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>
67 MODULE_AUTHOR("NTT Corp.");
68 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
70 MODULE_LICENSE("GPL");
72 struct kmem_cache *nilfs_inode_cachep;
73 struct kmem_cache *nilfs_transaction_cachep;
74 struct kmem_cache *nilfs_segbuf_cachep;
75 struct kmem_cache *nilfs_btree_path_cache;
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_common(struct the_nilfs *nilfs)
152 struct nilfs_inode_info *ii;
154 ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
159 ii->vfs_inode.i_version = 1;
160 nilfs_btnode_cache_init(&ii->i_btnode_cache, nilfs->ns_bdi);
161 return &ii->vfs_inode;
164 struct inode *nilfs_alloc_inode(struct super_block *sb)
166 return nilfs_alloc_inode_common(NILFS_SB(sb)->s_nilfs);
169 void nilfs_destroy_inode(struct inode *inode)
171 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
174 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int flag)
176 struct the_nilfs *nilfs = sbi->s_nilfs;
178 int barrier_done = 0;
180 if (nilfs_test_opt(sbi, BARRIER)) {
181 set_buffer_ordered(nilfs->ns_sbh[0]);
185 set_buffer_dirty(nilfs->ns_sbh[0]);
186 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
187 if (err == -EOPNOTSUPP && barrier_done) {
188 nilfs_warning(sbi->s_super, __func__,
189 "barrier-based sync failed. "
190 "disabling barriers\n");
191 nilfs_clear_opt(sbi, BARRIER);
193 clear_buffer_ordered(nilfs->ns_sbh[0]);
198 "NILFS: unable to write superblock (err=%d)\n", err);
199 if (err == -EIO && nilfs->ns_sbh[1]) {
201 * sbp[0] points to newer log than sbp[1],
202 * so copy sbp[0] to sbp[1] to take over sbp[0].
204 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
206 nilfs_fall_back_super_block(nilfs);
210 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
212 nilfs->ns_sbwcount++;
215 * The latest segment becomes trailable from the position
216 * written in superblock.
218 clear_nilfs_discontinued(nilfs);
220 /* update GC protection for recent segments */
221 if (nilfs->ns_sbh[1]) {
222 if (flag == NILFS_SB_COMMIT_ALL) {
223 set_buffer_dirty(nilfs->ns_sbh[1]);
224 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
227 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
228 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
229 sbp = nilfs->ns_sbp[1];
232 spin_lock(&nilfs->ns_last_segment_lock);
233 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
234 spin_unlock(&nilfs->ns_last_segment_lock);
240 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
241 struct the_nilfs *nilfs)
243 sector_t nfreeblocks;
245 /* nilfs->ns_sem must be locked by the caller. */
246 nilfs_count_free_blocks(nilfs, &nfreeblocks);
247 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
249 spin_lock(&nilfs->ns_last_segment_lock);
250 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
251 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
252 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
253 spin_unlock(&nilfs->ns_last_segment_lock);
256 struct nilfs_super_block **nilfs_prepare_super(struct nilfs_sb_info *sbi,
259 struct the_nilfs *nilfs = sbi->s_nilfs;
260 struct nilfs_super_block **sbp = nilfs->ns_sbp;
262 /* nilfs->ns_sem must be locked by the caller. */
263 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
265 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
266 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
268 printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
273 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
274 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
278 nilfs_swap_super_block(nilfs);
283 int nilfs_commit_super(struct nilfs_sb_info *sbi, int flag)
285 struct the_nilfs *nilfs = sbi->s_nilfs;
286 struct nilfs_super_block **sbp = nilfs->ns_sbp;
289 /* nilfs->ns_sem must be locked by the caller. */
291 nilfs->ns_sbwtime = t;
292 sbp[0]->s_wtime = cpu_to_le64(t);
294 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
295 (unsigned char *)sbp[0],
297 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
298 sbp[1]->s_wtime = sbp[0]->s_wtime;
300 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
301 (unsigned char *)sbp[1],
304 clear_nilfs_sb_dirty(nilfs);
305 return nilfs_sync_super(sbi, flag);
309 * nilfs_cleanup_super() - write filesystem state for cleanup
310 * @sbi: nilfs_sb_info to be unmounted or degraded to read-only
312 * This function restores state flags in the on-disk super block.
313 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
314 * filesystem was not clean previously.
316 int nilfs_cleanup_super(struct nilfs_sb_info *sbi)
318 struct nilfs_super_block **sbp;
319 int flag = NILFS_SB_COMMIT;
322 sbp = nilfs_prepare_super(sbi, 0);
324 sbp[0]->s_state = cpu_to_le16(sbi->s_nilfs->ns_mount_state);
325 nilfs_set_log_cursor(sbp[0], sbi->s_nilfs);
326 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
328 * make the "clean" flag also to the opposite
329 * super block if both super blocks point to
330 * the same checkpoint.
332 sbp[1]->s_state = sbp[0]->s_state;
333 flag = NILFS_SB_COMMIT_ALL;
335 ret = nilfs_commit_super(sbi, flag);
340 static void nilfs_put_super(struct super_block *sb)
342 struct nilfs_sb_info *sbi = NILFS_SB(sb);
343 struct the_nilfs *nilfs = sbi->s_nilfs;
347 nilfs_detach_segment_constructor(sbi);
349 if (!(sb->s_flags & MS_RDONLY)) {
350 down_write(&nilfs->ns_sem);
351 nilfs_cleanup_super(sbi);
352 up_write(&nilfs->ns_sem);
354 down_write(&nilfs->ns_super_sem);
355 if (nilfs->ns_current == sbi)
356 nilfs->ns_current = NULL;
357 up_write(&nilfs->ns_super_sem);
359 nilfs_detach_checkpoint(sbi);
360 put_nilfs(sbi->s_nilfs);
362 sb->s_fs_info = NULL;
363 nilfs_put_sbinfo(sbi);
368 static int nilfs_sync_fs(struct super_block *sb, int wait)
370 struct nilfs_sb_info *sbi = NILFS_SB(sb);
371 struct the_nilfs *nilfs = sbi->s_nilfs;
372 struct nilfs_super_block **sbp;
375 /* This function is called when super block should be written back */
377 err = nilfs_construct_segment(sb);
379 down_write(&nilfs->ns_sem);
380 if (nilfs_sb_dirty(nilfs)) {
381 sbp = nilfs_prepare_super(sbi, nilfs_sb_will_flip(nilfs));
383 nilfs_set_log_cursor(sbp[0], nilfs);
384 nilfs_commit_super(sbi, NILFS_SB_COMMIT);
387 up_write(&nilfs->ns_sem);
392 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
394 struct the_nilfs *nilfs = sbi->s_nilfs;
395 struct nilfs_checkpoint *raw_cp;
396 struct buffer_head *bh_cp;
399 down_write(&nilfs->ns_super_sem);
400 list_add(&sbi->s_list, &nilfs->ns_supers);
401 up_write(&nilfs->ns_super_sem);
403 sbi->s_ifile = nilfs_ifile_new(sbi, nilfs->ns_inode_size);
407 down_read(&nilfs->ns_segctor_sem);
408 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
410 up_read(&nilfs->ns_segctor_sem);
412 if (err == -ENOENT || err == -EINVAL) {
414 "NILFS: Invalid checkpoint "
415 "(checkpoint number=%llu)\n",
416 (unsigned long long)cno);
421 err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
424 atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
425 atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
427 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
431 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
433 nilfs_mdt_destroy(sbi->s_ifile);
436 down_write(&nilfs->ns_super_sem);
437 list_del_init(&sbi->s_list);
438 up_write(&nilfs->ns_super_sem);
443 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
445 struct the_nilfs *nilfs = sbi->s_nilfs;
447 nilfs_mdt_destroy(sbi->s_ifile);
449 down_write(&nilfs->ns_super_sem);
450 list_del_init(&sbi->s_list);
451 up_write(&nilfs->ns_super_sem);
454 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
456 struct super_block *sb = dentry->d_sb;
457 struct nilfs_sb_info *sbi = NILFS_SB(sb);
458 struct the_nilfs *nilfs = sbi->s_nilfs;
459 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
460 unsigned long long blocks;
461 unsigned long overhead;
462 unsigned long nrsvblocks;
463 sector_t nfreeblocks;
467 * Compute all of the segment blocks
469 * The blocks before first segment and after last segment
472 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
473 - nilfs->ns_first_data_block;
474 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
477 * Compute the overhead
479 * When distributing meta data blocks outside segment structure,
480 * We must count them as the overhead.
484 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
488 buf->f_type = NILFS_SUPER_MAGIC;
489 buf->f_bsize = sb->s_blocksize;
490 buf->f_blocks = blocks - overhead;
491 buf->f_bfree = nfreeblocks;
492 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
493 (buf->f_bfree - nrsvblocks) : 0;
494 buf->f_files = atomic_read(&sbi->s_inodes_count);
495 buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
496 buf->f_namelen = NILFS_NAME_LEN;
497 buf->f_fsid.val[0] = (u32)id;
498 buf->f_fsid.val[1] = (u32)(id >> 32);
503 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
505 struct super_block *sb = vfs->mnt_sb;
506 struct nilfs_sb_info *sbi = NILFS_SB(sb);
508 if (!nilfs_test_opt(sbi, BARRIER))
509 seq_puts(seq, ",nobarrier");
510 if (nilfs_test_opt(sbi, SNAPSHOT))
511 seq_printf(seq, ",cp=%llu",
512 (unsigned long long int)sbi->s_snapshot_cno);
513 if (nilfs_test_opt(sbi, ERRORS_PANIC))
514 seq_puts(seq, ",errors=panic");
515 if (nilfs_test_opt(sbi, ERRORS_CONT))
516 seq_puts(seq, ",errors=continue");
517 if (nilfs_test_opt(sbi, STRICT_ORDER))
518 seq_puts(seq, ",order=strict");
519 if (nilfs_test_opt(sbi, NORECOVERY))
520 seq_puts(seq, ",norecovery");
521 if (nilfs_test_opt(sbi, DISCARD))
522 seq_puts(seq, ",discard");
527 static const struct super_operations nilfs_sops = {
528 .alloc_inode = nilfs_alloc_inode,
529 .destroy_inode = nilfs_destroy_inode,
530 .dirty_inode = nilfs_dirty_inode,
531 /* .write_inode = nilfs_write_inode, */
532 /* .put_inode = nilfs_put_inode, */
533 /* .drop_inode = nilfs_drop_inode, */
534 .evict_inode = nilfs_evict_inode,
535 .put_super = nilfs_put_super,
536 /* .write_super = nilfs_write_super, */
537 .sync_fs = nilfs_sync_fs,
538 /* .write_super_lockfs */
540 .statfs = nilfs_statfs,
541 .remount_fs = nilfs_remount,
543 .show_options = nilfs_show_options
546 static struct inode *
547 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
551 if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
552 ino != NILFS_SKETCH_INO)
553 return ERR_PTR(-ESTALE);
555 inode = nilfs_iget(sb, ino);
557 return ERR_CAST(inode);
558 if (generation && inode->i_generation != generation) {
560 return ERR_PTR(-ESTALE);
566 static struct dentry *
567 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
570 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
571 nilfs_nfs_get_inode);
574 static struct dentry *
575 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
578 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
579 nilfs_nfs_get_inode);
582 static const struct export_operations nilfs_export_ops = {
583 .fh_to_dentry = nilfs_fh_to_dentry,
584 .fh_to_parent = nilfs_fh_to_parent,
585 .get_parent = nilfs_get_parent,
589 Opt_err_cont, Opt_err_panic, Opt_err_ro,
590 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
591 Opt_discard, Opt_nodiscard, Opt_err,
594 static match_table_t tokens = {
595 {Opt_err_cont, "errors=continue"},
596 {Opt_err_panic, "errors=panic"},
597 {Opt_err_ro, "errors=remount-ro"},
598 {Opt_barrier, "barrier"},
599 {Opt_nobarrier, "nobarrier"},
600 {Opt_snapshot, "cp=%u"},
601 {Opt_order, "order=%s"},
602 {Opt_norecovery, "norecovery"},
603 {Opt_discard, "discard"},
604 {Opt_nodiscard, "nodiscard"},
608 static int parse_options(char *options, struct super_block *sb, int is_remount)
610 struct nilfs_sb_info *sbi = NILFS_SB(sb);
612 substring_t args[MAX_OPT_ARGS];
618 while ((p = strsep(&options, ",")) != NULL) {
623 token = match_token(p, tokens, args);
626 nilfs_set_opt(sbi, BARRIER);
629 nilfs_clear_opt(sbi, BARRIER);
632 if (strcmp(args[0].from, "relaxed") == 0)
633 /* Ordered data semantics */
634 nilfs_clear_opt(sbi, STRICT_ORDER);
635 else if (strcmp(args[0].from, "strict") == 0)
636 /* Strict in-order semantics */
637 nilfs_set_opt(sbi, STRICT_ORDER);
642 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
645 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
648 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
651 if (match_int(&args[0], &option) || option <= 0)
654 if (!nilfs_test_opt(sbi, SNAPSHOT)) {
656 "NILFS: cannot change regular "
657 "mount to snapshot.\n");
659 } else if (option != sbi->s_snapshot_cno) {
661 "NILFS: cannot remount to a "
662 "different snapshot.\n");
667 if (!(sb->s_flags & MS_RDONLY)) {
668 printk(KERN_ERR "NILFS: cannot mount snapshot "
669 "read/write. A read-only option is "
673 sbi->s_snapshot_cno = option;
674 nilfs_set_opt(sbi, SNAPSHOT);
677 nilfs_set_opt(sbi, NORECOVERY);
680 nilfs_set_opt(sbi, DISCARD);
683 nilfs_clear_opt(sbi, DISCARD);
687 "NILFS: Unrecognized mount option \"%s\"\n", p);
695 nilfs_set_default_options(struct nilfs_sb_info *sbi,
696 struct nilfs_super_block *sbp)
699 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
702 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
704 struct the_nilfs *nilfs = sbi->s_nilfs;
705 struct nilfs_super_block **sbp;
709 /* nilfs->ns_sem must be locked by the caller. */
710 sbp = nilfs_prepare_super(sbi, 0);
714 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
715 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
717 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
719 "NILFS warning: mounting fs with errors\n");
721 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
723 "NILFS warning: maximal mount count reached\n");
727 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
729 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
731 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
732 sbp[0]->s_mtime = cpu_to_le64(get_seconds());
733 /* synchronize sbp[1] with sbp[0] */
734 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
735 return nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
738 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
739 u64 pos, int blocksize,
740 struct buffer_head **pbh)
742 unsigned long long sb_index = pos;
743 unsigned long offset;
745 offset = do_div(sb_index, blocksize);
746 *pbh = sb_bread(sb, sb_index);
749 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
752 int nilfs_store_magic_and_option(struct super_block *sb,
753 struct nilfs_super_block *sbp,
756 struct nilfs_sb_info *sbi = NILFS_SB(sb);
758 sb->s_magic = le16_to_cpu(sbp->s_magic);
760 /* FS independent flags */
761 #ifdef NILFS_ATIME_DISABLE
762 sb->s_flags |= MS_NOATIME;
765 nilfs_set_default_options(sbi, sbp);
767 sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
768 sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
769 sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
770 sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
772 return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
775 int nilfs_check_feature_compatibility(struct super_block *sb,
776 struct nilfs_super_block *sbp)
780 features = le64_to_cpu(sbp->s_feature_incompat) &
781 ~NILFS_FEATURE_INCOMPAT_SUPP;
783 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
784 "optional features (%llx)\n",
785 (unsigned long long)features);
788 features = le64_to_cpu(sbp->s_feature_compat_ro) &
789 ~NILFS_FEATURE_COMPAT_RO_SUPP;
790 if (!(sb->s_flags & MS_RDONLY) && features) {
791 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
792 "unsupported optional features (%llx)\n",
793 (unsigned long long)features);
800 * nilfs_fill_super() - initialize a super block instance
802 * @data: mount options
803 * @silent: silent mode flag
804 * @nilfs: the_nilfs struct
806 * This function is called exclusively by nilfs->ns_mount_mutex.
807 * So, the recovery process is protected from other simultaneous mounts.
810 nilfs_fill_super(struct super_block *sb, void *data, int silent,
811 struct the_nilfs *nilfs)
813 struct nilfs_sb_info *sbi;
818 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
825 sbi->s_nilfs = nilfs;
827 atomic_set(&sbi->s_count, 1);
829 err = init_nilfs(nilfs, sbi, (char *)data);
833 spin_lock_init(&sbi->s_inode_lock);
834 INIT_LIST_HEAD(&sbi->s_dirty_files);
835 INIT_LIST_HEAD(&sbi->s_list);
838 * Following initialization is overlapped because
839 * nilfs_sb_info structure has been cleared at the beginning.
840 * But we reserve them to keep our interest and make ready
841 * for the future change.
843 get_random_bytes(&sbi->s_next_generation,
844 sizeof(sbi->s_next_generation));
845 spin_lock_init(&sbi->s_next_gen_lock);
847 sb->s_op = &nilfs_sops;
848 sb->s_export_op = &nilfs_export_ops;
851 sb->s_bdi = nilfs->ns_bdi;
853 err = load_nilfs(nilfs, sbi);
857 cno = nilfs_last_cno(nilfs);
859 if (sb->s_flags & MS_RDONLY) {
860 if (nilfs_test_opt(sbi, SNAPSHOT)) {
861 down_read(&nilfs->ns_segctor_sem);
862 err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
863 sbi->s_snapshot_cno);
864 up_read(&nilfs->ns_segctor_sem);
872 "NILFS: The specified checkpoint is "
874 "(checkpoint number=%llu).\n",
875 (unsigned long long)sbi->s_snapshot_cno);
879 cno = sbi->s_snapshot_cno;
883 err = nilfs_attach_checkpoint(sbi, cno);
885 printk(KERN_ERR "NILFS: error loading a checkpoint"
886 " (checkpoint number=%llu).\n", (unsigned long long)cno);
890 if (!(sb->s_flags & MS_RDONLY)) {
891 err = nilfs_attach_segment_constructor(sbi);
893 goto failed_checkpoint;
896 root = nilfs_iget(sb, NILFS_ROOT_INO);
898 printk(KERN_ERR "NILFS: get root inode failed\n");
902 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
904 printk(KERN_ERR "NILFS: corrupt root inode.\n");
908 sb->s_root = d_alloc_root(root);
911 printk(KERN_ERR "NILFS: get root dentry failed\n");
916 if (!(sb->s_flags & MS_RDONLY)) {
917 down_write(&nilfs->ns_sem);
918 nilfs_setup_super(sbi);
919 up_write(&nilfs->ns_sem);
922 down_write(&nilfs->ns_super_sem);
923 if (!nilfs_test_opt(sbi, SNAPSHOT))
924 nilfs->ns_current = sbi;
925 up_write(&nilfs->ns_super_sem);
930 nilfs_detach_segment_constructor(sbi);
933 nilfs_detach_checkpoint(sbi);
937 sb->s_fs_info = NULL;
938 nilfs_put_sbinfo(sbi);
942 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
944 struct nilfs_sb_info *sbi = NILFS_SB(sb);
945 struct the_nilfs *nilfs = sbi->s_nilfs;
946 unsigned long old_sb_flags;
947 struct nilfs_mount_options old_opts;
948 int was_snapshot, err;
952 down_write(&nilfs->ns_super_sem);
953 old_sb_flags = sb->s_flags;
954 old_opts.mount_opt = sbi->s_mount_opt;
955 old_opts.snapshot_cno = sbi->s_snapshot_cno;
956 was_snapshot = nilfs_test_opt(sbi, SNAPSHOT);
958 if (!parse_options(data, sb, 1)) {
962 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
965 if (was_snapshot && !(*flags & MS_RDONLY)) {
966 printk(KERN_ERR "NILFS (device %s): cannot remount snapshot "
967 "read/write.\n", sb->s_id);
971 if (!nilfs_valid_fs(nilfs)) {
972 printk(KERN_WARNING "NILFS (device %s): couldn't "
973 "remount because the filesystem is in an "
974 "incomplete recovery state.\n", sb->s_id);
978 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
980 if (*flags & MS_RDONLY) {
981 /* Shutting down the segment constructor */
982 nilfs_detach_segment_constructor(sbi);
983 sb->s_flags |= MS_RDONLY;
986 * Remounting a valid RW partition RDONLY, so set
987 * the RDONLY flag and then mark the partition as valid again.
989 down_write(&nilfs->ns_sem);
990 nilfs_cleanup_super(sbi);
991 up_write(&nilfs->ns_sem);
996 * Mounting a RDONLY partition read-write, so reread and
997 * store the current valid flag. (It may have been changed
998 * by fsck since we originally mounted the partition.)
1000 down_read(&nilfs->ns_sem);
1001 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1002 ~NILFS_FEATURE_COMPAT_RO_SUPP;
1003 up_read(&nilfs->ns_sem);
1005 printk(KERN_WARNING "NILFS (device %s): couldn't "
1006 "remount RDWR because of unsupported optional "
1007 "features (%llx)\n",
1008 sb->s_id, (unsigned long long)features);
1013 sb->s_flags &= ~MS_RDONLY;
1015 err = nilfs_attach_segment_constructor(sbi);
1019 down_write(&nilfs->ns_sem);
1020 nilfs_setup_super(sbi);
1021 up_write(&nilfs->ns_sem);
1024 up_write(&nilfs->ns_super_sem);
1029 sb->s_flags = old_sb_flags;
1030 sbi->s_mount_opt = old_opts.mount_opt;
1031 sbi->s_snapshot_cno = old_opts.snapshot_cno;
1032 up_write(&nilfs->ns_super_sem);
1037 struct nilfs_super_data {
1038 struct block_device *bdev;
1039 struct nilfs_sb_info *sbi;
1045 * nilfs_identify - pre-read mount options needed to identify mount instance
1046 * @data: mount options
1047 * @sd: nilfs_super_data
1049 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1051 char *p, *options = data;
1052 substring_t args[MAX_OPT_ARGS];
1057 p = strsep(&options, ",");
1058 if (p != NULL && *p) {
1059 token = match_token(p, tokens, args);
1060 if (token == Opt_snapshot) {
1061 if (!(sd->flags & MS_RDONLY))
1064 ret = match_int(&args[0], &option);
1075 "NILFS: invalid mount option: %s\n", p);
1079 BUG_ON(options == data);
1080 *(options - 1) = ',';
1085 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1087 struct nilfs_super_data *sd = data;
1089 s->s_bdev = sd->bdev;
1090 s->s_dev = s->s_bdev->bd_dev;
1094 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1096 struct nilfs_super_data *sd = data;
1098 return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1102 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1103 const char *dev_name, void *data, struct vfsmount *mnt)
1105 struct nilfs_super_data sd;
1106 struct super_block *s;
1107 fmode_t mode = FMODE_READ;
1108 struct the_nilfs *nilfs;
1109 int err, need_to_close = 1;
1111 if (!(flags & MS_RDONLY))
1112 mode |= FMODE_WRITE;
1114 sd.bdev = open_bdev_exclusive(dev_name, mode, fs_type);
1115 if (IS_ERR(sd.bdev))
1116 return PTR_ERR(sd.bdev);
1119 * To get mount instance using sget() vfs-routine, NILFS needs
1120 * much more information than normal filesystems to identify mount
1121 * instance. For snapshot mounts, not only a mount type (ro-mount
1122 * or rw-mount) but also a checkpoint number is required.
1126 if (nilfs_identify((char *)data, &sd)) {
1131 nilfs = find_or_create_nilfs(sd.bdev);
1137 mutex_lock(&nilfs->ns_mount_mutex);
1141 * Check if an exclusive mount exists or not.
1142 * Snapshot mounts coexist with a current mount
1143 * (i.e. rw-mount or ro-mount), whereas rw-mount and
1144 * ro-mount are mutually exclusive.
1146 down_read(&nilfs->ns_super_sem);
1147 if (nilfs->ns_current &&
1148 ((nilfs->ns_current->s_super->s_flags ^ flags)
1150 up_read(&nilfs->ns_super_sem);
1154 up_read(&nilfs->ns_super_sem);
1158 * Find existing nilfs_sb_info struct
1160 sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1163 * Get super block instance holding the nilfs_sb_info struct.
1164 * A new instance is allocated if no existing mount is present or
1165 * existing instance has been unmounted.
1167 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1169 nilfs_put_sbinfo(sd.sbi);
1177 char b[BDEVNAME_SIZE];
1179 /* New superblock instance created */
1182 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1183 sb_set_blocksize(s, block_size(sd.bdev));
1185 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0,
1190 s->s_flags |= MS_ACTIVE;
1194 mutex_unlock(&nilfs->ns_mount_mutex);
1197 close_bdev_exclusive(sd.bdev, mode);
1198 simple_set_mnt(mnt, s);
1202 mutex_unlock(&nilfs->ns_mount_mutex);
1205 close_bdev_exclusive(sd.bdev, mode);
1210 /* Abandoning the newly allocated superblock */
1211 mutex_unlock(&nilfs->ns_mount_mutex);
1213 deactivate_locked_super(s);
1215 * deactivate_locked_super() invokes close_bdev_exclusive().
1216 * We must finish all post-cleaning before this call;
1217 * put_nilfs() needs the block device.
1222 struct file_system_type nilfs_fs_type = {
1223 .owner = THIS_MODULE,
1225 .get_sb = nilfs_get_sb,
1226 .kill_sb = kill_block_super,
1227 .fs_flags = FS_REQUIRES_DEV,
1230 static void nilfs_inode_init_once(void *obj)
1232 struct nilfs_inode_info *ii = obj;
1234 INIT_LIST_HEAD(&ii->i_dirty);
1235 #ifdef CONFIG_NILFS_XATTR
1236 init_rwsem(&ii->xattr_sem);
1238 nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
1239 ii->i_bmap = &ii->i_bmap_data;
1240 inode_init_once(&ii->vfs_inode);
1243 static void nilfs_segbuf_init_once(void *obj)
1245 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1248 static void nilfs_destroy_cachep(void)
1250 if (nilfs_inode_cachep)
1251 kmem_cache_destroy(nilfs_inode_cachep);
1252 if (nilfs_transaction_cachep)
1253 kmem_cache_destroy(nilfs_transaction_cachep);
1254 if (nilfs_segbuf_cachep)
1255 kmem_cache_destroy(nilfs_segbuf_cachep);
1256 if (nilfs_btree_path_cache)
1257 kmem_cache_destroy(nilfs_btree_path_cache);
1260 static int __init nilfs_init_cachep(void)
1262 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1263 sizeof(struct nilfs_inode_info), 0,
1264 SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1265 if (!nilfs_inode_cachep)
1268 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1269 sizeof(struct nilfs_transaction_info), 0,
1270 SLAB_RECLAIM_ACCOUNT, NULL);
1271 if (!nilfs_transaction_cachep)
1274 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1275 sizeof(struct nilfs_segment_buffer), 0,
1276 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1277 if (!nilfs_segbuf_cachep)
1280 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1281 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1283 if (!nilfs_btree_path_cache)
1289 nilfs_destroy_cachep();
1293 static int __init init_nilfs_fs(void)
1297 err = nilfs_init_cachep();
1301 err = register_filesystem(&nilfs_fs_type);
1305 printk(KERN_INFO "NILFS version 2 loaded\n");
1309 nilfs_destroy_cachep();
1314 static void __exit exit_nilfs_fs(void)
1316 nilfs_destroy_cachep();
1317 unregister_filesystem(&nilfs_fs_type);
1320 module_init(init_nilfs_fs)
1321 module_exit(exit_nilfs_fs)