Merge branch 'v2.6.36-rc8' into for-2.6.37/barrier
[linux-3.10.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  * 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/crc32.h>
29 #include "nilfs.h"
30 #include "segment.h"
31 #include "alloc.h"
32 #include "cpfile.h"
33 #include "sufile.h"
34 #include "dat.h"
35 #include "segbuf.h"
36
37
38 static LIST_HEAD(nilfs_objects);
39 static DEFINE_SPINLOCK(nilfs_lock);
40
41 static int nilfs_valid_sb(struct nilfs_super_block *sbp);
42
43 void nilfs_set_last_segment(struct the_nilfs *nilfs,
44                             sector_t start_blocknr, u64 seq, __u64 cno)
45 {
46         spin_lock(&nilfs->ns_last_segment_lock);
47         nilfs->ns_last_pseg = start_blocknr;
48         nilfs->ns_last_seq = seq;
49         nilfs->ns_last_cno = cno;
50
51         if (!nilfs_sb_dirty(nilfs)) {
52                 if (nilfs->ns_prev_seq == nilfs->ns_last_seq)
53                         goto stay_cursor;
54
55                 set_nilfs_sb_dirty(nilfs);
56         }
57         nilfs->ns_prev_seq = nilfs->ns_last_seq;
58
59  stay_cursor:
60         spin_unlock(&nilfs->ns_last_segment_lock);
61 }
62
63 /**
64  * alloc_nilfs - allocate the_nilfs structure
65  * @bdev: block device to which the_nilfs is related
66  *
67  * alloc_nilfs() allocates memory for the_nilfs and
68  * initializes its reference count and locks.
69  *
70  * Return Value: On success, pointer to the_nilfs is returned.
71  * On error, NULL is returned.
72  */
73 static struct the_nilfs *alloc_nilfs(struct block_device *bdev)
74 {
75         struct the_nilfs *nilfs;
76
77         nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
78         if (!nilfs)
79                 return NULL;
80
81         nilfs->ns_bdev = bdev;
82         atomic_set(&nilfs->ns_count, 1);
83         atomic_set(&nilfs->ns_ndirtyblks, 0);
84         init_rwsem(&nilfs->ns_sem);
85         init_rwsem(&nilfs->ns_super_sem);
86         mutex_init(&nilfs->ns_mount_mutex);
87         init_rwsem(&nilfs->ns_writer_sem);
88         INIT_LIST_HEAD(&nilfs->ns_list);
89         INIT_LIST_HEAD(&nilfs->ns_supers);
90         spin_lock_init(&nilfs->ns_last_segment_lock);
91         nilfs->ns_gc_inodes_h = NULL;
92         init_rwsem(&nilfs->ns_segctor_sem);
93
94         return nilfs;
95 }
96
97 /**
98  * find_or_create_nilfs - find or create nilfs object
99  * @bdev: block device to which the_nilfs is related
100  *
101  * find_nilfs() looks up an existent nilfs object created on the
102  * device and gets the reference count of the object.  If no nilfs object
103  * is found on the device, a new nilfs object is allocated.
104  *
105  * Return Value: On success, pointer to the nilfs object is returned.
106  * On error, NULL is returned.
107  */
108 struct the_nilfs *find_or_create_nilfs(struct block_device *bdev)
109 {
110         struct the_nilfs *nilfs, *new = NULL;
111
112  retry:
113         spin_lock(&nilfs_lock);
114         list_for_each_entry(nilfs, &nilfs_objects, ns_list) {
115                 if (nilfs->ns_bdev == bdev) {
116                         get_nilfs(nilfs);
117                         spin_unlock(&nilfs_lock);
118                         if (new)
119                                 put_nilfs(new);
120                         return nilfs; /* existing object */
121                 }
122         }
123         if (new) {
124                 list_add_tail(&new->ns_list, &nilfs_objects);
125                 spin_unlock(&nilfs_lock);
126                 return new; /* new object */
127         }
128         spin_unlock(&nilfs_lock);
129
130         new = alloc_nilfs(bdev);
131         if (new)
132                 goto retry;
133         return NULL; /* insufficient memory */
134 }
135
136 /**
137  * put_nilfs - release a reference to the_nilfs
138  * @nilfs: the_nilfs structure to be released
139  *
140  * put_nilfs() decrements a reference counter of the_nilfs.
141  * If the reference count reaches zero, the_nilfs is freed.
142  */
143 void put_nilfs(struct the_nilfs *nilfs)
144 {
145         spin_lock(&nilfs_lock);
146         if (!atomic_dec_and_test(&nilfs->ns_count)) {
147                 spin_unlock(&nilfs_lock);
148                 return;
149         }
150         list_del_init(&nilfs->ns_list);
151         spin_unlock(&nilfs_lock);
152
153         /*
154          * Increment of ns_count never occurs below because the caller
155          * of get_nilfs() holds at least one reference to the_nilfs.
156          * Thus its exclusion control is not required here.
157          */
158
159         might_sleep();
160         if (nilfs_loaded(nilfs)) {
161                 nilfs_mdt_destroy(nilfs->ns_sufile);
162                 nilfs_mdt_destroy(nilfs->ns_cpfile);
163                 nilfs_mdt_destroy(nilfs->ns_dat);
164                 nilfs_mdt_destroy(nilfs->ns_gc_dat);
165         }
166         if (nilfs_init(nilfs)) {
167                 nilfs_destroy_gccache(nilfs);
168                 brelse(nilfs->ns_sbh[0]);
169                 brelse(nilfs->ns_sbh[1]);
170         }
171         kfree(nilfs);
172 }
173
174 static int nilfs_load_super_root(struct the_nilfs *nilfs, sector_t sr_block)
175 {
176         struct buffer_head *bh_sr;
177         struct nilfs_super_root *raw_sr;
178         struct nilfs_super_block **sbp = nilfs->ns_sbp;
179         unsigned dat_entry_size, segment_usage_size, checkpoint_size;
180         unsigned inode_size;
181         int err;
182
183         err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
184         if (unlikely(err))
185                 return err;
186
187         down_read(&nilfs->ns_sem);
188         dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
189         checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
190         segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
191         up_read(&nilfs->ns_sem);
192
193         inode_size = nilfs->ns_inode_size;
194
195         err = -ENOMEM;
196         nilfs->ns_dat = nilfs_dat_new(nilfs, dat_entry_size);
197         if (unlikely(!nilfs->ns_dat))
198                 goto failed;
199
200         nilfs->ns_gc_dat = nilfs_dat_new(nilfs, dat_entry_size);
201         if (unlikely(!nilfs->ns_gc_dat))
202                 goto failed_dat;
203
204         nilfs->ns_cpfile = nilfs_cpfile_new(nilfs, checkpoint_size);
205         if (unlikely(!nilfs->ns_cpfile))
206                 goto failed_gc_dat;
207
208         nilfs->ns_sufile = nilfs_sufile_new(nilfs, segment_usage_size);
209         if (unlikely(!nilfs->ns_sufile))
210                 goto failed_cpfile;
211
212         nilfs_mdt_set_shadow(nilfs->ns_dat, nilfs->ns_gc_dat);
213
214         err = nilfs_dat_read(nilfs->ns_dat, (void *)bh_sr->b_data +
215                              NILFS_SR_DAT_OFFSET(inode_size));
216         if (unlikely(err))
217                 goto failed_sufile;
218
219         err = nilfs_cpfile_read(nilfs->ns_cpfile, (void *)bh_sr->b_data +
220                                 NILFS_SR_CPFILE_OFFSET(inode_size));
221         if (unlikely(err))
222                 goto failed_sufile;
223
224         err = nilfs_sufile_read(nilfs->ns_sufile, (void *)bh_sr->b_data +
225                                 NILFS_SR_SUFILE_OFFSET(inode_size));
226         if (unlikely(err))
227                 goto failed_sufile;
228
229         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
230         nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
231
232  failed:
233         brelse(bh_sr);
234         return err;
235
236  failed_sufile:
237         nilfs_mdt_destroy(nilfs->ns_sufile);
238
239  failed_cpfile:
240         nilfs_mdt_destroy(nilfs->ns_cpfile);
241
242  failed_gc_dat:
243         nilfs_mdt_destroy(nilfs->ns_gc_dat);
244
245  failed_dat:
246         nilfs_mdt_destroy(nilfs->ns_dat);
247         goto failed;
248 }
249
250 static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
251 {
252         memset(ri, 0, sizeof(*ri));
253         INIT_LIST_HEAD(&ri->ri_used_segments);
254 }
255
256 static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
257 {
258         nilfs_dispose_segment_list(&ri->ri_used_segments);
259 }
260
261 /**
262  * nilfs_store_log_cursor - load log cursor from a super block
263  * @nilfs: nilfs object
264  * @sbp: buffer storing super block to be read
265  *
266  * nilfs_store_log_cursor() reads the last position of the log
267  * containing a super root from a given super block, and initializes
268  * relevant information on the nilfs object preparatory for log
269  * scanning and recovery.
270  */
271 static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
272                                   struct nilfs_super_block *sbp)
273 {
274         int ret = 0;
275
276         nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
277         nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
278         nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
279
280         nilfs->ns_prev_seq = nilfs->ns_last_seq;
281         nilfs->ns_seg_seq = nilfs->ns_last_seq;
282         nilfs->ns_segnum =
283                 nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
284         nilfs->ns_cno = nilfs->ns_last_cno + 1;
285         if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
286                 printk(KERN_ERR "NILFS invalid last segment number.\n");
287                 ret = -EINVAL;
288         }
289         return ret;
290 }
291
292 /**
293  * load_nilfs - load and recover the nilfs
294  * @nilfs: the_nilfs structure to be released
295  * @sbi: nilfs_sb_info used to recover past segment
296  *
297  * load_nilfs() searches and load the latest super root,
298  * attaches the last segment, and does recovery if needed.
299  * The caller must call this exclusively for simultaneous mounts.
300  */
301 int load_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
302 {
303         struct nilfs_recovery_info ri;
304         unsigned int s_flags = sbi->s_super->s_flags;
305         int really_read_only = bdev_read_only(nilfs->ns_bdev);
306         int valid_fs = nilfs_valid_fs(nilfs);
307         int err;
308
309         if (nilfs_loaded(nilfs)) {
310                 if (valid_fs ||
311                     ((s_flags & MS_RDONLY) && nilfs_test_opt(sbi, NORECOVERY)))
312                         return 0;
313                 printk(KERN_ERR "NILFS: the filesystem is in an incomplete "
314                        "recovery state.\n");
315                 return -EINVAL;
316         }
317
318         if (!valid_fs) {
319                 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
320                 if (s_flags & MS_RDONLY) {
321                         printk(KERN_INFO "NILFS: INFO: recovery "
322                                "required for readonly filesystem.\n");
323                         printk(KERN_INFO "NILFS: write access will "
324                                "be enabled during recovery.\n");
325                 }
326         }
327
328         nilfs_init_recovery_info(&ri);
329
330         err = nilfs_search_super_root(nilfs, &ri);
331         if (unlikely(err)) {
332                 struct nilfs_super_block **sbp = nilfs->ns_sbp;
333                 int blocksize;
334
335                 if (err != -EINVAL)
336                         goto scan_error;
337
338                 if (!nilfs_valid_sb(sbp[1])) {
339                         printk(KERN_WARNING
340                                "NILFS warning: unable to fall back to spare"
341                                "super block\n");
342                         goto scan_error;
343                 }
344                 printk(KERN_INFO
345                        "NILFS: try rollback from an earlier position\n");
346
347                 /*
348                  * restore super block with its spare and reconfigure
349                  * relevant states of the nilfs object.
350                  */
351                 memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
352                 nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
353                 nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
354
355                 /* verify consistency between two super blocks */
356                 blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
357                 if (blocksize != nilfs->ns_blocksize) {
358                         printk(KERN_WARNING
359                                "NILFS warning: blocksize differs between "
360                                "two super blocks (%d != %d)\n",
361                                blocksize, nilfs->ns_blocksize);
362                         goto scan_error;
363                 }
364
365                 err = nilfs_store_log_cursor(nilfs, sbp[0]);
366                 if (err)
367                         goto scan_error;
368
369                 /* drop clean flag to allow roll-forward and recovery */
370                 nilfs->ns_mount_state &= ~NILFS_VALID_FS;
371                 valid_fs = 0;
372
373                 err = nilfs_search_super_root(nilfs, &ri);
374                 if (err)
375                         goto scan_error;
376         }
377
378         err = nilfs_load_super_root(nilfs, ri.ri_super_root);
379         if (unlikely(err)) {
380                 printk(KERN_ERR "NILFS: error loading super root.\n");
381                 goto failed;
382         }
383
384         if (valid_fs)
385                 goto skip_recovery;
386
387         if (s_flags & MS_RDONLY) {
388                 __u64 features;
389
390                 if (nilfs_test_opt(sbi, NORECOVERY)) {
391                         printk(KERN_INFO "NILFS: norecovery option specified. "
392                                "skipping roll-forward recovery\n");
393                         goto skip_recovery;
394                 }
395                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
396                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
397                 if (features) {
398                         printk(KERN_ERR "NILFS: couldn't proceed with "
399                                "recovery because of unsupported optional "
400                                "features (%llx)\n",
401                                (unsigned long long)features);
402                         err = -EROFS;
403                         goto failed_unload;
404                 }
405                 if (really_read_only) {
406                         printk(KERN_ERR "NILFS: write access "
407                                "unavailable, cannot proceed.\n");
408                         err = -EROFS;
409                         goto failed_unload;
410                 }
411                 sbi->s_super->s_flags &= ~MS_RDONLY;
412         } else if (nilfs_test_opt(sbi, NORECOVERY)) {
413                 printk(KERN_ERR "NILFS: recovery cancelled because norecovery "
414                        "option was specified for a read/write mount\n");
415                 err = -EINVAL;
416                 goto failed_unload;
417         }
418
419         err = nilfs_salvage_orphan_logs(nilfs, sbi, &ri);
420         if (err)
421                 goto failed_unload;
422
423         down_write(&nilfs->ns_sem);
424         nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
425         err = nilfs_cleanup_super(sbi);
426         up_write(&nilfs->ns_sem);
427
428         if (err) {
429                 printk(KERN_ERR "NILFS: failed to update super block. "
430                        "recovery unfinished.\n");
431                 goto failed_unload;
432         }
433         printk(KERN_INFO "NILFS: recovery complete.\n");
434
435  skip_recovery:
436         set_nilfs_loaded(nilfs);
437         nilfs_clear_recovery_info(&ri);
438         sbi->s_super->s_flags = s_flags;
439         return 0;
440
441  scan_error:
442         printk(KERN_ERR "NILFS: error searching super root.\n");
443         goto failed;
444
445  failed_unload:
446         nilfs_mdt_destroy(nilfs->ns_cpfile);
447         nilfs_mdt_destroy(nilfs->ns_sufile);
448         nilfs_mdt_destroy(nilfs->ns_dat);
449         nilfs_mdt_destroy(nilfs->ns_gc_dat);
450
451  failed:
452         nilfs_clear_recovery_info(&ri);
453         sbi->s_super->s_flags = s_flags;
454         return err;
455 }
456
457 static unsigned long long nilfs_max_size(unsigned int blkbits)
458 {
459         unsigned int max_bits;
460         unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
461
462         max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
463         if (max_bits < 64)
464                 res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
465         return res;
466 }
467
468 static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
469                                    struct nilfs_super_block *sbp)
470 {
471         if (le32_to_cpu(sbp->s_rev_level) != NILFS_CURRENT_REV) {
472                 printk(KERN_ERR "NILFS: revision mismatch "
473                        "(superblock rev.=%d.%d, current rev.=%d.%d). "
474                        "Please check the version of mkfs.nilfs.\n",
475                        le32_to_cpu(sbp->s_rev_level),
476                        le16_to_cpu(sbp->s_minor_rev_level),
477                        NILFS_CURRENT_REV, NILFS_MINOR_REV);
478                 return -EINVAL;
479         }
480         nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
481         if (nilfs->ns_sbsize > BLOCK_SIZE)
482                 return -EINVAL;
483
484         nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
485         nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
486
487         nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
488         if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
489                 printk(KERN_ERR "NILFS: too short segment.\n");
490                 return -EINVAL;
491         }
492
493         nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
494         nilfs->ns_nsegments = le64_to_cpu(sbp->s_nsegments);
495         nilfs->ns_r_segments_percentage =
496                 le32_to_cpu(sbp->s_r_segments_percentage);
497         nilfs->ns_nrsvsegs =
498                 max_t(unsigned long, NILFS_MIN_NRSVSEGS,
499                       DIV_ROUND_UP(nilfs->ns_nsegments *
500                                    nilfs->ns_r_segments_percentage, 100));
501         nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
502         return 0;
503 }
504
505 static int nilfs_valid_sb(struct nilfs_super_block *sbp)
506 {
507         static unsigned char sum[4];
508         const int sumoff = offsetof(struct nilfs_super_block, s_sum);
509         size_t bytes;
510         u32 crc;
511
512         if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
513                 return 0;
514         bytes = le16_to_cpu(sbp->s_bytes);
515         if (bytes > BLOCK_SIZE)
516                 return 0;
517         crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
518                        sumoff);
519         crc = crc32_le(crc, sum, 4);
520         crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
521                        bytes - sumoff - 4);
522         return crc == le32_to_cpu(sbp->s_sum);
523 }
524
525 static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
526 {
527         return offset < ((le64_to_cpu(sbp->s_nsegments) *
528                           le32_to_cpu(sbp->s_blocks_per_segment)) <<
529                          (le32_to_cpu(sbp->s_log_block_size) + 10));
530 }
531
532 static void nilfs_release_super_block(struct the_nilfs *nilfs)
533 {
534         int i;
535
536         for (i = 0; i < 2; i++) {
537                 if (nilfs->ns_sbp[i]) {
538                         brelse(nilfs->ns_sbh[i]);
539                         nilfs->ns_sbh[i] = NULL;
540                         nilfs->ns_sbp[i] = NULL;
541                 }
542         }
543 }
544
545 void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
546 {
547         brelse(nilfs->ns_sbh[0]);
548         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
549         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
550         nilfs->ns_sbh[1] = NULL;
551         nilfs->ns_sbp[1] = NULL;
552 }
553
554 void nilfs_swap_super_block(struct the_nilfs *nilfs)
555 {
556         struct buffer_head *tsbh = nilfs->ns_sbh[0];
557         struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
558
559         nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
560         nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
561         nilfs->ns_sbh[1] = tsbh;
562         nilfs->ns_sbp[1] = tsbp;
563 }
564
565 static int nilfs_load_super_block(struct the_nilfs *nilfs,
566                                   struct super_block *sb, int blocksize,
567                                   struct nilfs_super_block **sbpp)
568 {
569         struct nilfs_super_block **sbp = nilfs->ns_sbp;
570         struct buffer_head **sbh = nilfs->ns_sbh;
571         u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
572         int valid[2], swp = 0;
573
574         sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
575                                         &sbh[0]);
576         sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
577
578         if (!sbp[0]) {
579                 if (!sbp[1]) {
580                         printk(KERN_ERR "NILFS: unable to read superblock\n");
581                         return -EIO;
582                 }
583                 printk(KERN_WARNING
584                        "NILFS warning: unable to read primary superblock\n");
585         } else if (!sbp[1])
586                 printk(KERN_WARNING
587                        "NILFS warning: unable to read secondary superblock\n");
588
589         /*
590          * Compare two super blocks and set 1 in swp if the secondary
591          * super block is valid and newer.  Otherwise, set 0 in swp.
592          */
593         valid[0] = nilfs_valid_sb(sbp[0]);
594         valid[1] = nilfs_valid_sb(sbp[1]);
595         swp = valid[1] && (!valid[0] ||
596                            le64_to_cpu(sbp[1]->s_last_cno) >
597                            le64_to_cpu(sbp[0]->s_last_cno));
598
599         if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
600                 brelse(sbh[1]);
601                 sbh[1] = NULL;
602                 sbp[1] = NULL;
603                 swp = 0;
604         }
605         if (!valid[swp]) {
606                 nilfs_release_super_block(nilfs);
607                 printk(KERN_ERR "NILFS: Can't find nilfs on dev %s.\n",
608                        sb->s_id);
609                 return -EINVAL;
610         }
611
612         if (!valid[!swp])
613                 printk(KERN_WARNING "NILFS warning: broken superblock. "
614                        "using spare superblock.\n");
615         if (swp)
616                 nilfs_swap_super_block(nilfs);
617
618         nilfs->ns_sbwcount = 0;
619         nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
620         nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
621         *sbpp = sbp[0];
622         return 0;
623 }
624
625 /**
626  * init_nilfs - initialize a NILFS instance.
627  * @nilfs: the_nilfs structure
628  * @sbi: nilfs_sb_info
629  * @sb: super block
630  * @data: mount options
631  *
632  * init_nilfs() performs common initialization per block device (e.g.
633  * reading the super block, getting disk layout information, initializing
634  * shared fields in the_nilfs). It takes on some portion of the jobs
635  * typically done by a fill_super() routine. This division arises from
636  * the nature that multiple NILFS instances may be simultaneously
637  * mounted on a device.
638  * For multiple mounts on the same device, only the first mount
639  * invokes these tasks.
640  *
641  * Return Value: On success, 0 is returned. On error, a negative error
642  * code is returned.
643  */
644 int init_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi, char *data)
645 {
646         struct super_block *sb = sbi->s_super;
647         struct nilfs_super_block *sbp;
648         struct backing_dev_info *bdi;
649         int blocksize;
650         int err;
651
652         down_write(&nilfs->ns_sem);
653         if (nilfs_init(nilfs)) {
654                 /* Load values from existing the_nilfs */
655                 sbp = nilfs->ns_sbp[0];
656                 err = nilfs_store_magic_and_option(sb, sbp, data);
657                 if (err)
658                         goto out;
659
660                 err = nilfs_check_feature_compatibility(sb, sbp);
661                 if (err)
662                         goto out;
663
664                 blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
665                 if (sb->s_blocksize != blocksize &&
666                     !sb_set_blocksize(sb, blocksize)) {
667                         printk(KERN_ERR "NILFS: blocksize %d unfit to device\n",
668                                blocksize);
669                         err = -EINVAL;
670                 }
671                 sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
672                 goto out;
673         }
674
675         blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
676         if (!blocksize) {
677                 printk(KERN_ERR "NILFS: unable to set blocksize\n");
678                 err = -EINVAL;
679                 goto out;
680         }
681         err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
682         if (err)
683                 goto out;
684
685         err = nilfs_store_magic_and_option(sb, sbp, data);
686         if (err)
687                 goto failed_sbh;
688
689         err = nilfs_check_feature_compatibility(sb, sbp);
690         if (err)
691                 goto failed_sbh;
692
693         blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
694         if (blocksize < NILFS_MIN_BLOCK_SIZE ||
695             blocksize > NILFS_MAX_BLOCK_SIZE) {
696                 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
697                        "filesystem blocksize %d\n", blocksize);
698                 err = -EINVAL;
699                 goto failed_sbh;
700         }
701         if (sb->s_blocksize != blocksize) {
702                 int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
703
704                 if (blocksize < hw_blocksize) {
705                         printk(KERN_ERR
706                                "NILFS: blocksize %d too small for device "
707                                "(sector-size = %d).\n",
708                                blocksize, hw_blocksize);
709                         err = -EINVAL;
710                         goto failed_sbh;
711                 }
712                 nilfs_release_super_block(nilfs);
713                 sb_set_blocksize(sb, blocksize);
714
715                 err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
716                 if (err)
717                         goto out;
718                         /* not failed_sbh; sbh is released automatically
719                            when reloading fails. */
720         }
721         nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
722         nilfs->ns_blocksize = blocksize;
723
724         err = nilfs_store_disk_layout(nilfs, sbp);
725         if (err)
726                 goto failed_sbh;
727
728         sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
729
730         nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
731
732         bdi = nilfs->ns_bdev->bd_inode->i_mapping->backing_dev_info;
733         nilfs->ns_bdi = bdi ? : &default_backing_dev_info;
734
735         err = nilfs_store_log_cursor(nilfs, sbp);
736         if (err)
737                 goto failed_sbh;
738
739         /* Initialize gcinode cache */
740         err = nilfs_init_gccache(nilfs);
741         if (err)
742                 goto failed_sbh;
743
744         set_nilfs_init(nilfs);
745         err = 0;
746  out:
747         up_write(&nilfs->ns_sem);
748         return err;
749
750  failed_sbh:
751         nilfs_release_super_block(nilfs);
752         goto out;
753 }
754
755 int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
756                             size_t nsegs)
757 {
758         sector_t seg_start, seg_end;
759         sector_t start = 0, nblocks = 0;
760         unsigned int sects_per_block;
761         __u64 *sn;
762         int ret = 0;
763
764         sects_per_block = (1 << nilfs->ns_blocksize_bits) /
765                 bdev_logical_block_size(nilfs->ns_bdev);
766         for (sn = segnump; sn < segnump + nsegs; sn++) {
767                 nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
768
769                 if (!nblocks) {
770                         start = seg_start;
771                         nblocks = seg_end - seg_start + 1;
772                 } else if (start + nblocks == seg_start) {
773                         nblocks += seg_end - seg_start + 1;
774                 } else {
775                         ret = blkdev_issue_discard(nilfs->ns_bdev,
776                                                    start * sects_per_block,
777                                                    nblocks * sects_per_block,
778                                                    GFP_NOFS, 0);
779                         if (ret < 0)
780                                 return ret;
781                         nblocks = 0;
782                 }
783         }
784         if (nblocks)
785                 ret = blkdev_issue_discard(nilfs->ns_bdev,
786                                            start * sects_per_block,
787                                            nblocks * sects_per_block,
788                                            GFP_NOFS, 0);
789         return ret;
790 }
791
792 int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
793 {
794         struct inode *dat = nilfs_dat_inode(nilfs);
795         unsigned long ncleansegs;
796
797         down_read(&NILFS_MDT(dat)->mi_sem);     /* XXX */
798         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
799         up_read(&NILFS_MDT(dat)->mi_sem);       /* XXX */
800         *nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
801         return 0;
802 }
803
804 int nilfs_near_disk_full(struct the_nilfs *nilfs)
805 {
806         unsigned long ncleansegs, nincsegs;
807
808         ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
809         nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
810                 nilfs->ns_blocks_per_segment + 1;
811
812         return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
813 }
814
815 /**
816  * nilfs_find_sbinfo - find existing nilfs_sb_info structure
817  * @nilfs: nilfs object
818  * @rw_mount: mount type (non-zero value for read/write mount)
819  * @cno: checkpoint number (zero for read-only mount)
820  *
821  * nilfs_find_sbinfo() returns the nilfs_sb_info structure which
822  * @rw_mount and @cno (in case of snapshots) matched.  If no instance
823  * was found, NULL is returned.  Although the super block instance can
824  * be unmounted after this function returns, the nilfs_sb_info struct
825  * is kept on memory until nilfs_put_sbinfo() is called.
826  */
827 struct nilfs_sb_info *nilfs_find_sbinfo(struct the_nilfs *nilfs,
828                                         int rw_mount, __u64 cno)
829 {
830         struct nilfs_sb_info *sbi;
831
832         down_read(&nilfs->ns_super_sem);
833         /*
834          * The SNAPSHOT flag and sb->s_flags are supposed to be
835          * protected with nilfs->ns_super_sem.
836          */
837         sbi = nilfs->ns_current;
838         if (rw_mount) {
839                 if (sbi && !(sbi->s_super->s_flags & MS_RDONLY))
840                         goto found; /* read/write mount */
841                 else
842                         goto out;
843         } else if (cno == 0) {
844                 if (sbi && (sbi->s_super->s_flags & MS_RDONLY))
845                         goto found; /* read-only mount */
846                 else
847                         goto out;
848         }
849
850         list_for_each_entry(sbi, &nilfs->ns_supers, s_list) {
851                 if (nilfs_test_opt(sbi, SNAPSHOT) &&
852                     sbi->s_snapshot_cno == cno)
853                         goto found; /* snapshot mount */
854         }
855  out:
856         up_read(&nilfs->ns_super_sem);
857         return NULL;
858
859  found:
860         atomic_inc(&sbi->s_count);
861         up_read(&nilfs->ns_super_sem);
862         return sbi;
863 }
864
865 int nilfs_checkpoint_is_mounted(struct the_nilfs *nilfs, __u64 cno,
866                                 int snapshot_mount)
867 {
868         struct nilfs_sb_info *sbi;
869         int ret = 0;
870
871         down_read(&nilfs->ns_super_sem);
872         if (cno == 0 || cno > nilfs->ns_cno)
873                 goto out_unlock;
874
875         list_for_each_entry(sbi, &nilfs->ns_supers, s_list) {
876                 if (sbi->s_snapshot_cno == cno &&
877                     (!snapshot_mount || nilfs_test_opt(sbi, SNAPSHOT))) {
878                                         /* exclude read-only mounts */
879                         ret++;
880                         break;
881                 }
882         }
883         /* for protecting recent checkpoints */
884         if (cno >= nilfs_last_cno(nilfs))
885                 ret++;
886
887  out_unlock:
888         up_read(&nilfs->ns_super_sem);
889         return ret;
890 }