nilfs2: clean up nilfs_write_super
[safe/jmp/linux-2.6] / fs / nilfs2 / super.c
1 /*
2  * super.c - NILFS module and super block management.
3  *
4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
19  *
20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
21  */
22 /*
23  *  linux/fs/ext2/super.c
24  *
25  * Copyright (C) 1992, 1993, 1994, 1995
26  * Remy Card (card@masi.ibp.fr)
27  * Laboratoire MASI - Institut Blaise Pascal
28  * Universite Pierre et Marie Curie (Paris VI)
29  *
30  *  from
31  *
32  *  linux/fs/minix/inode.c
33  *
34  *  Copyright (C) 1991, 1992  Linus Torvalds
35  *
36  *  Big-endian to little-endian byte-swapping/bitmaps by
37  *        David S. Miller (davem@caip.rutgers.edu), 1995
38  */
39
40 #include <linux/module.h>
41 #include <linux/string.h>
42 #include <linux/slab.h>
43 #include <linux/init.h>
44 #include <linux/blkdev.h>
45 #include <linux/parser.h>
46 #include <linux/random.h>
47 #include <linux/crc32.h>
48 #include <linux/smp_lock.h>
49 #include <linux/vfs.h>
50 #include <linux/writeback.h>
51 #include <linux/kobject.h>
52 #include <linux/exportfs.h>
53 #include <linux/seq_file.h>
54 #include <linux/mount.h>
55 #include "nilfs.h"
56 #include "mdt.h"
57 #include "alloc.h"
58 #include "page.h"
59 #include "cpfile.h"
60 #include "ifile.h"
61 #include "dat.h"
62 #include "segment.h"
63 #include "segbuf.h"
64
65 MODULE_AUTHOR("NTT Corp.");
66 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
67                    "(NILFS)");
68 MODULE_LICENSE("GPL");
69
70 static void nilfs_write_super(struct super_block *sb);
71 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
72
73 /**
74  * nilfs_error() - report failure condition on a filesystem
75  *
76  * nilfs_error() sets an ERROR_FS flag on the superblock as well as
77  * reporting an error message.  It should be called when NILFS detects
78  * incoherences or defects of meta data on disk.  As for sustainable
79  * errors such as a single-shot I/O error, nilfs_warning() or the printk()
80  * function should be used instead.
81  *
82  * The segment constructor must not call this function because it can
83  * kill itself.
84  */
85 void nilfs_error(struct super_block *sb, const char *function,
86                  const char *fmt, ...)
87 {
88         struct nilfs_sb_info *sbi = NILFS_SB(sb);
89         va_list args;
90
91         va_start(args, fmt);
92         printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
93         vprintk(fmt, args);
94         printk("\n");
95         va_end(args);
96
97         if (!(sb->s_flags & MS_RDONLY)) {
98                 struct the_nilfs *nilfs = sbi->s_nilfs;
99
100                 if (!nilfs_test_opt(sbi, ERRORS_CONT))
101                         nilfs_detach_segment_constructor(sbi);
102
103                 down_write(&nilfs->ns_sem);
104                 if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
105                         nilfs->ns_mount_state |= NILFS_ERROR_FS;
106                         nilfs->ns_sbp[0]->s_state |=
107                                 cpu_to_le16(NILFS_ERROR_FS);
108                         nilfs_commit_super(sbi, 1);
109                 }
110                 up_write(&nilfs->ns_sem);
111
112                 if (nilfs_test_opt(sbi, ERRORS_RO)) {
113                         printk(KERN_CRIT "Remounting filesystem read-only\n");
114                         sb->s_flags |= MS_RDONLY;
115                 }
116         }
117
118         if (nilfs_test_opt(sbi, ERRORS_PANIC))
119                 panic("NILFS (device %s): panic forced after error\n",
120                       sb->s_id);
121 }
122
123 void nilfs_warning(struct super_block *sb, const char *function,
124                    const char *fmt, ...)
125 {
126         va_list args;
127
128         va_start(args, fmt);
129         printk(KERN_WARNING "NILFS warning (device %s): %s: ",
130                sb->s_id, function);
131         vprintk(fmt, args);
132         printk("\n");
133         va_end(args);
134 }
135
136 static struct kmem_cache *nilfs_inode_cachep;
137
138 struct inode *nilfs_alloc_inode_common(struct the_nilfs *nilfs)
139 {
140         struct nilfs_inode_info *ii;
141
142         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
143         if (!ii)
144                 return NULL;
145         ii->i_bh = NULL;
146         ii->i_state = 0;
147         ii->vfs_inode.i_version = 1;
148         nilfs_btnode_cache_init(&ii->i_btnode_cache, nilfs->ns_bdi);
149         return &ii->vfs_inode;
150 }
151
152 struct inode *nilfs_alloc_inode(struct super_block *sb)
153 {
154         return nilfs_alloc_inode_common(NILFS_SB(sb)->s_nilfs);
155 }
156
157 void nilfs_destroy_inode(struct inode *inode)
158 {
159         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
160 }
161
162 static void init_once(void *obj)
163 {
164         struct nilfs_inode_info *ii = obj;
165
166         INIT_LIST_HEAD(&ii->i_dirty);
167 #ifdef CONFIG_NILFS_XATTR
168         init_rwsem(&ii->xattr_sem);
169 #endif
170         nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
171         ii->i_bmap = (struct nilfs_bmap *)&ii->i_bmap_union;
172         inode_init_once(&ii->vfs_inode);
173 }
174
175 static int nilfs_init_inode_cache(void)
176 {
177         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
178                                                sizeof(struct nilfs_inode_info),
179                                                0, SLAB_RECLAIM_ACCOUNT,
180                                                init_once);
181
182         return (nilfs_inode_cachep == NULL) ? -ENOMEM : 0;
183 }
184
185 static inline void nilfs_destroy_inode_cache(void)
186 {
187         kmem_cache_destroy(nilfs_inode_cachep);
188 }
189
190 static void nilfs_clear_inode(struct inode *inode)
191 {
192         struct nilfs_inode_info *ii = NILFS_I(inode);
193
194         /*
195          * Free resources allocated in nilfs_read_inode(), here.
196          */
197         BUG_ON(!list_empty(&ii->i_dirty));
198         brelse(ii->i_bh);
199         ii->i_bh = NULL;
200
201         if (test_bit(NILFS_I_BMAP, &ii->i_state))
202                 nilfs_bmap_clear(ii->i_bmap);
203
204         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
205 }
206
207 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int dupsb)
208 {
209         struct the_nilfs *nilfs = sbi->s_nilfs;
210         int err;
211         int barrier_done = 0;
212
213         if (nilfs_test_opt(sbi, BARRIER)) {
214                 set_buffer_ordered(nilfs->ns_sbh[0]);
215                 barrier_done = 1;
216         }
217  retry:
218         set_buffer_dirty(nilfs->ns_sbh[0]);
219         err = sync_dirty_buffer(nilfs->ns_sbh[0]);
220         if (err == -EOPNOTSUPP && barrier_done) {
221                 nilfs_warning(sbi->s_super, __func__,
222                               "barrier-based sync failed. "
223                               "disabling barriers\n");
224                 nilfs_clear_opt(sbi, BARRIER);
225                 barrier_done = 0;
226                 clear_buffer_ordered(nilfs->ns_sbh[0]);
227                 goto retry;
228         }
229         if (unlikely(err)) {
230                 printk(KERN_ERR
231                        "NILFS: unable to write superblock (err=%d)\n", err);
232                 if (err == -EIO && nilfs->ns_sbh[1]) {
233                         nilfs_fall_back_super_block(nilfs);
234                         goto retry;
235                 }
236         } else {
237                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
238
239                 /*
240                  * The latest segment becomes trailable from the position
241                  * written in superblock.
242                  */
243                 clear_nilfs_discontinued(nilfs);
244
245                 /* update GC protection for recent segments */
246                 if (nilfs->ns_sbh[1]) {
247                         sbp = NULL;
248                         if (dupsb) {
249                                 set_buffer_dirty(nilfs->ns_sbh[1]);
250                                 if (!sync_dirty_buffer(nilfs->ns_sbh[1]))
251                                         sbp = nilfs->ns_sbp[1];
252                         }
253                 }
254                 if (sbp) {
255                         spin_lock(&nilfs->ns_last_segment_lock);
256                         nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
257                         spin_unlock(&nilfs->ns_last_segment_lock);
258                 }
259         }
260
261         return err;
262 }
263
264 int nilfs_commit_super(struct nilfs_sb_info *sbi, int dupsb)
265 {
266         struct the_nilfs *nilfs = sbi->s_nilfs;
267         struct nilfs_super_block **sbp = nilfs->ns_sbp;
268         sector_t nfreeblocks;
269         time_t t;
270         int err;
271
272         /* nilfs->sem must be locked by the caller. */
273         if (sbp[0]->s_magic != NILFS_SUPER_MAGIC) {
274                 if (sbp[1] && sbp[1]->s_magic == NILFS_SUPER_MAGIC)
275                         nilfs_swap_super_block(nilfs);
276                 else {
277                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
278                                sbi->s_super->s_id);
279                         return -EIO;
280                 }
281         }
282         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
283         if (unlikely(err)) {
284                 printk(KERN_ERR "NILFS: failed to count free blocks\n");
285                 return err;
286         }
287         spin_lock(&nilfs->ns_last_segment_lock);
288         sbp[0]->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
289         sbp[0]->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
290         sbp[0]->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
291         spin_unlock(&nilfs->ns_last_segment_lock);
292
293         t = get_seconds();
294         nilfs->ns_sbwtime[0] = t;
295         sbp[0]->s_free_blocks_count = cpu_to_le64(nfreeblocks);
296         sbp[0]->s_wtime = cpu_to_le64(t);
297         sbp[0]->s_sum = 0;
298         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
299                                              (unsigned char *)sbp[0],
300                                              nilfs->ns_sbsize));
301         if (dupsb && sbp[1]) {
302                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
303                 nilfs->ns_sbwtime[1] = t;
304         }
305         sbi->s_super->s_dirt = 0;
306         return nilfs_sync_super(sbi, dupsb);
307 }
308
309 static void nilfs_put_super(struct super_block *sb)
310 {
311         struct nilfs_sb_info *sbi = NILFS_SB(sb);
312         struct the_nilfs *nilfs = sbi->s_nilfs;
313
314         lock_kernel();
315
316         nilfs_detach_segment_constructor(sbi);
317
318         if (!(sb->s_flags & MS_RDONLY)) {
319                 down_write(&nilfs->ns_sem);
320                 nilfs->ns_sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
321                 nilfs_commit_super(sbi, 1);
322                 up_write(&nilfs->ns_sem);
323         }
324         down_write(&nilfs->ns_super_sem);
325         if (nilfs->ns_current == sbi)
326                 nilfs->ns_current = NULL;
327         up_write(&nilfs->ns_super_sem);
328
329         nilfs_detach_checkpoint(sbi);
330         put_nilfs(sbi->s_nilfs);
331         sbi->s_super = NULL;
332         sb->s_fs_info = NULL;
333         nilfs_put_sbinfo(sbi);
334
335         unlock_kernel();
336 }
337
338 /**
339  * nilfs_write_super - write super block(s) of NILFS
340  * @sb: super_block
341  *
342  * nilfs_write_super() gets a fs-dependent lock, writes super block(s), and
343  * clears s_dirt.  This function is called in the section protected by
344  * lock_super().
345  *
346  * The s_dirt flag is managed by each filesystem and we protect it by ns_sem
347  * of the struct the_nilfs.  Lock order must be as follows:
348  *
349  *   1. lock_super()
350  *   2.    down_write(&nilfs->ns_sem)
351  *
352  * Inside NILFS, locking ns_sem is enough to protect s_dirt and the buffer
353  * of the super block (nilfs->ns_sbp[]).
354  *
355  * In most cases, VFS functions call lock_super() before calling these
356  * methods.  So we must be careful not to bring on deadlocks when using
357  * lock_super();  see generic_shutdown_super(), write_super(), and so on.
358  *
359  * Note that order of lock_kernel() and lock_super() depends on contexts
360  * of VFS.  We should also note that lock_kernel() can be used in its
361  * protective section and only the outermost one has an effect.
362  */
363 static void nilfs_write_super(struct super_block *sb)
364 {
365         struct nilfs_sb_info *sbi = NILFS_SB(sb);
366         struct the_nilfs *nilfs = sbi->s_nilfs;
367
368         down_write(&nilfs->ns_sem);
369         if (!(sb->s_flags & MS_RDONLY)) {
370                 if (!nilfs_discontinued(nilfs) &&
371                     !nilfs_sb_need_update(nilfs)) {
372                         up_write(&nilfs->ns_sem);
373                         return;
374                 }
375                 nilfs_commit_super(sbi, nilfs_altsb_need_update(nilfs));
376         }
377         sb->s_dirt = 0;
378         up_write(&nilfs->ns_sem);
379 }
380
381 static int nilfs_sync_fs(struct super_block *sb, int wait)
382 {
383         struct nilfs_sb_info *sbi = NILFS_SB(sb);
384         struct the_nilfs *nilfs = sbi->s_nilfs;
385         int err = 0;
386
387         /* This function is called when super block should be written back */
388         if (wait)
389                 err = nilfs_construct_segment(sb);
390
391         down_write(&nilfs->ns_sem);
392         if (sb->s_dirt)
393                 nilfs_commit_super(sbi, 1);
394         up_write(&nilfs->ns_sem);
395
396         return err;
397 }
398
399 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
400 {
401         struct the_nilfs *nilfs = sbi->s_nilfs;
402         struct nilfs_checkpoint *raw_cp;
403         struct buffer_head *bh_cp;
404         int err;
405
406         down_write(&nilfs->ns_super_sem);
407         list_add(&sbi->s_list, &nilfs->ns_supers);
408         up_write(&nilfs->ns_super_sem);
409
410         sbi->s_ifile = nilfs_mdt_new(
411                 nilfs, sbi->s_super, NILFS_IFILE_INO, NILFS_IFILE_GFP);
412         if (!sbi->s_ifile)
413                 return -ENOMEM;
414
415         err = nilfs_palloc_init_blockgroup(sbi->s_ifile, nilfs->ns_inode_size);
416         if (unlikely(err))
417                 goto failed;
418
419         down_read(&nilfs->ns_segctor_sem);
420         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
421                                           &bh_cp);
422         up_read(&nilfs->ns_segctor_sem);
423         if (unlikely(err)) {
424                 if (err == -ENOENT || err == -EINVAL) {
425                         printk(KERN_ERR
426                                "NILFS: Invalid checkpoint "
427                                "(checkpoint number=%llu)\n",
428                                (unsigned long long)cno);
429                         err = -EINVAL;
430                 }
431                 goto failed;
432         }
433         err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
434         if (unlikely(err))
435                 goto failed_bh;
436         atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
437         atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
438
439         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
440         return 0;
441
442  failed_bh:
443         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
444  failed:
445         nilfs_mdt_destroy(sbi->s_ifile);
446         sbi->s_ifile = NULL;
447
448         down_write(&nilfs->ns_super_sem);
449         list_del_init(&sbi->s_list);
450         up_write(&nilfs->ns_super_sem);
451
452         return err;
453 }
454
455 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
456 {
457         struct the_nilfs *nilfs = sbi->s_nilfs;
458
459         nilfs_mdt_clear(sbi->s_ifile);
460         nilfs_mdt_destroy(sbi->s_ifile);
461         sbi->s_ifile = NULL;
462         down_write(&nilfs->ns_super_sem);
463         list_del_init(&sbi->s_list);
464         up_write(&nilfs->ns_super_sem);
465 }
466
467 static int nilfs_mark_recovery_complete(struct nilfs_sb_info *sbi)
468 {
469         struct the_nilfs *nilfs = sbi->s_nilfs;
470         int err = 0;
471
472         down_write(&nilfs->ns_sem);
473         if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
474                 nilfs->ns_mount_state |= NILFS_VALID_FS;
475                 err = nilfs_commit_super(sbi, 1);
476                 if (likely(!err))
477                         printk(KERN_INFO "NILFS: recovery complete.\n");
478         }
479         up_write(&nilfs->ns_sem);
480         return err;
481 }
482
483 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
484 {
485         struct super_block *sb = dentry->d_sb;
486         struct nilfs_sb_info *sbi = NILFS_SB(sb);
487         struct the_nilfs *nilfs = sbi->s_nilfs;
488         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
489         unsigned long long blocks;
490         unsigned long overhead;
491         unsigned long nrsvblocks;
492         sector_t nfreeblocks;
493         int err;
494
495         /*
496          * Compute all of the segment blocks
497          *
498          * The blocks before first segment and after last segment
499          * are excluded.
500          */
501         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
502                 - nilfs->ns_first_data_block;
503         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
504
505         /*
506          * Compute the overhead
507          *
508          * When distributing meta data blocks outside semgent structure,
509          * We must count them as the overhead.
510          */
511         overhead = 0;
512
513         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
514         if (unlikely(err))
515                 return err;
516
517         buf->f_type = NILFS_SUPER_MAGIC;
518         buf->f_bsize = sb->s_blocksize;
519         buf->f_blocks = blocks - overhead;
520         buf->f_bfree = nfreeblocks;
521         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
522                 (buf->f_bfree - nrsvblocks) : 0;
523         buf->f_files = atomic_read(&sbi->s_inodes_count);
524         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
525         buf->f_namelen = NILFS_NAME_LEN;
526         buf->f_fsid.val[0] = (u32)id;
527         buf->f_fsid.val[1] = (u32)(id >> 32);
528
529         return 0;
530 }
531
532 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
533 {
534         struct super_block *sb = vfs->mnt_sb;
535         struct nilfs_sb_info *sbi = NILFS_SB(sb);
536
537         if (!nilfs_test_opt(sbi, BARRIER))
538                 seq_printf(seq, ",barrier=off");
539         if (nilfs_test_opt(sbi, SNAPSHOT))
540                 seq_printf(seq, ",cp=%llu",
541                            (unsigned long long int)sbi->s_snapshot_cno);
542         if (nilfs_test_opt(sbi, ERRORS_RO))
543                 seq_printf(seq, ",errors=remount-ro");
544         if (nilfs_test_opt(sbi, ERRORS_PANIC))
545                 seq_printf(seq, ",errors=panic");
546         if (nilfs_test_opt(sbi, STRICT_ORDER))
547                 seq_printf(seq, ",order=strict");
548
549         return 0;
550 }
551
552 static struct super_operations nilfs_sops = {
553         .alloc_inode    = nilfs_alloc_inode,
554         .destroy_inode  = nilfs_destroy_inode,
555         .dirty_inode    = nilfs_dirty_inode,
556         /* .write_inode    = nilfs_write_inode, */
557         /* .put_inode      = nilfs_put_inode, */
558         /* .drop_inode    = nilfs_drop_inode, */
559         .delete_inode   = nilfs_delete_inode,
560         .put_super      = nilfs_put_super,
561         .write_super    = nilfs_write_super,
562         .sync_fs        = nilfs_sync_fs,
563         /* .write_super_lockfs */
564         /* .unlockfs */
565         .statfs         = nilfs_statfs,
566         .remount_fs     = nilfs_remount,
567         .clear_inode    = nilfs_clear_inode,
568         /* .umount_begin */
569         .show_options = nilfs_show_options
570 };
571
572 static struct inode *
573 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
574 {
575         struct inode *inode;
576
577         if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
578             ino != NILFS_SKETCH_INO)
579                 return ERR_PTR(-ESTALE);
580
581         inode = nilfs_iget(sb, ino);
582         if (IS_ERR(inode))
583                 return ERR_CAST(inode);
584         if (generation && inode->i_generation != generation) {
585                 iput(inode);
586                 return ERR_PTR(-ESTALE);
587         }
588
589         return inode;
590 }
591
592 static struct dentry *
593 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
594                    int fh_type)
595 {
596         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
597                                     nilfs_nfs_get_inode);
598 }
599
600 static struct dentry *
601 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
602                    int fh_type)
603 {
604         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
605                                     nilfs_nfs_get_inode);
606 }
607
608 static struct export_operations nilfs_export_ops = {
609         .fh_to_dentry = nilfs_fh_to_dentry,
610         .fh_to_parent = nilfs_fh_to_parent,
611         .get_parent = nilfs_get_parent,
612 };
613
614 enum {
615         Opt_err_cont, Opt_err_panic, Opt_err_ro,
616         Opt_barrier, Opt_snapshot, Opt_order,
617         Opt_err,
618 };
619
620 static match_table_t tokens = {
621         {Opt_err_cont, "errors=continue"},
622         {Opt_err_panic, "errors=panic"},
623         {Opt_err_ro, "errors=remount-ro"},
624         {Opt_barrier, "barrier=%s"},
625         {Opt_snapshot, "cp=%u"},
626         {Opt_order, "order=%s"},
627         {Opt_err, NULL}
628 };
629
630 static int match_bool(substring_t *s, int *result)
631 {
632         int len = s->to - s->from;
633
634         if (strncmp(s->from, "on", len) == 0)
635                 *result = 1;
636         else if (strncmp(s->from, "off", len) == 0)
637                 *result = 0;
638         else
639                 return 1;
640         return 0;
641 }
642
643 static int parse_options(char *options, struct super_block *sb)
644 {
645         struct nilfs_sb_info *sbi = NILFS_SB(sb);
646         char *p;
647         substring_t args[MAX_OPT_ARGS];
648         int option;
649
650         if (!options)
651                 return 1;
652
653         while ((p = strsep(&options, ",")) != NULL) {
654                 int token;
655                 if (!*p)
656                         continue;
657
658                 token = match_token(p, tokens, args);
659                 switch (token) {
660                 case Opt_barrier:
661                         if (match_bool(&args[0], &option))
662                                 return 0;
663                         if (option)
664                                 nilfs_set_opt(sbi, BARRIER);
665                         else
666                                 nilfs_clear_opt(sbi, BARRIER);
667                         break;
668                 case Opt_order:
669                         if (strcmp(args[0].from, "relaxed") == 0)
670                                 /* Ordered data semantics */
671                                 nilfs_clear_opt(sbi, STRICT_ORDER);
672                         else if (strcmp(args[0].from, "strict") == 0)
673                                 /* Strict in-order semantics */
674                                 nilfs_set_opt(sbi, STRICT_ORDER);
675                         else
676                                 return 0;
677                         break;
678                 case Opt_err_panic:
679                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
680                         break;
681                 case Opt_err_ro:
682                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
683                         break;
684                 case Opt_err_cont:
685                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
686                         break;
687                 case Opt_snapshot:
688                         if (match_int(&args[0], &option) || option <= 0)
689                                 return 0;
690                         if (!(sb->s_flags & MS_RDONLY))
691                                 return 0;
692                         sbi->s_snapshot_cno = option;
693                         nilfs_set_opt(sbi, SNAPSHOT);
694                         break;
695                 default:
696                         printk(KERN_ERR
697                                "NILFS: Unrecognized mount option \"%s\"\n", p);
698                         return 0;
699                 }
700         }
701         return 1;
702 }
703
704 static inline void
705 nilfs_set_default_options(struct nilfs_sb_info *sbi,
706                           struct nilfs_super_block *sbp)
707 {
708         sbi->s_mount_opt =
709                 NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER;
710 }
711
712 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
713 {
714         struct the_nilfs *nilfs = sbi->s_nilfs;
715         struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
716         int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
717         int mnt_count = le16_to_cpu(sbp->s_mnt_count);
718
719         /* nilfs->sem must be locked by the caller. */
720         if (!(nilfs->ns_mount_state & NILFS_VALID_FS)) {
721                 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
722         } else if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
723                 printk(KERN_WARNING
724                        "NILFS warning: mounting fs with errors\n");
725 #if 0
726         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
727                 printk(KERN_WARNING
728                        "NILFS warning: maximal mount count reached\n");
729 #endif
730         }
731         if (!max_mnt_count)
732                 sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
733
734         sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
735         sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
736         sbp->s_mtime = cpu_to_le64(get_seconds());
737         return nilfs_commit_super(sbi, 1);
738 }
739
740 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
741                                                  u64 pos, int blocksize,
742                                                  struct buffer_head **pbh)
743 {
744         unsigned long long sb_index = pos;
745         unsigned long offset;
746
747         offset = do_div(sb_index, blocksize);
748         *pbh = sb_bread(sb, sb_index);
749         if (!*pbh)
750                 return NULL;
751         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
752 }
753
754 int nilfs_store_magic_and_option(struct super_block *sb,
755                                  struct nilfs_super_block *sbp,
756                                  char *data)
757 {
758         struct nilfs_sb_info *sbi = NILFS_SB(sb);
759
760         sb->s_magic = le16_to_cpu(sbp->s_magic);
761
762         /* FS independent flags */
763 #ifdef NILFS_ATIME_DISABLE
764         sb->s_flags |= MS_NOATIME;
765 #endif
766
767         nilfs_set_default_options(sbi, sbp);
768
769         sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
770         sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
771         sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
772         sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
773
774         return !parse_options(data, sb) ? -EINVAL : 0 ;
775 }
776
777 /**
778  * nilfs_fill_super() - initialize a super block instance
779  * @sb: super_block
780  * @data: mount options
781  * @silent: silent mode flag
782  * @nilfs: the_nilfs struct
783  *
784  * This function is called exclusively by nilfs->ns_mount_mutex.
785  * So, the recovery process is protected from other simultaneous mounts.
786  */
787 static int
788 nilfs_fill_super(struct super_block *sb, void *data, int silent,
789                  struct the_nilfs *nilfs)
790 {
791         struct nilfs_sb_info *sbi;
792         struct inode *root;
793         __u64 cno;
794         int err;
795
796         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
797         if (!sbi)
798                 return -ENOMEM;
799
800         sb->s_fs_info = sbi;
801
802         get_nilfs(nilfs);
803         sbi->s_nilfs = nilfs;
804         sbi->s_super = sb;
805         atomic_set(&sbi->s_count, 1);
806
807         err = init_nilfs(nilfs, sbi, (char *)data);
808         if (err)
809                 goto failed_sbi;
810
811         spin_lock_init(&sbi->s_inode_lock);
812         INIT_LIST_HEAD(&sbi->s_dirty_files);
813         INIT_LIST_HEAD(&sbi->s_list);
814
815         /*
816          * Following initialization is overlapped because
817          * nilfs_sb_info structure has been cleared at the beginning.
818          * But we reserve them to keep our interest and make ready
819          * for the future change.
820          */
821         get_random_bytes(&sbi->s_next_generation,
822                          sizeof(sbi->s_next_generation));
823         spin_lock_init(&sbi->s_next_gen_lock);
824
825         sb->s_op = &nilfs_sops;
826         sb->s_export_op = &nilfs_export_ops;
827         sb->s_root = NULL;
828         sb->s_time_gran = 1;
829
830         if (!nilfs_loaded(nilfs)) {
831                 err = load_nilfs(nilfs, sbi);
832                 if (err)
833                         goto failed_sbi;
834         }
835         cno = nilfs_last_cno(nilfs);
836
837         if (sb->s_flags & MS_RDONLY) {
838                 if (nilfs_test_opt(sbi, SNAPSHOT)) {
839                         err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
840                                                        sbi->s_snapshot_cno);
841                         if (err < 0)
842                                 goto failed_sbi;
843                         if (!err) {
844                                 printk(KERN_ERR
845                                        "NILFS: The specified checkpoint is "
846                                        "not a snapshot "
847                                        "(checkpoint number=%llu).\n",
848                                        (unsigned long long)sbi->s_snapshot_cno);
849                                 err = -EINVAL;
850                                 goto failed_sbi;
851                         }
852                         cno = sbi->s_snapshot_cno;
853                 } else
854                         /* Read-only mount */
855                         sbi->s_snapshot_cno = cno;
856         }
857
858         err = nilfs_attach_checkpoint(sbi, cno);
859         if (err) {
860                 printk(KERN_ERR "NILFS: error loading a checkpoint"
861                        " (checkpoint number=%llu).\n", (unsigned long long)cno);
862                 goto failed_sbi;
863         }
864
865         if (!(sb->s_flags & MS_RDONLY)) {
866                 err = nilfs_attach_segment_constructor(sbi);
867                 if (err)
868                         goto failed_checkpoint;
869         }
870
871         root = nilfs_iget(sb, NILFS_ROOT_INO);
872         if (IS_ERR(root)) {
873                 printk(KERN_ERR "NILFS: get root inode failed\n");
874                 err = PTR_ERR(root);
875                 goto failed_segctor;
876         }
877         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
878                 iput(root);
879                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
880                 err = -EINVAL;
881                 goto failed_segctor;
882         }
883         sb->s_root = d_alloc_root(root);
884         if (!sb->s_root) {
885                 iput(root);
886                 printk(KERN_ERR "NILFS: get root dentry failed\n");
887                 err = -ENOMEM;
888                 goto failed_segctor;
889         }
890
891         if (!(sb->s_flags & MS_RDONLY)) {
892                 down_write(&nilfs->ns_sem);
893                 nilfs_setup_super(sbi);
894                 up_write(&nilfs->ns_sem);
895         }
896
897         err = nilfs_mark_recovery_complete(sbi);
898         if (unlikely(err)) {
899                 printk(KERN_ERR "NILFS: recovery failed.\n");
900                 goto failed_root;
901         }
902
903         down_write(&nilfs->ns_super_sem);
904         if (!nilfs_test_opt(sbi, SNAPSHOT))
905                 nilfs->ns_current = sbi;
906         up_write(&nilfs->ns_super_sem);
907
908         return 0;
909
910  failed_root:
911         dput(sb->s_root);
912         sb->s_root = NULL;
913
914  failed_segctor:
915         nilfs_detach_segment_constructor(sbi);
916
917  failed_checkpoint:
918         nilfs_detach_checkpoint(sbi);
919
920  failed_sbi:
921         put_nilfs(nilfs);
922         sb->s_fs_info = NULL;
923         nilfs_put_sbinfo(sbi);
924         return err;
925 }
926
927 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
928 {
929         struct nilfs_sb_info *sbi = NILFS_SB(sb);
930         struct nilfs_super_block *sbp;
931         struct the_nilfs *nilfs = sbi->s_nilfs;
932         unsigned long old_sb_flags;
933         struct nilfs_mount_options old_opts;
934         int err;
935
936         lock_kernel();
937
938         down_write(&nilfs->ns_super_sem);
939         old_sb_flags = sb->s_flags;
940         old_opts.mount_opt = sbi->s_mount_opt;
941         old_opts.snapshot_cno = sbi->s_snapshot_cno;
942
943         if (!parse_options(data, sb)) {
944                 err = -EINVAL;
945                 goto restore_opts;
946         }
947         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
948
949         if ((*flags & MS_RDONLY) &&
950             sbi->s_snapshot_cno != old_opts.snapshot_cno) {
951                 printk(KERN_WARNING "NILFS (device %s): couldn't "
952                        "remount to a different snapshot. \n",
953                        sb->s_id);
954                 err = -EINVAL;
955                 goto restore_opts;
956         }
957
958         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
959                 goto out;
960         if (*flags & MS_RDONLY) {
961                 /* Shutting down the segment constructor */
962                 nilfs_detach_segment_constructor(sbi);
963                 sb->s_flags |= MS_RDONLY;
964
965                 sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
966                 /* nilfs_set_opt(sbi, SNAPSHOT); */
967
968                 /*
969                  * Remounting a valid RW partition RDONLY, so set
970                  * the RDONLY flag and then mark the partition as valid again.
971                  */
972                 down_write(&nilfs->ns_sem);
973                 sbp = nilfs->ns_sbp[0];
974                 if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
975                     (nilfs->ns_mount_state & NILFS_VALID_FS))
976                         sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
977                 sbp->s_mtime = cpu_to_le64(get_seconds());
978                 nilfs_commit_super(sbi, 1);
979                 up_write(&nilfs->ns_sem);
980         } else {
981                 /*
982                  * Mounting a RDONLY partition read-write, so reread and
983                  * store the current valid flag.  (It may have been changed
984                  * by fsck since we originally mounted the partition.)
985                  */
986                 if (nilfs->ns_current && nilfs->ns_current != sbi) {
987                         printk(KERN_WARNING "NILFS (device %s): couldn't "
988                                "remount because an RW-mount exists.\n",
989                                sb->s_id);
990                         err = -EBUSY;
991                         goto restore_opts;
992                 }
993                 if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
994                         printk(KERN_WARNING "NILFS (device %s): couldn't "
995                                "remount because the current RO-mount is not "
996                                "the latest one.\n",
997                                sb->s_id);
998                         err = -EINVAL;
999                         goto restore_opts;
1000                 }
1001                 sb->s_flags &= ~MS_RDONLY;
1002                 nilfs_clear_opt(sbi, SNAPSHOT);
1003                 sbi->s_snapshot_cno = 0;
1004
1005                 err = nilfs_attach_segment_constructor(sbi);
1006                 if (err)
1007                         goto restore_opts;
1008
1009                 down_write(&nilfs->ns_sem);
1010                 nilfs_setup_super(sbi);
1011                 up_write(&nilfs->ns_sem);
1012
1013                 nilfs->ns_current = sbi;
1014         }
1015  out:
1016         up_write(&nilfs->ns_super_sem);
1017         unlock_kernel();
1018         return 0;
1019
1020  restore_opts:
1021         sb->s_flags = old_sb_flags;
1022         sbi->s_mount_opt = old_opts.mount_opt;
1023         sbi->s_snapshot_cno = old_opts.snapshot_cno;
1024         up_write(&nilfs->ns_super_sem);
1025         unlock_kernel();
1026         return err;
1027 }
1028
1029 struct nilfs_super_data {
1030         struct block_device *bdev;
1031         struct nilfs_sb_info *sbi;
1032         __u64 cno;
1033         int flags;
1034 };
1035
1036 /**
1037  * nilfs_identify - pre-read mount options needed to identify mount instance
1038  * @data: mount options
1039  * @sd: nilfs_super_data
1040  */
1041 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1042 {
1043         char *p, *options = data;
1044         substring_t args[MAX_OPT_ARGS];
1045         int option, token;
1046         int ret = 0;
1047
1048         do {
1049                 p = strsep(&options, ",");
1050                 if (p != NULL && *p) {
1051                         token = match_token(p, tokens, args);
1052                         if (token == Opt_snapshot) {
1053                                 if (!(sd->flags & MS_RDONLY))
1054                                         ret++;
1055                                 else {
1056                                         ret = match_int(&args[0], &option);
1057                                         if (!ret) {
1058                                                 if (option > 0)
1059                                                         sd->cno = option;
1060                                                 else
1061                                                         ret++;
1062                                         }
1063                                 }
1064                         }
1065                         if (ret)
1066                                 printk(KERN_ERR
1067                                        "NILFS: invalid mount option: %s\n", p);
1068                 }
1069                 if (!options)
1070                         break;
1071                 BUG_ON(options == data);
1072                 *(options - 1) = ',';
1073         } while (!ret);
1074         return ret;
1075 }
1076
1077 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1078 {
1079         struct nilfs_super_data *sd = data;
1080
1081         s->s_bdev = sd->bdev;
1082         s->s_dev = s->s_bdev->bd_dev;
1083         return 0;
1084 }
1085
1086 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1087 {
1088         struct nilfs_super_data *sd = data;
1089
1090         return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1091 }
1092
1093 static int
1094 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1095              const char *dev_name, void *data, struct vfsmount *mnt)
1096 {
1097         struct nilfs_super_data sd;
1098         struct super_block *s;
1099         struct the_nilfs *nilfs;
1100         int err, need_to_close = 1;
1101
1102         sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1103         if (IS_ERR(sd.bdev))
1104                 return PTR_ERR(sd.bdev);
1105
1106         /*
1107          * To get mount instance using sget() vfs-routine, NILFS needs
1108          * much more information than normal filesystems to identify mount
1109          * instance.  For snapshot mounts, not only a mount type (ro-mount
1110          * or rw-mount) but also a checkpoint number is required.
1111          */
1112         sd.cno = 0;
1113         sd.flags = flags;
1114         if (nilfs_identify((char *)data, &sd)) {
1115                 err = -EINVAL;
1116                 goto failed;
1117         }
1118
1119         nilfs = find_or_create_nilfs(sd.bdev);
1120         if (!nilfs) {
1121                 err = -ENOMEM;
1122                 goto failed;
1123         }
1124
1125         mutex_lock(&nilfs->ns_mount_mutex);
1126
1127         if (!sd.cno) {
1128                 /*
1129                  * Check if an exclusive mount exists or not.
1130                  * Snapshot mounts coexist with a current mount
1131                  * (i.e. rw-mount or ro-mount), whereas rw-mount and
1132                  * ro-mount are mutually exclusive.
1133                  */
1134                 down_read(&nilfs->ns_super_sem);
1135                 if (nilfs->ns_current &&
1136                     ((nilfs->ns_current->s_super->s_flags ^ flags)
1137                      & MS_RDONLY)) {
1138                         up_read(&nilfs->ns_super_sem);
1139                         err = -EBUSY;
1140                         goto failed_unlock;
1141                 }
1142                 up_read(&nilfs->ns_super_sem);
1143         }
1144
1145         /*
1146          * Find existing nilfs_sb_info struct
1147          */
1148         sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1149
1150         if (!sd.cno)
1151                 /* trying to get the latest checkpoint.  */
1152                 sd.cno = nilfs_last_cno(nilfs);
1153
1154         /*
1155          * Get super block instance holding the nilfs_sb_info struct.
1156          * A new instance is allocated if no existing mount is present or
1157          * existing instance has been unmounted.
1158          */
1159         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1160         if (sd.sbi)
1161                 nilfs_put_sbinfo(sd.sbi);
1162
1163         if (IS_ERR(s)) {
1164                 err = PTR_ERR(s);
1165                 goto failed_unlock;
1166         }
1167
1168         if (!s->s_root) {
1169                 char b[BDEVNAME_SIZE];
1170
1171                 /* New superblock instance created */
1172                 s->s_flags = flags;
1173                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1174                 sb_set_blocksize(s, block_size(sd.bdev));
1175
1176                 err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1177                 if (err)
1178                         goto cancel_new;
1179
1180                 s->s_flags |= MS_ACTIVE;
1181                 need_to_close = 0;
1182         }
1183
1184         mutex_unlock(&nilfs->ns_mount_mutex);
1185         put_nilfs(nilfs);
1186         if (need_to_close)
1187                 close_bdev_exclusive(sd.bdev, flags);
1188         simple_set_mnt(mnt, s);
1189         return 0;
1190
1191  failed_unlock:
1192         mutex_unlock(&nilfs->ns_mount_mutex);
1193         put_nilfs(nilfs);
1194  failed:
1195         close_bdev_exclusive(sd.bdev, flags);
1196
1197         return err;
1198
1199  cancel_new:
1200         /* Abandoning the newly allocated superblock */
1201         mutex_unlock(&nilfs->ns_mount_mutex);
1202         put_nilfs(nilfs);
1203         up_write(&s->s_umount);
1204         deactivate_super(s);
1205         /*
1206          * deactivate_super() invokes close_bdev_exclusive().
1207          * We must finish all post-cleaning before this call;
1208          * put_nilfs() needs the block device.
1209          */
1210         return err;
1211 }
1212
1213 struct file_system_type nilfs_fs_type = {
1214         .owner    = THIS_MODULE,
1215         .name     = "nilfs2",
1216         .get_sb   = nilfs_get_sb,
1217         .kill_sb  = kill_block_super,
1218         .fs_flags = FS_REQUIRES_DEV,
1219 };
1220
1221 static int __init init_nilfs_fs(void)
1222 {
1223         int err;
1224
1225         err = nilfs_init_inode_cache();
1226         if (err)
1227                 goto failed;
1228
1229         err = nilfs_init_transaction_cache();
1230         if (err)
1231                 goto failed_inode_cache;
1232
1233         err = nilfs_init_segbuf_cache();
1234         if (err)
1235                 goto failed_transaction_cache;
1236
1237         err = nilfs_btree_path_cache_init();
1238         if (err)
1239                 goto failed_segbuf_cache;
1240
1241         err = register_filesystem(&nilfs_fs_type);
1242         if (err)
1243                 goto failed_btree_path_cache;
1244
1245         return 0;
1246
1247  failed_btree_path_cache:
1248         nilfs_btree_path_cache_destroy();
1249
1250  failed_segbuf_cache:
1251         nilfs_destroy_segbuf_cache();
1252
1253  failed_transaction_cache:
1254         nilfs_destroy_transaction_cache();
1255
1256  failed_inode_cache:
1257         nilfs_destroy_inode_cache();
1258
1259  failed:
1260         return err;
1261 }
1262
1263 static void __exit exit_nilfs_fs(void)
1264 {
1265         nilfs_destroy_segbuf_cache();
1266         nilfs_destroy_transaction_cache();
1267         nilfs_destroy_inode_cache();
1268         nilfs_btree_path_cache_destroy();
1269         unregister_filesystem(&nilfs_fs_type);
1270 }
1271
1272 module_init(init_nilfs_fs)
1273 module_exit(exit_nilfs_fs)