]> bbs.cooldavid.org Git - net-next-2.6.git/blob - fs/nilfs2/super.c
nilfs2: see state of root dentry for mount check of snapshots
[net-next-2.6.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/vfs.h>
49 #include <linux/writeback.h>
50 #include <linux/kobject.h>
51 #include <linux/seq_file.h>
52 #include <linux/mount.h>
53 #include "nilfs.h"
54 #include "export.h"
55 #include "mdt.h"
56 #include "alloc.h"
57 #include "btree.h"
58 #include "btnode.h"
59 #include "page.h"
60 #include "cpfile.h"
61 #include "ifile.h"
62 #include "dat.h"
63 #include "segment.h"
64 #include "segbuf.h"
65
66 MODULE_AUTHOR("NTT Corp.");
67 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
68                    "(NILFS)");
69 MODULE_LICENSE("GPL");
70
71 struct kmem_cache *nilfs_inode_cachep;
72 struct kmem_cache *nilfs_transaction_cachep;
73 struct kmem_cache *nilfs_segbuf_cachep;
74 struct kmem_cache *nilfs_btree_path_cache;
75
76 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
77
78 static void nilfs_set_error(struct nilfs_sb_info *sbi)
79 {
80         struct the_nilfs *nilfs = sbi->s_nilfs;
81         struct nilfs_super_block **sbp;
82
83         down_write(&nilfs->ns_sem);
84         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
85                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
86                 sbp = nilfs_prepare_super(sbi, 0);
87                 if (likely(sbp)) {
88                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
89                         if (sbp[1])
90                                 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
91                         nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
92                 }
93         }
94         up_write(&nilfs->ns_sem);
95 }
96
97 /**
98  * nilfs_error() - report failure condition on a filesystem
99  *
100  * nilfs_error() sets an ERROR_FS flag on the superblock as well as
101  * reporting an error message.  It should be called when NILFS detects
102  * incoherences or defects of meta data on disk.  As for sustainable
103  * errors such as a single-shot I/O error, nilfs_warning() or the printk()
104  * function should be used instead.
105  *
106  * The segment constructor must not call this function because it can
107  * kill itself.
108  */
109 void nilfs_error(struct super_block *sb, const char *function,
110                  const char *fmt, ...)
111 {
112         struct nilfs_sb_info *sbi = NILFS_SB(sb);
113         va_list args;
114
115         va_start(args, fmt);
116         printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
117         vprintk(fmt, args);
118         printk("\n");
119         va_end(args);
120
121         if (!(sb->s_flags & MS_RDONLY)) {
122                 nilfs_set_error(sbi);
123
124                 if (nilfs_test_opt(sbi, ERRORS_RO)) {
125                         printk(KERN_CRIT "Remounting filesystem read-only\n");
126                         sb->s_flags |= MS_RDONLY;
127                 }
128         }
129
130         if (nilfs_test_opt(sbi, ERRORS_PANIC))
131                 panic("NILFS (device %s): panic forced after error\n",
132                       sb->s_id);
133 }
134
135 void nilfs_warning(struct super_block *sb, const char *function,
136                    const char *fmt, ...)
137 {
138         va_list args;
139
140         va_start(args, fmt);
141         printk(KERN_WARNING "NILFS warning (device %s): %s: ",
142                sb->s_id, function);
143         vprintk(fmt, args);
144         printk("\n");
145         va_end(args);
146 }
147
148
149 struct inode *nilfs_alloc_inode_common(struct the_nilfs *nilfs)
150 {
151         struct nilfs_inode_info *ii;
152
153         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
154         if (!ii)
155                 return NULL;
156         ii->i_bh = NULL;
157         ii->i_state = 0;
158         ii->i_cno = 0;
159         ii->vfs_inode.i_version = 1;
160         nilfs_btnode_cache_init(&ii->i_btnode_cache, nilfs->ns_bdi);
161         return &ii->vfs_inode;
162 }
163
164 struct inode *nilfs_alloc_inode(struct super_block *sb)
165 {
166         return nilfs_alloc_inode_common(NILFS_SB(sb)->s_nilfs);
167 }
168
169 void nilfs_destroy_inode(struct inode *inode)
170 {
171         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
172
173         if (mdi) {
174                 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
175                 kfree(mdi);
176         }
177         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
178 }
179
180 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int flag)
181 {
182         struct the_nilfs *nilfs = sbi->s_nilfs;
183         int err;
184
185  retry:
186         set_buffer_dirty(nilfs->ns_sbh[0]);
187
188         if (nilfs_test_opt(sbi, BARRIER)) {
189                 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
190                                           WRITE_SYNC | WRITE_BARRIER);
191                 if (err == -EOPNOTSUPP) {
192                         nilfs_warning(sbi->s_super, __func__,
193                                       "barrier-based sync failed. "
194                                       "disabling barriers\n");
195                         nilfs_clear_opt(sbi, BARRIER);
196                         goto retry;
197                 }
198         } else {
199                 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
200         }
201
202         if (unlikely(err)) {
203                 printk(KERN_ERR
204                        "NILFS: unable to write superblock (err=%d)\n", err);
205                 if (err == -EIO && nilfs->ns_sbh[1]) {
206                         /*
207                          * sbp[0] points to newer log than sbp[1],
208                          * so copy sbp[0] to sbp[1] to take over sbp[0].
209                          */
210                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
211                                nilfs->ns_sbsize);
212                         nilfs_fall_back_super_block(nilfs);
213                         goto retry;
214                 }
215         } else {
216                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
217
218                 nilfs->ns_sbwcount++;
219
220                 /*
221                  * The latest segment becomes trailable from the position
222                  * written in superblock.
223                  */
224                 clear_nilfs_discontinued(nilfs);
225
226                 /* update GC protection for recent segments */
227                 if (nilfs->ns_sbh[1]) {
228                         if (flag == NILFS_SB_COMMIT_ALL) {
229                                 set_buffer_dirty(nilfs->ns_sbh[1]);
230                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
231                                         goto out;
232                         }
233                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
234                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
235                                 sbp = nilfs->ns_sbp[1];
236                 }
237
238                 spin_lock(&nilfs->ns_last_segment_lock);
239                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
240                 spin_unlock(&nilfs->ns_last_segment_lock);
241         }
242  out:
243         return err;
244 }
245
246 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
247                           struct the_nilfs *nilfs)
248 {
249         sector_t nfreeblocks;
250
251         /* nilfs->ns_sem must be locked by the caller. */
252         nilfs_count_free_blocks(nilfs, &nfreeblocks);
253         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
254
255         spin_lock(&nilfs->ns_last_segment_lock);
256         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
257         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
258         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
259         spin_unlock(&nilfs->ns_last_segment_lock);
260 }
261
262 struct nilfs_super_block **nilfs_prepare_super(struct nilfs_sb_info *sbi,
263                                                int flip)
264 {
265         struct the_nilfs *nilfs = sbi->s_nilfs;
266         struct nilfs_super_block **sbp = nilfs->ns_sbp;
267
268         /* nilfs->ns_sem must be locked by the caller. */
269         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
270                 if (sbp[1] &&
271                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
272                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
273                 } else {
274                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
275                                sbi->s_super->s_id);
276                         return NULL;
277                 }
278         } else if (sbp[1] &&
279                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
280                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
281         }
282
283         if (flip && sbp[1])
284                 nilfs_swap_super_block(nilfs);
285
286         return sbp;
287 }
288
289 int nilfs_commit_super(struct nilfs_sb_info *sbi, int flag)
290 {
291         struct the_nilfs *nilfs = sbi->s_nilfs;
292         struct nilfs_super_block **sbp = nilfs->ns_sbp;
293         time_t t;
294
295         /* nilfs->ns_sem must be locked by the caller. */
296         t = get_seconds();
297         nilfs->ns_sbwtime = t;
298         sbp[0]->s_wtime = cpu_to_le64(t);
299         sbp[0]->s_sum = 0;
300         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
301                                              (unsigned char *)sbp[0],
302                                              nilfs->ns_sbsize));
303         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
304                 sbp[1]->s_wtime = sbp[0]->s_wtime;
305                 sbp[1]->s_sum = 0;
306                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
307                                             (unsigned char *)sbp[1],
308                                             nilfs->ns_sbsize));
309         }
310         clear_nilfs_sb_dirty(nilfs);
311         return nilfs_sync_super(sbi, flag);
312 }
313
314 /**
315  * nilfs_cleanup_super() - write filesystem state for cleanup
316  * @sbi: nilfs_sb_info to be unmounted or degraded to read-only
317  *
318  * This function restores state flags in the on-disk super block.
319  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
320  * filesystem was not clean previously.
321  */
322 int nilfs_cleanup_super(struct nilfs_sb_info *sbi)
323 {
324         struct nilfs_super_block **sbp;
325         int flag = NILFS_SB_COMMIT;
326         int ret = -EIO;
327
328         sbp = nilfs_prepare_super(sbi, 0);
329         if (sbp) {
330                 sbp[0]->s_state = cpu_to_le16(sbi->s_nilfs->ns_mount_state);
331                 nilfs_set_log_cursor(sbp[0], sbi->s_nilfs);
332                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
333                         /*
334                          * make the "clean" flag also to the opposite
335                          * super block if both super blocks point to
336                          * the same checkpoint.
337                          */
338                         sbp[1]->s_state = sbp[0]->s_state;
339                         flag = NILFS_SB_COMMIT_ALL;
340                 }
341                 ret = nilfs_commit_super(sbi, flag);
342         }
343         return ret;
344 }
345
346 static void nilfs_put_super(struct super_block *sb)
347 {
348         struct nilfs_sb_info *sbi = NILFS_SB(sb);
349         struct the_nilfs *nilfs = sbi->s_nilfs;
350
351         nilfs_detach_segment_constructor(sbi);
352
353         if (!(sb->s_flags & MS_RDONLY)) {
354                 down_write(&nilfs->ns_sem);
355                 nilfs_cleanup_super(sbi);
356                 up_write(&nilfs->ns_sem);
357         }
358
359         iput(nilfs->ns_sufile);
360         iput(nilfs->ns_cpfile);
361         iput(nilfs->ns_dat);
362
363         destroy_nilfs(nilfs);
364         sbi->s_super = NULL;
365         sb->s_fs_info = NULL;
366         kfree(sbi);
367 }
368
369 static int nilfs_sync_fs(struct super_block *sb, int wait)
370 {
371         struct nilfs_sb_info *sbi = NILFS_SB(sb);
372         struct the_nilfs *nilfs = sbi->s_nilfs;
373         struct nilfs_super_block **sbp;
374         int err = 0;
375
376         /* This function is called when super block should be written back */
377         if (wait)
378                 err = nilfs_construct_segment(sb);
379
380         down_write(&nilfs->ns_sem);
381         if (nilfs_sb_dirty(nilfs)) {
382                 sbp = nilfs_prepare_super(sbi, nilfs_sb_will_flip(nilfs));
383                 if (likely(sbp)) {
384                         nilfs_set_log_cursor(sbp[0], nilfs);
385                         nilfs_commit_super(sbi, NILFS_SB_COMMIT);
386                 }
387         }
388         up_write(&nilfs->ns_sem);
389
390         return err;
391 }
392
393 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno, int curr_mnt,
394                             struct nilfs_root **rootp)
395 {
396         struct the_nilfs *nilfs = sbi->s_nilfs;
397         struct nilfs_root *root;
398         struct nilfs_checkpoint *raw_cp;
399         struct buffer_head *bh_cp;
400         int err = -ENOMEM;
401
402         root = nilfs_find_or_create_root(
403                 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
404         if (!root)
405                 return err;
406
407         if (root->ifile)
408                 goto reuse; /* already attached checkpoint */
409
410         down_read(&nilfs->ns_segctor_sem);
411         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
412                                           &bh_cp);
413         up_read(&nilfs->ns_segctor_sem);
414         if (unlikely(err)) {
415                 if (err == -ENOENT || err == -EINVAL) {
416                         printk(KERN_ERR
417                                "NILFS: Invalid checkpoint "
418                                "(checkpoint number=%llu)\n",
419                                (unsigned long long)cno);
420                         err = -EINVAL;
421                 }
422                 goto failed;
423         }
424
425         err = nilfs_ifile_read(sbi->s_super, root, nilfs->ns_inode_size,
426                                &raw_cp->cp_ifile_inode, &root->ifile);
427         if (err)
428                 goto failed_bh;
429
430         atomic_set(&root->inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
431         atomic_set(&root->blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
432
433         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
434
435  reuse:
436         *rootp = root;
437         return 0;
438
439  failed_bh:
440         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
441  failed:
442         nilfs_put_root(root);
443
444         return err;
445 }
446
447 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
448 {
449         struct super_block *sb = dentry->d_sb;
450         struct nilfs_root *root = NILFS_I(dentry->d_inode)->i_root;
451         struct the_nilfs *nilfs = root->nilfs;
452         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
453         unsigned long long blocks;
454         unsigned long overhead;
455         unsigned long nrsvblocks;
456         sector_t nfreeblocks;
457         int err;
458
459         /*
460          * Compute all of the segment blocks
461          *
462          * The blocks before first segment and after last segment
463          * are excluded.
464          */
465         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
466                 - nilfs->ns_first_data_block;
467         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
468
469         /*
470          * Compute the overhead
471          *
472          * When distributing meta data blocks outside segment structure,
473          * We must count them as the overhead.
474          */
475         overhead = 0;
476
477         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
478         if (unlikely(err))
479                 return err;
480
481         buf->f_type = NILFS_SUPER_MAGIC;
482         buf->f_bsize = sb->s_blocksize;
483         buf->f_blocks = blocks - overhead;
484         buf->f_bfree = nfreeblocks;
485         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
486                 (buf->f_bfree - nrsvblocks) : 0;
487         buf->f_files = atomic_read(&root->inodes_count);
488         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
489         buf->f_namelen = NILFS_NAME_LEN;
490         buf->f_fsid.val[0] = (u32)id;
491         buf->f_fsid.val[1] = (u32)(id >> 32);
492
493         return 0;
494 }
495
496 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
497 {
498         struct super_block *sb = vfs->mnt_sb;
499         struct nilfs_sb_info *sbi = NILFS_SB(sb);
500         struct nilfs_root *root = NILFS_I(vfs->mnt_root->d_inode)->i_root;
501
502         if (!nilfs_test_opt(sbi, BARRIER))
503                 seq_puts(seq, ",nobarrier");
504         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
505                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
506         if (nilfs_test_opt(sbi, ERRORS_PANIC))
507                 seq_puts(seq, ",errors=panic");
508         if (nilfs_test_opt(sbi, ERRORS_CONT))
509                 seq_puts(seq, ",errors=continue");
510         if (nilfs_test_opt(sbi, STRICT_ORDER))
511                 seq_puts(seq, ",order=strict");
512         if (nilfs_test_opt(sbi, NORECOVERY))
513                 seq_puts(seq, ",norecovery");
514         if (nilfs_test_opt(sbi, DISCARD))
515                 seq_puts(seq, ",discard");
516
517         return 0;
518 }
519
520 static const struct super_operations nilfs_sops = {
521         .alloc_inode    = nilfs_alloc_inode,
522         .destroy_inode  = nilfs_destroy_inode,
523         .dirty_inode    = nilfs_dirty_inode,
524         /* .write_inode    = nilfs_write_inode, */
525         /* .put_inode      = nilfs_put_inode, */
526         /* .drop_inode    = nilfs_drop_inode, */
527         .evict_inode    = nilfs_evict_inode,
528         .put_super      = nilfs_put_super,
529         /* .write_super    = nilfs_write_super, */
530         .sync_fs        = nilfs_sync_fs,
531         /* .write_super_lockfs */
532         /* .unlockfs */
533         .statfs         = nilfs_statfs,
534         .remount_fs     = nilfs_remount,
535         /* .umount_begin */
536         .show_options = nilfs_show_options
537 };
538
539 enum {
540         Opt_err_cont, Opt_err_panic, Opt_err_ro,
541         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
542         Opt_discard, Opt_nodiscard, Opt_err,
543 };
544
545 static match_table_t tokens = {
546         {Opt_err_cont, "errors=continue"},
547         {Opt_err_panic, "errors=panic"},
548         {Opt_err_ro, "errors=remount-ro"},
549         {Opt_barrier, "barrier"},
550         {Opt_nobarrier, "nobarrier"},
551         {Opt_snapshot, "cp=%u"},
552         {Opt_order, "order=%s"},
553         {Opt_norecovery, "norecovery"},
554         {Opt_discard, "discard"},
555         {Opt_nodiscard, "nodiscard"},
556         {Opt_err, NULL}
557 };
558
559 static int parse_options(char *options, struct super_block *sb, int is_remount)
560 {
561         struct nilfs_sb_info *sbi = NILFS_SB(sb);
562         char *p;
563         substring_t args[MAX_OPT_ARGS];
564         int option;
565
566         if (!options)
567                 return 1;
568
569         while ((p = strsep(&options, ",")) != NULL) {
570                 int token;
571                 if (!*p)
572                         continue;
573
574                 token = match_token(p, tokens, args);
575                 switch (token) {
576                 case Opt_barrier:
577                         nilfs_set_opt(sbi, BARRIER);
578                         break;
579                 case Opt_nobarrier:
580                         nilfs_clear_opt(sbi, BARRIER);
581                         break;
582                 case Opt_order:
583                         if (strcmp(args[0].from, "relaxed") == 0)
584                                 /* Ordered data semantics */
585                                 nilfs_clear_opt(sbi, STRICT_ORDER);
586                         else if (strcmp(args[0].from, "strict") == 0)
587                                 /* Strict in-order semantics */
588                                 nilfs_set_opt(sbi, STRICT_ORDER);
589                         else
590                                 return 0;
591                         break;
592                 case Opt_err_panic:
593                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
594                         break;
595                 case Opt_err_ro:
596                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
597                         break;
598                 case Opt_err_cont:
599                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
600                         break;
601                 case Opt_snapshot:
602                         if (match_int(&args[0], &option) || option <= 0)
603                                 return 0;
604                         if (is_remount) {
605                                 printk(KERN_ERR
606                                        "NILFS: \"%s\" option is invalid "
607                                        "for remount.\n", p);
608                                 return 0;
609                         }
610                         break;
611                 case Opt_norecovery:
612                         nilfs_set_opt(sbi, NORECOVERY);
613                         break;
614                 case Opt_discard:
615                         nilfs_set_opt(sbi, DISCARD);
616                         break;
617                 case Opt_nodiscard:
618                         nilfs_clear_opt(sbi, DISCARD);
619                         break;
620                 default:
621                         printk(KERN_ERR
622                                "NILFS: Unrecognized mount option \"%s\"\n", p);
623                         return 0;
624                 }
625         }
626         return 1;
627 }
628
629 static inline void
630 nilfs_set_default_options(struct nilfs_sb_info *sbi,
631                           struct nilfs_super_block *sbp)
632 {
633         sbi->s_mount_opt =
634                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
635 }
636
637 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
638 {
639         struct the_nilfs *nilfs = sbi->s_nilfs;
640         struct nilfs_super_block **sbp;
641         int max_mnt_count;
642         int mnt_count;
643
644         /* nilfs->ns_sem must be locked by the caller. */
645         sbp = nilfs_prepare_super(sbi, 0);
646         if (!sbp)
647                 return -EIO;
648
649         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
650         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
651
652         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
653                 printk(KERN_WARNING
654                        "NILFS warning: mounting fs with errors\n");
655 #if 0
656         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
657                 printk(KERN_WARNING
658                        "NILFS warning: maximal mount count reached\n");
659 #endif
660         }
661         if (!max_mnt_count)
662                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
663
664         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
665         sbp[0]->s_state =
666                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
667         sbp[0]->s_mtime = cpu_to_le64(get_seconds());
668         /* synchronize sbp[1] with sbp[0] */
669         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
670         return nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
671 }
672
673 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
674                                                  u64 pos, int blocksize,
675                                                  struct buffer_head **pbh)
676 {
677         unsigned long long sb_index = pos;
678         unsigned long offset;
679
680         offset = do_div(sb_index, blocksize);
681         *pbh = sb_bread(sb, sb_index);
682         if (!*pbh)
683                 return NULL;
684         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
685 }
686
687 int nilfs_store_magic_and_option(struct super_block *sb,
688                                  struct nilfs_super_block *sbp,
689                                  char *data)
690 {
691         struct nilfs_sb_info *sbi = NILFS_SB(sb);
692
693         sb->s_magic = le16_to_cpu(sbp->s_magic);
694
695         /* FS independent flags */
696 #ifdef NILFS_ATIME_DISABLE
697         sb->s_flags |= MS_NOATIME;
698 #endif
699
700         nilfs_set_default_options(sbi, sbp);
701
702         sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
703         sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
704         sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
705         sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
706
707         return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
708 }
709
710 int nilfs_check_feature_compatibility(struct super_block *sb,
711                                       struct nilfs_super_block *sbp)
712 {
713         __u64 features;
714
715         features = le64_to_cpu(sbp->s_feature_incompat) &
716                 ~NILFS_FEATURE_INCOMPAT_SUPP;
717         if (features) {
718                 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
719                        "optional features (%llx)\n",
720                        (unsigned long long)features);
721                 return -EINVAL;
722         }
723         features = le64_to_cpu(sbp->s_feature_compat_ro) &
724                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
725         if (!(sb->s_flags & MS_RDONLY) && features) {
726                 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
727                        "unsupported optional features (%llx)\n",
728                        (unsigned long long)features);
729                 return -EINVAL;
730         }
731         return 0;
732 }
733
734 static int nilfs_get_root_dentry(struct super_block *sb,
735                                  struct nilfs_root *root,
736                                  struct dentry **root_dentry)
737 {
738         struct inode *inode;
739         struct dentry *dentry;
740         int ret = 0;
741
742         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
743         if (IS_ERR(inode)) {
744                 printk(KERN_ERR "NILFS: get root inode failed\n");
745                 ret = PTR_ERR(inode);
746                 goto out;
747         }
748         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
749                 iput(inode);
750                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
751                 ret = -EINVAL;
752                 goto out;
753         }
754
755         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
756                 dentry = d_find_alias(inode);
757                 if (!dentry) {
758                         dentry = d_alloc_root(inode);
759                         if (!dentry) {
760                                 iput(inode);
761                                 ret = -ENOMEM;
762                                 goto failed_dentry;
763                         }
764                 } else {
765                         iput(inode);
766                 }
767         } else {
768                 dentry = d_obtain_alias(inode);
769                 if (IS_ERR(dentry)) {
770                         ret = PTR_ERR(dentry);
771                         goto failed_dentry;
772                 }
773         }
774         *root_dentry = dentry;
775  out:
776         return ret;
777
778  failed_dentry:
779         printk(KERN_ERR "NILFS: get root dentry failed\n");
780         goto out;
781 }
782
783 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
784                                  struct dentry **root_dentry)
785 {
786         struct the_nilfs *nilfs = NILFS_SB(s)->s_nilfs;
787         struct nilfs_root *root;
788         int ret;
789
790         down_read(&nilfs->ns_segctor_sem);
791         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
792         up_read(&nilfs->ns_segctor_sem);
793         if (ret < 0) {
794                 ret = (ret == -ENOENT) ? -EINVAL : ret;
795                 goto out;
796         } else if (!ret) {
797                 printk(KERN_ERR "NILFS: The specified checkpoint is "
798                        "not a snapshot (checkpoint number=%llu).\n",
799                        (unsigned long long)cno);
800                 ret = -EINVAL;
801                 goto out;
802         }
803
804         ret = nilfs_attach_checkpoint(NILFS_SB(s), cno, false, &root);
805         if (ret) {
806                 printk(KERN_ERR "NILFS: error loading snapshot "
807                        "(checkpoint number=%llu).\n",
808                (unsigned long long)cno);
809                 goto out;
810         }
811         ret = nilfs_get_root_dentry(s, root, root_dentry);
812         nilfs_put_root(root);
813  out:
814         return ret;
815 }
816
817 static int nilfs_tree_was_touched(struct dentry *root_dentry)
818 {
819         return atomic_read(&root_dentry->d_count) > 1;
820 }
821
822 /**
823  * nilfs_try_to_shrink_tree() - try to shrink dentries of a checkpoint
824  * @root_dentry: root dentry of the tree to be shrunk
825  *
826  * This function returns true if the tree was in-use.
827  */
828 static int nilfs_try_to_shrink_tree(struct dentry *root_dentry)
829 {
830         if (have_submounts(root_dentry))
831                 return true;
832         shrink_dcache_parent(root_dentry);
833         return nilfs_tree_was_touched(root_dentry);
834 }
835
836 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
837 {
838         struct the_nilfs *nilfs = NILFS_SB(sb)->s_nilfs;
839         struct nilfs_root *root;
840         struct inode *inode;
841         struct dentry *dentry;
842         int ret;
843
844         if (cno < 0 || cno > nilfs->ns_cno)
845                 return false;
846
847         if (cno >= nilfs_last_cno(nilfs))
848                 return true;    /* protect recent checkpoints */
849
850         ret = false;
851         root = nilfs_lookup_root(NILFS_SB(sb)->s_nilfs, cno);
852         if (root) {
853                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
854                 if (inode) {
855                         dentry = d_find_alias(inode);
856                         if (dentry) {
857                                 if (nilfs_tree_was_touched(dentry))
858                                         ret = nilfs_try_to_shrink_tree(dentry);
859                                 dput(dentry);
860                         }
861                         iput(inode);
862                 }
863                 nilfs_put_root(root);
864         }
865         return ret;
866 }
867
868 /**
869  * nilfs_fill_super() - initialize a super block instance
870  * @sb: super_block
871  * @data: mount options
872  * @silent: silent mode flag
873  *
874  * This function is called exclusively by nilfs->ns_mount_mutex.
875  * So, the recovery process is protected from other simultaneous mounts.
876  */
877 static int
878 nilfs_fill_super(struct super_block *sb, void *data, int silent)
879 {
880         struct the_nilfs *nilfs;
881         struct nilfs_sb_info *sbi;
882         struct nilfs_root *fsroot;
883         __u64 cno;
884         int err;
885
886         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
887         if (!sbi)
888                 return -ENOMEM;
889
890         sb->s_fs_info = sbi;
891         sbi->s_super = sb;
892
893         nilfs = alloc_nilfs(sb->s_bdev);
894         if (!nilfs) {
895                 err = -ENOMEM;
896                 goto failed_sbi;
897         }
898         sbi->s_nilfs = nilfs;
899
900         err = init_nilfs(nilfs, sbi, (char *)data);
901         if (err)
902                 goto failed_nilfs;
903
904         spin_lock_init(&sbi->s_inode_lock);
905         INIT_LIST_HEAD(&sbi->s_dirty_files);
906
907         /*
908          * Following initialization is overlapped because
909          * nilfs_sb_info structure has been cleared at the beginning.
910          * But we reserve them to keep our interest and make ready
911          * for the future change.
912          */
913         get_random_bytes(&sbi->s_next_generation,
914                          sizeof(sbi->s_next_generation));
915         spin_lock_init(&sbi->s_next_gen_lock);
916
917         sb->s_op = &nilfs_sops;
918         sb->s_export_op = &nilfs_export_ops;
919         sb->s_root = NULL;
920         sb->s_time_gran = 1;
921         sb->s_bdi = nilfs->ns_bdi;
922
923         err = load_nilfs(nilfs, sbi);
924         if (err)
925                 goto failed_nilfs;
926
927         cno = nilfs_last_cno(nilfs);
928         err = nilfs_attach_checkpoint(sbi, cno, true, &fsroot);
929         if (err) {
930                 printk(KERN_ERR "NILFS: error loading last checkpoint "
931                        "(checkpoint number=%llu).\n", (unsigned long long)cno);
932                 goto failed_unload;
933         }
934
935         if (!(sb->s_flags & MS_RDONLY)) {
936                 err = nilfs_attach_segment_constructor(sbi, fsroot);
937                 if (err)
938                         goto failed_checkpoint;
939         }
940
941         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
942         if (err)
943                 goto failed_segctor;
944
945         nilfs_put_root(fsroot);
946
947         if (!(sb->s_flags & MS_RDONLY)) {
948                 down_write(&nilfs->ns_sem);
949                 nilfs_setup_super(sbi);
950                 up_write(&nilfs->ns_sem);
951         }
952
953         return 0;
954
955  failed_segctor:
956         nilfs_detach_segment_constructor(sbi);
957
958  failed_checkpoint:
959         nilfs_put_root(fsroot);
960
961  failed_unload:
962         iput(nilfs->ns_sufile);
963         iput(nilfs->ns_cpfile);
964         iput(nilfs->ns_dat);
965
966  failed_nilfs:
967         destroy_nilfs(nilfs);
968
969  failed_sbi:
970         sb->s_fs_info = NULL;
971         kfree(sbi);
972         return err;
973 }
974
975 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
976 {
977         struct nilfs_sb_info *sbi = NILFS_SB(sb);
978         struct the_nilfs *nilfs = sbi->s_nilfs;
979         unsigned long old_sb_flags;
980         struct nilfs_mount_options old_opts;
981         int err;
982
983         old_sb_flags = sb->s_flags;
984         old_opts.mount_opt = sbi->s_mount_opt;
985
986         if (!parse_options(data, sb, 1)) {
987                 err = -EINVAL;
988                 goto restore_opts;
989         }
990         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
991
992         err = -EINVAL;
993
994         if (!nilfs_valid_fs(nilfs)) {
995                 printk(KERN_WARNING "NILFS (device %s): couldn't "
996                        "remount because the filesystem is in an "
997                        "incomplete recovery state.\n", sb->s_id);
998                 goto restore_opts;
999         }
1000
1001         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1002                 goto out;
1003         if (*flags & MS_RDONLY) {
1004                 /* Shutting down the segment constructor */
1005                 nilfs_detach_segment_constructor(sbi);
1006                 sb->s_flags |= MS_RDONLY;
1007
1008                 /*
1009                  * Remounting a valid RW partition RDONLY, so set
1010                  * the RDONLY flag and then mark the partition as valid again.
1011                  */
1012                 down_write(&nilfs->ns_sem);
1013                 nilfs_cleanup_super(sbi);
1014                 up_write(&nilfs->ns_sem);
1015         } else {
1016                 __u64 features;
1017                 struct nilfs_root *root;
1018
1019                 /*
1020                  * Mounting a RDONLY partition read-write, so reread and
1021                  * store the current valid flag.  (It may have been changed
1022                  * by fsck since we originally mounted the partition.)
1023                  */
1024                 down_read(&nilfs->ns_sem);
1025                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1026                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1027                 up_read(&nilfs->ns_sem);
1028                 if (features) {
1029                         printk(KERN_WARNING "NILFS (device %s): couldn't "
1030                                "remount RDWR because of unsupported optional "
1031                                "features (%llx)\n",
1032                                sb->s_id, (unsigned long long)features);
1033                         err = -EROFS;
1034                         goto restore_opts;
1035                 }
1036
1037                 sb->s_flags &= ~MS_RDONLY;
1038
1039                 root = NILFS_I(sb->s_root->d_inode)->i_root;
1040                 err = nilfs_attach_segment_constructor(sbi, root);
1041                 if (err)
1042                         goto restore_opts;
1043
1044                 down_write(&nilfs->ns_sem);
1045                 nilfs_setup_super(sbi);
1046                 up_write(&nilfs->ns_sem);
1047         }
1048  out:
1049         return 0;
1050
1051  restore_opts:
1052         sb->s_flags = old_sb_flags;
1053         sbi->s_mount_opt = old_opts.mount_opt;
1054         return err;
1055 }
1056
1057 struct nilfs_super_data {
1058         struct block_device *bdev;
1059         struct nilfs_sb_info *sbi;
1060         __u64 cno;
1061         int flags;
1062 };
1063
1064 /**
1065  * nilfs_identify - pre-read mount options needed to identify mount instance
1066  * @data: mount options
1067  * @sd: nilfs_super_data
1068  */
1069 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1070 {
1071         char *p, *options = data;
1072         substring_t args[MAX_OPT_ARGS];
1073         int option, token;
1074         int ret = 0;
1075
1076         do {
1077                 p = strsep(&options, ",");
1078                 if (p != NULL && *p) {
1079                         token = match_token(p, tokens, args);
1080                         if (token == Opt_snapshot) {
1081                                 if (!(sd->flags & MS_RDONLY))
1082                                         ret++;
1083                                 else {
1084                                         ret = match_int(&args[0], &option);
1085                                         if (!ret) {
1086                                                 if (option > 0)
1087                                                         sd->cno = option;
1088                                                 else
1089                                                         ret++;
1090                                         }
1091                                 }
1092                         }
1093                         if (ret)
1094                                 printk(KERN_ERR
1095                                        "NILFS: invalid mount option: %s\n", p);
1096                 }
1097                 if (!options)
1098                         break;
1099                 BUG_ON(options == data);
1100                 *(options - 1) = ',';
1101         } while (!ret);
1102         return ret;
1103 }
1104
1105 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1106 {
1107         s->s_bdev = data;
1108         s->s_dev = s->s_bdev->bd_dev;
1109         return 0;
1110 }
1111
1112 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1113 {
1114         return (void *)s->s_bdev == data;
1115 }
1116
1117 static int
1118 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1119              const char *dev_name, void *data, struct vfsmount *mnt)
1120 {
1121         struct nilfs_super_data sd;
1122         struct super_block *s;
1123         fmode_t mode = FMODE_READ;
1124         struct dentry *root_dentry;
1125         int err, s_new = false;
1126
1127         if (!(flags & MS_RDONLY))
1128                 mode |= FMODE_WRITE;
1129
1130         sd.bdev = open_bdev_exclusive(dev_name, mode, fs_type);
1131         if (IS_ERR(sd.bdev))
1132                 return PTR_ERR(sd.bdev);
1133
1134         sd.cno = 0;
1135         sd.flags = flags;
1136         if (nilfs_identify((char *)data, &sd)) {
1137                 err = -EINVAL;
1138                 goto failed;
1139         }
1140
1141         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, sd.bdev);
1142         if (IS_ERR(s)) {
1143                 err = PTR_ERR(s);
1144                 goto failed;
1145         }
1146
1147         if (!s->s_root) {
1148                 char b[BDEVNAME_SIZE];
1149
1150                 s_new = true;
1151
1152                 /* New superblock instance created */
1153                 s->s_flags = flags;
1154                 s->s_mode = mode;
1155                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1156                 sb_set_blocksize(s, block_size(sd.bdev));
1157
1158                 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1159                 if (err)
1160                         goto failed_super;
1161
1162                 s->s_flags |= MS_ACTIVE;
1163         } else if (!sd.cno) {
1164                 int busy = false;
1165
1166                 if (nilfs_tree_was_touched(s->s_root)) {
1167                         busy = nilfs_try_to_shrink_tree(s->s_root);
1168                         if (busy && (flags ^ s->s_flags) & MS_RDONLY) {
1169                                 printk(KERN_ERR "NILFS: the device already "
1170                                        "has a %s mount.\n",
1171                                        (s->s_flags & MS_RDONLY) ?
1172                                        "read-only" : "read/write");
1173                                 err = -EBUSY;
1174                                 goto failed_super;
1175                         }
1176                 }
1177                 if (!busy) {
1178                         /*
1179                          * Try remount to setup mount states if the current
1180                          * tree is not mounted and only snapshots use this sb.
1181                          */
1182                         err = nilfs_remount(s, &flags, data);
1183                         if (err)
1184                                 goto failed_super;
1185                 }
1186         }
1187
1188         if (sd.cno) {
1189                 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1190                 if (err)
1191                         goto failed_super;
1192         } else {
1193                 root_dentry = dget(s->s_root);
1194         }
1195
1196         if (!s_new)
1197                 close_bdev_exclusive(sd.bdev, mode);
1198
1199         mnt->mnt_sb = s;
1200         mnt->mnt_root = root_dentry;
1201         return 0;
1202
1203  failed_super:
1204         deactivate_locked_super(s);
1205
1206  failed:
1207         if (!s_new)
1208                 close_bdev_exclusive(sd.bdev, mode);
1209         return err;
1210 }
1211
1212 struct file_system_type nilfs_fs_type = {
1213         .owner    = THIS_MODULE,
1214         .name     = "nilfs2",
1215         .get_sb   = nilfs_get_sb,
1216         .kill_sb  = kill_block_super,
1217         .fs_flags = FS_REQUIRES_DEV,
1218 };
1219
1220 static void nilfs_inode_init_once(void *obj)
1221 {
1222         struct nilfs_inode_info *ii = obj;
1223
1224         INIT_LIST_HEAD(&ii->i_dirty);
1225 #ifdef CONFIG_NILFS_XATTR
1226         init_rwsem(&ii->xattr_sem);
1227 #endif
1228         nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
1229         ii->i_bmap = &ii->i_bmap_data;
1230         inode_init_once(&ii->vfs_inode);
1231 }
1232
1233 static void nilfs_segbuf_init_once(void *obj)
1234 {
1235         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1236 }
1237
1238 static void nilfs_destroy_cachep(void)
1239 {
1240         if (nilfs_inode_cachep)
1241                 kmem_cache_destroy(nilfs_inode_cachep);
1242         if (nilfs_transaction_cachep)
1243                 kmem_cache_destroy(nilfs_transaction_cachep);
1244         if (nilfs_segbuf_cachep)
1245                 kmem_cache_destroy(nilfs_segbuf_cachep);
1246         if (nilfs_btree_path_cache)
1247                 kmem_cache_destroy(nilfs_btree_path_cache);
1248 }
1249
1250 static int __init nilfs_init_cachep(void)
1251 {
1252         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1253                         sizeof(struct nilfs_inode_info), 0,
1254                         SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1255         if (!nilfs_inode_cachep)
1256                 goto fail;
1257
1258         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1259                         sizeof(struct nilfs_transaction_info), 0,
1260                         SLAB_RECLAIM_ACCOUNT, NULL);
1261         if (!nilfs_transaction_cachep)
1262                 goto fail;
1263
1264         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1265                         sizeof(struct nilfs_segment_buffer), 0,
1266                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1267         if (!nilfs_segbuf_cachep)
1268                 goto fail;
1269
1270         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1271                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1272                         0, 0, NULL);
1273         if (!nilfs_btree_path_cache)
1274                 goto fail;
1275
1276         return 0;
1277
1278 fail:
1279         nilfs_destroy_cachep();
1280         return -ENOMEM;
1281 }
1282
1283 static int __init init_nilfs_fs(void)
1284 {
1285         int err;
1286
1287         err = nilfs_init_cachep();
1288         if (err)
1289                 goto fail;
1290
1291         err = register_filesystem(&nilfs_fs_type);
1292         if (err)
1293                 goto free_cachep;
1294
1295         printk(KERN_INFO "NILFS version 2 loaded\n");
1296         return 0;
1297
1298 free_cachep:
1299         nilfs_destroy_cachep();
1300 fail:
1301         return err;
1302 }
1303
1304 static void __exit exit_nilfs_fs(void)
1305 {
1306         nilfs_destroy_cachep();
1307         unregister_filesystem(&nilfs_fs_type);
1308 }
1309
1310 module_init(init_nilfs_fs)
1311 module_exit(exit_nilfs_fs)