xref: /linux/fs/nilfs2/super.c (revision e26207a3819684e9b4450a2d30bdd065fa92d9c7)
1 /*
2  * super.c - NILFS module and super block management.
3  *
4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * This program is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
19  *
20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
21  */
22 /*
23  *  linux/fs/ext2/super.c
24  *
25  * Copyright (C) 1992, 1993, 1994, 1995
26  * Remy Card (card@masi.ibp.fr)
27  * Laboratoire MASI - Institut Blaise Pascal
28  * Universite Pierre et Marie Curie (Paris VI)
29  *
30  *  from
31  *
32  *  linux/fs/minix/inode.c
33  *
34  *  Copyright (C) 1991, 1992  Linus Torvalds
35  *
36  *  Big-endian to little-endian byte-swapping/bitmaps by
37  *        David S. Miller (davem@caip.rutgers.edu), 1995
38  */
39 
40 #include <linux/module.h>
41 #include <linux/string.h>
42 #include <linux/slab.h>
43 #include <linux/init.h>
44 #include <linux/blkdev.h>
45 #include <linux/parser.h>
46 #include <linux/random.h>
47 #include <linux/crc32.h>
48 #include <linux/smp_lock.h>
49 #include <linux/vfs.h>
50 #include <linux/writeback.h>
51 #include <linux/kobject.h>
52 #include <linux/exportfs.h>
53 #include <linux/seq_file.h>
54 #include <linux/mount.h>
55 #include "nilfs.h"
56 #include "mdt.h"
57 #include "alloc.h"
58 #include "page.h"
59 #include "cpfile.h"
60 #include "ifile.h"
61 #include "dat.h"
62 #include "segment.h"
63 #include "segbuf.h"
64 
65 MODULE_AUTHOR("NTT Corp.");
66 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
67 		   "(NILFS)");
68 MODULE_LICENSE("GPL");
69 
70 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
71 
72 /**
73  * nilfs_error() - report failure condition on a filesystem
74  *
75  * nilfs_error() sets an ERROR_FS flag on the superblock as well as
76  * reporting an error message.  It should be called when NILFS detects
77  * incoherences or defects of meta data on disk.  As for sustainable
78  * errors such as a single-shot I/O error, nilfs_warning() or the printk()
79  * function should be used instead.
80  *
81  * The segment constructor must not call this function because it can
82  * kill itself.
83  */
84 void nilfs_error(struct super_block *sb, const char *function,
85 		 const char *fmt, ...)
86 {
87 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
88 	va_list args;
89 
90 	va_start(args, fmt);
91 	printk(KERN_CRIT "NILFS error (device %s): %s: ", sb->s_id, function);
92 	vprintk(fmt, args);
93 	printk("\n");
94 	va_end(args);
95 
96 	if (!(sb->s_flags & MS_RDONLY)) {
97 		struct the_nilfs *nilfs = sbi->s_nilfs;
98 
99 		if (!nilfs_test_opt(sbi, ERRORS_CONT))
100 			nilfs_detach_segment_constructor(sbi);
101 
102 		down_write(&nilfs->ns_sem);
103 		if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
104 			nilfs->ns_mount_state |= NILFS_ERROR_FS;
105 			nilfs->ns_sbp[0]->s_state |=
106 				cpu_to_le16(NILFS_ERROR_FS);
107 			nilfs_commit_super(sbi, 1);
108 		}
109 		up_write(&nilfs->ns_sem);
110 
111 		if (nilfs_test_opt(sbi, ERRORS_RO)) {
112 			printk(KERN_CRIT "Remounting filesystem read-only\n");
113 			sb->s_flags |= MS_RDONLY;
114 		}
115 	}
116 
117 	if (nilfs_test_opt(sbi, ERRORS_PANIC))
118 		panic("NILFS (device %s): panic forced after error\n",
119 		      sb->s_id);
120 }
121 
122 void nilfs_warning(struct super_block *sb, const char *function,
123 		   const char *fmt, ...)
124 {
125 	va_list args;
126 
127 	va_start(args, fmt);
128 	printk(KERN_WARNING "NILFS warning (device %s): %s: ",
129 	       sb->s_id, function);
130 	vprintk(fmt, args);
131 	printk("\n");
132 	va_end(args);
133 }
134 
135 static struct kmem_cache *nilfs_inode_cachep;
136 
137 struct inode *nilfs_alloc_inode_common(struct the_nilfs *nilfs)
138 {
139 	struct nilfs_inode_info *ii;
140 
141 	ii = kmem_cache_alloc(nilfs_inode_cachep, GFP_NOFS);
142 	if (!ii)
143 		return NULL;
144 	ii->i_bh = NULL;
145 	ii->i_state = 0;
146 	ii->vfs_inode.i_version = 1;
147 	nilfs_btnode_cache_init(&ii->i_btnode_cache, nilfs->ns_bdi);
148 	return &ii->vfs_inode;
149 }
150 
151 struct inode *nilfs_alloc_inode(struct super_block *sb)
152 {
153 	return nilfs_alloc_inode_common(NILFS_SB(sb)->s_nilfs);
154 }
155 
156 void nilfs_destroy_inode(struct inode *inode)
157 {
158 	kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
159 }
160 
161 static void init_once(void *obj)
162 {
163 	struct nilfs_inode_info *ii = obj;
164 
165 	INIT_LIST_HEAD(&ii->i_dirty);
166 #ifdef CONFIG_NILFS_XATTR
167 	init_rwsem(&ii->xattr_sem);
168 #endif
169 	nilfs_btnode_cache_init_once(&ii->i_btnode_cache);
170 	ii->i_bmap = (struct nilfs_bmap *)&ii->i_bmap_union;
171 	inode_init_once(&ii->vfs_inode);
172 }
173 
174 static int nilfs_init_inode_cache(void)
175 {
176 	nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
177 					       sizeof(struct nilfs_inode_info),
178 					       0, SLAB_RECLAIM_ACCOUNT,
179 					       init_once);
180 
181 	return (nilfs_inode_cachep == NULL) ? -ENOMEM : 0;
182 }
183 
184 static inline void nilfs_destroy_inode_cache(void)
185 {
186 	kmem_cache_destroy(nilfs_inode_cachep);
187 }
188 
189 static void nilfs_clear_inode(struct inode *inode)
190 {
191 	struct nilfs_inode_info *ii = NILFS_I(inode);
192 
193 	/*
194 	 * Free resources allocated in nilfs_read_inode(), here.
195 	 */
196 	BUG_ON(!list_empty(&ii->i_dirty));
197 	brelse(ii->i_bh);
198 	ii->i_bh = NULL;
199 
200 	if (test_bit(NILFS_I_BMAP, &ii->i_state))
201 		nilfs_bmap_clear(ii->i_bmap);
202 
203 	nilfs_btnode_cache_clear(&ii->i_btnode_cache);
204 }
205 
206 static int nilfs_sync_super(struct nilfs_sb_info *sbi, int dupsb)
207 {
208 	struct the_nilfs *nilfs = sbi->s_nilfs;
209 	int err;
210 	int barrier_done = 0;
211 
212 	if (nilfs_test_opt(sbi, BARRIER)) {
213 		set_buffer_ordered(nilfs->ns_sbh[0]);
214 		barrier_done = 1;
215 	}
216  retry:
217 	set_buffer_dirty(nilfs->ns_sbh[0]);
218 	err = sync_dirty_buffer(nilfs->ns_sbh[0]);
219 	if (err == -EOPNOTSUPP && barrier_done) {
220 		nilfs_warning(sbi->s_super, __func__,
221 			      "barrier-based sync failed. "
222 			      "disabling barriers\n");
223 		nilfs_clear_opt(sbi, BARRIER);
224 		barrier_done = 0;
225 		clear_buffer_ordered(nilfs->ns_sbh[0]);
226 		goto retry;
227 	}
228 	if (unlikely(err)) {
229 		printk(KERN_ERR
230 		       "NILFS: unable to write superblock (err=%d)\n", err);
231 		if (err == -EIO && nilfs->ns_sbh[1]) {
232 			nilfs_fall_back_super_block(nilfs);
233 			goto retry;
234 		}
235 	} else {
236 		struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
237 
238 		/*
239 		 * The latest segment becomes trailable from the position
240 		 * written in superblock.
241 		 */
242 		clear_nilfs_discontinued(nilfs);
243 
244 		/* update GC protection for recent segments */
245 		if (nilfs->ns_sbh[1]) {
246 			sbp = NULL;
247 			if (dupsb) {
248 				set_buffer_dirty(nilfs->ns_sbh[1]);
249 				if (!sync_dirty_buffer(nilfs->ns_sbh[1]))
250 					sbp = nilfs->ns_sbp[1];
251 			}
252 		}
253 		if (sbp) {
254 			spin_lock(&nilfs->ns_last_segment_lock);
255 			nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
256 			spin_unlock(&nilfs->ns_last_segment_lock);
257 		}
258 	}
259 
260 	return err;
261 }
262 
263 int nilfs_commit_super(struct nilfs_sb_info *sbi, int dupsb)
264 {
265 	struct the_nilfs *nilfs = sbi->s_nilfs;
266 	struct nilfs_super_block **sbp = nilfs->ns_sbp;
267 	sector_t nfreeblocks;
268 	time_t t;
269 	int err;
270 
271 	/* nilfs->sem must be locked by the caller. */
272 	if (sbp[0]->s_magic != NILFS_SUPER_MAGIC) {
273 		if (sbp[1] && sbp[1]->s_magic == NILFS_SUPER_MAGIC)
274 			nilfs_swap_super_block(nilfs);
275 		else {
276 			printk(KERN_CRIT "NILFS: superblock broke on dev %s\n",
277 			       sbi->s_super->s_id);
278 			return -EIO;
279 		}
280 	}
281 	err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
282 	if (unlikely(err)) {
283 		printk(KERN_ERR "NILFS: failed to count free blocks\n");
284 		return err;
285 	}
286 	spin_lock(&nilfs->ns_last_segment_lock);
287 	sbp[0]->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
288 	sbp[0]->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
289 	sbp[0]->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
290 	spin_unlock(&nilfs->ns_last_segment_lock);
291 
292 	t = get_seconds();
293 	nilfs->ns_sbwtime[0] = t;
294 	sbp[0]->s_free_blocks_count = cpu_to_le64(nfreeblocks);
295 	sbp[0]->s_wtime = cpu_to_le64(t);
296 	sbp[0]->s_sum = 0;
297 	sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
298 					     (unsigned char *)sbp[0],
299 					     nilfs->ns_sbsize));
300 	if (dupsb && sbp[1]) {
301 		memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
302 		nilfs->ns_sbwtime[1] = t;
303 	}
304 	sbi->s_super->s_dirt = 0;
305 	return nilfs_sync_super(sbi, dupsb);
306 }
307 
308 static void nilfs_put_super(struct super_block *sb)
309 {
310 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
311 	struct the_nilfs *nilfs = sbi->s_nilfs;
312 
313 	lock_kernel();
314 
315 	nilfs_detach_segment_constructor(sbi);
316 
317 	if (!(sb->s_flags & MS_RDONLY)) {
318 		down_write(&nilfs->ns_sem);
319 		nilfs->ns_sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
320 		nilfs_commit_super(sbi, 1);
321 		up_write(&nilfs->ns_sem);
322 	}
323 	down_write(&nilfs->ns_super_sem);
324 	if (nilfs->ns_current == sbi)
325 		nilfs->ns_current = NULL;
326 	up_write(&nilfs->ns_super_sem);
327 
328 	nilfs_detach_checkpoint(sbi);
329 	put_nilfs(sbi->s_nilfs);
330 	sbi->s_super = NULL;
331 	sb->s_fs_info = NULL;
332 	nilfs_put_sbinfo(sbi);
333 
334 	unlock_kernel();
335 }
336 
337 static int nilfs_sync_fs(struct super_block *sb, int wait)
338 {
339 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
340 	struct the_nilfs *nilfs = sbi->s_nilfs;
341 	int err = 0;
342 
343 	/* This function is called when super block should be written back */
344 	if (wait)
345 		err = nilfs_construct_segment(sb);
346 
347 	down_write(&nilfs->ns_sem);
348 	if (sb->s_dirt)
349 		nilfs_commit_super(sbi, 1);
350 	up_write(&nilfs->ns_sem);
351 
352 	return err;
353 }
354 
355 int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
356 {
357 	struct the_nilfs *nilfs = sbi->s_nilfs;
358 	struct nilfs_checkpoint *raw_cp;
359 	struct buffer_head *bh_cp;
360 	int err;
361 
362 	down_write(&nilfs->ns_super_sem);
363 	list_add(&sbi->s_list, &nilfs->ns_supers);
364 	up_write(&nilfs->ns_super_sem);
365 
366 	sbi->s_ifile = nilfs_ifile_new(sbi, nilfs->ns_inode_size);
367 	if (!sbi->s_ifile)
368 		return -ENOMEM;
369 
370 	down_read(&nilfs->ns_segctor_sem);
371 	err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
372 					  &bh_cp);
373 	up_read(&nilfs->ns_segctor_sem);
374 	if (unlikely(err)) {
375 		if (err == -ENOENT || err == -EINVAL) {
376 			printk(KERN_ERR
377 			       "NILFS: Invalid checkpoint "
378 			       "(checkpoint number=%llu)\n",
379 			       (unsigned long long)cno);
380 			err = -EINVAL;
381 		}
382 		goto failed;
383 	}
384 	err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode);
385 	if (unlikely(err))
386 		goto failed_bh;
387 	atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
388 	atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
389 
390 	nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
391 	return 0;
392 
393  failed_bh:
394 	nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
395  failed:
396 	nilfs_mdt_destroy(sbi->s_ifile);
397 	sbi->s_ifile = NULL;
398 
399 	down_write(&nilfs->ns_super_sem);
400 	list_del_init(&sbi->s_list);
401 	up_write(&nilfs->ns_super_sem);
402 
403 	return err;
404 }
405 
406 void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi)
407 {
408 	struct the_nilfs *nilfs = sbi->s_nilfs;
409 
410 	nilfs_mdt_destroy(sbi->s_ifile);
411 	sbi->s_ifile = NULL;
412 	down_write(&nilfs->ns_super_sem);
413 	list_del_init(&sbi->s_list);
414 	up_write(&nilfs->ns_super_sem);
415 }
416 
417 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
418 {
419 	struct super_block *sb = dentry->d_sb;
420 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
421 	struct the_nilfs *nilfs = sbi->s_nilfs;
422 	u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
423 	unsigned long long blocks;
424 	unsigned long overhead;
425 	unsigned long nrsvblocks;
426 	sector_t nfreeblocks;
427 	int err;
428 
429 	/*
430 	 * Compute all of the segment blocks
431 	 *
432 	 * The blocks before first segment and after last segment
433 	 * are excluded.
434 	 */
435 	blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
436 		- nilfs->ns_first_data_block;
437 	nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
438 
439 	/*
440 	 * Compute the overhead
441 	 *
442 	 * When distributing meta data blocks outside semgent structure,
443 	 * We must count them as the overhead.
444 	 */
445 	overhead = 0;
446 
447 	err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
448 	if (unlikely(err))
449 		return err;
450 
451 	buf->f_type = NILFS_SUPER_MAGIC;
452 	buf->f_bsize = sb->s_blocksize;
453 	buf->f_blocks = blocks - overhead;
454 	buf->f_bfree = nfreeblocks;
455 	buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
456 		(buf->f_bfree - nrsvblocks) : 0;
457 	buf->f_files = atomic_read(&sbi->s_inodes_count);
458 	buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
459 	buf->f_namelen = NILFS_NAME_LEN;
460 	buf->f_fsid.val[0] = (u32)id;
461 	buf->f_fsid.val[1] = (u32)(id >> 32);
462 
463 	return 0;
464 }
465 
466 static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
467 {
468 	struct super_block *sb = vfs->mnt_sb;
469 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
470 
471 	if (!nilfs_test_opt(sbi, BARRIER))
472 		seq_printf(seq, ",nobarrier");
473 	if (nilfs_test_opt(sbi, SNAPSHOT))
474 		seq_printf(seq, ",cp=%llu",
475 			   (unsigned long long int)sbi->s_snapshot_cno);
476 	if (nilfs_test_opt(sbi, ERRORS_RO))
477 		seq_printf(seq, ",errors=remount-ro");
478 	if (nilfs_test_opt(sbi, ERRORS_PANIC))
479 		seq_printf(seq, ",errors=panic");
480 	if (nilfs_test_opt(sbi, STRICT_ORDER))
481 		seq_printf(seq, ",order=strict");
482 	if (nilfs_test_opt(sbi, NORECOVERY))
483 		seq_printf(seq, ",norecovery");
484 
485 	return 0;
486 }
487 
488 static const struct super_operations nilfs_sops = {
489 	.alloc_inode    = nilfs_alloc_inode,
490 	.destroy_inode  = nilfs_destroy_inode,
491 	.dirty_inode    = nilfs_dirty_inode,
492 	/* .write_inode    = nilfs_write_inode, */
493 	/* .put_inode      = nilfs_put_inode, */
494 	/* .drop_inode	  = nilfs_drop_inode, */
495 	.delete_inode   = nilfs_delete_inode,
496 	.put_super      = nilfs_put_super,
497 	/* .write_super    = nilfs_write_super, */
498 	.sync_fs        = nilfs_sync_fs,
499 	/* .write_super_lockfs */
500 	/* .unlockfs */
501 	.statfs         = nilfs_statfs,
502 	.remount_fs     = nilfs_remount,
503 	.clear_inode    = nilfs_clear_inode,
504 	/* .umount_begin */
505 	.show_options = nilfs_show_options
506 };
507 
508 static struct inode *
509 nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
510 {
511 	struct inode *inode;
512 
513 	if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
514 	    ino != NILFS_SKETCH_INO)
515 		return ERR_PTR(-ESTALE);
516 
517 	inode = nilfs_iget(sb, ino);
518 	if (IS_ERR(inode))
519 		return ERR_CAST(inode);
520 	if (generation && inode->i_generation != generation) {
521 		iput(inode);
522 		return ERR_PTR(-ESTALE);
523 	}
524 
525 	return inode;
526 }
527 
528 static struct dentry *
529 nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
530 		   int fh_type)
531 {
532 	return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
533 				    nilfs_nfs_get_inode);
534 }
535 
536 static struct dentry *
537 nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
538 		   int fh_type)
539 {
540 	return generic_fh_to_parent(sb, fid, fh_len, fh_type,
541 				    nilfs_nfs_get_inode);
542 }
543 
544 static const struct export_operations nilfs_export_ops = {
545 	.fh_to_dentry = nilfs_fh_to_dentry,
546 	.fh_to_parent = nilfs_fh_to_parent,
547 	.get_parent = nilfs_get_parent,
548 };
549 
550 enum {
551 	Opt_err_cont, Opt_err_panic, Opt_err_ro,
552 	Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
553 	Opt_err,
554 };
555 
556 static match_table_t tokens = {
557 	{Opt_err_cont, "errors=continue"},
558 	{Opt_err_panic, "errors=panic"},
559 	{Opt_err_ro, "errors=remount-ro"},
560 	{Opt_nobarrier, "nobarrier"},
561 	{Opt_snapshot, "cp=%u"},
562 	{Opt_order, "order=%s"},
563 	{Opt_norecovery, "norecovery"},
564 	{Opt_err, NULL}
565 };
566 
567 static int parse_options(char *options, struct super_block *sb)
568 {
569 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
570 	char *p;
571 	substring_t args[MAX_OPT_ARGS];
572 	int option;
573 
574 	if (!options)
575 		return 1;
576 
577 	while ((p = strsep(&options, ",")) != NULL) {
578 		int token;
579 		if (!*p)
580 			continue;
581 
582 		token = match_token(p, tokens, args);
583 		switch (token) {
584 		case Opt_nobarrier:
585 			nilfs_clear_opt(sbi, BARRIER);
586 			break;
587 		case Opt_order:
588 			if (strcmp(args[0].from, "relaxed") == 0)
589 				/* Ordered data semantics */
590 				nilfs_clear_opt(sbi, STRICT_ORDER);
591 			else if (strcmp(args[0].from, "strict") == 0)
592 				/* Strict in-order semantics */
593 				nilfs_set_opt(sbi, STRICT_ORDER);
594 			else
595 				return 0;
596 			break;
597 		case Opt_err_panic:
598 			nilfs_write_opt(sbi, ERROR_MODE, ERRORS_PANIC);
599 			break;
600 		case Opt_err_ro:
601 			nilfs_write_opt(sbi, ERROR_MODE, ERRORS_RO);
602 			break;
603 		case Opt_err_cont:
604 			nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
605 			break;
606 		case Opt_snapshot:
607 			if (match_int(&args[0], &option) || option <= 0)
608 				return 0;
609 			if (!(sb->s_flags & MS_RDONLY))
610 				return 0;
611 			sbi->s_snapshot_cno = option;
612 			nilfs_set_opt(sbi, SNAPSHOT);
613 			break;
614 		case Opt_norecovery:
615 			nilfs_set_opt(sbi, NORECOVERY);
616 			break;
617 		default:
618 			printk(KERN_ERR
619 			       "NILFS: Unrecognized mount option \"%s\"\n", p);
620 			return 0;
621 		}
622 	}
623 	return 1;
624 }
625 
626 static inline void
627 nilfs_set_default_options(struct nilfs_sb_info *sbi,
628 			  struct nilfs_super_block *sbp)
629 {
630 	sbi->s_mount_opt =
631 		NILFS_MOUNT_ERRORS_CONT | NILFS_MOUNT_BARRIER;
632 }
633 
634 static int nilfs_setup_super(struct nilfs_sb_info *sbi)
635 {
636 	struct the_nilfs *nilfs = sbi->s_nilfs;
637 	struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
638 	int max_mnt_count = le16_to_cpu(sbp->s_max_mnt_count);
639 	int mnt_count = le16_to_cpu(sbp->s_mnt_count);
640 
641 	/* nilfs->sem must be locked by the caller. */
642 	if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
643 		printk(KERN_WARNING
644 		       "NILFS warning: mounting fs with errors\n");
645 #if 0
646 	} else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
647 		printk(KERN_WARNING
648 		       "NILFS warning: maximal mount count reached\n");
649 #endif
650 	}
651 	if (!max_mnt_count)
652 		sbp->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
653 
654 	sbp->s_mnt_count = cpu_to_le16(mnt_count + 1);
655 	sbp->s_state = cpu_to_le16(le16_to_cpu(sbp->s_state) & ~NILFS_VALID_FS);
656 	sbp->s_mtime = cpu_to_le64(get_seconds());
657 	return nilfs_commit_super(sbi, 1);
658 }
659 
660 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
661 						 u64 pos, int blocksize,
662 						 struct buffer_head **pbh)
663 {
664 	unsigned long long sb_index = pos;
665 	unsigned long offset;
666 
667 	offset = do_div(sb_index, blocksize);
668 	*pbh = sb_bread(sb, sb_index);
669 	if (!*pbh)
670 		return NULL;
671 	return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
672 }
673 
674 int nilfs_store_magic_and_option(struct super_block *sb,
675 				 struct nilfs_super_block *sbp,
676 				 char *data)
677 {
678 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
679 
680 	sb->s_magic = le16_to_cpu(sbp->s_magic);
681 
682 	/* FS independent flags */
683 #ifdef NILFS_ATIME_DISABLE
684 	sb->s_flags |= MS_NOATIME;
685 #endif
686 
687 	nilfs_set_default_options(sbi, sbp);
688 
689 	sbi->s_resuid = le16_to_cpu(sbp->s_def_resuid);
690 	sbi->s_resgid = le16_to_cpu(sbp->s_def_resgid);
691 	sbi->s_interval = le32_to_cpu(sbp->s_c_interval);
692 	sbi->s_watermark = le32_to_cpu(sbp->s_c_block_max);
693 
694 	return !parse_options(data, sb) ? -EINVAL : 0 ;
695 }
696 
697 /**
698  * nilfs_fill_super() - initialize a super block instance
699  * @sb: super_block
700  * @data: mount options
701  * @silent: silent mode flag
702  * @nilfs: the_nilfs struct
703  *
704  * This function is called exclusively by nilfs->ns_mount_mutex.
705  * So, the recovery process is protected from other simultaneous mounts.
706  */
707 static int
708 nilfs_fill_super(struct super_block *sb, void *data, int silent,
709 		 struct the_nilfs *nilfs)
710 {
711 	struct nilfs_sb_info *sbi;
712 	struct inode *root;
713 	__u64 cno;
714 	int err;
715 
716 	sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
717 	if (!sbi)
718 		return -ENOMEM;
719 
720 	sb->s_fs_info = sbi;
721 
722 	get_nilfs(nilfs);
723 	sbi->s_nilfs = nilfs;
724 	sbi->s_super = sb;
725 	atomic_set(&sbi->s_count, 1);
726 
727 	err = init_nilfs(nilfs, sbi, (char *)data);
728 	if (err)
729 		goto failed_sbi;
730 
731 	spin_lock_init(&sbi->s_inode_lock);
732 	INIT_LIST_HEAD(&sbi->s_dirty_files);
733 	INIT_LIST_HEAD(&sbi->s_list);
734 
735 	/*
736 	 * Following initialization is overlapped because
737 	 * nilfs_sb_info structure has been cleared at the beginning.
738 	 * But we reserve them to keep our interest and make ready
739 	 * for the future change.
740 	 */
741 	get_random_bytes(&sbi->s_next_generation,
742 			 sizeof(sbi->s_next_generation));
743 	spin_lock_init(&sbi->s_next_gen_lock);
744 
745 	sb->s_op = &nilfs_sops;
746 	sb->s_export_op = &nilfs_export_ops;
747 	sb->s_root = NULL;
748 	sb->s_time_gran = 1;
749 
750 	err = load_nilfs(nilfs, sbi);
751 	if (err)
752 		goto failed_sbi;
753 
754 	cno = nilfs_last_cno(nilfs);
755 
756 	if (sb->s_flags & MS_RDONLY) {
757 		if (nilfs_test_opt(sbi, SNAPSHOT)) {
758 			down_read(&nilfs->ns_segctor_sem);
759 			err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
760 						       sbi->s_snapshot_cno);
761 			up_read(&nilfs->ns_segctor_sem);
762 			if (err < 0) {
763 				if (err == -ENOENT)
764 					err = -EINVAL;
765 				goto failed_sbi;
766 			}
767 			if (!err) {
768 				printk(KERN_ERR
769 				       "NILFS: The specified checkpoint is "
770 				       "not a snapshot "
771 				       "(checkpoint number=%llu).\n",
772 				       (unsigned long long)sbi->s_snapshot_cno);
773 				err = -EINVAL;
774 				goto failed_sbi;
775 			}
776 			cno = sbi->s_snapshot_cno;
777 		} else
778 			/* Read-only mount */
779 			sbi->s_snapshot_cno = cno;
780 	}
781 
782 	err = nilfs_attach_checkpoint(sbi, cno);
783 	if (err) {
784 		printk(KERN_ERR "NILFS: error loading a checkpoint"
785 		       " (checkpoint number=%llu).\n", (unsigned long long)cno);
786 		goto failed_sbi;
787 	}
788 
789 	if (!(sb->s_flags & MS_RDONLY)) {
790 		err = nilfs_attach_segment_constructor(sbi);
791 		if (err)
792 			goto failed_checkpoint;
793 	}
794 
795 	root = nilfs_iget(sb, NILFS_ROOT_INO);
796 	if (IS_ERR(root)) {
797 		printk(KERN_ERR "NILFS: get root inode failed\n");
798 		err = PTR_ERR(root);
799 		goto failed_segctor;
800 	}
801 	if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
802 		iput(root);
803 		printk(KERN_ERR "NILFS: corrupt root inode.\n");
804 		err = -EINVAL;
805 		goto failed_segctor;
806 	}
807 	sb->s_root = d_alloc_root(root);
808 	if (!sb->s_root) {
809 		iput(root);
810 		printk(KERN_ERR "NILFS: get root dentry failed\n");
811 		err = -ENOMEM;
812 		goto failed_segctor;
813 	}
814 
815 	if (!(sb->s_flags & MS_RDONLY)) {
816 		down_write(&nilfs->ns_sem);
817 		nilfs_setup_super(sbi);
818 		up_write(&nilfs->ns_sem);
819 	}
820 
821 	down_write(&nilfs->ns_super_sem);
822 	if (!nilfs_test_opt(sbi, SNAPSHOT))
823 		nilfs->ns_current = sbi;
824 	up_write(&nilfs->ns_super_sem);
825 
826 	return 0;
827 
828  failed_segctor:
829 	nilfs_detach_segment_constructor(sbi);
830 
831  failed_checkpoint:
832 	nilfs_detach_checkpoint(sbi);
833 
834  failed_sbi:
835 	put_nilfs(nilfs);
836 	sb->s_fs_info = NULL;
837 	nilfs_put_sbinfo(sbi);
838 	return err;
839 }
840 
841 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
842 {
843 	struct nilfs_sb_info *sbi = NILFS_SB(sb);
844 	struct nilfs_super_block *sbp;
845 	struct the_nilfs *nilfs = sbi->s_nilfs;
846 	unsigned long old_sb_flags;
847 	struct nilfs_mount_options old_opts;
848 	int err;
849 
850 	lock_kernel();
851 
852 	down_write(&nilfs->ns_super_sem);
853 	old_sb_flags = sb->s_flags;
854 	old_opts.mount_opt = sbi->s_mount_opt;
855 	old_opts.snapshot_cno = sbi->s_snapshot_cno;
856 
857 	if (!parse_options(data, sb)) {
858 		err = -EINVAL;
859 		goto restore_opts;
860 	}
861 	sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
862 
863 	if ((*flags & MS_RDONLY) &&
864 	    sbi->s_snapshot_cno != old_opts.snapshot_cno) {
865 		printk(KERN_WARNING "NILFS (device %s): couldn't "
866 		       "remount to a different snapshot. \n",
867 		       sb->s_id);
868 		err = -EINVAL;
869 		goto restore_opts;
870 	}
871 
872 	if (!nilfs_valid_fs(nilfs)) {
873 		printk(KERN_WARNING "NILFS (device %s): couldn't "
874 		       "remount because the filesystem is in an "
875 		       "incomplete recovery state.\n", sb->s_id);
876 		err = -EINVAL;
877 		goto restore_opts;
878 	}
879 
880 	if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
881 		goto out;
882 	if (*flags & MS_RDONLY) {
883 		/* Shutting down the segment constructor */
884 		nilfs_detach_segment_constructor(sbi);
885 		sb->s_flags |= MS_RDONLY;
886 
887 		sbi->s_snapshot_cno = nilfs_last_cno(nilfs);
888 		/* nilfs_set_opt(sbi, SNAPSHOT); */
889 
890 		/*
891 		 * Remounting a valid RW partition RDONLY, so set
892 		 * the RDONLY flag and then mark the partition as valid again.
893 		 */
894 		down_write(&nilfs->ns_sem);
895 		sbp = nilfs->ns_sbp[0];
896 		if (!(sbp->s_state & le16_to_cpu(NILFS_VALID_FS)) &&
897 		    (nilfs->ns_mount_state & NILFS_VALID_FS))
898 			sbp->s_state = cpu_to_le16(nilfs->ns_mount_state);
899 		sbp->s_mtime = cpu_to_le64(get_seconds());
900 		nilfs_commit_super(sbi, 1);
901 		up_write(&nilfs->ns_sem);
902 	} else {
903 		/*
904 		 * Mounting a RDONLY partition read-write, so reread and
905 		 * store the current valid flag.  (It may have been changed
906 		 * by fsck since we originally mounted the partition.)
907 		 */
908 		if (nilfs->ns_current && nilfs->ns_current != sbi) {
909 			printk(KERN_WARNING "NILFS (device %s): couldn't "
910 			       "remount because an RW-mount exists.\n",
911 			       sb->s_id);
912 			err = -EBUSY;
913 			goto restore_opts;
914 		}
915 		if (sbi->s_snapshot_cno != nilfs_last_cno(nilfs)) {
916 			printk(KERN_WARNING "NILFS (device %s): couldn't "
917 			       "remount because the current RO-mount is not "
918 			       "the latest one.\n",
919 			       sb->s_id);
920 			err = -EINVAL;
921 			goto restore_opts;
922 		}
923 		sb->s_flags &= ~MS_RDONLY;
924 		nilfs_clear_opt(sbi, SNAPSHOT);
925 		sbi->s_snapshot_cno = 0;
926 
927 		err = nilfs_attach_segment_constructor(sbi);
928 		if (err)
929 			goto restore_opts;
930 
931 		down_write(&nilfs->ns_sem);
932 		nilfs_setup_super(sbi);
933 		up_write(&nilfs->ns_sem);
934 
935 		nilfs->ns_current = sbi;
936 	}
937  out:
938 	up_write(&nilfs->ns_super_sem);
939 	unlock_kernel();
940 	return 0;
941 
942  restore_opts:
943 	sb->s_flags = old_sb_flags;
944 	sbi->s_mount_opt = old_opts.mount_opt;
945 	sbi->s_snapshot_cno = old_opts.snapshot_cno;
946 	up_write(&nilfs->ns_super_sem);
947 	unlock_kernel();
948 	return err;
949 }
950 
951 struct nilfs_super_data {
952 	struct block_device *bdev;
953 	struct nilfs_sb_info *sbi;
954 	__u64 cno;
955 	int flags;
956 };
957 
958 /**
959  * nilfs_identify - pre-read mount options needed to identify mount instance
960  * @data: mount options
961  * @sd: nilfs_super_data
962  */
963 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
964 {
965 	char *p, *options = data;
966 	substring_t args[MAX_OPT_ARGS];
967 	int option, token;
968 	int ret = 0;
969 
970 	do {
971 		p = strsep(&options, ",");
972 		if (p != NULL && *p) {
973 			token = match_token(p, tokens, args);
974 			if (token == Opt_snapshot) {
975 				if (!(sd->flags & MS_RDONLY))
976 					ret++;
977 				else {
978 					ret = match_int(&args[0], &option);
979 					if (!ret) {
980 						if (option > 0)
981 							sd->cno = option;
982 						else
983 							ret++;
984 					}
985 				}
986 			}
987 			if (ret)
988 				printk(KERN_ERR
989 				       "NILFS: invalid mount option: %s\n", p);
990 		}
991 		if (!options)
992 			break;
993 		BUG_ON(options == data);
994 		*(options - 1) = ',';
995 	} while (!ret);
996 	return ret;
997 }
998 
999 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1000 {
1001 	struct nilfs_super_data *sd = data;
1002 
1003 	s->s_bdev = sd->bdev;
1004 	s->s_dev = s->s_bdev->bd_dev;
1005 	return 0;
1006 }
1007 
1008 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1009 {
1010 	struct nilfs_super_data *sd = data;
1011 
1012 	return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1013 }
1014 
1015 static int
1016 nilfs_get_sb(struct file_system_type *fs_type, int flags,
1017 	     const char *dev_name, void *data, struct vfsmount *mnt)
1018 {
1019 	struct nilfs_super_data sd;
1020 	struct super_block *s;
1021 	struct the_nilfs *nilfs;
1022 	int err, need_to_close = 1;
1023 
1024 	sd.bdev = open_bdev_exclusive(dev_name, flags, fs_type);
1025 	if (IS_ERR(sd.bdev))
1026 		return PTR_ERR(sd.bdev);
1027 
1028 	/*
1029 	 * To get mount instance using sget() vfs-routine, NILFS needs
1030 	 * much more information than normal filesystems to identify mount
1031 	 * instance.  For snapshot mounts, not only a mount type (ro-mount
1032 	 * or rw-mount) but also a checkpoint number is required.
1033 	 */
1034 	sd.cno = 0;
1035 	sd.flags = flags;
1036 	if (nilfs_identify((char *)data, &sd)) {
1037 		err = -EINVAL;
1038 		goto failed;
1039 	}
1040 
1041 	nilfs = find_or_create_nilfs(sd.bdev);
1042 	if (!nilfs) {
1043 		err = -ENOMEM;
1044 		goto failed;
1045 	}
1046 
1047 	mutex_lock(&nilfs->ns_mount_mutex);
1048 
1049 	if (!sd.cno) {
1050 		/*
1051 		 * Check if an exclusive mount exists or not.
1052 		 * Snapshot mounts coexist with a current mount
1053 		 * (i.e. rw-mount or ro-mount), whereas rw-mount and
1054 		 * ro-mount are mutually exclusive.
1055 		 */
1056 		down_read(&nilfs->ns_super_sem);
1057 		if (nilfs->ns_current &&
1058 		    ((nilfs->ns_current->s_super->s_flags ^ flags)
1059 		     & MS_RDONLY)) {
1060 			up_read(&nilfs->ns_super_sem);
1061 			err = -EBUSY;
1062 			goto failed_unlock;
1063 		}
1064 		up_read(&nilfs->ns_super_sem);
1065 	}
1066 
1067 	/*
1068 	 * Find existing nilfs_sb_info struct
1069 	 */
1070 	sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1071 
1072 	/*
1073 	 * Get super block instance holding the nilfs_sb_info struct.
1074 	 * A new instance is allocated if no existing mount is present or
1075 	 * existing instance has been unmounted.
1076 	 */
1077 	s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd);
1078 	if (sd.sbi)
1079 		nilfs_put_sbinfo(sd.sbi);
1080 
1081 	if (IS_ERR(s)) {
1082 		err = PTR_ERR(s);
1083 		goto failed_unlock;
1084 	}
1085 
1086 	if (!s->s_root) {
1087 		char b[BDEVNAME_SIZE];
1088 
1089 		/* New superblock instance created */
1090 		s->s_flags = flags;
1091 		strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1092 		sb_set_blocksize(s, block_size(sd.bdev));
1093 
1094 		err = nilfs_fill_super(s, data, flags & MS_VERBOSE, nilfs);
1095 		if (err)
1096 			goto cancel_new;
1097 
1098 		s->s_flags |= MS_ACTIVE;
1099 		need_to_close = 0;
1100 	}
1101 
1102 	mutex_unlock(&nilfs->ns_mount_mutex);
1103 	put_nilfs(nilfs);
1104 	if (need_to_close)
1105 		close_bdev_exclusive(sd.bdev, flags);
1106 	simple_set_mnt(mnt, s);
1107 	return 0;
1108 
1109  failed_unlock:
1110 	mutex_unlock(&nilfs->ns_mount_mutex);
1111 	put_nilfs(nilfs);
1112  failed:
1113 	close_bdev_exclusive(sd.bdev, flags);
1114 
1115 	return err;
1116 
1117  cancel_new:
1118 	/* Abandoning the newly allocated superblock */
1119 	mutex_unlock(&nilfs->ns_mount_mutex);
1120 	put_nilfs(nilfs);
1121 	deactivate_locked_super(s);
1122 	/*
1123 	 * deactivate_super() invokes close_bdev_exclusive().
1124 	 * We must finish all post-cleaning before this call;
1125 	 * put_nilfs() needs the block device.
1126 	 */
1127 	return err;
1128 }
1129 
1130 struct file_system_type nilfs_fs_type = {
1131 	.owner    = THIS_MODULE,
1132 	.name     = "nilfs2",
1133 	.get_sb   = nilfs_get_sb,
1134 	.kill_sb  = kill_block_super,
1135 	.fs_flags = FS_REQUIRES_DEV,
1136 };
1137 
1138 static int __init init_nilfs_fs(void)
1139 {
1140 	int err;
1141 
1142 	err = nilfs_init_inode_cache();
1143 	if (err)
1144 		goto failed;
1145 
1146 	err = nilfs_init_transaction_cache();
1147 	if (err)
1148 		goto failed_inode_cache;
1149 
1150 	err = nilfs_init_segbuf_cache();
1151 	if (err)
1152 		goto failed_transaction_cache;
1153 
1154 	err = nilfs_btree_path_cache_init();
1155 	if (err)
1156 		goto failed_segbuf_cache;
1157 
1158 	err = register_filesystem(&nilfs_fs_type);
1159 	if (err)
1160 		goto failed_btree_path_cache;
1161 
1162 	return 0;
1163 
1164  failed_btree_path_cache:
1165 	nilfs_btree_path_cache_destroy();
1166 
1167  failed_segbuf_cache:
1168 	nilfs_destroy_segbuf_cache();
1169 
1170  failed_transaction_cache:
1171 	nilfs_destroy_transaction_cache();
1172 
1173  failed_inode_cache:
1174 	nilfs_destroy_inode_cache();
1175 
1176  failed:
1177 	return err;
1178 }
1179 
1180 static void __exit exit_nilfs_fs(void)
1181 {
1182 	nilfs_destroy_segbuf_cache();
1183 	nilfs_destroy_transaction_cache();
1184 	nilfs_destroy_inode_cache();
1185 	nilfs_btree_path_cache_destroy();
1186 	unregister_filesystem(&nilfs_fs_type);
1187 }
1188 
1189 module_init(init_nilfs_fs)
1190 module_exit(exit_nilfs_fs)
1191