xref: /openbmc/linux/fs/fuse/inode.c (revision 2d6bed9c)
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 	unsigned user_id;
64 	unsigned 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 	INIT_LIST_HEAD(&fi->write_files);
96 	INIT_LIST_HEAD(&fi->queued_writes);
97 	INIT_LIST_HEAD(&fi->writepages);
98 	init_waitqueue_head(&fi->page_waitq);
99 	fi->forget = fuse_alloc_forget();
100 	if (!fi->forget) {
101 		kmem_cache_free(fuse_inode_cachep, inode);
102 		return NULL;
103 	}
104 
105 	return inode;
106 }
107 
108 static void fuse_i_callback(struct rcu_head *head)
109 {
110 	struct inode *inode = container_of(head, struct inode, i_rcu);
111 	kmem_cache_free(fuse_inode_cachep, inode);
112 }
113 
114 static void fuse_destroy_inode(struct inode *inode)
115 {
116 	struct fuse_inode *fi = get_fuse_inode(inode);
117 	BUG_ON(!list_empty(&fi->write_files));
118 	BUG_ON(!list_empty(&fi->queued_writes));
119 	kfree(fi->forget);
120 	call_rcu(&inode->i_rcu, fuse_i_callback);
121 }
122 
123 static void fuse_evict_inode(struct inode *inode)
124 {
125 	truncate_inode_pages(&inode->i_data, 0);
126 	clear_inode(inode);
127 	if (inode->i_sb->s_flags & MS_ACTIVE) {
128 		struct fuse_conn *fc = get_fuse_conn(inode);
129 		struct fuse_inode *fi = get_fuse_inode(inode);
130 		fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
131 		fi->forget = NULL;
132 	}
133 }
134 
135 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
136 {
137 	if (*flags & MS_MANDLOCK)
138 		return -EINVAL;
139 
140 	return 0;
141 }
142 
143 /*
144  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
145  * so that it will fit.
146  */
147 static ino_t fuse_squash_ino(u64 ino64)
148 {
149 	ino_t ino = (ino_t) ino64;
150 	if (sizeof(ino_t) < sizeof(u64))
151 		ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
152 	return ino;
153 }
154 
155 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
156 				   u64 attr_valid)
157 {
158 	struct fuse_conn *fc = get_fuse_conn(inode);
159 	struct fuse_inode *fi = get_fuse_inode(inode);
160 
161 	fi->attr_version = ++fc->attr_version;
162 	fi->i_time = attr_valid;
163 
164 	inode->i_ino     = fuse_squash_ino(attr->ino);
165 	inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
166 	set_nlink(inode, attr->nlink);
167 	inode->i_uid     = attr->uid;
168 	inode->i_gid     = attr->gid;
169 	inode->i_blocks  = attr->blocks;
170 	inode->i_atime.tv_sec   = attr->atime;
171 	inode->i_atime.tv_nsec  = attr->atimensec;
172 	inode->i_mtime.tv_sec   = attr->mtime;
173 	inode->i_mtime.tv_nsec  = attr->mtimensec;
174 	inode->i_ctime.tv_sec   = attr->ctime;
175 	inode->i_ctime.tv_nsec  = attr->ctimensec;
176 
177 	if (attr->blksize != 0)
178 		inode->i_blkbits = ilog2(attr->blksize);
179 	else
180 		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
181 
182 	/*
183 	 * Don't set the sticky bit in i_mode, unless we want the VFS
184 	 * to check permissions.  This prevents failures due to the
185 	 * check in may_delete().
186 	 */
187 	fi->orig_i_mode = inode->i_mode;
188 	if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
189 		inode->i_mode &= ~S_ISVTX;
190 
191 	fi->orig_ino = attr->ino;
192 }
193 
194 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
195 			    u64 attr_valid, u64 attr_version)
196 {
197 	struct fuse_conn *fc = get_fuse_conn(inode);
198 	struct fuse_inode *fi = get_fuse_inode(inode);
199 	loff_t oldsize;
200 	struct timespec old_mtime;
201 
202 	spin_lock(&fc->lock);
203 	if (attr_version != 0 && fi->attr_version > attr_version) {
204 		spin_unlock(&fc->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 	i_size_write(inode, attr->size);
213 	spin_unlock(&fc->lock);
214 
215 	if (S_ISREG(inode->i_mode)) {
216 		bool inval = false;
217 
218 		if (oldsize != attr->size) {
219 			truncate_pagecache(inode, oldsize, attr->size);
220 			inval = true;
221 		} else if (fc->auto_inval_data) {
222 			struct timespec new_mtime = {
223 				.tv_sec = attr->mtime,
224 				.tv_nsec = attr->mtimensec,
225 			};
226 
227 			/*
228 			 * Auto inval mode also checks and invalidates if mtime
229 			 * has changed.
230 			 */
231 			if (!timespec_equal(&old_mtime, &new_mtime))
232 				inval = true;
233 		}
234 
235 		if (inval)
236 			invalidate_inode_pages2(inode->i_mapping);
237 	}
238 }
239 
240 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
241 {
242 	inode->i_mode = attr->mode & S_IFMT;
243 	inode->i_size = attr->size;
244 	if (S_ISREG(inode->i_mode)) {
245 		fuse_init_common(inode);
246 		fuse_init_file_inode(inode);
247 	} else if (S_ISDIR(inode->i_mode))
248 		fuse_init_dir(inode);
249 	else if (S_ISLNK(inode->i_mode))
250 		fuse_init_symlink(inode);
251 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
252 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
253 		fuse_init_common(inode);
254 		init_special_inode(inode, inode->i_mode,
255 				   new_decode_dev(attr->rdev));
256 	} else
257 		BUG();
258 }
259 
260 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
261 {
262 	u64 nodeid = *(u64 *) _nodeidp;
263 	if (get_node_id(inode) == nodeid)
264 		return 1;
265 	else
266 		return 0;
267 }
268 
269 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
270 {
271 	u64 nodeid = *(u64 *) _nodeidp;
272 	get_fuse_inode(inode)->nodeid = nodeid;
273 	return 0;
274 }
275 
276 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
277 			int generation, struct fuse_attr *attr,
278 			u64 attr_valid, u64 attr_version)
279 {
280 	struct inode *inode;
281 	struct fuse_inode *fi;
282 	struct fuse_conn *fc = get_fuse_conn_super(sb);
283 
284  retry:
285 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
286 	if (!inode)
287 		return NULL;
288 
289 	if ((inode->i_state & I_NEW)) {
290 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
291 		inode->i_generation = generation;
292 		inode->i_data.backing_dev_info = &fc->bdi;
293 		fuse_init_inode(inode, attr);
294 		unlock_new_inode(inode);
295 	} else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
296 		/* Inode has changed type, any I/O on the old should fail */
297 		make_bad_inode(inode);
298 		iput(inode);
299 		goto retry;
300 	}
301 
302 	fi = get_fuse_inode(inode);
303 	spin_lock(&fc->lock);
304 	fi->nlookup++;
305 	spin_unlock(&fc->lock);
306 	fuse_change_attributes(inode, attr, attr_valid, attr_version);
307 
308 	return inode;
309 }
310 
311 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
312 			     loff_t offset, loff_t len)
313 {
314 	struct inode *inode;
315 	pgoff_t pg_start;
316 	pgoff_t pg_end;
317 
318 	inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
319 	if (!inode)
320 		return -ENOENT;
321 
322 	fuse_invalidate_attr(inode);
323 	if (offset >= 0) {
324 		pg_start = offset >> PAGE_CACHE_SHIFT;
325 		if (len <= 0)
326 			pg_end = -1;
327 		else
328 			pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
329 		invalidate_inode_pages2_range(inode->i_mapping,
330 					      pg_start, pg_end);
331 	}
332 	iput(inode);
333 	return 0;
334 }
335 
336 static void fuse_umount_begin(struct super_block *sb)
337 {
338 	fuse_abort_conn(get_fuse_conn_super(sb));
339 }
340 
341 static void fuse_send_destroy(struct fuse_conn *fc)
342 {
343 	struct fuse_req *req = fc->destroy_req;
344 	if (req && fc->conn_init) {
345 		fc->destroy_req = NULL;
346 		req->in.h.opcode = FUSE_DESTROY;
347 		req->force = 1;
348 		fuse_request_send(fc, req);
349 		fuse_put_request(fc, req);
350 	}
351 }
352 
353 static void fuse_bdi_destroy(struct fuse_conn *fc)
354 {
355 	if (fc->bdi_initialized)
356 		bdi_destroy(&fc->bdi);
357 }
358 
359 void fuse_conn_kill(struct fuse_conn *fc)
360 {
361 	spin_lock(&fc->lock);
362 	fc->connected = 0;
363 	fc->blocked = 0;
364 	spin_unlock(&fc->lock);
365 	/* Flush all readers on this fs */
366 	kill_fasync(&fc->fasync, SIGIO, POLL_IN);
367 	wake_up_all(&fc->waitq);
368 	wake_up_all(&fc->blocked_waitq);
369 	wake_up_all(&fc->reserved_req_waitq);
370 }
371 EXPORT_SYMBOL_GPL(fuse_conn_kill);
372 
373 static void fuse_put_super(struct super_block *sb)
374 {
375 	struct fuse_conn *fc = get_fuse_conn_super(sb);
376 
377 	fuse_send_destroy(fc);
378 
379 	fuse_conn_kill(fc);
380 	mutex_lock(&fuse_mutex);
381 	list_del(&fc->entry);
382 	fuse_ctl_remove_conn(fc);
383 	mutex_unlock(&fuse_mutex);
384 	fuse_bdi_destroy(fc);
385 
386 	fuse_conn_put(fc);
387 }
388 
389 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
390 {
391 	stbuf->f_type    = FUSE_SUPER_MAGIC;
392 	stbuf->f_bsize   = attr->bsize;
393 	stbuf->f_frsize  = attr->frsize;
394 	stbuf->f_blocks  = attr->blocks;
395 	stbuf->f_bfree   = attr->bfree;
396 	stbuf->f_bavail  = attr->bavail;
397 	stbuf->f_files   = attr->files;
398 	stbuf->f_ffree   = attr->ffree;
399 	stbuf->f_namelen = attr->namelen;
400 	/* fsid is left zero */
401 }
402 
403 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
404 {
405 	struct super_block *sb = dentry->d_sb;
406 	struct fuse_conn *fc = get_fuse_conn_super(sb);
407 	struct fuse_req *req;
408 	struct fuse_statfs_out outarg;
409 	int err;
410 
411 	if (!fuse_allow_task(fc, current)) {
412 		buf->f_type = FUSE_SUPER_MAGIC;
413 		return 0;
414 	}
415 
416 	req = fuse_get_req(fc);
417 	if (IS_ERR(req))
418 		return PTR_ERR(req);
419 
420 	memset(&outarg, 0, sizeof(outarg));
421 	req->in.numargs = 0;
422 	req->in.h.opcode = FUSE_STATFS;
423 	req->in.h.nodeid = get_node_id(dentry->d_inode);
424 	req->out.numargs = 1;
425 	req->out.args[0].size =
426 		fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
427 	req->out.args[0].value = &outarg;
428 	fuse_request_send(fc, req);
429 	err = req->out.h.error;
430 	if (!err)
431 		convert_fuse_statfs(buf, &outarg.st);
432 	fuse_put_request(fc, req);
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 parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
461 {
462 	char *p;
463 	memset(d, 0, sizeof(struct fuse_mount_data));
464 	d->max_read = ~0;
465 	d->blksize = FUSE_DEFAULT_BLKSIZE;
466 
467 	while ((p = strsep(&opt, ",")) != NULL) {
468 		int token;
469 		int value;
470 		substring_t args[MAX_OPT_ARGS];
471 		if (!*p)
472 			continue;
473 
474 		token = match_token(p, tokens, args);
475 		switch (token) {
476 		case OPT_FD:
477 			if (match_int(&args[0], &value))
478 				return 0;
479 			d->fd = value;
480 			d->fd_present = 1;
481 			break;
482 
483 		case OPT_ROOTMODE:
484 			if (match_octal(&args[0], &value))
485 				return 0;
486 			if (!fuse_valid_type(value))
487 				return 0;
488 			d->rootmode = value;
489 			d->rootmode_present = 1;
490 			break;
491 
492 		case OPT_USER_ID:
493 			if (match_int(&args[0], &value))
494 				return 0;
495 			d->user_id = value;
496 			d->user_id_present = 1;
497 			break;
498 
499 		case OPT_GROUP_ID:
500 			if (match_int(&args[0], &value))
501 				return 0;
502 			d->group_id = value;
503 			d->group_id_present = 1;
504 			break;
505 
506 		case OPT_DEFAULT_PERMISSIONS:
507 			d->flags |= FUSE_DEFAULT_PERMISSIONS;
508 			break;
509 
510 		case OPT_ALLOW_OTHER:
511 			d->flags |= FUSE_ALLOW_OTHER;
512 			break;
513 
514 		case OPT_MAX_READ:
515 			if (match_int(&args[0], &value))
516 				return 0;
517 			d->max_read = value;
518 			break;
519 
520 		case OPT_BLKSIZE:
521 			if (!is_bdev || match_int(&args[0], &value))
522 				return 0;
523 			d->blksize = value;
524 			break;
525 
526 		default:
527 			return 0;
528 		}
529 	}
530 
531 	if (!d->fd_present || !d->rootmode_present ||
532 	    !d->user_id_present || !d->group_id_present)
533 		return 0;
534 
535 	return 1;
536 }
537 
538 static int fuse_show_options(struct seq_file *m, struct dentry *root)
539 {
540 	struct super_block *sb = root->d_sb;
541 	struct fuse_conn *fc = get_fuse_conn_super(sb);
542 
543 	seq_printf(m, ",user_id=%u", fc->user_id);
544 	seq_printf(m, ",group_id=%u", fc->group_id);
545 	if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
546 		seq_puts(m, ",default_permissions");
547 	if (fc->flags & FUSE_ALLOW_OTHER)
548 		seq_puts(m, ",allow_other");
549 	if (fc->max_read != ~0)
550 		seq_printf(m, ",max_read=%u", fc->max_read);
551 	if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
552 		seq_printf(m, ",blksize=%lu", sb->s_blocksize);
553 	return 0;
554 }
555 
556 void fuse_conn_init(struct fuse_conn *fc)
557 {
558 	memset(fc, 0, sizeof(*fc));
559 	spin_lock_init(&fc->lock);
560 	mutex_init(&fc->inst_mutex);
561 	init_rwsem(&fc->killsb);
562 	atomic_set(&fc->count, 1);
563 	init_waitqueue_head(&fc->waitq);
564 	init_waitqueue_head(&fc->blocked_waitq);
565 	init_waitqueue_head(&fc->reserved_req_waitq);
566 	INIT_LIST_HEAD(&fc->pending);
567 	INIT_LIST_HEAD(&fc->processing);
568 	INIT_LIST_HEAD(&fc->io);
569 	INIT_LIST_HEAD(&fc->interrupts);
570 	INIT_LIST_HEAD(&fc->bg_queue);
571 	INIT_LIST_HEAD(&fc->entry);
572 	fc->forget_list_tail = &fc->forget_list_head;
573 	atomic_set(&fc->num_waiting, 0);
574 	fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
575 	fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
576 	fc->khctr = 0;
577 	fc->polled_files = RB_ROOT;
578 	fc->reqctr = 0;
579 	fc->blocked = 1;
580 	fc->attr_version = 1;
581 	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
582 }
583 EXPORT_SYMBOL_GPL(fuse_conn_init);
584 
585 void fuse_conn_put(struct fuse_conn *fc)
586 {
587 	if (atomic_dec_and_test(&fc->count)) {
588 		if (fc->destroy_req)
589 			fuse_request_free(fc->destroy_req);
590 		mutex_destroy(&fc->inst_mutex);
591 		fc->release(fc);
592 	}
593 }
594 EXPORT_SYMBOL_GPL(fuse_conn_put);
595 
596 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
597 {
598 	atomic_inc(&fc->count);
599 	return fc;
600 }
601 EXPORT_SYMBOL_GPL(fuse_conn_get);
602 
603 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
604 {
605 	struct fuse_attr attr;
606 	memset(&attr, 0, sizeof(attr));
607 
608 	attr.mode = mode;
609 	attr.ino = FUSE_ROOT_ID;
610 	attr.nlink = 1;
611 	return fuse_iget(sb, 1, 0, &attr, 0, 0);
612 }
613 
614 struct fuse_inode_handle {
615 	u64 nodeid;
616 	u32 generation;
617 };
618 
619 static struct dentry *fuse_get_dentry(struct super_block *sb,
620 				      struct fuse_inode_handle *handle)
621 {
622 	struct fuse_conn *fc = get_fuse_conn_super(sb);
623 	struct inode *inode;
624 	struct dentry *entry;
625 	int err = -ESTALE;
626 
627 	if (handle->nodeid == 0)
628 		goto out_err;
629 
630 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
631 	if (!inode) {
632 		struct fuse_entry_out outarg;
633 		struct qstr name;
634 
635 		if (!fc->export_support)
636 			goto out_err;
637 
638 		name.len = 1;
639 		name.name = ".";
640 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
641 				       &inode);
642 		if (err && err != -ENOENT)
643 			goto out_err;
644 		if (err || !inode) {
645 			err = -ESTALE;
646 			goto out_err;
647 		}
648 		err = -EIO;
649 		if (get_node_id(inode) != handle->nodeid)
650 			goto out_iput;
651 	}
652 	err = -ESTALE;
653 	if (inode->i_generation != handle->generation)
654 		goto out_iput;
655 
656 	entry = d_obtain_alias(inode);
657 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
658 		fuse_invalidate_entry_cache(entry);
659 
660 	return entry;
661 
662  out_iput:
663 	iput(inode);
664  out_err:
665 	return ERR_PTR(err);
666 }
667 
668 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
669 			   struct inode *parent)
670 {
671 	int len = parent ? 6 : 3;
672 	u64 nodeid;
673 	u32 generation;
674 
675 	if (*max_len < len) {
676 		*max_len = len;
677 		return  255;
678 	}
679 
680 	nodeid = get_fuse_inode(inode)->nodeid;
681 	generation = inode->i_generation;
682 
683 	fh[0] = (u32)(nodeid >> 32);
684 	fh[1] = (u32)(nodeid & 0xffffffff);
685 	fh[2] = generation;
686 
687 	if (parent) {
688 		nodeid = get_fuse_inode(parent)->nodeid;
689 		generation = parent->i_generation;
690 
691 		fh[3] = (u32)(nodeid >> 32);
692 		fh[4] = (u32)(nodeid & 0xffffffff);
693 		fh[5] = generation;
694 	}
695 
696 	*max_len = len;
697 	return parent ? 0x82 : 0x81;
698 }
699 
700 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
701 		struct fid *fid, int fh_len, int fh_type)
702 {
703 	struct fuse_inode_handle handle;
704 
705 	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
706 		return NULL;
707 
708 	handle.nodeid = (u64) fid->raw[0] << 32;
709 	handle.nodeid |= (u64) fid->raw[1];
710 	handle.generation = fid->raw[2];
711 	return fuse_get_dentry(sb, &handle);
712 }
713 
714 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
715 		struct fid *fid, int fh_len, int fh_type)
716 {
717 	struct fuse_inode_handle parent;
718 
719 	if (fh_type != 0x82 || fh_len < 6)
720 		return NULL;
721 
722 	parent.nodeid = (u64) fid->raw[3] << 32;
723 	parent.nodeid |= (u64) fid->raw[4];
724 	parent.generation = fid->raw[5];
725 	return fuse_get_dentry(sb, &parent);
726 }
727 
728 static struct dentry *fuse_get_parent(struct dentry *child)
729 {
730 	struct inode *child_inode = child->d_inode;
731 	struct fuse_conn *fc = get_fuse_conn(child_inode);
732 	struct inode *inode;
733 	struct dentry *parent;
734 	struct fuse_entry_out outarg;
735 	struct qstr name;
736 	int err;
737 
738 	if (!fc->export_support)
739 		return ERR_PTR(-ESTALE);
740 
741 	name.len = 2;
742 	name.name = "..";
743 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
744 			       &name, &outarg, &inode);
745 	if (err) {
746 		if (err == -ENOENT)
747 			return ERR_PTR(-ESTALE);
748 		return ERR_PTR(err);
749 	}
750 
751 	parent = d_obtain_alias(inode);
752 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
753 		fuse_invalidate_entry_cache(parent);
754 
755 	return parent;
756 }
757 
758 static const struct export_operations fuse_export_operations = {
759 	.fh_to_dentry	= fuse_fh_to_dentry,
760 	.fh_to_parent	= fuse_fh_to_parent,
761 	.encode_fh	= fuse_encode_fh,
762 	.get_parent	= fuse_get_parent,
763 };
764 
765 static const struct super_operations fuse_super_operations = {
766 	.alloc_inode    = fuse_alloc_inode,
767 	.destroy_inode  = fuse_destroy_inode,
768 	.evict_inode	= fuse_evict_inode,
769 	.drop_inode	= generic_delete_inode,
770 	.remount_fs	= fuse_remount_fs,
771 	.put_super	= fuse_put_super,
772 	.umount_begin	= fuse_umount_begin,
773 	.statfs		= fuse_statfs,
774 	.show_options	= fuse_show_options,
775 };
776 
777 static void sanitize_global_limit(unsigned *limit)
778 {
779 	if (*limit == 0)
780 		*limit = ((num_physpages << PAGE_SHIFT) >> 13) /
781 			 sizeof(struct fuse_req);
782 
783 	if (*limit >= 1 << 16)
784 		*limit = (1 << 16) - 1;
785 }
786 
787 static int set_global_limit(const char *val, struct kernel_param *kp)
788 {
789 	int rv;
790 
791 	rv = param_set_uint(val, kp);
792 	if (rv)
793 		return rv;
794 
795 	sanitize_global_limit((unsigned *)kp->arg);
796 
797 	return 0;
798 }
799 
800 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
801 {
802 	int cap_sys_admin = capable(CAP_SYS_ADMIN);
803 
804 	if (arg->minor < 13)
805 		return;
806 
807 	sanitize_global_limit(&max_user_bgreq);
808 	sanitize_global_limit(&max_user_congthresh);
809 
810 	if (arg->max_background) {
811 		fc->max_background = arg->max_background;
812 
813 		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
814 			fc->max_background = max_user_bgreq;
815 	}
816 	if (arg->congestion_threshold) {
817 		fc->congestion_threshold = arg->congestion_threshold;
818 
819 		if (!cap_sys_admin &&
820 		    fc->congestion_threshold > max_user_congthresh)
821 			fc->congestion_threshold = max_user_congthresh;
822 	}
823 }
824 
825 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
826 {
827 	struct fuse_init_out *arg = &req->misc.init_out;
828 
829 	if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
830 		fc->conn_error = 1;
831 	else {
832 		unsigned long ra_pages;
833 
834 		process_init_limits(fc, arg);
835 
836 		if (arg->minor >= 6) {
837 			ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
838 			if (arg->flags & FUSE_ASYNC_READ)
839 				fc->async_read = 1;
840 			if (!(arg->flags & FUSE_POSIX_LOCKS))
841 				fc->no_lock = 1;
842 			if (arg->minor >= 17) {
843 				if (!(arg->flags & FUSE_FLOCK_LOCKS))
844 					fc->no_flock = 1;
845 			} else {
846 				if (!(arg->flags & FUSE_POSIX_LOCKS))
847 					fc->no_flock = 1;
848 			}
849 			if (arg->flags & FUSE_ATOMIC_O_TRUNC)
850 				fc->atomic_o_trunc = 1;
851 			if (arg->minor >= 9) {
852 				/* LOOKUP has dependency on proto version */
853 				if (arg->flags & FUSE_EXPORT_SUPPORT)
854 					fc->export_support = 1;
855 			}
856 			if (arg->flags & FUSE_BIG_WRITES)
857 				fc->big_writes = 1;
858 			if (arg->flags & FUSE_DONT_MASK)
859 				fc->dont_mask = 1;
860 			if (arg->flags & FUSE_AUTO_INVAL_DATA)
861 				fc->auto_inval_data = 1;
862 		} else {
863 			ra_pages = fc->max_read / PAGE_CACHE_SIZE;
864 			fc->no_lock = 1;
865 			fc->no_flock = 1;
866 		}
867 
868 		fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
869 		fc->minor = arg->minor;
870 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
871 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
872 		fc->conn_init = 1;
873 	}
874 	fc->blocked = 0;
875 	wake_up_all(&fc->blocked_waitq);
876 }
877 
878 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
879 {
880 	struct fuse_init_in *arg = &req->misc.init_in;
881 
882 	arg->major = FUSE_KERNEL_VERSION;
883 	arg->minor = FUSE_KERNEL_MINOR_VERSION;
884 	arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
885 	arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
886 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
887 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
888 		FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA;
889 	req->in.h.opcode = FUSE_INIT;
890 	req->in.numargs = 1;
891 	req->in.args[0].size = sizeof(*arg);
892 	req->in.args[0].value = arg;
893 	req->out.numargs = 1;
894 	/* Variable length argument used for backward compatibility
895 	   with interface version < 7.5.  Rest of init_out is zeroed
896 	   by do_get_request(), so a short reply is not a problem */
897 	req->out.argvar = 1;
898 	req->out.args[0].size = sizeof(struct fuse_init_out);
899 	req->out.args[0].value = &req->misc.init_out;
900 	req->end = process_init_reply;
901 	fuse_request_send_background(fc, req);
902 }
903 
904 static void fuse_free_conn(struct fuse_conn *fc)
905 {
906 	kfree(fc);
907 }
908 
909 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
910 {
911 	int err;
912 
913 	fc->bdi.name = "fuse";
914 	fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
915 	/* fuse does it's own writeback accounting */
916 	fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
917 
918 	err = bdi_init(&fc->bdi);
919 	if (err)
920 		return err;
921 
922 	fc->bdi_initialized = 1;
923 
924 	if (sb->s_bdev) {
925 		err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
926 				    MAJOR(fc->dev), MINOR(fc->dev));
927 	} else {
928 		err = bdi_register_dev(&fc->bdi, fc->dev);
929 	}
930 
931 	if (err)
932 		return err;
933 
934 	/*
935 	 * For a single fuse filesystem use max 1% of dirty +
936 	 * writeback threshold.
937 	 *
938 	 * This gives about 1M of write buffer for memory maps on a
939 	 * machine with 1G and 10% dirty_ratio, which should be more
940 	 * than enough.
941 	 *
942 	 * Privileged users can raise it by writing to
943 	 *
944 	 *    /sys/class/bdi/<bdi>/max_ratio
945 	 */
946 	bdi_set_max_ratio(&fc->bdi, 1);
947 
948 	return 0;
949 }
950 
951 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
952 {
953 	struct fuse_conn *fc;
954 	struct inode *root;
955 	struct fuse_mount_data d;
956 	struct file *file;
957 	struct dentry *root_dentry;
958 	struct fuse_req *init_req;
959 	int err;
960 	int is_bdev = sb->s_bdev != NULL;
961 
962 	err = -EINVAL;
963 	if (sb->s_flags & MS_MANDLOCK)
964 		goto err;
965 
966 	sb->s_flags &= ~MS_NOSEC;
967 
968 	if (!parse_fuse_opt((char *) data, &d, is_bdev))
969 		goto err;
970 
971 	if (is_bdev) {
972 #ifdef CONFIG_BLOCK
973 		err = -EINVAL;
974 		if (!sb_set_blocksize(sb, d.blksize))
975 			goto err;
976 #endif
977 	} else {
978 		sb->s_blocksize = PAGE_CACHE_SIZE;
979 		sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
980 	}
981 	sb->s_magic = FUSE_SUPER_MAGIC;
982 	sb->s_op = &fuse_super_operations;
983 	sb->s_maxbytes = MAX_LFS_FILESIZE;
984 	sb->s_time_gran = 1;
985 	sb->s_export_op = &fuse_export_operations;
986 
987 	file = fget(d.fd);
988 	err = -EINVAL;
989 	if (!file)
990 		goto err;
991 
992 	if (file->f_op != &fuse_dev_operations)
993 		goto err_fput;
994 
995 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
996 	err = -ENOMEM;
997 	if (!fc)
998 		goto err_fput;
999 
1000 	fuse_conn_init(fc);
1001 
1002 	fc->dev = sb->s_dev;
1003 	fc->sb = sb;
1004 	err = fuse_bdi_init(fc, sb);
1005 	if (err)
1006 		goto err_put_conn;
1007 
1008 	sb->s_bdi = &fc->bdi;
1009 
1010 	/* Handle umasking inside the fuse code */
1011 	if (sb->s_flags & MS_POSIXACL)
1012 		fc->dont_mask = 1;
1013 	sb->s_flags |= MS_POSIXACL;
1014 
1015 	fc->release = fuse_free_conn;
1016 	fc->flags = d.flags;
1017 	fc->user_id = d.user_id;
1018 	fc->group_id = d.group_id;
1019 	fc->max_read = max_t(unsigned, 4096, d.max_read);
1020 
1021 	/* Used by get_root_inode() */
1022 	sb->s_fs_info = fc;
1023 
1024 	err = -ENOMEM;
1025 	root = fuse_get_root_inode(sb, d.rootmode);
1026 	root_dentry = d_make_root(root);
1027 	if (!root_dentry)
1028 		goto err_put_conn;
1029 	/* only now - we want root dentry with NULL ->d_op */
1030 	sb->s_d_op = &fuse_dentry_operations;
1031 
1032 	init_req = fuse_request_alloc();
1033 	if (!init_req)
1034 		goto err_put_root;
1035 
1036 	if (is_bdev) {
1037 		fc->destroy_req = fuse_request_alloc();
1038 		if (!fc->destroy_req)
1039 			goto err_free_init_req;
1040 	}
1041 
1042 	mutex_lock(&fuse_mutex);
1043 	err = -EINVAL;
1044 	if (file->private_data)
1045 		goto err_unlock;
1046 
1047 	err = fuse_ctl_add_conn(fc);
1048 	if (err)
1049 		goto err_unlock;
1050 
1051 	list_add_tail(&fc->entry, &fuse_conn_list);
1052 	sb->s_root = root_dentry;
1053 	fc->connected = 1;
1054 	file->private_data = fuse_conn_get(fc);
1055 	mutex_unlock(&fuse_mutex);
1056 	/*
1057 	 * atomic_dec_and_test() in fput() provides the necessary
1058 	 * memory barrier for file->private_data to be visible on all
1059 	 * CPUs after this
1060 	 */
1061 	fput(file);
1062 
1063 	fuse_send_init(fc, init_req);
1064 
1065 	return 0;
1066 
1067  err_unlock:
1068 	mutex_unlock(&fuse_mutex);
1069  err_free_init_req:
1070 	fuse_request_free(init_req);
1071  err_put_root:
1072 	dput(root_dentry);
1073  err_put_conn:
1074 	fuse_bdi_destroy(fc);
1075 	fuse_conn_put(fc);
1076  err_fput:
1077 	fput(file);
1078  err:
1079 	return err;
1080 }
1081 
1082 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1083 		       int flags, const char *dev_name,
1084 		       void *raw_data)
1085 {
1086 	return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1087 }
1088 
1089 static void fuse_kill_sb_anon(struct super_block *sb)
1090 {
1091 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1092 
1093 	if (fc) {
1094 		down_write(&fc->killsb);
1095 		fc->sb = NULL;
1096 		up_write(&fc->killsb);
1097 	}
1098 
1099 	kill_anon_super(sb);
1100 }
1101 
1102 static struct file_system_type fuse_fs_type = {
1103 	.owner		= THIS_MODULE,
1104 	.name		= "fuse",
1105 	.fs_flags	= FS_HAS_SUBTYPE,
1106 	.mount		= fuse_mount,
1107 	.kill_sb	= fuse_kill_sb_anon,
1108 };
1109 
1110 #ifdef CONFIG_BLOCK
1111 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1112 			   int flags, const char *dev_name,
1113 			   void *raw_data)
1114 {
1115 	return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1116 }
1117 
1118 static void fuse_kill_sb_blk(struct super_block *sb)
1119 {
1120 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1121 
1122 	if (fc) {
1123 		down_write(&fc->killsb);
1124 		fc->sb = NULL;
1125 		up_write(&fc->killsb);
1126 	}
1127 
1128 	kill_block_super(sb);
1129 }
1130 
1131 static struct file_system_type fuseblk_fs_type = {
1132 	.owner		= THIS_MODULE,
1133 	.name		= "fuseblk",
1134 	.mount		= fuse_mount_blk,
1135 	.kill_sb	= fuse_kill_sb_blk,
1136 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1137 };
1138 
1139 static inline int register_fuseblk(void)
1140 {
1141 	return register_filesystem(&fuseblk_fs_type);
1142 }
1143 
1144 static inline void unregister_fuseblk(void)
1145 {
1146 	unregister_filesystem(&fuseblk_fs_type);
1147 }
1148 #else
1149 static inline int register_fuseblk(void)
1150 {
1151 	return 0;
1152 }
1153 
1154 static inline void unregister_fuseblk(void)
1155 {
1156 }
1157 #endif
1158 
1159 static void fuse_inode_init_once(void *foo)
1160 {
1161 	struct inode *inode = foo;
1162 
1163 	inode_init_once(inode);
1164 }
1165 
1166 static int __init fuse_fs_init(void)
1167 {
1168 	int err;
1169 
1170 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
1171 					      sizeof(struct fuse_inode),
1172 					      0, SLAB_HWCACHE_ALIGN,
1173 					      fuse_inode_init_once);
1174 	err = -ENOMEM;
1175 	if (!fuse_inode_cachep)
1176 		goto out;
1177 
1178 	err = register_fuseblk();
1179 	if (err)
1180 		goto out2;
1181 
1182 	err = register_filesystem(&fuse_fs_type);
1183 	if (err)
1184 		goto out3;
1185 
1186 	return 0;
1187 
1188  out3:
1189 	unregister_fuseblk();
1190  out2:
1191 	kmem_cache_destroy(fuse_inode_cachep);
1192  out:
1193 	return err;
1194 }
1195 
1196 static void fuse_fs_cleanup(void)
1197 {
1198 	unregister_filesystem(&fuse_fs_type);
1199 	unregister_fuseblk();
1200 
1201 	/*
1202 	 * Make sure all delayed rcu free inodes are flushed before we
1203 	 * destroy cache.
1204 	 */
1205 	rcu_barrier();
1206 	kmem_cache_destroy(fuse_inode_cachep);
1207 }
1208 
1209 static struct kobject *fuse_kobj;
1210 static struct kobject *connections_kobj;
1211 
1212 static int fuse_sysfs_init(void)
1213 {
1214 	int err;
1215 
1216 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1217 	if (!fuse_kobj) {
1218 		err = -ENOMEM;
1219 		goto out_err;
1220 	}
1221 
1222 	connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1223 	if (!connections_kobj) {
1224 		err = -ENOMEM;
1225 		goto out_fuse_unregister;
1226 	}
1227 
1228 	return 0;
1229 
1230  out_fuse_unregister:
1231 	kobject_put(fuse_kobj);
1232  out_err:
1233 	return err;
1234 }
1235 
1236 static void fuse_sysfs_cleanup(void)
1237 {
1238 	kobject_put(connections_kobj);
1239 	kobject_put(fuse_kobj);
1240 }
1241 
1242 static int __init fuse_init(void)
1243 {
1244 	int res;
1245 
1246 	printk(KERN_INFO "fuse init (API version %i.%i)\n",
1247 	       FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1248 
1249 	INIT_LIST_HEAD(&fuse_conn_list);
1250 	res = fuse_fs_init();
1251 	if (res)
1252 		goto err;
1253 
1254 	res = fuse_dev_init();
1255 	if (res)
1256 		goto err_fs_cleanup;
1257 
1258 	res = fuse_sysfs_init();
1259 	if (res)
1260 		goto err_dev_cleanup;
1261 
1262 	res = fuse_ctl_init();
1263 	if (res)
1264 		goto err_sysfs_cleanup;
1265 
1266 	sanitize_global_limit(&max_user_bgreq);
1267 	sanitize_global_limit(&max_user_congthresh);
1268 
1269 	return 0;
1270 
1271  err_sysfs_cleanup:
1272 	fuse_sysfs_cleanup();
1273  err_dev_cleanup:
1274 	fuse_dev_cleanup();
1275  err_fs_cleanup:
1276 	fuse_fs_cleanup();
1277  err:
1278 	return res;
1279 }
1280 
1281 static void __exit fuse_exit(void)
1282 {
1283 	printk(KERN_DEBUG "fuse exit\n");
1284 
1285 	fuse_ctl_cleanup();
1286 	fuse_sysfs_cleanup();
1287 	fuse_fs_cleanup();
1288 	fuse_dev_cleanup();
1289 }
1290 
1291 module_init(fuse_init);
1292 module_exit(fuse_exit);
1293