]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - fs/ext4/super.c
3858767ec672ef214ad5f288f30932ee0b5d933a
[~shefty/rdma-dev.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         INIT_LIST_HEAD(&inode->i_dentry);
934         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
935 }
936
937 static void ext4_destroy_inode(struct inode *inode)
938 {
939         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
940                 ext4_msg(inode->i_sb, KERN_ERR,
941                          "Inode %lu (%p): orphan list check failed!",
942                          inode->i_ino, EXT4_I(inode));
943                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
944                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
945                                 true);
946                 dump_stack();
947         }
948         call_rcu(&inode->i_rcu, ext4_i_callback);
949 }
950
951 static void init_once(void *foo)
952 {
953         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
954
955         INIT_LIST_HEAD(&ei->i_orphan);
956 #ifdef CONFIG_EXT4_FS_XATTR
957         init_rwsem(&ei->xattr_sem);
958 #endif
959         init_rwsem(&ei->i_data_sem);
960         inode_init_once(&ei->vfs_inode);
961 }
962
963 static int init_inodecache(void)
964 {
965         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
966                                              sizeof(struct ext4_inode_info),
967                                              0, (SLAB_RECLAIM_ACCOUNT|
968                                                 SLAB_MEM_SPREAD),
969                                              init_once);
970         if (ext4_inode_cachep == NULL)
971                 return -ENOMEM;
972         return 0;
973 }
974
975 static void destroy_inodecache(void)
976 {
977         kmem_cache_destroy(ext4_inode_cachep);
978 }
979
980 void ext4_clear_inode(struct inode *inode)
981 {
982         invalidate_inode_buffers(inode);
983         end_writeback(inode);
984         dquot_drop(inode);
985         ext4_discard_preallocations(inode);
986         if (EXT4_I(inode)->jinode) {
987                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
988                                                EXT4_I(inode)->jinode);
989                 jbd2_free_inode(EXT4_I(inode)->jinode);
990                 EXT4_I(inode)->jinode = NULL;
991         }
992 }
993
994 static inline void ext4_show_quota_options(struct seq_file *seq,
995                                            struct super_block *sb)
996 {
997 #if defined(CONFIG_QUOTA)
998         struct ext4_sb_info *sbi = EXT4_SB(sb);
999
1000         if (sbi->s_jquota_fmt) {
1001                 char *fmtname = "";
1002
1003                 switch (sbi->s_jquota_fmt) {
1004                 case QFMT_VFS_OLD:
1005                         fmtname = "vfsold";
1006                         break;
1007                 case QFMT_VFS_V0:
1008                         fmtname = "vfsv0";
1009                         break;
1010                 case QFMT_VFS_V1:
1011                         fmtname = "vfsv1";
1012                         break;
1013                 }
1014                 seq_printf(seq, ",jqfmt=%s", fmtname);
1015         }
1016
1017         if (sbi->s_qf_names[USRQUOTA])
1018                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1019
1020         if (sbi->s_qf_names[GRPQUOTA])
1021                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1022
1023         if (test_opt(sb, USRQUOTA))
1024                 seq_puts(seq, ",usrquota");
1025
1026         if (test_opt(sb, GRPQUOTA))
1027                 seq_puts(seq, ",grpquota");
1028 #endif
1029 }
1030
1031 /*
1032  * Show an option if
1033  *  - it's set to a non-default value OR
1034  *  - if the per-sb default is different from the global default
1035  */
1036 static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
1037 {
1038         int def_errors;
1039         unsigned long def_mount_opts;
1040         struct super_block *sb = vfs->mnt_sb;
1041         struct ext4_sb_info *sbi = EXT4_SB(sb);
1042         struct ext4_super_block *es = sbi->s_es;
1043
1044         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1045         def_errors     = le16_to_cpu(es->s_errors);
1046
1047         if (sbi->s_sb_block != 1)
1048                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1049         if (test_opt(sb, MINIX_DF))
1050                 seq_puts(seq, ",minixdf");
1051         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1052                 seq_puts(seq, ",grpid");
1053         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1054                 seq_puts(seq, ",nogrpid");
1055         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1056             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1057                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1058         }
1059         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1060             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1061                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1062         }
1063         if (test_opt(sb, ERRORS_RO)) {
1064                 if (def_errors == EXT4_ERRORS_PANIC ||
1065                     def_errors == EXT4_ERRORS_CONTINUE) {
1066                         seq_puts(seq, ",errors=remount-ro");
1067                 }
1068         }
1069         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1070                 seq_puts(seq, ",errors=continue");
1071         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1072                 seq_puts(seq, ",errors=panic");
1073         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1074                 seq_puts(seq, ",nouid32");
1075         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1076                 seq_puts(seq, ",debug");
1077 #ifdef CONFIG_EXT4_FS_XATTR
1078         if (test_opt(sb, XATTR_USER))
1079                 seq_puts(seq, ",user_xattr");
1080         if (!test_opt(sb, XATTR_USER))
1081                 seq_puts(seq, ",nouser_xattr");
1082 #endif
1083 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1084         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1085                 seq_puts(seq, ",acl");
1086         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1087                 seq_puts(seq, ",noacl");
1088 #endif
1089         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1090                 seq_printf(seq, ",commit=%u",
1091                            (unsigned) (sbi->s_commit_interval / HZ));
1092         }
1093         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1094                 seq_printf(seq, ",min_batch_time=%u",
1095                            (unsigned) sbi->s_min_batch_time);
1096         }
1097         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1098                 seq_printf(seq, ",max_batch_time=%u",
1099                            (unsigned) sbi->s_min_batch_time);
1100         }
1101
1102         /*
1103          * We're changing the default of barrier mount option, so
1104          * let's always display its mount state so it's clear what its
1105          * status is.
1106          */
1107         seq_puts(seq, ",barrier=");
1108         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1109         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1110                 seq_puts(seq, ",journal_async_commit");
1111         else if (test_opt(sb, JOURNAL_CHECKSUM))
1112                 seq_puts(seq, ",journal_checksum");
1113         if (test_opt(sb, I_VERSION))
1114                 seq_puts(seq, ",i_version");
1115         if (!test_opt(sb, DELALLOC) &&
1116             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1117                 seq_puts(seq, ",nodelalloc");
1118
1119         if (!test_opt(sb, MBLK_IO_SUBMIT))
1120                 seq_puts(seq, ",nomblk_io_submit");
1121         if (sbi->s_stripe)
1122                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1123         /*
1124          * journal mode get enabled in different ways
1125          * So just print the value even if we didn't specify it
1126          */
1127         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1128                 seq_puts(seq, ",data=journal");
1129         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1130                 seq_puts(seq, ",data=ordered");
1131         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1132                 seq_puts(seq, ",data=writeback");
1133
1134         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1135                 seq_printf(seq, ",inode_readahead_blks=%u",
1136                            sbi->s_inode_readahead_blks);
1137
1138         if (test_opt(sb, DATA_ERR_ABORT))
1139                 seq_puts(seq, ",data_err=abort");
1140
1141         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1142                 seq_puts(seq, ",noauto_da_alloc");
1143
1144         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1145                 seq_puts(seq, ",discard");
1146
1147         if (test_opt(sb, NOLOAD))
1148                 seq_puts(seq, ",norecovery");
1149
1150         if (test_opt(sb, DIOREAD_NOLOCK))
1151                 seq_puts(seq, ",dioread_nolock");
1152
1153         if (test_opt(sb, BLOCK_VALIDITY) &&
1154             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1155                 seq_puts(seq, ",block_validity");
1156
1157         if (!test_opt(sb, INIT_INODE_TABLE))
1158                 seq_puts(seq, ",noinit_inode_table");
1159         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1160                 seq_printf(seq, ",init_inode_table=%u",
1161                            (unsigned) sbi->s_li_wait_mult);
1162
1163         ext4_show_quota_options(seq, sb);
1164
1165         return 0;
1166 }
1167
1168 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1169                                         u64 ino, u32 generation)
1170 {
1171         struct inode *inode;
1172
1173         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1174                 return ERR_PTR(-ESTALE);
1175         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1176                 return ERR_PTR(-ESTALE);
1177
1178         /* iget isn't really right if the inode is currently unallocated!!
1179          *
1180          * ext4_read_inode will return a bad_inode if the inode had been
1181          * deleted, so we should be safe.
1182          *
1183          * Currently we don't know the generation for parent directory, so
1184          * a generation of 0 means "accept any"
1185          */
1186         inode = ext4_iget(sb, ino);
1187         if (IS_ERR(inode))
1188                 return ERR_CAST(inode);
1189         if (generation && inode->i_generation != generation) {
1190                 iput(inode);
1191                 return ERR_PTR(-ESTALE);
1192         }
1193
1194         return inode;
1195 }
1196
1197 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1198                                         int fh_len, int fh_type)
1199 {
1200         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1201                                     ext4_nfs_get_inode);
1202 }
1203
1204 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1205                                         int fh_len, int fh_type)
1206 {
1207         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1208                                     ext4_nfs_get_inode);
1209 }
1210
1211 /*
1212  * Try to release metadata pages (indirect blocks, directories) which are
1213  * mapped via the block device.  Since these pages could have journal heads
1214  * which would prevent try_to_free_buffers() from freeing them, we must use
1215  * jbd2 layer's try_to_free_buffers() function to release them.
1216  */
1217 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1218                                  gfp_t wait)
1219 {
1220         journal_t *journal = EXT4_SB(sb)->s_journal;
1221
1222         WARN_ON(PageChecked(page));
1223         if (!page_has_buffers(page))
1224                 return 0;
1225         if (journal)
1226                 return jbd2_journal_try_to_free_buffers(journal, page,
1227                                                         wait & ~__GFP_WAIT);
1228         return try_to_free_buffers(page);
1229 }
1230
1231 #ifdef CONFIG_QUOTA
1232 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1233 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1234
1235 static int ext4_write_dquot(struct dquot *dquot);
1236 static int ext4_acquire_dquot(struct dquot *dquot);
1237 static int ext4_release_dquot(struct dquot *dquot);
1238 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1239 static int ext4_write_info(struct super_block *sb, int type);
1240 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1241                          struct path *path);
1242 static int ext4_quota_off(struct super_block *sb, int type);
1243 static int ext4_quota_on_mount(struct super_block *sb, int type);
1244 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1245                                size_t len, loff_t off);
1246 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1247                                 const char *data, size_t len, loff_t off);
1248
1249 static const struct dquot_operations ext4_quota_operations = {
1250         .get_reserved_space = ext4_get_reserved_space,
1251         .write_dquot    = ext4_write_dquot,
1252         .acquire_dquot  = ext4_acquire_dquot,
1253         .release_dquot  = ext4_release_dquot,
1254         .mark_dirty     = ext4_mark_dquot_dirty,
1255         .write_info     = ext4_write_info,
1256         .alloc_dquot    = dquot_alloc,
1257         .destroy_dquot  = dquot_destroy,
1258 };
1259
1260 static const struct quotactl_ops ext4_qctl_operations = {
1261         .quota_on       = ext4_quota_on,
1262         .quota_off      = ext4_quota_off,
1263         .quota_sync     = dquot_quota_sync,
1264         .get_info       = dquot_get_dqinfo,
1265         .set_info       = dquot_set_dqinfo,
1266         .get_dqblk      = dquot_get_dqblk,
1267         .set_dqblk      = dquot_set_dqblk
1268 };
1269 #endif
1270
1271 static const struct super_operations ext4_sops = {
1272         .alloc_inode    = ext4_alloc_inode,
1273         .destroy_inode  = ext4_destroy_inode,
1274         .write_inode    = ext4_write_inode,
1275         .dirty_inode    = ext4_dirty_inode,
1276         .drop_inode     = ext4_drop_inode,
1277         .evict_inode    = ext4_evict_inode,
1278         .put_super      = ext4_put_super,
1279         .sync_fs        = ext4_sync_fs,
1280         .freeze_fs      = ext4_freeze,
1281         .unfreeze_fs    = ext4_unfreeze,
1282         .statfs         = ext4_statfs,
1283         .remount_fs     = ext4_remount,
1284         .show_options   = ext4_show_options,
1285 #ifdef CONFIG_QUOTA
1286         .quota_read     = ext4_quota_read,
1287         .quota_write    = ext4_quota_write,
1288 #endif
1289         .bdev_try_to_free_page = bdev_try_to_free_page,
1290 };
1291
1292 static const struct super_operations ext4_nojournal_sops = {
1293         .alloc_inode    = ext4_alloc_inode,
1294         .destroy_inode  = ext4_destroy_inode,
1295         .write_inode    = ext4_write_inode,
1296         .dirty_inode    = ext4_dirty_inode,
1297         .drop_inode     = ext4_drop_inode,
1298         .evict_inode    = ext4_evict_inode,
1299         .write_super    = ext4_write_super,
1300         .put_super      = ext4_put_super,
1301         .statfs         = ext4_statfs,
1302         .remount_fs     = ext4_remount,
1303         .show_options   = ext4_show_options,
1304 #ifdef CONFIG_QUOTA
1305         .quota_read     = ext4_quota_read,
1306         .quota_write    = ext4_quota_write,
1307 #endif
1308         .bdev_try_to_free_page = bdev_try_to_free_page,
1309 };
1310
1311 static const struct export_operations ext4_export_ops = {
1312         .fh_to_dentry = ext4_fh_to_dentry,
1313         .fh_to_parent = ext4_fh_to_parent,
1314         .get_parent = ext4_get_parent,
1315 };
1316
1317 enum {
1318         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1319         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1320         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1321         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1322         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1323         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1324         Opt_journal_update, Opt_journal_dev,
1325         Opt_journal_checksum, Opt_journal_async_commit,
1326         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1327         Opt_data_err_abort, Opt_data_err_ignore,
1328         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1329         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1330         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1331         Opt_resize, Opt_usrquota, Opt_grpquota, Opt_i_version,
1332         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1333         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1334         Opt_inode_readahead_blks, Opt_journal_ioprio,
1335         Opt_dioread_nolock, Opt_dioread_lock,
1336         Opt_discard, Opt_nodiscard,
1337         Opt_init_inode_table, Opt_noinit_inode_table,
1338 };
1339
1340 static const match_table_t tokens = {
1341         {Opt_bsd_df, "bsddf"},
1342         {Opt_minix_df, "minixdf"},
1343         {Opt_grpid, "grpid"},
1344         {Opt_grpid, "bsdgroups"},
1345         {Opt_nogrpid, "nogrpid"},
1346         {Opt_nogrpid, "sysvgroups"},
1347         {Opt_resgid, "resgid=%u"},
1348         {Opt_resuid, "resuid=%u"},
1349         {Opt_sb, "sb=%u"},
1350         {Opt_err_cont, "errors=continue"},
1351         {Opt_err_panic, "errors=panic"},
1352         {Opt_err_ro, "errors=remount-ro"},
1353         {Opt_nouid32, "nouid32"},
1354         {Opt_debug, "debug"},
1355         {Opt_oldalloc, "oldalloc"},
1356         {Opt_orlov, "orlov"},
1357         {Opt_user_xattr, "user_xattr"},
1358         {Opt_nouser_xattr, "nouser_xattr"},
1359         {Opt_acl, "acl"},
1360         {Opt_noacl, "noacl"},
1361         {Opt_noload, "noload"},
1362         {Opt_noload, "norecovery"},
1363         {Opt_nobh, "nobh"},
1364         {Opt_bh, "bh"},
1365         {Opt_commit, "commit=%u"},
1366         {Opt_min_batch_time, "min_batch_time=%u"},
1367         {Opt_max_batch_time, "max_batch_time=%u"},
1368         {Opt_journal_update, "journal=update"},
1369         {Opt_journal_dev, "journal_dev=%u"},
1370         {Opt_journal_checksum, "journal_checksum"},
1371         {Opt_journal_async_commit, "journal_async_commit"},
1372         {Opt_abort, "abort"},
1373         {Opt_data_journal, "data=journal"},
1374         {Opt_data_ordered, "data=ordered"},
1375         {Opt_data_writeback, "data=writeback"},
1376         {Opt_data_err_abort, "data_err=abort"},
1377         {Opt_data_err_ignore, "data_err=ignore"},
1378         {Opt_offusrjquota, "usrjquota="},
1379         {Opt_usrjquota, "usrjquota=%s"},
1380         {Opt_offgrpjquota, "grpjquota="},
1381         {Opt_grpjquota, "grpjquota=%s"},
1382         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1383         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1384         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1385         {Opt_grpquota, "grpquota"},
1386         {Opt_noquota, "noquota"},
1387         {Opt_quota, "quota"},
1388         {Opt_usrquota, "usrquota"},
1389         {Opt_barrier, "barrier=%u"},
1390         {Opt_barrier, "barrier"},
1391         {Opt_nobarrier, "nobarrier"},
1392         {Opt_i_version, "i_version"},
1393         {Opt_stripe, "stripe=%u"},
1394         {Opt_resize, "resize"},
1395         {Opt_delalloc, "delalloc"},
1396         {Opt_nodelalloc, "nodelalloc"},
1397         {Opt_mblk_io_submit, "mblk_io_submit"},
1398         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1399         {Opt_block_validity, "block_validity"},
1400         {Opt_noblock_validity, "noblock_validity"},
1401         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1402         {Opt_journal_ioprio, "journal_ioprio=%u"},
1403         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1404         {Opt_auto_da_alloc, "auto_da_alloc"},
1405         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1406         {Opt_dioread_nolock, "dioread_nolock"},
1407         {Opt_dioread_lock, "dioread_lock"},
1408         {Opt_discard, "discard"},
1409         {Opt_nodiscard, "nodiscard"},
1410         {Opt_init_inode_table, "init_itable=%u"},
1411         {Opt_init_inode_table, "init_itable"},
1412         {Opt_noinit_inode_table, "noinit_itable"},
1413         {Opt_err, NULL},
1414 };
1415
1416 static ext4_fsblk_t get_sb_block(void **data)
1417 {
1418         ext4_fsblk_t    sb_block;
1419         char            *options = (char *) *data;
1420
1421         if (!options || strncmp(options, "sb=", 3) != 0)
1422                 return 1;       /* Default location */
1423
1424         options += 3;
1425         /* TODO: use simple_strtoll with >32bit ext4 */
1426         sb_block = simple_strtoul(options, &options, 0);
1427         if (*options && *options != ',') {
1428                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1429                        (char *) *data);
1430                 return 1;
1431         }
1432         if (*options == ',')
1433                 options++;
1434         *data = (void *) options;
1435
1436         return sb_block;
1437 }
1438
1439 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1440 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1441         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1442
1443 #ifdef CONFIG_QUOTA
1444 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1445 {
1446         struct ext4_sb_info *sbi = EXT4_SB(sb);
1447         char *qname;
1448
1449         if (sb_any_quota_loaded(sb) &&
1450                 !sbi->s_qf_names[qtype]) {
1451                 ext4_msg(sb, KERN_ERR,
1452                         "Cannot change journaled "
1453                         "quota options when quota turned on");
1454                 return 0;
1455         }
1456         qname = match_strdup(args);
1457         if (!qname) {
1458                 ext4_msg(sb, KERN_ERR,
1459                         "Not enough memory for storing quotafile name");
1460                 return 0;
1461         }
1462         if (sbi->s_qf_names[qtype] &&
1463                 strcmp(sbi->s_qf_names[qtype], qname)) {
1464                 ext4_msg(sb, KERN_ERR,
1465                         "%s quota file already specified", QTYPE2NAME(qtype));
1466                 kfree(qname);
1467                 return 0;
1468         }
1469         sbi->s_qf_names[qtype] = qname;
1470         if (strchr(sbi->s_qf_names[qtype], '/')) {
1471                 ext4_msg(sb, KERN_ERR,
1472                         "quotafile must be on filesystem root");
1473                 kfree(sbi->s_qf_names[qtype]);
1474                 sbi->s_qf_names[qtype] = NULL;
1475                 return 0;
1476         }
1477         set_opt(sb, QUOTA);
1478         return 1;
1479 }
1480
1481 static int clear_qf_name(struct super_block *sb, int qtype)
1482 {
1483
1484         struct ext4_sb_info *sbi = EXT4_SB(sb);
1485
1486         if (sb_any_quota_loaded(sb) &&
1487                 sbi->s_qf_names[qtype]) {
1488                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1489                         " when quota turned on");
1490                 return 0;
1491         }
1492         /*
1493          * The space will be released later when all options are confirmed
1494          * to be correct
1495          */
1496         sbi->s_qf_names[qtype] = NULL;
1497         return 1;
1498 }
1499 #endif
1500
1501 static int parse_options(char *options, struct super_block *sb,
1502                          unsigned long *journal_devnum,
1503                          unsigned int *journal_ioprio,
1504                          ext4_fsblk_t *n_blocks_count, int is_remount)
1505 {
1506         struct ext4_sb_info *sbi = EXT4_SB(sb);
1507         char *p;
1508         substring_t args[MAX_OPT_ARGS];
1509         int data_opt = 0;
1510         int option;
1511 #ifdef CONFIG_QUOTA
1512         int qfmt;
1513 #endif
1514
1515         if (!options)
1516                 return 1;
1517
1518         while ((p = strsep(&options, ",")) != NULL) {
1519                 int token;
1520                 if (!*p)
1521                         continue;
1522
1523                 /*
1524                  * Initialize args struct so we know whether arg was
1525                  * found; some options take optional arguments.
1526                  */
1527                 args[0].to = args[0].from = NULL;
1528                 token = match_token(p, tokens, args);
1529                 switch (token) {
1530                 case Opt_bsd_df:
1531                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1532                         clear_opt(sb, MINIX_DF);
1533                         break;
1534                 case Opt_minix_df:
1535                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1536                         set_opt(sb, MINIX_DF);
1537
1538                         break;
1539                 case Opt_grpid:
1540                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1541                         set_opt(sb, GRPID);
1542
1543                         break;
1544                 case Opt_nogrpid:
1545                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1546                         clear_opt(sb, GRPID);
1547
1548                         break;
1549                 case Opt_resuid:
1550                         if (match_int(&args[0], &option))
1551                                 return 0;
1552                         sbi->s_resuid = option;
1553                         break;
1554                 case Opt_resgid:
1555                         if (match_int(&args[0], &option))
1556                                 return 0;
1557                         sbi->s_resgid = option;
1558                         break;
1559                 case Opt_sb:
1560                         /* handled by get_sb_block() instead of here */
1561                         /* *sb_block = match_int(&args[0]); */
1562                         break;
1563                 case Opt_err_panic:
1564                         clear_opt(sb, ERRORS_CONT);
1565                         clear_opt(sb, ERRORS_RO);
1566                         set_opt(sb, ERRORS_PANIC);
1567                         break;
1568                 case Opt_err_ro:
1569                         clear_opt(sb, ERRORS_CONT);
1570                         clear_opt(sb, ERRORS_PANIC);
1571                         set_opt(sb, ERRORS_RO);
1572                         break;
1573                 case Opt_err_cont:
1574                         clear_opt(sb, ERRORS_RO);
1575                         clear_opt(sb, ERRORS_PANIC);
1576                         set_opt(sb, ERRORS_CONT);
1577                         break;
1578                 case Opt_nouid32:
1579                         set_opt(sb, NO_UID32);
1580                         break;
1581                 case Opt_debug:
1582                         set_opt(sb, DEBUG);
1583                         break;
1584                 case Opt_oldalloc:
1585                         ext4_msg(sb, KERN_WARNING,
1586                                  "Ignoring deprecated oldalloc option");
1587                         break;
1588                 case Opt_orlov:
1589                         ext4_msg(sb, KERN_WARNING,
1590                                  "Ignoring deprecated orlov option");
1591                         break;
1592 #ifdef CONFIG_EXT4_FS_XATTR
1593                 case Opt_user_xattr:
1594                         set_opt(sb, XATTR_USER);
1595                         break;
1596                 case Opt_nouser_xattr:
1597                         clear_opt(sb, XATTR_USER);
1598                         break;
1599 #else
1600                 case Opt_user_xattr:
1601                 case Opt_nouser_xattr:
1602                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1603                         break;
1604 #endif
1605 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1606                 case Opt_acl:
1607                         set_opt(sb, POSIX_ACL);
1608                         break;
1609                 case Opt_noacl:
1610                         clear_opt(sb, POSIX_ACL);
1611                         break;
1612 #else
1613                 case Opt_acl:
1614                 case Opt_noacl:
1615                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1616                         break;
1617 #endif
1618                 case Opt_journal_update:
1619                         /* @@@ FIXME */
1620                         /* Eventually we will want to be able to create
1621                            a journal file here.  For now, only allow the
1622                            user to specify an existing inode to be the
1623                            journal file. */
1624                         if (is_remount) {
1625                                 ext4_msg(sb, KERN_ERR,
1626                                          "Cannot specify journal on remount");
1627                                 return 0;
1628                         }
1629                         set_opt(sb, UPDATE_JOURNAL);
1630                         break;
1631                 case Opt_journal_dev:
1632                         if (is_remount) {
1633                                 ext4_msg(sb, KERN_ERR,
1634                                         "Cannot specify journal on remount");
1635                                 return 0;
1636                         }
1637                         if (match_int(&args[0], &option))
1638                                 return 0;
1639                         *journal_devnum = option;
1640                         break;
1641                 case Opt_journal_checksum:
1642                         set_opt(sb, JOURNAL_CHECKSUM);
1643                         break;
1644                 case Opt_journal_async_commit:
1645                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1646                         set_opt(sb, JOURNAL_CHECKSUM);
1647                         break;
1648                 case Opt_noload:
1649                         set_opt(sb, NOLOAD);
1650                         break;
1651                 case Opt_commit:
1652                         if (match_int(&args[0], &option))
1653                                 return 0;
1654                         if (option < 0)
1655                                 return 0;
1656                         if (option == 0)
1657                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1658                         sbi->s_commit_interval = HZ * option;
1659                         break;
1660                 case Opt_max_batch_time:
1661                         if (match_int(&args[0], &option))
1662                                 return 0;
1663                         if (option < 0)
1664                                 return 0;
1665                         if (option == 0)
1666                                 option = EXT4_DEF_MAX_BATCH_TIME;
1667                         sbi->s_max_batch_time = option;
1668                         break;
1669                 case Opt_min_batch_time:
1670                         if (match_int(&args[0], &option))
1671                                 return 0;
1672                         if (option < 0)
1673                                 return 0;
1674                         sbi->s_min_batch_time = option;
1675                         break;
1676                 case Opt_data_journal:
1677                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1678                         goto datacheck;
1679                 case Opt_data_ordered:
1680                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1681                         goto datacheck;
1682                 case Opt_data_writeback:
1683                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1684                 datacheck:
1685                         if (is_remount) {
1686                                 if (!sbi->s_journal)
1687                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1688                                 else if (test_opt(sb, DATA_FLAGS) != data_opt) {
1689                                         ext4_msg(sb, KERN_ERR,
1690                                                 "Cannot change data mode on remount");
1691                                         return 0;
1692                                 }
1693                         } else {
1694                                 clear_opt(sb, DATA_FLAGS);
1695                                 sbi->s_mount_opt |= data_opt;
1696                         }
1697                         break;
1698                 case Opt_data_err_abort:
1699                         set_opt(sb, DATA_ERR_ABORT);
1700                         break;
1701                 case Opt_data_err_ignore:
1702                         clear_opt(sb, DATA_ERR_ABORT);
1703                         break;
1704 #ifdef CONFIG_QUOTA
1705                 case Opt_usrjquota:
1706                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1707                                 return 0;
1708                         break;
1709                 case Opt_grpjquota:
1710                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1711                                 return 0;
1712                         break;
1713                 case Opt_offusrjquota:
1714                         if (!clear_qf_name(sb, USRQUOTA))
1715                                 return 0;
1716                         break;
1717                 case Opt_offgrpjquota:
1718                         if (!clear_qf_name(sb, GRPQUOTA))
1719                                 return 0;
1720                         break;
1721
1722                 case Opt_jqfmt_vfsold:
1723                         qfmt = QFMT_VFS_OLD;
1724                         goto set_qf_format;
1725                 case Opt_jqfmt_vfsv0:
1726                         qfmt = QFMT_VFS_V0;
1727                         goto set_qf_format;
1728                 case Opt_jqfmt_vfsv1:
1729                         qfmt = QFMT_VFS_V1;
1730 set_qf_format:
1731                         if (sb_any_quota_loaded(sb) &&
1732                             sbi->s_jquota_fmt != qfmt) {
1733                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1734                                         "journaled quota options when "
1735                                         "quota turned on");
1736                                 return 0;
1737                         }
1738                         sbi->s_jquota_fmt = qfmt;
1739                         break;
1740                 case Opt_quota:
1741                 case Opt_usrquota:
1742                         set_opt(sb, QUOTA);
1743                         set_opt(sb, USRQUOTA);
1744                         break;
1745                 case Opt_grpquota:
1746                         set_opt(sb, QUOTA);
1747                         set_opt(sb, GRPQUOTA);
1748                         break;
1749                 case Opt_noquota:
1750                         if (sb_any_quota_loaded(sb)) {
1751                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1752                                         "options when quota turned on");
1753                                 return 0;
1754                         }
1755                         clear_opt(sb, QUOTA);
1756                         clear_opt(sb, USRQUOTA);
1757                         clear_opt(sb, GRPQUOTA);
1758                         break;
1759 #else
1760                 case Opt_quota:
1761                 case Opt_usrquota:
1762                 case Opt_grpquota:
1763                         ext4_msg(sb, KERN_ERR,
1764                                 "quota options not supported");
1765                         break;
1766                 case Opt_usrjquota:
1767                 case Opt_grpjquota:
1768                 case Opt_offusrjquota:
1769                 case Opt_offgrpjquota:
1770                 case Opt_jqfmt_vfsold:
1771                 case Opt_jqfmt_vfsv0:
1772                 case Opt_jqfmt_vfsv1:
1773                         ext4_msg(sb, KERN_ERR,
1774                                 "journaled quota options not supported");
1775                         break;
1776                 case Opt_noquota:
1777                         break;
1778 #endif
1779                 case Opt_abort:
1780                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1781                         break;
1782                 case Opt_nobarrier:
1783                         clear_opt(sb, BARRIER);
1784                         break;
1785                 case Opt_barrier:
1786                         if (args[0].from) {
1787                                 if (match_int(&args[0], &option))
1788                                         return 0;
1789                         } else
1790                                 option = 1;     /* No argument, default to 1 */
1791                         if (option)
1792                                 set_opt(sb, BARRIER);
1793                         else
1794                                 clear_opt(sb, BARRIER);
1795                         break;
1796                 case Opt_ignore:
1797                         break;
1798                 case Opt_resize:
1799                         if (!is_remount) {
1800                                 ext4_msg(sb, KERN_ERR,
1801                                         "resize option only available "
1802                                         "for remount");
1803                                 return 0;
1804                         }
1805                         if (match_int(&args[0], &option) != 0)
1806                                 return 0;
1807                         *n_blocks_count = option;
1808                         break;
1809                 case Opt_nobh:
1810                         ext4_msg(sb, KERN_WARNING,
1811                                  "Ignoring deprecated nobh option");
1812                         break;
1813                 case Opt_bh:
1814                         ext4_msg(sb, KERN_WARNING,
1815                                  "Ignoring deprecated bh option");
1816                         break;
1817                 case Opt_i_version:
1818                         set_opt(sb, I_VERSION);
1819                         sb->s_flags |= MS_I_VERSION;
1820                         break;
1821                 case Opt_nodelalloc:
1822                         clear_opt(sb, DELALLOC);
1823                         clear_opt2(sb, EXPLICIT_DELALLOC);
1824                         break;
1825                 case Opt_mblk_io_submit:
1826                         set_opt(sb, MBLK_IO_SUBMIT);
1827                         break;
1828                 case Opt_nomblk_io_submit:
1829                         clear_opt(sb, MBLK_IO_SUBMIT);
1830                         break;
1831                 case Opt_stripe:
1832                         if (match_int(&args[0], &option))
1833                                 return 0;
1834                         if (option < 0)
1835                                 return 0;
1836                         sbi->s_stripe = option;
1837                         break;
1838                 case Opt_delalloc:
1839                         set_opt(sb, DELALLOC);
1840                         set_opt2(sb, EXPLICIT_DELALLOC);
1841                         break;
1842                 case Opt_block_validity:
1843                         set_opt(sb, BLOCK_VALIDITY);
1844                         break;
1845                 case Opt_noblock_validity:
1846                         clear_opt(sb, BLOCK_VALIDITY);
1847                         break;
1848                 case Opt_inode_readahead_blks:
1849                         if (match_int(&args[0], &option))
1850                                 return 0;
1851                         if (option < 0 || option > (1 << 30))
1852                                 return 0;
1853                         if (option && !is_power_of_2(option)) {
1854                                 ext4_msg(sb, KERN_ERR,
1855                                          "EXT4-fs: inode_readahead_blks"
1856                                          " must be a power of 2");
1857                                 return 0;
1858                         }
1859                         sbi->s_inode_readahead_blks = option;
1860                         break;
1861                 case Opt_journal_ioprio:
1862                         if (match_int(&args[0], &option))
1863                                 return 0;
1864                         if (option < 0 || option > 7)
1865                                 break;
1866                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1867                                                             option);
1868                         break;
1869                 case Opt_noauto_da_alloc:
1870                         set_opt(sb, NO_AUTO_DA_ALLOC);
1871                         break;
1872                 case Opt_auto_da_alloc:
1873                         if (args[0].from) {
1874                                 if (match_int(&args[0], &option))
1875                                         return 0;
1876                         } else
1877                                 option = 1;     /* No argument, default to 1 */
1878                         if (option)
1879                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1880                         else
1881                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1882                         break;
1883                 case Opt_discard:
1884                         set_opt(sb, DISCARD);
1885                         break;
1886                 case Opt_nodiscard:
1887                         clear_opt(sb, DISCARD);
1888                         break;
1889                 case Opt_dioread_nolock:
1890                         set_opt(sb, DIOREAD_NOLOCK);
1891                         break;
1892                 case Opt_dioread_lock:
1893                         clear_opt(sb, DIOREAD_NOLOCK);
1894                         break;
1895                 case Opt_init_inode_table:
1896                         set_opt(sb, INIT_INODE_TABLE);
1897                         if (args[0].from) {
1898                                 if (match_int(&args[0], &option))
1899                                         return 0;
1900                         } else
1901                                 option = EXT4_DEF_LI_WAIT_MULT;
1902                         if (option < 0)
1903                                 return 0;
1904                         sbi->s_li_wait_mult = option;
1905                         break;
1906                 case Opt_noinit_inode_table:
1907                         clear_opt(sb, INIT_INODE_TABLE);
1908                         break;
1909                 default:
1910                         ext4_msg(sb, KERN_ERR,
1911                                "Unrecognized mount option \"%s\" "
1912                                "or missing value", p);
1913                         return 0;
1914                 }
1915         }
1916 #ifdef CONFIG_QUOTA
1917         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1918                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1919                         clear_opt(sb, USRQUOTA);
1920
1921                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1922                         clear_opt(sb, GRPQUOTA);
1923
1924                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1925                         ext4_msg(sb, KERN_ERR, "old and new quota "
1926                                         "format mixing");
1927                         return 0;
1928                 }
1929
1930                 if (!sbi->s_jquota_fmt) {
1931                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1932                                         "not specified");
1933                         return 0;
1934                 }
1935         } else {
1936                 if (sbi->s_jquota_fmt) {
1937                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1938                                         "specified with no journaling "
1939                                         "enabled");
1940                         return 0;
1941                 }
1942         }
1943 #endif
1944         return 1;
1945 }
1946
1947 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1948                             int read_only)
1949 {
1950         struct ext4_sb_info *sbi = EXT4_SB(sb);
1951         int res = 0;
1952
1953         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1954                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1955                          "forcing read-only mode");
1956                 res = MS_RDONLY;
1957         }
1958         if (read_only)
1959                 goto done;
1960         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1961                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1962                          "running e2fsck is recommended");
1963         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1964                 ext4_msg(sb, KERN_WARNING,
1965                          "warning: mounting fs with errors, "
1966                          "running e2fsck is recommended");
1967         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1968                  le16_to_cpu(es->s_mnt_count) >=
1969                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1970                 ext4_msg(sb, KERN_WARNING,
1971                          "warning: maximal mount count reached, "
1972                          "running e2fsck is recommended");
1973         else if (le32_to_cpu(es->s_checkinterval) &&
1974                 (le32_to_cpu(es->s_lastcheck) +
1975                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1976                 ext4_msg(sb, KERN_WARNING,
1977                          "warning: checktime reached, "
1978                          "running e2fsck is recommended");
1979         if (!sbi->s_journal)
1980                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1981         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1982                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1983         le16_add_cpu(&es->s_mnt_count, 1);
1984         es->s_mtime = cpu_to_le32(get_seconds());
1985         ext4_update_dynamic_rev(sb);
1986         if (sbi->s_journal)
1987                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1988
1989         ext4_commit_super(sb, 1);
1990 done:
1991         if (test_opt(sb, DEBUG))
1992                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1993                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1994                         sb->s_blocksize,
1995                         sbi->s_groups_count,
1996                         EXT4_BLOCKS_PER_GROUP(sb),
1997                         EXT4_INODES_PER_GROUP(sb),
1998                         sbi->s_mount_opt, sbi->s_mount_opt2);
1999
2000         cleancache_init_fs(sb);
2001         return res;
2002 }
2003
2004 static int ext4_fill_flex_info(struct super_block *sb)
2005 {
2006         struct ext4_sb_info *sbi = EXT4_SB(sb);
2007         struct ext4_group_desc *gdp = NULL;
2008         ext4_group_t flex_group_count;
2009         ext4_group_t flex_group;
2010         int groups_per_flex = 0;
2011         size_t size;
2012         int i;
2013
2014         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2015         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
2016
2017         if (groups_per_flex < 2) {
2018                 sbi->s_log_groups_per_flex = 0;
2019                 return 1;
2020         }
2021
2022         /* We allocate both existing and potentially added groups */
2023         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2024                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2025                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2026         size = flex_group_count * sizeof(struct flex_groups);
2027         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2028         if (sbi->s_flex_groups == NULL) {
2029                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2030                          flex_group_count);
2031                 goto failed;
2032         }
2033
2034         for (i = 0; i < sbi->s_groups_count; i++) {
2035                 gdp = ext4_get_group_desc(sb, i, NULL);
2036
2037                 flex_group = ext4_flex_group(sbi, i);
2038                 atomic_add(ext4_free_inodes_count(sb, gdp),
2039                            &sbi->s_flex_groups[flex_group].free_inodes);
2040                 atomic_add(ext4_free_group_clusters(sb, gdp),
2041                            &sbi->s_flex_groups[flex_group].free_clusters);
2042                 atomic_add(ext4_used_dirs_count(sb, gdp),
2043                            &sbi->s_flex_groups[flex_group].used_dirs);
2044         }
2045
2046         return 1;
2047 failed:
2048         return 0;
2049 }
2050
2051 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2052                             struct ext4_group_desc *gdp)
2053 {
2054         __u16 crc = 0;
2055
2056         if (sbi->s_es->s_feature_ro_compat &
2057             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2058                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2059                 __le32 le_group = cpu_to_le32(block_group);
2060
2061                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2062                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2063                 crc = crc16(crc, (__u8 *)gdp, offset);
2064                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2065                 /* for checksum of struct ext4_group_desc do the rest...*/
2066                 if ((sbi->s_es->s_feature_incompat &
2067                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2068                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2069                         crc = crc16(crc, (__u8 *)gdp + offset,
2070                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2071                                         offset);
2072         }
2073
2074         return cpu_to_le16(crc);
2075 }
2076
2077 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2078                                 struct ext4_group_desc *gdp)
2079 {
2080         if ((sbi->s_es->s_feature_ro_compat &
2081              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2082             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2083                 return 0;
2084
2085         return 1;
2086 }
2087
2088 /* Called at mount-time, super-block is locked */
2089 static int ext4_check_descriptors(struct super_block *sb,
2090                                   ext4_group_t *first_not_zeroed)
2091 {
2092         struct ext4_sb_info *sbi = EXT4_SB(sb);
2093         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2094         ext4_fsblk_t last_block;
2095         ext4_fsblk_t block_bitmap;
2096         ext4_fsblk_t inode_bitmap;
2097         ext4_fsblk_t inode_table;
2098         int flexbg_flag = 0;
2099         ext4_group_t i, grp = sbi->s_groups_count;
2100
2101         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2102                 flexbg_flag = 1;
2103
2104         ext4_debug("Checking group descriptors");
2105
2106         for (i = 0; i < sbi->s_groups_count; i++) {
2107                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2108
2109                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2110                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2111                 else
2112                         last_block = first_block +
2113                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2114
2115                 if ((grp == sbi->s_groups_count) &&
2116                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2117                         grp = i;
2118
2119                 block_bitmap = ext4_block_bitmap(sb, gdp);
2120                 if (block_bitmap < first_block || block_bitmap > last_block) {
2121                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2122                                "Block bitmap for group %u not in group "
2123                                "(block %llu)!", i, block_bitmap);
2124                         return 0;
2125                 }
2126                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2127                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2128                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2129                                "Inode bitmap for group %u not in group "
2130                                "(block %llu)!", i, inode_bitmap);
2131                         return 0;
2132                 }
2133                 inode_table = ext4_inode_table(sb, gdp);
2134                 if (inode_table < first_block ||
2135                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2136                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2137                                "Inode table for group %u not in group "
2138                                "(block %llu)!", i, inode_table);
2139                         return 0;
2140                 }
2141                 ext4_lock_group(sb, i);
2142                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2143                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2144                                  "Checksum for group %u failed (%u!=%u)",
2145                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2146                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2147                         if (!(sb->s_flags & MS_RDONLY)) {
2148                                 ext4_unlock_group(sb, i);
2149                                 return 0;
2150                         }
2151                 }
2152                 ext4_unlock_group(sb, i);
2153                 if (!flexbg_flag)
2154                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2155         }
2156         if (NULL != first_not_zeroed)
2157                 *first_not_zeroed = grp;
2158
2159         ext4_free_blocks_count_set(sbi->s_es,
2160                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2161         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2162         return 1;
2163 }
2164
2165 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2166  * the superblock) which were deleted from all directories, but held open by
2167  * a process at the time of a crash.  We walk the list and try to delete these
2168  * inodes at recovery time (only with a read-write filesystem).
2169  *
2170  * In order to keep the orphan inode chain consistent during traversal (in
2171  * case of crash during recovery), we link each inode into the superblock
2172  * orphan list_head and handle it the same way as an inode deletion during
2173  * normal operation (which journals the operations for us).
2174  *
2175  * We only do an iget() and an iput() on each inode, which is very safe if we
2176  * accidentally point at an in-use or already deleted inode.  The worst that
2177  * can happen in this case is that we get a "bit already cleared" message from
2178  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2179  * e2fsck was run on this filesystem, and it must have already done the orphan
2180  * inode cleanup for us, so we can safely abort without any further action.
2181  */
2182 static void ext4_orphan_cleanup(struct super_block *sb,
2183                                 struct ext4_super_block *es)
2184 {
2185         unsigned int s_flags = sb->s_flags;
2186         int nr_orphans = 0, nr_truncates = 0;
2187 #ifdef CONFIG_QUOTA
2188         int i;
2189 #endif
2190         if (!es->s_last_orphan) {
2191                 jbd_debug(4, "no orphan inodes to clean up\n");
2192                 return;
2193         }
2194
2195         if (bdev_read_only(sb->s_bdev)) {
2196                 ext4_msg(sb, KERN_ERR, "write access "
2197                         "unavailable, skipping orphan cleanup");
2198                 return;
2199         }
2200
2201         /* Check if feature set would not allow a r/w mount */
2202         if (!ext4_feature_set_ok(sb, 0)) {
2203                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2204                          "unknown ROCOMPAT features");
2205                 return;
2206         }
2207
2208         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2209                 if (es->s_last_orphan)
2210                         jbd_debug(1, "Errors on filesystem, "
2211                                   "clearing orphan list.\n");
2212                 es->s_last_orphan = 0;
2213                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2214                 return;
2215         }
2216
2217         if (s_flags & MS_RDONLY) {
2218                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2219                 sb->s_flags &= ~MS_RDONLY;
2220         }
2221 #ifdef CONFIG_QUOTA
2222         /* Needed for iput() to work correctly and not trash data */
2223         sb->s_flags |= MS_ACTIVE;
2224         /* Turn on quotas so that they are updated correctly */
2225         for (i = 0; i < MAXQUOTAS; i++) {
2226                 if (EXT4_SB(sb)->s_qf_names[i]) {
2227                         int ret = ext4_quota_on_mount(sb, i);
2228                         if (ret < 0)
2229                                 ext4_msg(sb, KERN_ERR,
2230                                         "Cannot turn on journaled "
2231                                         "quota: error %d", ret);
2232                 }
2233         }
2234 #endif
2235
2236         while (es->s_last_orphan) {
2237                 struct inode *inode;
2238
2239                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2240                 if (IS_ERR(inode)) {
2241                         es->s_last_orphan = 0;
2242                         break;
2243                 }
2244
2245                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2246                 dquot_initialize(inode);
2247                 if (inode->i_nlink) {
2248                         ext4_msg(sb, KERN_DEBUG,
2249                                 "%s: truncating inode %lu to %lld bytes",
2250                                 __func__, inode->i_ino, inode->i_size);
2251                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2252                                   inode->i_ino, inode->i_size);
2253                         ext4_truncate(inode);
2254                         nr_truncates++;
2255                 } else {
2256                         ext4_msg(sb, KERN_DEBUG,
2257                                 "%s: deleting unreferenced inode %lu",
2258                                 __func__, inode->i_ino);
2259                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2260                                   inode->i_ino);
2261                         nr_orphans++;
2262                 }
2263                 iput(inode);  /* The delete magic happens here! */
2264         }
2265
2266 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2267
2268         if (nr_orphans)
2269                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2270                        PLURAL(nr_orphans));
2271         if (nr_truncates)
2272                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2273                        PLURAL(nr_truncates));
2274 #ifdef CONFIG_QUOTA
2275         /* Turn quotas off */
2276         for (i = 0; i < MAXQUOTAS; i++) {
2277                 if (sb_dqopt(sb)->files[i])
2278                         dquot_quota_off(sb, i);
2279         }
2280 #endif
2281         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2282 }
2283
2284 /*
2285  * Maximal extent format file size.
2286  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2287  * extent format containers, within a sector_t, and within i_blocks
2288  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2289  * so that won't be a limiting factor.
2290  *
2291  * However there is other limiting factor. We do store extents in the form
2292  * of starting block and length, hence the resulting length of the extent
2293  * covering maximum file size must fit into on-disk format containers as
2294  * well. Given that length is always by 1 unit bigger than max unit (because
2295  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2296  *
2297  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2298  */
2299 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2300 {
2301         loff_t res;
2302         loff_t upper_limit = MAX_LFS_FILESIZE;
2303
2304         /* small i_blocks in vfs inode? */
2305         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2306                 /*
2307                  * CONFIG_LBDAF is not enabled implies the inode
2308                  * i_block represent total blocks in 512 bytes
2309                  * 32 == size of vfs inode i_blocks * 8
2310                  */
2311                 upper_limit = (1LL << 32) - 1;
2312
2313                 /* total blocks in file system block size */
2314                 upper_limit >>= (blkbits - 9);
2315                 upper_limit <<= blkbits;
2316         }
2317
2318         /*
2319          * 32-bit extent-start container, ee_block. We lower the maxbytes
2320          * by one fs block, so ee_len can cover the extent of maximum file
2321          * size
2322          */
2323         res = (1LL << 32) - 1;
2324         res <<= blkbits;
2325
2326         /* Sanity check against vm- & vfs- imposed limits */
2327         if (res > upper_limit)
2328                 res = upper_limit;
2329
2330         return res;
2331 }
2332
2333 /*
2334  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2335  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2336  * We need to be 1 filesystem block less than the 2^48 sector limit.
2337  */
2338 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2339 {
2340         loff_t res = EXT4_NDIR_BLOCKS;
2341         int meta_blocks;
2342         loff_t upper_limit;
2343         /* This is calculated to be the largest file size for a dense, block
2344          * mapped file such that the file's total number of 512-byte sectors,
2345          * including data and all indirect blocks, does not exceed (2^48 - 1).
2346          *
2347          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2348          * number of 512-byte sectors of the file.
2349          */
2350
2351         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2352                 /*
2353                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2354                  * the inode i_block field represents total file blocks in
2355                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2356                  */
2357                 upper_limit = (1LL << 32) - 1;
2358
2359                 /* total blocks in file system block size */
2360                 upper_limit >>= (bits - 9);
2361
2362         } else {
2363                 /*
2364                  * We use 48 bit ext4_inode i_blocks
2365                  * With EXT4_HUGE_FILE_FL set the i_blocks
2366                  * represent total number of blocks in
2367                  * file system block size
2368                  */
2369                 upper_limit = (1LL << 48) - 1;
2370
2371         }
2372
2373         /* indirect blocks */
2374         meta_blocks = 1;
2375         /* double indirect blocks */
2376         meta_blocks += 1 + (1LL << (bits-2));
2377         /* tripple indirect blocks */
2378         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2379
2380         upper_limit -= meta_blocks;
2381         upper_limit <<= bits;
2382
2383         res += 1LL << (bits-2);
2384         res += 1LL << (2*(bits-2));
2385         res += 1LL << (3*(bits-2));
2386         res <<= bits;
2387         if (res > upper_limit)
2388                 res = upper_limit;
2389
2390         if (res > MAX_LFS_FILESIZE)
2391                 res = MAX_LFS_FILESIZE;
2392
2393         return res;
2394 }
2395
2396 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2397                                    ext4_fsblk_t logical_sb_block, int nr)
2398 {
2399         struct ext4_sb_info *sbi = EXT4_SB(sb);
2400         ext4_group_t bg, first_meta_bg;
2401         int has_super = 0;
2402
2403         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2404
2405         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2406             nr < first_meta_bg)
2407                 return logical_sb_block + nr + 1;
2408         bg = sbi->s_desc_per_block * nr;
2409         if (ext4_bg_has_super(sb, bg))
2410                 has_super = 1;
2411
2412         return (has_super + ext4_group_first_block_no(sb, bg));
2413 }
2414
2415 /**
2416  * ext4_get_stripe_size: Get the stripe size.
2417  * @sbi: In memory super block info
2418  *
2419  * If we have specified it via mount option, then
2420  * use the mount option value. If the value specified at mount time is
2421  * greater than the blocks per group use the super block value.
2422  * If the super block value is greater than blocks per group return 0.
2423  * Allocator needs it be less than blocks per group.
2424  *
2425  */
2426 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2427 {
2428         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2429         unsigned long stripe_width =
2430                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2431         int ret;
2432
2433         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2434                 ret = sbi->s_stripe;
2435         else if (stripe_width <= sbi->s_blocks_per_group)
2436                 ret = stripe_width;
2437         else if (stride <= sbi->s_blocks_per_group)
2438                 ret = stride;
2439         else
2440                 ret = 0;
2441
2442         /*
2443          * If the stripe width is 1, this makes no sense and
2444          * we set it to 0 to turn off stripe handling code.
2445          */
2446         if (ret <= 1)
2447                 ret = 0;
2448
2449         return ret;
2450 }
2451
2452 /* sysfs supprt */
2453
2454 struct ext4_attr {
2455         struct attribute attr;
2456         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2457         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2458                          const char *, size_t);
2459         int offset;
2460 };
2461
2462 static int parse_strtoul(const char *buf,
2463                 unsigned long max, unsigned long *value)
2464 {
2465         char *endp;
2466
2467         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2468         endp = skip_spaces(endp);
2469         if (*endp || *value > max)
2470                 return -EINVAL;
2471
2472         return 0;
2473 }
2474
2475 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2476                                               struct ext4_sb_info *sbi,
2477                                               char *buf)
2478 {
2479         return snprintf(buf, PAGE_SIZE, "%llu\n",
2480                 (s64) EXT4_C2B(sbi,
2481                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2482 }
2483
2484 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2485                                          struct ext4_sb_info *sbi, char *buf)
2486 {
2487         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2488
2489         if (!sb->s_bdev->bd_part)
2490                 return snprintf(buf, PAGE_SIZE, "0\n");
2491         return snprintf(buf, PAGE_SIZE, "%lu\n",
2492                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2493                          sbi->s_sectors_written_start) >> 1);
2494 }
2495
2496 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2497                                           struct ext4_sb_info *sbi, char *buf)
2498 {
2499         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2500
2501         if (!sb->s_bdev->bd_part)
2502                 return snprintf(buf, PAGE_SIZE, "0\n");
2503         return snprintf(buf, PAGE_SIZE, "%llu\n",
2504                         (unsigned long long)(sbi->s_kbytes_written +
2505                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2506                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2507 }
2508
2509 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2510                                       struct ext4_sb_info *sbi, char *buf)
2511 {
2512         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2513 }
2514
2515 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2516                                         struct ext4_sb_info *sbi, char *buf)
2517 {
2518         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2519 }
2520
2521 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2522                                           struct ext4_sb_info *sbi,
2523                                           const char *buf, size_t count)
2524 {
2525         unsigned long t;
2526
2527         if (parse_strtoul(buf, 0x40000000, &t))
2528                 return -EINVAL;
2529
2530         if (t && !is_power_of_2(t))
2531                 return -EINVAL;
2532
2533         sbi->s_inode_readahead_blks = t;
2534         return count;
2535 }
2536
2537 static ssize_t sbi_ui_show(struct ext4_attr *a,
2538                            struct ext4_sb_info *sbi, char *buf)
2539 {
2540         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2541
2542         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2543 }
2544
2545 static ssize_t sbi_ui_store(struct ext4_attr *a,
2546                             struct ext4_sb_info *sbi,
2547                             const char *buf, size_t count)
2548 {
2549         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2550         unsigned long t;
2551
2552         if (parse_strtoul(buf, 0xffffffff, &t))
2553                 return -EINVAL;
2554         *ui = t;
2555         return count;
2556 }
2557
2558 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2559 static struct ext4_attr ext4_attr_##_name = {                   \
2560         .attr = {.name = __stringify(_name), .mode = _mode },   \
2561         .show   = _show,                                        \
2562         .store  = _store,                                       \
2563         .offset = offsetof(struct ext4_sb_info, _elname),       \
2564 }
2565 #define EXT4_ATTR(name, mode, show, store) \
2566 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2567
2568 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2569 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2570 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2571 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2572         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2573 #define ATTR_LIST(name) &ext4_attr_##name.attr
2574
2575 EXT4_RO_ATTR(delayed_allocation_blocks);
2576 EXT4_RO_ATTR(session_write_kbytes);
2577 EXT4_RO_ATTR(lifetime_write_kbytes);
2578 EXT4_RO_ATTR(extent_cache_hits);
2579 EXT4_RO_ATTR(extent_cache_misses);
2580 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2581                  inode_readahead_blks_store, s_inode_readahead_blks);
2582 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2583 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2584 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2585 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2586 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2587 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2588 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2589 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2590
2591 static struct attribute *ext4_attrs[] = {
2592         ATTR_LIST(delayed_allocation_blocks),
2593         ATTR_LIST(session_write_kbytes),
2594         ATTR_LIST(lifetime_write_kbytes),
2595         ATTR_LIST(extent_cache_hits),
2596         ATTR_LIST(extent_cache_misses),
2597         ATTR_LIST(inode_readahead_blks),
2598         ATTR_LIST(inode_goal),
2599         ATTR_LIST(mb_stats),
2600         ATTR_LIST(mb_max_to_scan),
2601         ATTR_LIST(mb_min_to_scan),
2602         ATTR_LIST(mb_order2_req),
2603         ATTR_LIST(mb_stream_req),
2604         ATTR_LIST(mb_group_prealloc),
2605         ATTR_LIST(max_writeback_mb_bump),
2606         NULL,
2607 };
2608
2609 /* Features this copy of ext4 supports */
2610 EXT4_INFO_ATTR(lazy_itable_init);
2611 EXT4_INFO_ATTR(batched_discard);
2612
2613 static struct attribute *ext4_feat_attrs[] = {
2614         ATTR_LIST(lazy_itable_init),
2615         ATTR_LIST(batched_discard),
2616         NULL,
2617 };
2618
2619 static ssize_t ext4_attr_show(struct kobject *kobj,
2620                               struct attribute *attr, char *buf)
2621 {
2622         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2623                                                 s_kobj);
2624         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2625
2626         return a->show ? a->show(a, sbi, buf) : 0;
2627 }
2628
2629 static ssize_t ext4_attr_store(struct kobject *kobj,
2630                                struct attribute *attr,
2631                                const char *buf, size_t len)
2632 {
2633         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2634                                                 s_kobj);
2635         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2636
2637         return a->store ? a->store(a, sbi, buf, len) : 0;
2638 }
2639
2640 static void ext4_sb_release(struct kobject *kobj)
2641 {
2642         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2643                                                 s_kobj);
2644         complete(&sbi->s_kobj_unregister);
2645 }
2646
2647 static const struct sysfs_ops ext4_attr_ops = {
2648         .show   = ext4_attr_show,
2649         .store  = ext4_attr_store,
2650 };
2651
2652 static struct kobj_type ext4_ktype = {
2653         .default_attrs  = ext4_attrs,
2654         .sysfs_ops      = &ext4_attr_ops,
2655         .release        = ext4_sb_release,
2656 };
2657
2658 static void ext4_feat_release(struct kobject *kobj)
2659 {
2660         complete(&ext4_feat->f_kobj_unregister);
2661 }
2662
2663 static struct kobj_type ext4_feat_ktype = {
2664         .default_attrs  = ext4_feat_attrs,
2665         .sysfs_ops      = &ext4_attr_ops,
2666         .release        = ext4_feat_release,
2667 };
2668
2669 /*
2670  * Check whether this filesystem can be mounted based on
2671  * the features present and the RDONLY/RDWR mount requested.
2672  * Returns 1 if this filesystem can be mounted as requested,
2673  * 0 if it cannot be.
2674  */
2675 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2676 {
2677         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2678                 ext4_msg(sb, KERN_ERR,
2679                         "Couldn't mount because of "
2680                         "unsupported optional features (%x)",
2681                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2682                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2683                 return 0;
2684         }
2685
2686         if (readonly)
2687                 return 1;
2688
2689         /* Check that feature set is OK for a read-write mount */
2690         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2691                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2692                          "unsupported optional features (%x)",
2693                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2694                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2695                 return 0;
2696         }
2697         /*
2698          * Large file size enabled file system can only be mounted
2699          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2700          */
2701         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2702                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2703                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2704                                  "cannot be mounted RDWR without "
2705                                  "CONFIG_LBDAF");
2706                         return 0;
2707                 }
2708         }
2709         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2710             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2711                 ext4_msg(sb, KERN_ERR,
2712                          "Can't support bigalloc feature without "
2713                          "extents feature\n");
2714                 return 0;
2715         }
2716         return 1;
2717 }
2718
2719 /*
2720  * This function is called once a day if we have errors logged
2721  * on the file system
2722  */
2723 static void print_daily_error_info(unsigned long arg)
2724 {
2725         struct super_block *sb = (struct super_block *) arg;
2726         struct ext4_sb_info *sbi;
2727         struct ext4_super_block *es;
2728
2729         sbi = EXT4_SB(sb);
2730         es = sbi->s_es;
2731
2732         if (es->s_error_count)
2733                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2734                          le32_to_cpu(es->s_error_count));
2735         if (es->s_first_error_time) {
2736                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2737                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2738                        (int) sizeof(es->s_first_error_func),
2739                        es->s_first_error_func,
2740                        le32_to_cpu(es->s_first_error_line));
2741                 if (es->s_first_error_ino)
2742                         printk(": inode %u",
2743                                le32_to_cpu(es->s_first_error_ino));
2744                 if (es->s_first_error_block)
2745                         printk(": block %llu", (unsigned long long)
2746                                le64_to_cpu(es->s_first_error_block));
2747                 printk("\n");
2748         }
2749         if (es->s_last_error_time) {
2750                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2751                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2752                        (int) sizeof(es->s_last_error_func),
2753                        es->s_last_error_func,
2754                        le32_to_cpu(es->s_last_error_line));
2755                 if (es->s_last_error_ino)
2756                         printk(": inode %u",
2757                                le32_to_cpu(es->s_last_error_ino));
2758                 if (es->s_last_error_block)
2759                         printk(": block %llu", (unsigned long long)
2760                                le64_to_cpu(es->s_last_error_block));
2761                 printk("\n");
2762         }
2763         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2764 }
2765
2766 /* Find next suitable group and run ext4_init_inode_table */
2767 static int ext4_run_li_request(struct ext4_li_request *elr)
2768 {
2769         struct ext4_group_desc *gdp = NULL;
2770         ext4_group_t group, ngroups;
2771         struct super_block *sb;
2772         unsigned long timeout = 0;
2773         int ret = 0;
2774
2775         sb = elr->lr_super;
2776         ngroups = EXT4_SB(sb)->s_groups_count;
2777
2778         for (group = elr->lr_next_group; group < ngroups; group++) {
2779                 gdp = ext4_get_group_desc(sb, group, NULL);
2780                 if (!gdp) {
2781                         ret = 1;
2782                         break;
2783                 }
2784
2785                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2786                         break;
2787         }
2788
2789         if (group == ngroups)
2790                 ret = 1;
2791
2792         if (!ret) {
2793                 timeout = jiffies;
2794                 ret = ext4_init_inode_table(sb, group,
2795                                             elr->lr_timeout ? 0 : 1);
2796                 if (elr->lr_timeout == 0) {
2797                         timeout = (jiffies - timeout) *
2798                                   elr->lr_sbi->s_li_wait_mult;
2799                         elr->lr_timeout = timeout;
2800                 }
2801                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2802                 elr->lr_next_group = group + 1;
2803         }
2804
2805         return ret;
2806 }
2807
2808 /*
2809  * Remove lr_request from the list_request and free the
2810  * request structure. Should be called with li_list_mtx held
2811  */
2812 static void ext4_remove_li_request(struct ext4_li_request *elr)
2813 {
2814         struct ext4_sb_info *sbi;
2815
2816         if (!elr)
2817                 return;
2818
2819         sbi = elr->lr_sbi;
2820
2821         list_del(&elr->lr_request);
2822         sbi->s_li_request = NULL;
2823         kfree(elr);
2824 }
2825
2826 static void ext4_unregister_li_request(struct super_block *sb)
2827 {
2828         mutex_lock(&ext4_li_mtx);
2829         if (!ext4_li_info) {
2830                 mutex_unlock(&ext4_li_mtx);
2831                 return;
2832         }
2833
2834         mutex_lock(&ext4_li_info->li_list_mtx);
2835         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2836         mutex_unlock(&ext4_li_info->li_list_mtx);
2837         mutex_unlock(&ext4_li_mtx);
2838 }
2839
2840 static struct task_struct *ext4_lazyinit_task;
2841
2842 /*
2843  * This is the function where ext4lazyinit thread lives. It walks
2844  * through the request list searching for next scheduled filesystem.
2845  * When such a fs is found, run the lazy initialization request
2846  * (ext4_rn_li_request) and keep track of the time spend in this
2847  * function. Based on that time we compute next schedule time of
2848  * the request. When walking through the list is complete, compute
2849  * next waking time and put itself into sleep.
2850  */
2851 static int ext4_lazyinit_thread(void *arg)
2852 {
2853         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2854         struct list_head *pos, *n;
2855         struct ext4_li_request *elr;
2856         unsigned long next_wakeup, cur;
2857
2858         BUG_ON(NULL == eli);
2859
2860 cont_thread:
2861         while (true) {
2862                 next_wakeup = MAX_JIFFY_OFFSET;
2863
2864                 mutex_lock(&eli->li_list_mtx);
2865                 if (list_empty(&eli->li_request_list)) {
2866                         mutex_unlock(&eli->li_list_mtx);
2867                         goto exit_thread;
2868                 }
2869
2870                 list_for_each_safe(pos, n, &eli->li_request_list) {
2871                         elr = list_entry(pos, struct ext4_li_request,
2872                                          lr_request);
2873
2874                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2875                                 if (ext4_run_li_request(elr) != 0) {
2876                                         /* error, remove the lazy_init job */
2877                                         ext4_remove_li_request(elr);
2878                                         continue;
2879                                 }
2880                         }
2881
2882                         if (time_before(elr->lr_next_sched, next_wakeup))
2883                                 next_wakeup = elr->lr_next_sched;
2884                 }
2885                 mutex_unlock(&eli->li_list_mtx);
2886
2887                 if (freezing(current))
2888                         refrigerator();
2889
2890                 cur = jiffies;
2891                 if ((time_after_eq(cur, next_wakeup)) ||
2892                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2893                         cond_resched();
2894                         continue;
2895                 }
2896
2897                 schedule_timeout_interruptible(next_wakeup - cur);
2898
2899                 if (kthread_should_stop()) {
2900                         ext4_clear_request_list();
2901                         goto exit_thread;
2902                 }
2903         }
2904
2905 exit_thread:
2906         /*
2907          * It looks like the request list is empty, but we need
2908          * to check it under the li_list_mtx lock, to prevent any
2909          * additions into it, and of course we should lock ext4_li_mtx
2910          * to atomically free the list and ext4_li_info, because at
2911          * this point another ext4 filesystem could be registering
2912          * new one.
2913          */
2914         mutex_lock(&ext4_li_mtx);
2915         mutex_lock(&eli->li_list_mtx);
2916         if (!list_empty(&eli->li_request_list)) {
2917                 mutex_unlock(&eli->li_list_mtx);
2918                 mutex_unlock(&ext4_li_mtx);
2919                 goto cont_thread;
2920         }
2921         mutex_unlock(&eli->li_list_mtx);
2922         kfree(ext4_li_info);
2923         ext4_li_info = NULL;
2924         mutex_unlock(&ext4_li_mtx);
2925
2926         return 0;
2927 }
2928
2929 static void ext4_clear_request_list(void)
2930 {
2931         struct list_head *pos, *n;
2932         struct ext4_li_request *elr;
2933
2934         mutex_lock(&ext4_li_info->li_list_mtx);
2935         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2936                 elr = list_entry(pos, struct ext4_li_request,
2937                                  lr_request);
2938                 ext4_remove_li_request(elr);
2939         }
2940         mutex_unlock(&ext4_li_info->li_list_mtx);
2941 }
2942
2943 static int ext4_run_lazyinit_thread(void)
2944 {
2945         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2946                                          ext4_li_info, "ext4lazyinit");
2947         if (IS_ERR(ext4_lazyinit_task)) {
2948                 int err = PTR_ERR(ext4_lazyinit_task);
2949                 ext4_clear_request_list();
2950                 kfree(ext4_li_info);
2951                 ext4_li_info = NULL;
2952                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2953                                  "initialization thread\n",
2954                                  err);
2955                 return err;
2956         }
2957         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2958         return 0;
2959 }
2960
2961 /*
2962  * Check whether it make sense to run itable init. thread or not.
2963  * If there is at least one uninitialized inode table, return
2964  * corresponding group number, else the loop goes through all
2965  * groups and return total number of groups.
2966  */
2967 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2968 {
2969         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2970         struct ext4_group_desc *gdp = NULL;
2971
2972         for (group = 0; group < ngroups; group++) {
2973                 gdp = ext4_get_group_desc(sb, group, NULL);
2974                 if (!gdp)
2975                         continue;
2976
2977                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2978                         break;
2979         }
2980
2981         return group;
2982 }
2983
2984 static int ext4_li_info_new(void)
2985 {
2986         struct ext4_lazy_init *eli = NULL;
2987
2988         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2989         if (!eli)
2990                 return -ENOMEM;
2991
2992         INIT_LIST_HEAD(&eli->li_request_list);
2993         mutex_init(&eli->li_list_mtx);
2994
2995         eli->li_state |= EXT4_LAZYINIT_QUIT;
2996
2997         ext4_li_info = eli;
2998
2999         return 0;
3000 }
3001
3002 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3003                                             ext4_group_t start)
3004 {
3005         struct ext4_sb_info *sbi = EXT4_SB(sb);
3006         struct ext4_li_request *elr;
3007         unsigned long rnd;
3008
3009         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3010         if (!elr)
3011                 return NULL;
3012
3013         elr->lr_super = sb;
3014         elr->lr_sbi = sbi;
3015         elr->lr_next_group = start;
3016
3017         /*
3018          * Randomize first schedule time of the request to
3019          * spread the inode table initialization requests
3020          * better.
3021          */
3022         get_random_bytes(&rnd, sizeof(rnd));
3023         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3024                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3025
3026         return elr;
3027 }
3028
3029 static int ext4_register_li_request(struct super_block *sb,
3030                                     ext4_group_t first_not_zeroed)
3031 {
3032         struct ext4_sb_info *sbi = EXT4_SB(sb);
3033         struct ext4_li_request *elr;
3034         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3035         int ret = 0;
3036
3037         if (sbi->s_li_request != NULL) {
3038                 /*
3039                  * Reset timeout so it can be computed again, because
3040                  * s_li_wait_mult might have changed.
3041                  */
3042                 sbi->s_li_request->lr_timeout = 0;
3043                 return 0;
3044         }
3045
3046         if (first_not_zeroed == ngroups ||
3047             (sb->s_flags & MS_RDONLY) ||
3048             !test_opt(sb, INIT_INODE_TABLE))
3049                 return 0;
3050
3051         elr = ext4_li_request_new(sb, first_not_zeroed);
3052         if (!elr)
3053                 return -ENOMEM;
3054
3055         mutex_lock(&ext4_li_mtx);
3056
3057         if (NULL == ext4_li_info) {
3058                 ret = ext4_li_info_new();
3059                 if (ret)
3060                         goto out;
3061         }
3062
3063         mutex_lock(&ext4_li_info->li_list_mtx);
3064         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3065         mutex_unlock(&ext4_li_info->li_list_mtx);
3066
3067         sbi->s_li_request = elr;
3068         /*
3069          * set elr to NULL here since it has been inserted to
3070          * the request_list and the removal and free of it is
3071          * handled by ext4_clear_request_list from now on.
3072          */
3073         elr = NULL;
3074
3075         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3076                 ret = ext4_run_lazyinit_thread();
3077                 if (ret)
3078                         goto out;
3079         }
3080 out:
3081         mutex_unlock(&ext4_li_mtx);
3082         if (ret)
3083                 kfree(elr);
3084         return ret;
3085 }
3086
3087 /*
3088  * We do not need to lock anything since this is called on
3089  * module unload.
3090  */
3091 static void ext4_destroy_lazyinit_thread(void)
3092 {
3093         /*
3094          * If thread exited earlier
3095          * there's nothing to be done.
3096          */
3097         if (!ext4_li_info || !ext4_lazyinit_task)
3098                 return;
3099
3100         kthread_stop(ext4_lazyinit_task);
3101 }
3102
3103 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3104 {
3105         char *orig_data = kstrdup(data, GFP_KERNEL);
3106         struct buffer_head *bh;
3107         struct ext4_super_block *es = NULL;
3108         struct ext4_sb_info *sbi;
3109         ext4_fsblk_t block;
3110         ext4_fsblk_t sb_block = get_sb_block(&data);
3111         ext4_fsblk_t logical_sb_block;
3112         unsigned long offset = 0;
3113         unsigned long journal_devnum = 0;
3114         unsigned long def_mount_opts;
3115         struct inode *root;
3116         char *cp;
3117         const char *descr;
3118         int ret = -ENOMEM;
3119         int blocksize, clustersize;
3120         unsigned int db_count;
3121         unsigned int i;
3122         int needs_recovery, has_huge_files, has_bigalloc;
3123         __u64 blocks_count;
3124         int err;
3125         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3126         ext4_group_t first_not_zeroed;
3127
3128         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3129         if (!sbi)
3130                 goto out_free_orig;
3131
3132         sbi->s_blockgroup_lock =
3133                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3134         if (!sbi->s_blockgroup_lock) {
3135                 kfree(sbi);
3136                 goto out_free_orig;
3137         }
3138         sb->s_fs_info = sbi;
3139         sbi->s_mount_opt = 0;
3140         sbi->s_resuid = EXT4_DEF_RESUID;
3141         sbi->s_resgid = EXT4_DEF_RESGID;
3142         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3143         sbi->s_sb_block = sb_block;
3144         if (sb->s_bdev->bd_part)
3145                 sbi->s_sectors_written_start =
3146                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3147
3148         /* Cleanup superblock name */
3149         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3150                 *cp = '!';
3151
3152         ret = -EINVAL;
3153         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3154         if (!blocksize) {
3155                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3156                 goto out_fail;
3157         }
3158
3159         /*
3160          * The ext4 superblock will not be buffer aligned for other than 1kB
3161          * block sizes.  We need to calculate the offset from buffer start.
3162          */
3163         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3164                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3165                 offset = do_div(logical_sb_block, blocksize);
3166         } else {
3167                 logical_sb_block = sb_block;
3168         }
3169
3170         if (!(bh = sb_bread(sb, logical_sb_block))) {
3171                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3172                 goto out_fail;
3173         }
3174         /*
3175          * Note: s_es must be initialized as soon as possible because
3176          *       some ext4 macro-instructions depend on its value
3177          */
3178         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3179         sbi->s_es = es;
3180         sb->s_magic = le16_to_cpu(es->s_magic);
3181         if (sb->s_magic != EXT4_SUPER_MAGIC)
3182                 goto cantfind_ext4;
3183         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3184
3185         /* Set defaults before we parse the mount options */
3186         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3187         set_opt(sb, INIT_INODE_TABLE);
3188         if (def_mount_opts & EXT4_DEFM_DEBUG)
3189                 set_opt(sb, DEBUG);
3190         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3191                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3192                         "2.6.38");
3193                 set_opt(sb, GRPID);
3194         }
3195         if (def_mount_opts & EXT4_DEFM_UID16)
3196                 set_opt(sb, NO_UID32);
3197         /* xattr user namespace & acls are now defaulted on */
3198 #ifdef CONFIG_EXT4_FS_XATTR
3199         set_opt(sb, XATTR_USER);
3200 #endif
3201 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3202         set_opt(sb, POSIX_ACL);
3203 #endif
3204         set_opt(sb, MBLK_IO_SUBMIT);
3205         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3206                 set_opt(sb, JOURNAL_DATA);
3207         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3208                 set_opt(sb, ORDERED_DATA);
3209         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3210                 set_opt(sb, WRITEBACK_DATA);
3211
3212         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3213                 set_opt(sb, ERRORS_PANIC);
3214         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3215                 set_opt(sb, ERRORS_CONT);
3216         else
3217                 set_opt(sb, ERRORS_RO);
3218         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3219                 set_opt(sb, BLOCK_VALIDITY);
3220         if (def_mount_opts & EXT4_DEFM_DISCARD)
3221                 set_opt(sb, DISCARD);
3222
3223         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3224         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3225         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3226         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3227         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3228
3229         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3230                 set_opt(sb, BARRIER);
3231
3232         /*
3233          * enable delayed allocation by default
3234          * Use -o nodelalloc to turn it off
3235          */
3236         if (!IS_EXT3_SB(sb) &&
3237             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3238                 set_opt(sb, DELALLOC);
3239
3240         /*
3241          * set default s_li_wait_mult for lazyinit, for the case there is
3242          * no mount option specified.
3243          */
3244         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3245
3246         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3247                            &journal_devnum, &journal_ioprio, NULL, 0)) {
3248                 ext4_msg(sb, KERN_WARNING,
3249                          "failed to parse options in superblock: %s",
3250                          sbi->s_es->s_mount_opts);
3251         }
3252         if (!parse_options((char *) data, sb, &journal_devnum,
3253                            &journal_ioprio, NULL, 0))
3254                 goto failed_mount;
3255
3256         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3257                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3258                             "with data=journal disables delayed "
3259                             "allocation and O_DIRECT support!\n");
3260                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3261                         ext4_msg(sb, KERN_ERR, "can't mount with "
3262                                  "both data=journal and delalloc");
3263                         goto failed_mount;
3264                 }
3265                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3266                         ext4_msg(sb, KERN_ERR, "can't mount with "
3267                                  "both data=journal and delalloc");
3268                         goto failed_mount;
3269                 }
3270                 if (test_opt(sb, DELALLOC))
3271                         clear_opt(sb, DELALLOC);
3272         }
3273
3274         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3275         if (test_opt(sb, DIOREAD_NOLOCK)) {
3276                 if (blocksize < PAGE_SIZE) {
3277                         ext4_msg(sb, KERN_ERR, "can't mount with "
3278                                  "dioread_nolock if block size != PAGE_SIZE");
3279                         goto failed_mount;
3280                 }
3281         }
3282
3283         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3284                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3285
3286         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3287             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3288              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3289              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3290                 ext4_msg(sb, KERN_WARNING,
3291                        "feature flags set on rev 0 fs, "
3292                        "running e2fsck is recommended");
3293
3294         if (IS_EXT2_SB(sb)) {
3295                 if (ext2_feature_set_ok(sb))
3296                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3297                                  "using the ext4 subsystem");
3298                 else {
3299                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3300                                  "to feature incompatibilities");
3301                         goto failed_mount;
3302                 }
3303         }
3304
3305         if (IS_EXT3_SB(sb)) {
3306                 if (ext3_feature_set_ok(sb))
3307                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3308                                  "using the ext4 subsystem");
3309                 else {
3310                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3311                                  "to feature incompatibilities");
3312                         goto failed_mount;
3313                 }
3314         }
3315
3316         /*
3317          * Check feature flags regardless of the revision level, since we
3318          * previously didn't change the revision level when setting the flags,
3319          * so there is a chance incompat flags are set on a rev 0 filesystem.
3320          */
3321         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3322                 goto failed_mount;
3323
3324         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3325             blocksize > EXT4_MAX_BLOCK_SIZE) {
3326                 ext4_msg(sb, KERN_ERR,
3327                        "Unsupported filesystem blocksize %d", blocksize);
3328                 goto failed_mount;
3329         }
3330
3331         if (sb->s_blocksize != blocksize) {
3332                 /* Validate the filesystem blocksize */
3333                 if (!sb_set_blocksize(sb, blocksize)) {
3334                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3335                                         blocksize);
3336                         goto failed_mount;
3337                 }
3338
3339                 brelse(bh);
3340                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3341                 offset = do_div(logical_sb_block, blocksize);
3342                 bh = sb_bread(sb, logical_sb_block);
3343                 if (!bh) {
3344                         ext4_msg(sb, KERN_ERR,
3345                                "Can't read superblock on 2nd try");
3346                         goto failed_mount;
3347                 }
3348                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3349                 sbi->s_es = es;
3350                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3351                         ext4_msg(sb, KERN_ERR,
3352                                "Magic mismatch, very weird!");
3353                         goto failed_mount;
3354                 }
3355         }
3356
3357         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3358                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3359         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3360                                                       has_huge_files);
3361         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3362
3363         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3364                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3365                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3366         } else {
3367                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3368                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3369                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3370                     (!is_power_of_2(sbi->s_inode_size)) ||
3371                     (sbi->s_inode_size > blocksize)) {
3372                         ext4_msg(sb, KERN_ERR,
3373                                "unsupported inode size: %d",
3374                                sbi->s_inode_size);
3375                         goto failed_mount;
3376                 }
3377                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3378                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3379         }
3380
3381         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3382         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3383                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3384                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3385                     !is_power_of_2(sbi->s_desc_size)) {
3386                         ext4_msg(sb, KERN_ERR,
3387                                "unsupported descriptor size %lu",
3388                                sbi->s_desc_size);
3389                         goto failed_mount;
3390                 }
3391         } else
3392                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3393
3394         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3395         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3396         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3397                 goto cantfind_ext4;
3398
3399         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3400         if (sbi->s_inodes_per_block == 0)
3401                 goto cantfind_ext4;
3402         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3403                                         sbi->s_inodes_per_block;
3404         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3405         sbi->s_sbh = bh;
3406         sbi->s_mount_state = le16_to_cpu(es->s_state);
3407         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3408         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3409
3410         for (i = 0; i < 4; i++)
3411                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3412         sbi->s_def_hash_version = es->s_def_hash_version;
3413         i = le32_to_cpu(es->s_flags);
3414         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3415                 sbi->s_hash_unsigned = 3;
3416         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3417 #ifdef __CHAR_UNSIGNED__
3418                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3419                 sbi->s_hash_unsigned = 3;
3420 #else
3421                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3422 #endif
3423                 sb->s_dirt = 1;
3424         }
3425
3426         /* Handle clustersize */
3427         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3428         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3429                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3430         if (has_bigalloc) {
3431                 if (clustersize < blocksize) {
3432                         ext4_msg(sb, KERN_ERR,
3433                                  "cluster size (%d) smaller than "
3434                                  "block size (%d)", clustersize, blocksize);
3435                         goto failed_mount;
3436                 }
3437                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3438                         le32_to_cpu(es->s_log_block_size);
3439                 sbi->s_clusters_per_group =
3440                         le32_to_cpu(es->s_clusters_per_group);
3441                 if (sbi->s_clusters_per_group > blocksize * 8) {
3442                         ext4_msg(sb, KERN_ERR,
3443                                  "#clusters per group too big: %lu",
3444                                  sbi->s_clusters_per_group);
3445                         goto failed_mount;
3446                 }
3447                 if (sbi->s_blocks_per_group !=
3448                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3449                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3450                                  "clusters per group (%lu) inconsistent",
3451                                  sbi->s_blocks_per_group,
3452                                  sbi->s_clusters_per_group);
3453                         goto failed_mount;
3454                 }
3455         } else {
3456                 if (clustersize != blocksize) {
3457                         ext4_warning(sb, "fragment/cluster size (%d) != "
3458                                      "block size (%d)", clustersize,
3459                                      blocksize);
3460                         clustersize = blocksize;
3461                 }
3462                 if (sbi->s_blocks_per_group > blocksize * 8) {
3463                         ext4_msg(sb, KERN_ERR,
3464                                  "#blocks per group too big: %lu",
3465                                  sbi->s_blocks_per_group);
3466                         goto failed_mount;
3467                 }
3468                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3469                 sbi->s_cluster_bits = 0;
3470         }
3471         sbi->s_cluster_ratio = clustersize / blocksize;
3472
3473         if (sbi->s_inodes_per_group > blocksize * 8) {
3474                 ext4_msg(sb, KERN_ERR,
3475                        "#inodes per group too big: %lu",
3476                        sbi->s_inodes_per_group);
3477                 goto failed_mount;
3478         }
3479
3480         /*
3481          * Test whether we have more sectors than will fit in sector_t,
3482          * and whether the max offset is addressable by the page cache.
3483          */
3484         err = generic_check_addressable(sb->s_blocksize_bits,
3485                                         ext4_blocks_count(es));
3486         if (err) {
3487                 ext4_msg(sb, KERN_ERR, "filesystem"
3488                          " too large to mount safely on this system");
3489                 if (sizeof(sector_t) < 8)
3490                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3491                 ret = err;
3492                 goto failed_mount;
3493         }
3494
3495         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3496                 goto cantfind_ext4;
3497
3498         /* check blocks count against device size */
3499         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3500         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3501                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3502                        "exceeds size of device (%llu blocks)",
3503                        ext4_blocks_count(es), blocks_count);
3504                 goto failed_mount;
3505         }
3506
3507         /*
3508          * It makes no sense for the first data block to be beyond the end
3509          * of the filesystem.
3510          */
3511         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3512                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data"
3513                          "block %u is beyond end of filesystem (%llu)",
3514                          le32_to_cpu(es->s_first_data_block),
3515                          ext4_blocks_count(es));
3516                 goto failed_mount;
3517         }
3518         blocks_count = (ext4_blocks_count(es) -
3519                         le32_to_cpu(es->s_first_data_block) +
3520                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3521         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3522         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3523                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3524                        "(block count %llu, first data block %u, "
3525                        "blocks per group %lu)", sbi->s_groups_count,
3526                        ext4_blocks_count(es),
3527                        le32_to_cpu(es->s_first_data_block),
3528                        EXT4_BLOCKS_PER_GROUP(sb));
3529                 goto failed_mount;
3530         }
3531         sbi->s_groups_count = blocks_count;
3532         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3533                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3534         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3535                    EXT4_DESC_PER_BLOCK(sb);
3536         sbi->s_group_desc = ext4_kvmalloc(db_count *
3537                                           sizeof(struct buffer_head *),
3538                                           GFP_KERNEL);
3539         if (sbi->s_group_desc == NULL) {
3540                 ext4_msg(sb, KERN_ERR, "not enough memory");
3541                 goto failed_mount;
3542         }
3543
3544         if (ext4_proc_root)
3545                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3546
3547         bgl_lock_init(sbi->s_blockgroup_lock);
3548
3549         for (i = 0; i < db_count; i++) {
3550                 block = descriptor_loc(sb, logical_sb_block, i);
3551                 sbi->s_group_desc[i] = sb_bread(sb, block);
3552                 if (!sbi->s_group_desc[i]) {
3553                         ext4_msg(sb, KERN_ERR,
3554                                "can't read group descriptor %d", i);
3555                         db_count = i;
3556                         goto failed_mount2;
3557                 }
3558         }
3559         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3560                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3561                 goto failed_mount2;
3562         }
3563         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3564                 if (!ext4_fill_flex_info(sb)) {
3565                         ext4_msg(sb, KERN_ERR,
3566                                "unable to initialize "
3567                                "flex_bg meta info!");
3568                         goto failed_mount2;
3569                 }
3570
3571         sbi->s_gdb_count = db_count;
3572         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3573         spin_lock_init(&sbi->s_next_gen_lock);
3574
3575         init_timer(&sbi->s_err_report);
3576         sbi->s_err_report.function = print_daily_error_info;
3577         sbi->s_err_report.data = (unsigned long) sb;
3578
3579         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3580                         ext4_count_free_clusters(sb));
3581         if (!err) {
3582                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3583                                 ext4_count_free_inodes(sb));
3584         }
3585         if (!err) {
3586                 err = percpu_counter_init(&sbi->s_dirs_counter,
3587                                 ext4_count_dirs(sb));
3588         }
3589         if (!err) {
3590                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3591         }
3592         if (err) {
3593                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3594                 goto failed_mount3;
3595         }
3596
3597         sbi->s_stripe = ext4_get_stripe_size(sbi);
3598         sbi->s_max_writeback_mb_bump = 128;
3599
3600         /*
3601          * set up enough so that it can read an inode
3602          */
3603         if (!test_opt(sb, NOLOAD) &&
3604             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3605                 sb->s_op = &ext4_sops;
3606         else
3607                 sb->s_op = &ext4_nojournal_sops;
3608         sb->s_export_op = &ext4_export_ops;
3609         sb->s_xattr = ext4_xattr_handlers;
3610 #ifdef CONFIG_QUOTA
3611         sb->s_qcop = &ext4_qctl_operations;
3612         sb->dq_op = &ext4_quota_operations;
3613 #endif
3614         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3615
3616         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3617         mutex_init(&sbi->s_orphan_lock);
3618         sbi->s_resize_flags = 0;
3619
3620         sb->s_root = NULL;
3621
3622         needs_recovery = (es->s_last_orphan != 0 ||
3623                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3624                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3625
3626         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3627             !(sb->s_flags & MS_RDONLY))
3628                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3629                         goto failed_mount3;
3630
3631         /*
3632          * The first inode we look at is the journal inode.  Don't try
3633          * root first: it may be modified in the journal!
3634          */
3635         if (!test_opt(sb, NOLOAD) &&
3636             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3637                 if (ext4_load_journal(sb, es, journal_devnum))
3638                         goto failed_mount3;
3639         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3640               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3641                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3642                        "suppressed and not mounted read-only");
3643                 goto failed_mount_wq;
3644         } else {
3645                 clear_opt(sb, DATA_FLAGS);
3646                 sbi->s_journal = NULL;
3647                 needs_recovery = 0;
3648                 goto no_journal;
3649         }
3650
3651         if (ext4_blocks_count(es) > 0xffffffffULL &&
3652             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3653                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3654                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3655                 goto failed_mount_wq;
3656         }
3657
3658         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3659                 jbd2_journal_set_features(sbi->s_journal,
3660                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3661                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3662         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3663                 jbd2_journal_set_features(sbi->s_journal,
3664                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3665                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3666                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3667         } else {
3668                 jbd2_journal_clear_features(sbi->s_journal,
3669                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3670                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3671         }
3672
3673         /* We have now updated the journal if required, so we can
3674          * validate the data journaling mode. */
3675         switch (test_opt(sb, DATA_FLAGS)) {
3676         case 0:
3677                 /* No mode set, assume a default based on the journal
3678                  * capabilities: ORDERED_DATA if the journal can
3679                  * cope, else JOURNAL_DATA
3680                  */
3681                 if (jbd2_journal_check_available_features
3682                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3683                         set_opt(sb, ORDERED_DATA);
3684                 else
3685                         set_opt(sb, JOURNAL_DATA);
3686                 break;
3687
3688         case EXT4_MOUNT_ORDERED_DATA:
3689         case EXT4_MOUNT_WRITEBACK_DATA:
3690                 if (!jbd2_journal_check_available_features
3691                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3692                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3693                                "requested data journaling mode");
3694                         goto failed_mount_wq;
3695                 }
3696         default:
3697                 break;
3698         }
3699         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3700
3701         /*
3702          * The journal may have updated the bg summary counts, so we
3703          * need to update the global counters.
3704          */
3705         percpu_counter_set(&sbi->s_freeclusters_counter,
3706                            ext4_count_free_clusters(sb));
3707         percpu_counter_set(&sbi->s_freeinodes_counter,
3708                            ext4_count_free_inodes(sb));
3709         percpu_counter_set(&sbi->s_dirs_counter,
3710                            ext4_count_dirs(sb));
3711         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3712
3713 no_journal:
3714         /*
3715          * The maximum number of concurrent works can be high and
3716          * concurrency isn't really necessary.  Limit it to 1.
3717          */
3718         EXT4_SB(sb)->dio_unwritten_wq =
3719                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3720         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3721                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3722                 goto failed_mount_wq;
3723         }
3724
3725         /*
3726          * The jbd2_journal_load will have done any necessary log recovery,
3727          * so we can safely mount the rest of the filesystem now.
3728          */
3729
3730         root = ext4_iget(sb, EXT4_ROOT_INO);
3731         if (IS_ERR(root)) {
3732                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3733                 ret = PTR_ERR(root);
3734                 root = NULL;
3735                 goto failed_mount4;
3736         }
3737         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3738                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3739                 goto failed_mount4;
3740         }
3741         sb->s_root = d_alloc_root(root);
3742         if (!sb->s_root) {
3743                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3744                 ret = -ENOMEM;
3745                 goto failed_mount4;
3746         }
3747
3748         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3749
3750         /* determine the minimum size of new large inodes, if present */
3751         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3752                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3753                                                      EXT4_GOOD_OLD_INODE_SIZE;
3754                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3755                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3756                         if (sbi->s_want_extra_isize <
3757                             le16_to_cpu(es->s_want_extra_isize))
3758                                 sbi->s_want_extra_isize =
3759                                         le16_to_cpu(es->s_want_extra_isize);
3760                         if (sbi->s_want_extra_isize <
3761                             le16_to_cpu(es->s_min_extra_isize))
3762                                 sbi->s_want_extra_isize =
3763                                         le16_to_cpu(es->s_min_extra_isize);
3764                 }
3765         }
3766         /* Check if enough inode space is available */
3767         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3768                                                         sbi->s_inode_size) {
3769                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3770                                                        EXT4_GOOD_OLD_INODE_SIZE;
3771                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3772                          "available");
3773         }
3774
3775         err = ext4_setup_system_zone(sb);
3776         if (err) {
3777                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3778                          "zone (%d)", err);
3779                 goto failed_mount4;
3780         }
3781
3782         ext4_ext_init(sb);
3783         err = ext4_mb_init(sb, needs_recovery);
3784         if (err) {
3785                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3786                          err);
3787                 goto failed_mount5;
3788         }
3789
3790         err = ext4_register_li_request(sb, first_not_zeroed);
3791         if (err)
3792                 goto failed_mount6;
3793
3794         sbi->s_kobj.kset = ext4_kset;
3795         init_completion(&sbi->s_kobj_unregister);
3796         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3797                                    "%s", sb->s_id);
3798         if (err)
3799                 goto failed_mount7;
3800
3801         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3802         ext4_orphan_cleanup(sb, es);
3803         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3804         if (needs_recovery) {
3805                 ext4_msg(sb, KERN_INFO, "recovery complete");
3806                 ext4_mark_recovery_complete(sb, es);
3807         }
3808         if (EXT4_SB(sb)->s_journal) {
3809                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3810                         descr = " journalled data mode";
3811                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3812                         descr = " ordered data mode";
3813                 else
3814                         descr = " writeback data mode";
3815         } else
3816                 descr = "out journal";
3817
3818         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3819                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3820                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3821
3822         if (es->s_error_count)
3823                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3824
3825         kfree(orig_data);
3826         return 0;
3827
3828 cantfind_ext4:
3829         if (!silent)
3830                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3831         goto failed_mount;
3832
3833 failed_mount7:
3834         ext4_unregister_li_request(sb);