block: avoid ordered task state change for polled IO
[linux/fpc-iii.git] / kernel / bpf / inode.c
blob2ada5e21dfa62175d6cf9667ed4636e4c4ada659
1 /*
2 * Minimal file system backend for holding eBPF maps and programs,
3 * used by bpf(2) object pinning.
5 * Authors:
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>
19 #include <linux/fs.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>
26 enum bpf_type {
27 BPF_TYPE_UNSPEC = 0,
28 BPF_TYPE_PROG,
29 BPF_TYPE_MAP,
32 static void *bpf_any_get(void *raw, enum bpf_type type)
34 switch (type) {
35 case BPF_TYPE_PROG:
36 raw = bpf_prog_inc(raw);
37 break;
38 case BPF_TYPE_MAP:
39 raw = bpf_map_inc(raw, true);
40 break;
41 default:
42 WARN_ON_ONCE(1);
43 break;
46 return raw;
49 static void bpf_any_put(void *raw, enum bpf_type type)
51 switch (type) {
52 case BPF_TYPE_PROG:
53 bpf_prog_put(raw);
54 break;
55 case BPF_TYPE_MAP:
56 bpf_map_put_with_uref(raw);
57 break;
58 default:
59 WARN_ON_ONCE(1);
60 break;
64 static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
66 void *raw;
68 *type = BPF_TYPE_MAP;
69 raw = bpf_map_get_with_uref(ufd);
70 if (IS_ERR(raw)) {
71 *type = BPF_TYPE_PROG;
72 raw = bpf_prog_get(ufd);
75 return raw;
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,
85 umode_t mode)
87 struct inode *inode;
89 switch (mode & S_IFMT) {
90 case S_IFDIR:
91 case S_IFREG:
92 case S_IFLNK:
93 break;
94 default:
95 return ERR_PTR(-EINVAL);
98 inode = new_inode(sb);
99 if (!inode)
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);
109 return inode;
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;
119 else
120 return -EACCES;
122 return 0;
125 static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
126 struct inode *dir)
128 d_instantiate(dentry, inode);
129 dget(dentry);
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)
137 struct inode *inode;
139 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
140 if (IS_ERR(inode))
141 return PTR_ERR(inode);
143 inode->i_op = &bpf_dir_iops;
144 inode->i_fop = &simple_dir_operations;
146 inc_nlink(inode);
147 inc_nlink(dir);
149 bpf_dentry_finalize(dentry, inode, dir);
150 return 0;
153 struct map_iter {
154 void *key;
155 bool done;
158 static struct map_iter *map_iter(struct seq_file *m)
160 return m->private;
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)
170 if (iter) {
171 kfree(iter->key);
172 kfree(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);
181 if (!iter)
182 goto error;
184 iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN);
185 if (!iter->key)
186 goto error;
188 return iter;
190 error:
191 map_iter_free(iter);
192 return NULL;
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;
199 void *prev_key;
201 if (map_iter(m)->done)
202 return NULL;
204 if (unlikely(v == SEQ_START_TOKEN))
205 prev_key = NULL;
206 else
207 prev_key = key;
209 if (map->ops->map_get_next_key(map, prev_key, key)) {
210 map_iter(m)->done = true;
211 return NULL;
214 ++(*pos);
215 return key;
218 static void *map_seq_start(struct seq_file *m, loff_t *pos)
220 if (map_iter(m)->done)
221 return NULL;
223 return *pos ? map_iter(m)->key : SEQ_START_TOKEN;
226 static void map_seq_stop(struct seq_file *m, void *v)
230 static int map_seq_show(struct seq_file *m, void *v)
232 struct bpf_map *map = seq_file_to_map(m);
233 void *key = map_iter(m)->key;
235 if (unlikely(v == SEQ_START_TOKEN)) {
236 seq_puts(m, "# WARNING!! The output is for debug purpose only\n");
237 seq_puts(m, "# WARNING!! The output format will change\n");
238 } else {
239 map->ops->map_seq_show_elem(map, key, m);
242 return 0;
245 static const struct seq_operations bpffs_map_seq_ops = {
246 .start = map_seq_start,
247 .next = map_seq_next,
248 .show = map_seq_show,
249 .stop = map_seq_stop,
252 static int bpffs_map_open(struct inode *inode, struct file *file)
254 struct bpf_map *map = inode->i_private;
255 struct map_iter *iter;
256 struct seq_file *m;
257 int err;
259 iter = map_iter_alloc(map);
260 if (!iter)
261 return -ENOMEM;
263 err = seq_open(file, &bpffs_map_seq_ops);
264 if (err) {
265 map_iter_free(iter);
266 return err;
269 m = file->private_data;
270 m->private = iter;
272 return 0;
275 static int bpffs_map_release(struct inode *inode, struct file *file)
277 struct seq_file *m = file->private_data;
279 map_iter_free(map_iter(m));
281 return seq_release(inode, file);
284 /* bpffs_map_fops should only implement the basic
285 * read operation for a BPF map. The purpose is to
286 * provide a simple user intuitive way to do
287 * "cat bpffs/pathto/a-pinned-map".
289 * Other operations (e.g. write, lookup...) should be realized by
290 * the userspace tools (e.g. bpftool) through the
291 * BPF_OBJ_GET_INFO_BY_FD and the map's lookup/update
292 * interface.
294 static const struct file_operations bpffs_map_fops = {
295 .open = bpffs_map_open,
296 .read = seq_read,
297 .release = bpffs_map_release,
300 static int bpffs_obj_open(struct inode *inode, struct file *file)
302 return -EIO;
305 static const struct file_operations bpffs_obj_fops = {
306 .open = bpffs_obj_open,
309 static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
310 const struct inode_operations *iops,
311 const struct file_operations *fops)
313 struct inode *dir = dentry->d_parent->d_inode;
314 struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
315 if (IS_ERR(inode))
316 return PTR_ERR(inode);
318 inode->i_op = iops;
319 inode->i_fop = fops;
320 inode->i_private = raw;
322 bpf_dentry_finalize(dentry, inode, dir);
323 return 0;
326 static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
328 return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops,
329 &bpffs_obj_fops);
332 static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
334 struct bpf_map *map = arg;
336 return bpf_mkobj_ops(dentry, mode, arg, &bpf_map_iops,
337 bpf_map_support_seq_show(map) ?
338 &bpffs_map_fops : &bpffs_obj_fops);
341 static struct dentry *
342 bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
344 /* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
345 * extensions.
347 if (strchr(dentry->d_name.name, '.'))
348 return ERR_PTR(-EPERM);
350 return simple_lookup(dir, dentry, flags);
353 static int bpf_symlink(struct inode *dir, struct dentry *dentry,
354 const char *target)
356 char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
357 struct inode *inode;
359 if (!link)
360 return -ENOMEM;
362 inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
363 if (IS_ERR(inode)) {
364 kfree(link);
365 return PTR_ERR(inode);
368 inode->i_op = &simple_symlink_inode_operations;
369 inode->i_link = link;
371 bpf_dentry_finalize(dentry, inode, dir);
372 return 0;
375 static const struct inode_operations bpf_dir_iops = {
376 .lookup = bpf_lookup,
377 .mkdir = bpf_mkdir,
378 .symlink = bpf_symlink,
379 .rmdir = simple_rmdir,
380 .rename = simple_rename,
381 .link = simple_link,
382 .unlink = simple_unlink,
385 static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
386 enum bpf_type type)
388 struct dentry *dentry;
389 struct inode *dir;
390 struct path path;
391 umode_t mode;
392 int ret;
394 dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
395 if (IS_ERR(dentry))
396 return PTR_ERR(dentry);
398 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
400 ret = security_path_mknod(&path, dentry, mode, 0);
401 if (ret)
402 goto out;
404 dir = d_inode(path.dentry);
405 if (dir->i_op != &bpf_dir_iops) {
406 ret = -EPERM;
407 goto out;
410 switch (type) {
411 case BPF_TYPE_PROG:
412 ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
413 break;
414 case BPF_TYPE_MAP:
415 ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
416 break;
417 default:
418 ret = -EPERM;
420 out:
421 done_path_create(&path, dentry);
422 return ret;
425 int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
427 struct filename *pname;
428 enum bpf_type type;
429 void *raw;
430 int ret;
432 pname = getname(pathname);
433 if (IS_ERR(pname))
434 return PTR_ERR(pname);
436 raw = bpf_fd_probe_obj(ufd, &type);
437 if (IS_ERR(raw)) {
438 ret = PTR_ERR(raw);
439 goto out;
442 ret = bpf_obj_do_pin(pname, raw, type);
443 if (ret != 0)
444 bpf_any_put(raw, type);
445 out:
446 putname(pname);
447 return ret;
450 static void *bpf_obj_do_get(const struct filename *pathname,
451 enum bpf_type *type, int flags)
453 struct inode *inode;
454 struct path path;
455 void *raw;
456 int ret;
458 ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
459 if (ret)
460 return ERR_PTR(ret);
462 inode = d_backing_inode(path.dentry);
463 ret = inode_permission(inode, ACC_MODE(flags));
464 if (ret)
465 goto out;
467 ret = bpf_inode_type(inode, type);
468 if (ret)
469 goto out;
471 raw = bpf_any_get(inode->i_private, *type);
472 if (!IS_ERR(raw))
473 touch_atime(&path);
475 path_put(&path);
476 return raw;
477 out:
478 path_put(&path);
479 return ERR_PTR(ret);
482 int bpf_obj_get_user(const char __user *pathname, int flags)
484 enum bpf_type type = BPF_TYPE_UNSPEC;
485 struct filename *pname;
486 int ret = -ENOENT;
487 int f_flags;
488 void *raw;
490 f_flags = bpf_get_file_flag(flags);
491 if (f_flags < 0)
492 return f_flags;
494 pname = getname(pathname);
495 if (IS_ERR(pname))
496 return PTR_ERR(pname);
498 raw = bpf_obj_do_get(pname, &type, f_flags);
499 if (IS_ERR(raw)) {
500 ret = PTR_ERR(raw);
501 goto out;
504 if (type == BPF_TYPE_PROG)
505 ret = bpf_prog_new_fd(raw);
506 else if (type == BPF_TYPE_MAP)
507 ret = bpf_map_new_fd(raw, f_flags);
508 else
509 goto out;
511 if (ret < 0)
512 bpf_any_put(raw, type);
513 out:
514 putname(pname);
515 return ret;
518 static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
520 struct bpf_prog *prog;
521 int ret = inode_permission(inode, MAY_READ | MAY_WRITE);
522 if (ret)
523 return ERR_PTR(ret);
525 if (inode->i_op == &bpf_map_iops)
526 return ERR_PTR(-EINVAL);
527 if (inode->i_op != &bpf_prog_iops)
528 return ERR_PTR(-EACCES);
530 prog = inode->i_private;
532 ret = security_bpf_prog(prog);
533 if (ret < 0)
534 return ERR_PTR(ret);
536 if (!bpf_prog_get_ok(prog, &type, false))
537 return ERR_PTR(-EINVAL);
539 return bpf_prog_inc(prog);
542 struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
544 struct bpf_prog *prog;
545 struct path path;
546 int ret = kern_path(name, LOOKUP_FOLLOW, &path);
547 if (ret)
548 return ERR_PTR(ret);
549 prog = __get_prog_inode(d_backing_inode(path.dentry), type);
550 if (!IS_ERR(prog))
551 touch_atime(&path);
552 path_put(&path);
553 return prog;
555 EXPORT_SYMBOL(bpf_prog_get_type_path);
557 static void bpf_evict_inode(struct inode *inode)
559 enum bpf_type type;
561 truncate_inode_pages_final(&inode->i_data);
562 clear_inode(inode);
564 if (S_ISLNK(inode->i_mode))
565 kfree(inode->i_link);
566 if (!bpf_inode_type(inode, &type))
567 bpf_any_put(inode->i_private, type);
571 * Display the mount options in /proc/mounts.
573 static int bpf_show_options(struct seq_file *m, struct dentry *root)
575 umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
577 if (mode != S_IRWXUGO)
578 seq_printf(m, ",mode=%o", mode);
579 return 0;
582 static const struct super_operations bpf_super_ops = {
583 .statfs = simple_statfs,
584 .drop_inode = generic_delete_inode,
585 .show_options = bpf_show_options,
586 .evict_inode = bpf_evict_inode,
589 enum {
590 OPT_MODE,
591 OPT_ERR,
594 static const match_table_t bpf_mount_tokens = {
595 { OPT_MODE, "mode=%o" },
596 { OPT_ERR, NULL },
599 struct bpf_mount_opts {
600 umode_t mode;
603 static int bpf_parse_options(char *data, struct bpf_mount_opts *opts)
605 substring_t args[MAX_OPT_ARGS];
606 int option, token;
607 char *ptr;
609 opts->mode = S_IRWXUGO;
611 while ((ptr = strsep(&data, ",")) != NULL) {
612 if (!*ptr)
613 continue;
615 token = match_token(ptr, bpf_mount_tokens, args);
616 switch (token) {
617 case OPT_MODE:
618 if (match_octal(&args[0], &option))
619 return -EINVAL;
620 opts->mode = option & S_IALLUGO;
621 break;
622 /* We might like to report bad mount options here, but
623 * traditionally we've ignored all mount options, so we'd
624 * better continue to ignore non-existing options for bpf.
629 return 0;
632 static int bpf_fill_super(struct super_block *sb, void *data, int silent)
634 static const struct tree_descr bpf_rfiles[] = { { "" } };
635 struct bpf_mount_opts opts;
636 struct inode *inode;
637 int ret;
639 ret = bpf_parse_options(data, &opts);
640 if (ret)
641 return ret;
643 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
644 if (ret)
645 return ret;
647 sb->s_op = &bpf_super_ops;
649 inode = sb->s_root->d_inode;
650 inode->i_op = &bpf_dir_iops;
651 inode->i_mode &= ~S_IALLUGO;
652 inode->i_mode |= S_ISVTX | opts.mode;
654 return 0;
657 static struct dentry *bpf_mount(struct file_system_type *type, int flags,
658 const char *dev_name, void *data)
660 return mount_nodev(type, flags, data, bpf_fill_super);
663 static struct file_system_type bpf_fs_type = {
664 .owner = THIS_MODULE,
665 .name = "bpf",
666 .mount = bpf_mount,
667 .kill_sb = kill_litter_super,
670 static int __init bpf_init(void)
672 int ret;
674 ret = sysfs_create_mount_point(fs_kobj, "bpf");
675 if (ret)
676 return ret;
678 ret = register_filesystem(&bpf_fs_type);
679 if (ret)
680 sysfs_remove_mount_point(fs_kobj, "bpf");
682 return ret;
684 fs_initcall(bpf_init);