xref: /openbmc/linux/fs/fuse/inode.c (revision e1cd7b80)
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 	spin_lock_init(&fi->lock);
89 	fi->forget = fuse_alloc_forget();
90 	if (!fi->forget) {
91 		kmem_cache_free(fuse_inode_cachep, fi);
92 		return NULL;
93 	}
94 
95 	return &fi->inode;
96 }
97 
98 static void fuse_free_inode(struct inode *inode)
99 {
100 	struct fuse_inode *fi = get_fuse_inode(inode);
101 
102 	mutex_destroy(&fi->mutex);
103 	kfree(fi->forget);
104 	kmem_cache_free(fuse_inode_cachep, fi);
105 }
106 
107 static void fuse_evict_inode(struct inode *inode)
108 {
109 	struct fuse_inode *fi = get_fuse_inode(inode);
110 
111 	truncate_inode_pages_final(&inode->i_data);
112 	clear_inode(inode);
113 	if (inode->i_sb->s_flags & SB_ACTIVE) {
114 		struct fuse_conn *fc = get_fuse_conn(inode);
115 		fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
116 		fi->forget = NULL;
117 	}
118 	if (S_ISREG(inode->i_mode) && !is_bad_inode(inode)) {
119 		WARN_ON(!list_empty(&fi->write_files));
120 		WARN_ON(!list_empty(&fi->queued_writes));
121 	}
122 }
123 
124 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
125 {
126 	sync_filesystem(sb);
127 	if (*flags & SB_MANDLOCK)
128 		return -EINVAL;
129 
130 	return 0;
131 }
132 
133 /*
134  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
135  * so that it will fit.
136  */
137 static ino_t fuse_squash_ino(u64 ino64)
138 {
139 	ino_t ino = (ino_t) ino64;
140 	if (sizeof(ino_t) < sizeof(u64))
141 		ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
142 	return ino;
143 }
144 
145 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
146 				   u64 attr_valid)
147 {
148 	struct fuse_conn *fc = get_fuse_conn(inode);
149 	struct fuse_inode *fi = get_fuse_inode(inode);
150 
151 	lockdep_assert_held(&fi->lock);
152 
153 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
154 	fi->i_time = attr_valid;
155 	WRITE_ONCE(fi->inval_mask, 0);
156 
157 	inode->i_ino     = fuse_squash_ino(attr->ino);
158 	inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
159 	set_nlink(inode, attr->nlink);
160 	inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
161 	inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
162 	inode->i_blocks  = attr->blocks;
163 	inode->i_atime.tv_sec   = attr->atime;
164 	inode->i_atime.tv_nsec  = attr->atimensec;
165 	/* mtime from server may be stale due to local buffered write */
166 	if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
167 		inode->i_mtime.tv_sec   = attr->mtime;
168 		inode->i_mtime.tv_nsec  = attr->mtimensec;
169 		inode->i_ctime.tv_sec   = attr->ctime;
170 		inode->i_ctime.tv_nsec  = attr->ctimensec;
171 	}
172 
173 	if (attr->blksize != 0)
174 		inode->i_blkbits = ilog2(attr->blksize);
175 	else
176 		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
177 
178 	/*
179 	 * Don't set the sticky bit in i_mode, unless we want the VFS
180 	 * to check permissions.  This prevents failures due to the
181 	 * check in may_delete().
182 	 */
183 	fi->orig_i_mode = inode->i_mode;
184 	if (!fc->default_permissions)
185 		inode->i_mode &= ~S_ISVTX;
186 
187 	fi->orig_ino = attr->ino;
188 }
189 
190 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
191 			    u64 attr_valid, u64 attr_version)
192 {
193 	struct fuse_conn *fc = get_fuse_conn(inode);
194 	struct fuse_inode *fi = get_fuse_inode(inode);
195 	bool is_wb = fc->writeback_cache;
196 	loff_t oldsize;
197 	struct timespec64 old_mtime;
198 
199 	spin_lock(&fi->lock);
200 	if ((attr_version != 0 && fi->attr_version > attr_version) ||
201 	    test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
202 		spin_unlock(&fi->lock);
203 		return;
204 	}
205 
206 	old_mtime = inode->i_mtime;
207 	fuse_change_attributes_common(inode, attr, attr_valid);
208 
209 	oldsize = inode->i_size;
210 	/*
211 	 * In case of writeback_cache enabled, the cached writes beyond EOF
212 	 * extend local i_size without keeping userspace server in sync. So,
213 	 * attr->size coming from server can be stale. We cannot trust it.
214 	 */
215 	if (!is_wb || !S_ISREG(inode->i_mode))
216 		i_size_write(inode, attr->size);
217 	spin_unlock(&fi->lock);
218 
219 	if (!is_wb && S_ISREG(inode->i_mode)) {
220 		bool inval = false;
221 
222 		if (oldsize != attr->size) {
223 			truncate_pagecache(inode, attr->size);
224 			if (!fc->explicit_inval_data)
225 				inval = true;
226 		} else if (fc->auto_inval_data) {
227 			struct timespec64 new_mtime = {
228 				.tv_sec = attr->mtime,
229 				.tv_nsec = attr->mtimensec,
230 			};
231 
232 			/*
233 			 * Auto inval mode also checks and invalidates if mtime
234 			 * has changed.
235 			 */
236 			if (!timespec64_equal(&old_mtime, &new_mtime))
237 				inval = true;
238 		}
239 
240 		if (inval)
241 			invalidate_inode_pages2(inode->i_mapping);
242 	}
243 }
244 
245 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
246 {
247 	inode->i_mode = attr->mode & S_IFMT;
248 	inode->i_size = attr->size;
249 	inode->i_mtime.tv_sec  = attr->mtime;
250 	inode->i_mtime.tv_nsec = attr->mtimensec;
251 	inode->i_ctime.tv_sec  = attr->ctime;
252 	inode->i_ctime.tv_nsec = attr->ctimensec;
253 	if (S_ISREG(inode->i_mode)) {
254 		fuse_init_common(inode);
255 		fuse_init_file_inode(inode);
256 	} else if (S_ISDIR(inode->i_mode))
257 		fuse_init_dir(inode);
258 	else if (S_ISLNK(inode->i_mode))
259 		fuse_init_symlink(inode);
260 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
261 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
262 		fuse_init_common(inode);
263 		init_special_inode(inode, inode->i_mode,
264 				   new_decode_dev(attr->rdev));
265 	} else
266 		BUG();
267 }
268 
269 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
270 {
271 	u64 nodeid = *(u64 *) _nodeidp;
272 	if (get_node_id(inode) == nodeid)
273 		return 1;
274 	else
275 		return 0;
276 }
277 
278 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
279 {
280 	u64 nodeid = *(u64 *) _nodeidp;
281 	get_fuse_inode(inode)->nodeid = nodeid;
282 	return 0;
283 }
284 
285 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
286 			int generation, struct fuse_attr *attr,
287 			u64 attr_valid, u64 attr_version)
288 {
289 	struct inode *inode;
290 	struct fuse_inode *fi;
291 	struct fuse_conn *fc = get_fuse_conn_super(sb);
292 
293  retry:
294 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
295 	if (!inode)
296 		return NULL;
297 
298 	if ((inode->i_state & I_NEW)) {
299 		inode->i_flags |= S_NOATIME;
300 		if (!fc->writeback_cache || !S_ISREG(attr->mode))
301 			inode->i_flags |= S_NOCMTIME;
302 		inode->i_generation = generation;
303 		fuse_init_inode(inode, attr);
304 		unlock_new_inode(inode);
305 	} else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
306 		/* Inode has changed type, any I/O on the old should fail */
307 		make_bad_inode(inode);
308 		iput(inode);
309 		goto retry;
310 	}
311 
312 	fi = get_fuse_inode(inode);
313 	spin_lock(&fi->lock);
314 	fi->nlookup++;
315 	spin_unlock(&fi->lock);
316 	fuse_change_attributes(inode, attr, attr_valid, attr_version);
317 
318 	return inode;
319 }
320 
321 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
322 			     loff_t offset, loff_t len)
323 {
324 	struct inode *inode;
325 	pgoff_t pg_start;
326 	pgoff_t pg_end;
327 
328 	inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
329 	if (!inode)
330 		return -ENOENT;
331 
332 	fuse_invalidate_attr(inode);
333 	forget_all_cached_acls(inode);
334 	if (offset >= 0) {
335 		pg_start = offset >> PAGE_SHIFT;
336 		if (len <= 0)
337 			pg_end = -1;
338 		else
339 			pg_end = (offset + len - 1) >> PAGE_SHIFT;
340 		invalidate_inode_pages2_range(inode->i_mapping,
341 					      pg_start, pg_end);
342 	}
343 	iput(inode);
344 	return 0;
345 }
346 
347 bool fuse_lock_inode(struct inode *inode)
348 {
349 	bool locked = false;
350 
351 	if (!get_fuse_conn(inode)->parallel_dirops) {
352 		mutex_lock(&get_fuse_inode(inode)->mutex);
353 		locked = true;
354 	}
355 
356 	return locked;
357 }
358 
359 void fuse_unlock_inode(struct inode *inode, bool locked)
360 {
361 	if (locked)
362 		mutex_unlock(&get_fuse_inode(inode)->mutex);
363 }
364 
365 static void fuse_umount_begin(struct super_block *sb)
366 {
367 	struct fuse_conn *fc = get_fuse_conn_super(sb);
368 
369 	if (!fc->no_force_umount)
370 		fuse_abort_conn(fc);
371 }
372 
373 static void fuse_send_destroy(struct fuse_conn *fc)
374 {
375 	if (fc->conn_init) {
376 		FUSE_ARGS(args);
377 
378 		args.opcode = FUSE_DESTROY;
379 		args.force = true;
380 		args.nocreds = true;
381 		fuse_simple_request(fc, &args);
382 	}
383 }
384 
385 static void fuse_put_super(struct super_block *sb)
386 {
387 	struct fuse_conn *fc = get_fuse_conn_super(sb);
388 
389 	mutex_lock(&fuse_mutex);
390 	list_del(&fc->entry);
391 	fuse_ctl_remove_conn(fc);
392 	mutex_unlock(&fuse_mutex);
393 
394 	fuse_conn_put(fc);
395 }
396 
397 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
398 {
399 	stbuf->f_type    = FUSE_SUPER_MAGIC;
400 	stbuf->f_bsize   = attr->bsize;
401 	stbuf->f_frsize  = attr->frsize;
402 	stbuf->f_blocks  = attr->blocks;
403 	stbuf->f_bfree   = attr->bfree;
404 	stbuf->f_bavail  = attr->bavail;
405 	stbuf->f_files   = attr->files;
406 	stbuf->f_ffree   = attr->ffree;
407 	stbuf->f_namelen = attr->namelen;
408 	/* fsid is left zero */
409 }
410 
411 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
412 {
413 	struct super_block *sb = dentry->d_sb;
414 	struct fuse_conn *fc = get_fuse_conn_super(sb);
415 	FUSE_ARGS(args);
416 	struct fuse_statfs_out outarg;
417 	int err;
418 
419 	if (!fuse_allow_current_process(fc)) {
420 		buf->f_type = FUSE_SUPER_MAGIC;
421 		return 0;
422 	}
423 
424 	memset(&outarg, 0, sizeof(outarg));
425 	args.in_numargs = 0;
426 	args.opcode = FUSE_STATFS;
427 	args.nodeid = get_node_id(d_inode(dentry));
428 	args.out_numargs = 1;
429 	args.out_args[0].size = sizeof(outarg);
430 	args.out_args[0].value = &outarg;
431 	err = fuse_simple_request(fc, &args);
432 	if (!err)
433 		convert_fuse_statfs(buf, &outarg.st);
434 	return err;
435 }
436 
437 enum {
438 	OPT_SOURCE,
439 	OPT_SUBTYPE,
440 	OPT_FD,
441 	OPT_ROOTMODE,
442 	OPT_USER_ID,
443 	OPT_GROUP_ID,
444 	OPT_DEFAULT_PERMISSIONS,
445 	OPT_ALLOW_OTHER,
446 	OPT_MAX_READ,
447 	OPT_BLKSIZE,
448 	OPT_ERR
449 };
450 
451 static const struct fs_parameter_spec fuse_param_specs[] = {
452 	fsparam_string	("source",		OPT_SOURCE),
453 	fsparam_u32	("fd",			OPT_FD),
454 	fsparam_u32oct	("rootmode",		OPT_ROOTMODE),
455 	fsparam_u32	("user_id",		OPT_USER_ID),
456 	fsparam_u32	("group_id",		OPT_GROUP_ID),
457 	fsparam_flag	("default_permissions",	OPT_DEFAULT_PERMISSIONS),
458 	fsparam_flag	("allow_other",		OPT_ALLOW_OTHER),
459 	fsparam_u32	("max_read",		OPT_MAX_READ),
460 	fsparam_u32	("blksize",		OPT_BLKSIZE),
461 	fsparam_string	("subtype",		OPT_SUBTYPE),
462 	{}
463 };
464 
465 static const struct fs_parameter_description fuse_fs_parameters = {
466 	.name		= "fuse",
467 	.specs		= fuse_param_specs,
468 };
469 
470 static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param)
471 {
472 	struct fs_parse_result result;
473 	struct fuse_fs_context *ctx = fc->fs_private;
474 	int opt;
475 
476 	opt = fs_parse(fc, &fuse_fs_parameters, param, &result);
477 	if (opt < 0)
478 		return opt;
479 
480 	switch (opt) {
481 	case OPT_SOURCE:
482 		if (fc->source)
483 			return invalf(fc, "fuse: Multiple sources specified");
484 		fc->source = param->string;
485 		param->string = NULL;
486 		break;
487 
488 	case OPT_SUBTYPE:
489 		if (ctx->subtype)
490 			return invalf(fc, "fuse: Multiple subtypes specified");
491 		ctx->subtype = param->string;
492 		param->string = NULL;
493 		return 0;
494 
495 	case OPT_FD:
496 		ctx->fd = result.uint_32;
497 		ctx->fd_present = 1;
498 		break;
499 
500 	case OPT_ROOTMODE:
501 		if (!fuse_valid_type(result.uint_32))
502 			return invalf(fc, "fuse: Invalid rootmode");
503 		ctx->rootmode = result.uint_32;
504 		ctx->rootmode_present = 1;
505 		break;
506 
507 	case OPT_USER_ID:
508 		ctx->user_id = make_kuid(fc->user_ns, result.uint_32);
509 		if (!uid_valid(ctx->user_id))
510 			return invalf(fc, "fuse: Invalid user_id");
511 		ctx->user_id_present = 1;
512 		break;
513 
514 	case OPT_GROUP_ID:
515 		ctx->group_id = make_kgid(fc->user_ns, result.uint_32);
516 		if (!gid_valid(ctx->group_id))
517 			return invalf(fc, "fuse: Invalid group_id");
518 		ctx->group_id_present = 1;
519 		break;
520 
521 	case OPT_DEFAULT_PERMISSIONS:
522 		ctx->default_permissions = 1;
523 		break;
524 
525 	case OPT_ALLOW_OTHER:
526 		ctx->allow_other = 1;
527 		break;
528 
529 	case OPT_MAX_READ:
530 		ctx->max_read = result.uint_32;
531 		break;
532 
533 	case OPT_BLKSIZE:
534 		if (!ctx->is_bdev)
535 			return invalf(fc, "fuse: blksize only supported for fuseblk");
536 		ctx->blksize = result.uint_32;
537 		break;
538 
539 	default:
540 		return -EINVAL;
541 	}
542 
543 	return 0;
544 }
545 
546 static void fuse_free_fc(struct fs_context *fc)
547 {
548 	struct fuse_fs_context *ctx = fc->fs_private;
549 
550 	if (ctx) {
551 		kfree(ctx->subtype);
552 		kfree(ctx);
553 	}
554 }
555 
556 static int fuse_show_options(struct seq_file *m, struct dentry *root)
557 {
558 	struct super_block *sb = root->d_sb;
559 	struct fuse_conn *fc = get_fuse_conn_super(sb);
560 
561 	if (fc->no_mount_options)
562 		return 0;
563 
564 	seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
565 	seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
566 	if (fc->default_permissions)
567 		seq_puts(m, ",default_permissions");
568 	if (fc->allow_other)
569 		seq_puts(m, ",allow_other");
570 	if (fc->max_read != ~0)
571 		seq_printf(m, ",max_read=%u", fc->max_read);
572 	if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
573 		seq_printf(m, ",blksize=%lu", sb->s_blocksize);
574 	return 0;
575 }
576 
577 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
578 			     const struct fuse_iqueue_ops *ops,
579 			     void *priv)
580 {
581 	memset(fiq, 0, sizeof(struct fuse_iqueue));
582 	spin_lock_init(&fiq->lock);
583 	init_waitqueue_head(&fiq->waitq);
584 	INIT_LIST_HEAD(&fiq->pending);
585 	INIT_LIST_HEAD(&fiq->interrupts);
586 	fiq->forget_list_tail = &fiq->forget_list_head;
587 	fiq->connected = 1;
588 	fiq->ops = ops;
589 	fiq->priv = priv;
590 }
591 
592 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
593 {
594 	unsigned int i;
595 
596 	spin_lock_init(&fpq->lock);
597 	for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
598 		INIT_LIST_HEAD(&fpq->processing[i]);
599 	INIT_LIST_HEAD(&fpq->io);
600 	fpq->connected = 1;
601 }
602 
603 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns,
604 		    const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
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 	fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
614 	INIT_LIST_HEAD(&fc->bg_queue);
615 	INIT_LIST_HEAD(&fc->entry);
616 	INIT_LIST_HEAD(&fc->devices);
617 	atomic_set(&fc->num_waiting, 0);
618 	fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
619 	fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
620 	atomic64_set(&fc->khctr, 0);
621 	fc->polled_files = RB_ROOT;
622 	fc->blocked = 0;
623 	fc->initialized = 0;
624 	fc->connected = 1;
625 	atomic64_set(&fc->attr_version, 1);
626 	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
627 	fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
628 	fc->user_ns = get_user_ns(user_ns);
629 	fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
630 }
631 EXPORT_SYMBOL_GPL(fuse_conn_init);
632 
633 void fuse_conn_put(struct fuse_conn *fc)
634 {
635 	if (refcount_dec_and_test(&fc->count)) {
636 		struct fuse_iqueue *fiq = &fc->iq;
637 
638 		if (fiq->ops->release)
639 			fiq->ops->release(fiq);
640 		put_pid_ns(fc->pid_ns);
641 		put_user_ns(fc->user_ns);
642 		fc->release(fc);
643 	}
644 }
645 EXPORT_SYMBOL_GPL(fuse_conn_put);
646 
647 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
648 {
649 	refcount_inc(&fc->count);
650 	return fc;
651 }
652 EXPORT_SYMBOL_GPL(fuse_conn_get);
653 
654 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
655 {
656 	struct fuse_attr attr;
657 	memset(&attr, 0, sizeof(attr));
658 
659 	attr.mode = mode;
660 	attr.ino = FUSE_ROOT_ID;
661 	attr.nlink = 1;
662 	return fuse_iget(sb, 1, 0, &attr, 0, 0);
663 }
664 
665 struct fuse_inode_handle {
666 	u64 nodeid;
667 	u32 generation;
668 };
669 
670 static struct dentry *fuse_get_dentry(struct super_block *sb,
671 				      struct fuse_inode_handle *handle)
672 {
673 	struct fuse_conn *fc = get_fuse_conn_super(sb);
674 	struct inode *inode;
675 	struct dentry *entry;
676 	int err = -ESTALE;
677 
678 	if (handle->nodeid == 0)
679 		goto out_err;
680 
681 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
682 	if (!inode) {
683 		struct fuse_entry_out outarg;
684 		const struct qstr name = QSTR_INIT(".", 1);
685 
686 		if (!fc->export_support)
687 			goto out_err;
688 
689 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
690 				       &inode);
691 		if (err && err != -ENOENT)
692 			goto out_err;
693 		if (err || !inode) {
694 			err = -ESTALE;
695 			goto out_err;
696 		}
697 		err = -EIO;
698 		if (get_node_id(inode) != handle->nodeid)
699 			goto out_iput;
700 	}
701 	err = -ESTALE;
702 	if (inode->i_generation != handle->generation)
703 		goto out_iput;
704 
705 	entry = d_obtain_alias(inode);
706 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
707 		fuse_invalidate_entry_cache(entry);
708 
709 	return entry;
710 
711  out_iput:
712 	iput(inode);
713  out_err:
714 	return ERR_PTR(err);
715 }
716 
717 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
718 			   struct inode *parent)
719 {
720 	int len = parent ? 6 : 3;
721 	u64 nodeid;
722 	u32 generation;
723 
724 	if (*max_len < len) {
725 		*max_len = len;
726 		return  FILEID_INVALID;
727 	}
728 
729 	nodeid = get_fuse_inode(inode)->nodeid;
730 	generation = inode->i_generation;
731 
732 	fh[0] = (u32)(nodeid >> 32);
733 	fh[1] = (u32)(nodeid & 0xffffffff);
734 	fh[2] = generation;
735 
736 	if (parent) {
737 		nodeid = get_fuse_inode(parent)->nodeid;
738 		generation = parent->i_generation;
739 
740 		fh[3] = (u32)(nodeid >> 32);
741 		fh[4] = (u32)(nodeid & 0xffffffff);
742 		fh[5] = generation;
743 	}
744 
745 	*max_len = len;
746 	return parent ? 0x82 : 0x81;
747 }
748 
749 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
750 		struct fid *fid, int fh_len, int fh_type)
751 {
752 	struct fuse_inode_handle handle;
753 
754 	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
755 		return NULL;
756 
757 	handle.nodeid = (u64) fid->raw[0] << 32;
758 	handle.nodeid |= (u64) fid->raw[1];
759 	handle.generation = fid->raw[2];
760 	return fuse_get_dentry(sb, &handle);
761 }
762 
763 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
764 		struct fid *fid, int fh_len, int fh_type)
765 {
766 	struct fuse_inode_handle parent;
767 
768 	if (fh_type != 0x82 || fh_len < 6)
769 		return NULL;
770 
771 	parent.nodeid = (u64) fid->raw[3] << 32;
772 	parent.nodeid |= (u64) fid->raw[4];
773 	parent.generation = fid->raw[5];
774 	return fuse_get_dentry(sb, &parent);
775 }
776 
777 static struct dentry *fuse_get_parent(struct dentry *child)
778 {
779 	struct inode *child_inode = d_inode(child);
780 	struct fuse_conn *fc = get_fuse_conn(child_inode);
781 	struct inode *inode;
782 	struct dentry *parent;
783 	struct fuse_entry_out outarg;
784 	const struct qstr name = QSTR_INIT("..", 2);
785 	int err;
786 
787 	if (!fc->export_support)
788 		return ERR_PTR(-ESTALE);
789 
790 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
791 			       &name, &outarg, &inode);
792 	if (err) {
793 		if (err == -ENOENT)
794 			return ERR_PTR(-ESTALE);
795 		return ERR_PTR(err);
796 	}
797 
798 	parent = d_obtain_alias(inode);
799 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
800 		fuse_invalidate_entry_cache(parent);
801 
802 	return parent;
803 }
804 
805 static const struct export_operations fuse_export_operations = {
806 	.fh_to_dentry	= fuse_fh_to_dentry,
807 	.fh_to_parent	= fuse_fh_to_parent,
808 	.encode_fh	= fuse_encode_fh,
809 	.get_parent	= fuse_get_parent,
810 };
811 
812 static const struct super_operations fuse_super_operations = {
813 	.alloc_inode    = fuse_alloc_inode,
814 	.free_inode     = fuse_free_inode,
815 	.evict_inode	= fuse_evict_inode,
816 	.write_inode	= fuse_write_inode,
817 	.drop_inode	= generic_delete_inode,
818 	.remount_fs	= fuse_remount_fs,
819 	.put_super	= fuse_put_super,
820 	.umount_begin	= fuse_umount_begin,
821 	.statfs		= fuse_statfs,
822 	.show_options	= fuse_show_options,
823 };
824 
825 static void sanitize_global_limit(unsigned *limit)
826 {
827 	/*
828 	 * The default maximum number of async requests is calculated to consume
829 	 * 1/2^13 of the total memory, assuming 392 bytes per request.
830 	 */
831 	if (*limit == 0)
832 		*limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
833 
834 	if (*limit >= 1 << 16)
835 		*limit = (1 << 16) - 1;
836 }
837 
838 static int set_global_limit(const char *val, const struct kernel_param *kp)
839 {
840 	int rv;
841 
842 	rv = param_set_uint(val, kp);
843 	if (rv)
844 		return rv;
845 
846 	sanitize_global_limit((unsigned *)kp->arg);
847 
848 	return 0;
849 }
850 
851 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
852 {
853 	int cap_sys_admin = capable(CAP_SYS_ADMIN);
854 
855 	if (arg->minor < 13)
856 		return;
857 
858 	sanitize_global_limit(&max_user_bgreq);
859 	sanitize_global_limit(&max_user_congthresh);
860 
861 	spin_lock(&fc->bg_lock);
862 	if (arg->max_background) {
863 		fc->max_background = arg->max_background;
864 
865 		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
866 			fc->max_background = max_user_bgreq;
867 	}
868 	if (arg->congestion_threshold) {
869 		fc->congestion_threshold = arg->congestion_threshold;
870 
871 		if (!cap_sys_admin &&
872 		    fc->congestion_threshold > max_user_congthresh)
873 			fc->congestion_threshold = max_user_congthresh;
874 	}
875 	spin_unlock(&fc->bg_lock);
876 }
877 
878 struct fuse_init_args {
879 	struct fuse_args args;
880 	struct fuse_init_in in;
881 	struct fuse_init_out out;
882 };
883 
884 static void process_init_reply(struct fuse_conn *fc, struct fuse_args *args,
885 			       int error)
886 {
887 	struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
888 	struct fuse_init_out *arg = &ia->out;
889 
890 	if (error || arg->major != FUSE_KERNEL_VERSION)
891 		fc->conn_error = 1;
892 	else {
893 		unsigned long ra_pages;
894 
895 		process_init_limits(fc, arg);
896 
897 		if (arg->minor >= 6) {
898 			ra_pages = arg->max_readahead / PAGE_SIZE;
899 			if (arg->flags & FUSE_ASYNC_READ)
900 				fc->async_read = 1;
901 			if (!(arg->flags & FUSE_POSIX_LOCKS))
902 				fc->no_lock = 1;
903 			if (arg->minor >= 17) {
904 				if (!(arg->flags & FUSE_FLOCK_LOCKS))
905 					fc->no_flock = 1;
906 			} else {
907 				if (!(arg->flags & FUSE_POSIX_LOCKS))
908 					fc->no_flock = 1;
909 			}
910 			if (arg->flags & FUSE_ATOMIC_O_TRUNC)
911 				fc->atomic_o_trunc = 1;
912 			if (arg->minor >= 9) {
913 				/* LOOKUP has dependency on proto version */
914 				if (arg->flags & FUSE_EXPORT_SUPPORT)
915 					fc->export_support = 1;
916 			}
917 			if (arg->flags & FUSE_BIG_WRITES)
918 				fc->big_writes = 1;
919 			if (arg->flags & FUSE_DONT_MASK)
920 				fc->dont_mask = 1;
921 			if (arg->flags & FUSE_AUTO_INVAL_DATA)
922 				fc->auto_inval_data = 1;
923 			else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
924 				fc->explicit_inval_data = 1;
925 			if (arg->flags & FUSE_DO_READDIRPLUS) {
926 				fc->do_readdirplus = 1;
927 				if (arg->flags & FUSE_READDIRPLUS_AUTO)
928 					fc->readdirplus_auto = 1;
929 			}
930 			if (arg->flags & FUSE_ASYNC_DIO)
931 				fc->async_dio = 1;
932 			if (arg->flags & FUSE_WRITEBACK_CACHE)
933 				fc->writeback_cache = 1;
934 			if (arg->flags & FUSE_PARALLEL_DIROPS)
935 				fc->parallel_dirops = 1;
936 			if (arg->flags & FUSE_HANDLE_KILLPRIV)
937 				fc->handle_killpriv = 1;
938 			if (arg->time_gran && arg->time_gran <= 1000000000)
939 				fc->sb->s_time_gran = arg->time_gran;
940 			if ((arg->flags & FUSE_POSIX_ACL)) {
941 				fc->default_permissions = 1;
942 				fc->posix_acl = 1;
943 				fc->sb->s_xattr = fuse_acl_xattr_handlers;
944 			}
945 			if (arg->flags & FUSE_CACHE_SYMLINKS)
946 				fc->cache_symlinks = 1;
947 			if (arg->flags & FUSE_ABORT_ERROR)
948 				fc->abort_err = 1;
949 			if (arg->flags & FUSE_MAX_PAGES) {
950 				fc->max_pages =
951 					min_t(unsigned int, FUSE_MAX_MAX_PAGES,
952 					max_t(unsigned int, arg->max_pages, 1));
953 			}
954 		} else {
955 			ra_pages = fc->max_read / PAGE_SIZE;
956 			fc->no_lock = 1;
957 			fc->no_flock = 1;
958 		}
959 
960 		fc->sb->s_bdi->ra_pages =
961 				min(fc->sb->s_bdi->ra_pages, ra_pages);
962 		fc->minor = arg->minor;
963 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
964 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
965 		fc->conn_init = 1;
966 	}
967 	kfree(ia);
968 
969 	fuse_set_initialized(fc);
970 	wake_up_all(&fc->blocked_waitq);
971 }
972 
973 void fuse_send_init(struct fuse_conn *fc)
974 {
975 	struct fuse_init_args *ia;
976 
977 	ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
978 
979 	ia->in.major = FUSE_KERNEL_VERSION;
980 	ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
981 	ia->in.max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
982 	ia->in.flags |=
983 		FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
984 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
985 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
986 		FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
987 		FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
988 		FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
989 		FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
990 		FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
991 		FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA;
992 	ia->args.opcode = FUSE_INIT;
993 	ia->args.in_numargs = 1;
994 	ia->args.in_args[0].size = sizeof(ia->in);
995 	ia->args.in_args[0].value = &ia->in;
996 	ia->args.out_numargs = 1;
997 	/* Variable length argument used for backward compatibility
998 	   with interface version < 7.5.  Rest of init_out is zeroed
999 	   by do_get_request(), so a short reply is not a problem */
1000 	ia->args.out_argvar = 1;
1001 	ia->args.out_args[0].size = sizeof(ia->out);
1002 	ia->args.out_args[0].value = &ia->out;
1003 	ia->args.force = true;
1004 	ia->args.nocreds = true;
1005 	ia->args.end = process_init_reply;
1006 
1007 	if (fuse_simple_background(fc, &ia->args, GFP_KERNEL) != 0)
1008 		process_init_reply(fc, &ia->args, -ENOTCONN);
1009 }
1010 EXPORT_SYMBOL_GPL(fuse_send_init);
1011 
1012 void fuse_free_conn(struct fuse_conn *fc)
1013 {
1014 	WARN_ON(!list_empty(&fc->devices));
1015 	kfree_rcu(fc, rcu);
1016 }
1017 EXPORT_SYMBOL_GPL(fuse_free_conn);
1018 
1019 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1020 {
1021 	int err;
1022 	char *suffix = "";
1023 
1024 	if (sb->s_bdev) {
1025 		suffix = "-fuseblk";
1026 		/*
1027 		 * sb->s_bdi points to blkdev's bdi however we want to redirect
1028 		 * it to our private bdi...
1029 		 */
1030 		bdi_put(sb->s_bdi);
1031 		sb->s_bdi = &noop_backing_dev_info;
1032 	}
1033 	err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1034 				   MINOR(fc->dev), suffix);
1035 	if (err)
1036 		return err;
1037 
1038 	sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1039 	/* fuse does it's own writeback accounting */
1040 	sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1041 
1042 	/*
1043 	 * For a single fuse filesystem use max 1% of dirty +
1044 	 * writeback threshold.
1045 	 *
1046 	 * This gives about 1M of write buffer for memory maps on a
1047 	 * machine with 1G and 10% dirty_ratio, which should be more
1048 	 * than enough.
1049 	 *
1050 	 * Privileged users can raise it by writing to
1051 	 *
1052 	 *    /sys/class/bdi/<bdi>/max_ratio
1053 	 */
1054 	bdi_set_max_ratio(sb->s_bdi, 1);
1055 
1056 	return 0;
1057 }
1058 
1059 struct fuse_dev *fuse_dev_alloc(void)
1060 {
1061 	struct fuse_dev *fud;
1062 	struct list_head *pq;
1063 
1064 	fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1065 	if (!fud)
1066 		return NULL;
1067 
1068 	pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1069 	if (!pq) {
1070 		kfree(fud);
1071 		return NULL;
1072 	}
1073 
1074 	fud->pq.processing = pq;
1075 	fuse_pqueue_init(&fud->pq);
1076 
1077 	return fud;
1078 }
1079 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1080 
1081 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1082 {
1083 	fud->fc = fuse_conn_get(fc);
1084 	spin_lock(&fc->lock);
1085 	list_add_tail(&fud->entry, &fc->devices);
1086 	spin_unlock(&fc->lock);
1087 }
1088 EXPORT_SYMBOL_GPL(fuse_dev_install);
1089 
1090 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1091 {
1092 	struct fuse_dev *fud;
1093 
1094 	fud = fuse_dev_alloc();
1095 	if (!fud)
1096 		return NULL;
1097 
1098 	fuse_dev_install(fud, fc);
1099 	return fud;
1100 }
1101 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1102 
1103 void fuse_dev_free(struct fuse_dev *fud)
1104 {
1105 	struct fuse_conn *fc = fud->fc;
1106 
1107 	if (fc) {
1108 		spin_lock(&fc->lock);
1109 		list_del(&fud->entry);
1110 		spin_unlock(&fc->lock);
1111 
1112 		fuse_conn_put(fc);
1113 	}
1114 	kfree(fud->pq.processing);
1115 	kfree(fud);
1116 }
1117 EXPORT_SYMBOL_GPL(fuse_dev_free);
1118 
1119 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1120 {
1121 	struct fuse_dev *fud;
1122 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1123 	struct inode *root;
1124 	struct dentry *root_dentry;
1125 	int err;
1126 
1127 	err = -EINVAL;
1128 	if (sb->s_flags & SB_MANDLOCK)
1129 		goto err;
1130 
1131 	sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1132 
1133 	if (ctx->is_bdev) {
1134 #ifdef CONFIG_BLOCK
1135 		err = -EINVAL;
1136 		if (!sb_set_blocksize(sb, ctx->blksize))
1137 			goto err;
1138 #endif
1139 	} else {
1140 		sb->s_blocksize = PAGE_SIZE;
1141 		sb->s_blocksize_bits = PAGE_SHIFT;
1142 	}
1143 
1144 	sb->s_subtype = ctx->subtype;
1145 	ctx->subtype = NULL;
1146 	sb->s_magic = FUSE_SUPER_MAGIC;
1147 	sb->s_op = &fuse_super_operations;
1148 	sb->s_xattr = fuse_xattr_handlers;
1149 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1150 	sb->s_time_gran = 1;
1151 	sb->s_export_op = &fuse_export_operations;
1152 	sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1153 	if (sb->s_user_ns != &init_user_ns)
1154 		sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1155 
1156 	/*
1157 	 * If we are not in the initial user namespace posix
1158 	 * acls must be translated.
1159 	 */
1160 	if (sb->s_user_ns != &init_user_ns)
1161 		sb->s_xattr = fuse_no_acl_xattr_handlers;
1162 
1163 	fud = fuse_dev_alloc_install(fc);
1164 	if (!fud)
1165 		goto err;
1166 
1167 	fc->dev = sb->s_dev;
1168 	fc->sb = sb;
1169 	err = fuse_bdi_init(fc, sb);
1170 	if (err)
1171 		goto err_dev_free;
1172 
1173 	/* Handle umasking inside the fuse code */
1174 	if (sb->s_flags & SB_POSIXACL)
1175 		fc->dont_mask = 1;
1176 	sb->s_flags |= SB_POSIXACL;
1177 
1178 	fc->default_permissions = ctx->default_permissions;
1179 	fc->allow_other = ctx->allow_other;
1180 	fc->user_id = ctx->user_id;
1181 	fc->group_id = ctx->group_id;
1182 	fc->max_read = max_t(unsigned, 4096, ctx->max_read);
1183 	fc->destroy = ctx->destroy;
1184 	fc->no_control = ctx->no_control;
1185 	fc->no_force_umount = ctx->no_force_umount;
1186 	fc->no_mount_options = ctx->no_mount_options;
1187 
1188 	err = -ENOMEM;
1189 	root = fuse_get_root_inode(sb, ctx->rootmode);
1190 	sb->s_d_op = &fuse_root_dentry_operations;
1191 	root_dentry = d_make_root(root);
1192 	if (!root_dentry)
1193 		goto err_dev_free;
1194 	/* Root dentry doesn't have .d_revalidate */
1195 	sb->s_d_op = &fuse_dentry_operations;
1196 
1197 	mutex_lock(&fuse_mutex);
1198 	err = -EINVAL;
1199 	if (*ctx->fudptr)
1200 		goto err_unlock;
1201 
1202 	err = fuse_ctl_add_conn(fc);
1203 	if (err)
1204 		goto err_unlock;
1205 
1206 	list_add_tail(&fc->entry, &fuse_conn_list);
1207 	sb->s_root = root_dentry;
1208 	*ctx->fudptr = fud;
1209 	mutex_unlock(&fuse_mutex);
1210 	return 0;
1211 
1212  err_unlock:
1213 	mutex_unlock(&fuse_mutex);
1214 	dput(root_dentry);
1215  err_dev_free:
1216 	fuse_dev_free(fud);
1217  err:
1218 	return err;
1219 }
1220 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1221 
1222 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1223 {
1224 	struct fuse_fs_context *ctx = fsc->fs_private;
1225 	struct file *file;
1226 	int err;
1227 	struct fuse_conn *fc;
1228 
1229 	err = -EINVAL;
1230 	file = fget(ctx->fd);
1231 	if (!file)
1232 		goto err;
1233 
1234 	/*
1235 	 * Require mount to happen from the same user namespace which
1236 	 * opened /dev/fuse to prevent potential attacks.
1237 	 */
1238 	if ((file->f_op != &fuse_dev_operations) ||
1239 	    (file->f_cred->user_ns != sb->s_user_ns))
1240 		goto err_fput;
1241 	ctx->fudptr = &file->private_data;
1242 
1243 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1244 	err = -ENOMEM;
1245 	if (!fc)
1246 		goto err_fput;
1247 
1248 	fuse_conn_init(fc, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1249 	fc->release = fuse_free_conn;
1250 	sb->s_fs_info = fc;
1251 
1252 	err = fuse_fill_super_common(sb, ctx);
1253 	if (err)
1254 		goto err_put_conn;
1255 	/*
1256 	 * atomic_dec_and_test() in fput() provides the necessary
1257 	 * memory barrier for file->private_data to be visible on all
1258 	 * CPUs after this
1259 	 */
1260 	fput(file);
1261 	fuse_send_init(get_fuse_conn_super(sb));
1262 	return 0;
1263 
1264  err_put_conn:
1265 	fuse_conn_put(fc);
1266 	sb->s_fs_info = NULL;
1267  err_fput:
1268 	fput(file);
1269  err:
1270 	return err;
1271 }
1272 
1273 static int fuse_get_tree(struct fs_context *fc)
1274 {
1275 	struct fuse_fs_context *ctx = fc->fs_private;
1276 
1277 	if (!ctx->fd_present || !ctx->rootmode_present ||
1278 	    !ctx->user_id_present || !ctx->group_id_present)
1279 		return -EINVAL;
1280 
1281 #ifdef CONFIG_BLOCK
1282 	if (ctx->is_bdev)
1283 		return get_tree_bdev(fc, fuse_fill_super);
1284 #endif
1285 
1286 	return get_tree_nodev(fc, fuse_fill_super);
1287 }
1288 
1289 static const struct fs_context_operations fuse_context_ops = {
1290 	.free		= fuse_free_fc,
1291 	.parse_param	= fuse_parse_param,
1292 	.get_tree	= fuse_get_tree,
1293 };
1294 
1295 /*
1296  * Set up the filesystem mount context.
1297  */
1298 static int fuse_init_fs_context(struct fs_context *fc)
1299 {
1300 	struct fuse_fs_context *ctx;
1301 
1302 	ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1303 	if (!ctx)
1304 		return -ENOMEM;
1305 
1306 	ctx->max_read = ~0;
1307 	ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1308 
1309 #ifdef CONFIG_BLOCK
1310 	if (fc->fs_type == &fuseblk_fs_type) {
1311 		ctx->is_bdev = true;
1312 		ctx->destroy = true;
1313 	}
1314 #endif
1315 
1316 	fc->fs_private = ctx;
1317 	fc->ops = &fuse_context_ops;
1318 	return 0;
1319 }
1320 
1321 static void fuse_sb_destroy(struct super_block *sb)
1322 {
1323 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1324 
1325 	if (fc) {
1326 		if (fc->destroy)
1327 			fuse_send_destroy(fc);
1328 
1329 		fuse_abort_conn(fc);
1330 		fuse_wait_aborted(fc);
1331 
1332 		down_write(&fc->killsb);
1333 		fc->sb = NULL;
1334 		up_write(&fc->killsb);
1335 	}
1336 }
1337 
1338 void fuse_kill_sb_anon(struct super_block *sb)
1339 {
1340 	fuse_sb_destroy(sb);
1341 	kill_anon_super(sb);
1342 }
1343 EXPORT_SYMBOL_GPL(fuse_kill_sb_anon);
1344 
1345 static struct file_system_type fuse_fs_type = {
1346 	.owner		= THIS_MODULE,
1347 	.name		= "fuse",
1348 	.fs_flags	= FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1349 	.init_fs_context = fuse_init_fs_context,
1350 	.parameters	= &fuse_fs_parameters,
1351 	.kill_sb	= fuse_kill_sb_anon,
1352 };
1353 MODULE_ALIAS_FS("fuse");
1354 
1355 #ifdef CONFIG_BLOCK
1356 static void fuse_kill_sb_blk(struct super_block *sb)
1357 {
1358 	fuse_sb_destroy(sb);
1359 	kill_block_super(sb);
1360 }
1361 
1362 static struct file_system_type fuseblk_fs_type = {
1363 	.owner		= THIS_MODULE,
1364 	.name		= "fuseblk",
1365 	.init_fs_context = fuse_init_fs_context,
1366 	.parameters	= &fuse_fs_parameters,
1367 	.kill_sb	= fuse_kill_sb_blk,
1368 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1369 };
1370 MODULE_ALIAS_FS("fuseblk");
1371 
1372 static inline int register_fuseblk(void)
1373 {
1374 	return register_filesystem(&fuseblk_fs_type);
1375 }
1376 
1377 static inline void unregister_fuseblk(void)
1378 {
1379 	unregister_filesystem(&fuseblk_fs_type);
1380 }
1381 #else
1382 static inline int register_fuseblk(void)
1383 {
1384 	return 0;
1385 }
1386 
1387 static inline void unregister_fuseblk(void)
1388 {
1389 }
1390 #endif
1391 
1392 static void fuse_inode_init_once(void *foo)
1393 {
1394 	struct inode *inode = foo;
1395 
1396 	inode_init_once(inode);
1397 }
1398 
1399 static int __init fuse_fs_init(void)
1400 {
1401 	int err;
1402 
1403 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
1404 			sizeof(struct fuse_inode), 0,
1405 			SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1406 			fuse_inode_init_once);
1407 	err = -ENOMEM;
1408 	if (!fuse_inode_cachep)
1409 		goto out;
1410 
1411 	err = register_fuseblk();
1412 	if (err)
1413 		goto out2;
1414 
1415 	err = register_filesystem(&fuse_fs_type);
1416 	if (err)
1417 		goto out3;
1418 
1419 	return 0;
1420 
1421  out3:
1422 	unregister_fuseblk();
1423  out2:
1424 	kmem_cache_destroy(fuse_inode_cachep);
1425  out:
1426 	return err;
1427 }
1428 
1429 static void fuse_fs_cleanup(void)
1430 {
1431 	unregister_filesystem(&fuse_fs_type);
1432 	unregister_fuseblk();
1433 
1434 	/*
1435 	 * Make sure all delayed rcu free inodes are flushed before we
1436 	 * destroy cache.
1437 	 */
1438 	rcu_barrier();
1439 	kmem_cache_destroy(fuse_inode_cachep);
1440 }
1441 
1442 static struct kobject *fuse_kobj;
1443 
1444 static int fuse_sysfs_init(void)
1445 {
1446 	int err;
1447 
1448 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1449 	if (!fuse_kobj) {
1450 		err = -ENOMEM;
1451 		goto out_err;
1452 	}
1453 
1454 	err = sysfs_create_mount_point(fuse_kobj, "connections");
1455 	if (err)
1456 		goto out_fuse_unregister;
1457 
1458 	return 0;
1459 
1460  out_fuse_unregister:
1461 	kobject_put(fuse_kobj);
1462  out_err:
1463 	return err;
1464 }
1465 
1466 static void fuse_sysfs_cleanup(void)
1467 {
1468 	sysfs_remove_mount_point(fuse_kobj, "connections");
1469 	kobject_put(fuse_kobj);
1470 }
1471 
1472 static int __init fuse_init(void)
1473 {
1474 	int res;
1475 
1476 	pr_info("init (API version %i.%i)\n",
1477 		FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1478 
1479 	INIT_LIST_HEAD(&fuse_conn_list);
1480 	res = fuse_fs_init();
1481 	if (res)
1482 		goto err;
1483 
1484 	res = fuse_dev_init();
1485 	if (res)
1486 		goto err_fs_cleanup;
1487 
1488 	res = fuse_sysfs_init();
1489 	if (res)
1490 		goto err_dev_cleanup;
1491 
1492 	res = fuse_ctl_init();
1493 	if (res)
1494 		goto err_sysfs_cleanup;
1495 
1496 	sanitize_global_limit(&max_user_bgreq);
1497 	sanitize_global_limit(&max_user_congthresh);
1498 
1499 	return 0;
1500 
1501  err_sysfs_cleanup:
1502 	fuse_sysfs_cleanup();
1503  err_dev_cleanup:
1504 	fuse_dev_cleanup();
1505  err_fs_cleanup:
1506 	fuse_fs_cleanup();
1507  err:
1508 	return res;
1509 }
1510 
1511 static void __exit fuse_exit(void)
1512 {
1513 	pr_debug("exit\n");
1514 
1515 	fuse_ctl_cleanup();
1516 	fuse_sysfs_cleanup();
1517 	fuse_fs_cleanup();
1518 	fuse_dev_cleanup();
1519 }
1520 
1521 module_init(fuse_init);
1522 module_exit(fuse_exit);
1523