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
24enum bpf_type {
25 BPF_TYPE_UNSPEC = 0,
26 BPF_TYPE_PROG,
27 BPF_TYPE_MAP,
28};
29
30static void *bpf_any_get(void *raw, enum bpf_type type)
31{
32 switch (type) {
33 case BPF_TYPE_PROG:
34 raw = bpf_prog_inc(raw);
35 break;
36 case BPF_TYPE_MAP:
37 raw = bpf_map_inc(raw, true);
38 break;
39 default:
40 WARN_ON_ONCE(1);
41 break;
42 }
43
44 return raw;
45}
46
47static void bpf_any_put(void *raw, enum bpf_type type)
48{
49 switch (type) {
50 case BPF_TYPE_PROG:
51 bpf_prog_put(raw);
52 break;
53 case BPF_TYPE_MAP:
54 bpf_map_put_with_uref(raw);
55 break;
56 default:
57 WARN_ON_ONCE(1);
58 break;
59 }
60}
61
62static void *bpf_fd_probe_obj(u32 ufd, enum bpf_type *type)
63{
64 void *raw;
65
66 *type = BPF_TYPE_MAP;
67 raw = bpf_map_get_with_uref(ufd);
68 if (IS_ERR(raw)) {
69 *type = BPF_TYPE_PROG;
70 raw = bpf_prog_get(ufd);
71 }
72
73 return raw;
74}
75
76static const struct inode_operations bpf_dir_iops;
77
78static const struct inode_operations bpf_prog_iops = { };
79static const struct inode_operations bpf_map_iops = { };
80
81static struct inode *bpf_get_inode(struct super_block *sb,
82 const struct inode *dir,
83 umode_t mode)
84{
85 struct inode *inode;
86
87 switch (mode & S_IFMT) {
88 case S_IFDIR:
89 case S_IFREG:
90 case S_IFLNK:
91 break;
92 default:
93 return ERR_PTR(-EINVAL);
94 }
95
96 inode = new_inode(sb);
97 if (!inode)
98 return ERR_PTR(-ENOSPC);
99
100 inode->i_ino = get_next_ino();
101 inode->i_atime = current_time(inode);
102 inode->i_mtime = inode->i_atime;
103 inode->i_ctime = inode->i_atime;
104
105 inode_init_owner(inode, dir, mode);
106
107 return inode;
108}
109
110static int bpf_inode_type(const struct inode *inode, enum bpf_type *type)
111{
112 *type = BPF_TYPE_UNSPEC;
113 if (inode->i_op == &bpf_prog_iops)
114 *type = BPF_TYPE_PROG;
115 else if (inode->i_op == &bpf_map_iops)
116 *type = BPF_TYPE_MAP;
117 else
118 return -EACCES;
119
120 return 0;
121}
122
123static void bpf_dentry_finalize(struct dentry *dentry, struct inode *inode,
124 struct inode *dir)
125{
126 d_instantiate(dentry, inode);
127 dget(dentry);
128
129 dir->i_mtime = current_time(dir);
130 dir->i_ctime = dir->i_mtime;
131}
132
133static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
134{
135 struct inode *inode;
136
137 inode = bpf_get_inode(dir->i_sb, dir, mode | S_IFDIR);
138 if (IS_ERR(inode))
139 return PTR_ERR(inode);
140
141 inode->i_op = &bpf_dir_iops;
142 inode->i_fop = &simple_dir_operations;
143
144 inc_nlink(inode);
145 inc_nlink(dir);
146
147 bpf_dentry_finalize(dentry, inode, dir);
148 return 0;
149}
150
151struct map_iter {
152 void *key;
153 bool done;
154};
155
156static struct map_iter *map_iter(struct seq_file *m)
157{
158 return m->private;
159}
160
161static struct bpf_map *seq_file_to_map(struct seq_file *m)
162{
163 return file_inode(m->file)->i_private;
164}
165
166static void map_iter_free(struct map_iter *iter)
167{
168 if (iter) {
169 kfree(iter->key);
170 kfree(iter);
171 }
172}
173
174static struct map_iter *map_iter_alloc(struct bpf_map *map)
175{
176 struct map_iter *iter;
177
178 iter = kzalloc(sizeof(*iter), GFP_KERNEL | __GFP_NOWARN);
179 if (!iter)
180 goto error;
181
182 iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN);
183 if (!iter->key)
184 goto error;
185
186 return iter;
187
188error:
189 map_iter_free(iter);
190 return NULL;
191}
192
193static void *map_seq_next(struct seq_file *m, void *v, loff_t *pos)
194{
195 struct bpf_map *map = seq_file_to_map(m);
196 void *key = map_iter(m)->key;
197 void *prev_key;
198
199 if (map_iter(m)->done)
200 return NULL;
201
202 if (unlikely(v == SEQ_START_TOKEN))
203 prev_key = NULL;
204 else
205 prev_key = key;
206
207 if (map->ops->map_get_next_key(map, prev_key, key)) {
208 map_iter(m)->done = true;
209 return NULL;
210 }
211
212 ++(*pos);
213 return key;
214}
215
216static void *map_seq_start(struct seq_file *m, loff_t *pos)
217{
218 if (map_iter(m)->done)
219 return NULL;
220
221 return *pos ? map_iter(m)->key : SEQ_START_TOKEN;
222}
223
224static void map_seq_stop(struct seq_file *m, void *v)
225{
226}
227
228static int map_seq_show(struct seq_file *m, void *v)
229{
230 struct bpf_map *map = seq_file_to_map(m);
231 void *key = map_iter(m)->key;
232
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");
236 } else {
237 map->ops->map_seq_show_elem(map, key, m);
238 }
239
240 return 0;
241}
242
243static 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,
248};
249
250static int bpffs_map_open(struct inode *inode, struct file *file)
251{
252 struct bpf_map *map = inode->i_private;
253 struct map_iter *iter;
254 struct seq_file *m;
255 int err;
256
257 iter = map_iter_alloc(map);
258 if (!iter)
259 return -ENOMEM;
260
261 err = seq_open(file, &bpffs_map_seq_ops);
262 if (err) {
263 map_iter_free(iter);
264 return err;
265 }
266
267 m = file->private_data;
268 m->private = iter;
269
270 return 0;
271}
272
273static int bpffs_map_release(struct inode *inode, struct file *file)
274{
275 struct seq_file *m = file->private_data;
276
277 map_iter_free(map_iter(m));
278
279 return seq_release(inode, file);
280}
281
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".
286 *
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
290 * interface.
291 */
292static const struct file_operations bpffs_map_fops = {
293 .open = bpffs_map_open,
294 .read = seq_read,
295 .release = bpffs_map_release,
296};
297
298static int bpffs_obj_open(struct inode *inode, struct file *file)
299{
300 return -EIO;
301}
302
303static const struct file_operations bpffs_obj_fops = {
304 .open = bpffs_obj_open,
305};
306
307static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
308 const struct inode_operations *iops,
309 const struct file_operations *fops)
310{
311 struct inode *dir = dentry->d_parent->d_inode;
312 struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
313 if (IS_ERR(inode))
314 return PTR_ERR(inode);
315
316 inode->i_op = iops;
317 inode->i_fop = fops;
318 inode->i_private = raw;
319
320 bpf_dentry_finalize(dentry, inode, dir);
321 return 0;
322}
323
324static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
325{
326 return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops,
327 &bpffs_obj_fops);
328}
329
330static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
331{
332 struct bpf_map *map = arg;
333
334 return bpf_mkobj_ops(dentry, mode, arg, &bpf_map_iops,
335 bpf_map_support_seq_show(map) ?
336 &bpffs_map_fops : &bpffs_obj_fops);
337}
338
339static struct dentry *
340bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
341{
342 /* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
343 * extensions.
344 */
345 if (strchr(dentry->d_name.name, '.'))
346 return ERR_PTR(-EPERM);
347
348 return simple_lookup(dir, dentry, flags);
349}
350
351static int bpf_symlink(struct inode *dir, struct dentry *dentry,
352 const char *target)
353{
354 char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
355 struct inode *inode;
356
357 if (!link)
358 return -ENOMEM;
359
360 inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
361 if (IS_ERR(inode)) {
362 kfree(link);
363 return PTR_ERR(inode);
364 }
365
366 inode->i_op = &simple_symlink_inode_operations;
367 inode->i_link = link;
368
369 bpf_dentry_finalize(dentry, inode, dir);
370 return 0;
371}
372
373static const struct inode_operations bpf_dir_iops = {
374 .lookup = bpf_lookup,
375 .mkdir = bpf_mkdir,
376 .symlink = bpf_symlink,
377 .rmdir = simple_rmdir,
378 .rename = simple_rename,
379 .link = simple_link,
380 .unlink = simple_unlink,
381};
382
383static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
384 enum bpf_type type)
385{
386 struct dentry *dentry;
387 struct inode *dir;
388 struct path path;
389 umode_t mode;
390 int ret;
391
392 dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
393 if (IS_ERR(dentry))
394 return PTR_ERR(dentry);
395
396 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
397
398 ret = security_path_mknod(&path, dentry, mode, 0);
399 if (ret)
400 goto out;
401
402 dir = d_inode(path.dentry);
403 if (dir->i_op != &bpf_dir_iops) {
404 ret = -EPERM;
405 goto out;
406 }
407
408 switch (type) {
409 case BPF_TYPE_PROG:
410 ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
411 break;
412 case BPF_TYPE_MAP:
413 ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
414 break;
415 default:
416 ret = -EPERM;
417 }
418out:
419 done_path_create(&path, dentry);
420 return ret;
421}
422
423int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
424{
425 struct filename *pname;
426 enum bpf_type type;
427 void *raw;
428 int ret;
429
430 pname = getname(pathname);
431 if (IS_ERR(pname))
432 return PTR_ERR(pname);
433
434 raw = bpf_fd_probe_obj(ufd, &type);
435 if (IS_ERR(raw)) {
436 ret = PTR_ERR(raw);
437 goto out;
438 }
439
440 ret = bpf_obj_do_pin(pname, raw, type);
441 if (ret != 0)
442 bpf_any_put(raw, type);
443out:
444 putname(pname);
445 return ret;
446}
447
448static void *bpf_obj_do_get(const struct filename *pathname,
449 enum bpf_type *type, int flags)
450{
451 struct inode *inode;
452 struct path path;
453 void *raw;
454 int ret;
455
456 ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
457 if (ret)
458 return ERR_PTR(ret);
459
460 inode = d_backing_inode(path.dentry);
461 ret = inode_permission(inode, ACC_MODE(flags));
462 if (ret)
463 goto out;
464
465 ret = bpf_inode_type(inode, type);
466 if (ret)
467 goto out;
468
469 raw = bpf_any_get(inode->i_private, *type);
470 if (!IS_ERR(raw))
471 touch_atime(&path);
472
473 path_put(&path);
474 return raw;
475out:
476 path_put(&path);
477 return ERR_PTR(ret);
478}
479
480int bpf_obj_get_user(const char __user *pathname, int flags)
481{
482 enum bpf_type type = BPF_TYPE_UNSPEC;
483 struct filename *pname;
484 int ret = -ENOENT;
485 int f_flags;
486 void *raw;
487
488 f_flags = bpf_get_file_flag(flags);
489 if (f_flags < 0)
490 return f_flags;
491
492 pname = getname(pathname);
493 if (IS_ERR(pname))
494 return PTR_ERR(pname);
495
496 raw = bpf_obj_do_get(pname, &type, f_flags);
497 if (IS_ERR(raw)) {
498 ret = PTR_ERR(raw);
499 goto out;
500 }
501
502 if (type == BPF_TYPE_PROG)
503 ret = bpf_prog_new_fd(raw);
504 else if (type == BPF_TYPE_MAP)
505 ret = bpf_map_new_fd(raw, f_flags);
506 else
507 goto out;
508
509 if (ret < 0)
510 bpf_any_put(raw, type);
511out:
512 putname(pname);
513 return ret;
514}
515
516static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
517{
518 struct bpf_prog *prog;
519 int ret = inode_permission(inode, MAY_READ);
520 if (ret)
521 return ERR_PTR(ret);
522
523 if (inode->i_op == &bpf_map_iops)
524 return ERR_PTR(-EINVAL);
525 if (inode->i_op != &bpf_prog_iops)
526 return ERR_PTR(-EACCES);
527
528 prog = inode->i_private;
529
530 ret = security_bpf_prog(prog);
531 if (ret < 0)
532 return ERR_PTR(ret);
533
534 if (!bpf_prog_get_ok(prog, &type, false))
535 return ERR_PTR(-EINVAL);
536
537 return bpf_prog_inc(prog);
538}
539
540struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
541{
542 struct bpf_prog *prog;
543 struct path path;
544 int ret = kern_path(name, LOOKUP_FOLLOW, &path);
545 if (ret)
546 return ERR_PTR(ret);
547 prog = __get_prog_inode(d_backing_inode(path.dentry), type);
548 if (!IS_ERR(prog))
549 touch_atime(&path);
550 path_put(&path);
551 return prog;
552}
553EXPORT_SYMBOL(bpf_prog_get_type_path);
554
555/*
556 * Display the mount options in /proc/mounts.
557 */
558static int bpf_show_options(struct seq_file *m, struct dentry *root)
559{
560 umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
561
562 if (mode != S_IRWXUGO)
563 seq_printf(m, ",mode=%o", mode);
564 return 0;
565}
566
567static void bpf_free_inode(struct inode *inode)
568{
569 enum bpf_type type;
570
571 if (S_ISLNK(inode->i_mode))
572 kfree(inode->i_link);
573 if (!bpf_inode_type(inode, &type))
574 bpf_any_put(inode->i_private, type);
575 free_inode_nonrcu(inode);
576}
577
578static const struct super_operations bpf_super_ops = {
579 .statfs = simple_statfs,
580 .drop_inode = generic_delete_inode,
581 .show_options = bpf_show_options,
582 .free_inode = bpf_free_inode,
583};
584
585enum {
586 OPT_MODE,
587};
588
589static const struct fs_parameter_spec bpf_param_specs[] = {
590 fsparam_u32oct ("mode", OPT_MODE),
591 {}
592};
593
594static const struct fs_parameter_description bpf_fs_parameters = {
595 .name = "bpf",
596 .specs = bpf_param_specs,
597};
598
599struct bpf_mount_opts {
600 umode_t mode;
601};
602
603static int bpf_parse_param(struct fs_context *fc, struct fs_parameter *param)
604{
605 struct bpf_mount_opts *opts = fc->fs_private;
606 struct fs_parse_result result;
607 int opt;
608
609 opt = fs_parse(fc, &bpf_fs_parameters, param, &result);
610 if (opt < 0)
611 /* We might like to report bad mount options here, but
612 * traditionally we've ignored all mount options, so we'd
613 * better continue to ignore non-existing options for bpf.
614 */
615 return opt == -ENOPARAM ? 0 : opt;
616
617 switch (opt) {
618 case OPT_MODE:
619 opts->mode = result.uint_32 & S_IALLUGO;
620 break;
621 }
622
623 return 0;
624}
625
626static int bpf_fill_super(struct super_block *sb, struct fs_context *fc)
627{
628 static const struct tree_descr bpf_rfiles[] = { { "" } };
629 struct bpf_mount_opts *opts = fc->fs_private;
630 struct inode *inode;
631 int ret;
632
633 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
634 if (ret)
635 return ret;
636
637 sb->s_op = &bpf_super_ops;
638
639 inode = sb->s_root->d_inode;
640 inode->i_op = &bpf_dir_iops;
641 inode->i_mode &= ~S_IALLUGO;
642 inode->i_mode |= S_ISVTX | opts->mode;
643
644 return 0;
645}
646
647static int bpf_get_tree(struct fs_context *fc)
648{
649 return get_tree_nodev(fc, bpf_fill_super);
650}
651
652static void bpf_free_fc(struct fs_context *fc)
653{
654 kfree(fc->fs_private);
655}
656
657static const struct fs_context_operations bpf_context_ops = {
658 .free = bpf_free_fc,
659 .parse_param = bpf_parse_param,
660 .get_tree = bpf_get_tree,
661};
662
663/*
664 * Set up the filesystem mount context.
665 */
666static int bpf_init_fs_context(struct fs_context *fc)
667{
668 struct bpf_mount_opts *opts;
669
670 opts = kzalloc(sizeof(struct bpf_mount_opts), GFP_KERNEL);
671 if (!opts)
672 return -ENOMEM;
673
674 opts->mode = S_IRWXUGO;
675
676 fc->fs_private = opts;
677 fc->ops = &bpf_context_ops;
678 return 0;
679}
680
681static struct file_system_type bpf_fs_type = {
682 .owner = THIS_MODULE,
683 .name = "bpf",
684 .init_fs_context = bpf_init_fs_context,
685 .parameters = &bpf_fs_parameters,
686 .kill_sb = kill_litter_super,
687};
688
689static int __init bpf_init(void)
690{
691 int ret;
692
693 ret = sysfs_create_mount_point(fs_kobj, "bpf");
694 if (ret)
695 return ret;
696
697 ret = register_filesystem(&bpf_fs_type);
698 if (ret)
699 sysfs_remove_mount_point(fs_kobj, "bpf");
700
701 return ret;
702}
703fs_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);