Loading...
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}
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}