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/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
nilfs_set_last_segment(struct the_nilfs * nilfs,sector_t start_blocknr,u64 seq,__u64 cno)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 */
alloc_nilfs(struct super_block * sb)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 */
destroy_nilfs(struct the_nilfs * nilfs)86 void destroy_nilfs(struct the_nilfs *nilfs)
87 {
88 might_sleep();
89 if (nilfs_init(nilfs)) {
90 brelse(nilfs->ns_sbh[0]);
91 brelse(nilfs->ns_sbh[1]);
92 }
93 kfree(nilfs);
94 }
95
nilfs_load_super_root(struct the_nilfs * nilfs,struct super_block * sb,sector_t sr_block)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
nilfs_init_recovery_info(struct nilfs_recovery_info * ri)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
nilfs_clear_recovery_info(struct nilfs_recovery_info * ri)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 */
nilfs_store_log_cursor(struct the_nilfs * nilfs,struct nilfs_super_block * sbp)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 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 */
load_nilfs(struct the_nilfs * nilfs,struct super_block * sb)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 err = nilfs_sysfs_create_device_group(sb);
276 if (unlikely(err))
277 goto sysfs_error;
278
279 if (valid_fs)
280 goto skip_recovery;
281
282 if (s_flags & SB_RDONLY) {
283 __u64 features;
284
285 if (nilfs_test_opt(nilfs, NORECOVERY)) {
286 nilfs_info(sb,
287 "norecovery option specified, skipping roll-forward recovery");
288 goto skip_recovery;
289 }
290 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
291 ~NILFS_FEATURE_COMPAT_RO_SUPP;
292 if (features) {
293 nilfs_err(sb,
294 "couldn't proceed with recovery because of unsupported optional features (%llx)",
295 (unsigned long long)features);
296 err = -EROFS;
297 goto failed_unload;
298 }
299 if (really_read_only) {
300 nilfs_err(sb,
301 "write access unavailable, cannot proceed");
302 err = -EROFS;
303 goto failed_unload;
304 }
305 sb->s_flags &= ~SB_RDONLY;
306 } else if (nilfs_test_opt(nilfs, NORECOVERY)) {
307 nilfs_err(sb,
308 "recovery cancelled because norecovery option was specified for a read/write mount");
309 err = -EINVAL;
310 goto failed_unload;
311 }
312
313 err = nilfs_salvage_orphan_logs(nilfs, sb, &ri);
314 if (err)
315 goto failed_unload;
316
317 down_write(&nilfs->ns_sem);
318 nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
319 err = nilfs_cleanup_super(sb);
320 up_write(&nilfs->ns_sem);
321
322 if (err) {
323 nilfs_err(sb,
324 "error %d updating super block. recovery unfinished.",
325 err);
326 goto failed_unload;
327 }
328 nilfs_info(sb, "recovery complete");
329
330 skip_recovery:
331 nilfs_clear_recovery_info(&ri);
332 sb->s_flags = s_flags;
333 return 0;
334
335 scan_error:
336 nilfs_err(sb, "error %d while searching super root", err);
337 goto failed;
338
339 failed_unload:
340 nilfs_sysfs_delete_device_group(nilfs);
341
342 sysfs_error:
343 iput(nilfs->ns_cpfile);
344 iput(nilfs->ns_sufile);
345 iput(nilfs->ns_dat);
346
347 failed:
348 nilfs_clear_recovery_info(&ri);
349 sb->s_flags = s_flags;
350 return err;
351 }
352
nilfs_max_size(unsigned int blkbits)353 static unsigned long long nilfs_max_size(unsigned int blkbits)
354 {
355 unsigned int max_bits;
356 unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
357
358 max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
359 if (max_bits < 64)
360 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
361 return res;
362 }
363
364 /**
365 * nilfs_nrsvsegs - calculate the number of reserved segments
366 * @nilfs: nilfs object
367 * @nsegs: total number of segments
368 */
nilfs_nrsvsegs(struct the_nilfs * nilfs,unsigned long nsegs)369 unsigned long nilfs_nrsvsegs(struct the_nilfs *nilfs, unsigned long nsegs)
370 {
371 return max_t(unsigned long, NILFS_MIN_NRSVSEGS,
372 DIV_ROUND_UP(nsegs * nilfs->ns_r_segments_percentage,
373 100));
374 }
375
376 /**
377 * nilfs_max_segment_count - calculate the maximum number of segments
378 * @nilfs: nilfs object
379 */
nilfs_max_segment_count(struct the_nilfs * nilfs)380 static u64 nilfs_max_segment_count(struct the_nilfs *nilfs)
381 {
382 u64 max_count = U64_MAX;
383
384 do_div(max_count, nilfs->ns_blocks_per_segment);
385 return min_t(u64, max_count, ULONG_MAX);
386 }
387
nilfs_set_nsegments(struct the_nilfs * nilfs,unsigned long nsegs)388 void nilfs_set_nsegments(struct the_nilfs *nilfs, unsigned long nsegs)
389 {
390 nilfs->ns_nsegments = nsegs;
391 nilfs->ns_nrsvsegs = nilfs_nrsvsegs(nilfs, nsegs);
392 }
393
nilfs_store_disk_layout(struct the_nilfs * nilfs,struct nilfs_super_block * sbp)394 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
395 struct nilfs_super_block *sbp)
396 {
397 u64 nsegments, nblocks;
398
399 if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
400 nilfs_err(nilfs->ns_sb,
401 "unsupported revision (superblock rev.=%d.%d, current rev.=%d.%d). Please check the version of mkfs.nilfs(2).",
402 le32_to_cpu(sbp->s_rev_level),
403 le16_to_cpu(sbp->s_minor_rev_level),
404 NILFS_CURRENT_REV, NILFS_MINOR_REV);
405 return -EINVAL;
406 }
407 nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
408 if (nilfs->ns_sbsize > BLOCK_SIZE)
409 return -EINVAL;
410
411 nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
412 if (nilfs->ns_inode_size > nilfs->ns_blocksize) {
413 nilfs_err(nilfs->ns_sb, "too large inode size: %d bytes",
414 nilfs->ns_inode_size);
415 return -EINVAL;
416 } else if (nilfs->ns_inode_size < NILFS_MIN_INODE_SIZE) {
417 nilfs_err(nilfs->ns_sb, "too small inode size: %d bytes",
418 nilfs->ns_inode_size);
419 return -EINVAL;
420 }
421
422 nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
423
424 nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
425 if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
426 nilfs_err(nilfs->ns_sb, "too short segment: %lu blocks",
427 nilfs->ns_blocks_per_segment);
428 return -EINVAL;
429 }
430
431 nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
432 nilfs->ns_r_segments_percentage =
433 le32_to_cpu(sbp->s_r_segments_percentage);
434 if (nilfs->ns_r_segments_percentage < 1 ||
435 nilfs->ns_r_segments_percentage > 99) {
436 nilfs_err(nilfs->ns_sb,
437 "invalid reserved segments percentage: %lu",
438 nilfs->ns_r_segments_percentage);
439 return -EINVAL;
440 }
441
442 nsegments = le64_to_cpu(sbp->s_nsegments);
443 if (nsegments > nilfs_max_segment_count(nilfs)) {
444 nilfs_msg(nilfs->ns_sb, KERN_ERR,
445 "segment count %llu exceeds upper limit (%llu segments)",
446 (unsigned long long)nsegments,
447 (unsigned long long)nilfs_max_segment_count(nilfs));
448 return -EINVAL;
449 }
450
451 nblocks = (u64)i_size_read(nilfs->ns_sb->s_bdev->bd_inode) >>
452 nilfs->ns_sb->s_blocksize_bits;
453 if (nblocks) {
454 u64 min_block_count = nsegments * nilfs->ns_blocks_per_segment;
455 /*
456 * To avoid failing to mount early device images without a
457 * second superblock, exclude that block count from the
458 * "min_block_count" calculation.
459 */
460
461 if (nblocks < min_block_count) {
462 nilfs_msg(nilfs->ns_sb, KERN_ERR,
463 "total number of segment blocks %llu exceeds device size (%llu blocks)",
464 (unsigned long long)min_block_count,
465 (unsigned long long)nblocks);
466 return -EINVAL;
467 }
468 }
469
470 nilfs_set_nsegments(nilfs, nsegments);
471 nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
472 return 0;
473 }
474
nilfs_valid_sb(struct nilfs_super_block * sbp)475 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
476 {
477 static unsigned char sum[4];
478 const int sumoff = offsetof(struct nilfs_super_block, s_sum);
479 size_t bytes;
480 u32 crc;
481
482 if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
483 return 0;
484 bytes = le16_to_cpu(sbp->s_bytes);
485 if (bytes < sumoff + 4 || bytes > BLOCK_SIZE)
486 return 0;
487 crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
488 sumoff);
489 crc = crc32_le(crc, sum, 4);
490 crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
491 bytes - sumoff - 4);
492 return crc == le32_to_cpu(sbp->s_sum);
493 }
494
495 /**
496 * nilfs_sb2_bad_offset - check the location of the second superblock
497 * @sbp: superblock raw data buffer
498 * @offset: byte offset of second superblock calculated from device size
499 *
500 * nilfs_sb2_bad_offset() checks if the position on the second
501 * superblock is valid or not based on the filesystem parameters
502 * stored in @sbp. If @offset points to a location within the segment
503 * area, or if the parameters themselves are not normal, it is
504 * determined to be invalid.
505 *
506 * Return Value: true if invalid, false if valid.
507 */
nilfs_sb2_bad_offset(struct nilfs_super_block * sbp,u64 offset)508 static bool nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
509 {
510 unsigned int shift_bits = le32_to_cpu(sbp->s_log_block_size);
511 u32 blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
512 u64 nsegments = le64_to_cpu(sbp->s_nsegments);
513 u64 index;
514
515 if (blocks_per_segment < NILFS_SEG_MIN_BLOCKS ||
516 shift_bits > ilog2(NILFS_MAX_BLOCK_SIZE) - BLOCK_SIZE_BITS)
517 return true;
518
519 index = offset >> (shift_bits + BLOCK_SIZE_BITS);
520 do_div(index, blocks_per_segment);
521 return index < nsegments;
522 }
523
nilfs_release_super_block(struct the_nilfs * nilfs)524 static void nilfs_release_super_block(struct the_nilfs *nilfs)
525 {
526 int i;
527
528 for (i = 0; i < 2; i++) {
529 if (nilfs->ns_sbp[i]) {
530 brelse(nilfs->ns_sbh[i]);
531 nilfs->ns_sbh[i] = NULL;
532 nilfs->ns_sbp[i] = NULL;
533 }
534 }
535 }
536
nilfs_fall_back_super_block(struct the_nilfs * nilfs)537 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
538 {
539 brelse(nilfs->ns_sbh[0]);
540 nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
541 nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
542 nilfs->ns_sbh[1] = NULL;
543 nilfs->ns_sbp[1] = NULL;
544 }
545
nilfs_swap_super_block(struct the_nilfs * nilfs)546 void nilfs_swap_super_block(struct the_nilfs *nilfs)
547 {
548 struct buffer_head *tsbh = nilfs->ns_sbh[0];
549 struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
550
551 nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
552 nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
553 nilfs->ns_sbh[1] = tsbh;
554 nilfs->ns_sbp[1] = tsbp;
555 }
556
nilfs_load_super_block(struct the_nilfs * nilfs,struct super_block * sb,int blocksize,struct nilfs_super_block ** sbpp)557 static int nilfs_load_super_block(struct the_nilfs *nilfs,
558 struct super_block *sb, int blocksize,
559 struct nilfs_super_block **sbpp)
560 {
561 struct nilfs_super_block **sbp = nilfs->ns_sbp;
562 struct buffer_head **sbh = nilfs->ns_sbh;
563 u64 sb2off, devsize = nilfs->ns_bdev->bd_inode->i_size;
564 int valid[2], swp = 0;
565
566 if (devsize < NILFS_SEG_MIN_BLOCKS * NILFS_MIN_BLOCK_SIZE + 4096) {
567 nilfs_msg(sb, KERN_ERR, "device size too small");
568 return -EINVAL;
569 }
570 sb2off = NILFS_SB2_OFFSET_BYTES(devsize);
571
572 sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
573 &sbh[0]);
574 sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
575
576 if (!sbp[0]) {
577 if (!sbp[1]) {
578 nilfs_err(sb, "unable to read superblock");
579 return -EIO;
580 }
581 nilfs_warn(sb,
582 "unable to read primary superblock (blocksize = %d)",
583 blocksize);
584 } else if (!sbp[1]) {
585 nilfs_warn(sb,
586 "unable to read secondary superblock (blocksize = %d)",
587 blocksize);
588 }
589
590 /*
591 * Compare two super blocks and set 1 in swp if the secondary
592 * super block is valid and newer. Otherwise, set 0 in swp.
593 */
594 valid[0] = nilfs_valid_sb(sbp[0]);
595 valid[1] = nilfs_valid_sb(sbp[1]);
596 swp = valid[1] && (!valid[0] ||
597 le64_to_cpu(sbp[1]->s_last_cno) >
598 le64_to_cpu(sbp[0]->s_last_cno));
599
600 if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
601 brelse(sbh[1]);
602 sbh[1] = NULL;
603 sbp[1] = NULL;
604 valid[1] = 0;
605 swp = 0;
606 }
607 if (!valid[swp]) {
608 nilfs_release_super_block(nilfs);
609 nilfs_err(sb, "couldn't find nilfs on the device");
610 return -EINVAL;
611 }
612
613 if (!valid[!swp])
614 nilfs_warn(sb,
615 "broken superblock, retrying with spare superblock (blocksize = %d)",
616 blocksize);
617 if (swp)
618 nilfs_swap_super_block(nilfs);
619
620 nilfs->ns_sbwcount = 0;
621 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
622 nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
623 *sbpp = sbp[0];
624 return 0;
625 }
626
627 /**
628 * init_nilfs - initialize a NILFS instance.
629 * @nilfs: the_nilfs structure
630 * @sb: super block
631 * @data: mount options
632 *
633 * init_nilfs() performs common initialization per block device (e.g.
634 * reading the super block, getting disk layout information, initializing
635 * shared fields in the_nilfs).
636 *
637 * Return Value: On success, 0 is returned. On error, a negative error
638 * code is returned.
639 */
init_nilfs(struct the_nilfs * nilfs,struct super_block * sb,char * data)640 int init_nilfs(struct the_nilfs *nilfs, struct super_block *sb, char *data)
641 {
642 struct nilfs_super_block *sbp;
643 int blocksize;
644 int err;
645
646 down_write(&nilfs->ns_sem);
647
648 blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
649 if (!blocksize) {
650 nilfs_err(sb, "unable to set blocksize");
651 err = -EINVAL;
652 goto out;
653 }
654 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
655 if (err)
656 goto out;
657
658 err = nilfs_store_magic_and_option(sb, sbp, data);
659 if (err)
660 goto failed_sbh;
661
662 err = nilfs_check_feature_compatibility(sb, sbp);
663 if (err)
664 goto failed_sbh;
665
666 blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
667 if (blocksize < NILFS_MIN_BLOCK_SIZE ||
668 blocksize > NILFS_MAX_BLOCK_SIZE) {
669 nilfs_err(sb,
670 "couldn't mount because of unsupported filesystem blocksize %d",
671 blocksize);
672 err = -EINVAL;
673 goto failed_sbh;
674 }
675 if (sb->s_blocksize != blocksize) {
676 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
677
678 if (blocksize < hw_blocksize) {
679 nilfs_err(sb,
680 "blocksize %d too small for device (sector-size = %d)",
681 blocksize, hw_blocksize);
682 err = -EINVAL;
683 goto failed_sbh;
684 }
685 nilfs_release_super_block(nilfs);
686 if (!sb_set_blocksize(sb, blocksize)) {
687 nilfs_msg(sb, KERN_ERR, "bad blocksize %d", blocksize);
688 err = -EINVAL;
689 goto out;
690 }
691
692 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
693 if (err)
694 goto out;
695 /*
696 * Not to failed_sbh; sbh is released automatically
697 * when reloading fails.
698 */
699 }
700 nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
701 nilfs->ns_blocksize = blocksize;
702
703 get_random_bytes(&nilfs->ns_next_generation,
704 sizeof(nilfs->ns_next_generation));
705
706 err = nilfs_store_disk_layout(nilfs, sbp);
707 if (err)
708 goto failed_sbh;
709
710 sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
711
712 nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
713
714 err = nilfs_store_log_cursor(nilfs, sbp);
715 if (err)
716 goto failed_sbh;
717
718 set_nilfs_init(nilfs);
719 err = 0;
720 out:
721 up_write(&nilfs->ns_sem);
722 return err;
723
724 failed_sbh:
725 nilfs_release_super_block(nilfs);
726 goto out;
727 }
728
nilfs_discard_segments(struct the_nilfs * nilfs,__u64 * segnump,size_t nsegs)729 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
730 size_t nsegs)
731 {
732 sector_t seg_start, seg_end;
733 sector_t start = 0, nblocks = 0;
734 unsigned int sects_per_block;
735 __u64 *sn;
736 int ret = 0;
737
738 sects_per_block = (1 << nilfs->ns_blocksize_bits) /
739 bdev_logical_block_size(nilfs->ns_bdev);
740 for (sn = segnump; sn < segnump + nsegs; sn++) {
741 nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
742
743 if (!nblocks) {
744 start = seg_start;
745 nblocks = seg_end - seg_start + 1;
746 } else if (start + nblocks == seg_start) {
747 nblocks += seg_end - seg_start + 1;
748 } else {
749 ret = blkdev_issue_discard(nilfs->ns_bdev,
750 start * sects_per_block,
751 nblocks * sects_per_block,
752 GFP_NOFS, 0);
753 if (ret < 0)
754 return ret;
755 nblocks = 0;
756 }
757 }
758 if (nblocks)
759 ret = blkdev_issue_discard(nilfs->ns_bdev,
760 start * sects_per_block,
761 nblocks * sects_per_block,
762 GFP_NOFS, 0);
763 return ret;
764 }
765
nilfs_count_free_blocks(struct the_nilfs * nilfs,sector_t * nblocks)766 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
767 {
768 unsigned long ncleansegs;
769
770 ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
771 *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
772 return 0;
773 }
774
nilfs_near_disk_full(struct the_nilfs * nilfs)775 int nilfs_near_disk_full(struct the_nilfs *nilfs)
776 {
777 unsigned long ncleansegs, nincsegs;
778
779 ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
780 nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
781 nilfs->ns_blocks_per_segment + 1;
782
783 return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
784 }
785
nilfs_lookup_root(struct the_nilfs * nilfs,__u64 cno)786 struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
787 {
788 struct rb_node *n;
789 struct nilfs_root *root;
790
791 spin_lock(&nilfs->ns_cptree_lock);
792 n = nilfs->ns_cptree.rb_node;
793 while (n) {
794 root = rb_entry(n, struct nilfs_root, rb_node);
795
796 if (cno < root->cno) {
797 n = n->rb_left;
798 } else if (cno > root->cno) {
799 n = n->rb_right;
800 } else {
801 refcount_inc(&root->count);
802 spin_unlock(&nilfs->ns_cptree_lock);
803 return root;
804 }
805 }
806 spin_unlock(&nilfs->ns_cptree_lock);
807
808 return NULL;
809 }
810
811 struct nilfs_root *
nilfs_find_or_create_root(struct the_nilfs * nilfs,__u64 cno)812 nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
813 {
814 struct rb_node **p, *parent;
815 struct nilfs_root *root, *new;
816 int err;
817
818 root = nilfs_lookup_root(nilfs, cno);
819 if (root)
820 return root;
821
822 new = kzalloc(sizeof(*root), GFP_KERNEL);
823 if (!new)
824 return NULL;
825
826 spin_lock(&nilfs->ns_cptree_lock);
827
828 p = &nilfs->ns_cptree.rb_node;
829 parent = NULL;
830
831 while (*p) {
832 parent = *p;
833 root = rb_entry(parent, struct nilfs_root, rb_node);
834
835 if (cno < root->cno) {
836 p = &(*p)->rb_left;
837 } else if (cno > root->cno) {
838 p = &(*p)->rb_right;
839 } else {
840 refcount_inc(&root->count);
841 spin_unlock(&nilfs->ns_cptree_lock);
842 kfree(new);
843 return root;
844 }
845 }
846
847 new->cno = cno;
848 new->ifile = NULL;
849 new->nilfs = nilfs;
850 refcount_set(&new->count, 1);
851 atomic64_set(&new->inodes_count, 0);
852 atomic64_set(&new->blocks_count, 0);
853
854 rb_link_node(&new->rb_node, parent, p);
855 rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
856
857 spin_unlock(&nilfs->ns_cptree_lock);
858
859 err = nilfs_sysfs_create_snapshot_group(new);
860 if (err) {
861 kfree(new);
862 new = NULL;
863 }
864
865 return new;
866 }
867
nilfs_put_root(struct nilfs_root * root)868 void nilfs_put_root(struct nilfs_root *root)
869 {
870 struct the_nilfs *nilfs = root->nilfs;
871
872 if (refcount_dec_and_lock(&root->count, &nilfs->ns_cptree_lock)) {
873 rb_erase(&root->rb_node, &nilfs->ns_cptree);
874 spin_unlock(&nilfs->ns_cptree_lock);
875
876 nilfs_sysfs_delete_snapshot_group(root);
877 iput(root->ifile);
878
879 kfree(root);
880 }
881 }
882