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
bpf_any_get(void * raw,enum bpf_type type)30 static void *bpf_any_get(void *raw, enum bpf_type type)
31 {
32 switch (type) {
33 case BPF_TYPE_PROG:
34 raw = bpf_prog_inc(raw);
35 break;
36 case BPF_TYPE_MAP:
37 raw = bpf_map_inc(raw, true);
38 break;
39 default:
40 WARN_ON_ONCE(1);
41 break;
42 }
43
44 return raw;
45 }
46
bpf_any_put(void * raw,enum bpf_type type)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:
54 bpf_map_put_with_uref(raw);
55 break;
56 default:
57 WARN_ON_ONCE(1);
58 break;
59 }
60 }
61
bpf_fd_probe_obj(u32 ufd,enum bpf_type * type)62 static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
63 {
64 void *raw;
65
66 *type = BPF_TYPE_MAP;
67 raw = bpf_map_get_with_uref(ufd);
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
bpf_get_inode(struct super_block * sb,const struct inode * dir,umode_t mode)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
bpf_inode_type(const struct inode * inode,enum bpf_type * type)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
bpf_dname_reserved(const struct dentry * dentry)122 static bool bpf_dname_reserved(const struct dentry *dentry)
123 {
124 return strchr(dentry->d_name.name, '.');
125 }
126
bpf_mkdir(struct inode * dir,struct dentry * dentry,umode_t mode)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
bpf_mkobj_ops(struct inode * dir,struct dentry * dentry,umode_t mode,const struct inode_operations * iops)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
bpf_mkobj(struct inode * dir,struct dentry * dentry,umode_t mode,dev_t devt)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
bpf_obj_do_pin(const struct filename * pathname,void * raw,enum bpf_type type)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
bpf_obj_pin_user(u32 ufd,const char __user * pathname)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
bpf_obj_do_get(const struct filename * pathname,enum bpf_type * type)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 if (!IS_ERR(raw))
281 touch_atime(&path);
282
283 path_put(&path);
284 return raw;
285 out:
286 path_put(&path);
287 return ERR_PTR(ret);
288 }
289
bpf_obj_get_user(const char __user * pathname)290 int bpf_obj_get_user(const char __user *pathname)
291 {
292 enum bpf_type type = BPF_TYPE_UNSPEC;
293 struct filename *pname;
294 int ret = -ENOENT;
295 void *raw;
296
297 pname = getname(pathname);
298 if (IS_ERR(pname))
299 return PTR_ERR(pname);
300
301 raw = bpf_obj_do_get(pname, &type);
302 if (IS_ERR(raw)) {
303 ret = PTR_ERR(raw);
304 goto out;
305 }
306
307 if (type == BPF_TYPE_PROG)
308 ret = bpf_prog_new_fd(raw);
309 else if (type == BPF_TYPE_MAP)
310 ret = bpf_map_new_fd(raw);
311 else
312 goto out;
313
314 if (ret < 0)
315 bpf_any_put(raw, type);
316 out:
317 putname(pname);
318 return ret;
319 }
320
bpf_evict_inode(struct inode * inode)321 static void bpf_evict_inode(struct inode *inode)
322 {
323 enum bpf_type type;
324
325 truncate_inode_pages_final(&inode->i_data);
326 clear_inode(inode);
327
328 if (!bpf_inode_type(inode, &type))
329 bpf_any_put(inode->i_private, type);
330 }
331
332 static const struct super_operations bpf_super_ops = {
333 .statfs = simple_statfs,
334 .drop_inode = generic_delete_inode,
335 .evict_inode = bpf_evict_inode,
336 };
337
bpf_fill_super(struct super_block * sb,void * data,int silent)338 static int bpf_fill_super(struct super_block *sb, void *data, int silent)
339 {
340 static struct tree_descr bpf_rfiles[] = { { "" } };
341 struct inode *inode;
342 int ret;
343
344 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
345 if (ret)
346 return ret;
347
348 sb->s_op = &bpf_super_ops;
349
350 inode = sb->s_root->d_inode;
351 inode->i_op = &bpf_dir_iops;
352 inode->i_mode &= ~S_IALLUGO;
353 inode->i_mode |= S_ISVTX | S_IRWXUGO;
354
355 return 0;
356 }
357
bpf_mount(struct file_system_type * type,int flags,const char * dev_name,void * data)358 static struct dentry *bpf_mount(struct file_system_type *type, int flags,
359 const char *dev_name, void *data)
360 {
361 return mount_nodev(type, flags, data, bpf_fill_super);
362 }
363
364 static struct file_system_type bpf_fs_type = {
365 .owner = THIS_MODULE,
366 .name = "bpf",
367 .mount = bpf_mount,
368 .kill_sb = kill_litter_super,
369 };
370
371 MODULE_ALIAS_FS("bpf");
372
bpf_init(void)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);
388