xref: /linux/fs/nilfs2/the_nilfs.c (revision bd628c1bed7902ec1f24ba0fe70758949146abbe)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * the_nilfs.c - 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_msg(nilfs->ns_sb, KERN_ERR,
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 isntance 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_msg(sb, KERN_WARNING, "mounting unchecked fs");
214 		if (s_flags & SB_RDONLY) {
215 			nilfs_msg(sb, KERN_INFO,
216 				  "recovery required for readonly filesystem");
217 			nilfs_msg(sb, KERN_INFO,
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_msg(sb, KERN_WARNING,
234 				  "unable to fall back to spare super block");
235 			goto scan_error;
236 		}
237 		nilfs_msg(sb, KERN_INFO,
238 			  "trying rollback from an earlier position");
239 
240 		/*
241 		 * restore super block with its spare and reconfigure
242 		 * relevant states of the nilfs object.
243 		 */
244 		memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
245 		nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
246 		nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
247 
248 		/* verify consistency between two super blocks */
249 		blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
250 		if (blocksize != nilfs->ns_blocksize) {
251 			nilfs_msg(sb, KERN_WARNING,
252 				  "blocksize differs between two super blocks (%d != %d)",
253 				  blocksize, nilfs->ns_blocksize);
254 			goto scan_error;
255 		}
256 
257 		err = nilfs_store_log_cursor(nilfs, sbp[0]);
258 		if (err)
259 			goto scan_error;
260 
261 		/* drop clean flag to allow roll-forward and recovery */
262 		nilfs->ns_mount_state &= ~NILFS_VALID_FS;
263 		valid_fs = 0;
264 
265 		err = nilfs_search_super_root(nilfs, &ri);
266 		if (err)
267 			goto scan_error;
268 	}
269 
270 	err = nilfs_load_super_root(nilfs, sb, ri.ri_super_root);
271 	if (unlikely(err)) {
272 		nilfs_msg(sb, KERN_ERR, "error %d while loading super root",
273 			  err);
274 		goto failed;
275 	}
276 
277 	if (valid_fs)
278 		goto skip_recovery;
279 
280 	if (s_flags & SB_RDONLY) {
281 		__u64 features;
282 
283 		if (nilfs_test_opt(nilfs, NORECOVERY)) {
284 			nilfs_msg(sb, KERN_INFO,
285 				  "norecovery option specified, skipping roll-forward recovery");
286 			goto skip_recovery;
287 		}
288 		features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
289 			~NILFS_FEATURE_COMPAT_RO_SUPP;
290 		if (features) {
291 			nilfs_msg(sb, KERN_ERR,
292 				  "couldn't proceed with recovery because of unsupported optional features (%llx)",
293 				  (unsigned long long)features);
294 			err = -EROFS;
295 			goto failed_unload;
296 		}
297 		if (really_read_only) {
298 			nilfs_msg(sb, KERN_ERR,
299 				  "write access unavailable, cannot proceed");
300 			err = -EROFS;
301 			goto failed_unload;
302 		}
303 		sb->s_flags &= ~SB_RDONLY;
304 	} else if (nilfs_test_opt(nilfs, NORECOVERY)) {
305 		nilfs_msg(sb, KERN_ERR,
306 			  "recovery cancelled because norecovery option was specified for a read/write mount");
307 		err = -EINVAL;
308 		goto failed_unload;
309 	}
310 
311 	err = nilfs_salvage_orphan_logs(nilfs, sb, &ri);
312 	if (err)
313 		goto failed_unload;
314 
315 	down_write(&nilfs->ns_sem);
316 	nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
317 	err = nilfs_cleanup_super(sb);
318 	up_write(&nilfs->ns_sem);
319 
320 	if (err) {
321 		nilfs_msg(sb, KERN_ERR,
322 			  "error %d updating super block. recovery unfinished.",
323 			  err);
324 		goto failed_unload;
325 	}
326 	nilfs_msg(sb, KERN_INFO, "recovery complete");
327 
328  skip_recovery:
329 	nilfs_clear_recovery_info(&ri);
330 	sb->s_flags = s_flags;
331 	return 0;
332 
333  scan_error:
334 	nilfs_msg(sb, KERN_ERR, "error %d while searching super root", err);
335 	goto failed;
336 
337  failed_unload:
338 	iput(nilfs->ns_cpfile);
339 	iput(nilfs->ns_sufile);
340 	iput(nilfs->ns_dat);
341 
342  failed:
343 	nilfs_clear_recovery_info(&ri);
344 	sb->s_flags = s_flags;
345 	return err;
346 }
347 
348 static unsigned long long nilfs_max_size(unsigned int blkbits)
349 {
350 	unsigned int max_bits;
351 	unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
352 
353 	max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
354 	if (max_bits < 64)
355 		res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
356 	return res;
357 }
358 
359 /**
360  * nilfs_nrsvsegs - calculate the number of reserved segments
361  * @nilfs: nilfs object
362  * @nsegs: total number of segments
363  */
364 unsigned long nilfs_nrsvsegs(struct the_nilfs *nilfs, unsigned long nsegs)
365 {
366 	return max_t(unsigned long, NILFS_MIN_NRSVSEGS,
367 		     DIV_ROUND_UP(nsegs * nilfs->ns_r_segments_percentage,
368 				  100));
369 }
370 
371 void nilfs_set_nsegments(struct the_nilfs *nilfs, unsigned long nsegs)
372 {
373 	nilfs->ns_nsegments = nsegs;
374 	nilfs->ns_nrsvsegs = nilfs_nrsvsegs(nilfs, nsegs);
375 }
376 
377 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
378 				   struct nilfs_super_block *sbp)
379 {
380 	if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
381 		nilfs_msg(nilfs->ns_sb, KERN_ERR,
382 			  "unsupported revision (superblock rev.=%d.%d, current rev.=%d.%d). Please check the version of mkfs.nilfs(2).",
383 			  le32_to_cpu(sbp->s_rev_level),
384 			  le16_to_cpu(sbp->s_minor_rev_level),
385 			  NILFS_CURRENT_REV, NILFS_MINOR_REV);
386 		return -EINVAL;
387 	}
388 	nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
389 	if (nilfs->ns_sbsize > BLOCK_SIZE)
390 		return -EINVAL;
391 
392 	nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
393 	if (nilfs->ns_inode_size > nilfs->ns_blocksize) {
394 		nilfs_msg(nilfs->ns_sb, KERN_ERR,
395 			  "too large inode size: %d bytes",
396 			  nilfs->ns_inode_size);
397 		return -EINVAL;
398 	} else if (nilfs->ns_inode_size < NILFS_MIN_INODE_SIZE) {
399 		nilfs_msg(nilfs->ns_sb, KERN_ERR,
400 			  "too small inode size: %d bytes",
401 			  nilfs->ns_inode_size);
402 		return -EINVAL;
403 	}
404 
405 	nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
406 
407 	nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
408 	if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
409 		nilfs_msg(nilfs->ns_sb, KERN_ERR,
410 			  "too short segment: %lu blocks",
411 			  nilfs->ns_blocks_per_segment);
412 		return -EINVAL;
413 	}
414 
415 	nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
416 	nilfs->ns_r_segments_percentage =
417 		le32_to_cpu(sbp->s_r_segments_percentage);
418 	if (nilfs->ns_r_segments_percentage < 1 ||
419 	    nilfs->ns_r_segments_percentage > 99) {
420 		nilfs_msg(nilfs->ns_sb, KERN_ERR,
421 			  "invalid reserved segments percentage: %lu",
422 			  nilfs->ns_r_segments_percentage);
423 		return -EINVAL;
424 	}
425 
426 	nilfs_set_nsegments(nilfs, le64_to_cpu(sbp->s_nsegments));
427 	nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
428 	return 0;
429 }
430 
431 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
432 {
433 	static unsigned char sum[4];
434 	const int sumoff = offsetof(struct nilfs_super_block, s_sum);
435 	size_t bytes;
436 	u32 crc;
437 
438 	if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
439 		return 0;
440 	bytes = le16_to_cpu(sbp->s_bytes);
441 	if (bytes < sumoff + 4 || bytes > BLOCK_SIZE)
442 		return 0;
443 	crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
444 		       sumoff);
445 	crc = crc32_le(crc, sum, 4);
446 	crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
447 		       bytes - sumoff - 4);
448 	return crc == le32_to_cpu(sbp->s_sum);
449 }
450 
451 static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
452 {
453 	return offset < ((le64_to_cpu(sbp->s_nsegments) *
454 			  le32_to_cpu(sbp->s_blocks_per_segment)) <<
455 			 (le32_to_cpu(sbp->s_log_block_size) + 10));
456 }
457 
458 static void nilfs_release_super_block(struct the_nilfs *nilfs)
459 {
460 	int i;
461 
462 	for (i = 0; i < 2; i++) {
463 		if (nilfs->ns_sbp[i]) {
464 			brelse(nilfs->ns_sbh[i]);
465 			nilfs->ns_sbh[i] = NULL;
466 			nilfs->ns_sbp[i] = NULL;
467 		}
468 	}
469 }
470 
471 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
472 {
473 	brelse(nilfs->ns_sbh[0]);
474 	nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
475 	nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
476 	nilfs->ns_sbh[1] = NULL;
477 	nilfs->ns_sbp[1] = NULL;
478 }
479 
480 void nilfs_swap_super_block(struct the_nilfs *nilfs)
481 {
482 	struct buffer_head *tsbh = nilfs->ns_sbh[0];
483 	struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
484 
485 	nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
486 	nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
487 	nilfs->ns_sbh[1] = tsbh;
488 	nilfs->ns_sbp[1] = tsbp;
489 }
490 
491 static int nilfs_load_super_block(struct the_nilfs *nilfs,
492 				  struct super_block *sb, int blocksize,
493 				  struct nilfs_super_block **sbpp)
494 {
495 	struct nilfs_super_block **sbp = nilfs->ns_sbp;
496 	struct buffer_head **sbh = nilfs->ns_sbh;
497 	u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
498 	int valid[2], swp = 0;
499 
500 	sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
501 					&sbh[0]);
502 	sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
503 
504 	if (!sbp[0]) {
505 		if (!sbp[1]) {
506 			nilfs_msg(sb, KERN_ERR, "unable to read superblock");
507 			return -EIO;
508 		}
509 		nilfs_msg(sb, KERN_WARNING,
510 			  "unable to read primary superblock (blocksize = %d)",
511 			  blocksize);
512 	} else if (!sbp[1]) {
513 		nilfs_msg(sb, KERN_WARNING,
514 			  "unable to read secondary superblock (blocksize = %d)",
515 			  blocksize);
516 	}
517 
518 	/*
519 	 * Compare two super blocks and set 1 in swp if the secondary
520 	 * super block is valid and newer.  Otherwise, set 0 in swp.
521 	 */
522 	valid[0] = nilfs_valid_sb(sbp[0]);
523 	valid[1] = nilfs_valid_sb(sbp[1]);
524 	swp = valid[1] && (!valid[0] ||
525 			   le64_to_cpu(sbp[1]->s_last_cno) >
526 			   le64_to_cpu(sbp[0]->s_last_cno));
527 
528 	if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
529 		brelse(sbh[1]);
530 		sbh[1] = NULL;
531 		sbp[1] = NULL;
532 		valid[1] = 0;
533 		swp = 0;
534 	}
535 	if (!valid[swp]) {
536 		nilfs_release_super_block(nilfs);
537 		nilfs_msg(sb, KERN_ERR, "couldn't find nilfs on the device");
538 		return -EINVAL;
539 	}
540 
541 	if (!valid[!swp])
542 		nilfs_msg(sb, KERN_WARNING,
543 			  "broken superblock, retrying with spare superblock (blocksize = %d)",
544 			  blocksize);
545 	if (swp)
546 		nilfs_swap_super_block(nilfs);
547 
548 	nilfs->ns_sbwcount = 0;
549 	nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
550 	nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
551 	*sbpp = sbp[0];
552 	return 0;
553 }
554 
555 /**
556  * init_nilfs - initialize a NILFS instance.
557  * @nilfs: the_nilfs structure
558  * @sb: super block
559  * @data: mount options
560  *
561  * init_nilfs() performs common initialization per block device (e.g.
562  * reading the super block, getting disk layout information, initializing
563  * shared fields in the_nilfs).
564  *
565  * Return Value: On success, 0 is returned. On error, a negative error
566  * code is returned.
567  */
568 int init_nilfs(struct the_nilfs *nilfs, struct super_block *sb, char *data)
569 {
570 	struct nilfs_super_block *sbp;
571 	int blocksize;
572 	int err;
573 
574 	down_write(&nilfs->ns_sem);
575 
576 	blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
577 	if (!blocksize) {
578 		nilfs_msg(sb, KERN_ERR, "unable to set blocksize");
579 		err = -EINVAL;
580 		goto out;
581 	}
582 	err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
583 	if (err)
584 		goto out;
585 
586 	err = nilfs_store_magic_and_option(sb, sbp, data);
587 	if (err)
588 		goto failed_sbh;
589 
590 	err = nilfs_check_feature_compatibility(sb, sbp);
591 	if (err)
592 		goto failed_sbh;
593 
594 	blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
595 	if (blocksize < NILFS_MIN_BLOCK_SIZE ||
596 	    blocksize > NILFS_MAX_BLOCK_SIZE) {
597 		nilfs_msg(sb, KERN_ERR,
598 			  "couldn't mount because of unsupported filesystem blocksize %d",
599 			  blocksize);
600 		err = -EINVAL;
601 		goto failed_sbh;
602 	}
603 	if (sb->s_blocksize != blocksize) {
604 		int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
605 
606 		if (blocksize < hw_blocksize) {
607 			nilfs_msg(sb, KERN_ERR,
608 				  "blocksize %d too small for device (sector-size = %d)",
609 				  blocksize, hw_blocksize);
610 			err = -EINVAL;
611 			goto failed_sbh;
612 		}
613 		nilfs_release_super_block(nilfs);
614 		sb_set_blocksize(sb, blocksize);
615 
616 		err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
617 		if (err)
618 			goto out;
619 			/*
620 			 * Not to failed_sbh; sbh is released automatically
621 			 * when reloading fails.
622 			 */
623 	}
624 	nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
625 	nilfs->ns_blocksize = blocksize;
626 
627 	get_random_bytes(&nilfs->ns_next_generation,
628 			 sizeof(nilfs->ns_next_generation));
629 
630 	err = nilfs_store_disk_layout(nilfs, sbp);
631 	if (err)
632 		goto failed_sbh;
633 
634 	sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
635 
636 	nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
637 
638 	err = nilfs_store_log_cursor(nilfs, sbp);
639 	if (err)
640 		goto failed_sbh;
641 
642 	err = nilfs_sysfs_create_device_group(sb);
643 	if (err)
644 		goto failed_sbh;
645 
646 	set_nilfs_init(nilfs);
647 	err = 0;
648  out:
649 	up_write(&nilfs->ns_sem);
650 	return err;
651 
652  failed_sbh:
653 	nilfs_release_super_block(nilfs);
654 	goto out;
655 }
656 
657 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
658 			    size_t nsegs)
659 {
660 	sector_t seg_start, seg_end;
661 	sector_t start = 0, nblocks = 0;
662 	unsigned int sects_per_block;
663 	__u64 *sn;
664 	int ret = 0;
665 
666 	sects_per_block = (1 << nilfs->ns_blocksize_bits) /
667 		bdev_logical_block_size(nilfs->ns_bdev);
668 	for (sn = segnump; sn < segnump + nsegs; sn++) {
669 		nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
670 
671 		if (!nblocks) {
672 			start = seg_start;
673 			nblocks = seg_end - seg_start + 1;
674 		} else if (start + nblocks == seg_start) {
675 			nblocks += seg_end - seg_start + 1;
676 		} else {
677 			ret = blkdev_issue_discard(nilfs->ns_bdev,
678 						   start * sects_per_block,
679 						   nblocks * sects_per_block,
680 						   GFP_NOFS, 0);
681 			if (ret < 0)
682 				return ret;
683 			nblocks = 0;
684 		}
685 	}
686 	if (nblocks)
687 		ret = blkdev_issue_discard(nilfs->ns_bdev,
688 					   start * sects_per_block,
689 					   nblocks * sects_per_block,
690 					   GFP_NOFS, 0);
691 	return ret;
692 }
693 
694 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
695 {
696 	unsigned long ncleansegs;
697 
698 	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
699 	ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
700 	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
701 	*nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
702 	return 0;
703 }
704 
705 int nilfs_near_disk_full(struct the_nilfs *nilfs)
706 {
707 	unsigned long ncleansegs, nincsegs;
708 
709 	ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
710 	nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
711 		nilfs->ns_blocks_per_segment + 1;
712 
713 	return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
714 }
715 
716 struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
717 {
718 	struct rb_node *n;
719 	struct nilfs_root *root;
720 
721 	spin_lock(&nilfs->ns_cptree_lock);
722 	n = nilfs->ns_cptree.rb_node;
723 	while (n) {
724 		root = rb_entry(n, struct nilfs_root, rb_node);
725 
726 		if (cno < root->cno) {
727 			n = n->rb_left;
728 		} else if (cno > root->cno) {
729 			n = n->rb_right;
730 		} else {
731 			refcount_inc(&root->count);
732 			spin_unlock(&nilfs->ns_cptree_lock);
733 			return root;
734 		}
735 	}
736 	spin_unlock(&nilfs->ns_cptree_lock);
737 
738 	return NULL;
739 }
740 
741 struct nilfs_root *
742 nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
743 {
744 	struct rb_node **p, *parent;
745 	struct nilfs_root *root, *new;
746 	int err;
747 
748 	root = nilfs_lookup_root(nilfs, cno);
749 	if (root)
750 		return root;
751 
752 	new = kzalloc(sizeof(*root), GFP_KERNEL);
753 	if (!new)
754 		return NULL;
755 
756 	spin_lock(&nilfs->ns_cptree_lock);
757 
758 	p = &nilfs->ns_cptree.rb_node;
759 	parent = NULL;
760 
761 	while (*p) {
762 		parent = *p;
763 		root = rb_entry(parent, struct nilfs_root, rb_node);
764 
765 		if (cno < root->cno) {
766 			p = &(*p)->rb_left;
767 		} else if (cno > root->cno) {
768 			p = &(*p)->rb_right;
769 		} else {
770 			refcount_inc(&root->count);
771 			spin_unlock(&nilfs->ns_cptree_lock);
772 			kfree(new);
773 			return root;
774 		}
775 	}
776 
777 	new->cno = cno;
778 	new->ifile = NULL;
779 	new->nilfs = nilfs;
780 	refcount_set(&new->count, 1);
781 	atomic64_set(&new->inodes_count, 0);
782 	atomic64_set(&new->blocks_count, 0);
783 
784 	rb_link_node(&new->rb_node, parent, p);
785 	rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
786 
787 	spin_unlock(&nilfs->ns_cptree_lock);
788 
789 	err = nilfs_sysfs_create_snapshot_group(new);
790 	if (err) {
791 		kfree(new);
792 		new = NULL;
793 	}
794 
795 	return new;
796 }
797 
798 void nilfs_put_root(struct nilfs_root *root)
799 {
800 	if (refcount_dec_and_test(&root->count)) {
801 		struct the_nilfs *nilfs = root->nilfs;
802 
803 		nilfs_sysfs_delete_snapshot_group(root);
804 
805 		spin_lock(&nilfs->ns_cptree_lock);
806 		rb_erase(&root->rb_node, &nilfs->ns_cptree);
807 		spin_unlock(&nilfs->ns_cptree_lock);
808 		iput(root->ifile);
809 
810 		kfree(root);
811 	}
812 }
813