xref: /linux/fs/fuse/inode.c (revision 73b0140bf0fe9df90fb267c00673c4b9bf285430)
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/parser.h>
19 #include <linux/statfs.h>
20 #include <linux/random.h>
21 #include <linux/sched.h>
22 #include <linux/exportfs.h>
23 #include <linux/posix_acl.h>
24 #include <linux/pid_namespace.h>
25 
26 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
27 MODULE_DESCRIPTION("Filesystem in Userspace");
28 MODULE_LICENSE("GPL");
29 
30 static struct kmem_cache *fuse_inode_cachep;
31 struct list_head fuse_conn_list;
32 DEFINE_MUTEX(fuse_mutex);
33 
34 static int set_global_limit(const char *val, const struct kernel_param *kp);
35 
36 unsigned max_user_bgreq;
37 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
38 		  &max_user_bgreq, 0644);
39 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
40 MODULE_PARM_DESC(max_user_bgreq,
41  "Global limit for the maximum number of backgrounded requests an "
42  "unprivileged user can set");
43 
44 unsigned max_user_congthresh;
45 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
46 		  &max_user_congthresh, 0644);
47 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
48 MODULE_PARM_DESC(max_user_congthresh,
49  "Global limit for the maximum congestion threshold an "
50  "unprivileged user can set");
51 
52 #define FUSE_SUPER_MAGIC 0x65735546
53 
54 #define FUSE_DEFAULT_BLKSIZE 512
55 
56 /** Maximum number of outstanding background requests */
57 #define FUSE_DEFAULT_MAX_BACKGROUND 12
58 
59 /** Congestion starts at 75% of maximum */
60 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
61 
62 struct fuse_mount_data {
63 	int fd;
64 	unsigned rootmode;
65 	kuid_t user_id;
66 	kgid_t group_id;
67 	unsigned fd_present:1;
68 	unsigned rootmode_present:1;
69 	unsigned user_id_present:1;
70 	unsigned group_id_present:1;
71 	unsigned default_permissions:1;
72 	unsigned allow_other:1;
73 	unsigned max_read;
74 	unsigned blksize;
75 };
76 
77 struct fuse_forget_link *fuse_alloc_forget(void)
78 {
79 	return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
80 }
81 
82 static struct inode *fuse_alloc_inode(struct super_block *sb)
83 {
84 	struct inode *inode;
85 	struct fuse_inode *fi;
86 
87 	inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
88 	if (!inode)
89 		return NULL;
90 
91 	fi = get_fuse_inode(inode);
92 	fi->i_time = 0;
93 	fi->inval_mask = 0;
94 	fi->nodeid = 0;
95 	fi->nlookup = 0;
96 	fi->attr_version = 0;
97 	fi->orig_ino = 0;
98 	fi->state = 0;
99 	mutex_init(&fi->mutex);
100 	spin_lock_init(&fi->lock);
101 	fi->forget = fuse_alloc_forget();
102 	if (!fi->forget) {
103 		kmem_cache_free(fuse_inode_cachep, inode);
104 		return NULL;
105 	}
106 
107 	return inode;
108 }
109 
110 static void fuse_free_inode(struct inode *inode)
111 {
112 	struct fuse_inode *fi = get_fuse_inode(inode);
113 
114 	mutex_destroy(&fi->mutex);
115 	kfree(fi->forget);
116 	kmem_cache_free(fuse_inode_cachep, fi);
117 }
118 
119 static void fuse_evict_inode(struct inode *inode)
120 {
121 	struct fuse_inode *fi = get_fuse_inode(inode);
122 
123 	truncate_inode_pages_final(&inode->i_data);
124 	clear_inode(inode);
125 	if (inode->i_sb->s_flags & SB_ACTIVE) {
126 		struct fuse_conn *fc = get_fuse_conn(inode);
127 		fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
128 		fi->forget = NULL;
129 	}
130 	if (S_ISREG(inode->i_mode) && !is_bad_inode(inode)) {
131 		WARN_ON(!list_empty(&fi->write_files));
132 		WARN_ON(!list_empty(&fi->queued_writes));
133 	}
134 }
135 
136 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
137 {
138 	sync_filesystem(sb);
139 	if (*flags & SB_MANDLOCK)
140 		return -EINVAL;
141 
142 	return 0;
143 }
144 
145 /*
146  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
147  * so that it will fit.
148  */
149 static ino_t fuse_squash_ino(u64 ino64)
150 {
151 	ino_t ino = (ino_t) ino64;
152 	if (sizeof(ino_t) < sizeof(u64))
153 		ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
154 	return ino;
155 }
156 
157 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
158 				   u64 attr_valid)
159 {
160 	struct fuse_conn *fc = get_fuse_conn(inode);
161 	struct fuse_inode *fi = get_fuse_inode(inode);
162 
163 	lockdep_assert_held(&fi->lock);
164 
165 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
166 	fi->i_time = attr_valid;
167 	WRITE_ONCE(fi->inval_mask, 0);
168 
169 	inode->i_ino     = fuse_squash_ino(attr->ino);
170 	inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
171 	set_nlink(inode, attr->nlink);
172 	inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
173 	inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
174 	inode->i_blocks  = attr->blocks;
175 	inode->i_atime.tv_sec   = attr->atime;
176 	inode->i_atime.tv_nsec  = attr->atimensec;
177 	/* mtime from server may be stale due to local buffered write */
178 	if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
179 		inode->i_mtime.tv_sec   = attr->mtime;
180 		inode->i_mtime.tv_nsec  = attr->mtimensec;
181 		inode->i_ctime.tv_sec   = attr->ctime;
182 		inode->i_ctime.tv_nsec  = attr->ctimensec;
183 	}
184 
185 	if (attr->blksize != 0)
186 		inode->i_blkbits = ilog2(attr->blksize);
187 	else
188 		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
189 
190 	/*
191 	 * Don't set the sticky bit in i_mode, unless we want the VFS
192 	 * to check permissions.  This prevents failures due to the
193 	 * check in may_delete().
194 	 */
195 	fi->orig_i_mode = inode->i_mode;
196 	if (!fc->default_permissions)
197 		inode->i_mode &= ~S_ISVTX;
198 
199 	fi->orig_ino = attr->ino;
200 }
201 
202 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
203 			    u64 attr_valid, u64 attr_version)
204 {
205 	struct fuse_conn *fc = get_fuse_conn(inode);
206 	struct fuse_inode *fi = get_fuse_inode(inode);
207 	bool is_wb = fc->writeback_cache;
208 	loff_t oldsize;
209 	struct timespec64 old_mtime;
210 
211 	spin_lock(&fi->lock);
212 	if ((attr_version != 0 && fi->attr_version > attr_version) ||
213 	    test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
214 		spin_unlock(&fi->lock);
215 		return;
216 	}
217 
218 	old_mtime = inode->i_mtime;
219 	fuse_change_attributes_common(inode, attr, attr_valid);
220 
221 	oldsize = inode->i_size;
222 	/*
223 	 * In case of writeback_cache enabled, the cached writes beyond EOF
224 	 * extend local i_size without keeping userspace server in sync. So,
225 	 * attr->size coming from server can be stale. We cannot trust it.
226 	 */
227 	if (!is_wb || !S_ISREG(inode->i_mode))
228 		i_size_write(inode, attr->size);
229 	spin_unlock(&fi->lock);
230 
231 	if (!is_wb && S_ISREG(inode->i_mode)) {
232 		bool inval = false;
233 
234 		if (oldsize != attr->size) {
235 			truncate_pagecache(inode, attr->size);
236 			inval = true;
237 		} else if (fc->auto_inval_data) {
238 			struct timespec64 new_mtime = {
239 				.tv_sec = attr->mtime,
240 				.tv_nsec = attr->mtimensec,
241 			};
242 
243 			/*
244 			 * Auto inval mode also checks and invalidates if mtime
245 			 * has changed.
246 			 */
247 			if (!timespec64_equal(&old_mtime, &new_mtime))
248 				inval = true;
249 		}
250 
251 		if (inval)
252 			invalidate_inode_pages2(inode->i_mapping);
253 	}
254 }
255 
256 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
257 {
258 	inode->i_mode = attr->mode & S_IFMT;
259 	inode->i_size = attr->size;
260 	inode->i_mtime.tv_sec  = attr->mtime;
261 	inode->i_mtime.tv_nsec = attr->mtimensec;
262 	inode->i_ctime.tv_sec  = attr->ctime;
263 	inode->i_ctime.tv_nsec = attr->ctimensec;
264 	if (S_ISREG(inode->i_mode)) {
265 		fuse_init_common(inode);
266 		fuse_init_file_inode(inode);
267 	} else if (S_ISDIR(inode->i_mode))
268 		fuse_init_dir(inode);
269 	else if (S_ISLNK(inode->i_mode))
270 		fuse_init_symlink(inode);
271 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
272 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
273 		fuse_init_common(inode);
274 		init_special_inode(inode, inode->i_mode,
275 				   new_decode_dev(attr->rdev));
276 	} else
277 		BUG();
278 }
279 
280 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
281 {
282 	u64 nodeid = *(u64 *) _nodeidp;
283 	if (get_node_id(inode) == nodeid)
284 		return 1;
285 	else
286 		return 0;
287 }
288 
289 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
290 {
291 	u64 nodeid = *(u64 *) _nodeidp;
292 	get_fuse_inode(inode)->nodeid = nodeid;
293 	return 0;
294 }
295 
296 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
297 			int generation, struct fuse_attr *attr,
298 			u64 attr_valid, u64 attr_version)
299 {
300 	struct inode *inode;
301 	struct fuse_inode *fi;
302 	struct fuse_conn *fc = get_fuse_conn_super(sb);
303 
304  retry:
305 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
306 	if (!inode)
307 		return NULL;
308 
309 	if ((inode->i_state & I_NEW)) {
310 		inode->i_flags |= S_NOATIME;
311 		if (!fc->writeback_cache || !S_ISREG(attr->mode))
312 			inode->i_flags |= S_NOCMTIME;
313 		inode->i_generation = generation;
314 		fuse_init_inode(inode, attr);
315 		unlock_new_inode(inode);
316 	} else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
317 		/* Inode has changed type, any I/O on the old should fail */
318 		make_bad_inode(inode);
319 		iput(inode);
320 		goto retry;
321 	}
322 
323 	fi = get_fuse_inode(inode);
324 	spin_lock(&fi->lock);
325 	fi->nlookup++;
326 	spin_unlock(&fi->lock);
327 	fuse_change_attributes(inode, attr, attr_valid, attr_version);
328 
329 	return inode;
330 }
331 
332 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
333 			     loff_t offset, loff_t len)
334 {
335 	struct inode *inode;
336 	pgoff_t pg_start;
337 	pgoff_t pg_end;
338 
339 	inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
340 	if (!inode)
341 		return -ENOENT;
342 
343 	fuse_invalidate_attr(inode);
344 	forget_all_cached_acls(inode);
345 	if (offset >= 0) {
346 		pg_start = offset >> PAGE_SHIFT;
347 		if (len <= 0)
348 			pg_end = -1;
349 		else
350 			pg_end = (offset + len - 1) >> PAGE_SHIFT;
351 		invalidate_inode_pages2_range(inode->i_mapping,
352 					      pg_start, pg_end);
353 	}
354 	iput(inode);
355 	return 0;
356 }
357 
358 bool fuse_lock_inode(struct inode *inode)
359 {
360 	bool locked = false;
361 
362 	if (!get_fuse_conn(inode)->parallel_dirops) {
363 		mutex_lock(&get_fuse_inode(inode)->mutex);
364 		locked = true;
365 	}
366 
367 	return locked;
368 }
369 
370 void fuse_unlock_inode(struct inode *inode, bool locked)
371 {
372 	if (locked)
373 		mutex_unlock(&get_fuse_inode(inode)->mutex);
374 }
375 
376 static void fuse_umount_begin(struct super_block *sb)
377 {
378 	fuse_abort_conn(get_fuse_conn_super(sb));
379 }
380 
381 static void fuse_send_destroy(struct fuse_conn *fc)
382 {
383 	struct fuse_req *req = fc->destroy_req;
384 	if (req && fc->conn_init) {
385 		fc->destroy_req = NULL;
386 		req->in.h.opcode = FUSE_DESTROY;
387 		__set_bit(FR_FORCE, &req->flags);
388 		__clear_bit(FR_BACKGROUND, &req->flags);
389 		fuse_request_send(fc, req);
390 		fuse_put_request(fc, req);
391 	}
392 }
393 
394 static void fuse_put_super(struct super_block *sb)
395 {
396 	struct fuse_conn *fc = get_fuse_conn_super(sb);
397 
398 	mutex_lock(&fuse_mutex);
399 	list_del(&fc->entry);
400 	fuse_ctl_remove_conn(fc);
401 	mutex_unlock(&fuse_mutex);
402 
403 	fuse_conn_put(fc);
404 }
405 
406 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
407 {
408 	stbuf->f_type    = FUSE_SUPER_MAGIC;
409 	stbuf->f_bsize   = attr->bsize;
410 	stbuf->f_frsize  = attr->frsize;
411 	stbuf->f_blocks  = attr->blocks;
412 	stbuf->f_bfree   = attr->bfree;
413 	stbuf->f_bavail  = attr->bavail;
414 	stbuf->f_files   = attr->files;
415 	stbuf->f_ffree   = attr->ffree;
416 	stbuf->f_namelen = attr->namelen;
417 	/* fsid is left zero */
418 }
419 
420 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
421 {
422 	struct super_block *sb = dentry->d_sb;
423 	struct fuse_conn *fc = get_fuse_conn_super(sb);
424 	FUSE_ARGS(args);
425 	struct fuse_statfs_out outarg;
426 	int err;
427 
428 	if (!fuse_allow_current_process(fc)) {
429 		buf->f_type = FUSE_SUPER_MAGIC;
430 		return 0;
431 	}
432 
433 	memset(&outarg, 0, sizeof(outarg));
434 	args.in.numargs = 0;
435 	args.in.h.opcode = FUSE_STATFS;
436 	args.in.h.nodeid = get_node_id(d_inode(dentry));
437 	args.out.numargs = 1;
438 	args.out.args[0].size = sizeof(outarg);
439 	args.out.args[0].value = &outarg;
440 	err = fuse_simple_request(fc, &args);
441 	if (!err)
442 		convert_fuse_statfs(buf, &outarg.st);
443 	return err;
444 }
445 
446 enum {
447 	OPT_FD,
448 	OPT_ROOTMODE,
449 	OPT_USER_ID,
450 	OPT_GROUP_ID,
451 	OPT_DEFAULT_PERMISSIONS,
452 	OPT_ALLOW_OTHER,
453 	OPT_MAX_READ,
454 	OPT_BLKSIZE,
455 	OPT_ERR
456 };
457 
458 static const match_table_t tokens = {
459 	{OPT_FD,			"fd=%u"},
460 	{OPT_ROOTMODE,			"rootmode=%o"},
461 	{OPT_USER_ID,			"user_id=%u"},
462 	{OPT_GROUP_ID,			"group_id=%u"},
463 	{OPT_DEFAULT_PERMISSIONS,	"default_permissions"},
464 	{OPT_ALLOW_OTHER,		"allow_other"},
465 	{OPT_MAX_READ,			"max_read=%u"},
466 	{OPT_BLKSIZE,			"blksize=%u"},
467 	{OPT_ERR,			NULL}
468 };
469 
470 static int fuse_match_uint(substring_t *s, unsigned int *res)
471 {
472 	int err = -ENOMEM;
473 	char *buf = match_strdup(s);
474 	if (buf) {
475 		err = kstrtouint(buf, 10, res);
476 		kfree(buf);
477 	}
478 	return err;
479 }
480 
481 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev,
482 			  struct user_namespace *user_ns)
483 {
484 	char *p;
485 	memset(d, 0, sizeof(struct fuse_mount_data));
486 	d->max_read = ~0;
487 	d->blksize = FUSE_DEFAULT_BLKSIZE;
488 
489 	while ((p = strsep(&opt, ",")) != NULL) {
490 		int token;
491 		int value;
492 		unsigned uv;
493 		substring_t args[MAX_OPT_ARGS];
494 		if (!*p)
495 			continue;
496 
497 		token = match_token(p, tokens, args);
498 		switch (token) {
499 		case OPT_FD:
500 			if (match_int(&args[0], &value))
501 				return 0;
502 			d->fd = value;
503 			d->fd_present = 1;
504 			break;
505 
506 		case OPT_ROOTMODE:
507 			if (match_octal(&args[0], &value))
508 				return 0;
509 			if (!fuse_valid_type(value))
510 				return 0;
511 			d->rootmode = value;
512 			d->rootmode_present = 1;
513 			break;
514 
515 		case OPT_USER_ID:
516 			if (fuse_match_uint(&args[0], &uv))
517 				return 0;
518 			d->user_id = make_kuid(user_ns, uv);
519 			if (!uid_valid(d->user_id))
520 				return 0;
521 			d->user_id_present = 1;
522 			break;
523 
524 		case OPT_GROUP_ID:
525 			if (fuse_match_uint(&args[0], &uv))
526 				return 0;
527 			d->group_id = make_kgid(user_ns, uv);
528 			if (!gid_valid(d->group_id))
529 				return 0;
530 			d->group_id_present = 1;
531 			break;
532 
533 		case OPT_DEFAULT_PERMISSIONS:
534 			d->default_permissions = 1;
535 			break;
536 
537 		case OPT_ALLOW_OTHER:
538 			d->allow_other = 1;
539 			break;
540 
541 		case OPT_MAX_READ:
542 			if (match_int(&args[0], &value))
543 				return 0;
544 			d->max_read = value;
545 			break;
546 
547 		case OPT_BLKSIZE:
548 			if (!is_bdev || match_int(&args[0], &value))
549 				return 0;
550 			d->blksize = value;
551 			break;
552 
553 		default:
554 			return 0;
555 		}
556 	}
557 
558 	if (!d->fd_present || !d->rootmode_present ||
559 	    !d->user_id_present || !d->group_id_present)
560 		return 0;
561 
562 	return 1;
563 }
564 
565 static int fuse_show_options(struct seq_file *m, struct dentry *root)
566 {
567 	struct super_block *sb = root->d_sb;
568 	struct fuse_conn *fc = get_fuse_conn_super(sb);
569 
570 	seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
571 	seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
572 	if (fc->default_permissions)
573 		seq_puts(m, ",default_permissions");
574 	if (fc->allow_other)
575 		seq_puts(m, ",allow_other");
576 	if (fc->max_read != ~0)
577 		seq_printf(m, ",max_read=%u", fc->max_read);
578 	if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
579 		seq_printf(m, ",blksize=%lu", sb->s_blocksize);
580 	return 0;
581 }
582 
583 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
584 {
585 	memset(fiq, 0, sizeof(struct fuse_iqueue));
586 	init_waitqueue_head(&fiq->waitq);
587 	INIT_LIST_HEAD(&fiq->pending);
588 	INIT_LIST_HEAD(&fiq->interrupts);
589 	fiq->forget_list_tail = &fiq->forget_list_head;
590 	fiq->connected = 1;
591 }
592 
593 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
594 {
595 	unsigned int i;
596 
597 	spin_lock_init(&fpq->lock);
598 	for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
599 		INIT_LIST_HEAD(&fpq->processing[i]);
600 	INIT_LIST_HEAD(&fpq->io);
601 	fpq->connected = 1;
602 }
603 
604 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns)
605 {
606 	memset(fc, 0, sizeof(*fc));
607 	spin_lock_init(&fc->lock);
608 	spin_lock_init(&fc->bg_lock);
609 	init_rwsem(&fc->killsb);
610 	refcount_set(&fc->count, 1);
611 	atomic_set(&fc->dev_count, 1);
612 	init_waitqueue_head(&fc->blocked_waitq);
613 	init_waitqueue_head(&fc->reserved_req_waitq);
614 	fuse_iqueue_init(&fc->iq);
615 	INIT_LIST_HEAD(&fc->bg_queue);
616 	INIT_LIST_HEAD(&fc->entry);
617 	INIT_LIST_HEAD(&fc->devices);
618 	atomic_set(&fc->num_waiting, 0);
619 	fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
620 	fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
621 	atomic64_set(&fc->khctr, 0);
622 	fc->polled_files = RB_ROOT;
623 	fc->blocked = 0;
624 	fc->initialized = 0;
625 	fc->connected = 1;
626 	atomic64_set(&fc->attr_version, 1);
627 	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
628 	fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
629 	fc->user_ns = get_user_ns(user_ns);
630 	fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
631 }
632 EXPORT_SYMBOL_GPL(fuse_conn_init);
633 
634 void fuse_conn_put(struct fuse_conn *fc)
635 {
636 	if (refcount_dec_and_test(&fc->count)) {
637 		if (fc->destroy_req)
638 			fuse_request_free(fc->destroy_req);
639 		put_pid_ns(fc->pid_ns);
640 		put_user_ns(fc->user_ns);
641 		fc->release(fc);
642 	}
643 }
644 EXPORT_SYMBOL_GPL(fuse_conn_put);
645 
646 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
647 {
648 	refcount_inc(&fc->count);
649 	return fc;
650 }
651 EXPORT_SYMBOL_GPL(fuse_conn_get);
652 
653 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
654 {
655 	struct fuse_attr attr;
656 	memset(&attr, 0, sizeof(attr));
657 
658 	attr.mode = mode;
659 	attr.ino = FUSE_ROOT_ID;
660 	attr.nlink = 1;
661 	return fuse_iget(sb, 1, 0, &attr, 0, 0);
662 }
663 
664 struct fuse_inode_handle {
665 	u64 nodeid;
666 	u32 generation;
667 };
668 
669 static struct dentry *fuse_get_dentry(struct super_block *sb,
670 				      struct fuse_inode_handle *handle)
671 {
672 	struct fuse_conn *fc = get_fuse_conn_super(sb);
673 	struct inode *inode;
674 	struct dentry *entry;
675 	int err = -ESTALE;
676 
677 	if (handle->nodeid == 0)
678 		goto out_err;
679 
680 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
681 	if (!inode) {
682 		struct fuse_entry_out outarg;
683 		const struct qstr name = QSTR_INIT(".", 1);
684 
685 		if (!fc->export_support)
686 			goto out_err;
687 
688 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
689 				       &inode);
690 		if (err && err != -ENOENT)
691 			goto out_err;
692 		if (err || !inode) {
693 			err = -ESTALE;
694 			goto out_err;
695 		}
696 		err = -EIO;
697 		if (get_node_id(inode) != handle->nodeid)
698 			goto out_iput;
699 	}
700 	err = -ESTALE;
701 	if (inode->i_generation != handle->generation)
702 		goto out_iput;
703 
704 	entry = d_obtain_alias(inode);
705 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
706 		fuse_invalidate_entry_cache(entry);
707 
708 	return entry;
709 
710  out_iput:
711 	iput(inode);
712  out_err:
713 	return ERR_PTR(err);
714 }
715 
716 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
717 			   struct inode *parent)
718 {
719 	int len = parent ? 6 : 3;
720 	u64 nodeid;
721 	u32 generation;
722 
723 	if (*max_len < len) {
724 		*max_len = len;
725 		return  FILEID_INVALID;
726 	}
727 
728 	nodeid = get_fuse_inode(inode)->nodeid;
729 	generation = inode->i_generation;
730 
731 	fh[0] = (u32)(nodeid >> 32);
732 	fh[1] = (u32)(nodeid & 0xffffffff);
733 	fh[2] = generation;
734 
735 	if (parent) {
736 		nodeid = get_fuse_inode(parent)->nodeid;
737 		generation = parent->i_generation;
738 
739 		fh[3] = (u32)(nodeid >> 32);
740 		fh[4] = (u32)(nodeid & 0xffffffff);
741 		fh[5] = generation;
742 	}
743 
744 	*max_len = len;
745 	return parent ? 0x82 : 0x81;
746 }
747 
748 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
749 		struct fid *fid, int fh_len, int fh_type)
750 {
751 	struct fuse_inode_handle handle;
752 
753 	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
754 		return NULL;
755 
756 	handle.nodeid = (u64) fid->raw[0] << 32;
757 	handle.nodeid |= (u64) fid->raw[1];
758 	handle.generation = fid->raw[2];
759 	return fuse_get_dentry(sb, &handle);
760 }
761 
762 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
763 		struct fid *fid, int fh_len, int fh_type)
764 {
765 	struct fuse_inode_handle parent;
766 
767 	if (fh_type != 0x82 || fh_len < 6)
768 		return NULL;
769 
770 	parent.nodeid = (u64) fid->raw[3] << 32;
771 	parent.nodeid |= (u64) fid->raw[4];
772 	parent.generation = fid->raw[5];
773 	return fuse_get_dentry(sb, &parent);
774 }
775 
776 static struct dentry *fuse_get_parent(struct dentry *child)
777 {
778 	struct inode *child_inode = d_inode(child);
779 	struct fuse_conn *fc = get_fuse_conn(child_inode);
780 	struct inode *inode;
781 	struct dentry *parent;
782 	struct fuse_entry_out outarg;
783 	const struct qstr name = QSTR_INIT("..", 2);
784 	int err;
785 
786 	if (!fc->export_support)
787 		return ERR_PTR(-ESTALE);
788 
789 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
790 			       &name, &outarg, &inode);
791 	if (err) {
792 		if (err == -ENOENT)
793 			return ERR_PTR(-ESTALE);
794 		return ERR_PTR(err);
795 	}
796 
797 	parent = d_obtain_alias(inode);
798 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
799 		fuse_invalidate_entry_cache(parent);
800 
801 	return parent;
802 }
803 
804 static const struct export_operations fuse_export_operations = {
805 	.fh_to_dentry	= fuse_fh_to_dentry,
806 	.fh_to_parent	= fuse_fh_to_parent,
807 	.encode_fh	= fuse_encode_fh,
808 	.get_parent	= fuse_get_parent,
809 };
810 
811 static const struct super_operations fuse_super_operations = {
812 	.alloc_inode    = fuse_alloc_inode,
813 	.free_inode     = fuse_free_inode,
814 	.evict_inode	= fuse_evict_inode,
815 	.write_inode	= fuse_write_inode,
816 	.drop_inode	= generic_delete_inode,
817 	.remount_fs	= fuse_remount_fs,
818 	.put_super	= fuse_put_super,
819 	.umount_begin	= fuse_umount_begin,
820 	.statfs		= fuse_statfs,
821 	.show_options	= fuse_show_options,
822 };
823 
824 static void sanitize_global_limit(unsigned *limit)
825 {
826 	if (*limit == 0)
827 		*limit = ((totalram_pages() << PAGE_SHIFT) >> 13) /
828 			 sizeof(struct fuse_req);
829 
830 	if (*limit >= 1 << 16)
831 		*limit = (1 << 16) - 1;
832 }
833 
834 static int set_global_limit(const char *val, const struct kernel_param *kp)
835 {
836 	int rv;
837 
838 	rv = param_set_uint(val, kp);
839 	if (rv)
840 		return rv;
841 
842 	sanitize_global_limit((unsigned *)kp->arg);
843 
844 	return 0;
845 }
846 
847 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
848 {
849 	int cap_sys_admin = capable(CAP_SYS_ADMIN);
850 
851 	if (arg->minor < 13)
852 		return;
853 
854 	sanitize_global_limit(&max_user_bgreq);
855 	sanitize_global_limit(&max_user_congthresh);
856 
857 	spin_lock(&fc->bg_lock);
858 	if (arg->max_background) {
859 		fc->max_background = arg->max_background;
860 
861 		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
862 			fc->max_background = max_user_bgreq;
863 	}
864 	if (arg->congestion_threshold) {
865 		fc->congestion_threshold = arg->congestion_threshold;
866 
867 		if (!cap_sys_admin &&
868 		    fc->congestion_threshold > max_user_congthresh)
869 			fc->congestion_threshold = max_user_congthresh;
870 	}
871 	spin_unlock(&fc->bg_lock);
872 }
873 
874 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
875 {
876 	struct fuse_init_out *arg = &req->misc.init_out;
877 
878 	if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
879 		fc->conn_error = 1;
880 	else {
881 		unsigned long ra_pages;
882 
883 		process_init_limits(fc, arg);
884 
885 		if (arg->minor >= 6) {
886 			ra_pages = arg->max_readahead / PAGE_SIZE;
887 			if (arg->flags & FUSE_ASYNC_READ)
888 				fc->async_read = 1;
889 			if (!(arg->flags & FUSE_POSIX_LOCKS))
890 				fc->no_lock = 1;
891 			if (arg->minor >= 17) {
892 				if (!(arg->flags & FUSE_FLOCK_LOCKS))
893 					fc->no_flock = 1;
894 			} else {
895 				if (!(arg->flags & FUSE_POSIX_LOCKS))
896 					fc->no_flock = 1;
897 			}
898 			if (arg->flags & FUSE_ATOMIC_O_TRUNC)
899 				fc->atomic_o_trunc = 1;
900 			if (arg->minor >= 9) {
901 				/* LOOKUP has dependency on proto version */
902 				if (arg->flags & FUSE_EXPORT_SUPPORT)
903 					fc->export_support = 1;
904 			}
905 			if (arg->flags & FUSE_BIG_WRITES)
906 				fc->big_writes = 1;
907 			if (arg->flags & FUSE_DONT_MASK)
908 				fc->dont_mask = 1;
909 			if (arg->flags & FUSE_AUTO_INVAL_DATA)
910 				fc->auto_inval_data = 1;
911 			if (arg->flags & FUSE_DO_READDIRPLUS) {
912 				fc->do_readdirplus = 1;
913 				if (arg->flags & FUSE_READDIRPLUS_AUTO)
914 					fc->readdirplus_auto = 1;
915 			}
916 			if (arg->flags & FUSE_ASYNC_DIO)
917 				fc->async_dio = 1;
918 			if (arg->flags & FUSE_WRITEBACK_CACHE)
919 				fc->writeback_cache = 1;
920 			if (arg->flags & FUSE_PARALLEL_DIROPS)
921 				fc->parallel_dirops = 1;
922 			if (arg->flags & FUSE_HANDLE_KILLPRIV)
923 				fc->handle_killpriv = 1;
924 			if (arg->time_gran && arg->time_gran <= 1000000000)
925 				fc->sb->s_time_gran = arg->time_gran;
926 			if ((arg->flags & FUSE_POSIX_ACL)) {
927 				fc->default_permissions = 1;
928 				fc->posix_acl = 1;
929 				fc->sb->s_xattr = fuse_acl_xattr_handlers;
930 			}
931 			if (arg->flags & FUSE_CACHE_SYMLINKS)
932 				fc->cache_symlinks = 1;
933 			if (arg->flags & FUSE_ABORT_ERROR)
934 				fc->abort_err = 1;
935 			if (arg->flags & FUSE_MAX_PAGES) {
936 				fc->max_pages =
937 					min_t(unsigned int, FUSE_MAX_MAX_PAGES,
938 					max_t(unsigned int, arg->max_pages, 1));
939 			}
940 		} else {
941 			ra_pages = fc->max_read / PAGE_SIZE;
942 			fc->no_lock = 1;
943 			fc->no_flock = 1;
944 		}
945 
946 		fc->sb->s_bdi->ra_pages =
947 				min(fc->sb->s_bdi->ra_pages, ra_pages);
948 		fc->minor = arg->minor;
949 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
950 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
951 		fc->conn_init = 1;
952 	}
953 	fuse_set_initialized(fc);
954 	wake_up_all(&fc->blocked_waitq);
955 }
956 
957 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
958 {
959 	struct fuse_init_in *arg = &req->misc.init_in;
960 
961 	arg->major = FUSE_KERNEL_VERSION;
962 	arg->minor = FUSE_KERNEL_MINOR_VERSION;
963 	arg->max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
964 	arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
965 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
966 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
967 		FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
968 		FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
969 		FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
970 		FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
971 		FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
972 		FUSE_NO_OPENDIR_SUPPORT;
973 	req->in.h.opcode = FUSE_INIT;
974 	req->in.numargs = 1;
975 	req->in.args[0].size = sizeof(*arg);
976 	req->in.args[0].value = arg;
977 	req->out.numargs = 1;
978 	/* Variable length argument used for backward compatibility
979 	   with interface version < 7.5.  Rest of init_out is zeroed
980 	   by do_get_request(), so a short reply is not a problem */
981 	req->out.argvar = 1;
982 	req->out.args[0].size = sizeof(struct fuse_init_out);
983 	req->out.args[0].value = &req->misc.init_out;
984 	req->end = process_init_reply;
985 	fuse_request_send_background(fc, req);
986 }
987 
988 static void fuse_free_conn(struct fuse_conn *fc)
989 {
990 	WARN_ON(!list_empty(&fc->devices));
991 	kfree_rcu(fc, rcu);
992 }
993 
994 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
995 {
996 	int err;
997 	char *suffix = "";
998 
999 	if (sb->s_bdev) {
1000 		suffix = "-fuseblk";
1001 		/*
1002 		 * sb->s_bdi points to blkdev's bdi however we want to redirect
1003 		 * it to our private bdi...
1004 		 */
1005 		bdi_put(sb->s_bdi);
1006 		sb->s_bdi = &noop_backing_dev_info;
1007 	}
1008 	err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1009 				   MINOR(fc->dev), suffix);
1010 	if (err)
1011 		return err;
1012 
1013 	sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1014 	/* fuse does it's own writeback accounting */
1015 	sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1016 
1017 	/*
1018 	 * For a single fuse filesystem use max 1% of dirty +
1019 	 * writeback threshold.
1020 	 *
1021 	 * This gives about 1M of write buffer for memory maps on a
1022 	 * machine with 1G and 10% dirty_ratio, which should be more
1023 	 * than enough.
1024 	 *
1025 	 * Privileged users can raise it by writing to
1026 	 *
1027 	 *    /sys/class/bdi/<bdi>/max_ratio
1028 	 */
1029 	bdi_set_max_ratio(sb->s_bdi, 1);
1030 
1031 	return 0;
1032 }
1033 
1034 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1035 {
1036 	struct fuse_dev *fud;
1037 	struct list_head *pq;
1038 
1039 	fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1040 	if (!fud)
1041 		return NULL;
1042 
1043 	pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1044 	if (!pq) {
1045 		kfree(fud);
1046 		return NULL;
1047 	}
1048 
1049 	fud->pq.processing = pq;
1050 	fud->fc = fuse_conn_get(fc);
1051 	fuse_pqueue_init(&fud->pq);
1052 
1053 	spin_lock(&fc->lock);
1054 	list_add_tail(&fud->entry, &fc->devices);
1055 	spin_unlock(&fc->lock);
1056 
1057 	return fud;
1058 }
1059 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1060 
1061 void fuse_dev_free(struct fuse_dev *fud)
1062 {
1063 	struct fuse_conn *fc = fud->fc;
1064 
1065 	if (fc) {
1066 		spin_lock(&fc->lock);
1067 		list_del(&fud->entry);
1068 		spin_unlock(&fc->lock);
1069 
1070 		fuse_conn_put(fc);
1071 	}
1072 	kfree(fud->pq.processing);
1073 	kfree(fud);
1074 }
1075 EXPORT_SYMBOL_GPL(fuse_dev_free);
1076 
1077 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1078 {
1079 	struct fuse_dev *fud;
1080 	struct fuse_conn *fc;
1081 	struct inode *root;
1082 	struct fuse_mount_data d;
1083 	struct file *file;
1084 	struct dentry *root_dentry;
1085 	struct fuse_req *init_req;
1086 	int err;
1087 	int is_bdev = sb->s_bdev != NULL;
1088 
1089 	err = -EINVAL;
1090 	if (sb->s_flags & SB_MANDLOCK)
1091 		goto err;
1092 
1093 	sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1094 
1095 	if (!parse_fuse_opt(data, &d, is_bdev, sb->s_user_ns))
1096 		goto err;
1097 
1098 	if (is_bdev) {
1099 #ifdef CONFIG_BLOCK
1100 		err = -EINVAL;
1101 		if (!sb_set_blocksize(sb, d.blksize))
1102 			goto err;
1103 #endif
1104 	} else {
1105 		sb->s_blocksize = PAGE_SIZE;
1106 		sb->s_blocksize_bits = PAGE_SHIFT;
1107 	}
1108 	sb->s_magic = FUSE_SUPER_MAGIC;
1109 	sb->s_op = &fuse_super_operations;
1110 	sb->s_xattr = fuse_xattr_handlers;
1111 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1112 	sb->s_time_gran = 1;
1113 	sb->s_export_op = &fuse_export_operations;
1114 	sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1115 	if (sb->s_user_ns != &init_user_ns)
1116 		sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1117 
1118 	file = fget(d.fd);
1119 	err = -EINVAL;
1120 	if (!file)
1121 		goto err;
1122 
1123 	/*
1124 	 * Require mount to happen from the same user namespace which
1125 	 * opened /dev/fuse to prevent potential attacks.
1126 	 */
1127 	if (file->f_op != &fuse_dev_operations ||
1128 	    file->f_cred->user_ns != sb->s_user_ns)
1129 		goto err_fput;
1130 
1131 	/*
1132 	 * If we are not in the initial user namespace posix
1133 	 * acls must be translated.
1134 	 */
1135 	if (sb->s_user_ns != &init_user_ns)
1136 		sb->s_xattr = fuse_no_acl_xattr_handlers;
1137 
1138 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1139 	err = -ENOMEM;
1140 	if (!fc)
1141 		goto err_fput;
1142 
1143 	fuse_conn_init(fc, sb->s_user_ns);
1144 	fc->release = fuse_free_conn;
1145 
1146 	fud = fuse_dev_alloc(fc);
1147 	if (!fud)
1148 		goto err_put_conn;
1149 
1150 	fc->dev = sb->s_dev;
1151 	fc->sb = sb;
1152 	err = fuse_bdi_init(fc, sb);
1153 	if (err)
1154 		goto err_dev_free;
1155 
1156 	/* Handle umasking inside the fuse code */
1157 	if (sb->s_flags & SB_POSIXACL)
1158 		fc->dont_mask = 1;
1159 	sb->s_flags |= SB_POSIXACL;
1160 
1161 	fc->default_permissions = d.default_permissions;
1162 	fc->allow_other = d.allow_other;
1163 	fc->user_id = d.user_id;
1164 	fc->group_id = d.group_id;
1165 	fc->max_read = max_t(unsigned, 4096, d.max_read);
1166 
1167 	/* Used by get_root_inode() */
1168 	sb->s_fs_info = fc;
1169 
1170 	err = -ENOMEM;
1171 	root = fuse_get_root_inode(sb, d.rootmode);
1172 	sb->s_d_op = &fuse_root_dentry_operations;
1173 	root_dentry = d_make_root(root);
1174 	if (!root_dentry)
1175 		goto err_dev_free;
1176 	/* Root dentry doesn't have .d_revalidate */
1177 	sb->s_d_op = &fuse_dentry_operations;
1178 
1179 	init_req = fuse_request_alloc(0);
1180 	if (!init_req)
1181 		goto err_put_root;
1182 	__set_bit(FR_BACKGROUND, &init_req->flags);
1183 
1184 	if (is_bdev) {
1185 		fc->destroy_req = fuse_request_alloc(0);
1186 		if (!fc->destroy_req)
1187 			goto err_free_init_req;
1188 	}
1189 
1190 	mutex_lock(&fuse_mutex);
1191 	err = -EINVAL;
1192 	if (file->private_data)
1193 		goto err_unlock;
1194 
1195 	err = fuse_ctl_add_conn(fc);
1196 	if (err)
1197 		goto err_unlock;
1198 
1199 	list_add_tail(&fc->entry, &fuse_conn_list);
1200 	sb->s_root = root_dentry;
1201 	file->private_data = fud;
1202 	mutex_unlock(&fuse_mutex);
1203 	/*
1204 	 * atomic_dec_and_test() in fput() provides the necessary
1205 	 * memory barrier for file->private_data to be visible on all
1206 	 * CPUs after this
1207 	 */
1208 	fput(file);
1209 
1210 	fuse_send_init(fc, init_req);
1211 
1212 	return 0;
1213 
1214  err_unlock:
1215 	mutex_unlock(&fuse_mutex);
1216  err_free_init_req:
1217 	fuse_request_free(init_req);
1218  err_put_root:
1219 	dput(root_dentry);
1220  err_dev_free:
1221 	fuse_dev_free(fud);
1222  err_put_conn:
1223 	fuse_conn_put(fc);
1224 	sb->s_fs_info = NULL;
1225  err_fput:
1226 	fput(file);
1227  err:
1228 	return err;
1229 }
1230 
1231 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1232 		       int flags, const char *dev_name,
1233 		       void *raw_data)
1234 {
1235 	return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1236 }
1237 
1238 static void fuse_sb_destroy(struct super_block *sb)
1239 {
1240 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1241 
1242 	if (fc) {
1243 		fuse_send_destroy(fc);
1244 
1245 		fuse_abort_conn(fc);
1246 		fuse_wait_aborted(fc);
1247 
1248 		down_write(&fc->killsb);
1249 		fc->sb = NULL;
1250 		up_write(&fc->killsb);
1251 	}
1252 }
1253 
1254 static void fuse_kill_sb_anon(struct super_block *sb)
1255 {
1256 	fuse_sb_destroy(sb);
1257 	kill_anon_super(sb);
1258 }
1259 
1260 static struct file_system_type fuse_fs_type = {
1261 	.owner		= THIS_MODULE,
1262 	.name		= "fuse",
1263 	.fs_flags	= FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1264 	.mount		= fuse_mount,
1265 	.kill_sb	= fuse_kill_sb_anon,
1266 };
1267 MODULE_ALIAS_FS("fuse");
1268 
1269 #ifdef CONFIG_BLOCK
1270 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1271 			   int flags, const char *dev_name,
1272 			   void *raw_data)
1273 {
1274 	return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1275 }
1276 
1277 static void fuse_kill_sb_blk(struct super_block *sb)
1278 {
1279 	fuse_sb_destroy(sb);
1280 	kill_block_super(sb);
1281 }
1282 
1283 static struct file_system_type fuseblk_fs_type = {
1284 	.owner		= THIS_MODULE,
1285 	.name		= "fuseblk",
1286 	.mount		= fuse_mount_blk,
1287 	.kill_sb	= fuse_kill_sb_blk,
1288 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1289 };
1290 MODULE_ALIAS_FS("fuseblk");
1291 
1292 static inline int register_fuseblk(void)
1293 {
1294 	return register_filesystem(&fuseblk_fs_type);
1295 }
1296 
1297 static inline void unregister_fuseblk(void)
1298 {
1299 	unregister_filesystem(&fuseblk_fs_type);
1300 }
1301 #else
1302 static inline int register_fuseblk(void)
1303 {
1304 	return 0;
1305 }
1306 
1307 static inline void unregister_fuseblk(void)
1308 {
1309 }
1310 #endif
1311 
1312 static void fuse_inode_init_once(void *foo)
1313 {
1314 	struct inode *inode = foo;
1315 
1316 	inode_init_once(inode);
1317 }
1318 
1319 static int __init fuse_fs_init(void)
1320 {
1321 	int err;
1322 
1323 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
1324 			sizeof(struct fuse_inode), 0,
1325 			SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1326 			fuse_inode_init_once);
1327 	err = -ENOMEM;
1328 	if (!fuse_inode_cachep)
1329 		goto out;
1330 
1331 	err = register_fuseblk();
1332 	if (err)
1333 		goto out2;
1334 
1335 	err = register_filesystem(&fuse_fs_type);
1336 	if (err)
1337 		goto out3;
1338 
1339 	return 0;
1340 
1341  out3:
1342 	unregister_fuseblk();
1343  out2:
1344 	kmem_cache_destroy(fuse_inode_cachep);
1345  out:
1346 	return err;
1347 }
1348 
1349 static void fuse_fs_cleanup(void)
1350 {
1351 	unregister_filesystem(&fuse_fs_type);
1352 	unregister_fuseblk();
1353 
1354 	/*
1355 	 * Make sure all delayed rcu free inodes are flushed before we
1356 	 * destroy cache.
1357 	 */
1358 	rcu_barrier();
1359 	kmem_cache_destroy(fuse_inode_cachep);
1360 }
1361 
1362 static struct kobject *fuse_kobj;
1363 
1364 static int fuse_sysfs_init(void)
1365 {
1366 	int err;
1367 
1368 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1369 	if (!fuse_kobj) {
1370 		err = -ENOMEM;
1371 		goto out_err;
1372 	}
1373 
1374 	err = sysfs_create_mount_point(fuse_kobj, "connections");
1375 	if (err)
1376 		goto out_fuse_unregister;
1377 
1378 	return 0;
1379 
1380  out_fuse_unregister:
1381 	kobject_put(fuse_kobj);
1382  out_err:
1383 	return err;
1384 }
1385 
1386 static void fuse_sysfs_cleanup(void)
1387 {
1388 	sysfs_remove_mount_point(fuse_kobj, "connections");
1389 	kobject_put(fuse_kobj);
1390 }
1391 
1392 static int __init fuse_init(void)
1393 {
1394 	int res;
1395 
1396 	printk(KERN_INFO "fuse init (API version %i.%i)\n",
1397 	       FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1398 
1399 	INIT_LIST_HEAD(&fuse_conn_list);
1400 	res = fuse_fs_init();
1401 	if (res)
1402 		goto err;
1403 
1404 	res = fuse_dev_init();
1405 	if (res)
1406 		goto err_fs_cleanup;
1407 
1408 	res = fuse_sysfs_init();
1409 	if (res)
1410 		goto err_dev_cleanup;
1411 
1412 	res = fuse_ctl_init();
1413 	if (res)
1414 		goto err_sysfs_cleanup;
1415 
1416 	sanitize_global_limit(&max_user_bgreq);
1417 	sanitize_global_limit(&max_user_congthresh);
1418 
1419 	return 0;
1420 
1421  err_sysfs_cleanup:
1422 	fuse_sysfs_cleanup();
1423  err_dev_cleanup:
1424 	fuse_dev_cleanup();
1425  err_fs_cleanup:
1426 	fuse_fs_cleanup();
1427  err:
1428 	return res;
1429 }
1430 
1431 static void __exit fuse_exit(void)
1432 {
1433 	printk(KERN_DEBUG "fuse exit\n");
1434 
1435 	fuse_ctl_cleanup();
1436 	fuse_sysfs_cleanup();
1437 	fuse_fs_cleanup();
1438 	fuse_dev_cleanup();
1439 }
1440 
1441 module_init(fuse_init);
1442 module_exit(fuse_exit);
1443