• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/fs_context.h>
19 #include <linux/fs_parser.h>
20 #include <linux/statfs.h>
21 #include <linux/random.h>
22 #include <linux/sched.h>
23 #include <linux/exportfs.h>
24 #include <linux/posix_acl.h>
25 #include <linux/pid_namespace.h>
26 
27 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
28 MODULE_DESCRIPTION("Filesystem in Userspace");
29 MODULE_LICENSE("GPL");
30 MODULE_IMPORT_NS(ANDROID_GKI_VFS_EXPORT_ONLY);
31 
32 static struct kmem_cache *fuse_inode_cachep;
33 struct list_head fuse_conn_list;
34 DEFINE_MUTEX(fuse_mutex);
35 
36 static int set_global_limit(const char *val, const struct kernel_param *kp);
37 
38 unsigned max_user_bgreq;
39 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
40 		  &max_user_bgreq, 0644);
41 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
42 MODULE_PARM_DESC(max_user_bgreq,
43  "Global limit for the maximum number of backgrounded requests an "
44  "unprivileged user can set");
45 
46 unsigned max_user_congthresh;
47 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
48 		  &max_user_congthresh, 0644);
49 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
50 MODULE_PARM_DESC(max_user_congthresh,
51  "Global limit for the maximum congestion threshold an "
52  "unprivileged user can set");
53 
54 #define FUSE_DEFAULT_BLKSIZE 512
55 
56 /** Maximum number of outstanding background requests */
57 #define FUSE_DEFAULT_MAX_BACKGROUND 12
58 
59 /** Congestion starts at 75% of maximum */
60 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
61 
62 #ifdef CONFIG_BLOCK
63 static struct file_system_type fuseblk_fs_type;
64 #endif
65 
fuse_alloc_forget(void)66 struct fuse_forget_link *fuse_alloc_forget(void)
67 {
68 	return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
69 }
70 
fuse_alloc_inode(struct super_block * sb)71 static struct inode *fuse_alloc_inode(struct super_block *sb)
72 {
73 	struct fuse_inode *fi;
74 
75 	fi = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
76 	if (!fi)
77 		return NULL;
78 
79 	fi->i_time = 0;
80 	fi->inval_mask = 0;
81 #ifdef CONFIG_FUSE_BPF
82 	fi->backing_inode = NULL;
83 	fi->bpf = NULL;
84 #endif
85 	fi->nodeid = 0;
86 	fi->nlookup = 0;
87 	fi->attr_version = 0;
88 	fi->orig_ino = 0;
89 	fi->state = 0;
90 	mutex_init(&fi->mutex);
91 	init_rwsem(&fi->i_mmap_sem);
92 	spin_lock_init(&fi->lock);
93 	fi->forget = fuse_alloc_forget();
94 	if (!fi->forget)
95 		goto out_free;
96 
97 	if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
98 		goto out_free_forget;
99 
100 	return &fi->inode;
101 
102 out_free_forget:
103 	kfree(fi->forget);
104 out_free:
105 	kmem_cache_free(fuse_inode_cachep, fi);
106 	return NULL;
107 }
108 
fuse_free_inode(struct inode * inode)109 static void fuse_free_inode(struct inode *inode)
110 {
111 	struct fuse_inode *fi = get_fuse_inode(inode);
112 
113 	mutex_destroy(&fi->mutex);
114 	kfree(fi->forget);
115 #ifdef CONFIG_FUSE_DAX
116 	kfree(fi->dax);
117 #endif
118 #ifdef CONFIG_FUSE_BPF
119 	if (fi->bpf)
120 		bpf_prog_put(fi->bpf);
121 #endif
122 	kmem_cache_free(fuse_inode_cachep, fi);
123 }
124 
fuse_evict_inode(struct inode * inode)125 static void fuse_evict_inode(struct inode *inode)
126 {
127 	struct fuse_inode *fi = get_fuse_inode(inode);
128 
129 	/* Will write inode on close/munmap and in all other dirtiers */
130 	WARN_ON(inode->i_state & I_DIRTY_INODE);
131 
132 	truncate_inode_pages_final(&inode->i_data);
133 	clear_inode(inode);
134 	if (inode->i_sb->s_flags & SB_ACTIVE) {
135 		struct fuse_conn *fc = get_fuse_conn(inode);
136 
137 		if (FUSE_IS_DAX(inode))
138 			fuse_dax_inode_cleanup(inode);
139 		if (fi->nlookup) {
140 			fuse_queue_forget(fc, fi->forget, fi->nodeid,
141 					  fi->nlookup);
142 			fi->forget = NULL;
143 		}
144 	}
145 	if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
146 		WARN_ON(!list_empty(&fi->write_files));
147 		WARN_ON(!list_empty(&fi->queued_writes));
148 	}
149 }
150 
151 #ifdef CONFIG_FUSE_BPF
fuse_destroy_inode(struct inode * inode)152 static void fuse_destroy_inode(struct inode *inode)
153 {
154 	struct fuse_inode *fi = get_fuse_inode(inode);
155 
156 	iput(fi->backing_inode);
157 }
158 #endif
159 
fuse_reconfigure(struct fs_context * fc)160 static int fuse_reconfigure(struct fs_context *fc)
161 {
162 	struct super_block *sb = fc->root->d_sb;
163 
164 	sync_filesystem(sb);
165 	if (fc->sb_flags & SB_MANDLOCK)
166 		return -EINVAL;
167 
168 	return 0;
169 }
170 
171 /*
172  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
173  * so that it will fit.
174  */
fuse_squash_ino(u64 ino64)175 static ino_t fuse_squash_ino(u64 ino64)
176 {
177 	ino_t ino = (ino_t) ino64;
178 	if (sizeof(ino_t) < sizeof(u64))
179 		ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
180 	return ino;
181 }
182 
fuse_fill_attr_from_inode(struct fuse_attr * attr,const struct inode * inode)183 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
184 				      const struct inode *inode)
185 {
186 	*attr = (struct fuse_attr){
187 		.ino		= inode->i_ino,
188 		.size		= inode->i_size,
189 		.blocks		= inode->i_blocks,
190 		.atime		= inode->i_atime.tv_sec,
191 		.mtime		= inode->i_mtime.tv_sec,
192 		.ctime		= inode->i_ctime.tv_sec,
193 		.atimensec	= inode->i_atime.tv_nsec,
194 		.mtimensec	= inode->i_mtime.tv_nsec,
195 		.ctimensec	= inode->i_ctime.tv_nsec,
196 		.mode		= inode->i_mode,
197 		.nlink		= inode->i_nlink,
198 		.uid		= inode->i_uid.val,
199 		.gid		= inode->i_gid.val,
200 		.rdev		= inode->i_rdev,
201 		.blksize	= 1u << inode->i_blkbits,
202 	};
203 }
204 
fuse_change_attributes_common(struct inode * inode,struct fuse_attr * attr,u64 attr_valid)205 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
206 				   u64 attr_valid)
207 {
208 	struct fuse_conn *fc = get_fuse_conn(inode);
209 	struct fuse_inode *fi = get_fuse_inode(inode);
210 
211 	lockdep_assert_held(&fi->lock);
212 
213 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
214 	fi->i_time = attr_valid;
215 	WRITE_ONCE(fi->inval_mask, 0);
216 
217 	inode->i_ino     = fuse_squash_ino(attr->ino);
218 	inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
219 	set_nlink(inode, attr->nlink);
220 	inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
221 	inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
222 	inode->i_blocks  = attr->blocks;
223 
224 	/* Sanitize nsecs */
225 	attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
226 	attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
227 	attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
228 
229 	inode->i_atime.tv_sec   = attr->atime;
230 	inode->i_atime.tv_nsec  = attr->atimensec;
231 	/* mtime from server may be stale due to local buffered write */
232 	if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
233 		inode->i_mtime.tv_sec   = attr->mtime;
234 		inode->i_mtime.tv_nsec  = attr->mtimensec;
235 		inode->i_ctime.tv_sec   = attr->ctime;
236 		inode->i_ctime.tv_nsec  = attr->ctimensec;
237 	}
238 
239 	if (attr->blksize != 0)
240 		inode->i_blkbits = ilog2(attr->blksize);
241 	else
242 		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
243 
244 	/*
245 	 * Don't set the sticky bit in i_mode, unless we want the VFS
246 	 * to check permissions.  This prevents failures due to the
247 	 * check in may_delete().
248 	 */
249 	fi->orig_i_mode = inode->i_mode;
250 	if (!fc->default_permissions)
251 		inode->i_mode &= ~S_ISVTX;
252 
253 	fi->orig_ino = attr->ino;
254 }
255 
fuse_change_attributes(struct inode * inode,struct fuse_attr * attr,u64 attr_valid,u64 attr_version)256 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
257 			    u64 attr_valid, u64 attr_version)
258 {
259 	struct fuse_conn *fc = get_fuse_conn(inode);
260 	struct fuse_inode *fi = get_fuse_inode(inode);
261 	bool is_wb = fc->writeback_cache;
262 	loff_t oldsize;
263 	struct timespec64 old_mtime;
264 
265 	spin_lock(&fi->lock);
266 	if ((attr_version != 0 && fi->attr_version > attr_version) ||
267 	    test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
268 		spin_unlock(&fi->lock);
269 		return;
270 	}
271 
272 	old_mtime = inode->i_mtime;
273 	fuse_change_attributes_common(inode, attr, attr_valid);
274 
275 	oldsize = inode->i_size;
276 	/*
277 	 * In case of writeback_cache enabled, the cached writes beyond EOF
278 	 * extend local i_size without keeping userspace server in sync. So,
279 	 * attr->size coming from server can be stale. We cannot trust it.
280 	 */
281 	if (!is_wb || !S_ISREG(inode->i_mode))
282 		i_size_write(inode, attr->size);
283 	spin_unlock(&fi->lock);
284 
285 	if (!is_wb && S_ISREG(inode->i_mode)) {
286 		bool inval = false;
287 
288 		if (oldsize != attr->size) {
289 			truncate_pagecache(inode, attr->size);
290 			if (!fc->explicit_inval_data)
291 				inval = true;
292 		} else if (fc->auto_inval_data) {
293 			struct timespec64 new_mtime = {
294 				.tv_sec = attr->mtime,
295 				.tv_nsec = attr->mtimensec,
296 			};
297 
298 			/*
299 			 * Auto inval mode also checks and invalidates if mtime
300 			 * has changed.
301 			 */
302 			if (!timespec64_equal(&old_mtime, &new_mtime))
303 				inval = true;
304 		}
305 
306 		if (inval)
307 			invalidate_inode_pages2(inode->i_mapping);
308 	}
309 }
310 
fuse_init_inode(struct inode * inode,struct fuse_attr * attr)311 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
312 {
313 	inode->i_mode = attr->mode & S_IFMT;
314 	inode->i_size = attr->size;
315 	inode->i_mtime.tv_sec  = attr->mtime;
316 	inode->i_mtime.tv_nsec = attr->mtimensec;
317 	inode->i_ctime.tv_sec  = attr->ctime;
318 	inode->i_ctime.tv_nsec = attr->ctimensec;
319 	if (S_ISREG(inode->i_mode)) {
320 		fuse_init_common(inode);
321 		fuse_init_file_inode(inode);
322 	} else if (S_ISDIR(inode->i_mode))
323 		fuse_init_dir(inode);
324 	else if (S_ISLNK(inode->i_mode))
325 		fuse_init_symlink(inode);
326 	else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
327 		 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
328 		fuse_init_common(inode);
329 		init_special_inode(inode, inode->i_mode, attr->rdev);
330 	} else
331 		BUG();
332 }
333 
334 struct fuse_inode_identifier {
335 	u64 nodeid;
336 	struct inode *backing_inode;
337 };
338 
fuse_inode_eq(struct inode * inode,void * _nodeidp)339 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
340 {
341 	struct fuse_inode_identifier *fii =
342 		(struct fuse_inode_identifier *) _nodeidp;
343 	struct fuse_inode *fi = get_fuse_inode(inode);
344 
345 	return fii->nodeid == fi->nodeid;
346 }
347 
fuse_inode_backing_eq(struct inode * inode,void * _nodeidp)348 static int fuse_inode_backing_eq(struct inode *inode, void *_nodeidp)
349 {
350 	struct fuse_inode_identifier *fii =
351 		(struct fuse_inode_identifier *) _nodeidp;
352 	struct fuse_inode *fi = get_fuse_inode(inode);
353 
354 	return fii->nodeid == fi->nodeid
355 #ifdef CONFIG_FUSE_BPF
356 		&& fii->backing_inode == fi->backing_inode
357 #endif
358 		;
359 }
360 
fuse_inode_set(struct inode * inode,void * _nodeidp)361 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
362 {
363 	struct fuse_inode_identifier *fii =
364 		(struct fuse_inode_identifier *) _nodeidp;
365 	struct fuse_inode *fi = get_fuse_inode(inode);
366 
367 	fi->nodeid = fii->nodeid;
368 
369 	return 0;
370 }
371 
fuse_inode_backing_set(struct inode * inode,void * _nodeidp)372 static int fuse_inode_backing_set(struct inode *inode, void *_nodeidp)
373 {
374 	struct fuse_inode_identifier *fii =
375 		(struct fuse_inode_identifier *) _nodeidp;
376 	struct fuse_inode *fi = get_fuse_inode(inode);
377 
378 	fi->nodeid = fii->nodeid;
379 #ifdef CONFIG_FUSE_BPF
380 	fi->backing_inode = fii->backing_inode;
381 	if (fi->backing_inode)
382 		ihold(fi->backing_inode);
383 #endif
384 
385 	return 0;
386 }
387 
fuse_iget_backing(struct super_block * sb,u64 nodeid,struct inode * backing_inode)388 struct inode *fuse_iget_backing(struct super_block *sb, u64 nodeid,
389 				struct inode *backing_inode)
390 {
391 	struct inode *inode;
392 	struct fuse_inode *fi;
393 	struct fuse_conn *fc = get_fuse_conn_super(sb);
394 	struct fuse_inode_identifier fii = {
395 		.nodeid = nodeid,
396 		.backing_inode = backing_inode,
397 	};
398 	struct fuse_attr attr;
399 	unsigned long hash = (unsigned long) backing_inode;
400 
401 	if (nodeid)
402 		hash = nodeid;
403 
404 	fuse_fill_attr_from_inode(&attr, backing_inode);
405 	inode = iget5_locked(sb, hash, fuse_inode_backing_eq,
406 			     fuse_inode_backing_set, &fii);
407 	if (!inode)
408 		return NULL;
409 
410 	if ((inode->i_state & I_NEW)) {
411 		inode->i_flags |= S_NOATIME;
412 		if (!fc->writeback_cache)
413 			inode->i_flags |= S_NOCMTIME;
414 		fuse_init_common(inode);
415 		unlock_new_inode(inode);
416 	}
417 
418 	fi = get_fuse_inode(inode);
419 	fuse_init_inode(inode, &attr);
420 	spin_lock(&fi->lock);
421 	fi->nlookup++;
422 	spin_unlock(&fi->lock);
423 
424 	return inode;
425 }
426 
fuse_iget(struct super_block * sb,u64 nodeid,int generation,struct fuse_attr * attr,u64 attr_valid,u64 attr_version)427 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
428 			int generation, struct fuse_attr *attr,
429 			u64 attr_valid, u64 attr_version)
430 {
431 	struct inode *inode;
432 	struct fuse_inode *fi;
433 	struct fuse_conn *fc = get_fuse_conn_super(sb);
434 	struct fuse_inode_identifier fii = {
435 		.nodeid = nodeid,
436 	};
437 
438 	/*
439 	 * Auto mount points get their node id from the submount root, which is
440 	 * not a unique identifier within this filesystem.
441 	 *
442 	 * To avoid conflicts, do not place submount points into the inode hash
443 	 * table.
444 	 */
445 	if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
446 	    S_ISDIR(attr->mode)) {
447 		inode = new_inode(sb);
448 		if (!inode)
449 			return NULL;
450 
451 		fuse_init_inode(inode, attr);
452 		get_fuse_inode(inode)->nodeid = nodeid;
453 		inode->i_flags |= S_AUTOMOUNT;
454 		goto done;
455 	}
456 
457 retry:
458 	inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &fii);
459 	if (!inode)
460 		return NULL;
461 
462 	if ((inode->i_state & I_NEW)) {
463 		inode->i_flags |= S_NOATIME;
464 		if (!fc->writeback_cache || !S_ISREG(attr->mode))
465 			inode->i_flags |= S_NOCMTIME;
466 		inode->i_generation = generation;
467 		fuse_init_inode(inode, attr);
468 		unlock_new_inode(inode);
469 	} else if (fuse_stale_inode(inode, generation, attr)) {
470 		/* nodeid was reused, any I/O on the old inode should fail */
471 		fuse_make_bad(inode);
472 		iput(inode);
473 		goto retry;
474 	}
475 done:
476 	fi = get_fuse_inode(inode);
477 	spin_lock(&fi->lock);
478 	fi->nlookup++;
479 	spin_unlock(&fi->lock);
480 	fuse_change_attributes(inode, attr, attr_valid, attr_version);
481 
482 	return inode;
483 }
484 
fuse_ilookup(struct fuse_conn * fc,u64 nodeid,struct fuse_mount ** fm)485 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
486 			   struct fuse_mount **fm)
487 {
488 	struct fuse_mount *fm_iter;
489 	struct inode *inode;
490 	struct fuse_inode_identifier fii = {
491 		.nodeid = nodeid,
492 	};
493 
494 	WARN_ON(!rwsem_is_locked(&fc->killsb));
495 	list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
496 		if (!fm_iter->sb)
497 			continue;
498 
499 		inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &fii);
500 		if (inode) {
501 			if (fm)
502 				*fm = fm_iter;
503 			return inode;
504 		}
505 	}
506 
507 	return NULL;
508 }
509 
fuse_reverse_inval_inode(struct fuse_conn * fc,u64 nodeid,loff_t offset,loff_t len)510 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
511 			     loff_t offset, loff_t len)
512 {
513 	struct fuse_inode *fi;
514 	struct inode *inode;
515 	pgoff_t pg_start;
516 	pgoff_t pg_end;
517 
518 	inode = fuse_ilookup(fc, nodeid, NULL);
519 	if (!inode)
520 		return -ENOENT;
521 
522 	fi = get_fuse_inode(inode);
523 	spin_lock(&fi->lock);
524 	fi->attr_version = atomic64_inc_return(&fc->attr_version);
525 	spin_unlock(&fi->lock);
526 
527 	fuse_invalidate_attr(inode);
528 	forget_all_cached_acls(inode);
529 	if (offset >= 0) {
530 		pg_start = offset >> PAGE_SHIFT;
531 		if (len <= 0)
532 			pg_end = -1;
533 		else
534 			pg_end = (offset + len - 1) >> PAGE_SHIFT;
535 		invalidate_inode_pages2_range(inode->i_mapping,
536 					      pg_start, pg_end);
537 	}
538 	iput(inode);
539 	return 0;
540 }
541 
fuse_lock_inode(struct inode * inode)542 bool fuse_lock_inode(struct inode *inode)
543 {
544 	bool locked = false;
545 
546 	if (!get_fuse_conn(inode)->parallel_dirops) {
547 		mutex_lock(&get_fuse_inode(inode)->mutex);
548 		locked = true;
549 	}
550 
551 	return locked;
552 }
553 
fuse_unlock_inode(struct inode * inode,bool locked)554 void fuse_unlock_inode(struct inode *inode, bool locked)
555 {
556 	if (locked)
557 		mutex_unlock(&get_fuse_inode(inode)->mutex);
558 }
559 
fuse_umount_begin(struct super_block * sb)560 static void fuse_umount_begin(struct super_block *sb)
561 {
562 	struct fuse_conn *fc = get_fuse_conn_super(sb);
563 
564 	if (!fc->no_force_umount)
565 		fuse_abort_conn(fc);
566 }
567 
fuse_send_destroy(struct fuse_mount * fm)568 static void fuse_send_destroy(struct fuse_mount *fm)
569 {
570 	if (fm->fc->conn_init) {
571 		FUSE_ARGS(args);
572 
573 		args.opcode = FUSE_DESTROY;
574 		args.force = true;
575 		args.nocreds = true;
576 		fuse_simple_request(fm, &args);
577 	}
578 }
579 
fuse_put_super(struct super_block * sb)580 static void fuse_put_super(struct super_block *sb)
581 {
582 	struct fuse_mount *fm = get_fuse_mount_super(sb);
583 
584 	fuse_mount_put(fm);
585 }
586 
fuse_statfs(struct dentry * dentry,struct kstatfs * buf)587 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
588 {
589 	struct super_block *sb = dentry->d_sb;
590 	struct fuse_mount *fm = get_fuse_mount_super(sb);
591 	FUSE_ARGS(args);
592 	struct fuse_statfs_out outarg;
593 	int err;
594 #ifdef CONFIG_FUSE_BPF
595 	struct fuse_err_ret fer;
596 #endif
597 
598 	if (!fuse_allow_current_process(fm->fc)) {
599 		buf->f_type = FUSE_SUPER_MAGIC;
600 		return 0;
601 	}
602 
603 #ifdef CONFIG_FUSE_BPF
604 	fer = fuse_bpf_backing(dentry->d_inode, struct fuse_statfs_out,
605 			       fuse_statfs_initialize, fuse_statfs_backing,
606 			       fuse_statfs_finalize,
607 			       dentry, buf);
608 	if (fer.ret)
609 		return PTR_ERR(fer.result);
610 #endif
611 
612 	memset(&outarg, 0, sizeof(outarg));
613 	args.in_numargs = 0;
614 	args.opcode = FUSE_STATFS;
615 	args.nodeid = get_node_id(d_inode(dentry));
616 	args.out_numargs = 1;
617 	args.out_args[0].size = sizeof(outarg);
618 	args.out_args[0].value = &outarg;
619 	err = fuse_simple_request(fm, &args);
620 	if (!err)
621 		convert_fuse_statfs(buf, &outarg.st);
622 	return err;
623 }
624 
625 enum {
626 	OPT_SOURCE,
627 	OPT_SUBTYPE,
628 	OPT_FD,
629 	OPT_ROOTMODE,
630 	OPT_USER_ID,
631 	OPT_GROUP_ID,
632 	OPT_DEFAULT_PERMISSIONS,
633 	OPT_ALLOW_OTHER,
634 	OPT_MAX_READ,
635 	OPT_BLKSIZE,
636 	OPT_ROOT_BPF,
637 	OPT_ROOT_DIR,
638 	OPT_NO_DAEMON,
639 	OPT_ERR
640 };
641 
642 static const struct fs_parameter_spec fuse_fs_parameters[] = {
643 	fsparam_string	("source",		OPT_SOURCE),
644 	fsparam_u32	("fd",			OPT_FD),
645 	fsparam_u32oct	("rootmode",		OPT_ROOTMODE),
646 	fsparam_u32	("user_id",		OPT_USER_ID),
647 	fsparam_u32	("group_id",		OPT_GROUP_ID),
648 	fsparam_flag	("default_permissions",	OPT_DEFAULT_PERMISSIONS),
649 	fsparam_flag	("allow_other",		OPT_ALLOW_OTHER),
650 	fsparam_u32	("max_read",		OPT_MAX_READ),
651 	fsparam_u32	("blksize",		OPT_BLKSIZE),
652 	fsparam_string	("subtype",		OPT_SUBTYPE),
653 	fsparam_u32	("root_bpf",		OPT_ROOT_BPF),
654 	fsparam_u32	("root_dir",		OPT_ROOT_DIR),
655 	fsparam_flag	("no_daemon",		OPT_NO_DAEMON),
656 	{}
657 };
658 
fuse_parse_param(struct fs_context * fc,struct fs_parameter * param)659 static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param)
660 {
661 	struct fs_parse_result result;
662 	struct fuse_fs_context *ctx = fc->fs_private;
663 	int opt;
664 
665 	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
666 		/*
667 		 * Ignore options coming from mount(MS_REMOUNT) for backward
668 		 * compatibility.
669 		 */
670 		if (fc->oldapi)
671 			return 0;
672 
673 		return invalfc(fc, "No changes allowed in reconfigure");
674 	}
675 
676 	opt = fs_parse(fc, fuse_fs_parameters, param, &result);
677 	if (opt < 0)
678 		return opt;
679 
680 	switch (opt) {
681 	case OPT_SOURCE:
682 		if (fc->source)
683 			return invalfc(fc, "Multiple sources specified");
684 		fc->source = param->string;
685 		param->string = NULL;
686 		break;
687 
688 	case OPT_SUBTYPE:
689 		if (ctx->subtype)
690 			return invalfc(fc, "Multiple subtypes specified");
691 		ctx->subtype = param->string;
692 		param->string = NULL;
693 		return 0;
694 
695 	case OPT_FD:
696 		ctx->fd = result.uint_32;
697 		ctx->fd_present = true;
698 		break;
699 
700 	case OPT_ROOTMODE:
701 		if (!fuse_valid_type(result.uint_32))
702 			return invalfc(fc, "Invalid rootmode");
703 		ctx->rootmode = result.uint_32;
704 		ctx->rootmode_present = true;
705 		break;
706 
707 	case OPT_USER_ID:
708 		ctx->user_id = make_kuid(fc->user_ns, result.uint_32);
709 		if (!uid_valid(ctx->user_id))
710 			return invalfc(fc, "Invalid user_id");
711 		ctx->user_id_present = true;
712 		break;
713 
714 	case OPT_GROUP_ID:
715 		ctx->group_id = make_kgid(fc->user_ns, result.uint_32);
716 		if (!gid_valid(ctx->group_id))
717 			return invalfc(fc, "Invalid group_id");
718 		ctx->group_id_present = true;
719 		break;
720 
721 	case OPT_DEFAULT_PERMISSIONS:
722 		ctx->default_permissions = true;
723 		break;
724 
725 	case OPT_ALLOW_OTHER:
726 		ctx->allow_other = true;
727 		break;
728 
729 	case OPT_MAX_READ:
730 		ctx->max_read = result.uint_32;
731 		break;
732 
733 	case OPT_BLKSIZE:
734 		if (!ctx->is_bdev)
735 			return invalfc(fc, "blksize only supported for fuseblk");
736 		ctx->blksize = result.uint_32;
737 		break;
738 
739 	case OPT_ROOT_BPF:
740 		ctx->root_bpf = bpf_prog_get_type_dev(result.uint_32,
741 						BPF_PROG_TYPE_FUSE, false);
742 		if (IS_ERR(ctx->root_bpf)) {
743 			ctx->root_bpf = NULL;
744 			return invalfc(fc, "Unable to open bpf program");
745 		}
746 		break;
747 
748 	case OPT_ROOT_DIR:
749 		ctx->root_dir = fget(result.uint_32);
750 		if (!ctx->root_dir)
751 			return invalfc(fc, "Unable to open root directory");
752 		break;
753 
754 	case OPT_NO_DAEMON:
755 		ctx->no_daemon = true;
756 		ctx->fd_present = true;
757 		break;
758 
759 	default:
760 		return -EINVAL;
761 	}
762 
763 	return 0;
764 }
765 
fuse_free_fc(struct fs_context * fc)766 static void fuse_free_fc(struct fs_context *fc)
767 {
768 	struct fuse_fs_context *ctx = fc->fs_private;
769 
770 	if (ctx) {
771 		if (ctx->root_dir)
772 			fput(ctx->root_dir);
773 		if (ctx->root_bpf)
774 			bpf_prog_put(ctx->root_bpf);
775 		kfree(ctx->subtype);
776 		kfree(ctx);
777 	}
778 }
779 
fuse_show_options(struct seq_file * m,struct dentry * root)780 static int fuse_show_options(struct seq_file *m, struct dentry *root)
781 {
782 	struct super_block *sb = root->d_sb;
783 	struct fuse_conn *fc = get_fuse_conn_super(sb);
784 
785 	if (fc->legacy_opts_show) {
786 		seq_printf(m, ",user_id=%u",
787 			   from_kuid_munged(fc->user_ns, fc->user_id));
788 		seq_printf(m, ",group_id=%u",
789 			   from_kgid_munged(fc->user_ns, fc->group_id));
790 		if (fc->default_permissions)
791 			seq_puts(m, ",default_permissions");
792 		if (fc->allow_other)
793 			seq_puts(m, ",allow_other");
794 		if (fc->max_read != ~0)
795 			seq_printf(m, ",max_read=%u", fc->max_read);
796 		if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
797 			seq_printf(m, ",blksize=%lu", sb->s_blocksize);
798 	}
799 #ifdef CONFIG_FUSE_DAX
800 	if (fc->dax)
801 		seq_puts(m, ",dax");
802 #endif
803 
804 	return 0;
805 }
806 
fuse_iqueue_init(struct fuse_iqueue * fiq,const struct fuse_iqueue_ops * ops,void * priv)807 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
808 			     const struct fuse_iqueue_ops *ops,
809 			     void *priv)
810 {
811 	memset(fiq, 0, sizeof(struct fuse_iqueue));
812 	spin_lock_init(&fiq->lock);
813 	init_waitqueue_head(&fiq->waitq);
814 	INIT_LIST_HEAD(&fiq->pending);
815 	INIT_LIST_HEAD(&fiq->interrupts);
816 	fiq->forget_list_tail = &fiq->forget_list_head;
817 	fiq->connected = 1;
818 	fiq->ops = ops;
819 	fiq->priv = priv;
820 }
821 
fuse_pqueue_init(struct fuse_pqueue * fpq)822 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
823 {
824 	unsigned int i;
825 
826 	spin_lock_init(&fpq->lock);
827 	for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
828 		INIT_LIST_HEAD(&fpq->processing[i]);
829 	INIT_LIST_HEAD(&fpq->io);
830 	fpq->connected = 1;
831 }
832 
fuse_conn_init(struct fuse_conn * fc,struct fuse_mount * fm,struct user_namespace * user_ns,const struct fuse_iqueue_ops * fiq_ops,void * fiq_priv)833 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
834 		    struct user_namespace *user_ns,
835 		    const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
836 {
837 	memset(fc, 0, sizeof(*fc));
838 	spin_lock_init(&fc->lock);
839 	spin_lock_init(&fc->bg_lock);
840 	spin_lock_init(&fc->passthrough_req_lock);
841 	init_rwsem(&fc->killsb);
842 	refcount_set(&fc->count, 1);
843 	atomic_set(&fc->dev_count, 1);
844 	init_waitqueue_head(&fc->blocked_waitq);
845 	fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
846 	INIT_LIST_HEAD(&fc->bg_queue);
847 	INIT_LIST_HEAD(&fc->entry);
848 	INIT_LIST_HEAD(&fc->devices);
849 	idr_init(&fc->passthrough_req);
850 	atomic_set(&fc->num_waiting, 0);
851 	fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
852 	fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
853 	atomic64_set(&fc->khctr, 0);
854 	fc->polled_files = RB_ROOT;
855 	fc->blocked = 0;
856 	fc->initialized = 0;
857 	fc->connected = 1;
858 	atomic64_set(&fc->attr_version, 1);
859 	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
860 	fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
861 	fc->user_ns = get_user_ns(user_ns);
862 	fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
863 	fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
864 
865 	INIT_LIST_HEAD(&fc->mounts);
866 	list_add(&fm->fc_entry, &fc->mounts);
867 	fm->fc = fc;
868 	refcount_set(&fm->count, 1);
869 }
870 EXPORT_SYMBOL_GPL(fuse_conn_init);
871 
fuse_conn_put(struct fuse_conn * fc)872 void fuse_conn_put(struct fuse_conn *fc)
873 {
874 	if (refcount_dec_and_test(&fc->count)) {
875 		struct fuse_iqueue *fiq = &fc->iq;
876 
877 		if (IS_ENABLED(CONFIG_FUSE_DAX))
878 			fuse_dax_conn_free(fc);
879 		if (fiq->ops->release)
880 			fiq->ops->release(fiq);
881 		put_pid_ns(fc->pid_ns);
882 		put_user_ns(fc->user_ns);
883 		fc->release(fc);
884 	}
885 }
886 EXPORT_SYMBOL_GPL(fuse_conn_put);
887 
fuse_conn_get(struct fuse_conn * fc)888 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
889 {
890 	refcount_inc(&fc->count);
891 	return fc;
892 }
893 EXPORT_SYMBOL_GPL(fuse_conn_get);
894 
fuse_mount_put(struct fuse_mount * fm)895 void fuse_mount_put(struct fuse_mount *fm)
896 {
897 	if (refcount_dec_and_test(&fm->count)) {
898 		if (fm->fc)
899 			fuse_conn_put(fm->fc);
900 		kfree(fm);
901 	}
902 }
903 EXPORT_SYMBOL_GPL(fuse_mount_put);
904 
fuse_mount_get(struct fuse_mount * fm)905 struct fuse_mount *fuse_mount_get(struct fuse_mount *fm)
906 {
907 	refcount_inc(&fm->count);
908 	return fm;
909 }
910 EXPORT_SYMBOL_GPL(fuse_mount_get);
911 
fuse_get_root_inode(struct super_block * sb,unsigned int mode,struct bpf_prog * root_bpf,struct file * backing_fd)912 static struct inode *fuse_get_root_inode(struct super_block *sb,
913 					 unsigned int mode,
914 					 struct bpf_prog *root_bpf,
915 					 struct file *backing_fd)
916 {
917 	struct fuse_attr attr;
918 	struct inode *inode;
919 
920 	memset(&attr, 0, sizeof(attr));
921 	attr.mode = mode;
922 	attr.ino = FUSE_ROOT_ID;
923 	attr.nlink = 1;
924 	inode = fuse_iget(sb, 1, 0, &attr, 0, 0);
925 	if (!inode)
926 		return NULL;
927 
928 #ifdef CONFIG_FUSE_BPF
929 	get_fuse_inode(inode)->bpf = root_bpf;
930 	if (root_bpf)
931 		bpf_prog_inc(root_bpf);
932 
933 	if (backing_fd) {
934 		get_fuse_inode(inode)->backing_inode = backing_fd->f_inode;
935 		ihold(backing_fd->f_inode);
936 	}
937 #endif
938 
939 	return inode;
940 }
941 
942 struct fuse_inode_handle {
943 	u64 nodeid;
944 	u32 generation;
945 };
946 
fuse_get_dentry(struct super_block * sb,struct fuse_inode_handle * handle)947 static struct dentry *fuse_get_dentry(struct super_block *sb,
948 				      struct fuse_inode_handle *handle)
949 {
950 	struct fuse_conn *fc = get_fuse_conn_super(sb);
951 	struct inode *inode;
952 	struct dentry *entry;
953 	int err = -ESTALE;
954 	struct fuse_inode_identifier fii = {
955 		.nodeid = handle->nodeid,
956 	};
957 
958 	if (handle->nodeid == 0)
959 		goto out_err;
960 
961 	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &fii);
962 	if (!inode) {
963 		struct fuse_entry_out outarg;
964 		const struct qstr name = QSTR_INIT(".", 1);
965 
966 		if (!fc->export_support)
967 			goto out_err;
968 
969 		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
970 				       NULL, &inode);
971 		if (err && err != -ENOENT)
972 			goto out_err;
973 		if (err || !inode) {
974 			err = -ESTALE;
975 			goto out_err;
976 		}
977 		err = -EIO;
978 		if (get_node_id(inode) != handle->nodeid)
979 			goto out_iput;
980 	}
981 	err = -ESTALE;
982 	if (inode->i_generation != handle->generation)
983 		goto out_iput;
984 
985 	entry = d_obtain_alias(inode);
986 	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
987 		fuse_invalidate_entry_cache(entry);
988 
989 	return entry;
990 
991  out_iput:
992 	iput(inode);
993  out_err:
994 	return ERR_PTR(err);
995 }
996 
fuse_encode_fh(struct inode * inode,u32 * fh,int * max_len,struct inode * parent)997 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
998 			   struct inode *parent)
999 {
1000 	int len = parent ? 6 : 3;
1001 	u64 nodeid;
1002 	u32 generation;
1003 
1004 	if (*max_len < len) {
1005 		*max_len = len;
1006 		return  FILEID_INVALID;
1007 	}
1008 
1009 	nodeid = get_fuse_inode(inode)->nodeid;
1010 	generation = inode->i_generation;
1011 
1012 	fh[0] = (u32)(nodeid >> 32);
1013 	fh[1] = (u32)(nodeid & 0xffffffff);
1014 	fh[2] = generation;
1015 
1016 	if (parent) {
1017 		nodeid = get_fuse_inode(parent)->nodeid;
1018 		generation = parent->i_generation;
1019 
1020 		fh[3] = (u32)(nodeid >> 32);
1021 		fh[4] = (u32)(nodeid & 0xffffffff);
1022 		fh[5] = generation;
1023 	}
1024 
1025 	*max_len = len;
1026 	return parent ? 0x82 : 0x81;
1027 }
1028 
fuse_fh_to_dentry(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)1029 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
1030 		struct fid *fid, int fh_len, int fh_type)
1031 {
1032 	struct fuse_inode_handle handle;
1033 
1034 	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
1035 		return NULL;
1036 
1037 	handle.nodeid = (u64) fid->raw[0] << 32;
1038 	handle.nodeid |= (u64) fid->raw[1];
1039 	handle.generation = fid->raw[2];
1040 	return fuse_get_dentry(sb, &handle);
1041 }
1042 
fuse_fh_to_parent(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)1043 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
1044 		struct fid *fid, int fh_len, int fh_type)
1045 {
1046 	struct fuse_inode_handle parent;
1047 
1048 	if (fh_type != 0x82 || fh_len < 6)
1049 		return NULL;
1050 
1051 	parent.nodeid = (u64) fid->raw[3] << 32;
1052 	parent.nodeid |= (u64) fid->raw[4];
1053 	parent.generation = fid->raw[5];
1054 	return fuse_get_dentry(sb, &parent);
1055 }
1056 
fuse_get_parent(struct dentry * child)1057 static struct dentry *fuse_get_parent(struct dentry *child)
1058 {
1059 	struct inode *child_inode = d_inode(child);
1060 	struct fuse_conn *fc = get_fuse_conn(child_inode);
1061 	struct inode *inode;
1062 	struct dentry *parent;
1063 	struct fuse_entry_out outarg;
1064 	const struct qstr name = QSTR_INIT("..", 2);
1065 	int err;
1066 
1067 	if (!fc->export_support)
1068 		return ERR_PTR(-ESTALE);
1069 
1070 	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
1071 			       &name, &outarg, NULL, &inode);
1072 	if (err) {
1073 		if (err == -ENOENT)
1074 			return ERR_PTR(-ESTALE);
1075 		return ERR_PTR(err);
1076 	}
1077 
1078 	parent = d_obtain_alias(inode);
1079 	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
1080 		fuse_invalidate_entry_cache(parent);
1081 
1082 	return parent;
1083 }
1084 
1085 static const struct export_operations fuse_export_operations = {
1086 	.fh_to_dentry	= fuse_fh_to_dentry,
1087 	.fh_to_parent	= fuse_fh_to_parent,
1088 	.encode_fh	= fuse_encode_fh,
1089 	.get_parent	= fuse_get_parent,
1090 };
1091 
1092 static const struct super_operations fuse_super_operations = {
1093 	.alloc_inode    = fuse_alloc_inode,
1094 #ifdef CONFIG_FUSE_BPF
1095 	.destroy_inode  = fuse_destroy_inode,
1096 #endif
1097 	.free_inode     = fuse_free_inode,
1098 	.evict_inode	= fuse_evict_inode,
1099 	.write_inode	= fuse_write_inode,
1100 	.drop_inode	= generic_delete_inode,
1101 	.put_super	= fuse_put_super,
1102 	.umount_begin	= fuse_umount_begin,
1103 	.statfs		= fuse_statfs,
1104 	.show_options	= fuse_show_options,
1105 };
1106 
sanitize_global_limit(unsigned * limit)1107 static void sanitize_global_limit(unsigned *limit)
1108 {
1109 	/*
1110 	 * The default maximum number of async requests is calculated to consume
1111 	 * 1/2^13 of the total memory, assuming 392 bytes per request.
1112 	 */
1113 	if (*limit == 0)
1114 		*limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1115 
1116 	if (*limit >= 1 << 16)
1117 		*limit = (1 << 16) - 1;
1118 }
1119 
set_global_limit(const char * val,const struct kernel_param * kp)1120 static int set_global_limit(const char *val, const struct kernel_param *kp)
1121 {
1122 	int rv;
1123 
1124 	rv = param_set_uint(val, kp);
1125 	if (rv)
1126 		return rv;
1127 
1128 	sanitize_global_limit((unsigned *)kp->arg);
1129 
1130 	return 0;
1131 }
1132 
process_init_limits(struct fuse_conn * fc,struct fuse_init_out * arg)1133 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1134 {
1135 	int cap_sys_admin = capable(CAP_SYS_ADMIN);
1136 
1137 	if (arg->minor < 13)
1138 		return;
1139 
1140 	sanitize_global_limit(&max_user_bgreq);
1141 	sanitize_global_limit(&max_user_congthresh);
1142 
1143 	spin_lock(&fc->bg_lock);
1144 	if (arg->max_background) {
1145 		fc->max_background = arg->max_background;
1146 
1147 		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1148 			fc->max_background = max_user_bgreq;
1149 	}
1150 	if (arg->congestion_threshold) {
1151 		fc->congestion_threshold = arg->congestion_threshold;
1152 
1153 		if (!cap_sys_admin &&
1154 		    fc->congestion_threshold > max_user_congthresh)
1155 			fc->congestion_threshold = max_user_congthresh;
1156 	}
1157 	spin_unlock(&fc->bg_lock);
1158 }
1159 
1160 struct fuse_init_args {
1161 	struct fuse_args args;
1162 	struct fuse_init_in in;
1163 	struct fuse_init_out out;
1164 };
1165 
process_init_reply(struct fuse_mount * fm,struct fuse_args * args,int error)1166 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1167 			       int error)
1168 {
1169 	struct fuse_conn *fc = fm->fc;
1170 	struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1171 	struct fuse_init_out *arg = &ia->out;
1172 	bool ok = true;
1173 
1174 	if (error || arg->major != FUSE_KERNEL_VERSION)
1175 		ok = false;
1176 	else {
1177 		unsigned long ra_pages;
1178 
1179 		process_init_limits(fc, arg);
1180 
1181 		if (arg->minor >= 6) {
1182 			ra_pages = arg->max_readahead / PAGE_SIZE;
1183 			if (arg->flags & FUSE_ASYNC_READ)
1184 				fc->async_read = 1;
1185 			if (!(arg->flags & FUSE_POSIX_LOCKS))
1186 				fc->no_lock = 1;
1187 			if (arg->minor >= 17) {
1188 				if (!(arg->flags & FUSE_FLOCK_LOCKS))
1189 					fc->no_flock = 1;
1190 			} else {
1191 				if (!(arg->flags & FUSE_POSIX_LOCKS))
1192 					fc->no_flock = 1;
1193 			}
1194 			if (arg->flags & FUSE_ATOMIC_O_TRUNC)
1195 				fc->atomic_o_trunc = 1;
1196 			if (arg->minor >= 9) {
1197 				/* LOOKUP has dependency on proto version */
1198 				if (arg->flags & FUSE_EXPORT_SUPPORT)
1199 					fc->export_support = 1;
1200 			}
1201 			if (arg->flags & FUSE_BIG_WRITES)
1202 				fc->big_writes = 1;
1203 			if (arg->flags & FUSE_DONT_MASK)
1204 				fc->dont_mask = 1;
1205 			if (arg->flags & FUSE_AUTO_INVAL_DATA)
1206 				fc->auto_inval_data = 1;
1207 			else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
1208 				fc->explicit_inval_data = 1;
1209 			if (arg->flags & FUSE_DO_READDIRPLUS) {
1210 				fc->do_readdirplus = 1;
1211 				if (arg->flags & FUSE_READDIRPLUS_AUTO)
1212 					fc->readdirplus_auto = 1;
1213 			}
1214 			if (arg->flags & FUSE_ASYNC_DIO)
1215 				fc->async_dio = 1;
1216 			if (arg->flags & FUSE_WRITEBACK_CACHE)
1217 				fc->writeback_cache = 1;
1218 			if (arg->flags & FUSE_PARALLEL_DIROPS)
1219 				fc->parallel_dirops = 1;
1220 			if (arg->flags & FUSE_HANDLE_KILLPRIV)
1221 				fc->handle_killpriv = 1;
1222 			if (arg->time_gran && arg->time_gran <= 1000000000)
1223 				fm->sb->s_time_gran = arg->time_gran;
1224 			if ((arg->flags & FUSE_POSIX_ACL)) {
1225 				fc->default_permissions = 1;
1226 				fc->posix_acl = 1;
1227 				fm->sb->s_xattr = fuse_acl_xattr_handlers;
1228 			}
1229 			if (arg->flags & FUSE_CACHE_SYMLINKS)
1230 				fc->cache_symlinks = 1;
1231 			if (arg->flags & FUSE_ABORT_ERROR)
1232 				fc->abort_err = 1;
1233 			if (arg->flags & FUSE_MAX_PAGES) {
1234 				fc->max_pages =
1235 					min_t(unsigned int, fc->max_pages_limit,
1236 					max_t(unsigned int, arg->max_pages, 1));
1237 			}
1238 			if (IS_ENABLED(CONFIG_FUSE_DAX) &&
1239 			    arg->flags & FUSE_MAP_ALIGNMENT &&
1240 			    !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1241 				ok = false;
1242 			}
1243 			if (arg->flags & FUSE_PASSTHROUGH) {
1244 				fc->passthrough = 1;
1245 				/* Prevent further stacking */
1246 				fm->sb->s_stack_depth =
1247 					FILESYSTEM_MAX_STACK_DEPTH;
1248 			}
1249 		} else {
1250 			ra_pages = fc->max_read / PAGE_SIZE;
1251 			fc->no_lock = 1;
1252 			fc->no_flock = 1;
1253 		}
1254 
1255 		fm->sb->s_bdi->ra_pages =
1256 				min(fm->sb->s_bdi->ra_pages, ra_pages);
1257 		fc->minor = arg->minor;
1258 		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1259 		fc->max_write = max_t(unsigned, 4096, fc->max_write);
1260 		fc->conn_init = 1;
1261 	}
1262 	kfree(ia);
1263 
1264 	if (!ok) {
1265 		fc->conn_init = 0;
1266 		fc->conn_error = 1;
1267 	}
1268 
1269 	fuse_set_initialized(fc);
1270 	wake_up_all(&fc->blocked_waitq);
1271 }
1272 
fuse_send_init(struct fuse_mount * fm)1273 void fuse_send_init(struct fuse_mount *fm)
1274 {
1275 	struct fuse_init_args *ia;
1276 
1277 	ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1278 
1279 	ia->in.major = FUSE_KERNEL_VERSION;
1280 	ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1281 	ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1282 	ia->in.flags |=
1283 		FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1284 		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1285 		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1286 		FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1287 		FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1288 		FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1289 		FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1290 		FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1291 		FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1292 		FUSE_PASSTHROUGH;
1293 #ifdef CONFIG_FUSE_DAX
1294 	if (fm->fc->dax)
1295 		ia->in.flags |= FUSE_MAP_ALIGNMENT;
1296 #endif
1297 	if (fm->fc->auto_submounts)
1298 		ia->in.flags |= FUSE_SUBMOUNTS;
1299 
1300 	ia->args.opcode = FUSE_INIT;
1301 	ia->args.in_numargs = 1;
1302 	ia->args.in_args[0].size = sizeof(ia->in);
1303 	ia->args.in_args[0].value = &ia->in;
1304 	ia->args.out_numargs = 1;
1305 	/* Variable length argument used for backward compatibility
1306 	   with interface version < 7.5.  Rest of init_out is zeroed
1307 	   by do_get_request(), so a short reply is not a problem */
1308 	ia->args.out_argvar = true;
1309 	ia->args.out_args[0].size = sizeof(ia->out);
1310 	ia->args.out_args[0].value = &ia->out;
1311 	ia->args.force = true;
1312 	ia->args.nocreds = true;
1313 	ia->args.end = process_init_reply;
1314 
1315 	if (unlikely(fm->fc->no_daemon) || fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1316 		process_init_reply(fm, &ia->args, -ENOTCONN);
1317 }
1318 EXPORT_SYMBOL_GPL(fuse_send_init);
1319 
free_fuse_passthrough(int id,void * p,void * data)1320 static int free_fuse_passthrough(int id, void *p, void *data)
1321 {
1322 	struct fuse_passthrough *passthrough = (struct fuse_passthrough *)p;
1323 
1324 	fuse_passthrough_release(passthrough);
1325 	kfree(p);
1326 
1327 	return 0;
1328 }
1329 
fuse_free_conn(struct fuse_conn * fc)1330 void fuse_free_conn(struct fuse_conn *fc)
1331 {
1332 	WARN_ON(!list_empty(&fc->devices));
1333 	idr_for_each(&fc->passthrough_req, free_fuse_passthrough, NULL);
1334 	idr_destroy(&fc->passthrough_req);
1335 	kfree_rcu(fc, rcu);
1336 }
1337 EXPORT_SYMBOL_GPL(fuse_free_conn);
1338 
fuse_bdi_init(struct fuse_conn * fc,struct super_block * sb)1339 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1340 {
1341 	int err;
1342 	char *suffix = "";
1343 
1344 	if (sb->s_bdev) {
1345 		suffix = "-fuseblk";
1346 		/*
1347 		 * sb->s_bdi points to blkdev's bdi however we want to redirect
1348 		 * it to our private bdi...
1349 		 */
1350 		bdi_put(sb->s_bdi);
1351 		sb->s_bdi = &noop_backing_dev_info;
1352 	}
1353 	err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1354 				   MINOR(fc->dev), suffix);
1355 	if (err)
1356 		return err;
1357 
1358 	/* fuse does it's own writeback accounting */
1359 	sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1360 	sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1361 
1362 	/*
1363 	 * For a single fuse filesystem use max 1% of dirty +
1364 	 * writeback threshold.
1365 	 *
1366 	 * This gives about 1M of write buffer for memory maps on a
1367 	 * machine with 1G and 10% dirty_ratio, which should be more
1368 	 * than enough.
1369 	 *
1370 	 * Privileged users can raise it by writing to
1371 	 *
1372 	 *    /sys/class/bdi/<bdi>/max_ratio
1373 	 */
1374 	bdi_set_max_ratio(sb->s_bdi, 1);
1375 
1376 	return 0;
1377 }
1378 
fuse_dev_alloc(void)1379 struct fuse_dev *fuse_dev_alloc(void)
1380 {
1381 	struct fuse_dev *fud;
1382 	struct list_head *pq;
1383 
1384 	fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1385 	if (!fud)
1386 		return NULL;
1387 
1388 	pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1389 	if (!pq) {
1390 		kfree(fud);
1391 		return NULL;
1392 	}
1393 
1394 	fud->pq.processing = pq;
1395 	fuse_pqueue_init(&fud->pq);
1396 
1397 	return fud;
1398 }
1399 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1400 
fuse_dev_install(struct fuse_dev * fud,struct fuse_conn * fc)1401 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1402 {
1403 	fud->fc = fuse_conn_get(fc);
1404 	spin_lock(&fc->lock);
1405 	list_add_tail(&fud->entry, &fc->devices);
1406 	spin_unlock(&fc->lock);
1407 }
1408 EXPORT_SYMBOL_GPL(fuse_dev_install);
1409 
fuse_dev_alloc_install(struct fuse_conn * fc)1410 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1411 {
1412 	struct fuse_dev *fud;
1413 
1414 	fud = fuse_dev_alloc();
1415 	if (!fud)
1416 		return NULL;
1417 
1418 	fuse_dev_install(fud, fc);
1419 	return fud;
1420 }
1421 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1422 
fuse_dev_free(struct fuse_dev * fud)1423 void fuse_dev_free(struct fuse_dev *fud)
1424 {
1425 	struct fuse_conn *fc = fud->fc;
1426 
1427 	if (fc) {
1428 		spin_lock(&fc->lock);
1429 		list_del(&fud->entry);
1430 		spin_unlock(&fc->lock);
1431 
1432 		fuse_conn_put(fc);
1433 	}
1434 	kfree(fud->pq.processing);
1435 	kfree(fud);
1436 }
1437 EXPORT_SYMBOL_GPL(fuse_dev_free);
1438 
fuse_sb_defaults(struct super_block * sb)1439 static void fuse_sb_defaults(struct super_block *sb)
1440 {
1441 	sb->s_magic = FUSE_SUPER_MAGIC;
1442 	sb->s_op = &fuse_super_operations;
1443 	sb->s_xattr = fuse_xattr_handlers;
1444 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1445 	sb->s_time_gran = 1;
1446 	sb->s_export_op = &fuse_export_operations;
1447 	sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1448 	if (sb->s_user_ns != &init_user_ns)
1449 		sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1450 	sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1451 
1452 	/*
1453 	 * If we are not in the initial user namespace posix
1454 	 * acls must be translated.
1455 	 */
1456 	if (sb->s_user_ns != &init_user_ns)
1457 		sb->s_xattr = fuse_no_acl_xattr_handlers;
1458 }
1459 
fuse_fill_super_submount(struct super_block * sb,struct fuse_inode * parent_fi)1460 int fuse_fill_super_submount(struct super_block *sb,
1461 			     struct fuse_inode *parent_fi)
1462 {
1463 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1464 	struct super_block *parent_sb = parent_fi->inode.i_sb;
1465 	struct fuse_attr root_attr;
1466 	struct inode *root;
1467 
1468 	fuse_sb_defaults(sb);
1469 	fm->sb = sb;
1470 
1471 	WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1472 	sb->s_bdi = bdi_get(parent_sb->s_bdi);
1473 
1474 	sb->s_xattr = parent_sb->s_xattr;
1475 	sb->s_time_gran = parent_sb->s_time_gran;
1476 	sb->s_blocksize = parent_sb->s_blocksize;
1477 	sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1478 	sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1479 	if (parent_sb->s_subtype && !sb->s_subtype)
1480 		return -ENOMEM;
1481 
1482 	fuse_fill_attr_from_inode(&root_attr, &parent_fi->inode);
1483 	root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1484 	/*
1485 	 * This inode is just a duplicate, so it is not looked up and
1486 	 * its nlookup should not be incremented.  fuse_iget() does
1487 	 * that, though, so undo it here.
1488 	 */
1489 	get_fuse_inode(root)->nlookup--;
1490 	sb->s_d_op = &fuse_dentry_operations;
1491 	sb->s_root = d_make_root(root);
1492 	if (!sb->s_root)
1493 		return -ENOMEM;
1494 
1495 	return 0;
1496 }
1497 
fuse_fill_super_common(struct super_block * sb,struct fuse_fs_context * ctx)1498 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1499 {
1500 	struct fuse_dev *fud = NULL;
1501 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1502 	struct fuse_conn *fc = fm->fc;
1503 	struct inode *root;
1504 	struct dentry *root_dentry;
1505 	int err;
1506 
1507 	err = -EINVAL;
1508 	if (sb->s_flags & SB_MANDLOCK)
1509 		goto err;
1510 
1511 	fuse_sb_defaults(sb);
1512 
1513 	if (ctx->is_bdev) {
1514 #ifdef CONFIG_BLOCK
1515 		err = -EINVAL;
1516 		if (!sb_set_blocksize(sb, ctx->blksize))
1517 			goto err;
1518 #endif
1519 	} else {
1520 		sb->s_blocksize = PAGE_SIZE;
1521 		sb->s_blocksize_bits = PAGE_SHIFT;
1522 	}
1523 
1524 	sb->s_subtype = ctx->subtype;
1525 	ctx->subtype = NULL;
1526 	if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1527 		err = fuse_dax_conn_alloc(fc, ctx->dax_dev);
1528 		if (err)
1529 			goto err;
1530 	}
1531 
1532 	if (ctx->fudptr) {
1533 		err = -ENOMEM;
1534 		fud = fuse_dev_alloc_install(fc);
1535 		if (!fud)
1536 			goto err_free_dax;
1537 	}
1538 
1539 	fc->dev = sb->s_dev;
1540 	fm->sb = sb;
1541 	err = fuse_bdi_init(fc, sb);
1542 	if (err)
1543 		goto err_dev_free;
1544 
1545 	/* Handle umasking inside the fuse code */
1546 	if (sb->s_flags & SB_POSIXACL)
1547 		fc->dont_mask = 1;
1548 	sb->s_flags |= SB_POSIXACL;
1549 
1550 	fc->default_permissions = ctx->default_permissions;
1551 	fc->allow_other = ctx->allow_other;
1552 	fc->user_id = ctx->user_id;
1553 	fc->group_id = ctx->group_id;
1554 	fc->legacy_opts_show = ctx->legacy_opts_show;
1555 	fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1556 	fc->destroy = ctx->destroy;
1557 	fc->no_control = ctx->no_control;
1558 	fc->no_force_umount = ctx->no_force_umount;
1559 	fc->no_daemon = ctx->no_daemon;
1560 
1561 	err = -ENOMEM;
1562 	root = fuse_get_root_inode(sb, ctx->rootmode, ctx->root_bpf,
1563 				   ctx->root_dir);
1564 	sb->s_d_op = &fuse_root_dentry_operations;
1565 	root_dentry = d_make_root(root);
1566 	if (!root_dentry)
1567 		goto err_dev_free;
1568 	fuse_init_dentry_root(root_dentry, ctx->root_dir);
1569 	/* Root dentry doesn't have .d_revalidate */
1570 	sb->s_d_op = &fuse_dentry_operations;
1571 
1572 	mutex_lock(&fuse_mutex);
1573 	err = -EINVAL;
1574 	if (ctx->fudptr && *ctx->fudptr)
1575 		goto err_unlock;
1576 
1577 	err = fuse_ctl_add_conn(fc);
1578 	if (err)
1579 		goto err_unlock;
1580 
1581 	list_add_tail(&fc->entry, &fuse_conn_list);
1582 	sb->s_root = root_dentry;
1583 	if (ctx->fudptr)
1584 		*ctx->fudptr = fud;
1585 	mutex_unlock(&fuse_mutex);
1586 	return 0;
1587 
1588  err_unlock:
1589 	mutex_unlock(&fuse_mutex);
1590 	dput(root_dentry);
1591  err_dev_free:
1592 	if (fud)
1593 		fuse_dev_free(fud);
1594  err_free_dax:
1595 	if (IS_ENABLED(CONFIG_FUSE_DAX))
1596 		fuse_dax_conn_free(fc);
1597  err:
1598 	return err;
1599 }
1600 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1601 
fuse_fill_super(struct super_block * sb,struct fs_context * fsc)1602 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1603 {
1604 	struct fuse_fs_context *ctx = fsc->fs_private;
1605 	struct file *file;
1606 	int err;
1607 	struct fuse_conn *fc;
1608 	struct fuse_mount *fm;
1609 
1610 	err = -EINVAL;
1611 	if (!ctx->no_daemon) {
1612 		file = fget(ctx->fd);
1613 		if (!file)
1614 			goto err;
1615 
1616 		/*
1617 		 * Require mount to happen from the same user namespace which
1618 		 * opened /dev/fuse to prevent potential attacks.
1619 		 */
1620 		if ((file->f_op != &fuse_dev_operations) ||
1621 		    (file->f_cred->user_ns != sb->s_user_ns))
1622 			goto err_fput;
1623 		ctx->fudptr = &file->private_data;
1624 	}
1625 
1626 	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1627 	err = -ENOMEM;
1628 	if (!fc)
1629 		goto err_fput;
1630 
1631 	fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1632 	if (!fm) {
1633 		kfree(fc);
1634 		goto err_fput;
1635 	}
1636 
1637 	fuse_conn_init(fc, fm, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1638 	fc->release = fuse_free_conn;
1639 
1640 	sb->s_fs_info = fm;
1641 
1642 	err = fuse_fill_super_common(sb, ctx);
1643 	if (err)
1644 		goto err_put_conn;
1645 	/*
1646 	 * atomic_dec_and_test() in fput() provides the necessary
1647 	 * memory barrier for file->private_data to be visible on all
1648 	 * CPUs after this
1649 	 */
1650 	if (!ctx->no_daemon)
1651 		fput(file);
1652 	fuse_send_init(get_fuse_mount_super(sb));
1653 	return 0;
1654 
1655  err_put_conn:
1656 	fuse_mount_put(fm);
1657 	sb->s_fs_info = NULL;
1658  err_fput:
1659 	if (!ctx->no_daemon)
1660 		fput(file);
1661  err:
1662 	return err;
1663 }
1664 
fuse_get_tree(struct fs_context * fc)1665 static int fuse_get_tree(struct fs_context *fc)
1666 {
1667 	struct fuse_fs_context *ctx = fc->fs_private;
1668 
1669 	if (!ctx->fd_present || !ctx->rootmode_present ||
1670 	    !ctx->user_id_present || !ctx->group_id_present)
1671 		return -EINVAL;
1672 
1673 #ifdef CONFIG_BLOCK
1674 	if (ctx->is_bdev)
1675 		return get_tree_bdev(fc, fuse_fill_super);
1676 #endif
1677 
1678 	return get_tree_nodev(fc, fuse_fill_super);
1679 }
1680 
1681 static const struct fs_context_operations fuse_context_ops = {
1682 	.free		= fuse_free_fc,
1683 	.parse_param	= fuse_parse_param,
1684 	.reconfigure	= fuse_reconfigure,
1685 	.get_tree	= fuse_get_tree,
1686 };
1687 
1688 /*
1689  * Set up the filesystem mount context.
1690  */
fuse_init_fs_context(struct fs_context * fc)1691 static int fuse_init_fs_context(struct fs_context *fc)
1692 {
1693 	struct fuse_fs_context *ctx;
1694 
1695 	ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1696 	if (!ctx)
1697 		return -ENOMEM;
1698 
1699 	ctx->max_read = ~0;
1700 	ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1701 	ctx->legacy_opts_show = true;
1702 
1703 #ifdef CONFIG_BLOCK
1704 	if (fc->fs_type == &fuseblk_fs_type) {
1705 		ctx->is_bdev = true;
1706 		ctx->destroy = true;
1707 	}
1708 #endif
1709 
1710 	fc->fs_private = ctx;
1711 	fc->ops = &fuse_context_ops;
1712 	return 0;
1713 }
1714 
fuse_mount_remove(struct fuse_mount * fm)1715 bool fuse_mount_remove(struct fuse_mount *fm)
1716 {
1717 	struct fuse_conn *fc = fm->fc;
1718 	bool last = false;
1719 
1720 	down_write(&fc->killsb);
1721 	list_del_init(&fm->fc_entry);
1722 	if (list_empty(&fc->mounts))
1723 		last = true;
1724 	up_write(&fc->killsb);
1725 
1726 	return last;
1727 }
1728 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1729 
fuse_conn_destroy(struct fuse_mount * fm)1730 void fuse_conn_destroy(struct fuse_mount *fm)
1731 {
1732 	struct fuse_conn *fc = fm->fc;
1733 
1734 	if (fc->destroy)
1735 		fuse_send_destroy(fm);
1736 
1737 	fuse_abort_conn(fc);
1738 	fuse_wait_aborted(fc);
1739 
1740 	if (!list_empty(&fc->entry)) {
1741 		mutex_lock(&fuse_mutex);
1742 		list_del(&fc->entry);
1743 		fuse_ctl_remove_conn(fc);
1744 		mutex_unlock(&fuse_mutex);
1745 	}
1746 }
1747 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1748 
fuse_kill_sb_anon(struct super_block * sb)1749 static void fuse_kill_sb_anon(struct super_block *sb)
1750 {
1751 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1752 	bool last;
1753 
1754 	if (fm) {
1755 		last = fuse_mount_remove(fm);
1756 		if (last)
1757 			fuse_conn_destroy(fm);
1758 	}
1759 	kill_anon_super(sb);
1760 }
1761 
1762 static struct file_system_type fuse_fs_type = {
1763 	.owner		= THIS_MODULE,
1764 	.name		= "fuse",
1765 	.fs_flags	= FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1766 	.init_fs_context = fuse_init_fs_context,
1767 	.parameters	= fuse_fs_parameters,
1768 	.kill_sb	= fuse_kill_sb_anon,
1769 };
1770 MODULE_ALIAS_FS("fuse");
1771 
1772 #ifdef CONFIG_BLOCK
fuse_kill_sb_blk(struct super_block * sb)1773 static void fuse_kill_sb_blk(struct super_block *sb)
1774 {
1775 	struct fuse_mount *fm = get_fuse_mount_super(sb);
1776 	bool last;
1777 
1778 	if (sb->s_root) {
1779 		last = fuse_mount_remove(fm);
1780 		if (last)
1781 			fuse_conn_destroy(fm);
1782 	}
1783 	kill_block_super(sb);
1784 }
1785 
1786 static struct file_system_type fuseblk_fs_type = {
1787 	.owner		= THIS_MODULE,
1788 	.name		= "fuseblk",
1789 	.init_fs_context = fuse_init_fs_context,
1790 	.parameters	= fuse_fs_parameters,
1791 	.kill_sb	= fuse_kill_sb_blk,
1792 	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1793 };
1794 MODULE_ALIAS_FS("fuseblk");
1795 
register_fuseblk(void)1796 static inline int register_fuseblk(void)
1797 {
1798 	return register_filesystem(&fuseblk_fs_type);
1799 }
1800 
unregister_fuseblk(void)1801 static inline void unregister_fuseblk(void)
1802 {
1803 	unregister_filesystem(&fuseblk_fs_type);
1804 }
1805 #else
register_fuseblk(void)1806 static inline int register_fuseblk(void)
1807 {
1808 	return 0;
1809 }
1810 
unregister_fuseblk(void)1811 static inline void unregister_fuseblk(void)
1812 {
1813 }
1814 #endif
1815 
fuse_inode_init_once(void * foo)1816 static void fuse_inode_init_once(void *foo)
1817 {
1818 	struct inode *inode = foo;
1819 
1820 	inode_init_once(inode);
1821 }
1822 
fuse_fs_init(void)1823 static int __init fuse_fs_init(void)
1824 {
1825 	int err;
1826 
1827 	fuse_inode_cachep = kmem_cache_create("fuse_inode",
1828 			sizeof(struct fuse_inode), 0,
1829 			SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1830 			fuse_inode_init_once);
1831 	err = -ENOMEM;
1832 	if (!fuse_inode_cachep)
1833 		goto out;
1834 
1835 	err = register_fuseblk();
1836 	if (err)
1837 		goto out2;
1838 
1839 	err = register_filesystem(&fuse_fs_type);
1840 	if (err)
1841 		goto out3;
1842 
1843 	return 0;
1844 
1845  out3:
1846 	unregister_fuseblk();
1847  out2:
1848 	kmem_cache_destroy(fuse_inode_cachep);
1849  out:
1850 	return err;
1851 }
1852 
fuse_fs_cleanup(void)1853 static void fuse_fs_cleanup(void)
1854 {
1855 	unregister_filesystem(&fuse_fs_type);
1856 	unregister_fuseblk();
1857 
1858 	/*
1859 	 * Make sure all delayed rcu free inodes are flushed before we
1860 	 * destroy cache.
1861 	 */
1862 	rcu_barrier();
1863 	kmem_cache_destroy(fuse_inode_cachep);
1864 }
1865 
1866 static struct kobject *fuse_kobj;
1867 
fuse_bpf_show(struct kobject * kobj,struct kobj_attribute * attr,char * buff)1868 static ssize_t fuse_bpf_show(struct kobject *kobj,
1869 				       struct kobj_attribute *attr, char *buff)
1870 {
1871 	return sysfs_emit(buff, "supported\n");
1872 }
1873 
1874 static struct kobj_attribute fuse_bpf_attr =
1875 		__ATTR_RO(fuse_bpf);
1876 
1877 static struct attribute *bpf_features[] = {
1878 	&fuse_bpf_attr.attr,
1879 	NULL,
1880 };
1881 
1882 static const struct attribute_group bpf_features_group = {
1883 	.name = "features",
1884 	.attrs = bpf_features,
1885 };
1886 
1887 /*
1888  * TODO Remove this once fuse-bpf is upstreamed
1889  *
1890  * bpf_prog_type_fuse exports the bpf_prog_type_fuse 'constant', which cannot be
1891  * constant until the code is upstreamed
1892  */
bpf_prog_type_fuse_show(struct kobject * kobj,struct kobj_attribute * attr,char * buff)1893 static ssize_t bpf_prog_type_fuse_show(struct kobject *kobj,
1894 				       struct kobj_attribute *attr, char *buff)
1895 {
1896 	return sysfs_emit(buff, "%d\n", BPF_PROG_TYPE_FUSE);
1897 }
1898 
1899 static struct kobj_attribute bpf_prog_type_fuse_attr =
1900 		__ATTR_RO(bpf_prog_type_fuse);
1901 
1902 static struct attribute *bpf_attributes[] = {
1903 	&bpf_prog_type_fuse_attr.attr,
1904 	NULL,
1905 };
1906 
1907 static const struct attribute_group bpf_attr_group = {
1908 	.attrs = bpf_attributes,
1909 };
1910 
1911 static const struct attribute_group *attribute_groups[] = {
1912 	&bpf_features_group,
1913 	&bpf_attr_group,
1914 	NULL
1915 };
1916 
1917 /* TODO remove to here */
1918 
fuse_sysfs_init(void)1919 static int fuse_sysfs_init(void)
1920 {
1921 	int err;
1922 
1923 	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1924 	if (!fuse_kobj) {
1925 		err = -ENOMEM;
1926 		goto out_err;
1927 	}
1928 
1929 	err = sysfs_create_mount_point(fuse_kobj, "connections");
1930 	if (err)
1931 		goto out_fuse_unregister;
1932 
1933 	/* TODO Remove when BPF_PROG_TYPE_FUSE is upstreamed */
1934 	err = sysfs_create_groups(fuse_kobj, attribute_groups);
1935 	if (err)
1936 		goto out_fuse_remove_mount_point;
1937 
1938 	return 0;
1939 
1940  out_fuse_remove_mount_point:
1941 	sysfs_remove_mount_point(fuse_kobj, "connections");
1942  out_fuse_unregister:
1943 	kobject_put(fuse_kobj);
1944  out_err:
1945 	return err;
1946 }
1947 
fuse_sysfs_cleanup(void)1948 static void fuse_sysfs_cleanup(void)
1949 {
1950 	sysfs_remove_groups(fuse_kobj, attribute_groups);
1951 	sysfs_remove_mount_point(fuse_kobj, "connections");
1952 	kobject_put(fuse_kobj);
1953 }
1954 
fuse_init(void)1955 static int __init fuse_init(void)
1956 {
1957 	int res;
1958 
1959 	pr_info("init (API version %i.%i)\n",
1960 		FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1961 
1962 	INIT_LIST_HEAD(&fuse_conn_list);
1963 	res = fuse_fs_init();
1964 	if (res)
1965 		goto err;
1966 
1967 	res = fuse_dev_init();
1968 	if (res)
1969 		goto err_fs_cleanup;
1970 
1971 	res = fuse_sysfs_init();
1972 	if (res)
1973 		goto err_dev_cleanup;
1974 
1975 	res = fuse_ctl_init();
1976 	if (res)
1977 		goto err_sysfs_cleanup;
1978 
1979 #ifdef CONFIG_FUSE_BPF
1980 	res = fuse_bpf_init();
1981 	if (res)
1982 		goto err_ctl_cleanup;
1983 #endif
1984 
1985 	sanitize_global_limit(&max_user_bgreq);
1986 	sanitize_global_limit(&max_user_congthresh);
1987 
1988 	return 0;
1989 
1990 #ifdef CONFIG_FUSE_BPF
1991  err_ctl_cleanup:
1992 	fuse_ctl_cleanup();
1993 #endif
1994  err_sysfs_cleanup:
1995 	fuse_sysfs_cleanup();
1996  err_dev_cleanup:
1997 	fuse_dev_cleanup();
1998  err_fs_cleanup:
1999 	fuse_fs_cleanup();
2000  err:
2001 	return res;
2002 }
2003 
fuse_exit(void)2004 static void __exit fuse_exit(void)
2005 {
2006 	pr_debug("exit\n");
2007 
2008 	fuse_ctl_cleanup();
2009 	fuse_sysfs_cleanup();
2010 	fuse_fs_cleanup();
2011 #ifdef CONFIG_FUSE_BPF
2012 	fuse_bpf_cleanup();
2013 #endif
2014 	fuse_dev_cleanup();
2015 }
2016 
2017 module_init(fuse_init);
2018 module_exit(fuse_exit);
2019