xref: /openbmc/linux/fs/fuse/inode.c (revision 4582a4ab2a0e7218449fb2e895d0aae9ea753c94)
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(&inode->i_data, 0);
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 	if (*flags & MS_MANDLOCK)
139 		return -EINVAL;
140 
141 	return 0;
142 }
143 
144 /*
145  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
146  * so that it will fit.
147  */
148 static ino_t fuse_squash_ino(u64 ino64)
149 {
150 	ino_t ino = (ino_t) ino64;
151 	if (sizeof(ino_t) < sizeof(u64))
152 		ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
153 	return ino;
154 }
155 
156 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
157 				   u64 attr_valid)
158 {
159 	struct fuse_conn *fc = get_fuse_conn(inode);
160 	struct fuse_inode *fi = get_fuse_inode(inode);
161 
162 	fi->attr_version = ++fc->attr_version;
163 	fi->i_time = attr_valid;
164 
165 	inode->i_ino     = fuse_squash_ino(attr->ino);
166 	inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
167 	set_nlink(inode, attr->nlink);
168 	inode->i_uid     = make_kuid(&init_user_ns, attr->uid);
169 	inode->i_gid     = make_kgid(&init_user_ns, attr->gid);
170 	inode->i_blocks  = attr->blocks;
171 	inode->i_atime.tv_sec   = attr->atime;
172 	inode->i_atime.tv_nsec  = attr->atimensec;
173 	inode->i_mtime.tv_sec   = attr->mtime;
174 	inode->i_mtime.tv_nsec  = attr->mtimensec;
175 	inode->i_ctime.tv_sec   = attr->ctime;
176 	inode->i_ctime.tv_nsec  = attr->ctimensec;
177 
178 	if (attr->blksize != 0)
179 		inode->i_blkbits = ilog2(attr->blksize);
180 	else
181 		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
182 
183 	/*
184 	 * Don't set the sticky bit in i_mode, unless we want the VFS
185 	 * to check permissions.  This prevents failures due to the
186 	 * check in may_delete().
187 	 */
188 	fi->orig_i_mode = inode->i_mode;
189 	if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
190 		inode->i_mode &= ~S_ISVTX;
191 
192 	fi->orig_ino = attr->ino;
193 }
194 
195 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
196 			    u64 attr_valid, u64 attr_version)
197 {
198 	struct fuse_conn *fc = get_fuse_conn(inode);
199 	struct fuse_inode *fi = get_fuse_inode(inode);
200 	loff_t oldsize;
201 	struct timespec old_mtime;
202 
203 	spin_lock(&fc->lock);
204 	if (attr_version != 0 && fi->attr_version > attr_version) {
205 		spin_unlock(&fc->lock);
206 		return;
207 	}
208 
209 	old_mtime = inode->i_mtime;
210 	fuse_change_attributes_common(inode, attr, attr_valid);
211 
212 	oldsize = inode->i_size;
213 	i_size_write(inode, attr->size);
214 	spin_unlock(&fc->lock);
215 
216 	if (S_ISREG(inode->i_mode)) {
217 		bool inval = false;
218 
219 		if (oldsize != attr->size) {
220 			truncate_pagecache(inode, oldsize, attr->size);
221 			inval = true;
222 		} else if (fc->auto_inval_data) {
223 			struct timespec new_mtime = {
224 				.tv_sec = attr->mtime,
225 				.tv_nsec = attr->mtimensec,
226 			};
227 
228 			/*
229 			 * Auto inval mode also checks and invalidates if mtime
230 			 * has changed.
231 			 */
232 			if (!timespec_equal(&old_mtime, &new_mtime))
233 				inval = true;
234 		}
235 
236 		if (inval)
237 			invalidate_inode_pages2(inode->i_mapping);
238 	}
239 }
240 
241 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
242 {
243 	inode->i_mode = attr->mode & S_IFMT;
244 	inode->i_size = attr->size;
245 	if (S_ISREG(inode->i_mode)) {
246 		fuse_init_common(inode);
247 		fuse_init_file_inode(inode);
248 	} else if (S_ISDIR(inode->i_mode))
249 		fuse_init_dir(inode);
250 	else if (S_ISLNK(inode->i_mode))
251 		fuse_init_symlink(inode);
252 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
253 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
254 		fuse_init_common(inode);
255 		init_special_inode(inode, inode->i_mode,
256 				   new_decode_dev(attr->rdev));
257 	} else
258 		BUG();
259 }
260 
261 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
262 {
263 	u64 nodeid = *(u64 *) _nodeidp;
264 	if (get_node_id(inode) == nodeid)
265 		return 1;
266 	else
267 		return 0;
268 }
269 
270 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
271 {
272 	u64 nodeid = *(u64 *) _nodeidp;
273 	get_fuse_inode(inode)->nodeid = nodeid;
274 	return 0;
275 }
276 
277 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
278 			int generation, struct fuse_attr *attr,
279 			u64 attr_valid, u64 attr_version)
280 {
281 	struct inode *inode;
282 	struct fuse_inode *fi;
283 	struct fuse_conn *fc = get_fuse_conn_super(sb);
284 
285  retry:
286 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
287 	if (!inode)
288 		return NULL;
289 
290 	if ((inode->i_state & I_NEW)) {
291 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
292 		inode->i_generation = generation;
293 		inode->i_data.backing_dev_info = &fc->bdi;
294 		fuse_init_inode(inode, attr);
295 		unlock_new_inode(inode);
296 	} else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
297 		/* Inode has changed type, any I/O on the old should fail */
298 		make_bad_inode(inode);
299 		iput(inode);
300 		goto retry;
301 	}
302 
303 	fi = get_fuse_inode(inode);
304 	spin_lock(&fc->lock);
305 	fi->nlookup++;
306 	spin_unlock(&fc->lock);
307 	fuse_change_attributes(inode, attr, attr_valid, attr_version);
308 
309 	return inode;
310 }
311 
312 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
313 			     loff_t offset, loff_t len)
314 {
315 	struct inode *inode;
316 	pgoff_t pg_start;
317 	pgoff_t pg_end;
318 
319 	inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
320 	if (!inode)
321 		return -ENOENT;
322 
323 	fuse_invalidate_attr(inode);
324 	if (offset >= 0) {
325 		pg_start = offset >> PAGE_CACHE_SHIFT;
326 		if (len <= 0)
327 			pg_end = -1;
328 		else
329 			pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
330 		invalidate_inode_pages2_range(inode->i_mapping,
331 					      pg_start, pg_end);
332 	}
333 	iput(inode);
334 	return 0;
335 }
336 
337 static void fuse_umount_begin(struct super_block *sb)
338 {
339 	fuse_abort_conn(get_fuse_conn_super(sb));
340 }
341 
342 static void fuse_send_destroy(struct fuse_conn *fc)
343 {
344 	struct fuse_req *req = fc->destroy_req;
345 	if (req && fc->conn_init) {
346 		fc->destroy_req = NULL;
347 		req->in.h.opcode = FUSE_DESTROY;
348 		req->force = 1;
349 		fuse_request_send(fc, req);
350 		fuse_put_request(fc, req);
351 	}
352 }
353 
354 static void fuse_bdi_destroy(struct fuse_conn *fc)
355 {
356 	if (fc->bdi_initialized)
357 		bdi_destroy(&fc->bdi);
358 }
359 
360 void fuse_conn_kill(struct fuse_conn *fc)
361 {
362 	spin_lock(&fc->lock);
363 	fc->connected = 0;
364 	fc->blocked = 0;
365 	spin_unlock(&fc->lock);
366 	/* Flush all readers on this fs */
367 	kill_fasync(&fc->fasync, SIGIO, POLL_IN);
368 	wake_up_all(&fc->waitq);
369 	wake_up_all(&fc->blocked_waitq);
370 	wake_up_all(&fc->reserved_req_waitq);
371 }
372 EXPORT_SYMBOL_GPL(fuse_conn_kill);
373 
374 static void fuse_put_super(struct super_block *sb)
375 {
376 	struct fuse_conn *fc = get_fuse_conn_super(sb);
377 
378 	fuse_send_destroy(fc);
379 
380 	fuse_conn_kill(fc);
381 	mutex_lock(&fuse_mutex);
382 	list_del(&fc->entry);
383 	fuse_ctl_remove_conn(fc);
384 	mutex_unlock(&fuse_mutex);
385 	fuse_bdi_destroy(fc);
386 
387 	fuse_conn_put(fc);
388 }
389 
390 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
391 {
392 	stbuf->f_type    = FUSE_SUPER_MAGIC;
393 	stbuf->f_bsize   = attr->bsize;
394 	stbuf->f_frsize  = attr->frsize;
395 	stbuf->f_blocks  = attr->blocks;
396 	stbuf->f_bfree   = attr->bfree;
397 	stbuf->f_bavail  = attr->bavail;
398 	stbuf->f_files   = attr->files;
399 	stbuf->f_ffree   = attr->ffree;
400 	stbuf->f_namelen = attr->namelen;
401 	/* fsid is left zero */
402 }
403 
404 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
405 {
406 	struct super_block *sb = dentry->d_sb;
407 	struct fuse_conn *fc = get_fuse_conn_super(sb);
408 	struct fuse_req *req;
409 	struct fuse_statfs_out outarg;
410 	int err;
411 
412 	if (!fuse_allow_current_process(fc)) {
413 		buf->f_type = FUSE_SUPER_MAGIC;
414 		return 0;
415 	}
416 
417 	req = fuse_get_req_nopages(fc);
418 	if (IS_ERR(req))
419 		return PTR_ERR(req);
420 
421 	memset(&outarg, 0, sizeof(outarg));
422 	req->in.numargs = 0;
423 	req->in.h.opcode = FUSE_STATFS;
424 	req->in.h.nodeid = get_node_id(dentry->d_inode);
425 	req->out.numargs = 1;
426 	req->out.args[0].size =
427 		fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
428 	req->out.args[0].value = &outarg;
429 	fuse_request_send(fc, req);
430 	err = req->out.h.error;
431 	if (!err)
432 		convert_fuse_statfs(buf, &outarg.st);
433 	fuse_put_request(fc, req);
434 	return err;
435 }
436 
437 enum {
438 	OPT_FD,
439 	OPT_ROOTMODE,
440 	OPT_USER_ID,
441 	OPT_GROUP_ID,
442 	OPT_DEFAULT_PERMISSIONS,
443 	OPT_ALLOW_OTHER,
444 	OPT_MAX_READ,
445 	OPT_BLKSIZE,
446 	OPT_ERR
447 };
448 
449 static const match_table_t tokens = {
450 	{OPT_FD,			"fd=%u"},
451 	{OPT_ROOTMODE,			"rootmode=%o"},
452 	{OPT_USER_ID,			"user_id=%u"},
453 	{OPT_GROUP_ID,			"group_id=%u"},
454 	{OPT_DEFAULT_PERMISSIONS,	"default_permissions"},
455 	{OPT_ALLOW_OTHER,		"allow_other"},
456 	{OPT_MAX_READ,			"max_read=%u"},
457 	{OPT_BLKSIZE,			"blksize=%u"},
458 	{OPT_ERR,			NULL}
459 };
460 
461 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
462 {
463 	char *p;
464 	memset(d, 0, sizeof(struct fuse_mount_data));
465 	d->max_read = ~0;
466 	d->blksize = FUSE_DEFAULT_BLKSIZE;
467 
468 	while ((p = strsep(&opt, ",")) != NULL) {
469 		int token;
470 		int value;
471 		substring_t args[MAX_OPT_ARGS];
472 		if (!*p)
473 			continue;
474 
475 		token = match_token(p, tokens, args);
476 		switch (token) {
477 		case OPT_FD:
478 			if (match_int(&args[0], &value))
479 				return 0;
480 			d->fd = value;
481 			d->fd_present = 1;
482 			break;
483 
484 		case OPT_ROOTMODE:
485 			if (match_octal(&args[0], &value))
486 				return 0;
487 			if (!fuse_valid_type(value))
488 				return 0;
489 			d->rootmode = value;
490 			d->rootmode_present = 1;
491 			break;
492 
493 		case OPT_USER_ID:
494 			if (match_int(&args[0], &value))
495 				return 0;
496 			d->user_id = make_kuid(current_user_ns(), value);
497 			if (!uid_valid(d->user_id))
498 				return 0;
499 			d->user_id_present = 1;
500 			break;
501 
502 		case OPT_GROUP_ID:
503 			if (match_int(&args[0], &value))
504 				return 0;
505 			d->group_id = make_kgid(current_user_ns(), value);
506 			if (!gid_valid(d->group_id))
507 				return 0;
508 			d->group_id_present = 1;
509 			break;
510 
511 		case OPT_DEFAULT_PERMISSIONS:
512 			d->flags |= FUSE_DEFAULT_PERMISSIONS;
513 			break;
514 
515 		case OPT_ALLOW_OTHER:
516 			d->flags |= FUSE_ALLOW_OTHER;
517 			break;
518 
519 		case OPT_MAX_READ:
520 			if (match_int(&args[0], &value))
521 				return 0;
522 			d->max_read = value;
523 			break;
524 
525 		case OPT_BLKSIZE:
526 			if (!is_bdev || match_int(&args[0], &value))
527 				return 0;
528 			d->blksize = value;
529 			break;
530 
531 		default:
532 			return 0;
533 		}
534 	}
535 
536 	if (!d->fd_present || !d->rootmode_present ||
537 	    !d->user_id_present || !d->group_id_present)
538 		return 0;
539 
540 	return 1;
541 }
542 
543 static int fuse_show_options(struct seq_file *m, struct dentry *root)
544 {
545 	struct super_block *sb = root->d_sb;
546 	struct fuse_conn *fc = get_fuse_conn_super(sb);
547 
548 	seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
549 	seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
550 	if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
551 		seq_puts(m, ",default_permissions");
552 	if (fc->flags & FUSE_ALLOW_OTHER)
553 		seq_puts(m, ",allow_other");
554 	if (fc->max_read != ~0)
555 		seq_printf(m, ",max_read=%u", fc->max_read);
556 	if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
557 		seq_printf(m, ",blksize=%lu", sb->s_blocksize);
558 	return 0;
559 }
560 
561 void fuse_conn_init(struct fuse_conn *fc)
562 {
563 	memset(fc, 0, sizeof(*fc));
564 	spin_lock_init(&fc->lock);
565 	mutex_init(&fc->inst_mutex);
566 	init_rwsem(&fc->killsb);
567 	atomic_set(&fc->count, 1);
568 	init_waitqueue_head(&fc->waitq);
569 	init_waitqueue_head(&fc->blocked_waitq);
570 	init_waitqueue_head(&fc->reserved_req_waitq);
571 	INIT_LIST_HEAD(&fc->pending);
572 	INIT_LIST_HEAD(&fc->processing);
573 	INIT_LIST_HEAD(&fc->io);
574 	INIT_LIST_HEAD(&fc->interrupts);
575 	INIT_LIST_HEAD(&fc->bg_queue);
576 	INIT_LIST_HEAD(&fc->entry);
577 	fc->forget_list_tail = &fc->forget_list_head;
578 	atomic_set(&fc->num_waiting, 0);
579 	fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
580 	fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
581 	fc->khctr = 0;
582 	fc->polled_files = RB_ROOT;
583 	fc->reqctr = 0;
584 	fc->blocked = 1;
585 	fc->attr_version = 1;
586 	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
587 }
588 EXPORT_SYMBOL_GPL(fuse_conn_init);
589 
590 void fuse_conn_put(struct fuse_conn *fc)
591 {
592 	if (atomic_dec_and_test(&fc->count)) {
593 		if (fc->destroy_req)
594 			fuse_request_free(fc->destroy_req);
595 		mutex_destroy(&fc->inst_mutex);
596 		fc->release(fc);
597 	}
598 }
599 EXPORT_SYMBOL_GPL(fuse_conn_put);
600 
601 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
602 {
603 	atomic_inc(&fc->count);
604 	return fc;
605 }
606 EXPORT_SYMBOL_GPL(fuse_conn_get);
607 
608 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
609 {
610 	struct fuse_attr attr;
611 	memset(&attr, 0, sizeof(attr));
612 
613 	attr.mode = mode;
614 	attr.ino = FUSE_ROOT_ID;
615 	attr.nlink = 1;
616 	return fuse_iget(sb, 1, 0, &attr, 0, 0);
617 }
618 
619 struct fuse_inode_handle {
620 	u64 nodeid;
621 	u32 generation;
622 };
623 
624 static struct dentry *fuse_get_dentry(struct super_block *sb,
625 				      struct fuse_inode_handle *handle)
626 {
627 	struct fuse_conn *fc = get_fuse_conn_super(sb);
628 	struct inode *inode;
629 	struct dentry *entry;
630 	int err = -ESTALE;
631 
632 	if (handle->nodeid == 0)
633 		goto out_err;
634 
635 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
636 	if (!inode) {
637 		struct fuse_entry_out outarg;
638 		struct qstr name;
639 
640 		if (!fc->export_support)
641 			goto out_err;
642 
643 		name.len = 1;
644 		name.name = ".";
645 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
646 				       &inode);
647 		if (err && err != -ENOENT)
648 			goto out_err;
649 		if (err || !inode) {
650 			err = -ESTALE;
651 			goto out_err;
652 		}
653 		err = -EIO;
654 		if (get_node_id(inode) != handle->nodeid)
655 			goto out_iput;
656 	}
657 	err = -ESTALE;
658 	if (inode->i_generation != handle->generation)
659 		goto out_iput;
660 
661 	entry = d_obtain_alias(inode);
662 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
663 		fuse_invalidate_entry_cache(entry);
664 
665 	return entry;
666 
667  out_iput:
668 	iput(inode);
669  out_err:
670 	return ERR_PTR(err);
671 }
672 
673 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
674 			   struct inode *parent)
675 {
676 	int len = parent ? 6 : 3;
677 	u64 nodeid;
678 	u32 generation;
679 
680 	if (*max_len < len) {
681 		*max_len = len;
682 		return  255;
683 	}
684 
685 	nodeid = get_fuse_inode(inode)->nodeid;
686 	generation = inode->i_generation;
687 
688 	fh[0] = (u32)(nodeid >> 32);
689 	fh[1] = (u32)(nodeid & 0xffffffff);
690 	fh[2] = generation;
691 
692 	if (parent) {
693 		nodeid = get_fuse_inode(parent)->nodeid;
694 		generation = parent->i_generation;
695 
696 		fh[3] = (u32)(nodeid >> 32);
697 		fh[4] = (u32)(nodeid & 0xffffffff);
698 		fh[5] = generation;
699 	}
700 
701 	*max_len = len;
702 	return parent ? 0x82 : 0x81;
703 }
704 
705 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
706 		struct fid *fid, int fh_len, int fh_type)
707 {
708 	struct fuse_inode_handle handle;
709 
710 	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
711 		return NULL;
712 
713 	handle.nodeid = (u64) fid->raw[0] << 32;
714 	handle.nodeid |= (u64) fid->raw[1];
715 	handle.generation = fid->raw[2];
716 	return fuse_get_dentry(sb, &handle);
717 }
718 
719 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
720 		struct fid *fid, int fh_len, int fh_type)
721 {
722 	struct fuse_inode_handle parent;
723 
724 	if (fh_type != 0x82 || fh_len < 6)
725 		return NULL;
726 
727 	parent.nodeid = (u64) fid->raw[3] << 32;
728 	parent.nodeid |= (u64) fid->raw[4];
729 	parent.generation = fid->raw[5];
730 	return fuse_get_dentry(sb, &parent);
731 }
732 
733 static struct dentry *fuse_get_parent(struct dentry *child)
734 {
735 	struct inode *child_inode = child->d_inode;
736 	struct fuse_conn *fc = get_fuse_conn(child_inode);
737 	struct inode *inode;
738 	struct dentry *parent;
739 	struct fuse_entry_out outarg;
740 	struct qstr name;
741 	int err;
742 
743 	if (!fc->export_support)
744 		return ERR_PTR(-ESTALE);
745 
746 	name.len = 2;
747 	name.name = "..";
748 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
749 			       &name, &outarg, &inode);
750 	if (err) {
751 		if (err == -ENOENT)
752 			return ERR_PTR(-ESTALE);
753 		return ERR_PTR(err);
754 	}
755 
756 	parent = d_obtain_alias(inode);
757 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
758 		fuse_invalidate_entry_cache(parent);
759 
760 	return parent;
761 }
762 
763 static const struct export_operations fuse_export_operations = {
764 	.fh_to_dentry	= fuse_fh_to_dentry,
765 	.fh_to_parent	= fuse_fh_to_parent,
766 	.encode_fh	= fuse_encode_fh,
767 	.get_parent	= fuse_get_parent,
768 };
769 
770 static const struct super_operations fuse_super_operations = {
771 	.alloc_inode    = fuse_alloc_inode,
772 	.destroy_inode  = fuse_destroy_inode,
773 	.evict_inode	= fuse_evict_inode,
774 	.drop_inode	= generic_delete_inode,
775 	.remount_fs	= fuse_remount_fs,
776 	.put_super	= fuse_put_super,
777 	.umount_begin	= fuse_umount_begin,
778 	.statfs		= fuse_statfs,
779 	.show_options	= fuse_show_options,
780 };
781 
782 static void sanitize_global_limit(unsigned *limit)
783 {
784 	if (*limit == 0)
785 		*limit = ((num_physpages << PAGE_SHIFT) >> 13) /
786 			 sizeof(struct fuse_req);
787 
788 	if (*limit >= 1 << 16)
789 		*limit = (1 << 16) - 1;
790 }
791 
792 static int set_global_limit(const char *val, struct kernel_param *kp)
793 {
794 	int rv;
795 
796 	rv = param_set_uint(val, kp);
797 	if (rv)
798 		return rv;
799 
800 	sanitize_global_limit((unsigned *)kp->arg);
801 
802 	return 0;
803 }
804 
805 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
806 {
807 	int cap_sys_admin = capable(CAP_SYS_ADMIN);
808 
809 	if (arg->minor < 13)
810 		return;
811 
812 	sanitize_global_limit(&max_user_bgreq);
813 	sanitize_global_limit(&max_user_congthresh);
814 
815 	if (arg->max_background) {
816 		fc->max_background = arg->max_background;
817 
818 		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
819 			fc->max_background = max_user_bgreq;
820 	}
821 	if (arg->congestion_threshold) {
822 		fc->congestion_threshold = arg->congestion_threshold;
823 
824 		if (!cap_sys_admin &&
825 		    fc->congestion_threshold > max_user_congthresh)
826 			fc->congestion_threshold = max_user_congthresh;
827 	}
828 }
829 
830 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
831 {
832 	struct fuse_init_out *arg = &req->misc.init_out;
833 
834 	if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
835 		fc->conn_error = 1;
836 	else {
837 		unsigned long ra_pages;
838 
839 		process_init_limits(fc, arg);
840 
841 		if (arg->minor >= 6) {
842 			ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
843 			if (arg->flags & FUSE_ASYNC_READ)
844 				fc->async_read = 1;
845 			if (!(arg->flags & FUSE_POSIX_LOCKS))
846 				fc->no_lock = 1;
847 			if (arg->minor >= 17) {
848 				if (!(arg->flags & FUSE_FLOCK_LOCKS))
849 					fc->no_flock = 1;
850 			} else {
851 				if (!(arg->flags & FUSE_POSIX_LOCKS))
852 					fc->no_flock = 1;
853 			}
854 			if (arg->flags & FUSE_ATOMIC_O_TRUNC)
855 				fc->atomic_o_trunc = 1;
856 			if (arg->minor >= 9) {
857 				/* LOOKUP has dependency on proto version */
858 				if (arg->flags & FUSE_EXPORT_SUPPORT)
859 					fc->export_support = 1;
860 			}
861 			if (arg->flags & FUSE_BIG_WRITES)
862 				fc->big_writes = 1;
863 			if (arg->flags & FUSE_DONT_MASK)
864 				fc->dont_mask = 1;
865 			if (arg->flags & FUSE_AUTO_INVAL_DATA)
866 				fc->auto_inval_data = 1;
867 			if (arg->flags & FUSE_DO_READDIRPLUS)
868 				fc->do_readdirplus = 1;
869 		} else {
870 			ra_pages = fc->max_read / PAGE_CACHE_SIZE;
871 			fc->no_lock = 1;
872 			fc->no_flock = 1;
873 		}
874 
875 		fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
876 		fc->minor = arg->minor;
877 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
878 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
879 		fc->conn_init = 1;
880 	}
881 	fc->blocked = 0;
882 	wake_up_all(&fc->blocked_waitq);
883 }
884 
885 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
886 {
887 	struct fuse_init_in *arg = &req->misc.init_in;
888 
889 	arg->major = FUSE_KERNEL_VERSION;
890 	arg->minor = FUSE_KERNEL_MINOR_VERSION;
891 	arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
892 	arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
893 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
894 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
895 		FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
896 		FUSE_DO_READDIRPLUS;
897 	req->in.h.opcode = FUSE_INIT;
898 	req->in.numargs = 1;
899 	req->in.args[0].size = sizeof(*arg);
900 	req->in.args[0].value = arg;
901 	req->out.numargs = 1;
902 	/* Variable length argument used for backward compatibility
903 	   with interface version < 7.5.  Rest of init_out is zeroed
904 	   by do_get_request(), so a short reply is not a problem */
905 	req->out.argvar = 1;
906 	req->out.args[0].size = sizeof(struct fuse_init_out);
907 	req->out.args[0].value = &req->misc.init_out;
908 	req->end = process_init_reply;
909 	fuse_request_send_background(fc, req);
910 }
911 
912 static void fuse_free_conn(struct fuse_conn *fc)
913 {
914 	kfree(fc);
915 }
916 
917 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
918 {
919 	int err;
920 
921 	fc->bdi.name = "fuse";
922 	fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
923 	/* fuse does it's own writeback accounting */
924 	fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
925 
926 	err = bdi_init(&fc->bdi);
927 	if (err)
928 		return err;
929 
930 	fc->bdi_initialized = 1;
931 
932 	if (sb->s_bdev) {
933 		err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
934 				    MAJOR(fc->dev), MINOR(fc->dev));
935 	} else {
936 		err = bdi_register_dev(&fc->bdi, fc->dev);
937 	}
938 
939 	if (err)
940 		return err;
941 
942 	/*
943 	 * For a single fuse filesystem use max 1% of dirty +
944 	 * writeback threshold.
945 	 *
946 	 * This gives about 1M of write buffer for memory maps on a
947 	 * machine with 1G and 10% dirty_ratio, which should be more
948 	 * than enough.
949 	 *
950 	 * Privileged users can raise it by writing to
951 	 *
952 	 *    /sys/class/bdi/<bdi>/max_ratio
953 	 */
954 	bdi_set_max_ratio(&fc->bdi, 1);
955 
956 	return 0;
957 }
958 
959 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
960 {
961 	struct fuse_conn *fc;
962 	struct inode *root;
963 	struct fuse_mount_data d;
964 	struct file *file;
965 	struct dentry *root_dentry;
966 	struct fuse_req *init_req;
967 	int err;
968 	int is_bdev = sb->s_bdev != NULL;
969 
970 	err = -EINVAL;
971 	if (sb->s_flags & MS_MANDLOCK)
972 		goto err;
973 
974 	sb->s_flags &= ~MS_NOSEC;
975 
976 	if (!parse_fuse_opt((char *) data, &d, is_bdev))
977 		goto err;
978 
979 	if (is_bdev) {
980 #ifdef CONFIG_BLOCK
981 		err = -EINVAL;
982 		if (!sb_set_blocksize(sb, d.blksize))
983 			goto err;
984 #endif
985 	} else {
986 		sb->s_blocksize = PAGE_CACHE_SIZE;
987 		sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
988 	}
989 	sb->s_magic = FUSE_SUPER_MAGIC;
990 	sb->s_op = &fuse_super_operations;
991 	sb->s_maxbytes = MAX_LFS_FILESIZE;
992 	sb->s_time_gran = 1;
993 	sb->s_export_op = &fuse_export_operations;
994 
995 	file = fget(d.fd);
996 	err = -EINVAL;
997 	if (!file)
998 		goto err;
999 
1000 	if ((file->f_op != &fuse_dev_operations) ||
1001 	    (file->f_cred->user_ns != &init_user_ns))
1002 		goto err_fput;
1003 
1004 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1005 	err = -ENOMEM;
1006 	if (!fc)
1007 		goto err_fput;
1008 
1009 	fuse_conn_init(fc);
1010 
1011 	fc->dev = sb->s_dev;
1012 	fc->sb = sb;
1013 	err = fuse_bdi_init(fc, sb);
1014 	if (err)
1015 		goto err_put_conn;
1016 
1017 	sb->s_bdi = &fc->bdi;
1018 
1019 	/* Handle umasking inside the fuse code */
1020 	if (sb->s_flags & MS_POSIXACL)
1021 		fc->dont_mask = 1;
1022 	sb->s_flags |= MS_POSIXACL;
1023 
1024 	fc->release = fuse_free_conn;
1025 	fc->flags = d.flags;
1026 	fc->user_id = d.user_id;
1027 	fc->group_id = d.group_id;
1028 	fc->max_read = max_t(unsigned, 4096, d.max_read);
1029 
1030 	/* Used by get_root_inode() */
1031 	sb->s_fs_info = fc;
1032 
1033 	err = -ENOMEM;
1034 	root = fuse_get_root_inode(sb, d.rootmode);
1035 	root_dentry = d_make_root(root);
1036 	if (!root_dentry)
1037 		goto err_put_conn;
1038 	/* only now - we want root dentry with NULL ->d_op */
1039 	sb->s_d_op = &fuse_dentry_operations;
1040 
1041 	init_req = fuse_request_alloc(0);
1042 	if (!init_req)
1043 		goto err_put_root;
1044 
1045 	if (is_bdev) {
1046 		fc->destroy_req = fuse_request_alloc(0);
1047 		if (!fc->destroy_req)
1048 			goto err_free_init_req;
1049 	}
1050 
1051 	mutex_lock(&fuse_mutex);
1052 	err = -EINVAL;
1053 	if (file->private_data)
1054 		goto err_unlock;
1055 
1056 	err = fuse_ctl_add_conn(fc);
1057 	if (err)
1058 		goto err_unlock;
1059 
1060 	list_add_tail(&fc->entry, &fuse_conn_list);
1061 	sb->s_root = root_dentry;
1062 	fc->connected = 1;
1063 	file->private_data = fuse_conn_get(fc);
1064 	mutex_unlock(&fuse_mutex);
1065 	/*
1066 	 * atomic_dec_and_test() in fput() provides the necessary
1067 	 * memory barrier for file->private_data to be visible on all
1068 	 * CPUs after this
1069 	 */
1070 	fput(file);
1071 
1072 	fuse_send_init(fc, init_req);
1073 
1074 	return 0;
1075 
1076  err_unlock:
1077 	mutex_unlock(&fuse_mutex);
1078  err_free_init_req:
1079 	fuse_request_free(init_req);
1080  err_put_root:
1081 	dput(root_dentry);
1082  err_put_conn:
1083 	fuse_bdi_destroy(fc);
1084 	fuse_conn_put(fc);
1085  err_fput:
1086 	fput(file);
1087  err:
1088 	return err;
1089 }
1090 
1091 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1092 		       int flags, const char *dev_name,
1093 		       void *raw_data)
1094 {
1095 	return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1096 }
1097 
1098 static void fuse_kill_sb_anon(struct super_block *sb)
1099 {
1100 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1101 
1102 	if (fc) {
1103 		down_write(&fc->killsb);
1104 		fc->sb = NULL;
1105 		up_write(&fc->killsb);
1106 	}
1107 
1108 	kill_anon_super(sb);
1109 }
1110 
1111 static struct file_system_type fuse_fs_type = {
1112 	.owner		= THIS_MODULE,
1113 	.name		= "fuse",
1114 	.fs_flags	= FS_HAS_SUBTYPE,
1115 	.mount		= fuse_mount,
1116 	.kill_sb	= fuse_kill_sb_anon,
1117 };
1118 
1119 #ifdef CONFIG_BLOCK
1120 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1121 			   int flags, const char *dev_name,
1122 			   void *raw_data)
1123 {
1124 	return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1125 }
1126 
1127 static void fuse_kill_sb_blk(struct super_block *sb)
1128 {
1129 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1130 
1131 	if (fc) {
1132 		down_write(&fc->killsb);
1133 		fc->sb = NULL;
1134 		up_write(&fc->killsb);
1135 	}
1136 
1137 	kill_block_super(sb);
1138 }
1139 
1140 static struct file_system_type fuseblk_fs_type = {
1141 	.owner		= THIS_MODULE,
1142 	.name		= "fuseblk",
1143 	.mount		= fuse_mount_blk,
1144 	.kill_sb	= fuse_kill_sb_blk,
1145 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1146 };
1147 
1148 static inline int register_fuseblk(void)
1149 {
1150 	return register_filesystem(&fuseblk_fs_type);
1151 }
1152 
1153 static inline void unregister_fuseblk(void)
1154 {
1155 	unregister_filesystem(&fuseblk_fs_type);
1156 }
1157 #else
1158 static inline int register_fuseblk(void)
1159 {
1160 	return 0;
1161 }
1162 
1163 static inline void unregister_fuseblk(void)
1164 {
1165 }
1166 #endif
1167 
1168 static void fuse_inode_init_once(void *foo)
1169 {
1170 	struct inode *inode = foo;
1171 
1172 	inode_init_once(inode);
1173 }
1174 
1175 static int __init fuse_fs_init(void)
1176 {
1177 	int err;
1178 
1179 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
1180 					      sizeof(struct fuse_inode),
1181 					      0, SLAB_HWCACHE_ALIGN,
1182 					      fuse_inode_init_once);
1183 	err = -ENOMEM;
1184 	if (!fuse_inode_cachep)
1185 		goto out;
1186 
1187 	err = register_fuseblk();
1188 	if (err)
1189 		goto out2;
1190 
1191 	err = register_filesystem(&fuse_fs_type);
1192 	if (err)
1193 		goto out3;
1194 
1195 	return 0;
1196 
1197  out3:
1198 	unregister_fuseblk();
1199  out2:
1200 	kmem_cache_destroy(fuse_inode_cachep);
1201  out:
1202 	return err;
1203 }
1204 
1205 static void fuse_fs_cleanup(void)
1206 {
1207 	unregister_filesystem(&fuse_fs_type);
1208 	unregister_fuseblk();
1209 
1210 	/*
1211 	 * Make sure all delayed rcu free inodes are flushed before we
1212 	 * destroy cache.
1213 	 */
1214 	rcu_barrier();
1215 	kmem_cache_destroy(fuse_inode_cachep);
1216 }
1217 
1218 static struct kobject *fuse_kobj;
1219 static struct kobject *connections_kobj;
1220 
1221 static int fuse_sysfs_init(void)
1222 {
1223 	int err;
1224 
1225 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1226 	if (!fuse_kobj) {
1227 		err = -ENOMEM;
1228 		goto out_err;
1229 	}
1230 
1231 	connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1232 	if (!connections_kobj) {
1233 		err = -ENOMEM;
1234 		goto out_fuse_unregister;
1235 	}
1236 
1237 	return 0;
1238 
1239  out_fuse_unregister:
1240 	kobject_put(fuse_kobj);
1241  out_err:
1242 	return err;
1243 }
1244 
1245 static void fuse_sysfs_cleanup(void)
1246 {
1247 	kobject_put(connections_kobj);
1248 	kobject_put(fuse_kobj);
1249 }
1250 
1251 static int __init fuse_init(void)
1252 {
1253 	int res;
1254 
1255 	printk(KERN_INFO "fuse init (API version %i.%i)\n",
1256 	       FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1257 
1258 	INIT_LIST_HEAD(&fuse_conn_list);
1259 	res = fuse_fs_init();
1260 	if (res)
1261 		goto err;
1262 
1263 	res = fuse_dev_init();
1264 	if (res)
1265 		goto err_fs_cleanup;
1266 
1267 	res = fuse_sysfs_init();
1268 	if (res)
1269 		goto err_dev_cleanup;
1270 
1271 	res = fuse_ctl_init();
1272 	if (res)
1273 		goto err_sysfs_cleanup;
1274 
1275 	sanitize_global_limit(&max_user_bgreq);
1276 	sanitize_global_limit(&max_user_congthresh);
1277 
1278 	return 0;
1279 
1280  err_sysfs_cleanup:
1281 	fuse_sysfs_cleanup();
1282  err_dev_cleanup:
1283 	fuse_dev_cleanup();
1284  err_fs_cleanup:
1285 	fuse_fs_cleanup();
1286  err:
1287 	return res;
1288 }
1289 
1290 static void __exit fuse_exit(void)
1291 {
1292 	printk(KERN_DEBUG "fuse exit\n");
1293 
1294 	fuse_ctl_cleanup();
1295 	fuse_sysfs_cleanup();
1296 	fuse_fs_cleanup();
1297 	fuse_dev_cleanup();
1298 }
1299 
1300 module_init(fuse_init);
1301 module_exit(fuse_exit);
1302