xref: /openbmc/linux/net/bpf/test_run.c (revision dc6a81c3)
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 
14 #define CREATE_TRACE_POINTS
15 #include <trace/events/bpf_test_run.h>
16 
17 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
18 			u32 *retval, u32 *time, bool xdp)
19 {
20 	struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE] = { NULL };
21 	enum bpf_cgroup_storage_type stype;
22 	u64 time_start, time_spent = 0;
23 	int ret = 0;
24 	u32 i;
25 
26 	for_each_cgroup_storage_type(stype) {
27 		storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
28 		if (IS_ERR(storage[stype])) {
29 			storage[stype] = NULL;
30 			for_each_cgroup_storage_type(stype)
31 				bpf_cgroup_storage_free(storage[stype]);
32 			return -ENOMEM;
33 		}
34 	}
35 
36 	if (!repeat)
37 		repeat = 1;
38 
39 	rcu_read_lock();
40 	preempt_disable();
41 	time_start = ktime_get_ns();
42 	for (i = 0; i < repeat; i++) {
43 		bpf_cgroup_storage_set(storage);
44 
45 		if (xdp)
46 			*retval = bpf_prog_run_xdp(prog, ctx);
47 		else
48 			*retval = BPF_PROG_RUN(prog, ctx);
49 
50 		if (signal_pending(current)) {
51 			ret = -EINTR;
52 			break;
53 		}
54 
55 		if (need_resched()) {
56 			time_spent += ktime_get_ns() - time_start;
57 			preempt_enable();
58 			rcu_read_unlock();
59 
60 			cond_resched();
61 
62 			rcu_read_lock();
63 			preempt_disable();
64 			time_start = ktime_get_ns();
65 		}
66 	}
67 	time_spent += ktime_get_ns() - time_start;
68 	preempt_enable();
69 	rcu_read_unlock();
70 
71 	do_div(time_spent, repeat);
72 	*time = time_spent > U32_MAX ? U32_MAX : (u32)time_spent;
73 
74 	for_each_cgroup_storage_type(stype)
75 		bpf_cgroup_storage_free(storage[stype]);
76 
77 	return ret;
78 }
79 
80 static int bpf_test_finish(const union bpf_attr *kattr,
81 			   union bpf_attr __user *uattr, const void *data,
82 			   u32 size, u32 retval, u32 duration)
83 {
84 	void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
85 	int err = -EFAULT;
86 	u32 copy_size = size;
87 
88 	/* Clamp copy if the user has provided a size hint, but copy the full
89 	 * buffer if not to retain old behaviour.
90 	 */
91 	if (kattr->test.data_size_out &&
92 	    copy_size > kattr->test.data_size_out) {
93 		copy_size = kattr->test.data_size_out;
94 		err = -ENOSPC;
95 	}
96 
97 	if (data_out && copy_to_user(data_out, data, copy_size))
98 		goto out;
99 	if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
100 		goto out;
101 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
102 		goto out;
103 	if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
104 		goto out;
105 	if (err != -ENOSPC)
106 		err = 0;
107 out:
108 	trace_bpf_test_finish(&err);
109 	return err;
110 }
111 
112 /* Integer types of various sizes and pointer combinations cover variety of
113  * architecture dependent calling conventions. 7+ can be supported in the
114  * future.
115  */
116 int noinline bpf_fentry_test1(int a)
117 {
118 	return a + 1;
119 }
120 
121 int noinline bpf_fentry_test2(int a, u64 b)
122 {
123 	return a + b;
124 }
125 
126 int noinline bpf_fentry_test3(char a, int b, u64 c)
127 {
128 	return a + b + c;
129 }
130 
131 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
132 {
133 	return (long)a + b + c + d;
134 }
135 
136 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
137 {
138 	return a + (long)b + c + d + e;
139 }
140 
141 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
142 {
143 	return a + (long)b + c + d + (long)e + f;
144 }
145 
146 static void *bpf_test_init(const union bpf_attr *kattr, u32 size,
147 			   u32 headroom, u32 tailroom)
148 {
149 	void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
150 	void *data;
151 
152 	if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
153 		return ERR_PTR(-EINVAL);
154 
155 	data = kzalloc(size + headroom + tailroom, GFP_USER);
156 	if (!data)
157 		return ERR_PTR(-ENOMEM);
158 
159 	if (copy_from_user(data + headroom, data_in, size)) {
160 		kfree(data);
161 		return ERR_PTR(-EFAULT);
162 	}
163 	if (bpf_fentry_test1(1) != 2 ||
164 	    bpf_fentry_test2(2, 3) != 5 ||
165 	    bpf_fentry_test3(4, 5, 6) != 15 ||
166 	    bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
167 	    bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
168 	    bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111) {
169 		kfree(data);
170 		return ERR_PTR(-EFAULT);
171 	}
172 	return data;
173 }
174 
175 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
176 {
177 	void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
178 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
179 	u32 size = kattr->test.ctx_size_in;
180 	void *data;
181 	int err;
182 
183 	if (!data_in && !data_out)
184 		return NULL;
185 
186 	data = kzalloc(max_size, GFP_USER);
187 	if (!data)
188 		return ERR_PTR(-ENOMEM);
189 
190 	if (data_in) {
191 		err = bpf_check_uarg_tail_zero(data_in, max_size, size);
192 		if (err) {
193 			kfree(data);
194 			return ERR_PTR(err);
195 		}
196 
197 		size = min_t(u32, max_size, size);
198 		if (copy_from_user(data, data_in, size)) {
199 			kfree(data);
200 			return ERR_PTR(-EFAULT);
201 		}
202 	}
203 	return data;
204 }
205 
206 static int bpf_ctx_finish(const union bpf_attr *kattr,
207 			  union bpf_attr __user *uattr, const void *data,
208 			  u32 size)
209 {
210 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
211 	int err = -EFAULT;
212 	u32 copy_size = size;
213 
214 	if (!data || !data_out)
215 		return 0;
216 
217 	if (copy_size > kattr->test.ctx_size_out) {
218 		copy_size = kattr->test.ctx_size_out;
219 		err = -ENOSPC;
220 	}
221 
222 	if (copy_to_user(data_out, data, copy_size))
223 		goto out;
224 	if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
225 		goto out;
226 	if (err != -ENOSPC)
227 		err = 0;
228 out:
229 	return err;
230 }
231 
232 /**
233  * range_is_zero - test whether buffer is initialized
234  * @buf: buffer to check
235  * @from: check from this position
236  * @to: check up until (excluding) this position
237  *
238  * This function returns true if the there is a non-zero byte
239  * in the buf in the range [from,to).
240  */
241 static inline bool range_is_zero(void *buf, size_t from, size_t to)
242 {
243 	return !memchr_inv((u8 *)buf + from, 0, to - from);
244 }
245 
246 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
247 {
248 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
249 
250 	if (!__skb)
251 		return 0;
252 
253 	/* make sure the fields we don't use are zeroed */
254 	if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
255 		return -EINVAL;
256 
257 	/* mark is allowed */
258 
259 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
260 			   offsetof(struct __sk_buff, priority)))
261 		return -EINVAL;
262 
263 	/* priority is allowed */
264 
265 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, priority),
266 			   offsetof(struct __sk_buff, cb)))
267 		return -EINVAL;
268 
269 	/* cb is allowed */
270 
271 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
272 			   offsetof(struct __sk_buff, tstamp)))
273 		return -EINVAL;
274 
275 	/* tstamp is allowed */
276 	/* wire_len is allowed */
277 	/* gso_segs is allowed */
278 
279 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
280 			   sizeof(struct __sk_buff)))
281 		return -EINVAL;
282 
283 	skb->mark = __skb->mark;
284 	skb->priority = __skb->priority;
285 	skb->tstamp = __skb->tstamp;
286 	memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
287 
288 	if (__skb->wire_len == 0) {
289 		cb->pkt_len = skb->len;
290 	} else {
291 		if (__skb->wire_len < skb->len ||
292 		    __skb->wire_len > GSO_MAX_SIZE)
293 			return -EINVAL;
294 		cb->pkt_len = __skb->wire_len;
295 	}
296 
297 	if (__skb->gso_segs > GSO_MAX_SEGS)
298 		return -EINVAL;
299 	skb_shinfo(skb)->gso_segs = __skb->gso_segs;
300 
301 	return 0;
302 }
303 
304 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
305 {
306 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
307 
308 	if (!__skb)
309 		return;
310 
311 	__skb->mark = skb->mark;
312 	__skb->priority = skb->priority;
313 	__skb->tstamp = skb->tstamp;
314 	memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
315 	__skb->wire_len = cb->pkt_len;
316 	__skb->gso_segs = skb_shinfo(skb)->gso_segs;
317 }
318 
319 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
320 			  union bpf_attr __user *uattr)
321 {
322 	bool is_l2 = false, is_direct_pkt_access = false;
323 	u32 size = kattr->test.data_size_in;
324 	u32 repeat = kattr->test.repeat;
325 	struct __sk_buff *ctx = NULL;
326 	u32 retval, duration;
327 	int hh_len = ETH_HLEN;
328 	struct sk_buff *skb;
329 	struct sock *sk;
330 	void *data;
331 	int ret;
332 
333 	data = bpf_test_init(kattr, size, NET_SKB_PAD + NET_IP_ALIGN,
334 			     SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
335 	if (IS_ERR(data))
336 		return PTR_ERR(data);
337 
338 	ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
339 	if (IS_ERR(ctx)) {
340 		kfree(data);
341 		return PTR_ERR(ctx);
342 	}
343 
344 	switch (prog->type) {
345 	case BPF_PROG_TYPE_SCHED_CLS:
346 	case BPF_PROG_TYPE_SCHED_ACT:
347 		is_l2 = true;
348 		/* fall through */
349 	case BPF_PROG_TYPE_LWT_IN:
350 	case BPF_PROG_TYPE_LWT_OUT:
351 	case BPF_PROG_TYPE_LWT_XMIT:
352 		is_direct_pkt_access = true;
353 		break;
354 	default:
355 		break;
356 	}
357 
358 	sk = kzalloc(sizeof(struct sock), GFP_USER);
359 	if (!sk) {
360 		kfree(data);
361 		kfree(ctx);
362 		return -ENOMEM;
363 	}
364 	sock_net_set(sk, current->nsproxy->net_ns);
365 	sock_init_data(NULL, sk);
366 
367 	skb = build_skb(data, 0);
368 	if (!skb) {
369 		kfree(data);
370 		kfree(ctx);
371 		kfree(sk);
372 		return -ENOMEM;
373 	}
374 	skb->sk = sk;
375 
376 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
377 	__skb_put(skb, size);
378 	skb->protocol = eth_type_trans(skb, current->nsproxy->net_ns->loopback_dev);
379 	skb_reset_network_header(skb);
380 
381 	if (is_l2)
382 		__skb_push(skb, hh_len);
383 	if (is_direct_pkt_access)
384 		bpf_compute_data_pointers(skb);
385 	ret = convert___skb_to_skb(skb, ctx);
386 	if (ret)
387 		goto out;
388 	ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
389 	if (ret)
390 		goto out;
391 	if (!is_l2) {
392 		if (skb_headroom(skb) < hh_len) {
393 			int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
394 
395 			if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
396 				ret = -ENOMEM;
397 				goto out;
398 			}
399 		}
400 		memset(__skb_push(skb, hh_len), 0, hh_len);
401 	}
402 	convert_skb_to___skb(skb, ctx);
403 
404 	size = skb->len;
405 	/* bpf program can never convert linear skb to non-linear */
406 	if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
407 		size = skb_headlen(skb);
408 	ret = bpf_test_finish(kattr, uattr, skb->data, size, retval, duration);
409 	if (!ret)
410 		ret = bpf_ctx_finish(kattr, uattr, ctx,
411 				     sizeof(struct __sk_buff));
412 out:
413 	kfree_skb(skb);
414 	bpf_sk_storage_free(sk);
415 	kfree(sk);
416 	kfree(ctx);
417 	return ret;
418 }
419 
420 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
421 			  union bpf_attr __user *uattr)
422 {
423 	u32 size = kattr->test.data_size_in;
424 	u32 repeat = kattr->test.repeat;
425 	struct netdev_rx_queue *rxqueue;
426 	struct xdp_buff xdp = {};
427 	u32 retval, duration;
428 	void *data;
429 	int ret;
430 
431 	if (kattr->test.ctx_in || kattr->test.ctx_out)
432 		return -EINVAL;
433 
434 	data = bpf_test_init(kattr, size, XDP_PACKET_HEADROOM + NET_IP_ALIGN, 0);
435 	if (IS_ERR(data))
436 		return PTR_ERR(data);
437 
438 	xdp.data_hard_start = data;
439 	xdp.data = data + XDP_PACKET_HEADROOM + NET_IP_ALIGN;
440 	xdp.data_meta = xdp.data;
441 	xdp.data_end = xdp.data + size;
442 
443 	rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
444 	xdp.rxq = &rxqueue->xdp_rxq;
445 	bpf_prog_change_xdp(NULL, prog);
446 	ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
447 	if (ret)
448 		goto out;
449 	if (xdp.data != data + XDP_PACKET_HEADROOM + NET_IP_ALIGN ||
450 	    xdp.data_end != xdp.data + size)
451 		size = xdp.data_end - xdp.data;
452 	ret = bpf_test_finish(kattr, uattr, xdp.data, size, retval, duration);
453 out:
454 	bpf_prog_change_xdp(prog, NULL);
455 	kfree(data);
456 	return ret;
457 }
458 
459 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
460 {
461 	/* make sure the fields we don't use are zeroed */
462 	if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
463 		return -EINVAL;
464 
465 	/* flags is allowed */
466 
467 	if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
468 			   sizeof(struct bpf_flow_keys)))
469 		return -EINVAL;
470 
471 	return 0;
472 }
473 
474 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
475 				     const union bpf_attr *kattr,
476 				     union bpf_attr __user *uattr)
477 {
478 	u32 size = kattr->test.data_size_in;
479 	struct bpf_flow_dissector ctx = {};
480 	u32 repeat = kattr->test.repeat;
481 	struct bpf_flow_keys *user_ctx;
482 	struct bpf_flow_keys flow_keys;
483 	u64 time_start, time_spent = 0;
484 	const struct ethhdr *eth;
485 	unsigned int flags = 0;
486 	u32 retval, duration;
487 	void *data;
488 	int ret;
489 	u32 i;
490 
491 	if (prog->type != BPF_PROG_TYPE_FLOW_DISSECTOR)
492 		return -EINVAL;
493 
494 	if (size < ETH_HLEN)
495 		return -EINVAL;
496 
497 	data = bpf_test_init(kattr, size, 0, 0);
498 	if (IS_ERR(data))
499 		return PTR_ERR(data);
500 
501 	eth = (struct ethhdr *)data;
502 
503 	if (!repeat)
504 		repeat = 1;
505 
506 	user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
507 	if (IS_ERR(user_ctx)) {
508 		kfree(data);
509 		return PTR_ERR(user_ctx);
510 	}
511 	if (user_ctx) {
512 		ret = verify_user_bpf_flow_keys(user_ctx);
513 		if (ret)
514 			goto out;
515 		flags = user_ctx->flags;
516 	}
517 
518 	ctx.flow_keys = &flow_keys;
519 	ctx.data = data;
520 	ctx.data_end = (__u8 *)data + size;
521 
522 	rcu_read_lock();
523 	preempt_disable();
524 	time_start = ktime_get_ns();
525 	for (i = 0; i < repeat; i++) {
526 		retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
527 					  size, flags);
528 
529 		if (signal_pending(current)) {
530 			preempt_enable();
531 			rcu_read_unlock();
532 
533 			ret = -EINTR;
534 			goto out;
535 		}
536 
537 		if (need_resched()) {
538 			time_spent += ktime_get_ns() - time_start;
539 			preempt_enable();
540 			rcu_read_unlock();
541 
542 			cond_resched();
543 
544 			rcu_read_lock();
545 			preempt_disable();
546 			time_start = ktime_get_ns();
547 		}
548 	}
549 	time_spent += ktime_get_ns() - time_start;
550 	preempt_enable();
551 	rcu_read_unlock();
552 
553 	do_div(time_spent, repeat);
554 	duration = time_spent > U32_MAX ? U32_MAX : (u32)time_spent;
555 
556 	ret = bpf_test_finish(kattr, uattr, &flow_keys, sizeof(flow_keys),
557 			      retval, duration);
558 	if (!ret)
559 		ret = bpf_ctx_finish(kattr, uattr, user_ctx,
560 				     sizeof(struct bpf_flow_keys));
561 
562 out:
563 	kfree(user_ctx);
564 	kfree(data);
565 	return ret;
566 }
567