xref: /openbmc/linux/net/bpf/test_run.c (revision 9659281c)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2017 Facebook
3  */
4 #include <linux/bpf.h>
5 #include <linux/btf_ids.h>
6 #include <linux/slab.h>
7 #include <linux/vmalloc.h>
8 #include <linux/etherdevice.h>
9 #include <linux/filter.h>
10 #include <linux/sched/signal.h>
11 #include <net/bpf_sk_storage.h>
12 #include <net/sock.h>
13 #include <net/tcp.h>
14 #include <net/net_namespace.h>
15 #include <linux/error-injection.h>
16 #include <linux/smp.h>
17 #include <linux/sock_diag.h>
18 
19 #define CREATE_TRACE_POINTS
20 #include <trace/events/bpf_test_run.h>
21 
22 struct bpf_test_timer {
23 	enum { NO_PREEMPT, NO_MIGRATE } mode;
24 	u32 i;
25 	u64 time_start, time_spent;
26 };
27 
28 static void bpf_test_timer_enter(struct bpf_test_timer *t)
29 	__acquires(rcu)
30 {
31 	rcu_read_lock();
32 	if (t->mode == NO_PREEMPT)
33 		preempt_disable();
34 	else
35 		migrate_disable();
36 
37 	t->time_start = ktime_get_ns();
38 }
39 
40 static void bpf_test_timer_leave(struct bpf_test_timer *t)
41 	__releases(rcu)
42 {
43 	t->time_start = 0;
44 
45 	if (t->mode == NO_PREEMPT)
46 		preempt_enable();
47 	else
48 		migrate_enable();
49 	rcu_read_unlock();
50 }
51 
52 static bool bpf_test_timer_continue(struct bpf_test_timer *t, u32 repeat, int *err, u32 *duration)
53 	__must_hold(rcu)
54 {
55 	t->i++;
56 	if (t->i >= repeat) {
57 		/* We're done. */
58 		t->time_spent += ktime_get_ns() - t->time_start;
59 		do_div(t->time_spent, t->i);
60 		*duration = t->time_spent > U32_MAX ? U32_MAX : (u32)t->time_spent;
61 		*err = 0;
62 		goto reset;
63 	}
64 
65 	if (signal_pending(current)) {
66 		/* During iteration: we've been cancelled, abort. */
67 		*err = -EINTR;
68 		goto reset;
69 	}
70 
71 	if (need_resched()) {
72 		/* During iteration: we need to reschedule between runs. */
73 		t->time_spent += ktime_get_ns() - t->time_start;
74 		bpf_test_timer_leave(t);
75 		cond_resched();
76 		bpf_test_timer_enter(t);
77 	}
78 
79 	/* Do another round. */
80 	return true;
81 
82 reset:
83 	t->i = 0;
84 	return false;
85 }
86 
87 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
88 			u32 *retval, u32 *time, bool xdp)
89 {
90 	struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE] = { NULL };
91 	struct bpf_test_timer t = { NO_MIGRATE };
92 	enum bpf_cgroup_storage_type stype;
93 	int ret;
94 
95 	for_each_cgroup_storage_type(stype) {
96 		storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
97 		if (IS_ERR(storage[stype])) {
98 			storage[stype] = NULL;
99 			for_each_cgroup_storage_type(stype)
100 				bpf_cgroup_storage_free(storage[stype]);
101 			return -ENOMEM;
102 		}
103 	}
104 
105 	if (!repeat)
106 		repeat = 1;
107 
108 	bpf_test_timer_enter(&t);
109 	do {
110 		ret = bpf_cgroup_storage_set(storage);
111 		if (ret)
112 			break;
113 
114 		if (xdp)
115 			*retval = bpf_prog_run_xdp(prog, ctx);
116 		else
117 			*retval = BPF_PROG_RUN(prog, ctx);
118 
119 		bpf_cgroup_storage_unset();
120 	} while (bpf_test_timer_continue(&t, repeat, &ret, time));
121 	bpf_test_timer_leave(&t);
122 
123 	for_each_cgroup_storage_type(stype)
124 		bpf_cgroup_storage_free(storage[stype]);
125 
126 	return ret;
127 }
128 
129 static int bpf_test_finish(const union bpf_attr *kattr,
130 			   union bpf_attr __user *uattr, const void *data,
131 			   u32 size, u32 retval, u32 duration)
132 {
133 	void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
134 	int err = -EFAULT;
135 	u32 copy_size = size;
136 
137 	/* Clamp copy if the user has provided a size hint, but copy the full
138 	 * buffer if not to retain old behaviour.
139 	 */
140 	if (kattr->test.data_size_out &&
141 	    copy_size > kattr->test.data_size_out) {
142 		copy_size = kattr->test.data_size_out;
143 		err = -ENOSPC;
144 	}
145 
146 	if (data_out && copy_to_user(data_out, data, copy_size))
147 		goto out;
148 	if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
149 		goto out;
150 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
151 		goto out;
152 	if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
153 		goto out;
154 	if (err != -ENOSPC)
155 		err = 0;
156 out:
157 	trace_bpf_test_finish(&err);
158 	return err;
159 }
160 
161 /* Integer types of various sizes and pointer combinations cover variety of
162  * architecture dependent calling conventions. 7+ can be supported in the
163  * future.
164  */
165 __diag_push();
166 __diag_ignore(GCC, 8, "-Wmissing-prototypes",
167 	      "Global functions as their definitions will be in vmlinux BTF");
168 int noinline bpf_fentry_test1(int a)
169 {
170 	return a + 1;
171 }
172 
173 int noinline bpf_fentry_test2(int a, u64 b)
174 {
175 	return a + b;
176 }
177 
178 int noinline bpf_fentry_test3(char a, int b, u64 c)
179 {
180 	return a + b + c;
181 }
182 
183 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
184 {
185 	return (long)a + b + c + d;
186 }
187 
188 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
189 {
190 	return a + (long)b + c + d + e;
191 }
192 
193 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
194 {
195 	return a + (long)b + c + d + (long)e + f;
196 }
197 
198 struct bpf_fentry_test_t {
199 	struct bpf_fentry_test_t *a;
200 };
201 
202 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
203 {
204 	return (long)arg;
205 }
206 
207 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
208 {
209 	return (long)arg->a;
210 }
211 
212 int noinline bpf_modify_return_test(int a, int *b)
213 {
214 	*b += 1;
215 	return a + *b;
216 }
217 
218 u64 noinline bpf_kfunc_call_test1(struct sock *sk, u32 a, u64 b, u32 c, u64 d)
219 {
220 	return a + b + c + d;
221 }
222 
223 int noinline bpf_kfunc_call_test2(struct sock *sk, u32 a, u32 b)
224 {
225 	return a + b;
226 }
227 
228 struct sock * noinline bpf_kfunc_call_test3(struct sock *sk)
229 {
230 	return sk;
231 }
232 
233 __diag_pop();
234 
235 ALLOW_ERROR_INJECTION(bpf_modify_return_test, ERRNO);
236 
237 BTF_SET_START(test_sk_kfunc_ids)
238 BTF_ID(func, bpf_kfunc_call_test1)
239 BTF_ID(func, bpf_kfunc_call_test2)
240 BTF_ID(func, bpf_kfunc_call_test3)
241 BTF_SET_END(test_sk_kfunc_ids)
242 
243 bool bpf_prog_test_check_kfunc_call(u32 kfunc_id)
244 {
245 	return btf_id_set_contains(&test_sk_kfunc_ids, kfunc_id);
246 }
247 
248 static void *bpf_test_init(const union bpf_attr *kattr, u32 size,
249 			   u32 headroom, u32 tailroom)
250 {
251 	void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
252 	u32 user_size = kattr->test.data_size_in;
253 	void *data;
254 
255 	if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
256 		return ERR_PTR(-EINVAL);
257 
258 	if (user_size > size)
259 		return ERR_PTR(-EMSGSIZE);
260 
261 	data = kzalloc(size + headroom + tailroom, GFP_USER);
262 	if (!data)
263 		return ERR_PTR(-ENOMEM);
264 
265 	if (copy_from_user(data + headroom, data_in, user_size)) {
266 		kfree(data);
267 		return ERR_PTR(-EFAULT);
268 	}
269 
270 	return data;
271 }
272 
273 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
274 			      const union bpf_attr *kattr,
275 			      union bpf_attr __user *uattr)
276 {
277 	struct bpf_fentry_test_t arg = {};
278 	u16 side_effect = 0, ret = 0;
279 	int b = 2, err = -EFAULT;
280 	u32 retval = 0;
281 
282 	if (kattr->test.flags || kattr->test.cpu)
283 		return -EINVAL;
284 
285 	switch (prog->expected_attach_type) {
286 	case BPF_TRACE_FENTRY:
287 	case BPF_TRACE_FEXIT:
288 		if (bpf_fentry_test1(1) != 2 ||
289 		    bpf_fentry_test2(2, 3) != 5 ||
290 		    bpf_fentry_test3(4, 5, 6) != 15 ||
291 		    bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
292 		    bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
293 		    bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
294 		    bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
295 		    bpf_fentry_test8(&arg) != 0)
296 			goto out;
297 		break;
298 	case BPF_MODIFY_RETURN:
299 		ret = bpf_modify_return_test(1, &b);
300 		if (b != 2)
301 			side_effect = 1;
302 		break;
303 	default:
304 		goto out;
305 	}
306 
307 	retval = ((u32)side_effect << 16) | ret;
308 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
309 		goto out;
310 
311 	err = 0;
312 out:
313 	trace_bpf_test_finish(&err);
314 	return err;
315 }
316 
317 struct bpf_raw_tp_test_run_info {
318 	struct bpf_prog *prog;
319 	void *ctx;
320 	u32 retval;
321 };
322 
323 static void
324 __bpf_prog_test_run_raw_tp(void *data)
325 {
326 	struct bpf_raw_tp_test_run_info *info = data;
327 
328 	rcu_read_lock();
329 	info->retval = BPF_PROG_RUN(info->prog, info->ctx);
330 	rcu_read_unlock();
331 }
332 
333 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
334 			     const union bpf_attr *kattr,
335 			     union bpf_attr __user *uattr)
336 {
337 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
338 	__u32 ctx_size_in = kattr->test.ctx_size_in;
339 	struct bpf_raw_tp_test_run_info info;
340 	int cpu = kattr->test.cpu, err = 0;
341 	int current_cpu;
342 
343 	/* doesn't support data_in/out, ctx_out, duration, or repeat */
344 	if (kattr->test.data_in || kattr->test.data_out ||
345 	    kattr->test.ctx_out || kattr->test.duration ||
346 	    kattr->test.repeat)
347 		return -EINVAL;
348 
349 	if (ctx_size_in < prog->aux->max_ctx_offset ||
350 	    ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
351 		return -EINVAL;
352 
353 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
354 		return -EINVAL;
355 
356 	if (ctx_size_in) {
357 		info.ctx = kzalloc(ctx_size_in, GFP_USER);
358 		if (!info.ctx)
359 			return -ENOMEM;
360 		if (copy_from_user(info.ctx, ctx_in, ctx_size_in)) {
361 			err = -EFAULT;
362 			goto out;
363 		}
364 	} else {
365 		info.ctx = NULL;
366 	}
367 
368 	info.prog = prog;
369 
370 	current_cpu = get_cpu();
371 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
372 	    cpu == current_cpu) {
373 		__bpf_prog_test_run_raw_tp(&info);
374 	} else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
375 		/* smp_call_function_single() also checks cpu_online()
376 		 * after csd_lock(). However, since cpu is from user
377 		 * space, let's do an extra quick check to filter out
378 		 * invalid value before smp_call_function_single().
379 		 */
380 		err = -ENXIO;
381 	} else {
382 		err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
383 					       &info, 1);
384 	}
385 	put_cpu();
386 
387 	if (!err &&
388 	    copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
389 		err = -EFAULT;
390 
391 out:
392 	kfree(info.ctx);
393 	return err;
394 }
395 
396 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
397 {
398 	void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
399 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
400 	u32 size = kattr->test.ctx_size_in;
401 	void *data;
402 	int err;
403 
404 	if (!data_in && !data_out)
405 		return NULL;
406 
407 	data = kzalloc(max_size, GFP_USER);
408 	if (!data)
409 		return ERR_PTR(-ENOMEM);
410 
411 	if (data_in) {
412 		err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
413 		if (err) {
414 			kfree(data);
415 			return ERR_PTR(err);
416 		}
417 
418 		size = min_t(u32, max_size, size);
419 		if (copy_from_user(data, data_in, size)) {
420 			kfree(data);
421 			return ERR_PTR(-EFAULT);
422 		}
423 	}
424 	return data;
425 }
426 
427 static int bpf_ctx_finish(const union bpf_attr *kattr,
428 			  union bpf_attr __user *uattr, const void *data,
429 			  u32 size)
430 {
431 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
432 	int err = -EFAULT;
433 	u32 copy_size = size;
434 
435 	if (!data || !data_out)
436 		return 0;
437 
438 	if (copy_size > kattr->test.ctx_size_out) {
439 		copy_size = kattr->test.ctx_size_out;
440 		err = -ENOSPC;
441 	}
442 
443 	if (copy_to_user(data_out, data, copy_size))
444 		goto out;
445 	if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
446 		goto out;
447 	if (err != -ENOSPC)
448 		err = 0;
449 out:
450 	return err;
451 }
452 
453 /**
454  * range_is_zero - test whether buffer is initialized
455  * @buf: buffer to check
456  * @from: check from this position
457  * @to: check up until (excluding) this position
458  *
459  * This function returns true if the there is a non-zero byte
460  * in the buf in the range [from,to).
461  */
462 static inline bool range_is_zero(void *buf, size_t from, size_t to)
463 {
464 	return !memchr_inv((u8 *)buf + from, 0, to - from);
465 }
466 
467 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
468 {
469 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
470 
471 	if (!__skb)
472 		return 0;
473 
474 	/* make sure the fields we don't use are zeroed */
475 	if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
476 		return -EINVAL;
477 
478 	/* mark is allowed */
479 
480 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
481 			   offsetof(struct __sk_buff, priority)))
482 		return -EINVAL;
483 
484 	/* priority is allowed */
485 
486 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, priority),
487 			   offsetof(struct __sk_buff, ifindex)))
488 		return -EINVAL;
489 
490 	/* ifindex is allowed */
491 
492 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
493 			   offsetof(struct __sk_buff, cb)))
494 		return -EINVAL;
495 
496 	/* cb is allowed */
497 
498 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
499 			   offsetof(struct __sk_buff, tstamp)))
500 		return -EINVAL;
501 
502 	/* tstamp is allowed */
503 	/* wire_len is allowed */
504 	/* gso_segs is allowed */
505 
506 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
507 			   offsetof(struct __sk_buff, gso_size)))
508 		return -EINVAL;
509 
510 	/* gso_size is allowed */
511 
512 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
513 			   sizeof(struct __sk_buff)))
514 		return -EINVAL;
515 
516 	skb->mark = __skb->mark;
517 	skb->priority = __skb->priority;
518 	skb->tstamp = __skb->tstamp;
519 	memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
520 
521 	if (__skb->wire_len == 0) {
522 		cb->pkt_len = skb->len;
523 	} else {
524 		if (__skb->wire_len < skb->len ||
525 		    __skb->wire_len > GSO_MAX_SIZE)
526 			return -EINVAL;
527 		cb->pkt_len = __skb->wire_len;
528 	}
529 
530 	if (__skb->gso_segs > GSO_MAX_SEGS)
531 		return -EINVAL;
532 	skb_shinfo(skb)->gso_segs = __skb->gso_segs;
533 	skb_shinfo(skb)->gso_size = __skb->gso_size;
534 
535 	return 0;
536 }
537 
538 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
539 {
540 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
541 
542 	if (!__skb)
543 		return;
544 
545 	__skb->mark = skb->mark;
546 	__skb->priority = skb->priority;
547 	__skb->ifindex = skb->dev->ifindex;
548 	__skb->tstamp = skb->tstamp;
549 	memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
550 	__skb->wire_len = cb->pkt_len;
551 	__skb->gso_segs = skb_shinfo(skb)->gso_segs;
552 }
553 
554 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
555 			  union bpf_attr __user *uattr)
556 {
557 	bool is_l2 = false, is_direct_pkt_access = false;
558 	struct net *net = current->nsproxy->net_ns;
559 	struct net_device *dev = net->loopback_dev;
560 	u32 size = kattr->test.data_size_in;
561 	u32 repeat = kattr->test.repeat;
562 	struct __sk_buff *ctx = NULL;
563 	u32 retval, duration;
564 	int hh_len = ETH_HLEN;
565 	struct sk_buff *skb;
566 	struct sock *sk;
567 	void *data;
568 	int ret;
569 
570 	if (kattr->test.flags || kattr->test.cpu)
571 		return -EINVAL;
572 
573 	data = bpf_test_init(kattr, size, NET_SKB_PAD + NET_IP_ALIGN,
574 			     SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
575 	if (IS_ERR(data))
576 		return PTR_ERR(data);
577 
578 	ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
579 	if (IS_ERR(ctx)) {
580 		kfree(data);
581 		return PTR_ERR(ctx);
582 	}
583 
584 	switch (prog->type) {
585 	case BPF_PROG_TYPE_SCHED_CLS:
586 	case BPF_PROG_TYPE_SCHED_ACT:
587 		is_l2 = true;
588 		fallthrough;
589 	case BPF_PROG_TYPE_LWT_IN:
590 	case BPF_PROG_TYPE_LWT_OUT:
591 	case BPF_PROG_TYPE_LWT_XMIT:
592 		is_direct_pkt_access = true;
593 		break;
594 	default:
595 		break;
596 	}
597 
598 	sk = kzalloc(sizeof(struct sock), GFP_USER);
599 	if (!sk) {
600 		kfree(data);
601 		kfree(ctx);
602 		return -ENOMEM;
603 	}
604 	sock_net_set(sk, net);
605 	sock_init_data(NULL, sk);
606 
607 	skb = build_skb(data, 0);
608 	if (!skb) {
609 		kfree(data);
610 		kfree(ctx);
611 		kfree(sk);
612 		return -ENOMEM;
613 	}
614 	skb->sk = sk;
615 
616 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
617 	__skb_put(skb, size);
618 	if (ctx && ctx->ifindex > 1) {
619 		dev = dev_get_by_index(net, ctx->ifindex);
620 		if (!dev) {
621 			ret = -ENODEV;
622 			goto out;
623 		}
624 	}
625 	skb->protocol = eth_type_trans(skb, dev);
626 	skb_reset_network_header(skb);
627 
628 	switch (skb->protocol) {
629 	case htons(ETH_P_IP):
630 		sk->sk_family = AF_INET;
631 		if (sizeof(struct iphdr) <= skb_headlen(skb)) {
632 			sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
633 			sk->sk_daddr = ip_hdr(skb)->daddr;
634 		}
635 		break;
636 #if IS_ENABLED(CONFIG_IPV6)
637 	case htons(ETH_P_IPV6):
638 		sk->sk_family = AF_INET6;
639 		if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
640 			sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
641 			sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
642 		}
643 		break;
644 #endif
645 	default:
646 		break;
647 	}
648 
649 	if (is_l2)
650 		__skb_push(skb, hh_len);
651 	if (is_direct_pkt_access)
652 		bpf_compute_data_pointers(skb);
653 	ret = convert___skb_to_skb(skb, ctx);
654 	if (ret)
655 		goto out;
656 	ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
657 	if (ret)
658 		goto out;
659 	if (!is_l2) {
660 		if (skb_headroom(skb) < hh_len) {
661 			int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
662 
663 			if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
664 				ret = -ENOMEM;
665 				goto out;
666 			}
667 		}
668 		memset(__skb_push(skb, hh_len), 0, hh_len);
669 	}
670 	convert_skb_to___skb(skb, ctx);
671 
672 	size = skb->len;
673 	/* bpf program can never convert linear skb to non-linear */
674 	if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
675 		size = skb_headlen(skb);
676 	ret = bpf_test_finish(kattr, uattr, skb->data, size, retval, duration);
677 	if (!ret)
678 		ret = bpf_ctx_finish(kattr, uattr, ctx,
679 				     sizeof(struct __sk_buff));
680 out:
681 	if (dev && dev != net->loopback_dev)
682 		dev_put(dev);
683 	kfree_skb(skb);
684 	bpf_sk_storage_free(sk);
685 	kfree(sk);
686 	kfree(ctx);
687 	return ret;
688 }
689 
690 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
691 			  union bpf_attr __user *uattr)
692 {
693 	u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
694 	u32 headroom = XDP_PACKET_HEADROOM;
695 	u32 size = kattr->test.data_size_in;
696 	u32 repeat = kattr->test.repeat;
697 	struct netdev_rx_queue *rxqueue;
698 	struct xdp_buff xdp = {};
699 	u32 retval, duration;
700 	u32 max_data_sz;
701 	void *data;
702 	int ret;
703 
704 	if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
705 	    prog->expected_attach_type == BPF_XDP_CPUMAP)
706 		return -EINVAL;
707 	if (kattr->test.ctx_in || kattr->test.ctx_out)
708 		return -EINVAL;
709 
710 	/* XDP have extra tailroom as (most) drivers use full page */
711 	max_data_sz = 4096 - headroom - tailroom;
712 
713 	data = bpf_test_init(kattr, max_data_sz, headroom, tailroom);
714 	if (IS_ERR(data))
715 		return PTR_ERR(data);
716 
717 	rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
718 	xdp_init_buff(&xdp, headroom + max_data_sz + tailroom,
719 		      &rxqueue->xdp_rxq);
720 	xdp_prepare_buff(&xdp, data, headroom, size, true);
721 
722 	bpf_prog_change_xdp(NULL, prog);
723 	ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
724 	if (ret)
725 		goto out;
726 	if (xdp.data != data + headroom || xdp.data_end != xdp.data + size)
727 		size = xdp.data_end - xdp.data;
728 	ret = bpf_test_finish(kattr, uattr, xdp.data, size, retval, duration);
729 out:
730 	bpf_prog_change_xdp(prog, NULL);
731 	kfree(data);
732 	return ret;
733 }
734 
735 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
736 {
737 	/* make sure the fields we don't use are zeroed */
738 	if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
739 		return -EINVAL;
740 
741 	/* flags is allowed */
742 
743 	if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
744 			   sizeof(struct bpf_flow_keys)))
745 		return -EINVAL;
746 
747 	return 0;
748 }
749 
750 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
751 				     const union bpf_attr *kattr,
752 				     union bpf_attr __user *uattr)
753 {
754 	struct bpf_test_timer t = { NO_PREEMPT };
755 	u32 size = kattr->test.data_size_in;
756 	struct bpf_flow_dissector ctx = {};
757 	u32 repeat = kattr->test.repeat;
758 	struct bpf_flow_keys *user_ctx;
759 	struct bpf_flow_keys flow_keys;
760 	const struct ethhdr *eth;
761 	unsigned int flags = 0;
762 	u32 retval, duration;
763 	void *data;
764 	int ret;
765 
766 	if (prog->type != BPF_PROG_TYPE_FLOW_DISSECTOR)
767 		return -EINVAL;
768 
769 	if (kattr->test.flags || kattr->test.cpu)
770 		return -EINVAL;
771 
772 	if (size < ETH_HLEN)
773 		return -EINVAL;
774 
775 	data = bpf_test_init(kattr, size, 0, 0);
776 	if (IS_ERR(data))
777 		return PTR_ERR(data);
778 
779 	eth = (struct ethhdr *)data;
780 
781 	if (!repeat)
782 		repeat = 1;
783 
784 	user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
785 	if (IS_ERR(user_ctx)) {
786 		kfree(data);
787 		return PTR_ERR(user_ctx);
788 	}
789 	if (user_ctx) {
790 		ret = verify_user_bpf_flow_keys(user_ctx);
791 		if (ret)
792 			goto out;
793 		flags = user_ctx->flags;
794 	}
795 
796 	ctx.flow_keys = &flow_keys;
797 	ctx.data = data;
798 	ctx.data_end = (__u8 *)data + size;
799 
800 	bpf_test_timer_enter(&t);
801 	do {
802 		retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
803 					  size, flags);
804 	} while (bpf_test_timer_continue(&t, repeat, &ret, &duration));
805 	bpf_test_timer_leave(&t);
806 
807 	if (ret < 0)
808 		goto out;
809 
810 	ret = bpf_test_finish(kattr, uattr, &flow_keys, sizeof(flow_keys),
811 			      retval, duration);
812 	if (!ret)
813 		ret = bpf_ctx_finish(kattr, uattr, user_ctx,
814 				     sizeof(struct bpf_flow_keys));
815 
816 out:
817 	kfree(user_ctx);
818 	kfree(data);
819 	return ret;
820 }
821 
822 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
823 				union bpf_attr __user *uattr)
824 {
825 	struct bpf_test_timer t = { NO_PREEMPT };
826 	struct bpf_prog_array *progs = NULL;
827 	struct bpf_sk_lookup_kern ctx = {};
828 	u32 repeat = kattr->test.repeat;
829 	struct bpf_sk_lookup *user_ctx;
830 	u32 retval, duration;
831 	int ret = -EINVAL;
832 
833 	if (prog->type != BPF_PROG_TYPE_SK_LOOKUP)
834 		return -EINVAL;
835 
836 	if (kattr->test.flags || kattr->test.cpu)
837 		return -EINVAL;
838 
839 	if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
840 	    kattr->test.data_size_out)
841 		return -EINVAL;
842 
843 	if (!repeat)
844 		repeat = 1;
845 
846 	user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
847 	if (IS_ERR(user_ctx))
848 		return PTR_ERR(user_ctx);
849 
850 	if (!user_ctx)
851 		return -EINVAL;
852 
853 	if (user_ctx->sk)
854 		goto out;
855 
856 	if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
857 		goto out;
858 
859 	if (user_ctx->local_port > U16_MAX || user_ctx->remote_port > U16_MAX) {
860 		ret = -ERANGE;
861 		goto out;
862 	}
863 
864 	ctx.family = (u16)user_ctx->family;
865 	ctx.protocol = (u16)user_ctx->protocol;
866 	ctx.dport = (u16)user_ctx->local_port;
867 	ctx.sport = (__force __be16)user_ctx->remote_port;
868 
869 	switch (ctx.family) {
870 	case AF_INET:
871 		ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
872 		ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
873 		break;
874 
875 #if IS_ENABLED(CONFIG_IPV6)
876 	case AF_INET6:
877 		ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
878 		ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
879 		break;
880 #endif
881 
882 	default:
883 		ret = -EAFNOSUPPORT;
884 		goto out;
885 	}
886 
887 	progs = bpf_prog_array_alloc(1, GFP_KERNEL);
888 	if (!progs) {
889 		ret = -ENOMEM;
890 		goto out;
891 	}
892 
893 	progs->items[0].prog = prog;
894 
895 	bpf_test_timer_enter(&t);
896 	do {
897 		ctx.selected_sk = NULL;
898 		retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, BPF_PROG_RUN);
899 	} while (bpf_test_timer_continue(&t, repeat, &ret, &duration));
900 	bpf_test_timer_leave(&t);
901 
902 	if (ret < 0)
903 		goto out;
904 
905 	user_ctx->cookie = 0;
906 	if (ctx.selected_sk) {
907 		if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
908 			ret = -EOPNOTSUPP;
909 			goto out;
910 		}
911 
912 		user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
913 	}
914 
915 	ret = bpf_test_finish(kattr, uattr, NULL, 0, retval, duration);
916 	if (!ret)
917 		ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
918 
919 out:
920 	bpf_prog_array_free(progs);
921 	kfree(user_ctx);
922 	return ret;
923 }
924 
925 int bpf_prog_test_run_syscall(struct bpf_prog *prog,
926 			      const union bpf_attr *kattr,
927 			      union bpf_attr __user *uattr)
928 {
929 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
930 	__u32 ctx_size_in = kattr->test.ctx_size_in;
931 	void *ctx = NULL;
932 	u32 retval;
933 	int err = 0;
934 
935 	/* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
936 	if (kattr->test.data_in || kattr->test.data_out ||
937 	    kattr->test.ctx_out || kattr->test.duration ||
938 	    kattr->test.repeat || kattr->test.flags)
939 		return -EINVAL;
940 
941 	if (ctx_size_in < prog->aux->max_ctx_offset ||
942 	    ctx_size_in > U16_MAX)
943 		return -EINVAL;
944 
945 	if (ctx_size_in) {
946 		ctx = kzalloc(ctx_size_in, GFP_USER);
947 		if (!ctx)
948 			return -ENOMEM;
949 		if (copy_from_user(ctx, ctx_in, ctx_size_in)) {
950 			err = -EFAULT;
951 			goto out;
952 		}
953 	}
954 	retval = bpf_prog_run_pin_on_cpu(prog, ctx);
955 
956 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
957 		err = -EFAULT;
958 		goto out;
959 	}
960 	if (ctx_size_in)
961 		if (copy_to_user(ctx_in, ctx, ctx_size_in))
962 			err = -EFAULT;
963 out:
964 	kfree(ctx);
965 	return err;
966 }
967