xref: /linux/fs/fuse/inode.c (revision db10cb9b)
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->i_atime.tv_sec   = attr->atime;
192 	inode->i_atime.tv_nsec  = attr->atimensec;
193 	/* mtime from server may be stale due to local buffered write */
194 	if (!(cache_mask & STATX_MTIME)) {
195 		inode->i_mtime.tv_sec   = attr->mtime;
196 		inode->i_mtime.tv_nsec  = attr->mtimensec;
197 	}
198 	if (!(cache_mask & STATX_CTIME)) {
199 		inode_set_ctime(inode, attr->ctime, attr->ctimensec);
200 	}
201 	if (sx) {
202 		/* Sanitize nsecs */
203 		sx->btime.tv_nsec =
204 			min_t(u32, sx->btime.tv_nsec, NSEC_PER_SEC - 1);
205 
206 		/*
207 		 * Btime has been queried, cache is valid (whether or not btime
208 		 * is available or not) so clear STATX_BTIME from inval_mask.
209 		 *
210 		 * Availability of the btime attribute is indicated in
211 		 * FUSE_I_BTIME
212 		 */
213 		set_mask_bits(&fi->inval_mask, STATX_BTIME, 0);
214 		if (sx->mask & STATX_BTIME) {
215 			set_bit(FUSE_I_BTIME, &fi->state);
216 			fi->i_btime.tv_sec = sx->btime.tv_sec;
217 			fi->i_btime.tv_nsec = sx->btime.tv_nsec;
218 		}
219 	}
220 
221 	if (attr->blksize != 0)
222 		inode->i_blkbits = ilog2(attr->blksize);
223 	else
224 		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
225 
226 	/*
227 	 * Don't set the sticky bit in i_mode, unless we want the VFS
228 	 * to check permissions.  This prevents failures due to the
229 	 * check in may_delete().
230 	 */
231 	fi->orig_i_mode = inode->i_mode;
232 	if (!fc->default_permissions)
233 		inode->i_mode &= ~S_ISVTX;
234 
235 	fi->orig_ino = attr->ino;
236 
237 	/*
238 	 * We are refreshing inode data and it is possible that another
239 	 * client set suid/sgid or security.capability xattr. So clear
240 	 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
241 	 * was set or if security.capability xattr was set. But we don't
242 	 * know if security.capability has been set or not. So clear it
243 	 * anyway. Its less efficient but should be safe.
244 	 */
245 	inode->i_flags &= ~S_NOSEC;
246 }
247 
248 u32 fuse_get_cache_mask(struct inode *inode)
249 {
250 	struct fuse_conn *fc = get_fuse_conn(inode);
251 
252 	if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
253 		return 0;
254 
255 	return STATX_MTIME | STATX_CTIME | STATX_SIZE;
256 }
257 
258 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
259 			    struct fuse_statx *sx,
260 			    u64 attr_valid, u64 attr_version)
261 {
262 	struct fuse_conn *fc = get_fuse_conn(inode);
263 	struct fuse_inode *fi = get_fuse_inode(inode);
264 	u32 cache_mask;
265 	loff_t oldsize;
266 	struct timespec64 old_mtime;
267 
268 	spin_lock(&fi->lock);
269 	/*
270 	 * In case of writeback_cache enabled, writes update mtime, ctime and
271 	 * may update i_size.  In these cases trust the cached value in the
272 	 * inode.
273 	 */
274 	cache_mask = fuse_get_cache_mask(inode);
275 	if (cache_mask & STATX_SIZE)
276 		attr->size = i_size_read(inode);
277 
278 	if (cache_mask & STATX_MTIME) {
279 		attr->mtime = inode->i_mtime.tv_sec;
280 		attr->mtimensec = inode->i_mtime.tv_nsec;
281 	}
282 	if (cache_mask & STATX_CTIME) {
283 		attr->ctime = inode_get_ctime(inode).tv_sec;
284 		attr->ctimensec = inode_get_ctime(inode).tv_nsec;
285 	}
286 
287 	if ((attr_version != 0 && fi->attr_version > attr_version) ||
288 	    test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
289 		spin_unlock(&fi->lock);
290 		return;
291 	}
292 
293 	old_mtime = inode->i_mtime;
294 	fuse_change_attributes_common(inode, attr, sx, attr_valid, cache_mask);
295 
296 	oldsize = inode->i_size;
297 	/*
298 	 * In case of writeback_cache enabled, the cached writes beyond EOF
299 	 * extend local i_size without keeping userspace server in sync. So,
300 	 * attr->size coming from server can be stale. We cannot trust it.
301 	 */
302 	if (!(cache_mask & STATX_SIZE))
303 		i_size_write(inode, attr->size);
304 	spin_unlock(&fi->lock);
305 
306 	if (!cache_mask && S_ISREG(inode->i_mode)) {
307 		bool inval = false;
308 
309 		if (oldsize != attr->size) {
310 			truncate_pagecache(inode, attr->size);
311 			if (!fc->explicit_inval_data)
312 				inval = true;
313 		} else if (fc->auto_inval_data) {
314 			struct timespec64 new_mtime = {
315 				.tv_sec = attr->mtime,
316 				.tv_nsec = attr->mtimensec,
317 			};
318 
319 			/*
320 			 * Auto inval mode also checks and invalidates if mtime
321 			 * has changed.
322 			 */
323 			if (!timespec64_equal(&old_mtime, &new_mtime))
324 				inval = true;
325 		}
326 
327 		if (inval)
328 			invalidate_inode_pages2(inode->i_mapping);
329 	}
330 
331 	if (IS_ENABLED(CONFIG_FUSE_DAX))
332 		fuse_dax_dontcache(inode, attr->flags);
333 }
334 
335 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr,
336 			    struct fuse_conn *fc)
337 {
338 	inode->i_mode = attr->mode & S_IFMT;
339 	inode->i_size = attr->size;
340 	inode->i_mtime.tv_sec  = attr->mtime;
341 	inode->i_mtime.tv_nsec = attr->mtimensec;
342 	inode_set_ctime(inode, attr->ctime, attr->ctimensec);
343 	if (S_ISREG(inode->i_mode)) {
344 		fuse_init_common(inode);
345 		fuse_init_file_inode(inode, attr->flags);
346 	} else if (S_ISDIR(inode->i_mode))
347 		fuse_init_dir(inode);
348 	else if (S_ISLNK(inode->i_mode))
349 		fuse_init_symlink(inode);
350 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
351 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
352 		fuse_init_common(inode);
353 		init_special_inode(inode, inode->i_mode,
354 				   new_decode_dev(attr->rdev));
355 	} else
356 		BUG();
357 	/*
358 	 * Ensure that we don't cache acls for daemons without FUSE_POSIX_ACL
359 	 * so they see the exact same behavior as before.
360 	 */
361 	if (!fc->posix_acl)
362 		inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
363 }
364 
365 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
366 {
367 	u64 nodeid = *(u64 *) _nodeidp;
368 	if (get_node_id(inode) == nodeid)
369 		return 1;
370 	else
371 		return 0;
372 }
373 
374 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
375 {
376 	u64 nodeid = *(u64 *) _nodeidp;
377 	get_fuse_inode(inode)->nodeid = nodeid;
378 	return 0;
379 }
380 
381 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
382 			int generation, struct fuse_attr *attr,
383 			u64 attr_valid, u64 attr_version)
384 {
385 	struct inode *inode;
386 	struct fuse_inode *fi;
387 	struct fuse_conn *fc = get_fuse_conn_super(sb);
388 
389 	/*
390 	 * Auto mount points get their node id from the submount root, which is
391 	 * not a unique identifier within this filesystem.
392 	 *
393 	 * To avoid conflicts, do not place submount points into the inode hash
394 	 * table.
395 	 */
396 	if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
397 	    S_ISDIR(attr->mode)) {
398 		inode = new_inode(sb);
399 		if (!inode)
400 			return NULL;
401 
402 		fuse_init_inode(inode, attr, fc);
403 		get_fuse_inode(inode)->nodeid = nodeid;
404 		inode->i_flags |= S_AUTOMOUNT;
405 		goto done;
406 	}
407 
408 retry:
409 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
410 	if (!inode)
411 		return NULL;
412 
413 	if ((inode->i_state & I_NEW)) {
414 		inode->i_flags |= S_NOATIME;
415 		if (!fc->writeback_cache || !S_ISREG(attr->mode))
416 			inode->i_flags |= S_NOCMTIME;
417 		inode->i_generation = generation;
418 		fuse_init_inode(inode, attr, fc);
419 		unlock_new_inode(inode);
420 	} else if (fuse_stale_inode(inode, generation, attr)) {
421 		/* nodeid was reused, any I/O on the old inode should fail */
422 		fuse_make_bad(inode);
423 		iput(inode);
424 		goto retry;
425 	}
426 done:
427 	fi = get_fuse_inode(inode);
428 	spin_lock(&fi->lock);
429 	fi->nlookup++;
430 	spin_unlock(&fi->lock);
431 	fuse_change_attributes(inode, attr, NULL, attr_valid, attr_version);
432 
433 	return inode;
434 }
435 
436 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
437 			   struct fuse_mount **fm)
438 {
439 	struct fuse_mount *fm_iter;
440 	struct inode *inode;
441 
442 	WARN_ON(!rwsem_is_locked(&fc->killsb));
443 	list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
444 		if (!fm_iter->sb)
445 			continue;
446 
447 		inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
448 		if (inode) {
449 			if (fm)
450 				*fm = fm_iter;
451 			return inode;
452 		}
453 	}
454 
455 	return NULL;
456 }
457 
458 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
459 			     loff_t offset, loff_t len)
460 {
461 	struct fuse_inode *fi;
462 	struct inode *inode;
463 	pgoff_t pg_start;
464 	pgoff_t pg_end;
465 
466 	inode = fuse_ilookup(fc, nodeid, NULL);
467 	if (!inode)
468 		return -ENOENT;
469 
470 	fi = get_fuse_inode(inode);
471 	spin_lock(&fi->lock);
472 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
473 	spin_unlock(&fi->lock);
474 
475 	fuse_invalidate_attr(inode);
476 	forget_all_cached_acls(inode);
477 	if (offset >= 0) {
478 		pg_start = offset >> PAGE_SHIFT;
479 		if (len <= 0)
480 			pg_end = -1;
481 		else
482 			pg_end = (offset + len - 1) >> PAGE_SHIFT;
483 		invalidate_inode_pages2_range(inode->i_mapping,
484 					      pg_start, pg_end);
485 	}
486 	iput(inode);
487 	return 0;
488 }
489 
490 bool fuse_lock_inode(struct inode *inode)
491 {
492 	bool locked = false;
493 
494 	if (!get_fuse_conn(inode)->parallel_dirops) {
495 		mutex_lock(&get_fuse_inode(inode)->mutex);
496 		locked = true;
497 	}
498 
499 	return locked;
500 }
501 
502 void fuse_unlock_inode(struct inode *inode, bool locked)
503 {
504 	if (locked)
505 		mutex_unlock(&get_fuse_inode(inode)->mutex);
506 }
507 
508 static void fuse_umount_begin(struct super_block *sb)
509 {
510 	struct fuse_conn *fc = get_fuse_conn_super(sb);
511 
512 	if (fc->no_force_umount)
513 		return;
514 
515 	fuse_abort_conn(fc);
516 
517 	// Only retire block-device-based superblocks.
518 	if (sb->s_bdev != NULL)
519 		retire_super(sb);
520 }
521 
522 static void fuse_send_destroy(struct fuse_mount *fm)
523 {
524 	if (fm->fc->conn_init) {
525 		FUSE_ARGS(args);
526 
527 		args.opcode = FUSE_DESTROY;
528 		args.force = true;
529 		args.nocreds = true;
530 		fuse_simple_request(fm, &args);
531 	}
532 }
533 
534 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
535 {
536 	stbuf->f_type    = FUSE_SUPER_MAGIC;
537 	stbuf->f_bsize   = attr->bsize;
538 	stbuf->f_frsize  = attr->frsize;
539 	stbuf->f_blocks  = attr->blocks;
540 	stbuf->f_bfree   = attr->bfree;
541 	stbuf->f_bavail  = attr->bavail;
542 	stbuf->f_files   = attr->files;
543 	stbuf->f_ffree   = attr->ffree;
544 	stbuf->f_namelen = attr->namelen;
545 	/* fsid is left zero */
546 }
547 
548 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
549 {
550 	struct super_block *sb = dentry->d_sb;
551 	struct fuse_mount *fm = get_fuse_mount_super(sb);
552 	FUSE_ARGS(args);
553 	struct fuse_statfs_out outarg;
554 	int err;
555 
556 	if (!fuse_allow_current_process(fm->fc)) {
557 		buf->f_type = FUSE_SUPER_MAGIC;
558 		return 0;
559 	}
560 
561 	memset(&outarg, 0, sizeof(outarg));
562 	args.in_numargs = 0;
563 	args.opcode = FUSE_STATFS;
564 	args.nodeid = get_node_id(d_inode(dentry));
565 	args.out_numargs = 1;
566 	args.out_args[0].size = sizeof(outarg);
567 	args.out_args[0].value = &outarg;
568 	err = fuse_simple_request(fm, &args);
569 	if (!err)
570 		convert_fuse_statfs(buf, &outarg.st);
571 	return err;
572 }
573 
574 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
575 {
576 	struct fuse_sync_bucket *bucket;
577 
578 	bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
579 	if (bucket) {
580 		init_waitqueue_head(&bucket->waitq);
581 		/* Initial active count */
582 		atomic_set(&bucket->count, 1);
583 	}
584 	return bucket;
585 }
586 
587 static void fuse_sync_fs_writes(struct fuse_conn *fc)
588 {
589 	struct fuse_sync_bucket *bucket, *new_bucket;
590 	int count;
591 
592 	new_bucket = fuse_sync_bucket_alloc();
593 	spin_lock(&fc->lock);
594 	bucket = rcu_dereference_protected(fc->curr_bucket, 1);
595 	count = atomic_read(&bucket->count);
596 	WARN_ON(count < 1);
597 	/* No outstanding writes? */
598 	if (count == 1) {
599 		spin_unlock(&fc->lock);
600 		kfree(new_bucket);
601 		return;
602 	}
603 
604 	/*
605 	 * Completion of new bucket depends on completion of this bucket, so add
606 	 * one more count.
607 	 */
608 	atomic_inc(&new_bucket->count);
609 	rcu_assign_pointer(fc->curr_bucket, new_bucket);
610 	spin_unlock(&fc->lock);
611 	/*
612 	 * Drop initial active count.  At this point if all writes in this and
613 	 * ancestor buckets complete, the count will go to zero and this task
614 	 * will be woken up.
615 	 */
616 	atomic_dec(&bucket->count);
617 
618 	wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
619 
620 	/* Drop temp count on descendant bucket */
621 	fuse_sync_bucket_dec(new_bucket);
622 	kfree_rcu(bucket, rcu);
623 }
624 
625 static int fuse_sync_fs(struct super_block *sb, int wait)
626 {
627 	struct fuse_mount *fm = get_fuse_mount_super(sb);
628 	struct fuse_conn *fc = fm->fc;
629 	struct fuse_syncfs_in inarg;
630 	FUSE_ARGS(args);
631 	int err;
632 
633 	/*
634 	 * Userspace cannot handle the wait == 0 case.  Avoid a
635 	 * gratuitous roundtrip.
636 	 */
637 	if (!wait)
638 		return 0;
639 
640 	/* The filesystem is being unmounted.  Nothing to do. */
641 	if (!sb->s_root)
642 		return 0;
643 
644 	if (!fc->sync_fs)
645 		return 0;
646 
647 	fuse_sync_fs_writes(fc);
648 
649 	memset(&inarg, 0, sizeof(inarg));
650 	args.in_numargs = 1;
651 	args.in_args[0].size = sizeof(inarg);
652 	args.in_args[0].value = &inarg;
653 	args.opcode = FUSE_SYNCFS;
654 	args.nodeid = get_node_id(sb->s_root->d_inode);
655 	args.out_numargs = 0;
656 
657 	err = fuse_simple_request(fm, &args);
658 	if (err == -ENOSYS) {
659 		fc->sync_fs = 0;
660 		err = 0;
661 	}
662 
663 	return err;
664 }
665 
666 enum {
667 	OPT_SOURCE,
668 	OPT_SUBTYPE,
669 	OPT_FD,
670 	OPT_ROOTMODE,
671 	OPT_USER_ID,
672 	OPT_GROUP_ID,
673 	OPT_DEFAULT_PERMISSIONS,
674 	OPT_ALLOW_OTHER,
675 	OPT_MAX_READ,
676 	OPT_BLKSIZE,
677 	OPT_ERR
678 };
679 
680 static const struct fs_parameter_spec fuse_fs_parameters[] = {
681 	fsparam_string	("source",		OPT_SOURCE),
682 	fsparam_u32	("fd",			OPT_FD),
683 	fsparam_u32oct	("rootmode",		OPT_ROOTMODE),
684 	fsparam_u32	("user_id",		OPT_USER_ID),
685 	fsparam_u32	("group_id",		OPT_GROUP_ID),
686 	fsparam_flag	("default_permissions",	OPT_DEFAULT_PERMISSIONS),
687 	fsparam_flag	("allow_other",		OPT_ALLOW_OTHER),
688 	fsparam_u32	("max_read",		OPT_MAX_READ),
689 	fsparam_u32	("blksize",		OPT_BLKSIZE),
690 	fsparam_string	("subtype",		OPT_SUBTYPE),
691 	{}
692 };
693 
694 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
695 {
696 	struct fs_parse_result result;
697 	struct fuse_fs_context *ctx = fsc->fs_private;
698 	int opt;
699 
700 	if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
701 		/*
702 		 * Ignore options coming from mount(MS_REMOUNT) for backward
703 		 * compatibility.
704 		 */
705 		if (fsc->oldapi)
706 			return 0;
707 
708 		return invalfc(fsc, "No changes allowed in reconfigure");
709 	}
710 
711 	opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
712 	if (opt < 0)
713 		return opt;
714 
715 	switch (opt) {
716 	case OPT_SOURCE:
717 		if (fsc->source)
718 			return invalfc(fsc, "Multiple sources specified");
719 		fsc->source = param->string;
720 		param->string = NULL;
721 		break;
722 
723 	case OPT_SUBTYPE:
724 		if (ctx->subtype)
725 			return invalfc(fsc, "Multiple subtypes specified");
726 		ctx->subtype = param->string;
727 		param->string = NULL;
728 		return 0;
729 
730 	case OPT_FD:
731 		ctx->fd = result.uint_32;
732 		ctx->fd_present = true;
733 		break;
734 
735 	case OPT_ROOTMODE:
736 		if (!fuse_valid_type(result.uint_32))
737 			return invalfc(fsc, "Invalid rootmode");
738 		ctx->rootmode = result.uint_32;
739 		ctx->rootmode_present = true;
740 		break;
741 
742 	case OPT_USER_ID:
743 		ctx->user_id = make_kuid(fsc->user_ns, result.uint_32);
744 		if (!uid_valid(ctx->user_id))
745 			return invalfc(fsc, "Invalid user_id");
746 		ctx->user_id_present = true;
747 		break;
748 
749 	case OPT_GROUP_ID:
750 		ctx->group_id = make_kgid(fsc->user_ns, result.uint_32);
751 		if (!gid_valid(ctx->group_id))
752 			return invalfc(fsc, "Invalid group_id");
753 		ctx->group_id_present = true;
754 		break;
755 
756 	case OPT_DEFAULT_PERMISSIONS:
757 		ctx->default_permissions = true;
758 		break;
759 
760 	case OPT_ALLOW_OTHER:
761 		ctx->allow_other = true;
762 		break;
763 
764 	case OPT_MAX_READ:
765 		ctx->max_read = result.uint_32;
766 		break;
767 
768 	case OPT_BLKSIZE:
769 		if (!ctx->is_bdev)
770 			return invalfc(fsc, "blksize only supported for fuseblk");
771 		ctx->blksize = result.uint_32;
772 		break;
773 
774 	default:
775 		return -EINVAL;
776 	}
777 
778 	return 0;
779 }
780 
781 static void fuse_free_fsc(struct fs_context *fsc)
782 {
783 	struct fuse_fs_context *ctx = fsc->fs_private;
784 
785 	if (ctx) {
786 		kfree(ctx->subtype);
787 		kfree(ctx);
788 	}
789 }
790 
791 static int fuse_show_options(struct seq_file *m, struct dentry *root)
792 {
793 	struct super_block *sb = root->d_sb;
794 	struct fuse_conn *fc = get_fuse_conn_super(sb);
795 
796 	if (fc->legacy_opts_show) {
797 		seq_printf(m, ",user_id=%u",
798 			   from_kuid_munged(fc->user_ns, fc->user_id));
799 		seq_printf(m, ",group_id=%u",
800 			   from_kgid_munged(fc->user_ns, fc->group_id));
801 		if (fc->default_permissions)
802 			seq_puts(m, ",default_permissions");
803 		if (fc->allow_other)
804 			seq_puts(m, ",allow_other");
805 		if (fc->max_read != ~0)
806 			seq_printf(m, ",max_read=%u", fc->max_read);
807 		if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
808 			seq_printf(m, ",blksize=%lu", sb->s_blocksize);
809 	}
810 #ifdef CONFIG_FUSE_DAX
811 	if (fc->dax_mode == FUSE_DAX_ALWAYS)
812 		seq_puts(m, ",dax=always");
813 	else if (fc->dax_mode == FUSE_DAX_NEVER)
814 		seq_puts(m, ",dax=never");
815 	else if (fc->dax_mode == FUSE_DAX_INODE_USER)
816 		seq_puts(m, ",dax=inode");
817 #endif
818 
819 	return 0;
820 }
821 
822 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
823 			     const struct fuse_iqueue_ops *ops,
824 			     void *priv)
825 {
826 	memset(fiq, 0, sizeof(struct fuse_iqueue));
827 	spin_lock_init(&fiq->lock);
828 	init_waitqueue_head(&fiq->waitq);
829 	INIT_LIST_HEAD(&fiq->pending);
830 	INIT_LIST_HEAD(&fiq->interrupts);
831 	fiq->forget_list_tail = &fiq->forget_list_head;
832 	fiq->connected = 1;
833 	fiq->ops = ops;
834 	fiq->priv = priv;
835 }
836 
837 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
838 {
839 	unsigned int i;
840 
841 	spin_lock_init(&fpq->lock);
842 	for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
843 		INIT_LIST_HEAD(&fpq->processing[i]);
844 	INIT_LIST_HEAD(&fpq->io);
845 	fpq->connected = 1;
846 }
847 
848 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
849 		    struct user_namespace *user_ns,
850 		    const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
851 {
852 	memset(fc, 0, sizeof(*fc));
853 	spin_lock_init(&fc->lock);
854 	spin_lock_init(&fc->bg_lock);
855 	init_rwsem(&fc->killsb);
856 	refcount_set(&fc->count, 1);
857 	atomic_set(&fc->dev_count, 1);
858 	init_waitqueue_head(&fc->blocked_waitq);
859 	fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
860 	INIT_LIST_HEAD(&fc->bg_queue);
861 	INIT_LIST_HEAD(&fc->entry);
862 	INIT_LIST_HEAD(&fc->devices);
863 	atomic_set(&fc->num_waiting, 0);
864 	fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
865 	fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
866 	atomic64_set(&fc->khctr, 0);
867 	fc->polled_files = RB_ROOT;
868 	fc->blocked = 0;
869 	fc->initialized = 0;
870 	fc->connected = 1;
871 	atomic64_set(&fc->attr_version, 1);
872 	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
873 	fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
874 	fc->user_ns = get_user_ns(user_ns);
875 	fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
876 	fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
877 
878 	INIT_LIST_HEAD(&fc->mounts);
879 	list_add(&fm->fc_entry, &fc->mounts);
880 	fm->fc = fc;
881 }
882 EXPORT_SYMBOL_GPL(fuse_conn_init);
883 
884 void fuse_conn_put(struct fuse_conn *fc)
885 {
886 	if (refcount_dec_and_test(&fc->count)) {
887 		struct fuse_iqueue *fiq = &fc->iq;
888 		struct fuse_sync_bucket *bucket;
889 
890 		if (IS_ENABLED(CONFIG_FUSE_DAX))
891 			fuse_dax_conn_free(fc);
892 		if (fiq->ops->release)
893 			fiq->ops->release(fiq);
894 		put_pid_ns(fc->pid_ns);
895 		put_user_ns(fc->user_ns);
896 		bucket = rcu_dereference_protected(fc->curr_bucket, 1);
897 		if (bucket) {
898 			WARN_ON(atomic_read(&bucket->count) != 1);
899 			kfree(bucket);
900 		}
901 		fc->release(fc);
902 	}
903 }
904 EXPORT_SYMBOL_GPL(fuse_conn_put);
905 
906 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
907 {
908 	refcount_inc(&fc->count);
909 	return fc;
910 }
911 EXPORT_SYMBOL_GPL(fuse_conn_get);
912 
913 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
914 {
915 	struct fuse_attr attr;
916 	memset(&attr, 0, sizeof(attr));
917 
918 	attr.mode = mode;
919 	attr.ino = FUSE_ROOT_ID;
920 	attr.nlink = 1;
921 	return fuse_iget(sb, 1, 0, &attr, 0, 0);
922 }
923 
924 struct fuse_inode_handle {
925 	u64 nodeid;
926 	u32 generation;
927 };
928 
929 static struct dentry *fuse_get_dentry(struct super_block *sb,
930 				      struct fuse_inode_handle *handle)
931 {
932 	struct fuse_conn *fc = get_fuse_conn_super(sb);
933 	struct inode *inode;
934 	struct dentry *entry;
935 	int err = -ESTALE;
936 
937 	if (handle->nodeid == 0)
938 		goto out_err;
939 
940 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
941 	if (!inode) {
942 		struct fuse_entry_out outarg;
943 		const struct qstr name = QSTR_INIT(".", 1);
944 
945 		if (!fc->export_support)
946 			goto out_err;
947 
948 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
949 				       &inode);
950 		if (err && err != -ENOENT)
951 			goto out_err;
952 		if (err || !inode) {
953 			err = -ESTALE;
954 			goto out_err;
955 		}
956 		err = -EIO;
957 		if (get_node_id(inode) != handle->nodeid)
958 			goto out_iput;
959 	}
960 	err = -ESTALE;
961 	if (inode->i_generation != handle->generation)
962 		goto out_iput;
963 
964 	entry = d_obtain_alias(inode);
965 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
966 		fuse_invalidate_entry_cache(entry);
967 
968 	return entry;
969 
970  out_iput:
971 	iput(inode);
972  out_err:
973 	return ERR_PTR(err);
974 }
975 
976 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
977 			   struct inode *parent)
978 {
979 	int len = parent ? 6 : 3;
980 	u64 nodeid;
981 	u32 generation;
982 
983 	if (*max_len < len) {
984 		*max_len = len;
985 		return  FILEID_INVALID;
986 	}
987 
988 	nodeid = get_fuse_inode(inode)->nodeid;
989 	generation = inode->i_generation;
990 
991 	fh[0] = (u32)(nodeid >> 32);
992 	fh[1] = (u32)(nodeid & 0xffffffff);
993 	fh[2] = generation;
994 
995 	if (parent) {
996 		nodeid = get_fuse_inode(parent)->nodeid;
997 		generation = parent->i_generation;
998 
999 		fh[3] = (u32)(nodeid >> 32);
1000 		fh[4] = (u32)(nodeid & 0xffffffff);
1001 		fh[5] = generation;
1002 	}
1003 
1004 	*max_len = len;
1005 	return parent ? 0x82 : 0x81;
1006 }
1007 
1008 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
1009 		struct fid *fid, int fh_len, int fh_type)
1010 {
1011 	struct fuse_inode_handle handle;
1012 
1013 	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
1014 		return NULL;
1015 
1016 	handle.nodeid = (u64) fid->raw[0] << 32;
1017 	handle.nodeid |= (u64) fid->raw[1];
1018 	handle.generation = fid->raw[2];
1019 	return fuse_get_dentry(sb, &handle);
1020 }
1021 
1022 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
1023 		struct fid *fid, int fh_len, int fh_type)
1024 {
1025 	struct fuse_inode_handle parent;
1026 
1027 	if (fh_type != 0x82 || fh_len < 6)
1028 		return NULL;
1029 
1030 	parent.nodeid = (u64) fid->raw[3] << 32;
1031 	parent.nodeid |= (u64) fid->raw[4];
1032 	parent.generation = fid->raw[5];
1033 	return fuse_get_dentry(sb, &parent);
1034 }
1035 
1036 static struct dentry *fuse_get_parent(struct dentry *child)
1037 {
1038 	struct inode *child_inode = d_inode(child);
1039 	struct fuse_conn *fc = get_fuse_conn(child_inode);
1040 	struct inode *inode;
1041 	struct dentry *parent;
1042 	struct fuse_entry_out outarg;
1043 	int err;
1044 
1045 	if (!fc->export_support)
1046 		return ERR_PTR(-ESTALE);
1047 
1048 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
1049 			       &dotdot_name, &outarg, &inode);
1050 	if (err) {
1051 		if (err == -ENOENT)
1052 			return ERR_PTR(-ESTALE);
1053 		return ERR_PTR(err);
1054 	}
1055 
1056 	parent = d_obtain_alias(inode);
1057 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
1058 		fuse_invalidate_entry_cache(parent);
1059 
1060 	return parent;
1061 }
1062 
1063 static const struct export_operations fuse_export_operations = {
1064 	.fh_to_dentry	= fuse_fh_to_dentry,
1065 	.fh_to_parent	= fuse_fh_to_parent,
1066 	.encode_fh	= fuse_encode_fh,
1067 	.get_parent	= fuse_get_parent,
1068 };
1069 
1070 static const struct super_operations fuse_super_operations = {
1071 	.alloc_inode    = fuse_alloc_inode,
1072 	.free_inode     = fuse_free_inode,
1073 	.evict_inode	= fuse_evict_inode,
1074 	.write_inode	= fuse_write_inode,
1075 	.drop_inode	= generic_delete_inode,
1076 	.umount_begin	= fuse_umount_begin,
1077 	.statfs		= fuse_statfs,
1078 	.sync_fs	= fuse_sync_fs,
1079 	.show_options	= fuse_show_options,
1080 };
1081 
1082 static void sanitize_global_limit(unsigned *limit)
1083 {
1084 	/*
1085 	 * The default maximum number of async requests is calculated to consume
1086 	 * 1/2^13 of the total memory, assuming 392 bytes per request.
1087 	 */
1088 	if (*limit == 0)
1089 		*limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1090 
1091 	if (*limit >= 1 << 16)
1092 		*limit = (1 << 16) - 1;
1093 }
1094 
1095 static int set_global_limit(const char *val, const struct kernel_param *kp)
1096 {
1097 	int rv;
1098 
1099 	rv = param_set_uint(val, kp);
1100 	if (rv)
1101 		return rv;
1102 
1103 	sanitize_global_limit((unsigned *)kp->arg);
1104 
1105 	return 0;
1106 }
1107 
1108 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1109 {
1110 	int cap_sys_admin = capable(CAP_SYS_ADMIN);
1111 
1112 	if (arg->minor < 13)
1113 		return;
1114 
1115 	sanitize_global_limit(&max_user_bgreq);
1116 	sanitize_global_limit(&max_user_congthresh);
1117 
1118 	spin_lock(&fc->bg_lock);
1119 	if (arg->max_background) {
1120 		fc->max_background = arg->max_background;
1121 
1122 		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1123 			fc->max_background = max_user_bgreq;
1124 	}
1125 	if (arg->congestion_threshold) {
1126 		fc->congestion_threshold = arg->congestion_threshold;
1127 
1128 		if (!cap_sys_admin &&
1129 		    fc->congestion_threshold > max_user_congthresh)
1130 			fc->congestion_threshold = max_user_congthresh;
1131 	}
1132 	spin_unlock(&fc->bg_lock);
1133 }
1134 
1135 struct fuse_init_args {
1136 	struct fuse_args args;
1137 	struct fuse_init_in in;
1138 	struct fuse_init_out out;
1139 };
1140 
1141 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1142 			       int error)
1143 {
1144 	struct fuse_conn *fc = fm->fc;
1145 	struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1146 	struct fuse_init_out *arg = &ia->out;
1147 	bool ok = true;
1148 
1149 	if (error || arg->major != FUSE_KERNEL_VERSION)
1150 		ok = false;
1151 	else {
1152 		unsigned long ra_pages;
1153 
1154 		process_init_limits(fc, arg);
1155 
1156 		if (arg->minor >= 6) {
1157 			u64 flags = arg->flags;
1158 
1159 			if (flags & FUSE_INIT_EXT)
1160 				flags |= (u64) arg->flags2 << 32;
1161 
1162 			ra_pages = arg->max_readahead / PAGE_SIZE;
1163 			if (flags & FUSE_ASYNC_READ)
1164 				fc->async_read = 1;
1165 			if (!(flags & FUSE_POSIX_LOCKS))
1166 				fc->no_lock = 1;
1167 			if (arg->minor >= 17) {
1168 				if (!(flags & FUSE_FLOCK_LOCKS))
1169 					fc->no_flock = 1;
1170 			} else {
1171 				if (!(flags & FUSE_POSIX_LOCKS))
1172 					fc->no_flock = 1;
1173 			}
1174 			if (flags & FUSE_ATOMIC_O_TRUNC)
1175 				fc->atomic_o_trunc = 1;
1176 			if (arg->minor >= 9) {
1177 				/* LOOKUP has dependency on proto version */
1178 				if (flags & FUSE_EXPORT_SUPPORT)
1179 					fc->export_support = 1;
1180 			}
1181 			if (flags & FUSE_BIG_WRITES)
1182 				fc->big_writes = 1;
1183 			if (flags & FUSE_DONT_MASK)
1184 				fc->dont_mask = 1;
1185 			if (flags & FUSE_AUTO_INVAL_DATA)
1186 				fc->auto_inval_data = 1;
1187 			else if (flags & FUSE_EXPLICIT_INVAL_DATA)
1188 				fc->explicit_inval_data = 1;
1189 			if (flags & FUSE_DO_READDIRPLUS) {
1190 				fc->do_readdirplus = 1;
1191 				if (flags & FUSE_READDIRPLUS_AUTO)
1192 					fc->readdirplus_auto = 1;
1193 			}
1194 			if (flags & FUSE_ASYNC_DIO)
1195 				fc->async_dio = 1;
1196 			if (flags & FUSE_WRITEBACK_CACHE)
1197 				fc->writeback_cache = 1;
1198 			if (flags & FUSE_PARALLEL_DIROPS)
1199 				fc->parallel_dirops = 1;
1200 			if (flags & FUSE_HANDLE_KILLPRIV)
1201 				fc->handle_killpriv = 1;
1202 			if (arg->time_gran && arg->time_gran <= 1000000000)
1203 				fm->sb->s_time_gran = arg->time_gran;
1204 			if ((flags & FUSE_POSIX_ACL)) {
1205 				fc->default_permissions = 1;
1206 				fc->posix_acl = 1;
1207 			}
1208 			if (flags & FUSE_CACHE_SYMLINKS)
1209 				fc->cache_symlinks = 1;
1210 			if (flags & FUSE_ABORT_ERROR)
1211 				fc->abort_err = 1;
1212 			if (flags & FUSE_MAX_PAGES) {
1213 				fc->max_pages =
1214 					min_t(unsigned int, fc->max_pages_limit,
1215 					max_t(unsigned int, arg->max_pages, 1));
1216 			}
1217 			if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1218 				if (flags & FUSE_MAP_ALIGNMENT &&
1219 				    !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1220 					ok = false;
1221 				}
1222 				if (flags & FUSE_HAS_INODE_DAX)
1223 					fc->inode_dax = 1;
1224 			}
1225 			if (flags & FUSE_HANDLE_KILLPRIV_V2) {
1226 				fc->handle_killpriv_v2 = 1;
1227 				fm->sb->s_flags |= SB_NOSEC;
1228 			}
1229 			if (flags & FUSE_SETXATTR_EXT)
1230 				fc->setxattr_ext = 1;
1231 			if (flags & FUSE_SECURITY_CTX)
1232 				fc->init_security = 1;
1233 			if (flags & FUSE_CREATE_SUPP_GROUP)
1234 				fc->create_supp_group = 1;
1235 			if (flags & FUSE_DIRECT_IO_RELAX)
1236 				fc->direct_io_relax = 1;
1237 		} else {
1238 			ra_pages = fc->max_read / PAGE_SIZE;
1239 			fc->no_lock = 1;
1240 			fc->no_flock = 1;
1241 		}
1242 
1243 		fm->sb->s_bdi->ra_pages =
1244 				min(fm->sb->s_bdi->ra_pages, ra_pages);
1245 		fc->minor = arg->minor;
1246 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1247 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
1248 		fc->conn_init = 1;
1249 	}
1250 	kfree(ia);
1251 
1252 	if (!ok) {
1253 		fc->conn_init = 0;
1254 		fc->conn_error = 1;
1255 	}
1256 
1257 	fuse_set_initialized(fc);
1258 	wake_up_all(&fc->blocked_waitq);
1259 }
1260 
1261 void fuse_send_init(struct fuse_mount *fm)
1262 {
1263 	struct fuse_init_args *ia;
1264 	u64 flags;
1265 
1266 	ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1267 
1268 	ia->in.major = FUSE_KERNEL_VERSION;
1269 	ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1270 	ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1271 	flags =
1272 		FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1273 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1274 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1275 		FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1276 		FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1277 		FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1278 		FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1279 		FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1280 		FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1281 		FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT | FUSE_INIT_EXT |
1282 		FUSE_SECURITY_CTX | FUSE_CREATE_SUPP_GROUP |
1283 		FUSE_HAS_EXPIRE_ONLY | FUSE_DIRECT_IO_RELAX;
1284 #ifdef CONFIG_FUSE_DAX
1285 	if (fm->fc->dax)
1286 		flags |= FUSE_MAP_ALIGNMENT;
1287 	if (fuse_is_inode_dax_mode(fm->fc->dax_mode))
1288 		flags |= FUSE_HAS_INODE_DAX;
1289 #endif
1290 	if (fm->fc->auto_submounts)
1291 		flags |= FUSE_SUBMOUNTS;
1292 
1293 	ia->in.flags = flags;
1294 	ia->in.flags2 = flags >> 32;
1295 
1296 	ia->args.opcode = FUSE_INIT;
1297 	ia->args.in_numargs = 1;
1298 	ia->args.in_args[0].size = sizeof(ia->in);
1299 	ia->args.in_args[0].value = &ia->in;
1300 	ia->args.out_numargs = 1;
1301 	/* Variable length argument used for backward compatibility
1302 	   with interface version < 7.5.  Rest of init_out is zeroed
1303 	   by do_get_request(), so a short reply is not a problem */
1304 	ia->args.out_argvar = true;
1305 	ia->args.out_args[0].size = sizeof(ia->out);
1306 	ia->args.out_args[0].value = &ia->out;
1307 	ia->args.force = true;
1308 	ia->args.nocreds = true;
1309 	ia->args.end = process_init_reply;
1310 
1311 	if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1312 		process_init_reply(fm, &ia->args, -ENOTCONN);
1313 }
1314 EXPORT_SYMBOL_GPL(fuse_send_init);
1315 
1316 void fuse_free_conn(struct fuse_conn *fc)
1317 {
1318 	WARN_ON(!list_empty(&fc->devices));
1319 	kfree_rcu(fc, rcu);
1320 }
1321 EXPORT_SYMBOL_GPL(fuse_free_conn);
1322 
1323 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1324 {
1325 	int err;
1326 	char *suffix = "";
1327 
1328 	if (sb->s_bdev) {
1329 		suffix = "-fuseblk";
1330 		/*
1331 		 * sb->s_bdi points to blkdev's bdi however we want to redirect
1332 		 * it to our private bdi...
1333 		 */
1334 		bdi_put(sb->s_bdi);
1335 		sb->s_bdi = &noop_backing_dev_info;
1336 	}
1337 	err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1338 				   MINOR(fc->dev), suffix);
1339 	if (err)
1340 		return err;
1341 
1342 	/* fuse does it's own writeback accounting */
1343 	sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1344 	sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1345 
1346 	/*
1347 	 * For a single fuse filesystem use max 1% of dirty +
1348 	 * writeback threshold.
1349 	 *
1350 	 * This gives about 1M of write buffer for memory maps on a
1351 	 * machine with 1G and 10% dirty_ratio, which should be more
1352 	 * than enough.
1353 	 *
1354 	 * Privileged users can raise it by writing to
1355 	 *
1356 	 *    /sys/class/bdi/<bdi>/max_ratio
1357 	 */
1358 	bdi_set_max_ratio(sb->s_bdi, 1);
1359 
1360 	return 0;
1361 }
1362 
1363 struct fuse_dev *fuse_dev_alloc(void)
1364 {
1365 	struct fuse_dev *fud;
1366 	struct list_head *pq;
1367 
1368 	fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1369 	if (!fud)
1370 		return NULL;
1371 
1372 	pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1373 	if (!pq) {
1374 		kfree(fud);
1375 		return NULL;
1376 	}
1377 
1378 	fud->pq.processing = pq;
1379 	fuse_pqueue_init(&fud->pq);
1380 
1381 	return fud;
1382 }
1383 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1384 
1385 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1386 {
1387 	fud->fc = fuse_conn_get(fc);
1388 	spin_lock(&fc->lock);
1389 	list_add_tail(&fud->entry, &fc->devices);
1390 	spin_unlock(&fc->lock);
1391 }
1392 EXPORT_SYMBOL_GPL(fuse_dev_install);
1393 
1394 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1395 {
1396 	struct fuse_dev *fud;
1397 
1398 	fud = fuse_dev_alloc();
1399 	if (!fud)
1400 		return NULL;
1401 
1402 	fuse_dev_install(fud, fc);
1403 	return fud;
1404 }
1405 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1406 
1407 void fuse_dev_free(struct fuse_dev *fud)
1408 {
1409 	struct fuse_conn *fc = fud->fc;
1410 
1411 	if (fc) {
1412 		spin_lock(&fc->lock);
1413 		list_del(&fud->entry);
1414 		spin_unlock(&fc->lock);
1415 
1416 		fuse_conn_put(fc);
1417 	}
1418 	kfree(fud->pq.processing);
1419 	kfree(fud);
1420 }
1421 EXPORT_SYMBOL_GPL(fuse_dev_free);
1422 
1423 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1424 				      const struct fuse_inode *fi)
1425 {
1426 	struct timespec64 ctime = inode_get_ctime(&fi->inode);
1427 
1428 	*attr = (struct fuse_attr){
1429 		.ino		= fi->inode.i_ino,
1430 		.size		= fi->inode.i_size,
1431 		.blocks		= fi->inode.i_blocks,
1432 		.atime		= fi->inode.i_atime.tv_sec,
1433 		.mtime		= fi->inode.i_mtime.tv_sec,
1434 		.ctime		= ctime.tv_sec,
1435 		.atimensec	= fi->inode.i_atime.tv_nsec,
1436 		.mtimensec	= fi->inode.i_mtime.tv_nsec,
1437 		.ctimensec	= ctime.tv_nsec,
1438 		.mode		= fi->inode.i_mode,
1439 		.nlink		= fi->inode.i_nlink,
1440 		.uid		= fi->inode.i_uid.val,
1441 		.gid		= fi->inode.i_gid.val,
1442 		.rdev		= fi->inode.i_rdev,
1443 		.blksize	= 1u << fi->inode.i_blkbits,
1444 	};
1445 }
1446 
1447 static void fuse_sb_defaults(struct super_block *sb)
1448 {
1449 	sb->s_magic = FUSE_SUPER_MAGIC;
1450 	sb->s_op = &fuse_super_operations;
1451 	sb->s_xattr = fuse_xattr_handlers;
1452 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1453 	sb->s_time_gran = 1;
1454 	sb->s_export_op = &fuse_export_operations;
1455 	sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1456 	if (sb->s_user_ns != &init_user_ns)
1457 		sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1458 	sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1459 }
1460 
1461 static int fuse_fill_super_submount(struct super_block *sb,
1462 				    struct fuse_inode *parent_fi)
1463 {
1464 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1465 	struct super_block *parent_sb = parent_fi->inode.i_sb;
1466 	struct fuse_attr root_attr;
1467 	struct inode *root;
1468 
1469 	fuse_sb_defaults(sb);
1470 	fm->sb = sb;
1471 
1472 	WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1473 	sb->s_bdi = bdi_get(parent_sb->s_bdi);
1474 
1475 	sb->s_xattr = parent_sb->s_xattr;
1476 	sb->s_time_gran = parent_sb->s_time_gran;
1477 	sb->s_blocksize = parent_sb->s_blocksize;
1478 	sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1479 	sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1480 	if (parent_sb->s_subtype && !sb->s_subtype)
1481 		return -ENOMEM;
1482 
1483 	fuse_fill_attr_from_inode(&root_attr, parent_fi);
1484 	root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1485 	/*
1486 	 * This inode is just a duplicate, so it is not looked up and
1487 	 * its nlookup should not be incremented.  fuse_iget() does
1488 	 * that, though, so undo it here.
1489 	 */
1490 	get_fuse_inode(root)->nlookup--;
1491 	sb->s_d_op = &fuse_dentry_operations;
1492 	sb->s_root = d_make_root(root);
1493 	if (!sb->s_root)
1494 		return -ENOMEM;
1495 
1496 	return 0;
1497 }
1498 
1499 /* Filesystem context private data holds the FUSE inode of the mount point */
1500 static int fuse_get_tree_submount(struct fs_context *fsc)
1501 {
1502 	struct fuse_mount *fm;
1503 	struct fuse_inode *mp_fi = fsc->fs_private;
1504 	struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1505 	struct super_block *sb;
1506 	int err;
1507 
1508 	fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1509 	if (!fm)
1510 		return -ENOMEM;
1511 
1512 	fm->fc = fuse_conn_get(fc);
1513 	fsc->s_fs_info = fm;
1514 	sb = sget_fc(fsc, NULL, set_anon_super_fc);
1515 	if (fsc->s_fs_info)
1516 		fuse_mount_destroy(fm);
1517 	if (IS_ERR(sb))
1518 		return PTR_ERR(sb);
1519 
1520 	/* Initialize superblock, making @mp_fi its root */
1521 	err = fuse_fill_super_submount(sb, mp_fi);
1522 	if (err) {
1523 		deactivate_locked_super(sb);
1524 		return err;
1525 	}
1526 
1527 	down_write(&fc->killsb);
1528 	list_add_tail(&fm->fc_entry, &fc->mounts);
1529 	up_write(&fc->killsb);
1530 
1531 	sb->s_flags |= SB_ACTIVE;
1532 	fsc->root = dget(sb->s_root);
1533 
1534 	return 0;
1535 }
1536 
1537 static const struct fs_context_operations fuse_context_submount_ops = {
1538 	.get_tree	= fuse_get_tree_submount,
1539 };
1540 
1541 int fuse_init_fs_context_submount(struct fs_context *fsc)
1542 {
1543 	fsc->ops = &fuse_context_submount_ops;
1544 	return 0;
1545 }
1546 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1547 
1548 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1549 {
1550 	struct fuse_dev *fud = NULL;
1551 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1552 	struct fuse_conn *fc = fm->fc;
1553 	struct inode *root;
1554 	struct dentry *root_dentry;
1555 	int err;
1556 
1557 	err = -EINVAL;
1558 	if (sb->s_flags & SB_MANDLOCK)
1559 		goto err;
1560 
1561 	rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1562 	fuse_sb_defaults(sb);
1563 
1564 	if (ctx->is_bdev) {
1565 #ifdef CONFIG_BLOCK
1566 		err = -EINVAL;
1567 		if (!sb_set_blocksize(sb, ctx->blksize))
1568 			goto err;
1569 #endif
1570 	} else {
1571 		sb->s_blocksize = PAGE_SIZE;
1572 		sb->s_blocksize_bits = PAGE_SHIFT;
1573 	}
1574 
1575 	sb->s_subtype = ctx->subtype;
1576 	ctx->subtype = NULL;
1577 	if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1578 		err = fuse_dax_conn_alloc(fc, ctx->dax_mode, ctx->dax_dev);
1579 		if (err)
1580 			goto err;
1581 	}
1582 
1583 	if (ctx->fudptr) {
1584 		err = -ENOMEM;
1585 		fud = fuse_dev_alloc_install(fc);
1586 		if (!fud)
1587 			goto err_free_dax;
1588 	}
1589 
1590 	fc->dev = sb->s_dev;
1591 	fm->sb = sb;
1592 	err = fuse_bdi_init(fc, sb);
1593 	if (err)
1594 		goto err_dev_free;
1595 
1596 	/* Handle umasking inside the fuse code */
1597 	if (sb->s_flags & SB_POSIXACL)
1598 		fc->dont_mask = 1;
1599 	sb->s_flags |= SB_POSIXACL;
1600 
1601 	fc->default_permissions = ctx->default_permissions;
1602 	fc->allow_other = ctx->allow_other;
1603 	fc->user_id = ctx->user_id;
1604 	fc->group_id = ctx->group_id;
1605 	fc->legacy_opts_show = ctx->legacy_opts_show;
1606 	fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1607 	fc->destroy = ctx->destroy;
1608 	fc->no_control = ctx->no_control;
1609 	fc->no_force_umount = ctx->no_force_umount;
1610 
1611 	err = -ENOMEM;
1612 	root = fuse_get_root_inode(sb, ctx->rootmode);
1613 	sb->s_d_op = &fuse_root_dentry_operations;
1614 	root_dentry = d_make_root(root);
1615 	if (!root_dentry)
1616 		goto err_dev_free;
1617 	/* Root dentry doesn't have .d_revalidate */
1618 	sb->s_d_op = &fuse_dentry_operations;
1619 
1620 	mutex_lock(&fuse_mutex);
1621 	err = -EINVAL;
1622 	if (ctx->fudptr && *ctx->fudptr)
1623 		goto err_unlock;
1624 
1625 	err = fuse_ctl_add_conn(fc);
1626 	if (err)
1627 		goto err_unlock;
1628 
1629 	list_add_tail(&fc->entry, &fuse_conn_list);
1630 	sb->s_root = root_dentry;
1631 	if (ctx->fudptr)
1632 		*ctx->fudptr = fud;
1633 	mutex_unlock(&fuse_mutex);
1634 	return 0;
1635 
1636  err_unlock:
1637 	mutex_unlock(&fuse_mutex);
1638 	dput(root_dentry);
1639  err_dev_free:
1640 	if (fud)
1641 		fuse_dev_free(fud);
1642  err_free_dax:
1643 	if (IS_ENABLED(CONFIG_FUSE_DAX))
1644 		fuse_dax_conn_free(fc);
1645  err:
1646 	return err;
1647 }
1648 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1649 
1650 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1651 {
1652 	struct fuse_fs_context *ctx = fsc->fs_private;
1653 	int err;
1654 
1655 	if (!ctx->file || !ctx->rootmode_present ||
1656 	    !ctx->user_id_present || !ctx->group_id_present)
1657 		return -EINVAL;
1658 
1659 	/*
1660 	 * Require mount to happen from the same user namespace which
1661 	 * opened /dev/fuse to prevent potential attacks.
1662 	 */
1663 	if ((ctx->file->f_op != &fuse_dev_operations) ||
1664 	    (ctx->file->f_cred->user_ns != sb->s_user_ns))
1665 		return -EINVAL;
1666 	ctx->fudptr = &ctx->file->private_data;
1667 
1668 	err = fuse_fill_super_common(sb, ctx);
1669 	if (err)
1670 		return err;
1671 	/* file->private_data shall be visible on all CPUs after this */
1672 	smp_mb();
1673 	fuse_send_init(get_fuse_mount_super(sb));
1674 	return 0;
1675 }
1676 
1677 /*
1678  * This is the path where user supplied an already initialized fuse dev.  In
1679  * this case never create a new super if the old one is gone.
1680  */
1681 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1682 {
1683 	return -ENOTCONN;
1684 }
1685 
1686 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1687 {
1688 
1689 	return fsc->sget_key == get_fuse_conn_super(sb);
1690 }
1691 
1692 static int fuse_get_tree(struct fs_context *fsc)
1693 {
1694 	struct fuse_fs_context *ctx = fsc->fs_private;
1695 	struct fuse_dev *fud;
1696 	struct fuse_conn *fc;
1697 	struct fuse_mount *fm;
1698 	struct super_block *sb;
1699 	int err;
1700 
1701 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1702 	if (!fc)
1703 		return -ENOMEM;
1704 
1705 	fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1706 	if (!fm) {
1707 		kfree(fc);
1708 		return -ENOMEM;
1709 	}
1710 
1711 	fuse_conn_init(fc, fm, fsc->user_ns, &fuse_dev_fiq_ops, NULL);
1712 	fc->release = fuse_free_conn;
1713 
1714 	fsc->s_fs_info = fm;
1715 
1716 	if (ctx->fd_present)
1717 		ctx->file = fget(ctx->fd);
1718 
1719 	if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1720 		err = get_tree_bdev(fsc, fuse_fill_super);
1721 		goto out;
1722 	}
1723 	/*
1724 	 * While block dev mount can be initialized with a dummy device fd
1725 	 * (found by device name), normal fuse mounts can't
1726 	 */
1727 	err = -EINVAL;
1728 	if (!ctx->file)
1729 		goto out;
1730 
1731 	/*
1732 	 * Allow creating a fuse mount with an already initialized fuse
1733 	 * connection
1734 	 */
1735 	fud = READ_ONCE(ctx->file->private_data);
1736 	if (ctx->file->f_op == &fuse_dev_operations && fud) {
1737 		fsc->sget_key = fud->fc;
1738 		sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1739 		err = PTR_ERR_OR_ZERO(sb);
1740 		if (!IS_ERR(sb))
1741 			fsc->root = dget(sb->s_root);
1742 	} else {
1743 		err = get_tree_nodev(fsc, fuse_fill_super);
1744 	}
1745 out:
1746 	if (fsc->s_fs_info)
1747 		fuse_mount_destroy(fm);
1748 	if (ctx->file)
1749 		fput(ctx->file);
1750 	return err;
1751 }
1752 
1753 static const struct fs_context_operations fuse_context_ops = {
1754 	.free		= fuse_free_fsc,
1755 	.parse_param	= fuse_parse_param,
1756 	.reconfigure	= fuse_reconfigure,
1757 	.get_tree	= fuse_get_tree,
1758 };
1759 
1760 /*
1761  * Set up the filesystem mount context.
1762  */
1763 static int fuse_init_fs_context(struct fs_context *fsc)
1764 {
1765 	struct fuse_fs_context *ctx;
1766 
1767 	ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1768 	if (!ctx)
1769 		return -ENOMEM;
1770 
1771 	ctx->max_read = ~0;
1772 	ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1773 	ctx->legacy_opts_show = true;
1774 
1775 #ifdef CONFIG_BLOCK
1776 	if (fsc->fs_type == &fuseblk_fs_type) {
1777 		ctx->is_bdev = true;
1778 		ctx->destroy = true;
1779 	}
1780 #endif
1781 
1782 	fsc->fs_private = ctx;
1783 	fsc->ops = &fuse_context_ops;
1784 	return 0;
1785 }
1786 
1787 bool fuse_mount_remove(struct fuse_mount *fm)
1788 {
1789 	struct fuse_conn *fc = fm->fc;
1790 	bool last = false;
1791 
1792 	down_write(&fc->killsb);
1793 	list_del_init(&fm->fc_entry);
1794 	if (list_empty(&fc->mounts))
1795 		last = true;
1796 	up_write(&fc->killsb);
1797 
1798 	return last;
1799 }
1800 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1801 
1802 void fuse_conn_destroy(struct fuse_mount *fm)
1803 {
1804 	struct fuse_conn *fc = fm->fc;
1805 
1806 	if (fc->destroy)
1807 		fuse_send_destroy(fm);
1808 
1809 	fuse_abort_conn(fc);
1810 	fuse_wait_aborted(fc);
1811 
1812 	if (!list_empty(&fc->entry)) {
1813 		mutex_lock(&fuse_mutex);
1814 		list_del(&fc->entry);
1815 		fuse_ctl_remove_conn(fc);
1816 		mutex_unlock(&fuse_mutex);
1817 	}
1818 }
1819 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1820 
1821 static void fuse_sb_destroy(struct super_block *sb)
1822 {
1823 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1824 	bool last;
1825 
1826 	if (sb->s_root) {
1827 		last = fuse_mount_remove(fm);
1828 		if (last)
1829 			fuse_conn_destroy(fm);
1830 	}
1831 }
1832 
1833 void fuse_mount_destroy(struct fuse_mount *fm)
1834 {
1835 	fuse_conn_put(fm->fc);
1836 	kfree(fm);
1837 }
1838 EXPORT_SYMBOL(fuse_mount_destroy);
1839 
1840 static void fuse_kill_sb_anon(struct super_block *sb)
1841 {
1842 	fuse_sb_destroy(sb);
1843 	kill_anon_super(sb);
1844 	fuse_mount_destroy(get_fuse_mount_super(sb));
1845 }
1846 
1847 static struct file_system_type fuse_fs_type = {
1848 	.owner		= THIS_MODULE,
1849 	.name		= "fuse",
1850 	.fs_flags	= FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1851 	.init_fs_context = fuse_init_fs_context,
1852 	.parameters	= fuse_fs_parameters,
1853 	.kill_sb	= fuse_kill_sb_anon,
1854 };
1855 MODULE_ALIAS_FS("fuse");
1856 
1857 #ifdef CONFIG_BLOCK
1858 static void fuse_kill_sb_blk(struct super_block *sb)
1859 {
1860 	fuse_sb_destroy(sb);
1861 	kill_block_super(sb);
1862 	fuse_mount_destroy(get_fuse_mount_super(sb));
1863 }
1864 
1865 static struct file_system_type fuseblk_fs_type = {
1866 	.owner		= THIS_MODULE,
1867 	.name		= "fuseblk",
1868 	.init_fs_context = fuse_init_fs_context,
1869 	.parameters	= fuse_fs_parameters,
1870 	.kill_sb	= fuse_kill_sb_blk,
1871 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1872 };
1873 MODULE_ALIAS_FS("fuseblk");
1874 
1875 static inline int register_fuseblk(void)
1876 {
1877 	return register_filesystem(&fuseblk_fs_type);
1878 }
1879 
1880 static inline void unregister_fuseblk(void)
1881 {
1882 	unregister_filesystem(&fuseblk_fs_type);
1883 }
1884 #else
1885 static inline int register_fuseblk(void)
1886 {
1887 	return 0;
1888 }
1889 
1890 static inline void unregister_fuseblk(void)
1891 {
1892 }
1893 #endif
1894 
1895 static void fuse_inode_init_once(void *foo)
1896 {
1897 	struct inode *inode = foo;
1898 
1899 	inode_init_once(inode);
1900 }
1901 
1902 static int __init fuse_fs_init(void)
1903 {
1904 	int err;
1905 
1906 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
1907 			sizeof(struct fuse_inode), 0,
1908 			SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1909 			fuse_inode_init_once);
1910 	err = -ENOMEM;
1911 	if (!fuse_inode_cachep)
1912 		goto out;
1913 
1914 	err = register_fuseblk();
1915 	if (err)
1916 		goto out2;
1917 
1918 	err = register_filesystem(&fuse_fs_type);
1919 	if (err)
1920 		goto out3;
1921 
1922 	return 0;
1923 
1924  out3:
1925 	unregister_fuseblk();
1926  out2:
1927 	kmem_cache_destroy(fuse_inode_cachep);
1928  out:
1929 	return err;
1930 }
1931 
1932 static void fuse_fs_cleanup(void)
1933 {
1934 	unregister_filesystem(&fuse_fs_type);
1935 	unregister_fuseblk();
1936 
1937 	/*
1938 	 * Make sure all delayed rcu free inodes are flushed before we
1939 	 * destroy cache.
1940 	 */
1941 	rcu_barrier();
1942 	kmem_cache_destroy(fuse_inode_cachep);
1943 }
1944 
1945 static struct kobject *fuse_kobj;
1946 
1947 static int fuse_sysfs_init(void)
1948 {
1949 	int err;
1950 
1951 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1952 	if (!fuse_kobj) {
1953 		err = -ENOMEM;
1954 		goto out_err;
1955 	}
1956 
1957 	err = sysfs_create_mount_point(fuse_kobj, "connections");
1958 	if (err)
1959 		goto out_fuse_unregister;
1960 
1961 	return 0;
1962 
1963  out_fuse_unregister:
1964 	kobject_put(fuse_kobj);
1965  out_err:
1966 	return err;
1967 }
1968 
1969 static void fuse_sysfs_cleanup(void)
1970 {
1971 	sysfs_remove_mount_point(fuse_kobj, "connections");
1972 	kobject_put(fuse_kobj);
1973 }
1974 
1975 static int __init fuse_init(void)
1976 {
1977 	int res;
1978 
1979 	pr_info("init (API version %i.%i)\n",
1980 		FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1981 
1982 	INIT_LIST_HEAD(&fuse_conn_list);
1983 	res = fuse_fs_init();
1984 	if (res)
1985 		goto err;
1986 
1987 	res = fuse_dev_init();
1988 	if (res)
1989 		goto err_fs_cleanup;
1990 
1991 	res = fuse_sysfs_init();
1992 	if (res)
1993 		goto err_dev_cleanup;
1994 
1995 	res = fuse_ctl_init();
1996 	if (res)
1997 		goto err_sysfs_cleanup;
1998 
1999 	sanitize_global_limit(&max_user_bgreq);
2000 	sanitize_global_limit(&max_user_congthresh);
2001 
2002 	return 0;
2003 
2004  err_sysfs_cleanup:
2005 	fuse_sysfs_cleanup();
2006  err_dev_cleanup:
2007 	fuse_dev_cleanup();
2008  err_fs_cleanup:
2009 	fuse_fs_cleanup();
2010  err:
2011 	return res;
2012 }
2013 
2014 static void __exit fuse_exit(void)
2015 {
2016 	pr_debug("exit\n");
2017 
2018 	fuse_ctl_cleanup();
2019 	fuse_sysfs_cleanup();
2020 	fuse_fs_cleanup();
2021 	fuse_dev_cleanup();
2022 }
2023 
2024 module_init(fuse_init);
2025 module_exit(fuse_exit);
2026