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 * Written by Ryusuke Konishi.
19 * linux/fs/ext2/super.c
21 * Copyright (C) 1992, 1993, 1994, 1995
22 * Remy Card (card@masi.ibp.fr)
23 * Laboratoire MASI - Institut Blaise Pascal
24 * Universite Pierre et Marie Curie (Paris VI)
28 * linux/fs/minix/inode.c
30 * Copyright (C) 1991, 1992 Linus Torvalds
32 * Big-endian to little-endian byte-swapping/bitmaps by
33 * David S. Miller (davem@caip.rutgers.edu), 1995
36 #include <linux/module.h>
37 #include <linux/string.h>
38 #include <linux/slab.h>
39 #include <linux/init.h>
40 #include <linux/blkdev.h>
41 #include <linux/parser.h>
42 #include <linux/crc32.h>
43 #include <linux/vfs.h>
44 #include <linux/writeback.h>
45 #include <linux/seq_file.h>
46 #include <linux/mount.h>
55 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
61 MODULE_AUTHOR("NTT Corp.");
62 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
64 MODULE_LICENSE("GPL");
66 static struct kmem_cache *nilfs_inode_cachep;
67 struct kmem_cache *nilfs_transaction_cachep;
68 struct kmem_cache *nilfs_segbuf_cachep;
69 struct kmem_cache *nilfs_btree_path_cache;
71 static int nilfs_setup_super(struct super_block *sb, int is_mount);
72 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
74 void __nilfs_msg(struct super_block *sb, const char *level, const char *fmt,
84 printk("%sNILFS (%s): %pV\n", level, sb->s_id, &vaf);
86 printk("%sNILFS: %pV\n", level, &vaf);
90 static void nilfs_set_error(struct super_block *sb)
92 struct the_nilfs *nilfs = sb->s_fs_info;
93 struct nilfs_super_block **sbp;
95 down_write(&nilfs->ns_sem);
96 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
97 nilfs->ns_mount_state |= NILFS_ERROR_FS;
98 sbp = nilfs_prepare_super(sb, 0);
100 sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
102 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
103 nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
106 up_write(&nilfs->ns_sem);
110 * __nilfs_error() - report failure condition on a filesystem
112 * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
113 * reporting an error message. This function should be called when
114 * NILFS detects incoherences or defects of meta data on disk.
116 * This implements the body of nilfs_error() macro. Normally,
117 * nilfs_error() should be used. As for sustainable errors such as a
118 * single-shot I/O error, nilfs_msg() should be used instead.
120 * Callers should not add a trailing newline since this will do it.
122 void __nilfs_error(struct super_block *sb, const char *function,
123 const char *fmt, ...)
125 struct the_nilfs *nilfs = sb->s_fs_info;
126 struct va_format vaf;
134 printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
135 sb->s_id, function, &vaf);
139 if (!sb_rdonly(sb)) {
142 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
143 printk(KERN_CRIT "Remounting filesystem read-only\n");
144 sb->s_flags |= MS_RDONLY;
148 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
149 panic("NILFS (device %s): panic forced after error\n",
153 struct inode *nilfs_alloc_inode(struct super_block *sb)
155 struct nilfs_inode_info *ii;
157 ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
163 ii->vfs_inode.i_version = 1;
164 ii->i_assoc_inode = NULL;
165 ii->i_bmap = &ii->i_bmap_data;
166 return &ii->vfs_inode;
169 static void nilfs_i_callback(struct rcu_head *head)
171 struct inode *inode = container_of(head, struct inode, i_rcu);
173 if (nilfs_is_metadata_file_inode(inode))
174 nilfs_mdt_destroy(inode);
176 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
179 void nilfs_destroy_inode(struct inode *inode)
181 call_rcu(&inode->i_rcu, nilfs_i_callback);
184 static int nilfs_sync_super(struct super_block *sb, int flag)
186 struct the_nilfs *nilfs = sb->s_fs_info;
190 set_buffer_dirty(nilfs->ns_sbh[0]);
191 if (nilfs_test_opt(nilfs, BARRIER)) {
192 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
193 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
195 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
199 nilfs_msg(sb, KERN_ERR, "unable to write superblock: err=%d",
201 if (err == -EIO && nilfs->ns_sbh[1]) {
203 * sbp[0] points to newer log than sbp[1],
204 * so copy sbp[0] to sbp[1] to take over sbp[0].
206 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
208 nilfs_fall_back_super_block(nilfs);
212 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
214 nilfs->ns_sbwcount++;
217 * The latest segment becomes trailable from the position
218 * written in superblock.
220 clear_nilfs_discontinued(nilfs);
222 /* update GC protection for recent segments */
223 if (nilfs->ns_sbh[1]) {
224 if (flag == NILFS_SB_COMMIT_ALL) {
225 set_buffer_dirty(nilfs->ns_sbh[1]);
226 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
229 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
230 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
231 sbp = nilfs->ns_sbp[1];
234 spin_lock(&nilfs->ns_last_segment_lock);
235 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
236 spin_unlock(&nilfs->ns_last_segment_lock);
242 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
243 struct the_nilfs *nilfs)
245 sector_t nfreeblocks;
247 /* nilfs->ns_sem must be locked by the caller. */
248 nilfs_count_free_blocks(nilfs, &nfreeblocks);
249 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
251 spin_lock(&nilfs->ns_last_segment_lock);
252 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
253 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
254 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
255 spin_unlock(&nilfs->ns_last_segment_lock);
258 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
261 struct the_nilfs *nilfs = sb->s_fs_info;
262 struct nilfs_super_block **sbp = nilfs->ns_sbp;
264 /* nilfs->ns_sem must be locked by the caller. */
265 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
267 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
268 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
270 nilfs_msg(sb, KERN_CRIT, "superblock broke");
274 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
275 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
279 nilfs_swap_super_block(nilfs);
284 int nilfs_commit_super(struct super_block *sb, int flag)
286 struct the_nilfs *nilfs = sb->s_fs_info;
287 struct nilfs_super_block **sbp = nilfs->ns_sbp;
290 /* nilfs->ns_sem must be locked by the caller. */
292 nilfs->ns_sbwtime = t;
293 sbp[0]->s_wtime = cpu_to_le64(t);
295 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
296 (unsigned char *)sbp[0],
298 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
299 sbp[1]->s_wtime = sbp[0]->s_wtime;
301 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
302 (unsigned char *)sbp[1],
305 clear_nilfs_sb_dirty(nilfs);
306 nilfs->ns_flushed_device = 1;
307 /* make sure store to ns_flushed_device cannot be reordered */
309 return nilfs_sync_super(sb, flag);
313 * nilfs_cleanup_super() - write filesystem state for cleanup
314 * @sb: super block instance to be unmounted or degraded to read-only
316 * This function restores state flags in the on-disk super block.
317 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
318 * filesystem was not clean previously.
320 int nilfs_cleanup_super(struct super_block *sb)
322 struct the_nilfs *nilfs = sb->s_fs_info;
323 struct nilfs_super_block **sbp;
324 int flag = NILFS_SB_COMMIT;
327 sbp = nilfs_prepare_super(sb, 0);
329 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
330 nilfs_set_log_cursor(sbp[0], nilfs);
331 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
333 * make the "clean" flag also to the opposite
334 * super block if both super blocks point to
335 * the same checkpoint.
337 sbp[1]->s_state = sbp[0]->s_state;
338 flag = NILFS_SB_COMMIT_ALL;
340 ret = nilfs_commit_super(sb, flag);
346 * nilfs_move_2nd_super - relocate secondary super block
347 * @sb: super block instance
348 * @sb2off: new offset of the secondary super block (in bytes)
350 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
352 struct the_nilfs *nilfs = sb->s_fs_info;
353 struct buffer_head *nsbh;
354 struct nilfs_super_block *nsbp;
355 sector_t blocknr, newblocknr;
356 unsigned long offset;
357 int sb2i; /* array index of the secondary superblock */
360 /* nilfs->ns_sem must be locked by the caller. */
361 if (nilfs->ns_sbh[1] &&
362 nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
364 blocknr = nilfs->ns_sbh[1]->b_blocknr;
365 } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
367 blocknr = nilfs->ns_sbh[0]->b_blocknr;
372 if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
373 goto out; /* super block location is unchanged */
375 /* Get new super block buffer */
376 newblocknr = sb2off >> nilfs->ns_blocksize_bits;
377 offset = sb2off & (nilfs->ns_blocksize - 1);
378 nsbh = sb_getblk(sb, newblocknr);
380 nilfs_msg(sb, KERN_WARNING,
381 "unable to move secondary superblock to block %llu",
382 (unsigned long long)newblocknr);
386 nsbp = (void *)nsbh->b_data + offset;
387 memset(nsbp, 0, nilfs->ns_blocksize);
390 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
391 brelse(nilfs->ns_sbh[sb2i]);
392 nilfs->ns_sbh[sb2i] = nsbh;
393 nilfs->ns_sbp[sb2i] = nsbp;
394 } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
395 /* secondary super block will be restored to index 1 */
396 nilfs->ns_sbh[1] = nsbh;
397 nilfs->ns_sbp[1] = nsbp;
406 * nilfs_resize_fs - resize the filesystem
407 * @sb: super block instance
408 * @newsize: new size of the filesystem (in bytes)
410 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
412 struct the_nilfs *nilfs = sb->s_fs_info;
413 struct nilfs_super_block **sbp;
414 __u64 devsize, newnsegs;
419 devsize = i_size_read(sb->s_bdev->bd_inode);
420 if (newsize > devsize)
424 * Prevent underflow in second superblock position calculation.
425 * The exact minimum size check is done in nilfs_sufile_resize().
427 if (newsize < 4096) {
433 * Write lock is required to protect some functions depending
434 * on the number of segments, the number of reserved segments,
437 down_write(&nilfs->ns_segctor_sem);
439 sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
440 newnsegs = sb2off >> nilfs->ns_blocksize_bits;
441 do_div(newnsegs, nilfs->ns_blocks_per_segment);
443 ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
444 up_write(&nilfs->ns_segctor_sem);
448 ret = nilfs_construct_segment(sb);
452 down_write(&nilfs->ns_sem);
453 nilfs_move_2nd_super(sb, sb2off);
455 sbp = nilfs_prepare_super(sb, 0);
457 nilfs_set_log_cursor(sbp[0], nilfs);
459 * Drop NILFS_RESIZE_FS flag for compatibility with
460 * mount-time resize which may be implemented in a
463 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
465 sbp[0]->s_dev_size = cpu_to_le64(newsize);
466 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
468 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
469 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
471 up_write(&nilfs->ns_sem);
474 * Reset the range of allocatable segments last. This order
475 * is important in the case of expansion because the secondary
476 * superblock must be protected from log write until migration
480 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
485 static void nilfs_put_super(struct super_block *sb)
487 struct the_nilfs *nilfs = sb->s_fs_info;
489 nilfs_detach_log_writer(sb);
491 if (!sb_rdonly(sb)) {
492 down_write(&nilfs->ns_sem);
493 nilfs_cleanup_super(sb);
494 up_write(&nilfs->ns_sem);
497 nilfs_sysfs_delete_device_group(nilfs);
498 iput(nilfs->ns_sufile);
499 iput(nilfs->ns_cpfile);
502 destroy_nilfs(nilfs);
503 sb->s_fs_info = NULL;
506 static int nilfs_sync_fs(struct super_block *sb, int wait)
508 struct the_nilfs *nilfs = sb->s_fs_info;
509 struct nilfs_super_block **sbp;
512 /* This function is called when super block should be written back */
514 err = nilfs_construct_segment(sb);
516 down_write(&nilfs->ns_sem);
517 if (nilfs_sb_dirty(nilfs)) {
518 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
520 nilfs_set_log_cursor(sbp[0], nilfs);
521 nilfs_commit_super(sb, NILFS_SB_COMMIT);
524 up_write(&nilfs->ns_sem);
527 err = nilfs_flush_device(nilfs);
532 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
533 struct nilfs_root **rootp)
535 struct the_nilfs *nilfs = sb->s_fs_info;
536 struct nilfs_root *root;
537 struct nilfs_checkpoint *raw_cp;
538 struct buffer_head *bh_cp;
541 root = nilfs_find_or_create_root(
542 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
547 goto reuse; /* already attached checkpoint */
549 down_read(&nilfs->ns_segctor_sem);
550 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
552 up_read(&nilfs->ns_segctor_sem);
554 if (err == -ENOENT || err == -EINVAL) {
555 nilfs_msg(sb, KERN_ERR,
556 "Invalid checkpoint (checkpoint number=%llu)",
557 (unsigned long long)cno);
563 err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
564 &raw_cp->cp_ifile_inode, &root->ifile);
568 atomic64_set(&root->inodes_count,
569 le64_to_cpu(raw_cp->cp_inodes_count));
570 atomic64_set(&root->blocks_count,
571 le64_to_cpu(raw_cp->cp_blocks_count));
573 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
580 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
582 nilfs_put_root(root);
587 static int nilfs_freeze(struct super_block *sb)
589 struct the_nilfs *nilfs = sb->s_fs_info;
595 /* Mark super block clean */
596 down_write(&nilfs->ns_sem);
597 err = nilfs_cleanup_super(sb);
598 up_write(&nilfs->ns_sem);
602 static int nilfs_unfreeze(struct super_block *sb)
604 struct the_nilfs *nilfs = sb->s_fs_info;
609 down_write(&nilfs->ns_sem);
610 nilfs_setup_super(sb, false);
611 up_write(&nilfs->ns_sem);
615 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
617 struct super_block *sb = dentry->d_sb;
618 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
619 struct the_nilfs *nilfs = root->nilfs;
620 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
621 unsigned long long blocks;
622 unsigned long overhead;
623 unsigned long nrsvblocks;
624 sector_t nfreeblocks;
625 u64 nmaxinodes, nfreeinodes;
629 * Compute all of the segment blocks
631 * The blocks before first segment and after last segment
634 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
635 - nilfs->ns_first_data_block;
636 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
639 * Compute the overhead
641 * When distributing meta data blocks outside segment structure,
642 * We must count them as the overhead.
646 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
650 err = nilfs_ifile_count_free_inodes(root->ifile,
651 &nmaxinodes, &nfreeinodes);
653 nilfs_msg(sb, KERN_WARNING,
654 "failed to count free inodes: err=%d", err);
655 if (err == -ERANGE) {
657 * If nilfs_palloc_count_max_entries() returns
658 * -ERANGE error code then we simply treat
659 * curent inodes count as maximum possible and
660 * zero as free inodes value.
662 nmaxinodes = atomic64_read(&root->inodes_count);
669 buf->f_type = NILFS_SUPER_MAGIC;
670 buf->f_bsize = sb->s_blocksize;
671 buf->f_blocks = blocks - overhead;
672 buf->f_bfree = nfreeblocks;
673 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
674 (buf->f_bfree - nrsvblocks) : 0;
675 buf->f_files = nmaxinodes;
676 buf->f_ffree = nfreeinodes;
677 buf->f_namelen = NILFS_NAME_LEN;
678 buf->f_fsid.val[0] = (u32)id;
679 buf->f_fsid.val[1] = (u32)(id >> 32);
684 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
686 struct super_block *sb = dentry->d_sb;
687 struct the_nilfs *nilfs = sb->s_fs_info;
688 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
690 if (!nilfs_test_opt(nilfs, BARRIER))
691 seq_puts(seq, ",nobarrier");
692 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
693 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
694 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
695 seq_puts(seq, ",errors=panic");
696 if (nilfs_test_opt(nilfs, ERRORS_CONT))
697 seq_puts(seq, ",errors=continue");
698 if (nilfs_test_opt(nilfs, STRICT_ORDER))
699 seq_puts(seq, ",order=strict");
700 if (nilfs_test_opt(nilfs, NORECOVERY))
701 seq_puts(seq, ",norecovery");
702 if (nilfs_test_opt(nilfs, DISCARD))
703 seq_puts(seq, ",discard");
708 static const struct super_operations nilfs_sops = {
709 .alloc_inode = nilfs_alloc_inode,
710 .destroy_inode = nilfs_destroy_inode,
711 .dirty_inode = nilfs_dirty_inode,
712 .evict_inode = nilfs_evict_inode,
713 .put_super = nilfs_put_super,
714 .sync_fs = nilfs_sync_fs,
715 .freeze_fs = nilfs_freeze,
716 .unfreeze_fs = nilfs_unfreeze,
717 .statfs = nilfs_statfs,
718 .remount_fs = nilfs_remount,
719 .show_options = nilfs_show_options
723 Opt_err_cont, Opt_err_panic, Opt_err_ro,
724 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
725 Opt_discard, Opt_nodiscard, Opt_err,
728 static match_table_t tokens = {
729 {Opt_err_cont, "errors=continue"},
730 {Opt_err_panic, "errors=panic"},
731 {Opt_err_ro, "errors=remount-ro"},
732 {Opt_barrier, "barrier"},
733 {Opt_nobarrier, "nobarrier"},
734 {Opt_snapshot, "cp=%u"},
735 {Opt_order, "order=%s"},
736 {Opt_norecovery, "norecovery"},
737 {Opt_discard, "discard"},
738 {Opt_nodiscard, "nodiscard"},
742 static int parse_options(char *options, struct super_block *sb, int is_remount)
744 struct the_nilfs *nilfs = sb->s_fs_info;
746 substring_t args[MAX_OPT_ARGS];
751 while ((p = strsep(&options, ",")) != NULL) {
757 token = match_token(p, tokens, args);
760 nilfs_set_opt(nilfs, BARRIER);
763 nilfs_clear_opt(nilfs, BARRIER);
766 if (strcmp(args[0].from, "relaxed") == 0)
767 /* Ordered data semantics */
768 nilfs_clear_opt(nilfs, STRICT_ORDER);
769 else if (strcmp(args[0].from, "strict") == 0)
770 /* Strict in-order semantics */
771 nilfs_set_opt(nilfs, STRICT_ORDER);
776 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
779 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
782 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
786 nilfs_msg(sb, KERN_ERR,
787 "\"%s\" option is invalid for remount",
793 nilfs_set_opt(nilfs, NORECOVERY);
796 nilfs_set_opt(nilfs, DISCARD);
799 nilfs_clear_opt(nilfs, DISCARD);
802 nilfs_msg(sb, KERN_ERR,
803 "unrecognized mount option \"%s\"", p);
811 nilfs_set_default_options(struct super_block *sb,
812 struct nilfs_super_block *sbp)
814 struct the_nilfs *nilfs = sb->s_fs_info;
816 nilfs->ns_mount_opt =
817 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
820 static int nilfs_setup_super(struct super_block *sb, int is_mount)
822 struct the_nilfs *nilfs = sb->s_fs_info;
823 struct nilfs_super_block **sbp;
827 /* nilfs->ns_sem must be locked by the caller. */
828 sbp = nilfs_prepare_super(sb, 0);
833 goto skip_mount_setup;
835 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
836 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
838 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
839 nilfs_msg(sb, KERN_WARNING, "mounting fs with errors");
841 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
842 nilfs_msg(sb, KERN_WARNING, "maximal mount count reached");
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 nilfs_msg(sb, KERN_ERR,
906 "couldn't mount because of unsupported optional features (%llx)",
907 (unsigned long long)features);
910 features = le64_to_cpu(sbp->s_feature_compat_ro) &
911 ~NILFS_FEATURE_COMPAT_RO_SUPP;
912 if (!sb_rdonly(sb) && features) {
913 nilfs_msg(sb, KERN_ERR,
914 "couldn't mount RDWR because of unsupported optional features (%llx)",
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 ret = PTR_ERR(inode);
932 nilfs_msg(sb, KERN_ERR, "error %d getting root inode", ret);
935 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
937 nilfs_msg(sb, KERN_ERR, "corrupt root inode");
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 nilfs_msg(sb, KERN_ERR, "error %d getting root dentry", ret);
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 nilfs_msg(s, KERN_ERR,
986 "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
987 (unsigned long long)cno);
992 ret = nilfs_attach_checkpoint(s, cno, false, &root);
994 nilfs_msg(s, KERN_ERR,
995 "error %d while loading snapshot (checkpoint number=%llu)",
996 ret, (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);
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 = bdi_get(sb->s_bdev->bd_bdi);
1084 err = load_nilfs(nilfs, sb);
1088 cno = nilfs_last_cno(nilfs);
1089 err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1091 nilfs_msg(sb, KERN_ERR,
1092 "error %d while loading last checkpoint (checkpoint number=%llu)",
1093 err, (unsigned long long)cno);
1097 if (!sb_rdonly(sb)) {
1098 err = nilfs_attach_log_writer(sb, fsroot);
1100 goto failed_checkpoint;
1103 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1105 goto failed_segctor;
1107 nilfs_put_root(fsroot);
1109 if (!sb_rdonly(sb)) {
1110 down_write(&nilfs->ns_sem);
1111 nilfs_setup_super(sb, true);
1112 up_write(&nilfs->ns_sem);
1118 nilfs_detach_log_writer(sb);
1121 nilfs_put_root(fsroot);
1124 nilfs_sysfs_delete_device_group(nilfs);
1125 iput(nilfs->ns_sufile);
1126 iput(nilfs->ns_cpfile);
1127 iput(nilfs->ns_dat);
1130 destroy_nilfs(nilfs);
1134 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1136 struct the_nilfs *nilfs = sb->s_fs_info;
1137 unsigned long old_sb_flags;
1138 unsigned long old_mount_opt;
1141 sync_filesystem(sb);
1142 old_sb_flags = sb->s_flags;
1143 old_mount_opt = nilfs->ns_mount_opt;
1145 if (!parse_options(data, sb, 1)) {
1149 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
1153 if (!nilfs_valid_fs(nilfs)) {
1154 nilfs_msg(sb, KERN_WARNING,
1155 "couldn't remount because the filesystem is in an incomplete recovery state");
1159 if ((bool)(*flags & MS_RDONLY) == sb_rdonly(sb))
1161 if (*flags & MS_RDONLY) {
1162 sb->s_flags |= MS_RDONLY;
1165 * Remounting a valid RW partition RDONLY, so set
1166 * the RDONLY flag and then mark the partition as valid again.
1168 down_write(&nilfs->ns_sem);
1169 nilfs_cleanup_super(sb);
1170 up_write(&nilfs->ns_sem);
1173 struct nilfs_root *root;
1176 * Mounting a RDONLY partition read-write, so reread and
1177 * store the current valid flag. (It may have been changed
1178 * by fsck since we originally mounted the partition.)
1180 down_read(&nilfs->ns_sem);
1181 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1182 ~NILFS_FEATURE_COMPAT_RO_SUPP;
1183 up_read(&nilfs->ns_sem);
1185 nilfs_msg(sb, KERN_WARNING,
1186 "couldn't remount RDWR because of unsupported optional features (%llx)",
1187 (unsigned long long)features);
1192 sb->s_flags &= ~MS_RDONLY;
1194 root = NILFS_I(d_inode(sb->s_root))->i_root;
1195 err = nilfs_attach_log_writer(sb, root);
1199 down_write(&nilfs->ns_sem);
1200 nilfs_setup_super(sb, true);
1201 up_write(&nilfs->ns_sem);
1207 sb->s_flags = old_sb_flags;
1208 nilfs->ns_mount_opt = old_mount_opt;
1212 struct nilfs_super_data {
1213 struct block_device *bdev;
1218 static int nilfs_parse_snapshot_option(const char *option,
1219 const substring_t *arg,
1220 struct nilfs_super_data *sd)
1222 unsigned long long val;
1223 const char *msg = NULL;
1226 if (!(sd->flags & MS_RDONLY)) {
1227 msg = "read-only option is not specified";
1231 err = kstrtoull(arg->from, 0, &val);
1234 msg = "too large checkpoint number";
1236 msg = "malformed argument";
1238 } else if (val == 0) {
1239 msg = "invalid checkpoint number 0";
1246 nilfs_msg(NULL, KERN_ERR, "invalid option \"%s\": %s", option, msg);
1251 * nilfs_identify - pre-read mount options needed to identify mount instance
1252 * @data: mount options
1253 * @sd: nilfs_super_data
1255 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1257 char *p, *options = data;
1258 substring_t args[MAX_OPT_ARGS];
1263 p = strsep(&options, ",");
1264 if (p != NULL && *p) {
1265 token = match_token(p, tokens, args);
1266 if (token == Opt_snapshot)
1267 ret = nilfs_parse_snapshot_option(p, &args[0],
1272 BUG_ON(options == data);
1273 *(options - 1) = ',';
1278 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1281 s->s_dev = s->s_bdev->bd_dev;
1285 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1287 return (void *)s->s_bdev == data;
1290 static struct dentry *
1291 nilfs_mount(struct file_system_type *fs_type, int flags,
1292 const char *dev_name, void *data)
1294 struct nilfs_super_data sd;
1295 struct super_block *s;
1296 fmode_t mode = FMODE_READ | FMODE_EXCL;
1297 struct dentry *root_dentry;
1298 int err, s_new = false;
1300 if (!(flags & MS_RDONLY))
1301 mode |= FMODE_WRITE;
1303 sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1304 if (IS_ERR(sd.bdev))
1305 return ERR_CAST(sd.bdev);
1309 if (nilfs_identify((char *)data, &sd)) {
1315 * once the super is inserted into the list by sget, s_umount
1316 * will protect the lockfs code from trying to start a snapshot
1317 * while we are mounting
1319 mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1320 if (sd.bdev->bd_fsfreeze_count > 0) {
1321 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1325 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1327 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1336 /* New superblock instance created */
1338 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1339 sb_set_blocksize(s, block_size(sd.bdev));
1341 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1345 s->s_flags |= MS_ACTIVE;
1346 } else if (!sd.cno) {
1347 if (nilfs_tree_is_busy(s->s_root)) {
1348 if ((flags ^ s->s_flags) & MS_RDONLY) {
1349 nilfs_msg(s, KERN_ERR,
1350 "the device already has a %s mount.",
1351 sb_rdonly(s) ? "read-only" : "read/write");
1357 * Try remount to setup mount states if the current
1358 * tree is not mounted and only snapshots use this sb.
1360 err = nilfs_remount(s, &flags, data);
1367 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1371 root_dentry = dget(s->s_root);
1375 blkdev_put(sd.bdev, mode);
1380 deactivate_locked_super(s);
1384 blkdev_put(sd.bdev, mode);
1385 return ERR_PTR(err);
1388 struct file_system_type nilfs_fs_type = {
1389 .owner = THIS_MODULE,
1391 .mount = nilfs_mount,
1392 .kill_sb = kill_block_super,
1393 .fs_flags = FS_REQUIRES_DEV,
1395 MODULE_ALIAS_FS("nilfs2");
1397 static void nilfs_inode_init_once(void *obj)
1399 struct nilfs_inode_info *ii = obj;
1401 INIT_LIST_HEAD(&ii->i_dirty);
1402 #ifdef CONFIG_NILFS_XATTR
1403 init_rwsem(&ii->xattr_sem);
1405 inode_init_once(&ii->vfs_inode);
1408 static void nilfs_segbuf_init_once(void *obj)
1410 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1413 static void nilfs_destroy_cachep(void)
1416 * Make sure all delayed rcu free inodes are flushed before we
1421 kmem_cache_destroy(nilfs_inode_cachep);
1422 kmem_cache_destroy(nilfs_transaction_cachep);
1423 kmem_cache_destroy(nilfs_segbuf_cachep);
1424 kmem_cache_destroy(nilfs_btree_path_cache);
1427 static int __init nilfs_init_cachep(void)
1429 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1430 sizeof(struct nilfs_inode_info), 0,
1431 SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1432 nilfs_inode_init_once);
1433 if (!nilfs_inode_cachep)
1436 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1437 sizeof(struct nilfs_transaction_info), 0,
1438 SLAB_RECLAIM_ACCOUNT, NULL);
1439 if (!nilfs_transaction_cachep)
1442 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1443 sizeof(struct nilfs_segment_buffer), 0,
1444 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1445 if (!nilfs_segbuf_cachep)
1448 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1449 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1451 if (!nilfs_btree_path_cache)
1457 nilfs_destroy_cachep();
1461 static int __init init_nilfs_fs(void)
1465 err = nilfs_init_cachep();
1469 err = nilfs_sysfs_init();
1473 err = register_filesystem(&nilfs_fs_type);
1475 goto deinit_sysfs_entry;
1477 printk(KERN_INFO "NILFS version 2 loaded\n");
1483 nilfs_destroy_cachep();
1488 static void __exit exit_nilfs_fs(void)
1490 nilfs_destroy_cachep();
1492 unregister_filesystem(&nilfs_fs_type);
1495 module_init(init_nilfs_fs)
1496 module_exit(exit_nilfs_fs)