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/vfs.h>
49 #include <linux/writeback.h>
50 #include <linux/kobject.h>
51 #include <linux/seq_file.h>
52 #include <linux/mount.h>
66 MODULE_AUTHOR("NTT Corp.");
67 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
69 MODULE_LICENSE("GPL");
71 static struct kmem_cache *nilfs_inode_cachep;
72 struct kmem_cache *nilfs_transaction_cachep;
73 struct kmem_cache *nilfs_segbuf_cachep;
74 struct kmem_cache *nilfs_btree_path_cache;
76 static int nilfs_setup_super(struct nilfs_sb_info *sbi, int is_mount);
77 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
79 static void nilfs_set_error(struct nilfs_sb_info *sbi)
81 struct the_nilfs *nilfs = sbi->s_nilfs;
82 struct nilfs_super_block **sbp;
84 down_write(&nilfs->ns_sem);
85 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
86 nilfs->ns_mount_state |= NILFS_ERROR_FS;
87 sbp = nilfs_prepare_super(sbi, 0);
89 sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
91 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
92 nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
95 up_write(&nilfs->ns_sem);
99 * nilfs_error() - report failure condition on a filesystem
101 * nilfs_error() sets an ERROR_FS flag on the superblock as well as
102 * reporting an error message. It should be called when NILFS detects
103 * incoherences or defects of meta data on disk. As for sustainable
104 * errors such as a single-shot I/O error, nilfs_warning() or the printk()
105 * function should be used instead.
107 * The segment constructor must not call this function because it can
110 void nilfs_error(struct super_block *sb, const char *function,
111 const char *fmt, ...)
113 struct nilfs_sb_info *sbi = NILFS_SB(sb);
117 printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
122 if (!(sb->s_flags & MS_RDONLY)) {
123 nilfs_set_error(sbi);
125 if (nilfs_test_opt(sbi, ERRORS_RO)) {
126 printk(KERN_CRIT "Remounting filesystem read-only\n");
127 sb->s_flags |= MS_RDONLY;
131 if (nilfs_test_opt(sbi, ERRORS_PANIC))
132 panic("NILFS (device %s): panic forced after error\n",
136 void nilfs_warning(struct super_block *sb, const char *function,
137 const char *fmt, ...)
142 printk(KERN_WARNING "NILFS warning (device %s): %s: ",
150 struct inode *nilfs_alloc_inode(struct super_block *sb)
152 struct nilfs_inode_info *ii;
154 ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
160 ii->vfs_inode.i_version = 1;
161 nilfs_btnode_cache_init(&ii->i_btnode_cache, sb->s_bdi);
162 return &ii->vfs_inode;
165 void nilfs_destroy_inode(struct inode *inode)
167 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
170 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
173 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
176 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int flag)
178 struct the_nilfs *nilfs = sbi->s_nilfs;
182 set_buffer_dirty(nilfs->ns_sbh[0]);
183 if (nilfs_test_opt(sbi, BARRIER)) {
184 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
185 WRITE_SYNC | WRITE_FLUSH_FUA);
187 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
192 "NILFS: unable to write superblock (err=%d)\n", err);
193 if (err == -EIO && nilfs->ns_sbh[1]) {
195 * sbp[0] points to newer log than sbp[1],
196 * so copy sbp[0] to sbp[1] to take over sbp[0].
198 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
200 nilfs_fall_back_super_block(nilfs);
204 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
206 nilfs->ns_sbwcount++;
209 * The latest segment becomes trailable from the position
210 * written in superblock.
212 clear_nilfs_discontinued(nilfs);
214 /* update GC protection for recent segments */
215 if (nilfs->ns_sbh[1]) {
216 if (flag == NILFS_SB_COMMIT_ALL) {
217 set_buffer_dirty(nilfs->ns_sbh[1]);
218 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
221 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
222 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
223 sbp = nilfs->ns_sbp[1];
226 spin_lock(&nilfs->ns_last_segment_lock);
227 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
228 spin_unlock(&nilfs->ns_last_segment_lock);
234 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
235 struct the_nilfs *nilfs)
237 sector_t nfreeblocks;
239 /* nilfs->ns_sem must be locked by the caller. */
240 nilfs_count_free_blocks(nilfs, &nfreeblocks);
241 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
243 spin_lock(&nilfs->ns_last_segment_lock);
244 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
245 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
246 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
247 spin_unlock(&nilfs->ns_last_segment_lock);
250 struct nilfs_super_block **nilfs_prepare_super(struct nilfs_sb_info *sbi,
253 struct the_nilfs *nilfs = sbi->s_nilfs;
254 struct nilfs_super_block **sbp = nilfs->ns_sbp;
256 /* nilfs->ns_sem must be locked by the caller. */
257 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
259 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
260 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
262 printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
267 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
268 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
272 nilfs_swap_super_block(nilfs);
277 int nilfs_commit_super(struct nilfs_sb_info *sbi, int flag)
279 struct the_nilfs *nilfs = sbi->s_nilfs;
280 struct nilfs_super_block **sbp = nilfs->ns_sbp;
283 /* nilfs->ns_sem must be locked by the caller. */
285 nilfs->ns_sbwtime = t;
286 sbp[0]->s_wtime = cpu_to_le64(t);
288 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
289 (unsigned char *)sbp[0],
291 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
292 sbp[1]->s_wtime = sbp[0]->s_wtime;
294 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
295 (unsigned char *)sbp[1],
298 clear_nilfs_sb_dirty(nilfs);
299 return nilfs_sync_super(sbi, flag);
303 * nilfs_cleanup_super() - write filesystem state for cleanup
304 * @sbi: nilfs_sb_info to be unmounted or degraded to read-only
306 * This function restores state flags in the on-disk super block.
307 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
308 * filesystem was not clean previously.
310 int nilfs_cleanup_super(struct nilfs_sb_info *sbi)
312 struct nilfs_super_block **sbp;
313 int flag = NILFS_SB_COMMIT;
316 sbp = nilfs_prepare_super(sbi, 0);
318 sbp[0]->s_state = cpu_to_le16(sbi->s_nilfs->ns_mount_state);
319 nilfs_set_log_cursor(sbp[0], sbi->s_nilfs);
320 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
322 * make the "clean" flag also to the opposite
323 * super block if both super blocks point to
324 * the same checkpoint.
326 sbp[1]->s_state = sbp[0]->s_state;
327 flag = NILFS_SB_COMMIT_ALL;
329 ret = nilfs_commit_super(sbi, flag);
334 static void nilfs_put_super(struct super_block *sb)
336 struct nilfs_sb_info *sbi = NILFS_SB(sb);
337 struct the_nilfs *nilfs = sbi->s_nilfs;
339 nilfs_detach_segment_constructor(sbi);
341 if (!(sb->s_flags & MS_RDONLY)) {
342 down_write(&nilfs->ns_sem);
343 nilfs_cleanup_super(sbi);
344 up_write(&nilfs->ns_sem);
347 iput(nilfs->ns_sufile);
348 iput(nilfs->ns_cpfile);
351 destroy_nilfs(nilfs);
353 sb->s_fs_info = NULL;
357 static int nilfs_sync_fs(struct super_block *sb, int wait)
359 struct nilfs_sb_info *sbi = NILFS_SB(sb);
360 struct the_nilfs *nilfs = sbi->s_nilfs;
361 struct nilfs_super_block **sbp;
364 /* This function is called when super block should be written back */
366 err = nilfs_construct_segment(sb);
368 down_write(&nilfs->ns_sem);
369 if (nilfs_sb_dirty(nilfs)) {
370 sbp = nilfs_prepare_super(sbi, nilfs_sb_will_flip(nilfs));
372 nilfs_set_log_cursor(sbp[0], nilfs);
373 nilfs_commit_super(sbi, NILFS_SB_COMMIT);
376 up_write(&nilfs->ns_sem);
381 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno, int curr_mnt,
382 struct nilfs_root **rootp)
384 struct the_nilfs *nilfs = sbi->s_nilfs;
385 struct nilfs_root *root;
386 struct nilfs_checkpoint *raw_cp;
387 struct buffer_head *bh_cp;
390 root = nilfs_find_or_create_root(
391 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
396 goto reuse; /* already attached checkpoint */
398 down_read(&nilfs->ns_segctor_sem);
399 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
401 up_read(&nilfs->ns_segctor_sem);
403 if (err == -ENOENT || err == -EINVAL) {
405 "NILFS: Invalid checkpoint "
406 "(checkpoint number=%llu)\n",
407 (unsigned long long)cno);
413 err = nilfs_ifile_read(sbi->s_super, root, nilfs->ns_inode_size,
414 &raw_cp->cp_ifile_inode, &root->ifile);
418 atomic_set(&root->inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
419 atomic_set(&root->blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
421 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
428 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
430 nilfs_put_root(root);
435 static int nilfs_freeze(struct super_block *sb)
437 struct nilfs_sb_info *sbi = NILFS_SB(sb);
438 struct the_nilfs *nilfs = sbi->s_nilfs;
441 if (sb->s_flags & MS_RDONLY)
444 /* Mark super block clean */
445 down_write(&nilfs->ns_sem);
446 err = nilfs_cleanup_super(sbi);
447 up_write(&nilfs->ns_sem);
451 static int nilfs_unfreeze(struct super_block *sb)
453 struct nilfs_sb_info *sbi = NILFS_SB(sb);
454 struct the_nilfs *nilfs = sbi->s_nilfs;
456 if (sb->s_flags & MS_RDONLY)
459 down_write(&nilfs->ns_sem);
460 nilfs_setup_super(sbi, false);
461 up_write(&nilfs->ns_sem);
465 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
467 struct super_block *sb = dentry->d_sb;
468 struct nilfs_root *root = NILFS_I(dentry->d_inode)->i_root;
469 struct the_nilfs *nilfs = root->nilfs;
470 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
471 unsigned long long blocks;
472 unsigned long overhead;
473 unsigned long nrsvblocks;
474 sector_t nfreeblocks;
478 * Compute all of the segment blocks
480 * The blocks before first segment and after last segment
483 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
484 - nilfs->ns_first_data_block;
485 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
488 * Compute the overhead
490 * When distributing meta data blocks outside segment structure,
491 * We must count them as the overhead.
495 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
499 buf->f_type = NILFS_SUPER_MAGIC;
500 buf->f_bsize = sb->s_blocksize;
501 buf->f_blocks = blocks - overhead;
502 buf->f_bfree = nfreeblocks;
503 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
504 (buf->f_bfree - nrsvblocks) : 0;
505 buf->f_files = atomic_read(&root->inodes_count);
506 buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
507 buf->f_namelen = NILFS_NAME_LEN;
508 buf->f_fsid.val[0] = (u32)id;
509 buf->f_fsid.val[1] = (u32)(id >> 32);
514 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
516 struct super_block *sb = vfs->mnt_sb;
517 struct nilfs_sb_info *sbi = NILFS_SB(sb);
518 struct nilfs_root *root = NILFS_I(vfs->mnt_root->d_inode)->i_root;
520 if (!nilfs_test_opt(sbi, BARRIER))
521 seq_puts(seq, ",nobarrier");
522 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
523 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
524 if (nilfs_test_opt(sbi, ERRORS_PANIC))
525 seq_puts(seq, ",errors=panic");
526 if (nilfs_test_opt(sbi, ERRORS_CONT))
527 seq_puts(seq, ",errors=continue");
528 if (nilfs_test_opt(sbi, STRICT_ORDER))
529 seq_puts(seq, ",order=strict");
530 if (nilfs_test_opt(sbi, NORECOVERY))
531 seq_puts(seq, ",norecovery");
532 if (nilfs_test_opt(sbi, DISCARD))
533 seq_puts(seq, ",discard");
538 static const struct super_operations nilfs_sops = {
539 .alloc_inode = nilfs_alloc_inode,
540 .destroy_inode = nilfs_destroy_inode,
541 .dirty_inode = nilfs_dirty_inode,
542 /* .write_inode = nilfs_write_inode, */
543 /* .put_inode = nilfs_put_inode, */
544 /* .drop_inode = nilfs_drop_inode, */
545 .evict_inode = nilfs_evict_inode,
546 .put_super = nilfs_put_super,
547 /* .write_super = nilfs_write_super, */
548 .sync_fs = nilfs_sync_fs,
549 .freeze_fs = nilfs_freeze,
550 .unfreeze_fs = nilfs_unfreeze,
551 /* .write_super_lockfs */
553 .statfs = nilfs_statfs,
554 .remount_fs = nilfs_remount,
556 .show_options = nilfs_show_options
560 Opt_err_cont, Opt_err_panic, Opt_err_ro,
561 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
562 Opt_discard, Opt_nodiscard, Opt_err,
565 static match_table_t tokens = {
566 {Opt_err_cont, "errors=continue"},
567 {Opt_err_panic, "errors=panic"},
568 {Opt_err_ro, "errors=remount-ro"},
569 {Opt_barrier, "barrier"},
570 {Opt_nobarrier, "nobarrier"},
571 {Opt_snapshot, "cp=%u"},
572 {Opt_order, "order=%s"},
573 {Opt_norecovery, "norecovery"},
574 {Opt_discard, "discard"},
575 {Opt_nodiscard, "nodiscard"},
579 static int parse_options(char *options, struct super_block *sb, int is_remount)
581 struct nilfs_sb_info *sbi = NILFS_SB(sb);
583 substring_t args[MAX_OPT_ARGS];
588 while ((p = strsep(&options, ",")) != NULL) {
593 token = match_token(p, tokens, args);
596 nilfs_set_opt(sbi, BARRIER);
599 nilfs_clear_opt(sbi, BARRIER);
602 if (strcmp(args[0].from, "relaxed") == 0)
603 /* Ordered data semantics */
604 nilfs_clear_opt(sbi, STRICT_ORDER);
605 else if (strcmp(args[0].from, "strict") == 0)
606 /* Strict in-order semantics */
607 nilfs_set_opt(sbi, STRICT_ORDER);
612 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
615 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
618 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
623 "NILFS: \"%s\" option is invalid "
624 "for remount.\n", p);
629 nilfs_set_opt(sbi, NORECOVERY);
632 nilfs_set_opt(sbi, DISCARD);
635 nilfs_clear_opt(sbi, DISCARD);
639 "NILFS: Unrecognized mount option \"%s\"\n", p);
647 nilfs_set_default_options(struct nilfs_sb_info *sbi,
648 struct nilfs_super_block *sbp)
651 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
654 static int nilfs_setup_super(struct nilfs_sb_info *sbi, int is_mount)
656 struct the_nilfs *nilfs = sbi->s_nilfs;
657 struct nilfs_super_block **sbp;
661 /* nilfs->ns_sem must be locked by the caller. */
662 sbp = nilfs_prepare_super(sbi, 0);
667 goto skip_mount_setup;
669 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
670 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
672 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
674 "NILFS warning: mounting fs with errors\n");
676 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
678 "NILFS warning: maximal mount count reached\n");
682 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
684 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
685 sbp[0]->s_mtime = cpu_to_le64(get_seconds());
689 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
690 /* synchronize sbp[1] with sbp[0] */
691 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
692 return nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
695 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
696 u64 pos, int blocksize,
697 struct buffer_head **pbh)
699 unsigned long long sb_index = pos;
700 unsigned long offset;
702 offset = do_div(sb_index, blocksize);
703 *pbh = sb_bread(sb, sb_index);
706 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
709 int nilfs_store_magic_and_option(struct super_block *sb,
710 struct nilfs_super_block *sbp,
713 struct nilfs_sb_info *sbi = NILFS_SB(sb);
715 sb->s_magic = le16_to_cpu(sbp->s_magic);
717 /* FS independent flags */
718 #ifdef NILFS_ATIME_DISABLE
719 sb->s_flags |= MS_NOATIME;
722 nilfs_set_default_options(sbi, sbp);
724 sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
725 sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
726 sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
727 sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
729 return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
732 int nilfs_check_feature_compatibility(struct super_block *sb,
733 struct nilfs_super_block *sbp)
737 features = le64_to_cpu(sbp->s_feature_incompat) &
738 ~NILFS_FEATURE_INCOMPAT_SUPP;
740 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
741 "optional features (%llx)\n",
742 (unsigned long long)features);
745 features = le64_to_cpu(sbp->s_feature_compat_ro) &
746 ~NILFS_FEATURE_COMPAT_RO_SUPP;
747 if (!(sb->s_flags & MS_RDONLY) && features) {
748 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
749 "unsupported optional features (%llx)\n",
750 (unsigned long long)features);
756 static int nilfs_get_root_dentry(struct super_block *sb,
757 struct nilfs_root *root,
758 struct dentry **root_dentry)
761 struct dentry *dentry;
764 inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
766 printk(KERN_ERR "NILFS: get root inode failed\n");
767 ret = PTR_ERR(inode);
770 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
772 printk(KERN_ERR "NILFS: corrupt root inode.\n");
777 if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
778 dentry = d_find_alias(inode);
780 dentry = d_alloc_root(inode);
790 dentry = d_obtain_alias(inode);
791 if (IS_ERR(dentry)) {
792 ret = PTR_ERR(dentry);
796 *root_dentry = dentry;
801 printk(KERN_ERR "NILFS: get root dentry failed\n");
805 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
806 struct dentry **root_dentry)
808 struct the_nilfs *nilfs = NILFS_SB(s)->s_nilfs;
809 struct nilfs_root *root;
812 down_read(&nilfs->ns_segctor_sem);
813 ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
814 up_read(&nilfs->ns_segctor_sem);
816 ret = (ret == -ENOENT) ? -EINVAL : ret;
819 printk(KERN_ERR "NILFS: The specified checkpoint is "
820 "not a snapshot (checkpoint number=%llu).\n",
821 (unsigned long long)cno);
826 ret = nilfs_attach_checkpoint(NILFS_SB(s), cno, false, &root);
828 printk(KERN_ERR "NILFS: error loading snapshot "
829 "(checkpoint number=%llu).\n",
830 (unsigned long long)cno);
833 ret = nilfs_get_root_dentry(s, root, root_dentry);
834 nilfs_put_root(root);
839 static int nilfs_tree_was_touched(struct dentry *root_dentry)
841 return root_dentry->d_count > 1;
845 * nilfs_try_to_shrink_tree() - try to shrink dentries of a checkpoint
846 * @root_dentry: root dentry of the tree to be shrunk
848 * This function returns true if the tree was in-use.
850 static int nilfs_try_to_shrink_tree(struct dentry *root_dentry)
852 if (have_submounts(root_dentry))
854 shrink_dcache_parent(root_dentry);
855 return nilfs_tree_was_touched(root_dentry);
858 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
860 struct the_nilfs *nilfs = NILFS_SB(sb)->s_nilfs;
861 struct nilfs_root *root;
863 struct dentry *dentry;
866 if (cno < 0 || cno > nilfs->ns_cno)
869 if (cno >= nilfs_last_cno(nilfs))
870 return true; /* protect recent checkpoints */
873 root = nilfs_lookup_root(NILFS_SB(sb)->s_nilfs, cno);
875 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
877 dentry = d_find_alias(inode);
879 if (nilfs_tree_was_touched(dentry))
880 ret = nilfs_try_to_shrink_tree(dentry);
885 nilfs_put_root(root);
891 * nilfs_fill_super() - initialize a super block instance
893 * @data: mount options
894 * @silent: silent mode flag
896 * This function is called exclusively by nilfs->ns_mount_mutex.
897 * So, the recovery process is protected from other simultaneous mounts.
900 nilfs_fill_super(struct super_block *sb, void *data, int silent)
902 struct the_nilfs *nilfs;
903 struct nilfs_sb_info *sbi;
904 struct nilfs_root *fsroot;
905 struct backing_dev_info *bdi;
909 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
916 nilfs = alloc_nilfs(sb->s_bdev);
921 sbi->s_nilfs = nilfs;
923 err = init_nilfs(nilfs, sbi, (char *)data);
927 spin_lock_init(&sbi->s_inode_lock);
928 INIT_LIST_HEAD(&sbi->s_dirty_files);
931 * Following initialization is overlapped because
932 * nilfs_sb_info structure has been cleared at the beginning.
933 * But we reserve them to keep our interest and make ready
934 * for the future change.
936 get_random_bytes(&sbi->s_next_generation,
937 sizeof(sbi->s_next_generation));
938 spin_lock_init(&sbi->s_next_gen_lock);
940 sb->s_op = &nilfs_sops;
941 sb->s_export_op = &nilfs_export_ops;
945 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
946 sb->s_bdi = bdi ? : &default_backing_dev_info;
948 err = load_nilfs(nilfs, sbi);
952 cno = nilfs_last_cno(nilfs);
953 err = nilfs_attach_checkpoint(sbi, cno, true, &fsroot);
955 printk(KERN_ERR "NILFS: error loading last checkpoint "
956 "(checkpoint number=%llu).\n", (unsigned long long)cno);
960 if (!(sb->s_flags & MS_RDONLY)) {
961 err = nilfs_attach_segment_constructor(sbi, fsroot);
963 goto failed_checkpoint;
966 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
970 nilfs_put_root(fsroot);
972 if (!(sb->s_flags & MS_RDONLY)) {
973 down_write(&nilfs->ns_sem);
974 nilfs_setup_super(sbi, true);
975 up_write(&nilfs->ns_sem);
981 nilfs_detach_segment_constructor(sbi);
984 nilfs_put_root(fsroot);
987 iput(nilfs->ns_sufile);
988 iput(nilfs->ns_cpfile);
992 destroy_nilfs(nilfs);
995 sb->s_fs_info = NULL;
1000 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1002 struct nilfs_sb_info *sbi = NILFS_SB(sb);
1003 struct the_nilfs *nilfs = sbi->s_nilfs;
1004 unsigned long old_sb_flags;
1005 struct nilfs_mount_options old_opts;
1008 old_sb_flags = sb->s_flags;
1009 old_opts.mount_opt = sbi->s_mount_opt;
1011 if (!parse_options(data, sb, 1)) {
1015 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
1019 if (!nilfs_valid_fs(nilfs)) {
1020 printk(KERN_WARNING "NILFS (device %s): couldn't "
1021 "remount because the filesystem is in an "
1022 "incomplete recovery state.\n", sb->s_id);
1026 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1028 if (*flags & MS_RDONLY) {
1029 /* Shutting down the segment constructor */
1030 nilfs_detach_segment_constructor(sbi);
1031 sb->s_flags |= MS_RDONLY;
1034 * Remounting a valid RW partition RDONLY, so set
1035 * the RDONLY flag and then mark the partition as valid again.
1037 down_write(&nilfs->ns_sem);
1038 nilfs_cleanup_super(sbi);
1039 up_write(&nilfs->ns_sem);
1042 struct nilfs_root *root;
1045 * Mounting a RDONLY partition read-write, so reread and
1046 * store the current valid flag. (It may have been changed
1047 * by fsck since we originally mounted the partition.)
1049 down_read(&nilfs->ns_sem);
1050 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1051 ~NILFS_FEATURE_COMPAT_RO_SUPP;
1052 up_read(&nilfs->ns_sem);
1054 printk(KERN_WARNING "NILFS (device %s): couldn't "
1055 "remount RDWR because of unsupported optional "
1056 "features (%llx)\n",
1057 sb->s_id, (unsigned long long)features);
1062 sb->s_flags &= ~MS_RDONLY;
1064 root = NILFS_I(sb->s_root->d_inode)->i_root;
1065 err = nilfs_attach_segment_constructor(sbi, root);
1069 down_write(&nilfs->ns_sem);
1070 nilfs_setup_super(sbi, true);
1071 up_write(&nilfs->ns_sem);
1077 sb->s_flags = old_sb_flags;
1078 sbi->s_mount_opt = old_opts.mount_opt;
1082 struct nilfs_super_data {
1083 struct block_device *bdev;
1084 struct nilfs_sb_info *sbi;
1090 * nilfs_identify - pre-read mount options needed to identify mount instance
1091 * @data: mount options
1092 * @sd: nilfs_super_data
1094 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1096 char *p, *options = data;
1097 substring_t args[MAX_OPT_ARGS];
1102 p = strsep(&options, ",");
1103 if (p != NULL && *p) {
1104 token = match_token(p, tokens, args);
1105 if (token == Opt_snapshot) {
1106 if (!(sd->flags & MS_RDONLY)) {
1109 sd->cno = simple_strtoull(args[0].from,
1112 * No need to see the end pointer;
1113 * match_token() has done syntax
1122 "NILFS: invalid mount option: %s\n", p);
1126 BUG_ON(options == data);
1127 *(options - 1) = ',';
1132 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1135 s->s_dev = s->s_bdev->bd_dev;
1139 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1141 return (void *)s->s_bdev == data;
1144 static struct dentry *
1145 nilfs_mount(struct file_system_type *fs_type, int flags,
1146 const char *dev_name, void *data)
1148 struct nilfs_super_data sd;
1149 struct super_block *s;
1150 fmode_t mode = FMODE_READ;
1151 struct dentry *root_dentry;
1152 int err, s_new = false;
1154 if (!(flags & MS_RDONLY))
1155 mode |= FMODE_WRITE;
1157 sd.bdev = open_bdev_exclusive(dev_name, mode, fs_type);
1158 if (IS_ERR(sd.bdev))
1159 return ERR_CAST(sd.bdev);
1163 if (nilfs_identify((char *)data, &sd)) {
1169 * once the super is inserted into the list by sget, s_umount
1170 * will protect the lockfs code from trying to start a snapshot
1171 * while we are mounting
1173 mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1174 if (sd.bdev->bd_fsfreeze_count > 0) {
1175 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1179 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, sd.bdev);
1180 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1187 char b[BDEVNAME_SIZE];
1191 /* New superblock instance created */
1194 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1195 sb_set_blocksize(s, block_size(sd.bdev));
1197 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1201 s->s_flags |= MS_ACTIVE;
1202 } else if (!sd.cno) {
1205 if (nilfs_tree_was_touched(s->s_root)) {
1206 busy = nilfs_try_to_shrink_tree(s->s_root);
1207 if (busy && (flags ^ s->s_flags) & MS_RDONLY) {
1208 printk(KERN_ERR "NILFS: the device already "
1209 "has a %s mount.\n",
1210 (s->s_flags & MS_RDONLY) ?
1211 "read-only" : "read/write");
1218 * Try remount to setup mount states if the current
1219 * tree is not mounted and only snapshots use this sb.
1221 err = nilfs_remount(s, &flags, data);
1228 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1232 root_dentry = dget(s->s_root);
1236 close_bdev_exclusive(sd.bdev, mode);
1241 deactivate_locked_super(s);
1245 close_bdev_exclusive(sd.bdev, mode);
1246 return ERR_PTR(err);
1249 struct file_system_type nilfs_fs_type = {
1250 .owner = THIS_MODULE,
1252 .mount = nilfs_mount,
1253 .kill_sb = kill_block_super,
1254 .fs_flags = FS_REQUIRES_DEV,
1257 static void nilfs_inode_init_once(void *obj)
1259 struct nilfs_inode_info *ii = obj;
1261 INIT_LIST_HEAD(&ii->i_dirty);
1262 #ifdef CONFIG_NILFS_XATTR
1263 init_rwsem(&ii->xattr_sem);
1265 nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
1266 ii->i_bmap = &ii->i_bmap_data;
1267 inode_init_once(&ii->vfs_inode);
1270 static void nilfs_segbuf_init_once(void *obj)
1272 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1275 static void nilfs_destroy_cachep(void)
1277 if (nilfs_inode_cachep)
1278 kmem_cache_destroy(nilfs_inode_cachep);
1279 if (nilfs_transaction_cachep)
1280 kmem_cache_destroy(nilfs_transaction_cachep);
1281 if (nilfs_segbuf_cachep)
1282 kmem_cache_destroy(nilfs_segbuf_cachep);
1283 if (nilfs_btree_path_cache)
1284 kmem_cache_destroy(nilfs_btree_path_cache);
1287 static int __init nilfs_init_cachep(void)
1289 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1290 sizeof(struct nilfs_inode_info), 0,
1291 SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1292 if (!nilfs_inode_cachep)
1295 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1296 sizeof(struct nilfs_transaction_info), 0,
1297 SLAB_RECLAIM_ACCOUNT, NULL);
1298 if (!nilfs_transaction_cachep)
1301 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1302 sizeof(struct nilfs_segment_buffer), 0,
1303 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1304 if (!nilfs_segbuf_cachep)
1307 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1308 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1310 if (!nilfs_btree_path_cache)
1316 nilfs_destroy_cachep();
1320 static int __init init_nilfs_fs(void)
1324 err = nilfs_init_cachep();
1328 err = register_filesystem(&nilfs_fs_type);
1332 printk(KERN_INFO "NILFS version 2 loaded\n");
1336 nilfs_destroy_cachep();
1341 static void __exit exit_nilfs_fs(void)
1343 nilfs_destroy_cachep();
1344 unregister_filesystem(&nilfs_fs_type);
1347 module_init(init_nilfs_fs)
1348 module_exit(exit_nilfs_fs)