Loading...
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
57static struct proc_dir_entry *ext4_proc_root;
58static struct kset *ext4_kset;
59static struct ext4_lazy_init *ext4_li_info;
60static struct mutex ext4_li_mtx;
61static struct ext4_features *ext4_feat;
62
63static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
64 unsigned long journal_devnum);
65static int ext4_show_options(struct seq_file *seq, struct dentry *root);
66static int ext4_commit_super(struct super_block *sb, int sync);
67static void ext4_mark_recovery_complete(struct super_block *sb,
68 struct ext4_super_block *es);
69static void ext4_clear_journal_err(struct super_block *sb,
70 struct ext4_super_block *es);
71static int ext4_sync_fs(struct super_block *sb, int wait);
72static const char *ext4_decode_error(struct super_block *sb, int errno,
73 char nbuf[16]);
74static int ext4_remount(struct super_block *sb, int *flags, char *data);
75static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
76static int ext4_unfreeze(struct super_block *sb);
77static void ext4_write_super(struct super_block *sb);
78static int ext4_freeze(struct super_block *sb);
79static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
80 const char *dev_name, void *data);
81static inline int ext2_feature_set_ok(struct super_block *sb);
82static inline int ext3_feature_set_ok(struct super_block *sb);
83static int ext4_feature_set_ok(struct super_block *sb, int readonly);
84static void ext4_destroy_lazyinit_thread(void);
85static void ext4_unregister_li_request(struct super_block *sb);
86static void ext4_clear_request_list(void);
87
88#if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
89static struct file_system_type ext2_fs_type = {
90 .owner = THIS_MODULE,
91 .name = "ext2",
92 .mount = ext4_mount,
93 .kill_sb = kill_block_super,
94 .fs_flags = FS_REQUIRES_DEV,
95};
96#define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
97#else
98#define IS_EXT2_SB(sb) (0)
99#endif
100
101
102#if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
103static struct file_system_type ext3_fs_type = {
104 .owner = THIS_MODULE,
105 .name = "ext3",
106 .mount = ext4_mount,
107 .kill_sb = kill_block_super,
108 .fs_flags = FS_REQUIRES_DEV,
109};
110#define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
111#else
112#define IS_EXT3_SB(sb) (0)
113#endif
114
115static int ext4_verify_csum_type(struct super_block *sb,
116 struct ext4_super_block *es)
117{
118 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
119 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
120 return 1;
121
122 return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
123}
124
125static __le32 ext4_superblock_csum(struct super_block *sb,
126 struct ext4_super_block *es)
127{
128 struct ext4_sb_info *sbi = EXT4_SB(sb);
129 int offset = offsetof(struct ext4_super_block, s_checksum);
130 __u32 csum;
131
132 csum = ext4_chksum(sbi, ~0, (char *)es, offset);
133
134 return cpu_to_le32(csum);
135}
136
137int ext4_superblock_csum_verify(struct super_block *sb,
138 struct ext4_super_block *es)
139{
140 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
141 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
142 return 1;
143
144 return es->s_checksum == ext4_superblock_csum(sb, es);
145}
146
147void ext4_superblock_csum_set(struct super_block *sb,
148 struct ext4_super_block *es)
149{
150 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
151 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
152 return;
153
154 es->s_checksum = ext4_superblock_csum(sb, es);
155}
156
157void *ext4_kvmalloc(size_t size, gfp_t flags)
158{
159 void *ret;
160
161 ret = kmalloc(size, flags);
162 if (!ret)
163 ret = __vmalloc(size, flags, PAGE_KERNEL);
164 return ret;
165}
166
167void *ext4_kvzalloc(size_t size, gfp_t flags)
168{
169 void *ret;
170
171 ret = kzalloc(size, flags);
172 if (!ret)
173 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
174 return ret;
175}
176
177void ext4_kvfree(void *ptr)
178{
179 if (is_vmalloc_addr(ptr))
180 vfree(ptr);
181 else
182 kfree(ptr);
183
184}
185
186ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
187 struct ext4_group_desc *bg)
188{
189 return le32_to_cpu(bg->bg_block_bitmap_lo) |
190 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
191 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
192}
193
194ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
195 struct ext4_group_desc *bg)
196{
197 return le32_to_cpu(bg->bg_inode_bitmap_lo) |
198 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
199 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
200}
201
202ext4_fsblk_t ext4_inode_table(struct super_block *sb,
203 struct ext4_group_desc *bg)
204{
205 return le32_to_cpu(bg->bg_inode_table_lo) |
206 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
207 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
208}
209
210__u32 ext4_free_group_clusters(struct super_block *sb,
211 struct ext4_group_desc *bg)
212{
213 return le16_to_cpu(bg->bg_free_blocks_count_lo) |
214 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
215 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
216}
217
218__u32 ext4_free_inodes_count(struct super_block *sb,
219 struct ext4_group_desc *bg)
220{
221 return le16_to_cpu(bg->bg_free_inodes_count_lo) |
222 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
223 (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
224}
225
226__u32 ext4_used_dirs_count(struct super_block *sb,
227 struct ext4_group_desc *bg)
228{
229 return le16_to_cpu(bg->bg_used_dirs_count_lo) |
230 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
231 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
232}
233
234__u32 ext4_itable_unused_count(struct super_block *sb,
235 struct ext4_group_desc *bg)
236{
237 return le16_to_cpu(bg->bg_itable_unused_lo) |
238 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
239 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
240}
241
242void ext4_block_bitmap_set(struct super_block *sb,
243 struct ext4_group_desc *bg, ext4_fsblk_t blk)
244{
245 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
246 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
247 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
248}
249
250void ext4_inode_bitmap_set(struct super_block *sb,
251 struct ext4_group_desc *bg, ext4_fsblk_t blk)
252{
253 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk);
254 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
255 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
256}
257
258void ext4_inode_table_set(struct super_block *sb,
259 struct ext4_group_desc *bg, ext4_fsblk_t blk)
260{
261 bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
262 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
263 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
264}
265
266void ext4_free_group_clusters_set(struct super_block *sb,
267 struct ext4_group_desc *bg, __u32 count)
268{
269 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
270 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
271 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
272}
273
274void ext4_free_inodes_set(struct super_block *sb,
275 struct ext4_group_desc *bg, __u32 count)
276{
277 bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
278 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
279 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
280}
281
282void ext4_used_dirs_set(struct super_block *sb,
283 struct ext4_group_desc *bg, __u32 count)
284{
285 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
286 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
287 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
288}
289
290void ext4_itable_unused_set(struct super_block *sb,
291 struct ext4_group_desc *bg, __u32 count)
292{
293 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
294 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
295 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
296}
297
298
299/* Just increment the non-pointer handle value */
300static handle_t *ext4_get_nojournal(void)
301{
302 handle_t *handle = current->journal_info;
303 unsigned long ref_cnt = (unsigned long)handle;
304
305 BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
306
307 ref_cnt++;
308 handle = (handle_t *)ref_cnt;
309
310 current->journal_info = handle;
311 return handle;
312}
313
314
315/* Decrement the non-pointer handle value */
316static void ext4_put_nojournal(handle_t *handle)
317{
318 unsigned long ref_cnt = (unsigned long)handle;
319
320 BUG_ON(ref_cnt == 0);
321
322 ref_cnt--;
323 handle = (handle_t *)ref_cnt;
324
325 current->journal_info = handle;
326}
327
328/*
329 * Wrappers for jbd2_journal_start/end.
330 *
331 * The only special thing we need to do here is to make sure that all
332 * journal_end calls result in the superblock being marked dirty, so
333 * that sync() will call the filesystem's write_super callback if
334 * appropriate.
335 *
336 * To avoid j_barrier hold in userspace when a user calls freeze(),
337 * ext4 prevents a new handle from being started by s_frozen, which
338 * is in an upper layer.
339 */
340handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
341{
342 journal_t *journal;
343 handle_t *handle;
344
345 trace_ext4_journal_start(sb, nblocks, _RET_IP_);
346 if (sb->s_flags & MS_RDONLY)
347 return ERR_PTR(-EROFS);
348
349 journal = EXT4_SB(sb)->s_journal;
350 handle = ext4_journal_current_handle();
351
352 /*
353 * If a handle has been started, it should be allowed to
354 * finish, otherwise deadlock could happen between freeze
355 * and others(e.g. truncate) due to the restart of the
356 * journal handle if the filesystem is forzen and active
357 * handles are not stopped.
358 */
359 if (!handle)
360 vfs_check_frozen(sb, SB_FREEZE_TRANS);
361
362 if (!journal)
363 return ext4_get_nojournal();
364 /*
365 * Special case here: if the journal has aborted behind our
366 * backs (eg. EIO in the commit thread), then we still need to
367 * take the FS itself readonly cleanly.
368 */
369 if (is_journal_aborted(journal)) {
370 ext4_abort(sb, "Detected aborted journal");
371 return ERR_PTR(-EROFS);
372 }
373 return jbd2_journal_start(journal, nblocks);
374}
375
376/*
377 * The only special thing we need to do here is to make sure that all
378 * jbd2_journal_stop calls result in the superblock being marked dirty, so
379 * that sync() will call the filesystem's write_super callback if
380 * appropriate.
381 */
382int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
383{
384 struct super_block *sb;
385 int err;
386 int rc;
387
388 if (!ext4_handle_valid(handle)) {
389 ext4_put_nojournal(handle);
390 return 0;
391 }
392 sb = handle->h_transaction->t_journal->j_private;
393 err = handle->h_err;
394 rc = jbd2_journal_stop(handle);
395
396 if (!err)
397 err = rc;
398 if (err)
399 __ext4_std_error(sb, where, line, err);
400 return err;
401}
402
403void ext4_journal_abort_handle(const char *caller, unsigned int line,
404 const char *err_fn, struct buffer_head *bh,
405 handle_t *handle, int err)
406{
407 char nbuf[16];
408 const char *errstr = ext4_decode_error(NULL, err, nbuf);
409
410 BUG_ON(!ext4_handle_valid(handle));
411
412 if (bh)
413 BUFFER_TRACE(bh, "abort");
414
415 if (!handle->h_err)
416 handle->h_err = err;
417
418 if (is_handle_aborted(handle))
419 return;
420
421 printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
422 caller, line, errstr, err_fn);
423
424 jbd2_journal_abort_handle(handle);
425}
426
427static void __save_error_info(struct super_block *sb, const char *func,
428 unsigned int line)
429{
430 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
431
432 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
433 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
434 es->s_last_error_time = cpu_to_le32(get_seconds());
435 strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
436 es->s_last_error_line = cpu_to_le32(line);
437 if (!es->s_first_error_time) {
438 es->s_first_error_time = es->s_last_error_time;
439 strncpy(es->s_first_error_func, func,
440 sizeof(es->s_first_error_func));
441 es->s_first_error_line = cpu_to_le32(line);
442 es->s_first_error_ino = es->s_last_error_ino;
443 es->s_first_error_block = es->s_last_error_block;
444 }
445 /*
446 * Start the daily error reporting function if it hasn't been
447 * started already
448 */
449 if (!es->s_error_count)
450 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
451 es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
452}
453
454static void save_error_info(struct super_block *sb, const char *func,
455 unsigned int line)
456{
457 __save_error_info(sb, func, line);
458 ext4_commit_super(sb, 1);
459}
460
461/*
462 * The del_gendisk() function uninitializes the disk-specific data
463 * structures, including the bdi structure, without telling anyone
464 * else. Once this happens, any attempt to call mark_buffer_dirty()
465 * (for example, by ext4_commit_super), will cause a kernel OOPS.
466 * This is a kludge to prevent these oops until we can put in a proper
467 * hook in del_gendisk() to inform the VFS and file system layers.
468 */
469static int block_device_ejected(struct super_block *sb)
470{
471 struct inode *bd_inode = sb->s_bdev->bd_inode;
472 struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
473
474 return bdi->dev == NULL;
475}
476
477static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
478{
479 struct super_block *sb = journal->j_private;
480 struct ext4_sb_info *sbi = EXT4_SB(sb);
481 int error = is_journal_aborted(journal);
482 struct ext4_journal_cb_entry *jce, *tmp;
483
484 spin_lock(&sbi->s_md_lock);
485 list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
486 list_del_init(&jce->jce_list);
487 spin_unlock(&sbi->s_md_lock);
488 jce->jce_func(sb, jce, error);
489 spin_lock(&sbi->s_md_lock);
490 }
491 spin_unlock(&sbi->s_md_lock);
492}
493
494/* Deal with the reporting of failure conditions on a filesystem such as
495 * inconsistencies detected or read IO failures.
496 *
497 * On ext2, we can store the error state of the filesystem in the
498 * superblock. That is not possible on ext4, because we may have other
499 * write ordering constraints on the superblock which prevent us from
500 * writing it out straight away; and given that the journal is about to
501 * be aborted, we can't rely on the current, or future, transactions to
502 * write out the superblock safely.
503 *
504 * We'll just use the jbd2_journal_abort() error code to record an error in
505 * the journal instead. On recovery, the journal will complain about
506 * that error until we've noted it down and cleared it.
507 */
508
509static void ext4_handle_error(struct super_block *sb)
510{
511 if (sb->s_flags & MS_RDONLY)
512 return;
513
514 if (!test_opt(sb, ERRORS_CONT)) {
515 journal_t *journal = EXT4_SB(sb)->s_journal;
516
517 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
518 if (journal)
519 jbd2_journal_abort(journal, -EIO);
520 }
521 if (test_opt(sb, ERRORS_RO)) {
522 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
523 sb->s_flags |= MS_RDONLY;
524 }
525 if (test_opt(sb, ERRORS_PANIC))
526 panic("EXT4-fs (device %s): panic forced after error\n",
527 sb->s_id);
528}
529
530void __ext4_error(struct super_block *sb, const char *function,
531 unsigned int line, const char *fmt, ...)
532{
533 struct va_format vaf;
534 va_list args;
535
536 va_start(args, fmt);
537 vaf.fmt = fmt;
538 vaf.va = &args;
539 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
540 sb->s_id, function, line, current->comm, &vaf);
541 va_end(args);
542 save_error_info(sb, function, line);
543
544 ext4_handle_error(sb);
545}
546
547void ext4_error_inode(struct inode *inode, const char *function,
548 unsigned int line, ext4_fsblk_t block,
549 const char *fmt, ...)
550{
551 va_list args;
552 struct va_format vaf;
553 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
554
555 es->s_last_error_ino = cpu_to_le32(inode->i_ino);
556 es->s_last_error_block = cpu_to_le64(block);
557 save_error_info(inode->i_sb, function, line);
558 va_start(args, fmt);
559 vaf.fmt = fmt;
560 vaf.va = &args;
561 if (block)
562 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
563 "inode #%lu: block %llu: comm %s: %pV\n",
564 inode->i_sb->s_id, function, line, inode->i_ino,
565 block, current->comm, &vaf);
566 else
567 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
568 "inode #%lu: comm %s: %pV\n",
569 inode->i_sb->s_id, function, line, inode->i_ino,
570 current->comm, &vaf);
571 va_end(args);
572
573 ext4_handle_error(inode->i_sb);
574}
575
576void ext4_error_file(struct file *file, const char *function,
577 unsigned int line, ext4_fsblk_t block,
578 const char *fmt, ...)
579{
580 va_list args;
581 struct va_format vaf;
582 struct ext4_super_block *es;
583 struct inode *inode = file->f_dentry->d_inode;
584 char pathname[80], *path;
585
586 es = EXT4_SB(inode->i_sb)->s_es;
587 es->s_last_error_ino = cpu_to_le32(inode->i_ino);
588 save_error_info(inode->i_sb, function, line);
589 path = d_path(&(file->f_path), pathname, sizeof(pathname));
590 if (IS_ERR(path))
591 path = "(unknown)";
592 va_start(args, fmt);
593 vaf.fmt = fmt;
594 vaf.va = &args;
595 if (block)
596 printk(KERN_CRIT
597 "EXT4-fs error (device %s): %s:%d: inode #%lu: "
598 "block %llu: comm %s: path %s: %pV\n",
599 inode->i_sb->s_id, function, line, inode->i_ino,
600 block, current->comm, path, &vaf);
601 else
602 printk(KERN_CRIT
603 "EXT4-fs error (device %s): %s:%d: inode #%lu: "
604 "comm %s: path %s: %pV\n",
605 inode->i_sb->s_id, function, line, inode->i_ino,
606 current->comm, path, &vaf);
607 va_end(args);
608
609 ext4_handle_error(inode->i_sb);
610}
611
612static const char *ext4_decode_error(struct super_block *sb, int errno,
613 char nbuf[16])
614{
615 char *errstr = NULL;
616
617 switch (errno) {
618 case -EIO:
619 errstr = "IO failure";
620 break;
621 case -ENOMEM:
622 errstr = "Out of memory";
623 break;
624 case -EROFS:
625 if (!sb || (EXT4_SB(sb)->s_journal &&
626 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
627 errstr = "Journal has aborted";
628 else
629 errstr = "Readonly filesystem";
630 break;
631 default:
632 /* If the caller passed in an extra buffer for unknown
633 * errors, textualise them now. Else we just return
634 * NULL. */
635 if (nbuf) {
636 /* Check for truncated error codes... */
637 if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
638 errstr = nbuf;
639 }
640 break;
641 }
642
643 return errstr;
644}
645
646/* __ext4_std_error decodes expected errors from journaling functions
647 * automatically and invokes the appropriate error response. */
648
649void __ext4_std_error(struct super_block *sb, const char *function,
650 unsigned int line, int errno)
651{
652 char nbuf[16];
653 const char *errstr;
654
655 /* Special case: if the error is EROFS, and we're not already
656 * inside a transaction, then there's really no point in logging
657 * an error. */
658 if (errno == -EROFS && journal_current_handle() == NULL &&
659 (sb->s_flags & MS_RDONLY))
660 return;
661
662 errstr = ext4_decode_error(sb, errno, nbuf);
663 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
664 sb->s_id, function, line, errstr);
665 save_error_info(sb, function, line);
666
667 ext4_handle_error(sb);
668}
669
670/*
671 * ext4_abort is a much stronger failure handler than ext4_error. The
672 * abort function may be used to deal with unrecoverable failures such
673 * as journal IO errors or ENOMEM at a critical moment in log management.
674 *
675 * We unconditionally force the filesystem into an ABORT|READONLY state,
676 * unless the error response on the fs has been set to panic in which
677 * case we take the easy way out and panic immediately.
678 */
679
680void __ext4_abort(struct super_block *sb, const char *function,
681 unsigned int line, const char *fmt, ...)
682{
683 va_list args;
684
685 save_error_info(sb, function, line);
686 va_start(args, fmt);
687 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
688 function, line);
689 vprintk(fmt, args);
690 printk("\n");
691 va_end(args);
692
693 if ((sb->s_flags & MS_RDONLY) == 0) {
694 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
695 sb->s_flags |= MS_RDONLY;
696 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
697 if (EXT4_SB(sb)->s_journal)
698 jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
699 save_error_info(sb, function, line);
700 }
701 if (test_opt(sb, ERRORS_PANIC))
702 panic("EXT4-fs panic from previous error\n");
703}
704
705void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
706{
707 struct va_format vaf;
708 va_list args;
709
710 va_start(args, fmt);
711 vaf.fmt = fmt;
712 vaf.va = &args;
713 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
714 va_end(args);
715}
716
717void __ext4_warning(struct super_block *sb, const char *function,
718 unsigned int line, const char *fmt, ...)
719{
720 struct va_format vaf;
721 va_list args;
722
723 va_start(args, fmt);
724 vaf.fmt = fmt;
725 vaf.va = &args;
726 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
727 sb->s_id, function, line, &vaf);
728 va_end(args);
729}
730
731void __ext4_grp_locked_error(const char *function, unsigned int line,
732 struct super_block *sb, ext4_group_t grp,
733 unsigned long ino, ext4_fsblk_t block,
734 const char *fmt, ...)
735__releases(bitlock)
736__acquires(bitlock)
737{
738 struct va_format vaf;
739 va_list args;
740 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
741
742 es->s_last_error_ino = cpu_to_le32(ino);
743 es->s_last_error_block = cpu_to_le64(block);
744 __save_error_info(sb, function, line);
745
746 va_start(args, fmt);
747
748 vaf.fmt = fmt;
749 vaf.va = &args;
750 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
751 sb->s_id, function, line, grp);
752 if (ino)
753 printk(KERN_CONT "inode %lu: ", ino);
754 if (block)
755 printk(KERN_CONT "block %llu:", (unsigned long long) block);
756 printk(KERN_CONT "%pV\n", &vaf);
757 va_end(args);
758
759 if (test_opt(sb, ERRORS_CONT)) {
760 ext4_commit_super(sb, 0);
761 return;
762 }
763
764 ext4_unlock_group(sb, grp);
765 ext4_handle_error(sb);
766 /*
767 * We only get here in the ERRORS_RO case; relocking the group
768 * may be dangerous, but nothing bad will happen since the
769 * filesystem will have already been marked read/only and the
770 * journal has been aborted. We return 1 as a hint to callers
771 * who might what to use the return value from
772 * ext4_grp_locked_error() to distinguish between the
773 * ERRORS_CONT and ERRORS_RO case, and perhaps return more
774 * aggressively from the ext4 function in question, with a
775 * more appropriate error code.
776 */
777 ext4_lock_group(sb, grp);
778 return;
779}
780
781void ext4_update_dynamic_rev(struct super_block *sb)
782{
783 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
784
785 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
786 return;
787
788 ext4_warning(sb,
789 "updating to rev %d because of new feature flag, "
790 "running e2fsck is recommended",
791 EXT4_DYNAMIC_REV);
792
793 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
794 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
795 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
796 /* leave es->s_feature_*compat flags alone */
797 /* es->s_uuid will be set by e2fsck if empty */
798
799 /*
800 * The rest of the superblock fields should be zero, and if not it
801 * means they are likely already in use, so leave them alone. We
802 * can leave it up to e2fsck to clean up any inconsistencies there.
803 */
804}
805
806/*
807 * Open the external journal device
808 */
809static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
810{
811 struct block_device *bdev;
812 char b[BDEVNAME_SIZE];
813
814 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
815 if (IS_ERR(bdev))
816 goto fail;
817 return bdev;
818
819fail:
820 ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
821 __bdevname(dev, b), PTR_ERR(bdev));
822 return NULL;
823}
824
825/*
826 * Release the journal device
827 */
828static int ext4_blkdev_put(struct block_device *bdev)
829{
830 return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
831}
832
833static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
834{
835 struct block_device *bdev;
836 int ret = -ENODEV;
837
838 bdev = sbi->journal_bdev;
839 if (bdev) {
840 ret = ext4_blkdev_put(bdev);
841 sbi->journal_bdev = NULL;
842 }
843 return ret;
844}
845
846static inline struct inode *orphan_list_entry(struct list_head *l)
847{
848 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
849}
850
851static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
852{
853 struct list_head *l;
854
855 ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
856 le32_to_cpu(sbi->s_es->s_last_orphan));
857
858 printk(KERN_ERR "sb_info orphan list:\n");
859 list_for_each(l, &sbi->s_orphan) {
860 struct inode *inode = orphan_list_entry(l);
861 printk(KERN_ERR " "
862 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
863 inode->i_sb->s_id, inode->i_ino, inode,
864 inode->i_mode, inode->i_nlink,
865 NEXT_ORPHAN(inode));
866 }
867}
868
869static void ext4_put_super(struct super_block *sb)
870{
871 struct ext4_sb_info *sbi = EXT4_SB(sb);
872 struct ext4_super_block *es = sbi->s_es;
873 int i, err;
874
875 ext4_unregister_li_request(sb);
876 dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
877
878 flush_workqueue(sbi->dio_unwritten_wq);
879 destroy_workqueue(sbi->dio_unwritten_wq);
880
881 lock_super(sb);
882 if (sbi->s_journal) {
883 err = jbd2_journal_destroy(sbi->s_journal);
884 sbi->s_journal = NULL;
885 if (err < 0)
886 ext4_abort(sb, "Couldn't clean up the journal");
887 }
888
889 del_timer(&sbi->s_err_report);
890 ext4_release_system_zone(sb);
891 ext4_mb_release(sb);
892 ext4_ext_release(sb);
893 ext4_xattr_put_super(sb);
894
895 if (!(sb->s_flags & MS_RDONLY)) {
896 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
897 es->s_state = cpu_to_le16(sbi->s_mount_state);
898 }
899 if (sb->s_dirt || !(sb->s_flags & MS_RDONLY))
900 ext4_commit_super(sb, 1);
901
902 if (sbi->s_proc) {
903 remove_proc_entry("options", sbi->s_proc);
904 remove_proc_entry(sb->s_id, ext4_proc_root);
905 }
906 kobject_del(&sbi->s_kobj);
907
908 for (i = 0; i < sbi->s_gdb_count; i++)
909 brelse(sbi->s_group_desc[i]);
910 ext4_kvfree(sbi->s_group_desc);
911 ext4_kvfree(sbi->s_flex_groups);
912 percpu_counter_destroy(&sbi->s_freeclusters_counter);
913 percpu_counter_destroy(&sbi->s_freeinodes_counter);
914 percpu_counter_destroy(&sbi->s_dirs_counter);
915 percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
916 brelse(sbi->s_sbh);
917#ifdef CONFIG_QUOTA
918 for (i = 0; i < MAXQUOTAS; i++)
919 kfree(sbi->s_qf_names[i]);
920#endif
921
922 /* Debugging code just in case the in-memory inode orphan list
923 * isn't empty. The on-disk one can be non-empty if we've
924 * detected an error and taken the fs readonly, but the
925 * in-memory list had better be clean by this point. */
926 if (!list_empty(&sbi->s_orphan))
927 dump_orphan_list(sb, sbi);
928 J_ASSERT(list_empty(&sbi->s_orphan));
929
930 invalidate_bdev(sb->s_bdev);
931 if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
932 /*
933 * Invalidate the journal device's buffers. We don't want them
934 * floating about in memory - the physical journal device may
935 * hotswapped, and it breaks the `ro-after' testing code.
936 */
937 sync_blockdev(sbi->journal_bdev);
938 invalidate_bdev(sbi->journal_bdev);
939 ext4_blkdev_remove(sbi);
940 }
941 if (sbi->s_mmp_tsk)
942 kthread_stop(sbi->s_mmp_tsk);
943 sb->s_fs_info = NULL;
944 /*
945 * Now that we are completely done shutting down the
946 * superblock, we need to actually destroy the kobject.
947 */
948 unlock_super(sb);
949 kobject_put(&sbi->s_kobj);
950 wait_for_completion(&sbi->s_kobj_unregister);
951 if (sbi->s_chksum_driver)
952 crypto_free_shash(sbi->s_chksum_driver);
953 kfree(sbi->s_blockgroup_lock);
954 kfree(sbi);
955}
956
957static struct kmem_cache *ext4_inode_cachep;
958
959/*
960 * Called inside transaction, so use GFP_NOFS
961 */
962static struct inode *ext4_alloc_inode(struct super_block *sb)
963{
964 struct ext4_inode_info *ei;
965
966 ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
967 if (!ei)
968 return NULL;
969
970 ei->vfs_inode.i_version = 1;
971 ei->vfs_inode.i_data.writeback_index = 0;
972 memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
973 INIT_LIST_HEAD(&ei->i_prealloc_list);
974 spin_lock_init(&ei->i_prealloc_lock);
975 ei->i_reserved_data_blocks = 0;
976 ei->i_reserved_meta_blocks = 0;
977 ei->i_allocated_meta_blocks = 0;
978 ei->i_da_metadata_calc_len = 0;
979 ei->i_da_metadata_calc_last_lblock = 0;
980 spin_lock_init(&(ei->i_block_reservation_lock));
981#ifdef CONFIG_QUOTA
982 ei->i_reserved_quota = 0;
983#endif
984 ei->jinode = NULL;
985 INIT_LIST_HEAD(&ei->i_completed_io_list);
986 spin_lock_init(&ei->i_completed_io_lock);
987 ei->cur_aio_dio = NULL;
988 ei->i_sync_tid = 0;
989 ei->i_datasync_tid = 0;
990 atomic_set(&ei->i_ioend_count, 0);
991 atomic_set(&ei->i_aiodio_unwritten, 0);
992
993 return &ei->vfs_inode;
994}
995
996static int ext4_drop_inode(struct inode *inode)
997{
998 int drop = generic_drop_inode(inode);
999
1000 trace_ext4_drop_inode(inode, drop);
1001 return drop;
1002}
1003
1004static void ext4_i_callback(struct rcu_head *head)
1005{
1006 struct inode *inode = container_of(head, struct inode, i_rcu);
1007 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1008}
1009
1010static void ext4_destroy_inode(struct inode *inode)
1011{
1012 if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1013 ext4_msg(inode->i_sb, KERN_ERR,
1014 "Inode %lu (%p): orphan list check failed!",
1015 inode->i_ino, EXT4_I(inode));
1016 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1017 EXT4_I(inode), sizeof(struct ext4_inode_info),
1018 true);
1019 dump_stack();
1020 }
1021 call_rcu(&inode->i_rcu, ext4_i_callback);
1022}
1023
1024static void init_once(void *foo)
1025{
1026 struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1027
1028 INIT_LIST_HEAD(&ei->i_orphan);
1029#ifdef CONFIG_EXT4_FS_XATTR
1030 init_rwsem(&ei->xattr_sem);
1031#endif
1032 init_rwsem(&ei->i_data_sem);
1033 inode_init_once(&ei->vfs_inode);
1034}
1035
1036static int init_inodecache(void)
1037{
1038 ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1039 sizeof(struct ext4_inode_info),
1040 0, (SLAB_RECLAIM_ACCOUNT|
1041 SLAB_MEM_SPREAD),
1042 init_once);
1043 if (ext4_inode_cachep == NULL)
1044 return -ENOMEM;
1045 return 0;
1046}
1047
1048static void destroy_inodecache(void)
1049{
1050 kmem_cache_destroy(ext4_inode_cachep);
1051}
1052
1053void ext4_clear_inode(struct inode *inode)
1054{
1055 invalidate_inode_buffers(inode);
1056 clear_inode(inode);
1057 dquot_drop(inode);
1058 ext4_discard_preallocations(inode);
1059 if (EXT4_I(inode)->jinode) {
1060 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1061 EXT4_I(inode)->jinode);
1062 jbd2_free_inode(EXT4_I(inode)->jinode);
1063 EXT4_I(inode)->jinode = NULL;
1064 }
1065}
1066
1067static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1068 u64 ino, u32 generation)
1069{
1070 struct inode *inode;
1071
1072 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1073 return ERR_PTR(-ESTALE);
1074 if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1075 return ERR_PTR(-ESTALE);
1076
1077 /* iget isn't really right if the inode is currently unallocated!!
1078 *
1079 * ext4_read_inode will return a bad_inode if the inode had been
1080 * deleted, so we should be safe.
1081 *
1082 * Currently we don't know the generation for parent directory, so
1083 * a generation of 0 means "accept any"
1084 */
1085 inode = ext4_iget(sb, ino);
1086 if (IS_ERR(inode))
1087 return ERR_CAST(inode);
1088 if (generation && inode->i_generation != generation) {
1089 iput(inode);
1090 return ERR_PTR(-ESTALE);
1091 }
1092
1093 return inode;
1094}
1095
1096static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1097 int fh_len, int fh_type)
1098{
1099 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1100 ext4_nfs_get_inode);
1101}
1102
1103static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1104 int fh_len, int fh_type)
1105{
1106 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1107 ext4_nfs_get_inode);
1108}
1109
1110/*
1111 * Try to release metadata pages (indirect blocks, directories) which are
1112 * mapped via the block device. Since these pages could have journal heads
1113 * which would prevent try_to_free_buffers() from freeing them, we must use
1114 * jbd2 layer's try_to_free_buffers() function to release them.
1115 */
1116static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1117 gfp_t wait)
1118{
1119 journal_t *journal = EXT4_SB(sb)->s_journal;
1120
1121 WARN_ON(PageChecked(page));
1122 if (!page_has_buffers(page))
1123 return 0;
1124 if (journal)
1125 return jbd2_journal_try_to_free_buffers(journal, page,
1126 wait & ~__GFP_WAIT);
1127 return try_to_free_buffers(page);
1128}
1129
1130#ifdef CONFIG_QUOTA
1131#define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1132#define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1133
1134static int ext4_write_dquot(struct dquot *dquot);
1135static int ext4_acquire_dquot(struct dquot *dquot);
1136static int ext4_release_dquot(struct dquot *dquot);
1137static int ext4_mark_dquot_dirty(struct dquot *dquot);
1138static int ext4_write_info(struct super_block *sb, int type);
1139static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1140 struct path *path);
1141static int ext4_quota_off(struct super_block *sb, int type);
1142static int ext4_quota_on_mount(struct super_block *sb, int type);
1143static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1144 size_t len, loff_t off);
1145static ssize_t ext4_quota_write(struct super_block *sb, int type,
1146 const char *data, size_t len, loff_t off);
1147
1148static const struct dquot_operations ext4_quota_operations = {
1149 .get_reserved_space = ext4_get_reserved_space,
1150 .write_dquot = ext4_write_dquot,
1151 .acquire_dquot = ext4_acquire_dquot,
1152 .release_dquot = ext4_release_dquot,
1153 .mark_dirty = ext4_mark_dquot_dirty,
1154 .write_info = ext4_write_info,
1155 .alloc_dquot = dquot_alloc,
1156 .destroy_dquot = dquot_destroy,
1157};
1158
1159static const struct quotactl_ops ext4_qctl_operations = {
1160 .quota_on = ext4_quota_on,
1161 .quota_off = ext4_quota_off,
1162 .quota_sync = dquot_quota_sync,
1163 .get_info = dquot_get_dqinfo,
1164 .set_info = dquot_set_dqinfo,
1165 .get_dqblk = dquot_get_dqblk,
1166 .set_dqblk = dquot_set_dqblk
1167};
1168#endif
1169
1170static const struct super_operations ext4_sops = {
1171 .alloc_inode = ext4_alloc_inode,
1172 .destroy_inode = ext4_destroy_inode,
1173 .write_inode = ext4_write_inode,
1174 .dirty_inode = ext4_dirty_inode,
1175 .drop_inode = ext4_drop_inode,
1176 .evict_inode = ext4_evict_inode,
1177 .put_super = ext4_put_super,
1178 .sync_fs = ext4_sync_fs,
1179 .freeze_fs = ext4_freeze,
1180 .unfreeze_fs = ext4_unfreeze,
1181 .statfs = ext4_statfs,
1182 .remount_fs = ext4_remount,
1183 .show_options = ext4_show_options,
1184#ifdef CONFIG_QUOTA
1185 .quota_read = ext4_quota_read,
1186 .quota_write = ext4_quota_write,
1187#endif
1188 .bdev_try_to_free_page = bdev_try_to_free_page,
1189};
1190
1191static const struct super_operations ext4_nojournal_sops = {
1192 .alloc_inode = ext4_alloc_inode,
1193 .destroy_inode = ext4_destroy_inode,
1194 .write_inode = ext4_write_inode,
1195 .dirty_inode = ext4_dirty_inode,
1196 .drop_inode = ext4_drop_inode,
1197 .evict_inode = ext4_evict_inode,
1198 .write_super = ext4_write_super,
1199 .put_super = ext4_put_super,
1200 .statfs = ext4_statfs,
1201 .remount_fs = ext4_remount,
1202 .show_options = ext4_show_options,
1203#ifdef CONFIG_QUOTA
1204 .quota_read = ext4_quota_read,
1205 .quota_write = ext4_quota_write,
1206#endif
1207 .bdev_try_to_free_page = bdev_try_to_free_page,
1208};
1209
1210static const struct export_operations ext4_export_ops = {
1211 .fh_to_dentry = ext4_fh_to_dentry,
1212 .fh_to_parent = ext4_fh_to_parent,
1213 .get_parent = ext4_get_parent,
1214};
1215
1216enum {
1217 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1218 Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1219 Opt_nouid32, Opt_debug, Opt_removed,
1220 Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1221 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1222 Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1223 Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1224 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1225 Opt_data_err_abort, Opt_data_err_ignore,
1226 Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1227 Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1228 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1229 Opt_usrquota, Opt_grpquota, Opt_i_version,
1230 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1231 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1232 Opt_inode_readahead_blks, Opt_journal_ioprio,
1233 Opt_dioread_nolock, Opt_dioread_lock,
1234 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1235};
1236
1237static const match_table_t tokens = {
1238 {Opt_bsd_df, "bsddf"},
1239 {Opt_minix_df, "minixdf"},
1240 {Opt_grpid, "grpid"},
1241 {Opt_grpid, "bsdgroups"},
1242 {Opt_nogrpid, "nogrpid"},
1243 {Opt_nogrpid, "sysvgroups"},
1244 {Opt_resgid, "resgid=%u"},
1245 {Opt_resuid, "resuid=%u"},
1246 {Opt_sb, "sb=%u"},
1247 {Opt_err_cont, "errors=continue"},
1248 {Opt_err_panic, "errors=panic"},
1249 {Opt_err_ro, "errors=remount-ro"},
1250 {Opt_nouid32, "nouid32"},
1251 {Opt_debug, "debug"},
1252 {Opt_removed, "oldalloc"},
1253 {Opt_removed, "orlov"},
1254 {Opt_user_xattr, "user_xattr"},
1255 {Opt_nouser_xattr, "nouser_xattr"},
1256 {Opt_acl, "acl"},
1257 {Opt_noacl, "noacl"},
1258 {Opt_noload, "norecovery"},
1259 {Opt_noload, "noload"},
1260 {Opt_removed, "nobh"},
1261 {Opt_removed, "bh"},
1262 {Opt_commit, "commit=%u"},
1263 {Opt_min_batch_time, "min_batch_time=%u"},
1264 {Opt_max_batch_time, "max_batch_time=%u"},
1265 {Opt_journal_dev, "journal_dev=%u"},
1266 {Opt_journal_checksum, "journal_checksum"},
1267 {Opt_journal_async_commit, "journal_async_commit"},
1268 {Opt_abort, "abort"},
1269 {Opt_data_journal, "data=journal"},
1270 {Opt_data_ordered, "data=ordered"},
1271 {Opt_data_writeback, "data=writeback"},
1272 {Opt_data_err_abort, "data_err=abort"},
1273 {Opt_data_err_ignore, "data_err=ignore"},
1274 {Opt_offusrjquota, "usrjquota="},
1275 {Opt_usrjquota, "usrjquota=%s"},
1276 {Opt_offgrpjquota, "grpjquota="},
1277 {Opt_grpjquota, "grpjquota=%s"},
1278 {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1279 {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1280 {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1281 {Opt_grpquota, "grpquota"},
1282 {Opt_noquota, "noquota"},
1283 {Opt_quota, "quota"},
1284 {Opt_usrquota, "usrquota"},
1285 {Opt_barrier, "barrier=%u"},
1286 {Opt_barrier, "barrier"},
1287 {Opt_nobarrier, "nobarrier"},
1288 {Opt_i_version, "i_version"},
1289 {Opt_stripe, "stripe=%u"},
1290 {Opt_delalloc, "delalloc"},
1291 {Opt_nodelalloc, "nodelalloc"},
1292 {Opt_mblk_io_submit, "mblk_io_submit"},
1293 {Opt_nomblk_io_submit, "nomblk_io_submit"},
1294 {Opt_block_validity, "block_validity"},
1295 {Opt_noblock_validity, "noblock_validity"},
1296 {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1297 {Opt_journal_ioprio, "journal_ioprio=%u"},
1298 {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1299 {Opt_auto_da_alloc, "auto_da_alloc"},
1300 {Opt_noauto_da_alloc, "noauto_da_alloc"},
1301 {Opt_dioread_nolock, "dioread_nolock"},
1302 {Opt_dioread_lock, "dioread_lock"},
1303 {Opt_discard, "discard"},
1304 {Opt_nodiscard, "nodiscard"},
1305 {Opt_init_itable, "init_itable=%u"},
1306 {Opt_init_itable, "init_itable"},
1307 {Opt_noinit_itable, "noinit_itable"},
1308 {Opt_removed, "check=none"}, /* mount option from ext2/3 */
1309 {Opt_removed, "nocheck"}, /* mount option from ext2/3 */
1310 {Opt_removed, "reservation"}, /* mount option from ext2/3 */
1311 {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1312 {Opt_removed, "journal=%u"}, /* mount option from ext2/3 */
1313 {Opt_err, NULL},
1314};
1315
1316static ext4_fsblk_t get_sb_block(void **data)
1317{
1318 ext4_fsblk_t sb_block;
1319 char *options = (char *) *data;
1320
1321 if (!options || strncmp(options, "sb=", 3) != 0)
1322 return 1; /* Default location */
1323
1324 options += 3;
1325 /* TODO: use simple_strtoll with >32bit ext4 */
1326 sb_block = simple_strtoul(options, &options, 0);
1327 if (*options && *options != ',') {
1328 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1329 (char *) *data);
1330 return 1;
1331 }
1332 if (*options == ',')
1333 options++;
1334 *data = (void *) options;
1335
1336 return sb_block;
1337}
1338
1339#define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1340static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1341 "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1342
1343#ifdef CONFIG_QUOTA
1344static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1345{
1346 struct ext4_sb_info *sbi = EXT4_SB(sb);
1347 char *qname;
1348
1349 if (sb_any_quota_loaded(sb) &&
1350 !sbi->s_qf_names[qtype]) {
1351 ext4_msg(sb, KERN_ERR,
1352 "Cannot change journaled "
1353 "quota options when quota turned on");
1354 return -1;
1355 }
1356 qname = match_strdup(args);
1357 if (!qname) {
1358 ext4_msg(sb, KERN_ERR,
1359 "Not enough memory for storing quotafile name");
1360 return -1;
1361 }
1362 if (sbi->s_qf_names[qtype] &&
1363 strcmp(sbi->s_qf_names[qtype], qname)) {
1364 ext4_msg(sb, KERN_ERR,
1365 "%s quota file already specified", QTYPE2NAME(qtype));
1366 kfree(qname);
1367 return -1;
1368 }
1369 sbi->s_qf_names[qtype] = qname;
1370 if (strchr(sbi->s_qf_names[qtype], '/')) {
1371 ext4_msg(sb, KERN_ERR,
1372 "quotafile must be on filesystem root");
1373 kfree(sbi->s_qf_names[qtype]);
1374 sbi->s_qf_names[qtype] = NULL;
1375 return -1;
1376 }
1377 set_opt(sb, QUOTA);
1378 return 1;
1379}
1380
1381static int clear_qf_name(struct super_block *sb, int qtype)
1382{
1383
1384 struct ext4_sb_info *sbi = EXT4_SB(sb);
1385
1386 if (sb_any_quota_loaded(sb) &&
1387 sbi->s_qf_names[qtype]) {
1388 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1389 " when quota turned on");
1390 return -1;
1391 }
1392 /*
1393 * The space will be released later when all options are confirmed
1394 * to be correct
1395 */
1396 sbi->s_qf_names[qtype] = NULL;
1397 return 1;
1398}
1399#endif
1400
1401#define MOPT_SET 0x0001
1402#define MOPT_CLEAR 0x0002
1403#define MOPT_NOSUPPORT 0x0004
1404#define MOPT_EXPLICIT 0x0008
1405#define MOPT_CLEAR_ERR 0x0010
1406#define MOPT_GTE0 0x0020
1407#ifdef CONFIG_QUOTA
1408#define MOPT_Q 0
1409#define MOPT_QFMT 0x0040
1410#else
1411#define MOPT_Q MOPT_NOSUPPORT
1412#define MOPT_QFMT MOPT_NOSUPPORT
1413#endif
1414#define MOPT_DATAJ 0x0080
1415
1416static const struct mount_opts {
1417 int token;
1418 int mount_opt;
1419 int flags;
1420} ext4_mount_opts[] = {
1421 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1422 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1423 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1424 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1425 {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1426 {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1427 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1428 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1429 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1430 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1431 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1432 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1433 {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1434 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1435 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1436 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1437 EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1438 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1439 {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1440 {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1441 {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1442 {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1443 {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1444 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1445 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1446 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1447 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1448 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1449 {Opt_commit, 0, MOPT_GTE0},
1450 {Opt_max_batch_time, 0, MOPT_GTE0},
1451 {Opt_min_batch_time, 0, MOPT_GTE0},
1452 {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1453 {Opt_init_itable, 0, MOPT_GTE0},
1454 {Opt_stripe, 0, MOPT_GTE0},
1455 {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1456 {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1457 {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1458#ifdef CONFIG_EXT4_FS_XATTR
1459 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1460 {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1461#else
1462 {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1463 {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1464#endif
1465#ifdef CONFIG_EXT4_FS_POSIX_ACL
1466 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1467 {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1468#else
1469 {Opt_acl, 0, MOPT_NOSUPPORT},
1470 {Opt_noacl, 0, MOPT_NOSUPPORT},
1471#endif
1472 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1473 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1474 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1475 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1476 MOPT_SET | MOPT_Q},
1477 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1478 MOPT_SET | MOPT_Q},
1479 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1480 EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1481 {Opt_usrjquota, 0, MOPT_Q},
1482 {Opt_grpjquota, 0, MOPT_Q},
1483 {Opt_offusrjquota, 0, MOPT_Q},
1484 {Opt_offgrpjquota, 0, MOPT_Q},
1485 {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1486 {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1487 {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1488 {Opt_err, 0, 0}
1489};
1490
1491static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1492 substring_t *args, unsigned long *journal_devnum,
1493 unsigned int *journal_ioprio, int is_remount)
1494{
1495 struct ext4_sb_info *sbi = EXT4_SB(sb);
1496 const struct mount_opts *m;
1497 kuid_t uid;
1498 kgid_t gid;
1499 int arg = 0;
1500
1501#ifdef CONFIG_QUOTA
1502 if (token == Opt_usrjquota)
1503 return set_qf_name(sb, USRQUOTA, &args[0]);
1504 else if (token == Opt_grpjquota)
1505 return set_qf_name(sb, GRPQUOTA, &args[0]);
1506 else if (token == Opt_offusrjquota)
1507 return clear_qf_name(sb, USRQUOTA);
1508 else if (token == Opt_offgrpjquota)
1509 return clear_qf_name(sb, GRPQUOTA);
1510#endif
1511 if (args->from && match_int(args, &arg))
1512 return -1;
1513 switch (token) {
1514 case Opt_noacl:
1515 case Opt_nouser_xattr:
1516 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1517 break;
1518 case Opt_sb:
1519 return 1; /* handled by get_sb_block() */
1520 case Opt_removed:
1521 ext4_msg(sb, KERN_WARNING,
1522 "Ignoring removed %s option", opt);
1523 return 1;
1524 case Opt_resuid:
1525 uid = make_kuid(current_user_ns(), arg);
1526 if (!uid_valid(uid)) {
1527 ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1528 return -1;
1529 }
1530 sbi->s_resuid = uid;
1531 return 1;
1532 case Opt_resgid:
1533 gid = make_kgid(current_user_ns(), arg);
1534 if (!gid_valid(gid)) {
1535 ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1536 return -1;
1537 }
1538 sbi->s_resgid = gid;
1539 return 1;
1540 case Opt_abort:
1541 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1542 return 1;
1543 case Opt_i_version:
1544 sb->s_flags |= MS_I_VERSION;
1545 return 1;
1546 case Opt_journal_dev:
1547 if (is_remount) {
1548 ext4_msg(sb, KERN_ERR,
1549 "Cannot specify journal on remount");
1550 return -1;
1551 }
1552 *journal_devnum = arg;
1553 return 1;
1554 case Opt_journal_ioprio:
1555 if (arg < 0 || arg > 7)
1556 return -1;
1557 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1558 return 1;
1559 }
1560
1561 for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1562 if (token != m->token)
1563 continue;
1564 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1565 return -1;
1566 if (m->flags & MOPT_EXPLICIT)
1567 set_opt2(sb, EXPLICIT_DELALLOC);
1568 if (m->flags & MOPT_CLEAR_ERR)
1569 clear_opt(sb, ERRORS_MASK);
1570 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1571 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1572 "options when quota turned on");
1573 return -1;
1574 }
1575
1576 if (m->flags & MOPT_NOSUPPORT) {
1577 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1578 } else if (token == Opt_commit) {
1579 if (arg == 0)
1580 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1581 sbi->s_commit_interval = HZ * arg;
1582 } else if (token == Opt_max_batch_time) {
1583 if (arg == 0)
1584 arg = EXT4_DEF_MAX_BATCH_TIME;
1585 sbi->s_max_batch_time = arg;
1586 } else if (token == Opt_min_batch_time) {
1587 sbi->s_min_batch_time = arg;
1588 } else if (token == Opt_inode_readahead_blks) {
1589 if (arg > (1 << 30))
1590 return -1;
1591 if (arg && !is_power_of_2(arg)) {
1592 ext4_msg(sb, KERN_ERR,
1593 "EXT4-fs: inode_readahead_blks"
1594 " must be a power of 2");
1595 return -1;
1596 }
1597 sbi->s_inode_readahead_blks = arg;
1598 } else if (token == Opt_init_itable) {
1599 set_opt(sb, INIT_INODE_TABLE);
1600 if (!args->from)
1601 arg = EXT4_DEF_LI_WAIT_MULT;
1602 sbi->s_li_wait_mult = arg;
1603 } else if (token == Opt_stripe) {
1604 sbi->s_stripe = arg;
1605 } else if (m->flags & MOPT_DATAJ) {
1606 if (is_remount) {
1607 if (!sbi->s_journal)
1608 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1609 else if (test_opt(sb, DATA_FLAGS) !=
1610 m->mount_opt) {
1611 ext4_msg(sb, KERN_ERR,
1612 "Cannot change data mode on remount");
1613 return -1;
1614 }
1615 } else {
1616 clear_opt(sb, DATA_FLAGS);
1617 sbi->s_mount_opt |= m->mount_opt;
1618 }
1619#ifdef CONFIG_QUOTA
1620 } else if (m->flags & MOPT_QFMT) {
1621 if (sb_any_quota_loaded(sb) &&
1622 sbi->s_jquota_fmt != m->mount_opt) {
1623 ext4_msg(sb, KERN_ERR, "Cannot "
1624 "change journaled quota options "
1625 "when quota turned on");
1626 return -1;
1627 }
1628 sbi->s_jquota_fmt = m->mount_opt;
1629#endif
1630 } else {
1631 if (!args->from)
1632 arg = 1;
1633 if (m->flags & MOPT_CLEAR)
1634 arg = !arg;
1635 else if (unlikely(!(m->flags & MOPT_SET))) {
1636 ext4_msg(sb, KERN_WARNING,
1637 "buggy handling of option %s", opt);
1638 WARN_ON(1);
1639 return -1;
1640 }
1641 if (arg != 0)
1642 sbi->s_mount_opt |= m->mount_opt;
1643 else
1644 sbi->s_mount_opt &= ~m->mount_opt;
1645 }
1646 return 1;
1647 }
1648 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1649 "or missing value", opt);
1650 return -1;
1651}
1652
1653static int parse_options(char *options, struct super_block *sb,
1654 unsigned long *journal_devnum,
1655 unsigned int *journal_ioprio,
1656 int is_remount)
1657{
1658#ifdef CONFIG_QUOTA
1659 struct ext4_sb_info *sbi = EXT4_SB(sb);
1660#endif
1661 char *p;
1662 substring_t args[MAX_OPT_ARGS];
1663 int token;
1664
1665 if (!options)
1666 return 1;
1667
1668 while ((p = strsep(&options, ",")) != NULL) {
1669 if (!*p)
1670 continue;
1671 /*
1672 * Initialize args struct so we know whether arg was
1673 * found; some options take optional arguments.
1674 */
1675 args[0].to = args[0].from = 0;
1676 token = match_token(p, tokens, args);
1677 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1678 journal_ioprio, is_remount) < 0)
1679 return 0;
1680 }
1681#ifdef CONFIG_QUOTA
1682 if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1683 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1684 clear_opt(sb, USRQUOTA);
1685
1686 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1687 clear_opt(sb, GRPQUOTA);
1688
1689 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1690 ext4_msg(sb, KERN_ERR, "old and new quota "
1691 "format mixing");
1692 return 0;
1693 }
1694
1695 if (!sbi->s_jquota_fmt) {
1696 ext4_msg(sb, KERN_ERR, "journaled quota format "
1697 "not specified");
1698 return 0;
1699 }
1700 } else {
1701 if (sbi->s_jquota_fmt) {
1702 ext4_msg(sb, KERN_ERR, "journaled quota format "
1703 "specified with no journaling "
1704 "enabled");
1705 return 0;
1706 }
1707 }
1708#endif
1709 return 1;
1710}
1711
1712static inline void ext4_show_quota_options(struct seq_file *seq,
1713 struct super_block *sb)
1714{
1715#if defined(CONFIG_QUOTA)
1716 struct ext4_sb_info *sbi = EXT4_SB(sb);
1717
1718 if (sbi->s_jquota_fmt) {
1719 char *fmtname = "";
1720
1721 switch (sbi->s_jquota_fmt) {
1722 case QFMT_VFS_OLD:
1723 fmtname = "vfsold";
1724 break;
1725 case QFMT_VFS_V0:
1726 fmtname = "vfsv0";
1727 break;
1728 case QFMT_VFS_V1:
1729 fmtname = "vfsv1";
1730 break;
1731 }
1732 seq_printf(seq, ",jqfmt=%s", fmtname);
1733 }
1734
1735 if (sbi->s_qf_names[USRQUOTA])
1736 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1737
1738 if (sbi->s_qf_names[GRPQUOTA])
1739 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1740
1741 if (test_opt(sb, USRQUOTA))
1742 seq_puts(seq, ",usrquota");
1743
1744 if (test_opt(sb, GRPQUOTA))
1745 seq_puts(seq, ",grpquota");
1746#endif
1747}
1748
1749static const char *token2str(int token)
1750{
1751 static const struct match_token *t;
1752
1753 for (t = tokens; t->token != Opt_err; t++)
1754 if (t->token == token && !strchr(t->pattern, '='))
1755 break;
1756 return t->pattern;
1757}
1758
1759/*
1760 * Show an option if
1761 * - it's set to a non-default value OR
1762 * - if the per-sb default is different from the global default
1763 */
1764static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1765 int nodefs)
1766{
1767 struct ext4_sb_info *sbi = EXT4_SB(sb);
1768 struct ext4_super_block *es = sbi->s_es;
1769 int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1770 const struct mount_opts *m;
1771 char sep = nodefs ? '\n' : ',';
1772
1773#define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1774#define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1775
1776 if (sbi->s_sb_block != 1)
1777 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1778
1779 for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1780 int want_set = m->flags & MOPT_SET;
1781 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1782 (m->flags & MOPT_CLEAR_ERR))
1783 continue;
1784 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1785 continue; /* skip if same as the default */
1786 if ((want_set &&
1787 (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1788 (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1789 continue; /* select Opt_noFoo vs Opt_Foo */
1790 SEQ_OPTS_PRINT("%s", token2str(m->token));
1791 }
1792
1793 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1794 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1795 SEQ_OPTS_PRINT("resuid=%u",
1796 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1797 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1798 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1799 SEQ_OPTS_PRINT("resgid=%u",
1800 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1801 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1802 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1803 SEQ_OPTS_PUTS("errors=remount-ro");
1804 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1805 SEQ_OPTS_PUTS("errors=continue");
1806 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1807 SEQ_OPTS_PUTS("errors=panic");
1808 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1809 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1810 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1811 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1812 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1813 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1814 if (sb->s_flags & MS_I_VERSION)
1815 SEQ_OPTS_PUTS("i_version");
1816 if (nodefs || sbi->s_stripe)
1817 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1818 if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1819 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1820 SEQ_OPTS_PUTS("data=journal");
1821 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1822 SEQ_OPTS_PUTS("data=ordered");
1823 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1824 SEQ_OPTS_PUTS("data=writeback");
1825 }
1826 if (nodefs ||
1827 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1828 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1829 sbi->s_inode_readahead_blks);
1830
1831 if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1832 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1833 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1834
1835 ext4_show_quota_options(seq, sb);
1836 return 0;
1837}
1838
1839static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1840{
1841 return _ext4_show_options(seq, root->d_sb, 0);
1842}
1843
1844static int options_seq_show(struct seq_file *seq, void *offset)
1845{
1846 struct super_block *sb = seq->private;
1847 int rc;
1848
1849 seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1850 rc = _ext4_show_options(seq, sb, 1);
1851 seq_puts(seq, "\n");
1852 return rc;
1853}
1854
1855static int options_open_fs(struct inode *inode, struct file *file)
1856{
1857 return single_open(file, options_seq_show, PDE(inode)->data);
1858}
1859
1860static const struct file_operations ext4_seq_options_fops = {
1861 .owner = THIS_MODULE,
1862 .open = options_open_fs,
1863 .read = seq_read,
1864 .llseek = seq_lseek,
1865 .release = single_release,
1866};
1867
1868static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1869 int read_only)
1870{
1871 struct ext4_sb_info *sbi = EXT4_SB(sb);
1872 int res = 0;
1873
1874 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1875 ext4_msg(sb, KERN_ERR, "revision level too high, "
1876 "forcing read-only mode");
1877 res = MS_RDONLY;
1878 }
1879 if (read_only)
1880 goto done;
1881 if (!(sbi->s_mount_state & EXT4_VALID_FS))
1882 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1883 "running e2fsck is recommended");
1884 else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1885 ext4_msg(sb, KERN_WARNING,
1886 "warning: mounting fs with errors, "
1887 "running e2fsck is recommended");
1888 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1889 le16_to_cpu(es->s_mnt_count) >=
1890 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1891 ext4_msg(sb, KERN_WARNING,
1892 "warning: maximal mount count reached, "
1893 "running e2fsck is recommended");
1894 else if (le32_to_cpu(es->s_checkinterval) &&
1895 (le32_to_cpu(es->s_lastcheck) +
1896 le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1897 ext4_msg(sb, KERN_WARNING,
1898 "warning: checktime reached, "
1899 "running e2fsck is recommended");
1900 if (!sbi->s_journal)
1901 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1902 if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1903 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1904 le16_add_cpu(&es->s_mnt_count, 1);
1905 es->s_mtime = cpu_to_le32(get_seconds());
1906 ext4_update_dynamic_rev(sb);
1907 if (sbi->s_journal)
1908 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1909
1910 ext4_commit_super(sb, 1);
1911done:
1912 if (test_opt(sb, DEBUG))
1913 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1914 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1915 sb->s_blocksize,
1916 sbi->s_groups_count,
1917 EXT4_BLOCKS_PER_GROUP(sb),
1918 EXT4_INODES_PER_GROUP(sb),
1919 sbi->s_mount_opt, sbi->s_mount_opt2);
1920
1921 cleancache_init_fs(sb);
1922 return res;
1923}
1924
1925static int ext4_fill_flex_info(struct super_block *sb)
1926{
1927 struct ext4_sb_info *sbi = EXT4_SB(sb);
1928 struct ext4_group_desc *gdp = NULL;
1929 ext4_group_t flex_group_count;
1930 ext4_group_t flex_group;
1931 unsigned int groups_per_flex = 0;
1932 size_t size;
1933 int i;
1934
1935 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1936 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1937 sbi->s_log_groups_per_flex = 0;
1938 return 1;
1939 }
1940 groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1941
1942 /* We allocate both existing and potentially added groups */
1943 flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1944 ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1945 EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1946 size = flex_group_count * sizeof(struct flex_groups);
1947 sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1948 if (sbi->s_flex_groups == NULL) {
1949 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1950 flex_group_count);
1951 goto failed;
1952 }
1953
1954 for (i = 0; i < sbi->s_groups_count; i++) {
1955 gdp = ext4_get_group_desc(sb, i, NULL);
1956
1957 flex_group = ext4_flex_group(sbi, i);
1958 atomic_add(ext4_free_inodes_count(sb, gdp),
1959 &sbi->s_flex_groups[flex_group].free_inodes);
1960 atomic_add(ext4_free_group_clusters(sb, gdp),
1961 &sbi->s_flex_groups[flex_group].free_clusters);
1962 atomic_add(ext4_used_dirs_count(sb, gdp),
1963 &sbi->s_flex_groups[flex_group].used_dirs);
1964 }
1965
1966 return 1;
1967failed:
1968 return 0;
1969}
1970
1971static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1972 struct ext4_group_desc *gdp)
1973{
1974 int offset;
1975 __u16 crc = 0;
1976 __le32 le_group = cpu_to_le32(block_group);
1977
1978 if ((sbi->s_es->s_feature_ro_compat &
1979 cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1980 /* Use new metadata_csum algorithm */
1981 __u16 old_csum;
1982 __u32 csum32;
1983
1984 old_csum = gdp->bg_checksum;
1985 gdp->bg_checksum = 0;
1986 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1987 sizeof(le_group));
1988 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
1989 sbi->s_desc_size);
1990 gdp->bg_checksum = old_csum;
1991
1992 crc = csum32 & 0xFFFF;
1993 goto out;
1994 }
1995
1996 /* old crc16 code */
1997 offset = offsetof(struct ext4_group_desc, bg_checksum);
1998
1999 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2000 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2001 crc = crc16(crc, (__u8 *)gdp, offset);
2002 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2003 /* for checksum of struct ext4_group_desc do the rest...*/
2004 if ((sbi->s_es->s_feature_incompat &
2005 cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2006 offset < le16_to_cpu(sbi->s_es->s_desc_size))
2007 crc = crc16(crc, (__u8 *)gdp + offset,
2008 le16_to_cpu(sbi->s_es->s_desc_size) -
2009 offset);
2010
2011out:
2012 return cpu_to_le16(crc);
2013}
2014
2015int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2016 struct ext4_group_desc *gdp)
2017{
2018 if (ext4_has_group_desc_csum(sb) &&
2019 (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2020 block_group, gdp)))
2021 return 0;
2022
2023 return 1;
2024}
2025
2026void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2027 struct ext4_group_desc *gdp)
2028{
2029 if (!ext4_has_group_desc_csum(sb))
2030 return;
2031 gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2032}
2033
2034/* Called at mount-time, super-block is locked */
2035static int ext4_check_descriptors(struct super_block *sb,
2036 ext4_group_t *first_not_zeroed)
2037{
2038 struct ext4_sb_info *sbi = EXT4_SB(sb);
2039 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2040 ext4_fsblk_t last_block;
2041 ext4_fsblk_t block_bitmap;
2042 ext4_fsblk_t inode_bitmap;
2043 ext4_fsblk_t inode_table;
2044 int flexbg_flag = 0;
2045 ext4_group_t i, grp = sbi->s_groups_count;
2046
2047 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2048 flexbg_flag = 1;
2049
2050 ext4_debug("Checking group descriptors");
2051
2052 for (i = 0; i < sbi->s_groups_count; i++) {
2053 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2054
2055 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2056 last_block = ext4_blocks_count(sbi->s_es) - 1;
2057 else
2058 last_block = first_block +
2059 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2060
2061 if ((grp == sbi->s_groups_count) &&
2062 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2063 grp = i;
2064
2065 block_bitmap = ext4_block_bitmap(sb, gdp);
2066 if (block_bitmap < first_block || block_bitmap > last_block) {
2067 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2068 "Block bitmap for group %u not in group "
2069 "(block %llu)!", i, block_bitmap);
2070 return 0;
2071 }
2072 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2073 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2074 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2075 "Inode bitmap for group %u not in group "
2076 "(block %llu)!", i, inode_bitmap);
2077 return 0;
2078 }
2079 inode_table = ext4_inode_table(sb, gdp);
2080 if (inode_table < first_block ||
2081 inode_table + sbi->s_itb_per_group - 1 > last_block) {
2082 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2083 "Inode table for group %u not in group "
2084 "(block %llu)!", i, inode_table);
2085 return 0;
2086 }
2087 ext4_lock_group(sb, i);
2088 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2089 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2090 "Checksum for group %u failed (%u!=%u)",
2091 i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2092 gdp)), le16_to_cpu(gdp->bg_checksum));
2093 if (!(sb->s_flags & MS_RDONLY)) {
2094 ext4_unlock_group(sb, i);
2095 return 0;
2096 }
2097 }
2098 ext4_unlock_group(sb, i);
2099 if (!flexbg_flag)
2100 first_block += EXT4_BLOCKS_PER_GROUP(sb);
2101 }
2102 if (NULL != first_not_zeroed)
2103 *first_not_zeroed = grp;
2104
2105 ext4_free_blocks_count_set(sbi->s_es,
2106 EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2107 sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2108 return 1;
2109}
2110
2111/* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2112 * the superblock) which were deleted from all directories, but held open by
2113 * a process at the time of a crash. We walk the list and try to delete these
2114 * inodes at recovery time (only with a read-write filesystem).
2115 *
2116 * In order to keep the orphan inode chain consistent during traversal (in
2117 * case of crash during recovery), we link each inode into the superblock
2118 * orphan list_head and handle it the same way as an inode deletion during
2119 * normal operation (which journals the operations for us).
2120 *
2121 * We only do an iget() and an iput() on each inode, which is very safe if we
2122 * accidentally point at an in-use or already deleted inode. The worst that
2123 * can happen in this case is that we get a "bit already cleared" message from
2124 * ext4_free_inode(). The only reason we would point at a wrong inode is if
2125 * e2fsck was run on this filesystem, and it must have already done the orphan
2126 * inode cleanup for us, so we can safely abort without any further action.
2127 */
2128static void ext4_orphan_cleanup(struct super_block *sb,
2129 struct ext4_super_block *es)
2130{
2131 unsigned int s_flags = sb->s_flags;
2132 int nr_orphans = 0, nr_truncates = 0;
2133#ifdef CONFIG_QUOTA
2134 int i;
2135#endif
2136 if (!es->s_last_orphan) {
2137 jbd_debug(4, "no orphan inodes to clean up\n");
2138 return;
2139 }
2140
2141 if (bdev_read_only(sb->s_bdev)) {
2142 ext4_msg(sb, KERN_ERR, "write access "
2143 "unavailable, skipping orphan cleanup");
2144 return;
2145 }
2146
2147 /* Check if feature set would not allow a r/w mount */
2148 if (!ext4_feature_set_ok(sb, 0)) {
2149 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2150 "unknown ROCOMPAT features");
2151 return;
2152 }
2153
2154 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2155 if (es->s_last_orphan)
2156 jbd_debug(1, "Errors on filesystem, "
2157 "clearing orphan list.\n");
2158 es->s_last_orphan = 0;
2159 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2160 return;
2161 }
2162
2163 if (s_flags & MS_RDONLY) {
2164 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2165 sb->s_flags &= ~MS_RDONLY;
2166 }
2167#ifdef CONFIG_QUOTA
2168 /* Needed for iput() to work correctly and not trash data */
2169 sb->s_flags |= MS_ACTIVE;
2170 /* Turn on quotas so that they are updated correctly */
2171 for (i = 0; i < MAXQUOTAS; i++) {
2172 if (EXT4_SB(sb)->s_qf_names[i]) {
2173 int ret = ext4_quota_on_mount(sb, i);
2174 if (ret < 0)
2175 ext4_msg(sb, KERN_ERR,
2176 "Cannot turn on journaled "
2177 "quota: error %d", ret);
2178 }
2179 }
2180#endif
2181
2182 while (es->s_last_orphan) {
2183 struct inode *inode;
2184
2185 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2186 if (IS_ERR(inode)) {
2187 es->s_last_orphan = 0;
2188 break;
2189 }
2190
2191 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2192 dquot_initialize(inode);
2193 if (inode->i_nlink) {
2194 ext4_msg(sb, KERN_DEBUG,
2195 "%s: truncating inode %lu to %lld bytes",
2196 __func__, inode->i_ino, inode->i_size);
2197 jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2198 inode->i_ino, inode->i_size);
2199 ext4_truncate(inode);
2200 nr_truncates++;
2201 } else {
2202 ext4_msg(sb, KERN_DEBUG,
2203 "%s: deleting unreferenced inode %lu",
2204 __func__, inode->i_ino);
2205 jbd_debug(2, "deleting unreferenced inode %lu\n",
2206 inode->i_ino);
2207 nr_orphans++;
2208 }
2209 iput(inode); /* The delete magic happens here! */
2210 }
2211
2212#define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2213
2214 if (nr_orphans)
2215 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2216 PLURAL(nr_orphans));
2217 if (nr_truncates)
2218 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2219 PLURAL(nr_truncates));
2220#ifdef CONFIG_QUOTA
2221 /* Turn quotas off */
2222 for (i = 0; i < MAXQUOTAS; i++) {
2223 if (sb_dqopt(sb)->files[i])
2224 dquot_quota_off(sb, i);
2225 }
2226#endif
2227 sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2228}
2229
2230/*
2231 * Maximal extent format file size.
2232 * Resulting logical blkno at s_maxbytes must fit in our on-disk
2233 * extent format containers, within a sector_t, and within i_blocks
2234 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units,
2235 * so that won't be a limiting factor.
2236 *
2237 * However there is other limiting factor. We do store extents in the form
2238 * of starting block and length, hence the resulting length of the extent
2239 * covering maximum file size must fit into on-disk format containers as
2240 * well. Given that length is always by 1 unit bigger than max unit (because
2241 * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2242 *
2243 * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2244 */
2245static loff_t ext4_max_size(int blkbits, int has_huge_files)
2246{
2247 loff_t res;
2248 loff_t upper_limit = MAX_LFS_FILESIZE;
2249
2250 /* small i_blocks in vfs inode? */
2251 if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2252 /*
2253 * CONFIG_LBDAF is not enabled implies the inode
2254 * i_block represent total blocks in 512 bytes
2255 * 32 == size of vfs inode i_blocks * 8
2256 */
2257 upper_limit = (1LL << 32) - 1;
2258
2259 /* total blocks in file system block size */
2260 upper_limit >>= (blkbits - 9);
2261 upper_limit <<= blkbits;
2262 }
2263
2264 /*
2265 * 32-bit extent-start container, ee_block. We lower the maxbytes
2266 * by one fs block, so ee_len can cover the extent of maximum file
2267 * size
2268 */
2269 res = (1LL << 32) - 1;
2270 res <<= blkbits;
2271
2272 /* Sanity check against vm- & vfs- imposed limits */
2273 if (res > upper_limit)
2274 res = upper_limit;
2275
2276 return res;
2277}
2278
2279/*
2280 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect
2281 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2282 * We need to be 1 filesystem block less than the 2^48 sector limit.
2283 */
2284static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2285{
2286 loff_t res = EXT4_NDIR_BLOCKS;
2287 int meta_blocks;
2288 loff_t upper_limit;
2289 /* This is calculated to be the largest file size for a dense, block
2290 * mapped file such that the file's total number of 512-byte sectors,
2291 * including data and all indirect blocks, does not exceed (2^48 - 1).
2292 *
2293 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2294 * number of 512-byte sectors of the file.
2295 */
2296
2297 if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2298 /*
2299 * !has_huge_files or CONFIG_LBDAF not enabled implies that
2300 * the inode i_block field represents total file blocks in
2301 * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2302 */
2303 upper_limit = (1LL << 32) - 1;
2304
2305 /* total blocks in file system block size */
2306 upper_limit >>= (bits - 9);
2307
2308 } else {
2309 /*
2310 * We use 48 bit ext4_inode i_blocks
2311 * With EXT4_HUGE_FILE_FL set the i_blocks
2312 * represent total number of blocks in
2313 * file system block size
2314 */
2315 upper_limit = (1LL << 48) - 1;
2316
2317 }
2318
2319 /* indirect blocks */
2320 meta_blocks = 1;
2321 /* double indirect blocks */
2322 meta_blocks += 1 + (1LL << (bits-2));
2323 /* tripple indirect blocks */
2324 meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2325
2326 upper_limit -= meta_blocks;
2327 upper_limit <<= bits;
2328
2329 res += 1LL << (bits-2);
2330 res += 1LL << (2*(bits-2));
2331 res += 1LL << (3*(bits-2));
2332 res <<= bits;
2333 if (res > upper_limit)
2334 res = upper_limit;
2335
2336 if (res > MAX_LFS_FILESIZE)
2337 res = MAX_LFS_FILESIZE;
2338
2339 return res;
2340}
2341
2342static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2343 ext4_fsblk_t logical_sb_block, int nr)
2344{
2345 struct ext4_sb_info *sbi = EXT4_SB(sb);
2346 ext4_group_t bg, first_meta_bg;
2347 int has_super = 0;
2348
2349 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2350
2351 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2352 nr < first_meta_bg)
2353 return logical_sb_block + nr + 1;
2354 bg = sbi->s_desc_per_block * nr;
2355 if (ext4_bg_has_super(sb, bg))
2356 has_super = 1;
2357
2358 return (has_super + ext4_group_first_block_no(sb, bg));
2359}
2360
2361/**
2362 * ext4_get_stripe_size: Get the stripe size.
2363 * @sbi: In memory super block info
2364 *
2365 * If we have specified it via mount option, then
2366 * use the mount option value. If the value specified at mount time is
2367 * greater than the blocks per group use the super block value.
2368 * If the super block value is greater than blocks per group return 0.
2369 * Allocator needs it be less than blocks per group.
2370 *
2371 */
2372static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2373{
2374 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2375 unsigned long stripe_width =
2376 le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2377 int ret;
2378
2379 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2380 ret = sbi->s_stripe;
2381 else if (stripe_width <= sbi->s_blocks_per_group)
2382 ret = stripe_width;
2383 else if (stride <= sbi->s_blocks_per_group)
2384 ret = stride;
2385 else
2386 ret = 0;
2387
2388 /*
2389 * If the stripe width is 1, this makes no sense and
2390 * we set it to 0 to turn off stripe handling code.
2391 */
2392 if (ret <= 1)
2393 ret = 0;
2394
2395 return ret;
2396}
2397
2398/* sysfs supprt */
2399
2400struct ext4_attr {
2401 struct attribute attr;
2402 ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2403 ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2404 const char *, size_t);
2405 int offset;
2406};
2407
2408static int parse_strtoul(const char *buf,
2409 unsigned long max, unsigned long *value)
2410{
2411 char *endp;
2412
2413 *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2414 endp = skip_spaces(endp);
2415 if (*endp || *value > max)
2416 return -EINVAL;
2417
2418 return 0;
2419}
2420
2421static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2422 struct ext4_sb_info *sbi,
2423 char *buf)
2424{
2425 return snprintf(buf, PAGE_SIZE, "%llu\n",
2426 (s64) EXT4_C2B(sbi,
2427 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2428}
2429
2430static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2431 struct ext4_sb_info *sbi, char *buf)
2432{
2433 struct super_block *sb = sbi->s_buddy_cache->i_sb;
2434
2435 if (!sb->s_bdev->bd_part)
2436 return snprintf(buf, PAGE_SIZE, "0\n");
2437 return snprintf(buf, PAGE_SIZE, "%lu\n",
2438 (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2439 sbi->s_sectors_written_start) >> 1);
2440}
2441
2442static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2443 struct ext4_sb_info *sbi, char *buf)
2444{
2445 struct super_block *sb = sbi->s_buddy_cache->i_sb;
2446
2447 if (!sb->s_bdev->bd_part)
2448 return snprintf(buf, PAGE_SIZE, "0\n");
2449 return snprintf(buf, PAGE_SIZE, "%llu\n",
2450 (unsigned long long)(sbi->s_kbytes_written +
2451 ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2452 EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2453}
2454
2455static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2456 struct ext4_sb_info *sbi,
2457 const char *buf, size_t count)
2458{
2459 unsigned long t;
2460
2461 if (parse_strtoul(buf, 0x40000000, &t))
2462 return -EINVAL;
2463
2464 if (t && !is_power_of_2(t))
2465 return -EINVAL;
2466
2467 sbi->s_inode_readahead_blks = t;
2468 return count;
2469}
2470
2471static ssize_t sbi_ui_show(struct ext4_attr *a,
2472 struct ext4_sb_info *sbi, char *buf)
2473{
2474 unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2475
2476 return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2477}
2478
2479static ssize_t sbi_ui_store(struct ext4_attr *a,
2480 struct ext4_sb_info *sbi,
2481 const char *buf, size_t count)
2482{
2483 unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2484 unsigned long t;
2485
2486 if (parse_strtoul(buf, 0xffffffff, &t))
2487 return -EINVAL;
2488 *ui = t;
2489 return count;
2490}
2491
2492static ssize_t trigger_test_error(struct ext4_attr *a,
2493 struct ext4_sb_info *sbi,
2494 const char *buf, size_t count)
2495{
2496 int len = count;
2497
2498 if (!capable(CAP_SYS_ADMIN))
2499 return -EPERM;
2500
2501 if (len && buf[len-1] == '\n')
2502 len--;
2503
2504 if (len)
2505 ext4_error(sbi->s_sb, "%.*s", len, buf);
2506 return count;
2507}
2508
2509#define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2510static struct ext4_attr ext4_attr_##_name = { \
2511 .attr = {.name = __stringify(_name), .mode = _mode }, \
2512 .show = _show, \
2513 .store = _store, \
2514 .offset = offsetof(struct ext4_sb_info, _elname), \
2515}
2516#define EXT4_ATTR(name, mode, show, store) \
2517static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2518
2519#define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2520#define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2521#define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2522#define EXT4_RW_ATTR_SBI_UI(name, elname) \
2523 EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2524#define ATTR_LIST(name) &ext4_attr_##name.attr
2525
2526EXT4_RO_ATTR(delayed_allocation_blocks);
2527EXT4_RO_ATTR(session_write_kbytes);
2528EXT4_RO_ATTR(lifetime_write_kbytes);
2529EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2530 inode_readahead_blks_store, s_inode_readahead_blks);
2531EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2532EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2533EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2534EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2535EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2536EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2537EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2538EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2539EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2540
2541static struct attribute *ext4_attrs[] = {
2542 ATTR_LIST(delayed_allocation_blocks),
2543 ATTR_LIST(session_write_kbytes),
2544 ATTR_LIST(lifetime_write_kbytes),
2545 ATTR_LIST(inode_readahead_blks),
2546 ATTR_LIST(inode_goal),
2547 ATTR_LIST(mb_stats),
2548 ATTR_LIST(mb_max_to_scan),
2549 ATTR_LIST(mb_min_to_scan),
2550 ATTR_LIST(mb_order2_req),
2551 ATTR_LIST(mb_stream_req),
2552 ATTR_LIST(mb_group_prealloc),
2553 ATTR_LIST(max_writeback_mb_bump),
2554 ATTR_LIST(trigger_fs_error),
2555 NULL,
2556};
2557
2558/* Features this copy of ext4 supports */
2559EXT4_INFO_ATTR(lazy_itable_init);
2560EXT4_INFO_ATTR(batched_discard);
2561
2562static struct attribute *ext4_feat_attrs[] = {
2563 ATTR_LIST(lazy_itable_init),
2564 ATTR_LIST(batched_discard),
2565 NULL,
2566};
2567
2568static ssize_t ext4_attr_show(struct kobject *kobj,
2569 struct attribute *attr, char *buf)
2570{
2571 struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2572 s_kobj);
2573 struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2574
2575 return a->show ? a->show(a, sbi, buf) : 0;
2576}
2577
2578static ssize_t ext4_attr_store(struct kobject *kobj,
2579 struct attribute *attr,
2580 const char *buf, size_t len)
2581{
2582 struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2583 s_kobj);
2584 struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2585
2586 return a->store ? a->store(a, sbi, buf, len) : 0;
2587}
2588
2589static void ext4_sb_release(struct kobject *kobj)
2590{
2591 struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2592 s_kobj);
2593 complete(&sbi->s_kobj_unregister);
2594}
2595
2596static const struct sysfs_ops ext4_attr_ops = {
2597 .show = ext4_attr_show,
2598 .store = ext4_attr_store,
2599};
2600
2601static struct kobj_type ext4_ktype = {
2602 .default_attrs = ext4_attrs,
2603 .sysfs_ops = &ext4_attr_ops,
2604 .release = ext4_sb_release,
2605};
2606
2607static void ext4_feat_release(struct kobject *kobj)
2608{
2609 complete(&ext4_feat->f_kobj_unregister);
2610}
2611
2612static struct kobj_type ext4_feat_ktype = {
2613 .default_attrs = ext4_feat_attrs,
2614 .sysfs_ops = &ext4_attr_ops,
2615 .release = ext4_feat_release,
2616};
2617
2618/*
2619 * Check whether this filesystem can be mounted based on
2620 * the features present and the RDONLY/RDWR mount requested.
2621 * Returns 1 if this filesystem can be mounted as requested,
2622 * 0 if it cannot be.
2623 */
2624static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2625{
2626 if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2627 ext4_msg(sb, KERN_ERR,
2628 "Couldn't mount because of "
2629 "unsupported optional features (%x)",
2630 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2631 ~EXT4_FEATURE_INCOMPAT_SUPP));
2632 return 0;
2633 }
2634
2635 if (readonly)
2636 return 1;
2637
2638 /* Check that feature set is OK for a read-write mount */
2639 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2640 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2641 "unsupported optional features (%x)",
2642 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2643 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2644 return 0;
2645 }
2646 /*
2647 * Large file size enabled file system can only be mounted
2648 * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2649 */
2650 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2651 if (sizeof(blkcnt_t) < sizeof(u64)) {
2652 ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2653 "cannot be mounted RDWR without "
2654 "CONFIG_LBDAF");
2655 return 0;
2656 }
2657 }
2658 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2659 !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2660 ext4_msg(sb, KERN_ERR,
2661 "Can't support bigalloc feature without "
2662 "extents feature\n");
2663 return 0;
2664 }
2665 return 1;
2666}
2667
2668/*
2669 * This function is called once a day if we have errors logged
2670 * on the file system
2671 */
2672static void print_daily_error_info(unsigned long arg)
2673{
2674 struct super_block *sb = (struct super_block *) arg;
2675 struct ext4_sb_info *sbi;
2676 struct ext4_super_block *es;
2677
2678 sbi = EXT4_SB(sb);
2679 es = sbi->s_es;
2680
2681 if (es->s_error_count)
2682 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2683 le32_to_cpu(es->s_error_count));
2684 if (es->s_first_error_time) {
2685 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2686 sb->s_id, le32_to_cpu(es->s_first_error_time),
2687 (int) sizeof(es->s_first_error_func),
2688 es->s_first_error_func,
2689 le32_to_cpu(es->s_first_error_line));
2690 if (es->s_first_error_ino)
2691 printk(": inode %u",
2692 le32_to_cpu(es->s_first_error_ino));
2693 if (es->s_first_error_block)
2694 printk(": block %llu", (unsigned long long)
2695 le64_to_cpu(es->s_first_error_block));
2696 printk("\n");
2697 }
2698 if (es->s_last_error_time) {
2699 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2700 sb->s_id, le32_to_cpu(es->s_last_error_time),
2701 (int) sizeof(es->s_last_error_func),
2702 es->s_last_error_func,
2703 le32_to_cpu(es->s_last_error_line));
2704 if (es->s_last_error_ino)
2705 printk(": inode %u",
2706 le32_to_cpu(es->s_last_error_ino));
2707 if (es->s_last_error_block)
2708 printk(": block %llu", (unsigned long long)
2709 le64_to_cpu(es->s_last_error_block));
2710 printk("\n");
2711 }
2712 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */
2713}
2714
2715/* Find next suitable group and run ext4_init_inode_table */
2716static int ext4_run_li_request(struct ext4_li_request *elr)
2717{
2718 struct ext4_group_desc *gdp = NULL;
2719 ext4_group_t group, ngroups;
2720 struct super_block *sb;
2721 unsigned long timeout = 0;
2722 int ret = 0;
2723
2724 sb = elr->lr_super;
2725 ngroups = EXT4_SB(sb)->s_groups_count;
2726
2727 for (group = elr->lr_next_group; group < ngroups; group++) {
2728 gdp = ext4_get_group_desc(sb, group, NULL);
2729 if (!gdp) {
2730 ret = 1;
2731 break;
2732 }
2733
2734 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2735 break;
2736 }
2737
2738 if (group == ngroups)
2739 ret = 1;
2740
2741 if (!ret) {
2742 timeout = jiffies;
2743 ret = ext4_init_inode_table(sb, group,
2744 elr->lr_timeout ? 0 : 1);
2745 if (elr->lr_timeout == 0) {
2746 timeout = (jiffies - timeout) *
2747 elr->lr_sbi->s_li_wait_mult;
2748 elr->lr_timeout = timeout;
2749 }
2750 elr->lr_next_sched = jiffies + elr->lr_timeout;
2751 elr->lr_next_group = group + 1;
2752 }
2753
2754 return ret;
2755}
2756
2757/*
2758 * Remove lr_request from the list_request and free the
2759 * request structure. Should be called with li_list_mtx held
2760 */
2761static void ext4_remove_li_request(struct ext4_li_request *elr)
2762{
2763 struct ext4_sb_info *sbi;
2764
2765 if (!elr)
2766 return;
2767
2768 sbi = elr->lr_sbi;
2769
2770 list_del(&elr->lr_request);
2771 sbi->s_li_request = NULL;
2772 kfree(elr);
2773}
2774
2775static void ext4_unregister_li_request(struct super_block *sb)
2776{
2777 mutex_lock(&ext4_li_mtx);
2778 if (!ext4_li_info) {
2779 mutex_unlock(&ext4_li_mtx);
2780 return;
2781 }
2782
2783 mutex_lock(&ext4_li_info->li_list_mtx);
2784 ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2785 mutex_unlock(&ext4_li_info->li_list_mtx);
2786 mutex_unlock(&ext4_li_mtx);
2787}
2788
2789static struct task_struct *ext4_lazyinit_task;
2790
2791/*
2792 * This is the function where ext4lazyinit thread lives. It walks
2793 * through the request list searching for next scheduled filesystem.
2794 * When such a fs is found, run the lazy initialization request
2795 * (ext4_rn_li_request) and keep track of the time spend in this
2796 * function. Based on that time we compute next schedule time of
2797 * the request. When walking through the list is complete, compute
2798 * next waking time and put itself into sleep.
2799 */
2800static int ext4_lazyinit_thread(void *arg)
2801{
2802 struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2803 struct list_head *pos, *n;
2804 struct ext4_li_request *elr;
2805 unsigned long next_wakeup, cur;
2806
2807 BUG_ON(NULL == eli);
2808
2809cont_thread:
2810 while (true) {
2811 next_wakeup = MAX_JIFFY_OFFSET;
2812
2813 mutex_lock(&eli->li_list_mtx);
2814 if (list_empty(&eli->li_request_list)) {
2815 mutex_unlock(&eli->li_list_mtx);
2816 goto exit_thread;
2817 }
2818
2819 list_for_each_safe(pos, n, &eli->li_request_list) {
2820 elr = list_entry(pos, struct ext4_li_request,
2821 lr_request);
2822
2823 if (time_after_eq(jiffies, elr->lr_next_sched)) {
2824 if (ext4_run_li_request(elr) != 0) {
2825 /* error, remove the lazy_init job */
2826 ext4_remove_li_request(elr);
2827 continue;
2828 }
2829 }
2830
2831 if (time_before(elr->lr_next_sched, next_wakeup))
2832 next_wakeup = elr->lr_next_sched;
2833 }
2834 mutex_unlock(&eli->li_list_mtx);
2835
2836 try_to_freeze();
2837
2838 cur = jiffies;
2839 if ((time_after_eq(cur, next_wakeup)) ||
2840 (MAX_JIFFY_OFFSET == next_wakeup)) {
2841 cond_resched();
2842 continue;
2843 }
2844
2845 schedule_timeout_interruptible(next_wakeup - cur);
2846
2847 if (kthread_should_stop()) {
2848 ext4_clear_request_list();
2849 goto exit_thread;
2850 }
2851 }
2852
2853exit_thread:
2854 /*
2855 * It looks like the request list is empty, but we need
2856 * to check it under the li_list_mtx lock, to prevent any
2857 * additions into it, and of course we should lock ext4_li_mtx
2858 * to atomically free the list and ext4_li_info, because at
2859 * this point another ext4 filesystem could be registering
2860 * new one.
2861 */
2862 mutex_lock(&ext4_li_mtx);
2863 mutex_lock(&eli->li_list_mtx);
2864 if (!list_empty(&eli->li_request_list)) {
2865 mutex_unlock(&eli->li_list_mtx);
2866 mutex_unlock(&ext4_li_mtx);
2867 goto cont_thread;
2868 }
2869 mutex_unlock(&eli->li_list_mtx);
2870 kfree(ext4_li_info);
2871 ext4_li_info = NULL;
2872 mutex_unlock(&ext4_li_mtx);
2873
2874 return 0;
2875}
2876
2877static void ext4_clear_request_list(void)
2878{
2879 struct list_head *pos, *n;
2880 struct ext4_li_request *elr;
2881
2882 mutex_lock(&ext4_li_info->li_list_mtx);
2883 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2884 elr = list_entry(pos, struct ext4_li_request,
2885 lr_request);
2886 ext4_remove_li_request(elr);
2887 }
2888 mutex_unlock(&ext4_li_info->li_list_mtx);
2889}
2890
2891static int ext4_run_lazyinit_thread(void)
2892{
2893 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2894 ext4_li_info, "ext4lazyinit");
2895 if (IS_ERR(ext4_lazyinit_task)) {
2896 int err = PTR_ERR(ext4_lazyinit_task);
2897 ext4_clear_request_list();
2898 kfree(ext4_li_info);
2899 ext4_li_info = NULL;
2900 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2901 "initialization thread\n",
2902 err);
2903 return err;
2904 }
2905 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2906 return 0;
2907}
2908
2909/*
2910 * Check whether it make sense to run itable init. thread or not.
2911 * If there is at least one uninitialized inode table, return
2912 * corresponding group number, else the loop goes through all
2913 * groups and return total number of groups.
2914 */
2915static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2916{
2917 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2918 struct ext4_group_desc *gdp = NULL;
2919
2920 for (group = 0; group < ngroups; group++) {
2921 gdp = ext4_get_group_desc(sb, group, NULL);
2922 if (!gdp)
2923 continue;
2924
2925 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2926 break;
2927 }
2928
2929 return group;
2930}
2931
2932static int ext4_li_info_new(void)
2933{
2934 struct ext4_lazy_init *eli = NULL;
2935
2936 eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2937 if (!eli)
2938 return -ENOMEM;
2939
2940 INIT_LIST_HEAD(&eli->li_request_list);
2941 mutex_init(&eli->li_list_mtx);
2942
2943 eli->li_state |= EXT4_LAZYINIT_QUIT;
2944
2945 ext4_li_info = eli;
2946
2947 return 0;
2948}
2949
2950static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2951 ext4_group_t start)
2952{
2953 struct ext4_sb_info *sbi = EXT4_SB(sb);
2954 struct ext4_li_request *elr;
2955 unsigned long rnd;
2956
2957 elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2958 if (!elr)
2959 return NULL;
2960
2961 elr->lr_super = sb;
2962 elr->lr_sbi = sbi;
2963 elr->lr_next_group = start;
2964
2965 /*
2966 * Randomize first schedule time of the request to
2967 * spread the inode table initialization requests
2968 * better.
2969 */
2970 get_random_bytes(&rnd, sizeof(rnd));
2971 elr->lr_next_sched = jiffies + (unsigned long)rnd %
2972 (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2973
2974 return elr;
2975}
2976
2977static int ext4_register_li_request(struct super_block *sb,
2978 ext4_group_t first_not_zeroed)
2979{
2980 struct ext4_sb_info *sbi = EXT4_SB(sb);
2981 struct ext4_li_request *elr;
2982 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2983 int ret = 0;
2984
2985 if (sbi->s_li_request != NULL) {
2986 /*
2987 * Reset timeout so it can be computed again, because
2988 * s_li_wait_mult might have changed.
2989 */
2990 sbi->s_li_request->lr_timeout = 0;
2991 return 0;
2992 }
2993
2994 if (first_not_zeroed == ngroups ||
2995 (sb->s_flags & MS_RDONLY) ||
2996 !test_opt(sb, INIT_INODE_TABLE))
2997 return 0;
2998
2999 elr = ext4_li_request_new(sb, first_not_zeroed);
3000 if (!elr)
3001 return -ENOMEM;
3002
3003 mutex_lock(&ext4_li_mtx);
3004
3005 if (NULL == ext4_li_info) {
3006 ret = ext4_li_info_new();
3007 if (ret)
3008 goto out;
3009 }
3010
3011 mutex_lock(&ext4_li_info->li_list_mtx);
3012 list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3013 mutex_unlock(&ext4_li_info->li_list_mtx);
3014
3015 sbi->s_li_request = elr;
3016 /*
3017 * set elr to NULL here since it has been inserted to
3018 * the request_list and the removal and free of it is
3019 * handled by ext4_clear_request_list from now on.
3020 */
3021 elr = NULL;
3022
3023 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3024 ret = ext4_run_lazyinit_thread();
3025 if (ret)
3026 goto out;
3027 }
3028out:
3029 mutex_unlock(&ext4_li_mtx);
3030 if (ret)
3031 kfree(elr);
3032 return ret;
3033}
3034
3035/*
3036 * We do not need to lock anything since this is called on
3037 * module unload.
3038 */
3039static void ext4_destroy_lazyinit_thread(void)
3040{
3041 /*
3042 * If thread exited earlier
3043 * there's nothing to be done.
3044 */
3045 if (!ext4_li_info || !ext4_lazyinit_task)
3046 return;
3047
3048 kthread_stop(ext4_lazyinit_task);
3049}
3050
3051static int set_journal_csum_feature_set(struct super_block *sb)
3052{
3053 int ret = 1;
3054 int compat, incompat;
3055 struct ext4_sb_info *sbi = EXT4_SB(sb);
3056
3057 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3058 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3059 /* journal checksum v2 */
3060 compat = 0;
3061 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3062 } else {
3063 /* journal checksum v1 */
3064 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3065 incompat = 0;
3066 }
3067
3068 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3069 ret = jbd2_journal_set_features(sbi->s_journal,
3070 compat, 0,
3071 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3072 incompat);
3073 } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3074 ret = jbd2_journal_set_features(sbi->s_journal,
3075 compat, 0,
3076 incompat);
3077 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3078 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3079 } else {
3080 jbd2_journal_clear_features(sbi->s_journal,
3081 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3082 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3083 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3084 }
3085
3086 return ret;
3087}
3088
3089/*
3090 * Note: calculating the overhead so we can be compatible with
3091 * historical BSD practice is quite difficult in the face of
3092 * clusters/bigalloc. This is because multiple metadata blocks from
3093 * different block group can end up in the same allocation cluster.
3094 * Calculating the exact overhead in the face of clustered allocation
3095 * requires either O(all block bitmaps) in memory or O(number of block
3096 * groups**2) in time. We will still calculate the superblock for
3097 * older file systems --- and if we come across with a bigalloc file
3098 * system with zero in s_overhead_clusters the estimate will be close to
3099 * correct especially for very large cluster sizes --- but for newer
3100 * file systems, it's better to calculate this figure once at mkfs
3101 * time, and store it in the superblock. If the superblock value is
3102 * present (even for non-bigalloc file systems), we will use it.
3103 */
3104static int count_overhead(struct super_block *sb, ext4_group_t grp,
3105 char *buf)
3106{
3107 struct ext4_sb_info *sbi = EXT4_SB(sb);
3108 struct ext4_group_desc *gdp;
3109 ext4_fsblk_t first_block, last_block, b;
3110 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3111 int s, j, count = 0;
3112
3113 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3114 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3115 sbi->s_itb_per_group + 2);
3116
3117 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3118 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3119 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3120 for (i = 0; i < ngroups; i++) {
3121 gdp = ext4_get_group_desc(sb, i, NULL);
3122 b = ext4_block_bitmap(sb, gdp);
3123 if (b >= first_block && b <= last_block) {
3124 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3125 count++;
3126 }
3127 b = ext4_inode_bitmap(sb, gdp);
3128 if (b >= first_block && b <= last_block) {
3129 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3130 count++;
3131 }
3132 b = ext4_inode_table(sb, gdp);
3133 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3134 for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3135 int c = EXT4_B2C(sbi, b - first_block);
3136 ext4_set_bit(c, buf);
3137 count++;
3138 }
3139 if (i != grp)
3140 continue;
3141 s = 0;
3142 if (ext4_bg_has_super(sb, grp)) {
3143 ext4_set_bit(s++, buf);
3144 count++;
3145 }
3146 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3147 ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3148 count++;
3149 }
3150 }
3151 if (!count)
3152 return 0;
3153 return EXT4_CLUSTERS_PER_GROUP(sb) -
3154 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3155}
3156
3157/*
3158 * Compute the overhead and stash it in sbi->s_overhead
3159 */
3160int ext4_calculate_overhead(struct super_block *sb)
3161{
3162 struct ext4_sb_info *sbi = EXT4_SB(sb);
3163 struct ext4_super_block *es = sbi->s_es;
3164 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3165 ext4_fsblk_t overhead = 0;
3166 char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3167
3168 memset(buf, 0, PAGE_SIZE);
3169 if (!buf)
3170 return -ENOMEM;
3171
3172 /*
3173 * Compute the overhead (FS structures). This is constant
3174 * for a given filesystem unless the number of block groups
3175 * changes so we cache the previous value until it does.
3176 */
3177
3178 /*
3179 * All of the blocks before first_data_block are overhead
3180 */
3181 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3182
3183 /*
3184 * Add the overhead found in each block group
3185 */
3186 for (i = 0; i < ngroups; i++) {
3187 int blks;
3188
3189 blks = count_overhead(sb, i, buf);
3190 overhead += blks;
3191 if (blks)
3192 memset(buf, 0, PAGE_SIZE);
3193 cond_resched();
3194 }
3195 sbi->s_overhead = overhead;
3196 smp_wmb();
3197 free_page((unsigned long) buf);
3198 return 0;
3199}
3200
3201static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3202{
3203 char *orig_data = kstrdup(data, GFP_KERNEL);
3204 struct buffer_head *bh;
3205 struct ext4_super_block *es = NULL;
3206 struct ext4_sb_info *sbi;
3207 ext4_fsblk_t block;
3208 ext4_fsblk_t sb_block = get_sb_block(&data);
3209 ext4_fsblk_t logical_sb_block;
3210 unsigned long offset = 0;
3211 unsigned long journal_devnum = 0;
3212 unsigned long def_mount_opts;
3213 struct inode *root;
3214 char *cp;
3215 const char *descr;
3216 int ret = -ENOMEM;
3217 int blocksize, clustersize;
3218 unsigned int db_count;
3219 unsigned int i;
3220 int needs_recovery, has_huge_files, has_bigalloc;
3221 __u64 blocks_count;
3222 int err;
3223 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3224 ext4_group_t first_not_zeroed;
3225
3226 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3227 if (!sbi)
3228 goto out_free_orig;
3229
3230 sbi->s_blockgroup_lock =
3231 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3232 if (!sbi->s_blockgroup_lock) {
3233 kfree(sbi);
3234 goto out_free_orig;
3235 }
3236 sb->s_fs_info = sbi;
3237 sbi->s_sb = sb;
3238 sbi->s_mount_opt = 0;
3239 sbi->s_resuid = make_kuid(&init_user_ns, EXT4_DEF_RESUID);
3240 sbi->s_resgid = make_kgid(&init_user_ns, EXT4_DEF_RESGID);
3241 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3242 sbi->s_sb_block = sb_block;
3243 if (sb->s_bdev->bd_part)
3244 sbi->s_sectors_written_start =
3245 part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3246
3247 /* Cleanup superblock name */
3248 for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3249 *cp = '!';
3250
3251 ret = -EINVAL;
3252 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3253 if (!blocksize) {
3254 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3255 goto out_fail;
3256 }
3257
3258 /*
3259 * The ext4 superblock will not be buffer aligned for other than 1kB
3260 * block sizes. We need to calculate the offset from buffer start.
3261 */
3262 if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3263 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3264 offset = do_div(logical_sb_block, blocksize);
3265 } else {
3266 logical_sb_block = sb_block;
3267 }
3268
3269 if (!(bh = sb_bread(sb, logical_sb_block))) {
3270 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3271 goto out_fail;
3272 }
3273 /*
3274 * Note: s_es must be initialized as soon as possible because
3275 * some ext4 macro-instructions depend on its value
3276 */
3277 es = (struct ext4_super_block *) (bh->b_data + offset);
3278 sbi->s_es = es;
3279 sb->s_magic = le16_to_cpu(es->s_magic);
3280 if (sb->s_magic != EXT4_SUPER_MAGIC)
3281 goto cantfind_ext4;
3282 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3283
3284 /* Warn if metadata_csum and gdt_csum are both set. */
3285 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3286 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3287 EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3288 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3289 "redundant flags; please run fsck.");
3290
3291 /* Check for a known checksum algorithm */
3292 if (!ext4_verify_csum_type(sb, es)) {
3293 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3294 "unknown checksum algorithm.");
3295 silent = 1;
3296 goto cantfind_ext4;
3297 }
3298
3299 /* Load the checksum driver */
3300 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3301 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3302 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3303 if (IS_ERR(sbi->s_chksum_driver)) {
3304 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3305 ret = PTR_ERR(sbi->s_chksum_driver);
3306 sbi->s_chksum_driver = NULL;
3307 goto failed_mount;
3308 }
3309 }
3310
3311 /* Check superblock checksum */
3312 if (!ext4_superblock_csum_verify(sb, es)) {
3313 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3314 "invalid superblock checksum. Run e2fsck?");
3315 silent = 1;
3316 goto cantfind_ext4;
3317 }
3318
3319 /* Precompute checksum seed for all metadata */
3320 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3321 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3322 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3323 sizeof(es->s_uuid));
3324
3325 /* Set defaults before we parse the mount options */
3326 def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3327 set_opt(sb, INIT_INODE_TABLE);
3328 if (def_mount_opts & EXT4_DEFM_DEBUG)
3329 set_opt(sb, DEBUG);
3330 if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3331 set_opt(sb, GRPID);
3332 if (def_mount_opts & EXT4_DEFM_UID16)
3333 set_opt(sb, NO_UID32);
3334 /* xattr user namespace & acls are now defaulted on */
3335#ifdef CONFIG_EXT4_FS_XATTR
3336 set_opt(sb, XATTR_USER);
3337#endif
3338#ifdef CONFIG_EXT4_FS_POSIX_ACL
3339 set_opt(sb, POSIX_ACL);
3340#endif
3341 set_opt(sb, MBLK_IO_SUBMIT);
3342 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3343 set_opt(sb, JOURNAL_DATA);
3344 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3345 set_opt(sb, ORDERED_DATA);
3346 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3347 set_opt(sb, WRITEBACK_DATA);
3348
3349 if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3350 set_opt(sb, ERRORS_PANIC);
3351 else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3352 set_opt(sb, ERRORS_CONT);
3353 else
3354 set_opt(sb, ERRORS_RO);
3355 if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3356 set_opt(sb, BLOCK_VALIDITY);
3357 if (def_mount_opts & EXT4_DEFM_DISCARD)
3358 set_opt(sb, DISCARD);
3359
3360 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3361 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3362 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3363 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3364 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3365
3366 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3367 set_opt(sb, BARRIER);
3368
3369 /*
3370 * enable delayed allocation by default
3371 * Use -o nodelalloc to turn it off
3372 */
3373 if (!IS_EXT3_SB(sb) &&
3374 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3375 set_opt(sb, DELALLOC);
3376
3377 /*
3378 * set default s_li_wait_mult for lazyinit, for the case there is
3379 * no mount option specified.
3380 */
3381 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3382
3383 if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3384 &journal_devnum, &journal_ioprio, 0)) {
3385 ext4_msg(sb, KERN_WARNING,
3386 "failed to parse options in superblock: %s",
3387 sbi->s_es->s_mount_opts);
3388 }
3389 sbi->s_def_mount_opt = sbi->s_mount_opt;
3390 if (!parse_options((char *) data, sb, &journal_devnum,
3391 &journal_ioprio, 0))
3392 goto failed_mount;
3393
3394 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3395 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3396 "with data=journal disables delayed "
3397 "allocation and O_DIRECT support!\n");
3398 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3399 ext4_msg(sb, KERN_ERR, "can't mount with "
3400 "both data=journal and delalloc");
3401 goto failed_mount;
3402 }
3403 if (test_opt(sb, DIOREAD_NOLOCK)) {
3404 ext4_msg(sb, KERN_ERR, "can't mount with "
3405 "both data=journal and delalloc");
3406 goto failed_mount;
3407 }
3408 if (test_opt(sb, DELALLOC))
3409 clear_opt(sb, DELALLOC);
3410 }
3411
3412 blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3413 if (test_opt(sb, DIOREAD_NOLOCK)) {
3414 if (blocksize < PAGE_SIZE) {
3415 ext4_msg(sb, KERN_ERR, "can't mount with "
3416 "dioread_nolock if block size != PAGE_SIZE");
3417 goto failed_mount;
3418 }
3419 }
3420
3421 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3422 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3423
3424 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3425 (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3426 EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3427 EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3428 ext4_msg(sb, KERN_WARNING,
3429 "feature flags set on rev 0 fs, "
3430 "running e2fsck is recommended");
3431
3432 if (IS_EXT2_SB(sb)) {
3433 if (ext2_feature_set_ok(sb))
3434 ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3435 "using the ext4 subsystem");
3436 else {
3437 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3438 "to feature incompatibilities");
3439 goto failed_mount;
3440 }
3441 }
3442
3443 if (IS_EXT3_SB(sb)) {
3444 if (ext3_feature_set_ok(sb))
3445 ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3446 "using the ext4 subsystem");
3447 else {
3448 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3449 "to feature incompatibilities");
3450 goto failed_mount;
3451 }
3452 }
3453
3454 /*
3455 * Check feature flags regardless of the revision level, since we
3456 * previously didn't change the revision level when setting the flags,
3457 * so there is a chance incompat flags are set on a rev 0 filesystem.
3458 */
3459 if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3460 goto failed_mount;
3461
3462 if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3463 blocksize > EXT4_MAX_BLOCK_SIZE) {
3464 ext4_msg(sb, KERN_ERR,
3465 "Unsupported filesystem blocksize %d", blocksize);
3466 goto failed_mount;
3467 }
3468
3469 if (sb->s_blocksize != blocksize) {
3470 /* Validate the filesystem blocksize */
3471 if (!sb_set_blocksize(sb, blocksize)) {
3472 ext4_msg(sb, KERN_ERR, "bad block size %d",
3473 blocksize);
3474 goto failed_mount;
3475 }
3476
3477 brelse(bh);
3478 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3479 offset = do_div(logical_sb_block, blocksize);
3480 bh = sb_bread(sb, logical_sb_block);
3481 if (!bh) {
3482 ext4_msg(sb, KERN_ERR,
3483 "Can't read superblock on 2nd try");
3484 goto failed_mount;
3485 }
3486 es = (struct ext4_super_block *)(bh->b_data + offset);
3487 sbi->s_es = es;
3488 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3489 ext4_msg(sb, KERN_ERR,
3490 "Magic mismatch, very weird!");
3491 goto failed_mount;
3492 }
3493 }
3494
3495 has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3496 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3497 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3498 has_huge_files);
3499 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3500
3501 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3502 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3503 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3504 } else {
3505 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3506 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3507 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3508 (!is_power_of_2(sbi->s_inode_size)) ||
3509 (sbi->s_inode_size > blocksize)) {
3510 ext4_msg(sb, KERN_ERR,
3511 "unsupported inode size: %d",
3512 sbi->s_inode_size);
3513 goto failed_mount;
3514 }
3515 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3516 sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3517 }
3518
3519 sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3520 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3521 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3522 sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3523 !is_power_of_2(sbi->s_desc_size)) {
3524 ext4_msg(sb, KERN_ERR,
3525 "unsupported descriptor size %lu",
3526 sbi->s_desc_size);
3527 goto failed_mount;
3528 }
3529 } else
3530 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3531
3532 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3533 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3534 if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3535 goto cantfind_ext4;
3536
3537 sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3538 if (sbi->s_inodes_per_block == 0)
3539 goto cantfind_ext4;
3540 sbi->s_itb_per_group = sbi->s_inodes_per_group /
3541 sbi->s_inodes_per_block;
3542 sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3543 sbi->s_sbh = bh;
3544 sbi->s_mount_state = le16_to_cpu(es->s_state);
3545 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3546 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3547
3548 for (i = 0; i < 4; i++)
3549 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3550 sbi->s_def_hash_version = es->s_def_hash_version;
3551 i = le32_to_cpu(es->s_flags);
3552 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3553 sbi->s_hash_unsigned = 3;
3554 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3555#ifdef __CHAR_UNSIGNED__
3556 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3557 sbi->s_hash_unsigned = 3;
3558#else
3559 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3560#endif
3561 }
3562
3563 /* Handle clustersize */
3564 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3565 has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3566 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3567 if (has_bigalloc) {
3568 if (clustersize < blocksize) {
3569 ext4_msg(sb, KERN_ERR,
3570 "cluster size (%d) smaller than "
3571 "block size (%d)", clustersize, blocksize);
3572 goto failed_mount;
3573 }
3574 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3575 le32_to_cpu(es->s_log_block_size);
3576 sbi->s_clusters_per_group =
3577 le32_to_cpu(es->s_clusters_per_group);
3578 if (sbi->s_clusters_per_group > blocksize * 8) {
3579 ext4_msg(sb, KERN_ERR,
3580 "#clusters per group too big: %lu",
3581 sbi->s_clusters_per_group);
3582 goto failed_mount;
3583 }
3584 if (sbi->s_blocks_per_group !=
3585 (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3586 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3587 "clusters per group (%lu) inconsistent",
3588 sbi->s_blocks_per_group,
3589 sbi->s_clusters_per_group);
3590 goto failed_mount;
3591 }
3592 } else {
3593 if (clustersize != blocksize) {
3594 ext4_warning(sb, "fragment/cluster size (%d) != "
3595 "block size (%d)", clustersize,
3596 blocksize);
3597 clustersize = blocksize;
3598 }
3599 if (sbi->s_blocks_per_group > blocksize * 8) {
3600 ext4_msg(sb, KERN_ERR,
3601 "#blocks per group too big: %lu",
3602 sbi->s_blocks_per_group);
3603 goto failed_mount;
3604 }
3605 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3606 sbi->s_cluster_bits = 0;
3607 }
3608 sbi->s_cluster_ratio = clustersize / blocksize;
3609
3610 if (sbi->s_inodes_per_group > blocksize * 8) {
3611 ext4_msg(sb, KERN_ERR,
3612 "#inodes per group too big: %lu",
3613 sbi->s_inodes_per_group);
3614 goto failed_mount;
3615 }
3616
3617 /*
3618 * Test whether we have more sectors than will fit in sector_t,
3619 * and whether the max offset is addressable by the page cache.
3620 */
3621 err = generic_check_addressable(sb->s_blocksize_bits,
3622 ext4_blocks_count(es));
3623 if (err) {
3624 ext4_msg(sb, KERN_ERR, "filesystem"
3625 " too large to mount safely on this system");
3626 if (sizeof(sector_t) < 8)
3627 ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3628 ret = err;
3629 goto failed_mount;
3630 }
3631
3632 if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3633 goto cantfind_ext4;
3634
3635 /* check blocks count against device size */
3636 blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3637 if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3638 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3639 "exceeds size of device (%llu blocks)",
3640 ext4_blocks_count(es), blocks_count);
3641 goto failed_mount;
3642 }
3643
3644 /*
3645 * It makes no sense for the first data block to be beyond the end
3646 * of the filesystem.
3647 */
3648 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3649 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3650 "block %u is beyond end of filesystem (%llu)",
3651 le32_to_cpu(es->s_first_data_block),
3652 ext4_blocks_count(es));
3653 goto failed_mount;
3654 }
3655 blocks_count = (ext4_blocks_count(es) -
3656 le32_to_cpu(es->s_first_data_block) +
3657 EXT4_BLOCKS_PER_GROUP(sb) - 1);
3658 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3659 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3660 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3661 "(block count %llu, first data block %u, "
3662 "blocks per group %lu)", sbi->s_groups_count,
3663 ext4_blocks_count(es),
3664 le32_to_cpu(es->s_first_data_block),
3665 EXT4_BLOCKS_PER_GROUP(sb));
3666 goto failed_mount;
3667 }
3668 sbi->s_groups_count = blocks_count;
3669 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3670 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3671 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3672 EXT4_DESC_PER_BLOCK(sb);
3673 sbi->s_group_desc = ext4_kvmalloc(db_count *
3674 sizeof(struct buffer_head *),
3675 GFP_KERNEL);
3676 if (sbi->s_group_desc == NULL) {
3677 ext4_msg(sb, KERN_ERR, "not enough memory");
3678 ret = -ENOMEM;
3679 goto failed_mount;
3680 }
3681
3682 if (ext4_proc_root)
3683 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3684
3685 if (sbi->s_proc)
3686 proc_create_data("options", S_IRUGO, sbi->s_proc,
3687 &ext4_seq_options_fops, sb);
3688
3689 bgl_lock_init(sbi->s_blockgroup_lock);
3690
3691 for (i = 0; i < db_count; i++) {
3692 block = descriptor_loc(sb, logical_sb_block, i);
3693 sbi->s_group_desc[i] = sb_bread(sb, block);
3694 if (!sbi->s_group_desc[i]) {
3695 ext4_msg(sb, KERN_ERR,
3696 "can't read group descriptor %d", i);
3697 db_count = i;
3698 goto failed_mount2;
3699 }
3700 }
3701 if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3702 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3703 goto failed_mount2;
3704 }
3705 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3706 if (!ext4_fill_flex_info(sb)) {
3707 ext4_msg(sb, KERN_ERR,
3708 "unable to initialize "
3709 "flex_bg meta info!");
3710 goto failed_mount2;
3711 }
3712
3713 sbi->s_gdb_count = db_count;
3714 get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3715 spin_lock_init(&sbi->s_next_gen_lock);
3716
3717 init_timer(&sbi->s_err_report);
3718 sbi->s_err_report.function = print_daily_error_info;
3719 sbi->s_err_report.data = (unsigned long) sb;
3720
3721 err = percpu_counter_init(&sbi->s_freeclusters_counter,
3722 ext4_count_free_clusters(sb));
3723 if (!err) {
3724 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3725 ext4_count_free_inodes(sb));
3726 }
3727 if (!err) {
3728 err = percpu_counter_init(&sbi->s_dirs_counter,
3729 ext4_count_dirs(sb));
3730 }
3731 if (!err) {
3732 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3733 }
3734 if (err) {
3735 ext4_msg(sb, KERN_ERR, "insufficient memory");
3736 ret = err;
3737 goto failed_mount3;
3738 }
3739
3740 sbi->s_stripe = ext4_get_stripe_size(sbi);
3741 sbi->s_max_writeback_mb_bump = 128;
3742
3743 /*
3744 * set up enough so that it can read an inode
3745 */
3746 if (!test_opt(sb, NOLOAD) &&
3747 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3748 sb->s_op = &ext4_sops;
3749 else
3750 sb->s_op = &ext4_nojournal_sops;
3751 sb->s_export_op = &ext4_export_ops;
3752 sb->s_xattr = ext4_xattr_handlers;
3753#ifdef CONFIG_QUOTA
3754 sb->s_qcop = &ext4_qctl_operations;
3755 sb->dq_op = &ext4_quota_operations;
3756#endif
3757 memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3758
3759 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3760 mutex_init(&sbi->s_orphan_lock);
3761 sbi->s_resize_flags = 0;
3762
3763 sb->s_root = NULL;
3764
3765 needs_recovery = (es->s_last_orphan != 0 ||
3766 EXT4_HAS_INCOMPAT_FEATURE(sb,
3767 EXT4_FEATURE_INCOMPAT_RECOVER));
3768
3769 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3770 !(sb->s_flags & MS_RDONLY))
3771 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3772 goto failed_mount3;
3773
3774 /*
3775 * The first inode we look at is the journal inode. Don't try
3776 * root first: it may be modified in the journal!
3777 */
3778 if (!test_opt(sb, NOLOAD) &&
3779 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3780 if (ext4_load_journal(sb, es, journal_devnum))
3781 goto failed_mount3;
3782 } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3783 EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3784 ext4_msg(sb, KERN_ERR, "required journal recovery "
3785 "suppressed and not mounted read-only");
3786 goto failed_mount_wq;
3787 } else {
3788 clear_opt(sb, DATA_FLAGS);
3789 sbi->s_journal = NULL;
3790 needs_recovery = 0;
3791 goto no_journal;
3792 }
3793
3794 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3795 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3796 JBD2_FEATURE_INCOMPAT_64BIT)) {
3797 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3798 goto failed_mount_wq;
3799 }
3800
3801 if (!set_journal_csum_feature_set(sb)) {
3802 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3803 "feature set");
3804 goto failed_mount_wq;
3805 }
3806
3807 /* We have now updated the journal if required, so we can
3808 * validate the data journaling mode. */
3809 switch (test_opt(sb, DATA_FLAGS)) {
3810 case 0:
3811 /* No mode set, assume a default based on the journal
3812 * capabilities: ORDERED_DATA if the journal can
3813 * cope, else JOURNAL_DATA
3814 */
3815 if (jbd2_journal_check_available_features
3816 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3817 set_opt(sb, ORDERED_DATA);
3818 else
3819 set_opt(sb, JOURNAL_DATA);
3820 break;
3821
3822 case EXT4_MOUNT_ORDERED_DATA:
3823 case EXT4_MOUNT_WRITEBACK_DATA:
3824 if (!jbd2_journal_check_available_features
3825 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3826 ext4_msg(sb, KERN_ERR, "Journal does not support "
3827 "requested data journaling mode");
3828 goto failed_mount_wq;
3829 }
3830 default:
3831 break;
3832 }
3833 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3834
3835 sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3836
3837 /*
3838 * The journal may have updated the bg summary counts, so we
3839 * need to update the global counters.
3840 */
3841 percpu_counter_set(&sbi->s_freeclusters_counter,
3842 ext4_count_free_clusters(sb));
3843 percpu_counter_set(&sbi->s_freeinodes_counter,
3844 ext4_count_free_inodes(sb));
3845 percpu_counter_set(&sbi->s_dirs_counter,
3846 ext4_count_dirs(sb));
3847 percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3848
3849no_journal:
3850 /*
3851 * Get the # of file system overhead blocks from the
3852 * superblock if present.
3853 */
3854 if (es->s_overhead_clusters)
3855 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3856 else {
3857 ret = ext4_calculate_overhead(sb);
3858 if (ret)
3859 goto failed_mount_wq;
3860 }
3861
3862 /*
3863 * The maximum number of concurrent works can be high and
3864 * concurrency isn't really necessary. Limit it to 1.
3865 */
3866 EXT4_SB(sb)->dio_unwritten_wq =
3867 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3868 if (!EXT4_SB(sb)->dio_unwritten_wq) {
3869 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3870 goto failed_mount_wq;
3871 }
3872
3873 /*
3874 * The jbd2_journal_load will have done any necessary log recovery,
3875 * so we can safely mount the rest of the filesystem now.
3876 */
3877
3878 root = ext4_iget(sb, EXT4_ROOT_INO);
3879 if (IS_ERR(root)) {
3880 ext4_msg(sb, KERN_ERR, "get root inode failed");
3881 ret = PTR_ERR(root);
3882 root = NULL;
3883 goto failed_mount4;
3884 }
3885 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3886 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3887 iput(root);
3888 goto failed_mount4;
3889 }
3890 sb->s_root = d_make_root(root);
3891 if (!sb->s_root) {
3892 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3893 ret = -ENOMEM;
3894 goto failed_mount4;
3895 }
3896
3897 if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3898 sb->s_flags |= MS_RDONLY;
3899
3900 /* determine the minimum size of new large inodes, if present */
3901 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3902 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3903 EXT4_GOOD_OLD_INODE_SIZE;
3904 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3905 EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3906 if (sbi->s_want_extra_isize <
3907 le16_to_cpu(es->s_want_extra_isize))
3908 sbi->s_want_extra_isize =
3909 le16_to_cpu(es->s_want_extra_isize);
3910 if (sbi->s_want_extra_isize <
3911 le16_to_cpu(es->s_min_extra_isize))
3912 sbi->s_want_extra_isize =
3913 le16_to_cpu(es->s_min_extra_isize);
3914 }
3915 }
3916 /* Check if enough inode space is available */
3917 if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3918 sbi->s_inode_size) {
3919 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3920 EXT4_GOOD_OLD_INODE_SIZE;
3921 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3922 "available");
3923 }
3924
3925 err = ext4_setup_system_zone(sb);
3926 if (err) {
3927 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3928 "zone (%d)", err);
3929 goto failed_mount4a;
3930 }
3931
3932 ext4_ext_init(sb);
3933 err = ext4_mb_init(sb);
3934 if (err) {
3935 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3936 err);
3937 goto failed_mount5;
3938 }
3939
3940 err = ext4_register_li_request(sb, first_not_zeroed);
3941 if (err)
3942 goto failed_mount6;
3943
3944 sbi->s_kobj.kset = ext4_kset;
3945 init_completion(&sbi->s_kobj_unregister);
3946 err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3947 "%s", sb->s_id);
3948 if (err)
3949 goto failed_mount7;
3950
3951 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3952 ext4_orphan_cleanup(sb, es);
3953 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3954 if (needs_recovery) {
3955 ext4_msg(sb, KERN_INFO, "recovery complete");
3956 ext4_mark_recovery_complete(sb, es);
3957 }
3958 if (EXT4_SB(sb)->s_journal) {
3959 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3960 descr = " journalled data mode";
3961 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3962 descr = " ordered data mode";
3963 else
3964 descr = " writeback data mode";
3965 } else
3966 descr = "out journal";
3967
3968 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3969 "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3970 *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3971
3972 if (es->s_error_count)
3973 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3974
3975 kfree(orig_data);
3976 return 0;
3977
3978cantfind_ext4:
3979 if (!silent)
3980 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3981 goto failed_mount;
3982
3983failed_mount7:
3984 ext4_unregister_li_request(sb);
3985failed_mount6:
3986 ext4_mb_release(sb);
3987failed_mount5:
3988 ext4_ext_release(sb);
3989 ext4_release_system_zone(sb);
3990failed_mount4a:
3991 dput(sb->s_root);
3992 sb->s_root = NULL;
3993failed_mount4:
3994 ext4_msg(sb, KERN_ERR, "mount failed");
3995 destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3996failed_mount_wq:
3997 if (sbi->s_journal) {
3998 jbd2_journal_destroy(sbi->s_journal);
3999 sbi->s_journal = NULL;
4000 }
4001failed_mount3:
4002 del_timer(&sbi->s_err_report);
4003 if (sbi->s_flex_groups)
4004 ext4_kvfree(sbi->s_flex_groups);
4005 percpu_counter_destroy(&sbi->s_freeclusters_counter);
4006 percpu_counter_destroy(&sbi->s_freeinodes_counter);
4007 percpu_counter_destroy(&sbi->s_dirs_counter);
4008 percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4009 if (sbi->s_mmp_tsk)
4010 kthread_stop(sbi->s_mmp_tsk);
4011failed_mount2:
4012 for (i = 0; i < db_count; i++)
4013 brelse(sbi->s_group_desc[i]);
4014 ext4_kvfree(sbi->s_group_desc);
4015failed_mount:
4016 if (sbi->s_chksum_driver)
4017 crypto_free_shash(sbi->s_chksum_driver);
4018 if (sbi->s_proc) {
4019 remove_proc_entry("options", sbi->s_proc);
4020 remove_proc_entry(sb->s_id, ext4_proc_root);
4021 }
4022#ifdef CONFIG_QUOTA
4023 for (i = 0; i < MAXQUOTAS; i++)
4024 kfree(sbi->s_qf_names[i]);
4025#endif
4026 ext4_blkdev_remove(sbi);
4027 brelse(bh);
4028out_fail:
4029 sb->s_fs_info = NULL;
4030 kfree(sbi->s_blockgroup_lock);
4031 kfree(sbi);
4032out_free_orig:
4033 kfree(orig_data);
4034 return ret;
4035}
4036
4037/*
4038 * Setup any per-fs journal parameters now. We'll do this both on
4039 * initial mount, once the journal has been initialised but before we've
4040 * done any recovery; and again on any subsequent remount.
4041 */
4042static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4043{
4044 struct ext4_sb_info *sbi = EXT4_SB(sb);
4045
4046 journal->j_commit_interval = sbi->s_commit_interval;
4047 journal->j_min_batch_time = sbi->s_min_batch_time;
4048 journal->j_max_batch_time = sbi->s_max_batch_time;
4049
4050 write_lock(&journal->j_state_lock);
4051 if (test_opt(sb, BARRIER))
4052 journal->j_flags |= JBD2_BARRIER;
4053 else
4054 journal->j_flags &= ~JBD2_BARRIER;
4055 if (test_opt(sb, DATA_ERR_ABORT))
4056 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4057 else
4058 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4059 write_unlock(&journal->j_state_lock);
4060}
4061
4062static journal_t *ext4_get_journal(struct super_block *sb,
4063 unsigned int journal_inum)
4064{
4065 struct inode *journal_inode;
4066 journal_t *journal;
4067
4068 BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4069
4070 /* First, test for the existence of a valid inode on disk. Bad
4071 * things happen if we iget() an unused inode, as the subsequent
4072 * iput() will try to delete it. */
4073
4074 journal_inode = ext4_iget(sb, journal_inum);
4075 if (IS_ERR(journal_inode)) {
4076 ext4_msg(sb, KERN_ERR, "no journal found");
4077 return NULL;
4078 }
4079 if (!journal_inode->i_nlink) {
4080 make_bad_inode(journal_inode);
4081 iput(journal_inode);
4082 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4083 return NULL;
4084 }
4085
4086 jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4087 journal_inode, journal_inode->i_size);
4088 if (!S_ISREG(journal_inode->i_mode)) {
4089 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4090 iput(journal_inode);
4091 return NULL;
4092 }
4093
4094 journal = jbd2_journal_init_inode(journal_inode);
4095 if (!journal) {
4096 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4097 iput(journal_inode);
4098 return NULL;
4099 }
4100 journal->j_private = sb;
4101 ext4_init_journal_params(sb, journal);
4102 return journal;
4103}
4104
4105static journal_t *ext4_get_dev_journal(struct super_block *sb,
4106 dev_t j_dev)
4107{
4108 struct buffer_head *bh;
4109 journal_t *journal;
4110 ext4_fsblk_t start;
4111 ext4_fsblk_t len;
4112 int hblock, blocksize;
4113 ext4_fsblk_t sb_block;
4114 unsigned long offset;
4115 struct ext4_super_block *es;
4116 struct block_device *bdev;
4117
4118 BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4119
4120 bdev = ext4_blkdev_get(j_dev, sb);
4121 if (bdev == NULL)
4122 return NULL;
4123
4124 blocksize = sb->s_blocksize;
4125 hblock = bdev_logical_block_size(bdev);
4126 if (blocksize < hblock) {
4127 ext4_msg(sb, KERN_ERR,
4128 "blocksize too small for journal device");
4129 goto out_bdev;
4130 }
4131
4132 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4133 offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4134 set_blocksize(bdev, blocksize);
4135 if (!(bh = __bread(bdev, sb_block, blocksize))) {
4136 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4137 "external journal");
4138 goto out_bdev;
4139 }
4140
4141 es = (struct ext4_super_block *) (bh->b_data + offset);
4142 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4143 !(le32_to_cpu(es->s_feature_incompat) &
4144 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4145 ext4_msg(sb, KERN_ERR, "external journal has "
4146 "bad superblock");
4147 brelse(bh);
4148 goto out_bdev;
4149 }
4150
4151 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4152 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4153 brelse(bh);
4154 goto out_bdev;
4155 }
4156
4157 len = ext4_blocks_count(es);
4158 start = sb_block + 1;
4159 brelse(bh); /* we're done with the superblock */
4160
4161 journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4162 start, len, blocksize);
4163 if (!journal) {
4164 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4165 goto out_bdev;
4166 }
4167 journal->j_private = sb;
4168 ll_rw_block(READ, 1, &journal->j_sb_buffer);
4169 wait_on_buffer(journal->j_sb_buffer);
4170 if (!buffer_uptodate(journal->j_sb_buffer)) {
4171 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4172 goto out_journal;
4173 }
4174 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4175 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4176 "user (unsupported) - %d",
4177 be32_to_cpu(journal->j_superblock->s_nr_users));
4178 goto out_journal;
4179 }
4180 EXT4_SB(sb)->journal_bdev = bdev;
4181 ext4_init_journal_params(sb, journal);
4182 return journal;
4183
4184out_journal:
4185 jbd2_journal_destroy(journal);
4186out_bdev:
4187 ext4_blkdev_put(bdev);
4188 return NULL;
4189}
4190
4191static int ext4_load_journal(struct super_block *sb,
4192 struct ext4_super_block *es,
4193 unsigned long journal_devnum)
4194{
4195 journal_t *journal;
4196 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4197 dev_t journal_dev;
4198 int err = 0;
4199 int really_read_only;
4200
4201 BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4202
4203 if (journal_devnum &&
4204 journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4205 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4206 "numbers have changed");
4207 journal_dev = new_decode_dev(journal_devnum);
4208 } else
4209 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4210
4211 really_read_only = bdev_read_only(sb->s_bdev);
4212
4213 /*
4214 * Are we loading a blank journal or performing recovery after a
4215 * crash? For recovery, we need to check in advance whether we
4216 * can get read-write access to the device.
4217 */
4218 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4219 if (sb->s_flags & MS_RDONLY) {
4220 ext4_msg(sb, KERN_INFO, "INFO: recovery "
4221 "required on readonly filesystem");
4222 if (really_read_only) {
4223 ext4_msg(sb, KERN_ERR, "write access "
4224 "unavailable, cannot proceed");
4225 return -EROFS;
4226 }
4227 ext4_msg(sb, KERN_INFO, "write access will "
4228 "be enabled during recovery");
4229 }
4230 }
4231
4232 if (journal_inum && journal_dev) {
4233 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4234 "and inode journals!");
4235 return -EINVAL;
4236 }
4237
4238 if (journal_inum) {
4239 if (!(journal = ext4_get_journal(sb, journal_inum)))
4240 return -EINVAL;
4241 } else {
4242 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4243 return -EINVAL;
4244 }
4245
4246 if (!(journal->j_flags & JBD2_BARRIER))
4247 ext4_msg(sb, KERN_INFO, "barriers disabled");
4248
4249 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4250 err = jbd2_journal_wipe(journal, !really_read_only);
4251 if (!err) {
4252 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4253 if (save)
4254 memcpy(save, ((char *) es) +
4255 EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4256 err = jbd2_journal_load(journal);
4257 if (save)
4258 memcpy(((char *) es) + EXT4_S_ERR_START,
4259 save, EXT4_S_ERR_LEN);
4260 kfree(save);
4261 }
4262
4263 if (err) {
4264 ext4_msg(sb, KERN_ERR, "error loading journal");
4265 jbd2_journal_destroy(journal);
4266 return err;
4267 }
4268
4269 EXT4_SB(sb)->s_journal = journal;
4270 ext4_clear_journal_err(sb, es);
4271
4272 if (!really_read_only && journal_devnum &&
4273 journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4274 es->s_journal_dev = cpu_to_le32(journal_devnum);
4275
4276 /* Make sure we flush the recovery flag to disk. */
4277 ext4_commit_super(sb, 1);
4278 }
4279
4280 return 0;
4281}
4282
4283static int ext4_commit_super(struct super_block *sb, int sync)
4284{
4285 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4286 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4287 int error = 0;
4288
4289 if (!sbh || block_device_ejected(sb))
4290 return error;
4291 if (buffer_write_io_error(sbh)) {
4292 /*
4293 * Oh, dear. A previous attempt to write the
4294 * superblock failed. This could happen because the
4295 * USB device was yanked out. Or it could happen to
4296 * be a transient write error and maybe the block will
4297 * be remapped. Nothing we can do but to retry the
4298 * write and hope for the best.
4299 */
4300 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4301 "superblock detected");
4302 clear_buffer_write_io_error(sbh);
4303 set_buffer_uptodate(sbh);
4304 }
4305 /*
4306 * If the file system is mounted read-only, don't update the
4307 * superblock write time. This avoids updating the superblock
4308 * write time when we are mounting the root file system
4309 * read/only but we need to replay the journal; at that point,
4310 * for people who are east of GMT and who make their clock
4311 * tick in localtime for Windows bug-for-bug compatibility,
4312 * the clock is set in the future, and this will cause e2fsck
4313 * to complain and force a full file system check.
4314 */
4315 if (!(sb->s_flags & MS_RDONLY))
4316 es->s_wtime = cpu_to_le32(get_seconds());
4317 if (sb->s_bdev->bd_part)
4318 es->s_kbytes_written =
4319 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4320 ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4321 EXT4_SB(sb)->s_sectors_written_start) >> 1));
4322 else
4323 es->s_kbytes_written =
4324 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4325 ext4_free_blocks_count_set(es,
4326 EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4327 &EXT4_SB(sb)->s_freeclusters_counter)));
4328 es->s_free_inodes_count =
4329 cpu_to_le32(percpu_counter_sum_positive(
4330 &EXT4_SB(sb)->s_freeinodes_counter));
4331 sb->s_dirt = 0;
4332 BUFFER_TRACE(sbh, "marking dirty");
4333 ext4_superblock_csum_set(sb, es);
4334 mark_buffer_dirty(sbh);
4335 if (sync) {
4336 error = sync_dirty_buffer(sbh);
4337 if (error)
4338 return error;
4339
4340 error = buffer_write_io_error(sbh);
4341 if (error) {
4342 ext4_msg(sb, KERN_ERR, "I/O error while writing "
4343 "superblock");
4344 clear_buffer_write_io_error(sbh);
4345 set_buffer_uptodate(sbh);
4346 }
4347 }
4348 return error;
4349}
4350
4351/*
4352 * Have we just finished recovery? If so, and if we are mounting (or
4353 * remounting) the filesystem readonly, then we will end up with a
4354 * consistent fs on disk. Record that fact.
4355 */
4356static void ext4_mark_recovery_complete(struct super_block *sb,
4357 struct ext4_super_block *es)
4358{
4359 journal_t *journal = EXT4_SB(sb)->s_journal;
4360
4361 if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4362 BUG_ON(journal != NULL);
4363 return;
4364 }
4365 jbd2_journal_lock_updates(journal);
4366 if (jbd2_journal_flush(journal) < 0)
4367 goto out;
4368
4369 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4370 sb->s_flags & MS_RDONLY) {
4371 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4372 ext4_commit_super(sb, 1);
4373 }
4374
4375out:
4376 jbd2_journal_unlock_updates(journal);
4377}
4378
4379/*
4380 * If we are mounting (or read-write remounting) a filesystem whose journal
4381 * has recorded an error from a previous lifetime, move that error to the
4382 * main filesystem now.
4383 */
4384static void ext4_clear_journal_err(struct super_block *sb,
4385 struct ext4_super_block *es)
4386{
4387 journal_t *journal;
4388 int j_errno;
4389 const char *errstr;
4390
4391 BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4392
4393 journal = EXT4_SB(sb)->s_journal;
4394
4395 /*
4396 * Now check for any error status which may have been recorded in the
4397 * journal by a prior ext4_error() or ext4_abort()
4398 */
4399
4400 j_errno = jbd2_journal_errno(journal);
4401 if (j_errno) {
4402 char nbuf[16];
4403
4404 errstr = ext4_decode_error(sb, j_errno, nbuf);
4405 ext4_warning(sb, "Filesystem error recorded "
4406 "from previous mount: %s", errstr);
4407 ext4_warning(sb, "Marking fs in need of filesystem check.");
4408
4409 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4410 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4411 ext4_commit_super(sb, 1);
4412
4413 jbd2_journal_clear_err(journal);
4414 jbd2_journal_update_sb_errno(journal);
4415 }
4416}
4417
4418/*
4419 * Force the running and committing transactions to commit,
4420 * and wait on the commit.
4421 */
4422int ext4_force_commit(struct super_block *sb)
4423{
4424 journal_t *journal;
4425 int ret = 0;
4426
4427 if (sb->s_flags & MS_RDONLY)
4428 return 0;
4429
4430 journal = EXT4_SB(sb)->s_journal;
4431 if (journal) {
4432 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4433 ret = ext4_journal_force_commit(journal);
4434 }
4435
4436 return ret;
4437}
4438
4439static void ext4_write_super(struct super_block *sb)
4440{
4441 lock_super(sb);
4442 ext4_commit_super(sb, 1);
4443 unlock_super(sb);
4444}
4445
4446static int ext4_sync_fs(struct super_block *sb, int wait)
4447{
4448 int ret = 0;
4449 tid_t target;
4450 struct ext4_sb_info *sbi = EXT4_SB(sb);
4451
4452 trace_ext4_sync_fs(sb, wait);
4453 flush_workqueue(sbi->dio_unwritten_wq);
4454 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4455 if (wait)
4456 jbd2_log_wait_commit(sbi->s_journal, target);
4457 }
4458 return ret;
4459}
4460
4461/*
4462 * LVM calls this function before a (read-only) snapshot is created. This
4463 * gives us a chance to flush the journal completely and mark the fs clean.
4464 *
4465 * Note that only this function cannot bring a filesystem to be in a clean
4466 * state independently, because ext4 prevents a new handle from being started
4467 * by @sb->s_frozen, which stays in an upper layer. It thus needs help from
4468 * the upper layer.
4469 */
4470static int ext4_freeze(struct super_block *sb)
4471{
4472 int error = 0;
4473 journal_t *journal;
4474
4475 if (sb->s_flags & MS_RDONLY)
4476 return 0;
4477
4478 journal = EXT4_SB(sb)->s_journal;
4479
4480 /* Now we set up the journal barrier. */
4481 jbd2_journal_lock_updates(journal);
4482
4483 /*
4484 * Don't clear the needs_recovery flag if we failed to flush
4485 * the journal.
4486 */
4487 error = jbd2_journal_flush(journal);
4488 if (error < 0)
4489 goto out;
4490
4491 /* Journal blocked and flushed, clear needs_recovery flag. */
4492 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4493 error = ext4_commit_super(sb, 1);
4494out:
4495 /* we rely on s_frozen to stop further updates */
4496 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4497 return error;
4498}
4499
4500/*
4501 * Called by LVM after the snapshot is done. We need to reset the RECOVER
4502 * flag here, even though the filesystem is not technically dirty yet.
4503 */
4504static int ext4_unfreeze(struct super_block *sb)
4505{
4506 if (sb->s_flags & MS_RDONLY)
4507 return 0;
4508
4509 lock_super(sb);
4510 /* Reset the needs_recovery flag before the fs is unlocked. */
4511 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4512 ext4_commit_super(sb, 1);
4513 unlock_super(sb);
4514 return 0;
4515}
4516
4517/*
4518 * Structure to save mount options for ext4_remount's benefit
4519 */
4520struct ext4_mount_options {
4521 unsigned long s_mount_opt;
4522 unsigned long s_mount_opt2;
4523 kuid_t s_resuid;
4524 kgid_t s_resgid;
4525 unsigned long s_commit_interval;
4526 u32 s_min_batch_time, s_max_batch_time;
4527#ifdef CONFIG_QUOTA
4528 int s_jquota_fmt;
4529 char *s_qf_names[MAXQUOTAS];
4530#endif
4531};
4532
4533static int ext4_remount(struct super_block *sb, int *flags, char *data)
4534{
4535 struct ext4_super_block *es;
4536 struct ext4_sb_info *sbi = EXT4_SB(sb);
4537 unsigned long old_sb_flags;
4538 struct ext4_mount_options old_opts;
4539 int enable_quota = 0;
4540 ext4_group_t g;
4541 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4542 int err = 0;
4543#ifdef CONFIG_QUOTA
4544 int i;
4545#endif
4546 char *orig_data = kstrdup(data, GFP_KERNEL);
4547
4548 /* Store the original options */
4549 lock_super(sb);
4550 old_sb_flags = sb->s_flags;
4551 old_opts.s_mount_opt = sbi->s_mount_opt;
4552 old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4553 old_opts.s_resuid = sbi->s_resuid;
4554 old_opts.s_resgid = sbi->s_resgid;
4555 old_opts.s_commit_interval = sbi->s_commit_interval;
4556 old_opts.s_min_batch_time = sbi->s_min_batch_time;
4557 old_opts.s_max_batch_time = sbi->s_max_batch_time;
4558#ifdef CONFIG_QUOTA
4559 old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4560 for (i = 0; i < MAXQUOTAS; i++)
4561 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4562#endif
4563 if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4564 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4565
4566 /*
4567 * Allow the "check" option to be passed as a remount option.
4568 */
4569 if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4570 err = -EINVAL;
4571 goto restore_opts;
4572 }
4573
4574 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4575 ext4_abort(sb, "Abort forced by user");
4576
4577 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4578 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4579
4580 es = sbi->s_es;
4581
4582 if (sbi->s_journal) {
4583 ext4_init_journal_params(sb, sbi->s_journal);
4584 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4585 }
4586
4587 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4588 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4589 err = -EROFS;
4590 goto restore_opts;
4591 }
4592
4593 if (*flags & MS_RDONLY) {
4594 err = dquot_suspend(sb, -1);
4595 if (err < 0)
4596 goto restore_opts;
4597
4598 /*
4599 * First of all, the unconditional stuff we have to do
4600 * to disable replay of the journal when we next remount
4601 */
4602 sb->s_flags |= MS_RDONLY;
4603
4604 /*
4605 * OK, test if we are remounting a valid rw partition
4606 * readonly, and if so set the rdonly flag and then
4607 * mark the partition as valid again.
4608 */
4609 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4610 (sbi->s_mount_state & EXT4_VALID_FS))
4611 es->s_state = cpu_to_le16(sbi->s_mount_state);
4612
4613 if (sbi->s_journal)
4614 ext4_mark_recovery_complete(sb, es);
4615 } else {
4616 /* Make sure we can mount this feature set readwrite */
4617 if (!ext4_feature_set_ok(sb, 0)) {
4618 err = -EROFS;
4619 goto restore_opts;
4620 }
4621 /*
4622 * Make sure the group descriptor checksums
4623 * are sane. If they aren't, refuse to remount r/w.
4624 */
4625 for (g = 0; g < sbi->s_groups_count; g++) {
4626 struct ext4_group_desc *gdp =
4627 ext4_get_group_desc(sb, g, NULL);
4628
4629 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4630 ext4_msg(sb, KERN_ERR,
4631 "ext4_remount: Checksum for group %u failed (%u!=%u)",
4632 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4633 le16_to_cpu(gdp->bg_checksum));
4634 err = -EINVAL;
4635 goto restore_opts;
4636 }
4637 }
4638
4639 /*
4640 * If we have an unprocessed orphan list hanging
4641 * around from a previously readonly bdev mount,
4642 * require a full umount/remount for now.
4643 */
4644 if (es->s_last_orphan) {
4645 ext4_msg(sb, KERN_WARNING, "Couldn't "
4646 "remount RDWR because of unprocessed "
4647 "orphan inode list. Please "
4648 "umount/remount instead");
4649 err = -EINVAL;
4650 goto restore_opts;
4651 }
4652
4653 /*
4654 * Mounting a RDONLY partition read-write, so reread
4655 * and store the current valid flag. (It may have
4656 * been changed by e2fsck since we originally mounted
4657 * the partition.)
4658 */
4659 if (sbi->s_journal)
4660 ext4_clear_journal_err(sb, es);
4661 sbi->s_mount_state = le16_to_cpu(es->s_state);
4662 if (!ext4_setup_super(sb, es, 0))
4663 sb->s_flags &= ~MS_RDONLY;
4664 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4665 EXT4_FEATURE_INCOMPAT_MMP))
4666 if (ext4_multi_mount_protect(sb,
4667 le64_to_cpu(es->s_mmp_block))) {
4668 err = -EROFS;
4669 goto restore_opts;
4670 }
4671 enable_quota = 1;
4672 }
4673 }
4674
4675 /*
4676 * Reinitialize lazy itable initialization thread based on
4677 * current settings
4678 */
4679 if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4680 ext4_unregister_li_request(sb);
4681 else {
4682 ext4_group_t first_not_zeroed;
4683 first_not_zeroed = ext4_has_uninit_itable(sb);
4684 ext4_register_li_request(sb, first_not_zeroed);
4685 }
4686
4687 ext4_setup_system_zone(sb);
4688 if (sbi->s_journal == NULL)
4689 ext4_commit_super(sb, 1);
4690
4691#ifdef CONFIG_QUOTA
4692 /* Release old quota file names */
4693 for (i = 0; i < MAXQUOTAS; i++)
4694 if (old_opts.s_qf_names[i] &&
4695 old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4696 kfree(old_opts.s_qf_names[i]);
4697#endif
4698 unlock_super(sb);
4699 if (enable_quota)
4700 dquot_resume(sb, -1);
4701
4702 ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4703 kfree(orig_data);
4704 return 0;
4705
4706restore_opts:
4707 sb->s_flags = old_sb_flags;
4708 sbi->s_mount_opt = old_opts.s_mount_opt;
4709 sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4710 sbi->s_resuid = old_opts.s_resuid;
4711 sbi->s_resgid = old_opts.s_resgid;
4712 sbi->s_commit_interval = old_opts.s_commit_interval;
4713 sbi->s_min_batch_time = old_opts.s_min_batch_time;
4714 sbi->s_max_batch_time = old_opts.s_max_batch_time;
4715#ifdef CONFIG_QUOTA
4716 sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4717 for (i = 0; i < MAXQUOTAS; i++) {
4718 if (sbi->s_qf_names[i] &&
4719 old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4720 kfree(sbi->s_qf_names[i]);
4721 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4722 }
4723#endif
4724 unlock_super(sb);
4725 kfree(orig_data);
4726 return err;
4727}
4728
4729static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4730{
4731 struct super_block *sb = dentry->d_sb;
4732 struct ext4_sb_info *sbi = EXT4_SB(sb);
4733 struct ext4_super_block *es = sbi->s_es;
4734 ext4_fsblk_t overhead = 0;
4735 u64 fsid;
4736 s64 bfree;
4737
4738 if (!test_opt(sb, MINIX_DF))
4739 overhead = sbi->s_overhead;
4740
4741 buf->f_type = EXT4_SUPER_MAGIC;
4742 buf->f_bsize = sb->s_blocksize;
4743 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, sbi->s_overhead);
4744 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4745 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4746 /* prevent underflow in case that few free space is available */
4747 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4748 buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4749 if (buf->f_bfree < ext4_r_blocks_count(es))
4750 buf->f_bavail = 0;
4751 buf->f_files = le32_to_cpu(es->s_inodes_count);
4752 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4753 buf->f_namelen = EXT4_NAME_LEN;
4754 fsid = le64_to_cpup((void *)es->s_uuid) ^
4755 le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4756 buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4757 buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4758
4759 return 0;
4760}
4761
4762/* Helper function for writing quotas on sync - we need to start transaction
4763 * before quota file is locked for write. Otherwise the are possible deadlocks:
4764 * Process 1 Process 2
4765 * ext4_create() quota_sync()
4766 * jbd2_journal_start() write_dquot()
4767 * dquot_initialize() down(dqio_mutex)
4768 * down(dqio_mutex) jbd2_journal_start()
4769 *
4770 */
4771
4772#ifdef CONFIG_QUOTA
4773
4774static inline struct inode *dquot_to_inode(struct dquot *dquot)
4775{
4776 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4777}
4778
4779static int ext4_write_dquot(struct dquot *dquot)
4780{
4781 int ret, err;
4782 handle_t *handle;
4783 struct inode *inode;
4784
4785 inode = dquot_to_inode(dquot);
4786 handle = ext4_journal_start(inode,
4787 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4788 if (IS_ERR(handle))
4789 return PTR_ERR(handle);
4790 ret = dquot_commit(dquot);
4791 err = ext4_journal_stop(handle);
4792 if (!ret)
4793 ret = err;
4794 return ret;
4795}
4796
4797static int ext4_acquire_dquot(struct dquot *dquot)
4798{
4799 int ret, err;
4800 handle_t *handle;
4801
4802 handle = ext4_journal_start(dquot_to_inode(dquot),
4803 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4804 if (IS_ERR(handle))
4805 return PTR_ERR(handle);
4806 ret = dquot_acquire(dquot);
4807 err = ext4_journal_stop(handle);
4808 if (!ret)
4809 ret = err;
4810 return ret;
4811}
4812
4813static int ext4_release_dquot(struct dquot *dquot)
4814{
4815 int ret, err;
4816 handle_t *handle;
4817
4818 handle = ext4_journal_start(dquot_to_inode(dquot),
4819 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4820 if (IS_ERR(handle)) {
4821 /* Release dquot anyway to avoid endless cycle in dqput() */
4822 dquot_release(dquot);
4823 return PTR_ERR(handle);
4824 }
4825 ret = dquot_release(dquot);
4826 err = ext4_journal_stop(handle);
4827 if (!ret)
4828 ret = err;
4829 return ret;
4830}
4831
4832static int ext4_mark_dquot_dirty(struct dquot *dquot)
4833{
4834 /* Are we journaling quotas? */
4835 if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4836 EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4837 dquot_mark_dquot_dirty(dquot);
4838 return ext4_write_dquot(dquot);
4839 } else {
4840 return dquot_mark_dquot_dirty(dquot);
4841 }
4842}
4843
4844static int ext4_write_info(struct super_block *sb, int type)
4845{
4846 int ret, err;
4847 handle_t *handle;
4848
4849 /* Data block + inode block */
4850 handle = ext4_journal_start(sb->s_root->d_inode, 2);
4851 if (IS_ERR(handle))
4852 return PTR_ERR(handle);
4853 ret = dquot_commit_info(sb, type);
4854 err = ext4_journal_stop(handle);
4855 if (!ret)
4856 ret = err;
4857 return ret;
4858}
4859
4860/*
4861 * Turn on quotas during mount time - we need to find
4862 * the quota file and such...
4863 */
4864static int ext4_quota_on_mount(struct super_block *sb, int type)
4865{
4866 return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4867 EXT4_SB(sb)->s_jquota_fmt, type);
4868}
4869
4870/*
4871 * Standard function to be called on quota_on
4872 */
4873static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4874 struct path *path)
4875{
4876 int err;
4877
4878 if (!test_opt(sb, QUOTA))
4879 return -EINVAL;
4880
4881 /* Quotafile not on the same filesystem? */
4882 if (path->dentry->d_sb != sb)
4883 return -EXDEV;
4884 /* Journaling quota? */
4885 if (EXT4_SB(sb)->s_qf_names[type]) {
4886 /* Quotafile not in fs root? */
4887 if (path->dentry->d_parent != sb->s_root)
4888 ext4_msg(sb, KERN_WARNING,
4889 "Quota file not on filesystem root. "
4890 "Journaled quota will not work");
4891 }
4892
4893 /*
4894 * When we journal data on quota file, we have to flush journal to see
4895 * all updates to the file when we bypass pagecache...
4896 */
4897 if (EXT4_SB(sb)->s_journal &&
4898 ext4_should_journal_data(path->dentry->d_inode)) {
4899 /*
4900 * We don't need to lock updates but journal_flush() could
4901 * otherwise be livelocked...
4902 */
4903 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4904 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4905 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4906 if (err)
4907 return err;
4908 }
4909
4910 return dquot_quota_on(sb, type, format_id, path);
4911}
4912
4913static int ext4_quota_off(struct super_block *sb, int type)
4914{
4915 struct inode *inode = sb_dqopt(sb)->files[type];
4916 handle_t *handle;
4917
4918 /* Force all delayed allocation blocks to be allocated.
4919 * Caller already holds s_umount sem */
4920 if (test_opt(sb, DELALLOC))
4921 sync_filesystem(sb);
4922
4923 if (!inode)
4924 goto out;
4925
4926 /* Update modification times of quota files when userspace can
4927 * start looking at them */
4928 handle = ext4_journal_start(inode, 1);
4929 if (IS_ERR(handle))
4930 goto out;
4931 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4932 ext4_mark_inode_dirty(handle, inode);
4933 ext4_journal_stop(handle);
4934
4935out:
4936 return dquot_quota_off(sb, type);
4937}
4938
4939/* Read data from quotafile - avoid pagecache and such because we cannot afford
4940 * acquiring the locks... As quota files are never truncated and quota code
4941 * itself serializes the operations (and no one else should touch the files)
4942 * we don't have to be afraid of races */
4943static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4944 size_t len, loff_t off)
4945{
4946 struct inode *inode = sb_dqopt(sb)->files[type];
4947 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4948 int err = 0;
4949 int offset = off & (sb->s_blocksize - 1);
4950 int tocopy;
4951 size_t toread;
4952 struct buffer_head *bh;
4953 loff_t i_size = i_size_read(inode);
4954
4955 if (off > i_size)
4956 return 0;
4957 if (off+len > i_size)
4958 len = i_size-off;
4959 toread = len;
4960 while (toread > 0) {
4961 tocopy = sb->s_blocksize - offset < toread ?
4962 sb->s_blocksize - offset : toread;
4963 bh = ext4_bread(NULL, inode, blk, 0, &err);
4964 if (err)
4965 return err;
4966 if (!bh) /* A hole? */
4967 memset(data, 0, tocopy);
4968 else
4969 memcpy(data, bh->b_data+offset, tocopy);
4970 brelse(bh);
4971 offset = 0;
4972 toread -= tocopy;
4973 data += tocopy;
4974 blk++;
4975 }
4976 return len;
4977}
4978
4979/* Write to quotafile (we know the transaction is already started and has
4980 * enough credits) */
4981static ssize_t ext4_quota_write(struct super_block *sb, int type,
4982 const char *data, size_t len, loff_t off)
4983{
4984 struct inode *inode = sb_dqopt(sb)->files[type];
4985 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4986 int err = 0;
4987 int offset = off & (sb->s_blocksize - 1);
4988 struct buffer_head *bh;
4989 handle_t *handle = journal_current_handle();
4990
4991 if (EXT4_SB(sb)->s_journal && !handle) {
4992 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4993 " cancelled because transaction is not started",
4994 (unsigned long long)off, (unsigned long long)len);
4995 return -EIO;
4996 }
4997 /*
4998 * Since we account only one data block in transaction credits,
4999 * then it is impossible to cross a block boundary.
5000 */
5001 if (sb->s_blocksize - offset < len) {
5002 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5003 " cancelled because not block aligned",
5004 (unsigned long long)off, (unsigned long long)len);
5005 return -EIO;
5006 }
5007
5008 bh = ext4_bread(handle, inode, blk, 1, &err);
5009 if (!bh)
5010 goto out;
5011 err = ext4_journal_get_write_access(handle, bh);
5012 if (err) {
5013 brelse(bh);
5014 goto out;
5015 }
5016 lock_buffer(bh);
5017 memcpy(bh->b_data+offset, data, len);
5018 flush_dcache_page(bh->b_page);
5019 unlock_buffer(bh);
5020 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5021 brelse(bh);
5022out:
5023 if (err)
5024 return err;
5025 if (inode->i_size < off + len) {
5026 i_size_write(inode, off + len);
5027 EXT4_I(inode)->i_disksize = inode->i_size;
5028 ext4_mark_inode_dirty(handle, inode);
5029 }
5030 return len;
5031}
5032
5033#endif
5034
5035static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5036 const char *dev_name, void *data)
5037{
5038 return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5039}
5040
5041#if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5042static inline void register_as_ext2(void)
5043{
5044 int err = register_filesystem(&ext2_fs_type);
5045 if (err)
5046 printk(KERN_WARNING
5047 "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5048}
5049
5050static inline void unregister_as_ext2(void)
5051{
5052 unregister_filesystem(&ext2_fs_type);
5053}
5054
5055static inline int ext2_feature_set_ok(struct super_block *sb)
5056{
5057 if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5058 return 0;
5059 if (sb->s_flags & MS_RDONLY)
5060 return 1;
5061 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5062 return 0;
5063 return 1;
5064}
5065MODULE_ALIAS("ext2");
5066#else
5067static inline void register_as_ext2(void) { }
5068static inline void unregister_as_ext2(void) { }
5069static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5070#endif
5071
5072#if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5073static inline void register_as_ext3(void)
5074{
5075 int err = register_filesystem(&ext3_fs_type);
5076 if (err)
5077 printk(KERN_WARNING
5078 "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5079}
5080
5081static inline void unregister_as_ext3(void)
5082{
5083 unregister_filesystem(&ext3_fs_type);
5084}
5085
5086static inline int ext3_feature_set_ok(struct super_block *sb)
5087{
5088 if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5089 return 0;
5090 if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5091 return 0;
5092 if (sb->s_flags & MS_RDONLY)
5093 return 1;
5094 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5095 return 0;
5096 return 1;
5097}
5098MODULE_ALIAS("ext3");
5099#else
5100static inline void register_as_ext3(void) { }
5101static inline void unregister_as_ext3(void) { }
5102static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5103#endif
5104
5105static struct file_system_type ext4_fs_type = {
5106 .owner = THIS_MODULE,
5107 .name = "ext4",
5108 .mount = ext4_mount,
5109 .kill_sb = kill_block_super,
5110 .fs_flags = FS_REQUIRES_DEV,
5111};
5112
5113static int __init ext4_init_feat_adverts(void)
5114{
5115 struct ext4_features *ef;
5116 int ret = -ENOMEM;
5117
5118 ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5119 if (!ef)
5120 goto out;
5121
5122 ef->f_kobj.kset = ext4_kset;
5123 init_completion(&ef->f_kobj_unregister);
5124 ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5125 "features");
5126 if (ret) {
5127 kfree(ef);
5128 goto out;
5129 }
5130
5131 ext4_feat = ef;
5132 ret = 0;
5133out:
5134 return ret;
5135}
5136
5137static void ext4_exit_feat_adverts(void)
5138{
5139 kobject_put(&ext4_feat->f_kobj);
5140 wait_for_completion(&ext4_feat->f_kobj_unregister);
5141 kfree(ext4_feat);
5142}
5143
5144/* Shared across all ext4 file systems */
5145wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5146struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5147
5148static int __init ext4_init_fs(void)
5149{
5150 int i, err;
5151
5152 ext4_li_info = NULL;
5153 mutex_init(&ext4_li_mtx);
5154
5155 ext4_check_flag_values();
5156
5157 for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5158 mutex_init(&ext4__aio_mutex[i]);
5159 init_waitqueue_head(&ext4__ioend_wq[i]);
5160 }
5161
5162 err = ext4_init_pageio();
5163 if (err)
5164 return err;
5165 err = ext4_init_system_zone();
5166 if (err)
5167 goto out6;
5168 ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5169 if (!ext4_kset)
5170 goto out5;
5171 ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5172
5173 err = ext4_init_feat_adverts();
5174 if (err)
5175 goto out4;
5176
5177 err = ext4_init_mballoc();
5178 if (err)
5179 goto out3;
5180
5181 err = ext4_init_xattr();
5182 if (err)
5183 goto out2;
5184 err = init_inodecache();
5185 if (err)
5186 goto out1;
5187 register_as_ext3();
5188 register_as_ext2();
5189 err = register_filesystem(&ext4_fs_type);
5190 if (err)
5191 goto out;
5192
5193 return 0;
5194out:
5195 unregister_as_ext2();
5196 unregister_as_ext3();
5197 destroy_inodecache();
5198out1:
5199 ext4_exit_xattr();
5200out2:
5201 ext4_exit_mballoc();
5202out3:
5203 ext4_exit_feat_adverts();
5204out4:
5205 if (ext4_proc_root)
5206 remove_proc_entry("fs/ext4", NULL);
5207 kset_unregister(ext4_kset);
5208out5:
5209 ext4_exit_system_zone();
5210out6:
5211 ext4_exit_pageio();
5212 return err;
5213}
5214
5215static void __exit ext4_exit_fs(void)
5216{
5217 ext4_destroy_lazyinit_thread();
5218 unregister_as_ext2();
5219 unregister_as_ext3();
5220 unregister_filesystem(&ext4_fs_type);
5221 destroy_inodecache();
5222 ext4_exit_xattr();
5223 ext4_exit_mballoc();
5224 ext4_exit_feat_adverts();
5225 remove_proc_entry("fs/ext4", NULL);
5226 kset_unregister(ext4_kset);
5227 ext4_exit_system_zone();
5228 ext4_exit_pageio();
5229}
5230
5231MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5232MODULE_DESCRIPTION("Fourth Extended Filesystem");
5233MODULE_LICENSE("GPL");
5234module_init(ext4_init_fs)
5235module_exit(ext4_exit_fs)
1/*
2 * linux/fs/ext4/super.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/inode.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 */
18
19#include <linux/module.h>
20#include <linux/string.h>
21#include <linux/fs.h>
22#include <linux/time.h>
23#include <linux/vmalloc.h>
24#include <linux/slab.h>
25#include <linux/init.h>
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
28#include <linux/parser.h>
29#include <linux/buffer_head.h>
30#include <linux/exportfs.h>
31#include <linux/vfs.h>
32#include <linux/random.h>
33#include <linux/mount.h>
34#include <linux/namei.h>
35#include <linux/quotaops.h>
36#include <linux/seq_file.h>
37#include <linux/ctype.h>
38#include <linux/log2.h>
39#include <linux/crc16.h>
40#include <linux/cleancache.h>
41#include <asm/uaccess.h>
42
43#include <linux/kthread.h>
44#include <linux/freezer.h>
45
46#include "ext4.h"
47#include "ext4_extents.h" /* Needed for trace points definition */
48#include "ext4_jbd2.h"
49#include "xattr.h"
50#include "acl.h"
51#include "mballoc.h"
52
53#define CREATE_TRACE_POINTS
54#include <trace/events/ext4.h>
55
56static struct ext4_lazy_init *ext4_li_info;
57static struct mutex ext4_li_mtx;
58static struct ratelimit_state ext4_mount_msg_ratelimit;
59
60static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
61 unsigned long journal_devnum);
62static int ext4_show_options(struct seq_file *seq, struct dentry *root);
63static int ext4_commit_super(struct super_block *sb, int sync);
64static void ext4_mark_recovery_complete(struct super_block *sb,
65 struct ext4_super_block *es);
66static void ext4_clear_journal_err(struct super_block *sb,
67 struct ext4_super_block *es);
68static int ext4_sync_fs(struct super_block *sb, int wait);
69static int ext4_remount(struct super_block *sb, int *flags, char *data);
70static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
71static int ext4_unfreeze(struct super_block *sb);
72static int ext4_freeze(struct super_block *sb);
73static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
74 const char *dev_name, void *data);
75static inline int ext2_feature_set_ok(struct super_block *sb);
76static inline int ext3_feature_set_ok(struct super_block *sb);
77static int ext4_feature_set_ok(struct super_block *sb, int readonly);
78static void ext4_destroy_lazyinit_thread(void);
79static void ext4_unregister_li_request(struct super_block *sb);
80static void ext4_clear_request_list(void);
81
82/*
83 * Lock ordering
84 *
85 * Note the difference between i_mmap_sem (EXT4_I(inode)->i_mmap_sem) and
86 * i_mmap_rwsem (inode->i_mmap_rwsem)!
87 *
88 * page fault path:
89 * mmap_sem -> sb_start_pagefault -> i_mmap_sem (r) -> transaction start ->
90 * page lock -> i_data_sem (rw)
91 *
92 * buffered write path:
93 * sb_start_write -> i_mutex -> mmap_sem
94 * sb_start_write -> i_mutex -> transaction start -> page lock ->
95 * i_data_sem (rw)
96 *
97 * truncate:
98 * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
99 * i_mmap_rwsem (w) -> page lock
100 * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (w) -> i_mmap_sem (w) ->
101 * transaction start -> i_data_sem (rw)
102 *
103 * direct IO:
104 * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) -> mmap_sem
105 * sb_start_write -> i_mutex -> EXT4_STATE_DIOREAD_LOCK (r) ->
106 * transaction start -> i_data_sem (rw)
107 *
108 * writepages:
109 * transaction start -> page lock(s) -> i_data_sem (rw)
110 */
111
112#if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
113static struct file_system_type ext2_fs_type = {
114 .owner = THIS_MODULE,
115 .name = "ext2",
116 .mount = ext4_mount,
117 .kill_sb = kill_block_super,
118 .fs_flags = FS_REQUIRES_DEV,
119};
120MODULE_ALIAS_FS("ext2");
121MODULE_ALIAS("ext2");
122#define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
123#else
124#define IS_EXT2_SB(sb) (0)
125#endif
126
127
128static struct file_system_type ext3_fs_type = {
129 .owner = THIS_MODULE,
130 .name = "ext3",
131 .mount = ext4_mount,
132 .kill_sb = kill_block_super,
133 .fs_flags = FS_REQUIRES_DEV,
134};
135MODULE_ALIAS_FS("ext3");
136MODULE_ALIAS("ext3");
137#define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
138
139static int ext4_verify_csum_type(struct super_block *sb,
140 struct ext4_super_block *es)
141{
142 if (!ext4_has_feature_metadata_csum(sb))
143 return 1;
144
145 return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
146}
147
148static __le32 ext4_superblock_csum(struct super_block *sb,
149 struct ext4_super_block *es)
150{
151 struct ext4_sb_info *sbi = EXT4_SB(sb);
152 int offset = offsetof(struct ext4_super_block, s_checksum);
153 __u32 csum;
154
155 csum = ext4_chksum(sbi, ~0, (char *)es, offset);
156
157 return cpu_to_le32(csum);
158}
159
160static int ext4_superblock_csum_verify(struct super_block *sb,
161 struct ext4_super_block *es)
162{
163 if (!ext4_has_metadata_csum(sb))
164 return 1;
165
166 return es->s_checksum == ext4_superblock_csum(sb, es);
167}
168
169void ext4_superblock_csum_set(struct super_block *sb)
170{
171 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
172
173 if (!ext4_has_metadata_csum(sb))
174 return;
175
176 es->s_checksum = ext4_superblock_csum(sb, es);
177}
178
179void *ext4_kvmalloc(size_t size, gfp_t flags)
180{
181 void *ret;
182
183 ret = kmalloc(size, flags | __GFP_NOWARN);
184 if (!ret)
185 ret = __vmalloc(size, flags, PAGE_KERNEL);
186 return ret;
187}
188
189void *ext4_kvzalloc(size_t size, gfp_t flags)
190{
191 void *ret;
192
193 ret = kzalloc(size, flags | __GFP_NOWARN);
194 if (!ret)
195 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
196 return ret;
197}
198
199ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
200 struct ext4_group_desc *bg)
201{
202 return le32_to_cpu(bg->bg_block_bitmap_lo) |
203 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
204 (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
205}
206
207ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
208 struct ext4_group_desc *bg)
209{
210 return le32_to_cpu(bg->bg_inode_bitmap_lo) |
211 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
212 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
213}
214
215ext4_fsblk_t ext4_inode_table(struct super_block *sb,
216 struct ext4_group_desc *bg)
217{
218 return le32_to_cpu(bg->bg_inode_table_lo) |
219 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
220 (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
221}
222
223__u32 ext4_free_group_clusters(struct super_block *sb,
224 struct ext4_group_desc *bg)
225{
226 return le16_to_cpu(bg->bg_free_blocks_count_lo) |
227 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
228 (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
229}
230
231__u32 ext4_free_inodes_count(struct super_block *sb,
232 struct ext4_group_desc *bg)
233{
234 return le16_to_cpu(bg->bg_free_inodes_count_lo) |
235 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
236 (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
237}
238
239__u32 ext4_used_dirs_count(struct super_block *sb,
240 struct ext4_group_desc *bg)
241{
242 return le16_to_cpu(bg->bg_used_dirs_count_lo) |
243 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
244 (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
245}
246
247__u32 ext4_itable_unused_count(struct super_block *sb,
248 struct ext4_group_desc *bg)
249{
250 return le16_to_cpu(bg->bg_itable_unused_lo) |
251 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
252 (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
253}
254
255void ext4_block_bitmap_set(struct super_block *sb,
256 struct ext4_group_desc *bg, ext4_fsblk_t blk)
257{
258 bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
259 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
260 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
261}
262
263void ext4_inode_bitmap_set(struct super_block *sb,
264 struct ext4_group_desc *bg, ext4_fsblk_t blk)
265{
266 bg->bg_inode_bitmap_lo = cpu_to_le32((u32)blk);
267 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
268 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
269}
270
271void ext4_inode_table_set(struct super_block *sb,
272 struct ext4_group_desc *bg, ext4_fsblk_t blk)
273{
274 bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
275 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
276 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
277}
278
279void ext4_free_group_clusters_set(struct super_block *sb,
280 struct ext4_group_desc *bg, __u32 count)
281{
282 bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
283 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
284 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
285}
286
287void ext4_free_inodes_set(struct super_block *sb,
288 struct ext4_group_desc *bg, __u32 count)
289{
290 bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
291 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
292 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
293}
294
295void ext4_used_dirs_set(struct super_block *sb,
296 struct ext4_group_desc *bg, __u32 count)
297{
298 bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
299 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
300 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
301}
302
303void ext4_itable_unused_set(struct super_block *sb,
304 struct ext4_group_desc *bg, __u32 count)
305{
306 bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
307 if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
308 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
309}
310
311
312static void __save_error_info(struct super_block *sb, const char *func,
313 unsigned int line)
314{
315 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
316
317 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
318 if (bdev_read_only(sb->s_bdev))
319 return;
320 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
321 es->s_last_error_time = cpu_to_le32(get_seconds());
322 strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
323 es->s_last_error_line = cpu_to_le32(line);
324 if (!es->s_first_error_time) {
325 es->s_first_error_time = es->s_last_error_time;
326 strncpy(es->s_first_error_func, func,
327 sizeof(es->s_first_error_func));
328 es->s_first_error_line = cpu_to_le32(line);
329 es->s_first_error_ino = es->s_last_error_ino;
330 es->s_first_error_block = es->s_last_error_block;
331 }
332 /*
333 * Start the daily error reporting function if it hasn't been
334 * started already
335 */
336 if (!es->s_error_count)
337 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
338 le32_add_cpu(&es->s_error_count, 1);
339}
340
341static void save_error_info(struct super_block *sb, const char *func,
342 unsigned int line)
343{
344 __save_error_info(sb, func, line);
345 ext4_commit_super(sb, 1);
346}
347
348/*
349 * The del_gendisk() function uninitializes the disk-specific data
350 * structures, including the bdi structure, without telling anyone
351 * else. Once this happens, any attempt to call mark_buffer_dirty()
352 * (for example, by ext4_commit_super), will cause a kernel OOPS.
353 * This is a kludge to prevent these oops until we can put in a proper
354 * hook in del_gendisk() to inform the VFS and file system layers.
355 */
356static int block_device_ejected(struct super_block *sb)
357{
358 struct inode *bd_inode = sb->s_bdev->bd_inode;
359 struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
360
361 return bdi->dev == NULL;
362}
363
364static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
365{
366 struct super_block *sb = journal->j_private;
367 struct ext4_sb_info *sbi = EXT4_SB(sb);
368 int error = is_journal_aborted(journal);
369 struct ext4_journal_cb_entry *jce;
370
371 BUG_ON(txn->t_state == T_FINISHED);
372 spin_lock(&sbi->s_md_lock);
373 while (!list_empty(&txn->t_private_list)) {
374 jce = list_entry(txn->t_private_list.next,
375 struct ext4_journal_cb_entry, jce_list);
376 list_del_init(&jce->jce_list);
377 spin_unlock(&sbi->s_md_lock);
378 jce->jce_func(sb, jce, error);
379 spin_lock(&sbi->s_md_lock);
380 }
381 spin_unlock(&sbi->s_md_lock);
382}
383
384/* Deal with the reporting of failure conditions on a filesystem such as
385 * inconsistencies detected or read IO failures.
386 *
387 * On ext2, we can store the error state of the filesystem in the
388 * superblock. That is not possible on ext4, because we may have other
389 * write ordering constraints on the superblock which prevent us from
390 * writing it out straight away; and given that the journal is about to
391 * be aborted, we can't rely on the current, or future, transactions to
392 * write out the superblock safely.
393 *
394 * We'll just use the jbd2_journal_abort() error code to record an error in
395 * the journal instead. On recovery, the journal will complain about
396 * that error until we've noted it down and cleared it.
397 */
398
399static void ext4_handle_error(struct super_block *sb)
400{
401 if (sb->s_flags & MS_RDONLY)
402 return;
403
404 if (!test_opt(sb, ERRORS_CONT)) {
405 journal_t *journal = EXT4_SB(sb)->s_journal;
406
407 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
408 if (journal)
409 jbd2_journal_abort(journal, -EIO);
410 }
411 if (test_opt(sb, ERRORS_RO)) {
412 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
413 /*
414 * Make sure updated value of ->s_mount_flags will be visible
415 * before ->s_flags update
416 */
417 smp_wmb();
418 sb->s_flags |= MS_RDONLY;
419 }
420 if (test_opt(sb, ERRORS_PANIC)) {
421 if (EXT4_SB(sb)->s_journal &&
422 !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
423 return;
424 panic("EXT4-fs (device %s): panic forced after error\n",
425 sb->s_id);
426 }
427}
428
429#define ext4_error_ratelimit(sb) \
430 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state), \
431 "EXT4-fs error")
432
433void __ext4_error(struct super_block *sb, const char *function,
434 unsigned int line, const char *fmt, ...)
435{
436 struct va_format vaf;
437 va_list args;
438
439 if (ext4_error_ratelimit(sb)) {
440 va_start(args, fmt);
441 vaf.fmt = fmt;
442 vaf.va = &args;
443 printk(KERN_CRIT
444 "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
445 sb->s_id, function, line, current->comm, &vaf);
446 va_end(args);
447 }
448 save_error_info(sb, function, line);
449 ext4_handle_error(sb);
450}
451
452void __ext4_error_inode(struct inode *inode, const char *function,
453 unsigned int line, ext4_fsblk_t block,
454 const char *fmt, ...)
455{
456 va_list args;
457 struct va_format vaf;
458 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
459
460 es->s_last_error_ino = cpu_to_le32(inode->i_ino);
461 es->s_last_error_block = cpu_to_le64(block);
462 if (ext4_error_ratelimit(inode->i_sb)) {
463 va_start(args, fmt);
464 vaf.fmt = fmt;
465 vaf.va = &args;
466 if (block)
467 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
468 "inode #%lu: block %llu: comm %s: %pV\n",
469 inode->i_sb->s_id, function, line, inode->i_ino,
470 block, current->comm, &vaf);
471 else
472 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
473 "inode #%lu: comm %s: %pV\n",
474 inode->i_sb->s_id, function, line, inode->i_ino,
475 current->comm, &vaf);
476 va_end(args);
477 }
478 save_error_info(inode->i_sb, function, line);
479 ext4_handle_error(inode->i_sb);
480}
481
482void __ext4_error_file(struct file *file, const char *function,
483 unsigned int line, ext4_fsblk_t block,
484 const char *fmt, ...)
485{
486 va_list args;
487 struct va_format vaf;
488 struct ext4_super_block *es;
489 struct inode *inode = file_inode(file);
490 char pathname[80], *path;
491
492 es = EXT4_SB(inode->i_sb)->s_es;
493 es->s_last_error_ino = cpu_to_le32(inode->i_ino);
494 if (ext4_error_ratelimit(inode->i_sb)) {
495 path = file_path(file, pathname, sizeof(pathname));
496 if (IS_ERR(path))
497 path = "(unknown)";
498 va_start(args, fmt);
499 vaf.fmt = fmt;
500 vaf.va = &args;
501 if (block)
502 printk(KERN_CRIT
503 "EXT4-fs error (device %s): %s:%d: inode #%lu: "
504 "block %llu: comm %s: path %s: %pV\n",
505 inode->i_sb->s_id, function, line, inode->i_ino,
506 block, current->comm, path, &vaf);
507 else
508 printk(KERN_CRIT
509 "EXT4-fs error (device %s): %s:%d: inode #%lu: "
510 "comm %s: path %s: %pV\n",
511 inode->i_sb->s_id, function, line, inode->i_ino,
512 current->comm, path, &vaf);
513 va_end(args);
514 }
515 save_error_info(inode->i_sb, function, line);
516 ext4_handle_error(inode->i_sb);
517}
518
519const char *ext4_decode_error(struct super_block *sb, int errno,
520 char nbuf[16])
521{
522 char *errstr = NULL;
523
524 switch (errno) {
525 case -EFSCORRUPTED:
526 errstr = "Corrupt filesystem";
527 break;
528 case -EFSBADCRC:
529 errstr = "Filesystem failed CRC";
530 break;
531 case -EIO:
532 errstr = "IO failure";
533 break;
534 case -ENOMEM:
535 errstr = "Out of memory";
536 break;
537 case -EROFS:
538 if (!sb || (EXT4_SB(sb)->s_journal &&
539 EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
540 errstr = "Journal has aborted";
541 else
542 errstr = "Readonly filesystem";
543 break;
544 default:
545 /* If the caller passed in an extra buffer for unknown
546 * errors, textualise them now. Else we just return
547 * NULL. */
548 if (nbuf) {
549 /* Check for truncated error codes... */
550 if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
551 errstr = nbuf;
552 }
553 break;
554 }
555
556 return errstr;
557}
558
559/* __ext4_std_error decodes expected errors from journaling functions
560 * automatically and invokes the appropriate error response. */
561
562void __ext4_std_error(struct super_block *sb, const char *function,
563 unsigned int line, int errno)
564{
565 char nbuf[16];
566 const char *errstr;
567
568 /* Special case: if the error is EROFS, and we're not already
569 * inside a transaction, then there's really no point in logging
570 * an error. */
571 if (errno == -EROFS && journal_current_handle() == NULL &&
572 (sb->s_flags & MS_RDONLY))
573 return;
574
575 if (ext4_error_ratelimit(sb)) {
576 errstr = ext4_decode_error(sb, errno, nbuf);
577 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
578 sb->s_id, function, line, errstr);
579 }
580
581 save_error_info(sb, function, line);
582 ext4_handle_error(sb);
583}
584
585/*
586 * ext4_abort is a much stronger failure handler than ext4_error. The
587 * abort function may be used to deal with unrecoverable failures such
588 * as journal IO errors or ENOMEM at a critical moment in log management.
589 *
590 * We unconditionally force the filesystem into an ABORT|READONLY state,
591 * unless the error response on the fs has been set to panic in which
592 * case we take the easy way out and panic immediately.
593 */
594
595void __ext4_abort(struct super_block *sb, const char *function,
596 unsigned int line, const char *fmt, ...)
597{
598 va_list args;
599
600 save_error_info(sb, function, line);
601 va_start(args, fmt);
602 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
603 function, line);
604 vprintk(fmt, args);
605 printk("\n");
606 va_end(args);
607
608 if ((sb->s_flags & MS_RDONLY) == 0) {
609 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
610 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
611 /*
612 * Make sure updated value of ->s_mount_flags will be visible
613 * before ->s_flags update
614 */
615 smp_wmb();
616 sb->s_flags |= MS_RDONLY;
617 if (EXT4_SB(sb)->s_journal)
618 jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
619 save_error_info(sb, function, line);
620 }
621 if (test_opt(sb, ERRORS_PANIC)) {
622 if (EXT4_SB(sb)->s_journal &&
623 !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
624 return;
625 panic("EXT4-fs panic from previous error\n");
626 }
627}
628
629void __ext4_msg(struct super_block *sb,
630 const char *prefix, const char *fmt, ...)
631{
632 struct va_format vaf;
633 va_list args;
634
635 if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
636 return;
637
638 va_start(args, fmt);
639 vaf.fmt = fmt;
640 vaf.va = &args;
641 printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
642 va_end(args);
643}
644
645#define ext4_warning_ratelimit(sb) \
646 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
647 "EXT4-fs warning")
648
649void __ext4_warning(struct super_block *sb, const char *function,
650 unsigned int line, const char *fmt, ...)
651{
652 struct va_format vaf;
653 va_list args;
654
655 if (!ext4_warning_ratelimit(sb))
656 return;
657
658 va_start(args, fmt);
659 vaf.fmt = fmt;
660 vaf.va = &args;
661 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
662 sb->s_id, function, line, &vaf);
663 va_end(args);
664}
665
666void __ext4_warning_inode(const struct inode *inode, const char *function,
667 unsigned int line, const char *fmt, ...)
668{
669 struct va_format vaf;
670 va_list args;
671
672 if (!ext4_warning_ratelimit(inode->i_sb))
673 return;
674
675 va_start(args, fmt);
676 vaf.fmt = fmt;
677 vaf.va = &args;
678 printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
679 "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
680 function, line, inode->i_ino, current->comm, &vaf);
681 va_end(args);
682}
683
684void __ext4_grp_locked_error(const char *function, unsigned int line,
685 struct super_block *sb, ext4_group_t grp,
686 unsigned long ino, ext4_fsblk_t block,
687 const char *fmt, ...)
688__releases(bitlock)
689__acquires(bitlock)
690{
691 struct va_format vaf;
692 va_list args;
693 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
694
695 es->s_last_error_ino = cpu_to_le32(ino);
696 es->s_last_error_block = cpu_to_le64(block);
697 __save_error_info(sb, function, line);
698
699 if (ext4_error_ratelimit(sb)) {
700 va_start(args, fmt);
701 vaf.fmt = fmt;
702 vaf.va = &args;
703 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
704 sb->s_id, function, line, grp);
705 if (ino)
706 printk(KERN_CONT "inode %lu: ", ino);
707 if (block)
708 printk(KERN_CONT "block %llu:",
709 (unsigned long long) block);
710 printk(KERN_CONT "%pV\n", &vaf);
711 va_end(args);
712 }
713
714 if (test_opt(sb, ERRORS_CONT)) {
715 ext4_commit_super(sb, 0);
716 return;
717 }
718
719 ext4_unlock_group(sb, grp);
720 ext4_handle_error(sb);
721 /*
722 * We only get here in the ERRORS_RO case; relocking the group
723 * may be dangerous, but nothing bad will happen since the
724 * filesystem will have already been marked read/only and the
725 * journal has been aborted. We return 1 as a hint to callers
726 * who might what to use the return value from
727 * ext4_grp_locked_error() to distinguish between the
728 * ERRORS_CONT and ERRORS_RO case, and perhaps return more
729 * aggressively from the ext4 function in question, with a
730 * more appropriate error code.
731 */
732 ext4_lock_group(sb, grp);
733 return;
734}
735
736void ext4_update_dynamic_rev(struct super_block *sb)
737{
738 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
739
740 if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
741 return;
742
743 ext4_warning(sb,
744 "updating to rev %d because of new feature flag, "
745 "running e2fsck is recommended",
746 EXT4_DYNAMIC_REV);
747
748 es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
749 es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
750 es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
751 /* leave es->s_feature_*compat flags alone */
752 /* es->s_uuid will be set by e2fsck if empty */
753
754 /*
755 * The rest of the superblock fields should be zero, and if not it
756 * means they are likely already in use, so leave them alone. We
757 * can leave it up to e2fsck to clean up any inconsistencies there.
758 */
759}
760
761/*
762 * Open the external journal device
763 */
764static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
765{
766 struct block_device *bdev;
767 char b[BDEVNAME_SIZE];
768
769 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
770 if (IS_ERR(bdev))
771 goto fail;
772 return bdev;
773
774fail:
775 ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
776 __bdevname(dev, b), PTR_ERR(bdev));
777 return NULL;
778}
779
780/*
781 * Release the journal device
782 */
783static void ext4_blkdev_put(struct block_device *bdev)
784{
785 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
786}
787
788static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
789{
790 struct block_device *bdev;
791 bdev = sbi->journal_bdev;
792 if (bdev) {
793 ext4_blkdev_put(bdev);
794 sbi->journal_bdev = NULL;
795 }
796}
797
798static inline struct inode *orphan_list_entry(struct list_head *l)
799{
800 return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
801}
802
803static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
804{
805 struct list_head *l;
806
807 ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
808 le32_to_cpu(sbi->s_es->s_last_orphan));
809
810 printk(KERN_ERR "sb_info orphan list:\n");
811 list_for_each(l, &sbi->s_orphan) {
812 struct inode *inode = orphan_list_entry(l);
813 printk(KERN_ERR " "
814 "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
815 inode->i_sb->s_id, inode->i_ino, inode,
816 inode->i_mode, inode->i_nlink,
817 NEXT_ORPHAN(inode));
818 }
819}
820
821static void ext4_put_super(struct super_block *sb)
822{
823 struct ext4_sb_info *sbi = EXT4_SB(sb);
824 struct ext4_super_block *es = sbi->s_es;
825 int i, err;
826
827 ext4_unregister_li_request(sb);
828 dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
829
830 flush_workqueue(sbi->rsv_conversion_wq);
831 destroy_workqueue(sbi->rsv_conversion_wq);
832
833 if (sbi->s_journal) {
834 err = jbd2_journal_destroy(sbi->s_journal);
835 sbi->s_journal = NULL;
836 if (err < 0)
837 ext4_abort(sb, "Couldn't clean up the journal");
838 }
839
840 ext4_unregister_sysfs(sb);
841 ext4_es_unregister_shrinker(sbi);
842 del_timer_sync(&sbi->s_err_report);
843 ext4_release_system_zone(sb);
844 ext4_mb_release(sb);
845 ext4_ext_release(sb);
846
847 if (!(sb->s_flags & MS_RDONLY)) {
848 ext4_clear_feature_journal_needs_recovery(sb);
849 es->s_state = cpu_to_le16(sbi->s_mount_state);
850 }
851 if (!(sb->s_flags & MS_RDONLY))
852 ext4_commit_super(sb, 1);
853
854 for (i = 0; i < sbi->s_gdb_count; i++)
855 brelse(sbi->s_group_desc[i]);
856 kvfree(sbi->s_group_desc);
857 kvfree(sbi->s_flex_groups);
858 percpu_counter_destroy(&sbi->s_freeclusters_counter);
859 percpu_counter_destroy(&sbi->s_freeinodes_counter);
860 percpu_counter_destroy(&sbi->s_dirs_counter);
861 percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
862 brelse(sbi->s_sbh);
863#ifdef CONFIG_QUOTA
864 for (i = 0; i < EXT4_MAXQUOTAS; i++)
865 kfree(sbi->s_qf_names[i]);
866#endif
867
868 /* Debugging code just in case the in-memory inode orphan list
869 * isn't empty. The on-disk one can be non-empty if we've
870 * detected an error and taken the fs readonly, but the
871 * in-memory list had better be clean by this point. */
872 if (!list_empty(&sbi->s_orphan))
873 dump_orphan_list(sb, sbi);
874 J_ASSERT(list_empty(&sbi->s_orphan));
875
876 sync_blockdev(sb->s_bdev);
877 invalidate_bdev(sb->s_bdev);
878 if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
879 /*
880 * Invalidate the journal device's buffers. We don't want them
881 * floating about in memory - the physical journal device may
882 * hotswapped, and it breaks the `ro-after' testing code.
883 */
884 sync_blockdev(sbi->journal_bdev);
885 invalidate_bdev(sbi->journal_bdev);
886 ext4_blkdev_remove(sbi);
887 }
888 if (sbi->s_mb_cache) {
889 ext4_xattr_destroy_cache(sbi->s_mb_cache);
890 sbi->s_mb_cache = NULL;
891 }
892 if (sbi->s_mmp_tsk)
893 kthread_stop(sbi->s_mmp_tsk);
894 sb->s_fs_info = NULL;
895 /*
896 * Now that we are completely done shutting down the
897 * superblock, we need to actually destroy the kobject.
898 */
899 kobject_put(&sbi->s_kobj);
900 wait_for_completion(&sbi->s_kobj_unregister);
901 if (sbi->s_chksum_driver)
902 crypto_free_shash(sbi->s_chksum_driver);
903 kfree(sbi->s_blockgroup_lock);
904 kfree(sbi);
905}
906
907static struct kmem_cache *ext4_inode_cachep;
908
909/*
910 * Called inside transaction, so use GFP_NOFS
911 */
912static struct inode *ext4_alloc_inode(struct super_block *sb)
913{
914 struct ext4_inode_info *ei;
915
916 ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
917 if (!ei)
918 return NULL;
919
920 ei->vfs_inode.i_version = 1;
921 spin_lock_init(&ei->i_raw_lock);
922 INIT_LIST_HEAD(&ei->i_prealloc_list);
923 spin_lock_init(&ei->i_prealloc_lock);
924 ext4_es_init_tree(&ei->i_es_tree);
925 rwlock_init(&ei->i_es_lock);
926 INIT_LIST_HEAD(&ei->i_es_list);
927 ei->i_es_all_nr = 0;
928 ei->i_es_shk_nr = 0;
929 ei->i_es_shrink_lblk = 0;
930 ei->i_reserved_data_blocks = 0;
931 ei->i_reserved_meta_blocks = 0;
932 ei->i_allocated_meta_blocks = 0;
933 ei->i_da_metadata_calc_len = 0;
934 ei->i_da_metadata_calc_last_lblock = 0;
935 spin_lock_init(&(ei->i_block_reservation_lock));
936#ifdef CONFIG_QUOTA
937 ei->i_reserved_quota = 0;
938 memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
939#endif
940 ei->jinode = NULL;
941 INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
942 spin_lock_init(&ei->i_completed_io_lock);
943 ei->i_sync_tid = 0;
944 ei->i_datasync_tid = 0;
945 atomic_set(&ei->i_unwritten, 0);
946 INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
947#ifdef CONFIG_EXT4_FS_ENCRYPTION
948 ei->i_crypt_info = NULL;
949#endif
950 return &ei->vfs_inode;
951}
952
953static int ext4_drop_inode(struct inode *inode)
954{
955 int drop = generic_drop_inode(inode);
956
957 trace_ext4_drop_inode(inode, drop);
958 return drop;
959}
960
961static void ext4_i_callback(struct rcu_head *head)
962{
963 struct inode *inode = container_of(head, struct inode, i_rcu);
964 kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
965}
966
967static void ext4_destroy_inode(struct inode *inode)
968{
969 if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
970 ext4_msg(inode->i_sb, KERN_ERR,
971 "Inode %lu (%p): orphan list check failed!",
972 inode->i_ino, EXT4_I(inode));
973 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
974 EXT4_I(inode), sizeof(struct ext4_inode_info),
975 true);
976 dump_stack();
977 }
978 call_rcu(&inode->i_rcu, ext4_i_callback);
979}
980
981static void init_once(void *foo)
982{
983 struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
984
985 INIT_LIST_HEAD(&ei->i_orphan);
986 init_rwsem(&ei->xattr_sem);
987 init_rwsem(&ei->i_data_sem);
988 init_rwsem(&ei->i_mmap_sem);
989 inode_init_once(&ei->vfs_inode);
990}
991
992static int __init init_inodecache(void)
993{
994 ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
995 sizeof(struct ext4_inode_info),
996 0, (SLAB_RECLAIM_ACCOUNT|
997 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
998 init_once);
999 if (ext4_inode_cachep == NULL)
1000 return -ENOMEM;
1001 return 0;
1002}
1003
1004static void destroy_inodecache(void)
1005{
1006 /*
1007 * Make sure all delayed rcu free inodes are flushed before we
1008 * destroy cache.
1009 */
1010 rcu_barrier();
1011 kmem_cache_destroy(ext4_inode_cachep);
1012}
1013
1014void ext4_clear_inode(struct inode *inode)
1015{
1016 invalidate_inode_buffers(inode);
1017 clear_inode(inode);
1018 dquot_drop(inode);
1019 ext4_discard_preallocations(inode);
1020 ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1021 if (EXT4_I(inode)->jinode) {
1022 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1023 EXT4_I(inode)->jinode);
1024 jbd2_free_inode(EXT4_I(inode)->jinode);
1025 EXT4_I(inode)->jinode = NULL;
1026 }
1027#ifdef CONFIG_EXT4_FS_ENCRYPTION
1028 if (EXT4_I(inode)->i_crypt_info)
1029 ext4_free_encryption_info(inode, EXT4_I(inode)->i_crypt_info);
1030#endif
1031}
1032
1033static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1034 u64 ino, u32 generation)
1035{
1036 struct inode *inode;
1037
1038 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1039 return ERR_PTR(-ESTALE);
1040 if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1041 return ERR_PTR(-ESTALE);
1042
1043 /* iget isn't really right if the inode is currently unallocated!!
1044 *
1045 * ext4_read_inode will return a bad_inode if the inode had been
1046 * deleted, so we should be safe.
1047 *
1048 * Currently we don't know the generation for parent directory, so
1049 * a generation of 0 means "accept any"
1050 */
1051 inode = ext4_iget_normal(sb, ino);
1052 if (IS_ERR(inode))
1053 return ERR_CAST(inode);
1054 if (generation && inode->i_generation != generation) {
1055 iput(inode);
1056 return ERR_PTR(-ESTALE);
1057 }
1058
1059 return inode;
1060}
1061
1062static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1063 int fh_len, int fh_type)
1064{
1065 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1066 ext4_nfs_get_inode);
1067}
1068
1069static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1070 int fh_len, int fh_type)
1071{
1072 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1073 ext4_nfs_get_inode);
1074}
1075
1076/*
1077 * Try to release metadata pages (indirect blocks, directories) which are
1078 * mapped via the block device. Since these pages could have journal heads
1079 * which would prevent try_to_free_buffers() from freeing them, we must use
1080 * jbd2 layer's try_to_free_buffers() function to release them.
1081 */
1082static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1083 gfp_t wait)
1084{
1085 journal_t *journal = EXT4_SB(sb)->s_journal;
1086
1087 WARN_ON(PageChecked(page));
1088 if (!page_has_buffers(page))
1089 return 0;
1090 if (journal)
1091 return jbd2_journal_try_to_free_buffers(journal, page,
1092 wait & ~__GFP_DIRECT_RECLAIM);
1093 return try_to_free_buffers(page);
1094}
1095
1096#ifdef CONFIG_QUOTA
1097static char *quotatypes[] = INITQFNAMES;
1098#define QTYPE2NAME(t) (quotatypes[t])
1099
1100static int ext4_write_dquot(struct dquot *dquot);
1101static int ext4_acquire_dquot(struct dquot *dquot);
1102static int ext4_release_dquot(struct dquot *dquot);
1103static int ext4_mark_dquot_dirty(struct dquot *dquot);
1104static int ext4_write_info(struct super_block *sb, int type);
1105static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1106 struct path *path);
1107static int ext4_quota_off(struct super_block *sb, int type);
1108static int ext4_quota_on_mount(struct super_block *sb, int type);
1109static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1110 size_t len, loff_t off);
1111static ssize_t ext4_quota_write(struct super_block *sb, int type,
1112 const char *data, size_t len, loff_t off);
1113static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1114 unsigned int flags);
1115static int ext4_enable_quotas(struct super_block *sb);
1116static int ext4_get_next_id(struct super_block *sb, struct kqid *qid);
1117
1118static struct dquot **ext4_get_dquots(struct inode *inode)
1119{
1120 return EXT4_I(inode)->i_dquot;
1121}
1122
1123static const struct dquot_operations ext4_quota_operations = {
1124 .get_reserved_space = ext4_get_reserved_space,
1125 .write_dquot = ext4_write_dquot,
1126 .acquire_dquot = ext4_acquire_dquot,
1127 .release_dquot = ext4_release_dquot,
1128 .mark_dirty = ext4_mark_dquot_dirty,
1129 .write_info = ext4_write_info,
1130 .alloc_dquot = dquot_alloc,
1131 .destroy_dquot = dquot_destroy,
1132 .get_projid = ext4_get_projid,
1133 .get_next_id = ext4_get_next_id,
1134};
1135
1136static const struct quotactl_ops ext4_qctl_operations = {
1137 .quota_on = ext4_quota_on,
1138 .quota_off = ext4_quota_off,
1139 .quota_sync = dquot_quota_sync,
1140 .get_state = dquot_get_state,
1141 .set_info = dquot_set_dqinfo,
1142 .get_dqblk = dquot_get_dqblk,
1143 .set_dqblk = dquot_set_dqblk,
1144 .get_nextdqblk = dquot_get_next_dqblk,
1145};
1146#endif
1147
1148static const struct super_operations ext4_sops = {
1149 .alloc_inode = ext4_alloc_inode,
1150 .destroy_inode = ext4_destroy_inode,
1151 .write_inode = ext4_write_inode,
1152 .dirty_inode = ext4_dirty_inode,
1153 .drop_inode = ext4_drop_inode,
1154 .evict_inode = ext4_evict_inode,
1155 .put_super = ext4_put_super,
1156 .sync_fs = ext4_sync_fs,
1157 .freeze_fs = ext4_freeze,
1158 .unfreeze_fs = ext4_unfreeze,
1159 .statfs = ext4_statfs,
1160 .remount_fs = ext4_remount,
1161 .show_options = ext4_show_options,
1162#ifdef CONFIG_QUOTA
1163 .quota_read = ext4_quota_read,
1164 .quota_write = ext4_quota_write,
1165 .get_dquots = ext4_get_dquots,
1166#endif
1167 .bdev_try_to_free_page = bdev_try_to_free_page,
1168};
1169
1170static const struct export_operations ext4_export_ops = {
1171 .fh_to_dentry = ext4_fh_to_dentry,
1172 .fh_to_parent = ext4_fh_to_parent,
1173 .get_parent = ext4_get_parent,
1174};
1175
1176enum {
1177 Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1178 Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1179 Opt_nouid32, Opt_debug, Opt_removed,
1180 Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1181 Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1182 Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1183 Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1184 Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1185 Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1186 Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1187 Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1188 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1189 Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
1190 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1191 Opt_lazytime, Opt_nolazytime,
1192 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1193 Opt_inode_readahead_blks, Opt_journal_ioprio,
1194 Opt_dioread_nolock, Opt_dioread_lock,
1195 Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1196 Opt_max_dir_size_kb, Opt_nojournal_checksum,
1197};
1198
1199static const match_table_t tokens = {
1200 {Opt_bsd_df, "bsddf"},
1201 {Opt_minix_df, "minixdf"},
1202 {Opt_grpid, "grpid"},
1203 {Opt_grpid, "bsdgroups"},
1204 {Opt_nogrpid, "nogrpid"},
1205 {Opt_nogrpid, "sysvgroups"},
1206 {Opt_resgid, "resgid=%u"},
1207 {Opt_resuid, "resuid=%u"},
1208 {Opt_sb, "sb=%u"},
1209 {Opt_err_cont, "errors=continue"},
1210 {Opt_err_panic, "errors=panic"},
1211 {Opt_err_ro, "errors=remount-ro"},
1212 {Opt_nouid32, "nouid32"},
1213 {Opt_debug, "debug"},
1214 {Opt_removed, "oldalloc"},
1215 {Opt_removed, "orlov"},
1216 {Opt_user_xattr, "user_xattr"},
1217 {Opt_nouser_xattr, "nouser_xattr"},
1218 {Opt_acl, "acl"},
1219 {Opt_noacl, "noacl"},
1220 {Opt_noload, "norecovery"},
1221 {Opt_noload, "noload"},
1222 {Opt_removed, "nobh"},
1223 {Opt_removed, "bh"},
1224 {Opt_commit, "commit=%u"},
1225 {Opt_min_batch_time, "min_batch_time=%u"},
1226 {Opt_max_batch_time, "max_batch_time=%u"},
1227 {Opt_journal_dev, "journal_dev=%u"},
1228 {Opt_journal_path, "journal_path=%s"},
1229 {Opt_journal_checksum, "journal_checksum"},
1230 {Opt_nojournal_checksum, "nojournal_checksum"},
1231 {Opt_journal_async_commit, "journal_async_commit"},
1232 {Opt_abort, "abort"},
1233 {Opt_data_journal, "data=journal"},
1234 {Opt_data_ordered, "data=ordered"},
1235 {Opt_data_writeback, "data=writeback"},
1236 {Opt_data_err_abort, "data_err=abort"},
1237 {Opt_data_err_ignore, "data_err=ignore"},
1238 {Opt_offusrjquota, "usrjquota="},
1239 {Opt_usrjquota, "usrjquota=%s"},
1240 {Opt_offgrpjquota, "grpjquota="},
1241 {Opt_grpjquota, "grpjquota=%s"},
1242 {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1243 {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1244 {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1245 {Opt_grpquota, "grpquota"},
1246 {Opt_noquota, "noquota"},
1247 {Opt_quota, "quota"},
1248 {Opt_usrquota, "usrquota"},
1249 {Opt_barrier, "barrier=%u"},
1250 {Opt_barrier, "barrier"},
1251 {Opt_nobarrier, "nobarrier"},
1252 {Opt_i_version, "i_version"},
1253 {Opt_dax, "dax"},
1254 {Opt_stripe, "stripe=%u"},
1255 {Opt_delalloc, "delalloc"},
1256 {Opt_lazytime, "lazytime"},
1257 {Opt_nolazytime, "nolazytime"},
1258 {Opt_nodelalloc, "nodelalloc"},
1259 {Opt_removed, "mblk_io_submit"},
1260 {Opt_removed, "nomblk_io_submit"},
1261 {Opt_block_validity, "block_validity"},
1262 {Opt_noblock_validity, "noblock_validity"},
1263 {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1264 {Opt_journal_ioprio, "journal_ioprio=%u"},
1265 {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1266 {Opt_auto_da_alloc, "auto_da_alloc"},
1267 {Opt_noauto_da_alloc, "noauto_da_alloc"},
1268 {Opt_dioread_nolock, "dioread_nolock"},
1269 {Opt_dioread_lock, "dioread_lock"},
1270 {Opt_discard, "discard"},
1271 {Opt_nodiscard, "nodiscard"},
1272 {Opt_init_itable, "init_itable=%u"},
1273 {Opt_init_itable, "init_itable"},
1274 {Opt_noinit_itable, "noinit_itable"},
1275 {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1276 {Opt_test_dummy_encryption, "test_dummy_encryption"},
1277 {Opt_removed, "check=none"}, /* mount option from ext2/3 */
1278 {Opt_removed, "nocheck"}, /* mount option from ext2/3 */
1279 {Opt_removed, "reservation"}, /* mount option from ext2/3 */
1280 {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1281 {Opt_removed, "journal=%u"}, /* mount option from ext2/3 */
1282 {Opt_err, NULL},
1283};
1284
1285static ext4_fsblk_t get_sb_block(void **data)
1286{
1287 ext4_fsblk_t sb_block;
1288 char *options = (char *) *data;
1289
1290 if (!options || strncmp(options, "sb=", 3) != 0)
1291 return 1; /* Default location */
1292
1293 options += 3;
1294 /* TODO: use simple_strtoll with >32bit ext4 */
1295 sb_block = simple_strtoul(options, &options, 0);
1296 if (*options && *options != ',') {
1297 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1298 (char *) *data);
1299 return 1;
1300 }
1301 if (*options == ',')
1302 options++;
1303 *data = (void *) options;
1304
1305 return sb_block;
1306}
1307
1308#define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1309static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1310 "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1311
1312#ifdef CONFIG_QUOTA
1313static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1314{
1315 struct ext4_sb_info *sbi = EXT4_SB(sb);
1316 char *qname;
1317 int ret = -1;
1318
1319 if (sb_any_quota_loaded(sb) &&
1320 !sbi->s_qf_names[qtype]) {
1321 ext4_msg(sb, KERN_ERR,
1322 "Cannot change journaled "
1323 "quota options when quota turned on");
1324 return -1;
1325 }
1326 if (ext4_has_feature_quota(sb)) {
1327 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1328 "ignored when QUOTA feature is enabled");
1329 return 1;
1330 }
1331 qname = match_strdup(args);
1332 if (!qname) {
1333 ext4_msg(sb, KERN_ERR,
1334 "Not enough memory for storing quotafile name");
1335 return -1;
1336 }
1337 if (sbi->s_qf_names[qtype]) {
1338 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1339 ret = 1;
1340 else
1341 ext4_msg(sb, KERN_ERR,
1342 "%s quota file already specified",
1343 QTYPE2NAME(qtype));
1344 goto errout;
1345 }
1346 if (strchr(qname, '/')) {
1347 ext4_msg(sb, KERN_ERR,
1348 "quotafile must be on filesystem root");
1349 goto errout;
1350 }
1351 sbi->s_qf_names[qtype] = qname;
1352 set_opt(sb, QUOTA);
1353 return 1;
1354errout:
1355 kfree(qname);
1356 return ret;
1357}
1358
1359static int clear_qf_name(struct super_block *sb, int qtype)
1360{
1361
1362 struct ext4_sb_info *sbi = EXT4_SB(sb);
1363
1364 if (sb_any_quota_loaded(sb) &&
1365 sbi->s_qf_names[qtype]) {
1366 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1367 " when quota turned on");
1368 return -1;
1369 }
1370 kfree(sbi->s_qf_names[qtype]);
1371 sbi->s_qf_names[qtype] = NULL;
1372 return 1;
1373}
1374#endif
1375
1376#define MOPT_SET 0x0001
1377#define MOPT_CLEAR 0x0002
1378#define MOPT_NOSUPPORT 0x0004
1379#define MOPT_EXPLICIT 0x0008
1380#define MOPT_CLEAR_ERR 0x0010
1381#define MOPT_GTE0 0x0020
1382#ifdef CONFIG_QUOTA
1383#define MOPT_Q 0
1384#define MOPT_QFMT 0x0040
1385#else
1386#define MOPT_Q MOPT_NOSUPPORT
1387#define MOPT_QFMT MOPT_NOSUPPORT
1388#endif
1389#define MOPT_DATAJ 0x0080
1390#define MOPT_NO_EXT2 0x0100
1391#define MOPT_NO_EXT3 0x0200
1392#define MOPT_EXT4_ONLY (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1393#define MOPT_STRING 0x0400
1394
1395static const struct mount_opts {
1396 int token;
1397 int mount_opt;
1398 int flags;
1399} ext4_mount_opts[] = {
1400 {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1401 {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1402 {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1403 {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1404 {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1405 {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1406 {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1407 MOPT_EXT4_ONLY | MOPT_SET},
1408 {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1409 MOPT_EXT4_ONLY | MOPT_CLEAR},
1410 {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1411 {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1412 {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1413 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1414 {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1415 MOPT_EXT4_ONLY | MOPT_CLEAR},
1416 {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1417 MOPT_EXT4_ONLY | MOPT_CLEAR},
1418 {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1419 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1420 {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1421 EXT4_MOUNT_JOURNAL_CHECKSUM),
1422 MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1423 {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1424 {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1425 {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1426 {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1427 {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1428 MOPT_NO_EXT2},
1429 {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1430 MOPT_NO_EXT2},
1431 {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1432 {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1433 {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1434 {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1435 {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1436 {Opt_commit, 0, MOPT_GTE0},
1437 {Opt_max_batch_time, 0, MOPT_GTE0},
1438 {Opt_min_batch_time, 0, MOPT_GTE0},
1439 {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1440 {Opt_init_itable, 0, MOPT_GTE0},
1441 {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1442 {Opt_stripe, 0, MOPT_GTE0},
1443 {Opt_resuid, 0, MOPT_GTE0},
1444 {Opt_resgid, 0, MOPT_GTE0},
1445 {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1446 {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1447 {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1448 {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1449 {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1450 {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1451 MOPT_NO_EXT2 | MOPT_DATAJ},
1452 {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1453 {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1454#ifdef CONFIG_EXT4_FS_POSIX_ACL
1455 {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1456 {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1457#else
1458 {Opt_acl, 0, MOPT_NOSUPPORT},
1459 {Opt_noacl, 0, MOPT_NOSUPPORT},
1460#endif
1461 {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1462 {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1463 {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1464 {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1465 MOPT_SET | MOPT_Q},
1466 {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1467 MOPT_SET | MOPT_Q},
1468 {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1469 EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1470 {Opt_usrjquota, 0, MOPT_Q},
1471 {Opt_grpjquota, 0, MOPT_Q},
1472 {Opt_offusrjquota, 0, MOPT_Q},
1473 {Opt_offgrpjquota, 0, MOPT_Q},
1474 {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1475 {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1476 {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1477 {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1478 {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1479 {Opt_err, 0, 0}
1480};
1481
1482static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1483 substring_t *args, unsigned long *journal_devnum,
1484 unsigned int *journal_ioprio, int is_remount)
1485{
1486 struct ext4_sb_info *sbi = EXT4_SB(sb);
1487 const struct mount_opts *m;
1488 kuid_t uid;
1489 kgid_t gid;
1490 int arg = 0;
1491
1492#ifdef CONFIG_QUOTA
1493 if (token == Opt_usrjquota)
1494 return set_qf_name(sb, USRQUOTA, &args[0]);
1495 else if (token == Opt_grpjquota)
1496 return set_qf_name(sb, GRPQUOTA, &args[0]);
1497 else if (token == Opt_offusrjquota)
1498 return clear_qf_name(sb, USRQUOTA);
1499 else if (token == Opt_offgrpjquota)
1500 return clear_qf_name(sb, GRPQUOTA);
1501#endif
1502 switch (token) {
1503 case Opt_noacl:
1504 case Opt_nouser_xattr:
1505 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1506 break;
1507 case Opt_sb:
1508 return 1; /* handled by get_sb_block() */
1509 case Opt_removed:
1510 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1511 return 1;
1512 case Opt_abort:
1513 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1514 return 1;
1515 case Opt_i_version:
1516 sb->s_flags |= MS_I_VERSION;
1517 return 1;
1518 case Opt_lazytime:
1519 sb->s_flags |= MS_LAZYTIME;
1520 return 1;
1521 case Opt_nolazytime:
1522 sb->s_flags &= ~MS_LAZYTIME;
1523 return 1;
1524 }
1525
1526 for (m = ext4_mount_opts; m->token != Opt_err; m++)
1527 if (token == m->token)
1528 break;
1529
1530 if (m->token == Opt_err) {
1531 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1532 "or missing value", opt);
1533 return -1;
1534 }
1535
1536 if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1537 ext4_msg(sb, KERN_ERR,
1538 "Mount option \"%s\" incompatible with ext2", opt);
1539 return -1;
1540 }
1541 if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1542 ext4_msg(sb, KERN_ERR,
1543 "Mount option \"%s\" incompatible with ext3", opt);
1544 return -1;
1545 }
1546
1547 if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1548 return -1;
1549 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1550 return -1;
1551 if (m->flags & MOPT_EXPLICIT) {
1552 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1553 set_opt2(sb, EXPLICIT_DELALLOC);
1554 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1555 set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1556 } else
1557 return -1;
1558 }
1559 if (m->flags & MOPT_CLEAR_ERR)
1560 clear_opt(sb, ERRORS_MASK);
1561 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1562 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1563 "options when quota turned on");
1564 return -1;
1565 }
1566
1567 if (m->flags & MOPT_NOSUPPORT) {
1568 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1569 } else if (token == Opt_commit) {
1570 if (arg == 0)
1571 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1572 sbi->s_commit_interval = HZ * arg;
1573 } else if (token == Opt_max_batch_time) {
1574 sbi->s_max_batch_time = arg;
1575 } else if (token == Opt_min_batch_time) {
1576 sbi->s_min_batch_time = arg;
1577 } else if (token == Opt_inode_readahead_blks) {
1578 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1579 ext4_msg(sb, KERN_ERR,
1580 "EXT4-fs: inode_readahead_blks must be "
1581 "0 or a power of 2 smaller than 2^31");
1582 return -1;
1583 }
1584 sbi->s_inode_readahead_blks = arg;
1585 } else if (token == Opt_init_itable) {
1586 set_opt(sb, INIT_INODE_TABLE);
1587 if (!args->from)
1588 arg = EXT4_DEF_LI_WAIT_MULT;
1589 sbi->s_li_wait_mult = arg;
1590 } else if (token == Opt_max_dir_size_kb) {
1591 sbi->s_max_dir_size_kb = arg;
1592 } else if (token == Opt_stripe) {
1593 sbi->s_stripe = arg;
1594 } else if (token == Opt_resuid) {
1595 uid = make_kuid(current_user_ns(), arg);
1596 if (!uid_valid(uid)) {
1597 ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1598 return -1;
1599 }
1600 sbi->s_resuid = uid;
1601 } else if (token == Opt_resgid) {
1602 gid = make_kgid(current_user_ns(), arg);
1603 if (!gid_valid(gid)) {
1604 ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1605 return -1;
1606 }
1607 sbi->s_resgid = gid;
1608 } else if (token == Opt_journal_dev) {
1609 if (is_remount) {
1610 ext4_msg(sb, KERN_ERR,
1611 "Cannot specify journal on remount");
1612 return -1;
1613 }
1614 *journal_devnum = arg;
1615 } else if (token == Opt_journal_path) {
1616 char *journal_path;
1617 struct inode *journal_inode;
1618 struct path path;
1619 int error;
1620
1621 if (is_remount) {
1622 ext4_msg(sb, KERN_ERR,
1623 "Cannot specify journal on remount");
1624 return -1;
1625 }
1626 journal_path = match_strdup(&args[0]);
1627 if (!journal_path) {
1628 ext4_msg(sb, KERN_ERR, "error: could not dup "
1629 "journal device string");
1630 return -1;
1631 }
1632
1633 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1634 if (error) {
1635 ext4_msg(sb, KERN_ERR, "error: could not find "
1636 "journal device path: error %d", error);
1637 kfree(journal_path);
1638 return -1;
1639 }
1640
1641 journal_inode = d_inode(path.dentry);
1642 if (!S_ISBLK(journal_inode->i_mode)) {
1643 ext4_msg(sb, KERN_ERR, "error: journal path %s "
1644 "is not a block device", journal_path);
1645 path_put(&path);
1646 kfree(journal_path);
1647 return -1;
1648 }
1649
1650 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1651 path_put(&path);
1652 kfree(journal_path);
1653 } else if (token == Opt_journal_ioprio) {
1654 if (arg > 7) {
1655 ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1656 " (must be 0-7)");
1657 return -1;
1658 }
1659 *journal_ioprio =
1660 IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1661 } else if (token == Opt_test_dummy_encryption) {
1662#ifdef CONFIG_EXT4_FS_ENCRYPTION
1663 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1664 ext4_msg(sb, KERN_WARNING,
1665 "Test dummy encryption mode enabled");
1666#else
1667 ext4_msg(sb, KERN_WARNING,
1668 "Test dummy encryption mount option ignored");
1669#endif
1670 } else if (m->flags & MOPT_DATAJ) {
1671 if (is_remount) {
1672 if (!sbi->s_journal)
1673 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1674 else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1675 ext4_msg(sb, KERN_ERR,
1676 "Cannot change data mode on remount");
1677 return -1;
1678 }
1679 } else {
1680 clear_opt(sb, DATA_FLAGS);
1681 sbi->s_mount_opt |= m->mount_opt;
1682 }
1683#ifdef CONFIG_QUOTA
1684 } else if (m->flags & MOPT_QFMT) {
1685 if (sb_any_quota_loaded(sb) &&
1686 sbi->s_jquota_fmt != m->mount_opt) {
1687 ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1688 "quota options when quota turned on");
1689 return -1;
1690 }
1691 if (ext4_has_feature_quota(sb)) {
1692 ext4_msg(sb, KERN_INFO,
1693 "Quota format mount options ignored "
1694 "when QUOTA feature is enabled");
1695 return 1;
1696 }
1697 sbi->s_jquota_fmt = m->mount_opt;
1698#endif
1699 } else if (token == Opt_dax) {
1700#ifdef CONFIG_FS_DAX
1701 ext4_msg(sb, KERN_WARNING,
1702 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1703 sbi->s_mount_opt |= m->mount_opt;
1704#else
1705 ext4_msg(sb, KERN_INFO, "dax option not supported");
1706 return -1;
1707#endif
1708 } else if (token == Opt_data_err_abort) {
1709 sbi->s_mount_opt |= m->mount_opt;
1710 } else if (token == Opt_data_err_ignore) {
1711 sbi->s_mount_opt &= ~m->mount_opt;
1712 } else {
1713 if (!args->from)
1714 arg = 1;
1715 if (m->flags & MOPT_CLEAR)
1716 arg = !arg;
1717 else if (unlikely(!(m->flags & MOPT_SET))) {
1718 ext4_msg(sb, KERN_WARNING,
1719 "buggy handling of option %s", opt);
1720 WARN_ON(1);
1721 return -1;
1722 }
1723 if (arg != 0)
1724 sbi->s_mount_opt |= m->mount_opt;
1725 else
1726 sbi->s_mount_opt &= ~m->mount_opt;
1727 }
1728 return 1;
1729}
1730
1731static int parse_options(char *options, struct super_block *sb,
1732 unsigned long *journal_devnum,
1733 unsigned int *journal_ioprio,
1734 int is_remount)
1735{
1736 struct ext4_sb_info *sbi = EXT4_SB(sb);
1737 char *p;
1738 substring_t args[MAX_OPT_ARGS];
1739 int token;
1740
1741 if (!options)
1742 return 1;
1743
1744 while ((p = strsep(&options, ",")) != NULL) {
1745 if (!*p)
1746 continue;
1747 /*
1748 * Initialize args struct so we know whether arg was
1749 * found; some options take optional arguments.
1750 */
1751 args[0].to = args[0].from = NULL;
1752 token = match_token(p, tokens, args);
1753 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1754 journal_ioprio, is_remount) < 0)
1755 return 0;
1756 }
1757#ifdef CONFIG_QUOTA
1758 if (ext4_has_feature_quota(sb) &&
1759 (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1760 ext4_msg(sb, KERN_INFO, "Quota feature enabled, usrquota and grpquota "
1761 "mount options ignored.");
1762 clear_opt(sb, USRQUOTA);
1763 clear_opt(sb, GRPQUOTA);
1764 } else if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1765 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1766 clear_opt(sb, USRQUOTA);
1767
1768 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1769 clear_opt(sb, GRPQUOTA);
1770
1771 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1772 ext4_msg(sb, KERN_ERR, "old and new quota "
1773 "format mixing");
1774 return 0;
1775 }
1776
1777 if (!sbi->s_jquota_fmt) {
1778 ext4_msg(sb, KERN_ERR, "journaled quota format "
1779 "not specified");
1780 return 0;
1781 }
1782 }
1783#endif
1784 if (test_opt(sb, DIOREAD_NOLOCK)) {
1785 int blocksize =
1786 BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1787
1788 if (blocksize < PAGE_SIZE) {
1789 ext4_msg(sb, KERN_ERR, "can't mount with "
1790 "dioread_nolock if block size != PAGE_SIZE");
1791 return 0;
1792 }
1793 }
1794 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
1795 test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
1796 ext4_msg(sb, KERN_ERR, "can't mount with journal_async_commit "
1797 "in data=ordered mode");
1798 return 0;
1799 }
1800 return 1;
1801}
1802
1803static inline void ext4_show_quota_options(struct seq_file *seq,
1804 struct super_block *sb)
1805{
1806#if defined(CONFIG_QUOTA)
1807 struct ext4_sb_info *sbi = EXT4_SB(sb);
1808
1809 if (sbi->s_jquota_fmt) {
1810 char *fmtname = "";
1811
1812 switch (sbi->s_jquota_fmt) {
1813 case QFMT_VFS_OLD:
1814 fmtname = "vfsold";
1815 break;
1816 case QFMT_VFS_V0:
1817 fmtname = "vfsv0";
1818 break;
1819 case QFMT_VFS_V1:
1820 fmtname = "vfsv1";
1821 break;
1822 }
1823 seq_printf(seq, ",jqfmt=%s", fmtname);
1824 }
1825
1826 if (sbi->s_qf_names[USRQUOTA])
1827 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1828
1829 if (sbi->s_qf_names[GRPQUOTA])
1830 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1831#endif
1832}
1833
1834static const char *token2str(int token)
1835{
1836 const struct match_token *t;
1837
1838 for (t = tokens; t->token != Opt_err; t++)
1839 if (t->token == token && !strchr(t->pattern, '='))
1840 break;
1841 return t->pattern;
1842}
1843
1844/*
1845 * Show an option if
1846 * - it's set to a non-default value OR
1847 * - if the per-sb default is different from the global default
1848 */
1849static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1850 int nodefs)
1851{
1852 struct ext4_sb_info *sbi = EXT4_SB(sb);
1853 struct ext4_super_block *es = sbi->s_es;
1854 int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1855 const struct mount_opts *m;
1856 char sep = nodefs ? '\n' : ',';
1857
1858#define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1859#define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1860
1861 if (sbi->s_sb_block != 1)
1862 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1863
1864 for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1865 int want_set = m->flags & MOPT_SET;
1866 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1867 (m->flags & MOPT_CLEAR_ERR))
1868 continue;
1869 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1870 continue; /* skip if same as the default */
1871 if ((want_set &&
1872 (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1873 (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1874 continue; /* select Opt_noFoo vs Opt_Foo */
1875 SEQ_OPTS_PRINT("%s", token2str(m->token));
1876 }
1877
1878 if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1879 le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1880 SEQ_OPTS_PRINT("resuid=%u",
1881 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1882 if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1883 le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1884 SEQ_OPTS_PRINT("resgid=%u",
1885 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1886 def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1887 if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1888 SEQ_OPTS_PUTS("errors=remount-ro");
1889 if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1890 SEQ_OPTS_PUTS("errors=continue");
1891 if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1892 SEQ_OPTS_PUTS("errors=panic");
1893 if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1894 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1895 if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1896 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1897 if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1898 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1899 if (sb->s_flags & MS_I_VERSION)
1900 SEQ_OPTS_PUTS("i_version");
1901 if (nodefs || sbi->s_stripe)
1902 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1903 if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1904 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1905 SEQ_OPTS_PUTS("data=journal");
1906 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1907 SEQ_OPTS_PUTS("data=ordered");
1908 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1909 SEQ_OPTS_PUTS("data=writeback");
1910 }
1911 if (nodefs ||
1912 sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1913 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1914 sbi->s_inode_readahead_blks);
1915
1916 if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1917 (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1918 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1919 if (nodefs || sbi->s_max_dir_size_kb)
1920 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1921 if (test_opt(sb, DATA_ERR_ABORT))
1922 SEQ_OPTS_PUTS("data_err=abort");
1923
1924 ext4_show_quota_options(seq, sb);
1925 return 0;
1926}
1927
1928static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1929{
1930 return _ext4_show_options(seq, root->d_sb, 0);
1931}
1932
1933int ext4_seq_options_show(struct seq_file *seq, void *offset)
1934{
1935 struct super_block *sb = seq->private;
1936 int rc;
1937
1938 seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1939 rc = _ext4_show_options(seq, sb, 1);
1940 seq_puts(seq, "\n");
1941 return rc;
1942}
1943
1944static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1945 int read_only)
1946{
1947 struct ext4_sb_info *sbi = EXT4_SB(sb);
1948 int res = 0;
1949
1950 if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1951 ext4_msg(sb, KERN_ERR, "revision level too high, "
1952 "forcing read-only mode");
1953 res = MS_RDONLY;
1954 }
1955 if (read_only)
1956 goto done;
1957 if (!(sbi->s_mount_state & EXT4_VALID_FS))
1958 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1959 "running e2fsck is recommended");
1960 else if (sbi->s_mount_state & EXT4_ERROR_FS)
1961 ext4_msg(sb, KERN_WARNING,
1962 "warning: mounting fs with errors, "
1963 "running e2fsck is recommended");
1964 else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1965 le16_to_cpu(es->s_mnt_count) >=
1966 (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1967 ext4_msg(sb, KERN_WARNING,
1968 "warning: maximal mount count reached, "
1969 "running e2fsck is recommended");
1970 else if (le32_to_cpu(es->s_checkinterval) &&
1971 (le32_to_cpu(es->s_lastcheck) +
1972 le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1973 ext4_msg(sb, KERN_WARNING,
1974 "warning: checktime reached, "
1975 "running e2fsck is recommended");
1976 if (!sbi->s_journal)
1977 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1978 if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1979 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1980 le16_add_cpu(&es->s_mnt_count, 1);
1981 es->s_mtime = cpu_to_le32(get_seconds());
1982 ext4_update_dynamic_rev(sb);
1983 if (sbi->s_journal)
1984 ext4_set_feature_journal_needs_recovery(sb);
1985
1986 ext4_commit_super(sb, 1);
1987done:
1988 if (test_opt(sb, DEBUG))
1989 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1990 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1991 sb->s_blocksize,
1992 sbi->s_groups_count,
1993 EXT4_BLOCKS_PER_GROUP(sb),
1994 EXT4_INODES_PER_GROUP(sb),
1995 sbi->s_mount_opt, sbi->s_mount_opt2);
1996
1997 cleancache_init_fs(sb);
1998 return res;
1999}
2000
2001int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
2002{
2003 struct ext4_sb_info *sbi = EXT4_SB(sb);
2004 struct flex_groups *new_groups;
2005 int size;
2006
2007 if (!sbi->s_log_groups_per_flex)
2008 return 0;
2009
2010 size = ext4_flex_group(sbi, ngroup - 1) + 1;
2011 if (size <= sbi->s_flex_groups_allocated)
2012 return 0;
2013
2014 size = roundup_pow_of_two(size * sizeof(struct flex_groups));
2015 new_groups = ext4_kvzalloc(size, GFP_KERNEL);
2016 if (!new_groups) {
2017 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
2018 size / (int) sizeof(struct flex_groups));
2019 return -ENOMEM;
2020 }
2021
2022 if (sbi->s_flex_groups) {
2023 memcpy(new_groups, sbi->s_flex_groups,
2024 (sbi->s_flex_groups_allocated *
2025 sizeof(struct flex_groups)));
2026 kvfree(sbi->s_flex_groups);
2027 }
2028 sbi->s_flex_groups = new_groups;
2029 sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
2030 return 0;
2031}
2032
2033static int ext4_fill_flex_info(struct super_block *sb)
2034{
2035 struct ext4_sb_info *sbi = EXT4_SB(sb);
2036 struct ext4_group_desc *gdp = NULL;
2037 ext4_group_t flex_group;
2038 int i, err;
2039
2040 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2041 if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2042 sbi->s_log_groups_per_flex = 0;
2043 return 1;
2044 }
2045
2046 err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2047 if (err)
2048 goto failed;
2049
2050 for (i = 0; i < sbi->s_groups_count; i++) {
2051 gdp = ext4_get_group_desc(sb, i, NULL);
2052
2053 flex_group = ext4_flex_group(sbi, i);
2054 atomic_add(ext4_free_inodes_count(sb, gdp),
2055 &sbi->s_flex_groups[flex_group].free_inodes);
2056 atomic64_add(ext4_free_group_clusters(sb, gdp),
2057 &sbi->s_flex_groups[flex_group].free_clusters);
2058 atomic_add(ext4_used_dirs_count(sb, gdp),
2059 &sbi->s_flex_groups[flex_group].used_dirs);
2060 }
2061
2062 return 1;
2063failed:
2064 return 0;
2065}
2066
2067static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2068 struct ext4_group_desc *gdp)
2069{
2070 int offset;
2071 __u16 crc = 0;
2072 __le32 le_group = cpu_to_le32(block_group);
2073 struct ext4_sb_info *sbi = EXT4_SB(sb);
2074
2075 if (ext4_has_metadata_csum(sbi->s_sb)) {
2076 /* Use new metadata_csum algorithm */
2077 __le16 save_csum;
2078 __u32 csum32;
2079
2080 save_csum = gdp->bg_checksum;
2081 gdp->bg_checksum = 0;
2082 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2083 sizeof(le_group));
2084 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2085 sbi->s_desc_size);
2086 gdp->bg_checksum = save_csum;
2087
2088 crc = csum32 & 0xFFFF;
2089 goto out;
2090 }
2091
2092 /* old crc16 code */
2093 if (!ext4_has_feature_gdt_csum(sb))
2094 return 0;
2095
2096 offset = offsetof(struct ext4_group_desc, bg_checksum);
2097
2098 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2099 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2100 crc = crc16(crc, (__u8 *)gdp, offset);
2101 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2102 /* for checksum of struct ext4_group_desc do the rest...*/
2103 if (ext4_has_feature_64bit(sb) &&
2104 offset < le16_to_cpu(sbi->s_es->s_desc_size))
2105 crc = crc16(crc, (__u8 *)gdp + offset,
2106 le16_to_cpu(sbi->s_es->s_desc_size) -
2107 offset);
2108
2109out:
2110 return cpu_to_le16(crc);
2111}
2112
2113int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2114 struct ext4_group_desc *gdp)
2115{
2116 if (ext4_has_group_desc_csum(sb) &&
2117 (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2118 return 0;
2119
2120 return 1;
2121}
2122
2123void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2124 struct ext4_group_desc *gdp)
2125{
2126 if (!ext4_has_group_desc_csum(sb))
2127 return;
2128 gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2129}
2130
2131/* Called at mount-time, super-block is locked */
2132static int ext4_check_descriptors(struct super_block *sb,
2133 ext4_group_t *first_not_zeroed)
2134{
2135 struct ext4_sb_info *sbi = EXT4_SB(sb);
2136 ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2137 ext4_fsblk_t last_block;
2138 ext4_fsblk_t block_bitmap;
2139 ext4_fsblk_t inode_bitmap;
2140 ext4_fsblk_t inode_table;
2141 int flexbg_flag = 0;
2142 ext4_group_t i, grp = sbi->s_groups_count;
2143
2144 if (ext4_has_feature_flex_bg(sb))
2145 flexbg_flag = 1;
2146
2147 ext4_debug("Checking group descriptors");
2148
2149 for (i = 0; i < sbi->s_groups_count; i++) {
2150 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2151
2152 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2153 last_block = ext4_blocks_count(sbi->s_es) - 1;
2154 else
2155 last_block = first_block +
2156 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2157
2158 if ((grp == sbi->s_groups_count) &&
2159 !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2160 grp = i;
2161
2162 block_bitmap = ext4_block_bitmap(sb, gdp);
2163 if (block_bitmap < first_block || block_bitmap > last_block) {
2164 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2165 "Block bitmap for group %u not in group "
2166 "(block %llu)!", i, block_bitmap);
2167 return 0;
2168 }
2169 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2170 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2171 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2172 "Inode bitmap for group %u not in group "
2173 "(block %llu)!", i, inode_bitmap);
2174 return 0;
2175 }
2176 inode_table = ext4_inode_table(sb, gdp);
2177 if (inode_table < first_block ||
2178 inode_table + sbi->s_itb_per_group - 1 > last_block) {
2179 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2180 "Inode table for group %u not in group "
2181 "(block %llu)!", i, inode_table);
2182 return 0;
2183 }
2184 ext4_lock_group(sb, i);
2185 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2186 ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2187 "Checksum for group %u failed (%u!=%u)",
2188 i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2189 gdp)), le16_to_cpu(gdp->bg_checksum));
2190 if (!(sb->s_flags & MS_RDONLY)) {
2191 ext4_unlock_group(sb, i);
2192 return 0;
2193 }
2194 }
2195 ext4_unlock_group(sb, i);
2196 if (!flexbg_flag)
2197 first_block += EXT4_BLOCKS_PER_GROUP(sb);
2198 }
2199 if (NULL != first_not_zeroed)
2200 *first_not_zeroed = grp;
2201 return 1;
2202}
2203
2204/* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2205 * the superblock) which were deleted from all directories, but held open by
2206 * a process at the time of a crash. We walk the list and try to delete these
2207 * inodes at recovery time (only with a read-write filesystem).
2208 *
2209 * In order to keep the orphan inode chain consistent during traversal (in
2210 * case of crash during recovery), we link each inode into the superblock
2211 * orphan list_head and handle it the same way as an inode deletion during
2212 * normal operation (which journals the operations for us).
2213 *
2214 * We only do an iget() and an iput() on each inode, which is very safe if we
2215 * accidentally point at an in-use or already deleted inode. The worst that
2216 * can happen in this case is that we get a "bit already cleared" message from
2217 * ext4_free_inode(). The only reason we would point at a wrong inode is if
2218 * e2fsck was run on this filesystem, and it must have already done the orphan
2219 * inode cleanup for us, so we can safely abort without any further action.
2220 */
2221static void ext4_orphan_cleanup(struct super_block *sb,
2222 struct ext4_super_block *es)
2223{
2224 unsigned int s_flags = sb->s_flags;
2225 int nr_orphans = 0, nr_truncates = 0;
2226#ifdef CONFIG_QUOTA
2227 int i;
2228#endif
2229 if (!es->s_last_orphan) {
2230 jbd_debug(4, "no orphan inodes to clean up\n");
2231 return;
2232 }
2233
2234 if (bdev_read_only(sb->s_bdev)) {
2235 ext4_msg(sb, KERN_ERR, "write access "
2236 "unavailable, skipping orphan cleanup");
2237 return;
2238 }
2239
2240 /* Check if feature set would not allow a r/w mount */
2241 if (!ext4_feature_set_ok(sb, 0)) {
2242 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2243 "unknown ROCOMPAT features");
2244 return;
2245 }
2246
2247 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2248 /* don't clear list on RO mount w/ errors */
2249 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2250 ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2251 "clearing orphan list.\n");
2252 es->s_last_orphan = 0;
2253 }
2254 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2255 return;
2256 }
2257
2258 if (s_flags & MS_RDONLY) {
2259 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2260 sb->s_flags &= ~MS_RDONLY;
2261 }
2262#ifdef CONFIG_QUOTA
2263 /* Needed for iput() to work correctly and not trash data */
2264 sb->s_flags |= MS_ACTIVE;
2265 /* Turn on quotas so that they are updated correctly */
2266 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2267 if (EXT4_SB(sb)->s_qf_names[i]) {
2268 int ret = ext4_quota_on_mount(sb, i);
2269 if (ret < 0)
2270 ext4_msg(sb, KERN_ERR,
2271 "Cannot turn on journaled "
2272 "quota: error %d", ret);
2273 }
2274 }
2275#endif
2276
2277 while (es->s_last_orphan) {
2278 struct inode *inode;
2279
2280 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2281 if (IS_ERR(inode)) {
2282 es->s_last_orphan = 0;
2283 break;
2284 }
2285
2286 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2287 dquot_initialize(inode);
2288 if (inode->i_nlink) {
2289 if (test_opt(sb, DEBUG))
2290 ext4_msg(sb, KERN_DEBUG,
2291 "%s: truncating inode %lu to %lld bytes",
2292 __func__, inode->i_ino, inode->i_size);
2293 jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2294 inode->i_ino, inode->i_size);
2295 inode_lock(inode);
2296 truncate_inode_pages(inode->i_mapping, inode->i_size);
2297 ext4_truncate(inode);
2298 inode_unlock(inode);
2299 nr_truncates++;
2300 } else {
2301 if (test_opt(sb, DEBUG))
2302 ext4_msg(sb, KERN_DEBUG,
2303 "%s: deleting unreferenced inode %lu",
2304 __func__, inode->i_ino);
2305 jbd_debug(2, "deleting unreferenced inode %lu\n",
2306 inode->i_ino);
2307 nr_orphans++;
2308 }
2309 iput(inode); /* The delete magic happens here! */
2310 }
2311
2312#define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2313
2314 if (nr_orphans)
2315 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2316 PLURAL(nr_orphans));
2317 if (nr_truncates)
2318 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2319 PLURAL(nr_truncates));
2320#ifdef CONFIG_QUOTA
2321 /* Turn quotas off */
2322 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2323 if (sb_dqopt(sb)->files[i])
2324 dquot_quota_off(sb, i);
2325 }
2326#endif
2327 sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2328}
2329
2330/*
2331 * Maximal extent format file size.
2332 * Resulting logical blkno at s_maxbytes must fit in our on-disk
2333 * extent format containers, within a sector_t, and within i_blocks
2334 * in the vfs. ext4 inode has 48 bits of i_block in fsblock units,
2335 * so that won't be a limiting factor.
2336 *
2337 * However there is other limiting factor. We do store extents in the form
2338 * of starting block and length, hence the resulting length of the extent
2339 * covering maximum file size must fit into on-disk format containers as
2340 * well. Given that length is always by 1 unit bigger than max unit (because
2341 * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2342 *
2343 * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2344 */
2345static loff_t ext4_max_size(int blkbits, int has_huge_files)
2346{
2347 loff_t res;
2348 loff_t upper_limit = MAX_LFS_FILESIZE;
2349
2350 /* small i_blocks in vfs inode? */
2351 if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2352 /*
2353 * CONFIG_LBDAF is not enabled implies the inode
2354 * i_block represent total blocks in 512 bytes
2355 * 32 == 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 >>= (blkbits - 9);
2361 upper_limit <<= blkbits;
2362 }
2363
2364 /*
2365 * 32-bit extent-start container, ee_block. We lower the maxbytes
2366 * by one fs block, so ee_len can cover the extent of maximum file
2367 * size
2368 */
2369 res = (1LL << 32) - 1;
2370 res <<= blkbits;
2371
2372 /* Sanity check against vm- & vfs- imposed limits */
2373 if (res > upper_limit)
2374 res = upper_limit;
2375
2376 return res;
2377}
2378
2379/*
2380 * Maximal bitmap file size. There is a direct, and {,double-,triple-}indirect
2381 * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2382 * We need to be 1 filesystem block less than the 2^48 sector limit.
2383 */
2384static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2385{
2386 loff_t res = EXT4_NDIR_BLOCKS;
2387 int meta_blocks;
2388 loff_t upper_limit;
2389 /* This is calculated to be the largest file size for a dense, block
2390 * mapped file such that the file's total number of 512-byte sectors,
2391 * including data and all indirect blocks, does not exceed (2^48 - 1).
2392 *
2393 * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2394 * number of 512-byte sectors of the file.
2395 */
2396
2397 if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2398 /*
2399 * !has_huge_files or CONFIG_LBDAF not enabled implies that
2400 * the inode i_block field represents total file blocks in
2401 * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2402 */
2403 upper_limit = (1LL << 32) - 1;
2404
2405 /* total blocks in file system block size */
2406 upper_limit >>= (bits - 9);
2407
2408 } else {
2409 /*
2410 * We use 48 bit ext4_inode i_blocks
2411 * With EXT4_HUGE_FILE_FL set the i_blocks
2412 * represent total number of blocks in
2413 * file system block size
2414 */
2415 upper_limit = (1LL << 48) - 1;
2416
2417 }
2418
2419 /* indirect blocks */
2420 meta_blocks = 1;
2421 /* double indirect blocks */
2422 meta_blocks += 1 + (1LL << (bits-2));
2423 /* tripple indirect blocks */
2424 meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2425
2426 upper_limit -= meta_blocks;
2427 upper_limit <<= bits;
2428
2429 res += 1LL << (bits-2);
2430 res += 1LL << (2*(bits-2));
2431 res += 1LL << (3*(bits-2));
2432 res <<= bits;
2433 if (res > upper_limit)
2434 res = upper_limit;
2435
2436 if (res > MAX_LFS_FILESIZE)
2437 res = MAX_LFS_FILESIZE;
2438
2439 return res;
2440}
2441
2442static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2443 ext4_fsblk_t logical_sb_block, int nr)
2444{
2445 struct ext4_sb_info *sbi = EXT4_SB(sb);
2446 ext4_group_t bg, first_meta_bg;
2447 int has_super = 0;
2448
2449 first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2450
2451 if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2452 return logical_sb_block + nr + 1;
2453 bg = sbi->s_desc_per_block * nr;
2454 if (ext4_bg_has_super(sb, bg))
2455 has_super = 1;
2456
2457 /*
2458 * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2459 * block 2, not 1. If s_first_data_block == 0 (bigalloc is enabled
2460 * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2461 * compensate.
2462 */
2463 if (sb->s_blocksize == 1024 && nr == 0 &&
2464 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2465 has_super++;
2466
2467 return (has_super + ext4_group_first_block_no(sb, bg));
2468}
2469
2470/**
2471 * ext4_get_stripe_size: Get the stripe size.
2472 * @sbi: In memory super block info
2473 *
2474 * If we have specified it via mount option, then
2475 * use the mount option value. If the value specified at mount time is
2476 * greater than the blocks per group use the super block value.
2477 * If the super block value is greater than blocks per group return 0.
2478 * Allocator needs it be less than blocks per group.
2479 *
2480 */
2481static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2482{
2483 unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2484 unsigned long stripe_width =
2485 le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2486 int ret;
2487
2488 if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2489 ret = sbi->s_stripe;
2490 else if (stripe_width <= sbi->s_blocks_per_group)
2491 ret = stripe_width;
2492 else if (stride <= sbi->s_blocks_per_group)
2493 ret = stride;
2494 else
2495 ret = 0;
2496
2497 /*
2498 * If the stripe width is 1, this makes no sense and
2499 * we set it to 0 to turn off stripe handling code.
2500 */
2501 if (ret <= 1)
2502 ret = 0;
2503
2504 return ret;
2505}
2506
2507/*
2508 * Check whether this filesystem can be mounted based on
2509 * the features present and the RDONLY/RDWR mount requested.
2510 * Returns 1 if this filesystem can be mounted as requested,
2511 * 0 if it cannot be.
2512 */
2513static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2514{
2515 if (ext4_has_unknown_ext4_incompat_features(sb)) {
2516 ext4_msg(sb, KERN_ERR,
2517 "Couldn't mount because of "
2518 "unsupported optional features (%x)",
2519 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2520 ~EXT4_FEATURE_INCOMPAT_SUPP));
2521 return 0;
2522 }
2523
2524 if (readonly)
2525 return 1;
2526
2527 if (ext4_has_feature_readonly(sb)) {
2528 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2529 sb->s_flags |= MS_RDONLY;
2530 return 1;
2531 }
2532
2533 /* Check that feature set is OK for a read-write mount */
2534 if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2535 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2536 "unsupported optional features (%x)",
2537 (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2538 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2539 return 0;
2540 }
2541 /*
2542 * Large file size enabled file system can only be mounted
2543 * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2544 */
2545 if (ext4_has_feature_huge_file(sb)) {
2546 if (sizeof(blkcnt_t) < sizeof(u64)) {
2547 ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2548 "cannot be mounted RDWR without "
2549 "CONFIG_LBDAF");
2550 return 0;
2551 }
2552 }
2553 if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2554 ext4_msg(sb, KERN_ERR,
2555 "Can't support bigalloc feature without "
2556 "extents feature\n");
2557 return 0;
2558 }
2559
2560#ifndef CONFIG_QUOTA
2561 if (ext4_has_feature_quota(sb) && !readonly) {
2562 ext4_msg(sb, KERN_ERR,
2563 "Filesystem with quota feature cannot be mounted RDWR "
2564 "without CONFIG_QUOTA");
2565 return 0;
2566 }
2567 if (ext4_has_feature_project(sb) && !readonly) {
2568 ext4_msg(sb, KERN_ERR,
2569 "Filesystem with project quota feature cannot be mounted RDWR "
2570 "without CONFIG_QUOTA");
2571 return 0;
2572 }
2573#endif /* CONFIG_QUOTA */
2574 return 1;
2575}
2576
2577/*
2578 * This function is called once a day if we have errors logged
2579 * on the file system
2580 */
2581static void print_daily_error_info(unsigned long arg)
2582{
2583 struct super_block *sb = (struct super_block *) arg;
2584 struct ext4_sb_info *sbi;
2585 struct ext4_super_block *es;
2586
2587 sbi = EXT4_SB(sb);
2588 es = sbi->s_es;
2589
2590 if (es->s_error_count)
2591 /* fsck newer than v1.41.13 is needed to clean this condition. */
2592 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2593 le32_to_cpu(es->s_error_count));
2594 if (es->s_first_error_time) {
2595 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2596 sb->s_id, le32_to_cpu(es->s_first_error_time),
2597 (int) sizeof(es->s_first_error_func),
2598 es->s_first_error_func,
2599 le32_to_cpu(es->s_first_error_line));
2600 if (es->s_first_error_ino)
2601 printk(": inode %u",
2602 le32_to_cpu(es->s_first_error_ino));
2603 if (es->s_first_error_block)
2604 printk(": block %llu", (unsigned long long)
2605 le64_to_cpu(es->s_first_error_block));
2606 printk("\n");
2607 }
2608 if (es->s_last_error_time) {
2609 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2610 sb->s_id, le32_to_cpu(es->s_last_error_time),
2611 (int) sizeof(es->s_last_error_func),
2612 es->s_last_error_func,
2613 le32_to_cpu(es->s_last_error_line));
2614 if (es->s_last_error_ino)
2615 printk(": inode %u",
2616 le32_to_cpu(es->s_last_error_ino));
2617 if (es->s_last_error_block)
2618 printk(": block %llu", (unsigned long long)
2619 le64_to_cpu(es->s_last_error_block));
2620 printk("\n");
2621 }
2622 mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ); /* Once a day */
2623}
2624
2625/* Find next suitable group and run ext4_init_inode_table */
2626static int ext4_run_li_request(struct ext4_li_request *elr)
2627{
2628 struct ext4_group_desc *gdp = NULL;
2629 ext4_group_t group, ngroups;
2630 struct super_block *sb;
2631 unsigned long timeout = 0;
2632 int ret = 0;
2633
2634 sb = elr->lr_super;
2635 ngroups = EXT4_SB(sb)->s_groups_count;
2636
2637 sb_start_write(sb);
2638 for (group = elr->lr_next_group; group < ngroups; group++) {
2639 gdp = ext4_get_group_desc(sb, group, NULL);
2640 if (!gdp) {
2641 ret = 1;
2642 break;
2643 }
2644
2645 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2646 break;
2647 }
2648
2649 if (group >= ngroups)
2650 ret = 1;
2651
2652 if (!ret) {
2653 timeout = jiffies;
2654 ret = ext4_init_inode_table(sb, group,
2655 elr->lr_timeout ? 0 : 1);
2656 if (elr->lr_timeout == 0) {
2657 timeout = (jiffies - timeout) *
2658 elr->lr_sbi->s_li_wait_mult;
2659 elr->lr_timeout = timeout;
2660 }
2661 elr->lr_next_sched = jiffies + elr->lr_timeout;
2662 elr->lr_next_group = group + 1;
2663 }
2664 sb_end_write(sb);
2665
2666 return ret;
2667}
2668
2669/*
2670 * Remove lr_request from the list_request and free the
2671 * request structure. Should be called with li_list_mtx held
2672 */
2673static void ext4_remove_li_request(struct ext4_li_request *elr)
2674{
2675 struct ext4_sb_info *sbi;
2676
2677 if (!elr)
2678 return;
2679
2680 sbi = elr->lr_sbi;
2681
2682 list_del(&elr->lr_request);
2683 sbi->s_li_request = NULL;
2684 kfree(elr);
2685}
2686
2687static void ext4_unregister_li_request(struct super_block *sb)
2688{
2689 mutex_lock(&ext4_li_mtx);
2690 if (!ext4_li_info) {
2691 mutex_unlock(&ext4_li_mtx);
2692 return;
2693 }
2694
2695 mutex_lock(&ext4_li_info->li_list_mtx);
2696 ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2697 mutex_unlock(&ext4_li_info->li_list_mtx);
2698 mutex_unlock(&ext4_li_mtx);
2699}
2700
2701static struct task_struct *ext4_lazyinit_task;
2702
2703/*
2704 * This is the function where ext4lazyinit thread lives. It walks
2705 * through the request list searching for next scheduled filesystem.
2706 * When such a fs is found, run the lazy initialization request
2707 * (ext4_rn_li_request) and keep track of the time spend in this
2708 * function. Based on that time we compute next schedule time of
2709 * the request. When walking through the list is complete, compute
2710 * next waking time and put itself into sleep.
2711 */
2712static int ext4_lazyinit_thread(void *arg)
2713{
2714 struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2715 struct list_head *pos, *n;
2716 struct ext4_li_request *elr;
2717 unsigned long next_wakeup, cur;
2718
2719 BUG_ON(NULL == eli);
2720
2721cont_thread:
2722 while (true) {
2723 next_wakeup = MAX_JIFFY_OFFSET;
2724
2725 mutex_lock(&eli->li_list_mtx);
2726 if (list_empty(&eli->li_request_list)) {
2727 mutex_unlock(&eli->li_list_mtx);
2728 goto exit_thread;
2729 }
2730
2731 list_for_each_safe(pos, n, &eli->li_request_list) {
2732 elr = list_entry(pos, struct ext4_li_request,
2733 lr_request);
2734
2735 if (time_after_eq(jiffies, elr->lr_next_sched)) {
2736 if (ext4_run_li_request(elr) != 0) {
2737 /* error, remove the lazy_init job */
2738 ext4_remove_li_request(elr);
2739 continue;
2740 }
2741 }
2742
2743 if (time_before(elr->lr_next_sched, next_wakeup))
2744 next_wakeup = elr->lr_next_sched;
2745 }
2746 mutex_unlock(&eli->li_list_mtx);
2747
2748 try_to_freeze();
2749
2750 cur = jiffies;
2751 if ((time_after_eq(cur, next_wakeup)) ||
2752 (MAX_JIFFY_OFFSET == next_wakeup)) {
2753 cond_resched();
2754 continue;
2755 }
2756
2757 schedule_timeout_interruptible(next_wakeup - cur);
2758
2759 if (kthread_should_stop()) {
2760 ext4_clear_request_list();
2761 goto exit_thread;
2762 }
2763 }
2764
2765exit_thread:
2766 /*
2767 * It looks like the request list is empty, but we need
2768 * to check it under the li_list_mtx lock, to prevent any
2769 * additions into it, and of course we should lock ext4_li_mtx
2770 * to atomically free the list and ext4_li_info, because at
2771 * this point another ext4 filesystem could be registering
2772 * new one.
2773 */
2774 mutex_lock(&ext4_li_mtx);
2775 mutex_lock(&eli->li_list_mtx);
2776 if (!list_empty(&eli->li_request_list)) {
2777 mutex_unlock(&eli->li_list_mtx);
2778 mutex_unlock(&ext4_li_mtx);
2779 goto cont_thread;
2780 }
2781 mutex_unlock(&eli->li_list_mtx);
2782 kfree(ext4_li_info);
2783 ext4_li_info = NULL;
2784 mutex_unlock(&ext4_li_mtx);
2785
2786 return 0;
2787}
2788
2789static void ext4_clear_request_list(void)
2790{
2791 struct list_head *pos, *n;
2792 struct ext4_li_request *elr;
2793
2794 mutex_lock(&ext4_li_info->li_list_mtx);
2795 list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2796 elr = list_entry(pos, struct ext4_li_request,
2797 lr_request);
2798 ext4_remove_li_request(elr);
2799 }
2800 mutex_unlock(&ext4_li_info->li_list_mtx);
2801}
2802
2803static int ext4_run_lazyinit_thread(void)
2804{
2805 ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2806 ext4_li_info, "ext4lazyinit");
2807 if (IS_ERR(ext4_lazyinit_task)) {
2808 int err = PTR_ERR(ext4_lazyinit_task);
2809 ext4_clear_request_list();
2810 kfree(ext4_li_info);
2811 ext4_li_info = NULL;
2812 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2813 "initialization thread\n",
2814 err);
2815 return err;
2816 }
2817 ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2818 return 0;
2819}
2820
2821/*
2822 * Check whether it make sense to run itable init. thread or not.
2823 * If there is at least one uninitialized inode table, return
2824 * corresponding group number, else the loop goes through all
2825 * groups and return total number of groups.
2826 */
2827static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2828{
2829 ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2830 struct ext4_group_desc *gdp = NULL;
2831
2832 for (group = 0; group < ngroups; group++) {
2833 gdp = ext4_get_group_desc(sb, group, NULL);
2834 if (!gdp)
2835 continue;
2836
2837 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2838 break;
2839 }
2840
2841 return group;
2842}
2843
2844static int ext4_li_info_new(void)
2845{
2846 struct ext4_lazy_init *eli = NULL;
2847
2848 eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2849 if (!eli)
2850 return -ENOMEM;
2851
2852 INIT_LIST_HEAD(&eli->li_request_list);
2853 mutex_init(&eli->li_list_mtx);
2854
2855 eli->li_state |= EXT4_LAZYINIT_QUIT;
2856
2857 ext4_li_info = eli;
2858
2859 return 0;
2860}
2861
2862static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2863 ext4_group_t start)
2864{
2865 struct ext4_sb_info *sbi = EXT4_SB(sb);
2866 struct ext4_li_request *elr;
2867
2868 elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2869 if (!elr)
2870 return NULL;
2871
2872 elr->lr_super = sb;
2873 elr->lr_sbi = sbi;
2874 elr->lr_next_group = start;
2875
2876 /*
2877 * Randomize first schedule time of the request to
2878 * spread the inode table initialization requests
2879 * better.
2880 */
2881 elr->lr_next_sched = jiffies + (prandom_u32() %
2882 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2883 return elr;
2884}
2885
2886int ext4_register_li_request(struct super_block *sb,
2887 ext4_group_t first_not_zeroed)
2888{
2889 struct ext4_sb_info *sbi = EXT4_SB(sb);
2890 struct ext4_li_request *elr = NULL;
2891 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2892 int ret = 0;
2893
2894 mutex_lock(&ext4_li_mtx);
2895 if (sbi->s_li_request != NULL) {
2896 /*
2897 * Reset timeout so it can be computed again, because
2898 * s_li_wait_mult might have changed.
2899 */
2900 sbi->s_li_request->lr_timeout = 0;
2901 goto out;
2902 }
2903
2904 if (first_not_zeroed == ngroups ||
2905 (sb->s_flags & MS_RDONLY) ||
2906 !test_opt(sb, INIT_INODE_TABLE))
2907 goto out;
2908
2909 elr = ext4_li_request_new(sb, first_not_zeroed);
2910 if (!elr) {
2911 ret = -ENOMEM;
2912 goto out;
2913 }
2914
2915 if (NULL == ext4_li_info) {
2916 ret = ext4_li_info_new();
2917 if (ret)
2918 goto out;
2919 }
2920
2921 mutex_lock(&ext4_li_info->li_list_mtx);
2922 list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2923 mutex_unlock(&ext4_li_info->li_list_mtx);
2924
2925 sbi->s_li_request = elr;
2926 /*
2927 * set elr to NULL here since it has been inserted to
2928 * the request_list and the removal and free of it is
2929 * handled by ext4_clear_request_list from now on.
2930 */
2931 elr = NULL;
2932
2933 if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2934 ret = ext4_run_lazyinit_thread();
2935 if (ret)
2936 goto out;
2937 }
2938out:
2939 mutex_unlock(&ext4_li_mtx);
2940 if (ret)
2941 kfree(elr);
2942 return ret;
2943}
2944
2945/*
2946 * We do not need to lock anything since this is called on
2947 * module unload.
2948 */
2949static void ext4_destroy_lazyinit_thread(void)
2950{
2951 /*
2952 * If thread exited earlier
2953 * there's nothing to be done.
2954 */
2955 if (!ext4_li_info || !ext4_lazyinit_task)
2956 return;
2957
2958 kthread_stop(ext4_lazyinit_task);
2959}
2960
2961static int set_journal_csum_feature_set(struct super_block *sb)
2962{
2963 int ret = 1;
2964 int compat, incompat;
2965 struct ext4_sb_info *sbi = EXT4_SB(sb);
2966
2967 if (ext4_has_metadata_csum(sb)) {
2968 /* journal checksum v3 */
2969 compat = 0;
2970 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
2971 } else {
2972 /* journal checksum v1 */
2973 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
2974 incompat = 0;
2975 }
2976
2977 jbd2_journal_clear_features(sbi->s_journal,
2978 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2979 JBD2_FEATURE_INCOMPAT_CSUM_V3 |
2980 JBD2_FEATURE_INCOMPAT_CSUM_V2);
2981 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2982 ret = jbd2_journal_set_features(sbi->s_journal,
2983 compat, 0,
2984 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
2985 incompat);
2986 } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2987 ret = jbd2_journal_set_features(sbi->s_journal,
2988 compat, 0,
2989 incompat);
2990 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2991 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2992 } else {
2993 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2994 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2995 }
2996
2997 return ret;
2998}
2999
3000/*
3001 * Note: calculating the overhead so we can be compatible with
3002 * historical BSD practice is quite difficult in the face of
3003 * clusters/bigalloc. This is because multiple metadata blocks from
3004 * different block group can end up in the same allocation cluster.
3005 * Calculating the exact overhead in the face of clustered allocation
3006 * requires either O(all block bitmaps) in memory or O(number of block
3007 * groups**2) in time. We will still calculate the superblock for
3008 * older file systems --- and if we come across with a bigalloc file
3009 * system with zero in s_overhead_clusters the estimate will be close to
3010 * correct especially for very large cluster sizes --- but for newer
3011 * file systems, it's better to calculate this figure once at mkfs
3012 * time, and store it in the superblock. If the superblock value is
3013 * present (even for non-bigalloc file systems), we will use it.
3014 */
3015static int count_overhead(struct super_block *sb, ext4_group_t grp,
3016 char *buf)
3017{
3018 struct ext4_sb_info *sbi = EXT4_SB(sb);
3019 struct ext4_group_desc *gdp;
3020 ext4_fsblk_t first_block, last_block, b;
3021 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3022 int s, j, count = 0;
3023
3024 if (!ext4_has_feature_bigalloc(sb))
3025 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3026 sbi->s_itb_per_group + 2);
3027
3028 first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3029 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3030 last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3031 for (i = 0; i < ngroups; i++) {
3032 gdp = ext4_get_group_desc(sb, i, NULL);
3033 b = ext4_block_bitmap(sb, gdp);
3034 if (b >= first_block && b <= last_block) {
3035 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3036 count++;
3037 }
3038 b = ext4_inode_bitmap(sb, gdp);
3039 if (b >= first_block && b <= last_block) {
3040 ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3041 count++;
3042 }
3043 b = ext4_inode_table(sb, gdp);
3044 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3045 for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3046 int c = EXT4_B2C(sbi, b - first_block);
3047 ext4_set_bit(c, buf);
3048 count++;
3049 }
3050 if (i != grp)
3051 continue;
3052 s = 0;
3053 if (ext4_bg_has_super(sb, grp)) {
3054 ext4_set_bit(s++, buf);
3055 count++;
3056 }
3057 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3058 ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3059 count++;
3060 }
3061 }
3062 if (!count)
3063 return 0;
3064 return EXT4_CLUSTERS_PER_GROUP(sb) -
3065 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3066}
3067
3068/*
3069 * Compute the overhead and stash it in sbi->s_overhead
3070 */
3071int ext4_calculate_overhead(struct super_block *sb)
3072{
3073 struct ext4_sb_info *sbi = EXT4_SB(sb);
3074 struct ext4_super_block *es = sbi->s_es;
3075 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3076 ext4_fsblk_t overhead = 0;
3077 char *buf = (char *) get_zeroed_page(GFP_NOFS);
3078
3079 if (!buf)
3080 return -ENOMEM;
3081
3082 /*
3083 * Compute the overhead (FS structures). This is constant
3084 * for a given filesystem unless the number of block groups
3085 * changes so we cache the previous value until it does.
3086 */
3087
3088 /*
3089 * All of the blocks before first_data_block are overhead
3090 */
3091 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3092
3093 /*
3094 * Add the overhead found in each block group
3095 */
3096 for (i = 0; i < ngroups; i++) {
3097 int blks;
3098
3099 blks = count_overhead(sb, i, buf);
3100 overhead += blks;
3101 if (blks)
3102 memset(buf, 0, PAGE_SIZE);
3103 cond_resched();
3104 }
3105 /* Add the internal journal blocks as well */
3106 if (sbi->s_journal && !sbi->journal_bdev)
3107 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3108
3109 sbi->s_overhead = overhead;
3110 smp_wmb();
3111 free_page((unsigned long) buf);
3112 return 0;
3113}
3114
3115static void ext4_set_resv_clusters(struct super_block *sb)
3116{
3117 ext4_fsblk_t resv_clusters;
3118 struct ext4_sb_info *sbi = EXT4_SB(sb);
3119
3120 /*
3121 * There's no need to reserve anything when we aren't using extents.
3122 * The space estimates are exact, there are no unwritten extents,
3123 * hole punching doesn't need new metadata... This is needed especially
3124 * to keep ext2/3 backward compatibility.
3125 */
3126 if (!ext4_has_feature_extents(sb))
3127 return;
3128 /*
3129 * By default we reserve 2% or 4096 clusters, whichever is smaller.
3130 * This should cover the situations where we can not afford to run
3131 * out of space like for example punch hole, or converting
3132 * unwritten extents in delalloc path. In most cases such
3133 * allocation would require 1, or 2 blocks, higher numbers are
3134 * very rare.
3135 */
3136 resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3137 sbi->s_cluster_bits);
3138
3139 do_div(resv_clusters, 50);
3140 resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3141
3142 atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3143}
3144
3145static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3146{
3147 char *orig_data = kstrdup(data, GFP_KERNEL);
3148 struct buffer_head *bh;
3149 struct ext4_super_block *es = NULL;
3150 struct ext4_sb_info *sbi;
3151 ext4_fsblk_t block;
3152 ext4_fsblk_t sb_block = get_sb_block(&data);
3153 ext4_fsblk_t logical_sb_block;
3154 unsigned long offset = 0;
3155 unsigned long journal_devnum = 0;
3156 unsigned long def_mount_opts;
3157 struct inode *root;
3158 const char *descr;
3159 int ret = -ENOMEM;
3160 int blocksize, clustersize;
3161 unsigned int db_count;
3162 unsigned int i;
3163 int needs_recovery, has_huge_files, has_bigalloc;
3164 __u64 blocks_count;
3165 int err = 0;
3166 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3167 ext4_group_t first_not_zeroed;
3168
3169 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3170 if (!sbi)
3171 goto out_free_orig;
3172
3173 sbi->s_blockgroup_lock =
3174 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3175 if (!sbi->s_blockgroup_lock) {
3176 kfree(sbi);
3177 goto out_free_orig;
3178 }
3179 sb->s_fs_info = sbi;
3180 sbi->s_sb = sb;
3181 sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3182 sbi->s_sb_block = sb_block;
3183 if (sb->s_bdev->bd_part)
3184 sbi->s_sectors_written_start =
3185 part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3186
3187 /* Cleanup superblock name */
3188 strreplace(sb->s_id, '/', '!');
3189
3190 /* -EINVAL is default */
3191 ret = -EINVAL;
3192 blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3193 if (!blocksize) {
3194 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3195 goto out_fail;
3196 }
3197
3198 /*
3199 * The ext4 superblock will not be buffer aligned for other than 1kB
3200 * block sizes. We need to calculate the offset from buffer start.
3201 */
3202 if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3203 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3204 offset = do_div(logical_sb_block, blocksize);
3205 } else {
3206 logical_sb_block = sb_block;
3207 }
3208
3209 if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3210 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3211 goto out_fail;
3212 }
3213 /*
3214 * Note: s_es must be initialized as soon as possible because
3215 * some ext4 macro-instructions depend on its value
3216 */
3217 es = (struct ext4_super_block *) (bh->b_data + offset);
3218 sbi->s_es = es;
3219 sb->s_magic = le16_to_cpu(es->s_magic);
3220 if (sb->s_magic != EXT4_SUPER_MAGIC)
3221 goto cantfind_ext4;
3222 sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3223
3224 /* Warn if metadata_csum and gdt_csum are both set. */
3225 if (ext4_has_feature_metadata_csum(sb) &&
3226 ext4_has_feature_gdt_csum(sb))
3227 ext4_warning(sb, "metadata_csum and uninit_bg are "
3228 "redundant flags; please run fsck.");
3229
3230 /* Check for a known checksum algorithm */
3231 if (!ext4_verify_csum_type(sb, es)) {
3232 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3233 "unknown checksum algorithm.");
3234 silent = 1;
3235 goto cantfind_ext4;
3236 }
3237
3238 /* Load the checksum driver */
3239 if (ext4_has_feature_metadata_csum(sb)) {
3240 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3241 if (IS_ERR(sbi->s_chksum_driver)) {
3242 ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3243 ret = PTR_ERR(sbi->s_chksum_driver);
3244 sbi->s_chksum_driver = NULL;
3245 goto failed_mount;
3246 }
3247 }
3248
3249 /* Check superblock checksum */
3250 if (!ext4_superblock_csum_verify(sb, es)) {
3251 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3252 "invalid superblock checksum. Run e2fsck?");
3253 silent = 1;
3254 ret = -EFSBADCRC;
3255 goto cantfind_ext4;
3256 }
3257
3258 /* Precompute checksum seed for all metadata */
3259 if (ext4_has_feature_csum_seed(sb))
3260 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3261 else if (ext4_has_metadata_csum(sb))
3262 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3263 sizeof(es->s_uuid));
3264
3265 /* Set defaults before we parse the mount options */
3266 def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3267 set_opt(sb, INIT_INODE_TABLE);
3268 if (def_mount_opts & EXT4_DEFM_DEBUG)
3269 set_opt(sb, DEBUG);
3270 if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3271 set_opt(sb, GRPID);
3272 if (def_mount_opts & EXT4_DEFM_UID16)
3273 set_opt(sb, NO_UID32);
3274 /* xattr user namespace & acls are now defaulted on */
3275 set_opt(sb, XATTR_USER);
3276#ifdef CONFIG_EXT4_FS_POSIX_ACL
3277 set_opt(sb, POSIX_ACL);
3278#endif
3279 /* don't forget to enable journal_csum when metadata_csum is enabled. */
3280 if (ext4_has_metadata_csum(sb))
3281 set_opt(sb, JOURNAL_CHECKSUM);
3282
3283 if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3284 set_opt(sb, JOURNAL_DATA);
3285 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3286 set_opt(sb, ORDERED_DATA);
3287 else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3288 set_opt(sb, WRITEBACK_DATA);
3289
3290 if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3291 set_opt(sb, ERRORS_PANIC);
3292 else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3293 set_opt(sb, ERRORS_CONT);
3294 else
3295 set_opt(sb, ERRORS_RO);
3296 /* block_validity enabled by default; disable with noblock_validity */
3297 set_opt(sb, BLOCK_VALIDITY);
3298 if (def_mount_opts & EXT4_DEFM_DISCARD)
3299 set_opt(sb, DISCARD);
3300
3301 sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3302 sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3303 sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3304 sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3305 sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3306
3307 if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3308 set_opt(sb, BARRIER);
3309
3310 /*
3311 * enable delayed allocation by default
3312 * Use -o nodelalloc to turn it off
3313 */
3314 if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3315 ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3316 set_opt(sb, DELALLOC);
3317
3318 /*
3319 * set default s_li_wait_mult for lazyinit, for the case there is
3320 * no mount option specified.
3321 */
3322 sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3323
3324 if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3325 &journal_devnum, &journal_ioprio, 0)) {
3326 ext4_msg(sb, KERN_WARNING,
3327 "failed to parse options in superblock: %s",
3328 sbi->s_es->s_mount_opts);
3329 }
3330 sbi->s_def_mount_opt = sbi->s_mount_opt;
3331 if (!parse_options((char *) data, sb, &journal_devnum,
3332 &journal_ioprio, 0))
3333 goto failed_mount;
3334
3335 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3336 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3337 "with data=journal disables delayed "
3338 "allocation and O_DIRECT support!\n");
3339 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3340 ext4_msg(sb, KERN_ERR, "can't mount with "
3341 "both data=journal and delalloc");
3342 goto failed_mount;
3343 }
3344 if (test_opt(sb, DIOREAD_NOLOCK)) {
3345 ext4_msg(sb, KERN_ERR, "can't mount with "
3346 "both data=journal and dioread_nolock");
3347 goto failed_mount;
3348 }
3349 if (test_opt(sb, DAX)) {
3350 ext4_msg(sb, KERN_ERR, "can't mount with "
3351 "both data=journal and dax");
3352 goto failed_mount;
3353 }
3354 if (test_opt(sb, DELALLOC))
3355 clear_opt(sb, DELALLOC);
3356 } else {
3357 sb->s_iflags |= SB_I_CGROUPWB;
3358 }
3359
3360 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3361 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3362
3363 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3364 (ext4_has_compat_features(sb) ||
3365 ext4_has_ro_compat_features(sb) ||
3366 ext4_has_incompat_features(sb)))
3367 ext4_msg(sb, KERN_WARNING,
3368 "feature flags set on rev 0 fs, "
3369 "running e2fsck is recommended");
3370
3371 if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3372 set_opt2(sb, HURD_COMPAT);
3373 if (ext4_has_feature_64bit(sb)) {
3374 ext4_msg(sb, KERN_ERR,
3375 "The Hurd can't support 64-bit file systems");
3376 goto failed_mount;
3377 }
3378 }
3379
3380 if (IS_EXT2_SB(sb)) {
3381 if (ext2_feature_set_ok(sb))
3382 ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3383 "using the ext4 subsystem");
3384 else {
3385 ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3386 "to feature incompatibilities");
3387 goto failed_mount;
3388 }
3389 }
3390
3391 if (IS_EXT3_SB(sb)) {
3392 if (ext3_feature_set_ok(sb))
3393 ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3394 "using the ext4 subsystem");
3395 else {
3396 ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3397 "to feature incompatibilities");
3398 goto failed_mount;
3399 }
3400 }
3401
3402 /*
3403 * Check feature flags regardless of the revision level, since we
3404 * previously didn't change the revision level when setting the flags,
3405 * so there is a chance incompat flags are set on a rev 0 filesystem.
3406 */
3407 if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3408 goto failed_mount;
3409
3410 blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3411 if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3412 blocksize > EXT4_MAX_BLOCK_SIZE) {
3413 ext4_msg(sb, KERN_ERR,
3414 "Unsupported filesystem blocksize %d", blocksize);
3415 goto failed_mount;
3416 }
3417
3418 if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3419 if (blocksize != PAGE_SIZE) {
3420 ext4_msg(sb, KERN_ERR,
3421 "error: unsupported blocksize for dax");
3422 goto failed_mount;
3423 }
3424 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3425 ext4_msg(sb, KERN_ERR,
3426 "error: device does not support dax");
3427 goto failed_mount;
3428 }
3429 }
3430
3431 if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3432 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3433 es->s_encryption_level);
3434 goto failed_mount;
3435 }
3436
3437 if (sb->s_blocksize != blocksize) {
3438 /* Validate the filesystem blocksize */
3439 if (!sb_set_blocksize(sb, blocksize)) {
3440 ext4_msg(sb, KERN_ERR, "bad block size %d",
3441 blocksize);
3442 goto failed_mount;
3443 }
3444
3445 brelse(bh);
3446 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3447 offset = do_div(logical_sb_block, blocksize);
3448 bh = sb_bread_unmovable(sb, logical_sb_block);
3449 if (!bh) {
3450 ext4_msg(sb, KERN_ERR,
3451 "Can't read superblock on 2nd try");
3452 goto failed_mount;
3453 }
3454 es = (struct ext4_super_block *)(bh->b_data + offset);
3455 sbi->s_es = es;
3456 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3457 ext4_msg(sb, KERN_ERR,
3458 "Magic mismatch, very weird!");
3459 goto failed_mount;
3460 }
3461 }
3462
3463 has_huge_files = ext4_has_feature_huge_file(sb);
3464 sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3465 has_huge_files);
3466 sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3467
3468 if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3469 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3470 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3471 } else {
3472 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3473 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3474 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3475 (!is_power_of_2(sbi->s_inode_size)) ||
3476 (sbi->s_inode_size > blocksize)) {
3477 ext4_msg(sb, KERN_ERR,
3478 "unsupported inode size: %d",
3479 sbi->s_inode_size);
3480 goto failed_mount;
3481 }
3482 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3483 sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3484 }
3485
3486 sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3487 if (ext4_has_feature_64bit(sb)) {
3488 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3489 sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3490 !is_power_of_2(sbi->s_desc_size)) {
3491 ext4_msg(sb, KERN_ERR,
3492 "unsupported descriptor size %lu",
3493 sbi->s_desc_size);
3494 goto failed_mount;
3495 }
3496 } else
3497 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3498
3499 sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3500 sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3501 if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3502 goto cantfind_ext4;
3503
3504 sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3505 if (sbi->s_inodes_per_block == 0)
3506 goto cantfind_ext4;
3507 sbi->s_itb_per_group = sbi->s_inodes_per_group /
3508 sbi->s_inodes_per_block;
3509 sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3510 sbi->s_sbh = bh;
3511 sbi->s_mount_state = le16_to_cpu(es->s_state);
3512 sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3513 sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3514
3515 for (i = 0; i < 4; i++)
3516 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3517 sbi->s_def_hash_version = es->s_def_hash_version;
3518 if (ext4_has_feature_dir_index(sb)) {
3519 i = le32_to_cpu(es->s_flags);
3520 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3521 sbi->s_hash_unsigned = 3;
3522 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3523#ifdef __CHAR_UNSIGNED__
3524 if (!(sb->s_flags & MS_RDONLY))
3525 es->s_flags |=
3526 cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3527 sbi->s_hash_unsigned = 3;
3528#else
3529 if (!(sb->s_flags & MS_RDONLY))
3530 es->s_flags |=
3531 cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3532#endif
3533 }
3534 }
3535
3536 /* Handle clustersize */
3537 clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3538 has_bigalloc = ext4_has_feature_bigalloc(sb);
3539 if (has_bigalloc) {
3540 if (clustersize < blocksize) {
3541 ext4_msg(sb, KERN_ERR,
3542 "cluster size (%d) smaller than "
3543 "block size (%d)", clustersize, blocksize);
3544 goto failed_mount;
3545 }
3546 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3547 le32_to_cpu(es->s_log_block_size);
3548 sbi->s_clusters_per_group =
3549 le32_to_cpu(es->s_clusters_per_group);
3550 if (sbi->s_clusters_per_group > blocksize * 8) {
3551 ext4_msg(sb, KERN_ERR,
3552 "#clusters per group too big: %lu",
3553 sbi->s_clusters_per_group);
3554 goto failed_mount;
3555 }
3556 if (sbi->s_blocks_per_group !=
3557 (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3558 ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3559 "clusters per group (%lu) inconsistent",
3560 sbi->s_blocks_per_group,
3561 sbi->s_clusters_per_group);
3562 goto failed_mount;
3563 }
3564 } else {
3565 if (clustersize != blocksize) {
3566 ext4_warning(sb, "fragment/cluster size (%d) != "
3567 "block size (%d)", clustersize,
3568 blocksize);
3569 clustersize = blocksize;
3570 }
3571 if (sbi->s_blocks_per_group > blocksize * 8) {
3572 ext4_msg(sb, KERN_ERR,
3573 "#blocks per group too big: %lu",
3574 sbi->s_blocks_per_group);
3575 goto failed_mount;
3576 }
3577 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3578 sbi->s_cluster_bits = 0;
3579 }
3580 sbi->s_cluster_ratio = clustersize / blocksize;
3581
3582 if (sbi->s_inodes_per_group > blocksize * 8) {
3583 ext4_msg(sb, KERN_ERR,
3584 "#inodes per group too big: %lu",
3585 sbi->s_inodes_per_group);
3586 goto failed_mount;
3587 }
3588
3589 /* Do we have standard group size of clustersize * 8 blocks ? */
3590 if (sbi->s_blocks_per_group == clustersize << 3)
3591 set_opt2(sb, STD_GROUP_SIZE);
3592
3593 /*
3594 * Test whether we have more sectors than will fit in sector_t,
3595 * and whether the max offset is addressable by the page cache.
3596 */
3597 err = generic_check_addressable(sb->s_blocksize_bits,
3598 ext4_blocks_count(es));
3599 if (err) {
3600 ext4_msg(sb, KERN_ERR, "filesystem"
3601 " too large to mount safely on this system");
3602 if (sizeof(sector_t) < 8)
3603 ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3604 goto failed_mount;
3605 }
3606
3607 if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3608 goto cantfind_ext4;
3609
3610 /* check blocks count against device size */
3611 blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3612 if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3613 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3614 "exceeds size of device (%llu blocks)",
3615 ext4_blocks_count(es), blocks_count);
3616 goto failed_mount;
3617 }
3618
3619 /*
3620 * It makes no sense for the first data block to be beyond the end
3621 * of the filesystem.
3622 */
3623 if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3624 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3625 "block %u is beyond end of filesystem (%llu)",
3626 le32_to_cpu(es->s_first_data_block),
3627 ext4_blocks_count(es));
3628 goto failed_mount;
3629 }
3630 blocks_count = (ext4_blocks_count(es) -
3631 le32_to_cpu(es->s_first_data_block) +
3632 EXT4_BLOCKS_PER_GROUP(sb) - 1);
3633 do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3634 if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3635 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3636 "(block count %llu, first data block %u, "
3637 "blocks per group %lu)", sbi->s_groups_count,
3638 ext4_blocks_count(es),
3639 le32_to_cpu(es->s_first_data_block),
3640 EXT4_BLOCKS_PER_GROUP(sb));
3641 goto failed_mount;
3642 }
3643 sbi->s_groups_count = blocks_count;
3644 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3645 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3646 db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3647 EXT4_DESC_PER_BLOCK(sb);
3648 sbi->s_group_desc = ext4_kvmalloc(db_count *
3649 sizeof(struct buffer_head *),
3650 GFP_KERNEL);
3651 if (sbi->s_group_desc == NULL) {
3652 ext4_msg(sb, KERN_ERR, "not enough memory");
3653 ret = -ENOMEM;
3654 goto failed_mount;
3655 }
3656
3657 bgl_lock_init(sbi->s_blockgroup_lock);
3658
3659 for (i = 0; i < db_count; i++) {
3660 block = descriptor_loc(sb, logical_sb_block, i);
3661 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3662 if (!sbi->s_group_desc[i]) {
3663 ext4_msg(sb, KERN_ERR,
3664 "can't read group descriptor %d", i);
3665 db_count = i;
3666 goto failed_mount2;
3667 }
3668 }
3669 if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3670 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3671 ret = -EFSCORRUPTED;
3672 goto failed_mount2;
3673 }
3674
3675 sbi->s_gdb_count = db_count;
3676 get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3677 spin_lock_init(&sbi->s_next_gen_lock);
3678
3679 setup_timer(&sbi->s_err_report, print_daily_error_info,
3680 (unsigned long) sb);
3681
3682 /* Register extent status tree shrinker */
3683 if (ext4_es_register_shrinker(sbi))
3684 goto failed_mount3;
3685
3686 sbi->s_stripe = ext4_get_stripe_size(sbi);
3687 sbi->s_extent_max_zeroout_kb = 32;
3688
3689 /*
3690 * set up enough so that it can read an inode
3691 */
3692 sb->s_op = &ext4_sops;
3693 sb->s_export_op = &ext4_export_ops;
3694 sb->s_xattr = ext4_xattr_handlers;
3695#ifdef CONFIG_QUOTA
3696 sb->dq_op = &ext4_quota_operations;
3697 if (ext4_has_feature_quota(sb))
3698 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3699 else
3700 sb->s_qcop = &ext4_qctl_operations;
3701 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
3702#endif
3703 memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3704
3705 INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3706 mutex_init(&sbi->s_orphan_lock);
3707
3708 sb->s_root = NULL;
3709
3710 needs_recovery = (es->s_last_orphan != 0 ||
3711 ext4_has_feature_journal_needs_recovery(sb));
3712
3713 if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3714 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3715 goto failed_mount3a;
3716
3717 /*
3718 * The first inode we look at is the journal inode. Don't try
3719 * root first: it may be modified in the journal!
3720 */
3721 if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3722 if (ext4_load_journal(sb, es, journal_devnum))
3723 goto failed_mount3a;
3724 } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3725 ext4_has_feature_journal_needs_recovery(sb)) {
3726 ext4_msg(sb, KERN_ERR, "required journal recovery "
3727 "suppressed and not mounted read-only");
3728 goto failed_mount_wq;
3729 } else {
3730 /* Nojournal mode, all journal mount options are illegal */
3731 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
3732 ext4_msg(sb, KERN_ERR, "can't mount with "
3733 "journal_checksum, fs mounted w/o journal");
3734 goto failed_mount_wq;
3735 }
3736 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3737 ext4_msg(sb, KERN_ERR, "can't mount with "
3738 "journal_async_commit, fs mounted w/o journal");
3739 goto failed_mount_wq;
3740 }
3741 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
3742 ext4_msg(sb, KERN_ERR, "can't mount with "
3743 "commit=%lu, fs mounted w/o journal",
3744 sbi->s_commit_interval / HZ);
3745 goto failed_mount_wq;
3746 }
3747 if (EXT4_MOUNT_DATA_FLAGS &
3748 (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
3749 ext4_msg(sb, KERN_ERR, "can't mount with "
3750 "data=, fs mounted w/o journal");
3751 goto failed_mount_wq;
3752 }
3753 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
3754 clear_opt(sb, JOURNAL_CHECKSUM);
3755 clear_opt(sb, DATA_FLAGS);
3756 sbi->s_journal = NULL;
3757 needs_recovery = 0;
3758 goto no_journal;
3759 }
3760
3761 if (ext4_has_feature_64bit(sb) &&
3762 !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3763 JBD2_FEATURE_INCOMPAT_64BIT)) {
3764 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3765 goto failed_mount_wq;
3766 }
3767
3768 if (!set_journal_csum_feature_set(sb)) {
3769 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3770 "feature set");
3771 goto failed_mount_wq;
3772 }
3773
3774 /* We have now updated the journal if required, so we can
3775 * validate the data journaling mode. */
3776 switch (test_opt(sb, DATA_FLAGS)) {
3777 case 0:
3778 /* No mode set, assume a default based on the journal
3779 * capabilities: ORDERED_DATA if the journal can
3780 * cope, else JOURNAL_DATA
3781 */
3782 if (jbd2_journal_check_available_features
3783 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3784 set_opt(sb, ORDERED_DATA);
3785 else
3786 set_opt(sb, JOURNAL_DATA);
3787 break;
3788
3789 case EXT4_MOUNT_ORDERED_DATA:
3790 case EXT4_MOUNT_WRITEBACK_DATA:
3791 if (!jbd2_journal_check_available_features
3792 (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3793 ext4_msg(sb, KERN_ERR, "Journal does not support "
3794 "requested data journaling mode");
3795 goto failed_mount_wq;
3796 }
3797 default:
3798 break;
3799 }
3800 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3801
3802 sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3803
3804no_journal:
3805 sbi->s_mb_cache = ext4_xattr_create_cache();
3806 if (!sbi->s_mb_cache) {
3807 ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
3808 goto failed_mount_wq;
3809 }
3810
3811 if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
3812 (blocksize != PAGE_SIZE)) {
3813 ext4_msg(sb, KERN_ERR,
3814 "Unsupported blocksize for fs encryption");
3815 goto failed_mount_wq;
3816 }
3817
3818 if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
3819 !ext4_has_feature_encrypt(sb)) {
3820 ext4_set_feature_encrypt(sb);
3821 ext4_commit_super(sb, 1);
3822 }
3823
3824 /*
3825 * Get the # of file system overhead blocks from the
3826 * superblock if present.
3827 */
3828 if (es->s_overhead_clusters)
3829 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3830 else {
3831 err = ext4_calculate_overhead(sb);
3832 if (err)
3833 goto failed_mount_wq;
3834 }
3835
3836 /*
3837 * The maximum number of concurrent works can be high and
3838 * concurrency isn't really necessary. Limit it to 1.
3839 */
3840 EXT4_SB(sb)->rsv_conversion_wq =
3841 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3842 if (!EXT4_SB(sb)->rsv_conversion_wq) {
3843 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3844 ret = -ENOMEM;
3845 goto failed_mount4;
3846 }
3847
3848 /*
3849 * The jbd2_journal_load will have done any necessary log recovery,
3850 * so we can safely mount the rest of the filesystem now.
3851 */
3852
3853 root = ext4_iget(sb, EXT4_ROOT_INO);
3854 if (IS_ERR(root)) {
3855 ext4_msg(sb, KERN_ERR, "get root inode failed");
3856 ret = PTR_ERR(root);
3857 root = NULL;
3858 goto failed_mount4;
3859 }
3860 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3861 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3862 iput(root);
3863 goto failed_mount4;
3864 }
3865 sb->s_root = d_make_root(root);
3866 if (!sb->s_root) {
3867 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3868 ret = -ENOMEM;
3869 goto failed_mount4;
3870 }
3871
3872 if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3873 sb->s_flags |= MS_RDONLY;
3874
3875 /* determine the minimum size of new large inodes, if present */
3876 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3877 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3878 EXT4_GOOD_OLD_INODE_SIZE;
3879 if (ext4_has_feature_extra_isize(sb)) {
3880 if (sbi->s_want_extra_isize <
3881 le16_to_cpu(es->s_want_extra_isize))
3882 sbi->s_want_extra_isize =
3883 le16_to_cpu(es->s_want_extra_isize);
3884 if (sbi->s_want_extra_isize <
3885 le16_to_cpu(es->s_min_extra_isize))
3886 sbi->s_want_extra_isize =
3887 le16_to_cpu(es->s_min_extra_isize);
3888 }
3889 }
3890 /* Check if enough inode space is available */
3891 if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3892 sbi->s_inode_size) {
3893 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3894 EXT4_GOOD_OLD_INODE_SIZE;
3895 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3896 "available");
3897 }
3898
3899 ext4_set_resv_clusters(sb);
3900
3901 err = ext4_setup_system_zone(sb);
3902 if (err) {
3903 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3904 "zone (%d)", err);
3905 goto failed_mount4a;
3906 }
3907
3908 ext4_ext_init(sb);
3909 err = ext4_mb_init(sb);
3910 if (err) {
3911 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3912 err);
3913 goto failed_mount5;
3914 }
3915
3916 block = ext4_count_free_clusters(sb);
3917 ext4_free_blocks_count_set(sbi->s_es,
3918 EXT4_C2B(sbi, block));
3919 err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
3920 GFP_KERNEL);
3921 if (!err) {
3922 unsigned long freei = ext4_count_free_inodes(sb);
3923 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
3924 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
3925 GFP_KERNEL);
3926 }
3927 if (!err)
3928 err = percpu_counter_init(&sbi->s_dirs_counter,
3929 ext4_count_dirs(sb), GFP_KERNEL);
3930 if (!err)
3931 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
3932 GFP_KERNEL);
3933 if (err) {
3934 ext4_msg(sb, KERN_ERR, "insufficient memory");
3935 goto failed_mount6;
3936 }
3937
3938 if (ext4_has_feature_flex_bg(sb))
3939 if (!ext4_fill_flex_info(sb)) {
3940 ext4_msg(sb, KERN_ERR,
3941 "unable to initialize "
3942 "flex_bg meta info!");
3943 goto failed_mount6;
3944 }
3945
3946 err = ext4_register_li_request(sb, first_not_zeroed);
3947 if (err)
3948 goto failed_mount6;
3949
3950 err = ext4_register_sysfs(sb);
3951 if (err)
3952 goto failed_mount7;
3953
3954#ifdef CONFIG_QUOTA
3955 /* Enable quota usage during mount. */
3956 if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
3957 err = ext4_enable_quotas(sb);
3958 if (err)
3959 goto failed_mount8;
3960 }
3961#endif /* CONFIG_QUOTA */
3962
3963 EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3964 ext4_orphan_cleanup(sb, es);
3965 EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3966 if (needs_recovery) {
3967 ext4_msg(sb, KERN_INFO, "recovery complete");
3968 ext4_mark_recovery_complete(sb, es);
3969 }
3970 if (EXT4_SB(sb)->s_journal) {
3971 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3972 descr = " journalled data mode";
3973 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3974 descr = " ordered data mode";
3975 else
3976 descr = " writeback data mode";
3977 } else
3978 descr = "out journal";
3979
3980 if (test_opt(sb, DISCARD)) {
3981 struct request_queue *q = bdev_get_queue(sb->s_bdev);
3982 if (!blk_queue_discard(q))
3983 ext4_msg(sb, KERN_WARNING,
3984 "mounting with \"discard\" option, but "
3985 "the device does not support discard");
3986 }
3987
3988 if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
3989 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3990 "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3991 *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3992
3993 if (es->s_error_count)
3994 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3995
3996 /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
3997 ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
3998 ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
3999 ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4000
4001 kfree(orig_data);
4002 return 0;
4003
4004cantfind_ext4:
4005 if (!silent)
4006 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4007 goto failed_mount;
4008
4009#ifdef CONFIG_QUOTA
4010failed_mount8:
4011 ext4_unregister_sysfs(sb);
4012#endif
4013failed_mount7:
4014 ext4_unregister_li_request(sb);
4015failed_mount6:
4016 ext4_mb_release(sb);
4017 if (sbi->s_flex_groups)
4018 kvfree(sbi->s_flex_groups);
4019 percpu_counter_destroy(&sbi->s_freeclusters_counter);
4020 percpu_counter_destroy(&sbi->s_freeinodes_counter);
4021 percpu_counter_destroy(&sbi->s_dirs_counter);
4022 percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4023failed_mount5:
4024 ext4_ext_release(sb);
4025 ext4_release_system_zone(sb);
4026failed_mount4a:
4027 dput(sb->s_root);
4028 sb->s_root = NULL;
4029failed_mount4:
4030 ext4_msg(sb, KERN_ERR, "mount failed");
4031 if (EXT4_SB(sb)->rsv_conversion_wq)
4032 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4033failed_mount_wq:
4034 if (sbi->s_mb_cache) {
4035 ext4_xattr_destroy_cache(sbi->s_mb_cache);
4036 sbi->s_mb_cache = NULL;
4037 }
4038 if (sbi->s_journal) {
4039 jbd2_journal_destroy(sbi->s_journal);
4040 sbi->s_journal = NULL;
4041 }
4042failed_mount3a:
4043 ext4_es_unregister_shrinker(sbi);
4044failed_mount3:
4045 del_timer_sync(&sbi->s_err_report);
4046 if (sbi->s_mmp_tsk)
4047 kthread_stop(sbi->s_mmp_tsk);
4048failed_mount2:
4049 for (i = 0; i < db_count; i++)
4050 brelse(sbi->s_group_desc[i]);
4051 kvfree(sbi->s_group_desc);
4052failed_mount:
4053 if (sbi->s_chksum_driver)
4054 crypto_free_shash(sbi->s_chksum_driver);
4055#ifdef CONFIG_QUOTA
4056 for (i = 0; i < EXT4_MAXQUOTAS; i++)
4057 kfree(sbi->s_qf_names[i]);
4058#endif
4059 ext4_blkdev_remove(sbi);
4060 brelse(bh);
4061out_fail:
4062 sb->s_fs_info = NULL;
4063 kfree(sbi->s_blockgroup_lock);
4064 kfree(sbi);
4065out_free_orig:
4066 kfree(orig_data);
4067 return err ? err : ret;
4068}
4069
4070/*
4071 * Setup any per-fs journal parameters now. We'll do this both on
4072 * initial mount, once the journal has been initialised but before we've
4073 * done any recovery; and again on any subsequent remount.
4074 */
4075static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4076{
4077 struct ext4_sb_info *sbi = EXT4_SB(sb);
4078
4079 journal->j_commit_interval = sbi->s_commit_interval;
4080 journal->j_min_batch_time = sbi->s_min_batch_time;
4081 journal->j_max_batch_time = sbi->s_max_batch_time;
4082
4083 write_lock(&journal->j_state_lock);
4084 if (test_opt(sb, BARRIER))
4085 journal->j_flags |= JBD2_BARRIER;
4086 else
4087 journal->j_flags &= ~JBD2_BARRIER;
4088 if (test_opt(sb, DATA_ERR_ABORT))
4089 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4090 else
4091 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4092 write_unlock(&journal->j_state_lock);
4093}
4094
4095static journal_t *ext4_get_journal(struct super_block *sb,
4096 unsigned int journal_inum)
4097{
4098 struct inode *journal_inode;
4099 journal_t *journal;
4100
4101 BUG_ON(!ext4_has_feature_journal(sb));
4102
4103 /* First, test for the existence of a valid inode on disk. Bad
4104 * things happen if we iget() an unused inode, as the subsequent
4105 * iput() will try to delete it. */
4106
4107 journal_inode = ext4_iget(sb, journal_inum);
4108 if (IS_ERR(journal_inode)) {
4109 ext4_msg(sb, KERN_ERR, "no journal found");
4110 return NULL;
4111 }
4112 if (!journal_inode->i_nlink) {
4113 make_bad_inode(journal_inode);
4114 iput(journal_inode);
4115 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4116 return NULL;
4117 }
4118
4119 jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4120 journal_inode, journal_inode->i_size);
4121 if (!S_ISREG(journal_inode->i_mode)) {
4122 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4123 iput(journal_inode);
4124 return NULL;
4125 }
4126
4127 journal = jbd2_journal_init_inode(journal_inode);
4128 if (!journal) {
4129 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4130 iput(journal_inode);
4131 return NULL;
4132 }
4133 journal->j_private = sb;
4134 ext4_init_journal_params(sb, journal);
4135 return journal;
4136}
4137
4138static journal_t *ext4_get_dev_journal(struct super_block *sb,
4139 dev_t j_dev)
4140{
4141 struct buffer_head *bh;
4142 journal_t *journal;
4143 ext4_fsblk_t start;
4144 ext4_fsblk_t len;
4145 int hblock, blocksize;
4146 ext4_fsblk_t sb_block;
4147 unsigned long offset;
4148 struct ext4_super_block *es;
4149 struct block_device *bdev;
4150
4151 BUG_ON(!ext4_has_feature_journal(sb));
4152
4153 bdev = ext4_blkdev_get(j_dev, sb);
4154 if (bdev == NULL)
4155 return NULL;
4156
4157 blocksize = sb->s_blocksize;
4158 hblock = bdev_logical_block_size(bdev);
4159 if (blocksize < hblock) {
4160 ext4_msg(sb, KERN_ERR,
4161 "blocksize too small for journal device");
4162 goto out_bdev;
4163 }
4164
4165 sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4166 offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4167 set_blocksize(bdev, blocksize);
4168 if (!(bh = __bread(bdev, sb_block, blocksize))) {
4169 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4170 "external journal");
4171 goto out_bdev;
4172 }
4173
4174 es = (struct ext4_super_block *) (bh->b_data + offset);
4175 if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4176 !(le32_to_cpu(es->s_feature_incompat) &
4177 EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4178 ext4_msg(sb, KERN_ERR, "external journal has "
4179 "bad superblock");
4180 brelse(bh);
4181 goto out_bdev;
4182 }
4183
4184 if ((le32_to_cpu(es->s_feature_ro_compat) &
4185 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4186 es->s_checksum != ext4_superblock_csum(sb, es)) {
4187 ext4_msg(sb, KERN_ERR, "external journal has "
4188 "corrupt superblock");
4189 brelse(bh);
4190 goto out_bdev;
4191 }
4192
4193 if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4194 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4195 brelse(bh);
4196 goto out_bdev;
4197 }
4198
4199 len = ext4_blocks_count(es);
4200 start = sb_block + 1;
4201 brelse(bh); /* we're done with the superblock */
4202
4203 journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4204 start, len, blocksize);
4205 if (!journal) {
4206 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4207 goto out_bdev;
4208 }
4209 journal->j_private = sb;
4210 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4211 wait_on_buffer(journal->j_sb_buffer);
4212 if (!buffer_uptodate(journal->j_sb_buffer)) {
4213 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4214 goto out_journal;
4215 }
4216 if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4217 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4218 "user (unsupported) - %d",
4219 be32_to_cpu(journal->j_superblock->s_nr_users));
4220 goto out_journal;
4221 }
4222 EXT4_SB(sb)->journal_bdev = bdev;
4223 ext4_init_journal_params(sb, journal);
4224 return journal;
4225
4226out_journal:
4227 jbd2_journal_destroy(journal);
4228out_bdev:
4229 ext4_blkdev_put(bdev);
4230 return NULL;
4231}
4232
4233static int ext4_load_journal(struct super_block *sb,
4234 struct ext4_super_block *es,
4235 unsigned long journal_devnum)
4236{
4237 journal_t *journal;
4238 unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4239 dev_t journal_dev;
4240 int err = 0;
4241 int really_read_only;
4242
4243 BUG_ON(!ext4_has_feature_journal(sb));
4244
4245 if (journal_devnum &&
4246 journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4247 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4248 "numbers have changed");
4249 journal_dev = new_decode_dev(journal_devnum);
4250 } else
4251 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4252
4253 really_read_only = bdev_read_only(sb->s_bdev);
4254
4255 /*
4256 * Are we loading a blank journal or performing recovery after a
4257 * crash? For recovery, we need to check in advance whether we
4258 * can get read-write access to the device.
4259 */
4260 if (ext4_has_feature_journal_needs_recovery(sb)) {
4261 if (sb->s_flags & MS_RDONLY) {
4262 ext4_msg(sb, KERN_INFO, "INFO: recovery "
4263 "required on readonly filesystem");
4264 if (really_read_only) {
4265 ext4_msg(sb, KERN_ERR, "write access "
4266 "unavailable, cannot proceed");
4267 return -EROFS;
4268 }
4269 ext4_msg(sb, KERN_INFO, "write access will "
4270 "be enabled during recovery");
4271 }
4272 }
4273
4274 if (journal_inum && journal_dev) {
4275 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4276 "and inode journals!");
4277 return -EINVAL;
4278 }
4279
4280 if (journal_inum) {
4281 if (!(journal = ext4_get_journal(sb, journal_inum)))
4282 return -EINVAL;
4283 } else {
4284 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4285 return -EINVAL;
4286 }
4287
4288 if (!(journal->j_flags & JBD2_BARRIER))
4289 ext4_msg(sb, KERN_INFO, "barriers disabled");
4290
4291 if (!ext4_has_feature_journal_needs_recovery(sb))
4292 err = jbd2_journal_wipe(journal, !really_read_only);
4293 if (!err) {
4294 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4295 if (save)
4296 memcpy(save, ((char *) es) +
4297 EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4298 err = jbd2_journal_load(journal);
4299 if (save)
4300 memcpy(((char *) es) + EXT4_S_ERR_START,
4301 save, EXT4_S_ERR_LEN);
4302 kfree(save);
4303 }
4304
4305 if (err) {
4306 ext4_msg(sb, KERN_ERR, "error loading journal");
4307 jbd2_journal_destroy(journal);
4308 return err;
4309 }
4310
4311 EXT4_SB(sb)->s_journal = journal;
4312 ext4_clear_journal_err(sb, es);
4313
4314 if (!really_read_only && journal_devnum &&
4315 journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4316 es->s_journal_dev = cpu_to_le32(journal_devnum);
4317
4318 /* Make sure we flush the recovery flag to disk. */
4319 ext4_commit_super(sb, 1);
4320 }
4321
4322 return 0;
4323}
4324
4325static int ext4_commit_super(struct super_block *sb, int sync)
4326{
4327 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4328 struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4329 int error = 0;
4330
4331 if (!sbh || block_device_ejected(sb))
4332 return error;
4333 if (buffer_write_io_error(sbh)) {
4334 /*
4335 * Oh, dear. A previous attempt to write the
4336 * superblock failed. This could happen because the
4337 * USB device was yanked out. Or it could happen to
4338 * be a transient write error and maybe the block will
4339 * be remapped. Nothing we can do but to retry the
4340 * write and hope for the best.
4341 */
4342 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4343 "superblock detected");
4344 clear_buffer_write_io_error(sbh);
4345 set_buffer_uptodate(sbh);
4346 }
4347 /*
4348 * If the file system is mounted read-only, don't update the
4349 * superblock write time. This avoids updating the superblock
4350 * write time when we are mounting the root file system
4351 * read/only but we need to replay the journal; at that point,
4352 * for people who are east of GMT and who make their clock
4353 * tick in localtime for Windows bug-for-bug compatibility,
4354 * the clock is set in the future, and this will cause e2fsck
4355 * to complain and force a full file system check.
4356 */
4357 if (!(sb->s_flags & MS_RDONLY))
4358 es->s_wtime = cpu_to_le32(get_seconds());
4359 if (sb->s_bdev->bd_part)
4360 es->s_kbytes_written =
4361 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4362 ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4363 EXT4_SB(sb)->s_sectors_written_start) >> 1));
4364 else
4365 es->s_kbytes_written =
4366 cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4367 if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4368 ext4_free_blocks_count_set(es,
4369 EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4370 &EXT4_SB(sb)->s_freeclusters_counter)));
4371 if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4372 es->s_free_inodes_count =
4373 cpu_to_le32(percpu_counter_sum_positive(
4374 &EXT4_SB(sb)->s_freeinodes_counter));
4375 BUFFER_TRACE(sbh, "marking dirty");
4376 ext4_superblock_csum_set(sb);
4377 mark_buffer_dirty(sbh);
4378 if (sync) {
4379 error = __sync_dirty_buffer(sbh,
4380 test_opt(sb, BARRIER) ? WRITE_FUA : WRITE_SYNC);
4381 if (error)
4382 return error;
4383
4384 error = buffer_write_io_error(sbh);
4385 if (error) {
4386 ext4_msg(sb, KERN_ERR, "I/O error while writing "
4387 "superblock");
4388 clear_buffer_write_io_error(sbh);
4389 set_buffer_uptodate(sbh);
4390 }
4391 }
4392 return error;
4393}
4394
4395/*
4396 * Have we just finished recovery? If so, and if we are mounting (or
4397 * remounting) the filesystem readonly, then we will end up with a
4398 * consistent fs on disk. Record that fact.
4399 */
4400static void ext4_mark_recovery_complete(struct super_block *sb,
4401 struct ext4_super_block *es)
4402{
4403 journal_t *journal = EXT4_SB(sb)->s_journal;
4404
4405 if (!ext4_has_feature_journal(sb)) {
4406 BUG_ON(journal != NULL);
4407 return;
4408 }
4409 jbd2_journal_lock_updates(journal);
4410 if (jbd2_journal_flush(journal) < 0)
4411 goto out;
4412
4413 if (ext4_has_feature_journal_needs_recovery(sb) &&
4414 sb->s_flags & MS_RDONLY) {
4415 ext4_clear_feature_journal_needs_recovery(sb);
4416 ext4_commit_super(sb, 1);
4417 }
4418
4419out:
4420 jbd2_journal_unlock_updates(journal);
4421}
4422
4423/*
4424 * If we are mounting (or read-write remounting) a filesystem whose journal
4425 * has recorded an error from a previous lifetime, move that error to the
4426 * main filesystem now.
4427 */
4428static void ext4_clear_journal_err(struct super_block *sb,
4429 struct ext4_super_block *es)
4430{
4431 journal_t *journal;
4432 int j_errno;
4433 const char *errstr;
4434
4435 BUG_ON(!ext4_has_feature_journal(sb));
4436
4437 journal = EXT4_SB(sb)->s_journal;
4438
4439 /*
4440 * Now check for any error status which may have been recorded in the
4441 * journal by a prior ext4_error() or ext4_abort()
4442 */
4443
4444 j_errno = jbd2_journal_errno(journal);
4445 if (j_errno) {
4446 char nbuf[16];
4447
4448 errstr = ext4_decode_error(sb, j_errno, nbuf);
4449 ext4_warning(sb, "Filesystem error recorded "
4450 "from previous mount: %s", errstr);
4451 ext4_warning(sb, "Marking fs in need of filesystem check.");
4452
4453 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4454 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4455 ext4_commit_super(sb, 1);
4456
4457 jbd2_journal_clear_err(journal);
4458 jbd2_journal_update_sb_errno(journal);
4459 }
4460}
4461
4462/*
4463 * Force the running and committing transactions to commit,
4464 * and wait on the commit.
4465 */
4466int ext4_force_commit(struct super_block *sb)
4467{
4468 journal_t *journal;
4469
4470 if (sb->s_flags & MS_RDONLY)
4471 return 0;
4472
4473 journal = EXT4_SB(sb)->s_journal;
4474 return ext4_journal_force_commit(journal);
4475}
4476
4477static int ext4_sync_fs(struct super_block *sb, int wait)
4478{
4479 int ret = 0;
4480 tid_t target;
4481 bool needs_barrier = false;
4482 struct ext4_sb_info *sbi = EXT4_SB(sb);
4483
4484 trace_ext4_sync_fs(sb, wait);
4485 flush_workqueue(sbi->rsv_conversion_wq);
4486 /*
4487 * Writeback quota in non-journalled quota case - journalled quota has
4488 * no dirty dquots
4489 */
4490 dquot_writeback_dquots(sb, -1);
4491 /*
4492 * Data writeback is possible w/o journal transaction, so barrier must
4493 * being sent at the end of the function. But we can skip it if
4494 * transaction_commit will do it for us.
4495 */
4496 if (sbi->s_journal) {
4497 target = jbd2_get_latest_transaction(sbi->s_journal);
4498 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4499 !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4500 needs_barrier = true;
4501
4502 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4503 if (wait)
4504 ret = jbd2_log_wait_commit(sbi->s_journal,
4505 target);
4506 }
4507 } else if (wait && test_opt(sb, BARRIER))
4508 needs_barrier = true;
4509 if (needs_barrier) {
4510 int err;
4511 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4512 if (!ret)
4513 ret = err;
4514 }
4515
4516 return ret;
4517}
4518
4519/*
4520 * LVM calls this function before a (read-only) snapshot is created. This
4521 * gives us a chance to flush the journal completely and mark the fs clean.
4522 *
4523 * Note that only this function cannot bring a filesystem to be in a clean
4524 * state independently. It relies on upper layer to stop all data & metadata
4525 * modifications.
4526 */
4527static int ext4_freeze(struct super_block *sb)
4528{
4529 int error = 0;
4530 journal_t *journal;
4531
4532 if (sb->s_flags & MS_RDONLY)
4533 return 0;
4534
4535 journal = EXT4_SB(sb)->s_journal;
4536
4537 if (journal) {
4538 /* Now we set up the journal barrier. */
4539 jbd2_journal_lock_updates(journal);
4540
4541 /*
4542 * Don't clear the needs_recovery flag if we failed to
4543 * flush the journal.
4544 */
4545 error = jbd2_journal_flush(journal);
4546 if (error < 0)
4547 goto out;
4548
4549 /* Journal blocked and flushed, clear needs_recovery flag. */
4550 ext4_clear_feature_journal_needs_recovery(sb);
4551 }
4552
4553 error = ext4_commit_super(sb, 1);
4554out:
4555 if (journal)
4556 /* we rely on upper layer to stop further updates */
4557 jbd2_journal_unlock_updates(journal);
4558 return error;
4559}
4560
4561/*
4562 * Called by LVM after the snapshot is done. We need to reset the RECOVER
4563 * flag here, even though the filesystem is not technically dirty yet.
4564 */
4565static int ext4_unfreeze(struct super_block *sb)
4566{
4567 if (sb->s_flags & MS_RDONLY)
4568 return 0;
4569
4570 if (EXT4_SB(sb)->s_journal) {
4571 /* Reset the needs_recovery flag before the fs is unlocked. */
4572 ext4_set_feature_journal_needs_recovery(sb);
4573 }
4574
4575 ext4_commit_super(sb, 1);
4576 return 0;
4577}
4578
4579/*
4580 * Structure to save mount options for ext4_remount's benefit
4581 */
4582struct ext4_mount_options {
4583 unsigned long s_mount_opt;
4584 unsigned long s_mount_opt2;
4585 kuid_t s_resuid;
4586 kgid_t s_resgid;
4587 unsigned long s_commit_interval;
4588 u32 s_min_batch_time, s_max_batch_time;
4589#ifdef CONFIG_QUOTA
4590 int s_jquota_fmt;
4591 char *s_qf_names[EXT4_MAXQUOTAS];
4592#endif
4593};
4594
4595static int ext4_remount(struct super_block *sb, int *flags, char *data)
4596{
4597 struct ext4_super_block *es;
4598 struct ext4_sb_info *sbi = EXT4_SB(sb);
4599 unsigned long old_sb_flags;
4600 struct ext4_mount_options old_opts;
4601 int enable_quota = 0;
4602 ext4_group_t g;
4603 unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4604 int err = 0;
4605#ifdef CONFIG_QUOTA
4606 int i, j;
4607#endif
4608 char *orig_data = kstrdup(data, GFP_KERNEL);
4609
4610 /* Store the original options */
4611 old_sb_flags = sb->s_flags;
4612 old_opts.s_mount_opt = sbi->s_mount_opt;
4613 old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4614 old_opts.s_resuid = sbi->s_resuid;
4615 old_opts.s_resgid = sbi->s_resgid;
4616 old_opts.s_commit_interval = sbi->s_commit_interval;
4617 old_opts.s_min_batch_time = sbi->s_min_batch_time;
4618 old_opts.s_max_batch_time = sbi->s_max_batch_time;
4619#ifdef CONFIG_QUOTA
4620 old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4621 for (i = 0; i < EXT4_MAXQUOTAS; i++)
4622 if (sbi->s_qf_names[i]) {
4623 old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4624 GFP_KERNEL);
4625 if (!old_opts.s_qf_names[i]) {
4626 for (j = 0; j < i; j++)
4627 kfree(old_opts.s_qf_names[j]);
4628 kfree(orig_data);
4629 return -ENOMEM;
4630 }
4631 } else
4632 old_opts.s_qf_names[i] = NULL;
4633#endif
4634 if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4635 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4636
4637 if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4638 err = -EINVAL;
4639 goto restore_opts;
4640 }
4641
4642 if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4643 test_opt(sb, JOURNAL_CHECKSUM)) {
4644 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4645 "during remount not supported; ignoring");
4646 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4647 }
4648
4649 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4650 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4651 ext4_msg(sb, KERN_ERR, "can't mount with "
4652 "both data=journal and delalloc");
4653 err = -EINVAL;
4654 goto restore_opts;
4655 }
4656 if (test_opt(sb, DIOREAD_NOLOCK)) {
4657 ext4_msg(sb, KERN_ERR, "can't mount with "
4658 "both data=journal and dioread_nolock");
4659 err = -EINVAL;
4660 goto restore_opts;
4661 }
4662 if (test_opt(sb, DAX)) {
4663 ext4_msg(sb, KERN_ERR, "can't mount with "
4664 "both data=journal and dax");
4665 err = -EINVAL;
4666 goto restore_opts;
4667 }
4668 }
4669
4670 if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4671 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4672 "dax flag with busy inodes while remounting");
4673 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4674 }
4675
4676 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4677 ext4_abort(sb, "Abort forced by user");
4678
4679 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4680 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4681
4682 es = sbi->s_es;
4683
4684 if (sbi->s_journal) {
4685 ext4_init_journal_params(sb, sbi->s_journal);
4686 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4687 }
4688
4689 if (*flags & MS_LAZYTIME)
4690 sb->s_flags |= MS_LAZYTIME;
4691
4692 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4693 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4694 err = -EROFS;
4695 goto restore_opts;
4696 }
4697
4698 if (*flags & MS_RDONLY) {
4699 err = sync_filesystem(sb);
4700 if (err < 0)
4701 goto restore_opts;
4702 err = dquot_suspend(sb, -1);
4703 if (err < 0)
4704 goto restore_opts;
4705
4706 /*
4707 * First of all, the unconditional stuff we have to do
4708 * to disable replay of the journal when we next remount
4709 */
4710 sb->s_flags |= MS_RDONLY;
4711
4712 /*
4713 * OK, test if we are remounting a valid rw partition
4714 * readonly, and if so set the rdonly flag and then
4715 * mark the partition as valid again.
4716 */
4717 if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4718 (sbi->s_mount_state & EXT4_VALID_FS))
4719 es->s_state = cpu_to_le16(sbi->s_mount_state);
4720
4721 if (sbi->s_journal)
4722 ext4_mark_recovery_complete(sb, es);
4723 } else {
4724 /* Make sure we can mount this feature set readwrite */
4725 if (ext4_has_feature_readonly(sb) ||
4726 !ext4_feature_set_ok(sb, 0)) {
4727 err = -EROFS;
4728 goto restore_opts;
4729 }
4730 /*
4731 * Make sure the group descriptor checksums
4732 * are sane. If they aren't, refuse to remount r/w.
4733 */
4734 for (g = 0; g < sbi->s_groups_count; g++) {
4735 struct ext4_group_desc *gdp =
4736 ext4_get_group_desc(sb, g, NULL);
4737
4738 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4739 ext4_msg(sb, KERN_ERR,
4740 "ext4_remount: Checksum for group %u failed (%u!=%u)",
4741 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
4742 le16_to_cpu(gdp->bg_checksum));
4743 err = -EFSBADCRC;
4744 goto restore_opts;
4745 }
4746 }
4747
4748 /*
4749 * If we have an unprocessed orphan list hanging
4750 * around from a previously readonly bdev mount,
4751 * require a full umount/remount for now.
4752 */
4753 if (es->s_last_orphan) {
4754 ext4_msg(sb, KERN_WARNING, "Couldn't "
4755 "remount RDWR because of unprocessed "
4756 "orphan inode list. Please "
4757 "umount/remount instead");
4758 err = -EINVAL;
4759 goto restore_opts;
4760 }
4761
4762 /*
4763 * Mounting a RDONLY partition read-write, so reread
4764 * and store the current valid flag. (It may have
4765 * been changed by e2fsck since we originally mounted
4766 * the partition.)
4767 */
4768 if (sbi->s_journal)
4769 ext4_clear_journal_err(sb, es);
4770 sbi->s_mount_state = le16_to_cpu(es->s_state);
4771 if (!ext4_setup_super(sb, es, 0))
4772 sb->s_flags &= ~MS_RDONLY;
4773 if (ext4_has_feature_mmp(sb))
4774 if (ext4_multi_mount_protect(sb,
4775 le64_to_cpu(es->s_mmp_block))) {
4776 err = -EROFS;
4777 goto restore_opts;
4778 }
4779 enable_quota = 1;
4780 }
4781 }
4782
4783 /*
4784 * Reinitialize lazy itable initialization thread based on
4785 * current settings
4786 */
4787 if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4788 ext4_unregister_li_request(sb);
4789 else {
4790 ext4_group_t first_not_zeroed;
4791 first_not_zeroed = ext4_has_uninit_itable(sb);
4792 ext4_register_li_request(sb, first_not_zeroed);
4793 }
4794
4795 ext4_setup_system_zone(sb);
4796 if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4797 ext4_commit_super(sb, 1);
4798
4799#ifdef CONFIG_QUOTA
4800 /* Release old quota file names */
4801 for (i = 0; i < EXT4_MAXQUOTAS; i++)
4802 kfree(old_opts.s_qf_names[i]);
4803 if (enable_quota) {
4804 if (sb_any_quota_suspended(sb))
4805 dquot_resume(sb, -1);
4806 else if (ext4_has_feature_quota(sb)) {
4807 err = ext4_enable_quotas(sb);
4808 if (err)
4809 goto restore_opts;
4810 }
4811 }
4812#endif
4813
4814 *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
4815 ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4816 kfree(orig_data);
4817 return 0;
4818
4819restore_opts:
4820 sb->s_flags = old_sb_flags;
4821 sbi->s_mount_opt = old_opts.s_mount_opt;
4822 sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4823 sbi->s_resuid = old_opts.s_resuid;
4824 sbi->s_resgid = old_opts.s_resgid;
4825 sbi->s_commit_interval = old_opts.s_commit_interval;
4826 sbi->s_min_batch_time = old_opts.s_min_batch_time;
4827 sbi->s_max_batch_time = old_opts.s_max_batch_time;
4828#ifdef CONFIG_QUOTA
4829 sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4830 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
4831 kfree(sbi->s_qf_names[i]);
4832 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4833 }
4834#endif
4835 kfree(orig_data);
4836 return err;
4837}
4838
4839#ifdef CONFIG_QUOTA
4840static int ext4_statfs_project(struct super_block *sb,
4841 kprojid_t projid, struct kstatfs *buf)
4842{
4843 struct kqid qid;
4844 struct dquot *dquot;
4845 u64 limit;
4846 u64 curblock;
4847
4848 qid = make_kqid_projid(projid);
4849 dquot = dqget(sb, qid);
4850 if (IS_ERR(dquot))
4851 return PTR_ERR(dquot);
4852 spin_lock(&dq_data_lock);
4853
4854 limit = (dquot->dq_dqb.dqb_bsoftlimit ?
4855 dquot->dq_dqb.dqb_bsoftlimit :
4856 dquot->dq_dqb.dqb_bhardlimit) >> sb->s_blocksize_bits;
4857 if (limit && buf->f_blocks > limit) {
4858 curblock = dquot->dq_dqb.dqb_curspace >> sb->s_blocksize_bits;
4859 buf->f_blocks = limit;
4860 buf->f_bfree = buf->f_bavail =
4861 (buf->f_blocks > curblock) ?
4862 (buf->f_blocks - curblock) : 0;
4863 }
4864
4865 limit = dquot->dq_dqb.dqb_isoftlimit ?
4866 dquot->dq_dqb.dqb_isoftlimit :
4867 dquot->dq_dqb.dqb_ihardlimit;
4868 if (limit && buf->f_files > limit) {
4869 buf->f_files = limit;
4870 buf->f_ffree =
4871 (buf->f_files > dquot->dq_dqb.dqb_curinodes) ?
4872 (buf->f_files - dquot->dq_dqb.dqb_curinodes) : 0;
4873 }
4874
4875 spin_unlock(&dq_data_lock);
4876 dqput(dquot);
4877 return 0;
4878}
4879#endif
4880
4881static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4882{
4883 struct super_block *sb = dentry->d_sb;
4884 struct ext4_sb_info *sbi = EXT4_SB(sb);
4885 struct ext4_super_block *es = sbi->s_es;
4886 ext4_fsblk_t overhead = 0, resv_blocks;
4887 u64 fsid;
4888 s64 bfree;
4889 resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4890
4891 if (!test_opt(sb, MINIX_DF))
4892 overhead = sbi->s_overhead;
4893
4894 buf->f_type = EXT4_SUPER_MAGIC;
4895 buf->f_bsize = sb->s_blocksize;
4896 buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4897 bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4898 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4899 /* prevent underflow in case that few free space is available */
4900 buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4901 buf->f_bavail = buf->f_bfree -
4902 (ext4_r_blocks_count(es) + resv_blocks);
4903 if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4904 buf->f_bavail = 0;
4905 buf->f_files = le32_to_cpu(es->s_inodes_count);
4906 buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4907 buf->f_namelen = EXT4_NAME_LEN;
4908 fsid = le64_to_cpup((void *)es->s_uuid) ^
4909 le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4910 buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4911 buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4912
4913#ifdef CONFIG_QUOTA
4914 if (ext4_test_inode_flag(dentry->d_inode, EXT4_INODE_PROJINHERIT) &&
4915 sb_has_quota_limits_enabled(sb, PRJQUOTA))
4916 ext4_statfs_project(sb, EXT4_I(dentry->d_inode)->i_projid, buf);
4917#endif
4918 return 0;
4919}
4920
4921/* Helper function for writing quotas on sync - we need to start transaction
4922 * before quota file is locked for write. Otherwise the are possible deadlocks:
4923 * Process 1 Process 2
4924 * ext4_create() quota_sync()
4925 * jbd2_journal_start() write_dquot()
4926 * dquot_initialize() down(dqio_mutex)
4927 * down(dqio_mutex) jbd2_journal_start()
4928 *
4929 */
4930
4931#ifdef CONFIG_QUOTA
4932
4933static inline struct inode *dquot_to_inode(struct dquot *dquot)
4934{
4935 return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4936}
4937
4938static int ext4_write_dquot(struct dquot *dquot)
4939{
4940 int ret, err;
4941 handle_t *handle;
4942 struct inode *inode;
4943
4944 inode = dquot_to_inode(dquot);
4945 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4946 EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4947 if (IS_ERR(handle))
4948 return PTR_ERR(handle);
4949 ret = dquot_commit(dquot);
4950 err = ext4_journal_stop(handle);
4951 if (!ret)
4952 ret = err;
4953 return ret;
4954}
4955
4956static int ext4_acquire_dquot(struct dquot *dquot)
4957{
4958 int ret, err;
4959 handle_t *handle;
4960
4961 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4962 EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4963 if (IS_ERR(handle))
4964 return PTR_ERR(handle);
4965 ret = dquot_acquire(dquot);
4966 err = ext4_journal_stop(handle);
4967 if (!ret)
4968 ret = err;
4969 return ret;
4970}
4971
4972static int ext4_release_dquot(struct dquot *dquot)
4973{
4974 int ret, err;
4975 handle_t *handle;
4976
4977 handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4978 EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4979 if (IS_ERR(handle)) {
4980 /* Release dquot anyway to avoid endless cycle in dqput() */
4981 dquot_release(dquot);
4982 return PTR_ERR(handle);
4983 }
4984 ret = dquot_release(dquot);
4985 err = ext4_journal_stop(handle);
4986 if (!ret)
4987 ret = err;
4988 return ret;
4989}
4990
4991static int ext4_mark_dquot_dirty(struct dquot *dquot)
4992{
4993 struct super_block *sb = dquot->dq_sb;
4994 struct ext4_sb_info *sbi = EXT4_SB(sb);
4995
4996 /* Are we journaling quotas? */
4997 if (ext4_has_feature_quota(sb) ||
4998 sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
4999 dquot_mark_dquot_dirty(dquot);
5000 return ext4_write_dquot(dquot);
5001 } else {
5002 return dquot_mark_dquot_dirty(dquot);
5003 }
5004}
5005
5006static int ext4_write_info(struct super_block *sb, int type)
5007{
5008 int ret, err;
5009 handle_t *handle;
5010
5011 /* Data block + inode block */
5012 handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5013 if (IS_ERR(handle))
5014 return PTR_ERR(handle);
5015 ret = dquot_commit_info(sb, type);
5016 err = ext4_journal_stop(handle);
5017 if (!ret)
5018 ret = err;
5019 return ret;
5020}
5021
5022/*
5023 * Turn on quotas during mount time - we need to find
5024 * the quota file and such...
5025 */
5026static int ext4_quota_on_mount(struct super_block *sb, int type)
5027{
5028 return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5029 EXT4_SB(sb)->s_jquota_fmt, type);
5030}
5031
5032static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5033{
5034 struct ext4_inode_info *ei = EXT4_I(inode);
5035
5036 /* The first argument of lockdep_set_subclass has to be
5037 * *exactly* the same as the argument to init_rwsem() --- in
5038 * this case, in init_once() --- or lockdep gets unhappy
5039 * because the name of the lock is set using the
5040 * stringification of the argument to init_rwsem().
5041 */
5042 (void) ei; /* shut up clang warning if !CONFIG_LOCKDEP */
5043 lockdep_set_subclass(&ei->i_data_sem, subclass);
5044}
5045
5046/*
5047 * Standard function to be called on quota_on
5048 */
5049static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5050 struct path *path)
5051{
5052 int err;
5053
5054 if (!test_opt(sb, QUOTA))
5055 return -EINVAL;
5056
5057 /* Quotafile not on the same filesystem? */
5058 if (path->dentry->d_sb != sb)
5059 return -EXDEV;
5060 /* Journaling quota? */
5061 if (EXT4_SB(sb)->s_qf_names[type]) {
5062 /* Quotafile not in fs root? */
5063 if (path->dentry->d_parent != sb->s_root)
5064 ext4_msg(sb, KERN_WARNING,
5065 "Quota file not on filesystem root. "
5066 "Journaled quota will not work");
5067 }
5068
5069 /*
5070 * When we journal data on quota file, we have to flush journal to see
5071 * all updates to the file when we bypass pagecache...
5072 */
5073 if (EXT4_SB(sb)->s_journal &&
5074 ext4_should_journal_data(d_inode(path->dentry))) {
5075 /*
5076 * We don't need to lock updates but journal_flush() could
5077 * otherwise be livelocked...
5078 */
5079 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5080 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5081 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5082 if (err)
5083 return err;
5084 }
5085 lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5086 err = dquot_quota_on(sb, type, format_id, path);
5087 if (err)
5088 lockdep_set_quota_inode(path->dentry->d_inode,
5089 I_DATA_SEM_NORMAL);
5090 return err;
5091}
5092
5093static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5094 unsigned int flags)
5095{
5096 int err;
5097 struct inode *qf_inode;
5098 unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5099 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5100 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5101 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5102 };
5103
5104 BUG_ON(!ext4_has_feature_quota(sb));
5105
5106 if (!qf_inums[type])
5107 return -EPERM;
5108
5109 qf_inode = ext4_iget(sb, qf_inums[type]);
5110 if (IS_ERR(qf_inode)) {
5111 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5112 return PTR_ERR(qf_inode);
5113 }
5114
5115 /* Don't account quota for quota files to avoid recursion */
5116 qf_inode->i_flags |= S_NOQUOTA;
5117 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5118 err = dquot_enable(qf_inode, type, format_id, flags);
5119 iput(qf_inode);
5120 if (err)
5121 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5122
5123 return err;
5124}
5125
5126/* Enable usage tracking for all quota types. */
5127static int ext4_enable_quotas(struct super_block *sb)
5128{
5129 int type, err = 0;
5130 unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5131 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5132 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum),
5133 le32_to_cpu(EXT4_SB(sb)->s_es->s_prj_quota_inum)
5134 };
5135
5136 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5137 for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5138 if (qf_inums[type]) {
5139 err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5140 DQUOT_USAGE_ENABLED);
5141 if (err) {
5142 ext4_warning(sb,
5143 "Failed to enable quota tracking "
5144 "(type=%d, err=%d). Please run "
5145 "e2fsck to fix.", type, err);
5146 return err;
5147 }
5148 }
5149 }
5150 return 0;
5151}
5152
5153static int ext4_quota_off(struct super_block *sb, int type)
5154{
5155 struct inode *inode = sb_dqopt(sb)->files[type];
5156 handle_t *handle;
5157
5158 /* Force all delayed allocation blocks to be allocated.
5159 * Caller already holds s_umount sem */
5160 if (test_opt(sb, DELALLOC))
5161 sync_filesystem(sb);
5162
5163 if (!inode)
5164 goto out;
5165
5166 /* Update modification times of quota files when userspace can
5167 * start looking at them */
5168 handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5169 if (IS_ERR(handle))
5170 goto out;
5171 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5172 ext4_mark_inode_dirty(handle, inode);
5173 ext4_journal_stop(handle);
5174
5175out:
5176 return dquot_quota_off(sb, type);
5177}
5178
5179/* Read data from quotafile - avoid pagecache and such because we cannot afford
5180 * acquiring the locks... As quota files are never truncated and quota code
5181 * itself serializes the operations (and no one else should touch the files)
5182 * we don't have to be afraid of races */
5183static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5184 size_t len, loff_t off)
5185{
5186 struct inode *inode = sb_dqopt(sb)->files[type];
5187 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5188 int offset = off & (sb->s_blocksize - 1);
5189 int tocopy;
5190 size_t toread;
5191 struct buffer_head *bh;
5192 loff_t i_size = i_size_read(inode);
5193
5194 if (off > i_size)
5195 return 0;
5196 if (off+len > i_size)
5197 len = i_size-off;
5198 toread = len;
5199 while (toread > 0) {
5200 tocopy = sb->s_blocksize - offset < toread ?
5201 sb->s_blocksize - offset : toread;
5202 bh = ext4_bread(NULL, inode, blk, 0);
5203 if (IS_ERR(bh))
5204 return PTR_ERR(bh);
5205 if (!bh) /* A hole? */
5206 memset(data, 0, tocopy);
5207 else
5208 memcpy(data, bh->b_data+offset, tocopy);
5209 brelse(bh);
5210 offset = 0;
5211 toread -= tocopy;
5212 data += tocopy;
5213 blk++;
5214 }
5215 return len;
5216}
5217
5218/* Write to quotafile (we know the transaction is already started and has
5219 * enough credits) */
5220static ssize_t ext4_quota_write(struct super_block *sb, int type,
5221 const char *data, size_t len, loff_t off)
5222{
5223 struct inode *inode = sb_dqopt(sb)->files[type];
5224 ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5225 int err, offset = off & (sb->s_blocksize - 1);
5226 int retries = 0;
5227 struct buffer_head *bh;
5228 handle_t *handle = journal_current_handle();
5229
5230 if (EXT4_SB(sb)->s_journal && !handle) {
5231 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5232 " cancelled because transaction is not started",
5233 (unsigned long long)off, (unsigned long long)len);
5234 return -EIO;
5235 }
5236 /*
5237 * Since we account only one data block in transaction credits,
5238 * then it is impossible to cross a block boundary.
5239 */
5240 if (sb->s_blocksize - offset < len) {
5241 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5242 " cancelled because not block aligned",
5243 (unsigned long long)off, (unsigned long long)len);
5244 return -EIO;
5245 }
5246
5247 do {
5248 bh = ext4_bread(handle, inode, blk,
5249 EXT4_GET_BLOCKS_CREATE |
5250 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5251 } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5252 ext4_should_retry_alloc(inode->i_sb, &retries));
5253 if (IS_ERR(bh))
5254 return PTR_ERR(bh);
5255 if (!bh)
5256 goto out;
5257 BUFFER_TRACE(bh, "get write access");
5258 err = ext4_journal_get_write_access(handle, bh);
5259 if (err) {
5260 brelse(bh);
5261 return err;
5262 }
5263 lock_buffer(bh);
5264 memcpy(bh->b_data+offset, data, len);
5265 flush_dcache_page(bh->b_page);
5266 unlock_buffer(bh);
5267 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5268 brelse(bh);
5269out:
5270 if (inode->i_size < off + len) {
5271 i_size_write(inode, off + len);
5272 EXT4_I(inode)->i_disksize = inode->i_size;
5273 ext4_mark_inode_dirty(handle, inode);
5274 }
5275 return len;
5276}
5277
5278static int ext4_get_next_id(struct super_block *sb, struct kqid *qid)
5279{
5280 const struct quota_format_ops *ops;
5281
5282 if (!sb_has_quota_loaded(sb, qid->type))
5283 return -ESRCH;
5284 ops = sb_dqopt(sb)->ops[qid->type];
5285 if (!ops || !ops->get_next_id)
5286 return -ENOSYS;
5287 return dquot_get_next_id(sb, qid);
5288}
5289#endif
5290
5291static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5292 const char *dev_name, void *data)
5293{
5294 return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5295}
5296
5297#if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5298static inline void register_as_ext2(void)
5299{
5300 int err = register_filesystem(&ext2_fs_type);
5301 if (err)
5302 printk(KERN_WARNING
5303 "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5304}
5305
5306static inline void unregister_as_ext2(void)
5307{
5308 unregister_filesystem(&ext2_fs_type);
5309}
5310
5311static inline int ext2_feature_set_ok(struct super_block *sb)
5312{
5313 if (ext4_has_unknown_ext2_incompat_features(sb))
5314 return 0;
5315 if (sb->s_flags & MS_RDONLY)
5316 return 1;
5317 if (ext4_has_unknown_ext2_ro_compat_features(sb))
5318 return 0;
5319 return 1;
5320}
5321#else
5322static inline void register_as_ext2(void) { }
5323static inline void unregister_as_ext2(void) { }
5324static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5325#endif
5326
5327static inline void register_as_ext3(void)
5328{
5329 int err = register_filesystem(&ext3_fs_type);
5330 if (err)
5331 printk(KERN_WARNING
5332 "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5333}
5334
5335static inline void unregister_as_ext3(void)
5336{
5337 unregister_filesystem(&ext3_fs_type);
5338}
5339
5340static inline int ext3_feature_set_ok(struct super_block *sb)
5341{
5342 if (ext4_has_unknown_ext3_incompat_features(sb))
5343 return 0;
5344 if (!ext4_has_feature_journal(sb))
5345 return 0;
5346 if (sb->s_flags & MS_RDONLY)
5347 return 1;
5348 if (ext4_has_unknown_ext3_ro_compat_features(sb))
5349 return 0;
5350 return 1;
5351}
5352
5353static struct file_system_type ext4_fs_type = {
5354 .owner = THIS_MODULE,
5355 .name = "ext4",
5356 .mount = ext4_mount,
5357 .kill_sb = kill_block_super,
5358 .fs_flags = FS_REQUIRES_DEV,
5359};
5360MODULE_ALIAS_FS("ext4");
5361
5362/* Shared across all ext4 file systems */
5363wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5364
5365static int __init ext4_init_fs(void)
5366{
5367 int i, err;
5368
5369 ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5370 ext4_li_info = NULL;
5371 mutex_init(&ext4_li_mtx);
5372
5373 /* Build-time check for flags consistency */
5374 ext4_check_flag_values();
5375
5376 for (i = 0; i < EXT4_WQ_HASH_SZ; i++)
5377 init_waitqueue_head(&ext4__ioend_wq[i]);
5378
5379 err = ext4_init_es();
5380 if (err)
5381 return err;
5382
5383 err = ext4_init_pageio();
5384 if (err)
5385 goto out5;
5386
5387 err = ext4_init_system_zone();
5388 if (err)
5389 goto out4;
5390
5391 err = ext4_init_sysfs();
5392 if (err)
5393 goto out3;
5394
5395 err = ext4_init_mballoc();
5396 if (err)
5397 goto out2;
5398 err = init_inodecache();
5399 if (err)
5400 goto out1;
5401 register_as_ext3();
5402 register_as_ext2();
5403 err = register_filesystem(&ext4_fs_type);
5404 if (err)
5405 goto out;
5406
5407 return 0;
5408out:
5409 unregister_as_ext2();
5410 unregister_as_ext3();
5411 destroy_inodecache();
5412out1:
5413 ext4_exit_mballoc();
5414out2:
5415 ext4_exit_sysfs();
5416out3:
5417 ext4_exit_system_zone();
5418out4:
5419 ext4_exit_pageio();
5420out5:
5421 ext4_exit_es();
5422
5423 return err;
5424}
5425
5426static void __exit ext4_exit_fs(void)
5427{
5428 ext4_exit_crypto();
5429 ext4_destroy_lazyinit_thread();
5430 unregister_as_ext2();
5431 unregister_as_ext3();
5432 unregister_filesystem(&ext4_fs_type);
5433 destroy_inodecache();
5434 ext4_exit_mballoc();
5435 ext4_exit_sysfs();
5436 ext4_exit_system_zone();
5437 ext4_exit_pageio();
5438 ext4_exit_es();
5439}
5440
5441MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5442MODULE_DESCRIPTION("Fourth Extended Filesystem");
5443MODULE_LICENSE("GPL");
5444module_init(ext4_init_fs)
5445module_exit(ext4_exit_fs)