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 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_remount(struct super_block *sb, int *flags, char *data);
78 static void nilfs_set_error(struct nilfs_sb_info *sbi)
80 struct the_nilfs *nilfs = sbi->s_nilfs;
81 struct nilfs_super_block **sbp;
83 down_write(&nilfs->ns_sem);
84 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
85 nilfs->ns_mount_state |= NILFS_ERROR_FS;
86 sbp = nilfs_prepare_super(sbi, 0);
88 sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
90 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
91 nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
94 up_write(&nilfs->ns_sem);
98 * nilfs_error() - report failure condition on a filesystem
100 * nilfs_error() sets an ERROR_FS flag on the superblock as well as
101 * reporting an error message. It should be called when NILFS detects
102 * incoherences or defects of meta data on disk. As for sustainable
103 * errors such as a single-shot I/O error, nilfs_warning() or the printk()
104 * function should be used instead.
106 * The segment constructor must not call this function because it can
109 void nilfs_error(struct super_block *sb, const char *function,
110 const char *fmt, ...)
112 struct nilfs_sb_info *sbi = NILFS_SB(sb);
116 printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
121 if (!(sb->s_flags & MS_RDONLY)) {
122 nilfs_set_error(sbi);
124 if (nilfs_test_opt(sbi, ERRORS_RO)) {
125 printk(KERN_CRIT "Remounting filesystem read-only\n");
126 sb->s_flags |= MS_RDONLY;
130 if (nilfs_test_opt(sbi, ERRORS_PANIC))
131 panic("NILFS (device %s): panic forced after error\n",
135 void nilfs_warning(struct super_block *sb, const char *function,
136 const char *fmt, ...)
141 printk(KERN_WARNING "NILFS warning (device %s): %s: ",
149 struct inode *nilfs_alloc_inode_common(struct the_nilfs *nilfs)
151 struct nilfs_inode_info *ii;
153 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 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
174 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
177 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
180 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int flag)
182 struct the_nilfs *nilfs = sbi->s_nilfs;
186 set_buffer_dirty(nilfs->ns_sbh[0]);
188 if (nilfs_test_opt(sbi, BARRIER)) {
189 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
190 WRITE_SYNC | WRITE_BARRIER);
191 if (err == -EOPNOTSUPP) {
192 nilfs_warning(sbi->s_super, __func__,
193 "barrier-based sync failed. "
194 "disabling barriers\n");
195 nilfs_clear_opt(sbi, BARRIER);
199 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
204 "NILFS: unable to write superblock (err=%d)\n", err);
205 if (err == -EIO && nilfs->ns_sbh[1]) {
207 * sbp[0] points to newer log than sbp[1],
208 * so copy sbp[0] to sbp[1] to take over sbp[0].
210 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
212 nilfs_fall_back_super_block(nilfs);
216 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
218 nilfs->ns_sbwcount++;
221 * The latest segment becomes trailable from the position
222 * written in superblock.
224 clear_nilfs_discontinued(nilfs);
226 /* update GC protection for recent segments */
227 if (nilfs->ns_sbh[1]) {
228 if (flag == NILFS_SB_COMMIT_ALL) {
229 set_buffer_dirty(nilfs->ns_sbh[1]);
230 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
233 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
234 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
235 sbp = nilfs->ns_sbp[1];
238 spin_lock(&nilfs->ns_last_segment_lock);
239 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
240 spin_unlock(&nilfs->ns_last_segment_lock);
246 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
247 struct the_nilfs *nilfs)
249 sector_t nfreeblocks;
251 /* nilfs->ns_sem must be locked by the caller. */
252 nilfs_count_free_blocks(nilfs, &nfreeblocks);
253 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
255 spin_lock(&nilfs->ns_last_segment_lock);
256 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
257 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
258 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
259 spin_unlock(&nilfs->ns_last_segment_lock);
262 struct nilfs_super_block **nilfs_prepare_super(struct nilfs_sb_info *sbi,
265 struct the_nilfs *nilfs = sbi->s_nilfs;
266 struct nilfs_super_block **sbp = nilfs->ns_sbp;
268 /* nilfs->ns_sem must be locked by the caller. */
269 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
271 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
272 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
274 printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
279 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
280 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
284 nilfs_swap_super_block(nilfs);
289 int nilfs_commit_super(struct nilfs_sb_info *sbi, int flag)
291 struct the_nilfs *nilfs = sbi->s_nilfs;
292 struct nilfs_super_block **sbp = nilfs->ns_sbp;
295 /* nilfs->ns_sem must be locked by the caller. */
297 nilfs->ns_sbwtime = t;
298 sbp[0]->s_wtime = cpu_to_le64(t);
300 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
301 (unsigned char *)sbp[0],
303 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
304 sbp[1]->s_wtime = sbp[0]->s_wtime;
306 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
307 (unsigned char *)sbp[1],
310 clear_nilfs_sb_dirty(nilfs);
311 return nilfs_sync_super(sbi, flag);
315 * nilfs_cleanup_super() - write filesystem state for cleanup
316 * @sbi: nilfs_sb_info to be unmounted or degraded to read-only
318 * This function restores state flags in the on-disk super block.
319 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
320 * filesystem was not clean previously.
322 int nilfs_cleanup_super(struct nilfs_sb_info *sbi)
324 struct nilfs_super_block **sbp;
325 int flag = NILFS_SB_COMMIT;
328 sbp = nilfs_prepare_super(sbi, 0);
330 sbp[0]->s_state = cpu_to_le16(sbi->s_nilfs->ns_mount_state);
331 nilfs_set_log_cursor(sbp[0], sbi->s_nilfs);
332 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
334 * make the "clean" flag also to the opposite
335 * super block if both super blocks point to
336 * the same checkpoint.
338 sbp[1]->s_state = sbp[0]->s_state;
339 flag = NILFS_SB_COMMIT_ALL;
341 ret = nilfs_commit_super(sbi, flag);
346 static void nilfs_put_super(struct super_block *sb)
348 struct nilfs_sb_info *sbi = NILFS_SB(sb);
349 struct the_nilfs *nilfs = sbi->s_nilfs;
351 nilfs_detach_segment_constructor(sbi);
353 if (!(sb->s_flags & MS_RDONLY)) {
354 down_write(&nilfs->ns_sem);
355 nilfs_cleanup_super(sbi);
356 up_write(&nilfs->ns_sem);
358 down_write(&nilfs->ns_super_sem);
359 if (nilfs->ns_current == sbi)
360 nilfs->ns_current = NULL;
361 up_write(&nilfs->ns_super_sem);
363 nilfs_detach_checkpoint(sbi);
364 put_nilfs(sbi->s_nilfs);
366 sb->s_fs_info = NULL;
367 nilfs_put_sbinfo(sbi);
370 static int nilfs_sync_fs(struct super_block *sb, int wait)
372 struct nilfs_sb_info *sbi = NILFS_SB(sb);
373 struct the_nilfs *nilfs = sbi->s_nilfs;
374 struct nilfs_super_block **sbp;
377 /* This function is called when super block should be written back */
379 err = nilfs_construct_segment(sb);
381 down_write(&nilfs->ns_sem);
382 if (nilfs_sb_dirty(nilfs)) {
383 sbp = nilfs_prepare_super(sbi, nilfs_sb_will_flip(nilfs));
385 nilfs_set_log_cursor(sbp[0], nilfs);
386 nilfs_commit_super(sbi, NILFS_SB_COMMIT);
389 up_write(&nilfs->ns_sem);
394 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno, int curr_mnt,
395 struct nilfs_root **rootp)
397 struct the_nilfs *nilfs = sbi->s_nilfs;
398 struct nilfs_root *root;
399 struct nilfs_checkpoint *raw_cp;
400 struct buffer_head *bh_cp;
403 root = nilfs_find_or_create_root(
404 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
408 down_write(&nilfs->ns_super_sem);
409 list_add(&sbi->s_list, &nilfs->ns_supers);
410 up_write(&nilfs->ns_super_sem);
412 sbi->s_ifile = nilfs_ifile_new(sbi, nilfs->ns_inode_size);
416 down_read(&nilfs->ns_segctor_sem);
417 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
419 up_read(&nilfs->ns_segctor_sem);
421 if (err == -ENOENT || err == -EINVAL) {
423 "NILFS: Invalid checkpoint "
424 "(checkpoint number=%llu)\n",
425 (unsigned long long)cno);
430 err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
433 atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
434 atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
436 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
442 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
444 nilfs_mdt_destroy(sbi->s_ifile);
448 down_write(&nilfs->ns_super_sem);
449 list_del_init(&sbi->s_list);
450 up_write(&nilfs->ns_super_sem);
451 nilfs_put_root(root);
456 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
458 struct the_nilfs *nilfs = sbi->s_nilfs;
460 nilfs_mdt_destroy(sbi->s_ifile);
462 down_write(&nilfs->ns_super_sem);
463 list_del_init(&sbi->s_list);
464 up_write(&nilfs->ns_super_sem);
467 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
469 struct super_block *sb = dentry->d_sb;
470 struct nilfs_sb_info *sbi = NILFS_SB(sb);
471 struct the_nilfs *nilfs = sbi->s_nilfs;
472 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
473 unsigned long long blocks;
474 unsigned long overhead;
475 unsigned long nrsvblocks;
476 sector_t nfreeblocks;
480 * Compute all of the segment blocks
482 * The blocks before first segment and after last segment
485 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
486 - nilfs->ns_first_data_block;
487 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
490 * Compute the overhead
492 * When distributing meta data blocks outside segment structure,
493 * We must count them as the overhead.
497 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
501 buf->f_type = NILFS_SUPER_MAGIC;
502 buf->f_bsize = sb->s_blocksize;
503 buf->f_blocks = blocks - overhead;
504 buf->f_bfree = nfreeblocks;
505 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
506 (buf->f_bfree - nrsvblocks) : 0;
507 buf->f_files = atomic_read(&sbi->s_inodes_count);
508 buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
509 buf->f_namelen = NILFS_NAME_LEN;
510 buf->f_fsid.val[0] = (u32)id;
511 buf->f_fsid.val[1] = (u32)(id >> 32);
516 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
518 struct super_block *sb = vfs->mnt_sb;
519 struct nilfs_sb_info *sbi = NILFS_SB(sb);
521 if (!nilfs_test_opt(sbi, BARRIER))
522 seq_puts(seq, ",nobarrier");
523 if (nilfs_test_opt(sbi, SNAPSHOT))
524 seq_printf(seq, ",cp=%llu",
525 (unsigned long long int)sbi->s_snapshot_cno);
526 if (nilfs_test_opt(sbi, ERRORS_PANIC))
527 seq_puts(seq, ",errors=panic");
528 if (nilfs_test_opt(sbi, ERRORS_CONT))
529 seq_puts(seq, ",errors=continue");
530 if (nilfs_test_opt(sbi, STRICT_ORDER))
531 seq_puts(seq, ",order=strict");
532 if (nilfs_test_opt(sbi, NORECOVERY))
533 seq_puts(seq, ",norecovery");
534 if (nilfs_test_opt(sbi, DISCARD))
535 seq_puts(seq, ",discard");
540 static const struct super_operations nilfs_sops = {
541 .alloc_inode = nilfs_alloc_inode,
542 .destroy_inode = nilfs_destroy_inode,
543 .dirty_inode = nilfs_dirty_inode,
544 /* .write_inode = nilfs_write_inode, */
545 /* .put_inode = nilfs_put_inode, */
546 /* .drop_inode = nilfs_drop_inode, */
547 .evict_inode = nilfs_evict_inode,
548 .put_super = nilfs_put_super,
549 /* .write_super = nilfs_write_super, */
550 .sync_fs = nilfs_sync_fs,
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];
589 while ((p = strsep(&options, ",")) != NULL) {
594 token = match_token(p, tokens, args);
597 nilfs_set_opt(sbi, BARRIER);
600 nilfs_clear_opt(sbi, BARRIER);
603 if (strcmp(args[0].from, "relaxed") == 0)
604 /* Ordered data semantics */
605 nilfs_clear_opt(sbi, STRICT_ORDER);
606 else if (strcmp(args[0].from, "strict") == 0)
607 /* Strict in-order semantics */
608 nilfs_set_opt(sbi, STRICT_ORDER);
613 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
616 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
619 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
622 if (match_int(&args[0], &option) || option <= 0)
625 if (!nilfs_test_opt(sbi, SNAPSHOT)) {
627 "NILFS: cannot change regular "
628 "mount to snapshot.\n");
630 } else if (option != sbi->s_snapshot_cno) {
632 "NILFS: cannot remount to a "
633 "different snapshot.\n");
638 if (!(sb->s_flags & MS_RDONLY)) {
639 printk(KERN_ERR "NILFS: cannot mount snapshot "
640 "read/write. A read-only option is "
644 sbi->s_snapshot_cno = option;
645 nilfs_set_opt(sbi, SNAPSHOT);
648 nilfs_set_opt(sbi, NORECOVERY);
651 nilfs_set_opt(sbi, DISCARD);
654 nilfs_clear_opt(sbi, DISCARD);
658 "NILFS: Unrecognized mount option \"%s\"\n", p);
666 nilfs_set_default_options(struct nilfs_sb_info *sbi,
667 struct nilfs_super_block *sbp)
670 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
673 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
675 struct the_nilfs *nilfs = sbi->s_nilfs;
676 struct nilfs_super_block **sbp;
680 /* nilfs->ns_sem must be locked by the caller. */
681 sbp = nilfs_prepare_super(sbi, 0);
685 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
686 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
688 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
690 "NILFS warning: mounting fs with errors\n");
692 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
694 "NILFS warning: maximal mount count reached\n");
698 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
700 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
702 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
703 sbp[0]->s_mtime = cpu_to_le64(get_seconds());
704 /* synchronize sbp[1] with sbp[0] */
705 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
706 return nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
709 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
710 u64 pos, int blocksize,
711 struct buffer_head **pbh)
713 unsigned long long sb_index = pos;
714 unsigned long offset;
716 offset = do_div(sb_index, blocksize);
717 *pbh = sb_bread(sb, sb_index);
720 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
723 int nilfs_store_magic_and_option(struct super_block *sb,
724 struct nilfs_super_block *sbp,
727 struct nilfs_sb_info *sbi = NILFS_SB(sb);
729 sb->s_magic = le16_to_cpu(sbp->s_magic);
731 /* FS independent flags */
732 #ifdef NILFS_ATIME_DISABLE
733 sb->s_flags |= MS_NOATIME;
736 nilfs_set_default_options(sbi, sbp);
738 sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
739 sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
740 sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
741 sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
743 return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
746 int nilfs_check_feature_compatibility(struct super_block *sb,
747 struct nilfs_super_block *sbp)
751 features = le64_to_cpu(sbp->s_feature_incompat) &
752 ~NILFS_FEATURE_INCOMPAT_SUPP;
754 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
755 "optional features (%llx)\n",
756 (unsigned long long)features);
759 features = le64_to_cpu(sbp->s_feature_compat_ro) &
760 ~NILFS_FEATURE_COMPAT_RO_SUPP;
761 if (!(sb->s_flags & MS_RDONLY) && features) {
762 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
763 "unsupported optional features (%llx)\n",
764 (unsigned long long)features);
771 * nilfs_fill_super() - initialize a super block instance
773 * @data: mount options
774 * @silent: silent mode flag
775 * @nilfs: the_nilfs struct
777 * This function is called exclusively by nilfs->ns_mount_mutex.
778 * So, the recovery process is protected from other simultaneous mounts.
781 nilfs_fill_super(struct super_block *sb, void *data, int silent,
782 struct the_nilfs *nilfs)
784 struct nilfs_sb_info *sbi;
785 struct nilfs_root *fsroot;
790 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
797 sbi->s_nilfs = nilfs;
799 atomic_set(&sbi->s_count, 1);
801 err = init_nilfs(nilfs, sbi, (char *)data);
805 spin_lock_init(&sbi->s_inode_lock);
806 INIT_LIST_HEAD(&sbi->s_dirty_files);
807 INIT_LIST_HEAD(&sbi->s_list);
810 * Following initialization is overlapped because
811 * nilfs_sb_info structure has been cleared at the beginning.
812 * But we reserve them to keep our interest and make ready
813 * for the future change.
815 get_random_bytes(&sbi->s_next_generation,
816 sizeof(sbi->s_next_generation));
817 spin_lock_init(&sbi->s_next_gen_lock);
819 sb->s_op = &nilfs_sops;
820 sb->s_export_op = &nilfs_export_ops;
823 sb->s_bdi = nilfs->ns_bdi;
825 err = load_nilfs(nilfs, sbi);
829 cno = nilfs_last_cno(nilfs);
832 if (sb->s_flags & MS_RDONLY) {
833 if (nilfs_test_opt(sbi, SNAPSHOT)) {
834 down_read(&nilfs->ns_segctor_sem);
835 err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
836 sbi->s_snapshot_cno);
837 up_read(&nilfs->ns_segctor_sem);
845 "NILFS: The specified checkpoint is "
847 "(checkpoint number=%llu).\n",
848 (unsigned long long)sbi->s_snapshot_cno);
852 cno = sbi->s_snapshot_cno;
857 err = nilfs_attach_checkpoint(sbi, cno, curr_mnt, &fsroot);
859 printk(KERN_ERR "NILFS: error loading a checkpoint"
860 " (checkpoint number=%llu).\n", (unsigned long long)cno);
864 if (!(sb->s_flags & MS_RDONLY)) {
865 err = nilfs_attach_segment_constructor(sbi);
867 goto failed_checkpoint;
870 root = nilfs_iget(sb, fsroot, NILFS_ROOT_INO);
872 printk(KERN_ERR "NILFS: get root inode failed\n");
876 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
878 printk(KERN_ERR "NILFS: corrupt root inode.\n");
882 sb->s_root = d_alloc_root(root);
885 printk(KERN_ERR "NILFS: get root dentry failed\n");
890 nilfs_put_root(fsroot);
892 if (!(sb->s_flags & MS_RDONLY)) {
893 down_write(&nilfs->ns_sem);
894 nilfs_setup_super(sbi);
895 up_write(&nilfs->ns_sem);
898 down_write(&nilfs->ns_super_sem);
899 if (!nilfs_test_opt(sbi, SNAPSHOT))
900 nilfs->ns_current = sbi;
901 up_write(&nilfs->ns_super_sem);
906 nilfs_detach_segment_constructor(sbi);
909 nilfs_detach_checkpoint(sbi);
910 nilfs_put_root(fsroot);
914 sb->s_fs_info = NULL;
915 nilfs_put_sbinfo(sbi);
919 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
921 struct nilfs_sb_info *sbi = NILFS_SB(sb);
922 struct the_nilfs *nilfs = sbi->s_nilfs;
923 unsigned long old_sb_flags;
924 struct nilfs_mount_options old_opts;
925 int was_snapshot, err;
927 down_write(&nilfs->ns_super_sem);
928 old_sb_flags = sb->s_flags;
929 old_opts.mount_opt = sbi->s_mount_opt;
930 old_opts.snapshot_cno = sbi->s_snapshot_cno;
931 was_snapshot = nilfs_test_opt(sbi, SNAPSHOT);
933 if (!parse_options(data, sb, 1)) {
937 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
940 if (was_snapshot && !(*flags & MS_RDONLY)) {
941 printk(KERN_ERR "NILFS (device %s): cannot remount snapshot "
942 "read/write.\n", sb->s_id);
946 if (!nilfs_valid_fs(nilfs)) {
947 printk(KERN_WARNING "NILFS (device %s): couldn't "
948 "remount because the filesystem is in an "
949 "incomplete recovery state.\n", sb->s_id);
953 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
955 if (*flags & MS_RDONLY) {
956 /* Shutting down the segment constructor */
957 nilfs_detach_segment_constructor(sbi);
958 sb->s_flags |= MS_RDONLY;
961 * Remounting a valid RW partition RDONLY, so set
962 * the RDONLY flag and then mark the partition as valid again.
964 down_write(&nilfs->ns_sem);
965 nilfs_cleanup_super(sbi);
966 up_write(&nilfs->ns_sem);
971 * Mounting a RDONLY partition read-write, so reread and
972 * store the current valid flag. (It may have been changed
973 * by fsck since we originally mounted the partition.)
975 down_read(&nilfs->ns_sem);
976 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
977 ~NILFS_FEATURE_COMPAT_RO_SUPP;
978 up_read(&nilfs->ns_sem);
980 printk(KERN_WARNING "NILFS (device %s): couldn't "
981 "remount RDWR because of unsupported optional "
983 sb->s_id, (unsigned long long)features);
988 sb->s_flags &= ~MS_RDONLY;
990 err = nilfs_attach_segment_constructor(sbi);
994 down_write(&nilfs->ns_sem);
995 nilfs_setup_super(sbi);
996 up_write(&nilfs->ns_sem);
999 up_write(&nilfs->ns_super_sem);
1003 sb->s_flags = old_sb_flags;
1004 sbi->s_mount_opt = old_opts.mount_opt;
1005 sbi->s_snapshot_cno = old_opts.snapshot_cno;
1006 up_write(&nilfs->ns_super_sem);
1010 struct nilfs_super_data {
1011 struct block_device *bdev;
1012 struct nilfs_sb_info *sbi;
1018 * nilfs_identify - pre-read mount options needed to identify mount instance
1019 * @data: mount options
1020 * @sd: nilfs_super_data
1022 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1024 char *p, *options = data;
1025 substring_t args[MAX_OPT_ARGS];
1030 p = strsep(&options, ",");
1031 if (p != NULL && *p) {
1032 token = match_token(p, tokens, args);
1033 if (token == Opt_snapshot) {
1034 if (!(sd->flags & MS_RDONLY))
1037 ret = match_int(&args[0], &option);
1048 "NILFS: invalid mount option: %s\n", p);
1052 BUG_ON(options == data);
1053 *(options - 1) = ',';
1058 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1060 struct nilfs_super_data *sd = data;
1062 s->s_bdev = sd->bdev;
1063 s->s_dev = s->s_bdev->bd_dev;
1067 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1069 struct nilfs_super_data *sd = data;
1071 return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1075 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1076 const char *dev_name, void *data, struct vfsmount *mnt)
1078 struct nilfs_super_data sd;
1079 struct super_block *s;
1080 fmode_t mode = FMODE_READ;
1081 struct the_nilfs *nilfs;
1082 int err, need_to_close = 1;
1084 if (!(flags & MS_RDONLY))
1085 mode |= FMODE_WRITE;
1087 sd.bdev = open_bdev_exclusive(dev_name, mode, fs_type);
1088 if (IS_ERR(sd.bdev))
1089 return PTR_ERR(sd.bdev);
1092 * To get mount instance using sget() vfs-routine, NILFS needs
1093 * much more information than normal filesystems to identify mount
1094 * instance. For snapshot mounts, not only a mount type (ro-mount
1095 * or rw-mount) but also a checkpoint number is required.
1099 if (nilfs_identify((char *)data, &sd)) {
1104 nilfs = find_or_create_nilfs(sd.bdev);
1110 mutex_lock(&nilfs->ns_mount_mutex);
1114 * Check if an exclusive mount exists or not.
1115 * Snapshot mounts coexist with a current mount
1116 * (i.e. rw-mount or ro-mount), whereas rw-mount and
1117 * ro-mount are mutually exclusive.
1119 down_read(&nilfs->ns_super_sem);
1120 if (nilfs->ns_current &&
1121 ((nilfs->ns_current->s_super->s_flags ^ flags)
1123 up_read(&nilfs->ns_super_sem);
1127 up_read(&nilfs->ns_super_sem);
1131 * Find existing nilfs_sb_info struct
1133 sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1136 * Get super block instance holding the nilfs_sb_info struct.
1137 * A new instance is allocated if no existing mount is present or
1138 * existing instance has been unmounted.
1140 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1142 nilfs_put_sbinfo(sd.sbi);
1150 char b[BDEVNAME_SIZE];
1152 /* New superblock instance created */
1155 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1156 sb_set_blocksize(s, block_size(sd.bdev));
1158 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0,
1163 s->s_flags |= MS_ACTIVE;
1167 mutex_unlock(&nilfs->ns_mount_mutex);
1170 close_bdev_exclusive(sd.bdev, mode);
1171 simple_set_mnt(mnt, s);
1175 mutex_unlock(&nilfs->ns_mount_mutex);
1178 close_bdev_exclusive(sd.bdev, mode);
1182 /* Abandoning the newly allocated superblock */
1183 mutex_unlock(&nilfs->ns_mount_mutex);
1185 deactivate_locked_super(s);
1187 * deactivate_locked_super() invokes close_bdev_exclusive().
1188 * We must finish all post-cleaning before this call;
1189 * put_nilfs() needs the block device.
1194 struct file_system_type nilfs_fs_type = {
1195 .owner = THIS_MODULE,
1197 .get_sb = nilfs_get_sb,
1198 .kill_sb = kill_block_super,
1199 .fs_flags = FS_REQUIRES_DEV,
1202 static void nilfs_inode_init_once(void *obj)
1204 struct nilfs_inode_info *ii = obj;
1206 INIT_LIST_HEAD(&ii->i_dirty);
1207 #ifdef CONFIG_NILFS_XATTR
1208 init_rwsem(&ii->xattr_sem);
1210 nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
1211 ii->i_bmap = &ii->i_bmap_data;
1212 inode_init_once(&ii->vfs_inode);
1215 static void nilfs_segbuf_init_once(void *obj)
1217 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1220 static void nilfs_destroy_cachep(void)
1222 if (nilfs_inode_cachep)
1223 kmem_cache_destroy(nilfs_inode_cachep);
1224 if (nilfs_transaction_cachep)
1225 kmem_cache_destroy(nilfs_transaction_cachep);
1226 if (nilfs_segbuf_cachep)
1227 kmem_cache_destroy(nilfs_segbuf_cachep);
1228 if (nilfs_btree_path_cache)
1229 kmem_cache_destroy(nilfs_btree_path_cache);
1232 static int __init nilfs_init_cachep(void)
1234 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1235 sizeof(struct nilfs_inode_info), 0,
1236 SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1237 if (!nilfs_inode_cachep)
1240 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1241 sizeof(struct nilfs_transaction_info), 0,
1242 SLAB_RECLAIM_ACCOUNT, NULL);
1243 if (!nilfs_transaction_cachep)
1246 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1247 sizeof(struct nilfs_segment_buffer), 0,
1248 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1249 if (!nilfs_segbuf_cachep)
1252 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1253 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1255 if (!nilfs_btree_path_cache)
1261 nilfs_destroy_cachep();
1265 static int __init init_nilfs_fs(void)
1269 err = nilfs_init_cachep();
1273 err = register_filesystem(&nilfs_fs_type);
1277 printk(KERN_INFO "NILFS version 2 loaded\n");
1281 nilfs_destroy_cachep();
1286 static void __exit exit_nilfs_fs(void)
1288 nilfs_destroy_cachep();
1289 unregister_filesystem(&nilfs_fs_type);
1292 module_init(init_nilfs_fs)
1293 module_exit(exit_nilfs_fs)