flow_dissector: switch kernel context to struct bpf_flow_dissector
[linux-2.6-block.git] / net / bpf / test_run.c
CommitLineData
1cf1cae9
AS
1/* Copyright (c) 2017 Facebook
2 *
3 * This program is free software; you can redistribute it and/or
4 * modify it under the terms of version 2 of the GNU General Public
5 * License as published by the Free Software Foundation.
6 */
7#include <linux/bpf.h>
8#include <linux/slab.h>
9#include <linux/vmalloc.h>
10#include <linux/etherdevice.h>
11#include <linux/filter.h>
12#include <linux/sched/signal.h>
2cb494a3
SL
13#include <net/sock.h>
14#include <net/tcp.h>
1cf1cae9 15
df1a2cb7
SF
16static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
17 u32 *retval, u32 *time)
1cf1cae9 18{
71b91a50 19 struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE] = { NULL };
8bad74f9 20 enum bpf_cgroup_storage_type stype;
1cf1cae9 21 u64 time_start, time_spent = 0;
df1a2cb7 22 int ret = 0;
dcb40590 23 u32 i;
1cf1cae9 24
8bad74f9
RG
25 for_each_cgroup_storage_type(stype) {
26 storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
27 if (IS_ERR(storage[stype])) {
28 storage[stype] = NULL;
29 for_each_cgroup_storage_type(stype)
30 bpf_cgroup_storage_free(storage[stype]);
31 return -ENOMEM;
32 }
33 }
f42ee093 34
1cf1cae9
AS
35 if (!repeat)
36 repeat = 1;
df1a2cb7
SF
37
38 rcu_read_lock();
39 preempt_disable();
1cf1cae9
AS
40 time_start = ktime_get_ns();
41 for (i = 0; i < repeat; i++) {
df1a2cb7
SF
42 bpf_cgroup_storage_set(storage);
43 *retval = BPF_PROG_RUN(prog, ctx);
44
45 if (signal_pending(current)) {
46 ret = -EINTR;
47 break;
48 }
49
1cf1cae9 50 if (need_resched()) {
1cf1cae9 51 time_spent += ktime_get_ns() - time_start;
df1a2cb7
SF
52 preempt_enable();
53 rcu_read_unlock();
54
1cf1cae9 55 cond_resched();
df1a2cb7
SF
56
57 rcu_read_lock();
58 preempt_disable();
1cf1cae9
AS
59 time_start = ktime_get_ns();
60 }
61 }
62 time_spent += ktime_get_ns() - time_start;
df1a2cb7
SF
63 preempt_enable();
64 rcu_read_unlock();
65
1cf1cae9
AS
66 do_div(time_spent, repeat);
67 *time = time_spent > U32_MAX ? U32_MAX : (u32)time_spent;
68
8bad74f9
RG
69 for_each_cgroup_storage_type(stype)
70 bpf_cgroup_storage_free(storage[stype]);
f42ee093 71
df1a2cb7 72 return ret;
1cf1cae9
AS
73}
74
78e52272
DM
75static int bpf_test_finish(const union bpf_attr *kattr,
76 union bpf_attr __user *uattr, const void *data,
1cf1cae9
AS
77 u32 size, u32 retval, u32 duration)
78{
78e52272 79 void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
1cf1cae9 80 int err = -EFAULT;
b5a36b1e 81 u32 copy_size = size;
1cf1cae9 82
b5a36b1e
LB
83 /* Clamp copy if the user has provided a size hint, but copy the full
84 * buffer if not to retain old behaviour.
85 */
86 if (kattr->test.data_size_out &&
87 copy_size > kattr->test.data_size_out) {
88 copy_size = kattr->test.data_size_out;
89 err = -ENOSPC;
90 }
91
92 if (data_out && copy_to_user(data_out, data, copy_size))
1cf1cae9
AS
93 goto out;
94 if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
95 goto out;
96 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
97 goto out;
98 if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
99 goto out;
b5a36b1e
LB
100 if (err != -ENOSPC)
101 err = 0;
1cf1cae9
AS
102out:
103 return err;
104}
105
106static void *bpf_test_init(const union bpf_attr *kattr, u32 size,
107 u32 headroom, u32 tailroom)
108{
109 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
110 void *data;
111
112 if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
113 return ERR_PTR(-EINVAL);
114
115 data = kzalloc(size + headroom + tailroom, GFP_USER);
116 if (!data)
117 return ERR_PTR(-ENOMEM);
118
119 if (copy_from_user(data + headroom, data_in, size)) {
120 kfree(data);
121 return ERR_PTR(-EFAULT);
122 }
123 return data;
124}
125
b0b9395d
SF
126static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
127{
128 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
129 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
130 u32 size = kattr->test.ctx_size_in;
131 void *data;
132 int err;
133
134 if (!data_in && !data_out)
135 return NULL;
136
137 data = kzalloc(max_size, GFP_USER);
138 if (!data)
139 return ERR_PTR(-ENOMEM);
140
141 if (data_in) {
142 err = bpf_check_uarg_tail_zero(data_in, max_size, size);
143 if (err) {
144 kfree(data);
145 return ERR_PTR(err);
146 }
147
148 size = min_t(u32, max_size, size);
149 if (copy_from_user(data, data_in, size)) {
150 kfree(data);
151 return ERR_PTR(-EFAULT);
152 }
153 }
154 return data;
155}
156
157static int bpf_ctx_finish(const union bpf_attr *kattr,
158 union bpf_attr __user *uattr, const void *data,
159 u32 size)
160{
161 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
162 int err = -EFAULT;
163 u32 copy_size = size;
164
165 if (!data || !data_out)
166 return 0;
167
168 if (copy_size > kattr->test.ctx_size_out) {
169 copy_size = kattr->test.ctx_size_out;
170 err = -ENOSPC;
171 }
172
173 if (copy_to_user(data_out, data, copy_size))
174 goto out;
175 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
176 goto out;
177 if (err != -ENOSPC)
178 err = 0;
179out:
180 return err;
181}
182
183/**
184 * range_is_zero - test whether buffer is initialized
185 * @buf: buffer to check
186 * @from: check from this position
187 * @to: check up until (excluding) this position
188 *
189 * This function returns true if the there is a non-zero byte
190 * in the buf in the range [from,to).
191 */
192static inline bool range_is_zero(void *buf, size_t from, size_t to)
193{
194 return !memchr_inv((u8 *)buf + from, 0, to - from);
195}
196
197static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
198{
199 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
200
201 if (!__skb)
202 return 0;
203
204 /* make sure the fields we don't use are zeroed */
205 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, priority)))
206 return -EINVAL;
207
208 /* priority is allowed */
209
210 if (!range_is_zero(__skb, offsetof(struct __sk_buff, priority) +
211 FIELD_SIZEOF(struct __sk_buff, priority),
212 offsetof(struct __sk_buff, cb)))
213 return -EINVAL;
214
215 /* cb is allowed */
216
217 if (!range_is_zero(__skb, offsetof(struct __sk_buff, cb) +
218 FIELD_SIZEOF(struct __sk_buff, cb),
219 sizeof(struct __sk_buff)))
220 return -EINVAL;
221
222 skb->priority = __skb->priority;
223 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
224
225 return 0;
226}
227
228static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
229{
230 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
231
232 if (!__skb)
233 return;
234
235 __skb->priority = skb->priority;
236 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
237}
238
1cf1cae9
AS
239int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
240 union bpf_attr __user *uattr)
241{
242 bool is_l2 = false, is_direct_pkt_access = false;
243 u32 size = kattr->test.data_size_in;
244 u32 repeat = kattr->test.repeat;
b0b9395d 245 struct __sk_buff *ctx = NULL;
1cf1cae9 246 u32 retval, duration;
6e6fddc7 247 int hh_len = ETH_HLEN;
1cf1cae9 248 struct sk_buff *skb;
2cb494a3 249 struct sock *sk;
1cf1cae9
AS
250 void *data;
251 int ret;
252
586f8525 253 data = bpf_test_init(kattr, size, NET_SKB_PAD + NET_IP_ALIGN,
1cf1cae9
AS
254 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
255 if (IS_ERR(data))
256 return PTR_ERR(data);
257
b0b9395d
SF
258 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
259 if (IS_ERR(ctx)) {
260 kfree(data);
261 return PTR_ERR(ctx);
262 }
263
1cf1cae9
AS
264 switch (prog->type) {
265 case BPF_PROG_TYPE_SCHED_CLS:
266 case BPF_PROG_TYPE_SCHED_ACT:
267 is_l2 = true;
268 /* fall through */
269 case BPF_PROG_TYPE_LWT_IN:
270 case BPF_PROG_TYPE_LWT_OUT:
271 case BPF_PROG_TYPE_LWT_XMIT:
272 is_direct_pkt_access = true;
273 break;
274 default:
275 break;
276 }
277
2cb494a3
SL
278 sk = kzalloc(sizeof(struct sock), GFP_USER);
279 if (!sk) {
280 kfree(data);
b0b9395d 281 kfree(ctx);
2cb494a3
SL
282 return -ENOMEM;
283 }
284 sock_net_set(sk, current->nsproxy->net_ns);
285 sock_init_data(NULL, sk);
286
1cf1cae9
AS
287 skb = build_skb(data, 0);
288 if (!skb) {
289 kfree(data);
b0b9395d 290 kfree(ctx);
2cb494a3 291 kfree(sk);
1cf1cae9
AS
292 return -ENOMEM;
293 }
2cb494a3 294 skb->sk = sk;
1cf1cae9 295
586f8525 296 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1cf1cae9
AS
297 __skb_put(skb, size);
298 skb->protocol = eth_type_trans(skb, current->nsproxy->net_ns->loopback_dev);
299 skb_reset_network_header(skb);
300
301 if (is_l2)
6e6fddc7 302 __skb_push(skb, hh_len);
1cf1cae9 303 if (is_direct_pkt_access)
6aaae2b6 304 bpf_compute_data_pointers(skb);
b0b9395d
SF
305 ret = convert___skb_to_skb(skb, ctx);
306 if (ret)
307 goto out;
dcb40590 308 ret = bpf_test_run(prog, skb, repeat, &retval, &duration);
b0b9395d
SF
309 if (ret)
310 goto out;
6e6fddc7
DB
311 if (!is_l2) {
312 if (skb_headroom(skb) < hh_len) {
313 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
314
315 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
b0b9395d
SF
316 ret = -ENOMEM;
317 goto out;
6e6fddc7
DB
318 }
319 }
320 memset(__skb_push(skb, hh_len), 0, hh_len);
321 }
b0b9395d 322 convert_skb_to___skb(skb, ctx);
6e6fddc7 323
1cf1cae9
AS
324 size = skb->len;
325 /* bpf program can never convert linear skb to non-linear */
326 if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
327 size = skb_headlen(skb);
78e52272 328 ret = bpf_test_finish(kattr, uattr, skb->data, size, retval, duration);
b0b9395d
SF
329 if (!ret)
330 ret = bpf_ctx_finish(kattr, uattr, ctx,
331 sizeof(struct __sk_buff));
332out:
1cf1cae9 333 kfree_skb(skb);
2cb494a3 334 kfree(sk);
b0b9395d 335 kfree(ctx);
1cf1cae9
AS
336 return ret;
337}
338
339int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
340 union bpf_attr __user *uattr)
341{
342 u32 size = kattr->test.data_size_in;
343 u32 repeat = kattr->test.repeat;
65073a67 344 struct netdev_rx_queue *rxqueue;
1cf1cae9
AS
345 struct xdp_buff xdp = {};
346 u32 retval, duration;
347 void *data;
348 int ret;
349
947e8b59
SF
350 if (kattr->test.ctx_in || kattr->test.ctx_out)
351 return -EINVAL;
352
586f8525 353 data = bpf_test_init(kattr, size, XDP_PACKET_HEADROOM + NET_IP_ALIGN, 0);
1cf1cae9
AS
354 if (IS_ERR(data))
355 return PTR_ERR(data);
356
357 xdp.data_hard_start = data;
586f8525 358 xdp.data = data + XDP_PACKET_HEADROOM + NET_IP_ALIGN;
de8f3a83 359 xdp.data_meta = xdp.data;
1cf1cae9
AS
360 xdp.data_end = xdp.data + size;
361
65073a67
DB
362 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
363 xdp.rxq = &rxqueue->xdp_rxq;
364
dcb40590
RG
365 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration);
366 if (ret)
367 goto out;
587b80cc
NS
368 if (xdp.data != data + XDP_PACKET_HEADROOM + NET_IP_ALIGN ||
369 xdp.data_end != xdp.data + size)
1cf1cae9 370 size = xdp.data_end - xdp.data;
78e52272 371 ret = bpf_test_finish(kattr, uattr, xdp.data, size, retval, duration);
dcb40590 372out:
1cf1cae9
AS
373 kfree(data);
374 return ret;
375}
b7a1848e
SF
376
377int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
378 const union bpf_attr *kattr,
379 union bpf_attr __user *uattr)
380{
381 u32 size = kattr->test.data_size_in;
382 u32 repeat = kattr->test.repeat;
383 struct bpf_flow_keys flow_keys;
384 u64 time_start, time_spent = 0;
b7a1848e
SF
385 u32 retval, duration;
386 struct sk_buff *skb;
387 struct sock *sk;
388 void *data;
389 int ret;
390 u32 i;
391
392 if (prog->type != BPF_PROG_TYPE_FLOW_DISSECTOR)
393 return -EINVAL;
394
947e8b59
SF
395 if (kattr->test.ctx_in || kattr->test.ctx_out)
396 return -EINVAL;
397
b7a1848e
SF
398 data = bpf_test_init(kattr, size, NET_SKB_PAD + NET_IP_ALIGN,
399 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
400 if (IS_ERR(data))
401 return PTR_ERR(data);
402
403 sk = kzalloc(sizeof(*sk), GFP_USER);
404 if (!sk) {
405 kfree(data);
406 return -ENOMEM;
407 }
408 sock_net_set(sk, current->nsproxy->net_ns);
409 sock_init_data(NULL, sk);
410
411 skb = build_skb(data, 0);
412 if (!skb) {
413 kfree(data);
414 kfree(sk);
415 return -ENOMEM;
416 }
417 skb->sk = sk;
418
419 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
420 __skb_put(skb, size);
421 skb->protocol = eth_type_trans(skb,
422 current->nsproxy->net_ns->loopback_dev);
423 skb_reset_network_header(skb);
424
b7a1848e
SF
425 if (!repeat)
426 repeat = 1;
427
a439184d
SF
428 rcu_read_lock();
429 preempt_disable();
b7a1848e
SF
430 time_start = ktime_get_ns();
431 for (i = 0; i < repeat; i++) {
b7a1848e
SF
432 retval = __skb_flow_bpf_dissect(prog, skb,
433 &flow_keys_dissector,
434 &flow_keys);
a439184d
SF
435
436 if (signal_pending(current)) {
437 preempt_enable();
438 rcu_read_unlock();
439
440 ret = -EINTR;
441 goto out;
442 }
b7a1848e
SF
443
444 if (need_resched()) {
b7a1848e 445 time_spent += ktime_get_ns() - time_start;
a439184d
SF
446 preempt_enable();
447 rcu_read_unlock();
448
b7a1848e 449 cond_resched();
a439184d
SF
450
451 rcu_read_lock();
452 preempt_disable();
b7a1848e
SF
453 time_start = ktime_get_ns();
454 }
455 }
456 time_spent += ktime_get_ns() - time_start;
a439184d
SF
457 preempt_enable();
458 rcu_read_unlock();
459
b7a1848e
SF
460 do_div(time_spent, repeat);
461 duration = time_spent > U32_MAX ? U32_MAX : (u32)time_spent;
462
463 ret = bpf_test_finish(kattr, uattr, &flow_keys, sizeof(flow_keys),
464 retval, duration);
465
a439184d 466out:
b7a1848e
SF
467 kfree_skb(skb);
468 kfree(sk);
469 return ret;
470}