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);
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);
212 * nilfs_update_last_segment - change pointer to the latest segment
213 * @sbi: nilfs_sb_info
214 * @update_cno: flag whether to update checkpoint number.
216 * nilfs_update_last_segment() changes information in the super block
217 * after a partial segment is written out successfully. The super
218 * block is marked dirty. It will be written out at the next VFS sync
219 * operations such as sync_supers() and generic_shutdown_super().
221 void nilfs_update_last_segment(struct nilfs_sb_info *sbi, int update_cno)
223 struct the_nilfs *nilfs = sbi->s_nilfs;
224 struct nilfs_super_block *sbp = nilfs->ns_sbp;
226 /* nilfs->sem must be locked by the caller. */
227 spin_lock(&nilfs->ns_last_segment_lock);
229 nilfs->ns_last_cno = nilfs->ns_cno++;
230 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
231 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
232 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
233 spin_unlock(&nilfs->ns_last_segment_lock);
235 sbi->s_super->s_dirt = 1; /* must be set if delaying the call of
236 nilfs_commit_super() */
239 static int nilfs_sync_super(struct nilfs_sb_info *sbi)
241 struct the_nilfs *nilfs = sbi->s_nilfs;
243 int barrier_done = 0;
245 if (nilfs_test_opt(sbi, BARRIER)) {
246 set_buffer_ordered(nilfs->ns_sbh);
250 set_buffer_dirty(nilfs->ns_sbh);
251 err = sync_dirty_buffer(nilfs->ns_sbh);
252 if (err == -EOPNOTSUPP && barrier_done) {
253 nilfs_warning(sbi->s_super, __func__,
254 "barrier-based sync failed. "
255 "disabling barriers\n");
256 nilfs_clear_opt(sbi, BARRIER);
258 clear_buffer_ordered(nilfs->ns_sbh);
263 "NILFS: unable to write superblock (err=%d)\n", err);
265 nilfs_dispose_used_segments(nilfs);
266 clear_nilfs_discontinued(nilfs);
272 int nilfs_commit_super(struct nilfs_sb_info *sbi)
274 struct the_nilfs *nilfs = sbi->s_nilfs;
275 struct nilfs_super_block *sbp = nilfs->ns_sbp;
276 sector_t nfreeblocks;
279 /* nilfs->sem must be locked by the caller. */
280 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
282 printk(KERN_ERR "NILFS: failed to count free blocks\n");
285 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
286 sbp->s_wtime = cpu_to_le64(get_seconds());
288 sbp->s_sum = crc32_le(nilfs->ns_crc_seed, (unsigned char *)sbp,
289 le16_to_cpu(sbp->s_bytes));
291 sbi->s_super->s_dirt = 0;
292 return nilfs_sync_super(sbi);
295 static void nilfs_put_super(struct super_block *sb)
297 struct nilfs_sb_info *sbi = NILFS_SB(sb);
298 struct the_nilfs *nilfs = sbi->s_nilfs;
300 nilfs_detach_segment_constructor(sbi);
302 if (!(sb->s_flags & MS_RDONLY)) {
303 down_write(&nilfs->ns_sem);
304 nilfs->ns_sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
305 nilfs_commit_super(sbi);
306 up_write(&nilfs->ns_sem);
309 nilfs_detach_checkpoint(sbi);
310 put_nilfs(sbi->s_nilfs);
312 sb->s_fs_info = NULL;
317 * nilfs_write_super - write super block(s) of NILFS
320 * nilfs_write_super() gets a fs-dependent lock, writes super block(s), and
321 * clears s_dirt. This function is called in the section protected by
324 * The s_dirt flag is managed by each filesystem and we protect it by ns_sem
325 * of the struct the_nilfs. Lock order must be as follows:
328 * 2. down_write(&nilfs->ns_sem)
330 * Inside NILFS, locking ns_sem is enough to protect s_dirt and the buffer
331 * of the super block (nilfs->ns_sbp).
333 * In most cases, VFS functions call lock_super() before calling these
334 * methods. So we must be careful not to bring on deadlocks when using
335 * lock_super(); see generic_shutdown_super(), write_super(), and so on.
337 * Note that order of lock_kernel() and lock_super() depends on contexts
338 * of VFS. We should also note that lock_kernel() can be used in its
339 * protective section and only the outermost one has an effect.
341 static void nilfs_write_super(struct super_block *sb)
343 struct nilfs_sb_info *sbi = NILFS_SB(sb);
344 struct the_nilfs *nilfs = sbi->s_nilfs;
346 down_write(&nilfs->ns_sem);
347 if (!(sb->s_flags & MS_RDONLY))
348 nilfs_commit_super(sbi);
350 up_write(&nilfs->ns_sem);
353 static int nilfs_sync_fs(struct super_block *sb, int wait)
357 /* This function is called when super block should be written back */
359 err = nilfs_construct_segment(sb);
363 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
365 struct the_nilfs *nilfs = sbi->s_nilfs;
366 struct nilfs_checkpoint *raw_cp;
367 struct buffer_head *bh_cp;
370 down_write(&nilfs->ns_sem);
371 list_add(&sbi->s_list, &nilfs->ns_supers);
372 up_write(&nilfs->ns_sem);
374 sbi->s_ifile = nilfs_mdt_new(
375 nilfs, sbi->s_super, NILFS_IFILE_INO, NILFS_IFILE_GFP);
379 err = nilfs_palloc_init_blockgroup(sbi->s_ifile, nilfs->ns_inode_size);
383 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
386 if (err == -ENOENT || err == -EINVAL) {
388 "NILFS: Invalid checkpoint "
389 "(checkpoint number=%llu)\n",
390 (unsigned long long)cno);
395 err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
398 atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
399 atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
401 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
405 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
407 nilfs_mdt_destroy(sbi->s_ifile);
410 down_write(&nilfs->ns_sem);
411 list_del_init(&sbi->s_list);
412 up_write(&nilfs->ns_sem);
417 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
419 struct the_nilfs *nilfs = sbi->s_nilfs;
421 nilfs_mdt_clear(sbi->s_ifile);
422 nilfs_mdt_destroy(sbi->s_ifile);
424 down_write(&nilfs->ns_sem);
425 list_del_init(&sbi->s_list);
426 up_write(&nilfs->ns_sem);
429 static int nilfs_mark_recovery_complete(struct nilfs_sb_info *sbi)
431 struct the_nilfs *nilfs = sbi->s_nilfs;
434 down_write(&nilfs->ns_sem);
435 if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
436 nilfs->ns_mount_state |= NILFS_VALID_FS;
437 err = nilfs_commit_super(sbi);
439 printk(KERN_INFO "NILFS: recovery complete.\n");
441 up_write(&nilfs->ns_sem);
445 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
447 struct super_block *sb = dentry->d_sb;
448 struct nilfs_sb_info *sbi = NILFS_SB(sb);
449 unsigned long long blocks;
450 unsigned long overhead;
451 unsigned long nrsvblocks;
452 sector_t nfreeblocks;
453 struct the_nilfs *nilfs = sbi->s_nilfs;
457 * Compute all of the segment blocks
459 * The blocks before first segment and after last segment
462 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
463 - nilfs->ns_first_data_block;
464 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
467 * Compute the overhead
469 * When distributing meta data blocks outside semgent structure,
470 * We must count them as the overhead.
474 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
478 buf->f_type = NILFS_SUPER_MAGIC;
479 buf->f_bsize = sb->s_blocksize;
480 buf->f_blocks = blocks - overhead;
481 buf->f_bfree = nfreeblocks;
482 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
483 (buf->f_bfree - nrsvblocks) : 0;
484 buf->f_files = atomic_read(&sbi->s_inodes_count);
485 buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
486 buf->f_namelen = NILFS_NAME_LEN;
490 static struct super_operations nilfs_sops = {
491 .alloc_inode = nilfs_alloc_inode,
492 .destroy_inode = nilfs_destroy_inode,
493 .dirty_inode = nilfs_dirty_inode,
494 /* .write_inode = nilfs_write_inode, */
495 /* .put_inode = nilfs_put_inode, */
496 /* .drop_inode = nilfs_drop_inode, */
497 .delete_inode = nilfs_delete_inode,
498 .put_super = nilfs_put_super,
499 .write_super = nilfs_write_super,
500 .sync_fs = nilfs_sync_fs,
501 /* .write_super_lockfs */
503 .statfs = nilfs_statfs,
504 .remount_fs = nilfs_remount,
505 .clear_inode = nilfs_clear_inode,
510 static struct inode *
511 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
515 if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
516 ino != NILFS_SKETCH_INO)
517 return ERR_PTR(-ESTALE);
519 inode = nilfs_iget(sb, ino);
521 return ERR_CAST(inode);
522 if (generation && inode->i_generation != generation) {
524 return ERR_PTR(-ESTALE);
530 static struct dentry *
531 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
534 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
535 nilfs_nfs_get_inode);
538 static struct dentry *
539 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
542 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
543 nilfs_nfs_get_inode);
546 static struct export_operations nilfs_export_ops = {
547 .fh_to_dentry = nilfs_fh_to_dentry,
548 .fh_to_parent = nilfs_fh_to_parent,
549 .get_parent = nilfs_get_parent,
553 Opt_err_cont, Opt_err_panic, Opt_err_ro,
554 Opt_barrier, Opt_snapshot, Opt_order,
558 static match_table_t tokens = {
559 {Opt_err_cont, "errors=continue"},
560 {Opt_err_panic, "errors=panic"},
561 {Opt_err_ro, "errors=remount-ro"},
562 {Opt_barrier, "barrier=%s"},
563 {Opt_snapshot, "cp=%u"},
564 {Opt_order, "order=%s"},
568 static int match_bool(substring_t *s, int *result)
570 int len = s->to - s->from;
572 if (strncmp(s->from, "on", len) == 0)
574 else if (strncmp(s->from, "off", len) == 0)
581 static int parse_options(char *options, struct super_block *sb)
583 struct nilfs_sb_info *sbi = NILFS_SB(sb);
585 substring_t args[MAX_OPT_ARGS];
591 while ((p = strsep(&options, ",")) != NULL) {
596 token = match_token(p, tokens, args);
599 if (match_bool(&args[0], &option))
602 nilfs_set_opt(sbi, BARRIER);
604 nilfs_clear_opt(sbi, BARRIER);
607 if (strcmp(args[0].from, "relaxed") == 0)
608 /* Ordered data semantics */
609 nilfs_clear_opt(sbi, STRICT_ORDER);
610 else if (strcmp(args[0].from, "strict") == 0)
611 /* Strict in-order semantics */
612 nilfs_set_opt(sbi, STRICT_ORDER);
617 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
620 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
623 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
626 if (match_int(&args[0], &option) || option <= 0)
628 if (!(sb->s_flags & MS_RDONLY))
630 sbi->s_snapshot_cno = option;
631 nilfs_set_opt(sbi, SNAPSHOT);
635 "NILFS: Unrecognized mount option \"%s\"\n", p);
643 nilfs_set_default_options(struct nilfs_sb_info *sbi,
644 struct nilfs_super_block *sbp)
647 NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER;
650 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
652 struct the_nilfs *nilfs = sbi->s_nilfs;
653 struct nilfs_super_block *sbp = nilfs->ns_sbp;
654 int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
655 int mnt_count = le16_to_cpu(sbp->s_mnt_count);
657 /* nilfs->sem must be locked by the caller. */
658 if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
659 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
660 } else if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
662 "NILFS warning: mounting fs with errors\n");
664 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
666 "NILFS warning: maximal mount count reached\n");
670 sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
672 sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
673 sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
674 sbp->s_mtime = cpu_to_le64(get_seconds());
675 return nilfs_commit_super(sbi);
678 struct nilfs_super_block *
679 nilfs_load_super_block(struct super_block *sb, struct buffer_head **pbh)
682 unsigned long offset, sb_index;
685 * Adjusting block size
686 * Blocksize will be enlarged when it is smaller than hardware
688 * Disk format of superblock does not change.
690 blocksize = sb_min_blocksize(sb, BLOCK_SIZE);
693 "NILFS: unable to set blocksize of superblock\n");
696 sb_index = NILFS_SB_OFFSET_BYTES / blocksize;
697 offset = NILFS_SB_OFFSET_BYTES % blocksize;
699 *pbh = sb_bread(sb, sb_index);
701 printk(KERN_ERR "NILFS: unable to read superblock\n");
704 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
707 struct nilfs_super_block *
708 nilfs_reload_super_block(struct super_block *sb, struct buffer_head **pbh,
711 struct nilfs_super_block *sbp;
712 unsigned long offset, sb_index;
713 int hw_blocksize = bdev_hardsect_size(sb->s_bdev);
715 if (blocksize < hw_blocksize) {
717 "NILFS: blocksize %d too small for device "
718 "(sector-size = %d).\n",
719 blocksize, hw_blocksize);
723 sb_set_blocksize(sb, blocksize);
725 sb_index = NILFS_SB_OFFSET_BYTES / blocksize;
726 offset = NILFS_SB_OFFSET_BYTES % blocksize;
728 *pbh = sb_bread(sb, sb_index);
731 "NILFS: cannot read superblock on 2nd try.\n");
735 sbp = (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
736 if (sbp->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
738 "NILFS: !? Magic mismatch on 2nd try.\n");
750 int nilfs_store_magic_and_option(struct super_block *sb,
751 struct nilfs_super_block *sbp,
754 struct nilfs_sb_info *sbi = NILFS_SB(sb);
756 /* trying to fill super (1st stage) */
757 sb->s_magic = le16_to_cpu(sbp->s_magic);
759 /* FS independent flags */
760 #ifdef NILFS_ATIME_DISABLE
761 sb->s_flags |= MS_NOATIME;
764 if (sb->s_magic != NILFS_SUPER_MAGIC) {
765 printk("NILFS: Can't find nilfs on dev %s.\n", sb->s_id);
769 nilfs_set_default_options(sbi, sbp);
771 sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
772 sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
773 sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
774 sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
776 if (!parse_options(data, sb))
783 * nilfs_fill_super() - initialize a super block instance
785 * @data: mount options
786 * @silent: silent mode flag
787 * @nilfs: the_nilfs struct
789 * This function is called exclusively by bd_mount_mutex.
790 * So, the recovery process is protected from other simultaneous mounts.
793 nilfs_fill_super(struct super_block *sb, void *data, int silent,
794 struct the_nilfs *nilfs)
796 struct nilfs_sb_info *sbi;
801 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
808 sbi->s_nilfs = nilfs;
811 err = init_nilfs(nilfs, sbi, (char *)data);
815 spin_lock_init(&sbi->s_inode_lock);
816 INIT_LIST_HEAD(&sbi->s_dirty_files);
817 INIT_LIST_HEAD(&sbi->s_list);
820 * Following initialization is overlapped because
821 * nilfs_sb_info structure has been cleared at the beginning.
822 * But we reserve them to keep our interest and make ready
823 * for the future change.
825 get_random_bytes(&sbi->s_next_generation,
826 sizeof(sbi->s_next_generation));
827 spin_lock_init(&sbi->s_next_gen_lock);
829 sb->s_op = &nilfs_sops;
830 sb->s_export_op = &nilfs_export_ops;
833 if (!nilfs_loaded(nilfs)) {
834 err = load_nilfs(nilfs, sbi);
838 cno = nilfs_last_cno(nilfs);
840 if (sb->s_flags & MS_RDONLY) {
841 if (nilfs_test_opt(sbi, SNAPSHOT)) {
842 if (!nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
843 sbi->s_snapshot_cno)) {
845 "NILFS: The specified checkpoint is "
847 "(checkpoint number=%llu).\n",
848 (unsigned long long)sbi->s_snapshot_cno);
852 cno = sbi->s_snapshot_cno;
854 /* Read-only mount */
855 sbi->s_snapshot_cno = cno;
858 err = nilfs_attach_checkpoint(sbi, cno);
860 printk(KERN_ERR "NILFS: error loading a checkpoint"
861 " (checkpoint number=%llu).\n", (unsigned long long)cno);
865 if (!(sb->s_flags & MS_RDONLY)) {
866 err = nilfs_attach_segment_constructor(sbi, NULL);
868 goto failed_checkpoint;
871 root = nilfs_iget(sb, NILFS_ROOT_INO);
873 printk(KERN_ERR "NILFS: get root inode failed\n");
877 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
879 printk(KERN_ERR "NILFS: corrupt root inode.\n");
883 sb->s_root = d_alloc_root(root);
886 printk(KERN_ERR "NILFS: get root dentry failed\n");
891 if (!(sb->s_flags & MS_RDONLY)) {
892 down_write(&nilfs->ns_sem);
893 nilfs_setup_super(sbi);
894 up_write(&nilfs->ns_sem);
897 err = nilfs_mark_recovery_complete(sbi);
899 printk(KERN_ERR "NILFS: recovery failed.\n");
910 nilfs_detach_segment_constructor(sbi);
913 nilfs_detach_checkpoint(sbi);
917 sb->s_fs_info = NULL;
922 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
924 struct nilfs_sb_info *sbi = NILFS_SB(sb);
925 struct nilfs_super_block *sbp;
926 struct the_nilfs *nilfs = sbi->s_nilfs;
927 unsigned long old_sb_flags;
928 struct nilfs_mount_options old_opts;
931 old_sb_flags = sb->s_flags;
932 old_opts.mount_opt = sbi->s_mount_opt;
933 old_opts.snapshot_cno = sbi->s_snapshot_cno;
935 if (!parse_options(data, sb)) {
939 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
941 if ((*flags & MS_RDONLY) &&
942 sbi->s_snapshot_cno != old_opts.snapshot_cno) {
943 printk(KERN_WARNING "NILFS (device %s): couldn't "
944 "remount to a different snapshot. \n",
950 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
952 if (*flags & MS_RDONLY) {
953 /* Shutting down the segment constructor */
954 nilfs_detach_segment_constructor(sbi);
955 sb->s_flags |= MS_RDONLY;
957 sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
958 /* nilfs_set_opt(sbi, SNAPSHOT); */
961 * Remounting a valid RW partition RDONLY, so set
962 * the RDONLY flag and then mark the partition as valid again.
964 down_write(&nilfs->ns_sem);
966 if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
967 (nilfs->ns_mount_state & NILFS_VALID_FS))
968 sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
969 sbp->s_mtime = cpu_to_le64(get_seconds());
970 nilfs_commit_super(sbi);
971 up_write(&nilfs->ns_sem);
974 * Mounting a RDONLY partition read-write, so reread and
975 * store the current valid flag. (It may have been changed
976 * by fsck since we originally mounted the partition.)
978 down(&sb->s_bdev->bd_mount_sem);
979 /* Check existing RW-mount */
980 if (test_exclusive_mount(sb->s_type, sb->s_bdev, 0)) {
981 printk(KERN_WARNING "NILFS (device %s): couldn't "
982 "remount because a RW-mount exists.\n",
985 goto rw_remount_failed;
987 if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
988 printk(KERN_WARNING "NILFS (device %s): couldn't "
989 "remount because the current RO-mount is not "
993 goto rw_remount_failed;
995 sb->s_flags &= ~MS_RDONLY;
996 nilfs_clear_opt(sbi, SNAPSHOT);
997 sbi->s_snapshot_cno = 0;
999 err = nilfs_attach_segment_constructor(sbi, NULL);
1001 goto rw_remount_failed;
1003 down_write(&nilfs->ns_sem);
1004 nilfs_setup_super(sbi);
1005 up_write(&nilfs->ns_sem);
1007 up(&sb->s_bdev->bd_mount_sem);
1013 up(&sb->s_bdev->bd_mount_sem);
1015 sb->s_flags = old_sb_flags;
1016 sbi->s_mount_opt = old_opts.mount_opt;
1017 sbi->s_snapshot_cno = old_opts.snapshot_cno;
1021 struct nilfs_super_data {
1022 struct block_device *bdev;
1028 * nilfs_identify - pre-read mount options needed to identify mount instance
1029 * @data: mount options
1030 * @sd: nilfs_super_data
1032 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1034 char *p, *options = data;
1035 substring_t args[MAX_OPT_ARGS];
1040 p = strsep(&options, ",");
1041 if (p != NULL && *p) {
1042 token = match_token(p, tokens, args);
1043 if (token == Opt_snapshot) {
1044 if (!(sd->flags & MS_RDONLY))
1047 ret = match_int(&args[0], &option);
1058 "NILFS: invalid mount option: %s\n", p);
1062 BUG_ON(options == data);
1063 *(options - 1) = ',';
1068 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1070 struct nilfs_super_data *sd = data;
1072 s->s_bdev = sd->bdev;
1073 s->s_dev = s->s_bdev->bd_dev;
1077 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1079 struct nilfs_super_data *sd = data;
1081 return s->s_bdev == sd->bdev;
1084 static int nilfs_test_bdev_super2(struct super_block *s, void *data)
1086 struct nilfs_super_data *sd = data;
1089 if (s->s_bdev != sd->bdev)
1092 if (!((s->s_flags | sd->flags) & MS_RDONLY))
1093 return 1; /* Reuse an old R/W-mode super_block */
1095 if (s->s_flags & sd->flags & MS_RDONLY) {
1096 if (down_read_trylock(&s->s_umount)) {
1098 (sd->cno == NILFS_SB(s)->s_snapshot_cno);
1099 up_read(&s->s_umount);
1101 * This path is locked with sb_lock by sget().
1102 * So, drop_super() causes deadlock.
1111 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1112 const char *dev_name, void *data, struct vfsmount *mnt)
1114 struct nilfs_super_data sd;
1115 struct super_block *s, *s2;
1116 struct the_nilfs *nilfs = NULL;
1117 int err, need_to_close = 1;
1119 sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1120 if (IS_ERR(sd.bdev))
1121 return PTR_ERR(sd.bdev);
1124 * To get mount instance using sget() vfs-routine, NILFS needs
1125 * much more information than normal filesystems to identify mount
1126 * instance. For snapshot mounts, not only a mount type (ro-mount
1127 * or rw-mount) but also a checkpoint number is required.
1128 * The results are passed in sget() using nilfs_super_data.
1132 if (nilfs_identify((char *)data, &sd)) {
1138 * once the super is inserted into the list by sget, s_umount
1139 * will protect the lockfs code from trying to start a snapshot
1140 * while we are mounting
1142 down(&sd.bdev->bd_mount_sem);
1144 (err = test_exclusive_mount(fs_type, sd.bdev, flags ^ MS_RDONLY))) {
1145 err = (err < 0) ? : -EBUSY;
1150 * Phase-1: search any existent instance and get the_nilfs
1152 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1158 nilfs = alloc_nilfs(sd.bdev);
1162 struct nilfs_sb_info *sbi = NILFS_SB(s);
1164 BUG_ON(!sbi || !sbi->s_nilfs);
1166 * s_umount protects super_block from unmount process;
1167 * It covers pointers of nilfs_sb_info and the_nilfs.
1169 nilfs = sbi->s_nilfs;
1171 up_write(&s->s_umount);
1174 * Phase-2: search specified snapshot or R/W mode super_block
1177 /* trying to get the latest checkpoint. */
1178 sd.cno = nilfs_last_cno(nilfs);
1180 s2 = sget(fs_type, nilfs_test_bdev_super2,
1181 nilfs_set_bdev_super, &sd);
1182 deactivate_super(s);
1184 * Although deactivate_super() invokes close_bdev_exclusive() at
1185 * kill_block_super(). Here, s is an existent mount; we need
1186 * one more close_bdev_exclusive() call.
1194 char b[BDEVNAME_SIZE];
1197 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1198 sb_set_blocksize(s, block_size(sd.bdev));
1200 err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1204 s->s_flags |= MS_ACTIVE;
1206 } else if (!(s->s_flags & MS_RDONLY)) {
1210 up(&sd.bdev->bd_mount_sem);
1213 close_bdev_exclusive(sd.bdev, flags);
1214 simple_set_mnt(mnt, s);
1218 up(&sd.bdev->bd_mount_sem);
1221 close_bdev_exclusive(sd.bdev, flags);
1225 up(&sd.bdev->bd_mount_sem);
1227 close_bdev_exclusive(sd.bdev, flags);
1232 /* Abandoning the newly allocated superblock */
1233 up(&sd.bdev->bd_mount_sem);
1236 up_write(&s->s_umount);
1237 deactivate_super(s);
1239 * deactivate_super() invokes close_bdev_exclusive().
1240 * We must finish all post-cleaning before this call;
1241 * put_nilfs() and unlocking bd_mount_sem need the block device.
1246 static int nilfs_test_bdev_super3(struct super_block *s, void *data)
1248 struct nilfs_super_data *sd = data;
1251 if (s->s_bdev != sd->bdev)
1253 if (down_read_trylock(&s->s_umount)) {
1254 ret = (s->s_flags & MS_RDONLY) && s->s_root &&
1255 nilfs_test_opt(NILFS_SB(s), SNAPSHOT);
1256 up_read(&s->s_umount);
1258 return 0; /* ignore snapshot mounts */
1260 return !((sd->flags ^ s->s_flags) & MS_RDONLY);
1263 static int __false_bdev_super(struct super_block *s, void *data)
1265 #if 0 /* XXX: workaround for lock debug. This is not good idea */
1266 up_write(&s->s_umount);
1272 * test_exclusive_mount - check whether an exclusive RW/RO mount exists or not.
1273 * fs_type: filesystem type
1274 * bdev: block device
1275 * flag: 0 (check rw-mount) or MS_RDONLY (check ro-mount)
1276 * res: pointer to an integer to store result
1278 * This function must be called within a section protected by bd_mount_mutex.
1280 static int test_exclusive_mount(struct file_system_type *fs_type,
1281 struct block_device *bdev, int flags)
1283 struct super_block *s;
1284 struct nilfs_super_data sd = { .flags = flags, .bdev = bdev };
1286 s = sget(fs_type, nilfs_test_bdev_super3, __false_bdev_super, &sd);
1288 if (PTR_ERR(s) != -EFAULT)
1290 return 0; /* Not found */
1292 up_write(&s->s_umount);
1293 deactivate_super(s);
1294 return 1; /* Found */
1297 struct file_system_type nilfs_fs_type = {
1298 .owner = THIS_MODULE,
1300 .get_sb = nilfs_get_sb,
1301 .kill_sb = kill_block_super,
1302 .fs_flags = FS_REQUIRES_DEV,
1305 static int __init init_nilfs_fs(void)
1309 err = nilfs_init_inode_cache();
1313 err = nilfs_init_transaction_cache();
1315 goto failed_inode_cache;
1317 err = nilfs_init_segbuf_cache();
1319 goto failed_transaction_cache;
1321 err = nilfs_btree_path_cache_init();
1323 goto failed_segbuf_cache;
1325 err = register_filesystem(&nilfs_fs_type);
1327 goto failed_btree_path_cache;
1331 failed_btree_path_cache:
1332 nilfs_btree_path_cache_destroy();
1334 failed_segbuf_cache:
1335 nilfs_destroy_segbuf_cache();
1337 failed_transaction_cache:
1338 nilfs_destroy_transaction_cache();
1341 nilfs_destroy_inode_cache();
1347 static void __exit exit_nilfs_fs(void)
1349 nilfs_destroy_segbuf_cache();
1350 nilfs_destroy_transaction_cache();
1351 nilfs_destroy_inode_cache();
1352 nilfs_btree_path_cache_destroy();
1353 unregister_filesystem(&nilfs_fs_type);
1356 module_init(init_nilfs_fs)
1357 module_exit(exit_nilfs_fs)