xref: /openbmc/linux/fs/fuse/inode.c (revision 77d84ff8)
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 	    test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
206 		spin_unlock(&fc->lock);
207 		return;
208 	}
209 
210 	old_mtime = inode->i_mtime;
211 	fuse_change_attributes_common(inode, attr, attr_valid);
212 
213 	oldsize = inode->i_size;
214 	i_size_write(inode, attr->size);
215 	spin_unlock(&fc->lock);
216 
217 	if (S_ISREG(inode->i_mode)) {
218 		bool inval = false;
219 
220 		if (oldsize != attr->size) {
221 			truncate_pagecache(inode, attr->size);
222 			inval = true;
223 		} else if (fc->auto_inval_data) {
224 			struct timespec new_mtime = {
225 				.tv_sec = attr->mtime,
226 				.tv_nsec = attr->mtimensec,
227 			};
228 
229 			/*
230 			 * Auto inval mode also checks and invalidates if mtime
231 			 * has changed.
232 			 */
233 			if (!timespec_equal(&old_mtime, &new_mtime))
234 				inval = true;
235 		}
236 
237 		if (inval)
238 			invalidate_inode_pages2(inode->i_mapping);
239 	}
240 }
241 
242 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
243 {
244 	inode->i_mode = attr->mode & S_IFMT;
245 	inode->i_size = attr->size;
246 	if (S_ISREG(inode->i_mode)) {
247 		fuse_init_common(inode);
248 		fuse_init_file_inode(inode);
249 	} else if (S_ISDIR(inode->i_mode))
250 		fuse_init_dir(inode);
251 	else if (S_ISLNK(inode->i_mode))
252 		fuse_init_symlink(inode);
253 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
254 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
255 		fuse_init_common(inode);
256 		init_special_inode(inode, inode->i_mode,
257 				   new_decode_dev(attr->rdev));
258 	} else
259 		BUG();
260 }
261 
262 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
263 {
264 	u64 nodeid = *(u64 *) _nodeidp;
265 	if (get_node_id(inode) == nodeid)
266 		return 1;
267 	else
268 		return 0;
269 }
270 
271 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
272 {
273 	u64 nodeid = *(u64 *) _nodeidp;
274 	get_fuse_inode(inode)->nodeid = nodeid;
275 	return 0;
276 }
277 
278 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
279 			int generation, struct fuse_attr *attr,
280 			u64 attr_valid, u64 attr_version)
281 {
282 	struct inode *inode;
283 	struct fuse_inode *fi;
284 	struct fuse_conn *fc = get_fuse_conn_super(sb);
285 
286  retry:
287 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
288 	if (!inode)
289 		return NULL;
290 
291 	if ((inode->i_state & I_NEW)) {
292 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
293 		inode->i_generation = generation;
294 		inode->i_data.backing_dev_info = &fc->bdi;
295 		fuse_init_inode(inode, attr);
296 		unlock_new_inode(inode);
297 	} else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
298 		/* Inode has changed type, any I/O on the old should fail */
299 		make_bad_inode(inode);
300 		iput(inode);
301 		goto retry;
302 	}
303 
304 	fi = get_fuse_inode(inode);
305 	spin_lock(&fc->lock);
306 	fi->nlookup++;
307 	spin_unlock(&fc->lock);
308 	fuse_change_attributes(inode, attr, attr_valid, attr_version);
309 
310 	return inode;
311 }
312 
313 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
314 			     loff_t offset, loff_t len)
315 {
316 	struct inode *inode;
317 	pgoff_t pg_start;
318 	pgoff_t pg_end;
319 
320 	inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
321 	if (!inode)
322 		return -ENOENT;
323 
324 	fuse_invalidate_attr(inode);
325 	if (offset >= 0) {
326 		pg_start = offset >> PAGE_CACHE_SHIFT;
327 		if (len <= 0)
328 			pg_end = -1;
329 		else
330 			pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
331 		invalidate_inode_pages2_range(inode->i_mapping,
332 					      pg_start, pg_end);
333 	}
334 	iput(inode);
335 	return 0;
336 }
337 
338 static void fuse_umount_begin(struct super_block *sb)
339 {
340 	fuse_abort_conn(get_fuse_conn_super(sb));
341 }
342 
343 static void fuse_send_destroy(struct fuse_conn *fc)
344 {
345 	struct fuse_req *req = fc->destroy_req;
346 	if (req && fc->conn_init) {
347 		fc->destroy_req = NULL;
348 		req->in.h.opcode = FUSE_DESTROY;
349 		req->force = 1;
350 		req->background = 0;
351 		fuse_request_send(fc, req);
352 		fuse_put_request(fc, req);
353 	}
354 }
355 
356 static void fuse_bdi_destroy(struct fuse_conn *fc)
357 {
358 	if (fc->bdi_initialized)
359 		bdi_destroy(&fc->bdi);
360 }
361 
362 void fuse_conn_kill(struct fuse_conn *fc)
363 {
364 	spin_lock(&fc->lock);
365 	fc->connected = 0;
366 	fc->blocked = 0;
367 	fc->initialized = 1;
368 	spin_unlock(&fc->lock);
369 	/* Flush all readers on this fs */
370 	kill_fasync(&fc->fasync, SIGIO, POLL_IN);
371 	wake_up_all(&fc->waitq);
372 	wake_up_all(&fc->blocked_waitq);
373 	wake_up_all(&fc->reserved_req_waitq);
374 }
375 EXPORT_SYMBOL_GPL(fuse_conn_kill);
376 
377 static void fuse_put_super(struct super_block *sb)
378 {
379 	struct fuse_conn *fc = get_fuse_conn_super(sb);
380 
381 	fuse_send_destroy(fc);
382 
383 	fuse_conn_kill(fc);
384 	mutex_lock(&fuse_mutex);
385 	list_del(&fc->entry);
386 	fuse_ctl_remove_conn(fc);
387 	mutex_unlock(&fuse_mutex);
388 	fuse_bdi_destroy(fc);
389 
390 	fuse_conn_put(fc);
391 }
392 
393 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
394 {
395 	stbuf->f_type    = FUSE_SUPER_MAGIC;
396 	stbuf->f_bsize   = attr->bsize;
397 	stbuf->f_frsize  = attr->frsize;
398 	stbuf->f_blocks  = attr->blocks;
399 	stbuf->f_bfree   = attr->bfree;
400 	stbuf->f_bavail  = attr->bavail;
401 	stbuf->f_files   = attr->files;
402 	stbuf->f_ffree   = attr->ffree;
403 	stbuf->f_namelen = attr->namelen;
404 	/* fsid is left zero */
405 }
406 
407 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
408 {
409 	struct super_block *sb = dentry->d_sb;
410 	struct fuse_conn *fc = get_fuse_conn_super(sb);
411 	struct fuse_req *req;
412 	struct fuse_statfs_out outarg;
413 	int err;
414 
415 	if (!fuse_allow_current_process(fc)) {
416 		buf->f_type = FUSE_SUPER_MAGIC;
417 		return 0;
418 	}
419 
420 	req = fuse_get_req_nopages(fc);
421 	if (IS_ERR(req))
422 		return PTR_ERR(req);
423 
424 	memset(&outarg, 0, sizeof(outarg));
425 	req->in.numargs = 0;
426 	req->in.h.opcode = FUSE_STATFS;
427 	req->in.h.nodeid = get_node_id(dentry->d_inode);
428 	req->out.numargs = 1;
429 	req->out.args[0].size =
430 		fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
431 	req->out.args[0].value = &outarg;
432 	fuse_request_send(fc, req);
433 	err = req->out.h.error;
434 	if (!err)
435 		convert_fuse_statfs(buf, &outarg.st);
436 	fuse_put_request(fc, req);
437 	return err;
438 }
439 
440 enum {
441 	OPT_FD,
442 	OPT_ROOTMODE,
443 	OPT_USER_ID,
444 	OPT_GROUP_ID,
445 	OPT_DEFAULT_PERMISSIONS,
446 	OPT_ALLOW_OTHER,
447 	OPT_MAX_READ,
448 	OPT_BLKSIZE,
449 	OPT_ERR
450 };
451 
452 static const match_table_t tokens = {
453 	{OPT_FD,			"fd=%u"},
454 	{OPT_ROOTMODE,			"rootmode=%o"},
455 	{OPT_USER_ID,			"user_id=%u"},
456 	{OPT_GROUP_ID,			"group_id=%u"},
457 	{OPT_DEFAULT_PERMISSIONS,	"default_permissions"},
458 	{OPT_ALLOW_OTHER,		"allow_other"},
459 	{OPT_MAX_READ,			"max_read=%u"},
460 	{OPT_BLKSIZE,			"blksize=%u"},
461 	{OPT_ERR,			NULL}
462 };
463 
464 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
465 {
466 	char *p;
467 	memset(d, 0, sizeof(struct fuse_mount_data));
468 	d->max_read = ~0;
469 	d->blksize = FUSE_DEFAULT_BLKSIZE;
470 
471 	while ((p = strsep(&opt, ",")) != NULL) {
472 		int token;
473 		int value;
474 		substring_t args[MAX_OPT_ARGS];
475 		if (!*p)
476 			continue;
477 
478 		token = match_token(p, tokens, args);
479 		switch (token) {
480 		case OPT_FD:
481 			if (match_int(&args[0], &value))
482 				return 0;
483 			d->fd = value;
484 			d->fd_present = 1;
485 			break;
486 
487 		case OPT_ROOTMODE:
488 			if (match_octal(&args[0], &value))
489 				return 0;
490 			if (!fuse_valid_type(value))
491 				return 0;
492 			d->rootmode = value;
493 			d->rootmode_present = 1;
494 			break;
495 
496 		case OPT_USER_ID:
497 			if (match_int(&args[0], &value))
498 				return 0;
499 			d->user_id = make_kuid(current_user_ns(), value);
500 			if (!uid_valid(d->user_id))
501 				return 0;
502 			d->user_id_present = 1;
503 			break;
504 
505 		case OPT_GROUP_ID:
506 			if (match_int(&args[0], &value))
507 				return 0;
508 			d->group_id = make_kgid(current_user_ns(), value);
509 			if (!gid_valid(d->group_id))
510 				return 0;
511 			d->group_id_present = 1;
512 			break;
513 
514 		case OPT_DEFAULT_PERMISSIONS:
515 			d->flags |= FUSE_DEFAULT_PERMISSIONS;
516 			break;
517 
518 		case OPT_ALLOW_OTHER:
519 			d->flags |= FUSE_ALLOW_OTHER;
520 			break;
521 
522 		case OPT_MAX_READ:
523 			if (match_int(&args[0], &value))
524 				return 0;
525 			d->max_read = value;
526 			break;
527 
528 		case OPT_BLKSIZE:
529 			if (!is_bdev || match_int(&args[0], &value))
530 				return 0;
531 			d->blksize = value;
532 			break;
533 
534 		default:
535 			return 0;
536 		}
537 	}
538 
539 	if (!d->fd_present || !d->rootmode_present ||
540 	    !d->user_id_present || !d->group_id_present)
541 		return 0;
542 
543 	return 1;
544 }
545 
546 static int fuse_show_options(struct seq_file *m, struct dentry *root)
547 {
548 	struct super_block *sb = root->d_sb;
549 	struct fuse_conn *fc = get_fuse_conn_super(sb);
550 
551 	seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
552 	seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
553 	if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
554 		seq_puts(m, ",default_permissions");
555 	if (fc->flags & FUSE_ALLOW_OTHER)
556 		seq_puts(m, ",allow_other");
557 	if (fc->max_read != ~0)
558 		seq_printf(m, ",max_read=%u", fc->max_read);
559 	if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
560 		seq_printf(m, ",blksize=%lu", sb->s_blocksize);
561 	return 0;
562 }
563 
564 void fuse_conn_init(struct fuse_conn *fc)
565 {
566 	memset(fc, 0, sizeof(*fc));
567 	spin_lock_init(&fc->lock);
568 	init_rwsem(&fc->killsb);
569 	atomic_set(&fc->count, 1);
570 	init_waitqueue_head(&fc->waitq);
571 	init_waitqueue_head(&fc->blocked_waitq);
572 	init_waitqueue_head(&fc->reserved_req_waitq);
573 	INIT_LIST_HEAD(&fc->pending);
574 	INIT_LIST_HEAD(&fc->processing);
575 	INIT_LIST_HEAD(&fc->io);
576 	INIT_LIST_HEAD(&fc->interrupts);
577 	INIT_LIST_HEAD(&fc->bg_queue);
578 	INIT_LIST_HEAD(&fc->entry);
579 	fc->forget_list_tail = &fc->forget_list_head;
580 	atomic_set(&fc->num_waiting, 0);
581 	fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
582 	fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
583 	fc->khctr = 0;
584 	fc->polled_files = RB_ROOT;
585 	fc->reqctr = 0;
586 	fc->blocked = 0;
587 	fc->initialized = 0;
588 	fc->attr_version = 1;
589 	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
590 }
591 EXPORT_SYMBOL_GPL(fuse_conn_init);
592 
593 void fuse_conn_put(struct fuse_conn *fc)
594 {
595 	if (atomic_dec_and_test(&fc->count)) {
596 		if (fc->destroy_req)
597 			fuse_request_free(fc->destroy_req);
598 		fc->release(fc);
599 	}
600 }
601 EXPORT_SYMBOL_GPL(fuse_conn_put);
602 
603 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
604 {
605 	atomic_inc(&fc->count);
606 	return fc;
607 }
608 EXPORT_SYMBOL_GPL(fuse_conn_get);
609 
610 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
611 {
612 	struct fuse_attr attr;
613 	memset(&attr, 0, sizeof(attr));
614 
615 	attr.mode = mode;
616 	attr.ino = FUSE_ROOT_ID;
617 	attr.nlink = 1;
618 	return fuse_iget(sb, 1, 0, &attr, 0, 0);
619 }
620 
621 struct fuse_inode_handle {
622 	u64 nodeid;
623 	u32 generation;
624 };
625 
626 static struct dentry *fuse_get_dentry(struct super_block *sb,
627 				      struct fuse_inode_handle *handle)
628 {
629 	struct fuse_conn *fc = get_fuse_conn_super(sb);
630 	struct inode *inode;
631 	struct dentry *entry;
632 	int err = -ESTALE;
633 
634 	if (handle->nodeid == 0)
635 		goto out_err;
636 
637 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
638 	if (!inode) {
639 		struct fuse_entry_out outarg;
640 		struct qstr name;
641 
642 		if (!fc->export_support)
643 			goto out_err;
644 
645 		name.len = 1;
646 		name.name = ".";
647 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
648 				       &inode);
649 		if (err && err != -ENOENT)
650 			goto out_err;
651 		if (err || !inode) {
652 			err = -ESTALE;
653 			goto out_err;
654 		}
655 		err = -EIO;
656 		if (get_node_id(inode) != handle->nodeid)
657 			goto out_iput;
658 	}
659 	err = -ESTALE;
660 	if (inode->i_generation != handle->generation)
661 		goto out_iput;
662 
663 	entry = d_obtain_alias(inode);
664 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
665 		fuse_invalidate_entry_cache(entry);
666 
667 	return entry;
668 
669  out_iput:
670 	iput(inode);
671  out_err:
672 	return ERR_PTR(err);
673 }
674 
675 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
676 			   struct inode *parent)
677 {
678 	int len = parent ? 6 : 3;
679 	u64 nodeid;
680 	u32 generation;
681 
682 	if (*max_len < len) {
683 		*max_len = len;
684 		return  FILEID_INVALID;
685 	}
686 
687 	nodeid = get_fuse_inode(inode)->nodeid;
688 	generation = inode->i_generation;
689 
690 	fh[0] = (u32)(nodeid >> 32);
691 	fh[1] = (u32)(nodeid & 0xffffffff);
692 	fh[2] = generation;
693 
694 	if (parent) {
695 		nodeid = get_fuse_inode(parent)->nodeid;
696 		generation = parent->i_generation;
697 
698 		fh[3] = (u32)(nodeid >> 32);
699 		fh[4] = (u32)(nodeid & 0xffffffff);
700 		fh[5] = generation;
701 	}
702 
703 	*max_len = len;
704 	return parent ? 0x82 : 0x81;
705 }
706 
707 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
708 		struct fid *fid, int fh_len, int fh_type)
709 {
710 	struct fuse_inode_handle handle;
711 
712 	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
713 		return NULL;
714 
715 	handle.nodeid = (u64) fid->raw[0] << 32;
716 	handle.nodeid |= (u64) fid->raw[1];
717 	handle.generation = fid->raw[2];
718 	return fuse_get_dentry(sb, &handle);
719 }
720 
721 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
722 		struct fid *fid, int fh_len, int fh_type)
723 {
724 	struct fuse_inode_handle parent;
725 
726 	if (fh_type != 0x82 || fh_len < 6)
727 		return NULL;
728 
729 	parent.nodeid = (u64) fid->raw[3] << 32;
730 	parent.nodeid |= (u64) fid->raw[4];
731 	parent.generation = fid->raw[5];
732 	return fuse_get_dentry(sb, &parent);
733 }
734 
735 static struct dentry *fuse_get_parent(struct dentry *child)
736 {
737 	struct inode *child_inode = child->d_inode;
738 	struct fuse_conn *fc = get_fuse_conn(child_inode);
739 	struct inode *inode;
740 	struct dentry *parent;
741 	struct fuse_entry_out outarg;
742 	struct qstr name;
743 	int err;
744 
745 	if (!fc->export_support)
746 		return ERR_PTR(-ESTALE);
747 
748 	name.len = 2;
749 	name.name = "..";
750 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
751 			       &name, &outarg, &inode);
752 	if (err) {
753 		if (err == -ENOENT)
754 			return ERR_PTR(-ESTALE);
755 		return ERR_PTR(err);
756 	}
757 
758 	parent = d_obtain_alias(inode);
759 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
760 		fuse_invalidate_entry_cache(parent);
761 
762 	return parent;
763 }
764 
765 static const struct export_operations fuse_export_operations = {
766 	.fh_to_dentry	= fuse_fh_to_dentry,
767 	.fh_to_parent	= fuse_fh_to_parent,
768 	.encode_fh	= fuse_encode_fh,
769 	.get_parent	= fuse_get_parent,
770 };
771 
772 static const struct super_operations fuse_super_operations = {
773 	.alloc_inode    = fuse_alloc_inode,
774 	.destroy_inode  = fuse_destroy_inode,
775 	.evict_inode	= fuse_evict_inode,
776 	.drop_inode	= generic_delete_inode,
777 	.remount_fs	= fuse_remount_fs,
778 	.put_super	= fuse_put_super,
779 	.umount_begin	= fuse_umount_begin,
780 	.statfs		= fuse_statfs,
781 	.show_options	= fuse_show_options,
782 };
783 
784 static void sanitize_global_limit(unsigned *limit)
785 {
786 	if (*limit == 0)
787 		*limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
788 			 sizeof(struct fuse_req);
789 
790 	if (*limit >= 1 << 16)
791 		*limit = (1 << 16) - 1;
792 }
793 
794 static int set_global_limit(const char *val, struct kernel_param *kp)
795 {
796 	int rv;
797 
798 	rv = param_set_uint(val, kp);
799 	if (rv)
800 		return rv;
801 
802 	sanitize_global_limit((unsigned *)kp->arg);
803 
804 	return 0;
805 }
806 
807 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
808 {
809 	int cap_sys_admin = capable(CAP_SYS_ADMIN);
810 
811 	if (arg->minor < 13)
812 		return;
813 
814 	sanitize_global_limit(&max_user_bgreq);
815 	sanitize_global_limit(&max_user_congthresh);
816 
817 	if (arg->max_background) {
818 		fc->max_background = arg->max_background;
819 
820 		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
821 			fc->max_background = max_user_bgreq;
822 	}
823 	if (arg->congestion_threshold) {
824 		fc->congestion_threshold = arg->congestion_threshold;
825 
826 		if (!cap_sys_admin &&
827 		    fc->congestion_threshold > max_user_congthresh)
828 			fc->congestion_threshold = max_user_congthresh;
829 	}
830 }
831 
832 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
833 {
834 	struct fuse_init_out *arg = &req->misc.init_out;
835 
836 	if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
837 		fc->conn_error = 1;
838 	else {
839 		unsigned long ra_pages;
840 
841 		process_init_limits(fc, arg);
842 
843 		if (arg->minor >= 6) {
844 			ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
845 			if (arg->flags & FUSE_ASYNC_READ)
846 				fc->async_read = 1;
847 			if (!(arg->flags & FUSE_POSIX_LOCKS))
848 				fc->no_lock = 1;
849 			if (arg->minor >= 17) {
850 				if (!(arg->flags & FUSE_FLOCK_LOCKS))
851 					fc->no_flock = 1;
852 			} else {
853 				if (!(arg->flags & FUSE_POSIX_LOCKS))
854 					fc->no_flock = 1;
855 			}
856 			if (arg->flags & FUSE_ATOMIC_O_TRUNC)
857 				fc->atomic_o_trunc = 1;
858 			if (arg->minor >= 9) {
859 				/* LOOKUP has dependency on proto version */
860 				if (arg->flags & FUSE_EXPORT_SUPPORT)
861 					fc->export_support = 1;
862 			}
863 			if (arg->flags & FUSE_BIG_WRITES)
864 				fc->big_writes = 1;
865 			if (arg->flags & FUSE_DONT_MASK)
866 				fc->dont_mask = 1;
867 			if (arg->flags & FUSE_AUTO_INVAL_DATA)
868 				fc->auto_inval_data = 1;
869 			if (arg->flags & FUSE_DO_READDIRPLUS) {
870 				fc->do_readdirplus = 1;
871 				if (arg->flags & FUSE_READDIRPLUS_AUTO)
872 					fc->readdirplus_auto = 1;
873 			}
874 			if (arg->flags & FUSE_ASYNC_DIO)
875 				fc->async_dio = 1;
876 		} else {
877 			ra_pages = fc->max_read / PAGE_CACHE_SIZE;
878 			fc->no_lock = 1;
879 			fc->no_flock = 1;
880 		}
881 
882 		fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
883 		fc->minor = arg->minor;
884 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
885 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
886 		fc->conn_init = 1;
887 	}
888 	fc->initialized = 1;
889 	wake_up_all(&fc->blocked_waitq);
890 }
891 
892 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
893 {
894 	struct fuse_init_in *arg = &req->misc.init_in;
895 
896 	arg->major = FUSE_KERNEL_VERSION;
897 	arg->minor = FUSE_KERNEL_MINOR_VERSION;
898 	arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
899 	arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
900 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
901 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
902 		FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
903 		FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO;
904 	req->in.h.opcode = FUSE_INIT;
905 	req->in.numargs = 1;
906 	req->in.args[0].size = sizeof(*arg);
907 	req->in.args[0].value = arg;
908 	req->out.numargs = 1;
909 	/* Variable length argument used for backward compatibility
910 	   with interface version < 7.5.  Rest of init_out is zeroed
911 	   by do_get_request(), so a short reply is not a problem */
912 	req->out.argvar = 1;
913 	req->out.args[0].size = sizeof(struct fuse_init_out);
914 	req->out.args[0].value = &req->misc.init_out;
915 	req->end = process_init_reply;
916 	fuse_request_send_background(fc, req);
917 }
918 
919 static void fuse_free_conn(struct fuse_conn *fc)
920 {
921 	kfree_rcu(fc, rcu);
922 }
923 
924 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
925 {
926 	int err;
927 
928 	fc->bdi.name = "fuse";
929 	fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
930 	/* fuse does it's own writeback accounting */
931 	fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
932 
933 	err = bdi_init(&fc->bdi);
934 	if (err)
935 		return err;
936 
937 	fc->bdi_initialized = 1;
938 
939 	if (sb->s_bdev) {
940 		err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
941 				    MAJOR(fc->dev), MINOR(fc->dev));
942 	} else {
943 		err = bdi_register_dev(&fc->bdi, fc->dev);
944 	}
945 
946 	if (err)
947 		return err;
948 
949 	/*
950 	 * For a single fuse filesystem use max 1% of dirty +
951 	 * writeback threshold.
952 	 *
953 	 * This gives about 1M of write buffer for memory maps on a
954 	 * machine with 1G and 10% dirty_ratio, which should be more
955 	 * than enough.
956 	 *
957 	 * Privileged users can raise it by writing to
958 	 *
959 	 *    /sys/class/bdi/<bdi>/max_ratio
960 	 */
961 	bdi_set_max_ratio(&fc->bdi, 1);
962 
963 	return 0;
964 }
965 
966 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
967 {
968 	struct fuse_conn *fc;
969 	struct inode *root;
970 	struct fuse_mount_data d;
971 	struct file *file;
972 	struct dentry *root_dentry;
973 	struct fuse_req *init_req;
974 	int err;
975 	int is_bdev = sb->s_bdev != NULL;
976 
977 	err = -EINVAL;
978 	if (sb->s_flags & MS_MANDLOCK)
979 		goto err;
980 
981 	sb->s_flags &= ~MS_NOSEC;
982 
983 	if (!parse_fuse_opt((char *) data, &d, is_bdev))
984 		goto err;
985 
986 	if (is_bdev) {
987 #ifdef CONFIG_BLOCK
988 		err = -EINVAL;
989 		if (!sb_set_blocksize(sb, d.blksize))
990 			goto err;
991 #endif
992 	} else {
993 		sb->s_blocksize = PAGE_CACHE_SIZE;
994 		sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
995 	}
996 	sb->s_magic = FUSE_SUPER_MAGIC;
997 	sb->s_op = &fuse_super_operations;
998 	sb->s_maxbytes = MAX_LFS_FILESIZE;
999 	sb->s_time_gran = 1;
1000 	sb->s_export_op = &fuse_export_operations;
1001 
1002 	file = fget(d.fd);
1003 	err = -EINVAL;
1004 	if (!file)
1005 		goto err;
1006 
1007 	if ((file->f_op != &fuse_dev_operations) ||
1008 	    (file->f_cred->user_ns != &init_user_ns))
1009 		goto err_fput;
1010 
1011 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1012 	err = -ENOMEM;
1013 	if (!fc)
1014 		goto err_fput;
1015 
1016 	fuse_conn_init(fc);
1017 
1018 	fc->dev = sb->s_dev;
1019 	fc->sb = sb;
1020 	err = fuse_bdi_init(fc, sb);
1021 	if (err)
1022 		goto err_put_conn;
1023 
1024 	sb->s_bdi = &fc->bdi;
1025 
1026 	/* Handle umasking inside the fuse code */
1027 	if (sb->s_flags & MS_POSIXACL)
1028 		fc->dont_mask = 1;
1029 	sb->s_flags |= MS_POSIXACL;
1030 
1031 	fc->release = fuse_free_conn;
1032 	fc->flags = d.flags;
1033 	fc->user_id = d.user_id;
1034 	fc->group_id = d.group_id;
1035 	fc->max_read = max_t(unsigned, 4096, d.max_read);
1036 
1037 	/* Used by get_root_inode() */
1038 	sb->s_fs_info = fc;
1039 
1040 	err = -ENOMEM;
1041 	root = fuse_get_root_inode(sb, d.rootmode);
1042 	root_dentry = d_make_root(root);
1043 	if (!root_dentry)
1044 		goto err_put_conn;
1045 	/* only now - we want root dentry with NULL ->d_op */
1046 	sb->s_d_op = &fuse_dentry_operations;
1047 
1048 	init_req = fuse_request_alloc(0);
1049 	if (!init_req)
1050 		goto err_put_root;
1051 	init_req->background = 1;
1052 
1053 	if (is_bdev) {
1054 		fc->destroy_req = fuse_request_alloc(0);
1055 		if (!fc->destroy_req)
1056 			goto err_free_init_req;
1057 	}
1058 
1059 	mutex_lock(&fuse_mutex);
1060 	err = -EINVAL;
1061 	if (file->private_data)
1062 		goto err_unlock;
1063 
1064 	err = fuse_ctl_add_conn(fc);
1065 	if (err)
1066 		goto err_unlock;
1067 
1068 	list_add_tail(&fc->entry, &fuse_conn_list);
1069 	sb->s_root = root_dentry;
1070 	fc->connected = 1;
1071 	file->private_data = fuse_conn_get(fc);
1072 	mutex_unlock(&fuse_mutex);
1073 	/*
1074 	 * atomic_dec_and_test() in fput() provides the necessary
1075 	 * memory barrier for file->private_data to be visible on all
1076 	 * CPUs after this
1077 	 */
1078 	fput(file);
1079 
1080 	fuse_send_init(fc, init_req);
1081 
1082 	return 0;
1083 
1084  err_unlock:
1085 	mutex_unlock(&fuse_mutex);
1086  err_free_init_req:
1087 	fuse_request_free(init_req);
1088  err_put_root:
1089 	dput(root_dentry);
1090  err_put_conn:
1091 	fuse_bdi_destroy(fc);
1092 	fuse_conn_put(fc);
1093  err_fput:
1094 	fput(file);
1095  err:
1096 	return err;
1097 }
1098 
1099 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1100 		       int flags, const char *dev_name,
1101 		       void *raw_data)
1102 {
1103 	return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1104 }
1105 
1106 static void fuse_kill_sb_anon(struct super_block *sb)
1107 {
1108 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1109 
1110 	if (fc) {
1111 		down_write(&fc->killsb);
1112 		fc->sb = NULL;
1113 		up_write(&fc->killsb);
1114 	}
1115 
1116 	kill_anon_super(sb);
1117 }
1118 
1119 static struct file_system_type fuse_fs_type = {
1120 	.owner		= THIS_MODULE,
1121 	.name		= "fuse",
1122 	.fs_flags	= FS_HAS_SUBTYPE,
1123 	.mount		= fuse_mount,
1124 	.kill_sb	= fuse_kill_sb_anon,
1125 };
1126 MODULE_ALIAS_FS("fuse");
1127 
1128 #ifdef CONFIG_BLOCK
1129 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1130 			   int flags, const char *dev_name,
1131 			   void *raw_data)
1132 {
1133 	return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1134 }
1135 
1136 static void fuse_kill_sb_blk(struct super_block *sb)
1137 {
1138 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1139 
1140 	if (fc) {
1141 		down_write(&fc->killsb);
1142 		fc->sb = NULL;
1143 		up_write(&fc->killsb);
1144 	}
1145 
1146 	kill_block_super(sb);
1147 }
1148 
1149 static struct file_system_type fuseblk_fs_type = {
1150 	.owner		= THIS_MODULE,
1151 	.name		= "fuseblk",
1152 	.mount		= fuse_mount_blk,
1153 	.kill_sb	= fuse_kill_sb_blk,
1154 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1155 };
1156 MODULE_ALIAS_FS("fuseblk");
1157 
1158 static inline int register_fuseblk(void)
1159 {
1160 	return register_filesystem(&fuseblk_fs_type);
1161 }
1162 
1163 static inline void unregister_fuseblk(void)
1164 {
1165 	unregister_filesystem(&fuseblk_fs_type);
1166 }
1167 #else
1168 static inline int register_fuseblk(void)
1169 {
1170 	return 0;
1171 }
1172 
1173 static inline void unregister_fuseblk(void)
1174 {
1175 }
1176 #endif
1177 
1178 static void fuse_inode_init_once(void *foo)
1179 {
1180 	struct inode *inode = foo;
1181 
1182 	inode_init_once(inode);
1183 }
1184 
1185 static int __init fuse_fs_init(void)
1186 {
1187 	int err;
1188 
1189 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
1190 					      sizeof(struct fuse_inode),
1191 					      0, SLAB_HWCACHE_ALIGN,
1192 					      fuse_inode_init_once);
1193 	err = -ENOMEM;
1194 	if (!fuse_inode_cachep)
1195 		goto out;
1196 
1197 	err = register_fuseblk();
1198 	if (err)
1199 		goto out2;
1200 
1201 	err = register_filesystem(&fuse_fs_type);
1202 	if (err)
1203 		goto out3;
1204 
1205 	return 0;
1206 
1207  out3:
1208 	unregister_fuseblk();
1209  out2:
1210 	kmem_cache_destroy(fuse_inode_cachep);
1211  out:
1212 	return err;
1213 }
1214 
1215 static void fuse_fs_cleanup(void)
1216 {
1217 	unregister_filesystem(&fuse_fs_type);
1218 	unregister_fuseblk();
1219 
1220 	/*
1221 	 * Make sure all delayed rcu free inodes are flushed before we
1222 	 * destroy cache.
1223 	 */
1224 	rcu_barrier();
1225 	kmem_cache_destroy(fuse_inode_cachep);
1226 }
1227 
1228 static struct kobject *fuse_kobj;
1229 static struct kobject *connections_kobj;
1230 
1231 static int fuse_sysfs_init(void)
1232 {
1233 	int err;
1234 
1235 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1236 	if (!fuse_kobj) {
1237 		err = -ENOMEM;
1238 		goto out_err;
1239 	}
1240 
1241 	connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1242 	if (!connections_kobj) {
1243 		err = -ENOMEM;
1244 		goto out_fuse_unregister;
1245 	}
1246 
1247 	return 0;
1248 
1249  out_fuse_unregister:
1250 	kobject_put(fuse_kobj);
1251  out_err:
1252 	return err;
1253 }
1254 
1255 static void fuse_sysfs_cleanup(void)
1256 {
1257 	kobject_put(connections_kobj);
1258 	kobject_put(fuse_kobj);
1259 }
1260 
1261 static int __init fuse_init(void)
1262 {
1263 	int res;
1264 
1265 	printk(KERN_INFO "fuse init (API version %i.%i)\n",
1266 	       FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1267 
1268 	INIT_LIST_HEAD(&fuse_conn_list);
1269 	res = fuse_fs_init();
1270 	if (res)
1271 		goto err;
1272 
1273 	res = fuse_dev_init();
1274 	if (res)
1275 		goto err_fs_cleanup;
1276 
1277 	res = fuse_sysfs_init();
1278 	if (res)
1279 		goto err_dev_cleanup;
1280 
1281 	res = fuse_ctl_init();
1282 	if (res)
1283 		goto err_sysfs_cleanup;
1284 
1285 	sanitize_global_limit(&max_user_bgreq);
1286 	sanitize_global_limit(&max_user_congthresh);
1287 
1288 	return 0;
1289 
1290  err_sysfs_cleanup:
1291 	fuse_sysfs_cleanup();
1292  err_dev_cleanup:
1293 	fuse_dev_cleanup();
1294  err_fs_cleanup:
1295 	fuse_fs_cleanup();
1296  err:
1297 	return res;
1298 }
1299 
1300 static void __exit fuse_exit(void)
1301 {
1302 	printk(KERN_DEBUG "fuse exit\n");
1303 
1304 	fuse_ctl_cleanup();
1305 	fuse_sysfs_cleanup();
1306 	fuse_fs_cleanup();
1307 	fuse_dev_cleanup();
1308 }
1309 
1310 module_init(fuse_init);
1311 module_exit(fuse_exit);
1312