1 // SPDX-License-Identifier: GPL-2.0+
3 * super.c - NILFS module and super block management.
5 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
7 * Written by Ryusuke Konishi.
10 * linux/fs/ext2/super.c
12 * Copyright (C) 1992, 1993, 1994, 1995
13 * Remy Card (card@masi.ibp.fr)
14 * Laboratoire MASI - Institut Blaise Pascal
15 * Universite Pierre et Marie Curie (Paris VI)
19 * linux/fs/minix/inode.c
21 * Copyright (C) 1991, 1992 Linus Torvalds
23 * Big-endian to little-endian byte-swapping/bitmaps by
24 * David S. Miller (davem@caip.rutgers.edu), 1995
27 #include <linux/module.h>
28 #include <linux/string.h>
29 #include <linux/slab.h>
30 #include <linux/init.h>
31 #include <linux/blkdev.h>
32 #include <linux/parser.h>
33 #include <linux/crc32.h>
34 #include <linux/vfs.h>
35 #include <linux/writeback.h>
36 #include <linux/seq_file.h>
37 #include <linux/mount.h>
46 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
52 MODULE_AUTHOR("NTT Corp.");
53 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
55 MODULE_LICENSE("GPL");
57 static struct kmem_cache *nilfs_inode_cachep;
58 struct kmem_cache *nilfs_transaction_cachep;
59 struct kmem_cache *nilfs_segbuf_cachep;
60 struct kmem_cache *nilfs_btree_path_cache;
62 static int nilfs_setup_super(struct super_block *sb, int is_mount);
63 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
65 void __nilfs_msg(struct super_block *sb, const char *level, const char *fmt,
75 printk("%sNILFS (%s): %pV\n", level, sb->s_id, &vaf);
77 printk("%sNILFS: %pV\n", level, &vaf);
81 static void nilfs_set_error(struct super_block *sb)
83 struct the_nilfs *nilfs = sb->s_fs_info;
84 struct nilfs_super_block **sbp;
86 down_write(&nilfs->ns_sem);
87 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
88 nilfs->ns_mount_state |= NILFS_ERROR_FS;
89 sbp = nilfs_prepare_super(sb, 0);
91 sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
93 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
94 nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
97 up_write(&nilfs->ns_sem);
101 * __nilfs_error() - report failure condition on a filesystem
103 * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
104 * reporting an error message. This function should be called when
105 * NILFS detects incoherences or defects of meta data on disk.
107 * This implements the body of nilfs_error() macro. Normally,
108 * nilfs_error() should be used. As for sustainable errors such as a
109 * single-shot I/O error, nilfs_msg() should be used instead.
111 * Callers should not add a trailing newline since this will do it.
113 void __nilfs_error(struct super_block *sb, const char *function,
114 const char *fmt, ...)
116 struct the_nilfs *nilfs = sb->s_fs_info;
117 struct va_format vaf;
125 printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
126 sb->s_id, function, &vaf);
130 if (!sb_rdonly(sb)) {
133 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
134 printk(KERN_CRIT "Remounting filesystem read-only\n");
135 sb->s_flags |= SB_RDONLY;
139 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
140 panic("NILFS (device %s): panic forced after error\n",
144 struct inode *nilfs_alloc_inode(struct super_block *sb)
146 struct nilfs_inode_info *ii;
148 ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
154 ii->i_assoc_inode = NULL;
155 ii->i_bmap = &ii->i_bmap_data;
156 return &ii->vfs_inode;
159 static void nilfs_free_inode(struct inode *inode)
161 if (nilfs_is_metadata_file_inode(inode))
162 nilfs_mdt_destroy(inode);
164 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
167 static int nilfs_sync_super(struct super_block *sb, int flag)
169 struct the_nilfs *nilfs = sb->s_fs_info;
173 set_buffer_dirty(nilfs->ns_sbh[0]);
174 if (nilfs_test_opt(nilfs, BARRIER)) {
175 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
176 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
178 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
182 nilfs_msg(sb, KERN_ERR, "unable to write superblock: err=%d",
184 if (err == -EIO && nilfs->ns_sbh[1]) {
186 * sbp[0] points to newer log than sbp[1],
187 * so copy sbp[0] to sbp[1] to take over sbp[0].
189 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
191 nilfs_fall_back_super_block(nilfs);
195 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
197 nilfs->ns_sbwcount++;
200 * The latest segment becomes trailable from the position
201 * written in superblock.
203 clear_nilfs_discontinued(nilfs);
205 /* update GC protection for recent segments */
206 if (nilfs->ns_sbh[1]) {
207 if (flag == NILFS_SB_COMMIT_ALL) {
208 set_buffer_dirty(nilfs->ns_sbh[1]);
209 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
212 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
213 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
214 sbp = nilfs->ns_sbp[1];
217 spin_lock(&nilfs->ns_last_segment_lock);
218 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
219 spin_unlock(&nilfs->ns_last_segment_lock);
225 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
226 struct the_nilfs *nilfs)
228 sector_t nfreeblocks;
230 /* nilfs->ns_sem must be locked by the caller. */
231 nilfs_count_free_blocks(nilfs, &nfreeblocks);
232 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
234 spin_lock(&nilfs->ns_last_segment_lock);
235 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
236 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
237 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
238 spin_unlock(&nilfs->ns_last_segment_lock);
241 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
244 struct the_nilfs *nilfs = sb->s_fs_info;
245 struct nilfs_super_block **sbp = nilfs->ns_sbp;
247 /* nilfs->ns_sem must be locked by the caller. */
248 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
250 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
251 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
253 nilfs_msg(sb, KERN_CRIT, "superblock broke");
257 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
258 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
262 nilfs_swap_super_block(nilfs);
267 int nilfs_commit_super(struct super_block *sb, int flag)
269 struct the_nilfs *nilfs = sb->s_fs_info;
270 struct nilfs_super_block **sbp = nilfs->ns_sbp;
273 /* nilfs->ns_sem must be locked by the caller. */
274 t = ktime_get_real_seconds();
275 nilfs->ns_sbwtime = t;
276 sbp[0]->s_wtime = cpu_to_le64(t);
278 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
279 (unsigned char *)sbp[0],
281 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
282 sbp[1]->s_wtime = sbp[0]->s_wtime;
284 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
285 (unsigned char *)sbp[1],
288 clear_nilfs_sb_dirty(nilfs);
289 nilfs->ns_flushed_device = 1;
290 /* make sure store to ns_flushed_device cannot be reordered */
292 return nilfs_sync_super(sb, flag);
296 * nilfs_cleanup_super() - write filesystem state for cleanup
297 * @sb: super block instance to be unmounted or degraded to read-only
299 * This function restores state flags in the on-disk super block.
300 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
301 * filesystem was not clean previously.
303 int nilfs_cleanup_super(struct super_block *sb)
305 struct the_nilfs *nilfs = sb->s_fs_info;
306 struct nilfs_super_block **sbp;
307 int flag = NILFS_SB_COMMIT;
310 sbp = nilfs_prepare_super(sb, 0);
312 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
313 nilfs_set_log_cursor(sbp[0], nilfs);
314 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
316 * make the "clean" flag also to the opposite
317 * super block if both super blocks point to
318 * the same checkpoint.
320 sbp[1]->s_state = sbp[0]->s_state;
321 flag = NILFS_SB_COMMIT_ALL;
323 ret = nilfs_commit_super(sb, flag);
329 * nilfs_move_2nd_super - relocate secondary super block
330 * @sb: super block instance
331 * @sb2off: new offset of the secondary super block (in bytes)
333 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
335 struct the_nilfs *nilfs = sb->s_fs_info;
336 struct buffer_head *nsbh;
337 struct nilfs_super_block *nsbp;
338 sector_t blocknr, newblocknr;
339 unsigned long offset;
340 int sb2i; /* array index of the secondary superblock */
343 /* nilfs->ns_sem must be locked by the caller. */
344 if (nilfs->ns_sbh[1] &&
345 nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
347 blocknr = nilfs->ns_sbh[1]->b_blocknr;
348 } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
350 blocknr = nilfs->ns_sbh[0]->b_blocknr;
355 if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
356 goto out; /* super block location is unchanged */
358 /* Get new super block buffer */
359 newblocknr = sb2off >> nilfs->ns_blocksize_bits;
360 offset = sb2off & (nilfs->ns_blocksize - 1);
361 nsbh = sb_getblk(sb, newblocknr);
363 nilfs_msg(sb, KERN_WARNING,
364 "unable to move secondary superblock to block %llu",
365 (unsigned long long)newblocknr);
369 nsbp = (void *)nsbh->b_data + offset;
374 * The position of the second superblock only changes by 4KiB,
375 * which is larger than the maximum superblock data size
376 * (= 1KiB), so there is no need to use memmove() to allow
377 * overlap between source and destination.
379 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
382 * Zero fill after copy to avoid overwriting in case of move
383 * within the same block.
385 memset(nsbh->b_data, 0, offset);
386 memset((void *)nsbp + nilfs->ns_sbsize, 0,
387 nsbh->b_size - offset - nilfs->ns_sbsize);
389 memset(nsbh->b_data, 0, nsbh->b_size);
391 set_buffer_uptodate(nsbh);
395 brelse(nilfs->ns_sbh[sb2i]);
396 nilfs->ns_sbh[sb2i] = nsbh;
397 nilfs->ns_sbp[sb2i] = nsbp;
398 } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
399 /* secondary super block will be restored to index 1 */
400 nilfs->ns_sbh[1] = nsbh;
401 nilfs->ns_sbp[1] = nsbp;
410 * nilfs_resize_fs - resize the filesystem
411 * @sb: super block instance
412 * @newsize: new size of the filesystem (in bytes)
414 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
416 struct the_nilfs *nilfs = sb->s_fs_info;
417 struct nilfs_super_block **sbp;
418 __u64 devsize, newnsegs;
423 devsize = i_size_read(sb->s_bdev->bd_inode);
424 if (newsize > devsize)
428 * Prevent underflow in second superblock position calculation.
429 * The exact minimum size check is done in nilfs_sufile_resize().
431 if (newsize < 4096) {
437 * Write lock is required to protect some functions depending
438 * on the number of segments, the number of reserved segments,
441 down_write(&nilfs->ns_segctor_sem);
443 sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
444 newnsegs = sb2off >> nilfs->ns_blocksize_bits;
445 do_div(newnsegs, nilfs->ns_blocks_per_segment);
447 ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
448 up_write(&nilfs->ns_segctor_sem);
452 ret = nilfs_construct_segment(sb);
456 down_write(&nilfs->ns_sem);
457 nilfs_move_2nd_super(sb, sb2off);
459 sbp = nilfs_prepare_super(sb, 0);
461 nilfs_set_log_cursor(sbp[0], nilfs);
463 * Drop NILFS_RESIZE_FS flag for compatibility with
464 * mount-time resize which may be implemented in a
467 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
469 sbp[0]->s_dev_size = cpu_to_le64(newsize);
470 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
472 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
473 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
475 up_write(&nilfs->ns_sem);
478 * Reset the range of allocatable segments last. This order
479 * is important in the case of expansion because the secondary
480 * superblock must be protected from log write until migration
484 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
489 static void nilfs_put_super(struct super_block *sb)
491 struct the_nilfs *nilfs = sb->s_fs_info;
493 nilfs_detach_log_writer(sb);
495 if (!sb_rdonly(sb)) {
496 down_write(&nilfs->ns_sem);
497 nilfs_cleanup_super(sb);
498 up_write(&nilfs->ns_sem);
501 nilfs_sysfs_delete_device_group(nilfs);
502 iput(nilfs->ns_sufile);
503 iput(nilfs->ns_cpfile);
506 destroy_nilfs(nilfs);
507 sb->s_fs_info = NULL;
510 static int nilfs_sync_fs(struct super_block *sb, int wait)
512 struct the_nilfs *nilfs = sb->s_fs_info;
513 struct nilfs_super_block **sbp;
516 /* This function is called when super block should be written back */
518 err = nilfs_construct_segment(sb);
520 down_write(&nilfs->ns_sem);
521 if (nilfs_sb_dirty(nilfs)) {
522 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
524 nilfs_set_log_cursor(sbp[0], nilfs);
525 nilfs_commit_super(sb, NILFS_SB_COMMIT);
528 up_write(&nilfs->ns_sem);
531 err = nilfs_flush_device(nilfs);
536 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
537 struct nilfs_root **rootp)
539 struct the_nilfs *nilfs = sb->s_fs_info;
540 struct nilfs_root *root;
541 struct nilfs_checkpoint *raw_cp;
542 struct buffer_head *bh_cp;
545 root = nilfs_find_or_create_root(
546 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
551 goto reuse; /* already attached checkpoint */
553 down_read(&nilfs->ns_segctor_sem);
554 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
556 up_read(&nilfs->ns_segctor_sem);
558 if (err == -ENOENT || err == -EINVAL) {
559 nilfs_msg(sb, KERN_ERR,
560 "Invalid checkpoint (checkpoint number=%llu)",
561 (unsigned long long)cno);
567 err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
568 &raw_cp->cp_ifile_inode, &root->ifile);
572 atomic64_set(&root->inodes_count,
573 le64_to_cpu(raw_cp->cp_inodes_count));
574 atomic64_set(&root->blocks_count,
575 le64_to_cpu(raw_cp->cp_blocks_count));
577 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
584 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
586 nilfs_put_root(root);
591 static int nilfs_freeze(struct super_block *sb)
593 struct the_nilfs *nilfs = sb->s_fs_info;
599 /* Mark super block clean */
600 down_write(&nilfs->ns_sem);
601 err = nilfs_cleanup_super(sb);
602 up_write(&nilfs->ns_sem);
606 static int nilfs_unfreeze(struct super_block *sb)
608 struct the_nilfs *nilfs = sb->s_fs_info;
613 down_write(&nilfs->ns_sem);
614 nilfs_setup_super(sb, false);
615 up_write(&nilfs->ns_sem);
619 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
621 struct super_block *sb = dentry->d_sb;
622 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
623 struct the_nilfs *nilfs = root->nilfs;
624 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
625 unsigned long long blocks;
626 unsigned long overhead;
627 unsigned long nrsvblocks;
628 sector_t nfreeblocks;
629 u64 nmaxinodes, nfreeinodes;
633 * Compute all of the segment blocks
635 * The blocks before first segment and after last segment
638 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
639 - nilfs->ns_first_data_block;
640 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
643 * Compute the overhead
645 * When distributing meta data blocks outside segment structure,
646 * We must count them as the overhead.
650 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
654 err = nilfs_ifile_count_free_inodes(root->ifile,
655 &nmaxinodes, &nfreeinodes);
657 nilfs_msg(sb, KERN_WARNING,
658 "failed to count free inodes: err=%d", err);
659 if (err == -ERANGE) {
661 * If nilfs_palloc_count_max_entries() returns
662 * -ERANGE error code then we simply treat
663 * curent inodes count as maximum possible and
664 * zero as free inodes value.
666 nmaxinodes = atomic64_read(&root->inodes_count);
673 buf->f_type = NILFS_SUPER_MAGIC;
674 buf->f_bsize = sb->s_blocksize;
675 buf->f_blocks = blocks - overhead;
676 buf->f_bfree = nfreeblocks;
677 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
678 (buf->f_bfree - nrsvblocks) : 0;
679 buf->f_files = nmaxinodes;
680 buf->f_ffree = nfreeinodes;
681 buf->f_namelen = NILFS_NAME_LEN;
682 buf->f_fsid.val[0] = (u32)id;
683 buf->f_fsid.val[1] = (u32)(id >> 32);
688 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
690 struct super_block *sb = dentry->d_sb;
691 struct the_nilfs *nilfs = sb->s_fs_info;
692 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
694 if (!nilfs_test_opt(nilfs, BARRIER))
695 seq_puts(seq, ",nobarrier");
696 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
697 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
698 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
699 seq_puts(seq, ",errors=panic");
700 if (nilfs_test_opt(nilfs, ERRORS_CONT))
701 seq_puts(seq, ",errors=continue");
702 if (nilfs_test_opt(nilfs, STRICT_ORDER))
703 seq_puts(seq, ",order=strict");
704 if (nilfs_test_opt(nilfs, NORECOVERY))
705 seq_puts(seq, ",norecovery");
706 if (nilfs_test_opt(nilfs, DISCARD))
707 seq_puts(seq, ",discard");
712 static const struct super_operations nilfs_sops = {
713 .alloc_inode = nilfs_alloc_inode,
714 .free_inode = nilfs_free_inode,
715 .dirty_inode = nilfs_dirty_inode,
716 .evict_inode = nilfs_evict_inode,
717 .put_super = nilfs_put_super,
718 .sync_fs = nilfs_sync_fs,
719 .freeze_fs = nilfs_freeze,
720 .unfreeze_fs = nilfs_unfreeze,
721 .statfs = nilfs_statfs,
722 .remount_fs = nilfs_remount,
723 .show_options = nilfs_show_options
727 Opt_err_cont, Opt_err_panic, Opt_err_ro,
728 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
729 Opt_discard, Opt_nodiscard, Opt_err,
732 static match_table_t tokens = {
733 {Opt_err_cont, "errors=continue"},
734 {Opt_err_panic, "errors=panic"},
735 {Opt_err_ro, "errors=remount-ro"},
736 {Opt_barrier, "barrier"},
737 {Opt_nobarrier, "nobarrier"},
738 {Opt_snapshot, "cp=%u"},
739 {Opt_order, "order=%s"},
740 {Opt_norecovery, "norecovery"},
741 {Opt_discard, "discard"},
742 {Opt_nodiscard, "nodiscard"},
746 static int parse_options(char *options, struct super_block *sb, int is_remount)
748 struct the_nilfs *nilfs = sb->s_fs_info;
750 substring_t args[MAX_OPT_ARGS];
755 while ((p = strsep(&options, ",")) != NULL) {
761 token = match_token(p, tokens, args);
764 nilfs_set_opt(nilfs, BARRIER);
767 nilfs_clear_opt(nilfs, BARRIER);
770 if (strcmp(args[0].from, "relaxed") == 0)
771 /* Ordered data semantics */
772 nilfs_clear_opt(nilfs, STRICT_ORDER);
773 else if (strcmp(args[0].from, "strict") == 0)
774 /* Strict in-order semantics */
775 nilfs_set_opt(nilfs, STRICT_ORDER);
780 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
783 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
786 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
790 nilfs_msg(sb, KERN_ERR,
791 "\"%s\" option is invalid for remount",
797 nilfs_set_opt(nilfs, NORECOVERY);
800 nilfs_set_opt(nilfs, DISCARD);
803 nilfs_clear_opt(nilfs, DISCARD);
806 nilfs_msg(sb, KERN_ERR,
807 "unrecognized mount option \"%s\"", p);
815 nilfs_set_default_options(struct super_block *sb,
816 struct nilfs_super_block *sbp)
818 struct the_nilfs *nilfs = sb->s_fs_info;
820 nilfs->ns_mount_opt =
821 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
824 static int nilfs_setup_super(struct super_block *sb, int is_mount)
826 struct the_nilfs *nilfs = sb->s_fs_info;
827 struct nilfs_super_block **sbp;
831 /* nilfs->ns_sem must be locked by the caller. */
832 sbp = nilfs_prepare_super(sb, 0);
837 goto skip_mount_setup;
839 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
840 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
842 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
843 nilfs_msg(sb, KERN_WARNING, "mounting fs with errors");
845 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
846 nilfs_msg(sb, KERN_WARNING, "maximal mount count reached");
850 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
852 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
853 sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds());
857 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
858 /* synchronize sbp[1] with sbp[0] */
860 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
861 return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
864 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
865 u64 pos, int blocksize,
866 struct buffer_head **pbh)
868 unsigned long long sb_index = pos;
869 unsigned long offset;
871 offset = do_div(sb_index, blocksize);
872 *pbh = sb_bread(sb, sb_index);
875 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
878 int nilfs_store_magic_and_option(struct super_block *sb,
879 struct nilfs_super_block *sbp,
882 struct the_nilfs *nilfs = sb->s_fs_info;
884 sb->s_magic = le16_to_cpu(sbp->s_magic);
886 /* FS independent flags */
887 #ifdef NILFS_ATIME_DISABLE
888 sb->s_flags |= SB_NOATIME;
891 nilfs_set_default_options(sb, sbp);
893 nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
894 nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
895 nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
896 nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
898 return !parse_options(data, sb, 0) ? -EINVAL : 0;
901 int nilfs_check_feature_compatibility(struct super_block *sb,
902 struct nilfs_super_block *sbp)
906 features = le64_to_cpu(sbp->s_feature_incompat) &
907 ~NILFS_FEATURE_INCOMPAT_SUPP;
909 nilfs_msg(sb, KERN_ERR,
910 "couldn't mount because of unsupported optional features (%llx)",
911 (unsigned long long)features);
914 features = le64_to_cpu(sbp->s_feature_compat_ro) &
915 ~NILFS_FEATURE_COMPAT_RO_SUPP;
916 if (!sb_rdonly(sb) && features) {
917 nilfs_msg(sb, KERN_ERR,
918 "couldn't mount RDWR because of unsupported optional features (%llx)",
919 (unsigned long long)features);
925 static int nilfs_get_root_dentry(struct super_block *sb,
926 struct nilfs_root *root,
927 struct dentry **root_dentry)
930 struct dentry *dentry;
933 inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
935 ret = PTR_ERR(inode);
936 nilfs_msg(sb, KERN_ERR, "error %d getting root inode", ret);
939 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
941 nilfs_msg(sb, KERN_ERR, "corrupt root inode");
946 if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
947 dentry = d_find_alias(inode);
949 dentry = d_make_root(inode);
958 dentry = d_obtain_root(inode);
959 if (IS_ERR(dentry)) {
960 ret = PTR_ERR(dentry);
964 *root_dentry = dentry;
969 nilfs_msg(sb, KERN_ERR, "error %d getting root dentry", ret);
973 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
974 struct dentry **root_dentry)
976 struct the_nilfs *nilfs = s->s_fs_info;
977 struct nilfs_root *root;
980 mutex_lock(&nilfs->ns_snapshot_mount_mutex);
982 down_read(&nilfs->ns_segctor_sem);
983 ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
984 up_read(&nilfs->ns_segctor_sem);
986 ret = (ret == -ENOENT) ? -EINVAL : ret;
989 nilfs_msg(s, KERN_ERR,
990 "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
991 (unsigned long long)cno);
996 ret = nilfs_attach_checkpoint(s, cno, false, &root);
998 nilfs_msg(s, KERN_ERR,
999 "error %d while loading snapshot (checkpoint number=%llu)",
1000 ret, (unsigned long long)cno);
1003 ret = nilfs_get_root_dentry(s, root, root_dentry);
1004 nilfs_put_root(root);
1006 mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
1011 * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
1012 * @root_dentry: root dentry of the tree to be shrunk
1014 * This function returns true if the tree was in-use.
1016 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
1018 shrink_dcache_parent(root_dentry);
1019 return d_count(root_dentry) > 1;
1022 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1024 struct the_nilfs *nilfs = sb->s_fs_info;
1025 struct nilfs_root *root;
1026 struct inode *inode;
1027 struct dentry *dentry;
1030 if (cno > nilfs->ns_cno)
1033 if (cno >= nilfs_last_cno(nilfs))
1034 return true; /* protect recent checkpoints */
1037 root = nilfs_lookup_root(nilfs, cno);
1039 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1041 dentry = d_find_alias(inode);
1043 ret = nilfs_tree_is_busy(dentry);
1048 nilfs_put_root(root);
1054 * nilfs_fill_super() - initialize a super block instance
1056 * @data: mount options
1057 * @silent: silent mode flag
1059 * This function is called exclusively by nilfs->ns_mount_mutex.
1060 * So, the recovery process is protected from other simultaneous mounts.
1063 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1065 struct the_nilfs *nilfs;
1066 struct nilfs_root *fsroot;
1070 nilfs = alloc_nilfs(sb);
1074 sb->s_fs_info = nilfs;
1076 err = init_nilfs(nilfs, sb, (char *)data);
1080 sb->s_op = &nilfs_sops;
1081 sb->s_export_op = &nilfs_export_ops;
1083 sb->s_time_gran = 1;
1084 sb->s_max_links = NILFS_LINK_MAX;
1086 sb->s_bdi = bdi_get(sb->s_bdev->bd_bdi);
1088 err = load_nilfs(nilfs, sb);
1092 cno = nilfs_last_cno(nilfs);
1093 err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1095 nilfs_msg(sb, KERN_ERR,
1096 "error %d while loading last checkpoint (checkpoint number=%llu)",
1097 err, (unsigned long long)cno);
1101 if (!sb_rdonly(sb)) {
1102 err = nilfs_attach_log_writer(sb, fsroot);
1104 goto failed_checkpoint;
1107 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1109 goto failed_segctor;
1111 nilfs_put_root(fsroot);
1113 if (!sb_rdonly(sb)) {
1114 down_write(&nilfs->ns_sem);
1115 nilfs_setup_super(sb, true);
1116 up_write(&nilfs->ns_sem);
1122 nilfs_detach_log_writer(sb);
1125 nilfs_put_root(fsroot);
1128 nilfs_sysfs_delete_device_group(nilfs);
1129 iput(nilfs->ns_sufile);
1130 iput(nilfs->ns_cpfile);
1131 iput(nilfs->ns_dat);
1134 destroy_nilfs(nilfs);
1138 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1140 struct the_nilfs *nilfs = sb->s_fs_info;
1141 unsigned long old_sb_flags;
1142 unsigned long old_mount_opt;
1145 sync_filesystem(sb);
1146 old_sb_flags = sb->s_flags;
1147 old_mount_opt = nilfs->ns_mount_opt;
1149 if (!parse_options(data, sb, 1)) {
1153 sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1157 if (!nilfs_valid_fs(nilfs)) {
1158 nilfs_msg(sb, KERN_WARNING,
1159 "couldn't remount because the filesystem is in an incomplete recovery state");
1163 if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1165 if (*flags & SB_RDONLY) {
1166 sb->s_flags |= SB_RDONLY;
1169 * Remounting a valid RW partition RDONLY, so set
1170 * the RDONLY flag and then mark the partition as valid again.
1172 down_write(&nilfs->ns_sem);
1173 nilfs_cleanup_super(sb);
1174 up_write(&nilfs->ns_sem);
1177 struct nilfs_root *root;
1180 * Mounting a RDONLY partition read-write, so reread and
1181 * store the current valid flag. (It may have been changed
1182 * by fsck since we originally mounted the partition.)
1184 down_read(&nilfs->ns_sem);
1185 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1186 ~NILFS_FEATURE_COMPAT_RO_SUPP;
1187 up_read(&nilfs->ns_sem);
1189 nilfs_msg(sb, KERN_WARNING,
1190 "couldn't remount RDWR because of unsupported optional features (%llx)",
1191 (unsigned long long)features);
1196 sb->s_flags &= ~SB_RDONLY;
1198 root = NILFS_I(d_inode(sb->s_root))->i_root;
1199 err = nilfs_attach_log_writer(sb, root);
1203 down_write(&nilfs->ns_sem);
1204 nilfs_setup_super(sb, true);
1205 up_write(&nilfs->ns_sem);
1211 sb->s_flags = old_sb_flags;
1212 nilfs->ns_mount_opt = old_mount_opt;
1216 struct nilfs_super_data {
1217 struct block_device *bdev;
1222 static int nilfs_parse_snapshot_option(const char *option,
1223 const substring_t *arg,
1224 struct nilfs_super_data *sd)
1226 unsigned long long val;
1227 const char *msg = NULL;
1230 if (!(sd->flags & SB_RDONLY)) {
1231 msg = "read-only option is not specified";
1235 err = kstrtoull(arg->from, 0, &val);
1238 msg = "too large checkpoint number";
1240 msg = "malformed argument";
1242 } else if (val == 0) {
1243 msg = "invalid checkpoint number 0";
1250 nilfs_msg(NULL, KERN_ERR, "invalid option \"%s\": %s", option, msg);
1255 * nilfs_identify - pre-read mount options needed to identify mount instance
1256 * @data: mount options
1257 * @sd: nilfs_super_data
1259 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1261 char *p, *options = data;
1262 substring_t args[MAX_OPT_ARGS];
1267 p = strsep(&options, ",");
1268 if (p != NULL && *p) {
1269 token = match_token(p, tokens, args);
1270 if (token == Opt_snapshot)
1271 ret = nilfs_parse_snapshot_option(p, &args[0],
1276 BUG_ON(options == data);
1277 *(options - 1) = ',';
1282 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1285 s->s_dev = s->s_bdev->bd_dev;
1289 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1291 return (void *)s->s_bdev == data;
1294 static struct dentry *
1295 nilfs_mount(struct file_system_type *fs_type, int flags,
1296 const char *dev_name, void *data)
1298 struct nilfs_super_data sd;
1299 struct super_block *s;
1300 fmode_t mode = FMODE_READ | FMODE_EXCL;
1301 struct dentry *root_dentry;
1302 int err, s_new = false;
1304 if (!(flags & SB_RDONLY))
1305 mode |= FMODE_WRITE;
1307 sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1308 if (IS_ERR(sd.bdev))
1309 return ERR_CAST(sd.bdev);
1313 if (nilfs_identify((char *)data, &sd)) {
1319 * once the super is inserted into the list by sget, s_umount
1320 * will protect the lockfs code from trying to start a snapshot
1321 * while we are mounting
1323 mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1324 if (sd.bdev->bd_fsfreeze_count > 0) {
1325 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1329 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1331 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1340 /* New superblock instance created */
1342 snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1343 sb_set_blocksize(s, block_size(sd.bdev));
1345 err = nilfs_fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1349 s->s_flags |= SB_ACTIVE;
1350 } else if (!sd.cno) {
1351 if (nilfs_tree_is_busy(s->s_root)) {
1352 if ((flags ^ s->s_flags) & SB_RDONLY) {
1353 nilfs_msg(s, KERN_ERR,
1354 "the device already has a %s mount.",
1355 sb_rdonly(s) ? "read-only" : "read/write");
1361 * Try remount to setup mount states if the current
1362 * tree is not mounted and only snapshots use this sb.
1364 err = nilfs_remount(s, &flags, data);
1371 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1375 root_dentry = dget(s->s_root);
1379 blkdev_put(sd.bdev, mode);
1384 deactivate_locked_super(s);
1388 blkdev_put(sd.bdev, mode);
1389 return ERR_PTR(err);
1392 struct file_system_type nilfs_fs_type = {
1393 .owner = THIS_MODULE,
1395 .mount = nilfs_mount,
1396 .kill_sb = kill_block_super,
1397 .fs_flags = FS_REQUIRES_DEV,
1399 MODULE_ALIAS_FS("nilfs2");
1401 static void nilfs_inode_init_once(void *obj)
1403 struct nilfs_inode_info *ii = obj;
1405 INIT_LIST_HEAD(&ii->i_dirty);
1406 #ifdef CONFIG_NILFS_XATTR
1407 init_rwsem(&ii->xattr_sem);
1409 inode_init_once(&ii->vfs_inode);
1412 static void nilfs_segbuf_init_once(void *obj)
1414 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1417 static void nilfs_destroy_cachep(void)
1420 * Make sure all delayed rcu free inodes are flushed before we
1425 kmem_cache_destroy(nilfs_inode_cachep);
1426 kmem_cache_destroy(nilfs_transaction_cachep);
1427 kmem_cache_destroy(nilfs_segbuf_cachep);
1428 kmem_cache_destroy(nilfs_btree_path_cache);
1431 static int __init nilfs_init_cachep(void)
1433 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1434 sizeof(struct nilfs_inode_info), 0,
1435 SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1436 nilfs_inode_init_once);
1437 if (!nilfs_inode_cachep)
1440 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1441 sizeof(struct nilfs_transaction_info), 0,
1442 SLAB_RECLAIM_ACCOUNT, NULL);
1443 if (!nilfs_transaction_cachep)
1446 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1447 sizeof(struct nilfs_segment_buffer), 0,
1448 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1449 if (!nilfs_segbuf_cachep)
1452 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1453 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1455 if (!nilfs_btree_path_cache)
1461 nilfs_destroy_cachep();
1465 static int __init init_nilfs_fs(void)
1469 err = nilfs_init_cachep();
1473 err = nilfs_sysfs_init();
1477 err = register_filesystem(&nilfs_fs_type);
1479 goto deinit_sysfs_entry;
1481 printk(KERN_INFO "NILFS version 2 loaded\n");
1487 nilfs_destroy_cachep();
1492 static void __exit exit_nilfs_fs(void)
1494 nilfs_destroy_cachep();
1496 unregister_filesystem(&nilfs_fs_type);
1499 module_init(init_nilfs_fs)
1500 module_exit(exit_nilfs_fs)