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