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 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 * Written by Ryusuke Konishi <ryusuke@osrg.net>
23 * linux/fs/ext2/super.c
25 * Copyright (C) 1992, 1993, 1994, 1995
26 * Remy Card (card@masi.ibp.fr)
27 * Laboratoire MASI - Institut Blaise Pascal
28 * Universite Pierre et Marie Curie (Paris VI)
32 * linux/fs/minix/inode.c
34 * Copyright (C) 1991, 1992 Linus Torvalds
36 * Big-endian to little-endian byte-swapping/bitmaps by
37 * David S. Miller (davem@caip.rutgers.edu), 1995
40 #include <linux/module.h>
41 #include <linux/string.h>
42 #include <linux/slab.h>
43 #include <linux/init.h>
44 #include <linux/blkdev.h>
45 #include <linux/parser.h>
46 #include <linux/crc32.h>
47 #include <linux/vfs.h>
48 #include <linux/writeback.h>
49 #include <linux/seq_file.h>
50 #include <linux/mount.h>
59 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
65 MODULE_AUTHOR("NTT Corp.");
66 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
68 MODULE_LICENSE("GPL");
70 static struct kmem_cache
*nilfs_inode_cachep
;
71 struct kmem_cache
*nilfs_transaction_cachep
;
72 struct kmem_cache
*nilfs_segbuf_cachep
;
73 struct kmem_cache
*nilfs_btree_path_cache
;
75 static int nilfs_setup_super(struct super_block
*sb
, int is_mount
);
76 static int nilfs_remount(struct super_block
*sb
, int *flags
, char *data
);
78 static void nilfs_set_error(struct super_block
*sb
)
80 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
81 struct nilfs_super_block
**sbp
;
83 down_write(&nilfs
->ns_sem
);
84 if (!(nilfs
->ns_mount_state
& NILFS_ERROR_FS
)) {
85 nilfs
->ns_mount_state
|= NILFS_ERROR_FS
;
86 sbp
= nilfs_prepare_super(sb
, 0);
88 sbp
[0]->s_state
|= cpu_to_le16(NILFS_ERROR_FS
);
90 sbp
[1]->s_state
|= cpu_to_le16(NILFS_ERROR_FS
);
91 nilfs_commit_super(sb
, NILFS_SB_COMMIT_ALL
);
94 up_write(&nilfs
->ns_sem
);
98 * nilfs_error() - report failure condition on a filesystem
100 * nilfs_error() sets an ERROR_FS flag on the superblock as well as
101 * reporting an error message. It should be called when NILFS detects
102 * incoherences or defects of meta data on disk. As for sustainable
103 * errors such as a single-shot I/O error, nilfs_warning() or the printk()
104 * function should be used instead.
106 * The segment constructor must not call this function because it can
109 void nilfs_error(struct super_block
*sb
, const char *function
,
110 const char *fmt
, ...)
112 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
113 struct va_format vaf
;
121 printk(KERN_CRIT
"NILFS error (device %s): %s: %pV\n",
122 sb
->s_id
, function
, &vaf
);
126 if (!(sb
->s_flags
& MS_RDONLY
)) {
129 if (nilfs_test_opt(nilfs
, ERRORS_RO
)) {
130 printk(KERN_CRIT
"Remounting filesystem read-only\n");
131 sb
->s_flags
|= MS_RDONLY
;
135 if (nilfs_test_opt(nilfs
, ERRORS_PANIC
))
136 panic("NILFS (device %s): panic forced after error\n",
140 void nilfs_warning(struct super_block
*sb
, const char *function
,
141 const char *fmt
, ...)
143 struct va_format vaf
;
151 printk(KERN_WARNING
"NILFS warning (device %s): %s: %pV\n",
152 sb
->s_id
, function
, &vaf
);
158 struct inode
*nilfs_alloc_inode(struct super_block
*sb
)
160 struct nilfs_inode_info
*ii
;
162 ii
= kmem_cache_alloc(nilfs_inode_cachep
, GFP_NOFS
);
168 ii
->vfs_inode
.i_version
= 1;
169 nilfs_mapping_init(&ii
->i_btnode_cache
, &ii
->vfs_inode
);
170 return &ii
->vfs_inode
;
173 static void nilfs_i_callback(struct rcu_head
*head
)
175 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
176 struct nilfs_mdt_info
*mdi
= NILFS_MDT(inode
);
179 kfree(mdi
->mi_bgl
); /* kfree(NULL) is safe */
182 kmem_cache_free(nilfs_inode_cachep
, NILFS_I(inode
));
185 void nilfs_destroy_inode(struct inode
*inode
)
187 call_rcu(&inode
->i_rcu
, nilfs_i_callback
);
190 static int nilfs_sync_super(struct super_block
*sb
, int flag
)
192 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
196 set_buffer_dirty(nilfs
->ns_sbh
[0]);
197 if (nilfs_test_opt(nilfs
, BARRIER
)) {
198 err
= __sync_dirty_buffer(nilfs
->ns_sbh
[0],
199 WRITE_SYNC
| WRITE_FLUSH_FUA
);
201 err
= sync_dirty_buffer(nilfs
->ns_sbh
[0]);
206 "NILFS: unable to write superblock (err=%d)\n", err
);
207 if (err
== -EIO
&& nilfs
->ns_sbh
[1]) {
209 * sbp[0] points to newer log than sbp[1],
210 * so copy sbp[0] to sbp[1] to take over sbp[0].
212 memcpy(nilfs
->ns_sbp
[1], nilfs
->ns_sbp
[0],
214 nilfs_fall_back_super_block(nilfs
);
218 struct nilfs_super_block
*sbp
= nilfs
->ns_sbp
[0];
220 nilfs
->ns_sbwcount
++;
223 * The latest segment becomes trailable from the position
224 * written in superblock.
226 clear_nilfs_discontinued(nilfs
);
228 /* update GC protection for recent segments */
229 if (nilfs
->ns_sbh
[1]) {
230 if (flag
== NILFS_SB_COMMIT_ALL
) {
231 set_buffer_dirty(nilfs
->ns_sbh
[1]);
232 if (sync_dirty_buffer(nilfs
->ns_sbh
[1]) < 0)
235 if (le64_to_cpu(nilfs
->ns_sbp
[1]->s_last_cno
) <
236 le64_to_cpu(nilfs
->ns_sbp
[0]->s_last_cno
))
237 sbp
= nilfs
->ns_sbp
[1];
240 spin_lock(&nilfs
->ns_last_segment_lock
);
241 nilfs
->ns_prot_seq
= le64_to_cpu(sbp
->s_last_seq
);
242 spin_unlock(&nilfs
->ns_last_segment_lock
);
248 void nilfs_set_log_cursor(struct nilfs_super_block
*sbp
,
249 struct the_nilfs
*nilfs
)
251 sector_t nfreeblocks
;
253 /* nilfs->ns_sem must be locked by the caller. */
254 nilfs_count_free_blocks(nilfs
, &nfreeblocks
);
255 sbp
->s_free_blocks_count
= cpu_to_le64(nfreeblocks
);
257 spin_lock(&nilfs
->ns_last_segment_lock
);
258 sbp
->s_last_seq
= cpu_to_le64(nilfs
->ns_last_seq
);
259 sbp
->s_last_pseg
= cpu_to_le64(nilfs
->ns_last_pseg
);
260 sbp
->s_last_cno
= cpu_to_le64(nilfs
->ns_last_cno
);
261 spin_unlock(&nilfs
->ns_last_segment_lock
);
264 struct nilfs_super_block
**nilfs_prepare_super(struct super_block
*sb
,
267 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
268 struct nilfs_super_block
**sbp
= nilfs
->ns_sbp
;
270 /* nilfs->ns_sem must be locked by the caller. */
271 if (sbp
[0]->s_magic
!= cpu_to_le16(NILFS_SUPER_MAGIC
)) {
273 sbp
[1]->s_magic
== cpu_to_le16(NILFS_SUPER_MAGIC
)) {
274 memcpy(sbp
[0], sbp
[1], nilfs
->ns_sbsize
);
276 printk(KERN_CRIT
"NILFS: superblock broke on dev %s\n",
281 sbp
[1]->s_magic
!= cpu_to_le16(NILFS_SUPER_MAGIC
)) {
282 memcpy(sbp
[1], sbp
[0], nilfs
->ns_sbsize
);
286 nilfs_swap_super_block(nilfs
);
291 int nilfs_commit_super(struct super_block
*sb
, int flag
)
293 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
294 struct nilfs_super_block
**sbp
= nilfs
->ns_sbp
;
297 /* nilfs->ns_sem must be locked by the caller. */
299 nilfs
->ns_sbwtime
= t
;
300 sbp
[0]->s_wtime
= cpu_to_le64(t
);
302 sbp
[0]->s_sum
= cpu_to_le32(crc32_le(nilfs
->ns_crc_seed
,
303 (unsigned char *)sbp
[0],
305 if (flag
== NILFS_SB_COMMIT_ALL
&& sbp
[1]) {
306 sbp
[1]->s_wtime
= sbp
[0]->s_wtime
;
308 sbp
[1]->s_sum
= cpu_to_le32(crc32_le(nilfs
->ns_crc_seed
,
309 (unsigned char *)sbp
[1],
312 clear_nilfs_sb_dirty(nilfs
);
313 nilfs
->ns_flushed_device
= 1;
314 /* make sure store to ns_flushed_device cannot be reordered */
316 return nilfs_sync_super(sb
, flag
);
320 * nilfs_cleanup_super() - write filesystem state for cleanup
321 * @sb: super block instance to be unmounted or degraded to read-only
323 * This function restores state flags in the on-disk super block.
324 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
325 * filesystem was not clean previously.
327 int nilfs_cleanup_super(struct super_block
*sb
)
329 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
330 struct nilfs_super_block
**sbp
;
331 int flag
= NILFS_SB_COMMIT
;
334 sbp
= nilfs_prepare_super(sb
, 0);
336 sbp
[0]->s_state
= cpu_to_le16(nilfs
->ns_mount_state
);
337 nilfs_set_log_cursor(sbp
[0], nilfs
);
338 if (sbp
[1] && sbp
[0]->s_last_cno
== sbp
[1]->s_last_cno
) {
340 * make the "clean" flag also to the opposite
341 * super block if both super blocks point to
342 * the same checkpoint.
344 sbp
[1]->s_state
= sbp
[0]->s_state
;
345 flag
= NILFS_SB_COMMIT_ALL
;
347 ret
= nilfs_commit_super(sb
, flag
);
353 * nilfs_move_2nd_super - relocate secondary super block
354 * @sb: super block instance
355 * @sb2off: new offset of the secondary super block (in bytes)
357 static int nilfs_move_2nd_super(struct super_block
*sb
, loff_t sb2off
)
359 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
360 struct buffer_head
*nsbh
;
361 struct nilfs_super_block
*nsbp
;
362 sector_t blocknr
, newblocknr
;
363 unsigned long offset
;
364 int sb2i
= -1; /* array index of the secondary superblock */
367 /* nilfs->ns_sem must be locked by the caller. */
368 if (nilfs
->ns_sbh
[1] &&
369 nilfs
->ns_sbh
[1]->b_blocknr
> nilfs
->ns_first_data_block
) {
371 blocknr
= nilfs
->ns_sbh
[1]->b_blocknr
;
372 } else if (nilfs
->ns_sbh
[0]->b_blocknr
> nilfs
->ns_first_data_block
) {
374 blocknr
= nilfs
->ns_sbh
[0]->b_blocknr
;
376 if (sb2i
>= 0 && (u64
)blocknr
<< nilfs
->ns_blocksize_bits
== sb2off
)
377 goto out
; /* super block location is unchanged */
379 /* Get new super block buffer */
380 newblocknr
= sb2off
>> nilfs
->ns_blocksize_bits
;
381 offset
= sb2off
& (nilfs
->ns_blocksize
- 1);
382 nsbh
= sb_getblk(sb
, newblocknr
);
385 "NILFS warning: unable to move secondary superblock "
386 "to block %llu\n", (unsigned long long)newblocknr
);
390 nsbp
= (void *)nsbh
->b_data
+ offset
;
391 memset(nsbp
, 0, nilfs
->ns_blocksize
);
394 memcpy(nsbp
, nilfs
->ns_sbp
[sb2i
], nilfs
->ns_sbsize
);
395 brelse(nilfs
->ns_sbh
[sb2i
]);
396 nilfs
->ns_sbh
[sb2i
] = nsbh
;
397 nilfs
->ns_sbp
[sb2i
] = nsbp
;
398 } else if (nilfs
->ns_sbh
[0]->b_blocknr
< nilfs
->ns_first_data_block
) {
399 /* secondary super block will be restored to index 1 */
400 nilfs
->ns_sbh
[1] = nsbh
;
401 nilfs
->ns_sbp
[1] = nsbp
;
410 * nilfs_resize_fs - resize the filesystem
411 * @sb: super block instance
412 * @newsize: new size of the filesystem (in bytes)
414 int nilfs_resize_fs(struct super_block
*sb
, __u64 newsize
)
416 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
417 struct nilfs_super_block
**sbp
;
418 __u64 devsize
, newnsegs
;
423 devsize
= i_size_read(sb
->s_bdev
->bd_inode
);
424 if (newsize
> devsize
)
428 * Write lock is required to protect some functions depending
429 * on the number of segments, the number of reserved segments,
432 down_write(&nilfs
->ns_segctor_sem
);
434 sb2off
= NILFS_SB2_OFFSET_BYTES(newsize
);
435 newnsegs
= sb2off
>> nilfs
->ns_blocksize_bits
;
436 do_div(newnsegs
, nilfs
->ns_blocks_per_segment
);
438 ret
= nilfs_sufile_resize(nilfs
->ns_sufile
, newnsegs
);
439 up_write(&nilfs
->ns_segctor_sem
);
443 ret
= nilfs_construct_segment(sb
);
447 down_write(&nilfs
->ns_sem
);
448 nilfs_move_2nd_super(sb
, sb2off
);
450 sbp
= nilfs_prepare_super(sb
, 0);
452 nilfs_set_log_cursor(sbp
[0], nilfs
);
454 * Drop NILFS_RESIZE_FS flag for compatibility with
455 * mount-time resize which may be implemented in a
458 sbp
[0]->s_state
= cpu_to_le16(le16_to_cpu(sbp
[0]->s_state
) &
460 sbp
[0]->s_dev_size
= cpu_to_le64(newsize
);
461 sbp
[0]->s_nsegments
= cpu_to_le64(nilfs
->ns_nsegments
);
463 memcpy(sbp
[1], sbp
[0], nilfs
->ns_sbsize
);
464 ret
= nilfs_commit_super(sb
, NILFS_SB_COMMIT_ALL
);
466 up_write(&nilfs
->ns_sem
);
469 * Reset the range of allocatable segments last. This order
470 * is important in the case of expansion because the secondary
471 * superblock must be protected from log write until migration
475 nilfs_sufile_set_alloc_range(nilfs
->ns_sufile
, 0, newnsegs
- 1);
480 static void nilfs_put_super(struct super_block
*sb
)
482 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
484 nilfs_detach_log_writer(sb
);
486 if (!(sb
->s_flags
& MS_RDONLY
)) {
487 down_write(&nilfs
->ns_sem
);
488 nilfs_cleanup_super(sb
);
489 up_write(&nilfs
->ns_sem
);
492 iput(nilfs
->ns_sufile
);
493 iput(nilfs
->ns_cpfile
);
496 destroy_nilfs(nilfs
);
497 sb
->s_fs_info
= NULL
;
500 static int nilfs_sync_fs(struct super_block
*sb
, int wait
)
502 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
503 struct nilfs_super_block
**sbp
;
506 /* This function is called when super block should be written back */
508 err
= nilfs_construct_segment(sb
);
510 down_write(&nilfs
->ns_sem
);
511 if (nilfs_sb_dirty(nilfs
)) {
512 sbp
= nilfs_prepare_super(sb
, nilfs_sb_will_flip(nilfs
));
514 nilfs_set_log_cursor(sbp
[0], nilfs
);
515 nilfs_commit_super(sb
, NILFS_SB_COMMIT
);
518 up_write(&nilfs
->ns_sem
);
521 err
= nilfs_flush_device(nilfs
);
526 int nilfs_attach_checkpoint(struct super_block
*sb
, __u64 cno
, int curr_mnt
,
527 struct nilfs_root
**rootp
)
529 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
530 struct nilfs_root
*root
;
531 struct nilfs_checkpoint
*raw_cp
;
532 struct buffer_head
*bh_cp
;
535 root
= nilfs_find_or_create_root(
536 nilfs
, curr_mnt
? NILFS_CPTREE_CURRENT_CNO
: cno
);
541 goto reuse
; /* already attached checkpoint */
543 down_read(&nilfs
->ns_segctor_sem
);
544 err
= nilfs_cpfile_get_checkpoint(nilfs
->ns_cpfile
, cno
, 0, &raw_cp
,
546 up_read(&nilfs
->ns_segctor_sem
);
548 if (err
== -ENOENT
|| err
== -EINVAL
) {
550 "NILFS: Invalid checkpoint "
551 "(checkpoint number=%llu)\n",
552 (unsigned long long)cno
);
558 err
= nilfs_ifile_read(sb
, root
, nilfs
->ns_inode_size
,
559 &raw_cp
->cp_ifile_inode
, &root
->ifile
);
563 atomic64_set(&root
->inodes_count
,
564 le64_to_cpu(raw_cp
->cp_inodes_count
));
565 atomic64_set(&root
->blocks_count
,
566 le64_to_cpu(raw_cp
->cp_blocks_count
));
568 nilfs_cpfile_put_checkpoint(nilfs
->ns_cpfile
, cno
, bh_cp
);
575 nilfs_cpfile_put_checkpoint(nilfs
->ns_cpfile
, cno
, bh_cp
);
577 nilfs_put_root(root
);
582 static int nilfs_freeze(struct super_block
*sb
)
584 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
587 if (sb
->s_flags
& MS_RDONLY
)
590 /* Mark super block clean */
591 down_write(&nilfs
->ns_sem
);
592 err
= nilfs_cleanup_super(sb
);
593 up_write(&nilfs
->ns_sem
);
597 static int nilfs_unfreeze(struct super_block
*sb
)
599 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
601 if (sb
->s_flags
& MS_RDONLY
)
604 down_write(&nilfs
->ns_sem
);
605 nilfs_setup_super(sb
, false);
606 up_write(&nilfs
->ns_sem
);
610 static int nilfs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
612 struct super_block
*sb
= dentry
->d_sb
;
613 struct nilfs_root
*root
= NILFS_I(d_inode(dentry
))->i_root
;
614 struct the_nilfs
*nilfs
= root
->nilfs
;
615 u64 id
= huge_encode_dev(sb
->s_bdev
->bd_dev
);
616 unsigned long long blocks
;
617 unsigned long overhead
;
618 unsigned long nrsvblocks
;
619 sector_t nfreeblocks
;
620 u64 nmaxinodes
, nfreeinodes
;
624 * Compute all of the segment blocks
626 * The blocks before first segment and after last segment
629 blocks
= nilfs
->ns_blocks_per_segment
* nilfs
->ns_nsegments
630 - nilfs
->ns_first_data_block
;
631 nrsvblocks
= nilfs
->ns_nrsvsegs
* nilfs
->ns_blocks_per_segment
;
634 * Compute the overhead
636 * When distributing meta data blocks outside segment structure,
637 * We must count them as the overhead.
641 err
= nilfs_count_free_blocks(nilfs
, &nfreeblocks
);
645 err
= nilfs_ifile_count_free_inodes(root
->ifile
,
646 &nmaxinodes
, &nfreeinodes
);
649 "NILFS warning: fail to count free inodes: err %d.\n",
651 if (err
== -ERANGE
) {
653 * If nilfs_palloc_count_max_entries() returns
654 * -ERANGE error code then we simply treat
655 * curent inodes count as maximum possible and
656 * zero as free inodes value.
658 nmaxinodes
= atomic64_read(&root
->inodes_count
);
665 buf
->f_type
= NILFS_SUPER_MAGIC
;
666 buf
->f_bsize
= sb
->s_blocksize
;
667 buf
->f_blocks
= blocks
- overhead
;
668 buf
->f_bfree
= nfreeblocks
;
669 buf
->f_bavail
= (buf
->f_bfree
>= nrsvblocks
) ?
670 (buf
->f_bfree
- nrsvblocks
) : 0;
671 buf
->f_files
= nmaxinodes
;
672 buf
->f_ffree
= nfreeinodes
;
673 buf
->f_namelen
= NILFS_NAME_LEN
;
674 buf
->f_fsid
.val
[0] = (u32
)id
;
675 buf
->f_fsid
.val
[1] = (u32
)(id
>> 32);
680 static int nilfs_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
682 struct super_block
*sb
= dentry
->d_sb
;
683 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
684 struct nilfs_root
*root
= NILFS_I(d_inode(dentry
))->i_root
;
686 if (!nilfs_test_opt(nilfs
, BARRIER
))
687 seq_puts(seq
, ",nobarrier");
688 if (root
->cno
!= NILFS_CPTREE_CURRENT_CNO
)
689 seq_printf(seq
, ",cp=%llu", (unsigned long long)root
->cno
);
690 if (nilfs_test_opt(nilfs
, ERRORS_PANIC
))
691 seq_puts(seq
, ",errors=panic");
692 if (nilfs_test_opt(nilfs
, ERRORS_CONT
))
693 seq_puts(seq
, ",errors=continue");
694 if (nilfs_test_opt(nilfs
, STRICT_ORDER
))
695 seq_puts(seq
, ",order=strict");
696 if (nilfs_test_opt(nilfs
, NORECOVERY
))
697 seq_puts(seq
, ",norecovery");
698 if (nilfs_test_opt(nilfs
, DISCARD
))
699 seq_puts(seq
, ",discard");
704 static const struct super_operations nilfs_sops
= {
705 .alloc_inode
= nilfs_alloc_inode
,
706 .destroy_inode
= nilfs_destroy_inode
,
707 .dirty_inode
= nilfs_dirty_inode
,
708 .evict_inode
= nilfs_evict_inode
,
709 .put_super
= nilfs_put_super
,
710 .sync_fs
= nilfs_sync_fs
,
711 .freeze_fs
= nilfs_freeze
,
712 .unfreeze_fs
= nilfs_unfreeze
,
713 .statfs
= nilfs_statfs
,
714 .remount_fs
= nilfs_remount
,
715 .show_options
= nilfs_show_options
719 Opt_err_cont
, Opt_err_panic
, Opt_err_ro
,
720 Opt_barrier
, Opt_nobarrier
, Opt_snapshot
, Opt_order
, Opt_norecovery
,
721 Opt_discard
, Opt_nodiscard
, Opt_err
,
724 static match_table_t tokens
= {
725 {Opt_err_cont
, "errors=continue"},
726 {Opt_err_panic
, "errors=panic"},
727 {Opt_err_ro
, "errors=remount-ro"},
728 {Opt_barrier
, "barrier"},
729 {Opt_nobarrier
, "nobarrier"},
730 {Opt_snapshot
, "cp=%u"},
731 {Opt_order
, "order=%s"},
732 {Opt_norecovery
, "norecovery"},
733 {Opt_discard
, "discard"},
734 {Opt_nodiscard
, "nodiscard"},
738 static int parse_options(char *options
, struct super_block
*sb
, int is_remount
)
740 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
742 substring_t args
[MAX_OPT_ARGS
];
747 while ((p
= strsep(&options
, ",")) != NULL
) {
752 token
= match_token(p
, tokens
, args
);
755 nilfs_set_opt(nilfs
, BARRIER
);
758 nilfs_clear_opt(nilfs
, BARRIER
);
761 if (strcmp(args
[0].from
, "relaxed") == 0)
762 /* Ordered data semantics */
763 nilfs_clear_opt(nilfs
, STRICT_ORDER
);
764 else if (strcmp(args
[0].from
, "strict") == 0)
765 /* Strict in-order semantics */
766 nilfs_set_opt(nilfs
, STRICT_ORDER
);
771 nilfs_write_opt(nilfs
, ERROR_MODE
, ERRORS_PANIC
);
774 nilfs_write_opt(nilfs
, ERROR_MODE
, ERRORS_RO
);
777 nilfs_write_opt(nilfs
, ERROR_MODE
, ERRORS_CONT
);
782 "NILFS: \"%s\" option is invalid "
783 "for remount.\n", p
);
788 nilfs_set_opt(nilfs
, NORECOVERY
);
791 nilfs_set_opt(nilfs
, DISCARD
);
794 nilfs_clear_opt(nilfs
, DISCARD
);
798 "NILFS: Unrecognized mount option \"%s\"\n", p
);
806 nilfs_set_default_options(struct super_block
*sb
,
807 struct nilfs_super_block
*sbp
)
809 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
811 nilfs
->ns_mount_opt
=
812 NILFS_MOUNT_ERRORS_RO
| NILFS_MOUNT_BARRIER
;
815 static int nilfs_setup_super(struct super_block
*sb
, int is_mount
)
817 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
818 struct nilfs_super_block
**sbp
;
822 /* nilfs->ns_sem must be locked by the caller. */
823 sbp
= nilfs_prepare_super(sb
, 0);
828 goto skip_mount_setup
;
830 max_mnt_count
= le16_to_cpu(sbp
[0]->s_max_mnt_count
);
831 mnt_count
= le16_to_cpu(sbp
[0]->s_mnt_count
);
833 if (nilfs
->ns_mount_state
& NILFS_ERROR_FS
) {
835 "NILFS warning: mounting fs with errors\n");
837 } else if (max_mnt_count
>= 0 && mnt_count
>= max_mnt_count
) {
839 "NILFS warning: maximal mount count reached\n");
843 sbp
[0]->s_max_mnt_count
= cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT
);
845 sbp
[0]->s_mnt_count
= cpu_to_le16(mnt_count
+ 1);
846 sbp
[0]->s_mtime
= cpu_to_le64(get_seconds());
850 cpu_to_le16(le16_to_cpu(sbp
[0]->s_state
) & ~NILFS_VALID_FS
);
851 /* synchronize sbp[1] with sbp[0] */
853 memcpy(sbp
[1], sbp
[0], nilfs
->ns_sbsize
);
854 return nilfs_commit_super(sb
, NILFS_SB_COMMIT_ALL
);
857 struct nilfs_super_block
*nilfs_read_super_block(struct super_block
*sb
,
858 u64 pos
, int blocksize
,
859 struct buffer_head
**pbh
)
861 unsigned long long sb_index
= pos
;
862 unsigned long offset
;
864 offset
= do_div(sb_index
, blocksize
);
865 *pbh
= sb_bread(sb
, sb_index
);
868 return (struct nilfs_super_block
*)((char *)(*pbh
)->b_data
+ offset
);
871 int nilfs_store_magic_and_option(struct super_block
*sb
,
872 struct nilfs_super_block
*sbp
,
875 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
877 sb
->s_magic
= le16_to_cpu(sbp
->s_magic
);
879 /* FS independent flags */
880 #ifdef NILFS_ATIME_DISABLE
881 sb
->s_flags
|= MS_NOATIME
;
884 nilfs_set_default_options(sb
, sbp
);
886 nilfs
->ns_resuid
= le16_to_cpu(sbp
->s_def_resuid
);
887 nilfs
->ns_resgid
= le16_to_cpu(sbp
->s_def_resgid
);
888 nilfs
->ns_interval
= le32_to_cpu(sbp
->s_c_interval
);
889 nilfs
->ns_watermark
= le32_to_cpu(sbp
->s_c_block_max
);
891 return !parse_options(data
, sb
, 0) ? -EINVAL
: 0 ;
894 int nilfs_check_feature_compatibility(struct super_block
*sb
,
895 struct nilfs_super_block
*sbp
)
899 features
= le64_to_cpu(sbp
->s_feature_incompat
) &
900 ~NILFS_FEATURE_INCOMPAT_SUPP
;
902 printk(KERN_ERR
"NILFS: couldn't mount because of unsupported "
903 "optional features (%llx)\n",
904 (unsigned long long)features
);
907 features
= le64_to_cpu(sbp
->s_feature_compat_ro
) &
908 ~NILFS_FEATURE_COMPAT_RO_SUPP
;
909 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
910 printk(KERN_ERR
"NILFS: couldn't mount RDWR because of "
911 "unsupported optional features (%llx)\n",
912 (unsigned long long)features
);
918 static int nilfs_get_root_dentry(struct super_block
*sb
,
919 struct nilfs_root
*root
,
920 struct dentry
**root_dentry
)
923 struct dentry
*dentry
;
926 inode
= nilfs_iget(sb
, root
, NILFS_ROOT_INO
);
928 printk(KERN_ERR
"NILFS: get root inode failed\n");
929 ret
= PTR_ERR(inode
);
932 if (!S_ISDIR(inode
->i_mode
) || !inode
->i_blocks
|| !inode
->i_size
) {
934 printk(KERN_ERR
"NILFS: corrupt root inode.\n");
939 if (root
->cno
== NILFS_CPTREE_CURRENT_CNO
) {
940 dentry
= d_find_alias(inode
);
942 dentry
= d_make_root(inode
);
951 dentry
= d_obtain_root(inode
);
952 if (IS_ERR(dentry
)) {
953 ret
= PTR_ERR(dentry
);
957 *root_dentry
= dentry
;
962 printk(KERN_ERR
"NILFS: get root dentry failed\n");
966 static int nilfs_attach_snapshot(struct super_block
*s
, __u64 cno
,
967 struct dentry
**root_dentry
)
969 struct the_nilfs
*nilfs
= s
->s_fs_info
;
970 struct nilfs_root
*root
;
973 mutex_lock(&nilfs
->ns_snapshot_mount_mutex
);
975 down_read(&nilfs
->ns_segctor_sem
);
976 ret
= nilfs_cpfile_is_snapshot(nilfs
->ns_cpfile
, cno
);
977 up_read(&nilfs
->ns_segctor_sem
);
979 ret
= (ret
== -ENOENT
) ? -EINVAL
: ret
;
982 printk(KERN_ERR
"NILFS: The specified checkpoint is "
983 "not a snapshot (checkpoint number=%llu).\n",
984 (unsigned long long)cno
);
989 ret
= nilfs_attach_checkpoint(s
, cno
, false, &root
);
991 printk(KERN_ERR
"NILFS: error loading snapshot "
992 "(checkpoint number=%llu).\n",
993 (unsigned long long)cno
);
996 ret
= nilfs_get_root_dentry(s
, root
, root_dentry
);
997 nilfs_put_root(root
);
999 mutex_unlock(&nilfs
->ns_snapshot_mount_mutex
);
1004 * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
1005 * @root_dentry: root dentry of the tree to be shrunk
1007 * This function returns true if the tree was in-use.
1009 static bool nilfs_tree_is_busy(struct dentry
*root_dentry
)
1011 shrink_dcache_parent(root_dentry
);
1012 return d_count(root_dentry
) > 1;
1015 int nilfs_checkpoint_is_mounted(struct super_block
*sb
, __u64 cno
)
1017 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
1018 struct nilfs_root
*root
;
1019 struct inode
*inode
;
1020 struct dentry
*dentry
;
1023 if (cno
> nilfs
->ns_cno
)
1026 if (cno
>= nilfs_last_cno(nilfs
))
1027 return true; /* protect recent checkpoints */
1030 root
= nilfs_lookup_root(nilfs
, cno
);
1032 inode
= nilfs_ilookup(sb
, root
, NILFS_ROOT_INO
);
1034 dentry
= d_find_alias(inode
);
1036 ret
= nilfs_tree_is_busy(dentry
);
1041 nilfs_put_root(root
);
1047 * nilfs_fill_super() - initialize a super block instance
1049 * @data: mount options
1050 * @silent: silent mode flag
1052 * This function is called exclusively by nilfs->ns_mount_mutex.
1053 * So, the recovery process is protected from other simultaneous mounts.
1056 nilfs_fill_super(struct super_block
*sb
, void *data
, int silent
)
1058 struct the_nilfs
*nilfs
;
1059 struct nilfs_root
*fsroot
;
1063 nilfs
= alloc_nilfs(sb
->s_bdev
);
1067 sb
->s_fs_info
= nilfs
;
1069 err
= init_nilfs(nilfs
, sb
, (char *)data
);
1073 sb
->s_op
= &nilfs_sops
;
1074 sb
->s_export_op
= &nilfs_export_ops
;
1076 sb
->s_time_gran
= 1;
1077 sb
->s_max_links
= NILFS_LINK_MAX
;
1079 sb
->s_bdi
= &bdev_get_queue(sb
->s_bdev
)->backing_dev_info
;
1081 err
= load_nilfs(nilfs
, sb
);
1085 cno
= nilfs_last_cno(nilfs
);
1086 err
= nilfs_attach_checkpoint(sb
, cno
, true, &fsroot
);
1088 printk(KERN_ERR
"NILFS: error loading last checkpoint "
1089 "(checkpoint number=%llu).\n", (unsigned long long)cno
);
1093 if (!(sb
->s_flags
& MS_RDONLY
)) {
1094 err
= nilfs_attach_log_writer(sb
, fsroot
);
1096 goto failed_checkpoint
;
1099 err
= nilfs_get_root_dentry(sb
, fsroot
, &sb
->s_root
);
1101 goto failed_segctor
;
1103 nilfs_put_root(fsroot
);
1105 if (!(sb
->s_flags
& MS_RDONLY
)) {
1106 down_write(&nilfs
->ns_sem
);
1107 nilfs_setup_super(sb
, true);
1108 up_write(&nilfs
->ns_sem
);
1114 nilfs_detach_log_writer(sb
);
1117 nilfs_put_root(fsroot
);
1120 iput(nilfs
->ns_sufile
);
1121 iput(nilfs
->ns_cpfile
);
1122 iput(nilfs
->ns_dat
);
1125 destroy_nilfs(nilfs
);
1129 static int nilfs_remount(struct super_block
*sb
, int *flags
, char *data
)
1131 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
1132 unsigned long old_sb_flags
;
1133 unsigned long old_mount_opt
;
1136 sync_filesystem(sb
);
1137 old_sb_flags
= sb
->s_flags
;
1138 old_mount_opt
= nilfs
->ns_mount_opt
;
1140 if (!parse_options(data
, sb
, 1)) {
1144 sb
->s_flags
= (sb
->s_flags
& ~MS_POSIXACL
);
1148 if (!nilfs_valid_fs(nilfs
)) {
1149 printk(KERN_WARNING
"NILFS (device %s): couldn't "
1150 "remount because the filesystem is in an "
1151 "incomplete recovery state.\n", sb
->s_id
);
1155 if ((*flags
& MS_RDONLY
) == (sb
->s_flags
& MS_RDONLY
))
1157 if (*flags
& MS_RDONLY
) {
1158 /* Shutting down log writer */
1159 nilfs_detach_log_writer(sb
);
1160 sb
->s_flags
|= MS_RDONLY
;
1163 * Remounting a valid RW partition RDONLY, so set
1164 * the RDONLY flag and then mark the partition as valid again.
1166 down_write(&nilfs
->ns_sem
);
1167 nilfs_cleanup_super(sb
);
1168 up_write(&nilfs
->ns_sem
);
1171 struct nilfs_root
*root
;
1174 * Mounting a RDONLY partition read-write, so reread and
1175 * store the current valid flag. (It may have been changed
1176 * by fsck since we originally mounted the partition.)
1178 down_read(&nilfs
->ns_sem
);
1179 features
= le64_to_cpu(nilfs
->ns_sbp
[0]->s_feature_compat_ro
) &
1180 ~NILFS_FEATURE_COMPAT_RO_SUPP
;
1181 up_read(&nilfs
->ns_sem
);
1183 printk(KERN_WARNING
"NILFS (device %s): couldn't "
1184 "remount RDWR because of unsupported optional "
1185 "features (%llx)\n",
1186 sb
->s_id
, (unsigned long long)features
);
1191 sb
->s_flags
&= ~MS_RDONLY
;
1193 root
= NILFS_I(d_inode(sb
->s_root
))->i_root
;
1194 err
= nilfs_attach_log_writer(sb
, root
);
1198 down_write(&nilfs
->ns_sem
);
1199 nilfs_setup_super(sb
, true);
1200 up_write(&nilfs
->ns_sem
);
1206 sb
->s_flags
= old_sb_flags
;
1207 nilfs
->ns_mount_opt
= old_mount_opt
;
1211 struct nilfs_super_data
{
1212 struct block_device
*bdev
;
1218 * nilfs_identify - pre-read mount options needed to identify mount instance
1219 * @data: mount options
1220 * @sd: nilfs_super_data
1222 static int nilfs_identify(char *data
, struct nilfs_super_data
*sd
)
1224 char *p
, *options
= data
;
1225 substring_t args
[MAX_OPT_ARGS
];
1230 p
= strsep(&options
, ",");
1231 if (p
!= NULL
&& *p
) {
1232 token
= match_token(p
, tokens
, args
);
1233 if (token
== Opt_snapshot
) {
1234 if (!(sd
->flags
& MS_RDONLY
)) {
1237 sd
->cno
= simple_strtoull(args
[0].from
,
1240 * No need to see the end pointer;
1241 * match_token() has done syntax
1250 "NILFS: invalid mount option: %s\n", p
);
1254 BUG_ON(options
== data
);
1255 *(options
- 1) = ',';
1260 static int nilfs_set_bdev_super(struct super_block
*s
, void *data
)
1263 s
->s_dev
= s
->s_bdev
->bd_dev
;
1267 static int nilfs_test_bdev_super(struct super_block
*s
, void *data
)
1269 return (void *)s
->s_bdev
== data
;
1272 static struct dentry
*
1273 nilfs_mount(struct file_system_type
*fs_type
, int flags
,
1274 const char *dev_name
, void *data
)
1276 struct nilfs_super_data sd
;
1277 struct super_block
*s
;
1278 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1279 struct dentry
*root_dentry
;
1280 int err
, s_new
= false;
1282 if (!(flags
& MS_RDONLY
))
1283 mode
|= FMODE_WRITE
;
1285 sd
.bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1286 if (IS_ERR(sd
.bdev
))
1287 return ERR_CAST(sd
.bdev
);
1291 if (nilfs_identify((char *)data
, &sd
)) {
1297 * once the super is inserted into the list by sget, s_umount
1298 * will protect the lockfs code from trying to start a snapshot
1299 * while we are mounting
1301 mutex_lock(&sd
.bdev
->bd_fsfreeze_mutex
);
1302 if (sd
.bdev
->bd_fsfreeze_count
> 0) {
1303 mutex_unlock(&sd
.bdev
->bd_fsfreeze_mutex
);
1307 s
= sget(fs_type
, nilfs_test_bdev_super
, nilfs_set_bdev_super
, flags
,
1309 mutex_unlock(&sd
.bdev
->bd_fsfreeze_mutex
);
1316 char b
[BDEVNAME_SIZE
];
1320 /* New superblock instance created */
1322 strlcpy(s
->s_id
, bdevname(sd
.bdev
, b
), sizeof(s
->s_id
));
1323 sb_set_blocksize(s
, block_size(sd
.bdev
));
1325 err
= nilfs_fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1329 s
->s_flags
|= MS_ACTIVE
;
1330 } else if (!sd
.cno
) {
1331 if (nilfs_tree_is_busy(s
->s_root
)) {
1332 if ((flags
^ s
->s_flags
) & MS_RDONLY
) {
1333 printk(KERN_ERR
"NILFS: the device already "
1334 "has a %s mount.\n",
1335 (s
->s_flags
& MS_RDONLY
) ?
1336 "read-only" : "read/write");
1342 * Try remount to setup mount states if the current
1343 * tree is not mounted and only snapshots use this sb.
1345 err
= nilfs_remount(s
, &flags
, data
);
1352 err
= nilfs_attach_snapshot(s
, sd
.cno
, &root_dentry
);
1356 root_dentry
= dget(s
->s_root
);
1360 blkdev_put(sd
.bdev
, mode
);
1365 deactivate_locked_super(s
);
1369 blkdev_put(sd
.bdev
, mode
);
1370 return ERR_PTR(err
);
1373 struct file_system_type nilfs_fs_type
= {
1374 .owner
= THIS_MODULE
,
1376 .mount
= nilfs_mount
,
1377 .kill_sb
= kill_block_super
,
1378 .fs_flags
= FS_REQUIRES_DEV
,
1380 MODULE_ALIAS_FS("nilfs2");
1382 static void nilfs_inode_init_once(void *obj
)
1384 struct nilfs_inode_info
*ii
= obj
;
1386 INIT_LIST_HEAD(&ii
->i_dirty
);
1387 #ifdef CONFIG_NILFS_XATTR
1388 init_rwsem(&ii
->xattr_sem
);
1390 address_space_init_once(&ii
->i_btnode_cache
);
1391 ii
->i_bmap
= &ii
->i_bmap_data
;
1392 inode_init_once(&ii
->vfs_inode
);
1395 static void nilfs_segbuf_init_once(void *obj
)
1397 memset(obj
, 0, sizeof(struct nilfs_segment_buffer
));
1400 static void nilfs_destroy_cachep(void)
1403 * Make sure all delayed rcu free inodes are flushed before we
1408 if (nilfs_inode_cachep
)
1409 kmem_cache_destroy(nilfs_inode_cachep
);
1410 if (nilfs_transaction_cachep
)
1411 kmem_cache_destroy(nilfs_transaction_cachep
);
1412 if (nilfs_segbuf_cachep
)
1413 kmem_cache_destroy(nilfs_segbuf_cachep
);
1414 if (nilfs_btree_path_cache
)
1415 kmem_cache_destroy(nilfs_btree_path_cache
);
1418 static int __init
nilfs_init_cachep(void)
1420 nilfs_inode_cachep
= kmem_cache_create("nilfs2_inode_cache",
1421 sizeof(struct nilfs_inode_info
), 0,
1422 SLAB_RECLAIM_ACCOUNT
, nilfs_inode_init_once
);
1423 if (!nilfs_inode_cachep
)
1426 nilfs_transaction_cachep
= kmem_cache_create("nilfs2_transaction_cache",
1427 sizeof(struct nilfs_transaction_info
), 0,
1428 SLAB_RECLAIM_ACCOUNT
, NULL
);
1429 if (!nilfs_transaction_cachep
)
1432 nilfs_segbuf_cachep
= kmem_cache_create("nilfs2_segbuf_cache",
1433 sizeof(struct nilfs_segment_buffer
), 0,
1434 SLAB_RECLAIM_ACCOUNT
, nilfs_segbuf_init_once
);
1435 if (!nilfs_segbuf_cachep
)
1438 nilfs_btree_path_cache
= kmem_cache_create("nilfs2_btree_path_cache",
1439 sizeof(struct nilfs_btree_path
) * NILFS_BTREE_LEVEL_MAX
,
1441 if (!nilfs_btree_path_cache
)
1447 nilfs_destroy_cachep();
1451 static int __init
init_nilfs_fs(void)
1455 err
= nilfs_init_cachep();
1459 err
= nilfs_sysfs_init();
1463 err
= register_filesystem(&nilfs_fs_type
);
1465 goto deinit_sysfs_entry
;
1467 printk(KERN_INFO
"NILFS version 2 loaded\n");
1473 nilfs_destroy_cachep();
1478 static void __exit
exit_nilfs_fs(void)
1480 nilfs_destroy_cachep();
1482 unregister_filesystem(&nilfs_fs_type
);
1485 module_init(init_nilfs_fs
)
1486 module_exit(exit_nilfs_fs
)