xref: /linux/kernel/bpf/inode.c (revision e58e871becec2d3b04ed91c0c16fe8deac9c9dfa)
1 /*
2  * Minimal file system backend for holding eBPF maps and programs,
3  * used by bpf(2) object pinning.
4  *
5  * Authors:
6  *
7  *	Daniel Borkmann <daniel@iogearbox.net>
8  *
9  * This program is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU General Public License
11  * version 2 as published by the Free Software Foundation.
12  */
13 
14 #include <linux/init.h>
15 #include <linux/magic.h>
16 #include <linux/major.h>
17 #include <linux/mount.h>
18 #include <linux/namei.h>
19 #include <linux/fs.h>
20 #include <linux/kdev_t.h>
21 #include <linux/parser.h>
22 #include <linux/filter.h>
23 #include <linux/bpf.h>
24 #include <linux/bpf_trace.h>
25 
26 enum bpf_type {
27 	BPF_TYPE_UNSPEC	= 0,
28 	BPF_TYPE_PROG,
29 	BPF_TYPE_MAP,
30 };
31 
32 static void *bpf_any_get(void *raw, enum bpf_type type)
33 {
34 	switch (type) {
35 	case BPF_TYPE_PROG:
36 		raw = bpf_prog_inc(raw);
37 		break;
38 	case BPF_TYPE_MAP:
39 		raw = bpf_map_inc(raw, true);
40 		break;
41 	default:
42 		WARN_ON_ONCE(1);
43 		break;
44 	}
45 
46 	return raw;
47 }
48 
49 static void bpf_any_put(void *raw, enum bpf_type type)
50 {
51 	switch (type) {
52 	case BPF_TYPE_PROG:
53 		bpf_prog_put(raw);
54 		break;
55 	case BPF_TYPE_MAP:
56 		bpf_map_put_with_uref(raw);
57 		break;
58 	default:
59 		WARN_ON_ONCE(1);
60 		break;
61 	}
62 }
63 
64 static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
65 {
66 	void *raw;
67 
68 	*type = BPF_TYPE_MAP;
69 	raw = bpf_map_get_with_uref(ufd);
70 	if (IS_ERR(raw)) {
71 		*type = BPF_TYPE_PROG;
72 		raw = bpf_prog_get(ufd);
73 	}
74 
75 	return raw;
76 }
77 
78 static const struct inode_operations bpf_dir_iops;
79 
80 static const struct inode_operations bpf_prog_iops = { };
81 static const struct inode_operations bpf_map_iops  = { };
82 
83 static struct inode *bpf_get_inode(struct super_block *sb,
84 				   const struct inode *dir,
85 				   umode_t mode)
86 {
87 	struct inode *inode;
88 
89 	switch (mode & S_IFMT) {
90 	case S_IFDIR:
91 	case S_IFREG:
92 	case S_IFLNK:
93 		break;
94 	default:
95 		return ERR_PTR(-EINVAL);
96 	}
97 
98 	inode = new_inode(sb);
99 	if (!inode)
100 		return ERR_PTR(-ENOSPC);
101 
102 	inode->i_ino = get_next_ino();
103 	inode->i_atime = current_time(inode);
104 	inode->i_mtime = inode->i_atime;
105 	inode->i_ctime = inode->i_atime;
106 
107 	inode_init_owner(inode, dir, mode);
108 
109 	return inode;
110 }
111 
112 static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
113 {
114 	*type = BPF_TYPE_UNSPEC;
115 	if (inode->i_op == &bpf_prog_iops)
116 		*type = BPF_TYPE_PROG;
117 	else if (inode->i_op == &bpf_map_iops)
118 		*type = BPF_TYPE_MAP;
119 	else
120 		return -EACCES;
121 
122 	return 0;
123 }
124 
125 static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
126 				struct inode *dir)
127 {
128 	d_instantiate(dentry, inode);
129 	dget(dentry);
130 
131 	dir->i_mtime = current_time(dir);
132 	dir->i_ctime = dir->i_mtime;
133 }
134 
135 static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
136 {
137 	struct inode *inode;
138 
139 	inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
140 	if (IS_ERR(inode))
141 		return PTR_ERR(inode);
142 
143 	inode->i_op = &bpf_dir_iops;
144 	inode->i_fop = &simple_dir_operations;
145 
146 	inc_nlink(inode);
147 	inc_nlink(dir);
148 
149 	bpf_dentry_finalize(dentry, inode, dir);
150 	return 0;
151 }
152 
153 static int bpf_mkobj_ops(struct inode *dir, struct dentry *dentry,
154 			 umode_t mode, const struct inode_operations *iops)
155 {
156 	struct inode *inode;
157 
158 	inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFREG);
159 	if (IS_ERR(inode))
160 		return PTR_ERR(inode);
161 
162 	inode->i_op = iops;
163 	inode->i_private = dentry->d_fsdata;
164 
165 	bpf_dentry_finalize(dentry, inode, dir);
166 	return 0;
167 }
168 
169 static int bpf_mkobj(struct inode *dir, struct dentry *dentry, umode_t mode,
170 		     dev_t devt)
171 {
172 	enum bpf_type type = MINOR(devt);
173 
174 	if (MAJOR(devt) != UNNAMED_MAJOR || !S_ISREG(mode) ||
175 	    dentry->d_fsdata == NULL)
176 		return -EPERM;
177 
178 	switch (type) {
179 	case BPF_TYPE_PROG:
180 		return bpf_mkobj_ops(dir, dentry, mode, &bpf_prog_iops);
181 	case BPF_TYPE_MAP:
182 		return bpf_mkobj_ops(dir, dentry, mode, &bpf_map_iops);
183 	default:
184 		return -EPERM;
185 	}
186 }
187 
188 static struct dentry *
189 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
190 {
191 	if (strchr(dentry->d_name.name, '.'))
192 		return ERR_PTR(-EPERM);
193 
194 	return simple_lookup(dir, dentry, flags);
195 }
196 
197 static int bpf_symlink(struct inode *dir, struct dentry *dentry,
198 		       const char *target)
199 {
200 	char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
201 	struct inode *inode;
202 
203 	if (!link)
204 		return -ENOMEM;
205 
206 	inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
207 	if (IS_ERR(inode)) {
208 		kfree(link);
209 		return PTR_ERR(inode);
210 	}
211 
212 	inode->i_op = &simple_symlink_inode_operations;
213 	inode->i_link = link;
214 
215 	bpf_dentry_finalize(dentry, inode, dir);
216 	return 0;
217 }
218 
219 static const struct inode_operations bpf_dir_iops = {
220 	.lookup		= bpf_lookup,
221 	.mknod		= bpf_mkobj,
222 	.mkdir		= bpf_mkdir,
223 	.symlink	= bpf_symlink,
224 	.rmdir		= simple_rmdir,
225 	.rename		= simple_rename,
226 	.link		= simple_link,
227 	.unlink		= simple_unlink,
228 };
229 
230 static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
231 			  enum bpf_type type)
232 {
233 	struct dentry *dentry;
234 	struct inode *dir;
235 	struct path path;
236 	umode_t mode;
237 	dev_t devt;
238 	int ret;
239 
240 	dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
241 	if (IS_ERR(dentry))
242 		return PTR_ERR(dentry);
243 
244 	mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
245 	devt = MKDEV(UNNAMED_MAJOR, type);
246 
247 	ret = security_path_mknod(&path, dentry, mode, devt);
248 	if (ret)
249 		goto out;
250 
251 	dir = d_inode(path.dentry);
252 	if (dir->i_op != &bpf_dir_iops) {
253 		ret = -EPERM;
254 		goto out;
255 	}
256 
257 	dentry->d_fsdata = raw;
258 	ret = vfs_mknod(dir, dentry, mode, devt);
259 	dentry->d_fsdata = NULL;
260 out:
261 	done_path_create(&path, dentry);
262 	return ret;
263 }
264 
265 int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
266 {
267 	struct filename *pname;
268 	enum bpf_type type;
269 	void *raw;
270 	int ret;
271 
272 	pname = getname(pathname);
273 	if (IS_ERR(pname))
274 		return PTR_ERR(pname);
275 
276 	raw = bpf_fd_probe_obj(ufd, &type);
277 	if (IS_ERR(raw)) {
278 		ret = PTR_ERR(raw);
279 		goto out;
280 	}
281 
282 	ret = bpf_obj_do_pin(pname, raw, type);
283 	if (ret != 0)
284 		bpf_any_put(raw, type);
285 	if ((trace_bpf_obj_pin_prog_enabled() ||
286 	     trace_bpf_obj_pin_map_enabled()) && !ret) {
287 		if (type == BPF_TYPE_PROG)
288 			trace_bpf_obj_pin_prog(raw, ufd, pname);
289 		if (type == BPF_TYPE_MAP)
290 			trace_bpf_obj_pin_map(raw, ufd, pname);
291 	}
292 out:
293 	putname(pname);
294 	return ret;
295 }
296 
297 static void *bpf_obj_do_get(const struct filename *pathname,
298 			    enum bpf_type *type)
299 {
300 	struct inode *inode;
301 	struct path path;
302 	void *raw;
303 	int ret;
304 
305 	ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
306 	if (ret)
307 		return ERR_PTR(ret);
308 
309 	inode = d_backing_inode(path.dentry);
310 	ret = inode_permission(inode, MAY_WRITE);
311 	if (ret)
312 		goto out;
313 
314 	ret = bpf_inode_type(inode, type);
315 	if (ret)
316 		goto out;
317 
318 	raw = bpf_any_get(inode->i_private, *type);
319 	if (!IS_ERR(raw))
320 		touch_atime(&path);
321 
322 	path_put(&path);
323 	return raw;
324 out:
325 	path_put(&path);
326 	return ERR_PTR(ret);
327 }
328 
329 int bpf_obj_get_user(const char __user *pathname)
330 {
331 	enum bpf_type type = BPF_TYPE_UNSPEC;
332 	struct filename *pname;
333 	int ret = -ENOENT;
334 	void *raw;
335 
336 	pname = getname(pathname);
337 	if (IS_ERR(pname))
338 		return PTR_ERR(pname);
339 
340 	raw = bpf_obj_do_get(pname, &type);
341 	if (IS_ERR(raw)) {
342 		ret = PTR_ERR(raw);
343 		goto out;
344 	}
345 
346 	if (type == BPF_TYPE_PROG)
347 		ret = bpf_prog_new_fd(raw);
348 	else if (type == BPF_TYPE_MAP)
349 		ret = bpf_map_new_fd(raw);
350 	else
351 		goto out;
352 
353 	if (ret < 0) {
354 		bpf_any_put(raw, type);
355 	} else if (trace_bpf_obj_get_prog_enabled() ||
356 		   trace_bpf_obj_get_map_enabled()) {
357 		if (type == BPF_TYPE_PROG)
358 			trace_bpf_obj_get_prog(raw, ret, pname);
359 		if (type == BPF_TYPE_MAP)
360 			trace_bpf_obj_get_map(raw, ret, pname);
361 	}
362 out:
363 	putname(pname);
364 	return ret;
365 }
366 
367 static void bpf_evict_inode(struct inode *inode)
368 {
369 	enum bpf_type type;
370 
371 	truncate_inode_pages_final(&inode->i_data);
372 	clear_inode(inode);
373 
374 	if (S_ISLNK(inode->i_mode))
375 		kfree(inode->i_link);
376 	if (!bpf_inode_type(inode, &type))
377 		bpf_any_put(inode->i_private, type);
378 }
379 
380 static const struct super_operations bpf_super_ops = {
381 	.statfs		= simple_statfs,
382 	.drop_inode	= generic_delete_inode,
383 	.show_options	= generic_show_options,
384 	.evict_inode	= bpf_evict_inode,
385 };
386 
387 enum {
388 	OPT_MODE,
389 	OPT_ERR,
390 };
391 
392 static const match_table_t bpf_mount_tokens = {
393 	{ OPT_MODE, "mode=%o" },
394 	{ OPT_ERR, NULL },
395 };
396 
397 struct bpf_mount_opts {
398 	umode_t mode;
399 };
400 
401 static int bpf_parse_options(char *data, struct bpf_mount_opts *opts)
402 {
403 	substring_t args[MAX_OPT_ARGS];
404 	int option, token;
405 	char *ptr;
406 
407 	opts->mode = S_IRWXUGO;
408 
409 	while ((ptr = strsep(&data, ",")) != NULL) {
410 		if (!*ptr)
411 			continue;
412 
413 		token = match_token(ptr, bpf_mount_tokens, args);
414 		switch (token) {
415 		case OPT_MODE:
416 			if (match_octal(&args[0], &option))
417 				return -EINVAL;
418 			opts->mode = option & S_IALLUGO;
419 			break;
420 		/* We might like to report bad mount options here, but
421 		 * traditionally we've ignored all mount options, so we'd
422 		 * better continue to ignore non-existing options for bpf.
423 		 */
424 		}
425 	}
426 
427 	return 0;
428 }
429 
430 static int bpf_fill_super(struct super_block *sb, void *data, int silent)
431 {
432 	static const struct tree_descr bpf_rfiles[] = { { "" } };
433 	struct bpf_mount_opts opts;
434 	struct inode *inode;
435 	int ret;
436 
437 	save_mount_options(sb, data);
438 
439 	ret = bpf_parse_options(data, &opts);
440 	if (ret)
441 		return ret;
442 
443 	ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
444 	if (ret)
445 		return ret;
446 
447 	sb->s_op = &bpf_super_ops;
448 
449 	inode = sb->s_root->d_inode;
450 	inode->i_op = &bpf_dir_iops;
451 	inode->i_mode &= ~S_IALLUGO;
452 	inode->i_mode |= S_ISVTX | opts.mode;
453 
454 	return 0;
455 }
456 
457 static struct dentry *bpf_mount(struct file_system_type *type, int flags,
458 				const char *dev_name, void *data)
459 {
460 	return mount_nodev(type, flags, data, bpf_fill_super);
461 }
462 
463 static struct file_system_type bpf_fs_type = {
464 	.owner		= THIS_MODULE,
465 	.name		= "bpf",
466 	.mount		= bpf_mount,
467 	.kill_sb	= kill_litter_super,
468 };
469 
470 static int __init bpf_init(void)
471 {
472 	int ret;
473 
474 	ret = sysfs_create_mount_point(fs_kobj, "bpf");
475 	if (ret)
476 		return ret;
477 
478 	ret = register_filesystem(&bpf_fs_type);
479 	if (ret)
480 		sysfs_remove_mount_point(fs_kobj, "bpf");
481 
482 	return ret;
483 }
484 fs_initcall(bpf_init);
485