Loading...
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/rcupdate_trace.h>
11#include <linux/sched/signal.h>
12#include <net/bpf_sk_storage.h>
13#include <net/sock.h>
14#include <net/tcp.h>
15#include <net/net_namespace.h>
16#include <linux/error-injection.h>
17#include <linux/smp.h>
18#include <linux/sock_diag.h>
19
20#define CREATE_TRACE_POINTS
21#include <trace/events/bpf_test_run.h>
22
23struct bpf_test_timer {
24 enum { NO_PREEMPT, NO_MIGRATE } mode;
25 u32 i;
26 u64 time_start, time_spent;
27};
28
29static void bpf_test_timer_enter(struct bpf_test_timer *t)
30 __acquires(rcu)
31{
32 rcu_read_lock();
33 if (t->mode == NO_PREEMPT)
34 preempt_disable();
35 else
36 migrate_disable();
37
38 t->time_start = ktime_get_ns();
39}
40
41static void bpf_test_timer_leave(struct bpf_test_timer *t)
42 __releases(rcu)
43{
44 t->time_start = 0;
45
46 if (t->mode == NO_PREEMPT)
47 preempt_enable();
48 else
49 migrate_enable();
50 rcu_read_unlock();
51}
52
53static bool bpf_test_timer_continue(struct bpf_test_timer *t, u32 repeat, int *err, u32 *duration)
54 __must_hold(rcu)
55{
56 t->i++;
57 if (t->i >= repeat) {
58 /* We're done. */
59 t->time_spent += ktime_get_ns() - t->time_start;
60 do_div(t->time_spent, t->i);
61 *duration = t->time_spent > U32_MAX ? U32_MAX : (u32)t->time_spent;
62 *err = 0;
63 goto reset;
64 }
65
66 if (signal_pending(current)) {
67 /* During iteration: we've been cancelled, abort. */
68 *err = -EINTR;
69 goto reset;
70 }
71
72 if (need_resched()) {
73 /* During iteration: we need to reschedule between runs. */
74 t->time_spent += ktime_get_ns() - t->time_start;
75 bpf_test_timer_leave(t);
76 cond_resched();
77 bpf_test_timer_enter(t);
78 }
79
80 /* Do another round. */
81 return true;
82
83reset:
84 t->i = 0;
85 return false;
86}
87
88static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
89 u32 *retval, u32 *time, bool xdp)
90{
91 struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE] = { NULL };
92 struct bpf_test_timer t = { NO_MIGRATE };
93 enum bpf_cgroup_storage_type stype;
94 int ret;
95
96 for_each_cgroup_storage_type(stype) {
97 storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
98 if (IS_ERR(storage[stype])) {
99 storage[stype] = NULL;
100 for_each_cgroup_storage_type(stype)
101 bpf_cgroup_storage_free(storage[stype]);
102 return -ENOMEM;
103 }
104 }
105
106 if (!repeat)
107 repeat = 1;
108
109 bpf_test_timer_enter(&t);
110 do {
111 ret = bpf_cgroup_storage_set(storage);
112 if (ret)
113 break;
114
115 if (xdp)
116 *retval = bpf_prog_run_xdp(prog, ctx);
117 else
118 *retval = BPF_PROG_RUN(prog, ctx);
119
120 bpf_cgroup_storage_unset();
121 } while (bpf_test_timer_continue(&t, repeat, &ret, time));
122 bpf_test_timer_leave(&t);
123
124 for_each_cgroup_storage_type(stype)
125 bpf_cgroup_storage_free(storage[stype]);
126
127 return ret;
128}
129
130static int bpf_test_finish(const union bpf_attr *kattr,
131 union bpf_attr __user *uattr, const void *data,
132 u32 size, u32 retval, u32 duration)
133{
134 void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
135 int err = -EFAULT;
136 u32 copy_size = size;
137
138 /* Clamp copy if the user has provided a size hint, but copy the full
139 * buffer if not to retain old behaviour.
140 */
141 if (kattr->test.data_size_out &&
142 copy_size > kattr->test.data_size_out) {
143 copy_size = kattr->test.data_size_out;
144 err = -ENOSPC;
145 }
146
147 if (data_out && copy_to_user(data_out, data, copy_size))
148 goto out;
149 if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
150 goto out;
151 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
152 goto out;
153 if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
154 goto out;
155 if (err != -ENOSPC)
156 err = 0;
157out:
158 trace_bpf_test_finish(&err);
159 return err;
160}
161
162/* Integer types of various sizes and pointer combinations cover variety of
163 * architecture dependent calling conventions. 7+ can be supported in the
164 * future.
165 */
166__diag_push();
167__diag_ignore(GCC, 8, "-Wmissing-prototypes",
168 "Global functions as their definitions will be in vmlinux BTF");
169int noinline bpf_fentry_test1(int a)
170{
171 return a + 1;
172}
173
174int noinline bpf_fentry_test2(int a, u64 b)
175{
176 return a + b;
177}
178
179int noinline bpf_fentry_test3(char a, int b, u64 c)
180{
181 return a + b + c;
182}
183
184int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
185{
186 return (long)a + b + c + d;
187}
188
189int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
190{
191 return a + (long)b + c + d + e;
192}
193
194int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
195{
196 return a + (long)b + c + d + (long)e + f;
197}
198
199struct bpf_fentry_test_t {
200 struct bpf_fentry_test_t *a;
201};
202
203int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
204{
205 return (long)arg;
206}
207
208int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
209{
210 return (long)arg->a;
211}
212
213int noinline bpf_modify_return_test(int a, int *b)
214{
215 *b += 1;
216 return a + *b;
217}
218
219u64 noinline bpf_kfunc_call_test1(struct sock *sk, u32 a, u64 b, u32 c, u64 d)
220{
221 return a + b + c + d;
222}
223
224int noinline bpf_kfunc_call_test2(struct sock *sk, u32 a, u32 b)
225{
226 return a + b;
227}
228
229struct sock * noinline bpf_kfunc_call_test3(struct sock *sk)
230{
231 return sk;
232}
233
234__diag_pop();
235
236ALLOW_ERROR_INJECTION(bpf_modify_return_test, ERRNO);
237
238BTF_SET_START(test_sk_kfunc_ids)
239BTF_ID(func, bpf_kfunc_call_test1)
240BTF_ID(func, bpf_kfunc_call_test2)
241BTF_ID(func, bpf_kfunc_call_test3)
242BTF_SET_END(test_sk_kfunc_ids)
243
244bool bpf_prog_test_check_kfunc_call(u32 kfunc_id)
245{
246 return btf_id_set_contains(&test_sk_kfunc_ids, kfunc_id);
247}
248
249static void *bpf_test_init(const union bpf_attr *kattr, u32 size,
250 u32 headroom, u32 tailroom)
251{
252 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
253 u32 user_size = kattr->test.data_size_in;
254 void *data;
255
256 if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
257 return ERR_PTR(-EINVAL);
258
259 if (user_size > size)
260 return ERR_PTR(-EMSGSIZE);
261
262 data = kzalloc(size + headroom + tailroom, GFP_USER);
263 if (!data)
264 return ERR_PTR(-ENOMEM);
265
266 if (copy_from_user(data + headroom, data_in, user_size)) {
267 kfree(data);
268 return ERR_PTR(-EFAULT);
269 }
270
271 return data;
272}
273
274int bpf_prog_test_run_tracing(struct bpf_prog *prog,
275 const union bpf_attr *kattr,
276 union bpf_attr __user *uattr)
277{
278 struct bpf_fentry_test_t arg = {};
279 u16 side_effect = 0, ret = 0;
280 int b = 2, err = -EFAULT;
281 u32 retval = 0;
282
283 if (kattr->test.flags || kattr->test.cpu)
284 return -EINVAL;
285
286 switch (prog->expected_attach_type) {
287 case BPF_TRACE_FENTRY:
288 case BPF_TRACE_FEXIT:
289 if (bpf_fentry_test1(1) != 2 ||
290 bpf_fentry_test2(2, 3) != 5 ||
291 bpf_fentry_test3(4, 5, 6) != 15 ||
292 bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
293 bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
294 bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
295 bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
296 bpf_fentry_test8(&arg) != 0)
297 goto out;
298 break;
299 case BPF_MODIFY_RETURN:
300 ret = bpf_modify_return_test(1, &b);
301 if (b != 2)
302 side_effect = 1;
303 break;
304 default:
305 goto out;
306 }
307
308 retval = ((u32)side_effect << 16) | ret;
309 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
310 goto out;
311
312 err = 0;
313out:
314 trace_bpf_test_finish(&err);
315 return err;
316}
317
318struct bpf_raw_tp_test_run_info {
319 struct bpf_prog *prog;
320 void *ctx;
321 u32 retval;
322};
323
324static void
325__bpf_prog_test_run_raw_tp(void *data)
326{
327 struct bpf_raw_tp_test_run_info *info = data;
328
329 rcu_read_lock();
330 info->retval = BPF_PROG_RUN(info->prog, info->ctx);
331 rcu_read_unlock();
332}
333
334int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
335 const union bpf_attr *kattr,
336 union bpf_attr __user *uattr)
337{
338 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
339 __u32 ctx_size_in = kattr->test.ctx_size_in;
340 struct bpf_raw_tp_test_run_info info;
341 int cpu = kattr->test.cpu, err = 0;
342 int current_cpu;
343
344 /* doesn't support data_in/out, ctx_out, duration, or repeat */
345 if (kattr->test.data_in || kattr->test.data_out ||
346 kattr->test.ctx_out || kattr->test.duration ||
347 kattr->test.repeat)
348 return -EINVAL;
349
350 if (ctx_size_in < prog->aux->max_ctx_offset ||
351 ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
352 return -EINVAL;
353
354 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
355 return -EINVAL;
356
357 if (ctx_size_in) {
358 info.ctx = kzalloc(ctx_size_in, GFP_USER);
359 if (!info.ctx)
360 return -ENOMEM;
361 if (copy_from_user(info.ctx, ctx_in, ctx_size_in)) {
362 err = -EFAULT;
363 goto out;
364 }
365 } else {
366 info.ctx = NULL;
367 }
368
369 info.prog = prog;
370
371 current_cpu = get_cpu();
372 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
373 cpu == current_cpu) {
374 __bpf_prog_test_run_raw_tp(&info);
375 } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
376 /* smp_call_function_single() also checks cpu_online()
377 * after csd_lock(). However, since cpu is from user
378 * space, let's do an extra quick check to filter out
379 * invalid value before smp_call_function_single().
380 */
381 err = -ENXIO;
382 } else {
383 err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
384 &info, 1);
385 }
386 put_cpu();
387
388 if (!err &&
389 copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
390 err = -EFAULT;
391
392out:
393 kfree(info.ctx);
394 return err;
395}
396
397static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
398{
399 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
400 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
401 u32 size = kattr->test.ctx_size_in;
402 void *data;
403 int err;
404
405 if (!data_in && !data_out)
406 return NULL;
407
408 data = kzalloc(max_size, GFP_USER);
409 if (!data)
410 return ERR_PTR(-ENOMEM);
411
412 if (data_in) {
413 err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
414 if (err) {
415 kfree(data);
416 return ERR_PTR(err);
417 }
418
419 size = min_t(u32, max_size, size);
420 if (copy_from_user(data, data_in, size)) {
421 kfree(data);
422 return ERR_PTR(-EFAULT);
423 }
424 }
425 return data;
426}
427
428static int bpf_ctx_finish(const union bpf_attr *kattr,
429 union bpf_attr __user *uattr, const void *data,
430 u32 size)
431{
432 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
433 int err = -EFAULT;
434 u32 copy_size = size;
435
436 if (!data || !data_out)
437 return 0;
438
439 if (copy_size > kattr->test.ctx_size_out) {
440 copy_size = kattr->test.ctx_size_out;
441 err = -ENOSPC;
442 }
443
444 if (copy_to_user(data_out, data, copy_size))
445 goto out;
446 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
447 goto out;
448 if (err != -ENOSPC)
449 err = 0;
450out:
451 return err;
452}
453
454/**
455 * range_is_zero - test whether buffer is initialized
456 * @buf: buffer to check
457 * @from: check from this position
458 * @to: check up until (excluding) this position
459 *
460 * This function returns true if the there is a non-zero byte
461 * in the buf in the range [from,to).
462 */
463static inline bool range_is_zero(void *buf, size_t from, size_t to)
464{
465 return !memchr_inv((u8 *)buf + from, 0, to - from);
466}
467
468static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
469{
470 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
471
472 if (!__skb)
473 return 0;
474
475 /* make sure the fields we don't use are zeroed */
476 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
477 return -EINVAL;
478
479 /* mark is allowed */
480
481 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
482 offsetof(struct __sk_buff, priority)))
483 return -EINVAL;
484
485 /* priority is allowed */
486
487 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, priority),
488 offsetof(struct __sk_buff, ifindex)))
489 return -EINVAL;
490
491 /* ifindex is allowed */
492
493 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
494 offsetof(struct __sk_buff, cb)))
495 return -EINVAL;
496
497 /* cb is allowed */
498
499 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
500 offsetof(struct __sk_buff, tstamp)))
501 return -EINVAL;
502
503 /* tstamp is allowed */
504 /* wire_len is allowed */
505 /* gso_segs is allowed */
506
507 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
508 offsetof(struct __sk_buff, gso_size)))
509 return -EINVAL;
510
511 /* gso_size is allowed */
512
513 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
514 sizeof(struct __sk_buff)))
515 return -EINVAL;
516
517 skb->mark = __skb->mark;
518 skb->priority = __skb->priority;
519 skb->tstamp = __skb->tstamp;
520 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
521
522 if (__skb->wire_len == 0) {
523 cb->pkt_len = skb->len;
524 } else {
525 if (__skb->wire_len < skb->len ||
526 __skb->wire_len > GSO_MAX_SIZE)
527 return -EINVAL;
528 cb->pkt_len = __skb->wire_len;
529 }
530
531 if (__skb->gso_segs > GSO_MAX_SEGS)
532 return -EINVAL;
533 skb_shinfo(skb)->gso_segs = __skb->gso_segs;
534 skb_shinfo(skb)->gso_size = __skb->gso_size;
535
536 return 0;
537}
538
539static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
540{
541 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
542
543 if (!__skb)
544 return;
545
546 __skb->mark = skb->mark;
547 __skb->priority = skb->priority;
548 __skb->ifindex = skb->dev->ifindex;
549 __skb->tstamp = skb->tstamp;
550 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
551 __skb->wire_len = cb->pkt_len;
552 __skb->gso_segs = skb_shinfo(skb)->gso_segs;
553}
554
555static struct proto bpf_dummy_proto = {
556 .name = "bpf_dummy",
557 .owner = THIS_MODULE,
558 .obj_size = sizeof(struct sock),
559};
560
561int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
562 union bpf_attr __user *uattr)
563{
564 bool is_l2 = false, is_direct_pkt_access = false;
565 struct net *net = current->nsproxy->net_ns;
566 struct net_device *dev = net->loopback_dev;
567 u32 size = kattr->test.data_size_in;
568 u32 repeat = kattr->test.repeat;
569 struct __sk_buff *ctx = NULL;
570 u32 retval, duration;
571 int hh_len = ETH_HLEN;
572 struct sk_buff *skb;
573 struct sock *sk;
574 void *data;
575 int ret;
576
577 if (kattr->test.flags || kattr->test.cpu)
578 return -EINVAL;
579
580 data = bpf_test_init(kattr, size, NET_SKB_PAD + NET_IP_ALIGN,
581 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
582 if (IS_ERR(data))
583 return PTR_ERR(data);
584
585 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
586 if (IS_ERR(ctx)) {
587 kfree(data);
588 return PTR_ERR(ctx);
589 }
590
591 switch (prog->type) {
592 case BPF_PROG_TYPE_SCHED_CLS:
593 case BPF_PROG_TYPE_SCHED_ACT:
594 is_l2 = true;
595 fallthrough;
596 case BPF_PROG_TYPE_LWT_IN:
597 case BPF_PROG_TYPE_LWT_OUT:
598 case BPF_PROG_TYPE_LWT_XMIT:
599 is_direct_pkt_access = true;
600 break;
601 default:
602 break;
603 }
604
605 sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
606 if (!sk) {
607 kfree(data);
608 kfree(ctx);
609 return -ENOMEM;
610 }
611 sock_init_data(NULL, sk);
612
613 skb = build_skb(data, 0);
614 if (!skb) {
615 kfree(data);
616 kfree(ctx);
617 sk_free(sk);
618 return -ENOMEM;
619 }
620 skb->sk = sk;
621
622 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
623 __skb_put(skb, size);
624 if (ctx && ctx->ifindex > 1) {
625 dev = dev_get_by_index(net, ctx->ifindex);
626 if (!dev) {
627 ret = -ENODEV;
628 goto out;
629 }
630 }
631 skb->protocol = eth_type_trans(skb, dev);
632 skb_reset_network_header(skb);
633
634 switch (skb->protocol) {
635 case htons(ETH_P_IP):
636 sk->sk_family = AF_INET;
637 if (sizeof(struct iphdr) <= skb_headlen(skb)) {
638 sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
639 sk->sk_daddr = ip_hdr(skb)->daddr;
640 }
641 break;
642#if IS_ENABLED(CONFIG_IPV6)
643 case htons(ETH_P_IPV6):
644 sk->sk_family = AF_INET6;
645 if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
646 sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
647 sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
648 }
649 break;
650#endif
651 default:
652 break;
653 }
654
655 if (is_l2)
656 __skb_push(skb, hh_len);
657 if (is_direct_pkt_access)
658 bpf_compute_data_pointers(skb);
659 ret = convert___skb_to_skb(skb, ctx);
660 if (ret)
661 goto out;
662 ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
663 if (ret)
664 goto out;
665 if (!is_l2) {
666 if (skb_headroom(skb) < hh_len) {
667 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
668
669 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
670 ret = -ENOMEM;
671 goto out;
672 }
673 }
674 memset(__skb_push(skb, hh_len), 0, hh_len);
675 }
676 convert_skb_to___skb(skb, ctx);
677
678 size = skb->len;
679 /* bpf program can never convert linear skb to non-linear */
680 if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
681 size = skb_headlen(skb);
682 ret = bpf_test_finish(kattr, uattr, skb->data, size, retval, duration);
683 if (!ret)
684 ret = bpf_ctx_finish(kattr, uattr, ctx,
685 sizeof(struct __sk_buff));
686out:
687 if (dev && dev != net->loopback_dev)
688 dev_put(dev);
689 kfree_skb(skb);
690 sk_free(sk);
691 kfree(ctx);
692 return ret;
693}
694
695int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
696 union bpf_attr __user *uattr)
697{
698 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
699 u32 headroom = XDP_PACKET_HEADROOM;
700 u32 size = kattr->test.data_size_in;
701 u32 repeat = kattr->test.repeat;
702 struct netdev_rx_queue *rxqueue;
703 struct xdp_buff xdp = {};
704 u32 retval, duration;
705 u32 max_data_sz;
706 void *data;
707 int ret;
708
709 if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
710 prog->expected_attach_type == BPF_XDP_CPUMAP)
711 return -EINVAL;
712 if (kattr->test.ctx_in || kattr->test.ctx_out)
713 return -EINVAL;
714
715 /* XDP have extra tailroom as (most) drivers use full page */
716 max_data_sz = 4096 - headroom - tailroom;
717
718 data = bpf_test_init(kattr, max_data_sz, headroom, tailroom);
719 if (IS_ERR(data))
720 return PTR_ERR(data);
721
722 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
723 xdp_init_buff(&xdp, headroom + max_data_sz + tailroom,
724 &rxqueue->xdp_rxq);
725 xdp_prepare_buff(&xdp, data, headroom, size, true);
726
727 bpf_prog_change_xdp(NULL, prog);
728 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
729 if (ret)
730 goto out;
731 if (xdp.data != data + headroom || xdp.data_end != xdp.data + size)
732 size = xdp.data_end - xdp.data;
733 ret = bpf_test_finish(kattr, uattr, xdp.data, size, retval, duration);
734out:
735 bpf_prog_change_xdp(prog, NULL);
736 kfree(data);
737 return ret;
738}
739
740static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
741{
742 /* make sure the fields we don't use are zeroed */
743 if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
744 return -EINVAL;
745
746 /* flags is allowed */
747
748 if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
749 sizeof(struct bpf_flow_keys)))
750 return -EINVAL;
751
752 return 0;
753}
754
755int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
756 const union bpf_attr *kattr,
757 union bpf_attr __user *uattr)
758{
759 struct bpf_test_timer t = { NO_PREEMPT };
760 u32 size = kattr->test.data_size_in;
761 struct bpf_flow_dissector ctx = {};
762 u32 repeat = kattr->test.repeat;
763 struct bpf_flow_keys *user_ctx;
764 struct bpf_flow_keys flow_keys;
765 const struct ethhdr *eth;
766 unsigned int flags = 0;
767 u32 retval, duration;
768 void *data;
769 int ret;
770
771 if (prog->type != BPF_PROG_TYPE_FLOW_DISSECTOR)
772 return -EINVAL;
773
774 if (kattr->test.flags || kattr->test.cpu)
775 return -EINVAL;
776
777 if (size < ETH_HLEN)
778 return -EINVAL;
779
780 data = bpf_test_init(kattr, size, 0, 0);
781 if (IS_ERR(data))
782 return PTR_ERR(data);
783
784 eth = (struct ethhdr *)data;
785
786 if (!repeat)
787 repeat = 1;
788
789 user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
790 if (IS_ERR(user_ctx)) {
791 kfree(data);
792 return PTR_ERR(user_ctx);
793 }
794 if (user_ctx) {
795 ret = verify_user_bpf_flow_keys(user_ctx);
796 if (ret)
797 goto out;
798 flags = user_ctx->flags;
799 }
800
801 ctx.flow_keys = &flow_keys;
802 ctx.data = data;
803 ctx.data_end = (__u8 *)data + size;
804
805 bpf_test_timer_enter(&t);
806 do {
807 retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
808 size, flags);
809 } while (bpf_test_timer_continue(&t, repeat, &ret, &duration));
810 bpf_test_timer_leave(&t);
811
812 if (ret < 0)
813 goto out;
814
815 ret = bpf_test_finish(kattr, uattr, &flow_keys, sizeof(flow_keys),
816 retval, duration);
817 if (!ret)
818 ret = bpf_ctx_finish(kattr, uattr, user_ctx,
819 sizeof(struct bpf_flow_keys));
820
821out:
822 kfree(user_ctx);
823 kfree(data);
824 return ret;
825}
826
827int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
828 union bpf_attr __user *uattr)
829{
830 struct bpf_test_timer t = { NO_PREEMPT };
831 struct bpf_prog_array *progs = NULL;
832 struct bpf_sk_lookup_kern ctx = {};
833 u32 repeat = kattr->test.repeat;
834 struct bpf_sk_lookup *user_ctx;
835 u32 retval, duration;
836 int ret = -EINVAL;
837
838 if (prog->type != BPF_PROG_TYPE_SK_LOOKUP)
839 return -EINVAL;
840
841 if (kattr->test.flags || kattr->test.cpu)
842 return -EINVAL;
843
844 if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
845 kattr->test.data_size_out)
846 return -EINVAL;
847
848 if (!repeat)
849 repeat = 1;
850
851 user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
852 if (IS_ERR(user_ctx))
853 return PTR_ERR(user_ctx);
854
855 if (!user_ctx)
856 return -EINVAL;
857
858 if (user_ctx->sk)
859 goto out;
860
861 if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
862 goto out;
863
864 if (user_ctx->local_port > U16_MAX || user_ctx->remote_port > U16_MAX) {
865 ret = -ERANGE;
866 goto out;
867 }
868
869 ctx.family = (u16)user_ctx->family;
870 ctx.protocol = (u16)user_ctx->protocol;
871 ctx.dport = (u16)user_ctx->local_port;
872 ctx.sport = (__force __be16)user_ctx->remote_port;
873
874 switch (ctx.family) {
875 case AF_INET:
876 ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
877 ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
878 break;
879
880#if IS_ENABLED(CONFIG_IPV6)
881 case AF_INET6:
882 ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
883 ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
884 break;
885#endif
886
887 default:
888 ret = -EAFNOSUPPORT;
889 goto out;
890 }
891
892 progs = bpf_prog_array_alloc(1, GFP_KERNEL);
893 if (!progs) {
894 ret = -ENOMEM;
895 goto out;
896 }
897
898 progs->items[0].prog = prog;
899
900 bpf_test_timer_enter(&t);
901 do {
902 ctx.selected_sk = NULL;
903 retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, BPF_PROG_RUN);
904 } while (bpf_test_timer_continue(&t, repeat, &ret, &duration));
905 bpf_test_timer_leave(&t);
906
907 if (ret < 0)
908 goto out;
909
910 user_ctx->cookie = 0;
911 if (ctx.selected_sk) {
912 if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
913 ret = -EOPNOTSUPP;
914 goto out;
915 }
916
917 user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
918 }
919
920 ret = bpf_test_finish(kattr, uattr, NULL, 0, retval, duration);
921 if (!ret)
922 ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
923
924out:
925 bpf_prog_array_free(progs);
926 kfree(user_ctx);
927 return ret;
928}
929
930int bpf_prog_test_run_syscall(struct bpf_prog *prog,
931 const union bpf_attr *kattr,
932 union bpf_attr __user *uattr)
933{
934 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
935 __u32 ctx_size_in = kattr->test.ctx_size_in;
936 void *ctx = NULL;
937 u32 retval;
938 int err = 0;
939
940 /* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
941 if (kattr->test.data_in || kattr->test.data_out ||
942 kattr->test.ctx_out || kattr->test.duration ||
943 kattr->test.repeat || kattr->test.flags)
944 return -EINVAL;
945
946 if (ctx_size_in < prog->aux->max_ctx_offset ||
947 ctx_size_in > U16_MAX)
948 return -EINVAL;
949
950 if (ctx_size_in) {
951 ctx = kzalloc(ctx_size_in, GFP_USER);
952 if (!ctx)
953 return -ENOMEM;
954 if (copy_from_user(ctx, ctx_in, ctx_size_in)) {
955 err = -EFAULT;
956 goto out;
957 }
958 }
959
960 rcu_read_lock_trace();
961 retval = bpf_prog_run_pin_on_cpu(prog, ctx);
962 rcu_read_unlock_trace();
963
964 if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
965 err = -EFAULT;
966 goto out;
967 }
968 if (ctx_size_in)
969 if (copy_to_user(ctx_in, ctx, ctx_size_in))
970 err = -EFAULT;
971out:
972 kfree(ctx);
973 return err;
974}
1// SPDX-License-Identifier: GPL-2.0-only
2/* Copyright (c) 2017 Facebook
3 */
4#include <linux/bpf.h>
5#include <linux/btf.h>
6#include <linux/btf_ids.h>
7#include <linux/slab.h>
8#include <linux/init.h>
9#include <linux/vmalloc.h>
10#include <linux/etherdevice.h>
11#include <linux/filter.h>
12#include <linux/rcupdate_trace.h>
13#include <linux/sched/signal.h>
14#include <net/bpf_sk_storage.h>
15#include <net/hotdata.h>
16#include <net/sock.h>
17#include <net/tcp.h>
18#include <net/net_namespace.h>
19#include <net/page_pool/helpers.h>
20#include <linux/error-injection.h>
21#include <linux/smp.h>
22#include <linux/sock_diag.h>
23#include <linux/netfilter.h>
24#include <net/netdev_rx_queue.h>
25#include <net/xdp.h>
26#include <net/netfilter/nf_bpf_link.h>
27
28#define CREATE_TRACE_POINTS
29#include <trace/events/bpf_test_run.h>
30
31struct bpf_test_timer {
32 enum { NO_PREEMPT, NO_MIGRATE } mode;
33 u32 i;
34 u64 time_start, time_spent;
35};
36
37static void bpf_test_timer_enter(struct bpf_test_timer *t)
38 __acquires(rcu)
39{
40 rcu_read_lock();
41 if (t->mode == NO_PREEMPT)
42 preempt_disable();
43 else
44 migrate_disable();
45
46 t->time_start = ktime_get_ns();
47}
48
49static void bpf_test_timer_leave(struct bpf_test_timer *t)
50 __releases(rcu)
51{
52 t->time_start = 0;
53
54 if (t->mode == NO_PREEMPT)
55 preempt_enable();
56 else
57 migrate_enable();
58 rcu_read_unlock();
59}
60
61static bool bpf_test_timer_continue(struct bpf_test_timer *t, int iterations,
62 u32 repeat, int *err, u32 *duration)
63 __must_hold(rcu)
64{
65 t->i += iterations;
66 if (t->i >= repeat) {
67 /* We're done. */
68 t->time_spent += ktime_get_ns() - t->time_start;
69 do_div(t->time_spent, t->i);
70 *duration = t->time_spent > U32_MAX ? U32_MAX : (u32)t->time_spent;
71 *err = 0;
72 goto reset;
73 }
74
75 if (signal_pending(current)) {
76 /* During iteration: we've been cancelled, abort. */
77 *err = -EINTR;
78 goto reset;
79 }
80
81 if (need_resched()) {
82 /* During iteration: we need to reschedule between runs. */
83 t->time_spent += ktime_get_ns() - t->time_start;
84 bpf_test_timer_leave(t);
85 cond_resched();
86 bpf_test_timer_enter(t);
87 }
88
89 /* Do another round. */
90 return true;
91
92reset:
93 t->i = 0;
94 return false;
95}
96
97/* We put this struct at the head of each page with a context and frame
98 * initialised when the page is allocated, so we don't have to do this on each
99 * repetition of the test run.
100 */
101struct xdp_page_head {
102 struct xdp_buff orig_ctx;
103 struct xdp_buff ctx;
104 union {
105 /* ::data_hard_start starts here */
106 DECLARE_FLEX_ARRAY(struct xdp_frame, frame);
107 DECLARE_FLEX_ARRAY(u8, data);
108 };
109};
110
111struct xdp_test_data {
112 struct xdp_buff *orig_ctx;
113 struct xdp_rxq_info rxq;
114 struct net_device *dev;
115 struct page_pool *pp;
116 struct xdp_frame **frames;
117 struct sk_buff **skbs;
118 struct xdp_mem_info mem;
119 u32 batch_size;
120 u32 frame_cnt;
121};
122
123/* tools/testing/selftests/bpf/prog_tests/xdp_do_redirect.c:%MAX_PKT_SIZE
124 * must be updated accordingly this gets changed, otherwise BPF selftests
125 * will fail.
126 */
127#define TEST_XDP_FRAME_SIZE (PAGE_SIZE - sizeof(struct xdp_page_head))
128#define TEST_XDP_MAX_BATCH 256
129
130static void xdp_test_run_init_page(struct page *page, void *arg)
131{
132 struct xdp_page_head *head = phys_to_virt(page_to_phys(page));
133 struct xdp_buff *new_ctx, *orig_ctx;
134 u32 headroom = XDP_PACKET_HEADROOM;
135 struct xdp_test_data *xdp = arg;
136 size_t frm_len, meta_len;
137 struct xdp_frame *frm;
138 void *data;
139
140 orig_ctx = xdp->orig_ctx;
141 frm_len = orig_ctx->data_end - orig_ctx->data_meta;
142 meta_len = orig_ctx->data - orig_ctx->data_meta;
143 headroom -= meta_len;
144
145 new_ctx = &head->ctx;
146 frm = head->frame;
147 data = head->data;
148 memcpy(data + headroom, orig_ctx->data_meta, frm_len);
149
150 xdp_init_buff(new_ctx, TEST_XDP_FRAME_SIZE, &xdp->rxq);
151 xdp_prepare_buff(new_ctx, data, headroom, frm_len, true);
152 new_ctx->data = new_ctx->data_meta + meta_len;
153
154 xdp_update_frame_from_buff(new_ctx, frm);
155 frm->mem = new_ctx->rxq->mem;
156
157 memcpy(&head->orig_ctx, new_ctx, sizeof(head->orig_ctx));
158}
159
160static int xdp_test_run_setup(struct xdp_test_data *xdp, struct xdp_buff *orig_ctx)
161{
162 struct page_pool *pp;
163 int err = -ENOMEM;
164 struct page_pool_params pp_params = {
165 .order = 0,
166 .flags = 0,
167 .pool_size = xdp->batch_size,
168 .nid = NUMA_NO_NODE,
169 .init_callback = xdp_test_run_init_page,
170 .init_arg = xdp,
171 };
172
173 xdp->frames = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
174 if (!xdp->frames)
175 return -ENOMEM;
176
177 xdp->skbs = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
178 if (!xdp->skbs)
179 goto err_skbs;
180
181 pp = page_pool_create(&pp_params);
182 if (IS_ERR(pp)) {
183 err = PTR_ERR(pp);
184 goto err_pp;
185 }
186
187 /* will copy 'mem.id' into pp->xdp_mem_id */
188 err = xdp_reg_mem_model(&xdp->mem, MEM_TYPE_PAGE_POOL, pp);
189 if (err)
190 goto err_mmodel;
191
192 xdp->pp = pp;
193
194 /* We create a 'fake' RXQ referencing the original dev, but with an
195 * xdp_mem_info pointing to our page_pool
196 */
197 xdp_rxq_info_reg(&xdp->rxq, orig_ctx->rxq->dev, 0, 0);
198 xdp->rxq.mem.type = MEM_TYPE_PAGE_POOL;
199 xdp->rxq.mem.id = pp->xdp_mem_id;
200 xdp->dev = orig_ctx->rxq->dev;
201 xdp->orig_ctx = orig_ctx;
202
203 return 0;
204
205err_mmodel:
206 page_pool_destroy(pp);
207err_pp:
208 kvfree(xdp->skbs);
209err_skbs:
210 kvfree(xdp->frames);
211 return err;
212}
213
214static void xdp_test_run_teardown(struct xdp_test_data *xdp)
215{
216 xdp_unreg_mem_model(&xdp->mem);
217 page_pool_destroy(xdp->pp);
218 kfree(xdp->frames);
219 kfree(xdp->skbs);
220}
221
222static bool frame_was_changed(const struct xdp_page_head *head)
223{
224 /* xdp_scrub_frame() zeroes the data pointer, flags is the last field,
225 * i.e. has the highest chances to be overwritten. If those two are
226 * untouched, it's most likely safe to skip the context reset.
227 */
228 return head->frame->data != head->orig_ctx.data ||
229 head->frame->flags != head->orig_ctx.flags;
230}
231
232static bool ctx_was_changed(struct xdp_page_head *head)
233{
234 return head->orig_ctx.data != head->ctx.data ||
235 head->orig_ctx.data_meta != head->ctx.data_meta ||
236 head->orig_ctx.data_end != head->ctx.data_end;
237}
238
239static void reset_ctx(struct xdp_page_head *head)
240{
241 if (likely(!frame_was_changed(head) && !ctx_was_changed(head)))
242 return;
243
244 head->ctx.data = head->orig_ctx.data;
245 head->ctx.data_meta = head->orig_ctx.data_meta;
246 head->ctx.data_end = head->orig_ctx.data_end;
247 xdp_update_frame_from_buff(&head->ctx, head->frame);
248}
249
250static int xdp_recv_frames(struct xdp_frame **frames, int nframes,
251 struct sk_buff **skbs,
252 struct net_device *dev)
253{
254 gfp_t gfp = __GFP_ZERO | GFP_ATOMIC;
255 int i, n;
256 LIST_HEAD(list);
257
258 n = kmem_cache_alloc_bulk(net_hotdata.skbuff_cache, gfp, nframes,
259 (void **)skbs);
260 if (unlikely(n == 0)) {
261 for (i = 0; i < nframes; i++)
262 xdp_return_frame(frames[i]);
263 return -ENOMEM;
264 }
265
266 for (i = 0; i < nframes; i++) {
267 struct xdp_frame *xdpf = frames[i];
268 struct sk_buff *skb = skbs[i];
269
270 skb = __xdp_build_skb_from_frame(xdpf, skb, dev);
271 if (!skb) {
272 xdp_return_frame(xdpf);
273 continue;
274 }
275
276 list_add_tail(&skb->list, &list);
277 }
278 netif_receive_skb_list(&list);
279
280 return 0;
281}
282
283static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
284 u32 repeat)
285{
286 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
287 int err = 0, act, ret, i, nframes = 0, batch_sz;
288 struct xdp_frame **frames = xdp->frames;
289 struct xdp_page_head *head;
290 struct xdp_frame *frm;
291 bool redirect = false;
292 struct xdp_buff *ctx;
293 struct page *page;
294
295 batch_sz = min_t(u32, repeat, xdp->batch_size);
296
297 local_bh_disable();
298 xdp_set_return_frame_no_direct();
299
300 for (i = 0; i < batch_sz; i++) {
301 page = page_pool_dev_alloc_pages(xdp->pp);
302 if (!page) {
303 err = -ENOMEM;
304 goto out;
305 }
306
307 head = phys_to_virt(page_to_phys(page));
308 reset_ctx(head);
309 ctx = &head->ctx;
310 frm = head->frame;
311 xdp->frame_cnt++;
312
313 act = bpf_prog_run_xdp(prog, ctx);
314
315 /* if program changed pkt bounds we need to update the xdp_frame */
316 if (unlikely(ctx_was_changed(head))) {
317 ret = xdp_update_frame_from_buff(ctx, frm);
318 if (ret) {
319 xdp_return_buff(ctx);
320 continue;
321 }
322 }
323
324 switch (act) {
325 case XDP_TX:
326 /* we can't do a real XDP_TX since we're not in the
327 * driver, so turn it into a REDIRECT back to the same
328 * index
329 */
330 ri->tgt_index = xdp->dev->ifindex;
331 ri->map_id = INT_MAX;
332 ri->map_type = BPF_MAP_TYPE_UNSPEC;
333 fallthrough;
334 case XDP_REDIRECT:
335 redirect = true;
336 ret = xdp_do_redirect_frame(xdp->dev, ctx, frm, prog);
337 if (ret)
338 xdp_return_buff(ctx);
339 break;
340 case XDP_PASS:
341 frames[nframes++] = frm;
342 break;
343 default:
344 bpf_warn_invalid_xdp_action(NULL, prog, act);
345 fallthrough;
346 case XDP_DROP:
347 xdp_return_buff(ctx);
348 break;
349 }
350 }
351
352out:
353 if (redirect)
354 xdp_do_flush();
355 if (nframes) {
356 ret = xdp_recv_frames(frames, nframes, xdp->skbs, xdp->dev);
357 if (ret)
358 err = ret;
359 }
360
361 xdp_clear_return_frame_no_direct();
362 local_bh_enable();
363 return err;
364}
365
366static int bpf_test_run_xdp_live(struct bpf_prog *prog, struct xdp_buff *ctx,
367 u32 repeat, u32 batch_size, u32 *time)
368
369{
370 struct xdp_test_data xdp = { .batch_size = batch_size };
371 struct bpf_test_timer t = { .mode = NO_MIGRATE };
372 int ret;
373
374 if (!repeat)
375 repeat = 1;
376
377 ret = xdp_test_run_setup(&xdp, ctx);
378 if (ret)
379 return ret;
380
381 bpf_test_timer_enter(&t);
382 do {
383 xdp.frame_cnt = 0;
384 ret = xdp_test_run_batch(&xdp, prog, repeat - t.i);
385 if (unlikely(ret < 0))
386 break;
387 } while (bpf_test_timer_continue(&t, xdp.frame_cnt, repeat, &ret, time));
388 bpf_test_timer_leave(&t);
389
390 xdp_test_run_teardown(&xdp);
391 return ret;
392}
393
394static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
395 u32 *retval, u32 *time, bool xdp)
396{
397 struct bpf_prog_array_item item = {.prog = prog};
398 struct bpf_run_ctx *old_ctx;
399 struct bpf_cg_run_ctx run_ctx;
400 struct bpf_test_timer t = { NO_MIGRATE };
401 enum bpf_cgroup_storage_type stype;
402 int ret;
403
404 for_each_cgroup_storage_type(stype) {
405 item.cgroup_storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
406 if (IS_ERR(item.cgroup_storage[stype])) {
407 item.cgroup_storage[stype] = NULL;
408 for_each_cgroup_storage_type(stype)
409 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
410 return -ENOMEM;
411 }
412 }
413
414 if (!repeat)
415 repeat = 1;
416
417 bpf_test_timer_enter(&t);
418 old_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
419 do {
420 run_ctx.prog_item = &item;
421 local_bh_disable();
422 if (xdp)
423 *retval = bpf_prog_run_xdp(prog, ctx);
424 else
425 *retval = bpf_prog_run(prog, ctx);
426 local_bh_enable();
427 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, time));
428 bpf_reset_run_ctx(old_ctx);
429 bpf_test_timer_leave(&t);
430
431 for_each_cgroup_storage_type(stype)
432 bpf_cgroup_storage_free(item.cgroup_storage[stype]);
433
434 return ret;
435}
436
437static int bpf_test_finish(const union bpf_attr *kattr,
438 union bpf_attr __user *uattr, const void *data,
439 struct skb_shared_info *sinfo, u32 size,
440 u32 retval, u32 duration)
441{
442 void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
443 int err = -EFAULT;
444 u32 copy_size = size;
445
446 /* Clamp copy if the user has provided a size hint, but copy the full
447 * buffer if not to retain old behaviour.
448 */
449 if (kattr->test.data_size_out &&
450 copy_size > kattr->test.data_size_out) {
451 copy_size = kattr->test.data_size_out;
452 err = -ENOSPC;
453 }
454
455 if (data_out) {
456 int len = sinfo ? copy_size - sinfo->xdp_frags_size : copy_size;
457
458 if (len < 0) {
459 err = -ENOSPC;
460 goto out;
461 }
462
463 if (copy_to_user(data_out, data, len))
464 goto out;
465
466 if (sinfo) {
467 int i, offset = len;
468 u32 data_len;
469
470 for (i = 0; i < sinfo->nr_frags; i++) {
471 skb_frag_t *frag = &sinfo->frags[i];
472
473 if (offset >= copy_size) {
474 err = -ENOSPC;
475 break;
476 }
477
478 data_len = min_t(u32, copy_size - offset,
479 skb_frag_size(frag));
480
481 if (copy_to_user(data_out + offset,
482 skb_frag_address(frag),
483 data_len))
484 goto out;
485
486 offset += data_len;
487 }
488 }
489 }
490
491 if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
492 goto out;
493 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
494 goto out;
495 if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
496 goto out;
497 if (err != -ENOSPC)
498 err = 0;
499out:
500 trace_bpf_test_finish(&err);
501 return err;
502}
503
504/* Integer types of various sizes and pointer combinations cover variety of
505 * architecture dependent calling conventions. 7+ can be supported in the
506 * future.
507 */
508__bpf_kfunc_start_defs();
509
510__bpf_kfunc int bpf_fentry_test1(int a)
511{
512 return a + 1;
513}
514EXPORT_SYMBOL_GPL(bpf_fentry_test1);
515
516int noinline bpf_fentry_test2(int a, u64 b)
517{
518 return a + b;
519}
520
521int noinline bpf_fentry_test3(char a, int b, u64 c)
522{
523 return a + b + c;
524}
525
526int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
527{
528 return (long)a + b + c + d;
529}
530
531int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
532{
533 return a + (long)b + c + d + e;
534}
535
536int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
537{
538 return a + (long)b + c + d + (long)e + f;
539}
540
541struct bpf_fentry_test_t {
542 struct bpf_fentry_test_t *a;
543};
544
545int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
546{
547 asm volatile ("": "+r"(arg));
548 return (long)arg;
549}
550
551int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
552{
553 return (long)arg->a;
554}
555
556__bpf_kfunc u32 bpf_fentry_test9(u32 *a)
557{
558 return *a;
559}
560
561void noinline bpf_fentry_test_sinfo(struct skb_shared_info *sinfo)
562{
563}
564
565__bpf_kfunc int bpf_modify_return_test(int a, int *b)
566{
567 *b += 1;
568 return a + *b;
569}
570
571__bpf_kfunc int bpf_modify_return_test2(int a, int *b, short c, int d,
572 void *e, char f, int g)
573{
574 *b += 1;
575 return a + *b + c + d + (long)e + f + g;
576}
577
578int noinline bpf_fentry_shadow_test(int a)
579{
580 return a + 1;
581}
582
583struct prog_test_member1 {
584 int a;
585};
586
587struct prog_test_member {
588 struct prog_test_member1 m;
589 int c;
590};
591
592struct prog_test_ref_kfunc {
593 int a;
594 int b;
595 struct prog_test_member memb;
596 struct prog_test_ref_kfunc *next;
597 refcount_t cnt;
598};
599
600__bpf_kfunc void bpf_kfunc_call_test_release(struct prog_test_ref_kfunc *p)
601{
602 refcount_dec(&p->cnt);
603}
604
605__bpf_kfunc void bpf_kfunc_call_test_release_dtor(void *p)
606{
607 bpf_kfunc_call_test_release(p);
608}
609CFI_NOSEAL(bpf_kfunc_call_test_release_dtor);
610
611__bpf_kfunc void bpf_kfunc_call_memb_release(struct prog_test_member *p)
612{
613}
614
615__bpf_kfunc void bpf_kfunc_call_memb_release_dtor(void *p)
616{
617}
618CFI_NOSEAL(bpf_kfunc_call_memb_release_dtor);
619
620__bpf_kfunc_end_defs();
621
622BTF_KFUNCS_START(bpf_test_modify_return_ids)
623BTF_ID_FLAGS(func, bpf_modify_return_test)
624BTF_ID_FLAGS(func, bpf_modify_return_test2)
625BTF_ID_FLAGS(func, bpf_fentry_test1, KF_SLEEPABLE)
626BTF_KFUNCS_END(bpf_test_modify_return_ids)
627
628static const struct btf_kfunc_id_set bpf_test_modify_return_set = {
629 .owner = THIS_MODULE,
630 .set = &bpf_test_modify_return_ids,
631};
632
633BTF_KFUNCS_START(test_sk_check_kfunc_ids)
634BTF_ID_FLAGS(func, bpf_kfunc_call_test_release, KF_RELEASE)
635BTF_ID_FLAGS(func, bpf_kfunc_call_memb_release, KF_RELEASE)
636BTF_KFUNCS_END(test_sk_check_kfunc_ids)
637
638static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
639 u32 size, u32 headroom, u32 tailroom)
640{
641 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
642 void *data;
643
644 if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
645 return ERR_PTR(-EINVAL);
646
647 if (user_size > size)
648 return ERR_PTR(-EMSGSIZE);
649
650 size = SKB_DATA_ALIGN(size);
651 data = kzalloc(size + headroom + tailroom, GFP_USER);
652 if (!data)
653 return ERR_PTR(-ENOMEM);
654
655 if (copy_from_user(data + headroom, data_in, user_size)) {
656 kfree(data);
657 return ERR_PTR(-EFAULT);
658 }
659
660 return data;
661}
662
663int bpf_prog_test_run_tracing(struct bpf_prog *prog,
664 const union bpf_attr *kattr,
665 union bpf_attr __user *uattr)
666{
667 struct bpf_fentry_test_t arg = {};
668 u16 side_effect = 0, ret = 0;
669 int b = 2, err = -EFAULT;
670 u32 retval = 0;
671
672 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
673 return -EINVAL;
674
675 switch (prog->expected_attach_type) {
676 case BPF_TRACE_FENTRY:
677 case BPF_TRACE_FEXIT:
678 if (bpf_fentry_test1(1) != 2 ||
679 bpf_fentry_test2(2, 3) != 5 ||
680 bpf_fentry_test3(4, 5, 6) != 15 ||
681 bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
682 bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
683 bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
684 bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
685 bpf_fentry_test8(&arg) != 0 ||
686 bpf_fentry_test9(&retval) != 0)
687 goto out;
688 break;
689 case BPF_MODIFY_RETURN:
690 ret = bpf_modify_return_test(1, &b);
691 if (b != 2)
692 side_effect++;
693 b = 2;
694 ret += bpf_modify_return_test2(1, &b, 3, 4, (void *)5, 6, 7);
695 if (b != 2)
696 side_effect++;
697 break;
698 default:
699 goto out;
700 }
701
702 retval = ((u32)side_effect << 16) | ret;
703 if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
704 goto out;
705
706 err = 0;
707out:
708 trace_bpf_test_finish(&err);
709 return err;
710}
711
712struct bpf_raw_tp_test_run_info {
713 struct bpf_prog *prog;
714 void *ctx;
715 u32 retval;
716};
717
718static void
719__bpf_prog_test_run_raw_tp(void *data)
720{
721 struct bpf_raw_tp_test_run_info *info = data;
722
723 rcu_read_lock();
724 info->retval = bpf_prog_run(info->prog, info->ctx);
725 rcu_read_unlock();
726}
727
728int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
729 const union bpf_attr *kattr,
730 union bpf_attr __user *uattr)
731{
732 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
733 __u32 ctx_size_in = kattr->test.ctx_size_in;
734 struct bpf_raw_tp_test_run_info info;
735 int cpu = kattr->test.cpu, err = 0;
736 int current_cpu;
737
738 /* doesn't support data_in/out, ctx_out, duration, or repeat */
739 if (kattr->test.data_in || kattr->test.data_out ||
740 kattr->test.ctx_out || kattr->test.duration ||
741 kattr->test.repeat || kattr->test.batch_size)
742 return -EINVAL;
743
744 if (ctx_size_in < prog->aux->max_ctx_offset ||
745 ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
746 return -EINVAL;
747
748 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
749 return -EINVAL;
750
751 if (ctx_size_in) {
752 info.ctx = memdup_user(ctx_in, ctx_size_in);
753 if (IS_ERR(info.ctx))
754 return PTR_ERR(info.ctx);
755 } else {
756 info.ctx = NULL;
757 }
758
759 info.prog = prog;
760
761 current_cpu = get_cpu();
762 if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
763 cpu == current_cpu) {
764 __bpf_prog_test_run_raw_tp(&info);
765 } else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
766 /* smp_call_function_single() also checks cpu_online()
767 * after csd_lock(). However, since cpu is from user
768 * space, let's do an extra quick check to filter out
769 * invalid value before smp_call_function_single().
770 */
771 err = -ENXIO;
772 } else {
773 err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
774 &info, 1);
775 }
776 put_cpu();
777
778 if (!err &&
779 copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
780 err = -EFAULT;
781
782 kfree(info.ctx);
783 return err;
784}
785
786static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
787{
788 void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
789 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
790 u32 size = kattr->test.ctx_size_in;
791 void *data;
792 int err;
793
794 if (!data_in && !data_out)
795 return NULL;
796
797 data = kzalloc(max_size, GFP_USER);
798 if (!data)
799 return ERR_PTR(-ENOMEM);
800
801 if (data_in) {
802 err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
803 if (err) {
804 kfree(data);
805 return ERR_PTR(err);
806 }
807
808 size = min_t(u32, max_size, size);
809 if (copy_from_user(data, data_in, size)) {
810 kfree(data);
811 return ERR_PTR(-EFAULT);
812 }
813 }
814 return data;
815}
816
817static int bpf_ctx_finish(const union bpf_attr *kattr,
818 union bpf_attr __user *uattr, const void *data,
819 u32 size)
820{
821 void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
822 int err = -EFAULT;
823 u32 copy_size = size;
824
825 if (!data || !data_out)
826 return 0;
827
828 if (copy_size > kattr->test.ctx_size_out) {
829 copy_size = kattr->test.ctx_size_out;
830 err = -ENOSPC;
831 }
832
833 if (copy_to_user(data_out, data, copy_size))
834 goto out;
835 if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
836 goto out;
837 if (err != -ENOSPC)
838 err = 0;
839out:
840 return err;
841}
842
843/**
844 * range_is_zero - test whether buffer is initialized
845 * @buf: buffer to check
846 * @from: check from this position
847 * @to: check up until (excluding) this position
848 *
849 * This function returns true if the there is a non-zero byte
850 * in the buf in the range [from,to).
851 */
852static inline bool range_is_zero(void *buf, size_t from, size_t to)
853{
854 return !memchr_inv((u8 *)buf + from, 0, to - from);
855}
856
857static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
858{
859 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
860
861 if (!__skb)
862 return 0;
863
864 /* make sure the fields we don't use are zeroed */
865 if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
866 return -EINVAL;
867
868 /* mark is allowed */
869
870 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
871 offsetof(struct __sk_buff, priority)))
872 return -EINVAL;
873
874 /* priority is allowed */
875 /* ingress_ifindex is allowed */
876 /* ifindex is allowed */
877
878 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
879 offsetof(struct __sk_buff, cb)))
880 return -EINVAL;
881
882 /* cb is allowed */
883
884 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
885 offsetof(struct __sk_buff, tstamp)))
886 return -EINVAL;
887
888 /* tstamp is allowed */
889 /* wire_len is allowed */
890 /* gso_segs is allowed */
891
892 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
893 offsetof(struct __sk_buff, gso_size)))
894 return -EINVAL;
895
896 /* gso_size is allowed */
897
898 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
899 offsetof(struct __sk_buff, hwtstamp)))
900 return -EINVAL;
901
902 /* hwtstamp is allowed */
903
904 if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp),
905 sizeof(struct __sk_buff)))
906 return -EINVAL;
907
908 skb->mark = __skb->mark;
909 skb->priority = __skb->priority;
910 skb->skb_iif = __skb->ingress_ifindex;
911 skb->tstamp = __skb->tstamp;
912 memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
913
914 if (__skb->wire_len == 0) {
915 cb->pkt_len = skb->len;
916 } else {
917 if (__skb->wire_len < skb->len ||
918 __skb->wire_len > GSO_LEGACY_MAX_SIZE)
919 return -EINVAL;
920 cb->pkt_len = __skb->wire_len;
921 }
922
923 if (__skb->gso_segs > GSO_MAX_SEGS)
924 return -EINVAL;
925 skb_shinfo(skb)->gso_segs = __skb->gso_segs;
926 skb_shinfo(skb)->gso_size = __skb->gso_size;
927 skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp;
928
929 return 0;
930}
931
932static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
933{
934 struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
935
936 if (!__skb)
937 return;
938
939 __skb->mark = skb->mark;
940 __skb->priority = skb->priority;
941 __skb->ingress_ifindex = skb->skb_iif;
942 __skb->ifindex = skb->dev->ifindex;
943 __skb->tstamp = skb->tstamp;
944 memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
945 __skb->wire_len = cb->pkt_len;
946 __skb->gso_segs = skb_shinfo(skb)->gso_segs;
947 __skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp;
948}
949
950static struct proto bpf_dummy_proto = {
951 .name = "bpf_dummy",
952 .owner = THIS_MODULE,
953 .obj_size = sizeof(struct sock),
954};
955
956int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
957 union bpf_attr __user *uattr)
958{
959 bool is_l2 = false, is_direct_pkt_access = false;
960 struct net *net = current->nsproxy->net_ns;
961 struct net_device *dev = net->loopback_dev;
962 u32 size = kattr->test.data_size_in;
963 u32 repeat = kattr->test.repeat;
964 struct __sk_buff *ctx = NULL;
965 u32 retval, duration;
966 int hh_len = ETH_HLEN;
967 struct sk_buff *skb;
968 struct sock *sk;
969 void *data;
970 int ret;
971
972 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
973 return -EINVAL;
974
975 data = bpf_test_init(kattr, kattr->test.data_size_in,
976 size, NET_SKB_PAD + NET_IP_ALIGN,
977 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
978 if (IS_ERR(data))
979 return PTR_ERR(data);
980
981 ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
982 if (IS_ERR(ctx)) {
983 kfree(data);
984 return PTR_ERR(ctx);
985 }
986
987 switch (prog->type) {
988 case BPF_PROG_TYPE_SCHED_CLS:
989 case BPF_PROG_TYPE_SCHED_ACT:
990 is_l2 = true;
991 fallthrough;
992 case BPF_PROG_TYPE_LWT_IN:
993 case BPF_PROG_TYPE_LWT_OUT:
994 case BPF_PROG_TYPE_LWT_XMIT:
995 is_direct_pkt_access = true;
996 break;
997 default:
998 break;
999 }
1000
1001 sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
1002 if (!sk) {
1003 kfree(data);
1004 kfree(ctx);
1005 return -ENOMEM;
1006 }
1007 sock_init_data(NULL, sk);
1008
1009 skb = slab_build_skb(data);
1010 if (!skb) {
1011 kfree(data);
1012 kfree(ctx);
1013 sk_free(sk);
1014 return -ENOMEM;
1015 }
1016 skb->sk = sk;
1017
1018 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1019 __skb_put(skb, size);
1020 if (ctx && ctx->ifindex > 1) {
1021 dev = dev_get_by_index(net, ctx->ifindex);
1022 if (!dev) {
1023 ret = -ENODEV;
1024 goto out;
1025 }
1026 }
1027 skb->protocol = eth_type_trans(skb, dev);
1028 skb_reset_network_header(skb);
1029
1030 switch (skb->protocol) {
1031 case htons(ETH_P_IP):
1032 sk->sk_family = AF_INET;
1033 if (sizeof(struct iphdr) <= skb_headlen(skb)) {
1034 sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
1035 sk->sk_daddr = ip_hdr(skb)->daddr;
1036 }
1037 break;
1038#if IS_ENABLED(CONFIG_IPV6)
1039 case htons(ETH_P_IPV6):
1040 sk->sk_family = AF_INET6;
1041 if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
1042 sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
1043 sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
1044 }
1045 break;
1046#endif
1047 default:
1048 break;
1049 }
1050
1051 if (is_l2)
1052 __skb_push(skb, hh_len);
1053 if (is_direct_pkt_access)
1054 bpf_compute_data_pointers(skb);
1055 ret = convert___skb_to_skb(skb, ctx);
1056 if (ret)
1057 goto out;
1058 ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
1059 if (ret)
1060 goto out;
1061 if (!is_l2) {
1062 if (skb_headroom(skb) < hh_len) {
1063 int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
1064
1065 if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
1066 ret = -ENOMEM;
1067 goto out;
1068 }
1069 }
1070 memset(__skb_push(skb, hh_len), 0, hh_len);
1071 }
1072 convert_skb_to___skb(skb, ctx);
1073
1074 size = skb->len;
1075 /* bpf program can never convert linear skb to non-linear */
1076 if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
1077 size = skb_headlen(skb);
1078 ret = bpf_test_finish(kattr, uattr, skb->data, NULL, size, retval,
1079 duration);
1080 if (!ret)
1081 ret = bpf_ctx_finish(kattr, uattr, ctx,
1082 sizeof(struct __sk_buff));
1083out:
1084 if (dev && dev != net->loopback_dev)
1085 dev_put(dev);
1086 kfree_skb(skb);
1087 sk_free(sk);
1088 kfree(ctx);
1089 return ret;
1090}
1091
1092static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp)
1093{
1094 unsigned int ingress_ifindex, rx_queue_index;
1095 struct netdev_rx_queue *rxqueue;
1096 struct net_device *device;
1097
1098 if (!xdp_md)
1099 return 0;
1100
1101 if (xdp_md->egress_ifindex != 0)
1102 return -EINVAL;
1103
1104 ingress_ifindex = xdp_md->ingress_ifindex;
1105 rx_queue_index = xdp_md->rx_queue_index;
1106
1107 if (!ingress_ifindex && rx_queue_index)
1108 return -EINVAL;
1109
1110 if (ingress_ifindex) {
1111 device = dev_get_by_index(current->nsproxy->net_ns,
1112 ingress_ifindex);
1113 if (!device)
1114 return -ENODEV;
1115
1116 if (rx_queue_index >= device->real_num_rx_queues)
1117 goto free_dev;
1118
1119 rxqueue = __netif_get_rx_queue(device, rx_queue_index);
1120
1121 if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq))
1122 goto free_dev;
1123
1124 xdp->rxq = &rxqueue->xdp_rxq;
1125 /* The device is now tracked in the xdp->rxq for later
1126 * dev_put()
1127 */
1128 }
1129
1130 xdp->data = xdp->data_meta + xdp_md->data;
1131 return 0;
1132
1133free_dev:
1134 dev_put(device);
1135 return -EINVAL;
1136}
1137
1138static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md)
1139{
1140 if (!xdp_md)
1141 return;
1142
1143 xdp_md->data = xdp->data - xdp->data_meta;
1144 xdp_md->data_end = xdp->data_end - xdp->data_meta;
1145
1146 if (xdp_md->ingress_ifindex)
1147 dev_put(xdp->rxq->dev);
1148}
1149
1150int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
1151 union bpf_attr __user *uattr)
1152{
1153 bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES);
1154 u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1155 u32 batch_size = kattr->test.batch_size;
1156 u32 retval = 0, duration, max_data_sz;
1157 u32 size = kattr->test.data_size_in;
1158 u32 headroom = XDP_PACKET_HEADROOM;
1159 u32 repeat = kattr->test.repeat;
1160 struct netdev_rx_queue *rxqueue;
1161 struct skb_shared_info *sinfo;
1162 struct xdp_buff xdp = {};
1163 int i, ret = -EINVAL;
1164 struct xdp_md *ctx;
1165 void *data;
1166
1167 if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
1168 prog->expected_attach_type == BPF_XDP_CPUMAP)
1169 return -EINVAL;
1170
1171 if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES)
1172 return -EINVAL;
1173
1174 if (bpf_prog_is_dev_bound(prog->aux))
1175 return -EINVAL;
1176
1177 if (do_live) {
1178 if (!batch_size)
1179 batch_size = NAPI_POLL_WEIGHT;
1180 else if (batch_size > TEST_XDP_MAX_BATCH)
1181 return -E2BIG;
1182
1183 headroom += sizeof(struct xdp_page_head);
1184 } else if (batch_size) {
1185 return -EINVAL;
1186 }
1187
1188 ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md));
1189 if (IS_ERR(ctx))
1190 return PTR_ERR(ctx);
1191
1192 if (ctx) {
1193 /* There can't be user provided data before the meta data */
1194 if (ctx->data_meta || ctx->data_end != size ||
1195 ctx->data > ctx->data_end ||
1196 unlikely(xdp_metalen_invalid(ctx->data)) ||
1197 (do_live && (kattr->test.data_out || kattr->test.ctx_out)))
1198 goto free_ctx;
1199 /* Meta data is allocated from the headroom */
1200 headroom -= ctx->data;
1201 }
1202
1203 max_data_sz = 4096 - headroom - tailroom;
1204 if (size > max_data_sz) {
1205 /* disallow live data mode for jumbo frames */
1206 if (do_live)
1207 goto free_ctx;
1208 size = max_data_sz;
1209 }
1210
1211 data = bpf_test_init(kattr, size, max_data_sz, headroom, tailroom);
1212 if (IS_ERR(data)) {
1213 ret = PTR_ERR(data);
1214 goto free_ctx;
1215 }
1216
1217 rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
1218 rxqueue->xdp_rxq.frag_size = headroom + max_data_sz + tailroom;
1219 xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq);
1220 xdp_prepare_buff(&xdp, data, headroom, size, true);
1221 sinfo = xdp_get_shared_info_from_buff(&xdp);
1222
1223 ret = xdp_convert_md_to_buff(ctx, &xdp);
1224 if (ret)
1225 goto free_data;
1226
1227 if (unlikely(kattr->test.data_size_in > size)) {
1228 void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
1229
1230 while (size < kattr->test.data_size_in) {
1231 struct page *page;
1232 skb_frag_t *frag;
1233 u32 data_len;
1234
1235 if (sinfo->nr_frags == MAX_SKB_FRAGS) {
1236 ret = -ENOMEM;
1237 goto out;
1238 }
1239
1240 page = alloc_page(GFP_KERNEL);
1241 if (!page) {
1242 ret = -ENOMEM;
1243 goto out;
1244 }
1245
1246 frag = &sinfo->frags[sinfo->nr_frags++];
1247
1248 data_len = min_t(u32, kattr->test.data_size_in - size,
1249 PAGE_SIZE);
1250 skb_frag_fill_page_desc(frag, page, 0, data_len);
1251
1252 if (copy_from_user(page_address(page), data_in + size,
1253 data_len)) {
1254 ret = -EFAULT;
1255 goto out;
1256 }
1257 sinfo->xdp_frags_size += data_len;
1258 size += data_len;
1259 }
1260 xdp_buff_set_frags_flag(&xdp);
1261 }
1262
1263 if (repeat > 1)
1264 bpf_prog_change_xdp(NULL, prog);
1265
1266 if (do_live)
1267 ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration);
1268 else
1269 ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
1270 /* We convert the xdp_buff back to an xdp_md before checking the return
1271 * code so the reference count of any held netdevice will be decremented
1272 * even if the test run failed.
1273 */
1274 xdp_convert_buff_to_md(&xdp, ctx);
1275 if (ret)
1276 goto out;
1277
1278 size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size;
1279 ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size,
1280 retval, duration);
1281 if (!ret)
1282 ret = bpf_ctx_finish(kattr, uattr, ctx,
1283 sizeof(struct xdp_md));
1284
1285out:
1286 if (repeat > 1)
1287 bpf_prog_change_xdp(prog, NULL);
1288free_data:
1289 for (i = 0; i < sinfo->nr_frags; i++)
1290 __free_page(skb_frag_page(&sinfo->frags[i]));
1291 kfree(data);
1292free_ctx:
1293 kfree(ctx);
1294 return ret;
1295}
1296
1297static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
1298{
1299 /* make sure the fields we don't use are zeroed */
1300 if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
1301 return -EINVAL;
1302
1303 /* flags is allowed */
1304
1305 if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
1306 sizeof(struct bpf_flow_keys)))
1307 return -EINVAL;
1308
1309 return 0;
1310}
1311
1312int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1313 const union bpf_attr *kattr,
1314 union bpf_attr __user *uattr)
1315{
1316 struct bpf_test_timer t = { NO_PREEMPT };
1317 u32 size = kattr->test.data_size_in;
1318 struct bpf_flow_dissector ctx = {};
1319 u32 repeat = kattr->test.repeat;
1320 struct bpf_flow_keys *user_ctx;
1321 struct bpf_flow_keys flow_keys;
1322 const struct ethhdr *eth;
1323 unsigned int flags = 0;
1324 u32 retval, duration;
1325 void *data;
1326 int ret;
1327
1328 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1329 return -EINVAL;
1330
1331 if (size < ETH_HLEN)
1332 return -EINVAL;
1333
1334 data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0);
1335 if (IS_ERR(data))
1336 return PTR_ERR(data);
1337
1338 eth = (struct ethhdr *)data;
1339
1340 if (!repeat)
1341 repeat = 1;
1342
1343 user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
1344 if (IS_ERR(user_ctx)) {
1345 kfree(data);
1346 return PTR_ERR(user_ctx);
1347 }
1348 if (user_ctx) {
1349 ret = verify_user_bpf_flow_keys(user_ctx);
1350 if (ret)
1351 goto out;
1352 flags = user_ctx->flags;
1353 }
1354
1355 ctx.flow_keys = &flow_keys;
1356 ctx.data = data;
1357 ctx.data_end = (__u8 *)data + size;
1358
1359 bpf_test_timer_enter(&t);
1360 do {
1361 retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
1362 size, flags);
1363 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1364 bpf_test_timer_leave(&t);
1365
1366 if (ret < 0)
1367 goto out;
1368
1369 ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL,
1370 sizeof(flow_keys), retval, duration);
1371 if (!ret)
1372 ret = bpf_ctx_finish(kattr, uattr, user_ctx,
1373 sizeof(struct bpf_flow_keys));
1374
1375out:
1376 kfree(user_ctx);
1377 kfree(data);
1378 return ret;
1379}
1380
1381int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
1382 union bpf_attr __user *uattr)
1383{
1384 struct bpf_test_timer t = { NO_PREEMPT };
1385 struct bpf_prog_array *progs = NULL;
1386 struct bpf_sk_lookup_kern ctx = {};
1387 u32 repeat = kattr->test.repeat;
1388 struct bpf_sk_lookup *user_ctx;
1389 u32 retval, duration;
1390 int ret = -EINVAL;
1391
1392 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1393 return -EINVAL;
1394
1395 if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
1396 kattr->test.data_size_out)
1397 return -EINVAL;
1398
1399 if (!repeat)
1400 repeat = 1;
1401
1402 user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
1403 if (IS_ERR(user_ctx))
1404 return PTR_ERR(user_ctx);
1405
1406 if (!user_ctx)
1407 return -EINVAL;
1408
1409 if (user_ctx->sk)
1410 goto out;
1411
1412 if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
1413 goto out;
1414
1415 if (user_ctx->local_port > U16_MAX) {
1416 ret = -ERANGE;
1417 goto out;
1418 }
1419
1420 ctx.family = (u16)user_ctx->family;
1421 ctx.protocol = (u16)user_ctx->protocol;
1422 ctx.dport = (u16)user_ctx->local_port;
1423 ctx.sport = user_ctx->remote_port;
1424
1425 switch (ctx.family) {
1426 case AF_INET:
1427 ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
1428 ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
1429 break;
1430
1431#if IS_ENABLED(CONFIG_IPV6)
1432 case AF_INET6:
1433 ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
1434 ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
1435 break;
1436#endif
1437
1438 default:
1439 ret = -EAFNOSUPPORT;
1440 goto out;
1441 }
1442
1443 progs = bpf_prog_array_alloc(1, GFP_KERNEL);
1444 if (!progs) {
1445 ret = -ENOMEM;
1446 goto out;
1447 }
1448
1449 progs->items[0].prog = prog;
1450
1451 bpf_test_timer_enter(&t);
1452 do {
1453 ctx.selected_sk = NULL;
1454 retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run);
1455 } while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1456 bpf_test_timer_leave(&t);
1457
1458 if (ret < 0)
1459 goto out;
1460
1461 user_ctx->cookie = 0;
1462 if (ctx.selected_sk) {
1463 if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
1464 ret = -EOPNOTSUPP;
1465 goto out;
1466 }
1467
1468 user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
1469 }
1470
1471 ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1472 if (!ret)
1473 ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
1474
1475out:
1476 bpf_prog_array_free(progs);
1477 kfree(user_ctx);
1478 return ret;
1479}
1480
1481int bpf_prog_test_run_syscall(struct bpf_prog *prog,
1482 const union bpf_attr *kattr,
1483 union bpf_attr __user *uattr)
1484{
1485 void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
1486 __u32 ctx_size_in = kattr->test.ctx_size_in;
1487 void *ctx = NULL;
1488 u32 retval;
1489 int err = 0;
1490
1491 /* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
1492 if (kattr->test.data_in || kattr->test.data_out ||
1493 kattr->test.ctx_out || kattr->test.duration ||
1494 kattr->test.repeat || kattr->test.flags ||
1495 kattr->test.batch_size)
1496 return -EINVAL;
1497
1498 if (ctx_size_in < prog->aux->max_ctx_offset ||
1499 ctx_size_in > U16_MAX)
1500 return -EINVAL;
1501
1502 if (ctx_size_in) {
1503 ctx = memdup_user(ctx_in, ctx_size_in);
1504 if (IS_ERR(ctx))
1505 return PTR_ERR(ctx);
1506 }
1507
1508 rcu_read_lock_trace();
1509 retval = bpf_prog_run_pin_on_cpu(prog, ctx);
1510 rcu_read_unlock_trace();
1511
1512 if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
1513 err = -EFAULT;
1514 goto out;
1515 }
1516 if (ctx_size_in)
1517 if (copy_to_user(ctx_in, ctx, ctx_size_in))
1518 err = -EFAULT;
1519out:
1520 kfree(ctx);
1521 return err;
1522}
1523
1524static int verify_and_copy_hook_state(struct nf_hook_state *state,
1525 const struct nf_hook_state *user,
1526 struct net_device *dev)
1527{
1528 if (user->in || user->out)
1529 return -EINVAL;
1530
1531 if (user->net || user->sk || user->okfn)
1532 return -EINVAL;
1533
1534 switch (user->pf) {
1535 case NFPROTO_IPV4:
1536 case NFPROTO_IPV6:
1537 switch (state->hook) {
1538 case NF_INET_PRE_ROUTING:
1539 state->in = dev;
1540 break;
1541 case NF_INET_LOCAL_IN:
1542 state->in = dev;
1543 break;
1544 case NF_INET_FORWARD:
1545 state->in = dev;
1546 state->out = dev;
1547 break;
1548 case NF_INET_LOCAL_OUT:
1549 state->out = dev;
1550 break;
1551 case NF_INET_POST_ROUTING:
1552 state->out = dev;
1553 break;
1554 }
1555
1556 break;
1557 default:
1558 return -EINVAL;
1559 }
1560
1561 state->pf = user->pf;
1562 state->hook = user->hook;
1563
1564 return 0;
1565}
1566
1567static __be16 nfproto_eth(int nfproto)
1568{
1569 switch (nfproto) {
1570 case NFPROTO_IPV4:
1571 return htons(ETH_P_IP);
1572 case NFPROTO_IPV6:
1573 break;
1574 }
1575
1576 return htons(ETH_P_IPV6);
1577}
1578
1579int bpf_prog_test_run_nf(struct bpf_prog *prog,
1580 const union bpf_attr *kattr,
1581 union bpf_attr __user *uattr)
1582{
1583 struct net *net = current->nsproxy->net_ns;
1584 struct net_device *dev = net->loopback_dev;
1585 struct nf_hook_state *user_ctx, hook_state = {
1586 .pf = NFPROTO_IPV4,
1587 .hook = NF_INET_LOCAL_OUT,
1588 };
1589 u32 size = kattr->test.data_size_in;
1590 u32 repeat = kattr->test.repeat;
1591 struct bpf_nf_ctx ctx = {
1592 .state = &hook_state,
1593 };
1594 struct sk_buff *skb = NULL;
1595 u32 retval, duration;
1596 void *data;
1597 int ret;
1598
1599 if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1600 return -EINVAL;
1601
1602 if (size < sizeof(struct iphdr))
1603 return -EINVAL;
1604
1605 data = bpf_test_init(kattr, kattr->test.data_size_in, size,
1606 NET_SKB_PAD + NET_IP_ALIGN,
1607 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1608 if (IS_ERR(data))
1609 return PTR_ERR(data);
1610
1611 if (!repeat)
1612 repeat = 1;
1613
1614 user_ctx = bpf_ctx_init(kattr, sizeof(struct nf_hook_state));
1615 if (IS_ERR(user_ctx)) {
1616 kfree(data);
1617 return PTR_ERR(user_ctx);
1618 }
1619
1620 if (user_ctx) {
1621 ret = verify_and_copy_hook_state(&hook_state, user_ctx, dev);
1622 if (ret)
1623 goto out;
1624 }
1625
1626 skb = slab_build_skb(data);
1627 if (!skb) {
1628 ret = -ENOMEM;
1629 goto out;
1630 }
1631
1632 data = NULL; /* data released via kfree_skb */
1633
1634 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1635 __skb_put(skb, size);
1636
1637 ret = -EINVAL;
1638
1639 if (hook_state.hook != NF_INET_LOCAL_OUT) {
1640 if (size < ETH_HLEN + sizeof(struct iphdr))
1641 goto out;
1642
1643 skb->protocol = eth_type_trans(skb, dev);
1644 switch (skb->protocol) {
1645 case htons(ETH_P_IP):
1646 if (hook_state.pf == NFPROTO_IPV4)
1647 break;
1648 goto out;
1649 case htons(ETH_P_IPV6):
1650 if (size < ETH_HLEN + sizeof(struct ipv6hdr))
1651 goto out;
1652 if (hook_state.pf == NFPROTO_IPV6)
1653 break;
1654 goto out;
1655 default:
1656 ret = -EPROTO;
1657 goto out;
1658 }
1659
1660 skb_reset_network_header(skb);
1661 } else {
1662 skb->protocol = nfproto_eth(hook_state.pf);
1663 }
1664
1665 ctx.skb = skb;
1666
1667 ret = bpf_test_run(prog, &ctx, repeat, &retval, &duration, false);
1668 if (ret)
1669 goto out;
1670
1671 ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1672
1673out:
1674 kfree(user_ctx);
1675 kfree_skb(skb);
1676 kfree(data);
1677 return ret;
1678}
1679
1680static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = {
1681 .owner = THIS_MODULE,
1682 .set = &test_sk_check_kfunc_ids,
1683};
1684
1685BTF_ID_LIST(bpf_prog_test_dtor_kfunc_ids)
1686BTF_ID(struct, prog_test_ref_kfunc)
1687BTF_ID(func, bpf_kfunc_call_test_release_dtor)
1688BTF_ID(struct, prog_test_member)
1689BTF_ID(func, bpf_kfunc_call_memb_release_dtor)
1690
1691static int __init bpf_prog_test_run_init(void)
1692{
1693 const struct btf_id_dtor_kfunc bpf_prog_test_dtor_kfunc[] = {
1694 {
1695 .btf_id = bpf_prog_test_dtor_kfunc_ids[0],
1696 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[1]
1697 },
1698 {
1699 .btf_id = bpf_prog_test_dtor_kfunc_ids[2],
1700 .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[3],
1701 },
1702 };
1703 int ret;
1704
1705 ret = register_btf_fmodret_id_set(&bpf_test_modify_return_set);
1706 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set);
1707 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_prog_test_kfunc_set);
1708 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SYSCALL, &bpf_prog_test_kfunc_set);
1709 return ret ?: register_btf_id_dtor_kfuncs(bpf_prog_test_dtor_kfunc,
1710 ARRAY_SIZE(bpf_prog_test_dtor_kfunc),
1711 THIS_MODULE);
1712}
1713late_initcall(bpf_prog_test_run_init);