xref: /linux/fs/fuse/inode.c (revision ba199dc909a20fe62270ae4e93f263987bb9d119)
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 	if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
115 		fuse_inode_backing_set(fi, NULL);
116 
117 	return &fi->inode;
118 
119 out_free_forget:
120 	kfree(fi->forget);
121 out_free:
122 	kmem_cache_free(fuse_inode_cachep, fi);
123 	return NULL;
124 }
125 
126 static void fuse_free_inode(struct inode *inode)
127 {
128 	struct fuse_inode *fi = get_fuse_inode(inode);
129 
130 	mutex_destroy(&fi->mutex);
131 	kfree(fi->forget);
132 #ifdef CONFIG_FUSE_DAX
133 	kfree(fi->dax);
134 #endif
135 	if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
136 		fuse_backing_put(fuse_inode_backing(fi));
137 
138 	kmem_cache_free(fuse_inode_cachep, fi);
139 }
140 
141 static void fuse_cleanup_submount_lookup(struct fuse_conn *fc,
142 					 struct fuse_submount_lookup *sl)
143 {
144 	if (!refcount_dec_and_test(&sl->count))
145 		return;
146 
147 	fuse_queue_forget(fc, sl->forget, sl->nodeid, 1);
148 	sl->forget = NULL;
149 	kfree(sl);
150 }
151 
152 static void fuse_evict_inode(struct inode *inode)
153 {
154 	struct fuse_inode *fi = get_fuse_inode(inode);
155 
156 	/* Will write inode on close/munmap and in all other dirtiers */
157 	WARN_ON(inode->i_state & I_DIRTY_INODE);
158 
159 	truncate_inode_pages_final(&inode->i_data);
160 	clear_inode(inode);
161 	if (inode->i_sb->s_flags & SB_ACTIVE) {
162 		struct fuse_conn *fc = get_fuse_conn(inode);
163 
164 		if (FUSE_IS_DAX(inode))
165 			fuse_dax_inode_cleanup(inode);
166 		if (fi->nlookup) {
167 			fuse_queue_forget(fc, fi->forget, fi->nodeid,
168 					  fi->nlookup);
169 			fi->forget = NULL;
170 		}
171 
172 		if (fi->submount_lookup) {
173 			fuse_cleanup_submount_lookup(fc, fi->submount_lookup);
174 			fi->submount_lookup = NULL;
175 		}
176 	}
177 	if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
178 		WARN_ON(fi->iocachectr != 0);
179 		WARN_ON(!list_empty(&fi->write_files));
180 		WARN_ON(!list_empty(&fi->queued_writes));
181 	}
182 }
183 
184 static int fuse_reconfigure(struct fs_context *fsc)
185 {
186 	struct super_block *sb = fsc->root->d_sb;
187 
188 	sync_filesystem(sb);
189 	if (fsc->sb_flags & SB_MANDLOCK)
190 		return -EINVAL;
191 
192 	return 0;
193 }
194 
195 /*
196  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
197  * so that it will fit.
198  */
199 static ino_t fuse_squash_ino(u64 ino64)
200 {
201 	ino_t ino = (ino_t) ino64;
202 	if (sizeof(ino_t) < sizeof(u64))
203 		ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
204 	return ino;
205 }
206 
207 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
208 				   struct fuse_statx *sx,
209 				   u64 attr_valid, u32 cache_mask)
210 {
211 	struct fuse_conn *fc = get_fuse_conn(inode);
212 	struct fuse_inode *fi = get_fuse_inode(inode);
213 
214 	lockdep_assert_held(&fi->lock);
215 
216 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
217 	fi->i_time = attr_valid;
218 	/* Clear basic stats from invalid mask */
219 	set_mask_bits(&fi->inval_mask, STATX_BASIC_STATS, 0);
220 
221 	inode->i_ino     = fuse_squash_ino(attr->ino);
222 	inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
223 	set_nlink(inode, attr->nlink);
224 	inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
225 	inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
226 	inode->i_blocks  = attr->blocks;
227 
228 	/* Sanitize nsecs */
229 	attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
230 	attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
231 	attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
232 
233 	inode_set_atime(inode, attr->atime, attr->atimensec);
234 	/* mtime from server may be stale due to local buffered write */
235 	if (!(cache_mask & STATX_MTIME)) {
236 		inode_set_mtime(inode, attr->mtime, attr->mtimensec);
237 	}
238 	if (!(cache_mask & STATX_CTIME)) {
239 		inode_set_ctime(inode, attr->ctime, attr->ctimensec);
240 	}
241 	if (sx) {
242 		/* Sanitize nsecs */
243 		sx->btime.tv_nsec =
244 			min_t(u32, sx->btime.tv_nsec, NSEC_PER_SEC - 1);
245 
246 		/*
247 		 * Btime has been queried, cache is valid (whether or not btime
248 		 * is available or not) so clear STATX_BTIME from inval_mask.
249 		 *
250 		 * Availability of the btime attribute is indicated in
251 		 * FUSE_I_BTIME
252 		 */
253 		set_mask_bits(&fi->inval_mask, STATX_BTIME, 0);
254 		if (sx->mask & STATX_BTIME) {
255 			set_bit(FUSE_I_BTIME, &fi->state);
256 			fi->i_btime.tv_sec = sx->btime.tv_sec;
257 			fi->i_btime.tv_nsec = sx->btime.tv_nsec;
258 		}
259 	}
260 
261 	if (attr->blksize != 0)
262 		inode->i_blkbits = ilog2(attr->blksize);
263 	else
264 		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
265 
266 	/*
267 	 * Don't set the sticky bit in i_mode, unless we want the VFS
268 	 * to check permissions.  This prevents failures due to the
269 	 * check in may_delete().
270 	 */
271 	fi->orig_i_mode = inode->i_mode;
272 	if (!fc->default_permissions)
273 		inode->i_mode &= ~S_ISVTX;
274 
275 	fi->orig_ino = attr->ino;
276 
277 	/*
278 	 * We are refreshing inode data and it is possible that another
279 	 * client set suid/sgid or security.capability xattr. So clear
280 	 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
281 	 * was set or if security.capability xattr was set. But we don't
282 	 * know if security.capability has been set or not. So clear it
283 	 * anyway. Its less efficient but should be safe.
284 	 */
285 	inode->i_flags &= ~S_NOSEC;
286 }
287 
288 u32 fuse_get_cache_mask(struct inode *inode)
289 {
290 	struct fuse_conn *fc = get_fuse_conn(inode);
291 
292 	if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
293 		return 0;
294 
295 	return STATX_MTIME | STATX_CTIME | STATX_SIZE;
296 }
297 
298 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
299 			    struct fuse_statx *sx,
300 			    u64 attr_valid, u64 attr_version)
301 {
302 	struct fuse_conn *fc = get_fuse_conn(inode);
303 	struct fuse_inode *fi = get_fuse_inode(inode);
304 	u32 cache_mask;
305 	loff_t oldsize;
306 	struct timespec64 old_mtime;
307 
308 	spin_lock(&fi->lock);
309 	/*
310 	 * In case of writeback_cache enabled, writes update mtime, ctime and
311 	 * may update i_size.  In these cases trust the cached value in the
312 	 * inode.
313 	 */
314 	cache_mask = fuse_get_cache_mask(inode);
315 	if (cache_mask & STATX_SIZE)
316 		attr->size = i_size_read(inode);
317 
318 	if (cache_mask & STATX_MTIME) {
319 		attr->mtime = inode_get_mtime_sec(inode);
320 		attr->mtimensec = inode_get_mtime_nsec(inode);
321 	}
322 	if (cache_mask & STATX_CTIME) {
323 		attr->ctime = inode_get_ctime_sec(inode);
324 		attr->ctimensec = inode_get_ctime_nsec(inode);
325 	}
326 
327 	if ((attr_version != 0 && fi->attr_version > attr_version) ||
328 	    test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
329 		spin_unlock(&fi->lock);
330 		return;
331 	}
332 
333 	old_mtime = inode_get_mtime(inode);
334 	fuse_change_attributes_common(inode, attr, sx, attr_valid, cache_mask);
335 
336 	oldsize = inode->i_size;
337 	/*
338 	 * In case of writeback_cache enabled, the cached writes beyond EOF
339 	 * extend local i_size without keeping userspace server in sync. So,
340 	 * attr->size coming from server can be stale. We cannot trust it.
341 	 */
342 	if (!(cache_mask & STATX_SIZE))
343 		i_size_write(inode, attr->size);
344 	spin_unlock(&fi->lock);
345 
346 	if (!cache_mask && S_ISREG(inode->i_mode)) {
347 		bool inval = false;
348 
349 		if (oldsize != attr->size) {
350 			truncate_pagecache(inode, attr->size);
351 			if (!fc->explicit_inval_data)
352 				inval = true;
353 		} else if (fc->auto_inval_data) {
354 			struct timespec64 new_mtime = {
355 				.tv_sec = attr->mtime,
356 				.tv_nsec = attr->mtimensec,
357 			};
358 
359 			/*
360 			 * Auto inval mode also checks and invalidates if mtime
361 			 * has changed.
362 			 */
363 			if (!timespec64_equal(&old_mtime, &new_mtime))
364 				inval = true;
365 		}
366 
367 		if (inval)
368 			invalidate_inode_pages2(inode->i_mapping);
369 	}
370 
371 	if (IS_ENABLED(CONFIG_FUSE_DAX))
372 		fuse_dax_dontcache(inode, attr->flags);
373 }
374 
375 static void fuse_init_submount_lookup(struct fuse_submount_lookup *sl,
376 				      u64 nodeid)
377 {
378 	sl->nodeid = nodeid;
379 	refcount_set(&sl->count, 1);
380 }
381 
382 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr,
383 			    struct fuse_conn *fc)
384 {
385 	inode->i_mode = attr->mode & S_IFMT;
386 	inode->i_size = attr->size;
387 	inode_set_mtime(inode, attr->mtime, attr->mtimensec);
388 	inode_set_ctime(inode, attr->ctime, attr->ctimensec);
389 	if (S_ISREG(inode->i_mode)) {
390 		fuse_init_common(inode);
391 		fuse_init_file_inode(inode, attr->flags);
392 	} else if (S_ISDIR(inode->i_mode))
393 		fuse_init_dir(inode);
394 	else if (S_ISLNK(inode->i_mode))
395 		fuse_init_symlink(inode);
396 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
397 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
398 		fuse_init_common(inode);
399 		init_special_inode(inode, inode->i_mode,
400 				   new_decode_dev(attr->rdev));
401 	} else
402 		BUG();
403 	/*
404 	 * Ensure that we don't cache acls for daemons without FUSE_POSIX_ACL
405 	 * so they see the exact same behavior as before.
406 	 */
407 	if (!fc->posix_acl)
408 		inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
409 }
410 
411 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
412 {
413 	u64 nodeid = *(u64 *) _nodeidp;
414 	if (get_node_id(inode) == nodeid)
415 		return 1;
416 	else
417 		return 0;
418 }
419 
420 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
421 {
422 	u64 nodeid = *(u64 *) _nodeidp;
423 	get_fuse_inode(inode)->nodeid = nodeid;
424 	return 0;
425 }
426 
427 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
428 			int generation, struct fuse_attr *attr,
429 			u64 attr_valid, u64 attr_version)
430 {
431 	struct inode *inode;
432 	struct fuse_inode *fi;
433 	struct fuse_conn *fc = get_fuse_conn_super(sb);
434 
435 	/*
436 	 * Auto mount points get their node id from the submount root, which is
437 	 * not a unique identifier within this filesystem.
438 	 *
439 	 * To avoid conflicts, do not place submount points into the inode hash
440 	 * table.
441 	 */
442 	if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
443 	    S_ISDIR(attr->mode)) {
444 		struct fuse_inode *fi;
445 
446 		inode = new_inode(sb);
447 		if (!inode)
448 			return NULL;
449 
450 		fuse_init_inode(inode, attr, fc);
451 		fi = get_fuse_inode(inode);
452 		fi->nodeid = nodeid;
453 		fi->submount_lookup = fuse_alloc_submount_lookup();
454 		if (!fi->submount_lookup) {
455 			iput(inode);
456 			return NULL;
457 		}
458 		/* Sets nlookup = 1 on fi->submount_lookup->nlookup */
459 		fuse_init_submount_lookup(fi->submount_lookup, nodeid);
460 		inode->i_flags |= S_AUTOMOUNT;
461 		goto done;
462 	}
463 
464 retry:
465 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
466 	if (!inode)
467 		return NULL;
468 
469 	if ((inode->i_state & I_NEW)) {
470 		inode->i_flags |= S_NOATIME;
471 		if (!fc->writeback_cache || !S_ISREG(attr->mode))
472 			inode->i_flags |= S_NOCMTIME;
473 		inode->i_generation = generation;
474 		fuse_init_inode(inode, attr, fc);
475 		unlock_new_inode(inode);
476 	} else if (fuse_stale_inode(inode, generation, attr)) {
477 		/* nodeid was reused, any I/O on the old inode should fail */
478 		fuse_make_bad(inode);
479 		if (inode != d_inode(sb->s_root)) {
480 			remove_inode_hash(inode);
481 			iput(inode);
482 			goto retry;
483 		}
484 	}
485 	fi = get_fuse_inode(inode);
486 	spin_lock(&fi->lock);
487 	fi->nlookup++;
488 	spin_unlock(&fi->lock);
489 done:
490 	fuse_change_attributes(inode, attr, NULL, attr_valid, attr_version);
491 
492 	return inode;
493 }
494 
495 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
496 			   struct fuse_mount **fm)
497 {
498 	struct fuse_mount *fm_iter;
499 	struct inode *inode;
500 
501 	WARN_ON(!rwsem_is_locked(&fc->killsb));
502 	list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
503 		if (!fm_iter->sb)
504 			continue;
505 
506 		inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
507 		if (inode) {
508 			if (fm)
509 				*fm = fm_iter;
510 			return inode;
511 		}
512 	}
513 
514 	return NULL;
515 }
516 
517 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
518 			     loff_t offset, loff_t len)
519 {
520 	struct fuse_inode *fi;
521 	struct inode *inode;
522 	pgoff_t pg_start;
523 	pgoff_t pg_end;
524 
525 	inode = fuse_ilookup(fc, nodeid, NULL);
526 	if (!inode)
527 		return -ENOENT;
528 
529 	fi = get_fuse_inode(inode);
530 	spin_lock(&fi->lock);
531 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
532 	spin_unlock(&fi->lock);
533 
534 	fuse_invalidate_attr(inode);
535 	forget_all_cached_acls(inode);
536 	if (offset >= 0) {
537 		pg_start = offset >> PAGE_SHIFT;
538 		if (len <= 0)
539 			pg_end = -1;
540 		else
541 			pg_end = (offset + len - 1) >> PAGE_SHIFT;
542 		invalidate_inode_pages2_range(inode->i_mapping,
543 					      pg_start, pg_end);
544 	}
545 	iput(inode);
546 	return 0;
547 }
548 
549 bool fuse_lock_inode(struct inode *inode)
550 {
551 	bool locked = false;
552 
553 	if (!get_fuse_conn(inode)->parallel_dirops) {
554 		mutex_lock(&get_fuse_inode(inode)->mutex);
555 		locked = true;
556 	}
557 
558 	return locked;
559 }
560 
561 void fuse_unlock_inode(struct inode *inode, bool locked)
562 {
563 	if (locked)
564 		mutex_unlock(&get_fuse_inode(inode)->mutex);
565 }
566 
567 static void fuse_umount_begin(struct super_block *sb)
568 {
569 	struct fuse_conn *fc = get_fuse_conn_super(sb);
570 
571 	if (fc->no_force_umount)
572 		return;
573 
574 	fuse_abort_conn(fc);
575 
576 	// Only retire block-device-based superblocks.
577 	if (sb->s_bdev != NULL)
578 		retire_super(sb);
579 }
580 
581 static void fuse_send_destroy(struct fuse_mount *fm)
582 {
583 	if (fm->fc->conn_init) {
584 		FUSE_ARGS(args);
585 
586 		args.opcode = FUSE_DESTROY;
587 		args.force = true;
588 		args.nocreds = true;
589 		fuse_simple_request(fm, &args);
590 	}
591 }
592 
593 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
594 {
595 	stbuf->f_type    = FUSE_SUPER_MAGIC;
596 	stbuf->f_bsize   = attr->bsize;
597 	stbuf->f_frsize  = attr->frsize;
598 	stbuf->f_blocks  = attr->blocks;
599 	stbuf->f_bfree   = attr->bfree;
600 	stbuf->f_bavail  = attr->bavail;
601 	stbuf->f_files   = attr->files;
602 	stbuf->f_ffree   = attr->ffree;
603 	stbuf->f_namelen = attr->namelen;
604 	/* fsid is left zero */
605 }
606 
607 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
608 {
609 	struct super_block *sb = dentry->d_sb;
610 	struct fuse_mount *fm = get_fuse_mount_super(sb);
611 	FUSE_ARGS(args);
612 	struct fuse_statfs_out outarg;
613 	int err;
614 
615 	if (!fuse_allow_current_process(fm->fc)) {
616 		buf->f_type = FUSE_SUPER_MAGIC;
617 		return 0;
618 	}
619 
620 	memset(&outarg, 0, sizeof(outarg));
621 	args.in_numargs = 0;
622 	args.opcode = FUSE_STATFS;
623 	args.nodeid = get_node_id(d_inode(dentry));
624 	args.out_numargs = 1;
625 	args.out_args[0].size = sizeof(outarg);
626 	args.out_args[0].value = &outarg;
627 	err = fuse_simple_request(fm, &args);
628 	if (!err)
629 		convert_fuse_statfs(buf, &outarg.st);
630 	return err;
631 }
632 
633 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
634 {
635 	struct fuse_sync_bucket *bucket;
636 
637 	bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
638 	if (bucket) {
639 		init_waitqueue_head(&bucket->waitq);
640 		/* Initial active count */
641 		atomic_set(&bucket->count, 1);
642 	}
643 	return bucket;
644 }
645 
646 static void fuse_sync_fs_writes(struct fuse_conn *fc)
647 {
648 	struct fuse_sync_bucket *bucket, *new_bucket;
649 	int count;
650 
651 	new_bucket = fuse_sync_bucket_alloc();
652 	spin_lock(&fc->lock);
653 	bucket = rcu_dereference_protected(fc->curr_bucket, 1);
654 	count = atomic_read(&bucket->count);
655 	WARN_ON(count < 1);
656 	/* No outstanding writes? */
657 	if (count == 1) {
658 		spin_unlock(&fc->lock);
659 		kfree(new_bucket);
660 		return;
661 	}
662 
663 	/*
664 	 * Completion of new bucket depends on completion of this bucket, so add
665 	 * one more count.
666 	 */
667 	atomic_inc(&new_bucket->count);
668 	rcu_assign_pointer(fc->curr_bucket, new_bucket);
669 	spin_unlock(&fc->lock);
670 	/*
671 	 * Drop initial active count.  At this point if all writes in this and
672 	 * ancestor buckets complete, the count will go to zero and this task
673 	 * will be woken up.
674 	 */
675 	atomic_dec(&bucket->count);
676 
677 	wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
678 
679 	/* Drop temp count on descendant bucket */
680 	fuse_sync_bucket_dec(new_bucket);
681 	kfree_rcu(bucket, rcu);
682 }
683 
684 static int fuse_sync_fs(struct super_block *sb, int wait)
685 {
686 	struct fuse_mount *fm = get_fuse_mount_super(sb);
687 	struct fuse_conn *fc = fm->fc;
688 	struct fuse_syncfs_in inarg;
689 	FUSE_ARGS(args);
690 	int err;
691 
692 	/*
693 	 * Userspace cannot handle the wait == 0 case.  Avoid a
694 	 * gratuitous roundtrip.
695 	 */
696 	if (!wait)
697 		return 0;
698 
699 	/* The filesystem is being unmounted.  Nothing to do. */
700 	if (!sb->s_root)
701 		return 0;
702 
703 	if (!fc->sync_fs)
704 		return 0;
705 
706 	fuse_sync_fs_writes(fc);
707 
708 	memset(&inarg, 0, sizeof(inarg));
709 	args.in_numargs = 1;
710 	args.in_args[0].size = sizeof(inarg);
711 	args.in_args[0].value = &inarg;
712 	args.opcode = FUSE_SYNCFS;
713 	args.nodeid = get_node_id(sb->s_root->d_inode);
714 	args.out_numargs = 0;
715 
716 	err = fuse_simple_request(fm, &args);
717 	if (err == -ENOSYS) {
718 		fc->sync_fs = 0;
719 		err = 0;
720 	}
721 
722 	return err;
723 }
724 
725 enum {
726 	OPT_SOURCE,
727 	OPT_SUBTYPE,
728 	OPT_FD,
729 	OPT_ROOTMODE,
730 	OPT_USER_ID,
731 	OPT_GROUP_ID,
732 	OPT_DEFAULT_PERMISSIONS,
733 	OPT_ALLOW_OTHER,
734 	OPT_MAX_READ,
735 	OPT_BLKSIZE,
736 	OPT_ERR
737 };
738 
739 static const struct fs_parameter_spec fuse_fs_parameters[] = {
740 	fsparam_string	("source",		OPT_SOURCE),
741 	fsparam_u32	("fd",			OPT_FD),
742 	fsparam_u32oct	("rootmode",		OPT_ROOTMODE),
743 	fsparam_uid	("user_id",		OPT_USER_ID),
744 	fsparam_gid	("group_id",		OPT_GROUP_ID),
745 	fsparam_flag	("default_permissions",	OPT_DEFAULT_PERMISSIONS),
746 	fsparam_flag	("allow_other",		OPT_ALLOW_OTHER),
747 	fsparam_u32	("max_read",		OPT_MAX_READ),
748 	fsparam_u32	("blksize",		OPT_BLKSIZE),
749 	fsparam_string	("subtype",		OPT_SUBTYPE),
750 	{}
751 };
752 
753 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
754 {
755 	struct fs_parse_result result;
756 	struct fuse_fs_context *ctx = fsc->fs_private;
757 	int opt;
758 	kuid_t kuid;
759 	kgid_t kgid;
760 
761 	if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
762 		/*
763 		 * Ignore options coming from mount(MS_REMOUNT) for backward
764 		 * compatibility.
765 		 */
766 		if (fsc->oldapi)
767 			return 0;
768 
769 		return invalfc(fsc, "No changes allowed in reconfigure");
770 	}
771 
772 	opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
773 	if (opt < 0)
774 		return opt;
775 
776 	switch (opt) {
777 	case OPT_SOURCE:
778 		if (fsc->source)
779 			return invalfc(fsc, "Multiple sources specified");
780 		fsc->source = param->string;
781 		param->string = NULL;
782 		break;
783 
784 	case OPT_SUBTYPE:
785 		if (ctx->subtype)
786 			return invalfc(fsc, "Multiple subtypes specified");
787 		ctx->subtype = param->string;
788 		param->string = NULL;
789 		return 0;
790 
791 	case OPT_FD:
792 		ctx->fd = result.uint_32;
793 		ctx->fd_present = true;
794 		break;
795 
796 	case OPT_ROOTMODE:
797 		if (!fuse_valid_type(result.uint_32))
798 			return invalfc(fsc, "Invalid rootmode");
799 		ctx->rootmode = result.uint_32;
800 		ctx->rootmode_present = true;
801 		break;
802 
803 	case OPT_USER_ID:
804 		kuid = result.uid;
805 		/*
806 		 * The requested uid must be representable in the
807 		 * filesystem's idmapping.
808 		 */
809 		if (!kuid_has_mapping(fsc->user_ns, kuid))
810 			return invalfc(fsc, "Invalid user_id");
811 		ctx->user_id = kuid;
812 		ctx->user_id_present = true;
813 		break;
814 
815 	case OPT_GROUP_ID:
816 		kgid = result.gid;
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 	if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
950 		fuse_backing_files_init(fc);
951 
952 	INIT_LIST_HEAD(&fc->mounts);
953 	list_add(&fm->fc_entry, &fc->mounts);
954 	fm->fc = fc;
955 }
956 EXPORT_SYMBOL_GPL(fuse_conn_init);
957 
958 static void delayed_release(struct rcu_head *p)
959 {
960 	struct fuse_conn *fc = container_of(p, struct fuse_conn, rcu);
961 
962 	put_user_ns(fc->user_ns);
963 	fc->release(fc);
964 }
965 
966 void fuse_conn_put(struct fuse_conn *fc)
967 {
968 	if (refcount_dec_and_test(&fc->count)) {
969 		struct fuse_iqueue *fiq = &fc->iq;
970 		struct fuse_sync_bucket *bucket;
971 
972 		if (IS_ENABLED(CONFIG_FUSE_DAX))
973 			fuse_dax_conn_free(fc);
974 		if (fiq->ops->release)
975 			fiq->ops->release(fiq);
976 		put_pid_ns(fc->pid_ns);
977 		bucket = rcu_dereference_protected(fc->curr_bucket, 1);
978 		if (bucket) {
979 			WARN_ON(atomic_read(&bucket->count) != 1);
980 			kfree(bucket);
981 		}
982 		if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
983 			fuse_backing_files_free(fc);
984 		call_rcu(&fc->rcu, delayed_release);
985 	}
986 }
987 EXPORT_SYMBOL_GPL(fuse_conn_put);
988 
989 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
990 {
991 	refcount_inc(&fc->count);
992 	return fc;
993 }
994 EXPORT_SYMBOL_GPL(fuse_conn_get);
995 
996 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
997 {
998 	struct fuse_attr attr;
999 	memset(&attr, 0, sizeof(attr));
1000 
1001 	attr.mode = mode;
1002 	attr.ino = FUSE_ROOT_ID;
1003 	attr.nlink = 1;
1004 	return fuse_iget(sb, FUSE_ROOT_ID, 0, &attr, 0, 0);
1005 }
1006 
1007 struct fuse_inode_handle {
1008 	u64 nodeid;
1009 	u32 generation;
1010 };
1011 
1012 static struct dentry *fuse_get_dentry(struct super_block *sb,
1013 				      struct fuse_inode_handle *handle)
1014 {
1015 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1016 	struct inode *inode;
1017 	struct dentry *entry;
1018 	int err = -ESTALE;
1019 
1020 	if (handle->nodeid == 0)
1021 		goto out_err;
1022 
1023 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
1024 	if (!inode) {
1025 		struct fuse_entry_out outarg;
1026 		const struct qstr name = QSTR_INIT(".", 1);
1027 
1028 		if (!fc->export_support)
1029 			goto out_err;
1030 
1031 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
1032 				       &inode);
1033 		if (err && err != -ENOENT)
1034 			goto out_err;
1035 		if (err || !inode) {
1036 			err = -ESTALE;
1037 			goto out_err;
1038 		}
1039 		err = -EIO;
1040 		if (get_node_id(inode) != handle->nodeid)
1041 			goto out_iput;
1042 	}
1043 	err = -ESTALE;
1044 	if (inode->i_generation != handle->generation)
1045 		goto out_iput;
1046 
1047 	entry = d_obtain_alias(inode);
1048 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
1049 		fuse_invalidate_entry_cache(entry);
1050 
1051 	return entry;
1052 
1053  out_iput:
1054 	iput(inode);
1055  out_err:
1056 	return ERR_PTR(err);
1057 }
1058 
1059 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
1060 			   struct inode *parent)
1061 {
1062 	int len = parent ? 6 : 3;
1063 	u64 nodeid;
1064 	u32 generation;
1065 
1066 	if (*max_len < len) {
1067 		*max_len = len;
1068 		return  FILEID_INVALID;
1069 	}
1070 
1071 	nodeid = get_fuse_inode(inode)->nodeid;
1072 	generation = inode->i_generation;
1073 
1074 	fh[0] = (u32)(nodeid >> 32);
1075 	fh[1] = (u32)(nodeid & 0xffffffff);
1076 	fh[2] = generation;
1077 
1078 	if (parent) {
1079 		nodeid = get_fuse_inode(parent)->nodeid;
1080 		generation = parent->i_generation;
1081 
1082 		fh[3] = (u32)(nodeid >> 32);
1083 		fh[4] = (u32)(nodeid & 0xffffffff);
1084 		fh[5] = generation;
1085 	}
1086 
1087 	*max_len = len;
1088 	return parent ? FILEID_INO64_GEN_PARENT : FILEID_INO64_GEN;
1089 }
1090 
1091 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
1092 		struct fid *fid, int fh_len, int fh_type)
1093 {
1094 	struct fuse_inode_handle handle;
1095 
1096 	if ((fh_type != FILEID_INO64_GEN &&
1097 	     fh_type != FILEID_INO64_GEN_PARENT) || fh_len < 3)
1098 		return NULL;
1099 
1100 	handle.nodeid = (u64) fid->raw[0] << 32;
1101 	handle.nodeid |= (u64) fid->raw[1];
1102 	handle.generation = fid->raw[2];
1103 	return fuse_get_dentry(sb, &handle);
1104 }
1105 
1106 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
1107 		struct fid *fid, int fh_len, int fh_type)
1108 {
1109 	struct fuse_inode_handle parent;
1110 
1111 	if (fh_type != FILEID_INO64_GEN_PARENT || fh_len < 6)
1112 		return NULL;
1113 
1114 	parent.nodeid = (u64) fid->raw[3] << 32;
1115 	parent.nodeid |= (u64) fid->raw[4];
1116 	parent.generation = fid->raw[5];
1117 	return fuse_get_dentry(sb, &parent);
1118 }
1119 
1120 static struct dentry *fuse_get_parent(struct dentry *child)
1121 {
1122 	struct inode *child_inode = d_inode(child);
1123 	struct fuse_conn *fc = get_fuse_conn(child_inode);
1124 	struct inode *inode;
1125 	struct dentry *parent;
1126 	struct fuse_entry_out outarg;
1127 	int err;
1128 
1129 	if (!fc->export_support)
1130 		return ERR_PTR(-ESTALE);
1131 
1132 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
1133 			       &dotdot_name, &outarg, &inode);
1134 	if (err) {
1135 		if (err == -ENOENT)
1136 			return ERR_PTR(-ESTALE);
1137 		return ERR_PTR(err);
1138 	}
1139 
1140 	parent = d_obtain_alias(inode);
1141 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
1142 		fuse_invalidate_entry_cache(parent);
1143 
1144 	return parent;
1145 }
1146 
1147 /* only for fid encoding; no support for file handle */
1148 static const struct export_operations fuse_export_fid_operations = {
1149 	.encode_fh	= fuse_encode_fh,
1150 };
1151 
1152 static const struct export_operations fuse_export_operations = {
1153 	.fh_to_dentry	= fuse_fh_to_dentry,
1154 	.fh_to_parent	= fuse_fh_to_parent,
1155 	.encode_fh	= fuse_encode_fh,
1156 	.get_parent	= fuse_get_parent,
1157 };
1158 
1159 static const struct super_operations fuse_super_operations = {
1160 	.alloc_inode    = fuse_alloc_inode,
1161 	.free_inode     = fuse_free_inode,
1162 	.evict_inode	= fuse_evict_inode,
1163 	.write_inode	= fuse_write_inode,
1164 	.drop_inode	= generic_delete_inode,
1165 	.umount_begin	= fuse_umount_begin,
1166 	.statfs		= fuse_statfs,
1167 	.sync_fs	= fuse_sync_fs,
1168 	.show_options	= fuse_show_options,
1169 };
1170 
1171 static void sanitize_global_limit(unsigned *limit)
1172 {
1173 	/*
1174 	 * The default maximum number of async requests is calculated to consume
1175 	 * 1/2^13 of the total memory, assuming 392 bytes per request.
1176 	 */
1177 	if (*limit == 0)
1178 		*limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1179 
1180 	if (*limit >= 1 << 16)
1181 		*limit = (1 << 16) - 1;
1182 }
1183 
1184 static int set_global_limit(const char *val, const struct kernel_param *kp)
1185 {
1186 	int rv;
1187 
1188 	rv = param_set_uint(val, kp);
1189 	if (rv)
1190 		return rv;
1191 
1192 	sanitize_global_limit((unsigned *)kp->arg);
1193 
1194 	return 0;
1195 }
1196 
1197 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1198 {
1199 	int cap_sys_admin = capable(CAP_SYS_ADMIN);
1200 
1201 	if (arg->minor < 13)
1202 		return;
1203 
1204 	sanitize_global_limit(&max_user_bgreq);
1205 	sanitize_global_limit(&max_user_congthresh);
1206 
1207 	spin_lock(&fc->bg_lock);
1208 	if (arg->max_background) {
1209 		fc->max_background = arg->max_background;
1210 
1211 		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1212 			fc->max_background = max_user_bgreq;
1213 	}
1214 	if (arg->congestion_threshold) {
1215 		fc->congestion_threshold = arg->congestion_threshold;
1216 
1217 		if (!cap_sys_admin &&
1218 		    fc->congestion_threshold > max_user_congthresh)
1219 			fc->congestion_threshold = max_user_congthresh;
1220 	}
1221 	spin_unlock(&fc->bg_lock);
1222 }
1223 
1224 struct fuse_init_args {
1225 	struct fuse_args args;
1226 	struct fuse_init_in in;
1227 	struct fuse_init_out out;
1228 };
1229 
1230 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1231 			       int error)
1232 {
1233 	struct fuse_conn *fc = fm->fc;
1234 	struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1235 	struct fuse_init_out *arg = &ia->out;
1236 	bool ok = true;
1237 
1238 	if (error || arg->major != FUSE_KERNEL_VERSION)
1239 		ok = false;
1240 	else {
1241 		unsigned long ra_pages;
1242 
1243 		process_init_limits(fc, arg);
1244 
1245 		if (arg->minor >= 6) {
1246 			u64 flags = arg->flags;
1247 
1248 			if (flags & FUSE_INIT_EXT)
1249 				flags |= (u64) arg->flags2 << 32;
1250 
1251 			ra_pages = arg->max_readahead / PAGE_SIZE;
1252 			if (flags & FUSE_ASYNC_READ)
1253 				fc->async_read = 1;
1254 			if (!(flags & FUSE_POSIX_LOCKS))
1255 				fc->no_lock = 1;
1256 			if (arg->minor >= 17) {
1257 				if (!(flags & FUSE_FLOCK_LOCKS))
1258 					fc->no_flock = 1;
1259 			} else {
1260 				if (!(flags & FUSE_POSIX_LOCKS))
1261 					fc->no_flock = 1;
1262 			}
1263 			if (flags & FUSE_ATOMIC_O_TRUNC)
1264 				fc->atomic_o_trunc = 1;
1265 			if (arg->minor >= 9) {
1266 				/* LOOKUP has dependency on proto version */
1267 				if (flags & FUSE_EXPORT_SUPPORT)
1268 					fc->export_support = 1;
1269 			}
1270 			if (flags & FUSE_BIG_WRITES)
1271 				fc->big_writes = 1;
1272 			if (flags & FUSE_DONT_MASK)
1273 				fc->dont_mask = 1;
1274 			if (flags & FUSE_AUTO_INVAL_DATA)
1275 				fc->auto_inval_data = 1;
1276 			else if (flags & FUSE_EXPLICIT_INVAL_DATA)
1277 				fc->explicit_inval_data = 1;
1278 			if (flags & FUSE_DO_READDIRPLUS) {
1279 				fc->do_readdirplus = 1;
1280 				if (flags & FUSE_READDIRPLUS_AUTO)
1281 					fc->readdirplus_auto = 1;
1282 			}
1283 			if (flags & FUSE_ASYNC_DIO)
1284 				fc->async_dio = 1;
1285 			if (flags & FUSE_WRITEBACK_CACHE)
1286 				fc->writeback_cache = 1;
1287 			if (flags & FUSE_PARALLEL_DIROPS)
1288 				fc->parallel_dirops = 1;
1289 			if (flags & FUSE_HANDLE_KILLPRIV)
1290 				fc->handle_killpriv = 1;
1291 			if (arg->time_gran && arg->time_gran <= 1000000000)
1292 				fm->sb->s_time_gran = arg->time_gran;
1293 			if ((flags & FUSE_POSIX_ACL)) {
1294 				fc->default_permissions = 1;
1295 				fc->posix_acl = 1;
1296 			}
1297 			if (flags & FUSE_CACHE_SYMLINKS)
1298 				fc->cache_symlinks = 1;
1299 			if (flags & FUSE_ABORT_ERROR)
1300 				fc->abort_err = 1;
1301 			if (flags & FUSE_MAX_PAGES) {
1302 				fc->max_pages =
1303 					min_t(unsigned int, fc->max_pages_limit,
1304 					max_t(unsigned int, arg->max_pages, 1));
1305 			}
1306 			if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1307 				if (flags & FUSE_MAP_ALIGNMENT &&
1308 				    !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1309 					ok = false;
1310 				}
1311 				if (flags & FUSE_HAS_INODE_DAX)
1312 					fc->inode_dax = 1;
1313 			}
1314 			if (flags & FUSE_HANDLE_KILLPRIV_V2) {
1315 				fc->handle_killpriv_v2 = 1;
1316 				fm->sb->s_flags |= SB_NOSEC;
1317 			}
1318 			if (flags & FUSE_SETXATTR_EXT)
1319 				fc->setxattr_ext = 1;
1320 			if (flags & FUSE_SECURITY_CTX)
1321 				fc->init_security = 1;
1322 			if (flags & FUSE_CREATE_SUPP_GROUP)
1323 				fc->create_supp_group = 1;
1324 			if (flags & FUSE_DIRECT_IO_ALLOW_MMAP)
1325 				fc->direct_io_allow_mmap = 1;
1326 			/*
1327 			 * max_stack_depth is the max stack depth of FUSE fs,
1328 			 * so it has to be at least 1 to support passthrough
1329 			 * to backing files.
1330 			 *
1331 			 * with max_stack_depth > 1, the backing files can be
1332 			 * on a stacked fs (e.g. overlayfs) themselves and with
1333 			 * max_stack_depth == 1, FUSE fs can be stacked as the
1334 			 * underlying fs of a stacked fs (e.g. overlayfs).
1335 			 */
1336 			if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH) &&
1337 			    (flags & FUSE_PASSTHROUGH) &&
1338 			    arg->max_stack_depth > 0 &&
1339 			    arg->max_stack_depth <= FILESYSTEM_MAX_STACK_DEPTH) {
1340 				fc->passthrough = 1;
1341 				fc->max_stack_depth = arg->max_stack_depth;
1342 				fm->sb->s_stack_depth = arg->max_stack_depth;
1343 			}
1344 			if (flags & FUSE_NO_EXPORT_SUPPORT)
1345 				fm->sb->s_export_op = &fuse_export_fid_operations;
1346 		} else {
1347 			ra_pages = fc->max_read / PAGE_SIZE;
1348 			fc->no_lock = 1;
1349 			fc->no_flock = 1;
1350 		}
1351 
1352 		fm->sb->s_bdi->ra_pages =
1353 				min(fm->sb->s_bdi->ra_pages, ra_pages);
1354 		fc->minor = arg->minor;
1355 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1356 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
1357 		fc->conn_init = 1;
1358 	}
1359 	kfree(ia);
1360 
1361 	if (!ok) {
1362 		fc->conn_init = 0;
1363 		fc->conn_error = 1;
1364 	}
1365 
1366 	fuse_set_initialized(fc);
1367 	wake_up_all(&fc->blocked_waitq);
1368 }
1369 
1370 void fuse_send_init(struct fuse_mount *fm)
1371 {
1372 	struct fuse_init_args *ia;
1373 	u64 flags;
1374 
1375 	ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1376 
1377 	ia->in.major = FUSE_KERNEL_VERSION;
1378 	ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1379 	ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1380 	flags =
1381 		FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1382 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1383 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1384 		FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1385 		FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1386 		FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1387 		FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1388 		FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1389 		FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1390 		FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT | FUSE_INIT_EXT |
1391 		FUSE_SECURITY_CTX | FUSE_CREATE_SUPP_GROUP |
1392 		FUSE_HAS_EXPIRE_ONLY | FUSE_DIRECT_IO_ALLOW_MMAP |
1393 		FUSE_NO_EXPORT_SUPPORT | FUSE_HAS_RESEND;
1394 #ifdef CONFIG_FUSE_DAX
1395 	if (fm->fc->dax)
1396 		flags |= FUSE_MAP_ALIGNMENT;
1397 	if (fuse_is_inode_dax_mode(fm->fc->dax_mode))
1398 		flags |= FUSE_HAS_INODE_DAX;
1399 #endif
1400 	if (fm->fc->auto_submounts)
1401 		flags |= FUSE_SUBMOUNTS;
1402 	if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
1403 		flags |= FUSE_PASSTHROUGH;
1404 
1405 	ia->in.flags = flags;
1406 	ia->in.flags2 = flags >> 32;
1407 
1408 	ia->args.opcode = FUSE_INIT;
1409 	ia->args.in_numargs = 1;
1410 	ia->args.in_args[0].size = sizeof(ia->in);
1411 	ia->args.in_args[0].value = &ia->in;
1412 	ia->args.out_numargs = 1;
1413 	/* Variable length argument used for backward compatibility
1414 	   with interface version < 7.5.  Rest of init_out is zeroed
1415 	   by do_get_request(), so a short reply is not a problem */
1416 	ia->args.out_argvar = true;
1417 	ia->args.out_args[0].size = sizeof(ia->out);
1418 	ia->args.out_args[0].value = &ia->out;
1419 	ia->args.force = true;
1420 	ia->args.nocreds = true;
1421 	ia->args.end = process_init_reply;
1422 
1423 	if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1424 		process_init_reply(fm, &ia->args, -ENOTCONN);
1425 }
1426 EXPORT_SYMBOL_GPL(fuse_send_init);
1427 
1428 void fuse_free_conn(struct fuse_conn *fc)
1429 {
1430 	WARN_ON(!list_empty(&fc->devices));
1431 	kfree(fc);
1432 }
1433 EXPORT_SYMBOL_GPL(fuse_free_conn);
1434 
1435 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1436 {
1437 	int err;
1438 	char *suffix = "";
1439 
1440 	if (sb->s_bdev) {
1441 		suffix = "-fuseblk";
1442 		/*
1443 		 * sb->s_bdi points to blkdev's bdi however we want to redirect
1444 		 * it to our private bdi...
1445 		 */
1446 		bdi_put(sb->s_bdi);
1447 		sb->s_bdi = &noop_backing_dev_info;
1448 	}
1449 	err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1450 				   MINOR(fc->dev), suffix);
1451 	if (err)
1452 		return err;
1453 
1454 	/* fuse does it's own writeback accounting */
1455 	sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1456 	sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1457 
1458 	/*
1459 	 * For a single fuse filesystem use max 1% of dirty +
1460 	 * writeback threshold.
1461 	 *
1462 	 * This gives about 1M of write buffer for memory maps on a
1463 	 * machine with 1G and 10% dirty_ratio, which should be more
1464 	 * than enough.
1465 	 *
1466 	 * Privileged users can raise it by writing to
1467 	 *
1468 	 *    /sys/class/bdi/<bdi>/max_ratio
1469 	 */
1470 	bdi_set_max_ratio(sb->s_bdi, 1);
1471 
1472 	return 0;
1473 }
1474 
1475 struct fuse_dev *fuse_dev_alloc(void)
1476 {
1477 	struct fuse_dev *fud;
1478 	struct list_head *pq;
1479 
1480 	fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1481 	if (!fud)
1482 		return NULL;
1483 
1484 	pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1485 	if (!pq) {
1486 		kfree(fud);
1487 		return NULL;
1488 	}
1489 
1490 	fud->pq.processing = pq;
1491 	fuse_pqueue_init(&fud->pq);
1492 
1493 	return fud;
1494 }
1495 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1496 
1497 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1498 {
1499 	fud->fc = fuse_conn_get(fc);
1500 	spin_lock(&fc->lock);
1501 	list_add_tail(&fud->entry, &fc->devices);
1502 	spin_unlock(&fc->lock);
1503 }
1504 EXPORT_SYMBOL_GPL(fuse_dev_install);
1505 
1506 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1507 {
1508 	struct fuse_dev *fud;
1509 
1510 	fud = fuse_dev_alloc();
1511 	if (!fud)
1512 		return NULL;
1513 
1514 	fuse_dev_install(fud, fc);
1515 	return fud;
1516 }
1517 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1518 
1519 void fuse_dev_free(struct fuse_dev *fud)
1520 {
1521 	struct fuse_conn *fc = fud->fc;
1522 
1523 	if (fc) {
1524 		spin_lock(&fc->lock);
1525 		list_del(&fud->entry);
1526 		spin_unlock(&fc->lock);
1527 
1528 		fuse_conn_put(fc);
1529 	}
1530 	kfree(fud->pq.processing);
1531 	kfree(fud);
1532 }
1533 EXPORT_SYMBOL_GPL(fuse_dev_free);
1534 
1535 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1536 				      const struct fuse_inode *fi)
1537 {
1538 	struct timespec64 atime = inode_get_atime(&fi->inode);
1539 	struct timespec64 mtime = inode_get_mtime(&fi->inode);
1540 	struct timespec64 ctime = inode_get_ctime(&fi->inode);
1541 
1542 	*attr = (struct fuse_attr){
1543 		.ino		= fi->inode.i_ino,
1544 		.size		= fi->inode.i_size,
1545 		.blocks		= fi->inode.i_blocks,
1546 		.atime		= atime.tv_sec,
1547 		.mtime		= mtime.tv_sec,
1548 		.ctime		= ctime.tv_sec,
1549 		.atimensec	= atime.tv_nsec,
1550 		.mtimensec	= mtime.tv_nsec,
1551 		.ctimensec	= ctime.tv_nsec,
1552 		.mode		= fi->inode.i_mode,
1553 		.nlink		= fi->inode.i_nlink,
1554 		.uid		= __kuid_val(fi->inode.i_uid),
1555 		.gid		= __kgid_val(fi->inode.i_gid),
1556 		.rdev		= fi->inode.i_rdev,
1557 		.blksize	= 1u << fi->inode.i_blkbits,
1558 	};
1559 }
1560 
1561 static void fuse_sb_defaults(struct super_block *sb)
1562 {
1563 	sb->s_magic = FUSE_SUPER_MAGIC;
1564 	sb->s_op = &fuse_super_operations;
1565 	sb->s_xattr = fuse_xattr_handlers;
1566 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1567 	sb->s_time_gran = 1;
1568 	sb->s_export_op = &fuse_export_operations;
1569 	sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1570 	if (sb->s_user_ns != &init_user_ns)
1571 		sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1572 	sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1573 }
1574 
1575 static int fuse_fill_super_submount(struct super_block *sb,
1576 				    struct fuse_inode *parent_fi)
1577 {
1578 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1579 	struct super_block *parent_sb = parent_fi->inode.i_sb;
1580 	struct fuse_attr root_attr;
1581 	struct inode *root;
1582 	struct fuse_submount_lookup *sl;
1583 	struct fuse_inode *fi;
1584 
1585 	fuse_sb_defaults(sb);
1586 	fm->sb = sb;
1587 
1588 	WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1589 	sb->s_bdi = bdi_get(parent_sb->s_bdi);
1590 
1591 	sb->s_xattr = parent_sb->s_xattr;
1592 	sb->s_export_op = parent_sb->s_export_op;
1593 	sb->s_time_gran = parent_sb->s_time_gran;
1594 	sb->s_blocksize = parent_sb->s_blocksize;
1595 	sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1596 	sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1597 	if (parent_sb->s_subtype && !sb->s_subtype)
1598 		return -ENOMEM;
1599 
1600 	fuse_fill_attr_from_inode(&root_attr, parent_fi);
1601 	root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1602 	/*
1603 	 * This inode is just a duplicate, so it is not looked up and
1604 	 * its nlookup should not be incremented.  fuse_iget() does
1605 	 * that, though, so undo it here.
1606 	 */
1607 	fi = get_fuse_inode(root);
1608 	fi->nlookup--;
1609 
1610 	sb->s_d_op = &fuse_dentry_operations;
1611 	sb->s_root = d_make_root(root);
1612 	if (!sb->s_root)
1613 		return -ENOMEM;
1614 
1615 	/*
1616 	 * Grab the parent's submount_lookup pointer and take a
1617 	 * reference on the shared nlookup from the parent.  This is to
1618 	 * prevent the last forget for this nodeid from getting
1619 	 * triggered until all users have finished with it.
1620 	 */
1621 	sl = parent_fi->submount_lookup;
1622 	WARN_ON(!sl);
1623 	if (sl) {
1624 		refcount_inc(&sl->count);
1625 		fi->submount_lookup = sl;
1626 	}
1627 
1628 	return 0;
1629 }
1630 
1631 /* Filesystem context private data holds the FUSE inode of the mount point */
1632 static int fuse_get_tree_submount(struct fs_context *fsc)
1633 {
1634 	struct fuse_mount *fm;
1635 	struct fuse_inode *mp_fi = fsc->fs_private;
1636 	struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1637 	struct super_block *sb;
1638 	int err;
1639 
1640 	fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1641 	if (!fm)
1642 		return -ENOMEM;
1643 
1644 	fm->fc = fuse_conn_get(fc);
1645 	fsc->s_fs_info = fm;
1646 	sb = sget_fc(fsc, NULL, set_anon_super_fc);
1647 	if (fsc->s_fs_info)
1648 		fuse_mount_destroy(fm);
1649 	if (IS_ERR(sb))
1650 		return PTR_ERR(sb);
1651 
1652 	/* Initialize superblock, making @mp_fi its root */
1653 	err = fuse_fill_super_submount(sb, mp_fi);
1654 	if (err) {
1655 		deactivate_locked_super(sb);
1656 		return err;
1657 	}
1658 
1659 	down_write(&fc->killsb);
1660 	list_add_tail(&fm->fc_entry, &fc->mounts);
1661 	up_write(&fc->killsb);
1662 
1663 	sb->s_flags |= SB_ACTIVE;
1664 	fsc->root = dget(sb->s_root);
1665 
1666 	return 0;
1667 }
1668 
1669 static const struct fs_context_operations fuse_context_submount_ops = {
1670 	.get_tree	= fuse_get_tree_submount,
1671 };
1672 
1673 int fuse_init_fs_context_submount(struct fs_context *fsc)
1674 {
1675 	fsc->ops = &fuse_context_submount_ops;
1676 	return 0;
1677 }
1678 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1679 
1680 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1681 {
1682 	struct fuse_dev *fud = NULL;
1683 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1684 	struct fuse_conn *fc = fm->fc;
1685 	struct inode *root;
1686 	struct dentry *root_dentry;
1687 	int err;
1688 
1689 	err = -EINVAL;
1690 	if (sb->s_flags & SB_MANDLOCK)
1691 		goto err;
1692 
1693 	rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1694 	fuse_sb_defaults(sb);
1695 
1696 	if (ctx->is_bdev) {
1697 #ifdef CONFIG_BLOCK
1698 		err = -EINVAL;
1699 		if (!sb_set_blocksize(sb, ctx->blksize))
1700 			goto err;
1701 #endif
1702 	} else {
1703 		sb->s_blocksize = PAGE_SIZE;
1704 		sb->s_blocksize_bits = PAGE_SHIFT;
1705 	}
1706 
1707 	sb->s_subtype = ctx->subtype;
1708 	ctx->subtype = NULL;
1709 	if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1710 		err = fuse_dax_conn_alloc(fc, ctx->dax_mode, ctx->dax_dev);
1711 		if (err)
1712 			goto err;
1713 	}
1714 
1715 	if (ctx->fudptr) {
1716 		err = -ENOMEM;
1717 		fud = fuse_dev_alloc_install(fc);
1718 		if (!fud)
1719 			goto err_free_dax;
1720 	}
1721 
1722 	fc->dev = sb->s_dev;
1723 	fm->sb = sb;
1724 	err = fuse_bdi_init(fc, sb);
1725 	if (err)
1726 		goto err_dev_free;
1727 
1728 	/* Handle umasking inside the fuse code */
1729 	if (sb->s_flags & SB_POSIXACL)
1730 		fc->dont_mask = 1;
1731 	sb->s_flags |= SB_POSIXACL;
1732 
1733 	fc->default_permissions = ctx->default_permissions;
1734 	fc->allow_other = ctx->allow_other;
1735 	fc->user_id = ctx->user_id;
1736 	fc->group_id = ctx->group_id;
1737 	fc->legacy_opts_show = ctx->legacy_opts_show;
1738 	fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1739 	fc->destroy = ctx->destroy;
1740 	fc->no_control = ctx->no_control;
1741 	fc->no_force_umount = ctx->no_force_umount;
1742 
1743 	err = -ENOMEM;
1744 	root = fuse_get_root_inode(sb, ctx->rootmode);
1745 	sb->s_d_op = &fuse_root_dentry_operations;
1746 	root_dentry = d_make_root(root);
1747 	if (!root_dentry)
1748 		goto err_dev_free;
1749 	/* Root dentry doesn't have .d_revalidate */
1750 	sb->s_d_op = &fuse_dentry_operations;
1751 
1752 	mutex_lock(&fuse_mutex);
1753 	err = -EINVAL;
1754 	if (ctx->fudptr && *ctx->fudptr)
1755 		goto err_unlock;
1756 
1757 	err = fuse_ctl_add_conn(fc);
1758 	if (err)
1759 		goto err_unlock;
1760 
1761 	list_add_tail(&fc->entry, &fuse_conn_list);
1762 	sb->s_root = root_dentry;
1763 	if (ctx->fudptr)
1764 		*ctx->fudptr = fud;
1765 	mutex_unlock(&fuse_mutex);
1766 	return 0;
1767 
1768  err_unlock:
1769 	mutex_unlock(&fuse_mutex);
1770 	dput(root_dentry);
1771  err_dev_free:
1772 	if (fud)
1773 		fuse_dev_free(fud);
1774  err_free_dax:
1775 	if (IS_ENABLED(CONFIG_FUSE_DAX))
1776 		fuse_dax_conn_free(fc);
1777  err:
1778 	return err;
1779 }
1780 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1781 
1782 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1783 {
1784 	struct fuse_fs_context *ctx = fsc->fs_private;
1785 	int err;
1786 
1787 	if (!ctx->file || !ctx->rootmode_present ||
1788 	    !ctx->user_id_present || !ctx->group_id_present)
1789 		return -EINVAL;
1790 
1791 	/*
1792 	 * Require mount to happen from the same user namespace which
1793 	 * opened /dev/fuse to prevent potential attacks.
1794 	 */
1795 	if ((ctx->file->f_op != &fuse_dev_operations) ||
1796 	    (ctx->file->f_cred->user_ns != sb->s_user_ns))
1797 		return -EINVAL;
1798 	ctx->fudptr = &ctx->file->private_data;
1799 
1800 	err = fuse_fill_super_common(sb, ctx);
1801 	if (err)
1802 		return err;
1803 	/* file->private_data shall be visible on all CPUs after this */
1804 	smp_mb();
1805 	fuse_send_init(get_fuse_mount_super(sb));
1806 	return 0;
1807 }
1808 
1809 /*
1810  * This is the path where user supplied an already initialized fuse dev.  In
1811  * this case never create a new super if the old one is gone.
1812  */
1813 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1814 {
1815 	return -ENOTCONN;
1816 }
1817 
1818 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1819 {
1820 
1821 	return fsc->sget_key == get_fuse_conn_super(sb);
1822 }
1823 
1824 static int fuse_get_tree(struct fs_context *fsc)
1825 {
1826 	struct fuse_fs_context *ctx = fsc->fs_private;
1827 	struct fuse_dev *fud;
1828 	struct fuse_conn *fc;
1829 	struct fuse_mount *fm;
1830 	struct super_block *sb;
1831 	int err;
1832 
1833 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1834 	if (!fc)
1835 		return -ENOMEM;
1836 
1837 	fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1838 	if (!fm) {
1839 		kfree(fc);
1840 		return -ENOMEM;
1841 	}
1842 
1843 	fuse_conn_init(fc, fm, fsc->user_ns, &fuse_dev_fiq_ops, NULL);
1844 	fc->release = fuse_free_conn;
1845 
1846 	fsc->s_fs_info = fm;
1847 
1848 	if (ctx->fd_present)
1849 		ctx->file = fget(ctx->fd);
1850 
1851 	if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1852 		err = get_tree_bdev(fsc, fuse_fill_super);
1853 		goto out;
1854 	}
1855 	/*
1856 	 * While block dev mount can be initialized with a dummy device fd
1857 	 * (found by device name), normal fuse mounts can't
1858 	 */
1859 	err = -EINVAL;
1860 	if (!ctx->file)
1861 		goto out;
1862 
1863 	/*
1864 	 * Allow creating a fuse mount with an already initialized fuse
1865 	 * connection
1866 	 */
1867 	fud = READ_ONCE(ctx->file->private_data);
1868 	if (ctx->file->f_op == &fuse_dev_operations && fud) {
1869 		fsc->sget_key = fud->fc;
1870 		sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1871 		err = PTR_ERR_OR_ZERO(sb);
1872 		if (!IS_ERR(sb))
1873 			fsc->root = dget(sb->s_root);
1874 	} else {
1875 		err = get_tree_nodev(fsc, fuse_fill_super);
1876 	}
1877 out:
1878 	if (fsc->s_fs_info)
1879 		fuse_mount_destroy(fm);
1880 	if (ctx->file)
1881 		fput(ctx->file);
1882 	return err;
1883 }
1884 
1885 static const struct fs_context_operations fuse_context_ops = {
1886 	.free		= fuse_free_fsc,
1887 	.parse_param	= fuse_parse_param,
1888 	.reconfigure	= fuse_reconfigure,
1889 	.get_tree	= fuse_get_tree,
1890 };
1891 
1892 /*
1893  * Set up the filesystem mount context.
1894  */
1895 static int fuse_init_fs_context(struct fs_context *fsc)
1896 {
1897 	struct fuse_fs_context *ctx;
1898 
1899 	ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1900 	if (!ctx)
1901 		return -ENOMEM;
1902 
1903 	ctx->max_read = ~0;
1904 	ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1905 	ctx->legacy_opts_show = true;
1906 
1907 #ifdef CONFIG_BLOCK
1908 	if (fsc->fs_type == &fuseblk_fs_type) {
1909 		ctx->is_bdev = true;
1910 		ctx->destroy = true;
1911 	}
1912 #endif
1913 
1914 	fsc->fs_private = ctx;
1915 	fsc->ops = &fuse_context_ops;
1916 	return 0;
1917 }
1918 
1919 bool fuse_mount_remove(struct fuse_mount *fm)
1920 {
1921 	struct fuse_conn *fc = fm->fc;
1922 	bool last = false;
1923 
1924 	down_write(&fc->killsb);
1925 	list_del_init(&fm->fc_entry);
1926 	if (list_empty(&fc->mounts))
1927 		last = true;
1928 	up_write(&fc->killsb);
1929 
1930 	return last;
1931 }
1932 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1933 
1934 void fuse_conn_destroy(struct fuse_mount *fm)
1935 {
1936 	struct fuse_conn *fc = fm->fc;
1937 
1938 	if (fc->destroy)
1939 		fuse_send_destroy(fm);
1940 
1941 	fuse_abort_conn(fc);
1942 	fuse_wait_aborted(fc);
1943 
1944 	if (!list_empty(&fc->entry)) {
1945 		mutex_lock(&fuse_mutex);
1946 		list_del(&fc->entry);
1947 		fuse_ctl_remove_conn(fc);
1948 		mutex_unlock(&fuse_mutex);
1949 	}
1950 }
1951 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1952 
1953 static void fuse_sb_destroy(struct super_block *sb)
1954 {
1955 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1956 	bool last;
1957 
1958 	if (sb->s_root) {
1959 		last = fuse_mount_remove(fm);
1960 		if (last)
1961 			fuse_conn_destroy(fm);
1962 	}
1963 }
1964 
1965 void fuse_mount_destroy(struct fuse_mount *fm)
1966 {
1967 	fuse_conn_put(fm->fc);
1968 	kfree_rcu(fm, rcu);
1969 }
1970 EXPORT_SYMBOL(fuse_mount_destroy);
1971 
1972 static void fuse_kill_sb_anon(struct super_block *sb)
1973 {
1974 	fuse_sb_destroy(sb);
1975 	kill_anon_super(sb);
1976 	fuse_mount_destroy(get_fuse_mount_super(sb));
1977 }
1978 
1979 static struct file_system_type fuse_fs_type = {
1980 	.owner		= THIS_MODULE,
1981 	.name		= "fuse",
1982 	.fs_flags	= FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1983 	.init_fs_context = fuse_init_fs_context,
1984 	.parameters	= fuse_fs_parameters,
1985 	.kill_sb	= fuse_kill_sb_anon,
1986 };
1987 MODULE_ALIAS_FS("fuse");
1988 
1989 #ifdef CONFIG_BLOCK
1990 static void fuse_kill_sb_blk(struct super_block *sb)
1991 {
1992 	fuse_sb_destroy(sb);
1993 	kill_block_super(sb);
1994 	fuse_mount_destroy(get_fuse_mount_super(sb));
1995 }
1996 
1997 static struct file_system_type fuseblk_fs_type = {
1998 	.owner		= THIS_MODULE,
1999 	.name		= "fuseblk",
2000 	.init_fs_context = fuse_init_fs_context,
2001 	.parameters	= fuse_fs_parameters,
2002 	.kill_sb	= fuse_kill_sb_blk,
2003 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
2004 };
2005 MODULE_ALIAS_FS("fuseblk");
2006 
2007 static inline int register_fuseblk(void)
2008 {
2009 	return register_filesystem(&fuseblk_fs_type);
2010 }
2011 
2012 static inline void unregister_fuseblk(void)
2013 {
2014 	unregister_filesystem(&fuseblk_fs_type);
2015 }
2016 #else
2017 static inline int register_fuseblk(void)
2018 {
2019 	return 0;
2020 }
2021 
2022 static inline void unregister_fuseblk(void)
2023 {
2024 }
2025 #endif
2026 
2027 static void fuse_inode_init_once(void *foo)
2028 {
2029 	struct inode *inode = foo;
2030 
2031 	inode_init_once(inode);
2032 }
2033 
2034 static int __init fuse_fs_init(void)
2035 {
2036 	int err;
2037 
2038 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
2039 			sizeof(struct fuse_inode), 0,
2040 			SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
2041 			fuse_inode_init_once);
2042 	err = -ENOMEM;
2043 	if (!fuse_inode_cachep)
2044 		goto out;
2045 
2046 	err = register_fuseblk();
2047 	if (err)
2048 		goto out2;
2049 
2050 	err = register_filesystem(&fuse_fs_type);
2051 	if (err)
2052 		goto out3;
2053 
2054 	return 0;
2055 
2056  out3:
2057 	unregister_fuseblk();
2058  out2:
2059 	kmem_cache_destroy(fuse_inode_cachep);
2060  out:
2061 	return err;
2062 }
2063 
2064 static void fuse_fs_cleanup(void)
2065 {
2066 	unregister_filesystem(&fuse_fs_type);
2067 	unregister_fuseblk();
2068 
2069 	/*
2070 	 * Make sure all delayed rcu free inodes are flushed before we
2071 	 * destroy cache.
2072 	 */
2073 	rcu_barrier();
2074 	kmem_cache_destroy(fuse_inode_cachep);
2075 }
2076 
2077 static struct kobject *fuse_kobj;
2078 
2079 static int fuse_sysfs_init(void)
2080 {
2081 	int err;
2082 
2083 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
2084 	if (!fuse_kobj) {
2085 		err = -ENOMEM;
2086 		goto out_err;
2087 	}
2088 
2089 	err = sysfs_create_mount_point(fuse_kobj, "connections");
2090 	if (err)
2091 		goto out_fuse_unregister;
2092 
2093 	return 0;
2094 
2095  out_fuse_unregister:
2096 	kobject_put(fuse_kobj);
2097  out_err:
2098 	return err;
2099 }
2100 
2101 static void fuse_sysfs_cleanup(void)
2102 {
2103 	sysfs_remove_mount_point(fuse_kobj, "connections");
2104 	kobject_put(fuse_kobj);
2105 }
2106 
2107 static int __init fuse_init(void)
2108 {
2109 	int res;
2110 
2111 	pr_info("init (API version %i.%i)\n",
2112 		FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
2113 
2114 	INIT_LIST_HEAD(&fuse_conn_list);
2115 	res = fuse_fs_init();
2116 	if (res)
2117 		goto err;
2118 
2119 	res = fuse_dev_init();
2120 	if (res)
2121 		goto err_fs_cleanup;
2122 
2123 	res = fuse_sysfs_init();
2124 	if (res)
2125 		goto err_dev_cleanup;
2126 
2127 	res = fuse_ctl_init();
2128 	if (res)
2129 		goto err_sysfs_cleanup;
2130 
2131 	sanitize_global_limit(&max_user_bgreq);
2132 	sanitize_global_limit(&max_user_congthresh);
2133 
2134 	return 0;
2135 
2136  err_sysfs_cleanup:
2137 	fuse_sysfs_cleanup();
2138  err_dev_cleanup:
2139 	fuse_dev_cleanup();
2140  err_fs_cleanup:
2141 	fuse_fs_cleanup();
2142  err:
2143 	return res;
2144 }
2145 
2146 static void __exit fuse_exit(void)
2147 {
2148 	pr_debug("exit\n");
2149 
2150 	fuse_ctl_cleanup();
2151 	fuse_sysfs_cleanup();
2152 	fuse_fs_cleanup();
2153 	fuse_dev_cleanup();
2154 }
2155 
2156 module_init(fuse_init);
2157 module_exit(fuse_exit);
2158