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 <ryusuke@osrg.net>
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 static void nilfs_set_error(struct super_block *sb)
76 struct the_nilfs *nilfs = sb->s_fs_info;
77 struct nilfs_super_block **sbp;
79 down_write(&nilfs->ns_sem);
80 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
81 nilfs->ns_mount_state |= NILFS_ERROR_FS;
82 sbp = nilfs_prepare_super(sb, 0);
84 sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
86 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
87 nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
90 up_write(&nilfs->ns_sem);
94 * nilfs_error() - report failure condition on a filesystem
96 * nilfs_error() sets an ERROR_FS flag on the superblock as well as
97 * reporting an error message. It should be called when NILFS detects
98 * incoherences or defects of meta data on disk. As for sustainable
99 * errors such as a single-shot I/O error, nilfs_warning() or the printk()
100 * function should be used instead.
102 * The segment constructor must not call this function because it can
105 void nilfs_error(struct super_block *sb, const char *function,
106 const char *fmt, ...)
108 struct the_nilfs *nilfs = sb->s_fs_info;
109 struct va_format vaf;
117 printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
118 sb->s_id, function, &vaf);
122 if (!(sb->s_flags & MS_RDONLY)) {
125 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
126 printk(KERN_CRIT "Remounting filesystem read-only\n");
127 sb->s_flags |= MS_RDONLY;
131 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
132 panic("NILFS (device %s): panic forced after error\n",
136 void nilfs_warning(struct super_block *sb, const char *function,
137 const char *fmt, ...)
139 struct va_format vaf;
147 printk(KERN_WARNING "NILFS warning (device %s): %s: %pV\n",
148 sb->s_id, function, &vaf);
154 struct inode *nilfs_alloc_inode(struct super_block *sb)
156 struct nilfs_inode_info *ii;
158 ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
164 ii->vfs_inode.i_version = 1;
165 nilfs_mapping_init(&ii->i_btnode_cache, &ii->vfs_inode);
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);
172 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
175 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
178 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
181 void nilfs_destroy_inode(struct inode *inode)
183 call_rcu(&inode->i_rcu, nilfs_i_callback);
186 static int nilfs_sync_super(struct super_block *sb, int flag)
188 struct the_nilfs *nilfs = sb->s_fs_info;
192 set_buffer_dirty(nilfs->ns_sbh[0]);
193 if (nilfs_test_opt(nilfs, BARRIER)) {
194 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
195 WRITE_SYNC | WRITE_FLUSH_FUA);
197 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
202 "NILFS: unable to write superblock (err=%d)\n", err);
203 if (err == -EIO && nilfs->ns_sbh[1]) {
205 * sbp[0] points to newer log than sbp[1],
206 * so copy sbp[0] to sbp[1] to take over sbp[0].
208 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
210 nilfs_fall_back_super_block(nilfs);
214 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
216 nilfs->ns_sbwcount++;
219 * The latest segment becomes trailable from the position
220 * written in superblock.
222 clear_nilfs_discontinued(nilfs);
224 /* update GC protection for recent segments */
225 if (nilfs->ns_sbh[1]) {
226 if (flag == NILFS_SB_COMMIT_ALL) {
227 set_buffer_dirty(nilfs->ns_sbh[1]);
228 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
231 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
232 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
233 sbp = nilfs->ns_sbp[1];
236 spin_lock(&nilfs->ns_last_segment_lock);
237 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
238 spin_unlock(&nilfs->ns_last_segment_lock);
244 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
245 struct the_nilfs *nilfs)
247 sector_t nfreeblocks;
249 /* nilfs->ns_sem must be locked by the caller. */
250 nilfs_count_free_blocks(nilfs, &nfreeblocks);
251 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
253 spin_lock(&nilfs->ns_last_segment_lock);
254 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
255 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
256 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
257 spin_unlock(&nilfs->ns_last_segment_lock);
260 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
263 struct the_nilfs *nilfs = sb->s_fs_info;
264 struct nilfs_super_block **sbp = nilfs->ns_sbp;
266 /* nilfs->ns_sem must be locked by the caller. */
267 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
269 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
270 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
272 printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
277 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
278 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
282 nilfs_swap_super_block(nilfs);
287 int nilfs_commit_super(struct super_block *sb, int flag)
289 struct the_nilfs *nilfs = sb->s_fs_info;
290 struct nilfs_super_block **sbp = nilfs->ns_sbp;
293 /* nilfs->ns_sem must be locked by the caller. */
295 nilfs->ns_sbwtime = t;
296 sbp[0]->s_wtime = cpu_to_le64(t);
298 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
299 (unsigned char *)sbp[0],
301 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
302 sbp[1]->s_wtime = sbp[0]->s_wtime;
304 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
305 (unsigned char *)sbp[1],
308 clear_nilfs_sb_dirty(nilfs);
309 nilfs->ns_flushed_device = 1;
310 /* make sure store to ns_flushed_device cannot be reordered */
312 return nilfs_sync_super(sb, flag);
316 * nilfs_cleanup_super() - write filesystem state for cleanup
317 * @sb: super block instance to be unmounted or degraded to read-only
319 * This function restores state flags in the on-disk super block.
320 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
321 * filesystem was not clean previously.
323 int nilfs_cleanup_super(struct super_block *sb)
325 struct the_nilfs *nilfs = sb->s_fs_info;
326 struct nilfs_super_block **sbp;
327 int flag = NILFS_SB_COMMIT;
330 sbp = nilfs_prepare_super(sb, 0);
332 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
333 nilfs_set_log_cursor(sbp[0], nilfs);
334 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
336 * make the "clean" flag also to the opposite
337 * super block if both super blocks point to
338 * the same checkpoint.
340 sbp[1]->s_state = sbp[0]->s_state;
341 flag = NILFS_SB_COMMIT_ALL;
343 ret = nilfs_commit_super(sb, flag);
349 * nilfs_move_2nd_super - relocate secondary super block
350 * @sb: super block instance
351 * @sb2off: new offset of the secondary super block (in bytes)
353 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
355 struct the_nilfs *nilfs = sb->s_fs_info;
356 struct buffer_head *nsbh;
357 struct nilfs_super_block *nsbp;
358 sector_t blocknr, newblocknr;
359 unsigned long offset;
360 int sb2i; /* array index of the secondary superblock */
363 /* nilfs->ns_sem must be locked by the caller. */
364 if (nilfs->ns_sbh[1] &&
365 nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
367 blocknr = nilfs->ns_sbh[1]->b_blocknr;
368 } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
370 blocknr = nilfs->ns_sbh[0]->b_blocknr;
375 if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
376 goto out; /* super block location is unchanged */
378 /* Get new super block buffer */
379 newblocknr = sb2off >> nilfs->ns_blocksize_bits;
380 offset = sb2off & (nilfs->ns_blocksize - 1);
381 nsbh = sb_getblk(sb, newblocknr);
384 "NILFS warning: unable to move secondary superblock "
385 "to block %llu\n", (unsigned long long)newblocknr);
389 nsbp = (void *)nsbh->b_data + offset;
390 memset(nsbp, 0, nilfs->ns_blocksize);
393 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
394 brelse(nilfs->ns_sbh[sb2i]);
395 nilfs->ns_sbh[sb2i] = nsbh;
396 nilfs->ns_sbp[sb2i] = nsbp;
397 } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
398 /* secondary super block will be restored to index 1 */
399 nilfs->ns_sbh[1] = nsbh;
400 nilfs->ns_sbp[1] = nsbp;
409 * nilfs_resize_fs - resize the filesystem
410 * @sb: super block instance
411 * @newsize: new size of the filesystem (in bytes)
413 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
415 struct the_nilfs *nilfs = sb->s_fs_info;
416 struct nilfs_super_block **sbp;
417 __u64 devsize, newnsegs;
422 devsize = i_size_read(sb->s_bdev->bd_inode);
423 if (newsize > devsize)
427 * Write lock is required to protect some functions depending
428 * on the number of segments, the number of reserved segments,
431 down_write(&nilfs->ns_segctor_sem);
433 sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
434 newnsegs = sb2off >> nilfs->ns_blocksize_bits;
435 do_div(newnsegs, nilfs->ns_blocks_per_segment);
437 ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
438 up_write(&nilfs->ns_segctor_sem);
442 ret = nilfs_construct_segment(sb);
446 down_write(&nilfs->ns_sem);
447 nilfs_move_2nd_super(sb, sb2off);
449 sbp = nilfs_prepare_super(sb, 0);
451 nilfs_set_log_cursor(sbp[0], nilfs);
453 * Drop NILFS_RESIZE_FS flag for compatibility with
454 * mount-time resize which may be implemented in a
457 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
459 sbp[0]->s_dev_size = cpu_to_le64(newsize);
460 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
462 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
463 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
465 up_write(&nilfs->ns_sem);
468 * Reset the range of allocatable segments last. This order
469 * is important in the case of expansion because the secondary
470 * superblock must be protected from log write until migration
474 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
479 static void nilfs_put_super(struct super_block *sb)
481 struct the_nilfs *nilfs = sb->s_fs_info;
483 nilfs_detach_log_writer(sb);
485 if (!(sb->s_flags & MS_RDONLY)) {
486 down_write(&nilfs->ns_sem);
487 nilfs_cleanup_super(sb);
488 up_write(&nilfs->ns_sem);
491 iput(nilfs->ns_sufile);
492 iput(nilfs->ns_cpfile);
495 destroy_nilfs(nilfs);
496 sb->s_fs_info = NULL;
499 static int nilfs_sync_fs(struct super_block *sb, int wait)
501 struct the_nilfs *nilfs = sb->s_fs_info;
502 struct nilfs_super_block **sbp;
505 /* This function is called when super block should be written back */
507 err = nilfs_construct_segment(sb);
509 down_write(&nilfs->ns_sem);
510 if (nilfs_sb_dirty(nilfs)) {
511 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
513 nilfs_set_log_cursor(sbp[0], nilfs);
514 nilfs_commit_super(sb, NILFS_SB_COMMIT);
517 up_write(&nilfs->ns_sem);
520 err = nilfs_flush_device(nilfs);
525 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
526 struct nilfs_root **rootp)
528 struct the_nilfs *nilfs = sb->s_fs_info;
529 struct nilfs_root *root;
530 struct nilfs_checkpoint *raw_cp;
531 struct buffer_head *bh_cp;
534 root = nilfs_find_or_create_root(
535 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
540 goto reuse; /* already attached checkpoint */
542 down_read(&nilfs->ns_segctor_sem);
543 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
545 up_read(&nilfs->ns_segctor_sem);
547 if (err == -ENOENT || err == -EINVAL) {
549 "NILFS: Invalid checkpoint "
550 "(checkpoint number=%llu)\n",
551 (unsigned long long)cno);
557 err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
558 &raw_cp->cp_ifile_inode, &root->ifile);
562 atomic64_set(&root->inodes_count,
563 le64_to_cpu(raw_cp->cp_inodes_count));
564 atomic64_set(&root->blocks_count,
565 le64_to_cpu(raw_cp->cp_blocks_count));
567 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
574 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
576 nilfs_put_root(root);
581 static int nilfs_freeze(struct super_block *sb)
583 struct the_nilfs *nilfs = sb->s_fs_info;
586 if (sb->s_flags & MS_RDONLY)
589 /* Mark super block clean */
590 down_write(&nilfs->ns_sem);
591 err = nilfs_cleanup_super(sb);
592 up_write(&nilfs->ns_sem);
596 static int nilfs_unfreeze(struct super_block *sb)
598 struct the_nilfs *nilfs = sb->s_fs_info;
600 if (sb->s_flags & MS_RDONLY)
603 down_write(&nilfs->ns_sem);
604 nilfs_setup_super(sb, false);
605 up_write(&nilfs->ns_sem);
609 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
611 struct super_block *sb = dentry->d_sb;
612 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
613 struct the_nilfs *nilfs = root->nilfs;
614 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
615 unsigned long long blocks;
616 unsigned long overhead;
617 unsigned long nrsvblocks;
618 sector_t nfreeblocks;
619 u64 nmaxinodes, nfreeinodes;
623 * Compute all of the segment blocks
625 * The blocks before first segment and after last segment
628 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
629 - nilfs->ns_first_data_block;
630 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
633 * Compute the overhead
635 * When distributing meta data blocks outside segment structure,
636 * We must count them as the overhead.
640 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
644 err = nilfs_ifile_count_free_inodes(root->ifile,
645 &nmaxinodes, &nfreeinodes);
648 "NILFS warning: fail to count free inodes: err %d.\n",
650 if (err == -ERANGE) {
652 * If nilfs_palloc_count_max_entries() returns
653 * -ERANGE error code then we simply treat
654 * curent inodes count as maximum possible and
655 * zero as free inodes value.
657 nmaxinodes = atomic64_read(&root->inodes_count);
664 buf->f_type = NILFS_SUPER_MAGIC;
665 buf->f_bsize = sb->s_blocksize;
666 buf->f_blocks = blocks - overhead;
667 buf->f_bfree = nfreeblocks;
668 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
669 (buf->f_bfree - nrsvblocks) : 0;
670 buf->f_files = nmaxinodes;
671 buf->f_ffree = nfreeinodes;
672 buf->f_namelen = NILFS_NAME_LEN;
673 buf->f_fsid.val[0] = (u32)id;
674 buf->f_fsid.val[1] = (u32)(id >> 32);
679 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
681 struct super_block *sb = dentry->d_sb;
682 struct the_nilfs *nilfs = sb->s_fs_info;
683 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
685 if (!nilfs_test_opt(nilfs, BARRIER))
686 seq_puts(seq, ",nobarrier");
687 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
688 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
689 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
690 seq_puts(seq, ",errors=panic");
691 if (nilfs_test_opt(nilfs, ERRORS_CONT))
692 seq_puts(seq, ",errors=continue");
693 if (nilfs_test_opt(nilfs, STRICT_ORDER))
694 seq_puts(seq, ",order=strict");
695 if (nilfs_test_opt(nilfs, NORECOVERY))
696 seq_puts(seq, ",norecovery");
697 if (nilfs_test_opt(nilfs, DISCARD))
698 seq_puts(seq, ",discard");
703 static const struct super_operations nilfs_sops = {
704 .alloc_inode = nilfs_alloc_inode,
705 .destroy_inode = nilfs_destroy_inode,
706 .dirty_inode = nilfs_dirty_inode,
707 .evict_inode = nilfs_evict_inode,
708 .put_super = nilfs_put_super,
709 .sync_fs = nilfs_sync_fs,
710 .freeze_fs = nilfs_freeze,
711 .unfreeze_fs = nilfs_unfreeze,
712 .statfs = nilfs_statfs,
713 .remount_fs = nilfs_remount,
714 .show_options = nilfs_show_options
718 Opt_err_cont, Opt_err_panic, Opt_err_ro,
719 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
720 Opt_discard, Opt_nodiscard, Opt_err,
723 static match_table_t tokens = {
724 {Opt_err_cont, "errors=continue"},
725 {Opt_err_panic, "errors=panic"},
726 {Opt_err_ro, "errors=remount-ro"},
727 {Opt_barrier, "barrier"},
728 {Opt_nobarrier, "nobarrier"},
729 {Opt_snapshot, "cp=%u"},
730 {Opt_order, "order=%s"},
731 {Opt_norecovery, "norecovery"},
732 {Opt_discard, "discard"},
733 {Opt_nodiscard, "nodiscard"},
737 static int parse_options(char *options, struct super_block *sb, int is_remount)
739 struct the_nilfs *nilfs = sb->s_fs_info;
741 substring_t args[MAX_OPT_ARGS];
746 while ((p = strsep(&options, ",")) != NULL) {
751 token = match_token(p, tokens, args);
754 nilfs_set_opt(nilfs, BARRIER);
757 nilfs_clear_opt(nilfs, BARRIER);
760 if (strcmp(args[0].from, "relaxed") == 0)
761 /* Ordered data semantics */
762 nilfs_clear_opt(nilfs, STRICT_ORDER);
763 else if (strcmp(args[0].from, "strict") == 0)
764 /* Strict in-order semantics */
765 nilfs_set_opt(nilfs, STRICT_ORDER);
770 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
773 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
776 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
781 "NILFS: \"%s\" option is invalid "
782 "for remount.\n", p);
787 nilfs_set_opt(nilfs, NORECOVERY);
790 nilfs_set_opt(nilfs, DISCARD);
793 nilfs_clear_opt(nilfs, DISCARD);
797 "NILFS: Unrecognized mount option \"%s\"\n", p);
805 nilfs_set_default_options(struct super_block *sb,
806 struct nilfs_super_block *sbp)
808 struct the_nilfs *nilfs = sb->s_fs_info;
810 nilfs->ns_mount_opt =
811 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
814 static int nilfs_setup_super(struct super_block *sb, int is_mount)
816 struct the_nilfs *nilfs = sb->s_fs_info;
817 struct nilfs_super_block **sbp;
821 /* nilfs->ns_sem must be locked by the caller. */
822 sbp = nilfs_prepare_super(sb, 0);
827 goto skip_mount_setup;
829 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
830 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
832 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
834 "NILFS warning: mounting fs with errors\n");
836 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
838 "NILFS warning: maximal mount count reached\n");
842 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
844 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
845 sbp[0]->s_mtime = cpu_to_le64(get_seconds());
849 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
850 /* synchronize sbp[1] with sbp[0] */
852 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
853 return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
856 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
857 u64 pos, int blocksize,
858 struct buffer_head **pbh)
860 unsigned long long sb_index = pos;
861 unsigned long offset;
863 offset = do_div(sb_index, blocksize);
864 *pbh = sb_bread(sb, sb_index);
867 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
870 int nilfs_store_magic_and_option(struct super_block *sb,
871 struct nilfs_super_block *sbp,
874 struct the_nilfs *nilfs = sb->s_fs_info;
876 sb->s_magic = le16_to_cpu(sbp->s_magic);
878 /* FS independent flags */
879 #ifdef NILFS_ATIME_DISABLE
880 sb->s_flags |= MS_NOATIME;
883 nilfs_set_default_options(sb, sbp);
885 nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
886 nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
887 nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
888 nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
890 return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
893 int nilfs_check_feature_compatibility(struct super_block *sb,
894 struct nilfs_super_block *sbp)
898 features = le64_to_cpu(sbp->s_feature_incompat) &
899 ~NILFS_FEATURE_INCOMPAT_SUPP;
901 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
902 "optional features (%llx)\n",
903 (unsigned long long)features);
906 features = le64_to_cpu(sbp->s_feature_compat_ro) &
907 ~NILFS_FEATURE_COMPAT_RO_SUPP;
908 if (!(sb->s_flags & MS_RDONLY) && features) {
909 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
910 "unsupported optional features (%llx)\n",
911 (unsigned long long)features);
917 static int nilfs_get_root_dentry(struct super_block *sb,
918 struct nilfs_root *root,
919 struct dentry **root_dentry)
922 struct dentry *dentry;
925 inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
927 printk(KERN_ERR "NILFS: get root inode failed\n");
928 ret = PTR_ERR(inode);
931 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
933 printk(KERN_ERR "NILFS: corrupt root inode.\n");
938 if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
939 dentry = d_find_alias(inode);
941 dentry = d_make_root(inode);
950 dentry = d_obtain_root(inode);
951 if (IS_ERR(dentry)) {
952 ret = PTR_ERR(dentry);
956 *root_dentry = dentry;
961 printk(KERN_ERR "NILFS: get root dentry failed\n");
965 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
966 struct dentry **root_dentry)
968 struct the_nilfs *nilfs = s->s_fs_info;
969 struct nilfs_root *root;
972 mutex_lock(&nilfs->ns_snapshot_mount_mutex);
974 down_read(&nilfs->ns_segctor_sem);
975 ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
976 up_read(&nilfs->ns_segctor_sem);
978 ret = (ret == -ENOENT) ? -EINVAL : ret;
981 printk(KERN_ERR "NILFS: The specified checkpoint is "
982 "not a snapshot (checkpoint number=%llu).\n",
983 (unsigned long long)cno);
988 ret = nilfs_attach_checkpoint(s, cno, false, &root);
990 printk(KERN_ERR "NILFS: error loading snapshot "
991 "(checkpoint number=%llu).\n",
992 (unsigned long long)cno);
995 ret = nilfs_get_root_dentry(s, root, root_dentry);
996 nilfs_put_root(root);
998 mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
1003 * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
1004 * @root_dentry: root dentry of the tree to be shrunk
1006 * This function returns true if the tree was in-use.
1008 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
1010 shrink_dcache_parent(root_dentry);
1011 return d_count(root_dentry) > 1;
1014 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1016 struct the_nilfs *nilfs = sb->s_fs_info;
1017 struct nilfs_root *root;
1018 struct inode *inode;
1019 struct dentry *dentry;
1022 if (cno > nilfs->ns_cno)
1025 if (cno >= nilfs_last_cno(nilfs))
1026 return true; /* protect recent checkpoints */
1029 root = nilfs_lookup_root(nilfs, cno);
1031 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1033 dentry = d_find_alias(inode);
1035 ret = nilfs_tree_is_busy(dentry);
1040 nilfs_put_root(root);
1046 * nilfs_fill_super() - initialize a super block instance
1048 * @data: mount options
1049 * @silent: silent mode flag
1051 * This function is called exclusively by nilfs->ns_mount_mutex.
1052 * So, the recovery process is protected from other simultaneous mounts.
1055 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1057 struct the_nilfs *nilfs;
1058 struct nilfs_root *fsroot;
1062 nilfs = alloc_nilfs(sb->s_bdev);
1066 sb->s_fs_info = nilfs;
1068 err = init_nilfs(nilfs, sb, (char *)data);
1072 sb->s_op = &nilfs_sops;
1073 sb->s_export_op = &nilfs_export_ops;
1075 sb->s_time_gran = 1;
1076 sb->s_max_links = NILFS_LINK_MAX;
1078 sb->s_bdi = &bdev_get_queue(sb->s_bdev)->backing_dev_info;
1080 err = load_nilfs(nilfs, sb);
1084 cno = nilfs_last_cno(nilfs);
1085 err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1087 printk(KERN_ERR "NILFS: error loading last checkpoint "
1088 "(checkpoint number=%llu).\n", (unsigned long long)cno);
1092 if (!(sb->s_flags & MS_RDONLY)) {
1093 err = nilfs_attach_log_writer(sb, fsroot);
1095 goto failed_checkpoint;
1098 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1100 goto failed_segctor;
1102 nilfs_put_root(fsroot);
1104 if (!(sb->s_flags & MS_RDONLY)) {
1105 down_write(&nilfs->ns_sem);
1106 nilfs_setup_super(sb, true);
1107 up_write(&nilfs->ns_sem);
1113 nilfs_detach_log_writer(sb);
1116 nilfs_put_root(fsroot);
1119 iput(nilfs->ns_sufile);
1120 iput(nilfs->ns_cpfile);
1121 iput(nilfs->ns_dat);
1124 destroy_nilfs(nilfs);
1128 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1130 struct the_nilfs *nilfs = sb->s_fs_info;
1131 unsigned long old_sb_flags;
1132 unsigned long old_mount_opt;
1135 sync_filesystem(sb);
1136 old_sb_flags = sb->s_flags;
1137 old_mount_opt = nilfs->ns_mount_opt;
1139 if (!parse_options(data, sb, 1)) {
1143 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
1147 if (!nilfs_valid_fs(nilfs)) {
1148 printk(KERN_WARNING "NILFS (device %s): couldn't "
1149 "remount because the filesystem is in an "
1150 "incomplete recovery state.\n", sb->s_id);
1154 if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1156 if (*flags & MS_RDONLY) {
1157 /* Shutting down log writer */
1158 nilfs_detach_log_writer(sb);
1159 sb->s_flags |= MS_RDONLY;
1162 * Remounting a valid RW partition RDONLY, so set
1163 * the RDONLY flag and then mark the partition as valid again.
1165 down_write(&nilfs->ns_sem);
1166 nilfs_cleanup_super(sb);
1167 up_write(&nilfs->ns_sem);
1170 struct nilfs_root *root;
1173 * Mounting a RDONLY partition read-write, so reread and
1174 * store the current valid flag. (It may have been changed
1175 * by fsck since we originally mounted the partition.)
1177 down_read(&nilfs->ns_sem);
1178 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1179 ~NILFS_FEATURE_COMPAT_RO_SUPP;
1180 up_read(&nilfs->ns_sem);
1182 printk(KERN_WARNING "NILFS (device %s): couldn't "
1183 "remount RDWR because of unsupported optional "
1184 "features (%llx)\n",
1185 sb->s_id, (unsigned long long)features);
1190 sb->s_flags &= ~MS_RDONLY;
1192 root = NILFS_I(d_inode(sb->s_root))->i_root;
1193 err = nilfs_attach_log_writer(sb, root);
1197 down_write(&nilfs->ns_sem);
1198 nilfs_setup_super(sb, true);
1199 up_write(&nilfs->ns_sem);
1205 sb->s_flags = old_sb_flags;
1206 nilfs->ns_mount_opt = old_mount_opt;
1210 struct nilfs_super_data {
1211 struct block_device *bdev;
1217 * nilfs_identify - pre-read mount options needed to identify mount instance
1218 * @data: mount options
1219 * @sd: nilfs_super_data
1221 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1223 char *p, *options = data;
1224 substring_t args[MAX_OPT_ARGS];
1229 p = strsep(&options, ",");
1230 if (p != NULL && *p) {
1231 token = match_token(p, tokens, args);
1232 if (token == Opt_snapshot) {
1233 if (!(sd->flags & MS_RDONLY)) {
1236 sd->cno = simple_strtoull(args[0].from,
1239 * No need to see the end pointer;
1240 * match_token() has done syntax
1249 "NILFS: invalid mount option: %s\n", p);
1253 BUG_ON(options == data);
1254 *(options - 1) = ',';
1259 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1262 s->s_dev = s->s_bdev->bd_dev;
1266 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1268 return (void *)s->s_bdev == data;
1271 static struct dentry *
1272 nilfs_mount(struct file_system_type *fs_type, int flags,
1273 const char *dev_name, void *data)
1275 struct nilfs_super_data sd;
1276 struct super_block *s;
1277 fmode_t mode = FMODE_READ | FMODE_EXCL;
1278 struct dentry *root_dentry;
1279 int err, s_new = false;
1281 if (!(flags & MS_RDONLY))
1282 mode |= FMODE_WRITE;
1284 sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1285 if (IS_ERR(sd.bdev))
1286 return ERR_CAST(sd.bdev);
1290 if (nilfs_identify((char *)data, &sd)) {
1296 * once the super is inserted into the list by sget, s_umount
1297 * will protect the lockfs code from trying to start a snapshot
1298 * while we are mounting
1300 mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1301 if (sd.bdev->bd_fsfreeze_count > 0) {
1302 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1306 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1308 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1317 /* New superblock instance created */
1319 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1320 sb_set_blocksize(s, block_size(sd.bdev));
1322 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1326 s->s_flags |= MS_ACTIVE;
1327 } else if (!sd.cno) {
1328 if (nilfs_tree_is_busy(s->s_root)) {
1329 if ((flags ^ s->s_flags) & MS_RDONLY) {
1330 printk(KERN_ERR "NILFS: the device already "
1331 "has a %s mount.\n",
1332 (s->s_flags & MS_RDONLY) ?
1333 "read-only" : "read/write");
1339 * Try remount to setup mount states if the current
1340 * tree is not mounted and only snapshots use this sb.
1342 err = nilfs_remount(s, &flags, data);
1349 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1353 root_dentry = dget(s->s_root);
1357 blkdev_put(sd.bdev, mode);
1362 deactivate_locked_super(s);
1366 blkdev_put(sd.bdev, mode);
1367 return ERR_PTR(err);
1370 struct file_system_type nilfs_fs_type = {
1371 .owner = THIS_MODULE,
1373 .mount = nilfs_mount,
1374 .kill_sb = kill_block_super,
1375 .fs_flags = FS_REQUIRES_DEV,
1377 MODULE_ALIAS_FS("nilfs2");
1379 static void nilfs_inode_init_once(void *obj)
1381 struct nilfs_inode_info *ii = obj;
1383 INIT_LIST_HEAD(&ii->i_dirty);
1384 #ifdef CONFIG_NILFS_XATTR
1385 init_rwsem(&ii->xattr_sem);
1387 address_space_init_once(&ii->i_btnode_cache);
1388 ii->i_bmap = &ii->i_bmap_data;
1389 inode_init_once(&ii->vfs_inode);
1392 static void nilfs_segbuf_init_once(void *obj)
1394 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1397 static void nilfs_destroy_cachep(void)
1400 * Make sure all delayed rcu free inodes are flushed before we
1405 kmem_cache_destroy(nilfs_inode_cachep);
1406 kmem_cache_destroy(nilfs_transaction_cachep);
1407 kmem_cache_destroy(nilfs_segbuf_cachep);
1408 kmem_cache_destroy(nilfs_btree_path_cache);
1411 static int __init nilfs_init_cachep(void)
1413 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1414 sizeof(struct nilfs_inode_info), 0,
1415 SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1416 nilfs_inode_init_once);
1417 if (!nilfs_inode_cachep)
1420 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1421 sizeof(struct nilfs_transaction_info), 0,
1422 SLAB_RECLAIM_ACCOUNT, NULL);
1423 if (!nilfs_transaction_cachep)
1426 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1427 sizeof(struct nilfs_segment_buffer), 0,
1428 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1429 if (!nilfs_segbuf_cachep)
1432 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1433 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1435 if (!nilfs_btree_path_cache)
1441 nilfs_destroy_cachep();
1445 static int __init init_nilfs_fs(void)
1449 err = nilfs_init_cachep();
1453 err = nilfs_sysfs_init();
1457 err = register_filesystem(&nilfs_fs_type);
1459 goto deinit_sysfs_entry;
1461 printk(KERN_INFO "NILFS version 2 loaded\n");
1467 nilfs_destroy_cachep();
1472 static void __exit exit_nilfs_fs(void)
1474 nilfs_destroy_cachep();
1476 unregister_filesystem(&nilfs_fs_type);
1479 module_init(init_nilfs_fs)
1480 module_exit(exit_nilfs_fs)