Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / 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  * 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
39 static int nilfs_valid_sb(struct nilfs_super_block *sbp);
40
41 void 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  */
68 struct 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         nilfs->ns_sb_update_freq = NILFS_SB_FREQ;
89
90         return nilfs;
91 }
92
93 /**
94  * destroy_nilfs - destroy nilfs object
95  * @nilfs: nilfs object to be released
96  */
97 void destroy_nilfs(struct the_nilfs *nilfs)
98 {
99         might_sleep();
100         if (nilfs_init(nilfs)) {
101                 nilfs_sysfs_delete_device_group(nilfs);
102                 brelse(nilfs->ns_sbh[0]);
103                 brelse(nilfs->ns_sbh[1]);
104         }
105         kfree(nilfs);
106 }
107
108 static int nilfs_load_super_root(struct the_nilfs *nilfs,
109                                  struct super_block *sb, sector_t sr_block)
110 {
111         struct buffer_head *bh_sr;
112         struct nilfs_super_root *raw_sr;
113         struct nilfs_super_block **sbp = nilfs->ns_sbp;
114         struct nilfs_inode *rawi;
115         unsigned dat_entry_size, segment_usage_size, checkpoint_size;
116         unsigned inode_size;
117         int err;
118
119         err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
120         if (unlikely(err))
121                 return err;
122
123         down_read(&nilfs->ns_sem);
124         dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
125         checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
126         segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
127         up_read(&nilfs->ns_sem);
128
129         inode_size = nilfs->ns_inode_size;
130
131         rawi = (void *)bh_sr->b_data + NILFS_SR_DAT_OFFSET(inode_size);
132         err = nilfs_dat_read(sb, dat_entry_size, rawi, &nilfs->ns_dat);
133         if (err)
134                 goto failed;
135
136         rawi = (void *)bh_sr->b_data + NILFS_SR_CPFILE_OFFSET(inode_size);
137         err = nilfs_cpfile_read(sb, checkpoint_size, rawi, &nilfs->ns_cpfile);
138         if (err)
139                 goto failed_dat;
140
141         rawi = (void *)bh_sr->b_data + NILFS_SR_SUFILE_OFFSET(inode_size);
142         err = nilfs_sufile_read(sb, segment_usage_size, rawi,
143                                 &nilfs->ns_sufile);
144         if (err)
145                 goto failed_cpfile;
146
147         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
148         nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
149
150  failed:
151         brelse(bh_sr);
152         return err;
153
154  failed_cpfile:
155         iput(nilfs->ns_cpfile);
156
157  failed_dat:
158         iput(nilfs->ns_dat);
159         goto failed;
160 }
161
162 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
163 {
164         memset(ri, 0, sizeof(*ri));
165         INIT_LIST_HEAD(&ri->ri_used_segments);
166 }
167
168 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
169 {
170         nilfs_dispose_segment_list(&ri->ri_used_segments);
171 }
172
173 /**
174  * nilfs_store_log_cursor - load log cursor from a super block
175  * @nilfs: nilfs object
176  * @sbp: buffer storing super block to be read
177  *
178  * nilfs_store_log_cursor() reads the last position of the log
179  * containing a super root from a given super block, and initializes
180  * relevant information on the nilfs object preparatory for log
181  * scanning and recovery.
182  */
183 static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
184                                   struct nilfs_super_block *sbp)
185 {
186         int ret = 0;
187
188         nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
189         nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
190         nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
191
192         nilfs->ns_prev_seq = nilfs->ns_last_seq;
193         nilfs->ns_seg_seq = nilfs->ns_last_seq;
194         nilfs->ns_segnum =
195                 nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
196         nilfs->ns_cno = nilfs->ns_last_cno + 1;
197         if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
198                 printk(KERN_ERR "NILFS invalid last segment number.\n");
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                 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
223                 if (s_flags & MS_RDONLY) {
224                         printk(KERN_INFO "NILFS: INFO: recovery "
225                                "required for readonly filesystem.\n");
226                         printk(KERN_INFO "NILFS: write access will "
227                                "be enabled during recovery.\n");
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                         printk(KERN_WARNING
243                                "NILFS warning: unable to fall back to spare"
244                                "super block\n");
245                         goto scan_error;
246                 }
247                 printk(KERN_INFO
248                        "NILFS: try rollback from an earlier position\n");
249
250                 /*
251                  * restore super block with its spare and reconfigure
252                  * relevant states of the nilfs object.
253                  */
254                 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
255                 nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
256                 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
257
258                 /* verify consistency between two super blocks */
259                 blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
260                 if (blocksize != nilfs->ns_blocksize) {
261                         printk(KERN_WARNING
262                                "NILFS warning: blocksize differs between "
263                                "two super blocks (%d != %d)\n",
264                                blocksize, nilfs->ns_blocksize);
265                         goto scan_error;
266                 }
267
268                 err = nilfs_store_log_cursor(nilfs, sbp[0]);
269                 if (err)
270                         goto scan_error;
271
272                 /* drop clean flag to allow roll-forward and recovery */
273                 nilfs->ns_mount_state &= ~NILFS_VALID_FS;
274                 valid_fs = 0;
275
276                 err = nilfs_search_super_root(nilfs, &ri);
277                 if (err)
278                         goto scan_error;
279         }
280
281         err = nilfs_load_super_root(nilfs, sb, ri.ri_super_root);
282         if (unlikely(err)) {
283                 printk(KERN_ERR "NILFS: error loading super root.\n");
284                 goto failed;
285         }
286
287         if (valid_fs)
288                 goto skip_recovery;
289
290         if (s_flags & MS_RDONLY) {
291                 __u64 features;
292
293                 if (nilfs_test_opt(nilfs, NORECOVERY)) {
294                         printk(KERN_INFO "NILFS: norecovery option specified. "
295                                "skipping roll-forward recovery\n");
296                         goto skip_recovery;
297                 }
298                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
299                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
300                 if (features) {
301                         printk(KERN_ERR "NILFS: couldn't proceed with "
302                                "recovery because of unsupported optional "
303                                "features (%llx)\n",
304                                (unsigned long long)features);
305                         err = -EROFS;
306                         goto failed_unload;
307                 }
308                 if (really_read_only) {
309                         printk(KERN_ERR "NILFS: write access "
310                                "unavailable, cannot proceed.\n");
311                         err = -EROFS;
312                         goto failed_unload;
313                 }
314                 sb->s_flags &= ~MS_RDONLY;
315         } else if (nilfs_test_opt(nilfs, NORECOVERY)) {
316                 printk(KERN_ERR "NILFS: recovery cancelled because norecovery "
317                        "option was specified for a read/write mount\n");
318                 err = -EINVAL;
319                 goto failed_unload;
320         }
321
322         err = nilfs_salvage_orphan_logs(nilfs, sb, &ri);
323         if (err)
324                 goto failed_unload;
325
326         down_write(&nilfs->ns_sem);
327         nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
328         err = nilfs_cleanup_super(sb);
329         up_write(&nilfs->ns_sem);
330
331         if (err) {
332                 printk(KERN_ERR "NILFS: failed to update super block. "
333                        "recovery unfinished.\n");
334                 goto failed_unload;
335         }
336         printk(KERN_INFO "NILFS: recovery complete.\n");
337
338  skip_recovery:
339         nilfs_clear_recovery_info(&ri);
340         sb->s_flags = s_flags;
341         return 0;
342
343  scan_error:
344         printk(KERN_ERR "NILFS: error searching super root.\n");
345         goto failed;
346
347  failed_unload:
348         iput(nilfs->ns_cpfile);
349         iput(nilfs->ns_sufile);
350         iput(nilfs->ns_dat);
351
352  failed:
353         nilfs_clear_recovery_info(&ri);
354         sb->s_flags = s_flags;
355         return err;
356 }
357
358 static unsigned long long nilfs_max_size(unsigned int blkbits)
359 {
360         unsigned int max_bits;
361         unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
362
363         max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
364         if (max_bits < 64)
365                 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
366         return res;
367 }
368
369 /**
370  * nilfs_nrsvsegs - calculate the number of reserved segments
371  * @nilfs: nilfs object
372  * @nsegs: total number of segments
373  */
374 unsigned long nilfs_nrsvsegs(struct the_nilfs *nilfs, unsigned long nsegs)
375 {
376         return max_t(unsigned long, NILFS_MIN_NRSVSEGS,
377                      DIV_ROUND_UP(nsegs * nilfs->ns_r_segments_percentage,
378                                   100));
379 }
380
381 void nilfs_set_nsegments(struct the_nilfs *nilfs, unsigned long nsegs)
382 {
383         nilfs->ns_nsegments = nsegs;
384         nilfs->ns_nrsvsegs = nilfs_nrsvsegs(nilfs, nsegs);
385 }
386
387 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
388                                    struct nilfs_super_block *sbp)
389 {
390         if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
391                 printk(KERN_ERR "NILFS: unsupported revision "
392                        "(superblock rev.=%d.%d, current rev.=%d.%d). "
393                        "Please check the version of mkfs.nilfs.\n",
394                        le32_to_cpu(sbp->s_rev_level),
395                        le16_to_cpu(sbp->s_minor_rev_level),
396                        NILFS_CURRENT_REV, NILFS_MINOR_REV);
397                 return -EINVAL;
398         }
399         nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
400         if (nilfs->ns_sbsize > BLOCK_SIZE)
401                 return -EINVAL;
402
403         nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
404         if (nilfs->ns_inode_size > nilfs->ns_blocksize) {
405                 printk(KERN_ERR "NILFS: too large inode size: %d bytes.\n",
406                        nilfs->ns_inode_size);
407                 return -EINVAL;
408         } else if (nilfs->ns_inode_size < NILFS_MIN_INODE_SIZE) {
409                 printk(KERN_ERR "NILFS: too small inode size: %d bytes.\n",
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                 printk(KERN_ERR "NILFS: too short segment.\n");
419                 return -EINVAL;
420         }
421
422         nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
423         nilfs->ns_r_segments_percentage =
424                 le32_to_cpu(sbp->s_r_segments_percentage);
425         if (nilfs->ns_r_segments_percentage < 1 ||
426             nilfs->ns_r_segments_percentage > 99) {
427                 printk(KERN_ERR "NILFS: invalid reserved segments percentage.\n");
428                 return -EINVAL;
429         }
430
431         nilfs_set_nsegments(nilfs, le64_to_cpu(sbp->s_nsegments));
432         nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
433         return 0;
434 }
435
436 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
437 {
438         static unsigned char sum[4];
439         const int sumoff = offsetof(struct nilfs_super_block, s_sum);
440         size_t bytes;
441         u32 crc;
442
443         if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
444                 return 0;
445         bytes = le16_to_cpu(sbp->s_bytes);
446         if (bytes > BLOCK_SIZE)
447                 return 0;
448         crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
449                        sumoff);
450         crc = crc32_le(crc, sum, 4);
451         crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
452                        bytes - sumoff - 4);
453         return crc == le32_to_cpu(sbp->s_sum);
454 }
455
456 static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
457 {
458         return offset < ((le64_to_cpu(sbp->s_nsegments) *
459                           le32_to_cpu(sbp->s_blocks_per_segment)) <<
460                          (le32_to_cpu(sbp->s_log_block_size) + 10));
461 }
462
463 static void nilfs_release_super_block(struct the_nilfs *nilfs)
464 {
465         int i;
466
467         for (i = 0; i < 2; i++) {
468                 if (nilfs->ns_sbp[i]) {
469                         brelse(nilfs->ns_sbh[i]);
470                         nilfs->ns_sbh[i] = NULL;
471                         nilfs->ns_sbp[i] = NULL;
472                 }
473         }
474 }
475
476 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
477 {
478         brelse(nilfs->ns_sbh[0]);
479         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
480         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
481         nilfs->ns_sbh[1] = NULL;
482         nilfs->ns_sbp[1] = NULL;
483 }
484
485 void nilfs_swap_super_block(struct the_nilfs *nilfs)
486 {
487         struct buffer_head *tsbh = nilfs->ns_sbh[0];
488         struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
489
490         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
491         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
492         nilfs->ns_sbh[1] = tsbh;
493         nilfs->ns_sbp[1] = tsbp;
494 }
495
496 static int nilfs_load_super_block(struct the_nilfs *nilfs,
497                                   struct super_block *sb, int blocksize,
498                                   struct nilfs_super_block **sbpp)
499 {
500         struct nilfs_super_block **sbp = nilfs->ns_sbp;
501         struct buffer_head **sbh = nilfs->ns_sbh;
502         u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
503         int valid[2], swp = 0;
504
505         sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
506                                         &sbh[0]);
507         sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
508
509         if (!sbp[0]) {
510                 if (!sbp[1]) {
511                         printk(KERN_ERR "NILFS: unable to read superblock\n");
512                         return -EIO;
513                 }
514                 printk(KERN_WARNING
515                        "NILFS warning: unable to read primary superblock "
516                        "(blocksize = %d)\n", blocksize);
517         } else if (!sbp[1]) {
518                 printk(KERN_WARNING
519                        "NILFS warning: unable to read secondary superblock "
520                        "(blocksize = %d)\n", blocksize);
521         }
522
523         /*
524          * Compare two super blocks and set 1 in swp if the secondary
525          * super block is valid and newer.  Otherwise, set 0 in swp.
526          */
527         valid[0] = nilfs_valid_sb(sbp[0]);
528         valid[1] = nilfs_valid_sb(sbp[1]);
529         swp = valid[1] && (!valid[0] ||
530                            le64_to_cpu(sbp[1]->s_last_cno) >
531                            le64_to_cpu(sbp[0]->s_last_cno));
532
533         if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
534                 brelse(sbh[1]);
535                 sbh[1] = NULL;
536                 sbp[1] = NULL;
537                 valid[1] = 0;
538                 swp = 0;
539         }
540         if (!valid[swp]) {
541                 nilfs_release_super_block(nilfs);
542                 printk(KERN_ERR "NILFS: Can't find nilfs on dev %s.\n",
543                        sb->s_id);
544                 return -EINVAL;
545         }
546
547         if (!valid[!swp])
548                 printk(KERN_WARNING "NILFS warning: broken superblock. "
549                        "using spare superblock (blocksize = %d).\n", blocksize);
550         if (swp)
551                 nilfs_swap_super_block(nilfs);
552
553         nilfs->ns_sbwcount = 0;
554         nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
555         nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
556         *sbpp = sbp[0];
557         return 0;
558 }
559
560 /**
561  * init_nilfs - initialize a NILFS instance.
562  * @nilfs: the_nilfs structure
563  * @sb: super block
564  * @data: mount options
565  *
566  * init_nilfs() performs common initialization per block device (e.g.
567  * reading the super block, getting disk layout information, initializing
568  * shared fields in the_nilfs).
569  *
570  * Return Value: On success, 0 is returned. On error, a negative error
571  * code is returned.
572  */
573 int init_nilfs(struct the_nilfs *nilfs, struct super_block *sb, char *data)
574 {
575         struct nilfs_super_block *sbp;
576         int blocksize;
577         int err;
578
579         down_write(&nilfs->ns_sem);
580
581         blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
582         if (!blocksize) {
583                 printk(KERN_ERR "NILFS: unable to set blocksize\n");
584                 err = -EINVAL;
585                 goto out;
586         }
587         err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
588         if (err)
589                 goto out;
590
591         err = nilfs_store_magic_and_option(sb, sbp, data);
592         if (err)
593                 goto failed_sbh;
594
595         err = nilfs_check_feature_compatibility(sb, sbp);
596         if (err)
597                 goto failed_sbh;
598
599         blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
600         if (blocksize < NILFS_MIN_BLOCK_SIZE ||
601             blocksize > NILFS_MAX_BLOCK_SIZE) {
602                 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
603                        "filesystem blocksize %d\n", blocksize);
604                 err = -EINVAL;
605                 goto failed_sbh;
606         }
607         if (sb->s_blocksize != blocksize) {
608                 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
609
610                 if (blocksize < hw_blocksize) {
611                         printk(KERN_ERR
612                                "NILFS: blocksize %d too small for device "
613                                "(sector-size = %d).\n",
614                                blocksize, hw_blocksize);
615                         err = -EINVAL;
616                         goto failed_sbh;
617                 }
618                 nilfs_release_super_block(nilfs);
619                 sb_set_blocksize(sb, blocksize);
620
621                 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
622                 if (err)
623                         goto out;
624                         /* not failed_sbh; sbh is released automatically
625                            when reloading fails. */
626         }
627         nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
628         nilfs->ns_blocksize = blocksize;
629
630         get_random_bytes(&nilfs->ns_next_generation,
631                          sizeof(nilfs->ns_next_generation));
632
633         err = nilfs_store_disk_layout(nilfs, sbp);
634         if (err)
635                 goto failed_sbh;
636
637         sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
638
639         nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
640
641         err = nilfs_store_log_cursor(nilfs, sbp);
642         if (err)
643                 goto failed_sbh;
644
645         err = nilfs_sysfs_create_device_group(sb);
646         if (err)
647                 goto failed_sbh;
648
649         set_nilfs_init(nilfs);
650         err = 0;
651  out:
652         up_write(&nilfs->ns_sem);
653         return err;
654
655  failed_sbh:
656         nilfs_release_super_block(nilfs);
657         goto out;
658 }
659
660 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
661                             size_t nsegs)
662 {
663         sector_t seg_start, seg_end;
664         sector_t start = 0, nblocks = 0;
665         unsigned int sects_per_block;
666         __u64 *sn;
667         int ret = 0;
668
669         sects_per_block = (1 << nilfs->ns_blocksize_bits) /
670                 bdev_logical_block_size(nilfs->ns_bdev);
671         for (sn = segnump; sn < segnump + nsegs; sn++) {
672                 nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
673
674                 if (!nblocks) {
675                         start = seg_start;
676                         nblocks = seg_end - seg_start + 1;
677                 } else if (start + nblocks == seg_start) {
678                         nblocks += seg_end - seg_start + 1;
679                 } else {
680                         ret = blkdev_issue_discard(nilfs->ns_bdev,
681                                                    start * sects_per_block,
682                                                    nblocks * sects_per_block,
683                                                    GFP_NOFS, 0);
684                         if (ret < 0)
685                                 return ret;
686                         nblocks = 0;
687                 }
688         }
689         if (nblocks)
690                 ret = blkdev_issue_discard(nilfs->ns_bdev,
691                                            start * sects_per_block,
692                                            nblocks * sects_per_block,
693                                            GFP_NOFS, 0);
694         return ret;
695 }
696
697 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
698 {
699         unsigned long ncleansegs;
700
701         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
702         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
703         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
704         *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
705         return 0;
706 }
707
708 int nilfs_near_disk_full(struct the_nilfs *nilfs)
709 {
710         unsigned long ncleansegs, nincsegs;
711
712         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
713         nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
714                 nilfs->ns_blocks_per_segment + 1;
715
716         return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
717 }
718
719 struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
720 {
721         struct rb_node *n;
722         struct nilfs_root *root;
723
724         spin_lock(&nilfs->ns_cptree_lock);
725         n = nilfs->ns_cptree.rb_node;
726         while (n) {
727                 root = rb_entry(n, struct nilfs_root, rb_node);
728
729                 if (cno < root->cno) {
730                         n = n->rb_left;
731                 } else if (cno > root->cno) {
732                         n = n->rb_right;
733                 } else {
734                         atomic_inc(&root->count);
735                         spin_unlock(&nilfs->ns_cptree_lock);
736                         return root;
737                 }
738         }
739         spin_unlock(&nilfs->ns_cptree_lock);
740
741         return NULL;
742 }
743
744 struct nilfs_root *
745 nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
746 {
747         struct rb_node **p, *parent;
748         struct nilfs_root *root, *new;
749         int err;
750
751         root = nilfs_lookup_root(nilfs, cno);
752         if (root)
753                 return root;
754
755         new = kzalloc(sizeof(*root), GFP_KERNEL);
756         if (!new)
757                 return NULL;
758
759         spin_lock(&nilfs->ns_cptree_lock);
760
761         p = &nilfs->ns_cptree.rb_node;
762         parent = NULL;
763
764         while (*p) {
765                 parent = *p;
766                 root = rb_entry(parent, struct nilfs_root, rb_node);
767
768                 if (cno < root->cno) {
769                         p = &(*p)->rb_left;
770                 } else if (cno > root->cno) {
771                         p = &(*p)->rb_right;
772                 } else {
773                         atomic_inc(&root->count);
774                         spin_unlock(&nilfs->ns_cptree_lock);
775                         kfree(new);
776                         return root;
777                 }
778         }
779
780         new->cno = cno;
781         new->ifile = NULL;
782         new->nilfs = nilfs;
783         atomic_set(&new->count, 1);
784         atomic64_set(&new->inodes_count, 0);
785         atomic64_set(&new->blocks_count, 0);
786
787         rb_link_node(&new->rb_node, parent, p);
788         rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
789
790         spin_unlock(&nilfs->ns_cptree_lock);
791
792         err = nilfs_sysfs_create_snapshot_group(new);
793         if (err) {
794                 kfree(new);
795                 new = NULL;
796         }
797
798         return new;
799 }
800
801 void nilfs_put_root(struct nilfs_root *root)
802 {
803         if (atomic_dec_and_test(&root->count)) {
804                 struct the_nilfs *nilfs = root->nilfs;
805
806                 nilfs_sysfs_delete_snapshot_group(root);
807
808                 spin_lock(&nilfs->ns_cptree_lock);
809                 rb_erase(&root->rb_node, &nilfs->ns_cptree);
810                 spin_unlock(&nilfs->ns_cptree_lock);
811                 iput(root->ifile);
812
813                 kfree(root);
814         }
815 }