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