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(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);
188 #ifdef CONFIG_NILFS_POSIX_ACL
189 if (ii->i_acl && ii->i_acl != NILFS_ACL_NOT_CACHED) {
190 posix_acl_release(ii->i_acl);
191 ii->i_acl = NILFS_ACL_NOT_CACHED;
193 if (ii->i_default_acl && ii->i_default_acl != NILFS_ACL_NOT_CACHED) {
194 posix_acl_release(ii->i_default_acl);
195 ii->i_default_acl = NILFS_ACL_NOT_CACHED;
199 * Free resources allocated in nilfs_read_inode(), here.
201 BUG_ON(!list_empty(&ii->i_dirty));
205 if (test_bit(NILFS_I_BMAP, &ii->i_state))
206 nilfs_bmap_clear(ii->i_bmap);
208 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
211 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int dupsb)
213 struct the_nilfs *nilfs = sbi->s_nilfs;
215 int barrier_done = 0;
217 if (nilfs_test_opt(sbi, BARRIER)) {
218 set_buffer_ordered(nilfs->ns_sbh[0]);
222 set_buffer_dirty(nilfs->ns_sbh[0]);
223 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
224 if (err == -EOPNOTSUPP && barrier_done) {
225 nilfs_warning(sbi->s_super, __func__,
226 "barrier-based sync failed. "
227 "disabling barriers\n");
228 nilfs_clear_opt(sbi, BARRIER);
230 clear_buffer_ordered(nilfs->ns_sbh[0]);
235 "NILFS: unable to write superblock (err=%d)\n", err);
236 if (err == -EIO && nilfs->ns_sbh[1]) {
237 nilfs_fall_back_super_block(nilfs);
241 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
244 * The latest segment becomes trailable from the position
245 * written in superblock.
247 clear_nilfs_discontinued(nilfs);
249 /* update GC protection for recent segments */
250 if (nilfs->ns_sbh[1]) {
253 set_buffer_dirty(nilfs->ns_sbh[1]);
254 if (!sync_dirty_buffer(nilfs->ns_sbh[1]))
255 sbp = nilfs->ns_sbp[1];
259 spin_lock(&nilfs->ns_last_segment_lock);
260 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
261 spin_unlock(&nilfs->ns_last_segment_lock);
268 int nilfs_commit_super(struct nilfs_sb_info *sbi, int dupsb)
270 struct the_nilfs *nilfs = sbi->s_nilfs;
271 struct nilfs_super_block **sbp = nilfs->ns_sbp;
272 sector_t nfreeblocks;
276 /* nilfs->sem must be locked by the caller. */
277 if (sbp[0]->s_magic != NILFS_SUPER_MAGIC) {
278 if (sbp[1] && sbp[1]->s_magic == NILFS_SUPER_MAGIC)
279 nilfs_swap_super_block(nilfs);
281 printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
286 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
288 printk(KERN_ERR "NILFS: failed to count free blocks\n");
291 spin_lock(&nilfs->ns_last_segment_lock);
292 sbp[0]->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
293 sbp[0]->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
294 sbp[0]->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
295 spin_unlock(&nilfs->ns_last_segment_lock);
298 nilfs->ns_sbwtime[0] = t;
299 sbp[0]->s_free_blocks_count = cpu_to_le64(nfreeblocks);
300 sbp[0]->s_wtime = cpu_to_le64(t);
302 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
303 (unsigned char *)sbp[0],
305 if (dupsb && sbp[1]) {
306 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
307 nilfs->ns_sbwtime[1] = t;
309 sbi->s_super->s_dirt = 0;
310 return nilfs_sync_super(sbi, dupsb);
313 static void nilfs_put_super(struct super_block *sb)
315 struct nilfs_sb_info *sbi = NILFS_SB(sb);
316 struct the_nilfs *nilfs = sbi->s_nilfs;
318 nilfs_detach_segment_constructor(sbi);
320 if (!(sb->s_flags & MS_RDONLY)) {
321 down_write(&nilfs->ns_sem);
322 nilfs->ns_sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
323 nilfs_commit_super(sbi, 1);
324 up_write(&nilfs->ns_sem);
327 nilfs_detach_checkpoint(sbi);
328 put_nilfs(sbi->s_nilfs);
330 sb->s_fs_info = NULL;
335 * nilfs_write_super - write super block(s) of NILFS
338 * nilfs_write_super() gets a fs-dependent lock, writes super block(s), and
339 * clears s_dirt. This function is called in the section protected by
342 * The s_dirt flag is managed by each filesystem and we protect it by ns_sem
343 * of the struct the_nilfs. Lock order must be as follows:
346 * 2. down_write(&nilfs->ns_sem)
348 * Inside NILFS, locking ns_sem is enough to protect s_dirt and the buffer
349 * of the super block (nilfs->ns_sbp[]).
351 * In most cases, VFS functions call lock_super() before calling these
352 * methods. So we must be careful not to bring on deadlocks when using
353 * lock_super(); see generic_shutdown_super(), write_super(), and so on.
355 * Note that order of lock_kernel() and lock_super() depends on contexts
356 * of VFS. We should also note that lock_kernel() can be used in its
357 * protective section and only the outermost one has an effect.
359 static void nilfs_write_super(struct super_block *sb)
361 struct nilfs_sb_info *sbi = NILFS_SB(sb);
362 struct the_nilfs *nilfs = sbi->s_nilfs;
364 down_write(&nilfs->ns_sem);
365 if (!(sb->s_flags & MS_RDONLY)) {
366 struct nilfs_super_block **sbp = nilfs->ns_sbp;
367 u64 t = get_seconds();
370 if (!nilfs_discontinued(nilfs) && t >= nilfs->ns_sbwtime[0] &&
371 t < nilfs->ns_sbwtime[0] + NILFS_SB_FREQ) {
372 up_write(&nilfs->ns_sem);
375 dupsb = sbp[1] && t > nilfs->ns_sbwtime[1] + NILFS_ALTSB_FREQ;
376 nilfs_commit_super(sbi, dupsb);
379 up_write(&nilfs->ns_sem);
382 static int nilfs_sync_fs(struct super_block *sb, int wait)
386 /* This function is called when super block should be written back */
388 err = nilfs_construct_segment(sb);
392 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
394 struct the_nilfs *nilfs = sbi->s_nilfs;
395 struct nilfs_checkpoint *raw_cp;
396 struct buffer_head *bh_cp;
399 down_write(&nilfs->ns_sem);
400 list_add(&sbi->s_list, &nilfs->ns_supers);
401 up_write(&nilfs->ns_sem);
403 sbi->s_ifile = nilfs_mdt_new(
404 nilfs, sbi->s_super, NILFS_IFILE_INO, NILFS_IFILE_GFP);
408 err = nilfs_palloc_init_blockgroup(sbi->s_ifile, nilfs->ns_inode_size);
412 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
415 if (err == -ENOENT || err == -EINVAL) {
417 "NILFS: Invalid checkpoint "
418 "(checkpoint number=%llu)\n",
419 (unsigned long long)cno);
424 err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
427 atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
428 atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
430 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
434 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
436 nilfs_mdt_destroy(sbi->s_ifile);
439 down_write(&nilfs->ns_sem);
440 list_del_init(&sbi->s_list);
441 up_write(&nilfs->ns_sem);
446 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
448 struct the_nilfs *nilfs = sbi->s_nilfs;
450 nilfs_mdt_clear(sbi->s_ifile);
451 nilfs_mdt_destroy(sbi->s_ifile);
453 down_write(&nilfs->ns_sem);
454 list_del_init(&sbi->s_list);
455 up_write(&nilfs->ns_sem);
458 static int nilfs_mark_recovery_complete(struct nilfs_sb_info *sbi)
460 struct the_nilfs *nilfs = sbi->s_nilfs;
463 down_write(&nilfs->ns_sem);
464 if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
465 nilfs->ns_mount_state |= NILFS_VALID_FS;
466 err = nilfs_commit_super(sbi, 1);
468 printk(KERN_INFO "NILFS: recovery complete.\n");
470 up_write(&nilfs->ns_sem);
474 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
476 struct super_block *sb = dentry->d_sb;
477 struct nilfs_sb_info *sbi = NILFS_SB(sb);
478 struct the_nilfs *nilfs = sbi->s_nilfs;
479 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
480 unsigned long long blocks;
481 unsigned long overhead;
482 unsigned long nrsvblocks;
483 sector_t nfreeblocks;
487 * Compute all of the segment blocks
489 * The blocks before first segment and after last segment
492 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
493 - nilfs->ns_first_data_block;
494 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
497 * Compute the overhead
499 * When distributing meta data blocks outside semgent structure,
500 * We must count them as the overhead.
504 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
508 buf->f_type = NILFS_SUPER_MAGIC;
509 buf->f_bsize = sb->s_blocksize;
510 buf->f_blocks = blocks - overhead;
511 buf->f_bfree = nfreeblocks;
512 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
513 (buf->f_bfree - nrsvblocks) : 0;
514 buf->f_files = atomic_read(&sbi->s_inodes_count);
515 buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
516 buf->f_namelen = NILFS_NAME_LEN;
517 buf->f_fsid.val[0] = (u32)id;
518 buf->f_fsid.val[1] = (u32)(id >> 32);
523 static struct super_operations nilfs_sops = {
524 .alloc_inode = nilfs_alloc_inode,
525 .destroy_inode = nilfs_destroy_inode,
526 .dirty_inode = nilfs_dirty_inode,
527 /* .write_inode = nilfs_write_inode, */
528 /* .put_inode = nilfs_put_inode, */
529 /* .drop_inode = nilfs_drop_inode, */
530 .delete_inode = nilfs_delete_inode,
531 .put_super = nilfs_put_super,
532 .write_super = nilfs_write_super,
533 .sync_fs = nilfs_sync_fs,
534 /* .write_super_lockfs */
536 .statfs = nilfs_statfs,
537 .remount_fs = nilfs_remount,
538 .clear_inode = nilfs_clear_inode,
543 static struct inode *
544 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
548 if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
549 ino != NILFS_SKETCH_INO)
550 return ERR_PTR(-ESTALE);
552 inode = nilfs_iget(sb, ino);
554 return ERR_CAST(inode);
555 if (generation && inode->i_generation != generation) {
557 return ERR_PTR(-ESTALE);
563 static struct dentry *
564 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
567 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
568 nilfs_nfs_get_inode);
571 static struct dentry *
572 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
575 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
576 nilfs_nfs_get_inode);
579 static struct export_operations nilfs_export_ops = {
580 .fh_to_dentry = nilfs_fh_to_dentry,
581 .fh_to_parent = nilfs_fh_to_parent,
582 .get_parent = nilfs_get_parent,
586 Opt_err_cont, Opt_err_panic, Opt_err_ro,
587 Opt_barrier, Opt_snapshot, Opt_order,
591 static match_table_t tokens = {
592 {Opt_err_cont, "errors=continue"},
593 {Opt_err_panic, "errors=panic"},
594 {Opt_err_ro, "errors=remount-ro"},
595 {Opt_barrier, "barrier=%s"},
596 {Opt_snapshot, "cp=%u"},
597 {Opt_order, "order=%s"},
601 static int match_bool(substring_t *s, int *result)
603 int len = s->to - s->from;
605 if (strncmp(s->from, "on", len) == 0)
607 else if (strncmp(s->from, "off", len) == 0)
614 static int parse_options(char *options, struct super_block *sb)
616 struct nilfs_sb_info *sbi = NILFS_SB(sb);
618 substring_t args[MAX_OPT_ARGS];
624 while ((p = strsep(&options, ",")) != NULL) {
629 token = match_token(p, tokens, args);
632 if (match_bool(&args[0], &option))
635 nilfs_set_opt(sbi, BARRIER);
637 nilfs_clear_opt(sbi, BARRIER);
640 if (strcmp(args[0].from, "relaxed") == 0)
641 /* Ordered data semantics */
642 nilfs_clear_opt(sbi, STRICT_ORDER);
643 else if (strcmp(args[0].from, "strict") == 0)
644 /* Strict in-order semantics */
645 nilfs_set_opt(sbi, STRICT_ORDER);
650 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
653 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
656 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
659 if (match_int(&args[0], &option) || option <= 0)
661 if (!(sb->s_flags & MS_RDONLY))
663 sbi->s_snapshot_cno = option;
664 nilfs_set_opt(sbi, SNAPSHOT);
668 "NILFS: Unrecognized mount option \"%s\"\n", p);
676 nilfs_set_default_options(struct nilfs_sb_info *sbi,
677 struct nilfs_super_block *sbp)
680 NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER;
683 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
685 struct the_nilfs *nilfs = sbi->s_nilfs;
686 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
687 int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
688 int mnt_count = le16_to_cpu(sbp->s_mnt_count);
690 /* nilfs->sem must be locked by the caller. */
691 if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
692 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
693 } else if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
695 "NILFS warning: mounting fs with errors\n");
697 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
699 "NILFS warning: maximal mount count reached\n");
703 sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
705 sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
706 sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
707 sbp->s_mtime = cpu_to_le64(get_seconds());
708 return nilfs_commit_super(sbi, 1);
711 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
712 u64 pos, int blocksize,
713 struct buffer_head **pbh)
715 unsigned long long sb_index = pos;
716 unsigned long offset;
718 offset = do_div(sb_index, blocksize);
719 *pbh = sb_bread(sb, sb_index);
722 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
725 int nilfs_store_magic_and_option(struct super_block *sb,
726 struct nilfs_super_block *sbp,
729 struct nilfs_sb_info *sbi = NILFS_SB(sb);
731 sb->s_magic = le16_to_cpu(sbp->s_magic);
733 /* FS independent flags */
734 #ifdef NILFS_ATIME_DISABLE
735 sb->s_flags |= MS_NOATIME;
738 nilfs_set_default_options(sbi, sbp);
740 sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
741 sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
742 sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
743 sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
745 return !parse_options(data, sb) ? -EINVAL : 0 ;
749 * nilfs_fill_super() - initialize a super block instance
751 * @data: mount options
752 * @silent: silent mode flag
753 * @nilfs: the_nilfs struct
755 * This function is called exclusively by bd_mount_mutex.
756 * So, the recovery process is protected from other simultaneous mounts.
759 nilfs_fill_super(struct super_block *sb, void *data, int silent,
760 struct the_nilfs *nilfs)
762 struct nilfs_sb_info *sbi;
767 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
774 sbi->s_nilfs = nilfs;
777 err = init_nilfs(nilfs, sbi, (char *)data);
781 spin_lock_init(&sbi->s_inode_lock);
782 INIT_LIST_HEAD(&sbi->s_dirty_files);
783 INIT_LIST_HEAD(&sbi->s_list);
786 * Following initialization is overlapped because
787 * nilfs_sb_info structure has been cleared at the beginning.
788 * But we reserve them to keep our interest and make ready
789 * for the future change.
791 get_random_bytes(&sbi->s_next_generation,
792 sizeof(sbi->s_next_generation));
793 spin_lock_init(&sbi->s_next_gen_lock);
795 sb->s_op = &nilfs_sops;
796 sb->s_export_op = &nilfs_export_ops;
800 if (!nilfs_loaded(nilfs)) {
801 err = load_nilfs(nilfs, sbi);
805 cno = nilfs_last_cno(nilfs);
807 if (sb->s_flags & MS_RDONLY) {
808 if (nilfs_test_opt(sbi, SNAPSHOT)) {
809 err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
810 sbi->s_snapshot_cno);
815 "NILFS: The specified checkpoint is "
817 "(checkpoint number=%llu).\n",
818 (unsigned long long)sbi->s_snapshot_cno);
822 cno = sbi->s_snapshot_cno;
824 /* Read-only mount */
825 sbi->s_snapshot_cno = cno;
828 err = nilfs_attach_checkpoint(sbi, cno);
830 printk(KERN_ERR "NILFS: error loading a checkpoint"
831 " (checkpoint number=%llu).\n", (unsigned long long)cno);
835 if (!(sb->s_flags & MS_RDONLY)) {
836 err = nilfs_attach_segment_constructor(sbi);
838 goto failed_checkpoint;
841 root = nilfs_iget(sb, NILFS_ROOT_INO);
843 printk(KERN_ERR "NILFS: get root inode failed\n");
847 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
849 printk(KERN_ERR "NILFS: corrupt root inode.\n");
853 sb->s_root = d_alloc_root(root);
856 printk(KERN_ERR "NILFS: get root dentry failed\n");
861 if (!(sb->s_flags & MS_RDONLY)) {
862 down_write(&nilfs->ns_sem);
863 nilfs_setup_super(sbi);
864 up_write(&nilfs->ns_sem);
867 err = nilfs_mark_recovery_complete(sbi);
869 printk(KERN_ERR "NILFS: recovery failed.\n");
880 nilfs_detach_segment_constructor(sbi);
883 nilfs_detach_checkpoint(sbi);
887 sb->s_fs_info = NULL;
892 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
894 struct nilfs_sb_info *sbi = NILFS_SB(sb);
895 struct nilfs_super_block *sbp;
896 struct the_nilfs *nilfs = sbi->s_nilfs;
897 unsigned long old_sb_flags;
898 struct nilfs_mount_options old_opts;
901 old_sb_flags = sb->s_flags;
902 old_opts.mount_opt = sbi->s_mount_opt;
903 old_opts.snapshot_cno = sbi->s_snapshot_cno;
905 if (!parse_options(data, sb)) {
909 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
911 if ((*flags & MS_RDONLY) &&
912 sbi->s_snapshot_cno != old_opts.snapshot_cno) {
913 printk(KERN_WARNING "NILFS (device %s): couldn't "
914 "remount to a different snapshot. \n",
920 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
922 if (*flags & MS_RDONLY) {
923 /* Shutting down the segment constructor */
924 nilfs_detach_segment_constructor(sbi);
925 sb->s_flags |= MS_RDONLY;
927 sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
928 /* nilfs_set_opt(sbi, SNAPSHOT); */
931 * Remounting a valid RW partition RDONLY, so set
932 * the RDONLY flag and then mark the partition as valid again.
934 down_write(&nilfs->ns_sem);
935 sbp = nilfs->ns_sbp[0];
936 if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
937 (nilfs->ns_mount_state & NILFS_VALID_FS))
938 sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
939 sbp->s_mtime = cpu_to_le64(get_seconds());
940 nilfs_commit_super(sbi, 1);
941 up_write(&nilfs->ns_sem);
944 * Mounting a RDONLY partition read-write, so reread and
945 * store the current valid flag. (It may have been changed
946 * by fsck since we originally mounted the partition.)
948 down(&sb->s_bdev->bd_mount_sem);
949 /* Check existing RW-mount */
950 if (test_exclusive_mount(sb->s_type, sb->s_bdev, 0)) {
951 printk(KERN_WARNING "NILFS (device %s): couldn't "
952 "remount because a RW-mount exists.\n",
955 goto rw_remount_failed;
957 if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
958 printk(KERN_WARNING "NILFS (device %s): couldn't "
959 "remount because the current RO-mount is not "
963 goto rw_remount_failed;
965 sb->s_flags &= ~MS_RDONLY;
966 nilfs_clear_opt(sbi, SNAPSHOT);
967 sbi->s_snapshot_cno = 0;
969 err = nilfs_attach_segment_constructor(sbi);
971 goto rw_remount_failed;
973 down_write(&nilfs->ns_sem);
974 nilfs_setup_super(sbi);
975 up_write(&nilfs->ns_sem);
977 up(&sb->s_bdev->bd_mount_sem);
983 up(&sb->s_bdev->bd_mount_sem);
985 sb->s_flags = old_sb_flags;
986 sbi->s_mount_opt = old_opts.mount_opt;
987 sbi->s_snapshot_cno = old_opts.snapshot_cno;
991 struct nilfs_super_data {
992 struct block_device *bdev;
998 * nilfs_identify - pre-read mount options needed to identify mount instance
999 * @data: mount options
1000 * @sd: nilfs_super_data
1002 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1004 char *p, *options = data;
1005 substring_t args[MAX_OPT_ARGS];
1010 p = strsep(&options, ",");
1011 if (p != NULL && *p) {
1012 token = match_token(p, tokens, args);
1013 if (token == Opt_snapshot) {
1014 if (!(sd->flags & MS_RDONLY))
1017 ret = match_int(&args[0], &option);
1028 "NILFS: invalid mount option: %s\n", p);
1032 BUG_ON(options == data);
1033 *(options - 1) = ',';
1038 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1040 struct nilfs_super_data *sd = data;
1042 s->s_bdev = sd->bdev;
1043 s->s_dev = s->s_bdev->bd_dev;
1047 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1049 struct nilfs_super_data *sd = data;
1051 return s->s_bdev == sd->bdev;
1054 static int nilfs_test_bdev_super2(struct super_block *s, void *data)
1056 struct nilfs_super_data *sd = data;
1059 if (s->s_bdev != sd->bdev)
1062 if (!((s->s_flags | sd->flags) & MS_RDONLY))
1063 return 1; /* Reuse an old R/W-mode super_block */
1065 if (s->s_flags & sd->flags & MS_RDONLY) {
1066 if (down_read_trylock(&s->s_umount)) {
1068 (sd->cno == NILFS_SB(s)->s_snapshot_cno);
1069 up_read(&s->s_umount);
1071 * This path is locked with sb_lock by sget().
1072 * So, drop_super() causes deadlock.
1081 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1082 const char *dev_name, void *data, struct vfsmount *mnt)
1084 struct nilfs_super_data sd;
1085 struct super_block *s, *s2;
1086 struct the_nilfs *nilfs = NULL;
1087 int err, need_to_close = 1;
1089 sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1090 if (IS_ERR(sd.bdev))
1091 return PTR_ERR(sd.bdev);
1094 * To get mount instance using sget() vfs-routine, NILFS needs
1095 * much more information than normal filesystems to identify mount
1096 * instance. For snapshot mounts, not only a mount type (ro-mount
1097 * or rw-mount) but also a checkpoint number is required.
1098 * The results are passed in sget() using nilfs_super_data.
1102 if (nilfs_identify((char *)data, &sd)) {
1108 * once the super is inserted into the list by sget, s_umount
1109 * will protect the lockfs code from trying to start a snapshot
1110 * while we are mounting
1112 down(&sd.bdev->bd_mount_sem);
1114 (err = test_exclusive_mount(fs_type, sd.bdev, flags ^ MS_RDONLY))) {
1115 err = (err < 0) ? : -EBUSY;
1120 * Phase-1: search any existent instance and get the_nilfs
1122 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1128 nilfs = alloc_nilfs(sd.bdev);
1132 struct nilfs_sb_info *sbi = NILFS_SB(s);
1135 * s_umount protects super_block from unmount process;
1136 * It covers pointers of nilfs_sb_info and the_nilfs.
1138 nilfs = sbi->s_nilfs;
1140 up_write(&s->s_umount);
1143 * Phase-2: search specified snapshot or R/W mode super_block
1146 /* trying to get the latest checkpoint. */
1147 sd.cno = nilfs_last_cno(nilfs);
1149 s2 = sget(fs_type, nilfs_test_bdev_super2,
1150 nilfs_set_bdev_super, &sd);
1151 deactivate_super(s);
1153 * Although deactivate_super() invokes close_bdev_exclusive() at
1154 * kill_block_super(). Here, s is an existent mount; we need
1155 * one more close_bdev_exclusive() call.
1163 char b[BDEVNAME_SIZE];
1166 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1167 sb_set_blocksize(s, block_size(sd.bdev));
1169 err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1173 s->s_flags |= MS_ACTIVE;
1175 } else if (!(s->s_flags & MS_RDONLY)) {
1179 up(&sd.bdev->bd_mount_sem);
1182 close_bdev_exclusive(sd.bdev, flags);
1183 simple_set_mnt(mnt, s);
1187 up(&sd.bdev->bd_mount_sem);
1190 close_bdev_exclusive(sd.bdev, flags);
1194 up(&sd.bdev->bd_mount_sem);
1196 close_bdev_exclusive(sd.bdev, flags);
1201 /* Abandoning the newly allocated superblock */
1202 up(&sd.bdev->bd_mount_sem);
1205 up_write(&s->s_umount);
1206 deactivate_super(s);
1208 * deactivate_super() invokes close_bdev_exclusive().
1209 * We must finish all post-cleaning before this call;
1210 * put_nilfs() and unlocking bd_mount_sem need the block device.
1215 static int nilfs_test_bdev_super3(struct super_block *s, void *data)
1217 struct nilfs_super_data *sd = data;
1220 if (s->s_bdev != sd->bdev)
1222 if (down_read_trylock(&s->s_umount)) {
1223 ret = (s->s_flags & MS_RDONLY) && s->s_root &&
1224 nilfs_test_opt(NILFS_SB(s), SNAPSHOT);
1225 up_read(&s->s_umount);
1227 return 0; /* ignore snapshot mounts */
1229 return !((sd->flags ^ s->s_flags) & MS_RDONLY);
1232 static int __false_bdev_super(struct super_block *s, void *data)
1234 #if 0 /* XXX: workaround for lock debug. This is not good idea */
1235 up_write(&s->s_umount);
1241 * test_exclusive_mount - check whether an exclusive RW/RO mount exists or not.
1242 * fs_type: filesystem type
1243 * bdev: block device
1244 * flag: 0 (check rw-mount) or MS_RDONLY (check ro-mount)
1245 * res: pointer to an integer to store result
1247 * This function must be called within a section protected by bd_mount_mutex.
1249 static int test_exclusive_mount(struct file_system_type *fs_type,
1250 struct block_device *bdev, int flags)
1252 struct super_block *s;
1253 struct nilfs_super_data sd = { .flags = flags, .bdev = bdev };
1255 s = sget(fs_type, nilfs_test_bdev_super3, __false_bdev_super, &sd);
1257 if (PTR_ERR(s) != -EFAULT)
1259 return 0; /* Not found */
1261 up_write(&s->s_umount);
1262 deactivate_super(s);
1263 return 1; /* Found */
1266 struct file_system_type nilfs_fs_type = {
1267 .owner = THIS_MODULE,
1269 .get_sb = nilfs_get_sb,
1270 .kill_sb = kill_block_super,
1271 .fs_flags = FS_REQUIRES_DEV,
1274 static int __init init_nilfs_fs(void)
1278 err = nilfs_init_inode_cache();
1282 err = nilfs_init_transaction_cache();
1284 goto failed_inode_cache;
1286 err = nilfs_init_segbuf_cache();
1288 goto failed_transaction_cache;
1290 err = nilfs_btree_path_cache_init();
1292 goto failed_segbuf_cache;
1294 err = register_filesystem(&nilfs_fs_type);
1296 goto failed_btree_path_cache;
1300 failed_btree_path_cache:
1301 nilfs_btree_path_cache_destroy();
1303 failed_segbuf_cache:
1304 nilfs_destroy_segbuf_cache();
1306 failed_transaction_cache:
1307 nilfs_destroy_transaction_cache();
1310 nilfs_destroy_inode_cache();
1316 static void __exit exit_nilfs_fs(void)
1318 nilfs_destroy_segbuf_cache();
1319 nilfs_destroy_transaction_cache();
1320 nilfs_destroy_inode_cache();
1321 nilfs_btree_path_cache_destroy();
1322 unregister_filesystem(&nilfs_fs_type);
1325 module_init(init_nilfs_fs)
1326 module_exit(exit_nilfs_fs)