xref: /linux/fs/fuse/inode.c (revision cdd30ebb1b9f36159d66f088b61aee264e649d7a)
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 int fuse_max_pages_limit = 256;
39 
40 unsigned max_user_bgreq;
41 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
42 		  &max_user_bgreq, 0644);
43 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
44 MODULE_PARM_DESC(max_user_bgreq,
45  "Global limit for the maximum number of backgrounded requests an "
46  "unprivileged user can set");
47 
48 unsigned max_user_congthresh;
49 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
50 		  &max_user_congthresh, 0644);
51 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
52 MODULE_PARM_DESC(max_user_congthresh,
53  "Global limit for the maximum congestion threshold an "
54  "unprivileged user can set");
55 
56 #define FUSE_DEFAULT_BLKSIZE 512
57 
58 /** Maximum number of outstanding background requests */
59 #define FUSE_DEFAULT_MAX_BACKGROUND 12
60 
61 /** Congestion starts at 75% of maximum */
62 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
63 
64 #ifdef CONFIG_BLOCK
65 static struct file_system_type fuseblk_fs_type;
66 #endif
67 
68 struct fuse_forget_link *fuse_alloc_forget(void)
69 {
70 	return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
71 }
72 
73 static struct fuse_submount_lookup *fuse_alloc_submount_lookup(void)
74 {
75 	struct fuse_submount_lookup *sl;
76 
77 	sl = kzalloc(sizeof(struct fuse_submount_lookup), GFP_KERNEL_ACCOUNT);
78 	if (!sl)
79 		return NULL;
80 	sl->forget = fuse_alloc_forget();
81 	if (!sl->forget)
82 		goto out_free;
83 
84 	return sl;
85 
86 out_free:
87 	kfree(sl);
88 	return NULL;
89 }
90 
91 static struct inode *fuse_alloc_inode(struct super_block *sb)
92 {
93 	struct fuse_inode *fi;
94 
95 	fi = alloc_inode_sb(sb, fuse_inode_cachep, GFP_KERNEL);
96 	if (!fi)
97 		return NULL;
98 
99 	fi->i_time = 0;
100 	fi->inval_mask = ~0;
101 	fi->nodeid = 0;
102 	fi->nlookup = 0;
103 	fi->attr_version = 0;
104 	fi->orig_ino = 0;
105 	fi->state = 0;
106 	fi->submount_lookup = NULL;
107 	mutex_init(&fi->mutex);
108 	spin_lock_init(&fi->lock);
109 	fi->forget = fuse_alloc_forget();
110 	if (!fi->forget)
111 		goto out_free;
112 
113 	if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
114 		goto out_free_forget;
115 
116 	if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
117 		fuse_inode_backing_set(fi, NULL);
118 
119 	return &fi->inode;
120 
121 out_free_forget:
122 	kfree(fi->forget);
123 out_free:
124 	kmem_cache_free(fuse_inode_cachep, fi);
125 	return NULL;
126 }
127 
128 static void fuse_free_inode(struct inode *inode)
129 {
130 	struct fuse_inode *fi = get_fuse_inode(inode);
131 
132 	mutex_destroy(&fi->mutex);
133 	kfree(fi->forget);
134 #ifdef CONFIG_FUSE_DAX
135 	kfree(fi->dax);
136 #endif
137 	if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
138 		fuse_backing_put(fuse_inode_backing(fi));
139 
140 	kmem_cache_free(fuse_inode_cachep, fi);
141 }
142 
143 static void fuse_cleanup_submount_lookup(struct fuse_conn *fc,
144 					 struct fuse_submount_lookup *sl)
145 {
146 	if (!refcount_dec_and_test(&sl->count))
147 		return;
148 
149 	fuse_queue_forget(fc, sl->forget, sl->nodeid, 1);
150 	sl->forget = NULL;
151 	kfree(sl);
152 }
153 
154 static void fuse_evict_inode(struct inode *inode)
155 {
156 	struct fuse_inode *fi = get_fuse_inode(inode);
157 
158 	/* Will write inode on close/munmap and in all other dirtiers */
159 	WARN_ON(inode->i_state & I_DIRTY_INODE);
160 
161 	truncate_inode_pages_final(&inode->i_data);
162 	clear_inode(inode);
163 	if (inode->i_sb->s_flags & SB_ACTIVE) {
164 		struct fuse_conn *fc = get_fuse_conn(inode);
165 
166 		if (FUSE_IS_DAX(inode))
167 			fuse_dax_inode_cleanup(inode);
168 		if (fi->nlookup) {
169 			fuse_queue_forget(fc, fi->forget, fi->nodeid,
170 					  fi->nlookup);
171 			fi->forget = NULL;
172 		}
173 
174 		if (fi->submount_lookup) {
175 			fuse_cleanup_submount_lookup(fc, fi->submount_lookup);
176 			fi->submount_lookup = NULL;
177 		}
178 		/*
179 		 * Evict of non-deleted inode may race with outstanding
180 		 * LOOKUP/READDIRPLUS requests and result in inconsistency when
181 		 * the request finishes.  Deal with that here by bumping a
182 		 * counter that can be compared to the starting value.
183 		 */
184 		if (inode->i_nlink > 0)
185 			atomic64_inc(&fc->evict_ctr);
186 	}
187 	if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
188 		WARN_ON(fi->iocachectr != 0);
189 		WARN_ON(!list_empty(&fi->write_files));
190 		WARN_ON(!list_empty(&fi->queued_writes));
191 	}
192 }
193 
194 static int fuse_reconfigure(struct fs_context *fsc)
195 {
196 	struct super_block *sb = fsc->root->d_sb;
197 
198 	sync_filesystem(sb);
199 	if (fsc->sb_flags & SB_MANDLOCK)
200 		return -EINVAL;
201 
202 	return 0;
203 }
204 
205 /*
206  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
207  * so that it will fit.
208  */
209 static ino_t fuse_squash_ino(u64 ino64)
210 {
211 	ino_t ino = (ino_t) ino64;
212 	if (sizeof(ino_t) < sizeof(u64))
213 		ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
214 	return ino;
215 }
216 
217 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
218 				   struct fuse_statx *sx,
219 				   u64 attr_valid, u32 cache_mask,
220 				   u64 evict_ctr)
221 {
222 	struct fuse_conn *fc = get_fuse_conn(inode);
223 	struct fuse_inode *fi = get_fuse_inode(inode);
224 
225 	lockdep_assert_held(&fi->lock);
226 
227 	/*
228 	 * Clear basic stats from invalid mask.
229 	 *
230 	 * Don't do this if this is coming from a fuse_iget() call and there
231 	 * might have been a racing evict which would've invalidated the result
232 	 * if the attr_version would've been preserved.
233 	 *
234 	 * !evict_ctr -> this is create
235 	 * fi->attr_version != 0 -> this is not a new inode
236 	 * evict_ctr == fuse_get_evict_ctr() -> no evicts while during request
237 	 */
238 	if (!evict_ctr || fi->attr_version || evict_ctr == fuse_get_evict_ctr(fc))
239 		set_mask_bits(&fi->inval_mask, STATX_BASIC_STATS, 0);
240 
241 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
242 	fi->i_time = attr_valid;
243 
244 	inode->i_ino     = fuse_squash_ino(attr->ino);
245 	inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
246 	set_nlink(inode, attr->nlink);
247 	inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
248 	inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
249 	inode->i_blocks  = attr->blocks;
250 
251 	/* Sanitize nsecs */
252 	attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
253 	attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
254 	attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
255 
256 	inode_set_atime(inode, attr->atime, attr->atimensec);
257 	/* mtime from server may be stale due to local buffered write */
258 	if (!(cache_mask & STATX_MTIME)) {
259 		inode_set_mtime(inode, attr->mtime, attr->mtimensec);
260 	}
261 	if (!(cache_mask & STATX_CTIME)) {
262 		inode_set_ctime(inode, attr->ctime, attr->ctimensec);
263 	}
264 	if (sx) {
265 		/* Sanitize nsecs */
266 		sx->btime.tv_nsec =
267 			min_t(u32, sx->btime.tv_nsec, NSEC_PER_SEC - 1);
268 
269 		/*
270 		 * Btime has been queried, cache is valid (whether or not btime
271 		 * is available or not) so clear STATX_BTIME from inval_mask.
272 		 *
273 		 * Availability of the btime attribute is indicated in
274 		 * FUSE_I_BTIME
275 		 */
276 		set_mask_bits(&fi->inval_mask, STATX_BTIME, 0);
277 		if (sx->mask & STATX_BTIME) {
278 			set_bit(FUSE_I_BTIME, &fi->state);
279 			fi->i_btime.tv_sec = sx->btime.tv_sec;
280 			fi->i_btime.tv_nsec = sx->btime.tv_nsec;
281 		}
282 	}
283 
284 	if (attr->blksize != 0)
285 		inode->i_blkbits = ilog2(attr->blksize);
286 	else
287 		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
288 
289 	/*
290 	 * Don't set the sticky bit in i_mode, unless we want the VFS
291 	 * to check permissions.  This prevents failures due to the
292 	 * check in may_delete().
293 	 */
294 	fi->orig_i_mode = inode->i_mode;
295 	if (!fc->default_permissions)
296 		inode->i_mode &= ~S_ISVTX;
297 
298 	fi->orig_ino = attr->ino;
299 
300 	/*
301 	 * We are refreshing inode data and it is possible that another
302 	 * client set suid/sgid or security.capability xattr. So clear
303 	 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
304 	 * was set or if security.capability xattr was set. But we don't
305 	 * know if security.capability has been set or not. So clear it
306 	 * anyway. Its less efficient but should be safe.
307 	 */
308 	inode->i_flags &= ~S_NOSEC;
309 }
310 
311 u32 fuse_get_cache_mask(struct inode *inode)
312 {
313 	struct fuse_conn *fc = get_fuse_conn(inode);
314 
315 	if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
316 		return 0;
317 
318 	return STATX_MTIME | STATX_CTIME | STATX_SIZE;
319 }
320 
321 static void fuse_change_attributes_i(struct inode *inode, struct fuse_attr *attr,
322 				     struct fuse_statx *sx, u64 attr_valid,
323 				     u64 attr_version, u64 evict_ctr)
324 {
325 	struct fuse_conn *fc = get_fuse_conn(inode);
326 	struct fuse_inode *fi = get_fuse_inode(inode);
327 	u32 cache_mask;
328 	loff_t oldsize;
329 	struct timespec64 old_mtime;
330 
331 	spin_lock(&fi->lock);
332 	/*
333 	 * In case of writeback_cache enabled, writes update mtime, ctime and
334 	 * may update i_size.  In these cases trust the cached value in the
335 	 * inode.
336 	 */
337 	cache_mask = fuse_get_cache_mask(inode);
338 	if (cache_mask & STATX_SIZE)
339 		attr->size = i_size_read(inode);
340 
341 	if (cache_mask & STATX_MTIME) {
342 		attr->mtime = inode_get_mtime_sec(inode);
343 		attr->mtimensec = inode_get_mtime_nsec(inode);
344 	}
345 	if (cache_mask & STATX_CTIME) {
346 		attr->ctime = inode_get_ctime_sec(inode);
347 		attr->ctimensec = inode_get_ctime_nsec(inode);
348 	}
349 
350 	if ((attr_version != 0 && fi->attr_version > attr_version) ||
351 	    test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
352 		spin_unlock(&fi->lock);
353 		return;
354 	}
355 
356 	old_mtime = inode_get_mtime(inode);
357 	fuse_change_attributes_common(inode, attr, sx, attr_valid, cache_mask,
358 				      evict_ctr);
359 
360 	oldsize = inode->i_size;
361 	/*
362 	 * In case of writeback_cache enabled, the cached writes beyond EOF
363 	 * extend local i_size without keeping userspace server in sync. So,
364 	 * attr->size coming from server can be stale. We cannot trust it.
365 	 */
366 	if (!(cache_mask & STATX_SIZE))
367 		i_size_write(inode, attr->size);
368 	spin_unlock(&fi->lock);
369 
370 	if (!cache_mask && S_ISREG(inode->i_mode)) {
371 		bool inval = false;
372 
373 		if (oldsize != attr->size) {
374 			truncate_pagecache(inode, attr->size);
375 			if (!fc->explicit_inval_data)
376 				inval = true;
377 		} else if (fc->auto_inval_data) {
378 			struct timespec64 new_mtime = {
379 				.tv_sec = attr->mtime,
380 				.tv_nsec = attr->mtimensec,
381 			};
382 
383 			/*
384 			 * Auto inval mode also checks and invalidates if mtime
385 			 * has changed.
386 			 */
387 			if (!timespec64_equal(&old_mtime, &new_mtime))
388 				inval = true;
389 		}
390 
391 		if (inval)
392 			invalidate_inode_pages2(inode->i_mapping);
393 	}
394 
395 	if (IS_ENABLED(CONFIG_FUSE_DAX))
396 		fuse_dax_dontcache(inode, attr->flags);
397 }
398 
399 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
400 			    struct fuse_statx *sx, u64 attr_valid,
401 			    u64 attr_version)
402 {
403 	fuse_change_attributes_i(inode, attr, sx, attr_valid, attr_version, 0);
404 }
405 
406 static void fuse_init_submount_lookup(struct fuse_submount_lookup *sl,
407 				      u64 nodeid)
408 {
409 	sl->nodeid = nodeid;
410 	refcount_set(&sl->count, 1);
411 }
412 
413 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr,
414 			    struct fuse_conn *fc)
415 {
416 	inode->i_mode = attr->mode & S_IFMT;
417 	inode->i_size = attr->size;
418 	inode_set_mtime(inode, attr->mtime, attr->mtimensec);
419 	inode_set_ctime(inode, attr->ctime, attr->ctimensec);
420 	if (S_ISREG(inode->i_mode)) {
421 		fuse_init_common(inode);
422 		fuse_init_file_inode(inode, attr->flags);
423 	} else if (S_ISDIR(inode->i_mode))
424 		fuse_init_dir(inode);
425 	else if (S_ISLNK(inode->i_mode))
426 		fuse_init_symlink(inode);
427 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
428 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
429 		fuse_init_common(inode);
430 		init_special_inode(inode, inode->i_mode,
431 				   new_decode_dev(attr->rdev));
432 	} else
433 		BUG();
434 	/*
435 	 * Ensure that we don't cache acls for daemons without FUSE_POSIX_ACL
436 	 * so they see the exact same behavior as before.
437 	 */
438 	if (!fc->posix_acl)
439 		inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
440 }
441 
442 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
443 {
444 	u64 nodeid = *(u64 *) _nodeidp;
445 	if (get_node_id(inode) == nodeid)
446 		return 1;
447 	else
448 		return 0;
449 }
450 
451 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
452 {
453 	u64 nodeid = *(u64 *) _nodeidp;
454 	get_fuse_inode(inode)->nodeid = nodeid;
455 	return 0;
456 }
457 
458 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
459 			int generation, struct fuse_attr *attr,
460 			u64 attr_valid, u64 attr_version,
461 			u64 evict_ctr)
462 {
463 	struct inode *inode;
464 	struct fuse_inode *fi;
465 	struct fuse_conn *fc = get_fuse_conn_super(sb);
466 
467 	/*
468 	 * Auto mount points get their node id from the submount root, which is
469 	 * not a unique identifier within this filesystem.
470 	 *
471 	 * To avoid conflicts, do not place submount points into the inode hash
472 	 * table.
473 	 */
474 	if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
475 	    S_ISDIR(attr->mode)) {
476 		struct fuse_inode *fi;
477 
478 		inode = new_inode(sb);
479 		if (!inode)
480 			return NULL;
481 
482 		fuse_init_inode(inode, attr, fc);
483 		fi = get_fuse_inode(inode);
484 		fi->nodeid = nodeid;
485 		fi->submount_lookup = fuse_alloc_submount_lookup();
486 		if (!fi->submount_lookup) {
487 			iput(inode);
488 			return NULL;
489 		}
490 		/* Sets nlookup = 1 on fi->submount_lookup->nlookup */
491 		fuse_init_submount_lookup(fi->submount_lookup, nodeid);
492 		inode->i_flags |= S_AUTOMOUNT;
493 		goto done;
494 	}
495 
496 retry:
497 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
498 	if (!inode)
499 		return NULL;
500 
501 	if ((inode->i_state & I_NEW)) {
502 		inode->i_flags |= S_NOATIME;
503 		if (!fc->writeback_cache || !S_ISREG(attr->mode))
504 			inode->i_flags |= S_NOCMTIME;
505 		inode->i_generation = generation;
506 		fuse_init_inode(inode, attr, fc);
507 		unlock_new_inode(inode);
508 	} else if (fuse_stale_inode(inode, generation, attr)) {
509 		/* nodeid was reused, any I/O on the old inode should fail */
510 		fuse_make_bad(inode);
511 		if (inode != d_inode(sb->s_root)) {
512 			remove_inode_hash(inode);
513 			iput(inode);
514 			goto retry;
515 		}
516 	}
517 	fi = get_fuse_inode(inode);
518 	spin_lock(&fi->lock);
519 	fi->nlookup++;
520 	spin_unlock(&fi->lock);
521 done:
522 	fuse_change_attributes_i(inode, attr, NULL, attr_valid, attr_version,
523 				 evict_ctr);
524 	return inode;
525 }
526 
527 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
528 			   struct fuse_mount **fm)
529 {
530 	struct fuse_mount *fm_iter;
531 	struct inode *inode;
532 
533 	WARN_ON(!rwsem_is_locked(&fc->killsb));
534 	list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
535 		if (!fm_iter->sb)
536 			continue;
537 
538 		inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
539 		if (inode) {
540 			if (fm)
541 				*fm = fm_iter;
542 			return inode;
543 		}
544 	}
545 
546 	return NULL;
547 }
548 
549 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
550 			     loff_t offset, loff_t len)
551 {
552 	struct fuse_inode *fi;
553 	struct inode *inode;
554 	pgoff_t pg_start;
555 	pgoff_t pg_end;
556 
557 	inode = fuse_ilookup(fc, nodeid, NULL);
558 	if (!inode)
559 		return -ENOENT;
560 
561 	fi = get_fuse_inode(inode);
562 	spin_lock(&fi->lock);
563 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
564 	spin_unlock(&fi->lock);
565 
566 	fuse_invalidate_attr(inode);
567 	forget_all_cached_acls(inode);
568 	if (offset >= 0) {
569 		pg_start = offset >> PAGE_SHIFT;
570 		if (len <= 0)
571 			pg_end = -1;
572 		else
573 			pg_end = (offset + len - 1) >> PAGE_SHIFT;
574 		invalidate_inode_pages2_range(inode->i_mapping,
575 					      pg_start, pg_end);
576 	}
577 	iput(inode);
578 	return 0;
579 }
580 
581 bool fuse_lock_inode(struct inode *inode)
582 {
583 	bool locked = false;
584 
585 	if (!get_fuse_conn(inode)->parallel_dirops) {
586 		mutex_lock(&get_fuse_inode(inode)->mutex);
587 		locked = true;
588 	}
589 
590 	return locked;
591 }
592 
593 void fuse_unlock_inode(struct inode *inode, bool locked)
594 {
595 	if (locked)
596 		mutex_unlock(&get_fuse_inode(inode)->mutex);
597 }
598 
599 static void fuse_umount_begin(struct super_block *sb)
600 {
601 	struct fuse_conn *fc = get_fuse_conn_super(sb);
602 
603 	if (fc->no_force_umount)
604 		return;
605 
606 	fuse_abort_conn(fc);
607 
608 	// Only retire block-device-based superblocks.
609 	if (sb->s_bdev != NULL)
610 		retire_super(sb);
611 }
612 
613 static void fuse_send_destroy(struct fuse_mount *fm)
614 {
615 	if (fm->fc->conn_init) {
616 		FUSE_ARGS(args);
617 
618 		args.opcode = FUSE_DESTROY;
619 		args.force = true;
620 		args.nocreds = true;
621 		fuse_simple_request(fm, &args);
622 	}
623 }
624 
625 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
626 {
627 	stbuf->f_type    = FUSE_SUPER_MAGIC;
628 	stbuf->f_bsize   = attr->bsize;
629 	stbuf->f_frsize  = attr->frsize;
630 	stbuf->f_blocks  = attr->blocks;
631 	stbuf->f_bfree   = attr->bfree;
632 	stbuf->f_bavail  = attr->bavail;
633 	stbuf->f_files   = attr->files;
634 	stbuf->f_ffree   = attr->ffree;
635 	stbuf->f_namelen = attr->namelen;
636 	/* fsid is left zero */
637 }
638 
639 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
640 {
641 	struct super_block *sb = dentry->d_sb;
642 	struct fuse_mount *fm = get_fuse_mount_super(sb);
643 	FUSE_ARGS(args);
644 	struct fuse_statfs_out outarg;
645 	int err;
646 
647 	if (!fuse_allow_current_process(fm->fc)) {
648 		buf->f_type = FUSE_SUPER_MAGIC;
649 		return 0;
650 	}
651 
652 	memset(&outarg, 0, sizeof(outarg));
653 	args.in_numargs = 0;
654 	args.opcode = FUSE_STATFS;
655 	args.nodeid = get_node_id(d_inode(dentry));
656 	args.out_numargs = 1;
657 	args.out_args[0].size = sizeof(outarg);
658 	args.out_args[0].value = &outarg;
659 	err = fuse_simple_request(fm, &args);
660 	if (!err)
661 		convert_fuse_statfs(buf, &outarg.st);
662 	return err;
663 }
664 
665 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
666 {
667 	struct fuse_sync_bucket *bucket;
668 
669 	bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
670 	if (bucket) {
671 		init_waitqueue_head(&bucket->waitq);
672 		/* Initial active count */
673 		atomic_set(&bucket->count, 1);
674 	}
675 	return bucket;
676 }
677 
678 static void fuse_sync_fs_writes(struct fuse_conn *fc)
679 {
680 	struct fuse_sync_bucket *bucket, *new_bucket;
681 	int count;
682 
683 	new_bucket = fuse_sync_bucket_alloc();
684 	spin_lock(&fc->lock);
685 	bucket = rcu_dereference_protected(fc->curr_bucket, 1);
686 	count = atomic_read(&bucket->count);
687 	WARN_ON(count < 1);
688 	/* No outstanding writes? */
689 	if (count == 1) {
690 		spin_unlock(&fc->lock);
691 		kfree(new_bucket);
692 		return;
693 	}
694 
695 	/*
696 	 * Completion of new bucket depends on completion of this bucket, so add
697 	 * one more count.
698 	 */
699 	atomic_inc(&new_bucket->count);
700 	rcu_assign_pointer(fc->curr_bucket, new_bucket);
701 	spin_unlock(&fc->lock);
702 	/*
703 	 * Drop initial active count.  At this point if all writes in this and
704 	 * ancestor buckets complete, the count will go to zero and this task
705 	 * will be woken up.
706 	 */
707 	atomic_dec(&bucket->count);
708 
709 	wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
710 
711 	/* Drop temp count on descendant bucket */
712 	fuse_sync_bucket_dec(new_bucket);
713 	kfree_rcu(bucket, rcu);
714 }
715 
716 static int fuse_sync_fs(struct super_block *sb, int wait)
717 {
718 	struct fuse_mount *fm = get_fuse_mount_super(sb);
719 	struct fuse_conn *fc = fm->fc;
720 	struct fuse_syncfs_in inarg;
721 	FUSE_ARGS(args);
722 	int err;
723 
724 	/*
725 	 * Userspace cannot handle the wait == 0 case.  Avoid a
726 	 * gratuitous roundtrip.
727 	 */
728 	if (!wait)
729 		return 0;
730 
731 	/* The filesystem is being unmounted.  Nothing to do. */
732 	if (!sb->s_root)
733 		return 0;
734 
735 	if (!fc->sync_fs)
736 		return 0;
737 
738 	fuse_sync_fs_writes(fc);
739 
740 	memset(&inarg, 0, sizeof(inarg));
741 	args.in_numargs = 1;
742 	args.in_args[0].size = sizeof(inarg);
743 	args.in_args[0].value = &inarg;
744 	args.opcode = FUSE_SYNCFS;
745 	args.nodeid = get_node_id(sb->s_root->d_inode);
746 	args.out_numargs = 0;
747 
748 	err = fuse_simple_request(fm, &args);
749 	if (err == -ENOSYS) {
750 		fc->sync_fs = 0;
751 		err = 0;
752 	}
753 
754 	return err;
755 }
756 
757 enum {
758 	OPT_SOURCE,
759 	OPT_SUBTYPE,
760 	OPT_FD,
761 	OPT_ROOTMODE,
762 	OPT_USER_ID,
763 	OPT_GROUP_ID,
764 	OPT_DEFAULT_PERMISSIONS,
765 	OPT_ALLOW_OTHER,
766 	OPT_MAX_READ,
767 	OPT_BLKSIZE,
768 	OPT_ERR
769 };
770 
771 static const struct fs_parameter_spec fuse_fs_parameters[] = {
772 	fsparam_string	("source",		OPT_SOURCE),
773 	fsparam_u32	("fd",			OPT_FD),
774 	fsparam_u32oct	("rootmode",		OPT_ROOTMODE),
775 	fsparam_uid	("user_id",		OPT_USER_ID),
776 	fsparam_gid	("group_id",		OPT_GROUP_ID),
777 	fsparam_flag	("default_permissions",	OPT_DEFAULT_PERMISSIONS),
778 	fsparam_flag	("allow_other",		OPT_ALLOW_OTHER),
779 	fsparam_u32	("max_read",		OPT_MAX_READ),
780 	fsparam_u32	("blksize",		OPT_BLKSIZE),
781 	fsparam_string	("subtype",		OPT_SUBTYPE),
782 	{}
783 };
784 
785 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
786 {
787 	struct fs_parse_result result;
788 	struct fuse_fs_context *ctx = fsc->fs_private;
789 	int opt;
790 	kuid_t kuid;
791 	kgid_t kgid;
792 
793 	if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
794 		/*
795 		 * Ignore options coming from mount(MS_REMOUNT) for backward
796 		 * compatibility.
797 		 */
798 		if (fsc->oldapi)
799 			return 0;
800 
801 		return invalfc(fsc, "No changes allowed in reconfigure");
802 	}
803 
804 	opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
805 	if (opt < 0)
806 		return opt;
807 
808 	switch (opt) {
809 	case OPT_SOURCE:
810 		if (fsc->source)
811 			return invalfc(fsc, "Multiple sources specified");
812 		fsc->source = param->string;
813 		param->string = NULL;
814 		break;
815 
816 	case OPT_SUBTYPE:
817 		if (ctx->subtype)
818 			return invalfc(fsc, "Multiple subtypes specified");
819 		ctx->subtype = param->string;
820 		param->string = NULL;
821 		return 0;
822 
823 	case OPT_FD:
824 		ctx->fd = result.uint_32;
825 		ctx->fd_present = true;
826 		break;
827 
828 	case OPT_ROOTMODE:
829 		if (!fuse_valid_type(result.uint_32))
830 			return invalfc(fsc, "Invalid rootmode");
831 		ctx->rootmode = result.uint_32;
832 		ctx->rootmode_present = true;
833 		break;
834 
835 	case OPT_USER_ID:
836 		kuid = result.uid;
837 		/*
838 		 * The requested uid must be representable in the
839 		 * filesystem's idmapping.
840 		 */
841 		if (!kuid_has_mapping(fsc->user_ns, kuid))
842 			return invalfc(fsc, "Invalid user_id");
843 		ctx->user_id = kuid;
844 		ctx->user_id_present = true;
845 		break;
846 
847 	case OPT_GROUP_ID:
848 		kgid = result.gid;
849 		/*
850 		 * The requested gid must be representable in the
851 		 * filesystem's idmapping.
852 		 */
853 		if (!kgid_has_mapping(fsc->user_ns, kgid))
854 			return invalfc(fsc, "Invalid group_id");
855 		ctx->group_id = kgid;
856 		ctx->group_id_present = true;
857 		break;
858 
859 	case OPT_DEFAULT_PERMISSIONS:
860 		ctx->default_permissions = true;
861 		break;
862 
863 	case OPT_ALLOW_OTHER:
864 		ctx->allow_other = true;
865 		break;
866 
867 	case OPT_MAX_READ:
868 		ctx->max_read = result.uint_32;
869 		break;
870 
871 	case OPT_BLKSIZE:
872 		if (!ctx->is_bdev)
873 			return invalfc(fsc, "blksize only supported for fuseblk");
874 		ctx->blksize = result.uint_32;
875 		break;
876 
877 	default:
878 		return -EINVAL;
879 	}
880 
881 	return 0;
882 }
883 
884 static void fuse_free_fsc(struct fs_context *fsc)
885 {
886 	struct fuse_fs_context *ctx = fsc->fs_private;
887 
888 	if (ctx) {
889 		kfree(ctx->subtype);
890 		kfree(ctx);
891 	}
892 }
893 
894 static int fuse_show_options(struct seq_file *m, struct dentry *root)
895 {
896 	struct super_block *sb = root->d_sb;
897 	struct fuse_conn *fc = get_fuse_conn_super(sb);
898 
899 	if (fc->legacy_opts_show) {
900 		seq_printf(m, ",user_id=%u",
901 			   from_kuid_munged(fc->user_ns, fc->user_id));
902 		seq_printf(m, ",group_id=%u",
903 			   from_kgid_munged(fc->user_ns, fc->group_id));
904 		if (fc->default_permissions)
905 			seq_puts(m, ",default_permissions");
906 		if (fc->allow_other)
907 			seq_puts(m, ",allow_other");
908 		if (fc->max_read != ~0)
909 			seq_printf(m, ",max_read=%u", fc->max_read);
910 		if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
911 			seq_printf(m, ",blksize=%lu", sb->s_blocksize);
912 	}
913 #ifdef CONFIG_FUSE_DAX
914 	if (fc->dax_mode == FUSE_DAX_ALWAYS)
915 		seq_puts(m, ",dax=always");
916 	else if (fc->dax_mode == FUSE_DAX_NEVER)
917 		seq_puts(m, ",dax=never");
918 	else if (fc->dax_mode == FUSE_DAX_INODE_USER)
919 		seq_puts(m, ",dax=inode");
920 #endif
921 
922 	return 0;
923 }
924 
925 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
926 			     const struct fuse_iqueue_ops *ops,
927 			     void *priv)
928 {
929 	memset(fiq, 0, sizeof(struct fuse_iqueue));
930 	spin_lock_init(&fiq->lock);
931 	init_waitqueue_head(&fiq->waitq);
932 	INIT_LIST_HEAD(&fiq->pending);
933 	INIT_LIST_HEAD(&fiq->interrupts);
934 	fiq->forget_list_tail = &fiq->forget_list_head;
935 	fiq->connected = 1;
936 	fiq->ops = ops;
937 	fiq->priv = priv;
938 }
939 
940 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
941 {
942 	unsigned int i;
943 
944 	spin_lock_init(&fpq->lock);
945 	for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
946 		INIT_LIST_HEAD(&fpq->processing[i]);
947 	INIT_LIST_HEAD(&fpq->io);
948 	fpq->connected = 1;
949 }
950 
951 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
952 		    struct user_namespace *user_ns,
953 		    const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
954 {
955 	memset(fc, 0, sizeof(*fc));
956 	spin_lock_init(&fc->lock);
957 	spin_lock_init(&fc->bg_lock);
958 	init_rwsem(&fc->killsb);
959 	refcount_set(&fc->count, 1);
960 	atomic_set(&fc->dev_count, 1);
961 	init_waitqueue_head(&fc->blocked_waitq);
962 	fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
963 	INIT_LIST_HEAD(&fc->bg_queue);
964 	INIT_LIST_HEAD(&fc->entry);
965 	INIT_LIST_HEAD(&fc->devices);
966 	atomic_set(&fc->num_waiting, 0);
967 	fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
968 	fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
969 	atomic64_set(&fc->khctr, 0);
970 	fc->polled_files = RB_ROOT;
971 	fc->blocked = 0;
972 	fc->initialized = 0;
973 	fc->connected = 1;
974 	atomic64_set(&fc->attr_version, 1);
975 	atomic64_set(&fc->evict_ctr, 1);
976 	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
977 	fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
978 	fc->user_ns = get_user_ns(user_ns);
979 	fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
980 	fc->max_pages_limit = fuse_max_pages_limit;
981 
982 	if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
983 		fuse_backing_files_init(fc);
984 
985 	INIT_LIST_HEAD(&fc->mounts);
986 	list_add(&fm->fc_entry, &fc->mounts);
987 	fm->fc = fc;
988 }
989 EXPORT_SYMBOL_GPL(fuse_conn_init);
990 
991 static void delayed_release(struct rcu_head *p)
992 {
993 	struct fuse_conn *fc = container_of(p, struct fuse_conn, rcu);
994 
995 	put_user_ns(fc->user_ns);
996 	fc->release(fc);
997 }
998 
999 void fuse_conn_put(struct fuse_conn *fc)
1000 {
1001 	if (refcount_dec_and_test(&fc->count)) {
1002 		struct fuse_iqueue *fiq = &fc->iq;
1003 		struct fuse_sync_bucket *bucket;
1004 
1005 		if (IS_ENABLED(CONFIG_FUSE_DAX))
1006 			fuse_dax_conn_free(fc);
1007 		if (fiq->ops->release)
1008 			fiq->ops->release(fiq);
1009 		put_pid_ns(fc->pid_ns);
1010 		bucket = rcu_dereference_protected(fc->curr_bucket, 1);
1011 		if (bucket) {
1012 			WARN_ON(atomic_read(&bucket->count) != 1);
1013 			kfree(bucket);
1014 		}
1015 		if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
1016 			fuse_backing_files_free(fc);
1017 		call_rcu(&fc->rcu, delayed_release);
1018 	}
1019 }
1020 EXPORT_SYMBOL_GPL(fuse_conn_put);
1021 
1022 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
1023 {
1024 	refcount_inc(&fc->count);
1025 	return fc;
1026 }
1027 EXPORT_SYMBOL_GPL(fuse_conn_get);
1028 
1029 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
1030 {
1031 	struct fuse_attr attr;
1032 	memset(&attr, 0, sizeof(attr));
1033 
1034 	attr.mode = mode;
1035 	attr.ino = FUSE_ROOT_ID;
1036 	attr.nlink = 1;
1037 	return fuse_iget(sb, FUSE_ROOT_ID, 0, &attr, 0, 0, 0);
1038 }
1039 
1040 struct fuse_inode_handle {
1041 	u64 nodeid;
1042 	u32 generation;
1043 };
1044 
1045 static struct dentry *fuse_get_dentry(struct super_block *sb,
1046 				      struct fuse_inode_handle *handle)
1047 {
1048 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1049 	struct inode *inode;
1050 	struct dentry *entry;
1051 	int err = -ESTALE;
1052 
1053 	if (handle->nodeid == 0)
1054 		goto out_err;
1055 
1056 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
1057 	if (!inode) {
1058 		struct fuse_entry_out outarg;
1059 		const struct qstr name = QSTR_INIT(".", 1);
1060 
1061 		if (!fc->export_support)
1062 			goto out_err;
1063 
1064 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
1065 				       &inode);
1066 		if (err && err != -ENOENT)
1067 			goto out_err;
1068 		if (err || !inode) {
1069 			err = -ESTALE;
1070 			goto out_err;
1071 		}
1072 		err = -EIO;
1073 		if (get_node_id(inode) != handle->nodeid)
1074 			goto out_iput;
1075 	}
1076 	err = -ESTALE;
1077 	if (inode->i_generation != handle->generation)
1078 		goto out_iput;
1079 
1080 	entry = d_obtain_alias(inode);
1081 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
1082 		fuse_invalidate_entry_cache(entry);
1083 
1084 	return entry;
1085 
1086  out_iput:
1087 	iput(inode);
1088  out_err:
1089 	return ERR_PTR(err);
1090 }
1091 
1092 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
1093 			   struct inode *parent)
1094 {
1095 	int len = parent ? 6 : 3;
1096 	u64 nodeid;
1097 	u32 generation;
1098 
1099 	if (*max_len < len) {
1100 		*max_len = len;
1101 		return  FILEID_INVALID;
1102 	}
1103 
1104 	nodeid = get_fuse_inode(inode)->nodeid;
1105 	generation = inode->i_generation;
1106 
1107 	fh[0] = (u32)(nodeid >> 32);
1108 	fh[1] = (u32)(nodeid & 0xffffffff);
1109 	fh[2] = generation;
1110 
1111 	if (parent) {
1112 		nodeid = get_fuse_inode(parent)->nodeid;
1113 		generation = parent->i_generation;
1114 
1115 		fh[3] = (u32)(nodeid >> 32);
1116 		fh[4] = (u32)(nodeid & 0xffffffff);
1117 		fh[5] = generation;
1118 	}
1119 
1120 	*max_len = len;
1121 	return parent ? FILEID_INO64_GEN_PARENT : FILEID_INO64_GEN;
1122 }
1123 
1124 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
1125 		struct fid *fid, int fh_len, int fh_type)
1126 {
1127 	struct fuse_inode_handle handle;
1128 
1129 	if ((fh_type != FILEID_INO64_GEN &&
1130 	     fh_type != FILEID_INO64_GEN_PARENT) || fh_len < 3)
1131 		return NULL;
1132 
1133 	handle.nodeid = (u64) fid->raw[0] << 32;
1134 	handle.nodeid |= (u64) fid->raw[1];
1135 	handle.generation = fid->raw[2];
1136 	return fuse_get_dentry(sb, &handle);
1137 }
1138 
1139 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
1140 		struct fid *fid, int fh_len, int fh_type)
1141 {
1142 	struct fuse_inode_handle parent;
1143 
1144 	if (fh_type != FILEID_INO64_GEN_PARENT || fh_len < 6)
1145 		return NULL;
1146 
1147 	parent.nodeid = (u64) fid->raw[3] << 32;
1148 	parent.nodeid |= (u64) fid->raw[4];
1149 	parent.generation = fid->raw[5];
1150 	return fuse_get_dentry(sb, &parent);
1151 }
1152 
1153 static struct dentry *fuse_get_parent(struct dentry *child)
1154 {
1155 	struct inode *child_inode = d_inode(child);
1156 	struct fuse_conn *fc = get_fuse_conn(child_inode);
1157 	struct inode *inode;
1158 	struct dentry *parent;
1159 	struct fuse_entry_out outarg;
1160 	int err;
1161 
1162 	if (!fc->export_support)
1163 		return ERR_PTR(-ESTALE);
1164 
1165 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
1166 			       &dotdot_name, &outarg, &inode);
1167 	if (err) {
1168 		if (err == -ENOENT)
1169 			return ERR_PTR(-ESTALE);
1170 		return ERR_PTR(err);
1171 	}
1172 
1173 	parent = d_obtain_alias(inode);
1174 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
1175 		fuse_invalidate_entry_cache(parent);
1176 
1177 	return parent;
1178 }
1179 
1180 /* only for fid encoding; no support for file handle */
1181 static const struct export_operations fuse_export_fid_operations = {
1182 	.encode_fh	= fuse_encode_fh,
1183 };
1184 
1185 static const struct export_operations fuse_export_operations = {
1186 	.fh_to_dentry	= fuse_fh_to_dentry,
1187 	.fh_to_parent	= fuse_fh_to_parent,
1188 	.encode_fh	= fuse_encode_fh,
1189 	.get_parent	= fuse_get_parent,
1190 };
1191 
1192 static const struct super_operations fuse_super_operations = {
1193 	.alloc_inode    = fuse_alloc_inode,
1194 	.free_inode     = fuse_free_inode,
1195 	.evict_inode	= fuse_evict_inode,
1196 	.write_inode	= fuse_write_inode,
1197 	.drop_inode	= generic_delete_inode,
1198 	.umount_begin	= fuse_umount_begin,
1199 	.statfs		= fuse_statfs,
1200 	.sync_fs	= fuse_sync_fs,
1201 	.show_options	= fuse_show_options,
1202 };
1203 
1204 static void sanitize_global_limit(unsigned *limit)
1205 {
1206 	/*
1207 	 * The default maximum number of async requests is calculated to consume
1208 	 * 1/2^13 of the total memory, assuming 392 bytes per request.
1209 	 */
1210 	if (*limit == 0)
1211 		*limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1212 
1213 	if (*limit >= 1 << 16)
1214 		*limit = (1 << 16) - 1;
1215 }
1216 
1217 static int set_global_limit(const char *val, const struct kernel_param *kp)
1218 {
1219 	int rv;
1220 
1221 	rv = param_set_uint(val, kp);
1222 	if (rv)
1223 		return rv;
1224 
1225 	sanitize_global_limit((unsigned *)kp->arg);
1226 
1227 	return 0;
1228 }
1229 
1230 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1231 {
1232 	int cap_sys_admin = capable(CAP_SYS_ADMIN);
1233 
1234 	if (arg->minor < 13)
1235 		return;
1236 
1237 	sanitize_global_limit(&max_user_bgreq);
1238 	sanitize_global_limit(&max_user_congthresh);
1239 
1240 	spin_lock(&fc->bg_lock);
1241 	if (arg->max_background) {
1242 		fc->max_background = arg->max_background;
1243 
1244 		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1245 			fc->max_background = max_user_bgreq;
1246 	}
1247 	if (arg->congestion_threshold) {
1248 		fc->congestion_threshold = arg->congestion_threshold;
1249 
1250 		if (!cap_sys_admin &&
1251 		    fc->congestion_threshold > max_user_congthresh)
1252 			fc->congestion_threshold = max_user_congthresh;
1253 	}
1254 	spin_unlock(&fc->bg_lock);
1255 }
1256 
1257 struct fuse_init_args {
1258 	struct fuse_args args;
1259 	struct fuse_init_in in;
1260 	struct fuse_init_out out;
1261 };
1262 
1263 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1264 			       int error)
1265 {
1266 	struct fuse_conn *fc = fm->fc;
1267 	struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1268 	struct fuse_init_out *arg = &ia->out;
1269 	bool ok = true;
1270 
1271 	if (error || arg->major != FUSE_KERNEL_VERSION)
1272 		ok = false;
1273 	else {
1274 		unsigned long ra_pages;
1275 
1276 		process_init_limits(fc, arg);
1277 
1278 		if (arg->minor >= 6) {
1279 			u64 flags = arg->flags;
1280 
1281 			if (flags & FUSE_INIT_EXT)
1282 				flags |= (u64) arg->flags2 << 32;
1283 
1284 			ra_pages = arg->max_readahead / PAGE_SIZE;
1285 			if (flags & FUSE_ASYNC_READ)
1286 				fc->async_read = 1;
1287 			if (!(flags & FUSE_POSIX_LOCKS))
1288 				fc->no_lock = 1;
1289 			if (arg->minor >= 17) {
1290 				if (!(flags & FUSE_FLOCK_LOCKS))
1291 					fc->no_flock = 1;
1292 			} else {
1293 				if (!(flags & FUSE_POSIX_LOCKS))
1294 					fc->no_flock = 1;
1295 			}
1296 			if (flags & FUSE_ATOMIC_O_TRUNC)
1297 				fc->atomic_o_trunc = 1;
1298 			if (arg->minor >= 9) {
1299 				/* LOOKUP has dependency on proto version */
1300 				if (flags & FUSE_EXPORT_SUPPORT)
1301 					fc->export_support = 1;
1302 			}
1303 			if (flags & FUSE_BIG_WRITES)
1304 				fc->big_writes = 1;
1305 			if (flags & FUSE_DONT_MASK)
1306 				fc->dont_mask = 1;
1307 			if (flags & FUSE_AUTO_INVAL_DATA)
1308 				fc->auto_inval_data = 1;
1309 			else if (flags & FUSE_EXPLICIT_INVAL_DATA)
1310 				fc->explicit_inval_data = 1;
1311 			if (flags & FUSE_DO_READDIRPLUS) {
1312 				fc->do_readdirplus = 1;
1313 				if (flags & FUSE_READDIRPLUS_AUTO)
1314 					fc->readdirplus_auto = 1;
1315 			}
1316 			if (flags & FUSE_ASYNC_DIO)
1317 				fc->async_dio = 1;
1318 			if (flags & FUSE_WRITEBACK_CACHE)
1319 				fc->writeback_cache = 1;
1320 			if (flags & FUSE_PARALLEL_DIROPS)
1321 				fc->parallel_dirops = 1;
1322 			if (flags & FUSE_HANDLE_KILLPRIV)
1323 				fc->handle_killpriv = 1;
1324 			if (arg->time_gran && arg->time_gran <= 1000000000)
1325 				fm->sb->s_time_gran = arg->time_gran;
1326 			if ((flags & FUSE_POSIX_ACL)) {
1327 				fc->default_permissions = 1;
1328 				fc->posix_acl = 1;
1329 			}
1330 			if (flags & FUSE_CACHE_SYMLINKS)
1331 				fc->cache_symlinks = 1;
1332 			if (flags & FUSE_ABORT_ERROR)
1333 				fc->abort_err = 1;
1334 			if (flags & FUSE_MAX_PAGES) {
1335 				fc->max_pages =
1336 					min_t(unsigned int, fc->max_pages_limit,
1337 					max_t(unsigned int, arg->max_pages, 1));
1338 			}
1339 			if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1340 				if (flags & FUSE_MAP_ALIGNMENT &&
1341 				    !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1342 					ok = false;
1343 				}
1344 				if (flags & FUSE_HAS_INODE_DAX)
1345 					fc->inode_dax = 1;
1346 			}
1347 			if (flags & FUSE_HANDLE_KILLPRIV_V2) {
1348 				fc->handle_killpriv_v2 = 1;
1349 				fm->sb->s_flags |= SB_NOSEC;
1350 			}
1351 			if (flags & FUSE_SETXATTR_EXT)
1352 				fc->setxattr_ext = 1;
1353 			if (flags & FUSE_SECURITY_CTX)
1354 				fc->init_security = 1;
1355 			if (flags & FUSE_CREATE_SUPP_GROUP)
1356 				fc->create_supp_group = 1;
1357 			if (flags & FUSE_DIRECT_IO_ALLOW_MMAP)
1358 				fc->direct_io_allow_mmap = 1;
1359 			/*
1360 			 * max_stack_depth is the max stack depth of FUSE fs,
1361 			 * so it has to be at least 1 to support passthrough
1362 			 * to backing files.
1363 			 *
1364 			 * with max_stack_depth > 1, the backing files can be
1365 			 * on a stacked fs (e.g. overlayfs) themselves and with
1366 			 * max_stack_depth == 1, FUSE fs can be stacked as the
1367 			 * underlying fs of a stacked fs (e.g. overlayfs).
1368 			 *
1369 			 * Also don't allow the combination of FUSE_PASSTHROUGH
1370 			 * and FUSE_WRITEBACK_CACHE, current design doesn't handle
1371 			 * them together.
1372 			 */
1373 			if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH) &&
1374 			    (flags & FUSE_PASSTHROUGH) &&
1375 			    arg->max_stack_depth > 0 &&
1376 			    arg->max_stack_depth <= FILESYSTEM_MAX_STACK_DEPTH &&
1377 			    !(flags & FUSE_WRITEBACK_CACHE))  {
1378 				fc->passthrough = 1;
1379 				fc->max_stack_depth = arg->max_stack_depth;
1380 				fm->sb->s_stack_depth = arg->max_stack_depth;
1381 			}
1382 			if (flags & FUSE_NO_EXPORT_SUPPORT)
1383 				fm->sb->s_export_op = &fuse_export_fid_operations;
1384 			if (flags & FUSE_ALLOW_IDMAP) {
1385 				if (fc->default_permissions)
1386 					fm->sb->s_iflags &= ~SB_I_NOIDMAP;
1387 				else
1388 					ok = false;
1389 			}
1390 		} else {
1391 			ra_pages = fc->max_read / PAGE_SIZE;
1392 			fc->no_lock = 1;
1393 			fc->no_flock = 1;
1394 		}
1395 
1396 		fm->sb->s_bdi->ra_pages =
1397 				min(fm->sb->s_bdi->ra_pages, ra_pages);
1398 		fc->minor = arg->minor;
1399 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1400 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
1401 		fc->conn_init = 1;
1402 	}
1403 	kfree(ia);
1404 
1405 	if (!ok) {
1406 		fc->conn_init = 0;
1407 		fc->conn_error = 1;
1408 	}
1409 
1410 	fuse_set_initialized(fc);
1411 	wake_up_all(&fc->blocked_waitq);
1412 }
1413 
1414 void fuse_send_init(struct fuse_mount *fm)
1415 {
1416 	struct fuse_init_args *ia;
1417 	u64 flags;
1418 
1419 	ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1420 
1421 	ia->in.major = FUSE_KERNEL_VERSION;
1422 	ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1423 	ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1424 	flags =
1425 		FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1426 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1427 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1428 		FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1429 		FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1430 		FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1431 		FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1432 		FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1433 		FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1434 		FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT | FUSE_INIT_EXT |
1435 		FUSE_SECURITY_CTX | FUSE_CREATE_SUPP_GROUP |
1436 		FUSE_HAS_EXPIRE_ONLY | FUSE_DIRECT_IO_ALLOW_MMAP |
1437 		FUSE_NO_EXPORT_SUPPORT | FUSE_HAS_RESEND | FUSE_ALLOW_IDMAP;
1438 #ifdef CONFIG_FUSE_DAX
1439 	if (fm->fc->dax)
1440 		flags |= FUSE_MAP_ALIGNMENT;
1441 	if (fuse_is_inode_dax_mode(fm->fc->dax_mode))
1442 		flags |= FUSE_HAS_INODE_DAX;
1443 #endif
1444 	if (fm->fc->auto_submounts)
1445 		flags |= FUSE_SUBMOUNTS;
1446 	if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
1447 		flags |= FUSE_PASSTHROUGH;
1448 
1449 	ia->in.flags = flags;
1450 	ia->in.flags2 = flags >> 32;
1451 
1452 	ia->args.opcode = FUSE_INIT;
1453 	ia->args.in_numargs = 1;
1454 	ia->args.in_args[0].size = sizeof(ia->in);
1455 	ia->args.in_args[0].value = &ia->in;
1456 	ia->args.out_numargs = 1;
1457 	/* Variable length argument used for backward compatibility
1458 	   with interface version < 7.5.  Rest of init_out is zeroed
1459 	   by do_get_request(), so a short reply is not a problem */
1460 	ia->args.out_argvar = true;
1461 	ia->args.out_args[0].size = sizeof(ia->out);
1462 	ia->args.out_args[0].value = &ia->out;
1463 	ia->args.force = true;
1464 	ia->args.nocreds = true;
1465 	ia->args.end = process_init_reply;
1466 
1467 	if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1468 		process_init_reply(fm, &ia->args, -ENOTCONN);
1469 }
1470 EXPORT_SYMBOL_GPL(fuse_send_init);
1471 
1472 void fuse_free_conn(struct fuse_conn *fc)
1473 {
1474 	WARN_ON(!list_empty(&fc->devices));
1475 	kfree(fc);
1476 }
1477 EXPORT_SYMBOL_GPL(fuse_free_conn);
1478 
1479 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1480 {
1481 	int err;
1482 	char *suffix = "";
1483 
1484 	if (sb->s_bdev) {
1485 		suffix = "-fuseblk";
1486 		/*
1487 		 * sb->s_bdi points to blkdev's bdi however we want to redirect
1488 		 * it to our private bdi...
1489 		 */
1490 		bdi_put(sb->s_bdi);
1491 		sb->s_bdi = &noop_backing_dev_info;
1492 	}
1493 	err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1494 				   MINOR(fc->dev), suffix);
1495 	if (err)
1496 		return err;
1497 
1498 	/* fuse does it's own writeback accounting */
1499 	sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1500 	sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1501 
1502 	/*
1503 	 * For a single fuse filesystem use max 1% of dirty +
1504 	 * writeback threshold.
1505 	 *
1506 	 * This gives about 1M of write buffer for memory maps on a
1507 	 * machine with 1G and 10% dirty_ratio, which should be more
1508 	 * than enough.
1509 	 *
1510 	 * Privileged users can raise it by writing to
1511 	 *
1512 	 *    /sys/class/bdi/<bdi>/max_ratio
1513 	 */
1514 	bdi_set_max_ratio(sb->s_bdi, 1);
1515 
1516 	return 0;
1517 }
1518 
1519 struct fuse_dev *fuse_dev_alloc(void)
1520 {
1521 	struct fuse_dev *fud;
1522 	struct list_head *pq;
1523 
1524 	fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1525 	if (!fud)
1526 		return NULL;
1527 
1528 	pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1529 	if (!pq) {
1530 		kfree(fud);
1531 		return NULL;
1532 	}
1533 
1534 	fud->pq.processing = pq;
1535 	fuse_pqueue_init(&fud->pq);
1536 
1537 	return fud;
1538 }
1539 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1540 
1541 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1542 {
1543 	fud->fc = fuse_conn_get(fc);
1544 	spin_lock(&fc->lock);
1545 	list_add_tail(&fud->entry, &fc->devices);
1546 	spin_unlock(&fc->lock);
1547 }
1548 EXPORT_SYMBOL_GPL(fuse_dev_install);
1549 
1550 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1551 {
1552 	struct fuse_dev *fud;
1553 
1554 	fud = fuse_dev_alloc();
1555 	if (!fud)
1556 		return NULL;
1557 
1558 	fuse_dev_install(fud, fc);
1559 	return fud;
1560 }
1561 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1562 
1563 void fuse_dev_free(struct fuse_dev *fud)
1564 {
1565 	struct fuse_conn *fc = fud->fc;
1566 
1567 	if (fc) {
1568 		spin_lock(&fc->lock);
1569 		list_del(&fud->entry);
1570 		spin_unlock(&fc->lock);
1571 
1572 		fuse_conn_put(fc);
1573 	}
1574 	kfree(fud->pq.processing);
1575 	kfree(fud);
1576 }
1577 EXPORT_SYMBOL_GPL(fuse_dev_free);
1578 
1579 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1580 				      const struct fuse_inode *fi)
1581 {
1582 	struct timespec64 atime = inode_get_atime(&fi->inode);
1583 	struct timespec64 mtime = inode_get_mtime(&fi->inode);
1584 	struct timespec64 ctime = inode_get_ctime(&fi->inode);
1585 
1586 	*attr = (struct fuse_attr){
1587 		.ino		= fi->inode.i_ino,
1588 		.size		= fi->inode.i_size,
1589 		.blocks		= fi->inode.i_blocks,
1590 		.atime		= atime.tv_sec,
1591 		.mtime		= mtime.tv_sec,
1592 		.ctime		= ctime.tv_sec,
1593 		.atimensec	= atime.tv_nsec,
1594 		.mtimensec	= mtime.tv_nsec,
1595 		.ctimensec	= ctime.tv_nsec,
1596 		.mode		= fi->inode.i_mode,
1597 		.nlink		= fi->inode.i_nlink,
1598 		.uid		= __kuid_val(fi->inode.i_uid),
1599 		.gid		= __kgid_val(fi->inode.i_gid),
1600 		.rdev		= fi->inode.i_rdev,
1601 		.blksize	= 1u << fi->inode.i_blkbits,
1602 	};
1603 }
1604 
1605 static void fuse_sb_defaults(struct super_block *sb)
1606 {
1607 	sb->s_magic = FUSE_SUPER_MAGIC;
1608 	sb->s_op = &fuse_super_operations;
1609 	sb->s_xattr = fuse_xattr_handlers;
1610 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1611 	sb->s_time_gran = 1;
1612 	sb->s_export_op = &fuse_export_operations;
1613 	sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1614 	sb->s_iflags |= SB_I_NOIDMAP;
1615 	if (sb->s_user_ns != &init_user_ns)
1616 		sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1617 	sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1618 }
1619 
1620 static int fuse_fill_super_submount(struct super_block *sb,
1621 				    struct fuse_inode *parent_fi)
1622 {
1623 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1624 	struct super_block *parent_sb = parent_fi->inode.i_sb;
1625 	struct fuse_attr root_attr;
1626 	struct inode *root;
1627 	struct fuse_submount_lookup *sl;
1628 	struct fuse_inode *fi;
1629 
1630 	fuse_sb_defaults(sb);
1631 	fm->sb = sb;
1632 
1633 	WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1634 	sb->s_bdi = bdi_get(parent_sb->s_bdi);
1635 
1636 	sb->s_xattr = parent_sb->s_xattr;
1637 	sb->s_export_op = parent_sb->s_export_op;
1638 	sb->s_time_gran = parent_sb->s_time_gran;
1639 	sb->s_blocksize = parent_sb->s_blocksize;
1640 	sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1641 	sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1642 	if (parent_sb->s_subtype && !sb->s_subtype)
1643 		return -ENOMEM;
1644 
1645 	fuse_fill_attr_from_inode(&root_attr, parent_fi);
1646 	root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0,
1647 			 fuse_get_evict_ctr(fm->fc));
1648 	/*
1649 	 * This inode is just a duplicate, so it is not looked up and
1650 	 * its nlookup should not be incremented.  fuse_iget() does
1651 	 * that, though, so undo it here.
1652 	 */
1653 	fi = get_fuse_inode(root);
1654 	fi->nlookup--;
1655 
1656 	sb->s_d_op = &fuse_dentry_operations;
1657 	sb->s_root = d_make_root(root);
1658 	if (!sb->s_root)
1659 		return -ENOMEM;
1660 
1661 	/*
1662 	 * Grab the parent's submount_lookup pointer and take a
1663 	 * reference on the shared nlookup from the parent.  This is to
1664 	 * prevent the last forget for this nodeid from getting
1665 	 * triggered until all users have finished with it.
1666 	 */
1667 	sl = parent_fi->submount_lookup;
1668 	WARN_ON(!sl);
1669 	if (sl) {
1670 		refcount_inc(&sl->count);
1671 		fi->submount_lookup = sl;
1672 	}
1673 
1674 	return 0;
1675 }
1676 
1677 /* Filesystem context private data holds the FUSE inode of the mount point */
1678 static int fuse_get_tree_submount(struct fs_context *fsc)
1679 {
1680 	struct fuse_mount *fm;
1681 	struct fuse_inode *mp_fi = fsc->fs_private;
1682 	struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1683 	struct super_block *sb;
1684 	int err;
1685 
1686 	fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1687 	if (!fm)
1688 		return -ENOMEM;
1689 
1690 	fm->fc = fuse_conn_get(fc);
1691 	fsc->s_fs_info = fm;
1692 	sb = sget_fc(fsc, NULL, set_anon_super_fc);
1693 	if (fsc->s_fs_info)
1694 		fuse_mount_destroy(fm);
1695 	if (IS_ERR(sb))
1696 		return PTR_ERR(sb);
1697 
1698 	/* Initialize superblock, making @mp_fi its root */
1699 	err = fuse_fill_super_submount(sb, mp_fi);
1700 	if (err) {
1701 		deactivate_locked_super(sb);
1702 		return err;
1703 	}
1704 
1705 	down_write(&fc->killsb);
1706 	list_add_tail(&fm->fc_entry, &fc->mounts);
1707 	up_write(&fc->killsb);
1708 
1709 	sb->s_flags |= SB_ACTIVE;
1710 	fsc->root = dget(sb->s_root);
1711 
1712 	return 0;
1713 }
1714 
1715 static const struct fs_context_operations fuse_context_submount_ops = {
1716 	.get_tree	= fuse_get_tree_submount,
1717 };
1718 
1719 int fuse_init_fs_context_submount(struct fs_context *fsc)
1720 {
1721 	fsc->ops = &fuse_context_submount_ops;
1722 	return 0;
1723 }
1724 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1725 
1726 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1727 {
1728 	struct fuse_dev *fud = NULL;
1729 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1730 	struct fuse_conn *fc = fm->fc;
1731 	struct inode *root;
1732 	struct dentry *root_dentry;
1733 	int err;
1734 
1735 	err = -EINVAL;
1736 	if (sb->s_flags & SB_MANDLOCK)
1737 		goto err;
1738 
1739 	rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1740 	fuse_sb_defaults(sb);
1741 
1742 	if (ctx->is_bdev) {
1743 #ifdef CONFIG_BLOCK
1744 		err = -EINVAL;
1745 		if (!sb_set_blocksize(sb, ctx->blksize))
1746 			goto err;
1747 #endif
1748 	} else {
1749 		sb->s_blocksize = PAGE_SIZE;
1750 		sb->s_blocksize_bits = PAGE_SHIFT;
1751 	}
1752 
1753 	sb->s_subtype = ctx->subtype;
1754 	ctx->subtype = NULL;
1755 	if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1756 		err = fuse_dax_conn_alloc(fc, ctx->dax_mode, ctx->dax_dev);
1757 		if (err)
1758 			goto err;
1759 	}
1760 
1761 	if (ctx->fudptr) {
1762 		err = -ENOMEM;
1763 		fud = fuse_dev_alloc_install(fc);
1764 		if (!fud)
1765 			goto err_free_dax;
1766 	}
1767 
1768 	fc->dev = sb->s_dev;
1769 	fm->sb = sb;
1770 	err = fuse_bdi_init(fc, sb);
1771 	if (err)
1772 		goto err_dev_free;
1773 
1774 	/* Handle umasking inside the fuse code */
1775 	if (sb->s_flags & SB_POSIXACL)
1776 		fc->dont_mask = 1;
1777 	sb->s_flags |= SB_POSIXACL;
1778 
1779 	fc->default_permissions = ctx->default_permissions;
1780 	fc->allow_other = ctx->allow_other;
1781 	fc->user_id = ctx->user_id;
1782 	fc->group_id = ctx->group_id;
1783 	fc->legacy_opts_show = ctx->legacy_opts_show;
1784 	fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1785 	fc->destroy = ctx->destroy;
1786 	fc->no_control = ctx->no_control;
1787 	fc->no_force_umount = ctx->no_force_umount;
1788 
1789 	err = -ENOMEM;
1790 	root = fuse_get_root_inode(sb, ctx->rootmode);
1791 	sb->s_d_op = &fuse_root_dentry_operations;
1792 	root_dentry = d_make_root(root);
1793 	if (!root_dentry)
1794 		goto err_dev_free;
1795 	/* Root dentry doesn't have .d_revalidate */
1796 	sb->s_d_op = &fuse_dentry_operations;
1797 
1798 	mutex_lock(&fuse_mutex);
1799 	err = -EINVAL;
1800 	if (ctx->fudptr && *ctx->fudptr)
1801 		goto err_unlock;
1802 
1803 	err = fuse_ctl_add_conn(fc);
1804 	if (err)
1805 		goto err_unlock;
1806 
1807 	list_add_tail(&fc->entry, &fuse_conn_list);
1808 	sb->s_root = root_dentry;
1809 	if (ctx->fudptr)
1810 		*ctx->fudptr = fud;
1811 	mutex_unlock(&fuse_mutex);
1812 	return 0;
1813 
1814  err_unlock:
1815 	mutex_unlock(&fuse_mutex);
1816 	dput(root_dentry);
1817  err_dev_free:
1818 	if (fud)
1819 		fuse_dev_free(fud);
1820  err_free_dax:
1821 	if (IS_ENABLED(CONFIG_FUSE_DAX))
1822 		fuse_dax_conn_free(fc);
1823  err:
1824 	return err;
1825 }
1826 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1827 
1828 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1829 {
1830 	struct fuse_fs_context *ctx = fsc->fs_private;
1831 	int err;
1832 
1833 	if (!ctx->file || !ctx->rootmode_present ||
1834 	    !ctx->user_id_present || !ctx->group_id_present)
1835 		return -EINVAL;
1836 
1837 	/*
1838 	 * Require mount to happen from the same user namespace which
1839 	 * opened /dev/fuse to prevent potential attacks.
1840 	 */
1841 	if ((ctx->file->f_op != &fuse_dev_operations) ||
1842 	    (ctx->file->f_cred->user_ns != sb->s_user_ns))
1843 		return -EINVAL;
1844 	ctx->fudptr = &ctx->file->private_data;
1845 
1846 	err = fuse_fill_super_common(sb, ctx);
1847 	if (err)
1848 		return err;
1849 	/* file->private_data shall be visible on all CPUs after this */
1850 	smp_mb();
1851 	fuse_send_init(get_fuse_mount_super(sb));
1852 	return 0;
1853 }
1854 
1855 /*
1856  * This is the path where user supplied an already initialized fuse dev.  In
1857  * this case never create a new super if the old one is gone.
1858  */
1859 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1860 {
1861 	return -ENOTCONN;
1862 }
1863 
1864 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1865 {
1866 
1867 	return fsc->sget_key == get_fuse_conn_super(sb);
1868 }
1869 
1870 static int fuse_get_tree(struct fs_context *fsc)
1871 {
1872 	struct fuse_fs_context *ctx = fsc->fs_private;
1873 	struct fuse_dev *fud;
1874 	struct fuse_conn *fc;
1875 	struct fuse_mount *fm;
1876 	struct super_block *sb;
1877 	int err;
1878 
1879 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1880 	if (!fc)
1881 		return -ENOMEM;
1882 
1883 	fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1884 	if (!fm) {
1885 		kfree(fc);
1886 		return -ENOMEM;
1887 	}
1888 
1889 	fuse_conn_init(fc, fm, fsc->user_ns, &fuse_dev_fiq_ops, NULL);
1890 	fc->release = fuse_free_conn;
1891 
1892 	fsc->s_fs_info = fm;
1893 
1894 	if (ctx->fd_present)
1895 		ctx->file = fget(ctx->fd);
1896 
1897 	if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1898 		err = get_tree_bdev(fsc, fuse_fill_super);
1899 		goto out;
1900 	}
1901 	/*
1902 	 * While block dev mount can be initialized with a dummy device fd
1903 	 * (found by device name), normal fuse mounts can't
1904 	 */
1905 	err = -EINVAL;
1906 	if (!ctx->file)
1907 		goto out;
1908 
1909 	/*
1910 	 * Allow creating a fuse mount with an already initialized fuse
1911 	 * connection
1912 	 */
1913 	fud = READ_ONCE(ctx->file->private_data);
1914 	if (ctx->file->f_op == &fuse_dev_operations && fud) {
1915 		fsc->sget_key = fud->fc;
1916 		sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1917 		err = PTR_ERR_OR_ZERO(sb);
1918 		if (!IS_ERR(sb))
1919 			fsc->root = dget(sb->s_root);
1920 	} else {
1921 		err = get_tree_nodev(fsc, fuse_fill_super);
1922 	}
1923 out:
1924 	if (fsc->s_fs_info)
1925 		fuse_mount_destroy(fm);
1926 	if (ctx->file)
1927 		fput(ctx->file);
1928 	return err;
1929 }
1930 
1931 static const struct fs_context_operations fuse_context_ops = {
1932 	.free		= fuse_free_fsc,
1933 	.parse_param	= fuse_parse_param,
1934 	.reconfigure	= fuse_reconfigure,
1935 	.get_tree	= fuse_get_tree,
1936 };
1937 
1938 /*
1939  * Set up the filesystem mount context.
1940  */
1941 static int fuse_init_fs_context(struct fs_context *fsc)
1942 {
1943 	struct fuse_fs_context *ctx;
1944 
1945 	ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1946 	if (!ctx)
1947 		return -ENOMEM;
1948 
1949 	ctx->max_read = ~0;
1950 	ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1951 	ctx->legacy_opts_show = true;
1952 
1953 #ifdef CONFIG_BLOCK
1954 	if (fsc->fs_type == &fuseblk_fs_type) {
1955 		ctx->is_bdev = true;
1956 		ctx->destroy = true;
1957 	}
1958 #endif
1959 
1960 	fsc->fs_private = ctx;
1961 	fsc->ops = &fuse_context_ops;
1962 	return 0;
1963 }
1964 
1965 bool fuse_mount_remove(struct fuse_mount *fm)
1966 {
1967 	struct fuse_conn *fc = fm->fc;
1968 	bool last = false;
1969 
1970 	down_write(&fc->killsb);
1971 	list_del_init(&fm->fc_entry);
1972 	if (list_empty(&fc->mounts))
1973 		last = true;
1974 	up_write(&fc->killsb);
1975 
1976 	return last;
1977 }
1978 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1979 
1980 void fuse_conn_destroy(struct fuse_mount *fm)
1981 {
1982 	struct fuse_conn *fc = fm->fc;
1983 
1984 	if (fc->destroy)
1985 		fuse_send_destroy(fm);
1986 
1987 	fuse_abort_conn(fc);
1988 	fuse_wait_aborted(fc);
1989 
1990 	if (!list_empty(&fc->entry)) {
1991 		mutex_lock(&fuse_mutex);
1992 		list_del(&fc->entry);
1993 		fuse_ctl_remove_conn(fc);
1994 		mutex_unlock(&fuse_mutex);
1995 	}
1996 }
1997 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1998 
1999 static void fuse_sb_destroy(struct super_block *sb)
2000 {
2001 	struct fuse_mount *fm = get_fuse_mount_super(sb);
2002 	bool last;
2003 
2004 	if (sb->s_root) {
2005 		last = fuse_mount_remove(fm);
2006 		if (last)
2007 			fuse_conn_destroy(fm);
2008 	}
2009 }
2010 
2011 void fuse_mount_destroy(struct fuse_mount *fm)
2012 {
2013 	fuse_conn_put(fm->fc);
2014 	kfree_rcu(fm, rcu);
2015 }
2016 EXPORT_SYMBOL(fuse_mount_destroy);
2017 
2018 static void fuse_kill_sb_anon(struct super_block *sb)
2019 {
2020 	fuse_sb_destroy(sb);
2021 	kill_anon_super(sb);
2022 	fuse_mount_destroy(get_fuse_mount_super(sb));
2023 }
2024 
2025 static struct file_system_type fuse_fs_type = {
2026 	.owner		= THIS_MODULE,
2027 	.name		= "fuse",
2028 	.fs_flags	= FS_HAS_SUBTYPE | FS_USERNS_MOUNT | FS_ALLOW_IDMAP,
2029 	.init_fs_context = fuse_init_fs_context,
2030 	.parameters	= fuse_fs_parameters,
2031 	.kill_sb	= fuse_kill_sb_anon,
2032 };
2033 MODULE_ALIAS_FS("fuse");
2034 
2035 #ifdef CONFIG_BLOCK
2036 static void fuse_kill_sb_blk(struct super_block *sb)
2037 {
2038 	fuse_sb_destroy(sb);
2039 	kill_block_super(sb);
2040 	fuse_mount_destroy(get_fuse_mount_super(sb));
2041 }
2042 
2043 static struct file_system_type fuseblk_fs_type = {
2044 	.owner		= THIS_MODULE,
2045 	.name		= "fuseblk",
2046 	.init_fs_context = fuse_init_fs_context,
2047 	.parameters	= fuse_fs_parameters,
2048 	.kill_sb	= fuse_kill_sb_blk,
2049 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE | FS_ALLOW_IDMAP,
2050 };
2051 MODULE_ALIAS_FS("fuseblk");
2052 
2053 static inline int register_fuseblk(void)
2054 {
2055 	return register_filesystem(&fuseblk_fs_type);
2056 }
2057 
2058 static inline void unregister_fuseblk(void)
2059 {
2060 	unregister_filesystem(&fuseblk_fs_type);
2061 }
2062 #else
2063 static inline int register_fuseblk(void)
2064 {
2065 	return 0;
2066 }
2067 
2068 static inline void unregister_fuseblk(void)
2069 {
2070 }
2071 #endif
2072 
2073 static void fuse_inode_init_once(void *foo)
2074 {
2075 	struct inode *inode = foo;
2076 
2077 	inode_init_once(inode);
2078 }
2079 
2080 static int __init fuse_fs_init(void)
2081 {
2082 	int err;
2083 
2084 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
2085 			sizeof(struct fuse_inode), 0,
2086 			SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
2087 			fuse_inode_init_once);
2088 	err = -ENOMEM;
2089 	if (!fuse_inode_cachep)
2090 		goto out;
2091 
2092 	err = register_fuseblk();
2093 	if (err)
2094 		goto out2;
2095 
2096 	err = register_filesystem(&fuse_fs_type);
2097 	if (err)
2098 		goto out3;
2099 
2100 	err = fuse_sysctl_register();
2101 	if (err)
2102 		goto out4;
2103 
2104 	return 0;
2105 
2106  out4:
2107 	unregister_filesystem(&fuse_fs_type);
2108  out3:
2109 	unregister_fuseblk();
2110  out2:
2111 	kmem_cache_destroy(fuse_inode_cachep);
2112  out:
2113 	return err;
2114 }
2115 
2116 static void fuse_fs_cleanup(void)
2117 {
2118 	fuse_sysctl_unregister();
2119 	unregister_filesystem(&fuse_fs_type);
2120 	unregister_fuseblk();
2121 
2122 	/*
2123 	 * Make sure all delayed rcu free inodes are flushed before we
2124 	 * destroy cache.
2125 	 */
2126 	rcu_barrier();
2127 	kmem_cache_destroy(fuse_inode_cachep);
2128 }
2129 
2130 static struct kobject *fuse_kobj;
2131 
2132 static int fuse_sysfs_init(void)
2133 {
2134 	int err;
2135 
2136 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
2137 	if (!fuse_kobj) {
2138 		err = -ENOMEM;
2139 		goto out_err;
2140 	}
2141 
2142 	err = sysfs_create_mount_point(fuse_kobj, "connections");
2143 	if (err)
2144 		goto out_fuse_unregister;
2145 
2146 	return 0;
2147 
2148  out_fuse_unregister:
2149 	kobject_put(fuse_kobj);
2150  out_err:
2151 	return err;
2152 }
2153 
2154 static void fuse_sysfs_cleanup(void)
2155 {
2156 	sysfs_remove_mount_point(fuse_kobj, "connections");
2157 	kobject_put(fuse_kobj);
2158 }
2159 
2160 static int __init fuse_init(void)
2161 {
2162 	int res;
2163 
2164 	pr_info("init (API version %i.%i)\n",
2165 		FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
2166 
2167 	INIT_LIST_HEAD(&fuse_conn_list);
2168 	res = fuse_fs_init();
2169 	if (res)
2170 		goto err;
2171 
2172 	res = fuse_dev_init();
2173 	if (res)
2174 		goto err_fs_cleanup;
2175 
2176 	res = fuse_sysfs_init();
2177 	if (res)
2178 		goto err_dev_cleanup;
2179 
2180 	res = fuse_ctl_init();
2181 	if (res)
2182 		goto err_sysfs_cleanup;
2183 
2184 	sanitize_global_limit(&max_user_bgreq);
2185 	sanitize_global_limit(&max_user_congthresh);
2186 
2187 	return 0;
2188 
2189  err_sysfs_cleanup:
2190 	fuse_sysfs_cleanup();
2191  err_dev_cleanup:
2192 	fuse_dev_cleanup();
2193  err_fs_cleanup:
2194 	fuse_fs_cleanup();
2195  err:
2196 	return res;
2197 }
2198 
2199 static void __exit fuse_exit(void)
2200 {
2201 	pr_debug("exit\n");
2202 
2203 	fuse_ctl_cleanup();
2204 	fuse_sysfs_cleanup();
2205 	fuse_fs_cleanup();
2206 	fuse_dev_cleanup();
2207 }
2208 
2209 module_init(fuse_init);
2210 module_exit(fuse_exit);
2211