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) {
93 return ERR_PTR(-EINVAL);
96 inode = new_inode(sb);
98 return ERR_PTR(-ENOSPC);
100 inode->i_ino = get_next_ino();
101 inode->i_atime = current_time(inode);
102 inode->i_mtime = inode->i_atime;
103 inode->i_ctime = inode->i_atime;
105 inode_init_owner(inode, dir, mode);
110 static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
112 *type = BPF_TYPE_UNSPEC;
113 if (inode->i_op == &bpf_prog_iops)
114 *type = BPF_TYPE_PROG;
115 else if (inode->i_op == &bpf_map_iops)
116 *type = BPF_TYPE_MAP;
123 static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
126 d_instantiate(dentry, inode);
129 dir->i_mtime = current_time(dir);
130 dir->i_ctime = dir->i_mtime;
133 static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
137 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
139 return PTR_ERR(inode);
141 inode->i_op = &bpf_dir_iops;
142 inode->i_fop = &simple_dir_operations;
147 bpf_dentry_finalize(dentry, inode, dir);
151 static int bpf_mkobj_ops(struct inode *dir, struct dentry *dentry,
152 umode_t mode, const struct inode_operations *iops)
156 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFREG);
158 return PTR_ERR(inode);
161 inode->i_private = dentry->d_fsdata;
163 bpf_dentry_finalize(dentry, inode, dir);
167 static int bpf_mkobj(struct inode *dir, struct dentry *dentry, umode_t mode,
170 enum bpf_type type = MINOR(devt);
172 if (MAJOR(devt) != UNNAMED_MAJOR || !S_ISREG(mode) ||
173 dentry->d_fsdata == NULL)
178 return bpf_mkobj_ops(dir, dentry, mode, &bpf_prog_iops);
180 return bpf_mkobj_ops(dir, dentry, mode, &bpf_map_iops);
186 static struct dentry *
187 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
189 if (strchr(dentry->d_name.name, '.'))
190 return ERR_PTR(-EPERM);
192 return simple_lookup(dir, dentry, flags);
195 static int bpf_symlink(struct inode *dir, struct dentry *dentry,
198 char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
204 inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
207 return PTR_ERR(inode);
210 inode->i_op = &simple_symlink_inode_operations;
211 inode->i_link = link;
213 bpf_dentry_finalize(dentry, inode, dir);
217 static const struct inode_operations bpf_dir_iops = {
218 .lookup = bpf_lookup,
221 .symlink = bpf_symlink,
222 .rmdir = simple_rmdir,
223 .rename = simple_rename,
225 .unlink = simple_unlink,
228 static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
231 struct dentry *dentry;
238 dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
240 return PTR_ERR(dentry);
242 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
243 devt = MKDEV(UNNAMED_MAJOR, type);
245 ret = security_path_mknod(&path, dentry, mode, devt);
249 dir = d_inode(path.dentry);
250 if (dir->i_op != &bpf_dir_iops) {
255 dentry->d_fsdata = raw;
256 ret = vfs_mknod(dir, dentry, mode, devt);
257 dentry->d_fsdata = NULL;
259 done_path_create(&path, dentry);
263 int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
265 struct filename *pname;
270 pname = getname(pathname);
272 return PTR_ERR(pname);
274 raw = bpf_fd_probe_obj(ufd, &type);
280 ret = bpf_obj_do_pin(pname, raw, type);
282 bpf_any_put(raw, type);
288 static void *bpf_obj_do_get(const struct filename *pathname,
296 ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
300 inode = d_backing_inode(path.dentry);
301 ret = inode_permission(inode, MAY_WRITE);
305 ret = bpf_inode_type(inode, type);
309 raw = bpf_any_get(inode->i_private, *type);
320 int bpf_obj_get_user(const char __user *pathname)
322 enum bpf_type type = BPF_TYPE_UNSPEC;
323 struct filename *pname;
327 pname = getname(pathname);
329 return PTR_ERR(pname);
331 raw = bpf_obj_do_get(pname, &type);
337 if (type == BPF_TYPE_PROG)
338 ret = bpf_prog_new_fd(raw);
339 else if (type == BPF_TYPE_MAP)
340 ret = bpf_map_new_fd(raw);
345 bpf_any_put(raw, type);
351 static void bpf_evict_inode(struct inode *inode)
355 truncate_inode_pages_final(&inode->i_data);
358 if (S_ISLNK(inode->i_mode))
359 kfree(inode->i_link);
360 if (!bpf_inode_type(inode, &type))
361 bpf_any_put(inode->i_private, type);
364 static const struct super_operations bpf_super_ops = {
365 .statfs = simple_statfs,
366 .drop_inode = generic_delete_inode,
367 .evict_inode = bpf_evict_inode,
370 static int bpf_fill_super(struct super_block *sb, void *data, int silent)
372 static struct tree_descr bpf_rfiles[] = { { "" } };
376 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
380 sb->s_op = &bpf_super_ops;
382 inode = sb->s_root->d_inode;
383 inode->i_op = &bpf_dir_iops;
384 inode->i_mode &= ~S_IALLUGO;
385 inode->i_mode |= S_ISVTX | S_IRWXUGO;
390 static struct dentry *bpf_mount(struct file_system_type *type, int flags,
391 const char *dev_name, void *data)
393 return mount_nodev(type, flags, data, bpf_fill_super);
396 static struct file_system_type bpf_fs_type = {
397 .owner = THIS_MODULE,
400 .kill_sb = kill_litter_super,
403 static int __init bpf_init(void)
407 ret = sysfs_create_mount_point(fs_kobj, "bpf");
411 ret = register_filesystem(&bpf_fs_type);
413 sysfs_remove_mount_point(fs_kobj, "bpf");
417 fs_initcall(bpf_init);