GNU Linux-libre 4.9.308-gnu1
[releases.git] / fs / nilfs2 / the_nilfs.c
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
35 static int nilfs_valid_sb(struct nilfs_super_block *sbp);
36
37 void 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  */
64 struct 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  */
94 void 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
105 static 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
159 static 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
165 static 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  */
180 static 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  */
213 int 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 & MS_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 & MS_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 &= ~MS_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
357 static 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  */
373 unsigned 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
380 void 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
386 static 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
440 static 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
460 static 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
467 static 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
480 void 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
489 void 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
500 static 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  */
577 int 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
666 int 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
703 int 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
714 int 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
725 struct 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                         atomic_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
750 struct nilfs_root *
751 nilfs_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                         atomic_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         atomic_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
807 void nilfs_put_root(struct nilfs_root *root)
808 {
809         if (atomic_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 }