nilfs2: introduce nilfs_prepare_super
[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 <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 struct kmem_cache *nilfs_inode_cachep;
71 struct kmem_cache *nilfs_transaction_cachep;
72 struct kmem_cache *nilfs_segbuf_cachep;
73 struct kmem_cache *nilfs_btree_path_cache;
74
75 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
76
77 static void nilfs_set_error(struct nilfs_sb_info *sbi)
78 {
79         struct the_nilfs *nilfs = sbi->s_nilfs;
80         struct nilfs_super_block **sbp;
81
82         down_write(&nilfs->ns_sem);
83         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
84                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
85                 sbp = nilfs_prepare_super(sbi);
86                 if (likely(sbp)) {
87                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
88                         nilfs_commit_super(sbi, 1);
89                 }
90         }
91         up_write(&nilfs->ns_sem);
92 }
93
94 /**
95  * nilfs_error() - report failure condition on a filesystem
96  *
97  * nilfs_error() sets an ERROR_FS flag on the superblock as well as
98  * reporting an error message.  It should be called when NILFS detects
99  * incoherences or defects of meta data on disk.  As for sustainable
100  * errors such as a single-shot I/O error, nilfs_warning() or the printk()
101  * function should be used instead.
102  *
103  * The segment constructor must not call this function because it can
104  * kill itself.
105  */
106 void nilfs_error(struct super_block *sb, const char *function,
107                  const char *fmt, ...)
108 {
109         struct nilfs_sb_info *sbi = NILFS_SB(sb);
110         va_list args;
111
112         va_start(args, fmt);
113         printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
114         vprintk(fmt, args);
115         printk("\n");
116         va_end(args);
117
118         if (!(sb->s_flags & MS_RDONLY)) {
119                 nilfs_set_error(sbi);
120
121                 if (nilfs_test_opt(sbi, ERRORS_RO)) {
122                         printk(KERN_CRIT "Remounting filesystem read-only\n");
123                         sb->s_flags |= MS_RDONLY;
124                 }
125         }
126
127         if (nilfs_test_opt(sbi, ERRORS_PANIC))
128                 panic("NILFS (device %s): panic forced after error\n",
129                       sb->s_id);
130 }
131
132 void nilfs_warning(struct super_block *sb, const char *function,
133                    const char *fmt, ...)
134 {
135         va_list args;
136
137         va_start(args, fmt);
138         printk(KERN_WARNING "NILFS warning (device %s): %s: ",
139                sb->s_id, function);
140         vprintk(fmt, args);
141         printk("\n");
142         va_end(args);
143 }
144
145
146 struct inode *nilfs_alloc_inode_common(struct the_nilfs *nilfs)
147 {
148         struct nilfs_inode_info *ii;
149
150         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
151         if (!ii)
152                 return NULL;
153         ii->i_bh = NULL;
154         ii->i_state = 0;
155         ii->vfs_inode.i_version = 1;
156         nilfs_btnode_cache_init(&ii->i_btnode_cache, nilfs->ns_bdi);
157         return &ii->vfs_inode;
158 }
159
160 struct inode *nilfs_alloc_inode(struct super_block *sb)
161 {
162         return nilfs_alloc_inode_common(NILFS_SB(sb)->s_nilfs);
163 }
164
165 void nilfs_destroy_inode(struct inode *inode)
166 {
167         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
168 }
169
170 static void nilfs_clear_inode(struct inode *inode)
171 {
172         struct nilfs_inode_info *ii = NILFS_I(inode);
173
174         /*
175          * Free resources allocated in nilfs_read_inode(), here.
176          */
177         BUG_ON(!list_empty(&ii->i_dirty));
178         brelse(ii->i_bh);
179         ii->i_bh = NULL;
180
181         if (test_bit(NILFS_I_BMAP, &ii->i_state))
182                 nilfs_bmap_clear(ii->i_bmap);
183
184         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
185 }
186
187 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int dupsb)
188 {
189         struct the_nilfs *nilfs = sbi->s_nilfs;
190         int err;
191         int barrier_done = 0;
192
193         if (nilfs_test_opt(sbi, BARRIER)) {
194                 set_buffer_ordered(nilfs->ns_sbh[0]);
195                 barrier_done = 1;
196         }
197  retry:
198         set_buffer_dirty(nilfs->ns_sbh[0]);
199         err = sync_dirty_buffer(nilfs->ns_sbh[0]);
200         if (err == -EOPNOTSUPP && barrier_done) {
201                 nilfs_warning(sbi->s_super, __func__,
202                               "barrier-based sync failed. "
203                               "disabling barriers\n");
204                 nilfs_clear_opt(sbi, BARRIER);
205                 barrier_done = 0;
206                 clear_buffer_ordered(nilfs->ns_sbh[0]);
207                 goto retry;
208         }
209         if (unlikely(err)) {
210                 printk(KERN_ERR
211                        "NILFS: unable to write superblock (err=%d)\n", err);
212                 if (err == -EIO && nilfs->ns_sbh[1]) {
213                         nilfs_fall_back_super_block(nilfs);
214                         goto retry;
215                 }
216         } else {
217                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
218
219                 /*
220                  * The latest segment becomes trailable from the position
221                  * written in superblock.
222                  */
223                 clear_nilfs_discontinued(nilfs);
224
225                 /* update GC protection for recent segments */
226                 if (nilfs->ns_sbh[1]) {
227                         sbp = NULL;
228                         if (dupsb) {
229                                 set_buffer_dirty(nilfs->ns_sbh[1]);
230                                 if (!sync_dirty_buffer(nilfs->ns_sbh[1]))
231                                         sbp = nilfs->ns_sbp[1];
232                         }
233                 }
234                 if (sbp) {
235                         spin_lock(&nilfs->ns_last_segment_lock);
236                         nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
237                         spin_unlock(&nilfs->ns_last_segment_lock);
238                 }
239         }
240
241         return err;
242 }
243
244 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
245                           struct the_nilfs *nilfs)
246 {
247         sector_t nfreeblocks;
248
249         /* nilfs->ns_sem must be locked by the caller. */
250         nilfs_count_free_blocks(nilfs, &nfreeblocks);
251         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
252
253         spin_lock(&nilfs->ns_last_segment_lock);
254         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
255         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
256         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
257         spin_unlock(&nilfs->ns_last_segment_lock);
258 }
259
260 struct nilfs_super_block **nilfs_prepare_super(struct nilfs_sb_info *sbi)
261 {
262         struct the_nilfs *nilfs = sbi->s_nilfs;
263         struct nilfs_super_block **sbp = nilfs->ns_sbp;
264
265         /* nilfs->ns_sem must be locked by the caller. */
266         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
267                 if (sbp[1] &&
268                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
269                         nilfs_swap_super_block(nilfs);
270                 } else {
271                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
272                                sbi->s_super->s_id);
273                         return NULL;
274                 }
275         }
276         return sbp;
277 }
278
279 int nilfs_commit_super(struct nilfs_sb_info *sbi, int dupsb)
280 {
281         struct the_nilfs *nilfs = sbi->s_nilfs;
282         struct nilfs_super_block **sbp = nilfs->ns_sbp;
283         time_t t;
284
285         /* nilfs->ns_sem must be locked by the caller. */
286         nilfs_set_log_cursor(sbp[0], nilfs);
287
288         t = get_seconds();
289         nilfs->ns_sbwtime[0] = t;
290         sbp[0]->s_wtime = cpu_to_le64(t);
291         sbp[0]->s_sum = 0;
292         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
293                                              (unsigned char *)sbp[0],
294                                              nilfs->ns_sbsize));
295         if (dupsb && sbp[1]) {
296                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
297                 nilfs->ns_sbwtime[1] = t;
298         }
299         clear_nilfs_sb_dirty(nilfs);
300         return nilfs_sync_super(sbi, dupsb);
301 }
302
303 /**
304  * nilfs_cleanup_super() - write filesystem state for cleanup
305  * @sbi: nilfs_sb_info to be unmounted or degraded to read-only
306  *
307  * This function restores state flags in the on-disk super block.
308  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
309  * filesystem was not clean previously.
310  */
311 int nilfs_cleanup_super(struct nilfs_sb_info *sbi)
312 {
313         struct nilfs_super_block **sbp;
314         int ret = -EIO;
315
316         sbp = nilfs_prepare_super(sbi);
317         if (sbp) {
318                 sbp[0]->s_state = cpu_to_le16(sbi->s_nilfs->ns_mount_state);
319                 ret = nilfs_commit_super(sbi, 1);
320         }
321         return ret;
322 }
323
324 static void nilfs_put_super(struct super_block *sb)
325 {
326         struct nilfs_sb_info *sbi = NILFS_SB(sb);
327         struct the_nilfs *nilfs = sbi->s_nilfs;
328
329         lock_kernel();
330
331         nilfs_detach_segment_constructor(sbi);
332
333         if (!(sb->s_flags & MS_RDONLY)) {
334                 down_write(&nilfs->ns_sem);
335                 nilfs_cleanup_super(sbi);
336                 up_write(&nilfs->ns_sem);
337         }
338         down_write(&nilfs->ns_super_sem);
339         if (nilfs->ns_current == sbi)
340                 nilfs->ns_current = NULL;
341         up_write(&nilfs->ns_super_sem);
342
343         nilfs_detach_checkpoint(sbi);
344         put_nilfs(sbi->s_nilfs);
345         sbi->s_super = NULL;
346         sb->s_fs_info = NULL;
347         nilfs_put_sbinfo(sbi);
348
349         unlock_kernel();
350 }
351
352 static int nilfs_sync_fs(struct super_block *sb, int wait)
353 {
354         struct nilfs_sb_info *sbi = NILFS_SB(sb);
355         struct the_nilfs *nilfs = sbi->s_nilfs;
356         struct nilfs_super_block **sbp;
357         int err = 0;
358
359         /* This function is called when super block should be written back */
360         if (wait)
361                 err = nilfs_construct_segment(sb);
362
363         down_write(&nilfs->ns_sem);
364         if (nilfs_sb_dirty(nilfs)) {
365                 sbp = nilfs_prepare_super(sbi);
366                 if (likely(sbp))
367                         nilfs_commit_super(sbi, 1);
368         }
369         up_write(&nilfs->ns_sem);
370
371         return err;
372 }
373
374 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
375 {
376         struct the_nilfs *nilfs = sbi->s_nilfs;
377         struct nilfs_checkpoint *raw_cp;
378         struct buffer_head *bh_cp;
379         int err;
380
381         down_write(&nilfs->ns_super_sem);
382         list_add(&sbi->s_list, &nilfs->ns_supers);
383         up_write(&nilfs->ns_super_sem);
384
385         sbi->s_ifile = nilfs_ifile_new(sbi, nilfs->ns_inode_size);
386         if (!sbi->s_ifile)
387                 return -ENOMEM;
388
389         down_read(&nilfs->ns_segctor_sem);
390         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
391                                           &bh_cp);
392         up_read(&nilfs->ns_segctor_sem);
393         if (unlikely(err)) {
394                 if (err == -ENOENT || err == -EINVAL) {
395                         printk(KERN_ERR
396                                "NILFS: Invalid checkpoint "
397                                "(checkpoint number=%llu)\n",
398                                (unsigned long long)cno);
399                         err = -EINVAL;
400                 }
401                 goto failed;
402         }
403         err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
404         if (unlikely(err))
405                 goto failed_bh;
406         atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
407         atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
408
409         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
410         return 0;
411
412  failed_bh:
413         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
414  failed:
415         nilfs_mdt_destroy(sbi->s_ifile);
416         sbi->s_ifile = NULL;
417
418         down_write(&nilfs->ns_super_sem);
419         list_del_init(&sbi->s_list);
420         up_write(&nilfs->ns_super_sem);
421
422         return err;
423 }
424
425 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
426 {
427         struct the_nilfs *nilfs = sbi->s_nilfs;
428
429         nilfs_mdt_destroy(sbi->s_ifile);
430         sbi->s_ifile = NULL;
431         down_write(&nilfs->ns_super_sem);
432         list_del_init(&sbi->s_list);
433         up_write(&nilfs->ns_super_sem);
434 }
435
436 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
437 {
438         struct super_block *sb = dentry->d_sb;
439         struct nilfs_sb_info *sbi = NILFS_SB(sb);
440         struct the_nilfs *nilfs = sbi->s_nilfs;
441         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
442         unsigned long long blocks;
443         unsigned long overhead;
444         unsigned long nrsvblocks;
445         sector_t nfreeblocks;
446         int err;
447
448         /*
449          * Compute all of the segment blocks
450          *
451          * The blocks before first segment and after last segment
452          * are excluded.
453          */
454         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
455                 - nilfs->ns_first_data_block;
456         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
457
458         /*
459          * Compute the overhead
460          *
461          * When distributing meta data blocks outside segment structure,
462          * We must count them as the overhead.
463          */
464         overhead = 0;
465
466         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
467         if (unlikely(err))
468                 return err;
469
470         buf->f_type = NILFS_SUPER_MAGIC;
471         buf->f_bsize = sb->s_blocksize;
472         buf->f_blocks = blocks - overhead;
473         buf->f_bfree = nfreeblocks;
474         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
475                 (buf->f_bfree - nrsvblocks) : 0;
476         buf->f_files = atomic_read(&sbi->s_inodes_count);
477         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
478         buf->f_namelen = NILFS_NAME_LEN;
479         buf->f_fsid.val[0] = (u32)id;
480         buf->f_fsid.val[1] = (u32)(id >> 32);
481
482         return 0;
483 }
484
485 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
486 {
487         struct super_block *sb = vfs->mnt_sb;
488         struct nilfs_sb_info *sbi = NILFS_SB(sb);
489
490         if (!nilfs_test_opt(sbi, BARRIER))
491                 seq_printf(seq, ",nobarrier");
492         if (nilfs_test_opt(sbi, SNAPSHOT))
493                 seq_printf(seq, ",cp=%llu",
494                            (unsigned long long int)sbi->s_snapshot_cno);
495         if (nilfs_test_opt(sbi, ERRORS_PANIC))
496                 seq_printf(seq, ",errors=panic");
497         if (nilfs_test_opt(sbi, ERRORS_CONT))
498                 seq_printf(seq, ",errors=continue");
499         if (nilfs_test_opt(sbi, STRICT_ORDER))
500                 seq_printf(seq, ",order=strict");
501         if (nilfs_test_opt(sbi, NORECOVERY))
502                 seq_printf(seq, ",norecovery");
503         if (nilfs_test_opt(sbi, DISCARD))
504                 seq_printf(seq, ",discard");
505
506         return 0;
507 }
508
509 static const struct super_operations nilfs_sops = {
510         .alloc_inode    = nilfs_alloc_inode,
511         .destroy_inode  = nilfs_destroy_inode,
512         .dirty_inode    = nilfs_dirty_inode,
513         /* .write_inode    = nilfs_write_inode, */
514         /* .put_inode      = nilfs_put_inode, */
515         /* .drop_inode    = nilfs_drop_inode, */
516         .delete_inode   = nilfs_delete_inode,
517         .put_super      = nilfs_put_super,
518         /* .write_super    = nilfs_write_super, */
519         .sync_fs        = nilfs_sync_fs,
520         /* .write_super_lockfs */
521         /* .unlockfs */
522         .statfs         = nilfs_statfs,
523         .remount_fs     = nilfs_remount,
524         .clear_inode    = nilfs_clear_inode,
525         /* .umount_begin */
526         .show_options = nilfs_show_options
527 };
528
529 static struct inode *
530 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
531 {
532         struct inode *inode;
533
534         if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
535             ino != NILFS_SKETCH_INO)
536                 return ERR_PTR(-ESTALE);
537
538         inode = nilfs_iget(sb, ino);
539         if (IS_ERR(inode))
540                 return ERR_CAST(inode);
541         if (generation && inode->i_generation != generation) {
542                 iput(inode);
543                 return ERR_PTR(-ESTALE);
544         }
545
546         return inode;
547 }
548
549 static struct dentry *
550 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
551                    int fh_type)
552 {
553         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
554                                     nilfs_nfs_get_inode);
555 }
556
557 static struct dentry *
558 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
559                    int fh_type)
560 {
561         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
562                                     nilfs_nfs_get_inode);
563 }
564
565 static const struct export_operations nilfs_export_ops = {
566         .fh_to_dentry = nilfs_fh_to_dentry,
567         .fh_to_parent = nilfs_fh_to_parent,
568         .get_parent = nilfs_get_parent,
569 };
570
571 enum {
572         Opt_err_cont, Opt_err_panic, Opt_err_ro,
573         Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
574         Opt_discard, Opt_err,
575 };
576
577 static match_table_t tokens = {
578         {Opt_err_cont, "errors=continue"},
579         {Opt_err_panic, "errors=panic"},
580         {Opt_err_ro, "errors=remount-ro"},
581         {Opt_nobarrier, "nobarrier"},
582         {Opt_snapshot, "cp=%u"},
583         {Opt_order, "order=%s"},
584         {Opt_norecovery, "norecovery"},
585         {Opt_discard, "discard"},
586         {Opt_err, NULL}
587 };
588
589 static int parse_options(char *options, struct super_block *sb)
590 {
591         struct nilfs_sb_info *sbi = NILFS_SB(sb);
592         char *p;
593         substring_t args[MAX_OPT_ARGS];
594         int option;
595
596         if (!options)
597                 return 1;
598
599         while ((p = strsep(&options, ",")) != NULL) {
600                 int token;
601                 if (!*p)
602                         continue;
603
604                 token = match_token(p, tokens, args);
605                 switch (token) {
606                 case Opt_nobarrier:
607                         nilfs_clear_opt(sbi, BARRIER);
608                         break;
609                 case Opt_order:
610                         if (strcmp(args[0].from, "relaxed") == 0)
611                                 /* Ordered data semantics */
612                                 nilfs_clear_opt(sbi, STRICT_ORDER);
613                         else if (strcmp(args[0].from, "strict") == 0)
614                                 /* Strict in-order semantics */
615                                 nilfs_set_opt(sbi, STRICT_ORDER);
616                         else
617                                 return 0;
618                         break;
619                 case Opt_err_panic:
620                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
621                         break;
622                 case Opt_err_ro:
623                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
624                         break;
625                 case Opt_err_cont:
626                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
627                         break;
628                 case Opt_snapshot:
629                         if (match_int(&args[0], &option) || option <= 0)
630                                 return 0;
631                         if (!(sb->s_flags & MS_RDONLY))
632                                 return 0;
633                         sbi->s_snapshot_cno = option;
634                         nilfs_set_opt(sbi, SNAPSHOT);
635                         break;
636                 case Opt_norecovery:
637                         nilfs_set_opt(sbi, NORECOVERY);
638                         break;
639                 case Opt_discard:
640                         nilfs_set_opt(sbi, DISCARD);
641                         break;
642                 default:
643                         printk(KERN_ERR
644                                "NILFS: Unrecognized mount option \"%s\"\n", p);
645                         return 0;
646                 }
647         }
648         return 1;
649 }
650
651 static inline void
652 nilfs_set_default_options(struct nilfs_sb_info *sbi,
653                           struct nilfs_super_block *sbp)
654 {
655         sbi->s_mount_opt =
656                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
657 }
658
659 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
660 {
661         struct the_nilfs *nilfs = sbi->s_nilfs;
662         struct nilfs_super_block **sbp;
663         int max_mnt_count;
664         int mnt_count;
665
666         /* nilfs->ns_sem must be locked by the caller. */
667         sbp = nilfs_prepare_super(sbi);
668         if (!sbp)
669                 return -EIO;
670
671         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
672         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
673
674         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
675                 printk(KERN_WARNING
676                        "NILFS warning: mounting fs with errors\n");
677 #if 0
678         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
679                 printk(KERN_WARNING
680                        "NILFS warning: maximal mount count reached\n");
681 #endif
682         }
683         if (!max_mnt_count)
684                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
685
686         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
687         sbp[0]->s_state =
688                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
689         sbp[0]->s_mtime = cpu_to_le64(get_seconds());
690         return nilfs_commit_super(sbi, 1);
691 }
692
693 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
694                                                  u64 pos, int blocksize,
695                                                  struct buffer_head **pbh)
696 {
697         unsigned long long sb_index = pos;
698         unsigned long offset;
699
700         offset = do_div(sb_index, blocksize);
701         *pbh = sb_bread(sb, sb_index);
702         if (!*pbh)
703                 return NULL;
704         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
705 }
706
707 int nilfs_store_magic_and_option(struct super_block *sb,
708                                  struct nilfs_super_block *sbp,
709                                  char *data)
710 {
711         struct nilfs_sb_info *sbi = NILFS_SB(sb);
712
713         sb->s_magic = le16_to_cpu(sbp->s_magic);
714
715         /* FS independent flags */
716 #ifdef NILFS_ATIME_DISABLE
717         sb->s_flags |= MS_NOATIME;
718 #endif
719
720         nilfs_set_default_options(sbi, sbp);
721
722         sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
723         sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
724         sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
725         sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
726
727         return !parse_options(data, sb) ? -EINVAL : 0 ;
728 }
729
730 /**
731  * nilfs_fill_super() - initialize a super block instance
732  * @sb: super_block
733  * @data: mount options
734  * @silent: silent mode flag
735  * @nilfs: the_nilfs struct
736  *
737  * This function is called exclusively by nilfs->ns_mount_mutex.
738  * So, the recovery process is protected from other simultaneous mounts.
739  */
740 static int
741 nilfs_fill_super(struct super_block *sb, void *data, int silent,
742                  struct the_nilfs *nilfs)
743 {
744         struct nilfs_sb_info *sbi;
745         struct inode *root;
746         __u64 cno;
747         int err;
748
749         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
750         if (!sbi)
751                 return -ENOMEM;
752
753         sb->s_fs_info = sbi;
754
755         get_nilfs(nilfs);
756         sbi->s_nilfs = nilfs;
757         sbi->s_super = sb;
758         atomic_set(&sbi->s_count, 1);
759
760         err = init_nilfs(nilfs, sbi, (char *)data);
761         if (err)
762                 goto failed_sbi;
763
764         spin_lock_init(&sbi->s_inode_lock);
765         INIT_LIST_HEAD(&sbi->s_dirty_files);
766         INIT_LIST_HEAD(&sbi->s_list);
767
768         /*
769          * Following initialization is overlapped because
770          * nilfs_sb_info structure has been cleared at the beginning.
771          * But we reserve them to keep our interest and make ready
772          * for the future change.
773          */
774         get_random_bytes(&sbi->s_next_generation,
775                          sizeof(sbi->s_next_generation));
776         spin_lock_init(&sbi->s_next_gen_lock);
777
778         sb->s_op = &nilfs_sops;
779         sb->s_export_op = &nilfs_export_ops;
780         sb->s_root = NULL;
781         sb->s_time_gran = 1;
782         sb->s_bdi = nilfs->ns_bdi;
783
784         err = load_nilfs(nilfs, sbi);
785         if (err)
786                 goto failed_sbi;
787
788         cno = nilfs_last_cno(nilfs);
789
790         if (sb->s_flags & MS_RDONLY) {
791                 if (nilfs_test_opt(sbi, SNAPSHOT)) {
792                         down_read(&nilfs->ns_segctor_sem);
793                         err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
794                                                        sbi->s_snapshot_cno);
795                         up_read(&nilfs->ns_segctor_sem);
796                         if (err < 0) {
797                                 if (err == -ENOENT)
798                                         err = -EINVAL;
799                                 goto failed_sbi;
800                         }
801                         if (!err) {
802                                 printk(KERN_ERR
803                                        "NILFS: The specified checkpoint is "
804                                        "not a snapshot "
805                                        "(checkpoint number=%llu).\n",
806                                        (unsigned long long)sbi->s_snapshot_cno);
807                                 err = -EINVAL;
808                                 goto failed_sbi;
809                         }
810                         cno = sbi->s_snapshot_cno;
811                 }
812         }
813
814         err = nilfs_attach_checkpoint(sbi, cno);
815         if (err) {
816                 printk(KERN_ERR "NILFS: error loading a checkpoint"
817                        " (checkpoint number=%llu).\n", (unsigned long long)cno);
818                 goto failed_sbi;
819         }
820
821         if (!(sb->s_flags & MS_RDONLY)) {
822                 err = nilfs_attach_segment_constructor(sbi);
823                 if (err)
824                         goto failed_checkpoint;
825         }
826
827         root = nilfs_iget(sb, NILFS_ROOT_INO);
828         if (IS_ERR(root)) {
829                 printk(KERN_ERR "NILFS: get root inode failed\n");
830                 err = PTR_ERR(root);
831                 goto failed_segctor;
832         }
833         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
834                 iput(root);
835                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
836                 err = -EINVAL;
837                 goto failed_segctor;
838         }
839         sb->s_root = d_alloc_root(root);
840         if (!sb->s_root) {
841                 iput(root);
842                 printk(KERN_ERR "NILFS: get root dentry failed\n");
843                 err = -ENOMEM;
844                 goto failed_segctor;
845         }
846
847         if (!(sb->s_flags & MS_RDONLY)) {
848                 down_write(&nilfs->ns_sem);
849                 nilfs_setup_super(sbi);
850                 up_write(&nilfs->ns_sem);
851         }
852
853         down_write(&nilfs->ns_super_sem);
854         if (!nilfs_test_opt(sbi, SNAPSHOT))
855                 nilfs->ns_current = sbi;
856         up_write(&nilfs->ns_super_sem);
857
858         return 0;
859
860  failed_segctor:
861         nilfs_detach_segment_constructor(sbi);
862
863  failed_checkpoint:
864         nilfs_detach_checkpoint(sbi);
865
866  failed_sbi:
867         put_nilfs(nilfs);
868         sb->s_fs_info = NULL;
869         nilfs_put_sbinfo(sbi);
870         return err;
871 }
872
873 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
874 {
875         struct nilfs_sb_info *sbi = NILFS_SB(sb);
876         struct the_nilfs *nilfs = sbi->s_nilfs;
877         unsigned long old_sb_flags;
878         struct nilfs_mount_options old_opts;
879         int was_snapshot, err;
880
881         lock_kernel();
882
883         down_write(&nilfs->ns_super_sem);
884         old_sb_flags = sb->s_flags;
885         old_opts.mount_opt = sbi->s_mount_opt;
886         old_opts.snapshot_cno = sbi->s_snapshot_cno;
887         was_snapshot = nilfs_test_opt(sbi, SNAPSHOT);
888
889         if (!parse_options(data, sb)) {
890                 err = -EINVAL;
891                 goto restore_opts;
892         }
893         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
894
895         err = -EINVAL;
896         if (was_snapshot) {
897                 if (!(*flags & MS_RDONLY)) {
898                         printk(KERN_ERR "NILFS (device %s): cannot remount "
899                                "snapshot read/write.\n",
900                                sb->s_id);
901                         goto restore_opts;
902                 } else if (sbi->s_snapshot_cno != old_opts.snapshot_cno) {
903                         printk(KERN_ERR "NILFS (device %s): cannot "
904                                "remount to a different snapshot.\n",
905                                sb->s_id);
906                         goto restore_opts;
907                 }
908         } else {
909                 if (nilfs_test_opt(sbi, SNAPSHOT)) {
910                         printk(KERN_ERR "NILFS (device %s): cannot change "
911                                "a regular mount to a snapshot.\n",
912                                sb->s_id);
913                         goto restore_opts;
914                 }
915         }
916
917         if (!nilfs_valid_fs(nilfs)) {
918                 printk(KERN_WARNING "NILFS (device %s): couldn't "
919                        "remount because the filesystem is in an "
920                        "incomplete recovery state.\n", sb->s_id);
921                 goto restore_opts;
922         }
923
924         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
925                 goto out;
926         if (*flags & MS_RDONLY) {
927                 /* Shutting down the segment constructor */
928                 nilfs_detach_segment_constructor(sbi);
929                 sb->s_flags |= MS_RDONLY;
930
931                 /*
932                  * Remounting a valid RW partition RDONLY, so set
933                  * the RDONLY flag and then mark the partition as valid again.
934                  */
935                 down_write(&nilfs->ns_sem);
936                 nilfs_cleanup_super(sbi);
937                 up_write(&nilfs->ns_sem);
938         } else {
939                 /*
940                  * Mounting a RDONLY partition read-write, so reread and
941                  * store the current valid flag.  (It may have been changed
942                  * by fsck since we originally mounted the partition.)
943                  */
944                 sb->s_flags &= ~MS_RDONLY;
945
946                 err = nilfs_attach_segment_constructor(sbi);
947                 if (err)
948                         goto restore_opts;
949
950                 down_write(&nilfs->ns_sem);
951                 nilfs_setup_super(sbi);
952                 up_write(&nilfs->ns_sem);
953         }
954  out:
955         up_write(&nilfs->ns_super_sem);
956         unlock_kernel();
957         return 0;
958
959  restore_opts:
960         sb->s_flags = old_sb_flags;
961         sbi->s_mount_opt = old_opts.mount_opt;
962         sbi->s_snapshot_cno = old_opts.snapshot_cno;
963         up_write(&nilfs->ns_super_sem);
964         unlock_kernel();
965         return err;
966 }
967
968 struct nilfs_super_data {
969         struct block_device *bdev;
970         struct nilfs_sb_info *sbi;
971         __u64 cno;
972         int flags;
973 };
974
975 /**
976  * nilfs_identify - pre-read mount options needed to identify mount instance
977  * @data: mount options
978  * @sd: nilfs_super_data
979  */
980 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
981 {
982         char *p, *options = data;
983         substring_t args[MAX_OPT_ARGS];
984         int option, token;
985         int ret = 0;
986
987         do {
988                 p = strsep(&options, ",");
989                 if (p != NULL && *p) {
990                         token = match_token(p, tokens, args);
991                         if (token == Opt_snapshot) {
992                                 if (!(sd->flags & MS_RDONLY))
993                                         ret++;
994                                 else {
995                                         ret = match_int(&args[0], &option);
996                                         if (!ret) {
997                                                 if (option > 0)
998                                                         sd->cno = option;
999                                                 else
1000                                                         ret++;
1001                                         }
1002                                 }
1003                         }
1004                         if (ret)
1005                                 printk(KERN_ERR
1006                                        "NILFS: invalid mount option: %s\n", p);
1007                 }
1008                 if (!options)
1009                         break;
1010                 BUG_ON(options == data);
1011                 *(options - 1) = ',';
1012         } while (!ret);
1013         return ret;
1014 }
1015
1016 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1017 {
1018         struct nilfs_super_data *sd = data;
1019
1020         s->s_bdev = sd->bdev;
1021         s->s_dev = s->s_bdev->bd_dev;
1022         return 0;
1023 }
1024
1025 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1026 {
1027         struct nilfs_super_data *sd = data;
1028
1029         return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1030 }
1031
1032 static int
1033 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1034              const char *dev_name, void *data, struct vfsmount *mnt)
1035 {
1036         struct nilfs_super_data sd;
1037         struct super_block *s;
1038         fmode_t mode = FMODE_READ;
1039         struct the_nilfs *nilfs;
1040         int err, need_to_close = 1;
1041
1042         if (!(flags & MS_RDONLY))
1043                 mode |= FMODE_WRITE;
1044
1045         sd.bdev = open_bdev_exclusive(dev_name, mode, fs_type);
1046         if (IS_ERR(sd.bdev))
1047                 return PTR_ERR(sd.bdev);
1048
1049         /*
1050          * To get mount instance using sget() vfs-routine, NILFS needs
1051          * much more information than normal filesystems to identify mount
1052          * instance.  For snapshot mounts, not only a mount type (ro-mount
1053          * or rw-mount) but also a checkpoint number is required.
1054          */
1055         sd.cno = 0;
1056         sd.flags = flags;
1057         if (nilfs_identify((char *)data, &sd)) {
1058                 err = -EINVAL;
1059                 goto failed;
1060         }
1061
1062         nilfs = find_or_create_nilfs(sd.bdev);
1063         if (!nilfs) {
1064                 err = -ENOMEM;
1065                 goto failed;
1066         }
1067
1068         mutex_lock(&nilfs->ns_mount_mutex);
1069
1070         if (!sd.cno) {
1071                 /*
1072                  * Check if an exclusive mount exists or not.
1073                  * Snapshot mounts coexist with a current mount
1074                  * (i.e. rw-mount or ro-mount), whereas rw-mount and
1075                  * ro-mount are mutually exclusive.
1076                  */
1077                 down_read(&nilfs->ns_super_sem);
1078                 if (nilfs->ns_current &&
1079                     ((nilfs->ns_current->s_super->s_flags ^ flags)
1080                      & MS_RDONLY)) {
1081                         up_read(&nilfs->ns_super_sem);
1082                         err = -EBUSY;
1083                         goto failed_unlock;
1084                 }
1085                 up_read(&nilfs->ns_super_sem);
1086         }
1087
1088         /*
1089          * Find existing nilfs_sb_info struct
1090          */
1091         sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1092
1093         /*
1094          * Get super block instance holding the nilfs_sb_info struct.
1095          * A new instance is allocated if no existing mount is present or
1096          * existing instance has been unmounted.
1097          */
1098         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1099         if (sd.sbi)
1100                 nilfs_put_sbinfo(sd.sbi);
1101
1102         if (IS_ERR(s)) {
1103                 err = PTR_ERR(s);
1104                 goto failed_unlock;
1105         }
1106
1107         if (!s->s_root) {
1108                 char b[BDEVNAME_SIZE];
1109
1110                 /* New superblock instance created */
1111                 s->s_flags = flags;
1112                 s->s_mode = mode;
1113                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1114                 sb_set_blocksize(s, block_size(sd.bdev));
1115
1116                 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0,
1117                                        nilfs);
1118                 if (err)
1119                         goto cancel_new;
1120
1121                 s->s_flags |= MS_ACTIVE;
1122                 need_to_close = 0;
1123         }
1124
1125         mutex_unlock(&nilfs->ns_mount_mutex);
1126         put_nilfs(nilfs);
1127         if (need_to_close)
1128                 close_bdev_exclusive(sd.bdev, mode);
1129         simple_set_mnt(mnt, s);
1130         return 0;
1131
1132  failed_unlock:
1133         mutex_unlock(&nilfs->ns_mount_mutex);
1134         put_nilfs(nilfs);
1135  failed:
1136         close_bdev_exclusive(sd.bdev, mode);
1137
1138         return err;
1139
1140  cancel_new:
1141         /* Abandoning the newly allocated superblock */
1142         mutex_unlock(&nilfs->ns_mount_mutex);
1143         put_nilfs(nilfs);
1144         deactivate_locked_super(s);
1145         /*
1146          * deactivate_locked_super() invokes close_bdev_exclusive().
1147          * We must finish all post-cleaning before this call;
1148          * put_nilfs() needs the block device.
1149          */
1150         return err;
1151 }
1152
1153 struct file_system_type nilfs_fs_type = {
1154         .owner    = THIS_MODULE,
1155         .name     = "nilfs2",
1156         .get_sb   = nilfs_get_sb,
1157         .kill_sb  = kill_block_super,
1158         .fs_flags = FS_REQUIRES_DEV,
1159 };
1160
1161 static void nilfs_inode_init_once(void *obj)
1162 {
1163         struct nilfs_inode_info *ii = obj;
1164
1165         INIT_LIST_HEAD(&ii->i_dirty);
1166 #ifdef CONFIG_NILFS_XATTR
1167         init_rwsem(&ii->xattr_sem);
1168 #endif
1169         nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
1170         ii->i_bmap = (struct nilfs_bmap *)&ii->i_bmap_union;
1171         inode_init_once(&ii->vfs_inode);
1172 }
1173
1174 static void nilfs_segbuf_init_once(void *obj)
1175 {
1176         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1177 }
1178
1179 static void nilfs_destroy_cachep(void)
1180 {
1181         if (nilfs_inode_cachep)
1182                 kmem_cache_destroy(nilfs_inode_cachep);
1183         if (nilfs_transaction_cachep)
1184                 kmem_cache_destroy(nilfs_transaction_cachep);
1185         if (nilfs_segbuf_cachep)
1186                 kmem_cache_destroy(nilfs_segbuf_cachep);
1187         if (nilfs_btree_path_cache)
1188                 kmem_cache_destroy(nilfs_btree_path_cache);
1189 }
1190
1191 static int __init nilfs_init_cachep(void)
1192 {
1193         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1194                         sizeof(struct nilfs_inode_info), 0,
1195                         SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1196         if (!nilfs_inode_cachep)
1197                 goto fail;
1198
1199         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1200                         sizeof(struct nilfs_transaction_info), 0,
1201                         SLAB_RECLAIM_ACCOUNT, NULL);
1202         if (!nilfs_transaction_cachep)
1203                 goto fail;
1204
1205         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1206                         sizeof(struct nilfs_segment_buffer), 0,
1207                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1208         if (!nilfs_segbuf_cachep)
1209                 goto fail;
1210
1211         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1212                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1213                         0, 0, NULL);
1214         if (!nilfs_btree_path_cache)
1215                 goto fail;
1216
1217         return 0;
1218
1219 fail:
1220         nilfs_destroy_cachep();
1221         return -ENOMEM;
1222 }
1223
1224 static int __init init_nilfs_fs(void)
1225 {
1226         int err;
1227
1228         err = nilfs_init_cachep();
1229         if (err)
1230                 goto fail;
1231
1232         err = register_filesystem(&nilfs_fs_type);
1233         if (err)
1234                 goto free_cachep;
1235
1236         printk(KERN_INFO "NILFS version 2 loaded\n");
1237         return 0;
1238
1239 free_cachep:
1240         nilfs_destroy_cachep();
1241 fail:
1242         return err;
1243 }
1244
1245 static void __exit exit_nilfs_fs(void)
1246 {
1247         nilfs_destroy_cachep();
1248         unregister_filesystem(&nilfs_fs_type);
1249 }
1250
1251 module_init(init_nilfs_fs)
1252 module_exit(exit_nilfs_fs)