xref: /linux/fs/nilfs2/the_nilfs.c (revision 660a5704603593393799b4cd116fac06d35471b2)
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/log2.h>
17 #include <linux/crc32.h>
18 #include "nilfs.h"
19 #include "segment.h"
20 #include "alloc.h"
21 #include "cpfile.h"
22 #include "sufile.h"
23 #include "dat.h"
24 #include "segbuf.h"
25 
26 
27 static int nilfs_valid_sb(struct nilfs_super_block *sbp);
28 
29 void nilfs_set_last_segment(struct the_nilfs *nilfs,
30 			    sector_t start_blocknr, u64 seq, __u64 cno)
31 {
32 	spin_lock(&nilfs->ns_last_segment_lock);
33 	nilfs->ns_last_pseg = start_blocknr;
34 	nilfs->ns_last_seq = seq;
35 	nilfs->ns_last_cno = cno;
36 
37 	if (!nilfs_sb_dirty(nilfs)) {
38 		if (nilfs->ns_prev_seq == nilfs->ns_last_seq)
39 			goto stay_cursor;
40 
41 		set_nilfs_sb_dirty(nilfs);
42 	}
43 	nilfs->ns_prev_seq = nilfs->ns_last_seq;
44 
45  stay_cursor:
46 	spin_unlock(&nilfs->ns_last_segment_lock);
47 }
48 
49 /**
50  * alloc_nilfs - allocate a nilfs object
51  * @sb: super block instance
52  *
53  * Return Value: On success, pointer to the_nilfs is returned.
54  * On error, NULL is returned.
55  */
56 struct the_nilfs *alloc_nilfs(struct super_block *sb)
57 {
58 	struct the_nilfs *nilfs;
59 
60 	nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
61 	if (!nilfs)
62 		return NULL;
63 
64 	nilfs->ns_sb = sb;
65 	nilfs->ns_bdev = sb->s_bdev;
66 	atomic_set(&nilfs->ns_ndirtyblks, 0);
67 	init_rwsem(&nilfs->ns_sem);
68 	mutex_init(&nilfs->ns_snapshot_mount_mutex);
69 	INIT_LIST_HEAD(&nilfs->ns_dirty_files);
70 	INIT_LIST_HEAD(&nilfs->ns_gc_inodes);
71 	spin_lock_init(&nilfs->ns_inode_lock);
72 	spin_lock_init(&nilfs->ns_next_gen_lock);
73 	spin_lock_init(&nilfs->ns_last_segment_lock);
74 	nilfs->ns_cptree = RB_ROOT;
75 	spin_lock_init(&nilfs->ns_cptree_lock);
76 	init_rwsem(&nilfs->ns_segctor_sem);
77 	nilfs->ns_sb_update_freq = NILFS_SB_FREQ;
78 
79 	return nilfs;
80 }
81 
82 /**
83  * destroy_nilfs - destroy nilfs object
84  * @nilfs: nilfs object to be released
85  */
86 void destroy_nilfs(struct the_nilfs *nilfs)
87 {
88 	might_sleep();
89 	if (nilfs_init(nilfs)) {
90 		nilfs_sysfs_delete_device_group(nilfs);
91 		brelse(nilfs->ns_sbh[0]);
92 		brelse(nilfs->ns_sbh[1]);
93 	}
94 	kfree(nilfs);
95 }
96 
97 static int nilfs_load_super_root(struct the_nilfs *nilfs,
98 				 struct super_block *sb, sector_t sr_block)
99 {
100 	struct buffer_head *bh_sr;
101 	struct nilfs_super_root *raw_sr;
102 	struct nilfs_super_block **sbp = nilfs->ns_sbp;
103 	struct nilfs_inode *rawi;
104 	unsigned int dat_entry_size, segment_usage_size, checkpoint_size;
105 	unsigned int inode_size;
106 	int err;
107 
108 	err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
109 	if (unlikely(err))
110 		return err;
111 
112 	down_read(&nilfs->ns_sem);
113 	dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
114 	checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
115 	segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
116 	up_read(&nilfs->ns_sem);
117 
118 	inode_size = nilfs->ns_inode_size;
119 
120 	rawi = (void *)bh_sr->b_data + NILFS_SR_DAT_OFFSET(inode_size);
121 	err = nilfs_dat_read(sb, dat_entry_size, rawi, &nilfs->ns_dat);
122 	if (err)
123 		goto failed;
124 
125 	rawi = (void *)bh_sr->b_data + NILFS_SR_CPFILE_OFFSET(inode_size);
126 	err = nilfs_cpfile_read(sb, checkpoint_size, rawi, &nilfs->ns_cpfile);
127 	if (err)
128 		goto failed_dat;
129 
130 	rawi = (void *)bh_sr->b_data + NILFS_SR_SUFILE_OFFSET(inode_size);
131 	err = nilfs_sufile_read(sb, segment_usage_size, rawi,
132 				&nilfs->ns_sufile);
133 	if (err)
134 		goto failed_cpfile;
135 
136 	raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
137 	nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
138 
139  failed:
140 	brelse(bh_sr);
141 	return err;
142 
143  failed_cpfile:
144 	iput(nilfs->ns_cpfile);
145 
146  failed_dat:
147 	iput(nilfs->ns_dat);
148 	goto failed;
149 }
150 
151 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
152 {
153 	memset(ri, 0, sizeof(*ri));
154 	INIT_LIST_HEAD(&ri->ri_used_segments);
155 }
156 
157 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
158 {
159 	nilfs_dispose_segment_list(&ri->ri_used_segments);
160 }
161 
162 /**
163  * nilfs_store_log_cursor - load log cursor from a super block
164  * @nilfs: nilfs object
165  * @sbp: buffer storing super block to be read
166  *
167  * nilfs_store_log_cursor() reads the last position of the log
168  * containing a super root from a given super block, and initializes
169  * relevant information on the nilfs object preparatory for log
170  * scanning and recovery.
171  */
172 static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
173 				  struct nilfs_super_block *sbp)
174 {
175 	int ret = 0;
176 
177 	nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
178 	nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
179 	nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
180 
181 	nilfs->ns_prev_seq = nilfs->ns_last_seq;
182 	nilfs->ns_seg_seq = nilfs->ns_last_seq;
183 	nilfs->ns_segnum =
184 		nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
185 	nilfs->ns_cno = nilfs->ns_last_cno + 1;
186 	if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
187 		nilfs_err(nilfs->ns_sb,
188 			  "pointed segment number is out of range: segnum=%llu, nsegments=%lu",
189 			  (unsigned long long)nilfs->ns_segnum,
190 			  nilfs->ns_nsegments);
191 		ret = -EINVAL;
192 	}
193 	return ret;
194 }
195 
196 /**
197  * nilfs_get_blocksize - get block size from raw superblock data
198  * @sb: super block instance
199  * @sbp: superblock raw data buffer
200  * @blocksize: place to store block size
201  *
202  * nilfs_get_blocksize() calculates the block size from the block size
203  * exponent information written in @sbp and stores it in @blocksize,
204  * or aborts with an error message if it's too large.
205  *
206  * Return Value: On success, 0 is returned. If the block size is too
207  * large, -EINVAL is returned.
208  */
209 static int nilfs_get_blocksize(struct super_block *sb,
210 			       struct nilfs_super_block *sbp, int *blocksize)
211 {
212 	unsigned int shift_bits = le32_to_cpu(sbp->s_log_block_size);
213 
214 	if (unlikely(shift_bits >
215 		     ilog2(NILFS_MAX_BLOCK_SIZE) - BLOCK_SIZE_BITS)) {
216 		nilfs_err(sb, "too large filesystem blocksize: 2 ^ %u KiB",
217 			  shift_bits);
218 		return -EINVAL;
219 	}
220 	*blocksize = BLOCK_SIZE << shift_bits;
221 	return 0;
222 }
223 
224 /**
225  * load_nilfs - load and recover the nilfs
226  * @nilfs: the_nilfs structure to be released
227  * @sb: super block instance used to recover past segment
228  *
229  * load_nilfs() searches and load the latest super root,
230  * attaches the last segment, and does recovery if needed.
231  * The caller must call this exclusively for simultaneous mounts.
232  */
233 int load_nilfs(struct the_nilfs *nilfs, struct super_block *sb)
234 {
235 	struct nilfs_recovery_info ri;
236 	unsigned int s_flags = sb->s_flags;
237 	int really_read_only = bdev_read_only(nilfs->ns_bdev);
238 	int valid_fs = nilfs_valid_fs(nilfs);
239 	int err;
240 
241 	if (!valid_fs) {
242 		nilfs_warn(sb, "mounting unchecked fs");
243 		if (s_flags & SB_RDONLY) {
244 			nilfs_info(sb,
245 				   "recovery required for readonly filesystem");
246 			nilfs_info(sb,
247 				   "write access will be enabled during recovery");
248 		}
249 	}
250 
251 	nilfs_init_recovery_info(&ri);
252 
253 	err = nilfs_search_super_root(nilfs, &ri);
254 	if (unlikely(err)) {
255 		struct nilfs_super_block **sbp = nilfs->ns_sbp;
256 		int blocksize;
257 
258 		if (err != -EINVAL)
259 			goto scan_error;
260 
261 		if (!nilfs_valid_sb(sbp[1])) {
262 			nilfs_warn(sb,
263 				   "unable to fall back to spare super block");
264 			goto scan_error;
265 		}
266 		nilfs_info(sb, "trying rollback from an earlier position");
267 
268 		/*
269 		 * restore super block with its spare and reconfigure
270 		 * relevant states of the nilfs object.
271 		 */
272 		memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
273 		nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
274 		nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
275 
276 		/* verify consistency between two super blocks */
277 		err = nilfs_get_blocksize(sb, sbp[0], &blocksize);
278 		if (err)
279 			goto scan_error;
280 
281 		if (blocksize != nilfs->ns_blocksize) {
282 			nilfs_warn(sb,
283 				   "blocksize differs between two super blocks (%d != %d)",
284 				   blocksize, nilfs->ns_blocksize);
285 			err = -EINVAL;
286 			goto scan_error;
287 		}
288 
289 		err = nilfs_store_log_cursor(nilfs, sbp[0]);
290 		if (err)
291 			goto scan_error;
292 
293 		/* drop clean flag to allow roll-forward and recovery */
294 		nilfs->ns_mount_state &= ~NILFS_VALID_FS;
295 		valid_fs = 0;
296 
297 		err = nilfs_search_super_root(nilfs, &ri);
298 		if (err)
299 			goto scan_error;
300 	}
301 
302 	err = nilfs_load_super_root(nilfs, sb, ri.ri_super_root);
303 	if (unlikely(err)) {
304 		nilfs_err(sb, "error %d while loading super root", err);
305 		goto failed;
306 	}
307 
308 	if (valid_fs)
309 		goto skip_recovery;
310 
311 	if (s_flags & SB_RDONLY) {
312 		__u64 features;
313 
314 		if (nilfs_test_opt(nilfs, NORECOVERY)) {
315 			nilfs_info(sb,
316 				   "norecovery option specified, skipping roll-forward recovery");
317 			goto skip_recovery;
318 		}
319 		features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
320 			~NILFS_FEATURE_COMPAT_RO_SUPP;
321 		if (features) {
322 			nilfs_err(sb,
323 				  "couldn't proceed with recovery because of unsupported optional features (%llx)",
324 				  (unsigned long long)features);
325 			err = -EROFS;
326 			goto failed_unload;
327 		}
328 		if (really_read_only) {
329 			nilfs_err(sb,
330 				  "write access unavailable, cannot proceed");
331 			err = -EROFS;
332 			goto failed_unload;
333 		}
334 		sb->s_flags &= ~SB_RDONLY;
335 	} else if (nilfs_test_opt(nilfs, NORECOVERY)) {
336 		nilfs_err(sb,
337 			  "recovery cancelled because norecovery option was specified for a read/write mount");
338 		err = -EINVAL;
339 		goto failed_unload;
340 	}
341 
342 	err = nilfs_salvage_orphan_logs(nilfs, sb, &ri);
343 	if (err)
344 		goto failed_unload;
345 
346 	down_write(&nilfs->ns_sem);
347 	nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
348 	err = nilfs_cleanup_super(sb);
349 	up_write(&nilfs->ns_sem);
350 
351 	if (err) {
352 		nilfs_err(sb,
353 			  "error %d updating super block. recovery unfinished.",
354 			  err);
355 		goto failed_unload;
356 	}
357 	nilfs_info(sb, "recovery complete");
358 
359  skip_recovery:
360 	nilfs_clear_recovery_info(&ri);
361 	sb->s_flags = s_flags;
362 	return 0;
363 
364  scan_error:
365 	nilfs_err(sb, "error %d while searching super root", err);
366 	goto failed;
367 
368  failed_unload:
369 	iput(nilfs->ns_cpfile);
370 	iput(nilfs->ns_sufile);
371 	iput(nilfs->ns_dat);
372 
373  failed:
374 	nilfs_clear_recovery_info(&ri);
375 	sb->s_flags = s_flags;
376 	return err;
377 }
378 
379 static unsigned long long nilfs_max_size(unsigned int blkbits)
380 {
381 	unsigned int max_bits;
382 	unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
383 
384 	max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
385 	if (max_bits < 64)
386 		res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
387 	return res;
388 }
389 
390 /**
391  * nilfs_nrsvsegs - calculate the number of reserved segments
392  * @nilfs: nilfs object
393  * @nsegs: total number of segments
394  */
395 unsigned long nilfs_nrsvsegs(struct the_nilfs *nilfs, unsigned long nsegs)
396 {
397 	return max_t(unsigned long, NILFS_MIN_NRSVSEGS,
398 		     DIV_ROUND_UP(nsegs * nilfs->ns_r_segments_percentage,
399 				  100));
400 }
401 
402 void nilfs_set_nsegments(struct the_nilfs *nilfs, unsigned long nsegs)
403 {
404 	nilfs->ns_nsegments = nsegs;
405 	nilfs->ns_nrsvsegs = nilfs_nrsvsegs(nilfs, nsegs);
406 }
407 
408 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
409 				   struct nilfs_super_block *sbp)
410 {
411 	if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
412 		nilfs_err(nilfs->ns_sb,
413 			  "unsupported revision (superblock rev.=%d.%d, current rev.=%d.%d). Please check the version of mkfs.nilfs(2).",
414 			  le32_to_cpu(sbp->s_rev_level),
415 			  le16_to_cpu(sbp->s_minor_rev_level),
416 			  NILFS_CURRENT_REV, NILFS_MINOR_REV);
417 		return -EINVAL;
418 	}
419 	nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
420 	if (nilfs->ns_sbsize > BLOCK_SIZE)
421 		return -EINVAL;
422 
423 	nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
424 	if (nilfs->ns_inode_size > nilfs->ns_blocksize) {
425 		nilfs_err(nilfs->ns_sb, "too large inode size: %d bytes",
426 			  nilfs->ns_inode_size);
427 		return -EINVAL;
428 	} else if (nilfs->ns_inode_size < NILFS_MIN_INODE_SIZE) {
429 		nilfs_err(nilfs->ns_sb, "too small inode size: %d bytes",
430 			  nilfs->ns_inode_size);
431 		return -EINVAL;
432 	}
433 
434 	nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
435 
436 	nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
437 	if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
438 		nilfs_err(nilfs->ns_sb, "too short segment: %lu blocks",
439 			  nilfs->ns_blocks_per_segment);
440 		return -EINVAL;
441 	}
442 
443 	nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
444 	nilfs->ns_r_segments_percentage =
445 		le32_to_cpu(sbp->s_r_segments_percentage);
446 	if (nilfs->ns_r_segments_percentage < 1 ||
447 	    nilfs->ns_r_segments_percentage > 99) {
448 		nilfs_err(nilfs->ns_sb,
449 			  "invalid reserved segments percentage: %lu",
450 			  nilfs->ns_r_segments_percentage);
451 		return -EINVAL;
452 	}
453 
454 	nilfs_set_nsegments(nilfs, le64_to_cpu(sbp->s_nsegments));
455 	nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
456 	return 0;
457 }
458 
459 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
460 {
461 	static unsigned char sum[4];
462 	const int sumoff = offsetof(struct nilfs_super_block, s_sum);
463 	size_t bytes;
464 	u32 crc;
465 
466 	if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
467 		return 0;
468 	bytes = le16_to_cpu(sbp->s_bytes);
469 	if (bytes < sumoff + 4 || bytes > BLOCK_SIZE)
470 		return 0;
471 	crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
472 		       sumoff);
473 	crc = crc32_le(crc, sum, 4);
474 	crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
475 		       bytes - sumoff - 4);
476 	return crc == le32_to_cpu(sbp->s_sum);
477 }
478 
479 /**
480  * nilfs_sb2_bad_offset - check the location of the second superblock
481  * @sbp: superblock raw data buffer
482  * @offset: byte offset of second superblock calculated from device size
483  *
484  * nilfs_sb2_bad_offset() checks if the position on the second
485  * superblock is valid or not based on the filesystem parameters
486  * stored in @sbp.  If @offset points to a location within the segment
487  * area, or if the parameters themselves are not normal, it is
488  * determined to be invalid.
489  *
490  * Return Value: true if invalid, false if valid.
491  */
492 static bool nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
493 {
494 	unsigned int shift_bits = le32_to_cpu(sbp->s_log_block_size);
495 	u32 blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
496 	u64 nsegments = le64_to_cpu(sbp->s_nsegments);
497 	u64 index;
498 
499 	if (blocks_per_segment < NILFS_SEG_MIN_BLOCKS ||
500 	    shift_bits > ilog2(NILFS_MAX_BLOCK_SIZE) - BLOCK_SIZE_BITS)
501 		return true;
502 
503 	index = offset >> (shift_bits + BLOCK_SIZE_BITS);
504 	do_div(index, blocks_per_segment);
505 	return index < nsegments;
506 }
507 
508 static void nilfs_release_super_block(struct the_nilfs *nilfs)
509 {
510 	int i;
511 
512 	for (i = 0; i < 2; i++) {
513 		if (nilfs->ns_sbp[i]) {
514 			brelse(nilfs->ns_sbh[i]);
515 			nilfs->ns_sbh[i] = NULL;
516 			nilfs->ns_sbp[i] = NULL;
517 		}
518 	}
519 }
520 
521 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
522 {
523 	brelse(nilfs->ns_sbh[0]);
524 	nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
525 	nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
526 	nilfs->ns_sbh[1] = NULL;
527 	nilfs->ns_sbp[1] = NULL;
528 }
529 
530 void nilfs_swap_super_block(struct the_nilfs *nilfs)
531 {
532 	struct buffer_head *tsbh = nilfs->ns_sbh[0];
533 	struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
534 
535 	nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
536 	nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
537 	nilfs->ns_sbh[1] = tsbh;
538 	nilfs->ns_sbp[1] = tsbp;
539 }
540 
541 static int nilfs_load_super_block(struct the_nilfs *nilfs,
542 				  struct super_block *sb, int blocksize,
543 				  struct nilfs_super_block **sbpp)
544 {
545 	struct nilfs_super_block **sbp = nilfs->ns_sbp;
546 	struct buffer_head **sbh = nilfs->ns_sbh;
547 	u64 sb2off = NILFS_SB2_OFFSET_BYTES(bdev_nr_bytes(nilfs->ns_bdev));
548 	int valid[2], swp = 0;
549 
550 	sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
551 					&sbh[0]);
552 	sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
553 
554 	if (!sbp[0]) {
555 		if (!sbp[1]) {
556 			nilfs_err(sb, "unable to read superblock");
557 			return -EIO;
558 		}
559 		nilfs_warn(sb,
560 			   "unable to read primary superblock (blocksize = %d)",
561 			   blocksize);
562 	} else if (!sbp[1]) {
563 		nilfs_warn(sb,
564 			   "unable to read secondary superblock (blocksize = %d)",
565 			   blocksize);
566 	}
567 
568 	/*
569 	 * Compare two super blocks and set 1 in swp if the secondary
570 	 * super block is valid and newer.  Otherwise, set 0 in swp.
571 	 */
572 	valid[0] = nilfs_valid_sb(sbp[0]);
573 	valid[1] = nilfs_valid_sb(sbp[1]);
574 	swp = valid[1] && (!valid[0] ||
575 			   le64_to_cpu(sbp[1]->s_last_cno) >
576 			   le64_to_cpu(sbp[0]->s_last_cno));
577 
578 	if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
579 		brelse(sbh[1]);
580 		sbh[1] = NULL;
581 		sbp[1] = NULL;
582 		valid[1] = 0;
583 		swp = 0;
584 	}
585 	if (!valid[swp]) {
586 		nilfs_release_super_block(nilfs);
587 		nilfs_err(sb, "couldn't find nilfs on the device");
588 		return -EINVAL;
589 	}
590 
591 	if (!valid[!swp])
592 		nilfs_warn(sb,
593 			   "broken superblock, retrying with spare superblock (blocksize = %d)",
594 			   blocksize);
595 	if (swp)
596 		nilfs_swap_super_block(nilfs);
597 
598 	nilfs->ns_sbwcount = 0;
599 	nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
600 	nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
601 	*sbpp = sbp[0];
602 	return 0;
603 }
604 
605 /**
606  * init_nilfs - initialize a NILFS instance.
607  * @nilfs: the_nilfs structure
608  * @sb: super block
609  * @data: mount options
610  *
611  * init_nilfs() performs common initialization per block device (e.g.
612  * reading the super block, getting disk layout information, initializing
613  * shared fields in the_nilfs).
614  *
615  * Return Value: On success, 0 is returned. On error, a negative error
616  * code is returned.
617  */
618 int init_nilfs(struct the_nilfs *nilfs, struct super_block *sb, char *data)
619 {
620 	struct nilfs_super_block *sbp;
621 	int blocksize;
622 	int err;
623 
624 	down_write(&nilfs->ns_sem);
625 
626 	blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
627 	if (!blocksize) {
628 		nilfs_err(sb, "unable to set blocksize");
629 		err = -EINVAL;
630 		goto out;
631 	}
632 	err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
633 	if (err)
634 		goto out;
635 
636 	err = nilfs_store_magic_and_option(sb, sbp, data);
637 	if (err)
638 		goto failed_sbh;
639 
640 	err = nilfs_check_feature_compatibility(sb, sbp);
641 	if (err)
642 		goto failed_sbh;
643 
644 	err = nilfs_get_blocksize(sb, sbp, &blocksize);
645 	if (err)
646 		goto failed_sbh;
647 
648 	if (blocksize < NILFS_MIN_BLOCK_SIZE) {
649 		nilfs_err(sb,
650 			  "couldn't mount because of unsupported filesystem blocksize %d",
651 			  blocksize);
652 		err = -EINVAL;
653 		goto failed_sbh;
654 	}
655 	if (sb->s_blocksize != blocksize) {
656 		int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
657 
658 		if (blocksize < hw_blocksize) {
659 			nilfs_err(sb,
660 				  "blocksize %d too small for device (sector-size = %d)",
661 				  blocksize, hw_blocksize);
662 			err = -EINVAL;
663 			goto failed_sbh;
664 		}
665 		nilfs_release_super_block(nilfs);
666 		sb_set_blocksize(sb, blocksize);
667 
668 		err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
669 		if (err)
670 			goto out;
671 			/*
672 			 * Not to failed_sbh; sbh is released automatically
673 			 * when reloading fails.
674 			 */
675 	}
676 	nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
677 	nilfs->ns_blocksize = blocksize;
678 
679 	get_random_bytes(&nilfs->ns_next_generation,
680 			 sizeof(nilfs->ns_next_generation));
681 
682 	err = nilfs_store_disk_layout(nilfs, sbp);
683 	if (err)
684 		goto failed_sbh;
685 
686 	sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
687 
688 	nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
689 
690 	err = nilfs_store_log_cursor(nilfs, sbp);
691 	if (err)
692 		goto failed_sbh;
693 
694 	err = nilfs_sysfs_create_device_group(sb);
695 	if (err)
696 		goto failed_sbh;
697 
698 	set_nilfs_init(nilfs);
699 	err = 0;
700  out:
701 	up_write(&nilfs->ns_sem);
702 	return err;
703 
704  failed_sbh:
705 	nilfs_release_super_block(nilfs);
706 	goto out;
707 }
708 
709 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
710 			    size_t nsegs)
711 {
712 	sector_t seg_start, seg_end;
713 	sector_t start = 0, nblocks = 0;
714 	unsigned int sects_per_block;
715 	__u64 *sn;
716 	int ret = 0;
717 
718 	sects_per_block = (1 << nilfs->ns_blocksize_bits) /
719 		bdev_logical_block_size(nilfs->ns_bdev);
720 	for (sn = segnump; sn < segnump + nsegs; sn++) {
721 		nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
722 
723 		if (!nblocks) {
724 			start = seg_start;
725 			nblocks = seg_end - seg_start + 1;
726 		} else if (start + nblocks == seg_start) {
727 			nblocks += seg_end - seg_start + 1;
728 		} else {
729 			ret = blkdev_issue_discard(nilfs->ns_bdev,
730 						   start * sects_per_block,
731 						   nblocks * sects_per_block,
732 						   GFP_NOFS);
733 			if (ret < 0)
734 				return ret;
735 			nblocks = 0;
736 		}
737 	}
738 	if (nblocks)
739 		ret = blkdev_issue_discard(nilfs->ns_bdev,
740 					   start * sects_per_block,
741 					   nblocks * sects_per_block,
742 					   GFP_NOFS);
743 	return ret;
744 }
745 
746 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
747 {
748 	unsigned long ncleansegs;
749 
750 	ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
751 	*nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
752 	return 0;
753 }
754 
755 int nilfs_near_disk_full(struct the_nilfs *nilfs)
756 {
757 	unsigned long ncleansegs, nincsegs;
758 
759 	ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
760 	nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
761 		nilfs->ns_blocks_per_segment + 1;
762 
763 	return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
764 }
765 
766 struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
767 {
768 	struct rb_node *n;
769 	struct nilfs_root *root;
770 
771 	spin_lock(&nilfs->ns_cptree_lock);
772 	n = nilfs->ns_cptree.rb_node;
773 	while (n) {
774 		root = rb_entry(n, struct nilfs_root, rb_node);
775 
776 		if (cno < root->cno) {
777 			n = n->rb_left;
778 		} else if (cno > root->cno) {
779 			n = n->rb_right;
780 		} else {
781 			refcount_inc(&root->count);
782 			spin_unlock(&nilfs->ns_cptree_lock);
783 			return root;
784 		}
785 	}
786 	spin_unlock(&nilfs->ns_cptree_lock);
787 
788 	return NULL;
789 }
790 
791 struct nilfs_root *
792 nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
793 {
794 	struct rb_node **p, *parent;
795 	struct nilfs_root *root, *new;
796 	int err;
797 
798 	root = nilfs_lookup_root(nilfs, cno);
799 	if (root)
800 		return root;
801 
802 	new = kzalloc(sizeof(*root), GFP_KERNEL);
803 	if (!new)
804 		return NULL;
805 
806 	spin_lock(&nilfs->ns_cptree_lock);
807 
808 	p = &nilfs->ns_cptree.rb_node;
809 	parent = NULL;
810 
811 	while (*p) {
812 		parent = *p;
813 		root = rb_entry(parent, struct nilfs_root, rb_node);
814 
815 		if (cno < root->cno) {
816 			p = &(*p)->rb_left;
817 		} else if (cno > root->cno) {
818 			p = &(*p)->rb_right;
819 		} else {
820 			refcount_inc(&root->count);
821 			spin_unlock(&nilfs->ns_cptree_lock);
822 			kfree(new);
823 			return root;
824 		}
825 	}
826 
827 	new->cno = cno;
828 	new->ifile = NULL;
829 	new->nilfs = nilfs;
830 	refcount_set(&new->count, 1);
831 	atomic64_set(&new->inodes_count, 0);
832 	atomic64_set(&new->blocks_count, 0);
833 
834 	rb_link_node(&new->rb_node, parent, p);
835 	rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
836 
837 	spin_unlock(&nilfs->ns_cptree_lock);
838 
839 	err = nilfs_sysfs_create_snapshot_group(new);
840 	if (err) {
841 		kfree(new);
842 		new = NULL;
843 	}
844 
845 	return new;
846 }
847 
848 void nilfs_put_root(struct nilfs_root *root)
849 {
850 	struct the_nilfs *nilfs = root->nilfs;
851 
852 	if (refcount_dec_and_lock(&root->count, &nilfs->ns_cptree_lock)) {
853 		rb_erase(&root->rb_node, &nilfs->ns_cptree);
854 		spin_unlock(&nilfs->ns_cptree_lock);
855 
856 		nilfs_sysfs_delete_snapshot_group(root);
857 		iput(root->ifile);
858 
859 		kfree(root);
860 	}
861 }
862