xref: /openbmc/linux/kernel/bpf/offload.c (revision ebd09753)
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 
29 /* Protects offdevs, members of bpf_offload_netdev and offload members
30  * of all progs.
31  * RTNL lock cannot be taken when holding this lock.
32  */
33 static DECLARE_RWSEM(bpf_devs_lock);
34 
35 struct bpf_offload_dev {
36 	struct list_head netdevs;
37 };
38 
39 struct bpf_offload_netdev {
40 	struct rhash_head l;
41 	struct net_device *netdev;
42 	struct bpf_offload_dev *offdev;
43 	struct list_head progs;
44 	struct list_head maps;
45 	struct list_head offdev_netdevs;
46 };
47 
48 static const struct rhashtable_params offdevs_params = {
49 	.nelem_hint		= 4,
50 	.key_len		= sizeof(struct net_device *),
51 	.key_offset		= offsetof(struct bpf_offload_netdev, netdev),
52 	.head_offset		= offsetof(struct bpf_offload_netdev, l),
53 	.automatic_shrinking	= true,
54 };
55 
56 static struct rhashtable offdevs;
57 static bool offdevs_inited;
58 
59 static int bpf_dev_offload_check(struct net_device *netdev)
60 {
61 	if (!netdev)
62 		return -EINVAL;
63 	if (!netdev->netdev_ops->ndo_bpf)
64 		return -EOPNOTSUPP;
65 	return 0;
66 }
67 
68 static struct bpf_offload_netdev *
69 bpf_offload_find_netdev(struct net_device *netdev)
70 {
71 	lockdep_assert_held(&bpf_devs_lock);
72 
73 	if (!offdevs_inited)
74 		return NULL;
75 	return rhashtable_lookup_fast(&offdevs, &netdev, offdevs_params);
76 }
77 
78 int bpf_prog_offload_init(struct bpf_prog *prog, union bpf_attr *attr)
79 {
80 	struct bpf_offload_netdev *ondev;
81 	struct bpf_prog_offload *offload;
82 	int err;
83 
84 	if (attr->prog_type != BPF_PROG_TYPE_SCHED_CLS &&
85 	    attr->prog_type != BPF_PROG_TYPE_XDP)
86 		return -EINVAL;
87 
88 	if (attr->prog_flags)
89 		return -EINVAL;
90 
91 	offload = kzalloc(sizeof(*offload), GFP_USER);
92 	if (!offload)
93 		return -ENOMEM;
94 
95 	offload->prog = prog;
96 
97 	offload->netdev = dev_get_by_index(current->nsproxy->net_ns,
98 					   attr->prog_ifindex);
99 	err = bpf_dev_offload_check(offload->netdev);
100 	if (err)
101 		goto err_maybe_put;
102 
103 	down_write(&bpf_devs_lock);
104 	ondev = bpf_offload_find_netdev(offload->netdev);
105 	if (!ondev) {
106 		err = -EINVAL;
107 		goto err_unlock;
108 	}
109 	prog->aux->offload = offload;
110 	list_add_tail(&offload->offloads, &ondev->progs);
111 	dev_put(offload->netdev);
112 	up_write(&bpf_devs_lock);
113 
114 	return 0;
115 err_unlock:
116 	up_write(&bpf_devs_lock);
117 err_maybe_put:
118 	if (offload->netdev)
119 		dev_put(offload->netdev);
120 	kfree(offload);
121 	return err;
122 }
123 
124 static int __bpf_offload_ndo(struct bpf_prog *prog, enum bpf_netdev_command cmd,
125 			     struct netdev_bpf *data)
126 {
127 	struct bpf_prog_offload *offload = prog->aux->offload;
128 	struct net_device *netdev;
129 
130 	ASSERT_RTNL();
131 
132 	if (!offload)
133 		return -ENODEV;
134 	netdev = offload->netdev;
135 
136 	data->command = cmd;
137 
138 	return netdev->netdev_ops->ndo_bpf(netdev, data);
139 }
140 
141 int bpf_prog_offload_verifier_prep(struct bpf_verifier_env *env)
142 {
143 	struct netdev_bpf data = {};
144 	int err;
145 
146 	data.verifier.prog = env->prog;
147 
148 	rtnl_lock();
149 	err = __bpf_offload_ndo(env->prog, BPF_OFFLOAD_VERIFIER_PREP, &data);
150 	if (err)
151 		goto exit_unlock;
152 
153 	env->prog->aux->offload->dev_ops = data.verifier.ops;
154 	env->prog->aux->offload->dev_state = true;
155 exit_unlock:
156 	rtnl_unlock();
157 	return err;
158 }
159 
160 int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env,
161 				 int insn_idx, int prev_insn_idx)
162 {
163 	struct bpf_prog_offload *offload;
164 	int ret = -ENODEV;
165 
166 	down_read(&bpf_devs_lock);
167 	offload = env->prog->aux->offload;
168 	if (offload)
169 		ret = offload->dev_ops->insn_hook(env, insn_idx, prev_insn_idx);
170 	up_read(&bpf_devs_lock);
171 
172 	return ret;
173 }
174 
175 int bpf_prog_offload_finalize(struct bpf_verifier_env *env)
176 {
177 	struct bpf_prog_offload *offload;
178 	int ret = -ENODEV;
179 
180 	down_read(&bpf_devs_lock);
181 	offload = env->prog->aux->offload;
182 	if (offload) {
183 		if (offload->dev_ops->finalize)
184 			ret = offload->dev_ops->finalize(env);
185 		else
186 			ret = 0;
187 	}
188 	up_read(&bpf_devs_lock);
189 
190 	return ret;
191 }
192 
193 static void __bpf_prog_offload_destroy(struct bpf_prog *prog)
194 {
195 	struct bpf_prog_offload *offload = prog->aux->offload;
196 	struct netdev_bpf data = {};
197 
198 	data.offload.prog = prog;
199 
200 	if (offload->dev_state)
201 		WARN_ON(__bpf_offload_ndo(prog, BPF_OFFLOAD_DESTROY, &data));
202 
203 	/* Make sure BPF_PROG_GET_NEXT_ID can't find this dead program */
204 	bpf_prog_free_id(prog, true);
205 
206 	list_del_init(&offload->offloads);
207 	kfree(offload);
208 	prog->aux->offload = NULL;
209 }
210 
211 void bpf_prog_offload_destroy(struct bpf_prog *prog)
212 {
213 	rtnl_lock();
214 	down_write(&bpf_devs_lock);
215 	if (prog->aux->offload)
216 		__bpf_prog_offload_destroy(prog);
217 	up_write(&bpf_devs_lock);
218 	rtnl_unlock();
219 }
220 
221 static int bpf_prog_offload_translate(struct bpf_prog *prog)
222 {
223 	struct netdev_bpf data = {};
224 	int ret;
225 
226 	data.offload.prog = prog;
227 
228 	rtnl_lock();
229 	ret = __bpf_offload_ndo(prog, BPF_OFFLOAD_TRANSLATE, &data);
230 	rtnl_unlock();
231 
232 	return ret;
233 }
234 
235 static unsigned int bpf_prog_warn_on_exec(const void *ctx,
236 					  const struct bpf_insn *insn)
237 {
238 	WARN(1, "attempt to execute device eBPF program on the host!");
239 	return 0;
240 }
241 
242 int bpf_prog_offload_compile(struct bpf_prog *prog)
243 {
244 	prog->bpf_func = bpf_prog_warn_on_exec;
245 
246 	return bpf_prog_offload_translate(prog);
247 }
248 
249 struct ns_get_path_bpf_prog_args {
250 	struct bpf_prog *prog;
251 	struct bpf_prog_info *info;
252 };
253 
254 static struct ns_common *bpf_prog_offload_info_fill_ns(void *private_data)
255 {
256 	struct ns_get_path_bpf_prog_args *args = private_data;
257 	struct bpf_prog_aux *aux = args->prog->aux;
258 	struct ns_common *ns;
259 	struct net *net;
260 
261 	rtnl_lock();
262 	down_read(&bpf_devs_lock);
263 
264 	if (aux->offload) {
265 		args->info->ifindex = aux->offload->netdev->ifindex;
266 		net = dev_net(aux->offload->netdev);
267 		get_net(net);
268 		ns = &net->ns;
269 	} else {
270 		args->info->ifindex = 0;
271 		ns = NULL;
272 	}
273 
274 	up_read(&bpf_devs_lock);
275 	rtnl_unlock();
276 
277 	return ns;
278 }
279 
280 int bpf_prog_offload_info_fill(struct bpf_prog_info *info,
281 			       struct bpf_prog *prog)
282 {
283 	struct ns_get_path_bpf_prog_args args = {
284 		.prog	= prog,
285 		.info	= info,
286 	};
287 	struct bpf_prog_aux *aux = prog->aux;
288 	struct inode *ns_inode;
289 	struct path ns_path;
290 	char __user *uinsns;
291 	void *res;
292 	u32 ulen;
293 
294 	res = ns_get_path_cb(&ns_path, bpf_prog_offload_info_fill_ns, &args);
295 	if (IS_ERR(res)) {
296 		if (!info->ifindex)
297 			return -ENODEV;
298 		return PTR_ERR(res);
299 	}
300 
301 	down_read(&bpf_devs_lock);
302 
303 	if (!aux->offload) {
304 		up_read(&bpf_devs_lock);
305 		return -ENODEV;
306 	}
307 
308 	ulen = info->jited_prog_len;
309 	info->jited_prog_len = aux->offload->jited_len;
310 	if (info->jited_prog_len & ulen) {
311 		uinsns = u64_to_user_ptr(info->jited_prog_insns);
312 		ulen = min_t(u32, info->jited_prog_len, ulen);
313 		if (copy_to_user(uinsns, aux->offload->jited_image, ulen)) {
314 			up_read(&bpf_devs_lock);
315 			return -EFAULT;
316 		}
317 	}
318 
319 	up_read(&bpf_devs_lock);
320 
321 	ns_inode = ns_path.dentry->d_inode;
322 	info->netns_dev = new_encode_dev(ns_inode->i_sb->s_dev);
323 	info->netns_ino = ns_inode->i_ino;
324 	path_put(&ns_path);
325 
326 	return 0;
327 }
328 
329 const struct bpf_prog_ops bpf_offload_prog_ops = {
330 };
331 
332 static int bpf_map_offload_ndo(struct bpf_offloaded_map *offmap,
333 			       enum bpf_netdev_command cmd)
334 {
335 	struct netdev_bpf data = {};
336 	struct net_device *netdev;
337 
338 	ASSERT_RTNL();
339 
340 	data.command = cmd;
341 	data.offmap = offmap;
342 	/* Caller must make sure netdev is valid */
343 	netdev = offmap->netdev;
344 
345 	return netdev->netdev_ops->ndo_bpf(netdev, &data);
346 }
347 
348 struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr)
349 {
350 	struct net *net = current->nsproxy->net_ns;
351 	struct bpf_offload_netdev *ondev;
352 	struct bpf_offloaded_map *offmap;
353 	int err;
354 
355 	if (!capable(CAP_SYS_ADMIN))
356 		return ERR_PTR(-EPERM);
357 	if (attr->map_type != BPF_MAP_TYPE_ARRAY &&
358 	    attr->map_type != BPF_MAP_TYPE_HASH)
359 		return ERR_PTR(-EINVAL);
360 
361 	offmap = kzalloc(sizeof(*offmap), GFP_USER);
362 	if (!offmap)
363 		return ERR_PTR(-ENOMEM);
364 
365 	bpf_map_init_from_attr(&offmap->map, attr);
366 
367 	rtnl_lock();
368 	down_write(&bpf_devs_lock);
369 	offmap->netdev = __dev_get_by_index(net, attr->map_ifindex);
370 	err = bpf_dev_offload_check(offmap->netdev);
371 	if (err)
372 		goto err_unlock;
373 
374 	ondev = bpf_offload_find_netdev(offmap->netdev);
375 	if (!ondev) {
376 		err = -EINVAL;
377 		goto err_unlock;
378 	}
379 
380 	err = bpf_map_offload_ndo(offmap, BPF_OFFLOAD_MAP_ALLOC);
381 	if (err)
382 		goto err_unlock;
383 
384 	list_add_tail(&offmap->offloads, &ondev->maps);
385 	up_write(&bpf_devs_lock);
386 	rtnl_unlock();
387 
388 	return &offmap->map;
389 
390 err_unlock:
391 	up_write(&bpf_devs_lock);
392 	rtnl_unlock();
393 	kfree(offmap);
394 	return ERR_PTR(err);
395 }
396 
397 static void __bpf_map_offload_destroy(struct bpf_offloaded_map *offmap)
398 {
399 	WARN_ON(bpf_map_offload_ndo(offmap, BPF_OFFLOAD_MAP_FREE));
400 	/* Make sure BPF_MAP_GET_NEXT_ID can't find this dead map */
401 	bpf_map_free_id(&offmap->map, true);
402 	list_del_init(&offmap->offloads);
403 	offmap->netdev = NULL;
404 }
405 
406 void bpf_map_offload_map_free(struct bpf_map *map)
407 {
408 	struct bpf_offloaded_map *offmap = map_to_offmap(map);
409 
410 	rtnl_lock();
411 	down_write(&bpf_devs_lock);
412 	if (offmap->netdev)
413 		__bpf_map_offload_destroy(offmap);
414 	up_write(&bpf_devs_lock);
415 	rtnl_unlock();
416 
417 	kfree(offmap);
418 }
419 
420 int bpf_map_offload_lookup_elem(struct bpf_map *map, void *key, void *value)
421 {
422 	struct bpf_offloaded_map *offmap = map_to_offmap(map);
423 	int ret = -ENODEV;
424 
425 	down_read(&bpf_devs_lock);
426 	if (offmap->netdev)
427 		ret = offmap->dev_ops->map_lookup_elem(offmap, key, value);
428 	up_read(&bpf_devs_lock);
429 
430 	return ret;
431 }
432 
433 int bpf_map_offload_update_elem(struct bpf_map *map,
434 				void *key, void *value, u64 flags)
435 {
436 	struct bpf_offloaded_map *offmap = map_to_offmap(map);
437 	int ret = -ENODEV;
438 
439 	if (unlikely(flags > BPF_EXIST))
440 		return -EINVAL;
441 
442 	down_read(&bpf_devs_lock);
443 	if (offmap->netdev)
444 		ret = offmap->dev_ops->map_update_elem(offmap, key, value,
445 						       flags);
446 	up_read(&bpf_devs_lock);
447 
448 	return ret;
449 }
450 
451 int bpf_map_offload_delete_elem(struct bpf_map *map, void *key)
452 {
453 	struct bpf_offloaded_map *offmap = map_to_offmap(map);
454 	int ret = -ENODEV;
455 
456 	down_read(&bpf_devs_lock);
457 	if (offmap->netdev)
458 		ret = offmap->dev_ops->map_delete_elem(offmap, key);
459 	up_read(&bpf_devs_lock);
460 
461 	return ret;
462 }
463 
464 int bpf_map_offload_get_next_key(struct bpf_map *map, void *key, void *next_key)
465 {
466 	struct bpf_offloaded_map *offmap = map_to_offmap(map);
467 	int ret = -ENODEV;
468 
469 	down_read(&bpf_devs_lock);
470 	if (offmap->netdev)
471 		ret = offmap->dev_ops->map_get_next_key(offmap, key, next_key);
472 	up_read(&bpf_devs_lock);
473 
474 	return ret;
475 }
476 
477 struct ns_get_path_bpf_map_args {
478 	struct bpf_offloaded_map *offmap;
479 	struct bpf_map_info *info;
480 };
481 
482 static struct ns_common *bpf_map_offload_info_fill_ns(void *private_data)
483 {
484 	struct ns_get_path_bpf_map_args *args = private_data;
485 	struct ns_common *ns;
486 	struct net *net;
487 
488 	rtnl_lock();
489 	down_read(&bpf_devs_lock);
490 
491 	if (args->offmap->netdev) {
492 		args->info->ifindex = args->offmap->netdev->ifindex;
493 		net = dev_net(args->offmap->netdev);
494 		get_net(net);
495 		ns = &net->ns;
496 	} else {
497 		args->info->ifindex = 0;
498 		ns = NULL;
499 	}
500 
501 	up_read(&bpf_devs_lock);
502 	rtnl_unlock();
503 
504 	return ns;
505 }
506 
507 int bpf_map_offload_info_fill(struct bpf_map_info *info, struct bpf_map *map)
508 {
509 	struct ns_get_path_bpf_map_args args = {
510 		.offmap	= map_to_offmap(map),
511 		.info	= info,
512 	};
513 	struct inode *ns_inode;
514 	struct path ns_path;
515 	void *res;
516 
517 	res = ns_get_path_cb(&ns_path, bpf_map_offload_info_fill_ns, &args);
518 	if (IS_ERR(res)) {
519 		if (!info->ifindex)
520 			return -ENODEV;
521 		return PTR_ERR(res);
522 	}
523 
524 	ns_inode = ns_path.dentry->d_inode;
525 	info->netns_dev = new_encode_dev(ns_inode->i_sb->s_dev);
526 	info->netns_ino = ns_inode->i_ino;
527 	path_put(&ns_path);
528 
529 	return 0;
530 }
531 
532 static bool __bpf_offload_dev_match(struct bpf_prog *prog,
533 				    struct net_device *netdev)
534 {
535 	struct bpf_offload_netdev *ondev1, *ondev2;
536 	struct bpf_prog_offload *offload;
537 
538 	if (!bpf_prog_is_dev_bound(prog->aux))
539 		return false;
540 
541 	offload = prog->aux->offload;
542 	if (!offload)
543 		return false;
544 	if (offload->netdev == netdev)
545 		return true;
546 
547 	ondev1 = bpf_offload_find_netdev(offload->netdev);
548 	ondev2 = bpf_offload_find_netdev(netdev);
549 
550 	return ondev1 && ondev2 && ondev1->offdev == ondev2->offdev;
551 }
552 
553 bool bpf_offload_dev_match(struct bpf_prog *prog, struct net_device *netdev)
554 {
555 	bool ret;
556 
557 	down_read(&bpf_devs_lock);
558 	ret = __bpf_offload_dev_match(prog, netdev);
559 	up_read(&bpf_devs_lock);
560 
561 	return ret;
562 }
563 EXPORT_SYMBOL_GPL(bpf_offload_dev_match);
564 
565 bool bpf_offload_prog_map_match(struct bpf_prog *prog, struct bpf_map *map)
566 {
567 	struct bpf_offloaded_map *offmap;
568 	bool ret;
569 
570 	if (!bpf_map_is_dev_bound(map))
571 		return bpf_map_offload_neutral(map);
572 	offmap = map_to_offmap(map);
573 
574 	down_read(&bpf_devs_lock);
575 	ret = __bpf_offload_dev_match(prog, offmap->netdev);
576 	up_read(&bpf_devs_lock);
577 
578 	return ret;
579 }
580 
581 int bpf_offload_dev_netdev_register(struct bpf_offload_dev *offdev,
582 				    struct net_device *netdev)
583 {
584 	struct bpf_offload_netdev *ondev;
585 	int err;
586 
587 	ondev = kzalloc(sizeof(*ondev), GFP_KERNEL);
588 	if (!ondev)
589 		return -ENOMEM;
590 
591 	ondev->netdev = netdev;
592 	ondev->offdev = offdev;
593 	INIT_LIST_HEAD(&ondev->progs);
594 	INIT_LIST_HEAD(&ondev->maps);
595 
596 	down_write(&bpf_devs_lock);
597 	err = rhashtable_insert_fast(&offdevs, &ondev->l, offdevs_params);
598 	if (err) {
599 		netdev_warn(netdev, "failed to register for BPF offload\n");
600 		goto err_unlock_free;
601 	}
602 
603 	list_add(&ondev->offdev_netdevs, &offdev->netdevs);
604 	up_write(&bpf_devs_lock);
605 	return 0;
606 
607 err_unlock_free:
608 	up_write(&bpf_devs_lock);
609 	kfree(ondev);
610 	return err;
611 }
612 EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_register);
613 
614 void bpf_offload_dev_netdev_unregister(struct bpf_offload_dev *offdev,
615 				       struct net_device *netdev)
616 {
617 	struct bpf_offload_netdev *ondev, *altdev;
618 	struct bpf_offloaded_map *offmap, *mtmp;
619 	struct bpf_prog_offload *offload, *ptmp;
620 
621 	ASSERT_RTNL();
622 
623 	down_write(&bpf_devs_lock);
624 	ondev = rhashtable_lookup_fast(&offdevs, &netdev, offdevs_params);
625 	if (WARN_ON(!ondev))
626 		goto unlock;
627 
628 	WARN_ON(rhashtable_remove_fast(&offdevs, &ondev->l, offdevs_params));
629 	list_del(&ondev->offdev_netdevs);
630 
631 	/* Try to move the objects to another netdev of the device */
632 	altdev = list_first_entry_or_null(&offdev->netdevs,
633 					  struct bpf_offload_netdev,
634 					  offdev_netdevs);
635 	if (altdev) {
636 		list_for_each_entry(offload, &ondev->progs, offloads)
637 			offload->netdev = altdev->netdev;
638 		list_splice_init(&ondev->progs, &altdev->progs);
639 
640 		list_for_each_entry(offmap, &ondev->maps, offloads)
641 			offmap->netdev = altdev->netdev;
642 		list_splice_init(&ondev->maps, &altdev->maps);
643 	} else {
644 		list_for_each_entry_safe(offload, ptmp, &ondev->progs, offloads)
645 			__bpf_prog_offload_destroy(offload->prog);
646 		list_for_each_entry_safe(offmap, mtmp, &ondev->maps, offloads)
647 			__bpf_map_offload_destroy(offmap);
648 	}
649 
650 	WARN_ON(!list_empty(&ondev->progs));
651 	WARN_ON(!list_empty(&ondev->maps));
652 	kfree(ondev);
653 unlock:
654 	up_write(&bpf_devs_lock);
655 }
656 EXPORT_SYMBOL_GPL(bpf_offload_dev_netdev_unregister);
657 
658 struct bpf_offload_dev *bpf_offload_dev_create(void)
659 {
660 	struct bpf_offload_dev *offdev;
661 	int err;
662 
663 	down_write(&bpf_devs_lock);
664 	if (!offdevs_inited) {
665 		err = rhashtable_init(&offdevs, &offdevs_params);
666 		if (err)
667 			return ERR_PTR(err);
668 		offdevs_inited = true;
669 	}
670 	up_write(&bpf_devs_lock);
671 
672 	offdev = kzalloc(sizeof(*offdev), GFP_KERNEL);
673 	if (!offdev)
674 		return ERR_PTR(-ENOMEM);
675 
676 	INIT_LIST_HEAD(&offdev->netdevs);
677 
678 	return offdev;
679 }
680 EXPORT_SYMBOL_GPL(bpf_offload_dev_create);
681 
682 void bpf_offload_dev_destroy(struct bpf_offload_dev *offdev)
683 {
684 	WARN_ON(!list_empty(&offdev->netdevs));
685 	kfree(offdev);
686 }
687 EXPORT_SYMBOL_GPL(bpf_offload_dev_destroy);
688