2 * the_nilfs.c - the_nilfs shared structure.
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 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 * Written by Ryusuke Konishi <ryusuke@osrg.net>
24 #include <linux/buffer_head.h>
25 #include <linux/slab.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/crc32.h>
38 static int nilfs_valid_sb(struct nilfs_super_block *sbp);
40 void nilfs_set_last_segment(struct the_nilfs *nilfs,
41 sector_t start_blocknr, u64 seq, __u64 cno)
43 spin_lock(&nilfs->ns_last_segment_lock);
44 nilfs->ns_last_pseg = start_blocknr;
45 nilfs->ns_last_seq = seq;
46 nilfs->ns_last_cno = cno;
48 if (!nilfs_sb_dirty(nilfs)) {
49 if (nilfs->ns_prev_seq == nilfs->ns_last_seq)
52 set_nilfs_sb_dirty(nilfs);
54 nilfs->ns_prev_seq = nilfs->ns_last_seq;
57 spin_unlock(&nilfs->ns_last_segment_lock);
61 * alloc_nilfs - allocate a nilfs object
62 * @bdev: block device to which the_nilfs is related
64 * Return Value: On success, pointer to the_nilfs is returned.
65 * On error, NULL is returned.
67 struct the_nilfs *alloc_nilfs(struct block_device *bdev)
69 struct the_nilfs *nilfs;
71 nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
75 nilfs->ns_bdev = bdev;
76 atomic_set(&nilfs->ns_ndirtyblks, 0);
77 init_rwsem(&nilfs->ns_sem);
78 init_rwsem(&nilfs->ns_writer_sem);
79 INIT_LIST_HEAD(&nilfs->ns_gc_inodes);
80 spin_lock_init(&nilfs->ns_last_segment_lock);
81 nilfs->ns_cptree = RB_ROOT;
82 spin_lock_init(&nilfs->ns_cptree_lock);
83 init_rwsem(&nilfs->ns_segctor_sem);
89 * destroy_nilfs - destroy nilfs object
90 * @nilfs: nilfs object to be released
92 void destroy_nilfs(struct the_nilfs *nilfs)
95 if (nilfs_init(nilfs)) {
96 brelse(nilfs->ns_sbh[0]);
97 brelse(nilfs->ns_sbh[1]);
102 static int nilfs_load_super_root(struct the_nilfs *nilfs,
103 struct super_block *sb, sector_t sr_block)
105 struct buffer_head *bh_sr;
106 struct nilfs_super_root *raw_sr;
107 struct nilfs_super_block **sbp = nilfs->ns_sbp;
108 struct nilfs_inode *rawi;
109 unsigned dat_entry_size, segment_usage_size, checkpoint_size;
113 err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
117 down_read(&nilfs->ns_sem);
118 dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
119 checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
120 segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
121 up_read(&nilfs->ns_sem);
123 inode_size = nilfs->ns_inode_size;
125 rawi = (void *)bh_sr->b_data + NILFS_SR_DAT_OFFSET(inode_size);
126 err = nilfs_dat_read(sb, dat_entry_size, rawi, &nilfs->ns_dat);
130 rawi = (void *)bh_sr->b_data + NILFS_SR_CPFILE_OFFSET(inode_size);
131 err = nilfs_cpfile_read(sb, checkpoint_size, rawi, &nilfs->ns_cpfile);
135 rawi = (void *)bh_sr->b_data + NILFS_SR_SUFILE_OFFSET(inode_size);
136 err = nilfs_sufile_read(sb, segment_usage_size, rawi,
141 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
142 nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
149 iput(nilfs->ns_cpfile);
156 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
158 memset(ri, 0, sizeof(*ri));
159 INIT_LIST_HEAD(&ri->ri_used_segments);
162 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
164 nilfs_dispose_segment_list(&ri->ri_used_segments);
168 * nilfs_store_log_cursor - load log cursor from a super block
169 * @nilfs: nilfs object
170 * @sbp: buffer storing super block to be read
172 * nilfs_store_log_cursor() reads the last position of the log
173 * containing a super root from a given super block, and initializes
174 * relevant information on the nilfs object preparatory for log
175 * scanning and recovery.
177 static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
178 struct nilfs_super_block *sbp)
182 nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
183 nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
184 nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
186 nilfs->ns_prev_seq = nilfs->ns_last_seq;
187 nilfs->ns_seg_seq = nilfs->ns_last_seq;
189 nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
190 nilfs->ns_cno = nilfs->ns_last_cno + 1;
191 if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
192 printk(KERN_ERR "NILFS invalid last segment number.\n");
199 * load_nilfs - load and recover the nilfs
200 * @nilfs: the_nilfs structure to be released
201 * @sbi: nilfs_sb_info used to recover past segment
203 * load_nilfs() searches and load the latest super root,
204 * attaches the last segment, and does recovery if needed.
205 * The caller must call this exclusively for simultaneous mounts.
207 int load_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
209 struct nilfs_recovery_info ri;
210 unsigned int s_flags = sbi->s_super->s_flags;
211 int really_read_only = bdev_read_only(nilfs->ns_bdev);
212 int valid_fs = nilfs_valid_fs(nilfs);
216 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
217 if (s_flags & MS_RDONLY) {
218 printk(KERN_INFO "NILFS: INFO: recovery "
219 "required for readonly filesystem.\n");
220 printk(KERN_INFO "NILFS: write access will "
221 "be enabled during recovery.\n");
225 nilfs_init_recovery_info(&ri);
227 err = nilfs_search_super_root(nilfs, &ri);
229 struct nilfs_super_block **sbp = nilfs->ns_sbp;
235 if (!nilfs_valid_sb(sbp[1])) {
237 "NILFS warning: unable to fall back to spare"
242 "NILFS: try rollback from an earlier position\n");
245 * restore super block with its spare and reconfigure
246 * relevant states of the nilfs object.
248 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
249 nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
250 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
252 /* verify consistency between two super blocks */
253 blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
254 if (blocksize != nilfs->ns_blocksize) {
256 "NILFS warning: blocksize differs between "
257 "two super blocks (%d != %d)\n",
258 blocksize, nilfs->ns_blocksize);
262 err = nilfs_store_log_cursor(nilfs, sbp[0]);
266 /* drop clean flag to allow roll-forward and recovery */
267 nilfs->ns_mount_state &= ~NILFS_VALID_FS;
270 err = nilfs_search_super_root(nilfs, &ri);
275 err = nilfs_load_super_root(nilfs, sbi->s_super, ri.ri_super_root);
277 printk(KERN_ERR "NILFS: error loading super root.\n");
284 if (s_flags & MS_RDONLY) {
287 if (nilfs_test_opt(sbi, NORECOVERY)) {
288 printk(KERN_INFO "NILFS: norecovery option specified. "
289 "skipping roll-forward recovery\n");
292 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
293 ~NILFS_FEATURE_COMPAT_RO_SUPP;
295 printk(KERN_ERR "NILFS: couldn't proceed with "
296 "recovery because of unsupported optional "
298 (unsigned long long)features);
302 if (really_read_only) {
303 printk(KERN_ERR "NILFS: write access "
304 "unavailable, cannot proceed.\n");
308 sbi->s_super->s_flags &= ~MS_RDONLY;
309 } else if (nilfs_test_opt(sbi, NORECOVERY)) {
310 printk(KERN_ERR "NILFS: recovery cancelled because norecovery "
311 "option was specified for a read/write mount\n");
316 err = nilfs_salvage_orphan_logs(nilfs, sbi, &ri);
320 down_write(&nilfs->ns_sem);
321 nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
322 err = nilfs_cleanup_super(sbi);
323 up_write(&nilfs->ns_sem);
326 printk(KERN_ERR "NILFS: failed to update super block. "
327 "recovery unfinished.\n");
330 printk(KERN_INFO "NILFS: recovery complete.\n");
333 set_nilfs_loaded(nilfs);
334 nilfs_clear_recovery_info(&ri);
335 sbi->s_super->s_flags = s_flags;
339 printk(KERN_ERR "NILFS: error searching super root.\n");
343 iput(nilfs->ns_cpfile);
344 iput(nilfs->ns_sufile);
348 nilfs_clear_recovery_info(&ri);
349 sbi->s_super->s_flags = s_flags;
353 static unsigned long long nilfs_max_size(unsigned int blkbits)
355 unsigned int max_bits;
356 unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
358 max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
360 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
364 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
365 struct nilfs_super_block *sbp)
367 if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
368 printk(KERN_ERR "NILFS: unsupported revision "
369 "(superblock rev.=%d.%d, current rev.=%d.%d). "
370 "Please check the version of mkfs.nilfs.\n",
371 le32_to_cpu(sbp->s_rev_level),
372 le16_to_cpu(sbp->s_minor_rev_level),
373 NILFS_CURRENT_REV, NILFS_MINOR_REV);
376 nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
377 if (nilfs->ns_sbsize > BLOCK_SIZE)
380 nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
381 nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
383 nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
384 if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
385 printk(KERN_ERR "NILFS: too short segment.\n");
389 nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
390 nilfs->ns_nsegments = le64_to_cpu(sbp->s_nsegments);
391 nilfs->ns_r_segments_percentage =
392 le32_to_cpu(sbp->s_r_segments_percentage);
394 max_t(unsigned long, NILFS_MIN_NRSVSEGS,
395 DIV_ROUND_UP(nilfs->ns_nsegments *
396 nilfs->ns_r_segments_percentage, 100));
397 nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
401 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
403 static unsigned char sum[4];
404 const int sumoff = offsetof(struct nilfs_super_block, s_sum);
408 if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
410 bytes = le16_to_cpu(sbp->s_bytes);
411 if (bytes > BLOCK_SIZE)
413 crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
415 crc = crc32_le(crc, sum, 4);
416 crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
418 return crc == le32_to_cpu(sbp->s_sum);
421 static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
423 return offset < ((le64_to_cpu(sbp->s_nsegments) *
424 le32_to_cpu(sbp->s_blocks_per_segment)) <<
425 (le32_to_cpu(sbp->s_log_block_size) + 10));
428 static void nilfs_release_super_block(struct the_nilfs *nilfs)
432 for (i = 0; i < 2; i++) {
433 if (nilfs->ns_sbp[i]) {
434 brelse(nilfs->ns_sbh[i]);
435 nilfs->ns_sbh[i] = NULL;
436 nilfs->ns_sbp[i] = NULL;
441 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
443 brelse(nilfs->ns_sbh[0]);
444 nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
445 nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
446 nilfs->ns_sbh[1] = NULL;
447 nilfs->ns_sbp[1] = NULL;
450 void nilfs_swap_super_block(struct the_nilfs *nilfs)
452 struct buffer_head *tsbh = nilfs->ns_sbh[0];
453 struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
455 nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
456 nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
457 nilfs->ns_sbh[1] = tsbh;
458 nilfs->ns_sbp[1] = tsbp;
461 static int nilfs_load_super_block(struct the_nilfs *nilfs,
462 struct super_block *sb, int blocksize,
463 struct nilfs_super_block **sbpp)
465 struct nilfs_super_block **sbp = nilfs->ns_sbp;
466 struct buffer_head **sbh = nilfs->ns_sbh;
467 u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
468 int valid[2], swp = 0;
470 sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
472 sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
476 printk(KERN_ERR "NILFS: unable to read superblock\n");
480 "NILFS warning: unable to read primary superblock\n");
483 "NILFS warning: unable to read secondary superblock\n");
486 * Compare two super blocks and set 1 in swp if the secondary
487 * super block is valid and newer. Otherwise, set 0 in swp.
489 valid[0] = nilfs_valid_sb(sbp[0]);
490 valid[1] = nilfs_valid_sb(sbp[1]);
491 swp = valid[1] && (!valid[0] ||
492 le64_to_cpu(sbp[1]->s_last_cno) >
493 le64_to_cpu(sbp[0]->s_last_cno));
495 if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
502 nilfs_release_super_block(nilfs);
503 printk(KERN_ERR "NILFS: Can't find nilfs on dev %s.\n",
509 printk(KERN_WARNING "NILFS warning: broken superblock. "
510 "using spare superblock.\n");
512 nilfs_swap_super_block(nilfs);
514 nilfs->ns_sbwcount = 0;
515 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
516 nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
522 * init_nilfs - initialize a NILFS instance.
523 * @nilfs: the_nilfs structure
524 * @sbi: nilfs_sb_info
526 * @data: mount options
528 * init_nilfs() performs common initialization per block device (e.g.
529 * reading the super block, getting disk layout information, initializing
530 * shared fields in the_nilfs).
532 * Return Value: On success, 0 is returned. On error, a negative error
535 int init_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi, char *data)
537 struct super_block *sb = sbi->s_super;
538 struct nilfs_super_block *sbp;
539 struct backing_dev_info *bdi;
543 down_write(&nilfs->ns_sem);
545 blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
547 printk(KERN_ERR "NILFS: unable to set blocksize\n");
551 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
555 err = nilfs_store_magic_and_option(sb, sbp, data);
559 err = nilfs_check_feature_compatibility(sb, sbp);
563 blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
564 if (blocksize < NILFS_MIN_BLOCK_SIZE ||
565 blocksize > NILFS_MAX_BLOCK_SIZE) {
566 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
567 "filesystem blocksize %d\n", blocksize);
571 if (sb->s_blocksize != blocksize) {
572 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
574 if (blocksize < hw_blocksize) {
576 "NILFS: blocksize %d too small for device "
577 "(sector-size = %d).\n",
578 blocksize, hw_blocksize);
582 nilfs_release_super_block(nilfs);
583 sb_set_blocksize(sb, blocksize);
585 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
588 /* not failed_sbh; sbh is released automatically
589 when reloading fails. */
591 nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
592 nilfs->ns_blocksize = blocksize;
594 err = nilfs_store_disk_layout(nilfs, sbp);
598 sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
600 nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
602 bdi = nilfs->ns_bdev->bd_inode->i_mapping->backing_dev_info;
603 nilfs->ns_bdi = bdi ? : &default_backing_dev_info;
605 err = nilfs_store_log_cursor(nilfs, sbp);
609 set_nilfs_init(nilfs);
612 up_write(&nilfs->ns_sem);
616 nilfs_release_super_block(nilfs);
620 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
623 sector_t seg_start, seg_end;
624 sector_t start = 0, nblocks = 0;
625 unsigned int sects_per_block;
629 sects_per_block = (1 << nilfs->ns_blocksize_bits) /
630 bdev_logical_block_size(nilfs->ns_bdev);
631 for (sn = segnump; sn < segnump + nsegs; sn++) {
632 nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
636 nblocks = seg_end - seg_start + 1;
637 } else if (start + nblocks == seg_start) {
638 nblocks += seg_end - seg_start + 1;
640 ret = blkdev_issue_discard(nilfs->ns_bdev,
641 start * sects_per_block,
642 nblocks * sects_per_block,
652 ret = blkdev_issue_discard(nilfs->ns_bdev,
653 start * sects_per_block,
654 nblocks * sects_per_block,
656 BLKDEV_IFL_WAIT | BLKDEV_IFL_BARRIER);
660 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
662 struct inode *dat = nilfs_dat_inode(nilfs);
663 unsigned long ncleansegs;
665 down_read(&NILFS_MDT(dat)->mi_sem); /* XXX */
666 ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
667 up_read(&NILFS_MDT(dat)->mi_sem); /* XXX */
668 *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
672 int nilfs_near_disk_full(struct the_nilfs *nilfs)
674 unsigned long ncleansegs, nincsegs;
676 ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
677 nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
678 nilfs->ns_blocks_per_segment + 1;
680 return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
683 struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
686 struct nilfs_root *root;
688 spin_lock(&nilfs->ns_cptree_lock);
689 n = nilfs->ns_cptree.rb_node;
691 root = rb_entry(n, struct nilfs_root, rb_node);
693 if (cno < root->cno) {
695 } else if (cno > root->cno) {
698 atomic_inc(&root->count);
699 spin_unlock(&nilfs->ns_cptree_lock);
703 spin_unlock(&nilfs->ns_cptree_lock);
709 nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
711 struct rb_node **p, *parent;
712 struct nilfs_root *root, *new;
714 root = nilfs_lookup_root(nilfs, cno);
718 new = kmalloc(sizeof(*root), GFP_KERNEL);
722 spin_lock(&nilfs->ns_cptree_lock);
724 p = &nilfs->ns_cptree.rb_node;
729 root = rb_entry(parent, struct nilfs_root, rb_node);
731 if (cno < root->cno) {
733 } else if (cno > root->cno) {
736 atomic_inc(&root->count);
737 spin_unlock(&nilfs->ns_cptree_lock);
746 atomic_set(&new->count, 1);
747 atomic_set(&new->inodes_count, 0);
748 atomic_set(&new->blocks_count, 0);
750 rb_link_node(&new->rb_node, parent, p);
751 rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
753 spin_unlock(&nilfs->ns_cptree_lock);
758 void nilfs_put_root(struct nilfs_root *root)
760 if (atomic_dec_and_test(&root->count)) {
761 struct the_nilfs *nilfs = root->nilfs;
763 spin_lock(&nilfs->ns_cptree_lock);
764 rb_erase(&root->rb_node, &nilfs->ns_cptree);
765 spin_unlock(&nilfs->ns_cptree_lock);
773 int nilfs_checkpoint_is_mounted(struct the_nilfs *nilfs, __u64 cno,
776 struct nilfs_root *root;
779 if (cno < 0 || cno > nilfs->ns_cno)
782 if (cno >= nilfs_last_cno(nilfs))
783 return true; /* protect recent checkpoints */
786 root = nilfs_lookup_root(nilfs, cno);
789 nilfs_put_root(root);