xref: /openbmc/linux/fs/nilfs2/super.c (revision 2a598d0b)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * NILFS module and super block management.
4  *
5  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
6  *
7  * Written by Ryusuke Konishi.
8  */
9 /*
10  *  linux/fs/ext2/super.c
11  *
12  * Copyright (C) 1992, 1993, 1994, 1995
13  * Remy Card (card@masi.ibp.fr)
14  * Laboratoire MASI - Institut Blaise Pascal
15  * Universite Pierre et Marie Curie (Paris VI)
16  *
17  *  from
18  *
19  *  linux/fs/minix/inode.c
20  *
21  *  Copyright (C) 1991, 1992  Linus Torvalds
22  *
23  *  Big-endian to little-endian byte-swapping/bitmaps by
24  *        David S. Miller (davem@caip.rutgers.edu), 1995
25  */
26 
27 #include <linux/module.h>
28 #include <linux/string.h>
29 #include <linux/slab.h>
30 #include <linux/init.h>
31 #include <linux/blkdev.h>
32 #include <linux/parser.h>
33 #include <linux/crc32.h>
34 #include <linux/vfs.h>
35 #include <linux/writeback.h>
36 #include <linux/seq_file.h>
37 #include <linux/mount.h>
38 #include "nilfs.h"
39 #include "export.h"
40 #include "mdt.h"
41 #include "alloc.h"
42 #include "btree.h"
43 #include "btnode.h"
44 #include "page.h"
45 #include "cpfile.h"
46 #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */
47 #include "ifile.h"
48 #include "dat.h"
49 #include "segment.h"
50 #include "segbuf.h"
51 
52 MODULE_AUTHOR("NTT Corp.");
53 MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
54 		   "(NILFS)");
55 MODULE_LICENSE("GPL");
56 
57 static struct kmem_cache *nilfs_inode_cachep;
58 struct kmem_cache *nilfs_transaction_cachep;
59 struct kmem_cache *nilfs_segbuf_cachep;
60 struct kmem_cache *nilfs_btree_path_cache;
61 
62 static int nilfs_setup_super(struct super_block *sb, int is_mount);
63 static int nilfs_remount(struct super_block *sb, int *flags, char *data);
64 
65 void __nilfs_msg(struct super_block *sb, const char *fmt, ...)
66 {
67 	struct va_format vaf;
68 	va_list args;
69 	int level;
70 
71 	va_start(args, fmt);
72 
73 	level = printk_get_level(fmt);
74 	vaf.fmt = printk_skip_level(fmt);
75 	vaf.va = &args;
76 
77 	if (sb)
78 		printk("%c%cNILFS (%s): %pV\n",
79 		       KERN_SOH_ASCII, level, sb->s_id, &vaf);
80 	else
81 		printk("%c%cNILFS: %pV\n",
82 		       KERN_SOH_ASCII, level, &vaf);
83 
84 	va_end(args);
85 }
86 
87 static void nilfs_set_error(struct super_block *sb)
88 {
89 	struct the_nilfs *nilfs = sb->s_fs_info;
90 	struct nilfs_super_block **sbp;
91 
92 	down_write(&nilfs->ns_sem);
93 	if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) {
94 		nilfs->ns_mount_state |= NILFS_ERROR_FS;
95 		sbp = nilfs_prepare_super(sb, 0);
96 		if (likely(sbp)) {
97 			sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
98 			if (sbp[1])
99 				sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS);
100 			nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
101 		}
102 	}
103 	up_write(&nilfs->ns_sem);
104 }
105 
106 /**
107  * __nilfs_error() - report failure condition on a filesystem
108  *
109  * __nilfs_error() sets an ERROR_FS flag on the superblock as well as
110  * reporting an error message.  This function should be called when
111  * NILFS detects incoherences or defects of meta data on disk.
112  *
113  * This implements the body of nilfs_error() macro.  Normally,
114  * nilfs_error() should be used.  As for sustainable errors such as a
115  * single-shot I/O error, nilfs_err() should be used instead.
116  *
117  * Callers should not add a trailing newline since this will do it.
118  */
119 void __nilfs_error(struct super_block *sb, const char *function,
120 		   const char *fmt, ...)
121 {
122 	struct the_nilfs *nilfs = sb->s_fs_info;
123 	struct va_format vaf;
124 	va_list args;
125 
126 	va_start(args, fmt);
127 
128 	vaf.fmt = fmt;
129 	vaf.va = &args;
130 
131 	printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n",
132 	       sb->s_id, function, &vaf);
133 
134 	va_end(args);
135 
136 	if (!sb_rdonly(sb)) {
137 		nilfs_set_error(sb);
138 
139 		if (nilfs_test_opt(nilfs, ERRORS_RO)) {
140 			printk(KERN_CRIT "Remounting filesystem read-only\n");
141 			sb->s_flags |= SB_RDONLY;
142 		}
143 	}
144 
145 	if (nilfs_test_opt(nilfs, ERRORS_PANIC))
146 		panic("NILFS (device %s): panic forced after error\n",
147 		      sb->s_id);
148 }
149 
150 struct inode *nilfs_alloc_inode(struct super_block *sb)
151 {
152 	struct nilfs_inode_info *ii;
153 
154 	ii = alloc_inode_sb(sb, nilfs_inode_cachep, GFP_NOFS);
155 	if (!ii)
156 		return NULL;
157 	ii->i_bh = NULL;
158 	ii->i_state = 0;
159 	ii->i_cno = 0;
160 	ii->i_assoc_inode = NULL;
161 	ii->i_bmap = &ii->i_bmap_data;
162 	return &ii->vfs_inode;
163 }
164 
165 static void nilfs_free_inode(struct inode *inode)
166 {
167 	if (nilfs_is_metadata_file_inode(inode))
168 		nilfs_mdt_destroy(inode);
169 
170 	kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
171 }
172 
173 static int nilfs_sync_super(struct super_block *sb, int flag)
174 {
175 	struct the_nilfs *nilfs = sb->s_fs_info;
176 	int err;
177 
178  retry:
179 	set_buffer_dirty(nilfs->ns_sbh[0]);
180 	if (nilfs_test_opt(nilfs, BARRIER)) {
181 		err = __sync_dirty_buffer(nilfs->ns_sbh[0],
182 					  REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
183 	} else {
184 		err = sync_dirty_buffer(nilfs->ns_sbh[0]);
185 	}
186 
187 	if (unlikely(err)) {
188 		nilfs_err(sb, "unable to write superblock: err=%d", err);
189 		if (err == -EIO && nilfs->ns_sbh[1]) {
190 			/*
191 			 * sbp[0] points to newer log than sbp[1],
192 			 * so copy sbp[0] to sbp[1] to take over sbp[0].
193 			 */
194 			memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0],
195 			       nilfs->ns_sbsize);
196 			nilfs_fall_back_super_block(nilfs);
197 			goto retry;
198 		}
199 	} else {
200 		struct nilfs_super_block *sbp = nilfs->ns_sbp[0];
201 
202 		nilfs->ns_sbwcount++;
203 
204 		/*
205 		 * The latest segment becomes trailable from the position
206 		 * written in superblock.
207 		 */
208 		clear_nilfs_discontinued(nilfs);
209 
210 		/* update GC protection for recent segments */
211 		if (nilfs->ns_sbh[1]) {
212 			if (flag == NILFS_SB_COMMIT_ALL) {
213 				set_buffer_dirty(nilfs->ns_sbh[1]);
214 				if (sync_dirty_buffer(nilfs->ns_sbh[1]) < 0)
215 					goto out;
216 			}
217 			if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) <
218 			    le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno))
219 				sbp = nilfs->ns_sbp[1];
220 		}
221 
222 		spin_lock(&nilfs->ns_last_segment_lock);
223 		nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq);
224 		spin_unlock(&nilfs->ns_last_segment_lock);
225 	}
226  out:
227 	return err;
228 }
229 
230 void nilfs_set_log_cursor(struct nilfs_super_block *sbp,
231 			  struct the_nilfs *nilfs)
232 {
233 	sector_t nfreeblocks;
234 
235 	/* nilfs->ns_sem must be locked by the caller. */
236 	nilfs_count_free_blocks(nilfs, &nfreeblocks);
237 	sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks);
238 
239 	spin_lock(&nilfs->ns_last_segment_lock);
240 	sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq);
241 	sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg);
242 	sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno);
243 	spin_unlock(&nilfs->ns_last_segment_lock);
244 }
245 
246 struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb,
247 					       int flip)
248 {
249 	struct the_nilfs *nilfs = sb->s_fs_info;
250 	struct nilfs_super_block **sbp = nilfs->ns_sbp;
251 
252 	/* nilfs->ns_sem must be locked by the caller. */
253 	if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
254 		if (sbp[1] &&
255 		    sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) {
256 			memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
257 		} else {
258 			nilfs_crit(sb, "superblock broke");
259 			return NULL;
260 		}
261 	} else if (sbp[1] &&
262 		   sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) {
263 		memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
264 	}
265 
266 	if (flip && sbp[1])
267 		nilfs_swap_super_block(nilfs);
268 
269 	return sbp;
270 }
271 
272 int nilfs_commit_super(struct super_block *sb, int flag)
273 {
274 	struct the_nilfs *nilfs = sb->s_fs_info;
275 	struct nilfs_super_block **sbp = nilfs->ns_sbp;
276 	time64_t t;
277 
278 	/* nilfs->ns_sem must be locked by the caller. */
279 	t = ktime_get_real_seconds();
280 	nilfs->ns_sbwtime = t;
281 	sbp[0]->s_wtime = cpu_to_le64(t);
282 	sbp[0]->s_sum = 0;
283 	sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
284 					     (unsigned char *)sbp[0],
285 					     nilfs->ns_sbsize));
286 	if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) {
287 		sbp[1]->s_wtime = sbp[0]->s_wtime;
288 		sbp[1]->s_sum = 0;
289 		sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed,
290 					    (unsigned char *)sbp[1],
291 					    nilfs->ns_sbsize));
292 	}
293 	clear_nilfs_sb_dirty(nilfs);
294 	nilfs->ns_flushed_device = 1;
295 	/* make sure store to ns_flushed_device cannot be reordered */
296 	smp_wmb();
297 	return nilfs_sync_super(sb, flag);
298 }
299 
300 /**
301  * nilfs_cleanup_super() - write filesystem state for cleanup
302  * @sb: super block instance to be unmounted or degraded to read-only
303  *
304  * This function restores state flags in the on-disk super block.
305  * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the
306  * filesystem was not clean previously.
307  */
308 int nilfs_cleanup_super(struct super_block *sb)
309 {
310 	struct the_nilfs *nilfs = sb->s_fs_info;
311 	struct nilfs_super_block **sbp;
312 	int flag = NILFS_SB_COMMIT;
313 	int ret = -EIO;
314 
315 	sbp = nilfs_prepare_super(sb, 0);
316 	if (sbp) {
317 		sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state);
318 		nilfs_set_log_cursor(sbp[0], nilfs);
319 		if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) {
320 			/*
321 			 * make the "clean" flag also to the opposite
322 			 * super block if both super blocks point to
323 			 * the same checkpoint.
324 			 */
325 			sbp[1]->s_state = sbp[0]->s_state;
326 			flag = NILFS_SB_COMMIT_ALL;
327 		}
328 		ret = nilfs_commit_super(sb, flag);
329 	}
330 	return ret;
331 }
332 
333 /**
334  * nilfs_move_2nd_super - relocate secondary super block
335  * @sb: super block instance
336  * @sb2off: new offset of the secondary super block (in bytes)
337  */
338 static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off)
339 {
340 	struct the_nilfs *nilfs = sb->s_fs_info;
341 	struct buffer_head *nsbh;
342 	struct nilfs_super_block *nsbp;
343 	sector_t blocknr, newblocknr;
344 	unsigned long offset;
345 	int sb2i;  /* array index of the secondary superblock */
346 	int ret = 0;
347 
348 	/* nilfs->ns_sem must be locked by the caller. */
349 	if (nilfs->ns_sbh[1] &&
350 	    nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) {
351 		sb2i = 1;
352 		blocknr = nilfs->ns_sbh[1]->b_blocknr;
353 	} else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) {
354 		sb2i = 0;
355 		blocknr = nilfs->ns_sbh[0]->b_blocknr;
356 	} else {
357 		sb2i = -1;
358 		blocknr = 0;
359 	}
360 	if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off)
361 		goto out;  /* super block location is unchanged */
362 
363 	/* Get new super block buffer */
364 	newblocknr = sb2off >> nilfs->ns_blocksize_bits;
365 	offset = sb2off & (nilfs->ns_blocksize - 1);
366 	nsbh = sb_getblk(sb, newblocknr);
367 	if (!nsbh) {
368 		nilfs_warn(sb,
369 			   "unable to move secondary superblock to block %llu",
370 			   (unsigned long long)newblocknr);
371 		ret = -EIO;
372 		goto out;
373 	}
374 	nsbp = (void *)nsbh->b_data + offset;
375 
376 	lock_buffer(nsbh);
377 	if (sb2i >= 0) {
378 		/*
379 		 * The position of the second superblock only changes by 4KiB,
380 		 * which is larger than the maximum superblock data size
381 		 * (= 1KiB), so there is no need to use memmove() to allow
382 		 * overlap between source and destination.
383 		 */
384 		memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize);
385 
386 		/*
387 		 * Zero fill after copy to avoid overwriting in case of move
388 		 * within the same block.
389 		 */
390 		memset(nsbh->b_data, 0, offset);
391 		memset((void *)nsbp + nilfs->ns_sbsize, 0,
392 		       nsbh->b_size - offset - nilfs->ns_sbsize);
393 	} else {
394 		memset(nsbh->b_data, 0, nsbh->b_size);
395 	}
396 	set_buffer_uptodate(nsbh);
397 	unlock_buffer(nsbh);
398 
399 	if (sb2i >= 0) {
400 		brelse(nilfs->ns_sbh[sb2i]);
401 		nilfs->ns_sbh[sb2i] = nsbh;
402 		nilfs->ns_sbp[sb2i] = nsbp;
403 	} else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) {
404 		/* secondary super block will be restored to index 1 */
405 		nilfs->ns_sbh[1] = nsbh;
406 		nilfs->ns_sbp[1] = nsbp;
407 	} else {
408 		brelse(nsbh);
409 	}
410 out:
411 	return ret;
412 }
413 
414 /**
415  * nilfs_resize_fs - resize the filesystem
416  * @sb: super block instance
417  * @newsize: new size of the filesystem (in bytes)
418  */
419 int nilfs_resize_fs(struct super_block *sb, __u64 newsize)
420 {
421 	struct the_nilfs *nilfs = sb->s_fs_info;
422 	struct nilfs_super_block **sbp;
423 	__u64 devsize, newnsegs;
424 	loff_t sb2off;
425 	int ret;
426 
427 	ret = -ERANGE;
428 	devsize = bdev_nr_bytes(sb->s_bdev);
429 	if (newsize > devsize)
430 		goto out;
431 
432 	/*
433 	 * Prevent underflow in second superblock position calculation.
434 	 * The exact minimum size check is done in nilfs_sufile_resize().
435 	 */
436 	if (newsize < 4096) {
437 		ret = -ENOSPC;
438 		goto out;
439 	}
440 
441 	/*
442 	 * Write lock is required to protect some functions depending
443 	 * on the number of segments, the number of reserved segments,
444 	 * and so forth.
445 	 */
446 	down_write(&nilfs->ns_segctor_sem);
447 
448 	sb2off = NILFS_SB2_OFFSET_BYTES(newsize);
449 	newnsegs = sb2off >> nilfs->ns_blocksize_bits;
450 	do_div(newnsegs, nilfs->ns_blocks_per_segment);
451 
452 	ret = nilfs_sufile_resize(nilfs->ns_sufile, newnsegs);
453 	up_write(&nilfs->ns_segctor_sem);
454 	if (ret < 0)
455 		goto out;
456 
457 	ret = nilfs_construct_segment(sb);
458 	if (ret < 0)
459 		goto out;
460 
461 	down_write(&nilfs->ns_sem);
462 	nilfs_move_2nd_super(sb, sb2off);
463 	ret = -EIO;
464 	sbp = nilfs_prepare_super(sb, 0);
465 	if (likely(sbp)) {
466 		nilfs_set_log_cursor(sbp[0], nilfs);
467 		/*
468 		 * Drop NILFS_RESIZE_FS flag for compatibility with
469 		 * mount-time resize which may be implemented in a
470 		 * future release.
471 		 */
472 		sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) &
473 					      ~NILFS_RESIZE_FS);
474 		sbp[0]->s_dev_size = cpu_to_le64(newsize);
475 		sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments);
476 		if (sbp[1])
477 			memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
478 		ret = nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
479 	}
480 	up_write(&nilfs->ns_sem);
481 
482 	/*
483 	 * Reset the range of allocatable segments last.  This order
484 	 * is important in the case of expansion because the secondary
485 	 * superblock must be protected from log write until migration
486 	 * completes.
487 	 */
488 	if (!ret)
489 		nilfs_sufile_set_alloc_range(nilfs->ns_sufile, 0, newnsegs - 1);
490 out:
491 	return ret;
492 }
493 
494 static void nilfs_put_super(struct super_block *sb)
495 {
496 	struct the_nilfs *nilfs = sb->s_fs_info;
497 
498 	nilfs_detach_log_writer(sb);
499 
500 	if (!sb_rdonly(sb)) {
501 		down_write(&nilfs->ns_sem);
502 		nilfs_cleanup_super(sb);
503 		up_write(&nilfs->ns_sem);
504 	}
505 
506 	nilfs_sysfs_delete_device_group(nilfs);
507 	iput(nilfs->ns_sufile);
508 	iput(nilfs->ns_cpfile);
509 	iput(nilfs->ns_dat);
510 
511 	destroy_nilfs(nilfs);
512 	sb->s_fs_info = NULL;
513 }
514 
515 static int nilfs_sync_fs(struct super_block *sb, int wait)
516 {
517 	struct the_nilfs *nilfs = sb->s_fs_info;
518 	struct nilfs_super_block **sbp;
519 	int err = 0;
520 
521 	/* This function is called when super block should be written back */
522 	if (wait)
523 		err = nilfs_construct_segment(sb);
524 
525 	down_write(&nilfs->ns_sem);
526 	if (nilfs_sb_dirty(nilfs)) {
527 		sbp = nilfs_prepare_super(sb, nilfs_sb_will_flip(nilfs));
528 		if (likely(sbp)) {
529 			nilfs_set_log_cursor(sbp[0], nilfs);
530 			nilfs_commit_super(sb, NILFS_SB_COMMIT);
531 		}
532 	}
533 	up_write(&nilfs->ns_sem);
534 
535 	if (!err)
536 		err = nilfs_flush_device(nilfs);
537 
538 	return err;
539 }
540 
541 int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt,
542 			    struct nilfs_root **rootp)
543 {
544 	struct the_nilfs *nilfs = sb->s_fs_info;
545 	struct nilfs_root *root;
546 	struct nilfs_checkpoint *raw_cp;
547 	struct buffer_head *bh_cp;
548 	int err = -ENOMEM;
549 
550 	root = nilfs_find_or_create_root(
551 		nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
552 	if (!root)
553 		return err;
554 
555 	if (root->ifile)
556 		goto reuse; /* already attached checkpoint */
557 
558 	down_read(&nilfs->ns_segctor_sem);
559 	err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
560 					  &bh_cp);
561 	up_read(&nilfs->ns_segctor_sem);
562 	if (unlikely(err)) {
563 		if (err == -ENOENT || err == -EINVAL) {
564 			nilfs_err(sb,
565 				  "Invalid checkpoint (checkpoint number=%llu)",
566 				  (unsigned long long)cno);
567 			err = -EINVAL;
568 		}
569 		goto failed;
570 	}
571 
572 	err = nilfs_ifile_read(sb, root, nilfs->ns_inode_size,
573 			       &raw_cp->cp_ifile_inode, &root->ifile);
574 	if (err)
575 		goto failed_bh;
576 
577 	atomic64_set(&root->inodes_count,
578 			le64_to_cpu(raw_cp->cp_inodes_count));
579 	atomic64_set(&root->blocks_count,
580 			le64_to_cpu(raw_cp->cp_blocks_count));
581 
582 	nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
583 
584  reuse:
585 	*rootp = root;
586 	return 0;
587 
588  failed_bh:
589 	nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
590  failed:
591 	nilfs_put_root(root);
592 
593 	return err;
594 }
595 
596 static int nilfs_freeze(struct super_block *sb)
597 {
598 	struct the_nilfs *nilfs = sb->s_fs_info;
599 	int err;
600 
601 	if (sb_rdonly(sb))
602 		return 0;
603 
604 	/* Mark super block clean */
605 	down_write(&nilfs->ns_sem);
606 	err = nilfs_cleanup_super(sb);
607 	up_write(&nilfs->ns_sem);
608 	return err;
609 }
610 
611 static int nilfs_unfreeze(struct super_block *sb)
612 {
613 	struct the_nilfs *nilfs = sb->s_fs_info;
614 
615 	if (sb_rdonly(sb))
616 		return 0;
617 
618 	down_write(&nilfs->ns_sem);
619 	nilfs_setup_super(sb, false);
620 	up_write(&nilfs->ns_sem);
621 	return 0;
622 }
623 
624 static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
625 {
626 	struct super_block *sb = dentry->d_sb;
627 	struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
628 	struct the_nilfs *nilfs = root->nilfs;
629 	u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
630 	unsigned long long blocks;
631 	unsigned long overhead;
632 	unsigned long nrsvblocks;
633 	sector_t nfreeblocks;
634 	u64 nmaxinodes, nfreeinodes;
635 	int err;
636 
637 	/*
638 	 * Compute all of the segment blocks
639 	 *
640 	 * The blocks before first segment and after last segment
641 	 * are excluded.
642 	 */
643 	blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments
644 		- nilfs->ns_first_data_block;
645 	nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment;
646 
647 	/*
648 	 * Compute the overhead
649 	 *
650 	 * When distributing meta data blocks outside segment structure,
651 	 * We must count them as the overhead.
652 	 */
653 	overhead = 0;
654 
655 	err = nilfs_count_free_blocks(nilfs, &nfreeblocks);
656 	if (unlikely(err))
657 		return err;
658 
659 	err = nilfs_ifile_count_free_inodes(root->ifile,
660 					    &nmaxinodes, &nfreeinodes);
661 	if (unlikely(err)) {
662 		nilfs_warn(sb, "failed to count free inodes: err=%d", err);
663 		if (err == -ERANGE) {
664 			/*
665 			 * If nilfs_palloc_count_max_entries() returns
666 			 * -ERANGE error code then we simply treat
667 			 * curent inodes count as maximum possible and
668 			 * zero as free inodes value.
669 			 */
670 			nmaxinodes = atomic64_read(&root->inodes_count);
671 			nfreeinodes = 0;
672 			err = 0;
673 		} else
674 			return err;
675 	}
676 
677 	buf->f_type = NILFS_SUPER_MAGIC;
678 	buf->f_bsize = sb->s_blocksize;
679 	buf->f_blocks = blocks - overhead;
680 	buf->f_bfree = nfreeblocks;
681 	buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
682 		(buf->f_bfree - nrsvblocks) : 0;
683 	buf->f_files = nmaxinodes;
684 	buf->f_ffree = nfreeinodes;
685 	buf->f_namelen = NILFS_NAME_LEN;
686 	buf->f_fsid = u64_to_fsid(id);
687 
688 	return 0;
689 }
690 
691 static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry)
692 {
693 	struct super_block *sb = dentry->d_sb;
694 	struct the_nilfs *nilfs = sb->s_fs_info;
695 	struct nilfs_root *root = NILFS_I(d_inode(dentry))->i_root;
696 
697 	if (!nilfs_test_opt(nilfs, BARRIER))
698 		seq_puts(seq, ",nobarrier");
699 	if (root->cno != NILFS_CPTREE_CURRENT_CNO)
700 		seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
701 	if (nilfs_test_opt(nilfs, ERRORS_PANIC))
702 		seq_puts(seq, ",errors=panic");
703 	if (nilfs_test_opt(nilfs, ERRORS_CONT))
704 		seq_puts(seq, ",errors=continue");
705 	if (nilfs_test_opt(nilfs, STRICT_ORDER))
706 		seq_puts(seq, ",order=strict");
707 	if (nilfs_test_opt(nilfs, NORECOVERY))
708 		seq_puts(seq, ",norecovery");
709 	if (nilfs_test_opt(nilfs, DISCARD))
710 		seq_puts(seq, ",discard");
711 
712 	return 0;
713 }
714 
715 static const struct super_operations nilfs_sops = {
716 	.alloc_inode    = nilfs_alloc_inode,
717 	.free_inode     = nilfs_free_inode,
718 	.dirty_inode    = nilfs_dirty_inode,
719 	.evict_inode    = nilfs_evict_inode,
720 	.put_super      = nilfs_put_super,
721 	.sync_fs        = nilfs_sync_fs,
722 	.freeze_fs	= nilfs_freeze,
723 	.unfreeze_fs	= nilfs_unfreeze,
724 	.statfs         = nilfs_statfs,
725 	.remount_fs     = nilfs_remount,
726 	.show_options = nilfs_show_options
727 };
728 
729 enum {
730 	Opt_err_cont, Opt_err_panic, Opt_err_ro,
731 	Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
732 	Opt_discard, Opt_nodiscard, Opt_err,
733 };
734 
735 static match_table_t tokens = {
736 	{Opt_err_cont, "errors=continue"},
737 	{Opt_err_panic, "errors=panic"},
738 	{Opt_err_ro, "errors=remount-ro"},
739 	{Opt_barrier, "barrier"},
740 	{Opt_nobarrier, "nobarrier"},
741 	{Opt_snapshot, "cp=%u"},
742 	{Opt_order, "order=%s"},
743 	{Opt_norecovery, "norecovery"},
744 	{Opt_discard, "discard"},
745 	{Opt_nodiscard, "nodiscard"},
746 	{Opt_err, NULL}
747 };
748 
749 static int parse_options(char *options, struct super_block *sb, int is_remount)
750 {
751 	struct the_nilfs *nilfs = sb->s_fs_info;
752 	char *p;
753 	substring_t args[MAX_OPT_ARGS];
754 
755 	if (!options)
756 		return 1;
757 
758 	while ((p = strsep(&options, ",")) != NULL) {
759 		int token;
760 
761 		if (!*p)
762 			continue;
763 
764 		token = match_token(p, tokens, args);
765 		switch (token) {
766 		case Opt_barrier:
767 			nilfs_set_opt(nilfs, BARRIER);
768 			break;
769 		case Opt_nobarrier:
770 			nilfs_clear_opt(nilfs, BARRIER);
771 			break;
772 		case Opt_order:
773 			if (strcmp(args[0].from, "relaxed") == 0)
774 				/* Ordered data semantics */
775 				nilfs_clear_opt(nilfs, STRICT_ORDER);
776 			else if (strcmp(args[0].from, "strict") == 0)
777 				/* Strict in-order semantics */
778 				nilfs_set_opt(nilfs, STRICT_ORDER);
779 			else
780 				return 0;
781 			break;
782 		case Opt_err_panic:
783 			nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_PANIC);
784 			break;
785 		case Opt_err_ro:
786 			nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_RO);
787 			break;
788 		case Opt_err_cont:
789 			nilfs_write_opt(nilfs, ERROR_MODE, ERRORS_CONT);
790 			break;
791 		case Opt_snapshot:
792 			if (is_remount) {
793 				nilfs_err(sb,
794 					  "\"%s\" option is invalid for remount",
795 					  p);
796 				return 0;
797 			}
798 			break;
799 		case Opt_norecovery:
800 			nilfs_set_opt(nilfs, NORECOVERY);
801 			break;
802 		case Opt_discard:
803 			nilfs_set_opt(nilfs, DISCARD);
804 			break;
805 		case Opt_nodiscard:
806 			nilfs_clear_opt(nilfs, DISCARD);
807 			break;
808 		default:
809 			nilfs_err(sb, "unrecognized mount option \"%s\"", p);
810 			return 0;
811 		}
812 	}
813 	return 1;
814 }
815 
816 static inline void
817 nilfs_set_default_options(struct super_block *sb,
818 			  struct nilfs_super_block *sbp)
819 {
820 	struct the_nilfs *nilfs = sb->s_fs_info;
821 
822 	nilfs->ns_mount_opt =
823 		NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
824 }
825 
826 static int nilfs_setup_super(struct super_block *sb, int is_mount)
827 {
828 	struct the_nilfs *nilfs = sb->s_fs_info;
829 	struct nilfs_super_block **sbp;
830 	int max_mnt_count;
831 	int mnt_count;
832 
833 	/* nilfs->ns_sem must be locked by the caller. */
834 	sbp = nilfs_prepare_super(sb, 0);
835 	if (!sbp)
836 		return -EIO;
837 
838 	if (!is_mount)
839 		goto skip_mount_setup;
840 
841 	max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
842 	mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
843 
844 	if (nilfs->ns_mount_state & NILFS_ERROR_FS) {
845 		nilfs_warn(sb, "mounting fs with errors");
846 #if 0
847 	} else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) {
848 		nilfs_warn(sb, "maximal mount count reached");
849 #endif
850 	}
851 	if (!max_mnt_count)
852 		sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
853 
854 	sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
855 	sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds());
856 
857 skip_mount_setup:
858 	sbp[0]->s_state =
859 		cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
860 	/* synchronize sbp[1] with sbp[0] */
861 	if (sbp[1])
862 		memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
863 	return nilfs_commit_super(sb, NILFS_SB_COMMIT_ALL);
864 }
865 
866 struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb,
867 						 u64 pos, int blocksize,
868 						 struct buffer_head **pbh)
869 {
870 	unsigned long long sb_index = pos;
871 	unsigned long offset;
872 
873 	offset = do_div(sb_index, blocksize);
874 	*pbh = sb_bread(sb, sb_index);
875 	if (!*pbh)
876 		return NULL;
877 	return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset);
878 }
879 
880 int nilfs_store_magic_and_option(struct super_block *sb,
881 				 struct nilfs_super_block *sbp,
882 				 char *data)
883 {
884 	struct the_nilfs *nilfs = sb->s_fs_info;
885 
886 	sb->s_magic = le16_to_cpu(sbp->s_magic);
887 
888 	/* FS independent flags */
889 #ifdef NILFS_ATIME_DISABLE
890 	sb->s_flags |= SB_NOATIME;
891 #endif
892 
893 	nilfs_set_default_options(sb, sbp);
894 
895 	nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid);
896 	nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid);
897 	nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval);
898 	nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max);
899 
900 	return !parse_options(data, sb, 0) ? -EINVAL : 0;
901 }
902 
903 int nilfs_check_feature_compatibility(struct super_block *sb,
904 				      struct nilfs_super_block *sbp)
905 {
906 	__u64 features;
907 
908 	features = le64_to_cpu(sbp->s_feature_incompat) &
909 		~NILFS_FEATURE_INCOMPAT_SUPP;
910 	if (features) {
911 		nilfs_err(sb,
912 			  "couldn't mount because of unsupported optional features (%llx)",
913 			  (unsigned long long)features);
914 		return -EINVAL;
915 	}
916 	features = le64_to_cpu(sbp->s_feature_compat_ro) &
917 		~NILFS_FEATURE_COMPAT_RO_SUPP;
918 	if (!sb_rdonly(sb) && features) {
919 		nilfs_err(sb,
920 			  "couldn't mount RDWR because of unsupported optional features (%llx)",
921 			  (unsigned long long)features);
922 		return -EINVAL;
923 	}
924 	return 0;
925 }
926 
927 static int nilfs_get_root_dentry(struct super_block *sb,
928 				 struct nilfs_root *root,
929 				 struct dentry **root_dentry)
930 {
931 	struct inode *inode;
932 	struct dentry *dentry;
933 	int ret = 0;
934 
935 	inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
936 	if (IS_ERR(inode)) {
937 		ret = PTR_ERR(inode);
938 		nilfs_err(sb, "error %d getting root inode", ret);
939 		goto out;
940 	}
941 	if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
942 		iput(inode);
943 		nilfs_err(sb, "corrupt root inode");
944 		ret = -EINVAL;
945 		goto out;
946 	}
947 
948 	if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
949 		dentry = d_find_alias(inode);
950 		if (!dentry) {
951 			dentry = d_make_root(inode);
952 			if (!dentry) {
953 				ret = -ENOMEM;
954 				goto failed_dentry;
955 			}
956 		} else {
957 			iput(inode);
958 		}
959 	} else {
960 		dentry = d_obtain_root(inode);
961 		if (IS_ERR(dentry)) {
962 			ret = PTR_ERR(dentry);
963 			goto failed_dentry;
964 		}
965 	}
966 	*root_dentry = dentry;
967  out:
968 	return ret;
969 
970  failed_dentry:
971 	nilfs_err(sb, "error %d getting root dentry", ret);
972 	goto out;
973 }
974 
975 static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
976 				 struct dentry **root_dentry)
977 {
978 	struct the_nilfs *nilfs = s->s_fs_info;
979 	struct nilfs_root *root;
980 	int ret;
981 
982 	mutex_lock(&nilfs->ns_snapshot_mount_mutex);
983 
984 	down_read(&nilfs->ns_segctor_sem);
985 	ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
986 	up_read(&nilfs->ns_segctor_sem);
987 	if (ret < 0) {
988 		ret = (ret == -ENOENT) ? -EINVAL : ret;
989 		goto out;
990 	} else if (!ret) {
991 		nilfs_err(s,
992 			  "The specified checkpoint is not a snapshot (checkpoint number=%llu)",
993 			  (unsigned long long)cno);
994 		ret = -EINVAL;
995 		goto out;
996 	}
997 
998 	ret = nilfs_attach_checkpoint(s, cno, false, &root);
999 	if (ret) {
1000 		nilfs_err(s,
1001 			  "error %d while loading snapshot (checkpoint number=%llu)",
1002 			  ret, (unsigned long long)cno);
1003 		goto out;
1004 	}
1005 	ret = nilfs_get_root_dentry(s, root, root_dentry);
1006 	nilfs_put_root(root);
1007  out:
1008 	mutex_unlock(&nilfs->ns_snapshot_mount_mutex);
1009 	return ret;
1010 }
1011 
1012 /**
1013  * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint
1014  * @root_dentry: root dentry of the tree to be shrunk
1015  *
1016  * This function returns true if the tree was in-use.
1017  */
1018 static bool nilfs_tree_is_busy(struct dentry *root_dentry)
1019 {
1020 	shrink_dcache_parent(root_dentry);
1021 	return d_count(root_dentry) > 1;
1022 }
1023 
1024 int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
1025 {
1026 	struct the_nilfs *nilfs = sb->s_fs_info;
1027 	struct nilfs_root *root;
1028 	struct inode *inode;
1029 	struct dentry *dentry;
1030 	int ret;
1031 
1032 	if (cno > nilfs->ns_cno)
1033 		return false;
1034 
1035 	if (cno >= nilfs_last_cno(nilfs))
1036 		return true;	/* protect recent checkpoints */
1037 
1038 	ret = false;
1039 	root = nilfs_lookup_root(nilfs, cno);
1040 	if (root) {
1041 		inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
1042 		if (inode) {
1043 			dentry = d_find_alias(inode);
1044 			if (dentry) {
1045 				ret = nilfs_tree_is_busy(dentry);
1046 				dput(dentry);
1047 			}
1048 			iput(inode);
1049 		}
1050 		nilfs_put_root(root);
1051 	}
1052 	return ret;
1053 }
1054 
1055 /**
1056  * nilfs_fill_super() - initialize a super block instance
1057  * @sb: super_block
1058  * @data: mount options
1059  * @silent: silent mode flag
1060  *
1061  * This function is called exclusively by nilfs->ns_mount_mutex.
1062  * So, the recovery process is protected from other simultaneous mounts.
1063  */
1064 static int
1065 nilfs_fill_super(struct super_block *sb, void *data, int silent)
1066 {
1067 	struct the_nilfs *nilfs;
1068 	struct nilfs_root *fsroot;
1069 	__u64 cno;
1070 	int err;
1071 
1072 	nilfs = alloc_nilfs(sb);
1073 	if (!nilfs)
1074 		return -ENOMEM;
1075 
1076 	sb->s_fs_info = nilfs;
1077 
1078 	err = init_nilfs(nilfs, sb, (char *)data);
1079 	if (err)
1080 		goto failed_nilfs;
1081 
1082 	sb->s_op = &nilfs_sops;
1083 	sb->s_export_op = &nilfs_export_ops;
1084 	sb->s_root = NULL;
1085 	sb->s_time_gran = 1;
1086 	sb->s_max_links = NILFS_LINK_MAX;
1087 
1088 	sb->s_bdi = bdi_get(sb->s_bdev->bd_disk->bdi);
1089 
1090 	err = load_nilfs(nilfs, sb);
1091 	if (err)
1092 		goto failed_nilfs;
1093 
1094 	cno = nilfs_last_cno(nilfs);
1095 	err = nilfs_attach_checkpoint(sb, cno, true, &fsroot);
1096 	if (err) {
1097 		nilfs_err(sb,
1098 			  "error %d while loading last checkpoint (checkpoint number=%llu)",
1099 			  err, (unsigned long long)cno);
1100 		goto failed_unload;
1101 	}
1102 
1103 	if (!sb_rdonly(sb)) {
1104 		err = nilfs_attach_log_writer(sb, fsroot);
1105 		if (err)
1106 			goto failed_checkpoint;
1107 	}
1108 
1109 	err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
1110 	if (err)
1111 		goto failed_segctor;
1112 
1113 	nilfs_put_root(fsroot);
1114 
1115 	if (!sb_rdonly(sb)) {
1116 		down_write(&nilfs->ns_sem);
1117 		nilfs_setup_super(sb, true);
1118 		up_write(&nilfs->ns_sem);
1119 	}
1120 
1121 	return 0;
1122 
1123  failed_segctor:
1124 	nilfs_detach_log_writer(sb);
1125 
1126  failed_checkpoint:
1127 	nilfs_put_root(fsroot);
1128 
1129  failed_unload:
1130 	nilfs_sysfs_delete_device_group(nilfs);
1131 	iput(nilfs->ns_sufile);
1132 	iput(nilfs->ns_cpfile);
1133 	iput(nilfs->ns_dat);
1134 
1135  failed_nilfs:
1136 	destroy_nilfs(nilfs);
1137 	return err;
1138 }
1139 
1140 static int nilfs_remount(struct super_block *sb, int *flags, char *data)
1141 {
1142 	struct the_nilfs *nilfs = sb->s_fs_info;
1143 	unsigned long old_sb_flags;
1144 	unsigned long old_mount_opt;
1145 	int err;
1146 
1147 	sync_filesystem(sb);
1148 	old_sb_flags = sb->s_flags;
1149 	old_mount_opt = nilfs->ns_mount_opt;
1150 
1151 	if (!parse_options(data, sb, 1)) {
1152 		err = -EINVAL;
1153 		goto restore_opts;
1154 	}
1155 	sb->s_flags = (sb->s_flags & ~SB_POSIXACL);
1156 
1157 	err = -EINVAL;
1158 
1159 	if (!nilfs_valid_fs(nilfs)) {
1160 		nilfs_warn(sb,
1161 			   "couldn't remount because the filesystem is in an incomplete recovery state");
1162 		goto restore_opts;
1163 	}
1164 
1165 	if ((bool)(*flags & SB_RDONLY) == sb_rdonly(sb))
1166 		goto out;
1167 	if (*flags & SB_RDONLY) {
1168 		sb->s_flags |= SB_RDONLY;
1169 
1170 		/*
1171 		 * Remounting a valid RW partition RDONLY, so set
1172 		 * the RDONLY flag and then mark the partition as valid again.
1173 		 */
1174 		down_write(&nilfs->ns_sem);
1175 		nilfs_cleanup_super(sb);
1176 		up_write(&nilfs->ns_sem);
1177 	} else {
1178 		__u64 features;
1179 		struct nilfs_root *root;
1180 
1181 		/*
1182 		 * Mounting a RDONLY partition read-write, so reread and
1183 		 * store the current valid flag.  (It may have been changed
1184 		 * by fsck since we originally mounted the partition.)
1185 		 */
1186 		down_read(&nilfs->ns_sem);
1187 		features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
1188 			~NILFS_FEATURE_COMPAT_RO_SUPP;
1189 		up_read(&nilfs->ns_sem);
1190 		if (features) {
1191 			nilfs_warn(sb,
1192 				   "couldn't remount RDWR because of unsupported optional features (%llx)",
1193 				   (unsigned long long)features);
1194 			err = -EROFS;
1195 			goto restore_opts;
1196 		}
1197 
1198 		sb->s_flags &= ~SB_RDONLY;
1199 
1200 		root = NILFS_I(d_inode(sb->s_root))->i_root;
1201 		err = nilfs_attach_log_writer(sb, root);
1202 		if (err)
1203 			goto restore_opts;
1204 
1205 		down_write(&nilfs->ns_sem);
1206 		nilfs_setup_super(sb, true);
1207 		up_write(&nilfs->ns_sem);
1208 	}
1209  out:
1210 	return 0;
1211 
1212  restore_opts:
1213 	sb->s_flags = old_sb_flags;
1214 	nilfs->ns_mount_opt = old_mount_opt;
1215 	return err;
1216 }
1217 
1218 struct nilfs_super_data {
1219 	struct block_device *bdev;
1220 	__u64 cno;
1221 	int flags;
1222 };
1223 
1224 static int nilfs_parse_snapshot_option(const char *option,
1225 				       const substring_t *arg,
1226 				       struct nilfs_super_data *sd)
1227 {
1228 	unsigned long long val;
1229 	const char *msg = NULL;
1230 	int err;
1231 
1232 	if (!(sd->flags & SB_RDONLY)) {
1233 		msg = "read-only option is not specified";
1234 		goto parse_error;
1235 	}
1236 
1237 	err = kstrtoull(arg->from, 0, &val);
1238 	if (err) {
1239 		if (err == -ERANGE)
1240 			msg = "too large checkpoint number";
1241 		else
1242 			msg = "malformed argument";
1243 		goto parse_error;
1244 	} else if (val == 0) {
1245 		msg = "invalid checkpoint number 0";
1246 		goto parse_error;
1247 	}
1248 	sd->cno = val;
1249 	return 0;
1250 
1251 parse_error:
1252 	nilfs_err(NULL, "invalid option \"%s\": %s", option, msg);
1253 	return 1;
1254 }
1255 
1256 /**
1257  * nilfs_identify - pre-read mount options needed to identify mount instance
1258  * @data: mount options
1259  * @sd: nilfs_super_data
1260  */
1261 static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1262 {
1263 	char *p, *options = data;
1264 	substring_t args[MAX_OPT_ARGS];
1265 	int token;
1266 	int ret = 0;
1267 
1268 	do {
1269 		p = strsep(&options, ",");
1270 		if (p != NULL && *p) {
1271 			token = match_token(p, tokens, args);
1272 			if (token == Opt_snapshot)
1273 				ret = nilfs_parse_snapshot_option(p, &args[0],
1274 								  sd);
1275 		}
1276 		if (!options)
1277 			break;
1278 		BUG_ON(options == data);
1279 		*(options - 1) = ',';
1280 	} while (!ret);
1281 	return ret;
1282 }
1283 
1284 static int nilfs_set_bdev_super(struct super_block *s, void *data)
1285 {
1286 	s->s_bdev = data;
1287 	s->s_dev = s->s_bdev->bd_dev;
1288 	return 0;
1289 }
1290 
1291 static int nilfs_test_bdev_super(struct super_block *s, void *data)
1292 {
1293 	return (void *)s->s_bdev == data;
1294 }
1295 
1296 static struct dentry *
1297 nilfs_mount(struct file_system_type *fs_type, int flags,
1298 	     const char *dev_name, void *data)
1299 {
1300 	struct nilfs_super_data sd;
1301 	struct super_block *s;
1302 	struct dentry *root_dentry;
1303 	int err, s_new = false;
1304 
1305 	sd.bdev = blkdev_get_by_path(dev_name, sb_open_mode(flags), fs_type,
1306 				     NULL);
1307 	if (IS_ERR(sd.bdev))
1308 		return ERR_CAST(sd.bdev);
1309 
1310 	sd.cno = 0;
1311 	sd.flags = flags;
1312 	if (nilfs_identify((char *)data, &sd)) {
1313 		err = -EINVAL;
1314 		goto failed;
1315 	}
1316 
1317 	/*
1318 	 * once the super is inserted into the list by sget, s_umount
1319 	 * will protect the lockfs code from trying to start a snapshot
1320 	 * while we are mounting
1321 	 */
1322 	mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1323 	if (sd.bdev->bd_fsfreeze_count > 0) {
1324 		mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1325 		err = -EBUSY;
1326 		goto failed;
1327 	}
1328 	s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, flags,
1329 		 sd.bdev);
1330 	mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1331 	if (IS_ERR(s)) {
1332 		err = PTR_ERR(s);
1333 		goto failed;
1334 	}
1335 
1336 	if (!s->s_root) {
1337 		s_new = true;
1338 
1339 		/* New superblock instance created */
1340 		snprintf(s->s_id, sizeof(s->s_id), "%pg", sd.bdev);
1341 		sb_set_blocksize(s, block_size(sd.bdev));
1342 
1343 		err = nilfs_fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1344 		if (err)
1345 			goto failed_super;
1346 
1347 		s->s_flags |= SB_ACTIVE;
1348 	} else if (!sd.cno) {
1349 		if (nilfs_tree_is_busy(s->s_root)) {
1350 			if ((flags ^ s->s_flags) & SB_RDONLY) {
1351 				nilfs_err(s,
1352 					  "the device already has a %s mount.",
1353 					  sb_rdonly(s) ? "read-only" : "read/write");
1354 				err = -EBUSY;
1355 				goto failed_super;
1356 			}
1357 		} else {
1358 			/*
1359 			 * Try remount to setup mount states if the current
1360 			 * tree is not mounted and only snapshots use this sb.
1361 			 */
1362 			err = nilfs_remount(s, &flags, data);
1363 			if (err)
1364 				goto failed_super;
1365 		}
1366 	}
1367 
1368 	if (sd.cno) {
1369 		err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1370 		if (err)
1371 			goto failed_super;
1372 	} else {
1373 		root_dentry = dget(s->s_root);
1374 	}
1375 
1376 	if (!s_new)
1377 		blkdev_put(sd.bdev, fs_type);
1378 
1379 	return root_dentry;
1380 
1381  failed_super:
1382 	deactivate_locked_super(s);
1383 
1384  failed:
1385 	if (!s_new)
1386 		blkdev_put(sd.bdev, fs_type);
1387 	return ERR_PTR(err);
1388 }
1389 
1390 struct file_system_type nilfs_fs_type = {
1391 	.owner    = THIS_MODULE,
1392 	.name     = "nilfs2",
1393 	.mount    = nilfs_mount,
1394 	.kill_sb  = kill_block_super,
1395 	.fs_flags = FS_REQUIRES_DEV,
1396 };
1397 MODULE_ALIAS_FS("nilfs2");
1398 
1399 static void nilfs_inode_init_once(void *obj)
1400 {
1401 	struct nilfs_inode_info *ii = obj;
1402 
1403 	INIT_LIST_HEAD(&ii->i_dirty);
1404 #ifdef CONFIG_NILFS_XATTR
1405 	init_rwsem(&ii->xattr_sem);
1406 #endif
1407 	inode_init_once(&ii->vfs_inode);
1408 }
1409 
1410 static void nilfs_segbuf_init_once(void *obj)
1411 {
1412 	memset(obj, 0, sizeof(struct nilfs_segment_buffer));
1413 }
1414 
1415 static void nilfs_destroy_cachep(void)
1416 {
1417 	/*
1418 	 * Make sure all delayed rcu free inodes are flushed before we
1419 	 * destroy cache.
1420 	 */
1421 	rcu_barrier();
1422 
1423 	kmem_cache_destroy(nilfs_inode_cachep);
1424 	kmem_cache_destroy(nilfs_transaction_cachep);
1425 	kmem_cache_destroy(nilfs_segbuf_cachep);
1426 	kmem_cache_destroy(nilfs_btree_path_cache);
1427 }
1428 
1429 static int __init nilfs_init_cachep(void)
1430 {
1431 	nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache",
1432 			sizeof(struct nilfs_inode_info), 0,
1433 			SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT,
1434 			nilfs_inode_init_once);
1435 	if (!nilfs_inode_cachep)
1436 		goto fail;
1437 
1438 	nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache",
1439 			sizeof(struct nilfs_transaction_info), 0,
1440 			SLAB_RECLAIM_ACCOUNT, NULL);
1441 	if (!nilfs_transaction_cachep)
1442 		goto fail;
1443 
1444 	nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache",
1445 			sizeof(struct nilfs_segment_buffer), 0,
1446 			SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once);
1447 	if (!nilfs_segbuf_cachep)
1448 		goto fail;
1449 
1450 	nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache",
1451 			sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX,
1452 			0, 0, NULL);
1453 	if (!nilfs_btree_path_cache)
1454 		goto fail;
1455 
1456 	return 0;
1457 
1458 fail:
1459 	nilfs_destroy_cachep();
1460 	return -ENOMEM;
1461 }
1462 
1463 static int __init init_nilfs_fs(void)
1464 {
1465 	int err;
1466 
1467 	err = nilfs_init_cachep();
1468 	if (err)
1469 		goto fail;
1470 
1471 	err = nilfs_sysfs_init();
1472 	if (err)
1473 		goto free_cachep;
1474 
1475 	err = register_filesystem(&nilfs_fs_type);
1476 	if (err)
1477 		goto deinit_sysfs_entry;
1478 
1479 	printk(KERN_INFO "NILFS version 2 loaded\n");
1480 	return 0;
1481 
1482 deinit_sysfs_entry:
1483 	nilfs_sysfs_exit();
1484 free_cachep:
1485 	nilfs_destroy_cachep();
1486 fail:
1487 	return err;
1488 }
1489 
1490 static void __exit exit_nilfs_fs(void)
1491 {
1492 	nilfs_destroy_cachep();
1493 	nilfs_sysfs_exit();
1494 	unregister_filesystem(&nilfs_fs_type);
1495 }
1496 
1497 module_init(init_nilfs_fs)
1498 module_exit(exit_nilfs_fs)
1499