xref: /linux/fs/nilfs2/super.c (revision c4ee0af3fa0dc65f690fc908f02b8355f9576ea0)
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 	atomic64_set(&root->inodes_count,
558 			le64_to_cpu(raw_cp->cp_inodes_count));
559 	atomic64_set(&root->blocks_count,
560 			le64_to_cpu(raw_cp->cp_blocks_count));
561 
562 	nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
563 
564  reuse:
565 	*rootp = root;
566 	return 0;
567 
568  failed_bh:
569 	nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
570  failed:
571 	nilfs_put_root(root);
572 
573 	return err;
574 }
575 
576 static int nilfs_freeze(struct super_block *sb)
577 {
578 	struct the_nilfs *nilfs = sb->s_fs_info;
579 	int err;
580 
581 	if (sb->s_flags & MS_RDONLY)
582 		return 0;
583 
584 	/* Mark super block clean */
585 	down_write(&nilfs->ns_sem);
586 	err = nilfs_cleanup_super(sb);
587 	up_write(&nilfs->ns_sem);
588 	return err;
589 }
590 
591 static int nilfs_unfreeze(struct super_block *sb)
592 {
593 	struct the_nilfs *nilfs = sb->s_fs_info;
594 
595 	if (sb->s_flags & MS_RDONLY)
596 		return 0;
597 
598 	down_write(&nilfs->ns_sem);
599 	nilfs_setup_super(sb, false);
600 	up_write(&nilfs->ns_sem);
601 	return 0;
602 }
603 
604 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
605 {
606 	struct super_block *sb = dentry->d_sb;
607 	struct nilfs_root *root = NILFS_I(dentry->d_inode)->i_root;
608 	struct the_nilfs *nilfs = root->nilfs;
609 	u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
610 	unsigned long long blocks;
611 	unsigned long overhead;
612 	unsigned long nrsvblocks;
613 	sector_t nfreeblocks;
614 	u64 nmaxinodes, nfreeinodes;
615 	int err;
616 
617 	/*
618 	 * Compute all of the segment blocks
619 	 *
620 	 * The blocks before first segment and after last segment
621 	 * are excluded.
622 	 */
623 	blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
624 		- nilfs->ns_first_data_block;
625 	nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
626 
627 	/*
628 	 * Compute the overhead
629 	 *
630 	 * When distributing meta data blocks outside segment structure,
631 	 * We must count them as the overhead.
632 	 */
633 	overhead = 0;
634 
635 	err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
636 	if (unlikely(err))
637 		return err;
638 
639 	err = nilfs_ifile_count_free_inodes(root->ifile,
640 					    &nmaxinodes, &nfreeinodes);
641 	if (unlikely(err)) {
642 		printk(KERN_WARNING
643 			"NILFS warning: fail to count free inodes: err %d.\n",
644 			err);
645 		if (err == -ERANGE) {
646 			/*
647 			 * If nilfs_palloc_count_max_entries() returns
648 			 * -ERANGE error code then we simply treat
649 			 * curent inodes count as maximum possible and
650 			 * zero as free inodes value.
651 			 */
652 			nmaxinodes = atomic64_read(&root->inodes_count);
653 			nfreeinodes = 0;
654 			err = 0;
655 		} else
656 			return err;
657 	}
658 
659 	buf->f_type = NILFS_SUPER_MAGIC;
660 	buf->f_bsize = sb->s_blocksize;
661 	buf->f_blocks = blocks - overhead;
662 	buf->f_bfree = nfreeblocks;
663 	buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
664 		(buf->f_bfree - nrsvblocks) : 0;
665 	buf->f_files = nmaxinodes;
666 	buf->f_ffree = nfreeinodes;
667 	buf->f_namelen = NILFS_NAME_LEN;
668 	buf->f_fsid.val[0] = (u32)id;
669 	buf->f_fsid.val[1] = (u32)(id >> 32);
670 
671 	return 0;
672 }
673 
674 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
675 {
676 	struct super_block *sb = dentry->d_sb;
677 	struct the_nilfs *nilfs = sb->s_fs_info;
678 	struct nilfs_root *root = NILFS_I(dentry->d_inode)->i_root;
679 
680 	if (!nilfs_test_opt(nilfs, BARRIER))
681 		seq_puts(seq, ",nobarrier");
682 	if (root->cno != NILFS_CPTREE_CURRENT_CNO)
683 		seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
684 	if (nilfs_test_opt(nilfs, ERRORS_PANIC))
685 		seq_puts(seq, ",errors=panic");
686 	if (nilfs_test_opt(nilfs, ERRORS_CONT))
687 		seq_puts(seq, ",errors=continue");
688 	if (nilfs_test_opt(nilfs, STRICT_ORDER))
689 		seq_puts(seq, ",order=strict");
690 	if (nilfs_test_opt(nilfs, NORECOVERY))
691 		seq_puts(seq, ",norecovery");
692 	if (nilfs_test_opt(nilfs, DISCARD))
693 		seq_puts(seq, ",discard");
694 
695 	return 0;
696 }
697 
698 static const struct super_operations nilfs_sops = {
699 	.alloc_inode    = nilfs_alloc_inode,
700 	.destroy_inode  = nilfs_destroy_inode,
701 	.dirty_inode    = nilfs_dirty_inode,
702 	.evict_inode    = nilfs_evict_inode,
703 	.put_super      = nilfs_put_super,
704 	.sync_fs        = nilfs_sync_fs,
705 	.freeze_fs	= nilfs_freeze,
706 	.unfreeze_fs	= nilfs_unfreeze,
707 	.statfs         = nilfs_statfs,
708 	.remount_fs     = nilfs_remount,
709 	.show_options = nilfs_show_options
710 };
711 
712 enum {
713 	Opt_err_cont, Opt_err_panic, Opt_err_ro,
714 	Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
715 	Opt_discard, Opt_nodiscard, Opt_err,
716 };
717 
718 static match_table_t tokens = {
719 	{Opt_err_cont, "errors=continue"},
720 	{Opt_err_panic, "errors=panic"},
721 	{Opt_err_ro, "errors=remount-ro"},
722 	{Opt_barrier, "barrier"},
723 	{Opt_nobarrier, "nobarrier"},
724 	{Opt_snapshot, "cp=%u"},
725 	{Opt_order, "order=%s"},
726 	{Opt_norecovery, "norecovery"},
727 	{Opt_discard, "discard"},
728 	{Opt_nodiscard, "nodiscard"},
729 	{Opt_err, NULL}
730 };
731 
732 static int parse_options(char *options, struct super_block *sb, int is_remount)
733 {
734 	struct the_nilfs *nilfs = sb->s_fs_info;
735 	char *p;
736 	substring_t args[MAX_OPT_ARGS];
737 
738 	if (!options)
739 		return 1;
740 
741 	while ((p = strsep(&options, ",")) != NULL) {
742 		int token;
743 		if (!*p)
744 			continue;
745 
746 		token = match_token(p, tokens, args);
747 		switch (token) {
748 		case Opt_barrier:
749 			nilfs_set_opt(nilfs, BARRIER);
750 			break;
751 		case Opt_nobarrier:
752 			nilfs_clear_opt(nilfs, BARRIER);
753 			break;
754 		case Opt_order:
755 			if (strcmp(args[0].from, "relaxed") == 0)
756 				/* Ordered data semantics */
757 				nilfs_clear_opt(nilfs, STRICT_ORDER);
758 			else if (strcmp(args[0].from, "strict") == 0)
759 				/* Strict in-order semantics */
760 				nilfs_set_opt(nilfs, STRICT_ORDER);
761 			else
762 				return 0;
763 			break;
764 		case Opt_err_panic:
765 			nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
766 			break;
767 		case Opt_err_ro:
768 			nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
769 			break;
770 		case Opt_err_cont:
771 			nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
772 			break;
773 		case Opt_snapshot:
774 			if (is_remount) {
775 				printk(KERN_ERR
776 				       "NILFS: \"%s\" option is invalid "
777 				       "for remount.\n", p);
778 				return 0;
779 			}
780 			break;
781 		case Opt_norecovery:
782 			nilfs_set_opt(nilfs, NORECOVERY);
783 			break;
784 		case Opt_discard:
785 			nilfs_set_opt(nilfs, DISCARD);
786 			break;
787 		case Opt_nodiscard:
788 			nilfs_clear_opt(nilfs, DISCARD);
789 			break;
790 		default:
791 			printk(KERN_ERR
792 			       "NILFS: Unrecognized mount option \"%s\"\n", p);
793 			return 0;
794 		}
795 	}
796 	return 1;
797 }
798 
799 static inline void
800 nilfs_set_default_options(struct super_block *sb,
801 			  struct nilfs_super_block *sbp)
802 {
803 	struct the_nilfs *nilfs = sb->s_fs_info;
804 
805 	nilfs->ns_mount_opt =
806 		NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
807 }
808 
809 static int nilfs_setup_super(struct super_block *sb, int is_mount)
810 {
811 	struct the_nilfs *nilfs = sb->s_fs_info;
812 	struct nilfs_super_block **sbp;
813 	int max_mnt_count;
814 	int mnt_count;
815 
816 	/* nilfs->ns_sem must be locked by the caller. */
817 	sbp = nilfs_prepare_super(sb, 0);
818 	if (!sbp)
819 		return -EIO;
820 
821 	if (!is_mount)
822 		goto skip_mount_setup;
823 
824 	max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
825 	mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
826 
827 	if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
828 		printk(KERN_WARNING
829 		       "NILFS warning: mounting fs with errors\n");
830 #if 0
831 	} else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
832 		printk(KERN_WARNING
833 		       "NILFS warning: maximal mount count reached\n");
834 #endif
835 	}
836 	if (!max_mnt_count)
837 		sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
838 
839 	sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
840 	sbp[0]->s_mtime = cpu_to_le64(get_seconds());
841 
842 skip_mount_setup:
843 	sbp[0]->s_state =
844 		cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
845 	/* synchronize sbp[1] with sbp[0] */
846 	if (sbp[1])
847 		memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
848 	return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
849 }
850 
851 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
852 						 u64 pos, int blocksize,
853 						 struct buffer_head **pbh)
854 {
855 	unsigned long long sb_index = pos;
856 	unsigned long offset;
857 
858 	offset = do_div(sb_index, blocksize);
859 	*pbh = sb_bread(sb, sb_index);
860 	if (!*pbh)
861 		return NULL;
862 	return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
863 }
864 
865 int nilfs_store_magic_and_option(struct super_block *sb,
866 				 struct nilfs_super_block *sbp,
867 				 char *data)
868 {
869 	struct the_nilfs *nilfs = sb->s_fs_info;
870 
871 	sb->s_magic = le16_to_cpu(sbp->s_magic);
872 
873 	/* FS independent flags */
874 #ifdef NILFS_ATIME_DISABLE
875 	sb->s_flags |= MS_NOATIME;
876 #endif
877 
878 	nilfs_set_default_options(sb, sbp);
879 
880 	nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
881 	nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
882 	nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
883 	nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
884 
885 	return !parse_options(data, sb, 0) ? -EINVAL : 0 ;
886 }
887 
888 int nilfs_check_feature_compatibility(struct super_block *sb,
889 				      struct nilfs_super_block *sbp)
890 {
891 	__u64 features;
892 
893 	features = le64_to_cpu(sbp->s_feature_incompat) &
894 		~NILFS_FEATURE_INCOMPAT_SUPP;
895 	if (features) {
896 		printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
897 		       "optional features (%llx)\n",
898 		       (unsigned long long)features);
899 		return -EINVAL;
900 	}
901 	features = le64_to_cpu(sbp->s_feature_compat_ro) &
902 		~NILFS_FEATURE_COMPAT_RO_SUPP;
903 	if (!(sb->s_flags & MS_RDONLY) && features) {
904 		printk(KERN_ERR "NILFS: couldn't mount RDWR because of "
905 		       "unsupported optional features (%llx)\n",
906 		       (unsigned long long)features);
907 		return -EINVAL;
908 	}
909 	return 0;
910 }
911 
912 static int nilfs_get_root_dentry(struct super_block *sb,
913 				 struct nilfs_root *root,
914 				 struct dentry **root_dentry)
915 {
916 	struct inode *inode;
917 	struct dentry *dentry;
918 	int ret = 0;
919 
920 	inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
921 	if (IS_ERR(inode)) {
922 		printk(KERN_ERR "NILFS: get root inode failed\n");
923 		ret = PTR_ERR(inode);
924 		goto out;
925 	}
926 	if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
927 		iput(inode);
928 		printk(KERN_ERR "NILFS: corrupt root inode.\n");
929 		ret = -EINVAL;
930 		goto out;
931 	}
932 
933 	if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
934 		dentry = d_find_alias(inode);
935 		if (!dentry) {
936 			dentry = d_make_root(inode);
937 			if (!dentry) {
938 				ret = -ENOMEM;
939 				goto failed_dentry;
940 			}
941 		} else {
942 			iput(inode);
943 		}
944 	} else {
945 		dentry = d_obtain_alias(inode);
946 		if (IS_ERR(dentry)) {
947 			ret = PTR_ERR(dentry);
948 			goto failed_dentry;
949 		}
950 	}
951 	*root_dentry = dentry;
952  out:
953 	return ret;
954 
955  failed_dentry:
956 	printk(KERN_ERR "NILFS: get root dentry failed\n");
957 	goto out;
958 }
959 
960 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
961 				 struct dentry **root_dentry)
962 {
963 	struct the_nilfs *nilfs = s->s_fs_info;
964 	struct nilfs_root *root;
965 	int ret;
966 
967 	mutex_lock(&nilfs->ns_snapshot_mount_mutex);
968 
969 	down_read(&nilfs->ns_segctor_sem);
970 	ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
971 	up_read(&nilfs->ns_segctor_sem);
972 	if (ret < 0) {
973 		ret = (ret == -ENOENT) ? -EINVAL : ret;
974 		goto out;
975 	} else if (!ret) {
976 		printk(KERN_ERR "NILFS: The specified checkpoint is "
977 		       "not a snapshot (checkpoint number=%llu).\n",
978 		       (unsigned long long)cno);
979 		ret = -EINVAL;
980 		goto out;
981 	}
982 
983 	ret = nilfs_attach_checkpoint(s, cno, false, &root);
984 	if (ret) {
985 		printk(KERN_ERR "NILFS: error loading snapshot "
986 		       "(checkpoint number=%llu).\n",
987 	       (unsigned long long)cno);
988 		goto out;
989 	}
990 	ret = nilfs_get_root_dentry(s, root, root_dentry);
991 	nilfs_put_root(root);
992  out:
993 	mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
994 	return ret;
995 }
996 
997 /**
998  * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
999  * @root_dentry: root dentry of the tree to be shrunk
1000  *
1001  * This function returns true if the tree was in-use.
1002  */
1003 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
1004 {
1005 	shrink_dcache_parent(root_dentry);
1006 	return d_count(root_dentry) > 1;
1007 }
1008 
1009 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1010 {
1011 	struct the_nilfs *nilfs = sb->s_fs_info;
1012 	struct nilfs_root *root;
1013 	struct inode *inode;
1014 	struct dentry *dentry;
1015 	int ret;
1016 
1017 	if (cno < 0 || cno > nilfs->ns_cno)
1018 		return false;
1019 
1020 	if (cno >= nilfs_last_cno(nilfs))
1021 		return true;	/* protect recent checkpoints */
1022 
1023 	ret = false;
1024 	root = nilfs_lookup_root(nilfs, cno);
1025 	if (root) {
1026 		inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1027 		if (inode) {
1028 			dentry = d_find_alias(inode);
1029 			if (dentry) {
1030 				ret = nilfs_tree_is_busy(dentry);
1031 				dput(dentry);
1032 			}
1033 			iput(inode);
1034 		}
1035 		nilfs_put_root(root);
1036 	}
1037 	return ret;
1038 }
1039 
1040 /**
1041  * nilfs_fill_super() - initialize a super block instance
1042  * @sb: super_block
1043  * @data: mount options
1044  * @silent: silent mode flag
1045  *
1046  * This function is called exclusively by nilfs->ns_mount_mutex.
1047  * So, the recovery process is protected from other simultaneous mounts.
1048  */
1049 static int
1050 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1051 {
1052 	struct the_nilfs *nilfs;
1053 	struct nilfs_root *fsroot;
1054 	struct backing_dev_info *bdi;
1055 	__u64 cno;
1056 	int err;
1057 
1058 	nilfs = alloc_nilfs(sb->s_bdev);
1059 	if (!nilfs)
1060 		return -ENOMEM;
1061 
1062 	sb->s_fs_info = nilfs;
1063 
1064 	err = init_nilfs(nilfs, sb, (char *)data);
1065 	if (err)
1066 		goto failed_nilfs;
1067 
1068 	sb->s_op = &nilfs_sops;
1069 	sb->s_export_op = &nilfs_export_ops;
1070 	sb->s_root = NULL;
1071 	sb->s_time_gran = 1;
1072 	sb->s_max_links = NILFS_LINK_MAX;
1073 
1074 	bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
1075 	sb->s_bdi = bdi ? : &default_backing_dev_info;
1076 
1077 	err = load_nilfs(nilfs, sb);
1078 	if (err)
1079 		goto failed_nilfs;
1080 
1081 	cno = nilfs_last_cno(nilfs);
1082 	err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1083 	if (err) {
1084 		printk(KERN_ERR "NILFS: error loading last checkpoint "
1085 		       "(checkpoint number=%llu).\n", (unsigned long long)cno);
1086 		goto failed_unload;
1087 	}
1088 
1089 	if (!(sb->s_flags & MS_RDONLY)) {
1090 		err = nilfs_attach_log_writer(sb, fsroot);
1091 		if (err)
1092 			goto failed_checkpoint;
1093 	}
1094 
1095 	err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1096 	if (err)
1097 		goto failed_segctor;
1098 
1099 	nilfs_put_root(fsroot);
1100 
1101 	if (!(sb->s_flags & MS_RDONLY)) {
1102 		down_write(&nilfs->ns_sem);
1103 		nilfs_setup_super(sb, true);
1104 		up_write(&nilfs->ns_sem);
1105 	}
1106 
1107 	return 0;
1108 
1109  failed_segctor:
1110 	nilfs_detach_log_writer(sb);
1111 
1112  failed_checkpoint:
1113 	nilfs_put_root(fsroot);
1114 
1115  failed_unload:
1116 	iput(nilfs->ns_sufile);
1117 	iput(nilfs->ns_cpfile);
1118 	iput(nilfs->ns_dat);
1119 
1120  failed_nilfs:
1121 	destroy_nilfs(nilfs);
1122 	return err;
1123 }
1124 
1125 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1126 {
1127 	struct the_nilfs *nilfs = sb->s_fs_info;
1128 	unsigned long old_sb_flags;
1129 	unsigned long old_mount_opt;
1130 	int err;
1131 
1132 	old_sb_flags = sb->s_flags;
1133 	old_mount_opt = nilfs->ns_mount_opt;
1134 
1135 	if (!parse_options(data, sb, 1)) {
1136 		err = -EINVAL;
1137 		goto restore_opts;
1138 	}
1139 	sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
1140 
1141 	err = -EINVAL;
1142 
1143 	if (!nilfs_valid_fs(nilfs)) {
1144 		printk(KERN_WARNING "NILFS (device %s): couldn't "
1145 		       "remount because the filesystem is in an "
1146 		       "incomplete recovery state.\n", sb->s_id);
1147 		goto restore_opts;
1148 	}
1149 
1150 	if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1151 		goto out;
1152 	if (*flags & MS_RDONLY) {
1153 		/* Shutting down log writer */
1154 		nilfs_detach_log_writer(sb);
1155 		sb->s_flags |= MS_RDONLY;
1156 
1157 		/*
1158 		 * Remounting a valid RW partition RDONLY, so set
1159 		 * the RDONLY flag and then mark the partition as valid again.
1160 		 */
1161 		down_write(&nilfs->ns_sem);
1162 		nilfs_cleanup_super(sb);
1163 		up_write(&nilfs->ns_sem);
1164 	} else {
1165 		__u64 features;
1166 		struct nilfs_root *root;
1167 
1168 		/*
1169 		 * Mounting a RDONLY partition read-write, so reread and
1170 		 * store the current valid flag.  (It may have been changed
1171 		 * by fsck since we originally mounted the partition.)
1172 		 */
1173 		down_read(&nilfs->ns_sem);
1174 		features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1175 			~NILFS_FEATURE_COMPAT_RO_SUPP;
1176 		up_read(&nilfs->ns_sem);
1177 		if (features) {
1178 			printk(KERN_WARNING "NILFS (device %s): couldn't "
1179 			       "remount RDWR because of unsupported optional "
1180 			       "features (%llx)\n",
1181 			       sb->s_id, (unsigned long long)features);
1182 			err = -EROFS;
1183 			goto restore_opts;
1184 		}
1185 
1186 		sb->s_flags &= ~MS_RDONLY;
1187 
1188 		root = NILFS_I(sb->s_root->d_inode)->i_root;
1189 		err = nilfs_attach_log_writer(sb, root);
1190 		if (err)
1191 			goto restore_opts;
1192 
1193 		down_write(&nilfs->ns_sem);
1194 		nilfs_setup_super(sb, true);
1195 		up_write(&nilfs->ns_sem);
1196 	}
1197  out:
1198 	return 0;
1199 
1200  restore_opts:
1201 	sb->s_flags = old_sb_flags;
1202 	nilfs->ns_mount_opt = old_mount_opt;
1203 	return err;
1204 }
1205 
1206 struct nilfs_super_data {
1207 	struct block_device *bdev;
1208 	__u64 cno;
1209 	int flags;
1210 };
1211 
1212 /**
1213  * nilfs_identify - pre-read mount options needed to identify mount instance
1214  * @data: mount options
1215  * @sd: nilfs_super_data
1216  */
1217 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1218 {
1219 	char *p, *options = data;
1220 	substring_t args[MAX_OPT_ARGS];
1221 	int token;
1222 	int ret = 0;
1223 
1224 	do {
1225 		p = strsep(&options, ",");
1226 		if (p != NULL && *p) {
1227 			token = match_token(p, tokens, args);
1228 			if (token == Opt_snapshot) {
1229 				if (!(sd->flags & MS_RDONLY)) {
1230 					ret++;
1231 				} else {
1232 					sd->cno = simple_strtoull(args[0].from,
1233 								  NULL, 0);
1234 					/*
1235 					 * No need to see the end pointer;
1236 					 * match_token() has done syntax
1237 					 * checking.
1238 					 */
1239 					if (sd->cno == 0)
1240 						ret++;
1241 				}
1242 			}
1243 			if (ret)
1244 				printk(KERN_ERR
1245 				       "NILFS: invalid mount option: %s\n", p);
1246 		}
1247 		if (!options)
1248 			break;
1249 		BUG_ON(options == data);
1250 		*(options - 1) = ',';
1251 	} while (!ret);
1252 	return ret;
1253 }
1254 
1255 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1256 {
1257 	s->s_bdev = data;
1258 	s->s_dev = s->s_bdev->bd_dev;
1259 	return 0;
1260 }
1261 
1262 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1263 {
1264 	return (void *)s->s_bdev == data;
1265 }
1266 
1267 static struct dentry *
1268 nilfs_mount(struct file_system_type *fs_type, int flags,
1269 	     const char *dev_name, void *data)
1270 {
1271 	struct nilfs_super_data sd;
1272 	struct super_block *s;
1273 	fmode_t mode = FMODE_READ | FMODE_EXCL;
1274 	struct dentry *root_dentry;
1275 	int err, s_new = false;
1276 
1277 	if (!(flags & MS_RDONLY))
1278 		mode |= FMODE_WRITE;
1279 
1280 	sd.bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1281 	if (IS_ERR(sd.bdev))
1282 		return ERR_CAST(sd.bdev);
1283 
1284 	sd.cno = 0;
1285 	sd.flags = flags;
1286 	if (nilfs_identify((char *)data, &sd)) {
1287 		err = -EINVAL;
1288 		goto failed;
1289 	}
1290 
1291 	/*
1292 	 * once the super is inserted into the list by sget, s_umount
1293 	 * will protect the lockfs code from trying to start a snapshot
1294 	 * while we are mounting
1295 	 */
1296 	mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1297 	if (sd.bdev->bd_fsfreeze_count > 0) {
1298 		mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1299 		err = -EBUSY;
1300 		goto failed;
1301 	}
1302 	s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1303 		 sd.bdev);
1304 	mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1305 	if (IS_ERR(s)) {
1306 		err = PTR_ERR(s);
1307 		goto failed;
1308 	}
1309 
1310 	if (!s->s_root) {
1311 		char b[BDEVNAME_SIZE];
1312 
1313 		s_new = true;
1314 
1315 		/* New superblock instance created */
1316 		s->s_mode = mode;
1317 		strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1318 		sb_set_blocksize(s, block_size(sd.bdev));
1319 
1320 		err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1321 		if (err)
1322 			goto failed_super;
1323 
1324 		s->s_flags |= MS_ACTIVE;
1325 	} else if (!sd.cno) {
1326 		if (nilfs_tree_is_busy(s->s_root)) {
1327 			if ((flags ^ s->s_flags) & MS_RDONLY) {
1328 				printk(KERN_ERR "NILFS: the device already "
1329 				       "has a %s mount.\n",
1330 				       (s->s_flags & MS_RDONLY) ?
1331 				       "read-only" : "read/write");
1332 				err = -EBUSY;
1333 				goto failed_super;
1334 			}
1335 		} else {
1336 			/*
1337 			 * Try remount to setup mount states if the current
1338 			 * tree is not mounted and only snapshots use this sb.
1339 			 */
1340 			err = nilfs_remount(s, &flags, data);
1341 			if (err)
1342 				goto failed_super;
1343 		}
1344 	}
1345 
1346 	if (sd.cno) {
1347 		err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1348 		if (err)
1349 			goto failed_super;
1350 	} else {
1351 		root_dentry = dget(s->s_root);
1352 	}
1353 
1354 	if (!s_new)
1355 		blkdev_put(sd.bdev, mode);
1356 
1357 	return root_dentry;
1358 
1359  failed_super:
1360 	deactivate_locked_super(s);
1361 
1362  failed:
1363 	if (!s_new)
1364 		blkdev_put(sd.bdev, mode);
1365 	return ERR_PTR(err);
1366 }
1367 
1368 struct file_system_type nilfs_fs_type = {
1369 	.owner    = THIS_MODULE,
1370 	.name     = "nilfs2",
1371 	.mount    = nilfs_mount,
1372 	.kill_sb  = kill_block_super,
1373 	.fs_flags = FS_REQUIRES_DEV,
1374 };
1375 MODULE_ALIAS_FS("nilfs2");
1376 
1377 static void nilfs_inode_init_once(void *obj)
1378 {
1379 	struct nilfs_inode_info *ii = obj;
1380 
1381 	INIT_LIST_HEAD(&ii->i_dirty);
1382 #ifdef CONFIG_NILFS_XATTR
1383 	init_rwsem(&ii->xattr_sem);
1384 #endif
1385 	address_space_init_once(&ii->i_btnode_cache);
1386 	ii->i_bmap = &ii->i_bmap_data;
1387 	inode_init_once(&ii->vfs_inode);
1388 }
1389 
1390 static void nilfs_segbuf_init_once(void *obj)
1391 {
1392 	memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1393 }
1394 
1395 static void nilfs_destroy_cachep(void)
1396 {
1397 	/*
1398 	 * Make sure all delayed rcu free inodes are flushed before we
1399 	 * destroy cache.
1400 	 */
1401 	rcu_barrier();
1402 
1403 	if (nilfs_inode_cachep)
1404 		kmem_cache_destroy(nilfs_inode_cachep);
1405 	if (nilfs_transaction_cachep)
1406 		kmem_cache_destroy(nilfs_transaction_cachep);
1407 	if (nilfs_segbuf_cachep)
1408 		kmem_cache_destroy(nilfs_segbuf_cachep);
1409 	if (nilfs_btree_path_cache)
1410 		kmem_cache_destroy(nilfs_btree_path_cache);
1411 }
1412 
1413 static int __init nilfs_init_cachep(void)
1414 {
1415 	nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1416 			sizeof(struct nilfs_inode_info), 0,
1417 			SLAB_RECLAIM_ACCOUNT, nilfs_inode_init_once);
1418 	if (!nilfs_inode_cachep)
1419 		goto fail;
1420 
1421 	nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1422 			sizeof(struct nilfs_transaction_info), 0,
1423 			SLAB_RECLAIM_ACCOUNT, NULL);
1424 	if (!nilfs_transaction_cachep)
1425 		goto fail;
1426 
1427 	nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1428 			sizeof(struct nilfs_segment_buffer), 0,
1429 			SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1430 	if (!nilfs_segbuf_cachep)
1431 		goto fail;
1432 
1433 	nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1434 			sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1435 			0, 0, NULL);
1436 	if (!nilfs_btree_path_cache)
1437 		goto fail;
1438 
1439 	return 0;
1440 
1441 fail:
1442 	nilfs_destroy_cachep();
1443 	return -ENOMEM;
1444 }
1445 
1446 static int __init init_nilfs_fs(void)
1447 {
1448 	int err;
1449 
1450 	err = nilfs_init_cachep();
1451 	if (err)
1452 		goto fail;
1453 
1454 	err = register_filesystem(&nilfs_fs_type);
1455 	if (err)
1456 		goto free_cachep;
1457 
1458 	printk(KERN_INFO "NILFS version 2 loaded\n");
1459 	return 0;
1460 
1461 free_cachep:
1462 	nilfs_destroy_cachep();
1463 fail:
1464 	return err;
1465 }
1466 
1467 static void __exit exit_nilfs_fs(void)
1468 {
1469 	nilfs_destroy_cachep();
1470 	unregister_filesystem(&nilfs_fs_type);
1471 }
1472 
1473 module_init(init_nilfs_fs)
1474 module_exit(exit_nilfs_fs)
1475