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
23 * linux/fs/ext2/super.c
25 * Copyright (C) 1992, 1993, 1994, 1995
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
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>
59 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
65 MODULE_AUTHOR("NTT Corp.");
66 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
68 MODULE_LICENSE("GPL");
70 static struct kmem_cache *nilfs_inode_cachep;
71 struct kmem_cache *nilfs_transaction_cachep;
72 struct kmem_cache *nilfs_segbuf_cachep;
73 struct kmem_cache *nilfs_btree_path_cache;
75 static int nilfs_setup_super(struct super_block *sb, int is_mount);
76 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
78 static void nilfs_set_error(struct super_block *sb)
80 struct the_nilfs *nilfs = sb->s_fs_info;
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(sb, 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(sb, 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 the_nilfs *nilfs = sb->s_fs_info;
113 struct va_format vaf;
121 printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
122 sb->s_id, function, &vaf);
126 if (!(sb->s_flags & MS_RDONLY)) {
129 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
130 printk(KERN_CRIT "Remounting filesystem read-only\n");
131 sb->s_flags |= MS_RDONLY;
135 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
136 panic("NILFS (device %s): panic forced after error\n",
140 void nilfs_warning(struct super_block *sb, const char *function,
141 const char *fmt, ...)
143 struct va_format vaf;
151 printk(KERN_WARNING "NILFS warning (device %s): %s: %pV\n",
152 sb->s_id, function, &vaf);
158 struct inode *nilfs_alloc_inode(struct super_block *sb)
160 struct nilfs_inode_info *ii;
162 ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
168 ii->vfs_inode.i_version = 1;
169 nilfs_mapping_init(&ii->i_btnode_cache, &ii->vfs_inode);
170 return &ii->vfs_inode;
173 static void nilfs_i_callback(struct rcu_head *head)
175 struct inode *inode = container_of(head, struct inode, i_rcu);
176 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
179 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
182 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
185 void nilfs_destroy_inode(struct inode *inode)
187 call_rcu(&inode->i_rcu, nilfs_i_callback);
190 static int nilfs_sync_super(struct super_block *sb, int flag)
192 struct the_nilfs *nilfs = sb->s_fs_info;
196 set_buffer_dirty(nilfs->ns_sbh[0]);
197 if (nilfs_test_opt(nilfs, BARRIER)) {
198 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
199 WRITE_SYNC | WRITE_FLUSH_FUA);
201 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
206 "NILFS: unable to write superblock (err=%d)\n", err);
207 if (err == -EIO && nilfs->ns_sbh[1]) {
209 * sbp[0] points to newer log than sbp[1],
210 * so copy sbp[0] to sbp[1] to take over sbp[0].
212 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
214 nilfs_fall_back_super_block(nilfs);
218 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
220 nilfs->ns_sbwcount++;
223 * The latest segment becomes trailable from the position
224 * written in superblock.
226 clear_nilfs_discontinued(nilfs);
228 /* update GC protection for recent segments */
229 if (nilfs->ns_sbh[1]) {
230 if (flag == NILFS_SB_COMMIT_ALL) {
231 set_buffer_dirty(nilfs->ns_sbh[1]);
232 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
235 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
236 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
237 sbp = nilfs->ns_sbp[1];
240 spin_lock(&nilfs->ns_last_segment_lock);
241 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
242 spin_unlock(&nilfs->ns_last_segment_lock);
248 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
249 struct the_nilfs *nilfs)
251 sector_t nfreeblocks;
253 /* nilfs->ns_sem must be locked by the caller. */
254 nilfs_count_free_blocks(nilfs, &nfreeblocks);
255 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
257 spin_lock(&nilfs->ns_last_segment_lock);
258 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
259 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
260 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
261 spin_unlock(&nilfs->ns_last_segment_lock);
264 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
267 struct the_nilfs *nilfs = sb->s_fs_info;
268 struct nilfs_super_block **sbp = nilfs->ns_sbp;
270 /* nilfs->ns_sem must be locked by the caller. */
271 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
273 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
274 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
276 printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
281 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
282 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
286 nilfs_swap_super_block(nilfs);
291 int nilfs_commit_super(struct super_block *sb, int flag)
293 struct the_nilfs *nilfs = sb->s_fs_info;
294 struct nilfs_super_block **sbp = nilfs->ns_sbp;
297 /* nilfs->ns_sem must be locked by the caller. */
299 nilfs->ns_sbwtime = t;
300 sbp[0]->s_wtime = cpu_to_le64(t);
302 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
303 (unsigned char *)sbp[0],
305 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
306 sbp[1]->s_wtime = sbp[0]->s_wtime;
308 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
309 (unsigned char *)sbp[1],
312 clear_nilfs_sb_dirty(nilfs);
313 nilfs->ns_flushed_device = 1;
314 /* make sure store to ns_flushed_device cannot be reordered */
316 return nilfs_sync_super(sb, flag);
320 * nilfs_cleanup_super() - write filesystem state for cleanup
321 * @sb: super block instance to be unmounted or degraded to read-only
323 * This function restores state flags in the on-disk super block.
324 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
325 * filesystem was not clean previously.
327 int nilfs_cleanup_super(struct super_block *sb)
329 struct the_nilfs *nilfs = sb->s_fs_info;
330 struct nilfs_super_block **sbp;
331 int flag = NILFS_SB_COMMIT;
334 sbp = nilfs_prepare_super(sb, 0);
336 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
337 nilfs_set_log_cursor(sbp[0], nilfs);
338 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
340 * make the "clean" flag also to the opposite
341 * super block if both super blocks point to
342 * the same checkpoint.
344 sbp[1]->s_state = sbp[0]->s_state;
345 flag = NILFS_SB_COMMIT_ALL;
347 ret = nilfs_commit_super(sb, flag);
353 * nilfs_move_2nd_super - relocate secondary super block
354 * @sb: super block instance
355 * @sb2off: new offset of the secondary super block (in bytes)
357 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
359 struct the_nilfs *nilfs = sb->s_fs_info;
360 struct buffer_head *nsbh;
361 struct nilfs_super_block *nsbp;
362 sector_t blocknr, newblocknr;
363 unsigned long offset;
364 int sb2i; /* array index of the secondary superblock */
367 /* nilfs->ns_sem must be locked by the caller. */
368 if (nilfs->ns_sbh[1] &&
369 nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
371 blocknr = nilfs->ns_sbh[1]->b_blocknr;
372 } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
374 blocknr = nilfs->ns_sbh[0]->b_blocknr;
379 if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
380 goto out; /* super block location is unchanged */
382 /* Get new super block buffer */
383 newblocknr = sb2off >> nilfs->ns_blocksize_bits;
384 offset = sb2off & (nilfs->ns_blocksize - 1);
385 nsbh = sb_getblk(sb, newblocknr);
388 "NILFS warning: unable to move secondary superblock "
389 "to block %llu\n", (unsigned long long)newblocknr);
393 nsbp = (void *)nsbh->b_data + offset;
394 memset(nsbp, 0, nilfs->ns_blocksize);
397 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
398 brelse(nilfs->ns_sbh[sb2i]);
399 nilfs->ns_sbh[sb2i] = nsbh;
400 nilfs->ns_sbp[sb2i] = nsbp;
401 } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
402 /* secondary super block will be restored to index 1 */
403 nilfs->ns_sbh[1] = nsbh;
404 nilfs->ns_sbp[1] = nsbp;
413 * nilfs_resize_fs - resize the filesystem
414 * @sb: super block instance
415 * @newsize: new size of the filesystem (in bytes)
417 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
419 struct the_nilfs *nilfs = sb->s_fs_info;
420 struct nilfs_super_block **sbp;
421 __u64 devsize, newnsegs;
426 devsize = i_size_read(sb->s_bdev->bd_inode);
427 if (newsize > devsize)
431 * Write lock is required to protect some functions depending
432 * on the number of segments, the number of reserved segments,
435 down_write(&nilfs->ns_segctor_sem);
437 sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
438 newnsegs = sb2off >> nilfs->ns_blocksize_bits;
439 do_div(newnsegs, nilfs->ns_blocks_per_segment);
441 ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
442 up_write(&nilfs->ns_segctor_sem);
446 ret = nilfs_construct_segment(sb);
450 down_write(&nilfs->ns_sem);
451 nilfs_move_2nd_super(sb, sb2off);
453 sbp = nilfs_prepare_super(sb, 0);
455 nilfs_set_log_cursor(sbp[0], nilfs);
457 * Drop NILFS_RESIZE_FS flag for compatibility with
458 * mount-time resize which may be implemented in a
461 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
463 sbp[0]->s_dev_size = cpu_to_le64(newsize);
464 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
466 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
467 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
469 up_write(&nilfs->ns_sem);
472 * Reset the range of allocatable segments last. This order
473 * is important in the case of expansion because the secondary
474 * superblock must be protected from log write until migration
478 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
483 static void nilfs_put_super(struct super_block *sb)
485 struct the_nilfs *nilfs = sb->s_fs_info;
487 nilfs_detach_log_writer(sb);
489 if (!(sb->s_flags & MS_RDONLY)) {
490 down_write(&nilfs->ns_sem);
491 nilfs_cleanup_super(sb);
492 up_write(&nilfs->ns_sem);
495 iput(nilfs->ns_sufile);
496 iput(nilfs->ns_cpfile);
499 destroy_nilfs(nilfs);
500 sb->s_fs_info = NULL;
503 static int nilfs_sync_fs(struct super_block *sb, int wait)
505 struct the_nilfs *nilfs = sb->s_fs_info;
506 struct nilfs_super_block **sbp;
509 /* This function is called when super block should be written back */
511 err = nilfs_construct_segment(sb);
513 down_write(&nilfs->ns_sem);
514 if (nilfs_sb_dirty(nilfs)) {
515 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
517 nilfs_set_log_cursor(sbp[0], nilfs);
518 nilfs_commit_super(sb, NILFS_SB_COMMIT);
521 up_write(&nilfs->ns_sem);
524 err = nilfs_flush_device(nilfs);
529 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
530 struct nilfs_root **rootp)
532 struct the_nilfs *nilfs = sb->s_fs_info;
533 struct nilfs_root *root;
534 struct nilfs_checkpoint *raw_cp;
535 struct buffer_head *bh_cp;
538 root = nilfs_find_or_create_root(
539 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
544 goto reuse; /* already attached checkpoint */
546 down_read(&nilfs->ns_segctor_sem);
547 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
549 up_read(&nilfs->ns_segctor_sem);
551 if (err == -ENOENT || err == -EINVAL) {
553 "NILFS: Invalid checkpoint "
554 "(checkpoint number=%llu)\n",
555 (unsigned long long)cno);
561 err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
562 &raw_cp->cp_ifile_inode, &root->ifile);
566 atomic64_set(&root->inodes_count,
567 le64_to_cpu(raw_cp->cp_inodes_count));
568 atomic64_set(&root->blocks_count,
569 le64_to_cpu(raw_cp->cp_blocks_count));
571 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
578 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
580 nilfs_put_root(root);
585 static int nilfs_freeze(struct super_block *sb)
587 struct the_nilfs *nilfs = sb->s_fs_info;
590 if (sb->s_flags & MS_RDONLY)
593 /* Mark super block clean */
594 down_write(&nilfs->ns_sem);
595 err = nilfs_cleanup_super(sb);
596 up_write(&nilfs->ns_sem);
600 static int nilfs_unfreeze(struct super_block *sb)
602 struct the_nilfs *nilfs = sb->s_fs_info;
604 if (sb->s_flags & MS_RDONLY)
607 down_write(&nilfs->ns_sem);
608 nilfs_setup_super(sb, false);
609 up_write(&nilfs->ns_sem);
613 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
615 struct super_block *sb = dentry->d_sb;
616 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
617 struct the_nilfs *nilfs = root->nilfs;
618 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
619 unsigned long long blocks;
620 unsigned long overhead;
621 unsigned long nrsvblocks;
622 sector_t nfreeblocks;
623 u64 nmaxinodes, nfreeinodes;
627 * Compute all of the segment blocks
629 * The blocks before first segment and after last segment
632 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
633 - nilfs->ns_first_data_block;
634 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
637 * Compute the overhead
639 * When distributing meta data blocks outside segment structure,
640 * We must count them as the overhead.
644 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
648 err = nilfs_ifile_count_free_inodes(root->ifile,
649 &nmaxinodes, &nfreeinodes);
652 "NILFS warning: fail to count free inodes: err %d.\n",
654 if (err == -ERANGE) {
656 * If nilfs_palloc_count_max_entries() returns
657 * -ERANGE error code then we simply treat
658 * curent inodes count as maximum possible and
659 * zero as free inodes value.
661 nmaxinodes = atomic64_read(&root->inodes_count);
668 buf->f_type = NILFS_SUPER_MAGIC;
669 buf->f_bsize = sb->s_blocksize;
670 buf->f_blocks = blocks - overhead;
671 buf->f_bfree = nfreeblocks;
672 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
673 (buf->f_bfree - nrsvblocks) : 0;
674 buf->f_files = nmaxinodes;
675 buf->f_ffree = nfreeinodes;
676 buf->f_namelen = NILFS_NAME_LEN;
677 buf->f_fsid.val[0] = (u32)id;
678 buf->f_fsid.val[1] = (u32)(id >> 32);
683 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
685 struct super_block *sb = dentry->d_sb;
686 struct the_nilfs *nilfs = sb->s_fs_info;
687 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
689 if (!nilfs_test_opt(nilfs, BARRIER))
690 seq_puts(seq, ",nobarrier");
691 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
692 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
693 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
694 seq_puts(seq, ",errors=panic");
695 if (nilfs_test_opt(nilfs, ERRORS_CONT))
696 seq_puts(seq, ",errors=continue");
697 if (nilfs_test_opt(nilfs, STRICT_ORDER))
698 seq_puts(seq, ",order=strict");
699 if (nilfs_test_opt(nilfs, NORECOVERY))
700 seq_puts(seq, ",norecovery");
701 if (nilfs_test_opt(nilfs, DISCARD))
702 seq_puts(seq, ",discard");
707 static const struct super_operations nilfs_sops = {
708 .alloc_inode = nilfs_alloc_inode,
709 .destroy_inode = nilfs_destroy_inode,
710 .dirty_inode = nilfs_dirty_inode,
711 .evict_inode = nilfs_evict_inode,
712 .put_super = nilfs_put_super,
713 .sync_fs = nilfs_sync_fs,
714 .freeze_fs = nilfs_freeze,
715 .unfreeze_fs = nilfs_unfreeze,
716 .statfs = nilfs_statfs,
717 .remount_fs = nilfs_remount,
718 .show_options = nilfs_show_options
722 Opt_err_cont, Opt_err_panic, Opt_err_ro,
723 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
724 Opt_discard, Opt_nodiscard, Opt_err,
727 static match_table_t tokens = {
728 {Opt_err_cont, "errors=continue"},
729 {Opt_err_panic, "errors=panic"},
730 {Opt_err_ro, "errors=remount-ro"},
731 {Opt_barrier, "barrier"},
732 {Opt_nobarrier, "nobarrier"},
733 {Opt_snapshot, "cp=%u"},
734 {Opt_order, "order=%s"},
735 {Opt_norecovery, "norecovery"},
736 {Opt_discard, "discard"},
737 {Opt_nodiscard, "nodiscard"},
741 static int parse_options(char *options, struct super_block *sb, int is_remount)
743 struct the_nilfs *nilfs = sb->s_fs_info;
745 substring_t args[MAX_OPT_ARGS];
750 while ((p = strsep(&options, ",")) != NULL) {
755 token = match_token(p, tokens, args);
758 nilfs_set_opt(nilfs, BARRIER);
761 nilfs_clear_opt(nilfs, BARRIER);
764 if (strcmp(args[0].from, "relaxed") == 0)
765 /* Ordered data semantics */
766 nilfs_clear_opt(nilfs, STRICT_ORDER);
767 else if (strcmp(args[0].from, "strict") == 0)
768 /* Strict in-order semantics */
769 nilfs_set_opt(nilfs, STRICT_ORDER);
774 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
777 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
780 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
785 "NILFS: \"%s\" option is invalid "
786 "for remount.\n", p);
791 nilfs_set_opt(nilfs, NORECOVERY);
794 nilfs_set_opt(nilfs, DISCARD);
797 nilfs_clear_opt(nilfs, DISCARD);
801 "NILFS: Unrecognized mount option \"%s\"\n", p);
809 nilfs_set_default_options(struct super_block *sb,
810 struct nilfs_super_block *sbp)
812 struct the_nilfs *nilfs = sb->s_fs_info;
814 nilfs->ns_mount_opt =
815 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
818 static int nilfs_setup_super(struct super_block *sb, int is_mount)
820 struct the_nilfs *nilfs = sb->s_fs_info;
821 struct nilfs_super_block **sbp;
825 /* nilfs->ns_sem must be locked by the caller. */
826 sbp = nilfs_prepare_super(sb, 0);
831 goto skip_mount_setup;
833 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
834 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
836 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
838 "NILFS warning: mounting fs with errors\n");
840 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
842 "NILFS warning: maximal mount count reached\n");
846 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
848 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
849 sbp[0]->s_mtime = cpu_to_le64(get_seconds());
853 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
854 /* synchronize sbp[1] with sbp[0] */
856 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
857 return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
860 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
861 u64 pos, int blocksize,
862 struct buffer_head **pbh)
864 unsigned long long sb_index = pos;
865 unsigned long offset;
867 offset = do_div(sb_index, blocksize);
868 *pbh = sb_bread(sb, sb_index);
871 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
874 int nilfs_store_magic_and_option(struct super_block *sb,
875 struct nilfs_super_block *sbp,
878 struct the_nilfs *nilfs = sb->s_fs_info;
880 sb->s_magic = le16_to_cpu(sbp->s_magic);
882 /* FS independent flags */
883 #ifdef NILFS_ATIME_DISABLE
884 sb->s_flags |= MS_NOATIME;
887 nilfs_set_default_options(sb, sbp);
889 nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
890 nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
891 nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
892 nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
894 return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
897 int nilfs_check_feature_compatibility(struct super_block *sb,
898 struct nilfs_super_block *sbp)
902 features = le64_to_cpu(sbp->s_feature_incompat) &
903 ~NILFS_FEATURE_INCOMPAT_SUPP;
905 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
906 "optional features (%llx)\n",
907 (unsigned long long)features);
910 features = le64_to_cpu(sbp->s_feature_compat_ro) &
911 ~NILFS_FEATURE_COMPAT_RO_SUPP;
912 if (!(sb->s_flags & MS_RDONLY) && features) {
913 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
914 "unsupported optional features (%llx)\n",
915 (unsigned long long)features);
921 static int nilfs_get_root_dentry(struct super_block *sb,
922 struct nilfs_root *root,
923 struct dentry **root_dentry)
926 struct dentry *dentry;
929 inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
931 printk(KERN_ERR "NILFS: get root inode failed\n");
932 ret = PTR_ERR(inode);
935 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
937 printk(KERN_ERR "NILFS: corrupt root inode.\n");
942 if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
943 dentry = d_find_alias(inode);
945 dentry = d_make_root(inode);
954 dentry = d_obtain_root(inode);
955 if (IS_ERR(dentry)) {
956 ret = PTR_ERR(dentry);
960 *root_dentry = dentry;
965 printk(KERN_ERR "NILFS: get root dentry failed\n");
969 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
970 struct dentry **root_dentry)
972 struct the_nilfs *nilfs = s->s_fs_info;
973 struct nilfs_root *root;
976 mutex_lock(&nilfs->ns_snapshot_mount_mutex);
978 down_read(&nilfs->ns_segctor_sem);
979 ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
980 up_read(&nilfs->ns_segctor_sem);
982 ret = (ret == -ENOENT) ? -EINVAL : ret;
985 printk(KERN_ERR "NILFS: The specified checkpoint is "
986 "not a snapshot (checkpoint number=%llu).\n",
987 (unsigned long long)cno);
992 ret = nilfs_attach_checkpoint(s, cno, false, &root);
994 printk(KERN_ERR "NILFS: error loading snapshot "
995 "(checkpoint number=%llu).\n",
996 (unsigned long long)cno);
999 ret = nilfs_get_root_dentry(s, root, root_dentry);
1000 nilfs_put_root(root);
1002 mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
1007 * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
1008 * @root_dentry: root dentry of the tree to be shrunk
1010 * This function returns true if the tree was in-use.
1012 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
1014 shrink_dcache_parent(root_dentry);
1015 return d_count(root_dentry) > 1;
1018 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1020 struct the_nilfs *nilfs = sb->s_fs_info;
1021 struct nilfs_root *root;
1022 struct inode *inode;
1023 struct dentry *dentry;
1026 if (cno > nilfs->ns_cno)
1029 if (cno >= nilfs_last_cno(nilfs))
1030 return true; /* protect recent checkpoints */
1033 root = nilfs_lookup_root(nilfs, cno);
1035 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1037 dentry = d_find_alias(inode);
1039 ret = nilfs_tree_is_busy(dentry);
1044 nilfs_put_root(root);
1050 * nilfs_fill_super() - initialize a super block instance
1052 * @data: mount options
1053 * @silent: silent mode flag
1055 * This function is called exclusively by nilfs->ns_mount_mutex.
1056 * So, the recovery process is protected from other simultaneous mounts.
1059 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1061 struct the_nilfs *nilfs;
1062 struct nilfs_root *fsroot;
1066 nilfs = alloc_nilfs(sb->s_bdev);
1070 sb->s_fs_info = nilfs;
1072 err = init_nilfs(nilfs, sb, (char *)data);
1076 sb->s_op = &nilfs_sops;
1077 sb->s_export_op = &nilfs_export_ops;
1079 sb->s_time_gran = 1;
1080 sb->s_max_links = NILFS_LINK_MAX;
1082 sb->s_bdi = &bdev_get_queue(sb->s_bdev)->backing_dev_info;
1084 err = load_nilfs(nilfs, sb);
1088 cno = nilfs_last_cno(nilfs);
1089 err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1091 printk(KERN_ERR "NILFS: error loading last checkpoint "
1092 "(checkpoint number=%llu).\n", (unsigned long long)cno);
1096 if (!(sb->s_flags & MS_RDONLY)) {
1097 err = nilfs_attach_log_writer(sb, fsroot);
1099 goto failed_checkpoint;
1102 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1104 goto failed_segctor;
1106 nilfs_put_root(fsroot);
1108 if (!(sb->s_flags & MS_RDONLY)) {
1109 down_write(&nilfs->ns_sem);
1110 nilfs_setup_super(sb, true);
1111 up_write(&nilfs->ns_sem);
1117 nilfs_detach_log_writer(sb);
1120 nilfs_put_root(fsroot);
1123 iput(nilfs->ns_sufile);
1124 iput(nilfs->ns_cpfile);
1125 iput(nilfs->ns_dat);
1128 destroy_nilfs(nilfs);
1132 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1134 struct the_nilfs *nilfs = sb->s_fs_info;
1135 unsigned long old_sb_flags;
1136 unsigned long old_mount_opt;
1139 sync_filesystem(sb);
1140 old_sb_flags = sb->s_flags;
1141 old_mount_opt = nilfs->ns_mount_opt;
1143 if (!parse_options(data, sb, 1)) {
1147 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
1151 if (!nilfs_valid_fs(nilfs)) {
1152 printk(KERN_WARNING "NILFS (device %s): couldn't "
1153 "remount because the filesystem is in an "
1154 "incomplete recovery state.\n", sb->s_id);
1158 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1160 if (*flags & MS_RDONLY) {
1161 /* Shutting down log writer */
1162 nilfs_detach_log_writer(sb);
1163 sb->s_flags |= MS_RDONLY;
1166 * Remounting a valid RW partition RDONLY, so set
1167 * the RDONLY flag and then mark the partition as valid again.
1169 down_write(&nilfs->ns_sem);
1170 nilfs_cleanup_super(sb);
1171 up_write(&nilfs->ns_sem);
1174 struct nilfs_root *root;
1177 * Mounting a RDONLY partition read-write, so reread and
1178 * store the current valid flag. (It may have been changed
1179 * by fsck since we originally mounted the partition.)
1181 down_read(&nilfs->ns_sem);
1182 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1183 ~NILFS_FEATURE_COMPAT_RO_SUPP;
1184 up_read(&nilfs->ns_sem);
1186 printk(KERN_WARNING "NILFS (device %s): couldn't "
1187 "remount RDWR because of unsupported optional "
1188 "features (%llx)\n",
1189 sb->s_id, (unsigned long long)features);
1194 sb->s_flags &= ~MS_RDONLY;
1196 root = NILFS_I(d_inode(sb->s_root))->i_root;
1197 err = nilfs_attach_log_writer(sb, root);
1201 down_write(&nilfs->ns_sem);
1202 nilfs_setup_super(sb, true);
1203 up_write(&nilfs->ns_sem);
1209 sb->s_flags = old_sb_flags;
1210 nilfs->ns_mount_opt = old_mount_opt;
1214 struct nilfs_super_data {
1215 struct block_device *bdev;
1221 * nilfs_identify - pre-read mount options needed to identify mount instance
1222 * @data: mount options
1223 * @sd: nilfs_super_data
1225 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1227 char *p, *options = data;
1228 substring_t args[MAX_OPT_ARGS];
1233 p = strsep(&options, ",");
1234 if (p != NULL && *p) {
1235 token = match_token(p, tokens, args);
1236 if (token == Opt_snapshot) {
1237 if (!(sd->flags & MS_RDONLY)) {
1240 sd->cno = simple_strtoull(args[0].from,
1243 * No need to see the end pointer;
1244 * match_token() has done syntax
1253 "NILFS: invalid mount option: %s\n", p);
1257 BUG_ON(options == data);
1258 *(options - 1) = ',';
1263 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1266 s->s_dev = s->s_bdev->bd_dev;
1270 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1272 return (void *)s->s_bdev == data;
1275 static struct dentry *
1276 nilfs_mount(struct file_system_type *fs_type, int flags,
1277 const char *dev_name, void *data)
1279 struct nilfs_super_data sd;
1280 struct super_block *s;
1281 fmode_t mode = FMODE_READ | FMODE_EXCL;
1282 struct dentry *root_dentry;
1283 int err, s_new = false;
1285 if (!(flags & MS_RDONLY))
1286 mode |= FMODE_WRITE;
1288 sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1289 if (IS_ERR(sd.bdev))
1290 return ERR_CAST(sd.bdev);
1294 if (nilfs_identify((char *)data, &sd)) {
1300 * once the super is inserted into the list by sget, s_umount
1301 * will protect the lockfs code from trying to start a snapshot
1302 * while we are mounting
1304 mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1305 if (sd.bdev->bd_fsfreeze_count > 0) {
1306 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1310 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1312 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1321 /* New superblock instance created */
1323 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1324 sb_set_blocksize(s, block_size(sd.bdev));
1326 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1330 s->s_flags |= MS_ACTIVE;
1331 } else if (!sd.cno) {
1332 if (nilfs_tree_is_busy(s->s_root)) {
1333 if ((flags ^ s->s_flags) & MS_RDONLY) {
1334 printk(KERN_ERR "NILFS: the device already "
1335 "has a %s mount.\n",
1336 (s->s_flags & MS_RDONLY) ?
1337 "read-only" : "read/write");
1343 * Try remount to setup mount states if the current
1344 * tree is not mounted and only snapshots use this sb.
1346 err = nilfs_remount(s, &flags, data);
1353 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1357 root_dentry = dget(s->s_root);
1361 blkdev_put(sd.bdev, mode);
1366 deactivate_locked_super(s);
1370 blkdev_put(sd.bdev, mode);
1371 return ERR_PTR(err);
1374 struct file_system_type nilfs_fs_type = {
1375 .owner = THIS_MODULE,
1377 .mount = nilfs_mount,
1378 .kill_sb = kill_block_super,
1379 .fs_flags = FS_REQUIRES_DEV,
1381 MODULE_ALIAS_FS("nilfs2");
1383 static void nilfs_inode_init_once(void *obj)
1385 struct nilfs_inode_info *ii = obj;
1387 INIT_LIST_HEAD(&ii->i_dirty);
1388 #ifdef CONFIG_NILFS_XATTR
1389 init_rwsem(&ii->xattr_sem);
1391 address_space_init_once(&ii->i_btnode_cache);
1392 ii->i_bmap = &ii->i_bmap_data;
1393 inode_init_once(&ii->vfs_inode);
1396 static void nilfs_segbuf_init_once(void *obj)
1398 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1401 static void nilfs_destroy_cachep(void)
1404 * Make sure all delayed rcu free inodes are flushed before we
1409 kmem_cache_destroy(nilfs_inode_cachep);
1410 kmem_cache_destroy(nilfs_transaction_cachep);
1411 kmem_cache_destroy(nilfs_segbuf_cachep);
1412 kmem_cache_destroy(nilfs_btree_path_cache);
1415 static int __init nilfs_init_cachep(void)
1417 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1418 sizeof(struct nilfs_inode_info), 0,
1419 SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1420 nilfs_inode_init_once);
1421 if (!nilfs_inode_cachep)
1424 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1425 sizeof(struct nilfs_transaction_info), 0,
1426 SLAB_RECLAIM_ACCOUNT, NULL);
1427 if (!nilfs_transaction_cachep)
1430 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1431 sizeof(struct nilfs_segment_buffer), 0,
1432 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1433 if (!nilfs_segbuf_cachep)
1436 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1437 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1439 if (!nilfs_btree_path_cache)
1445 nilfs_destroy_cachep();
1449 static int __init init_nilfs_fs(void)
1453 err = nilfs_init_cachep();
1457 err = nilfs_sysfs_init();
1461 err = register_filesystem(&nilfs_fs_type);
1463 goto deinit_sysfs_entry;
1465 printk(KERN_INFO "NILFS version 2 loaded\n");
1471 nilfs_destroy_cachep();
1476 static void __exit exit_nilfs_fs(void)
1478 nilfs_destroy_cachep();
1480 unregister_filesystem(&nilfs_fs_type);
1483 module_init(init_nilfs_fs)
1484 module_exit(exit_nilfs_fs)