Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Minimal file system backend for holding eBPF maps and programs,
4 * used by bpf(2) object pinning.
5 *
6 * Authors:
7 *
8 * Daniel Borkmann <daniel@iogearbox.net>
9 */
10
11#include <linux/init.h>
12#include <linux/magic.h>
13#include <linux/major.h>
14#include <linux/mount.h>
15#include <linux/namei.h>
16#include <linux/fs.h>
17#include <linux/fs_context.h>
18#include <linux/fs_parser.h>
19#include <linux/kdev_t.h>
20#include <linux/filter.h>
21#include <linux/bpf.h>
22#include <linux/bpf_trace.h>
23#include "preload/bpf_preload.h"
24
25enum bpf_type {
26 BPF_TYPE_UNSPEC = 0,
27 BPF_TYPE_PROG,
28 BPF_TYPE_MAP,
29 BPF_TYPE_LINK,
30};
31
32static void *bpf_any_get(void *raw, enum bpf_type type)
33{
34 switch (type) {
35 case BPF_TYPE_PROG:
36 bpf_prog_inc(raw);
37 break;
38 case BPF_TYPE_MAP:
39 bpf_map_inc_with_uref(raw);
40 break;
41 case BPF_TYPE_LINK:
42 bpf_link_inc(raw);
43 break;
44 default:
45 WARN_ON_ONCE(1);
46 break;
47 }
48
49 return raw;
50}
51
52static void bpf_any_put(void *raw, enum bpf_type type)
53{
54 switch (type) {
55 case BPF_TYPE_PROG:
56 bpf_prog_put(raw);
57 break;
58 case BPF_TYPE_MAP:
59 bpf_map_put_with_uref(raw);
60 break;
61 case BPF_TYPE_LINK:
62 bpf_link_put(raw);
63 break;
64 default:
65 WARN_ON_ONCE(1);
66 break;
67 }
68}
69
70static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
71{
72 void *raw;
73
74 raw = bpf_map_get_with_uref(ufd);
75 if (!IS_ERR(raw)) {
76 *type = BPF_TYPE_MAP;
77 return raw;
78 }
79
80 raw = bpf_prog_get(ufd);
81 if (!IS_ERR(raw)) {
82 *type = BPF_TYPE_PROG;
83 return raw;
84 }
85
86 raw = bpf_link_get_from_fd(ufd);
87 if (!IS_ERR(raw)) {
88 *type = BPF_TYPE_LINK;
89 return raw;
90 }
91
92 return ERR_PTR(-EINVAL);
93}
94
95static const struct inode_operations bpf_dir_iops;
96
97static const struct inode_operations bpf_prog_iops = { };
98static const struct inode_operations bpf_map_iops = { };
99static const struct inode_operations bpf_link_iops = { };
100
101static struct inode *bpf_get_inode(struct super_block *sb,
102 const struct inode *dir,
103 umode_t mode)
104{
105 struct inode *inode;
106
107 switch (mode & S_IFMT) {
108 case S_IFDIR:
109 case S_IFREG:
110 case S_IFLNK:
111 break;
112 default:
113 return ERR_PTR(-EINVAL);
114 }
115
116 inode = new_inode(sb);
117 if (!inode)
118 return ERR_PTR(-ENOSPC);
119
120 inode->i_ino = get_next_ino();
121 inode->i_atime = current_time(inode);
122 inode->i_mtime = inode->i_atime;
123 inode->i_ctime = inode->i_atime;
124
125 inode_init_owner(&init_user_ns, inode, dir, mode);
126
127 return inode;
128}
129
130static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
131{
132 *type = BPF_TYPE_UNSPEC;
133 if (inode->i_op == &bpf_prog_iops)
134 *type = BPF_TYPE_PROG;
135 else if (inode->i_op == &bpf_map_iops)
136 *type = BPF_TYPE_MAP;
137 else if (inode->i_op == &bpf_link_iops)
138 *type = BPF_TYPE_LINK;
139 else
140 return -EACCES;
141
142 return 0;
143}
144
145static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
146 struct inode *dir)
147{
148 d_instantiate(dentry, inode);
149 dget(dentry);
150
151 dir->i_mtime = current_time(dir);
152 dir->i_ctime = dir->i_mtime;
153}
154
155static int bpf_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
156 struct dentry *dentry, umode_t mode)
157{
158 struct inode *inode;
159
160 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
161 if (IS_ERR(inode))
162 return PTR_ERR(inode);
163
164 inode->i_op = &bpf_dir_iops;
165 inode->i_fop = &simple_dir_operations;
166
167 inc_nlink(inode);
168 inc_nlink(dir);
169
170 bpf_dentry_finalize(dentry, inode, dir);
171 return 0;
172}
173
174struct map_iter {
175 void *key;
176 bool done;
177};
178
179static struct map_iter *map_iter(struct seq_file *m)
180{
181 return m->private;
182}
183
184static struct bpf_map *seq_file_to_map(struct seq_file *m)
185{
186 return file_inode(m->file)->i_private;
187}
188
189static void map_iter_free(struct map_iter *iter)
190{
191 if (iter) {
192 kfree(iter->key);
193 kfree(iter);
194 }
195}
196
197static struct map_iter *map_iter_alloc(struct bpf_map *map)
198{
199 struct map_iter *iter;
200
201 iter = kzalloc(sizeof(*iter), GFP_KERNEL | __GFP_NOWARN);
202 if (!iter)
203 goto error;
204
205 iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN);
206 if (!iter->key)
207 goto error;
208
209 return iter;
210
211error:
212 map_iter_free(iter);
213 return NULL;
214}
215
216static void *map_seq_next(struct seq_file *m, void *v, loff_t *pos)
217{
218 struct bpf_map *map = seq_file_to_map(m);
219 void *key = map_iter(m)->key;
220 void *prev_key;
221
222 (*pos)++;
223 if (map_iter(m)->done)
224 return NULL;
225
226 if (unlikely(v == SEQ_START_TOKEN))
227 prev_key = NULL;
228 else
229 prev_key = key;
230
231 rcu_read_lock();
232 if (map->ops->map_get_next_key(map, prev_key, key)) {
233 map_iter(m)->done = true;
234 key = NULL;
235 }
236 rcu_read_unlock();
237 return key;
238}
239
240static void *map_seq_start(struct seq_file *m, loff_t *pos)
241{
242 if (map_iter(m)->done)
243 return NULL;
244
245 return *pos ? map_iter(m)->key : SEQ_START_TOKEN;
246}
247
248static void map_seq_stop(struct seq_file *m, void *v)
249{
250}
251
252static int map_seq_show(struct seq_file *m, void *v)
253{
254 struct bpf_map *map = seq_file_to_map(m);
255 void *key = map_iter(m)->key;
256
257 if (unlikely(v == SEQ_START_TOKEN)) {
258 seq_puts(m, "# WARNING!! The output is for debug purpose only\n");
259 seq_puts(m, "# WARNING!! The output format will change\n");
260 } else {
261 map->ops->map_seq_show_elem(map, key, m);
262 }
263
264 return 0;
265}
266
267static const struct seq_operations bpffs_map_seq_ops = {
268 .start = map_seq_start,
269 .next = map_seq_next,
270 .show = map_seq_show,
271 .stop = map_seq_stop,
272};
273
274static int bpffs_map_open(struct inode *inode, struct file *file)
275{
276 struct bpf_map *map = inode->i_private;
277 struct map_iter *iter;
278 struct seq_file *m;
279 int err;
280
281 iter = map_iter_alloc(map);
282 if (!iter)
283 return -ENOMEM;
284
285 err = seq_open(file, &bpffs_map_seq_ops);
286 if (err) {
287 map_iter_free(iter);
288 return err;
289 }
290
291 m = file->private_data;
292 m->private = iter;
293
294 return 0;
295}
296
297static int bpffs_map_release(struct inode *inode, struct file *file)
298{
299 struct seq_file *m = file->private_data;
300
301 map_iter_free(map_iter(m));
302
303 return seq_release(inode, file);
304}
305
306/* bpffs_map_fops should only implement the basic
307 * read operation for a BPF map. The purpose is to
308 * provide a simple user intuitive way to do
309 * "cat bpffs/pathto/a-pinned-map".
310 *
311 * Other operations (e.g. write, lookup...) should be realized by
312 * the userspace tools (e.g. bpftool) through the
313 * BPF_OBJ_GET_INFO_BY_FD and the map's lookup/update
314 * interface.
315 */
316static const struct file_operations bpffs_map_fops = {
317 .open = bpffs_map_open,
318 .read = seq_read,
319 .release = bpffs_map_release,
320};
321
322static int bpffs_obj_open(struct inode *inode, struct file *file)
323{
324 return -EIO;
325}
326
327static const struct file_operations bpffs_obj_fops = {
328 .open = bpffs_obj_open,
329};
330
331static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
332 const struct inode_operations *iops,
333 const struct file_operations *fops)
334{
335 struct inode *dir = dentry->d_parent->d_inode;
336 struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
337 if (IS_ERR(inode))
338 return PTR_ERR(inode);
339
340 inode->i_op = iops;
341 inode->i_fop = fops;
342 inode->i_private = raw;
343
344 bpf_dentry_finalize(dentry, inode, dir);
345 return 0;
346}
347
348static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
349{
350 return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops,
351 &bpffs_obj_fops);
352}
353
354static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
355{
356 struct bpf_map *map = arg;
357
358 return bpf_mkobj_ops(dentry, mode, arg, &bpf_map_iops,
359 bpf_map_support_seq_show(map) ?
360 &bpffs_map_fops : &bpffs_obj_fops);
361}
362
363static int bpf_mklink(struct dentry *dentry, umode_t mode, void *arg)
364{
365 struct bpf_link *link = arg;
366
367 return bpf_mkobj_ops(dentry, mode, arg, &bpf_link_iops,
368 bpf_link_is_iter(link) ?
369 &bpf_iter_fops : &bpffs_obj_fops);
370}
371
372static struct dentry *
373bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
374{
375 /* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
376 * extensions. That allows popoulate_bpffs() create special files.
377 */
378 if ((dir->i_mode & S_IALLUGO) &&
379 strchr(dentry->d_name.name, '.'))
380 return ERR_PTR(-EPERM);
381
382 return simple_lookup(dir, dentry, flags);
383}
384
385static int bpf_symlink(struct user_namespace *mnt_userns, struct inode *dir,
386 struct dentry *dentry, const char *target)
387{
388 char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
389 struct inode *inode;
390
391 if (!link)
392 return -ENOMEM;
393
394 inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
395 if (IS_ERR(inode)) {
396 kfree(link);
397 return PTR_ERR(inode);
398 }
399
400 inode->i_op = &simple_symlink_inode_operations;
401 inode->i_link = link;
402
403 bpf_dentry_finalize(dentry, inode, dir);
404 return 0;
405}
406
407static const struct inode_operations bpf_dir_iops = {
408 .lookup = bpf_lookup,
409 .mkdir = bpf_mkdir,
410 .symlink = bpf_symlink,
411 .rmdir = simple_rmdir,
412 .rename = simple_rename,
413 .link = simple_link,
414 .unlink = simple_unlink,
415};
416
417/* pin iterator link into bpffs */
418static int bpf_iter_link_pin_kernel(struct dentry *parent,
419 const char *name, struct bpf_link *link)
420{
421 umode_t mode = S_IFREG | S_IRUSR;
422 struct dentry *dentry;
423 int ret;
424
425 inode_lock(parent->d_inode);
426 dentry = lookup_one_len(name, parent, strlen(name));
427 if (IS_ERR(dentry)) {
428 inode_unlock(parent->d_inode);
429 return PTR_ERR(dentry);
430 }
431 ret = bpf_mkobj_ops(dentry, mode, link, &bpf_link_iops,
432 &bpf_iter_fops);
433 dput(dentry);
434 inode_unlock(parent->d_inode);
435 return ret;
436}
437
438static int bpf_obj_do_pin(const char __user *pathname, void *raw,
439 enum bpf_type type)
440{
441 struct dentry *dentry;
442 struct inode *dir;
443 struct path path;
444 umode_t mode;
445 int ret;
446
447 dentry = user_path_create(AT_FDCWD, pathname, &path, 0);
448 if (IS_ERR(dentry))
449 return PTR_ERR(dentry);
450
451 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
452
453 ret = security_path_mknod(&path, dentry, mode, 0);
454 if (ret)
455 goto out;
456
457 dir = d_inode(path.dentry);
458 if (dir->i_op != &bpf_dir_iops) {
459 ret = -EPERM;
460 goto out;
461 }
462
463 switch (type) {
464 case BPF_TYPE_PROG:
465 ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
466 break;
467 case BPF_TYPE_MAP:
468 ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
469 break;
470 case BPF_TYPE_LINK:
471 ret = vfs_mkobj(dentry, mode, bpf_mklink, raw);
472 break;
473 default:
474 ret = -EPERM;
475 }
476out:
477 done_path_create(&path, dentry);
478 return ret;
479}
480
481int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
482{
483 enum bpf_type type;
484 void *raw;
485 int ret;
486
487 raw = bpf_fd_probe_obj(ufd, &type);
488 if (IS_ERR(raw))
489 return PTR_ERR(raw);
490
491 ret = bpf_obj_do_pin(pathname, raw, type);
492 if (ret != 0)
493 bpf_any_put(raw, type);
494
495 return ret;
496}
497
498static void *bpf_obj_do_get(const char __user *pathname,
499 enum bpf_type *type, int flags)
500{
501 struct inode *inode;
502 struct path path;
503 void *raw;
504 int ret;
505
506 ret = user_path_at(AT_FDCWD, pathname, LOOKUP_FOLLOW, &path);
507 if (ret)
508 return ERR_PTR(ret);
509
510 inode = d_backing_inode(path.dentry);
511 ret = path_permission(&path, ACC_MODE(flags));
512 if (ret)
513 goto out;
514
515 ret = bpf_inode_type(inode, type);
516 if (ret)
517 goto out;
518
519 raw = bpf_any_get(inode->i_private, *type);
520 if (!IS_ERR(raw))
521 touch_atime(&path);
522
523 path_put(&path);
524 return raw;
525out:
526 path_put(&path);
527 return ERR_PTR(ret);
528}
529
530int bpf_obj_get_user(const char __user *pathname, int flags)
531{
532 enum bpf_type type = BPF_TYPE_UNSPEC;
533 int f_flags;
534 void *raw;
535 int ret;
536
537 f_flags = bpf_get_file_flag(flags);
538 if (f_flags < 0)
539 return f_flags;
540
541 raw = bpf_obj_do_get(pathname, &type, f_flags);
542 if (IS_ERR(raw))
543 return PTR_ERR(raw);
544
545 if (type == BPF_TYPE_PROG)
546 ret = bpf_prog_new_fd(raw);
547 else if (type == BPF_TYPE_MAP)
548 ret = bpf_map_new_fd(raw, f_flags);
549 else if (type == BPF_TYPE_LINK)
550 ret = (f_flags != O_RDWR) ? -EINVAL : bpf_link_new_fd(raw);
551 else
552 return -ENOENT;
553
554 if (ret < 0)
555 bpf_any_put(raw, type);
556 return ret;
557}
558
559static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
560{
561 struct bpf_prog *prog;
562 int ret = inode_permission(&init_user_ns, inode, MAY_READ);
563 if (ret)
564 return ERR_PTR(ret);
565
566 if (inode->i_op == &bpf_map_iops)
567 return ERR_PTR(-EINVAL);
568 if (inode->i_op == &bpf_link_iops)
569 return ERR_PTR(-EINVAL);
570 if (inode->i_op != &bpf_prog_iops)
571 return ERR_PTR(-EACCES);
572
573 prog = inode->i_private;
574
575 ret = security_bpf_prog(prog);
576 if (ret < 0)
577 return ERR_PTR(ret);
578
579 if (!bpf_prog_get_ok(prog, &type, false))
580 return ERR_PTR(-EINVAL);
581
582 bpf_prog_inc(prog);
583 return prog;
584}
585
586struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
587{
588 struct bpf_prog *prog;
589 struct path path;
590 int ret = kern_path(name, LOOKUP_FOLLOW, &path);
591 if (ret)
592 return ERR_PTR(ret);
593 prog = __get_prog_inode(d_backing_inode(path.dentry), type);
594 if (!IS_ERR(prog))
595 touch_atime(&path);
596 path_put(&path);
597 return prog;
598}
599EXPORT_SYMBOL(bpf_prog_get_type_path);
600
601/*
602 * Display the mount options in /proc/mounts.
603 */
604static int bpf_show_options(struct seq_file *m, struct dentry *root)
605{
606 umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
607
608 if (mode != S_IRWXUGO)
609 seq_printf(m, ",mode=%o", mode);
610 return 0;
611}
612
613static void bpf_free_inode(struct inode *inode)
614{
615 enum bpf_type type;
616
617 if (S_ISLNK(inode->i_mode))
618 kfree(inode->i_link);
619 if (!bpf_inode_type(inode, &type))
620 bpf_any_put(inode->i_private, type);
621 free_inode_nonrcu(inode);
622}
623
624static const struct super_operations bpf_super_ops = {
625 .statfs = simple_statfs,
626 .drop_inode = generic_delete_inode,
627 .show_options = bpf_show_options,
628 .free_inode = bpf_free_inode,
629};
630
631enum {
632 OPT_MODE,
633};
634
635static const struct fs_parameter_spec bpf_fs_parameters[] = {
636 fsparam_u32oct ("mode", OPT_MODE),
637 {}
638};
639
640struct bpf_mount_opts {
641 umode_t mode;
642};
643
644static int bpf_parse_param(struct fs_context *fc, struct fs_parameter *param)
645{
646 struct bpf_mount_opts *opts = fc->fs_private;
647 struct fs_parse_result result;
648 int opt;
649
650 opt = fs_parse(fc, bpf_fs_parameters, param, &result);
651 if (opt < 0) {
652 /* We might like to report bad mount options here, but
653 * traditionally we've ignored all mount options, so we'd
654 * better continue to ignore non-existing options for bpf.
655 */
656 if (opt == -ENOPARAM) {
657 opt = vfs_parse_fs_param_source(fc, param);
658 if (opt != -ENOPARAM)
659 return opt;
660
661 return 0;
662 }
663
664 if (opt < 0)
665 return opt;
666 }
667
668 switch (opt) {
669 case OPT_MODE:
670 opts->mode = result.uint_32 & S_IALLUGO;
671 break;
672 }
673
674 return 0;
675}
676
677struct bpf_preload_ops *bpf_preload_ops;
678EXPORT_SYMBOL_GPL(bpf_preload_ops);
679
680static bool bpf_preload_mod_get(void)
681{
682 /* If bpf_preload.ko wasn't loaded earlier then load it now.
683 * When bpf_preload is built into vmlinux the module's __init
684 * function will populate it.
685 */
686 if (!bpf_preload_ops) {
687 request_module("bpf_preload");
688 if (!bpf_preload_ops)
689 return false;
690 }
691 /* And grab the reference, so the module doesn't disappear while the
692 * kernel is interacting with the kernel module and its UMD.
693 */
694 if (!try_module_get(bpf_preload_ops->owner)) {
695 pr_err("bpf_preload module get failed.\n");
696 return false;
697 }
698 return true;
699}
700
701static void bpf_preload_mod_put(void)
702{
703 if (bpf_preload_ops)
704 /* now user can "rmmod bpf_preload" if necessary */
705 module_put(bpf_preload_ops->owner);
706}
707
708static DEFINE_MUTEX(bpf_preload_lock);
709
710static int populate_bpffs(struct dentry *parent)
711{
712 struct bpf_preload_info objs[BPF_PRELOAD_LINKS] = {};
713 int err = 0, i;
714
715 /* grab the mutex to make sure the kernel interactions with bpf_preload
716 * are serialized
717 */
718 mutex_lock(&bpf_preload_lock);
719
720 /* if bpf_preload.ko wasn't built into vmlinux then load it */
721 if (!bpf_preload_mod_get())
722 goto out;
723
724 err = bpf_preload_ops->preload(objs);
725 if (err)
726 goto out_put;
727 for (i = 0; i < BPF_PRELOAD_LINKS; i++) {
728 bpf_link_inc(objs[i].link);
729 err = bpf_iter_link_pin_kernel(parent,
730 objs[i].link_name, objs[i].link);
731 if (err) {
732 bpf_link_put(objs[i].link);
733 goto out_put;
734 }
735 }
736out_put:
737 bpf_preload_mod_put();
738out:
739 mutex_unlock(&bpf_preload_lock);
740 return err;
741}
742
743static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
744{
745 static const struct tree_descr bpf_rfiles[] = { { "" } };
746 struct bpf_mount_opts *opts = fc->fs_private;
747 struct inode *inode;
748 int ret;
749
750 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
751 if (ret)
752 return ret;
753
754 sb->s_op = &bpf_super_ops;
755
756 inode = sb->s_root->d_inode;
757 inode->i_op = &bpf_dir_iops;
758 inode->i_mode &= ~S_IALLUGO;
759 populate_bpffs(sb->s_root);
760 inode->i_mode |= S_ISVTX | opts->mode;
761 return 0;
762}
763
764static int bpf_get_tree(struct fs_context *fc)
765{
766 return get_tree_nodev(fc, bpf_fill_super);
767}
768
769static void bpf_free_fc(struct fs_context *fc)
770{
771 kfree(fc->fs_private);
772}
773
774static const struct fs_context_operations bpf_context_ops = {
775 .free = bpf_free_fc,
776 .parse_param = bpf_parse_param,
777 .get_tree = bpf_get_tree,
778};
779
780/*
781 * Set up the filesystem mount context.
782 */
783static int bpf_init_fs_context(struct fs_context *fc)
784{
785 struct bpf_mount_opts *opts;
786
787 opts = kzalloc(sizeof(struct bpf_mount_opts), GFP_KERNEL);
788 if (!opts)
789 return -ENOMEM;
790
791 opts->mode = S_IRWXUGO;
792
793 fc->fs_private = opts;
794 fc->ops = &bpf_context_ops;
795 return 0;
796}
797
798static struct file_system_type bpf_fs_type = {
799 .owner = THIS_MODULE,
800 .name = "bpf",
801 .init_fs_context = bpf_init_fs_context,
802 .parameters = bpf_fs_parameters,
803 .kill_sb = kill_litter_super,
804};
805
806static int __init bpf_init(void)
807{
808 int ret;
809
810 ret = sysfs_create_mount_point(fs_kobj, "bpf");
811 if (ret)
812 return ret;
813
814 ret = register_filesystem(&bpf_fs_type);
815 if (ret)
816 sysfs_remove_mount_point(fs_kobj, "bpf");
817
818 return ret;
819}
820fs_initcall(bpf_init);
1/*
2 * Minimal file system backend for holding eBPF maps and programs,
3 * used by bpf(2) object pinning.
4 *
5 * Authors:
6 *
7 * Daniel Borkmann <daniel@iogearbox.net>
8 *
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.
12 */
13
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>
25
26enum bpf_type {
27 BPF_TYPE_UNSPEC = 0,
28 BPF_TYPE_PROG,
29 BPF_TYPE_MAP,
30};
31
32static void *bpf_any_get(void *raw, enum bpf_type type)
33{
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;
44 }
45
46 return raw;
47}
48
49static void bpf_any_put(void *raw, enum bpf_type type)
50{
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;
61 }
62}
63
64static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
65{
66 void *raw;
67
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);
73 }
74
75 return raw;
76}
77
78static const struct inode_operations bpf_dir_iops;
79
80static const struct inode_operations bpf_prog_iops = { };
81static const struct inode_operations bpf_map_iops = { };
82
83static struct inode *bpf_get_inode(struct super_block *sb,
84 const struct inode *dir,
85 umode_t mode)
86{
87 struct inode *inode;
88
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);
96 }
97
98 inode = new_inode(sb);
99 if (!inode)
100 return ERR_PTR(-ENOSPC);
101
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;
106
107 inode_init_owner(inode, dir, mode);
108
109 return inode;
110}
111
112static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
113{
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;
121
122 return 0;
123}
124
125static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
126 struct inode *dir)
127{
128 d_instantiate(dentry, inode);
129 dget(dentry);
130
131 dir->i_mtime = current_time(dir);
132 dir->i_ctime = dir->i_mtime;
133}
134
135static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
136{
137 struct inode *inode;
138
139 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
140 if (IS_ERR(inode))
141 return PTR_ERR(inode);
142
143 inode->i_op = &bpf_dir_iops;
144 inode->i_fop = &simple_dir_operations;
145
146 inc_nlink(inode);
147 inc_nlink(dir);
148
149 bpf_dentry_finalize(dentry, inode, dir);
150 return 0;
151}
152
153static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
154 const struct inode_operations *iops)
155{
156 struct inode *dir = dentry->d_parent->d_inode;
157 struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
158 if (IS_ERR(inode))
159 return PTR_ERR(inode);
160
161 inode->i_op = iops;
162 inode->i_private = raw;
163
164 bpf_dentry_finalize(dentry, inode, dir);
165 return 0;
166}
167
168static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
169{
170 return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops);
171}
172
173static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
174{
175 return bpf_mkobj_ops(dentry, mode, arg, &bpf_map_iops);
176}
177
178static struct dentry *
179bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
180{
181 /* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
182 * extensions.
183 */
184 if (strchr(dentry->d_name.name, '.'))
185 return ERR_PTR(-EPERM);
186
187 return simple_lookup(dir, dentry, flags);
188}
189
190static int bpf_symlink(struct inode *dir, struct dentry *dentry,
191 const char *target)
192{
193 char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
194 struct inode *inode;
195
196 if (!link)
197 return -ENOMEM;
198
199 inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
200 if (IS_ERR(inode)) {
201 kfree(link);
202 return PTR_ERR(inode);
203 }
204
205 inode->i_op = &simple_symlink_inode_operations;
206 inode->i_link = link;
207
208 bpf_dentry_finalize(dentry, inode, dir);
209 return 0;
210}
211
212static const struct inode_operations bpf_dir_iops = {
213 .lookup = bpf_lookup,
214 .mkdir = bpf_mkdir,
215 .symlink = bpf_symlink,
216 .rmdir = simple_rmdir,
217 .rename = simple_rename,
218 .link = simple_link,
219 .unlink = simple_unlink,
220};
221
222static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
223 enum bpf_type type)
224{
225 struct dentry *dentry;
226 struct inode *dir;
227 struct path path;
228 umode_t mode;
229 int ret;
230
231 dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
232 if (IS_ERR(dentry))
233 return PTR_ERR(dentry);
234
235 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
236
237 ret = security_path_mknod(&path, dentry, mode, 0);
238 if (ret)
239 goto out;
240
241 dir = d_inode(path.dentry);
242 if (dir->i_op != &bpf_dir_iops) {
243 ret = -EPERM;
244 goto out;
245 }
246
247 switch (type) {
248 case BPF_TYPE_PROG:
249 ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
250 break;
251 case BPF_TYPE_MAP:
252 ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
253 break;
254 default:
255 ret = -EPERM;
256 }
257out:
258 done_path_create(&path, dentry);
259 return ret;
260}
261
262int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
263{
264 struct filename *pname;
265 enum bpf_type type;
266 void *raw;
267 int ret;
268
269 pname = getname(pathname);
270 if (IS_ERR(pname))
271 return PTR_ERR(pname);
272
273 raw = bpf_fd_probe_obj(ufd, &type);
274 if (IS_ERR(raw)) {
275 ret = PTR_ERR(raw);
276 goto out;
277 }
278
279 ret = bpf_obj_do_pin(pname, raw, type);
280 if (ret != 0)
281 bpf_any_put(raw, type);
282 if ((trace_bpf_obj_pin_prog_enabled() ||
283 trace_bpf_obj_pin_map_enabled()) && !ret) {
284 if (type == BPF_TYPE_PROG)
285 trace_bpf_obj_pin_prog(raw, ufd, pname);
286 if (type == BPF_TYPE_MAP)
287 trace_bpf_obj_pin_map(raw, ufd, pname);
288 }
289out:
290 putname(pname);
291 return ret;
292}
293
294static void *bpf_obj_do_get(const struct filename *pathname,
295 enum bpf_type *type, int flags)
296{
297 struct inode *inode;
298 struct path path;
299 void *raw;
300 int ret;
301
302 ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
303 if (ret)
304 return ERR_PTR(ret);
305
306 inode = d_backing_inode(path.dentry);
307 ret = inode_permission(inode, ACC_MODE(flags));
308 if (ret)
309 goto out;
310
311 ret = bpf_inode_type(inode, type);
312 if (ret)
313 goto out;
314
315 raw = bpf_any_get(inode->i_private, *type);
316 if (!IS_ERR(raw))
317 touch_atime(&path);
318
319 path_put(&path);
320 return raw;
321out:
322 path_put(&path);
323 return ERR_PTR(ret);
324}
325
326int bpf_obj_get_user(const char __user *pathname, int flags)
327{
328 enum bpf_type type = BPF_TYPE_UNSPEC;
329 struct filename *pname;
330 int ret = -ENOENT;
331 int f_flags;
332 void *raw;
333
334 f_flags = bpf_get_file_flag(flags);
335 if (f_flags < 0)
336 return f_flags;
337
338 pname = getname(pathname);
339 if (IS_ERR(pname))
340 return PTR_ERR(pname);
341
342 raw = bpf_obj_do_get(pname, &type, f_flags);
343 if (IS_ERR(raw)) {
344 ret = PTR_ERR(raw);
345 goto out;
346 }
347
348 if (type == BPF_TYPE_PROG)
349 ret = bpf_prog_new_fd(raw);
350 else if (type == BPF_TYPE_MAP)
351 ret = bpf_map_new_fd(raw, f_flags);
352 else
353 goto out;
354
355 if (ret < 0) {
356 bpf_any_put(raw, type);
357 } else if (trace_bpf_obj_get_prog_enabled() ||
358 trace_bpf_obj_get_map_enabled()) {
359 if (type == BPF_TYPE_PROG)
360 trace_bpf_obj_get_prog(raw, ret, pname);
361 if (type == BPF_TYPE_MAP)
362 trace_bpf_obj_get_map(raw, ret, pname);
363 }
364out:
365 putname(pname);
366 return ret;
367}
368
369static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
370{
371 struct bpf_prog *prog;
372 int ret = inode_permission(inode, MAY_READ | MAY_WRITE);
373 if (ret)
374 return ERR_PTR(ret);
375
376 if (inode->i_op == &bpf_map_iops)
377 return ERR_PTR(-EINVAL);
378 if (inode->i_op != &bpf_prog_iops)
379 return ERR_PTR(-EACCES);
380
381 prog = inode->i_private;
382
383 ret = security_bpf_prog(prog);
384 if (ret < 0)
385 return ERR_PTR(ret);
386
387 if (!bpf_prog_get_ok(prog, &type, false))
388 return ERR_PTR(-EINVAL);
389
390 return bpf_prog_inc(prog);
391}
392
393struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
394{
395 struct bpf_prog *prog;
396 struct path path;
397 int ret = kern_path(name, LOOKUP_FOLLOW, &path);
398 if (ret)
399 return ERR_PTR(ret);
400 prog = __get_prog_inode(d_backing_inode(path.dentry), type);
401 if (!IS_ERR(prog))
402 touch_atime(&path);
403 path_put(&path);
404 return prog;
405}
406EXPORT_SYMBOL(bpf_prog_get_type_path);
407
408static void bpf_evict_inode(struct inode *inode)
409{
410 enum bpf_type type;
411
412 truncate_inode_pages_final(&inode->i_data);
413 clear_inode(inode);
414
415 if (S_ISLNK(inode->i_mode))
416 kfree(inode->i_link);
417 if (!bpf_inode_type(inode, &type))
418 bpf_any_put(inode->i_private, type);
419}
420
421/*
422 * Display the mount options in /proc/mounts.
423 */
424static int bpf_show_options(struct seq_file *m, struct dentry *root)
425{
426 umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
427
428 if (mode != S_IRWXUGO)
429 seq_printf(m, ",mode=%o", mode);
430 return 0;
431}
432
433static const struct super_operations bpf_super_ops = {
434 .statfs = simple_statfs,
435 .drop_inode = generic_delete_inode,
436 .show_options = bpf_show_options,
437 .evict_inode = bpf_evict_inode,
438};
439
440enum {
441 OPT_MODE,
442 OPT_ERR,
443};
444
445static const match_table_t bpf_mount_tokens = {
446 { OPT_MODE, "mode=%o" },
447 { OPT_ERR, NULL },
448};
449
450struct bpf_mount_opts {
451 umode_t mode;
452};
453
454static int bpf_parse_options(char *data, struct bpf_mount_opts *opts)
455{
456 substring_t args[MAX_OPT_ARGS];
457 int option, token;
458 char *ptr;
459
460 opts->mode = S_IRWXUGO;
461
462 while ((ptr = strsep(&data, ",")) != NULL) {
463 if (!*ptr)
464 continue;
465
466 token = match_token(ptr, bpf_mount_tokens, args);
467 switch (token) {
468 case OPT_MODE:
469 if (match_octal(&args[0], &option))
470 return -EINVAL;
471 opts->mode = option & S_IALLUGO;
472 break;
473 /* We might like to report bad mount options here, but
474 * traditionally we've ignored all mount options, so we'd
475 * better continue to ignore non-existing options for bpf.
476 */
477 }
478 }
479
480 return 0;
481}
482
483static int bpf_fill_super(struct super_block *sb, void *data, int silent)
484{
485 static const struct tree_descr bpf_rfiles[] = { { "" } };
486 struct bpf_mount_opts opts;
487 struct inode *inode;
488 int ret;
489
490 ret = bpf_parse_options(data, &opts);
491 if (ret)
492 return ret;
493
494 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
495 if (ret)
496 return ret;
497
498 sb->s_op = &bpf_super_ops;
499
500 inode = sb->s_root->d_inode;
501 inode->i_op = &bpf_dir_iops;
502 inode->i_mode &= ~S_IALLUGO;
503 inode->i_mode |= S_ISVTX | opts.mode;
504
505 return 0;
506}
507
508static struct dentry *bpf_mount(struct file_system_type *type, int flags,
509 const char *dev_name, void *data)
510{
511 return mount_nodev(type, flags, data, bpf_fill_super);
512}
513
514static struct file_system_type bpf_fs_type = {
515 .owner = THIS_MODULE,
516 .name = "bpf",
517 .mount = bpf_mount,
518 .kill_sb = kill_litter_super,
519};
520
521static int __init bpf_init(void)
522{
523 int ret;
524
525 ret = sysfs_create_mount_point(fs_kobj, "bpf");
526 if (ret)
527 return ret;
528
529 ret = register_filesystem(&bpf_fs_type);
530 if (ret)
531 sysfs_remove_mount_point(fs_kobj, "bpf");
532
533 return ret;
534}
535fs_initcall(bpf_init);