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>
53 #include <linux/seq_file.h>
54 #include <linux/mount.h>
65 MODULE_AUTHOR("NTT Corp.");
66 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
68 MODULE_LICENSE("GPL");
70 struct kmem_cache *nilfs_inode_cachep;
71 struct kmem_cache *nilfs_transaction_cachep;
72 struct kmem_cache *nilfs_segbuf_cachep;
73 struct kmem_cache *nilfs_btree_path_cache;
75 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
77 static void nilfs_set_error(struct nilfs_sb_info *sbi)
79 struct the_nilfs *nilfs = sbi->s_nilfs;
80 struct nilfs_super_block **sbp;
82 down_write(&nilfs->ns_sem);
83 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
84 nilfs->ns_mount_state |= NILFS_ERROR_FS;
85 sbp = nilfs_prepare_super(sbi, 0);
87 sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
89 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
90 nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
93 up_write(&nilfs->ns_sem);
97 * nilfs_error() - report failure condition on a filesystem
99 * nilfs_error() sets an ERROR_FS flag on the superblock as well as
100 * reporting an error message. It should be called when NILFS detects
101 * incoherences or defects of meta data on disk. As for sustainable
102 * errors such as a single-shot I/O error, nilfs_warning() or the printk()
103 * function should be used instead.
105 * The segment constructor must not call this function because it can
108 void nilfs_error(struct super_block *sb, const char *function,
109 const char *fmt, ...)
111 struct nilfs_sb_info *sbi = NILFS_SB(sb);
115 printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
120 if (!(sb->s_flags & MS_RDONLY)) {
121 nilfs_set_error(sbi);
123 if (nilfs_test_opt(sbi, ERRORS_RO)) {
124 printk(KERN_CRIT "Remounting filesystem read-only\n");
125 sb->s_flags |= MS_RDONLY;
129 if (nilfs_test_opt(sbi, ERRORS_PANIC))
130 panic("NILFS (device %s): panic forced after error\n",
134 void nilfs_warning(struct super_block *sb, const char *function,
135 const char *fmt, ...)
140 printk(KERN_WARNING "NILFS warning (device %s): %s: ",
148 struct inode *nilfs_alloc_inode_common(struct the_nilfs *nilfs)
150 struct nilfs_inode_info *ii;
152 ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
157 ii->vfs_inode.i_version = 1;
158 nilfs_btnode_cache_init(&ii->i_btnode_cache, nilfs->ns_bdi);
159 return &ii->vfs_inode;
162 struct inode *nilfs_alloc_inode(struct super_block *sb)
164 return nilfs_alloc_inode_common(NILFS_SB(sb)->s_nilfs);
167 void nilfs_destroy_inode(struct inode *inode)
169 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
172 static void nilfs_clear_inode(struct inode *inode)
174 struct nilfs_inode_info *ii = NILFS_I(inode);
177 * Free resources allocated in nilfs_read_inode(), here.
179 BUG_ON(!list_empty(&ii->i_dirty));
183 if (test_bit(NILFS_I_BMAP, &ii->i_state))
184 nilfs_bmap_clear(ii->i_bmap);
186 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
189 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int flag)
191 struct the_nilfs *nilfs = sbi->s_nilfs;
193 int barrier_done = 0;
195 if (nilfs_test_opt(sbi, BARRIER)) {
196 set_buffer_ordered(nilfs->ns_sbh[0]);
200 set_buffer_dirty(nilfs->ns_sbh[0]);
201 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
202 if (err == -EOPNOTSUPP && barrier_done) {
203 nilfs_warning(sbi->s_super, __func__,
204 "barrier-based sync failed. "
205 "disabling barriers\n");
206 nilfs_clear_opt(sbi, BARRIER);
208 clear_buffer_ordered(nilfs->ns_sbh[0]);
213 "NILFS: unable to write superblock (err=%d)\n", err);
214 if (err == -EIO && nilfs->ns_sbh[1]) {
216 * sbp[0] points to newer log than sbp[1],
217 * so copy sbp[0] to sbp[1] to take over sbp[0].
219 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
221 nilfs_fall_back_super_block(nilfs);
225 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
227 nilfs->ns_sbwcount++;
230 * The latest segment becomes trailable from the position
231 * written in superblock.
233 clear_nilfs_discontinued(nilfs);
235 /* update GC protection for recent segments */
236 if (nilfs->ns_sbh[1]) {
237 if (flag == NILFS_SB_COMMIT_ALL) {
238 set_buffer_dirty(nilfs->ns_sbh[1]);
239 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
242 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
243 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
244 sbp = nilfs->ns_sbp[1];
247 spin_lock(&nilfs->ns_last_segment_lock);
248 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
249 spin_unlock(&nilfs->ns_last_segment_lock);
255 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
256 struct the_nilfs *nilfs)
258 sector_t nfreeblocks;
260 /* nilfs->ns_sem must be locked by the caller. */
261 nilfs_count_free_blocks(nilfs, &nfreeblocks);
262 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
264 spin_lock(&nilfs->ns_last_segment_lock);
265 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
266 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
267 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
268 spin_unlock(&nilfs->ns_last_segment_lock);
271 struct nilfs_super_block **nilfs_prepare_super(struct nilfs_sb_info *sbi,
274 struct the_nilfs *nilfs = sbi->s_nilfs;
275 struct nilfs_super_block **sbp = nilfs->ns_sbp;
277 /* nilfs->ns_sem must be locked by the caller. */
278 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
280 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
281 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
283 printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
288 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
289 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
293 nilfs_swap_super_block(nilfs);
298 int nilfs_commit_super(struct nilfs_sb_info *sbi, int flag)
300 struct the_nilfs *nilfs = sbi->s_nilfs;
301 struct nilfs_super_block **sbp = nilfs->ns_sbp;
304 /* nilfs->ns_sem must be locked by the caller. */
306 nilfs->ns_sbwtime = t;
307 sbp[0]->s_wtime = cpu_to_le64(t);
309 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
310 (unsigned char *)sbp[0],
312 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
313 sbp[1]->s_wtime = sbp[0]->s_wtime;
315 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
316 (unsigned char *)sbp[1],
319 clear_nilfs_sb_dirty(nilfs);
320 return nilfs_sync_super(sbi, flag);
324 * nilfs_cleanup_super() - write filesystem state for cleanup
325 * @sbi: nilfs_sb_info to be unmounted or degraded to read-only
327 * This function restores state flags in the on-disk super block.
328 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
329 * filesystem was not clean previously.
331 int nilfs_cleanup_super(struct nilfs_sb_info *sbi)
333 struct nilfs_super_block **sbp;
334 int flag = NILFS_SB_COMMIT;
337 sbp = nilfs_prepare_super(sbi, 0);
339 sbp[0]->s_state = cpu_to_le16(sbi->s_nilfs->ns_mount_state);
340 nilfs_set_log_cursor(sbp[0], sbi->s_nilfs);
341 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
343 * make the "clean" flag also to the opposite
344 * super block if both super blocks point to
345 * the same checkpoint.
347 sbp[1]->s_state = sbp[0]->s_state;
348 flag = NILFS_SB_COMMIT_ALL;
350 ret = nilfs_commit_super(sbi, flag);
355 static void nilfs_put_super(struct super_block *sb)
357 struct nilfs_sb_info *sbi = NILFS_SB(sb);
358 struct the_nilfs *nilfs = sbi->s_nilfs;
362 nilfs_detach_segment_constructor(sbi);
364 if (!(sb->s_flags & MS_RDONLY)) {
365 down_write(&nilfs->ns_sem);
366 nilfs_cleanup_super(sbi);
367 up_write(&nilfs->ns_sem);
369 down_write(&nilfs->ns_super_sem);
370 if (nilfs->ns_current == sbi)
371 nilfs->ns_current = NULL;
372 up_write(&nilfs->ns_super_sem);
374 nilfs_detach_checkpoint(sbi);
375 put_nilfs(sbi->s_nilfs);
377 sb->s_fs_info = NULL;
378 nilfs_put_sbinfo(sbi);
383 static int nilfs_sync_fs(struct super_block *sb, int wait)
385 struct nilfs_sb_info *sbi = NILFS_SB(sb);
386 struct the_nilfs *nilfs = sbi->s_nilfs;
387 struct nilfs_super_block **sbp;
390 /* This function is called when super block should be written back */
392 err = nilfs_construct_segment(sb);
394 down_write(&nilfs->ns_sem);
395 if (nilfs_sb_dirty(nilfs)) {
396 sbp = nilfs_prepare_super(sbi, nilfs_sb_will_flip(nilfs));
398 nilfs_set_log_cursor(sbp[0], nilfs);
399 nilfs_commit_super(sbi, NILFS_SB_COMMIT);
402 up_write(&nilfs->ns_sem);
407 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
409 struct the_nilfs *nilfs = sbi->s_nilfs;
410 struct nilfs_checkpoint *raw_cp;
411 struct buffer_head *bh_cp;
414 down_write(&nilfs->ns_super_sem);
415 list_add(&sbi->s_list, &nilfs->ns_supers);
416 up_write(&nilfs->ns_super_sem);
418 sbi->s_ifile = nilfs_ifile_new(sbi, nilfs->ns_inode_size);
422 down_read(&nilfs->ns_segctor_sem);
423 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
425 up_read(&nilfs->ns_segctor_sem);
427 if (err == -ENOENT || err == -EINVAL) {
429 "NILFS: Invalid checkpoint "
430 "(checkpoint number=%llu)\n",
431 (unsigned long long)cno);
436 err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
439 atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
440 atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
442 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
446 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
448 nilfs_mdt_destroy(sbi->s_ifile);
451 down_write(&nilfs->ns_super_sem);
452 list_del_init(&sbi->s_list);
453 up_write(&nilfs->ns_super_sem);
458 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
460 struct the_nilfs *nilfs = sbi->s_nilfs;
462 nilfs_mdt_destroy(sbi->s_ifile);
464 down_write(&nilfs->ns_super_sem);
465 list_del_init(&sbi->s_list);
466 up_write(&nilfs->ns_super_sem);
469 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
471 struct super_block *sb = dentry->d_sb;
472 struct nilfs_sb_info *sbi = NILFS_SB(sb);
473 struct the_nilfs *nilfs = sbi->s_nilfs;
474 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
475 unsigned long long blocks;
476 unsigned long overhead;
477 unsigned long nrsvblocks;
478 sector_t nfreeblocks;
482 * Compute all of the segment blocks
484 * The blocks before first segment and after last segment
487 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
488 - nilfs->ns_first_data_block;
489 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
492 * Compute the overhead
494 * When distributing meta data blocks outside segment structure,
495 * We must count them as the overhead.
499 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
503 buf->f_type = NILFS_SUPER_MAGIC;
504 buf->f_bsize = sb->s_blocksize;
505 buf->f_blocks = blocks - overhead;
506 buf->f_bfree = nfreeblocks;
507 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
508 (buf->f_bfree - nrsvblocks) : 0;
509 buf->f_files = atomic_read(&sbi->s_inodes_count);
510 buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
511 buf->f_namelen = NILFS_NAME_LEN;
512 buf->f_fsid.val[0] = (u32)id;
513 buf->f_fsid.val[1] = (u32)(id >> 32);
518 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
520 struct super_block *sb = vfs->mnt_sb;
521 struct nilfs_sb_info *sbi = NILFS_SB(sb);
523 if (!nilfs_test_opt(sbi, BARRIER))
524 seq_printf(seq, ",nobarrier");
525 if (nilfs_test_opt(sbi, SNAPSHOT))
526 seq_printf(seq, ",cp=%llu",
527 (unsigned long long int)sbi->s_snapshot_cno);
528 if (nilfs_test_opt(sbi, ERRORS_PANIC))
529 seq_printf(seq, ",errors=panic");
530 if (nilfs_test_opt(sbi, ERRORS_CONT))
531 seq_printf(seq, ",errors=continue");
532 if (nilfs_test_opt(sbi, STRICT_ORDER))
533 seq_printf(seq, ",order=strict");
534 if (nilfs_test_opt(sbi, NORECOVERY))
535 seq_printf(seq, ",norecovery");
536 if (nilfs_test_opt(sbi, DISCARD))
537 seq_printf(seq, ",discard");
542 static const struct super_operations nilfs_sops = {
543 .alloc_inode = nilfs_alloc_inode,
544 .destroy_inode = nilfs_destroy_inode,
545 .dirty_inode = nilfs_dirty_inode,
546 /* .write_inode = nilfs_write_inode, */
547 /* .put_inode = nilfs_put_inode, */
548 /* .drop_inode = nilfs_drop_inode, */
549 .delete_inode = nilfs_delete_inode,
550 .put_super = nilfs_put_super,
551 /* .write_super = nilfs_write_super, */
552 .sync_fs = nilfs_sync_fs,
553 /* .write_super_lockfs */
555 .statfs = nilfs_statfs,
556 .remount_fs = nilfs_remount,
557 .clear_inode = nilfs_clear_inode,
559 .show_options = nilfs_show_options
562 static struct inode *
563 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
567 if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
568 ino != NILFS_SKETCH_INO)
569 return ERR_PTR(-ESTALE);
571 inode = nilfs_iget(sb, ino);
573 return ERR_CAST(inode);
574 if (generation && inode->i_generation != generation) {
576 return ERR_PTR(-ESTALE);
582 static struct dentry *
583 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
586 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
587 nilfs_nfs_get_inode);
590 static struct dentry *
591 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
594 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
595 nilfs_nfs_get_inode);
598 static const struct export_operations nilfs_export_ops = {
599 .fh_to_dentry = nilfs_fh_to_dentry,
600 .fh_to_parent = nilfs_fh_to_parent,
601 .get_parent = nilfs_get_parent,
605 Opt_err_cont, Opt_err_panic, Opt_err_ro,
606 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
607 Opt_discard, Opt_err,
610 static match_table_t tokens = {
611 {Opt_err_cont, "errors=continue"},
612 {Opt_err_panic, "errors=panic"},
613 {Opt_err_ro, "errors=remount-ro"},
614 {Opt_barrier, "barrier"},
615 {Opt_nobarrier, "nobarrier"},
616 {Opt_snapshot, "cp=%u"},
617 {Opt_order, "order=%s"},
618 {Opt_norecovery, "norecovery"},
619 {Opt_discard, "discard"},
623 static int parse_options(char *options, struct super_block *sb)
625 struct nilfs_sb_info *sbi = NILFS_SB(sb);
627 substring_t args[MAX_OPT_ARGS];
633 while ((p = strsep(&options, ",")) != NULL) {
638 token = match_token(p, tokens, args);
641 nilfs_set_opt(sbi, BARRIER);
644 nilfs_clear_opt(sbi, BARRIER);
647 if (strcmp(args[0].from, "relaxed") == 0)
648 /* Ordered data semantics */
649 nilfs_clear_opt(sbi, STRICT_ORDER);
650 else if (strcmp(args[0].from, "strict") == 0)
651 /* Strict in-order semantics */
652 nilfs_set_opt(sbi, STRICT_ORDER);
657 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
660 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
663 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
666 if (match_int(&args[0], &option) || option <= 0)
668 if (!(sb->s_flags & MS_RDONLY))
670 sbi->s_snapshot_cno = option;
671 nilfs_set_opt(sbi, SNAPSHOT);
674 nilfs_set_opt(sbi, NORECOVERY);
677 nilfs_set_opt(sbi, DISCARD);
681 "NILFS: Unrecognized mount option \"%s\"\n", p);
689 nilfs_set_default_options(struct nilfs_sb_info *sbi,
690 struct nilfs_super_block *sbp)
693 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
696 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
698 struct the_nilfs *nilfs = sbi->s_nilfs;
699 struct nilfs_super_block **sbp;
703 /* nilfs->ns_sem must be locked by the caller. */
704 sbp = nilfs_prepare_super(sbi, 0);
708 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
709 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
711 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
713 "NILFS warning: mounting fs with errors\n");
715 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
717 "NILFS warning: maximal mount count reached\n");
721 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
723 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
725 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
726 sbp[0]->s_mtime = cpu_to_le64(get_seconds());
727 /* synchronize sbp[1] with sbp[0] */
728 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
729 return nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
732 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
733 u64 pos, int blocksize,
734 struct buffer_head **pbh)
736 unsigned long long sb_index = pos;
737 unsigned long offset;
739 offset = do_div(sb_index, blocksize);
740 *pbh = sb_bread(sb, sb_index);
743 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
746 int nilfs_store_magic_and_option(struct super_block *sb,
747 struct nilfs_super_block *sbp,
750 struct nilfs_sb_info *sbi = NILFS_SB(sb);
752 sb->s_magic = le16_to_cpu(sbp->s_magic);
754 /* FS independent flags */
755 #ifdef NILFS_ATIME_DISABLE
756 sb->s_flags |= MS_NOATIME;
759 nilfs_set_default_options(sbi, sbp);
761 sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
762 sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
763 sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
764 sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
766 return !parse_options(data, sb) ? -EINVAL : 0 ;
770 * nilfs_fill_super() - initialize a super block instance
772 * @data: mount options
773 * @silent: silent mode flag
774 * @nilfs: the_nilfs struct
776 * This function is called exclusively by nilfs->ns_mount_mutex.
777 * So, the recovery process is protected from other simultaneous mounts.
780 nilfs_fill_super(struct super_block *sb, void *data, int silent,
781 struct the_nilfs *nilfs)
783 struct nilfs_sb_info *sbi;
788 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
795 sbi->s_nilfs = nilfs;
797 atomic_set(&sbi->s_count, 1);
799 err = init_nilfs(nilfs, sbi, (char *)data);
803 spin_lock_init(&sbi->s_inode_lock);
804 INIT_LIST_HEAD(&sbi->s_dirty_files);
805 INIT_LIST_HEAD(&sbi->s_list);
808 * Following initialization is overlapped because
809 * nilfs_sb_info structure has been cleared at the beginning.
810 * But we reserve them to keep our interest and make ready
811 * for the future change.
813 get_random_bytes(&sbi->s_next_generation,
814 sizeof(sbi->s_next_generation));
815 spin_lock_init(&sbi->s_next_gen_lock);
817 sb->s_op = &nilfs_sops;
818 sb->s_export_op = &nilfs_export_ops;
821 sb->s_bdi = nilfs->ns_bdi;
823 err = load_nilfs(nilfs, sbi);
827 cno = nilfs_last_cno(nilfs);
829 if (sb->s_flags & MS_RDONLY) {
830 if (nilfs_test_opt(sbi, SNAPSHOT)) {
831 down_read(&nilfs->ns_segctor_sem);
832 err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
833 sbi->s_snapshot_cno);
834 up_read(&nilfs->ns_segctor_sem);
842 "NILFS: The specified checkpoint is "
844 "(checkpoint number=%llu).\n",
845 (unsigned long long)sbi->s_snapshot_cno);
849 cno = sbi->s_snapshot_cno;
853 err = nilfs_attach_checkpoint(sbi, cno);
855 printk(KERN_ERR "NILFS: error loading a checkpoint"
856 " (checkpoint number=%llu).\n", (unsigned long long)cno);
860 if (!(sb->s_flags & MS_RDONLY)) {
861 err = nilfs_attach_segment_constructor(sbi);
863 goto failed_checkpoint;
866 root = nilfs_iget(sb, NILFS_ROOT_INO);
868 printk(KERN_ERR "NILFS: get root inode failed\n");
872 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
874 printk(KERN_ERR "NILFS: corrupt root inode.\n");
878 sb->s_root = d_alloc_root(root);
881 printk(KERN_ERR "NILFS: get root dentry failed\n");
886 if (!(sb->s_flags & MS_RDONLY)) {
887 down_write(&nilfs->ns_sem);
888 nilfs_setup_super(sbi);
889 up_write(&nilfs->ns_sem);
892 down_write(&nilfs->ns_super_sem);
893 if (!nilfs_test_opt(sbi, SNAPSHOT))
894 nilfs->ns_current = sbi;
895 up_write(&nilfs->ns_super_sem);
900 nilfs_detach_segment_constructor(sbi);
903 nilfs_detach_checkpoint(sbi);
907 sb->s_fs_info = NULL;
908 nilfs_put_sbinfo(sbi);
912 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
914 struct nilfs_sb_info *sbi = NILFS_SB(sb);
915 struct the_nilfs *nilfs = sbi->s_nilfs;
916 unsigned long old_sb_flags;
917 struct nilfs_mount_options old_opts;
918 int was_snapshot, err;
922 down_write(&nilfs->ns_super_sem);
923 old_sb_flags = sb->s_flags;
924 old_opts.mount_opt = sbi->s_mount_opt;
925 old_opts.snapshot_cno = sbi->s_snapshot_cno;
926 was_snapshot = nilfs_test_opt(sbi, SNAPSHOT);
928 if (!parse_options(data, sb)) {
932 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
936 if (!(*flags & MS_RDONLY)) {
937 printk(KERN_ERR "NILFS (device %s): cannot remount "
938 "snapshot read/write.\n",
941 } else if (sbi->s_snapshot_cno != old_opts.snapshot_cno) {
942 printk(KERN_ERR "NILFS (device %s): cannot "
943 "remount to a different snapshot.\n",
948 if (nilfs_test_opt(sbi, SNAPSHOT)) {
949 printk(KERN_ERR "NILFS (device %s): cannot change "
950 "a regular mount to a snapshot.\n",
956 if (!nilfs_valid_fs(nilfs)) {
957 printk(KERN_WARNING "NILFS (device %s): couldn't "
958 "remount because the filesystem is in an "
959 "incomplete recovery state.\n", sb->s_id);
963 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
965 if (*flags & MS_RDONLY) {
966 /* Shutting down the segment constructor */
967 nilfs_detach_segment_constructor(sbi);
968 sb->s_flags |= MS_RDONLY;
971 * Remounting a valid RW partition RDONLY, so set
972 * the RDONLY flag and then mark the partition as valid again.
974 down_write(&nilfs->ns_sem);
975 nilfs_cleanup_super(sbi);
976 up_write(&nilfs->ns_sem);
979 * Mounting a RDONLY partition read-write, so reread and
980 * store the current valid flag. (It may have been changed
981 * by fsck since we originally mounted the partition.)
983 sb->s_flags &= ~MS_RDONLY;
985 err = nilfs_attach_segment_constructor(sbi);
989 down_write(&nilfs->ns_sem);
990 nilfs_setup_super(sbi);
991 up_write(&nilfs->ns_sem);
994 up_write(&nilfs->ns_super_sem);
999 sb->s_flags = old_sb_flags;
1000 sbi->s_mount_opt = old_opts.mount_opt;
1001 sbi->s_snapshot_cno = old_opts.snapshot_cno;
1002 up_write(&nilfs->ns_super_sem);
1007 struct nilfs_super_data {
1008 struct block_device *bdev;
1009 struct nilfs_sb_info *sbi;
1015 * nilfs_identify - pre-read mount options needed to identify mount instance
1016 * @data: mount options
1017 * @sd: nilfs_super_data
1019 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1021 char *p, *options = data;
1022 substring_t args[MAX_OPT_ARGS];
1027 p = strsep(&options, ",");
1028 if (p != NULL && *p) {
1029 token = match_token(p, tokens, args);
1030 if (token == Opt_snapshot) {
1031 if (!(sd->flags & MS_RDONLY))
1034 ret = match_int(&args[0], &option);
1045 "NILFS: invalid mount option: %s\n", p);
1049 BUG_ON(options == data);
1050 *(options - 1) = ',';
1055 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1057 struct nilfs_super_data *sd = data;
1059 s->s_bdev = sd->bdev;
1060 s->s_dev = s->s_bdev->bd_dev;
1064 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1066 struct nilfs_super_data *sd = data;
1068 return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1072 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1073 const char *dev_name, void *data, struct vfsmount *mnt)
1075 struct nilfs_super_data sd;
1076 struct super_block *s;
1077 fmode_t mode = FMODE_READ;
1078 struct the_nilfs *nilfs;
1079 int err, need_to_close = 1;
1081 if (!(flags & MS_RDONLY))
1082 mode |= FMODE_WRITE;
1084 sd.bdev = open_bdev_exclusive(dev_name, mode, fs_type);
1085 if (IS_ERR(sd.bdev))
1086 return PTR_ERR(sd.bdev);
1089 * To get mount instance using sget() vfs-routine, NILFS needs
1090 * much more information than normal filesystems to identify mount
1091 * instance. For snapshot mounts, not only a mount type (ro-mount
1092 * or rw-mount) but also a checkpoint number is required.
1096 if (nilfs_identify((char *)data, &sd)) {
1101 nilfs = find_or_create_nilfs(sd.bdev);
1107 mutex_lock(&nilfs->ns_mount_mutex);
1111 * Check if an exclusive mount exists or not.
1112 * Snapshot mounts coexist with a current mount
1113 * (i.e. rw-mount or ro-mount), whereas rw-mount and
1114 * ro-mount are mutually exclusive.
1116 down_read(&nilfs->ns_super_sem);
1117 if (nilfs->ns_current &&
1118 ((nilfs->ns_current->s_super->s_flags ^ flags)
1120 up_read(&nilfs->ns_super_sem);
1124 up_read(&nilfs->ns_super_sem);
1128 * Find existing nilfs_sb_info struct
1130 sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1133 * Get super block instance holding the nilfs_sb_info struct.
1134 * A new instance is allocated if no existing mount is present or
1135 * existing instance has been unmounted.
1137 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1139 nilfs_put_sbinfo(sd.sbi);
1147 char b[BDEVNAME_SIZE];
1149 /* New superblock instance created */
1152 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1153 sb_set_blocksize(s, block_size(sd.bdev));
1155 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0,
1160 s->s_flags |= MS_ACTIVE;
1164 mutex_unlock(&nilfs->ns_mount_mutex);
1167 close_bdev_exclusive(sd.bdev, mode);
1168 simple_set_mnt(mnt, s);
1172 mutex_unlock(&nilfs->ns_mount_mutex);
1175 close_bdev_exclusive(sd.bdev, mode);
1180 /* Abandoning the newly allocated superblock */
1181 mutex_unlock(&nilfs->ns_mount_mutex);
1183 deactivate_locked_super(s);
1185 * deactivate_locked_super() invokes close_bdev_exclusive().
1186 * We must finish all post-cleaning before this call;
1187 * put_nilfs() needs the block device.
1192 struct file_system_type nilfs_fs_type = {
1193 .owner = THIS_MODULE,
1195 .get_sb = nilfs_get_sb,
1196 .kill_sb = kill_block_super,
1197 .fs_flags = FS_REQUIRES_DEV,
1200 static void nilfs_inode_init_once(void *obj)
1202 struct nilfs_inode_info *ii = obj;
1204 INIT_LIST_HEAD(&ii->i_dirty);
1205 #ifdef CONFIG_NILFS_XATTR
1206 init_rwsem(&ii->xattr_sem);
1208 nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
1209 ii->i_bmap = (struct nilfs_bmap *)&ii->i_bmap_union;
1210 inode_init_once(&ii->vfs_inode);
1213 static void nilfs_segbuf_init_once(void *obj)
1215 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1218 static void nilfs_destroy_cachep(void)
1220 if (nilfs_inode_cachep)
1221 kmem_cache_destroy(nilfs_inode_cachep);
1222 if (nilfs_transaction_cachep)
1223 kmem_cache_destroy(nilfs_transaction_cachep);
1224 if (nilfs_segbuf_cachep)
1225 kmem_cache_destroy(nilfs_segbuf_cachep);
1226 if (nilfs_btree_path_cache)
1227 kmem_cache_destroy(nilfs_btree_path_cache);
1230 static int __init nilfs_init_cachep(void)
1232 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1233 sizeof(struct nilfs_inode_info), 0,
1234 SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1235 if (!nilfs_inode_cachep)
1238 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1239 sizeof(struct nilfs_transaction_info), 0,
1240 SLAB_RECLAIM_ACCOUNT, NULL);
1241 if (!nilfs_transaction_cachep)
1244 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1245 sizeof(struct nilfs_segment_buffer), 0,
1246 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1247 if (!nilfs_segbuf_cachep)
1250 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1251 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1253 if (!nilfs_btree_path_cache)
1259 nilfs_destroy_cachep();
1263 static int __init init_nilfs_fs(void)
1267 err = nilfs_init_cachep();
1271 err = register_filesystem(&nilfs_fs_type);
1275 printk(KERN_INFO "NILFS version 2 loaded\n");
1279 nilfs_destroy_cachep();
1284 static void __exit exit_nilfs_fs(void)
1286 nilfs_destroy_cachep();
1287 unregister_filesystem(&nilfs_fs_type);
1290 module_init(init_nilfs_fs)
1291 module_exit(exit_nilfs_fs)