[people/ms/linux.git] / kernel / bpf / inode.c

/*
 * Minimal file system backend for holding eBPF maps and programs,
 * used by bpf(2) object pinning.
 *
 * Authors:
 *
 *	Daniel Borkmann <daniel@iogearbox.net>
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * version 2 as published by the Free Software Foundation.
 */

#include <linux/module.h>
#include <linux/magic.h>
#include <linux/major.h>
#include <linux/mount.h>
#include <linux/namei.h>
#include <linux/fs.h>
#include <linux/kdev_t.h>
#include <linux/filter.h>
#include <linux/bpf.h>

enum bpf_type {
	BPF_TYPE_UNSPEC	= 0,
	BPF_TYPE_PROG,
	BPF_TYPE_MAP,
};

static void *bpf_any_get(void *raw, enum bpf_type type)
{
	switch (type) {
	case BPF_TYPE_PROG:
		atomic_inc(&((struct bpf_prog *)raw)->aux->refcnt);
		break;
	case BPF_TYPE_MAP:
		bpf_map_inc(raw, true);
		break;
	default:
		WARN_ON_ONCE(1);
		break;
	}

	return raw;
}

static void bpf_any_put(void *raw, enum bpf_type type)
{
	switch (type) {
	case BPF_TYPE_PROG:
		bpf_prog_put(raw);
		break;
	case BPF_TYPE_MAP:
		bpf_map_put_with_uref(raw);
		break;
	default:
		WARN_ON_ONCE(1);
		break;
	}
}

static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
{
	void *raw;

	*type = BPF_TYPE_MAP;
	raw = bpf_map_get_with_uref(ufd);
	if (IS_ERR(raw)) {
		*type = BPF_TYPE_PROG;
		raw = bpf_prog_get(ufd);
	}

	return raw;
}

static const struct inode_operations bpf_dir_iops;

static const struct inode_operations bpf_prog_iops = { };
static const struct inode_operations bpf_map_iops  = { };

static struct inode *bpf_get_inode(struct super_block *sb,
				   const struct inode *dir,
				   umode_t mode)
{
	struct inode *inode;

	switch (mode & S_IFMT) {
	case S_IFDIR:
	case S_IFREG:
		break;
	default:
		return ERR_PTR(-EINVAL);
	}

	inode = new_inode(sb);
	if (!inode)
		return ERR_PTR(-ENOSPC);

	inode->i_ino = get_next_ino();
	inode->i_atime = CURRENT_TIME;
	inode->i_mtime = inode->i_atime;
	inode->i_ctime = inode->i_atime;

	inode_init_owner(inode, dir, mode);

	return inode;
}

static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
{
	*type = BPF_TYPE_UNSPEC;
	if (inode->i_op == &bpf_prog_iops)
		*type = BPF_TYPE_PROG;
	else if (inode->i_op == &bpf_map_iops)
		*type = BPF_TYPE_MAP;
	else
		return -EACCES;

	return 0;
}

static bool bpf_dname_reserved(const struct dentry *dentry)
{
	return strchr(dentry->d_name.name, '.');
}

static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
{
	struct inode *inode;

	if (bpf_dname_reserved(dentry))
		return -EPERM;

	inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
	if (IS_ERR(inode))
		return PTR_ERR(inode);

	inode->i_op = &bpf_dir_iops;
	inode->i_fop = &simple_dir_operations;

	inc_nlink(inode);
	inc_nlink(dir);

	d_instantiate(dentry, inode);
	dget(dentry);

	return 0;
}

static int bpf_mkobj_ops(struct inode *dir, struct dentry *dentry,
			 umode_t mode, const struct inode_operations *iops)
{
	struct inode *inode;

	if (bpf_dname_reserved(dentry))
		return -EPERM;

	inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFREG);
	if (IS_ERR(inode))
		return PTR_ERR(inode);

	inode->i_op = iops;
	inode->i_private = dentry->d_fsdata;

	d_instantiate(dentry, inode);
	dget(dentry);

	return 0;
}

static int bpf_mkobj(struct inode *dir, struct dentry *dentry, umode_t mode,
		     dev_t devt)
{
	enum bpf_type type = MINOR(devt);

	if (MAJOR(devt) != UNNAMED_MAJOR || !S_ISREG(mode) ||
	    dentry->d_fsdata == NULL)
		return -EPERM;

	switch (type) {
	case BPF_TYPE_PROG:
		return bpf_mkobj_ops(dir, dentry, mode, &bpf_prog_iops);
	case BPF_TYPE_MAP:
		return bpf_mkobj_ops(dir, dentry, mode, &bpf_map_iops);
	default:
		return -EPERM;
	}
}

static const struct inode_operations bpf_dir_iops = {
	.lookup		= simple_lookup,
	.mknod		= bpf_mkobj,
	.mkdir		= bpf_mkdir,
	.rmdir		= simple_rmdir,
	.unlink		= simple_unlink,
};

static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
			  enum bpf_type type)
{
	struct dentry *dentry;
	struct inode *dir;
	struct path path;
	umode_t mode;
	dev_t devt;
	int ret;

	dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
	if (IS_ERR(dentry))
		return PTR_ERR(dentry);

	mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
	devt = MKDEV(UNNAMED_MAJOR, type);

	ret = security_path_mknod(&path, dentry, mode, devt);
	if (ret)
		goto out;

	dir = d_inode(path.dentry);
	if (dir->i_op != &bpf_dir_iops) {
		ret = -EPERM;
		goto out;
	}

	dentry->d_fsdata = raw;
	ret = vfs_mknod(dir, dentry, mode, devt);
	dentry->d_fsdata = NULL;
out:
	done_path_create(&path, dentry);
	return ret;
}

int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
{
	struct filename *pname;
	enum bpf_type type;
	void *raw;
	int ret;

	pname = getname(pathname);
	if (IS_ERR(pname))
		return PTR_ERR(pname);

	raw = bpf_fd_probe_obj(ufd, &type);
	if (IS_ERR(raw)) {
		ret = PTR_ERR(raw);
		goto out;
	}

	ret = bpf_obj_do_pin(pname, raw, type);
	if (ret != 0)
		bpf_any_put(raw, type);
out:
	putname(pname);
	return ret;
}

static void *bpf_obj_do_get(const struct filename *pathname,
			    enum bpf_type *type)
{
	struct inode *inode;
	struct path path;
	void *raw;
	int ret;

	ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
	if (ret)
		return ERR_PTR(ret);

	inode = d_backing_inode(path.dentry);
	ret = inode_permission(inode, MAY_WRITE);
	if (ret)
		goto out;

	ret = bpf_inode_type(inode, type);
	if (ret)
		goto out;

	raw = bpf_any_get(inode->i_private, *type);
	touch_atime(&path);

	path_put(&path);
	return raw;
out:
	path_put(&path);
	return ERR_PTR(ret);
}

int bpf_obj_get_user(const char __user *pathname)
{
	enum bpf_type type = BPF_TYPE_UNSPEC;
	struct filename *pname;
	int ret = -ENOENT;
	void *raw;

	pname = getname(pathname);
	if (IS_ERR(pname))
		return PTR_ERR(pname);

	raw = bpf_obj_do_get(pname, &type);
	if (IS_ERR(raw)) {
		ret = PTR_ERR(raw);
		goto out;
	}

	if (type == BPF_TYPE_PROG)
		ret = bpf_prog_new_fd(raw);
	else if (type == BPF_TYPE_MAP)
		ret = bpf_map_new_fd(raw);
	else
		goto out;

	if (ret < 0)
		bpf_any_put(raw, type);
out:
	putname(pname);
	return ret;
}

static void bpf_evict_inode(struct inode *inode)
{
	enum bpf_type type;

	truncate_inode_pages_final(&inode->i_data);
	clear_inode(inode);

	if (!bpf_inode_type(inode, &type))
		bpf_any_put(inode->i_private, type);
}

static const struct super_operations bpf_super_ops = {
	.statfs		= simple_statfs,
	.drop_inode	= generic_delete_inode,
	.evict_inode	= bpf_evict_inode,
};

static int bpf_fill_super(struct super_block *sb, void *data, int silent)
{
	static struct tree_descr bpf_rfiles[] = { { "" } };
	struct inode *inode;
	int ret;

	ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
	if (ret)
		return ret;

	sb->s_op = &bpf_super_ops;

	inode = sb->s_root->d_inode;
	inode->i_op = &bpf_dir_iops;
	inode->i_mode &= ~S_IALLUGO;
	inode->i_mode |= S_ISVTX | S_IRWXUGO;

	return 0;
}

static struct dentry *bpf_mount(struct file_system_type *type, int flags,
				const char *dev_name, void *data)
{
	return mount_ns(type, flags, current->nsproxy->mnt_ns, bpf_fill_super);
}

static struct file_system_type bpf_fs_type = {
	.owner		= THIS_MODULE,
	.name		= "bpf",
	.mount		= bpf_mount,
	.kill_sb	= kill_litter_super,
	.fs_flags	= FS_USERNS_MOUNT,
};

MODULE_ALIAS_FS("bpf");

static int __init bpf_init(void)
{
	int ret;

	ret = sysfs_create_mount_point(fs_kobj, "bpf");
	if (ret)
		return ret;

	ret = register_filesystem(&bpf_fs_type);
	if (ret)
		sysfs_remove_mount_point(fs_kobj, "bpf");

	return ret;
}
fs_initcall(bpf_init);
Commit	Line	Data
b2197755 DB	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/module.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/filter.h>
	22	#include <linux/bpf.h>
	23
	24	enum bpf_type {
	25	BPF_TYPE_UNSPEC = 0,
	26	BPF_TYPE_PROG,
	27	BPF_TYPE_MAP,
	28	};
	29
	30	static void bpf_any_get(void raw, enum bpf_type type)
	31	{
	32	switch (type) {
	33	case BPF_TYPE_PROG:
	34	atomic_inc(&((struct bpf_prog *)raw)->aux->refcnt);
	35	break;
	36	case BPF_TYPE_MAP:
c9da161c	37	bpf_map_inc(raw, true);
b2197755 DB	38	break;
	39	default:
	40	WARN_ON_ONCE(1);
	41	break;
	42	}
	43
	44	return raw;
	45	}
	46
	47	static void bpf_any_put(void *raw, enum bpf_type type)
	48	{
	49	switch (type) {
	50	case BPF_TYPE_PROG:
	51	bpf_prog_put(raw);
	52	break;
	53	case BPF_TYPE_MAP:
c9da161c	54	bpf_map_put_with_uref(raw);
b2197755 DB	55	break;
	56	default:
	57	WARN_ON_ONCE(1);
	58	break;
	59	}
	60	}
	61
	62	static void bpf_fd_probe_obj(u32 ufd, enum bpf_type type)
	63	{
	64	void *raw;
	65
	66	*type = BPF_TYPE_MAP;
c9da161c	67	raw = bpf_map_get_with_uref(ufd);
b2197755 DB	68	if (IS_ERR(raw)) {
	69	*type = BPF_TYPE_PROG;
	70	raw = bpf_prog_get(ufd);
	71	}
	72
	73	return raw;
	74	}
	75
	76	static const struct inode_operations bpf_dir_iops;
	77
	78	static const struct inode_operations bpf_prog_iops = { };
	79	static const struct inode_operations bpf_map_iops = { };
	80
	81	static struct inode bpf_get_inode(struct super_block sb,
	82	const struct inode *dir,
	83	umode_t mode)
	84	{
	85	struct inode *inode;
	86
	87	switch (mode & S_IFMT) {
	88	case S_IFDIR:
	89	case S_IFREG:
	90	break;
	91	default:
	92	return ERR_PTR(-EINVAL);
	93	}
	94
	95	inode = new_inode(sb);
	96	if (!inode)
	97	return ERR_PTR(-ENOSPC);
	98
	99	inode->i_ino = get_next_ino();
	100	inode->i_atime = CURRENT_TIME;
	101	inode->i_mtime = inode->i_atime;
	102	inode->i_ctime = inode->i_atime;
	103
	104	inode_init_owner(inode, dir, mode);
	105
	106	return inode;
	107	}
	108
	109	static int bpf_inode_type(const struct inode inode, enum bpf_type type)
	110	{
	111	*type = BPF_TYPE_UNSPEC;
	112	if (inode->i_op == &bpf_prog_iops)
	113	*type = BPF_TYPE_PROG;
	114	else if (inode->i_op == &bpf_map_iops)
	115	*type = BPF_TYPE_MAP;
	116	else
	117	return -EACCES;
	118
	119	return 0;
	120	}
	121
	122	static bool bpf_dname_reserved(const struct dentry *dentry)
	123	{
	124	return strchr(dentry->d_name.name, '.');
	125	}
	126
	127	static int bpf_mkdir(struct inode dir, struct dentry dentry, umode_t mode)
	128	{
	129	struct inode *inode;
	130
	131	if (bpf_dname_reserved(dentry))
132	return -EPERM;
133
134	inode = bpf_get_inode(dir->i_sb, dir, mode \| S_IFDIR);
135	if (IS_ERR(inode))
136	return PTR_ERR(inode);
137
138	inode->i_op = &bpf_dir_iops;
139	inode->i_fop = &simple_dir_operations;
140
141	inc_nlink(inode);
142	inc_nlink(dir);
143
144	d_instantiate(dentry, inode);
145	dget(dentry);
146
147	return 0;
148	}
149
150	static int bpf_mkobj_ops(struct inode dir, struct dentry dentry,
151	umode_t mode, const struct inode_operations *iops)
152	{
153	struct inode *inode;
154
155	if (bpf_dname_reserved(dentry))
156	return -EPERM;
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	d_instantiate(dentry, inode);
166	dget(dentry);
167
168	return 0;
169	}
170
171	static int bpf_mkobj(struct inode dir, struct dentry dentry, umode_t mode,
172	dev_t devt)
173	{
174	enum bpf_type type = MINOR(devt);
175
176	if (MAJOR(devt) != UNNAMED_MAJOR \|\| !S_ISREG(mode) \|\|
177	dentry->d_fsdata == NULL)
178	return -EPERM;
179
180	switch (type) {
181	case BPF_TYPE_PROG:
182	return bpf_mkobj_ops(dir, dentry, mode, &bpf_prog_iops);
183	case BPF_TYPE_MAP:
184	return bpf_mkobj_ops(dir, dentry, mode, &bpf_map_iops);
185	default:
186	return -EPERM;
187	}
188	}
189
190	static const struct inode_operations bpf_dir_iops = {
191	.lookup = simple_lookup,
192	.mknod = bpf_mkobj,
193	.mkdir = bpf_mkdir,
194	.rmdir = simple_rmdir,
195	.unlink = simple_unlink,
196	};
197
198	static int bpf_obj_do_pin(const struct filename pathname, void raw,
199	enum bpf_type type)
200	{
201	struct dentry *dentry;
202	struct inode *dir;
203	struct path path;
204	umode_t mode;
205	dev_t devt;
206	int ret;
207
208	dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
209	if (IS_ERR(dentry))
210	return PTR_ERR(dentry);
211
212	mode = S_IFREG \| ((S_IRUSR \| S_IWUSR) & ~current_umask());
213	devt = MKDEV(UNNAMED_MAJOR, type);
214
215	ret = security_path_mknod(&path, dentry, mode, devt);
216	if (ret)
217	goto out;
218
219	dir = d_inode(path.dentry);
220	if (dir->i_op != &bpf_dir_iops) {
221	ret = -EPERM;
222	goto out;
223	}
224
225	dentry->d_fsdata = raw;
226	ret = vfs_mknod(dir, dentry, mode, devt);
227	dentry->d_fsdata = NULL;
228	out:
229	done_path_create(&path, dentry);
230	return ret;
231	}
232
233	int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
234	{
235	struct filename *pname;
236	enum bpf_type type;
237	void *raw;
238	int ret;
239
240	pname = getname(pathname);
241	if (IS_ERR(pname))
242	return PTR_ERR(pname);
243
244	raw = bpf_fd_probe_obj(ufd, &type);
245	if (IS_ERR(raw)) {
246	ret = PTR_ERR(raw);
247	goto out;
248	}
249
250	ret = bpf_obj_do_pin(pname, raw, type);
251	if (ret != 0)
252	bpf_any_put(raw, type);
253	out:
254	putname(pname);
255	return ret;
256	}
257
258	static void bpf_obj_do_get(const struct filename pathname,
259	enum bpf_type *type)
260	{
261	struct inode *inode;
262	struct path path;
263	void *raw;
264	int ret;
265
266	ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
267	if (ret)
268	return ERR_PTR(ret);
269
270	inode = d_backing_inode(path.dentry);
271	ret = inode_permission(inode, MAY_WRITE);
272	if (ret)
273	goto out;
274
275	ret = bpf_inode_type(inode, type);
276	if (ret)
277	goto out;
278
279	raw = bpf_any_get(inode->i_private, *type);
280	touch_atime(&path);
281
282	path_put(&path);
283	return raw;
284	out:
285	path_put(&path);
286	return ERR_PTR(ret);
287	}
288
289	int bpf_obj_get_user(const char __user *pathname)
290	{
291	enum bpf_type type = BPF_TYPE_UNSPEC;
292	struct filename *pname;
293	int ret = -ENOENT;
294	void *raw;
295
296	pname = getname(pathname);
297	if (IS_ERR(pname))
298	return PTR_ERR(pname);
299
300	raw = bpf_obj_do_get(pname, &type);
301	if (IS_ERR(raw)) {
302	ret = PTR_ERR(raw);
303	goto out;
304	}
305
306	if (type == BPF_TYPE_PROG)
307	ret = bpf_prog_new_fd(raw);
308	else if (type == BPF_TYPE_MAP)
309	ret = bpf_map_new_fd(raw);
310	else
311	goto out;
312
313	if (ret < 0)
314	bpf_any_put(raw, type);
315	out:
316	putname(pname);
317	return ret;
318	}
319
320	static void bpf_evict_inode(struct inode *inode)
321	{
322	enum bpf_type type;
323
324	truncate_inode_pages_final(&inode->i_data);
325	clear_inode(inode);
326
327	if (!bpf_inode_type(inode, &type))
328	bpf_any_put(inode->i_private, type);
329	}
330
331	static const struct super_operations bpf_super_ops = {
332	.statfs = simple_statfs,
333	.drop_inode = generic_delete_inode,
334	.evict_inode = bpf_evict_inode,
335	};
336
337	static int bpf_fill_super(struct super_block sb, void data, int silent)
338	{
339	static struct tree_descr bpf_rfiles[] = { { "" } };
340	struct inode *inode;
341	int ret;
342
343	ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
344	if (ret)
345	return ret;
346
347	sb->s_op = &bpf_super_ops;
348
349	inode = sb->s_root->d_inode;
350	inode->i_op = &bpf_dir_iops;
351	inode->i_mode &= ~S_IALLUGO;
352	inode->i_mode \|= S_ISVTX \| S_IRWXUGO;
353
354	return 0;
355	}
356
357	static struct dentry bpf_mount(struct file_system_type type, int flags,
358	const char dev_name, void data)
359	{
360	return mount_ns(type, flags, current->nsproxy->mnt_ns, bpf_fill_super);
361	}
362
363	static struct file_system_type bpf_fs_type = {
364	.owner = THIS_MODULE,
365	.name = "bpf",
366	.mount = bpf_mount,
367	.kill_sb = kill_litter_super,
368	.fs_flags = FS_USERNS_MOUNT,
369	};
370
371	MODULE_ALIAS_FS("bpf");
372
373	static int __init bpf_init(void)
374	{
375	int ret;
376
377	ret = sysfs_create_mount_point(fs_kobj, "bpf");
378	if (ret)
379	return ret;
380
381	ret = register_filesystem(&bpf_fs_type);
382	if (ret)
383	sysfs_remove_mount_point(fs_kobj, "bpf");
384
385	return ret;
386	}
387	fs_initcall(bpf_init);