xref: /openbmc/linux/net/bpf/test_run.c (revision 62eab49f)
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 	    ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
277 		return -EINVAL;
278 
279 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
280 		return -EINVAL;
281 
282 	if (ctx_size_in) {
283 		info.ctx = kzalloc(ctx_size_in, GFP_USER);
284 		if (!info.ctx)
285 			return -ENOMEM;
286 		if (copy_from_user(info.ctx, ctx_in, ctx_size_in)) {
287 			err = -EFAULT;
288 			goto out;
289 		}
290 	} else {
291 		info.ctx = NULL;
292 	}
293 
294 	info.prog = prog;
295 
296 	current_cpu = get_cpu();
297 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
298 	    cpu == current_cpu) {
299 		__bpf_prog_test_run_raw_tp(&info);
300 	} else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
301 		/* smp_call_function_single() also checks cpu_online()
302 		 * after csd_lock(). However, since cpu is from user
303 		 * space, let's do an extra quick check to filter out
304 		 * invalid value before smp_call_function_single().
305 		 */
306 		err = -ENXIO;
307 	} else {
308 		err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
309 					       &info, 1);
310 	}
311 	put_cpu();
312 
313 	if (!err &&
314 	    copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
315 		err = -EFAULT;
316 
317 out:
318 	kfree(info.ctx);
319 	return err;
320 }
321 
322 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
323 {
324 	void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
325 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
326 	u32 size = kattr->test.ctx_size_in;
327 	void *data;
328 	int err;
329 
330 	if (!data_in && !data_out)
331 		return NULL;
332 
333 	data = kzalloc(max_size, GFP_USER);
334 	if (!data)
335 		return ERR_PTR(-ENOMEM);
336 
337 	if (data_in) {
338 		err = bpf_check_uarg_tail_zero(data_in, max_size, size);
339 		if (err) {
340 			kfree(data);
341 			return ERR_PTR(err);
342 		}
343 
344 		size = min_t(u32, max_size, size);
345 		if (copy_from_user(data, data_in, size)) {
346 			kfree(data);
347 			return ERR_PTR(-EFAULT);
348 		}
349 	}
350 	return data;
351 }
352 
353 static int bpf_ctx_finish(const union bpf_attr *kattr,
354 			  union bpf_attr __user *uattr, const void *data,
355 			  u32 size)
356 {
357 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
358 	int err = -EFAULT;
359 	u32 copy_size = size;
360 
361 	if (!data || !data_out)
362 		return 0;
363 
364 	if (copy_size > kattr->test.ctx_size_out) {
365 		copy_size = kattr->test.ctx_size_out;
366 		err = -ENOSPC;
367 	}
368 
369 	if (copy_to_user(data_out, data, copy_size))
370 		goto out;
371 	if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
372 		goto out;
373 	if (err != -ENOSPC)
374 		err = 0;
375 out:
376 	return err;
377 }
378 
379 /**
380  * range_is_zero - test whether buffer is initialized
381  * @buf: buffer to check
382  * @from: check from this position
383  * @to: check up until (excluding) this position
384  *
385  * This function returns true if the there is a non-zero byte
386  * in the buf in the range [from,to).
387  */
388 static inline bool range_is_zero(void *buf, size_t from, size_t to)
389 {
390 	return !memchr_inv((u8 *)buf + from, 0, to - from);
391 }
392 
393 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
394 {
395 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
396 
397 	if (!__skb)
398 		return 0;
399 
400 	/* make sure the fields we don't use are zeroed */
401 	if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
402 		return -EINVAL;
403 
404 	/* mark is allowed */
405 
406 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
407 			   offsetof(struct __sk_buff, priority)))
408 		return -EINVAL;
409 
410 	/* priority is allowed */
411 
412 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, priority),
413 			   offsetof(struct __sk_buff, ifindex)))
414 		return -EINVAL;
415 
416 	/* ifindex is allowed */
417 
418 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
419 			   offsetof(struct __sk_buff, cb)))
420 		return -EINVAL;
421 
422 	/* cb is allowed */
423 
424 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
425 			   offsetof(struct __sk_buff, tstamp)))
426 		return -EINVAL;
427 
428 	/* tstamp is allowed */
429 	/* wire_len is allowed */
430 	/* gso_segs is allowed */
431 
432 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
433 			   offsetof(struct __sk_buff, gso_size)))
434 		return -EINVAL;
435 
436 	/* gso_size is allowed */
437 
438 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
439 			   sizeof(struct __sk_buff)))
440 		return -EINVAL;
441 
442 	skb->mark = __skb->mark;
443 	skb->priority = __skb->priority;
444 	skb->tstamp = __skb->tstamp;
445 	memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
446 
447 	if (__skb->wire_len == 0) {
448 		cb->pkt_len = skb->len;
449 	} else {
450 		if (__skb->wire_len < skb->len ||
451 		    __skb->wire_len > GSO_MAX_SIZE)
452 			return -EINVAL;
453 		cb->pkt_len = __skb->wire_len;
454 	}
455 
456 	if (__skb->gso_segs > GSO_MAX_SEGS)
457 		return -EINVAL;
458 	skb_shinfo(skb)->gso_segs = __skb->gso_segs;
459 	skb_shinfo(skb)->gso_size = __skb->gso_size;
460 
461 	return 0;
462 }
463 
464 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
465 {
466 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
467 
468 	if (!__skb)
469 		return;
470 
471 	__skb->mark = skb->mark;
472 	__skb->priority = skb->priority;
473 	__skb->ifindex = skb->dev->ifindex;
474 	__skb->tstamp = skb->tstamp;
475 	memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
476 	__skb->wire_len = cb->pkt_len;
477 	__skb->gso_segs = skb_shinfo(skb)->gso_segs;
478 }
479 
480 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
481 			  union bpf_attr __user *uattr)
482 {
483 	bool is_l2 = false, is_direct_pkt_access = false;
484 	struct net *net = current->nsproxy->net_ns;
485 	struct net_device *dev = net->loopback_dev;
486 	u32 size = kattr->test.data_size_in;
487 	u32 repeat = kattr->test.repeat;
488 	struct __sk_buff *ctx = NULL;
489 	u32 retval, duration;
490 	int hh_len = ETH_HLEN;
491 	struct sk_buff *skb;
492 	struct sock *sk;
493 	void *data;
494 	int ret;
495 
496 	if (kattr->test.flags || kattr->test.cpu)
497 		return -EINVAL;
498 
499 	data = bpf_test_init(kattr, size, NET_SKB_PAD + NET_IP_ALIGN,
500 			     SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
501 	if (IS_ERR(data))
502 		return PTR_ERR(data);
503 
504 	ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
505 	if (IS_ERR(ctx)) {
506 		kfree(data);
507 		return PTR_ERR(ctx);
508 	}
509 
510 	switch (prog->type) {
511 	case BPF_PROG_TYPE_SCHED_CLS:
512 	case BPF_PROG_TYPE_SCHED_ACT:
513 		is_l2 = true;
514 		fallthrough;
515 	case BPF_PROG_TYPE_LWT_IN:
516 	case BPF_PROG_TYPE_LWT_OUT:
517 	case BPF_PROG_TYPE_LWT_XMIT:
518 		is_direct_pkt_access = true;
519 		break;
520 	default:
521 		break;
522 	}
523 
524 	sk = kzalloc(sizeof(struct sock), GFP_USER);
525 	if (!sk) {
526 		kfree(data);
527 		kfree(ctx);
528 		return -ENOMEM;
529 	}
530 	sock_net_set(sk, net);
531 	sock_init_data(NULL, sk);
532 
533 	skb = build_skb(data, 0);
534 	if (!skb) {
535 		kfree(data);
536 		kfree(ctx);
537 		kfree(sk);
538 		return -ENOMEM;
539 	}
540 	skb->sk = sk;
541 
542 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
543 	__skb_put(skb, size);
544 	if (ctx && ctx->ifindex > 1) {
545 		dev = dev_get_by_index(net, ctx->ifindex);
546 		if (!dev) {
547 			ret = -ENODEV;
548 			goto out;
549 		}
550 	}
551 	skb->protocol = eth_type_trans(skb, dev);
552 	skb_reset_network_header(skb);
553 
554 	switch (skb->protocol) {
555 	case htons(ETH_P_IP):
556 		sk->sk_family = AF_INET;
557 		if (sizeof(struct iphdr) <= skb_headlen(skb)) {
558 			sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
559 			sk->sk_daddr = ip_hdr(skb)->daddr;
560 		}
561 		break;
562 #if IS_ENABLED(CONFIG_IPV6)
563 	case htons(ETH_P_IPV6):
564 		sk->sk_family = AF_INET6;
565 		if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
566 			sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
567 			sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
568 		}
569 		break;
570 #endif
571 	default:
572 		break;
573 	}
574 
575 	if (is_l2)
576 		__skb_push(skb, hh_len);
577 	if (is_direct_pkt_access)
578 		bpf_compute_data_pointers(skb);
579 	ret = convert___skb_to_skb(skb, ctx);
580 	if (ret)
581 		goto out;
582 	ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
583 	if (ret)
584 		goto out;
585 	if (!is_l2) {
586 		if (skb_headroom(skb) < hh_len) {
587 			int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
588 
589 			if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
590 				ret = -ENOMEM;
591 				goto out;
592 			}
593 		}
594 		memset(__skb_push(skb, hh_len), 0, hh_len);
595 	}
596 	convert_skb_to___skb(skb, ctx);
597 
598 	size = skb->len;
599 	/* bpf program can never convert linear skb to non-linear */
600 	if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
601 		size = skb_headlen(skb);
602 	ret = bpf_test_finish(kattr, uattr, skb->data, size, retval, duration);
603 	if (!ret)
604 		ret = bpf_ctx_finish(kattr, uattr, ctx,
605 				     sizeof(struct __sk_buff));
606 out:
607 	if (dev && dev != net->loopback_dev)
608 		dev_put(dev);
609 	kfree_skb(skb);
610 	bpf_sk_storage_free(sk);
611 	kfree(sk);
612 	kfree(ctx);
613 	return ret;
614 }
615 
616 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
617 			  union bpf_attr __user *uattr)
618 {
619 	u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
620 	u32 headroom = XDP_PACKET_HEADROOM;
621 	u32 size = kattr->test.data_size_in;
622 	u32 repeat = kattr->test.repeat;
623 	struct netdev_rx_queue *rxqueue;
624 	struct xdp_buff xdp = {};
625 	u32 retval, duration;
626 	u32 max_data_sz;
627 	void *data;
628 	int ret;
629 
630 	if (kattr->test.ctx_in || kattr->test.ctx_out)
631 		return -EINVAL;
632 
633 	/* XDP have extra tailroom as (most) drivers use full page */
634 	max_data_sz = 4096 - headroom - tailroom;
635 
636 	data = bpf_test_init(kattr, max_data_sz, headroom, tailroom);
637 	if (IS_ERR(data))
638 		return PTR_ERR(data);
639 
640 	rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
641 	xdp_init_buff(&xdp, headroom + max_data_sz + tailroom,
642 		      &rxqueue->xdp_rxq);
643 	xdp_prepare_buff(&xdp, data, headroom, size, true);
644 
645 	bpf_prog_change_xdp(NULL, prog);
646 	ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
647 	if (ret)
648 		goto out;
649 	if (xdp.data != data + headroom || xdp.data_end != xdp.data + size)
650 		size = xdp.data_end - xdp.data;
651 	ret = bpf_test_finish(kattr, uattr, xdp.data, size, retval, duration);
652 out:
653 	bpf_prog_change_xdp(prog, NULL);
654 	kfree(data);
655 	return ret;
656 }
657 
658 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
659 {
660 	/* make sure the fields we don't use are zeroed */
661 	if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
662 		return -EINVAL;
663 
664 	/* flags is allowed */
665 
666 	if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
667 			   sizeof(struct bpf_flow_keys)))
668 		return -EINVAL;
669 
670 	return 0;
671 }
672 
673 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
674 				     const union bpf_attr *kattr,
675 				     union bpf_attr __user *uattr)
676 {
677 	u32 size = kattr->test.data_size_in;
678 	struct bpf_flow_dissector ctx = {};
679 	u32 repeat = kattr->test.repeat;
680 	struct bpf_flow_keys *user_ctx;
681 	struct bpf_flow_keys flow_keys;
682 	u64 time_start, time_spent = 0;
683 	const struct ethhdr *eth;
684 	unsigned int flags = 0;
685 	u32 retval, duration;
686 	void *data;
687 	int ret;
688 	u32 i;
689 
690 	if (prog->type != BPF_PROG_TYPE_FLOW_DISSECTOR)
691 		return -EINVAL;
692 
693 	if (kattr->test.flags || kattr->test.cpu)
694 		return -EINVAL;
695 
696 	if (size < ETH_HLEN)
697 		return -EINVAL;
698 
699 	data = bpf_test_init(kattr, size, 0, 0);
700 	if (IS_ERR(data))
701 		return PTR_ERR(data);
702 
703 	eth = (struct ethhdr *)data;
704 
705 	if (!repeat)
706 		repeat = 1;
707 
708 	user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
709 	if (IS_ERR(user_ctx)) {
710 		kfree(data);
711 		return PTR_ERR(user_ctx);
712 	}
713 	if (user_ctx) {
714 		ret = verify_user_bpf_flow_keys(user_ctx);
715 		if (ret)
716 			goto out;
717 		flags = user_ctx->flags;
718 	}
719 
720 	ctx.flow_keys = &flow_keys;
721 	ctx.data = data;
722 	ctx.data_end = (__u8 *)data + size;
723 
724 	rcu_read_lock();
725 	preempt_disable();
726 	time_start = ktime_get_ns();
727 	for (i = 0; i < repeat; i++) {
728 		retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
729 					  size, flags);
730 
731 		if (signal_pending(current)) {
732 			preempt_enable();
733 			rcu_read_unlock();
734 
735 			ret = -EINTR;
736 			goto out;
737 		}
738 
739 		if (need_resched()) {
740 			time_spent += ktime_get_ns() - time_start;
741 			preempt_enable();
742 			rcu_read_unlock();
743 
744 			cond_resched();
745 
746 			rcu_read_lock();
747 			preempt_disable();
748 			time_start = ktime_get_ns();
749 		}
750 	}
751 	time_spent += ktime_get_ns() - time_start;
752 	preempt_enable();
753 	rcu_read_unlock();
754 
755 	do_div(time_spent, repeat);
756 	duration = time_spent > U32_MAX ? U32_MAX : (u32)time_spent;
757 
758 	ret = bpf_test_finish(kattr, uattr, &flow_keys, sizeof(flow_keys),
759 			      retval, duration);
760 	if (!ret)
761 		ret = bpf_ctx_finish(kattr, uattr, user_ctx,
762 				     sizeof(struct bpf_flow_keys));
763 
764 out:
765 	kfree(user_ctx);
766 	kfree(data);
767 	return ret;
768 }
769