blob: a41343009ccce430b05e3df4430260fcb5dcb264 [file] [log] [blame]
Daniel Borkmannb2197752015-10-29 14:58:09 +01001/*
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
Paul Gortmakera536a6e2016-07-11 12:51:01 -040014#include <linux/init.h>
Daniel Borkmannb2197752015-10-29 14:58:09 +010015#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>
Daniel Borkmanna3af5f82016-11-26 01:28:08 +010021#include <linux/parser.h>
Daniel Borkmannb2197752015-10-29 14:58:09 +010022#include <linux/filter.h>
23#include <linux/bpf.h>
Daniel Borkmanna67edbf2017-01-25 02:28:18 +010024#include <linux/bpf_trace.h>
Daniel Borkmannb2197752015-10-29 14:58:09 +010025
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:
Alexei Starovoitov92117d82016-04-27 18:56:20 -070036 raw = bpf_prog_inc(raw);
Daniel Borkmannb2197752015-10-29 14:58:09 +010037 break;
38 case BPF_TYPE_MAP:
Alexei Starovoitov92117d82016-04-27 18:56:20 -070039 raw = bpf_map_inc(raw, true);
Daniel Borkmannb2197752015-10-29 14:58:09 +010040 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:
Daniel Borkmannc9da1612015-11-24 21:28:15 +010056 bpf_map_put_with_uref(raw);
Daniel Borkmannb2197752015-10-29 14:58:09 +010057 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;
Daniel Borkmannc9da1612015-11-24 21:28:15 +010069 raw = bpf_map_get_with_uref(ufd);
Daniel Borkmannb2197752015-10-29 14:58:09 +010070 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:
Daniel Borkmann0f986212016-10-29 02:30:46 +020092 case S_IFLNK:
Daniel Borkmannb2197752015-10-29 14:58:09 +010093 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();
Deepa Dinamani078cd822016-09-14 07:48:04 -0700103 inode->i_atime = current_time(inode);
Daniel Borkmannb2197752015-10-29 14:58:09 +0100104 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
Daniel Borkmann0f986212016-10-29 02:30:46 +0200125static 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
Daniel Borkmannb2197752015-10-29 14:58:09 +0100135static int bpf_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
136{
137 struct inode *inode;
138
Daniel Borkmannb2197752015-10-29 14:58:09 +0100139 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
Daniel Borkmann0f986212016-10-29 02:30:46 +0200149 bpf_dentry_finalize(dentry, inode, dir);
Daniel Borkmannb2197752015-10-29 14:58:09 +0100150 return 0;
151}
152
Martin KaFai Laua26ca7c2018-04-18 15:56:03 -0700153struct map_iter {
154 void *key;
155 bool done;
156};
157
158static struct map_iter *map_iter(struct seq_file *m)
159{
160 return m->private;
161}
162
163static struct bpf_map *seq_file_to_map(struct seq_file *m)
164{
165 return file_inode(m->file)->i_private;
166}
167
168static void map_iter_free(struct map_iter *iter)
169{
170 if (iter) {
171 kfree(iter->key);
172 kfree(iter);
173 }
174}
175
176static struct map_iter *map_iter_alloc(struct bpf_map *map)
177{
178 struct map_iter *iter;
179
180 iter = kzalloc(sizeof(*iter), GFP_KERNEL | __GFP_NOWARN);
181 if (!iter)
182 goto error;
183
184 iter->key = kzalloc(map->key_size, GFP_KERNEL | __GFP_NOWARN);
185 if (!iter->key)
186 goto error;
187
188 return iter;
189
190error:
191 map_iter_free(iter);
192 return NULL;
193}
194
195static void *map_seq_next(struct seq_file *m, void *v, loff_t *pos)
196{
197 struct bpf_map *map = seq_file_to_map(m);
198 void *key = map_iter(m)->key;
199
200 if (map_iter(m)->done)
201 return NULL;
202
203 if (unlikely(v == SEQ_START_TOKEN))
204 goto done;
205
206 if (map->ops->map_get_next_key(map, key, key)) {
207 map_iter(m)->done = true;
208 return NULL;
209 }
210
211done:
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
Al Viroa4a06832017-12-01 17:22:19 -0500298static int bpf_mkobj_ops(struct dentry *dentry, umode_t mode, void *raw,
Martin KaFai Laua26ca7c2018-04-18 15:56:03 -0700299 const struct inode_operations *iops,
300 const struct file_operations *fops)
Daniel Borkmannb2197752015-10-29 14:58:09 +0100301{
Al Viroa4a06832017-12-01 17:22:19 -0500302 struct inode *dir = dentry->d_parent->d_inode;
303 struct inode *inode = bpf_get_inode(dir->i_sb, dir, mode);
Daniel Borkmannb2197752015-10-29 14:58:09 +0100304 if (IS_ERR(inode))
305 return PTR_ERR(inode);
306
307 inode->i_op = iops;
Martin KaFai Laua26ca7c2018-04-18 15:56:03 -0700308 inode->i_fop = fops;
Al Viroa4a06832017-12-01 17:22:19 -0500309 inode->i_private = raw;
Daniel Borkmannb2197752015-10-29 14:58:09 +0100310
Daniel Borkmann0f986212016-10-29 02:30:46 +0200311 bpf_dentry_finalize(dentry, inode, dir);
Daniel Borkmannb2197752015-10-29 14:58:09 +0100312 return 0;
313}
314
Al Viroa4a06832017-12-01 17:22:19 -0500315static int bpf_mkprog(struct dentry *dentry, umode_t mode, void *arg)
Daniel Borkmannb2197752015-10-29 14:58:09 +0100316{
Martin KaFai Laua26ca7c2018-04-18 15:56:03 -0700317 return bpf_mkobj_ops(dentry, mode, arg, &bpf_prog_iops, NULL);
Al Viroa4a06832017-12-01 17:22:19 -0500318}
Daniel Borkmannb2197752015-10-29 14:58:09 +0100319
Al Viroa4a06832017-12-01 17:22:19 -0500320static int bpf_mkmap(struct dentry *dentry, umode_t mode, void *arg)
321{
Martin KaFai Laua26ca7c2018-04-18 15:56:03 -0700322 struct bpf_map *map = arg;
323
324 return bpf_mkobj_ops(dentry, mode, arg, &bpf_map_iops,
325 map->btf ? &bpffs_map_fops : NULL);
Daniel Borkmannb2197752015-10-29 14:58:09 +0100326}
327
Al Viro0c93b7d2016-03-25 12:06:51 -0400328static struct dentry *
329bpf_lookup(struct inode *dir, struct dentry *dentry, unsigned flags)
Daniel Borkmannbb35a6e2015-12-10 22:33:49 +0100330{
Quentin Monnet6d8cb042018-03-08 23:46:33 -0800331 /* Dots in names (e.g. "/sys/fs/bpf/foo.bar") are reserved for future
332 * extensions.
333 */
Al Viro0c93b7d2016-03-25 12:06:51 -0400334 if (strchr(dentry->d_name.name, '.'))
335 return ERR_PTR(-EPERM);
Daniel Borkmann0f986212016-10-29 02:30:46 +0200336
Al Viro0c93b7d2016-03-25 12:06:51 -0400337 return simple_lookup(dir, dentry, flags);
Daniel Borkmannbb35a6e2015-12-10 22:33:49 +0100338}
339
Daniel Borkmann0f986212016-10-29 02:30:46 +0200340static int bpf_symlink(struct inode *dir, struct dentry *dentry,
341 const char *target)
342{
343 char *link = kstrdup(target, GFP_USER | __GFP_NOWARN);
344 struct inode *inode;
345
346 if (!link)
347 return -ENOMEM;
348
349 inode = bpf_get_inode(dir->i_sb, dir, S_IRWXUGO | S_IFLNK);
350 if (IS_ERR(inode)) {
351 kfree(link);
352 return PTR_ERR(inode);
353 }
354
355 inode->i_op = &simple_symlink_inode_operations;
356 inode->i_link = link;
357
358 bpf_dentry_finalize(dentry, inode, dir);
359 return 0;
360}
361
Daniel Borkmannb2197752015-10-29 14:58:09 +0100362static const struct inode_operations bpf_dir_iops = {
Al Viro0c93b7d2016-03-25 12:06:51 -0400363 .lookup = bpf_lookup,
Daniel Borkmannb2197752015-10-29 14:58:09 +0100364 .mkdir = bpf_mkdir,
Daniel Borkmann0f986212016-10-29 02:30:46 +0200365 .symlink = bpf_symlink,
Daniel Borkmannb2197752015-10-29 14:58:09 +0100366 .rmdir = simple_rmdir,
Al Viro0c93b7d2016-03-25 12:06:51 -0400367 .rename = simple_rename,
368 .link = simple_link,
Daniel Borkmannb2197752015-10-29 14:58:09 +0100369 .unlink = simple_unlink,
370};
371
372static int bpf_obj_do_pin(const struct filename *pathname, void *raw,
373 enum bpf_type type)
374{
375 struct dentry *dentry;
376 struct inode *dir;
377 struct path path;
378 umode_t mode;
Daniel Borkmannb2197752015-10-29 14:58:09 +0100379 int ret;
380
381 dentry = kern_path_create(AT_FDCWD, pathname->name, &path, 0);
382 if (IS_ERR(dentry))
383 return PTR_ERR(dentry);
384
385 mode = S_IFREG | ((S_IRUSR | S_IWUSR) & ~current_umask());
Daniel Borkmannb2197752015-10-29 14:58:09 +0100386
Al Viroa4a06832017-12-01 17:22:19 -0500387 ret = security_path_mknod(&path, dentry, mode, 0);
Daniel Borkmannb2197752015-10-29 14:58:09 +0100388 if (ret)
389 goto out;
390
391 dir = d_inode(path.dentry);
392 if (dir->i_op != &bpf_dir_iops) {
393 ret = -EPERM;
394 goto out;
395 }
396
Al Viroa4a06832017-12-01 17:22:19 -0500397 switch (type) {
398 case BPF_TYPE_PROG:
399 ret = vfs_mkobj(dentry, mode, bpf_mkprog, raw);
400 break;
401 case BPF_TYPE_MAP:
402 ret = vfs_mkobj(dentry, mode, bpf_mkmap, raw);
403 break;
404 default:
405 ret = -EPERM;
406 }
Daniel Borkmannb2197752015-10-29 14:58:09 +0100407out:
408 done_path_create(&path, dentry);
409 return ret;
410}
411
412int bpf_obj_pin_user(u32 ufd, const char __user *pathname)
413{
414 struct filename *pname;
415 enum bpf_type type;
416 void *raw;
417 int ret;
418
419 pname = getname(pathname);
420 if (IS_ERR(pname))
421 return PTR_ERR(pname);
422
423 raw = bpf_fd_probe_obj(ufd, &type);
424 if (IS_ERR(raw)) {
425 ret = PTR_ERR(raw);
426 goto out;
427 }
428
429 ret = bpf_obj_do_pin(pname, raw, type);
430 if (ret != 0)
431 bpf_any_put(raw, type);
Daniel Borkmanna67edbf2017-01-25 02:28:18 +0100432 if ((trace_bpf_obj_pin_prog_enabled() ||
433 trace_bpf_obj_pin_map_enabled()) && !ret) {
434 if (type == BPF_TYPE_PROG)
435 trace_bpf_obj_pin_prog(raw, ufd, pname);
436 if (type == BPF_TYPE_MAP)
437 trace_bpf_obj_pin_map(raw, ufd, pname);
438 }
Daniel Borkmannb2197752015-10-29 14:58:09 +0100439out:
440 putname(pname);
441 return ret;
442}
443
444static void *bpf_obj_do_get(const struct filename *pathname,
Chenbo Feng6e71b042017-10-18 13:00:22 -0700445 enum bpf_type *type, int flags)
Daniel Borkmannb2197752015-10-29 14:58:09 +0100446{
447 struct inode *inode;
448 struct path path;
449 void *raw;
450 int ret;
451
452 ret = kern_path(pathname->name, LOOKUP_FOLLOW, &path);
453 if (ret)
454 return ERR_PTR(ret);
455
456 inode = d_backing_inode(path.dentry);
Chenbo Feng6e71b042017-10-18 13:00:22 -0700457 ret = inode_permission(inode, ACC_MODE(flags));
Daniel Borkmannb2197752015-10-29 14:58:09 +0100458 if (ret)
459 goto out;
460
461 ret = bpf_inode_type(inode, type);
462 if (ret)
463 goto out;
464
465 raw = bpf_any_get(inode->i_private, *type);
Alexei Starovoitov92117d82016-04-27 18:56:20 -0700466 if (!IS_ERR(raw))
467 touch_atime(&path);
Daniel Borkmannb2197752015-10-29 14:58:09 +0100468
469 path_put(&path);
470 return raw;
471out:
472 path_put(&path);
473 return ERR_PTR(ret);
474}
475
Chenbo Feng6e71b042017-10-18 13:00:22 -0700476int bpf_obj_get_user(const char __user *pathname, int flags)
Daniel Borkmannb2197752015-10-29 14:58:09 +0100477{
478 enum bpf_type type = BPF_TYPE_UNSPEC;
479 struct filename *pname;
480 int ret = -ENOENT;
Chenbo Feng6e71b042017-10-18 13:00:22 -0700481 int f_flags;
Daniel Borkmannb2197752015-10-29 14:58:09 +0100482 void *raw;
483
Chenbo Feng6e71b042017-10-18 13:00:22 -0700484 f_flags = bpf_get_file_flag(flags);
485 if (f_flags < 0)
486 return f_flags;
487
Daniel Borkmannb2197752015-10-29 14:58:09 +0100488 pname = getname(pathname);
489 if (IS_ERR(pname))
490 return PTR_ERR(pname);
491
Chenbo Feng6e71b042017-10-18 13:00:22 -0700492 raw = bpf_obj_do_get(pname, &type, f_flags);
Daniel Borkmannb2197752015-10-29 14:58:09 +0100493 if (IS_ERR(raw)) {
494 ret = PTR_ERR(raw);
495 goto out;
496 }
497
498 if (type == BPF_TYPE_PROG)
499 ret = bpf_prog_new_fd(raw);
500 else if (type == BPF_TYPE_MAP)
Chenbo Feng6e71b042017-10-18 13:00:22 -0700501 ret = bpf_map_new_fd(raw, f_flags);
Daniel Borkmannb2197752015-10-29 14:58:09 +0100502 else
503 goto out;
504
Daniel Borkmanna67edbf2017-01-25 02:28:18 +0100505 if (ret < 0) {
Daniel Borkmannb2197752015-10-29 14:58:09 +0100506 bpf_any_put(raw, type);
Daniel Borkmanna67edbf2017-01-25 02:28:18 +0100507 } else if (trace_bpf_obj_get_prog_enabled() ||
508 trace_bpf_obj_get_map_enabled()) {
509 if (type == BPF_TYPE_PROG)
510 trace_bpf_obj_get_prog(raw, ret, pname);
511 if (type == BPF_TYPE_MAP)
512 trace_bpf_obj_get_map(raw, ret, pname);
513 }
Daniel Borkmannb2197752015-10-29 14:58:09 +0100514out:
515 putname(pname);
516 return ret;
517}
Al Viro040ee692017-12-02 20:20:38 -0500518
519static struct bpf_prog *__get_prog_inode(struct inode *inode, enum bpf_prog_type type)
520{
521 struct bpf_prog *prog;
522 int ret = inode_permission(inode, MAY_READ | MAY_WRITE);
523 if (ret)
524 return ERR_PTR(ret);
525
526 if (inode->i_op == &bpf_map_iops)
527 return ERR_PTR(-EINVAL);
528 if (inode->i_op != &bpf_prog_iops)
529 return ERR_PTR(-EACCES);
530
531 prog = inode->i_private;
532
533 ret = security_bpf_prog(prog);
534 if (ret < 0)
535 return ERR_PTR(ret);
536
537 if (!bpf_prog_get_ok(prog, &type, false))
538 return ERR_PTR(-EINVAL);
539
540 return bpf_prog_inc(prog);
541}
542
543struct bpf_prog *bpf_prog_get_type_path(const char *name, enum bpf_prog_type type)
544{
545 struct bpf_prog *prog;
546 struct path path;
547 int ret = kern_path(name, LOOKUP_FOLLOW, &path);
548 if (ret)
549 return ERR_PTR(ret);
550 prog = __get_prog_inode(d_backing_inode(path.dentry), type);
551 if (!IS_ERR(prog))
552 touch_atime(&path);
553 path_put(&path);
554 return prog;
555}
556EXPORT_SYMBOL(bpf_prog_get_type_path);
Daniel Borkmannb2197752015-10-29 14:58:09 +0100557
558static void bpf_evict_inode(struct inode *inode)
559{
560 enum bpf_type type;
561
562 truncate_inode_pages_final(&inode->i_data);
563 clear_inode(inode);
564
Daniel Borkmann0f986212016-10-29 02:30:46 +0200565 if (S_ISLNK(inode->i_mode))
566 kfree(inode->i_link);
Daniel Borkmannb2197752015-10-29 14:58:09 +0100567 if (!bpf_inode_type(inode, &type))
568 bpf_any_put(inode->i_private, type);
569}
570
David Howells4cc7c182017-07-05 16:24:49 +0100571/*
572 * Display the mount options in /proc/mounts.
573 */
574static int bpf_show_options(struct seq_file *m, struct dentry *root)
575{
576 umode_t mode = d_inode(root)->i_mode & S_IALLUGO & ~S_ISVTX;
577
578 if (mode != S_IRWXUGO)
579 seq_printf(m, ",mode=%o", mode);
580 return 0;
581}
582
Daniel Borkmannb2197752015-10-29 14:58:09 +0100583static const struct super_operations bpf_super_ops = {
584 .statfs = simple_statfs,
585 .drop_inode = generic_delete_inode,
David Howells4cc7c182017-07-05 16:24:49 +0100586 .show_options = bpf_show_options,
Daniel Borkmannb2197752015-10-29 14:58:09 +0100587 .evict_inode = bpf_evict_inode,
588};
589
Daniel Borkmanna3af5f82016-11-26 01:28:08 +0100590enum {
591 OPT_MODE,
592 OPT_ERR,
593};
594
595static const match_table_t bpf_mount_tokens = {
596 { OPT_MODE, "mode=%o" },
597 { OPT_ERR, NULL },
598};
599
600struct bpf_mount_opts {
601 umode_t mode;
602};
603
604static int bpf_parse_options(char *data, struct bpf_mount_opts *opts)
605{
606 substring_t args[MAX_OPT_ARGS];
607 int option, token;
608 char *ptr;
609
610 opts->mode = S_IRWXUGO;
611
612 while ((ptr = strsep(&data, ",")) != NULL) {
613 if (!*ptr)
614 continue;
615
616 token = match_token(ptr, bpf_mount_tokens, args);
617 switch (token) {
618 case OPT_MODE:
619 if (match_octal(&args[0], &option))
620 return -EINVAL;
621 opts->mode = option & S_IALLUGO;
622 break;
623 /* We might like to report bad mount options here, but
624 * traditionally we've ignored all mount options, so we'd
625 * better continue to ignore non-existing options for bpf.
626 */
627 }
628 }
629
630 return 0;
631}
632
Daniel Borkmannb2197752015-10-29 14:58:09 +0100633static int bpf_fill_super(struct super_block *sb, void *data, int silent)
634{
Eric Biggerscda37122017-03-25 21:15:37 -0700635 static const struct tree_descr bpf_rfiles[] = { { "" } };
Daniel Borkmanna3af5f82016-11-26 01:28:08 +0100636 struct bpf_mount_opts opts;
Daniel Borkmannb2197752015-10-29 14:58:09 +0100637 struct inode *inode;
638 int ret;
639
Daniel Borkmanna3af5f82016-11-26 01:28:08 +0100640 ret = bpf_parse_options(data, &opts);
641 if (ret)
642 return ret;
643
Daniel Borkmannb2197752015-10-29 14:58:09 +0100644 ret = simple_fill_super(sb, BPF_FS_MAGIC, bpf_rfiles);
645 if (ret)
646 return ret;
647
648 sb->s_op = &bpf_super_ops;
649
650 inode = sb->s_root->d_inode;
651 inode->i_op = &bpf_dir_iops;
652 inode->i_mode &= ~S_IALLUGO;
Daniel Borkmanna3af5f82016-11-26 01:28:08 +0100653 inode->i_mode |= S_ISVTX | opts.mode;
Daniel Borkmannb2197752015-10-29 14:58:09 +0100654
655 return 0;
656}
657
658static struct dentry *bpf_mount(struct file_system_type *type, int flags,
659 const char *dev_name, void *data)
660{
Eric W. Biedermane27f4a92016-05-20 17:22:48 -0500661 return mount_nodev(type, flags, data, bpf_fill_super);
Daniel Borkmannb2197752015-10-29 14:58:09 +0100662}
663
664static struct file_system_type bpf_fs_type = {
665 .owner = THIS_MODULE,
666 .name = "bpf",
667 .mount = bpf_mount,
668 .kill_sb = kill_litter_super,
Daniel Borkmannb2197752015-10-29 14:58:09 +0100669};
670
Daniel Borkmannb2197752015-10-29 14:58:09 +0100671static int __init bpf_init(void)
672{
673 int ret;
674
675 ret = sysfs_create_mount_point(fs_kobj, "bpf");
676 if (ret)
677 return ret;
678
679 ret = register_filesystem(&bpf_fs_type);
680 if (ret)
681 sysfs_remove_mount_point(fs_kobj, "bpf");
682
683 return ret;
684}
685fs_initcall(bpf_init);