spin_lock_init(&(ei->i_block_reservation_lock));
907 #ifdef CONFIG_QUOTA
908         ei->i_reserved_quota = 0;
909 #endif
910         ei->jinode = NULL;
911         INIT_LIST_HEAD(&ei->i_completed_io_list);
912         spin_lock_init(&ei->i_completed_io_lock);
913         ei->cur_aio_dio = NULL;
914         ei->i_sync_tid = 0;
915         ei->i_datasync_tid = 0;
916         atomic_set(&ei->i_ioend_count, 0);
917         atomic_set(&ei->i_aiodio_unwritten, 0);
918
919         return &ei->vfs_inode;
920 }
921
922 static int ext4_drop_inode(struct inode *inode)
923 {
924         int drop = generic_drop_inode(inode);
925
926         trace_ext4_drop_inode(inode, drop);
927         return drop;
928 }
929
930 static void ext4_i_callback(struct rcu_head *head)
931 {
932         struct inode *inode = container_of(head, struct inode, i_rcu);
933         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
934 }
935
936 static void ext4_destroy_inode(struct inode *inode)
937 {
938         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
939                 ext4_msg(inode->i_sb, KERN_ERR,
940                          "Inode %lu (%p): orphan list check failed!",
941                          inode->i_ino, EXT4_I(inode));
942                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
943                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
944                                 true);
945                 dump_stack();
946         }
947         call_rcu(&inode->i_rcu, ext4_i_callback);
948 }
949
950 static void init_once(void *foo)
951 {
952         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
953
954         INIT_LIST_HEAD(&ei->i_orphan);
955 #ifdef CONFIG_EXT4_FS_XATTR
956         init_rwsem(&ei->xattr_sem);
957 #endif
958         init_rwsem(&ei->i_data_sem);
959         inode_init_once(&ei->vfs_inode);
960 }
961
962 static int init_inodecache(void)
963 {
964         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
965                                              sizeof(struct ext4_inode_info),
966                                              0, (SLAB_RECLAIM_ACCOUNT|
967                                                 SLAB_MEM_SPREAD),
968                                              init_once);
969         if (ext4_inode_cachep == NULL)
970                 return -ENOMEM;
971         return 0;
972 }
973
974 static void destroy_inodecache(void)
975 {
976         kmem_cache_destroy(ext4_inode_cachep);
977 }
978
979 void ext4_clear_inode(struct inode *inode)
980 {
981         invalidate_inode_buffers(inode);
982         end_writeback(inode);
983         dquot_drop(inode);
984         ext4_discard_preallocations(inode);
985         if (EXT4_I(inode)->jinode) {
986                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
987                                                EXT4_I(inode)->jinode);
988                 jbd2_free_inode(EXT4_I(inode)->jinode);
989                 EXT4_I(inode)->jinode = NULL;
990         }
991 }
992
993 static inline void ext4_show_quota_options(struct seq_file *seq,
994                                            struct super_block *sb)
995 {
996 #if defined(CONFIG_QUOTA)
997         struct ext4_sb_info *sbi = EXT4_SB(sb);
998
999         if (sbi->s_jquota_fmt) {
1000                 char *fmtname = "";
1001
1002                 switch (sbi->s_jquota_fmt) {
1003                 case QFMT_VFS_OLD:
1004                         fmtname = "vfsold";
1005                         break;
1006                 case QFMT_VFS_V0:
1007                         fmtname = "vfsv0";
1008                         break;
1009                 case QFMT_VFS_V1:
1010                         fmtname = "vfsv1";
1011                         break;
1012                 }
1013                 seq_printf(seq, ",jqfmt=%s", fmtname);
1014         }
1015
1016         if (sbi->s_qf_names[USRQUOTA])
1017                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1018
1019         if (sbi->s_qf_names[GRPQUOTA])
1020                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1021
1022         if (test_opt(sb, USRQUOTA))
1023                 seq_puts(seq, ",usrquota");
1024
1025         if (test_opt(sb, GRPQUOTA))
1026                 seq_puts(seq, ",grpquota");
1027 #endif
1028 }
1029
1030 /*
1031  * Show an option if
1032  *  - it's set to a non-default value OR
1033  *  - if the per-sb default is different from the global default
1034  */
1035 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1036 {
1037         int def_errors;
1038         unsigned long def_mount_opts;
1039         struct super_block *sb = root->d_sb;
1040         struct ext4_sb_info *sbi = EXT4_SB(sb);
1041         struct ext4_super_block *es = sbi->s_es;
1042
1043         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1044         def_errors     = le16_to_cpu(es->s_errors);
1045
1046         if (sbi->s_sb_block != 1)
1047                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1048         if (test_opt(sb, MINIX_DF))
1049                 seq_puts(seq, ",minixdf");
1050         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1051                 seq_puts(seq, ",grpid");
1052         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1053                 seq_puts(seq, ",nogrpid");
1054         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1055             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1056                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1057         }
1058         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1059             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1060                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1061         }
1062         if (test_opt(sb, ERRORS_RO)) {
1063                 if (def_errors == EXT4_ERRORS_PANIC ||
1064                     def_errors == EXT4_ERRORS_CONTINUE) {
1065                         seq_puts(seq, ",errors=remount-ro");
1066                 }
1067         }
1068         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1069                 seq_puts(seq, ",errors=continue");
1070         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1071                 seq_puts(seq, ",errors=panic");
1072         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1073                 seq_puts(seq, ",nouid32");
1074         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1075                 seq_puts(seq, ",debug");
1076 #ifdef CONFIG_EXT4_FS_XATTR
1077         if (test_opt(sb, XATTR_USER))
1078                 seq_puts(seq, ",user_xattr");
1079         if (!test_opt(sb, XATTR_USER))
1080                 seq_puts(seq, ",nouser_xattr");
1081 #endif
1082 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1083         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1084                 seq_puts(seq, ",acl");
1085         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1086                 seq_puts(seq, ",noacl");
1087 #endif
1088         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1089                 seq_printf(seq, ",commit=%u",
1090                            (unsigned) (sbi->s_commit_interval / HZ));
1091         }
1092         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1093                 seq_printf(seq, ",min_batch_time=%u",
1094                            (unsigned) sbi->s_min_batch_time);
1095         }
1096         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1097                 seq_printf(seq, ",max_batch_time=%u",
1098                            (unsigned) sbi->s_max_batch_time);
1099         }
1100
1101         /*
1102          * We're changing the default of barrier mount option, so
1103          * let's always display its mount state so it's clear what its
1104          * status is.
1105          */
1106         seq_puts(seq, ",barrier=");
1107         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1108         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1109                 seq_puts(seq, ",journal_async_commit");
1110         else if (test_opt(sb, JOURNAL_CHECKSUM))
1111                 seq_puts(seq, ",journal_checksum");
1112         if (test_opt(sb, I_VERSION))
1113                 seq_puts(seq, ",i_version");
1114         if (!test_opt(sb, DELALLOC) &&
1115             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1116                 seq_puts(seq, ",nodelalloc");
1117
1118         if (!test_opt(sb, MBLK_IO_SUBMIT))
1119                 seq_puts(seq, ",nomblk_io_submit");
1120         if (sbi->s_stripe)
1121                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1122         /*
1123          * journal mode get enabled in different ways
1124          * So just print the value even if we didn't specify it
1125          */
1126         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1127                 seq_puts(seq, ",data=journal");
1128         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1129                 seq_puts(seq, ",data=ordered");
1130         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1131                 seq_puts(seq, ",data=writeback");
1132
1133         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1134                 seq_printf(seq, ",inode_readahead_blks=%u",
1135                            sbi->s_inode_readahead_blks);
1136
1137         if (test_opt(sb, DATA_ERR_ABORT))
1138                 seq_puts(seq, ",data_err=abort");
1139
1140         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1141                 seq_puts(seq, ",noauto_da_alloc");
1142
1143         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1144                 seq_puts(seq, ",discard");
1145
1146         if (test_opt(sb, NOLOAD))
1147                 seq_puts(seq, ",norecovery");
1148
1149         if (test_opt(sb, DIOREAD_NOLOCK))
1150                 seq_puts(seq, ",dioread_nolock");
1151
1152         if (test_opt(sb, BLOCK_VALIDITY) &&
1153             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1154                 seq_puts(seq, ",block_validity");
1155
1156         if (!test_opt(sb, INIT_INODE_TABLE))
1157                 seq_puts(seq, ",noinit_itable");
1158         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1159                 seq_printf(seq, ",init_itable=%u",
1160                            (unsigned) sbi->s_li_wait_mult);
1161
1162         ext4_show_quota_options(seq, sb);
1163
1164         return 0;
1165 }
1166
1167 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1168                                         u64 ino, u32 generation)
1169 {
1170         struct inode *inode;
1171
1172         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1173                 return ERR_PTR(-ESTALE);
1174         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1175                 return ERR_PTR(-ESTALE);
1176
1177         /* iget isn't really right if the inode is currently unallocated!!
1178          *
1179          * ext4_read_inode will return a bad_inode if the inode had been
1180          * deleted, so we should be safe.
1181          *
1182          * Currently we don't know the generation for parent directory, so
1183          * a generation of 0 means "accept any"
1184          */
1185         inode = ext4_iget(sb, ino);
1186         if (IS_ERR(inode))
1187                 return ERR_CAST(inode);
1188         if (generation && inode->i_generation != generation) {
1189                 iput(inode);
1190                 return ERR_PTR(-ESTALE);
1191         }
1192
1193         return inode;
1194 }
1195
1196 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1197                                         int fh_len, int fh_type)
1198 {
1199         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1200                                     ext4_nfs_get_inode);
1201 }
1202
1203 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1204                                         int fh_len, int fh_type)
1205 {
1206         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1207                                     ext4_nfs_get_inode);
1208 }
1209
1210 /*
1211  * Try to release metadata pages (indirect blocks, directories) which are
1212  * mapped via the block device.  Since these pages could have journal heads
1213  * which would prevent try_to_free_buffers() from freeing them, we must use
1214  * jbd2 layer's try_to_free_buffers() function to release them.
1215  */
1216 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1217                                  gfp_t wait)
1218 {
1219         journal_t *journal = EXT4_SB(sb)->s_journal;
1220
1221         WARN_ON(PageChecked(page));
1222         if (!page_has_buffers(page))
1223                 return 0;
1224         if (journal)
1225                 return jbd2_journal_try_to_free_buffers(journal, page,
1226                                                         wait & ~__GFP_WAIT);
1227         return try_to_free_buffers(page);
1228 }
1229
1230 #ifdef CONFIG_QUOTA
1231 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1232 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1233
1234 static int ext4_write_dquot(struct dquot *dquot);
1235 static int ext4_acquire_dquot(struct dquot *dquot);
1236 static int ext4_release_dquot(struct dquot *dquot);
1237 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1238 static int ext4_write_info(struct super_block *sb, int type);
1239 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1240                          struct path *path);
1241 static int ext4_quota_off(struct super_block *sb, int type);
1242 static int ext4_quota_on_mount(struct super_block *sb, int type);
1243 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1244                                size_t len, loff_t off);
1245 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1246                                 const char *data, size_t len, loff_t off);
1247
1248 static const struct dquot_operations ext4_quota_operations = {
1249         .get_reserved_space = ext4_get_reserved_space,
1250         .write_dquot    = ext4_write_dquot,
1251         .acquire_dquot  = ext4_acquire_dquot,
1252         .release_dquot  = ext4_release_dquot,
1253         .mark_dirty     = ext4_mark_dquot_dirty,
1254         .write_info     = ext4_write_info,
1255         .alloc_dquot    = dquot_alloc,
1256         .destroy_dquot  = dquot_destroy,
1257 };
1258
1259 static const struct quotactl_ops ext4_qctl_operations = {
1260         .quota_on       = ext4_quota_on,
1261         .quota_off      = ext4_quota_off,
1262         .quota_sync     = dquot_quota_sync,
1263         .get_info       = dquot_get_dqinfo,
1264         .set_info       = dquot_set_dqinfo,
1265         .get_dqblk      = dquot_get_dqblk,
1266         .set_dqblk      = dquot_set_dqblk
1267 };
1268 #endif
1269
1270 static const struct super_operations ext4_sops = {
1271         .alloc_inode    = ext4_alloc_inode,
1272         .destroy_inode  = ext4_destroy_inode,
1273         .write_inode    = ext4_write_inode,
1274         .dirty_inode    = ext4_dirty_inode,
1275         .drop_inode     = ext4_drop_inode,
1276         .evict_inode    = ext4_evict_inode,
1277         .put_super      = ext4_put_super,
1278         .sync_fs        = ext4_sync_fs,
1279         .freeze_fs      = ext4_freeze,
1280         .unfreeze_fs    = ext4_unfreeze,
1281         .statfs         = ext4_statfs,
1282         .remount_fs     = ext4_remount,
1283         .show_options   = ext4_show_options,
1284 #ifdef CONFIG_QUOTA
1285         .quota_read     = ext4_quota_read,
1286         .quota_write    = ext4_quota_write,
1287 #endif
1288         .bdev_try_to_free_page = bdev_try_to_free_page,
1289 };
1290
1291 static const struct super_operations ext4_nojournal_sops = {
1292         .alloc_inode    = ext4_alloc_inode,
1293         .destroy_inode  = ext4_destroy_inode,
1294         .write_inode    = ext4_write_inode,
1295         .dirty_inode    = ext4_dirty_inode,
1296         .drop_inode     = ext4_drop_inode,
1297         .evict_inode    = ext4_evict_inode,
1298         .write_super    = ext4_write_super,
1299         .put_super      = ext4_put_super,
1300         .statfs         = ext4_statfs,
1301         .remount_fs     = ext4_remount,
1302         .show_options   = ext4_show_options,
1303 #ifdef CONFIG_QUOTA
1304         .quota_read     = ext4_quota_read,
1305         .quota_write    = ext4_quota_write,
1306 #endif
1307         .bdev_try_to_free_page = bdev_try_to_free_page,
1308 };
1309
1310 static const struct export_operations ext4_export_ops = {
1311         .fh_to_dentry = ext4_fh_to_dentry,
1312         .fh_to_parent = ext4_fh_to_parent,
1313         .get_parent = ext4_get_parent,
1314 };
1315
1316 enum {
1317         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1318         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1319         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1320         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1321         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1322         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1323         Opt_journal_update, Opt_journal_dev,
1324         Opt_journal_checksum, Opt_journal_async_commit,
1325         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1326         Opt_data_err_abort, Opt_data_err_ignore,
1327         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1328         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1329         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1330         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1331         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1332         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1333         Opt_inode_readahead_blks, Opt_journal_ioprio,
1334         Opt_dioread_nolock, Opt_dioread_lock,
1335         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1336 };
1337
1338 static const match_table_t tokens = {
1339         {Opt_bsd_df, "bsddf"},
1340         {Opt_minix_df, "minixdf"},
1341         {Opt_grpid, "grpid"},
1342         {Opt_grpid, "bsdgroups"},
1343         {Opt_nogrpid, "nogrpid"},
1344         {Opt_nogrpid, "sysvgroups"},
1345         {Opt_resgid, "resgid=%u"},
1346         {Opt_resuid, "resuid=%u"},
1347         {Opt_sb, "sb=%u"},
1348         {Opt_err_cont, "errors=continue"},
1349         {Opt_err_panic, "errors=panic"},
1350         {Opt_err_ro, "errors=remount-ro"},
1351         {Opt_nouid32, "nouid32"},
1352         {Opt_debug, "debug"},
1353         {Opt_oldalloc, "oldalloc"},
1354         {Opt_orlov, "orlov"},
1355         {Opt_user_xattr, "user_xattr"},
1356         {Opt_nouser_xattr, "nouser_xattr"},
1357         {Opt_acl, "acl"},
1358         {Opt_noacl, "noacl"},
1359         {Opt_noload, "noload"},
1360         {Opt_noload, "norecovery"},
1361         {Opt_nobh, "nobh"},
1362         {Opt_bh, "bh"},
1363         {Opt_commit, "commit=%u"},
1364         {Opt_min_batch_time, "min_batch_time=%u"},
1365         {Opt_max_batch_time, "max_batch_time=%u"},
1366         {Opt_journal_update, "journal=update"},
1367         {Opt_journal_dev, "journal_dev=%u"},
1368         {Opt_journal_checksum, "journal_checksum"},
1369         {Opt_journal_async_commit, "journal_async_commit"},
1370         {Opt_abort, "abort"},
1371         {Opt_data_journal, "data=journal"},
1372         {Opt_data_ordered, "data=ordered"},
1373         {Opt_data_writeback, "data=writeback"},
1374         {Opt_data_err_abort, "data_err=abort"},
1375         {Opt_data_err_ignore, "data_err=ignore"},
1376         {Opt_offusrjquota, "usrjquota="},
1377         {Opt_usrjquota, "usrjquota=%s"},
1378         {Opt_offgrpjquota, "grpjquota="},
1379         {Opt_grpjquota, "grpjquota=%s"},
1380         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1381         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1382         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1383         {Opt_grpquota, "grpquota"},
1384         {Opt_noquota, "noquota"},
1385         {Opt_quota, "quota"},
1386         {Opt_usrquota, "usrquota"},
1387         {Opt_barrier, "barrier=%u"},
1388         {Opt_barrier, "barrier"},
1389         {Opt_nobarrier, "nobarrier"},
1390         {Opt_i_version, "i_version"},
1391         {Opt_stripe, "stripe=%u"},
1392         {Opt_resize, "resize"},
1393         {Opt_delalloc, "delalloc"},
1394         {Opt_nodelalloc, "nodelalloc"},
1395         {Opt_mblk_io_submit, "mblk_io_submit"},
1396         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1397         {Opt_block_validity, "block_validity"},
1398         {Opt_noblock_validity, "noblock_validity"},
1399         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1400         {Opt_journal_ioprio, "journal_ioprio=%u"},
1401         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1402         {Opt_auto_da_alloc, "auto_da_alloc"},
1403         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1404         {Opt_dioread_nolock, "dioread_nolock"},
1405         {Opt_dioread_lock, "dioread_lock"},
1406         {Opt_discard, "discard"},
1407         {Opt_nodiscard, "nodiscard"},
1408         {Opt_init_itable, "init_itable=%u"},
1409         {Opt_init_itable, "init_itable"},
1410         {Opt_noinit_itable, "noinit_itable"},
1411         {Opt_err, NULL},
1412 };
1413
1414 static ext4_fsblk_t get_sb_block(void **data)
1415 {
1416         ext4_fsblk_t    sb_block;
1417         char            *options = (char *) *data;
1418
1419         if (!options || strncmp(options, "sb=", 3) != 0)
1420                 return 1;       /* Default location */
1421
1422         options += 3;
1423         /* TODO: use simple_strtoll with >32bit ext4 */
1424         sb_block = simple_strtoul(options, &options, 0);
1425         if (*options && *options != ',') {
1426                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1427                        (char *) *data);
1428                 return 1;
1429         }
1430         if (*options == ',')
1431                 options++;
1432         *data = (void *) options;
1433
1434         return sb_block;
1435 }
1436
1437 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1438 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1439         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1440
1441 #ifdef CONFIG_QUOTA
1442 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1443 {
1444         struct ext4_sb_info *sbi = EXT4_SB(sb);
1445         char *qname;
1446
1447         if (sb_any_quota_loaded(sb) &&
1448                 !sbi->s_qf_names[qtype]) {
1449                 ext4_msg(sb, KERN_ERR,
1450                         "Cannot change journaled "
1451                         "quota options when quota turned on");
1452                 return 0;
1453         }
1454         qname = match_strdup(args);
1455         if (!qname) {
1456                 ext4_msg(sb, KERN_ERR,
1457                         "Not enough memory for storing quotafile name");
1458                 return 0;
1459         }
1460         if (sbi->s_qf_names[qtype] &&
1461                 strcmp(sbi->s_qf_names[qtype], qname)) {
1462                 ext4_msg(sb, KERN_ERR,
1463                         "%s quota file already specified", QTYPE2NAME(qtype));
1464                 kfree(qname);
1465                 return 0;
1466         }
1467         sbi->s_qf_names[qtype] = qname;
1468         if (strchr(sbi->s_qf_names[qtype], '/')) {
1469                 ext4_msg(sb, KERN_ERR,
1470                         "quotafile must be on filesystem root");
1471                 kfree(sbi->s_qf_names[qtype]);
1472                 sbi->s_qf_names[qtype] = NULL;
1473                 return 0;
1474         }
1475         set_opt(sb, QUOTA);
1476         return 1;
1477 }
1478
1479 static int clear_qf_name(struct super_block *sb, int qtype)
1480 {
1481
1482         struct ext4_sb_info *sbi = EXT4_SB(sb);
1483
1484         if (sb_any_quota_loaded(sb) &&
1485                 sbi->s_qf_names[qtype]) {
1486                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1487                         " when quota turned on");
1488                 return 0;
1489         }
1490         /*
1491          * The space will be released later when all options are confirmed
1492          * to be correct
1493          */
1494         sbi->s_qf_names[qtype] = NULL;
1495         return 1;
1496 }
1497 #endif
1498
1499 static int parse_options(char *options, struct super_block *sb,
1500                          unsigned long *journal_devnum,
1501                          unsigned int *journal_ioprio,
1502                          ext4_fsblk_t *n_blocks_count, int is_remount)
1503 {
1504         struct ext4_sb_info *sbi = EXT4_SB(sb);
1505         char *p;
1506         substring_t args[MAX_OPT_ARGS];
1507         int data_opt = 0;
1508         int option;
1509 #ifdef CONFIG_QUOTA
1510         int qfmt;
1511 #endif
1512
1513         if (!options)
1514                 return 1;
1515
1516         while ((p = strsep(&options, ",")) != NULL) {
1517                 int token;
1518                 if (!*p)
1519                         continue;
1520
1521                 /*
1522                  * Initialize args struct so we know whether arg was
1523                  * found; some options take optional arguments.
1524                  */
1525                 args[0].to = args[0].from = NULL;
1526                 token = match_token(p, tokens, args);
1527                 switch (token) {
1528                 case Opt_bsd_df:
1529                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1530                         clear_opt(sb, MINIX_DF);
1531                         break;
1532                 case Opt_minix_df:
1533                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1534                         set_opt(sb, MINIX_DF);
1535
1536                         break;
1537                 case Opt_grpid:
1538                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1539                         set_opt(sb, GRPID);
1540
1541                         break;
1542                 case Opt_nogrpid:
1543                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1544                         clear_opt(sb, GRPID);
1545
1546                         break;
1547                 case Opt_resuid:
1548                         if (match_int(&args[0], &option))
1549                                 return 0;
1550                         sbi->s_resuid = option;
1551                         break;
1552                 case Opt_resgid:
1553                         if (match_int(&args[0], &option))
1554                                 return 0;
1555                         sbi->s_resgid = option;
1556                         break;
1557                 case Opt_sb:
1558                         /* handled by get_sb_block() instead of here */
1559                         /* *sb_block = match_int(&args[0]); */
1560                         break;
1561                 case Opt_err_panic:
1562                         clear_opt(sb, ERRORS_CONT);
1563                         clear_opt(sb, ERRORS_RO);
1564                         set_opt(sb, ERRORS_PANIC);
1565                         break;
1566                 case Opt_err_ro:
1567                         clear_opt(sb, ERRORS_CONT);
1568                         clear_opt(sb, ERRORS_PANIC);
1569                         set_opt(sb, ERRORS_RO);
1570                         break;
1571                 case Opt_err_cont:
1572                         clear_opt(sb, ERRORS_RO);
1573                         clear_opt(sb, ERRORS_PANIC);
1574                         set_opt(sb, ERRORS_CONT);
1575                         break;
1576                 case Opt_nouid32:
1577                         set_opt(sb, NO_UID32);
1578                         break;
1579                 case Opt_debug:
1580                         set_opt(sb, DEBUG);
1581                         break;
1582                 case Opt_oldalloc:
1583                         ext4_msg(sb, KERN_WARNING,
1584                                  "Ignoring deprecated oldalloc option");
1585                         break;
1586                 case Opt_orlov:
1587                         ext4_msg(sb, KERN_WARNING,
1588                                  "Ignoring deprecated orlov option");
1589                         break;
1590 #ifdef CONFIG_EXT4_FS_XATTR
1591                 case Opt_user_xattr:
1592                         set_opt(sb, XATTR_USER);
1593                         break;
1594                 case Opt_nouser_xattr:
1595                         clear_opt(sb, XATTR_USER);
1596                         break;
1597 #else
1598                 case Opt_user_xattr:
1599                 case Opt_nouser_xattr:
1600                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1601                         break;
1602 #endif
1603 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1604                 case Opt_acl:
1605                         set_opt(sb, POSIX_ACL);
1606                         break;
1607                 case Opt_noacl:
1608                         clear_opt(sb, POSIX_ACL);
1609                         break;
1610 #else
1611                 case Opt_acl:
1612                 case Opt_noacl:
1613                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1614                         break;
1615 #endif
1616                 case Opt_journal_update:
1617                         /* @@@ FIXME */
1618                         /* Eventually we will want to be able to create
1619                            a journal file here.  For now, only allow the
1620                            user to specify an existing inode to be the
1621                            journal file. */
1622                         if (is_remount) {
1623                                 ext4_msg(sb, KERN_ERR,
1624                                          "Cannot specify journal on remount");
1625                                 return 0;
1626                         }
1627                         set_opt(sb, UPDATE_JOURNAL);
1628                         break;
1629                 case Opt_journal_dev:
1630                         if (is_remount) {
1631                                 ext4_msg(sb, KERN_ERR,
1632                                         "Cannot specify journal on remount");
1633                                 return 0;
1634                         }
1635                         if (match_int(&args[0], &option))
1636                                 return 0;
1637                         *journal_devnum = option;
1638                         break;
1639                 case Opt_journal_checksum:
1640                         set_opt(sb, JOURNAL_CHECKSUM);
1641                         break;
1642                 case Opt_journal_async_commit:
1643                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1644                         set_opt(sb, JOURNAL_CHECKSUM);
1645                         break;
1646                 case Opt_noload:
1647                         set_opt(sb, NOLOAD);
1648                         break;
1649                 case Opt_commit:
1650                         if (match_int(&args[0], &option))
1651                                 return 0;
1652                         if (option < 0)
1653                                 return 0;
1654                         if (option == 0)
1655                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1656                         sbi->s_commit_interval = HZ * option;
1657                         break;
1658                 case Opt_max_batch_time:
1659                         if (match_int(&args[0], &option))
1660                                 return 0;
1661                         if (option < 0)
1662                                 return 0;
1663                         if (option == 0)
1664                                 option = EXT4_DEF_MAX_BATCH_TIME;
1665                         sbi->s_max_batch_time = option;
1666                         break;
1667                 case Opt_min_batch_time:
1668                         if (match_int(&args[0], &option))
1669                                 return 0;
1670                         if (option < 0)
1671                                 return 0;
1672                         sbi->s_min_batch_time = option;
1673                         break;
1674                 case Opt_data_journal:
1675                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1676                         goto datacheck;
1677                 case Opt_data_ordered:
1678                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1679                         goto datacheck;
1680                 case Opt_data_writeback:
1681                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1682                 datacheck:
1683                         if (is_remount) {
1684                                 if (!sbi->s_journal)
1685                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1686                                 else if (test_opt(sb, DATA_FLAGS) != data_opt) {
1687                                         ext4_msg(sb, KERN_ERR,
1688                                                 "Cannot change data mode on remount");
1689                                         return 0;
1690                                 }
1691                         } else {
1692                                 clear_opt(sb, DATA_FLAGS);
1693                                 sbi->s_mount_opt |= data_opt;
1694                         }
1695                         break;
1696                 case Opt_data_err_abort:
1697                         set_opt(sb, DATA_ERR_ABORT);
1698                         break;
1699                 case Opt_data_err_ignore:
1700                         clear_opt(sb, DATA_ERR_ABORT);
1701                         break;
1702 #ifdef CONFIG_QUOTA
1703                 case Opt_usrjquota:
1704                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1705                                 return 0;
1706                         break;
1707                 case Opt_grpjquota:
1708                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1709                                 return 0;
1710                         break;
1711                 case Opt_offusrjquota:
1712                         if (!clear_qf_name(sb, USRQUOTA))
1713                                 return 0;
1714                         break;
1715                 case Opt_offgrpjquota:
1716                         if (!clear_qf_name(sb, GRPQUOTA))
1717                                 return 0;
1718                         break;
1719
1720                 case Opt_jqfmt_vfsold:
1721                         qfmt = QFMT_VFS_OLD;
1722                         goto set_qf_format;
1723                 case Opt_jqfmt_vfsv0:
1724                         qfmt = QFMT_VFS_V0;
1725                         goto set_qf_format;
1726                 case Opt_jqfmt_vfsv1:
1727                         qfmt = QFMT_VFS_V1;
1728 set_qf_format:
1729                         if (sb_any_quota_loaded(sb) &&
1730                             sbi->s_jquota_fmt != qfmt) {
1731                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1732                                         "journaled quota options when "
1733                                         "quota turned on");
1734                                 return 0;
1735                         }
1736                         sbi->s_jquota_fmt = qfmt;
1737                         break;
1738                 case Opt_quota:
1739                 case Opt_usrquota:
1740                         set_opt(sb, QUOTA);
1741                         set_opt(sb, USRQUOTA);
1742                         break;
1743                 case Opt_grpquota:
1744                         set_opt(sb, QUOTA);
1745                         set_opt(sb, GRPQUOTA);
1746                         break;
1747                 case Opt_noquota:
1748                         if (sb_any_quota_loaded(sb)) {
1749                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1750                                         "options when quota turned on");
1751                                 return 0;
1752                         }
1753                         clear_opt(sb, QUOTA);
1754                         clear_opt(sb, USRQUOTA);
1755                         clear_opt(sb, GRPQUOTA);
1756                         break;
1757 #else
1758                 case Opt_quota:
1759                 case Opt_usrquota:
1760                 case Opt_grpquota:
1761                         ext4_msg(sb, KERN_ERR,
1762                                 "quota options not supported");
1763                         break;
1764                 case Opt_usrjquota:
1765                 case Opt_grpjquota:
1766                 case Opt_offusrjquota:
1767                 case Opt_offgrpjquota:
1768                 case Opt_jqfmt_vfsold:
1769                 case Opt_jqfmt_vfsv0:
1770                 case Opt_jqfmt_vfsv1:
1771                         ext4_msg(sb, KERN_ERR,
1772                                 "journaled quota options not supported");
1773                         break;
1774                 case Opt_noquota:
1775                         break;
1776 #endif
1777                 case Opt_abort:
1778                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1779                         break;
1780                 case Opt_nobarrier:
1781                         clear_opt(sb, BARRIER);
1782                         break;
1783                 case Opt_barrier:
1784                         if (args[0].from) {
1785                                 if (match_int(&args[0], &option))
1786                                         return 0;
1787                         } else
1788                                 option = 1;     /* No argument, default to 1 */
1789                         if (option)
1790                                 set_opt(sb, BARRIER);
1791                         else
1792                                 clear_opt(sb, BARRIER);
1793                         break;
1794                 case Opt_ignore:
1795                         break;
1796                 case Opt_resize:
1797                         if (!is_remount) {
1798                                 ext4_msg(sb, KERN_ERR,
1799                                         "resize option only available "
1800                                         "for remount");
1801                                 return 0;
1802                         }
1803                         if (match_int(&args[0], &option) != 0)
1804                                 return 0;
1805                         *n_blocks_count = option;
1806                         break;
1807                 case Opt_nobh:
1808                         ext4_msg(sb, KERN_WARNING,
1809                                  "Ignoring deprecated nobh option");
1810                         break;
1811                 case Opt_bh:
1812                         ext4_msg(sb, KERN_WARNING,
1813                                  "Ignoring deprecated bh option");
1814                         break;
1815                 case Opt_i_version:
1816                         set_opt(sb, I_VERSION);
1817                         sb->s_flags |= MS_I_VERSION;
1818                         break;
1819                 case Opt_nodelalloc:
1820                         clear_opt(sb, DELALLOC);
1821                         clear_opt2(sb, EXPLICIT_DELALLOC);
1822                         break;
1823                 case Opt_mblk_io_submit:
1824                         set_opt(sb, MBLK_IO_SUBMIT);
1825                         break;
1826                 case Opt_nomblk_io_submit:
1827                         clear_opt(sb, MBLK_IO_SUBMIT);
1828                         break;
1829                 case Opt_stripe:
1830                         if (match_int(&args[0], &option))
1831                                 return 0;
1832                         if (option < 0)
1833                                 return 0;
1834                         sbi->s_stripe = option;
1835                         break;
1836                 case Opt_delalloc:
1837                         set_opt(sb, DELALLOC);
1838                         set_opt2(sb, EXPLICIT_DELALLOC);
1839                         break;
1840                 case Opt_block_validity:
1841                         set_opt(sb, BLOCK_VALIDITY);
1842                         break;
1843                 case Opt_noblock_validity:
1844                         clear_opt(sb, BLOCK_VALIDITY);
1845                         break;
1846                 case Opt_inode_readahead_blks:
1847                         if (match_int(&args[0], &option))
1848                                 return 0;
1849                         if (option < 0 || option > (1 << 30))
1850                                 return 0;
1851                         if (option && !is_power_of_2(option)) {
1852                                 ext4_msg(sb, KERN_ERR,
1853                                          "EXT4-fs: inode_readahead_blks"
1854                                          " must be a power of 2");
1855                                 return 0;
1856                         }
1857                         sbi->s_inode_readahead_blks = option;
1858                         break;
1859                 case Opt_journal_ioprio:
1860                         if (match_int(&args[0], &option))
1861                                 return 0;
1862                         if (option < 0 || option > 7)
1863                                 break;
1864                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1865                                                             option);
1866                         break;
1867                 case Opt_noauto_da_alloc:
1868                         set_opt(sb, NO_AUTO_DA_ALLOC);
1869                         break;
1870                 case Opt_auto_da_alloc:
1871                         if (args[0].from) {
1872                                 if (match_int(&args[0], &option))
1873                                         return 0;
1874                         } else
1875                                 option = 1;     /* No argument, default to 1 */
1876                         if (option)
1877                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1878                         else
1879                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1880                         break;
1881                 case Opt_discard:
1882                         set_opt(sb, DISCARD);
1883                         break;
1884                 case Opt_nodiscard:
1885                         clear_opt(sb, DISCARD);
1886                         break;
1887                 case Opt_dioread_nolock:
1888                         set_opt(sb, DIOREAD_NOLOCK);
1889                         break;
1890                 case Opt_dioread_lock:
1891                         clear_opt(sb, DIOREAD_NOLOCK);
1892                         break;
1893                 case Opt_init_itable:
1894                         set_opt(sb, INIT_INODE_TABLE);
1895                         if (args[0].from) {
1896                                 if (match_int(&args[0], &option))
1897                                         return 0;
1898                         } else
1899                                 option = EXT4_DEF_LI_WAIT_MULT;
1900                         if (option < 0)
1901                                 return 0;
1902                         sbi->s_li_wait_mult = option;
1903                         break;
1904                 case Opt_noinit_itable:
1905                         clear_opt(sb, INIT_INODE_TABLE);
1906                         break;
1907                 default:
1908                         ext4_msg(sb, KERN_ERR,
1909                                "Unrecognized mount option \"%s\" "
1910                                "or missing value", p);
1911                         return 0;
1912                 }
1913         }
1914 #ifdef CONFIG_QUOTA
1915         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1916                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1917                         clear_opt(sb, USRQUOTA);
1918
1919                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1920                         clear_opt(sb, GRPQUOTA);
1921
1922                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1923                         ext4_msg(sb, KERN_ERR, "old and new quota "
1924                                         "format mixing");
1925                         return 0;
1926                 }
1927
1928                 if (!sbi->s_jquota_fmt) {
1929                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1930                                         "not specified");
1931                         return 0;
1932                 }
1933         } else {
1934                 if (sbi->s_jquota_fmt) {
1935                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1936                                         "specified with no journaling "
1937                                         "enabled");
1938                         return 0;
1939                 }
1940         }
1941 #endif
1942         return 1;
1943 }
1944
1945 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1946                             int read_only)
1947 {
1948         struct ext4_sb_info *sbi = EXT4_SB(sb);
1949         int res = 0;
1950
1951         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1952                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1953                          "forcing read-only mode");
1954                 res = MS_RDONLY;
1955         }
1956         if (read_only)
1957                 goto done;
1958         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1959                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1960                          "running e2fsck is recommended");
1961         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1962                 ext4_msg(sb, KERN_WARNING,
1963                          "warning: mounting fs with errors, "
1964                          "running e2fsck is recommended");
1965         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1966                  le16_to_cpu(es->s_mnt_count) >=
1967                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1968                 ext4_msg(sb, KERN_WARNING,
1969                          "warning: maximal mount count reached, "
1970                          "running e2fsck is recommended");
1971         else if (le32_to_cpu(es->s_checkinterval) &&
1972                 (le32_to_cpu(es->s_lastcheck) +
1973                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1974                 ext4_msg(sb, KERN_WARNING,
1975                          "warning: checktime reached, "
1976                          "running e2fsck is recommended");
1977         if (!sbi->s_journal)
1978                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1979         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1980                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1981         le16_add_cpu(&es->s_mnt_count, 1);
1982         es->s_mtime = cpu_to_le32(get_seconds());
1983         ext4_update_dynamic_rev(sb);
1984         if (sbi->s_journal)
1985                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1986
1987         ext4_commit_super(sb, 1);
1988 done:
1989         if (test_opt(sb, DEBUG))
1990                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1991                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1992                         sb->s_blocksize,
1993                         sbi->s_groups_count,
1994                         EXT4_BLOCKS_PER_GROUP(sb),
1995                         EXT4_INODES_PER_GROUP(sb),
1996                         sbi->s_mount_opt, sbi->s_mount_opt2);
1997
1998         cleancache_init_fs(sb);
1999         return res;
2000 }
2001
2002 static int ext4_fill_flex_info(struct super_block *sb)
2003 {
2004         struct ext4_sb_info *sbi = EXT4_SB(sb);
2005         struct ext4_group_desc *gdp = NULL;
2006         ext4_group_t flex_group_count;
2007         ext4_group_t flex_group;
2008         unsigned int groups_per_flex = 0;
2009         size_t size;
2010         int i;
2011
2012         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2013         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2014                 sbi->s_log_groups_per_flex = 0;
2015                 return 1;
2016         }
2017         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
2018
2019         /* We allocate both existing and potentially added groups */
2020         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2021                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2022                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2023         size = flex_group_count * sizeof(struct flex_groups);
2024         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2025         if (sbi->s_flex_groups == NULL) {
2026                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2027                          flex_group_count);
2028                 goto failed;
2029         }
2030
2031         for (i = 0; i < sbi->s_groups_count; i++) {
2032                 gdp = ext4_get_group_desc(sb, i, NULL);
2033
2034                 flex_group = ext4_flex_group(sbi, i);
2035                 atomic_add(ext4_free_inodes_count(sb, gdp),
2036                            &sbi->s_flex_groups[flex_group].free_inodes);
2037                 atomic_add(ext4_free_group_clusters(sb, gdp),
2038                            &sbi->s_flex_groups[flex_group].free_clusters);
2039                 atomic_add(ext4_used_dirs_count(sb, gdp),
2040                            &sbi->s_flex_groups[flex_group].used_dirs);
2041         }
2042
2043         return 1;
2044 failed:
2045         return 0;
2046 }
2047
2048 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2049                             struct ext4_group_desc *gdp)
2050 {
2051         __u16 crc = 0;
2052
2053         if (sbi->s_es->s_feature_ro_compat &
2054             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2055                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2056                 __le32 le_group = cpu_to_le32(block_group);
2057
2058                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2059                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2060                 crc = crc16(crc, (__u8 *)gdp, offset);
2061                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2062                 /* for checksum of struct ext4_group_desc do the rest...*/
2063                 if ((sbi->s_es->s_feature_incompat &
2064                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2065                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2066                         crc = crc16(crc, (__u8 *)gdp + offset,
2067                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2068                                         offset);
2069         }
2070
2071         return cpu_to_le16(crc);
2072 }
2073
2074 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2075                                 struct ext4_group_desc *gdp)
2076 {
2077         if ((sbi->s_es->s_feature_ro_compat &
2078              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2079             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2080                 return 0;
2081
2082         return 1;
2083 }
2084
2085 /* Called at mount-time, super-block is locked */
2086 static int ext4_check_descriptors(struct super_block *sb,
2087                                   ext4_group_t *first_not_zeroed)
2088 {
2089         struct ext4_sb_info *sbi = EXT4_SB(sb);
2090         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2091         ext4_fsblk_t last_block;
2092         ext4_fsblk_t block_bitmap;
2093         ext4_fsblk_t inode_bitmap;
2094         ext4_fsblk_t inode_table;
2095         int flexbg_flag = 0;
2096         ext4_group_t i, grp = sbi->s_groups_count;
2097
2098         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2099                 flexbg_flag = 1;
2100
2101         ext4_debug("Checking group descriptors");
2102
2103         for (i = 0; i < sbi->s_groups_count; i++) {
2104                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2105
2106                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2107                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2108                 else
2109                         last_block = first_block +
2110                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2111
2112                 if ((grp == sbi->s_groups_count) &&
2113                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2114                         grp = i;
2115
2116                 block_bitmap = ext4_block_bitmap(sb, gdp);
2117                 if (block_bitmap < first_block || block_bitmap > last_block) {
2118                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2119                                "Block bitmap for group %u not in group "
2120                                "(block %llu)!", i, block_bitmap);
2121                         return 0;
2122                 }
2123                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2124                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2125                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2126                                "Inode bitmap for group %u not in group "
2127                                "(block %llu)!", i, inode_bitmap);
2128                         return 0;
2129                 }
2130                 inode_table = ext4_inode_table(sb, gdp);
2131                 if (inode_table < first_block ||
2132                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2133                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2134                                "Inode table for group %u not in group "
2135                                "(block %llu)!", i, inode_table);
2136                         return 0;
2137                 }
2138                 ext4_lock_group(sb, i);
2139                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2140                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2141                                  "Checksum for group %u failed (%u!=%u)",
2142                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2143                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2144                         if (!(sb->s_flags & MS_RDONLY)) {
2145                                 ext4_unlock_group(sb, i);
2146                                 return 0;
2147                         }
2148                 }
2149                 ext4_unlock_group(sb, i);
2150                 if (!flexbg_flag)
2151                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2152         }
2153         if (NULL != first_not_zeroed)
2154                 *first_not_zeroed = grp;
2155
2156         ext4_free_blocks_count_set(sbi->s_es,
2157                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2158         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2159         return 1;
2160 }
2161
2162 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2163  * the superblock) which were deleted from all directories, but held open by
2164  * a process at the time of a crash.  We walk the list and try to delete these
2165  * inodes at recovery time (only with a read-write filesystem).
2166  *
2167  * In order to keep the orphan inode chain consistent during traversal (in
2168  * case of crash during recovery), we link each inode into the superblock
2169  * orphan list_head and handle it the same way as an inode deletion during
2170  * normal operation (which journals the operations for us).
2171  *
2172  * We only do an iget() and an iput() on each inode, which is very safe if we
2173  * accidentally point at an in-use or already deleted inode.  The worst that
2174  * can happen in this case is that we get a "bit already cleared" message from
2175  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2176  * e2fsck was run on this filesystem, and it must have already done the orphan
2177  * inode cleanup for us, so we can safely abort without any further action.
2178  */
2179 static void ext4_orphan_cleanup(struct super_block *sb,
2180                                 struct ext4_super_block *es)
2181 {
2182         unsigned int s_flags = sb->s_flags;
2183         int nr_orphans = 0, nr_truncates = 0;
2184 #ifdef CONFIG_QUOTA
2185         int i;
2186 #endif
2187         if (!es->s_last_orphan) {
2188                 jbd_debug(4, "no orphan inodes to clean up\n");
2189                 return;
2190         }
2191
2192         if (bdev_read_only(sb->s_bdev)) {
2193                 ext4_msg(sb, KERN_ERR, "write access "
2194                         "unavailable, skipping orphan cleanup");
2195                 return;
2196         }
2197
2198         /* Check if feature set would not allow a r/w mount */
2199         if (!ext4_feature_set_ok(sb, 0)) {
2200                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2201                          "unknown ROCOMPAT features");
2202                 return;
2203         }
2204
2205         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2206                 if (es->s_last_orphan)
2207                         jbd_debug(1, "Errors on filesystem, "
2208                                   "clearing orphan list.\n");
2209                 es->s_last_orphan = 0;
2210                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2211                 return;
2212         }
2213
2214         if (s_flags & MS_RDONLY) {
2215                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2216                 sb->s_flags &= ~MS_RDONLY;
2217         }
2218 #ifdef CONFIG_QUOTA
2219         /* Needed for iput() to work correctly and not trash data */
2220         sb->s_flags |= MS_ACTIVE;
2221         /* Turn on quotas so that they are updated correctly */
2222         for (i = 0; i < MAXQUOTAS; i++) {
2223                 if (EXT4_SB(sb)->s_qf_names[i]) {
2224                         int ret = ext4_quota_on_mount(sb, i);
2225                         if (ret < 0)
2226                                 ext4_msg(sb, KERN_ERR,
2227                                         "Cannot turn on journaled "
2228                                         "quota: error %d", ret);
2229                 }
2230         }
2231 #endif
2232
2233         while (es->s_last_orphan) {
2234                 struct inode *inode;
2235
2236                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2237                 if (IS_ERR(inode)) {
2238                         es->s_last_orphan = 0;
2239                         break;
2240                 }
2241
2242                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2243                 dquot_initialize(inode);
2244                 if (inode->i_nlink) {
2245                         ext4_msg(sb, KERN_DEBUG,
2246                                 "%s: truncating inode %lu to %lld bytes",
2247                                 __func__, inode->i_ino, inode->i_size);
2248                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2249                                   inode->i_ino, inode->i_size);
2250                         ext4_truncate(inode);
2251                         nr_truncates++;
2252                 } else {
2253                         ext4_msg(sb, KERN_DEBUG,
2254                                 "%s: deleting unreferenced inode %lu",
2255                                 __func__, inode->i_ino);
2256                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2257                                   inode->i_ino);
2258                         nr_orphans++;
2259                 }
2260                 iput(inode);  /* The delete magic happens here! */
2261         }
2262
2263 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2264
2265         if (nr_orphans)
2266                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2267                        PLURAL(nr_orphans));
2268         if (nr_truncates)
2269                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2270                        PLURAL(nr_truncates));
2271 #ifdef CONFIG_QUOTA
2272         /* Turn quotas off */
2273         for (i = 0; i < MAXQUOTAS; i++) {
2274                 if (sb_dqopt(sb)->files[i])
2275                         dquot_quota_off(sb, i);
2276         }
2277 #endif
2278         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2279 }
2280
2281 /*
2282  * Maximal extent format file size.
2283  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2284  * extent format containers, within a sector_t, and within i_blocks
2285  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2286  * so that won't be a limiting factor.
2287  *
2288  * However there is other limiting factor. We do store extents in the form
2289  * of starting block and length, hence the resulting length of the extent
2290  * covering maximum file size must fit into on-disk format containers as
2291  * well. Given that length is always by 1 unit bigger than max unit (because
2292  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2293  *
2294  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2295  */
2296 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2297 {
2298         loff_t res;
2299         loff_t upper_limit = MAX_LFS_FILESIZE;
2300
2301         /* small i_blocks in vfs inode? */
2302         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2303                 /*
2304                  * CONFIG_LBDAF is not enabled implies the inode
2305                  * i_block represent total blocks in 512 bytes
2306                  * 32 == size of vfs inode i_blocks * 8
2307                  */
2308                 upper_limit = (1LL << 32) - 1;
2309
2310                 /* total blocks in file system block size */
2311                 upper_limit >>= (blkbits - 9);
2312                 upper_limit <<= blkbits;
2313         }
2314
2315         /*
2316          * 32-bit extent-start container, ee_block. We lower the maxbytes
2317          * by one fs block, so ee_len can cover the extent of maximum file
2318          * size
2319          */
2320         res = (1LL << 32) - 1;
2321         res <<= blkbits;
2322
2323         /* Sanity check against vm- & vfs- imposed limits */
2324         if (res > upper_limit)
2325                 res = upper_limit;
2326
2327         return res;
2328 }
2329
2330 /*
2331  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2332  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2333  * We need to be 1 filesystem block less than the 2^48 sector limit.
2334  */
2335 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2336 {
2337         loff_t res = EXT4_NDIR_BLOCKS;
2338         int meta_blocks;
2339         loff_t upper_limit;
2340         /* This is calculated to be the largest file size for a dense, block
2341          * mapped file such that the file's total number of 512-byte sectors,
2342          * including data and all indirect blocks, does not exceed (2^48 - 1).
2343          *
2344          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2345          * number of 512-byte sectors of the file.
2346          */
2347
2348         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2349                 /*
2350                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2351                  * the inode i_block field represents total file blocks in
2352                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2353                  */
2354                 upper_limit = (1LL << 32) - 1;
2355
2356                 /* total blocks in file system block size */
2357                 upper_limit >>= (bits - 9);
2358
2359         } else {
2360                 /*
2361                  * We use 48 bit ext4_inode i_blocks
2362                  * With EXT4_HUGE_FILE_FL set the i_blocks
2363                  * represent total number of blocks in
2364                  * file system block size
2365                  */
2366                 upper_limit = (1LL << 48) - 1;
2367
2368         }
2369
2370         /* indirect blocks */
2371         meta_blocks = 1;
2372         /* double indirect blocks */
2373         meta_blocks += 1 + (1LL << (bits-2));
2374         /* tripple indirect blocks */
2375         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2376
2377         upper_limit -= meta_blocks;
2378         upper_limit <<= bits;
2379
2380         res += 1LL << (bits-2);
2381         res += 1LL << (2*(bits-2));
2382         res += 1LL << (3*(bits-2));
2383         res <<= bits;
2384         if (res > upper_limit)
2385                 res = upper_limit;
2386
2387         if (res > MAX_LFS_FILESIZE)
2388                 res = MAX_LFS_FILESIZE;
2389
2390         return res;
2391 }
2392
2393 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2394                                    ext4_fsblk_t logical_sb_block, int nr)
2395 {
2396         struct ext4_sb_info *sbi = EXT4_SB(sb);
2397         ext4_group_t bg, first_meta_bg;
2398         int has_super = 0;
2399
2400         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2401
2402         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2403             nr < first_meta_bg)
2404                 return logical_sb_block + nr + 1;
2405         bg = sbi->s_desc_per_block * nr;
2406         if (ext4_bg_has_super(sb, bg))
2407                 has_super = 1;
2408
2409         return (has_super + ext4_group_first_block_no(sb, bg));
2410 }
2411
2412 /**
2413  * ext4_get_stripe_size: Get the stripe size.
2414  * @sbi: In memory super block info
2415  *
2416  * If we have specified it via mount option, then
2417  * use the mount option value. If the value specified at mount time is
2418  * greater than the blocks per group use the super block value.
2419  * If the super block value is greater than blocks per group return 0.
2420  * Allocator needs it be less than blocks per group.
2421  *
2422  */
2423 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2424 {
2425         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2426         unsigned long stripe_width =
2427                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2428         int ret;
2429
2430         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2431                 ret = sbi->s_stripe;
2432         else if (stripe_width <= sbi->s_blocks_per_group)
2433                 ret = stripe_width;
2434         else if (stride <= sbi->s_blocks_per_group)
2435                 ret = stride;
2436         else
2437                 ret = 0;
2438
2439         /*
2440          * If the stripe width is 1, this makes no sense and
2441          * we set it to 0 to turn off stripe handling code.
2442          */
2443         if (ret <= 1)
2444                 ret = 0;
2445
2446         return ret;
2447 }
2448
2449 /* sysfs supprt */
2450
2451 struct ext4_attr {
2452         struct attribute attr;
2453         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2454         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2455                          const char *, size_t);
2456         int offset;
2457 };
2458
2459 static int parse_strtoul(const char *buf,
2460                 unsigned long max, unsigned long *value)
2461 {
2462         char *endp;
2463
2464         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2465         endp = skip_spaces(endp);
2466         if (*endp || *value > max)
2467                 return -EINVAL;
2468
2469         return 0;
2470 }
2471
2472 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2473                                               struct ext4_sb_info *sbi,
2474                                               char *buf)
2475 {
2476         return snprintf(buf, PAGE_SIZE, "%llu\n",
2477                 (s64) EXT4_C2B(sbi,
2478                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2479 }
2480
2481 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2482                                          struct ext4_sb_info *sbi, char *buf)
2483 {
2484         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2485
2486         if (!sb->s_bdev->bd_part)
2487                 return snprintf(buf, PAGE_SIZE, "0\n");
2488         return snprintf(buf, PAGE_SIZE, "%lu\n",
2489                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2490                          sbi->s_sectors_written_start) >> 1);
2491 }
2492
2493 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2494                                           struct ext4_sb_info *sbi, char *buf)
2495 {
2496         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2497
2498         if (!sb->s_bdev->bd_part)
2499                 return snprintf(buf, PAGE_SIZE, "0\n");
2500         return snprintf(buf, PAGE_SIZE, "%llu\n",
2501                         (unsigned long long)(sbi->s_kbytes_written +
2502                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2503                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2504 }
2505
2506 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2507                                       struct ext4_sb_info *sbi, char *buf)
2508 {
2509         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2510 }
2511
2512 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2513                                         struct ext4_sb_info *sbi, char *buf)
2514 {
2515         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2516 }
2517
2518 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2519                                           struct ext4_sb_info *sbi,
2520                                           const char *buf, size_t count)
2521 {
2522         unsigned long t;
2523
2524         if (parse_strtoul(buf, 0x40000000, &t))
2525                 return -EINVAL;
2526
2527         if (t && !is_power_of_2(t))
2528                 return -EINVAL;
2529
2530         sbi->s_inode_readahead_blks = t;
2531         return count;
2532 }
2533
2534 static ssize_t sbi_ui_show(struct ext4_attr *a,
2535                            struct ext4_sb_info *sbi, char *buf)
2536 {
2537         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2538
2539         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2540 }
2541
2542 static ssize_t sbi_ui_store(struct ext4_attr *a,
2543                             struct ext4_sb_info *sbi,
2544                             const char *buf, size_t count)
2545 {
2546         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2547         unsigned long t;
2548
2549         if (parse_strtoul(buf, 0xffffffff, &t))
2550                 return -EINVAL;
2551         *ui = t;
2552         return count;
2553 }
2554
2555 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2556 static struct ext4_attr ext4_attr_##_name = {                   \
2557         .attr = {.name = __stringify(_name), .mode = _mode },   \
2558         .show   = _show,                                        \
2559         .store  = _store,                                       \
2560         .offset = offsetof(struct ext4_sb_info, _elname),       \
2561 }
2562 #define EXT4_ATTR(name, mode, show, store) \
2563 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2564
2565 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2566 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2567 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2568 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2569         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2570 #define ATTR_LIST(name) &ext4_attr_##name.attr
2571
2572 EXT4_RO_ATTR(delayed_allocation_blocks);
2573 EXT4_RO_ATTR(session_write_kbytes);
2574 EXT4_RO_ATTR(lifetime_write_kbytes);
2575 EXT4_RO_ATTR(extent_cache_hits);
2576 EXT4_RO_ATTR(extent_cache_misses);
2577 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2578                  inode_readahead_blks_store, s_inode_readahead_blks);
2579 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2580 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2581 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2582 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2583 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2584 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2585 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2586 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2587
2588 static struct attribute *ext4_attrs[] = {
2589         ATTR_LIST(delayed_allocation_blocks),
2590         ATTR_LIST(session_write_kbytes),
2591         ATTR_LIST(lifetime_write_kbytes),
2592         ATTR_LIST(extent_cache_hits),
2593         ATTR_LIST(extent_cache_misses),
2594         ATTR_LIST(inode_readahead_blks),
2595         ATTR_LIST(inode_goal),
2596         ATTR_LIST(mb_stats),
2597         ATTR_LIST(mb_max_to_scan),
2598         ATTR_LIST(mb_min_to_scan),
2599         ATTR_LIST(mb_order2_req),
2600         ATTR_LIST(mb_stream_req),
2601         ATTR_LIST(mb_group_prealloc),
2602         ATTR_LIST(max_writeback_mb_bump),
2603         NULL,
2604 };
2605
2606 /* Features this copy of ext4 supports */
2607 EXT4_INFO_ATTR(lazy_itable_init);
2608 EXT4_INFO_ATTR(batched_discard);
2609
2610 static struct attribute *ext4_feat_attrs[] = {
2611         ATTR_LIST(lazy_itable_init),
2612         ATTR_LIST(batched_discard),
2613         NULL,
2614 };
2615
2616 static ssize_t ext4_attr_show(struct kobject *kobj,
2617                               struct attribute *attr, char *buf)
2618 {
2619         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2620                                                 s_kobj);
2621         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2622
2623         return a->show ? a->show(a, sbi, buf) : 0;
2624 }
2625
2626 static ssize_t ext4_attr_store(struct kobject *kobj,
2627                                struct attribute *attr,
2628                                const char *buf, size_t len)
2629 {
2630         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2631                                                 s_kobj);
2632         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2633
2634         return a->store ? a->store(a, sbi, buf, len) : 0;
2635 }
2636
2637 static void ext4_sb_release(struct kobject *kobj)
2638 {
2639         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2640                                                 s_kobj);
2641         complete(&sbi->s_kobj_unregister);
2642 }
2643
2644 static const struct sysfs_ops ext4_attr_ops = {
2645         .show   = ext4_attr_show,
2646         .store  = ext4_attr_store,
2647 };
2648
2649 static struct kobj_type ext4_ktype = {
2650         .default_attrs  = ext4_attrs,
2651         .sysfs_ops      = &ext4_attr_ops,
2652         .release        = ext4_sb_release,
2653 };
2654
2655 static void ext4_feat_release(struct kobject *kobj)
2656 {
2657         complete(&ext4_feat->f_kobj_unregister);
2658 }
2659
2660 static struct kobj_type ext4_feat_ktype = {
2661         .default_attrs  = ext4_feat_attrs,
2662         .sysfs_ops      = &ext4_attr_ops,
2663         .release        = ext4_feat_release,
2664 };
2665
2666 /*
2667  * Check whether this filesystem can be mounted based on
2668  * the features present and the RDONLY/RDWR mount requested.
2669  * Returns 1 if this filesystem can be mounted as requested,
2670  * 0 if it cannot be.
2671  */
2672 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2673 {
2674         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2675                 ext4_msg(sb, KERN_ERR,
2676                         "Couldn't mount because of "
2677                         "unsupported optional features (%x)",
2678                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2679                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2680                 return 0;
2681         }
2682
2683         if (readonly)
2684                 return 1;
2685
2686         /* Check that feature set is OK for a read-write mount */
2687         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2688                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2689                          "unsupported optional features (%x)",
2690                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2691                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2692                 return 0;
2693         }
2694         /*
2695          * Large file size enabled file system can only be mounted
2696          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2697          */
2698         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2699                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2700                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2701                                  "cannot be mounted RDWR without "
2702                                  "CONFIG_LBDAF");
2703                         return 0;
2704                 }
2705         }
2706         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2707             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2708                 ext4_msg(sb, KERN_ERR,
2709                          "Can't support bigalloc feature without "
2710                          "extents feature\n");
2711                 return 0;
2712         }
2713         return 1;
2714 }
2715
2716 /*
2717  * This function is called once a day if we have errors logged
2718  * on the file system
2719  */
2720 static void print_daily_error_info(unsigned long arg)
2721 {
2722         struct super_block *sb = (struct super_block *) arg;
2723         struct ext4_sb_info *sbi;
2724         struct ext4_super_block *es;
2725
2726         sbi = EXT4_SB(sb);
2727         es = sbi->s_es;
2728
2729         if (es->s_error_count)
2730                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2731                          le32_to_cpu(es->s_error_count));
2732         if (es->s_first_error_time) {
2733                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2734                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2735                        (int) sizeof(es->s_first_error_func),
2736                        es->s_first_error_func,
2737                        le32_to_cpu(es->s_first_error_line));
2738                 if (es->s_first_error_ino)
2739                         printk(": inode %u",
2740                                le32_to_cpu(es->s_first_error_ino));
2741                 if (es->s_first_error_block)
2742                         printk(": block %llu", (unsigned long long)
2743                                le64_to_cpu(es->s_first_error_block));
2744                 printk("\n");
2745         }
2746         if (es->s_last_error_time) {
2747                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2748                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2749                        (int) sizeof(es->s_last_error_func),
2750                        es->s_last_error_func,
2751                        le32_to_cpu(es->s_last_error_line));
2752                 if (es->s_last_error_ino)
2753                         printk(": inode %u",
2754                                le32_to_cpu(es->s_last_error_ino));
2755                 if (es->s_last_error_block)
2756                         printk(": block %llu", (unsigned long long)
2757                                le64_to_cpu(es->s_last_error_block));
2758                 printk("\n");
2759         }
2760         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2761 }
2762
2763 /* Find next suitable group and run ext4_init_inode_table */
2764 static int ext4_run_li_request(struct ext4_li_request *elr)
2765 {
2766         struct ext4_group_desc *gdp = NULL;
2767         ext4_group_t group, ngroups;
2768         struct super_block *sb;
2769         unsigned long timeout = 0;
2770         int ret = 0;
2771
2772         sb = elr->lr_super;
2773         ngroups = EXT4_SB(sb)->s_groups_count;
2774
2775         for (group = elr->lr_next_group; group < ngroups; group++) {
2776                 gdp = ext4_get_group_desc(sb, group, NULL);
2777                 if (!gdp) {
2778                         ret = 1;
2779                         break;
2780                 }
2781
2782                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2783                         break;
2784         }
2785
2786         if (group == ngroups)
2787                 ret = 1;
2788
2789         if (!ret) {
2790                 timeout = jiffies;
2791                 ret = ext4_init_inode_table(sb, group,
2792                                             elr->lr_timeout ? 0 : 1);
2793                 if (elr->lr_timeout == 0) {
2794                         timeout = (jiffies - timeout) *
2795                                   elr->lr_sbi->s_li_wait_mult;
2796                         elr->lr_timeout = timeout;
2797                 }
2798                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2799                 elr->lr_next_group = group + 1;
2800         }
2801
2802         return ret;
2803 }
2804
2805 /*
2806  * Remove lr_request from the list_request and free the
2807  * request structure. Should be called with li_list_mtx held
2808  */
2809 static void ext4_remove_li_request(struct ext4_li_request *elr)
2810 {
2811         struct ext4_sb_info *sbi;
2812
2813         if (!elr)
2814                 return;
2815
2816         sbi = elr->lr_sbi;
2817
2818         list_del(&elr->lr_request);
2819         sbi->s_li_request = NULL;
2820         kfree(elr);
2821 }
2822
2823 static void ext4_unregister_li_request(struct super_block *sb)
2824 {
2825         mutex_lock(&ext4_li_mtx);
2826         if (!ext4_li_info) {
2827                 mutex_unlock(&ext4_li_mtx);
2828                 return;
2829         }
2830
2831         mutex_lock(&ext4_li_info->li_list_mtx);
2832         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2833         mutex_unlock(&ext4_li_info->li_list_mtx);
2834         mutex_unlock(&ext4_li_mtx);
2835 }
2836
2837 static struct task_struct *ext4_lazyinit_task;
2838
2839 /*
2840  * This is the function where ext4lazyinit thread lives. It walks
2841  * through the request list searching for next scheduled filesystem.
2842  * When such a fs is found, run the lazy initialization request
2843  * (ext4_rn_li_request) and keep track of the time spend in this
2844  * function. Based on that time we compute next schedule time of
2845  * the request. When walking through the list is complete, compute
2846  * next waking time and put itself into sleep.
2847  */
2848 static int ext4_lazyinit_thread(void *arg)
2849 {
2850         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2851         struct list_head *pos, *n;
2852         struct ext4_li_request *elr;
2853         unsigned long next_wakeup, cur;
2854
2855         BUG_ON(NULL == eli);
2856
2857 cont_thread:
2858         while (true) {
2859                 next_wakeup = MAX_JIFFY_OFFSET;
2860
2861                 mutex_lock(&eli->li_list_mtx);
2862                 if (list_empty(&eli->li_request_list)) {
2863                         mutex_unlock(&eli->li_list_mtx);
2864                         goto exit_thread;
2865                 }
2866
2867                 list_for_each_safe(pos, n, &eli->li_request_list) {
2868                         elr = list_entry(pos, struct ext4_li_request,
2869                                          lr_request);
2870
2871                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2872                                 if (ext4_run_li_request(elr) != 0) {
2873                                         /* error, remove the lazy_init job */
2874                                         ext4_remove_li_request(elr);
2875                                         continue;
2876                                 }
2877                         }
2878
2879                         if (time_before(elr->lr_next_sched, next_wakeup))
2880                                 next_wakeup = elr->lr_next_sched;
2881                 }
2882                 mutex_unlock(&eli->li_list_mtx);
2883
2884                 try_to_freeze();
2885
2886                 cur = jiffies;
2887                 if ((time_after_eq(cur, next_wakeup)) ||
2888                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2889                         cond_resched();
2890                         continue;
2891                 }
2892
2893                 schedule_timeout_interruptible(next_wakeup - cur);
2894
2895                 if (kthread_should_stop()) {
2896                         ext4_clear_request_list();
2897                         goto exit_thread;
2898                 }
2899         }
2900
2901 exit_thread:
2902         /*
2903          * It looks like the request list is empty, but we need
2904          * to check it under the li_list_mtx lock, to prevent any
2905          * additions into it, and of course we should lock ext4_li_mtx
2906          * to atomically free the list and ext4_li_info, because at
2907          * this point another ext4 filesystem could be registering
2908          * new one.
2909          */
2910         mutex_lock(&ext4_li_mtx);
2911         mutex_lock(&eli->li_list_mtx);
2912         if (!list_empty(&eli->li_request_list)) {
2913                 mutex_unlock(&eli->li_list_mtx);
2914                 mutex_unlock(&ext4_li_mtx);
2915                 goto cont_thread;
2916         }
2917         mutex_unlock(&eli->li_list_mtx);
2918         kfree(ext4_li_info);
2919         ext4_li_info = NULL;
2920         mutex_unlock(&ext4_li_mtx);
2921
2922         return 0;
2923 }
2924
2925 static void ext4_clear_request_list(void)
2926 {
2927         struct list_head *pos, *n;
2928         struct ext4_li_request *elr;
2929
2930         mutex_lock(&ext4_li_info->li_list_mtx);
2931         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2932                 elr = list_entry(pos, struct ext4_li_request,
2933                                  lr_request);
2934                 ext4_remove_li_request(elr);
2935         }
2936         mutex_unlock(&ext4_li_info->li_list_mtx);
2937 }
2938
2939 static int ext4_run_lazyinit_thread(void)
2940 {
2941         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2942                                          ext4_li_info, "ext4lazyinit");
2943         if (IS_ERR(ext4_lazyinit_task)) {
2944                 int err = PTR_ERR(ext4_lazyinit_task);
2945                 ext4_clear_request_list();
2946                 kfree(ext4_li_info);
2947                 ext4_li_info = NULL;
2948                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2949                                  "initialization thread\n",
2950                                  err);
2951                 return err;
2952         }
2953         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2954         return 0;
2955 }
2956
2957 /*
2958  * Check whether it make sense to run itable init. thread or not.
2959  * If there is at least one uninitialized inode table, return
2960  * corresponding group number, else the loop goes through all
2961  * groups and return total number of groups.
2962  */
2963 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2964 {
2965         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2966         struct ext4_group_desc *gdp = NULL;
2967
2968         for (group = 0; group < ngroups; group++) {
2969                 gdp = ext4_get_group_desc(sb, group, NULL);
2970                 if (!gdp)
2971                         continue;
2972
2973                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2974                         break;
2975         }
2976
2977         return group;
2978 }
2979
2980 static int ext4_li_info_new(void)
2981 {
2982         struct ext4_lazy_init *eli = NULL;
2983
2984         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2985         if (!eli)
2986                 return -ENOMEM;
2987
2988         INIT_LIST_HEAD(&eli->li_request_list);
2989         mutex_init(&eli->li_list_mtx);
2990
2991         eli->li_state |= EXT4_LAZYINIT_QUIT;
2992
2993         ext4_li_info = eli;
2994
2995         return 0;
2996 }
2997
2998 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2999                                             ext4_group_t start)
3000 {
3001         struct ext4_sb_info *sbi = EXT4_SB(sb);
3002         struct ext4_li_request *elr;
3003         unsigned long rnd;
3004
3005         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3006         if (!elr)
3007                 return NULL;
3008
3009         elr->lr_super = sb;
3010         elr->lr_sbi = sbi;
3011         elr->lr_next_group = start;
3012
3013         /*
3014          * Randomize first schedule time of the request to
3015          * spread the inode table initialization requests
3016          * better.
3017          */
3018         get_random_bytes(&rnd, sizeof(rnd));
3019         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3020                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3021
3022         return elr;
3023 }
3024
3025 static int ext4_register_li_request(struct super_block *sb,
3026                                     ext4_group_t first_not_zeroed)
3027 {
3028         struct ext4_sb_info *sbi = EXT4_SB(sb);
3029         struct ext4_li_request *elr;
3030         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3031         int ret = 0;
3032
3033         if (sbi->s_li_request != NULL) {
3034                 /*
3035                  * Reset timeout so it can be computed again, because
3036                  * s_li_wait_mult might have changed.
3037                  */
3038                 sbi->s_li_request->lr_timeout = 0;
3039                 return 0;
3040         }
3041
3042         if (first_not_zeroed == ngroups ||
3043             (sb->s_flags & MS_RDONLY) ||
3044             !test_opt(sb, INIT_INODE_TABLE))
3045                 return 0;
3046
3047         elr = ext4_li_request_new(sb, first_not_zeroed);
3048         if (!elr)
3049                 return -ENOMEM;
3050
3051         mutex_lock(&ext4_li_mtx);
3052
3053         if (NULL == ext4_li_info) {
3054                 ret = ext4_li_info_new();
3055                 if (ret)
3056                         goto out;
3057         }
3058
3059         mutex_lock(&ext4_li_info->li_list_mtx);
3060         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3061         mutex_unlock(&ext4_li_info->li_list_mtx);
3062
3063         sbi->s_li_request = elr;
3064         /*
3065          * set elr to NULL here since it has been inserted to
3066          * the request_list and the removal and free of it is
3067          * handled by ext4_clear_request_list from now on.
3068          */
3069         elr = NULL;
3070
3071         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3072                 ret = ext4_run_lazyinit_thread();
3073                 if (ret)
3074                         goto out;
3075         }
3076 out:
3077         mutex_unlock(&ext4_li_mtx);
3078         if (ret)
3079                 kfree(elr);
3080         return ret;
3081 }
3082
3083 /*
3084  * We do not need to lock anything since this is called on
3085  * module unload.
3086  */
3087 static void ext4_destroy_lazyinit_thread(void)
3088 {
3089         /*
3090          * If thread exited earlier
3091          * there's nothing to be done.
3092          */
3093         if (!ext4_li_info || !ext4_lazyinit_task)
3094                 return;
3095
3096         kthread_stop(ext4_lazyinit_task);
3097 }
3098
3099 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3100 {
3101         char *orig_data = kstrdup(data, GFP_KERNEL);
3102         struct buffer_head *bh;
3103         struct ext4_super_block *es = NULL;
3104         struct ext4_sb_info *sbi;
3105         ext4_fsblk_t block;
3106         ext4_fsblk_t sb_block = get_sb_block(&data);
3107         ext4_fsblk_t logical_sb_block;
3108         unsigned long offset = 0;
3109         unsigned long journal_devnum = 0;
3110         unsigned long def_mount_opts;
3111         struct inode *root;
3112         char *cp;
3113         const char *descr;
3114         int ret = -ENOMEM;
3115         int blocksize, clustersize;
3116         unsigned int db_count;
3117         unsigned int i;
3118         int needs_recovery, has_huge_files, has_bigalloc;
3119         __u64 blocks_count;
3120         int err;
3121         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3122         ext4_group_t first_not_zeroed;
3123
3124         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3125         if (!sbi)
3126                 goto out_free_orig;
3127
3128         sbi->s_blockgroup_lock =
3129                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3130         if (!sbi->s_blockgroup_lock) {
3131                 kfree(sbi);
3132                 goto out_free_orig;
3133         }
3134         sb->s_fs_info = sbi;
3135         sbi->s_mount_opt = 0;
3136         sbi->s_resuid = EXT4_DEF_RESUID;
3137         sbi->s_resgid = EXT4_DEF_RESGID;
3138         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3139         sbi->s_sb_block = sb_block;
3140         if (sb->s_bdev->bd_part)
3141                 sbi->s_sectors_written_start =
3142                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3143
3144         /* Cleanup superblock name */
3145         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3146                 *cp = '!';
3147
3148         ret = -EINVAL;
3149         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3150         if (!blocksize) {
3151                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3152                 goto out_fail;
3153         }
3154
3155         /*
3156          * The ext4 superblock will not be buffer aligned for other than 1kB
3157          * block sizes.  We need to calculate the offset from buffer start.
3158          */
3159         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3160                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3161                 offset = do_div(logical_sb_block, blocksize);
3162         } else {
3163                 logical_sb_block = sb_block;
3164         }
3165
3166         if (!(bh = sb_bread(sb, logical_sb_block))) {
3167                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3168                 goto out_fail;
3169         }
3170         /*
3171          * Note: s_es must be initialized as soon as possible because
3172          *       some ext4 macro-instructions depend on its value
3173          */
3174         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3175         sbi->s_es = es;
3176         sb->s_magic = le16_to_cpu(es->s_magic);
3177         if (sb->s_magic != EXT4_SUPER_MAGIC)
3178                 goto cantfind_ext4;
3179         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3180
3181         /* Set defaults before we parse the mount options */
3182         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3183         set_opt(sb, INIT_INODE_TABLE);
3184         if (def_mount_opts & EXT4_DEFM_DEBUG)
3185                 set_opt(sb, DEBUG);
3186         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3187                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3188                         "2.6.38");
3189                 set_opt(sb, GRPID);
3190         }
3191         if (def_mount_opts & EXT4_DEFM_UID16)
3192                 set_opt(sb, NO_UID32);
3193         /* xattr user namespace & acls are now defaulted on */
3194 #ifdef CONFIG_EXT4_FS_XATTR
3195         set_opt(sb, XATTR_USER);
3196 #endif
3197 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3198         set_opt(sb, POSIX_ACL);
3199 #endif
3200         set_opt(sb, MBLK_IO_SUBMIT);
3201         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3202                 set_opt(sb, JOURNAL_DATA);
3203         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3204                 set_opt(sb, ORDERED_DATA);
3205         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3206                 set_opt(sb, WRITEBACK_DATA);
3207
3208         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3209                 set_opt(sb, ERRORS_PANIC);
3210         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3211                 set_opt(sb, ERRORS_CONT);
3212         else
3213                 set_opt(sb, ERRORS_RO);
3214         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3215                 set_opt(sb, BLOCK_VALIDITY);
3216         if (def_mount_opts & EXT4_DEFM_DISCARD)
3217                 set_opt(sb, DISCARD);
3218
3219         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3220         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3221         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3222         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3223         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3224
3225         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3226                 set_opt(sb, BARRIER);
3227
3228         /*
3229          * enable delayed allocation by default
3230          * Use -o nodelalloc to turn it off
3231          */
3232         if (!IS_EXT3_SB(sb) &&
3233             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3234                 set_opt(sb, DELALLOC);
3235
3236         /*
3237          * set default s_li_wait_mult for lazyinit, for the case there is
3238          * no mount option specified.
3239          */
3240         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3241
3242         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3243                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3244                 ext4_msg(sb, KERN_WARNING,
3245                          "failed to parse options in superblock: %s",
3246                          sbi->s_es->s_mount_opts);
3247         }
3248         if (!parse_options((char *) data, sb, &journal_devnum,
3249                            &journal_ioprio, NULL, 0))
3250                 goto failed_mount;
3251
3252         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3253                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3254                             "with data=journal disables delayed "
3255                             "allocation and O_DIRECT support!\n");
3256                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3257                         ext4_msg(sb, KERN_ERR, "can't mount with "
3258                                  "both data=journal and delalloc");
3259                         goto failed_mount;
3260                 }
3261                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3262                         ext4_msg(sb, KERN_ERR, "can't mount with "
3263                                  "both data=journal and delalloc");
3264                         goto failed_mount;
3265                 }
3266                 if (test_opt(sb, DELALLOC))
3267                         clear_opt(sb, DELALLOC);
3268         }
3269
3270         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3271         if (test_opt(sb, DIOREAD_NOLOCK)) {
3272                 if (blocksize < PAGE_SIZE) {
3273                         ext4_msg(sb, KERN_ERR, "can't mount with "
3274                                  "dioread_nolock if block size != PAGE_SIZE");
3275                         goto failed_mount;
3276                 }
3277         }
3278
3279         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3280                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3281
3282         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3283             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3284              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3285              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3286                 ext4_msg(sb, KERN_WARNING,
3287                        "feature flags set on rev 0 fs, "
3288                        "running e2fsck is recommended");
3289
3290         if (IS_EXT2_SB(sb)) {
3291                 if (ext2_feature_set_ok(sb))
3292                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3293                                  "using the ext4 subsystem");
3294                 else {
3295                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3296                                  "to feature incompatibilities");
3297                         goto failed_mount;
3298                 }
3299         }
3300
3301         if (IS_EXT3_SB(sb)) {
3302                 if (ext3_feature_set_ok(sb))
3303                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3304                                  "using the ext4 subsystem");
3305                 else {
3306                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3307                                  "to feature incompatibilities");
3308                         goto failed_mount;
3309                 }
3310         }
3311
3312         /*
3313          * Check feature flags regardless of the revision level, since we
3314          * previously didn't change the revision level when setting the flags,
3315          * so there is a chance incompat flags are set on a rev 0 filesystem.
3316          */
3317         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3318                 goto failed_mount;
3319
3320         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3321             blocksize > EXT4_MAX_BLOCK_SIZE) {
3322                 ext4_msg(sb, KERN_ERR,
3323                        "Unsupported filesystem blocksize %d", blocksize);
3324                 goto failed_mount;
3325         }
3326
3327         if (sb->s_blocksize != blocksize) {
3328                 /* Validate the filesystem blocksize */
3329                 if (!sb_set_blocksize(sb, blocksize)) {
3330                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3331                                         blocksize);
3332                         goto failed_mount;
3333                 }
3334
3335                 brelse(bh);
3336                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3337                 offset = do_div(logical_sb_block, blocksize);
3338                 bh = sb_bread(sb, logical_sb_block);
3339                 if (!bh) {
3340                         ext4_msg(sb, KERN_ERR,
3341                                "Can't read superblock on 2nd try");
3342                         goto failed_mount;
3343                 }
3344                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3345                 sbi->s_es = es;
3346                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3347                         ext4_msg(sb, KERN_ERR,
3348                                "Magic mismatch, very weird!");
3349                         goto failed_mount;
3350                 }
3351         }
3352
3353         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3354                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3355         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3356                                                       has_huge_files);
3357         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3358
3359         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3360                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3361                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3362         } else {
3363                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3364                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3365                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3366                     (!is_power_of_2(sbi->s_inode_size)) ||
3367                     (sbi->s_inode_size > blocksize)) {
3368                         ext4_msg(sb, KERN_ERR,
3369                                "unsupported inode size: %d",
3370                                sbi->s_inode_size);
3371                         goto failed_mount;
3372                 }
3373                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3374                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3375         }
3376
3377         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3378         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3379                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3380                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3381                     !is_power_of_2(sbi->s_desc_size)) {
3382                         ext4_msg(sb, KERN_ERR,
3383                                "unsupported descriptor size %lu",
3384                                sbi->s_desc_size);
3385                         goto failed_mount;
3386                 }
3387         } else
3388                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3389
3390         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3391         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3392         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3393                 goto cantfind_ext4;
3394
3395         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3396         if (sbi->s_inodes_per_block == 0)
3397                 goto cantfind_ext4;
3398         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3399                                         sbi->s_inodes_per_block;
3400         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3401         sbi->s_sbh = bh;
3402         sbi->s_mount_state = le16_to_cpu(es->s_state);
3403         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3404         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3405
3406         for (i = 0; i < 4; i++)
3407                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3408         sbi->s_def_hash_version = es->s_def_hash_version;
3409         i = le32_to_cpu(es->s_flags);
3410         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3411                 sbi->s_hash_unsigned = 3;
3412         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3413 #ifdef __CHAR_UNSIGNED__
3414                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3415                 sbi->s_hash_unsigned = 3;
3416 #else
3417                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3418 #endif
3419                 sb->s_dirt = 1;
3420         }
3421
3422         /* Handle clustersize */
3423         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3424         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3425                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3426         if (has_bigalloc) {
3427                 if (clustersize < blocksize) {
3428                         ext4_msg(sb, KERN_ERR,
3429                                  "cluster size (%d) smaller than "
3430                                  "block size (%d)", clustersize, blocksize);
3431                         goto failed_mount;
3432                 }
3433                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3434                         le32_to_cpu(es->s_log_block_size);
3435                 sbi->s_clusters_per_group =
3436                         le32_to_cpu(es->s_clusters_per_group);
3437                 if (sbi->s_clusters_per_group > blocksize * 8) {
3438                         ext4_msg(sb, KERN_ERR,
3439                                  "#clusters per group too big: %lu",
3440                                  sbi->s_clusters_per_group);
3441                         goto failed_mount;
3442                 }
3443                 if (sbi->s_blocks_per_group !=
3444                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3445                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3446                                  "clusters per group (%lu) inconsistent",
3447                                  sbi->s_blocks_per_group,
3448                                  sbi->s_clusters_per_group);
3449                         goto failed_mount;
3450                 }
3451         } else {
3452                 if (clustersize != blocksize) {
3453                         ext4_warning(sb, "fragment/cluster size (%d) != "
3454                                      "block size (%d)", clustersize,
3455                                      blocksize);
3456                         clustersize = blocksize;
3457                 }
3458                 if (sbi->s_blocks_per_group > blocksize * 8) {
3459                         ext4_msg(sb, KERN_ERR,
3460                                  "#blocks per group too big: %lu",
3461                                  sbi->s_blocks_per_group);
3462                         goto failed_mount;
3463                 }
3464                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3465                 sbi->s_cluster_bits = 0;
3466         }
3467         sbi->s_cluster_ratio = clustersize / blocksize;
3468
3469         if (sbi->s_inodes_per_group > blocksize * 8) {
3470                 ext4_msg(sb, KERN_ERR,
3471                        "#inodes per group too big: %lu",
3472                        sbi->s_inodes_per_group);
3473                 goto failed_mount;
3474         }
3475
3476         /*
3477          * Test whether we have more sectors than will fit in sector_t,
3478          * and whether the max offset is addressable by the page cache.
3479          */
3480         err = generic_check_addressable(sb->s_blocksize_bits,
3481                                         ext4_blocks_count(es));
3482         if (err) {
3483                 ext4_msg(sb, KERN_ERR, "filesystem"
3484                          " too large to mount safely on this system");
3485                 if (sizeof(sector_t) < 8)
3486                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3487                 ret = err;
3488                 goto failed_mount;
3489         }
3490
3491         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3492                 goto cantfind_ext4;
3493
3494         /* check blocks count against device size */
3495         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3496         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3497                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3498                        "exceeds size of device (%llu blocks)",
3499                        ext4_blocks_count(es), blocks_count);
3500                 goto failed_mount;
3501         }
3502
3503         /*
3504          * It makes no sense for the first data block to be beyond the end
3505          * of the filesystem.
3506          */
3507         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3508                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3509                          "block %u is beyond end of filesystem (%llu)",
3510                          le32_to_cpu(es->s_first_data_block),
3511                          ext4_blocks_count(es));
3512                 goto failed_mount;
3513         }
3514         blocks_count = (ext4_blocks_count(es) -
3515                         le32_to_cpu(es->s_first_data_block) +
3516                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3517         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3518         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3519                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3520                        "(block count %llu, first data block %u, "
3521                        "blocks per group %lu)", sbi->s_groups_count,
3522                        ext4_blocks_count(es),
3523                        le32_to_cpu(es->s_first_data_block),
3524                        EXT4_BLOCKS_PER_GROUP(sb));
3525                 goto failed_mount;
3526         }
3527         sbi->s_groups_count = blocks_count;
3528         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3529                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3530         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3531                    EXT4_DESC_PER_BLOCK(sb);
3532         sbi->s_group_desc = ext4_kvmalloc(db_count *
3533                                           sizeof(struct buffer_head *),
3534                                           GFP_KERNEL);
3535         if (sbi->s_group_desc == NULL) {
3536                 ext4_msg(sb, KERN_ERR, "not enough memory");
3537                 goto failed_mount;
3538         }
3539
3540         if (ext4_proc_root)
3541                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3542
3543         bgl_lock_init(sbi->s_blockgroup_lock);
3544
3545         for (i = 0; i < db_count; i++) {
3546                 block = descriptor_loc(sb, logical_sb_block, i);
3547                 sbi->s_group_desc[i] = sb_bread(sb, block);
3548                 if (!sbi->s_group_desc[i]) {
3549                         ext4_msg(sb, KERN_ERR,
3550                                "can't read group descriptor %d", i);
3551                         db_count = i;
3552                         goto failed_mount2;
3553                 }
3554         }
3555         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3556                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3557                 goto failed_mount2;
3558         }
3559         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3560                 if (!ext4_fill_flex_info(sb)) {
3561                         ext4_msg(sb, KERN_ERR,
3562                                "unable to initialize "
3563                                "flex_bg meta info!");
3564                         goto failed_mount2;
3565                 }
3566
3567         sbi->s_gdb_count = db_count;
3568         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3569         spin_lock_init(&sbi->s_next_gen_lock);
3570
3571         init_timer(&sbi->s_err_report);
3572         sbi->s_err_report.function = print_daily_error_info;
3573         sbi->s_err_report.data = (unsigned long) sb;
3574
3575         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3576                         ext4_count_free_clusters(sb));
3577         if (!err) {
3578                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3579                                 ext4_count_free_inodes(sb));
3580         }
3581         if (!err) {
3582                 err = percpu_counter_init(&sbi->s_dirs_counter,
3583                                 ext4_count_dirs(sb));
3584         }
3585         if (!err) {
3586                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3587         }
3588         if (err) {
3589                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3590                 goto failed_mount3;
3591         }
3592
3593         sbi->s_stripe = ext4_get_stripe_size(sbi);
3594         sbi->s_max_writeback_mb_bump = 128;
3595
3596         /*
3597          * set up enough so that it can read an inode
3598          */
3599         if (!test_opt(sb, NOLOAD) &&
3600             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3601                 sb->s_op = &ext4_sops;
3602         else
3603                 sb->s_op = &ext4_nojournal_sops;
3604         sb->s_export_op = &ext4_export_ops;
3605         sb->s_xattr = ext4_xattr_handlers;
3606 #ifdef CONFIG_QUOTA
3607         sb->s_qcop = &ext4_qctl_operations;
3608         sb->dq_op = &ext4_quota_operations;
3609 #endif
3610         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3611
3612         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3613         mutex_init(&sbi->s_orphan_lock);
3614         sbi->s_resize_flags = 0;
3615
3616         sb->s_root = NULL;
3617
3618         needs_recovery = (es->s_last_orphan != 0 ||
3619                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3620                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3621
3622         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3623             !(sb->s_flags & MS_RDONLY))
3624                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3625                         goto failed_mount3;
3626
3627         /*
3628          * The first inode we look at is the journal inode.  Don't try
3629          * root first: it may be modified in the journal!
3630          */
3631         if (!test_opt(sb, NOLOAD) &&
3632             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3633                 if (ext4_load_journal(sb, es, journal_devnum))
3634                         goto failed_mount3;
3635         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3636               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3637                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3638                        "suppressed and not mounted read-only");
3639                 goto failed_mount_wq;
3640         } else {
3641                 clear_opt(sb, DATA_FLAGS);
3642                 sbi->s_journal = NULL;
3643                 needs_recovery = 0;
3644                 goto no_journal;
3645         }
3646
3647         if (ext4_blocks_count(es) > 0xffffffffULL &&
3648             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3649                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3650                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3651                 goto failed_mount_wq;
3652         }
3653
3654         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3655                 jbd2_journal_set_features(sbi->s_journal,
3656                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3657                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3658         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3659                 jbd2_journal_set_features(sbi->s_journal,
3660                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3661                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3662                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3663         } else {
3664                 jbd2_journal_clear_features(sbi->s_journal,
3665                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3666                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3667         }
3668
3669         /* We have now updated the journal if required, so we can
3670          * validate the data journaling mode. */
3671         switch (test_opt(sb, DATA_FLAGS)) {
3672         case 0:
3673                 /* No mode set, assume a default based on the journal
3674                  * capabilities: ORDERED_DATA if the journal can
3675                  * cope, else JOURNAL_DATA
3676                  */
3677                 if (jbd2_journal_check_available_features
3678                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3679                         set_opt(sb, ORDERED_DATA);
3680                 else
3681                         set_opt(sb, JOURNAL_DATA);
3682                 break;
3683
3684         case EXT4_MOUNT_ORDERED_DATA:
3685         case EXT4_MOUNT_WRITEBACK_DATA:
3686                 if (!jbd2_journal_check_available_features
3687                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3688                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3689                                "requested data journaling mode");
3690                         goto failed_mount_wq;
3691                 }
3692         default:
3693                 break;
3694         }
3695         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3696
3697         /*
3698          * The journal may have updated the bg summary counts, so we
3699          * need to update the global counters.
3700          */
3701         percpu_counter_set(&sbi->s_freeclusters_counter,
3702                            ext4_count_free_clusters(sb));
3703         percpu_counter_set(&sbi->s_freeinodes_counter,
3704                            ext4_count_free_inodes(sb));
3705         percpu_counter_set(&sbi->s_dirs_counter,
3706                            ext4_count_dirs(sb));
3707         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3708
3709 no_journal:
3710         /*
3711          * The maximum number of concurrent works can be high and
3712          * concurrency isn't really necessary.  Limit it to 1.
3713          */
3714         EXT4_SB(sb)->dio_unwritten_wq =
3715                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3716         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3717                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3718                 goto failed_mount_wq;
3719         }
3720
3721         /*
3722          * The jbd2_journal_load will have done any necessary log recovery,
3723          * so we can safely mount the rest of the filesystem now.
3724          */
3725
3726         root = ext4_iget(sb, EXT4_ROOT_INO);
3727         if (IS_ERR(root)) {
3728                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3729                 ret = PTR_ERR(root);
3730                 root = NULL;
3731                 goto failed_mount4;
3732         }
3733         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3734                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3735                 iput(root);
3736                 goto failed_mount4;
3737         }
3738         sb->s_root = d_alloc_root(root);
3739         if (!sb->s_root) {
3740                 iput(root);
3741                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3742                 ret = -ENOMEM;
3743                 goto failed_mount4;
3744         }
3745
3746         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3747
3748         /* determine the minimum size of new large inodes, if present */
3749         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3750                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3751                                                      EXT4_GOOD_OLD_INODE_SIZE;
3752                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3753                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3754                         if (sbi->s_want_extra_isize <
3755                             le16_to_cpu(es->s_want_extra_isize))
3756                                 sbi->s_want_extra_isize =
3757                                         le16_to_cpu(es->s_want_extra_isize);
3758                         if (sbi->s_want_extra_isize <
3759                             le16_to_cpu(es->s_min_extra_isize))
3760                                 sbi->s_want_extra_isize =
3761                                         le16_to_cpu(es->s_min_extra_isize);
3762                 }
3763         }
3764         /* Check if enough inode space is available */
3765         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3766                                                         sbi->s_inode_size) {
3767                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3768                                                        EXT4_GOOD_OLD_INODE_SIZE;
3769                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3770                          "available");
3771         }
3772
3773         err = ext4_setup_system_zone(sb);
3774         if (err) {
3775                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3776                          "zone (%d)", err);
3777                 goto failed_mount4a;
3778         }
3779
3780         ext4_ext_init(sb);
3781         err = ext4_mb_init(sb, needs_recovery);
3782         if (err) {
3783                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3784                          err);
3785                 goto failed_mount5;
3786         }
3787
3788         err = ext4_register_li_request(sb, first_not_zeroed);
3789         if (err)
3790                 goto failed_mount6;
3791
3792         sbi->s_kobj.kset = ext4_kset;
3793         init_completion(&sbi->s_kobj_unregister);
3794         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3795                                    "%s", sb->s_id);
3796         if (err)
3797                 goto failed_mount7;
3798
3799         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3800         ext4_orphan_cleanup(sb, es);
3801         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3802         if (needs_recovery) {
3803                 ext4_msg(sb, KERN_INFO, "recovery complete");
3804                 ext4_mark_recovery_complete(sb, es);
3805         }
3806         if (EXT4_SB(sb)->s_journal) {
3807                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3808                         descr = " journalled data mode";
3809                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3810                         descr = " ordered data mode";
3811                 else
3812                         descr = " writeback data mode";
3813         } else
3814                 descr = "out journal";
3815
3816         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3817                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3818                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3819
3820         if (es->s_error_count)
3821                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3822
3823         kfree(orig_data);
3824         return 0;
3825
3826 cantfind_ext4:
3827         if (!silent)
3828                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3829         goto failed_mount;
3830
3831 failed_mount7:
3832         ext4_unregister_li_request(sb);
3833 failed_mount6:
3834         ext4_mb_release(sb);
3835 failed_mount5:
3836         ext4_ext_release(sb);
3837         ext4_release_system_zone(sb);
3838 failed_mount4a:
3839         dput(sb->s_root);
3840         sb->s_root = NULL;
3841 failed_mount4:
3842         ext4_msg(sb, KERN_ERR, "mount failed");
3843         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3844 failed_mount_wq:
3845         if (sbi->s_journal) {
3846                 jbd2_journal_destroy(sbi->s_journal);
3847                 sbi->s_journal = NULL;
3848         }
3849 failed_mount3:
3850         del_timer(&sbi->s_err_report);
3851         if (sbi->s_flex_groups)
3852                 ext4_kvfree(sbi->s_flex_groups);
3853         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3854         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3855         percpu_counter_destroy(&sbi->s_dirs_counter);
3856         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3857         if (sbi->s_mmp_tsk)
3858                 kthread_stop(sbi->s_mmp_tsk);
3859 failed_mount2:
3860         for (i = 0; i < db_count; i++)
3861                 brelse(sbi->s_group_desc[i]);
3862         ext4_kvfree(sbi->s_group_desc);
3863 failed_mount:
3864         if (sbi->s_proc) {
3865                 remove_proc_entry(sb->s_id, ext4_proc_root);
3866         }
3867 #ifdef CONFIG_QUOTA
3868         for (i = 0; i < MAXQUOTAS; i++)
3869                 kfree(sbi->s_qf_names[i]);
3870 #endif
3871         ext4_blkdev_remove(sbi);
3872         brelse(bh);
3873 out_fail:
3874         sb->s_fs_info = NULL;
3875         kfree(sbi->s_blockgroup_lock);
3876         kfree(sbi);
3877 out_free_orig:
3878         kfree(orig_data);
3879         return ret;
3880 }
3881
3882 /*
3883  * Setup any per-fs journal parameters now.  We'll do this both on
3884  * initial mount, once the journal has been initialised but before we've
3885  * done any recovery; and again on any subsequent remount.
3886  */
3887 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3888 {
3889         struct ext4_sb_info *sbi = EXT4_SB(sb);
3890
3891         journal->j_commit_interval = sbi->s_commit_interval;
3892         journal->j_min_batch_time = sbi->s_min_batch_time;
3893         journal->j_max_batch_time = sbi->s_max_batch_time;
3894
3895         write_lock(&journal->j_state_lock);
3896         if (test_opt(sb, BARRIER))
3897                 journal->j_flags |= JBD2_BARRIER;
3898         else
3899                 journal->j_flags &= ~JBD2_BARRIER;
3900         if (test_opt(sb, DATA_ERR_ABORT))
3901                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3902         else
3903                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3904         write_unlock(&journal->j_state_lock);
3905 }
3906
3907 static journal_t *ext4_get_journal(struct super_block *sb,
3908                                    unsigned int journal_inum)
3909 {
3910         struct inode *journal_inode;
3911         journal_t *journal;
3912
3913         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3914
3915         /* First, test for the existence of a valid inode on disk.  Bad
3916          * things happen if we iget() an unused inode, as the subsequent
3917          * iput() will try to delete it. */
3918
3919         journal_inode = ext4_iget(sb, journal_inum);
3920         if (IS_ERR(journal_inode)) {
3921                 ext4_msg(sb, KERN_ERR, "no journal found");
3922                 return NULL;
3923         }
3924         if (!journal_inode->i_nlink) {
3925                 make_bad_inode(journal_inode);
3926                 iput(journal_inode);
3927                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3928                 return NULL;
3929         }
3930
3931         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3932                   journal_inode, journal_inode->i_size);
3933         if (!S_ISREG(journal_inode->i_mode)) {
3934                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3935                 iput(journal_inode);
3936                 return NULL;
3937         }
3938
3939         journal = jbd2_journal_init_inode(journal_inode);
3940         if (!journal) {
3941                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3942                 iput(journal_inode);
3943                 return NULL;
3944         }
3945         journal->j_private = sb;
3946         ext4_init_journal_params(sb, journal);
3947         return journal;
3948 }
3949
3950 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3951                                        dev_t j_dev)
3952 {
3953         struct buffer_head *bh;
3954         journal_t *journal;
3955         ext4_fsblk_t start;
3956         ext4_fsblk_t len;
3957         int hblock, blocksize;
3958         ext4_fsblk_t sb_block;
3959         unsigned long offset;
3960         struct ext4_super_block *es;
3961         struct block_device *bdev;
3962
3963         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3964
3965         bdev = ext4_blkdev_get(j_dev, sb);
3966         if (bdev == NULL)
3967                 return NULL;
3968
3969         blocksize = sb->s_blocksize;
3970         hblock = bdev_logical_block_size(bdev);
3971         if (blocksize < hblock) {
3972                 ext4_msg(sb, KERN_ERR,
3973                         "blocksize too small for journal device");
3974                 goto out_bdev;
3975         }
3976
3977         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3978         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3979         set_blocksize(bdev, blocksize);
3980         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3981                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3982                        "external journal");
3983                 goto out_bdev;
3984         }
3985
3986         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3987         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3988             !(le32_to_cpu(es->s_feature_incompat) &
3989               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3990                 ext4_msg(sb, KERN_ERR, "external journal has "
3991                                         "bad superblock");
3992                 brelse(bh);
3993                 goto out_bdev;
3994         }
3995
3996         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3997                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3998                 brelse(bh);
3999                 goto out_bdev;
4000         }
4001
4002         len = ext4_blocks_count(es);
4003         start = sb_block + 1;
4004         brelse(bh);     /* we're done with the superblock */
4005
4006         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4007                                         start, len, blocksize);
4008         if (!journal) {
4009                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4010                 goto out_bdev;
4011         }
4012         journal->j_private = sb;
4013         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4014         wait_on_buffer(journal->j_sb_buffer);
4015         if (!buffer_uptodate(journal->j_sb_buffer)) {
4016                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4017                 goto out_journal;
4018         }
4019         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4020                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4021                                         "user (unsupported) - %d",
4022                         be32_to_cpu(journal->j_superblock->s_nr_users));
4023                 goto out_journal;
4024         }
4025         EXT4_SB(sb)->journal_bdev = bdev;
4026         ext4_init_journal_params(sb, journal);
4027         return journal;
4028
4029 out_journal:
4030         jbd2_journal_destroy(journal);
4031 out_bdev:
4032         ext4_blkdev_put(bdev);
4033         return NULL;
4034 }
4035
4036 static int ext4_load_journal(struct super_block *sb,
4037                              struct ext4_super_block *es,
4038                              unsigned long journal_devnum)
4039 {
4040         journal_t *journal;
4041         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4042         dev_t journal_dev;
4043         int err = 0;
4044         int really_read_only;
4045
4046         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4047
4048         if (journal_devnum &&
4049             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4050                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4051                         "numbers have changed");
4052                 journal_dev = new_decode_dev(journal_devnum);
4053         } else
4054                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4055
4056         really_read_only = bdev_read_only(sb->s_bdev);
4057
4058         /*
4059          * Are we loading a blank journal or performing recovery after a
4060          * crash?  For recovery, we need to check in advance whether we
4061          * can get read-write access to the device.
4062          */
4063         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4064                 if (sb->s_flags & MS_RDONLY) {
4065                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4066                                         "required on readonly filesystem");
4067                         if (really_read_only) {
4068                                 ext4_msg(sb, KERN_ERR, "write access "
4069                                         "unavailable, cannot proceed");
4070                                 return -EROFS;
4071                         }
4072                         ext4_msg(sb, KERN_INFO, "write access will "
4073                                "be enabled during recovery");
4074                 }
4075         }
4076
4077         if (journal_inum && journal_dev) {
4078                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4079                        "and inode journals!");
4080                 return -EINVAL;
4081         }
4082
4083         if (journal_inum) {
4084                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4085                         return -EINVAL;
4086         } else {
4087                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4088                         return -EINVAL;
4089         }
4090
4091         if (!(journal->j_flags & JBD2_BARRIER))
4092                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4093
4094         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
4095                 err = jbd2_journal_update_format(journal);
4096                 if (err)  {
4097                         ext4_msg(sb, KERN_ERR, "error updating journal");
4098                         jbd2_journal_destroy(journal);
4099                         return err;
4100                 }
4101         }
4102
4103         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4104                 err = jbd2_journal_wipe(journal, !really_read_only);
4105         if (!err) {
4106                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4107                 if (save)
4108                         memcpy(save, ((char *) es) +
4109                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4110                 err = jbd2_journal_load(journal);
4111                 if (save)
4112                         memcpy(((char *) es) + EXT4_S_ERR_START,
4113                                save, EXT4_S_ERR_LEN);
4114                 kfree(save);
4115         }
4116
4117         if (err) {
4118                 ext4_msg(sb, KERN_ERR, "error loading journal");
4119                 jbd2_journal_destroy(journal);
4120                 return err;
4121         }
4122
4123         EXT4_SB(sb)->s_journal = journal;
4124         ext4_clear_journal_err(sb, es);
4125
4126         if (!really_read_only && journal_devnum &&
4127             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4128                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4129
4130                 /* Make sure we flush the recovery flag to disk. */
4131                 ext4_commit_super(sb, 1);
4132         }
4133
4134         return 0;
4135 }
4136
4137 static int ext4_commit_super(struct super_block *sb, int sync)
4138 {
4139         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4140         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4141         int error = 0;
4142
4143         if (!sbh || block_device_ejected(sb))
4144                 return error;
4145         if (buffer_write_io_error(sbh)) {
4146                 /*
4147                  * Oh, dear.  A previous attempt to write the
4148                  * superblock failed.  This could happen because the
4149                  * USB device was yanked out.  Or it could happen to
4150                  * be a transient write error and maybe the block will
4151                  * be remapped.  Nothing we can do but to retry the
4152                  * write and hope for the best.
4153                  */
4154                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4155                        "superblock detected");
4156                 clear_buffer_write_io_error(sbh);
4157                 set_buffer_uptodate(sbh);
4158         }
4159         /*
4160          * If the file system is mounted read-only, don't update the
4161          * superblock write time.  This avoids updating the superblock
4162          * write time when we are mounting the root file system
4163          * read/only but we need to replay the journal; at that point,
4164          * for people who are east of GMT and who make their clock
4165          * tick in localtime for Windows bug-for-bug compatibility,
4166          * the clock is set in the future, and this will cause e2fsck
4167          * to complain and force a full file system check.
4168          */
4169         if (!(sb->s_flags & MS_RDONLY))
4170                 es->s_wtime = cpu_to_le32(get_seconds());
4171         if (sb->s_bdev->bd_part)
4172                 es->s_kbytes_written =
4173                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4174                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4175                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4176         else
4177                 es->s_kbytes_written =
4178                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4179         ext4_free_blocks_count_set(es,
4180                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4181                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4182         es->s_free_inodes_count =
4183                 cpu_to_le32(percpu_counter_sum_positive(
4184                                 &EXT4_SB(sb)->s_freeinodes_counter));
4185         sb->s_dirt = 0;
4186         BUFFER_TRACE(sbh, "marking dirty");
4187         mark_buffer_dirty(sbh);
4188         if (sync) {
4189                 error = sync_dirty_buffer(sbh);
4190                 if (error)
4191                         return error;
4192
4193                 error = buffer_write_io_error(sbh);
4194                 if (error) {
4195                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4196                                "superblock");
4197                         clear_buffer_write_io_error(sbh);
4198                         set_buffer_uptodate(sbh);
4199                 }
4200         }
4201         return error;
4202 }
4203
4204 /*
4205  * Have we just finished recovery?  If so, and if we are mounting (or
4206  * remounting) the filesystem readonly, then we will end up with a
4207  * consistent fs on disk.  Record that fact.
4208  */
4209 static void ext4_mark_recovery_complete(struct super_block *sb,
4210                                         struct ext4_super_block *es)
4211 {
4212         journal_t *journal = EXT4_SB(sb)->s_journal;
4213
4214         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4215                 BUG_ON(journal != NULL);
4216                 return;
4217         }
4218         jbd2_journal_lock_updates(journal);
4219         if (jbd2_journal_flush(journal) < 0)
4220                 goto out;
4221
4222         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4223             sb->s_flags & MS_RDONLY) {
4224                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4225                 ext4_commit_super(sb, 1);
4226         }
4227
4228 out:
4229         jbd2_journal_unlock_updates(journal);
4230 }
4231
4232 /*
4233  * If we are mounting (or read-write remounting) a filesystem whose journal
4234  * has recorded an error from a previous lifetime, move that error to the
4235  * main filesystem now.
4236  */
4237 static void ext4_clear_journal_err(struct super_block *sb,
4238                                    struct ext4_super_block *es)
4239 {
4240         journal_t *journal;
4241         int j_errno;
4242         const char *errstr;
4243
4244         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4245
4246         journal = EXT4_SB(sb)->s_journal;
4247
4248         /*
4249          * Now check for any error status which may have been recorded in the
4250          * journal by a prior ext4_error() or ext4_abort()
4251          */
4252
4253         j_errno = jbd2_journal_errno(journal);
4254         if (j_errno) {
4255                 char nbuf[16];
4256
4257                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4258                 ext4_warning(sb, "Filesystem error recorded "
4259                              "from previous mount: %s", errstr);
4260                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4261
4262                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4263                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4264                 ext4_commit_super(sb, 1);
4265
4266                 jbd2_journal_clear_err(journal);
4267         }
4268 }
4269
4270 /*
4271  * Force the running and committing transactions to commit,
4272  * and wait on the commit.
4273  */
4274 int ext4_force_commit(struct super_block *sb)
4275 {
4276         journal_t *journal;
4277         int ret = 0;
4278
4279         if (sb->s_flags & MS_RDONLY)
4280                 return 0;
4281
4282         journal = EXT4_SB(sb)->s_journal;
4283         if (journal) {
4284                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4285                 ret = ext4_journal_force_commit(journal);
4286         }
4287
4288         return ret;
4289 }
4290
4291 static void ext4_write_super(struct super_block *sb)
4292 {
4293         lock_super(sb);
4294         ext4_commit_super(sb, 1);
4295         unlock_super(sb);
4296 }
4297
4298 static int ext4_sync_fs(struct super_block *sb, int wait)
4299 {
4300         int ret = 0;
4301         tid_t target;
4302         struct ext4_sb_info *sbi = EXT4_SB(sb);
4303
4304         trace_ext4_sync_fs(sb, wait);
4305         flush_workqueue(sbi->dio_unwritten_wq);
4306         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4307                 if (wait)
4308                         jbd2_log_wait_commit(sbi->s_journal, target);
4309         }
4310         return ret;
4311 }
4312
4313 /*
4314  * LVM calls this function before a (read-only) snapshot is created.  This
4315  * gives us a chance to flush the journal completely and mark the fs clean.
4316  *
4317  * Note that only this function cannot bring a filesystem to be in a clean
4318  * state independently, because ext4 prevents a new handle from being started
4319  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4320  * the upper layer.
4321  */
4322 static int ext4_freeze(struct super_block *sb)
4323 {
4324         int error = 0;
4325         journal_t *journal;
4326
4327         if (sb->s_flags & MS_RDONLY)
4328                 return 0;
4329
4330         journal = EXT4_SB(sb)->s_journal;
4331
4332         /* Now we set up the journal barrier. */
4333         jbd2_journal_lock_updates(journal);
4334
4335         /*
4336          * Don't clear the needs_recovery flag if we failed to flush
4337          * the journal.
4338          */
4339         error = jbd2_journal_flush(journal);
4340         if (error < 0)
4341                 goto out;
4342
4343         /* Journal blocked and flushed, clear needs_recovery flag. */
4344         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4345         error = ext4_commit_super(sb, 1);
4346 out:
4347         /* we rely on s_frozen to stop further updates */
4348         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4349         return error;
4350 }
4351
4352 /*
4353  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4354  * flag here, even though the filesystem is not technically dirty yet.
4355  */
4356 static int ext4_unfreeze(struct super_block *sb)
4357 {
4358         if (sb->s_flags & MS_RDONLY)
4359                 return 0;
4360
4361         lock_super(sb);
4362         /* Reset the needs_recovery flag before the fs is unlocked. */
4363         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4364         ext4_commit_super(sb, 1);
4365         unlock_super(sb);
4366         return 0;
4367 }
4368
4369 /*
4370  * Structure to save mount options for ext4_remount's benefit
4371  */
4372 struct ext4_mount_options {
4373         unsigned long s_mount_opt;
4374         unsigned long s_mount_opt2;
4375         uid_t s_resuid;
4376         gid_t s_resgid;
4377         unsigned long s_commit_interval;
4378         u32 s_min_batch_time, s_max_batch_time;
4379 #ifdef CONFIG_QUOTA
4380         int s_jquota_fmt;
4381         char *s_qf_names[MAXQUOTAS];
4382 #endif
4383 };
4384
4385 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4386 {
4387         struct ext4_super_block *es;
4388         struct ext4_sb_info *sbi = EXT4_SB(sb);
4389         ext4_fsblk_t n_blocks_count = 0;
4390         unsigned long old_sb_flags;
4391         struct ext4_mount_options old_opts;
4392         int enable_quota = 0;
4393         ext4_group_t g;
4394         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4395         int err = 0;
4396 #ifdef CONFIG_QUOTA
4397         int i;
4398 #endif
4399         char *orig_data = kstrdup(data, GFP_KERNEL);
4400
4401         /* Store the original options */
4402         lock_super(sb);
4403         old_sb_flags = sb->s_flags;
4404         old_opts.s_mount_opt = sbi->s_mount_opt;
4405         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4406         old_opts.s_resuid = sbi->s_resuid;
4407         old_opts.s_resgid = sbi->s_resgid;
4408         old_opts.s_commit_interval = sbi->s_commit_interval;
4409         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4410         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4411 #ifdef CONFIG_QUOTA
4412         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4413         for (i = 0; i < MAXQUOTAS; i++)
4414                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4415 #endif
4416         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4417                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4418
4419         /*
4420          * Allow the "check" option to be passed as a remount option.
4421          */
4422         if (!parse_options(data, sb, NULL, &journal_ioprio,
4423                            &n_blocks_count, 1)) {
4424                 err = -EINVAL;
4425                 goto restore_opts;
4426         }
4427
4428         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4429                 ext4_abort(sb, "Abort forced by user");
4430
4431         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4432                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4433
4434         es = sbi->s_es;
4435
4436         if (sbi->s_journal) {
4437                 ext4_init_journal_params(sb, sbi->s_journal);
4438                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4439         }
4440
4441         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
4442                 n_blocks_count > ext4_blocks_count(es)) {
4443                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4444                         err = -EROFS;
4445                         goto restore_opts;
4446                 }
4447
4448                 if (*flags & MS_RDONLY) {
4449                         err = dquot_suspend(sb, -1);
4450                         if (err < 0)
4451                                 goto restore_opts;
4452
4453                         /*
4454                          * First of all, the unconditional stuff we have to do
4455                          * to disable replay of the journal when we next remount
4456                          */
4457                         sb->s_flags |= MS_RDONLY;
4458
4459                         /*
4460                          * OK, test if we are remounting a valid rw partition
4461                          * readonly, and if so set the rdonly flag and then
4462                          * mark the partition as valid again.
4463                          */
4464                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4465                             (sbi->s_mount_state & EXT4_VALID_FS))
4466                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4467
4468                         if (sbi->s_journal)
4469                                 ext4_mark_recovery_complete(sb, es);
4470                 } else {
4471                         /* Make sure we can mount this feature set readwrite */
4472                         if (!ext4_feature_set_ok(sb, 0)) {
4473                                 err = -EROFS;
4474                                 goto restore_opts;
4475                         }
4476                         /*
4477                          * Make sure the group descriptor checksums
4478                          * are sane.  If they aren't, refuse to remount r/w.
4479                          */
4480                         for (g = 0; g < sbi->s_groups_count; g++) {
4481                                 struct ext4_group_desc *gdp =
4482                                         ext4_get_group_desc(sb, g, NULL);
4483
4484                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4485                                         ext4_msg(sb, KERN_ERR,
4486                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4487                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4488                                                le16_to_cpu(gdp->bg_checksum));
4489                                         err = -EINVAL;
4490                                         goto restore_opts;
4491                                 }
4492                         }
4493
4494                         /*
4495                          * If we have an unprocessed orphan list hanging
4496                          * around from a previously readonly bdev mount,
4497                          * require a full umount/remount for now.
4498                          */
4499                         if (es->s_last_orphan) {
4500                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4501                                        "remount RDWR because of unprocessed "
4502                                        "orphan inode list.  Please "
4503                                        "umount/remount instead");
4504                                 err = -EINVAL;
4505                                 goto restore_opts;
4506                         }
4507
4508                         /*
4509                          * Mounting a RDONLY partition read-write, so reread
4510                          * and store the current valid flag.  (It may have
4511                          * been changed by e2fsck since we originally mounted
4512                          * the partition.)
4513                          */
4514                         if (sbi->s_journal)
4515                                 ext4_clear_journal_err(sb, es);
4516                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4517                         if ((err = ext4_group_extend(sb, es, n_blocks_count)))
4518                                 goto restore_opts;
4519                         if (!ext4_setup_super(sb, es, 0))
4520                                 sb->s_flags &= ~MS_RDONLY;
4521                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4522                                                      EXT4_FEATURE_INCOMPAT_MMP))
4523                                 if (ext4_multi_mount_protect(sb,
4524                                                 le64_to_cpu(es->s_mmp_block))) {
4525                                         err = -EROFS;
4526                                         goto restore_opts;
4527                                 }
4528                         enable_quota = 1;
4529                 }
4530         }
4531
4532         /*
4533          * Reinitialize lazy itable initialization thread based on
4534          * current settings
4535          */
4536         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4537                 ext4_unregister_li_request(sb);
4538         else {
4539                 ext4_group_t first_not_zeroed;
4540                 first_not_zeroed = ext4_has_uninit_itable(sb);
4541                 ext4_register_li_request(sb, first_not_zeroed);
4542         }
4543
4544         ext4_setup_system_zone(sb);
4545         if (sbi->s_journal == NULL)
4546                 ext4_commit_super(sb, 1);
4547
4548 #ifdef CONFIG_QUOTA
4549         /* Release old quota file names */
4550         for (i = 0; i < MAXQUOTAS; i++)
4551                 if (old_opts.s_qf_names[i] &&
4552                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4553                         kfree(old_opts.s_qf_names[i]);
4554 #endif
4555         unlock_super(sb);
4556         if (enable_quota)
4557                 dquot_resume(sb, -1);
4558
4559         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4560         kfree(orig_data);
4561         return 0;
4562
4563 restore_opts:
4564         sb->s_flags = old_sb_flags;
4565         sbi->s_mount_opt = old_opts.s_mount_opt;
4566         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4567         sbi->s_resuid = old_opts.s_resuid;
4568         sbi->s_resgid = old_opts.s_resgid;
4569         sbi->s_commit_interval = old_opts.s_commit_interval;
4570         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4571         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4572 #ifdef CONFIG_QUOTA
4573         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4574         for (i = 0; i < MAXQUOTAS; i++) {
4575                 if (sbi->s_qf_names[i] &&
4576                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4577                         kfree(sbi->s_qf_names[i]);
4578                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4579         }
4580 #endif
4581         unlock_super(sb);
4582         kfree(orig_data);
4583         return err;
4584 }
4585
4586 /*
4587  * Note: calculating the overhead so we can be compatible with
4588  * historical BSD practice is quite difficult in the face of
4589  * clusters/bigalloc.  This is because multiple metadata blocks from
4590  * different block group can end up in the same allocation cluster.
4591  * Calculating the exact overhead in the face of clustered allocation
4592  * requires either O(all block bitmaps) in memory or O(number of block
4593  * groups**2) in time.  We will still calculate the superblock for
4594  * older file systems --- and if we come across with a bigalloc file
4595  * system with zero in s_overhead_clusters the estimate will be close to
4596  * correct especially for very large cluster sizes --- but for newer
4597  * file systems, it's better to calculate this figure once at mkfs
4598  * time, and store it in the superblock.  If the superblock value is
4599  * present (even for non-bigalloc file systems), we will use it.
4600  */
4601 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4602 {
4603         struct super_block *sb = dentry->d_sb;
4604         struct ext4_sb_info *sbi = EXT4_SB(sb);
4605         struct ext4_super_block *es = sbi->s_es;
4606         struct ext4_group_desc *gdp;
4607         u64 fsid;
4608         s64 bfree;
4609
4610         if (test_opt(sb, MINIX_DF)) {
4611                 sbi->s_overhead_last = 0;
4612         } else if (es->s_overhead_clusters) {
4613                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4614         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4615                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4616                 ext4_fsblk_t overhead = 0;
4617
4618                 /*
4619                  * Compute the overhead (FS structures).  This is constant
4620                  * for a given filesystem unless the number of block groups
4621                  * changes so we cache the previous value until it does.
4622                  */
4623
4624                 /*
4625                  * All of the blocks before first_data_block are
4626                  * overhead
4627                  */
4628                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4629
4630                 /*
4631                  * Add the overhead found in each block group
4632                  */
4633                 for (i = 0; i < ngroups; i++) {
4634                         gdp = ext4_get_group_desc(sb, i, NULL);
4635                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4636                         cond_resched();
4637                 }
4638                 sbi->s_overhead_last = overhead;
4639                 smp_wmb();
4640                 sbi->s_blocks_last = ext4_blocks_count(es);
4641         }
4642
4643         buf->f_type = EXT4_SUPER_MAGIC;
4644         buf->f_bsize = sb->s_blocksize;
4645         buf->f_blocks = (ext4_blocks_count(es) -
4646                          EXT4_C2B(sbi, sbi->s_overhead_last));
4647         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4648                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4649         /* prevent underflow in case that few free space is available */
4650         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4651         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4652         if (buf->f_bfree < ext4_r_blocks_count(es))
4653                 buf->f_bavail = 0;
4654         buf->f_files = le32_to_cpu(es->s_inodes_count);
4655         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4656         buf->f_namelen = EXT4_NAME_LEN;
4657         fsid = le64_to_cpup((void *)es->s_uuid) ^
4658                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4659         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4660         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4661
4662         return 0;
4663 }
4664
4665 /* Helper function for writing quotas on sync - we need to start transaction
4666  * before quota file is locked for write. Otherwise the are possible deadlocks:
4667  * Process 1                         Process 2
4668  * ext4_create()                     quota_sync()
4669  *   jbd2_journal_start()                  write_dquot()
4670  *   dquot_initialize()                         down(dqio_mutex)
4671  *     down(dqio_mutex)                    jbd2_journal_start()
4672  *
4673  */
4674
4675 #ifdef CONFIG_QUOTA
4676
4677 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4678 {
4679         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4680 }
4681
4682 static int ext4_write_dquot(struct dquot *dquot)
4683 {
4684         int ret, err;
4685         handle_t *handle;
4686         struct inode *inode;
4687
4688         inode = dquot_to_inode(dquot);
4689         handle = ext4_journal_start(inode,
4690                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4691         if (IS_ERR(handle))
4692                 return PTR_ERR(handle);
4693         ret = dquot_commit(dquot);
4694         err = ext4_journal_stop(handle);
4695         if (!ret)
4696                 ret = err;
4697         return ret;
4698 }
4699
4700 static int ext4_acquire_dquot(struct dquot *dquot)
4701 {
4702         int ret, err;
4703         handle_t *handle;
4704
4705         handle = ext4_journal_start(dquot_to_inode(dquot),
4706                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4707         if (IS_ERR(handle))
4708                 return PTR_ERR(handle);
4709         ret = dquot_acquire(dquot);
4710         err = ext4_journal_stop(handle);
4711         if (!ret)
4712                 ret = err;
4713         return ret;
4714 }
4715
4716 static int ext4_release_dquot(struct dquot *dquot)
4717 {
4718         int ret, err;
4719         handle_t *handle;
4720
4721         handle = ext4_journal_start(dquot_to_inode(dquot),
4722                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4723         if (IS_ERR(handle)) {
4724                 /* Release dquot anyway to avoid endless cycle in dqput() */
4725                 dquot_release(dquot);
4726                 return PTR_ERR(handle);
4727         }
4728         ret = dquot_release(dquot);
4729         err = ext4_journal_stop(handle);
4730         if (!ret)
4731                 ret = err;
4732         return ret;
4733 }
4734
4735 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4736 {
4737         /* Are we journaling quotas? */
4738         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4739             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4740                 dquot_mark_dquot_dirty(dquot);
4741                 return ext4_write_dquot(dquot);
4742         } else {
4743                 return dquot_mark_dquot_dirty(dquot);
4744         }
4745 }
4746
4747 static int ext4_write_info(struct super_block *sb, int type)
4748 {
4749         int ret, err;
4750         handle_t *handle;
4751
4752         /* Data block + inode block */
4753         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4754         if (IS_ERR(handle))
4755                 return PTR_ERR(handle);
4756         ret = dquot_commit_info(sb, type);
4757         err = ext4_journal_stop(handle);
4758         if (!ret)
4759                 ret = err;
4760         return ret;
4761 }
4762
4763 /*
4764  * Turn on quotas during mount time - we need to find
4765  * the quota file and such...
4766  */
4767 static int ext4_quota_on_mount(struct super_block *sb, int type)
4768 {
4769         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4770                                         EXT4_SB(sb)->s_jquota_fmt, type);
4771 }
4772
4773 /*
4774  * Standard function to be called on quota_on
4775  */
4776 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4777                          struct path *path)
4778 {
4779         int err;
4780
4781         if (!test_opt(sb, QUOTA))
4782                 return -EINVAL;
4783
4784         /* Quotafile not on the same filesystem? */
4785         if (path->dentry->d_sb != sb)
4786                 return -EXDEV;
4787         /* Journaling quota? */
4788         if (EXT4_SB(sb)->s_qf_names[type]) {
4789                 /* Quotafile not in fs root? */
4790                 if (path->dentry->d_parent != sb->s_root)
4791                         ext4_msg(sb, KERN_WARNING,
4792                                 "Quota file not on filesystem root. "
4793                                 "Journaled quota will not work");
4794         }
4795
4796         /*
4797          * When we journal data on quota file, we have to flush journal to see
4798          * all updates to the file when we bypass pagecache...
4799          */
4800         if (EXT4_SB(sb)->s_journal &&
4801             ext4_should_journal_data(path->dentry->d_inode)) {
4802                 /*
4803                  * We don't need to lock updates but journal_flush() could
4804                  * otherwise be livelocked...
4805                  */
4806                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4807                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4808                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4809                 if (err)
4810                         return err;
4811         }
4812
4813         return dquot_quota_on(sb, type, format_id, path);
4814 }
4815
4816 static int ext4_quota_off(struct super_block *sb, int type)
4817 {
4818         struct inode *inode = sb_dqopt(sb)->files[type];
4819         handle_t *handle;
4820
4821         /* Force all delayed allocation blocks to be allocated.
4822          * Caller already holds s_umount sem */
4823         if (test_opt(sb, DELALLOC))
4824                 sync_filesystem(sb);
4825
4826         if (!inode)
4827                 goto out;
4828
4829         /* Update modification times of quota files when userspace can
4830          * start looking at them */
4831         handle = ext4_journal_start(inode, 1);
4832         if (IS_ERR(handle))
4833                 goto out;
4834         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4835         ext4_mark_inode_dirty(handle, inode);
4836         ext4_journal_stop(handle);
4837
4838 out:
4839         return dquot_quota_off(sb, type);
4840 }
4841
4842 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4843  * acquiring the locks... As quota files are never truncated and quota code
4844  * itself serializes the operations (and no one else should touch the files)
4845  * we don't have to be afraid of races */
4846 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4847                                size_t len, loff_t off)
4848 {
4849         struct inode *inode = sb_dqopt(sb)->files[type];
4850         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4851         int err = 0;
4852         int offset = off & (sb->s_blocksize - 1);
4853         int tocopy;
4854         size_t toread;
4855         struct buffer_head *bh;
4856         loff_t i_size = i_size_read(inode);
4857
4858         if (off > i_size)
4859                 return 0;
4860         if (off+len > i_size)
4861                 len = i_size-off;
4862         toread = len;
4863         while (toread > 0) {
4864                 tocopy = sb->s_blocksize - offset < toread ?
4865                                 sb->s_blocksize - offset : toread;
4866                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4867                 if (err)
4868                         return err;
4869                 if (!bh)        /* A hole? */
4870                         memset(data, 0, tocopy);
4871                 else
4872                         memcpy(data, bh->b_data+offset, tocopy);
4873                 brelse(bh);
4874                 offset = 0;
4875                 toread -= tocopy;
4876                 data += tocopy;
4877                 blk++;
4878         }
4879         return len;
4880 }
4881
4882 /* Write to quotafile (we know the transaction is already started and has
4883  * enough credits) */
4884 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4885                                 const char *data, size_t len, loff_t off)
4886 {
4887         struct inode *inode = sb_dqopt(sb)->files[type];
4888         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4889         int err = 0;
4890         int offset = off & (sb->s_blocksize - 1);
4891         struct buffer_head *bh;
4892         handle_t *handle = journal_current_handle();
4893
4894         if (EXT4_SB(sb)->s_journal && !handle) {
4895                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4896                         " cancelled because transaction is not started",
4897                         (unsigned long long)off, (unsigned long long)len);
4898                 return -EIO;
4899         }
4900         /*
4901          * Since we account only one data block in transaction credits,
4902          * then it is impossible to cross a block boundary.
4903          */
4904         if (sb->s_blocksize - offset < len) {
4905                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4906                         " cancelled because not block aligned",
4907                         (unsigned long long)off, (unsigned long long)len);
4908                 return -EIO;
4909         }
4910
4911         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4912         bh = ext4_bread(handle, inode, blk, 1, &err);
4913         if (!bh)
4914                 goto out;
4915         err = ext4_journal_get_write_access(handle, bh);
4916         if (err) {
4917                 brelse(bh);
4918                 goto out;
4919         }
4920         lock_buffer(bh);
4921         memcpy(bh->b_data+offset, data, len);
4922         flush_dcache_page(bh->b_page);
4923         unlock_buffer(bh);
4924         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4925         brelse(bh);
4926 out:
4927         if (err) {
4928                 mutex_unlock(&inode->i_mutex);
4929                 return err;
4930         }
4931         if (inode->i_size < off + len) {
4932                 i_size_write(inode, off + len);
4933                 EXT4_I(inode)->i_disksize = inode->i_size;
4934                 ext4_mark_inode_dirty(handle, inode);
4935         }
4936         mutex_unlock(&inode->i_mutex);
4937         return len;
4938 }
4939
4940 #endif
4941
4942 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4943                        const char *dev_name, void *data)
4944 {
4945         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4946 }
4947
4948 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4949 static inline void register_as_ext2(void)
4950 {
4951         int err = register_filesystem(&ext2_fs_type);
4952         if (err)
4953                 printk(KERN_WARNING
4954                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4955 }
4956
4957 static inline void unregister_as_ext2(void)
4958 {
4959         unregister_filesystem(&ext2_fs_type);
4960 }
4961
4962 static inline int ext2_feature_set_ok(struct super_block *sb)
4963 {
4964         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4965                 return 0;
4966         if (sb->s_flags & MS_RDONLY)
4967                 return 1;
4968         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4969                 return 0;
4970         return 1;
4971 }
4972 MODULE_ALIAS("ext2");
4973 #else
4974 static inline void register_as_ext2(void) { }
4975 static inline void unregister_as_ext2(void) { }
4976 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4977 #endif
4978
4979 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4980 static inline void register_as_ext3(void)
4981 {
4982         int err = register_filesystem(&ext3_fs_type);
4983         if (err)
4984                 printk(KERN_WARNING
4985                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4986 }
4987
4988 static inline void unregister_as_ext3(void)
4989 {
4990         unregister_filesystem(&ext3_fs_type);
4991 }
4992
4993 static inline int ext3_feature_set_ok(struct super_block *sb)
4994 {
4995         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4996                 return 0;
4997         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4998                 return 0;
4999         if (sb->s_flags & MS_RDONLY)
5000                 return 1;
5001         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5002                 return 0;
5003         return 1;
5004 }
5005 MODULE_ALIAS("ext3");
5006 #else
5007 static inline void register_as_ext3(void) { }
5008 static inline void unregister_as_ext3(void) { }
5009 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5010 #endif
5011
5012 static struct file_system_type ext4_fs_type = {
5013         .owner          = THIS_MODULE,
5014         .name           = "ext4",
5015         .mount          = ext4_mount,
5016         .kill_sb        = kill_block_super,
5017         .fs_flags       = FS_REQUIRES_DEV,
5018 };
5019
5020 static int __init ext4_init_feat_adverts(void)
5021 {
5022         struct ext4_features *ef;
5023         int ret = -ENOMEM;
5024
5025         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5026         if (!ef)
5027                 goto out;
5028
5029         ef->f_kobj.kset = ext4_kset;
5030         init_completion(&ef->f_kobj_unregister);
5031         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5032                                    "features");
5033         if (ret) {
5034                 kfree(ef);
5035                 goto out;
5036         }
5037
5038         ext4_feat = ef;
5039         ret = 0;
5040 out:
5041         return ret;
5042 }
5043
5044 static void ext4_exit_feat_adverts(void)
5045 {
5046         kobject_put(&ext4_feat->f_kobj);
5047         wait_for_completion(&ext4_feat->f_kobj_unregister);
5048         kfree(ext4_feat);
5049 }
5050
5051 /* Shared across all ext4 file systems */
5052 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5053 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5054
5055 static int __init ext4_init_fs(void)
5056 {
5057         int i, err;
5058
5059         ext4_check_flag_values();
5060
5061         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5062                 mutex_init(&ext4__aio_mutex[i]);
5063                 init_waitqueue_head(&ext4__ioend_wq[i]);
5064         }
5065
5066         err = ext4_init_pageio();
5067         if (err)
5068                 return err;
5069         err = ext4_init_system_zone();
5070         if (err)
5071                 goto out6;
5072         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5073         if (!ext4_kset)
5074                 goto out5;
5075         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5076
5077         err = ext4_init_feat_adverts();
5078         if (err)
5079                 goto out4;
5080
5081         err = ext4_init_mballoc();
5082         if (err)
5083                 goto out3;
5084
5085         err = ext4_init_xattr();
5086         if (err)
5087                 goto out2;
5088         err = init_inodecache();
5089         if (err)
5090                 goto out1;
5091         register_as_ext3();
5092         register_as_ext2();
5093         err = register_filesystem(&ext4_fs_type);
5094         if (err)
5095                 goto out;
5096
5097         ext4_li_info = NULL;
5098         mutex_init(&ext4_li_mtx);
5099         return 0;
5100 out:
5101         unregister_as_ext2();
5102         unregister_as_ext3();
5103         destroy_inodecache();
5104 out1:
5105         ext4_exit_xattr();
5106 out2:
5107         ext4_exit_mballoc();
5108 out3:
5109         ext4_exit_feat_adverts();
5110 out4:
5111         if (ext4_proc_root)
5112                 remove_proc_entry("fs/ext4", NULL);
5113         kset_unregister(ext4_kset);
5114 out5:
5115         ext4_exit_system_zone();
5116 out6:
5117         ext4_exit_pageio();
5118         return err;
5119 }
5120
5121 static void __exit ext4_exit_fs(void)
5122 {
5123         ext4_destroy_lazyinit_thread();
5124         unregister_as_ext2();
5125         unregister_as_ext3();
5126         unregister_filesystem(&ext4_fs_type);
5127         destroy_inodecache();
5128         ext4_exit_xattr();
5129         ext4_exit_mballoc();
5130         ext4_exit_feat_adverts();
5131         remove_proc_entry("fs/ext4", NULL);
5132         kset_unregister(ext4_kset);
5133         ext4_exit_system_zone();
5134         ext4_exit_pageio();
5135 }
5136
5137 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5138 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5139 MODULE_LICENSE("GPL");
5140 module_init(ext4_init_fs)
5141 module_exit(ext4_exit_fs)