xref: /openbmc/linux/net/bpf/test_run.c (revision 6a143a7c) 
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 <net/net_namespace.h>
14 #include <linux/error-injection.h>
15 #include <linux/smp.h>
16 #include <linux/sock_diag.h>
17 
18 #define CREATE_TRACE_POINTS
19 #include <trace/events/bpf_test_run.h>
20 
21 struct bpf_test_timer {
22 	enum { NO_PREEMPT, NO_MIGRATE } mode;
23 	u32 i;
24 	u64 time_start, time_spent;
25 };
26 
27 static void bpf_test_timer_enter(struct bpf_test_timer *t)
28 	__acquires(rcu)
29 {
30 	rcu_read_lock();
31 	if (t->mode == NO_PREEMPT)
32 		preempt_disable();
33 	else
34 		migrate_disable();
35 
36 	t->time_start = ktime_get_ns();
37 }
38 
39 static void bpf_test_timer_leave(struct bpf_test_timer *t)
40 	__releases(rcu)
41 {
42 	t->time_start = 0;
43 
44 	if (t->mode == NO_PREEMPT)
45 		preempt_enable();
46 	else
47 		migrate_enable();
48 	rcu_read_unlock();
49 }
50 
51 static bool bpf_test_timer_continue(struct bpf_test_timer *t, u32 repeat, int *err, u32 *duration)
52 	__must_hold(rcu)
53 {
54 	t->i++;
55 	if (t->i >= repeat) {
56 		/* We're done. */
57 		t->time_spent += ktime_get_ns() - t->time_start;
58 		do_div(t->time_spent, t->i);
59 		*duration = t->time_spent > U32_MAX ? U32_MAX : (u32)t->time_spent;
60 		*err = 0;
61 		goto reset;
62 	}
63 
64 	if (signal_pending(current)) {
65 		/* During iteration: we've been cancelled, abort. */
66 		*err = -EINTR;
67 		goto reset;
68 	}
69 
70 	if (need_resched()) {
71 		/* During iteration: we need to reschedule between runs. */
72 		t->time_spent += ktime_get_ns() - t->time_start;
73 		bpf_test_timer_leave(t);
74 		cond_resched();
75 		bpf_test_timer_enter(t);
76 	}
77 
78 	/* Do another round. */
79 	return true;
80 
81 reset:
82 	t->i = 0;
83 	return false;
84 }
85 
86 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
87 			u32 *retval, u32 *time, bool xdp)
88 {
89 	struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE] = { NULL };
90 	struct bpf_test_timer t = { NO_MIGRATE };
91 	enum bpf_cgroup_storage_type stype;
92 	int ret;
93 
94 	for_each_cgroup_storage_type(stype) {
95 		storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
96 		if (IS_ERR(storage[stype])) {
97 			storage[stype] = NULL;
98 			for_each_cgroup_storage_type(stype)
99 				bpf_cgroup_storage_free(storage[stype]);
100 			return -ENOMEM;
101 		}
102 	}
103 
104 	if (!repeat)
105 		repeat = 1;
106 
107 	bpf_test_timer_enter(&t);
108 	do {
109 		bpf_cgroup_storage_set(storage);
110 
111 		if (xdp)
112 			*retval = bpf_prog_run_xdp(prog, ctx);
113 		else
114 			*retval = BPF_PROG_RUN(prog, ctx);
115 	} while (bpf_test_timer_continue(&t, repeat, &ret, time));
116 	bpf_test_timer_leave(&t);
117 
118 	for_each_cgroup_storage_type(stype)
119 		bpf_cgroup_storage_free(storage[stype]);
120 
121 	return ret;
122 }
123 
124 static int bpf_test_finish(const union bpf_attr *kattr,
125 			   union bpf_attr __user *uattr, const void *data,
126 			   u32 size, u32 retval, u32 duration)
127 {
128 	void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
129 	int err = -EFAULT;
130 	u32 copy_size = size;
131 
132 	/* Clamp copy if the user has provided a size hint, but copy the full
133 	 * buffer if not to retain old behaviour.
134 	 */
135 	if (kattr->test.data_size_out &&
136 	    copy_size > kattr->test.data_size_out) {
137 		copy_size = kattr->test.data_size_out;
138 		err = -ENOSPC;
139 	}
140 
141 	if (data_out && copy_to_user(data_out, data, copy_size))
142 		goto out;
143 	if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
144 		goto out;
145 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
146 		goto out;
147 	if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
148 		goto out;
149 	if (err != -ENOSPC)
150 		err = 0;
151 out:
152 	trace_bpf_test_finish(&err);
153 	return err;
154 }
155 
156 /* Integer types of various sizes and pointer combinations cover variety of
157  * architecture dependent calling conventions. 7+ can be supported in the
158  * future.
159  */
160 __diag_push();
161 __diag_ignore(GCC, 8, "-Wmissing-prototypes",
162 	      "Global functions as their definitions will be in vmlinux BTF");
163 int noinline bpf_fentry_test1(int a)
164 {
165 	return a + 1;
166 }
167 
168 int noinline bpf_fentry_test2(int a, u64 b)
169 {
170 	return a + b;
171 }
172 
173 int noinline bpf_fentry_test3(char a, int b, u64 c)
174 {
175 	return a + b + c;
176 }
177 
178 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
179 {
180 	return (long)a + b + c + d;
181 }
182 
183 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
184 {
185 	return a + (long)b + c + d + e;
186 }
187 
188 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
189 {
190 	return a + (long)b + c + d + (long)e + f;
191 }
192 
193 struct bpf_fentry_test_t {
194 	struct bpf_fentry_test_t *a;
195 };
196 
197 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
198 {
199 	return (long)arg;
200 }
201 
202 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
203 {
204 	return (long)arg->a;
205 }
206 
207 int noinline bpf_modify_return_test(int a, int *b)
208 {
209 	*b += 1;
210 	return a + *b;
211 }
212 __diag_pop();
213 
214 ALLOW_ERROR_INJECTION(bpf_modify_return_test, ERRNO);
215 
216 static void *bpf_test_init(const union bpf_attr *kattr, u32 size,
217 			   u32 headroom, u32 tailroom)
218 {
219 	void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
220 	u32 user_size = kattr->test.data_size_in;
221 	void *data;
222 
223 	if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
224 		return ERR_PTR(-EINVAL);
225 
226 	if (user_size > size)
227 		return ERR_PTR(-EMSGSIZE);
228 
229 	data = kzalloc(size + headroom + tailroom, GFP_USER);
230 	if (!data)
231 		return ERR_PTR(-ENOMEM);
232 
233 	if (copy_from_user(data + headroom, data_in, user_size)) {
234 		kfree(data);
235 		return ERR_PTR(-EFAULT);
236 	}
237 
238 	return data;
239 }
240 
241 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
242 			      const union bpf_attr *kattr,
243 			      union bpf_attr __user *uattr)
244 {
245 	struct bpf_fentry_test_t arg = {};
246 	u16 side_effect = 0, ret = 0;
247 	int b = 2, err = -EFAULT;
248 	u32 retval = 0;
249 
250 	if (kattr->test.flags || kattr->test.cpu)
251 		return -EINVAL;
252 
253 	switch (prog->expected_attach_type) {
254 	case BPF_TRACE_FENTRY:
255 	case BPF_TRACE_FEXIT:
256 		if (bpf_fentry_test1(1) != 2 ||
257 		    bpf_fentry_test2(2, 3) != 5 ||
258 		    bpf_fentry_test3(4, 5, 6) != 15 ||
259 		    bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
260 		    bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
261 		    bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
262 		    bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
263 		    bpf_fentry_test8(&arg) != 0)
264 			goto out;
265 		break;
266 	case BPF_MODIFY_RETURN:
267 		ret = bpf_modify_return_test(1, &b);
268 		if (b != 2)
269 			side_effect = 1;
270 		break;
271 	default:
272 		goto out;
273 	}
274 
275 	retval = ((u32)side_effect << 16) | ret;
276 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
277 		goto out;
278 
279 	err = 0;
280 out:
281 	trace_bpf_test_finish(&err);
282 	return err;
283 }
284 
285 struct bpf_raw_tp_test_run_info {
286 	struct bpf_prog *prog;
287 	void *ctx;
288 	u32 retval;
289 };
290 
291 static void
292 __bpf_prog_test_run_raw_tp(void *data)
293 {
294 	struct bpf_raw_tp_test_run_info *info = data;
295 
296 	rcu_read_lock();
297 	info->retval = BPF_PROG_RUN(info->prog, info->ctx);
298 	rcu_read_unlock();
299 }
300 
301 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
302 			     const union bpf_attr *kattr,
303 			     union bpf_attr __user *uattr)
304 {
305 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
306 	__u32 ctx_size_in = kattr->test.ctx_size_in;
307 	struct bpf_raw_tp_test_run_info info;
308 	int cpu = kattr->test.cpu, err = 0;
309 	int current_cpu;
310 
311 	/* doesn't support data_in/out, ctx_out, duration, or repeat */
312 	if (kattr->test.data_in || kattr->test.data_out ||
313 	    kattr->test.ctx_out || kattr->test.duration ||
314 	    kattr->test.repeat)
315 		return -EINVAL;
316 
317 	if (ctx_size_in < prog->aux->max_ctx_offset ||
318 	    ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
319 		return -EINVAL;
320 
321 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
322 		return -EINVAL;
323 
324 	if (ctx_size_in) {
325 		info.ctx = kzalloc(ctx_size_in, GFP_USER);
326 		if (!info.ctx)
327 			return -ENOMEM;
328 		if (copy_from_user(info.ctx, ctx_in, ctx_size_in)) {
329 			err = -EFAULT;
330 			goto out;
331 		}
332 	} else {
333 		info.ctx = NULL;
334 	}
335 
336 	info.prog = prog;
337 
338 	current_cpu = get_cpu();
339 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
340 	    cpu == current_cpu) {
341 		__bpf_prog_test_run_raw_tp(&info);
342 	} else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
343 		/* smp_call_function_single() also checks cpu_online()
344 		 * after csd_lock(). However, since cpu is from user
345 		 * space, let's do an extra quick check to filter out
346 		 * invalid value before smp_call_function_single().
347 		 */
348 		err = -ENXIO;
349 	} else {
350 		err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
351 					       &info, 1);
352 	}
353 	put_cpu();
354 
355 	if (!err &&
356 	    copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
357 		err = -EFAULT;
358 
359 out:
360 	kfree(info.ctx);
361 	return err;
362 }
363 
364 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
365 {
366 	void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
367 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
368 	u32 size = kattr->test.ctx_size_in;
369 	void *data;
370 	int err;
371 
372 	if (!data_in && !data_out)
373 		return NULL;
374 
375 	data = kzalloc(max_size, GFP_USER);
376 	if (!data)
377 		return ERR_PTR(-ENOMEM);
378 
379 	if (data_in) {
380 		err = bpf_check_uarg_tail_zero(data_in, max_size, size);
381 		if (err) {
382 			kfree(data);
383 			return ERR_PTR(err);
384 		}
385 
386 		size = min_t(u32, max_size, size);
387 		if (copy_from_user(data, data_in, size)) {
388 			kfree(data);
389 			return ERR_PTR(-EFAULT);
390 		}
391 	}
392 	return data;
393 }
394 
395 static int bpf_ctx_finish(const union bpf_attr *kattr,
396 			  union bpf_attr __user *uattr, const void *data,
397 			  u32 size)
398 {
399 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
400 	int err = -EFAULT;
401 	u32 copy_size = size;
402 
403 	if (!data || !data_out)
404 		return 0;
405 
406 	if (copy_size > kattr->test.ctx_size_out) {
407 		copy_size = kattr->test.ctx_size_out;
408 		err = -ENOSPC;
409 	}
410 
411 	if (copy_to_user(data_out, data, copy_size))
412 		goto out;
413 	if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
414 		goto out;
415 	if (err != -ENOSPC)
416 		err = 0;
417 out:
418 	return err;
419 }
420 
421 /**
422  * range_is_zero - test whether buffer is initialized
423  * @buf: buffer to check
424  * @from: check from this position
425  * @to: check up until (excluding) this position
426  *
427  * This function returns true if the there is a non-zero byte
428  * in the buf in the range [from,to).
429  */
430 static inline bool range_is_zero(void *buf, size_t from, size_t to)
431 {
432 	return !memchr_inv((u8 *)buf + from, 0, to - from);
433 }
434 
435 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
436 {
437 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
438 
439 	if (!__skb)
440 		return 0;
441 
442 	/* make sure the fields we don't use are zeroed */
443 	if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
444 		return -EINVAL;
445 
446 	/* mark is allowed */
447 
448 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
449 			   offsetof(struct __sk_buff, priority)))
450 		return -EINVAL;
451 
452 	/* priority is allowed */
453 
454 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, priority),
455 			   offsetof(struct __sk_buff, ifindex)))
456 		return -EINVAL;
457 
458 	/* ifindex is allowed */
459 
460 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
461 			   offsetof(struct __sk_buff, cb)))
462 		return -EINVAL;
463 
464 	/* cb is allowed */
465 
466 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
467 			   offsetof(struct __sk_buff, tstamp)))
468 		return -EINVAL;
469 
470 	/* tstamp is allowed */
471 	/* wire_len is allowed */
472 	/* gso_segs is allowed */
473 
474 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
475 			   offsetof(struct __sk_buff, gso_size)))
476 		return -EINVAL;
477 
478 	/* gso_size is allowed */
479 
480 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
481 			   sizeof(struct __sk_buff)))
482 		return -EINVAL;
483 
484 	skb->mark = __skb->mark;
485 	skb->priority = __skb->priority;
486 	skb->tstamp = __skb->tstamp;
487 	memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
488 
489 	if (__skb->wire_len == 0) {
490 		cb->pkt_len = skb->len;
491 	} else {
492 		if (__skb->wire_len < skb->len ||
493 		    __skb->wire_len > GSO_MAX_SIZE)
494 			return -EINVAL;
495 		cb->pkt_len = __skb->wire_len;
496 	}
497 
498 	if (__skb->gso_segs > GSO_MAX_SEGS)
499 		return -EINVAL;
500 	skb_shinfo(skb)->gso_segs = __skb->gso_segs;
501 	skb_shinfo(skb)->gso_size = __skb->gso_size;
502 
503 	return 0;
504 }
505 
506 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
507 {
508 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
509 
510 	if (!__skb)
511 		return;
512 
513 	__skb->mark = skb->mark;
514 	__skb->priority = skb->priority;
515 	__skb->ifindex = skb->dev->ifindex;
516 	__skb->tstamp = skb->tstamp;
517 	memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
518 	__skb->wire_len = cb->pkt_len;
519 	__skb->gso_segs = skb_shinfo(skb)->gso_segs;
520 }
521 
522 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
523 			  union bpf_attr __user *uattr)
524 {
525 	bool is_l2 = false, is_direct_pkt_access = false;
526 	struct net *net = current->nsproxy->net_ns;
527 	struct net_device *dev = net->loopback_dev;
528 	u32 size = kattr->test.data_size_in;
529 	u32 repeat = kattr->test.repeat;
530 	struct __sk_buff *ctx = NULL;
531 	u32 retval, duration;
532 	int hh_len = ETH_HLEN;
533 	struct sk_buff *skb;
534 	struct sock *sk;
535 	void *data;
536 	int ret;
537 
538 	if (kattr->test.flags || kattr->test.cpu)
539 		return -EINVAL;
540 
541 	data = bpf_test_init(kattr, size, NET_SKB_PAD + NET_IP_ALIGN,
542 			     SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
543 	if (IS_ERR(data))
544 		return PTR_ERR(data);
545 
546 	ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
547 	if (IS_ERR(ctx)) {
548 		kfree(data);
549 		return PTR_ERR(ctx);
550 	}
551 
552 	switch (prog->type) {
553 	case BPF_PROG_TYPE_SCHED_CLS:
554 	case BPF_PROG_TYPE_SCHED_ACT:
555 		is_l2 = true;
556 		fallthrough;
557 	case BPF_PROG_TYPE_LWT_IN:
558 	case BPF_PROG_TYPE_LWT_OUT:
559 	case BPF_PROG_TYPE_LWT_XMIT:
560 		is_direct_pkt_access = true;
561 		break;
562 	default:
563 		break;
564 	}
565 
566 	sk = kzalloc(sizeof(struct sock), GFP_USER);
567 	if (!sk) {
568 		kfree(data);
569 		kfree(ctx);
570 		return -ENOMEM;
571 	}
572 	sock_net_set(sk, net);
573 	sock_init_data(NULL, sk);
574 
575 	skb = build_skb(data, 0);
576 	if (!skb) {
577 		kfree(data);
578 		kfree(ctx);
579 		kfree(sk);
580 		return -ENOMEM;
581 	}
582 	skb->sk = sk;
583 
584 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
585 	__skb_put(skb, size);
586 	if (ctx && ctx->ifindex > 1) {
587 		dev = dev_get_by_index(net, ctx->ifindex);
588 		if (!dev) {
589 			ret = -ENODEV;
590 			goto out;
591 		}
592 	}
593 	skb->protocol = eth_type_trans(skb, dev);
594 	skb_reset_network_header(skb);
595 
596 	switch (skb->protocol) {
597 	case htons(ETH_P_IP):
598 		sk->sk_family = AF_INET;
599 		if (sizeof(struct iphdr) <= skb_headlen(skb)) {
600 			sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
601 			sk->sk_daddr = ip_hdr(skb)->daddr;
602 		}
603 		break;
604 #if IS_ENABLED(CONFIG_IPV6)
605 	case htons(ETH_P_IPV6):
606 		sk->sk_family = AF_INET6;
607 		if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
608 			sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
609 			sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
610 		}
611 		break;
612 #endif
613 	default:
614 		break;
615 	}
616 
617 	if (is_l2)
618 		__skb_push(skb, hh_len);
619 	if (is_direct_pkt_access)
620 		bpf_compute_data_pointers(skb);
621 	ret = convert___skb_to_skb(skb, ctx);
622 	if (ret)
623 		goto out;
624 	ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
625 	if (ret)
626 		goto out;
627 	if (!is_l2) {
628 		if (skb_headroom(skb) < hh_len) {
629 			int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
630 
631 			if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
632 				ret = -ENOMEM;
633 				goto out;
634 			}
635 		}
636 		memset(__skb_push(skb, hh_len), 0, hh_len);
637 	}
638 	convert_skb_to___skb(skb, ctx);
639 
640 	size = skb->len;
641 	/* bpf program can never convert linear skb to non-linear */
642 	if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
643 		size = skb_headlen(skb);
644 	ret = bpf_test_finish(kattr, uattr, skb->data, size, retval, duration);
645 	if (!ret)
646 		ret = bpf_ctx_finish(kattr, uattr, ctx,
647 				     sizeof(struct __sk_buff));
648 out:
649 	if (dev && dev != net->loopback_dev)
650 		dev_put(dev);
651 	kfree_skb(skb);
652 	bpf_sk_storage_free(sk);
653 	kfree(sk);
654 	kfree(ctx);
655 	return ret;
656 }
657 
658 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
659 			  union bpf_attr __user *uattr)
660 {
661 	u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
662 	u32 headroom = XDP_PACKET_HEADROOM;
663 	u32 size = kattr->test.data_size_in;
664 	u32 repeat = kattr->test.repeat;
665 	struct netdev_rx_queue *rxqueue;
666 	struct xdp_buff xdp = {};
667 	u32 retval, duration;
668 	u32 max_data_sz;
669 	void *data;
670 	int ret;
671 
672 	if (kattr->test.ctx_in || kattr->test.ctx_out)
673 		return -EINVAL;
674 
675 	/* XDP have extra tailroom as (most) drivers use full page */
676 	max_data_sz = 4096 - headroom - tailroom;
677 
678 	data = bpf_test_init(kattr, max_data_sz, headroom, tailroom);
679 	if (IS_ERR(data))
680 		return PTR_ERR(data);
681 
682 	rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
683 	xdp_init_buff(&xdp, headroom + max_data_sz + tailroom,
684 		      &rxqueue->xdp_rxq);
685 	xdp_prepare_buff(&xdp, data, headroom, size, true);
686 
687 	bpf_prog_change_xdp(NULL, prog);
688 	ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
689 	if (ret)
690 		goto out;
691 	if (xdp.data != data + headroom || xdp.data_end != xdp.data + size)
692 		size = xdp.data_end - xdp.data;
693 	ret = bpf_test_finish(kattr, uattr, xdp.data, size, retval, duration);
694 out:
695 	bpf_prog_change_xdp(prog, NULL);
696 	kfree(data);
697 	return ret;
698 }
699 
700 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
701 {
702 	/* make sure the fields we don't use are zeroed */
703 	if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
704 		return -EINVAL;
705 
706 	/* flags is allowed */
707 
708 	if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
709 			   sizeof(struct bpf_flow_keys)))
710 		return -EINVAL;
711 
712 	return 0;
713 }
714 
715 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
716 				     const union bpf_attr *kattr,
717 				     union bpf_attr __user *uattr)
718 {
719 	struct bpf_test_timer t = { NO_PREEMPT };
720 	u32 size = kattr->test.data_size_in;
721 	struct bpf_flow_dissector ctx = {};
722 	u32 repeat = kattr->test.repeat;
723 	struct bpf_flow_keys *user_ctx;
724 	struct bpf_flow_keys flow_keys;
725 	const struct ethhdr *eth;
726 	unsigned int flags = 0;
727 	u32 retval, duration;
728 	void *data;
729 	int ret;
730 
731 	if (prog->type != BPF_PROG_TYPE_FLOW_DISSECTOR)
732 		return -EINVAL;
733 
734 	if (kattr->test.flags || kattr->test.cpu)
735 		return -EINVAL;
736 
737 	if (size < ETH_HLEN)
738 		return -EINVAL;
739 
740 	data = bpf_test_init(kattr, size, 0, 0);
741 	if (IS_ERR(data))
742 		return PTR_ERR(data);
743 
744 	eth = (struct ethhdr *)data;
745 
746 	if (!repeat)
747 		repeat = 1;
748 
749 	user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
750 	if (IS_ERR(user_ctx)) {
751 		kfree(data);
752 		return PTR_ERR(user_ctx);
753 	}
754 	if (user_ctx) {
755 		ret = verify_user_bpf_flow_keys(user_ctx);
756 		if (ret)
757 			goto out;
758 		flags = user_ctx->flags;
759 	}
760 
761 	ctx.flow_keys = &flow_keys;
762 	ctx.data = data;
763 	ctx.data_end = (__u8 *)data + size;
764 
765 	bpf_test_timer_enter(&t);
766 	do {
767 		retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
768 					  size, flags);
769 	} while (bpf_test_timer_continue(&t, repeat, &ret, &duration));
770 	bpf_test_timer_leave(&t);
771 
772 	if (ret < 0)
773 		goto out;
774 
775 	ret = bpf_test_finish(kattr, uattr, &flow_keys, sizeof(flow_keys),
776 			      retval, duration);
777 	if (!ret)
778 		ret = bpf_ctx_finish(kattr, uattr, user_ctx,
779 				     sizeof(struct bpf_flow_keys));
780 
781 out:
782 	kfree(user_ctx);
783 	kfree(data);
784 	return ret;
785 }
786 
787 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
788 				union bpf_attr __user *uattr)
789 {
790 	struct bpf_test_timer t = { NO_PREEMPT };
791 	struct bpf_prog_array *progs = NULL;
792 	struct bpf_sk_lookup_kern ctx = {};
793 	u32 repeat = kattr->test.repeat;
794 	struct bpf_sk_lookup *user_ctx;
795 	u32 retval, duration;
796 	int ret = -EINVAL;
797 
798 	if (prog->type != BPF_PROG_TYPE_SK_LOOKUP)
799 		return -EINVAL;
800 
801 	if (kattr->test.flags || kattr->test.cpu)
802 		return -EINVAL;
803 
804 	if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
805 	    kattr->test.data_size_out)
806 		return -EINVAL;
807 
808 	if (!repeat)
809 		repeat = 1;
810 
811 	user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
812 	if (IS_ERR(user_ctx))
813 		return PTR_ERR(user_ctx);
814 
815 	if (!user_ctx)
816 		return -EINVAL;
817 
818 	if (user_ctx->sk)
819 		goto out;
820 
821 	if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
822 		goto out;
823 
824 	if (user_ctx->local_port > U16_MAX || user_ctx->remote_port > U16_MAX) {
825 		ret = -ERANGE;
826 		goto out;
827 	}
828 
829 	ctx.family = (u16)user_ctx->family;
830 	ctx.protocol = (u16)user_ctx->protocol;
831 	ctx.dport = (u16)user_ctx->local_port;
832 	ctx.sport = (__force __be16)user_ctx->remote_port;
833 
834 	switch (ctx.family) {
835 	case AF_INET:
836 		ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
837 		ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
838 		break;
839 
840 #if IS_ENABLED(CONFIG_IPV6)
841 	case AF_INET6:
842 		ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
843 		ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
844 		break;
845 #endif
846 
847 	default:
848 		ret = -EAFNOSUPPORT;
849 		goto out;
850 	}
851 
852 	progs = bpf_prog_array_alloc(1, GFP_KERNEL);
853 	if (!progs) {
854 		ret = -ENOMEM;
855 		goto out;
856 	}
857 
858 	progs->items[0].prog = prog;
859 
860 	bpf_test_timer_enter(&t);
861 	do {
862 		ctx.selected_sk = NULL;
863 		retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, BPF_PROG_RUN);
864 	} while (bpf_test_timer_continue(&t, repeat, &ret, &duration));
865 	bpf_test_timer_leave(&t);
866 
867 	if (ret < 0)
868 		goto out;
869 
870 	user_ctx->cookie = 0;
871 	if (ctx.selected_sk) {
872 		if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
873 			ret = -EOPNOTSUPP;
874 			goto out;
875 		}
876 
877 		user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
878 	}
879 
880 	ret = bpf_test_finish(kattr, uattr, NULL, 0, retval, duration);
881 	if (!ret)
882 		ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
883 
884 out:
885 	bpf_prog_array_free(progs);
886 	kfree(user_ctx);
887 	return ret;
888 }
889