xref: /openbmc/linux/fs/fuse/inode.c (revision e481ff3f)
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 
27 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
28 MODULE_DESCRIPTION("Filesystem in Userspace");
29 MODULE_LICENSE("GPL");
30 
31 static struct kmem_cache *fuse_inode_cachep;
32 struct list_head fuse_conn_list;
33 DEFINE_MUTEX(fuse_mutex);
34 
35 static int set_global_limit(const char *val, const struct kernel_param *kp);
36 
37 unsigned max_user_bgreq;
38 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
39 		  &max_user_bgreq, 0644);
40 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
41 MODULE_PARM_DESC(max_user_bgreq,
42  "Global limit for the maximum number of backgrounded requests an "
43  "unprivileged user can set");
44 
45 unsigned max_user_congthresh;
46 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
47 		  &max_user_congthresh, 0644);
48 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
49 MODULE_PARM_DESC(max_user_congthresh,
50  "Global limit for the maximum congestion threshold an "
51  "unprivileged user can set");
52 
53 #define FUSE_SUPER_MAGIC 0x65735546
54 
55 #define FUSE_DEFAULT_BLKSIZE 512
56 
57 /** Maximum number of outstanding background requests */
58 #define FUSE_DEFAULT_MAX_BACKGROUND 12
59 
60 /** Congestion starts at 75% of maximum */
61 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
62 
63 #ifdef CONFIG_BLOCK
64 static struct file_system_type fuseblk_fs_type;
65 #endif
66 
67 struct fuse_forget_link *fuse_alloc_forget(void)
68 {
69 	return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
70 }
71 
72 static struct inode *fuse_alloc_inode(struct super_block *sb)
73 {
74 	struct fuse_inode *fi;
75 
76 	fi = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
77 	if (!fi)
78 		return NULL;
79 
80 	fi->i_time = 0;
81 	fi->inval_mask = 0;
82 	fi->nodeid = 0;
83 	fi->nlookup = 0;
84 	fi->attr_version = 0;
85 	fi->orig_ino = 0;
86 	fi->state = 0;
87 	mutex_init(&fi->mutex);
88 	init_rwsem(&fi->i_mmap_sem);
89 	spin_lock_init(&fi->lock);
90 	fi->forget = fuse_alloc_forget();
91 	if (!fi->forget)
92 		goto out_free;
93 
94 	if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
95 		goto out_free_forget;
96 
97 	return &fi->inode;
98 
99 out_free_forget:
100 	kfree(fi->forget);
101 out_free:
102 	kmem_cache_free(fuse_inode_cachep, fi);
103 	return NULL;
104 }
105 
106 static void fuse_free_inode(struct inode *inode)
107 {
108 	struct fuse_inode *fi = get_fuse_inode(inode);
109 
110 	mutex_destroy(&fi->mutex);
111 	kfree(fi->forget);
112 #ifdef CONFIG_FUSE_DAX
113 	kfree(fi->dax);
114 #endif
115 	kmem_cache_free(fuse_inode_cachep, fi);
116 }
117 
118 static void fuse_evict_inode(struct inode *inode)
119 {
120 	struct fuse_inode *fi = get_fuse_inode(inode);
121 
122 	truncate_inode_pages_final(&inode->i_data);
123 	clear_inode(inode);
124 	if (inode->i_sb->s_flags & SB_ACTIVE) {
125 		struct fuse_conn *fc = get_fuse_conn(inode);
126 
127 		if (FUSE_IS_DAX(inode))
128 			fuse_dax_inode_cleanup(inode);
129 		if (fi->nlookup) {
130 			fuse_queue_forget(fc, fi->forget, fi->nodeid,
131 					  fi->nlookup);
132 			fi->forget = NULL;
133 		}
134 	}
135 	if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
136 		WARN_ON(!list_empty(&fi->write_files));
137 		WARN_ON(!list_empty(&fi->queued_writes));
138 	}
139 }
140 
141 static int fuse_reconfigure(struct fs_context *fc)
142 {
143 	struct super_block *sb = fc->root->d_sb;
144 
145 	sync_filesystem(sb);
146 	if (fc->sb_flags & SB_MANDLOCK)
147 		return -EINVAL;
148 
149 	return 0;
150 }
151 
152 /*
153  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
154  * so that it will fit.
155  */
156 static ino_t fuse_squash_ino(u64 ino64)
157 {
158 	ino_t ino = (ino_t) ino64;
159 	if (sizeof(ino_t) < sizeof(u64))
160 		ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
161 	return ino;
162 }
163 
164 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
165 				   u64 attr_valid)
166 {
167 	struct fuse_conn *fc = get_fuse_conn(inode);
168 	struct fuse_inode *fi = get_fuse_inode(inode);
169 
170 	lockdep_assert_held(&fi->lock);
171 
172 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
173 	fi->i_time = attr_valid;
174 	WRITE_ONCE(fi->inval_mask, 0);
175 
176 	inode->i_ino     = fuse_squash_ino(attr->ino);
177 	inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
178 	set_nlink(inode, attr->nlink);
179 	inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
180 	inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
181 	inode->i_blocks  = attr->blocks;
182 	inode->i_atime.tv_sec   = attr->atime;
183 	inode->i_atime.tv_nsec  = attr->atimensec;
184 	/* mtime from server may be stale due to local buffered write */
185 	if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
186 		inode->i_mtime.tv_sec   = attr->mtime;
187 		inode->i_mtime.tv_nsec  = attr->mtimensec;
188 		inode->i_ctime.tv_sec   = attr->ctime;
189 		inode->i_ctime.tv_nsec  = attr->ctimensec;
190 	}
191 
192 	if (attr->blksize != 0)
193 		inode->i_blkbits = ilog2(attr->blksize);
194 	else
195 		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
196 
197 	/*
198 	 * Don't set the sticky bit in i_mode, unless we want the VFS
199 	 * to check permissions.  This prevents failures due to the
200 	 * check in may_delete().
201 	 */
202 	fi->orig_i_mode = inode->i_mode;
203 	if (!fc->default_permissions)
204 		inode->i_mode &= ~S_ISVTX;
205 
206 	fi->orig_ino = attr->ino;
207 
208 	/*
209 	 * We are refreshing inode data and it is possible that another
210 	 * client set suid/sgid or security.capability xattr. So clear
211 	 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
212 	 * was set or if security.capability xattr was set. But we don't
213 	 * know if security.capability has been set or not. So clear it
214 	 * anyway. Its less efficient but should be safe.
215 	 */
216 	inode->i_flags &= ~S_NOSEC;
217 }
218 
219 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
220 			    u64 attr_valid, u64 attr_version)
221 {
222 	struct fuse_conn *fc = get_fuse_conn(inode);
223 	struct fuse_inode *fi = get_fuse_inode(inode);
224 	bool is_wb = fc->writeback_cache;
225 	loff_t oldsize;
226 	struct timespec64 old_mtime;
227 
228 	spin_lock(&fi->lock);
229 	if ((attr_version != 0 && fi->attr_version > attr_version) ||
230 	    test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
231 		spin_unlock(&fi->lock);
232 		return;
233 	}
234 
235 	old_mtime = inode->i_mtime;
236 	fuse_change_attributes_common(inode, attr, attr_valid);
237 
238 	oldsize = inode->i_size;
239 	/*
240 	 * In case of writeback_cache enabled, the cached writes beyond EOF
241 	 * extend local i_size without keeping userspace server in sync. So,
242 	 * attr->size coming from server can be stale. We cannot trust it.
243 	 */
244 	if (!is_wb || !S_ISREG(inode->i_mode))
245 		i_size_write(inode, attr->size);
246 	spin_unlock(&fi->lock);
247 
248 	if (!is_wb && S_ISREG(inode->i_mode)) {
249 		bool inval = false;
250 
251 		if (oldsize != attr->size) {
252 			truncate_pagecache(inode, attr->size);
253 			if (!fc->explicit_inval_data)
254 				inval = true;
255 		} else if (fc->auto_inval_data) {
256 			struct timespec64 new_mtime = {
257 				.tv_sec = attr->mtime,
258 				.tv_nsec = attr->mtimensec,
259 			};
260 
261 			/*
262 			 * Auto inval mode also checks and invalidates if mtime
263 			 * has changed.
264 			 */
265 			if (!timespec64_equal(&old_mtime, &new_mtime))
266 				inval = true;
267 		}
268 
269 		if (inval)
270 			invalidate_inode_pages2(inode->i_mapping);
271 	}
272 }
273 
274 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
275 {
276 	inode->i_mode = attr->mode & S_IFMT;
277 	inode->i_size = attr->size;
278 	inode->i_mtime.tv_sec  = attr->mtime;
279 	inode->i_mtime.tv_nsec = attr->mtimensec;
280 	inode->i_ctime.tv_sec  = attr->ctime;
281 	inode->i_ctime.tv_nsec = attr->ctimensec;
282 	if (S_ISREG(inode->i_mode)) {
283 		fuse_init_common(inode);
284 		fuse_init_file_inode(inode);
285 	} else if (S_ISDIR(inode->i_mode))
286 		fuse_init_dir(inode);
287 	else if (S_ISLNK(inode->i_mode))
288 		fuse_init_symlink(inode);
289 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
290 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
291 		fuse_init_common(inode);
292 		init_special_inode(inode, inode->i_mode,
293 				   new_decode_dev(attr->rdev));
294 	} else
295 		BUG();
296 }
297 
298 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
299 {
300 	u64 nodeid = *(u64 *) _nodeidp;
301 	if (get_node_id(inode) == nodeid)
302 		return 1;
303 	else
304 		return 0;
305 }
306 
307 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
308 {
309 	u64 nodeid = *(u64 *) _nodeidp;
310 	get_fuse_inode(inode)->nodeid = nodeid;
311 	return 0;
312 }
313 
314 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
315 			int generation, struct fuse_attr *attr,
316 			u64 attr_valid, u64 attr_version)
317 {
318 	struct inode *inode;
319 	struct fuse_inode *fi;
320 	struct fuse_conn *fc = get_fuse_conn_super(sb);
321 
322 	/*
323 	 * Auto mount points get their node id from the submount root, which is
324 	 * not a unique identifier within this filesystem.
325 	 *
326 	 * To avoid conflicts, do not place submount points into the inode hash
327 	 * table.
328 	 */
329 	if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
330 	    S_ISDIR(attr->mode)) {
331 		inode = new_inode(sb);
332 		if (!inode)
333 			return NULL;
334 
335 		fuse_init_inode(inode, attr);
336 		get_fuse_inode(inode)->nodeid = nodeid;
337 		inode->i_flags |= S_AUTOMOUNT;
338 		goto done;
339 	}
340 
341 retry:
342 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
343 	if (!inode)
344 		return NULL;
345 
346 	if ((inode->i_state & I_NEW)) {
347 		inode->i_flags |= S_NOATIME;
348 		if (!fc->writeback_cache || !S_ISREG(attr->mode))
349 			inode->i_flags |= S_NOCMTIME;
350 		inode->i_generation = generation;
351 		fuse_init_inode(inode, attr);
352 		unlock_new_inode(inode);
353 	} else if (fuse_stale_inode(inode, generation, attr)) {
354 		/* nodeid was reused, any I/O on the old inode should fail */
355 		fuse_make_bad(inode);
356 		iput(inode);
357 		goto retry;
358 	}
359 done:
360 	fi = get_fuse_inode(inode);
361 	spin_lock(&fi->lock);
362 	fi->nlookup++;
363 	spin_unlock(&fi->lock);
364 	fuse_change_attributes(inode, attr, attr_valid, attr_version);
365 
366 	return inode;
367 }
368 
369 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
370 			   struct fuse_mount **fm)
371 {
372 	struct fuse_mount *fm_iter;
373 	struct inode *inode;
374 
375 	WARN_ON(!rwsem_is_locked(&fc->killsb));
376 	list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
377 		if (!fm_iter->sb)
378 			continue;
379 
380 		inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
381 		if (inode) {
382 			if (fm)
383 				*fm = fm_iter;
384 			return inode;
385 		}
386 	}
387 
388 	return NULL;
389 }
390 
391 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
392 			     loff_t offset, loff_t len)
393 {
394 	struct fuse_inode *fi;
395 	struct inode *inode;
396 	pgoff_t pg_start;
397 	pgoff_t pg_end;
398 
399 	inode = fuse_ilookup(fc, nodeid, NULL);
400 	if (!inode)
401 		return -ENOENT;
402 
403 	fi = get_fuse_inode(inode);
404 	spin_lock(&fi->lock);
405 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
406 	spin_unlock(&fi->lock);
407 
408 	fuse_invalidate_attr(inode);
409 	forget_all_cached_acls(inode);
410 	if (offset >= 0) {
411 		pg_start = offset >> PAGE_SHIFT;
412 		if (len <= 0)
413 			pg_end = -1;
414 		else
415 			pg_end = (offset + len - 1) >> PAGE_SHIFT;
416 		invalidate_inode_pages2_range(inode->i_mapping,
417 					      pg_start, pg_end);
418 	}
419 	iput(inode);
420 	return 0;
421 }
422 
423 bool fuse_lock_inode(struct inode *inode)
424 {
425 	bool locked = false;
426 
427 	if (!get_fuse_conn(inode)->parallel_dirops) {
428 		mutex_lock(&get_fuse_inode(inode)->mutex);
429 		locked = true;
430 	}
431 
432 	return locked;
433 }
434 
435 void fuse_unlock_inode(struct inode *inode, bool locked)
436 {
437 	if (locked)
438 		mutex_unlock(&get_fuse_inode(inode)->mutex);
439 }
440 
441 static void fuse_umount_begin(struct super_block *sb)
442 {
443 	struct fuse_conn *fc = get_fuse_conn_super(sb);
444 
445 	if (!fc->no_force_umount)
446 		fuse_abort_conn(fc);
447 }
448 
449 static void fuse_send_destroy(struct fuse_mount *fm)
450 {
451 	if (fm->fc->conn_init) {
452 		FUSE_ARGS(args);
453 
454 		args.opcode = FUSE_DESTROY;
455 		args.force = true;
456 		args.nocreds = true;
457 		fuse_simple_request(fm, &args);
458 	}
459 }
460 
461 static void fuse_put_super(struct super_block *sb)
462 {
463 	struct fuse_mount *fm = get_fuse_mount_super(sb);
464 
465 	fuse_conn_put(fm->fc);
466 	kfree(fm);
467 }
468 
469 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
470 {
471 	stbuf->f_type    = FUSE_SUPER_MAGIC;
472 	stbuf->f_bsize   = attr->bsize;
473 	stbuf->f_frsize  = attr->frsize;
474 	stbuf->f_blocks  = attr->blocks;
475 	stbuf->f_bfree   = attr->bfree;
476 	stbuf->f_bavail  = attr->bavail;
477 	stbuf->f_files   = attr->files;
478 	stbuf->f_ffree   = attr->ffree;
479 	stbuf->f_namelen = attr->namelen;
480 	/* fsid is left zero */
481 }
482 
483 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
484 {
485 	struct super_block *sb = dentry->d_sb;
486 	struct fuse_mount *fm = get_fuse_mount_super(sb);
487 	FUSE_ARGS(args);
488 	struct fuse_statfs_out outarg;
489 	int err;
490 
491 	if (!fuse_allow_current_process(fm->fc)) {
492 		buf->f_type = FUSE_SUPER_MAGIC;
493 		return 0;
494 	}
495 
496 	memset(&outarg, 0, sizeof(outarg));
497 	args.in_numargs = 0;
498 	args.opcode = FUSE_STATFS;
499 	args.nodeid = get_node_id(d_inode(dentry));
500 	args.out_numargs = 1;
501 	args.out_args[0].size = sizeof(outarg);
502 	args.out_args[0].value = &outarg;
503 	err = fuse_simple_request(fm, &args);
504 	if (!err)
505 		convert_fuse_statfs(buf, &outarg.st);
506 	return err;
507 }
508 
509 static int fuse_sync_fs(struct super_block *sb, int wait)
510 {
511 	struct fuse_mount *fm = get_fuse_mount_super(sb);
512 	struct fuse_conn *fc = fm->fc;
513 	struct fuse_syncfs_in inarg;
514 	FUSE_ARGS(args);
515 	int err;
516 
517 	/*
518 	 * Userspace cannot handle the wait == 0 case.  Avoid a
519 	 * gratuitous roundtrip.
520 	 */
521 	if (!wait)
522 		return 0;
523 
524 	/* The filesystem is being unmounted.  Nothing to do. */
525 	if (!sb->s_root)
526 		return 0;
527 
528 	if (!fc->sync_fs)
529 		return 0;
530 
531 	memset(&inarg, 0, sizeof(inarg));
532 	args.in_numargs = 1;
533 	args.in_args[0].size = sizeof(inarg);
534 	args.in_args[0].value = &inarg;
535 	args.opcode = FUSE_SYNCFS;
536 	args.nodeid = get_node_id(sb->s_root->d_inode);
537 	args.out_numargs = 0;
538 
539 	err = fuse_simple_request(fm, &args);
540 	if (err == -ENOSYS) {
541 		fc->sync_fs = 0;
542 		err = 0;
543 	}
544 
545 	return err;
546 }
547 
548 enum {
549 	OPT_SOURCE,
550 	OPT_SUBTYPE,
551 	OPT_FD,
552 	OPT_ROOTMODE,
553 	OPT_USER_ID,
554 	OPT_GROUP_ID,
555 	OPT_DEFAULT_PERMISSIONS,
556 	OPT_ALLOW_OTHER,
557 	OPT_MAX_READ,
558 	OPT_BLKSIZE,
559 	OPT_ERR
560 };
561 
562 static const struct fs_parameter_spec fuse_fs_parameters[] = {
563 	fsparam_string	("source",		OPT_SOURCE),
564 	fsparam_u32	("fd",			OPT_FD),
565 	fsparam_u32oct	("rootmode",		OPT_ROOTMODE),
566 	fsparam_u32	("user_id",		OPT_USER_ID),
567 	fsparam_u32	("group_id",		OPT_GROUP_ID),
568 	fsparam_flag	("default_permissions",	OPT_DEFAULT_PERMISSIONS),
569 	fsparam_flag	("allow_other",		OPT_ALLOW_OTHER),
570 	fsparam_u32	("max_read",		OPT_MAX_READ),
571 	fsparam_u32	("blksize",		OPT_BLKSIZE),
572 	fsparam_string	("subtype",		OPT_SUBTYPE),
573 	{}
574 };
575 
576 static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param)
577 {
578 	struct fs_parse_result result;
579 	struct fuse_fs_context *ctx = fc->fs_private;
580 	int opt;
581 
582 	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
583 		/*
584 		 * Ignore options coming from mount(MS_REMOUNT) for backward
585 		 * compatibility.
586 		 */
587 		if (fc->oldapi)
588 			return 0;
589 
590 		return invalfc(fc, "No changes allowed in reconfigure");
591 	}
592 
593 	opt = fs_parse(fc, fuse_fs_parameters, param, &result);
594 	if (opt < 0)
595 		return opt;
596 
597 	switch (opt) {
598 	case OPT_SOURCE:
599 		if (fc->source)
600 			return invalfc(fc, "Multiple sources specified");
601 		fc->source = param->string;
602 		param->string = NULL;
603 		break;
604 
605 	case OPT_SUBTYPE:
606 		if (ctx->subtype)
607 			return invalfc(fc, "Multiple subtypes specified");
608 		ctx->subtype = param->string;
609 		param->string = NULL;
610 		return 0;
611 
612 	case OPT_FD:
613 		ctx->fd = result.uint_32;
614 		ctx->fd_present = true;
615 		break;
616 
617 	case OPT_ROOTMODE:
618 		if (!fuse_valid_type(result.uint_32))
619 			return invalfc(fc, "Invalid rootmode");
620 		ctx->rootmode = result.uint_32;
621 		ctx->rootmode_present = true;
622 		break;
623 
624 	case OPT_USER_ID:
625 		ctx->user_id = make_kuid(fc->user_ns, result.uint_32);
626 		if (!uid_valid(ctx->user_id))
627 			return invalfc(fc, "Invalid user_id");
628 		ctx->user_id_present = true;
629 		break;
630 
631 	case OPT_GROUP_ID:
632 		ctx->group_id = make_kgid(fc->user_ns, result.uint_32);
633 		if (!gid_valid(ctx->group_id))
634 			return invalfc(fc, "Invalid group_id");
635 		ctx->group_id_present = true;
636 		break;
637 
638 	case OPT_DEFAULT_PERMISSIONS:
639 		ctx->default_permissions = true;
640 		break;
641 
642 	case OPT_ALLOW_OTHER:
643 		ctx->allow_other = true;
644 		break;
645 
646 	case OPT_MAX_READ:
647 		ctx->max_read = result.uint_32;
648 		break;
649 
650 	case OPT_BLKSIZE:
651 		if (!ctx->is_bdev)
652 			return invalfc(fc, "blksize only supported for fuseblk");
653 		ctx->blksize = result.uint_32;
654 		break;
655 
656 	default:
657 		return -EINVAL;
658 	}
659 
660 	return 0;
661 }
662 
663 static void fuse_free_fc(struct fs_context *fc)
664 {
665 	struct fuse_fs_context *ctx = fc->fs_private;
666 
667 	if (ctx) {
668 		kfree(ctx->subtype);
669 		kfree(ctx);
670 	}
671 }
672 
673 static int fuse_show_options(struct seq_file *m, struct dentry *root)
674 {
675 	struct super_block *sb = root->d_sb;
676 	struct fuse_conn *fc = get_fuse_conn_super(sb);
677 
678 	if (fc->legacy_opts_show) {
679 		seq_printf(m, ",user_id=%u",
680 			   from_kuid_munged(fc->user_ns, fc->user_id));
681 		seq_printf(m, ",group_id=%u",
682 			   from_kgid_munged(fc->user_ns, fc->group_id));
683 		if (fc->default_permissions)
684 			seq_puts(m, ",default_permissions");
685 		if (fc->allow_other)
686 			seq_puts(m, ",allow_other");
687 		if (fc->max_read != ~0)
688 			seq_printf(m, ",max_read=%u", fc->max_read);
689 		if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
690 			seq_printf(m, ",blksize=%lu", sb->s_blocksize);
691 	}
692 #ifdef CONFIG_FUSE_DAX
693 	if (fc->dax)
694 		seq_puts(m, ",dax");
695 #endif
696 
697 	return 0;
698 }
699 
700 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
701 			     const struct fuse_iqueue_ops *ops,
702 			     void *priv)
703 {
704 	memset(fiq, 0, sizeof(struct fuse_iqueue));
705 	spin_lock_init(&fiq->lock);
706 	init_waitqueue_head(&fiq->waitq);
707 	INIT_LIST_HEAD(&fiq->pending);
708 	INIT_LIST_HEAD(&fiq->interrupts);
709 	fiq->forget_list_tail = &fiq->forget_list_head;
710 	fiq->connected = 1;
711 	fiq->ops = ops;
712 	fiq->priv = priv;
713 }
714 
715 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
716 {
717 	unsigned int i;
718 
719 	spin_lock_init(&fpq->lock);
720 	for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
721 		INIT_LIST_HEAD(&fpq->processing[i]);
722 	INIT_LIST_HEAD(&fpq->io);
723 	fpq->connected = 1;
724 }
725 
726 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
727 		    struct user_namespace *user_ns,
728 		    const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
729 {
730 	memset(fc, 0, sizeof(*fc));
731 	spin_lock_init(&fc->lock);
732 	spin_lock_init(&fc->bg_lock);
733 	init_rwsem(&fc->killsb);
734 	refcount_set(&fc->count, 1);
735 	atomic_set(&fc->dev_count, 1);
736 	init_waitqueue_head(&fc->blocked_waitq);
737 	fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
738 	INIT_LIST_HEAD(&fc->bg_queue);
739 	INIT_LIST_HEAD(&fc->entry);
740 	INIT_LIST_HEAD(&fc->devices);
741 	atomic_set(&fc->num_waiting, 0);
742 	fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
743 	fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
744 	atomic64_set(&fc->khctr, 0);
745 	fc->polled_files = RB_ROOT;
746 	fc->blocked = 0;
747 	fc->initialized = 0;
748 	fc->connected = 1;
749 	atomic64_set(&fc->attr_version, 1);
750 	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
751 	fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
752 	fc->user_ns = get_user_ns(user_ns);
753 	fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
754 	fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
755 
756 	INIT_LIST_HEAD(&fc->mounts);
757 	list_add(&fm->fc_entry, &fc->mounts);
758 	fm->fc = fc;
759 }
760 EXPORT_SYMBOL_GPL(fuse_conn_init);
761 
762 void fuse_conn_put(struct fuse_conn *fc)
763 {
764 	if (refcount_dec_and_test(&fc->count)) {
765 		struct fuse_iqueue *fiq = &fc->iq;
766 
767 		if (IS_ENABLED(CONFIG_FUSE_DAX))
768 			fuse_dax_conn_free(fc);
769 		if (fiq->ops->release)
770 			fiq->ops->release(fiq);
771 		put_pid_ns(fc->pid_ns);
772 		put_user_ns(fc->user_ns);
773 		fc->release(fc);
774 	}
775 }
776 EXPORT_SYMBOL_GPL(fuse_conn_put);
777 
778 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
779 {
780 	refcount_inc(&fc->count);
781 	return fc;
782 }
783 EXPORT_SYMBOL_GPL(fuse_conn_get);
784 
785 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
786 {
787 	struct fuse_attr attr;
788 	memset(&attr, 0, sizeof(attr));
789 
790 	attr.mode = mode;
791 	attr.ino = FUSE_ROOT_ID;
792 	attr.nlink = 1;
793 	return fuse_iget(sb, 1, 0, &attr, 0, 0);
794 }
795 
796 struct fuse_inode_handle {
797 	u64 nodeid;
798 	u32 generation;
799 };
800 
801 static struct dentry *fuse_get_dentry(struct super_block *sb,
802 				      struct fuse_inode_handle *handle)
803 {
804 	struct fuse_conn *fc = get_fuse_conn_super(sb);
805 	struct inode *inode;
806 	struct dentry *entry;
807 	int err = -ESTALE;
808 
809 	if (handle->nodeid == 0)
810 		goto out_err;
811 
812 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
813 	if (!inode) {
814 		struct fuse_entry_out outarg;
815 		const struct qstr name = QSTR_INIT(".", 1);
816 
817 		if (!fc->export_support)
818 			goto out_err;
819 
820 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
821 				       &inode);
822 		if (err && err != -ENOENT)
823 			goto out_err;
824 		if (err || !inode) {
825 			err = -ESTALE;
826 			goto out_err;
827 		}
828 		err = -EIO;
829 		if (get_node_id(inode) != handle->nodeid)
830 			goto out_iput;
831 	}
832 	err = -ESTALE;
833 	if (inode->i_generation != handle->generation)
834 		goto out_iput;
835 
836 	entry = d_obtain_alias(inode);
837 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
838 		fuse_invalidate_entry_cache(entry);
839 
840 	return entry;
841 
842  out_iput:
843 	iput(inode);
844  out_err:
845 	return ERR_PTR(err);
846 }
847 
848 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
849 			   struct inode *parent)
850 {
851 	int len = parent ? 6 : 3;
852 	u64 nodeid;
853 	u32 generation;
854 
855 	if (*max_len < len) {
856 		*max_len = len;
857 		return  FILEID_INVALID;
858 	}
859 
860 	nodeid = get_fuse_inode(inode)->nodeid;
861 	generation = inode->i_generation;
862 
863 	fh[0] = (u32)(nodeid >> 32);
864 	fh[1] = (u32)(nodeid & 0xffffffff);
865 	fh[2] = generation;
866 
867 	if (parent) {
868 		nodeid = get_fuse_inode(parent)->nodeid;
869 		generation = parent->i_generation;
870 
871 		fh[3] = (u32)(nodeid >> 32);
872 		fh[4] = (u32)(nodeid & 0xffffffff);
873 		fh[5] = generation;
874 	}
875 
876 	*max_len = len;
877 	return parent ? 0x82 : 0x81;
878 }
879 
880 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
881 		struct fid *fid, int fh_len, int fh_type)
882 {
883 	struct fuse_inode_handle handle;
884 
885 	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
886 		return NULL;
887 
888 	handle.nodeid = (u64) fid->raw[0] << 32;
889 	handle.nodeid |= (u64) fid->raw[1];
890 	handle.generation = fid->raw[2];
891 	return fuse_get_dentry(sb, &handle);
892 }
893 
894 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
895 		struct fid *fid, int fh_len, int fh_type)
896 {
897 	struct fuse_inode_handle parent;
898 
899 	if (fh_type != 0x82 || fh_len < 6)
900 		return NULL;
901 
902 	parent.nodeid = (u64) fid->raw[3] << 32;
903 	parent.nodeid |= (u64) fid->raw[4];
904 	parent.generation = fid->raw[5];
905 	return fuse_get_dentry(sb, &parent);
906 }
907 
908 static struct dentry *fuse_get_parent(struct dentry *child)
909 {
910 	struct inode *child_inode = d_inode(child);
911 	struct fuse_conn *fc = get_fuse_conn(child_inode);
912 	struct inode *inode;
913 	struct dentry *parent;
914 	struct fuse_entry_out outarg;
915 	int err;
916 
917 	if (!fc->export_support)
918 		return ERR_PTR(-ESTALE);
919 
920 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
921 			       &dotdot_name, &outarg, &inode);
922 	if (err) {
923 		if (err == -ENOENT)
924 			return ERR_PTR(-ESTALE);
925 		return ERR_PTR(err);
926 	}
927 
928 	parent = d_obtain_alias(inode);
929 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
930 		fuse_invalidate_entry_cache(parent);
931 
932 	return parent;
933 }
934 
935 static const struct export_operations fuse_export_operations = {
936 	.fh_to_dentry	= fuse_fh_to_dentry,
937 	.fh_to_parent	= fuse_fh_to_parent,
938 	.encode_fh	= fuse_encode_fh,
939 	.get_parent	= fuse_get_parent,
940 };
941 
942 static const struct super_operations fuse_super_operations = {
943 	.alloc_inode    = fuse_alloc_inode,
944 	.free_inode     = fuse_free_inode,
945 	.evict_inode	= fuse_evict_inode,
946 	.write_inode	= fuse_write_inode,
947 	.drop_inode	= generic_delete_inode,
948 	.put_super	= fuse_put_super,
949 	.umount_begin	= fuse_umount_begin,
950 	.statfs		= fuse_statfs,
951 	.sync_fs	= fuse_sync_fs,
952 	.show_options	= fuse_show_options,
953 };
954 
955 static void sanitize_global_limit(unsigned *limit)
956 {
957 	/*
958 	 * The default maximum number of async requests is calculated to consume
959 	 * 1/2^13 of the total memory, assuming 392 bytes per request.
960 	 */
961 	if (*limit == 0)
962 		*limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
963 
964 	if (*limit >= 1 << 16)
965 		*limit = (1 << 16) - 1;
966 }
967 
968 static int set_global_limit(const char *val, const struct kernel_param *kp)
969 {
970 	int rv;
971 
972 	rv = param_set_uint(val, kp);
973 	if (rv)
974 		return rv;
975 
976 	sanitize_global_limit((unsigned *)kp->arg);
977 
978 	return 0;
979 }
980 
981 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
982 {
983 	int cap_sys_admin = capable(CAP_SYS_ADMIN);
984 
985 	if (arg->minor < 13)
986 		return;
987 
988 	sanitize_global_limit(&max_user_bgreq);
989 	sanitize_global_limit(&max_user_congthresh);
990 
991 	spin_lock(&fc->bg_lock);
992 	if (arg->max_background) {
993 		fc->max_background = arg->max_background;
994 
995 		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
996 			fc->max_background = max_user_bgreq;
997 	}
998 	if (arg->congestion_threshold) {
999 		fc->congestion_threshold = arg->congestion_threshold;
1000 
1001 		if (!cap_sys_admin &&
1002 		    fc->congestion_threshold > max_user_congthresh)
1003 			fc->congestion_threshold = max_user_congthresh;
1004 	}
1005 	spin_unlock(&fc->bg_lock);
1006 }
1007 
1008 struct fuse_init_args {
1009 	struct fuse_args args;
1010 	struct fuse_init_in in;
1011 	struct fuse_init_out out;
1012 };
1013 
1014 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1015 			       int error)
1016 {
1017 	struct fuse_conn *fc = fm->fc;
1018 	struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1019 	struct fuse_init_out *arg = &ia->out;
1020 	bool ok = true;
1021 
1022 	if (error || arg->major != FUSE_KERNEL_VERSION)
1023 		ok = false;
1024 	else {
1025 		unsigned long ra_pages;
1026 
1027 		process_init_limits(fc, arg);
1028 
1029 		if (arg->minor >= 6) {
1030 			ra_pages = arg->max_readahead / PAGE_SIZE;
1031 			if (arg->flags & FUSE_ASYNC_READ)
1032 				fc->async_read = 1;
1033 			if (!(arg->flags & FUSE_POSIX_LOCKS))
1034 				fc->no_lock = 1;
1035 			if (arg->minor >= 17) {
1036 				if (!(arg->flags & FUSE_FLOCK_LOCKS))
1037 					fc->no_flock = 1;
1038 			} else {
1039 				if (!(arg->flags & FUSE_POSIX_LOCKS))
1040 					fc->no_flock = 1;
1041 			}
1042 			if (arg->flags & FUSE_ATOMIC_O_TRUNC)
1043 				fc->atomic_o_trunc = 1;
1044 			if (arg->minor >= 9) {
1045 				/* LOOKUP has dependency on proto version */
1046 				if (arg->flags & FUSE_EXPORT_SUPPORT)
1047 					fc->export_support = 1;
1048 			}
1049 			if (arg->flags & FUSE_BIG_WRITES)
1050 				fc->big_writes = 1;
1051 			if (arg->flags & FUSE_DONT_MASK)
1052 				fc->dont_mask = 1;
1053 			if (arg->flags & FUSE_AUTO_INVAL_DATA)
1054 				fc->auto_inval_data = 1;
1055 			else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
1056 				fc->explicit_inval_data = 1;
1057 			if (arg->flags & FUSE_DO_READDIRPLUS) {
1058 				fc->do_readdirplus = 1;
1059 				if (arg->flags & FUSE_READDIRPLUS_AUTO)
1060 					fc->readdirplus_auto = 1;
1061 			}
1062 			if (arg->flags & FUSE_ASYNC_DIO)
1063 				fc->async_dio = 1;
1064 			if (arg->flags & FUSE_WRITEBACK_CACHE)
1065 				fc->writeback_cache = 1;
1066 			if (arg->flags & FUSE_PARALLEL_DIROPS)
1067 				fc->parallel_dirops = 1;
1068 			if (arg->flags & FUSE_HANDLE_KILLPRIV)
1069 				fc->handle_killpriv = 1;
1070 			if (arg->time_gran && arg->time_gran <= 1000000000)
1071 				fm->sb->s_time_gran = arg->time_gran;
1072 			if ((arg->flags & FUSE_POSIX_ACL)) {
1073 				fc->default_permissions = 1;
1074 				fc->posix_acl = 1;
1075 				fm->sb->s_xattr = fuse_acl_xattr_handlers;
1076 			}
1077 			if (arg->flags & FUSE_CACHE_SYMLINKS)
1078 				fc->cache_symlinks = 1;
1079 			if (arg->flags & FUSE_ABORT_ERROR)
1080 				fc->abort_err = 1;
1081 			if (arg->flags & FUSE_MAX_PAGES) {
1082 				fc->max_pages =
1083 					min_t(unsigned int, fc->max_pages_limit,
1084 					max_t(unsigned int, arg->max_pages, 1));
1085 			}
1086 			if (IS_ENABLED(CONFIG_FUSE_DAX) &&
1087 			    arg->flags & FUSE_MAP_ALIGNMENT &&
1088 			    !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1089 				ok = false;
1090 			}
1091 			if (arg->flags & FUSE_HANDLE_KILLPRIV_V2) {
1092 				fc->handle_killpriv_v2 = 1;
1093 				fm->sb->s_flags |= SB_NOSEC;
1094 			}
1095 			if (arg->flags & FUSE_SETXATTR_EXT)
1096 				fc->setxattr_ext = 1;
1097 		} else {
1098 			ra_pages = fc->max_read / PAGE_SIZE;
1099 			fc->no_lock = 1;
1100 			fc->no_flock = 1;
1101 		}
1102 
1103 		fm->sb->s_bdi->ra_pages =
1104 				min(fm->sb->s_bdi->ra_pages, ra_pages);
1105 		fc->minor = arg->minor;
1106 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1107 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
1108 		fc->conn_init = 1;
1109 	}
1110 	kfree(ia);
1111 
1112 	if (!ok) {
1113 		fc->conn_init = 0;
1114 		fc->conn_error = 1;
1115 	}
1116 
1117 	fuse_set_initialized(fc);
1118 	wake_up_all(&fc->blocked_waitq);
1119 }
1120 
1121 void fuse_send_init(struct fuse_mount *fm)
1122 {
1123 	struct fuse_init_args *ia;
1124 
1125 	ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1126 
1127 	ia->in.major = FUSE_KERNEL_VERSION;
1128 	ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1129 	ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1130 	ia->in.flags |=
1131 		FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1132 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1133 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1134 		FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1135 		FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1136 		FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1137 		FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1138 		FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1139 		FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1140 		FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT;
1141 #ifdef CONFIG_FUSE_DAX
1142 	if (fm->fc->dax)
1143 		ia->in.flags |= FUSE_MAP_ALIGNMENT;
1144 #endif
1145 	if (fm->fc->auto_submounts)
1146 		ia->in.flags |= FUSE_SUBMOUNTS;
1147 
1148 	ia->args.opcode = FUSE_INIT;
1149 	ia->args.in_numargs = 1;
1150 	ia->args.in_args[0].size = sizeof(ia->in);
1151 	ia->args.in_args[0].value = &ia->in;
1152 	ia->args.out_numargs = 1;
1153 	/* Variable length argument used for backward compatibility
1154 	   with interface version < 7.5.  Rest of init_out is zeroed
1155 	   by do_get_request(), so a short reply is not a problem */
1156 	ia->args.out_argvar = true;
1157 	ia->args.out_args[0].size = sizeof(ia->out);
1158 	ia->args.out_args[0].value = &ia->out;
1159 	ia->args.force = true;
1160 	ia->args.nocreds = true;
1161 	ia->args.end = process_init_reply;
1162 
1163 	if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1164 		process_init_reply(fm, &ia->args, -ENOTCONN);
1165 }
1166 EXPORT_SYMBOL_GPL(fuse_send_init);
1167 
1168 void fuse_free_conn(struct fuse_conn *fc)
1169 {
1170 	WARN_ON(!list_empty(&fc->devices));
1171 	kfree_rcu(fc, rcu);
1172 }
1173 EXPORT_SYMBOL_GPL(fuse_free_conn);
1174 
1175 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1176 {
1177 	int err;
1178 	char *suffix = "";
1179 
1180 	if (sb->s_bdev) {
1181 		suffix = "-fuseblk";
1182 		/*
1183 		 * sb->s_bdi points to blkdev's bdi however we want to redirect
1184 		 * it to our private bdi...
1185 		 */
1186 		bdi_put(sb->s_bdi);
1187 		sb->s_bdi = &noop_backing_dev_info;
1188 	}
1189 	err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1190 				   MINOR(fc->dev), suffix);
1191 	if (err)
1192 		return err;
1193 
1194 	/* fuse does it's own writeback accounting */
1195 	sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1196 	sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1197 
1198 	/*
1199 	 * For a single fuse filesystem use max 1% of dirty +
1200 	 * writeback threshold.
1201 	 *
1202 	 * This gives about 1M of write buffer for memory maps on a
1203 	 * machine with 1G and 10% dirty_ratio, which should be more
1204 	 * than enough.
1205 	 *
1206 	 * Privileged users can raise it by writing to
1207 	 *
1208 	 *    /sys/class/bdi/<bdi>/max_ratio
1209 	 */
1210 	bdi_set_max_ratio(sb->s_bdi, 1);
1211 
1212 	return 0;
1213 }
1214 
1215 struct fuse_dev *fuse_dev_alloc(void)
1216 {
1217 	struct fuse_dev *fud;
1218 	struct list_head *pq;
1219 
1220 	fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1221 	if (!fud)
1222 		return NULL;
1223 
1224 	pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1225 	if (!pq) {
1226 		kfree(fud);
1227 		return NULL;
1228 	}
1229 
1230 	fud->pq.processing = pq;
1231 	fuse_pqueue_init(&fud->pq);
1232 
1233 	return fud;
1234 }
1235 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1236 
1237 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1238 {
1239 	fud->fc = fuse_conn_get(fc);
1240 	spin_lock(&fc->lock);
1241 	list_add_tail(&fud->entry, &fc->devices);
1242 	spin_unlock(&fc->lock);
1243 }
1244 EXPORT_SYMBOL_GPL(fuse_dev_install);
1245 
1246 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1247 {
1248 	struct fuse_dev *fud;
1249 
1250 	fud = fuse_dev_alloc();
1251 	if (!fud)
1252 		return NULL;
1253 
1254 	fuse_dev_install(fud, fc);
1255 	return fud;
1256 }
1257 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1258 
1259 void fuse_dev_free(struct fuse_dev *fud)
1260 {
1261 	struct fuse_conn *fc = fud->fc;
1262 
1263 	if (fc) {
1264 		spin_lock(&fc->lock);
1265 		list_del(&fud->entry);
1266 		spin_unlock(&fc->lock);
1267 
1268 		fuse_conn_put(fc);
1269 	}
1270 	kfree(fud->pq.processing);
1271 	kfree(fud);
1272 }
1273 EXPORT_SYMBOL_GPL(fuse_dev_free);
1274 
1275 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1276 				      const struct fuse_inode *fi)
1277 {
1278 	*attr = (struct fuse_attr){
1279 		.ino		= fi->inode.i_ino,
1280 		.size		= fi->inode.i_size,
1281 		.blocks		= fi->inode.i_blocks,
1282 		.atime		= fi->inode.i_atime.tv_sec,
1283 		.mtime		= fi->inode.i_mtime.tv_sec,
1284 		.ctime		= fi->inode.i_ctime.tv_sec,
1285 		.atimensec	= fi->inode.i_atime.tv_nsec,
1286 		.mtimensec	= fi->inode.i_mtime.tv_nsec,
1287 		.ctimensec	= fi->inode.i_ctime.tv_nsec,
1288 		.mode		= fi->inode.i_mode,
1289 		.nlink		= fi->inode.i_nlink,
1290 		.uid		= fi->inode.i_uid.val,
1291 		.gid		= fi->inode.i_gid.val,
1292 		.rdev		= fi->inode.i_rdev,
1293 		.blksize	= 1u << fi->inode.i_blkbits,
1294 	};
1295 }
1296 
1297 static void fuse_sb_defaults(struct super_block *sb)
1298 {
1299 	sb->s_magic = FUSE_SUPER_MAGIC;
1300 	sb->s_op = &fuse_super_operations;
1301 	sb->s_xattr = fuse_xattr_handlers;
1302 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1303 	sb->s_time_gran = 1;
1304 	sb->s_export_op = &fuse_export_operations;
1305 	sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1306 	if (sb->s_user_ns != &init_user_ns)
1307 		sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1308 	sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1309 
1310 	/*
1311 	 * If we are not in the initial user namespace posix
1312 	 * acls must be translated.
1313 	 */
1314 	if (sb->s_user_ns != &init_user_ns)
1315 		sb->s_xattr = fuse_no_acl_xattr_handlers;
1316 }
1317 
1318 static int fuse_fill_super_submount(struct super_block *sb,
1319 				    struct fuse_inode *parent_fi)
1320 {
1321 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1322 	struct super_block *parent_sb = parent_fi->inode.i_sb;
1323 	struct fuse_attr root_attr;
1324 	struct inode *root;
1325 
1326 	fuse_sb_defaults(sb);
1327 	fm->sb = sb;
1328 
1329 	WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1330 	sb->s_bdi = bdi_get(parent_sb->s_bdi);
1331 
1332 	sb->s_xattr = parent_sb->s_xattr;
1333 	sb->s_time_gran = parent_sb->s_time_gran;
1334 	sb->s_blocksize = parent_sb->s_blocksize;
1335 	sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1336 	sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1337 	if (parent_sb->s_subtype && !sb->s_subtype)
1338 		return -ENOMEM;
1339 
1340 	fuse_fill_attr_from_inode(&root_attr, parent_fi);
1341 	root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1342 	/*
1343 	 * This inode is just a duplicate, so it is not looked up and
1344 	 * its nlookup should not be incremented.  fuse_iget() does
1345 	 * that, though, so undo it here.
1346 	 */
1347 	get_fuse_inode(root)->nlookup--;
1348 	sb->s_d_op = &fuse_dentry_operations;
1349 	sb->s_root = d_make_root(root);
1350 	if (!sb->s_root)
1351 		return -ENOMEM;
1352 
1353 	return 0;
1354 }
1355 
1356 /* Filesystem context private data holds the FUSE inode of the mount point */
1357 static int fuse_get_tree_submount(struct fs_context *fsc)
1358 {
1359 	struct fuse_mount *fm;
1360 	struct fuse_inode *mp_fi = fsc->fs_private;
1361 	struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1362 	struct super_block *sb;
1363 	int err;
1364 
1365 	fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1366 	if (!fm)
1367 		return -ENOMEM;
1368 
1369 	fsc->s_fs_info = fm;
1370 	sb = sget_fc(fsc, NULL, set_anon_super_fc);
1371 	if (IS_ERR(sb)) {
1372 		kfree(fm);
1373 		return PTR_ERR(sb);
1374 	}
1375 	fm->fc = fuse_conn_get(fc);
1376 
1377 	/* Initialize superblock, making @mp_fi its root */
1378 	err = fuse_fill_super_submount(sb, mp_fi);
1379 	if (err) {
1380 		fuse_conn_put(fc);
1381 		kfree(fm);
1382 		sb->s_fs_info = NULL;
1383 		deactivate_locked_super(sb);
1384 		return err;
1385 	}
1386 
1387 	down_write(&fc->killsb);
1388 	list_add_tail(&fm->fc_entry, &fc->mounts);
1389 	up_write(&fc->killsb);
1390 
1391 	sb->s_flags |= SB_ACTIVE;
1392 	fsc->root = dget(sb->s_root);
1393 
1394 	return 0;
1395 }
1396 
1397 static const struct fs_context_operations fuse_context_submount_ops = {
1398 	.get_tree	= fuse_get_tree_submount,
1399 };
1400 
1401 int fuse_init_fs_context_submount(struct fs_context *fsc)
1402 {
1403 	fsc->ops = &fuse_context_submount_ops;
1404 	return 0;
1405 }
1406 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1407 
1408 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1409 {
1410 	struct fuse_dev *fud = NULL;
1411 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1412 	struct fuse_conn *fc = fm->fc;
1413 	struct inode *root;
1414 	struct dentry *root_dentry;
1415 	int err;
1416 
1417 	err = -EINVAL;
1418 	if (sb->s_flags & SB_MANDLOCK)
1419 		goto err;
1420 
1421 	fuse_sb_defaults(sb);
1422 
1423 	if (ctx->is_bdev) {
1424 #ifdef CONFIG_BLOCK
1425 		err = -EINVAL;
1426 		if (!sb_set_blocksize(sb, ctx->blksize))
1427 			goto err;
1428 #endif
1429 	} else {
1430 		sb->s_blocksize = PAGE_SIZE;
1431 		sb->s_blocksize_bits = PAGE_SHIFT;
1432 	}
1433 
1434 	sb->s_subtype = ctx->subtype;
1435 	ctx->subtype = NULL;
1436 	if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1437 		err = fuse_dax_conn_alloc(fc, ctx->dax_dev);
1438 		if (err)
1439 			goto err;
1440 	}
1441 
1442 	if (ctx->fudptr) {
1443 		err = -ENOMEM;
1444 		fud = fuse_dev_alloc_install(fc);
1445 		if (!fud)
1446 			goto err_free_dax;
1447 	}
1448 
1449 	fc->dev = sb->s_dev;
1450 	fm->sb = sb;
1451 	err = fuse_bdi_init(fc, sb);
1452 	if (err)
1453 		goto err_dev_free;
1454 
1455 	/* Handle umasking inside the fuse code */
1456 	if (sb->s_flags & SB_POSIXACL)
1457 		fc->dont_mask = 1;
1458 	sb->s_flags |= SB_POSIXACL;
1459 
1460 	fc->default_permissions = ctx->default_permissions;
1461 	fc->allow_other = ctx->allow_other;
1462 	fc->user_id = ctx->user_id;
1463 	fc->group_id = ctx->group_id;
1464 	fc->legacy_opts_show = ctx->legacy_opts_show;
1465 	fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1466 	fc->destroy = ctx->destroy;
1467 	fc->no_control = ctx->no_control;
1468 	fc->no_force_umount = ctx->no_force_umount;
1469 
1470 	err = -ENOMEM;
1471 	root = fuse_get_root_inode(sb, ctx->rootmode);
1472 	sb->s_d_op = &fuse_root_dentry_operations;
1473 	root_dentry = d_make_root(root);
1474 	if (!root_dentry)
1475 		goto err_dev_free;
1476 	/* Root dentry doesn't have .d_revalidate */
1477 	sb->s_d_op = &fuse_dentry_operations;
1478 
1479 	mutex_lock(&fuse_mutex);
1480 	err = -EINVAL;
1481 	if (ctx->fudptr && *ctx->fudptr)
1482 		goto err_unlock;
1483 
1484 	err = fuse_ctl_add_conn(fc);
1485 	if (err)
1486 		goto err_unlock;
1487 
1488 	list_add_tail(&fc->entry, &fuse_conn_list);
1489 	sb->s_root = root_dentry;
1490 	if (ctx->fudptr)
1491 		*ctx->fudptr = fud;
1492 	mutex_unlock(&fuse_mutex);
1493 	return 0;
1494 
1495  err_unlock:
1496 	mutex_unlock(&fuse_mutex);
1497 	dput(root_dentry);
1498  err_dev_free:
1499 	if (fud)
1500 		fuse_dev_free(fud);
1501  err_free_dax:
1502 	if (IS_ENABLED(CONFIG_FUSE_DAX))
1503 		fuse_dax_conn_free(fc);
1504  err:
1505 	return err;
1506 }
1507 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1508 
1509 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1510 {
1511 	struct fuse_fs_context *ctx = fsc->fs_private;
1512 	struct file *file;
1513 	int err;
1514 	struct fuse_conn *fc;
1515 	struct fuse_mount *fm;
1516 
1517 	err = -EINVAL;
1518 	file = fget(ctx->fd);
1519 	if (!file)
1520 		goto err;
1521 
1522 	/*
1523 	 * Require mount to happen from the same user namespace which
1524 	 * opened /dev/fuse to prevent potential attacks.
1525 	 */
1526 	if ((file->f_op != &fuse_dev_operations) ||
1527 	    (file->f_cred->user_ns != sb->s_user_ns))
1528 		goto err_fput;
1529 	ctx->fudptr = &file->private_data;
1530 
1531 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1532 	err = -ENOMEM;
1533 	if (!fc)
1534 		goto err_fput;
1535 
1536 	fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1537 	if (!fm) {
1538 		kfree(fc);
1539 		goto err_fput;
1540 	}
1541 
1542 	fuse_conn_init(fc, fm, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1543 	fc->release = fuse_free_conn;
1544 
1545 	sb->s_fs_info = fm;
1546 
1547 	err = fuse_fill_super_common(sb, ctx);
1548 	if (err)
1549 		goto err_put_conn;
1550 	/*
1551 	 * atomic_dec_and_test() in fput() provides the necessary
1552 	 * memory barrier for file->private_data to be visible on all
1553 	 * CPUs after this
1554 	 */
1555 	fput(file);
1556 	fuse_send_init(get_fuse_mount_super(sb));
1557 	return 0;
1558 
1559  err_put_conn:
1560 	fuse_conn_put(fc);
1561 	kfree(fm);
1562 	sb->s_fs_info = NULL;
1563  err_fput:
1564 	fput(file);
1565  err:
1566 	return err;
1567 }
1568 
1569 static int fuse_get_tree(struct fs_context *fc)
1570 {
1571 	struct fuse_fs_context *ctx = fc->fs_private;
1572 
1573 	if (!ctx->fd_present || !ctx->rootmode_present ||
1574 	    !ctx->user_id_present || !ctx->group_id_present)
1575 		return -EINVAL;
1576 
1577 #ifdef CONFIG_BLOCK
1578 	if (ctx->is_bdev)
1579 		return get_tree_bdev(fc, fuse_fill_super);
1580 #endif
1581 
1582 	return get_tree_nodev(fc, fuse_fill_super);
1583 }
1584 
1585 static const struct fs_context_operations fuse_context_ops = {
1586 	.free		= fuse_free_fc,
1587 	.parse_param	= fuse_parse_param,
1588 	.reconfigure	= fuse_reconfigure,
1589 	.get_tree	= fuse_get_tree,
1590 };
1591 
1592 /*
1593  * Set up the filesystem mount context.
1594  */
1595 static int fuse_init_fs_context(struct fs_context *fc)
1596 {
1597 	struct fuse_fs_context *ctx;
1598 
1599 	ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1600 	if (!ctx)
1601 		return -ENOMEM;
1602 
1603 	ctx->max_read = ~0;
1604 	ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1605 	ctx->legacy_opts_show = true;
1606 
1607 #ifdef CONFIG_BLOCK
1608 	if (fc->fs_type == &fuseblk_fs_type) {
1609 		ctx->is_bdev = true;
1610 		ctx->destroy = true;
1611 	}
1612 #endif
1613 
1614 	fc->fs_private = ctx;
1615 	fc->ops = &fuse_context_ops;
1616 	return 0;
1617 }
1618 
1619 bool fuse_mount_remove(struct fuse_mount *fm)
1620 {
1621 	struct fuse_conn *fc = fm->fc;
1622 	bool last = false;
1623 
1624 	down_write(&fc->killsb);
1625 	list_del_init(&fm->fc_entry);
1626 	if (list_empty(&fc->mounts))
1627 		last = true;
1628 	up_write(&fc->killsb);
1629 
1630 	return last;
1631 }
1632 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1633 
1634 void fuse_conn_destroy(struct fuse_mount *fm)
1635 {
1636 	struct fuse_conn *fc = fm->fc;
1637 
1638 	if (fc->destroy)
1639 		fuse_send_destroy(fm);
1640 
1641 	fuse_abort_conn(fc);
1642 	fuse_wait_aborted(fc);
1643 
1644 	if (!list_empty(&fc->entry)) {
1645 		mutex_lock(&fuse_mutex);
1646 		list_del(&fc->entry);
1647 		fuse_ctl_remove_conn(fc);
1648 		mutex_unlock(&fuse_mutex);
1649 	}
1650 }
1651 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1652 
1653 static void fuse_sb_destroy(struct super_block *sb)
1654 {
1655 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1656 	bool last;
1657 
1658 	if (fm) {
1659 		last = fuse_mount_remove(fm);
1660 		if (last)
1661 			fuse_conn_destroy(fm);
1662 	}
1663 }
1664 
1665 static void fuse_kill_sb_anon(struct super_block *sb)
1666 {
1667 	fuse_sb_destroy(sb);
1668 	kill_anon_super(sb);
1669 }
1670 
1671 static struct file_system_type fuse_fs_type = {
1672 	.owner		= THIS_MODULE,
1673 	.name		= "fuse",
1674 	.fs_flags	= FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1675 	.init_fs_context = fuse_init_fs_context,
1676 	.parameters	= fuse_fs_parameters,
1677 	.kill_sb	= fuse_kill_sb_anon,
1678 };
1679 MODULE_ALIAS_FS("fuse");
1680 
1681 #ifdef CONFIG_BLOCK
1682 static void fuse_kill_sb_blk(struct super_block *sb)
1683 {
1684 	fuse_sb_destroy(sb);
1685 	kill_block_super(sb);
1686 }
1687 
1688 static struct file_system_type fuseblk_fs_type = {
1689 	.owner		= THIS_MODULE,
1690 	.name		= "fuseblk",
1691 	.init_fs_context = fuse_init_fs_context,
1692 	.parameters	= fuse_fs_parameters,
1693 	.kill_sb	= fuse_kill_sb_blk,
1694 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1695 };
1696 MODULE_ALIAS_FS("fuseblk");
1697 
1698 static inline int register_fuseblk(void)
1699 {
1700 	return register_filesystem(&fuseblk_fs_type);
1701 }
1702 
1703 static inline void unregister_fuseblk(void)
1704 {
1705 	unregister_filesystem(&fuseblk_fs_type);
1706 }
1707 #else
1708 static inline int register_fuseblk(void)
1709 {
1710 	return 0;
1711 }
1712 
1713 static inline void unregister_fuseblk(void)
1714 {
1715 }
1716 #endif
1717 
1718 static void fuse_inode_init_once(void *foo)
1719 {
1720 	struct inode *inode = foo;
1721 
1722 	inode_init_once(inode);
1723 }
1724 
1725 static int __init fuse_fs_init(void)
1726 {
1727 	int err;
1728 
1729 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
1730 			sizeof(struct fuse_inode), 0,
1731 			SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1732 			fuse_inode_init_once);
1733 	err = -ENOMEM;
1734 	if (!fuse_inode_cachep)
1735 		goto out;
1736 
1737 	err = register_fuseblk();
1738 	if (err)
1739 		goto out2;
1740 
1741 	err = register_filesystem(&fuse_fs_type);
1742 	if (err)
1743 		goto out3;
1744 
1745 	return 0;
1746 
1747  out3:
1748 	unregister_fuseblk();
1749  out2:
1750 	kmem_cache_destroy(fuse_inode_cachep);
1751  out:
1752 	return err;
1753 }
1754 
1755 static void fuse_fs_cleanup(void)
1756 {
1757 	unregister_filesystem(&fuse_fs_type);
1758 	unregister_fuseblk();
1759 
1760 	/*
1761 	 * Make sure all delayed rcu free inodes are flushed before we
1762 	 * destroy cache.
1763 	 */
1764 	rcu_barrier();
1765 	kmem_cache_destroy(fuse_inode_cachep);
1766 }
1767 
1768 static struct kobject *fuse_kobj;
1769 
1770 static int fuse_sysfs_init(void)
1771 {
1772 	int err;
1773 
1774 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1775 	if (!fuse_kobj) {
1776 		err = -ENOMEM;
1777 		goto out_err;
1778 	}
1779 
1780 	err = sysfs_create_mount_point(fuse_kobj, "connections");
1781 	if (err)
1782 		goto out_fuse_unregister;
1783 
1784 	return 0;
1785 
1786  out_fuse_unregister:
1787 	kobject_put(fuse_kobj);
1788  out_err:
1789 	return err;
1790 }
1791 
1792 static void fuse_sysfs_cleanup(void)
1793 {
1794 	sysfs_remove_mount_point(fuse_kobj, "connections");
1795 	kobject_put(fuse_kobj);
1796 }
1797 
1798 static int __init fuse_init(void)
1799 {
1800 	int res;
1801 
1802 	pr_info("init (API version %i.%i)\n",
1803 		FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1804 
1805 	INIT_LIST_HEAD(&fuse_conn_list);
1806 	res = fuse_fs_init();
1807 	if (res)
1808 		goto err;
1809 
1810 	res = fuse_dev_init();
1811 	if (res)
1812 		goto err_fs_cleanup;
1813 
1814 	res = fuse_sysfs_init();
1815 	if (res)
1816 		goto err_dev_cleanup;
1817 
1818 	res = fuse_ctl_init();
1819 	if (res)
1820 		goto err_sysfs_cleanup;
1821 
1822 	sanitize_global_limit(&max_user_bgreq);
1823 	sanitize_global_limit(&max_user_congthresh);
1824 
1825 	return 0;
1826 
1827  err_sysfs_cleanup:
1828 	fuse_sysfs_cleanup();
1829  err_dev_cleanup:
1830 	fuse_dev_cleanup();
1831  err_fs_cleanup:
1832 	fuse_fs_cleanup();
1833  err:
1834 	return res;
1835 }
1836 
1837 static void __exit fuse_exit(void)
1838 {
1839 	pr_debug("exit\n");
1840 
1841 	fuse_ctl_cleanup();
1842 	fuse_sysfs_cleanup();
1843 	fuse_fs_cleanup();
1844 	fuse_dev_cleanup();
1845 }
1846 
1847 module_init(fuse_init);
1848 module_exit(fuse_exit);
1849