xref: /openbmc/linux/net/bpf/test_run.c (revision a89a501c)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2017 Facebook
3  */
4 #include <linux/bpf.h>
5 #include <linux/slab.h>
6 #include <linux/vmalloc.h>
7 #include <linux/etherdevice.h>
8 #include <linux/filter.h>
9 #include <linux/sched/signal.h>
10 #include <net/bpf_sk_storage.h>
11 #include <net/sock.h>
12 #include <net/tcp.h>
13 #include <linux/error-injection.h>
14 #include <linux/smp.h>
15 
16 #define CREATE_TRACE_POINTS
17 #include <trace/events/bpf_test_run.h>
18 
19 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
20 			u32 *retval, u32 *time, bool xdp)
21 {
22 	struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE] = { NULL };
23 	enum bpf_cgroup_storage_type stype;
24 	u64 time_start, time_spent = 0;
25 	int ret = 0;
26 	u32 i;
27 
28 	for_each_cgroup_storage_type(stype) {
29 		storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
30 		if (IS_ERR(storage[stype])) {
31 			storage[stype] = NULL;
32 			for_each_cgroup_storage_type(stype)
33 				bpf_cgroup_storage_free(storage[stype]);
34 			return -ENOMEM;
35 		}
36 	}
37 
38 	if (!repeat)
39 		repeat = 1;
40 
41 	rcu_read_lock();
42 	migrate_disable();
43 	time_start = ktime_get_ns();
44 	for (i = 0; i < repeat; i++) {
45 		bpf_cgroup_storage_set(storage);
46 
47 		if (xdp)
48 			*retval = bpf_prog_run_xdp(prog, ctx);
49 		else
50 			*retval = BPF_PROG_RUN(prog, ctx);
51 
52 		if (signal_pending(current)) {
53 			ret = -EINTR;
54 			break;
55 		}
56 
57 		if (need_resched()) {
58 			time_spent += ktime_get_ns() - time_start;
59 			migrate_enable();
60 			rcu_read_unlock();
61 
62 			cond_resched();
63 
64 			rcu_read_lock();
65 			migrate_disable();
66 			time_start = ktime_get_ns();
67 		}
68 	}
69 	time_spent += ktime_get_ns() - time_start;
70 	migrate_enable();
71 	rcu_read_unlock();
72 
73 	do_div(time_spent, repeat);
74 	*time = time_spent > U32_MAX ? U32_MAX : (u32)time_spent;
75 
76 	for_each_cgroup_storage_type(stype)
77 		bpf_cgroup_storage_free(storage[stype]);
78 
79 	return ret;
80 }
81 
82 static int bpf_test_finish(const union bpf_attr *kattr,
83 			   union bpf_attr __user *uattr, const void *data,
84 			   u32 size, u32 retval, u32 duration)
85 {
86 	void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
87 	int err = -EFAULT;
88 	u32 copy_size = size;
89 
90 	/* Clamp copy if the user has provided a size hint, but copy the full
91 	 * buffer if not to retain old behaviour.
92 	 */
93 	if (kattr->test.data_size_out &&
94 	    copy_size > kattr->test.data_size_out) {
95 		copy_size = kattr->test.data_size_out;
96 		err = -ENOSPC;
97 	}
98 
99 	if (data_out && copy_to_user(data_out, data, copy_size))
100 		goto out;
101 	if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
102 		goto out;
103 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
104 		goto out;
105 	if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
106 		goto out;
107 	if (err != -ENOSPC)
108 		err = 0;
109 out:
110 	trace_bpf_test_finish(&err);
111 	return err;
112 }
113 
114 /* Integer types of various sizes and pointer combinations cover variety of
115  * architecture dependent calling conventions. 7+ can be supported in the
116  * future.
117  */
118 __diag_push();
119 __diag_ignore(GCC, 8, "-Wmissing-prototypes",
120 	      "Global functions as their definitions will be in vmlinux BTF");
121 int noinline bpf_fentry_test1(int a)
122 {
123 	return a + 1;
124 }
125 
126 int noinline bpf_fentry_test2(int a, u64 b)
127 {
128 	return a + b;
129 }
130 
131 int noinline bpf_fentry_test3(char a, int b, u64 c)
132 {
133 	return a + b + c;
134 }
135 
136 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
137 {
138 	return (long)a + b + c + d;
139 }
140 
141 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
142 {
143 	return a + (long)b + c + d + e;
144 }
145 
146 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
147 {
148 	return a + (long)b + c + d + (long)e + f;
149 }
150 
151 struct bpf_fentry_test_t {
152 	struct bpf_fentry_test_t *a;
153 };
154 
155 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
156 {
157 	return (long)arg;
158 }
159 
160 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
161 {
162 	return (long)arg->a;
163 }
164 
165 int noinline bpf_modify_return_test(int a, int *b)
166 {
167 	*b += 1;
168 	return a + *b;
169 }
170 __diag_pop();
171 
172 ALLOW_ERROR_INJECTION(bpf_modify_return_test, ERRNO);
173 
174 static void *bpf_test_init(const union bpf_attr *kattr, u32 size,
175 			   u32 headroom, u32 tailroom)
176 {
177 	void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
178 	u32 user_size = kattr->test.data_size_in;
179 	void *data;
180 
181 	if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
182 		return ERR_PTR(-EINVAL);
183 
184 	if (user_size > size)
185 		return ERR_PTR(-EMSGSIZE);
186 
187 	data = kzalloc(size + headroom + tailroom, GFP_USER);
188 	if (!data)
189 		return ERR_PTR(-ENOMEM);
190 
191 	if (copy_from_user(data + headroom, data_in, user_size)) {
192 		kfree(data);
193 		return ERR_PTR(-EFAULT);
194 	}
195 
196 	return data;
197 }
198 
199 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
200 			      const union bpf_attr *kattr,
201 			      union bpf_attr __user *uattr)
202 {
203 	struct bpf_fentry_test_t arg = {};
204 	u16 side_effect = 0, ret = 0;
205 	int b = 2, err = -EFAULT;
206 	u32 retval = 0;
207 
208 	if (kattr->test.flags || kattr->test.cpu)
209 		return -EINVAL;
210 
211 	switch (prog->expected_attach_type) {
212 	case BPF_TRACE_FENTRY:
213 	case BPF_TRACE_FEXIT:
214 		if (bpf_fentry_test1(1) != 2 ||
215 		    bpf_fentry_test2(2, 3) != 5 ||
216 		    bpf_fentry_test3(4, 5, 6) != 15 ||
217 		    bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
218 		    bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
219 		    bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
220 		    bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
221 		    bpf_fentry_test8(&arg) != 0)
222 			goto out;
223 		break;
224 	case BPF_MODIFY_RETURN:
225 		ret = bpf_modify_return_test(1, &b);
226 		if (b != 2)
227 			side_effect = 1;
228 		break;
229 	default:
230 		goto out;
231 	}
232 
233 	retval = ((u32)side_effect << 16) | ret;
234 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
235 		goto out;
236 
237 	err = 0;
238 out:
239 	trace_bpf_test_finish(&err);
240 	return err;
241 }
242 
243 struct bpf_raw_tp_test_run_info {
244 	struct bpf_prog *prog;
245 	void *ctx;
246 	u32 retval;
247 };
248 
249 static void
250 __bpf_prog_test_run_raw_tp(void *data)
251 {
252 	struct bpf_raw_tp_test_run_info *info = data;
253 
254 	rcu_read_lock();
255 	info->retval = BPF_PROG_RUN(info->prog, info->ctx);
256 	rcu_read_unlock();
257 }
258 
259 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
260 			     const union bpf_attr *kattr,
261 			     union bpf_attr __user *uattr)
262 {
263 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
264 	__u32 ctx_size_in = kattr->test.ctx_size_in;
265 	struct bpf_raw_tp_test_run_info info;
266 	int cpu = kattr->test.cpu, err = 0;
267 	int current_cpu;
268 
269 	/* doesn't support data_in/out, ctx_out, duration, or repeat */
270 	if (kattr->test.data_in || kattr->test.data_out ||
271 	    kattr->test.ctx_out || kattr->test.duration ||
272 	    kattr->test.repeat)
273 		return -EINVAL;
274 
275 	if (ctx_size_in < prog->aux->max_ctx_offset)
276 		return -EINVAL;
277 
278 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
279 		return -EINVAL;
280 
281 	if (ctx_size_in) {
282 		info.ctx = kzalloc(ctx_size_in, GFP_USER);
283 		if (!info.ctx)
284 			return -ENOMEM;
285 		if (copy_from_user(info.ctx, ctx_in, ctx_size_in)) {
286 			err = -EFAULT;
287 			goto out;
288 		}
289 	} else {
290 		info.ctx = NULL;
291 	}
292 
293 	info.prog = prog;
294 
295 	current_cpu = get_cpu();
296 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
297 	    cpu == current_cpu) {
298 		__bpf_prog_test_run_raw_tp(&info);
299 	} else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
300 		/* smp_call_function_single() also checks cpu_online()
301 		 * after csd_lock(). However, since cpu is from user
302 		 * space, let's do an extra quick check to filter out
303 		 * invalid value before smp_call_function_single().
304 		 */
305 		err = -ENXIO;
306 	} else {
307 		err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
308 					       &info, 1);
309 	}
310 	put_cpu();
311 
312 	if (!err &&
313 	    copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
314 		err = -EFAULT;
315 
316 out:
317 	kfree(info.ctx);
318 	return err;
319 }
320 
321 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
322 {
323 	void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
324 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
325 	u32 size = kattr->test.ctx_size_in;
326 	void *data;
327 	int err;
328 
329 	if (!data_in && !data_out)
330 		return NULL;
331 
332 	data = kzalloc(max_size, GFP_USER);
333 	if (!data)
334 		return ERR_PTR(-ENOMEM);
335 
336 	if (data_in) {
337 		err = bpf_check_uarg_tail_zero(data_in, max_size, size);
338 		if (err) {
339 			kfree(data);
340 			return ERR_PTR(err);
341 		}
342 
343 		size = min_t(u32, max_size, size);
344 		if (copy_from_user(data, data_in, size)) {
345 			kfree(data);
346 			return ERR_PTR(-EFAULT);
347 		}
348 	}
349 	return data;
350 }
351 
352 static int bpf_ctx_finish(const union bpf_attr *kattr,
353 			  union bpf_attr __user *uattr, const void *data,
354 			  u32 size)
355 {
356 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
357 	int err = -EFAULT;
358 	u32 copy_size = size;
359 
360 	if (!data || !data_out)
361 		return 0;
362 
363 	if (copy_size > kattr->test.ctx_size_out) {
364 		copy_size = kattr->test.ctx_size_out;
365 		err = -ENOSPC;
366 	}
367 
368 	if (copy_to_user(data_out, data, copy_size))
369 		goto out;
370 	if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
371 		goto out;
372 	if (err != -ENOSPC)
373 		err = 0;
374 out:
375 	return err;
376 }
377 
378 /**
379  * range_is_zero - test whether buffer is initialized
380  * @buf: buffer to check
381  * @from: check from this position
382  * @to: check up until (excluding) this position
383  *
384  * This function returns true if the there is a non-zero byte
385  * in the buf in the range [from,to).
386  */
387 static inline bool range_is_zero(void *buf, size_t from, size_t to)
388 {
389 	return !memchr_inv((u8 *)buf + from, 0, to - from);
390 }
391 
392 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
393 {
394 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
395 
396 	if (!__skb)
397 		return 0;
398 
399 	/* make sure the fields we don't use are zeroed */
400 	if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
401 		return -EINVAL;
402 
403 	/* mark is allowed */
404 
405 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
406 			   offsetof(struct __sk_buff, priority)))
407 		return -EINVAL;
408 
409 	/* priority is allowed */
410 
411 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, priority),
412 			   offsetof(struct __sk_buff, ifindex)))
413 		return -EINVAL;
414 
415 	/* ifindex is allowed */
416 
417 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
418 			   offsetof(struct __sk_buff, cb)))
419 		return -EINVAL;
420 
421 	/* cb is allowed */
422 
423 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
424 			   offsetof(struct __sk_buff, tstamp)))
425 		return -EINVAL;
426 
427 	/* tstamp is allowed */
428 	/* wire_len is allowed */
429 	/* gso_segs is allowed */
430 
431 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
432 			   offsetof(struct __sk_buff, gso_size)))
433 		return -EINVAL;
434 
435 	/* gso_size is allowed */
436 
437 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
438 			   sizeof(struct __sk_buff)))
439 		return -EINVAL;
440 
441 	skb->mark = __skb->mark;
442 	skb->priority = __skb->priority;
443 	skb->tstamp = __skb->tstamp;
444 	memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
445 
446 	if (__skb->wire_len == 0) {
447 		cb->pkt_len = skb->len;
448 	} else {
449 		if (__skb->wire_len < skb->len ||
450 		    __skb->wire_len > GSO_MAX_SIZE)
451 			return -EINVAL;
452 		cb->pkt_len = __skb->wire_len;
453 	}
454 
455 	if (__skb->gso_segs > GSO_MAX_SEGS)
456 		return -EINVAL;
457 	skb_shinfo(skb)->gso_segs = __skb->gso_segs;
458 	skb_shinfo(skb)->gso_size = __skb->gso_size;
459 
460 	return 0;
461 }
462 
463 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
464 {
465 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
466 
467 	if (!__skb)
468 		return;
469 
470 	__skb->mark = skb->mark;
471 	__skb->priority = skb->priority;
472 	__skb->ifindex = skb->dev->ifindex;
473 	__skb->tstamp = skb->tstamp;
474 	memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
475 	__skb->wire_len = cb->pkt_len;
476 	__skb->gso_segs = skb_shinfo(skb)->gso_segs;
477 }
478 
479 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
480 			  union bpf_attr __user *uattr)
481 {
482 	bool is_l2 = false, is_direct_pkt_access = false;
483 	struct net *net = current->nsproxy->net_ns;
484 	struct net_device *dev = net->loopback_dev;
485 	u32 size = kattr->test.data_size_in;
486 	u32 repeat = kattr->test.repeat;
487 	struct __sk_buff *ctx = NULL;
488 	u32 retval, duration;
489 	int hh_len = ETH_HLEN;
490 	struct sk_buff *skb;
491 	struct sock *sk;
492 	void *data;
493 	int ret;
494 
495 	if (kattr->test.flags || kattr->test.cpu)
496 		return -EINVAL;
497 
498 	data = bpf_test_init(kattr, size, NET_SKB_PAD + NET_IP_ALIGN,
499 			     SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
500 	if (IS_ERR(data))
501 		return PTR_ERR(data);
502 
503 	ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
504 	if (IS_ERR(ctx)) {
505 		kfree(data);
506 		return PTR_ERR(ctx);
507 	}
508 
509 	switch (prog->type) {
510 	case BPF_PROG_TYPE_SCHED_CLS:
511 	case BPF_PROG_TYPE_SCHED_ACT:
512 		is_l2 = true;
513 		fallthrough;
514 	case BPF_PROG_TYPE_LWT_IN:
515 	case BPF_PROG_TYPE_LWT_OUT:
516 	case BPF_PROG_TYPE_LWT_XMIT:
517 		is_direct_pkt_access = true;
518 		break;
519 	default:
520 		break;
521 	}
522 
523 	sk = kzalloc(sizeof(struct sock), GFP_USER);
524 	if (!sk) {
525 		kfree(data);
526 		kfree(ctx);
527 		return -ENOMEM;
528 	}
529 	sock_net_set(sk, net);
530 	sock_init_data(NULL, sk);
531 
532 	skb = build_skb(data, 0);
533 	if (!skb) {
534 		kfree(data);
535 		kfree(ctx);
536 		kfree(sk);
537 		return -ENOMEM;
538 	}
539 	skb->sk = sk;
540 
541 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
542 	__skb_put(skb, size);
543 	if (ctx && ctx->ifindex > 1) {
544 		dev = dev_get_by_index(net, ctx->ifindex);
545 		if (!dev) {
546 			ret = -ENODEV;
547 			goto out;
548 		}
549 	}
550 	skb->protocol = eth_type_trans(skb, dev);
551 	skb_reset_network_header(skb);
552 
553 	switch (skb->protocol) {
554 	case htons(ETH_P_IP):
555 		sk->sk_family = AF_INET;
556 		if (sizeof(struct iphdr) <= skb_headlen(skb)) {
557 			sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
558 			sk->sk_daddr = ip_hdr(skb)->daddr;
559 		}
560 		break;
561 #if IS_ENABLED(CONFIG_IPV6)
562 	case htons(ETH_P_IPV6):
563 		sk->sk_family = AF_INET6;
564 		if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
565 			sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
566 			sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
567 		}
568 		break;
569 #endif
570 	default:
571 		break;
572 	}
573 
574 	if (is_l2)
575 		__skb_push(skb, hh_len);
576 	if (is_direct_pkt_access)
577 		bpf_compute_data_pointers(skb);
578 	ret = convert___skb_to_skb(skb, ctx);
579 	if (ret)
580 		goto out;
581 	ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
582 	if (ret)
583 		goto out;
584 	if (!is_l2) {
585 		if (skb_headroom(skb) < hh_len) {
586 			int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
587 
588 			if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
589 				ret = -ENOMEM;
590 				goto out;
591 			}
592 		}
593 		memset(__skb_push(skb, hh_len), 0, hh_len);
594 	}
595 	convert_skb_to___skb(skb, ctx);
596 
597 	size = skb->len;
598 	/* bpf program can never convert linear skb to non-linear */
599 	if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
600 		size = skb_headlen(skb);
601 	ret = bpf_test_finish(kattr, uattr, skb->data, size, retval, duration);
602 	if (!ret)
603 		ret = bpf_ctx_finish(kattr, uattr, ctx,
604 				     sizeof(struct __sk_buff));
605 out:
606 	if (dev && dev != net->loopback_dev)
607 		dev_put(dev);
608 	kfree_skb(skb);
609 	bpf_sk_storage_free(sk);
610 	kfree(sk);
611 	kfree(ctx);
612 	return ret;
613 }
614 
615 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
616 			  union bpf_attr __user *uattr)
617 {
618 	u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
619 	u32 headroom = XDP_PACKET_HEADROOM;
620 	u32 size = kattr->test.data_size_in;
621 	u32 repeat = kattr->test.repeat;
622 	struct netdev_rx_queue *rxqueue;
623 	struct xdp_buff xdp = {};
624 	u32 retval, duration;
625 	u32 max_data_sz;
626 	void *data;
627 	int ret;
628 
629 	if (kattr->test.ctx_in || kattr->test.ctx_out)
630 		return -EINVAL;
631 
632 	/* XDP have extra tailroom as (most) drivers use full page */
633 	max_data_sz = 4096 - headroom - tailroom;
634 
635 	data = bpf_test_init(kattr, max_data_sz, headroom, tailroom);
636 	if (IS_ERR(data))
637 		return PTR_ERR(data);
638 
639 	xdp.data_hard_start = data;
640 	xdp.data = data + headroom;
641 	xdp.data_meta = xdp.data;
642 	xdp.data_end = xdp.data + size;
643 	xdp.frame_sz = headroom + max_data_sz + tailroom;
644 
645 	rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
646 	xdp.rxq = &rxqueue->xdp_rxq;
647 	bpf_prog_change_xdp(NULL, prog);
648 	ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
649 	if (ret)
650 		goto out;
651 	if (xdp.data != data + headroom || xdp.data_end != xdp.data + size)
652 		size = xdp.data_end - xdp.data;
653 	ret = bpf_test_finish(kattr, uattr, xdp.data, size, retval, duration);
654 out:
655 	bpf_prog_change_xdp(prog, NULL);
656 	kfree(data);
657 	return ret;
658 }
659 
660 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
661 {
662 	/* make sure the fields we don't use are zeroed */
663 	if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
664 		return -EINVAL;
665 
666 	/* flags is allowed */
667 
668 	if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
669 			   sizeof(struct bpf_flow_keys)))
670 		return -EINVAL;
671 
672 	return 0;
673 }
674 
675 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
676 				     const union bpf_attr *kattr,
677 				     union bpf_attr __user *uattr)
678 {
679 	u32 size = kattr->test.data_size_in;
680 	struct bpf_flow_dissector ctx = {};
681 	u32 repeat = kattr->test.repeat;
682 	struct bpf_flow_keys *user_ctx;
683 	struct bpf_flow_keys flow_keys;
684 	u64 time_start, time_spent = 0;
685 	const struct ethhdr *eth;
686 	unsigned int flags = 0;
687 	u32 retval, duration;
688 	void *data;
689 	int ret;
690 	u32 i;
691 
692 	if (prog->type != BPF_PROG_TYPE_FLOW_DISSECTOR)
693 		return -EINVAL;
694 
695 	if (kattr->test.flags || kattr->test.cpu)
696 		return -EINVAL;
697 
698 	if (size < ETH_HLEN)
699 		return -EINVAL;
700 
701 	data = bpf_test_init(kattr, size, 0, 0);
702 	if (IS_ERR(data))
703 		return PTR_ERR(data);
704 
705 	eth = (struct ethhdr *)data;
706 
707 	if (!repeat)
708 		repeat = 1;
709 
710 	user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
711 	if (IS_ERR(user_ctx)) {
712 		kfree(data);
713 		return PTR_ERR(user_ctx);
714 	}
715 	if (user_ctx) {
716 		ret = verify_user_bpf_flow_keys(user_ctx);
717 		if (ret)
718 			goto out;
719 		flags = user_ctx->flags;
720 	}
721 
722 	ctx.flow_keys = &flow_keys;
723 	ctx.data = data;
724 	ctx.data_end = (__u8 *)data + size;
725 
726 	rcu_read_lock();
727 	preempt_disable();
728 	time_start = ktime_get_ns();
729 	for (i = 0; i < repeat; i++) {
730 		retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
731 					  size, flags);
732 
733 		if (signal_pending(current)) {
734 			preempt_enable();
735 			rcu_read_unlock();
736 
737 			ret = -EINTR;
738 			goto out;
739 		}
740 
741 		if (need_resched()) {
742 			time_spent += ktime_get_ns() - time_start;
743 			preempt_enable();
744 			rcu_read_unlock();
745 
746 			cond_resched();
747 
748 			rcu_read_lock();
749 			preempt_disable();
750 			time_start = ktime_get_ns();
751 		}
752 	}
753 	time_spent += ktime_get_ns() - time_start;
754 	preempt_enable();
755 	rcu_read_unlock();
756 
757 	do_div(time_spent, repeat);
758 	duration = time_spent > U32_MAX ? U32_MAX : (u32)time_spent;
759 
760 	ret = bpf_test_finish(kattr, uattr, &flow_keys, sizeof(flow_keys),
761 			      retval, duration);
762 	if (!ret)
763 		ret = bpf_ctx_finish(kattr, uattr, user_ctx,
764 				     sizeof(struct bpf_flow_keys));
765 
766 out:
767 	kfree(user_ctx);
768 	kfree(data);
769 	return ret;
770 }
771