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 void __nilfs_msg(struct super_block
*sb
, const char *level
, const char *fmt
,
84 printk("%sNILFS (%s): %pV\n", level
, sb
->s_id
, &vaf
);
86 printk("%sNILFS: %pV\n", level
, &vaf
);
90 static void nilfs_set_error(struct super_block
*sb
)
92 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
93 struct nilfs_super_block
**sbp
;
95 down_write(&nilfs
->ns_sem
);
96 if (!(nilfs
->ns_mount_state
& NILFS_ERROR_FS
)) {
97 nilfs
->ns_mount_state
|= NILFS_ERROR_FS
;
98 sbp
= nilfs_prepare_super(sb
, 0);
100 sbp
[0]->s_state
|= cpu_to_le16(NILFS_ERROR_FS
);
102 sbp
[1]->s_state
|= cpu_to_le16(NILFS_ERROR_FS
);
103 nilfs_commit_super(sb
, NILFS_SB_COMMIT_ALL
);
106 up_write(&nilfs
->ns_sem
);
110 * __nilfs_error() - report failure condition on a filesystem
112 * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
113 * reporting an error message. This function should be called when
114 * NILFS detects incoherences or defects of meta data on disk.
116 * This implements the body of nilfs_error() macro. Normally,
117 * nilfs_error() should be used. As for sustainable errors such as a
118 * single-shot I/O error, nilfs_msg() should be used instead.
120 * Callers should not add a trailing newline since this will do it.
122 void __nilfs_error(struct super_block
*sb
, const char *function
,
123 const char *fmt
, ...)
125 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
126 struct va_format vaf
;
134 printk(KERN_CRIT
"NILFS error (device %s): %s: %pV\n",
135 sb
->s_id
, function
, &vaf
);
139 if (!(sb
->s_flags
& MS_RDONLY
)) {
142 if (nilfs_test_opt(nilfs
, ERRORS_RO
)) {
143 printk(KERN_CRIT
"Remounting filesystem read-only\n");
144 sb
->s_flags
|= MS_RDONLY
;
148 if (nilfs_test_opt(nilfs
, ERRORS_PANIC
))
149 panic("NILFS (device %s): panic forced after error\n",
153 struct inode
*nilfs_alloc_inode(struct super_block
*sb
)
155 struct nilfs_inode_info
*ii
;
157 ii
= kmem_cache_alloc(nilfs_inode_cachep
, GFP_NOFS
);
163 ii
->vfs_inode
.i_version
= 1;
164 nilfs_mapping_init(&ii
->i_btnode_cache
, &ii
->vfs_inode
);
165 return &ii
->vfs_inode
;
168 static void nilfs_i_callback(struct rcu_head
*head
)
170 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
172 if (nilfs_is_metadata_file_inode(inode
))
173 nilfs_mdt_destroy(inode
);
175 kmem_cache_free(nilfs_inode_cachep
, NILFS_I(inode
));
178 void nilfs_destroy_inode(struct inode
*inode
)
180 call_rcu(&inode
->i_rcu
, nilfs_i_callback
);
183 static int nilfs_sync_super(struct super_block
*sb
, int flag
)
185 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
189 set_buffer_dirty(nilfs
->ns_sbh
[0]);
190 if (nilfs_test_opt(nilfs
, BARRIER
)) {
191 err
= __sync_dirty_buffer(nilfs
->ns_sbh
[0],
192 REQ_SYNC
| REQ_PREFLUSH
| REQ_FUA
);
194 err
= sync_dirty_buffer(nilfs
->ns_sbh
[0]);
198 nilfs_msg(sb
, KERN_ERR
, "unable to write superblock: err=%d",
200 if (err
== -EIO
&& nilfs
->ns_sbh
[1]) {
202 * sbp[0] points to newer log than sbp[1],
203 * so copy sbp[0] to sbp[1] to take over sbp[0].
205 memcpy(nilfs
->ns_sbp
[1], nilfs
->ns_sbp
[0],
207 nilfs_fall_back_super_block(nilfs
);
211 struct nilfs_super_block
*sbp
= nilfs
->ns_sbp
[0];
213 nilfs
->ns_sbwcount
++;
216 * The latest segment becomes trailable from the position
217 * written in superblock.
219 clear_nilfs_discontinued(nilfs
);
221 /* update GC protection for recent segments */
222 if (nilfs
->ns_sbh
[1]) {
223 if (flag
== NILFS_SB_COMMIT_ALL
) {
224 set_buffer_dirty(nilfs
->ns_sbh
[1]);
225 if (sync_dirty_buffer(nilfs
->ns_sbh
[1]) < 0)
228 if (le64_to_cpu(nilfs
->ns_sbp
[1]->s_last_cno
) <
229 le64_to_cpu(nilfs
->ns_sbp
[0]->s_last_cno
))
230 sbp
= nilfs
->ns_sbp
[1];
233 spin_lock(&nilfs
->ns_last_segment_lock
);
234 nilfs
->ns_prot_seq
= le64_to_cpu(sbp
->s_last_seq
);
235 spin_unlock(&nilfs
->ns_last_segment_lock
);
241 void nilfs_set_log_cursor(struct nilfs_super_block
*sbp
,
242 struct the_nilfs
*nilfs
)
244 sector_t nfreeblocks
;
246 /* nilfs->ns_sem must be locked by the caller. */
247 nilfs_count_free_blocks(nilfs
, &nfreeblocks
);
248 sbp
->s_free_blocks_count
= cpu_to_le64(nfreeblocks
);
250 spin_lock(&nilfs
->ns_last_segment_lock
);
251 sbp
->s_last_seq
= cpu_to_le64(nilfs
->ns_last_seq
);
252 sbp
->s_last_pseg
= cpu_to_le64(nilfs
->ns_last_pseg
);
253 sbp
->s_last_cno
= cpu_to_le64(nilfs
->ns_last_cno
);
254 spin_unlock(&nilfs
->ns_last_segment_lock
);
257 struct nilfs_super_block
**nilfs_prepare_super(struct super_block
*sb
,
260 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
261 struct nilfs_super_block
**sbp
= nilfs
->ns_sbp
;
263 /* nilfs->ns_sem must be locked by the caller. */
264 if (sbp
[0]->s_magic
!= cpu_to_le16(NILFS_SUPER_MAGIC
)) {
266 sbp
[1]->s_magic
== cpu_to_le16(NILFS_SUPER_MAGIC
)) {
267 memcpy(sbp
[0], sbp
[1], nilfs
->ns_sbsize
);
269 nilfs_msg(sb
, KERN_CRIT
, "superblock broke");
273 sbp
[1]->s_magic
!= cpu_to_le16(NILFS_SUPER_MAGIC
)) {
274 memcpy(sbp
[1], sbp
[0], nilfs
->ns_sbsize
);
278 nilfs_swap_super_block(nilfs
);
283 int nilfs_commit_super(struct super_block
*sb
, int flag
)
285 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
286 struct nilfs_super_block
**sbp
= nilfs
->ns_sbp
;
289 /* nilfs->ns_sem must be locked by the caller. */
291 nilfs
->ns_sbwtime
= t
;
292 sbp
[0]->s_wtime
= cpu_to_le64(t
);
294 sbp
[0]->s_sum
= cpu_to_le32(crc32_le(nilfs
->ns_crc_seed
,
295 (unsigned char *)sbp
[0],
297 if (flag
== NILFS_SB_COMMIT_ALL
&& sbp
[1]) {
298 sbp
[1]->s_wtime
= sbp
[0]->s_wtime
;
300 sbp
[1]->s_sum
= cpu_to_le32(crc32_le(nilfs
->ns_crc_seed
,
301 (unsigned char *)sbp
[1],
304 clear_nilfs_sb_dirty(nilfs
);
305 nilfs
->ns_flushed_device
= 1;
306 /* make sure store to ns_flushed_device cannot be reordered */
308 return nilfs_sync_super(sb
, flag
);
312 * nilfs_cleanup_super() - write filesystem state for cleanup
313 * @sb: super block instance to be unmounted or degraded to read-only
315 * This function restores state flags in the on-disk super block.
316 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
317 * filesystem was not clean previously.
319 int nilfs_cleanup_super(struct super_block
*sb
)
321 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
322 struct nilfs_super_block
**sbp
;
323 int flag
= NILFS_SB_COMMIT
;
326 sbp
= nilfs_prepare_super(sb
, 0);
328 sbp
[0]->s_state
= cpu_to_le16(nilfs
->ns_mount_state
);
329 nilfs_set_log_cursor(sbp
[0], nilfs
);
330 if (sbp
[1] && sbp
[0]->s_last_cno
== sbp
[1]->s_last_cno
) {
332 * make the "clean" flag also to the opposite
333 * super block if both super blocks point to
334 * the same checkpoint.
336 sbp
[1]->s_state
= sbp
[0]->s_state
;
337 flag
= NILFS_SB_COMMIT_ALL
;
339 ret
= nilfs_commit_super(sb
, flag
);
345 * nilfs_move_2nd_super - relocate secondary super block
346 * @sb: super block instance
347 * @sb2off: new offset of the secondary super block (in bytes)
349 static int nilfs_move_2nd_super(struct super_block
*sb
, loff_t sb2off
)
351 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
352 struct buffer_head
*nsbh
;
353 struct nilfs_super_block
*nsbp
;
354 sector_t blocknr
, newblocknr
;
355 unsigned long offset
;
356 int sb2i
; /* array index of the secondary superblock */
359 /* nilfs->ns_sem must be locked by the caller. */
360 if (nilfs
->ns_sbh
[1] &&
361 nilfs
->ns_sbh
[1]->b_blocknr
> nilfs
->ns_first_data_block
) {
363 blocknr
= nilfs
->ns_sbh
[1]->b_blocknr
;
364 } else if (nilfs
->ns_sbh
[0]->b_blocknr
> nilfs
->ns_first_data_block
) {
366 blocknr
= nilfs
->ns_sbh
[0]->b_blocknr
;
371 if (sb2i
>= 0 && (u64
)blocknr
<< nilfs
->ns_blocksize_bits
== sb2off
)
372 goto out
; /* super block location is unchanged */
374 /* Get new super block buffer */
375 newblocknr
= sb2off
>> nilfs
->ns_blocksize_bits
;
376 offset
= sb2off
& (nilfs
->ns_blocksize
- 1);
377 nsbh
= sb_getblk(sb
, newblocknr
);
379 nilfs_msg(sb
, KERN_WARNING
,
380 "unable to move secondary superblock to block %llu",
381 (unsigned long long)newblocknr
);
385 nsbp
= (void *)nsbh
->b_data
+ offset
;
386 memset(nsbp
, 0, nilfs
->ns_blocksize
);
389 memcpy(nsbp
, nilfs
->ns_sbp
[sb2i
], nilfs
->ns_sbsize
);
390 brelse(nilfs
->ns_sbh
[sb2i
]);
391 nilfs
->ns_sbh
[sb2i
] = nsbh
;
392 nilfs
->ns_sbp
[sb2i
] = nsbp
;
393 } else if (nilfs
->ns_sbh
[0]->b_blocknr
< nilfs
->ns_first_data_block
) {
394 /* secondary super block will be restored to index 1 */
395 nilfs
->ns_sbh
[1] = nsbh
;
396 nilfs
->ns_sbp
[1] = nsbp
;
405 * nilfs_resize_fs - resize the filesystem
406 * @sb: super block instance
407 * @newsize: new size of the filesystem (in bytes)
409 int nilfs_resize_fs(struct super_block
*sb
, __u64 newsize
)
411 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
412 struct nilfs_super_block
**sbp
;
413 __u64 devsize
, newnsegs
;
418 devsize
= i_size_read(sb
->s_bdev
->bd_inode
);
419 if (newsize
> devsize
)
423 * Write lock is required to protect some functions depending
424 * on the number of segments, the number of reserved segments,
427 down_write(&nilfs
->ns_segctor_sem
);
429 sb2off
= NILFS_SB2_OFFSET_BYTES(newsize
);
430 newnsegs
= sb2off
>> nilfs
->ns_blocksize_bits
;
431 do_div(newnsegs
, nilfs
->ns_blocks_per_segment
);
433 ret
= nilfs_sufile_resize(nilfs
->ns_sufile
, newnsegs
);
434 up_write(&nilfs
->ns_segctor_sem
);
438 ret
= nilfs_construct_segment(sb
);
442 down_write(&nilfs
->ns_sem
);
443 nilfs_move_2nd_super(sb
, sb2off
);
445 sbp
= nilfs_prepare_super(sb
, 0);
447 nilfs_set_log_cursor(sbp
[0], nilfs
);
449 * Drop NILFS_RESIZE_FS flag for compatibility with
450 * mount-time resize which may be implemented in a
453 sbp
[0]->s_state
= cpu_to_le16(le16_to_cpu(sbp
[0]->s_state
) &
455 sbp
[0]->s_dev_size
= cpu_to_le64(newsize
);
456 sbp
[0]->s_nsegments
= cpu_to_le64(nilfs
->ns_nsegments
);
458 memcpy(sbp
[1], sbp
[0], nilfs
->ns_sbsize
);
459 ret
= nilfs_commit_super(sb
, NILFS_SB_COMMIT_ALL
);
461 up_write(&nilfs
->ns_sem
);
464 * Reset the range of allocatable segments last. This order
465 * is important in the case of expansion because the secondary
466 * superblock must be protected from log write until migration
470 nilfs_sufile_set_alloc_range(nilfs
->ns_sufile
, 0, newnsegs
- 1);
475 static void nilfs_put_super(struct super_block
*sb
)
477 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
479 nilfs_detach_log_writer(sb
);
481 if (!(sb
->s_flags
& MS_RDONLY
)) {
482 down_write(&nilfs
->ns_sem
);
483 nilfs_cleanup_super(sb
);
484 up_write(&nilfs
->ns_sem
);
487 iput(nilfs
->ns_sufile
);
488 iput(nilfs
->ns_cpfile
);
491 destroy_nilfs(nilfs
);
492 sb
->s_fs_info
= NULL
;
495 static int nilfs_sync_fs(struct super_block
*sb
, int wait
)
497 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
498 struct nilfs_super_block
**sbp
;
501 /* This function is called when super block should be written back */
503 err
= nilfs_construct_segment(sb
);
505 down_write(&nilfs
->ns_sem
);
506 if (nilfs_sb_dirty(nilfs
)) {
507 sbp
= nilfs_prepare_super(sb
, nilfs_sb_will_flip(nilfs
));
509 nilfs_set_log_cursor(sbp
[0], nilfs
);
510 nilfs_commit_super(sb
, NILFS_SB_COMMIT
);
513 up_write(&nilfs
->ns_sem
);
516 err
= nilfs_flush_device(nilfs
);
521 int nilfs_attach_checkpoint(struct super_block
*sb
, __u64 cno
, int curr_mnt
,
522 struct nilfs_root
**rootp
)
524 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
525 struct nilfs_root
*root
;
526 struct nilfs_checkpoint
*raw_cp
;
527 struct buffer_head
*bh_cp
;
530 root
= nilfs_find_or_create_root(
531 nilfs
, curr_mnt
? NILFS_CPTREE_CURRENT_CNO
: cno
);
536 goto reuse
; /* already attached checkpoint */
538 down_read(&nilfs
->ns_segctor_sem
);
539 err
= nilfs_cpfile_get_checkpoint(nilfs
->ns_cpfile
, cno
, 0, &raw_cp
,
541 up_read(&nilfs
->ns_segctor_sem
);
543 if (err
== -ENOENT
|| err
== -EINVAL
) {
544 nilfs_msg(sb
, KERN_ERR
,
545 "Invalid checkpoint (checkpoint number=%llu)",
546 (unsigned long long)cno
);
552 err
= nilfs_ifile_read(sb
, root
, nilfs
->ns_inode_size
,
553 &raw_cp
->cp_ifile_inode
, &root
->ifile
);
557 atomic64_set(&root
->inodes_count
,
558 le64_to_cpu(raw_cp
->cp_inodes_count
));
559 atomic64_set(&root
->blocks_count
,
560 le64_to_cpu(raw_cp
->cp_blocks_count
));
562 nilfs_cpfile_put_checkpoint(nilfs
->ns_cpfile
, cno
, bh_cp
);
569 nilfs_cpfile_put_checkpoint(nilfs
->ns_cpfile
, cno
, bh_cp
);
571 nilfs_put_root(root
);
576 static int nilfs_freeze(struct super_block
*sb
)
578 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
581 if (sb
->s_flags
& MS_RDONLY
)
584 /* Mark super block clean */
585 down_write(&nilfs
->ns_sem
);
586 err
= nilfs_cleanup_super(sb
);
587 up_write(&nilfs
->ns_sem
);
591 static int nilfs_unfreeze(struct super_block
*sb
)
593 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
595 if (sb
->s_flags
& MS_RDONLY
)
598 down_write(&nilfs
->ns_sem
);
599 nilfs_setup_super(sb
, false);
600 up_write(&nilfs
->ns_sem
);
604 static int nilfs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
606 struct super_block
*sb
= dentry
->d_sb
;
607 struct nilfs_root
*root
= NILFS_I(d_inode(dentry
))->i_root
;
608 struct the_nilfs
*nilfs
= root
->nilfs
;
609 u64 id
= huge_encode_dev(sb
->s_bdev
->bd_dev
);
610 unsigned long long blocks
;
611 unsigned long overhead
;
612 unsigned long nrsvblocks
;
613 sector_t nfreeblocks
;
614 u64 nmaxinodes
, nfreeinodes
;
618 * Compute all of the segment blocks
620 * The blocks before first segment and after last segment
623 blocks
= nilfs
->ns_blocks_per_segment
* nilfs
->ns_nsegments
624 - nilfs
->ns_first_data_block
;
625 nrsvblocks
= nilfs
->ns_nrsvsegs
* nilfs
->ns_blocks_per_segment
;
628 * Compute the overhead
630 * When distributing meta data blocks outside segment structure,
631 * We must count them as the overhead.
635 err
= nilfs_count_free_blocks(nilfs
, &nfreeblocks
);
639 err
= nilfs_ifile_count_free_inodes(root
->ifile
,
640 &nmaxinodes
, &nfreeinodes
);
642 nilfs_msg(sb
, KERN_WARNING
,
643 "failed to count free inodes: err=%d", err
);
644 if (err
== -ERANGE
) {
646 * If nilfs_palloc_count_max_entries() returns
647 * -ERANGE error code then we simply treat
648 * curent inodes count as maximum possible and
649 * zero as free inodes value.
651 nmaxinodes
= atomic64_read(&root
->inodes_count
);
658 buf
->f_type
= NILFS_SUPER_MAGIC
;
659 buf
->f_bsize
= sb
->s_blocksize
;
660 buf
->f_blocks
= blocks
- overhead
;
661 buf
->f_bfree
= nfreeblocks
;
662 buf
->f_bavail
= (buf
->f_bfree
>= nrsvblocks
) ?
663 (buf
->f_bfree
- nrsvblocks
) : 0;
664 buf
->f_files
= nmaxinodes
;
665 buf
->f_ffree
= nfreeinodes
;
666 buf
->f_namelen
= NILFS_NAME_LEN
;
667 buf
->f_fsid
.val
[0] = (u32
)id
;
668 buf
->f_fsid
.val
[1] = (u32
)(id
>> 32);
673 static int nilfs_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
675 struct super_block
*sb
= dentry
->d_sb
;
676 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
677 struct nilfs_root
*root
= NILFS_I(d_inode(dentry
))->i_root
;
679 if (!nilfs_test_opt(nilfs
, BARRIER
))
680 seq_puts(seq
, ",nobarrier");
681 if (root
->cno
!= NILFS_CPTREE_CURRENT_CNO
)
682 seq_printf(seq
, ",cp=%llu", (unsigned long long)root
->cno
);
683 if (nilfs_test_opt(nilfs
, ERRORS_PANIC
))
684 seq_puts(seq
, ",errors=panic");
685 if (nilfs_test_opt(nilfs
, ERRORS_CONT
))
686 seq_puts(seq
, ",errors=continue");
687 if (nilfs_test_opt(nilfs
, STRICT_ORDER
))
688 seq_puts(seq
, ",order=strict");
689 if (nilfs_test_opt(nilfs
, NORECOVERY
))
690 seq_puts(seq
, ",norecovery");
691 if (nilfs_test_opt(nilfs
, DISCARD
))
692 seq_puts(seq
, ",discard");
697 static const struct super_operations nilfs_sops
= {
698 .alloc_inode
= nilfs_alloc_inode
,
699 .destroy_inode
= nilfs_destroy_inode
,
700 .dirty_inode
= nilfs_dirty_inode
,
701 .evict_inode
= nilfs_evict_inode
,
702 .put_super
= nilfs_put_super
,
703 .sync_fs
= nilfs_sync_fs
,
704 .freeze_fs
= nilfs_freeze
,
705 .unfreeze_fs
= nilfs_unfreeze
,
706 .statfs
= nilfs_statfs
,
707 .remount_fs
= nilfs_remount
,
708 .show_options
= nilfs_show_options
712 Opt_err_cont
, Opt_err_panic
, Opt_err_ro
,
713 Opt_barrier
, Opt_nobarrier
, Opt_snapshot
, Opt_order
, Opt_norecovery
,
714 Opt_discard
, Opt_nodiscard
, Opt_err
,
717 static match_table_t tokens
= {
718 {Opt_err_cont
, "errors=continue"},
719 {Opt_err_panic
, "errors=panic"},
720 {Opt_err_ro
, "errors=remount-ro"},
721 {Opt_barrier
, "barrier"},
722 {Opt_nobarrier
, "nobarrier"},
723 {Opt_snapshot
, "cp=%u"},
724 {Opt_order
, "order=%s"},
725 {Opt_norecovery
, "norecovery"},
726 {Opt_discard
, "discard"},
727 {Opt_nodiscard
, "nodiscard"},
731 static int parse_options(char *options
, struct super_block
*sb
, int is_remount
)
733 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
735 substring_t args
[MAX_OPT_ARGS
];
740 while ((p
= strsep(&options
, ",")) != NULL
) {
746 token
= match_token(p
, tokens
, args
);
749 nilfs_set_opt(nilfs
, BARRIER
);
752 nilfs_clear_opt(nilfs
, BARRIER
);
755 if (strcmp(args
[0].from
, "relaxed") == 0)
756 /* Ordered data semantics */
757 nilfs_clear_opt(nilfs
, STRICT_ORDER
);
758 else if (strcmp(args
[0].from
, "strict") == 0)
759 /* Strict in-order semantics */
760 nilfs_set_opt(nilfs
, STRICT_ORDER
);
765 nilfs_write_opt(nilfs
, ERROR_MODE
, ERRORS_PANIC
);
768 nilfs_write_opt(nilfs
, ERROR_MODE
, ERRORS_RO
);
771 nilfs_write_opt(nilfs
, ERROR_MODE
, ERRORS_CONT
);
775 nilfs_msg(sb
, KERN_ERR
,
776 "\"%s\" option is invalid for remount",
782 nilfs_set_opt(nilfs
, NORECOVERY
);
785 nilfs_set_opt(nilfs
, DISCARD
);
788 nilfs_clear_opt(nilfs
, DISCARD
);
791 nilfs_msg(sb
, KERN_ERR
,
792 "unrecognized mount option \"%s\"", p
);
800 nilfs_set_default_options(struct super_block
*sb
,
801 struct nilfs_super_block
*sbp
)
803 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
805 nilfs
->ns_mount_opt
=
806 NILFS_MOUNT_ERRORS_RO
| NILFS_MOUNT_BARRIER
;
809 static int nilfs_setup_super(struct super_block
*sb
, int is_mount
)
811 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
812 struct nilfs_super_block
**sbp
;
816 /* nilfs->ns_sem must be locked by the caller. */
817 sbp
= nilfs_prepare_super(sb
, 0);
822 goto skip_mount_setup
;
824 max_mnt_count
= le16_to_cpu(sbp
[0]->s_max_mnt_count
);
825 mnt_count
= le16_to_cpu(sbp
[0]->s_mnt_count
);
827 if (nilfs
->ns_mount_state
& NILFS_ERROR_FS
) {
828 nilfs_msg(sb
, KERN_WARNING
, "mounting fs with errors");
830 } else if (max_mnt_count
>= 0 && mnt_count
>= max_mnt_count
) {
831 nilfs_msg(sb
, KERN_WARNING
, "maximal mount count reached");
835 sbp
[0]->s_max_mnt_count
= cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT
);
837 sbp
[0]->s_mnt_count
= cpu_to_le16(mnt_count
+ 1);
838 sbp
[0]->s_mtime
= cpu_to_le64(get_seconds());
842 cpu_to_le16(le16_to_cpu(sbp
[0]->s_state
) & ~NILFS_VALID_FS
);
843 /* synchronize sbp[1] with sbp[0] */
845 memcpy(sbp
[1], sbp
[0], nilfs
->ns_sbsize
);
846 return nilfs_commit_super(sb
, NILFS_SB_COMMIT_ALL
);
849 struct nilfs_super_block
*nilfs_read_super_block(struct super_block
*sb
,
850 u64 pos
, int blocksize
,
851 struct buffer_head
**pbh
)
853 unsigned long long sb_index
= pos
;
854 unsigned long offset
;
856 offset
= do_div(sb_index
, blocksize
);
857 *pbh
= sb_bread(sb
, sb_index
);
860 return (struct nilfs_super_block
*)((char *)(*pbh
)->b_data
+ offset
);
863 int nilfs_store_magic_and_option(struct super_block
*sb
,
864 struct nilfs_super_block
*sbp
,
867 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
869 sb
->s_magic
= le16_to_cpu(sbp
->s_magic
);
871 /* FS independent flags */
872 #ifdef NILFS_ATIME_DISABLE
873 sb
->s_flags
|= MS_NOATIME
;
876 nilfs_set_default_options(sb
, sbp
);
878 nilfs
->ns_resuid
= le16_to_cpu(sbp
->s_def_resuid
);
879 nilfs
->ns_resgid
= le16_to_cpu(sbp
->s_def_resgid
);
880 nilfs
->ns_interval
= le32_to_cpu(sbp
->s_c_interval
);
881 nilfs
->ns_watermark
= le32_to_cpu(sbp
->s_c_block_max
);
883 return !parse_options(data
, sb
, 0) ? -EINVAL
: 0;
886 int nilfs_check_feature_compatibility(struct super_block
*sb
,
887 struct nilfs_super_block
*sbp
)
891 features
= le64_to_cpu(sbp
->s_feature_incompat
) &
892 ~NILFS_FEATURE_INCOMPAT_SUPP
;
894 nilfs_msg(sb
, KERN_ERR
,
895 "couldn't mount because of unsupported optional features (%llx)",
896 (unsigned long long)features
);
899 features
= le64_to_cpu(sbp
->s_feature_compat_ro
) &
900 ~NILFS_FEATURE_COMPAT_RO_SUPP
;
901 if (!(sb
->s_flags
& MS_RDONLY
) && features
) {
902 nilfs_msg(sb
, KERN_ERR
,
903 "couldn't mount RDWR because of unsupported optional features (%llx)",
904 (unsigned long long)features
);
910 static int nilfs_get_root_dentry(struct super_block
*sb
,
911 struct nilfs_root
*root
,
912 struct dentry
**root_dentry
)
915 struct dentry
*dentry
;
918 inode
= nilfs_iget(sb
, root
, NILFS_ROOT_INO
);
920 ret
= PTR_ERR(inode
);
921 nilfs_msg(sb
, KERN_ERR
, "error %d getting root inode", ret
);
924 if (!S_ISDIR(inode
->i_mode
) || !inode
->i_blocks
|| !inode
->i_size
) {
926 nilfs_msg(sb
, KERN_ERR
, "corrupt root inode");
931 if (root
->cno
== NILFS_CPTREE_CURRENT_CNO
) {
932 dentry
= d_find_alias(inode
);
934 dentry
= d_make_root(inode
);
943 dentry
= d_obtain_root(inode
);
944 if (IS_ERR(dentry
)) {
945 ret
= PTR_ERR(dentry
);
949 *root_dentry
= dentry
;
954 nilfs_msg(sb
, KERN_ERR
, "error %d getting root dentry", ret
);
958 static int nilfs_attach_snapshot(struct super_block
*s
, __u64 cno
,
959 struct dentry
**root_dentry
)
961 struct the_nilfs
*nilfs
= s
->s_fs_info
;
962 struct nilfs_root
*root
;
965 mutex_lock(&nilfs
->ns_snapshot_mount_mutex
);
967 down_read(&nilfs
->ns_segctor_sem
);
968 ret
= nilfs_cpfile_is_snapshot(nilfs
->ns_cpfile
, cno
);
969 up_read(&nilfs
->ns_segctor_sem
);
971 ret
= (ret
== -ENOENT
) ? -EINVAL
: ret
;
974 nilfs_msg(s
, KERN_ERR
,
975 "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
976 (unsigned long long)cno
);
981 ret
= nilfs_attach_checkpoint(s
, cno
, false, &root
);
983 nilfs_msg(s
, KERN_ERR
,
984 "error %d while loading snapshot (checkpoint number=%llu)",
985 ret
, (unsigned long long)cno
);
988 ret
= nilfs_get_root_dentry(s
, root
, root_dentry
);
989 nilfs_put_root(root
);
991 mutex_unlock(&nilfs
->ns_snapshot_mount_mutex
);
996 * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
997 * @root_dentry: root dentry of the tree to be shrunk
999 * This function returns true if the tree was in-use.
1001 static bool nilfs_tree_is_busy(struct dentry
*root_dentry
)
1003 shrink_dcache_parent(root_dentry
);
1004 return d_count(root_dentry
) > 1;
1007 int nilfs_checkpoint_is_mounted(struct super_block
*sb
, __u64 cno
)
1009 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
1010 struct nilfs_root
*root
;
1011 struct inode
*inode
;
1012 struct dentry
*dentry
;
1015 if (cno
> nilfs
->ns_cno
)
1018 if (cno
>= nilfs_last_cno(nilfs
))
1019 return true; /* protect recent checkpoints */
1022 root
= nilfs_lookup_root(nilfs
, cno
);
1024 inode
= nilfs_ilookup(sb
, root
, NILFS_ROOT_INO
);
1026 dentry
= d_find_alias(inode
);
1028 ret
= nilfs_tree_is_busy(dentry
);
1033 nilfs_put_root(root
);
1039 * nilfs_fill_super() - initialize a super block instance
1041 * @data: mount options
1042 * @silent: silent mode flag
1044 * This function is called exclusively by nilfs->ns_mount_mutex.
1045 * So, the recovery process is protected from other simultaneous mounts.
1048 nilfs_fill_super(struct super_block
*sb
, void *data
, int silent
)
1050 struct the_nilfs
*nilfs
;
1051 struct nilfs_root
*fsroot
;
1055 nilfs
= alloc_nilfs(sb
);
1059 sb
->s_fs_info
= nilfs
;
1061 err
= init_nilfs(nilfs
, sb
, (char *)data
);
1065 sb
->s_op
= &nilfs_sops
;
1066 sb
->s_export_op
= &nilfs_export_ops
;
1068 sb
->s_time_gran
= 1;
1069 sb
->s_max_links
= NILFS_LINK_MAX
;
1071 sb
->s_bdi
= bdev_get_queue(sb
->s_bdev
)->backing_dev_info
;
1073 err
= load_nilfs(nilfs
, sb
);
1077 cno
= nilfs_last_cno(nilfs
);
1078 err
= nilfs_attach_checkpoint(sb
, cno
, true, &fsroot
);
1080 nilfs_msg(sb
, KERN_ERR
,
1081 "error %d while loading last checkpoint (checkpoint number=%llu)",
1082 err
, (unsigned long long)cno
);
1086 if (!(sb
->s_flags
& MS_RDONLY
)) {
1087 err
= nilfs_attach_log_writer(sb
, fsroot
);
1089 goto failed_checkpoint
;
1092 err
= nilfs_get_root_dentry(sb
, fsroot
, &sb
->s_root
);
1094 goto failed_segctor
;
1096 nilfs_put_root(fsroot
);
1098 if (!(sb
->s_flags
& MS_RDONLY
)) {
1099 down_write(&nilfs
->ns_sem
);
1100 nilfs_setup_super(sb
, true);
1101 up_write(&nilfs
->ns_sem
);
1107 nilfs_detach_log_writer(sb
);
1110 nilfs_put_root(fsroot
);
1113 iput(nilfs
->ns_sufile
);
1114 iput(nilfs
->ns_cpfile
);
1115 iput(nilfs
->ns_dat
);
1118 destroy_nilfs(nilfs
);
1122 static int nilfs_remount(struct super_block
*sb
, int *flags
, char *data
)
1124 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
1125 unsigned long old_sb_flags
;
1126 unsigned long old_mount_opt
;
1129 sync_filesystem(sb
);
1130 old_sb_flags
= sb
->s_flags
;
1131 old_mount_opt
= nilfs
->ns_mount_opt
;
1133 if (!parse_options(data
, sb
, 1)) {
1137 sb
->s_flags
= (sb
->s_flags
& ~MS_POSIXACL
);
1141 if (!nilfs_valid_fs(nilfs
)) {
1142 nilfs_msg(sb
, KERN_WARNING
,
1143 "couldn't remount because the filesystem is in an incomplete recovery state");
1147 if ((*flags
& MS_RDONLY
) == (sb
->s_flags
& MS_RDONLY
))
1149 if (*flags
& MS_RDONLY
) {
1150 /* Shutting down log writer */
1151 nilfs_detach_log_writer(sb
);
1152 sb
->s_flags
|= MS_RDONLY
;
1155 * Remounting a valid RW partition RDONLY, so set
1156 * the RDONLY flag and then mark the partition as valid again.
1158 down_write(&nilfs
->ns_sem
);
1159 nilfs_cleanup_super(sb
);
1160 up_write(&nilfs
->ns_sem
);
1163 struct nilfs_root
*root
;
1166 * Mounting a RDONLY partition read-write, so reread and
1167 * store the current valid flag. (It may have been changed
1168 * by fsck since we originally mounted the partition.)
1170 down_read(&nilfs
->ns_sem
);
1171 features
= le64_to_cpu(nilfs
->ns_sbp
[0]->s_feature_compat_ro
) &
1172 ~NILFS_FEATURE_COMPAT_RO_SUPP
;
1173 up_read(&nilfs
->ns_sem
);
1175 nilfs_msg(sb
, KERN_WARNING
,
1176 "couldn't remount RDWR because of unsupported optional features (%llx)",
1177 (unsigned long long)features
);
1182 sb
->s_flags
&= ~MS_RDONLY
;
1184 root
= NILFS_I(d_inode(sb
->s_root
))->i_root
;
1185 err
= nilfs_attach_log_writer(sb
, root
);
1189 down_write(&nilfs
->ns_sem
);
1190 nilfs_setup_super(sb
, true);
1191 up_write(&nilfs
->ns_sem
);
1197 sb
->s_flags
= old_sb_flags
;
1198 nilfs
->ns_mount_opt
= old_mount_opt
;
1202 struct nilfs_super_data
{
1203 struct block_device
*bdev
;
1208 static int nilfs_parse_snapshot_option(const char *option
,
1209 const substring_t
*arg
,
1210 struct nilfs_super_data
*sd
)
1212 unsigned long long val
;
1213 const char *msg
= NULL
;
1216 if (!(sd
->flags
& MS_RDONLY
)) {
1217 msg
= "read-only option is not specified";
1221 err
= kstrtoull(arg
->from
, 0, &val
);
1224 msg
= "too large checkpoint number";
1226 msg
= "malformed argument";
1228 } else if (val
== 0) {
1229 msg
= "invalid checkpoint number 0";
1236 nilfs_msg(NULL
, KERN_ERR
, "invalid option \"%s\": %s", option
, msg
);
1241 * nilfs_identify - pre-read mount options needed to identify mount instance
1242 * @data: mount options
1243 * @sd: nilfs_super_data
1245 static int nilfs_identify(char *data
, struct nilfs_super_data
*sd
)
1247 char *p
, *options
= data
;
1248 substring_t args
[MAX_OPT_ARGS
];
1253 p
= strsep(&options
, ",");
1254 if (p
!= NULL
&& *p
) {
1255 token
= match_token(p
, tokens
, args
);
1256 if (token
== Opt_snapshot
)
1257 ret
= nilfs_parse_snapshot_option(p
, &args
[0],
1262 BUG_ON(options
== data
);
1263 *(options
- 1) = ',';
1268 static int nilfs_set_bdev_super(struct super_block
*s
, void *data
)
1271 s
->s_dev
= s
->s_bdev
->bd_dev
;
1275 static int nilfs_test_bdev_super(struct super_block
*s
, void *data
)
1277 return (void *)s
->s_bdev
== data
;
1280 static struct dentry
*
1281 nilfs_mount(struct file_system_type
*fs_type
, int flags
,
1282 const char *dev_name
, void *data
)
1284 struct nilfs_super_data sd
;
1285 struct super_block
*s
;
1286 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1287 struct dentry
*root_dentry
;
1288 int err
, s_new
= false;
1290 if (!(flags
& MS_RDONLY
))
1291 mode
|= FMODE_WRITE
;
1293 sd
.bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1294 if (IS_ERR(sd
.bdev
))
1295 return ERR_CAST(sd
.bdev
);
1299 if (nilfs_identify((char *)data
, &sd
)) {
1305 * once the super is inserted into the list by sget, s_umount
1306 * will protect the lockfs code from trying to start a snapshot
1307 * while we are mounting
1309 mutex_lock(&sd
.bdev
->bd_fsfreeze_mutex
);
1310 if (sd
.bdev
->bd_fsfreeze_count
> 0) {
1311 mutex_unlock(&sd
.bdev
->bd_fsfreeze_mutex
);
1315 s
= sget(fs_type
, nilfs_test_bdev_super
, nilfs_set_bdev_super
, flags
,
1317 mutex_unlock(&sd
.bdev
->bd_fsfreeze_mutex
);
1326 /* New superblock instance created */
1328 snprintf(s
->s_id
, sizeof(s
->s_id
), "%pg", sd
.bdev
);
1329 sb_set_blocksize(s
, block_size(sd
.bdev
));
1331 err
= nilfs_fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1335 s
->s_flags
|= MS_ACTIVE
;
1336 } else if (!sd
.cno
) {
1337 if (nilfs_tree_is_busy(s
->s_root
)) {
1338 if ((flags
^ s
->s_flags
) & MS_RDONLY
) {
1339 nilfs_msg(s
, KERN_ERR
,
1340 "the device already has a %s mount.",
1341 (s
->s_flags
& MS_RDONLY
) ?
1342 "read-only" : "read/write");
1348 * Try remount to setup mount states if the current
1349 * tree is not mounted and only snapshots use this sb.
1351 err
= nilfs_remount(s
, &flags
, data
);
1358 err
= nilfs_attach_snapshot(s
, sd
.cno
, &root_dentry
);
1362 root_dentry
= dget(s
->s_root
);
1366 blkdev_put(sd
.bdev
, mode
);
1371 deactivate_locked_super(s
);
1375 blkdev_put(sd
.bdev
, mode
);
1376 return ERR_PTR(err
);
1379 struct file_system_type nilfs_fs_type
= {
1380 .owner
= THIS_MODULE
,
1382 .mount
= nilfs_mount
,
1383 .kill_sb
= kill_block_super
,
1384 .fs_flags
= FS_REQUIRES_DEV
,
1386 MODULE_ALIAS_FS("nilfs2");
1388 static void nilfs_inode_init_once(void *obj
)
1390 struct nilfs_inode_info
*ii
= obj
;
1392 INIT_LIST_HEAD(&ii
->i_dirty
);
1393 #ifdef CONFIG_NILFS_XATTR
1394 init_rwsem(&ii
->xattr_sem
);
1396 address_space_init_once(&ii
->i_btnode_cache
);
1397 ii
->i_bmap
= &ii
->i_bmap_data
;
1398 inode_init_once(&ii
->vfs_inode
);
1401 static void nilfs_segbuf_init_once(void *obj
)
1403 memset(obj
, 0, sizeof(struct nilfs_segment_buffer
));
1406 static void nilfs_destroy_cachep(void)
1409 * Make sure all delayed rcu free inodes are flushed before we
1414 kmem_cache_destroy(nilfs_inode_cachep
);
1415 kmem_cache_destroy(nilfs_transaction_cachep
);
1416 kmem_cache_destroy(nilfs_segbuf_cachep
);
1417 kmem_cache_destroy(nilfs_btree_path_cache
);
1420 static int __init
nilfs_init_cachep(void)
1422 nilfs_inode_cachep
= kmem_cache_create("nilfs2_inode_cache",
1423 sizeof(struct nilfs_inode_info
), 0,
1424 SLAB_RECLAIM_ACCOUNT
|SLAB_ACCOUNT
,
1425 nilfs_inode_init_once
);
1426 if (!nilfs_inode_cachep
)
1429 nilfs_transaction_cachep
= kmem_cache_create("nilfs2_transaction_cache",
1430 sizeof(struct nilfs_transaction_info
), 0,
1431 SLAB_RECLAIM_ACCOUNT
, NULL
);
1432 if (!nilfs_transaction_cachep
)
1435 nilfs_segbuf_cachep
= kmem_cache_create("nilfs2_segbuf_cache",
1436 sizeof(struct nilfs_segment_buffer
), 0,
1437 SLAB_RECLAIM_ACCOUNT
, nilfs_segbuf_init_once
);
1438 if (!nilfs_segbuf_cachep
)
1441 nilfs_btree_path_cache
= kmem_cache_create("nilfs2_btree_path_cache",
1442 sizeof(struct nilfs_btree_path
) * NILFS_BTREE_LEVEL_MAX
,
1444 if (!nilfs_btree_path_cache
)
1450 nilfs_destroy_cachep();
1454 static int __init
init_nilfs_fs(void)
1458 err
= nilfs_init_cachep();
1462 err
= nilfs_sysfs_init();
1466 err
= register_filesystem(&nilfs_fs_type
);
1468 goto deinit_sysfs_entry
;
1470 printk(KERN_INFO
"NILFS version 2 loaded\n");
1476 nilfs_destroy_cachep();
1481 static void __exit
exit_nilfs_fs(void)
1483 nilfs_destroy_cachep();
1485 unregister_filesystem(&nilfs_fs_type
);
1488 module_init(init_nilfs_fs
)
1489 module_exit(exit_nilfs_fs
)