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;
391 * The position of the second superblock only changes by 4KiB,
392 * which is larger than the maximum superblock data size
393 * (= 1KiB), so there is no need to use memmove() to allow
394 * overlap between source and destination.
396 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
399 * Zero fill after copy to avoid overwriting in case of move
400 * within the same block.
402 memset(nsbh->b_data, 0, offset);
403 memset((void *)nsbp + nilfs->ns_sbsize, 0,
404 nsbh->b_size - offset - nilfs->ns_sbsize);
406 memset(nsbh->b_data, 0, nsbh->b_size);
408 set_buffer_uptodate(nsbh);
412 brelse(nilfs->ns_sbh[sb2i]);
413 nilfs->ns_sbh[sb2i] = nsbh;
414 nilfs->ns_sbp[sb2i] = nsbp;
415 } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
416 /* secondary super block will be restored to index 1 */
417 nilfs->ns_sbh[1] = nsbh;
418 nilfs->ns_sbp[1] = nsbp;
427 * nilfs_resize_fs - resize the filesystem
428 * @sb: super block instance
429 * @newsize: new size of the filesystem (in bytes)
431 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
433 struct the_nilfs *nilfs = sb->s_fs_info;
434 struct nilfs_super_block **sbp;
435 __u64 devsize, newnsegs;
440 devsize = i_size_read(sb->s_bdev->bd_inode);
441 if (newsize > devsize)
445 * Prevent underflow in second superblock position calculation.
446 * The exact minimum size check is done in nilfs_sufile_resize().
448 if (newsize < 4096) {
454 * Write lock is required to protect some functions depending
455 * on the number of segments, the number of reserved segments,
458 down_write(&nilfs->ns_segctor_sem);
460 sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
461 newnsegs = sb2off >> nilfs->ns_blocksize_bits;
462 do_div(newnsegs, nilfs->ns_blocks_per_segment);
464 ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
465 up_write(&nilfs->ns_segctor_sem);
469 ret = nilfs_construct_segment(sb);
473 down_write(&nilfs->ns_sem);
474 nilfs_move_2nd_super(sb, sb2off);
476 sbp = nilfs_prepare_super(sb, 0);
478 nilfs_set_log_cursor(sbp[0], nilfs);
480 * Drop NILFS_RESIZE_FS flag for compatibility with
481 * mount-time resize which may be implemented in a
484 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
486 sbp[0]->s_dev_size = cpu_to_le64(newsize);
487 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
489 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
490 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
492 up_write(&nilfs->ns_sem);
495 * Reset the range of allocatable segments last. This order
496 * is important in the case of expansion because the secondary
497 * superblock must be protected from log write until migration
501 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
506 static void nilfs_put_super(struct super_block *sb)
508 struct the_nilfs *nilfs = sb->s_fs_info;
510 nilfs_detach_log_writer(sb);
512 if (!sb_rdonly(sb)) {
513 down_write(&nilfs->ns_sem);
514 nilfs_cleanup_super(sb);
515 up_write(&nilfs->ns_sem);
518 nilfs_sysfs_delete_device_group(nilfs);
519 iput(nilfs->ns_sufile);
520 iput(nilfs->ns_cpfile);
523 destroy_nilfs(nilfs);
524 sb->s_fs_info = NULL;
527 static int nilfs_sync_fs(struct super_block *sb, int wait)
529 struct the_nilfs *nilfs = sb->s_fs_info;
530 struct nilfs_super_block **sbp;
533 /* This function is called when super block should be written back */
535 err = nilfs_construct_segment(sb);
537 down_write(&nilfs->ns_sem);
538 if (nilfs_sb_dirty(nilfs)) {
539 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
541 nilfs_set_log_cursor(sbp[0], nilfs);
542 nilfs_commit_super(sb, NILFS_SB_COMMIT);
545 up_write(&nilfs->ns_sem);
548 err = nilfs_flush_device(nilfs);
553 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
554 struct nilfs_root **rootp)
556 struct the_nilfs *nilfs = sb->s_fs_info;
557 struct nilfs_root *root;
558 struct nilfs_checkpoint *raw_cp;
559 struct buffer_head *bh_cp;
562 root = nilfs_find_or_create_root(
563 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
568 goto reuse; /* already attached checkpoint */
570 down_read(&nilfs->ns_segctor_sem);
571 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
573 up_read(&nilfs->ns_segctor_sem);
575 if (err == -ENOENT || err == -EINVAL) {
576 nilfs_msg(sb, KERN_ERR,
577 "Invalid checkpoint (checkpoint number=%llu)",
578 (unsigned long long)cno);
584 err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
585 &raw_cp->cp_ifile_inode, &root->ifile);
589 atomic64_set(&root->inodes_count,
590 le64_to_cpu(raw_cp->cp_inodes_count));
591 atomic64_set(&root->blocks_count,
592 le64_to_cpu(raw_cp->cp_blocks_count));
594 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
601 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
603 nilfs_put_root(root);
608 static int nilfs_freeze(struct super_block *sb)
610 struct the_nilfs *nilfs = sb->s_fs_info;
616 /* Mark super block clean */
617 down_write(&nilfs->ns_sem);
618 err = nilfs_cleanup_super(sb);
619 up_write(&nilfs->ns_sem);
623 static int nilfs_unfreeze(struct super_block *sb)
625 struct the_nilfs *nilfs = sb->s_fs_info;
630 down_write(&nilfs->ns_sem);
631 nilfs_setup_super(sb, false);
632 up_write(&nilfs->ns_sem);
636 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
638 struct super_block *sb = dentry->d_sb;
639 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
640 struct the_nilfs *nilfs = root->nilfs;
641 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
642 unsigned long long blocks;
643 unsigned long overhead;
644 unsigned long nrsvblocks;
645 sector_t nfreeblocks;
646 u64 nmaxinodes, nfreeinodes;
650 * Compute all of the segment blocks
652 * The blocks before first segment and after last segment
655 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
656 - nilfs->ns_first_data_block;
657 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
660 * Compute the overhead
662 * When distributing meta data blocks outside segment structure,
663 * We must count them as the overhead.
667 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
671 err = nilfs_ifile_count_free_inodes(root->ifile,
672 &nmaxinodes, &nfreeinodes);
674 nilfs_msg(sb, KERN_WARNING,
675 "failed to count free inodes: err=%d", err);
676 if (err == -ERANGE) {
678 * If nilfs_palloc_count_max_entries() returns
679 * -ERANGE error code then we simply treat
680 * curent inodes count as maximum possible and
681 * zero as free inodes value.
683 nmaxinodes = atomic64_read(&root->inodes_count);
690 buf->f_type = NILFS_SUPER_MAGIC;
691 buf->f_bsize = sb->s_blocksize;
692 buf->f_blocks = blocks - overhead;
693 buf->f_bfree = nfreeblocks;
694 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
695 (buf->f_bfree - nrsvblocks) : 0;
696 buf->f_files = nmaxinodes;
697 buf->f_ffree = nfreeinodes;
698 buf->f_namelen = NILFS_NAME_LEN;
699 buf->f_fsid.val[0] = (u32)id;
700 buf->f_fsid.val[1] = (u32)(id >> 32);
705 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
707 struct super_block *sb = dentry->d_sb;
708 struct the_nilfs *nilfs = sb->s_fs_info;
709 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
711 if (!nilfs_test_opt(nilfs, BARRIER))
712 seq_puts(seq, ",nobarrier");
713 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
714 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
715 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
716 seq_puts(seq, ",errors=panic");
717 if (nilfs_test_opt(nilfs, ERRORS_CONT))
718 seq_puts(seq, ",errors=continue");
719 if (nilfs_test_opt(nilfs, STRICT_ORDER))
720 seq_puts(seq, ",order=strict");
721 if (nilfs_test_opt(nilfs, NORECOVERY))
722 seq_puts(seq, ",norecovery");
723 if (nilfs_test_opt(nilfs, DISCARD))
724 seq_puts(seq, ",discard");
729 static const struct super_operations nilfs_sops = {
730 .alloc_inode = nilfs_alloc_inode,
731 .destroy_inode = nilfs_destroy_inode,
732 .dirty_inode = nilfs_dirty_inode,
733 .evict_inode = nilfs_evict_inode,
734 .put_super = nilfs_put_super,
735 .sync_fs = nilfs_sync_fs,
736 .freeze_fs = nilfs_freeze,
737 .unfreeze_fs = nilfs_unfreeze,
738 .statfs = nilfs_statfs,
739 .remount_fs = nilfs_remount,
740 .show_options = nilfs_show_options
744 Opt_err_cont, Opt_err_panic, Opt_err_ro,
745 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
746 Opt_discard, Opt_nodiscard, Opt_err,
749 static match_table_t tokens = {
750 {Opt_err_cont, "errors=continue"},
751 {Opt_err_panic, "errors=panic"},
752 {Opt_err_ro, "errors=remount-ro"},
753 {Opt_barrier, "barrier"},
754 {Opt_nobarrier, "nobarrier"},
755 {Opt_snapshot, "cp=%u"},
756 {Opt_order, "order=%s"},
757 {Opt_norecovery, "norecovery"},
758 {Opt_discard, "discard"},
759 {Opt_nodiscard, "nodiscard"},
763 static int parse_options(char *options, struct super_block *sb, int is_remount)
765 struct the_nilfs *nilfs = sb->s_fs_info;
767 substring_t args[MAX_OPT_ARGS];
772 while ((p = strsep(&options, ",")) != NULL) {
778 token = match_token(p, tokens, args);
781 nilfs_set_opt(nilfs, BARRIER);
784 nilfs_clear_opt(nilfs, BARRIER);
787 if (strcmp(args[0].from, "relaxed") == 0)
788 /* Ordered data semantics */
789 nilfs_clear_opt(nilfs, STRICT_ORDER);
790 else if (strcmp(args[0].from, "strict") == 0)
791 /* Strict in-order semantics */
792 nilfs_set_opt(nilfs, STRICT_ORDER);
797 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
800 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
803 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
807 nilfs_msg(sb, KERN_ERR,
808 "\"%s\" option is invalid for remount",
814 nilfs_set_opt(nilfs, NORECOVERY);
817 nilfs_set_opt(nilfs, DISCARD);
820 nilfs_clear_opt(nilfs, DISCARD);
823 nilfs_msg(sb, KERN_ERR,
824 "unrecognized mount option \"%s\"", p);
832 nilfs_set_default_options(struct super_block *sb,
833 struct nilfs_super_block *sbp)
835 struct the_nilfs *nilfs = sb->s_fs_info;
837 nilfs->ns_mount_opt =
838 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
841 static int nilfs_setup_super(struct super_block *sb, int is_mount)
843 struct the_nilfs *nilfs = sb->s_fs_info;
844 struct nilfs_super_block **sbp;
848 /* nilfs->ns_sem must be locked by the caller. */
849 sbp = nilfs_prepare_super(sb, 0);
854 goto skip_mount_setup;
856 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
857 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
859 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
860 nilfs_msg(sb, KERN_WARNING, "mounting fs with errors");
862 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
863 nilfs_msg(sb, KERN_WARNING, "maximal mount count reached");
867 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
869 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
870 sbp[0]->s_mtime = cpu_to_le64(get_seconds());
874 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
875 /* synchronize sbp[1] with sbp[0] */
877 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
878 return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
881 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
882 u64 pos, int blocksize,
883 struct buffer_head **pbh)
885 unsigned long long sb_index = pos;
886 unsigned long offset;
888 offset = do_div(sb_index, blocksize);
889 *pbh = sb_bread(sb, sb_index);
892 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
895 int nilfs_store_magic_and_option(struct super_block *sb,
896 struct nilfs_super_block *sbp,
899 struct the_nilfs *nilfs = sb->s_fs_info;
901 sb->s_magic = le16_to_cpu(sbp->s_magic);
903 /* FS independent flags */
904 #ifdef NILFS_ATIME_DISABLE
905 sb->s_flags |= MS_NOATIME;
908 nilfs_set_default_options(sb, sbp);
910 nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
911 nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
912 nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
913 nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
915 return !parse_options(data, sb, 0) ? -EINVAL : 0;
918 int nilfs_check_feature_compatibility(struct super_block *sb,
919 struct nilfs_super_block *sbp)
923 features = le64_to_cpu(sbp->s_feature_incompat) &
924 ~NILFS_FEATURE_INCOMPAT_SUPP;
926 nilfs_msg(sb, KERN_ERR,
927 "couldn't mount because of unsupported optional features (%llx)",
928 (unsigned long long)features);
931 features = le64_to_cpu(sbp->s_feature_compat_ro) &
932 ~NILFS_FEATURE_COMPAT_RO_SUPP;
933 if (!sb_rdonly(sb) && features) {
934 nilfs_msg(sb, KERN_ERR,
935 "couldn't mount RDWR because of unsupported optional features (%llx)",
936 (unsigned long long)features);
942 static int nilfs_get_root_dentry(struct super_block *sb,
943 struct nilfs_root *root,
944 struct dentry **root_dentry)
947 struct dentry *dentry;
950 inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
952 ret = PTR_ERR(inode);
953 nilfs_msg(sb, KERN_ERR, "error %d getting root inode", ret);
956 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
958 nilfs_msg(sb, KERN_ERR, "corrupt root inode");
963 if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
964 dentry = d_find_alias(inode);
966 dentry = d_make_root(inode);
975 dentry = d_obtain_root(inode);
976 if (IS_ERR(dentry)) {
977 ret = PTR_ERR(dentry);
981 *root_dentry = dentry;
986 nilfs_msg(sb, KERN_ERR, "error %d getting root dentry", ret);
990 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
991 struct dentry **root_dentry)
993 struct the_nilfs *nilfs = s->s_fs_info;
994 struct nilfs_root *root;
997 mutex_lock(&nilfs->ns_snapshot_mount_mutex);
999 down_read(&nilfs->ns_segctor_sem);
1000 ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
1001 up_read(&nilfs->ns_segctor_sem);
1003 ret = (ret == -ENOENT) ? -EINVAL : ret;
1006 nilfs_msg(s, KERN_ERR,
1007 "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
1008 (unsigned long long)cno);
1013 ret = nilfs_attach_checkpoint(s, cno, false, &root);
1015 nilfs_msg(s, KERN_ERR,
1016 "error %d while loading snapshot (checkpoint number=%llu)",
1017 ret, (unsigned long long)cno);
1020 ret = nilfs_get_root_dentry(s, root, root_dentry);
1021 nilfs_put_root(root);
1023 mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
1028 * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
1029 * @root_dentry: root dentry of the tree to be shrunk
1031 * This function returns true if the tree was in-use.
1033 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
1035 shrink_dcache_parent(root_dentry);
1036 return d_count(root_dentry) > 1;
1039 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1041 struct the_nilfs *nilfs = sb->s_fs_info;
1042 struct nilfs_root *root;
1043 struct inode *inode;
1044 struct dentry *dentry;
1047 if (cno > nilfs->ns_cno)
1050 if (cno >= nilfs_last_cno(nilfs))
1051 return true; /* protect recent checkpoints */
1054 root = nilfs_lookup_root(nilfs, cno);
1056 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1058 dentry = d_find_alias(inode);
1060 ret = nilfs_tree_is_busy(dentry);
1065 nilfs_put_root(root);
1071 * nilfs_fill_super() - initialize a super block instance
1073 * @data: mount options
1074 * @silent: silent mode flag
1076 * This function is called exclusively by nilfs->ns_mount_mutex.
1077 * So, the recovery process is protected from other simultaneous mounts.
1080 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1082 struct the_nilfs *nilfs;
1083 struct nilfs_root *fsroot;
1087 nilfs = alloc_nilfs(sb);
1091 sb->s_fs_info = nilfs;
1093 err = init_nilfs(nilfs, sb, (char *)data);
1097 sb->s_op = &nilfs_sops;
1098 sb->s_export_op = &nilfs_export_ops;
1100 sb->s_time_gran = 1;
1101 sb->s_max_links = NILFS_LINK_MAX;
1103 sb->s_bdi = bdi_get(sb->s_bdev->bd_bdi);
1105 err = load_nilfs(nilfs, sb);
1109 cno = nilfs_last_cno(nilfs);
1110 err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1112 nilfs_msg(sb, KERN_ERR,
1113 "error %d while loading last checkpoint (checkpoint number=%llu)",
1114 err, (unsigned long long)cno);
1118 if (!sb_rdonly(sb)) {
1119 err = nilfs_attach_log_writer(sb, fsroot);
1121 goto failed_checkpoint;
1124 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1126 goto failed_segctor;
1128 nilfs_put_root(fsroot);
1130 if (!sb_rdonly(sb)) {
1131 down_write(&nilfs->ns_sem);
1132 nilfs_setup_super(sb, true);
1133 up_write(&nilfs->ns_sem);
1139 nilfs_detach_log_writer(sb);
1142 nilfs_put_root(fsroot);
1145 nilfs_sysfs_delete_device_group(nilfs);
1146 iput(nilfs->ns_sufile);
1147 iput(nilfs->ns_cpfile);
1148 iput(nilfs->ns_dat);
1151 destroy_nilfs(nilfs);
1155 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1157 struct the_nilfs *nilfs = sb->s_fs_info;
1158 unsigned long old_sb_flags;
1159 unsigned long old_mount_opt;
1162 sync_filesystem(sb);
1163 old_sb_flags = sb->s_flags;
1164 old_mount_opt = nilfs->ns_mount_opt;
1166 if (!parse_options(data, sb, 1)) {
1170 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
1174 if (!nilfs_valid_fs(nilfs)) {
1175 nilfs_msg(sb, KERN_WARNING,
1176 "couldn't remount because the filesystem is in an incomplete recovery state");
1180 if ((bool)(*flags & MS_RDONLY) == sb_rdonly(sb))
1182 if (*flags & MS_RDONLY) {
1183 sb->s_flags |= MS_RDONLY;
1186 * Remounting a valid RW partition RDONLY, so set
1187 * the RDONLY flag and then mark the partition as valid again.
1189 down_write(&nilfs->ns_sem);
1190 nilfs_cleanup_super(sb);
1191 up_write(&nilfs->ns_sem);
1194 struct nilfs_root *root;
1197 * Mounting a RDONLY partition read-write, so reread and
1198 * store the current valid flag. (It may have been changed
1199 * by fsck since we originally mounted the partition.)
1201 down_read(&nilfs->ns_sem);
1202 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1203 ~NILFS_FEATURE_COMPAT_RO_SUPP;
1204 up_read(&nilfs->ns_sem);
1206 nilfs_msg(sb, KERN_WARNING,
1207 "couldn't remount RDWR because of unsupported optional features (%llx)",
1208 (unsigned long long)features);
1213 sb->s_flags &= ~MS_RDONLY;
1215 root = NILFS_I(d_inode(sb->s_root))->i_root;
1216 err = nilfs_attach_log_writer(sb, root);
1220 down_write(&nilfs->ns_sem);
1221 nilfs_setup_super(sb, true);
1222 up_write(&nilfs->ns_sem);
1228 sb->s_flags = old_sb_flags;
1229 nilfs->ns_mount_opt = old_mount_opt;
1233 struct nilfs_super_data {
1234 struct block_device *bdev;
1239 static int nilfs_parse_snapshot_option(const char *option,
1240 const substring_t *arg,
1241 struct nilfs_super_data *sd)
1243 unsigned long long val;
1244 const char *msg = NULL;
1247 if (!(sd->flags & MS_RDONLY)) {
1248 msg = "read-only option is not specified";
1252 err = kstrtoull(arg->from, 0, &val);
1255 msg = "too large checkpoint number";
1257 msg = "malformed argument";
1259 } else if (val == 0) {
1260 msg = "invalid checkpoint number 0";
1267 nilfs_msg(NULL, KERN_ERR, "invalid option \"%s\": %s", option, msg);
1272 * nilfs_identify - pre-read mount options needed to identify mount instance
1273 * @data: mount options
1274 * @sd: nilfs_super_data
1276 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1278 char *p, *options = data;
1279 substring_t args[MAX_OPT_ARGS];
1284 p = strsep(&options, ",");
1285 if (p != NULL && *p) {
1286 token = match_token(p, tokens, args);
1287 if (token == Opt_snapshot)
1288 ret = nilfs_parse_snapshot_option(p, &args[0],
1293 BUG_ON(options == data);
1294 *(options - 1) = ',';
1299 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1302 s->s_dev = s->s_bdev->bd_dev;
1306 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1308 return (void *)s->s_bdev == data;
1311 static struct dentry *
1312 nilfs_mount(struct file_system_type *fs_type, int flags,
1313 const char *dev_name, void *data)
1315 struct nilfs_super_data sd;
1316 struct super_block *s;
1317 fmode_t mode = FMODE_READ | FMODE_EXCL;
1318 struct dentry *root_dentry;
1319 int err, s_new = false;
1321 if (!(flags & MS_RDONLY))
1322 mode |= FMODE_WRITE;
1324 sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1325 if (IS_ERR(sd.bdev))
1326 return ERR_CAST(sd.bdev);
1330 if (nilfs_identify((char *)data, &sd)) {
1336 * once the super is inserted into the list by sget, s_umount
1337 * will protect the lockfs code from trying to start a snapshot
1338 * while we are mounting
1340 mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1341 if (sd.bdev->bd_fsfreeze_count > 0) {
1342 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1346 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1348 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1357 /* New superblock instance created */
1359 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1360 sb_set_blocksize(s, block_size(sd.bdev));
1362 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1366 s->s_flags |= MS_ACTIVE;
1367 } else if (!sd.cno) {
1368 if (nilfs_tree_is_busy(s->s_root)) {
1369 if ((flags ^ s->s_flags) & MS_RDONLY) {
1370 nilfs_msg(s, KERN_ERR,
1371 "the device already has a %s mount.",
1372 sb_rdonly(s) ? "read-only" : "read/write");
1378 * Try remount to setup mount states if the current
1379 * tree is not mounted and only snapshots use this sb.
1381 err = nilfs_remount(s, &flags, data);
1388 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1392 root_dentry = dget(s->s_root);
1396 blkdev_put(sd.bdev, mode);
1401 deactivate_locked_super(s);
1405 blkdev_put(sd.bdev, mode);
1406 return ERR_PTR(err);
1409 struct file_system_type nilfs_fs_type = {
1410 .owner = THIS_MODULE,
1412 .mount = nilfs_mount,
1413 .kill_sb = kill_block_super,
1414 .fs_flags = FS_REQUIRES_DEV,
1416 MODULE_ALIAS_FS("nilfs2");
1418 static void nilfs_inode_init_once(void *obj)
1420 struct nilfs_inode_info *ii = obj;
1422 INIT_LIST_HEAD(&ii->i_dirty);
1423 #ifdef CONFIG_NILFS_XATTR
1424 init_rwsem(&ii->xattr_sem);
1426 inode_init_once(&ii->vfs_inode);
1429 static void nilfs_segbuf_init_once(void *obj)
1431 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1434 static void nilfs_destroy_cachep(void)
1437 * Make sure all delayed rcu free inodes are flushed before we
1442 kmem_cache_destroy(nilfs_inode_cachep);
1443 kmem_cache_destroy(nilfs_transaction_cachep);
1444 kmem_cache_destroy(nilfs_segbuf_cachep);
1445 kmem_cache_destroy(nilfs_btree_path_cache);
1448 static int __init nilfs_init_cachep(void)
1450 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1451 sizeof(struct nilfs_inode_info), 0,
1452 SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1453 nilfs_inode_init_once);
1454 if (!nilfs_inode_cachep)
1457 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1458 sizeof(struct nilfs_transaction_info), 0,
1459 SLAB_RECLAIM_ACCOUNT, NULL);
1460 if (!nilfs_transaction_cachep)
1463 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1464 sizeof(struct nilfs_segment_buffer), 0,
1465 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1466 if (!nilfs_segbuf_cachep)
1469 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1470 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1472 if (!nilfs_btree_path_cache)
1478 nilfs_destroy_cachep();
1482 static int __init init_nilfs_fs(void)
1486 err = nilfs_init_cachep();
1490 err = nilfs_sysfs_init();
1494 err = register_filesystem(&nilfs_fs_type);
1496 goto deinit_sysfs_entry;
1498 printk(KERN_INFO "NILFS version 2 loaded\n");
1504 nilfs_destroy_cachep();
1509 static void __exit exit_nilfs_fs(void)
1511 nilfs_destroy_cachep();
1513 unregister_filesystem(&nilfs_fs_type);
1516 module_init(init_nilfs_fs)
1517 module_exit(exit_nilfs_fs)