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/*
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 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 *
20 * Written by Ryusuke Konishi <ryusuke@osrg.net>
21 *
22 */
23
24#include <linux/buffer_head.h>
25#include <linux/slab.h>
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
28#include <linux/random.h>
29#include <linux/crc32.h>
30#include "nilfs.h"
31#include "segment.h"
32#include "alloc.h"
33#include "cpfile.h"
34#include "sufile.h"
35#include "dat.h"
36#include "segbuf.h"
37
38
39static int nilfs_valid_sb(struct nilfs_super_block *sbp);
40
41void nilfs_set_last_segment(struct the_nilfs *nilfs,
42 sector_t start_blocknr, u64 seq, __u64 cno)
43{
44 spin_lock(&nilfs->ns_last_segment_lock);
45 nilfs->ns_last_pseg = start_blocknr;
46 nilfs->ns_last_seq = seq;
47 nilfs->ns_last_cno = cno;
48
49 if (!nilfs_sb_dirty(nilfs)) {
50 if (nilfs->ns_prev_seq == nilfs->ns_last_seq)
51 goto stay_cursor;
52
53 set_nilfs_sb_dirty(nilfs);
54 }
55 nilfs->ns_prev_seq = nilfs->ns_last_seq;
56
57 stay_cursor:
58 spin_unlock(&nilfs->ns_last_segment_lock);
59}
60
61/**
62 * alloc_nilfs - allocate a nilfs object
63 * @bdev: block device to which the_nilfs is related
64 *
65 * Return Value: On success, pointer to the_nilfs is returned.
66 * On error, NULL is returned.
67 */
68struct the_nilfs *alloc_nilfs(struct block_device *bdev)
69{
70 struct the_nilfs *nilfs;
71
72 nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
73 if (!nilfs)
74 return NULL;
75
76 nilfs->ns_bdev = bdev;
77 atomic_set(&nilfs->ns_ndirtyblks, 0);
78 init_rwsem(&nilfs->ns_sem);
79 mutex_init(&nilfs->ns_snapshot_mount_mutex);
80 INIT_LIST_HEAD(&nilfs->ns_dirty_files);
81 INIT_LIST_HEAD(&nilfs->ns_gc_inodes);
82 spin_lock_init(&nilfs->ns_inode_lock);
83 spin_lock_init(&nilfs->ns_next_gen_lock);
84 spin_lock_init(&nilfs->ns_last_segment_lock);
85 nilfs->ns_cptree = RB_ROOT;
86 spin_lock_init(&nilfs->ns_cptree_lock);
87 init_rwsem(&nilfs->ns_segctor_sem);
88
89 return nilfs;
90}
91
92/**
93 * destroy_nilfs - destroy nilfs object
94 * @nilfs: nilfs object to be released
95 */
96void destroy_nilfs(struct the_nilfs *nilfs)
97{
98 might_sleep();
99 if (nilfs_init(nilfs)) {
100 brelse(nilfs->ns_sbh[0]);
101 brelse(nilfs->ns_sbh[1]);
102 }
103 kfree(nilfs);
104}
105
106static int nilfs_load_super_root(struct the_nilfs *nilfs,
107 struct super_block *sb, sector_t sr_block)
108{
109 struct buffer_head *bh_sr;
110 struct nilfs_super_root *raw_sr;
111 struct nilfs_super_block **sbp = nilfs->ns_sbp;
112 struct nilfs_inode *rawi;
113 unsigned dat_entry_size, segment_usage_size, checkpoint_size;
114 unsigned inode_size;
115 int err;
116
117 err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
118 if (unlikely(err))
119 return err;
120
121 down_read(&nilfs->ns_sem);
122 dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
123 checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
124 segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
125 up_read(&nilfs->ns_sem);
126
127 inode_size = nilfs->ns_inode_size;
128
129 rawi = (void *)bh_sr->b_data + NILFS_SR_DAT_OFFSET(inode_size);
130 err = nilfs_dat_read(sb, dat_entry_size, rawi, &nilfs->ns_dat);
131 if (err)
132 goto failed;
133
134 rawi = (void *)bh_sr->b_data + NILFS_SR_CPFILE_OFFSET(inode_size);
135 err = nilfs_cpfile_read(sb, checkpoint_size, rawi, &nilfs->ns_cpfile);
136 if (err)
137 goto failed_dat;
138
139 rawi = (void *)bh_sr->b_data + NILFS_SR_SUFILE_OFFSET(inode_size);
140 err = nilfs_sufile_read(sb, segment_usage_size, rawi,
141 &nilfs->ns_sufile);
142 if (err)
143 goto failed_cpfile;
144
145 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
146 nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
147
148 failed:
149 brelse(bh_sr);
150 return err;
151
152 failed_cpfile:
153 iput(nilfs->ns_cpfile);
154
155 failed_dat:
156 iput(nilfs->ns_dat);
157 goto failed;
158}
159
160static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
161{
162 memset(ri, 0, sizeof(*ri));
163 INIT_LIST_HEAD(&ri->ri_used_segments);
164}
165
166static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
167{
168 nilfs_dispose_segment_list(&ri->ri_used_segments);
169}
170
171/**
172 * nilfs_store_log_cursor - load log cursor from a super block
173 * @nilfs: nilfs object
174 * @sbp: buffer storing super block to be read
175 *
176 * nilfs_store_log_cursor() reads the last position of the log
177 * containing a super root from a given super block, and initializes
178 * relevant information on the nilfs object preparatory for log
179 * scanning and recovery.
180 */
181static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
182 struct nilfs_super_block *sbp)
183{
184 int ret = 0;
185
186 nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
187 nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
188 nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
189
190 nilfs->ns_prev_seq = nilfs->ns_last_seq;
191 nilfs->ns_seg_seq = nilfs->ns_last_seq;
192 nilfs->ns_segnum =
193 nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
194 nilfs->ns_cno = nilfs->ns_last_cno + 1;
195 if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
196 printk(KERN_ERR "NILFS invalid last segment number.\n");
197 ret = -EINVAL;
198 }
199 return ret;
200}
201
202/**
203 * load_nilfs - load and recover the nilfs
204 * @nilfs: the_nilfs structure to be released
205 * @sb: super block isntance used to recover past segment
206 *
207 * load_nilfs() searches and load the latest super root,
208 * attaches the last segment, and does recovery if needed.
209 * The caller must call this exclusively for simultaneous mounts.
210 */
211int load_nilfs(struct the_nilfs *nilfs, struct super_block *sb)
212{
213 struct nilfs_recovery_info ri;
214 unsigned int s_flags = sb->s_flags;
215 int really_read_only = bdev_read_only(nilfs->ns_bdev);
216 int valid_fs = nilfs_valid_fs(nilfs);
217 int err;
218
219 if (!valid_fs) {
220 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
221 if (s_flags & MS_RDONLY) {
222 printk(KERN_INFO "NILFS: INFO: recovery "
223 "required for readonly filesystem.\n");
224 printk(KERN_INFO "NILFS: write access will "
225 "be enabled during recovery.\n");
226 }
227 }
228
229 nilfs_init_recovery_info(&ri);
230
231 err = nilfs_search_super_root(nilfs, &ri);
232 if (unlikely(err)) {
233 struct nilfs_super_block **sbp = nilfs->ns_sbp;
234 int blocksize;
235
236 if (err != -EINVAL)
237 goto scan_error;
238
239 if (!nilfs_valid_sb(sbp[1])) {
240 printk(KERN_WARNING
241 "NILFS warning: unable to fall back to spare"
242 "super block\n");
243 goto scan_error;
244 }
245 printk(KERN_INFO
246 "NILFS: try rollback from an earlier position\n");
247
248 /*
249 * restore super block with its spare and reconfigure
250 * relevant states of the nilfs object.
251 */
252 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
253 nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
254 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
255
256 /* verify consistency between two super blocks */
257 blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
258 if (blocksize != nilfs->ns_blocksize) {
259 printk(KERN_WARNING
260 "NILFS warning: blocksize differs between "
261 "two super blocks (%d != %d)\n",
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 printk(KERN_ERR "NILFS: error loading super root.\n");
282 goto failed;
283 }
284
285 if (valid_fs)
286 goto skip_recovery;
287
288 if (s_flags & MS_RDONLY) {
289 __u64 features;
290
291 if (nilfs_test_opt(nilfs, NORECOVERY)) {
292 printk(KERN_INFO "NILFS: norecovery option specified. "
293 "skipping roll-forward recovery\n");
294 goto skip_recovery;
295 }
296 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
297 ~NILFS_FEATURE_COMPAT_RO_SUPP;
298 if (features) {
299 printk(KERN_ERR "NILFS: couldn't proceed with "
300 "recovery because of unsupported optional "
301 "features (%llx)\n",
302 (unsigned long long)features);
303 err = -EROFS;
304 goto failed_unload;
305 }
306 if (really_read_only) {
307 printk(KERN_ERR "NILFS: write access "
308 "unavailable, cannot proceed.\n");
309 err = -EROFS;
310 goto failed_unload;
311 }
312 sb->s_flags &= ~MS_RDONLY;
313 } else if (nilfs_test_opt(nilfs, NORECOVERY)) {
314 printk(KERN_ERR "NILFS: recovery cancelled because norecovery "
315 "option was specified for a read/write mount\n");
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 printk(KERN_ERR "NILFS: failed to update super block. "
331 "recovery unfinished.\n");
332 goto failed_unload;
333 }
334 printk(KERN_INFO "NILFS: recovery complete.\n");
335
336 skip_recovery:
337 nilfs_clear_recovery_info(&ri);
338 sb->s_flags = s_flags;
339 return 0;
340
341 scan_error:
342 printk(KERN_ERR "NILFS: error searching super root.\n");
343 goto failed;
344
345 failed_unload:
346 iput(nilfs->ns_cpfile);
347 iput(nilfs->ns_sufile);
348 iput(nilfs->ns_dat);
349
350 failed:
351 nilfs_clear_recovery_info(&ri);
352 sb->s_flags = s_flags;
353 return err;
354}
355
356static unsigned long long nilfs_max_size(unsigned int blkbits)
357{
358 unsigned int max_bits;
359 unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
360
361 max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
362 if (max_bits < 64)
363 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
364 return res;
365}
366
367/**
368 * nilfs_nrsvsegs - calculate the number of reserved segments
369 * @nilfs: nilfs object
370 * @nsegs: total number of segments
371 */
372unsigned long nilfs_nrsvsegs(struct the_nilfs *nilfs, unsigned long nsegs)
373{
374 return max_t(unsigned long, NILFS_MIN_NRSVSEGS,
375 DIV_ROUND_UP(nsegs * nilfs->ns_r_segments_percentage,
376 100));
377}
378
379void nilfs_set_nsegments(struct the_nilfs *nilfs, unsigned long nsegs)
380{
381 nilfs->ns_nsegments = nsegs;
382 nilfs->ns_nrsvsegs = nilfs_nrsvsegs(nilfs, nsegs);
383}
384
385static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
386 struct nilfs_super_block *sbp)
387{
388 if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
389 printk(KERN_ERR "NILFS: unsupported revision "
390 "(superblock rev.=%d.%d, current rev.=%d.%d). "
391 "Please check the version of mkfs.nilfs.\n",
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 printk(KERN_ERR "NILFS: too large inode size: %d bytes.\n",
404 nilfs->ns_inode_size);
405 return -EINVAL;
406 } else if (nilfs->ns_inode_size < NILFS_MIN_INODE_SIZE) {
407 printk(KERN_ERR "NILFS: too small inode size: %d bytes.\n",
408 nilfs->ns_inode_size);
409 return -EINVAL;
410 }
411
412 nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
413
414 nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
415 if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
416 printk(KERN_ERR "NILFS: too short segment.\n");
417 return -EINVAL;
418 }
419
420 nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
421 nilfs->ns_r_segments_percentage =
422 le32_to_cpu(sbp->s_r_segments_percentage);
423 if (nilfs->ns_r_segments_percentage < 1 ||
424 nilfs->ns_r_segments_percentage > 99) {
425 printk(KERN_ERR "NILFS: invalid reserved segments percentage.\n");
426 return -EINVAL;
427 }
428
429 nilfs_set_nsegments(nilfs, le64_to_cpu(sbp->s_nsegments));
430 nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
431 return 0;
432}
433
434static int nilfs_valid_sb(struct nilfs_super_block *sbp)
435{
436 static unsigned char sum[4];
437 const int sumoff = offsetof(struct nilfs_super_block, s_sum);
438 size_t bytes;
439 u32 crc;
440
441 if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
442 return 0;
443 bytes = le16_to_cpu(sbp->s_bytes);
444 if (bytes > BLOCK_SIZE)
445 return 0;
446 crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
447 sumoff);
448 crc = crc32_le(crc, sum, 4);
449 crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
450 bytes - sumoff - 4);
451 return crc == le32_to_cpu(sbp->s_sum);
452}
453
454static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
455{
456 return offset < ((le64_to_cpu(sbp->s_nsegments) *
457 le32_to_cpu(sbp->s_blocks_per_segment)) <<
458 (le32_to_cpu(sbp->s_log_block_size) + 10));
459}
460
461static void nilfs_release_super_block(struct the_nilfs *nilfs)
462{
463 int i;
464
465 for (i = 0; i < 2; i++) {
466 if (nilfs->ns_sbp[i]) {
467 brelse(nilfs->ns_sbh[i]);
468 nilfs->ns_sbh[i] = NULL;
469 nilfs->ns_sbp[i] = NULL;
470 }
471 }
472}
473
474void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
475{
476 brelse(nilfs->ns_sbh[0]);
477 nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
478 nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
479 nilfs->ns_sbh[1] = NULL;
480 nilfs->ns_sbp[1] = NULL;
481}
482
483void nilfs_swap_super_block(struct the_nilfs *nilfs)
484{
485 struct buffer_head *tsbh = nilfs->ns_sbh[0];
486 struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
487
488 nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
489 nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
490 nilfs->ns_sbh[1] = tsbh;
491 nilfs->ns_sbp[1] = tsbp;
492}
493
494static int nilfs_load_super_block(struct the_nilfs *nilfs,
495 struct super_block *sb, int blocksize,
496 struct nilfs_super_block **sbpp)
497{
498 struct nilfs_super_block **sbp = nilfs->ns_sbp;
499 struct buffer_head **sbh = nilfs->ns_sbh;
500 u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
501 int valid[2], swp = 0;
502
503 sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
504 &sbh[0]);
505 sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
506
507 if (!sbp[0]) {
508 if (!sbp[1]) {
509 printk(KERN_ERR "NILFS: unable to read superblock\n");
510 return -EIO;
511 }
512 printk(KERN_WARNING
513 "NILFS warning: unable to read primary superblock "
514 "(blocksize = %d)\n", blocksize);
515 } else if (!sbp[1]) {
516 printk(KERN_WARNING
517 "NILFS warning: unable to read secondary superblock "
518 "(blocksize = %d)\n", blocksize);
519 }
520
521 /*
522 * Compare two super blocks and set 1 in swp if the secondary
523 * super block is valid and newer. Otherwise, set 0 in swp.
524 */
525 valid[0] = nilfs_valid_sb(sbp[0]);
526 valid[1] = nilfs_valid_sb(sbp[1]);
527 swp = valid[1] && (!valid[0] ||
528 le64_to_cpu(sbp[1]->s_last_cno) >
529 le64_to_cpu(sbp[0]->s_last_cno));
530
531 if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
532 brelse(sbh[1]);
533 sbh[1] = NULL;
534 sbp[1] = NULL;
535 valid[1] = 0;
536 swp = 0;
537 }
538 if (!valid[swp]) {
539 nilfs_release_super_block(nilfs);
540 printk(KERN_ERR "NILFS: Can't find nilfs on dev %s.\n",
541 sb->s_id);
542 return -EINVAL;
543 }
544
545 if (!valid[!swp])
546 printk(KERN_WARNING "NILFS warning: broken superblock. "
547 "using spare superblock (blocksize = %d).\n", blocksize);
548 if (swp)
549 nilfs_swap_super_block(nilfs);
550
551 nilfs->ns_sbwcount = 0;
552 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
553 nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
554 *sbpp = sbp[0];
555 return 0;
556}
557
558/**
559 * init_nilfs - initialize a NILFS instance.
560 * @nilfs: the_nilfs structure
561 * @sb: super block
562 * @data: mount options
563 *
564 * init_nilfs() performs common initialization per block device (e.g.
565 * reading the super block, getting disk layout information, initializing
566 * shared fields in the_nilfs).
567 *
568 * Return Value: On success, 0 is returned. On error, a negative error
569 * code is returned.
570 */
571int init_nilfs(struct the_nilfs *nilfs, struct super_block *sb, char *data)
572{
573 struct nilfs_super_block *sbp;
574 int blocksize;
575 int err;
576
577 down_write(&nilfs->ns_sem);
578
579 blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
580 if (!blocksize) {
581 printk(KERN_ERR "NILFS: unable to set blocksize\n");
582 err = -EINVAL;
583 goto out;
584 }
585 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
586 if (err)
587 goto out;
588
589 err = nilfs_store_magic_and_option(sb, sbp, data);
590 if (err)
591 goto failed_sbh;
592
593 err = nilfs_check_feature_compatibility(sb, sbp);
594 if (err)
595 goto failed_sbh;
596
597 blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
598 if (blocksize < NILFS_MIN_BLOCK_SIZE ||
599 blocksize > NILFS_MAX_BLOCK_SIZE) {
600 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
601 "filesystem blocksize %d\n", blocksize);
602 err = -EINVAL;
603 goto failed_sbh;
604 }
605 if (sb->s_blocksize != blocksize) {
606 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
607
608 if (blocksize < hw_blocksize) {
609 printk(KERN_ERR
610 "NILFS: blocksize %d too small for device "
611 "(sector-size = %d).\n",
612 blocksize, hw_blocksize);
613 err = -EINVAL;
614 goto failed_sbh;
615 }
616 nilfs_release_super_block(nilfs);
617 sb_set_blocksize(sb, blocksize);
618
619 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
620 if (err)
621 goto out;
622 /* not failed_sbh; sbh is released automatically
623 when reloading fails. */
624 }
625 nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
626 nilfs->ns_blocksize = blocksize;
627
628 get_random_bytes(&nilfs->ns_next_generation,
629 sizeof(nilfs->ns_next_generation));
630
631 err = nilfs_store_disk_layout(nilfs, sbp);
632 if (err)
633 goto failed_sbh;
634
635 sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
636
637 nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
638
639 err = nilfs_store_log_cursor(nilfs, sbp);
640 if (err)
641 goto failed_sbh;
642
643 set_nilfs_init(nilfs);
644 err = 0;
645 out:
646 up_write(&nilfs->ns_sem);
647 return err;
648
649 failed_sbh:
650 nilfs_release_super_block(nilfs);
651 goto out;
652}
653
654int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
655 size_t nsegs)
656{
657 sector_t seg_start, seg_end;
658 sector_t start = 0, nblocks = 0;
659 unsigned int sects_per_block;
660 __u64 *sn;
661 int ret = 0;
662
663 sects_per_block = (1 << nilfs->ns_blocksize_bits) /
664 bdev_logical_block_size(nilfs->ns_bdev);
665 for (sn = segnump; sn < segnump + nsegs; sn++) {
666 nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
667
668 if (!nblocks) {
669 start = seg_start;
670 nblocks = seg_end - seg_start + 1;
671 } else if (start + nblocks == seg_start) {
672 nblocks += seg_end - seg_start + 1;
673 } else {
674 ret = blkdev_issue_discard(nilfs->ns_bdev,
675 start * sects_per_block,
676 nblocks * sects_per_block,
677 GFP_NOFS, 0);
678 if (ret < 0)
679 return ret;
680 nblocks = 0;
681 }
682 }
683 if (nblocks)
684 ret = blkdev_issue_discard(nilfs->ns_bdev,
685 start * sects_per_block,
686 nblocks * sects_per_block,
687 GFP_NOFS, 0);
688 return ret;
689}
690
691int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
692{
693 unsigned long ncleansegs;
694
695 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
696 ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
697 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
698 *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
699 return 0;
700}
701
702int nilfs_near_disk_full(struct the_nilfs *nilfs)
703{
704 unsigned long ncleansegs, nincsegs;
705
706 ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
707 nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
708 nilfs->ns_blocks_per_segment + 1;
709
710 return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
711}
712
713struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
714{
715 struct rb_node *n;
716 struct nilfs_root *root;
717
718 spin_lock(&nilfs->ns_cptree_lock);
719 n = nilfs->ns_cptree.rb_node;
720 while (n) {
721 root = rb_entry(n, struct nilfs_root, rb_node);
722
723 if (cno < root->cno) {
724 n = n->rb_left;
725 } else if (cno > root->cno) {
726 n = n->rb_right;
727 } else {
728 atomic_inc(&root->count);
729 spin_unlock(&nilfs->ns_cptree_lock);
730 return root;
731 }
732 }
733 spin_unlock(&nilfs->ns_cptree_lock);
734
735 return NULL;
736}
737
738struct nilfs_root *
739nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
740{
741 struct rb_node **p, *parent;
742 struct nilfs_root *root, *new;
743
744 root = nilfs_lookup_root(nilfs, cno);
745 if (root)
746 return root;
747
748 new = kmalloc(sizeof(*root), GFP_KERNEL);
749 if (!new)
750 return NULL;
751
752 spin_lock(&nilfs->ns_cptree_lock);
753
754 p = &nilfs->ns_cptree.rb_node;
755 parent = NULL;
756
757 while (*p) {
758 parent = *p;
759 root = rb_entry(parent, struct nilfs_root, rb_node);
760
761 if (cno < root->cno) {
762 p = &(*p)->rb_left;
763 } else if (cno > root->cno) {
764 p = &(*p)->rb_right;
765 } else {
766 atomic_inc(&root->count);
767 spin_unlock(&nilfs->ns_cptree_lock);
768 kfree(new);
769 return root;
770 }
771 }
772
773 new->cno = cno;
774 new->ifile = NULL;
775 new->nilfs = nilfs;
776 atomic_set(&new->count, 1);
777 atomic64_set(&new->inodes_count, 0);
778 atomic64_set(&new->blocks_count, 0);
779
780 rb_link_node(&new->rb_node, parent, p);
781 rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
782
783 spin_unlock(&nilfs->ns_cptree_lock);
784
785 return new;
786}
787
788void nilfs_put_root(struct nilfs_root *root)
789{
790 if (atomic_dec_and_test(&root->count)) {
791 struct the_nilfs *nilfs = root->nilfs;
792
793 spin_lock(&nilfs->ns_cptree_lock);
794 rb_erase(&root->rb_node, &nilfs->ns_cptree);
795 spin_unlock(&nilfs->ns_cptree_lock);
796 if (root->ifile)
797 iput(root->ifile);
798
799 kfree(root);
800 }
801}