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