xref: /openbmc/linux/fs/fuse/inode.c (revision 8684014d)
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 
24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
25 MODULE_DESCRIPTION("Filesystem in Userspace");
26 MODULE_LICENSE("GPL");
27 
28 static struct kmem_cache *fuse_inode_cachep;
29 struct list_head fuse_conn_list;
30 DEFINE_MUTEX(fuse_mutex);
31 
32 static int set_global_limit(const char *val, struct kernel_param *kp);
33 
34 unsigned max_user_bgreq;
35 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
36 		  &max_user_bgreq, 0644);
37 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
38 MODULE_PARM_DESC(max_user_bgreq,
39  "Global limit for the maximum number of backgrounded requests an "
40  "unprivileged user can set");
41 
42 unsigned max_user_congthresh;
43 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
44 		  &max_user_congthresh, 0644);
45 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
46 MODULE_PARM_DESC(max_user_congthresh,
47  "Global limit for the maximum congestion threshold an "
48  "unprivileged user can set");
49 
50 #define FUSE_SUPER_MAGIC 0x65735546
51 
52 #define FUSE_DEFAULT_BLKSIZE 512
53 
54 /** Maximum number of outstanding background requests */
55 #define FUSE_DEFAULT_MAX_BACKGROUND 12
56 
57 /** Congestion starts at 75% of maximum */
58 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
59 
60 struct fuse_mount_data {
61 	int fd;
62 	unsigned rootmode;
63 	kuid_t user_id;
64 	kgid_t group_id;
65 	unsigned fd_present:1;
66 	unsigned rootmode_present:1;
67 	unsigned user_id_present:1;
68 	unsigned group_id_present:1;
69 	unsigned flags;
70 	unsigned max_read;
71 	unsigned blksize;
72 };
73 
74 struct fuse_forget_link *fuse_alloc_forget(void)
75 {
76 	return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
77 }
78 
79 static struct inode *fuse_alloc_inode(struct super_block *sb)
80 {
81 	struct inode *inode;
82 	struct fuse_inode *fi;
83 
84 	inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
85 	if (!inode)
86 		return NULL;
87 
88 	fi = get_fuse_inode(inode);
89 	fi->i_time = 0;
90 	fi->nodeid = 0;
91 	fi->nlookup = 0;
92 	fi->attr_version = 0;
93 	fi->writectr = 0;
94 	fi->orig_ino = 0;
95 	fi->state = 0;
96 	INIT_LIST_HEAD(&fi->write_files);
97 	INIT_LIST_HEAD(&fi->queued_writes);
98 	INIT_LIST_HEAD(&fi->writepages);
99 	init_waitqueue_head(&fi->page_waitq);
100 	fi->forget = fuse_alloc_forget();
101 	if (!fi->forget) {
102 		kmem_cache_free(fuse_inode_cachep, inode);
103 		return NULL;
104 	}
105 
106 	return inode;
107 }
108 
109 static void fuse_i_callback(struct rcu_head *head)
110 {
111 	struct inode *inode = container_of(head, struct inode, i_rcu);
112 	kmem_cache_free(fuse_inode_cachep, inode);
113 }
114 
115 static void fuse_destroy_inode(struct inode *inode)
116 {
117 	struct fuse_inode *fi = get_fuse_inode(inode);
118 	BUG_ON(!list_empty(&fi->write_files));
119 	BUG_ON(!list_empty(&fi->queued_writes));
120 	kfree(fi->forget);
121 	call_rcu(&inode->i_rcu, fuse_i_callback);
122 }
123 
124 static void fuse_evict_inode(struct inode *inode)
125 {
126 	truncate_inode_pages_final(&inode->i_data);
127 	clear_inode(inode);
128 	if (inode->i_sb->s_flags & MS_ACTIVE) {
129 		struct fuse_conn *fc = get_fuse_conn(inode);
130 		struct fuse_inode *fi = get_fuse_inode(inode);
131 		fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
132 		fi->forget = NULL;
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 & MS_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 	fi->attr_version = ++fc->attr_version;
164 	fi->i_time = attr_valid;
165 
166 	inode->i_ino     = fuse_squash_ino(attr->ino);
167 	inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
168 	set_nlink(inode, attr->nlink);
169 	inode->i_uid     = make_kuid(&init_user_ns, attr->uid);
170 	inode->i_gid     = make_kgid(&init_user_ns, attr->gid);
171 	inode->i_blocks  = attr->blocks;
172 	inode->i_atime.tv_sec   = attr->atime;
173 	inode->i_atime.tv_nsec  = attr->atimensec;
174 	/* mtime from server may be stale due to local buffered write */
175 	if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
176 		inode->i_mtime.tv_sec   = attr->mtime;
177 		inode->i_mtime.tv_nsec  = attr->mtimensec;
178 		inode->i_ctime.tv_sec   = attr->ctime;
179 		inode->i_ctime.tv_nsec  = attr->ctimensec;
180 	}
181 
182 	if (attr->blksize != 0)
183 		inode->i_blkbits = ilog2(attr->blksize);
184 	else
185 		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
186 
187 	/*
188 	 * Don't set the sticky bit in i_mode, unless we want the VFS
189 	 * to check permissions.  This prevents failures due to the
190 	 * check in may_delete().
191 	 */
192 	fi->orig_i_mode = inode->i_mode;
193 	if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
194 		inode->i_mode &= ~S_ISVTX;
195 
196 	fi->orig_ino = attr->ino;
197 }
198 
199 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
200 			    u64 attr_valid, u64 attr_version)
201 {
202 	struct fuse_conn *fc = get_fuse_conn(inode);
203 	struct fuse_inode *fi = get_fuse_inode(inode);
204 	bool is_wb = fc->writeback_cache;
205 	loff_t oldsize;
206 	struct timespec old_mtime;
207 
208 	spin_lock(&fc->lock);
209 	if ((attr_version != 0 && fi->attr_version > attr_version) ||
210 	    test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
211 		spin_unlock(&fc->lock);
212 		return;
213 	}
214 
215 	old_mtime = inode->i_mtime;
216 	fuse_change_attributes_common(inode, attr, attr_valid);
217 
218 	oldsize = inode->i_size;
219 	/*
220 	 * In case of writeback_cache enabled, the cached writes beyond EOF
221 	 * extend local i_size without keeping userspace server in sync. So,
222 	 * attr->size coming from server can be stale. We cannot trust it.
223 	 */
224 	if (!is_wb || !S_ISREG(inode->i_mode))
225 		i_size_write(inode, attr->size);
226 	spin_unlock(&fc->lock);
227 
228 	if (!is_wb && S_ISREG(inode->i_mode)) {
229 		bool inval = false;
230 
231 		if (oldsize != attr->size) {
232 			truncate_pagecache(inode, attr->size);
233 			inval = true;
234 		} else if (fc->auto_inval_data) {
235 			struct timespec new_mtime = {
236 				.tv_sec = attr->mtime,
237 				.tv_nsec = attr->mtimensec,
238 			};
239 
240 			/*
241 			 * Auto inval mode also checks and invalidates if mtime
242 			 * has changed.
243 			 */
244 			if (!timespec_equal(&old_mtime, &new_mtime))
245 				inval = true;
246 		}
247 
248 		if (inval)
249 			invalidate_inode_pages2(inode->i_mapping);
250 	}
251 }
252 
253 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
254 {
255 	inode->i_mode = attr->mode & S_IFMT;
256 	inode->i_size = attr->size;
257 	inode->i_mtime.tv_sec  = attr->mtime;
258 	inode->i_mtime.tv_nsec = attr->mtimensec;
259 	inode->i_ctime.tv_sec  = attr->ctime;
260 	inode->i_ctime.tv_nsec = attr->ctimensec;
261 	if (S_ISREG(inode->i_mode)) {
262 		fuse_init_common(inode);
263 		fuse_init_file_inode(inode);
264 	} else if (S_ISDIR(inode->i_mode))
265 		fuse_init_dir(inode);
266 	else if (S_ISLNK(inode->i_mode))
267 		fuse_init_symlink(inode);
268 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
269 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
270 		fuse_init_common(inode);
271 		init_special_inode(inode, inode->i_mode,
272 				   new_decode_dev(attr->rdev));
273 	} else
274 		BUG();
275 }
276 
277 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
278 {
279 	u64 nodeid = *(u64 *) _nodeidp;
280 	if (get_node_id(inode) == nodeid)
281 		return 1;
282 	else
283 		return 0;
284 }
285 
286 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
287 {
288 	u64 nodeid = *(u64 *) _nodeidp;
289 	get_fuse_inode(inode)->nodeid = nodeid;
290 	return 0;
291 }
292 
293 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
294 			int generation, struct fuse_attr *attr,
295 			u64 attr_valid, u64 attr_version)
296 {
297 	struct inode *inode;
298 	struct fuse_inode *fi;
299 	struct fuse_conn *fc = get_fuse_conn_super(sb);
300 
301  retry:
302 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
303 	if (!inode)
304 		return NULL;
305 
306 	if ((inode->i_state & I_NEW)) {
307 		inode->i_flags |= S_NOATIME;
308 		if (!fc->writeback_cache || !S_ISREG(attr->mode))
309 			inode->i_flags |= S_NOCMTIME;
310 		inode->i_generation = generation;
311 		inode->i_data.backing_dev_info = &fc->bdi;
312 		fuse_init_inode(inode, attr);
313 		unlock_new_inode(inode);
314 	} else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
315 		/* Inode has changed type, any I/O on the old should fail */
316 		make_bad_inode(inode);
317 		iput(inode);
318 		goto retry;
319 	}
320 
321 	fi = get_fuse_inode(inode);
322 	spin_lock(&fc->lock);
323 	fi->nlookup++;
324 	spin_unlock(&fc->lock);
325 	fuse_change_attributes(inode, attr, attr_valid, attr_version);
326 
327 	return inode;
328 }
329 
330 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
331 			     loff_t offset, loff_t len)
332 {
333 	struct inode *inode;
334 	pgoff_t pg_start;
335 	pgoff_t pg_end;
336 
337 	inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
338 	if (!inode)
339 		return -ENOENT;
340 
341 	fuse_invalidate_attr(inode);
342 	if (offset >= 0) {
343 		pg_start = offset >> PAGE_CACHE_SHIFT;
344 		if (len <= 0)
345 			pg_end = -1;
346 		else
347 			pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
348 		invalidate_inode_pages2_range(inode->i_mapping,
349 					      pg_start, pg_end);
350 	}
351 	iput(inode);
352 	return 0;
353 }
354 
355 static void fuse_umount_begin(struct super_block *sb)
356 {
357 	fuse_abort_conn(get_fuse_conn_super(sb));
358 }
359 
360 static void fuse_send_destroy(struct fuse_conn *fc)
361 {
362 	struct fuse_req *req = fc->destroy_req;
363 	if (req && fc->conn_init) {
364 		fc->destroy_req = NULL;
365 		req->in.h.opcode = FUSE_DESTROY;
366 		req->force = 1;
367 		req->background = 0;
368 		fuse_request_send(fc, req);
369 		fuse_put_request(fc, req);
370 	}
371 }
372 
373 static void fuse_bdi_destroy(struct fuse_conn *fc)
374 {
375 	if (fc->bdi_initialized)
376 		bdi_destroy(&fc->bdi);
377 }
378 
379 static void fuse_put_super(struct super_block *sb)
380 {
381 	struct fuse_conn *fc = get_fuse_conn_super(sb);
382 
383 	fuse_send_destroy(fc);
384 
385 	fuse_abort_conn(fc);
386 	mutex_lock(&fuse_mutex);
387 	list_del(&fc->entry);
388 	fuse_ctl_remove_conn(fc);
389 	mutex_unlock(&fuse_mutex);
390 	fuse_bdi_destroy(fc);
391 
392 	fuse_conn_put(fc);
393 }
394 
395 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
396 {
397 	stbuf->f_type    = FUSE_SUPER_MAGIC;
398 	stbuf->f_bsize   = attr->bsize;
399 	stbuf->f_frsize  = attr->frsize;
400 	stbuf->f_blocks  = attr->blocks;
401 	stbuf->f_bfree   = attr->bfree;
402 	stbuf->f_bavail  = attr->bavail;
403 	stbuf->f_files   = attr->files;
404 	stbuf->f_ffree   = attr->ffree;
405 	stbuf->f_namelen = attr->namelen;
406 	/* fsid is left zero */
407 }
408 
409 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
410 {
411 	struct super_block *sb = dentry->d_sb;
412 	struct fuse_conn *fc = get_fuse_conn_super(sb);
413 	FUSE_ARGS(args);
414 	struct fuse_statfs_out outarg;
415 	int err;
416 
417 	if (!fuse_allow_current_process(fc)) {
418 		buf->f_type = FUSE_SUPER_MAGIC;
419 		return 0;
420 	}
421 
422 	memset(&outarg, 0, sizeof(outarg));
423 	args.in.numargs = 0;
424 	args.in.h.opcode = FUSE_STATFS;
425 	args.in.h.nodeid = get_node_id(dentry->d_inode);
426 	args.out.numargs = 1;
427 	args.out.args[0].size =
428 		fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
429 	args.out.args[0].value = &outarg;
430 	err = fuse_simple_request(fc, &args);
431 	if (!err)
432 		convert_fuse_statfs(buf, &outarg.st);
433 	return err;
434 }
435 
436 enum {
437 	OPT_FD,
438 	OPT_ROOTMODE,
439 	OPT_USER_ID,
440 	OPT_GROUP_ID,
441 	OPT_DEFAULT_PERMISSIONS,
442 	OPT_ALLOW_OTHER,
443 	OPT_MAX_READ,
444 	OPT_BLKSIZE,
445 	OPT_ERR
446 };
447 
448 static const match_table_t tokens = {
449 	{OPT_FD,			"fd=%u"},
450 	{OPT_ROOTMODE,			"rootmode=%o"},
451 	{OPT_USER_ID,			"user_id=%u"},
452 	{OPT_GROUP_ID,			"group_id=%u"},
453 	{OPT_DEFAULT_PERMISSIONS,	"default_permissions"},
454 	{OPT_ALLOW_OTHER,		"allow_other"},
455 	{OPT_MAX_READ,			"max_read=%u"},
456 	{OPT_BLKSIZE,			"blksize=%u"},
457 	{OPT_ERR,			NULL}
458 };
459 
460 static int fuse_match_uint(substring_t *s, unsigned int *res)
461 {
462 	int err = -ENOMEM;
463 	char *buf = match_strdup(s);
464 	if (buf) {
465 		err = kstrtouint(buf, 10, res);
466 		kfree(buf);
467 	}
468 	return err;
469 }
470 
471 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
472 {
473 	char *p;
474 	memset(d, 0, sizeof(struct fuse_mount_data));
475 	d->max_read = ~0;
476 	d->blksize = FUSE_DEFAULT_BLKSIZE;
477 
478 	while ((p = strsep(&opt, ",")) != NULL) {
479 		int token;
480 		int value;
481 		unsigned uv;
482 		substring_t args[MAX_OPT_ARGS];
483 		if (!*p)
484 			continue;
485 
486 		token = match_token(p, tokens, args);
487 		switch (token) {
488 		case OPT_FD:
489 			if (match_int(&args[0], &value))
490 				return 0;
491 			d->fd = value;
492 			d->fd_present = 1;
493 			break;
494 
495 		case OPT_ROOTMODE:
496 			if (match_octal(&args[0], &value))
497 				return 0;
498 			if (!fuse_valid_type(value))
499 				return 0;
500 			d->rootmode = value;
501 			d->rootmode_present = 1;
502 			break;
503 
504 		case OPT_USER_ID:
505 			if (fuse_match_uint(&args[0], &uv))
506 				return 0;
507 			d->user_id = make_kuid(current_user_ns(), uv);
508 			if (!uid_valid(d->user_id))
509 				return 0;
510 			d->user_id_present = 1;
511 			break;
512 
513 		case OPT_GROUP_ID:
514 			if (fuse_match_uint(&args[0], &uv))
515 				return 0;
516 			d->group_id = make_kgid(current_user_ns(), uv);
517 			if (!gid_valid(d->group_id))
518 				return 0;
519 			d->group_id_present = 1;
520 			break;
521 
522 		case OPT_DEFAULT_PERMISSIONS:
523 			d->flags |= FUSE_DEFAULT_PERMISSIONS;
524 			break;
525 
526 		case OPT_ALLOW_OTHER:
527 			d->flags |= FUSE_ALLOW_OTHER;
528 			break;
529 
530 		case OPT_MAX_READ:
531 			if (match_int(&args[0], &value))
532 				return 0;
533 			d->max_read = value;
534 			break;
535 
536 		case OPT_BLKSIZE:
537 			if (!is_bdev || match_int(&args[0], &value))
538 				return 0;
539 			d->blksize = value;
540 			break;
541 
542 		default:
543 			return 0;
544 		}
545 	}
546 
547 	if (!d->fd_present || !d->rootmode_present ||
548 	    !d->user_id_present || !d->group_id_present)
549 		return 0;
550 
551 	return 1;
552 }
553 
554 static int fuse_show_options(struct seq_file *m, struct dentry *root)
555 {
556 	struct super_block *sb = root->d_sb;
557 	struct fuse_conn *fc = get_fuse_conn_super(sb);
558 
559 	seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
560 	seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
561 	if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
562 		seq_puts(m, ",default_permissions");
563 	if (fc->flags & FUSE_ALLOW_OTHER)
564 		seq_puts(m, ",allow_other");
565 	if (fc->max_read != ~0)
566 		seq_printf(m, ",max_read=%u", fc->max_read);
567 	if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
568 		seq_printf(m, ",blksize=%lu", sb->s_blocksize);
569 	return 0;
570 }
571 
572 void fuse_conn_init(struct fuse_conn *fc)
573 {
574 	memset(fc, 0, sizeof(*fc));
575 	spin_lock_init(&fc->lock);
576 	init_rwsem(&fc->killsb);
577 	atomic_set(&fc->count, 1);
578 	init_waitqueue_head(&fc->waitq);
579 	init_waitqueue_head(&fc->blocked_waitq);
580 	init_waitqueue_head(&fc->reserved_req_waitq);
581 	INIT_LIST_HEAD(&fc->pending);
582 	INIT_LIST_HEAD(&fc->processing);
583 	INIT_LIST_HEAD(&fc->io);
584 	INIT_LIST_HEAD(&fc->interrupts);
585 	INIT_LIST_HEAD(&fc->bg_queue);
586 	INIT_LIST_HEAD(&fc->entry);
587 	fc->forget_list_tail = &fc->forget_list_head;
588 	atomic_set(&fc->num_waiting, 0);
589 	fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
590 	fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
591 	fc->khctr = 0;
592 	fc->polled_files = RB_ROOT;
593 	fc->reqctr = 0;
594 	fc->blocked = 0;
595 	fc->initialized = 0;
596 	fc->attr_version = 1;
597 	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
598 }
599 EXPORT_SYMBOL_GPL(fuse_conn_init);
600 
601 void fuse_conn_put(struct fuse_conn *fc)
602 {
603 	if (atomic_dec_and_test(&fc->count)) {
604 		if (fc->destroy_req)
605 			fuse_request_free(fc->destroy_req);
606 		fc->release(fc);
607 	}
608 }
609 EXPORT_SYMBOL_GPL(fuse_conn_put);
610 
611 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
612 {
613 	atomic_inc(&fc->count);
614 	return fc;
615 }
616 EXPORT_SYMBOL_GPL(fuse_conn_get);
617 
618 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
619 {
620 	struct fuse_attr attr;
621 	memset(&attr, 0, sizeof(attr));
622 
623 	attr.mode = mode;
624 	attr.ino = FUSE_ROOT_ID;
625 	attr.nlink = 1;
626 	return fuse_iget(sb, 1, 0, &attr, 0, 0);
627 }
628 
629 struct fuse_inode_handle {
630 	u64 nodeid;
631 	u32 generation;
632 };
633 
634 static struct dentry *fuse_get_dentry(struct super_block *sb,
635 				      struct fuse_inode_handle *handle)
636 {
637 	struct fuse_conn *fc = get_fuse_conn_super(sb);
638 	struct inode *inode;
639 	struct dentry *entry;
640 	int err = -ESTALE;
641 
642 	if (handle->nodeid == 0)
643 		goto out_err;
644 
645 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
646 	if (!inode) {
647 		struct fuse_entry_out outarg;
648 		struct qstr name;
649 
650 		if (!fc->export_support)
651 			goto out_err;
652 
653 		name.len = 1;
654 		name.name = ".";
655 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
656 				       &inode);
657 		if (err && err != -ENOENT)
658 			goto out_err;
659 		if (err || !inode) {
660 			err = -ESTALE;
661 			goto out_err;
662 		}
663 		err = -EIO;
664 		if (get_node_id(inode) != handle->nodeid)
665 			goto out_iput;
666 	}
667 	err = -ESTALE;
668 	if (inode->i_generation != handle->generation)
669 		goto out_iput;
670 
671 	entry = d_obtain_alias(inode);
672 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
673 		fuse_invalidate_entry_cache(entry);
674 
675 	return entry;
676 
677  out_iput:
678 	iput(inode);
679  out_err:
680 	return ERR_PTR(err);
681 }
682 
683 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
684 			   struct inode *parent)
685 {
686 	int len = parent ? 6 : 3;
687 	u64 nodeid;
688 	u32 generation;
689 
690 	if (*max_len < len) {
691 		*max_len = len;
692 		return  FILEID_INVALID;
693 	}
694 
695 	nodeid = get_fuse_inode(inode)->nodeid;
696 	generation = inode->i_generation;
697 
698 	fh[0] = (u32)(nodeid >> 32);
699 	fh[1] = (u32)(nodeid & 0xffffffff);
700 	fh[2] = generation;
701 
702 	if (parent) {
703 		nodeid = get_fuse_inode(parent)->nodeid;
704 		generation = parent->i_generation;
705 
706 		fh[3] = (u32)(nodeid >> 32);
707 		fh[4] = (u32)(nodeid & 0xffffffff);
708 		fh[5] = generation;
709 	}
710 
711 	*max_len = len;
712 	return parent ? 0x82 : 0x81;
713 }
714 
715 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
716 		struct fid *fid, int fh_len, int fh_type)
717 {
718 	struct fuse_inode_handle handle;
719 
720 	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
721 		return NULL;
722 
723 	handle.nodeid = (u64) fid->raw[0] << 32;
724 	handle.nodeid |= (u64) fid->raw[1];
725 	handle.generation = fid->raw[2];
726 	return fuse_get_dentry(sb, &handle);
727 }
728 
729 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
730 		struct fid *fid, int fh_len, int fh_type)
731 {
732 	struct fuse_inode_handle parent;
733 
734 	if (fh_type != 0x82 || fh_len < 6)
735 		return NULL;
736 
737 	parent.nodeid = (u64) fid->raw[3] << 32;
738 	parent.nodeid |= (u64) fid->raw[4];
739 	parent.generation = fid->raw[5];
740 	return fuse_get_dentry(sb, &parent);
741 }
742 
743 static struct dentry *fuse_get_parent(struct dentry *child)
744 {
745 	struct inode *child_inode = child->d_inode;
746 	struct fuse_conn *fc = get_fuse_conn(child_inode);
747 	struct inode *inode;
748 	struct dentry *parent;
749 	struct fuse_entry_out outarg;
750 	struct qstr name;
751 	int err;
752 
753 	if (!fc->export_support)
754 		return ERR_PTR(-ESTALE);
755 
756 	name.len = 2;
757 	name.name = "..";
758 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
759 			       &name, &outarg, &inode);
760 	if (err) {
761 		if (err == -ENOENT)
762 			return ERR_PTR(-ESTALE);
763 		return ERR_PTR(err);
764 	}
765 
766 	parent = d_obtain_alias(inode);
767 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
768 		fuse_invalidate_entry_cache(parent);
769 
770 	return parent;
771 }
772 
773 static const struct export_operations fuse_export_operations = {
774 	.fh_to_dentry	= fuse_fh_to_dentry,
775 	.fh_to_parent	= fuse_fh_to_parent,
776 	.encode_fh	= fuse_encode_fh,
777 	.get_parent	= fuse_get_parent,
778 };
779 
780 static const struct super_operations fuse_super_operations = {
781 	.alloc_inode    = fuse_alloc_inode,
782 	.destroy_inode  = fuse_destroy_inode,
783 	.evict_inode	= fuse_evict_inode,
784 	.write_inode	= fuse_write_inode,
785 	.drop_inode	= generic_delete_inode,
786 	.remount_fs	= fuse_remount_fs,
787 	.put_super	= fuse_put_super,
788 	.umount_begin	= fuse_umount_begin,
789 	.statfs		= fuse_statfs,
790 	.show_options	= fuse_show_options,
791 };
792 
793 static void sanitize_global_limit(unsigned *limit)
794 {
795 	if (*limit == 0)
796 		*limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
797 			 sizeof(struct fuse_req);
798 
799 	if (*limit >= 1 << 16)
800 		*limit = (1 << 16) - 1;
801 }
802 
803 static int set_global_limit(const char *val, struct kernel_param *kp)
804 {
805 	int rv;
806 
807 	rv = param_set_uint(val, kp);
808 	if (rv)
809 		return rv;
810 
811 	sanitize_global_limit((unsigned *)kp->arg);
812 
813 	return 0;
814 }
815 
816 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
817 {
818 	int cap_sys_admin = capable(CAP_SYS_ADMIN);
819 
820 	if (arg->minor < 13)
821 		return;
822 
823 	sanitize_global_limit(&max_user_bgreq);
824 	sanitize_global_limit(&max_user_congthresh);
825 
826 	if (arg->max_background) {
827 		fc->max_background = arg->max_background;
828 
829 		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
830 			fc->max_background = max_user_bgreq;
831 	}
832 	if (arg->congestion_threshold) {
833 		fc->congestion_threshold = arg->congestion_threshold;
834 
835 		if (!cap_sys_admin &&
836 		    fc->congestion_threshold > max_user_congthresh)
837 			fc->congestion_threshold = max_user_congthresh;
838 	}
839 }
840 
841 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
842 {
843 	struct fuse_init_out *arg = &req->misc.init_out;
844 
845 	if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
846 		fc->conn_error = 1;
847 	else {
848 		unsigned long ra_pages;
849 
850 		process_init_limits(fc, arg);
851 
852 		if (arg->minor >= 6) {
853 			ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
854 			if (arg->flags & FUSE_ASYNC_READ)
855 				fc->async_read = 1;
856 			if (!(arg->flags & FUSE_POSIX_LOCKS))
857 				fc->no_lock = 1;
858 			if (arg->minor >= 17) {
859 				if (!(arg->flags & FUSE_FLOCK_LOCKS))
860 					fc->no_flock = 1;
861 			} else {
862 				if (!(arg->flags & FUSE_POSIX_LOCKS))
863 					fc->no_flock = 1;
864 			}
865 			if (arg->flags & FUSE_ATOMIC_O_TRUNC)
866 				fc->atomic_o_trunc = 1;
867 			if (arg->minor >= 9) {
868 				/* LOOKUP has dependency on proto version */
869 				if (arg->flags & FUSE_EXPORT_SUPPORT)
870 					fc->export_support = 1;
871 			}
872 			if (arg->flags & FUSE_BIG_WRITES)
873 				fc->big_writes = 1;
874 			if (arg->flags & FUSE_DONT_MASK)
875 				fc->dont_mask = 1;
876 			if (arg->flags & FUSE_AUTO_INVAL_DATA)
877 				fc->auto_inval_data = 1;
878 			if (arg->flags & FUSE_DO_READDIRPLUS) {
879 				fc->do_readdirplus = 1;
880 				if (arg->flags & FUSE_READDIRPLUS_AUTO)
881 					fc->readdirplus_auto = 1;
882 			}
883 			if (arg->flags & FUSE_ASYNC_DIO)
884 				fc->async_dio = 1;
885 			if (arg->flags & FUSE_WRITEBACK_CACHE)
886 				fc->writeback_cache = 1;
887 			if (arg->time_gran && arg->time_gran <= 1000000000)
888 				fc->sb->s_time_gran = arg->time_gran;
889 		} else {
890 			ra_pages = fc->max_read / PAGE_CACHE_SIZE;
891 			fc->no_lock = 1;
892 			fc->no_flock = 1;
893 		}
894 
895 		fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
896 		fc->minor = arg->minor;
897 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
898 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
899 		fc->conn_init = 1;
900 	}
901 	fc->initialized = 1;
902 	wake_up_all(&fc->blocked_waitq);
903 }
904 
905 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
906 {
907 	struct fuse_init_in *arg = &req->misc.init_in;
908 
909 	arg->major = FUSE_KERNEL_VERSION;
910 	arg->minor = FUSE_KERNEL_MINOR_VERSION;
911 	arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
912 	arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
913 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
914 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
915 		FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
916 		FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
917 		FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT;
918 	req->in.h.opcode = FUSE_INIT;
919 	req->in.numargs = 1;
920 	req->in.args[0].size = sizeof(*arg);
921 	req->in.args[0].value = arg;
922 	req->out.numargs = 1;
923 	/* Variable length argument used for backward compatibility
924 	   with interface version < 7.5.  Rest of init_out is zeroed
925 	   by do_get_request(), so a short reply is not a problem */
926 	req->out.argvar = 1;
927 	req->out.args[0].size = sizeof(struct fuse_init_out);
928 	req->out.args[0].value = &req->misc.init_out;
929 	req->end = process_init_reply;
930 	fuse_request_send_background(fc, req);
931 }
932 
933 static void fuse_free_conn(struct fuse_conn *fc)
934 {
935 	kfree_rcu(fc, rcu);
936 }
937 
938 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
939 {
940 	int err;
941 
942 	fc->bdi.name = "fuse";
943 	fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
944 	/* fuse does it's own writeback accounting */
945 	fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
946 
947 	err = bdi_init(&fc->bdi);
948 	if (err)
949 		return err;
950 
951 	fc->bdi_initialized = 1;
952 
953 	if (sb->s_bdev) {
954 		err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
955 				    MAJOR(fc->dev), MINOR(fc->dev));
956 	} else {
957 		err = bdi_register_dev(&fc->bdi, fc->dev);
958 	}
959 
960 	if (err)
961 		return err;
962 
963 	/*
964 	 * For a single fuse filesystem use max 1% of dirty +
965 	 * writeback threshold.
966 	 *
967 	 * This gives about 1M of write buffer for memory maps on a
968 	 * machine with 1G and 10% dirty_ratio, which should be more
969 	 * than enough.
970 	 *
971 	 * Privileged users can raise it by writing to
972 	 *
973 	 *    /sys/class/bdi/<bdi>/max_ratio
974 	 */
975 	bdi_set_max_ratio(&fc->bdi, 1);
976 
977 	return 0;
978 }
979 
980 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
981 {
982 	struct fuse_conn *fc;
983 	struct inode *root;
984 	struct fuse_mount_data d;
985 	struct file *file;
986 	struct dentry *root_dentry;
987 	struct fuse_req *init_req;
988 	int err;
989 	int is_bdev = sb->s_bdev != NULL;
990 
991 	err = -EINVAL;
992 	if (sb->s_flags & MS_MANDLOCK)
993 		goto err;
994 
995 	sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION);
996 
997 	if (!parse_fuse_opt(data, &d, is_bdev))
998 		goto err;
999 
1000 	if (is_bdev) {
1001 #ifdef CONFIG_BLOCK
1002 		err = -EINVAL;
1003 		if (!sb_set_blocksize(sb, d.blksize))
1004 			goto err;
1005 #endif
1006 	} else {
1007 		sb->s_blocksize = PAGE_CACHE_SIZE;
1008 		sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1009 	}
1010 	sb->s_magic = FUSE_SUPER_MAGIC;
1011 	sb->s_op = &fuse_super_operations;
1012 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1013 	sb->s_time_gran = 1;
1014 	sb->s_export_op = &fuse_export_operations;
1015 
1016 	file = fget(d.fd);
1017 	err = -EINVAL;
1018 	if (!file)
1019 		goto err;
1020 
1021 	if ((file->f_op != &fuse_dev_operations) ||
1022 	    (file->f_cred->user_ns != &init_user_ns))
1023 		goto err_fput;
1024 
1025 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1026 	err = -ENOMEM;
1027 	if (!fc)
1028 		goto err_fput;
1029 
1030 	fuse_conn_init(fc);
1031 
1032 	fc->dev = sb->s_dev;
1033 	fc->sb = sb;
1034 	err = fuse_bdi_init(fc, sb);
1035 	if (err)
1036 		goto err_put_conn;
1037 
1038 	sb->s_bdi = &fc->bdi;
1039 
1040 	/* Handle umasking inside the fuse code */
1041 	if (sb->s_flags & MS_POSIXACL)
1042 		fc->dont_mask = 1;
1043 	sb->s_flags |= MS_POSIXACL;
1044 
1045 	fc->release = fuse_free_conn;
1046 	fc->flags = d.flags;
1047 	fc->user_id = d.user_id;
1048 	fc->group_id = d.group_id;
1049 	fc->max_read = max_t(unsigned, 4096, d.max_read);
1050 
1051 	/* Used by get_root_inode() */
1052 	sb->s_fs_info = fc;
1053 
1054 	err = -ENOMEM;
1055 	root = fuse_get_root_inode(sb, d.rootmode);
1056 	root_dentry = d_make_root(root);
1057 	if (!root_dentry)
1058 		goto err_put_conn;
1059 	/* only now - we want root dentry with NULL ->d_op */
1060 	sb->s_d_op = &fuse_dentry_operations;
1061 
1062 	init_req = fuse_request_alloc(0);
1063 	if (!init_req)
1064 		goto err_put_root;
1065 	init_req->background = 1;
1066 
1067 	if (is_bdev) {
1068 		fc->destroy_req = fuse_request_alloc(0);
1069 		if (!fc->destroy_req)
1070 			goto err_free_init_req;
1071 	}
1072 
1073 	mutex_lock(&fuse_mutex);
1074 	err = -EINVAL;
1075 	if (file->private_data)
1076 		goto err_unlock;
1077 
1078 	err = fuse_ctl_add_conn(fc);
1079 	if (err)
1080 		goto err_unlock;
1081 
1082 	list_add_tail(&fc->entry, &fuse_conn_list);
1083 	sb->s_root = root_dentry;
1084 	fc->connected = 1;
1085 	file->private_data = fuse_conn_get(fc);
1086 	mutex_unlock(&fuse_mutex);
1087 	/*
1088 	 * atomic_dec_and_test() in fput() provides the necessary
1089 	 * memory barrier for file->private_data to be visible on all
1090 	 * CPUs after this
1091 	 */
1092 	fput(file);
1093 
1094 	fuse_send_init(fc, init_req);
1095 
1096 	return 0;
1097 
1098  err_unlock:
1099 	mutex_unlock(&fuse_mutex);
1100  err_free_init_req:
1101 	fuse_request_free(init_req);
1102  err_put_root:
1103 	dput(root_dentry);
1104  err_put_conn:
1105 	fuse_bdi_destroy(fc);
1106 	fuse_conn_put(fc);
1107  err_fput:
1108 	fput(file);
1109  err:
1110 	return err;
1111 }
1112 
1113 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1114 		       int flags, const char *dev_name,
1115 		       void *raw_data)
1116 {
1117 	return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1118 }
1119 
1120 static void fuse_kill_sb_anon(struct super_block *sb)
1121 {
1122 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1123 
1124 	if (fc) {
1125 		down_write(&fc->killsb);
1126 		fc->sb = NULL;
1127 		up_write(&fc->killsb);
1128 	}
1129 
1130 	kill_anon_super(sb);
1131 }
1132 
1133 static struct file_system_type fuse_fs_type = {
1134 	.owner		= THIS_MODULE,
1135 	.name		= "fuse",
1136 	.fs_flags	= FS_HAS_SUBTYPE,
1137 	.mount		= fuse_mount,
1138 	.kill_sb	= fuse_kill_sb_anon,
1139 };
1140 MODULE_ALIAS_FS("fuse");
1141 
1142 #ifdef CONFIG_BLOCK
1143 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1144 			   int flags, const char *dev_name,
1145 			   void *raw_data)
1146 {
1147 	return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1148 }
1149 
1150 static void fuse_kill_sb_blk(struct super_block *sb)
1151 {
1152 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1153 
1154 	if (fc) {
1155 		down_write(&fc->killsb);
1156 		fc->sb = NULL;
1157 		up_write(&fc->killsb);
1158 	}
1159 
1160 	kill_block_super(sb);
1161 }
1162 
1163 static struct file_system_type fuseblk_fs_type = {
1164 	.owner		= THIS_MODULE,
1165 	.name		= "fuseblk",
1166 	.mount		= fuse_mount_blk,
1167 	.kill_sb	= fuse_kill_sb_blk,
1168 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1169 };
1170 MODULE_ALIAS_FS("fuseblk");
1171 
1172 static inline int register_fuseblk(void)
1173 {
1174 	return register_filesystem(&fuseblk_fs_type);
1175 }
1176 
1177 static inline void unregister_fuseblk(void)
1178 {
1179 	unregister_filesystem(&fuseblk_fs_type);
1180 }
1181 #else
1182 static inline int register_fuseblk(void)
1183 {
1184 	return 0;
1185 }
1186 
1187 static inline void unregister_fuseblk(void)
1188 {
1189 }
1190 #endif
1191 
1192 static void fuse_inode_init_once(void *foo)
1193 {
1194 	struct inode *inode = foo;
1195 
1196 	inode_init_once(inode);
1197 }
1198 
1199 static int __init fuse_fs_init(void)
1200 {
1201 	int err;
1202 
1203 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
1204 					      sizeof(struct fuse_inode),
1205 					      0, SLAB_HWCACHE_ALIGN,
1206 					      fuse_inode_init_once);
1207 	err = -ENOMEM;
1208 	if (!fuse_inode_cachep)
1209 		goto out;
1210 
1211 	err = register_fuseblk();
1212 	if (err)
1213 		goto out2;
1214 
1215 	err = register_filesystem(&fuse_fs_type);
1216 	if (err)
1217 		goto out3;
1218 
1219 	return 0;
1220 
1221  out3:
1222 	unregister_fuseblk();
1223  out2:
1224 	kmem_cache_destroy(fuse_inode_cachep);
1225  out:
1226 	return err;
1227 }
1228 
1229 static void fuse_fs_cleanup(void)
1230 {
1231 	unregister_filesystem(&fuse_fs_type);
1232 	unregister_fuseblk();
1233 
1234 	/*
1235 	 * Make sure all delayed rcu free inodes are flushed before we
1236 	 * destroy cache.
1237 	 */
1238 	rcu_barrier();
1239 	kmem_cache_destroy(fuse_inode_cachep);
1240 }
1241 
1242 static struct kobject *fuse_kobj;
1243 static struct kobject *connections_kobj;
1244 
1245 static int fuse_sysfs_init(void)
1246 {
1247 	int err;
1248 
1249 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1250 	if (!fuse_kobj) {
1251 		err = -ENOMEM;
1252 		goto out_err;
1253 	}
1254 
1255 	connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1256 	if (!connections_kobj) {
1257 		err = -ENOMEM;
1258 		goto out_fuse_unregister;
1259 	}
1260 
1261 	return 0;
1262 
1263  out_fuse_unregister:
1264 	kobject_put(fuse_kobj);
1265  out_err:
1266 	return err;
1267 }
1268 
1269 static void fuse_sysfs_cleanup(void)
1270 {
1271 	kobject_put(connections_kobj);
1272 	kobject_put(fuse_kobj);
1273 }
1274 
1275 static int __init fuse_init(void)
1276 {
1277 	int res;
1278 
1279 	printk(KERN_INFO "fuse init (API version %i.%i)\n",
1280 	       FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1281 
1282 	INIT_LIST_HEAD(&fuse_conn_list);
1283 	res = fuse_fs_init();
1284 	if (res)
1285 		goto err;
1286 
1287 	res = fuse_dev_init();
1288 	if (res)
1289 		goto err_fs_cleanup;
1290 
1291 	res = fuse_sysfs_init();
1292 	if (res)
1293 		goto err_dev_cleanup;
1294 
1295 	res = fuse_ctl_init();
1296 	if (res)
1297 		goto err_sysfs_cleanup;
1298 
1299 	sanitize_global_limit(&max_user_bgreq);
1300 	sanitize_global_limit(&max_user_congthresh);
1301 
1302 	return 0;
1303 
1304  err_sysfs_cleanup:
1305 	fuse_sysfs_cleanup();
1306  err_dev_cleanup:
1307 	fuse_dev_cleanup();
1308  err_fs_cleanup:
1309 	fuse_fs_cleanup();
1310  err:
1311 	return res;
1312 }
1313 
1314 static void __exit fuse_exit(void)
1315 {
1316 	printk(KERN_DEBUG "fuse exit\n");
1317 
1318 	fuse_ctl_cleanup();
1319 	fuse_sysfs_cleanup();
1320 	fuse_fs_cleanup();
1321 	fuse_dev_cleanup();
1322 }
1323 
1324 module_init(fuse_init);
1325 module_exit(fuse_exit);
1326