xref: /openbmc/linux/fs/fuse/inode.c (revision 9123e3a7)
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_reconfigure(struct fs_context *fc)
125 {
126 	struct super_block *sb = fc->root->d_sb;
127 
128 	sync_filesystem(sb);
129 	if (fc->sb_flags & SB_MANDLOCK)
130 		return -EINVAL;
131 
132 	return 0;
133 }
134 
135 /*
136  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
137  * so that it will fit.
138  */
139 static ino_t fuse_squash_ino(u64 ino64)
140 {
141 	ino_t ino = (ino_t) ino64;
142 	if (sizeof(ino_t) < sizeof(u64))
143 		ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
144 	return ino;
145 }
146 
147 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
148 				   u64 attr_valid)
149 {
150 	struct fuse_conn *fc = get_fuse_conn(inode);
151 	struct fuse_inode *fi = get_fuse_inode(inode);
152 
153 	lockdep_assert_held(&fi->lock);
154 
155 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
156 	fi->i_time = attr_valid;
157 	WRITE_ONCE(fi->inval_mask, 0);
158 
159 	inode->i_ino     = fuse_squash_ino(attr->ino);
160 	inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
161 	set_nlink(inode, attr->nlink);
162 	inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
163 	inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
164 	inode->i_blocks  = attr->blocks;
165 	inode->i_atime.tv_sec   = attr->atime;
166 	inode->i_atime.tv_nsec  = attr->atimensec;
167 	/* mtime from server may be stale due to local buffered write */
168 	if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
169 		inode->i_mtime.tv_sec   = attr->mtime;
170 		inode->i_mtime.tv_nsec  = attr->mtimensec;
171 		inode->i_ctime.tv_sec   = attr->ctime;
172 		inode->i_ctime.tv_nsec  = attr->ctimensec;
173 	}
174 
175 	if (attr->blksize != 0)
176 		inode->i_blkbits = ilog2(attr->blksize);
177 	else
178 		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
179 
180 	/*
181 	 * Don't set the sticky bit in i_mode, unless we want the VFS
182 	 * to check permissions.  This prevents failures due to the
183 	 * check in may_delete().
184 	 */
185 	fi->orig_i_mode = inode->i_mode;
186 	if (!fc->default_permissions)
187 		inode->i_mode &= ~S_ISVTX;
188 
189 	fi->orig_ino = attr->ino;
190 }
191 
192 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
193 			    u64 attr_valid, u64 attr_version)
194 {
195 	struct fuse_conn *fc = get_fuse_conn(inode);
196 	struct fuse_inode *fi = get_fuse_inode(inode);
197 	bool is_wb = fc->writeback_cache;
198 	loff_t oldsize;
199 	struct timespec64 old_mtime;
200 
201 	spin_lock(&fi->lock);
202 	if ((attr_version != 0 && fi->attr_version > attr_version) ||
203 	    test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
204 		spin_unlock(&fi->lock);
205 		return;
206 	}
207 
208 	old_mtime = inode->i_mtime;
209 	fuse_change_attributes_common(inode, attr, attr_valid);
210 
211 	oldsize = inode->i_size;
212 	/*
213 	 * In case of writeback_cache enabled, the cached writes beyond EOF
214 	 * extend local i_size without keeping userspace server in sync. So,
215 	 * attr->size coming from server can be stale. We cannot trust it.
216 	 */
217 	if (!is_wb || !S_ISREG(inode->i_mode))
218 		i_size_write(inode, attr->size);
219 	spin_unlock(&fi->lock);
220 
221 	if (!is_wb && S_ISREG(inode->i_mode)) {
222 		bool inval = false;
223 
224 		if (oldsize != attr->size) {
225 			truncate_pagecache(inode, attr->size);
226 			if (!fc->explicit_inval_data)
227 				inval = true;
228 		} else if (fc->auto_inval_data) {
229 			struct timespec64 new_mtime = {
230 				.tv_sec = attr->mtime,
231 				.tv_nsec = attr->mtimensec,
232 			};
233 
234 			/*
235 			 * Auto inval mode also checks and invalidates if mtime
236 			 * has changed.
237 			 */
238 			if (!timespec64_equal(&old_mtime, &new_mtime))
239 				inval = true;
240 		}
241 
242 		if (inval)
243 			invalidate_inode_pages2(inode->i_mapping);
244 	}
245 }
246 
247 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
248 {
249 	inode->i_mode = attr->mode & S_IFMT;
250 	inode->i_size = attr->size;
251 	inode->i_mtime.tv_sec  = attr->mtime;
252 	inode->i_mtime.tv_nsec = attr->mtimensec;
253 	inode->i_ctime.tv_sec  = attr->ctime;
254 	inode->i_ctime.tv_nsec = attr->ctimensec;
255 	if (S_ISREG(inode->i_mode)) {
256 		fuse_init_common(inode);
257 		fuse_init_file_inode(inode);
258 	} else if (S_ISDIR(inode->i_mode))
259 		fuse_init_dir(inode);
260 	else if (S_ISLNK(inode->i_mode))
261 		fuse_init_symlink(inode);
262 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
263 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
264 		fuse_init_common(inode);
265 		init_special_inode(inode, inode->i_mode,
266 				   new_decode_dev(attr->rdev));
267 	} else
268 		BUG();
269 }
270 
271 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
272 {
273 	u64 nodeid = *(u64 *) _nodeidp;
274 	if (get_node_id(inode) == nodeid)
275 		return 1;
276 	else
277 		return 0;
278 }
279 
280 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
281 {
282 	u64 nodeid = *(u64 *) _nodeidp;
283 	get_fuse_inode(inode)->nodeid = nodeid;
284 	return 0;
285 }
286 
287 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
288 			int generation, struct fuse_attr *attr,
289 			u64 attr_valid, u64 attr_version)
290 {
291 	struct inode *inode;
292 	struct fuse_inode *fi;
293 	struct fuse_conn *fc = get_fuse_conn_super(sb);
294 
295  retry:
296 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
297 	if (!inode)
298 		return NULL;
299 
300 	if ((inode->i_state & I_NEW)) {
301 		inode->i_flags |= S_NOATIME;
302 		if (!fc->writeback_cache || !S_ISREG(attr->mode))
303 			inode->i_flags |= S_NOCMTIME;
304 		inode->i_generation = generation;
305 		fuse_init_inode(inode, attr);
306 		unlock_new_inode(inode);
307 	} else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
308 		/* Inode has changed type, any I/O on the old should fail */
309 		make_bad_inode(inode);
310 		iput(inode);
311 		goto retry;
312 	}
313 
314 	fi = get_fuse_inode(inode);
315 	spin_lock(&fi->lock);
316 	fi->nlookup++;
317 	spin_unlock(&fi->lock);
318 	fuse_change_attributes(inode, attr, attr_valid, attr_version);
319 
320 	return inode;
321 }
322 
323 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
324 			     loff_t offset, loff_t len)
325 {
326 	struct fuse_conn *fc = get_fuse_conn_super(sb);
327 	struct fuse_inode *fi;
328 	struct inode *inode;
329 	pgoff_t pg_start;
330 	pgoff_t pg_end;
331 
332 	inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
333 	if (!inode)
334 		return -ENOENT;
335 
336 	fi = get_fuse_inode(inode);
337 	spin_lock(&fi->lock);
338 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
339 	spin_unlock(&fi->lock);
340 
341 	fuse_invalidate_attr(inode);
342 	forget_all_cached_acls(inode);
343 	if (offset >= 0) {
344 		pg_start = offset >> PAGE_SHIFT;
345 		if (len <= 0)
346 			pg_end = -1;
347 		else
348 			pg_end = (offset + len - 1) >> PAGE_SHIFT;
349 		invalidate_inode_pages2_range(inode->i_mapping,
350 					      pg_start, pg_end);
351 	}
352 	iput(inode);
353 	return 0;
354 }
355 
356 bool fuse_lock_inode(struct inode *inode)
357 {
358 	bool locked = false;
359 
360 	if (!get_fuse_conn(inode)->parallel_dirops) {
361 		mutex_lock(&get_fuse_inode(inode)->mutex);
362 		locked = true;
363 	}
364 
365 	return locked;
366 }
367 
368 void fuse_unlock_inode(struct inode *inode, bool locked)
369 {
370 	if (locked)
371 		mutex_unlock(&get_fuse_inode(inode)->mutex);
372 }
373 
374 static void fuse_umount_begin(struct super_block *sb)
375 {
376 	struct fuse_conn *fc = get_fuse_conn_super(sb);
377 
378 	if (!fc->no_force_umount)
379 		fuse_abort_conn(fc);
380 }
381 
382 static void fuse_send_destroy(struct fuse_conn *fc)
383 {
384 	if (fc->conn_init) {
385 		FUSE_ARGS(args);
386 
387 		args.opcode = FUSE_DESTROY;
388 		args.force = true;
389 		args.nocreds = true;
390 		fuse_simple_request(fc, &args);
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.opcode = FUSE_STATFS;
436 	args.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_SOURCE,
448 	OPT_SUBTYPE,
449 	OPT_FD,
450 	OPT_ROOTMODE,
451 	OPT_USER_ID,
452 	OPT_GROUP_ID,
453 	OPT_DEFAULT_PERMISSIONS,
454 	OPT_ALLOW_OTHER,
455 	OPT_MAX_READ,
456 	OPT_BLKSIZE,
457 	OPT_ERR
458 };
459 
460 static const struct fs_parameter_spec fuse_fs_parameters[] = {
461 	fsparam_string	("source",		OPT_SOURCE),
462 	fsparam_u32	("fd",			OPT_FD),
463 	fsparam_u32oct	("rootmode",		OPT_ROOTMODE),
464 	fsparam_u32	("user_id",		OPT_USER_ID),
465 	fsparam_u32	("group_id",		OPT_GROUP_ID),
466 	fsparam_flag	("default_permissions",	OPT_DEFAULT_PERMISSIONS),
467 	fsparam_flag	("allow_other",		OPT_ALLOW_OTHER),
468 	fsparam_u32	("max_read",		OPT_MAX_READ),
469 	fsparam_u32	("blksize",		OPT_BLKSIZE),
470 	fsparam_string	("subtype",		OPT_SUBTYPE),
471 	{}
472 };
473 
474 static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param)
475 {
476 	struct fs_parse_result result;
477 	struct fuse_fs_context *ctx = fc->fs_private;
478 	int opt;
479 
480 	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
481 		/*
482 		 * Ignore options coming from mount(MS_REMOUNT) for backward
483 		 * compatibility.
484 		 */
485 		if (fc->oldapi)
486 			return 0;
487 
488 		return invalfc(fc, "No changes allowed in reconfigure");
489 	}
490 
491 	opt = fs_parse(fc, fuse_fs_parameters, param, &result);
492 	if (opt < 0)
493 		return opt;
494 
495 	switch (opt) {
496 	case OPT_SOURCE:
497 		if (fc->source)
498 			return invalfc(fc, "Multiple sources specified");
499 		fc->source = param->string;
500 		param->string = NULL;
501 		break;
502 
503 	case OPT_SUBTYPE:
504 		if (ctx->subtype)
505 			return invalfc(fc, "Multiple subtypes specified");
506 		ctx->subtype = param->string;
507 		param->string = NULL;
508 		return 0;
509 
510 	case OPT_FD:
511 		ctx->fd = result.uint_32;
512 		ctx->fd_present = true;
513 		break;
514 
515 	case OPT_ROOTMODE:
516 		if (!fuse_valid_type(result.uint_32))
517 			return invalfc(fc, "Invalid rootmode");
518 		ctx->rootmode = result.uint_32;
519 		ctx->rootmode_present = true;
520 		break;
521 
522 	case OPT_USER_ID:
523 		ctx->user_id = make_kuid(fc->user_ns, result.uint_32);
524 		if (!uid_valid(ctx->user_id))
525 			return invalfc(fc, "Invalid user_id");
526 		ctx->user_id_present = true;
527 		break;
528 
529 	case OPT_GROUP_ID:
530 		ctx->group_id = make_kgid(fc->user_ns, result.uint_32);
531 		if (!gid_valid(ctx->group_id))
532 			return invalfc(fc, "Invalid group_id");
533 		ctx->group_id_present = true;
534 		break;
535 
536 	case OPT_DEFAULT_PERMISSIONS:
537 		ctx->default_permissions = true;
538 		break;
539 
540 	case OPT_ALLOW_OTHER:
541 		ctx->allow_other = true;
542 		break;
543 
544 	case OPT_MAX_READ:
545 		ctx->max_read = result.uint_32;
546 		break;
547 
548 	case OPT_BLKSIZE:
549 		if (!ctx->is_bdev)
550 			return invalfc(fc, "blksize only supported for fuseblk");
551 		ctx->blksize = result.uint_32;
552 		break;
553 
554 	default:
555 		return -EINVAL;
556 	}
557 
558 	return 0;
559 }
560 
561 static void fuse_free_fc(struct fs_context *fc)
562 {
563 	struct fuse_fs_context *ctx = fc->fs_private;
564 
565 	if (ctx) {
566 		kfree(ctx->subtype);
567 		kfree(ctx);
568 	}
569 }
570 
571 static int fuse_show_options(struct seq_file *m, struct dentry *root)
572 {
573 	struct super_block *sb = root->d_sb;
574 	struct fuse_conn *fc = get_fuse_conn_super(sb);
575 
576 	if (fc->no_mount_options)
577 		return 0;
578 
579 	seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
580 	seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
581 	if (fc->default_permissions)
582 		seq_puts(m, ",default_permissions");
583 	if (fc->allow_other)
584 		seq_puts(m, ",allow_other");
585 	if (fc->max_read != ~0)
586 		seq_printf(m, ",max_read=%u", fc->max_read);
587 	if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
588 		seq_printf(m, ",blksize=%lu", sb->s_blocksize);
589 	return 0;
590 }
591 
592 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
593 			     const struct fuse_iqueue_ops *ops,
594 			     void *priv)
595 {
596 	memset(fiq, 0, sizeof(struct fuse_iqueue));
597 	spin_lock_init(&fiq->lock);
598 	init_waitqueue_head(&fiq->waitq);
599 	INIT_LIST_HEAD(&fiq->pending);
600 	INIT_LIST_HEAD(&fiq->interrupts);
601 	fiq->forget_list_tail = &fiq->forget_list_head;
602 	fiq->connected = 1;
603 	fiq->ops = ops;
604 	fiq->priv = priv;
605 }
606 
607 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
608 {
609 	unsigned int i;
610 
611 	spin_lock_init(&fpq->lock);
612 	for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
613 		INIT_LIST_HEAD(&fpq->processing[i]);
614 	INIT_LIST_HEAD(&fpq->io);
615 	fpq->connected = 1;
616 }
617 
618 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns,
619 		    const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
620 {
621 	memset(fc, 0, sizeof(*fc));
622 	spin_lock_init(&fc->lock);
623 	spin_lock_init(&fc->bg_lock);
624 	init_rwsem(&fc->killsb);
625 	refcount_set(&fc->count, 1);
626 	atomic_set(&fc->dev_count, 1);
627 	init_waitqueue_head(&fc->blocked_waitq);
628 	fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
629 	INIT_LIST_HEAD(&fc->bg_queue);
630 	INIT_LIST_HEAD(&fc->entry);
631 	INIT_LIST_HEAD(&fc->devices);
632 	atomic_set(&fc->num_waiting, 0);
633 	fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
634 	fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
635 	atomic64_set(&fc->khctr, 0);
636 	fc->polled_files = RB_ROOT;
637 	fc->blocked = 0;
638 	fc->initialized = 0;
639 	fc->connected = 1;
640 	atomic64_set(&fc->attr_version, 1);
641 	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
642 	fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
643 	fc->user_ns = get_user_ns(user_ns);
644 	fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
645 }
646 EXPORT_SYMBOL_GPL(fuse_conn_init);
647 
648 void fuse_conn_put(struct fuse_conn *fc)
649 {
650 	if (refcount_dec_and_test(&fc->count)) {
651 		struct fuse_iqueue *fiq = &fc->iq;
652 
653 		if (fiq->ops->release)
654 			fiq->ops->release(fiq);
655 		put_pid_ns(fc->pid_ns);
656 		put_user_ns(fc->user_ns);
657 		fc->release(fc);
658 	}
659 }
660 EXPORT_SYMBOL_GPL(fuse_conn_put);
661 
662 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
663 {
664 	refcount_inc(&fc->count);
665 	return fc;
666 }
667 EXPORT_SYMBOL_GPL(fuse_conn_get);
668 
669 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
670 {
671 	struct fuse_attr attr;
672 	memset(&attr, 0, sizeof(attr));
673 
674 	attr.mode = mode;
675 	attr.ino = FUSE_ROOT_ID;
676 	attr.nlink = 1;
677 	return fuse_iget(sb, 1, 0, &attr, 0, 0);
678 }
679 
680 struct fuse_inode_handle {
681 	u64 nodeid;
682 	u32 generation;
683 };
684 
685 static struct dentry *fuse_get_dentry(struct super_block *sb,
686 				      struct fuse_inode_handle *handle)
687 {
688 	struct fuse_conn *fc = get_fuse_conn_super(sb);
689 	struct inode *inode;
690 	struct dentry *entry;
691 	int err = -ESTALE;
692 
693 	if (handle->nodeid == 0)
694 		goto out_err;
695 
696 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
697 	if (!inode) {
698 		struct fuse_entry_out outarg;
699 		const struct qstr name = QSTR_INIT(".", 1);
700 
701 		if (!fc->export_support)
702 			goto out_err;
703 
704 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
705 				       &inode);
706 		if (err && err != -ENOENT)
707 			goto out_err;
708 		if (err || !inode) {
709 			err = -ESTALE;
710 			goto out_err;
711 		}
712 		err = -EIO;
713 		if (get_node_id(inode) != handle->nodeid)
714 			goto out_iput;
715 	}
716 	err = -ESTALE;
717 	if (inode->i_generation != handle->generation)
718 		goto out_iput;
719 
720 	entry = d_obtain_alias(inode);
721 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
722 		fuse_invalidate_entry_cache(entry);
723 
724 	return entry;
725 
726  out_iput:
727 	iput(inode);
728  out_err:
729 	return ERR_PTR(err);
730 }
731 
732 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
733 			   struct inode *parent)
734 {
735 	int len = parent ? 6 : 3;
736 	u64 nodeid;
737 	u32 generation;
738 
739 	if (*max_len < len) {
740 		*max_len = len;
741 		return  FILEID_INVALID;
742 	}
743 
744 	nodeid = get_fuse_inode(inode)->nodeid;
745 	generation = inode->i_generation;
746 
747 	fh[0] = (u32)(nodeid >> 32);
748 	fh[1] = (u32)(nodeid & 0xffffffff);
749 	fh[2] = generation;
750 
751 	if (parent) {
752 		nodeid = get_fuse_inode(parent)->nodeid;
753 		generation = parent->i_generation;
754 
755 		fh[3] = (u32)(nodeid >> 32);
756 		fh[4] = (u32)(nodeid & 0xffffffff);
757 		fh[5] = generation;
758 	}
759 
760 	*max_len = len;
761 	return parent ? 0x82 : 0x81;
762 }
763 
764 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
765 		struct fid *fid, int fh_len, int fh_type)
766 {
767 	struct fuse_inode_handle handle;
768 
769 	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
770 		return NULL;
771 
772 	handle.nodeid = (u64) fid->raw[0] << 32;
773 	handle.nodeid |= (u64) fid->raw[1];
774 	handle.generation = fid->raw[2];
775 	return fuse_get_dentry(sb, &handle);
776 }
777 
778 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
779 		struct fid *fid, int fh_len, int fh_type)
780 {
781 	struct fuse_inode_handle parent;
782 
783 	if (fh_type != 0x82 || fh_len < 6)
784 		return NULL;
785 
786 	parent.nodeid = (u64) fid->raw[3] << 32;
787 	parent.nodeid |= (u64) fid->raw[4];
788 	parent.generation = fid->raw[5];
789 	return fuse_get_dentry(sb, &parent);
790 }
791 
792 static struct dentry *fuse_get_parent(struct dentry *child)
793 {
794 	struct inode *child_inode = d_inode(child);
795 	struct fuse_conn *fc = get_fuse_conn(child_inode);
796 	struct inode *inode;
797 	struct dentry *parent;
798 	struct fuse_entry_out outarg;
799 	const struct qstr name = QSTR_INIT("..", 2);
800 	int err;
801 
802 	if (!fc->export_support)
803 		return ERR_PTR(-ESTALE);
804 
805 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
806 			       &name, &outarg, &inode);
807 	if (err) {
808 		if (err == -ENOENT)
809 			return ERR_PTR(-ESTALE);
810 		return ERR_PTR(err);
811 	}
812 
813 	parent = d_obtain_alias(inode);
814 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
815 		fuse_invalidate_entry_cache(parent);
816 
817 	return parent;
818 }
819 
820 static const struct export_operations fuse_export_operations = {
821 	.fh_to_dentry	= fuse_fh_to_dentry,
822 	.fh_to_parent	= fuse_fh_to_parent,
823 	.encode_fh	= fuse_encode_fh,
824 	.get_parent	= fuse_get_parent,
825 };
826 
827 static const struct super_operations fuse_super_operations = {
828 	.alloc_inode    = fuse_alloc_inode,
829 	.free_inode     = fuse_free_inode,
830 	.evict_inode	= fuse_evict_inode,
831 	.write_inode	= fuse_write_inode,
832 	.drop_inode	= generic_delete_inode,
833 	.put_super	= fuse_put_super,
834 	.umount_begin	= fuse_umount_begin,
835 	.statfs		= fuse_statfs,
836 	.show_options	= fuse_show_options,
837 };
838 
839 static void sanitize_global_limit(unsigned *limit)
840 {
841 	/*
842 	 * The default maximum number of async requests is calculated to consume
843 	 * 1/2^13 of the total memory, assuming 392 bytes per request.
844 	 */
845 	if (*limit == 0)
846 		*limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
847 
848 	if (*limit >= 1 << 16)
849 		*limit = (1 << 16) - 1;
850 }
851 
852 static int set_global_limit(const char *val, const struct kernel_param *kp)
853 {
854 	int rv;
855 
856 	rv = param_set_uint(val, kp);
857 	if (rv)
858 		return rv;
859 
860 	sanitize_global_limit((unsigned *)kp->arg);
861 
862 	return 0;
863 }
864 
865 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
866 {
867 	int cap_sys_admin = capable(CAP_SYS_ADMIN);
868 
869 	if (arg->minor < 13)
870 		return;
871 
872 	sanitize_global_limit(&max_user_bgreq);
873 	sanitize_global_limit(&max_user_congthresh);
874 
875 	spin_lock(&fc->bg_lock);
876 	if (arg->max_background) {
877 		fc->max_background = arg->max_background;
878 
879 		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
880 			fc->max_background = max_user_bgreq;
881 	}
882 	if (arg->congestion_threshold) {
883 		fc->congestion_threshold = arg->congestion_threshold;
884 
885 		if (!cap_sys_admin &&
886 		    fc->congestion_threshold > max_user_congthresh)
887 			fc->congestion_threshold = max_user_congthresh;
888 	}
889 	spin_unlock(&fc->bg_lock);
890 }
891 
892 struct fuse_init_args {
893 	struct fuse_args args;
894 	struct fuse_init_in in;
895 	struct fuse_init_out out;
896 };
897 
898 static void process_init_reply(struct fuse_conn *fc, struct fuse_args *args,
899 			       int error)
900 {
901 	struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
902 	struct fuse_init_out *arg = &ia->out;
903 
904 	if (error || arg->major != FUSE_KERNEL_VERSION)
905 		fc->conn_error = 1;
906 	else {
907 		unsigned long ra_pages;
908 
909 		process_init_limits(fc, arg);
910 
911 		if (arg->minor >= 6) {
912 			ra_pages = arg->max_readahead / PAGE_SIZE;
913 			if (arg->flags & FUSE_ASYNC_READ)
914 				fc->async_read = 1;
915 			if (!(arg->flags & FUSE_POSIX_LOCKS))
916 				fc->no_lock = 1;
917 			if (arg->minor >= 17) {
918 				if (!(arg->flags & FUSE_FLOCK_LOCKS))
919 					fc->no_flock = 1;
920 			} else {
921 				if (!(arg->flags & FUSE_POSIX_LOCKS))
922 					fc->no_flock = 1;
923 			}
924 			if (arg->flags & FUSE_ATOMIC_O_TRUNC)
925 				fc->atomic_o_trunc = 1;
926 			if (arg->minor >= 9) {
927 				/* LOOKUP has dependency on proto version */
928 				if (arg->flags & FUSE_EXPORT_SUPPORT)
929 					fc->export_support = 1;
930 			}
931 			if (arg->flags & FUSE_BIG_WRITES)
932 				fc->big_writes = 1;
933 			if (arg->flags & FUSE_DONT_MASK)
934 				fc->dont_mask = 1;
935 			if (arg->flags & FUSE_AUTO_INVAL_DATA)
936 				fc->auto_inval_data = 1;
937 			else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
938 				fc->explicit_inval_data = 1;
939 			if (arg->flags & FUSE_DO_READDIRPLUS) {
940 				fc->do_readdirplus = 1;
941 				if (arg->flags & FUSE_READDIRPLUS_AUTO)
942 					fc->readdirplus_auto = 1;
943 			}
944 			if (arg->flags & FUSE_ASYNC_DIO)
945 				fc->async_dio = 1;
946 			if (arg->flags & FUSE_WRITEBACK_CACHE)
947 				fc->writeback_cache = 1;
948 			if (arg->flags & FUSE_PARALLEL_DIROPS)
949 				fc->parallel_dirops = 1;
950 			if (arg->flags & FUSE_HANDLE_KILLPRIV)
951 				fc->handle_killpriv = 1;
952 			if (arg->time_gran && arg->time_gran <= 1000000000)
953 				fc->sb->s_time_gran = arg->time_gran;
954 			if ((arg->flags & FUSE_POSIX_ACL)) {
955 				fc->default_permissions = 1;
956 				fc->posix_acl = 1;
957 				fc->sb->s_xattr = fuse_acl_xattr_handlers;
958 			}
959 			if (arg->flags & FUSE_CACHE_SYMLINKS)
960 				fc->cache_symlinks = 1;
961 			if (arg->flags & FUSE_ABORT_ERROR)
962 				fc->abort_err = 1;
963 			if (arg->flags & FUSE_MAX_PAGES) {
964 				fc->max_pages =
965 					min_t(unsigned int, FUSE_MAX_MAX_PAGES,
966 					max_t(unsigned int, arg->max_pages, 1));
967 			}
968 		} else {
969 			ra_pages = fc->max_read / PAGE_SIZE;
970 			fc->no_lock = 1;
971 			fc->no_flock = 1;
972 		}
973 
974 		fc->sb->s_bdi->ra_pages =
975 				min(fc->sb->s_bdi->ra_pages, ra_pages);
976 		fc->minor = arg->minor;
977 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
978 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
979 		fc->conn_init = 1;
980 	}
981 	kfree(ia);
982 
983 	fuse_set_initialized(fc);
984 	wake_up_all(&fc->blocked_waitq);
985 }
986 
987 void fuse_send_init(struct fuse_conn *fc)
988 {
989 	struct fuse_init_args *ia;
990 
991 	ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
992 
993 	ia->in.major = FUSE_KERNEL_VERSION;
994 	ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
995 	ia->in.max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
996 	ia->in.flags |=
997 		FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
998 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
999 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1000 		FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1001 		FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1002 		FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1003 		FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1004 		FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1005 		FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA;
1006 	ia->args.opcode = FUSE_INIT;
1007 	ia->args.in_numargs = 1;
1008 	ia->args.in_args[0].size = sizeof(ia->in);
1009 	ia->args.in_args[0].value = &ia->in;
1010 	ia->args.out_numargs = 1;
1011 	/* Variable length argument used for backward compatibility
1012 	   with interface version < 7.5.  Rest of init_out is zeroed
1013 	   by do_get_request(), so a short reply is not a problem */
1014 	ia->args.out_argvar = true;
1015 	ia->args.out_args[0].size = sizeof(ia->out);
1016 	ia->args.out_args[0].value = &ia->out;
1017 	ia->args.force = true;
1018 	ia->args.nocreds = true;
1019 	ia->args.end = process_init_reply;
1020 
1021 	if (fuse_simple_background(fc, &ia->args, GFP_KERNEL) != 0)
1022 		process_init_reply(fc, &ia->args, -ENOTCONN);
1023 }
1024 EXPORT_SYMBOL_GPL(fuse_send_init);
1025 
1026 void fuse_free_conn(struct fuse_conn *fc)
1027 {
1028 	WARN_ON(!list_empty(&fc->devices));
1029 	kfree_rcu(fc, rcu);
1030 }
1031 EXPORT_SYMBOL_GPL(fuse_free_conn);
1032 
1033 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1034 {
1035 	int err;
1036 	char *suffix = "";
1037 
1038 	if (sb->s_bdev) {
1039 		suffix = "-fuseblk";
1040 		/*
1041 		 * sb->s_bdi points to blkdev's bdi however we want to redirect
1042 		 * it to our private bdi...
1043 		 */
1044 		bdi_put(sb->s_bdi);
1045 		sb->s_bdi = &noop_backing_dev_info;
1046 	}
1047 	err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1048 				   MINOR(fc->dev), suffix);
1049 	if (err)
1050 		return err;
1051 
1052 	sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1053 	/* fuse does it's own writeback accounting */
1054 	sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1055 
1056 	/*
1057 	 * For a single fuse filesystem use max 1% of dirty +
1058 	 * writeback threshold.
1059 	 *
1060 	 * This gives about 1M of write buffer for memory maps on a
1061 	 * machine with 1G and 10% dirty_ratio, which should be more
1062 	 * than enough.
1063 	 *
1064 	 * Privileged users can raise it by writing to
1065 	 *
1066 	 *    /sys/class/bdi/<bdi>/max_ratio
1067 	 */
1068 	bdi_set_max_ratio(sb->s_bdi, 1);
1069 
1070 	return 0;
1071 }
1072 
1073 struct fuse_dev *fuse_dev_alloc(void)
1074 {
1075 	struct fuse_dev *fud;
1076 	struct list_head *pq;
1077 
1078 	fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1079 	if (!fud)
1080 		return NULL;
1081 
1082 	pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1083 	if (!pq) {
1084 		kfree(fud);
1085 		return NULL;
1086 	}
1087 
1088 	fud->pq.processing = pq;
1089 	fuse_pqueue_init(&fud->pq);
1090 
1091 	return fud;
1092 }
1093 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1094 
1095 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1096 {
1097 	fud->fc = fuse_conn_get(fc);
1098 	spin_lock(&fc->lock);
1099 	list_add_tail(&fud->entry, &fc->devices);
1100 	spin_unlock(&fc->lock);
1101 }
1102 EXPORT_SYMBOL_GPL(fuse_dev_install);
1103 
1104 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1105 {
1106 	struct fuse_dev *fud;
1107 
1108 	fud = fuse_dev_alloc();
1109 	if (!fud)
1110 		return NULL;
1111 
1112 	fuse_dev_install(fud, fc);
1113 	return fud;
1114 }
1115 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1116 
1117 void fuse_dev_free(struct fuse_dev *fud)
1118 {
1119 	struct fuse_conn *fc = fud->fc;
1120 
1121 	if (fc) {
1122 		spin_lock(&fc->lock);
1123 		list_del(&fud->entry);
1124 		spin_unlock(&fc->lock);
1125 
1126 		fuse_conn_put(fc);
1127 	}
1128 	kfree(fud->pq.processing);
1129 	kfree(fud);
1130 }
1131 EXPORT_SYMBOL_GPL(fuse_dev_free);
1132 
1133 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1134 {
1135 	struct fuse_dev *fud = NULL;
1136 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1137 	struct inode *root;
1138 	struct dentry *root_dentry;
1139 	int err;
1140 
1141 	err = -EINVAL;
1142 	if (sb->s_flags & SB_MANDLOCK)
1143 		goto err;
1144 
1145 	sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1146 
1147 	if (ctx->is_bdev) {
1148 #ifdef CONFIG_BLOCK
1149 		err = -EINVAL;
1150 		if (!sb_set_blocksize(sb, ctx->blksize))
1151 			goto err;
1152 #endif
1153 	} else {
1154 		sb->s_blocksize = PAGE_SIZE;
1155 		sb->s_blocksize_bits = PAGE_SHIFT;
1156 	}
1157 
1158 	sb->s_subtype = ctx->subtype;
1159 	ctx->subtype = NULL;
1160 	sb->s_magic = FUSE_SUPER_MAGIC;
1161 	sb->s_op = &fuse_super_operations;
1162 	sb->s_xattr = fuse_xattr_handlers;
1163 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1164 	sb->s_time_gran = 1;
1165 	sb->s_export_op = &fuse_export_operations;
1166 	sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1167 	if (sb->s_user_ns != &init_user_ns)
1168 		sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1169 
1170 	/*
1171 	 * If we are not in the initial user namespace posix
1172 	 * acls must be translated.
1173 	 */
1174 	if (sb->s_user_ns != &init_user_ns)
1175 		sb->s_xattr = fuse_no_acl_xattr_handlers;
1176 
1177 	if (ctx->fudptr) {
1178 		err = -ENOMEM;
1179 		fud = fuse_dev_alloc_install(fc);
1180 		if (!fud)
1181 			goto err;
1182 	}
1183 
1184 	fc->dev = sb->s_dev;
1185 	fc->sb = sb;
1186 	err = fuse_bdi_init(fc, sb);
1187 	if (err)
1188 		goto err_dev_free;
1189 
1190 	/* Handle umasking inside the fuse code */
1191 	if (sb->s_flags & SB_POSIXACL)
1192 		fc->dont_mask = 1;
1193 	sb->s_flags |= SB_POSIXACL;
1194 
1195 	fc->default_permissions = ctx->default_permissions;
1196 	fc->allow_other = ctx->allow_other;
1197 	fc->user_id = ctx->user_id;
1198 	fc->group_id = ctx->group_id;
1199 	fc->max_read = max_t(unsigned, 4096, ctx->max_read);
1200 	fc->destroy = ctx->destroy;
1201 	fc->no_control = ctx->no_control;
1202 	fc->no_force_umount = ctx->no_force_umount;
1203 	fc->no_mount_options = ctx->no_mount_options;
1204 
1205 	err = -ENOMEM;
1206 	root = fuse_get_root_inode(sb, ctx->rootmode);
1207 	sb->s_d_op = &fuse_root_dentry_operations;
1208 	root_dentry = d_make_root(root);
1209 	if (!root_dentry)
1210 		goto err_dev_free;
1211 	/* Root dentry doesn't have .d_revalidate */
1212 	sb->s_d_op = &fuse_dentry_operations;
1213 
1214 	mutex_lock(&fuse_mutex);
1215 	err = -EINVAL;
1216 	if (ctx->fudptr && *ctx->fudptr)
1217 		goto err_unlock;
1218 
1219 	err = fuse_ctl_add_conn(fc);
1220 	if (err)
1221 		goto err_unlock;
1222 
1223 	list_add_tail(&fc->entry, &fuse_conn_list);
1224 	sb->s_root = root_dentry;
1225 	if (ctx->fudptr)
1226 		*ctx->fudptr = fud;
1227 	mutex_unlock(&fuse_mutex);
1228 	return 0;
1229 
1230  err_unlock:
1231 	mutex_unlock(&fuse_mutex);
1232 	dput(root_dentry);
1233  err_dev_free:
1234 	if (fud)
1235 		fuse_dev_free(fud);
1236  err:
1237 	return err;
1238 }
1239 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1240 
1241 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1242 {
1243 	struct fuse_fs_context *ctx = fsc->fs_private;
1244 	struct file *file;
1245 	int err;
1246 	struct fuse_conn *fc;
1247 
1248 	err = -EINVAL;
1249 	file = fget(ctx->fd);
1250 	if (!file)
1251 		goto err;
1252 
1253 	/*
1254 	 * Require mount to happen from the same user namespace which
1255 	 * opened /dev/fuse to prevent potential attacks.
1256 	 */
1257 	if ((file->f_op != &fuse_dev_operations) ||
1258 	    (file->f_cred->user_ns != sb->s_user_ns))
1259 		goto err_fput;
1260 	ctx->fudptr = &file->private_data;
1261 
1262 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1263 	err = -ENOMEM;
1264 	if (!fc)
1265 		goto err_fput;
1266 
1267 	fuse_conn_init(fc, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1268 	fc->release = fuse_free_conn;
1269 	sb->s_fs_info = fc;
1270 
1271 	err = fuse_fill_super_common(sb, ctx);
1272 	if (err)
1273 		goto err_put_conn;
1274 	/*
1275 	 * atomic_dec_and_test() in fput() provides the necessary
1276 	 * memory barrier for file->private_data to be visible on all
1277 	 * CPUs after this
1278 	 */
1279 	fput(file);
1280 	fuse_send_init(get_fuse_conn_super(sb));
1281 	return 0;
1282 
1283  err_put_conn:
1284 	fuse_conn_put(fc);
1285 	sb->s_fs_info = NULL;
1286  err_fput:
1287 	fput(file);
1288  err:
1289 	return err;
1290 }
1291 
1292 static int fuse_get_tree(struct fs_context *fc)
1293 {
1294 	struct fuse_fs_context *ctx = fc->fs_private;
1295 
1296 	if (!ctx->fd_present || !ctx->rootmode_present ||
1297 	    !ctx->user_id_present || !ctx->group_id_present)
1298 		return -EINVAL;
1299 
1300 #ifdef CONFIG_BLOCK
1301 	if (ctx->is_bdev)
1302 		return get_tree_bdev(fc, fuse_fill_super);
1303 #endif
1304 
1305 	return get_tree_nodev(fc, fuse_fill_super);
1306 }
1307 
1308 static const struct fs_context_operations fuse_context_ops = {
1309 	.free		= fuse_free_fc,
1310 	.parse_param	= fuse_parse_param,
1311 	.reconfigure	= fuse_reconfigure,
1312 	.get_tree	= fuse_get_tree,
1313 };
1314 
1315 /*
1316  * Set up the filesystem mount context.
1317  */
1318 static int fuse_init_fs_context(struct fs_context *fc)
1319 {
1320 	struct fuse_fs_context *ctx;
1321 
1322 	ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1323 	if (!ctx)
1324 		return -ENOMEM;
1325 
1326 	ctx->max_read = ~0;
1327 	ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1328 
1329 #ifdef CONFIG_BLOCK
1330 	if (fc->fs_type == &fuseblk_fs_type) {
1331 		ctx->is_bdev = true;
1332 		ctx->destroy = true;
1333 	}
1334 #endif
1335 
1336 	fc->fs_private = ctx;
1337 	fc->ops = &fuse_context_ops;
1338 	return 0;
1339 }
1340 
1341 static void fuse_sb_destroy(struct super_block *sb)
1342 {
1343 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1344 
1345 	if (fc) {
1346 		if (fc->destroy)
1347 			fuse_send_destroy(fc);
1348 
1349 		fuse_abort_conn(fc);
1350 		fuse_wait_aborted(fc);
1351 
1352 		down_write(&fc->killsb);
1353 		fc->sb = NULL;
1354 		up_write(&fc->killsb);
1355 	}
1356 }
1357 
1358 void fuse_kill_sb_anon(struct super_block *sb)
1359 {
1360 	fuse_sb_destroy(sb);
1361 	kill_anon_super(sb);
1362 }
1363 EXPORT_SYMBOL_GPL(fuse_kill_sb_anon);
1364 
1365 static struct file_system_type fuse_fs_type = {
1366 	.owner		= THIS_MODULE,
1367 	.name		= "fuse",
1368 	.fs_flags	= FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1369 	.init_fs_context = fuse_init_fs_context,
1370 	.parameters	= fuse_fs_parameters,
1371 	.kill_sb	= fuse_kill_sb_anon,
1372 };
1373 MODULE_ALIAS_FS("fuse");
1374 
1375 #ifdef CONFIG_BLOCK
1376 static void fuse_kill_sb_blk(struct super_block *sb)
1377 {
1378 	fuse_sb_destroy(sb);
1379 	kill_block_super(sb);
1380 }
1381 
1382 static struct file_system_type fuseblk_fs_type = {
1383 	.owner		= THIS_MODULE,
1384 	.name		= "fuseblk",
1385 	.init_fs_context = fuse_init_fs_context,
1386 	.parameters	= fuse_fs_parameters,
1387 	.kill_sb	= fuse_kill_sb_blk,
1388 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1389 };
1390 MODULE_ALIAS_FS("fuseblk");
1391 
1392 static inline int register_fuseblk(void)
1393 {
1394 	return register_filesystem(&fuseblk_fs_type);
1395 }
1396 
1397 static inline void unregister_fuseblk(void)
1398 {
1399 	unregister_filesystem(&fuseblk_fs_type);
1400 }
1401 #else
1402 static inline int register_fuseblk(void)
1403 {
1404 	return 0;
1405 }
1406 
1407 static inline void unregister_fuseblk(void)
1408 {
1409 }
1410 #endif
1411 
1412 static void fuse_inode_init_once(void *foo)
1413 {
1414 	struct inode *inode = foo;
1415 
1416 	inode_init_once(inode);
1417 }
1418 
1419 static int __init fuse_fs_init(void)
1420 {
1421 	int err;
1422 
1423 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
1424 			sizeof(struct fuse_inode), 0,
1425 			SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1426 			fuse_inode_init_once);
1427 	err = -ENOMEM;
1428 	if (!fuse_inode_cachep)
1429 		goto out;
1430 
1431 	err = register_fuseblk();
1432 	if (err)
1433 		goto out2;
1434 
1435 	err = register_filesystem(&fuse_fs_type);
1436 	if (err)
1437 		goto out3;
1438 
1439 	return 0;
1440 
1441  out3:
1442 	unregister_fuseblk();
1443  out2:
1444 	kmem_cache_destroy(fuse_inode_cachep);
1445  out:
1446 	return err;
1447 }
1448 
1449 static void fuse_fs_cleanup(void)
1450 {
1451 	unregister_filesystem(&fuse_fs_type);
1452 	unregister_fuseblk();
1453 
1454 	/*
1455 	 * Make sure all delayed rcu free inodes are flushed before we
1456 	 * destroy cache.
1457 	 */
1458 	rcu_barrier();
1459 	kmem_cache_destroy(fuse_inode_cachep);
1460 }
1461 
1462 static struct kobject *fuse_kobj;
1463 
1464 static int fuse_sysfs_init(void)
1465 {
1466 	int err;
1467 
1468 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1469 	if (!fuse_kobj) {
1470 		err = -ENOMEM;
1471 		goto out_err;
1472 	}
1473 
1474 	err = sysfs_create_mount_point(fuse_kobj, "connections");
1475 	if (err)
1476 		goto out_fuse_unregister;
1477 
1478 	return 0;
1479 
1480  out_fuse_unregister:
1481 	kobject_put(fuse_kobj);
1482  out_err:
1483 	return err;
1484 }
1485 
1486 static void fuse_sysfs_cleanup(void)
1487 {
1488 	sysfs_remove_mount_point(fuse_kobj, "connections");
1489 	kobject_put(fuse_kobj);
1490 }
1491 
1492 static int __init fuse_init(void)
1493 {
1494 	int res;
1495 
1496 	pr_info("init (API version %i.%i)\n",
1497 		FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1498 
1499 	INIT_LIST_HEAD(&fuse_conn_list);
1500 	res = fuse_fs_init();
1501 	if (res)
1502 		goto err;
1503 
1504 	res = fuse_dev_init();
1505 	if (res)
1506 		goto err_fs_cleanup;
1507 
1508 	res = fuse_sysfs_init();
1509 	if (res)
1510 		goto err_dev_cleanup;
1511 
1512 	res = fuse_ctl_init();
1513 	if (res)
1514 		goto err_sysfs_cleanup;
1515 
1516 	sanitize_global_limit(&max_user_bgreq);
1517 	sanitize_global_limit(&max_user_congthresh);
1518 
1519 	return 0;
1520 
1521  err_sysfs_cleanup:
1522 	fuse_sysfs_cleanup();
1523  err_dev_cleanup:
1524 	fuse_dev_cleanup();
1525  err_fs_cleanup:
1526 	fuse_fs_cleanup();
1527  err:
1528 	return res;
1529 }
1530 
1531 static void __exit fuse_exit(void)
1532 {
1533 	pr_debug("exit\n");
1534 
1535 	fuse_ctl_cleanup();
1536 	fuse_sysfs_cleanup();
1537 	fuse_fs_cleanup();
1538 	fuse_dev_cleanup();
1539 }
1540 
1541 module_init(fuse_init);
1542 module_exit(fuse_exit);
1543