ab785f85aa501abb40691a61b24395e0419c62f8
[pandora-kernel.git] / fs / nilfs2 / super.c
1 /*
2  * super.c - NILFS module and super block management.
3  *
4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
19  *
20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
21  */
22 /*
23  *  linux/fs/ext2/super.c
24  *
25  * Copyright (C) 1992, 1993, 1994, 1995
26  * Remy Card (card@masi.ibp.fr)
27  * Laboratoire MASI - Institut Blaise Pascal
28  * Universite Pierre et Marie Curie (Paris VI)
29  *
30  *  from
31  *
32  *  linux/fs/minix/inode.c
33  *
34  *  Copyright (C) 1991, 1992  Linus Torvalds
35  *
36  *  Big-endian to little-endian byte-swapping/bitmaps by
37  *        David S. Miller (davem@caip.rutgers.edu), 1995
38  */
39
40 #include <linux/module.h>
41 #include <linux/string.h>
42 #include <linux/slab.h>
43 #include <linux/init.h>
44 #include <linux/blkdev.h>
45 #include <linux/parser.h>
46 #include <linux/random.h>
47 #include <linux/crc32.h>
48 #include <linux/smp_lock.h>
49 #include <linux/vfs.h>
50 #include <linux/writeback.h>
51 #include <linux/kobject.h>
52 #include <linux/exportfs.h>
53 #include "nilfs.h"
54 #include "mdt.h"
55 #include "alloc.h"
56 #include "page.h"
57 #include "cpfile.h"
58 #include "ifile.h"
59 #include "dat.h"
60 #include "segment.h"
61 #include "segbuf.h"
62
63 MODULE_AUTHOR("NTT Corp.");
64 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
65                    "(NILFS)");
66 MODULE_LICENSE("GPL");
67
68 static void nilfs_write_super(struct super_block *sb);
69 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
70
71 /**
72  * nilfs_error() - report failure condition on a filesystem
73  *
74  * nilfs_error() sets an ERROR_FS flag on the superblock as well as
75  * reporting an error message.  It should be called when NILFS detects
76  * incoherences or defects of meta data on disk.  As for sustainable
77  * errors such as a single-shot I/O error, nilfs_warning() or the printk()
78  * function should be used instead.
79  *
80  * The segment constructor must not call this function because it can
81  * kill itself.
82  */
83 void nilfs_error(struct super_block *sb, const char *function,
84                  const char *fmt, ...)
85 {
86         struct nilfs_sb_info *sbi = NILFS_SB(sb);
87         va_list args;
88
89         va_start(args, fmt);
90         printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
91         vprintk(fmt, args);
92         printk("\n");
93         va_end(args);
94
95         if (!(sb->s_flags & MS_RDONLY)) {
96                 struct the_nilfs *nilfs = sbi->s_nilfs;
97
98                 if (!nilfs_test_opt(sbi, ERRORS_CONT))
99                         nilfs_detach_segment_constructor(sbi);
100
101                 down_write(&nilfs->ns_sem);
102                 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
103                         nilfs->ns_mount_state |= NILFS_ERROR_FS;
104                         nilfs->ns_sbp[0]->s_state |=
105                                 cpu_to_le16(NILFS_ERROR_FS);
106                         nilfs_commit_super(sbi, 1);
107                 }
108                 up_write(&nilfs->ns_sem);
109
110                 if (nilfs_test_opt(sbi, ERRORS_RO)) {
111                         printk(KERN_CRIT "Remounting filesystem read-only\n");
112                         sb->s_flags |= MS_RDONLY;
113                 }
114         }
115
116         if (nilfs_test_opt(sbi, ERRORS_PANIC))
117                 panic("NILFS (device %s): panic forced after error\n",
118                       sb->s_id);
119 }
120
121 void nilfs_warning(struct super_block *sb, const char *function,
122                    const char *fmt, ...)
123 {
124         va_list args;
125
126         va_start(args, fmt);
127         printk(KERN_WARNING "NILFS warning (device %s): %s: ",
128                sb->s_id, function);
129         vprintk(fmt, args);
130         printk("\n");
131         va_end(args);
132 }
133
134 static struct kmem_cache *nilfs_inode_cachep;
135
136 struct inode *nilfs_alloc_inode_common(struct the_nilfs *nilfs)
137 {
138         struct nilfs_inode_info *ii;
139
140         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
141         if (!ii)
142                 return NULL;
143         ii->i_bh = NULL;
144         ii->i_state = 0;
145         ii->vfs_inode.i_version = 1;
146         nilfs_btnode_cache_init(&ii->i_btnode_cache, nilfs->ns_bdi);
147         return &ii->vfs_inode;
148 }
149
150 struct inode *nilfs_alloc_inode(struct super_block *sb)
151 {
152         return nilfs_alloc_inode_common(NILFS_SB(sb)->s_nilfs);
153 }
154
155 void nilfs_destroy_inode(struct inode *inode)
156 {
157         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
158 }
159
160 static void init_once(void *obj)
161 {
162         struct nilfs_inode_info *ii = obj;
163
164         INIT_LIST_HEAD(&ii->i_dirty);
165 #ifdef CONFIG_NILFS_XATTR
166         init_rwsem(&ii->xattr_sem);
167 #endif
168         nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
169         ii->i_bmap = (struct nilfs_bmap *)&ii->i_bmap_union;
170         inode_init_once(&ii->vfs_inode);
171 }
172
173 static int nilfs_init_inode_cache(void)
174 {
175         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
176                                                sizeof(struct nilfs_inode_info),
177                                                0, SLAB_RECLAIM_ACCOUNT,
178                                                init_once);
179
180         return (nilfs_inode_cachep == NULL) ? -ENOMEM : 0;
181 }
182
183 static inline void nilfs_destroy_inode_cache(void)
184 {
185         kmem_cache_destroy(nilfs_inode_cachep);
186 }
187
188 static void nilfs_clear_inode(struct inode *inode)
189 {
190         struct nilfs_inode_info *ii = NILFS_I(inode);
191
192 #ifdef CONFIG_NILFS_POSIX_ACL
193         if (ii->i_acl && ii->i_acl != NILFS_ACL_NOT_CACHED) {
194                 posix_acl_release(ii->i_acl);
195                 ii->i_acl = NILFS_ACL_NOT_CACHED;
196         }
197         if (ii->i_default_acl && ii->i_default_acl != NILFS_ACL_NOT_CACHED) {
198                 posix_acl_release(ii->i_default_acl);
199                 ii->i_default_acl = NILFS_ACL_NOT_CACHED;
200         }
201 #endif
202         /*
203          * Free resources allocated in nilfs_read_inode(), here.
204          */
205         BUG_ON(!list_empty(&ii->i_dirty));
206         brelse(ii->i_bh);
207         ii->i_bh = NULL;
208
209         if (test_bit(NILFS_I_BMAP, &ii->i_state))
210                 nilfs_bmap_clear(ii->i_bmap);
211
212         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
213 }
214
215 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int dupsb)
216 {
217         struct the_nilfs *nilfs = sbi->s_nilfs;
218         int err;
219         int barrier_done = 0;
220
221         if (nilfs_test_opt(sbi, BARRIER)) {
222                 set_buffer_ordered(nilfs->ns_sbh[0]);
223                 barrier_done = 1;
224         }
225  retry:
226         set_buffer_dirty(nilfs->ns_sbh[0]);
227         err = sync_dirty_buffer(nilfs->ns_sbh[0]);
228         if (err == -EOPNOTSUPP && barrier_done) {
229                 nilfs_warning(sbi->s_super, __func__,
230                               "barrier-based sync failed. "
231                               "disabling barriers\n");
232                 nilfs_clear_opt(sbi, BARRIER);
233                 barrier_done = 0;
234                 clear_buffer_ordered(nilfs->ns_sbh[0]);
235                 goto retry;
236         }
237         if (unlikely(err)) {
238                 printk(KERN_ERR
239                        "NILFS: unable to write superblock (err=%d)\n", err);
240                 if (err == -EIO && nilfs->ns_sbh[1]) {
241                         nilfs_fall_back_super_block(nilfs);
242                         goto retry;
243                 }
244         } else {
245                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
246
247                 /*
248                  * The latest segment becomes trailable from the position
249                  * written in superblock.
250                  */
251                 clear_nilfs_discontinued(nilfs);
252
253                 /* update GC protection for recent segments */
254                 if (nilfs->ns_sbh[1]) {
255                         sbp = NULL;
256                         if (dupsb) {
257                                 set_buffer_dirty(nilfs->ns_sbh[1]);
258                                 if (!sync_dirty_buffer(nilfs->ns_sbh[1]))
259                                         sbp = nilfs->ns_sbp[1];
260                         }
261                 }
262                 if (sbp) {
263                         spin_lock(&nilfs->ns_last_segment_lock);
264                         nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
265                         spin_unlock(&nilfs->ns_last_segment_lock);
266                 }
267         }
268
269         return err;
270 }
271
272 int nilfs_commit_super(struct nilfs_sb_info *sbi, int dupsb)
273 {
274         struct the_nilfs *nilfs = sbi->s_nilfs;
275         struct nilfs_super_block **sbp = nilfs->ns_sbp;
276         sector_t nfreeblocks;
277         time_t t;
278         int err;
279
280         /* nilfs->sem must be locked by the caller. */
281         if (sbp[0]->s_magic != NILFS_SUPER_MAGIC) {
282                 if (sbp[1] && sbp[1]->s_magic == NILFS_SUPER_MAGIC)
283                         nilfs_swap_super_block(nilfs);
284                 else {
285                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
286                                sbi->s_super->s_id);
287                         return -EIO;
288                 }
289         }
290         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
291         if (unlikely(err)) {
292                 printk(KERN_ERR "NILFS: failed to count free blocks\n");
293                 return err;
294         }
295         spin_lock(&nilfs->ns_last_segment_lock);
296         sbp[0]->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
297         sbp[0]->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
298         sbp[0]->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
299         spin_unlock(&nilfs->ns_last_segment_lock);
300
301         t = get_seconds();
302         nilfs->ns_sbwtime[0] = t;
303         sbp[0]->s_free_blocks_count = cpu_to_le64(nfreeblocks);
304         sbp[0]->s_wtime = cpu_to_le64(t);
305         sbp[0]->s_sum = 0;
306         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
307                                              (unsigned char *)sbp[0],
308                                              nilfs->ns_sbsize));
309         if (dupsb && sbp[1]) {
310                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
311                 nilfs->ns_sbwtime[1] = t;
312         }
313         sbi->s_super->s_dirt = 0;
314         return nilfs_sync_super(sbi, dupsb);
315 }
316
317 static void nilfs_put_super(struct super_block *sb)
318 {
319         struct nilfs_sb_info *sbi = NILFS_SB(sb);
320         struct the_nilfs *nilfs = sbi->s_nilfs;
321
322         lock_kernel();
323
324         if (sb->s_dirt)
325                 nilfs_write_super(sb);
326
327         nilfs_detach_segment_constructor(sbi);
328
329         if (!(sb->s_flags & MS_RDONLY)) {
330                 down_write(&nilfs->ns_sem);
331                 nilfs->ns_sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
332                 nilfs_commit_super(sbi, 1);
333                 up_write(&nilfs->ns_sem);
334         }
335         down_write(&nilfs->ns_super_sem);
336         if (nilfs->ns_current == sbi)
337                 nilfs->ns_current = NULL;
338         up_write(&nilfs->ns_super_sem);
339
340         nilfs_detach_checkpoint(sbi);
341         put_nilfs(sbi->s_nilfs);
342         sbi->s_super = NULL;
343         sb->s_fs_info = NULL;
344         nilfs_put_sbinfo(sbi);
345
346         unlock_kernel();
347 }
348
349 /**
350  * nilfs_write_super - write super block(s) of NILFS
351  * @sb: super_block
352  *
353  * nilfs_write_super() gets a fs-dependent lock, writes super block(s), and
354  * clears s_dirt.  This function is called in the section protected by
355  * lock_super().
356  *
357  * The s_dirt flag is managed by each filesystem and we protect it by ns_sem
358  * of the struct the_nilfs.  Lock order must be as follows:
359  *
360  *   1. lock_super()
361  *   2.    down_write(&nilfs->ns_sem)
362  *
363  * Inside NILFS, locking ns_sem is enough to protect s_dirt and the buffer
364  * of the super block (nilfs->ns_sbp[]).
365  *
366  * In most cases, VFS functions call lock_super() before calling these
367  * methods.  So we must be careful not to bring on deadlocks when using
368  * lock_super();  see generic_shutdown_super(), write_super(), and so on.
369  *
370  * Note that order of lock_kernel() and lock_super() depends on contexts
371  * of VFS.  We should also note that lock_kernel() can be used in its
372  * protective section and only the outermost one has an effect.
373  */
374 static void nilfs_write_super(struct super_block *sb)
375 {
376         struct nilfs_sb_info *sbi = NILFS_SB(sb);
377         struct the_nilfs *nilfs = sbi->s_nilfs;
378
379         down_write(&nilfs->ns_sem);
380         if (!(sb->s_flags & MS_RDONLY)) {
381                 struct nilfs_super_block **sbp = nilfs->ns_sbp;
382                 u64 t = get_seconds();
383                 int dupsb;
384
385                 if (!nilfs_discontinued(nilfs) && t >= nilfs->ns_sbwtime[0] &&
386                     t < nilfs->ns_sbwtime[0] + NILFS_SB_FREQ) {
387                         up_write(&nilfs->ns_sem);
388                         return;
389                 }
390                 dupsb = sbp[1] && t > nilfs->ns_sbwtime[1] + NILFS_ALTSB_FREQ;
391                 nilfs_commit_super(sbi, dupsb);
392         }
393         sb->s_dirt = 0;
394         up_write(&nilfs->ns_sem);
395 }
396
397 static int nilfs_sync_fs(struct super_block *sb, int wait)
398 {
399         int err = 0;
400
401         nilfs_write_super(sb);
402
403         /* This function is called when super block should be written back */
404         if (wait)
405                 err = nilfs_construct_segment(sb);
406         return err;
407 }
408
409 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
410 {
411         struct the_nilfs *nilfs = sbi->s_nilfs;
412         struct nilfs_checkpoint *raw_cp;
413         struct buffer_head *bh_cp;
414         int err;
415
416         down_write(&nilfs->ns_super_sem);
417         list_add(&sbi->s_list, &nilfs->ns_supers);
418         up_write(&nilfs->ns_super_sem);
419
420         sbi->s_ifile = nilfs_mdt_new(
421                 nilfs, sbi->s_super, NILFS_IFILE_INO, NILFS_IFILE_GFP);
422         if (!sbi->s_ifile)
423                 return -ENOMEM;
424
425         err = nilfs_palloc_init_blockgroup(sbi->s_ifile, nilfs->ns_inode_size);
426         if (unlikely(err))
427                 goto failed;
428
429         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
430                                           &bh_cp);
431         if (unlikely(err)) {
432                 if (err == -ENOENT || err == -EINVAL) {
433                         printk(KERN_ERR
434                                "NILFS: Invalid checkpoint "
435                                "(checkpoint number=%llu)\n",
436                                (unsigned long long)cno);
437                         err = -EINVAL;
438                 }
439                 goto failed;
440         }
441         err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
442         if (unlikely(err))
443                 goto failed_bh;
444         atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
445         atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
446
447         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
448         return 0;
449
450  failed_bh:
451         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
452  failed:
453         nilfs_mdt_destroy(sbi->s_ifile);
454         sbi->s_ifile = NULL;
455
456         down_write(&nilfs->ns_super_sem);
457         list_del_init(&sbi->s_list);
458         up_write(&nilfs->ns_super_sem);
459
460         return err;
461 }
462
463 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
464 {
465         struct the_nilfs *nilfs = sbi->s_nilfs;
466
467         nilfs_mdt_clear(sbi->s_ifile);
468         nilfs_mdt_destroy(sbi->s_ifile);
469         sbi->s_ifile = NULL;
470         down_write(&nilfs->ns_super_sem);
471         list_del_init(&sbi->s_list);
472         up_write(&nilfs->ns_super_sem);
473 }
474
475 static int nilfs_mark_recovery_complete(struct nilfs_sb_info *sbi)
476 {
477         struct the_nilfs *nilfs = sbi->s_nilfs;
478         int err = 0;
479
480         down_write(&nilfs->ns_sem);
481         if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
482                 nilfs->ns_mount_state |= NILFS_VALID_FS;
483                 err = nilfs_commit_super(sbi, 1);
484                 if (likely(!err))
485                         printk(KERN_INFO "NILFS: recovery complete.\n");
486         }
487         up_write(&nilfs->ns_sem);
488         return err;
489 }
490
491 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
492 {
493         struct super_block *sb = dentry->d_sb;
494         struct nilfs_sb_info *sbi = NILFS_SB(sb);
495         struct the_nilfs *nilfs = sbi->s_nilfs;
496         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
497         unsigned long long blocks;
498         unsigned long overhead;
499         unsigned long nrsvblocks;
500         sector_t nfreeblocks;
501         int err;
502
503         /*
504          * Compute all of the segment blocks
505          *
506          * The blocks before first segment and after last segment
507          * are excluded.
508          */
509         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
510                 - nilfs->ns_first_data_block;
511         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
512
513         /*
514          * Compute the overhead
515          *
516          * When distributing meta data blocks outside semgent structure,
517          * We must count them as the overhead.
518          */
519         overhead = 0;
520
521         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
522         if (unlikely(err))
523                 return err;
524
525         buf->f_type = NILFS_SUPER_MAGIC;
526         buf->f_bsize = sb->s_blocksize;
527         buf->f_blocks = blocks - overhead;
528         buf->f_bfree = nfreeblocks;
529         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
530                 (buf->f_bfree - nrsvblocks) : 0;
531         buf->f_files = atomic_read(&sbi->s_inodes_count);
532         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
533         buf->f_namelen = NILFS_NAME_LEN;
534         buf->f_fsid.val[0] = (u32)id;
535         buf->f_fsid.val[1] = (u32)(id >> 32);
536
537         return 0;
538 }
539
540 static struct super_operations nilfs_sops = {
541         .alloc_inode    = nilfs_alloc_inode,
542         .destroy_inode  = nilfs_destroy_inode,
543         .dirty_inode    = nilfs_dirty_inode,
544         /* .write_inode    = nilfs_write_inode, */
545         /* .put_inode      = nilfs_put_inode, */
546         /* .drop_inode    = nilfs_drop_inode, */
547         .delete_inode   = nilfs_delete_inode,
548         .put_super      = nilfs_put_super,
549         .write_super    = nilfs_write_super,
550         .sync_fs        = nilfs_sync_fs,
551         /* .write_super_lockfs */
552         /* .unlockfs */
553         .statfs         = nilfs_statfs,
554         .remount_fs     = nilfs_remount,
555         .clear_inode    = nilfs_clear_inode,
556         /* .umount_begin */
557         /* .show_options */
558 };
559
560 static struct inode *
561 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
562 {
563         struct inode *inode;
564
565         if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
566             ino != NILFS_SKETCH_INO)
567                 return ERR_PTR(-ESTALE);
568
569         inode = nilfs_iget(sb, ino);
570         if (IS_ERR(inode))
571                 return ERR_CAST(inode);
572         if (generation && inode->i_generation != generation) {
573                 iput(inode);
574                 return ERR_PTR(-ESTALE);
575         }
576
577         return inode;
578 }
579
580 static struct dentry *
581 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
582                    int fh_type)
583 {
584         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
585                                     nilfs_nfs_get_inode);
586 }
587
588 static struct dentry *
589 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
590                    int fh_type)
591 {
592         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
593                                     nilfs_nfs_get_inode);
594 }
595
596 static struct export_operations nilfs_export_ops = {
597         .fh_to_dentry = nilfs_fh_to_dentry,
598         .fh_to_parent = nilfs_fh_to_parent,
599         .get_parent = nilfs_get_parent,
600 };
601
602 enum {
603         Opt_err_cont, Opt_err_panic, Opt_err_ro,
604         Opt_barrier, Opt_snapshot, Opt_order,
605         Opt_err,
606 };
607
608 static match_table_t tokens = {
609         {Opt_err_cont, "errors=continue"},
610         {Opt_err_panic, "errors=panic"},
611         {Opt_err_ro, "errors=remount-ro"},
612         {Opt_barrier, "barrier=%s"},
613         {Opt_snapshot, "cp=%u"},
614         {Opt_order, "order=%s"},
615         {Opt_err, NULL}
616 };
617
618 static int match_bool(substring_t *s, int *result)
619 {
620         int len = s->to - s->from;
621
622         if (strncmp(s->from, "on", len) == 0)
623                 *result = 1;
624         else if (strncmp(s->from, "off", len) == 0)
625                 *result = 0;
626         else
627                 return 1;
628         return 0;
629 }
630
631 static int parse_options(char *options, struct super_block *sb)
632 {
633         struct nilfs_sb_info *sbi = NILFS_SB(sb);
634         char *p;
635         substring_t args[MAX_OPT_ARGS];
636         int option;
637
638         if (!options)
639                 return 1;
640
641         while ((p = strsep(&options, ",")) != NULL) {
642                 int token;
643                 if (!*p)
644                         continue;
645
646                 token = match_token(p, tokens, args);
647                 switch (token) {
648                 case Opt_barrier:
649                         if (match_bool(&args[0], &option))
650                                 return 0;
651                         if (option)
652                                 nilfs_set_opt(sbi, BARRIER);
653                         else
654                                 nilfs_clear_opt(sbi, BARRIER);
655                         break;
656                 case Opt_order:
657                         if (strcmp(args[0].from, "relaxed") == 0)
658                                 /* Ordered data semantics */
659                                 nilfs_clear_opt(sbi, STRICT_ORDER);
660                         else if (strcmp(args[0].from, "strict") == 0)
661                                 /* Strict in-order semantics */
662                                 nilfs_set_opt(sbi, STRICT_ORDER);
663                         else
664                                 return 0;
665                         break;
666                 case Opt_err_panic:
667                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
668                         break;
669                 case Opt_err_ro:
670                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
671                         break;
672                 case Opt_err_cont:
673                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
674                         break;
675                 case Opt_snapshot:
676                         if (match_int(&args[0], &option) || option <= 0)
677                                 return 0;
678                         if (!(sb->s_flags & MS_RDONLY))
679                                 return 0;
680                         sbi->s_snapshot_cno = option;
681                         nilfs_set_opt(sbi, SNAPSHOT);
682                         break;
683                 default:
684                         printk(KERN_ERR
685                                "NILFS: Unrecognized mount option \"%s\"\n", p);
686                         return 0;
687                 }
688         }
689         return 1;
690 }
691
692 static inline void
693 nilfs_set_default_options(struct nilfs_sb_info *sbi,
694                           struct nilfs_super_block *sbp)
695 {
696         sbi->s_mount_opt =
697                 NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER;
698 }
699
700 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
701 {
702         struct the_nilfs *nilfs = sbi->s_nilfs;
703         struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
704         int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
705         int mnt_count = le16_to_cpu(sbp->s_mnt_count);
706
707         /* nilfs->sem must be locked by the caller. */
708         if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
709                 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
710         } else if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
711                 printk(KERN_WARNING
712                        "NILFS warning: mounting fs with errors\n");
713 #if 0
714         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
715                 printk(KERN_WARNING
716                        "NILFS warning: maximal mount count reached\n");
717 #endif
718         }
719         if (!max_mnt_count)
720                 sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
721
722         sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
723         sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
724         sbp->s_mtime = cpu_to_le64(get_seconds());
725         return nilfs_commit_super(sbi, 1);
726 }
727
728 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
729                                                  u64 pos, int blocksize,
730                                                  struct buffer_head **pbh)
731 {
732         unsigned long long sb_index = pos;
733         unsigned long offset;
734
735         offset = do_div(sb_index, blocksize);
736         *pbh = sb_bread(sb, sb_index);
737         if (!*pbh)
738                 return NULL;
739         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
740 }
741
742 int nilfs_store_magic_and_option(struct super_block *sb,
743                                  struct nilfs_super_block *sbp,
744                                  char *data)
745 {
746         struct nilfs_sb_info *sbi = NILFS_SB(sb);
747
748         sb->s_magic = le16_to_cpu(sbp->s_magic);
749
750         /* FS independent flags */
751 #ifdef NILFS_ATIME_DISABLE
752         sb->s_flags |= MS_NOATIME;
753 #endif
754
755         nilfs_set_default_options(sbi, sbp);
756
757         sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
758         sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
759         sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
760         sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
761
762         return !parse_options(data, sb) ? -EINVAL : 0 ;
763 }
764
765 /**
766  * nilfs_fill_super() - initialize a super block instance
767  * @sb: super_block
768  * @data: mount options
769  * @silent: silent mode flag
770  * @nilfs: the_nilfs struct
771  *
772  * This function is called exclusively by nilfs->ns_mount_mutex.
773  * So, the recovery process is protected from other simultaneous mounts.
774  */
775 static int
776 nilfs_fill_super(struct super_block *sb, void *data, int silent,
777                  struct the_nilfs *nilfs)
778 {
779         struct nilfs_sb_info *sbi;
780         struct inode *root;
781         __u64 cno;
782         int err;
783
784         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
785         if (!sbi)
786                 return -ENOMEM;
787
788         sb->s_fs_info = sbi;
789
790         get_nilfs(nilfs);
791         sbi->s_nilfs = nilfs;
792         sbi->s_super = sb;
793         atomic_set(&sbi->s_count, 1);
794
795         err = init_nilfs(nilfs, sbi, (char *)data);
796         if (err)
797                 goto failed_sbi;
798
799         spin_lock_init(&sbi->s_inode_lock);
800         INIT_LIST_HEAD(&sbi->s_dirty_files);
801         INIT_LIST_HEAD(&sbi->s_list);
802
803         /*
804          * Following initialization is overlapped because
805          * nilfs_sb_info structure has been cleared at the beginning.
806          * But we reserve them to keep our interest and make ready
807          * for the future change.
808          */
809         get_random_bytes(&sbi->s_next_generation,
810                          sizeof(sbi->s_next_generation));
811         spin_lock_init(&sbi->s_next_gen_lock);
812
813         sb->s_op = &nilfs_sops;
814         sb->s_export_op = &nilfs_export_ops;
815         sb->s_root = NULL;
816         sb->s_time_gran = 1;
817
818         if (!nilfs_loaded(nilfs)) {
819                 err = load_nilfs(nilfs, sbi);
820                 if (err)
821                         goto failed_sbi;
822         }
823         cno = nilfs_last_cno(nilfs);
824
825         if (sb->s_flags & MS_RDONLY) {
826                 if (nilfs_test_opt(sbi, SNAPSHOT)) {
827                         err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
828                                                        sbi->s_snapshot_cno);
829                         if (err < 0)
830                                 goto failed_sbi;
831                         if (!err) {
832                                 printk(KERN_ERR
833                                        "NILFS: The specified checkpoint is "
834                                        "not a snapshot "
835                                        "(checkpoint number=%llu).\n",
836                                        (unsigned long long)sbi->s_snapshot_cno);
837                                 err = -EINVAL;
838                                 goto failed_sbi;
839                         }
840                         cno = sbi->s_snapshot_cno;
841                 } else
842                         /* Read-only mount */
843                         sbi->s_snapshot_cno = cno;
844         }
845
846         err = nilfs_attach_checkpoint(sbi, cno);
847         if (err) {
848                 printk(KERN_ERR "NILFS: error loading a checkpoint"
849                        " (checkpoint number=%llu).\n", (unsigned long long)cno);
850                 goto failed_sbi;
851         }
852
853         if (!(sb->s_flags & MS_RDONLY)) {
854                 err = nilfs_attach_segment_constructor(sbi);
855                 if (err)
856                         goto failed_checkpoint;
857         }
858
859         root = nilfs_iget(sb, NILFS_ROOT_INO);
860         if (IS_ERR(root)) {
861                 printk(KERN_ERR "NILFS: get root inode failed\n");
862                 err = PTR_ERR(root);
863                 goto failed_segctor;
864         }
865         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
866                 iput(root);
867                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
868                 err = -EINVAL;
869                 goto failed_segctor;
870         }
871         sb->s_root = d_alloc_root(root);
872         if (!sb->s_root) {
873                 iput(root);
874                 printk(KERN_ERR "NILFS: get root dentry failed\n");
875                 err = -ENOMEM;
876                 goto failed_segctor;
877         }
878
879         if (!(sb->s_flags & MS_RDONLY)) {
880                 down_write(&nilfs->ns_sem);
881                 nilfs_setup_super(sbi);
882                 up_write(&nilfs->ns_sem);
883         }
884
885         err = nilfs_mark_recovery_complete(sbi);
886         if (unlikely(err)) {
887                 printk(KERN_ERR "NILFS: recovery failed.\n");
888                 goto failed_root;
889         }
890
891         down_write(&nilfs->ns_super_sem);
892         if (!nilfs_test_opt(sbi, SNAPSHOT))
893                 nilfs->ns_current = sbi;
894         up_write(&nilfs->ns_super_sem);
895
896         return 0;
897
898  failed_root:
899         dput(sb->s_root);
900         sb->s_root = NULL;
901
902  failed_segctor:
903         nilfs_detach_segment_constructor(sbi);
904
905  failed_checkpoint:
906         nilfs_detach_checkpoint(sbi);
907
908  failed_sbi:
909         put_nilfs(nilfs);
910         sb->s_fs_info = NULL;
911         nilfs_put_sbinfo(sbi);
912         return err;
913 }
914
915 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
916 {
917         struct nilfs_sb_info *sbi = NILFS_SB(sb);
918         struct nilfs_super_block *sbp;
919         struct the_nilfs *nilfs = sbi->s_nilfs;
920         unsigned long old_sb_flags;
921         struct nilfs_mount_options old_opts;
922         int err;
923
924         lock_kernel();
925
926         down_write(&nilfs->ns_super_sem);
927         old_sb_flags = sb->s_flags;
928         old_opts.mount_opt = sbi->s_mount_opt;
929         old_opts.snapshot_cno = sbi->s_snapshot_cno;
930
931         if (!parse_options(data, sb)) {
932                 err = -EINVAL;
933                 goto restore_opts;
934         }
935         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
936
937         if ((*flags & MS_RDONLY) &&
938             sbi->s_snapshot_cno != old_opts.snapshot_cno) {
939                 printk(KERN_WARNING "NILFS (device %s): couldn't "
940                        "remount to a different snapshot. \n",
941                        sb->s_id);
942                 err = -EINVAL;
943                 goto restore_opts;
944         }
945
946         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
947                 goto out;
948         if (*flags & MS_RDONLY) {
949                 /* Shutting down the segment constructor */
950                 nilfs_detach_segment_constructor(sbi);
951                 sb->s_flags |= MS_RDONLY;
952
953                 sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
954                 /* nilfs_set_opt(sbi, SNAPSHOT); */
955
956                 /*
957                  * Remounting a valid RW partition RDONLY, so set
958                  * the RDONLY flag and then mark the partition as valid again.
959                  */
960                 down_write(&nilfs->ns_sem);
961                 sbp = nilfs->ns_sbp[0];
962                 if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
963                     (nilfs->ns_mount_state & NILFS_VALID_FS))
964                         sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
965                 sbp->s_mtime = cpu_to_le64(get_seconds());
966                 nilfs_commit_super(sbi, 1);
967                 up_write(&nilfs->ns_sem);
968         } else {
969                 /*
970                  * Mounting a RDONLY partition read-write, so reread and
971                  * store the current valid flag.  (It may have been changed
972                  * by fsck since we originally mounted the partition.)
973                  */
974                 if (nilfs->ns_current && nilfs->ns_current != sbi) {
975                         printk(KERN_WARNING "NILFS (device %s): couldn't "
976                                "remount because an RW-mount exists.\n",
977                                sb->s_id);
978                         err = -EBUSY;
979                         goto restore_opts;
980                 }
981                 if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
982                         printk(KERN_WARNING "NILFS (device %s): couldn't "
983                                "remount because the current RO-mount is not "
984                                "the latest one.\n",
985                                sb->s_id);
986                         err = -EINVAL;
987                         goto restore_opts;
988                 }
989                 sb->s_flags &= ~MS_RDONLY;
990                 nilfs_clear_opt(sbi, SNAPSHOT);
991                 sbi->s_snapshot_cno = 0;
992
993                 err = nilfs_attach_segment_constructor(sbi);
994                 if (err)
995                         goto restore_opts;
996
997                 down_write(&nilfs->ns_sem);
998                 nilfs_setup_super(sbi);
999                 up_write(&nilfs->ns_sem);
1000
1001                 nilfs->ns_current = sbi;
1002         }
1003  out:
1004         up_write(&nilfs->ns_super_sem);
1005         unlock_kernel();
1006         return 0;
1007
1008  restore_opts:
1009         sb->s_flags = old_sb_flags;
1010         sbi->s_mount_opt = old_opts.mount_opt;
1011         sbi->s_snapshot_cno = old_opts.snapshot_cno;
1012         up_write(&nilfs->ns_super_sem);
1013         unlock_kernel();
1014         return err;
1015 }
1016
1017 struct nilfs_super_data {
1018         struct block_device *bdev;
1019         struct nilfs_sb_info *sbi;
1020         __u64 cno;
1021         int flags;
1022 };
1023
1024 /**
1025  * nilfs_identify - pre-read mount options needed to identify mount instance
1026  * @data: mount options
1027  * @sd: nilfs_super_data
1028  */
1029 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1030 {
1031         char *p, *options = data;
1032         substring_t args[MAX_OPT_ARGS];
1033         int option, token;
1034         int ret = 0;
1035
1036         do {
1037                 p = strsep(&options, ",");
1038                 if (p != NULL && *p) {
1039                         token = match_token(p, tokens, args);
1040                         if (token == Opt_snapshot) {
1041                                 if (!(sd->flags & MS_RDONLY))
1042                                         ret++;
1043                                 else {
1044                                         ret = match_int(&args[0], &option);
1045                                         if (!ret) {
1046                                                 if (option > 0)
1047                                                         sd->cno = option;
1048                                                 else
1049                                                         ret++;
1050                                         }
1051                                 }
1052                         }
1053                         if (ret)
1054                                 printk(KERN_ERR
1055                                        "NILFS: invalid mount option: %s\n", p);
1056                 }
1057                 if (!options)
1058                         break;
1059                 BUG_ON(options == data);
1060                 *(options - 1) = ',';
1061         } while (!ret);
1062         return ret;
1063 }
1064
1065 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1066 {
1067         struct nilfs_super_data *sd = data;
1068
1069         s->s_bdev = sd->bdev;
1070         s->s_dev = s->s_bdev->bd_dev;
1071         return 0;
1072 }
1073
1074 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1075 {
1076         struct nilfs_super_data *sd = data;
1077
1078         return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1079 }
1080
1081 static int
1082 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1083              const char *dev_name, void *data, struct vfsmount *mnt)
1084 {
1085         struct nilfs_super_data sd;
1086         struct super_block *s;
1087         struct the_nilfs *nilfs;
1088         int err, need_to_close = 1;
1089
1090         sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1091         if (IS_ERR(sd.bdev))
1092                 return PTR_ERR(sd.bdev);
1093
1094         /*
1095          * To get mount instance using sget() vfs-routine, NILFS needs
1096          * much more information than normal filesystems to identify mount
1097          * instance.  For snapshot mounts, not only a mount type (ro-mount
1098          * or rw-mount) but also a checkpoint number is required.
1099          */
1100         sd.cno = 0;
1101         sd.flags = flags;
1102         if (nilfs_identify((char *)data, &sd)) {
1103                 err = -EINVAL;
1104                 goto failed;
1105         }
1106
1107         nilfs = find_or_create_nilfs(sd.bdev);
1108         if (!nilfs) {
1109                 err = -ENOMEM;
1110                 goto failed;
1111         }
1112
1113         mutex_lock(&nilfs->ns_mount_mutex);
1114
1115         if (!sd.cno) {
1116                 /*
1117                  * Check if an exclusive mount exists or not.
1118                  * Snapshot mounts coexist with a current mount
1119                  * (i.e. rw-mount or ro-mount), whereas rw-mount and
1120                  * ro-mount are mutually exclusive.
1121                  */
1122                 down_read(&nilfs->ns_super_sem);
1123                 if (nilfs->ns_current &&
1124                     ((nilfs->ns_current->s_super->s_flags ^ flags)
1125                      & MS_RDONLY)) {
1126                         up_read(&nilfs->ns_super_sem);
1127                         err = -EBUSY;
1128                         goto failed_unlock;
1129                 }
1130                 up_read(&nilfs->ns_super_sem);
1131         }
1132
1133         /*
1134          * Find existing nilfs_sb_info struct
1135          */
1136         sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1137
1138         if (!sd.cno)
1139                 /* trying to get the latest checkpoint.  */
1140                 sd.cno = nilfs_last_cno(nilfs);
1141
1142         /*
1143          * Get super block instance holding the nilfs_sb_info struct.
1144          * A new instance is allocated if no existing mount is present or
1145          * existing instance has been unmounted.
1146          */
1147         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1148         if (sd.sbi)
1149                 nilfs_put_sbinfo(sd.sbi);
1150
1151         if (IS_ERR(s)) {
1152                 err = PTR_ERR(s);
1153                 goto failed_unlock;
1154         }
1155
1156         if (!s->s_root) {
1157                 char b[BDEVNAME_SIZE];
1158
1159                 /* New superblock instance created */
1160                 s->s_flags = flags;
1161                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1162                 sb_set_blocksize(s, block_size(sd.bdev));
1163
1164                 err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1165                 if (err)
1166                         goto cancel_new;
1167
1168                 s->s_flags |= MS_ACTIVE;
1169                 need_to_close = 0;
1170         }
1171
1172         mutex_unlock(&nilfs->ns_mount_mutex);
1173         put_nilfs(nilfs);
1174         if (need_to_close)
1175                 close_bdev_exclusive(sd.bdev, flags);
1176         simple_set_mnt(mnt, s);
1177         return 0;
1178
1179  failed_unlock:
1180         mutex_unlock(&nilfs->ns_mount_mutex);
1181         put_nilfs(nilfs);
1182  failed:
1183         close_bdev_exclusive(sd.bdev, flags);
1184
1185         return err;
1186
1187  cancel_new:
1188         /* Abandoning the newly allocated superblock */
1189         mutex_unlock(&nilfs->ns_mount_mutex);
1190         put_nilfs(nilfs);
1191         up_write(&s->s_umount);
1192         deactivate_super(s);
1193         /*
1194          * deactivate_super() invokes close_bdev_exclusive().
1195          * We must finish all post-cleaning before this call;
1196          * put_nilfs() needs the block device.
1197          */
1198         return err;
1199 }
1200
1201 struct file_system_type nilfs_fs_type = {
1202         .owner    = THIS_MODULE,
1203         .name     = "nilfs2",
1204         .get_sb   = nilfs_get_sb,
1205         .kill_sb  = kill_block_super,
1206         .fs_flags = FS_REQUIRES_DEV,
1207 };
1208
1209 static int __init init_nilfs_fs(void)
1210 {
1211         int err;
1212
1213         err = nilfs_init_inode_cache();
1214         if (err)
1215                 goto failed;
1216
1217         err = nilfs_init_transaction_cache();
1218         if (err)
1219                 goto failed_inode_cache;
1220
1221         err = nilfs_init_segbuf_cache();
1222         if (err)
1223                 goto failed_transaction_cache;
1224
1225         err = nilfs_btree_path_cache_init();
1226         if (err)
1227                 goto failed_segbuf_cache;
1228
1229         err = register_filesystem(&nilfs_fs_type);
1230         if (err)
1231                 goto failed_btree_path_cache;
1232
1233         return 0;
1234
1235  failed_btree_path_cache:
1236         nilfs_btree_path_cache_destroy();
1237
1238  failed_segbuf_cache:
1239         nilfs_destroy_segbuf_cache();
1240
1241  failed_transaction_cache:
1242         nilfs_destroy_transaction_cache();
1243
1244  failed_inode_cache:
1245         nilfs_destroy_inode_cache();
1246
1247  failed:
1248         return err;
1249 }
1250
1251 static void __exit exit_nilfs_fs(void)
1252 {
1253         nilfs_destroy_segbuf_cache();
1254         nilfs_destroy_transaction_cache();
1255         nilfs_destroy_inode_cache();
1256         nilfs_btree_path_cache_destroy();
1257         unregister_filesystem(&nilfs_fs_type);
1258 }
1259
1260 module_init(init_nilfs_fs)
1261 module_exit(exit_nilfs_fs)