ramfs: fix memleak on no-mmu arch
[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/crc32.h>
47 #include <linux/vfs.h>
48 #include <linux/writeback.h>
49 #include <linux/seq_file.h>
50 #include <linux/mount.h>
51 #include "nilfs.h"
52 #include "export.h"
53 #include "mdt.h"
54 #include "alloc.h"
55 #include "btree.h"
56 #include "btnode.h"
57 #include "page.h"
58 #include "cpfile.h"
59 #include "ifile.h"
60 #include "dat.h"
61 #include "segment.h"
62 #include "segbuf.h"
63
64 MODULE_AUTHOR("NTT Corp.");
65 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
66                    "(NILFS)");
67 MODULE_LICENSE("GPL");
68
69 static struct kmem_cache *nilfs_inode_cachep;
70 struct kmem_cache *nilfs_transaction_cachep;
71 struct kmem_cache *nilfs_segbuf_cachep;
72 struct kmem_cache *nilfs_btree_path_cache;
73
74 static int nilfs_setup_super(struct super_block *sb, int is_mount);
75 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
76
77 static void nilfs_set_error(struct super_block *sb)
78 {
79         struct the_nilfs *nilfs = sb->s_fs_info;
80         struct nilfs_super_block **sbp;
81
82         down_write(&nilfs->ns_sem);
83         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
84                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
85                 sbp = nilfs_prepare_super(sb, 0);
86                 if (likely(sbp)) {
87                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
88                         if (sbp[1])
89                                 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
90                         nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
91                 }
92         }
93         up_write(&nilfs->ns_sem);
94 }
95
96 /**
97  * nilfs_error() - report failure condition on a filesystem
98  *
99  * nilfs_error() sets an ERROR_FS flag on the superblock as well as
100  * reporting an error message.  It should be called when NILFS detects
101  * incoherences or defects of meta data on disk.  As for sustainable
102  * errors such as a single-shot I/O error, nilfs_warning() or the printk()
103  * function should be used instead.
104  *
105  * The segment constructor must not call this function because it can
106  * kill itself.
107  */
108 void nilfs_error(struct super_block *sb, const char *function,
109                  const char *fmt, ...)
110 {
111         struct the_nilfs *nilfs = sb->s_fs_info;
112         struct va_format vaf;
113         va_list args;
114
115         va_start(args, fmt);
116
117         vaf.fmt = fmt;
118         vaf.va = &args;
119
120         printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
121                sb->s_id, function, &vaf);
122
123         va_end(args);
124
125         if (!(sb->s_flags & MS_RDONLY)) {
126                 nilfs_set_error(sb);
127
128                 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
129                         printk(KERN_CRIT "Remounting filesystem read-only\n");
130                         sb->s_flags |= MS_RDONLY;
131                 }
132         }
133
134         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
135                 panic("NILFS (device %s): panic forced after error\n",
136                       sb->s_id);
137 }
138
139 void nilfs_warning(struct super_block *sb, const char *function,
140                    const char *fmt, ...)
141 {
142         struct va_format vaf;
143         va_list args;
144
145         va_start(args, fmt);
146
147         vaf.fmt = fmt;
148         vaf.va = &args;
149
150         printk(KERN_WARNING "NILFS warning (device %s): %s: %pV\n",
151                sb->s_id, function, &vaf);
152
153         va_end(args);
154 }
155
156
157 struct inode *nilfs_alloc_inode(struct super_block *sb)
158 {
159         struct nilfs_inode_info *ii;
160
161         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
162         if (!ii)
163                 return NULL;
164         ii->i_bh = NULL;
165         ii->i_state = 0;
166         ii->i_cno = 0;
167         ii->vfs_inode.i_version = 1;
168         nilfs_btnode_cache_init(&ii->i_btnode_cache, sb->s_bdi);
169         return &ii->vfs_inode;
170 }
171
172 static void nilfs_i_callback(struct rcu_head *head)
173 {
174         struct inode *inode = container_of(head, struct inode, i_rcu);
175         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
176
177         INIT_LIST_HEAD(&inode->i_dentry);
178
179         if (mdi) {
180                 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
181                 kfree(mdi);
182         }
183         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
184 }
185
186 void nilfs_destroy_inode(struct inode *inode)
187 {
188         call_rcu(&inode->i_rcu, nilfs_i_callback);
189 }
190
191 static int nilfs_sync_super(struct super_block *sb, int flag)
192 {
193         struct the_nilfs *nilfs = sb->s_fs_info;
194         int err;
195
196  retry:
197         set_buffer_dirty(nilfs->ns_sbh[0]);
198         if (nilfs_test_opt(nilfs, BARRIER)) {
199                 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
200                                           WRITE_SYNC | WRITE_FLUSH_FUA);
201         } else {
202                 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
203         }
204
205         if (unlikely(err)) {
206                 printk(KERN_ERR
207                        "NILFS: unable to write superblock (err=%d)\n", err);
208                 if (err == -EIO && nilfs->ns_sbh[1]) {
209                         /*
210                          * sbp[0] points to newer log than sbp[1],
211                          * so copy sbp[0] to sbp[1] to take over sbp[0].
212                          */
213                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
214                                nilfs->ns_sbsize);
215                         nilfs_fall_back_super_block(nilfs);
216                         goto retry;
217                 }
218         } else {
219                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
220
221                 nilfs->ns_sbwcount++;
222
223                 /*
224                  * The latest segment becomes trailable from the position
225                  * written in superblock.
226                  */
227                 clear_nilfs_discontinued(nilfs);
228
229                 /* update GC protection for recent segments */
230                 if (nilfs->ns_sbh[1]) {
231                         if (flag == NILFS_SB_COMMIT_ALL) {
232                                 set_buffer_dirty(nilfs->ns_sbh[1]);
233                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
234                                         goto out;
235                         }
236                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
237                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
238                                 sbp = nilfs->ns_sbp[1];
239                 }
240
241                 spin_lock(&nilfs->ns_last_segment_lock);
242                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
243                 spin_unlock(&nilfs->ns_last_segment_lock);
244         }
245  out:
246         return err;
247 }
248
249 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
250                           struct the_nilfs *nilfs)
251 {
252         sector_t nfreeblocks;
253
254         /* nilfs->ns_sem must be locked by the caller. */
255         nilfs_count_free_blocks(nilfs, &nfreeblocks);
256         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
257
258         spin_lock(&nilfs->ns_last_segment_lock);
259         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
260         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
261         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
262         spin_unlock(&nilfs->ns_last_segment_lock);
263 }
264
265 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
266                                                int flip)
267 {
268         struct the_nilfs *nilfs = sb->s_fs_info;
269         struct nilfs_super_block **sbp = nilfs->ns_sbp;
270
271         /* nilfs->ns_sem must be locked by the caller. */
272         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
273                 if (sbp[1] &&
274                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
275                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
276                 } else {
277                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
278                                sb->s_id);
279                         return NULL;
280                 }
281         } else if (sbp[1] &&
282                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
283                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
284         }
285
286         if (flip && sbp[1])
287                 nilfs_swap_super_block(nilfs);
288
289         return sbp;
290 }
291
292 int nilfs_commit_super(struct super_block *sb, int flag)
293 {
294         struct the_nilfs *nilfs = sb->s_fs_info;
295         struct nilfs_super_block **sbp = nilfs->ns_sbp;
296         time_t t;
297
298         /* nilfs->ns_sem must be locked by the caller. */
299         t = get_seconds();
300         nilfs->ns_sbwtime = t;
301         sbp[0]->s_wtime = cpu_to_le64(t);
302         sbp[0]->s_sum = 0;
303         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
304                                              (unsigned char *)sbp[0],
305                                              nilfs->ns_sbsize));
306         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
307                 sbp[1]->s_wtime = sbp[0]->s_wtime;
308                 sbp[1]->s_sum = 0;
309                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
310                                             (unsigned char *)sbp[1],
311                                             nilfs->ns_sbsize));
312         }
313         clear_nilfs_sb_dirty(nilfs);
314         return nilfs_sync_super(sb, flag);
315 }
316
317 /**
318  * nilfs_cleanup_super() - write filesystem state for cleanup
319  * @sb: super block instance to be unmounted or degraded to read-only
320  *
321  * This function restores state flags in the on-disk super block.
322  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
323  * filesystem was not clean previously.
324  */
325 int nilfs_cleanup_super(struct super_block *sb)
326 {
327         struct the_nilfs *nilfs = sb->s_fs_info;
328         struct nilfs_super_block **sbp;
329         int flag = NILFS_SB_COMMIT;
330         int ret = -EIO;
331
332         sbp = nilfs_prepare_super(sb, 0);
333         if (sbp) {
334                 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
335                 nilfs_set_log_cursor(sbp[0], nilfs);
336                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
337                         /*
338                          * make the "clean" flag also to the opposite
339                          * super block if both super blocks point to
340                          * the same checkpoint.
341                          */
342                         sbp[1]->s_state = sbp[0]->s_state;
343                         flag = NILFS_SB_COMMIT_ALL;
344                 }
345                 ret = nilfs_commit_super(sb, flag);
346         }
347         return ret;
348 }
349
350 static void nilfs_put_super(struct super_block *sb)
351 {
352         struct the_nilfs *nilfs = sb->s_fs_info;
353
354         nilfs_detach_log_writer(sb);
355
356         if (!(sb->s_flags & MS_RDONLY)) {
357                 down_write(&nilfs->ns_sem);
358                 nilfs_cleanup_super(sb);
359                 up_write(&nilfs->ns_sem);
360         }
361
362         iput(nilfs->ns_sufile);
363         iput(nilfs->ns_cpfile);
364         iput(nilfs->ns_dat);
365
366         destroy_nilfs(nilfs);
367         sb->s_fs_info = NULL;
368 }
369
370 static int nilfs_sync_fs(struct super_block *sb, int wait)
371 {
372         struct the_nilfs *nilfs = sb->s_fs_info;
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(sb, nilfs_sb_will_flip(nilfs));
383                 if (likely(sbp)) {
384                         nilfs_set_log_cursor(sbp[0], nilfs);
385                         nilfs_commit_super(sb, NILFS_SB_COMMIT);
386                 }
387         }
388         up_write(&nilfs->ns_sem);
389
390         return err;
391 }
392
393 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
394                             struct nilfs_root **rootp)
395 {
396         struct the_nilfs *nilfs = sb->s_fs_info;
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(sb, 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_freeze(struct super_block *sb)
448 {
449         struct the_nilfs *nilfs = sb->s_fs_info;
450         int err;
451
452         if (sb->s_flags & MS_RDONLY)
453                 return 0;
454
455         /* Mark super block clean */
456         down_write(&nilfs->ns_sem);
457         err = nilfs_cleanup_super(sb);
458         up_write(&nilfs->ns_sem);
459         return err;
460 }
461
462 static int nilfs_unfreeze(struct super_block *sb)
463 {
464         struct the_nilfs *nilfs = sb->s_fs_info;
465
466         if (sb->s_flags & MS_RDONLY)
467                 return 0;
468
469         down_write(&nilfs->ns_sem);
470         nilfs_setup_super(sb, false);
471         up_write(&nilfs->ns_sem);
472         return 0;
473 }
474
475 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
476 {
477         struct super_block *sb = dentry->d_sb;
478         struct nilfs_root *root = NILFS_I(dentry->d_inode)->i_root;
479         struct the_nilfs *nilfs = root->nilfs;
480         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
481         unsigned long long blocks;
482         unsigned long overhead;
483         unsigned long nrsvblocks;
484         sector_t nfreeblocks;
485         int err;
486
487         /*
488          * Compute all of the segment blocks
489          *
490          * The blocks before first segment and after last segment
491          * are excluded.
492          */
493         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
494                 - nilfs->ns_first_data_block;
495         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
496
497         /*
498          * Compute the overhead
499          *
500          * When distributing meta data blocks outside segment structure,
501          * We must count them as the overhead.
502          */
503         overhead = 0;
504
505         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
506         if (unlikely(err))
507                 return err;
508
509         buf->f_type = NILFS_SUPER_MAGIC;
510         buf->f_bsize = sb->s_blocksize;
511         buf->f_blocks = blocks - overhead;
512         buf->f_bfree = nfreeblocks;
513         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
514                 (buf->f_bfree - nrsvblocks) : 0;
515         buf->f_files = atomic_read(&root->inodes_count);
516         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
517         buf->f_namelen = NILFS_NAME_LEN;
518         buf->f_fsid.val[0] = (u32)id;
519         buf->f_fsid.val[1] = (u32)(id >> 32);
520
521         return 0;
522 }
523
524 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
525 {
526         struct super_block *sb = vfs->mnt_sb;
527         struct the_nilfs *nilfs = sb->s_fs_info;
528         struct nilfs_root *root = NILFS_I(vfs->mnt_root->d_inode)->i_root;
529
530         if (!nilfs_test_opt(nilfs, BARRIER))
531                 seq_puts(seq, ",nobarrier");
532         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
533                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
534         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
535                 seq_puts(seq, ",errors=panic");
536         if (nilfs_test_opt(nilfs, ERRORS_CONT))
537                 seq_puts(seq, ",errors=continue");
538         if (nilfs_test_opt(nilfs, STRICT_ORDER))
539                 seq_puts(seq, ",order=strict");
540         if (nilfs_test_opt(nilfs, NORECOVERY))
541                 seq_puts(seq, ",norecovery");
542         if (nilfs_test_opt(nilfs, DISCARD))
543                 seq_puts(seq, ",discard");
544
545         return 0;
546 }
547
548 static const struct super_operations nilfs_sops = {
549         .alloc_inode    = nilfs_alloc_inode,
550         .destroy_inode  = nilfs_destroy_inode,
551         .dirty_inode    = nilfs_dirty_inode,
552         /* .write_inode    = nilfs_write_inode, */
553         /* .put_inode      = nilfs_put_inode, */
554         /* .drop_inode    = nilfs_drop_inode, */
555         .evict_inode    = nilfs_evict_inode,
556         .put_super      = nilfs_put_super,
557         /* .write_super    = nilfs_write_super, */
558         .sync_fs        = nilfs_sync_fs,
559         .freeze_fs      = nilfs_freeze,
560         .unfreeze_fs    = nilfs_unfreeze,
561         /* .write_super_lockfs */
562         /* .unlockfs */
563         .statfs         = nilfs_statfs,
564         .remount_fs     = nilfs_remount,
565         /* .umount_begin */
566         .show_options = nilfs_show_options
567 };
568
569 enum {
570         Opt_err_cont, Opt_err_panic, Opt_err_ro,
571         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
572         Opt_discard, Opt_nodiscard, Opt_err,
573 };
574
575 static match_table_t tokens = {
576         {Opt_err_cont, "errors=continue"},
577         {Opt_err_panic, "errors=panic"},
578         {Opt_err_ro, "errors=remount-ro"},
579         {Opt_barrier, "barrier"},
580         {Opt_nobarrier, "nobarrier"},
581         {Opt_snapshot, "cp=%u"},
582         {Opt_order, "order=%s"},
583         {Opt_norecovery, "norecovery"},
584         {Opt_discard, "discard"},
585         {Opt_nodiscard, "nodiscard"},
586         {Opt_err, NULL}
587 };
588
589 static int parse_options(char *options, struct super_block *sb, int is_remount)
590 {
591         struct the_nilfs *nilfs = sb->s_fs_info;
592         char *p;
593         substring_t args[MAX_OPT_ARGS];
594
595         if (!options)
596                 return 1;
597
598         while ((p = strsep(&options, ",")) != NULL) {
599                 int token;
600                 if (!*p)
601                         continue;
602
603                 token = match_token(p, tokens, args);
604                 switch (token) {
605                 case Opt_barrier:
606                         nilfs_set_opt(nilfs, BARRIER);
607                         break;
608                 case Opt_nobarrier:
609                         nilfs_clear_opt(nilfs, BARRIER);
610                         break;
611                 case Opt_order:
612                         if (strcmp(args[0].from, "relaxed") == 0)
613                                 /* Ordered data semantics */
614                                 nilfs_clear_opt(nilfs, STRICT_ORDER);
615                         else if (strcmp(args[0].from, "strict") == 0)
616                                 /* Strict in-order semantics */
617                                 nilfs_set_opt(nilfs, STRICT_ORDER);
618                         else
619                                 return 0;
620                         break;
621                 case Opt_err_panic:
622                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
623                         break;
624                 case Opt_err_ro:
625                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
626                         break;
627                 case Opt_err_cont:
628                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
629                         break;
630                 case Opt_snapshot:
631                         if (is_remount) {
632                                 printk(KERN_ERR
633                                        "NILFS: \"%s\" option is invalid "
634                                        "for remount.\n", p);
635                                 return 0;
636                         }
637                         break;
638                 case Opt_norecovery:
639                         nilfs_set_opt(nilfs, NORECOVERY);
640                         break;
641                 case Opt_discard:
642                         nilfs_set_opt(nilfs, DISCARD);
643                         break;
644                 case Opt_nodiscard:
645                         nilfs_clear_opt(nilfs, DISCARD);
646                         break;
647                 default:
648                         printk(KERN_ERR
649                                "NILFS: Unrecognized mount option \"%s\"\n", p);
650                         return 0;
651                 }
652         }
653         return 1;
654 }
655
656 static inline void
657 nilfs_set_default_options(struct super_block *sb,
658                           struct nilfs_super_block *sbp)
659 {
660         struct the_nilfs *nilfs = sb->s_fs_info;
661
662         nilfs->ns_mount_opt =
663                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
664 }
665
666 static int nilfs_setup_super(struct super_block *sb, int is_mount)
667 {
668         struct the_nilfs *nilfs = sb->s_fs_info;
669         struct nilfs_super_block **sbp;
670         int max_mnt_count;
671         int mnt_count;
672
673         /* nilfs->ns_sem must be locked by the caller. */
674         sbp = nilfs_prepare_super(sb, 0);
675         if (!sbp)
676                 return -EIO;
677
678         if (!is_mount)
679                 goto skip_mount_setup;
680
681         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
682         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
683
684         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
685                 printk(KERN_WARNING
686                        "NILFS warning: mounting fs with errors\n");
687 #if 0
688         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
689                 printk(KERN_WARNING
690                        "NILFS warning: maximal mount count reached\n");
691 #endif
692         }
693         if (!max_mnt_count)
694                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
695
696         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
697         sbp[0]->s_mtime = cpu_to_le64(get_seconds());
698
699 skip_mount_setup:
700         sbp[0]->s_state =
701                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
702         /* synchronize sbp[1] with sbp[0] */
703         if (sbp[1])
704                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
705         return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
706 }
707
708 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
709                                                  u64 pos, int blocksize,
710                                                  struct buffer_head **pbh)
711 {
712         unsigned long long sb_index = pos;
713         unsigned long offset;
714
715         offset = do_div(sb_index, blocksize);
716         *pbh = sb_bread(sb, sb_index);
717         if (!*pbh)
718                 return NULL;
719         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
720 }
721
722 int nilfs_store_magic_and_option(struct super_block *sb,
723                                  struct nilfs_super_block *sbp,
724                                  char *data)
725 {
726         struct the_nilfs *nilfs = sb->s_fs_info;
727
728         sb->s_magic = le16_to_cpu(sbp->s_magic);
729
730         /* FS independent flags */
731 #ifdef NILFS_ATIME_DISABLE
732         sb->s_flags |= MS_NOATIME;
733 #endif
734
735         nilfs_set_default_options(sb, sbp);
736
737         nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
738         nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
739         nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
740         nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
741
742         return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
743 }
744
745 int nilfs_check_feature_compatibility(struct super_block *sb,
746                                       struct nilfs_super_block *sbp)
747 {
748         __u64 features;
749
750         features = le64_to_cpu(sbp->s_feature_incompat) &
751                 ~NILFS_FEATURE_INCOMPAT_SUPP;
752         if (features) {
753                 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
754                        "optional features (%llx)\n",
755                        (unsigned long long)features);
756                 return -EINVAL;
757         }
758         features = le64_to_cpu(sbp->s_feature_compat_ro) &
759                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
760         if (!(sb->s_flags & MS_RDONLY) && features) {
761                 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
762                        "unsupported optional features (%llx)\n",
763                        (unsigned long long)features);
764                 return -EINVAL;
765         }
766         return 0;
767 }
768
769 static int nilfs_get_root_dentry(struct super_block *sb,
770                                  struct nilfs_root *root,
771                                  struct dentry **root_dentry)
772 {
773         struct inode *inode;
774         struct dentry *dentry;
775         int ret = 0;
776
777         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
778         if (IS_ERR(inode)) {
779                 printk(KERN_ERR "NILFS: get root inode failed\n");
780                 ret = PTR_ERR(inode);
781                 goto out;
782         }
783         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
784                 iput(inode);
785                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
786                 ret = -EINVAL;
787                 goto out;
788         }
789
790         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
791                 dentry = d_find_alias(inode);
792                 if (!dentry) {
793                         dentry = d_alloc_root(inode);
794                         if (!dentry) {
795                                 iput(inode);
796                                 ret = -ENOMEM;
797                                 goto failed_dentry;
798                         }
799                 } else {
800                         iput(inode);
801                 }
802         } else {
803                 dentry = d_obtain_alias(inode);
804                 if (IS_ERR(dentry)) {
805                         ret = PTR_ERR(dentry);
806                         goto failed_dentry;
807                 }
808         }
809         *root_dentry = dentry;
810  out:
811         return ret;
812
813  failed_dentry:
814         printk(KERN_ERR "NILFS: get root dentry failed\n");
815         goto out;
816 }
817
818 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
819                                  struct dentry **root_dentry)
820 {
821         struct the_nilfs *nilfs = s->s_fs_info;
822         struct nilfs_root *root;
823         int ret;
824
825         down_read(&nilfs->ns_segctor_sem);
826         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
827         up_read(&nilfs->ns_segctor_sem);
828         if (ret < 0) {
829                 ret = (ret == -ENOENT) ? -EINVAL : ret;
830                 goto out;
831         } else if (!ret) {
832                 printk(KERN_ERR "NILFS: The specified checkpoint is "
833                        "not a snapshot (checkpoint number=%llu).\n",
834                        (unsigned long long)cno);
835                 ret = -EINVAL;
836                 goto out;
837         }
838
839         ret = nilfs_attach_checkpoint(s, cno, false, &root);
840         if (ret) {
841                 printk(KERN_ERR "NILFS: error loading snapshot "
842                        "(checkpoint number=%llu).\n",
843                (unsigned long long)cno);
844                 goto out;
845         }
846         ret = nilfs_get_root_dentry(s, root, root_dentry);
847         nilfs_put_root(root);
848  out:
849         return ret;
850 }
851
852 static int nilfs_tree_was_touched(struct dentry *root_dentry)
853 {
854         return root_dentry->d_count > 1;
855 }
856
857 /**
858  * nilfs_try_to_shrink_tree() - try to shrink dentries of a checkpoint
859  * @root_dentry: root dentry of the tree to be shrunk
860  *
861  * This function returns true if the tree was in-use.
862  */
863 static int nilfs_try_to_shrink_tree(struct dentry *root_dentry)
864 {
865         if (have_submounts(root_dentry))
866                 return true;
867         shrink_dcache_parent(root_dentry);
868         return nilfs_tree_was_touched(root_dentry);
869 }
870
871 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
872 {
873         struct the_nilfs *nilfs = sb->s_fs_info;
874         struct nilfs_root *root;
875         struct inode *inode;
876         struct dentry *dentry;
877         int ret;
878
879         if (cno < 0 || cno > nilfs->ns_cno)
880                 return false;
881
882         if (cno >= nilfs_last_cno(nilfs))
883                 return true;    /* protect recent checkpoints */
884
885         ret = false;
886         root = nilfs_lookup_root(nilfs, cno);
887         if (root) {
888                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
889                 if (inode) {
890                         dentry = d_find_alias(inode);
891                         if (dentry) {
892                                 if (nilfs_tree_was_touched(dentry))
893                                         ret = nilfs_try_to_shrink_tree(dentry);
894                                 dput(dentry);
895                         }
896                         iput(inode);
897                 }
898                 nilfs_put_root(root);
899         }
900         return ret;
901 }
902
903 /**
904  * nilfs_fill_super() - initialize a super block instance
905  * @sb: super_block
906  * @data: mount options
907  * @silent: silent mode flag
908  *
909  * This function is called exclusively by nilfs->ns_mount_mutex.
910  * So, the recovery process is protected from other simultaneous mounts.
911  */
912 static int
913 nilfs_fill_super(struct super_block *sb, void *data, int silent)
914 {
915         struct the_nilfs *nilfs;
916         struct nilfs_root *fsroot;
917         struct backing_dev_info *bdi;
918         __u64 cno;
919         int err;
920
921         nilfs = alloc_nilfs(sb->s_bdev);
922         if (!nilfs)
923                 return -ENOMEM;
924
925         sb->s_fs_info = nilfs;
926
927         err = init_nilfs(nilfs, sb, (char *)data);
928         if (err)
929                 goto failed_nilfs;
930
931         sb->s_op = &nilfs_sops;
932         sb->s_export_op = &nilfs_export_ops;
933         sb->s_root = NULL;
934         sb->s_time_gran = 1;
935
936         bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
937         sb->s_bdi = bdi ? : &default_backing_dev_info;
938
939         err = load_nilfs(nilfs, sb);
940         if (err)
941                 goto failed_nilfs;
942
943         cno = nilfs_last_cno(nilfs);
944         err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
945         if (err) {
946                 printk(KERN_ERR "NILFS: error loading last checkpoint "
947                        "(checkpoint number=%llu).\n", (unsigned long long)cno);
948                 goto failed_unload;
949         }
950
951         if (!(sb->s_flags & MS_RDONLY)) {
952                 err = nilfs_attach_log_writer(sb, fsroot);
953                 if (err)
954                         goto failed_checkpoint;
955         }
956
957         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
958         if (err)
959                 goto failed_segctor;
960
961         nilfs_put_root(fsroot);
962
963         if (!(sb->s_flags & MS_RDONLY)) {
964                 down_write(&nilfs->ns_sem);
965                 nilfs_setup_super(sb, true);
966                 up_write(&nilfs->ns_sem);
967         }
968
969         return 0;
970
971  failed_segctor:
972         nilfs_detach_log_writer(sb);
973
974  failed_checkpoint:
975         nilfs_put_root(fsroot);
976
977  failed_unload:
978         iput(nilfs->ns_sufile);
979         iput(nilfs->ns_cpfile);
980         iput(nilfs->ns_dat);
981
982  failed_nilfs:
983         destroy_nilfs(nilfs);
984         return err;
985 }
986
987 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
988 {
989         struct the_nilfs *nilfs = sb->s_fs_info;
990         unsigned long old_sb_flags;
991         unsigned long old_mount_opt;
992         int err;
993
994         old_sb_flags = sb->s_flags;
995         old_mount_opt = nilfs->ns_mount_opt;
996
997         if (!parse_options(data, sb, 1)) {
998                 err = -EINVAL;
999                 goto restore_opts;
1000         }
1001         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
1002
1003         err = -EINVAL;
1004
1005         if (!nilfs_valid_fs(nilfs)) {
1006                 printk(KERN_WARNING "NILFS (device %s): couldn't "
1007                        "remount because the filesystem is in an "
1008                        "incomplete recovery state.\n", sb->s_id);
1009                 goto restore_opts;
1010         }
1011
1012         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1013                 goto out;
1014         if (*flags & MS_RDONLY) {
1015                 /* Shutting down log writer */
1016                 nilfs_detach_log_writer(sb);
1017                 sb->s_flags |= MS_RDONLY;
1018
1019                 /*
1020                  * Remounting a valid RW partition RDONLY, so set
1021                  * the RDONLY flag and then mark the partition as valid again.
1022                  */
1023                 down_write(&nilfs->ns_sem);
1024                 nilfs_cleanup_super(sb);
1025                 up_write(&nilfs->ns_sem);
1026         } else {
1027                 __u64 features;
1028                 struct nilfs_root *root;
1029
1030                 /*
1031                  * Mounting a RDONLY partition read-write, so reread and
1032                  * store the current valid flag.  (It may have been changed
1033                  * by fsck since we originally mounted the partition.)
1034                  */
1035                 down_read(&nilfs->ns_sem);
1036                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1037                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1038                 up_read(&nilfs->ns_sem);
1039                 if (features) {
1040                         printk(KERN_WARNING "NILFS (device %s): couldn't "
1041                                "remount RDWR because of unsupported optional "
1042                                "features (%llx)\n",
1043                                sb->s_id, (unsigned long long)features);
1044                         err = -EROFS;
1045                         goto restore_opts;
1046                 }
1047
1048                 sb->s_flags &= ~MS_RDONLY;
1049
1050                 root = NILFS_I(sb->s_root->d_inode)->i_root;
1051                 err = nilfs_attach_log_writer(sb, root);
1052                 if (err)
1053                         goto restore_opts;
1054
1055                 down_write(&nilfs->ns_sem);
1056                 nilfs_setup_super(sb, true);
1057                 up_write(&nilfs->ns_sem);
1058         }
1059  out:
1060         return 0;
1061
1062  restore_opts:
1063         sb->s_flags = old_sb_flags;
1064         nilfs->ns_mount_opt = old_mount_opt;
1065         return err;
1066 }
1067
1068 struct nilfs_super_data {
1069         struct block_device *bdev;
1070         __u64 cno;
1071         int flags;
1072 };
1073
1074 /**
1075  * nilfs_identify - pre-read mount options needed to identify mount instance
1076  * @data: mount options
1077  * @sd: nilfs_super_data
1078  */
1079 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1080 {
1081         char *p, *options = data;
1082         substring_t args[MAX_OPT_ARGS];
1083         int token;
1084         int ret = 0;
1085
1086         do {
1087                 p = strsep(&options, ",");
1088                 if (p != NULL && *p) {
1089                         token = match_token(p, tokens, args);
1090                         if (token == Opt_snapshot) {
1091                                 if (!(sd->flags & MS_RDONLY)) {
1092                                         ret++;
1093                                 } else {
1094                                         sd->cno = simple_strtoull(args[0].from,
1095                                                                   NULL, 0);
1096                                         /*
1097                                          * No need to see the end pointer;
1098                                          * match_token() has done syntax
1099                                          * checking.
1100                                          */
1101                                         if (sd->cno == 0)
1102                                                 ret++;
1103                                 }
1104                         }
1105                         if (ret)
1106                                 printk(KERN_ERR
1107                                        "NILFS: invalid mount option: %s\n", p);
1108                 }
1109                 if (!options)
1110                         break;
1111                 BUG_ON(options == data);
1112                 *(options - 1) = ',';
1113         } while (!ret);
1114         return ret;
1115 }
1116
1117 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1118 {
1119         s->s_bdev = data;
1120         s->s_dev = s->s_bdev->bd_dev;
1121         return 0;
1122 }
1123
1124 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1125 {
1126         return (void *)s->s_bdev == data;
1127 }
1128
1129 static struct dentry *
1130 nilfs_mount(struct file_system_type *fs_type, int flags,
1131              const char *dev_name, void *data)
1132 {
1133         struct nilfs_super_data sd;
1134         struct super_block *s;
1135         fmode_t mode = FMODE_READ | FMODE_EXCL;
1136         struct dentry *root_dentry;
1137         int err, s_new = false;
1138
1139         if (!(flags & MS_RDONLY))
1140                 mode |= FMODE_WRITE;
1141
1142         sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1143         if (IS_ERR(sd.bdev))
1144                 return ERR_CAST(sd.bdev);
1145
1146         sd.cno = 0;
1147         sd.flags = flags;
1148         if (nilfs_identify((char *)data, &sd)) {
1149                 err = -EINVAL;
1150                 goto failed;
1151         }
1152
1153         /*
1154          * once the super is inserted into the list by sget, s_umount
1155          * will protect the lockfs code from trying to start a snapshot
1156          * while we are mounting
1157          */
1158         mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1159         if (sd.bdev->bd_fsfreeze_count > 0) {
1160                 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1161                 err = -EBUSY;
1162                 goto failed;
1163         }
1164         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, sd.bdev);
1165         mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1166         if (IS_ERR(s)) {
1167                 err = PTR_ERR(s);
1168                 goto failed;
1169         }
1170
1171         if (!s->s_root) {
1172                 char b[BDEVNAME_SIZE];
1173
1174                 s_new = true;
1175
1176                 /* New superblock instance created */
1177                 s->s_flags = flags;
1178                 s->s_mode = mode;
1179                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1180                 sb_set_blocksize(s, block_size(sd.bdev));
1181
1182                 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1183                 if (err)
1184                         goto failed_super;
1185
1186                 s->s_flags |= MS_ACTIVE;
1187         } else if (!sd.cno) {
1188                 int busy = false;
1189
1190                 if (nilfs_tree_was_touched(s->s_root)) {
1191                         busy = nilfs_try_to_shrink_tree(s->s_root);
1192                         if (busy && (flags ^ s->s_flags) & MS_RDONLY) {
1193                                 printk(KERN_ERR "NILFS: the device already "
1194                                        "has a %s mount.\n",
1195                                        (s->s_flags & MS_RDONLY) ?
1196                                        "read-only" : "read/write");
1197                                 err = -EBUSY;
1198                                 goto failed_super;
1199                         }
1200                 }
1201                 if (!busy) {
1202                         /*
1203                          * Try remount to setup mount states if the current
1204                          * tree is not mounted and only snapshots use this sb.
1205                          */
1206                         err = nilfs_remount(s, &flags, data);
1207                         if (err)
1208                                 goto failed_super;
1209                 }
1210         }
1211
1212         if (sd.cno) {
1213                 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1214                 if (err)
1215                         goto failed_super;
1216         } else {
1217                 root_dentry = dget(s->s_root);
1218         }
1219
1220         if (!s_new)
1221                 blkdev_put(sd.bdev, mode);
1222
1223         return root_dentry;
1224
1225  failed_super:
1226         deactivate_locked_super(s);
1227
1228  failed:
1229         if (!s_new)
1230                 blkdev_put(sd.bdev, mode);
1231         return ERR_PTR(err);
1232 }
1233
1234 struct file_system_type nilfs_fs_type = {
1235         .owner    = THIS_MODULE,
1236         .name     = "nilfs2",
1237         .mount    = nilfs_mount,
1238         .kill_sb  = kill_block_super,
1239         .fs_flags = FS_REQUIRES_DEV,
1240 };
1241
1242 static void nilfs_inode_init_once(void *obj)
1243 {
1244         struct nilfs_inode_info *ii = obj;
1245
1246         INIT_LIST_HEAD(&ii->i_dirty);
1247 #ifdef CONFIG_NILFS_XATTR
1248         init_rwsem(&ii->xattr_sem);
1249 #endif
1250         address_space_init_once(&ii->i_btnode_cache);
1251         ii->i_bmap = &ii->i_bmap_data;
1252         inode_init_once(&ii->vfs_inode);
1253 }
1254
1255 static void nilfs_segbuf_init_once(void *obj)
1256 {
1257         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1258 }
1259
1260 static void nilfs_destroy_cachep(void)
1261 {
1262         if (nilfs_inode_cachep)
1263                 kmem_cache_destroy(nilfs_inode_cachep);
1264         if (nilfs_transaction_cachep)
1265                 kmem_cache_destroy(nilfs_transaction_cachep);
1266         if (nilfs_segbuf_cachep)
1267                 kmem_cache_destroy(nilfs_segbuf_cachep);
1268         if (nilfs_btree_path_cache)
1269                 kmem_cache_destroy(nilfs_btree_path_cache);
1270 }
1271
1272 static int __init nilfs_init_cachep(void)
1273 {
1274         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1275                         sizeof(struct nilfs_inode_info), 0,
1276                         SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1277         if (!nilfs_inode_cachep)
1278                 goto fail;
1279
1280         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1281                         sizeof(struct nilfs_transaction_info), 0,
1282                         SLAB_RECLAIM_ACCOUNT, NULL);
1283         if (!nilfs_transaction_cachep)
1284                 goto fail;
1285
1286         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1287                         sizeof(struct nilfs_segment_buffer), 0,
1288                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1289         if (!nilfs_segbuf_cachep)
1290                 goto fail;
1291
1292         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1293                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1294                         0, 0, NULL);
1295         if (!nilfs_btree_path_cache)
1296                 goto fail;
1297
1298         return 0;
1299
1300 fail:
1301         nilfs_destroy_cachep();
1302         return -ENOMEM;
1303 }
1304
1305 static int __init init_nilfs_fs(void)
1306 {
1307         int err;
1308
1309         err = nilfs_init_cachep();
1310         if (err)
1311                 goto fail;
1312
1313         err = register_filesystem(&nilfs_fs_type);
1314         if (err)
1315                 goto free_cachep;
1316
1317         printk(KERN_INFO "NILFS version 2 loaded\n");
1318         return 0;
1319
1320 free_cachep:
1321         nilfs_destroy_cachep();
1322 fail:
1323         return err;
1324 }
1325
1326 static void __exit exit_nilfs_fs(void)
1327 {
1328         nilfs_destroy_cachep();
1329         unregister_filesystem(&nilfs_fs_type);
1330 }
1331
1332 module_init(init_nilfs_fs)
1333 module_exit(exit_nilfs_fs)