2 * Minimal file system backend for holding eBPF maps and programs,
3 * used by bpf(2) object pinning.
7 * Daniel Borkmann <daniel@iogearbox.net>
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.
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>
20 #include <linux/kdev_t.h>
21 #include <linux/filter.h>
22 #include <linux/bpf.h>
30 static void *bpf_any_get(void *raw, enum bpf_type type)
34 raw = bpf_prog_inc(raw);
37 raw = bpf_map_inc(raw, true);
47 static void bpf_any_put(void *raw, enum bpf_type type)
54 bpf_map_put_with_uref(raw);
62 static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
67 raw = bpf_map_get_with_uref(ufd);
69 *type = BPF_TYPE_PROG;
70 raw = bpf_prog_get(ufd);
76 static const struct inode_operations bpf_dir_iops;
78 static const struct inode_operations bpf_prog_iops = { };
79 static const struct inode_operations bpf_map_iops = { };
81 static struct inode *bpf_get_inode(struct super_block *sb,
82 const struct inode *dir,
87 switch (mode & S_IFMT) {
92 return ERR_PTR(-EINVAL);
95 inode = new_inode(sb);
97 return ERR_PTR(-ENOSPC);
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;
104 inode_init_owner(inode, dir, mode);
109 static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
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;
122 static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
126 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
128 return PTR_ERR(inode);
130 inode->i_op = &bpf_dir_iops;
131 inode->i_fop = &simple_dir_operations;
136 d_instantiate(dentry, inode);
142 static int bpf_mkobj_ops(struct inode *dir, struct dentry *dentry,
143 umode_t mode, const struct inode_operations *iops)
147 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFREG);
149 return PTR_ERR(inode);
152 inode->i_private = dentry->d_fsdata;
154 d_instantiate(dentry, inode);
160 static int bpf_mkobj(struct inode *dir, struct dentry *dentry, umode_t mode,
163 enum bpf_type type = MINOR(devt);
165 if (MAJOR(devt) != UNNAMED_MAJOR || !S_ISREG(mode) ||
166 dentry->d_fsdata == NULL)
171 return bpf_mkobj_ops(dir, dentry, mode, &bpf_prog_iops);
173 return bpf_mkobj_ops(dir, dentry, mode, &bpf_map_iops);
179 static struct dentry *
180 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
182 if (strchr(dentry->d_name.name, '.'))
183 return ERR_PTR(-EPERM);
184 return simple_lookup(dir, dentry, flags);
187 static const struct inode_operations bpf_dir_iops = {
188 .lookup = bpf_lookup,
191 .rmdir = simple_rmdir,
192 .rename = simple_rename,
194 .unlink = simple_unlink,
197 static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
200 struct dentry *dentry;
207 dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
209 return PTR_ERR(dentry);
211 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
212 devt = MKDEV(UNNAMED_MAJOR, type);
214 ret = security_path_mknod(&path, dentry, mode, devt);
218 dir = d_inode(path.dentry);
219 if (dir->i_op != &bpf_dir_iops) {
224 dentry->d_fsdata = raw;
225 ret = vfs_mknod(dir, dentry, mode, devt);
226 dentry->d_fsdata = NULL;
228 done_path_create(&path, dentry);
232 int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
234 struct filename *pname;
239 pname = getname(pathname);
241 return PTR_ERR(pname);
243 raw = bpf_fd_probe_obj(ufd, &type);
249 ret = bpf_obj_do_pin(pname, raw, type);
251 bpf_any_put(raw, type);
257 static void *bpf_obj_do_get(const struct filename *pathname,
265 ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
269 inode = d_backing_inode(path.dentry);
270 ret = inode_permission(inode, MAY_WRITE);
274 ret = bpf_inode_type(inode, type);
278 raw = bpf_any_get(inode->i_private, *type);
289 int bpf_obj_get_user(const char __user *pathname)
291 enum bpf_type type = BPF_TYPE_UNSPEC;
292 struct filename *pname;
296 pname = getname(pathname);
298 return PTR_ERR(pname);
300 raw = bpf_obj_do_get(pname, &type);
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);
314 bpf_any_put(raw, type);
320 static void bpf_evict_inode(struct inode *inode)
324 truncate_inode_pages_final(&inode->i_data);
327 if (!bpf_inode_type(inode, &type))
328 bpf_any_put(inode->i_private, type);
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,
337 static int bpf_fill_super(struct super_block *sb, void *data, int silent)
339 static struct tree_descr bpf_rfiles[] = { { "" } };
343 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
347 sb->s_op = &bpf_super_ops;
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;
357 static struct dentry *bpf_mount(struct file_system_type *type, int flags,
358 const char *dev_name, void *data)
360 return mount_nodev(type, flags, data, bpf_fill_super);
363 static struct file_system_type bpf_fs_type = {
364 .owner = THIS_MODULE,
367 .kill_sb = kill_litter_super,
370 static int __init bpf_init(void)
374 ret = sysfs_create_mount_point(fs_kobj, "bpf");
378 ret = register_filesystem(&bpf_fs_type);
380 sysfs_remove_mount_point(fs_kobj, "bpf");
384 fs_initcall(bpf_init);