xref: /linux/fs/fuse/inode.c (revision 5e2cb28dd7e182dfa641550dfa225913509ad45d)
1 /*
2   FUSE: Filesystem in Userspace
3   Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
4 
5   This program can be distributed under the terms of the GNU GPL.
6   See the file COPYING.
7 */
8 
9 #include "fuse_i.h"
10 
11 #include <linux/pagemap.h>
12 #include <linux/slab.h>
13 #include <linux/file.h>
14 #include <linux/seq_file.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/moduleparam.h>
18 #include <linux/fs_context.h>
19 #include <linux/fs_parser.h>
20 #include <linux/statfs.h>
21 #include <linux/random.h>
22 #include <linux/sched.h>
23 #include <linux/exportfs.h>
24 #include <linux/posix_acl.h>
25 #include <linux/pid_namespace.h>
26 #include <uapi/linux/magic.h>
27 
28 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
29 MODULE_DESCRIPTION("Filesystem in Userspace");
30 MODULE_LICENSE("GPL");
31 
32 static struct kmem_cache *fuse_inode_cachep;
33 struct list_head fuse_conn_list;
34 DEFINE_MUTEX(fuse_mutex);
35 
36 static int set_global_limit(const char *val, const struct kernel_param *kp);
37 
38 unsigned max_user_bgreq;
39 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
40 		  &max_user_bgreq, 0644);
41 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
42 MODULE_PARM_DESC(max_user_bgreq,
43  "Global limit for the maximum number of backgrounded requests an "
44  "unprivileged user can set");
45 
46 unsigned max_user_congthresh;
47 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
48 		  &max_user_congthresh, 0644);
49 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
50 MODULE_PARM_DESC(max_user_congthresh,
51  "Global limit for the maximum congestion threshold an "
52  "unprivileged user can set");
53 
54 #define FUSE_DEFAULT_BLKSIZE 512
55 
56 /** Maximum number of outstanding background requests */
57 #define FUSE_DEFAULT_MAX_BACKGROUND 12
58 
59 /** Congestion starts at 75% of maximum */
60 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
61 
62 #ifdef CONFIG_BLOCK
63 static struct file_system_type fuseblk_fs_type;
64 #endif
65 
66 struct fuse_forget_link *fuse_alloc_forget(void)
67 {
68 	return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
69 }
70 
71 static struct inode *fuse_alloc_inode(struct super_block *sb)
72 {
73 	struct fuse_inode *fi;
74 
75 	fi = alloc_inode_sb(sb, fuse_inode_cachep, GFP_KERNEL);
76 	if (!fi)
77 		return NULL;
78 
79 	fi->i_time = 0;
80 	fi->inval_mask = ~0;
81 	fi->nodeid = 0;
82 	fi->nlookup = 0;
83 	fi->attr_version = 0;
84 	fi->orig_ino = 0;
85 	fi->state = 0;
86 	mutex_init(&fi->mutex);
87 	spin_lock_init(&fi->lock);
88 	fi->forget = fuse_alloc_forget();
89 	if (!fi->forget)
90 		goto out_free;
91 
92 	if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
93 		goto out_free_forget;
94 
95 	return &fi->inode;
96 
97 out_free_forget:
98 	kfree(fi->forget);
99 out_free:
100 	kmem_cache_free(fuse_inode_cachep, fi);
101 	return NULL;
102 }
103 
104 static void fuse_free_inode(struct inode *inode)
105 {
106 	struct fuse_inode *fi = get_fuse_inode(inode);
107 
108 	mutex_destroy(&fi->mutex);
109 	kfree(fi->forget);
110 #ifdef CONFIG_FUSE_DAX
111 	kfree(fi->dax);
112 #endif
113 	kmem_cache_free(fuse_inode_cachep, fi);
114 }
115 
116 static void fuse_evict_inode(struct inode *inode)
117 {
118 	struct fuse_inode *fi = get_fuse_inode(inode);
119 
120 	/* Will write inode on close/munmap and in all other dirtiers */
121 	WARN_ON(inode->i_state & I_DIRTY_INODE);
122 
123 	truncate_inode_pages_final(&inode->i_data);
124 	clear_inode(inode);
125 	if (inode->i_sb->s_flags & SB_ACTIVE) {
126 		struct fuse_conn *fc = get_fuse_conn(inode);
127 
128 		if (FUSE_IS_DAX(inode))
129 			fuse_dax_inode_cleanup(inode);
130 		if (fi->nlookup) {
131 			fuse_queue_forget(fc, fi->forget, fi->nodeid,
132 					  fi->nlookup);
133 			fi->forget = NULL;
134 		}
135 	}
136 	if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
137 		WARN_ON(!list_empty(&fi->write_files));
138 		WARN_ON(!list_empty(&fi->queued_writes));
139 	}
140 }
141 
142 static int fuse_reconfigure(struct fs_context *fsc)
143 {
144 	struct super_block *sb = fsc->root->d_sb;
145 
146 	sync_filesystem(sb);
147 	if (fsc->sb_flags & SB_MANDLOCK)
148 		return -EINVAL;
149 
150 	return 0;
151 }
152 
153 /*
154  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
155  * so that it will fit.
156  */
157 static ino_t fuse_squash_ino(u64 ino64)
158 {
159 	ino_t ino = (ino_t) ino64;
160 	if (sizeof(ino_t) < sizeof(u64))
161 		ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
162 	return ino;
163 }
164 
165 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
166 				   struct fuse_statx *sx,
167 				   u64 attr_valid, u32 cache_mask)
168 {
169 	struct fuse_conn *fc = get_fuse_conn(inode);
170 	struct fuse_inode *fi = get_fuse_inode(inode);
171 
172 	lockdep_assert_held(&fi->lock);
173 
174 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
175 	fi->i_time = attr_valid;
176 	/* Clear basic stats from invalid mask */
177 	set_mask_bits(&fi->inval_mask, STATX_BASIC_STATS, 0);
178 
179 	inode->i_ino     = fuse_squash_ino(attr->ino);
180 	inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
181 	set_nlink(inode, attr->nlink);
182 	inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
183 	inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
184 	inode->i_blocks  = attr->blocks;
185 
186 	/* Sanitize nsecs */
187 	attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
188 	attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
189 	attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
190 
191 	inode_set_atime(inode, attr->atime, attr->atimensec);
192 	/* mtime from server may be stale due to local buffered write */
193 	if (!(cache_mask & STATX_MTIME)) {
194 		inode_set_mtime(inode, attr->mtime, attr->mtimensec);
195 	}
196 	if (!(cache_mask & STATX_CTIME)) {
197 		inode_set_ctime(inode, attr->ctime, attr->ctimensec);
198 	}
199 	if (sx) {
200 		/* Sanitize nsecs */
201 		sx->btime.tv_nsec =
202 			min_t(u32, sx->btime.tv_nsec, NSEC_PER_SEC - 1);
203 
204 		/*
205 		 * Btime has been queried, cache is valid (whether or not btime
206 		 * is available or not) so clear STATX_BTIME from inval_mask.
207 		 *
208 		 * Availability of the btime attribute is indicated in
209 		 * FUSE_I_BTIME
210 		 */
211 		set_mask_bits(&fi->inval_mask, STATX_BTIME, 0);
212 		if (sx->mask & STATX_BTIME) {
213 			set_bit(FUSE_I_BTIME, &fi->state);
214 			fi->i_btime.tv_sec = sx->btime.tv_sec;
215 			fi->i_btime.tv_nsec = sx->btime.tv_nsec;
216 		}
217 	}
218 
219 	if (attr->blksize != 0)
220 		inode->i_blkbits = ilog2(attr->blksize);
221 	else
222 		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
223 
224 	/*
225 	 * Don't set the sticky bit in i_mode, unless we want the VFS
226 	 * to check permissions.  This prevents failures due to the
227 	 * check in may_delete().
228 	 */
229 	fi->orig_i_mode = inode->i_mode;
230 	if (!fc->default_permissions)
231 		inode->i_mode &= ~S_ISVTX;
232 
233 	fi->orig_ino = attr->ino;
234 
235 	/*
236 	 * We are refreshing inode data and it is possible that another
237 	 * client set suid/sgid or security.capability xattr. So clear
238 	 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
239 	 * was set or if security.capability xattr was set. But we don't
240 	 * know if security.capability has been set or not. So clear it
241 	 * anyway. Its less efficient but should be safe.
242 	 */
243 	inode->i_flags &= ~S_NOSEC;
244 }
245 
246 u32 fuse_get_cache_mask(struct inode *inode)
247 {
248 	struct fuse_conn *fc = get_fuse_conn(inode);
249 
250 	if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
251 		return 0;
252 
253 	return STATX_MTIME | STATX_CTIME | STATX_SIZE;
254 }
255 
256 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
257 			    struct fuse_statx *sx,
258 			    u64 attr_valid, u64 attr_version)
259 {
260 	struct fuse_conn *fc = get_fuse_conn(inode);
261 	struct fuse_inode *fi = get_fuse_inode(inode);
262 	u32 cache_mask;
263 	loff_t oldsize;
264 	struct timespec64 old_mtime;
265 
266 	spin_lock(&fi->lock);
267 	/*
268 	 * In case of writeback_cache enabled, writes update mtime, ctime and
269 	 * may update i_size.  In these cases trust the cached value in the
270 	 * inode.
271 	 */
272 	cache_mask = fuse_get_cache_mask(inode);
273 	if (cache_mask & STATX_SIZE)
274 		attr->size = i_size_read(inode);
275 
276 	if (cache_mask & STATX_MTIME) {
277 		attr->mtime = inode_get_mtime_sec(inode);
278 		attr->mtimensec = inode_get_mtime_nsec(inode);
279 	}
280 	if (cache_mask & STATX_CTIME) {
281 		attr->ctime = inode_get_ctime_sec(inode);
282 		attr->ctimensec = inode_get_ctime_nsec(inode);
283 	}
284 
285 	if ((attr_version != 0 && fi->attr_version > attr_version) ||
286 	    test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
287 		spin_unlock(&fi->lock);
288 		return;
289 	}
290 
291 	old_mtime = inode_get_mtime(inode);
292 	fuse_change_attributes_common(inode, attr, sx, attr_valid, cache_mask);
293 
294 	oldsize = inode->i_size;
295 	/*
296 	 * In case of writeback_cache enabled, the cached writes beyond EOF
297 	 * extend local i_size without keeping userspace server in sync. So,
298 	 * attr->size coming from server can be stale. We cannot trust it.
299 	 */
300 	if (!(cache_mask & STATX_SIZE))
301 		i_size_write(inode, attr->size);
302 	spin_unlock(&fi->lock);
303 
304 	if (!cache_mask && S_ISREG(inode->i_mode)) {
305 		bool inval = false;
306 
307 		if (oldsize != attr->size) {
308 			truncate_pagecache(inode, attr->size);
309 			if (!fc->explicit_inval_data)
310 				inval = true;
311 		} else if (fc->auto_inval_data) {
312 			struct timespec64 new_mtime = {
313 				.tv_sec = attr->mtime,
314 				.tv_nsec = attr->mtimensec,
315 			};
316 
317 			/*
318 			 * Auto inval mode also checks and invalidates if mtime
319 			 * has changed.
320 			 */
321 			if (!timespec64_equal(&old_mtime, &new_mtime))
322 				inval = true;
323 		}
324 
325 		if (inval)
326 			invalidate_inode_pages2(inode->i_mapping);
327 	}
328 
329 	if (IS_ENABLED(CONFIG_FUSE_DAX))
330 		fuse_dax_dontcache(inode, attr->flags);
331 }
332 
333 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr,
334 			    struct fuse_conn *fc)
335 {
336 	inode->i_mode = attr->mode & S_IFMT;
337 	inode->i_size = attr->size;
338 	inode_set_mtime(inode, attr->mtime, attr->mtimensec);
339 	inode_set_ctime(inode, attr->ctime, attr->ctimensec);
340 	if (S_ISREG(inode->i_mode)) {
341 		fuse_init_common(inode);
342 		fuse_init_file_inode(inode, attr->flags);
343 	} else if (S_ISDIR(inode->i_mode))
344 		fuse_init_dir(inode);
345 	else if (S_ISLNK(inode->i_mode))
346 		fuse_init_symlink(inode);
347 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
348 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
349 		fuse_init_common(inode);
350 		init_special_inode(inode, inode->i_mode,
351 				   new_decode_dev(attr->rdev));
352 	} else
353 		BUG();
354 	/*
355 	 * Ensure that we don't cache acls for daemons without FUSE_POSIX_ACL
356 	 * so they see the exact same behavior as before.
357 	 */
358 	if (!fc->posix_acl)
359 		inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
360 }
361 
362 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
363 {
364 	u64 nodeid = *(u64 *) _nodeidp;
365 	if (get_node_id(inode) == nodeid)
366 		return 1;
367 	else
368 		return 0;
369 }
370 
371 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
372 {
373 	u64 nodeid = *(u64 *) _nodeidp;
374 	get_fuse_inode(inode)->nodeid = nodeid;
375 	return 0;
376 }
377 
378 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
379 			int generation, struct fuse_attr *attr,
380 			u64 attr_valid, u64 attr_version)
381 {
382 	struct inode *inode;
383 	struct fuse_inode *fi;
384 	struct fuse_conn *fc = get_fuse_conn_super(sb);
385 
386 	/*
387 	 * Auto mount points get their node id from the submount root, which is
388 	 * not a unique identifier within this filesystem.
389 	 *
390 	 * To avoid conflicts, do not place submount points into the inode hash
391 	 * table.
392 	 */
393 	if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
394 	    S_ISDIR(attr->mode)) {
395 		inode = new_inode(sb);
396 		if (!inode)
397 			return NULL;
398 
399 		fuse_init_inode(inode, attr, fc);
400 		get_fuse_inode(inode)->nodeid = nodeid;
401 		inode->i_flags |= S_AUTOMOUNT;
402 		goto done;
403 	}
404 
405 retry:
406 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
407 	if (!inode)
408 		return NULL;
409 
410 	if ((inode->i_state & I_NEW)) {
411 		inode->i_flags |= S_NOATIME;
412 		if (!fc->writeback_cache || !S_ISREG(attr->mode))
413 			inode->i_flags |= S_NOCMTIME;
414 		inode->i_generation = generation;
415 		fuse_init_inode(inode, attr, fc);
416 		unlock_new_inode(inode);
417 	} else if (fuse_stale_inode(inode, generation, attr)) {
418 		/* nodeid was reused, any I/O on the old inode should fail */
419 		fuse_make_bad(inode);
420 		iput(inode);
421 		goto retry;
422 	}
423 done:
424 	fi = get_fuse_inode(inode);
425 	spin_lock(&fi->lock);
426 	fi->nlookup++;
427 	spin_unlock(&fi->lock);
428 	fuse_change_attributes(inode, attr, NULL, attr_valid, attr_version);
429 
430 	return inode;
431 }
432 
433 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
434 			   struct fuse_mount **fm)
435 {
436 	struct fuse_mount *fm_iter;
437 	struct inode *inode;
438 
439 	WARN_ON(!rwsem_is_locked(&fc->killsb));
440 	list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
441 		if (!fm_iter->sb)
442 			continue;
443 
444 		inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
445 		if (inode) {
446 			if (fm)
447 				*fm = fm_iter;
448 			return inode;
449 		}
450 	}
451 
452 	return NULL;
453 }
454 
455 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
456 			     loff_t offset, loff_t len)
457 {
458 	struct fuse_inode *fi;
459 	struct inode *inode;
460 	pgoff_t pg_start;
461 	pgoff_t pg_end;
462 
463 	inode = fuse_ilookup(fc, nodeid, NULL);
464 	if (!inode)
465 		return -ENOENT;
466 
467 	fi = get_fuse_inode(inode);
468 	spin_lock(&fi->lock);
469 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
470 	spin_unlock(&fi->lock);
471 
472 	fuse_invalidate_attr(inode);
473 	forget_all_cached_acls(inode);
474 	if (offset >= 0) {
475 		pg_start = offset >> PAGE_SHIFT;
476 		if (len <= 0)
477 			pg_end = -1;
478 		else
479 			pg_end = (offset + len - 1) >> PAGE_SHIFT;
480 		invalidate_inode_pages2_range(inode->i_mapping,
481 					      pg_start, pg_end);
482 	}
483 	iput(inode);
484 	return 0;
485 }
486 
487 bool fuse_lock_inode(struct inode *inode)
488 {
489 	bool locked = false;
490 
491 	if (!get_fuse_conn(inode)->parallel_dirops) {
492 		mutex_lock(&get_fuse_inode(inode)->mutex);
493 		locked = true;
494 	}
495 
496 	return locked;
497 }
498 
499 void fuse_unlock_inode(struct inode *inode, bool locked)
500 {
501 	if (locked)
502 		mutex_unlock(&get_fuse_inode(inode)->mutex);
503 }
504 
505 static void fuse_umount_begin(struct super_block *sb)
506 {
507 	struct fuse_conn *fc = get_fuse_conn_super(sb);
508 
509 	if (fc->no_force_umount)
510 		return;
511 
512 	fuse_abort_conn(fc);
513 
514 	// Only retire block-device-based superblocks.
515 	if (sb->s_bdev != NULL)
516 		retire_super(sb);
517 }
518 
519 static void fuse_send_destroy(struct fuse_mount *fm)
520 {
521 	if (fm->fc->conn_init) {
522 		FUSE_ARGS(args);
523 
524 		args.opcode = FUSE_DESTROY;
525 		args.force = true;
526 		args.nocreds = true;
527 		fuse_simple_request(fm, &args);
528 	}
529 }
530 
531 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
532 {
533 	stbuf->f_type    = FUSE_SUPER_MAGIC;
534 	stbuf->f_bsize   = attr->bsize;
535 	stbuf->f_frsize  = attr->frsize;
536 	stbuf->f_blocks  = attr->blocks;
537 	stbuf->f_bfree   = attr->bfree;
538 	stbuf->f_bavail  = attr->bavail;
539 	stbuf->f_files   = attr->files;
540 	stbuf->f_ffree   = attr->ffree;
541 	stbuf->f_namelen = attr->namelen;
542 	/* fsid is left zero */
543 }
544 
545 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
546 {
547 	struct super_block *sb = dentry->d_sb;
548 	struct fuse_mount *fm = get_fuse_mount_super(sb);
549 	FUSE_ARGS(args);
550 	struct fuse_statfs_out outarg;
551 	int err;
552 
553 	if (!fuse_allow_current_process(fm->fc)) {
554 		buf->f_type = FUSE_SUPER_MAGIC;
555 		return 0;
556 	}
557 
558 	memset(&outarg, 0, sizeof(outarg));
559 	args.in_numargs = 0;
560 	args.opcode = FUSE_STATFS;
561 	args.nodeid = get_node_id(d_inode(dentry));
562 	args.out_numargs = 1;
563 	args.out_args[0].size = sizeof(outarg);
564 	args.out_args[0].value = &outarg;
565 	err = fuse_simple_request(fm, &args);
566 	if (!err)
567 		convert_fuse_statfs(buf, &outarg.st);
568 	return err;
569 }
570 
571 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
572 {
573 	struct fuse_sync_bucket *bucket;
574 
575 	bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
576 	if (bucket) {
577 		init_waitqueue_head(&bucket->waitq);
578 		/* Initial active count */
579 		atomic_set(&bucket->count, 1);
580 	}
581 	return bucket;
582 }
583 
584 static void fuse_sync_fs_writes(struct fuse_conn *fc)
585 {
586 	struct fuse_sync_bucket *bucket, *new_bucket;
587 	int count;
588 
589 	new_bucket = fuse_sync_bucket_alloc();
590 	spin_lock(&fc->lock);
591 	bucket = rcu_dereference_protected(fc->curr_bucket, 1);
592 	count = atomic_read(&bucket->count);
593 	WARN_ON(count < 1);
594 	/* No outstanding writes? */
595 	if (count == 1) {
596 		spin_unlock(&fc->lock);
597 		kfree(new_bucket);
598 		return;
599 	}
600 
601 	/*
602 	 * Completion of new bucket depends on completion of this bucket, so add
603 	 * one more count.
604 	 */
605 	atomic_inc(&new_bucket->count);
606 	rcu_assign_pointer(fc->curr_bucket, new_bucket);
607 	spin_unlock(&fc->lock);
608 	/*
609 	 * Drop initial active count.  At this point if all writes in this and
610 	 * ancestor buckets complete, the count will go to zero and this task
611 	 * will be woken up.
612 	 */
613 	atomic_dec(&bucket->count);
614 
615 	wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
616 
617 	/* Drop temp count on descendant bucket */
618 	fuse_sync_bucket_dec(new_bucket);
619 	kfree_rcu(bucket, rcu);
620 }
621 
622 static int fuse_sync_fs(struct super_block *sb, int wait)
623 {
624 	struct fuse_mount *fm = get_fuse_mount_super(sb);
625 	struct fuse_conn *fc = fm->fc;
626 	struct fuse_syncfs_in inarg;
627 	FUSE_ARGS(args);
628 	int err;
629 
630 	/*
631 	 * Userspace cannot handle the wait == 0 case.  Avoid a
632 	 * gratuitous roundtrip.
633 	 */
634 	if (!wait)
635 		return 0;
636 
637 	/* The filesystem is being unmounted.  Nothing to do. */
638 	if (!sb->s_root)
639 		return 0;
640 
641 	if (!fc->sync_fs)
642 		return 0;
643 
644 	fuse_sync_fs_writes(fc);
645 
646 	memset(&inarg, 0, sizeof(inarg));
647 	args.in_numargs = 1;
648 	args.in_args[0].size = sizeof(inarg);
649 	args.in_args[0].value = &inarg;
650 	args.opcode = FUSE_SYNCFS;
651 	args.nodeid = get_node_id(sb->s_root->d_inode);
652 	args.out_numargs = 0;
653 
654 	err = fuse_simple_request(fm, &args);
655 	if (err == -ENOSYS) {
656 		fc->sync_fs = 0;
657 		err = 0;
658 	}
659 
660 	return err;
661 }
662 
663 enum {
664 	OPT_SOURCE,
665 	OPT_SUBTYPE,
666 	OPT_FD,
667 	OPT_ROOTMODE,
668 	OPT_USER_ID,
669 	OPT_GROUP_ID,
670 	OPT_DEFAULT_PERMISSIONS,
671 	OPT_ALLOW_OTHER,
672 	OPT_MAX_READ,
673 	OPT_BLKSIZE,
674 	OPT_ERR
675 };
676 
677 static const struct fs_parameter_spec fuse_fs_parameters[] = {
678 	fsparam_string	("source",		OPT_SOURCE),
679 	fsparam_u32	("fd",			OPT_FD),
680 	fsparam_u32oct	("rootmode",		OPT_ROOTMODE),
681 	fsparam_u32	("user_id",		OPT_USER_ID),
682 	fsparam_u32	("group_id",		OPT_GROUP_ID),
683 	fsparam_flag	("default_permissions",	OPT_DEFAULT_PERMISSIONS),
684 	fsparam_flag	("allow_other",		OPT_ALLOW_OTHER),
685 	fsparam_u32	("max_read",		OPT_MAX_READ),
686 	fsparam_u32	("blksize",		OPT_BLKSIZE),
687 	fsparam_string	("subtype",		OPT_SUBTYPE),
688 	{}
689 };
690 
691 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
692 {
693 	struct fs_parse_result result;
694 	struct fuse_fs_context *ctx = fsc->fs_private;
695 	int opt;
696 
697 	if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
698 		/*
699 		 * Ignore options coming from mount(MS_REMOUNT) for backward
700 		 * compatibility.
701 		 */
702 		if (fsc->oldapi)
703 			return 0;
704 
705 		return invalfc(fsc, "No changes allowed in reconfigure");
706 	}
707 
708 	opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
709 	if (opt < 0)
710 		return opt;
711 
712 	switch (opt) {
713 	case OPT_SOURCE:
714 		if (fsc->source)
715 			return invalfc(fsc, "Multiple sources specified");
716 		fsc->source = param->string;
717 		param->string = NULL;
718 		break;
719 
720 	case OPT_SUBTYPE:
721 		if (ctx->subtype)
722 			return invalfc(fsc, "Multiple subtypes specified");
723 		ctx->subtype = param->string;
724 		param->string = NULL;
725 		return 0;
726 
727 	case OPT_FD:
728 		ctx->fd = result.uint_32;
729 		ctx->fd_present = true;
730 		break;
731 
732 	case OPT_ROOTMODE:
733 		if (!fuse_valid_type(result.uint_32))
734 			return invalfc(fsc, "Invalid rootmode");
735 		ctx->rootmode = result.uint_32;
736 		ctx->rootmode_present = true;
737 		break;
738 
739 	case OPT_USER_ID:
740 		ctx->user_id = make_kuid(fsc->user_ns, result.uint_32);
741 		if (!uid_valid(ctx->user_id))
742 			return invalfc(fsc, "Invalid user_id");
743 		ctx->user_id_present = true;
744 		break;
745 
746 	case OPT_GROUP_ID:
747 		ctx->group_id = make_kgid(fsc->user_ns, result.uint_32);
748 		if (!gid_valid(ctx->group_id))
749 			return invalfc(fsc, "Invalid group_id");
750 		ctx->group_id_present = true;
751 		break;
752 
753 	case OPT_DEFAULT_PERMISSIONS:
754 		ctx->default_permissions = true;
755 		break;
756 
757 	case OPT_ALLOW_OTHER:
758 		ctx->allow_other = true;
759 		break;
760 
761 	case OPT_MAX_READ:
762 		ctx->max_read = result.uint_32;
763 		break;
764 
765 	case OPT_BLKSIZE:
766 		if (!ctx->is_bdev)
767 			return invalfc(fsc, "blksize only supported for fuseblk");
768 		ctx->blksize = result.uint_32;
769 		break;
770 
771 	default:
772 		return -EINVAL;
773 	}
774 
775 	return 0;
776 }
777 
778 static void fuse_free_fsc(struct fs_context *fsc)
779 {
780 	struct fuse_fs_context *ctx = fsc->fs_private;
781 
782 	if (ctx) {
783 		kfree(ctx->subtype);
784 		kfree(ctx);
785 	}
786 }
787 
788 static int fuse_show_options(struct seq_file *m, struct dentry *root)
789 {
790 	struct super_block *sb = root->d_sb;
791 	struct fuse_conn *fc = get_fuse_conn_super(sb);
792 
793 	if (fc->legacy_opts_show) {
794 		seq_printf(m, ",user_id=%u",
795 			   from_kuid_munged(fc->user_ns, fc->user_id));
796 		seq_printf(m, ",group_id=%u",
797 			   from_kgid_munged(fc->user_ns, fc->group_id));
798 		if (fc->default_permissions)
799 			seq_puts(m, ",default_permissions");
800 		if (fc->allow_other)
801 			seq_puts(m, ",allow_other");
802 		if (fc->max_read != ~0)
803 			seq_printf(m, ",max_read=%u", fc->max_read);
804 		if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
805 			seq_printf(m, ",blksize=%lu", sb->s_blocksize);
806 	}
807 #ifdef CONFIG_FUSE_DAX
808 	if (fc->dax_mode == FUSE_DAX_ALWAYS)
809 		seq_puts(m, ",dax=always");
810 	else if (fc->dax_mode == FUSE_DAX_NEVER)
811 		seq_puts(m, ",dax=never");
812 	else if (fc->dax_mode == FUSE_DAX_INODE_USER)
813 		seq_puts(m, ",dax=inode");
814 #endif
815 
816 	return 0;
817 }
818 
819 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
820 			     const struct fuse_iqueue_ops *ops,
821 			     void *priv)
822 {
823 	memset(fiq, 0, sizeof(struct fuse_iqueue));
824 	spin_lock_init(&fiq->lock);
825 	init_waitqueue_head(&fiq->waitq);
826 	INIT_LIST_HEAD(&fiq->pending);
827 	INIT_LIST_HEAD(&fiq->interrupts);
828 	fiq->forget_list_tail = &fiq->forget_list_head;
829 	fiq->connected = 1;
830 	fiq->ops = ops;
831 	fiq->priv = priv;
832 }
833 
834 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
835 {
836 	unsigned int i;
837 
838 	spin_lock_init(&fpq->lock);
839 	for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
840 		INIT_LIST_HEAD(&fpq->processing[i]);
841 	INIT_LIST_HEAD(&fpq->io);
842 	fpq->connected = 1;
843 }
844 
845 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
846 		    struct user_namespace *user_ns,
847 		    const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
848 {
849 	memset(fc, 0, sizeof(*fc));
850 	spin_lock_init(&fc->lock);
851 	spin_lock_init(&fc->bg_lock);
852 	init_rwsem(&fc->killsb);
853 	refcount_set(&fc->count, 1);
854 	atomic_set(&fc->dev_count, 1);
855 	init_waitqueue_head(&fc->blocked_waitq);
856 	fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
857 	INIT_LIST_HEAD(&fc->bg_queue);
858 	INIT_LIST_HEAD(&fc->entry);
859 	INIT_LIST_HEAD(&fc->devices);
860 	atomic_set(&fc->num_waiting, 0);
861 	fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
862 	fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
863 	atomic64_set(&fc->khctr, 0);
864 	fc->polled_files = RB_ROOT;
865 	fc->blocked = 0;
866 	fc->initialized = 0;
867 	fc->connected = 1;
868 	atomic64_set(&fc->attr_version, 1);
869 	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
870 	fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
871 	fc->user_ns = get_user_ns(user_ns);
872 	fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
873 	fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
874 
875 	INIT_LIST_HEAD(&fc->mounts);
876 	list_add(&fm->fc_entry, &fc->mounts);
877 	fm->fc = fc;
878 }
879 EXPORT_SYMBOL_GPL(fuse_conn_init);
880 
881 void fuse_conn_put(struct fuse_conn *fc)
882 {
883 	if (refcount_dec_and_test(&fc->count)) {
884 		struct fuse_iqueue *fiq = &fc->iq;
885 		struct fuse_sync_bucket *bucket;
886 
887 		if (IS_ENABLED(CONFIG_FUSE_DAX))
888 			fuse_dax_conn_free(fc);
889 		if (fiq->ops->release)
890 			fiq->ops->release(fiq);
891 		put_pid_ns(fc->pid_ns);
892 		put_user_ns(fc->user_ns);
893 		bucket = rcu_dereference_protected(fc->curr_bucket, 1);
894 		if (bucket) {
895 			WARN_ON(atomic_read(&bucket->count) != 1);
896 			kfree(bucket);
897 		}
898 		fc->release(fc);
899 	}
900 }
901 EXPORT_SYMBOL_GPL(fuse_conn_put);
902 
903 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
904 {
905 	refcount_inc(&fc->count);
906 	return fc;
907 }
908 EXPORT_SYMBOL_GPL(fuse_conn_get);
909 
910 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
911 {
912 	struct fuse_attr attr;
913 	memset(&attr, 0, sizeof(attr));
914 
915 	attr.mode = mode;
916 	attr.ino = FUSE_ROOT_ID;
917 	attr.nlink = 1;
918 	return fuse_iget(sb, 1, 0, &attr, 0, 0);
919 }
920 
921 struct fuse_inode_handle {
922 	u64 nodeid;
923 	u32 generation;
924 };
925 
926 static struct dentry *fuse_get_dentry(struct super_block *sb,
927 				      struct fuse_inode_handle *handle)
928 {
929 	struct fuse_conn *fc = get_fuse_conn_super(sb);
930 	struct inode *inode;
931 	struct dentry *entry;
932 	int err = -ESTALE;
933 
934 	if (handle->nodeid == 0)
935 		goto out_err;
936 
937 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
938 	if (!inode) {
939 		struct fuse_entry_out outarg;
940 		const struct qstr name = QSTR_INIT(".", 1);
941 
942 		if (!fc->export_support)
943 			goto out_err;
944 
945 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
946 				       &inode);
947 		if (err && err != -ENOENT)
948 			goto out_err;
949 		if (err || !inode) {
950 			err = -ESTALE;
951 			goto out_err;
952 		}
953 		err = -EIO;
954 		if (get_node_id(inode) != handle->nodeid)
955 			goto out_iput;
956 	}
957 	err = -ESTALE;
958 	if (inode->i_generation != handle->generation)
959 		goto out_iput;
960 
961 	entry = d_obtain_alias(inode);
962 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
963 		fuse_invalidate_entry_cache(entry);
964 
965 	return entry;
966 
967  out_iput:
968 	iput(inode);
969  out_err:
970 	return ERR_PTR(err);
971 }
972 
973 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
974 			   struct inode *parent)
975 {
976 	int len = parent ? 6 : 3;
977 	u64 nodeid;
978 	u32 generation;
979 
980 	if (*max_len < len) {
981 		*max_len = len;
982 		return  FILEID_INVALID;
983 	}
984 
985 	nodeid = get_fuse_inode(inode)->nodeid;
986 	generation = inode->i_generation;
987 
988 	fh[0] = (u32)(nodeid >> 32);
989 	fh[1] = (u32)(nodeid & 0xffffffff);
990 	fh[2] = generation;
991 
992 	if (parent) {
993 		nodeid = get_fuse_inode(parent)->nodeid;
994 		generation = parent->i_generation;
995 
996 		fh[3] = (u32)(nodeid >> 32);
997 		fh[4] = (u32)(nodeid & 0xffffffff);
998 		fh[5] = generation;
999 	}
1000 
1001 	*max_len = len;
1002 	return parent ? 0x82 : 0x81;
1003 }
1004 
1005 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
1006 		struct fid *fid, int fh_len, int fh_type)
1007 {
1008 	struct fuse_inode_handle handle;
1009 
1010 	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
1011 		return NULL;
1012 
1013 	handle.nodeid = (u64) fid->raw[0] << 32;
1014 	handle.nodeid |= (u64) fid->raw[1];
1015 	handle.generation = fid->raw[2];
1016 	return fuse_get_dentry(sb, &handle);
1017 }
1018 
1019 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
1020 		struct fid *fid, int fh_len, int fh_type)
1021 {
1022 	struct fuse_inode_handle parent;
1023 
1024 	if (fh_type != 0x82 || fh_len < 6)
1025 		return NULL;
1026 
1027 	parent.nodeid = (u64) fid->raw[3] << 32;
1028 	parent.nodeid |= (u64) fid->raw[4];
1029 	parent.generation = fid->raw[5];
1030 	return fuse_get_dentry(sb, &parent);
1031 }
1032 
1033 static struct dentry *fuse_get_parent(struct dentry *child)
1034 {
1035 	struct inode *child_inode = d_inode(child);
1036 	struct fuse_conn *fc = get_fuse_conn(child_inode);
1037 	struct inode *inode;
1038 	struct dentry *parent;
1039 	struct fuse_entry_out outarg;
1040 	int err;
1041 
1042 	if (!fc->export_support)
1043 		return ERR_PTR(-ESTALE);
1044 
1045 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
1046 			       &dotdot_name, &outarg, &inode);
1047 	if (err) {
1048 		if (err == -ENOENT)
1049 			return ERR_PTR(-ESTALE);
1050 		return ERR_PTR(err);
1051 	}
1052 
1053 	parent = d_obtain_alias(inode);
1054 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
1055 		fuse_invalidate_entry_cache(parent);
1056 
1057 	return parent;
1058 }
1059 
1060 static const struct export_operations fuse_export_operations = {
1061 	.fh_to_dentry	= fuse_fh_to_dentry,
1062 	.fh_to_parent	= fuse_fh_to_parent,
1063 	.encode_fh	= fuse_encode_fh,
1064 	.get_parent	= fuse_get_parent,
1065 };
1066 
1067 static const struct super_operations fuse_super_operations = {
1068 	.alloc_inode    = fuse_alloc_inode,
1069 	.free_inode     = fuse_free_inode,
1070 	.evict_inode	= fuse_evict_inode,
1071 	.write_inode	= fuse_write_inode,
1072 	.drop_inode	= generic_delete_inode,
1073 	.umount_begin	= fuse_umount_begin,
1074 	.statfs		= fuse_statfs,
1075 	.sync_fs	= fuse_sync_fs,
1076 	.show_options	= fuse_show_options,
1077 };
1078 
1079 static void sanitize_global_limit(unsigned *limit)
1080 {
1081 	/*
1082 	 * The default maximum number of async requests is calculated to consume
1083 	 * 1/2^13 of the total memory, assuming 392 bytes per request.
1084 	 */
1085 	if (*limit == 0)
1086 		*limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1087 
1088 	if (*limit >= 1 << 16)
1089 		*limit = (1 << 16) - 1;
1090 }
1091 
1092 static int set_global_limit(const char *val, const struct kernel_param *kp)
1093 {
1094 	int rv;
1095 
1096 	rv = param_set_uint(val, kp);
1097 	if (rv)
1098 		return rv;
1099 
1100 	sanitize_global_limit((unsigned *)kp->arg);
1101 
1102 	return 0;
1103 }
1104 
1105 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1106 {
1107 	int cap_sys_admin = capable(CAP_SYS_ADMIN);
1108 
1109 	if (arg->minor < 13)
1110 		return;
1111 
1112 	sanitize_global_limit(&max_user_bgreq);
1113 	sanitize_global_limit(&max_user_congthresh);
1114 
1115 	spin_lock(&fc->bg_lock);
1116 	if (arg->max_background) {
1117 		fc->max_background = arg->max_background;
1118 
1119 		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1120 			fc->max_background = max_user_bgreq;
1121 	}
1122 	if (arg->congestion_threshold) {
1123 		fc->congestion_threshold = arg->congestion_threshold;
1124 
1125 		if (!cap_sys_admin &&
1126 		    fc->congestion_threshold > max_user_congthresh)
1127 			fc->congestion_threshold = max_user_congthresh;
1128 	}
1129 	spin_unlock(&fc->bg_lock);
1130 }
1131 
1132 struct fuse_init_args {
1133 	struct fuse_args args;
1134 	struct fuse_init_in in;
1135 	struct fuse_init_out out;
1136 };
1137 
1138 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1139 			       int error)
1140 {
1141 	struct fuse_conn *fc = fm->fc;
1142 	struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1143 	struct fuse_init_out *arg = &ia->out;
1144 	bool ok = true;
1145 
1146 	if (error || arg->major != FUSE_KERNEL_VERSION)
1147 		ok = false;
1148 	else {
1149 		unsigned long ra_pages;
1150 
1151 		process_init_limits(fc, arg);
1152 
1153 		if (arg->minor >= 6) {
1154 			u64 flags = arg->flags;
1155 
1156 			if (flags & FUSE_INIT_EXT)
1157 				flags |= (u64) arg->flags2 << 32;
1158 
1159 			ra_pages = arg->max_readahead / PAGE_SIZE;
1160 			if (flags & FUSE_ASYNC_READ)
1161 				fc->async_read = 1;
1162 			if (!(flags & FUSE_POSIX_LOCKS))
1163 				fc->no_lock = 1;
1164 			if (arg->minor >= 17) {
1165 				if (!(flags & FUSE_FLOCK_LOCKS))
1166 					fc->no_flock = 1;
1167 			} else {
1168 				if (!(flags & FUSE_POSIX_LOCKS))
1169 					fc->no_flock = 1;
1170 			}
1171 			if (flags & FUSE_ATOMIC_O_TRUNC)
1172 				fc->atomic_o_trunc = 1;
1173 			if (arg->minor >= 9) {
1174 				/* LOOKUP has dependency on proto version */
1175 				if (flags & FUSE_EXPORT_SUPPORT)
1176 					fc->export_support = 1;
1177 			}
1178 			if (flags & FUSE_BIG_WRITES)
1179 				fc->big_writes = 1;
1180 			if (flags & FUSE_DONT_MASK)
1181 				fc->dont_mask = 1;
1182 			if (flags & FUSE_AUTO_INVAL_DATA)
1183 				fc->auto_inval_data = 1;
1184 			else if (flags & FUSE_EXPLICIT_INVAL_DATA)
1185 				fc->explicit_inval_data = 1;
1186 			if (flags & FUSE_DO_READDIRPLUS) {
1187 				fc->do_readdirplus = 1;
1188 				if (flags & FUSE_READDIRPLUS_AUTO)
1189 					fc->readdirplus_auto = 1;
1190 			}
1191 			if (flags & FUSE_ASYNC_DIO)
1192 				fc->async_dio = 1;
1193 			if (flags & FUSE_WRITEBACK_CACHE)
1194 				fc->writeback_cache = 1;
1195 			if (flags & FUSE_PARALLEL_DIROPS)
1196 				fc->parallel_dirops = 1;
1197 			if (flags & FUSE_HANDLE_KILLPRIV)
1198 				fc->handle_killpriv = 1;
1199 			if (arg->time_gran && arg->time_gran <= 1000000000)
1200 				fm->sb->s_time_gran = arg->time_gran;
1201 			if ((flags & FUSE_POSIX_ACL)) {
1202 				fc->default_permissions = 1;
1203 				fc->posix_acl = 1;
1204 			}
1205 			if (flags & FUSE_CACHE_SYMLINKS)
1206 				fc->cache_symlinks = 1;
1207 			if (flags & FUSE_ABORT_ERROR)
1208 				fc->abort_err = 1;
1209 			if (flags & FUSE_MAX_PAGES) {
1210 				fc->max_pages =
1211 					min_t(unsigned int, fc->max_pages_limit,
1212 					max_t(unsigned int, arg->max_pages, 1));
1213 			}
1214 			if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1215 				if (flags & FUSE_MAP_ALIGNMENT &&
1216 				    !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1217 					ok = false;
1218 				}
1219 				if (flags & FUSE_HAS_INODE_DAX)
1220 					fc->inode_dax = 1;
1221 			}
1222 			if (flags & FUSE_HANDLE_KILLPRIV_V2) {
1223 				fc->handle_killpriv_v2 = 1;
1224 				fm->sb->s_flags |= SB_NOSEC;
1225 			}
1226 			if (flags & FUSE_SETXATTR_EXT)
1227 				fc->setxattr_ext = 1;
1228 			if (flags & FUSE_SECURITY_CTX)
1229 				fc->init_security = 1;
1230 			if (flags & FUSE_CREATE_SUPP_GROUP)
1231 				fc->create_supp_group = 1;
1232 			if (flags & FUSE_DIRECT_IO_RELAX)
1233 				fc->direct_io_relax = 1;
1234 		} else {
1235 			ra_pages = fc->max_read / PAGE_SIZE;
1236 			fc->no_lock = 1;
1237 			fc->no_flock = 1;
1238 		}
1239 
1240 		fm->sb->s_bdi->ra_pages =
1241 				min(fm->sb->s_bdi->ra_pages, ra_pages);
1242 		fc->minor = arg->minor;
1243 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1244 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
1245 		fc->conn_init = 1;
1246 	}
1247 	kfree(ia);
1248 
1249 	if (!ok) {
1250 		fc->conn_init = 0;
1251 		fc->conn_error = 1;
1252 	}
1253 
1254 	fuse_set_initialized(fc);
1255 	wake_up_all(&fc->blocked_waitq);
1256 }
1257 
1258 void fuse_send_init(struct fuse_mount *fm)
1259 {
1260 	struct fuse_init_args *ia;
1261 	u64 flags;
1262 
1263 	ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1264 
1265 	ia->in.major = FUSE_KERNEL_VERSION;
1266 	ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1267 	ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1268 	flags =
1269 		FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1270 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1271 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1272 		FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1273 		FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1274 		FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1275 		FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1276 		FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1277 		FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1278 		FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT | FUSE_INIT_EXT |
1279 		FUSE_SECURITY_CTX | FUSE_CREATE_SUPP_GROUP |
1280 		FUSE_HAS_EXPIRE_ONLY | FUSE_DIRECT_IO_RELAX;
1281 #ifdef CONFIG_FUSE_DAX
1282 	if (fm->fc->dax)
1283 		flags |= FUSE_MAP_ALIGNMENT;
1284 	if (fuse_is_inode_dax_mode(fm->fc->dax_mode))
1285 		flags |= FUSE_HAS_INODE_DAX;
1286 #endif
1287 	if (fm->fc->auto_submounts)
1288 		flags |= FUSE_SUBMOUNTS;
1289 
1290 	ia->in.flags = flags;
1291 	ia->in.flags2 = flags >> 32;
1292 
1293 	ia->args.opcode = FUSE_INIT;
1294 	ia->args.in_numargs = 1;
1295 	ia->args.in_args[0].size = sizeof(ia->in);
1296 	ia->args.in_args[0].value = &ia->in;
1297 	ia->args.out_numargs = 1;
1298 	/* Variable length argument used for backward compatibility
1299 	   with interface version < 7.5.  Rest of init_out is zeroed
1300 	   by do_get_request(), so a short reply is not a problem */
1301 	ia->args.out_argvar = true;
1302 	ia->args.out_args[0].size = sizeof(ia->out);
1303 	ia->args.out_args[0].value = &ia->out;
1304 	ia->args.force = true;
1305 	ia->args.nocreds = true;
1306 	ia->args.end = process_init_reply;
1307 
1308 	if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1309 		process_init_reply(fm, &ia->args, -ENOTCONN);
1310 }
1311 EXPORT_SYMBOL_GPL(fuse_send_init);
1312 
1313 void fuse_free_conn(struct fuse_conn *fc)
1314 {
1315 	WARN_ON(!list_empty(&fc->devices));
1316 	kfree_rcu(fc, rcu);
1317 }
1318 EXPORT_SYMBOL_GPL(fuse_free_conn);
1319 
1320 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1321 {
1322 	int err;
1323 	char *suffix = "";
1324 
1325 	if (sb->s_bdev) {
1326 		suffix = "-fuseblk";
1327 		/*
1328 		 * sb->s_bdi points to blkdev's bdi however we want to redirect
1329 		 * it to our private bdi...
1330 		 */
1331 		bdi_put(sb->s_bdi);
1332 		sb->s_bdi = &noop_backing_dev_info;
1333 	}
1334 	err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1335 				   MINOR(fc->dev), suffix);
1336 	if (err)
1337 		return err;
1338 
1339 	/* fuse does it's own writeback accounting */
1340 	sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1341 	sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1342 
1343 	/*
1344 	 * For a single fuse filesystem use max 1% of dirty +
1345 	 * writeback threshold.
1346 	 *
1347 	 * This gives about 1M of write buffer for memory maps on a
1348 	 * machine with 1G and 10% dirty_ratio, which should be more
1349 	 * than enough.
1350 	 *
1351 	 * Privileged users can raise it by writing to
1352 	 *
1353 	 *    /sys/class/bdi/<bdi>/max_ratio
1354 	 */
1355 	bdi_set_max_ratio(sb->s_bdi, 1);
1356 
1357 	return 0;
1358 }
1359 
1360 struct fuse_dev *fuse_dev_alloc(void)
1361 {
1362 	struct fuse_dev *fud;
1363 	struct list_head *pq;
1364 
1365 	fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1366 	if (!fud)
1367 		return NULL;
1368 
1369 	pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1370 	if (!pq) {
1371 		kfree(fud);
1372 		return NULL;
1373 	}
1374 
1375 	fud->pq.processing = pq;
1376 	fuse_pqueue_init(&fud->pq);
1377 
1378 	return fud;
1379 }
1380 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1381 
1382 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1383 {
1384 	fud->fc = fuse_conn_get(fc);
1385 	spin_lock(&fc->lock);
1386 	list_add_tail(&fud->entry, &fc->devices);
1387 	spin_unlock(&fc->lock);
1388 }
1389 EXPORT_SYMBOL_GPL(fuse_dev_install);
1390 
1391 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1392 {
1393 	struct fuse_dev *fud;
1394 
1395 	fud = fuse_dev_alloc();
1396 	if (!fud)
1397 		return NULL;
1398 
1399 	fuse_dev_install(fud, fc);
1400 	return fud;
1401 }
1402 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1403 
1404 void fuse_dev_free(struct fuse_dev *fud)
1405 {
1406 	struct fuse_conn *fc = fud->fc;
1407 
1408 	if (fc) {
1409 		spin_lock(&fc->lock);
1410 		list_del(&fud->entry);
1411 		spin_unlock(&fc->lock);
1412 
1413 		fuse_conn_put(fc);
1414 	}
1415 	kfree(fud->pq.processing);
1416 	kfree(fud);
1417 }
1418 EXPORT_SYMBOL_GPL(fuse_dev_free);
1419 
1420 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1421 				      const struct fuse_inode *fi)
1422 {
1423 	struct timespec64 atime = inode_get_atime(&fi->inode);
1424 	struct timespec64 mtime = inode_get_mtime(&fi->inode);
1425 	struct timespec64 ctime = inode_get_ctime(&fi->inode);
1426 
1427 	*attr = (struct fuse_attr){
1428 		.ino		= fi->inode.i_ino,
1429 		.size		= fi->inode.i_size,
1430 		.blocks		= fi->inode.i_blocks,
1431 		.atime		= atime.tv_sec,
1432 		.mtime		= mtime.tv_sec,
1433 		.ctime		= ctime.tv_sec,
1434 		.atimensec	= atime.tv_nsec,
1435 		.mtimensec	= mtime.tv_nsec,
1436 		.ctimensec	= ctime.tv_nsec,
1437 		.mode		= fi->inode.i_mode,
1438 		.nlink		= fi->inode.i_nlink,
1439 		.uid		= fi->inode.i_uid.val,
1440 		.gid		= fi->inode.i_gid.val,
1441 		.rdev		= fi->inode.i_rdev,
1442 		.blksize	= 1u << fi->inode.i_blkbits,
1443 	};
1444 }
1445 
1446 static void fuse_sb_defaults(struct super_block *sb)
1447 {
1448 	sb->s_magic = FUSE_SUPER_MAGIC;
1449 	sb->s_op = &fuse_super_operations;
1450 	sb->s_xattr = fuse_xattr_handlers;
1451 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1452 	sb->s_time_gran = 1;
1453 	sb->s_export_op = &fuse_export_operations;
1454 	sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1455 	if (sb->s_user_ns != &init_user_ns)
1456 		sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1457 	sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1458 }
1459 
1460 static int fuse_fill_super_submount(struct super_block *sb,
1461 				    struct fuse_inode *parent_fi)
1462 {
1463 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1464 	struct super_block *parent_sb = parent_fi->inode.i_sb;
1465 	struct fuse_attr root_attr;
1466 	struct inode *root;
1467 
1468 	fuse_sb_defaults(sb);
1469 	fm->sb = sb;
1470 
1471 	WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1472 	sb->s_bdi = bdi_get(parent_sb->s_bdi);
1473 
1474 	sb->s_xattr = parent_sb->s_xattr;
1475 	sb->s_time_gran = parent_sb->s_time_gran;
1476 	sb->s_blocksize = parent_sb->s_blocksize;
1477 	sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1478 	sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1479 	if (parent_sb->s_subtype && !sb->s_subtype)
1480 		return -ENOMEM;
1481 
1482 	fuse_fill_attr_from_inode(&root_attr, parent_fi);
1483 	root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1484 	/*
1485 	 * This inode is just a duplicate, so it is not looked up and
1486 	 * its nlookup should not be incremented.  fuse_iget() does
1487 	 * that, though, so undo it here.
1488 	 */
1489 	get_fuse_inode(root)->nlookup--;
1490 	sb->s_d_op = &fuse_dentry_operations;
1491 	sb->s_root = d_make_root(root);
1492 	if (!sb->s_root)
1493 		return -ENOMEM;
1494 
1495 	return 0;
1496 }
1497 
1498 /* Filesystem context private data holds the FUSE inode of the mount point */
1499 static int fuse_get_tree_submount(struct fs_context *fsc)
1500 {
1501 	struct fuse_mount *fm;
1502 	struct fuse_inode *mp_fi = fsc->fs_private;
1503 	struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1504 	struct super_block *sb;
1505 	int err;
1506 
1507 	fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1508 	if (!fm)
1509 		return -ENOMEM;
1510 
1511 	fm->fc = fuse_conn_get(fc);
1512 	fsc->s_fs_info = fm;
1513 	sb = sget_fc(fsc, NULL, set_anon_super_fc);
1514 	if (fsc->s_fs_info)
1515 		fuse_mount_destroy(fm);
1516 	if (IS_ERR(sb))
1517 		return PTR_ERR(sb);
1518 
1519 	/* Initialize superblock, making @mp_fi its root */
1520 	err = fuse_fill_super_submount(sb, mp_fi);
1521 	if (err) {
1522 		deactivate_locked_super(sb);
1523 		return err;
1524 	}
1525 
1526 	down_write(&fc->killsb);
1527 	list_add_tail(&fm->fc_entry, &fc->mounts);
1528 	up_write(&fc->killsb);
1529 
1530 	sb->s_flags |= SB_ACTIVE;
1531 	fsc->root = dget(sb->s_root);
1532 
1533 	return 0;
1534 }
1535 
1536 static const struct fs_context_operations fuse_context_submount_ops = {
1537 	.get_tree	= fuse_get_tree_submount,
1538 };
1539 
1540 int fuse_init_fs_context_submount(struct fs_context *fsc)
1541 {
1542 	fsc->ops = &fuse_context_submount_ops;
1543 	return 0;
1544 }
1545 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1546 
1547 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1548 {
1549 	struct fuse_dev *fud = NULL;
1550 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1551 	struct fuse_conn *fc = fm->fc;
1552 	struct inode *root;
1553 	struct dentry *root_dentry;
1554 	int err;
1555 
1556 	err = -EINVAL;
1557 	if (sb->s_flags & SB_MANDLOCK)
1558 		goto err;
1559 
1560 	rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1561 	fuse_sb_defaults(sb);
1562 
1563 	if (ctx->is_bdev) {
1564 #ifdef CONFIG_BLOCK
1565 		err = -EINVAL;
1566 		if (!sb_set_blocksize(sb, ctx->blksize))
1567 			goto err;
1568 #endif
1569 	} else {
1570 		sb->s_blocksize = PAGE_SIZE;
1571 		sb->s_blocksize_bits = PAGE_SHIFT;
1572 	}
1573 
1574 	sb->s_subtype = ctx->subtype;
1575 	ctx->subtype = NULL;
1576 	if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1577 		err = fuse_dax_conn_alloc(fc, ctx->dax_mode, ctx->dax_dev);
1578 		if (err)
1579 			goto err;
1580 	}
1581 
1582 	if (ctx->fudptr) {
1583 		err = -ENOMEM;
1584 		fud = fuse_dev_alloc_install(fc);
1585 		if (!fud)
1586 			goto err_free_dax;
1587 	}
1588 
1589 	fc->dev = sb->s_dev;
1590 	fm->sb = sb;
1591 	err = fuse_bdi_init(fc, sb);
1592 	if (err)
1593 		goto err_dev_free;
1594 
1595 	/* Handle umasking inside the fuse code */
1596 	if (sb->s_flags & SB_POSIXACL)
1597 		fc->dont_mask = 1;
1598 	sb->s_flags |= SB_POSIXACL;
1599 
1600 	fc->default_permissions = ctx->default_permissions;
1601 	fc->allow_other = ctx->allow_other;
1602 	fc->user_id = ctx->user_id;
1603 	fc->group_id = ctx->group_id;
1604 	fc->legacy_opts_show = ctx->legacy_opts_show;
1605 	fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1606 	fc->destroy = ctx->destroy;
1607 	fc->no_control = ctx->no_control;
1608 	fc->no_force_umount = ctx->no_force_umount;
1609 
1610 	err = -ENOMEM;
1611 	root = fuse_get_root_inode(sb, ctx->rootmode);
1612 	sb->s_d_op = &fuse_root_dentry_operations;
1613 	root_dentry = d_make_root(root);
1614 	if (!root_dentry)
1615 		goto err_dev_free;
1616 	/* Root dentry doesn't have .d_revalidate */
1617 	sb->s_d_op = &fuse_dentry_operations;
1618 
1619 	mutex_lock(&fuse_mutex);
1620 	err = -EINVAL;
1621 	if (ctx->fudptr && *ctx->fudptr)
1622 		goto err_unlock;
1623 
1624 	err = fuse_ctl_add_conn(fc);
1625 	if (err)
1626 		goto err_unlock;
1627 
1628 	list_add_tail(&fc->entry, &fuse_conn_list);
1629 	sb->s_root = root_dentry;
1630 	if (ctx->fudptr)
1631 		*ctx->fudptr = fud;
1632 	mutex_unlock(&fuse_mutex);
1633 	return 0;
1634 
1635  err_unlock:
1636 	mutex_unlock(&fuse_mutex);
1637 	dput(root_dentry);
1638  err_dev_free:
1639 	if (fud)
1640 		fuse_dev_free(fud);
1641  err_free_dax:
1642 	if (IS_ENABLED(CONFIG_FUSE_DAX))
1643 		fuse_dax_conn_free(fc);
1644  err:
1645 	return err;
1646 }
1647 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1648 
1649 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1650 {
1651 	struct fuse_fs_context *ctx = fsc->fs_private;
1652 	int err;
1653 
1654 	if (!ctx->file || !ctx->rootmode_present ||
1655 	    !ctx->user_id_present || !ctx->group_id_present)
1656 		return -EINVAL;
1657 
1658 	/*
1659 	 * Require mount to happen from the same user namespace which
1660 	 * opened /dev/fuse to prevent potential attacks.
1661 	 */
1662 	if ((ctx->file->f_op != &fuse_dev_operations) ||
1663 	    (ctx->file->f_cred->user_ns != sb->s_user_ns))
1664 		return -EINVAL;
1665 	ctx->fudptr = &ctx->file->private_data;
1666 
1667 	err = fuse_fill_super_common(sb, ctx);
1668 	if (err)
1669 		return err;
1670 	/* file->private_data shall be visible on all CPUs after this */
1671 	smp_mb();
1672 	fuse_send_init(get_fuse_mount_super(sb));
1673 	return 0;
1674 }
1675 
1676 /*
1677  * This is the path where user supplied an already initialized fuse dev.  In
1678  * this case never create a new super if the old one is gone.
1679  */
1680 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1681 {
1682 	return -ENOTCONN;
1683 }
1684 
1685 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1686 {
1687 
1688 	return fsc->sget_key == get_fuse_conn_super(sb);
1689 }
1690 
1691 static int fuse_get_tree(struct fs_context *fsc)
1692 {
1693 	struct fuse_fs_context *ctx = fsc->fs_private;
1694 	struct fuse_dev *fud;
1695 	struct fuse_conn *fc;
1696 	struct fuse_mount *fm;
1697 	struct super_block *sb;
1698 	int err;
1699 
1700 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1701 	if (!fc)
1702 		return -ENOMEM;
1703 
1704 	fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1705 	if (!fm) {
1706 		kfree(fc);
1707 		return -ENOMEM;
1708 	}
1709 
1710 	fuse_conn_init(fc, fm, fsc->user_ns, &fuse_dev_fiq_ops, NULL);
1711 	fc->release = fuse_free_conn;
1712 
1713 	fsc->s_fs_info = fm;
1714 
1715 	if (ctx->fd_present)
1716 		ctx->file = fget(ctx->fd);
1717 
1718 	if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1719 		err = get_tree_bdev(fsc, fuse_fill_super);
1720 		goto out;
1721 	}
1722 	/*
1723 	 * While block dev mount can be initialized with a dummy device fd
1724 	 * (found by device name), normal fuse mounts can't
1725 	 */
1726 	err = -EINVAL;
1727 	if (!ctx->file)
1728 		goto out;
1729 
1730 	/*
1731 	 * Allow creating a fuse mount with an already initialized fuse
1732 	 * connection
1733 	 */
1734 	fud = READ_ONCE(ctx->file->private_data);
1735 	if (ctx->file->f_op == &fuse_dev_operations && fud) {
1736 		fsc->sget_key = fud->fc;
1737 		sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1738 		err = PTR_ERR_OR_ZERO(sb);
1739 		if (!IS_ERR(sb))
1740 			fsc->root = dget(sb->s_root);
1741 	} else {
1742 		err = get_tree_nodev(fsc, fuse_fill_super);
1743 	}
1744 out:
1745 	if (fsc->s_fs_info)
1746 		fuse_mount_destroy(fm);
1747 	if (ctx->file)
1748 		fput(ctx->file);
1749 	return err;
1750 }
1751 
1752 static const struct fs_context_operations fuse_context_ops = {
1753 	.free		= fuse_free_fsc,
1754 	.parse_param	= fuse_parse_param,
1755 	.reconfigure	= fuse_reconfigure,
1756 	.get_tree	= fuse_get_tree,
1757 };
1758 
1759 /*
1760  * Set up the filesystem mount context.
1761  */
1762 static int fuse_init_fs_context(struct fs_context *fsc)
1763 {
1764 	struct fuse_fs_context *ctx;
1765 
1766 	ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1767 	if (!ctx)
1768 		return -ENOMEM;
1769 
1770 	ctx->max_read = ~0;
1771 	ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1772 	ctx->legacy_opts_show = true;
1773 
1774 #ifdef CONFIG_BLOCK
1775 	if (fsc->fs_type == &fuseblk_fs_type) {
1776 		ctx->is_bdev = true;
1777 		ctx->destroy = true;
1778 	}
1779 #endif
1780 
1781 	fsc->fs_private = ctx;
1782 	fsc->ops = &fuse_context_ops;
1783 	return 0;
1784 }
1785 
1786 bool fuse_mount_remove(struct fuse_mount *fm)
1787 {
1788 	struct fuse_conn *fc = fm->fc;
1789 	bool last = false;
1790 
1791 	down_write(&fc->killsb);
1792 	list_del_init(&fm->fc_entry);
1793 	if (list_empty(&fc->mounts))
1794 		last = true;
1795 	up_write(&fc->killsb);
1796 
1797 	return last;
1798 }
1799 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1800 
1801 void fuse_conn_destroy(struct fuse_mount *fm)
1802 {
1803 	struct fuse_conn *fc = fm->fc;
1804 
1805 	if (fc->destroy)
1806 		fuse_send_destroy(fm);
1807 
1808 	fuse_abort_conn(fc);
1809 	fuse_wait_aborted(fc);
1810 
1811 	if (!list_empty(&fc->entry)) {
1812 		mutex_lock(&fuse_mutex);
1813 		list_del(&fc->entry);
1814 		fuse_ctl_remove_conn(fc);
1815 		mutex_unlock(&fuse_mutex);
1816 	}
1817 }
1818 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1819 
1820 static void fuse_sb_destroy(struct super_block *sb)
1821 {
1822 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1823 	bool last;
1824 
1825 	if (sb->s_root) {
1826 		last = fuse_mount_remove(fm);
1827 		if (last)
1828 			fuse_conn_destroy(fm);
1829 	}
1830 }
1831 
1832 void fuse_mount_destroy(struct fuse_mount *fm)
1833 {
1834 	fuse_conn_put(fm->fc);
1835 	kfree(fm);
1836 }
1837 EXPORT_SYMBOL(fuse_mount_destroy);
1838 
1839 static void fuse_kill_sb_anon(struct super_block *sb)
1840 {
1841 	fuse_sb_destroy(sb);
1842 	kill_anon_super(sb);
1843 	fuse_mount_destroy(get_fuse_mount_super(sb));
1844 }
1845 
1846 static struct file_system_type fuse_fs_type = {
1847 	.owner		= THIS_MODULE,
1848 	.name		= "fuse",
1849 	.fs_flags	= FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1850 	.init_fs_context = fuse_init_fs_context,
1851 	.parameters	= fuse_fs_parameters,
1852 	.kill_sb	= fuse_kill_sb_anon,
1853 };
1854 MODULE_ALIAS_FS("fuse");
1855 
1856 #ifdef CONFIG_BLOCK
1857 static void fuse_kill_sb_blk(struct super_block *sb)
1858 {
1859 	fuse_sb_destroy(sb);
1860 	kill_block_super(sb);
1861 	fuse_mount_destroy(get_fuse_mount_super(sb));
1862 }
1863 
1864 static struct file_system_type fuseblk_fs_type = {
1865 	.owner		= THIS_MODULE,
1866 	.name		= "fuseblk",
1867 	.init_fs_context = fuse_init_fs_context,
1868 	.parameters	= fuse_fs_parameters,
1869 	.kill_sb	= fuse_kill_sb_blk,
1870 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1871 };
1872 MODULE_ALIAS_FS("fuseblk");
1873 
1874 static inline int register_fuseblk(void)
1875 {
1876 	return register_filesystem(&fuseblk_fs_type);
1877 }
1878 
1879 static inline void unregister_fuseblk(void)
1880 {
1881 	unregister_filesystem(&fuseblk_fs_type);
1882 }
1883 #else
1884 static inline int register_fuseblk(void)
1885 {
1886 	return 0;
1887 }
1888 
1889 static inline void unregister_fuseblk(void)
1890 {
1891 }
1892 #endif
1893 
1894 static void fuse_inode_init_once(void *foo)
1895 {
1896 	struct inode *inode = foo;
1897 
1898 	inode_init_once(inode);
1899 }
1900 
1901 static int __init fuse_fs_init(void)
1902 {
1903 	int err;
1904 
1905 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
1906 			sizeof(struct fuse_inode), 0,
1907 			SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1908 			fuse_inode_init_once);
1909 	err = -ENOMEM;
1910 	if (!fuse_inode_cachep)
1911 		goto out;
1912 
1913 	err = register_fuseblk();
1914 	if (err)
1915 		goto out2;
1916 
1917 	err = register_filesystem(&fuse_fs_type);
1918 	if (err)
1919 		goto out3;
1920 
1921 	return 0;
1922 
1923  out3:
1924 	unregister_fuseblk();
1925  out2:
1926 	kmem_cache_destroy(fuse_inode_cachep);
1927  out:
1928 	return err;
1929 }
1930 
1931 static void fuse_fs_cleanup(void)
1932 {
1933 	unregister_filesystem(&fuse_fs_type);
1934 	unregister_fuseblk();
1935 
1936 	/*
1937 	 * Make sure all delayed rcu free inodes are flushed before we
1938 	 * destroy cache.
1939 	 */
1940 	rcu_barrier();
1941 	kmem_cache_destroy(fuse_inode_cachep);
1942 }
1943 
1944 static struct kobject *fuse_kobj;
1945 
1946 static int fuse_sysfs_init(void)
1947 {
1948 	int err;
1949 
1950 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1951 	if (!fuse_kobj) {
1952 		err = -ENOMEM;
1953 		goto out_err;
1954 	}
1955 
1956 	err = sysfs_create_mount_point(fuse_kobj, "connections");
1957 	if (err)
1958 		goto out_fuse_unregister;
1959 
1960 	return 0;
1961 
1962  out_fuse_unregister:
1963 	kobject_put(fuse_kobj);
1964  out_err:
1965 	return err;
1966 }
1967 
1968 static void fuse_sysfs_cleanup(void)
1969 {
1970 	sysfs_remove_mount_point(fuse_kobj, "connections");
1971 	kobject_put(fuse_kobj);
1972 }
1973 
1974 static int __init fuse_init(void)
1975 {
1976 	int res;
1977 
1978 	pr_info("init (API version %i.%i)\n",
1979 		FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1980 
1981 	INIT_LIST_HEAD(&fuse_conn_list);
1982 	res = fuse_fs_init();
1983 	if (res)
1984 		goto err;
1985 
1986 	res = fuse_dev_init();
1987 	if (res)
1988 		goto err_fs_cleanup;
1989 
1990 	res = fuse_sysfs_init();
1991 	if (res)
1992 		goto err_dev_cleanup;
1993 
1994 	res = fuse_ctl_init();
1995 	if (res)
1996 		goto err_sysfs_cleanup;
1997 
1998 	sanitize_global_limit(&max_user_bgreq);
1999 	sanitize_global_limit(&max_user_congthresh);
2000 
2001 	return 0;
2002 
2003  err_sysfs_cleanup:
2004 	fuse_sysfs_cleanup();
2005  err_dev_cleanup:
2006 	fuse_dev_cleanup();
2007  err_fs_cleanup:
2008 	fuse_fs_cleanup();
2009  err:
2010 	return res;
2011 }
2012 
2013 static void __exit fuse_exit(void)
2014 {
2015 	pr_debug("exit\n");
2016 
2017 	fuse_ctl_cleanup();
2018 	fuse_sysfs_cleanup();
2019 	fuse_fs_cleanup();
2020 	fuse_dev_cleanup();
2021 }
2022 
2023 module_init(fuse_init);
2024 module_exit(fuse_exit);
2025