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/parser.h>
22 #include <linux/filter.h>
23 #include <linux/bpf.h>
24 #include <linux/bpf_trace.h>
32 static void *bpf_any_get(void *raw, enum bpf_type type)
36 raw = bpf_prog_inc(raw);
39 raw = bpf_map_inc(raw, true);
49 static void bpf_any_put(void *raw, enum bpf_type type)
56 bpf_map_put_with_uref(raw);
64 static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
69 raw = bpf_map_get_with_uref(ufd);
71 *type = BPF_TYPE_PROG;
72 raw = bpf_prog_get(ufd);
78 static const struct inode_operations bpf_dir_iops;
80 static const struct inode_operations bpf_prog_iops = { };
81 static const struct inode_operations bpf_map_iops = { };
83 static struct inode *bpf_get_inode(struct super_block *sb,
84 const struct inode *dir,
89 switch (mode & S_IFMT) {
95 return ERR_PTR(-EINVAL);
98 inode = new_inode(sb);
100 return ERR_PTR(-ENOSPC);
102 inode->i_ino = get_next_ino();
103 inode->i_atime = current_time(inode);
104 inode->i_mtime = inode->i_atime;
105 inode->i_ctime = inode->i_atime;
107 inode_init_owner(inode, dir, mode);
112 static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
114 *type = BPF_TYPE_UNSPEC;
115 if (inode->i_op == &bpf_prog_iops)
116 *type = BPF_TYPE_PROG;
117 else if (inode->i_op == &bpf_map_iops)
118 *type = BPF_TYPE_MAP;
125 static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
128 d_instantiate(dentry, inode);
131 dir->i_mtime = current_time(dir);
132 dir->i_ctime = dir->i_mtime;
135 static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
139 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
141 return PTR_ERR(inode);
143 inode->i_op = &bpf_dir_iops;
144 inode->i_fop = &simple_dir_operations;
149 bpf_dentry_finalize(dentry, inode, dir);
153 static int bpf_mkobj_ops(struct inode *dir, struct dentry *dentry,
154 umode_t mode, const struct inode_operations *iops)
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 bpf_dentry_finalize(dentry, inode, dir);
169 static int bpf_mkobj(struct inode *dir, struct dentry *dentry, umode_t mode,
172 enum bpf_type type = MINOR(devt);
174 if (MAJOR(devt) != UNNAMED_MAJOR || !S_ISREG(mode) ||
175 dentry->d_fsdata == NULL)
180 return bpf_mkobj_ops(dir, dentry, mode, &bpf_prog_iops);
182 return bpf_mkobj_ops(dir, dentry, mode, &bpf_map_iops);
188 static struct dentry *
189 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
191 if (strchr(dentry->d_name.name, '.'))
192 return ERR_PTR(-EPERM);
194 return simple_lookup(dir, dentry, flags);
197 static int bpf_symlink(struct inode *dir, struct dentry *dentry,
200 char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
206 inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
209 return PTR_ERR(inode);
212 inode->i_op = &simple_symlink_inode_operations;
213 inode->i_link = link;
215 bpf_dentry_finalize(dentry, inode, dir);
219 static const struct inode_operations bpf_dir_iops = {
220 .lookup = bpf_lookup,
223 .symlink = bpf_symlink,
224 .rmdir = simple_rmdir,
225 .rename = simple_rename,
227 .unlink = simple_unlink,
230 static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
233 struct dentry *dentry;
240 dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
242 return PTR_ERR(dentry);
244 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
245 devt = MKDEV(UNNAMED_MAJOR, type);
247 ret = security_path_mknod(&path, dentry, mode, devt);
251 dir = d_inode(path.dentry);
252 if (dir->i_op != &bpf_dir_iops) {
257 dentry->d_fsdata = raw;
258 ret = vfs_mknod(dir, dentry, mode, devt);
259 dentry->d_fsdata = NULL;
261 done_path_create(&path, dentry);
265 int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
267 struct filename *pname;
272 pname = getname(pathname);
274 return PTR_ERR(pname);
276 raw = bpf_fd_probe_obj(ufd, &type);
282 ret = bpf_obj_do_pin(pname, raw, type);
284 bpf_any_put(raw, type);
285 if ((trace_bpf_obj_pin_prog_enabled() ||
286 trace_bpf_obj_pin_map_enabled()) && !ret) {
287 if (type == BPF_TYPE_PROG)
288 trace_bpf_obj_pin_prog(raw, ufd, pname);
289 if (type == BPF_TYPE_MAP)
290 trace_bpf_obj_pin_map(raw, ufd, pname);
297 static void *bpf_obj_do_get(const struct filename *pathname,
305 ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
309 inode = d_backing_inode(path.dentry);
310 ret = inode_permission(inode, MAY_WRITE);
314 ret = bpf_inode_type(inode, type);
318 raw = bpf_any_get(inode->i_private, *type);
329 int bpf_obj_get_user(const char __user *pathname)
331 enum bpf_type type = BPF_TYPE_UNSPEC;
332 struct filename *pname;
336 pname = getname(pathname);
338 return PTR_ERR(pname);
340 raw = bpf_obj_do_get(pname, &type);
346 if (type == BPF_TYPE_PROG)
347 ret = bpf_prog_new_fd(raw);
348 else if (type == BPF_TYPE_MAP)
349 ret = bpf_map_new_fd(raw);
354 bpf_any_put(raw, type);
355 } else if (trace_bpf_obj_get_prog_enabled() ||
356 trace_bpf_obj_get_map_enabled()) {
357 if (type == BPF_TYPE_PROG)
358 trace_bpf_obj_get_prog(raw, ret, pname);
359 if (type == BPF_TYPE_MAP)
360 trace_bpf_obj_get_map(raw, ret, pname);
367 static void bpf_evict_inode(struct inode *inode)
371 truncate_inode_pages_final(&inode->i_data);
374 if (S_ISLNK(inode->i_mode))
375 kfree(inode->i_link);
376 if (!bpf_inode_type(inode, &type))
377 bpf_any_put(inode->i_private, type);
380 static const struct super_operations bpf_super_ops = {
381 .statfs = simple_statfs,
382 .drop_inode = generic_delete_inode,
383 .show_options = generic_show_options,
384 .evict_inode = bpf_evict_inode,
392 static const match_table_t bpf_mount_tokens = {
393 { OPT_MODE, "mode=%o" },
397 struct bpf_mount_opts {
401 static int bpf_parse_options(char *data, struct bpf_mount_opts *opts)
403 substring_t args[MAX_OPT_ARGS];
407 opts->mode = S_IRWXUGO;
409 while ((ptr = strsep(&data, ",")) != NULL) {
413 token = match_token(ptr, bpf_mount_tokens, args);
416 if (match_octal(&args[0], &option))
418 opts->mode = option & S_IALLUGO;
420 /* We might like to report bad mount options here, but
421 * traditionally we've ignored all mount options, so we'd
422 * better continue to ignore non-existing options for bpf.
430 static int bpf_fill_super(struct super_block *sb, void *data, int silent)
432 static const struct tree_descr bpf_rfiles[] = { { "" } };
433 struct bpf_mount_opts opts;
437 save_mount_options(sb, data);
439 ret = bpf_parse_options(data, &opts);
443 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
447 sb->s_op = &bpf_super_ops;
449 inode = sb->s_root->d_inode;
450 inode->i_op = &bpf_dir_iops;
451 inode->i_mode &= ~S_IALLUGO;
452 inode->i_mode |= S_ISVTX | opts.mode;
457 static struct dentry *bpf_mount(struct file_system_type *type, int flags,
458 const char *dev_name, void *data)
460 return mount_nodev(type, flags, data, bpf_fill_super);
463 static struct file_system_type bpf_fs_type = {
464 .owner = THIS_MODULE,
467 .kill_sb = kill_litter_super,
470 static int __init bpf_init(void)
474 ret = sysfs_create_mount_point(fs_kobj, "bpf");
478 ret = register_filesystem(&bpf_fs_type);
480 sysfs_remove_mount_point(fs_kobj, "bpf");
484 fs_initcall(bpf_init);