Merge commit 'origin/master' into next
[linux-2.6] / 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 "nilfs.h"
54 #include "mdt.h"
55 #include "alloc.h"
56 #include "page.h"
57 #include "cpfile.h"
58 #include "ifile.h"
59 #include "dat.h"
60 #include "segment.h"
61 #include "segbuf.h"
62
63 MODULE_AUTHOR("NTT Corp.");
64 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
65                    "(NILFS)");
66 MODULE_LICENSE("GPL");
67
68 static void nilfs_write_super(struct super_block *sb);
69 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
70
71 /**
72  * nilfs_error() - report failure condition on a filesystem
73  *
74  * nilfs_error() sets an ERROR_FS flag on the superblock as well as
75  * reporting an error message.  It should be called when NILFS detects
76  * incoherences or defects of meta data on disk.  As for sustainable
77  * errors such as a single-shot I/O error, nilfs_warning() or the printk()
78  * function should be used instead.
79  *
80  * The segment constructor must not call this function because it can
81  * kill itself.
82  */
83 void nilfs_error(struct super_block *sb, const char *function,
84                  const char *fmt, ...)
85 {
86         struct nilfs_sb_info *sbi = NILFS_SB(sb);
87         va_list args;
88
89         va_start(args, fmt);
90         printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
91         vprintk(fmt, args);
92         printk("\n");
93         va_end(args);
94
95         if (!(sb->s_flags & MS_RDONLY)) {
96                 struct the_nilfs *nilfs = sbi->s_nilfs;
97
98                 if (!nilfs_test_opt(sbi, ERRORS_CONT))
99                         nilfs_detach_segment_constructor(sbi);
100
101                 down_write(&nilfs->ns_sem);
102                 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
103                         nilfs->ns_mount_state |= NILFS_ERROR_FS;
104                         nilfs->ns_sbp[0]->s_state |=
105                                 cpu_to_le16(NILFS_ERROR_FS);
106                         nilfs_commit_super(sbi, 1);
107                 }
108                 up_write(&nilfs->ns_sem);
109
110                 if (nilfs_test_opt(sbi, ERRORS_RO)) {
111                         printk(KERN_CRIT "Remounting filesystem read-only\n");
112                         sb->s_flags |= MS_RDONLY;
113                 }
114         }
115
116         if (nilfs_test_opt(sbi, ERRORS_PANIC))
117                 panic("NILFS (device %s): panic forced after error\n",
118                       sb->s_id);
119 }
120
121 void nilfs_warning(struct super_block *sb, const char *function,
122                    const char *fmt, ...)
123 {
124         va_list args;
125
126         va_start(args, fmt);
127         printk(KERN_WARNING "NILFS warning (device %s): %s: ",
128                sb->s_id, function);
129         vprintk(fmt, args);
130         printk("\n");
131         va_end(args);
132 }
133
134 static struct kmem_cache *nilfs_inode_cachep;
135
136 struct inode *nilfs_alloc_inode(struct super_block *sb)
137 {
138         struct nilfs_inode_info *ii;
139
140         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
141         if (!ii)
142                 return NULL;
143         ii->i_bh = NULL;
144         ii->i_state = 0;
145         ii->vfs_inode.i_version = 1;
146         nilfs_btnode_cache_init(&ii->i_btnode_cache);
147         return &ii->vfs_inode;
148 }
149
150 void nilfs_destroy_inode(struct inode *inode)
151 {
152         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
153 }
154
155 static void init_once(void *obj)
156 {
157         struct nilfs_inode_info *ii = obj;
158
159         INIT_LIST_HEAD(&ii->i_dirty);
160 #ifdef CONFIG_NILFS_XATTR
161         init_rwsem(&ii->xattr_sem);
162 #endif
163         nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
164         ii->i_bmap = (struct nilfs_bmap *)&ii->i_bmap_union;
165         inode_init_once(&ii->vfs_inode);
166 }
167
168 static int nilfs_init_inode_cache(void)
169 {
170         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
171                                                sizeof(struct nilfs_inode_info),
172                                                0, SLAB_RECLAIM_ACCOUNT,
173                                                init_once);
174
175         return (nilfs_inode_cachep == NULL) ? -ENOMEM : 0;
176 }
177
178 static inline void nilfs_destroy_inode_cache(void)
179 {
180         kmem_cache_destroy(nilfs_inode_cachep);
181 }
182
183 static void nilfs_clear_inode(struct inode *inode)
184 {
185         struct nilfs_inode_info *ii = NILFS_I(inode);
186
187 #ifdef CONFIG_NILFS_POSIX_ACL
188         if (ii->i_acl && ii->i_acl != NILFS_ACL_NOT_CACHED) {
189                 posix_acl_release(ii->i_acl);
190                 ii->i_acl = NILFS_ACL_NOT_CACHED;
191         }
192         if (ii->i_default_acl && ii->i_default_acl != NILFS_ACL_NOT_CACHED) {
193                 posix_acl_release(ii->i_default_acl);
194                 ii->i_default_acl = NILFS_ACL_NOT_CACHED;
195         }
196 #endif
197         /*
198          * Free resources allocated in nilfs_read_inode(), here.
199          */
200         BUG_ON(!list_empty(&ii->i_dirty));
201         brelse(ii->i_bh);
202         ii->i_bh = NULL;
203
204         if (test_bit(NILFS_I_BMAP, &ii->i_state))
205                 nilfs_bmap_clear(ii->i_bmap);
206
207         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
208 }
209
210 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int dupsb)
211 {
212         struct the_nilfs *nilfs = sbi->s_nilfs;
213         int err;
214         int barrier_done = 0;
215
216         if (nilfs_test_opt(sbi, BARRIER)) {
217                 set_buffer_ordered(nilfs->ns_sbh[0]);
218                 barrier_done = 1;
219         }
220  retry:
221         set_buffer_dirty(nilfs->ns_sbh[0]);
222         err = sync_dirty_buffer(nilfs->ns_sbh[0]);
223         if (err == -EOPNOTSUPP && barrier_done) {
224                 nilfs_warning(sbi->s_super, __func__,
225                               "barrier-based sync failed. "
226                               "disabling barriers\n");
227                 nilfs_clear_opt(sbi, BARRIER);
228                 barrier_done = 0;
229                 clear_buffer_ordered(nilfs->ns_sbh[0]);
230                 goto retry;
231         }
232         if (unlikely(err)) {
233                 printk(KERN_ERR
234                        "NILFS: unable to write superblock (err=%d)\n", err);
235                 if (err == -EIO && nilfs->ns_sbh[1]) {
236                         nilfs_fall_back_super_block(nilfs);
237                         goto retry;
238                 }
239         } else {
240                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
241
242                 /*
243                  * The latest segment becomes trailable from the position
244                  * written in superblock.
245                  */
246                 clear_nilfs_discontinued(nilfs);
247
248                 /* update GC protection for recent segments */
249                 if (nilfs->ns_sbh[1]) {
250                         sbp = NULL;
251                         if (dupsb) {
252                                 set_buffer_dirty(nilfs->ns_sbh[1]);
253                                 if (!sync_dirty_buffer(nilfs->ns_sbh[1]))
254                                         sbp = nilfs->ns_sbp[1];
255                         }
256                 }
257                 if (sbp) {
258                         spin_lock(&nilfs->ns_last_segment_lock);
259                         nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
260                         spin_unlock(&nilfs->ns_last_segment_lock);
261                 }
262         }
263
264         return err;
265 }
266
267 int nilfs_commit_super(struct nilfs_sb_info *sbi, int dupsb)
268 {
269         struct the_nilfs *nilfs = sbi->s_nilfs;
270         struct nilfs_super_block **sbp = nilfs->ns_sbp;
271         sector_t nfreeblocks;
272         time_t t;
273         int err;
274
275         /* nilfs->sem must be locked by the caller. */
276         if (sbp[0]->s_magic != NILFS_SUPER_MAGIC) {
277                 if (sbp[1] && sbp[1]->s_magic == NILFS_SUPER_MAGIC)
278                         nilfs_swap_super_block(nilfs);
279                 else {
280                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
281                                sbi->s_super->s_id);
282                         return -EIO;
283                 }
284         }
285         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
286         if (unlikely(err)) {
287                 printk(KERN_ERR "NILFS: failed to count free blocks\n");
288                 return err;
289         }
290         spin_lock(&nilfs->ns_last_segment_lock);
291         sbp[0]->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
292         sbp[0]->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
293         sbp[0]->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
294         spin_unlock(&nilfs->ns_last_segment_lock);
295
296         t = get_seconds();
297         nilfs->ns_sbwtime[0] = t;
298         sbp[0]->s_free_blocks_count = cpu_to_le64(nfreeblocks);
299         sbp[0]->s_wtime = cpu_to_le64(t);
300         sbp[0]->s_sum = 0;
301         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
302                                              (unsigned char *)sbp[0],
303                                              nilfs->ns_sbsize));
304         if (dupsb && sbp[1]) {
305                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
306                 nilfs->ns_sbwtime[1] = t;
307         }
308         sbi->s_super->s_dirt = 0;
309         return nilfs_sync_super(sbi, dupsb);
310 }
311
312 static void nilfs_put_super(struct super_block *sb)
313 {
314         struct nilfs_sb_info *sbi = NILFS_SB(sb);
315         struct the_nilfs *nilfs = sbi->s_nilfs;
316
317         lock_kernel();
318
319         if (sb->s_dirt)
320                 nilfs_write_super(sb);
321
322         nilfs_detach_segment_constructor(sbi);
323
324         if (!(sb->s_flags & MS_RDONLY)) {
325                 down_write(&nilfs->ns_sem);
326                 nilfs->ns_sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
327                 nilfs_commit_super(sbi, 1);
328                 up_write(&nilfs->ns_sem);
329         }
330         down_write(&nilfs->ns_super_sem);
331         if (nilfs->ns_current == sbi)
332                 nilfs->ns_current = NULL;
333         up_write(&nilfs->ns_super_sem);
334
335         nilfs_detach_checkpoint(sbi);
336         put_nilfs(sbi->s_nilfs);
337         sbi->s_super = NULL;
338         sb->s_fs_info = NULL;
339         nilfs_put_sbinfo(sbi);
340
341         unlock_kernel();
342 }
343
344 /**
345  * nilfs_write_super - write super block(s) of NILFS
346  * @sb: super_block
347  *
348  * nilfs_write_super() gets a fs-dependent lock, writes super block(s), and
349  * clears s_dirt.  This function is called in the section protected by
350  * lock_super().
351  *
352  * The s_dirt flag is managed by each filesystem and we protect it by ns_sem
353  * of the struct the_nilfs.  Lock order must be as follows:
354  *
355  *   1. lock_super()
356  *   2.    down_write(&nilfs->ns_sem)
357  *
358  * Inside NILFS, locking ns_sem is enough to protect s_dirt and the buffer
359  * of the super block (nilfs->ns_sbp[]).
360  *
361  * In most cases, VFS functions call lock_super() before calling these
362  * methods.  So we must be careful not to bring on deadlocks when using
363  * lock_super();  see generic_shutdown_super(), write_super(), and so on.
364  *
365  * Note that order of lock_kernel() and lock_super() depends on contexts
366  * of VFS.  We should also note that lock_kernel() can be used in its
367  * protective section and only the outermost one has an effect.
368  */
369 static void nilfs_write_super(struct super_block *sb)
370 {
371         struct nilfs_sb_info *sbi = NILFS_SB(sb);
372         struct the_nilfs *nilfs = sbi->s_nilfs;
373
374         down_write(&nilfs->ns_sem);
375         if (!(sb->s_flags & MS_RDONLY)) {
376                 struct nilfs_super_block **sbp = nilfs->ns_sbp;
377                 u64 t = get_seconds();
378                 int dupsb;
379
380                 if (!nilfs_discontinued(nilfs) && t >= nilfs->ns_sbwtime[0] &&
381                     t < nilfs->ns_sbwtime[0] + NILFS_SB_FREQ) {
382                         up_write(&nilfs->ns_sem);
383                         return;
384                 }
385                 dupsb = sbp[1] && t > nilfs->ns_sbwtime[1] + NILFS_ALTSB_FREQ;
386                 nilfs_commit_super(sbi, dupsb);
387         }
388         sb->s_dirt = 0;
389         up_write(&nilfs->ns_sem);
390 }
391
392 static int nilfs_sync_fs(struct super_block *sb, int wait)
393 {
394         int err = 0;
395
396         nilfs_write_super(sb);
397
398         /* This function is called when super block should be written back */
399         if (wait)
400                 err = nilfs_construct_segment(sb);
401         return err;
402 }
403
404 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
405 {
406         struct the_nilfs *nilfs = sbi->s_nilfs;
407         struct nilfs_checkpoint *raw_cp;
408         struct buffer_head *bh_cp;
409         int err;
410
411         down_write(&nilfs->ns_super_sem);
412         list_add(&sbi->s_list, &nilfs->ns_supers);
413         up_write(&nilfs->ns_super_sem);
414
415         sbi->s_ifile = nilfs_mdt_new(
416                 nilfs, sbi->s_super, NILFS_IFILE_INO, NILFS_IFILE_GFP);
417         if (!sbi->s_ifile)
418                 return -ENOMEM;
419
420         err = nilfs_palloc_init_blockgroup(sbi->s_ifile, nilfs->ns_inode_size);
421         if (unlikely(err))
422                 goto failed;
423
424         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
425                                           &bh_cp);
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_clear(sbi->s_ifile);
463         nilfs_mdt_destroy(sbi->s_ifile);
464         sbi->s_ifile = NULL;
465         down_write(&nilfs->ns_super_sem);
466         list_del_init(&sbi->s_list);
467         up_write(&nilfs->ns_super_sem);
468 }
469
470 static int nilfs_mark_recovery_complete(struct nilfs_sb_info *sbi)
471 {
472         struct the_nilfs *nilfs = sbi->s_nilfs;
473         int err = 0;
474
475         down_write(&nilfs->ns_sem);
476         if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
477                 nilfs->ns_mount_state |= NILFS_VALID_FS;
478                 err = nilfs_commit_super(sbi, 1);
479                 if (likely(!err))
480                         printk(KERN_INFO "NILFS: recovery complete.\n");
481         }
482         up_write(&nilfs->ns_sem);
483         return err;
484 }
485
486 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
487 {
488         struct super_block *sb = dentry->d_sb;
489         struct nilfs_sb_info *sbi = NILFS_SB(sb);
490         struct the_nilfs *nilfs = sbi->s_nilfs;
491         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
492         unsigned long long blocks;
493         unsigned long overhead;
494         unsigned long nrsvblocks;
495         sector_t nfreeblocks;
496         int err;
497
498         /*
499          * Compute all of the segment blocks
500          *
501          * The blocks before first segment and after last segment
502          * are excluded.
503          */
504         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
505                 - nilfs->ns_first_data_block;
506         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
507
508         /*
509          * Compute the overhead
510          *
511          * When distributing meta data blocks outside semgent structure,
512          * We must count them as the overhead.
513          */
514         overhead = 0;
515
516         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
517         if (unlikely(err))
518                 return err;
519
520         buf->f_type = NILFS_SUPER_MAGIC;
521         buf->f_bsize = sb->s_blocksize;
522         buf->f_blocks = blocks - overhead;
523         buf->f_bfree = nfreeblocks;
524         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
525                 (buf->f_bfree - nrsvblocks) : 0;
526         buf->f_files = atomic_read(&sbi->s_inodes_count);
527         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
528         buf->f_namelen = NILFS_NAME_LEN;
529         buf->f_fsid.val[0] = (u32)id;
530         buf->f_fsid.val[1] = (u32)(id >> 32);
531
532         return 0;
533 }
534
535 static struct super_operations nilfs_sops = {
536         .alloc_inode    = nilfs_alloc_inode,
537         .destroy_inode  = nilfs_destroy_inode,
538         .dirty_inode    = nilfs_dirty_inode,
539         /* .write_inode    = nilfs_write_inode, */
540         /* .put_inode      = nilfs_put_inode, */
541         /* .drop_inode    = nilfs_drop_inode, */
542         .delete_inode   = nilfs_delete_inode,
543         .put_super      = nilfs_put_super,
544         .write_super    = nilfs_write_super,
545         .sync_fs        = nilfs_sync_fs,
546         /* .write_super_lockfs */
547         /* .unlockfs */
548         .statfs         = nilfs_statfs,
549         .remount_fs     = nilfs_remount,
550         .clear_inode    = nilfs_clear_inode,
551         /* .umount_begin */
552         /* .show_options */
553 };
554
555 static struct inode *
556 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
557 {
558         struct inode *inode;
559
560         if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
561             ino != NILFS_SKETCH_INO)
562                 return ERR_PTR(-ESTALE);
563
564         inode = nilfs_iget(sb, ino);
565         if (IS_ERR(inode))
566                 return ERR_CAST(inode);
567         if (generation && inode->i_generation != generation) {
568                 iput(inode);
569                 return ERR_PTR(-ESTALE);
570         }
571
572         return inode;
573 }
574
575 static struct dentry *
576 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
577                    int fh_type)
578 {
579         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
580                                     nilfs_nfs_get_inode);
581 }
582
583 static struct dentry *
584 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
585                    int fh_type)
586 {
587         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
588                                     nilfs_nfs_get_inode);
589 }
590
591 static struct export_operations nilfs_export_ops = {
592         .fh_to_dentry = nilfs_fh_to_dentry,
593         .fh_to_parent = nilfs_fh_to_parent,
594         .get_parent = nilfs_get_parent,
595 };
596
597 enum {
598         Opt_err_cont, Opt_err_panic, Opt_err_ro,
599         Opt_barrier, Opt_snapshot, Opt_order,
600         Opt_err,
601 };
602
603 static match_table_t tokens = {
604         {Opt_err_cont, "errors=continue"},
605         {Opt_err_panic, "errors=panic"},
606         {Opt_err_ro, "errors=remount-ro"},
607         {Opt_barrier, "barrier=%s"},
608         {Opt_snapshot, "cp=%u"},
609         {Opt_order, "order=%s"},
610         {Opt_err, NULL}
611 };
612
613 static int match_bool(substring_t *s, int *result)
614 {
615         int len = s->to - s->from;
616
617         if (strncmp(s->from, "on", len) == 0)
618                 *result = 1;
619         else if (strncmp(s->from, "off", len) == 0)
620                 *result = 0;
621         else
622                 return 1;
623         return 0;
624 }
625
626 static int parse_options(char *options, struct super_block *sb)
627 {
628         struct nilfs_sb_info *sbi = NILFS_SB(sb);
629         char *p;
630         substring_t args[MAX_OPT_ARGS];
631         int option;
632
633         if (!options)
634                 return 1;
635
636         while ((p = strsep(&options, ",")) != NULL) {
637                 int token;
638                 if (!*p)
639                         continue;
640
641                 token = match_token(p, tokens, args);
642                 switch (token) {
643                 case Opt_barrier:
644                         if (match_bool(&args[0], &option))
645                                 return 0;
646                         if (option)
647                                 nilfs_set_opt(sbi, BARRIER);
648                         else
649                                 nilfs_clear_opt(sbi, BARRIER);
650                         break;
651                 case Opt_order:
652                         if (strcmp(args[0].from, "relaxed") == 0)
653                                 /* Ordered data semantics */
654                                 nilfs_clear_opt(sbi, STRICT_ORDER);
655                         else if (strcmp(args[0].from, "strict") == 0)
656                                 /* Strict in-order semantics */
657                                 nilfs_set_opt(sbi, STRICT_ORDER);
658                         else
659                                 return 0;
660                         break;
661                 case Opt_err_panic:
662                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
663                         break;
664                 case Opt_err_ro:
665                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
666                         break;
667                 case Opt_err_cont:
668                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
669                         break;
670                 case Opt_snapshot:
671                         if (match_int(&args[0], &option) || option <= 0)
672                                 return 0;
673                         if (!(sb->s_flags & MS_RDONLY))
674                                 return 0;
675                         sbi->s_snapshot_cno = option;
676                         nilfs_set_opt(sbi, SNAPSHOT);
677                         break;
678                 default:
679                         printk(KERN_ERR
680                                "NILFS: Unrecognized mount option \"%s\"\n", p);
681                         return 0;
682                 }
683         }
684         return 1;
685 }
686
687 static inline void
688 nilfs_set_default_options(struct nilfs_sb_info *sbi,
689                           struct nilfs_super_block *sbp)
690 {
691         sbi->s_mount_opt =
692                 NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER;
693 }
694
695 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
696 {
697         struct the_nilfs *nilfs = sbi->s_nilfs;
698         struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
699         int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
700         int mnt_count = le16_to_cpu(sbp->s_mnt_count);
701
702         /* nilfs->sem must be locked by the caller. */
703         if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
704                 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
705         } else if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
706                 printk(KERN_WARNING
707                        "NILFS warning: mounting fs with errors\n");
708 #if 0
709         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
710                 printk(KERN_WARNING
711                        "NILFS warning: maximal mount count reached\n");
712 #endif
713         }
714         if (!max_mnt_count)
715                 sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
716
717         sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
718         sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
719         sbp->s_mtime = cpu_to_le64(get_seconds());
720         return nilfs_commit_super(sbi, 1);
721 }
722
723 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
724                                                  u64 pos, int blocksize,
725                                                  struct buffer_head **pbh)
726 {
727         unsigned long long sb_index = pos;
728         unsigned long offset;
729
730         offset = do_div(sb_index, blocksize);
731         *pbh = sb_bread(sb, sb_index);
732         if (!*pbh)
733                 return NULL;
734         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
735 }
736
737 int nilfs_store_magic_and_option(struct super_block *sb,
738                                  struct nilfs_super_block *sbp,
739                                  char *data)
740 {
741         struct nilfs_sb_info *sbi = NILFS_SB(sb);
742
743         sb->s_magic = le16_to_cpu(sbp->s_magic);
744
745         /* FS independent flags */
746 #ifdef NILFS_ATIME_DISABLE
747         sb->s_flags |= MS_NOATIME;
748 #endif
749
750         nilfs_set_default_options(sbi, sbp);
751
752         sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
753         sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
754         sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
755         sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
756
757         return !parse_options(data, sb) ? -EINVAL : 0 ;
758 }
759
760 /**
761  * nilfs_fill_super() - initialize a super block instance
762  * @sb: super_block
763  * @data: mount options
764  * @silent: silent mode flag
765  * @nilfs: the_nilfs struct
766  *
767  * This function is called exclusively by nilfs->ns_mount_mutex.
768  * So, the recovery process is protected from other simultaneous mounts.
769  */
770 static int
771 nilfs_fill_super(struct super_block *sb, void *data, int silent,
772                  struct the_nilfs *nilfs)
773 {
774         struct nilfs_sb_info *sbi;
775         struct inode *root;
776         __u64 cno;
777         int err;
778
779         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
780         if (!sbi)
781                 return -ENOMEM;
782
783         sb->s_fs_info = sbi;
784
785         get_nilfs(nilfs);
786         sbi->s_nilfs = nilfs;
787         sbi->s_super = sb;
788         atomic_set(&sbi->s_count, 1);
789
790         err = init_nilfs(nilfs, sbi, (char *)data);
791         if (err)
792                 goto failed_sbi;
793
794         spin_lock_init(&sbi->s_inode_lock);
795         INIT_LIST_HEAD(&sbi->s_dirty_files);
796         INIT_LIST_HEAD(&sbi->s_list);
797
798         /*
799          * Following initialization is overlapped because
800          * nilfs_sb_info structure has been cleared at the beginning.
801          * But we reserve them to keep our interest and make ready
802          * for the future change.
803          */
804         get_random_bytes(&sbi->s_next_generation,
805                          sizeof(sbi->s_next_generation));
806         spin_lock_init(&sbi->s_next_gen_lock);
807
808         sb->s_op = &nilfs_sops;
809         sb->s_export_op = &nilfs_export_ops;
810         sb->s_root = NULL;
811         sb->s_time_gran = 1;
812
813         if (!nilfs_loaded(nilfs)) {
814                 err = load_nilfs(nilfs, sbi);
815                 if (err)
816                         goto failed_sbi;
817         }
818         cno = nilfs_last_cno(nilfs);
819
820         if (sb->s_flags & MS_RDONLY) {
821                 if (nilfs_test_opt(sbi, SNAPSHOT)) {
822                         err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
823                                                        sbi->s_snapshot_cno);
824                         if (err < 0)
825                                 goto failed_sbi;
826                         if (!err) {
827                                 printk(KERN_ERR
828                                        "NILFS: The specified checkpoint is "
829                                        "not a snapshot "
830                                        "(checkpoint number=%llu).\n",
831                                        (unsigned long long)sbi->s_snapshot_cno);
832                                 err = -EINVAL;
833                                 goto failed_sbi;
834                         }
835                         cno = sbi->s_snapshot_cno;
836                 } else
837                         /* Read-only mount */
838                         sbi->s_snapshot_cno = cno;
839         }
840
841         err = nilfs_attach_checkpoint(sbi, cno);
842         if (err) {
843                 printk(KERN_ERR "NILFS: error loading a checkpoint"
844                        " (checkpoint number=%llu).\n", (unsigned long long)cno);
845                 goto failed_sbi;
846         }
847
848         if (!(sb->s_flags & MS_RDONLY)) {
849                 err = nilfs_attach_segment_constructor(sbi);
850                 if (err)
851                         goto failed_checkpoint;
852         }
853
854         root = nilfs_iget(sb, NILFS_ROOT_INO);
855         if (IS_ERR(root)) {
856                 printk(KERN_ERR "NILFS: get root inode failed\n");
857                 err = PTR_ERR(root);
858                 goto failed_segctor;
859         }
860         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
861                 iput(root);
862                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
863                 err = -EINVAL;
864                 goto failed_segctor;
865         }
866         sb->s_root = d_alloc_root(root);
867         if (!sb->s_root) {
868                 iput(root);
869                 printk(KERN_ERR "NILFS: get root dentry failed\n");
870                 err = -ENOMEM;
871                 goto failed_segctor;
872         }
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         err = nilfs_mark_recovery_complete(sbi);
881         if (unlikely(err)) {
882                 printk(KERN_ERR "NILFS: recovery failed.\n");
883                 goto failed_root;
884         }
885
886         down_write(&nilfs->ns_super_sem);
887         if (!nilfs_test_opt(sbi, SNAPSHOT))
888                 nilfs->ns_current = sbi;
889         up_write(&nilfs->ns_super_sem);
890
891         return 0;
892
893  failed_root:
894         dput(sb->s_root);
895         sb->s_root = NULL;
896
897  failed_segctor:
898         nilfs_detach_segment_constructor(sbi);
899
900  failed_checkpoint:
901         nilfs_detach_checkpoint(sbi);
902
903  failed_sbi:
904         put_nilfs(nilfs);
905         sb->s_fs_info = NULL;
906         nilfs_put_sbinfo(sbi);
907         return err;
908 }
909
910 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
911 {
912         struct nilfs_sb_info *sbi = NILFS_SB(sb);
913         struct nilfs_super_block *sbp;
914         struct the_nilfs *nilfs = sbi->s_nilfs;
915         unsigned long old_sb_flags;
916         struct nilfs_mount_options old_opts;
917         int err;
918
919         lock_kernel();
920
921         down_write(&nilfs->ns_super_sem);
922         old_sb_flags = sb->s_flags;
923         old_opts.mount_opt = sbi->s_mount_opt;
924         old_opts.snapshot_cno = sbi->s_snapshot_cno;
925
926         if (!parse_options(data, sb)) {
927                 err = -EINVAL;
928                 goto restore_opts;
929         }
930         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
931
932         if ((*flags & MS_RDONLY) &&
933             sbi->s_snapshot_cno != old_opts.snapshot_cno) {
934                 printk(KERN_WARNING "NILFS (device %s): couldn't "
935                        "remount to a different snapshot. \n",
936                        sb->s_id);
937                 err = -EINVAL;
938                 goto restore_opts;
939         }
940
941         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
942                 goto out;
943         if (*flags & MS_RDONLY) {
944                 /* Shutting down the segment constructor */
945                 nilfs_detach_segment_constructor(sbi);
946                 sb->s_flags |= MS_RDONLY;
947
948                 sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
949                 /* nilfs_set_opt(sbi, SNAPSHOT); */
950
951                 /*
952                  * Remounting a valid RW partition RDONLY, so set
953                  * the RDONLY flag and then mark the partition as valid again.
954                  */
955                 down_write(&nilfs->ns_sem);
956                 sbp = nilfs->ns_sbp[0];
957                 if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
958                     (nilfs->ns_mount_state & NILFS_VALID_FS))
959                         sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
960                 sbp->s_mtime = cpu_to_le64(get_seconds());
961                 nilfs_commit_super(sbi, 1);
962                 up_write(&nilfs->ns_sem);
963         } else {
964                 /*
965                  * Mounting a RDONLY partition read-write, so reread and
966                  * store the current valid flag.  (It may have been changed
967                  * by fsck since we originally mounted the partition.)
968                  */
969                 if (nilfs->ns_current && nilfs->ns_current != sbi) {
970                         printk(KERN_WARNING "NILFS (device %s): couldn't "
971                                "remount because an RW-mount exists.\n",
972                                sb->s_id);
973                         err = -EBUSY;
974                         goto restore_opts;
975                 }
976                 if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
977                         printk(KERN_WARNING "NILFS (device %s): couldn't "
978                                "remount because the current RO-mount is not "
979                                "the latest one.\n",
980                                sb->s_id);
981                         err = -EINVAL;
982                         goto restore_opts;
983                 }
984                 sb->s_flags &= ~MS_RDONLY;
985                 nilfs_clear_opt(sbi, SNAPSHOT);
986                 sbi->s_snapshot_cno = 0;
987
988                 err = nilfs_attach_segment_constructor(sbi);
989                 if (err)
990                         goto restore_opts;
991
992                 down_write(&nilfs->ns_sem);
993                 nilfs_setup_super(sbi);
994                 up_write(&nilfs->ns_sem);
995
996                 nilfs->ns_current = sbi;
997         }
998  out:
999         up_write(&nilfs->ns_super_sem);
1000         unlock_kernel();
1001         return 0;
1002
1003  restore_opts:
1004         sb->s_flags = old_sb_flags;
1005         sbi->s_mount_opt = old_opts.mount_opt;
1006         sbi->s_snapshot_cno = old_opts.snapshot_cno;
1007         up_write(&nilfs->ns_super_sem);
1008         unlock_kernel();
1009         return err;
1010 }
1011
1012 struct nilfs_super_data {
1013         struct block_device *bdev;
1014         struct nilfs_sb_info *sbi;
1015         __u64 cno;
1016         int flags;
1017 };
1018
1019 /**
1020  * nilfs_identify - pre-read mount options needed to identify mount instance
1021  * @data: mount options
1022  * @sd: nilfs_super_data
1023  */
1024 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1025 {
1026         char *p, *options = data;
1027         substring_t args[MAX_OPT_ARGS];
1028         int option, token;
1029         int ret = 0;
1030
1031         do {
1032                 p = strsep(&options, ",");
1033                 if (p != NULL && *p) {
1034                         token = match_token(p, tokens, args);
1035                         if (token == Opt_snapshot) {
1036                                 if (!(sd->flags & MS_RDONLY))
1037                                         ret++;
1038                                 else {
1039                                         ret = match_int(&args[0], &option);
1040                                         if (!ret) {
1041                                                 if (option > 0)
1042                                                         sd->cno = option;
1043                                                 else
1044                                                         ret++;
1045                                         }
1046                                 }
1047                         }
1048                         if (ret)
1049                                 printk(KERN_ERR
1050                                        "NILFS: invalid mount option: %s\n", p);
1051                 }
1052                 if (!options)
1053                         break;
1054                 BUG_ON(options == data);
1055                 *(options - 1) = ',';
1056         } while (!ret);
1057         return ret;
1058 }
1059
1060 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1061 {
1062         struct nilfs_super_data *sd = data;
1063
1064         s->s_bdev = sd->bdev;
1065         s->s_dev = s->s_bdev->bd_dev;
1066         return 0;
1067 }
1068
1069 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1070 {
1071         struct nilfs_super_data *sd = data;
1072
1073         return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1074 }
1075
1076 static int
1077 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1078              const char *dev_name, void *data, struct vfsmount *mnt)
1079 {
1080         struct nilfs_super_data sd;
1081         struct super_block *s;
1082         struct the_nilfs *nilfs;
1083         int err, need_to_close = 1;
1084
1085         sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1086         if (IS_ERR(sd.bdev))
1087                 return PTR_ERR(sd.bdev);
1088
1089         /*
1090          * To get mount instance using sget() vfs-routine, NILFS needs
1091          * much more information than normal filesystems to identify mount
1092          * instance.  For snapshot mounts, not only a mount type (ro-mount
1093          * or rw-mount) but also a checkpoint number is required.
1094          */
1095         sd.cno = 0;
1096         sd.flags = flags;
1097         if (nilfs_identify((char *)data, &sd)) {
1098                 err = -EINVAL;
1099                 goto failed;
1100         }
1101
1102         nilfs = find_or_create_nilfs(sd.bdev);
1103         if (!nilfs) {
1104                 err = -ENOMEM;
1105                 goto failed;
1106         }
1107
1108         mutex_lock(&nilfs->ns_mount_mutex);
1109
1110         if (!sd.cno) {
1111                 /*
1112                  * Check if an exclusive mount exists or not.
1113                  * Snapshot mounts coexist with a current mount
1114                  * (i.e. rw-mount or ro-mount), whereas rw-mount and
1115                  * ro-mount are mutually exclusive.
1116                  */
1117                 down_read(&nilfs->ns_super_sem);
1118                 if (nilfs->ns_current &&
1119                     ((nilfs->ns_current->s_super->s_flags ^ flags)
1120                      & MS_RDONLY)) {
1121                         up_read(&nilfs->ns_super_sem);
1122                         err = -EBUSY;
1123                         goto failed_unlock;
1124                 }
1125                 up_read(&nilfs->ns_super_sem);
1126         }
1127
1128         /*
1129          * Find existing nilfs_sb_info struct
1130          */
1131         sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1132
1133         if (!sd.cno)
1134                 /* trying to get the latest checkpoint.  */
1135                 sd.cno = nilfs_last_cno(nilfs);
1136
1137         /*
1138          * Get super block instance holding the nilfs_sb_info struct.
1139          * A new instance is allocated if no existing mount is present or
1140          * existing instance has been unmounted.
1141          */
1142         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1143         if (sd.sbi)
1144                 nilfs_put_sbinfo(sd.sbi);
1145
1146         if (IS_ERR(s)) {
1147                 err = PTR_ERR(s);
1148                 goto failed_unlock;
1149         }
1150
1151         if (!s->s_root) {
1152                 char b[BDEVNAME_SIZE];
1153
1154                 /* New superblock instance created */
1155                 s->s_flags = flags;
1156                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1157                 sb_set_blocksize(s, block_size(sd.bdev));
1158
1159                 err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1160                 if (err)
1161                         goto cancel_new;
1162
1163                 s->s_flags |= MS_ACTIVE;
1164                 need_to_close = 0;
1165         }
1166
1167         mutex_unlock(&nilfs->ns_mount_mutex);
1168         put_nilfs(nilfs);
1169         if (need_to_close)
1170                 close_bdev_exclusive(sd.bdev, flags);
1171         simple_set_mnt(mnt, s);
1172         return 0;
1173
1174  failed_unlock:
1175         mutex_unlock(&nilfs->ns_mount_mutex);
1176         put_nilfs(nilfs);
1177  failed:
1178         close_bdev_exclusive(sd.bdev, flags);
1179
1180         return err;
1181
1182  cancel_new:
1183         /* Abandoning the newly allocated superblock */
1184         mutex_unlock(&nilfs->ns_mount_mutex);
1185         put_nilfs(nilfs);
1186         up_write(&s->s_umount);
1187         deactivate_super(s);
1188         /*
1189          * deactivate_super() invokes close_bdev_exclusive().
1190          * We must finish all post-cleaning before this call;
1191          * put_nilfs() needs the block device.
1192          */
1193         return err;
1194 }
1195
1196 struct file_system_type nilfs_fs_type = {
1197         .owner    = THIS_MODULE,
1198         .name     = "nilfs2",
1199         .get_sb   = nilfs_get_sb,
1200         .kill_sb  = kill_block_super,
1201         .fs_flags = FS_REQUIRES_DEV,
1202 };
1203
1204 static int __init init_nilfs_fs(void)
1205 {
1206         int err;
1207
1208         err = nilfs_init_inode_cache();
1209         if (err)
1210                 goto failed;
1211
1212         err = nilfs_init_transaction_cache();
1213         if (err)
1214                 goto failed_inode_cache;
1215
1216         err = nilfs_init_segbuf_cache();
1217         if (err)
1218                 goto failed_transaction_cache;
1219
1220         err = nilfs_btree_path_cache_init();
1221         if (err)
1222                 goto failed_segbuf_cache;
1223
1224         err = register_filesystem(&nilfs_fs_type);
1225         if (err)
1226                 goto failed_btree_path_cache;
1227
1228         return 0;
1229
1230  failed_btree_path_cache:
1231         nilfs_btree_path_cache_destroy();
1232
1233  failed_segbuf_cache:
1234         nilfs_destroy_segbuf_cache();
1235
1236  failed_transaction_cache:
1237         nilfs_destroy_transaction_cache();
1238
1239  failed_inode_cache:
1240         nilfs_destroy_inode_cache();
1241
1242  failed:
1243         return err;
1244 }
1245
1246 static void __exit exit_nilfs_fs(void)
1247 {
1248         nilfs_destroy_segbuf_cache();
1249         nilfs_destroy_transaction_cache();
1250         nilfs_destroy_inode_cache();
1251         nilfs_btree_path_cache_destroy();
1252         unregister_filesystem(&nilfs_fs_type);
1253 }
1254
1255 module_init(init_nilfs_fs)
1256 module_exit(exit_nilfs_fs)