xref: /linux/net/bpf/test_run.c (revision b590cb5f802dc20c72f507f7fbe6737222d0afba)
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)
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 		*retval = BPF_PROG_RUN(prog, ctx);
45 
46 		if (signal_pending(current)) {
47 			ret = -EINTR;
48 			break;
49 		}
50 
51 		if (need_resched()) {
52 			time_spent += ktime_get_ns() - time_start;
53 			preempt_enable();
54 			rcu_read_unlock();
55 
56 			cond_resched();
57 
58 			rcu_read_lock();
59 			preempt_disable();
60 			time_start = ktime_get_ns();
61 		}
62 	}
63 	time_spent += ktime_get_ns() - time_start;
64 	preempt_enable();
65 	rcu_read_unlock();
66 
67 	do_div(time_spent, repeat);
68 	*time = time_spent > U32_MAX ? U32_MAX : (u32)time_spent;
69 
70 	for_each_cgroup_storage_type(stype)
71 		bpf_cgroup_storage_free(storage[stype]);
72 
73 	return ret;
74 }
75 
76 static int bpf_test_finish(const union bpf_attr *kattr,
77 			   union bpf_attr __user *uattr, const void *data,
78 			   u32 size, u32 retval, u32 duration)
79 {
80 	void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
81 	int err = -EFAULT;
82 	u32 copy_size = size;
83 
84 	/* Clamp copy if the user has provided a size hint, but copy the full
85 	 * buffer if not to retain old behaviour.
86 	 */
87 	if (kattr->test.data_size_out &&
88 	    copy_size > kattr->test.data_size_out) {
89 		copy_size = kattr->test.data_size_out;
90 		err = -ENOSPC;
91 	}
92 
93 	if (data_out && copy_to_user(data_out, data, copy_size))
94 		goto out;
95 	if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
96 		goto out;
97 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
98 		goto out;
99 	if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
100 		goto out;
101 	if (err != -ENOSPC)
102 		err = 0;
103 out:
104 	trace_bpf_test_finish(&err);
105 	return err;
106 }
107 
108 /* Integer types of various sizes and pointer combinations cover variety of
109  * architecture dependent calling conventions. 7+ can be supported in the
110  * future.
111  */
112 int noinline bpf_fentry_test1(int a)
113 {
114 	return a + 1;
115 }
116 
117 int noinline bpf_fentry_test2(int a, u64 b)
118 {
119 	return a + b;
120 }
121 
122 int noinline bpf_fentry_test3(char a, int b, u64 c)
123 {
124 	return a + b + c;
125 }
126 
127 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
128 {
129 	return (long)a + b + c + d;
130 }
131 
132 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
133 {
134 	return a + (long)b + c + d + e;
135 }
136 
137 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
138 {
139 	return a + (long)b + c + d + (long)e + f;
140 }
141 
142 static void *bpf_test_init(const union bpf_attr *kattr, u32 size,
143 			   u32 headroom, u32 tailroom)
144 {
145 	void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
146 	void *data;
147 
148 	if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
149 		return ERR_PTR(-EINVAL);
150 
151 	data = kzalloc(size + headroom + tailroom, GFP_USER);
152 	if (!data)
153 		return ERR_PTR(-ENOMEM);
154 
155 	if (copy_from_user(data + headroom, data_in, size)) {
156 		kfree(data);
157 		return ERR_PTR(-EFAULT);
158 	}
159 	if (bpf_fentry_test1(1) != 2 ||
160 	    bpf_fentry_test2(2, 3) != 5 ||
161 	    bpf_fentry_test3(4, 5, 6) != 15 ||
162 	    bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
163 	    bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
164 	    bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111) {
165 		kfree(data);
166 		return ERR_PTR(-EFAULT);
167 	}
168 	return data;
169 }
170 
171 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
172 {
173 	void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
174 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
175 	u32 size = kattr->test.ctx_size_in;
176 	void *data;
177 	int err;
178 
179 	if (!data_in && !data_out)
180 		return NULL;
181 
182 	data = kzalloc(max_size, GFP_USER);
183 	if (!data)
184 		return ERR_PTR(-ENOMEM);
185 
186 	if (data_in) {
187 		err = bpf_check_uarg_tail_zero(data_in, max_size, size);
188 		if (err) {
189 			kfree(data);
190 			return ERR_PTR(err);
191 		}
192 
193 		size = min_t(u32, max_size, size);
194 		if (copy_from_user(data, data_in, size)) {
195 			kfree(data);
196 			return ERR_PTR(-EFAULT);
197 		}
198 	}
199 	return data;
200 }
201 
202 static int bpf_ctx_finish(const union bpf_attr *kattr,
203 			  union bpf_attr __user *uattr, const void *data,
204 			  u32 size)
205 {
206 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
207 	int err = -EFAULT;
208 	u32 copy_size = size;
209 
210 	if (!data || !data_out)
211 		return 0;
212 
213 	if (copy_size > kattr->test.ctx_size_out) {
214 		copy_size = kattr->test.ctx_size_out;
215 		err = -ENOSPC;
216 	}
217 
218 	if (copy_to_user(data_out, data, copy_size))
219 		goto out;
220 	if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
221 		goto out;
222 	if (err != -ENOSPC)
223 		err = 0;
224 out:
225 	return err;
226 }
227 
228 /**
229  * range_is_zero - test whether buffer is initialized
230  * @buf: buffer to check
231  * @from: check from this position
232  * @to: check up until (excluding) this position
233  *
234  * This function returns true if the there is a non-zero byte
235  * in the buf in the range [from,to).
236  */
237 static inline bool range_is_zero(void *buf, size_t from, size_t to)
238 {
239 	return !memchr_inv((u8 *)buf + from, 0, to - from);
240 }
241 
242 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
243 {
244 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
245 
246 	if (!__skb)
247 		return 0;
248 
249 	/* make sure the fields we don't use are zeroed */
250 	if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, priority)))
251 		return -EINVAL;
252 
253 	/* priority is allowed */
254 
255 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, priority),
256 			   offsetof(struct __sk_buff, cb)))
257 		return -EINVAL;
258 
259 	/* cb is allowed */
260 
261 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
262 			   offsetof(struct __sk_buff, tstamp)))
263 		return -EINVAL;
264 
265 	/* tstamp is allowed */
266 
267 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, tstamp),
268 			   sizeof(struct __sk_buff)))
269 		return -EINVAL;
270 
271 	skb->priority = __skb->priority;
272 	skb->tstamp = __skb->tstamp;
273 	memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
274 
275 	return 0;
276 }
277 
278 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
279 {
280 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
281 
282 	if (!__skb)
283 		return;
284 
285 	__skb->priority = skb->priority;
286 	__skb->tstamp = skb->tstamp;
287 	memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
288 }
289 
290 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
291 			  union bpf_attr __user *uattr)
292 {
293 	bool is_l2 = false, is_direct_pkt_access = false;
294 	u32 size = kattr->test.data_size_in;
295 	u32 repeat = kattr->test.repeat;
296 	struct __sk_buff *ctx = NULL;
297 	u32 retval, duration;
298 	int hh_len = ETH_HLEN;
299 	struct sk_buff *skb;
300 	struct sock *sk;
301 	void *data;
302 	int ret;
303 
304 	data = bpf_test_init(kattr, size, NET_SKB_PAD + NET_IP_ALIGN,
305 			     SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
306 	if (IS_ERR(data))
307 		return PTR_ERR(data);
308 
309 	ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
310 	if (IS_ERR(ctx)) {
311 		kfree(data);
312 		return PTR_ERR(ctx);
313 	}
314 
315 	switch (prog->type) {
316 	case BPF_PROG_TYPE_SCHED_CLS:
317 	case BPF_PROG_TYPE_SCHED_ACT:
318 		is_l2 = true;
319 		/* fall through */
320 	case BPF_PROG_TYPE_LWT_IN:
321 	case BPF_PROG_TYPE_LWT_OUT:
322 	case BPF_PROG_TYPE_LWT_XMIT:
323 		is_direct_pkt_access = true;
324 		break;
325 	default:
326 		break;
327 	}
328 
329 	sk = kzalloc(sizeof(struct sock), GFP_USER);
330 	if (!sk) {
331 		kfree(data);
332 		kfree(ctx);
333 		return -ENOMEM;
334 	}
335 	sock_net_set(sk, current->nsproxy->net_ns);
336 	sock_init_data(NULL, sk);
337 
338 	skb = build_skb(data, 0);
339 	if (!skb) {
340 		kfree(data);
341 		kfree(ctx);
342 		kfree(sk);
343 		return -ENOMEM;
344 	}
345 	skb->sk = sk;
346 
347 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
348 	__skb_put(skb, size);
349 	skb->protocol = eth_type_trans(skb, current->nsproxy->net_ns->loopback_dev);
350 	skb_reset_network_header(skb);
351 
352 	if (is_l2)
353 		__skb_push(skb, hh_len);
354 	if (is_direct_pkt_access)
355 		bpf_compute_data_pointers(skb);
356 	ret = convert___skb_to_skb(skb, ctx);
357 	if (ret)
358 		goto out;
359 	ret = bpf_test_run(prog, skb, repeat, &retval, &duration);
360 	if (ret)
361 		goto out;
362 	if (!is_l2) {
363 		if (skb_headroom(skb) < hh_len) {
364 			int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
365 
366 			if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
367 				ret = -ENOMEM;
368 				goto out;
369 			}
370 		}
371 		memset(__skb_push(skb, hh_len), 0, hh_len);
372 	}
373 	convert_skb_to___skb(skb, ctx);
374 
375 	size = skb->len;
376 	/* bpf program can never convert linear skb to non-linear */
377 	if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
378 		size = skb_headlen(skb);
379 	ret = bpf_test_finish(kattr, uattr, skb->data, size, retval, duration);
380 	if (!ret)
381 		ret = bpf_ctx_finish(kattr, uattr, ctx,
382 				     sizeof(struct __sk_buff));
383 out:
384 	kfree_skb(skb);
385 	bpf_sk_storage_free(sk);
386 	kfree(sk);
387 	kfree(ctx);
388 	return ret;
389 }
390 
391 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
392 			  union bpf_attr __user *uattr)
393 {
394 	u32 size = kattr->test.data_size_in;
395 	u32 repeat = kattr->test.repeat;
396 	struct netdev_rx_queue *rxqueue;
397 	struct xdp_buff xdp = {};
398 	u32 retval, duration;
399 	void *data;
400 	int ret;
401 
402 	if (kattr->test.ctx_in || kattr->test.ctx_out)
403 		return -EINVAL;
404 
405 	data = bpf_test_init(kattr, size, XDP_PACKET_HEADROOM + NET_IP_ALIGN, 0);
406 	if (IS_ERR(data))
407 		return PTR_ERR(data);
408 
409 	xdp.data_hard_start = data;
410 	xdp.data = data + XDP_PACKET_HEADROOM + NET_IP_ALIGN;
411 	xdp.data_meta = xdp.data;
412 	xdp.data_end = xdp.data + size;
413 
414 	rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
415 	xdp.rxq = &rxqueue->xdp_rxq;
416 
417 	ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration);
418 	if (ret)
419 		goto out;
420 	if (xdp.data != data + XDP_PACKET_HEADROOM + NET_IP_ALIGN ||
421 	    xdp.data_end != xdp.data + size)
422 		size = xdp.data_end - xdp.data;
423 	ret = bpf_test_finish(kattr, uattr, xdp.data, size, retval, duration);
424 out:
425 	kfree(data);
426 	return ret;
427 }
428 
429 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
430 {
431 	/* make sure the fields we don't use are zeroed */
432 	if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
433 		return -EINVAL;
434 
435 	/* flags is allowed */
436 
437 	if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
438 			   sizeof(struct bpf_flow_keys)))
439 		return -EINVAL;
440 
441 	return 0;
442 }
443 
444 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
445 				     const union bpf_attr *kattr,
446 				     union bpf_attr __user *uattr)
447 {
448 	u32 size = kattr->test.data_size_in;
449 	struct bpf_flow_dissector ctx = {};
450 	u32 repeat = kattr->test.repeat;
451 	struct bpf_flow_keys *user_ctx;
452 	struct bpf_flow_keys flow_keys;
453 	u64 time_start, time_spent = 0;
454 	const struct ethhdr *eth;
455 	unsigned int flags = 0;
456 	u32 retval, duration;
457 	void *data;
458 	int ret;
459 	u32 i;
460 
461 	if (prog->type != BPF_PROG_TYPE_FLOW_DISSECTOR)
462 		return -EINVAL;
463 
464 	if (size < ETH_HLEN)
465 		return -EINVAL;
466 
467 	data = bpf_test_init(kattr, size, 0, 0);
468 	if (IS_ERR(data))
469 		return PTR_ERR(data);
470 
471 	eth = (struct ethhdr *)data;
472 
473 	if (!repeat)
474 		repeat = 1;
475 
476 	user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
477 	if (IS_ERR(user_ctx)) {
478 		kfree(data);
479 		return PTR_ERR(user_ctx);
480 	}
481 	if (user_ctx) {
482 		ret = verify_user_bpf_flow_keys(user_ctx);
483 		if (ret)
484 			goto out;
485 		flags = user_ctx->flags;
486 	}
487 
488 	ctx.flow_keys = &flow_keys;
489 	ctx.data = data;
490 	ctx.data_end = (__u8 *)data + size;
491 
492 	rcu_read_lock();
493 	preempt_disable();
494 	time_start = ktime_get_ns();
495 	for (i = 0; i < repeat; i++) {
496 		retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
497 					  size, flags);
498 
499 		if (signal_pending(current)) {
500 			preempt_enable();
501 			rcu_read_unlock();
502 
503 			ret = -EINTR;
504 			goto out;
505 		}
506 
507 		if (need_resched()) {
508 			time_spent += ktime_get_ns() - time_start;
509 			preempt_enable();
510 			rcu_read_unlock();
511 
512 			cond_resched();
513 
514 			rcu_read_lock();
515 			preempt_disable();
516 			time_start = ktime_get_ns();
517 		}
518 	}
519 	time_spent += ktime_get_ns() - time_start;
520 	preempt_enable();
521 	rcu_read_unlock();
522 
523 	do_div(time_spent, repeat);
524 	duration = time_spent > U32_MAX ? U32_MAX : (u32)time_spent;
525 
526 	ret = bpf_test_finish(kattr, uattr, &flow_keys, sizeof(flow_keys),
527 			      retval, duration);
528 	if (!ret)
529 		ret = bpf_ctx_finish(kattr, uattr, user_ctx,
530 				     sizeof(struct bpf_flow_keys));
531 
532 out:
533 	kfree(user_ctx);
534 	kfree(data);
535 	return ret;
536 }
537