1 // SPDX-License-Identifier: GPL-2.0+
3 * super.c - 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
= kmem_cache_alloc(nilfs_inode_cachep
, GFP_NOFS
);
160 nilfs_mapping_init(&ii
->i_btnode_cache
, &ii
->vfs_inode
);
161 return &ii
->vfs_inode
;
164 static void nilfs_free_inode(struct inode
*inode
)
166 if (nilfs_is_metadata_file_inode(inode
))
167 nilfs_mdt_destroy(inode
);
169 kmem_cache_free(nilfs_inode_cachep
, NILFS_I(inode
));
172 static int nilfs_sync_super(struct super_block
*sb
, int flag
)
174 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
178 set_buffer_dirty(nilfs
->ns_sbh
[0]);
179 if (nilfs_test_opt(nilfs
, BARRIER
)) {
180 err
= __sync_dirty_buffer(nilfs
->ns_sbh
[0],
181 REQ_SYNC
| REQ_PREFLUSH
| REQ_FUA
);
183 err
= sync_dirty_buffer(nilfs
->ns_sbh
[0]);
187 nilfs_err(sb
, "unable to write superblock: err=%d", err
);
188 if (err
== -EIO
&& nilfs
->ns_sbh
[1]) {
190 * sbp[0] points to newer log than sbp[1],
191 * so copy sbp[0] to sbp[1] to take over sbp[0].
193 memcpy(nilfs
->ns_sbp
[1], nilfs
->ns_sbp
[0],
195 nilfs_fall_back_super_block(nilfs
);
199 struct nilfs_super_block
*sbp
= nilfs
->ns_sbp
[0];
201 nilfs
->ns_sbwcount
++;
204 * The latest segment becomes trailable from the position
205 * written in superblock.
207 clear_nilfs_discontinued(nilfs
);
209 /* update GC protection for recent segments */
210 if (nilfs
->ns_sbh
[1]) {
211 if (flag
== NILFS_SB_COMMIT_ALL
) {
212 set_buffer_dirty(nilfs
->ns_sbh
[1]);
213 if (sync_dirty_buffer(nilfs
->ns_sbh
[1]) < 0)
216 if (le64_to_cpu(nilfs
->ns_sbp
[1]->s_last_cno
) <
217 le64_to_cpu(nilfs
->ns_sbp
[0]->s_last_cno
))
218 sbp
= nilfs
->ns_sbp
[1];
221 spin_lock(&nilfs
->ns_last_segment_lock
);
222 nilfs
->ns_prot_seq
= le64_to_cpu(sbp
->s_last_seq
);
223 spin_unlock(&nilfs
->ns_last_segment_lock
);
229 void nilfs_set_log_cursor(struct nilfs_super_block
*sbp
,
230 struct the_nilfs
*nilfs
)
232 sector_t nfreeblocks
;
234 /* nilfs->ns_sem must be locked by the caller. */
235 nilfs_count_free_blocks(nilfs
, &nfreeblocks
);
236 sbp
->s_free_blocks_count
= cpu_to_le64(nfreeblocks
);
238 spin_lock(&nilfs
->ns_last_segment_lock
);
239 sbp
->s_last_seq
= cpu_to_le64(nilfs
->ns_last_seq
);
240 sbp
->s_last_pseg
= cpu_to_le64(nilfs
->ns_last_pseg
);
241 sbp
->s_last_cno
= cpu_to_le64(nilfs
->ns_last_cno
);
242 spin_unlock(&nilfs
->ns_last_segment_lock
);
245 struct nilfs_super_block
**nilfs_prepare_super(struct super_block
*sb
,
248 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
249 struct nilfs_super_block
**sbp
= nilfs
->ns_sbp
;
251 /* nilfs->ns_sem must be locked by the caller. */
252 if (sbp
[0]->s_magic
!= cpu_to_le16(NILFS_SUPER_MAGIC
)) {
254 sbp
[1]->s_magic
== cpu_to_le16(NILFS_SUPER_MAGIC
)) {
255 memcpy(sbp
[0], sbp
[1], nilfs
->ns_sbsize
);
257 nilfs_crit(sb
, "superblock broke");
261 sbp
[1]->s_magic
!= cpu_to_le16(NILFS_SUPER_MAGIC
)) {
262 memcpy(sbp
[1], sbp
[0], nilfs
->ns_sbsize
);
266 nilfs_swap_super_block(nilfs
);
271 int nilfs_commit_super(struct super_block
*sb
, int flag
)
273 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
274 struct nilfs_super_block
**sbp
= nilfs
->ns_sbp
;
277 /* nilfs->ns_sem must be locked by the caller. */
278 t
= ktime_get_real_seconds();
279 nilfs
->ns_sbwtime
= t
;
280 sbp
[0]->s_wtime
= cpu_to_le64(t
);
282 sbp
[0]->s_sum
= cpu_to_le32(crc32_le(nilfs
->ns_crc_seed
,
283 (unsigned char *)sbp
[0],
285 if (flag
== NILFS_SB_COMMIT_ALL
&& sbp
[1]) {
286 sbp
[1]->s_wtime
= sbp
[0]->s_wtime
;
288 sbp
[1]->s_sum
= cpu_to_le32(crc32_le(nilfs
->ns_crc_seed
,
289 (unsigned char *)sbp
[1],
292 clear_nilfs_sb_dirty(nilfs
);
293 nilfs
->ns_flushed_device
= 1;
294 /* make sure store to ns_flushed_device cannot be reordered */
296 return nilfs_sync_super(sb
, flag
);
300 * nilfs_cleanup_super() - write filesystem state for cleanup
301 * @sb: super block instance to be unmounted or degraded to read-only
303 * This function restores state flags in the on-disk super block.
304 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
305 * filesystem was not clean previously.
307 int nilfs_cleanup_super(struct super_block
*sb
)
309 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
310 struct nilfs_super_block
**sbp
;
311 int flag
= NILFS_SB_COMMIT
;
314 sbp
= nilfs_prepare_super(sb
, 0);
316 sbp
[0]->s_state
= cpu_to_le16(nilfs
->ns_mount_state
);
317 nilfs_set_log_cursor(sbp
[0], nilfs
);
318 if (sbp
[1] && sbp
[0]->s_last_cno
== sbp
[1]->s_last_cno
) {
320 * make the "clean" flag also to the opposite
321 * super block if both super blocks point to
322 * the same checkpoint.
324 sbp
[1]->s_state
= sbp
[0]->s_state
;
325 flag
= NILFS_SB_COMMIT_ALL
;
327 ret
= nilfs_commit_super(sb
, flag
);
333 * nilfs_move_2nd_super - relocate secondary super block
334 * @sb: super block instance
335 * @sb2off: new offset of the secondary super block (in bytes)
337 static int nilfs_move_2nd_super(struct super_block
*sb
, loff_t sb2off
)
339 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
340 struct buffer_head
*nsbh
;
341 struct nilfs_super_block
*nsbp
;
342 sector_t blocknr
, newblocknr
;
343 unsigned long offset
;
344 int sb2i
; /* array index of the secondary superblock */
347 /* nilfs->ns_sem must be locked by the caller. */
348 if (nilfs
->ns_sbh
[1] &&
349 nilfs
->ns_sbh
[1]->b_blocknr
> nilfs
->ns_first_data_block
) {
351 blocknr
= nilfs
->ns_sbh
[1]->b_blocknr
;
352 } else if (nilfs
->ns_sbh
[0]->b_blocknr
> nilfs
->ns_first_data_block
) {
354 blocknr
= nilfs
->ns_sbh
[0]->b_blocknr
;
359 if (sb2i
>= 0 && (u64
)blocknr
<< nilfs
->ns_blocksize_bits
== sb2off
)
360 goto out
; /* super block location is unchanged */
362 /* Get new super block buffer */
363 newblocknr
= sb2off
>> nilfs
->ns_blocksize_bits
;
364 offset
= sb2off
& (nilfs
->ns_blocksize
- 1);
365 nsbh
= sb_getblk(sb
, newblocknr
);
368 "unable to move secondary superblock to block %llu",
369 (unsigned long long)newblocknr
);
373 nsbp
= (void *)nsbh
->b_data
+ offset
;
374 memset(nsbp
, 0, nilfs
->ns_blocksize
);
377 memcpy(nsbp
, nilfs
->ns_sbp
[sb2i
], nilfs
->ns_sbsize
);
378 brelse(nilfs
->ns_sbh
[sb2i
]);
379 nilfs
->ns_sbh
[sb2i
] = nsbh
;
380 nilfs
->ns_sbp
[sb2i
] = nsbp
;
381 } else if (nilfs
->ns_sbh
[0]->b_blocknr
< nilfs
->ns_first_data_block
) {
382 /* secondary super block will be restored to index 1 */
383 nilfs
->ns_sbh
[1] = nsbh
;
384 nilfs
->ns_sbp
[1] = nsbp
;
393 * nilfs_resize_fs - resize the filesystem
394 * @sb: super block instance
395 * @newsize: new size of the filesystem (in bytes)
397 int nilfs_resize_fs(struct super_block
*sb
, __u64 newsize
)
399 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
400 struct nilfs_super_block
**sbp
;
401 __u64 devsize
, newnsegs
;
406 devsize
= i_size_read(sb
->s_bdev
->bd_inode
);
407 if (newsize
> devsize
)
411 * Write lock is required to protect some functions depending
412 * on the number of segments, the number of reserved segments,
415 down_write(&nilfs
->ns_segctor_sem
);
417 sb2off
= NILFS_SB2_OFFSET_BYTES(newsize
);
418 newnsegs
= sb2off
>> nilfs
->ns_blocksize_bits
;
419 do_div(newnsegs
, nilfs
->ns_blocks_per_segment
);
421 ret
= nilfs_sufile_resize(nilfs
->ns_sufile
, newnsegs
);
422 up_write(&nilfs
->ns_segctor_sem
);
426 ret
= nilfs_construct_segment(sb
);
430 down_write(&nilfs
->ns_sem
);
431 nilfs_move_2nd_super(sb
, sb2off
);
433 sbp
= nilfs_prepare_super(sb
, 0);
435 nilfs_set_log_cursor(sbp
[0], nilfs
);
437 * Drop NILFS_RESIZE_FS flag for compatibility with
438 * mount-time resize which may be implemented in a
441 sbp
[0]->s_state
= cpu_to_le16(le16_to_cpu(sbp
[0]->s_state
) &
443 sbp
[0]->s_dev_size
= cpu_to_le64(newsize
);
444 sbp
[0]->s_nsegments
= cpu_to_le64(nilfs
->ns_nsegments
);
446 memcpy(sbp
[1], sbp
[0], nilfs
->ns_sbsize
);
447 ret
= nilfs_commit_super(sb
, NILFS_SB_COMMIT_ALL
);
449 up_write(&nilfs
->ns_sem
);
452 * Reset the range of allocatable segments last. This order
453 * is important in the case of expansion because the secondary
454 * superblock must be protected from log write until migration
458 nilfs_sufile_set_alloc_range(nilfs
->ns_sufile
, 0, newnsegs
- 1);
463 static void nilfs_put_super(struct super_block
*sb
)
465 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
467 nilfs_detach_log_writer(sb
);
469 if (!sb_rdonly(sb
)) {
470 down_write(&nilfs
->ns_sem
);
471 nilfs_cleanup_super(sb
);
472 up_write(&nilfs
->ns_sem
);
475 iput(nilfs
->ns_sufile
);
476 iput(nilfs
->ns_cpfile
);
479 destroy_nilfs(nilfs
);
480 sb
->s_fs_info
= NULL
;
483 static int nilfs_sync_fs(struct super_block
*sb
, int wait
)
485 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
486 struct nilfs_super_block
**sbp
;
489 /* This function is called when super block should be written back */
491 err
= nilfs_construct_segment(sb
);
493 down_write(&nilfs
->ns_sem
);
494 if (nilfs_sb_dirty(nilfs
)) {
495 sbp
= nilfs_prepare_super(sb
, nilfs_sb_will_flip(nilfs
));
497 nilfs_set_log_cursor(sbp
[0], nilfs
);
498 nilfs_commit_super(sb
, NILFS_SB_COMMIT
);
501 up_write(&nilfs
->ns_sem
);
504 err
= nilfs_flush_device(nilfs
);
509 int nilfs_attach_checkpoint(struct super_block
*sb
, __u64 cno
, int curr_mnt
,
510 struct nilfs_root
**rootp
)
512 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
513 struct nilfs_root
*root
;
514 struct nilfs_checkpoint
*raw_cp
;
515 struct buffer_head
*bh_cp
;
518 root
= nilfs_find_or_create_root(
519 nilfs
, curr_mnt
? NILFS_CPTREE_CURRENT_CNO
: cno
);
524 goto reuse
; /* already attached checkpoint */
526 down_read(&nilfs
->ns_segctor_sem
);
527 err
= nilfs_cpfile_get_checkpoint(nilfs
->ns_cpfile
, cno
, 0, &raw_cp
,
529 up_read(&nilfs
->ns_segctor_sem
);
531 if (err
== -ENOENT
|| err
== -EINVAL
) {
533 "Invalid checkpoint (checkpoint number=%llu)",
534 (unsigned long long)cno
);
540 err
= nilfs_ifile_read(sb
, root
, nilfs
->ns_inode_size
,
541 &raw_cp
->cp_ifile_inode
, &root
->ifile
);
545 atomic64_set(&root
->inodes_count
,
546 le64_to_cpu(raw_cp
->cp_inodes_count
));
547 atomic64_set(&root
->blocks_count
,
548 le64_to_cpu(raw_cp
->cp_blocks_count
));
550 nilfs_cpfile_put_checkpoint(nilfs
->ns_cpfile
, cno
, bh_cp
);
557 nilfs_cpfile_put_checkpoint(nilfs
->ns_cpfile
, cno
, bh_cp
);
559 nilfs_put_root(root
);
564 static int nilfs_freeze(struct super_block
*sb
)
566 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
572 /* Mark super block clean */
573 down_write(&nilfs
->ns_sem
);
574 err
= nilfs_cleanup_super(sb
);
575 up_write(&nilfs
->ns_sem
);
579 static int nilfs_unfreeze(struct super_block
*sb
)
581 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
586 down_write(&nilfs
->ns_sem
);
587 nilfs_setup_super(sb
, false);
588 up_write(&nilfs
->ns_sem
);
592 static int nilfs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
594 struct super_block
*sb
= dentry
->d_sb
;
595 struct nilfs_root
*root
= NILFS_I(d_inode(dentry
))->i_root
;
596 struct the_nilfs
*nilfs
= root
->nilfs
;
597 u64 id
= huge_encode_dev(sb
->s_bdev
->bd_dev
);
598 unsigned long long blocks
;
599 unsigned long overhead
;
600 unsigned long nrsvblocks
;
601 sector_t nfreeblocks
;
602 u64 nmaxinodes
, nfreeinodes
;
606 * Compute all of the segment blocks
608 * The blocks before first segment and after last segment
611 blocks
= nilfs
->ns_blocks_per_segment
* nilfs
->ns_nsegments
612 - nilfs
->ns_first_data_block
;
613 nrsvblocks
= nilfs
->ns_nrsvsegs
* nilfs
->ns_blocks_per_segment
;
616 * Compute the overhead
618 * When distributing meta data blocks outside segment structure,
619 * We must count them as the overhead.
623 err
= nilfs_count_free_blocks(nilfs
, &nfreeblocks
);
627 err
= nilfs_ifile_count_free_inodes(root
->ifile
,
628 &nmaxinodes
, &nfreeinodes
);
630 nilfs_warn(sb
, "failed to count free inodes: err=%d", err
);
631 if (err
== -ERANGE
) {
633 * If nilfs_palloc_count_max_entries() returns
634 * -ERANGE error code then we simply treat
635 * curent inodes count as maximum possible and
636 * zero as free inodes value.
638 nmaxinodes
= atomic64_read(&root
->inodes_count
);
645 buf
->f_type
= NILFS_SUPER_MAGIC
;
646 buf
->f_bsize
= sb
->s_blocksize
;
647 buf
->f_blocks
= blocks
- overhead
;
648 buf
->f_bfree
= nfreeblocks
;
649 buf
->f_bavail
= (buf
->f_bfree
>= nrsvblocks
) ?
650 (buf
->f_bfree
- nrsvblocks
) : 0;
651 buf
->f_files
= nmaxinodes
;
652 buf
->f_ffree
= nfreeinodes
;
653 buf
->f_namelen
= NILFS_NAME_LEN
;
654 buf
->f_fsid
= u64_to_fsid(id
);
659 static int nilfs_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
661 struct super_block
*sb
= dentry
->d_sb
;
662 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
663 struct nilfs_root
*root
= NILFS_I(d_inode(dentry
))->i_root
;
665 if (!nilfs_test_opt(nilfs
, BARRIER
))
666 seq_puts(seq
, ",nobarrier");
667 if (root
->cno
!= NILFS_CPTREE_CURRENT_CNO
)
668 seq_printf(seq
, ",cp=%llu", (unsigned long long)root
->cno
);
669 if (nilfs_test_opt(nilfs
, ERRORS_PANIC
))
670 seq_puts(seq
, ",errors=panic");
671 if (nilfs_test_opt(nilfs
, ERRORS_CONT
))
672 seq_puts(seq
, ",errors=continue");
673 if (nilfs_test_opt(nilfs
, STRICT_ORDER
))
674 seq_puts(seq
, ",order=strict");
675 if (nilfs_test_opt(nilfs
, NORECOVERY
))
676 seq_puts(seq
, ",norecovery");
677 if (nilfs_test_opt(nilfs
, DISCARD
))
678 seq_puts(seq
, ",discard");
683 static const struct super_operations nilfs_sops
= {
684 .alloc_inode
= nilfs_alloc_inode
,
685 .free_inode
= nilfs_free_inode
,
686 .dirty_inode
= nilfs_dirty_inode
,
687 .evict_inode
= nilfs_evict_inode
,
688 .put_super
= nilfs_put_super
,
689 .sync_fs
= nilfs_sync_fs
,
690 .freeze_fs
= nilfs_freeze
,
691 .unfreeze_fs
= nilfs_unfreeze
,
692 .statfs
= nilfs_statfs
,
693 .remount_fs
= nilfs_remount
,
694 .show_options
= nilfs_show_options
698 Opt_err_cont
, Opt_err_panic
, Opt_err_ro
,
699 Opt_barrier
, Opt_nobarrier
, Opt_snapshot
, Opt_order
, Opt_norecovery
,
700 Opt_discard
, Opt_nodiscard
, Opt_err
,
703 static match_table_t tokens
= {
704 {Opt_err_cont
, "errors=continue"},
705 {Opt_err_panic
, "errors=panic"},
706 {Opt_err_ro
, "errors=remount-ro"},
707 {Opt_barrier
, "barrier"},
708 {Opt_nobarrier
, "nobarrier"},
709 {Opt_snapshot
, "cp=%u"},
710 {Opt_order
, "order=%s"},
711 {Opt_norecovery
, "norecovery"},
712 {Opt_discard
, "discard"},
713 {Opt_nodiscard
, "nodiscard"},
717 static int parse_options(char *options
, struct super_block
*sb
, int is_remount
)
719 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
721 substring_t args
[MAX_OPT_ARGS
];
726 while ((p
= strsep(&options
, ",")) != NULL
) {
732 token
= match_token(p
, tokens
, args
);
735 nilfs_set_opt(nilfs
, BARRIER
);
738 nilfs_clear_opt(nilfs
, BARRIER
);
741 if (strcmp(args
[0].from
, "relaxed") == 0)
742 /* Ordered data semantics */
743 nilfs_clear_opt(nilfs
, STRICT_ORDER
);
744 else if (strcmp(args
[0].from
, "strict") == 0)
745 /* Strict in-order semantics */
746 nilfs_set_opt(nilfs
, STRICT_ORDER
);
751 nilfs_write_opt(nilfs
, ERROR_MODE
, ERRORS_PANIC
);
754 nilfs_write_opt(nilfs
, ERROR_MODE
, ERRORS_RO
);
757 nilfs_write_opt(nilfs
, ERROR_MODE
, ERRORS_CONT
);
762 "\"%s\" option is invalid for remount",
768 nilfs_set_opt(nilfs
, NORECOVERY
);
771 nilfs_set_opt(nilfs
, DISCARD
);
774 nilfs_clear_opt(nilfs
, DISCARD
);
777 nilfs_err(sb
, "unrecognized mount option \"%s\"", p
);
785 nilfs_set_default_options(struct super_block
*sb
,
786 struct nilfs_super_block
*sbp
)
788 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
790 nilfs
->ns_mount_opt
=
791 NILFS_MOUNT_ERRORS_RO
| NILFS_MOUNT_BARRIER
;
794 static int nilfs_setup_super(struct super_block
*sb
, int is_mount
)
796 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
797 struct nilfs_super_block
**sbp
;
801 /* nilfs->ns_sem must be locked by the caller. */
802 sbp
= nilfs_prepare_super(sb
, 0);
807 goto skip_mount_setup
;
809 max_mnt_count
= le16_to_cpu(sbp
[0]->s_max_mnt_count
);
810 mnt_count
= le16_to_cpu(sbp
[0]->s_mnt_count
);
812 if (nilfs
->ns_mount_state
& NILFS_ERROR_FS
) {
813 nilfs_warn(sb
, "mounting fs with errors");
815 } else if (max_mnt_count
>= 0 && mnt_count
>= max_mnt_count
) {
816 nilfs_warn(sb
, "maximal mount count reached");
820 sbp
[0]->s_max_mnt_count
= cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT
);
822 sbp
[0]->s_mnt_count
= cpu_to_le16(mnt_count
+ 1);
823 sbp
[0]->s_mtime
= cpu_to_le64(ktime_get_real_seconds());
827 cpu_to_le16(le16_to_cpu(sbp
[0]->s_state
) & ~NILFS_VALID_FS
);
828 /* synchronize sbp[1] with sbp[0] */
830 memcpy(sbp
[1], sbp
[0], nilfs
->ns_sbsize
);
831 return nilfs_commit_super(sb
, NILFS_SB_COMMIT_ALL
);
834 struct nilfs_super_block
*nilfs_read_super_block(struct super_block
*sb
,
835 u64 pos
, int blocksize
,
836 struct buffer_head
**pbh
)
838 unsigned long long sb_index
= pos
;
839 unsigned long offset
;
841 offset
= do_div(sb_index
, blocksize
);
842 *pbh
= sb_bread(sb
, sb_index
);
845 return (struct nilfs_super_block
*)((char *)(*pbh
)->b_data
+ offset
);
848 int nilfs_store_magic_and_option(struct super_block
*sb
,
849 struct nilfs_super_block
*sbp
,
852 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
854 sb
->s_magic
= le16_to_cpu(sbp
->s_magic
);
856 /* FS independent flags */
857 #ifdef NILFS_ATIME_DISABLE
858 sb
->s_flags
|= SB_NOATIME
;
861 nilfs_set_default_options(sb
, sbp
);
863 nilfs
->ns_resuid
= le16_to_cpu(sbp
->s_def_resuid
);
864 nilfs
->ns_resgid
= le16_to_cpu(sbp
->s_def_resgid
);
865 nilfs
->ns_interval
= le32_to_cpu(sbp
->s_c_interval
);
866 nilfs
->ns_watermark
= le32_to_cpu(sbp
->s_c_block_max
);
868 return !parse_options(data
, sb
, 0) ? -EINVAL
: 0;
871 int nilfs_check_feature_compatibility(struct super_block
*sb
,
872 struct nilfs_super_block
*sbp
)
876 features
= le64_to_cpu(sbp
->s_feature_incompat
) &
877 ~NILFS_FEATURE_INCOMPAT_SUPP
;
880 "couldn't mount because of unsupported optional features (%llx)",
881 (unsigned long long)features
);
884 features
= le64_to_cpu(sbp
->s_feature_compat_ro
) &
885 ~NILFS_FEATURE_COMPAT_RO_SUPP
;
886 if (!sb_rdonly(sb
) && features
) {
888 "couldn't mount RDWR because of unsupported optional features (%llx)",
889 (unsigned long long)features
);
895 static int nilfs_get_root_dentry(struct super_block
*sb
,
896 struct nilfs_root
*root
,
897 struct dentry
**root_dentry
)
900 struct dentry
*dentry
;
903 inode
= nilfs_iget(sb
, root
, NILFS_ROOT_INO
);
905 ret
= PTR_ERR(inode
);
906 nilfs_err(sb
, "error %d getting root inode", ret
);
909 if (!S_ISDIR(inode
->i_mode
) || !inode
->i_blocks
|| !inode
->i_size
) {
911 nilfs_err(sb
, "corrupt root inode");
916 if (root
->cno
== NILFS_CPTREE_CURRENT_CNO
) {
917 dentry
= d_find_alias(inode
);
919 dentry
= d_make_root(inode
);
928 dentry
= d_obtain_root(inode
);
929 if (IS_ERR(dentry
)) {
930 ret
= PTR_ERR(dentry
);
934 *root_dentry
= dentry
;
939 nilfs_err(sb
, "error %d getting root dentry", ret
);
943 static int nilfs_attach_snapshot(struct super_block
*s
, __u64 cno
,
944 struct dentry
**root_dentry
)
946 struct the_nilfs
*nilfs
= s
->s_fs_info
;
947 struct nilfs_root
*root
;
950 mutex_lock(&nilfs
->ns_snapshot_mount_mutex
);
952 down_read(&nilfs
->ns_segctor_sem
);
953 ret
= nilfs_cpfile_is_snapshot(nilfs
->ns_cpfile
, cno
);
954 up_read(&nilfs
->ns_segctor_sem
);
956 ret
= (ret
== -ENOENT
) ? -EINVAL
: ret
;
960 "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
961 (unsigned long long)cno
);
966 ret
= nilfs_attach_checkpoint(s
, cno
, false, &root
);
969 "error %d while loading snapshot (checkpoint number=%llu)",
970 ret
, (unsigned long long)cno
);
973 ret
= nilfs_get_root_dentry(s
, root
, root_dentry
);
974 nilfs_put_root(root
);
976 mutex_unlock(&nilfs
->ns_snapshot_mount_mutex
);
981 * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
982 * @root_dentry: root dentry of the tree to be shrunk
984 * This function returns true if the tree was in-use.
986 static bool nilfs_tree_is_busy(struct dentry
*root_dentry
)
988 shrink_dcache_parent(root_dentry
);
989 return d_count(root_dentry
) > 1;
992 int nilfs_checkpoint_is_mounted(struct super_block
*sb
, __u64 cno
)
994 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
995 struct nilfs_root
*root
;
997 struct dentry
*dentry
;
1000 if (cno
> nilfs
->ns_cno
)
1003 if (cno
>= nilfs_last_cno(nilfs
))
1004 return true; /* protect recent checkpoints */
1007 root
= nilfs_lookup_root(nilfs
, cno
);
1009 inode
= nilfs_ilookup(sb
, root
, NILFS_ROOT_INO
);
1011 dentry
= d_find_alias(inode
);
1013 ret
= nilfs_tree_is_busy(dentry
);
1018 nilfs_put_root(root
);
1024 * nilfs_fill_super() - initialize a super block instance
1026 * @data: mount options
1027 * @silent: silent mode flag
1029 * This function is called exclusively by nilfs->ns_mount_mutex.
1030 * So, the recovery process is protected from other simultaneous mounts.
1033 nilfs_fill_super(struct super_block
*sb
, void *data
, int silent
)
1035 struct the_nilfs
*nilfs
;
1036 struct nilfs_root
*fsroot
;
1040 nilfs
= alloc_nilfs(sb
);
1044 sb
->s_fs_info
= nilfs
;
1046 err
= init_nilfs(nilfs
, sb
, (char *)data
);
1050 sb
->s_op
= &nilfs_sops
;
1051 sb
->s_export_op
= &nilfs_export_ops
;
1053 sb
->s_time_gran
= 1;
1054 sb
->s_max_links
= NILFS_LINK_MAX
;
1056 sb
->s_bdi
= bdi_get(sb
->s_bdev
->bd_bdi
);
1058 err
= load_nilfs(nilfs
, sb
);
1062 cno
= nilfs_last_cno(nilfs
);
1063 err
= nilfs_attach_checkpoint(sb
, cno
, true, &fsroot
);
1066 "error %d while loading last checkpoint (checkpoint number=%llu)",
1067 err
, (unsigned long long)cno
);
1071 if (!sb_rdonly(sb
)) {
1072 err
= nilfs_attach_log_writer(sb
, fsroot
);
1074 goto failed_checkpoint
;
1077 err
= nilfs_get_root_dentry(sb
, fsroot
, &sb
->s_root
);
1079 goto failed_segctor
;
1081 nilfs_put_root(fsroot
);
1083 if (!sb_rdonly(sb
)) {
1084 down_write(&nilfs
->ns_sem
);
1085 nilfs_setup_super(sb
, true);
1086 up_write(&nilfs
->ns_sem
);
1092 nilfs_detach_log_writer(sb
);
1095 nilfs_put_root(fsroot
);
1098 iput(nilfs
->ns_sufile
);
1099 iput(nilfs
->ns_cpfile
);
1100 iput(nilfs
->ns_dat
);
1103 destroy_nilfs(nilfs
);
1107 static int nilfs_remount(struct super_block
*sb
, int *flags
, char *data
)
1109 struct the_nilfs
*nilfs
= sb
->s_fs_info
;
1110 unsigned long old_sb_flags
;
1111 unsigned long old_mount_opt
;
1114 sync_filesystem(sb
);
1115 old_sb_flags
= sb
->s_flags
;
1116 old_mount_opt
= nilfs
->ns_mount_opt
;
1118 if (!parse_options(data
, sb
, 1)) {
1122 sb
->s_flags
= (sb
->s_flags
& ~SB_POSIXACL
);
1126 if (!nilfs_valid_fs(nilfs
)) {
1128 "couldn't remount because the filesystem is in an incomplete recovery state");
1132 if ((bool)(*flags
& SB_RDONLY
) == sb_rdonly(sb
))
1134 if (*flags
& SB_RDONLY
) {
1135 /* Shutting down log writer */
1136 nilfs_detach_log_writer(sb
);
1137 sb
->s_flags
|= SB_RDONLY
;
1140 * Remounting a valid RW partition RDONLY, so set
1141 * the RDONLY flag and then mark the partition as valid again.
1143 down_write(&nilfs
->ns_sem
);
1144 nilfs_cleanup_super(sb
);
1145 up_write(&nilfs
->ns_sem
);
1148 struct nilfs_root
*root
;
1151 * Mounting a RDONLY partition read-write, so reread and
1152 * store the current valid flag. (It may have been changed
1153 * by fsck since we originally mounted the partition.)
1155 down_read(&nilfs
->ns_sem
);
1156 features
= le64_to_cpu(nilfs
->ns_sbp
[0]->s_feature_compat_ro
) &
1157 ~NILFS_FEATURE_COMPAT_RO_SUPP
;
1158 up_read(&nilfs
->ns_sem
);
1161 "couldn't remount RDWR because of unsupported optional features (%llx)",
1162 (unsigned long long)features
);
1167 sb
->s_flags
&= ~SB_RDONLY
;
1169 root
= NILFS_I(d_inode(sb
->s_root
))->i_root
;
1170 err
= nilfs_attach_log_writer(sb
, root
);
1174 down_write(&nilfs
->ns_sem
);
1175 nilfs_setup_super(sb
, true);
1176 up_write(&nilfs
->ns_sem
);
1182 sb
->s_flags
= old_sb_flags
;
1183 nilfs
->ns_mount_opt
= old_mount_opt
;
1187 struct nilfs_super_data
{
1188 struct block_device
*bdev
;
1193 static int nilfs_parse_snapshot_option(const char *option
,
1194 const substring_t
*arg
,
1195 struct nilfs_super_data
*sd
)
1197 unsigned long long val
;
1198 const char *msg
= NULL
;
1201 if (!(sd
->flags
& SB_RDONLY
)) {
1202 msg
= "read-only option is not specified";
1206 err
= kstrtoull(arg
->from
, 0, &val
);
1209 msg
= "too large checkpoint number";
1211 msg
= "malformed argument";
1213 } else if (val
== 0) {
1214 msg
= "invalid checkpoint number 0";
1221 nilfs_err(NULL
, "invalid option \"%s\": %s", option
, msg
);
1226 * nilfs_identify - pre-read mount options needed to identify mount instance
1227 * @data: mount options
1228 * @sd: nilfs_super_data
1230 static int nilfs_identify(char *data
, struct nilfs_super_data
*sd
)
1232 char *p
, *options
= data
;
1233 substring_t args
[MAX_OPT_ARGS
];
1238 p
= strsep(&options
, ",");
1239 if (p
!= NULL
&& *p
) {
1240 token
= match_token(p
, tokens
, args
);
1241 if (token
== Opt_snapshot
)
1242 ret
= nilfs_parse_snapshot_option(p
, &args
[0],
1247 BUG_ON(options
== data
);
1248 *(options
- 1) = ',';
1253 static int nilfs_set_bdev_super(struct super_block
*s
, void *data
)
1256 s
->s_dev
= s
->s_bdev
->bd_dev
;
1260 static int nilfs_test_bdev_super(struct super_block
*s
, void *data
)
1262 return (void *)s
->s_bdev
== data
;
1265 static struct dentry
*
1266 nilfs_mount(struct file_system_type
*fs_type
, int flags
,
1267 const char *dev_name
, void *data
)
1269 struct nilfs_super_data sd
;
1270 struct super_block
*s
;
1271 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1272 struct dentry
*root_dentry
;
1273 int err
, s_new
= false;
1275 if (!(flags
& SB_RDONLY
))
1276 mode
|= FMODE_WRITE
;
1278 sd
.bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1279 if (IS_ERR(sd
.bdev
))
1280 return ERR_CAST(sd
.bdev
);
1284 if (nilfs_identify((char *)data
, &sd
)) {
1290 * once the super is inserted into the list by sget, s_umount
1291 * will protect the lockfs code from trying to start a snapshot
1292 * while we are mounting
1294 mutex_lock(&sd
.bdev
->bd_fsfreeze_mutex
);
1295 if (sd
.bdev
->bd_fsfreeze_count
> 0) {
1296 mutex_unlock(&sd
.bdev
->bd_fsfreeze_mutex
);
1300 s
= sget(fs_type
, nilfs_test_bdev_super
, nilfs_set_bdev_super
, flags
,
1302 mutex_unlock(&sd
.bdev
->bd_fsfreeze_mutex
);
1311 /* New superblock instance created */
1313 snprintf(s
->s_id
, sizeof(s
->s_id
), "%pg", sd
.bdev
);
1314 sb_set_blocksize(s
, block_size(sd
.bdev
));
1316 err
= nilfs_fill_super(s
, data
, flags
& SB_SILENT
? 1 : 0);
1320 s
->s_flags
|= SB_ACTIVE
;
1321 } else if (!sd
.cno
) {
1322 if (nilfs_tree_is_busy(s
->s_root
)) {
1323 if ((flags
^ s
->s_flags
) & SB_RDONLY
) {
1325 "the device already has a %s mount.",
1326 sb_rdonly(s
) ? "read-only" : "read/write");
1332 * Try remount to setup mount states if the current
1333 * tree is not mounted and only snapshots use this sb.
1335 err
= nilfs_remount(s
, &flags
, data
);
1342 err
= nilfs_attach_snapshot(s
, sd
.cno
, &root_dentry
);
1346 root_dentry
= dget(s
->s_root
);
1350 blkdev_put(sd
.bdev
, mode
);
1355 deactivate_locked_super(s
);
1359 blkdev_put(sd
.bdev
, mode
);
1360 return ERR_PTR(err
);
1363 struct file_system_type nilfs_fs_type
= {
1364 .owner
= THIS_MODULE
,
1366 .mount
= nilfs_mount
,
1367 .kill_sb
= kill_block_super
,
1368 .fs_flags
= FS_REQUIRES_DEV
,
1370 MODULE_ALIAS_FS("nilfs2");
1372 static void nilfs_inode_init_once(void *obj
)
1374 struct nilfs_inode_info
*ii
= obj
;
1376 INIT_LIST_HEAD(&ii
->i_dirty
);
1377 #ifdef CONFIG_NILFS_XATTR
1378 init_rwsem(&ii
->xattr_sem
);
1380 address_space_init_once(&ii
->i_btnode_cache
);
1381 ii
->i_bmap
= &ii
->i_bmap_data
;
1382 inode_init_once(&ii
->vfs_inode
);
1385 static void nilfs_segbuf_init_once(void *obj
)
1387 memset(obj
, 0, sizeof(struct nilfs_segment_buffer
));
1390 static void nilfs_destroy_cachep(void)
1393 * Make sure all delayed rcu free inodes are flushed before we
1398 kmem_cache_destroy(nilfs_inode_cachep
);
1399 kmem_cache_destroy(nilfs_transaction_cachep
);
1400 kmem_cache_destroy(nilfs_segbuf_cachep
);
1401 kmem_cache_destroy(nilfs_btree_path_cache
);
1404 static int __init
nilfs_init_cachep(void)
1406 nilfs_inode_cachep
= kmem_cache_create("nilfs2_inode_cache",
1407 sizeof(struct nilfs_inode_info
), 0,
1408 SLAB_RECLAIM_ACCOUNT
|SLAB_ACCOUNT
,
1409 nilfs_inode_init_once
);
1410 if (!nilfs_inode_cachep
)
1413 nilfs_transaction_cachep
= kmem_cache_create("nilfs2_transaction_cache",
1414 sizeof(struct nilfs_transaction_info
), 0,
1415 SLAB_RECLAIM_ACCOUNT
, NULL
);
1416 if (!nilfs_transaction_cachep
)
1419 nilfs_segbuf_cachep
= kmem_cache_create("nilfs2_segbuf_cache",
1420 sizeof(struct nilfs_segment_buffer
), 0,
1421 SLAB_RECLAIM_ACCOUNT
, nilfs_segbuf_init_once
);
1422 if (!nilfs_segbuf_cachep
)
1425 nilfs_btree_path_cache
= kmem_cache_create("nilfs2_btree_path_cache",
1426 sizeof(struct nilfs_btree_path
) * NILFS_BTREE_LEVEL_MAX
,
1428 if (!nilfs_btree_path_cache
)
1434 nilfs_destroy_cachep();
1438 static int __init
init_nilfs_fs(void)
1442 err
= nilfs_init_cachep();
1446 err
= nilfs_sysfs_init();
1450 err
= register_filesystem(&nilfs_fs_type
);
1452 goto deinit_sysfs_entry
;
1454 printk(KERN_INFO
"NILFS version 2 loaded\n");
1460 nilfs_destroy_cachep();
1465 static void __exit
exit_nilfs_fs(void)
1467 nilfs_destroy_cachep();
1469 unregister_filesystem(&nilfs_fs_type
);
1472 module_init(init_nilfs_fs
)
1473 module_exit(exit_nilfs_fs
)