1 // SPDX-License-Identifier: GPL-2.0+
3 * NILFS module and super block management.
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
7 * Written by Ryusuke Konishi.
10 * linux/fs/ext2/super.c
12 * Copyright (C) 1992, 1993, 1994, 1995
13 * Remy Card (card@masi.ibp.fr)
14 * Laboratoire MASI - Institut Blaise Pascal
15 * Universite Pierre et Marie Curie (Paris VI)
19 * linux/fs/minix/inode.c
21 * Copyright (C) 1991, 1992 Linus Torvalds
23 * Big-endian to little-endian byte-swapping/bitmaps by
24 * David S. Miller (davem@caip.rutgers.edu), 1995
27 #include <linux/module.h>
28 #include <linux/string.h>
29 #include <linux/slab.h>
30 #include <linux/init.h>
31 #include <linux/blkdev.h>
32 #include <linux/parser.h>
33 #include <linux/crc32.h>
34 #include <linux/vfs.h>
35 #include <linux/writeback.h>
36 #include <linux/seq_file.h>
37 #include <linux/mount.h>
46 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
52 MODULE_AUTHOR("NTT Corp.");
53 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
55 MODULE_LICENSE("GPL");
57 static struct kmem_cache
*nilfs_inode_cachep
;
58 struct kmem_cache
*nilfs_transaction_cachep
;
59 struct kmem_cache
*nilfs_segbuf_cachep
;
60 struct kmem_cache
*nilfs_btree_path_cache
;
62 static int nilfs_setup_super(struct super_block
*sb
, int is_mount
);
63 static int nilfs_remount(struct super_block
*sb
, int *flags
, char *data
);
65 void __nilfs_msg(struct super_block
*sb
, const char *fmt
, ...)
73 level
= printk_get_level(fmt
);
74 vaf
.fmt
= printk_skip_level(fmt
);
78 printk("%c%cNILFS (%s): %pV\n",
79 KERN_SOH_ASCII
, level
, sb
->s_id
, &vaf
);
81 printk("%c%cNILFS: %pV\n",
82 KERN_SOH_ASCII
, level
, &vaf
);
87 static void nilfs_set_error(struct super_block
*sb
)
89 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
90 struct nilfs_super_block
**sbp
;
92 down_write(&nilfs
->ns_sem
);
93 if (!(nilfs
->ns_mount_state
& NILFS_ERROR_FS
)) {
94 nilfs
->ns_mount_state
|= NILFS_ERROR_FS
;
95 sbp
= nilfs_prepare_super(sb
, 0);
97 sbp
[0]->s_state
|= cpu_to_le16(NILFS_ERROR_FS
);
99 sbp
[1]->s_state
|= cpu_to_le16(NILFS_ERROR_FS
);
100 nilfs_commit_super(sb
, NILFS_SB_COMMIT_ALL
);
103 up_write(&nilfs
->ns_sem
);
107 * __nilfs_error() - report failure condition on a filesystem
109 * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
110 * reporting an error message. This function should be called when
111 * NILFS detects incoherences or defects of meta data on disk.
113 * This implements the body of nilfs_error() macro. Normally,
114 * nilfs_error() should be used. As for sustainable errors such as a
115 * single-shot I/O error, nilfs_err() should be used instead.
117 * Callers should not add a trailing newline since this will do it.
119 void __nilfs_error(struct super_block
*sb
, const char *function
,
120 const char *fmt
, ...)
122 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
123 struct va_format vaf
;
131 printk(KERN_CRIT
"NILFS error (device %s): %s: %pV\n",
132 sb
->s_id
, function
, &vaf
);
136 if (!sb_rdonly(sb
)) {
139 if (nilfs_test_opt(nilfs
, ERRORS_RO
)) {
140 printk(KERN_CRIT
"Remounting filesystem read-only\n");
141 sb
->s_flags
|= SB_RDONLY
;
145 if (nilfs_test_opt(nilfs
, ERRORS_PANIC
))
146 panic("NILFS (device %s): panic forced after error\n",
150 struct inode
*nilfs_alloc_inode(struct super_block
*sb
)
152 struct nilfs_inode_info
*ii
;
154 ii
= alloc_inode_sb(sb
, nilfs_inode_cachep
, GFP_NOFS
);
160 ii
->i_assoc_inode
= NULL
;
161 ii
->i_bmap
= &ii
->i_bmap_data
;
162 return &ii
->vfs_inode
;
165 static void nilfs_free_inode(struct inode
*inode
)
167 if (nilfs_is_metadata_file_inode(inode
))
168 nilfs_mdt_destroy(inode
);
170 kmem_cache_free(nilfs_inode_cachep
, NILFS_I(inode
));
173 static int nilfs_sync_super(struct super_block
*sb
, int flag
)
175 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
179 set_buffer_dirty(nilfs
->ns_sbh
[0]);
180 if (nilfs_test_opt(nilfs
, BARRIER
)) {
181 err
= __sync_dirty_buffer(nilfs
->ns_sbh
[0],
182 REQ_SYNC
| REQ_PREFLUSH
| REQ_FUA
);
184 err
= sync_dirty_buffer(nilfs
->ns_sbh
[0]);
188 nilfs_err(sb
, "unable to write superblock: err=%d", err
);
189 if (err
== -EIO
&& nilfs
->ns_sbh
[1]) {
191 * sbp[0] points to newer log than sbp[1],
192 * so copy sbp[0] to sbp[1] to take over sbp[0].
194 memcpy(nilfs
->ns_sbp
[1], nilfs
->ns_sbp
[0],
196 nilfs_fall_back_super_block(nilfs
);
200 struct nilfs_super_block
*sbp
= nilfs
->ns_sbp
[0];
202 nilfs
->ns_sbwcount
++;
205 * The latest segment becomes trailable from the position
206 * written in superblock.
208 clear_nilfs_discontinued(nilfs
);
210 /* update GC protection for recent segments */
211 if (nilfs
->ns_sbh
[1]) {
212 if (flag
== NILFS_SB_COMMIT_ALL
) {
213 set_buffer_dirty(nilfs
->ns_sbh
[1]);
214 if (sync_dirty_buffer(nilfs
->ns_sbh
[1]) < 0)
217 if (le64_to_cpu(nilfs
->ns_sbp
[1]->s_last_cno
) <
218 le64_to_cpu(nilfs
->ns_sbp
[0]->s_last_cno
))
219 sbp
= nilfs
->ns_sbp
[1];
222 spin_lock(&nilfs
->ns_last_segment_lock
);
223 nilfs
->ns_prot_seq
= le64_to_cpu(sbp
->s_last_seq
);
224 spin_unlock(&nilfs
->ns_last_segment_lock
);
230 void nilfs_set_log_cursor(struct nilfs_super_block
*sbp
,
231 struct the_nilfs
*nilfs
)
233 sector_t nfreeblocks
;
235 /* nilfs->ns_sem must be locked by the caller. */
236 nilfs_count_free_blocks(nilfs
, &nfreeblocks
);
237 sbp
->s_free_blocks_count
= cpu_to_le64(nfreeblocks
);
239 spin_lock(&nilfs
->ns_last_segment_lock
);
240 sbp
->s_last_seq
= cpu_to_le64(nilfs
->ns_last_seq
);
241 sbp
->s_last_pseg
= cpu_to_le64(nilfs
->ns_last_pseg
);
242 sbp
->s_last_cno
= cpu_to_le64(nilfs
->ns_last_cno
);
243 spin_unlock(&nilfs
->ns_last_segment_lock
);
246 struct nilfs_super_block
**nilfs_prepare_super(struct super_block
*sb
,
249 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
250 struct nilfs_super_block
**sbp
= nilfs
->ns_sbp
;
252 /* nilfs->ns_sem must be locked by the caller. */
253 if (sbp
[0]->s_magic
!= cpu_to_le16(NILFS_SUPER_MAGIC
)) {
255 sbp
[1]->s_magic
== cpu_to_le16(NILFS_SUPER_MAGIC
)) {
256 memcpy(sbp
[0], sbp
[1], nilfs
->ns_sbsize
);
258 nilfs_crit(sb
, "superblock broke");
262 sbp
[1]->s_magic
!= cpu_to_le16(NILFS_SUPER_MAGIC
)) {
263 memcpy(sbp
[1], sbp
[0], nilfs
->ns_sbsize
);
267 nilfs_swap_super_block(nilfs
);
272 int nilfs_commit_super(struct super_block
*sb
, int flag
)
274 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
275 struct nilfs_super_block
**sbp
= nilfs
->ns_sbp
;
278 /* nilfs->ns_sem must be locked by the caller. */
279 t
= ktime_get_real_seconds();
280 nilfs
->ns_sbwtime
= t
;
281 sbp
[0]->s_wtime
= cpu_to_le64(t
);
283 sbp
[0]->s_sum
= cpu_to_le32(crc32_le(nilfs
->ns_crc_seed
,
284 (unsigned char *)sbp
[0],
286 if (flag
== NILFS_SB_COMMIT_ALL
&& sbp
[1]) {
287 sbp
[1]->s_wtime
= sbp
[0]->s_wtime
;
289 sbp
[1]->s_sum
= cpu_to_le32(crc32_le(nilfs
->ns_crc_seed
,
290 (unsigned char *)sbp
[1],
293 clear_nilfs_sb_dirty(nilfs
);
294 nilfs
->ns_flushed_device
= 1;
295 /* make sure store to ns_flushed_device cannot be reordered */
297 return nilfs_sync_super(sb
, flag
);
301 * nilfs_cleanup_super() - write filesystem state for cleanup
302 * @sb: super block instance to be unmounted or degraded to read-only
304 * This function restores state flags in the on-disk super block.
305 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
306 * filesystem was not clean previously.
308 int nilfs_cleanup_super(struct super_block
*sb
)
310 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
311 struct nilfs_super_block
**sbp
;
312 int flag
= NILFS_SB_COMMIT
;
315 sbp
= nilfs_prepare_super(sb
, 0);
317 sbp
[0]->s_state
= cpu_to_le16(nilfs
->ns_mount_state
);
318 nilfs_set_log_cursor(sbp
[0], nilfs
);
319 if (sbp
[1] && sbp
[0]->s_last_cno
== sbp
[1]->s_last_cno
) {
321 * make the "clean" flag also to the opposite
322 * super block if both super blocks point to
323 * the same checkpoint.
325 sbp
[1]->s_state
= sbp
[0]->s_state
;
326 flag
= NILFS_SB_COMMIT_ALL
;
328 ret
= nilfs_commit_super(sb
, flag
);
334 * nilfs_move_2nd_super - relocate secondary super block
335 * @sb: super block instance
336 * @sb2off: new offset of the secondary super block (in bytes)
338 static int nilfs_move_2nd_super(struct super_block
*sb
, loff_t sb2off
)
340 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
341 struct buffer_head
*nsbh
;
342 struct nilfs_super_block
*nsbp
;
343 sector_t blocknr
, newblocknr
;
344 unsigned long offset
;
345 int sb2i
; /* array index of the secondary superblock */
348 /* nilfs->ns_sem must be locked by the caller. */
349 if (nilfs
->ns_sbh
[1] &&
350 nilfs
->ns_sbh
[1]->b_blocknr
> nilfs
->ns_first_data_block
) {
352 blocknr
= nilfs
->ns_sbh
[1]->b_blocknr
;
353 } else if (nilfs
->ns_sbh
[0]->b_blocknr
> nilfs
->ns_first_data_block
) {
355 blocknr
= nilfs
->ns_sbh
[0]->b_blocknr
;
360 if (sb2i
>= 0 && (u64
)blocknr
<< nilfs
->ns_blocksize_bits
== sb2off
)
361 goto out
; /* super block location is unchanged */
363 /* Get new super block buffer */
364 newblocknr
= sb2off
>> nilfs
->ns_blocksize_bits
;
365 offset
= sb2off
& (nilfs
->ns_blocksize
- 1);
366 nsbh
= sb_getblk(sb
, newblocknr
);
369 "unable to move secondary superblock to block %llu",
370 (unsigned long long)newblocknr
);
374 nsbp
= (void *)nsbh
->b_data
+ offset
;
375 memset(nsbp
, 0, nilfs
->ns_blocksize
);
378 memcpy(nsbp
, nilfs
->ns_sbp
[sb2i
], nilfs
->ns_sbsize
);
379 brelse(nilfs
->ns_sbh
[sb2i
]);
380 nilfs
->ns_sbh
[sb2i
] = nsbh
;
381 nilfs
->ns_sbp
[sb2i
] = nsbp
;
382 } else if (nilfs
->ns_sbh
[0]->b_blocknr
< nilfs
->ns_first_data_block
) {
383 /* secondary super block will be restored to index 1 */
384 nilfs
->ns_sbh
[1] = nsbh
;
385 nilfs
->ns_sbp
[1] = nsbp
;
394 * nilfs_resize_fs - resize the filesystem
395 * @sb: super block instance
396 * @newsize: new size of the filesystem (in bytes)
398 int nilfs_resize_fs(struct super_block
*sb
, __u64 newsize
)
400 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
401 struct nilfs_super_block
**sbp
;
402 __u64 devsize
, newnsegs
;
407 devsize
= bdev_nr_bytes(sb
->s_bdev
);
408 if (newsize
> devsize
)
412 * Write lock is required to protect some functions depending
413 * on the number of segments, the number of reserved segments,
416 down_write(&nilfs
->ns_segctor_sem
);
418 sb2off
= NILFS_SB2_OFFSET_BYTES(newsize
);
419 newnsegs
= sb2off
>> nilfs
->ns_blocksize_bits
;
420 do_div(newnsegs
, nilfs
->ns_blocks_per_segment
);
422 ret
= nilfs_sufile_resize(nilfs
->ns_sufile
, newnsegs
);
423 up_write(&nilfs
->ns_segctor_sem
);
427 ret
= nilfs_construct_segment(sb
);
431 down_write(&nilfs
->ns_sem
);
432 nilfs_move_2nd_super(sb
, sb2off
);
434 sbp
= nilfs_prepare_super(sb
, 0);
436 nilfs_set_log_cursor(sbp
[0], nilfs
);
438 * Drop NILFS_RESIZE_FS flag for compatibility with
439 * mount-time resize which may be implemented in a
442 sbp
[0]->s_state
= cpu_to_le16(le16_to_cpu(sbp
[0]->s_state
) &
444 sbp
[0]->s_dev_size
= cpu_to_le64(newsize
);
445 sbp
[0]->s_nsegments
= cpu_to_le64(nilfs
->ns_nsegments
);
447 memcpy(sbp
[1], sbp
[0], nilfs
->ns_sbsize
);
448 ret
= nilfs_commit_super(sb
, NILFS_SB_COMMIT_ALL
);
450 up_write(&nilfs
->ns_sem
);
453 * Reset the range of allocatable segments last. This order
454 * is important in the case of expansion because the secondary
455 * superblock must be protected from log write until migration
459 nilfs_sufile_set_alloc_range(nilfs
->ns_sufile
, 0, newnsegs
- 1);
464 static void nilfs_put_super(struct super_block
*sb
)
466 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
468 nilfs_detach_log_writer(sb
);
470 if (!sb_rdonly(sb
)) {
471 down_write(&nilfs
->ns_sem
);
472 nilfs_cleanup_super(sb
);
473 up_write(&nilfs
->ns_sem
);
476 iput(nilfs
->ns_sufile
);
477 iput(nilfs
->ns_cpfile
);
480 destroy_nilfs(nilfs
);
481 sb
->s_fs_info
= NULL
;
484 static int nilfs_sync_fs(struct super_block
*sb
, int wait
)
486 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
487 struct nilfs_super_block
**sbp
;
490 /* This function is called when super block should be written back */
492 err
= nilfs_construct_segment(sb
);
494 down_write(&nilfs
->ns_sem
);
495 if (nilfs_sb_dirty(nilfs
)) {
496 sbp
= nilfs_prepare_super(sb
, nilfs_sb_will_flip(nilfs
));
498 nilfs_set_log_cursor(sbp
[0], nilfs
);
499 nilfs_commit_super(sb
, NILFS_SB_COMMIT
);
502 up_write(&nilfs
->ns_sem
);
505 err
= nilfs_flush_device(nilfs
);
510 int nilfs_attach_checkpoint(struct super_block
*sb
, __u64 cno
, int curr_mnt
,
511 struct nilfs_root
**rootp
)
513 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
514 struct nilfs_root
*root
;
515 struct nilfs_checkpoint
*raw_cp
;
516 struct buffer_head
*bh_cp
;
519 root
= nilfs_find_or_create_root(
520 nilfs
, curr_mnt
? NILFS_CPTREE_CURRENT_CNO
: cno
);
525 goto reuse
; /* already attached checkpoint */
527 down_read(&nilfs
->ns_segctor_sem
);
528 err
= nilfs_cpfile_get_checkpoint(nilfs
->ns_cpfile
, cno
, 0, &raw_cp
,
530 up_read(&nilfs
->ns_segctor_sem
);
532 if (err
== -ENOENT
|| err
== -EINVAL
) {
534 "Invalid checkpoint (checkpoint number=%llu)",
535 (unsigned long long)cno
);
541 err
= nilfs_ifile_read(sb
, root
, nilfs
->ns_inode_size
,
542 &raw_cp
->cp_ifile_inode
, &root
->ifile
);
546 atomic64_set(&root
->inodes_count
,
547 le64_to_cpu(raw_cp
->cp_inodes_count
));
548 atomic64_set(&root
->blocks_count
,
549 le64_to_cpu(raw_cp
->cp_blocks_count
));
551 nilfs_cpfile_put_checkpoint(nilfs
->ns_cpfile
, cno
, bh_cp
);
558 nilfs_cpfile_put_checkpoint(nilfs
->ns_cpfile
, cno
, bh_cp
);
560 nilfs_put_root(root
);
565 static int nilfs_freeze(struct super_block
*sb
)
567 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
573 /* Mark super block clean */
574 down_write(&nilfs
->ns_sem
);
575 err
= nilfs_cleanup_super(sb
);
576 up_write(&nilfs
->ns_sem
);
580 static int nilfs_unfreeze(struct super_block
*sb
)
582 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
587 down_write(&nilfs
->ns_sem
);
588 nilfs_setup_super(sb
, false);
589 up_write(&nilfs
->ns_sem
);
593 static int nilfs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
595 struct super_block
*sb
= dentry
->d_sb
;
596 struct nilfs_root
*root
= NILFS_I(d_inode(dentry
))->i_root
;
597 struct the_nilfs
*nilfs
= root
->nilfs
;
598 u64 id
= huge_encode_dev(sb
->s_bdev
->bd_dev
);
599 unsigned long long blocks
;
600 unsigned long overhead
;
601 unsigned long nrsvblocks
;
602 sector_t nfreeblocks
;
603 u64 nmaxinodes
, nfreeinodes
;
607 * Compute all of the segment blocks
609 * The blocks before first segment and after last segment
612 blocks
= nilfs
->ns_blocks_per_segment
* nilfs
->ns_nsegments
613 - nilfs
->ns_first_data_block
;
614 nrsvblocks
= nilfs
->ns_nrsvsegs
* nilfs
->ns_blocks_per_segment
;
617 * Compute the overhead
619 * When distributing meta data blocks outside segment structure,
620 * We must count them as the overhead.
624 err
= nilfs_count_free_blocks(nilfs
, &nfreeblocks
);
628 err
= nilfs_ifile_count_free_inodes(root
->ifile
,
629 &nmaxinodes
, &nfreeinodes
);
631 nilfs_warn(sb
, "failed to count free inodes: err=%d", err
);
632 if (err
== -ERANGE
) {
634 * If nilfs_palloc_count_max_entries() returns
635 * -ERANGE error code then we simply treat
636 * curent inodes count as maximum possible and
637 * zero as free inodes value.
639 nmaxinodes
= atomic64_read(&root
->inodes_count
);
646 buf
->f_type
= NILFS_SUPER_MAGIC
;
647 buf
->f_bsize
= sb
->s_blocksize
;
648 buf
->f_blocks
= blocks
- overhead
;
649 buf
->f_bfree
= nfreeblocks
;
650 buf
->f_bavail
= (buf
->f_bfree
>= nrsvblocks
) ?
651 (buf
->f_bfree
- nrsvblocks
) : 0;
652 buf
->f_files
= nmaxinodes
;
653 buf
->f_ffree
= nfreeinodes
;
654 buf
->f_namelen
= NILFS_NAME_LEN
;
655 buf
->f_fsid
= u64_to_fsid(id
);
660 static int nilfs_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
662 struct super_block
*sb
= dentry
->d_sb
;
663 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
664 struct nilfs_root
*root
= NILFS_I(d_inode(dentry
))->i_root
;
666 if (!nilfs_test_opt(nilfs
, BARRIER
))
667 seq_puts(seq
, ",nobarrier");
668 if (root
->cno
!= NILFS_CPTREE_CURRENT_CNO
)
669 seq_printf(seq
, ",cp=%llu", (unsigned long long)root
->cno
);
670 if (nilfs_test_opt(nilfs
, ERRORS_PANIC
))
671 seq_puts(seq
, ",errors=panic");
672 if (nilfs_test_opt(nilfs
, ERRORS_CONT
))
673 seq_puts(seq
, ",errors=continue");
674 if (nilfs_test_opt(nilfs
, STRICT_ORDER
))
675 seq_puts(seq
, ",order=strict");
676 if (nilfs_test_opt(nilfs
, NORECOVERY
))
677 seq_puts(seq
, ",norecovery");
678 if (nilfs_test_opt(nilfs
, DISCARD
))
679 seq_puts(seq
, ",discard");
684 static const struct super_operations nilfs_sops
= {
685 .alloc_inode
= nilfs_alloc_inode
,
686 .free_inode
= nilfs_free_inode
,
687 .dirty_inode
= nilfs_dirty_inode
,
688 .evict_inode
= nilfs_evict_inode
,
689 .put_super
= nilfs_put_super
,
690 .sync_fs
= nilfs_sync_fs
,
691 .freeze_fs
= nilfs_freeze
,
692 .unfreeze_fs
= nilfs_unfreeze
,
693 .statfs
= nilfs_statfs
,
694 .remount_fs
= nilfs_remount
,
695 .show_options
= nilfs_show_options
699 Opt_err_cont
, Opt_err_panic
, Opt_err_ro
,
700 Opt_barrier
, Opt_nobarrier
, Opt_snapshot
, Opt_order
, Opt_norecovery
,
701 Opt_discard
, Opt_nodiscard
, Opt_err
,
704 static match_table_t tokens
= {
705 {Opt_err_cont
, "errors=continue"},
706 {Opt_err_panic
, "errors=panic"},
707 {Opt_err_ro
, "errors=remount-ro"},
708 {Opt_barrier
, "barrier"},
709 {Opt_nobarrier
, "nobarrier"},
710 {Opt_snapshot
, "cp=%u"},
711 {Opt_order
, "order=%s"},
712 {Opt_norecovery
, "norecovery"},
713 {Opt_discard
, "discard"},
714 {Opt_nodiscard
, "nodiscard"},
718 static int parse_options(char *options
, struct super_block
*sb
, int is_remount
)
720 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
722 substring_t args
[MAX_OPT_ARGS
];
727 while ((p
= strsep(&options
, ",")) != NULL
) {
733 token
= match_token(p
, tokens
, args
);
736 nilfs_set_opt(nilfs
, BARRIER
);
739 nilfs_clear_opt(nilfs
, BARRIER
);
742 if (strcmp(args
[0].from
, "relaxed") == 0)
743 /* Ordered data semantics */
744 nilfs_clear_opt(nilfs
, STRICT_ORDER
);
745 else if (strcmp(args
[0].from
, "strict") == 0)
746 /* Strict in-order semantics */
747 nilfs_set_opt(nilfs
, STRICT_ORDER
);
752 nilfs_write_opt(nilfs
, ERROR_MODE
, ERRORS_PANIC
);
755 nilfs_write_opt(nilfs
, ERROR_MODE
, ERRORS_RO
);
758 nilfs_write_opt(nilfs
, ERROR_MODE
, ERRORS_CONT
);
763 "\"%s\" option is invalid for remount",
769 nilfs_set_opt(nilfs
, NORECOVERY
);
772 nilfs_set_opt(nilfs
, DISCARD
);
775 nilfs_clear_opt(nilfs
, DISCARD
);
778 nilfs_err(sb
, "unrecognized mount option \"%s\"", p
);
786 nilfs_set_default_options(struct super_block
*sb
,
787 struct nilfs_super_block
*sbp
)
789 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
791 nilfs
->ns_mount_opt
=
792 NILFS_MOUNT_ERRORS_RO
| NILFS_MOUNT_BARRIER
;
795 static int nilfs_setup_super(struct super_block
*sb
, int is_mount
)
797 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
798 struct nilfs_super_block
**sbp
;
802 /* nilfs->ns_sem must be locked by the caller. */
803 sbp
= nilfs_prepare_super(sb
, 0);
808 goto skip_mount_setup
;
810 max_mnt_count
= le16_to_cpu(sbp
[0]->s_max_mnt_count
);
811 mnt_count
= le16_to_cpu(sbp
[0]->s_mnt_count
);
813 if (nilfs
->ns_mount_state
& NILFS_ERROR_FS
) {
814 nilfs_warn(sb
, "mounting fs with errors");
816 } else if (max_mnt_count
>= 0 && mnt_count
>= max_mnt_count
) {
817 nilfs_warn(sb
, "maximal mount count reached");
821 sbp
[0]->s_max_mnt_count
= cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT
);
823 sbp
[0]->s_mnt_count
= cpu_to_le16(mnt_count
+ 1);
824 sbp
[0]->s_mtime
= cpu_to_le64(ktime_get_real_seconds());
828 cpu_to_le16(le16_to_cpu(sbp
[0]->s_state
) & ~NILFS_VALID_FS
);
829 /* synchronize sbp[1] with sbp[0] */
831 memcpy(sbp
[1], sbp
[0], nilfs
->ns_sbsize
);
832 return nilfs_commit_super(sb
, NILFS_SB_COMMIT_ALL
);
835 struct nilfs_super_block
*nilfs_read_super_block(struct super_block
*sb
,
836 u64 pos
, int blocksize
,
837 struct buffer_head
**pbh
)
839 unsigned long long sb_index
= pos
;
840 unsigned long offset
;
842 offset
= do_div(sb_index
, blocksize
);
843 *pbh
= sb_bread(sb
, sb_index
);
846 return (struct nilfs_super_block
*)((char *)(*pbh
)->b_data
+ offset
);
849 int nilfs_store_magic_and_option(struct super_block
*sb
,
850 struct nilfs_super_block
*sbp
,
853 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
855 sb
->s_magic
= le16_to_cpu(sbp
->s_magic
);
857 /* FS independent flags */
858 #ifdef NILFS_ATIME_DISABLE
859 sb
->s_flags
|= SB_NOATIME
;
862 nilfs_set_default_options(sb
, sbp
);
864 nilfs
->ns_resuid
= le16_to_cpu(sbp
->s_def_resuid
);
865 nilfs
->ns_resgid
= le16_to_cpu(sbp
->s_def_resgid
);
866 nilfs
->ns_interval
= le32_to_cpu(sbp
->s_c_interval
);
867 nilfs
->ns_watermark
= le32_to_cpu(sbp
->s_c_block_max
);
869 return !parse_options(data
, sb
, 0) ? -EINVAL
: 0;
872 int nilfs_check_feature_compatibility(struct super_block
*sb
,
873 struct nilfs_super_block
*sbp
)
877 features
= le64_to_cpu(sbp
->s_feature_incompat
) &
878 ~NILFS_FEATURE_INCOMPAT_SUPP
;
881 "couldn't mount because of unsupported optional features (%llx)",
882 (unsigned long long)features
);
885 features
= le64_to_cpu(sbp
->s_feature_compat_ro
) &
886 ~NILFS_FEATURE_COMPAT_RO_SUPP
;
887 if (!sb_rdonly(sb
) && features
) {
889 "couldn't mount RDWR because of unsupported optional features (%llx)",
890 (unsigned long long)features
);
896 static int nilfs_get_root_dentry(struct super_block
*sb
,
897 struct nilfs_root
*root
,
898 struct dentry
**root_dentry
)
901 struct dentry
*dentry
;
904 inode
= nilfs_iget(sb
, root
, NILFS_ROOT_INO
);
906 ret
= PTR_ERR(inode
);
907 nilfs_err(sb
, "error %d getting root inode", ret
);
910 if (!S_ISDIR(inode
->i_mode
) || !inode
->i_blocks
|| !inode
->i_size
) {
912 nilfs_err(sb
, "corrupt root inode");
917 if (root
->cno
== NILFS_CPTREE_CURRENT_CNO
) {
918 dentry
= d_find_alias(inode
);
920 dentry
= d_make_root(inode
);
929 dentry
= d_obtain_root(inode
);
930 if (IS_ERR(dentry
)) {
931 ret
= PTR_ERR(dentry
);
935 *root_dentry
= dentry
;
940 nilfs_err(sb
, "error %d getting root dentry", ret
);
944 static int nilfs_attach_snapshot(struct super_block
*s
, __u64 cno
,
945 struct dentry
**root_dentry
)
947 struct the_nilfs
*nilfs
= s
->s_fs_info
;
948 struct nilfs_root
*root
;
951 mutex_lock(&nilfs
->ns_snapshot_mount_mutex
);
953 down_read(&nilfs
->ns_segctor_sem
);
954 ret
= nilfs_cpfile_is_snapshot(nilfs
->ns_cpfile
, cno
);
955 up_read(&nilfs
->ns_segctor_sem
);
957 ret
= (ret
== -ENOENT
) ? -EINVAL
: ret
;
961 "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
962 (unsigned long long)cno
);
967 ret
= nilfs_attach_checkpoint(s
, cno
, false, &root
);
970 "error %d while loading snapshot (checkpoint number=%llu)",
971 ret
, (unsigned long long)cno
);
974 ret
= nilfs_get_root_dentry(s
, root
, root_dentry
);
975 nilfs_put_root(root
);
977 mutex_unlock(&nilfs
->ns_snapshot_mount_mutex
);
982 * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
983 * @root_dentry: root dentry of the tree to be shrunk
985 * This function returns true if the tree was in-use.
987 static bool nilfs_tree_is_busy(struct dentry
*root_dentry
)
989 shrink_dcache_parent(root_dentry
);
990 return d_count(root_dentry
) > 1;
993 int nilfs_checkpoint_is_mounted(struct super_block
*sb
, __u64 cno
)
995 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
996 struct nilfs_root
*root
;
998 struct dentry
*dentry
;
1001 if (cno
> nilfs
->ns_cno
)
1004 if (cno
>= nilfs_last_cno(nilfs
))
1005 return true; /* protect recent checkpoints */
1008 root
= nilfs_lookup_root(nilfs
, cno
);
1010 inode
= nilfs_ilookup(sb
, root
, NILFS_ROOT_INO
);
1012 dentry
= d_find_alias(inode
);
1014 ret
= nilfs_tree_is_busy(dentry
);
1019 nilfs_put_root(root
);
1025 * nilfs_fill_super() - initialize a super block instance
1027 * @data: mount options
1028 * @silent: silent mode flag
1030 * This function is called exclusively by nilfs->ns_mount_mutex.
1031 * So, the recovery process is protected from other simultaneous mounts.
1034 nilfs_fill_super(struct super_block
*sb
, void *data
, int silent
)
1036 struct the_nilfs
*nilfs
;
1037 struct nilfs_root
*fsroot
;
1041 nilfs
= alloc_nilfs(sb
);
1045 sb
->s_fs_info
= nilfs
;
1047 err
= init_nilfs(nilfs
, sb
, (char *)data
);
1051 sb
->s_op
= &nilfs_sops
;
1052 sb
->s_export_op
= &nilfs_export_ops
;
1054 sb
->s_time_gran
= 1;
1055 sb
->s_max_links
= NILFS_LINK_MAX
;
1057 sb
->s_bdi
= bdi_get(sb
->s_bdev
->bd_disk
->bdi
);
1059 err
= load_nilfs(nilfs
, sb
);
1063 cno
= nilfs_last_cno(nilfs
);
1064 err
= nilfs_attach_checkpoint(sb
, cno
, true, &fsroot
);
1067 "error %d while loading last checkpoint (checkpoint number=%llu)",
1068 err
, (unsigned long long)cno
);
1072 if (!sb_rdonly(sb
)) {
1073 err
= nilfs_attach_log_writer(sb
, fsroot
);
1075 goto failed_checkpoint
;
1078 err
= nilfs_get_root_dentry(sb
, fsroot
, &sb
->s_root
);
1080 goto failed_segctor
;
1082 nilfs_put_root(fsroot
);
1084 if (!sb_rdonly(sb
)) {
1085 down_write(&nilfs
->ns_sem
);
1086 nilfs_setup_super(sb
, true);
1087 up_write(&nilfs
->ns_sem
);
1093 nilfs_detach_log_writer(sb
);
1096 nilfs_put_root(fsroot
);
1099 iput(nilfs
->ns_sufile
);
1100 iput(nilfs
->ns_cpfile
);
1101 iput(nilfs
->ns_dat
);
1104 destroy_nilfs(nilfs
);
1108 static int nilfs_remount(struct super_block
*sb
, int *flags
, char *data
)
1110 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
1111 unsigned long old_sb_flags
;
1112 unsigned long old_mount_opt
;
1115 sync_filesystem(sb
);
1116 old_sb_flags
= sb
->s_flags
;
1117 old_mount_opt
= nilfs
->ns_mount_opt
;
1119 if (!parse_options(data
, sb
, 1)) {
1123 sb
->s_flags
= (sb
->s_flags
& ~SB_POSIXACL
);
1127 if (!nilfs_valid_fs(nilfs
)) {
1129 "couldn't remount because the filesystem is in an incomplete recovery state");
1133 if ((bool)(*flags
& SB_RDONLY
) == sb_rdonly(sb
))
1135 if (*flags
& SB_RDONLY
) {
1136 /* Shutting down log writer */
1137 nilfs_detach_log_writer(sb
);
1138 sb
->s_flags
|= SB_RDONLY
;
1141 * Remounting a valid RW partition RDONLY, so set
1142 * the RDONLY flag and then mark the partition as valid again.
1144 down_write(&nilfs
->ns_sem
);
1145 nilfs_cleanup_super(sb
);
1146 up_write(&nilfs
->ns_sem
);
1149 struct nilfs_root
*root
;
1152 * Mounting a RDONLY partition read-write, so reread and
1153 * store the current valid flag. (It may have been changed
1154 * by fsck since we originally mounted the partition.)
1156 down_read(&nilfs
->ns_sem
);
1157 features
= le64_to_cpu(nilfs
->ns_sbp
[0]->s_feature_compat_ro
) &
1158 ~NILFS_FEATURE_COMPAT_RO_SUPP
;
1159 up_read(&nilfs
->ns_sem
);
1162 "couldn't remount RDWR because of unsupported optional features (%llx)",
1163 (unsigned long long)features
);
1168 sb
->s_flags
&= ~SB_RDONLY
;
1170 root
= NILFS_I(d_inode(sb
->s_root
))->i_root
;
1171 err
= nilfs_attach_log_writer(sb
, root
);
1175 down_write(&nilfs
->ns_sem
);
1176 nilfs_setup_super(sb
, true);
1177 up_write(&nilfs
->ns_sem
);
1183 sb
->s_flags
= old_sb_flags
;
1184 nilfs
->ns_mount_opt
= old_mount_opt
;
1188 struct nilfs_super_data
{
1189 struct block_device
*bdev
;
1194 static int nilfs_parse_snapshot_option(const char *option
,
1195 const substring_t
*arg
,
1196 struct nilfs_super_data
*sd
)
1198 unsigned long long val
;
1199 const char *msg
= NULL
;
1202 if (!(sd
->flags
& SB_RDONLY
)) {
1203 msg
= "read-only option is not specified";
1207 err
= kstrtoull(arg
->from
, 0, &val
);
1210 msg
= "too large checkpoint number";
1212 msg
= "malformed argument";
1214 } else if (val
== 0) {
1215 msg
= "invalid checkpoint number 0";
1222 nilfs_err(NULL
, "invalid option \"%s\": %s", option
, msg
);
1227 * nilfs_identify - pre-read mount options needed to identify mount instance
1228 * @data: mount options
1229 * @sd: nilfs_super_data
1231 static int nilfs_identify(char *data
, struct nilfs_super_data
*sd
)
1233 char *p
, *options
= data
;
1234 substring_t args
[MAX_OPT_ARGS
];
1239 p
= strsep(&options
, ",");
1240 if (p
!= NULL
&& *p
) {
1241 token
= match_token(p
, tokens
, args
);
1242 if (token
== Opt_snapshot
)
1243 ret
= nilfs_parse_snapshot_option(p
, &args
[0],
1248 BUG_ON(options
== data
);
1249 *(options
- 1) = ',';
1254 static int nilfs_set_bdev_super(struct super_block
*s
, void *data
)
1257 s
->s_dev
= s
->s_bdev
->bd_dev
;
1261 static int nilfs_test_bdev_super(struct super_block
*s
, void *data
)
1263 return (void *)s
->s_bdev
== data
;
1266 static struct dentry
*
1267 nilfs_mount(struct file_system_type
*fs_type
, int flags
,
1268 const char *dev_name
, void *data
)
1270 struct nilfs_super_data sd
;
1271 struct super_block
*s
;
1272 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1273 struct dentry
*root_dentry
;
1274 int err
, s_new
= false;
1276 if (!(flags
& SB_RDONLY
))
1277 mode
|= FMODE_WRITE
;
1279 sd
.bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1280 if (IS_ERR(sd
.bdev
))
1281 return ERR_CAST(sd
.bdev
);
1285 if (nilfs_identify((char *)data
, &sd
)) {
1291 * once the super is inserted into the list by sget, s_umount
1292 * will protect the lockfs code from trying to start a snapshot
1293 * while we are mounting
1295 mutex_lock(&sd
.bdev
->bd_fsfreeze_mutex
);
1296 if (sd
.bdev
->bd_fsfreeze_count
> 0) {
1297 mutex_unlock(&sd
.bdev
->bd_fsfreeze_mutex
);
1301 s
= sget(fs_type
, nilfs_test_bdev_super
, nilfs_set_bdev_super
, flags
,
1303 mutex_unlock(&sd
.bdev
->bd_fsfreeze_mutex
);
1312 /* New superblock instance created */
1314 snprintf(s
->s_id
, sizeof(s
->s_id
), "%pg", sd
.bdev
);
1315 sb_set_blocksize(s
, block_size(sd
.bdev
));
1317 err
= nilfs_fill_super(s
, data
, flags
& SB_SILENT
? 1 : 0);
1321 s
->s_flags
|= SB_ACTIVE
;
1322 } else if (!sd
.cno
) {
1323 if (nilfs_tree_is_busy(s
->s_root
)) {
1324 if ((flags
^ s
->s_flags
) & SB_RDONLY
) {
1326 "the device already has a %s mount.",
1327 sb_rdonly(s
) ? "read-only" : "read/write");
1333 * Try remount to setup mount states if the current
1334 * tree is not mounted and only snapshots use this sb.
1336 err
= nilfs_remount(s
, &flags
, data
);
1343 err
= nilfs_attach_snapshot(s
, sd
.cno
, &root_dentry
);
1347 root_dentry
= dget(s
->s_root
);
1351 blkdev_put(sd
.bdev
, mode
);
1356 deactivate_locked_super(s
);
1360 blkdev_put(sd
.bdev
, mode
);
1361 return ERR_PTR(err
);
1364 struct file_system_type nilfs_fs_type
= {
1365 .owner
= THIS_MODULE
,
1367 .mount
= nilfs_mount
,
1368 .kill_sb
= kill_block_super
,
1369 .fs_flags
= FS_REQUIRES_DEV
,
1371 MODULE_ALIAS_FS("nilfs2");
1373 static void nilfs_inode_init_once(void *obj
)
1375 struct nilfs_inode_info
*ii
= obj
;
1377 INIT_LIST_HEAD(&ii
->i_dirty
);
1378 #ifdef CONFIG_NILFS_XATTR
1379 init_rwsem(&ii
->xattr_sem
);
1381 inode_init_once(&ii
->vfs_inode
);
1384 static void nilfs_segbuf_init_once(void *obj
)
1386 memset(obj
, 0, sizeof(struct nilfs_segment_buffer
));
1389 static void nilfs_destroy_cachep(void)
1392 * Make sure all delayed rcu free inodes are flushed before we
1397 kmem_cache_destroy(nilfs_inode_cachep
);
1398 kmem_cache_destroy(nilfs_transaction_cachep
);
1399 kmem_cache_destroy(nilfs_segbuf_cachep
);
1400 kmem_cache_destroy(nilfs_btree_path_cache
);
1403 static int __init
nilfs_init_cachep(void)
1405 nilfs_inode_cachep
= kmem_cache_create("nilfs2_inode_cache",
1406 sizeof(struct nilfs_inode_info
), 0,
1407 SLAB_RECLAIM_ACCOUNT
|SLAB_ACCOUNT
,
1408 nilfs_inode_init_once
);
1409 if (!nilfs_inode_cachep
)
1412 nilfs_transaction_cachep
= kmem_cache_create("nilfs2_transaction_cache",
1413 sizeof(struct nilfs_transaction_info
), 0,
1414 SLAB_RECLAIM_ACCOUNT
, NULL
);
1415 if (!nilfs_transaction_cachep
)
1418 nilfs_segbuf_cachep
= kmem_cache_create("nilfs2_segbuf_cache",
1419 sizeof(struct nilfs_segment_buffer
), 0,
1420 SLAB_RECLAIM_ACCOUNT
, nilfs_segbuf_init_once
);
1421 if (!nilfs_segbuf_cachep
)
1424 nilfs_btree_path_cache
= kmem_cache_create("nilfs2_btree_path_cache",
1425 sizeof(struct nilfs_btree_path
) * NILFS_BTREE_LEVEL_MAX
,
1427 if (!nilfs_btree_path_cache
)
1433 nilfs_destroy_cachep();
1437 static int __init
init_nilfs_fs(void)
1441 err
= nilfs_init_cachep();
1445 err
= nilfs_sysfs_init();
1449 err
= register_filesystem(&nilfs_fs_type
);
1451 goto deinit_sysfs_entry
;
1453 printk(KERN_INFO
"NILFS version 2 loaded\n");
1459 nilfs_destroy_cachep();
1464 static void __exit
exit_nilfs_fs(void)
1466 nilfs_destroy_cachep();
1468 unregister_filesystem(&nilfs_fs_type
);
1471 module_init(init_nilfs_fs
)
1472 module_exit(exit_nilfs_fs
)