2 * super.c - NILFS module and super block management.
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * Written by Ryusuke Konishi.
19 * linux/fs/ext2/super.c
21 * Copyright (C) 1992, 1993, 1994, 1995
22 * Remy Card (card@masi.ibp.fr)
23 * Laboratoire MASI - Institut Blaise Pascal
24 * Universite Pierre et Marie Curie (Paris VI)
28 * linux/fs/minix/inode.c
30 * Copyright (C) 1991, 1992 Linus Torvalds
32 * Big-endian to little-endian byte-swapping/bitmaps by
33 * David S. Miller (davem@caip.rutgers.edu), 1995
36 #include <linux/module.h>
37 #include <linux/string.h>
38 #include <linux/slab.h>
39 #include <linux/init.h>
40 #include <linux/blkdev.h>
41 #include <linux/parser.h>
42 #include <linux/crc32.h>
43 #include <linux/vfs.h>
44 #include <linux/writeback.h>
45 #include <linux/seq_file.h>
46 #include <linux/mount.h>
55 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
61 MODULE_AUTHOR("NTT Corp.");
62 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
64 MODULE_LICENSE("GPL");
66 static struct kmem_cache
*nilfs_inode_cachep
;
67 struct kmem_cache
*nilfs_transaction_cachep
;
68 struct kmem_cache
*nilfs_segbuf_cachep
;
69 struct kmem_cache
*nilfs_btree_path_cache
;
71 static int nilfs_setup_super(struct super_block
*sb
, int is_mount
);
72 static int nilfs_remount(struct super_block
*sb
, int *flags
, char *data
);
74 static void nilfs_set_error(struct super_block
*sb
)
76 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
77 struct nilfs_super_block
**sbp
;
79 down_write(&nilfs
->ns_sem
);
80 if (!(nilfs
->ns_mount_state
& NILFS_ERROR_FS
)) {
81 nilfs
->ns_mount_state
|= NILFS_ERROR_FS
;
82 sbp
= nilfs_prepare_super(sb
, 0);
84 sbp
[0]->s_state
|= cpu_to_le16(NILFS_ERROR_FS
);
86 sbp
[1]->s_state
|= cpu_to_le16(NILFS_ERROR_FS
);
87 nilfs_commit_super(sb
, NILFS_SB_COMMIT_ALL
);
90 up_write(&nilfs
->ns_sem
);
94 * nilfs_error() - report failure condition on a filesystem
96 * nilfs_error() sets an ERROR_FS flag on the superblock as well as
97 * reporting an error message. It should be called when NILFS detects
98 * incoherences or defects of meta data on disk. As for sustainable
99 * errors such as a single-shot I/O error, nilfs_warning() or the printk()
100 * function should be used instead.
102 * The segment constructor must not call this function because it can
105 void nilfs_error(struct super_block
*sb
, const char *function
,
106 const char *fmt
, ...)
108 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
109 struct va_format vaf
;
117 printk(KERN_CRIT
"NILFS error (device %s): %s: %pV\n",
118 sb
->s_id
, function
, &vaf
);
122 if (!(sb
->s_flags
& MS_RDONLY
)) {
125 if (nilfs_test_opt(nilfs
, ERRORS_RO
)) {
126 printk(KERN_CRIT
"Remounting filesystem read-only\n");
127 sb
->s_flags
|= MS_RDONLY
;
131 if (nilfs_test_opt(nilfs
, ERRORS_PANIC
))
132 panic("NILFS (device %s): panic forced after error\n",
136 void nilfs_warning(struct super_block
*sb
, const char *function
,
137 const char *fmt
, ...)
139 struct va_format vaf
;
147 printk(KERN_WARNING
"NILFS warning (device %s): %s: %pV\n",
148 sb
->s_id
, function
, &vaf
);
154 struct inode
*nilfs_alloc_inode(struct super_block
*sb
)
156 struct nilfs_inode_info
*ii
;
158 ii
= kmem_cache_alloc(nilfs_inode_cachep
, GFP_NOFS
);
164 ii
->vfs_inode
.i_version
= 1;
165 nilfs_mapping_init(&ii
->i_btnode_cache
, &ii
->vfs_inode
);
166 return &ii
->vfs_inode
;
169 static void nilfs_i_callback(struct rcu_head
*head
)
171 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
173 if (nilfs_is_metadata_file_inode(inode
))
174 nilfs_mdt_destroy(inode
);
176 kmem_cache_free(nilfs_inode_cachep
, NILFS_I(inode
));
179 void nilfs_destroy_inode(struct inode
*inode
)
181 call_rcu(&inode
->i_rcu
, nilfs_i_callback
);
184 static int nilfs_sync_super(struct super_block
*sb
, int flag
)
186 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
190 set_buffer_dirty(nilfs
->ns_sbh
[0]);
191 if (nilfs_test_opt(nilfs
, BARRIER
)) {
192 err
= __sync_dirty_buffer(nilfs
->ns_sbh
[0],
193 WRITE_SYNC
| WRITE_FLUSH_FUA
);
195 err
= sync_dirty_buffer(nilfs
->ns_sbh
[0]);
200 "NILFS: unable to write superblock (err=%d)\n", err
);
201 if (err
== -EIO
&& nilfs
->ns_sbh
[1]) {
203 * sbp[0] points to newer log than sbp[1],
204 * so copy sbp[0] to sbp[1] to take over sbp[0].
206 memcpy(nilfs
->ns_sbp
[1], nilfs
->ns_sbp
[0],
208 nilfs_fall_back_super_block(nilfs
);
212 struct nilfs_super_block
*sbp
= nilfs
->ns_sbp
[0];
214 nilfs
->ns_sbwcount
++;
217 * The latest segment becomes trailable from the position
218 * written in superblock.
220 clear_nilfs_discontinued(nilfs
);
222 /* update GC protection for recent segments */
223 if (nilfs
->ns_sbh
[1]) {
224 if (flag
== NILFS_SB_COMMIT_ALL
) {
225 set_buffer_dirty(nilfs
->ns_sbh
[1]);
226 if (sync_dirty_buffer(nilfs
->ns_sbh
[1]) < 0)
229 if (le64_to_cpu(nilfs
->ns_sbp
[1]->s_last_cno
) <
230 le64_to_cpu(nilfs
->ns_sbp
[0]->s_last_cno
))
231 sbp
= nilfs
->ns_sbp
[1];
234 spin_lock(&nilfs
->ns_last_segment_lock
);
235 nilfs
->ns_prot_seq
= le64_to_cpu(sbp
->s_last_seq
);
236 spin_unlock(&nilfs
->ns_last_segment_lock
);
242 void nilfs_set_log_cursor(struct nilfs_super_block
*sbp
,
243 struct the_nilfs
*nilfs
)
245 sector_t nfreeblocks
;
247 /* nilfs->ns_sem must be locked by the caller. */
248 nilfs_count_free_blocks(nilfs
, &nfreeblocks
);
249 sbp
->s_free_blocks_count
= cpu_to_le64(nfreeblocks
);
251 spin_lock(&nilfs
->ns_last_segment_lock
);
252 sbp
->s_last_seq
= cpu_to_le64(nilfs
->ns_last_seq
);
253 sbp
->s_last_pseg
= cpu_to_le64(nilfs
->ns_last_pseg
);
254 sbp
->s_last_cno
= cpu_to_le64(nilfs
->ns_last_cno
);
255 spin_unlock(&nilfs
->ns_last_segment_lock
);
258 struct nilfs_super_block
**nilfs_prepare_super(struct super_block
*sb
,
261 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
262 struct nilfs_super_block
**sbp
= nilfs
->ns_sbp
;
264 /* nilfs->ns_sem must be locked by the caller. */
265 if (sbp
[0]->s_magic
!= cpu_to_le16(NILFS_SUPER_MAGIC
)) {
267 sbp
[1]->s_magic
== cpu_to_le16(NILFS_SUPER_MAGIC
)) {
268 memcpy(sbp
[0], sbp
[1], nilfs
->ns_sbsize
);
270 printk(KERN_CRIT
"NILFS: superblock broke on dev %s\n",
275 sbp
[1]->s_magic
!= cpu_to_le16(NILFS_SUPER_MAGIC
)) {
276 memcpy(sbp
[1], sbp
[0], nilfs
->ns_sbsize
);
280 nilfs_swap_super_block(nilfs
);
285 int nilfs_commit_super(struct super_block
*sb
, int flag
)
287 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
288 struct nilfs_super_block
**sbp
= nilfs
->ns_sbp
;
291 /* nilfs->ns_sem must be locked by the caller. */
293 nilfs
->ns_sbwtime
= t
;
294 sbp
[0]->s_wtime
= cpu_to_le64(t
);
296 sbp
[0]->s_sum
= cpu_to_le32(crc32_le(nilfs
->ns_crc_seed
,
297 (unsigned char *)sbp
[0],
299 if (flag
== NILFS_SB_COMMIT_ALL
&& sbp
[1]) {
300 sbp
[1]->s_wtime
= sbp
[0]->s_wtime
;
302 sbp
[1]->s_sum
= cpu_to_le32(crc32_le(nilfs
->ns_crc_seed
,
303 (unsigned char *)sbp
[1],
306 clear_nilfs_sb_dirty(nilfs
);
307 nilfs
->ns_flushed_device
= 1;
308 /* make sure store to ns_flushed_device cannot be reordered */
310 return nilfs_sync_super(sb
, flag
);
314 * nilfs_cleanup_super() - write filesystem state for cleanup
315 * @sb: super block instance to be unmounted or degraded to read-only
317 * This function restores state flags in the on-disk super block.
318 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
319 * filesystem was not clean previously.
321 int nilfs_cleanup_super(struct super_block
*sb
)
323 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
324 struct nilfs_super_block
**sbp
;
325 int flag
= NILFS_SB_COMMIT
;
328 sbp
= nilfs_prepare_super(sb
, 0);
330 sbp
[0]->s_state
= cpu_to_le16(nilfs
->ns_mount_state
);
331 nilfs_set_log_cursor(sbp
[0], nilfs
);
332 if (sbp
[1] && sbp
[0]->s_last_cno
== sbp
[1]->s_last_cno
) {
334 * make the "clean" flag also to the opposite
335 * super block if both super blocks point to
336 * the same checkpoint.
338 sbp
[1]->s_state
= sbp
[0]->s_state
;
339 flag
= NILFS_SB_COMMIT_ALL
;
341 ret
= nilfs_commit_super(sb
, flag
);
347 * nilfs_move_2nd_super - relocate secondary super block
348 * @sb: super block instance
349 * @sb2off: new offset of the secondary super block (in bytes)
351 static int nilfs_move_2nd_super(struct super_block
*sb
, loff_t sb2off
)
353 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
354 struct buffer_head
*nsbh
;
355 struct nilfs_super_block
*nsbp
;
356 sector_t blocknr
, newblocknr
;
357 unsigned long offset
;
358 int sb2i
; /* array index of the secondary superblock */
361 /* nilfs->ns_sem must be locked by the caller. */
362 if (nilfs
->ns_sbh
[1] &&
363 nilfs
->ns_sbh
[1]->b_blocknr
> nilfs
->ns_first_data_block
) {
365 blocknr
= nilfs
->ns_sbh
[1]->b_blocknr
;
366 } else if (nilfs
->ns_sbh
[0]->b_blocknr
> nilfs
->ns_first_data_block
) {
368 blocknr
= nilfs
->ns_sbh
[0]->b_blocknr
;
373 if (sb2i
>= 0 && (u64
)blocknr
<< nilfs
->ns_blocksize_bits
== sb2off
)
374 goto out
; /* super block location is unchanged */
376 /* Get new super block buffer */
377 newblocknr
= sb2off
>> nilfs
->ns_blocksize_bits
;
378 offset
= sb2off
& (nilfs
->ns_blocksize
- 1);
379 nsbh
= sb_getblk(sb
, newblocknr
);
382 "NILFS warning: unable to move secondary superblock "
383 "to block %llu\n", (unsigned long long)newblocknr
);
387 nsbp
= (void *)nsbh
->b_data
+ offset
;
388 memset(nsbp
, 0, nilfs
->ns_blocksize
);
391 memcpy(nsbp
, nilfs
->ns_sbp
[sb2i
], nilfs
->ns_sbsize
);
392 brelse(nilfs
->ns_sbh
[sb2i
]);
393 nilfs
->ns_sbh
[sb2i
] = nsbh
;
394 nilfs
->ns_sbp
[sb2i
] = nsbp
;
395 } else if (nilfs
->ns_sbh
[0]->b_blocknr
< nilfs
->ns_first_data_block
) {
396 /* secondary super block will be restored to index 1 */
397 nilfs
->ns_sbh
[1] = nsbh
;
398 nilfs
->ns_sbp
[1] = nsbp
;
407 * nilfs_resize_fs - resize the filesystem
408 * @sb: super block instance
409 * @newsize: new size of the filesystem (in bytes)
411 int nilfs_resize_fs(struct super_block
*sb
, __u64 newsize
)
413 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
414 struct nilfs_super_block
**sbp
;
415 __u64 devsize
, newnsegs
;
420 devsize
= i_size_read(sb
->s_bdev
->bd_inode
);
421 if (newsize
> devsize
)
425 * Write lock is required to protect some functions depending
426 * on the number of segments, the number of reserved segments,
429 down_write(&nilfs
->ns_segctor_sem
);
431 sb2off
= NILFS_SB2_OFFSET_BYTES(newsize
);
432 newnsegs
= sb2off
>> nilfs
->ns_blocksize_bits
;
433 do_div(newnsegs
, nilfs
->ns_blocks_per_segment
);
435 ret
= nilfs_sufile_resize(nilfs
->ns_sufile
, newnsegs
);
436 up_write(&nilfs
->ns_segctor_sem
);
440 ret
= nilfs_construct_segment(sb
);
444 down_write(&nilfs
->ns_sem
);
445 nilfs_move_2nd_super(sb
, sb2off
);
447 sbp
= nilfs_prepare_super(sb
, 0);
449 nilfs_set_log_cursor(sbp
[0], nilfs
);
451 * Drop NILFS_RESIZE_FS flag for compatibility with
452 * mount-time resize which may be implemented in a
455 sbp
[0]->s_state
= cpu_to_le16(le16_to_cpu(sbp
[0]->s_state
) &
457 sbp
[0]->s_dev_size
= cpu_to_le64(newsize
);
458 sbp
[0]->s_nsegments
= cpu_to_le64(nilfs
->ns_nsegments
);
460 memcpy(sbp
[1], sbp
[0], nilfs
->ns_sbsize
);
461 ret
= nilfs_commit_super(sb
, NILFS_SB_COMMIT_ALL
);
463 up_write(&nilfs
->ns_sem
);
466 * Reset the range of allocatable segments last. This order
467 * is important in the case of expansion because the secondary
468 * superblock must be protected from log write until migration
472 nilfs_sufile_set_alloc_range(nilfs
->ns_sufile
, 0, newnsegs
- 1);
477 static void nilfs_put_super(struct super_block
*sb
)
479 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
481 nilfs_detach_log_writer(sb
);
483 if (!(sb
->s_flags
& MS_RDONLY
)) {
484 down_write(&nilfs
->ns_sem
);
485 nilfs_cleanup_super(sb
);
486 up_write(&nilfs
->ns_sem
);
489 iput(nilfs
->ns_sufile
);
490 iput(nilfs
->ns_cpfile
);
493 destroy_nilfs(nilfs
);
494 sb
->s_fs_info
= NULL
;
497 static int nilfs_sync_fs(struct super_block
*sb
, int wait
)
499 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
500 struct nilfs_super_block
**sbp
;
503 /* This function is called when super block should be written back */
505 err
= nilfs_construct_segment(sb
);
507 down_write(&nilfs
->ns_sem
);
508 if (nilfs_sb_dirty(nilfs
)) {
509 sbp
= nilfs_prepare_super(sb
, nilfs_sb_will_flip(nilfs
));
511 nilfs_set_log_cursor(sbp
[0], nilfs
);
512 nilfs_commit_super(sb
, NILFS_SB_COMMIT
);
515 up_write(&nilfs
->ns_sem
);
518 err
= nilfs_flush_device(nilfs
);
523 int nilfs_attach_checkpoint(struct super_block
*sb
, __u64 cno
, int curr_mnt
,
524 struct nilfs_root
**rootp
)
526 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
527 struct nilfs_root
*root
;
528 struct nilfs_checkpoint
*raw_cp
;
529 struct buffer_head
*bh_cp
;
532 root
= nilfs_find_or_create_root(
533 nilfs
, curr_mnt
? NILFS_CPTREE_CURRENT_CNO
: cno
);
538 goto reuse
; /* already attached checkpoint */
540 down_read(&nilfs
->ns_segctor_sem
);
541 err
= nilfs_cpfile_get_checkpoint(nilfs
->ns_cpfile
, cno
, 0, &raw_cp
,
543 up_read(&nilfs
->ns_segctor_sem
);
545 if (err
== -ENOENT
|| err
== -EINVAL
) {
547 "NILFS: Invalid checkpoint "
548 "(checkpoint number=%llu)\n",
549 (unsigned long long)cno
);
555 err
= nilfs_ifile_read(sb
, root
, nilfs
->ns_inode_size
,
556 &raw_cp
->cp_ifile_inode
, &root
->ifile
);
560 atomic64_set(&root
->inodes_count
,
561 le64_to_cpu(raw_cp
->cp_inodes_count
));
562 atomic64_set(&root
->blocks_count
,
563 le64_to_cpu(raw_cp
->cp_blocks_count
));
565 nilfs_cpfile_put_checkpoint(nilfs
->ns_cpfile
, cno
, bh_cp
);
572 nilfs_cpfile_put_checkpoint(nilfs
->ns_cpfile
, cno
, bh_cp
);
574 nilfs_put_root(root
);
579 static int nilfs_freeze(struct super_block
*sb
)
581 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
584 if (sb
->s_flags
& MS_RDONLY
)
587 /* Mark super block clean */
588 down_write(&nilfs
->ns_sem
);
589 err
= nilfs_cleanup_super(sb
);
590 up_write(&nilfs
->ns_sem
);
594 static int nilfs_unfreeze(struct super_block
*sb
)
596 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
598 if (sb
->s_flags
& MS_RDONLY
)
601 down_write(&nilfs
->ns_sem
);
602 nilfs_setup_super(sb
, false);
603 up_write(&nilfs
->ns_sem
);
607 static int nilfs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
609 struct super_block
*sb
= dentry
->d_sb
;
610 struct nilfs_root
*root
= NILFS_I(d_inode(dentry
))->i_root
;
611 struct the_nilfs
*nilfs
= root
->nilfs
;
612 u64 id
= huge_encode_dev(sb
->s_bdev
->bd_dev
);
613 unsigned long long blocks
;
614 unsigned long overhead
;
615 unsigned long nrsvblocks
;
616 sector_t nfreeblocks
;
617 u64 nmaxinodes
, nfreeinodes
;
621 * Compute all of the segment blocks
623 * The blocks before first segment and after last segment
626 blocks
= nilfs
->ns_blocks_per_segment
* nilfs
->ns_nsegments
627 - nilfs
->ns_first_data_block
;
628 nrsvblocks
= nilfs
->ns_nrsvsegs
* nilfs
->ns_blocks_per_segment
;
631 * Compute the overhead
633 * When distributing meta data blocks outside segment structure,
634 * We must count them as the overhead.
638 err
= nilfs_count_free_blocks(nilfs
, &nfreeblocks
);
642 err
= nilfs_ifile_count_free_inodes(root
->ifile
,
643 &nmaxinodes
, &nfreeinodes
);
646 "NILFS warning: fail to count free inodes: err %d.\n",
648 if (err
== -ERANGE
) {
650 * If nilfs_palloc_count_max_entries() returns
651 * -ERANGE error code then we simply treat
652 * curent inodes count as maximum possible and
653 * zero as free inodes value.
655 nmaxinodes
= atomic64_read(&root
->inodes_count
);
662 buf
->f_type
= NILFS_SUPER_MAGIC
;
663 buf
->f_bsize
= sb
->s_blocksize
;
664 buf
->f_blocks
= blocks
- overhead
;
665 buf
->f_bfree
= nfreeblocks
;
666 buf
->f_bavail
= (buf
->f_bfree
>= nrsvblocks
) ?
667 (buf
->f_bfree
- nrsvblocks
) : 0;
668 buf
->f_files
= nmaxinodes
;
669 buf
->f_ffree
= nfreeinodes
;
670 buf
->f_namelen
= NILFS_NAME_LEN
;
671 buf
->f_fsid
.val
[0] = (u32
)id
;
672 buf
->f_fsid
.val
[1] = (u32
)(id
>> 32);
677 static int nilfs_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
679 struct super_block
*sb
= dentry
->d_sb
;
680 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
681 struct nilfs_root
*root
= NILFS_I(d_inode(dentry
))->i_root
;
683 if (!nilfs_test_opt(nilfs
, BARRIER
))
684 seq_puts(seq
, ",nobarrier");
685 if (root
->cno
!= NILFS_CPTREE_CURRENT_CNO
)
686 seq_printf(seq
, ",cp=%llu", (unsigned long long)root
->cno
);
687 if (nilfs_test_opt(nilfs
, ERRORS_PANIC
))
688 seq_puts(seq
, ",errors=panic");
689 if (nilfs_test_opt(nilfs
, ERRORS_CONT
))
690 seq_puts(seq
, ",errors=continue");
691 if (nilfs_test_opt(nilfs
, STRICT_ORDER
))
692 seq_puts(seq
, ",order=strict");
693 if (nilfs_test_opt(nilfs
, NORECOVERY
))
694 seq_puts(seq
, ",norecovery");
695 if (nilfs_test_opt(nilfs
, DISCARD
))
696 seq_puts(seq
, ",discard");
701 static const struct super_operations nilfs_sops
= {
702 .alloc_inode
= nilfs_alloc_inode
,
703 .destroy_inode
= nilfs_destroy_inode
,
704 .dirty_inode
= nilfs_dirty_inode
,
705 .evict_inode
= nilfs_evict_inode
,
706 .put_super
= nilfs_put_super
,
707 .sync_fs
= nilfs_sync_fs
,
708 .freeze_fs
= nilfs_freeze
,
709 .unfreeze_fs
= nilfs_unfreeze
,
710 .statfs
= nilfs_statfs
,
711 .remount_fs
= nilfs_remount
,
712 .show_options
= nilfs_show_options
716 Opt_err_cont
, Opt_err_panic
, Opt_err_ro
,
717 Opt_barrier
, Opt_nobarrier
, Opt_snapshot
, Opt_order
, Opt_norecovery
,
718 Opt_discard
, Opt_nodiscard
, Opt_err
,
721 static match_table_t tokens
= {
722 {Opt_err_cont
, "errors=continue"},
723 {Opt_err_panic
, "errors=panic"},
724 {Opt_err_ro
, "errors=remount-ro"},
725 {Opt_barrier
, "barrier"},
726 {Opt_nobarrier
, "nobarrier"},
727 {Opt_snapshot
, "cp=%u"},
728 {Opt_order
, "order=%s"},
729 {Opt_norecovery
, "norecovery"},
730 {Opt_discard
, "discard"},
731 {Opt_nodiscard
, "nodiscard"},
735 static int parse_options(char *options
, struct super_block
*sb
, int is_remount
)
737 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
739 substring_t args
[MAX_OPT_ARGS
];
744 while ((p
= strsep(&options
, ",")) != NULL
) {
750 token
= match_token(p
, tokens
, args
);
753 nilfs_set_opt(nilfs
, BARRIER
);
756 nilfs_clear_opt(nilfs
, BARRIER
);
759 if (strcmp(args
[0].from
, "relaxed") == 0)
760 /* Ordered data semantics */
761 nilfs_clear_opt(nilfs
, STRICT_ORDER
);
762 else if (strcmp(args
[0].from
, "strict") == 0)
763 /* Strict in-order semantics */
764 nilfs_set_opt(nilfs
, STRICT_ORDER
);
769 nilfs_write_opt(nilfs
, ERROR_MODE
, ERRORS_PANIC
);
772 nilfs_write_opt(nilfs
, ERROR_MODE
, ERRORS_RO
);
775 nilfs_write_opt(nilfs
, ERROR_MODE
, ERRORS_CONT
);
780 "NILFS: \"%s\" option is invalid "
781 "for remount.\n", p
);
786 nilfs_set_opt(nilfs
, NORECOVERY
);
789 nilfs_set_opt(nilfs
, DISCARD
);
792 nilfs_clear_opt(nilfs
, DISCARD
);
796 "NILFS: Unrecognized mount option \"%s\"\n", p
);
804 nilfs_set_default_options(struct super_block
*sb
,
805 struct nilfs_super_block
*sbp
)
807 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
809 nilfs
->ns_mount_opt
=
810 NILFS_MOUNT_ERRORS_RO
| NILFS_MOUNT_BARRIER
;
813 static int nilfs_setup_super(struct super_block
*sb
, int is_mount
)
815 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
816 struct nilfs_super_block
**sbp
;
820 /* nilfs->ns_sem must be locked by the caller. */
821 sbp
= nilfs_prepare_super(sb
, 0);
826 goto skip_mount_setup
;
828 max_mnt_count
= le16_to_cpu(sbp
[0]->s_max_mnt_count
);
829 mnt_count
= le16_to_cpu(sbp
[0]->s_mnt_count
);
831 if (nilfs
->ns_mount_state
& NILFS_ERROR_FS
) {
833 "NILFS warning: mounting fs with errors\n");
835 } else if (max_mnt_count
>= 0 && mnt_count
>= max_mnt_count
) {
837 "NILFS warning: maximal mount count reached\n");
841 sbp
[0]->s_max_mnt_count
= cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT
);
843 sbp
[0]->s_mnt_count
= cpu_to_le16(mnt_count
+ 1);
844 sbp
[0]->s_mtime
= cpu_to_le64(get_seconds());
848 cpu_to_le16(le16_to_cpu(sbp
[0]->s_state
) & ~NILFS_VALID_FS
);
849 /* synchronize sbp[1] with sbp[0] */
851 memcpy(sbp
[1], sbp
[0], nilfs
->ns_sbsize
);
852 return nilfs_commit_super(sb
, NILFS_SB_COMMIT_ALL
);
855 struct nilfs_super_block
*nilfs_read_super_block(struct super_block
*sb
,
856 u64 pos
, int blocksize
,
857 struct buffer_head
**pbh
)
859 unsigned long long sb_index
= pos
;
860 unsigned long offset
;
862 offset
= do_div(sb_index
, blocksize
);
863 *pbh
= sb_bread(sb
, sb_index
);
866 return (struct nilfs_super_block
*)((char *)(*pbh
)->b_data
+ offset
);
869 int nilfs_store_magic_and_option(struct super_block
*sb
,
870 struct nilfs_super_block
*sbp
,
873 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
875 sb
->s_magic
= le16_to_cpu(sbp
->s_magic
);
877 /* FS independent flags */
878 #ifdef NILFS_ATIME_DISABLE
879 sb
->s_flags
|= MS_NOATIME
;
882 nilfs_set_default_options(sb
, sbp
);
884 nilfs
->ns_resuid
= le16_to_cpu(sbp
->s_def_resuid
);
885 nilfs
->ns_resgid
= le16_to_cpu(sbp
->s_def_resgid
);
886 nilfs
->ns_interval
= le32_to_cpu(sbp
->s_c_interval
);
887 nilfs
->ns_watermark
= le32_to_cpu(sbp
->s_c_block_max
);
889 return !parse_options(data
, sb
, 0) ? -EINVAL
: 0;
892 int nilfs_check_feature_compatibility(struct super_block
*sb
,
893 struct nilfs_super_block
*sbp
)
897 features
= le64_to_cpu(sbp
->s_feature_incompat
) &
898 ~NILFS_FEATURE_INCOMPAT_SUPP
;
900 printk(KERN_ERR
"NILFS: couldn't mount because of unsupported "
901 "optional features (%llx)\n",
902 (unsigned long long)features
);
905 features
= le64_to_cpu(sbp
->s_feature_compat_ro
) &
906 ~NILFS_FEATURE_COMPAT_RO_SUPP
;
907 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
908 printk(KERN_ERR
"NILFS: couldn't mount RDWR because of "
909 "unsupported optional features (%llx)\n",
910 (unsigned long long)features
);
916 static int nilfs_get_root_dentry(struct super_block
*sb
,
917 struct nilfs_root
*root
,
918 struct dentry
**root_dentry
)
921 struct dentry
*dentry
;
924 inode
= nilfs_iget(sb
, root
, NILFS_ROOT_INO
);
926 printk(KERN_ERR
"NILFS: get root inode failed\n");
927 ret
= PTR_ERR(inode
);
930 if (!S_ISDIR(inode
->i_mode
) || !inode
->i_blocks
|| !inode
->i_size
) {
932 printk(KERN_ERR
"NILFS: corrupt root inode.\n");
937 if (root
->cno
== NILFS_CPTREE_CURRENT_CNO
) {
938 dentry
= d_find_alias(inode
);
940 dentry
= d_make_root(inode
);
949 dentry
= d_obtain_root(inode
);
950 if (IS_ERR(dentry
)) {
951 ret
= PTR_ERR(dentry
);
955 *root_dentry
= dentry
;
960 printk(KERN_ERR
"NILFS: get root dentry failed\n");
964 static int nilfs_attach_snapshot(struct super_block
*s
, __u64 cno
,
965 struct dentry
**root_dentry
)
967 struct the_nilfs
*nilfs
= s
->s_fs_info
;
968 struct nilfs_root
*root
;
971 mutex_lock(&nilfs
->ns_snapshot_mount_mutex
);
973 down_read(&nilfs
->ns_segctor_sem
);
974 ret
= nilfs_cpfile_is_snapshot(nilfs
->ns_cpfile
, cno
);
975 up_read(&nilfs
->ns_segctor_sem
);
977 ret
= (ret
== -ENOENT
) ? -EINVAL
: ret
;
980 printk(KERN_ERR
"NILFS: The specified checkpoint is "
981 "not a snapshot (checkpoint number=%llu).\n",
982 (unsigned long long)cno
);
987 ret
= nilfs_attach_checkpoint(s
, cno
, false, &root
);
989 printk(KERN_ERR
"NILFS: error loading snapshot "
990 "(checkpoint number=%llu).\n",
991 (unsigned long long)cno
);
994 ret
= nilfs_get_root_dentry(s
, root
, root_dentry
);
995 nilfs_put_root(root
);
997 mutex_unlock(&nilfs
->ns_snapshot_mount_mutex
);
1002 * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
1003 * @root_dentry: root dentry of the tree to be shrunk
1005 * This function returns true if the tree was in-use.
1007 static bool nilfs_tree_is_busy(struct dentry
*root_dentry
)
1009 shrink_dcache_parent(root_dentry
);
1010 return d_count(root_dentry
) > 1;
1013 int nilfs_checkpoint_is_mounted(struct super_block
*sb
, __u64 cno
)
1015 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
1016 struct nilfs_root
*root
;
1017 struct inode
*inode
;
1018 struct dentry
*dentry
;
1021 if (cno
> nilfs
->ns_cno
)
1024 if (cno
>= nilfs_last_cno(nilfs
))
1025 return true; /* protect recent checkpoints */
1028 root
= nilfs_lookup_root(nilfs
, cno
);
1030 inode
= nilfs_ilookup(sb
, root
, NILFS_ROOT_INO
);
1032 dentry
= d_find_alias(inode
);
1034 ret
= nilfs_tree_is_busy(dentry
);
1039 nilfs_put_root(root
);
1045 * nilfs_fill_super() - initialize a super block instance
1047 * @data: mount options
1048 * @silent: silent mode flag
1050 * This function is called exclusively by nilfs->ns_mount_mutex.
1051 * So, the recovery process is protected from other simultaneous mounts.
1054 nilfs_fill_super(struct super_block
*sb
, void *data
, int silent
)
1056 struct the_nilfs
*nilfs
;
1057 struct nilfs_root
*fsroot
;
1061 nilfs
= alloc_nilfs(sb
->s_bdev
);
1065 sb
->s_fs_info
= nilfs
;
1067 err
= init_nilfs(nilfs
, sb
, (char *)data
);
1071 sb
->s_op
= &nilfs_sops
;
1072 sb
->s_export_op
= &nilfs_export_ops
;
1074 sb
->s_time_gran
= 1;
1075 sb
->s_max_links
= NILFS_LINK_MAX
;
1077 sb
->s_bdi
= &bdev_get_queue(sb
->s_bdev
)->backing_dev_info
;
1079 err
= load_nilfs(nilfs
, sb
);
1083 cno
= nilfs_last_cno(nilfs
);
1084 err
= nilfs_attach_checkpoint(sb
, cno
, true, &fsroot
);
1086 printk(KERN_ERR
"NILFS: error loading last checkpoint "
1087 "(checkpoint number=%llu).\n", (unsigned long long)cno
);
1091 if (!(sb
->s_flags
& MS_RDONLY
)) {
1092 err
= nilfs_attach_log_writer(sb
, fsroot
);
1094 goto failed_checkpoint
;
1097 err
= nilfs_get_root_dentry(sb
, fsroot
, &sb
->s_root
);
1099 goto failed_segctor
;
1101 nilfs_put_root(fsroot
);
1103 if (!(sb
->s_flags
& MS_RDONLY
)) {
1104 down_write(&nilfs
->ns_sem
);
1105 nilfs_setup_super(sb
, true);
1106 up_write(&nilfs
->ns_sem
);
1112 nilfs_detach_log_writer(sb
);
1115 nilfs_put_root(fsroot
);
1118 iput(nilfs
->ns_sufile
);
1119 iput(nilfs
->ns_cpfile
);
1120 iput(nilfs
->ns_dat
);
1123 destroy_nilfs(nilfs
);
1127 static int nilfs_remount(struct super_block
*sb
, int *flags
, char *data
)
1129 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
1130 unsigned long old_sb_flags
;
1131 unsigned long old_mount_opt
;
1134 sync_filesystem(sb
);
1135 old_sb_flags
= sb
->s_flags
;
1136 old_mount_opt
= nilfs
->ns_mount_opt
;
1138 if (!parse_options(data
, sb
, 1)) {
1142 sb
->s_flags
= (sb
->s_flags
& ~MS_POSIXACL
);
1146 if (!nilfs_valid_fs(nilfs
)) {
1147 printk(KERN_WARNING
"NILFS (device %s): couldn't "
1148 "remount because the filesystem is in an "
1149 "incomplete recovery state.\n", sb
->s_id
);
1153 if ((*flags
& MS_RDONLY
) == (sb
->s_flags
& MS_RDONLY
))
1155 if (*flags
& MS_RDONLY
) {
1156 /* Shutting down log writer */
1157 nilfs_detach_log_writer(sb
);
1158 sb
->s_flags
|= MS_RDONLY
;
1161 * Remounting a valid RW partition RDONLY, so set
1162 * the RDONLY flag and then mark the partition as valid again.
1164 down_write(&nilfs
->ns_sem
);
1165 nilfs_cleanup_super(sb
);
1166 up_write(&nilfs
->ns_sem
);
1169 struct nilfs_root
*root
;
1172 * Mounting a RDONLY partition read-write, so reread and
1173 * store the current valid flag. (It may have been changed
1174 * by fsck since we originally mounted the partition.)
1176 down_read(&nilfs
->ns_sem
);
1177 features
= le64_to_cpu(nilfs
->ns_sbp
[0]->s_feature_compat_ro
) &
1178 ~NILFS_FEATURE_COMPAT_RO_SUPP
;
1179 up_read(&nilfs
->ns_sem
);
1181 printk(KERN_WARNING
"NILFS (device %s): couldn't "
1182 "remount RDWR because of unsupported optional "
1183 "features (%llx)\n",
1184 sb
->s_id
, (unsigned long long)features
);
1189 sb
->s_flags
&= ~MS_RDONLY
;
1191 root
= NILFS_I(d_inode(sb
->s_root
))->i_root
;
1192 err
= nilfs_attach_log_writer(sb
, root
);
1196 down_write(&nilfs
->ns_sem
);
1197 nilfs_setup_super(sb
, true);
1198 up_write(&nilfs
->ns_sem
);
1204 sb
->s_flags
= old_sb_flags
;
1205 nilfs
->ns_mount_opt
= old_mount_opt
;
1209 struct nilfs_super_data
{
1210 struct block_device
*bdev
;
1216 * nilfs_identify - pre-read mount options needed to identify mount instance
1217 * @data: mount options
1218 * @sd: nilfs_super_data
1220 static int nilfs_identify(char *data
, struct nilfs_super_data
*sd
)
1222 char *p
, *options
= data
;
1223 substring_t args
[MAX_OPT_ARGS
];
1228 p
= strsep(&options
, ",");
1229 if (p
!= NULL
&& *p
) {
1230 token
= match_token(p
, tokens
, args
);
1231 if (token
== Opt_snapshot
) {
1232 if (!(sd
->flags
& MS_RDONLY
)) {
1235 sd
->cno
= simple_strtoull(args
[0].from
,
1238 * No need to see the end pointer;
1239 * match_token() has done syntax
1248 "NILFS: invalid mount option: %s\n", p
);
1252 BUG_ON(options
== data
);
1253 *(options
- 1) = ',';
1258 static int nilfs_set_bdev_super(struct super_block
*s
, void *data
)
1261 s
->s_dev
= s
->s_bdev
->bd_dev
;
1265 static int nilfs_test_bdev_super(struct super_block
*s
, void *data
)
1267 return (void *)s
->s_bdev
== data
;
1270 static struct dentry
*
1271 nilfs_mount(struct file_system_type
*fs_type
, int flags
,
1272 const char *dev_name
, void *data
)
1274 struct nilfs_super_data sd
;
1275 struct super_block
*s
;
1276 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1277 struct dentry
*root_dentry
;
1278 int err
, s_new
= false;
1280 if (!(flags
& MS_RDONLY
))
1281 mode
|= FMODE_WRITE
;
1283 sd
.bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1284 if (IS_ERR(sd
.bdev
))
1285 return ERR_CAST(sd
.bdev
);
1289 if (nilfs_identify((char *)data
, &sd
)) {
1295 * once the super is inserted into the list by sget, s_umount
1296 * will protect the lockfs code from trying to start a snapshot
1297 * while we are mounting
1299 mutex_lock(&sd
.bdev
->bd_fsfreeze_mutex
);
1300 if (sd
.bdev
->bd_fsfreeze_count
> 0) {
1301 mutex_unlock(&sd
.bdev
->bd_fsfreeze_mutex
);
1305 s
= sget(fs_type
, nilfs_test_bdev_super
, nilfs_set_bdev_super
, flags
,
1307 mutex_unlock(&sd
.bdev
->bd_fsfreeze_mutex
);
1316 /* New superblock instance created */
1318 snprintf(s
->s_id
, sizeof(s
->s_id
), "%pg", sd
.bdev
);
1319 sb_set_blocksize(s
, block_size(sd
.bdev
));
1321 err
= nilfs_fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1325 s
->s_flags
|= MS_ACTIVE
;
1326 } else if (!sd
.cno
) {
1327 if (nilfs_tree_is_busy(s
->s_root
)) {
1328 if ((flags
^ s
->s_flags
) & MS_RDONLY
) {
1329 printk(KERN_ERR
"NILFS: the device already "
1330 "has a %s mount.\n",
1331 (s
->s_flags
& MS_RDONLY
) ?
1332 "read-only" : "read/write");
1338 * Try remount to setup mount states if the current
1339 * tree is not mounted and only snapshots use this sb.
1341 err
= nilfs_remount(s
, &flags
, data
);
1348 err
= nilfs_attach_snapshot(s
, sd
.cno
, &root_dentry
);
1352 root_dentry
= dget(s
->s_root
);
1356 blkdev_put(sd
.bdev
, mode
);
1361 deactivate_locked_super(s
);
1365 blkdev_put(sd
.bdev
, mode
);
1366 return ERR_PTR(err
);
1369 struct file_system_type nilfs_fs_type
= {
1370 .owner
= THIS_MODULE
,
1372 .mount
= nilfs_mount
,
1373 .kill_sb
= kill_block_super
,
1374 .fs_flags
= FS_REQUIRES_DEV
,
1376 MODULE_ALIAS_FS("nilfs2");
1378 static void nilfs_inode_init_once(void *obj
)
1380 struct nilfs_inode_info
*ii
= obj
;
1382 INIT_LIST_HEAD(&ii
->i_dirty
);
1383 #ifdef CONFIG_NILFS_XATTR
1384 init_rwsem(&ii
->xattr_sem
);
1386 address_space_init_once(&ii
->i_btnode_cache
);
1387 ii
->i_bmap
= &ii
->i_bmap_data
;
1388 inode_init_once(&ii
->vfs_inode
);
1391 static void nilfs_segbuf_init_once(void *obj
)
1393 memset(obj
, 0, sizeof(struct nilfs_segment_buffer
));
1396 static void nilfs_destroy_cachep(void)
1399 * Make sure all delayed rcu free inodes are flushed before we
1404 kmem_cache_destroy(nilfs_inode_cachep
);
1405 kmem_cache_destroy(nilfs_transaction_cachep
);
1406 kmem_cache_destroy(nilfs_segbuf_cachep
);
1407 kmem_cache_destroy(nilfs_btree_path_cache
);
1410 static int __init
nilfs_init_cachep(void)
1412 nilfs_inode_cachep
= kmem_cache_create("nilfs2_inode_cache",
1413 sizeof(struct nilfs_inode_info
), 0,
1414 SLAB_RECLAIM_ACCOUNT
|SLAB_ACCOUNT
,
1415 nilfs_inode_init_once
);
1416 if (!nilfs_inode_cachep
)
1419 nilfs_transaction_cachep
= kmem_cache_create("nilfs2_transaction_cache",
1420 sizeof(struct nilfs_transaction_info
), 0,
1421 SLAB_RECLAIM_ACCOUNT
, NULL
);
1422 if (!nilfs_transaction_cachep
)
1425 nilfs_segbuf_cachep
= kmem_cache_create("nilfs2_segbuf_cache",
1426 sizeof(struct nilfs_segment_buffer
), 0,
1427 SLAB_RECLAIM_ACCOUNT
, nilfs_segbuf_init_once
);
1428 if (!nilfs_segbuf_cachep
)
1431 nilfs_btree_path_cache
= kmem_cache_create("nilfs2_btree_path_cache",
1432 sizeof(struct nilfs_btree_path
) * NILFS_BTREE_LEVEL_MAX
,
1434 if (!nilfs_btree_path_cache
)
1440 nilfs_destroy_cachep();
1444 static int __init
init_nilfs_fs(void)
1448 err
= nilfs_init_cachep();
1452 err
= nilfs_sysfs_init();
1456 err
= register_filesystem(&nilfs_fs_type
);
1458 goto deinit_sysfs_entry
;
1460 printk(KERN_INFO
"NILFS version 2 loaded\n");
1466 nilfs_destroy_cachep();
1471 static void __exit
exit_nilfs_fs(void)
1473 nilfs_destroy_cachep();
1475 unregister_filesystem(&nilfs_fs_type
);
1478 module_init(init_nilfs_fs
)
1479 module_exit(exit_nilfs_fs
)