Loading...
1/*
2 * the_nilfs.c - the_nilfs shared structure.
3 *
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * Written by Ryusuke Konishi.
17 *
18 */
19
20#include <linux/buffer_head.h>
21#include <linux/slab.h>
22#include <linux/blkdev.h>
23#include <linux/backing-dev.h>
24#include <linux/random.h>
25#include <linux/crc32.h>
26#include "nilfs.h"
27#include "segment.h"
28#include "alloc.h"
29#include "cpfile.h"
30#include "sufile.h"
31#include "dat.h"
32#include "segbuf.h"
33
34
35static int nilfs_valid_sb(struct nilfs_super_block *sbp);
36
37void nilfs_set_last_segment(struct the_nilfs *nilfs,
38 sector_t start_blocknr, u64 seq, __u64 cno)
39{
40 spin_lock(&nilfs->ns_last_segment_lock);
41 nilfs->ns_last_pseg = start_blocknr;
42 nilfs->ns_last_seq = seq;
43 nilfs->ns_last_cno = cno;
44
45 if (!nilfs_sb_dirty(nilfs)) {
46 if (nilfs->ns_prev_seq == nilfs->ns_last_seq)
47 goto stay_cursor;
48
49 set_nilfs_sb_dirty(nilfs);
50 }
51 nilfs->ns_prev_seq = nilfs->ns_last_seq;
52
53 stay_cursor:
54 spin_unlock(&nilfs->ns_last_segment_lock);
55}
56
57/**
58 * alloc_nilfs - allocate a nilfs object
59 * @sb: super block instance
60 *
61 * Return Value: On success, pointer to the_nilfs is returned.
62 * On error, NULL is returned.
63 */
64struct the_nilfs *alloc_nilfs(struct super_block *sb)
65{
66 struct the_nilfs *nilfs;
67
68 nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
69 if (!nilfs)
70 return NULL;
71
72 nilfs->ns_sb = sb;
73 nilfs->ns_bdev = sb->s_bdev;
74 atomic_set(&nilfs->ns_ndirtyblks, 0);
75 init_rwsem(&nilfs->ns_sem);
76 mutex_init(&nilfs->ns_snapshot_mount_mutex);
77 INIT_LIST_HEAD(&nilfs->ns_dirty_files);
78 INIT_LIST_HEAD(&nilfs->ns_gc_inodes);
79 spin_lock_init(&nilfs->ns_inode_lock);
80 spin_lock_init(&nilfs->ns_next_gen_lock);
81 spin_lock_init(&nilfs->ns_last_segment_lock);
82 nilfs->ns_cptree = RB_ROOT;
83 spin_lock_init(&nilfs->ns_cptree_lock);
84 init_rwsem(&nilfs->ns_segctor_sem);
85 nilfs->ns_sb_update_freq = NILFS_SB_FREQ;
86
87 return nilfs;
88}
89
90/**
91 * destroy_nilfs - destroy nilfs object
92 * @nilfs: nilfs object to be released
93 */
94void destroy_nilfs(struct the_nilfs *nilfs)
95{
96 might_sleep();
97 if (nilfs_init(nilfs)) {
98 nilfs_sysfs_delete_device_group(nilfs);
99 brelse(nilfs->ns_sbh[0]);
100 brelse(nilfs->ns_sbh[1]);
101 }
102 kfree(nilfs);
103}
104
105static int nilfs_load_super_root(struct the_nilfs *nilfs,
106 struct super_block *sb, sector_t sr_block)
107{
108 struct buffer_head *bh_sr;
109 struct nilfs_super_root *raw_sr;
110 struct nilfs_super_block **sbp = nilfs->ns_sbp;
111 struct nilfs_inode *rawi;
112 unsigned int dat_entry_size, segment_usage_size, checkpoint_size;
113 unsigned int inode_size;
114 int err;
115
116 err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
117 if (unlikely(err))
118 return err;
119
120 down_read(&nilfs->ns_sem);
121 dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
122 checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
123 segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
124 up_read(&nilfs->ns_sem);
125
126 inode_size = nilfs->ns_inode_size;
127
128 rawi = (void *)bh_sr->b_data + NILFS_SR_DAT_OFFSET(inode_size);
129 err = nilfs_dat_read(sb, dat_entry_size, rawi, &nilfs->ns_dat);
130 if (err)
131 goto failed;
132
133 rawi = (void *)bh_sr->b_data + NILFS_SR_CPFILE_OFFSET(inode_size);
134 err = nilfs_cpfile_read(sb, checkpoint_size, rawi, &nilfs->ns_cpfile);
135 if (err)
136 goto failed_dat;
137
138 rawi = (void *)bh_sr->b_data + NILFS_SR_SUFILE_OFFSET(inode_size);
139 err = nilfs_sufile_read(sb, segment_usage_size, rawi,
140 &nilfs->ns_sufile);
141 if (err)
142 goto failed_cpfile;
143
144 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
145 nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
146
147 failed:
148 brelse(bh_sr);
149 return err;
150
151 failed_cpfile:
152 iput(nilfs->ns_cpfile);
153
154 failed_dat:
155 iput(nilfs->ns_dat);
156 goto failed;
157}
158
159static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
160{
161 memset(ri, 0, sizeof(*ri));
162 INIT_LIST_HEAD(&ri->ri_used_segments);
163}
164
165static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
166{
167 nilfs_dispose_segment_list(&ri->ri_used_segments);
168}
169
170/**
171 * nilfs_store_log_cursor - load log cursor from a super block
172 * @nilfs: nilfs object
173 * @sbp: buffer storing super block to be read
174 *
175 * nilfs_store_log_cursor() reads the last position of the log
176 * containing a super root from a given super block, and initializes
177 * relevant information on the nilfs object preparatory for log
178 * scanning and recovery.
179 */
180static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
181 struct nilfs_super_block *sbp)
182{
183 int ret = 0;
184
185 nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
186 nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
187 nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
188
189 nilfs->ns_prev_seq = nilfs->ns_last_seq;
190 nilfs->ns_seg_seq = nilfs->ns_last_seq;
191 nilfs->ns_segnum =
192 nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
193 nilfs->ns_cno = nilfs->ns_last_cno + 1;
194 if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
195 nilfs_msg(nilfs->ns_sb, KERN_ERR,
196 "pointed segment number is out of range: segnum=%llu, nsegments=%lu",
197 (unsigned long long)nilfs->ns_segnum,
198 nilfs->ns_nsegments);
199 ret = -EINVAL;
200 }
201 return ret;
202}
203
204/**
205 * load_nilfs - load and recover the nilfs
206 * @nilfs: the_nilfs structure to be released
207 * @sb: super block isntance used to recover past segment
208 *
209 * load_nilfs() searches and load the latest super root,
210 * attaches the last segment, and does recovery if needed.
211 * The caller must call this exclusively for simultaneous mounts.
212 */
213int load_nilfs(struct the_nilfs *nilfs, struct super_block *sb)
214{
215 struct nilfs_recovery_info ri;
216 unsigned int s_flags = sb->s_flags;
217 int really_read_only = bdev_read_only(nilfs->ns_bdev);
218 int valid_fs = nilfs_valid_fs(nilfs);
219 int err;
220
221 if (!valid_fs) {
222 nilfs_msg(sb, KERN_WARNING, "mounting unchecked fs");
223 if (s_flags & SB_RDONLY) {
224 nilfs_msg(sb, KERN_INFO,
225 "recovery required for readonly filesystem");
226 nilfs_msg(sb, KERN_INFO,
227 "write access will be enabled during recovery");
228 }
229 }
230
231 nilfs_init_recovery_info(&ri);
232
233 err = nilfs_search_super_root(nilfs, &ri);
234 if (unlikely(err)) {
235 struct nilfs_super_block **sbp = nilfs->ns_sbp;
236 int blocksize;
237
238 if (err != -EINVAL)
239 goto scan_error;
240
241 if (!nilfs_valid_sb(sbp[1])) {
242 nilfs_msg(sb, KERN_WARNING,
243 "unable to fall back to spare super block");
244 goto scan_error;
245 }
246 nilfs_msg(sb, KERN_INFO,
247 "trying rollback from an earlier position");
248
249 /*
250 * restore super block with its spare and reconfigure
251 * relevant states of the nilfs object.
252 */
253 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
254 nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
255 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
256
257 /* verify consistency between two super blocks */
258 blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
259 if (blocksize != nilfs->ns_blocksize) {
260 nilfs_msg(sb, KERN_WARNING,
261 "blocksize differs between two super blocks (%d != %d)",
262 blocksize, nilfs->ns_blocksize);
263 goto scan_error;
264 }
265
266 err = nilfs_store_log_cursor(nilfs, sbp[0]);
267 if (err)
268 goto scan_error;
269
270 /* drop clean flag to allow roll-forward and recovery */
271 nilfs->ns_mount_state &= ~NILFS_VALID_FS;
272 valid_fs = 0;
273
274 err = nilfs_search_super_root(nilfs, &ri);
275 if (err)
276 goto scan_error;
277 }
278
279 err = nilfs_load_super_root(nilfs, sb, ri.ri_super_root);
280 if (unlikely(err)) {
281 nilfs_msg(sb, KERN_ERR, "error %d while loading super root",
282 err);
283 goto failed;
284 }
285
286 if (valid_fs)
287 goto skip_recovery;
288
289 if (s_flags & SB_RDONLY) {
290 __u64 features;
291
292 if (nilfs_test_opt(nilfs, NORECOVERY)) {
293 nilfs_msg(sb, KERN_INFO,
294 "norecovery option specified, skipping roll-forward recovery");
295 goto skip_recovery;
296 }
297 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
298 ~NILFS_FEATURE_COMPAT_RO_SUPP;
299 if (features) {
300 nilfs_msg(sb, KERN_ERR,
301 "couldn't proceed with recovery because of unsupported optional features (%llx)",
302 (unsigned long long)features);
303 err = -EROFS;
304 goto failed_unload;
305 }
306 if (really_read_only) {
307 nilfs_msg(sb, KERN_ERR,
308 "write access unavailable, cannot proceed");
309 err = -EROFS;
310 goto failed_unload;
311 }
312 sb->s_flags &= ~SB_RDONLY;
313 } else if (nilfs_test_opt(nilfs, NORECOVERY)) {
314 nilfs_msg(sb, KERN_ERR,
315 "recovery cancelled because norecovery option was specified for a read/write mount");
316 err = -EINVAL;
317 goto failed_unload;
318 }
319
320 err = nilfs_salvage_orphan_logs(nilfs, sb, &ri);
321 if (err)
322 goto failed_unload;
323
324 down_write(&nilfs->ns_sem);
325 nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
326 err = nilfs_cleanup_super(sb);
327 up_write(&nilfs->ns_sem);
328
329 if (err) {
330 nilfs_msg(sb, KERN_ERR,
331 "error %d updating super block. recovery unfinished.",
332 err);
333 goto failed_unload;
334 }
335 nilfs_msg(sb, KERN_INFO, "recovery complete");
336
337 skip_recovery:
338 nilfs_clear_recovery_info(&ri);
339 sb->s_flags = s_flags;
340 return 0;
341
342 scan_error:
343 nilfs_msg(sb, KERN_ERR, "error %d while searching super root", err);
344 goto failed;
345
346 failed_unload:
347 iput(nilfs->ns_cpfile);
348 iput(nilfs->ns_sufile);
349 iput(nilfs->ns_dat);
350
351 failed:
352 nilfs_clear_recovery_info(&ri);
353 sb->s_flags = s_flags;
354 return err;
355}
356
357static unsigned long long nilfs_max_size(unsigned int blkbits)
358{
359 unsigned int max_bits;
360 unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
361
362 max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
363 if (max_bits < 64)
364 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
365 return res;
366}
367
368/**
369 * nilfs_nrsvsegs - calculate the number of reserved segments
370 * @nilfs: nilfs object
371 * @nsegs: total number of segments
372 */
373unsigned long nilfs_nrsvsegs(struct the_nilfs *nilfs, unsigned long nsegs)
374{
375 return max_t(unsigned long, NILFS_MIN_NRSVSEGS,
376 DIV_ROUND_UP(nsegs * nilfs->ns_r_segments_percentage,
377 100));
378}
379
380void nilfs_set_nsegments(struct the_nilfs *nilfs, unsigned long nsegs)
381{
382 nilfs->ns_nsegments = nsegs;
383 nilfs->ns_nrsvsegs = nilfs_nrsvsegs(nilfs, nsegs);
384}
385
386static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
387 struct nilfs_super_block *sbp)
388{
389 if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
390 nilfs_msg(nilfs->ns_sb, KERN_ERR,
391 "unsupported revision (superblock rev.=%d.%d, current rev.=%d.%d). Please check the version of mkfs.nilfs(2).",
392 le32_to_cpu(sbp->s_rev_level),
393 le16_to_cpu(sbp->s_minor_rev_level),
394 NILFS_CURRENT_REV, NILFS_MINOR_REV);
395 return -EINVAL;
396 }
397 nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
398 if (nilfs->ns_sbsize > BLOCK_SIZE)
399 return -EINVAL;
400
401 nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
402 if (nilfs->ns_inode_size > nilfs->ns_blocksize) {
403 nilfs_msg(nilfs->ns_sb, KERN_ERR,
404 "too large inode size: %d bytes",
405 nilfs->ns_inode_size);
406 return -EINVAL;
407 } else if (nilfs->ns_inode_size < NILFS_MIN_INODE_SIZE) {
408 nilfs_msg(nilfs->ns_sb, KERN_ERR,
409 "too small inode size: %d bytes",
410 nilfs->ns_inode_size);
411 return -EINVAL;
412 }
413
414 nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
415
416 nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
417 if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
418 nilfs_msg(nilfs->ns_sb, KERN_ERR,
419 "too short segment: %lu blocks",
420 nilfs->ns_blocks_per_segment);
421 return -EINVAL;
422 }
423
424 nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
425 nilfs->ns_r_segments_percentage =
426 le32_to_cpu(sbp->s_r_segments_percentage);
427 if (nilfs->ns_r_segments_percentage < 1 ||
428 nilfs->ns_r_segments_percentage > 99) {
429 nilfs_msg(nilfs->ns_sb, KERN_ERR,
430 "invalid reserved segments percentage: %lu",
431 nilfs->ns_r_segments_percentage);
432 return -EINVAL;
433 }
434
435 nilfs_set_nsegments(nilfs, le64_to_cpu(sbp->s_nsegments));
436 nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
437 return 0;
438}
439
440static int nilfs_valid_sb(struct nilfs_super_block *sbp)
441{
442 static unsigned char sum[4];
443 const int sumoff = offsetof(struct nilfs_super_block, s_sum);
444 size_t bytes;
445 u32 crc;
446
447 if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
448 return 0;
449 bytes = le16_to_cpu(sbp->s_bytes);
450 if (bytes < sumoff + 4 || bytes > BLOCK_SIZE)
451 return 0;
452 crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
453 sumoff);
454 crc = crc32_le(crc, sum, 4);
455 crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
456 bytes - sumoff - 4);
457 return crc == le32_to_cpu(sbp->s_sum);
458}
459
460static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
461{
462 return offset < ((le64_to_cpu(sbp->s_nsegments) *
463 le32_to_cpu(sbp->s_blocks_per_segment)) <<
464 (le32_to_cpu(sbp->s_log_block_size) + 10));
465}
466
467static void nilfs_release_super_block(struct the_nilfs *nilfs)
468{
469 int i;
470
471 for (i = 0; i < 2; i++) {
472 if (nilfs->ns_sbp[i]) {
473 brelse(nilfs->ns_sbh[i]);
474 nilfs->ns_sbh[i] = NULL;
475 nilfs->ns_sbp[i] = NULL;
476 }
477 }
478}
479
480void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
481{
482 brelse(nilfs->ns_sbh[0]);
483 nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
484 nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
485 nilfs->ns_sbh[1] = NULL;
486 nilfs->ns_sbp[1] = NULL;
487}
488
489void nilfs_swap_super_block(struct the_nilfs *nilfs)
490{
491 struct buffer_head *tsbh = nilfs->ns_sbh[0];
492 struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
493
494 nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
495 nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
496 nilfs->ns_sbh[1] = tsbh;
497 nilfs->ns_sbp[1] = tsbp;
498}
499
500static int nilfs_load_super_block(struct the_nilfs *nilfs,
501 struct super_block *sb, int blocksize,
502 struct nilfs_super_block **sbpp)
503{
504 struct nilfs_super_block **sbp = nilfs->ns_sbp;
505 struct buffer_head **sbh = nilfs->ns_sbh;
506 u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
507 int valid[2], swp = 0;
508
509 sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
510 &sbh[0]);
511 sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
512
513 if (!sbp[0]) {
514 if (!sbp[1]) {
515 nilfs_msg(sb, KERN_ERR, "unable to read superblock");
516 return -EIO;
517 }
518 nilfs_msg(sb, KERN_WARNING,
519 "unable to read primary superblock (blocksize = %d)",
520 blocksize);
521 } else if (!sbp[1]) {
522 nilfs_msg(sb, KERN_WARNING,
523 "unable to read secondary superblock (blocksize = %d)",
524 blocksize);
525 }
526
527 /*
528 * Compare two super blocks and set 1 in swp if the secondary
529 * super block is valid and newer. Otherwise, set 0 in swp.
530 */
531 valid[0] = nilfs_valid_sb(sbp[0]);
532 valid[1] = nilfs_valid_sb(sbp[1]);
533 swp = valid[1] && (!valid[0] ||
534 le64_to_cpu(sbp[1]->s_last_cno) >
535 le64_to_cpu(sbp[0]->s_last_cno));
536
537 if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
538 brelse(sbh[1]);
539 sbh[1] = NULL;
540 sbp[1] = NULL;
541 valid[1] = 0;
542 swp = 0;
543 }
544 if (!valid[swp]) {
545 nilfs_release_super_block(nilfs);
546 nilfs_msg(sb, KERN_ERR, "couldn't find nilfs on the device");
547 return -EINVAL;
548 }
549
550 if (!valid[!swp])
551 nilfs_msg(sb, KERN_WARNING,
552 "broken superblock, retrying with spare superblock (blocksize = %d)",
553 blocksize);
554 if (swp)
555 nilfs_swap_super_block(nilfs);
556
557 nilfs->ns_sbwcount = 0;
558 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
559 nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
560 *sbpp = sbp[0];
561 return 0;
562}
563
564/**
565 * init_nilfs - initialize a NILFS instance.
566 * @nilfs: the_nilfs structure
567 * @sb: super block
568 * @data: mount options
569 *
570 * init_nilfs() performs common initialization per block device (e.g.
571 * reading the super block, getting disk layout information, initializing
572 * shared fields in the_nilfs).
573 *
574 * Return Value: On success, 0 is returned. On error, a negative error
575 * code is returned.
576 */
577int init_nilfs(struct the_nilfs *nilfs, struct super_block *sb, char *data)
578{
579 struct nilfs_super_block *sbp;
580 int blocksize;
581 int err;
582
583 down_write(&nilfs->ns_sem);
584
585 blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
586 if (!blocksize) {
587 nilfs_msg(sb, KERN_ERR, "unable to set blocksize");
588 err = -EINVAL;
589 goto out;
590 }
591 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
592 if (err)
593 goto out;
594
595 err = nilfs_store_magic_and_option(sb, sbp, data);
596 if (err)
597 goto failed_sbh;
598
599 err = nilfs_check_feature_compatibility(sb, sbp);
600 if (err)
601 goto failed_sbh;
602
603 blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
604 if (blocksize < NILFS_MIN_BLOCK_SIZE ||
605 blocksize > NILFS_MAX_BLOCK_SIZE) {
606 nilfs_msg(sb, KERN_ERR,
607 "couldn't mount because of unsupported filesystem blocksize %d",
608 blocksize);
609 err = -EINVAL;
610 goto failed_sbh;
611 }
612 if (sb->s_blocksize != blocksize) {
613 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
614
615 if (blocksize < hw_blocksize) {
616 nilfs_msg(sb, KERN_ERR,
617 "blocksize %d too small for device (sector-size = %d)",
618 blocksize, hw_blocksize);
619 err = -EINVAL;
620 goto failed_sbh;
621 }
622 nilfs_release_super_block(nilfs);
623 sb_set_blocksize(sb, blocksize);
624
625 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
626 if (err)
627 goto out;
628 /*
629 * Not to failed_sbh; sbh is released automatically
630 * when reloading fails.
631 */
632 }
633 nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
634 nilfs->ns_blocksize = blocksize;
635
636 get_random_bytes(&nilfs->ns_next_generation,
637 sizeof(nilfs->ns_next_generation));
638
639 err = nilfs_store_disk_layout(nilfs, sbp);
640 if (err)
641 goto failed_sbh;
642
643 sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
644
645 nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
646
647 err = nilfs_store_log_cursor(nilfs, sbp);
648 if (err)
649 goto failed_sbh;
650
651 err = nilfs_sysfs_create_device_group(sb);
652 if (err)
653 goto failed_sbh;
654
655 set_nilfs_init(nilfs);
656 err = 0;
657 out:
658 up_write(&nilfs->ns_sem);
659 return err;
660
661 failed_sbh:
662 nilfs_release_super_block(nilfs);
663 goto out;
664}
665
666int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
667 size_t nsegs)
668{
669 sector_t seg_start, seg_end;
670 sector_t start = 0, nblocks = 0;
671 unsigned int sects_per_block;
672 __u64 *sn;
673 int ret = 0;
674
675 sects_per_block = (1 << nilfs->ns_blocksize_bits) /
676 bdev_logical_block_size(nilfs->ns_bdev);
677 for (sn = segnump; sn < segnump + nsegs; sn++) {
678 nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
679
680 if (!nblocks) {
681 start = seg_start;
682 nblocks = seg_end - seg_start + 1;
683 } else if (start + nblocks == seg_start) {
684 nblocks += seg_end - seg_start + 1;
685 } else {
686 ret = blkdev_issue_discard(nilfs->ns_bdev,
687 start * sects_per_block,
688 nblocks * sects_per_block,
689 GFP_NOFS, 0);
690 if (ret < 0)
691 return ret;
692 nblocks = 0;
693 }
694 }
695 if (nblocks)
696 ret = blkdev_issue_discard(nilfs->ns_bdev,
697 start * sects_per_block,
698 nblocks * sects_per_block,
699 GFP_NOFS, 0);
700 return ret;
701}
702
703int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
704{
705 unsigned long ncleansegs;
706
707 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
708 ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
709 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
710 *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
711 return 0;
712}
713
714int nilfs_near_disk_full(struct the_nilfs *nilfs)
715{
716 unsigned long ncleansegs, nincsegs;
717
718 ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
719 nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
720 nilfs->ns_blocks_per_segment + 1;
721
722 return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
723}
724
725struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
726{
727 struct rb_node *n;
728 struct nilfs_root *root;
729
730 spin_lock(&nilfs->ns_cptree_lock);
731 n = nilfs->ns_cptree.rb_node;
732 while (n) {
733 root = rb_entry(n, struct nilfs_root, rb_node);
734
735 if (cno < root->cno) {
736 n = n->rb_left;
737 } else if (cno > root->cno) {
738 n = n->rb_right;
739 } else {
740 refcount_inc(&root->count);
741 spin_unlock(&nilfs->ns_cptree_lock);
742 return root;
743 }
744 }
745 spin_unlock(&nilfs->ns_cptree_lock);
746
747 return NULL;
748}
749
750struct nilfs_root *
751nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
752{
753 struct rb_node **p, *parent;
754 struct nilfs_root *root, *new;
755 int err;
756
757 root = nilfs_lookup_root(nilfs, cno);
758 if (root)
759 return root;
760
761 new = kzalloc(sizeof(*root), GFP_KERNEL);
762 if (!new)
763 return NULL;
764
765 spin_lock(&nilfs->ns_cptree_lock);
766
767 p = &nilfs->ns_cptree.rb_node;
768 parent = NULL;
769
770 while (*p) {
771 parent = *p;
772 root = rb_entry(parent, struct nilfs_root, rb_node);
773
774 if (cno < root->cno) {
775 p = &(*p)->rb_left;
776 } else if (cno > root->cno) {
777 p = &(*p)->rb_right;
778 } else {
779 refcount_inc(&root->count);
780 spin_unlock(&nilfs->ns_cptree_lock);
781 kfree(new);
782 return root;
783 }
784 }
785
786 new->cno = cno;
787 new->ifile = NULL;
788 new->nilfs = nilfs;
789 refcount_set(&new->count, 1);
790 atomic64_set(&new->inodes_count, 0);
791 atomic64_set(&new->blocks_count, 0);
792
793 rb_link_node(&new->rb_node, parent, p);
794 rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
795
796 spin_unlock(&nilfs->ns_cptree_lock);
797
798 err = nilfs_sysfs_create_snapshot_group(new);
799 if (err) {
800 kfree(new);
801 new = NULL;
802 }
803
804 return new;
805}
806
807void nilfs_put_root(struct nilfs_root *root)
808{
809 if (refcount_dec_and_test(&root->count)) {
810 struct the_nilfs *nilfs = root->nilfs;
811
812 nilfs_sysfs_delete_snapshot_group(root);
813
814 spin_lock(&nilfs->ns_cptree_lock);
815 rb_erase(&root->rb_node, &nilfs->ns_cptree);
816 spin_unlock(&nilfs->ns_cptree_lock);
817 iput(root->ifile);
818
819 kfree(root);
820 }
821}
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
27static int nilfs_valid_sb(struct nilfs_super_block *sbp);
28
29void 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 */
56struct 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 */
86void 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
96static 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
150static 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
156static 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 */
171static 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 */
208static 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 */
232int 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
385static 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 */
401unsigned 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 */
412static 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
420void 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
426static 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
456 nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
457 if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
458 nilfs_err(nilfs->ns_sb, "too short segment: %lu blocks",
459 nilfs->ns_blocks_per_segment);
460 return -EINVAL;
461 }
462
463 nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
464 nilfs->ns_r_segments_percentage =
465 le32_to_cpu(sbp->s_r_segments_percentage);
466 if (nilfs->ns_r_segments_percentage < 1 ||
467 nilfs->ns_r_segments_percentage > 99) {
468 nilfs_err(nilfs->ns_sb,
469 "invalid reserved segments percentage: %lu",
470 nilfs->ns_r_segments_percentage);
471 return -EINVAL;
472 }
473
474 nsegments = le64_to_cpu(sbp->s_nsegments);
475 if (nsegments > nilfs_max_segment_count(nilfs)) {
476 nilfs_err(nilfs->ns_sb,
477 "segment count %llu exceeds upper limit (%llu segments)",
478 (unsigned long long)nsegments,
479 (unsigned long long)nilfs_max_segment_count(nilfs));
480 return -EINVAL;
481 }
482
483 nblocks = sb_bdev_nr_blocks(nilfs->ns_sb);
484 if (nblocks) {
485 u64 min_block_count = nsegments * nilfs->ns_blocks_per_segment;
486 /*
487 * To avoid failing to mount early device images without a
488 * second superblock, exclude that block count from the
489 * "min_block_count" calculation.
490 */
491
492 if (nblocks < min_block_count) {
493 nilfs_err(nilfs->ns_sb,
494 "total number of segment blocks %llu exceeds device size (%llu blocks)",
495 (unsigned long long)min_block_count,
496 (unsigned long long)nblocks);
497 return -EINVAL;
498 }
499 }
500
501 nilfs_set_nsegments(nilfs, nsegments);
502 nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
503 return 0;
504}
505
506static int nilfs_valid_sb(struct nilfs_super_block *sbp)
507{
508 static unsigned char sum[4];
509 const int sumoff = offsetof(struct nilfs_super_block, s_sum);
510 size_t bytes;
511 u32 crc;
512
513 if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
514 return 0;
515 bytes = le16_to_cpu(sbp->s_bytes);
516 if (bytes < sumoff + 4 || bytes > BLOCK_SIZE)
517 return 0;
518 crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
519 sumoff);
520 crc = crc32_le(crc, sum, 4);
521 crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
522 bytes - sumoff - 4);
523 return crc == le32_to_cpu(sbp->s_sum);
524}
525
526/**
527 * nilfs_sb2_bad_offset - check the location of the second superblock
528 * @sbp: superblock raw data buffer
529 * @offset: byte offset of second superblock calculated from device size
530 *
531 * nilfs_sb2_bad_offset() checks if the position on the second
532 * superblock is valid or not based on the filesystem parameters
533 * stored in @sbp. If @offset points to a location within the segment
534 * area, or if the parameters themselves are not normal, it is
535 * determined to be invalid.
536 *
537 * Return Value: true if invalid, false if valid.
538 */
539static bool nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
540{
541 unsigned int shift_bits = le32_to_cpu(sbp->s_log_block_size);
542 u32 blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
543 u64 nsegments = le64_to_cpu(sbp->s_nsegments);
544 u64 index;
545
546 if (blocks_per_segment < NILFS_SEG_MIN_BLOCKS ||
547 shift_bits > ilog2(NILFS_MAX_BLOCK_SIZE) - BLOCK_SIZE_BITS)
548 return true;
549
550 index = offset >> (shift_bits + BLOCK_SIZE_BITS);
551 do_div(index, blocks_per_segment);
552 return index < nsegments;
553}
554
555static void nilfs_release_super_block(struct the_nilfs *nilfs)
556{
557 int i;
558
559 for (i = 0; i < 2; i++) {
560 if (nilfs->ns_sbp[i]) {
561 brelse(nilfs->ns_sbh[i]);
562 nilfs->ns_sbh[i] = NULL;
563 nilfs->ns_sbp[i] = NULL;
564 }
565 }
566}
567
568void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
569{
570 brelse(nilfs->ns_sbh[0]);
571 nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
572 nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
573 nilfs->ns_sbh[1] = NULL;
574 nilfs->ns_sbp[1] = NULL;
575}
576
577void nilfs_swap_super_block(struct the_nilfs *nilfs)
578{
579 struct buffer_head *tsbh = nilfs->ns_sbh[0];
580 struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
581
582 nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
583 nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
584 nilfs->ns_sbh[1] = tsbh;
585 nilfs->ns_sbp[1] = tsbp;
586}
587
588static int nilfs_load_super_block(struct the_nilfs *nilfs,
589 struct super_block *sb, int blocksize,
590 struct nilfs_super_block **sbpp)
591{
592 struct nilfs_super_block **sbp = nilfs->ns_sbp;
593 struct buffer_head **sbh = nilfs->ns_sbh;
594 u64 sb2off, devsize = bdev_nr_bytes(nilfs->ns_bdev);
595 int valid[2], swp = 0;
596
597 if (devsize < NILFS_SEG_MIN_BLOCKS * NILFS_MIN_BLOCK_SIZE + 4096) {
598 nilfs_err(sb, "device size too small");
599 return -EINVAL;
600 }
601 sb2off = NILFS_SB2_OFFSET_BYTES(devsize);
602
603 sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
604 &sbh[0]);
605 sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
606
607 if (!sbp[0]) {
608 if (!sbp[1]) {
609 nilfs_err(sb, "unable to read superblock");
610 return -EIO;
611 }
612 nilfs_warn(sb,
613 "unable to read primary superblock (blocksize = %d)",
614 blocksize);
615 } else if (!sbp[1]) {
616 nilfs_warn(sb,
617 "unable to read secondary superblock (blocksize = %d)",
618 blocksize);
619 }
620
621 /*
622 * Compare two super blocks and set 1 in swp if the secondary
623 * super block is valid and newer. Otherwise, set 0 in swp.
624 */
625 valid[0] = nilfs_valid_sb(sbp[0]);
626 valid[1] = nilfs_valid_sb(sbp[1]);
627 swp = valid[1] && (!valid[0] ||
628 le64_to_cpu(sbp[1]->s_last_cno) >
629 le64_to_cpu(sbp[0]->s_last_cno));
630
631 if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
632 brelse(sbh[1]);
633 sbh[1] = NULL;
634 sbp[1] = NULL;
635 valid[1] = 0;
636 swp = 0;
637 }
638 if (!valid[swp]) {
639 nilfs_release_super_block(nilfs);
640 nilfs_err(sb, "couldn't find nilfs on the device");
641 return -EINVAL;
642 }
643
644 if (!valid[!swp])
645 nilfs_warn(sb,
646 "broken superblock, retrying with spare superblock (blocksize = %d)",
647 blocksize);
648 if (swp)
649 nilfs_swap_super_block(nilfs);
650
651 nilfs->ns_sbwcount = 0;
652 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
653 nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
654 *sbpp = sbp[0];
655 return 0;
656}
657
658/**
659 * init_nilfs - initialize a NILFS instance.
660 * @nilfs: the_nilfs structure
661 * @sb: super block
662 * @data: mount options
663 *
664 * init_nilfs() performs common initialization per block device (e.g.
665 * reading the super block, getting disk layout information, initializing
666 * shared fields in the_nilfs).
667 *
668 * Return Value: On success, 0 is returned. On error, a negative error
669 * code is returned.
670 */
671int init_nilfs(struct the_nilfs *nilfs, struct super_block *sb, char *data)
672{
673 struct nilfs_super_block *sbp;
674 int blocksize;
675 int err;
676
677 down_write(&nilfs->ns_sem);
678
679 blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
680 if (!blocksize) {
681 nilfs_err(sb, "unable to set blocksize");
682 err = -EINVAL;
683 goto out;
684 }
685 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
686 if (err)
687 goto out;
688
689 err = nilfs_store_magic_and_option(sb, sbp, data);
690 if (err)
691 goto failed_sbh;
692
693 err = nilfs_check_feature_compatibility(sb, sbp);
694 if (err)
695 goto failed_sbh;
696
697 err = nilfs_get_blocksize(sb, sbp, &blocksize);
698 if (err)
699 goto failed_sbh;
700
701 if (blocksize < NILFS_MIN_BLOCK_SIZE) {
702 nilfs_err(sb,
703 "couldn't mount because of unsupported filesystem blocksize %d",
704 blocksize);
705 err = -EINVAL;
706 goto failed_sbh;
707 }
708 if (sb->s_blocksize != blocksize) {
709 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
710
711 if (blocksize < hw_blocksize) {
712 nilfs_err(sb,
713 "blocksize %d too small for device (sector-size = %d)",
714 blocksize, hw_blocksize);
715 err = -EINVAL;
716 goto failed_sbh;
717 }
718 nilfs_release_super_block(nilfs);
719 if (!sb_set_blocksize(sb, blocksize)) {
720 nilfs_err(sb, "bad blocksize %d", blocksize);
721 err = -EINVAL;
722 goto out;
723 }
724
725 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
726 if (err)
727 goto out;
728 /*
729 * Not to failed_sbh; sbh is released automatically
730 * when reloading fails.
731 */
732 }
733 nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
734 nilfs->ns_blocksize = blocksize;
735
736 get_random_bytes(&nilfs->ns_next_generation,
737 sizeof(nilfs->ns_next_generation));
738
739 err = nilfs_store_disk_layout(nilfs, sbp);
740 if (err)
741 goto failed_sbh;
742
743 sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
744
745 nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
746
747 err = nilfs_store_log_cursor(nilfs, sbp);
748 if (err)
749 goto failed_sbh;
750
751 set_nilfs_init(nilfs);
752 err = 0;
753 out:
754 up_write(&nilfs->ns_sem);
755 return err;
756
757 failed_sbh:
758 nilfs_release_super_block(nilfs);
759 goto out;
760}
761
762int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
763 size_t nsegs)
764{
765 sector_t seg_start, seg_end;
766 sector_t start = 0, nblocks = 0;
767 unsigned int sects_per_block;
768 __u64 *sn;
769 int ret = 0;
770
771 sects_per_block = (1 << nilfs->ns_blocksize_bits) /
772 bdev_logical_block_size(nilfs->ns_bdev);
773 for (sn = segnump; sn < segnump + nsegs; sn++) {
774 nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
775
776 if (!nblocks) {
777 start = seg_start;
778 nblocks = seg_end - seg_start + 1;
779 } else if (start + nblocks == seg_start) {
780 nblocks += seg_end - seg_start + 1;
781 } else {
782 ret = blkdev_issue_discard(nilfs->ns_bdev,
783 start * sects_per_block,
784 nblocks * sects_per_block,
785 GFP_NOFS);
786 if (ret < 0)
787 return ret;
788 nblocks = 0;
789 }
790 }
791 if (nblocks)
792 ret = blkdev_issue_discard(nilfs->ns_bdev,
793 start * sects_per_block,
794 nblocks * sects_per_block,
795 GFP_NOFS);
796 return ret;
797}
798
799int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
800{
801 unsigned long ncleansegs;
802
803 ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
804 *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
805 return 0;
806}
807
808int nilfs_near_disk_full(struct the_nilfs *nilfs)
809{
810 unsigned long ncleansegs, nincsegs;
811
812 ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
813 nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
814 nilfs->ns_blocks_per_segment + 1;
815
816 return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
817}
818
819struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
820{
821 struct rb_node *n;
822 struct nilfs_root *root;
823
824 spin_lock(&nilfs->ns_cptree_lock);
825 n = nilfs->ns_cptree.rb_node;
826 while (n) {
827 root = rb_entry(n, struct nilfs_root, rb_node);
828
829 if (cno < root->cno) {
830 n = n->rb_left;
831 } else if (cno > root->cno) {
832 n = n->rb_right;
833 } else {
834 refcount_inc(&root->count);
835 spin_unlock(&nilfs->ns_cptree_lock);
836 return root;
837 }
838 }
839 spin_unlock(&nilfs->ns_cptree_lock);
840
841 return NULL;
842}
843
844struct nilfs_root *
845nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
846{
847 struct rb_node **p, *parent;
848 struct nilfs_root *root, *new;
849 int err;
850
851 root = nilfs_lookup_root(nilfs, cno);
852 if (root)
853 return root;
854
855 new = kzalloc(sizeof(*root), GFP_KERNEL);
856 if (!new)
857 return NULL;
858
859 spin_lock(&nilfs->ns_cptree_lock);
860
861 p = &nilfs->ns_cptree.rb_node;
862 parent = NULL;
863
864 while (*p) {
865 parent = *p;
866 root = rb_entry(parent, struct nilfs_root, rb_node);
867
868 if (cno < root->cno) {
869 p = &(*p)->rb_left;
870 } else if (cno > root->cno) {
871 p = &(*p)->rb_right;
872 } else {
873 refcount_inc(&root->count);
874 spin_unlock(&nilfs->ns_cptree_lock);
875 kfree(new);
876 return root;
877 }
878 }
879
880 new->cno = cno;
881 new->ifile = NULL;
882 new->nilfs = nilfs;
883 refcount_set(&new->count, 1);
884 atomic64_set(&new->inodes_count, 0);
885 atomic64_set(&new->blocks_count, 0);
886
887 rb_link_node(&new->rb_node, parent, p);
888 rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
889
890 spin_unlock(&nilfs->ns_cptree_lock);
891
892 err = nilfs_sysfs_create_snapshot_group(new);
893 if (err) {
894 kfree(new);
895 new = NULL;
896 }
897
898 return new;
899}
900
901void nilfs_put_root(struct nilfs_root *root)
902{
903 struct the_nilfs *nilfs = root->nilfs;
904
905 if (refcount_dec_and_lock(&root->count, &nilfs->ns_cptree_lock)) {
906 rb_erase(&root->rb_node, &nilfs->ns_cptree);
907 spin_unlock(&nilfs->ns_cptree_lock);
908
909 nilfs_sysfs_delete_snapshot_group(root);
910 iput(root->ifile);
911
912 kfree(root);
913 }
914}