1 // SPDX-License-Identifier: GPL-2.0+
3 * 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>
38 #include <linux/fs_context.h>
47 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
53 MODULE_AUTHOR("NTT Corp.");
54 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
56 MODULE_LICENSE("GPL");
58 static struct kmem_cache *nilfs_inode_cachep;
59 struct kmem_cache *nilfs_transaction_cachep;
60 struct kmem_cache *nilfs_segbuf_cachep;
61 struct kmem_cache *nilfs_btree_path_cache;
63 static int nilfs_setup_super(struct super_block *sb, int is_mount);
64 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
66 void __nilfs_msg(struct super_block *sb, const char *fmt, ...)
74 level = printk_get_level(fmt);
75 vaf.fmt = printk_skip_level(fmt);
79 printk("%c%cNILFS (%s): %pV\n",
80 KERN_SOH_ASCII, level, sb->s_id, &vaf);
82 printk("%c%cNILFS: %pV\n",
83 KERN_SOH_ASCII, level, &vaf);
88 static void nilfs_set_error(struct super_block *sb)
90 struct the_nilfs *nilfs = sb->s_fs_info;
91 struct nilfs_super_block **sbp;
93 down_write(&nilfs->ns_sem);
94 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
95 nilfs->ns_mount_state |= NILFS_ERROR_FS;
96 sbp = nilfs_prepare_super(sb, 0);
98 sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
100 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
101 nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
104 up_write(&nilfs->ns_sem);
108 * __nilfs_error() - report failure condition on a filesystem
110 * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
111 * reporting an error message. This function should be called when
112 * NILFS detects incoherences or defects of meta data on disk.
114 * This implements the body of nilfs_error() macro. Normally,
115 * nilfs_error() should be used. As for sustainable errors such as a
116 * single-shot I/O error, nilfs_err() should be used instead.
118 * Callers should not add a trailing newline since this will do it.
120 void __nilfs_error(struct super_block *sb, const char *function,
121 const char *fmt, ...)
123 struct the_nilfs *nilfs = sb->s_fs_info;
124 struct va_format vaf;
132 printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
133 sb->s_id, function, &vaf);
137 if (!sb_rdonly(sb)) {
140 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
141 printk(KERN_CRIT "Remounting filesystem read-only\n");
142 sb->s_flags |= SB_RDONLY;
146 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
147 panic("NILFS (device %s): panic forced after error\n",
151 struct inode *nilfs_alloc_inode(struct super_block *sb)
153 struct nilfs_inode_info *ii;
155 ii = alloc_inode_sb(sb, nilfs_inode_cachep, GFP_NOFS);
161 ii->i_assoc_inode = NULL;
162 ii->i_bmap = &ii->i_bmap_data;
163 return &ii->vfs_inode;
166 static void nilfs_free_inode(struct inode *inode)
168 if (nilfs_is_metadata_file_inode(inode))
169 nilfs_mdt_destroy(inode);
171 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
174 static int nilfs_sync_super(struct super_block *sb, int flag)
176 struct the_nilfs *nilfs = sb->s_fs_info;
180 set_buffer_dirty(nilfs->ns_sbh[0]);
181 if (nilfs_test_opt(nilfs, BARRIER)) {
182 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
183 REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
185 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
189 nilfs_err(sb, "unable to write superblock: err=%d", err);
190 if (err == -EIO && nilfs->ns_sbh[1]) {
192 * sbp[0] points to newer log than sbp[1],
193 * so copy sbp[0] to sbp[1] to take over sbp[0].
195 memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
197 nilfs_fall_back_super_block(nilfs);
201 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
203 nilfs->ns_sbwcount++;
206 * The latest segment becomes trailable from the position
207 * written in superblock.
209 clear_nilfs_discontinued(nilfs);
211 /* update GC protection for recent segments */
212 if (nilfs->ns_sbh[1]) {
213 if (flag == NILFS_SB_COMMIT_ALL) {
214 set_buffer_dirty(nilfs->ns_sbh[1]);
215 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
218 if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
219 le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
220 sbp = nilfs->ns_sbp[1];
223 spin_lock(&nilfs->ns_last_segment_lock);
224 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
225 spin_unlock(&nilfs->ns_last_segment_lock);
231 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
232 struct the_nilfs *nilfs)
234 sector_t nfreeblocks;
236 /* nilfs->ns_sem must be locked by the caller. */
237 nilfs_count_free_blocks(nilfs, &nfreeblocks);
238 sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
240 spin_lock(&nilfs->ns_last_segment_lock);
241 sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
242 sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
243 sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
244 spin_unlock(&nilfs->ns_last_segment_lock);
247 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
250 struct the_nilfs *nilfs = sb->s_fs_info;
251 struct nilfs_super_block **sbp = nilfs->ns_sbp;
253 /* nilfs->ns_sem must be locked by the caller. */
254 if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
256 sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
257 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
259 nilfs_crit(sb, "superblock broke");
263 sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
264 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
268 nilfs_swap_super_block(nilfs);
273 int nilfs_commit_super(struct super_block *sb, int flag)
275 struct the_nilfs *nilfs = sb->s_fs_info;
276 struct nilfs_super_block **sbp = nilfs->ns_sbp;
279 /* nilfs->ns_sem must be locked by the caller. */
280 t = ktime_get_real_seconds();
281 nilfs->ns_sbwtime = t;
282 sbp[0]->s_wtime = cpu_to_le64(t);
284 sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
285 (unsigned char *)sbp[0],
287 if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
288 sbp[1]->s_wtime = sbp[0]->s_wtime;
290 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
291 (unsigned char *)sbp[1],
294 clear_nilfs_sb_dirty(nilfs);
295 nilfs->ns_flushed_device = 1;
296 /* make sure store to ns_flushed_device cannot be reordered */
298 return nilfs_sync_super(sb, flag);
302 * nilfs_cleanup_super() - write filesystem state for cleanup
303 * @sb: super block instance to be unmounted or degraded to read-only
305 * This function restores state flags in the on-disk super block.
306 * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
307 * filesystem was not clean previously.
309 int nilfs_cleanup_super(struct super_block *sb)
311 struct the_nilfs *nilfs = sb->s_fs_info;
312 struct nilfs_super_block **sbp;
313 int flag = NILFS_SB_COMMIT;
316 sbp = nilfs_prepare_super(sb, 0);
318 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
319 nilfs_set_log_cursor(sbp[0], nilfs);
320 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
322 * make the "clean" flag also to the opposite
323 * super block if both super blocks point to
324 * the same checkpoint.
326 sbp[1]->s_state = sbp[0]->s_state;
327 flag = NILFS_SB_COMMIT_ALL;
329 ret = nilfs_commit_super(sb, flag);
335 * nilfs_move_2nd_super - relocate secondary super block
336 * @sb: super block instance
337 * @sb2off: new offset of the secondary super block (in bytes)
339 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
341 struct the_nilfs *nilfs = sb->s_fs_info;
342 struct buffer_head *nsbh;
343 struct nilfs_super_block *nsbp;
344 sector_t blocknr, newblocknr;
345 unsigned long offset;
346 int sb2i; /* array index of the secondary superblock */
349 /* nilfs->ns_sem must be locked by the caller. */
350 if (nilfs->ns_sbh[1] &&
351 nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
353 blocknr = nilfs->ns_sbh[1]->b_blocknr;
354 } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
356 blocknr = nilfs->ns_sbh[0]->b_blocknr;
361 if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
362 goto out; /* super block location is unchanged */
364 /* Get new super block buffer */
365 newblocknr = sb2off >> nilfs->ns_blocksize_bits;
366 offset = sb2off & (nilfs->ns_blocksize - 1);
367 nsbh = sb_getblk(sb, newblocknr);
370 "unable to move secondary superblock to block %llu",
371 (unsigned long long)newblocknr);
375 nsbp = (void *)nsbh->b_data + offset;
380 * The position of the second superblock only changes by 4KiB,
381 * which is larger than the maximum superblock data size
382 * (= 1KiB), so there is no need to use memmove() to allow
383 * overlap between source and destination.
385 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
388 * Zero fill after copy to avoid overwriting in case of move
389 * within the same block.
391 memset(nsbh->b_data, 0, offset);
392 memset((void *)nsbp + nilfs->ns_sbsize, 0,
393 nsbh->b_size - offset - nilfs->ns_sbsize);
395 memset(nsbh->b_data, 0, nsbh->b_size);
397 set_buffer_uptodate(nsbh);
401 brelse(nilfs->ns_sbh[sb2i]);
402 nilfs->ns_sbh[sb2i] = nsbh;
403 nilfs->ns_sbp[sb2i] = nsbp;
404 } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
405 /* secondary super block will be restored to index 1 */
406 nilfs->ns_sbh[1] = nsbh;
407 nilfs->ns_sbp[1] = nsbp;
416 * nilfs_resize_fs - resize the filesystem
417 * @sb: super block instance
418 * @newsize: new size of the filesystem (in bytes)
420 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
422 struct the_nilfs *nilfs = sb->s_fs_info;
423 struct nilfs_super_block **sbp;
424 __u64 devsize, newnsegs;
429 devsize = bdev_nr_bytes(sb->s_bdev);
430 if (newsize > devsize)
434 * Prevent underflow in second superblock position calculation.
435 * The exact minimum size check is done in nilfs_sufile_resize().
437 if (newsize < 4096) {
443 * Write lock is required to protect some functions depending
444 * on the number of segments, the number of reserved segments,
447 down_write(&nilfs->ns_segctor_sem);
449 sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
450 newnsegs = sb2off >> nilfs->ns_blocksize_bits;
451 newnsegs = div64_ul(newnsegs, nilfs->ns_blocks_per_segment);
453 ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
454 up_write(&nilfs->ns_segctor_sem);
458 ret = nilfs_construct_segment(sb);
462 down_write(&nilfs->ns_sem);
463 nilfs_move_2nd_super(sb, sb2off);
465 sbp = nilfs_prepare_super(sb, 0);
467 nilfs_set_log_cursor(sbp[0], nilfs);
469 * Drop NILFS_RESIZE_FS flag for compatibility with
470 * mount-time resize which may be implemented in a
473 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
475 sbp[0]->s_dev_size = cpu_to_le64(newsize);
476 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
478 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
479 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
481 up_write(&nilfs->ns_sem);
484 * Reset the range of allocatable segments last. This order
485 * is important in the case of expansion because the secondary
486 * superblock must be protected from log write until migration
490 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
495 static void nilfs_put_super(struct super_block *sb)
497 struct the_nilfs *nilfs = sb->s_fs_info;
499 nilfs_detach_log_writer(sb);
501 if (!sb_rdonly(sb)) {
502 down_write(&nilfs->ns_sem);
503 nilfs_cleanup_super(sb);
504 up_write(&nilfs->ns_sem);
507 nilfs_sysfs_delete_device_group(nilfs);
508 iput(nilfs->ns_sufile);
509 iput(nilfs->ns_cpfile);
512 destroy_nilfs(nilfs);
513 sb->s_fs_info = NULL;
516 static int nilfs_sync_fs(struct super_block *sb, int wait)
518 struct the_nilfs *nilfs = sb->s_fs_info;
519 struct nilfs_super_block **sbp;
522 /* This function is called when super block should be written back */
524 err = nilfs_construct_segment(sb);
526 down_write(&nilfs->ns_sem);
527 if (nilfs_sb_dirty(nilfs)) {
528 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
530 nilfs_set_log_cursor(sbp[0], nilfs);
531 nilfs_commit_super(sb, NILFS_SB_COMMIT);
534 up_write(&nilfs->ns_sem);
537 err = nilfs_flush_device(nilfs);
542 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
543 struct nilfs_root **rootp)
545 struct the_nilfs *nilfs = sb->s_fs_info;
546 struct nilfs_root *root;
549 root = nilfs_find_or_create_root(
550 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
555 goto reuse; /* already attached checkpoint */
557 down_read(&nilfs->ns_segctor_sem);
558 err = nilfs_ifile_read(sb, root, cno, nilfs->ns_inode_size);
559 up_read(&nilfs->ns_segctor_sem);
569 nilfs_err(sb, "Invalid checkpoint (checkpoint number=%llu)",
570 (unsigned long long)cno);
571 nilfs_put_root(root);
576 static int nilfs_freeze(struct super_block *sb)
578 struct the_nilfs *nilfs = sb->s_fs_info;
584 /* Mark super block clean */
585 down_write(&nilfs->ns_sem);
586 err = nilfs_cleanup_super(sb);
587 up_write(&nilfs->ns_sem);
591 static int nilfs_unfreeze(struct super_block *sb)
593 struct the_nilfs *nilfs = sb->s_fs_info;
598 down_write(&nilfs->ns_sem);
599 nilfs_setup_super(sb, false);
600 up_write(&nilfs->ns_sem);
604 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
606 struct super_block *sb = dentry->d_sb;
607 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
608 struct the_nilfs *nilfs = root->nilfs;
609 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
610 unsigned long long blocks;
611 unsigned long overhead;
612 unsigned long nrsvblocks;
613 sector_t nfreeblocks;
614 u64 nmaxinodes, nfreeinodes;
618 * Compute all of the segment blocks
620 * The blocks before first segment and after last segment
623 blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
624 - nilfs->ns_first_data_block;
625 nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
628 * Compute the overhead
630 * When distributing meta data blocks outside segment structure,
631 * We must count them as the overhead.
635 err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
639 err = nilfs_ifile_count_free_inodes(root->ifile,
640 &nmaxinodes, &nfreeinodes);
642 nilfs_warn(sb, "failed to count free inodes: err=%d", err);
643 if (err == -ERANGE) {
645 * If nilfs_palloc_count_max_entries() returns
646 * -ERANGE error code then we simply treat
647 * curent inodes count as maximum possible and
648 * zero as free inodes value.
650 nmaxinodes = atomic64_read(&root->inodes_count);
657 buf->f_type = NILFS_SUPER_MAGIC;
658 buf->f_bsize = sb->s_blocksize;
659 buf->f_blocks = blocks - overhead;
660 buf->f_bfree = nfreeblocks;
661 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
662 (buf->f_bfree - nrsvblocks) : 0;
663 buf->f_files = nmaxinodes;
664 buf->f_ffree = nfreeinodes;
665 buf->f_namelen = NILFS_NAME_LEN;
666 buf->f_fsid = u64_to_fsid(id);
671 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
673 struct super_block *sb = dentry->d_sb;
674 struct the_nilfs *nilfs = sb->s_fs_info;
675 struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
677 if (!nilfs_test_opt(nilfs, BARRIER))
678 seq_puts(seq, ",nobarrier");
679 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
680 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
681 if (nilfs_test_opt(nilfs, ERRORS_PANIC))
682 seq_puts(seq, ",errors=panic");
683 if (nilfs_test_opt(nilfs, ERRORS_CONT))
684 seq_puts(seq, ",errors=continue");
685 if (nilfs_test_opt(nilfs, STRICT_ORDER))
686 seq_puts(seq, ",order=strict");
687 if (nilfs_test_opt(nilfs, NORECOVERY))
688 seq_puts(seq, ",norecovery");
689 if (nilfs_test_opt(nilfs, DISCARD))
690 seq_puts(seq, ",discard");
695 static const struct super_operations nilfs_sops = {
696 .alloc_inode = nilfs_alloc_inode,
697 .free_inode = nilfs_free_inode,
698 .dirty_inode = nilfs_dirty_inode,
699 .evict_inode = nilfs_evict_inode,
700 .put_super = nilfs_put_super,
701 .sync_fs = nilfs_sync_fs,
702 .freeze_fs = nilfs_freeze,
703 .unfreeze_fs = nilfs_unfreeze,
704 .statfs = nilfs_statfs,
705 .remount_fs = nilfs_remount,
706 .show_options = nilfs_show_options
710 Opt_err_cont, Opt_err_panic, Opt_err_ro,
711 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
712 Opt_discard, Opt_nodiscard, Opt_err,
715 static match_table_t tokens = {
716 {Opt_err_cont, "errors=continue"},
717 {Opt_err_panic, "errors=panic"},
718 {Opt_err_ro, "errors=remount-ro"},
719 {Opt_barrier, "barrier"},
720 {Opt_nobarrier, "nobarrier"},
721 {Opt_snapshot, "cp=%u"},
722 {Opt_order, "order=%s"},
723 {Opt_norecovery, "norecovery"},
724 {Opt_discard, "discard"},
725 {Opt_nodiscard, "nodiscard"},
729 static int parse_options(char *options, struct super_block *sb, int is_remount)
731 struct the_nilfs *nilfs = sb->s_fs_info;
733 substring_t args[MAX_OPT_ARGS];
738 while ((p = strsep(&options, ",")) != NULL) {
744 token = match_token(p, tokens, args);
747 nilfs_set_opt(nilfs, BARRIER);
750 nilfs_clear_opt(nilfs, BARRIER);
753 if (strcmp(args[0].from, "relaxed") == 0)
754 /* Ordered data semantics */
755 nilfs_clear_opt(nilfs, STRICT_ORDER);
756 else if (strcmp(args[0].from, "strict") == 0)
757 /* Strict in-order semantics */
758 nilfs_set_opt(nilfs, STRICT_ORDER);
763 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
766 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
769 nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
774 "\"%s\" option is invalid for remount",
780 nilfs_set_opt(nilfs, NORECOVERY);
783 nilfs_set_opt(nilfs, DISCARD);
786 nilfs_clear_opt(nilfs, DISCARD);
789 nilfs_err(sb, "unrecognized mount option \"%s\"", p);
797 nilfs_set_default_options(struct super_block *sb,
798 struct nilfs_super_block *sbp)
800 struct the_nilfs *nilfs = sb->s_fs_info;
802 nilfs->ns_mount_opt =
803 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
806 static int nilfs_setup_super(struct super_block *sb, int is_mount)
808 struct the_nilfs *nilfs = sb->s_fs_info;
809 struct nilfs_super_block **sbp;
813 /* nilfs->ns_sem must be locked by the caller. */
814 sbp = nilfs_prepare_super(sb, 0);
819 goto skip_mount_setup;
821 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
822 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
824 if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
825 nilfs_warn(sb, "mounting fs with errors");
827 } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
828 nilfs_warn(sb, "maximal mount count reached");
832 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
834 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
835 sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds());
839 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
840 /* synchronize sbp[1] with sbp[0] */
842 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
843 return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
846 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
847 u64 pos, int blocksize,
848 struct buffer_head **pbh)
850 unsigned long long sb_index = pos;
851 unsigned long offset;
853 offset = do_div(sb_index, blocksize);
854 *pbh = sb_bread(sb, sb_index);
857 return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
860 int nilfs_store_magic_and_option(struct super_block *sb,
861 struct nilfs_super_block *sbp,
864 struct the_nilfs *nilfs = sb->s_fs_info;
866 sb->s_magic = le16_to_cpu(sbp->s_magic);
868 /* FS independent flags */
869 #ifdef NILFS_ATIME_DISABLE
870 sb->s_flags |= SB_NOATIME;
873 nilfs_set_default_options(sb, sbp);
875 nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
876 nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
877 nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
878 nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
880 return !parse_options(data, sb, 0) ? -EINVAL : 0;
883 int nilfs_check_feature_compatibility(struct super_block *sb,
884 struct nilfs_super_block *sbp)
888 features = le64_to_cpu(sbp->s_feature_incompat) &
889 ~NILFS_FEATURE_INCOMPAT_SUPP;
892 "couldn't mount because of unsupported optional features (%llx)",
893 (unsigned long long)features);
896 features = le64_to_cpu(sbp->s_feature_compat_ro) &
897 ~NILFS_FEATURE_COMPAT_RO_SUPP;
898 if (!sb_rdonly(sb) && features) {
900 "couldn't mount RDWR because of unsupported optional features (%llx)",
901 (unsigned long long)features);
907 static int nilfs_get_root_dentry(struct super_block *sb,
908 struct nilfs_root *root,
909 struct dentry **root_dentry)
912 struct dentry *dentry;
915 inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
917 ret = PTR_ERR(inode);
918 nilfs_err(sb, "error %d getting root inode", ret);
921 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
923 nilfs_err(sb, "corrupt root inode");
928 if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
929 dentry = d_find_alias(inode);
931 dentry = d_make_root(inode);
940 dentry = d_obtain_root(inode);
941 if (IS_ERR(dentry)) {
942 ret = PTR_ERR(dentry);
946 *root_dentry = dentry;
951 nilfs_err(sb, "error %d getting root dentry", ret);
955 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
956 struct dentry **root_dentry)
958 struct the_nilfs *nilfs = s->s_fs_info;
959 struct nilfs_root *root;
962 mutex_lock(&nilfs->ns_snapshot_mount_mutex);
964 down_read(&nilfs->ns_segctor_sem);
965 ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
966 up_read(&nilfs->ns_segctor_sem);
968 ret = (ret == -ENOENT) ? -EINVAL : ret;
972 "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
973 (unsigned long long)cno);
978 ret = nilfs_attach_checkpoint(s, cno, false, &root);
981 "error %d while loading snapshot (checkpoint number=%llu)",
982 ret, (unsigned long long)cno);
985 ret = nilfs_get_root_dentry(s, root, root_dentry);
986 nilfs_put_root(root);
988 mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
993 * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
994 * @root_dentry: root dentry of the tree to be shrunk
996 * This function returns true if the tree was in-use.
998 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
1000 shrink_dcache_parent(root_dentry);
1001 return d_count(root_dentry) > 1;
1004 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1006 struct the_nilfs *nilfs = sb->s_fs_info;
1007 struct nilfs_root *root;
1008 struct inode *inode;
1009 struct dentry *dentry;
1012 if (cno > nilfs->ns_cno)
1015 if (cno >= nilfs_last_cno(nilfs))
1016 return true; /* protect recent checkpoints */
1019 root = nilfs_lookup_root(nilfs, cno);
1021 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1023 dentry = d_find_alias(inode);
1025 ret = nilfs_tree_is_busy(dentry);
1030 nilfs_put_root(root);
1036 * nilfs_fill_super() - initialize a super block instance
1038 * @data: mount options
1039 * @silent: silent mode flag
1041 * This function is called exclusively by nilfs->ns_mount_mutex.
1042 * So, the recovery process is protected from other simultaneous mounts.
1045 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1047 struct the_nilfs *nilfs;
1048 struct nilfs_root *fsroot;
1052 nilfs = alloc_nilfs(sb);
1056 sb->s_fs_info = nilfs;
1058 err = init_nilfs(nilfs, sb, (char *)data);
1062 sb->s_op = &nilfs_sops;
1063 sb->s_export_op = &nilfs_export_ops;
1065 sb->s_time_gran = 1;
1066 sb->s_max_links = NILFS_LINK_MAX;
1068 sb->s_bdi = bdi_get(sb->s_bdev->bd_disk->bdi);
1070 err = load_nilfs(nilfs, sb);
1074 cno = nilfs_last_cno(nilfs);
1075 err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1078 "error %d while loading last checkpoint (checkpoint number=%llu)",
1079 err, (unsigned long long)cno);
1083 if (!sb_rdonly(sb)) {
1084 err = nilfs_attach_log_writer(sb, fsroot);
1086 goto failed_checkpoint;
1089 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1091 goto failed_segctor;
1093 nilfs_put_root(fsroot);
1095 if (!sb_rdonly(sb)) {
1096 down_write(&nilfs->ns_sem);
1097 nilfs_setup_super(sb, true);
1098 up_write(&nilfs->ns_sem);
1104 nilfs_detach_log_writer(sb);
1107 nilfs_put_root(fsroot);
1110 nilfs_sysfs_delete_device_group(nilfs);
1111 iput(nilfs->ns_sufile);
1112 iput(nilfs->ns_cpfile);
1113 iput(nilfs->ns_dat);
1116 destroy_nilfs(nilfs);
1120 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1122 struct the_nilfs *nilfs = sb->s_fs_info;
1123 unsigned long old_sb_flags;
1124 unsigned long old_mount_opt;
1127 sync_filesystem(sb);
1128 old_sb_flags = sb->s_flags;
1129 old_mount_opt = nilfs->ns_mount_opt;
1131 if (!parse_options(data, sb, 1)) {
1135 sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1139 if (!nilfs_valid_fs(nilfs)) {
1141 "couldn't remount because the filesystem is in an incomplete recovery state");
1145 if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1147 if (*flags & SB_RDONLY) {
1148 sb->s_flags |= SB_RDONLY;
1151 * Remounting a valid RW partition RDONLY, so set
1152 * the RDONLY flag and then mark the partition as valid again.
1154 down_write(&nilfs->ns_sem);
1155 nilfs_cleanup_super(sb);
1156 up_write(&nilfs->ns_sem);
1159 struct nilfs_root *root;
1162 * Mounting a RDONLY partition read-write, so reread and
1163 * store the current valid flag. (It may have been changed
1164 * by fsck since we originally mounted the partition.)
1166 down_read(&nilfs->ns_sem);
1167 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1168 ~NILFS_FEATURE_COMPAT_RO_SUPP;
1169 up_read(&nilfs->ns_sem);
1172 "couldn't remount RDWR because of unsupported optional features (%llx)",
1173 (unsigned long long)features);
1178 sb->s_flags &= ~SB_RDONLY;
1180 root = NILFS_I(d_inode(sb->s_root))->i_root;
1181 err = nilfs_attach_log_writer(sb, root);
1185 down_write(&nilfs->ns_sem);
1186 nilfs_setup_super(sb, true);
1187 up_write(&nilfs->ns_sem);
1193 sb->s_flags = old_sb_flags;
1194 nilfs->ns_mount_opt = old_mount_opt;
1198 struct nilfs_super_data {
1203 static int nilfs_parse_snapshot_option(const char *option,
1204 const substring_t *arg,
1205 struct nilfs_super_data *sd)
1207 unsigned long long val;
1208 const char *msg = NULL;
1211 if (!(sd->flags & SB_RDONLY)) {
1212 msg = "read-only option is not specified";
1216 err = kstrtoull(arg->from, 0, &val);
1219 msg = "too large checkpoint number";
1221 msg = "malformed argument";
1223 } else if (val == 0) {
1224 msg = "invalid checkpoint number 0";
1231 nilfs_err(NULL, "invalid option \"%s\": %s", option, msg);
1236 * nilfs_identify - pre-read mount options needed to identify mount instance
1237 * @data: mount options
1238 * @sd: nilfs_super_data
1240 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1242 char *p, *options = data;
1243 substring_t args[MAX_OPT_ARGS];
1248 p = strsep(&options, ",");
1249 if (p != NULL && *p) {
1250 token = match_token(p, tokens, args);
1251 if (token == Opt_snapshot)
1252 ret = nilfs_parse_snapshot_option(p, &args[0],
1257 BUG_ON(options == data);
1258 *(options - 1) = ',';
1263 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1265 s->s_dev = *(dev_t *)data;
1269 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1271 return !(s->s_iflags & SB_I_RETIRED) && s->s_dev == *(dev_t *)data;
1274 static struct dentry *
1275 nilfs_mount(struct file_system_type *fs_type, int flags,
1276 const char *dev_name, void *data)
1278 struct nilfs_super_data sd = { .flags = flags };
1279 struct super_block *s;
1283 if (nilfs_identify(data, &sd))
1284 return ERR_PTR(-EINVAL);
1286 err = lookup_bdev(dev_name, &dev);
1288 return ERR_PTR(err);
1290 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1296 err = setup_bdev_super(s, flags, NULL);
1298 err = nilfs_fill_super(s, data,
1299 flags & SB_SILENT ? 1 : 0);
1303 s->s_flags |= SB_ACTIVE;
1304 } else if (!sd.cno) {
1305 if (nilfs_tree_is_busy(s->s_root)) {
1306 if ((flags ^ s->s_flags) & SB_RDONLY) {
1308 "the device already has a %s mount.",
1309 sb_rdonly(s) ? "read-only" : "read/write");
1315 * Try remount to setup mount states if the current
1316 * tree is not mounted and only snapshots use this sb.
1318 err = nilfs_remount(s, &flags, data);
1325 struct dentry *root_dentry;
1327 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1333 return dget(s->s_root);
1336 deactivate_locked_super(s);
1337 return ERR_PTR(err);
1340 struct file_system_type nilfs_fs_type = {
1341 .owner = THIS_MODULE,
1343 .mount = nilfs_mount,
1344 .kill_sb = kill_block_super,
1345 .fs_flags = FS_REQUIRES_DEV,
1347 MODULE_ALIAS_FS("nilfs2");
1349 static void nilfs_inode_init_once(void *obj)
1351 struct nilfs_inode_info *ii = obj;
1353 INIT_LIST_HEAD(&ii->i_dirty);
1354 #ifdef CONFIG_NILFS_XATTR
1355 init_rwsem(&ii->xattr_sem);
1357 inode_init_once(&ii->vfs_inode);
1360 static void nilfs_segbuf_init_once(void *obj)
1362 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1365 static void nilfs_destroy_cachep(void)
1368 * Make sure all delayed rcu free inodes are flushed before we
1373 kmem_cache_destroy(nilfs_inode_cachep);
1374 kmem_cache_destroy(nilfs_transaction_cachep);
1375 kmem_cache_destroy(nilfs_segbuf_cachep);
1376 kmem_cache_destroy(nilfs_btree_path_cache);
1379 static int __init nilfs_init_cachep(void)
1381 nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1382 sizeof(struct nilfs_inode_info), 0,
1383 SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1384 nilfs_inode_init_once);
1385 if (!nilfs_inode_cachep)
1388 nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1389 sizeof(struct nilfs_transaction_info), 0,
1390 SLAB_RECLAIM_ACCOUNT, NULL);
1391 if (!nilfs_transaction_cachep)
1394 nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1395 sizeof(struct nilfs_segment_buffer), 0,
1396 SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1397 if (!nilfs_segbuf_cachep)
1400 nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1401 sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1403 if (!nilfs_btree_path_cache)
1409 nilfs_destroy_cachep();
1413 static int __init init_nilfs_fs(void)
1417 err = nilfs_init_cachep();
1421 err = nilfs_sysfs_init();
1425 err = register_filesystem(&nilfs_fs_type);
1427 goto deinit_sysfs_entry;
1429 printk(KERN_INFO "NILFS version 2 loaded\n");
1435 nilfs_destroy_cachep();
1440 static void __exit exit_nilfs_fs(void)
1442 nilfs_destroy_cachep();
1444 unregister_filesystem(&nilfs_fs_type);
1447 module_init(init_nilfs_fs)
1448 module_exit(exit_nilfs_fs)