xref: /openbmc/linux/kernel/bpf/offload.c (revision e2ad626f)
1 /*
2  * Copyright (C) 2017-2018 Netronome Systems, Inc.
3  *
4  * This software is licensed under the GNU General License Version 2,
5  * June 1991 as shown in the file COPYING in the top-level directory of this
6  * source tree.
7  *
8  * THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS"
9  * WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING,
10  * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
11  * FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE
12  * OF THE PROGRAM IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME
13  * THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
14  */
15 
16 #include <linux/bpf.h>
17 #include <linux/bpf_verifier.h>
18 #include <linux/bug.h>
19 #include <linux/kdev_t.h>
20 #include <linux/list.h>
21 #include <linux/lockdep.h>
22 #include <linux/netdevice.h>
23 #include <linux/printk.h>
24 #include <linux/proc_ns.h>
25 #include <linux/rhashtable.h>
26 #include <linux/rtnetlink.h>
27 #include <linux/rwsem.h>
28 #include <net/xdp.h>
29 
30 /* Protects offdevs, members of bpf_offload_netdev and offload members
31  * of all progs.
32  * RTNL lock cannot be taken when holding this lock.
33  */
34 static DECLARE_RWSEM(bpf_devs_lock);
35 
36 struct bpf_offload_dev {
37 	const struct bpf_prog_offload_ops *ops;
38 	struct list_head netdevs;
39 	void *priv;
40 };
41 
42 struct bpf_offload_netdev {
43 	struct rhash_head l;
44 	struct net_device *netdev;
45 	struct bpf_offload_dev *offdev; /* NULL when bound-only */
46 	struct list_head progs;
47 	struct list_head maps;
48 	struct list_head offdev_netdevs;
49 };
50 
51 static const struct rhashtable_params offdevs_params = {
52 	.nelem_hint		= 4,
53 	.key_len		= sizeof(struct net_device *),
54 	.key_offset		= offsetof(struct bpf_offload_netdev, netdev),
55 	.head_offset		= offsetof(struct bpf_offload_netdev, l),
56 	.automatic_shrinking	= true,
57 };
58 
59 static struct rhashtable offdevs;
60 
61 static int bpf_dev_offload_check(struct net_device *netdev)
62 {
63 	if (!netdev)
64 		return -EINVAL;
65 	if (!netdev->netdev_ops->ndo_bpf)
66 		return -EOPNOTSUPP;
67 	return 0;
68 }
69 
70 static struct bpf_offload_netdev *
71 bpf_offload_find_netdev(struct net_device *netdev)
72 {
73 	lockdep_assert_held(&bpf_devs_lock);
74 
75 	return rhashtable_lookup_fast(&offdevs, &netdev, offdevs_params);
76 }
77 
78 static int __bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev,
79 					     struct net_device *netdev)
80 {
81 	struct bpf_offload_netdev *ondev;
82 	int err;
83 
84 	ondev = kzalloc(sizeof(*ondev), GFP_KERNEL);
85 	if (!ondev)
86 		return -ENOMEM;
87 
88 	ondev->netdev = netdev;
89 	ondev->offdev = offdev;
90 	INIT_LIST_HEAD(&ondev->progs);
91 	INIT_LIST_HEAD(&ondev->maps);
92 
93 	err = rhashtable_insert_fast(&offdevs, &ondev->l, offdevs_params);
94 	if (err) {
95 		netdev_warn(netdev, "failed to register for BPF offload\n");
96 		goto err_free;
97 	}
98 
99 	if (offdev)
100 		list_add(&ondev->offdev_netdevs, &offdev->netdevs);
101 	return 0;
102 
103 err_free:
104 	kfree(ondev);
105 	return err;
106 }
107 
108 static void __bpf_prog_offload_destroy(struct bpf_prog *prog)
109 {
110 	struct bpf_prog_offload *offload = prog->aux->offload;
111 
112 	if (offload->dev_state)
113 		offload->offdev->ops->destroy(prog);
114 
115 	list_del_init(&offload->offloads);
116 	kfree(offload);
117 	prog->aux->offload = NULL;
118 }
119 
120 static int bpf_map_offload_ndo(struct bpf_offloaded_map *offmap,
121 			       enum bpf_netdev_command cmd)
122 {
123 	struct netdev_bpf data = {};
124 	struct net_device *netdev;
125 
126 	ASSERT_RTNL();
127 
128 	data.command = cmd;
129 	data.offmap = offmap;
130 	/* Caller must make sure netdev is valid */
131 	netdev = offmap->netdev;
132 
133 	return netdev->netdev_ops->ndo_bpf(netdev, &data);
134 }
135 
136 static void __bpf_map_offload_destroy(struct bpf_offloaded_map *offmap)
137 {
138 	WARN_ON(bpf_map_offload_ndo(offmap, BPF_OFFLOAD_MAP_FREE));
139 	/* Make sure BPF_MAP_GET_NEXT_ID can't find this dead map */
140 	bpf_map_free_id(&offmap->map);
141 	list_del_init(&offmap->offloads);
142 	offmap->netdev = NULL;
143 }
144 
145 static void __bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev,
146 						struct net_device *netdev)
147 {
148 	struct bpf_offload_netdev *ondev, *altdev = NULL;
149 	struct bpf_offloaded_map *offmap, *mtmp;
150 	struct bpf_prog_offload *offload, *ptmp;
151 
152 	ASSERT_RTNL();
153 
154 	ondev = rhashtable_lookup_fast(&offdevs, &netdev, offdevs_params);
155 	if (WARN_ON(!ondev))
156 		return;
157 
158 	WARN_ON(rhashtable_remove_fast(&offdevs, &ondev->l, offdevs_params));
159 
160 	/* Try to move the objects to another netdev of the device */
161 	if (offdev) {
162 		list_del(&ondev->offdev_netdevs);
163 		altdev = list_first_entry_or_null(&offdev->netdevs,
164 						  struct bpf_offload_netdev,
165 						  offdev_netdevs);
166 	}
167 
168 	if (altdev) {
169 		list_for_each_entry(offload, &ondev->progs, offloads)
170 			offload->netdev = altdev->netdev;
171 		list_splice_init(&ondev->progs, &altdev->progs);
172 
173 		list_for_each_entry(offmap, &ondev->maps, offloads)
174 			offmap->netdev = altdev->netdev;
175 		list_splice_init(&ondev->maps, &altdev->maps);
176 	} else {
177 		list_for_each_entry_safe(offload, ptmp, &ondev->progs, offloads)
178 			__bpf_prog_offload_destroy(offload->prog);
179 		list_for_each_entry_safe(offmap, mtmp, &ondev->maps, offloads)
180 			__bpf_map_offload_destroy(offmap);
181 	}
182 
183 	WARN_ON(!list_empty(&ondev->progs));
184 	WARN_ON(!list_empty(&ondev->maps));
185 	kfree(ondev);
186 }
187 
188 static int __bpf_prog_dev_bound_init(struct bpf_prog *prog, struct net_device *netdev)
189 {
190 	struct bpf_offload_netdev *ondev;
191 	struct bpf_prog_offload *offload;
192 	int err;
193 
194 	offload = kzalloc(sizeof(*offload), GFP_USER);
195 	if (!offload)
196 		return -ENOMEM;
197 
198 	offload->prog = prog;
199 	offload->netdev = netdev;
200 
201 	ondev = bpf_offload_find_netdev(offload->netdev);
202 	if (!ondev) {
203 		if (bpf_prog_is_offloaded(prog->aux)) {
204 			err = -EINVAL;
205 			goto err_free;
206 		}
207 
208 		/* When only binding to the device, explicitly
209 		 * create an entry in the hashtable.
210 		 */
211 		err = __bpf_offload_dev_netdev_register(NULL, offload->netdev);
212 		if (err)
213 			goto err_free;
214 		ondev = bpf_offload_find_netdev(offload->netdev);
215 	}
216 	offload->offdev = ondev->offdev;
217 	prog->aux->offload = offload;
218 	list_add_tail(&offload->offloads, &ondev->progs);
219 
220 	return 0;
221 err_free:
222 	kfree(offload);
223 	return err;
224 }
225 
226 int bpf_prog_dev_bound_init(struct bpf_prog *prog, union bpf_attr *attr)
227 {
228 	struct net_device *netdev;
229 	int err;
230 
231 	if (attr->prog_type != BPF_PROG_TYPE_SCHED_CLS &&
232 	    attr->prog_type != BPF_PROG_TYPE_XDP)
233 		return -EINVAL;
234 
235 	if (attr->prog_flags & ~BPF_F_XDP_DEV_BOUND_ONLY)
236 		return -EINVAL;
237 
238 	if (attr->prog_type == BPF_PROG_TYPE_SCHED_CLS &&
239 	    attr->prog_flags & BPF_F_XDP_DEV_BOUND_ONLY)
240 		return -EINVAL;
241 
242 	netdev = dev_get_by_index(current->nsproxy->net_ns, attr->prog_ifindex);
243 	if (!netdev)
244 		return -EINVAL;
245 
246 	err = bpf_dev_offload_check(netdev);
247 	if (err)
248 		goto out;
249 
250 	prog->aux->offload_requested = !(attr->prog_flags & BPF_F_XDP_DEV_BOUND_ONLY);
251 
252 	down_write(&bpf_devs_lock);
253 	err = __bpf_prog_dev_bound_init(prog, netdev);
254 	up_write(&bpf_devs_lock);
255 
256 out:
257 	dev_put(netdev);
258 	return err;
259 }
260 
261 int bpf_prog_dev_bound_inherit(struct bpf_prog *new_prog, struct bpf_prog *old_prog)
262 {
263 	int err;
264 
265 	if (!bpf_prog_is_dev_bound(old_prog->aux))
266 		return 0;
267 
268 	if (bpf_prog_is_offloaded(old_prog->aux))
269 		return -EINVAL;
270 
271 	new_prog->aux->dev_bound = old_prog->aux->dev_bound;
272 	new_prog->aux->offload_requested = old_prog->aux->offload_requested;
273 
274 	down_write(&bpf_devs_lock);
275 	if (!old_prog->aux->offload) {
276 		err = -EINVAL;
277 		goto out;
278 	}
279 
280 	err = __bpf_prog_dev_bound_init(new_prog, old_prog->aux->offload->netdev);
281 
282 out:
283 	up_write(&bpf_devs_lock);
284 	return err;
285 }
286 
287 int bpf_prog_offload_verifier_prep(struct bpf_prog *prog)
288 {
289 	struct bpf_prog_offload *offload;
290 	int ret = -ENODEV;
291 
292 	down_read(&bpf_devs_lock);
293 	offload = prog->aux->offload;
294 	if (offload) {
295 		ret = offload->offdev->ops->prepare(prog);
296 		offload->dev_state = !ret;
297 	}
298 	up_read(&bpf_devs_lock);
299 
300 	return ret;
301 }
302 
303 int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env,
304 				 int insn_idx, int prev_insn_idx)
305 {
306 	struct bpf_prog_offload *offload;
307 	int ret = -ENODEV;
308 
309 	down_read(&bpf_devs_lock);
310 	offload = env->prog->aux->offload;
311 	if (offload)
312 		ret = offload->offdev->ops->insn_hook(env, insn_idx,
313 						      prev_insn_idx);
314 	up_read(&bpf_devs_lock);
315 
316 	return ret;
317 }
318 
319 int bpf_prog_offload_finalize(struct bpf_verifier_env *env)
320 {
321 	struct bpf_prog_offload *offload;
322 	int ret = -ENODEV;
323 
324 	down_read(&bpf_devs_lock);
325 	offload = env->prog->aux->offload;
326 	if (offload) {
327 		if (offload->offdev->ops->finalize)
328 			ret = offload->offdev->ops->finalize(env);
329 		else
330 			ret = 0;
331 	}
332 	up_read(&bpf_devs_lock);
333 
334 	return ret;
335 }
336 
337 void
338 bpf_prog_offload_replace_insn(struct bpf_verifier_env *env, u32 off,
339 			      struct bpf_insn *insn)
340 {
341 	const struct bpf_prog_offload_ops *ops;
342 	struct bpf_prog_offload *offload;
343 	int ret = -EOPNOTSUPP;
344 
345 	down_read(&bpf_devs_lock);
346 	offload = env->prog->aux->offload;
347 	if (offload) {
348 		ops = offload->offdev->ops;
349 		if (!offload->opt_failed && ops->replace_insn)
350 			ret = ops->replace_insn(env, off, insn);
351 		offload->opt_failed |= ret;
352 	}
353 	up_read(&bpf_devs_lock);
354 }
355 
356 void
357 bpf_prog_offload_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt)
358 {
359 	struct bpf_prog_offload *offload;
360 	int ret = -EOPNOTSUPP;
361 
362 	down_read(&bpf_devs_lock);
363 	offload = env->prog->aux->offload;
364 	if (offload) {
365 		if (!offload->opt_failed && offload->offdev->ops->remove_insns)
366 			ret = offload->offdev->ops->remove_insns(env, off, cnt);
367 		offload->opt_failed |= ret;
368 	}
369 	up_read(&bpf_devs_lock);
370 }
371 
372 void bpf_prog_dev_bound_destroy(struct bpf_prog *prog)
373 {
374 	struct bpf_offload_netdev *ondev;
375 	struct net_device *netdev;
376 
377 	rtnl_lock();
378 	down_write(&bpf_devs_lock);
379 	if (prog->aux->offload) {
380 		list_del_init(&prog->aux->offload->offloads);
381 
382 		netdev = prog->aux->offload->netdev;
383 		__bpf_prog_offload_destroy(prog);
384 
385 		ondev = bpf_offload_find_netdev(netdev);
386 		if (!ondev->offdev && list_empty(&ondev->progs))
387 			__bpf_offload_dev_netdev_unregister(NULL, netdev);
388 	}
389 	up_write(&bpf_devs_lock);
390 	rtnl_unlock();
391 }
392 
393 static int bpf_prog_offload_translate(struct bpf_prog *prog)
394 {
395 	struct bpf_prog_offload *offload;
396 	int ret = -ENODEV;
397 
398 	down_read(&bpf_devs_lock);
399 	offload = prog->aux->offload;
400 	if (offload)
401 		ret = offload->offdev->ops->translate(prog);
402 	up_read(&bpf_devs_lock);
403 
404 	return ret;
405 }
406 
407 static unsigned int bpf_prog_warn_on_exec(const void *ctx,
408 					  const struct bpf_insn *insn)
409 {
410 	WARN(1, "attempt to execute device eBPF program on the host!");
411 	return 0;
412 }
413 
414 int bpf_prog_offload_compile(struct bpf_prog *prog)
415 {
416 	prog->bpf_func = bpf_prog_warn_on_exec;
417 
418 	return bpf_prog_offload_translate(prog);
419 }
420 
421 struct ns_get_path_bpf_prog_args {
422 	struct bpf_prog *prog;
423 	struct bpf_prog_info *info;
424 };
425 
426 static struct ns_common *bpf_prog_offload_info_fill_ns(void *private_data)
427 {
428 	struct ns_get_path_bpf_prog_args *args = private_data;
429 	struct bpf_prog_aux *aux = args->prog->aux;
430 	struct ns_common *ns;
431 	struct net *net;
432 
433 	rtnl_lock();
434 	down_read(&bpf_devs_lock);
435 
436 	if (aux->offload) {
437 		args->info->ifindex = aux->offload->netdev->ifindex;
438 		net = dev_net(aux->offload->netdev);
439 		get_net(net);
440 		ns = &net->ns;
441 	} else {
442 		args->info->ifindex = 0;
443 		ns = NULL;
444 	}
445 
446 	up_read(&bpf_devs_lock);
447 	rtnl_unlock();
448 
449 	return ns;
450 }
451 
452 int bpf_prog_offload_info_fill(struct bpf_prog_info *info,
453 			       struct bpf_prog *prog)
454 {
455 	struct ns_get_path_bpf_prog_args args = {
456 		.prog	= prog,
457 		.info	= info,
458 	};
459 	struct bpf_prog_aux *aux = prog->aux;
460 	struct inode *ns_inode;
461 	struct path ns_path;
462 	char __user *uinsns;
463 	int res;
464 	u32 ulen;
465 
466 	res = ns_get_path_cb(&ns_path, bpf_prog_offload_info_fill_ns, &args);
467 	if (res) {
468 		if (!info->ifindex)
469 			return -ENODEV;
470 		return res;
471 	}
472 
473 	down_read(&bpf_devs_lock);
474 
475 	if (!aux->offload) {
476 		up_read(&bpf_devs_lock);
477 		return -ENODEV;
478 	}
479 
480 	ulen = info->jited_prog_len;
481 	info->jited_prog_len = aux->offload->jited_len;
482 	if (info->jited_prog_len && ulen) {
483 		uinsns = u64_to_user_ptr(info->jited_prog_insns);
484 		ulen = min_t(u32, info->jited_prog_len, ulen);
485 		if (copy_to_user(uinsns, aux->offload->jited_image, ulen)) {
486 			up_read(&bpf_devs_lock);
487 			return -EFAULT;
488 		}
489 	}
490 
491 	up_read(&bpf_devs_lock);
492 
493 	ns_inode = ns_path.dentry->d_inode;
494 	info->netns_dev = new_encode_dev(ns_inode->i_sb->s_dev);
495 	info->netns_ino = ns_inode->i_ino;
496 	path_put(&ns_path);
497 
498 	return 0;
499 }
500 
501 const struct bpf_prog_ops bpf_offload_prog_ops = {
502 };
503 
504 struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr)
505 {
506 	struct net *net = current->nsproxy->net_ns;
507 	struct bpf_offload_netdev *ondev;
508 	struct bpf_offloaded_map *offmap;
509 	int err;
510 
511 	if (!capable(CAP_SYS_ADMIN))
512 		return ERR_PTR(-EPERM);
513 	if (attr->map_type != BPF_MAP_TYPE_ARRAY &&
514 	    attr->map_type != BPF_MAP_TYPE_HASH)
515 		return ERR_PTR(-EINVAL);
516 
517 	offmap = bpf_map_area_alloc(sizeof(*offmap), NUMA_NO_NODE);
518 	if (!offmap)
519 		return ERR_PTR(-ENOMEM);
520 
521 	bpf_map_init_from_attr(&offmap->map, attr);
522 
523 	rtnl_lock();
524 	down_write(&bpf_devs_lock);
525 	offmap->netdev = __dev_get_by_index(net, attr->map_ifindex);
526 	err = bpf_dev_offload_check(offmap->netdev);
527 	if (err)
528 		goto err_unlock;
529 
530 	ondev = bpf_offload_find_netdev(offmap->netdev);
531 	if (!ondev) {
532 		err = -EINVAL;
533 		goto err_unlock;
534 	}
535 
536 	err = bpf_map_offload_ndo(offmap, BPF_OFFLOAD_MAP_ALLOC);
537 	if (err)
538 		goto err_unlock;
539 
540 	list_add_tail(&offmap->offloads, &ondev->maps);
541 	up_write(&bpf_devs_lock);
542 	rtnl_unlock();
543 
544 	return &offmap->map;
545 
546 err_unlock:
547 	up_write(&bpf_devs_lock);
548 	rtnl_unlock();
549 	bpf_map_area_free(offmap);
550 	return ERR_PTR(err);
551 }
552 
553 void bpf_map_offload_map_free(struct bpf_map *map)
554 {
555 	struct bpf_offloaded_map *offmap = map_to_offmap(map);
556 
557 	rtnl_lock();
558 	down_write(&bpf_devs_lock);
559 	if (offmap->netdev)
560 		__bpf_map_offload_destroy(offmap);
561 	up_write(&bpf_devs_lock);
562 	rtnl_unlock();
563 
564 	bpf_map_area_free(offmap);
565 }
566 
567 u64 bpf_map_offload_map_mem_usage(const struct bpf_map *map)
568 {
569 	/* The memory dynamically allocated in netdev dev_ops is not counted */
570 	return sizeof(struct bpf_offloaded_map);
571 }
572 
573 int bpf_map_offload_lookup_elem(struct bpf_map *map, void *key, void *value)
574 {
575 	struct bpf_offloaded_map *offmap = map_to_offmap(map);
576 	int ret = -ENODEV;
577 
578 	down_read(&bpf_devs_lock);
579 	if (offmap->netdev)
580 		ret = offmap->dev_ops->map_lookup_elem(offmap, key, value);
581 	up_read(&bpf_devs_lock);
582 
583 	return ret;
584 }
585 
586 int bpf_map_offload_update_elem(struct bpf_map *map,
587 				void *key, void *value, u64 flags)
588 {
589 	struct bpf_offloaded_map *offmap = map_to_offmap(map);
590 	int ret = -ENODEV;
591 
592 	if (unlikely(flags > BPF_EXIST))
593 		return -EINVAL;
594 
595 	down_read(&bpf_devs_lock);
596 	if (offmap->netdev)
597 		ret = offmap->dev_ops->map_update_elem(offmap, key, value,
598 						       flags);
599 	up_read(&bpf_devs_lock);
600 
601 	return ret;
602 }
603 
604 int bpf_map_offload_delete_elem(struct bpf_map *map, void *key)
605 {
606 	struct bpf_offloaded_map *offmap = map_to_offmap(map);
607 	int ret = -ENODEV;
608 
609 	down_read(&bpf_devs_lock);
610 	if (offmap->netdev)
611 		ret = offmap->dev_ops->map_delete_elem(offmap, key);
612 	up_read(&bpf_devs_lock);
613 
614 	return ret;
615 }
616 
617 int bpf_map_offload_get_next_key(struct bpf_map *map, void *key, void *next_key)
618 {
619 	struct bpf_offloaded_map *offmap = map_to_offmap(map);
620 	int ret = -ENODEV;
621 
622 	down_read(&bpf_devs_lock);
623 	if (offmap->netdev)
624 		ret = offmap->dev_ops->map_get_next_key(offmap, key, next_key);
625 	up_read(&bpf_devs_lock);
626 
627 	return ret;
628 }
629 
630 struct ns_get_path_bpf_map_args {
631 	struct bpf_offloaded_map *offmap;
632 	struct bpf_map_info *info;
633 };
634 
635 static struct ns_common *bpf_map_offload_info_fill_ns(void *private_data)
636 {
637 	struct ns_get_path_bpf_map_args *args = private_data;
638 	struct ns_common *ns;
639 	struct net *net;
640 
641 	rtnl_lock();
642 	down_read(&bpf_devs_lock);
643 
644 	if (args->offmap->netdev) {
645 		args->info->ifindex = args->offmap->netdev->ifindex;
646 		net = dev_net(args->offmap->netdev);
647 		get_net(net);
648 		ns = &net->ns;
649 	} else {
650 		args->info->ifindex = 0;
651 		ns = NULL;
652 	}
653 
654 	up_read(&bpf_devs_lock);
655 	rtnl_unlock();
656 
657 	return ns;
658 }
659 
660 int bpf_map_offload_info_fill(struct bpf_map_info *info, struct bpf_map *map)
661 {
662 	struct ns_get_path_bpf_map_args args = {
663 		.offmap	= map_to_offmap(map),
664 		.info	= info,
665 	};
666 	struct inode *ns_inode;
667 	struct path ns_path;
668 	int res;
669 
670 	res = ns_get_path_cb(&ns_path, bpf_map_offload_info_fill_ns, &args);
671 	if (res) {
672 		if (!info->ifindex)
673 			return -ENODEV;
674 		return res;
675 	}
676 
677 	ns_inode = ns_path.dentry->d_inode;
678 	info->netns_dev = new_encode_dev(ns_inode->i_sb->s_dev);
679 	info->netns_ino = ns_inode->i_ino;
680 	path_put(&ns_path);
681 
682 	return 0;
683 }
684 
685 static bool __bpf_offload_dev_match(struct bpf_prog *prog,
686 				    struct net_device *netdev)
687 {
688 	struct bpf_offload_netdev *ondev1, *ondev2;
689 	struct bpf_prog_offload *offload;
690 
691 	if (!bpf_prog_is_dev_bound(prog->aux))
692 		return false;
693 
694 	offload = prog->aux->offload;
695 	if (!offload)
696 		return false;
697 	if (offload->netdev == netdev)
698 		return true;
699 
700 	ondev1 = bpf_offload_find_netdev(offload->netdev);
701 	ondev2 = bpf_offload_find_netdev(netdev);
702 
703 	return ondev1 && ondev2 && ondev1->offdev == ondev2->offdev;
704 }
705 
706 bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev)
707 {
708 	bool ret;
709 
710 	down_read(&bpf_devs_lock);
711 	ret = __bpf_offload_dev_match(prog, netdev);
712 	up_read(&bpf_devs_lock);
713 
714 	return ret;
715 }
716 EXPORT_SYMBOL_GPL(bpf_offload_dev_match);
717 
718 bool bpf_prog_dev_bound_match(const struct bpf_prog *lhs, const struct bpf_prog *rhs)
719 {
720 	bool ret;
721 
722 	if (bpf_prog_is_offloaded(lhs->aux) != bpf_prog_is_offloaded(rhs->aux))
723 		return false;
724 
725 	down_read(&bpf_devs_lock);
726 	ret = lhs->aux->offload && rhs->aux->offload &&
727 	      lhs->aux->offload->netdev &&
728 	      lhs->aux->offload->netdev == rhs->aux->offload->netdev;
729 	up_read(&bpf_devs_lock);
730 
731 	return ret;
732 }
733 
734 bool bpf_offload_prog_map_match(struct bpf_prog *prog, struct bpf_map *map)
735 {
736 	struct bpf_offloaded_map *offmap;
737 	bool ret;
738 
739 	if (!bpf_map_is_offloaded(map))
740 		return bpf_map_offload_neutral(map);
741 	offmap = map_to_offmap(map);
742 
743 	down_read(&bpf_devs_lock);
744 	ret = __bpf_offload_dev_match(prog, offmap->netdev);
745 	up_read(&bpf_devs_lock);
746 
747 	return ret;
748 }
749 
750 int bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev,
751 				    struct net_device *netdev)
752 {
753 	int err;
754 
755 	down_write(&bpf_devs_lock);
756 	err = __bpf_offload_dev_netdev_register(offdev, netdev);
757 	up_write(&bpf_devs_lock);
758 	return err;
759 }
760 EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_register);
761 
762 void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev,
763 				       struct net_device *netdev)
764 {
765 	down_write(&bpf_devs_lock);
766 	__bpf_offload_dev_netdev_unregister(offdev, netdev);
767 	up_write(&bpf_devs_lock);
768 }
769 EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_unregister);
770 
771 struct bpf_offload_dev *
772 bpf_offload_dev_create(const struct bpf_prog_offload_ops *ops, void *priv)
773 {
774 	struct bpf_offload_dev *offdev;
775 
776 	offdev = kzalloc(sizeof(*offdev), GFP_KERNEL);
777 	if (!offdev)
778 		return ERR_PTR(-ENOMEM);
779 
780 	offdev->ops = ops;
781 	offdev->priv = priv;
782 	INIT_LIST_HEAD(&offdev->netdevs);
783 
784 	return offdev;
785 }
786 EXPORT_SYMBOL_GPL(bpf_offload_dev_create);
787 
788 void bpf_offload_dev_destroy(struct bpf_offload_dev *offdev)
789 {
790 	WARN_ON(!list_empty(&offdev->netdevs));
791 	kfree(offdev);
792 }
793 EXPORT_SYMBOL_GPL(bpf_offload_dev_destroy);
794 
795 void *bpf_offload_dev_priv(struct bpf_offload_dev *offdev)
796 {
797 	return offdev->priv;
798 }
799 EXPORT_SYMBOL_GPL(bpf_offload_dev_priv);
800 
801 void bpf_dev_bound_netdev_unregister(struct net_device *dev)
802 {
803 	struct bpf_offload_netdev *ondev;
804 
805 	ASSERT_RTNL();
806 
807 	down_write(&bpf_devs_lock);
808 	ondev = bpf_offload_find_netdev(dev);
809 	if (ondev && !ondev->offdev)
810 		__bpf_offload_dev_netdev_unregister(NULL, ondev->netdev);
811 	up_write(&bpf_devs_lock);
812 }
813 
814 int bpf_dev_bound_kfunc_check(struct bpf_verifier_log *log,
815 			      struct bpf_prog_aux *prog_aux)
816 {
817 	if (!bpf_prog_is_dev_bound(prog_aux)) {
818 		bpf_log(log, "metadata kfuncs require device-bound program\n");
819 		return -EINVAL;
820 	}
821 
822 	if (bpf_prog_is_offloaded(prog_aux)) {
823 		bpf_log(log, "metadata kfuncs can't be offloaded\n");
824 		return -EINVAL;
825 	}
826 
827 	return 0;
828 }
829 
830 void *bpf_dev_bound_resolve_kfunc(struct bpf_prog *prog, u32 func_id)
831 {
832 	const struct xdp_metadata_ops *ops;
833 	void *p = NULL;
834 
835 	/* We don't hold bpf_devs_lock while resolving several
836 	 * kfuncs and can race with the unregister_netdevice().
837 	 * We rely on bpf_dev_bound_match() check at attach
838 	 * to render this program unusable.
839 	 */
840 	down_read(&bpf_devs_lock);
841 	if (!prog->aux->offload)
842 		goto out;
843 
844 	ops = prog->aux->offload->netdev->xdp_metadata_ops;
845 	if (!ops)
846 		goto out;
847 
848 	if (func_id == bpf_xdp_metadata_kfunc_id(XDP_METADATA_KFUNC_RX_TIMESTAMP))
849 		p = ops->xmo_rx_timestamp;
850 	else if (func_id == bpf_xdp_metadata_kfunc_id(XDP_METADATA_KFUNC_RX_HASH))
851 		p = ops->xmo_rx_hash;
852 out:
853 	up_read(&bpf_devs_lock);
854 
855 	return p;
856 }
857 
858 static int __init bpf_offload_init(void)
859 {
860 	return rhashtable_init(&offdevs, &offdevs_params);
861 }
862 
863 core_initcall(bpf_offload_init);
864