xref: /openbmc/linux/fs/fuse/inode.c (revision e8e0929d)
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 static struct inode *fuse_alloc_inode(struct super_block *sb)
75 {
76 	struct inode *inode;
77 	struct fuse_inode *fi;
78 
79 	inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
80 	if (!inode)
81 		return NULL;
82 
83 	fi = get_fuse_inode(inode);
84 	fi->i_time = 0;
85 	fi->nodeid = 0;
86 	fi->nlookup = 0;
87 	fi->attr_version = 0;
88 	fi->writectr = 0;
89 	INIT_LIST_HEAD(&fi->write_files);
90 	INIT_LIST_HEAD(&fi->queued_writes);
91 	INIT_LIST_HEAD(&fi->writepages);
92 	init_waitqueue_head(&fi->page_waitq);
93 	fi->forget_req = fuse_request_alloc();
94 	if (!fi->forget_req) {
95 		kmem_cache_free(fuse_inode_cachep, inode);
96 		return NULL;
97 	}
98 
99 	return inode;
100 }
101 
102 static void fuse_destroy_inode(struct inode *inode)
103 {
104 	struct fuse_inode *fi = get_fuse_inode(inode);
105 	BUG_ON(!list_empty(&fi->write_files));
106 	BUG_ON(!list_empty(&fi->queued_writes));
107 	if (fi->forget_req)
108 		fuse_request_free(fi->forget_req);
109 	kmem_cache_free(fuse_inode_cachep, inode);
110 }
111 
112 void fuse_send_forget(struct fuse_conn *fc, struct fuse_req *req,
113 		      u64 nodeid, u64 nlookup)
114 {
115 	struct fuse_forget_in *inarg = &req->misc.forget_in;
116 	inarg->nlookup = nlookup;
117 	req->in.h.opcode = FUSE_FORGET;
118 	req->in.h.nodeid = nodeid;
119 	req->in.numargs = 1;
120 	req->in.args[0].size = sizeof(struct fuse_forget_in);
121 	req->in.args[0].value = inarg;
122 	fuse_request_send_noreply(fc, req);
123 }
124 
125 static void fuse_clear_inode(struct inode *inode)
126 {
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_send_forget(fc, fi->forget_req, fi->nodeid, fi->nlookup);
131 		fi->forget_req = 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 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
144 				   u64 attr_valid)
145 {
146 	struct fuse_conn *fc = get_fuse_conn(inode);
147 	struct fuse_inode *fi = get_fuse_inode(inode);
148 
149 	fi->attr_version = ++fc->attr_version;
150 	fi->i_time = attr_valid;
151 
152 	inode->i_ino     = attr->ino;
153 	inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
154 	inode->i_nlink   = attr->nlink;
155 	inode->i_uid     = attr->uid;
156 	inode->i_gid     = attr->gid;
157 	inode->i_blocks  = attr->blocks;
158 	inode->i_atime.tv_sec   = attr->atime;
159 	inode->i_atime.tv_nsec  = attr->atimensec;
160 	inode->i_mtime.tv_sec   = attr->mtime;
161 	inode->i_mtime.tv_nsec  = attr->mtimensec;
162 	inode->i_ctime.tv_sec   = attr->ctime;
163 	inode->i_ctime.tv_nsec  = attr->ctimensec;
164 
165 	if (attr->blksize != 0)
166 		inode->i_blkbits = ilog2(attr->blksize);
167 	else
168 		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
169 
170 	/*
171 	 * Don't set the sticky bit in i_mode, unless we want the VFS
172 	 * to check permissions.  This prevents failures due to the
173 	 * check in may_delete().
174 	 */
175 	fi->orig_i_mode = inode->i_mode;
176 	if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
177 		inode->i_mode &= ~S_ISVTX;
178 }
179 
180 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
181 			    u64 attr_valid, u64 attr_version)
182 {
183 	struct fuse_conn *fc = get_fuse_conn(inode);
184 	struct fuse_inode *fi = get_fuse_inode(inode);
185 	loff_t oldsize;
186 
187 	spin_lock(&fc->lock);
188 	if (attr_version != 0 && fi->attr_version > attr_version) {
189 		spin_unlock(&fc->lock);
190 		return;
191 	}
192 
193 	fuse_change_attributes_common(inode, attr, attr_valid);
194 
195 	oldsize = inode->i_size;
196 	i_size_write(inode, attr->size);
197 	spin_unlock(&fc->lock);
198 
199 	if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
200 		truncate_pagecache(inode, oldsize, attr->size);
201 		invalidate_inode_pages2(inode->i_mapping);
202 	}
203 }
204 
205 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
206 {
207 	inode->i_mode = attr->mode & S_IFMT;
208 	inode->i_size = attr->size;
209 	if (S_ISREG(inode->i_mode)) {
210 		fuse_init_common(inode);
211 		fuse_init_file_inode(inode);
212 	} else if (S_ISDIR(inode->i_mode))
213 		fuse_init_dir(inode);
214 	else if (S_ISLNK(inode->i_mode))
215 		fuse_init_symlink(inode);
216 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
217 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
218 		fuse_init_common(inode);
219 		init_special_inode(inode, inode->i_mode,
220 				   new_decode_dev(attr->rdev));
221 	} else
222 		BUG();
223 }
224 
225 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
226 {
227 	u64 nodeid = *(u64 *) _nodeidp;
228 	if (get_node_id(inode) == nodeid)
229 		return 1;
230 	else
231 		return 0;
232 }
233 
234 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
235 {
236 	u64 nodeid = *(u64 *) _nodeidp;
237 	get_fuse_inode(inode)->nodeid = nodeid;
238 	return 0;
239 }
240 
241 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
242 			int generation, struct fuse_attr *attr,
243 			u64 attr_valid, u64 attr_version)
244 {
245 	struct inode *inode;
246 	struct fuse_inode *fi;
247 	struct fuse_conn *fc = get_fuse_conn_super(sb);
248 
249  retry:
250 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
251 	if (!inode)
252 		return NULL;
253 
254 	if ((inode->i_state & I_NEW)) {
255 		inode->i_flags |= S_NOATIME|S_NOCMTIME;
256 		inode->i_generation = generation;
257 		inode->i_data.backing_dev_info = &fc->bdi;
258 		fuse_init_inode(inode, attr);
259 		unlock_new_inode(inode);
260 	} else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
261 		/* Inode has changed type, any I/O on the old should fail */
262 		make_bad_inode(inode);
263 		iput(inode);
264 		goto retry;
265 	}
266 
267 	fi = get_fuse_inode(inode);
268 	spin_lock(&fc->lock);
269 	fi->nlookup++;
270 	spin_unlock(&fc->lock);
271 	fuse_change_attributes(inode, attr, attr_valid, attr_version);
272 
273 	return inode;
274 }
275 
276 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
277 			     loff_t offset, loff_t len)
278 {
279 	struct inode *inode;
280 	pgoff_t pg_start;
281 	pgoff_t pg_end;
282 
283 	inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
284 	if (!inode)
285 		return -ENOENT;
286 
287 	fuse_invalidate_attr(inode);
288 	if (offset >= 0) {
289 		pg_start = offset >> PAGE_CACHE_SHIFT;
290 		if (len <= 0)
291 			pg_end = -1;
292 		else
293 			pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
294 		invalidate_inode_pages2_range(inode->i_mapping,
295 					      pg_start, pg_end);
296 	}
297 	iput(inode);
298 	return 0;
299 }
300 
301 static void fuse_umount_begin(struct super_block *sb)
302 {
303 	fuse_abort_conn(get_fuse_conn_super(sb));
304 }
305 
306 static void fuse_send_destroy(struct fuse_conn *fc)
307 {
308 	struct fuse_req *req = fc->destroy_req;
309 	if (req && fc->conn_init) {
310 		fc->destroy_req = NULL;
311 		req->in.h.opcode = FUSE_DESTROY;
312 		req->force = 1;
313 		fuse_request_send(fc, req);
314 		fuse_put_request(fc, req);
315 	}
316 }
317 
318 static void fuse_bdi_destroy(struct fuse_conn *fc)
319 {
320 	if (fc->bdi_initialized)
321 		bdi_destroy(&fc->bdi);
322 }
323 
324 void fuse_conn_kill(struct fuse_conn *fc)
325 {
326 	spin_lock(&fc->lock);
327 	fc->connected = 0;
328 	fc->blocked = 0;
329 	spin_unlock(&fc->lock);
330 	/* Flush all readers on this fs */
331 	kill_fasync(&fc->fasync, SIGIO, POLL_IN);
332 	wake_up_all(&fc->waitq);
333 	wake_up_all(&fc->blocked_waitq);
334 	wake_up_all(&fc->reserved_req_waitq);
335 	mutex_lock(&fuse_mutex);
336 	list_del(&fc->entry);
337 	fuse_ctl_remove_conn(fc);
338 	mutex_unlock(&fuse_mutex);
339 	fuse_bdi_destroy(fc);
340 }
341 EXPORT_SYMBOL_GPL(fuse_conn_kill);
342 
343 static void fuse_put_super(struct super_block *sb)
344 {
345 	struct fuse_conn *fc = get_fuse_conn_super(sb);
346 
347 	fuse_send_destroy(fc);
348 	fuse_conn_kill(fc);
349 	fuse_conn_put(fc);
350 }
351 
352 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
353 {
354 	stbuf->f_type    = FUSE_SUPER_MAGIC;
355 	stbuf->f_bsize   = attr->bsize;
356 	stbuf->f_frsize  = attr->frsize;
357 	stbuf->f_blocks  = attr->blocks;
358 	stbuf->f_bfree   = attr->bfree;
359 	stbuf->f_bavail  = attr->bavail;
360 	stbuf->f_files   = attr->files;
361 	stbuf->f_ffree   = attr->ffree;
362 	stbuf->f_namelen = attr->namelen;
363 	/* fsid is left zero */
364 }
365 
366 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
367 {
368 	struct super_block *sb = dentry->d_sb;
369 	struct fuse_conn *fc = get_fuse_conn_super(sb);
370 	struct fuse_req *req;
371 	struct fuse_statfs_out outarg;
372 	int err;
373 
374 	if (!fuse_allow_task(fc, current)) {
375 		buf->f_type = FUSE_SUPER_MAGIC;
376 		return 0;
377 	}
378 
379 	req = fuse_get_req(fc);
380 	if (IS_ERR(req))
381 		return PTR_ERR(req);
382 
383 	memset(&outarg, 0, sizeof(outarg));
384 	req->in.numargs = 0;
385 	req->in.h.opcode = FUSE_STATFS;
386 	req->in.h.nodeid = get_node_id(dentry->d_inode);
387 	req->out.numargs = 1;
388 	req->out.args[0].size =
389 		fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
390 	req->out.args[0].value = &outarg;
391 	fuse_request_send(fc, req);
392 	err = req->out.h.error;
393 	if (!err)
394 		convert_fuse_statfs(buf, &outarg.st);
395 	fuse_put_request(fc, req);
396 	return err;
397 }
398 
399 enum {
400 	OPT_FD,
401 	OPT_ROOTMODE,
402 	OPT_USER_ID,
403 	OPT_GROUP_ID,
404 	OPT_DEFAULT_PERMISSIONS,
405 	OPT_ALLOW_OTHER,
406 	OPT_MAX_READ,
407 	OPT_BLKSIZE,
408 	OPT_ERR
409 };
410 
411 static const match_table_t tokens = {
412 	{OPT_FD,			"fd=%u"},
413 	{OPT_ROOTMODE,			"rootmode=%o"},
414 	{OPT_USER_ID,			"user_id=%u"},
415 	{OPT_GROUP_ID,			"group_id=%u"},
416 	{OPT_DEFAULT_PERMISSIONS,	"default_permissions"},
417 	{OPT_ALLOW_OTHER,		"allow_other"},
418 	{OPT_MAX_READ,			"max_read=%u"},
419 	{OPT_BLKSIZE,			"blksize=%u"},
420 	{OPT_ERR,			NULL}
421 };
422 
423 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
424 {
425 	char *p;
426 	memset(d, 0, sizeof(struct fuse_mount_data));
427 	d->max_read = ~0;
428 	d->blksize = FUSE_DEFAULT_BLKSIZE;
429 
430 	while ((p = strsep(&opt, ",")) != NULL) {
431 		int token;
432 		int value;
433 		substring_t args[MAX_OPT_ARGS];
434 		if (!*p)
435 			continue;
436 
437 		token = match_token(p, tokens, args);
438 		switch (token) {
439 		case OPT_FD:
440 			if (match_int(&args[0], &value))
441 				return 0;
442 			d->fd = value;
443 			d->fd_present = 1;
444 			break;
445 
446 		case OPT_ROOTMODE:
447 			if (match_octal(&args[0], &value))
448 				return 0;
449 			if (!fuse_valid_type(value))
450 				return 0;
451 			d->rootmode = value;
452 			d->rootmode_present = 1;
453 			break;
454 
455 		case OPT_USER_ID:
456 			if (match_int(&args[0], &value))
457 				return 0;
458 			d->user_id = value;
459 			d->user_id_present = 1;
460 			break;
461 
462 		case OPT_GROUP_ID:
463 			if (match_int(&args[0], &value))
464 				return 0;
465 			d->group_id = value;
466 			d->group_id_present = 1;
467 			break;
468 
469 		case OPT_DEFAULT_PERMISSIONS:
470 			d->flags |= FUSE_DEFAULT_PERMISSIONS;
471 			break;
472 
473 		case OPT_ALLOW_OTHER:
474 			d->flags |= FUSE_ALLOW_OTHER;
475 			break;
476 
477 		case OPT_MAX_READ:
478 			if (match_int(&args[0], &value))
479 				return 0;
480 			d->max_read = value;
481 			break;
482 
483 		case OPT_BLKSIZE:
484 			if (!is_bdev || match_int(&args[0], &value))
485 				return 0;
486 			d->blksize = value;
487 			break;
488 
489 		default:
490 			return 0;
491 		}
492 	}
493 
494 	if (!d->fd_present || !d->rootmode_present ||
495 	    !d->user_id_present || !d->group_id_present)
496 		return 0;
497 
498 	return 1;
499 }
500 
501 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
502 {
503 	struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
504 
505 	seq_printf(m, ",user_id=%u", fc->user_id);
506 	seq_printf(m, ",group_id=%u", fc->group_id);
507 	if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
508 		seq_puts(m, ",default_permissions");
509 	if (fc->flags & FUSE_ALLOW_OTHER)
510 		seq_puts(m, ",allow_other");
511 	if (fc->max_read != ~0)
512 		seq_printf(m, ",max_read=%u", fc->max_read);
513 	if (mnt->mnt_sb->s_bdev &&
514 	    mnt->mnt_sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
515 		seq_printf(m, ",blksize=%lu", mnt->mnt_sb->s_blocksize);
516 	return 0;
517 }
518 
519 void fuse_conn_init(struct fuse_conn *fc)
520 {
521 	memset(fc, 0, sizeof(*fc));
522 	spin_lock_init(&fc->lock);
523 	mutex_init(&fc->inst_mutex);
524 	init_rwsem(&fc->killsb);
525 	atomic_set(&fc->count, 1);
526 	init_waitqueue_head(&fc->waitq);
527 	init_waitqueue_head(&fc->blocked_waitq);
528 	init_waitqueue_head(&fc->reserved_req_waitq);
529 	INIT_LIST_HEAD(&fc->pending);
530 	INIT_LIST_HEAD(&fc->processing);
531 	INIT_LIST_HEAD(&fc->io);
532 	INIT_LIST_HEAD(&fc->interrupts);
533 	INIT_LIST_HEAD(&fc->bg_queue);
534 	INIT_LIST_HEAD(&fc->entry);
535 	atomic_set(&fc->num_waiting, 0);
536 	fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
537 	fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
538 	fc->khctr = 0;
539 	fc->polled_files = RB_ROOT;
540 	fc->reqctr = 0;
541 	fc->blocked = 1;
542 	fc->attr_version = 1;
543 	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
544 }
545 EXPORT_SYMBOL_GPL(fuse_conn_init);
546 
547 void fuse_conn_put(struct fuse_conn *fc)
548 {
549 	if (atomic_dec_and_test(&fc->count)) {
550 		if (fc->destroy_req)
551 			fuse_request_free(fc->destroy_req);
552 		mutex_destroy(&fc->inst_mutex);
553 		fc->release(fc);
554 	}
555 }
556 EXPORT_SYMBOL_GPL(fuse_conn_put);
557 
558 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
559 {
560 	atomic_inc(&fc->count);
561 	return fc;
562 }
563 EXPORT_SYMBOL_GPL(fuse_conn_get);
564 
565 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
566 {
567 	struct fuse_attr attr;
568 	memset(&attr, 0, sizeof(attr));
569 
570 	attr.mode = mode;
571 	attr.ino = FUSE_ROOT_ID;
572 	attr.nlink = 1;
573 	return fuse_iget(sb, 1, 0, &attr, 0, 0);
574 }
575 
576 struct fuse_inode_handle {
577 	u64 nodeid;
578 	u32 generation;
579 };
580 
581 static struct dentry *fuse_get_dentry(struct super_block *sb,
582 				      struct fuse_inode_handle *handle)
583 {
584 	struct fuse_conn *fc = get_fuse_conn_super(sb);
585 	struct inode *inode;
586 	struct dentry *entry;
587 	int err = -ESTALE;
588 
589 	if (handle->nodeid == 0)
590 		goto out_err;
591 
592 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
593 	if (!inode) {
594 		struct fuse_entry_out outarg;
595 		struct qstr name;
596 
597 		if (!fc->export_support)
598 			goto out_err;
599 
600 		name.len = 1;
601 		name.name = ".";
602 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
603 				       &inode);
604 		if (err && err != -ENOENT)
605 			goto out_err;
606 		if (err || !inode) {
607 			err = -ESTALE;
608 			goto out_err;
609 		}
610 		err = -EIO;
611 		if (get_node_id(inode) != handle->nodeid)
612 			goto out_iput;
613 	}
614 	err = -ESTALE;
615 	if (inode->i_generation != handle->generation)
616 		goto out_iput;
617 
618 	entry = d_obtain_alias(inode);
619 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID) {
620 		entry->d_op = &fuse_dentry_operations;
621 		fuse_invalidate_entry_cache(entry);
622 	}
623 
624 	return entry;
625 
626  out_iput:
627 	iput(inode);
628  out_err:
629 	return ERR_PTR(err);
630 }
631 
632 static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
633 			   int connectable)
634 {
635 	struct inode *inode = dentry->d_inode;
636 	bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
637 	int len = encode_parent ? 6 : 3;
638 	u64 nodeid;
639 	u32 generation;
640 
641 	if (*max_len < len)
642 		return  255;
643 
644 	nodeid = get_fuse_inode(inode)->nodeid;
645 	generation = inode->i_generation;
646 
647 	fh[0] = (u32)(nodeid >> 32);
648 	fh[1] = (u32)(nodeid & 0xffffffff);
649 	fh[2] = generation;
650 
651 	if (encode_parent) {
652 		struct inode *parent;
653 
654 		spin_lock(&dentry->d_lock);
655 		parent = dentry->d_parent->d_inode;
656 		nodeid = get_fuse_inode(parent)->nodeid;
657 		generation = parent->i_generation;
658 		spin_unlock(&dentry->d_lock);
659 
660 		fh[3] = (u32)(nodeid >> 32);
661 		fh[4] = (u32)(nodeid & 0xffffffff);
662 		fh[5] = generation;
663 	}
664 
665 	*max_len = len;
666 	return encode_parent ? 0x82 : 0x81;
667 }
668 
669 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
670 		struct fid *fid, int fh_len, int fh_type)
671 {
672 	struct fuse_inode_handle handle;
673 
674 	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
675 		return NULL;
676 
677 	handle.nodeid = (u64) fid->raw[0] << 32;
678 	handle.nodeid |= (u64) fid->raw[1];
679 	handle.generation = fid->raw[2];
680 	return fuse_get_dentry(sb, &handle);
681 }
682 
683 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
684 		struct fid *fid, int fh_len, int fh_type)
685 {
686 	struct fuse_inode_handle parent;
687 
688 	if (fh_type != 0x82 || fh_len < 6)
689 		return NULL;
690 
691 	parent.nodeid = (u64) fid->raw[3] << 32;
692 	parent.nodeid |= (u64) fid->raw[4];
693 	parent.generation = fid->raw[5];
694 	return fuse_get_dentry(sb, &parent);
695 }
696 
697 static struct dentry *fuse_get_parent(struct dentry *child)
698 {
699 	struct inode *child_inode = child->d_inode;
700 	struct fuse_conn *fc = get_fuse_conn(child_inode);
701 	struct inode *inode;
702 	struct dentry *parent;
703 	struct fuse_entry_out outarg;
704 	struct qstr name;
705 	int err;
706 
707 	if (!fc->export_support)
708 		return ERR_PTR(-ESTALE);
709 
710 	name.len = 2;
711 	name.name = "..";
712 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
713 			       &name, &outarg, &inode);
714 	if (err) {
715 		if (err == -ENOENT)
716 			return ERR_PTR(-ESTALE);
717 		return ERR_PTR(err);
718 	}
719 
720 	parent = d_obtain_alias(inode);
721 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID) {
722 		parent->d_op = &fuse_dentry_operations;
723 		fuse_invalidate_entry_cache(parent);
724 	}
725 
726 	return parent;
727 }
728 
729 static const struct export_operations fuse_export_operations = {
730 	.fh_to_dentry	= fuse_fh_to_dentry,
731 	.fh_to_parent	= fuse_fh_to_parent,
732 	.encode_fh	= fuse_encode_fh,
733 	.get_parent	= fuse_get_parent,
734 };
735 
736 static const struct super_operations fuse_super_operations = {
737 	.alloc_inode    = fuse_alloc_inode,
738 	.destroy_inode  = fuse_destroy_inode,
739 	.clear_inode	= fuse_clear_inode,
740 	.drop_inode	= generic_delete_inode,
741 	.remount_fs	= fuse_remount_fs,
742 	.put_super	= fuse_put_super,
743 	.umount_begin	= fuse_umount_begin,
744 	.statfs		= fuse_statfs,
745 	.show_options	= fuse_show_options,
746 };
747 
748 static void sanitize_global_limit(unsigned *limit)
749 {
750 	if (*limit == 0)
751 		*limit = ((num_physpages << PAGE_SHIFT) >> 13) /
752 			 sizeof(struct fuse_req);
753 
754 	if (*limit >= 1 << 16)
755 		*limit = (1 << 16) - 1;
756 }
757 
758 static int set_global_limit(const char *val, struct kernel_param *kp)
759 {
760 	int rv;
761 
762 	rv = param_set_uint(val, kp);
763 	if (rv)
764 		return rv;
765 
766 	sanitize_global_limit((unsigned *)kp->arg);
767 
768 	return 0;
769 }
770 
771 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
772 {
773 	int cap_sys_admin = capable(CAP_SYS_ADMIN);
774 
775 	if (arg->minor < 13)
776 		return;
777 
778 	sanitize_global_limit(&max_user_bgreq);
779 	sanitize_global_limit(&max_user_congthresh);
780 
781 	if (arg->max_background) {
782 		fc->max_background = arg->max_background;
783 
784 		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
785 			fc->max_background = max_user_bgreq;
786 	}
787 	if (arg->congestion_threshold) {
788 		fc->congestion_threshold = arg->congestion_threshold;
789 
790 		if (!cap_sys_admin &&
791 		    fc->congestion_threshold > max_user_congthresh)
792 			fc->congestion_threshold = max_user_congthresh;
793 	}
794 }
795 
796 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
797 {
798 	struct fuse_init_out *arg = &req->misc.init_out;
799 
800 	if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
801 		fc->conn_error = 1;
802 	else {
803 		unsigned long ra_pages;
804 
805 		process_init_limits(fc, arg);
806 
807 		if (arg->minor >= 6) {
808 			ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
809 			if (arg->flags & FUSE_ASYNC_READ)
810 				fc->async_read = 1;
811 			if (!(arg->flags & FUSE_POSIX_LOCKS))
812 				fc->no_lock = 1;
813 			if (arg->flags & FUSE_ATOMIC_O_TRUNC)
814 				fc->atomic_o_trunc = 1;
815 			if (arg->minor >= 9) {
816 				/* LOOKUP has dependency on proto version */
817 				if (arg->flags & FUSE_EXPORT_SUPPORT)
818 					fc->export_support = 1;
819 			}
820 			if (arg->flags & FUSE_BIG_WRITES)
821 				fc->big_writes = 1;
822 			if (arg->flags & FUSE_DONT_MASK)
823 				fc->dont_mask = 1;
824 		} else {
825 			ra_pages = fc->max_read / PAGE_CACHE_SIZE;
826 			fc->no_lock = 1;
827 		}
828 
829 		fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
830 		fc->minor = arg->minor;
831 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
832 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
833 		fc->conn_init = 1;
834 	}
835 	fc->blocked = 0;
836 	wake_up_all(&fc->blocked_waitq);
837 }
838 
839 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
840 {
841 	struct fuse_init_in *arg = &req->misc.init_in;
842 
843 	arg->major = FUSE_KERNEL_VERSION;
844 	arg->minor = FUSE_KERNEL_MINOR_VERSION;
845 	arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
846 	arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
847 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK;
848 	req->in.h.opcode = FUSE_INIT;
849 	req->in.numargs = 1;
850 	req->in.args[0].size = sizeof(*arg);
851 	req->in.args[0].value = arg;
852 	req->out.numargs = 1;
853 	/* Variable length arguement used for backward compatibility
854 	   with interface version < 7.5.  Rest of init_out is zeroed
855 	   by do_get_request(), so a short reply is not a problem */
856 	req->out.argvar = 1;
857 	req->out.args[0].size = sizeof(struct fuse_init_out);
858 	req->out.args[0].value = &req->misc.init_out;
859 	req->end = process_init_reply;
860 	fuse_request_send_background(fc, req);
861 }
862 
863 static void fuse_free_conn(struct fuse_conn *fc)
864 {
865 	kfree(fc);
866 }
867 
868 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
869 {
870 	int err;
871 
872 	fc->bdi.name = "fuse";
873 	fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
874 	fc->bdi.unplug_io_fn = default_unplug_io_fn;
875 	/* fuse does it's own writeback accounting */
876 	fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
877 
878 	err = bdi_init(&fc->bdi);
879 	if (err)
880 		return err;
881 
882 	fc->bdi_initialized = 1;
883 
884 	if (sb->s_bdev) {
885 		err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
886 				    MAJOR(fc->dev), MINOR(fc->dev));
887 	} else {
888 		err = bdi_register_dev(&fc->bdi, fc->dev);
889 	}
890 
891 	if (err)
892 		return err;
893 
894 	/*
895 	 * For a single fuse filesystem use max 1% of dirty +
896 	 * writeback threshold.
897 	 *
898 	 * This gives about 1M of write buffer for memory maps on a
899 	 * machine with 1G and 10% dirty_ratio, which should be more
900 	 * than enough.
901 	 *
902 	 * Privileged users can raise it by writing to
903 	 *
904 	 *    /sys/class/bdi/<bdi>/max_ratio
905 	 */
906 	bdi_set_max_ratio(&fc->bdi, 1);
907 
908 	return 0;
909 }
910 
911 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
912 {
913 	struct fuse_conn *fc;
914 	struct inode *root;
915 	struct fuse_mount_data d;
916 	struct file *file;
917 	struct dentry *root_dentry;
918 	struct fuse_req *init_req;
919 	int err;
920 	int is_bdev = sb->s_bdev != NULL;
921 
922 	err = -EINVAL;
923 	if (sb->s_flags & MS_MANDLOCK)
924 		goto err;
925 
926 	if (!parse_fuse_opt((char *) data, &d, is_bdev))
927 		goto err;
928 
929 	if (is_bdev) {
930 #ifdef CONFIG_BLOCK
931 		err = -EINVAL;
932 		if (!sb_set_blocksize(sb, d.blksize))
933 			goto err;
934 #endif
935 	} else {
936 		sb->s_blocksize = PAGE_CACHE_SIZE;
937 		sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
938 	}
939 	sb->s_magic = FUSE_SUPER_MAGIC;
940 	sb->s_op = &fuse_super_operations;
941 	sb->s_maxbytes = MAX_LFS_FILESIZE;
942 	sb->s_export_op = &fuse_export_operations;
943 
944 	file = fget(d.fd);
945 	err = -EINVAL;
946 	if (!file)
947 		goto err;
948 
949 	if (file->f_op != &fuse_dev_operations)
950 		goto err_fput;
951 
952 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
953 	err = -ENOMEM;
954 	if (!fc)
955 		goto err_fput;
956 
957 	fuse_conn_init(fc);
958 
959 	fc->dev = sb->s_dev;
960 	fc->sb = sb;
961 	err = fuse_bdi_init(fc, sb);
962 	if (err)
963 		goto err_put_conn;
964 
965 	sb->s_bdi = &fc->bdi;
966 
967 	/* Handle umasking inside the fuse code */
968 	if (sb->s_flags & MS_POSIXACL)
969 		fc->dont_mask = 1;
970 	sb->s_flags |= MS_POSIXACL;
971 
972 	fc->release = fuse_free_conn;
973 	fc->flags = d.flags;
974 	fc->user_id = d.user_id;
975 	fc->group_id = d.group_id;
976 	fc->max_read = max_t(unsigned, 4096, d.max_read);
977 
978 	/* Used by get_root_inode() */
979 	sb->s_fs_info = fc;
980 
981 	err = -ENOMEM;
982 	root = fuse_get_root_inode(sb, d.rootmode);
983 	if (!root)
984 		goto err_put_conn;
985 
986 	root_dentry = d_alloc_root(root);
987 	if (!root_dentry) {
988 		iput(root);
989 		goto err_put_conn;
990 	}
991 
992 	init_req = fuse_request_alloc();
993 	if (!init_req)
994 		goto err_put_root;
995 
996 	if (is_bdev) {
997 		fc->destroy_req = fuse_request_alloc();
998 		if (!fc->destroy_req)
999 			goto err_free_init_req;
1000 	}
1001 
1002 	mutex_lock(&fuse_mutex);
1003 	err = -EINVAL;
1004 	if (file->private_data)
1005 		goto err_unlock;
1006 
1007 	err = fuse_ctl_add_conn(fc);
1008 	if (err)
1009 		goto err_unlock;
1010 
1011 	list_add_tail(&fc->entry, &fuse_conn_list);
1012 	sb->s_root = root_dentry;
1013 	fc->connected = 1;
1014 	file->private_data = fuse_conn_get(fc);
1015 	mutex_unlock(&fuse_mutex);
1016 	/*
1017 	 * atomic_dec_and_test() in fput() provides the necessary
1018 	 * memory barrier for file->private_data to be visible on all
1019 	 * CPUs after this
1020 	 */
1021 	fput(file);
1022 
1023 	fuse_send_init(fc, init_req);
1024 
1025 	return 0;
1026 
1027  err_unlock:
1028 	mutex_unlock(&fuse_mutex);
1029  err_free_init_req:
1030 	fuse_request_free(init_req);
1031  err_put_root:
1032 	dput(root_dentry);
1033  err_put_conn:
1034 	fuse_bdi_destroy(fc);
1035 	fuse_conn_put(fc);
1036  err_fput:
1037 	fput(file);
1038  err:
1039 	return err;
1040 }
1041 
1042 static int fuse_get_sb(struct file_system_type *fs_type,
1043 		       int flags, const char *dev_name,
1044 		       void *raw_data, struct vfsmount *mnt)
1045 {
1046 	return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
1047 }
1048 
1049 static void fuse_kill_sb_anon(struct super_block *sb)
1050 {
1051 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1052 
1053 	if (fc) {
1054 		down_write(&fc->killsb);
1055 		fc->sb = NULL;
1056 		up_write(&fc->killsb);
1057 	}
1058 
1059 	kill_anon_super(sb);
1060 }
1061 
1062 static struct file_system_type fuse_fs_type = {
1063 	.owner		= THIS_MODULE,
1064 	.name		= "fuse",
1065 	.fs_flags	= FS_HAS_SUBTYPE,
1066 	.get_sb		= fuse_get_sb,
1067 	.kill_sb	= fuse_kill_sb_anon,
1068 };
1069 
1070 #ifdef CONFIG_BLOCK
1071 static int fuse_get_sb_blk(struct file_system_type *fs_type,
1072 			   int flags, const char *dev_name,
1073 			   void *raw_data, struct vfsmount *mnt)
1074 {
1075 	return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
1076 			   mnt);
1077 }
1078 
1079 static void fuse_kill_sb_blk(struct super_block *sb)
1080 {
1081 	struct fuse_conn *fc = get_fuse_conn_super(sb);
1082 
1083 	if (fc) {
1084 		down_write(&fc->killsb);
1085 		fc->sb = NULL;
1086 		up_write(&fc->killsb);
1087 	}
1088 
1089 	kill_block_super(sb);
1090 }
1091 
1092 static struct file_system_type fuseblk_fs_type = {
1093 	.owner		= THIS_MODULE,
1094 	.name		= "fuseblk",
1095 	.get_sb		= fuse_get_sb_blk,
1096 	.kill_sb	= fuse_kill_sb_blk,
1097 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1098 };
1099 
1100 static inline int register_fuseblk(void)
1101 {
1102 	return register_filesystem(&fuseblk_fs_type);
1103 }
1104 
1105 static inline void unregister_fuseblk(void)
1106 {
1107 	unregister_filesystem(&fuseblk_fs_type);
1108 }
1109 #else
1110 static inline int register_fuseblk(void)
1111 {
1112 	return 0;
1113 }
1114 
1115 static inline void unregister_fuseblk(void)
1116 {
1117 }
1118 #endif
1119 
1120 static void fuse_inode_init_once(void *foo)
1121 {
1122 	struct inode *inode = foo;
1123 
1124 	inode_init_once(inode);
1125 }
1126 
1127 static int __init fuse_fs_init(void)
1128 {
1129 	int err;
1130 
1131 	err = register_filesystem(&fuse_fs_type);
1132 	if (err)
1133 		goto out;
1134 
1135 	err = register_fuseblk();
1136 	if (err)
1137 		goto out_unreg;
1138 
1139 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
1140 					      sizeof(struct fuse_inode),
1141 					      0, SLAB_HWCACHE_ALIGN,
1142 					      fuse_inode_init_once);
1143 	err = -ENOMEM;
1144 	if (!fuse_inode_cachep)
1145 		goto out_unreg2;
1146 
1147 	return 0;
1148 
1149  out_unreg2:
1150 	unregister_fuseblk();
1151  out_unreg:
1152 	unregister_filesystem(&fuse_fs_type);
1153  out:
1154 	return err;
1155 }
1156 
1157 static void fuse_fs_cleanup(void)
1158 {
1159 	unregister_filesystem(&fuse_fs_type);
1160 	unregister_fuseblk();
1161 	kmem_cache_destroy(fuse_inode_cachep);
1162 }
1163 
1164 static struct kobject *fuse_kobj;
1165 static struct kobject *connections_kobj;
1166 
1167 static int fuse_sysfs_init(void)
1168 {
1169 	int err;
1170 
1171 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1172 	if (!fuse_kobj) {
1173 		err = -ENOMEM;
1174 		goto out_err;
1175 	}
1176 
1177 	connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1178 	if (!connections_kobj) {
1179 		err = -ENOMEM;
1180 		goto out_fuse_unregister;
1181 	}
1182 
1183 	return 0;
1184 
1185  out_fuse_unregister:
1186 	kobject_put(fuse_kobj);
1187  out_err:
1188 	return err;
1189 }
1190 
1191 static void fuse_sysfs_cleanup(void)
1192 {
1193 	kobject_put(connections_kobj);
1194 	kobject_put(fuse_kobj);
1195 }
1196 
1197 static int __init fuse_init(void)
1198 {
1199 	int res;
1200 
1201 	printk(KERN_INFO "fuse init (API version %i.%i)\n",
1202 	       FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1203 
1204 	INIT_LIST_HEAD(&fuse_conn_list);
1205 	res = fuse_fs_init();
1206 	if (res)
1207 		goto err;
1208 
1209 	res = fuse_dev_init();
1210 	if (res)
1211 		goto err_fs_cleanup;
1212 
1213 	res = fuse_sysfs_init();
1214 	if (res)
1215 		goto err_dev_cleanup;
1216 
1217 	res = fuse_ctl_init();
1218 	if (res)
1219 		goto err_sysfs_cleanup;
1220 
1221 	sanitize_global_limit(&max_user_bgreq);
1222 	sanitize_global_limit(&max_user_congthresh);
1223 
1224 	return 0;
1225 
1226  err_sysfs_cleanup:
1227 	fuse_sysfs_cleanup();
1228  err_dev_cleanup:
1229 	fuse_dev_cleanup();
1230  err_fs_cleanup:
1231 	fuse_fs_cleanup();
1232  err:
1233 	return res;
1234 }
1235 
1236 static void __exit fuse_exit(void)
1237 {
1238 	printk(KERN_DEBUG "fuse exit\n");
1239 
1240 	fuse_ctl_cleanup();
1241 	fuse_sysfs_cleanup();
1242 	fuse_fs_cleanup();
1243 	fuse_dev_cleanup();
1244 }
1245 
1246 module_init(fuse_init);
1247 module_exit(fuse_exit);
1248