2 * Minimal file system backend for holding eBPF maps and programs,
3 * used by bpf(2) object pinning.
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/parser.h>
22 #include <linux/filter.h>
23 #include <linux/bpf.h>
31 static void *bpf_any_get(void *raw, enum bpf_type type)
35 raw = bpf_prog_inc(raw);
38 raw = bpf_map_inc(raw, true);
48 static void bpf_any_put(void *raw, enum bpf_type type)
55 bpf_map_put_with_uref(raw);
63 static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
68 raw = bpf_map_get_with_uref(ufd);
70 *type = BPF_TYPE_PROG;
71 raw = bpf_prog_get(ufd);
77 static const struct inode_operations bpf_dir_iops;
79 static const struct inode_operations bpf_prog_iops = { };
80 static const struct inode_operations bpf_map_iops = { };
82 static struct inode *bpf_get_inode(struct super_block *sb,
83 const struct inode *dir,
88 switch (mode & S_IFMT) {
94 return ERR_PTR(-EINVAL);
97 inode = new_inode(sb);
99 return ERR_PTR(-ENOSPC);
101 inode->i_ino = get_next_ino();
102 inode->i_atime = current_time(inode);
103 inode->i_mtime = inode->i_atime;
104 inode->i_ctime = inode->i_atime;
106 inode_init_owner(inode, dir, mode);
111 static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
113 *type = BPF_TYPE_UNSPEC;
114 if (inode->i_op == &bpf_prog_iops)
115 *type = BPF_TYPE_PROG;
116 else if (inode->i_op == &bpf_map_iops)
117 *type = BPF_TYPE_MAP;
124 static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
127 d_instantiate(dentry, inode);
130 dir->i_mtime = current_time(dir);
131 dir->i_ctime = dir->i_mtime;
134 static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
138 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
140 return PTR_ERR(inode);
142 inode->i_op = &bpf_dir_iops;
143 inode->i_fop = &simple_dir_operations;
148 bpf_dentry_finalize(dentry, inode, dir);
152 static int bpf_mkobj_ops(struct inode *dir, struct dentry *dentry,
153 umode_t mode, const struct inode_operations *iops)
157 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFREG);
159 return PTR_ERR(inode);
162 inode->i_private = dentry->d_fsdata;
164 bpf_dentry_finalize(dentry, inode, dir);
168 static int bpf_mkobj(struct inode *dir, struct dentry *dentry, umode_t mode,
171 enum bpf_type type = MINOR(devt);
173 if (MAJOR(devt) != UNNAMED_MAJOR || !S_ISREG(mode) ||
174 dentry->d_fsdata == NULL)
179 return bpf_mkobj_ops(dir, dentry, mode, &bpf_prog_iops);
181 return bpf_mkobj_ops(dir, dentry, mode, &bpf_map_iops);
187 static struct dentry *
188 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
190 if (strchr(dentry->d_name.name, '.'))
191 return ERR_PTR(-EPERM);
193 return simple_lookup(dir, dentry, flags);
196 static int bpf_symlink(struct inode *dir, struct dentry *dentry,
199 char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
205 inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
208 return PTR_ERR(inode);
211 inode->i_op = &simple_symlink_inode_operations;
212 inode->i_link = link;
214 bpf_dentry_finalize(dentry, inode, dir);
218 static const struct inode_operations bpf_dir_iops = {
219 .lookup = bpf_lookup,
222 .symlink = bpf_symlink,
223 .rmdir = simple_rmdir,
224 .rename = simple_rename,
226 .unlink = simple_unlink,
229 static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
232 struct dentry *dentry;
239 dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
241 return PTR_ERR(dentry);
243 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
244 devt = MKDEV(UNNAMED_MAJOR, type);
246 ret = security_path_mknod(&path, dentry, mode, devt);
250 dir = d_inode(path.dentry);
251 if (dir->i_op != &bpf_dir_iops) {
256 dentry->d_fsdata = raw;
257 ret = vfs_mknod(dir, dentry, mode, devt);
258 dentry->d_fsdata = NULL;
260 done_path_create(&path, dentry);
264 int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
266 struct filename *pname;
271 pname = getname(pathname);
273 return PTR_ERR(pname);
275 raw = bpf_fd_probe_obj(ufd, &type);
281 ret = bpf_obj_do_pin(pname, raw, type);
283 bpf_any_put(raw, type);
289 static void *bpf_obj_do_get(const struct filename *pathname,
297 ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
301 inode = d_backing_inode(path.dentry);
302 ret = inode_permission(inode, MAY_WRITE);
306 ret = bpf_inode_type(inode, type);
310 raw = bpf_any_get(inode->i_private, *type);
321 int bpf_obj_get_user(const char __user *pathname)
323 enum bpf_type type = BPF_TYPE_UNSPEC;
324 struct filename *pname;
328 pname = getname(pathname);
330 return PTR_ERR(pname);
332 raw = bpf_obj_do_get(pname, &type);
338 if (type == BPF_TYPE_PROG)
339 ret = bpf_prog_new_fd(raw);
340 else if (type == BPF_TYPE_MAP)
341 ret = bpf_map_new_fd(raw);
346 bpf_any_put(raw, type);
352 static void bpf_evict_inode(struct inode *inode)
356 truncate_inode_pages_final(&inode->i_data);
359 if (S_ISLNK(inode->i_mode))
360 kfree(inode->i_link);
361 if (!bpf_inode_type(inode, &type))
362 bpf_any_put(inode->i_private, type);
365 static const struct super_operations bpf_super_ops = {
366 .statfs = simple_statfs,
367 .drop_inode = generic_delete_inode,
368 .show_options = generic_show_options,
369 .evict_inode = bpf_evict_inode,
377 static const match_table_t bpf_mount_tokens = {
378 { OPT_MODE, "mode=%o" },
382 struct bpf_mount_opts {
386 static int bpf_parse_options(char *data, struct bpf_mount_opts *opts)
388 substring_t args[MAX_OPT_ARGS];
392 opts->mode = S_IRWXUGO;
394 while ((ptr = strsep(&data, ",")) != NULL) {
398 token = match_token(ptr, bpf_mount_tokens, args);
401 if (match_octal(&args[0], &option))
403 opts->mode = option & S_IALLUGO;
405 /* We might like to report bad mount options here, but
406 * traditionally we've ignored all mount options, so we'd
407 * better continue to ignore non-existing options for bpf.
415 static int bpf_fill_super(struct super_block *sb, void *data, int silent)
417 static struct tree_descr bpf_rfiles[] = { { "" } };
418 struct bpf_mount_opts opts;
422 save_mount_options(sb, data);
424 ret = bpf_parse_options(data, &opts);
428 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
432 sb->s_op = &bpf_super_ops;
434 inode = sb->s_root->d_inode;
435 inode->i_op = &bpf_dir_iops;
436 inode->i_mode &= ~S_IALLUGO;
437 inode->i_mode |= S_ISVTX | opts.mode;
442 static struct dentry *bpf_mount(struct file_system_type *type, int flags,
443 const char *dev_name, void *data)
445 return mount_nodev(type, flags, data, bpf_fill_super);
448 static struct file_system_type bpf_fs_type = {
449 .owner = THIS_MODULE,
452 .kill_sb = kill_litter_super,
455 static int __init bpf_init(void)
459 ret = sysfs_create_mount_point(fs_kobj, "bpf");
463 ret = register_filesystem(&bpf_fs_type);
465 sysfs_remove_mount_point(fs_kobj, "bpf");
469 fs_initcall(bpf_init);