7d257e78fbad197701f8ea2a25ba12c7494d85b3
[platform/adaptation/renesas_rcar/renesas_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 "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
60 #include "ifile.h"
61 #include "dat.h"
62 #include "segment.h"
63 #include "segbuf.h"
64
65 MODULE_AUTHOR("NTT Corp.");
66 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
67                    "(NILFS)");
68 MODULE_LICENSE("GPL");
69
70 static struct kmem_cache *nilfs_inode_cachep;
71 struct kmem_cache *nilfs_transaction_cachep;
72 struct kmem_cache *nilfs_segbuf_cachep;
73 struct kmem_cache *nilfs_btree_path_cache;
74
75 static int nilfs_setup_super(struct super_block *sb, int is_mount);
76 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
77
78 static void nilfs_set_error(struct super_block *sb)
79 {
80         struct the_nilfs *nilfs = sb->s_fs_info;
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(sb, 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(sb, 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 the_nilfs *nilfs = sb->s_fs_info;
113         struct va_format vaf;
114         va_list args;
115
116         va_start(args, fmt);
117
118         vaf.fmt = fmt;
119         vaf.va = &args;
120
121         printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
122                sb->s_id, function, &vaf);
123
124         va_end(args);
125
126         if (!(sb->s_flags & MS_RDONLY)) {
127                 nilfs_set_error(sb);
128
129                 if (nilfs_test_opt(nilfs, ERRORS_RO)) {
130                         printk(KERN_CRIT "Remounting filesystem read-only\n");
131                         sb->s_flags |= MS_RDONLY;
132                 }
133         }
134
135         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
136                 panic("NILFS (device %s): panic forced after error\n",
137                       sb->s_id);
138 }
139
140 void nilfs_warning(struct super_block *sb, const char *function,
141                    const char *fmt, ...)
142 {
143         struct va_format vaf;
144         va_list args;
145
146         va_start(args, fmt);
147
148         vaf.fmt = fmt;
149         vaf.va = &args;
150
151         printk(KERN_WARNING "NILFS warning (device %s): %s: %pV\n",
152                sb->s_id, function, &vaf);
153
154         va_end(args);
155 }
156
157
158 struct inode *nilfs_alloc_inode(struct super_block *sb)
159 {
160         struct nilfs_inode_info *ii;
161
162         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
163         if (!ii)
164                 return NULL;
165         ii->i_bh = NULL;
166         ii->i_state = 0;
167         ii->i_cno = 0;
168         ii->vfs_inode.i_version = 1;
169         nilfs_mapping_init(&ii->i_btnode_cache, &ii->vfs_inode, sb->s_bdi);
170         return &ii->vfs_inode;
171 }
172
173 static void nilfs_i_callback(struct rcu_head *head)
174 {
175         struct inode *inode = container_of(head, struct inode, i_rcu);
176         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
177
178         if (mdi) {
179                 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
180                 kfree(mdi);
181         }
182         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
183 }
184
185 void nilfs_destroy_inode(struct inode *inode)
186 {
187         call_rcu(&inode->i_rcu, nilfs_i_callback);
188 }
189
190 static int nilfs_sync_super(struct super_block *sb, int flag)
191 {
192         struct the_nilfs *nilfs = sb->s_fs_info;
193         int err;
194
195  retry:
196         set_buffer_dirty(nilfs->ns_sbh[0]);
197         if (nilfs_test_opt(nilfs, BARRIER)) {
198                 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
199                                           WRITE_SYNC | WRITE_FLUSH_FUA);
200         } else {
201                 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
202         }
203
204         if (unlikely(err)) {
205                 printk(KERN_ERR
206                        "NILFS: unable to write superblock (err=%d)\n", err);
207                 if (err == -EIO && nilfs->ns_sbh[1]) {
208                         /*
209                          * sbp[0] points to newer log than sbp[1],
210                          * so copy sbp[0] to sbp[1] to take over sbp[0].
211                          */
212                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
213                                nilfs->ns_sbsize);
214                         nilfs_fall_back_super_block(nilfs);
215                         goto retry;
216                 }
217         } else {
218                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
219
220                 nilfs->ns_sbwcount++;
221
222                 /*
223                  * The latest segment becomes trailable from the position
224                  * written in superblock.
225                  */
226                 clear_nilfs_discontinued(nilfs);
227
228                 /* update GC protection for recent segments */
229                 if (nilfs->ns_sbh[1]) {
230                         if (flag == NILFS_SB_COMMIT_ALL) {
231                                 set_buffer_dirty(nilfs->ns_sbh[1]);
232                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
233                                         goto out;
234                         }
235                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
236                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
237                                 sbp = nilfs->ns_sbp[1];
238                 }
239
240                 spin_lock(&nilfs->ns_last_segment_lock);
241                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
242                 spin_unlock(&nilfs->ns_last_segment_lock);
243         }
244  out:
245         return err;
246 }
247
248 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
249                           struct the_nilfs *nilfs)
250 {
251         sector_t nfreeblocks;
252
253         /* nilfs->ns_sem must be locked by the caller. */
254         nilfs_count_free_blocks(nilfs, &nfreeblocks);
255         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
256
257         spin_lock(&nilfs->ns_last_segment_lock);
258         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
259         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
260         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
261         spin_unlock(&nilfs->ns_last_segment_lock);
262 }
263
264 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
265                                                int flip)
266 {
267         struct the_nilfs *nilfs = sb->s_fs_info;
268         struct nilfs_super_block **sbp = nilfs->ns_sbp;
269
270         /* nilfs->ns_sem must be locked by the caller. */
271         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
272                 if (sbp[1] &&
273                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
274                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
275                 } else {
276                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
277                                sb->s_id);
278                         return NULL;
279                 }
280         } else if (sbp[1] &&
281                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
282                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
283         }
284
285         if (flip && sbp[1])
286                 nilfs_swap_super_block(nilfs);
287
288         return sbp;
289 }
290
291 int nilfs_commit_super(struct super_block *sb, int flag)
292 {
293         struct the_nilfs *nilfs = sb->s_fs_info;
294         struct nilfs_super_block **sbp = nilfs->ns_sbp;
295         time_t t;
296
297         /* nilfs->ns_sem must be locked by the caller. */
298         t = get_seconds();
299         nilfs->ns_sbwtime = t;
300         sbp[0]->s_wtime = cpu_to_le64(t);
301         sbp[0]->s_sum = 0;
302         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
303                                              (unsigned char *)sbp[0],
304                                              nilfs->ns_sbsize));
305         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
306                 sbp[1]->s_wtime = sbp[0]->s_wtime;
307                 sbp[1]->s_sum = 0;
308                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
309                                             (unsigned char *)sbp[1],
310                                             nilfs->ns_sbsize));
311         }
312         clear_nilfs_sb_dirty(nilfs);
313         return nilfs_sync_super(sb, flag);
314 }
315
316 /**
317  * nilfs_cleanup_super() - write filesystem state for cleanup
318  * @sb: super block instance to be unmounted or degraded to read-only
319  *
320  * This function restores state flags in the on-disk super block.
321  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
322  * filesystem was not clean previously.
323  */
324 int nilfs_cleanup_super(struct super_block *sb)
325 {
326         struct the_nilfs *nilfs = sb->s_fs_info;
327         struct nilfs_super_block **sbp;
328         int flag = NILFS_SB_COMMIT;
329         int ret = -EIO;
330
331         sbp = nilfs_prepare_super(sb, 0);
332         if (sbp) {
333                 sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
334                 nilfs_set_log_cursor(sbp[0], nilfs);
335                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
336                         /*
337                          * make the "clean" flag also to the opposite
338                          * super block if both super blocks point to
339                          * the same checkpoint.
340                          */
341                         sbp[1]->s_state = sbp[0]->s_state;
342                         flag = NILFS_SB_COMMIT_ALL;
343                 }
344                 ret = nilfs_commit_super(sb, flag);
345         }
346         return ret;
347 }
348
349 /**
350  * nilfs_move_2nd_super - relocate secondary super block
351  * @sb: super block instance
352  * @sb2off: new offset of the secondary super block (in bytes)
353  */
354 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
355 {
356         struct the_nilfs *nilfs = sb->s_fs_info;
357         struct buffer_head *nsbh;
358         struct nilfs_super_block *nsbp;
359         sector_t blocknr, newblocknr;
360         unsigned long offset;
361         int sb2i = -1;  /* array index of the secondary superblock */
362         int ret = 0;
363
364         /* nilfs->ns_sem must be locked by the caller. */
365         if (nilfs->ns_sbh[1] &&
366             nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
367                 sb2i = 1;
368                 blocknr = nilfs->ns_sbh[1]->b_blocknr;
369         } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
370                 sb2i = 0;
371                 blocknr = nilfs->ns_sbh[0]->b_blocknr;
372         }
373         if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
374                 goto out;  /* super block location is unchanged */
375
376         /* Get new super block buffer */
377         newblocknr = sb2off >> nilfs->ns_blocksize_bits;
378         offset = sb2off & (nilfs->ns_blocksize - 1);
379         nsbh = sb_getblk(sb, newblocknr);
380         if (!nsbh) {
381                 printk(KERN_WARNING
382                        "NILFS warning: unable to move secondary superblock "
383                        "to block %llu\n", (unsigned long long)newblocknr);
384                 ret = -EIO;
385                 goto out;
386         }
387         nsbp = (void *)nsbh->b_data + offset;
388         memset(nsbp, 0, nilfs->ns_blocksize);
389
390         if (sb2i >= 0) {
391                 memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
392                 brelse(nilfs->ns_sbh[sb2i]);
393                 nilfs->ns_sbh[sb2i] = nsbh;
394                 nilfs->ns_sbp[sb2i] = nsbp;
395         } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
396                 /* secondary super block will be restored to index 1 */
397                 nilfs->ns_sbh[1] = nsbh;
398                 nilfs->ns_sbp[1] = nsbp;
399         } else {
400                 brelse(nsbh);
401         }
402 out:
403         return ret;
404 }
405
406 /**
407  * nilfs_resize_fs - resize the filesystem
408  * @sb: super block instance
409  * @newsize: new size of the filesystem (in bytes)
410  */
411 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
412 {
413         struct the_nilfs *nilfs = sb->s_fs_info;
414         struct nilfs_super_block **sbp;
415         __u64 devsize, newnsegs;
416         loff_t sb2off;
417         int ret;
418
419         ret = -ERANGE;
420         devsize = i_size_read(sb->s_bdev->bd_inode);
421         if (newsize > devsize)
422                 goto out;
423
424         /*
425          * Write lock is required to protect some functions depending
426          * on the number of segments, the number of reserved segments,
427          * and so forth.
428          */
429         down_write(&nilfs->ns_segctor_sem);
430
431         sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
432         newnsegs = sb2off >> nilfs->ns_blocksize_bits;
433         do_div(newnsegs, nilfs->ns_blocks_per_segment);
434
435         ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
436         up_write(&nilfs->ns_segctor_sem);
437         if (ret < 0)
438                 goto out;
439
440         ret = nilfs_construct_segment(sb);
441         if (ret < 0)
442                 goto out;
443
444         down_write(&nilfs->ns_sem);
445         nilfs_move_2nd_super(sb, sb2off);
446         ret = -EIO;
447         sbp = nilfs_prepare_super(sb, 0);
448         if (likely(sbp)) {
449                 nilfs_set_log_cursor(sbp[0], nilfs);
450                 /*
451                  * Drop NILFS_RESIZE_FS flag for compatibility with
452                  * mount-time resize which may be implemented in a
453                  * future release.
454                  */
455                 sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
456                                               ~NILFS_RESIZE_FS);
457                 sbp[0]->s_dev_size = cpu_to_le64(newsize);
458                 sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
459                 if (sbp[1])
460                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
461                 ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
462         }
463         up_write(&nilfs->ns_sem);
464
465         /*
466          * Reset the range of allocatable segments last.  This order
467          * is important in the case of expansion because the secondary
468          * superblock must be protected from log write until migration
469          * completes.
470          */
471         if (!ret)
472                 nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
473 out:
474         return ret;
475 }
476
477 static void nilfs_put_super(struct super_block *sb)
478 {
479         struct the_nilfs *nilfs = sb->s_fs_info;
480
481         nilfs_detach_log_writer(sb);
482
483         if (!(sb->s_flags & MS_RDONLY)) {
484                 down_write(&nilfs->ns_sem);
485                 nilfs_cleanup_super(sb);
486                 up_write(&nilfs->ns_sem);
487         }
488
489         iput(nilfs->ns_sufile);
490         iput(nilfs->ns_cpfile);
491         iput(nilfs->ns_dat);
492
493         destroy_nilfs(nilfs);
494         sb->s_fs_info = NULL;
495 }
496
497 static int nilfs_sync_fs(struct super_block *sb, int wait)
498 {
499         struct the_nilfs *nilfs = sb->s_fs_info;
500         struct nilfs_super_block **sbp;
501         int err = 0;
502
503         /* This function is called when super block should be written back */
504         if (wait)
505                 err = nilfs_construct_segment(sb);
506
507         down_write(&nilfs->ns_sem);
508         if (nilfs_sb_dirty(nilfs)) {
509                 sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
510                 if (likely(sbp)) {
511                         nilfs_set_log_cursor(sbp[0], nilfs);
512                         nilfs_commit_super(sb, NILFS_SB_COMMIT);
513                 }
514         }
515         up_write(&nilfs->ns_sem);
516
517         return err;
518 }
519
520 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
521                             struct nilfs_root **rootp)
522 {
523         struct the_nilfs *nilfs = sb->s_fs_info;
524         struct nilfs_root *root;
525         struct nilfs_checkpoint *raw_cp;
526         struct buffer_head *bh_cp;
527         int err = -ENOMEM;
528
529         root = nilfs_find_or_create_root(
530                 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
531         if (!root)
532                 return err;
533
534         if (root->ifile)
535                 goto reuse; /* already attached checkpoint */
536
537         down_read(&nilfs->ns_segctor_sem);
538         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
539                                           &bh_cp);
540         up_read(&nilfs->ns_segctor_sem);
541         if (unlikely(err)) {
542                 if (err == -ENOENT || err == -EINVAL) {
543                         printk(KERN_ERR
544                                "NILFS: Invalid checkpoint "
545                                "(checkpoint number=%llu)\n",
546                                (unsigned long long)cno);
547                         err = -EINVAL;
548                 }
549                 goto failed;
550         }
551
552         err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
553                                &raw_cp->cp_ifile_inode, &root->ifile);
554         if (err)
555                 goto failed_bh;
556
557         atomic_set(&root->inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
558         atomic_set(&root->blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
559
560         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
561
562  reuse:
563         *rootp = root;
564         return 0;
565
566  failed_bh:
567         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
568  failed:
569         nilfs_put_root(root);
570
571         return err;
572 }
573
574 static int nilfs_freeze(struct super_block *sb)
575 {
576         struct the_nilfs *nilfs = sb->s_fs_info;
577         int err;
578
579         if (sb->s_flags & MS_RDONLY)
580                 return 0;
581
582         /* Mark super block clean */
583         down_write(&nilfs->ns_sem);
584         err = nilfs_cleanup_super(sb);
585         up_write(&nilfs->ns_sem);
586         return err;
587 }
588
589 static int nilfs_unfreeze(struct super_block *sb)
590 {
591         struct the_nilfs *nilfs = sb->s_fs_info;
592
593         if (sb->s_flags & MS_RDONLY)
594                 return 0;
595
596         down_write(&nilfs->ns_sem);
597         nilfs_setup_super(sb, false);
598         up_write(&nilfs->ns_sem);
599         return 0;
600 }
601
602 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
603 {
604         struct super_block *sb = dentry->d_sb;
605         struct nilfs_root *root = NILFS_I(dentry->d_inode)->i_root;
606         struct the_nilfs *nilfs = root->nilfs;
607         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
608         unsigned long long blocks;
609         unsigned long overhead;
610         unsigned long nrsvblocks;
611         sector_t nfreeblocks;
612         u64 nmaxinodes, nfreeinodes;
613         int err;
614
615         /*
616          * Compute all of the segment blocks
617          *
618          * The blocks before first segment and after last segment
619          * are excluded.
620          */
621         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
622                 - nilfs->ns_first_data_block;
623         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
624
625         /*
626          * Compute the overhead
627          *
628          * When distributing meta data blocks outside segment structure,
629          * We must count them as the overhead.
630          */
631         overhead = 0;
632
633         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
634         if (unlikely(err))
635                 return err;
636
637         err = nilfs_ifile_count_free_inodes(root->ifile,
638                                             &nmaxinodes, &nfreeinodes);
639         if (unlikely(err)) {
640                 printk(KERN_WARNING
641                         "NILFS warning: fail to count free inodes: err %d.\n",
642                         err);
643                 if (err == -ERANGE) {
644                         /*
645                          * If nilfs_palloc_count_max_entries() returns
646                          * -ERANGE error code then we simply treat
647                          * curent inodes count as maximum possible and
648                          * zero as free inodes value.
649                          */
650                         nmaxinodes = atomic_read(&root->inodes_count);
651                         nfreeinodes = 0;
652                         err = 0;
653                 } else
654                         return err;
655         }
656
657         buf->f_type = NILFS_SUPER_MAGIC;
658         buf->f_bsize = sb->s_blocksize;
659         buf->f_blocks = blocks - overhead;
660         buf->f_bfree = nfreeblocks;
661         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
662                 (buf->f_bfree - nrsvblocks) : 0;
663         buf->f_files = nmaxinodes;
664         buf->f_ffree = nfreeinodes;
665         buf->f_namelen = NILFS_NAME_LEN;
666         buf->f_fsid.val[0] = (u32)id;
667         buf->f_fsid.val[1] = (u32)(id >> 32);
668
669         return 0;
670 }
671
672 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
673 {
674         struct super_block *sb = dentry->d_sb;
675         struct the_nilfs *nilfs = sb->s_fs_info;
676         struct nilfs_root *root = NILFS_I(dentry->d_inode)->i_root;
677
678         if (!nilfs_test_opt(nilfs, BARRIER))
679                 seq_puts(seq, ",nobarrier");
680         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
681                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
682         if (nilfs_test_opt(nilfs, ERRORS_PANIC))
683                 seq_puts(seq, ",errors=panic");
684         if (nilfs_test_opt(nilfs, ERRORS_CONT))
685                 seq_puts(seq, ",errors=continue");
686         if (nilfs_test_opt(nilfs, STRICT_ORDER))
687                 seq_puts(seq, ",order=strict");
688         if (nilfs_test_opt(nilfs, NORECOVERY))
689                 seq_puts(seq, ",norecovery");
690         if (nilfs_test_opt(nilfs, DISCARD))
691                 seq_puts(seq, ",discard");
692
693         return 0;
694 }
695
696 static const struct super_operations nilfs_sops = {
697         .alloc_inode    = nilfs_alloc_inode,
698         .destroy_inode  = nilfs_destroy_inode,
699         .dirty_inode    = nilfs_dirty_inode,
700         .evict_inode    = nilfs_evict_inode,
701         .put_super      = nilfs_put_super,
702         .sync_fs        = nilfs_sync_fs,
703         .freeze_fs      = nilfs_freeze,
704         .unfreeze_fs    = nilfs_unfreeze,
705         .statfs         = nilfs_statfs,
706         .remount_fs     = nilfs_remount,
707         .show_options = nilfs_show_options
708 };
709
710 enum {
711         Opt_err_cont, Opt_err_panic, Opt_err_ro,
712         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
713         Opt_discard, Opt_nodiscard, Opt_err,
714 };
715
716 static match_table_t tokens = {
717         {Opt_err_cont, "errors=continue"},
718         {Opt_err_panic, "errors=panic"},
719         {Opt_err_ro, "errors=remount-ro"},
720         {Opt_barrier, "barrier"},
721         {Opt_nobarrier, "nobarrier"},
722         {Opt_snapshot, "cp=%u"},
723         {Opt_order, "order=%s"},
724         {Opt_norecovery, "norecovery"},
725         {Opt_discard, "discard"},
726         {Opt_nodiscard, "nodiscard"},
727         {Opt_err, NULL}
728 };
729
730 static int parse_options(char *options, struct super_block *sb, int is_remount)
731 {
732         struct the_nilfs *nilfs = sb->s_fs_info;
733         char *p;
734         substring_t args[MAX_OPT_ARGS];
735
736         if (!options)
737                 return 1;
738
739         while ((p = strsep(&options, ",")) != NULL) {
740                 int token;
741                 if (!*p)
742                         continue;
743
744                 token = match_token(p, tokens, args);
745                 switch (token) {
746                 case Opt_barrier:
747                         nilfs_set_opt(nilfs, BARRIER);
748                         break;
749                 case Opt_nobarrier:
750                         nilfs_clear_opt(nilfs, BARRIER);
751                         break;
752                 case Opt_order:
753                         if (strcmp(args[0].from, "relaxed") == 0)
754                                 /* Ordered data semantics */
755                                 nilfs_clear_opt(nilfs, STRICT_ORDER);
756                         else if (strcmp(args[0].from, "strict") == 0)
757                                 /* Strict in-order semantics */
758                                 nilfs_set_opt(nilfs, STRICT_ORDER);
759                         else
760                                 return 0;
761                         break;
762                 case Opt_err_panic:
763                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
764                         break;
765                 case Opt_err_ro:
766                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
767                         break;
768                 case Opt_err_cont:
769                         nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
770                         break;
771                 case Opt_snapshot:
772                         if (is_remount) {
773                                 printk(KERN_ERR
774                                        "NILFS: \"%s\" option is invalid "
775                                        "for remount.\n", p);
776                                 return 0;
777                         }
778                         break;
779                 case Opt_norecovery:
780                         nilfs_set_opt(nilfs, NORECOVERY);
781                         break;
782                 case Opt_discard:
783                         nilfs_set_opt(nilfs, DISCARD);
784                         break;
785                 case Opt_nodiscard:
786                         nilfs_clear_opt(nilfs, DISCARD);
787                         break;
788                 default:
789                         printk(KERN_ERR
790                                "NILFS: Unrecognized mount option \"%s\"\n", p);
791                         return 0;
792                 }
793         }
794         return 1;
795 }
796
797 static inline void
798 nilfs_set_default_options(struct super_block *sb,
799                           struct nilfs_super_block *sbp)
800 {
801         struct the_nilfs *nilfs = sb->s_fs_info;
802
803         nilfs->ns_mount_opt =
804                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
805 }
806
807 static int nilfs_setup_super(struct super_block *sb, int is_mount)
808 {
809         struct the_nilfs *nilfs = sb->s_fs_info;
810         struct nilfs_super_block **sbp;
811         int max_mnt_count;
812         int mnt_count;
813
814         /* nilfs->ns_sem must be locked by the caller. */
815         sbp = nilfs_prepare_super(sb, 0);
816         if (!sbp)
817                 return -EIO;
818
819         if (!is_mount)
820                 goto skip_mount_setup;
821
822         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
823         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
824
825         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
826                 printk(KERN_WARNING
827                        "NILFS warning: mounting fs with errors\n");
828 #if 0
829         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
830                 printk(KERN_WARNING
831                        "NILFS warning: maximal mount count reached\n");
832 #endif
833         }
834         if (!max_mnt_count)
835                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
836
837         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
838         sbp[0]->s_mtime = cpu_to_le64(get_seconds());
839
840 skip_mount_setup:
841         sbp[0]->s_state =
842                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
843         /* synchronize sbp[1] with sbp[0] */
844         if (sbp[1])
845                 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
846         return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
847 }
848
849 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
850                                                  u64 pos, int blocksize,
851                                                  struct buffer_head **pbh)
852 {
853         unsigned long long sb_index = pos;
854         unsigned long offset;
855
856         offset = do_div(sb_index, blocksize);
857         *pbh = sb_bread(sb, sb_index);
858         if (!*pbh)
859                 return NULL;
860         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
861 }
862
863 int nilfs_store_magic_and_option(struct super_block *sb,
864                                  struct nilfs_super_block *sbp,
865                                  char *data)
866 {
867         struct the_nilfs *nilfs = sb->s_fs_info;
868
869         sb->s_magic = le16_to_cpu(sbp->s_magic);
870
871         /* FS independent flags */
872 #ifdef NILFS_ATIME_DISABLE
873         sb->s_flags |= MS_NOATIME;
874 #endif
875
876         nilfs_set_default_options(sb, sbp);
877
878         nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
879         nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
880         nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
881         nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
882
883         return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
884 }
885
886 int nilfs_check_feature_compatibility(struct super_block *sb,
887                                       struct nilfs_super_block *sbp)
888 {
889         __u64 features;
890
891         features = le64_to_cpu(sbp->s_feature_incompat) &
892                 ~NILFS_FEATURE_INCOMPAT_SUPP;
893         if (features) {
894                 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
895                        "optional features (%llx)\n",
896                        (unsigned long long)features);
897                 return -EINVAL;
898         }
899         features = le64_to_cpu(sbp->s_feature_compat_ro) &
900                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
901         if (!(sb->s_flags & MS_RDONLY) && features) {
902                 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
903                        "unsupported optional features (%llx)\n",
904                        (unsigned long long)features);
905                 return -EINVAL;
906         }
907         return 0;
908 }
909
910 static int nilfs_get_root_dentry(struct super_block *sb,
911                                  struct nilfs_root *root,
912                                  struct dentry **root_dentry)
913 {
914         struct inode *inode;
915         struct dentry *dentry;
916         int ret = 0;
917
918         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
919         if (IS_ERR(inode)) {
920                 printk(KERN_ERR "NILFS: get root inode failed\n");
921                 ret = PTR_ERR(inode);
922                 goto out;
923         }
924         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
925                 iput(inode);
926                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
927                 ret = -EINVAL;
928                 goto out;
929         }
930
931         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
932                 dentry = d_find_alias(inode);
933                 if (!dentry) {
934                         dentry = d_make_root(inode);
935                         if (!dentry) {
936                                 ret = -ENOMEM;
937                                 goto failed_dentry;
938                         }
939                 } else {
940                         iput(inode);
941                 }
942         } else {
943                 dentry = d_obtain_alias(inode);
944                 if (IS_ERR(dentry)) {
945                         ret = PTR_ERR(dentry);
946                         goto failed_dentry;
947                 }
948         }
949         *root_dentry = dentry;
950  out:
951         return ret;
952
953  failed_dentry:
954         printk(KERN_ERR "NILFS: get root dentry failed\n");
955         goto out;
956 }
957
958 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
959                                  struct dentry **root_dentry)
960 {
961         struct the_nilfs *nilfs = s->s_fs_info;
962         struct nilfs_root *root;
963         int ret;
964
965         mutex_lock(&nilfs->ns_snapshot_mount_mutex);
966
967         down_read(&nilfs->ns_segctor_sem);
968         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
969         up_read(&nilfs->ns_segctor_sem);
970         if (ret < 0) {
971                 ret = (ret == -ENOENT) ? -EINVAL : ret;
972                 goto out;
973         } else if (!ret) {
974                 printk(KERN_ERR "NILFS: The specified checkpoint is "
975                        "not a snapshot (checkpoint number=%llu).\n",
976                        (unsigned long long)cno);
977                 ret = -EINVAL;
978                 goto out;
979         }
980
981         ret = nilfs_attach_checkpoint(s, cno, false, &root);
982         if (ret) {
983                 printk(KERN_ERR "NILFS: error loading snapshot "
984                        "(checkpoint number=%llu).\n",
985                (unsigned long long)cno);
986                 goto out;
987         }
988         ret = nilfs_get_root_dentry(s, root, root_dentry);
989         nilfs_put_root(root);
990  out:
991         mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
992         return ret;
993 }
994
995 static int nilfs_tree_was_touched(struct dentry *root_dentry)
996 {
997         return root_dentry->d_count > 1;
998 }
999
1000 /**
1001  * nilfs_try_to_shrink_tree() - try to shrink dentries of a checkpoint
1002  * @root_dentry: root dentry of the tree to be shrunk
1003  *
1004  * This function returns true if the tree was in-use.
1005  */
1006 static int nilfs_try_to_shrink_tree(struct dentry *root_dentry)
1007 {
1008         if (have_submounts(root_dentry))
1009                 return true;
1010         shrink_dcache_parent(root_dentry);
1011         return nilfs_tree_was_touched(root_dentry);
1012 }
1013
1014 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1015 {
1016         struct the_nilfs *nilfs = sb->s_fs_info;
1017         struct nilfs_root *root;
1018         struct inode *inode;
1019         struct dentry *dentry;
1020         int ret;
1021
1022         if (cno < 0 || cno > nilfs->ns_cno)
1023                 return false;
1024
1025         if (cno >= nilfs_last_cno(nilfs))
1026                 return true;    /* protect recent checkpoints */
1027
1028         ret = false;
1029         root = nilfs_lookup_root(nilfs, cno);
1030         if (root) {
1031                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1032                 if (inode) {
1033                         dentry = d_find_alias(inode);
1034                         if (dentry) {
1035                                 if (nilfs_tree_was_touched(dentry))
1036                                         ret = nilfs_try_to_shrink_tree(dentry);
1037                                 dput(dentry);
1038                         }
1039                         iput(inode);
1040                 }
1041                 nilfs_put_root(root);
1042         }
1043         return ret;
1044 }
1045
1046 /**
1047  * nilfs_fill_super() - initialize a super block instance
1048  * @sb: super_block
1049  * @data: mount options
1050  * @silent: silent mode flag
1051  *
1052  * This function is called exclusively by nilfs->ns_mount_mutex.
1053  * So, the recovery process is protected from other simultaneous mounts.
1054  */
1055 static int
1056 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1057 {
1058         struct the_nilfs *nilfs;
1059         struct nilfs_root *fsroot;
1060         struct backing_dev_info *bdi;
1061         __u64 cno;
1062         int err;
1063
1064         nilfs = alloc_nilfs(sb->s_bdev);
1065         if (!nilfs)
1066                 return -ENOMEM;
1067
1068         sb->s_fs_info = nilfs;
1069
1070         err = init_nilfs(nilfs, sb, (char *)data);
1071         if (err)
1072                 goto failed_nilfs;
1073
1074         sb->s_op = &nilfs_sops;
1075         sb->s_export_op = &nilfs_export_ops;
1076         sb->s_root = NULL;
1077         sb->s_time_gran = 1;
1078         sb->s_max_links = NILFS_LINK_MAX;
1079
1080         bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
1081         sb->s_bdi = bdi ? : &default_backing_dev_info;
1082
1083         err = load_nilfs(nilfs, sb);
1084         if (err)
1085                 goto failed_nilfs;
1086
1087         cno = nilfs_last_cno(nilfs);
1088         err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1089         if (err) {
1090                 printk(KERN_ERR "NILFS: error loading last checkpoint "
1091                        "(checkpoint number=%llu).\n", (unsigned long long)cno);
1092                 goto failed_unload;
1093         }
1094
1095         if (!(sb->s_flags & MS_RDONLY)) {
1096                 err = nilfs_attach_log_writer(sb, fsroot);
1097                 if (err)
1098                         goto failed_checkpoint;
1099         }
1100
1101         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1102         if (err)
1103                 goto failed_segctor;
1104
1105         nilfs_put_root(fsroot);
1106
1107         if (!(sb->s_flags & MS_RDONLY)) {
1108                 down_write(&nilfs->ns_sem);
1109                 nilfs_setup_super(sb, true);
1110                 up_write(&nilfs->ns_sem);
1111         }
1112
1113         return 0;
1114
1115  failed_segctor:
1116         nilfs_detach_log_writer(sb);
1117
1118  failed_checkpoint:
1119         nilfs_put_root(fsroot);
1120
1121  failed_unload:
1122         iput(nilfs->ns_sufile);
1123         iput(nilfs->ns_cpfile);
1124         iput(nilfs->ns_dat);
1125
1126  failed_nilfs:
1127         destroy_nilfs(nilfs);
1128         return err;
1129 }
1130
1131 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1132 {
1133         struct the_nilfs *nilfs = sb->s_fs_info;
1134         unsigned long old_sb_flags;
1135         unsigned long old_mount_opt;
1136         int err;
1137
1138         old_sb_flags = sb->s_flags;
1139         old_mount_opt = nilfs->ns_mount_opt;
1140
1141         if (!parse_options(data, sb, 1)) {
1142                 err = -EINVAL;
1143                 goto restore_opts;
1144         }
1145         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
1146
1147         err = -EINVAL;
1148
1149         if (!nilfs_valid_fs(nilfs)) {
1150                 printk(KERN_WARNING "NILFS (device %s): couldn't "
1151                        "remount because the filesystem is in an "
1152                        "incomplete recovery state.\n", sb->s_id);
1153                 goto restore_opts;
1154         }
1155
1156         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1157                 goto out;
1158         if (*flags & MS_RDONLY) {
1159                 /* Shutting down log writer */
1160                 nilfs_detach_log_writer(sb);
1161                 sb->s_flags |= MS_RDONLY;
1162
1163                 /*
1164                  * Remounting a valid RW partition RDONLY, so set
1165                  * the RDONLY flag and then mark the partition as valid again.
1166                  */
1167                 down_write(&nilfs->ns_sem);
1168                 nilfs_cleanup_super(sb);
1169                 up_write(&nilfs->ns_sem);
1170         } else {
1171                 __u64 features;
1172                 struct nilfs_root *root;
1173
1174                 /*
1175                  * Mounting a RDONLY partition read-write, so reread and
1176                  * store the current valid flag.  (It may have been changed
1177                  * by fsck since we originally mounted the partition.)
1178                  */
1179                 down_read(&nilfs->ns_sem);
1180                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1181                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1182                 up_read(&nilfs->ns_sem);
1183                 if (features) {
1184                         printk(KERN_WARNING "NILFS (device %s): couldn't "
1185                                "remount RDWR because of unsupported optional "
1186                                "features (%llx)\n",
1187                                sb->s_id, (unsigned long long)features);
1188                         err = -EROFS;
1189                         goto restore_opts;
1190                 }
1191
1192                 sb->s_flags &= ~MS_RDONLY;
1193
1194                 root = NILFS_I(sb->s_root->d_inode)->i_root;
1195                 err = nilfs_attach_log_writer(sb, root);
1196                 if (err)
1197                         goto restore_opts;
1198
1199                 down_write(&nilfs->ns_sem);
1200                 nilfs_setup_super(sb, true);
1201                 up_write(&nilfs->ns_sem);
1202         }
1203  out:
1204         return 0;
1205
1206  restore_opts:
1207         sb->s_flags = old_sb_flags;
1208         nilfs->ns_mount_opt = old_mount_opt;
1209         return err;
1210 }
1211
1212 struct nilfs_super_data {
1213         struct block_device *bdev;
1214         __u64 cno;
1215         int flags;
1216 };
1217
1218 /**
1219  * nilfs_identify - pre-read mount options needed to identify mount instance
1220  * @data: mount options
1221  * @sd: nilfs_super_data
1222  */
1223 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1224 {
1225         char *p, *options = data;
1226         substring_t args[MAX_OPT_ARGS];
1227         int token;
1228         int ret = 0;
1229
1230         do {
1231                 p = strsep(&options, ",");
1232                 if (p != NULL && *p) {
1233                         token = match_token(p, tokens, args);
1234                         if (token == Opt_snapshot) {
1235                                 if (!(sd->flags & MS_RDONLY)) {
1236                                         ret++;
1237                                 } else {
1238                                         sd->cno = simple_strtoull(args[0].from,
1239                                                                   NULL, 0);
1240                                         /*
1241                                          * No need to see the end pointer;
1242                                          * match_token() has done syntax
1243                                          * checking.
1244                                          */
1245                                         if (sd->cno == 0)
1246                                                 ret++;
1247                                 }
1248                         }
1249                         if (ret)
1250                                 printk(KERN_ERR
1251                                        "NILFS: invalid mount option: %s\n", p);
1252                 }
1253                 if (!options)
1254                         break;
1255                 BUG_ON(options == data);
1256                 *(options - 1) = ',';
1257         } while (!ret);
1258         return ret;
1259 }
1260
1261 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1262 {
1263         s->s_bdev = data;
1264         s->s_dev = s->s_bdev->bd_dev;
1265         return 0;
1266 }
1267
1268 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1269 {
1270         return (void *)s->s_bdev == data;
1271 }
1272
1273 static struct dentry *
1274 nilfs_mount(struct file_system_type *fs_type, int flags,
1275              const char *dev_name, void *data)
1276 {
1277         struct nilfs_super_data sd;
1278         struct super_block *s;
1279         fmode_t mode = FMODE_READ | FMODE_EXCL;
1280         struct dentry *root_dentry;
1281         int err, s_new = false;
1282
1283         if (!(flags & MS_RDONLY))
1284                 mode |= FMODE_WRITE;
1285
1286         sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1287         if (IS_ERR(sd.bdev))
1288                 return ERR_CAST(sd.bdev);
1289
1290         sd.cno = 0;
1291         sd.flags = flags;
1292         if (nilfs_identify((char *)data, &sd)) {
1293                 err = -EINVAL;
1294                 goto failed;
1295         }
1296
1297         /*
1298          * once the super is inserted into the list by sget, s_umount
1299          * will protect the lockfs code from trying to start a snapshot
1300          * while we are mounting
1301          */
1302         mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1303         if (sd.bdev->bd_fsfreeze_count > 0) {
1304                 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1305                 err = -EBUSY;
1306                 goto failed;
1307         }
1308         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1309                  sd.bdev);
1310         mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1311         if (IS_ERR(s)) {
1312                 err = PTR_ERR(s);
1313                 goto failed;
1314         }
1315
1316         if (!s->s_root) {
1317                 char b[BDEVNAME_SIZE];
1318
1319                 s_new = true;
1320
1321                 /* New superblock instance created */
1322                 s->s_mode = mode;
1323                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1324                 sb_set_blocksize(s, block_size(sd.bdev));
1325
1326                 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1327                 if (err)
1328                         goto failed_super;
1329
1330                 s->s_flags |= MS_ACTIVE;
1331         } else if (!sd.cno) {
1332                 int busy = false;
1333
1334                 if (nilfs_tree_was_touched(s->s_root)) {
1335                         busy = nilfs_try_to_shrink_tree(s->s_root);
1336                         if (busy && (flags ^ s->s_flags) & MS_RDONLY) {
1337                                 printk(KERN_ERR "NILFS: the device already "
1338                                        "has a %s mount.\n",
1339                                        (s->s_flags & MS_RDONLY) ?
1340                                        "read-only" : "read/write");
1341                                 err = -EBUSY;
1342                                 goto failed_super;
1343                         }
1344                 }
1345                 if (!busy) {
1346                         /*
1347                          * Try remount to setup mount states if the current
1348                          * tree is not mounted and only snapshots use this sb.
1349                          */
1350                         err = nilfs_remount(s, &flags, data);
1351                         if (err)
1352                                 goto failed_super;
1353                 }
1354         }
1355
1356         if (sd.cno) {
1357                 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1358                 if (err)
1359                         goto failed_super;
1360         } else {
1361                 root_dentry = dget(s->s_root);
1362         }
1363
1364         if (!s_new)
1365                 blkdev_put(sd.bdev, mode);
1366
1367         return root_dentry;
1368
1369  failed_super:
1370         deactivate_locked_super(s);
1371
1372  failed:
1373         if (!s_new)
1374                 blkdev_put(sd.bdev, mode);
1375         return ERR_PTR(err);
1376 }
1377
1378 struct file_system_type nilfs_fs_type = {
1379         .owner    = THIS_MODULE,
1380         .name     = "nilfs2",
1381         .mount    = nilfs_mount,
1382         .kill_sb  = kill_block_super,
1383         .fs_flags = FS_REQUIRES_DEV,
1384 };
1385 MODULE_ALIAS_FS("nilfs2");
1386
1387 static void nilfs_inode_init_once(void *obj)
1388 {
1389         struct nilfs_inode_info *ii = obj;
1390
1391         INIT_LIST_HEAD(&ii->i_dirty);
1392 #ifdef CONFIG_NILFS_XATTR
1393         init_rwsem(&ii->xattr_sem);
1394 #endif
1395         address_space_init_once(&ii->i_btnode_cache);
1396         ii->i_bmap = &ii->i_bmap_data;
1397         inode_init_once(&ii->vfs_inode);
1398 }
1399
1400 static void nilfs_segbuf_init_once(void *obj)
1401 {
1402         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1403 }
1404
1405 static void nilfs_destroy_cachep(void)
1406 {
1407         /*
1408          * Make sure all delayed rcu free inodes are flushed before we
1409          * destroy cache.
1410          */
1411         rcu_barrier();
1412
1413         if (nilfs_inode_cachep)
1414                 kmem_cache_destroy(nilfs_inode_cachep);
1415         if (nilfs_transaction_cachep)
1416                 kmem_cache_destroy(nilfs_transaction_cachep);
1417         if (nilfs_segbuf_cachep)
1418                 kmem_cache_destroy(nilfs_segbuf_cachep);
1419         if (nilfs_btree_path_cache)
1420                 kmem_cache_destroy(nilfs_btree_path_cache);
1421 }
1422
1423 static int __init nilfs_init_cachep(void)
1424 {
1425         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1426                         sizeof(struct nilfs_inode_info), 0,
1427                         SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1428         if (!nilfs_inode_cachep)
1429                 goto fail;
1430
1431         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1432                         sizeof(struct nilfs_transaction_info), 0,
1433                         SLAB_RECLAIM_ACCOUNT, NULL);
1434         if (!nilfs_transaction_cachep)
1435                 goto fail;
1436
1437         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1438                         sizeof(struct nilfs_segment_buffer), 0,
1439                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1440         if (!nilfs_segbuf_cachep)
1441                 goto fail;
1442
1443         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1444                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1445                         0, 0, NULL);
1446         if (!nilfs_btree_path_cache)
1447                 goto fail;
1448
1449         return 0;
1450
1451 fail:
1452         nilfs_destroy_cachep();
1453         return -ENOMEM;
1454 }
1455
1456 static int __init init_nilfs_fs(void)
1457 {
1458         int err;
1459
1460         err = nilfs_init_cachep();
1461         if (err)
1462                 goto fail;
1463
1464         err = register_filesystem(&nilfs_fs_type);
1465         if (err)
1466                 goto free_cachep;
1467
1468         printk(KERN_INFO "NILFS version 2 loaded\n");
1469         return 0;
1470
1471 free_cachep:
1472         nilfs_destroy_cachep();
1473 fail:
1474         return err;
1475 }
1476
1477 static void __exit exit_nilfs_fs(void)
1478 {
1479         nilfs_destroy_cachep();
1480         unregister_filesystem(&nilfs_fs_type);
1481 }
1482
1483 module_init(init_nilfs_fs)
1484 module_exit(exit_nilfs_fs)