xref: /openbmc/linux/fs/nilfs2/the_nilfs.c (revision 97e1a530)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * the_nilfs shared structure.
4  *
5  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6  *
7  * Written by Ryusuke Konishi.
8  *
9  */
10 
11 #include <linux/buffer_head.h>
12 #include <linux/slab.h>
13 #include <linux/blkdev.h>
14 #include <linux/backing-dev.h>
15 #include <linux/random.h>
16 #include <linux/crc32.h>
17 #include "nilfs.h"
18 #include "segment.h"
19 #include "alloc.h"
20 #include "cpfile.h"
21 #include "sufile.h"
22 #include "dat.h"
23 #include "segbuf.h"
24 
25 
26 static int nilfs_valid_sb(struct nilfs_super_block *sbp);
27 
28 void nilfs_set_last_segment(struct the_nilfs *nilfs,
29 			    sector_t start_blocknr, u64 seq, __u64 cno)
30 {
31 	spin_lock(&nilfs->ns_last_segment_lock);
32 	nilfs->ns_last_pseg = start_blocknr;
33 	nilfs->ns_last_seq = seq;
34 	nilfs->ns_last_cno = cno;
35 
36 	if (!nilfs_sb_dirty(nilfs)) {
37 		if (nilfs->ns_prev_seq == nilfs->ns_last_seq)
38 			goto stay_cursor;
39 
40 		set_nilfs_sb_dirty(nilfs);
41 	}
42 	nilfs->ns_prev_seq = nilfs->ns_last_seq;
43 
44  stay_cursor:
45 	spin_unlock(&nilfs->ns_last_segment_lock);
46 }
47 
48 /**
49  * alloc_nilfs - allocate a nilfs object
50  * @sb: super block instance
51  *
52  * Return Value: On success, pointer to the_nilfs is returned.
53  * On error, NULL is returned.
54  */
55 struct the_nilfs *alloc_nilfs(struct super_block *sb)
56 {
57 	struct the_nilfs *nilfs;
58 
59 	nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
60 	if (!nilfs)
61 		return NULL;
62 
63 	nilfs->ns_sb = sb;
64 	nilfs->ns_bdev = sb->s_bdev;
65 	atomic_set(&nilfs->ns_ndirtyblks, 0);
66 	init_rwsem(&nilfs->ns_sem);
67 	mutex_init(&nilfs->ns_snapshot_mount_mutex);
68 	INIT_LIST_HEAD(&nilfs->ns_dirty_files);
69 	INIT_LIST_HEAD(&nilfs->ns_gc_inodes);
70 	spin_lock_init(&nilfs->ns_inode_lock);
71 	spin_lock_init(&nilfs->ns_next_gen_lock);
72 	spin_lock_init(&nilfs->ns_last_segment_lock);
73 	nilfs->ns_cptree = RB_ROOT;
74 	spin_lock_init(&nilfs->ns_cptree_lock);
75 	init_rwsem(&nilfs->ns_segctor_sem);
76 	nilfs->ns_sb_update_freq = NILFS_SB_FREQ;
77 
78 	return nilfs;
79 }
80 
81 /**
82  * destroy_nilfs - destroy nilfs object
83  * @nilfs: nilfs object to be released
84  */
85 void destroy_nilfs(struct the_nilfs *nilfs)
86 {
87 	might_sleep();
88 	if (nilfs_init(nilfs)) {
89 		nilfs_sysfs_delete_device_group(nilfs);
90 		brelse(nilfs->ns_sbh[0]);
91 		brelse(nilfs->ns_sbh[1]);
92 	}
93 	kfree(nilfs);
94 }
95 
96 static int nilfs_load_super_root(struct the_nilfs *nilfs,
97 				 struct super_block *sb, sector_t sr_block)
98 {
99 	struct buffer_head *bh_sr;
100 	struct nilfs_super_root *raw_sr;
101 	struct nilfs_super_block **sbp = nilfs->ns_sbp;
102 	struct nilfs_inode *rawi;
103 	unsigned int dat_entry_size, segment_usage_size, checkpoint_size;
104 	unsigned int inode_size;
105 	int err;
106 
107 	err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
108 	if (unlikely(err))
109 		return err;
110 
111 	down_read(&nilfs->ns_sem);
112 	dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
113 	checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
114 	segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
115 	up_read(&nilfs->ns_sem);
116 
117 	inode_size = nilfs->ns_inode_size;
118 
119 	rawi = (void *)bh_sr->b_data + NILFS_SR_DAT_OFFSET(inode_size);
120 	err = nilfs_dat_read(sb, dat_entry_size, rawi, &nilfs->ns_dat);
121 	if (err)
122 		goto failed;
123 
124 	rawi = (void *)bh_sr->b_data + NILFS_SR_CPFILE_OFFSET(inode_size);
125 	err = nilfs_cpfile_read(sb, checkpoint_size, rawi, &nilfs->ns_cpfile);
126 	if (err)
127 		goto failed_dat;
128 
129 	rawi = (void *)bh_sr->b_data + NILFS_SR_SUFILE_OFFSET(inode_size);
130 	err = nilfs_sufile_read(sb, segment_usage_size, rawi,
131 				&nilfs->ns_sufile);
132 	if (err)
133 		goto failed_cpfile;
134 
135 	raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
136 	nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
137 
138  failed:
139 	brelse(bh_sr);
140 	return err;
141 
142  failed_cpfile:
143 	iput(nilfs->ns_cpfile);
144 
145  failed_dat:
146 	iput(nilfs->ns_dat);
147 	goto failed;
148 }
149 
150 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
151 {
152 	memset(ri, 0, sizeof(*ri));
153 	INIT_LIST_HEAD(&ri->ri_used_segments);
154 }
155 
156 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
157 {
158 	nilfs_dispose_segment_list(&ri->ri_used_segments);
159 }
160 
161 /**
162  * nilfs_store_log_cursor - load log cursor from a super block
163  * @nilfs: nilfs object
164  * @sbp: buffer storing super block to be read
165  *
166  * nilfs_store_log_cursor() reads the last position of the log
167  * containing a super root from a given super block, and initializes
168  * relevant information on the nilfs object preparatory for log
169  * scanning and recovery.
170  */
171 static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
172 				  struct nilfs_super_block *sbp)
173 {
174 	int ret = 0;
175 
176 	nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
177 	nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
178 	nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
179 
180 	nilfs->ns_prev_seq = nilfs->ns_last_seq;
181 	nilfs->ns_seg_seq = nilfs->ns_last_seq;
182 	nilfs->ns_segnum =
183 		nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
184 	nilfs->ns_cno = nilfs->ns_last_cno + 1;
185 	if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
186 		nilfs_err(nilfs->ns_sb,
187 			  "pointed segment number is out of range: segnum=%llu, nsegments=%lu",
188 			  (unsigned long long)nilfs->ns_segnum,
189 			  nilfs->ns_nsegments);
190 		ret = -EINVAL;
191 	}
192 	return ret;
193 }
194 
195 /**
196  * load_nilfs - load and recover the nilfs
197  * @nilfs: the_nilfs structure to be released
198  * @sb: super block instance used to recover past segment
199  *
200  * load_nilfs() searches and load the latest super root,
201  * attaches the last segment, and does recovery if needed.
202  * The caller must call this exclusively for simultaneous mounts.
203  */
204 int load_nilfs(struct the_nilfs *nilfs, struct super_block *sb)
205 {
206 	struct nilfs_recovery_info ri;
207 	unsigned int s_flags = sb->s_flags;
208 	int really_read_only = bdev_read_only(nilfs->ns_bdev);
209 	int valid_fs = nilfs_valid_fs(nilfs);
210 	int err;
211 
212 	if (!valid_fs) {
213 		nilfs_warn(sb, "mounting unchecked fs");
214 		if (s_flags & SB_RDONLY) {
215 			nilfs_info(sb,
216 				   "recovery required for readonly filesystem");
217 			nilfs_info(sb,
218 				   "write access will be enabled during recovery");
219 		}
220 	}
221 
222 	nilfs_init_recovery_info(&ri);
223 
224 	err = nilfs_search_super_root(nilfs, &ri);
225 	if (unlikely(err)) {
226 		struct nilfs_super_block **sbp = nilfs->ns_sbp;
227 		int blocksize;
228 
229 		if (err != -EINVAL)
230 			goto scan_error;
231 
232 		if (!nilfs_valid_sb(sbp[1])) {
233 			nilfs_warn(sb,
234 				   "unable to fall back to spare super block");
235 			goto scan_error;
236 		}
237 		nilfs_info(sb, "trying rollback from an earlier position");
238 
239 		/*
240 		 * restore super block with its spare and reconfigure
241 		 * relevant states of the nilfs object.
242 		 */
243 		memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
244 		nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
245 		nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
246 
247 		/* verify consistency between two super blocks */
248 		blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
249 		if (blocksize != nilfs->ns_blocksize) {
250 			nilfs_warn(sb,
251 				   "blocksize differs between two super blocks (%d != %d)",
252 				   blocksize, nilfs->ns_blocksize);
253 			goto scan_error;
254 		}
255 
256 		err = nilfs_store_log_cursor(nilfs, sbp[0]);
257 		if (err)
258 			goto scan_error;
259 
260 		/* drop clean flag to allow roll-forward and recovery */
261 		nilfs->ns_mount_state &= ~NILFS_VALID_FS;
262 		valid_fs = 0;
263 
264 		err = nilfs_search_super_root(nilfs, &ri);
265 		if (err)
266 			goto scan_error;
267 	}
268 
269 	err = nilfs_load_super_root(nilfs, sb, ri.ri_super_root);
270 	if (unlikely(err)) {
271 		nilfs_err(sb, "error %d while loading super root", err);
272 		goto failed;
273 	}
274 
275 	if (valid_fs)
276 		goto skip_recovery;
277 
278 	if (s_flags & SB_RDONLY) {
279 		__u64 features;
280 
281 		if (nilfs_test_opt(nilfs, NORECOVERY)) {
282 			nilfs_info(sb,
283 				   "norecovery option specified, skipping roll-forward recovery");
284 			goto skip_recovery;
285 		}
286 		features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
287 			~NILFS_FEATURE_COMPAT_RO_SUPP;
288 		if (features) {
289 			nilfs_err(sb,
290 				  "couldn't proceed with recovery because of unsupported optional features (%llx)",
291 				  (unsigned long long)features);
292 			err = -EROFS;
293 			goto failed_unload;
294 		}
295 		if (really_read_only) {
296 			nilfs_err(sb,
297 				  "write access unavailable, cannot proceed");
298 			err = -EROFS;
299 			goto failed_unload;
300 		}
301 		sb->s_flags &= ~SB_RDONLY;
302 	} else if (nilfs_test_opt(nilfs, NORECOVERY)) {
303 		nilfs_err(sb,
304 			  "recovery cancelled because norecovery option was specified for a read/write mount");
305 		err = -EINVAL;
306 		goto failed_unload;
307 	}
308 
309 	err = nilfs_salvage_orphan_logs(nilfs, sb, &ri);
310 	if (err)
311 		goto failed_unload;
312 
313 	down_write(&nilfs->ns_sem);
314 	nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
315 	err = nilfs_cleanup_super(sb);
316 	up_write(&nilfs->ns_sem);
317 
318 	if (err) {
319 		nilfs_err(sb,
320 			  "error %d updating super block. recovery unfinished.",
321 			  err);
322 		goto failed_unload;
323 	}
324 	nilfs_info(sb, "recovery complete");
325 
326  skip_recovery:
327 	nilfs_clear_recovery_info(&ri);
328 	sb->s_flags = s_flags;
329 	return 0;
330 
331  scan_error:
332 	nilfs_err(sb, "error %d while searching super root", err);
333 	goto failed;
334 
335  failed_unload:
336 	iput(nilfs->ns_cpfile);
337 	iput(nilfs->ns_sufile);
338 	iput(nilfs->ns_dat);
339 
340  failed:
341 	nilfs_clear_recovery_info(&ri);
342 	sb->s_flags = s_flags;
343 	return err;
344 }
345 
346 static unsigned long long nilfs_max_size(unsigned int blkbits)
347 {
348 	unsigned int max_bits;
349 	unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
350 
351 	max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
352 	if (max_bits < 64)
353 		res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
354 	return res;
355 }
356 
357 /**
358  * nilfs_nrsvsegs - calculate the number of reserved segments
359  * @nilfs: nilfs object
360  * @nsegs: total number of segments
361  */
362 unsigned long nilfs_nrsvsegs(struct the_nilfs *nilfs, unsigned long nsegs)
363 {
364 	return max_t(unsigned long, NILFS_MIN_NRSVSEGS,
365 		     DIV_ROUND_UP(nsegs * nilfs->ns_r_segments_percentage,
366 				  100));
367 }
368 
369 void nilfs_set_nsegments(struct the_nilfs *nilfs, unsigned long nsegs)
370 {
371 	nilfs->ns_nsegments = nsegs;
372 	nilfs->ns_nrsvsegs = nilfs_nrsvsegs(nilfs, nsegs);
373 }
374 
375 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
376 				   struct nilfs_super_block *sbp)
377 {
378 	if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
379 		nilfs_err(nilfs->ns_sb,
380 			  "unsupported revision (superblock rev.=%d.%d, current rev.=%d.%d). Please check the version of mkfs.nilfs(2).",
381 			  le32_to_cpu(sbp->s_rev_level),
382 			  le16_to_cpu(sbp->s_minor_rev_level),
383 			  NILFS_CURRENT_REV, NILFS_MINOR_REV);
384 		return -EINVAL;
385 	}
386 	nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
387 	if (nilfs->ns_sbsize > BLOCK_SIZE)
388 		return -EINVAL;
389 
390 	nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
391 	if (nilfs->ns_inode_size > nilfs->ns_blocksize) {
392 		nilfs_err(nilfs->ns_sb, "too large inode size: %d bytes",
393 			  nilfs->ns_inode_size);
394 		return -EINVAL;
395 	} else if (nilfs->ns_inode_size < NILFS_MIN_INODE_SIZE) {
396 		nilfs_err(nilfs->ns_sb, "too small inode size: %d bytes",
397 			  nilfs->ns_inode_size);
398 		return -EINVAL;
399 	}
400 
401 	nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
402 
403 	nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
404 	if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
405 		nilfs_err(nilfs->ns_sb, "too short segment: %lu blocks",
406 			  nilfs->ns_blocks_per_segment);
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 		nilfs_err(nilfs->ns_sb,
416 			  "invalid reserved segments percentage: %lu",
417 			  nilfs->ns_r_segments_percentage);
418 		return -EINVAL;
419 	}
420 
421 	nilfs_set_nsegments(nilfs, le64_to_cpu(sbp->s_nsegments));
422 	nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
423 	return 0;
424 }
425 
426 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
427 {
428 	static unsigned char sum[4];
429 	const int sumoff = offsetof(struct nilfs_super_block, s_sum);
430 	size_t bytes;
431 	u32 crc;
432 
433 	if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
434 		return 0;
435 	bytes = le16_to_cpu(sbp->s_bytes);
436 	if (bytes < sumoff + 4 || bytes > BLOCK_SIZE)
437 		return 0;
438 	crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
439 		       sumoff);
440 	crc = crc32_le(crc, sum, 4);
441 	crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
442 		       bytes - sumoff - 4);
443 	return crc == le32_to_cpu(sbp->s_sum);
444 }
445 
446 static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
447 {
448 	return offset < ((le64_to_cpu(sbp->s_nsegments) *
449 			  le32_to_cpu(sbp->s_blocks_per_segment)) <<
450 			 (le32_to_cpu(sbp->s_log_block_size) + 10));
451 }
452 
453 static void nilfs_release_super_block(struct the_nilfs *nilfs)
454 {
455 	int i;
456 
457 	for (i = 0; i < 2; i++) {
458 		if (nilfs->ns_sbp[i]) {
459 			brelse(nilfs->ns_sbh[i]);
460 			nilfs->ns_sbh[i] = NULL;
461 			nilfs->ns_sbp[i] = NULL;
462 		}
463 	}
464 }
465 
466 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
467 {
468 	brelse(nilfs->ns_sbh[0]);
469 	nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
470 	nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
471 	nilfs->ns_sbh[1] = NULL;
472 	nilfs->ns_sbp[1] = NULL;
473 }
474 
475 void nilfs_swap_super_block(struct the_nilfs *nilfs)
476 {
477 	struct buffer_head *tsbh = nilfs->ns_sbh[0];
478 	struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
479 
480 	nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
481 	nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
482 	nilfs->ns_sbh[1] = tsbh;
483 	nilfs->ns_sbp[1] = tsbp;
484 }
485 
486 static int nilfs_load_super_block(struct the_nilfs *nilfs,
487 				  struct super_block *sb, int blocksize,
488 				  struct nilfs_super_block **sbpp)
489 {
490 	struct nilfs_super_block **sbp = nilfs->ns_sbp;
491 	struct buffer_head **sbh = nilfs->ns_sbh;
492 	u64 sb2off = NILFS_SB2_OFFSET_BYTES(bdev_nr_bytes(nilfs->ns_bdev));
493 	int valid[2], swp = 0;
494 
495 	sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
496 					&sbh[0]);
497 	sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
498 
499 	if (!sbp[0]) {
500 		if (!sbp[1]) {
501 			nilfs_err(sb, "unable to read superblock");
502 			return -EIO;
503 		}
504 		nilfs_warn(sb,
505 			   "unable to read primary superblock (blocksize = %d)",
506 			   blocksize);
507 	} else if (!sbp[1]) {
508 		nilfs_warn(sb,
509 			   "unable to read secondary superblock (blocksize = %d)",
510 			   blocksize);
511 	}
512 
513 	/*
514 	 * Compare two super blocks and set 1 in swp if the secondary
515 	 * super block is valid and newer.  Otherwise, set 0 in swp.
516 	 */
517 	valid[0] = nilfs_valid_sb(sbp[0]);
518 	valid[1] = nilfs_valid_sb(sbp[1]);
519 	swp = valid[1] && (!valid[0] ||
520 			   le64_to_cpu(sbp[1]->s_last_cno) >
521 			   le64_to_cpu(sbp[0]->s_last_cno));
522 
523 	if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
524 		brelse(sbh[1]);
525 		sbh[1] = NULL;
526 		sbp[1] = NULL;
527 		valid[1] = 0;
528 		swp = 0;
529 	}
530 	if (!valid[swp]) {
531 		nilfs_release_super_block(nilfs);
532 		nilfs_err(sb, "couldn't find nilfs on the device");
533 		return -EINVAL;
534 	}
535 
536 	if (!valid[!swp])
537 		nilfs_warn(sb,
538 			   "broken superblock, retrying with spare superblock (blocksize = %d)",
539 			   blocksize);
540 	if (swp)
541 		nilfs_swap_super_block(nilfs);
542 
543 	nilfs->ns_sbwcount = 0;
544 	nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
545 	nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
546 	*sbpp = sbp[0];
547 	return 0;
548 }
549 
550 /**
551  * init_nilfs - initialize a NILFS instance.
552  * @nilfs: the_nilfs structure
553  * @sb: super block
554  * @data: mount options
555  *
556  * init_nilfs() performs common initialization per block device (e.g.
557  * reading the super block, getting disk layout information, initializing
558  * shared fields in the_nilfs).
559  *
560  * Return Value: On success, 0 is returned. On error, a negative error
561  * code is returned.
562  */
563 int init_nilfs(struct the_nilfs *nilfs, struct super_block *sb, char *data)
564 {
565 	struct nilfs_super_block *sbp;
566 	int blocksize;
567 	int err;
568 
569 	down_write(&nilfs->ns_sem);
570 
571 	blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
572 	if (!blocksize) {
573 		nilfs_err(sb, "unable to set blocksize");
574 		err = -EINVAL;
575 		goto out;
576 	}
577 	err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
578 	if (err)
579 		goto out;
580 
581 	err = nilfs_store_magic_and_option(sb, sbp, data);
582 	if (err)
583 		goto failed_sbh;
584 
585 	err = nilfs_check_feature_compatibility(sb, sbp);
586 	if (err)
587 		goto failed_sbh;
588 
589 	blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
590 	if (blocksize < NILFS_MIN_BLOCK_SIZE ||
591 	    blocksize > NILFS_MAX_BLOCK_SIZE) {
592 		nilfs_err(sb,
593 			  "couldn't mount because of unsupported filesystem blocksize %d",
594 			  blocksize);
595 		err = -EINVAL;
596 		goto failed_sbh;
597 	}
598 	if (sb->s_blocksize != blocksize) {
599 		int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
600 
601 		if (blocksize < hw_blocksize) {
602 			nilfs_err(sb,
603 				  "blocksize %d too small for device (sector-size = %d)",
604 				  blocksize, hw_blocksize);
605 			err = -EINVAL;
606 			goto failed_sbh;
607 		}
608 		nilfs_release_super_block(nilfs);
609 		sb_set_blocksize(sb, blocksize);
610 
611 		err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
612 		if (err)
613 			goto out;
614 			/*
615 			 * Not to failed_sbh; sbh is released automatically
616 			 * when reloading fails.
617 			 */
618 	}
619 	nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
620 	nilfs->ns_blocksize = blocksize;
621 
622 	get_random_bytes(&nilfs->ns_next_generation,
623 			 sizeof(nilfs->ns_next_generation));
624 
625 	err = nilfs_store_disk_layout(nilfs, sbp);
626 	if (err)
627 		goto failed_sbh;
628 
629 	sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
630 
631 	nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
632 
633 	err = nilfs_store_log_cursor(nilfs, sbp);
634 	if (err)
635 		goto failed_sbh;
636 
637 	err = nilfs_sysfs_create_device_group(sb);
638 	if (err)
639 		goto failed_sbh;
640 
641 	set_nilfs_init(nilfs);
642 	err = 0;
643  out:
644 	up_write(&nilfs->ns_sem);
645 	return err;
646 
647  failed_sbh:
648 	nilfs_release_super_block(nilfs);
649 	goto out;
650 }
651 
652 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
653 			    size_t nsegs)
654 {
655 	sector_t seg_start, seg_end;
656 	sector_t start = 0, nblocks = 0;
657 	unsigned int sects_per_block;
658 	__u64 *sn;
659 	int ret = 0;
660 
661 	sects_per_block = (1 << nilfs->ns_blocksize_bits) /
662 		bdev_logical_block_size(nilfs->ns_bdev);
663 	for (sn = segnump; sn < segnump + nsegs; sn++) {
664 		nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
665 
666 		if (!nblocks) {
667 			start = seg_start;
668 			nblocks = seg_end - seg_start + 1;
669 		} else if (start + nblocks == seg_start) {
670 			nblocks += seg_end - seg_start + 1;
671 		} else {
672 			ret = blkdev_issue_discard(nilfs->ns_bdev,
673 						   start * sects_per_block,
674 						   nblocks * sects_per_block,
675 						   GFP_NOFS);
676 			if (ret < 0)
677 				return ret;
678 			nblocks = 0;
679 		}
680 	}
681 	if (nblocks)
682 		ret = blkdev_issue_discard(nilfs->ns_bdev,
683 					   start * sects_per_block,
684 					   nblocks * sects_per_block,
685 					   GFP_NOFS);
686 	return ret;
687 }
688 
689 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
690 {
691 	unsigned long ncleansegs;
692 
693 	ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
694 	*nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
695 	return 0;
696 }
697 
698 int nilfs_near_disk_full(struct the_nilfs *nilfs)
699 {
700 	unsigned long ncleansegs, nincsegs;
701 
702 	ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
703 	nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
704 		nilfs->ns_blocks_per_segment + 1;
705 
706 	return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
707 }
708 
709 struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
710 {
711 	struct rb_node *n;
712 	struct nilfs_root *root;
713 
714 	spin_lock(&nilfs->ns_cptree_lock);
715 	n = nilfs->ns_cptree.rb_node;
716 	while (n) {
717 		root = rb_entry(n, struct nilfs_root, rb_node);
718 
719 		if (cno < root->cno) {
720 			n = n->rb_left;
721 		} else if (cno > root->cno) {
722 			n = n->rb_right;
723 		} else {
724 			refcount_inc(&root->count);
725 			spin_unlock(&nilfs->ns_cptree_lock);
726 			return root;
727 		}
728 	}
729 	spin_unlock(&nilfs->ns_cptree_lock);
730 
731 	return NULL;
732 }
733 
734 struct nilfs_root *
735 nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
736 {
737 	struct rb_node **p, *parent;
738 	struct nilfs_root *root, *new;
739 	int err;
740 
741 	root = nilfs_lookup_root(nilfs, cno);
742 	if (root)
743 		return root;
744 
745 	new = kzalloc(sizeof(*root), GFP_KERNEL);
746 	if (!new)
747 		return NULL;
748 
749 	spin_lock(&nilfs->ns_cptree_lock);
750 
751 	p = &nilfs->ns_cptree.rb_node;
752 	parent = NULL;
753 
754 	while (*p) {
755 		parent = *p;
756 		root = rb_entry(parent, struct nilfs_root, rb_node);
757 
758 		if (cno < root->cno) {
759 			p = &(*p)->rb_left;
760 		} else if (cno > root->cno) {
761 			p = &(*p)->rb_right;
762 		} else {
763 			refcount_inc(&root->count);
764 			spin_unlock(&nilfs->ns_cptree_lock);
765 			kfree(new);
766 			return root;
767 		}
768 	}
769 
770 	new->cno = cno;
771 	new->ifile = NULL;
772 	new->nilfs = nilfs;
773 	refcount_set(&new->count, 1);
774 	atomic64_set(&new->inodes_count, 0);
775 	atomic64_set(&new->blocks_count, 0);
776 
777 	rb_link_node(&new->rb_node, parent, p);
778 	rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
779 
780 	spin_unlock(&nilfs->ns_cptree_lock);
781 
782 	err = nilfs_sysfs_create_snapshot_group(new);
783 	if (err) {
784 		kfree(new);
785 		new = NULL;
786 	}
787 
788 	return new;
789 }
790 
791 void nilfs_put_root(struct nilfs_root *root)
792 {
793 	struct the_nilfs *nilfs = root->nilfs;
794 
795 	if (refcount_dec_and_lock(&root->count, &nilfs->ns_cptree_lock)) {
796 		rb_erase(&root->rb_node, &nilfs->ns_cptree);
797 		spin_unlock(&nilfs->ns_cptree_lock);
798 
799 		nilfs_sysfs_delete_snapshot_group(root);
800 		iput(root->ifile);
801 
802 		kfree(root);
803 	}
804 }
805