2 * super.c - NILFS module and super block management.
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 * Written by Ryusuke Konishi <ryusuke@osrg.net>
23 * linux/fs/ext2/super.c
25 * Copyright (C) 1992, 1993, 1994, 1995
26 * Remy Card (card@masi.ibp.fr)
27 * Laboratoire MASI - Institut Blaise Pascal
28 * Universite Pierre et Marie Curie (Paris VI)
32 * linux/fs/minix/inode.c
34 * Copyright (C) 1991, 1992 Linus Torvalds
36 * Big-endian to little-endian byte-swapping/bitmaps by
37 * David S. Miller (davem@caip.rutgers.edu), 1995
40 #include <linux/module.h>
41 #include <linux/string.h>
42 #include <linux/slab.h>
43 #include <linux/init.h>
44 #include <linux/blkdev.h>
45 #include <linux/parser.h>
46 #include <linux/random.h>
47 #include <linux/crc32.h>
48 #include <linux/smp_lock.h>
49 #include <linux/vfs.h>
50 #include <linux/writeback.h>
51 #include <linux/kobject.h>
52 #include <linux/exportfs.h>
63 MODULE_AUTHOR("NTT Corp.");
64 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
66 MODULE_VERSION(NILFS_VERSION);
67 MODULE_LICENSE("GPL");
69 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
70 static int test_exclusive_mount(struct file_system_type *fs_type,
71 struct block_device *bdev, int flags);
74 * nilfs_error() - report failure condition on a filesystem
76 * nilfs_error() sets an ERROR_FS flag on the superblock as well as
77 * reporting an error message. It should be called when NILFS detects
78 * incoherences or defects of meta data on disk. As for sustainable
79 * errors such as a single-shot I/O error, nilfs_warning() or the printk()
80 * function should be used instead.
82 * The segment constructor must not call this function because it can
85 void nilfs_error(struct super_block *sb, const char *function,
88 struct nilfs_sb_info *sbi = NILFS_SB(sb);
92 printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
97 if (!(sb->s_flags & MS_RDONLY)) {
98 struct the_nilfs *nilfs = sbi->s_nilfs;
100 if (!nilfs_test_opt(sbi, ERRORS_CONT))
101 nilfs_detach_segment_constructor(sbi);
103 down_write(&nilfs->ns_sem);
104 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
105 nilfs->ns_mount_state |= NILFS_ERROR_FS;
106 nilfs->ns_sbp->s_state |= cpu_to_le16(NILFS_ERROR_FS);
107 nilfs_commit_super(sbi);
109 up_write(&nilfs->ns_sem);
111 if (nilfs_test_opt(sbi, ERRORS_RO)) {
112 printk(KERN_CRIT "Remounting filesystem read-only\n");
113 sb->s_flags |= MS_RDONLY;
117 if (nilfs_test_opt(sbi, ERRORS_PANIC))
118 panic("NILFS (device %s): panic forced after error\n",
122 void nilfs_warning(struct super_block *sb, const char *function,
123 const char *fmt, ...)
128 printk(KERN_WARNING "NILFS warning (device %s): %s: ",
135 static struct kmem_cache *nilfs_inode_cachep;
137 struct inode *nilfs_alloc_inode(struct super_block *sb)
139 struct nilfs_inode_info *ii;
141 ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
146 ii->vfs_inode.i_version = 1;
147 nilfs_btnode_cache_init(&ii->i_btnode_cache);
148 return &ii->vfs_inode;
151 void nilfs_destroy_inode(struct inode *inode)
153 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
156 static void init_once(void *obj)
158 struct nilfs_inode_info *ii = obj;
160 INIT_LIST_HEAD(&ii->i_dirty);
161 #ifdef CONFIG_NILFS_XATTR
162 init_rwsem(&ii->xattr_sem);
164 nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
165 ii->i_bmap = (struct nilfs_bmap *)&ii->i_bmap_union;
166 inode_init_once(&ii->vfs_inode);
169 static int nilfs_init_inode_cache(void)
171 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
172 sizeof(struct nilfs_inode_info),
173 0, SLAB_RECLAIM_ACCOUNT,
176 return (nilfs_inode_cachep == NULL) ? -ENOMEM : 0;
179 static inline void nilfs_destroy_inode_cache(void)
181 kmem_cache_destroy(nilfs_inode_cachep);
184 static void nilfs_clear_inode(struct inode *inode)
186 struct nilfs_inode_info *ii = NILFS_I(inode);
187 struct nilfs_transaction_info ti;
188 struct nilfs_sb_info *sbi = NILFS_SB(inode->i_sb);
190 #ifdef CONFIG_NILFS_POSIX_ACL
191 if (ii->i_acl && ii->i_acl != NILFS_ACL_NOT_CACHED) {
192 posix_acl_release(ii->i_acl);
193 ii->i_acl = NILFS_ACL_NOT_CACHED;
195 if (ii->i_default_acl && ii->i_default_acl != NILFS_ACL_NOT_CACHED) {
196 posix_acl_release(ii->i_default_acl);
197 ii->i_default_acl = NILFS_ACL_NOT_CACHED;
201 * Free resources allocated in nilfs_read_inode(), here.
203 nilfs_transaction_begin(inode->i_sb, &ti, 0);
205 spin_lock(&sbi->s_inode_lock);
206 if (!list_empty(&ii->i_dirty))
207 list_del_init(&ii->i_dirty);
210 spin_unlock(&sbi->s_inode_lock);
212 if (test_bit(NILFS_I_BMAP, &ii->i_state))
213 nilfs_bmap_clear(ii->i_bmap);
215 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
217 nilfs_transaction_end(inode->i_sb, 0);
221 * nilfs_update_last_segment - change pointer to the latest segment
222 * @sbi: nilfs_sb_info
223 * @update_cno: flag whether to update checkpoint number.
225 * nilfs_update_last_segment() changes information in the super block
226 * after a partial segment is written out successfully. The super
227 * block is marked dirty. It will be written out at the next VFS sync
228 * operations such as sync_supers() and generic_shutdown_super().
230 void nilfs_update_last_segment(struct nilfs_sb_info *sbi, int update_cno)
232 struct the_nilfs *nilfs = sbi->s_nilfs;
233 struct nilfs_super_block *sbp = nilfs->ns_sbp;
235 /* nilfs->sem must be locked by the caller. */
236 spin_lock(&nilfs->ns_last_segment_lock);
238 nilfs->ns_last_cno = nilfs->ns_cno++;
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);
244 sbi->s_super->s_dirt = 1; /* must be set if delaying the call of
245 nilfs_commit_super() */
248 static int nilfs_sync_super(struct nilfs_sb_info *sbi)
250 struct the_nilfs *nilfs = sbi->s_nilfs;
252 int barrier_done = 0;
254 if (nilfs_test_opt(sbi, BARRIER)) {
255 set_buffer_ordered(nilfs->ns_sbh);
259 set_buffer_dirty(nilfs->ns_sbh);
260 err = sync_dirty_buffer(nilfs->ns_sbh);
261 if (err == -EOPNOTSUPP && barrier_done) {
262 nilfs_warning(sbi->s_super, __func__,
263 "barrier-based sync failed. "
264 "disabling barriers\n");
265 nilfs_clear_opt(sbi, BARRIER);
267 clear_buffer_ordered(nilfs->ns_sbh);
272 "NILFS: unable to write superblock (err=%d)\n", err);
274 nilfs_dispose_used_segments(nilfs);
275 clear_nilfs_discontinued(nilfs);
281 int nilfs_commit_super(struct nilfs_sb_info *sbi)
283 struct the_nilfs *nilfs = sbi->s_nilfs;
284 struct nilfs_super_block *sbp = nilfs->ns_sbp;
285 sector_t nfreeblocks;
288 /* nilfs->sem must be locked by the caller. */
289 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
291 printk(KERN_ERR "NILFS: failed to count free blocks\n");
294 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
295 sbp->s_wtime = cpu_to_le64(get_seconds());
297 sbp->s_sum = crc32_le(nilfs->ns_crc_seed, (unsigned char *)sbp,
298 le16_to_cpu(sbp->s_bytes));
300 sbi->s_super->s_dirt = 0;
301 return nilfs_sync_super(sbi);
304 static void nilfs_put_super(struct super_block *sb)
306 struct nilfs_sb_info *sbi = NILFS_SB(sb);
307 struct the_nilfs *nilfs = sbi->s_nilfs;
309 nilfs_detach_segment_constructor(sbi);
311 if (!(sb->s_flags & MS_RDONLY)) {
312 down_write(&nilfs->ns_sem);
313 nilfs->ns_sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
314 nilfs_commit_super(sbi);
315 up_write(&nilfs->ns_sem);
318 nilfs_detach_checkpoint(sbi);
319 put_nilfs(sbi->s_nilfs);
321 sb->s_fs_info = NULL;
326 * nilfs_write_super - write super block(s) of NILFS
329 * nilfs_write_super() gets a fs-dependent lock, writes super block(s), and
330 * clears s_dirt. This function is called in the section protected by
333 * The s_dirt flag is managed by each filesystem and we protect it by ns_sem
334 * of the struct the_nilfs. Lock order must be as follows:
337 * 2. down_write(&nilfs->ns_sem)
339 * Inside NILFS, locking ns_sem is enough to protect s_dirt and the buffer
340 * of the super block (nilfs->ns_sbp).
342 * In most cases, VFS functions call lock_super() before calling these
343 * methods. So we must be careful not to bring on deadlocks when using
344 * lock_super(); see generic_shutdown_super(), write_super(), and so on.
346 * Note that order of lock_kernel() and lock_super() depends on contexts
347 * of VFS. We should also note that lock_kernel() can be used in its
348 * protective section and only the outermost one has an effect.
350 static void nilfs_write_super(struct super_block *sb)
352 struct nilfs_sb_info *sbi = NILFS_SB(sb);
353 struct the_nilfs *nilfs = sbi->s_nilfs;
355 down_write(&nilfs->ns_sem);
356 if (!(sb->s_flags & MS_RDONLY))
357 nilfs_commit_super(sbi);
359 up_write(&nilfs->ns_sem);
362 static int nilfs_sync_fs(struct super_block *sb, int wait)
366 /* This function is called when super block should be written back */
368 err = nilfs_construct_segment(sb);
372 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
374 struct the_nilfs *nilfs = sbi->s_nilfs;
375 struct nilfs_checkpoint *raw_cp;
376 struct buffer_head *bh_cp;
379 down_write(&nilfs->ns_sem);
380 list_add(&sbi->s_list, &nilfs->ns_supers);
381 up_write(&nilfs->ns_sem);
383 sbi->s_ifile = nilfs_mdt_new(
384 nilfs, sbi->s_super, NILFS_IFILE_INO, NILFS_IFILE_GFP);
388 err = nilfs_palloc_init_blockgroup(sbi->s_ifile, nilfs->ns_inode_size);
392 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
395 if (err == -ENOENT || err == -EINVAL) {
397 "NILFS: Invalid checkpoint "
398 "(checkpoint number=%llu)\n",
399 (unsigned long long)cno);
404 err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
407 atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
408 atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
410 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
414 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
416 nilfs_mdt_destroy(sbi->s_ifile);
419 down_write(&nilfs->ns_sem);
420 list_del_init(&sbi->s_list);
421 up_write(&nilfs->ns_sem);
426 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
428 struct the_nilfs *nilfs = sbi->s_nilfs;
430 nilfs_mdt_clear(sbi->s_ifile);
431 nilfs_mdt_destroy(sbi->s_ifile);
433 down_write(&nilfs->ns_sem);
434 list_del_init(&sbi->s_list);
435 up_write(&nilfs->ns_sem);
438 static int nilfs_mark_recovery_complete(struct nilfs_sb_info *sbi)
440 struct the_nilfs *nilfs = sbi->s_nilfs;
443 down_write(&nilfs->ns_sem);
444 if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
445 nilfs->ns_mount_state |= NILFS_VALID_FS;
446 err = nilfs_commit_super(sbi);
448 printk(KERN_INFO "NILFS: recovery complete.\n");
450 up_write(&nilfs->ns_sem);
454 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
456 struct super_block *sb = dentry->d_sb;
457 struct nilfs_sb_info *sbi = NILFS_SB(sb);
458 unsigned long long blocks;
459 unsigned long overhead;
460 unsigned long nrsvblocks;
461 sector_t nfreeblocks;
462 struct the_nilfs *nilfs = sbi->s_nilfs;
466 * Compute all of the segment blocks
468 * The blocks before first segment and after last segment
471 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
472 - nilfs->ns_first_data_block;
473 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
476 * Compute the overhead
478 * When distributing meta data blocks outside semgent structure,
479 * We must count them as the overhead.
483 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
487 buf->f_type = NILFS_SUPER_MAGIC;
488 buf->f_bsize = sb->s_blocksize;
489 buf->f_blocks = blocks - overhead;
490 buf->f_bfree = nfreeblocks;
491 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
492 (buf->f_bfree - nrsvblocks) : 0;
493 buf->f_files = atomic_read(&sbi->s_inodes_count);
494 buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
495 buf->f_namelen = NILFS_NAME_LEN;
499 static struct super_operations nilfs_sops = {
500 .alloc_inode = nilfs_alloc_inode,
501 .destroy_inode = nilfs_destroy_inode,
502 .dirty_inode = nilfs_dirty_inode,
503 /* .write_inode = nilfs_write_inode, */
504 /* .put_inode = nilfs_put_inode, */
505 /* .drop_inode = nilfs_drop_inode, */
506 .delete_inode = nilfs_delete_inode,
507 .put_super = nilfs_put_super,
508 .write_super = nilfs_write_super,
509 .sync_fs = nilfs_sync_fs,
510 /* .write_super_lockfs */
512 .statfs = nilfs_statfs,
513 .remount_fs = nilfs_remount,
514 .clear_inode = nilfs_clear_inode,
519 static struct inode *
520 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
524 if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
525 ino != NILFS_SKETCH_INO)
526 return ERR_PTR(-ESTALE);
528 inode = nilfs_iget(sb, ino);
530 return ERR_CAST(inode);
531 if (generation && inode->i_generation != generation) {
533 return ERR_PTR(-ESTALE);
539 static struct dentry *
540 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
543 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
544 nilfs_nfs_get_inode);
547 static struct dentry *
548 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
551 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
552 nilfs_nfs_get_inode);
555 static struct export_operations nilfs_export_ops = {
556 .fh_to_dentry = nilfs_fh_to_dentry,
557 .fh_to_parent = nilfs_fh_to_parent,
558 .get_parent = nilfs_get_parent,
562 Opt_err_cont, Opt_err_panic, Opt_err_ro,
563 Opt_barrier, Opt_snapshot, Opt_order,
567 static match_table_t tokens = {
568 {Opt_err_cont, "errors=continue"},
569 {Opt_err_panic, "errors=panic"},
570 {Opt_err_ro, "errors=remount-ro"},
571 {Opt_barrier, "barrier=%s"},
572 {Opt_snapshot, "cp=%u"},
573 {Opt_order, "order=%s"},
577 static int match_bool(substring_t *s, int *result)
579 int len = s->to - s->from;
581 if (strncmp(s->from, "on", len) == 0)
583 else if (strncmp(s->from, "off", len) == 0)
590 static int parse_options(char *options, struct super_block *sb)
592 struct nilfs_sb_info *sbi = NILFS_SB(sb);
594 substring_t args[MAX_OPT_ARGS];
600 while ((p = strsep(&options, ",")) != NULL) {
605 token = match_token(p, tokens, args);
608 if (match_bool(&args[0], &option))
611 nilfs_set_opt(sbi, BARRIER);
613 nilfs_clear_opt(sbi, BARRIER);
616 if (strcmp(args[0].from, "relaxed") == 0)
617 /* Ordered data semantics */
618 nilfs_clear_opt(sbi, STRICT_ORDER);
619 else if (strcmp(args[0].from, "strict") == 0)
620 /* Strict in-order semantics */
621 nilfs_set_opt(sbi, STRICT_ORDER);
626 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
629 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
632 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
635 if (match_int(&args[0], &option) || option <= 0)
637 if (!(sb->s_flags & MS_RDONLY))
639 sbi->s_snapshot_cno = option;
640 nilfs_set_opt(sbi, SNAPSHOT);
644 "NILFS: Unrecognized mount option \"%s\"\n", p);
652 nilfs_set_default_options(struct nilfs_sb_info *sbi,
653 struct nilfs_super_block *sbp)
656 NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER;
659 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
661 struct the_nilfs *nilfs = sbi->s_nilfs;
662 struct nilfs_super_block *sbp = nilfs->ns_sbp;
663 int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
664 int mnt_count = le16_to_cpu(sbp->s_mnt_count);
666 /* nilfs->sem must be locked by the caller. */
667 if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
668 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
669 } else if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
671 "NILFS warning: mounting fs with errors\n");
673 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
675 "NILFS warning: maximal mount count reached\n");
679 sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
681 sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
682 sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
683 sbp->s_mtime = cpu_to_le64(get_seconds());
684 return nilfs_commit_super(sbi);
687 struct nilfs_super_block *
688 nilfs_load_super_block(struct super_block *sb, struct buffer_head **pbh)
691 unsigned long offset, sb_index;
694 * Adjusting block size
695 * Blocksize will be enlarged when it is smaller than hardware
697 * Disk format of superblock does not change.
699 blocksize = sb_min_blocksize(sb, BLOCK_SIZE);
702 "NILFS: unable to set blocksize of superblock\n");
705 sb_index = NILFS_SB_OFFSET_BYTES / blocksize;
706 offset = NILFS_SB_OFFSET_BYTES % blocksize;
708 *pbh = sb_bread(sb, sb_index);
710 printk(KERN_ERR "NILFS: unable to read superblock\n");
713 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
716 struct nilfs_super_block *
717 nilfs_reload_super_block(struct super_block *sb, struct buffer_head **pbh,
720 struct nilfs_super_block *sbp;
721 unsigned long offset, sb_index;
722 int hw_blocksize = bdev_hardsect_size(sb->s_bdev);
724 if (blocksize < hw_blocksize) {
726 "NILFS: blocksize %d too small for device "
727 "(sector-size = %d).\n",
728 blocksize, hw_blocksize);
732 sb_set_blocksize(sb, blocksize);
734 sb_index = NILFS_SB_OFFSET_BYTES / blocksize;
735 offset = NILFS_SB_OFFSET_BYTES % blocksize;
737 *pbh = sb_bread(sb, sb_index);
740 "NILFS: cannot read superblock on 2nd try.\n");
744 sbp = (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
745 if (sbp->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
747 "NILFS: !? Magic mismatch on 2nd try.\n");
759 int nilfs_store_magic_and_option(struct super_block *sb,
760 struct nilfs_super_block *sbp,
763 struct nilfs_sb_info *sbi = NILFS_SB(sb);
765 /* trying to fill super (1st stage) */
766 sb->s_magic = le16_to_cpu(sbp->s_magic);
768 /* FS independent flags */
769 #ifdef NILFS_ATIME_DISABLE
770 sb->s_flags |= MS_NOATIME;
773 if (sb->s_magic != NILFS_SUPER_MAGIC) {
774 printk("NILFS: Can't find nilfs on dev %s.\n", sb->s_id);
778 nilfs_set_default_options(sbi, sbp);
780 sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
781 sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
782 sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
783 sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
785 if (!parse_options(data, sb))
792 * nilfs_fill_super() - initialize a super block instance
794 * @data: mount options
795 * @silent: silent mode flag
796 * @nilfs: the_nilfs struct
798 * This function is called exclusively by bd_mount_mutex.
799 * So, the recovery process is protected from other simultaneous mounts.
802 nilfs_fill_super(struct super_block *sb, void *data, int silent,
803 struct the_nilfs *nilfs)
805 struct nilfs_sb_info *sbi;
810 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
817 sbi->s_nilfs = nilfs;
820 err = init_nilfs(nilfs, sbi, (char *)data);
824 spin_lock_init(&sbi->s_inode_lock);
825 INIT_LIST_HEAD(&sbi->s_dirty_files);
826 INIT_LIST_HEAD(&sbi->s_list);
829 * Following initialization is overlapped because
830 * nilfs_sb_info structure has been cleared at the beginning.
831 * But we reserve them to keep our interest and make ready
832 * for the future change.
834 get_random_bytes(&sbi->s_next_generation,
835 sizeof(sbi->s_next_generation));
836 spin_lock_init(&sbi->s_next_gen_lock);
838 sb->s_op = &nilfs_sops;
839 sb->s_export_op = &nilfs_export_ops;
842 if (!nilfs_loaded(nilfs)) {
843 err = load_nilfs(nilfs, sbi);
847 cno = nilfs_last_cno(nilfs);
849 if (sb->s_flags & MS_RDONLY) {
850 if (nilfs_test_opt(sbi, SNAPSHOT)) {
851 if (!nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
852 sbi->s_snapshot_cno)) {
854 "NILFS: The specified checkpoint is "
856 "(checkpoint number=%llu).\n",
857 (unsigned long long)sbi->s_snapshot_cno);
861 cno = sbi->s_snapshot_cno;
863 /* Read-only mount */
864 sbi->s_snapshot_cno = cno;
867 err = nilfs_attach_checkpoint(sbi, cno);
869 printk(KERN_ERR "NILFS: error loading a checkpoint"
870 " (checkpoint number=%llu).\n", (unsigned long long)cno);
874 if (!(sb->s_flags & MS_RDONLY)) {
875 err = nilfs_attach_segment_constructor(sbi, NULL);
877 goto failed_checkpoint;
880 root = nilfs_iget(sb, NILFS_ROOT_INO);
882 printk(KERN_ERR "NILFS: get root inode failed\n");
886 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
888 printk(KERN_ERR "NILFS: corrupt root inode.\n");
892 sb->s_root = d_alloc_root(root);
895 printk(KERN_ERR "NILFS: get root dentry failed\n");
900 if (!(sb->s_flags & MS_RDONLY)) {
901 down_write(&nilfs->ns_sem);
902 nilfs_setup_super(sbi);
903 up_write(&nilfs->ns_sem);
906 err = nilfs_mark_recovery_complete(sbi);
908 printk(KERN_ERR "NILFS: recovery failed.\n");
919 nilfs_detach_segment_constructor(sbi);
922 nilfs_detach_checkpoint(sbi);
926 sb->s_fs_info = NULL;
931 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
933 struct nilfs_sb_info *sbi = NILFS_SB(sb);
934 struct nilfs_super_block *sbp;
935 struct the_nilfs *nilfs = sbi->s_nilfs;
936 unsigned long old_sb_flags;
937 struct nilfs_mount_options old_opts;
940 old_sb_flags = sb->s_flags;
941 old_opts.mount_opt = sbi->s_mount_opt;
942 old_opts.snapshot_cno = sbi->s_snapshot_cno;
944 if (!parse_options(data, sb)) {
948 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
950 if ((*flags & MS_RDONLY) &&
951 sbi->s_snapshot_cno != old_opts.snapshot_cno) {
952 printk(KERN_WARNING "NILFS (device %s): couldn't "
953 "remount to a different snapshot. \n",
959 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
961 if (*flags & MS_RDONLY) {
962 /* Shutting down the segment constructor */
963 nilfs_detach_segment_constructor(sbi);
964 sb->s_flags |= MS_RDONLY;
966 sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
967 /* nilfs_set_opt(sbi, SNAPSHOT); */
970 * Remounting a valid RW partition RDONLY, so set
971 * the RDONLY flag and then mark the partition as valid again.
973 down_write(&nilfs->ns_sem);
975 if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
976 (nilfs->ns_mount_state & NILFS_VALID_FS))
977 sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
978 sbp->s_mtime = cpu_to_le64(get_seconds());
979 nilfs_commit_super(sbi);
980 up_write(&nilfs->ns_sem);
983 * Mounting a RDONLY partition read-write, so reread and
984 * store the current valid flag. (It may have been changed
985 * by fsck since we originally mounted the partition.)
987 down(&sb->s_bdev->bd_mount_sem);
988 /* Check existing RW-mount */
989 if (test_exclusive_mount(sb->s_type, sb->s_bdev, 0)) {
990 printk(KERN_WARNING "NILFS (device %s): couldn't "
991 "remount because a RW-mount exists.\n",
994 goto rw_remount_failed;
996 if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
997 printk(KERN_WARNING "NILFS (device %s): couldn't "
998 "remount because the current RO-mount is not "
1002 goto rw_remount_failed;
1004 sb->s_flags &= ~MS_RDONLY;
1005 nilfs_clear_opt(sbi, SNAPSHOT);
1006 sbi->s_snapshot_cno = 0;
1008 err = nilfs_attach_segment_constructor(sbi, NULL);
1010 goto rw_remount_failed;
1012 down_write(&nilfs->ns_sem);
1013 nilfs_setup_super(sbi);
1014 up_write(&nilfs->ns_sem);
1016 up(&sb->s_bdev->bd_mount_sem);
1022 up(&sb->s_bdev->bd_mount_sem);
1024 sb->s_flags = old_sb_flags;
1025 sbi->s_mount_opt = old_opts.mount_opt;
1026 sbi->s_snapshot_cno = old_opts.snapshot_cno;
1030 struct nilfs_super_data {
1031 struct block_device *bdev;
1037 * nilfs_identify - pre-read mount options needed to identify mount instance
1038 * @data: mount options
1039 * @sd: nilfs_super_data
1041 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1043 char *p, *options = data;
1044 substring_t args[MAX_OPT_ARGS];
1049 p = strsep(&options, ",");
1050 if (p != NULL && *p) {
1051 token = match_token(p, tokens, args);
1052 if (token == Opt_snapshot) {
1053 if (!(sd->flags & MS_RDONLY))
1056 ret = match_int(&args[0], &option);
1067 "NILFS: invalid mount option: %s\n", p);
1071 BUG_ON(options == data);
1072 *(options - 1) = ',';
1077 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1079 struct nilfs_super_data *sd = data;
1081 s->s_bdev = sd->bdev;
1082 s->s_dev = s->s_bdev->bd_dev;
1086 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1088 struct nilfs_super_data *sd = data;
1090 return s->s_bdev == sd->bdev;
1093 static int nilfs_test_bdev_super2(struct super_block *s, void *data)
1095 struct nilfs_super_data *sd = data;
1098 if (s->s_bdev != sd->bdev)
1101 if (!((s->s_flags | sd->flags) & MS_RDONLY))
1102 return 1; /* Reuse an old R/W-mode super_block */
1104 if (s->s_flags & sd->flags & MS_RDONLY) {
1105 if (down_read_trylock(&s->s_umount)) {
1107 (sd->cno == NILFS_SB(s)->s_snapshot_cno);
1108 up_read(&s->s_umount);
1110 * This path is locked with sb_lock by sget().
1111 * So, drop_super() causes deadlock.
1120 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1121 const char *dev_name, void *data, struct vfsmount *mnt)
1123 struct nilfs_super_data sd;
1124 struct super_block *s, *s2;
1125 struct the_nilfs *nilfs = NULL;
1126 int err, need_to_close = 1;
1128 sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1129 if (IS_ERR(sd.bdev))
1130 return PTR_ERR(sd.bdev);
1133 * To get mount instance using sget() vfs-routine, NILFS needs
1134 * much more information than normal filesystems to identify mount
1135 * instance. For snapshot mounts, not only a mount type (ro-mount
1136 * or rw-mount) but also a checkpoint number is required.
1137 * The results are passed in sget() using nilfs_super_data.
1141 if (nilfs_identify((char *)data, &sd)) {
1147 * once the super is inserted into the list by sget, s_umount
1148 * will protect the lockfs code from trying to start a snapshot
1149 * while we are mounting
1151 down(&sd.bdev->bd_mount_sem);
1153 (err = test_exclusive_mount(fs_type, sd.bdev, flags ^ MS_RDONLY))) {
1154 err = (err < 0) ? : -EBUSY;
1159 * Phase-1: search any existent instance and get the_nilfs
1161 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1167 nilfs = alloc_nilfs(sd.bdev);
1171 struct nilfs_sb_info *sbi = NILFS_SB(s);
1173 BUG_ON(!sbi || !sbi->s_nilfs);
1175 * s_umount protects super_block from unmount process;
1176 * It covers pointers of nilfs_sb_info and the_nilfs.
1178 nilfs = sbi->s_nilfs;
1180 up_write(&s->s_umount);
1183 * Phase-2: search specified snapshot or R/W mode super_block
1186 /* trying to get the latest checkpoint. */
1187 sd.cno = nilfs_last_cno(nilfs);
1189 s2 = sget(fs_type, nilfs_test_bdev_super2,
1190 nilfs_set_bdev_super, &sd);
1191 deactivate_super(s);
1193 * Although deactivate_super() invokes close_bdev_exclusive() at
1194 * kill_block_super(). Here, s is an existent mount; we need
1195 * one more close_bdev_exclusive() call.
1203 char b[BDEVNAME_SIZE];
1206 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1207 sb_set_blocksize(s, block_size(sd.bdev));
1209 err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1213 s->s_flags |= MS_ACTIVE;
1215 } else if (!(s->s_flags & MS_RDONLY)) {
1219 up(&sd.bdev->bd_mount_sem);
1222 close_bdev_exclusive(sd.bdev, flags);
1223 simple_set_mnt(mnt, s);
1227 up(&sd.bdev->bd_mount_sem);
1230 close_bdev_exclusive(sd.bdev, flags);
1234 up(&sd.bdev->bd_mount_sem);
1236 close_bdev_exclusive(sd.bdev, flags);
1241 /* Abandoning the newly allocated superblock */
1242 up(&sd.bdev->bd_mount_sem);
1245 up_write(&s->s_umount);
1246 deactivate_super(s);
1248 * deactivate_super() invokes close_bdev_exclusive().
1249 * We must finish all post-cleaning before this call;
1250 * put_nilfs() and unlocking bd_mount_sem need the block device.
1255 static int nilfs_test_bdev_super3(struct super_block *s, void *data)
1257 struct nilfs_super_data *sd = data;
1260 if (s->s_bdev != sd->bdev)
1262 if (down_read_trylock(&s->s_umount)) {
1263 ret = (s->s_flags & MS_RDONLY) && s->s_root &&
1264 nilfs_test_opt(NILFS_SB(s), SNAPSHOT);
1265 up_read(&s->s_umount);
1267 return 0; /* ignore snapshot mounts */
1269 return !((sd->flags ^ s->s_flags) & MS_RDONLY);
1272 static int __false_bdev_super(struct super_block *s, void *data)
1274 #if 0 /* XXX: workaround for lock debug. This is not good idea */
1275 up_write(&s->s_umount);
1281 * test_exclusive_mount - check whether an exclusive RW/RO mount exists or not.
1282 * fs_type: filesystem type
1283 * bdev: block device
1284 * flag: 0 (check rw-mount) or MS_RDONLY (check ro-mount)
1285 * res: pointer to an integer to store result
1287 * This function must be called within a section protected by bd_mount_mutex.
1289 static int test_exclusive_mount(struct file_system_type *fs_type,
1290 struct block_device *bdev, int flags)
1292 struct super_block *s;
1293 struct nilfs_super_data sd = { .flags = flags, .bdev = bdev };
1295 s = sget(fs_type, nilfs_test_bdev_super3, __false_bdev_super, &sd);
1297 if (PTR_ERR(s) != -EFAULT)
1299 return 0; /* Not found */
1301 up_write(&s->s_umount);
1302 deactivate_super(s);
1303 return 1; /* Found */
1306 struct file_system_type nilfs_fs_type = {
1307 .owner = THIS_MODULE,
1309 .get_sb = nilfs_get_sb,
1310 .kill_sb = kill_block_super,
1311 .fs_flags = FS_REQUIRES_DEV,
1314 static int __init init_nilfs_fs(void)
1318 err = nilfs_init_inode_cache();
1322 err = nilfs_init_transaction_cache();
1324 goto failed_inode_cache;
1326 err = nilfs_init_segbuf_cache();
1328 goto failed_transaction_cache;
1330 err = nilfs_btree_path_cache_init();
1332 goto failed_segbuf_cache;
1334 err = register_filesystem(&nilfs_fs_type);
1336 goto failed_btree_path_cache;
1340 failed_btree_path_cache:
1341 nilfs_btree_path_cache_destroy();
1343 failed_segbuf_cache:
1344 nilfs_destroy_segbuf_cache();
1346 failed_transaction_cache:
1347 nilfs_destroy_transaction_cache();
1350 nilfs_destroy_inode_cache();
1356 static void __exit exit_nilfs_fs(void)
1358 nilfs_destroy_segbuf_cache();
1359 nilfs_destroy_transaction_cache();
1360 nilfs_destroy_inode_cache();
1361 nilfs_btree_path_cache_destroy();
1362 unregister_filesystem(&nilfs_fs_type);
1365 module_init(init_nilfs_fs)
1366 module_exit(exit_nilfs_fs)