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