nilfs2: implement nilfs_show_options to display mount options in /proc/mounts
[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         if (sb->s_dirt)
317                 nilfs_write_super(sb);
318
319         nilfs_detach_segment_constructor(sbi);
320
321         if (!(sb->s_flags & MS_RDONLY)) {
322                 down_write(&nilfs->ns_sem);
323                 nilfs->ns_sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
324                 nilfs_commit_super(sbi, 1);
325                 up_write(&nilfs->ns_sem);
326         }
327         down_write(&nilfs->ns_super_sem);
328         if (nilfs->ns_current == sbi)
329                 nilfs->ns_current = NULL;
330         up_write(&nilfs->ns_super_sem);
331
332         nilfs_detach_checkpoint(sbi);
333         put_nilfs(sbi->s_nilfs);
334         sbi->s_super = NULL;
335         sb->s_fs_info = NULL;
336         nilfs_put_sbinfo(sbi);
337
338         unlock_kernel();
339 }
340
341 /**
342  * nilfs_write_super - write super block(s) of NILFS
343  * @sb: super_block
344  *
345  * nilfs_write_super() gets a fs-dependent lock, writes super block(s), and
346  * clears s_dirt.  This function is called in the section protected by
347  * lock_super().
348  *
349  * The s_dirt flag is managed by each filesystem and we protect it by ns_sem
350  * of the struct the_nilfs.  Lock order must be as follows:
351  *
352  *   1. lock_super()
353  *   2.    down_write(&nilfs->ns_sem)
354  *
355  * Inside NILFS, locking ns_sem is enough to protect s_dirt and the buffer
356  * of the super block (nilfs->ns_sbp[]).
357  *
358  * In most cases, VFS functions call lock_super() before calling these
359  * methods.  So we must be careful not to bring on deadlocks when using
360  * lock_super();  see generic_shutdown_super(), write_super(), and so on.
361  *
362  * Note that order of lock_kernel() and lock_super() depends on contexts
363  * of VFS.  We should also note that lock_kernel() can be used in its
364  * protective section and only the outermost one has an effect.
365  */
366 static void nilfs_write_super(struct super_block *sb)
367 {
368         struct nilfs_sb_info *sbi = NILFS_SB(sb);
369         struct the_nilfs *nilfs = sbi->s_nilfs;
370
371         down_write(&nilfs->ns_sem);
372         if (!(sb->s_flags & MS_RDONLY)) {
373                 struct nilfs_super_block **sbp = nilfs->ns_sbp;
374                 u64 t = get_seconds();
375                 int dupsb;
376
377                 if (!nilfs_discontinued(nilfs) && t >= nilfs->ns_sbwtime[0] &&
378                     t < nilfs->ns_sbwtime[0] + NILFS_SB_FREQ) {
379                         up_write(&nilfs->ns_sem);
380                         return;
381                 }
382                 dupsb = sbp[1] && t > nilfs->ns_sbwtime[1] + NILFS_ALTSB_FREQ;
383                 nilfs_commit_super(sbi, dupsb);
384         }
385         sb->s_dirt = 0;
386         up_write(&nilfs->ns_sem);
387 }
388
389 static int nilfs_sync_fs(struct super_block *sb, int wait)
390 {
391         int err = 0;
392
393         nilfs_write_super(sb);
394
395         /* This function is called when super block should be written back */
396         if (wait)
397                 err = nilfs_construct_segment(sb);
398         return err;
399 }
400
401 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
402 {
403         struct the_nilfs *nilfs = sbi->s_nilfs;
404         struct nilfs_checkpoint *raw_cp;
405         struct buffer_head *bh_cp;
406         int err;
407
408         down_write(&nilfs->ns_super_sem);
409         list_add(&sbi->s_list, &nilfs->ns_supers);
410         up_write(&nilfs->ns_super_sem);
411
412         sbi->s_ifile = nilfs_mdt_new(
413                 nilfs, sbi->s_super, NILFS_IFILE_INO, NILFS_IFILE_GFP);
414         if (!sbi->s_ifile)
415                 return -ENOMEM;
416
417         err = nilfs_palloc_init_blockgroup(sbi->s_ifile, nilfs->ns_inode_size);
418         if (unlikely(err))
419                 goto failed;
420
421         down_read(&nilfs->ns_segctor_sem);
422         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
423                                           &bh_cp);
424         up_read(&nilfs->ns_segctor_sem);
425         if (unlikely(err)) {
426                 if (err == -ENOENT || err == -EINVAL) {
427                         printk(KERN_ERR
428                                "NILFS: Invalid checkpoint "
429                                "(checkpoint number=%llu)\n",
430                                (unsigned long long)cno);
431                         err = -EINVAL;
432                 }
433                 goto failed;
434         }
435         err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
436         if (unlikely(err))
437                 goto failed_bh;
438         atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
439         atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
440
441         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
442         return 0;
443
444  failed_bh:
445         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
446  failed:
447         nilfs_mdt_destroy(sbi->s_ifile);
448         sbi->s_ifile = NULL;
449
450         down_write(&nilfs->ns_super_sem);
451         list_del_init(&sbi->s_list);
452         up_write(&nilfs->ns_super_sem);
453
454         return err;
455 }
456
457 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
458 {
459         struct the_nilfs *nilfs = sbi->s_nilfs;
460
461         nilfs_mdt_clear(sbi->s_ifile);
462         nilfs_mdt_destroy(sbi->s_ifile);
463         sbi->s_ifile = NULL;
464         down_write(&nilfs->ns_super_sem);
465         list_del_init(&sbi->s_list);
466         up_write(&nilfs->ns_super_sem);
467 }
468
469 static int nilfs_mark_recovery_complete(struct nilfs_sb_info *sbi)
470 {
471         struct the_nilfs *nilfs = sbi->s_nilfs;
472         int err = 0;
473
474         down_write(&nilfs->ns_sem);
475         if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
476                 nilfs->ns_mount_state |= NILFS_VALID_FS;
477                 err = nilfs_commit_super(sbi, 1);
478                 if (likely(!err))
479                         printk(KERN_INFO "NILFS: recovery complete.\n");
480         }
481         up_write(&nilfs->ns_sem);
482         return err;
483 }
484
485 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
486 {
487         struct super_block *sb = dentry->d_sb;
488         struct nilfs_sb_info *sbi = NILFS_SB(sb);
489         struct the_nilfs *nilfs = sbi->s_nilfs;
490         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
491         unsigned long long blocks;
492         unsigned long overhead;
493         unsigned long nrsvblocks;
494         sector_t nfreeblocks;
495         int err;
496
497         /*
498          * Compute all of the segment blocks
499          *
500          * The blocks before first segment and after last segment
501          * are excluded.
502          */
503         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
504                 - nilfs->ns_first_data_block;
505         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
506
507         /*
508          * Compute the overhead
509          *
510          * When distributing meta data blocks outside semgent structure,
511          * We must count them as the overhead.
512          */
513         overhead = 0;
514
515         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
516         if (unlikely(err))
517                 return err;
518
519         buf->f_type = NILFS_SUPER_MAGIC;
520         buf->f_bsize = sb->s_blocksize;
521         buf->f_blocks = blocks - overhead;
522         buf->f_bfree = nfreeblocks;
523         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
524                 (buf->f_bfree - nrsvblocks) : 0;
525         buf->f_files = atomic_read(&sbi->s_inodes_count);
526         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
527         buf->f_namelen = NILFS_NAME_LEN;
528         buf->f_fsid.val[0] = (u32)id;
529         buf->f_fsid.val[1] = (u32)(id >> 32);
530
531         return 0;
532 }
533
534 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
535 {
536         struct super_block *sb = vfs->mnt_sb;
537         struct nilfs_sb_info *sbi = NILFS_SB(sb);
538
539         if (!nilfs_test_opt(sbi, BARRIER))
540                 seq_printf(seq, ",barrier=off");
541         if (nilfs_test_opt(sbi, SNAPSHOT))
542                 seq_printf(seq, ",cp=%llu",
543                            (unsigned long long int)sbi->s_snapshot_cno);
544         if (nilfs_test_opt(sbi, ERRORS_RO))
545                 seq_printf(seq, ",errors=remount-ro");
546         if (nilfs_test_opt(sbi, ERRORS_PANIC))
547                 seq_printf(seq, ",errors=panic");
548         if (nilfs_test_opt(sbi, STRICT_ORDER))
549                 seq_printf(seq, ",order=strict");
550
551         return 0;
552 }
553
554 static struct super_operations nilfs_sops = {
555         .alloc_inode    = nilfs_alloc_inode,
556         .destroy_inode  = nilfs_destroy_inode,
557         .dirty_inode    = nilfs_dirty_inode,
558         /* .write_inode    = nilfs_write_inode, */
559         /* .put_inode      = nilfs_put_inode, */
560         /* .drop_inode    = nilfs_drop_inode, */
561         .delete_inode   = nilfs_delete_inode,
562         .put_super      = nilfs_put_super,
563         .write_super    = nilfs_write_super,
564         .sync_fs        = nilfs_sync_fs,
565         /* .write_super_lockfs */
566         /* .unlockfs */
567         .statfs         = nilfs_statfs,
568         .remount_fs     = nilfs_remount,
569         .clear_inode    = nilfs_clear_inode,
570         /* .umount_begin */
571         .show_options = nilfs_show_options
572 };
573
574 static struct inode *
575 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
576 {
577         struct inode *inode;
578
579         if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
580             ino != NILFS_SKETCH_INO)
581                 return ERR_PTR(-ESTALE);
582
583         inode = nilfs_iget(sb, ino);
584         if (IS_ERR(inode))
585                 return ERR_CAST(inode);
586         if (generation && inode->i_generation != generation) {
587                 iput(inode);
588                 return ERR_PTR(-ESTALE);
589         }
590
591         return inode;
592 }
593
594 static struct dentry *
595 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
596                    int fh_type)
597 {
598         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
599                                     nilfs_nfs_get_inode);
600 }
601
602 static struct dentry *
603 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
604                    int fh_type)
605 {
606         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
607                                     nilfs_nfs_get_inode);
608 }
609
610 static struct export_operations nilfs_export_ops = {
611         .fh_to_dentry = nilfs_fh_to_dentry,
612         .fh_to_parent = nilfs_fh_to_parent,
613         .get_parent = nilfs_get_parent,
614 };
615
616 enum {
617         Opt_err_cont, Opt_err_panic, Opt_err_ro,
618         Opt_barrier, Opt_snapshot, Opt_order,
619         Opt_err,
620 };
621
622 static match_table_t tokens = {
623         {Opt_err_cont, "errors=continue"},
624         {Opt_err_panic, "errors=panic"},
625         {Opt_err_ro, "errors=remount-ro"},
626         {Opt_barrier, "barrier=%s"},
627         {Opt_snapshot, "cp=%u"},
628         {Opt_order, "order=%s"},
629         {Opt_err, NULL}
630 };
631
632 static int match_bool(substring_t *s, int *result)
633 {
634         int len = s->to - s->from;
635
636         if (strncmp(s->from, "on", len) == 0)
637                 *result = 1;
638         else if (strncmp(s->from, "off", len) == 0)
639                 *result = 0;
640         else
641                 return 1;
642         return 0;
643 }
644
645 static int parse_options(char *options, struct super_block *sb)
646 {
647         struct nilfs_sb_info *sbi = NILFS_SB(sb);
648         char *p;
649         substring_t args[MAX_OPT_ARGS];
650         int option;
651
652         if (!options)
653                 return 1;
654
655         while ((p = strsep(&options, ",")) != NULL) {
656                 int token;
657                 if (!*p)
658                         continue;
659
660                 token = match_token(p, tokens, args);
661                 switch (token) {
662                 case Opt_barrier:
663                         if (match_bool(&args[0], &option))
664                                 return 0;
665                         if (option)
666                                 nilfs_set_opt(sbi, BARRIER);
667                         else
668                                 nilfs_clear_opt(sbi, BARRIER);
669                         break;
670                 case Opt_order:
671                         if (strcmp(args[0].from, "relaxed") == 0)
672                                 /* Ordered data semantics */
673                                 nilfs_clear_opt(sbi, STRICT_ORDER);
674                         else if (strcmp(args[0].from, "strict") == 0)
675                                 /* Strict in-order semantics */
676                                 nilfs_set_opt(sbi, STRICT_ORDER);
677                         else
678                                 return 0;
679                         break;
680                 case Opt_err_panic:
681                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
682                         break;
683                 case Opt_err_ro:
684                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
685                         break;
686                 case Opt_err_cont:
687                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
688                         break;
689                 case Opt_snapshot:
690                         if (match_int(&args[0], &option) || option <= 0)
691                                 return 0;
692                         if (!(sb->s_flags & MS_RDONLY))
693                                 return 0;
694                         sbi->s_snapshot_cno = option;
695                         nilfs_set_opt(sbi, SNAPSHOT);
696                         break;
697                 default:
698                         printk(KERN_ERR
699                                "NILFS: Unrecognized mount option \"%s\"\n", p);
700                         return 0;
701                 }
702         }
703         return 1;
704 }
705
706 static inline void
707 nilfs_set_default_options(struct nilfs_sb_info *sbi,
708                           struct nilfs_super_block *sbp)
709 {
710         sbi->s_mount_opt =
711                 NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER;
712 }
713
714 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
715 {
716         struct the_nilfs *nilfs = sbi->s_nilfs;
717         struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
718         int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
719         int mnt_count = le16_to_cpu(sbp->s_mnt_count);
720
721         /* nilfs->sem must be locked by the caller. */
722         if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
723                 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
724         } else if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
725                 printk(KERN_WARNING
726                        "NILFS warning: mounting fs with errors\n");
727 #if 0
728         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
729                 printk(KERN_WARNING
730                        "NILFS warning: maximal mount count reached\n");
731 #endif
732         }
733         if (!max_mnt_count)
734                 sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
735
736         sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
737         sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
738         sbp->s_mtime = cpu_to_le64(get_seconds());
739         return nilfs_commit_super(sbi, 1);
740 }
741
742 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
743                                                  u64 pos, int blocksize,
744                                                  struct buffer_head **pbh)
745 {
746         unsigned long long sb_index = pos;
747         unsigned long offset;
748
749         offset = do_div(sb_index, blocksize);
750         *pbh = sb_bread(sb, sb_index);
751         if (!*pbh)
752                 return NULL;
753         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
754 }
755
756 int nilfs_store_magic_and_option(struct super_block *sb,
757                                  struct nilfs_super_block *sbp,
758                                  char *data)
759 {
760         struct nilfs_sb_info *sbi = NILFS_SB(sb);
761
762         sb->s_magic = le16_to_cpu(sbp->s_magic);
763
764         /* FS independent flags */
765 #ifdef NILFS_ATIME_DISABLE
766         sb->s_flags |= MS_NOATIME;
767 #endif
768
769         nilfs_set_default_options(sbi, sbp);
770
771         sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
772         sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
773         sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
774         sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
775
776         return !parse_options(data, sb) ? -EINVAL : 0 ;
777 }
778
779 /**
780  * nilfs_fill_super() - initialize a super block instance
781  * @sb: super_block
782  * @data: mount options
783  * @silent: silent mode flag
784  * @nilfs: the_nilfs struct
785  *
786  * This function is called exclusively by nilfs->ns_mount_mutex.
787  * So, the recovery process is protected from other simultaneous mounts.
788  */
789 static int
790 nilfs_fill_super(struct super_block *sb, void *data, int silent,
791                  struct the_nilfs *nilfs)
792 {
793         struct nilfs_sb_info *sbi;
794         struct inode *root;
795         __u64 cno;
796         int err;
797
798         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
799         if (!sbi)
800                 return -ENOMEM;
801
802         sb->s_fs_info = sbi;
803
804         get_nilfs(nilfs);
805         sbi->s_nilfs = nilfs;
806         sbi->s_super = sb;
807         atomic_set(&sbi->s_count, 1);
808
809         err = init_nilfs(nilfs, sbi, (char *)data);
810         if (err)
811                 goto failed_sbi;
812
813         spin_lock_init(&sbi->s_inode_lock);
814         INIT_LIST_HEAD(&sbi->s_dirty_files);
815         INIT_LIST_HEAD(&sbi->s_list);
816
817         /*
818          * Following initialization is overlapped because
819          * nilfs_sb_info structure has been cleared at the beginning.
820          * But we reserve them to keep our interest and make ready
821          * for the future change.
822          */
823         get_random_bytes(&sbi->s_next_generation,
824                          sizeof(sbi->s_next_generation));
825         spin_lock_init(&sbi->s_next_gen_lock);
826
827         sb->s_op = &nilfs_sops;
828         sb->s_export_op = &nilfs_export_ops;
829         sb->s_root = NULL;
830         sb->s_time_gran = 1;
831
832         if (!nilfs_loaded(nilfs)) {
833                 err = load_nilfs(nilfs, sbi);
834                 if (err)
835                         goto failed_sbi;
836         }
837         cno = nilfs_last_cno(nilfs);
838
839         if (sb->s_flags & MS_RDONLY) {
840                 if (nilfs_test_opt(sbi, SNAPSHOT)) {
841                         err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
842                                                        sbi->s_snapshot_cno);
843                         if (err < 0)
844                                 goto failed_sbi;
845                         if (!err) {
846                                 printk(KERN_ERR
847                                        "NILFS: The specified checkpoint is "
848                                        "not a snapshot "
849                                        "(checkpoint number=%llu).\n",
850                                        (unsigned long long)sbi->s_snapshot_cno);
851                                 err = -EINVAL;
852                                 goto failed_sbi;
853                         }
854                         cno = sbi->s_snapshot_cno;
855                 } else
856                         /* Read-only mount */
857                         sbi->s_snapshot_cno = cno;
858         }
859
860         err = nilfs_attach_checkpoint(sbi, cno);
861         if (err) {
862                 printk(KERN_ERR "NILFS: error loading a checkpoint"
863                        " (checkpoint number=%llu).\n", (unsigned long long)cno);
864                 goto failed_sbi;
865         }
866
867         if (!(sb->s_flags & MS_RDONLY)) {
868                 err = nilfs_attach_segment_constructor(sbi);
869                 if (err)
870                         goto failed_checkpoint;
871         }
872
873         root = nilfs_iget(sb, NILFS_ROOT_INO);
874         if (IS_ERR(root)) {
875                 printk(KERN_ERR "NILFS: get root inode failed\n");
876                 err = PTR_ERR(root);
877                 goto failed_segctor;
878         }
879         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
880                 iput(root);
881                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
882                 err = -EINVAL;
883                 goto failed_segctor;
884         }
885         sb->s_root = d_alloc_root(root);
886         if (!sb->s_root) {
887                 iput(root);
888                 printk(KERN_ERR "NILFS: get root dentry failed\n");
889                 err = -ENOMEM;
890                 goto failed_segctor;
891         }
892
893         if (!(sb->s_flags & MS_RDONLY)) {
894                 down_write(&nilfs->ns_sem);
895                 nilfs_setup_super(sbi);
896                 up_write(&nilfs->ns_sem);
897         }
898
899         err = nilfs_mark_recovery_complete(sbi);
900         if (unlikely(err)) {
901                 printk(KERN_ERR "NILFS: recovery failed.\n");
902                 goto failed_root;
903         }
904
905         down_write(&nilfs->ns_super_sem);
906         if (!nilfs_test_opt(sbi, SNAPSHOT))
907                 nilfs->ns_current = sbi;
908         up_write(&nilfs->ns_super_sem);
909
910         return 0;
911
912  failed_root:
913         dput(sb->s_root);
914         sb->s_root = NULL;
915
916  failed_segctor:
917         nilfs_detach_segment_constructor(sbi);
918
919  failed_checkpoint:
920         nilfs_detach_checkpoint(sbi);
921
922  failed_sbi:
923         put_nilfs(nilfs);
924         sb->s_fs_info = NULL;
925         nilfs_put_sbinfo(sbi);
926         return err;
927 }
928
929 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
930 {
931         struct nilfs_sb_info *sbi = NILFS_SB(sb);
932         struct nilfs_super_block *sbp;
933         struct the_nilfs *nilfs = sbi->s_nilfs;
934         unsigned long old_sb_flags;
935         struct nilfs_mount_options old_opts;
936         int err;
937
938         lock_kernel();
939
940         down_write(&nilfs->ns_super_sem);
941         old_sb_flags = sb->s_flags;
942         old_opts.mount_opt = sbi->s_mount_opt;
943         old_opts.snapshot_cno = sbi->s_snapshot_cno;
944
945         if (!parse_options(data, sb)) {
946                 err = -EINVAL;
947                 goto restore_opts;
948         }
949         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
950
951         if ((*flags & MS_RDONLY) &&
952             sbi->s_snapshot_cno != old_opts.snapshot_cno) {
953                 printk(KERN_WARNING "NILFS (device %s): couldn't "
954                        "remount to a different snapshot. \n",
955                        sb->s_id);
956                 err = -EINVAL;
957                 goto restore_opts;
958         }
959
960         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
961                 goto out;
962         if (*flags & MS_RDONLY) {
963                 /* Shutting down the segment constructor */
964                 nilfs_detach_segment_constructor(sbi);
965                 sb->s_flags |= MS_RDONLY;
966
967                 sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
968                 /* nilfs_set_opt(sbi, SNAPSHOT); */
969
970                 /*
971                  * Remounting a valid RW partition RDONLY, so set
972                  * the RDONLY flag and then mark the partition as valid again.
973                  */
974                 down_write(&nilfs->ns_sem);
975                 sbp = nilfs->ns_sbp[0];
976                 if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
977                     (nilfs->ns_mount_state & NILFS_VALID_FS))
978                         sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
979                 sbp->s_mtime = cpu_to_le64(get_seconds());
980                 nilfs_commit_super(sbi, 1);
981                 up_write(&nilfs->ns_sem);
982         } else {
983                 /*
984                  * Mounting a RDONLY partition read-write, so reread and
985                  * store the current valid flag.  (It may have been changed
986                  * by fsck since we originally mounted the partition.)
987                  */
988                 if (nilfs->ns_current && nilfs->ns_current != sbi) {
989                         printk(KERN_WARNING "NILFS (device %s): couldn't "
990                                "remount because an RW-mount exists.\n",
991                                sb->s_id);
992                         err = -EBUSY;
993                         goto restore_opts;
994                 }
995                 if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
996                         printk(KERN_WARNING "NILFS (device %s): couldn't "
997                                "remount because the current RO-mount is not "
998                                "the latest one.\n",
999                                sb->s_id);
1000                         err = -EINVAL;
1001                         goto restore_opts;
1002                 }
1003                 sb->s_flags &= ~MS_RDONLY;
1004                 nilfs_clear_opt(sbi, SNAPSHOT);
1005                 sbi->s_snapshot_cno = 0;
1006
1007                 err = nilfs_attach_segment_constructor(sbi);
1008                 if (err)
1009                         goto restore_opts;
1010
1011                 down_write(&nilfs->ns_sem);
1012                 nilfs_setup_super(sbi);
1013                 up_write(&nilfs->ns_sem);
1014
1015                 nilfs->ns_current = sbi;
1016         }
1017  out:
1018         up_write(&nilfs->ns_super_sem);
1019         unlock_kernel();
1020         return 0;
1021
1022  restore_opts:
1023         sb->s_flags = old_sb_flags;
1024         sbi->s_mount_opt = old_opts.mount_opt;
1025         sbi->s_snapshot_cno = old_opts.snapshot_cno;
1026         up_write(&nilfs->ns_super_sem);
1027         unlock_kernel();
1028         return err;
1029 }
1030
1031 struct nilfs_super_data {
1032         struct block_device *bdev;
1033         struct nilfs_sb_info *sbi;
1034         __u64 cno;
1035         int flags;
1036 };
1037
1038 /**
1039  * nilfs_identify - pre-read mount options needed to identify mount instance
1040  * @data: mount options
1041  * @sd: nilfs_super_data
1042  */
1043 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1044 {
1045         char *p, *options = data;
1046         substring_t args[MAX_OPT_ARGS];
1047         int option, token;
1048         int ret = 0;
1049
1050         do {
1051                 p = strsep(&options, ",");
1052                 if (p != NULL && *p) {
1053                         token = match_token(p, tokens, args);
1054                         if (token == Opt_snapshot) {
1055                                 if (!(sd->flags & MS_RDONLY))
1056                                         ret++;
1057                                 else {
1058                                         ret = match_int(&args[0], &option);
1059                                         if (!ret) {
1060                                                 if (option > 0)
1061                                                         sd->cno = option;
1062                                                 else
1063                                                         ret++;
1064                                         }
1065                                 }
1066                         }
1067                         if (ret)
1068                                 printk(KERN_ERR
1069                                        "NILFS: invalid mount option: %s\n", p);
1070                 }
1071                 if (!options)
1072                         break;
1073                 BUG_ON(options == data);
1074                 *(options - 1) = ',';
1075         } while (!ret);
1076         return ret;
1077 }
1078
1079 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1080 {
1081         struct nilfs_super_data *sd = data;
1082
1083         s->s_bdev = sd->bdev;
1084         s->s_dev = s->s_bdev->bd_dev;
1085         return 0;
1086 }
1087
1088 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1089 {
1090         struct nilfs_super_data *sd = data;
1091
1092         return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1093 }
1094
1095 static int
1096 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1097              const char *dev_name, void *data, struct vfsmount *mnt)
1098 {
1099         struct nilfs_super_data sd;
1100         struct super_block *s;
1101         struct the_nilfs *nilfs;
1102         int err, need_to_close = 1;
1103
1104         sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1105         if (IS_ERR(sd.bdev))
1106                 return PTR_ERR(sd.bdev);
1107
1108         /*
1109          * To get mount instance using sget() vfs-routine, NILFS needs
1110          * much more information than normal filesystems to identify mount
1111          * instance.  For snapshot mounts, not only a mount type (ro-mount
1112          * or rw-mount) but also a checkpoint number is required.
1113          */
1114         sd.cno = 0;
1115         sd.flags = flags;
1116         if (nilfs_identify((char *)data, &sd)) {
1117                 err = -EINVAL;
1118                 goto failed;
1119         }
1120
1121         nilfs = find_or_create_nilfs(sd.bdev);
1122         if (!nilfs) {
1123                 err = -ENOMEM;
1124                 goto failed;
1125         }
1126
1127         mutex_lock(&nilfs->ns_mount_mutex);
1128
1129         if (!sd.cno) {
1130                 /*
1131                  * Check if an exclusive mount exists or not.
1132                  * Snapshot mounts coexist with a current mount
1133                  * (i.e. rw-mount or ro-mount), whereas rw-mount and
1134                  * ro-mount are mutually exclusive.
1135                  */
1136                 down_read(&nilfs->ns_super_sem);
1137                 if (nilfs->ns_current &&
1138                     ((nilfs->ns_current->s_super->s_flags ^ flags)
1139                      & MS_RDONLY)) {
1140                         up_read(&nilfs->ns_super_sem);
1141                         err = -EBUSY;
1142                         goto failed_unlock;
1143                 }
1144                 up_read(&nilfs->ns_super_sem);
1145         }
1146
1147         /*
1148          * Find existing nilfs_sb_info struct
1149          */
1150         sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1151
1152         if (!sd.cno)
1153                 /* trying to get the latest checkpoint.  */
1154                 sd.cno = nilfs_last_cno(nilfs);
1155
1156         /*
1157          * Get super block instance holding the nilfs_sb_info struct.
1158          * A new instance is allocated if no existing mount is present or
1159          * existing instance has been unmounted.
1160          */
1161         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1162         if (sd.sbi)
1163                 nilfs_put_sbinfo(sd.sbi);
1164
1165         if (IS_ERR(s)) {
1166                 err = PTR_ERR(s);
1167                 goto failed_unlock;
1168         }
1169
1170         if (!s->s_root) {
1171                 char b[BDEVNAME_SIZE];
1172
1173                 /* New superblock instance created */
1174                 s->s_flags = flags;
1175                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1176                 sb_set_blocksize(s, block_size(sd.bdev));
1177
1178                 err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1179                 if (err)
1180                         goto cancel_new;
1181
1182                 s->s_flags |= MS_ACTIVE;
1183                 need_to_close = 0;
1184         }
1185
1186         mutex_unlock(&nilfs->ns_mount_mutex);
1187         put_nilfs(nilfs);
1188         if (need_to_close)
1189                 close_bdev_exclusive(sd.bdev, flags);
1190         simple_set_mnt(mnt, s);
1191         return 0;
1192
1193  failed_unlock:
1194         mutex_unlock(&nilfs->ns_mount_mutex);
1195         put_nilfs(nilfs);
1196  failed:
1197         close_bdev_exclusive(sd.bdev, flags);
1198
1199         return err;
1200
1201  cancel_new:
1202         /* Abandoning the newly allocated superblock */
1203         mutex_unlock(&nilfs->ns_mount_mutex);
1204         put_nilfs(nilfs);
1205         up_write(&s->s_umount);
1206         deactivate_super(s);
1207         /*
1208          * deactivate_super() invokes close_bdev_exclusive().
1209          * We must finish all post-cleaning before this call;
1210          * put_nilfs() needs the block device.
1211          */
1212         return err;
1213 }
1214
1215 struct file_system_type nilfs_fs_type = {
1216         .owner    = THIS_MODULE,
1217         .name     = "nilfs2",
1218         .get_sb   = nilfs_get_sb,
1219         .kill_sb  = kill_block_super,
1220         .fs_flags = FS_REQUIRES_DEV,
1221 };
1222
1223 static int __init init_nilfs_fs(void)
1224 {
1225         int err;
1226
1227         err = nilfs_init_inode_cache();
1228         if (err)
1229                 goto failed;
1230
1231         err = nilfs_init_transaction_cache();
1232         if (err)
1233                 goto failed_inode_cache;
1234
1235         err = nilfs_init_segbuf_cache();
1236         if (err)
1237                 goto failed_transaction_cache;
1238
1239         err = nilfs_btree_path_cache_init();
1240         if (err)
1241                 goto failed_segbuf_cache;
1242
1243         err = register_filesystem(&nilfs_fs_type);
1244         if (err)
1245                 goto failed_btree_path_cache;
1246
1247         return 0;
1248
1249  failed_btree_path_cache:
1250         nilfs_btree_path_cache_destroy();
1251
1252  failed_segbuf_cache:
1253         nilfs_destroy_segbuf_cache();
1254
1255  failed_transaction_cache:
1256         nilfs_destroy_transaction_cache();
1257
1258  failed_inode_cache:
1259         nilfs_destroy_inode_cache();
1260
1261  failed:
1262         return err;
1263 }
1264
1265 static void __exit exit_nilfs_fs(void)
1266 {
1267         nilfs_destroy_segbuf_cache();
1268         nilfs_destroy_transaction_cache();
1269         nilfs_destroy_inode_cache();
1270         nilfs_btree_path_cache_destroy();
1271         unregister_filesystem(&nilfs_fs_type);
1272 }
1273
1274 module_init(init_nilfs_fs)
1275 module_exit(exit_nilfs_fs)