nilfs2: sync super blocks in turns
[pandora-kernel.git] / fs / nilfs2 / super.c
1 /*
2  * super.c - NILFS module and super block management.
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  *  linux/fs/ext2/super.c
24  *
25  * Copyright (C) 1992, 1993, 1994, 1995
26  * Remy Card (card@masi.ibp.fr)
27  * Laboratoire MASI - Institut Blaise Pascal
28  * Universite Pierre et Marie Curie (Paris VI)
29  *
30  *  from
31  *
32  *  linux/fs/minix/inode.c
33  *
34  *  Copyright (C) 1991, 1992  Linus Torvalds
35  *
36  *  Big-endian to little-endian byte-swapping/bitmaps by
37  *        David S. Miller (davem@caip.rutgers.edu), 1995
38  */
39
40 #include <linux/module.h>
41 #include <linux/string.h>
42 #include <linux/slab.h>
43 #include <linux/init.h>
44 #include <linux/blkdev.h>
45 #include <linux/parser.h>
46 #include <linux/random.h>
47 #include <linux/crc32.h>
48 #include <linux/smp_lock.h>
49 #include <linux/vfs.h>
50 #include <linux/writeback.h>
51 #include <linux/kobject.h>
52 #include <linux/exportfs.h>
53 #include <linux/seq_file.h>
54 #include <linux/mount.h>
55 #include "nilfs.h"
56 #include "mdt.h"
57 #include "alloc.h"
58 #include "page.h"
59 #include "cpfile.h"
60 #include "ifile.h"
61 #include "dat.h"
62 #include "segment.h"
63 #include "segbuf.h"
64
65 MODULE_AUTHOR("NTT Corp.");
66 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
67                    "(NILFS)");
68 MODULE_LICENSE("GPL");
69
70 struct kmem_cache *nilfs_inode_cachep;
71 struct kmem_cache *nilfs_transaction_cachep;
72 struct kmem_cache *nilfs_segbuf_cachep;
73 struct kmem_cache *nilfs_btree_path_cache;
74
75 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
76
77 static void nilfs_set_error(struct nilfs_sb_info *sbi)
78 {
79         struct the_nilfs *nilfs = sbi->s_nilfs;
80         struct nilfs_super_block **sbp;
81
82         down_write(&nilfs->ns_sem);
83         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
84                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
85                 sbp = nilfs_prepare_super(sbi, 0);
86                 if (likely(sbp)) {
87                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
88                         if (sbp[1])
89                                 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
90                         nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
91                 }
92         }
93         up_write(&nilfs->ns_sem);
94 }
95
96 /**
97  * nilfs_error() - report failure condition on a filesystem
98  *
99  * nilfs_error() sets an ERROR_FS flag on the superblock as well as
100  * reporting an error message.  It should be called when NILFS detects
101  * incoherences or defects of meta data on disk.  As for sustainable
102  * errors such as a single-shot I/O error, nilfs_warning() or the printk()
103  * function should be used instead.
104  *
105  * The segment constructor must not call this function because it can
106  * kill itself.
107  */
108 void nilfs_error(struct super_block *sb, const char *function,
109                  const char *fmt, ...)
110 {
111         struct nilfs_sb_info *sbi = NILFS_SB(sb);
112         va_list args;
113
114         va_start(args, fmt);
115         printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
116         vprintk(fmt, args);
117         printk("\n");
118         va_end(args);
119
120         if (!(sb->s_flags & MS_RDONLY)) {
121                 nilfs_set_error(sbi);
122
123                 if (nilfs_test_opt(sbi, ERRORS_RO)) {
124                         printk(KERN_CRIT "Remounting filesystem read-only\n");
125                         sb->s_flags |= MS_RDONLY;
126                 }
127         }
128
129         if (nilfs_test_opt(sbi, ERRORS_PANIC))
130                 panic("NILFS (device %s): panic forced after error\n",
131                       sb->s_id);
132 }
133
134 void nilfs_warning(struct super_block *sb, const char *function,
135                    const char *fmt, ...)
136 {
137         va_list args;
138
139         va_start(args, fmt);
140         printk(KERN_WARNING "NILFS warning (device %s): %s: ",
141                sb->s_id, function);
142         vprintk(fmt, args);
143         printk("\n");
144         va_end(args);
145 }
146
147
148 struct inode *nilfs_alloc_inode_common(struct the_nilfs *nilfs)
149 {
150         struct nilfs_inode_info *ii;
151
152         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
153         if (!ii)
154                 return NULL;
155         ii->i_bh = NULL;
156         ii->i_state = 0;
157         ii->vfs_inode.i_version = 1;
158         nilfs_btnode_cache_init(&ii->i_btnode_cache, nilfs->ns_bdi);
159         return &ii->vfs_inode;
160 }
161
162 struct inode *nilfs_alloc_inode(struct super_block *sb)
163 {
164         return nilfs_alloc_inode_common(NILFS_SB(sb)->s_nilfs);
165 }
166
167 void nilfs_destroy_inode(struct inode *inode)
168 {
169         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
170 }
171
172 static void nilfs_clear_inode(struct inode *inode)
173 {
174         struct nilfs_inode_info *ii = NILFS_I(inode);
175
176         /*
177          * Free resources allocated in nilfs_read_inode(), here.
178          */
179         BUG_ON(!list_empty(&ii->i_dirty));
180         brelse(ii->i_bh);
181         ii->i_bh = NULL;
182
183         if (test_bit(NILFS_I_BMAP, &ii->i_state))
184                 nilfs_bmap_clear(ii->i_bmap);
185
186         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
187 }
188
189 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int flag)
190 {
191         struct the_nilfs *nilfs = sbi->s_nilfs;
192         int err;
193         int barrier_done = 0;
194
195         if (nilfs_test_opt(sbi, BARRIER)) {
196                 set_buffer_ordered(nilfs->ns_sbh[0]);
197                 barrier_done = 1;
198         }
199  retry:
200         set_buffer_dirty(nilfs->ns_sbh[0]);
201         err = sync_dirty_buffer(nilfs->ns_sbh[0]);
202         if (err == -EOPNOTSUPP && barrier_done) {
203                 nilfs_warning(sbi->s_super, __func__,
204                               "barrier-based sync failed. "
205                               "disabling barriers\n");
206                 nilfs_clear_opt(sbi, BARRIER);
207                 barrier_done = 0;
208                 clear_buffer_ordered(nilfs->ns_sbh[0]);
209                 goto retry;
210         }
211         if (unlikely(err)) {
212                 printk(KERN_ERR
213                        "NILFS: unable to write superblock (err=%d)\n", err);
214                 if (err == -EIO && nilfs->ns_sbh[1]) {
215                         /*
216                          * sbp[0] points to newer log than sbp[1],
217                          * so copy sbp[0] to sbp[1] to take over sbp[0].
218                          */
219                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
220                                nilfs->ns_sbsize);
221                         nilfs_fall_back_super_block(nilfs);
222                         goto retry;
223                 }
224         } else {
225                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
226
227                 nilfs->ns_sbwcount++;
228
229                 /*
230                  * The latest segment becomes trailable from the position
231                  * written in superblock.
232                  */
233                 clear_nilfs_discontinued(nilfs);
234
235                 /* update GC protection for recent segments */
236                 if (nilfs->ns_sbh[1]) {
237                         if (flag == NILFS_SB_COMMIT_ALL) {
238                                 set_buffer_dirty(nilfs->ns_sbh[1]);
239                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
240                                         goto out;
241                         }
242                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
243                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
244                                 sbp = nilfs->ns_sbp[1];
245                 }
246
247                 spin_lock(&nilfs->ns_last_segment_lock);
248                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
249                 spin_unlock(&nilfs->ns_last_segment_lock);
250         }
251  out:
252         return err;
253 }
254
255 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
256                           struct the_nilfs *nilfs)
257 {
258         sector_t nfreeblocks;
259
260         /* nilfs->ns_sem must be locked by the caller. */
261         nilfs_count_free_blocks(nilfs, &nfreeblocks);
262         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
263
264         spin_lock(&nilfs->ns_last_segment_lock);
265         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
266         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
267         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
268         spin_unlock(&nilfs->ns_last_segment_lock);
269 }
270
271 struct nilfs_super_block **nilfs_prepare_super(struct nilfs_sb_info *sbi,
272                                                int flip)
273 {
274         struct the_nilfs *nilfs = sbi->s_nilfs;
275         struct nilfs_super_block **sbp = nilfs->ns_sbp;
276
277         /* nilfs->ns_sem must be locked by the caller. */
278         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
279                 if (sbp[1] &&
280                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
281                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
282                 } else {
283                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
284                                sbi->s_super->s_id);
285                         return NULL;
286                 }
287         } else if (sbp[1] &&
288                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
289                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
290         }
291
292         if (flip && sbp[1])
293                 nilfs_swap_super_block(nilfs);
294
295         return sbp;
296 }
297
298 int nilfs_commit_super(struct nilfs_sb_info *sbi, int flag)
299 {
300         struct the_nilfs *nilfs = sbi->s_nilfs;
301         struct nilfs_super_block **sbp = nilfs->ns_sbp;
302         time_t t;
303
304         /* nilfs->ns_sem must be locked by the caller. */
305         t = get_seconds();
306         nilfs->ns_sbwtime = t;
307         sbp[0]->s_wtime = cpu_to_le64(t);
308         sbp[0]->s_sum = 0;
309         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
310                                              (unsigned char *)sbp[0],
311                                              nilfs->ns_sbsize));
312         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
313                 sbp[1]->s_wtime = sbp[0]->s_wtime;
314                 sbp[1]->s_sum = 0;
315                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
316                                             (unsigned char *)sbp[1],
317                                             nilfs->ns_sbsize));
318         }
319         clear_nilfs_sb_dirty(nilfs);
320         return nilfs_sync_super(sbi, flag);
321 }
322
323 /**
324  * nilfs_cleanup_super() - write filesystem state for cleanup
325  * @sbi: nilfs_sb_info to be unmounted or degraded to read-only
326  *
327  * This function restores state flags in the on-disk super block.
328  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
329  * filesystem was not clean previously.
330  */
331 int nilfs_cleanup_super(struct nilfs_sb_info *sbi)
332 {
333         struct nilfs_super_block **sbp;
334         int flag = NILFS_SB_COMMIT;
335         int ret = -EIO;
336
337         sbp = nilfs_prepare_super(sbi, 0);
338         if (sbp) {
339                 sbp[0]->s_state = cpu_to_le16(sbi->s_nilfs->ns_mount_state);
340                 nilfs_set_log_cursor(sbp[0], sbi->s_nilfs);
341                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
342                         /*
343                          * make the "clean" flag also to the opposite
344                          * super block if both super blocks point to
345                          * the same checkpoint.
346                          */
347                         sbp[1]->s_state = sbp[0]->s_state;
348                         flag = NILFS_SB_COMMIT_ALL;
349                 }
350                 ret = nilfs_commit_super(sbi, flag);
351         }
352         return ret;
353 }
354
355 static void nilfs_put_super(struct super_block *sb)
356 {
357         struct nilfs_sb_info *sbi = NILFS_SB(sb);
358         struct the_nilfs *nilfs = sbi->s_nilfs;
359
360         lock_kernel();
361
362         nilfs_detach_segment_constructor(sbi);
363
364         if (!(sb->s_flags & MS_RDONLY)) {
365                 down_write(&nilfs->ns_sem);
366                 nilfs_cleanup_super(sbi);
367                 up_write(&nilfs->ns_sem);
368         }
369         down_write(&nilfs->ns_super_sem);
370         if (nilfs->ns_current == sbi)
371                 nilfs->ns_current = NULL;
372         up_write(&nilfs->ns_super_sem);
373
374         nilfs_detach_checkpoint(sbi);
375         put_nilfs(sbi->s_nilfs);
376         sbi->s_super = NULL;
377         sb->s_fs_info = NULL;
378         nilfs_put_sbinfo(sbi);
379
380         unlock_kernel();
381 }
382
383 static int nilfs_sync_fs(struct super_block *sb, int wait)
384 {
385         struct nilfs_sb_info *sbi = NILFS_SB(sb);
386         struct the_nilfs *nilfs = sbi->s_nilfs;
387         struct nilfs_super_block **sbp;
388         int err = 0;
389
390         /* This function is called when super block should be written back */
391         if (wait)
392                 err = nilfs_construct_segment(sb);
393
394         down_write(&nilfs->ns_sem);
395         if (nilfs_sb_dirty(nilfs)) {
396                 sbp = nilfs_prepare_super(sbi, nilfs_sb_will_flip(nilfs));
397                 if (likely(sbp)) {
398                         nilfs_set_log_cursor(sbp[0], nilfs);
399                         nilfs_commit_super(sbi, NILFS_SB_COMMIT);
400                 }
401         }
402         up_write(&nilfs->ns_sem);
403
404         return err;
405 }
406
407 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
408 {
409         struct the_nilfs *nilfs = sbi->s_nilfs;
410         struct nilfs_checkpoint *raw_cp;
411         struct buffer_head *bh_cp;
412         int err;
413
414         down_write(&nilfs->ns_super_sem);
415         list_add(&sbi->s_list, &nilfs->ns_supers);
416         up_write(&nilfs->ns_super_sem);
417
418         sbi->s_ifile = nilfs_ifile_new(sbi, nilfs->ns_inode_size);
419         if (!sbi->s_ifile)
420                 return -ENOMEM;
421
422         down_read(&nilfs->ns_segctor_sem);
423         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
424                                           &bh_cp);
425         up_read(&nilfs->ns_segctor_sem);
426         if (unlikely(err)) {
427                 if (err == -ENOENT || err == -EINVAL) {
428                         printk(KERN_ERR
429                                "NILFS: Invalid checkpoint "
430                                "(checkpoint number=%llu)\n",
431                                (unsigned long long)cno);
432                         err = -EINVAL;
433                 }
434                 goto failed;
435         }
436         err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
437         if (unlikely(err))
438                 goto failed_bh;
439         atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
440         atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
441
442         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
443         return 0;
444
445  failed_bh:
446         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
447  failed:
448         nilfs_mdt_destroy(sbi->s_ifile);
449         sbi->s_ifile = NULL;
450
451         down_write(&nilfs->ns_super_sem);
452         list_del_init(&sbi->s_list);
453         up_write(&nilfs->ns_super_sem);
454
455         return err;
456 }
457
458 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
459 {
460         struct the_nilfs *nilfs = sbi->s_nilfs;
461
462         nilfs_mdt_destroy(sbi->s_ifile);
463         sbi->s_ifile = NULL;
464         down_write(&nilfs->ns_super_sem);
465         list_del_init(&sbi->s_list);
466         up_write(&nilfs->ns_super_sem);
467 }
468
469 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
470 {
471         struct super_block *sb = dentry->d_sb;
472         struct nilfs_sb_info *sbi = NILFS_SB(sb);
473         struct the_nilfs *nilfs = sbi->s_nilfs;
474         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
475         unsigned long long blocks;
476         unsigned long overhead;
477         unsigned long nrsvblocks;
478         sector_t nfreeblocks;
479         int err;
480
481         /*
482          * Compute all of the segment blocks
483          *
484          * The blocks before first segment and after last segment
485          * are excluded.
486          */
487         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
488                 - nilfs->ns_first_data_block;
489         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
490
491         /*
492          * Compute the overhead
493          *
494          * When distributing meta data blocks outside segment structure,
495          * We must count them as the overhead.
496          */
497         overhead = 0;
498
499         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
500         if (unlikely(err))
501                 return err;
502
503         buf->f_type = NILFS_SUPER_MAGIC;
504         buf->f_bsize = sb->s_blocksize;
505         buf->f_blocks = blocks - overhead;
506         buf->f_bfree = nfreeblocks;
507         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
508                 (buf->f_bfree - nrsvblocks) : 0;
509         buf->f_files = atomic_read(&sbi->s_inodes_count);
510         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
511         buf->f_namelen = NILFS_NAME_LEN;
512         buf->f_fsid.val[0] = (u32)id;
513         buf->f_fsid.val[1] = (u32)(id >> 32);
514
515         return 0;
516 }
517
518 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
519 {
520         struct super_block *sb = vfs->mnt_sb;
521         struct nilfs_sb_info *sbi = NILFS_SB(sb);
522
523         if (!nilfs_test_opt(sbi, BARRIER))
524                 seq_printf(seq, ",nobarrier");
525         if (nilfs_test_opt(sbi, SNAPSHOT))
526                 seq_printf(seq, ",cp=%llu",
527                            (unsigned long long int)sbi->s_snapshot_cno);
528         if (nilfs_test_opt(sbi, ERRORS_PANIC))
529                 seq_printf(seq, ",errors=panic");
530         if (nilfs_test_opt(sbi, ERRORS_CONT))
531                 seq_printf(seq, ",errors=continue");
532         if (nilfs_test_opt(sbi, STRICT_ORDER))
533                 seq_printf(seq, ",order=strict");
534         if (nilfs_test_opt(sbi, NORECOVERY))
535                 seq_printf(seq, ",norecovery");
536         if (nilfs_test_opt(sbi, DISCARD))
537                 seq_printf(seq, ",discard");
538
539         return 0;
540 }
541
542 static const struct super_operations nilfs_sops = {
543         .alloc_inode    = nilfs_alloc_inode,
544         .destroy_inode  = nilfs_destroy_inode,
545         .dirty_inode    = nilfs_dirty_inode,
546         /* .write_inode    = nilfs_write_inode, */
547         /* .put_inode      = nilfs_put_inode, */
548         /* .drop_inode    = nilfs_drop_inode, */
549         .delete_inode   = nilfs_delete_inode,
550         .put_super      = nilfs_put_super,
551         /* .write_super    = nilfs_write_super, */
552         .sync_fs        = nilfs_sync_fs,
553         /* .write_super_lockfs */
554         /* .unlockfs */
555         .statfs         = nilfs_statfs,
556         .remount_fs     = nilfs_remount,
557         .clear_inode    = nilfs_clear_inode,
558         /* .umount_begin */
559         .show_options = nilfs_show_options
560 };
561
562 static struct inode *
563 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
564 {
565         struct inode *inode;
566
567         if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
568             ino != NILFS_SKETCH_INO)
569                 return ERR_PTR(-ESTALE);
570
571         inode = nilfs_iget(sb, ino);
572         if (IS_ERR(inode))
573                 return ERR_CAST(inode);
574         if (generation && inode->i_generation != generation) {
575                 iput(inode);
576                 return ERR_PTR(-ESTALE);
577         }
578
579         return inode;
580 }
581
582 static struct dentry *
583 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
584                    int fh_type)
585 {
586         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
587                                     nilfs_nfs_get_inode);
588 }
589
590 static struct dentry *
591 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
592                    int fh_type)
593 {
594         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
595                                     nilfs_nfs_get_inode);
596 }
597
598 static const struct export_operations nilfs_export_ops = {
599         .fh_to_dentry = nilfs_fh_to_dentry,
600         .fh_to_parent = nilfs_fh_to_parent,
601         .get_parent = nilfs_get_parent,
602 };
603
604 enum {
605         Opt_err_cont, Opt_err_panic, Opt_err_ro,
606         Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
607         Opt_discard, Opt_err,
608 };
609
610 static match_table_t tokens = {
611         {Opt_err_cont, "errors=continue"},
612         {Opt_err_panic, "errors=panic"},
613         {Opt_err_ro, "errors=remount-ro"},
614         {Opt_nobarrier, "nobarrier"},
615         {Opt_snapshot, "cp=%u"},
616         {Opt_order, "order=%s"},
617         {Opt_norecovery, "norecovery"},
618         {Opt_discard, "discard"},
619         {Opt_err, NULL}
620 };
621
622 static int parse_options(char *options, struct super_block *sb)
623 {
624         struct nilfs_sb_info *sbi = NILFS_SB(sb);
625         char *p;
626         substring_t args[MAX_OPT_ARGS];
627         int option;
628
629         if (!options)
630                 return 1;
631
632         while ((p = strsep(&options, ",")) != NULL) {
633                 int token;
634                 if (!*p)
635                         continue;
636
637                 token = match_token(p, tokens, args);
638                 switch (token) {
639                 case Opt_nobarrier:
640                         nilfs_clear_opt(sbi, BARRIER);
641                         break;
642                 case Opt_order:
643                         if (strcmp(args[0].from, "relaxed") == 0)
644                                 /* Ordered data semantics */
645                                 nilfs_clear_opt(sbi, STRICT_ORDER);
646                         else if (strcmp(args[0].from, "strict") == 0)
647                                 /* Strict in-order semantics */
648                                 nilfs_set_opt(sbi, STRICT_ORDER);
649                         else
650                                 return 0;
651                         break;
652                 case Opt_err_panic:
653                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
654                         break;
655                 case Opt_err_ro:
656                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
657                         break;
658                 case Opt_err_cont:
659                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
660                         break;
661                 case Opt_snapshot:
662                         if (match_int(&args[0], &option) || option <= 0)
663                                 return 0;
664                         if (!(sb->s_flags & MS_RDONLY))
665                                 return 0;
666                         sbi->s_snapshot_cno = option;
667                         nilfs_set_opt(sbi, SNAPSHOT);
668                         break;
669                 case Opt_norecovery:
670                         nilfs_set_opt(sbi, NORECOVERY);
671                         break;
672                 case Opt_discard:
673                         nilfs_set_opt(sbi, DISCARD);
674                         break;
675                 default:
676                         printk(KERN_ERR
677                                "NILFS: Unrecognized mount option \"%s\"\n", p);
678                         return 0;
679                 }
680         }
681         return 1;
682 }
683
684 static inline void
685 nilfs_set_default_options(struct nilfs_sb_info *sbi,
686                           struct nilfs_super_block *sbp)
687 {
688         sbi->s_mount_opt =
689                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
690 }
691
692 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
693 {
694         struct the_nilfs *nilfs = sbi->s_nilfs;
695         struct nilfs_super_block **sbp;
696         int max_mnt_count;
697         int mnt_count;
698
699         /* nilfs->ns_sem must be locked by the caller. */
700         sbp = nilfs_prepare_super(sbi, 0);
701         if (!sbp)
702                 return -EIO;
703
704         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
705         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
706
707         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
708                 printk(KERN_WARNING
709                        "NILFS warning: mounting fs with errors\n");
710 #if 0
711         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
712                 printk(KERN_WARNING
713                        "NILFS warning: maximal mount count reached\n");
714 #endif
715         }
716         if (!max_mnt_count)
717                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
718
719         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
720         sbp[0]->s_state =
721                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
722         sbp[0]->s_mtime = cpu_to_le64(get_seconds());
723         /* synchronize sbp[1] with sbp[0] */
724         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
725         return nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
726 }
727
728 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
729                                                  u64 pos, int blocksize,
730                                                  struct buffer_head **pbh)
731 {
732         unsigned long long sb_index = pos;
733         unsigned long offset;
734
735         offset = do_div(sb_index, blocksize);
736         *pbh = sb_bread(sb, sb_index);
737         if (!*pbh)
738                 return NULL;
739         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
740 }
741
742 int nilfs_store_magic_and_option(struct super_block *sb,
743                                  struct nilfs_super_block *sbp,
744                                  char *data)
745 {
746         struct nilfs_sb_info *sbi = NILFS_SB(sb);
747
748         sb->s_magic = le16_to_cpu(sbp->s_magic);
749
750         /* FS independent flags */
751 #ifdef NILFS_ATIME_DISABLE
752         sb->s_flags |= MS_NOATIME;
753 #endif
754
755         nilfs_set_default_options(sbi, sbp);
756
757         sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
758         sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
759         sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
760         sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
761
762         return !parse_options(data, sb) ? -EINVAL : 0 ;
763 }
764
765 /**
766  * nilfs_fill_super() - initialize a super block instance
767  * @sb: super_block
768  * @data: mount options
769  * @silent: silent mode flag
770  * @nilfs: the_nilfs struct
771  *
772  * This function is called exclusively by nilfs->ns_mount_mutex.
773  * So, the recovery process is protected from other simultaneous mounts.
774  */
775 static int
776 nilfs_fill_super(struct super_block *sb, void *data, int silent,
777                  struct the_nilfs *nilfs)
778 {
779         struct nilfs_sb_info *sbi;
780         struct inode *root;
781         __u64 cno;
782         int err;
783
784         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
785         if (!sbi)
786                 return -ENOMEM;
787
788         sb->s_fs_info = sbi;
789
790         get_nilfs(nilfs);
791         sbi->s_nilfs = nilfs;
792         sbi->s_super = sb;
793         atomic_set(&sbi->s_count, 1);
794
795         err = init_nilfs(nilfs, sbi, (char *)data);
796         if (err)
797                 goto failed_sbi;
798
799         spin_lock_init(&sbi->s_inode_lock);
800         INIT_LIST_HEAD(&sbi->s_dirty_files);
801         INIT_LIST_HEAD(&sbi->s_list);
802
803         /*
804          * Following initialization is overlapped because
805          * nilfs_sb_info structure has been cleared at the beginning.
806          * But we reserve them to keep our interest and make ready
807          * for the future change.
808          */
809         get_random_bytes(&sbi->s_next_generation,
810                          sizeof(sbi->s_next_generation));
811         spin_lock_init(&sbi->s_next_gen_lock);
812
813         sb->s_op = &nilfs_sops;
814         sb->s_export_op = &nilfs_export_ops;
815         sb->s_root = NULL;
816         sb->s_time_gran = 1;
817         sb->s_bdi = nilfs->ns_bdi;
818
819         err = load_nilfs(nilfs, sbi);
820         if (err)
821                 goto failed_sbi;
822
823         cno = nilfs_last_cno(nilfs);
824
825         if (sb->s_flags & MS_RDONLY) {
826                 if (nilfs_test_opt(sbi, SNAPSHOT)) {
827                         down_read(&nilfs->ns_segctor_sem);
828                         err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
829                                                        sbi->s_snapshot_cno);
830                         up_read(&nilfs->ns_segctor_sem);
831                         if (err < 0) {
832                                 if (err == -ENOENT)
833                                         err = -EINVAL;
834                                 goto failed_sbi;
835                         }
836                         if (!err) {
837                                 printk(KERN_ERR
838                                        "NILFS: The specified checkpoint is "
839                                        "not a snapshot "
840                                        "(checkpoint number=%llu).\n",
841                                        (unsigned long long)sbi->s_snapshot_cno);
842                                 err = -EINVAL;
843                                 goto failed_sbi;
844                         }
845                         cno = sbi->s_snapshot_cno;
846                 }
847         }
848
849         err = nilfs_attach_checkpoint(sbi, cno);
850         if (err) {
851                 printk(KERN_ERR "NILFS: error loading a checkpoint"
852                        " (checkpoint number=%llu).\n", (unsigned long long)cno);
853                 goto failed_sbi;
854         }
855
856         if (!(sb->s_flags & MS_RDONLY)) {
857                 err = nilfs_attach_segment_constructor(sbi);
858                 if (err)
859                         goto failed_checkpoint;
860         }
861
862         root = nilfs_iget(sb, NILFS_ROOT_INO);
863         if (IS_ERR(root)) {
864                 printk(KERN_ERR "NILFS: get root inode failed\n");
865                 err = PTR_ERR(root);
866                 goto failed_segctor;
867         }
868         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
869                 iput(root);
870                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
871                 err = -EINVAL;
872                 goto failed_segctor;
873         }
874         sb->s_root = d_alloc_root(root);
875         if (!sb->s_root) {
876                 iput(root);
877                 printk(KERN_ERR "NILFS: get root dentry failed\n");
878                 err = -ENOMEM;
879                 goto failed_segctor;
880         }
881
882         if (!(sb->s_flags & MS_RDONLY)) {
883                 down_write(&nilfs->ns_sem);
884                 nilfs_setup_super(sbi);
885                 up_write(&nilfs->ns_sem);
886         }
887
888         down_write(&nilfs->ns_super_sem);
889         if (!nilfs_test_opt(sbi, SNAPSHOT))
890                 nilfs->ns_current = sbi;
891         up_write(&nilfs->ns_super_sem);
892
893         return 0;
894
895  failed_segctor:
896         nilfs_detach_segment_constructor(sbi);
897
898  failed_checkpoint:
899         nilfs_detach_checkpoint(sbi);
900
901  failed_sbi:
902         put_nilfs(nilfs);
903         sb->s_fs_info = NULL;
904         nilfs_put_sbinfo(sbi);
905         return err;
906 }
907
908 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
909 {
910         struct nilfs_sb_info *sbi = NILFS_SB(sb);
911         struct the_nilfs *nilfs = sbi->s_nilfs;
912         unsigned long old_sb_flags;
913         struct nilfs_mount_options old_opts;
914         int was_snapshot, err;
915
916         lock_kernel();
917
918         down_write(&nilfs->ns_super_sem);
919         old_sb_flags = sb->s_flags;
920         old_opts.mount_opt = sbi->s_mount_opt;
921         old_opts.snapshot_cno = sbi->s_snapshot_cno;
922         was_snapshot = nilfs_test_opt(sbi, SNAPSHOT);
923
924         if (!parse_options(data, sb)) {
925                 err = -EINVAL;
926                 goto restore_opts;
927         }
928         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
929
930         err = -EINVAL;
931         if (was_snapshot) {
932                 if (!(*flags & MS_RDONLY)) {
933                         printk(KERN_ERR "NILFS (device %s): cannot remount "
934                                "snapshot read/write.\n",
935                                sb->s_id);
936                         goto restore_opts;
937                 } else if (sbi->s_snapshot_cno != old_opts.snapshot_cno) {
938                         printk(KERN_ERR "NILFS (device %s): cannot "
939                                "remount to a different snapshot.\n",
940                                sb->s_id);
941                         goto restore_opts;
942                 }
943         } else {
944                 if (nilfs_test_opt(sbi, SNAPSHOT)) {
945                         printk(KERN_ERR "NILFS (device %s): cannot change "
946                                "a regular mount to a snapshot.\n",
947                                sb->s_id);
948                         goto restore_opts;
949                 }
950         }
951
952         if (!nilfs_valid_fs(nilfs)) {
953                 printk(KERN_WARNING "NILFS (device %s): couldn't "
954                        "remount because the filesystem is in an "
955                        "incomplete recovery state.\n", sb->s_id);
956                 goto restore_opts;
957         }
958
959         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
960                 goto out;
961         if (*flags & MS_RDONLY) {
962                 /* Shutting down the segment constructor */
963                 nilfs_detach_segment_constructor(sbi);
964                 sb->s_flags |= MS_RDONLY;
965
966                 /*
967                  * Remounting a valid RW partition RDONLY, so set
968                  * the RDONLY flag and then mark the partition as valid again.
969                  */
970                 down_write(&nilfs->ns_sem);
971                 nilfs_cleanup_super(sbi);
972                 up_write(&nilfs->ns_sem);
973         } else {
974                 /*
975                  * Mounting a RDONLY partition read-write, so reread and
976                  * store the current valid flag.  (It may have been changed
977                  * by fsck since we originally mounted the partition.)
978                  */
979                 sb->s_flags &= ~MS_RDONLY;
980
981                 err = nilfs_attach_segment_constructor(sbi);
982                 if (err)
983                         goto restore_opts;
984
985                 down_write(&nilfs->ns_sem);
986                 nilfs_setup_super(sbi);
987                 up_write(&nilfs->ns_sem);
988         }
989  out:
990         up_write(&nilfs->ns_super_sem);
991         unlock_kernel();
992         return 0;
993
994  restore_opts:
995         sb->s_flags = old_sb_flags;
996         sbi->s_mount_opt = old_opts.mount_opt;
997         sbi->s_snapshot_cno = old_opts.snapshot_cno;
998         up_write(&nilfs->ns_super_sem);
999         unlock_kernel();
1000         return err;
1001 }
1002
1003 struct nilfs_super_data {
1004         struct block_device *bdev;
1005         struct nilfs_sb_info *sbi;
1006         __u64 cno;
1007         int flags;
1008 };
1009
1010 /**
1011  * nilfs_identify - pre-read mount options needed to identify mount instance
1012  * @data: mount options
1013  * @sd: nilfs_super_data
1014  */
1015 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1016 {
1017         char *p, *options = data;
1018         substring_t args[MAX_OPT_ARGS];
1019         int option, token;
1020         int ret = 0;
1021
1022         do {
1023                 p = strsep(&options, ",");
1024                 if (p != NULL && *p) {
1025                         token = match_token(p, tokens, args);
1026                         if (token == Opt_snapshot) {
1027                                 if (!(sd->flags & MS_RDONLY))
1028                                         ret++;
1029                                 else {
1030                                         ret = match_int(&args[0], &option);
1031                                         if (!ret) {
1032                                                 if (option > 0)
1033                                                         sd->cno = option;
1034                                                 else
1035                                                         ret++;
1036                                         }
1037                                 }
1038                         }
1039                         if (ret)
1040                                 printk(KERN_ERR
1041                                        "NILFS: invalid mount option: %s\n", p);
1042                 }
1043                 if (!options)
1044                         break;
1045                 BUG_ON(options == data);
1046                 *(options - 1) = ',';
1047         } while (!ret);
1048         return ret;
1049 }
1050
1051 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1052 {
1053         struct nilfs_super_data *sd = data;
1054
1055         s->s_bdev = sd->bdev;
1056         s->s_dev = s->s_bdev->bd_dev;
1057         return 0;
1058 }
1059
1060 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1061 {
1062         struct nilfs_super_data *sd = data;
1063
1064         return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1065 }
1066
1067 static int
1068 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1069              const char *dev_name, void *data, struct vfsmount *mnt)
1070 {
1071         struct nilfs_super_data sd;
1072         struct super_block *s;
1073         fmode_t mode = FMODE_READ;
1074         struct the_nilfs *nilfs;
1075         int err, need_to_close = 1;
1076
1077         if (!(flags & MS_RDONLY))
1078                 mode |= FMODE_WRITE;
1079
1080         sd.bdev = open_bdev_exclusive(dev_name, mode, fs_type);
1081         if (IS_ERR(sd.bdev))
1082                 return PTR_ERR(sd.bdev);
1083
1084         /*
1085          * To get mount instance using sget() vfs-routine, NILFS needs
1086          * much more information than normal filesystems to identify mount
1087          * instance.  For snapshot mounts, not only a mount type (ro-mount
1088          * or rw-mount) but also a checkpoint number is required.
1089          */
1090         sd.cno = 0;
1091         sd.flags = flags;
1092         if (nilfs_identify((char *)data, &sd)) {
1093                 err = -EINVAL;
1094                 goto failed;
1095         }
1096
1097         nilfs = find_or_create_nilfs(sd.bdev);
1098         if (!nilfs) {
1099                 err = -ENOMEM;
1100                 goto failed;
1101         }
1102
1103         mutex_lock(&nilfs->ns_mount_mutex);
1104
1105         if (!sd.cno) {
1106                 /*
1107                  * Check if an exclusive mount exists or not.
1108                  * Snapshot mounts coexist with a current mount
1109                  * (i.e. rw-mount or ro-mount), whereas rw-mount and
1110                  * ro-mount are mutually exclusive.
1111                  */
1112                 down_read(&nilfs->ns_super_sem);
1113                 if (nilfs->ns_current &&
1114                     ((nilfs->ns_current->s_super->s_flags ^ flags)
1115                      & MS_RDONLY)) {
1116                         up_read(&nilfs->ns_super_sem);
1117                         err = -EBUSY;
1118                         goto failed_unlock;
1119                 }
1120                 up_read(&nilfs->ns_super_sem);
1121         }
1122
1123         /*
1124          * Find existing nilfs_sb_info struct
1125          */
1126         sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1127
1128         /*
1129          * Get super block instance holding the nilfs_sb_info struct.
1130          * A new instance is allocated if no existing mount is present or
1131          * existing instance has been unmounted.
1132          */
1133         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1134         if (sd.sbi)
1135                 nilfs_put_sbinfo(sd.sbi);
1136
1137         if (IS_ERR(s)) {
1138                 err = PTR_ERR(s);
1139                 goto failed_unlock;
1140         }
1141
1142         if (!s->s_root) {
1143                 char b[BDEVNAME_SIZE];
1144
1145                 /* New superblock instance created */
1146                 s->s_flags = flags;
1147                 s->s_mode = mode;
1148                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1149                 sb_set_blocksize(s, block_size(sd.bdev));
1150
1151                 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0,
1152                                        nilfs);
1153                 if (err)
1154                         goto cancel_new;
1155
1156                 s->s_flags |= MS_ACTIVE;
1157                 need_to_close = 0;
1158         }
1159
1160         mutex_unlock(&nilfs->ns_mount_mutex);
1161         put_nilfs(nilfs);
1162         if (need_to_close)
1163                 close_bdev_exclusive(sd.bdev, mode);
1164         simple_set_mnt(mnt, s);
1165         return 0;
1166
1167  failed_unlock:
1168         mutex_unlock(&nilfs->ns_mount_mutex);
1169         put_nilfs(nilfs);
1170  failed:
1171         close_bdev_exclusive(sd.bdev, mode);
1172
1173         return err;
1174
1175  cancel_new:
1176         /* Abandoning the newly allocated superblock */
1177         mutex_unlock(&nilfs->ns_mount_mutex);
1178         put_nilfs(nilfs);
1179         deactivate_locked_super(s);
1180         /*
1181          * deactivate_locked_super() invokes close_bdev_exclusive().
1182          * We must finish all post-cleaning before this call;
1183          * put_nilfs() needs the block device.
1184          */
1185         return err;
1186 }
1187
1188 struct file_system_type nilfs_fs_type = {
1189         .owner    = THIS_MODULE,
1190         .name     = "nilfs2",
1191         .get_sb   = nilfs_get_sb,
1192         .kill_sb  = kill_block_super,
1193         .fs_flags = FS_REQUIRES_DEV,
1194 };
1195
1196 static void nilfs_inode_init_once(void *obj)
1197 {
1198         struct nilfs_inode_info *ii = obj;
1199
1200         INIT_LIST_HEAD(&ii->i_dirty);
1201 #ifdef CONFIG_NILFS_XATTR
1202         init_rwsem(&ii->xattr_sem);
1203 #endif
1204         nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
1205         ii->i_bmap = (struct nilfs_bmap *)&ii->i_bmap_union;
1206         inode_init_once(&ii->vfs_inode);
1207 }
1208
1209 static void nilfs_segbuf_init_once(void *obj)
1210 {
1211         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1212 }
1213
1214 static void nilfs_destroy_cachep(void)
1215 {
1216         if (nilfs_inode_cachep)
1217                 kmem_cache_destroy(nilfs_inode_cachep);
1218         if (nilfs_transaction_cachep)
1219                 kmem_cache_destroy(nilfs_transaction_cachep);
1220         if (nilfs_segbuf_cachep)
1221                 kmem_cache_destroy(nilfs_segbuf_cachep);
1222         if (nilfs_btree_path_cache)
1223                 kmem_cache_destroy(nilfs_btree_path_cache);
1224 }
1225
1226 static int __init nilfs_init_cachep(void)
1227 {
1228         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1229                         sizeof(struct nilfs_inode_info), 0,
1230                         SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1231         if (!nilfs_inode_cachep)
1232                 goto fail;
1233
1234         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1235                         sizeof(struct nilfs_transaction_info), 0,
1236                         SLAB_RECLAIM_ACCOUNT, NULL);
1237         if (!nilfs_transaction_cachep)
1238                 goto fail;
1239
1240         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1241                         sizeof(struct nilfs_segment_buffer), 0,
1242                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1243         if (!nilfs_segbuf_cachep)
1244                 goto fail;
1245
1246         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1247                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1248                         0, 0, NULL);
1249         if (!nilfs_btree_path_cache)
1250                 goto fail;
1251
1252         return 0;
1253
1254 fail:
1255         nilfs_destroy_cachep();
1256         return -ENOMEM;
1257 }
1258
1259 static int __init init_nilfs_fs(void)
1260 {
1261         int err;
1262
1263         err = nilfs_init_cachep();
1264         if (err)
1265                 goto fail;
1266
1267         err = register_filesystem(&nilfs_fs_type);
1268         if (err)
1269                 goto free_cachep;
1270
1271         printk(KERN_INFO "NILFS version 2 loaded\n");
1272         return 0;
1273
1274 free_cachep:
1275         nilfs_destroy_cachep();
1276 fail:
1277         return err;
1278 }
1279
1280 static void __exit exit_nilfs_fs(void)
1281 {
1282         nilfs_destroy_cachep();
1283         unregister_filesystem(&nilfs_fs_type);
1284 }
1285
1286 module_init(init_nilfs_fs)
1287 module_exit(exit_nilfs_fs)