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/module.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 atomic_inc(&((struct bpf_prog *)raw)->aux->refcnt);
37 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 bool bpf_dname_reserved(const struct dentry *dentry)
124 return strchr(dentry->d_name.name, '.');
127 static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
131 if (bpf_dname_reserved(dentry))
134 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
136 return PTR_ERR(inode);
138 inode->i_op = &bpf_dir_iops;
139 inode->i_fop = &simple_dir_operations;
144 d_instantiate(dentry, inode);
150 static int bpf_mkobj_ops(struct inode *dir, struct dentry *dentry,
151 umode_t mode, const struct inode_operations *iops)
155 if (bpf_dname_reserved(dentry))
158 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFREG);
160 return PTR_ERR(inode);
163 inode->i_private = dentry->d_fsdata;
165 d_instantiate(dentry, inode);
171 static int bpf_mkobj(struct inode *dir, struct dentry *dentry, umode_t mode,
174 enum bpf_type type = MINOR(devt);
176 if (MAJOR(devt) != UNNAMED_MAJOR || !S_ISREG(mode) ||
177 dentry->d_fsdata == NULL)
182 return bpf_mkobj_ops(dir, dentry, mode, &bpf_prog_iops);
184 return bpf_mkobj_ops(dir, dentry, mode, &bpf_map_iops);
190 static int bpf_link(struct dentry *old_dentry, struct inode *dir,
191 struct dentry *new_dentry)
193 if (bpf_dname_reserved(new_dentry))
196 return simple_link(old_dentry, dir, new_dentry);
199 static int bpf_rename(struct inode *old_dir, struct dentry *old_dentry,
200 struct inode *new_dir, struct dentry *new_dentry)
202 if (bpf_dname_reserved(new_dentry))
205 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
208 static const struct inode_operations bpf_dir_iops = {
209 .lookup = simple_lookup,
212 .rmdir = simple_rmdir,
213 .rename = bpf_rename,
215 .unlink = simple_unlink,
218 static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
221 struct dentry *dentry;
228 dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
230 return PTR_ERR(dentry);
232 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
233 devt = MKDEV(UNNAMED_MAJOR, type);
235 ret = security_path_mknod(&path, dentry, mode, devt);
239 dir = d_inode(path.dentry);
240 if (dir->i_op != &bpf_dir_iops) {
245 dentry->d_fsdata = raw;
246 ret = vfs_mknod(dir, dentry, mode, devt);
247 dentry->d_fsdata = NULL;
249 done_path_create(&path, dentry);
253 int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
255 struct filename *pname;
260 pname = getname(pathname);
262 return PTR_ERR(pname);
264 raw = bpf_fd_probe_obj(ufd, &type);
270 ret = bpf_obj_do_pin(pname, raw, type);
272 bpf_any_put(raw, type);
278 static void *bpf_obj_do_get(const struct filename *pathname,
286 ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
290 inode = d_backing_inode(path.dentry);
291 ret = inode_permission(inode, MAY_WRITE);
295 ret = bpf_inode_type(inode, type);
299 raw = bpf_any_get(inode->i_private, *type);
309 int bpf_obj_get_user(const char __user *pathname)
311 enum bpf_type type = BPF_TYPE_UNSPEC;
312 struct filename *pname;
316 pname = getname(pathname);
318 return PTR_ERR(pname);
320 raw = bpf_obj_do_get(pname, &type);
326 if (type == BPF_TYPE_PROG)
327 ret = bpf_prog_new_fd(raw);
328 else if (type == BPF_TYPE_MAP)
329 ret = bpf_map_new_fd(raw);
334 bpf_any_put(raw, type);
340 static void bpf_evict_inode(struct inode *inode)
344 truncate_inode_pages_final(&inode->i_data);
347 if (!bpf_inode_type(inode, &type))
348 bpf_any_put(inode->i_private, type);
351 static const struct super_operations bpf_super_ops = {
352 .statfs = simple_statfs,
353 .drop_inode = generic_delete_inode,
354 .evict_inode = bpf_evict_inode,
357 static int bpf_fill_super(struct super_block *sb, void *data, int silent)
359 static struct tree_descr bpf_rfiles[] = { { "" } };
363 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
367 sb->s_op = &bpf_super_ops;
369 inode = sb->s_root->d_inode;
370 inode->i_op = &bpf_dir_iops;
371 inode->i_mode &= ~S_IALLUGO;
372 inode->i_mode |= S_ISVTX | S_IRWXUGO;
377 static struct dentry *bpf_mount(struct file_system_type *type, int flags,
378 const char *dev_name, void *data)
380 return mount_ns(type, flags, current->nsproxy->mnt_ns, bpf_fill_super);
383 static struct file_system_type bpf_fs_type = {
384 .owner = THIS_MODULE,
387 .kill_sb = kill_litter_super,
388 .fs_flags = FS_USERNS_MOUNT,
391 MODULE_ALIAS_FS("bpf");
393 static int __init bpf_init(void)
397 ret = sysfs_create_mount_point(fs_kobj, "bpf");
401 ret = register_filesystem(&bpf_fs_type);
403 sysfs_remove_mount_point(fs_kobj, "bpf");
407 fs_initcall(bpf_init);