xref: /linux/fs/nilfs2/the_nilfs.c (revision 55f1b540d893da740a81200450014c45a8103f54)
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/log2.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_last_segment_lock);
72 	nilfs->ns_cptree = RB_ROOT;
73 	spin_lock_init(&nilfs->ns_cptree_lock);
74 	init_rwsem(&nilfs->ns_segctor_sem);
75 	nilfs->ns_sb_update_freq = NILFS_SB_FREQ;
76 
77 	return nilfs;
78 }
79 
80 /**
81  * destroy_nilfs - destroy nilfs object
82  * @nilfs: nilfs object to be released
83  */
84 void destroy_nilfs(struct the_nilfs *nilfs)
85 {
86 	might_sleep();
87 	if (nilfs_init(nilfs)) {
88 		brelse(nilfs->ns_sbh[0]);
89 		brelse(nilfs->ns_sbh[1]);
90 	}
91 	kfree(nilfs);
92 }
93 
94 static int nilfs_load_super_root(struct the_nilfs *nilfs,
95 				 struct super_block *sb, sector_t sr_block)
96 {
97 	struct buffer_head *bh_sr;
98 	struct nilfs_super_root *raw_sr;
99 	struct nilfs_super_block **sbp = nilfs->ns_sbp;
100 	struct nilfs_inode *rawi;
101 	unsigned int dat_entry_size, segment_usage_size, checkpoint_size;
102 	unsigned int inode_size;
103 	int err;
104 
105 	err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
106 	if (unlikely(err))
107 		return err;
108 
109 	down_read(&nilfs->ns_sem);
110 	dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
111 	checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
112 	segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
113 	up_read(&nilfs->ns_sem);
114 
115 	inode_size = nilfs->ns_inode_size;
116 
117 	rawi = (void *)bh_sr->b_data + NILFS_SR_DAT_OFFSET(inode_size);
118 	err = nilfs_dat_read(sb, dat_entry_size, rawi, &nilfs->ns_dat);
119 	if (err)
120 		goto failed;
121 
122 	rawi = (void *)bh_sr->b_data + NILFS_SR_CPFILE_OFFSET(inode_size);
123 	err = nilfs_cpfile_read(sb, checkpoint_size, rawi, &nilfs->ns_cpfile);
124 	if (err)
125 		goto failed_dat;
126 
127 	rawi = (void *)bh_sr->b_data + NILFS_SR_SUFILE_OFFSET(inode_size);
128 	err = nilfs_sufile_read(sb, segment_usage_size, rawi,
129 				&nilfs->ns_sufile);
130 	if (err)
131 		goto failed_cpfile;
132 
133 	raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
134 	nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
135 
136  failed:
137 	brelse(bh_sr);
138 	return err;
139 
140  failed_cpfile:
141 	iput(nilfs->ns_cpfile);
142 
143  failed_dat:
144 	iput(nilfs->ns_dat);
145 	goto failed;
146 }
147 
148 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
149 {
150 	memset(ri, 0, sizeof(*ri));
151 	INIT_LIST_HEAD(&ri->ri_used_segments);
152 }
153 
154 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
155 {
156 	nilfs_dispose_segment_list(&ri->ri_used_segments);
157 }
158 
159 /**
160  * nilfs_store_log_cursor - load log cursor from a super block
161  * @nilfs: nilfs object
162  * @sbp: buffer storing super block to be read
163  *
164  * nilfs_store_log_cursor() reads the last position of the log
165  * containing a super root from a given super block, and initializes
166  * relevant information on the nilfs object preparatory for log
167  * scanning and recovery.
168  */
169 static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
170 				  struct nilfs_super_block *sbp)
171 {
172 	int ret = 0;
173 
174 	nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
175 	nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
176 	nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
177 
178 	nilfs->ns_prev_seq = nilfs->ns_last_seq;
179 	nilfs->ns_seg_seq = nilfs->ns_last_seq;
180 	nilfs->ns_segnum =
181 		nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
182 	nilfs->ns_cno = nilfs->ns_last_cno + 1;
183 	if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
184 		nilfs_err(nilfs->ns_sb,
185 			  "pointed segment number is out of range: segnum=%llu, nsegments=%lu",
186 			  (unsigned long long)nilfs->ns_segnum,
187 			  nilfs->ns_nsegments);
188 		ret = -EINVAL;
189 	}
190 	return ret;
191 }
192 
193 /**
194  * nilfs_get_blocksize - get block size from raw superblock data
195  * @sb: super block instance
196  * @sbp: superblock raw data buffer
197  * @blocksize: place to store block size
198  *
199  * nilfs_get_blocksize() calculates the block size from the block size
200  * exponent information written in @sbp and stores it in @blocksize,
201  * or aborts with an error message if it's too large.
202  *
203  * Return Value: On success, 0 is returned. If the block size is too
204  * large, -EINVAL is returned.
205  */
206 static int nilfs_get_blocksize(struct super_block *sb,
207 			       struct nilfs_super_block *sbp, int *blocksize)
208 {
209 	unsigned int shift_bits = le32_to_cpu(sbp->s_log_block_size);
210 
211 	if (unlikely(shift_bits >
212 		     ilog2(NILFS_MAX_BLOCK_SIZE) - BLOCK_SIZE_BITS)) {
213 		nilfs_err(sb, "too large filesystem blocksize: 2 ^ %u KiB",
214 			  shift_bits);
215 		return -EINVAL;
216 	}
217 	*blocksize = BLOCK_SIZE << shift_bits;
218 	return 0;
219 }
220 
221 /**
222  * load_nilfs - load and recover the nilfs
223  * @nilfs: the_nilfs structure to be released
224  * @sb: super block instance used to recover past segment
225  *
226  * load_nilfs() searches and load the latest super root,
227  * attaches the last segment, and does recovery if needed.
228  * The caller must call this exclusively for simultaneous mounts.
229  */
230 int load_nilfs(struct the_nilfs *nilfs, struct super_block *sb)
231 {
232 	struct nilfs_recovery_info ri;
233 	unsigned int s_flags = sb->s_flags;
234 	int really_read_only = bdev_read_only(nilfs->ns_bdev);
235 	int valid_fs = nilfs_valid_fs(nilfs);
236 	int err;
237 
238 	if (!valid_fs) {
239 		nilfs_warn(sb, "mounting unchecked fs");
240 		if (s_flags & SB_RDONLY) {
241 			nilfs_info(sb,
242 				   "recovery required for readonly filesystem");
243 			nilfs_info(sb,
244 				   "write access will be enabled during recovery");
245 		}
246 	}
247 
248 	nilfs_init_recovery_info(&ri);
249 
250 	err = nilfs_search_super_root(nilfs, &ri);
251 	if (unlikely(err)) {
252 		struct nilfs_super_block **sbp = nilfs->ns_sbp;
253 		int blocksize;
254 
255 		if (err != -EINVAL)
256 			goto scan_error;
257 
258 		if (!nilfs_valid_sb(sbp[1])) {
259 			nilfs_warn(sb,
260 				   "unable to fall back to spare super block");
261 			goto scan_error;
262 		}
263 		nilfs_info(sb, "trying rollback from an earlier position");
264 
265 		/*
266 		 * restore super block with its spare and reconfigure
267 		 * relevant states of the nilfs object.
268 		 */
269 		memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
270 		nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
271 		nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
272 
273 		/* verify consistency between two super blocks */
274 		err = nilfs_get_blocksize(sb, sbp[0], &blocksize);
275 		if (err)
276 			goto scan_error;
277 
278 		if (blocksize != nilfs->ns_blocksize) {
279 			nilfs_warn(sb,
280 				   "blocksize differs between two super blocks (%d != %d)",
281 				   blocksize, nilfs->ns_blocksize);
282 			err = -EINVAL;
283 			goto scan_error;
284 		}
285 
286 		err = nilfs_store_log_cursor(nilfs, sbp[0]);
287 		if (err)
288 			goto scan_error;
289 
290 		/* drop clean flag to allow roll-forward and recovery */
291 		nilfs->ns_mount_state &= ~NILFS_VALID_FS;
292 		valid_fs = 0;
293 
294 		err = nilfs_search_super_root(nilfs, &ri);
295 		if (err)
296 			goto scan_error;
297 	}
298 
299 	err = nilfs_load_super_root(nilfs, sb, ri.ri_super_root);
300 	if (unlikely(err)) {
301 		nilfs_err(sb, "error %d while loading super root", err);
302 		goto failed;
303 	}
304 
305 	err = nilfs_sysfs_create_device_group(sb);
306 	if (unlikely(err))
307 		goto sysfs_error;
308 
309 	if (valid_fs)
310 		goto skip_recovery;
311 
312 	if (s_flags & SB_RDONLY) {
313 		__u64 features;
314 
315 		if (nilfs_test_opt(nilfs, NORECOVERY)) {
316 			nilfs_info(sb,
317 				   "norecovery option specified, skipping roll-forward recovery");
318 			goto skip_recovery;
319 		}
320 		features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
321 			~NILFS_FEATURE_COMPAT_RO_SUPP;
322 		if (features) {
323 			nilfs_err(sb,
324 				  "couldn't proceed with recovery because of unsupported optional features (%llx)",
325 				  (unsigned long long)features);
326 			err = -EROFS;
327 			goto failed_unload;
328 		}
329 		if (really_read_only) {
330 			nilfs_err(sb,
331 				  "write access unavailable, cannot proceed");
332 			err = -EROFS;
333 			goto failed_unload;
334 		}
335 		sb->s_flags &= ~SB_RDONLY;
336 	} else if (nilfs_test_opt(nilfs, NORECOVERY)) {
337 		nilfs_err(sb,
338 			  "recovery cancelled because norecovery option was specified for a read/write mount");
339 		err = -EINVAL;
340 		goto failed_unload;
341 	}
342 
343 	err = nilfs_salvage_orphan_logs(nilfs, sb, &ri);
344 	if (err)
345 		goto failed_unload;
346 
347 	down_write(&nilfs->ns_sem);
348 	nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
349 	err = nilfs_cleanup_super(sb);
350 	up_write(&nilfs->ns_sem);
351 
352 	if (err) {
353 		nilfs_err(sb,
354 			  "error %d updating super block. recovery unfinished.",
355 			  err);
356 		goto failed_unload;
357 	}
358 	nilfs_info(sb, "recovery complete");
359 
360  skip_recovery:
361 	nilfs_clear_recovery_info(&ri);
362 	sb->s_flags = s_flags;
363 	return 0;
364 
365  scan_error:
366 	nilfs_err(sb, "error %d while searching super root", err);
367 	goto failed;
368 
369  failed_unload:
370 	nilfs_sysfs_delete_device_group(nilfs);
371 
372  sysfs_error:
373 	iput(nilfs->ns_cpfile);
374 	iput(nilfs->ns_sufile);
375 	iput(nilfs->ns_dat);
376 
377  failed:
378 	nilfs_clear_recovery_info(&ri);
379 	sb->s_flags = s_flags;
380 	return err;
381 }
382 
383 static unsigned long long nilfs_max_size(unsigned int blkbits)
384 {
385 	unsigned int max_bits;
386 	unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
387 
388 	max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
389 	if (max_bits < 64)
390 		res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
391 	return res;
392 }
393 
394 /**
395  * nilfs_nrsvsegs - calculate the number of reserved segments
396  * @nilfs: nilfs object
397  * @nsegs: total number of segments
398  */
399 unsigned long nilfs_nrsvsegs(struct the_nilfs *nilfs, unsigned long nsegs)
400 {
401 	return max_t(unsigned long, NILFS_MIN_NRSVSEGS,
402 		     DIV_ROUND_UP(nsegs * nilfs->ns_r_segments_percentage,
403 				  100));
404 }
405 
406 /**
407  * nilfs_max_segment_count - calculate the maximum number of segments
408  * @nilfs: nilfs object
409  */
410 static u64 nilfs_max_segment_count(struct the_nilfs *nilfs)
411 {
412 	u64 max_count = U64_MAX;
413 
414 	max_count = div64_ul(max_count, nilfs->ns_blocks_per_segment);
415 	return min_t(u64, max_count, ULONG_MAX);
416 }
417 
418 void nilfs_set_nsegments(struct the_nilfs *nilfs, unsigned long nsegs)
419 {
420 	nilfs->ns_nsegments = nsegs;
421 	nilfs->ns_nrsvsegs = nilfs_nrsvsegs(nilfs, nsegs);
422 }
423 
424 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
425 				   struct nilfs_super_block *sbp)
426 {
427 	u64 nsegments, nblocks;
428 
429 	if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
430 		nilfs_err(nilfs->ns_sb,
431 			  "unsupported revision (superblock rev.=%d.%d, current rev.=%d.%d). Please check the version of mkfs.nilfs(2).",
432 			  le32_to_cpu(sbp->s_rev_level),
433 			  le16_to_cpu(sbp->s_minor_rev_level),
434 			  NILFS_CURRENT_REV, NILFS_MINOR_REV);
435 		return -EINVAL;
436 	}
437 	nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
438 	if (nilfs->ns_sbsize > BLOCK_SIZE)
439 		return -EINVAL;
440 
441 	nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
442 	if (nilfs->ns_inode_size > nilfs->ns_blocksize) {
443 		nilfs_err(nilfs->ns_sb, "too large inode size: %d bytes",
444 			  nilfs->ns_inode_size);
445 		return -EINVAL;
446 	} else if (nilfs->ns_inode_size < NILFS_MIN_INODE_SIZE) {
447 		nilfs_err(nilfs->ns_sb, "too small inode size: %d bytes",
448 			  nilfs->ns_inode_size);
449 		return -EINVAL;
450 	}
451 
452 	nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
453 	if (nilfs->ns_first_ino < NILFS_USER_INO) {
454 		nilfs_err(nilfs->ns_sb,
455 			  "too small lower limit for non-reserved inode numbers: %u",
456 			  nilfs->ns_first_ino);
457 		return -EINVAL;
458 	}
459 
460 	nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
461 	if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
462 		nilfs_err(nilfs->ns_sb, "too short segment: %lu blocks",
463 			  nilfs->ns_blocks_per_segment);
464 		return -EINVAL;
465 	}
466 
467 	nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
468 	nilfs->ns_r_segments_percentage =
469 		le32_to_cpu(sbp->s_r_segments_percentage);
470 	if (nilfs->ns_r_segments_percentage < 1 ||
471 	    nilfs->ns_r_segments_percentage > 99) {
472 		nilfs_err(nilfs->ns_sb,
473 			  "invalid reserved segments percentage: %lu",
474 			  nilfs->ns_r_segments_percentage);
475 		return -EINVAL;
476 	}
477 
478 	nsegments = le64_to_cpu(sbp->s_nsegments);
479 	if (nsegments > nilfs_max_segment_count(nilfs)) {
480 		nilfs_err(nilfs->ns_sb,
481 			  "segment count %llu exceeds upper limit (%llu segments)",
482 			  (unsigned long long)nsegments,
483 			  (unsigned long long)nilfs_max_segment_count(nilfs));
484 		return -EINVAL;
485 	}
486 
487 	nblocks = sb_bdev_nr_blocks(nilfs->ns_sb);
488 	if (nblocks) {
489 		u64 min_block_count = nsegments * nilfs->ns_blocks_per_segment;
490 		/*
491 		 * To avoid failing to mount early device images without a
492 		 * second superblock, exclude that block count from the
493 		 * "min_block_count" calculation.
494 		 */
495 
496 		if (nblocks < min_block_count) {
497 			nilfs_err(nilfs->ns_sb,
498 				  "total number of segment blocks %llu exceeds device size (%llu blocks)",
499 				  (unsigned long long)min_block_count,
500 				  (unsigned long long)nblocks);
501 			return -EINVAL;
502 		}
503 	}
504 
505 	nilfs_set_nsegments(nilfs, nsegments);
506 	nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
507 	return 0;
508 }
509 
510 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
511 {
512 	static unsigned char sum[4];
513 	const int sumoff = offsetof(struct nilfs_super_block, s_sum);
514 	size_t bytes;
515 	u32 crc;
516 
517 	if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
518 		return 0;
519 	bytes = le16_to_cpu(sbp->s_bytes);
520 	if (bytes < sumoff + 4 || bytes > BLOCK_SIZE)
521 		return 0;
522 	crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
523 		       sumoff);
524 	crc = crc32_le(crc, sum, 4);
525 	crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
526 		       bytes - sumoff - 4);
527 	return crc == le32_to_cpu(sbp->s_sum);
528 }
529 
530 /**
531  * nilfs_sb2_bad_offset - check the location of the second superblock
532  * @sbp: superblock raw data buffer
533  * @offset: byte offset of second superblock calculated from device size
534  *
535  * nilfs_sb2_bad_offset() checks if the position on the second
536  * superblock is valid or not based on the filesystem parameters
537  * stored in @sbp.  If @offset points to a location within the segment
538  * area, or if the parameters themselves are not normal, it is
539  * determined to be invalid.
540  *
541  * Return Value: true if invalid, false if valid.
542  */
543 static bool nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
544 {
545 	unsigned int shift_bits = le32_to_cpu(sbp->s_log_block_size);
546 	u32 blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
547 	u64 nsegments = le64_to_cpu(sbp->s_nsegments);
548 	u64 index;
549 
550 	if (blocks_per_segment < NILFS_SEG_MIN_BLOCKS ||
551 	    shift_bits > ilog2(NILFS_MAX_BLOCK_SIZE) - BLOCK_SIZE_BITS)
552 		return true;
553 
554 	index = offset >> (shift_bits + BLOCK_SIZE_BITS);
555 	do_div(index, blocks_per_segment);
556 	return index < nsegments;
557 }
558 
559 static void nilfs_release_super_block(struct the_nilfs *nilfs)
560 {
561 	int i;
562 
563 	for (i = 0; i < 2; i++) {
564 		if (nilfs->ns_sbp[i]) {
565 			brelse(nilfs->ns_sbh[i]);
566 			nilfs->ns_sbh[i] = NULL;
567 			nilfs->ns_sbp[i] = NULL;
568 		}
569 	}
570 }
571 
572 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
573 {
574 	brelse(nilfs->ns_sbh[0]);
575 	nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
576 	nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
577 	nilfs->ns_sbh[1] = NULL;
578 	nilfs->ns_sbp[1] = NULL;
579 }
580 
581 void nilfs_swap_super_block(struct the_nilfs *nilfs)
582 {
583 	struct buffer_head *tsbh = nilfs->ns_sbh[0];
584 	struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
585 
586 	nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
587 	nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
588 	nilfs->ns_sbh[1] = tsbh;
589 	nilfs->ns_sbp[1] = tsbp;
590 }
591 
592 static int nilfs_load_super_block(struct the_nilfs *nilfs,
593 				  struct super_block *sb, int blocksize,
594 				  struct nilfs_super_block **sbpp)
595 {
596 	struct nilfs_super_block **sbp = nilfs->ns_sbp;
597 	struct buffer_head **sbh = nilfs->ns_sbh;
598 	u64 sb2off, devsize = bdev_nr_bytes(nilfs->ns_bdev);
599 	int valid[2], swp = 0, older;
600 
601 	if (devsize < NILFS_SEG_MIN_BLOCKS * NILFS_MIN_BLOCK_SIZE + 4096) {
602 		nilfs_err(sb, "device size too small");
603 		return -EINVAL;
604 	}
605 	sb2off = NILFS_SB2_OFFSET_BYTES(devsize);
606 
607 	sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
608 					&sbh[0]);
609 	sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
610 
611 	if (!sbp[0]) {
612 		if (!sbp[1]) {
613 			nilfs_err(sb, "unable to read superblock");
614 			return -EIO;
615 		}
616 		nilfs_warn(sb,
617 			   "unable to read primary superblock (blocksize = %d)",
618 			   blocksize);
619 	} else if (!sbp[1]) {
620 		nilfs_warn(sb,
621 			   "unable to read secondary superblock (blocksize = %d)",
622 			   blocksize);
623 	}
624 
625 	/*
626 	 * Compare two super blocks and set 1 in swp if the secondary
627 	 * super block is valid and newer.  Otherwise, set 0 in swp.
628 	 */
629 	valid[0] = nilfs_valid_sb(sbp[0]);
630 	valid[1] = nilfs_valid_sb(sbp[1]);
631 	swp = valid[1] && (!valid[0] ||
632 			   le64_to_cpu(sbp[1]->s_last_cno) >
633 			   le64_to_cpu(sbp[0]->s_last_cno));
634 
635 	if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
636 		brelse(sbh[1]);
637 		sbh[1] = NULL;
638 		sbp[1] = NULL;
639 		valid[1] = 0;
640 		swp = 0;
641 	}
642 	if (!valid[swp]) {
643 		nilfs_release_super_block(nilfs);
644 		nilfs_err(sb, "couldn't find nilfs on the device");
645 		return -EINVAL;
646 	}
647 
648 	if (!valid[!swp])
649 		nilfs_warn(sb,
650 			   "broken superblock, retrying with spare superblock (blocksize = %d)",
651 			   blocksize);
652 	if (swp)
653 		nilfs_swap_super_block(nilfs);
654 
655 	/*
656 	 * Calculate the array index of the older superblock data.
657 	 * If one has been dropped, set index 0 pointing to the remaining one,
658 	 * otherwise set index 1 pointing to the old one (including if both
659 	 * are the same).
660 	 *
661 	 *  Divided case             valid[0]  valid[1]  swp  ->  older
662 	 *  -------------------------------------------------------------
663 	 *  Both SBs are invalid        0         0       N/A (Error)
664 	 *  SB1 is invalid              0         1       1         0
665 	 *  SB2 is invalid              1         0       0         0
666 	 *  SB2 is newer                1         1       1         0
667 	 *  SB2 is older or the same    1         1       0         1
668 	 */
669 	older = valid[1] ^ swp;
670 
671 	nilfs->ns_sbwcount = 0;
672 	nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
673 	nilfs->ns_prot_seq = le64_to_cpu(sbp[older]->s_last_seq);
674 	*sbpp = sbp[0];
675 	return 0;
676 }
677 
678 /**
679  * init_nilfs - initialize a NILFS instance.
680  * @nilfs: the_nilfs structure
681  * @sb: super block
682  *
683  * init_nilfs() performs common initialization per block device (e.g.
684  * reading the super block, getting disk layout information, initializing
685  * shared fields in the_nilfs).
686  *
687  * Return Value: On success, 0 is returned. On error, a negative error
688  * code is returned.
689  */
690 int init_nilfs(struct the_nilfs *nilfs, struct super_block *sb)
691 {
692 	struct nilfs_super_block *sbp;
693 	int blocksize;
694 	int err;
695 
696 	down_write(&nilfs->ns_sem);
697 
698 	blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
699 	if (!blocksize) {
700 		nilfs_err(sb, "unable to set blocksize");
701 		err = -EINVAL;
702 		goto out;
703 	}
704 	err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
705 	if (err)
706 		goto out;
707 
708 	err = nilfs_store_magic(sb, sbp);
709 	if (err)
710 		goto failed_sbh;
711 
712 	err = nilfs_check_feature_compatibility(sb, sbp);
713 	if (err)
714 		goto failed_sbh;
715 
716 	err = nilfs_get_blocksize(sb, sbp, &blocksize);
717 	if (err)
718 		goto failed_sbh;
719 
720 	if (blocksize < NILFS_MIN_BLOCK_SIZE) {
721 		nilfs_err(sb,
722 			  "couldn't mount because of unsupported filesystem blocksize %d",
723 			  blocksize);
724 		err = -EINVAL;
725 		goto failed_sbh;
726 	}
727 	if (sb->s_blocksize != blocksize) {
728 		int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
729 
730 		if (blocksize < hw_blocksize) {
731 			nilfs_err(sb,
732 				  "blocksize %d too small for device (sector-size = %d)",
733 				  blocksize, hw_blocksize);
734 			err = -EINVAL;
735 			goto failed_sbh;
736 		}
737 		nilfs_release_super_block(nilfs);
738 		if (!sb_set_blocksize(sb, blocksize)) {
739 			nilfs_err(sb, "bad blocksize %d", blocksize);
740 			err = -EINVAL;
741 			goto out;
742 		}
743 
744 		err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
745 		if (err)
746 			goto out;
747 			/*
748 			 * Not to failed_sbh; sbh is released automatically
749 			 * when reloading fails.
750 			 */
751 	}
752 	nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
753 	nilfs->ns_blocksize = blocksize;
754 
755 	err = nilfs_store_disk_layout(nilfs, sbp);
756 	if (err)
757 		goto failed_sbh;
758 
759 	sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
760 
761 	nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
762 
763 	err = nilfs_store_log_cursor(nilfs, sbp);
764 	if (err)
765 		goto failed_sbh;
766 
767 	set_nilfs_init(nilfs);
768 	err = 0;
769  out:
770 	up_write(&nilfs->ns_sem);
771 	return err;
772 
773  failed_sbh:
774 	nilfs_release_super_block(nilfs);
775 	goto out;
776 }
777 
778 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
779 			    size_t nsegs)
780 {
781 	sector_t seg_start, seg_end;
782 	sector_t start = 0, nblocks = 0;
783 	unsigned int sects_per_block;
784 	__u64 *sn;
785 	int ret = 0;
786 
787 	sects_per_block = (1 << nilfs->ns_blocksize_bits) /
788 		bdev_logical_block_size(nilfs->ns_bdev);
789 	for (sn = segnump; sn < segnump + nsegs; sn++) {
790 		nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
791 
792 		if (!nblocks) {
793 			start = seg_start;
794 			nblocks = seg_end - seg_start + 1;
795 		} else if (start + nblocks == seg_start) {
796 			nblocks += seg_end - seg_start + 1;
797 		} else {
798 			ret = blkdev_issue_discard(nilfs->ns_bdev,
799 						   start * sects_per_block,
800 						   nblocks * sects_per_block,
801 						   GFP_NOFS);
802 			if (ret < 0)
803 				return ret;
804 			nblocks = 0;
805 		}
806 	}
807 	if (nblocks)
808 		ret = blkdev_issue_discard(nilfs->ns_bdev,
809 					   start * sects_per_block,
810 					   nblocks * sects_per_block,
811 					   GFP_NOFS);
812 	return ret;
813 }
814 
815 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
816 {
817 	unsigned long ncleansegs;
818 
819 	ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
820 	*nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
821 	return 0;
822 }
823 
824 int nilfs_near_disk_full(struct the_nilfs *nilfs)
825 {
826 	unsigned long ncleansegs, nincsegs;
827 
828 	ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
829 	nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
830 		nilfs->ns_blocks_per_segment + 1;
831 
832 	return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
833 }
834 
835 struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
836 {
837 	struct rb_node *n;
838 	struct nilfs_root *root;
839 
840 	spin_lock(&nilfs->ns_cptree_lock);
841 	n = nilfs->ns_cptree.rb_node;
842 	while (n) {
843 		root = rb_entry(n, struct nilfs_root, rb_node);
844 
845 		if (cno < root->cno) {
846 			n = n->rb_left;
847 		} else if (cno > root->cno) {
848 			n = n->rb_right;
849 		} else {
850 			refcount_inc(&root->count);
851 			spin_unlock(&nilfs->ns_cptree_lock);
852 			return root;
853 		}
854 	}
855 	spin_unlock(&nilfs->ns_cptree_lock);
856 
857 	return NULL;
858 }
859 
860 struct nilfs_root *
861 nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
862 {
863 	struct rb_node **p, *parent;
864 	struct nilfs_root *root, *new;
865 	int err;
866 
867 	root = nilfs_lookup_root(nilfs, cno);
868 	if (root)
869 		return root;
870 
871 	new = kzalloc(sizeof(*root), GFP_KERNEL);
872 	if (!new)
873 		return NULL;
874 
875 	spin_lock(&nilfs->ns_cptree_lock);
876 
877 	p = &nilfs->ns_cptree.rb_node;
878 	parent = NULL;
879 
880 	while (*p) {
881 		parent = *p;
882 		root = rb_entry(parent, struct nilfs_root, rb_node);
883 
884 		if (cno < root->cno) {
885 			p = &(*p)->rb_left;
886 		} else if (cno > root->cno) {
887 			p = &(*p)->rb_right;
888 		} else {
889 			refcount_inc(&root->count);
890 			spin_unlock(&nilfs->ns_cptree_lock);
891 			kfree(new);
892 			return root;
893 		}
894 	}
895 
896 	new->cno = cno;
897 	new->ifile = NULL;
898 	new->nilfs = nilfs;
899 	refcount_set(&new->count, 1);
900 	atomic64_set(&new->inodes_count, 0);
901 	atomic64_set(&new->blocks_count, 0);
902 
903 	rb_link_node(&new->rb_node, parent, p);
904 	rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
905 
906 	spin_unlock(&nilfs->ns_cptree_lock);
907 
908 	err = nilfs_sysfs_create_snapshot_group(new);
909 	if (err) {
910 		kfree(new);
911 		new = NULL;
912 	}
913 
914 	return new;
915 }
916 
917 void nilfs_put_root(struct nilfs_root *root)
918 {
919 	struct the_nilfs *nilfs = root->nilfs;
920 
921 	if (refcount_dec_and_lock(&root->count, &nilfs->ns_cptree_lock)) {
922 		rb_erase(&root->rb_node, &nilfs->ns_cptree);
923 		spin_unlock(&nilfs->ns_cptree_lock);
924 
925 		nilfs_sysfs_delete_snapshot_group(root);
926 		iput(root->ifile);
927 
928 		kfree(root);
929 	}
930 }
931