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_LICENSE("GPL");
68 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
69 static int test_exclusive_mount(struct file_system_type *fs_type,
70 struct block_device *bdev, int flags);
73 * nilfs_error() - report failure condition on a filesystem
75 * nilfs_error() sets an ERROR_FS flag on the superblock as well as
76 * reporting an error message. It should be called when NILFS detects
77 * incoherences or defects of meta data on disk. As for sustainable
78 * errors such as a single-shot I/O error, nilfs_warning() or the printk()
79 * function should be used instead.
81 * The segment constructor must not call this function because it can
84 void nilfs_error(struct super_block *sb, const char *function,
87 struct nilfs_sb_info *sbi = NILFS_SB(sb);
91 printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
96 if (!(sb->s_flags & MS_RDONLY)) {
97 struct the_nilfs *nilfs = sbi->s_nilfs;
99 if (!nilfs_test_opt(sbi, ERRORS_CONT))
100 nilfs_detach_segment_constructor(sbi);
102 down_write(&nilfs->ns_sem);
103 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
104 nilfs->ns_mount_state |= NILFS_ERROR_FS;
105 nilfs->ns_sbp[0]->s_state |=
106 cpu_to_le16(NILFS_ERROR_FS);
107 nilfs_commit_super(sbi, 1);
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_common(struct the_nilfs *nilfs)
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, nilfs->ns_bdi);
148 return &ii->vfs_inode;
151 struct inode *nilfs_alloc_inode(struct super_block *sb)
153 return nilfs_alloc_inode_common(NILFS_SB(sb)->s_nilfs);
156 void nilfs_destroy_inode(struct inode *inode)
158 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
161 static void init_once(void *obj)
163 struct nilfs_inode_info *ii = obj;
165 INIT_LIST_HEAD(&ii->i_dirty);
166 #ifdef CONFIG_NILFS_XATTR
167 init_rwsem(&ii->xattr_sem);
169 nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
170 ii->i_bmap = (struct nilfs_bmap *)&ii->i_bmap_union;
171 inode_init_once(&ii->vfs_inode);
174 static int nilfs_init_inode_cache(void)
176 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
177 sizeof(struct nilfs_inode_info),
178 0, SLAB_RECLAIM_ACCOUNT,
181 return (nilfs_inode_cachep == NULL) ? -ENOMEM : 0;
184 static inline void nilfs_destroy_inode_cache(void)
186 kmem_cache_destroy(nilfs_inode_cachep);
189 static void nilfs_clear_inode(struct inode *inode)
191 struct nilfs_inode_info *ii = NILFS_I(inode);
193 #ifdef CONFIG_NILFS_POSIX_ACL
194 if (ii->i_acl && ii->i_acl != NILFS_ACL_NOT_CACHED) {
195 posix_acl_release(ii->i_acl);
196 ii->i_acl = NILFS_ACL_NOT_CACHED;
198 if (ii->i_default_acl && ii->i_default_acl != NILFS_ACL_NOT_CACHED) {
199 posix_acl_release(ii->i_default_acl);
200 ii->i_default_acl = NILFS_ACL_NOT_CACHED;
204 * Free resources allocated in nilfs_read_inode(), here.
206 BUG_ON(!list_empty(&ii->i_dirty));
210 if (test_bit(NILFS_I_BMAP, &ii->i_state))
211 nilfs_bmap_clear(ii->i_bmap);
213 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
216 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int dupsb)
218 struct the_nilfs *nilfs = sbi->s_nilfs;
220 int barrier_done = 0;
222 if (nilfs_test_opt(sbi, BARRIER)) {
223 set_buffer_ordered(nilfs->ns_sbh[0]);
227 set_buffer_dirty(nilfs->ns_sbh[0]);
228 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
229 if (err == -EOPNOTSUPP && barrier_done) {
230 nilfs_warning(sbi->s_super, __func__,
231 "barrier-based sync failed. "
232 "disabling barriers\n");
233 nilfs_clear_opt(sbi, BARRIER);
235 clear_buffer_ordered(nilfs->ns_sbh[0]);
240 "NILFS: unable to write superblock (err=%d)\n", err);
241 if (err == -EIO && nilfs->ns_sbh[1]) {
242 nilfs_fall_back_super_block(nilfs);
246 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
249 * The latest segment becomes trailable from the position
250 * written in superblock.
252 clear_nilfs_discontinued(nilfs);
254 /* update GC protection for recent segments */
255 if (nilfs->ns_sbh[1]) {
258 set_buffer_dirty(nilfs->ns_sbh[1]);
259 if (!sync_dirty_buffer(nilfs->ns_sbh[1]))
260 sbp = nilfs->ns_sbp[1];
264 spin_lock(&nilfs->ns_last_segment_lock);
265 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
266 spin_unlock(&nilfs->ns_last_segment_lock);
273 int nilfs_commit_super(struct nilfs_sb_info *sbi, int dupsb)
275 struct the_nilfs *nilfs = sbi->s_nilfs;
276 struct nilfs_super_block **sbp = nilfs->ns_sbp;
277 sector_t nfreeblocks;
281 /* nilfs->sem must be locked by the caller. */
282 if (sbp[0]->s_magic != NILFS_SUPER_MAGIC) {
283 if (sbp[1] && sbp[1]->s_magic == NILFS_SUPER_MAGIC)
284 nilfs_swap_super_block(nilfs);
286 printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
291 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
293 printk(KERN_ERR "NILFS: failed to count free blocks\n");
296 spin_lock(&nilfs->ns_last_segment_lock);
297 sbp[0]->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
298 sbp[0]->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
299 sbp[0]->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
300 spin_unlock(&nilfs->ns_last_segment_lock);
303 nilfs->ns_sbwtime[0] = t;
304 sbp[0]->s_free_blocks_count = cpu_to_le64(nfreeblocks);
305 sbp[0]->s_wtime = cpu_to_le64(t);
307 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
308 (unsigned char *)sbp[0],
310 if (dupsb && sbp[1]) {
311 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
312 nilfs->ns_sbwtime[1] = t;
314 sbi->s_super->s_dirt = 0;
315 return nilfs_sync_super(sbi, dupsb);
318 static void nilfs_put_super(struct super_block *sb)
320 struct nilfs_sb_info *sbi = NILFS_SB(sb);
321 struct the_nilfs *nilfs = sbi->s_nilfs;
323 nilfs_detach_segment_constructor(sbi);
325 if (!(sb->s_flags & MS_RDONLY)) {
326 down_write(&nilfs->ns_sem);
327 nilfs->ns_sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
328 nilfs_commit_super(sbi, 1);
329 up_write(&nilfs->ns_sem);
332 nilfs_detach_checkpoint(sbi);
333 put_nilfs(sbi->s_nilfs);
335 sb->s_fs_info = NULL;
340 * nilfs_write_super - write super block(s) of NILFS
343 * nilfs_write_super() gets a fs-dependent lock, writes super block(s), and
344 * clears s_dirt. This function is called in the section protected by
347 * The s_dirt flag is managed by each filesystem and we protect it by ns_sem
348 * of the struct the_nilfs. Lock order must be as follows:
351 * 2. down_write(&nilfs->ns_sem)
353 * Inside NILFS, locking ns_sem is enough to protect s_dirt and the buffer
354 * of the super block (nilfs->ns_sbp[]).
356 * In most cases, VFS functions call lock_super() before calling these
357 * methods. So we must be careful not to bring on deadlocks when using
358 * lock_super(); see generic_shutdown_super(), write_super(), and so on.
360 * Note that order of lock_kernel() and lock_super() depends on contexts
361 * of VFS. We should also note that lock_kernel() can be used in its
362 * protective section and only the outermost one has an effect.
364 static void nilfs_write_super(struct super_block *sb)
366 struct nilfs_sb_info *sbi = NILFS_SB(sb);
367 struct the_nilfs *nilfs = sbi->s_nilfs;
369 down_write(&nilfs->ns_sem);
370 if (!(sb->s_flags & MS_RDONLY)) {
371 struct nilfs_super_block **sbp = nilfs->ns_sbp;
372 u64 t = get_seconds();
375 if (!nilfs_discontinued(nilfs) && t >= nilfs->ns_sbwtime[0] &&
376 t < nilfs->ns_sbwtime[0] + NILFS_SB_FREQ) {
377 up_write(&nilfs->ns_sem);
380 dupsb = sbp[1] && t > nilfs->ns_sbwtime[1] + NILFS_ALTSB_FREQ;
381 nilfs_commit_super(sbi, dupsb);
384 up_write(&nilfs->ns_sem);
387 static int nilfs_sync_fs(struct super_block *sb, int wait)
391 /* This function is called when super block should be written back */
393 err = nilfs_construct_segment(sb);
397 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
399 struct the_nilfs *nilfs = sbi->s_nilfs;
400 struct nilfs_checkpoint *raw_cp;
401 struct buffer_head *bh_cp;
404 down_write(&nilfs->ns_sem);
405 list_add(&sbi->s_list, &nilfs->ns_supers);
406 up_write(&nilfs->ns_sem);
408 sbi->s_ifile = nilfs_mdt_new(
409 nilfs, sbi->s_super, NILFS_IFILE_INO, NILFS_IFILE_GFP);
413 err = nilfs_palloc_init_blockgroup(sbi->s_ifile, nilfs->ns_inode_size);
417 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
420 if (err == -ENOENT || err == -EINVAL) {
422 "NILFS: Invalid checkpoint "
423 "(checkpoint number=%llu)\n",
424 (unsigned long long)cno);
429 err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
432 atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
433 atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
435 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
439 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
441 nilfs_mdt_destroy(sbi->s_ifile);
444 down_write(&nilfs->ns_sem);
445 list_del_init(&sbi->s_list);
446 up_write(&nilfs->ns_sem);
451 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
453 struct the_nilfs *nilfs = sbi->s_nilfs;
455 nilfs_mdt_clear(sbi->s_ifile);
456 nilfs_mdt_destroy(sbi->s_ifile);
458 down_write(&nilfs->ns_sem);
459 list_del_init(&sbi->s_list);
460 up_write(&nilfs->ns_sem);
463 static int nilfs_mark_recovery_complete(struct nilfs_sb_info *sbi)
465 struct the_nilfs *nilfs = sbi->s_nilfs;
468 down_write(&nilfs->ns_sem);
469 if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
470 nilfs->ns_mount_state |= NILFS_VALID_FS;
471 err = nilfs_commit_super(sbi, 1);
473 printk(KERN_INFO "NILFS: recovery complete.\n");
475 up_write(&nilfs->ns_sem);
479 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
481 struct super_block *sb = dentry->d_sb;
482 struct nilfs_sb_info *sbi = NILFS_SB(sb);
483 struct the_nilfs *nilfs = sbi->s_nilfs;
484 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
485 unsigned long long blocks;
486 unsigned long overhead;
487 unsigned long nrsvblocks;
488 sector_t nfreeblocks;
492 * Compute all of the segment blocks
494 * The blocks before first segment and after last segment
497 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
498 - nilfs->ns_first_data_block;
499 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
502 * Compute the overhead
504 * When distributing meta data blocks outside semgent structure,
505 * We must count them as the overhead.
509 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
513 buf->f_type = NILFS_SUPER_MAGIC;
514 buf->f_bsize = sb->s_blocksize;
515 buf->f_blocks = blocks - overhead;
516 buf->f_bfree = nfreeblocks;
517 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
518 (buf->f_bfree - nrsvblocks) : 0;
519 buf->f_files = atomic_read(&sbi->s_inodes_count);
520 buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
521 buf->f_namelen = NILFS_NAME_LEN;
522 buf->f_fsid.val[0] = (u32)id;
523 buf->f_fsid.val[1] = (u32)(id >> 32);
528 static struct super_operations nilfs_sops = {
529 .alloc_inode = nilfs_alloc_inode,
530 .destroy_inode = nilfs_destroy_inode,
531 .dirty_inode = nilfs_dirty_inode,
532 /* .write_inode = nilfs_write_inode, */
533 /* .put_inode = nilfs_put_inode, */
534 /* .drop_inode = nilfs_drop_inode, */
535 .delete_inode = nilfs_delete_inode,
536 .put_super = nilfs_put_super,
537 .write_super = nilfs_write_super,
538 .sync_fs = nilfs_sync_fs,
539 /* .write_super_lockfs */
541 .statfs = nilfs_statfs,
542 .remount_fs = nilfs_remount,
543 .clear_inode = nilfs_clear_inode,
548 static struct inode *
549 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
553 if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
554 ino != NILFS_SKETCH_INO)
555 return ERR_PTR(-ESTALE);
557 inode = nilfs_iget(sb, ino);
559 return ERR_CAST(inode);
560 if (generation && inode->i_generation != generation) {
562 return ERR_PTR(-ESTALE);
568 static struct dentry *
569 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
572 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
573 nilfs_nfs_get_inode);
576 static struct dentry *
577 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
580 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
581 nilfs_nfs_get_inode);
584 static struct export_operations nilfs_export_ops = {
585 .fh_to_dentry = nilfs_fh_to_dentry,
586 .fh_to_parent = nilfs_fh_to_parent,
587 .get_parent = nilfs_get_parent,
591 Opt_err_cont, Opt_err_panic, Opt_err_ro,
592 Opt_barrier, Opt_snapshot, Opt_order,
596 static match_table_t tokens = {
597 {Opt_err_cont, "errors=continue"},
598 {Opt_err_panic, "errors=panic"},
599 {Opt_err_ro, "errors=remount-ro"},
600 {Opt_barrier, "barrier=%s"},
601 {Opt_snapshot, "cp=%u"},
602 {Opt_order, "order=%s"},
606 static int match_bool(substring_t *s, int *result)
608 int len = s->to - s->from;
610 if (strncmp(s->from, "on", len) == 0)
612 else if (strncmp(s->from, "off", len) == 0)
619 static int parse_options(char *options, struct super_block *sb)
621 struct nilfs_sb_info *sbi = NILFS_SB(sb);
623 substring_t args[MAX_OPT_ARGS];
629 while ((p = strsep(&options, ",")) != NULL) {
634 token = match_token(p, tokens, args);
637 if (match_bool(&args[0], &option))
640 nilfs_set_opt(sbi, BARRIER);
642 nilfs_clear_opt(sbi, BARRIER);
645 if (strcmp(args[0].from, "relaxed") == 0)
646 /* Ordered data semantics */
647 nilfs_clear_opt(sbi, STRICT_ORDER);
648 else if (strcmp(args[0].from, "strict") == 0)
649 /* Strict in-order semantics */
650 nilfs_set_opt(sbi, STRICT_ORDER);
655 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
658 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
661 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
664 if (match_int(&args[0], &option) || option <= 0)
666 if (!(sb->s_flags & MS_RDONLY))
668 sbi->s_snapshot_cno = option;
669 nilfs_set_opt(sbi, SNAPSHOT);
673 "NILFS: Unrecognized mount option \"%s\"\n", p);
681 nilfs_set_default_options(struct nilfs_sb_info *sbi,
682 struct nilfs_super_block *sbp)
685 NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER;
688 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
690 struct the_nilfs *nilfs = sbi->s_nilfs;
691 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
692 int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
693 int mnt_count = le16_to_cpu(sbp->s_mnt_count);
695 /* nilfs->sem must be locked by the caller. */
696 if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
697 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
698 } else if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
700 "NILFS warning: mounting fs with errors\n");
702 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
704 "NILFS warning: maximal mount count reached\n");
708 sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
710 sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
711 sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
712 sbp->s_mtime = cpu_to_le64(get_seconds());
713 return nilfs_commit_super(sbi, 1);
716 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
717 u64 pos, int blocksize,
718 struct buffer_head **pbh)
720 unsigned long long sb_index = pos;
721 unsigned long offset;
723 offset = do_div(sb_index, blocksize);
724 *pbh = sb_bread(sb, sb_index);
727 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
730 int nilfs_store_magic_and_option(struct super_block *sb,
731 struct nilfs_super_block *sbp,
734 struct nilfs_sb_info *sbi = NILFS_SB(sb);
736 sb->s_magic = le16_to_cpu(sbp->s_magic);
738 /* FS independent flags */
739 #ifdef NILFS_ATIME_DISABLE
740 sb->s_flags |= MS_NOATIME;
743 nilfs_set_default_options(sbi, sbp);
745 sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
746 sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
747 sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
748 sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
750 return !parse_options(data, sb) ? -EINVAL : 0 ;
754 * nilfs_fill_super() - initialize a super block instance
756 * @data: mount options
757 * @silent: silent mode flag
758 * @nilfs: the_nilfs struct
760 * This function is called exclusively by bd_mount_mutex.
761 * So, the recovery process is protected from other simultaneous mounts.
764 nilfs_fill_super(struct super_block *sb, void *data, int silent,
765 struct the_nilfs *nilfs)
767 struct nilfs_sb_info *sbi;
772 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
779 sbi->s_nilfs = nilfs;
782 err = init_nilfs(nilfs, sbi, (char *)data);
786 spin_lock_init(&sbi->s_inode_lock);
787 INIT_LIST_HEAD(&sbi->s_dirty_files);
788 INIT_LIST_HEAD(&sbi->s_list);
791 * Following initialization is overlapped because
792 * nilfs_sb_info structure has been cleared at the beginning.
793 * But we reserve them to keep our interest and make ready
794 * for the future change.
796 get_random_bytes(&sbi->s_next_generation,
797 sizeof(sbi->s_next_generation));
798 spin_lock_init(&sbi->s_next_gen_lock);
800 sb->s_op = &nilfs_sops;
801 sb->s_export_op = &nilfs_export_ops;
805 if (!nilfs_loaded(nilfs)) {
806 err = load_nilfs(nilfs, sbi);
810 cno = nilfs_last_cno(nilfs);
812 if (sb->s_flags & MS_RDONLY) {
813 if (nilfs_test_opt(sbi, SNAPSHOT)) {
814 err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
815 sbi->s_snapshot_cno);
820 "NILFS: The specified checkpoint is "
822 "(checkpoint number=%llu).\n",
823 (unsigned long long)sbi->s_snapshot_cno);
827 cno = sbi->s_snapshot_cno;
829 /* Read-only mount */
830 sbi->s_snapshot_cno = cno;
833 err = nilfs_attach_checkpoint(sbi, cno);
835 printk(KERN_ERR "NILFS: error loading a checkpoint"
836 " (checkpoint number=%llu).\n", (unsigned long long)cno);
840 if (!(sb->s_flags & MS_RDONLY)) {
841 err = nilfs_attach_segment_constructor(sbi);
843 goto failed_checkpoint;
846 root = nilfs_iget(sb, NILFS_ROOT_INO);
848 printk(KERN_ERR "NILFS: get root inode failed\n");
852 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
854 printk(KERN_ERR "NILFS: corrupt root inode.\n");
858 sb->s_root = d_alloc_root(root);
861 printk(KERN_ERR "NILFS: get root dentry failed\n");
866 if (!(sb->s_flags & MS_RDONLY)) {
867 down_write(&nilfs->ns_sem);
868 nilfs_setup_super(sbi);
869 up_write(&nilfs->ns_sem);
872 err = nilfs_mark_recovery_complete(sbi);
874 printk(KERN_ERR "NILFS: recovery failed.\n");
885 nilfs_detach_segment_constructor(sbi);
888 nilfs_detach_checkpoint(sbi);
892 sb->s_fs_info = NULL;
897 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
899 struct nilfs_sb_info *sbi = NILFS_SB(sb);
900 struct nilfs_super_block *sbp;
901 struct the_nilfs *nilfs = sbi->s_nilfs;
902 unsigned long old_sb_flags;
903 struct nilfs_mount_options old_opts;
906 old_sb_flags = sb->s_flags;
907 old_opts.mount_opt = sbi->s_mount_opt;
908 old_opts.snapshot_cno = sbi->s_snapshot_cno;
910 if (!parse_options(data, sb)) {
914 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
916 if ((*flags & MS_RDONLY) &&
917 sbi->s_snapshot_cno != old_opts.snapshot_cno) {
918 printk(KERN_WARNING "NILFS (device %s): couldn't "
919 "remount to a different snapshot. \n",
925 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
927 if (*flags & MS_RDONLY) {
928 /* Shutting down the segment constructor */
929 nilfs_detach_segment_constructor(sbi);
930 sb->s_flags |= MS_RDONLY;
932 sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
933 /* nilfs_set_opt(sbi, SNAPSHOT); */
936 * Remounting a valid RW partition RDONLY, so set
937 * the RDONLY flag and then mark the partition as valid again.
939 down_write(&nilfs->ns_sem);
940 sbp = nilfs->ns_sbp[0];
941 if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
942 (nilfs->ns_mount_state & NILFS_VALID_FS))
943 sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
944 sbp->s_mtime = cpu_to_le64(get_seconds());
945 nilfs_commit_super(sbi, 1);
946 up_write(&nilfs->ns_sem);
949 * Mounting a RDONLY partition read-write, so reread and
950 * store the current valid flag. (It may have been changed
951 * by fsck since we originally mounted the partition.)
953 down(&sb->s_bdev->bd_mount_sem);
954 /* Check existing RW-mount */
955 if (test_exclusive_mount(sb->s_type, sb->s_bdev, 0)) {
956 printk(KERN_WARNING "NILFS (device %s): couldn't "
957 "remount because a RW-mount exists.\n",
960 goto rw_remount_failed;
962 if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
963 printk(KERN_WARNING "NILFS (device %s): couldn't "
964 "remount because the current RO-mount is not "
968 goto rw_remount_failed;
970 sb->s_flags &= ~MS_RDONLY;
971 nilfs_clear_opt(sbi, SNAPSHOT);
972 sbi->s_snapshot_cno = 0;
974 err = nilfs_attach_segment_constructor(sbi);
976 goto rw_remount_failed;
978 down_write(&nilfs->ns_sem);
979 nilfs_setup_super(sbi);
980 up_write(&nilfs->ns_sem);
982 up(&sb->s_bdev->bd_mount_sem);
988 up(&sb->s_bdev->bd_mount_sem);
990 sb->s_flags = old_sb_flags;
991 sbi->s_mount_opt = old_opts.mount_opt;
992 sbi->s_snapshot_cno = old_opts.snapshot_cno;
996 struct nilfs_super_data {
997 struct block_device *bdev;
1003 * nilfs_identify - pre-read mount options needed to identify mount instance
1004 * @data: mount options
1005 * @sd: nilfs_super_data
1007 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1009 char *p, *options = data;
1010 substring_t args[MAX_OPT_ARGS];
1015 p = strsep(&options, ",");
1016 if (p != NULL && *p) {
1017 token = match_token(p, tokens, args);
1018 if (token == Opt_snapshot) {
1019 if (!(sd->flags & MS_RDONLY))
1022 ret = match_int(&args[0], &option);
1033 "NILFS: invalid mount option: %s\n", p);
1037 BUG_ON(options == data);
1038 *(options - 1) = ',';
1043 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1045 struct nilfs_super_data *sd = data;
1047 s->s_bdev = sd->bdev;
1048 s->s_dev = s->s_bdev->bd_dev;
1052 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1054 struct nilfs_super_data *sd = data;
1056 return s->s_bdev == sd->bdev;
1059 static int nilfs_test_bdev_super2(struct super_block *s, void *data)
1061 struct nilfs_super_data *sd = data;
1064 if (s->s_bdev != sd->bdev)
1067 if (!((s->s_flags | sd->flags) & MS_RDONLY))
1068 return 1; /* Reuse an old R/W-mode super_block */
1070 if (s->s_flags & sd->flags & MS_RDONLY) {
1071 if (down_read_trylock(&s->s_umount)) {
1073 (sd->cno == NILFS_SB(s)->s_snapshot_cno);
1074 up_read(&s->s_umount);
1076 * This path is locked with sb_lock by sget().
1077 * So, drop_super() causes deadlock.
1086 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1087 const char *dev_name, void *data, struct vfsmount *mnt)
1089 struct nilfs_super_data sd;
1090 struct super_block *s, *s2;
1091 struct the_nilfs *nilfs = NULL;
1092 int err, need_to_close = 1;
1094 sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1095 if (IS_ERR(sd.bdev))
1096 return PTR_ERR(sd.bdev);
1099 * To get mount instance using sget() vfs-routine, NILFS needs
1100 * much more information than normal filesystems to identify mount
1101 * instance. For snapshot mounts, not only a mount type (ro-mount
1102 * or rw-mount) but also a checkpoint number is required.
1103 * The results are passed in sget() using nilfs_super_data.
1107 if (nilfs_identify((char *)data, &sd)) {
1113 * once the super is inserted into the list by sget, s_umount
1114 * will protect the lockfs code from trying to start a snapshot
1115 * while we are mounting
1117 down(&sd.bdev->bd_mount_sem);
1119 (err = test_exclusive_mount(fs_type, sd.bdev, flags ^ MS_RDONLY))) {
1120 err = (err < 0) ? : -EBUSY;
1125 * Phase-1: search any existent instance and get the_nilfs
1127 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1133 nilfs = alloc_nilfs(sd.bdev);
1137 struct nilfs_sb_info *sbi = NILFS_SB(s);
1140 * s_umount protects super_block from unmount process;
1141 * It covers pointers of nilfs_sb_info and the_nilfs.
1143 nilfs = sbi->s_nilfs;
1145 up_write(&s->s_umount);
1148 * Phase-2: search specified snapshot or R/W mode super_block
1151 /* trying to get the latest checkpoint. */
1152 sd.cno = nilfs_last_cno(nilfs);
1154 s2 = sget(fs_type, nilfs_test_bdev_super2,
1155 nilfs_set_bdev_super, &sd);
1156 deactivate_super(s);
1158 * Although deactivate_super() invokes close_bdev_exclusive() at
1159 * kill_block_super(). Here, s is an existent mount; we need
1160 * one more close_bdev_exclusive() call.
1168 char b[BDEVNAME_SIZE];
1171 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1172 sb_set_blocksize(s, block_size(sd.bdev));
1174 err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1178 s->s_flags |= MS_ACTIVE;
1180 } else if (!(s->s_flags & MS_RDONLY)) {
1184 up(&sd.bdev->bd_mount_sem);
1187 close_bdev_exclusive(sd.bdev, flags);
1188 simple_set_mnt(mnt, s);
1192 up(&sd.bdev->bd_mount_sem);
1195 close_bdev_exclusive(sd.bdev, flags);
1199 up(&sd.bdev->bd_mount_sem);
1201 close_bdev_exclusive(sd.bdev, flags);
1206 /* Abandoning the newly allocated superblock */
1207 up(&sd.bdev->bd_mount_sem);
1210 up_write(&s->s_umount);
1211 deactivate_super(s);
1213 * deactivate_super() invokes close_bdev_exclusive().
1214 * We must finish all post-cleaning before this call;
1215 * put_nilfs() and unlocking bd_mount_sem need the block device.
1220 static int nilfs_test_bdev_super3(struct super_block *s, void *data)
1222 struct nilfs_super_data *sd = data;
1225 if (s->s_bdev != sd->bdev)
1227 if (down_read_trylock(&s->s_umount)) {
1228 ret = (s->s_flags & MS_RDONLY) && s->s_root &&
1229 nilfs_test_opt(NILFS_SB(s), SNAPSHOT);
1230 up_read(&s->s_umount);
1232 return 0; /* ignore snapshot mounts */
1234 return !((sd->flags ^ s->s_flags) & MS_RDONLY);
1237 static int __false_bdev_super(struct super_block *s, void *data)
1239 #if 0 /* XXX: workaround for lock debug. This is not good idea */
1240 up_write(&s->s_umount);
1246 * test_exclusive_mount - check whether an exclusive RW/RO mount exists or not.
1247 * fs_type: filesystem type
1248 * bdev: block device
1249 * flag: 0 (check rw-mount) or MS_RDONLY (check ro-mount)
1250 * res: pointer to an integer to store result
1252 * This function must be called within a section protected by bd_mount_mutex.
1254 static int test_exclusive_mount(struct file_system_type *fs_type,
1255 struct block_device *bdev, int flags)
1257 struct super_block *s;
1258 struct nilfs_super_data sd = { .flags = flags, .bdev = bdev };
1260 s = sget(fs_type, nilfs_test_bdev_super3, __false_bdev_super, &sd);
1262 if (PTR_ERR(s) != -EFAULT)
1264 return 0; /* Not found */
1266 up_write(&s->s_umount);
1267 deactivate_super(s);
1268 return 1; /* Found */
1271 struct file_system_type nilfs_fs_type = {
1272 .owner = THIS_MODULE,
1274 .get_sb = nilfs_get_sb,
1275 .kill_sb = kill_block_super,
1276 .fs_flags = FS_REQUIRES_DEV,
1279 static int __init init_nilfs_fs(void)
1283 err = nilfs_init_inode_cache();
1287 err = nilfs_init_transaction_cache();
1289 goto failed_inode_cache;
1291 err = nilfs_init_segbuf_cache();
1293 goto failed_transaction_cache;
1295 err = nilfs_btree_path_cache_init();
1297 goto failed_segbuf_cache;
1299 err = register_filesystem(&nilfs_fs_type);
1301 goto failed_btree_path_cache;
1305 failed_btree_path_cache:
1306 nilfs_btree_path_cache_destroy();
1308 failed_segbuf_cache:
1309 nilfs_destroy_segbuf_cache();
1311 failed_transaction_cache:
1312 nilfs_destroy_transaction_cache();
1315 nilfs_destroy_inode_cache();
1321 static void __exit exit_nilfs_fs(void)
1323 nilfs_destroy_segbuf_cache();
1324 nilfs_destroy_transaction_cache();
1325 nilfs_destroy_inode_cache();
1326 nilfs_btree_path_cache_destroy();
1327 unregister_filesystem(&nilfs_fs_type);
1330 module_init(init_nilfs_fs)
1331 module_exit(exit_nilfs_fs)