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>
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);
158 static struct map_iter *map_iter(struct seq_file *m)
163 static struct bpf_map *seq_file_to_map(struct seq_file *m)
165 return file_inode(m->file)->i_private;
168 static void map_iter_free(struct map_iter *iter)
176 static struct map_iter *map_iter_alloc(struct bpf_map *map)
178 struct map_iter *iter;
180 iter = kzalloc(sizeof(*iter), GFP_KERNEL | __GFP_NOWARN);
184 iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN);
195 static void *map_seq_next(struct seq_file *m, void *v, loff_t *pos)
197 struct bpf_map *map = seq_file_to_map(m);
198 void *key = map_iter(m)->key;
200 if (map_iter(m)->done)
203 if (unlikely(v == SEQ_START_TOKEN))
206 if (map->ops->map_get_next_key(map, key, key)) {
207 map_iter(m)->done = true;
216 static void *map_seq_start(struct seq_file *m, loff_t *pos)
218 if (map_iter(m)->done)
221 return *pos ? map_iter(m)->key : SEQ_START_TOKEN;
224 static void map_seq_stop(struct seq_file *m, void *v)
228 static int map_seq_show(struct seq_file *m, void *v)
230 struct bpf_map *map = seq_file_to_map(m);
231 void *key = map_iter(m)->key;
233 if (unlikely(v == SEQ_START_TOKEN)) {
234 seq_puts(m, "# WARNING!! The output is for debug purpose only\n");
235 seq_puts(m, "# WARNING!! The output format will change\n");
237 map->ops->map_seq_show_elem(map, key, m);
243 static const struct seq_operations bpffs_map_seq_ops = {
244 .start = map_seq_start,
245 .next = map_seq_next,
246 .show = map_seq_show,
247 .stop = map_seq_stop,
250 static int bpffs_map_open(struct inode *inode, struct file *file)
252 struct bpf_map *map = inode->i_private;
253 struct map_iter *iter;
257 iter = map_iter_alloc(map);
261 err = seq_open(file, &bpffs_map_seq_ops);
267 m = file->private_data;
273 static int bpffs_map_release(struct inode *inode, struct file *file)
275 struct seq_file *m = file->private_data;
277 map_iter_free(map_iter(m));
279 return seq_release(inode, file);
282 /* bpffs_map_fops should only implement the basic
283 * read operation for a BPF map. The purpose is to
284 * provide a simple user intuitive way to do
285 * "cat bpffs/pathto/a-pinned-map".
287 * Other operations (e.g. write, lookup...) should be realized by
288 * the userspace tools (e.g. bpftool) through the
289 * BPF_OBJ_GET_INFO_BY_FD and the map's lookup/update
292 static const struct file_operations bpffs_map_fops = {
293 .open = bpffs_map_open,
295 .release = bpffs_map_release,
298 static int bpffs_obj_open(struct inode *inode, struct file *file)
303 static const struct file_operations bpffs_obj_fops = {
304 .open = bpffs_obj_open,
307 static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
308 const struct inode_operations *iops,
309 const struct file_operations *fops)
311 struct inode *dir = dentry->d_parent->d_inode;
312 struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
314 return PTR_ERR(inode);
318 inode->i_private = raw;
320 bpf_dentry_finalize(dentry, inode, dir);
324 static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
326 return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops,
330 static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
332 struct bpf_map *map = arg;
334 return bpf_mkobj_ops(dentry, mode, arg, &bpf_map_iops,
335 map->btf ? &bpffs_map_fops : &bpffs_obj_fops);
338 static struct dentry *
339 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
341 /* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
344 if (strchr(dentry->d_name.name, '.'))
345 return ERR_PTR(-EPERM);
347 return simple_lookup(dir, dentry, flags);
350 static int bpf_symlink(struct inode *dir, struct dentry *dentry,
353 char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
359 inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
362 return PTR_ERR(inode);
365 inode->i_op = &simple_symlink_inode_operations;
366 inode->i_link = link;
368 bpf_dentry_finalize(dentry, inode, dir);
372 static const struct inode_operations bpf_dir_iops = {
373 .lookup = bpf_lookup,
375 .symlink = bpf_symlink,
376 .rmdir = simple_rmdir,
377 .rename = simple_rename,
379 .unlink = simple_unlink,
382 static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
385 struct dentry *dentry;
391 dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
393 return PTR_ERR(dentry);
395 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
397 ret = security_path_mknod(&path, dentry, mode, 0);
401 dir = d_inode(path.dentry);
402 if (dir->i_op != &bpf_dir_iops) {
409 ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
412 ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
418 done_path_create(&path, dentry);
422 int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
424 struct filename *pname;
429 pname = getname(pathname);
431 return PTR_ERR(pname);
433 raw = bpf_fd_probe_obj(ufd, &type);
439 ret = bpf_obj_do_pin(pname, raw, type);
441 bpf_any_put(raw, type);
447 static void *bpf_obj_do_get(const struct filename *pathname,
448 enum bpf_type *type, int flags)
455 ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
459 inode = d_backing_inode(path.dentry);
460 ret = inode_permission(inode, ACC_MODE(flags));
464 ret = bpf_inode_type(inode, type);
468 raw = bpf_any_get(inode->i_private, *type);
479 int bpf_obj_get_user(const char __user *pathname, int flags)
481 enum bpf_type type = BPF_TYPE_UNSPEC;
482 struct filename *pname;
487 f_flags = bpf_get_file_flag(flags);
491 pname = getname(pathname);
493 return PTR_ERR(pname);
495 raw = bpf_obj_do_get(pname, &type, f_flags);
501 if (type == BPF_TYPE_PROG)
502 ret = bpf_prog_new_fd(raw);
503 else if (type == BPF_TYPE_MAP)
504 ret = bpf_map_new_fd(raw, f_flags);
509 bpf_any_put(raw, type);
515 static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
517 struct bpf_prog *prog;
518 int ret = inode_permission(inode, MAY_READ | MAY_WRITE);
522 if (inode->i_op == &bpf_map_iops)
523 return ERR_PTR(-EINVAL);
524 if (inode->i_op != &bpf_prog_iops)
525 return ERR_PTR(-EACCES);
527 prog = inode->i_private;
529 ret = security_bpf_prog(prog);
533 if (!bpf_prog_get_ok(prog, &type, false))
534 return ERR_PTR(-EINVAL);
536 return bpf_prog_inc(prog);
539 struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
541 struct bpf_prog *prog;
543 int ret = kern_path(name, LOOKUP_FOLLOW, &path);
546 prog = __get_prog_inode(d_backing_inode(path.dentry), type);
552 EXPORT_SYMBOL(bpf_prog_get_type_path);
554 static void bpf_evict_inode(struct inode *inode)
558 truncate_inode_pages_final(&inode->i_data);
561 if (S_ISLNK(inode->i_mode))
562 kfree(inode->i_link);
563 if (!bpf_inode_type(inode, &type))
564 bpf_any_put(inode->i_private, type);
568 * Display the mount options in /proc/mounts.
570 static int bpf_show_options(struct seq_file *m, struct dentry *root)
572 umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
574 if (mode != S_IRWXUGO)
575 seq_printf(m, ",mode=%o", mode);
579 static const struct super_operations bpf_super_ops = {
580 .statfs = simple_statfs,
581 .drop_inode = generic_delete_inode,
582 .show_options = bpf_show_options,
583 .evict_inode = bpf_evict_inode,
591 static const match_table_t bpf_mount_tokens = {
592 { OPT_MODE, "mode=%o" },
596 struct bpf_mount_opts {
600 static int bpf_parse_options(char *data, struct bpf_mount_opts *opts)
602 substring_t args[MAX_OPT_ARGS];
606 opts->mode = S_IRWXUGO;
608 while ((ptr = strsep(&data, ",")) != NULL) {
612 token = match_token(ptr, bpf_mount_tokens, args);
615 if (match_octal(&args[0], &option))
617 opts->mode = option & S_IALLUGO;
619 /* We might like to report bad mount options here, but
620 * traditionally we've ignored all mount options, so we'd
621 * better continue to ignore non-existing options for bpf.
629 static int bpf_fill_super(struct super_block *sb, void *data, int silent)
631 static const struct tree_descr bpf_rfiles[] = { { "" } };
632 struct bpf_mount_opts opts;
636 ret = bpf_parse_options(data, &opts);
640 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
644 sb->s_op = &bpf_super_ops;
646 inode = sb->s_root->d_inode;
647 inode->i_op = &bpf_dir_iops;
648 inode->i_mode &= ~S_IALLUGO;
649 inode->i_mode |= S_ISVTX | opts.mode;
654 static struct dentry *bpf_mount(struct file_system_type *type, int flags,
655 const char *dev_name, void *data)
657 return mount_nodev(type, flags, data, bpf_fill_super);
660 static struct file_system_type bpf_fs_type = {
661 .owner = THIS_MODULE,
664 .kill_sb = kill_litter_super,
667 static int __init bpf_init(void)
671 ret = sysfs_create_mount_point(fs_kobj, "bpf");
675 ret = register_filesystem(&bpf_fs_type);
677 sysfs_remove_mount_point(fs_kobj, "bpf");
681 fs_initcall(bpf_init);