xref: /openbmc/linux/fs/fuse/inode.c (revision 2f4e3926)
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 fuse_submount_lookup *fuse_alloc_submount_lookup(void)
72 {
73 	struct fuse_submount_lookup *sl;
74 
75 	sl = kzalloc(sizeof(struct fuse_submount_lookup), GFP_KERNEL_ACCOUNT);
76 	if (!sl)
77 		return NULL;
78 	sl->forget = fuse_alloc_forget();
79 	if (!sl->forget)
80 		goto out_free;
81 
82 	return sl;
83 
84 out_free:
85 	kfree(sl);
86 	return NULL;
87 }
88 
89 static struct inode *fuse_alloc_inode(struct super_block *sb)
90 {
91 	struct fuse_inode *fi;
92 
93 	fi = alloc_inode_sb(sb, fuse_inode_cachep, GFP_KERNEL);
94 	if (!fi)
95 		return NULL;
96 
97 	fi->i_time = 0;
98 	fi->inval_mask = ~0;
99 	fi->nodeid = 0;
100 	fi->nlookup = 0;
101 	fi->attr_version = 0;
102 	fi->orig_ino = 0;
103 	fi->state = 0;
104 	fi->submount_lookup = NULL;
105 	mutex_init(&fi->mutex);
106 	spin_lock_init(&fi->lock);
107 	fi->forget = fuse_alloc_forget();
108 	if (!fi->forget)
109 		goto out_free;
110 
111 	if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
112 		goto out_free_forget;
113 
114 	return &fi->inode;
115 
116 out_free_forget:
117 	kfree(fi->forget);
118 out_free:
119 	kmem_cache_free(fuse_inode_cachep, fi);
120 	return NULL;
121 }
122 
123 static void fuse_free_inode(struct inode *inode)
124 {
125 	struct fuse_inode *fi = get_fuse_inode(inode);
126 
127 	mutex_destroy(&fi->mutex);
128 	kfree(fi->forget);
129 #ifdef CONFIG_FUSE_DAX
130 	kfree(fi->dax);
131 #endif
132 	kmem_cache_free(fuse_inode_cachep, fi);
133 }
134 
135 static void fuse_cleanup_submount_lookup(struct fuse_conn *fc,
136 					 struct fuse_submount_lookup *sl)
137 {
138 	if (!refcount_dec_and_test(&sl->count))
139 		return;
140 
141 	fuse_queue_forget(fc, sl->forget, sl->nodeid, 1);
142 	sl->forget = NULL;
143 	kfree(sl);
144 }
145 
146 static void fuse_evict_inode(struct inode *inode)
147 {
148 	struct fuse_inode *fi = get_fuse_inode(inode);
149 
150 	/* Will write inode on close/munmap and in all other dirtiers */
151 	WARN_ON(inode->i_state & I_DIRTY_INODE);
152 
153 	truncate_inode_pages_final(&inode->i_data);
154 	clear_inode(inode);
155 	if (inode->i_sb->s_flags & SB_ACTIVE) {
156 		struct fuse_conn *fc = get_fuse_conn(inode);
157 
158 		if (FUSE_IS_DAX(inode))
159 			fuse_dax_inode_cleanup(inode);
160 		if (fi->nlookup) {
161 			fuse_queue_forget(fc, fi->forget, fi->nodeid,
162 					  fi->nlookup);
163 			fi->forget = NULL;
164 		}
165 
166 		if (fi->submount_lookup) {
167 			fuse_cleanup_submount_lookup(fc, fi->submount_lookup);
168 			fi->submount_lookup = NULL;
169 		}
170 	}
171 	if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
172 		WARN_ON(!list_empty(&fi->write_files));
173 		WARN_ON(!list_empty(&fi->queued_writes));
174 	}
175 }
176 
177 static int fuse_reconfigure(struct fs_context *fsc)
178 {
179 	struct super_block *sb = fsc->root->d_sb;
180 
181 	sync_filesystem(sb);
182 	if (fsc->sb_flags & SB_MANDLOCK)
183 		return -EINVAL;
184 
185 	return 0;
186 }
187 
188 /*
189  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
190  * so that it will fit.
191  */
192 static ino_t fuse_squash_ino(u64 ino64)
193 {
194 	ino_t ino = (ino_t) ino64;
195 	if (sizeof(ino_t) < sizeof(u64))
196 		ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
197 	return ino;
198 }
199 
200 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
201 				   struct fuse_statx *sx,
202 				   u64 attr_valid, u32 cache_mask)
203 {
204 	struct fuse_conn *fc = get_fuse_conn(inode);
205 	struct fuse_inode *fi = get_fuse_inode(inode);
206 
207 	lockdep_assert_held(&fi->lock);
208 
209 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
210 	fi->i_time = attr_valid;
211 	/* Clear basic stats from invalid mask */
212 	set_mask_bits(&fi->inval_mask, STATX_BASIC_STATS, 0);
213 
214 	inode->i_ino     = fuse_squash_ino(attr->ino);
215 	inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
216 	set_nlink(inode, attr->nlink);
217 	inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
218 	inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
219 	inode->i_blocks  = attr->blocks;
220 
221 	/* Sanitize nsecs */
222 	attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
223 	attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
224 	attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
225 
226 	inode->i_atime.tv_sec   = attr->atime;
227 	inode->i_atime.tv_nsec  = attr->atimensec;
228 	/* mtime from server may be stale due to local buffered write */
229 	if (!(cache_mask & STATX_MTIME)) {
230 		inode->i_mtime.tv_sec   = attr->mtime;
231 		inode->i_mtime.tv_nsec  = attr->mtimensec;
232 	}
233 	if (!(cache_mask & STATX_CTIME)) {
234 		inode_set_ctime(inode, attr->ctime, attr->ctimensec);
235 	}
236 	if (sx) {
237 		/* Sanitize nsecs */
238 		sx->btime.tv_nsec =
239 			min_t(u32, sx->btime.tv_nsec, NSEC_PER_SEC - 1);
240 
241 		/*
242 		 * Btime has been queried, cache is valid (whether or not btime
243 		 * is available or not) so clear STATX_BTIME from inval_mask.
244 		 *
245 		 * Availability of the btime attribute is indicated in
246 		 * FUSE_I_BTIME
247 		 */
248 		set_mask_bits(&fi->inval_mask, STATX_BTIME, 0);
249 		if (sx->mask & STATX_BTIME) {
250 			set_bit(FUSE_I_BTIME, &fi->state);
251 			fi->i_btime.tv_sec = sx->btime.tv_sec;
252 			fi->i_btime.tv_nsec = sx->btime.tv_nsec;
253 		}
254 	}
255 
256 	if (attr->blksize != 0)
257 		inode->i_blkbits = ilog2(attr->blksize);
258 	else
259 		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
260 
261 	/*
262 	 * Don't set the sticky bit in i_mode, unless we want the VFS
263 	 * to check permissions.  This prevents failures due to the
264 	 * check in may_delete().
265 	 */
266 	fi->orig_i_mode = inode->i_mode;
267 	if (!fc->default_permissions)
268 		inode->i_mode &= ~S_ISVTX;
269 
270 	fi->orig_ino = attr->ino;
271 
272 	/*
273 	 * We are refreshing inode data and it is possible that another
274 	 * client set suid/sgid or security.capability xattr. So clear
275 	 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
276 	 * was set or if security.capability xattr was set. But we don't
277 	 * know if security.capability has been set or not. So clear it
278 	 * anyway. Its less efficient but should be safe.
279 	 */
280 	inode->i_flags &= ~S_NOSEC;
281 }
282 
283 u32 fuse_get_cache_mask(struct inode *inode)
284 {
285 	struct fuse_conn *fc = get_fuse_conn(inode);
286 
287 	if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
288 		return 0;
289 
290 	return STATX_MTIME | STATX_CTIME | STATX_SIZE;
291 }
292 
293 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
294 			    struct fuse_statx *sx,
295 			    u64 attr_valid, u64 attr_version)
296 {
297 	struct fuse_conn *fc = get_fuse_conn(inode);
298 	struct fuse_inode *fi = get_fuse_inode(inode);
299 	u32 cache_mask;
300 	loff_t oldsize;
301 	struct timespec64 old_mtime;
302 
303 	spin_lock(&fi->lock);
304 	/*
305 	 * In case of writeback_cache enabled, writes update mtime, ctime and
306 	 * may update i_size.  In these cases trust the cached value in the
307 	 * inode.
308 	 */
309 	cache_mask = fuse_get_cache_mask(inode);
310 	if (cache_mask & STATX_SIZE)
311 		attr->size = i_size_read(inode);
312 
313 	if (cache_mask & STATX_MTIME) {
314 		attr->mtime = inode->i_mtime.tv_sec;
315 		attr->mtimensec = inode->i_mtime.tv_nsec;
316 	}
317 	if (cache_mask & STATX_CTIME) {
318 		attr->ctime = inode_get_ctime(inode).tv_sec;
319 		attr->ctimensec = inode_get_ctime(inode).tv_nsec;
320 	}
321 
322 	if ((attr_version != 0 && fi->attr_version > attr_version) ||
323 	    test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
324 		spin_unlock(&fi->lock);
325 		return;
326 	}
327 
328 	old_mtime = inode->i_mtime;
329 	fuse_change_attributes_common(inode, attr, sx, attr_valid, cache_mask);
330 
331 	oldsize = inode->i_size;
332 	/*
333 	 * In case of writeback_cache enabled, the cached writes beyond EOF
334 	 * extend local i_size without keeping userspace server in sync. So,
335 	 * attr->size coming from server can be stale. We cannot trust it.
336 	 */
337 	if (!(cache_mask & STATX_SIZE))
338 		i_size_write(inode, attr->size);
339 	spin_unlock(&fi->lock);
340 
341 	if (!cache_mask && S_ISREG(inode->i_mode)) {
342 		bool inval = false;
343 
344 		if (oldsize != attr->size) {
345 			truncate_pagecache(inode, attr->size);
346 			if (!fc->explicit_inval_data)
347 				inval = true;
348 		} else if (fc->auto_inval_data) {
349 			struct timespec64 new_mtime = {
350 				.tv_sec = attr->mtime,
351 				.tv_nsec = attr->mtimensec,
352 			};
353 
354 			/*
355 			 * Auto inval mode also checks and invalidates if mtime
356 			 * has changed.
357 			 */
358 			if (!timespec64_equal(&old_mtime, &new_mtime))
359 				inval = true;
360 		}
361 
362 		if (inval)
363 			invalidate_inode_pages2(inode->i_mapping);
364 	}
365 
366 	if (IS_ENABLED(CONFIG_FUSE_DAX))
367 		fuse_dax_dontcache(inode, attr->flags);
368 }
369 
370 static void fuse_init_submount_lookup(struct fuse_submount_lookup *sl,
371 				      u64 nodeid)
372 {
373 	sl->nodeid = nodeid;
374 	refcount_set(&sl->count, 1);
375 }
376 
377 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr,
378 			    struct fuse_conn *fc)
379 {
380 	inode->i_mode = attr->mode & S_IFMT;
381 	inode->i_size = attr->size;
382 	inode->i_mtime.tv_sec  = attr->mtime;
383 	inode->i_mtime.tv_nsec = attr->mtimensec;
384 	inode_set_ctime(inode, attr->ctime, attr->ctimensec);
385 	if (S_ISREG(inode->i_mode)) {
386 		fuse_init_common(inode);
387 		fuse_init_file_inode(inode, attr->flags);
388 	} else if (S_ISDIR(inode->i_mode))
389 		fuse_init_dir(inode);
390 	else if (S_ISLNK(inode->i_mode))
391 		fuse_init_symlink(inode);
392 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
393 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
394 		fuse_init_common(inode);
395 		init_special_inode(inode, inode->i_mode,
396 				   new_decode_dev(attr->rdev));
397 	} else
398 		BUG();
399 	/*
400 	 * Ensure that we don't cache acls for daemons without FUSE_POSIX_ACL
401 	 * so they see the exact same behavior as before.
402 	 */
403 	if (!fc->posix_acl)
404 		inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
405 }
406 
407 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
408 {
409 	u64 nodeid = *(u64 *) _nodeidp;
410 	if (get_node_id(inode) == nodeid)
411 		return 1;
412 	else
413 		return 0;
414 }
415 
416 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
417 {
418 	u64 nodeid = *(u64 *) _nodeidp;
419 	get_fuse_inode(inode)->nodeid = nodeid;
420 	return 0;
421 }
422 
423 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
424 			int generation, struct fuse_attr *attr,
425 			u64 attr_valid, u64 attr_version)
426 {
427 	struct inode *inode;
428 	struct fuse_inode *fi;
429 	struct fuse_conn *fc = get_fuse_conn_super(sb);
430 
431 	/*
432 	 * Auto mount points get their node id from the submount root, which is
433 	 * not a unique identifier within this filesystem.
434 	 *
435 	 * To avoid conflicts, do not place submount points into the inode hash
436 	 * table.
437 	 */
438 	if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
439 	    S_ISDIR(attr->mode)) {
440 		struct fuse_inode *fi;
441 
442 		inode = new_inode(sb);
443 		if (!inode)
444 			return NULL;
445 
446 		fuse_init_inode(inode, attr, fc);
447 		fi = get_fuse_inode(inode);
448 		fi->nodeid = nodeid;
449 		fi->submount_lookup = fuse_alloc_submount_lookup();
450 		if (!fi->submount_lookup) {
451 			iput(inode);
452 			return NULL;
453 		}
454 		/* Sets nlookup = 1 on fi->submount_lookup->nlookup */
455 		fuse_init_submount_lookup(fi->submount_lookup, nodeid);
456 		inode->i_flags |= S_AUTOMOUNT;
457 		goto done;
458 	}
459 
460 retry:
461 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
462 	if (!inode)
463 		return NULL;
464 
465 	if ((inode->i_state & I_NEW)) {
466 		inode->i_flags |= S_NOATIME;
467 		if (!fc->writeback_cache || !S_ISREG(attr->mode))
468 			inode->i_flags |= S_NOCMTIME;
469 		inode->i_generation = generation;
470 		fuse_init_inode(inode, attr, fc);
471 		unlock_new_inode(inode);
472 	} else if (fuse_stale_inode(inode, generation, attr)) {
473 		/* nodeid was reused, any I/O on the old inode should fail */
474 		fuse_make_bad(inode);
475 		if (inode != d_inode(sb->s_root)) {
476 			remove_inode_hash(inode);
477 			iput(inode);
478 			goto retry;
479 		}
480 	}
481 	fi = get_fuse_inode(inode);
482 	spin_lock(&fi->lock);
483 	fi->nlookup++;
484 	spin_unlock(&fi->lock);
485 done:
486 	fuse_change_attributes(inode, attr, NULL, attr_valid, attr_version);
487 
488 	return inode;
489 }
490 
491 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
492 			   struct fuse_mount **fm)
493 {
494 	struct fuse_mount *fm_iter;
495 	struct inode *inode;
496 
497 	WARN_ON(!rwsem_is_locked(&fc->killsb));
498 	list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
499 		if (!fm_iter->sb)
500 			continue;
501 
502 		inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
503 		if (inode) {
504 			if (fm)
505 				*fm = fm_iter;
506 			return inode;
507 		}
508 	}
509 
510 	return NULL;
511 }
512 
513 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
514 			     loff_t offset, loff_t len)
515 {
516 	struct fuse_inode *fi;
517 	struct inode *inode;
518 	pgoff_t pg_start;
519 	pgoff_t pg_end;
520 
521 	inode = fuse_ilookup(fc, nodeid, NULL);
522 	if (!inode)
523 		return -ENOENT;
524 
525 	fi = get_fuse_inode(inode);
526 	spin_lock(&fi->lock);
527 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
528 	spin_unlock(&fi->lock);
529 
530 	fuse_invalidate_attr(inode);
531 	forget_all_cached_acls(inode);
532 	if (offset >= 0) {
533 		pg_start = offset >> PAGE_SHIFT;
534 		if (len <= 0)
535 			pg_end = -1;
536 		else
537 			pg_end = (offset + len - 1) >> PAGE_SHIFT;
538 		invalidate_inode_pages2_range(inode->i_mapping,
539 					      pg_start, pg_end);
540 	}
541 	iput(inode);
542 	return 0;
543 }
544 
545 bool fuse_lock_inode(struct inode *inode)
546 {
547 	bool locked = false;
548 
549 	if (!get_fuse_conn(inode)->parallel_dirops) {
550 		mutex_lock(&get_fuse_inode(inode)->mutex);
551 		locked = true;
552 	}
553 
554 	return locked;
555 }
556 
557 void fuse_unlock_inode(struct inode *inode, bool locked)
558 {
559 	if (locked)
560 		mutex_unlock(&get_fuse_inode(inode)->mutex);
561 }
562 
563 static void fuse_umount_begin(struct super_block *sb)
564 {
565 	struct fuse_conn *fc = get_fuse_conn_super(sb);
566 
567 	if (fc->no_force_umount)
568 		return;
569 
570 	fuse_abort_conn(fc);
571 
572 	// Only retire block-device-based superblocks.
573 	if (sb->s_bdev != NULL)
574 		retire_super(sb);
575 }
576 
577 static void fuse_send_destroy(struct fuse_mount *fm)
578 {
579 	if (fm->fc->conn_init) {
580 		FUSE_ARGS(args);
581 
582 		args.opcode = FUSE_DESTROY;
583 		args.force = true;
584 		args.nocreds = true;
585 		fuse_simple_request(fm, &args);
586 	}
587 }
588 
589 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
590 {
591 	stbuf->f_type    = FUSE_SUPER_MAGIC;
592 	stbuf->f_bsize   = attr->bsize;
593 	stbuf->f_frsize  = attr->frsize;
594 	stbuf->f_blocks  = attr->blocks;
595 	stbuf->f_bfree   = attr->bfree;
596 	stbuf->f_bavail  = attr->bavail;
597 	stbuf->f_files   = attr->files;
598 	stbuf->f_ffree   = attr->ffree;
599 	stbuf->f_namelen = attr->namelen;
600 	/* fsid is left zero */
601 }
602 
603 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
604 {
605 	struct super_block *sb = dentry->d_sb;
606 	struct fuse_mount *fm = get_fuse_mount_super(sb);
607 	FUSE_ARGS(args);
608 	struct fuse_statfs_out outarg;
609 	int err;
610 
611 	if (!fuse_allow_current_process(fm->fc)) {
612 		buf->f_type = FUSE_SUPER_MAGIC;
613 		return 0;
614 	}
615 
616 	memset(&outarg, 0, sizeof(outarg));
617 	args.in_numargs = 0;
618 	args.opcode = FUSE_STATFS;
619 	args.nodeid = get_node_id(d_inode(dentry));
620 	args.out_numargs = 1;
621 	args.out_args[0].size = sizeof(outarg);
622 	args.out_args[0].value = &outarg;
623 	err = fuse_simple_request(fm, &args);
624 	if (!err)
625 		convert_fuse_statfs(buf, &outarg.st);
626 	return err;
627 }
628 
629 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
630 {
631 	struct fuse_sync_bucket *bucket;
632 
633 	bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
634 	if (bucket) {
635 		init_waitqueue_head(&bucket->waitq);
636 		/* Initial active count */
637 		atomic_set(&bucket->count, 1);
638 	}
639 	return bucket;
640 }
641 
642 static void fuse_sync_fs_writes(struct fuse_conn *fc)
643 {
644 	struct fuse_sync_bucket *bucket, *new_bucket;
645 	int count;
646 
647 	new_bucket = fuse_sync_bucket_alloc();
648 	spin_lock(&fc->lock);
649 	bucket = rcu_dereference_protected(fc->curr_bucket, 1);
650 	count = atomic_read(&bucket->count);
651 	WARN_ON(count < 1);
652 	/* No outstanding writes? */
653 	if (count == 1) {
654 		spin_unlock(&fc->lock);
655 		kfree(new_bucket);
656 		return;
657 	}
658 
659 	/*
660 	 * Completion of new bucket depends on completion of this bucket, so add
661 	 * one more count.
662 	 */
663 	atomic_inc(&new_bucket->count);
664 	rcu_assign_pointer(fc->curr_bucket, new_bucket);
665 	spin_unlock(&fc->lock);
666 	/*
667 	 * Drop initial active count.  At this point if all writes in this and
668 	 * ancestor buckets complete, the count will go to zero and this task
669 	 * will be woken up.
670 	 */
671 	atomic_dec(&bucket->count);
672 
673 	wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
674 
675 	/* Drop temp count on descendant bucket */
676 	fuse_sync_bucket_dec(new_bucket);
677 	kfree_rcu(bucket, rcu);
678 }
679 
680 static int fuse_sync_fs(struct super_block *sb, int wait)
681 {
682 	struct fuse_mount *fm = get_fuse_mount_super(sb);
683 	struct fuse_conn *fc = fm->fc;
684 	struct fuse_syncfs_in inarg;
685 	FUSE_ARGS(args);
686 	int err;
687 
688 	/*
689 	 * Userspace cannot handle the wait == 0 case.  Avoid a
690 	 * gratuitous roundtrip.
691 	 */
692 	if (!wait)
693 		return 0;
694 
695 	/* The filesystem is being unmounted.  Nothing to do. */
696 	if (!sb->s_root)
697 		return 0;
698 
699 	if (!fc->sync_fs)
700 		return 0;
701 
702 	fuse_sync_fs_writes(fc);
703 
704 	memset(&inarg, 0, sizeof(inarg));
705 	args.in_numargs = 1;
706 	args.in_args[0].size = sizeof(inarg);
707 	args.in_args[0].value = &inarg;
708 	args.opcode = FUSE_SYNCFS;
709 	args.nodeid = get_node_id(sb->s_root->d_inode);
710 	args.out_numargs = 0;
711 
712 	err = fuse_simple_request(fm, &args);
713 	if (err == -ENOSYS) {
714 		fc->sync_fs = 0;
715 		err = 0;
716 	}
717 
718 	return err;
719 }
720 
721 enum {
722 	OPT_SOURCE,
723 	OPT_SUBTYPE,
724 	OPT_FD,
725 	OPT_ROOTMODE,
726 	OPT_USER_ID,
727 	OPT_GROUP_ID,
728 	OPT_DEFAULT_PERMISSIONS,
729 	OPT_ALLOW_OTHER,
730 	OPT_MAX_READ,
731 	OPT_BLKSIZE,
732 	OPT_ERR
733 };
734 
735 static const struct fs_parameter_spec fuse_fs_parameters[] = {
736 	fsparam_string	("source",		OPT_SOURCE),
737 	fsparam_u32	("fd",			OPT_FD),
738 	fsparam_u32oct	("rootmode",		OPT_ROOTMODE),
739 	fsparam_u32	("user_id",		OPT_USER_ID),
740 	fsparam_u32	("group_id",		OPT_GROUP_ID),
741 	fsparam_flag	("default_permissions",	OPT_DEFAULT_PERMISSIONS),
742 	fsparam_flag	("allow_other",		OPT_ALLOW_OTHER),
743 	fsparam_u32	("max_read",		OPT_MAX_READ),
744 	fsparam_u32	("blksize",		OPT_BLKSIZE),
745 	fsparam_string	("subtype",		OPT_SUBTYPE),
746 	{}
747 };
748 
749 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
750 {
751 	struct fs_parse_result result;
752 	struct fuse_fs_context *ctx = fsc->fs_private;
753 	int opt;
754 	kuid_t kuid;
755 	kgid_t kgid;
756 
757 	if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
758 		/*
759 		 * Ignore options coming from mount(MS_REMOUNT) for backward
760 		 * compatibility.
761 		 */
762 		if (fsc->oldapi)
763 			return 0;
764 
765 		return invalfc(fsc, "No changes allowed in reconfigure");
766 	}
767 
768 	opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
769 	if (opt < 0)
770 		return opt;
771 
772 	switch (opt) {
773 	case OPT_SOURCE:
774 		if (fsc->source)
775 			return invalfc(fsc, "Multiple sources specified");
776 		fsc->source = param->string;
777 		param->string = NULL;
778 		break;
779 
780 	case OPT_SUBTYPE:
781 		if (ctx->subtype)
782 			return invalfc(fsc, "Multiple subtypes specified");
783 		ctx->subtype = param->string;
784 		param->string = NULL;
785 		return 0;
786 
787 	case OPT_FD:
788 		ctx->fd = result.uint_32;
789 		ctx->fd_present = true;
790 		break;
791 
792 	case OPT_ROOTMODE:
793 		if (!fuse_valid_type(result.uint_32))
794 			return invalfc(fsc, "Invalid rootmode");
795 		ctx->rootmode = result.uint_32;
796 		ctx->rootmode_present = true;
797 		break;
798 
799 	case OPT_USER_ID:
800 		kuid =  make_kuid(fsc->user_ns, result.uint_32);
801 		if (!uid_valid(kuid))
802 			return invalfc(fsc, "Invalid user_id");
803 		/*
804 		 * The requested uid must be representable in the
805 		 * filesystem's idmapping.
806 		 */
807 		if (!kuid_has_mapping(fsc->user_ns, kuid))
808 			return invalfc(fsc, "Invalid user_id");
809 		ctx->user_id = kuid;
810 		ctx->user_id_present = true;
811 		break;
812 
813 	case OPT_GROUP_ID:
814 		kgid = make_kgid(fsc->user_ns, result.uint_32);;
815 		if (!gid_valid(kgid))
816 			return invalfc(fsc, "Invalid group_id");
817 		/*
818 		 * The requested gid must be representable in the
819 		 * filesystem's idmapping.
820 		 */
821 		if (!kgid_has_mapping(fsc->user_ns, kgid))
822 			return invalfc(fsc, "Invalid group_id");
823 		ctx->group_id = kgid;
824 		ctx->group_id_present = true;
825 		break;
826 
827 	case OPT_DEFAULT_PERMISSIONS:
828 		ctx->default_permissions = true;
829 		break;
830 
831 	case OPT_ALLOW_OTHER:
832 		ctx->allow_other = true;
833 		break;
834 
835 	case OPT_MAX_READ:
836 		ctx->max_read = result.uint_32;
837 		break;
838 
839 	case OPT_BLKSIZE:
840 		if (!ctx->is_bdev)
841 			return invalfc(fsc, "blksize only supported for fuseblk");
842 		ctx->blksize = result.uint_32;
843 		break;
844 
845 	default:
846 		return -EINVAL;
847 	}
848 
849 	return 0;
850 }
851 
852 static void fuse_free_fsc(struct fs_context *fsc)
853 {
854 	struct fuse_fs_context *ctx = fsc->fs_private;
855 
856 	if (ctx) {
857 		kfree(ctx->subtype);
858 		kfree(ctx);
859 	}
860 }
861 
862 static int fuse_show_options(struct seq_file *m, struct dentry *root)
863 {
864 	struct super_block *sb = root->d_sb;
865 	struct fuse_conn *fc = get_fuse_conn_super(sb);
866 
867 	if (fc->legacy_opts_show) {
868 		seq_printf(m, ",user_id=%u",
869 			   from_kuid_munged(fc->user_ns, fc->user_id));
870 		seq_printf(m, ",group_id=%u",
871 			   from_kgid_munged(fc->user_ns, fc->group_id));
872 		if (fc->default_permissions)
873 			seq_puts(m, ",default_permissions");
874 		if (fc->allow_other)
875 			seq_puts(m, ",allow_other");
876 		if (fc->max_read != ~0)
877 			seq_printf(m, ",max_read=%u", fc->max_read);
878 		if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
879 			seq_printf(m, ",blksize=%lu", sb->s_blocksize);
880 	}
881 #ifdef CONFIG_FUSE_DAX
882 	if (fc->dax_mode == FUSE_DAX_ALWAYS)
883 		seq_puts(m, ",dax=always");
884 	else if (fc->dax_mode == FUSE_DAX_NEVER)
885 		seq_puts(m, ",dax=never");
886 	else if (fc->dax_mode == FUSE_DAX_INODE_USER)
887 		seq_puts(m, ",dax=inode");
888 #endif
889 
890 	return 0;
891 }
892 
893 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
894 			     const struct fuse_iqueue_ops *ops,
895 			     void *priv)
896 {
897 	memset(fiq, 0, sizeof(struct fuse_iqueue));
898 	spin_lock_init(&fiq->lock);
899 	init_waitqueue_head(&fiq->waitq);
900 	INIT_LIST_HEAD(&fiq->pending);
901 	INIT_LIST_HEAD(&fiq->interrupts);
902 	fiq->forget_list_tail = &fiq->forget_list_head;
903 	fiq->connected = 1;
904 	fiq->ops = ops;
905 	fiq->priv = priv;
906 }
907 
908 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
909 {
910 	unsigned int i;
911 
912 	spin_lock_init(&fpq->lock);
913 	for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
914 		INIT_LIST_HEAD(&fpq->processing[i]);
915 	INIT_LIST_HEAD(&fpq->io);
916 	fpq->connected = 1;
917 }
918 
919 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
920 		    struct user_namespace *user_ns,
921 		    const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
922 {
923 	memset(fc, 0, sizeof(*fc));
924 	spin_lock_init(&fc->lock);
925 	spin_lock_init(&fc->bg_lock);
926 	init_rwsem(&fc->killsb);
927 	refcount_set(&fc->count, 1);
928 	atomic_set(&fc->dev_count, 1);
929 	init_waitqueue_head(&fc->blocked_waitq);
930 	fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
931 	INIT_LIST_HEAD(&fc->bg_queue);
932 	INIT_LIST_HEAD(&fc->entry);
933 	INIT_LIST_HEAD(&fc->devices);
934 	atomic_set(&fc->num_waiting, 0);
935 	fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
936 	fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
937 	atomic64_set(&fc->khctr, 0);
938 	fc->polled_files = RB_ROOT;
939 	fc->blocked = 0;
940 	fc->initialized = 0;
941 	fc->connected = 1;
942 	atomic64_set(&fc->attr_version, 1);
943 	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
944 	fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
945 	fc->user_ns = get_user_ns(user_ns);
946 	fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
947 	fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
948 
949 	INIT_LIST_HEAD(&fc->mounts);
950 	list_add(&fm->fc_entry, &fc->mounts);
951 	fm->fc = fc;
952 }
953 EXPORT_SYMBOL_GPL(fuse_conn_init);
954 
955 static void delayed_release(struct rcu_head *p)
956 {
957 	struct fuse_conn *fc = container_of(p, struct fuse_conn, rcu);
958 
959 	put_user_ns(fc->user_ns);
960 	fc->release(fc);
961 }
962 
963 void fuse_conn_put(struct fuse_conn *fc)
964 {
965 	if (refcount_dec_and_test(&fc->count)) {
966 		struct fuse_iqueue *fiq = &fc->iq;
967 		struct fuse_sync_bucket *bucket;
968 
969 		if (IS_ENABLED(CONFIG_FUSE_DAX))
970 			fuse_dax_conn_free(fc);
971 		if (fiq->ops->release)
972 			fiq->ops->release(fiq);
973 		put_pid_ns(fc->pid_ns);
974 		bucket = rcu_dereference_protected(fc->curr_bucket, 1);
975 		if (bucket) {
976 			WARN_ON(atomic_read(&bucket->count) != 1);
977 			kfree(bucket);
978 		}
979 		call_rcu(&fc->rcu, delayed_release);
980 	}
981 }
982 EXPORT_SYMBOL_GPL(fuse_conn_put);
983 
984 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
985 {
986 	refcount_inc(&fc->count);
987 	return fc;
988 }
989 EXPORT_SYMBOL_GPL(fuse_conn_get);
990 
991 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
992 {
993 	struct fuse_attr attr;
994 	memset(&attr, 0, sizeof(attr));
995 
996 	attr.mode = mode;
997 	attr.ino = FUSE_ROOT_ID;
998 	attr.nlink = 1;
999 	return fuse_iget(sb, 1, 0, &attr, 0, 0);
1000 }
1001 
1002 struct fuse_inode_handle {
1003 	u64 nodeid;
1004 	u32 generation;
1005 };
1006 
1007 static struct dentry *fuse_get_dentry(struct super_block *sb,
1008 				      struct fuse_inode_handle *handle)
1009 {
1010 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1011 	struct inode *inode;
1012 	struct dentry *entry;
1013 	int err = -ESTALE;
1014 
1015 	if (handle->nodeid == 0)
1016 		goto out_err;
1017 
1018 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
1019 	if (!inode) {
1020 		struct fuse_entry_out outarg;
1021 		const struct qstr name = QSTR_INIT(".", 1);
1022 
1023 		if (!fc->export_support)
1024 			goto out_err;
1025 
1026 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
1027 				       &inode);
1028 		if (err && err != -ENOENT)
1029 			goto out_err;
1030 		if (err || !inode) {
1031 			err = -ESTALE;
1032 			goto out_err;
1033 		}
1034 		err = -EIO;
1035 		if (get_node_id(inode) != handle->nodeid)
1036 			goto out_iput;
1037 	}
1038 	err = -ESTALE;
1039 	if (inode->i_generation != handle->generation)
1040 		goto out_iput;
1041 
1042 	entry = d_obtain_alias(inode);
1043 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
1044 		fuse_invalidate_entry_cache(entry);
1045 
1046 	return entry;
1047 
1048  out_iput:
1049 	iput(inode);
1050  out_err:
1051 	return ERR_PTR(err);
1052 }
1053 
1054 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
1055 			   struct inode *parent)
1056 {
1057 	int len = parent ? 6 : 3;
1058 	u64 nodeid;
1059 	u32 generation;
1060 
1061 	if (*max_len < len) {
1062 		*max_len = len;
1063 		return  FILEID_INVALID;
1064 	}
1065 
1066 	nodeid = get_fuse_inode(inode)->nodeid;
1067 	generation = inode->i_generation;
1068 
1069 	fh[0] = (u32)(nodeid >> 32);
1070 	fh[1] = (u32)(nodeid & 0xffffffff);
1071 	fh[2] = generation;
1072 
1073 	if (parent) {
1074 		nodeid = get_fuse_inode(parent)->nodeid;
1075 		generation = parent->i_generation;
1076 
1077 		fh[3] = (u32)(nodeid >> 32);
1078 		fh[4] = (u32)(nodeid & 0xffffffff);
1079 		fh[5] = generation;
1080 	}
1081 
1082 	*max_len = len;
1083 	return parent ? 0x82 : 0x81;
1084 }
1085 
1086 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
1087 		struct fid *fid, int fh_len, int fh_type)
1088 {
1089 	struct fuse_inode_handle handle;
1090 
1091 	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
1092 		return NULL;
1093 
1094 	handle.nodeid = (u64) fid->raw[0] << 32;
1095 	handle.nodeid |= (u64) fid->raw[1];
1096 	handle.generation = fid->raw[2];
1097 	return fuse_get_dentry(sb, &handle);
1098 }
1099 
1100 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
1101 		struct fid *fid, int fh_len, int fh_type)
1102 {
1103 	struct fuse_inode_handle parent;
1104 
1105 	if (fh_type != 0x82 || fh_len < 6)
1106 		return NULL;
1107 
1108 	parent.nodeid = (u64) fid->raw[3] << 32;
1109 	parent.nodeid |= (u64) fid->raw[4];
1110 	parent.generation = fid->raw[5];
1111 	return fuse_get_dentry(sb, &parent);
1112 }
1113 
1114 static struct dentry *fuse_get_parent(struct dentry *child)
1115 {
1116 	struct inode *child_inode = d_inode(child);
1117 	struct fuse_conn *fc = get_fuse_conn(child_inode);
1118 	struct inode *inode;
1119 	struct dentry *parent;
1120 	struct fuse_entry_out outarg;
1121 	int err;
1122 
1123 	if (!fc->export_support)
1124 		return ERR_PTR(-ESTALE);
1125 
1126 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
1127 			       &dotdot_name, &outarg, &inode);
1128 	if (err) {
1129 		if (err == -ENOENT)
1130 			return ERR_PTR(-ESTALE);
1131 		return ERR_PTR(err);
1132 	}
1133 
1134 	parent = d_obtain_alias(inode);
1135 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
1136 		fuse_invalidate_entry_cache(parent);
1137 
1138 	return parent;
1139 }
1140 
1141 static const struct export_operations fuse_export_operations = {
1142 	.fh_to_dentry	= fuse_fh_to_dentry,
1143 	.fh_to_parent	= fuse_fh_to_parent,
1144 	.encode_fh	= fuse_encode_fh,
1145 	.get_parent	= fuse_get_parent,
1146 };
1147 
1148 static const struct super_operations fuse_super_operations = {
1149 	.alloc_inode    = fuse_alloc_inode,
1150 	.free_inode     = fuse_free_inode,
1151 	.evict_inode	= fuse_evict_inode,
1152 	.write_inode	= fuse_write_inode,
1153 	.drop_inode	= generic_delete_inode,
1154 	.umount_begin	= fuse_umount_begin,
1155 	.statfs		= fuse_statfs,
1156 	.sync_fs	= fuse_sync_fs,
1157 	.show_options	= fuse_show_options,
1158 };
1159 
1160 static void sanitize_global_limit(unsigned *limit)
1161 {
1162 	/*
1163 	 * The default maximum number of async requests is calculated to consume
1164 	 * 1/2^13 of the total memory, assuming 392 bytes per request.
1165 	 */
1166 	if (*limit == 0)
1167 		*limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1168 
1169 	if (*limit >= 1 << 16)
1170 		*limit = (1 << 16) - 1;
1171 }
1172 
1173 static int set_global_limit(const char *val, const struct kernel_param *kp)
1174 {
1175 	int rv;
1176 
1177 	rv = param_set_uint(val, kp);
1178 	if (rv)
1179 		return rv;
1180 
1181 	sanitize_global_limit((unsigned *)kp->arg);
1182 
1183 	return 0;
1184 }
1185 
1186 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1187 {
1188 	int cap_sys_admin = capable(CAP_SYS_ADMIN);
1189 
1190 	if (arg->minor < 13)
1191 		return;
1192 
1193 	sanitize_global_limit(&max_user_bgreq);
1194 	sanitize_global_limit(&max_user_congthresh);
1195 
1196 	spin_lock(&fc->bg_lock);
1197 	if (arg->max_background) {
1198 		fc->max_background = arg->max_background;
1199 
1200 		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1201 			fc->max_background = max_user_bgreq;
1202 	}
1203 	if (arg->congestion_threshold) {
1204 		fc->congestion_threshold = arg->congestion_threshold;
1205 
1206 		if (!cap_sys_admin &&
1207 		    fc->congestion_threshold > max_user_congthresh)
1208 			fc->congestion_threshold = max_user_congthresh;
1209 	}
1210 	spin_unlock(&fc->bg_lock);
1211 }
1212 
1213 struct fuse_init_args {
1214 	struct fuse_args args;
1215 	struct fuse_init_in in;
1216 	struct fuse_init_out out;
1217 };
1218 
1219 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1220 			       int error)
1221 {
1222 	struct fuse_conn *fc = fm->fc;
1223 	struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1224 	struct fuse_init_out *arg = &ia->out;
1225 	bool ok = true;
1226 
1227 	if (error || arg->major != FUSE_KERNEL_VERSION)
1228 		ok = false;
1229 	else {
1230 		unsigned long ra_pages;
1231 
1232 		process_init_limits(fc, arg);
1233 
1234 		if (arg->minor >= 6) {
1235 			u64 flags = arg->flags;
1236 
1237 			if (flags & FUSE_INIT_EXT)
1238 				flags |= (u64) arg->flags2 << 32;
1239 
1240 			ra_pages = arg->max_readahead / PAGE_SIZE;
1241 			if (flags & FUSE_ASYNC_READ)
1242 				fc->async_read = 1;
1243 			if (!(flags & FUSE_POSIX_LOCKS))
1244 				fc->no_lock = 1;
1245 			if (arg->minor >= 17) {
1246 				if (!(flags & FUSE_FLOCK_LOCKS))
1247 					fc->no_flock = 1;
1248 			} else {
1249 				if (!(flags & FUSE_POSIX_LOCKS))
1250 					fc->no_flock = 1;
1251 			}
1252 			if (flags & FUSE_ATOMIC_O_TRUNC)
1253 				fc->atomic_o_trunc = 1;
1254 			if (arg->minor >= 9) {
1255 				/* LOOKUP has dependency on proto version */
1256 				if (flags & FUSE_EXPORT_SUPPORT)
1257 					fc->export_support = 1;
1258 			}
1259 			if (flags & FUSE_BIG_WRITES)
1260 				fc->big_writes = 1;
1261 			if (flags & FUSE_DONT_MASK)
1262 				fc->dont_mask = 1;
1263 			if (flags & FUSE_AUTO_INVAL_DATA)
1264 				fc->auto_inval_data = 1;
1265 			else if (flags & FUSE_EXPLICIT_INVAL_DATA)
1266 				fc->explicit_inval_data = 1;
1267 			if (flags & FUSE_DO_READDIRPLUS) {
1268 				fc->do_readdirplus = 1;
1269 				if (flags & FUSE_READDIRPLUS_AUTO)
1270 					fc->readdirplus_auto = 1;
1271 			}
1272 			if (flags & FUSE_ASYNC_DIO)
1273 				fc->async_dio = 1;
1274 			if (flags & FUSE_WRITEBACK_CACHE)
1275 				fc->writeback_cache = 1;
1276 			if (flags & FUSE_PARALLEL_DIROPS)
1277 				fc->parallel_dirops = 1;
1278 			if (flags & FUSE_HANDLE_KILLPRIV)
1279 				fc->handle_killpriv = 1;
1280 			if (arg->time_gran && arg->time_gran <= 1000000000)
1281 				fm->sb->s_time_gran = arg->time_gran;
1282 			if ((flags & FUSE_POSIX_ACL)) {
1283 				fc->default_permissions = 1;
1284 				fc->posix_acl = 1;
1285 			}
1286 			if (flags & FUSE_CACHE_SYMLINKS)
1287 				fc->cache_symlinks = 1;
1288 			if (flags & FUSE_ABORT_ERROR)
1289 				fc->abort_err = 1;
1290 			if (flags & FUSE_MAX_PAGES) {
1291 				fc->max_pages =
1292 					min_t(unsigned int, fc->max_pages_limit,
1293 					max_t(unsigned int, arg->max_pages, 1));
1294 			}
1295 			if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1296 				if (flags & FUSE_MAP_ALIGNMENT &&
1297 				    !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1298 					ok = false;
1299 				}
1300 				if (flags & FUSE_HAS_INODE_DAX)
1301 					fc->inode_dax = 1;
1302 			}
1303 			if (flags & FUSE_HANDLE_KILLPRIV_V2) {
1304 				fc->handle_killpriv_v2 = 1;
1305 				fm->sb->s_flags |= SB_NOSEC;
1306 			}
1307 			if (flags & FUSE_SETXATTR_EXT)
1308 				fc->setxattr_ext = 1;
1309 			if (flags & FUSE_SECURITY_CTX)
1310 				fc->init_security = 1;
1311 			if (flags & FUSE_CREATE_SUPP_GROUP)
1312 				fc->create_supp_group = 1;
1313 			if (flags & FUSE_DIRECT_IO_ALLOW_MMAP)
1314 				fc->direct_io_allow_mmap = 1;
1315 		} else {
1316 			ra_pages = fc->max_read / PAGE_SIZE;
1317 			fc->no_lock = 1;
1318 			fc->no_flock = 1;
1319 		}
1320 
1321 		fm->sb->s_bdi->ra_pages =
1322 				min(fm->sb->s_bdi->ra_pages, ra_pages);
1323 		fc->minor = arg->minor;
1324 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1325 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
1326 		fc->conn_init = 1;
1327 	}
1328 	kfree(ia);
1329 
1330 	if (!ok) {
1331 		fc->conn_init = 0;
1332 		fc->conn_error = 1;
1333 	}
1334 
1335 	fuse_set_initialized(fc);
1336 	wake_up_all(&fc->blocked_waitq);
1337 }
1338 
1339 void fuse_send_init(struct fuse_mount *fm)
1340 {
1341 	struct fuse_init_args *ia;
1342 	u64 flags;
1343 
1344 	ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1345 
1346 	ia->in.major = FUSE_KERNEL_VERSION;
1347 	ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1348 	ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1349 	flags =
1350 		FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1351 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1352 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1353 		FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1354 		FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1355 		FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1356 		FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1357 		FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1358 		FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1359 		FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT | FUSE_INIT_EXT |
1360 		FUSE_SECURITY_CTX | FUSE_CREATE_SUPP_GROUP |
1361 		FUSE_HAS_EXPIRE_ONLY | FUSE_DIRECT_IO_ALLOW_MMAP;
1362 #ifdef CONFIG_FUSE_DAX
1363 	if (fm->fc->dax)
1364 		flags |= FUSE_MAP_ALIGNMENT;
1365 	if (fuse_is_inode_dax_mode(fm->fc->dax_mode))
1366 		flags |= FUSE_HAS_INODE_DAX;
1367 #endif
1368 	if (fm->fc->auto_submounts)
1369 		flags |= FUSE_SUBMOUNTS;
1370 
1371 	ia->in.flags = flags;
1372 	ia->in.flags2 = flags >> 32;
1373 
1374 	ia->args.opcode = FUSE_INIT;
1375 	ia->args.in_numargs = 1;
1376 	ia->args.in_args[0].size = sizeof(ia->in);
1377 	ia->args.in_args[0].value = &ia->in;
1378 	ia->args.out_numargs = 1;
1379 	/* Variable length argument used for backward compatibility
1380 	   with interface version < 7.5.  Rest of init_out is zeroed
1381 	   by do_get_request(), so a short reply is not a problem */
1382 	ia->args.out_argvar = true;
1383 	ia->args.out_args[0].size = sizeof(ia->out);
1384 	ia->args.out_args[0].value = &ia->out;
1385 	ia->args.force = true;
1386 	ia->args.nocreds = true;
1387 	ia->args.end = process_init_reply;
1388 
1389 	if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1390 		process_init_reply(fm, &ia->args, -ENOTCONN);
1391 }
1392 EXPORT_SYMBOL_GPL(fuse_send_init);
1393 
1394 void fuse_free_conn(struct fuse_conn *fc)
1395 {
1396 	WARN_ON(!list_empty(&fc->devices));
1397 	kfree(fc);
1398 }
1399 EXPORT_SYMBOL_GPL(fuse_free_conn);
1400 
1401 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1402 {
1403 	int err;
1404 	char *suffix = "";
1405 
1406 	if (sb->s_bdev) {
1407 		suffix = "-fuseblk";
1408 		/*
1409 		 * sb->s_bdi points to blkdev's bdi however we want to redirect
1410 		 * it to our private bdi...
1411 		 */
1412 		bdi_put(sb->s_bdi);
1413 		sb->s_bdi = &noop_backing_dev_info;
1414 	}
1415 	err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1416 				   MINOR(fc->dev), suffix);
1417 	if (err)
1418 		return err;
1419 
1420 	/* fuse does it's own writeback accounting */
1421 	sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1422 	sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1423 
1424 	/*
1425 	 * For a single fuse filesystem use max 1% of dirty +
1426 	 * writeback threshold.
1427 	 *
1428 	 * This gives about 1M of write buffer for memory maps on a
1429 	 * machine with 1G and 10% dirty_ratio, which should be more
1430 	 * than enough.
1431 	 *
1432 	 * Privileged users can raise it by writing to
1433 	 *
1434 	 *    /sys/class/bdi/<bdi>/max_ratio
1435 	 */
1436 	bdi_set_max_ratio(sb->s_bdi, 1);
1437 
1438 	return 0;
1439 }
1440 
1441 struct fuse_dev *fuse_dev_alloc(void)
1442 {
1443 	struct fuse_dev *fud;
1444 	struct list_head *pq;
1445 
1446 	fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1447 	if (!fud)
1448 		return NULL;
1449 
1450 	pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1451 	if (!pq) {
1452 		kfree(fud);
1453 		return NULL;
1454 	}
1455 
1456 	fud->pq.processing = pq;
1457 	fuse_pqueue_init(&fud->pq);
1458 
1459 	return fud;
1460 }
1461 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1462 
1463 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1464 {
1465 	fud->fc = fuse_conn_get(fc);
1466 	spin_lock(&fc->lock);
1467 	list_add_tail(&fud->entry, &fc->devices);
1468 	spin_unlock(&fc->lock);
1469 }
1470 EXPORT_SYMBOL_GPL(fuse_dev_install);
1471 
1472 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1473 {
1474 	struct fuse_dev *fud;
1475 
1476 	fud = fuse_dev_alloc();
1477 	if (!fud)
1478 		return NULL;
1479 
1480 	fuse_dev_install(fud, fc);
1481 	return fud;
1482 }
1483 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1484 
1485 void fuse_dev_free(struct fuse_dev *fud)
1486 {
1487 	struct fuse_conn *fc = fud->fc;
1488 
1489 	if (fc) {
1490 		spin_lock(&fc->lock);
1491 		list_del(&fud->entry);
1492 		spin_unlock(&fc->lock);
1493 
1494 		fuse_conn_put(fc);
1495 	}
1496 	kfree(fud->pq.processing);
1497 	kfree(fud);
1498 }
1499 EXPORT_SYMBOL_GPL(fuse_dev_free);
1500 
1501 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1502 				      const struct fuse_inode *fi)
1503 {
1504 	struct timespec64 ctime = inode_get_ctime(&fi->inode);
1505 
1506 	*attr = (struct fuse_attr){
1507 		.ino		= fi->inode.i_ino,
1508 		.size		= fi->inode.i_size,
1509 		.blocks		= fi->inode.i_blocks,
1510 		.atime		= fi->inode.i_atime.tv_sec,
1511 		.mtime		= fi->inode.i_mtime.tv_sec,
1512 		.ctime		= ctime.tv_sec,
1513 		.atimensec	= fi->inode.i_atime.tv_nsec,
1514 		.mtimensec	= fi->inode.i_mtime.tv_nsec,
1515 		.ctimensec	= ctime.tv_nsec,
1516 		.mode		= fi->inode.i_mode,
1517 		.nlink		= fi->inode.i_nlink,
1518 		.uid		= fi->inode.i_uid.val,
1519 		.gid		= fi->inode.i_gid.val,
1520 		.rdev		= fi->inode.i_rdev,
1521 		.blksize	= 1u << fi->inode.i_blkbits,
1522 	};
1523 }
1524 
1525 static void fuse_sb_defaults(struct super_block *sb)
1526 {
1527 	sb->s_magic = FUSE_SUPER_MAGIC;
1528 	sb->s_op = &fuse_super_operations;
1529 	sb->s_xattr = fuse_xattr_handlers;
1530 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1531 	sb->s_time_gran = 1;
1532 	sb->s_export_op = &fuse_export_operations;
1533 	sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1534 	if (sb->s_user_ns != &init_user_ns)
1535 		sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1536 	sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1537 }
1538 
1539 static int fuse_fill_super_submount(struct super_block *sb,
1540 				    struct fuse_inode *parent_fi)
1541 {
1542 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1543 	struct super_block *parent_sb = parent_fi->inode.i_sb;
1544 	struct fuse_attr root_attr;
1545 	struct inode *root;
1546 	struct fuse_submount_lookup *sl;
1547 	struct fuse_inode *fi;
1548 
1549 	fuse_sb_defaults(sb);
1550 	fm->sb = sb;
1551 
1552 	WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1553 	sb->s_bdi = bdi_get(parent_sb->s_bdi);
1554 
1555 	sb->s_xattr = parent_sb->s_xattr;
1556 	sb->s_time_gran = parent_sb->s_time_gran;
1557 	sb->s_blocksize = parent_sb->s_blocksize;
1558 	sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1559 	sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1560 	if (parent_sb->s_subtype && !sb->s_subtype)
1561 		return -ENOMEM;
1562 
1563 	fuse_fill_attr_from_inode(&root_attr, parent_fi);
1564 	root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1565 	/*
1566 	 * This inode is just a duplicate, so it is not looked up and
1567 	 * its nlookup should not be incremented.  fuse_iget() does
1568 	 * that, though, so undo it here.
1569 	 */
1570 	fi = get_fuse_inode(root);
1571 	fi->nlookup--;
1572 
1573 	sb->s_d_op = &fuse_dentry_operations;
1574 	sb->s_root = d_make_root(root);
1575 	if (!sb->s_root)
1576 		return -ENOMEM;
1577 
1578 	/*
1579 	 * Grab the parent's submount_lookup pointer and take a
1580 	 * reference on the shared nlookup from the parent.  This is to
1581 	 * prevent the last forget for this nodeid from getting
1582 	 * triggered until all users have finished with it.
1583 	 */
1584 	sl = parent_fi->submount_lookup;
1585 	WARN_ON(!sl);
1586 	if (sl) {
1587 		refcount_inc(&sl->count);
1588 		fi->submount_lookup = sl;
1589 	}
1590 
1591 	return 0;
1592 }
1593 
1594 /* Filesystem context private data holds the FUSE inode of the mount point */
1595 static int fuse_get_tree_submount(struct fs_context *fsc)
1596 {
1597 	struct fuse_mount *fm;
1598 	struct fuse_inode *mp_fi = fsc->fs_private;
1599 	struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1600 	struct super_block *sb;
1601 	int err;
1602 
1603 	fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1604 	if (!fm)
1605 		return -ENOMEM;
1606 
1607 	fm->fc = fuse_conn_get(fc);
1608 	fsc->s_fs_info = fm;
1609 	sb = sget_fc(fsc, NULL, set_anon_super_fc);
1610 	if (fsc->s_fs_info)
1611 		fuse_mount_destroy(fm);
1612 	if (IS_ERR(sb))
1613 		return PTR_ERR(sb);
1614 
1615 	/* Initialize superblock, making @mp_fi its root */
1616 	err = fuse_fill_super_submount(sb, mp_fi);
1617 	if (err) {
1618 		deactivate_locked_super(sb);
1619 		return err;
1620 	}
1621 
1622 	down_write(&fc->killsb);
1623 	list_add_tail(&fm->fc_entry, &fc->mounts);
1624 	up_write(&fc->killsb);
1625 
1626 	sb->s_flags |= SB_ACTIVE;
1627 	fsc->root = dget(sb->s_root);
1628 
1629 	return 0;
1630 }
1631 
1632 static const struct fs_context_operations fuse_context_submount_ops = {
1633 	.get_tree	= fuse_get_tree_submount,
1634 };
1635 
1636 int fuse_init_fs_context_submount(struct fs_context *fsc)
1637 {
1638 	fsc->ops = &fuse_context_submount_ops;
1639 	return 0;
1640 }
1641 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1642 
1643 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1644 {
1645 	struct fuse_dev *fud = NULL;
1646 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1647 	struct fuse_conn *fc = fm->fc;
1648 	struct inode *root;
1649 	struct dentry *root_dentry;
1650 	int err;
1651 
1652 	err = -EINVAL;
1653 	if (sb->s_flags & SB_MANDLOCK)
1654 		goto err;
1655 
1656 	rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1657 	fuse_sb_defaults(sb);
1658 
1659 	if (ctx->is_bdev) {
1660 #ifdef CONFIG_BLOCK
1661 		err = -EINVAL;
1662 		if (!sb_set_blocksize(sb, ctx->blksize))
1663 			goto err;
1664 #endif
1665 	} else {
1666 		sb->s_blocksize = PAGE_SIZE;
1667 		sb->s_blocksize_bits = PAGE_SHIFT;
1668 	}
1669 
1670 	sb->s_subtype = ctx->subtype;
1671 	ctx->subtype = NULL;
1672 	if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1673 		err = fuse_dax_conn_alloc(fc, ctx->dax_mode, ctx->dax_dev);
1674 		if (err)
1675 			goto err;
1676 	}
1677 
1678 	if (ctx->fudptr) {
1679 		err = -ENOMEM;
1680 		fud = fuse_dev_alloc_install(fc);
1681 		if (!fud)
1682 			goto err_free_dax;
1683 	}
1684 
1685 	fc->dev = sb->s_dev;
1686 	fm->sb = sb;
1687 	err = fuse_bdi_init(fc, sb);
1688 	if (err)
1689 		goto err_dev_free;
1690 
1691 	/* Handle umasking inside the fuse code */
1692 	if (sb->s_flags & SB_POSIXACL)
1693 		fc->dont_mask = 1;
1694 	sb->s_flags |= SB_POSIXACL;
1695 
1696 	fc->default_permissions = ctx->default_permissions;
1697 	fc->allow_other = ctx->allow_other;
1698 	fc->user_id = ctx->user_id;
1699 	fc->group_id = ctx->group_id;
1700 	fc->legacy_opts_show = ctx->legacy_opts_show;
1701 	fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1702 	fc->destroy = ctx->destroy;
1703 	fc->no_control = ctx->no_control;
1704 	fc->no_force_umount = ctx->no_force_umount;
1705 
1706 	err = -ENOMEM;
1707 	root = fuse_get_root_inode(sb, ctx->rootmode);
1708 	sb->s_d_op = &fuse_root_dentry_operations;
1709 	root_dentry = d_make_root(root);
1710 	if (!root_dentry)
1711 		goto err_dev_free;
1712 	/* Root dentry doesn't have .d_revalidate */
1713 	sb->s_d_op = &fuse_dentry_operations;
1714 
1715 	mutex_lock(&fuse_mutex);
1716 	err = -EINVAL;
1717 	if (ctx->fudptr && *ctx->fudptr)
1718 		goto err_unlock;
1719 
1720 	err = fuse_ctl_add_conn(fc);
1721 	if (err)
1722 		goto err_unlock;
1723 
1724 	list_add_tail(&fc->entry, &fuse_conn_list);
1725 	sb->s_root = root_dentry;
1726 	if (ctx->fudptr)
1727 		*ctx->fudptr = fud;
1728 	mutex_unlock(&fuse_mutex);
1729 	return 0;
1730 
1731  err_unlock:
1732 	mutex_unlock(&fuse_mutex);
1733 	dput(root_dentry);
1734  err_dev_free:
1735 	if (fud)
1736 		fuse_dev_free(fud);
1737  err_free_dax:
1738 	if (IS_ENABLED(CONFIG_FUSE_DAX))
1739 		fuse_dax_conn_free(fc);
1740  err:
1741 	return err;
1742 }
1743 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1744 
1745 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1746 {
1747 	struct fuse_fs_context *ctx = fsc->fs_private;
1748 	int err;
1749 
1750 	if (!ctx->file || !ctx->rootmode_present ||
1751 	    !ctx->user_id_present || !ctx->group_id_present)
1752 		return -EINVAL;
1753 
1754 	/*
1755 	 * Require mount to happen from the same user namespace which
1756 	 * opened /dev/fuse to prevent potential attacks.
1757 	 */
1758 	if ((ctx->file->f_op != &fuse_dev_operations) ||
1759 	    (ctx->file->f_cred->user_ns != sb->s_user_ns))
1760 		return -EINVAL;
1761 	ctx->fudptr = &ctx->file->private_data;
1762 
1763 	err = fuse_fill_super_common(sb, ctx);
1764 	if (err)
1765 		return err;
1766 	/* file->private_data shall be visible on all CPUs after this */
1767 	smp_mb();
1768 	fuse_send_init(get_fuse_mount_super(sb));
1769 	return 0;
1770 }
1771 
1772 /*
1773  * This is the path where user supplied an already initialized fuse dev.  In
1774  * this case never create a new super if the old one is gone.
1775  */
1776 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1777 {
1778 	return -ENOTCONN;
1779 }
1780 
1781 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1782 {
1783 
1784 	return fsc->sget_key == get_fuse_conn_super(sb);
1785 }
1786 
1787 static int fuse_get_tree(struct fs_context *fsc)
1788 {
1789 	struct fuse_fs_context *ctx = fsc->fs_private;
1790 	struct fuse_dev *fud;
1791 	struct fuse_conn *fc;
1792 	struct fuse_mount *fm;
1793 	struct super_block *sb;
1794 	int err;
1795 
1796 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1797 	if (!fc)
1798 		return -ENOMEM;
1799 
1800 	fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1801 	if (!fm) {
1802 		kfree(fc);
1803 		return -ENOMEM;
1804 	}
1805 
1806 	fuse_conn_init(fc, fm, fsc->user_ns, &fuse_dev_fiq_ops, NULL);
1807 	fc->release = fuse_free_conn;
1808 
1809 	fsc->s_fs_info = fm;
1810 
1811 	if (ctx->fd_present)
1812 		ctx->file = fget(ctx->fd);
1813 
1814 	if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1815 		err = get_tree_bdev(fsc, fuse_fill_super);
1816 		goto out;
1817 	}
1818 	/*
1819 	 * While block dev mount can be initialized with a dummy device fd
1820 	 * (found by device name), normal fuse mounts can't
1821 	 */
1822 	err = -EINVAL;
1823 	if (!ctx->file)
1824 		goto out;
1825 
1826 	/*
1827 	 * Allow creating a fuse mount with an already initialized fuse
1828 	 * connection
1829 	 */
1830 	fud = READ_ONCE(ctx->file->private_data);
1831 	if (ctx->file->f_op == &fuse_dev_operations && fud) {
1832 		fsc->sget_key = fud->fc;
1833 		sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1834 		err = PTR_ERR_OR_ZERO(sb);
1835 		if (!IS_ERR(sb))
1836 			fsc->root = dget(sb->s_root);
1837 	} else {
1838 		err = get_tree_nodev(fsc, fuse_fill_super);
1839 	}
1840 out:
1841 	if (fsc->s_fs_info)
1842 		fuse_mount_destroy(fm);
1843 	if (ctx->file)
1844 		fput(ctx->file);
1845 	return err;
1846 }
1847 
1848 static const struct fs_context_operations fuse_context_ops = {
1849 	.free		= fuse_free_fsc,
1850 	.parse_param	= fuse_parse_param,
1851 	.reconfigure	= fuse_reconfigure,
1852 	.get_tree	= fuse_get_tree,
1853 };
1854 
1855 /*
1856  * Set up the filesystem mount context.
1857  */
1858 static int fuse_init_fs_context(struct fs_context *fsc)
1859 {
1860 	struct fuse_fs_context *ctx;
1861 
1862 	ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1863 	if (!ctx)
1864 		return -ENOMEM;
1865 
1866 	ctx->max_read = ~0;
1867 	ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1868 	ctx->legacy_opts_show = true;
1869 
1870 #ifdef CONFIG_BLOCK
1871 	if (fsc->fs_type == &fuseblk_fs_type) {
1872 		ctx->is_bdev = true;
1873 		ctx->destroy = true;
1874 	}
1875 #endif
1876 
1877 	fsc->fs_private = ctx;
1878 	fsc->ops = &fuse_context_ops;
1879 	return 0;
1880 }
1881 
1882 bool fuse_mount_remove(struct fuse_mount *fm)
1883 {
1884 	struct fuse_conn *fc = fm->fc;
1885 	bool last = false;
1886 
1887 	down_write(&fc->killsb);
1888 	list_del_init(&fm->fc_entry);
1889 	if (list_empty(&fc->mounts))
1890 		last = true;
1891 	up_write(&fc->killsb);
1892 
1893 	return last;
1894 }
1895 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1896 
1897 void fuse_conn_destroy(struct fuse_mount *fm)
1898 {
1899 	struct fuse_conn *fc = fm->fc;
1900 
1901 	if (fc->destroy)
1902 		fuse_send_destroy(fm);
1903 
1904 	fuse_abort_conn(fc);
1905 	fuse_wait_aborted(fc);
1906 
1907 	if (!list_empty(&fc->entry)) {
1908 		mutex_lock(&fuse_mutex);
1909 		list_del(&fc->entry);
1910 		fuse_ctl_remove_conn(fc);
1911 		mutex_unlock(&fuse_mutex);
1912 	}
1913 }
1914 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1915 
1916 static void fuse_sb_destroy(struct super_block *sb)
1917 {
1918 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1919 	bool last;
1920 
1921 	if (sb->s_root) {
1922 		last = fuse_mount_remove(fm);
1923 		if (last)
1924 			fuse_conn_destroy(fm);
1925 	}
1926 }
1927 
1928 void fuse_mount_destroy(struct fuse_mount *fm)
1929 {
1930 	fuse_conn_put(fm->fc);
1931 	kfree_rcu(fm, rcu);
1932 }
1933 EXPORT_SYMBOL(fuse_mount_destroy);
1934 
1935 static void fuse_kill_sb_anon(struct super_block *sb)
1936 {
1937 	fuse_sb_destroy(sb);
1938 	kill_anon_super(sb);
1939 	fuse_mount_destroy(get_fuse_mount_super(sb));
1940 }
1941 
1942 static struct file_system_type fuse_fs_type = {
1943 	.owner		= THIS_MODULE,
1944 	.name		= "fuse",
1945 	.fs_flags	= FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1946 	.init_fs_context = fuse_init_fs_context,
1947 	.parameters	= fuse_fs_parameters,
1948 	.kill_sb	= fuse_kill_sb_anon,
1949 };
1950 MODULE_ALIAS_FS("fuse");
1951 
1952 #ifdef CONFIG_BLOCK
1953 static void fuse_kill_sb_blk(struct super_block *sb)
1954 {
1955 	fuse_sb_destroy(sb);
1956 	kill_block_super(sb);
1957 	fuse_mount_destroy(get_fuse_mount_super(sb));
1958 }
1959 
1960 static struct file_system_type fuseblk_fs_type = {
1961 	.owner		= THIS_MODULE,
1962 	.name		= "fuseblk",
1963 	.init_fs_context = fuse_init_fs_context,
1964 	.parameters	= fuse_fs_parameters,
1965 	.kill_sb	= fuse_kill_sb_blk,
1966 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1967 };
1968 MODULE_ALIAS_FS("fuseblk");
1969 
1970 static inline int register_fuseblk(void)
1971 {
1972 	return register_filesystem(&fuseblk_fs_type);
1973 }
1974 
1975 static inline void unregister_fuseblk(void)
1976 {
1977 	unregister_filesystem(&fuseblk_fs_type);
1978 }
1979 #else
1980 static inline int register_fuseblk(void)
1981 {
1982 	return 0;
1983 }
1984 
1985 static inline void unregister_fuseblk(void)
1986 {
1987 }
1988 #endif
1989 
1990 static void fuse_inode_init_once(void *foo)
1991 {
1992 	struct inode *inode = foo;
1993 
1994 	inode_init_once(inode);
1995 }
1996 
1997 static int __init fuse_fs_init(void)
1998 {
1999 	int err;
2000 
2001 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
2002 			sizeof(struct fuse_inode), 0,
2003 			SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
2004 			fuse_inode_init_once);
2005 	err = -ENOMEM;
2006 	if (!fuse_inode_cachep)
2007 		goto out;
2008 
2009 	err = register_fuseblk();
2010 	if (err)
2011 		goto out2;
2012 
2013 	err = register_filesystem(&fuse_fs_type);
2014 	if (err)
2015 		goto out3;
2016 
2017 	return 0;
2018 
2019  out3:
2020 	unregister_fuseblk();
2021  out2:
2022 	kmem_cache_destroy(fuse_inode_cachep);
2023  out:
2024 	return err;
2025 }
2026 
2027 static void fuse_fs_cleanup(void)
2028 {
2029 	unregister_filesystem(&fuse_fs_type);
2030 	unregister_fuseblk();
2031 
2032 	/*
2033 	 * Make sure all delayed rcu free inodes are flushed before we
2034 	 * destroy cache.
2035 	 */
2036 	rcu_barrier();
2037 	kmem_cache_destroy(fuse_inode_cachep);
2038 }
2039 
2040 static struct kobject *fuse_kobj;
2041 
2042 static int fuse_sysfs_init(void)
2043 {
2044 	int err;
2045 
2046 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
2047 	if (!fuse_kobj) {
2048 		err = -ENOMEM;
2049 		goto out_err;
2050 	}
2051 
2052 	err = sysfs_create_mount_point(fuse_kobj, "connections");
2053 	if (err)
2054 		goto out_fuse_unregister;
2055 
2056 	return 0;
2057 
2058  out_fuse_unregister:
2059 	kobject_put(fuse_kobj);
2060  out_err:
2061 	return err;
2062 }
2063 
2064 static void fuse_sysfs_cleanup(void)
2065 {
2066 	sysfs_remove_mount_point(fuse_kobj, "connections");
2067 	kobject_put(fuse_kobj);
2068 }
2069 
2070 static int __init fuse_init(void)
2071 {
2072 	int res;
2073 
2074 	pr_info("init (API version %i.%i)\n",
2075 		FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
2076 
2077 	INIT_LIST_HEAD(&fuse_conn_list);
2078 	res = fuse_fs_init();
2079 	if (res)
2080 		goto err;
2081 
2082 	res = fuse_dev_init();
2083 	if (res)
2084 		goto err_fs_cleanup;
2085 
2086 	res = fuse_sysfs_init();
2087 	if (res)
2088 		goto err_dev_cleanup;
2089 
2090 	res = fuse_ctl_init();
2091 	if (res)
2092 		goto err_sysfs_cleanup;
2093 
2094 	sanitize_global_limit(&max_user_bgreq);
2095 	sanitize_global_limit(&max_user_congthresh);
2096 
2097 	return 0;
2098 
2099  err_sysfs_cleanup:
2100 	fuse_sysfs_cleanup();
2101  err_dev_cleanup:
2102 	fuse_dev_cleanup();
2103  err_fs_cleanup:
2104 	fuse_fs_cleanup();
2105  err:
2106 	return res;
2107 }
2108 
2109 static void __exit fuse_exit(void)
2110 {
2111 	pr_debug("exit\n");
2112 
2113 	fuse_ctl_cleanup();
2114 	fuse_sysfs_cleanup();
2115 	fuse_fs_cleanup();
2116 	fuse_dev_cleanup();
2117 }
2118 
2119 module_init(fuse_init);
2120 module_exit(fuse_exit);
2121