xref: /openbmc/linux/fs/nilfs2/the_nilfs.c (revision e23feb16)
1 /*
2  * the_nilfs.c - the_nilfs shared structure.
3  *
4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5  *
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.
10  *
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.
15  *
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
19  *
20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
21  *
22  */
23 
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/random.h>
29 #include <linux/crc32.h>
30 #include "nilfs.h"
31 #include "segment.h"
32 #include "alloc.h"
33 #include "cpfile.h"
34 #include "sufile.h"
35 #include "dat.h"
36 #include "segbuf.h"
37 
38 
39 static int nilfs_valid_sb(struct nilfs_super_block *sbp);
40 
41 void nilfs_set_last_segment(struct the_nilfs *nilfs,
42 			    sector_t start_blocknr, u64 seq, __u64 cno)
43 {
44 	spin_lock(&nilfs->ns_last_segment_lock);
45 	nilfs->ns_last_pseg = start_blocknr;
46 	nilfs->ns_last_seq = seq;
47 	nilfs->ns_last_cno = cno;
48 
49 	if (!nilfs_sb_dirty(nilfs)) {
50 		if (nilfs->ns_prev_seq == nilfs->ns_last_seq)
51 			goto stay_cursor;
52 
53 		set_nilfs_sb_dirty(nilfs);
54 	}
55 	nilfs->ns_prev_seq = nilfs->ns_last_seq;
56 
57  stay_cursor:
58 	spin_unlock(&nilfs->ns_last_segment_lock);
59 }
60 
61 /**
62  * alloc_nilfs - allocate a nilfs object
63  * @bdev: block device to which the_nilfs is related
64  *
65  * Return Value: On success, pointer to the_nilfs is returned.
66  * On error, NULL is returned.
67  */
68 struct the_nilfs *alloc_nilfs(struct block_device *bdev)
69 {
70 	struct the_nilfs *nilfs;
71 
72 	nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
73 	if (!nilfs)
74 		return NULL;
75 
76 	nilfs->ns_bdev = bdev;
77 	atomic_set(&nilfs->ns_ndirtyblks, 0);
78 	init_rwsem(&nilfs->ns_sem);
79 	mutex_init(&nilfs->ns_snapshot_mount_mutex);
80 	INIT_LIST_HEAD(&nilfs->ns_dirty_files);
81 	INIT_LIST_HEAD(&nilfs->ns_gc_inodes);
82 	spin_lock_init(&nilfs->ns_inode_lock);
83 	spin_lock_init(&nilfs->ns_next_gen_lock);
84 	spin_lock_init(&nilfs->ns_last_segment_lock);
85 	nilfs->ns_cptree = RB_ROOT;
86 	spin_lock_init(&nilfs->ns_cptree_lock);
87 	init_rwsem(&nilfs->ns_segctor_sem);
88 
89 	return nilfs;
90 }
91 
92 /**
93  * destroy_nilfs - destroy nilfs object
94  * @nilfs: nilfs object to be released
95  */
96 void destroy_nilfs(struct the_nilfs *nilfs)
97 {
98 	might_sleep();
99 	if (nilfs_init(nilfs)) {
100 		brelse(nilfs->ns_sbh[0]);
101 		brelse(nilfs->ns_sbh[1]);
102 	}
103 	kfree(nilfs);
104 }
105 
106 static int nilfs_load_super_root(struct the_nilfs *nilfs,
107 				 struct super_block *sb, sector_t sr_block)
108 {
109 	struct buffer_head *bh_sr;
110 	struct nilfs_super_root *raw_sr;
111 	struct nilfs_super_block **sbp = nilfs->ns_sbp;
112 	struct nilfs_inode *rawi;
113 	unsigned dat_entry_size, segment_usage_size, checkpoint_size;
114 	unsigned inode_size;
115 	int err;
116 
117 	err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
118 	if (unlikely(err))
119 		return err;
120 
121 	down_read(&nilfs->ns_sem);
122 	dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
123 	checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
124 	segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
125 	up_read(&nilfs->ns_sem);
126 
127 	inode_size = nilfs->ns_inode_size;
128 
129 	rawi = (void *)bh_sr->b_data + NILFS_SR_DAT_OFFSET(inode_size);
130 	err = nilfs_dat_read(sb, dat_entry_size, rawi, &nilfs->ns_dat);
131 	if (err)
132 		goto failed;
133 
134 	rawi = (void *)bh_sr->b_data + NILFS_SR_CPFILE_OFFSET(inode_size);
135 	err = nilfs_cpfile_read(sb, checkpoint_size, rawi, &nilfs->ns_cpfile);
136 	if (err)
137 		goto failed_dat;
138 
139 	rawi = (void *)bh_sr->b_data + NILFS_SR_SUFILE_OFFSET(inode_size);
140 	err = nilfs_sufile_read(sb, segment_usage_size, rawi,
141 				&nilfs->ns_sufile);
142 	if (err)
143 		goto failed_cpfile;
144 
145 	raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
146 	nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
147 
148  failed:
149 	brelse(bh_sr);
150 	return err;
151 
152  failed_cpfile:
153 	iput(nilfs->ns_cpfile);
154 
155  failed_dat:
156 	iput(nilfs->ns_dat);
157 	goto failed;
158 }
159 
160 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
161 {
162 	memset(ri, 0, sizeof(*ri));
163 	INIT_LIST_HEAD(&ri->ri_used_segments);
164 }
165 
166 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
167 {
168 	nilfs_dispose_segment_list(&ri->ri_used_segments);
169 }
170 
171 /**
172  * nilfs_store_log_cursor - load log cursor from a super block
173  * @nilfs: nilfs object
174  * @sbp: buffer storing super block to be read
175  *
176  * nilfs_store_log_cursor() reads the last position of the log
177  * containing a super root from a given super block, and initializes
178  * relevant information on the nilfs object preparatory for log
179  * scanning and recovery.
180  */
181 static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
182 				  struct nilfs_super_block *sbp)
183 {
184 	int ret = 0;
185 
186 	nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
187 	nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
188 	nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
189 
190 	nilfs->ns_prev_seq = nilfs->ns_last_seq;
191 	nilfs->ns_seg_seq = nilfs->ns_last_seq;
192 	nilfs->ns_segnum =
193 		nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
194 	nilfs->ns_cno = nilfs->ns_last_cno + 1;
195 	if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
196 		printk(KERN_ERR "NILFS invalid last segment number.\n");
197 		ret = -EINVAL;
198 	}
199 	return ret;
200 }
201 
202 /**
203  * load_nilfs - load and recover the nilfs
204  * @nilfs: the_nilfs structure to be released
205  * @sb: super block isntance used to recover past segment
206  *
207  * load_nilfs() searches and load the latest super root,
208  * attaches the last segment, and does recovery if needed.
209  * The caller must call this exclusively for simultaneous mounts.
210  */
211 int load_nilfs(struct the_nilfs *nilfs, struct super_block *sb)
212 {
213 	struct nilfs_recovery_info ri;
214 	unsigned int s_flags = sb->s_flags;
215 	int really_read_only = bdev_read_only(nilfs->ns_bdev);
216 	int valid_fs = nilfs_valid_fs(nilfs);
217 	int err;
218 
219 	if (!valid_fs) {
220 		printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
221 		if (s_flags & MS_RDONLY) {
222 			printk(KERN_INFO "NILFS: INFO: recovery "
223 			       "required for readonly filesystem.\n");
224 			printk(KERN_INFO "NILFS: write access will "
225 			       "be enabled during recovery.\n");
226 		}
227 	}
228 
229 	nilfs_init_recovery_info(&ri);
230 
231 	err = nilfs_search_super_root(nilfs, &ri);
232 	if (unlikely(err)) {
233 		struct nilfs_super_block **sbp = nilfs->ns_sbp;
234 		int blocksize;
235 
236 		if (err != -EINVAL)
237 			goto scan_error;
238 
239 		if (!nilfs_valid_sb(sbp[1])) {
240 			printk(KERN_WARNING
241 			       "NILFS warning: unable to fall back to spare"
242 			       "super block\n");
243 			goto scan_error;
244 		}
245 		printk(KERN_INFO
246 		       "NILFS: try rollback from an earlier position\n");
247 
248 		/*
249 		 * restore super block with its spare and reconfigure
250 		 * relevant states of the nilfs object.
251 		 */
252 		memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
253 		nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
254 		nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
255 
256 		/* verify consistency between two super blocks */
257 		blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
258 		if (blocksize != nilfs->ns_blocksize) {
259 			printk(KERN_WARNING
260 			       "NILFS warning: blocksize differs between "
261 			       "two super blocks (%d != %d)\n",
262 			       blocksize, nilfs->ns_blocksize);
263 			goto scan_error;
264 		}
265 
266 		err = nilfs_store_log_cursor(nilfs, sbp[0]);
267 		if (err)
268 			goto scan_error;
269 
270 		/* drop clean flag to allow roll-forward and recovery */
271 		nilfs->ns_mount_state &= ~NILFS_VALID_FS;
272 		valid_fs = 0;
273 
274 		err = nilfs_search_super_root(nilfs, &ri);
275 		if (err)
276 			goto scan_error;
277 	}
278 
279 	err = nilfs_load_super_root(nilfs, sb, ri.ri_super_root);
280 	if (unlikely(err)) {
281 		printk(KERN_ERR "NILFS: error loading super root.\n");
282 		goto failed;
283 	}
284 
285 	if (valid_fs)
286 		goto skip_recovery;
287 
288 	if (s_flags & MS_RDONLY) {
289 		__u64 features;
290 
291 		if (nilfs_test_opt(nilfs, NORECOVERY)) {
292 			printk(KERN_INFO "NILFS: norecovery option specified. "
293 			       "skipping roll-forward recovery\n");
294 			goto skip_recovery;
295 		}
296 		features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
297 			~NILFS_FEATURE_COMPAT_RO_SUPP;
298 		if (features) {
299 			printk(KERN_ERR "NILFS: couldn't proceed with "
300 			       "recovery because of unsupported optional "
301 			       "features (%llx)\n",
302 			       (unsigned long long)features);
303 			err = -EROFS;
304 			goto failed_unload;
305 		}
306 		if (really_read_only) {
307 			printk(KERN_ERR "NILFS: write access "
308 			       "unavailable, cannot proceed.\n");
309 			err = -EROFS;
310 			goto failed_unload;
311 		}
312 		sb->s_flags &= ~MS_RDONLY;
313 	} else if (nilfs_test_opt(nilfs, NORECOVERY)) {
314 		printk(KERN_ERR "NILFS: recovery cancelled because norecovery "
315 		       "option was specified for a read/write mount\n");
316 		err = -EINVAL;
317 		goto failed_unload;
318 	}
319 
320 	err = nilfs_salvage_orphan_logs(nilfs, sb, &ri);
321 	if (err)
322 		goto failed_unload;
323 
324 	down_write(&nilfs->ns_sem);
325 	nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
326 	err = nilfs_cleanup_super(sb);
327 	up_write(&nilfs->ns_sem);
328 
329 	if (err) {
330 		printk(KERN_ERR "NILFS: failed to update super block. "
331 		       "recovery unfinished.\n");
332 		goto failed_unload;
333 	}
334 	printk(KERN_INFO "NILFS: recovery complete.\n");
335 
336  skip_recovery:
337 	nilfs_clear_recovery_info(&ri);
338 	sb->s_flags = s_flags;
339 	return 0;
340 
341  scan_error:
342 	printk(KERN_ERR "NILFS: error searching super root.\n");
343 	goto failed;
344 
345  failed_unload:
346 	iput(nilfs->ns_cpfile);
347 	iput(nilfs->ns_sufile);
348 	iput(nilfs->ns_dat);
349 
350  failed:
351 	nilfs_clear_recovery_info(&ri);
352 	sb->s_flags = s_flags;
353 	return err;
354 }
355 
356 static unsigned long long nilfs_max_size(unsigned int blkbits)
357 {
358 	unsigned int max_bits;
359 	unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
360 
361 	max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
362 	if (max_bits < 64)
363 		res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
364 	return res;
365 }
366 
367 /**
368  * nilfs_nrsvsegs - calculate the number of reserved segments
369  * @nilfs: nilfs object
370  * @nsegs: total number of segments
371  */
372 unsigned long nilfs_nrsvsegs(struct the_nilfs *nilfs, unsigned long nsegs)
373 {
374 	return max_t(unsigned long, NILFS_MIN_NRSVSEGS,
375 		     DIV_ROUND_UP(nsegs * nilfs->ns_r_segments_percentage,
376 				  100));
377 }
378 
379 void nilfs_set_nsegments(struct the_nilfs *nilfs, unsigned long nsegs)
380 {
381 	nilfs->ns_nsegments = nsegs;
382 	nilfs->ns_nrsvsegs = nilfs_nrsvsegs(nilfs, nsegs);
383 }
384 
385 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
386 				   struct nilfs_super_block *sbp)
387 {
388 	if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
389 		printk(KERN_ERR "NILFS: unsupported revision "
390 		       "(superblock rev.=%d.%d, current rev.=%d.%d). "
391 		       "Please check the version of mkfs.nilfs.\n",
392 		       le32_to_cpu(sbp->s_rev_level),
393 		       le16_to_cpu(sbp->s_minor_rev_level),
394 		       NILFS_CURRENT_REV, NILFS_MINOR_REV);
395 		return -EINVAL;
396 	}
397 	nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
398 	if (nilfs->ns_sbsize > BLOCK_SIZE)
399 		return -EINVAL;
400 
401 	nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
402 	nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
403 
404 	nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
405 	if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
406 		printk(KERN_ERR "NILFS: too short segment.\n");
407 		return -EINVAL;
408 	}
409 
410 	nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
411 	nilfs->ns_r_segments_percentage =
412 		le32_to_cpu(sbp->s_r_segments_percentage);
413 	if (nilfs->ns_r_segments_percentage < 1 ||
414 	    nilfs->ns_r_segments_percentage > 99) {
415 		printk(KERN_ERR "NILFS: invalid reserved segments percentage.\n");
416 		return -EINVAL;
417 	}
418 
419 	nilfs_set_nsegments(nilfs, le64_to_cpu(sbp->s_nsegments));
420 	nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
421 	return 0;
422 }
423 
424 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
425 {
426 	static unsigned char sum[4];
427 	const int sumoff = offsetof(struct nilfs_super_block, s_sum);
428 	size_t bytes;
429 	u32 crc;
430 
431 	if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
432 		return 0;
433 	bytes = le16_to_cpu(sbp->s_bytes);
434 	if (bytes > BLOCK_SIZE)
435 		return 0;
436 	crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
437 		       sumoff);
438 	crc = crc32_le(crc, sum, 4);
439 	crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
440 		       bytes - sumoff - 4);
441 	return crc == le32_to_cpu(sbp->s_sum);
442 }
443 
444 static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
445 {
446 	return offset < ((le64_to_cpu(sbp->s_nsegments) *
447 			  le32_to_cpu(sbp->s_blocks_per_segment)) <<
448 			 (le32_to_cpu(sbp->s_log_block_size) + 10));
449 }
450 
451 static void nilfs_release_super_block(struct the_nilfs *nilfs)
452 {
453 	int i;
454 
455 	for (i = 0; i < 2; i++) {
456 		if (nilfs->ns_sbp[i]) {
457 			brelse(nilfs->ns_sbh[i]);
458 			nilfs->ns_sbh[i] = NULL;
459 			nilfs->ns_sbp[i] = NULL;
460 		}
461 	}
462 }
463 
464 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
465 {
466 	brelse(nilfs->ns_sbh[0]);
467 	nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
468 	nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
469 	nilfs->ns_sbh[1] = NULL;
470 	nilfs->ns_sbp[1] = NULL;
471 }
472 
473 void nilfs_swap_super_block(struct the_nilfs *nilfs)
474 {
475 	struct buffer_head *tsbh = nilfs->ns_sbh[0];
476 	struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
477 
478 	nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
479 	nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
480 	nilfs->ns_sbh[1] = tsbh;
481 	nilfs->ns_sbp[1] = tsbp;
482 }
483 
484 static int nilfs_load_super_block(struct the_nilfs *nilfs,
485 				  struct super_block *sb, int blocksize,
486 				  struct nilfs_super_block **sbpp)
487 {
488 	struct nilfs_super_block **sbp = nilfs->ns_sbp;
489 	struct buffer_head **sbh = nilfs->ns_sbh;
490 	u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
491 	int valid[2], swp = 0;
492 
493 	sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
494 					&sbh[0]);
495 	sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
496 
497 	if (!sbp[0]) {
498 		if (!sbp[1]) {
499 			printk(KERN_ERR "NILFS: unable to read superblock\n");
500 			return -EIO;
501 		}
502 		printk(KERN_WARNING
503 		       "NILFS warning: unable to read primary superblock "
504 		       "(blocksize = %d)\n", blocksize);
505 	} else if (!sbp[1]) {
506 		printk(KERN_WARNING
507 		       "NILFS warning: unable to read secondary superblock "
508 		       "(blocksize = %d)\n", blocksize);
509 	}
510 
511 	/*
512 	 * Compare two super blocks and set 1 in swp if the secondary
513 	 * super block is valid and newer.  Otherwise, set 0 in swp.
514 	 */
515 	valid[0] = nilfs_valid_sb(sbp[0]);
516 	valid[1] = nilfs_valid_sb(sbp[1]);
517 	swp = valid[1] && (!valid[0] ||
518 			   le64_to_cpu(sbp[1]->s_last_cno) >
519 			   le64_to_cpu(sbp[0]->s_last_cno));
520 
521 	if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
522 		brelse(sbh[1]);
523 		sbh[1] = NULL;
524 		sbp[1] = NULL;
525 		valid[1] = 0;
526 		swp = 0;
527 	}
528 	if (!valid[swp]) {
529 		nilfs_release_super_block(nilfs);
530 		printk(KERN_ERR "NILFS: Can't find nilfs on dev %s.\n",
531 		       sb->s_id);
532 		return -EINVAL;
533 	}
534 
535 	if (!valid[!swp])
536 		printk(KERN_WARNING "NILFS warning: broken superblock. "
537 		       "using spare superblock (blocksize = %d).\n", blocksize);
538 	if (swp)
539 		nilfs_swap_super_block(nilfs);
540 
541 	nilfs->ns_sbwcount = 0;
542 	nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
543 	nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
544 	*sbpp = sbp[0];
545 	return 0;
546 }
547 
548 /**
549  * init_nilfs - initialize a NILFS instance.
550  * @nilfs: the_nilfs structure
551  * @sb: super block
552  * @data: mount options
553  *
554  * init_nilfs() performs common initialization per block device (e.g.
555  * reading the super block, getting disk layout information, initializing
556  * shared fields in the_nilfs).
557  *
558  * Return Value: On success, 0 is returned. On error, a negative error
559  * code is returned.
560  */
561 int init_nilfs(struct the_nilfs *nilfs, struct super_block *sb, char *data)
562 {
563 	struct nilfs_super_block *sbp;
564 	int blocksize;
565 	int err;
566 
567 	down_write(&nilfs->ns_sem);
568 
569 	blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
570 	if (!blocksize) {
571 		printk(KERN_ERR "NILFS: unable to set blocksize\n");
572 		err = -EINVAL;
573 		goto out;
574 	}
575 	err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
576 	if (err)
577 		goto out;
578 
579 	err = nilfs_store_magic_and_option(sb, sbp, data);
580 	if (err)
581 		goto failed_sbh;
582 
583 	err = nilfs_check_feature_compatibility(sb, sbp);
584 	if (err)
585 		goto failed_sbh;
586 
587 	blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
588 	if (blocksize < NILFS_MIN_BLOCK_SIZE ||
589 	    blocksize > NILFS_MAX_BLOCK_SIZE) {
590 		printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
591 		       "filesystem blocksize %d\n", blocksize);
592 		err = -EINVAL;
593 		goto failed_sbh;
594 	}
595 	if (sb->s_blocksize != blocksize) {
596 		int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
597 
598 		if (blocksize < hw_blocksize) {
599 			printk(KERN_ERR
600 			       "NILFS: blocksize %d too small for device "
601 			       "(sector-size = %d).\n",
602 			       blocksize, hw_blocksize);
603 			err = -EINVAL;
604 			goto failed_sbh;
605 		}
606 		nilfs_release_super_block(nilfs);
607 		sb_set_blocksize(sb, blocksize);
608 
609 		err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
610 		if (err)
611 			goto out;
612 			/* not failed_sbh; sbh is released automatically
613 			   when reloading fails. */
614 	}
615 	nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
616 	nilfs->ns_blocksize = blocksize;
617 
618 	get_random_bytes(&nilfs->ns_next_generation,
619 			 sizeof(nilfs->ns_next_generation));
620 
621 	err = nilfs_store_disk_layout(nilfs, sbp);
622 	if (err)
623 		goto failed_sbh;
624 
625 	sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
626 
627 	nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
628 
629 	err = nilfs_store_log_cursor(nilfs, sbp);
630 	if (err)
631 		goto failed_sbh;
632 
633 	set_nilfs_init(nilfs);
634 	err = 0;
635  out:
636 	up_write(&nilfs->ns_sem);
637 	return err;
638 
639  failed_sbh:
640 	nilfs_release_super_block(nilfs);
641 	goto out;
642 }
643 
644 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
645 			    size_t nsegs)
646 {
647 	sector_t seg_start, seg_end;
648 	sector_t start = 0, nblocks = 0;
649 	unsigned int sects_per_block;
650 	__u64 *sn;
651 	int ret = 0;
652 
653 	sects_per_block = (1 << nilfs->ns_blocksize_bits) /
654 		bdev_logical_block_size(nilfs->ns_bdev);
655 	for (sn = segnump; sn < segnump + nsegs; sn++) {
656 		nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
657 
658 		if (!nblocks) {
659 			start = seg_start;
660 			nblocks = seg_end - seg_start + 1;
661 		} else if (start + nblocks == seg_start) {
662 			nblocks += seg_end - seg_start + 1;
663 		} else {
664 			ret = blkdev_issue_discard(nilfs->ns_bdev,
665 						   start * sects_per_block,
666 						   nblocks * sects_per_block,
667 						   GFP_NOFS, 0);
668 			if (ret < 0)
669 				return ret;
670 			nblocks = 0;
671 		}
672 	}
673 	if (nblocks)
674 		ret = blkdev_issue_discard(nilfs->ns_bdev,
675 					   start * sects_per_block,
676 					   nblocks * sects_per_block,
677 					   GFP_NOFS, 0);
678 	return ret;
679 }
680 
681 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
682 {
683 	unsigned long ncleansegs;
684 
685 	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
686 	ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
687 	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
688 	*nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
689 	return 0;
690 }
691 
692 int nilfs_near_disk_full(struct the_nilfs *nilfs)
693 {
694 	unsigned long ncleansegs, nincsegs;
695 
696 	ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
697 	nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
698 		nilfs->ns_blocks_per_segment + 1;
699 
700 	return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
701 }
702 
703 struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
704 {
705 	struct rb_node *n;
706 	struct nilfs_root *root;
707 
708 	spin_lock(&nilfs->ns_cptree_lock);
709 	n = nilfs->ns_cptree.rb_node;
710 	while (n) {
711 		root = rb_entry(n, struct nilfs_root, rb_node);
712 
713 		if (cno < root->cno) {
714 			n = n->rb_left;
715 		} else if (cno > root->cno) {
716 			n = n->rb_right;
717 		} else {
718 			atomic_inc(&root->count);
719 			spin_unlock(&nilfs->ns_cptree_lock);
720 			return root;
721 		}
722 	}
723 	spin_unlock(&nilfs->ns_cptree_lock);
724 
725 	return NULL;
726 }
727 
728 struct nilfs_root *
729 nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
730 {
731 	struct rb_node **p, *parent;
732 	struct nilfs_root *root, *new;
733 
734 	root = nilfs_lookup_root(nilfs, cno);
735 	if (root)
736 		return root;
737 
738 	new = kmalloc(sizeof(*root), GFP_KERNEL);
739 	if (!new)
740 		return NULL;
741 
742 	spin_lock(&nilfs->ns_cptree_lock);
743 
744 	p = &nilfs->ns_cptree.rb_node;
745 	parent = NULL;
746 
747 	while (*p) {
748 		parent = *p;
749 		root = rb_entry(parent, struct nilfs_root, rb_node);
750 
751 		if (cno < root->cno) {
752 			p = &(*p)->rb_left;
753 		} else if (cno > root->cno) {
754 			p = &(*p)->rb_right;
755 		} else {
756 			atomic_inc(&root->count);
757 			spin_unlock(&nilfs->ns_cptree_lock);
758 			kfree(new);
759 			return root;
760 		}
761 	}
762 
763 	new->cno = cno;
764 	new->ifile = NULL;
765 	new->nilfs = nilfs;
766 	atomic_set(&new->count, 1);
767 	atomic64_set(&new->inodes_count, 0);
768 	atomic64_set(&new->blocks_count, 0);
769 
770 	rb_link_node(&new->rb_node, parent, p);
771 	rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
772 
773 	spin_unlock(&nilfs->ns_cptree_lock);
774 
775 	return new;
776 }
777 
778 void nilfs_put_root(struct nilfs_root *root)
779 {
780 	if (atomic_dec_and_test(&root->count)) {
781 		struct the_nilfs *nilfs = root->nilfs;
782 
783 		spin_lock(&nilfs->ns_cptree_lock);
784 		rb_erase(&root->rb_node, &nilfs->ns_cptree);
785 		spin_unlock(&nilfs->ns_cptree_lock);
786 		if (root->ifile)
787 			iput(root->ifile);
788 
789 		kfree(root);
790 	}
791 }
792