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 void nilfs_set_last_segment(struct the_nilfs *nilfs,
39 sector_t start_blocknr, u64 seq, __u64 cno)
41 spin_lock(&nilfs->ns_last_segment_lock);
42 nilfs->ns_last_pseg = start_blocknr;
43 nilfs->ns_last_seq = seq;
44 nilfs->ns_last_cno = cno;
45 spin_unlock(&nilfs->ns_last_segment_lock);
49 * alloc_nilfs - allocate the_nilfs structure
50 * @bdev: block device to which the_nilfs is related
52 * alloc_nilfs() allocates memory for the_nilfs and
53 * initializes its reference count and locks.
55 * Return Value: On success, pointer to the_nilfs is returned.
56 * On error, NULL is returned.
58 struct the_nilfs *alloc_nilfs(struct block_device *bdev)
60 struct the_nilfs *nilfs;
62 nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
66 nilfs->ns_bdev = bdev;
67 atomic_set(&nilfs->ns_count, 1);
68 atomic_set(&nilfs->ns_writer_refcount, -1);
69 atomic_set(&nilfs->ns_ndirtyblks, 0);
70 init_rwsem(&nilfs->ns_sem);
71 mutex_init(&nilfs->ns_writer_mutex);
72 INIT_LIST_HEAD(&nilfs->ns_supers);
73 spin_lock_init(&nilfs->ns_last_segment_lock);
74 nilfs->ns_gc_inodes_h = NULL;
75 init_rwsem(&nilfs->ns_segctor_sem);
81 * put_nilfs - release a reference to the_nilfs
82 * @nilfs: the_nilfs structure to be released
84 * put_nilfs() decrements a reference counter of the_nilfs.
85 * If the reference count reaches zero, the_nilfs is freed.
87 void put_nilfs(struct the_nilfs *nilfs)
89 if (!atomic_dec_and_test(&nilfs->ns_count))
92 * Increment of ns_count never occur below because the caller
93 * of get_nilfs() holds at least one reference to the_nilfs.
94 * Thus its exclusion control is not required here.
97 if (nilfs_loaded(nilfs)) {
98 nilfs_mdt_clear(nilfs->ns_sufile);
99 nilfs_mdt_destroy(nilfs->ns_sufile);
100 nilfs_mdt_clear(nilfs->ns_cpfile);
101 nilfs_mdt_destroy(nilfs->ns_cpfile);
102 nilfs_mdt_clear(nilfs->ns_dat);
103 nilfs_mdt_destroy(nilfs->ns_dat);
104 /* XXX: how and when to clear nilfs->ns_gc_dat? */
105 nilfs_mdt_destroy(nilfs->ns_gc_dat);
107 if (nilfs_init(nilfs)) {
108 nilfs_destroy_gccache(nilfs);
109 brelse(nilfs->ns_sbh[0]);
110 brelse(nilfs->ns_sbh[1]);
115 static int nilfs_load_super_root(struct the_nilfs *nilfs,
116 struct nilfs_sb_info *sbi, sector_t sr_block)
118 static struct lock_class_key dat_lock_key;
119 struct buffer_head *bh_sr;
120 struct nilfs_super_root *raw_sr;
121 struct nilfs_super_block **sbp = nilfs->ns_sbp;
122 unsigned dat_entry_size, segment_usage_size, checkpoint_size;
126 err = nilfs_read_super_root_block(sbi->s_super, sr_block, &bh_sr, 1);
130 down_read(&nilfs->ns_sem);
131 dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
132 checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
133 segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
134 up_read(&nilfs->ns_sem);
136 inode_size = nilfs->ns_inode_size;
139 nilfs->ns_dat = nilfs_mdt_new(
140 nilfs, NULL, NILFS_DAT_INO, NILFS_DAT_GFP);
141 if (unlikely(!nilfs->ns_dat))
144 nilfs->ns_gc_dat = nilfs_mdt_new(
145 nilfs, NULL, NILFS_DAT_INO, NILFS_DAT_GFP);
146 if (unlikely(!nilfs->ns_gc_dat))
149 nilfs->ns_cpfile = nilfs_mdt_new(
150 nilfs, NULL, NILFS_CPFILE_INO, NILFS_CPFILE_GFP);
151 if (unlikely(!nilfs->ns_cpfile))
154 nilfs->ns_sufile = nilfs_mdt_new(
155 nilfs, NULL, NILFS_SUFILE_INO, NILFS_SUFILE_GFP);
156 if (unlikely(!nilfs->ns_sufile))
159 err = nilfs_palloc_init_blockgroup(nilfs->ns_dat, dat_entry_size);
163 err = nilfs_palloc_init_blockgroup(nilfs->ns_gc_dat, dat_entry_size);
167 lockdep_set_class(&NILFS_MDT(nilfs->ns_dat)->mi_sem, &dat_lock_key);
168 lockdep_set_class(&NILFS_MDT(nilfs->ns_gc_dat)->mi_sem, &dat_lock_key);
170 nilfs_mdt_set_shadow(nilfs->ns_dat, nilfs->ns_gc_dat);
171 nilfs_mdt_set_entry_size(nilfs->ns_cpfile, checkpoint_size,
172 sizeof(struct nilfs_cpfile_header));
173 nilfs_mdt_set_entry_size(nilfs->ns_sufile, segment_usage_size,
174 sizeof(struct nilfs_sufile_header));
176 err = nilfs_mdt_read_inode_direct(
177 nilfs->ns_dat, bh_sr, NILFS_SR_DAT_OFFSET(inode_size));
181 err = nilfs_mdt_read_inode_direct(
182 nilfs->ns_cpfile, bh_sr, NILFS_SR_CPFILE_OFFSET(inode_size));
186 err = nilfs_mdt_read_inode_direct(
187 nilfs->ns_sufile, bh_sr, NILFS_SR_SUFILE_OFFSET(inode_size));
191 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
192 nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
199 nilfs_mdt_destroy(nilfs->ns_sufile);
202 nilfs_mdt_destroy(nilfs->ns_cpfile);
205 nilfs_mdt_destroy(nilfs->ns_gc_dat);
208 nilfs_mdt_destroy(nilfs->ns_dat);
212 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
214 memset(ri, 0, sizeof(*ri));
215 INIT_LIST_HEAD(&ri->ri_used_segments);
218 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
220 nilfs_dispose_segment_list(&ri->ri_used_segments);
224 * load_nilfs - load and recover the nilfs
225 * @nilfs: the_nilfs structure to be released
226 * @sbi: nilfs_sb_info used to recover past segment
228 * load_nilfs() searches and load the latest super root,
229 * attaches the last segment, and does recovery if needed.
230 * The caller must call this exclusively for simultaneous mounts.
232 int load_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
234 struct nilfs_recovery_info ri;
235 unsigned int s_flags = sbi->s_super->s_flags;
236 int really_read_only = bdev_read_only(nilfs->ns_bdev);
240 nilfs_init_recovery_info(&ri);
242 down_write(&nilfs->ns_sem);
243 valid_fs = (nilfs->ns_mount_state & NILFS_VALID_FS);
244 up_write(&nilfs->ns_sem);
246 if (!valid_fs && (s_flags & MS_RDONLY)) {
247 printk(KERN_INFO "NILFS: INFO: recovery "
248 "required for readonly filesystem.\n");
249 if (really_read_only) {
250 printk(KERN_ERR "NILFS: write access "
251 "unavailable, cannot proceed.\n");
255 printk(KERN_INFO "NILFS: write access will "
256 "be enabled during recovery.\n");
257 sbi->s_super->s_flags &= ~MS_RDONLY;
260 err = nilfs_search_super_root(nilfs, sbi, &ri);
262 printk(KERN_ERR "NILFS: error searching super root.\n");
266 err = nilfs_load_super_root(nilfs, sbi, ri.ri_super_root);
268 printk(KERN_ERR "NILFS: error loading super root.\n");
273 err = nilfs_recover_logical_segments(nilfs, sbi, &ri);
275 nilfs_mdt_destroy(nilfs->ns_cpfile);
276 nilfs_mdt_destroy(nilfs->ns_sufile);
277 nilfs_mdt_destroy(nilfs->ns_dat);
280 if (ri.ri_need_recovery == NILFS_RECOVERY_SR_UPDATED)
281 sbi->s_super->s_dirt = 1;
284 set_nilfs_loaded(nilfs);
287 nilfs_clear_recovery_info(&ri);
288 sbi->s_super->s_flags = s_flags;
292 static unsigned long long nilfs_max_size(unsigned int blkbits)
294 unsigned int max_bits;
295 unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
297 max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
299 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
303 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
304 struct nilfs_super_block *sbp)
306 if (le32_to_cpu(sbp->s_rev_level) != NILFS_CURRENT_REV) {
307 printk(KERN_ERR "NILFS: revision mismatch "
308 "(superblock rev.=%d.%d, current rev.=%d.%d). "
309 "Please check the version of mkfs.nilfs.\n",
310 le32_to_cpu(sbp->s_rev_level),
311 le16_to_cpu(sbp->s_minor_rev_level),
312 NILFS_CURRENT_REV, NILFS_MINOR_REV);
315 nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
316 if (nilfs->ns_sbsize > BLOCK_SIZE)
319 nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
320 nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
322 nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
323 if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
324 printk(KERN_ERR "NILFS: too short segment. \n");
328 nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
329 nilfs->ns_nsegments = le64_to_cpu(sbp->s_nsegments);
330 nilfs->ns_r_segments_percentage =
331 le32_to_cpu(sbp->s_r_segments_percentage);
333 max_t(unsigned long, NILFS_MIN_NRSVSEGS,
334 DIV_ROUND_UP(nilfs->ns_nsegments *
335 nilfs->ns_r_segments_percentage, 100));
336 nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
340 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
342 static unsigned char sum[4];
343 const int sumoff = offsetof(struct nilfs_super_block, s_sum);
347 if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
349 bytes = le16_to_cpu(sbp->s_bytes);
350 if (bytes > BLOCK_SIZE)
352 crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
354 crc = crc32_le(crc, sum, 4);
355 crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
357 return crc == le32_to_cpu(sbp->s_sum);
360 static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
362 return offset < ((le64_to_cpu(sbp->s_nsegments) *
363 le32_to_cpu(sbp->s_blocks_per_segment)) <<
364 (le32_to_cpu(sbp->s_log_block_size) + 10));
367 static void nilfs_release_super_block(struct the_nilfs *nilfs)
371 for (i = 0; i < 2; i++) {
372 if (nilfs->ns_sbp[i]) {
373 brelse(nilfs->ns_sbh[i]);
374 nilfs->ns_sbh[i] = NULL;
375 nilfs->ns_sbp[i] = NULL;
380 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
382 brelse(nilfs->ns_sbh[0]);
383 nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
384 nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
385 nilfs->ns_sbh[1] = NULL;
386 nilfs->ns_sbp[1] = NULL;
389 void nilfs_swap_super_block(struct the_nilfs *nilfs)
391 struct buffer_head *tsbh = nilfs->ns_sbh[0];
392 struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
394 nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
395 nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
396 nilfs->ns_sbh[1] = tsbh;
397 nilfs->ns_sbp[1] = tsbp;
400 static int nilfs_load_super_block(struct the_nilfs *nilfs,
401 struct super_block *sb, int blocksize,
402 struct nilfs_super_block **sbpp)
404 struct nilfs_super_block **sbp = nilfs->ns_sbp;
405 struct buffer_head **sbh = nilfs->ns_sbh;
406 u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
407 int valid[2], swp = 0;
409 sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
411 sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
415 printk(KERN_ERR "NILFS: unable to read superblock\n");
419 "NILFS warning: unable to read primary superblock\n");
422 "NILFS warning: unable to read secondary superblock\n");
424 valid[0] = nilfs_valid_sb(sbp[0]);
425 valid[1] = nilfs_valid_sb(sbp[1]);
428 le64_to_cpu(sbp[1]->s_wtime) > le64_to_cpu(sbp[0]->s_wtime));
430 if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
437 nilfs_release_super_block(nilfs);
438 printk(KERN_ERR "NILFS: Can't find nilfs on dev %s.\n",
444 printk(KERN_WARNING "NILFS warning: broken superblock. "
445 "using spare superblock.\n");
446 nilfs_swap_super_block(nilfs);
449 nilfs->ns_sbwtime[0] = le64_to_cpu(sbp[0]->s_wtime);
450 nilfs->ns_sbwtime[1] = valid[!swp] ? le64_to_cpu(sbp[1]->s_wtime) : 0;
451 nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
457 * init_nilfs - initialize a NILFS instance.
458 * @nilfs: the_nilfs structure
459 * @sbi: nilfs_sb_info
461 * @data: mount options
463 * init_nilfs() performs common initialization per block device (e.g.
464 * reading the super block, getting disk layout information, initializing
465 * shared fields in the_nilfs). It takes on some portion of the jobs
466 * typically done by a fill_super() routine. This division arises from
467 * the nature that multiple NILFS instances may be simultaneously
468 * mounted on a device.
469 * For multiple mounts on the same device, only the first mount
470 * invokes these tasks.
472 * Return Value: On success, 0 is returned. On error, a negative error
475 int init_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi, char *data)
477 struct super_block *sb = sbi->s_super;
478 struct nilfs_super_block *sbp;
479 struct backing_dev_info *bdi;
483 down_write(&nilfs->ns_sem);
484 if (nilfs_init(nilfs)) {
485 /* Load values from existing the_nilfs */
486 sbp = nilfs->ns_sbp[0];
487 err = nilfs_store_magic_and_option(sb, sbp, data);
491 blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
492 if (sb->s_blocksize != blocksize &&
493 !sb_set_blocksize(sb, blocksize)) {
494 printk(KERN_ERR "NILFS: blocksize %d unfit to device\n",
498 sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
502 blocksize = sb_min_blocksize(sb, BLOCK_SIZE);
504 printk(KERN_ERR "NILFS: unable to set blocksize\n");
508 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
512 err = nilfs_store_magic_and_option(sb, sbp, data);
516 blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
517 if (sb->s_blocksize != blocksize) {
518 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
520 if (blocksize < hw_blocksize) {
522 "NILFS: blocksize %d too small for device "
523 "(sector-size = %d).\n",
524 blocksize, hw_blocksize);
528 nilfs_release_super_block(nilfs);
529 sb_set_blocksize(sb, blocksize);
531 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
534 /* not failed_sbh; sbh is released automatically
535 when reloading fails. */
537 nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
539 err = nilfs_store_disk_layout(nilfs, sbp);
543 sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
545 nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
547 bdi = nilfs->ns_bdev->bd_inode_backing_dev_info;
549 bdi = nilfs->ns_bdev->bd_inode->i_mapping->backing_dev_info;
550 nilfs->ns_bdi = bdi ? : &default_backing_dev_info;
552 /* Finding last segment */
553 nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
554 nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
555 nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
557 nilfs->ns_seg_seq = nilfs->ns_last_seq;
559 nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
560 nilfs->ns_cno = nilfs->ns_last_cno + 1;
561 if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
562 printk(KERN_ERR "NILFS invalid last segment number.\n");
567 nilfs->ns_free_segments_count =
568 nilfs->ns_nsegments - (nilfs->ns_segnum + 1);
570 /* Initialize gcinode cache */
571 err = nilfs_init_gccache(nilfs);
575 set_nilfs_init(nilfs);
578 up_write(&nilfs->ns_sem);
582 nilfs_release_super_block(nilfs);
586 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
588 struct inode *dat = nilfs_dat_inode(nilfs);
589 unsigned long ncleansegs;
592 down_read(&NILFS_MDT(dat)->mi_sem); /* XXX */
593 err = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile, &ncleansegs);
594 up_read(&NILFS_MDT(dat)->mi_sem); /* XXX */
596 *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
600 int nilfs_near_disk_full(struct the_nilfs *nilfs)
602 struct inode *sufile = nilfs->ns_sufile;
603 unsigned long ncleansegs, nincsegs;
606 ret = nilfs_sufile_get_ncleansegs(sufile, &ncleansegs);
608 nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
609 nilfs->ns_blocks_per_segment + 1;
610 if (ncleansegs <= nilfs->ns_nrsvsegs + nincsegs)
616 int nilfs_checkpoint_is_mounted(struct the_nilfs *nilfs, __u64 cno,
619 struct nilfs_sb_info *sbi;
622 down_read(&nilfs->ns_sem);
623 if (cno == 0 || cno > nilfs->ns_cno)
626 list_for_each_entry(sbi, &nilfs->ns_supers, s_list) {
627 if (sbi->s_snapshot_cno == cno &&
628 (!snapshot_mount || nilfs_test_opt(sbi, SNAPSHOT))) {
629 /* exclude read-only mounts */
634 /* for protecting recent checkpoints */
635 if (cno >= nilfs_last_cno(nilfs))
639 up_read(&nilfs->ns_sem);