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/crc32.h>
47 #include <linux/vfs.h>
48 #include <linux/writeback.h>
49 #include <linux/seq_file.h>
50 #include <linux/mount.h>
64 MODULE_AUTHOR("NTT Corp.");
65 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
67 MODULE_LICENSE("GPL");
69 static struct kmem_cache *nilfs_inode_cachep;
70 struct kmem_cache *nilfs_transaction_cachep;
71 struct kmem_cache *nilfs_segbuf_cachep;
72 struct kmem_cache *nilfs_btree_path_cache;
74 static int nilfs_setup_super(struct nilfs_sb_info *sbi, int is_mount);
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 the_nilfs *nilfs = sbi->s_nilfs;
112 struct va_format vaf;
120 printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
121 sb->s_id, function, &vaf);
125 if (!(sb->s_flags & MS_RDONLY)) {
126 nilfs_set_error(sbi);
128 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
129 printk(KERN_CRIT "Remounting filesystem read-only\n");
130 sb->s_flags |= MS_RDONLY;
134 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
135 panic("NILFS (device %s): panic forced after error\n",
139 void nilfs_warning(struct super_block *sb, const char *function,
140 const char *fmt, ...)
142 struct va_format vaf;
150 printk(KERN_WARNING "NILFS warning (device %s): %s: %pV\n",
151 sb->s_id, function, &vaf);
157 struct inode *nilfs_alloc_inode(struct super_block *sb)
159 struct nilfs_inode_info *ii;
161 ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
167 ii->vfs_inode.i_version = 1;
168 nilfs_btnode_cache_init(&ii->i_btnode_cache, sb->s_bdi);
169 return &ii->vfs_inode;
172 static void nilfs_i_callback(struct rcu_head *head)
174 struct inode *inode = container_of(head, struct inode, i_rcu);
175 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
177 INIT_LIST_HEAD(&inode->i_dentry);
180 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
183 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
186 void nilfs_destroy_inode(struct inode *inode)
188 call_rcu(&inode->i_rcu, nilfs_i_callback);
191 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int flag)
193 struct the_nilfs *nilfs = sbi->s_nilfs;
197 set_buffer_dirty(nilfs->ns_sbh[0]);
198 if (nilfs_test_opt(nilfs, BARRIER)) {
199 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
200 WRITE_SYNC | WRITE_FLUSH_FUA);
202 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
207 "NILFS: unable to write superblock (err=%d)\n", err);
208 if (err == -EIO && nilfs->ns_sbh[1]) {
210 * sbp[0] points to newer log than sbp[1],
211 * so copy sbp[0] to sbp[1] to take over sbp[0].
213 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
215 nilfs_fall_back_super_block(nilfs);
219 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
221 nilfs->ns_sbwcount++;
224 * The latest segment becomes trailable from the position
225 * written in superblock.
227 clear_nilfs_discontinued(nilfs);
229 /* update GC protection for recent segments */
230 if (nilfs->ns_sbh[1]) {
231 if (flag == NILFS_SB_COMMIT_ALL) {
232 set_buffer_dirty(nilfs->ns_sbh[1]);
233 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
236 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
237 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
238 sbp = nilfs->ns_sbp[1];
241 spin_lock(&nilfs->ns_last_segment_lock);
242 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
243 spin_unlock(&nilfs->ns_last_segment_lock);
249 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
250 struct the_nilfs *nilfs)
252 sector_t nfreeblocks;
254 /* nilfs->ns_sem must be locked by the caller. */
255 nilfs_count_free_blocks(nilfs, &nfreeblocks);
256 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
258 spin_lock(&nilfs->ns_last_segment_lock);
259 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
260 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
261 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
262 spin_unlock(&nilfs->ns_last_segment_lock);
265 struct nilfs_super_block **nilfs_prepare_super(struct nilfs_sb_info *sbi,
268 struct the_nilfs *nilfs = sbi->s_nilfs;
269 struct nilfs_super_block **sbp = nilfs->ns_sbp;
271 /* nilfs->ns_sem must be locked by the caller. */
272 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
274 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
275 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
277 printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
282 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
283 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
287 nilfs_swap_super_block(nilfs);
292 int nilfs_commit_super(struct nilfs_sb_info *sbi, int flag)
294 struct the_nilfs *nilfs = sbi->s_nilfs;
295 struct nilfs_super_block **sbp = nilfs->ns_sbp;
298 /* nilfs->ns_sem must be locked by the caller. */
300 nilfs->ns_sbwtime = t;
301 sbp[0]->s_wtime = cpu_to_le64(t);
303 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
304 (unsigned char *)sbp[0],
306 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
307 sbp[1]->s_wtime = sbp[0]->s_wtime;
309 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
310 (unsigned char *)sbp[1],
313 clear_nilfs_sb_dirty(nilfs);
314 return nilfs_sync_super(sbi, flag);
318 * nilfs_cleanup_super() - write filesystem state for cleanup
319 * @sbi: nilfs_sb_info to be unmounted or degraded to read-only
321 * This function restores state flags in the on-disk super block.
322 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
323 * filesystem was not clean previously.
325 int nilfs_cleanup_super(struct nilfs_sb_info *sbi)
327 struct nilfs_super_block **sbp;
328 int flag = NILFS_SB_COMMIT;
331 sbp = nilfs_prepare_super(sbi, 0);
333 sbp[0]->s_state = cpu_to_le16(sbi->s_nilfs->ns_mount_state);
334 nilfs_set_log_cursor(sbp[0], sbi->s_nilfs);
335 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
337 * make the "clean" flag also to the opposite
338 * super block if both super blocks point to
339 * the same checkpoint.
341 sbp[1]->s_state = sbp[0]->s_state;
342 flag = NILFS_SB_COMMIT_ALL;
344 ret = nilfs_commit_super(sbi, flag);
349 static void nilfs_put_super(struct super_block *sb)
351 struct nilfs_sb_info *sbi = NILFS_SB(sb);
352 struct the_nilfs *nilfs = sbi->s_nilfs;
354 nilfs_detach_segment_constructor(sbi);
356 if (!(sb->s_flags & MS_RDONLY)) {
357 down_write(&nilfs->ns_sem);
358 nilfs_cleanup_super(sbi);
359 up_write(&nilfs->ns_sem);
362 iput(nilfs->ns_sufile);
363 iput(nilfs->ns_cpfile);
366 destroy_nilfs(nilfs);
368 sb->s_fs_info = NULL;
372 static int nilfs_sync_fs(struct super_block *sb, int wait)
374 struct nilfs_sb_info *sbi = NILFS_SB(sb);
375 struct the_nilfs *nilfs = sbi->s_nilfs;
376 struct nilfs_super_block **sbp;
379 /* This function is called when super block should be written back */
381 err = nilfs_construct_segment(sb);
383 down_write(&nilfs->ns_sem);
384 if (nilfs_sb_dirty(nilfs)) {
385 sbp = nilfs_prepare_super(sbi, nilfs_sb_will_flip(nilfs));
387 nilfs_set_log_cursor(sbp[0], nilfs);
388 nilfs_commit_super(sbi, NILFS_SB_COMMIT);
391 up_write(&nilfs->ns_sem);
396 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno, int curr_mnt,
397 struct nilfs_root **rootp)
399 struct the_nilfs *nilfs = sbi->s_nilfs;
400 struct nilfs_root *root;
401 struct nilfs_checkpoint *raw_cp;
402 struct buffer_head *bh_cp;
405 root = nilfs_find_or_create_root(
406 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
411 goto reuse; /* already attached checkpoint */
413 down_read(&nilfs->ns_segctor_sem);
414 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
416 up_read(&nilfs->ns_segctor_sem);
418 if (err == -ENOENT || err == -EINVAL) {
420 "NILFS: Invalid checkpoint "
421 "(checkpoint number=%llu)\n",
422 (unsigned long long)cno);
428 err = nilfs_ifile_read(sbi->s_super, root, nilfs->ns_inode_size,
429 &raw_cp->cp_ifile_inode, &root->ifile);
433 atomic_set(&root->inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
434 atomic_set(&root->blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
436 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
443 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
445 nilfs_put_root(root);
450 static int nilfs_freeze(struct super_block *sb)
452 struct nilfs_sb_info *sbi = NILFS_SB(sb);
453 struct the_nilfs *nilfs = sbi->s_nilfs;
456 if (sb->s_flags & MS_RDONLY)
459 /* Mark super block clean */
460 down_write(&nilfs->ns_sem);
461 err = nilfs_cleanup_super(sbi);
462 up_write(&nilfs->ns_sem);
466 static int nilfs_unfreeze(struct super_block *sb)
468 struct nilfs_sb_info *sbi = NILFS_SB(sb);
469 struct the_nilfs *nilfs = sbi->s_nilfs;
471 if (sb->s_flags & MS_RDONLY)
474 down_write(&nilfs->ns_sem);
475 nilfs_setup_super(sbi, false);
476 up_write(&nilfs->ns_sem);
480 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
482 struct super_block *sb = dentry->d_sb;
483 struct nilfs_root *root = NILFS_I(dentry->d_inode)->i_root;
484 struct the_nilfs *nilfs = root->nilfs;
485 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
486 unsigned long long blocks;
487 unsigned long overhead;
488 unsigned long nrsvblocks;
489 sector_t nfreeblocks;
493 * Compute all of the segment blocks
495 * The blocks before first segment and after last segment
498 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
499 - nilfs->ns_first_data_block;
500 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
503 * Compute the overhead
505 * When distributing meta data blocks outside segment structure,
506 * We must count them as the overhead.
510 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
514 buf->f_type = NILFS_SUPER_MAGIC;
515 buf->f_bsize = sb->s_blocksize;
516 buf->f_blocks = blocks - overhead;
517 buf->f_bfree = nfreeblocks;
518 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
519 (buf->f_bfree - nrsvblocks) : 0;
520 buf->f_files = atomic_read(&root->inodes_count);
521 buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
522 buf->f_namelen = NILFS_NAME_LEN;
523 buf->f_fsid.val[0] = (u32)id;
524 buf->f_fsid.val[1] = (u32)(id >> 32);
529 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
531 struct super_block *sb = vfs->mnt_sb;
532 struct the_nilfs *nilfs = NILFS_SB(sb)->s_nilfs;
533 struct nilfs_root *root = NILFS_I(vfs->mnt_root->d_inode)->i_root;
535 if (!nilfs_test_opt(nilfs, BARRIER))
536 seq_puts(seq, ",nobarrier");
537 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
538 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
539 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
540 seq_puts(seq, ",errors=panic");
541 if (nilfs_test_opt(nilfs, ERRORS_CONT))
542 seq_puts(seq, ",errors=continue");
543 if (nilfs_test_opt(nilfs, STRICT_ORDER))
544 seq_puts(seq, ",order=strict");
545 if (nilfs_test_opt(nilfs, NORECOVERY))
546 seq_puts(seq, ",norecovery");
547 if (nilfs_test_opt(nilfs, DISCARD))
548 seq_puts(seq, ",discard");
553 static const struct super_operations nilfs_sops = {
554 .alloc_inode = nilfs_alloc_inode,
555 .destroy_inode = nilfs_destroy_inode,
556 .dirty_inode = nilfs_dirty_inode,
557 /* .write_inode = nilfs_write_inode, */
558 /* .put_inode = nilfs_put_inode, */
559 /* .drop_inode = nilfs_drop_inode, */
560 .evict_inode = nilfs_evict_inode,
561 .put_super = nilfs_put_super,
562 /* .write_super = nilfs_write_super, */
563 .sync_fs = nilfs_sync_fs,
564 .freeze_fs = nilfs_freeze,
565 .unfreeze_fs = nilfs_unfreeze,
566 /* .write_super_lockfs */
568 .statfs = nilfs_statfs,
569 .remount_fs = nilfs_remount,
571 .show_options = nilfs_show_options
575 Opt_err_cont, Opt_err_panic, Opt_err_ro,
576 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
577 Opt_discard, Opt_nodiscard, Opt_err,
580 static match_table_t tokens = {
581 {Opt_err_cont, "errors=continue"},
582 {Opt_err_panic, "errors=panic"},
583 {Opt_err_ro, "errors=remount-ro"},
584 {Opt_barrier, "barrier"},
585 {Opt_nobarrier, "nobarrier"},
586 {Opt_snapshot, "cp=%u"},
587 {Opt_order, "order=%s"},
588 {Opt_norecovery, "norecovery"},
589 {Opt_discard, "discard"},
590 {Opt_nodiscard, "nodiscard"},
594 static int parse_options(char *options, struct super_block *sb, int is_remount)
596 struct the_nilfs *nilfs = NILFS_SB(sb)->s_nilfs;
598 substring_t args[MAX_OPT_ARGS];
603 while ((p = strsep(&options, ",")) != NULL) {
608 token = match_token(p, tokens, args);
611 nilfs_set_opt(nilfs, BARRIER);
614 nilfs_clear_opt(nilfs, BARRIER);
617 if (strcmp(args[0].from, "relaxed") == 0)
618 /* Ordered data semantics */
619 nilfs_clear_opt(nilfs, STRICT_ORDER);
620 else if (strcmp(args[0].from, "strict") == 0)
621 /* Strict in-order semantics */
622 nilfs_set_opt(nilfs, STRICT_ORDER);
627 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
630 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
633 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
638 "NILFS: \"%s\" option is invalid "
639 "for remount.\n", p);
644 nilfs_set_opt(nilfs, NORECOVERY);
647 nilfs_set_opt(nilfs, DISCARD);
650 nilfs_clear_opt(nilfs, DISCARD);
654 "NILFS: Unrecognized mount option \"%s\"\n", p);
662 nilfs_set_default_options(struct super_block *sb,
663 struct nilfs_super_block *sbp)
665 struct the_nilfs *nilfs = NILFS_SB(sb)->s_nilfs;
667 nilfs->ns_mount_opt =
668 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
671 static int nilfs_setup_super(struct nilfs_sb_info *sbi, int is_mount)
673 struct the_nilfs *nilfs = sbi->s_nilfs;
674 struct nilfs_super_block **sbp;
678 /* nilfs->ns_sem must be locked by the caller. */
679 sbp = nilfs_prepare_super(sbi, 0);
684 goto skip_mount_setup;
686 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
687 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
689 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
691 "NILFS warning: mounting fs with errors\n");
693 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
695 "NILFS warning: maximal mount count reached\n");
699 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
701 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
702 sbp[0]->s_mtime = cpu_to_le64(get_seconds());
706 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
707 /* synchronize sbp[1] with sbp[0] */
709 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
710 return nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
713 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
714 u64 pos, int blocksize,
715 struct buffer_head **pbh)
717 unsigned long long sb_index = pos;
718 unsigned long offset;
720 offset = do_div(sb_index, blocksize);
721 *pbh = sb_bread(sb, sb_index);
724 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
727 int nilfs_store_magic_and_option(struct super_block *sb,
728 struct nilfs_super_block *sbp,
731 struct the_nilfs *nilfs = NILFS_SB(sb)->s_nilfs;
733 sb->s_magic = le16_to_cpu(sbp->s_magic);
735 /* FS independent flags */
736 #ifdef NILFS_ATIME_DISABLE
737 sb->s_flags |= MS_NOATIME;
740 nilfs_set_default_options(sb, sbp);
742 nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
743 nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
744 nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
745 nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
747 return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
750 int nilfs_check_feature_compatibility(struct super_block *sb,
751 struct nilfs_super_block *sbp)
755 features = le64_to_cpu(sbp->s_feature_incompat) &
756 ~NILFS_FEATURE_INCOMPAT_SUPP;
758 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
759 "optional features (%llx)\n",
760 (unsigned long long)features);
763 features = le64_to_cpu(sbp->s_feature_compat_ro) &
764 ~NILFS_FEATURE_COMPAT_RO_SUPP;
765 if (!(sb->s_flags & MS_RDONLY) && features) {
766 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
767 "unsupported optional features (%llx)\n",
768 (unsigned long long)features);
774 static int nilfs_get_root_dentry(struct super_block *sb,
775 struct nilfs_root *root,
776 struct dentry **root_dentry)
779 struct dentry *dentry;
782 inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
784 printk(KERN_ERR "NILFS: get root inode failed\n");
785 ret = PTR_ERR(inode);
788 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
790 printk(KERN_ERR "NILFS: corrupt root inode.\n");
795 if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
796 dentry = d_find_alias(inode);
798 dentry = d_alloc_root(inode);
808 dentry = d_obtain_alias(inode);
809 if (IS_ERR(dentry)) {
810 ret = PTR_ERR(dentry);
814 *root_dentry = dentry;
819 printk(KERN_ERR "NILFS: get root dentry failed\n");
823 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
824 struct dentry **root_dentry)
826 struct the_nilfs *nilfs = NILFS_SB(s)->s_nilfs;
827 struct nilfs_root *root;
830 down_read(&nilfs->ns_segctor_sem);
831 ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
832 up_read(&nilfs->ns_segctor_sem);
834 ret = (ret == -ENOENT) ? -EINVAL : ret;
837 printk(KERN_ERR "NILFS: The specified checkpoint is "
838 "not a snapshot (checkpoint number=%llu).\n",
839 (unsigned long long)cno);
844 ret = nilfs_attach_checkpoint(NILFS_SB(s), cno, false, &root);
846 printk(KERN_ERR "NILFS: error loading snapshot "
847 "(checkpoint number=%llu).\n",
848 (unsigned long long)cno);
851 ret = nilfs_get_root_dentry(s, root, root_dentry);
852 nilfs_put_root(root);
857 static int nilfs_tree_was_touched(struct dentry *root_dentry)
859 return root_dentry->d_count > 1;
863 * nilfs_try_to_shrink_tree() - try to shrink dentries of a checkpoint
864 * @root_dentry: root dentry of the tree to be shrunk
866 * This function returns true if the tree was in-use.
868 static int nilfs_try_to_shrink_tree(struct dentry *root_dentry)
870 if (have_submounts(root_dentry))
872 shrink_dcache_parent(root_dentry);
873 return nilfs_tree_was_touched(root_dentry);
876 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
878 struct the_nilfs *nilfs = NILFS_SB(sb)->s_nilfs;
879 struct nilfs_root *root;
881 struct dentry *dentry;
884 if (cno < 0 || cno > nilfs->ns_cno)
887 if (cno >= nilfs_last_cno(nilfs))
888 return true; /* protect recent checkpoints */
891 root = nilfs_lookup_root(NILFS_SB(sb)->s_nilfs, cno);
893 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
895 dentry = d_find_alias(inode);
897 if (nilfs_tree_was_touched(dentry))
898 ret = nilfs_try_to_shrink_tree(dentry);
903 nilfs_put_root(root);
909 * nilfs_fill_super() - initialize a super block instance
911 * @data: mount options
912 * @silent: silent mode flag
914 * This function is called exclusively by nilfs->ns_mount_mutex.
915 * So, the recovery process is protected from other simultaneous mounts.
918 nilfs_fill_super(struct super_block *sb, void *data, int silent)
920 struct the_nilfs *nilfs;
921 struct nilfs_sb_info *sbi;
922 struct nilfs_root *fsroot;
923 struct backing_dev_info *bdi;
927 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
934 nilfs = alloc_nilfs(sb->s_bdev);
939 sbi->s_nilfs = nilfs;
941 err = init_nilfs(nilfs, sbi, (char *)data);
945 sb->s_op = &nilfs_sops;
946 sb->s_export_op = &nilfs_export_ops;
950 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
951 sb->s_bdi = bdi ? : &default_backing_dev_info;
953 err = load_nilfs(nilfs, sbi);
957 cno = nilfs_last_cno(nilfs);
958 err = nilfs_attach_checkpoint(sbi, cno, true, &fsroot);
960 printk(KERN_ERR "NILFS: error loading last checkpoint "
961 "(checkpoint number=%llu).\n", (unsigned long long)cno);
965 if (!(sb->s_flags & MS_RDONLY)) {
966 err = nilfs_attach_segment_constructor(sbi, fsroot);
968 goto failed_checkpoint;
971 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
975 nilfs_put_root(fsroot);
977 if (!(sb->s_flags & MS_RDONLY)) {
978 down_write(&nilfs->ns_sem);
979 nilfs_setup_super(sbi, true);
980 up_write(&nilfs->ns_sem);
986 nilfs_detach_segment_constructor(sbi);
989 nilfs_put_root(fsroot);
992 iput(nilfs->ns_sufile);
993 iput(nilfs->ns_cpfile);
997 destroy_nilfs(nilfs);
1000 sb->s_fs_info = NULL;
1005 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1007 struct nilfs_sb_info *sbi = NILFS_SB(sb);
1008 struct the_nilfs *nilfs = sbi->s_nilfs;
1009 unsigned long old_sb_flags;
1010 unsigned long old_mount_opt;
1013 old_sb_flags = sb->s_flags;
1014 old_mount_opt = nilfs->ns_mount_opt;
1016 if (!parse_options(data, sb, 1)) {
1020 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
1024 if (!nilfs_valid_fs(nilfs)) {
1025 printk(KERN_WARNING "NILFS (device %s): couldn't "
1026 "remount because the filesystem is in an "
1027 "incomplete recovery state.\n", sb->s_id);
1031 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1033 if (*flags & MS_RDONLY) {
1034 /* Shutting down the segment constructor */
1035 nilfs_detach_segment_constructor(sbi);
1036 sb->s_flags |= MS_RDONLY;
1039 * Remounting a valid RW partition RDONLY, so set
1040 * the RDONLY flag and then mark the partition as valid again.
1042 down_write(&nilfs->ns_sem);
1043 nilfs_cleanup_super(sbi);
1044 up_write(&nilfs->ns_sem);
1047 struct nilfs_root *root;
1050 * Mounting a RDONLY partition read-write, so reread and
1051 * store the current valid flag. (It may have been changed
1052 * by fsck since we originally mounted the partition.)
1054 down_read(&nilfs->ns_sem);
1055 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1056 ~NILFS_FEATURE_COMPAT_RO_SUPP;
1057 up_read(&nilfs->ns_sem);
1059 printk(KERN_WARNING "NILFS (device %s): couldn't "
1060 "remount RDWR because of unsupported optional "
1061 "features (%llx)\n",
1062 sb->s_id, (unsigned long long)features);
1067 sb->s_flags &= ~MS_RDONLY;
1069 root = NILFS_I(sb->s_root->d_inode)->i_root;
1070 err = nilfs_attach_segment_constructor(sbi, root);
1074 down_write(&nilfs->ns_sem);
1075 nilfs_setup_super(sbi, true);
1076 up_write(&nilfs->ns_sem);
1082 sb->s_flags = old_sb_flags;
1083 nilfs->ns_mount_opt = old_mount_opt;
1087 struct nilfs_super_data {
1088 struct block_device *bdev;
1089 struct nilfs_sb_info *sbi;
1095 * nilfs_identify - pre-read mount options needed to identify mount instance
1096 * @data: mount options
1097 * @sd: nilfs_super_data
1099 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1101 char *p, *options = data;
1102 substring_t args[MAX_OPT_ARGS];
1107 p = strsep(&options, ",");
1108 if (p != NULL && *p) {
1109 token = match_token(p, tokens, args);
1110 if (token == Opt_snapshot) {
1111 if (!(sd->flags & MS_RDONLY)) {
1114 sd->cno = simple_strtoull(args[0].from,
1117 * No need to see the end pointer;
1118 * match_token() has done syntax
1127 "NILFS: invalid mount option: %s\n", p);
1131 BUG_ON(options == data);
1132 *(options - 1) = ',';
1137 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1140 s->s_dev = s->s_bdev->bd_dev;
1144 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1146 return (void *)s->s_bdev == data;
1149 static struct dentry *
1150 nilfs_mount(struct file_system_type *fs_type, int flags,
1151 const char *dev_name, void *data)
1153 struct nilfs_super_data sd;
1154 struct super_block *s;
1155 fmode_t mode = FMODE_READ | FMODE_EXCL;
1156 struct dentry *root_dentry;
1157 int err, s_new = false;
1159 if (!(flags & MS_RDONLY))
1160 mode |= FMODE_WRITE;
1162 sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1163 if (IS_ERR(sd.bdev))
1164 return ERR_CAST(sd.bdev);
1168 if (nilfs_identify((char *)data, &sd)) {
1174 * once the super is inserted into the list by sget, s_umount
1175 * will protect the lockfs code from trying to start a snapshot
1176 * while we are mounting
1178 mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1179 if (sd.bdev->bd_fsfreeze_count > 0) {
1180 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1184 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, sd.bdev);
1185 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1192 char b[BDEVNAME_SIZE];
1196 /* New superblock instance created */
1199 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1200 sb_set_blocksize(s, block_size(sd.bdev));
1202 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1206 s->s_flags |= MS_ACTIVE;
1207 } else if (!sd.cno) {
1210 if (nilfs_tree_was_touched(s->s_root)) {
1211 busy = nilfs_try_to_shrink_tree(s->s_root);
1212 if (busy && (flags ^ s->s_flags) & MS_RDONLY) {
1213 printk(KERN_ERR "NILFS: the device already "
1214 "has a %s mount.\n",
1215 (s->s_flags & MS_RDONLY) ?
1216 "read-only" : "read/write");
1223 * Try remount to setup mount states if the current
1224 * tree is not mounted and only snapshots use this sb.
1226 err = nilfs_remount(s, &flags, data);
1233 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1237 root_dentry = dget(s->s_root);
1241 blkdev_put(sd.bdev, mode);
1246 deactivate_locked_super(s);
1250 blkdev_put(sd.bdev, mode);
1251 return ERR_PTR(err);
1254 struct file_system_type nilfs_fs_type = {
1255 .owner = THIS_MODULE,
1257 .mount = nilfs_mount,
1258 .kill_sb = kill_block_super,
1259 .fs_flags = FS_REQUIRES_DEV,
1262 static void nilfs_inode_init_once(void *obj)
1264 struct nilfs_inode_info *ii = obj;
1266 INIT_LIST_HEAD(&ii->i_dirty);
1267 #ifdef CONFIG_NILFS_XATTR
1268 init_rwsem(&ii->xattr_sem);
1270 address_space_init_once(&ii->i_btnode_cache);
1271 ii->i_bmap = &ii->i_bmap_data;
1272 inode_init_once(&ii->vfs_inode);
1275 static void nilfs_segbuf_init_once(void *obj)
1277 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1280 static void nilfs_destroy_cachep(void)
1282 if (nilfs_inode_cachep)
1283 kmem_cache_destroy(nilfs_inode_cachep);
1284 if (nilfs_transaction_cachep)
1285 kmem_cache_destroy(nilfs_transaction_cachep);
1286 if (nilfs_segbuf_cachep)
1287 kmem_cache_destroy(nilfs_segbuf_cachep);
1288 if (nilfs_btree_path_cache)
1289 kmem_cache_destroy(nilfs_btree_path_cache);
1292 static int __init nilfs_init_cachep(void)
1294 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1295 sizeof(struct nilfs_inode_info), 0,
1296 SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1297 if (!nilfs_inode_cachep)
1300 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1301 sizeof(struct nilfs_transaction_info), 0,
1302 SLAB_RECLAIM_ACCOUNT, NULL);
1303 if (!nilfs_transaction_cachep)
1306 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1307 sizeof(struct nilfs_segment_buffer), 0,
1308 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1309 if (!nilfs_segbuf_cachep)
1312 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1313 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1315 if (!nilfs_btree_path_cache)
1321 nilfs_destroy_cachep();
1325 static int __init init_nilfs_fs(void)
1329 err = nilfs_init_cachep();
1333 err = register_filesystem(&nilfs_fs_type);
1337 printk(KERN_INFO "NILFS version 2 loaded\n");
1341 nilfs_destroy_cachep();
1346 static void __exit exit_nilfs_fs(void)
1348 nilfs_destroy_cachep();
1349 unregister_filesystem(&nilfs_fs_type);
1352 module_init(init_nilfs_fs)
1353 module_exit(exit_nilfs_fs)