xref: /linux/net/bpf/test_run.c (revision 2c63221cd9e5c0dad0424029aeb1c40faada8330)
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 static void *bpf_test_init(const union bpf_attr *kattr, u32 size,
109 			   u32 headroom, u32 tailroom)
110 {
111 	void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
112 	void *data;
113 
114 	if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
115 		return ERR_PTR(-EINVAL);
116 
117 	data = kzalloc(size + headroom + tailroom, GFP_USER);
118 	if (!data)
119 		return ERR_PTR(-ENOMEM);
120 
121 	if (copy_from_user(data + headroom, data_in, size)) {
122 		kfree(data);
123 		return ERR_PTR(-EFAULT);
124 	}
125 	return data;
126 }
127 
128 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
129 {
130 	void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
131 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
132 	u32 size = kattr->test.ctx_size_in;
133 	void *data;
134 	int err;
135 
136 	if (!data_in && !data_out)
137 		return NULL;
138 
139 	data = kzalloc(max_size, GFP_USER);
140 	if (!data)
141 		return ERR_PTR(-ENOMEM);
142 
143 	if (data_in) {
144 		err = bpf_check_uarg_tail_zero(data_in, max_size, size);
145 		if (err) {
146 			kfree(data);
147 			return ERR_PTR(err);
148 		}
149 
150 		size = min_t(u32, max_size, size);
151 		if (copy_from_user(data, data_in, size)) {
152 			kfree(data);
153 			return ERR_PTR(-EFAULT);
154 		}
155 	}
156 	return data;
157 }
158 
159 static int bpf_ctx_finish(const union bpf_attr *kattr,
160 			  union bpf_attr __user *uattr, const void *data,
161 			  u32 size)
162 {
163 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
164 	int err = -EFAULT;
165 	u32 copy_size = size;
166 
167 	if (!data || !data_out)
168 		return 0;
169 
170 	if (copy_size > kattr->test.ctx_size_out) {
171 		copy_size = kattr->test.ctx_size_out;
172 		err = -ENOSPC;
173 	}
174 
175 	if (copy_to_user(data_out, data, copy_size))
176 		goto out;
177 	if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
178 		goto out;
179 	if (err != -ENOSPC)
180 		err = 0;
181 out:
182 	return err;
183 }
184 
185 /**
186  * range_is_zero - test whether buffer is initialized
187  * @buf: buffer to check
188  * @from: check from this position
189  * @to: check up until (excluding) this position
190  *
191  * This function returns true if the there is a non-zero byte
192  * in the buf in the range [from,to).
193  */
194 static inline bool range_is_zero(void *buf, size_t from, size_t to)
195 {
196 	return !memchr_inv((u8 *)buf + from, 0, to - from);
197 }
198 
199 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
200 {
201 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
202 
203 	if (!__skb)
204 		return 0;
205 
206 	/* make sure the fields we don't use are zeroed */
207 	if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, priority)))
208 		return -EINVAL;
209 
210 	/* priority is allowed */
211 
212 	if (!range_is_zero(__skb, offsetof(struct __sk_buff, priority) +
213 			   FIELD_SIZEOF(struct __sk_buff, priority),
214 			   offsetof(struct __sk_buff, cb)))
215 		return -EINVAL;
216 
217 	/* cb is allowed */
218 
219 	if (!range_is_zero(__skb, offsetof(struct __sk_buff, cb) +
220 			   FIELD_SIZEOF(struct __sk_buff, cb),
221 			   offsetof(struct __sk_buff, tstamp)))
222 		return -EINVAL;
223 
224 	/* tstamp is allowed */
225 
226 	if (!range_is_zero(__skb, offsetof(struct __sk_buff, tstamp) +
227 			   FIELD_SIZEOF(struct __sk_buff, tstamp),
228 			   sizeof(struct __sk_buff)))
229 		return -EINVAL;
230 
231 	skb->priority = __skb->priority;
232 	skb->tstamp = __skb->tstamp;
233 	memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
234 
235 	return 0;
236 }
237 
238 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
239 {
240 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
241 
242 	if (!__skb)
243 		return;
244 
245 	__skb->priority = skb->priority;
246 	__skb->tstamp = skb->tstamp;
247 	memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
248 }
249 
250 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
251 			  union bpf_attr __user *uattr)
252 {
253 	bool is_l2 = false, is_direct_pkt_access = false;
254 	u32 size = kattr->test.data_size_in;
255 	u32 repeat = kattr->test.repeat;
256 	struct __sk_buff *ctx = NULL;
257 	u32 retval, duration;
258 	int hh_len = ETH_HLEN;
259 	struct sk_buff *skb;
260 	struct sock *sk;
261 	void *data;
262 	int ret;
263 
264 	data = bpf_test_init(kattr, size, NET_SKB_PAD + NET_IP_ALIGN,
265 			     SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
266 	if (IS_ERR(data))
267 		return PTR_ERR(data);
268 
269 	ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
270 	if (IS_ERR(ctx)) {
271 		kfree(data);
272 		return PTR_ERR(ctx);
273 	}
274 
275 	switch (prog->type) {
276 	case BPF_PROG_TYPE_SCHED_CLS:
277 	case BPF_PROG_TYPE_SCHED_ACT:
278 		is_l2 = true;
279 		/* fall through */
280 	case BPF_PROG_TYPE_LWT_IN:
281 	case BPF_PROG_TYPE_LWT_OUT:
282 	case BPF_PROG_TYPE_LWT_XMIT:
283 		is_direct_pkt_access = true;
284 		break;
285 	default:
286 		break;
287 	}
288 
289 	sk = kzalloc(sizeof(struct sock), GFP_USER);
290 	if (!sk) {
291 		kfree(data);
292 		kfree(ctx);
293 		return -ENOMEM;
294 	}
295 	sock_net_set(sk, current->nsproxy->net_ns);
296 	sock_init_data(NULL, sk);
297 
298 	skb = build_skb(data, 0);
299 	if (!skb) {
300 		kfree(data);
301 		kfree(ctx);
302 		kfree(sk);
303 		return -ENOMEM;
304 	}
305 	skb->sk = sk;
306 
307 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
308 	__skb_put(skb, size);
309 	skb->protocol = eth_type_trans(skb, current->nsproxy->net_ns->loopback_dev);
310 	skb_reset_network_header(skb);
311 
312 	if (is_l2)
313 		__skb_push(skb, hh_len);
314 	if (is_direct_pkt_access)
315 		bpf_compute_data_pointers(skb);
316 	ret = convert___skb_to_skb(skb, ctx);
317 	if (ret)
318 		goto out;
319 	ret = bpf_test_run(prog, skb, repeat, &retval, &duration);
320 	if (ret)
321 		goto out;
322 	if (!is_l2) {
323 		if (skb_headroom(skb) < hh_len) {
324 			int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
325 
326 			if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
327 				ret = -ENOMEM;
328 				goto out;
329 			}
330 		}
331 		memset(__skb_push(skb, hh_len), 0, hh_len);
332 	}
333 	convert_skb_to___skb(skb, ctx);
334 
335 	size = skb->len;
336 	/* bpf program can never convert linear skb to non-linear */
337 	if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
338 		size = skb_headlen(skb);
339 	ret = bpf_test_finish(kattr, uattr, skb->data, size, retval, duration);
340 	if (!ret)
341 		ret = bpf_ctx_finish(kattr, uattr, ctx,
342 				     sizeof(struct __sk_buff));
343 out:
344 	kfree_skb(skb);
345 	bpf_sk_storage_free(sk);
346 	kfree(sk);
347 	kfree(ctx);
348 	return ret;
349 }
350 
351 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
352 			  union bpf_attr __user *uattr)
353 {
354 	u32 size = kattr->test.data_size_in;
355 	u32 repeat = kattr->test.repeat;
356 	struct netdev_rx_queue *rxqueue;
357 	struct xdp_buff xdp = {};
358 	u32 retval, duration;
359 	void *data;
360 	int ret;
361 
362 	if (kattr->test.ctx_in || kattr->test.ctx_out)
363 		return -EINVAL;
364 
365 	data = bpf_test_init(kattr, size, XDP_PACKET_HEADROOM + NET_IP_ALIGN, 0);
366 	if (IS_ERR(data))
367 		return PTR_ERR(data);
368 
369 	xdp.data_hard_start = data;
370 	xdp.data = data + XDP_PACKET_HEADROOM + NET_IP_ALIGN;
371 	xdp.data_meta = xdp.data;
372 	xdp.data_end = xdp.data + size;
373 
374 	rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
375 	xdp.rxq = &rxqueue->xdp_rxq;
376 
377 	ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration);
378 	if (ret)
379 		goto out;
380 	if (xdp.data != data + XDP_PACKET_HEADROOM + NET_IP_ALIGN ||
381 	    xdp.data_end != xdp.data + size)
382 		size = xdp.data_end - xdp.data;
383 	ret = bpf_test_finish(kattr, uattr, xdp.data, size, retval, duration);
384 out:
385 	kfree(data);
386 	return ret;
387 }
388 
389 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
390 {
391 	/* make sure the fields we don't use are zeroed */
392 	if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
393 		return -EINVAL;
394 
395 	/* flags is allowed */
396 
397 	if (!range_is_zero(ctx, offsetof(struct bpf_flow_keys, flags) +
398 			   FIELD_SIZEOF(struct bpf_flow_keys, flags),
399 			   sizeof(struct bpf_flow_keys)))
400 		return -EINVAL;
401 
402 	return 0;
403 }
404 
405 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
406 				     const union bpf_attr *kattr,
407 				     union bpf_attr __user *uattr)
408 {
409 	u32 size = kattr->test.data_size_in;
410 	struct bpf_flow_dissector ctx = {};
411 	u32 repeat = kattr->test.repeat;
412 	struct bpf_flow_keys *user_ctx;
413 	struct bpf_flow_keys flow_keys;
414 	u64 time_start, time_spent = 0;
415 	const struct ethhdr *eth;
416 	unsigned int flags = 0;
417 	u32 retval, duration;
418 	void *data;
419 	int ret;
420 	u32 i;
421 
422 	if (prog->type != BPF_PROG_TYPE_FLOW_DISSECTOR)
423 		return -EINVAL;
424 
425 	if (size < ETH_HLEN)
426 		return -EINVAL;
427 
428 	data = bpf_test_init(kattr, size, 0, 0);
429 	if (IS_ERR(data))
430 		return PTR_ERR(data);
431 
432 	eth = (struct ethhdr *)data;
433 
434 	if (!repeat)
435 		repeat = 1;
436 
437 	user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
438 	if (IS_ERR(user_ctx)) {
439 		kfree(data);
440 		return PTR_ERR(user_ctx);
441 	}
442 	if (user_ctx) {
443 		ret = verify_user_bpf_flow_keys(user_ctx);
444 		if (ret)
445 			goto out;
446 		flags = user_ctx->flags;
447 	}
448 
449 	ctx.flow_keys = &flow_keys;
450 	ctx.data = data;
451 	ctx.data_end = (__u8 *)data + size;
452 
453 	rcu_read_lock();
454 	preempt_disable();
455 	time_start = ktime_get_ns();
456 	for (i = 0; i < repeat; i++) {
457 		retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
458 					  size, flags);
459 
460 		if (signal_pending(current)) {
461 			preempt_enable();
462 			rcu_read_unlock();
463 
464 			ret = -EINTR;
465 			goto out;
466 		}
467 
468 		if (need_resched()) {
469 			time_spent += ktime_get_ns() - time_start;
470 			preempt_enable();
471 			rcu_read_unlock();
472 
473 			cond_resched();
474 
475 			rcu_read_lock();
476 			preempt_disable();
477 			time_start = ktime_get_ns();
478 		}
479 	}
480 	time_spent += ktime_get_ns() - time_start;
481 	preempt_enable();
482 	rcu_read_unlock();
483 
484 	do_div(time_spent, repeat);
485 	duration = time_spent > U32_MAX ? U32_MAX : (u32)time_spent;
486 
487 	ret = bpf_test_finish(kattr, uattr, &flow_keys, sizeof(flow_keys),
488 			      retval, duration);
489 	if (!ret)
490 		ret = bpf_ctx_finish(kattr, uattr, user_ctx,
491 				     sizeof(struct bpf_flow_keys));
492 
493 out:
494 	kfree(user_ctx);
495 	kfree(data);
496 	return ret;
497 }
498