xref: /linux/fs/fuse/inode.c (revision a0b54e256d513ed99e456bea6e4e188ff92e7c46)
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/parser.h>
18 #include <linux/statfs.h>
19 #include <linux/random.h>
20 #include <linux/sched.h>
21 #include <linux/exportfs.h>
22 
23 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
24 MODULE_DESCRIPTION("Filesystem in Userspace");
25 MODULE_LICENSE("GPL");
26 
27 static struct kmem_cache *fuse_inode_cachep;
28 struct list_head fuse_conn_list;
29 DEFINE_MUTEX(fuse_mutex);
30 
31 #define FUSE_SUPER_MAGIC 0x65735546
32 
33 #define FUSE_DEFAULT_BLKSIZE 512
34 
35 struct fuse_mount_data {
36 	int fd;
37 	unsigned rootmode;
38 	unsigned user_id;
39 	unsigned group_id;
40 	unsigned fd_present:1;
41 	unsigned rootmode_present:1;
42 	unsigned user_id_present:1;
43 	unsigned group_id_present:1;
44 	unsigned flags;
45 	unsigned max_read;
46 	unsigned blksize;
47 };
48 
49 static struct inode *fuse_alloc_inode(struct super_block *sb)
50 {
51 	struct inode *inode;
52 	struct fuse_inode *fi;
53 
54 	inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
55 	if (!inode)
56 		return NULL;
57 
58 	fi = get_fuse_inode(inode);
59 	fi->i_time = 0;
60 	fi->nodeid = 0;
61 	fi->nlookup = 0;
62 	fi->attr_version = 0;
63 	fi->writectr = 0;
64 	INIT_LIST_HEAD(&fi->write_files);
65 	INIT_LIST_HEAD(&fi->queued_writes);
66 	INIT_LIST_HEAD(&fi->writepages);
67 	init_waitqueue_head(&fi->page_waitq);
68 	fi->forget_req = fuse_request_alloc();
69 	if (!fi->forget_req) {
70 		kmem_cache_free(fuse_inode_cachep, inode);
71 		return NULL;
72 	}
73 
74 	return inode;
75 }
76 
77 static void fuse_destroy_inode(struct inode *inode)
78 {
79 	struct fuse_inode *fi = get_fuse_inode(inode);
80 	BUG_ON(!list_empty(&fi->write_files));
81 	BUG_ON(!list_empty(&fi->queued_writes));
82 	if (fi->forget_req)
83 		fuse_request_free(fi->forget_req);
84 	kmem_cache_free(fuse_inode_cachep, inode);
85 }
86 
87 void fuse_send_forget(struct fuse_conn *fc, struct fuse_req *req,
88 		      u64 nodeid, u64 nlookup)
89 {
90 	struct fuse_forget_in *inarg = &req->misc.forget_in;
91 	inarg->nlookup = nlookup;
92 	req->in.h.opcode = FUSE_FORGET;
93 	req->in.h.nodeid = nodeid;
94 	req->in.numargs = 1;
95 	req->in.args[0].size = sizeof(struct fuse_forget_in);
96 	req->in.args[0].value = inarg;
97 	fuse_request_send_noreply(fc, req);
98 }
99 
100 static void fuse_clear_inode(struct inode *inode)
101 {
102 	if (inode->i_sb->s_flags & MS_ACTIVE) {
103 		struct fuse_conn *fc = get_fuse_conn(inode);
104 		struct fuse_inode *fi = get_fuse_inode(inode);
105 		fuse_send_forget(fc, fi->forget_req, fi->nodeid, fi->nlookup);
106 		fi->forget_req = NULL;
107 	}
108 }
109 
110 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
111 {
112 	if (*flags & MS_MANDLOCK)
113 		return -EINVAL;
114 
115 	return 0;
116 }
117 
118 void fuse_truncate(struct address_space *mapping, loff_t offset)
119 {
120 	/* See vmtruncate() */
121 	unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
122 	truncate_inode_pages(mapping, offset);
123 	unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
124 }
125 
126 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
127 				   u64 attr_valid)
128 {
129 	struct fuse_conn *fc = get_fuse_conn(inode);
130 	struct fuse_inode *fi = get_fuse_inode(inode);
131 
132 	fi->attr_version = ++fc->attr_version;
133 	fi->i_time = attr_valid;
134 
135 	inode->i_ino     = attr->ino;
136 	inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
137 	inode->i_nlink   = attr->nlink;
138 	inode->i_uid     = attr->uid;
139 	inode->i_gid     = attr->gid;
140 	inode->i_blocks  = attr->blocks;
141 	inode->i_atime.tv_sec   = attr->atime;
142 	inode->i_atime.tv_nsec  = attr->atimensec;
143 	inode->i_mtime.tv_sec   = attr->mtime;
144 	inode->i_mtime.tv_nsec  = attr->mtimensec;
145 	inode->i_ctime.tv_sec   = attr->ctime;
146 	inode->i_ctime.tv_nsec  = attr->ctimensec;
147 
148 	if (attr->blksize != 0)
149 		inode->i_blkbits = ilog2(attr->blksize);
150 	else
151 		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
152 
153 	/*
154 	 * Don't set the sticky bit in i_mode, unless we want the VFS
155 	 * to check permissions.  This prevents failures due to the
156 	 * check in may_delete().
157 	 */
158 	fi->orig_i_mode = inode->i_mode;
159 	if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
160 		inode->i_mode &= ~S_ISVTX;
161 }
162 
163 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
164 			    u64 attr_valid, u64 attr_version)
165 {
166 	struct fuse_conn *fc = get_fuse_conn(inode);
167 	struct fuse_inode *fi = get_fuse_inode(inode);
168 	loff_t oldsize;
169 
170 	spin_lock(&fc->lock);
171 	if (attr_version != 0 && fi->attr_version > attr_version) {
172 		spin_unlock(&fc->lock);
173 		return;
174 	}
175 
176 	fuse_change_attributes_common(inode, attr, attr_valid);
177 
178 	oldsize = inode->i_size;
179 	i_size_write(inode, attr->size);
180 	spin_unlock(&fc->lock);
181 
182 	if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
183 		if (attr->size < oldsize)
184 			fuse_truncate(inode->i_mapping, attr->size);
185 		invalidate_inode_pages2(inode->i_mapping);
186 	}
187 }
188 
189 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
190 {
191 	inode->i_mode = attr->mode & S_IFMT;
192 	inode->i_size = attr->size;
193 	if (S_ISREG(inode->i_mode)) {
194 		fuse_init_common(inode);
195 		fuse_init_file_inode(inode);
196 	} else if (S_ISDIR(inode->i_mode))
197 		fuse_init_dir(inode);
198 	else if (S_ISLNK(inode->i_mode))
199 		fuse_init_symlink(inode);
200 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
201 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
202 		fuse_init_common(inode);
203 		init_special_inode(inode, inode->i_mode,
204 				   new_decode_dev(attr->rdev));
205 	} else
206 		BUG();
207 }
208 
209 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
210 {
211 	u64 nodeid = *(u64 *) _nodeidp;
212 	if (get_node_id(inode) == nodeid)
213 		return 1;
214 	else
215 		return 0;
216 }
217 
218 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
219 {
220 	u64 nodeid = *(u64 *) _nodeidp;
221 	get_fuse_inode(inode)->nodeid = nodeid;
222 	return 0;
223 }
224 
225 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
226 			int generation, struct fuse_attr *attr,
227 			u64 attr_valid, u64 attr_version)
228 {
229 	struct inode *inode;
230 	struct fuse_inode *fi;
231 	struct fuse_conn *fc = get_fuse_conn_super(sb);
232 
233  retry:
234 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
235 	if (!inode)
236 		return NULL;
237 
238 	if ((inode->i_state & I_NEW)) {
239 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
240 		inode->i_generation = generation;
241 		inode->i_data.backing_dev_info = &fc->bdi;
242 		fuse_init_inode(inode, attr);
243 		unlock_new_inode(inode);
244 	} else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
245 		/* Inode has changed type, any I/O on the old should fail */
246 		make_bad_inode(inode);
247 		iput(inode);
248 		goto retry;
249 	}
250 
251 	fi = get_fuse_inode(inode);
252 	spin_lock(&fc->lock);
253 	fi->nlookup++;
254 	spin_unlock(&fc->lock);
255 	fuse_change_attributes(inode, attr, attr_valid, attr_version);
256 
257 	return inode;
258 }
259 
260 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
261 			     loff_t offset, loff_t len)
262 {
263 	struct inode *inode;
264 	pgoff_t pg_start;
265 	pgoff_t pg_end;
266 
267 	inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
268 	if (!inode)
269 		return -ENOENT;
270 
271 	fuse_invalidate_attr(inode);
272 	if (offset >= 0) {
273 		pg_start = offset >> PAGE_CACHE_SHIFT;
274 		if (len <= 0)
275 			pg_end = -1;
276 		else
277 			pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
278 		invalidate_inode_pages2_range(inode->i_mapping,
279 					      pg_start, pg_end);
280 	}
281 	iput(inode);
282 	return 0;
283 }
284 
285 static void fuse_umount_begin(struct super_block *sb)
286 {
287 	fuse_abort_conn(get_fuse_conn_super(sb));
288 }
289 
290 static void fuse_send_destroy(struct fuse_conn *fc)
291 {
292 	struct fuse_req *req = fc->destroy_req;
293 	if (req && fc->conn_init) {
294 		fc->destroy_req = NULL;
295 		req->in.h.opcode = FUSE_DESTROY;
296 		req->force = 1;
297 		fuse_request_send(fc, req);
298 		fuse_put_request(fc, req);
299 	}
300 }
301 
302 static void fuse_bdi_destroy(struct fuse_conn *fc)
303 {
304 	if (fc->bdi_initialized)
305 		bdi_destroy(&fc->bdi);
306 }
307 
308 void fuse_conn_kill(struct fuse_conn *fc)
309 {
310 	spin_lock(&fc->lock);
311 	fc->connected = 0;
312 	fc->blocked = 0;
313 	spin_unlock(&fc->lock);
314 	/* Flush all readers on this fs */
315 	kill_fasync(&fc->fasync, SIGIO, POLL_IN);
316 	wake_up_all(&fc->waitq);
317 	wake_up_all(&fc->blocked_waitq);
318 	wake_up_all(&fc->reserved_req_waitq);
319 	mutex_lock(&fuse_mutex);
320 	list_del(&fc->entry);
321 	fuse_ctl_remove_conn(fc);
322 	mutex_unlock(&fuse_mutex);
323 	fuse_bdi_destroy(fc);
324 }
325 EXPORT_SYMBOL_GPL(fuse_conn_kill);
326 
327 static void fuse_put_super(struct super_block *sb)
328 {
329 	struct fuse_conn *fc = get_fuse_conn_super(sb);
330 
331 	fuse_send_destroy(fc);
332 	fuse_conn_kill(fc);
333 	fuse_conn_put(fc);
334 }
335 
336 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
337 {
338 	stbuf->f_type    = FUSE_SUPER_MAGIC;
339 	stbuf->f_bsize   = attr->bsize;
340 	stbuf->f_frsize  = attr->frsize;
341 	stbuf->f_blocks  = attr->blocks;
342 	stbuf->f_bfree   = attr->bfree;
343 	stbuf->f_bavail  = attr->bavail;
344 	stbuf->f_files   = attr->files;
345 	stbuf->f_ffree   = attr->ffree;
346 	stbuf->f_namelen = attr->namelen;
347 	/* fsid is left zero */
348 }
349 
350 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
351 {
352 	struct super_block *sb = dentry->d_sb;
353 	struct fuse_conn *fc = get_fuse_conn_super(sb);
354 	struct fuse_req *req;
355 	struct fuse_statfs_out outarg;
356 	int err;
357 
358 	if (!fuse_allow_task(fc, current)) {
359 		buf->f_type = FUSE_SUPER_MAGIC;
360 		return 0;
361 	}
362 
363 	req = fuse_get_req(fc);
364 	if (IS_ERR(req))
365 		return PTR_ERR(req);
366 
367 	memset(&outarg, 0, sizeof(outarg));
368 	req->in.numargs = 0;
369 	req->in.h.opcode = FUSE_STATFS;
370 	req->in.h.nodeid = get_node_id(dentry->d_inode);
371 	req->out.numargs = 1;
372 	req->out.args[0].size =
373 		fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
374 	req->out.args[0].value = &outarg;
375 	fuse_request_send(fc, req);
376 	err = req->out.h.error;
377 	if (!err)
378 		convert_fuse_statfs(buf, &outarg.st);
379 	fuse_put_request(fc, req);
380 	return err;
381 }
382 
383 enum {
384 	OPT_FD,
385 	OPT_ROOTMODE,
386 	OPT_USER_ID,
387 	OPT_GROUP_ID,
388 	OPT_DEFAULT_PERMISSIONS,
389 	OPT_ALLOW_OTHER,
390 	OPT_MAX_READ,
391 	OPT_BLKSIZE,
392 	OPT_ERR
393 };
394 
395 static const match_table_t tokens = {
396 	{OPT_FD,			"fd=%u"},
397 	{OPT_ROOTMODE,			"rootmode=%o"},
398 	{OPT_USER_ID,			"user_id=%u"},
399 	{OPT_GROUP_ID,			"group_id=%u"},
400 	{OPT_DEFAULT_PERMISSIONS,	"default_permissions"},
401 	{OPT_ALLOW_OTHER,		"allow_other"},
402 	{OPT_MAX_READ,			"max_read=%u"},
403 	{OPT_BLKSIZE,			"blksize=%u"},
404 	{OPT_ERR,			NULL}
405 };
406 
407 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
408 {
409 	char *p;
410 	memset(d, 0, sizeof(struct fuse_mount_data));
411 	d->max_read = ~0;
412 	d->blksize = FUSE_DEFAULT_BLKSIZE;
413 
414 	while ((p = strsep(&opt, ",")) != NULL) {
415 		int token;
416 		int value;
417 		substring_t args[MAX_OPT_ARGS];
418 		if (!*p)
419 			continue;
420 
421 		token = match_token(p, tokens, args);
422 		switch (token) {
423 		case OPT_FD:
424 			if (match_int(&args[0], &value))
425 				return 0;
426 			d->fd = value;
427 			d->fd_present = 1;
428 			break;
429 
430 		case OPT_ROOTMODE:
431 			if (match_octal(&args[0], &value))
432 				return 0;
433 			if (!fuse_valid_type(value))
434 				return 0;
435 			d->rootmode = value;
436 			d->rootmode_present = 1;
437 			break;
438 
439 		case OPT_USER_ID:
440 			if (match_int(&args[0], &value))
441 				return 0;
442 			d->user_id = value;
443 			d->user_id_present = 1;
444 			break;
445 
446 		case OPT_GROUP_ID:
447 			if (match_int(&args[0], &value))
448 				return 0;
449 			d->group_id = value;
450 			d->group_id_present = 1;
451 			break;
452 
453 		case OPT_DEFAULT_PERMISSIONS:
454 			d->flags |= FUSE_DEFAULT_PERMISSIONS;
455 			break;
456 
457 		case OPT_ALLOW_OTHER:
458 			d->flags |= FUSE_ALLOW_OTHER;
459 			break;
460 
461 		case OPT_MAX_READ:
462 			if (match_int(&args[0], &value))
463 				return 0;
464 			d->max_read = value;
465 			break;
466 
467 		case OPT_BLKSIZE:
468 			if (!is_bdev || match_int(&args[0], &value))
469 				return 0;
470 			d->blksize = value;
471 			break;
472 
473 		default:
474 			return 0;
475 		}
476 	}
477 
478 	if (!d->fd_present || !d->rootmode_present ||
479 	    !d->user_id_present || !d->group_id_present)
480 		return 0;
481 
482 	return 1;
483 }
484 
485 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
486 {
487 	struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
488 
489 	seq_printf(m, ",user_id=%u", fc->user_id);
490 	seq_printf(m, ",group_id=%u", fc->group_id);
491 	if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
492 		seq_puts(m, ",default_permissions");
493 	if (fc->flags & FUSE_ALLOW_OTHER)
494 		seq_puts(m, ",allow_other");
495 	if (fc->max_read != ~0)
496 		seq_printf(m, ",max_read=%u", fc->max_read);
497 	if (mnt->mnt_sb->s_bdev &&
498 	    mnt->mnt_sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
499 		seq_printf(m, ",blksize=%lu", mnt->mnt_sb->s_blocksize);
500 	return 0;
501 }
502 
503 void fuse_conn_init(struct fuse_conn *fc)
504 {
505 	memset(fc, 0, sizeof(*fc));
506 	spin_lock_init(&fc->lock);
507 	mutex_init(&fc->inst_mutex);
508 	init_rwsem(&fc->killsb);
509 	atomic_set(&fc->count, 1);
510 	init_waitqueue_head(&fc->waitq);
511 	init_waitqueue_head(&fc->blocked_waitq);
512 	init_waitqueue_head(&fc->reserved_req_waitq);
513 	INIT_LIST_HEAD(&fc->pending);
514 	INIT_LIST_HEAD(&fc->processing);
515 	INIT_LIST_HEAD(&fc->io);
516 	INIT_LIST_HEAD(&fc->interrupts);
517 	INIT_LIST_HEAD(&fc->bg_queue);
518 	INIT_LIST_HEAD(&fc->entry);
519 	atomic_set(&fc->num_waiting, 0);
520 	fc->khctr = 0;
521 	fc->polled_files = RB_ROOT;
522 	fc->reqctr = 0;
523 	fc->blocked = 1;
524 	fc->attr_version = 1;
525 	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
526 }
527 EXPORT_SYMBOL_GPL(fuse_conn_init);
528 
529 void fuse_conn_put(struct fuse_conn *fc)
530 {
531 	if (atomic_dec_and_test(&fc->count)) {
532 		if (fc->destroy_req)
533 			fuse_request_free(fc->destroy_req);
534 		mutex_destroy(&fc->inst_mutex);
535 		fc->release(fc);
536 	}
537 }
538 EXPORT_SYMBOL_GPL(fuse_conn_put);
539 
540 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
541 {
542 	atomic_inc(&fc->count);
543 	return fc;
544 }
545 EXPORT_SYMBOL_GPL(fuse_conn_get);
546 
547 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
548 {
549 	struct fuse_attr attr;
550 	memset(&attr, 0, sizeof(attr));
551 
552 	attr.mode = mode;
553 	attr.ino = FUSE_ROOT_ID;
554 	attr.nlink = 1;
555 	return fuse_iget(sb, 1, 0, &attr, 0, 0);
556 }
557 
558 struct fuse_inode_handle {
559 	u64 nodeid;
560 	u32 generation;
561 };
562 
563 static struct dentry *fuse_get_dentry(struct super_block *sb,
564 				      struct fuse_inode_handle *handle)
565 {
566 	struct fuse_conn *fc = get_fuse_conn_super(sb);
567 	struct inode *inode;
568 	struct dentry *entry;
569 	int err = -ESTALE;
570 
571 	if (handle->nodeid == 0)
572 		goto out_err;
573 
574 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
575 	if (!inode) {
576 		struct fuse_entry_out outarg;
577 		struct qstr name;
578 
579 		if (!fc->export_support)
580 			goto out_err;
581 
582 		name.len = 1;
583 		name.name = ".";
584 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
585 				       &inode);
586 		if (err && err != -ENOENT)
587 			goto out_err;
588 		if (err || !inode) {
589 			err = -ESTALE;
590 			goto out_err;
591 		}
592 		err = -EIO;
593 		if (get_node_id(inode) != handle->nodeid)
594 			goto out_iput;
595 	}
596 	err = -ESTALE;
597 	if (inode->i_generation != handle->generation)
598 		goto out_iput;
599 
600 	entry = d_obtain_alias(inode);
601 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID) {
602 		entry->d_op = &fuse_dentry_operations;
603 		fuse_invalidate_entry_cache(entry);
604 	}
605 
606 	return entry;
607 
608  out_iput:
609 	iput(inode);
610  out_err:
611 	return ERR_PTR(err);
612 }
613 
614 static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
615 			   int connectable)
616 {
617 	struct inode *inode = dentry->d_inode;
618 	bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
619 	int len = encode_parent ? 6 : 3;
620 	u64 nodeid;
621 	u32 generation;
622 
623 	if (*max_len < len)
624 		return  255;
625 
626 	nodeid = get_fuse_inode(inode)->nodeid;
627 	generation = inode->i_generation;
628 
629 	fh[0] = (u32)(nodeid >> 32);
630 	fh[1] = (u32)(nodeid & 0xffffffff);
631 	fh[2] = generation;
632 
633 	if (encode_parent) {
634 		struct inode *parent;
635 
636 		spin_lock(&dentry->d_lock);
637 		parent = dentry->d_parent->d_inode;
638 		nodeid = get_fuse_inode(parent)->nodeid;
639 		generation = parent->i_generation;
640 		spin_unlock(&dentry->d_lock);
641 
642 		fh[3] = (u32)(nodeid >> 32);
643 		fh[4] = (u32)(nodeid & 0xffffffff);
644 		fh[5] = generation;
645 	}
646 
647 	*max_len = len;
648 	return encode_parent ? 0x82 : 0x81;
649 }
650 
651 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
652 		struct fid *fid, int fh_len, int fh_type)
653 {
654 	struct fuse_inode_handle handle;
655 
656 	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
657 		return NULL;
658 
659 	handle.nodeid = (u64) fid->raw[0] << 32;
660 	handle.nodeid |= (u64) fid->raw[1];
661 	handle.generation = fid->raw[2];
662 	return fuse_get_dentry(sb, &handle);
663 }
664 
665 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
666 		struct fid *fid, int fh_len, int fh_type)
667 {
668 	struct fuse_inode_handle parent;
669 
670 	if (fh_type != 0x82 || fh_len < 6)
671 		return NULL;
672 
673 	parent.nodeid = (u64) fid->raw[3] << 32;
674 	parent.nodeid |= (u64) fid->raw[4];
675 	parent.generation = fid->raw[5];
676 	return fuse_get_dentry(sb, &parent);
677 }
678 
679 static struct dentry *fuse_get_parent(struct dentry *child)
680 {
681 	struct inode *child_inode = child->d_inode;
682 	struct fuse_conn *fc = get_fuse_conn(child_inode);
683 	struct inode *inode;
684 	struct dentry *parent;
685 	struct fuse_entry_out outarg;
686 	struct qstr name;
687 	int err;
688 
689 	if (!fc->export_support)
690 		return ERR_PTR(-ESTALE);
691 
692 	name.len = 2;
693 	name.name = "..";
694 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
695 			       &name, &outarg, &inode);
696 	if (err) {
697 		if (err == -ENOENT)
698 			return ERR_PTR(-ESTALE);
699 		return ERR_PTR(err);
700 	}
701 
702 	parent = d_obtain_alias(inode);
703 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID) {
704 		parent->d_op = &fuse_dentry_operations;
705 		fuse_invalidate_entry_cache(parent);
706 	}
707 
708 	return parent;
709 }
710 
711 static const struct export_operations fuse_export_operations = {
712 	.fh_to_dentry	= fuse_fh_to_dentry,
713 	.fh_to_parent	= fuse_fh_to_parent,
714 	.encode_fh	= fuse_encode_fh,
715 	.get_parent	= fuse_get_parent,
716 };
717 
718 static const struct super_operations fuse_super_operations = {
719 	.alloc_inode    = fuse_alloc_inode,
720 	.destroy_inode  = fuse_destroy_inode,
721 	.clear_inode	= fuse_clear_inode,
722 	.drop_inode	= generic_delete_inode,
723 	.remount_fs	= fuse_remount_fs,
724 	.put_super	= fuse_put_super,
725 	.umount_begin	= fuse_umount_begin,
726 	.statfs		= fuse_statfs,
727 	.show_options	= fuse_show_options,
728 };
729 
730 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
731 {
732 	struct fuse_init_out *arg = &req->misc.init_out;
733 
734 	if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
735 		fc->conn_error = 1;
736 	else {
737 		unsigned long ra_pages;
738 
739 		if (arg->minor >= 6) {
740 			ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
741 			if (arg->flags & FUSE_ASYNC_READ)
742 				fc->async_read = 1;
743 			if (!(arg->flags & FUSE_POSIX_LOCKS))
744 				fc->no_lock = 1;
745 			if (arg->flags & FUSE_ATOMIC_O_TRUNC)
746 				fc->atomic_o_trunc = 1;
747 			if (arg->minor >= 9) {
748 				/* LOOKUP has dependency on proto version */
749 				if (arg->flags & FUSE_EXPORT_SUPPORT)
750 					fc->export_support = 1;
751 			}
752 			if (arg->flags & FUSE_BIG_WRITES)
753 				fc->big_writes = 1;
754 			if (arg->flags & FUSE_DONT_MASK)
755 				fc->dont_mask = 1;
756 		} else {
757 			ra_pages = fc->max_read / PAGE_CACHE_SIZE;
758 			fc->no_lock = 1;
759 		}
760 
761 		fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
762 		fc->minor = arg->minor;
763 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
764 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
765 		fc->conn_init = 1;
766 	}
767 	fc->blocked = 0;
768 	wake_up_all(&fc->blocked_waitq);
769 }
770 
771 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
772 {
773 	struct fuse_init_in *arg = &req->misc.init_in;
774 
775 	arg->major = FUSE_KERNEL_VERSION;
776 	arg->minor = FUSE_KERNEL_MINOR_VERSION;
777 	arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
778 	arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
779 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK;
780 	req->in.h.opcode = FUSE_INIT;
781 	req->in.numargs = 1;
782 	req->in.args[0].size = sizeof(*arg);
783 	req->in.args[0].value = arg;
784 	req->out.numargs = 1;
785 	/* Variable length arguement used for backward compatibility
786 	   with interface version < 7.5.  Rest of init_out is zeroed
787 	   by do_get_request(), so a short reply is not a problem */
788 	req->out.argvar = 1;
789 	req->out.args[0].size = sizeof(struct fuse_init_out);
790 	req->out.args[0].value = &req->misc.init_out;
791 	req->end = process_init_reply;
792 	fuse_request_send_background(fc, req);
793 }
794 
795 static void fuse_free_conn(struct fuse_conn *fc)
796 {
797 	kfree(fc);
798 }
799 
800 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
801 {
802 	int err;
803 
804 	fc->bdi.name = "fuse";
805 	fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
806 	fc->bdi.unplug_io_fn = default_unplug_io_fn;
807 	/* fuse does it's own writeback accounting */
808 	fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
809 
810 	err = bdi_init(&fc->bdi);
811 	if (err)
812 		return err;
813 
814 	fc->bdi_initialized = 1;
815 
816 	if (sb->s_bdev) {
817 		err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
818 				    MAJOR(fc->dev), MINOR(fc->dev));
819 	} else {
820 		err = bdi_register_dev(&fc->bdi, fc->dev);
821 	}
822 
823 	if (err)
824 		return err;
825 
826 	/*
827 	 * For a single fuse filesystem use max 1% of dirty +
828 	 * writeback threshold.
829 	 *
830 	 * This gives about 1M of write buffer for memory maps on a
831 	 * machine with 1G and 10% dirty_ratio, which should be more
832 	 * than enough.
833 	 *
834 	 * Privileged users can raise it by writing to
835 	 *
836 	 *    /sys/class/bdi/<bdi>/max_ratio
837 	 */
838 	bdi_set_max_ratio(&fc->bdi, 1);
839 
840 	return 0;
841 }
842 
843 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
844 {
845 	struct fuse_conn *fc;
846 	struct inode *root;
847 	struct fuse_mount_data d;
848 	struct file *file;
849 	struct dentry *root_dentry;
850 	struct fuse_req *init_req;
851 	int err;
852 	int is_bdev = sb->s_bdev != NULL;
853 
854 	err = -EINVAL;
855 	if (sb->s_flags & MS_MANDLOCK)
856 		goto err;
857 
858 	if (!parse_fuse_opt((char *) data, &d, is_bdev))
859 		goto err;
860 
861 	if (is_bdev) {
862 #ifdef CONFIG_BLOCK
863 		err = -EINVAL;
864 		if (!sb_set_blocksize(sb, d.blksize))
865 			goto err;
866 #endif
867 	} else {
868 		sb->s_blocksize = PAGE_CACHE_SIZE;
869 		sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
870 	}
871 	sb->s_magic = FUSE_SUPER_MAGIC;
872 	sb->s_op = &fuse_super_operations;
873 	sb->s_maxbytes = MAX_LFS_FILESIZE;
874 	sb->s_export_op = &fuse_export_operations;
875 
876 	file = fget(d.fd);
877 	err = -EINVAL;
878 	if (!file)
879 		goto err;
880 
881 	if (file->f_op != &fuse_dev_operations)
882 		goto err_fput;
883 
884 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
885 	err = -ENOMEM;
886 	if (!fc)
887 		goto err_fput;
888 
889 	fuse_conn_init(fc);
890 
891 	fc->dev = sb->s_dev;
892 	fc->sb = sb;
893 	err = fuse_bdi_init(fc, sb);
894 	if (err)
895 		goto err_put_conn;
896 
897 	sb->s_bdi = &fc->bdi;
898 
899 	/* Handle umasking inside the fuse code */
900 	if (sb->s_flags & MS_POSIXACL)
901 		fc->dont_mask = 1;
902 	sb->s_flags |= MS_POSIXACL;
903 
904 	fc->release = fuse_free_conn;
905 	fc->flags = d.flags;
906 	fc->user_id = d.user_id;
907 	fc->group_id = d.group_id;
908 	fc->max_read = max_t(unsigned, 4096, d.max_read);
909 
910 	/* Used by get_root_inode() */
911 	sb->s_fs_info = fc;
912 
913 	err = -ENOMEM;
914 	root = fuse_get_root_inode(sb, d.rootmode);
915 	if (!root)
916 		goto err_put_conn;
917 
918 	root_dentry = d_alloc_root(root);
919 	if (!root_dentry) {
920 		iput(root);
921 		goto err_put_conn;
922 	}
923 
924 	init_req = fuse_request_alloc();
925 	if (!init_req)
926 		goto err_put_root;
927 
928 	if (is_bdev) {
929 		fc->destroy_req = fuse_request_alloc();
930 		if (!fc->destroy_req)
931 			goto err_free_init_req;
932 	}
933 
934 	mutex_lock(&fuse_mutex);
935 	err = -EINVAL;
936 	if (file->private_data)
937 		goto err_unlock;
938 
939 	err = fuse_ctl_add_conn(fc);
940 	if (err)
941 		goto err_unlock;
942 
943 	list_add_tail(&fc->entry, &fuse_conn_list);
944 	sb->s_root = root_dentry;
945 	fc->connected = 1;
946 	file->private_data = fuse_conn_get(fc);
947 	mutex_unlock(&fuse_mutex);
948 	/*
949 	 * atomic_dec_and_test() in fput() provides the necessary
950 	 * memory barrier for file->private_data to be visible on all
951 	 * CPUs after this
952 	 */
953 	fput(file);
954 
955 	fuse_send_init(fc, init_req);
956 
957 	return 0;
958 
959  err_unlock:
960 	mutex_unlock(&fuse_mutex);
961  err_free_init_req:
962 	fuse_request_free(init_req);
963  err_put_root:
964 	dput(root_dentry);
965  err_put_conn:
966 	fuse_bdi_destroy(fc);
967 	fuse_conn_put(fc);
968  err_fput:
969 	fput(file);
970  err:
971 	return err;
972 }
973 
974 static int fuse_get_sb(struct file_system_type *fs_type,
975 		       int flags, const char *dev_name,
976 		       void *raw_data, struct vfsmount *mnt)
977 {
978 	return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
979 }
980 
981 static void fuse_kill_sb_anon(struct super_block *sb)
982 {
983 	struct fuse_conn *fc = get_fuse_conn_super(sb);
984 
985 	if (fc) {
986 		down_write(&fc->killsb);
987 		fc->sb = NULL;
988 		up_write(&fc->killsb);
989 	}
990 
991 	kill_anon_super(sb);
992 }
993 
994 static struct file_system_type fuse_fs_type = {
995 	.owner		= THIS_MODULE,
996 	.name		= "fuse",
997 	.fs_flags	= FS_HAS_SUBTYPE,
998 	.get_sb		= fuse_get_sb,
999 	.kill_sb	= fuse_kill_sb_anon,
1000 };
1001 
1002 #ifdef CONFIG_BLOCK
1003 static int fuse_get_sb_blk(struct file_system_type *fs_type,
1004 			   int flags, const char *dev_name,
1005 			   void *raw_data, struct vfsmount *mnt)
1006 {
1007 	return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
1008 			   mnt);
1009 }
1010 
1011 static void fuse_kill_sb_blk(struct super_block *sb)
1012 {
1013 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1014 
1015 	if (fc) {
1016 		down_write(&fc->killsb);
1017 		fc->sb = NULL;
1018 		up_write(&fc->killsb);
1019 	}
1020 
1021 	kill_block_super(sb);
1022 }
1023 
1024 static struct file_system_type fuseblk_fs_type = {
1025 	.owner		= THIS_MODULE,
1026 	.name		= "fuseblk",
1027 	.get_sb		= fuse_get_sb_blk,
1028 	.kill_sb	= fuse_kill_sb_blk,
1029 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1030 };
1031 
1032 static inline int register_fuseblk(void)
1033 {
1034 	return register_filesystem(&fuseblk_fs_type);
1035 }
1036 
1037 static inline void unregister_fuseblk(void)
1038 {
1039 	unregister_filesystem(&fuseblk_fs_type);
1040 }
1041 #else
1042 static inline int register_fuseblk(void)
1043 {
1044 	return 0;
1045 }
1046 
1047 static inline void unregister_fuseblk(void)
1048 {
1049 }
1050 #endif
1051 
1052 static void fuse_inode_init_once(void *foo)
1053 {
1054 	struct inode *inode = foo;
1055 
1056 	inode_init_once(inode);
1057 }
1058 
1059 static int __init fuse_fs_init(void)
1060 {
1061 	int err;
1062 
1063 	err = register_filesystem(&fuse_fs_type);
1064 	if (err)
1065 		goto out;
1066 
1067 	err = register_fuseblk();
1068 	if (err)
1069 		goto out_unreg;
1070 
1071 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
1072 					      sizeof(struct fuse_inode),
1073 					      0, SLAB_HWCACHE_ALIGN,
1074 					      fuse_inode_init_once);
1075 	err = -ENOMEM;
1076 	if (!fuse_inode_cachep)
1077 		goto out_unreg2;
1078 
1079 	return 0;
1080 
1081  out_unreg2:
1082 	unregister_fuseblk();
1083  out_unreg:
1084 	unregister_filesystem(&fuse_fs_type);
1085  out:
1086 	return err;
1087 }
1088 
1089 static void fuse_fs_cleanup(void)
1090 {
1091 	unregister_filesystem(&fuse_fs_type);
1092 	unregister_fuseblk();
1093 	kmem_cache_destroy(fuse_inode_cachep);
1094 }
1095 
1096 static struct kobject *fuse_kobj;
1097 static struct kobject *connections_kobj;
1098 
1099 static int fuse_sysfs_init(void)
1100 {
1101 	int err;
1102 
1103 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1104 	if (!fuse_kobj) {
1105 		err = -ENOMEM;
1106 		goto out_err;
1107 	}
1108 
1109 	connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1110 	if (!connections_kobj) {
1111 		err = -ENOMEM;
1112 		goto out_fuse_unregister;
1113 	}
1114 
1115 	return 0;
1116 
1117  out_fuse_unregister:
1118 	kobject_put(fuse_kobj);
1119  out_err:
1120 	return err;
1121 }
1122 
1123 static void fuse_sysfs_cleanup(void)
1124 {
1125 	kobject_put(connections_kobj);
1126 	kobject_put(fuse_kobj);
1127 }
1128 
1129 static int __init fuse_init(void)
1130 {
1131 	int res;
1132 
1133 	printk(KERN_INFO "fuse init (API version %i.%i)\n",
1134 	       FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1135 
1136 	INIT_LIST_HEAD(&fuse_conn_list);
1137 	res = fuse_fs_init();
1138 	if (res)
1139 		goto err;
1140 
1141 	res = fuse_dev_init();
1142 	if (res)
1143 		goto err_fs_cleanup;
1144 
1145 	res = fuse_sysfs_init();
1146 	if (res)
1147 		goto err_dev_cleanup;
1148 
1149 	res = fuse_ctl_init();
1150 	if (res)
1151 		goto err_sysfs_cleanup;
1152 
1153 	return 0;
1154 
1155  err_sysfs_cleanup:
1156 	fuse_sysfs_cleanup();
1157  err_dev_cleanup:
1158 	fuse_dev_cleanup();
1159  err_fs_cleanup:
1160 	fuse_fs_cleanup();
1161  err:
1162 	return res;
1163 }
1164 
1165 static void __exit fuse_exit(void)
1166 {
1167 	printk(KERN_DEBUG "fuse exit\n");
1168 
1169 	fuse_ctl_cleanup();
1170 	fuse_sysfs_cleanup();
1171 	fuse_fs_cleanup();
1172 	fuse_dev_cleanup();
1173 }
1174 
1175 module_init(fuse_init);
1176 module_exit(fuse_exit);
1177