nilfs2: remove own inode allocator and destructor for metadata files
[pandora-kernel.git] / fs / nilfs2 / super.c
1 /*
2  * super.c - NILFS module and super block management.
3  *
4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
19  *
20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
21  */
22 /*
23  *  linux/fs/ext2/super.c
24  *
25  * Copyright (C) 1992, 1993, 1994, 1995
26  * Remy Card (card@masi.ibp.fr)
27  * Laboratoire MASI - Institut Blaise Pascal
28  * Universite Pierre et Marie Curie (Paris VI)
29  *
30  *  from
31  *
32  *  linux/fs/minix/inode.c
33  *
34  *  Copyright (C) 1991, 1992  Linus Torvalds
35  *
36  *  Big-endian to little-endian byte-swapping/bitmaps by
37  *        David S. Miller (davem@caip.rutgers.edu), 1995
38  */
39
40 #include <linux/module.h>
41 #include <linux/string.h>
42 #include <linux/slab.h>
43 #include <linux/init.h>
44 #include <linux/blkdev.h>
45 #include <linux/parser.h>
46 #include <linux/random.h>
47 #include <linux/crc32.h>
48 #include <linux/vfs.h>
49 #include <linux/writeback.h>
50 #include <linux/kobject.h>
51 #include <linux/seq_file.h>
52 #include <linux/mount.h>
53 #include "nilfs.h"
54 #include "export.h"
55 #include "mdt.h"
56 #include "alloc.h"
57 #include "btree.h"
58 #include "btnode.h"
59 #include "page.h"
60 #include "cpfile.h"
61 #include "ifile.h"
62 #include "dat.h"
63 #include "segment.h"
64 #include "segbuf.h"
65
66 MODULE_AUTHOR("NTT Corp.");
67 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
68                    "(NILFS)");
69 MODULE_LICENSE("GPL");
70
71 struct kmem_cache *nilfs_inode_cachep;
72 struct kmem_cache *nilfs_transaction_cachep;
73 struct kmem_cache *nilfs_segbuf_cachep;
74 struct kmem_cache *nilfs_btree_path_cache;
75
76 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
77
78 static void nilfs_set_error(struct nilfs_sb_info *sbi)
79 {
80         struct the_nilfs *nilfs = sbi->s_nilfs;
81         struct nilfs_super_block **sbp;
82
83         down_write(&nilfs->ns_sem);
84         if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
85                 nilfs->ns_mount_state |= NILFS_ERROR_FS;
86                 sbp = nilfs_prepare_super(sbi, 0);
87                 if (likely(sbp)) {
88                         sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
89                         if (sbp[1])
90                                 sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
91                         nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
92                 }
93         }
94         up_write(&nilfs->ns_sem);
95 }
96
97 /**
98  * nilfs_error() - report failure condition on a filesystem
99  *
100  * nilfs_error() sets an ERROR_FS flag on the superblock as well as
101  * reporting an error message.  It should be called when NILFS detects
102  * incoherences or defects of meta data on disk.  As for sustainable
103  * errors such as a single-shot I/O error, nilfs_warning() or the printk()
104  * function should be used instead.
105  *
106  * The segment constructor must not call this function because it can
107  * kill itself.
108  */
109 void nilfs_error(struct super_block *sb, const char *function,
110                  const char *fmt, ...)
111 {
112         struct nilfs_sb_info *sbi = NILFS_SB(sb);
113         va_list args;
114
115         va_start(args, fmt);
116         printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
117         vprintk(fmt, args);
118         printk("\n");
119         va_end(args);
120
121         if (!(sb->s_flags & MS_RDONLY)) {
122                 nilfs_set_error(sbi);
123
124                 if (nilfs_test_opt(sbi, ERRORS_RO)) {
125                         printk(KERN_CRIT "Remounting filesystem read-only\n");
126                         sb->s_flags |= MS_RDONLY;
127                 }
128         }
129
130         if (nilfs_test_opt(sbi, ERRORS_PANIC))
131                 panic("NILFS (device %s): panic forced after error\n",
132                       sb->s_id);
133 }
134
135 void nilfs_warning(struct super_block *sb, const char *function,
136                    const char *fmt, ...)
137 {
138         va_list args;
139
140         va_start(args, fmt);
141         printk(KERN_WARNING "NILFS warning (device %s): %s: ",
142                sb->s_id, function);
143         vprintk(fmt, args);
144         printk("\n");
145         va_end(args);
146 }
147
148
149 struct inode *nilfs_alloc_inode(struct super_block *sb)
150 {
151         struct nilfs_inode_info *ii;
152
153         ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
154         if (!ii)
155                 return NULL;
156         ii->i_bh = NULL;
157         ii->i_state = 0;
158         ii->i_cno = 0;
159         ii->vfs_inode.i_version = 1;
160         nilfs_btnode_cache_init(&ii->i_btnode_cache, sb->s_bdi);
161         return &ii->vfs_inode;
162 }
163
164 void nilfs_destroy_inode(struct inode *inode)
165 {
166         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
167
168         if (mdi) {
169                 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
170                 kfree(mdi);
171         }
172         kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
173 }
174
175 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int flag)
176 {
177         struct the_nilfs *nilfs = sbi->s_nilfs;
178         int err;
179
180  retry:
181         set_buffer_dirty(nilfs->ns_sbh[0]);
182
183         if (nilfs_test_opt(sbi, BARRIER)) {
184                 err = __sync_dirty_buffer(nilfs->ns_sbh[0],
185                                           WRITE_SYNC | WRITE_BARRIER);
186                 if (err == -EOPNOTSUPP) {
187                         nilfs_warning(sbi->s_super, __func__,
188                                       "barrier-based sync failed. "
189                                       "disabling barriers\n");
190                         nilfs_clear_opt(sbi, BARRIER);
191                         goto retry;
192                 }
193         } else {
194                 err = sync_dirty_buffer(nilfs->ns_sbh[0]);
195         }
196
197         if (unlikely(err)) {
198                 printk(KERN_ERR
199                        "NILFS: unable to write superblock (err=%d)\n", err);
200                 if (err == -EIO && nilfs->ns_sbh[1]) {
201                         /*
202                          * sbp[0] points to newer log than sbp[1],
203                          * so copy sbp[0] to sbp[1] to take over sbp[0].
204                          */
205                         memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
206                                nilfs->ns_sbsize);
207                         nilfs_fall_back_super_block(nilfs);
208                         goto retry;
209                 }
210         } else {
211                 struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
212
213                 nilfs->ns_sbwcount++;
214
215                 /*
216                  * The latest segment becomes trailable from the position
217                  * written in superblock.
218                  */
219                 clear_nilfs_discontinued(nilfs);
220
221                 /* update GC protection for recent segments */
222                 if (nilfs->ns_sbh[1]) {
223                         if (flag == NILFS_SB_COMMIT_ALL) {
224                                 set_buffer_dirty(nilfs->ns_sbh[1]);
225                                 if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
226                                         goto out;
227                         }
228                         if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
229                             le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
230                                 sbp = nilfs->ns_sbp[1];
231                 }
232
233                 spin_lock(&nilfs->ns_last_segment_lock);
234                 nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
235                 spin_unlock(&nilfs->ns_last_segment_lock);
236         }
237  out:
238         return err;
239 }
240
241 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
242                           struct the_nilfs *nilfs)
243 {
244         sector_t nfreeblocks;
245
246         /* nilfs->ns_sem must be locked by the caller. */
247         nilfs_count_free_blocks(nilfs, &nfreeblocks);
248         sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
249
250         spin_lock(&nilfs->ns_last_segment_lock);
251         sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
252         sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
253         sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
254         spin_unlock(&nilfs->ns_last_segment_lock);
255 }
256
257 struct nilfs_super_block **nilfs_prepare_super(struct nilfs_sb_info *sbi,
258                                                int flip)
259 {
260         struct the_nilfs *nilfs = sbi->s_nilfs;
261         struct nilfs_super_block **sbp = nilfs->ns_sbp;
262
263         /* nilfs->ns_sem must be locked by the caller. */
264         if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
265                 if (sbp[1] &&
266                     sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
267                         memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
268                 } else {
269                         printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
270                                sbi->s_super->s_id);
271                         return NULL;
272                 }
273         } else if (sbp[1] &&
274                    sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
275                         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
276         }
277
278         if (flip && sbp[1])
279                 nilfs_swap_super_block(nilfs);
280
281         return sbp;
282 }
283
284 int nilfs_commit_super(struct nilfs_sb_info *sbi, int flag)
285 {
286         struct the_nilfs *nilfs = sbi->s_nilfs;
287         struct nilfs_super_block **sbp = nilfs->ns_sbp;
288         time_t t;
289
290         /* nilfs->ns_sem must be locked by the caller. */
291         t = get_seconds();
292         nilfs->ns_sbwtime = t;
293         sbp[0]->s_wtime = cpu_to_le64(t);
294         sbp[0]->s_sum = 0;
295         sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
296                                              (unsigned char *)sbp[0],
297                                              nilfs->ns_sbsize));
298         if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
299                 sbp[1]->s_wtime = sbp[0]->s_wtime;
300                 sbp[1]->s_sum = 0;
301                 sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
302                                             (unsigned char *)sbp[1],
303                                             nilfs->ns_sbsize));
304         }
305         clear_nilfs_sb_dirty(nilfs);
306         return nilfs_sync_super(sbi, flag);
307 }
308
309 /**
310  * nilfs_cleanup_super() - write filesystem state for cleanup
311  * @sbi: nilfs_sb_info to be unmounted or degraded to read-only
312  *
313  * This function restores state flags in the on-disk super block.
314  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
315  * filesystem was not clean previously.
316  */
317 int nilfs_cleanup_super(struct nilfs_sb_info *sbi)
318 {
319         struct nilfs_super_block **sbp;
320         int flag = NILFS_SB_COMMIT;
321         int ret = -EIO;
322
323         sbp = nilfs_prepare_super(sbi, 0);
324         if (sbp) {
325                 sbp[0]->s_state = cpu_to_le16(sbi->s_nilfs->ns_mount_state);
326                 nilfs_set_log_cursor(sbp[0], sbi->s_nilfs);
327                 if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
328                         /*
329                          * make the "clean" flag also to the opposite
330                          * super block if both super blocks point to
331                          * the same checkpoint.
332                          */
333                         sbp[1]->s_state = sbp[0]->s_state;
334                         flag = NILFS_SB_COMMIT_ALL;
335                 }
336                 ret = nilfs_commit_super(sbi, flag);
337         }
338         return ret;
339 }
340
341 static void nilfs_put_super(struct super_block *sb)
342 {
343         struct nilfs_sb_info *sbi = NILFS_SB(sb);
344         struct the_nilfs *nilfs = sbi->s_nilfs;
345
346         nilfs_detach_segment_constructor(sbi);
347
348         if (!(sb->s_flags & MS_RDONLY)) {
349                 down_write(&nilfs->ns_sem);
350                 nilfs_cleanup_super(sbi);
351                 up_write(&nilfs->ns_sem);
352         }
353
354         iput(nilfs->ns_sufile);
355         iput(nilfs->ns_cpfile);
356         iput(nilfs->ns_dat);
357
358         destroy_nilfs(nilfs);
359         sbi->s_super = NULL;
360         sb->s_fs_info = NULL;
361         kfree(sbi);
362 }
363
364 static int nilfs_sync_fs(struct super_block *sb, int wait)
365 {
366         struct nilfs_sb_info *sbi = NILFS_SB(sb);
367         struct the_nilfs *nilfs = sbi->s_nilfs;
368         struct nilfs_super_block **sbp;
369         int err = 0;
370
371         /* This function is called when super block should be written back */
372         if (wait)
373                 err = nilfs_construct_segment(sb);
374
375         down_write(&nilfs->ns_sem);
376         if (nilfs_sb_dirty(nilfs)) {
377                 sbp = nilfs_prepare_super(sbi, nilfs_sb_will_flip(nilfs));
378                 if (likely(sbp)) {
379                         nilfs_set_log_cursor(sbp[0], nilfs);
380                         nilfs_commit_super(sbi, NILFS_SB_COMMIT);
381                 }
382         }
383         up_write(&nilfs->ns_sem);
384
385         return err;
386 }
387
388 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno, int curr_mnt,
389                             struct nilfs_root **rootp)
390 {
391         struct the_nilfs *nilfs = sbi->s_nilfs;
392         struct nilfs_root *root;
393         struct nilfs_checkpoint *raw_cp;
394         struct buffer_head *bh_cp;
395         int err = -ENOMEM;
396
397         root = nilfs_find_or_create_root(
398                 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
399         if (!root)
400                 return err;
401
402         if (root->ifile)
403                 goto reuse; /* already attached checkpoint */
404
405         down_read(&nilfs->ns_segctor_sem);
406         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
407                                           &bh_cp);
408         up_read(&nilfs->ns_segctor_sem);
409         if (unlikely(err)) {
410                 if (err == -ENOENT || err == -EINVAL) {
411                         printk(KERN_ERR
412                                "NILFS: Invalid checkpoint "
413                                "(checkpoint number=%llu)\n",
414                                (unsigned long long)cno);
415                         err = -EINVAL;
416                 }
417                 goto failed;
418         }
419
420         err = nilfs_ifile_read(sbi->s_super, root, nilfs->ns_inode_size,
421                                &raw_cp->cp_ifile_inode, &root->ifile);
422         if (err)
423                 goto failed_bh;
424
425         atomic_set(&root->inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
426         atomic_set(&root->blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
427
428         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
429
430  reuse:
431         *rootp = root;
432         return 0;
433
434  failed_bh:
435         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
436  failed:
437         nilfs_put_root(root);
438
439         return err;
440 }
441
442 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
443 {
444         struct super_block *sb = dentry->d_sb;
445         struct nilfs_root *root = NILFS_I(dentry->d_inode)->i_root;
446         struct the_nilfs *nilfs = root->nilfs;
447         u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
448         unsigned long long blocks;
449         unsigned long overhead;
450         unsigned long nrsvblocks;
451         sector_t nfreeblocks;
452         int err;
453
454         /*
455          * Compute all of the segment blocks
456          *
457          * The blocks before first segment and after last segment
458          * are excluded.
459          */
460         blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
461                 - nilfs->ns_first_data_block;
462         nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
463
464         /*
465          * Compute the overhead
466          *
467          * When distributing meta data blocks outside segment structure,
468          * We must count them as the overhead.
469          */
470         overhead = 0;
471
472         err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
473         if (unlikely(err))
474                 return err;
475
476         buf->f_type = NILFS_SUPER_MAGIC;
477         buf->f_bsize = sb->s_blocksize;
478         buf->f_blocks = blocks - overhead;
479         buf->f_bfree = nfreeblocks;
480         buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
481                 (buf->f_bfree - nrsvblocks) : 0;
482         buf->f_files = atomic_read(&root->inodes_count);
483         buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
484         buf->f_namelen = NILFS_NAME_LEN;
485         buf->f_fsid.val[0] = (u32)id;
486         buf->f_fsid.val[1] = (u32)(id >> 32);
487
488         return 0;
489 }
490
491 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
492 {
493         struct super_block *sb = vfs->mnt_sb;
494         struct nilfs_sb_info *sbi = NILFS_SB(sb);
495         struct nilfs_root *root = NILFS_I(vfs->mnt_root->d_inode)->i_root;
496
497         if (!nilfs_test_opt(sbi, BARRIER))
498                 seq_puts(seq, ",nobarrier");
499         if (root->cno != NILFS_CPTREE_CURRENT_CNO)
500                 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
501         if (nilfs_test_opt(sbi, ERRORS_PANIC))
502                 seq_puts(seq, ",errors=panic");
503         if (nilfs_test_opt(sbi, ERRORS_CONT))
504                 seq_puts(seq, ",errors=continue");
505         if (nilfs_test_opt(sbi, STRICT_ORDER))
506                 seq_puts(seq, ",order=strict");
507         if (nilfs_test_opt(sbi, NORECOVERY))
508                 seq_puts(seq, ",norecovery");
509         if (nilfs_test_opt(sbi, DISCARD))
510                 seq_puts(seq, ",discard");
511
512         return 0;
513 }
514
515 static const struct super_operations nilfs_sops = {
516         .alloc_inode    = nilfs_alloc_inode,
517         .destroy_inode  = nilfs_destroy_inode,
518         .dirty_inode    = nilfs_dirty_inode,
519         /* .write_inode    = nilfs_write_inode, */
520         /* .put_inode      = nilfs_put_inode, */
521         /* .drop_inode    = nilfs_drop_inode, */
522         .evict_inode    = nilfs_evict_inode,
523         .put_super      = nilfs_put_super,
524         /* .write_super    = nilfs_write_super, */
525         .sync_fs        = nilfs_sync_fs,
526         /* .write_super_lockfs */
527         /* .unlockfs */
528         .statfs         = nilfs_statfs,
529         .remount_fs     = nilfs_remount,
530         /* .umount_begin */
531         .show_options = nilfs_show_options
532 };
533
534 enum {
535         Opt_err_cont, Opt_err_panic, Opt_err_ro,
536         Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
537         Opt_discard, Opt_nodiscard, Opt_err,
538 };
539
540 static match_table_t tokens = {
541         {Opt_err_cont, "errors=continue"},
542         {Opt_err_panic, "errors=panic"},
543         {Opt_err_ro, "errors=remount-ro"},
544         {Opt_barrier, "barrier"},
545         {Opt_nobarrier, "nobarrier"},
546         {Opt_snapshot, "cp=%u"},
547         {Opt_order, "order=%s"},
548         {Opt_norecovery, "norecovery"},
549         {Opt_discard, "discard"},
550         {Opt_nodiscard, "nodiscard"},
551         {Opt_err, NULL}
552 };
553
554 static int parse_options(char *options, struct super_block *sb, int is_remount)
555 {
556         struct nilfs_sb_info *sbi = NILFS_SB(sb);
557         char *p;
558         substring_t args[MAX_OPT_ARGS];
559         int option;
560
561         if (!options)
562                 return 1;
563
564         while ((p = strsep(&options, ",")) != NULL) {
565                 int token;
566                 if (!*p)
567                         continue;
568
569                 token = match_token(p, tokens, args);
570                 switch (token) {
571                 case Opt_barrier:
572                         nilfs_set_opt(sbi, BARRIER);
573                         break;
574                 case Opt_nobarrier:
575                         nilfs_clear_opt(sbi, BARRIER);
576                         break;
577                 case Opt_order:
578                         if (strcmp(args[0].from, "relaxed") == 0)
579                                 /* Ordered data semantics */
580                                 nilfs_clear_opt(sbi, STRICT_ORDER);
581                         else if (strcmp(args[0].from, "strict") == 0)
582                                 /* Strict in-order semantics */
583                                 nilfs_set_opt(sbi, STRICT_ORDER);
584                         else
585                                 return 0;
586                         break;
587                 case Opt_err_panic:
588                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
589                         break;
590                 case Opt_err_ro:
591                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
592                         break;
593                 case Opt_err_cont:
594                         nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
595                         break;
596                 case Opt_snapshot:
597                         if (match_int(&args[0], &option) || option <= 0)
598                                 return 0;
599                         if (is_remount) {
600                                 printk(KERN_ERR
601                                        "NILFS: \"%s\" option is invalid "
602                                        "for remount.\n", p);
603                                 return 0;
604                         }
605                         break;
606                 case Opt_norecovery:
607                         nilfs_set_opt(sbi, NORECOVERY);
608                         break;
609                 case Opt_discard:
610                         nilfs_set_opt(sbi, DISCARD);
611                         break;
612                 case Opt_nodiscard:
613                         nilfs_clear_opt(sbi, DISCARD);
614                         break;
615                 default:
616                         printk(KERN_ERR
617                                "NILFS: Unrecognized mount option \"%s\"\n", p);
618                         return 0;
619                 }
620         }
621         return 1;
622 }
623
624 static inline void
625 nilfs_set_default_options(struct nilfs_sb_info *sbi,
626                           struct nilfs_super_block *sbp)
627 {
628         sbi->s_mount_opt =
629                 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
630 }
631
632 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
633 {
634         struct the_nilfs *nilfs = sbi->s_nilfs;
635         struct nilfs_super_block **sbp;
636         int max_mnt_count;
637         int mnt_count;
638
639         /* nilfs->ns_sem must be locked by the caller. */
640         sbp = nilfs_prepare_super(sbi, 0);
641         if (!sbp)
642                 return -EIO;
643
644         max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
645         mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
646
647         if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
648                 printk(KERN_WARNING
649                        "NILFS warning: mounting fs with errors\n");
650 #if 0
651         } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
652                 printk(KERN_WARNING
653                        "NILFS warning: maximal mount count reached\n");
654 #endif
655         }
656         if (!max_mnt_count)
657                 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
658
659         sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
660         sbp[0]->s_state =
661                 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
662         sbp[0]->s_mtime = cpu_to_le64(get_seconds());
663         /* synchronize sbp[1] with sbp[0] */
664         memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
665         return nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
666 }
667
668 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
669                                                  u64 pos, int blocksize,
670                                                  struct buffer_head **pbh)
671 {
672         unsigned long long sb_index = pos;
673         unsigned long offset;
674
675         offset = do_div(sb_index, blocksize);
676         *pbh = sb_bread(sb, sb_index);
677         if (!*pbh)
678                 return NULL;
679         return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
680 }
681
682 int nilfs_store_magic_and_option(struct super_block *sb,
683                                  struct nilfs_super_block *sbp,
684                                  char *data)
685 {
686         struct nilfs_sb_info *sbi = NILFS_SB(sb);
687
688         sb->s_magic = le16_to_cpu(sbp->s_magic);
689
690         /* FS independent flags */
691 #ifdef NILFS_ATIME_DISABLE
692         sb->s_flags |= MS_NOATIME;
693 #endif
694
695         nilfs_set_default_options(sbi, sbp);
696
697         sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
698         sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
699         sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
700         sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
701
702         return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
703 }
704
705 int nilfs_check_feature_compatibility(struct super_block *sb,
706                                       struct nilfs_super_block *sbp)
707 {
708         __u64 features;
709
710         features = le64_to_cpu(sbp->s_feature_incompat) &
711                 ~NILFS_FEATURE_INCOMPAT_SUPP;
712         if (features) {
713                 printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
714                        "optional features (%llx)\n",
715                        (unsigned long long)features);
716                 return -EINVAL;
717         }
718         features = le64_to_cpu(sbp->s_feature_compat_ro) &
719                 ~NILFS_FEATURE_COMPAT_RO_SUPP;
720         if (!(sb->s_flags & MS_RDONLY) && features) {
721                 printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
722                        "unsupported optional features (%llx)\n",
723                        (unsigned long long)features);
724                 return -EINVAL;
725         }
726         return 0;
727 }
728
729 static int nilfs_get_root_dentry(struct super_block *sb,
730                                  struct nilfs_root *root,
731                                  struct dentry **root_dentry)
732 {
733         struct inode *inode;
734         struct dentry *dentry;
735         int ret = 0;
736
737         inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
738         if (IS_ERR(inode)) {
739                 printk(KERN_ERR "NILFS: get root inode failed\n");
740                 ret = PTR_ERR(inode);
741                 goto out;
742         }
743         if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
744                 iput(inode);
745                 printk(KERN_ERR "NILFS: corrupt root inode.\n");
746                 ret = -EINVAL;
747                 goto out;
748         }
749
750         if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
751                 dentry = d_find_alias(inode);
752                 if (!dentry) {
753                         dentry = d_alloc_root(inode);
754                         if (!dentry) {
755                                 iput(inode);
756                                 ret = -ENOMEM;
757                                 goto failed_dentry;
758                         }
759                 } else {
760                         iput(inode);
761                 }
762         } else {
763                 dentry = d_obtain_alias(inode);
764                 if (IS_ERR(dentry)) {
765                         ret = PTR_ERR(dentry);
766                         goto failed_dentry;
767                 }
768         }
769         *root_dentry = dentry;
770  out:
771         return ret;
772
773  failed_dentry:
774         printk(KERN_ERR "NILFS: get root dentry failed\n");
775         goto out;
776 }
777
778 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
779                                  struct dentry **root_dentry)
780 {
781         struct the_nilfs *nilfs = NILFS_SB(s)->s_nilfs;
782         struct nilfs_root *root;
783         int ret;
784
785         down_read(&nilfs->ns_segctor_sem);
786         ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
787         up_read(&nilfs->ns_segctor_sem);
788         if (ret < 0) {
789                 ret = (ret == -ENOENT) ? -EINVAL : ret;
790                 goto out;
791         } else if (!ret) {
792                 printk(KERN_ERR "NILFS: The specified checkpoint is "
793                        "not a snapshot (checkpoint number=%llu).\n",
794                        (unsigned long long)cno);
795                 ret = -EINVAL;
796                 goto out;
797         }
798
799         ret = nilfs_attach_checkpoint(NILFS_SB(s), cno, false, &root);
800         if (ret) {
801                 printk(KERN_ERR "NILFS: error loading snapshot "
802                        "(checkpoint number=%llu).\n",
803                (unsigned long long)cno);
804                 goto out;
805         }
806         ret = nilfs_get_root_dentry(s, root, root_dentry);
807         nilfs_put_root(root);
808  out:
809         return ret;
810 }
811
812 static int nilfs_tree_was_touched(struct dentry *root_dentry)
813 {
814         return atomic_read(&root_dentry->d_count) > 1;
815 }
816
817 /**
818  * nilfs_try_to_shrink_tree() - try to shrink dentries of a checkpoint
819  * @root_dentry: root dentry of the tree to be shrunk
820  *
821  * This function returns true if the tree was in-use.
822  */
823 static int nilfs_try_to_shrink_tree(struct dentry *root_dentry)
824 {
825         if (have_submounts(root_dentry))
826                 return true;
827         shrink_dcache_parent(root_dentry);
828         return nilfs_tree_was_touched(root_dentry);
829 }
830
831 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
832 {
833         struct the_nilfs *nilfs = NILFS_SB(sb)->s_nilfs;
834         struct nilfs_root *root;
835         struct inode *inode;
836         struct dentry *dentry;
837         int ret;
838
839         if (cno < 0 || cno > nilfs->ns_cno)
840                 return false;
841
842         if (cno >= nilfs_last_cno(nilfs))
843                 return true;    /* protect recent checkpoints */
844
845         ret = false;
846         root = nilfs_lookup_root(NILFS_SB(sb)->s_nilfs, cno);
847         if (root) {
848                 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
849                 if (inode) {
850                         dentry = d_find_alias(inode);
851                         if (dentry) {
852                                 if (nilfs_tree_was_touched(dentry))
853                                         ret = nilfs_try_to_shrink_tree(dentry);
854                                 dput(dentry);
855                         }
856                         iput(inode);
857                 }
858                 nilfs_put_root(root);
859         }
860         return ret;
861 }
862
863 /**
864  * nilfs_fill_super() - initialize a super block instance
865  * @sb: super_block
866  * @data: mount options
867  * @silent: silent mode flag
868  *
869  * This function is called exclusively by nilfs->ns_mount_mutex.
870  * So, the recovery process is protected from other simultaneous mounts.
871  */
872 static int
873 nilfs_fill_super(struct super_block *sb, void *data, int silent)
874 {
875         struct the_nilfs *nilfs;
876         struct nilfs_sb_info *sbi;
877         struct nilfs_root *fsroot;
878         __u64 cno;
879         int err;
880
881         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
882         if (!sbi)
883                 return -ENOMEM;
884
885         sb->s_fs_info = sbi;
886         sbi->s_super = sb;
887
888         nilfs = alloc_nilfs(sb->s_bdev);
889         if (!nilfs) {
890                 err = -ENOMEM;
891                 goto failed_sbi;
892         }
893         sbi->s_nilfs = nilfs;
894
895         err = init_nilfs(nilfs, sbi, (char *)data);
896         if (err)
897                 goto failed_nilfs;
898
899         spin_lock_init(&sbi->s_inode_lock);
900         INIT_LIST_HEAD(&sbi->s_dirty_files);
901
902         /*
903          * Following initialization is overlapped because
904          * nilfs_sb_info structure has been cleared at the beginning.
905          * But we reserve them to keep our interest and make ready
906          * for the future change.
907          */
908         get_random_bytes(&sbi->s_next_generation,
909                          sizeof(sbi->s_next_generation));
910         spin_lock_init(&sbi->s_next_gen_lock);
911
912         sb->s_op = &nilfs_sops;
913         sb->s_export_op = &nilfs_export_ops;
914         sb->s_root = NULL;
915         sb->s_time_gran = 1;
916         sb->s_bdi = nilfs->ns_bdi;
917
918         err = load_nilfs(nilfs, sbi);
919         if (err)
920                 goto failed_nilfs;
921
922         cno = nilfs_last_cno(nilfs);
923         err = nilfs_attach_checkpoint(sbi, cno, true, &fsroot);
924         if (err) {
925                 printk(KERN_ERR "NILFS: error loading last checkpoint "
926                        "(checkpoint number=%llu).\n", (unsigned long long)cno);
927                 goto failed_unload;
928         }
929
930         if (!(sb->s_flags & MS_RDONLY)) {
931                 err = nilfs_attach_segment_constructor(sbi, fsroot);
932                 if (err)
933                         goto failed_checkpoint;
934         }
935
936         err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
937         if (err)
938                 goto failed_segctor;
939
940         nilfs_put_root(fsroot);
941
942         if (!(sb->s_flags & MS_RDONLY)) {
943                 down_write(&nilfs->ns_sem);
944                 nilfs_setup_super(sbi);
945                 up_write(&nilfs->ns_sem);
946         }
947
948         return 0;
949
950  failed_segctor:
951         nilfs_detach_segment_constructor(sbi);
952
953  failed_checkpoint:
954         nilfs_put_root(fsroot);
955
956  failed_unload:
957         iput(nilfs->ns_sufile);
958         iput(nilfs->ns_cpfile);
959         iput(nilfs->ns_dat);
960
961  failed_nilfs:
962         destroy_nilfs(nilfs);
963
964  failed_sbi:
965         sb->s_fs_info = NULL;
966         kfree(sbi);
967         return err;
968 }
969
970 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
971 {
972         struct nilfs_sb_info *sbi = NILFS_SB(sb);
973         struct the_nilfs *nilfs = sbi->s_nilfs;
974         unsigned long old_sb_flags;
975         struct nilfs_mount_options old_opts;
976         int err;
977
978         old_sb_flags = sb->s_flags;
979         old_opts.mount_opt = sbi->s_mount_opt;
980
981         if (!parse_options(data, sb, 1)) {
982                 err = -EINVAL;
983                 goto restore_opts;
984         }
985         sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
986
987         err = -EINVAL;
988
989         if (!nilfs_valid_fs(nilfs)) {
990                 printk(KERN_WARNING "NILFS (device %s): couldn't "
991                        "remount because the filesystem is in an "
992                        "incomplete recovery state.\n", sb->s_id);
993                 goto restore_opts;
994         }
995
996         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
997                 goto out;
998         if (*flags & MS_RDONLY) {
999                 /* Shutting down the segment constructor */
1000                 nilfs_detach_segment_constructor(sbi);
1001                 sb->s_flags |= MS_RDONLY;
1002
1003                 /*
1004                  * Remounting a valid RW partition RDONLY, so set
1005                  * the RDONLY flag and then mark the partition as valid again.
1006                  */
1007                 down_write(&nilfs->ns_sem);
1008                 nilfs_cleanup_super(sbi);
1009                 up_write(&nilfs->ns_sem);
1010         } else {
1011                 __u64 features;
1012                 struct nilfs_root *root;
1013
1014                 /*
1015                  * Mounting a RDONLY partition read-write, so reread and
1016                  * store the current valid flag.  (It may have been changed
1017                  * by fsck since we originally mounted the partition.)
1018                  */
1019                 down_read(&nilfs->ns_sem);
1020                 features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1021                         ~NILFS_FEATURE_COMPAT_RO_SUPP;
1022                 up_read(&nilfs->ns_sem);
1023                 if (features) {
1024                         printk(KERN_WARNING "NILFS (device %s): couldn't "
1025                                "remount RDWR because of unsupported optional "
1026                                "features (%llx)\n",
1027                                sb->s_id, (unsigned long long)features);
1028                         err = -EROFS;
1029                         goto restore_opts;
1030                 }
1031
1032                 sb->s_flags &= ~MS_RDONLY;
1033
1034                 root = NILFS_I(sb->s_root->d_inode)->i_root;
1035                 err = nilfs_attach_segment_constructor(sbi, root);
1036                 if (err)
1037                         goto restore_opts;
1038
1039                 down_write(&nilfs->ns_sem);
1040                 nilfs_setup_super(sbi);
1041                 up_write(&nilfs->ns_sem);
1042         }
1043  out:
1044         return 0;
1045
1046  restore_opts:
1047         sb->s_flags = old_sb_flags;
1048         sbi->s_mount_opt = old_opts.mount_opt;
1049         return err;
1050 }
1051
1052 struct nilfs_super_data {
1053         struct block_device *bdev;
1054         struct nilfs_sb_info *sbi;
1055         __u64 cno;
1056         int flags;
1057 };
1058
1059 /**
1060  * nilfs_identify - pre-read mount options needed to identify mount instance
1061  * @data: mount options
1062  * @sd: nilfs_super_data
1063  */
1064 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1065 {
1066         char *p, *options = data;
1067         substring_t args[MAX_OPT_ARGS];
1068         int option, token;
1069         int ret = 0;
1070
1071         do {
1072                 p = strsep(&options, ",");
1073                 if (p != NULL && *p) {
1074                         token = match_token(p, tokens, args);
1075                         if (token == Opt_snapshot) {
1076                                 if (!(sd->flags & MS_RDONLY))
1077                                         ret++;
1078                                 else {
1079                                         ret = match_int(&args[0], &option);
1080                                         if (!ret) {
1081                                                 if (option > 0)
1082                                                         sd->cno = option;
1083                                                 else
1084                                                         ret++;
1085                                         }
1086                                 }
1087                         }
1088                         if (ret)
1089                                 printk(KERN_ERR
1090                                        "NILFS: invalid mount option: %s\n", p);
1091                 }
1092                 if (!options)
1093                         break;
1094                 BUG_ON(options == data);
1095                 *(options - 1) = ',';
1096         } while (!ret);
1097         return ret;
1098 }
1099
1100 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1101 {
1102         s->s_bdev = data;
1103         s->s_dev = s->s_bdev->bd_dev;
1104         return 0;
1105 }
1106
1107 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1108 {
1109         return (void *)s->s_bdev == data;
1110 }
1111
1112 static int
1113 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1114              const char *dev_name, void *data, struct vfsmount *mnt)
1115 {
1116         struct nilfs_super_data sd;
1117         struct super_block *s;
1118         fmode_t mode = FMODE_READ;
1119         struct dentry *root_dentry;
1120         int err, s_new = false;
1121
1122         if (!(flags & MS_RDONLY))
1123                 mode |= FMODE_WRITE;
1124
1125         sd.bdev = open_bdev_exclusive(dev_name, mode, fs_type);
1126         if (IS_ERR(sd.bdev))
1127                 return PTR_ERR(sd.bdev);
1128
1129         sd.cno = 0;
1130         sd.flags = flags;
1131         if (nilfs_identify((char *)data, &sd)) {
1132                 err = -EINVAL;
1133                 goto failed;
1134         }
1135
1136         s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, sd.bdev);
1137         if (IS_ERR(s)) {
1138                 err = PTR_ERR(s);
1139                 goto failed;
1140         }
1141
1142         if (!s->s_root) {
1143                 char b[BDEVNAME_SIZE];
1144
1145                 s_new = true;
1146
1147                 /* New superblock instance created */
1148                 s->s_flags = flags;
1149                 s->s_mode = mode;
1150                 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1151                 sb_set_blocksize(s, block_size(sd.bdev));
1152
1153                 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1154                 if (err)
1155                         goto failed_super;
1156
1157                 s->s_flags |= MS_ACTIVE;
1158         } else if (!sd.cno) {
1159                 int busy = false;
1160
1161                 if (nilfs_tree_was_touched(s->s_root)) {
1162                         busy = nilfs_try_to_shrink_tree(s->s_root);
1163                         if (busy && (flags ^ s->s_flags) & MS_RDONLY) {
1164                                 printk(KERN_ERR "NILFS: the device already "
1165                                        "has a %s mount.\n",
1166                                        (s->s_flags & MS_RDONLY) ?
1167                                        "read-only" : "read/write");
1168                                 err = -EBUSY;
1169                                 goto failed_super;
1170                         }
1171                 }
1172                 if (!busy) {
1173                         /*
1174                          * Try remount to setup mount states if the current
1175                          * tree is not mounted and only snapshots use this sb.
1176                          */
1177                         err = nilfs_remount(s, &flags, data);
1178                         if (err)
1179                                 goto failed_super;
1180                 }
1181         }
1182
1183         if (sd.cno) {
1184                 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1185                 if (err)
1186                         goto failed_super;
1187         } else {
1188                 root_dentry = dget(s->s_root);
1189         }
1190
1191         if (!s_new)
1192                 close_bdev_exclusive(sd.bdev, mode);
1193
1194         mnt->mnt_sb = s;
1195         mnt->mnt_root = root_dentry;
1196         return 0;
1197
1198  failed_super:
1199         deactivate_locked_super(s);
1200
1201  failed:
1202         if (!s_new)
1203                 close_bdev_exclusive(sd.bdev, mode);
1204         return err;
1205 }
1206
1207 struct file_system_type nilfs_fs_type = {
1208         .owner    = THIS_MODULE,
1209         .name     = "nilfs2",
1210         .get_sb   = nilfs_get_sb,
1211         .kill_sb  = kill_block_super,
1212         .fs_flags = FS_REQUIRES_DEV,
1213 };
1214
1215 static void nilfs_inode_init_once(void *obj)
1216 {
1217         struct nilfs_inode_info *ii = obj;
1218
1219         INIT_LIST_HEAD(&ii->i_dirty);
1220 #ifdef CONFIG_NILFS_XATTR
1221         init_rwsem(&ii->xattr_sem);
1222 #endif
1223         nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
1224         ii->i_bmap = &ii->i_bmap_data;
1225         inode_init_once(&ii->vfs_inode);
1226 }
1227
1228 static void nilfs_segbuf_init_once(void *obj)
1229 {
1230         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1231 }
1232
1233 static void nilfs_destroy_cachep(void)
1234 {
1235         if (nilfs_inode_cachep)
1236                 kmem_cache_destroy(nilfs_inode_cachep);
1237         if (nilfs_transaction_cachep)
1238                 kmem_cache_destroy(nilfs_transaction_cachep);
1239         if (nilfs_segbuf_cachep)
1240                 kmem_cache_destroy(nilfs_segbuf_cachep);
1241         if (nilfs_btree_path_cache)
1242                 kmem_cache_destroy(nilfs_btree_path_cache);
1243 }
1244
1245 static int __init nilfs_init_cachep(void)
1246 {
1247         nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1248                         sizeof(struct nilfs_inode_info), 0,
1249                         SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1250         if (!nilfs_inode_cachep)
1251                 goto fail;
1252
1253         nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1254                         sizeof(struct nilfs_transaction_info), 0,
1255                         SLAB_RECLAIM_ACCOUNT, NULL);
1256         if (!nilfs_transaction_cachep)
1257                 goto fail;
1258
1259         nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1260                         sizeof(struct nilfs_segment_buffer), 0,
1261                         SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1262         if (!nilfs_segbuf_cachep)
1263                 goto fail;
1264
1265         nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1266                         sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1267                         0, 0, NULL);
1268         if (!nilfs_btree_path_cache)
1269                 goto fail;
1270
1271         return 0;
1272
1273 fail:
1274         nilfs_destroy_cachep();
1275         return -ENOMEM;
1276 }
1277
1278 static int __init init_nilfs_fs(void)
1279 {
1280         int err;
1281
1282         err = nilfs_init_cachep();
1283         if (err)
1284                 goto fail;
1285
1286         err = register_filesystem(&nilfs_fs_type);
1287         if (err)
1288                 goto free_cachep;
1289
1290         printk(KERN_INFO "NILFS version 2 loaded\n");
1291         return 0;
1292
1293 free_cachep:
1294         nilfs_destroy_cachep();
1295 fail:
1296         return err;
1297 }
1298
1299 static void __exit exit_nilfs_fs(void)
1300 {
1301         nilfs_destroy_cachep();
1302         unregister_filesystem(&nilfs_fs_type);
1303 }
1304
1305 module_init(init_nilfs_fs)
1306 module_exit(exit_nilfs_fs)