xref: /openbmc/linux/net/bpf/test_run.c (revision 4b33b5ff)
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/sock.h>
16 #include <net/tcp.h>
17 #include <net/net_namespace.h>
18 #include <net/page_pool.h>
19 #include <linux/error-injection.h>
20 #include <linux/smp.h>
21 #include <linux/sock_diag.h>
22 #include <net/xdp.h>
23 
24 #define CREATE_TRACE_POINTS
25 #include <trace/events/bpf_test_run.h>
26 
27 struct bpf_test_timer {
28 	enum { NO_PREEMPT, NO_MIGRATE } mode;
29 	u32 i;
30 	u64 time_start, time_spent;
31 };
32 
33 static void bpf_test_timer_enter(struct bpf_test_timer *t)
34 	__acquires(rcu)
35 {
36 	rcu_read_lock();
37 	if (t->mode == NO_PREEMPT)
38 		preempt_disable();
39 	else
40 		migrate_disable();
41 
42 	t->time_start = ktime_get_ns();
43 }
44 
45 static void bpf_test_timer_leave(struct bpf_test_timer *t)
46 	__releases(rcu)
47 {
48 	t->time_start = 0;
49 
50 	if (t->mode == NO_PREEMPT)
51 		preempt_enable();
52 	else
53 		migrate_enable();
54 	rcu_read_unlock();
55 }
56 
57 static bool bpf_test_timer_continue(struct bpf_test_timer *t, int iterations,
58 				    u32 repeat, int *err, u32 *duration)
59 	__must_hold(rcu)
60 {
61 	t->i += iterations;
62 	if (t->i >= repeat) {
63 		/* We're done. */
64 		t->time_spent += ktime_get_ns() - t->time_start;
65 		do_div(t->time_spent, t->i);
66 		*duration = t->time_spent > U32_MAX ? U32_MAX : (u32)t->time_spent;
67 		*err = 0;
68 		goto reset;
69 	}
70 
71 	if (signal_pending(current)) {
72 		/* During iteration: we've been cancelled, abort. */
73 		*err = -EINTR;
74 		goto reset;
75 	}
76 
77 	if (need_resched()) {
78 		/* During iteration: we need to reschedule between runs. */
79 		t->time_spent += ktime_get_ns() - t->time_start;
80 		bpf_test_timer_leave(t);
81 		cond_resched();
82 		bpf_test_timer_enter(t);
83 	}
84 
85 	/* Do another round. */
86 	return true;
87 
88 reset:
89 	t->i = 0;
90 	return false;
91 }
92 
93 /* We put this struct at the head of each page with a context and frame
94  * initialised when the page is allocated, so we don't have to do this on each
95  * repetition of the test run.
96  */
97 struct xdp_page_head {
98 	struct xdp_buff orig_ctx;
99 	struct xdp_buff ctx;
100 	struct xdp_frame frm;
101 	u8 data[];
102 };
103 
104 struct xdp_test_data {
105 	struct xdp_buff *orig_ctx;
106 	struct xdp_rxq_info rxq;
107 	struct net_device *dev;
108 	struct page_pool *pp;
109 	struct xdp_frame **frames;
110 	struct sk_buff **skbs;
111 	struct xdp_mem_info mem;
112 	u32 batch_size;
113 	u32 frame_cnt;
114 };
115 
116 #define TEST_XDP_FRAME_SIZE (PAGE_SIZE - sizeof(struct xdp_page_head))
117 #define TEST_XDP_MAX_BATCH 256
118 
119 static void xdp_test_run_init_page(struct page *page, void *arg)
120 {
121 	struct xdp_page_head *head = phys_to_virt(page_to_phys(page));
122 	struct xdp_buff *new_ctx, *orig_ctx;
123 	u32 headroom = XDP_PACKET_HEADROOM;
124 	struct xdp_test_data *xdp = arg;
125 	size_t frm_len, meta_len;
126 	struct xdp_frame *frm;
127 	void *data;
128 
129 	orig_ctx = xdp->orig_ctx;
130 	frm_len = orig_ctx->data_end - orig_ctx->data_meta;
131 	meta_len = orig_ctx->data - orig_ctx->data_meta;
132 	headroom -= meta_len;
133 
134 	new_ctx = &head->ctx;
135 	frm = &head->frm;
136 	data = &head->data;
137 	memcpy(data + headroom, orig_ctx->data_meta, frm_len);
138 
139 	xdp_init_buff(new_ctx, TEST_XDP_FRAME_SIZE, &xdp->rxq);
140 	xdp_prepare_buff(new_ctx, data, headroom, frm_len, true);
141 	new_ctx->data = new_ctx->data_meta + meta_len;
142 
143 	xdp_update_frame_from_buff(new_ctx, frm);
144 	frm->mem = new_ctx->rxq->mem;
145 
146 	memcpy(&head->orig_ctx, new_ctx, sizeof(head->orig_ctx));
147 }
148 
149 static int xdp_test_run_setup(struct xdp_test_data *xdp, struct xdp_buff *orig_ctx)
150 {
151 	struct page_pool *pp;
152 	int err = -ENOMEM;
153 	struct page_pool_params pp_params = {
154 		.order = 0,
155 		.flags = 0,
156 		.pool_size = xdp->batch_size,
157 		.nid = NUMA_NO_NODE,
158 		.init_callback = xdp_test_run_init_page,
159 		.init_arg = xdp,
160 	};
161 
162 	xdp->frames = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
163 	if (!xdp->frames)
164 		return -ENOMEM;
165 
166 	xdp->skbs = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
167 	if (!xdp->skbs)
168 		goto err_skbs;
169 
170 	pp = page_pool_create(&pp_params);
171 	if (IS_ERR(pp)) {
172 		err = PTR_ERR(pp);
173 		goto err_pp;
174 	}
175 
176 	/* will copy 'mem.id' into pp->xdp_mem_id */
177 	err = xdp_reg_mem_model(&xdp->mem, MEM_TYPE_PAGE_POOL, pp);
178 	if (err)
179 		goto err_mmodel;
180 
181 	xdp->pp = pp;
182 
183 	/* We create a 'fake' RXQ referencing the original dev, but with an
184 	 * xdp_mem_info pointing to our page_pool
185 	 */
186 	xdp_rxq_info_reg(&xdp->rxq, orig_ctx->rxq->dev, 0, 0);
187 	xdp->rxq.mem.type = MEM_TYPE_PAGE_POOL;
188 	xdp->rxq.mem.id = pp->xdp_mem_id;
189 	xdp->dev = orig_ctx->rxq->dev;
190 	xdp->orig_ctx = orig_ctx;
191 
192 	return 0;
193 
194 err_mmodel:
195 	page_pool_destroy(pp);
196 err_pp:
197 	kvfree(xdp->skbs);
198 err_skbs:
199 	kvfree(xdp->frames);
200 	return err;
201 }
202 
203 static void xdp_test_run_teardown(struct xdp_test_data *xdp)
204 {
205 	xdp_unreg_mem_model(&xdp->mem);
206 	page_pool_destroy(xdp->pp);
207 	kfree(xdp->frames);
208 	kfree(xdp->skbs);
209 }
210 
211 static bool ctx_was_changed(struct xdp_page_head *head)
212 {
213 	return head->orig_ctx.data != head->ctx.data ||
214 		head->orig_ctx.data_meta != head->ctx.data_meta ||
215 		head->orig_ctx.data_end != head->ctx.data_end;
216 }
217 
218 static void reset_ctx(struct xdp_page_head *head)
219 {
220 	if (likely(!ctx_was_changed(head)))
221 		return;
222 
223 	head->ctx.data = head->orig_ctx.data;
224 	head->ctx.data_meta = head->orig_ctx.data_meta;
225 	head->ctx.data_end = head->orig_ctx.data_end;
226 	xdp_update_frame_from_buff(&head->ctx, &head->frm);
227 }
228 
229 static int xdp_recv_frames(struct xdp_frame **frames, int nframes,
230 			   struct sk_buff **skbs,
231 			   struct net_device *dev)
232 {
233 	gfp_t gfp = __GFP_ZERO | GFP_ATOMIC;
234 	int i, n;
235 	LIST_HEAD(list);
236 
237 	n = kmem_cache_alloc_bulk(skbuff_head_cache, gfp, nframes, (void **)skbs);
238 	if (unlikely(n == 0)) {
239 		for (i = 0; i < nframes; i++)
240 			xdp_return_frame(frames[i]);
241 		return -ENOMEM;
242 	}
243 
244 	for (i = 0; i < nframes; i++) {
245 		struct xdp_frame *xdpf = frames[i];
246 		struct sk_buff *skb = skbs[i];
247 
248 		skb = __xdp_build_skb_from_frame(xdpf, skb, dev);
249 		if (!skb) {
250 			xdp_return_frame(xdpf);
251 			continue;
252 		}
253 
254 		list_add_tail(&skb->list, &list);
255 	}
256 	netif_receive_skb_list(&list);
257 
258 	return 0;
259 }
260 
261 static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
262 			      u32 repeat)
263 {
264 	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
265 	int err = 0, act, ret, i, nframes = 0, batch_sz;
266 	struct xdp_frame **frames = xdp->frames;
267 	struct xdp_page_head *head;
268 	struct xdp_frame *frm;
269 	bool redirect = false;
270 	struct xdp_buff *ctx;
271 	struct page *page;
272 
273 	batch_sz = min_t(u32, repeat, xdp->batch_size);
274 
275 	local_bh_disable();
276 	xdp_set_return_frame_no_direct();
277 
278 	for (i = 0; i < batch_sz; i++) {
279 		page = page_pool_dev_alloc_pages(xdp->pp);
280 		if (!page) {
281 			err = -ENOMEM;
282 			goto out;
283 		}
284 
285 		head = phys_to_virt(page_to_phys(page));
286 		reset_ctx(head);
287 		ctx = &head->ctx;
288 		frm = &head->frm;
289 		xdp->frame_cnt++;
290 
291 		act = bpf_prog_run_xdp(prog, ctx);
292 
293 		/* if program changed pkt bounds we need to update the xdp_frame */
294 		if (unlikely(ctx_was_changed(head))) {
295 			ret = xdp_update_frame_from_buff(ctx, frm);
296 			if (ret) {
297 				xdp_return_buff(ctx);
298 				continue;
299 			}
300 		}
301 
302 		switch (act) {
303 		case XDP_TX:
304 			/* we can't do a real XDP_TX since we're not in the
305 			 * driver, so turn it into a REDIRECT back to the same
306 			 * index
307 			 */
308 			ri->tgt_index = xdp->dev->ifindex;
309 			ri->map_id = INT_MAX;
310 			ri->map_type = BPF_MAP_TYPE_UNSPEC;
311 			fallthrough;
312 		case XDP_REDIRECT:
313 			redirect = true;
314 			ret = xdp_do_redirect_frame(xdp->dev, ctx, frm, prog);
315 			if (ret)
316 				xdp_return_buff(ctx);
317 			break;
318 		case XDP_PASS:
319 			frames[nframes++] = frm;
320 			break;
321 		default:
322 			bpf_warn_invalid_xdp_action(NULL, prog, act);
323 			fallthrough;
324 		case XDP_DROP:
325 			xdp_return_buff(ctx);
326 			break;
327 		}
328 	}
329 
330 out:
331 	if (redirect)
332 		xdp_do_flush();
333 	if (nframes) {
334 		ret = xdp_recv_frames(frames, nframes, xdp->skbs, xdp->dev);
335 		if (ret)
336 			err = ret;
337 	}
338 
339 	xdp_clear_return_frame_no_direct();
340 	local_bh_enable();
341 	return err;
342 }
343 
344 static int bpf_test_run_xdp_live(struct bpf_prog *prog, struct xdp_buff *ctx,
345 				 u32 repeat, u32 batch_size, u32 *time)
346 
347 {
348 	struct xdp_test_data xdp = { .batch_size = batch_size };
349 	struct bpf_test_timer t = { .mode = NO_MIGRATE };
350 	int ret;
351 
352 	if (!repeat)
353 		repeat = 1;
354 
355 	ret = xdp_test_run_setup(&xdp, ctx);
356 	if (ret)
357 		return ret;
358 
359 	bpf_test_timer_enter(&t);
360 	do {
361 		xdp.frame_cnt = 0;
362 		ret = xdp_test_run_batch(&xdp, prog, repeat - t.i);
363 		if (unlikely(ret < 0))
364 			break;
365 	} while (bpf_test_timer_continue(&t, xdp.frame_cnt, repeat, &ret, time));
366 	bpf_test_timer_leave(&t);
367 
368 	xdp_test_run_teardown(&xdp);
369 	return ret;
370 }
371 
372 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
373 			u32 *retval, u32 *time, bool xdp)
374 {
375 	struct bpf_prog_array_item item = {.prog = prog};
376 	struct bpf_run_ctx *old_ctx;
377 	struct bpf_cg_run_ctx run_ctx;
378 	struct bpf_test_timer t = { NO_MIGRATE };
379 	enum bpf_cgroup_storage_type stype;
380 	int ret;
381 
382 	for_each_cgroup_storage_type(stype) {
383 		item.cgroup_storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
384 		if (IS_ERR(item.cgroup_storage[stype])) {
385 			item.cgroup_storage[stype] = NULL;
386 			for_each_cgroup_storage_type(stype)
387 				bpf_cgroup_storage_free(item.cgroup_storage[stype]);
388 			return -ENOMEM;
389 		}
390 	}
391 
392 	if (!repeat)
393 		repeat = 1;
394 
395 	bpf_test_timer_enter(&t);
396 	old_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
397 	do {
398 		run_ctx.prog_item = &item;
399 		if (xdp)
400 			*retval = bpf_prog_run_xdp(prog, ctx);
401 		else
402 			*retval = bpf_prog_run(prog, ctx);
403 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, time));
404 	bpf_reset_run_ctx(old_ctx);
405 	bpf_test_timer_leave(&t);
406 
407 	for_each_cgroup_storage_type(stype)
408 		bpf_cgroup_storage_free(item.cgroup_storage[stype]);
409 
410 	return ret;
411 }
412 
413 static int bpf_test_finish(const union bpf_attr *kattr,
414 			   union bpf_attr __user *uattr, const void *data,
415 			   struct skb_shared_info *sinfo, u32 size,
416 			   u32 retval, u32 duration)
417 {
418 	void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
419 	int err = -EFAULT;
420 	u32 copy_size = size;
421 
422 	/* Clamp copy if the user has provided a size hint, but copy the full
423 	 * buffer if not to retain old behaviour.
424 	 */
425 	if (kattr->test.data_size_out &&
426 	    copy_size > kattr->test.data_size_out) {
427 		copy_size = kattr->test.data_size_out;
428 		err = -ENOSPC;
429 	}
430 
431 	if (data_out) {
432 		int len = sinfo ? copy_size - sinfo->xdp_frags_size : copy_size;
433 
434 		if (len < 0) {
435 			err = -ENOSPC;
436 			goto out;
437 		}
438 
439 		if (copy_to_user(data_out, data, len))
440 			goto out;
441 
442 		if (sinfo) {
443 			int i, offset = len;
444 			u32 data_len;
445 
446 			for (i = 0; i < sinfo->nr_frags; i++) {
447 				skb_frag_t *frag = &sinfo->frags[i];
448 
449 				if (offset >= copy_size) {
450 					err = -ENOSPC;
451 					break;
452 				}
453 
454 				data_len = min_t(u32, copy_size - offset,
455 						 skb_frag_size(frag));
456 
457 				if (copy_to_user(data_out + offset,
458 						 skb_frag_address(frag),
459 						 data_len))
460 					goto out;
461 
462 				offset += data_len;
463 			}
464 		}
465 	}
466 
467 	if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
468 		goto out;
469 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
470 		goto out;
471 	if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
472 		goto out;
473 	if (err != -ENOSPC)
474 		err = 0;
475 out:
476 	trace_bpf_test_finish(&err);
477 	return err;
478 }
479 
480 /* Integer types of various sizes and pointer combinations cover variety of
481  * architecture dependent calling conventions. 7+ can be supported in the
482  * future.
483  */
484 __diag_push();
485 __diag_ignore_all("-Wmissing-prototypes",
486 		  "Global functions as their definitions will be in vmlinux BTF");
487 int noinline bpf_fentry_test1(int a)
488 {
489 	return a + 1;
490 }
491 EXPORT_SYMBOL_GPL(bpf_fentry_test1);
492 ALLOW_ERROR_INJECTION(bpf_fentry_test1, ERRNO);
493 
494 int noinline bpf_fentry_test2(int a, u64 b)
495 {
496 	return a + b;
497 }
498 
499 int noinline bpf_fentry_test3(char a, int b, u64 c)
500 {
501 	return a + b + c;
502 }
503 
504 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
505 {
506 	return (long)a + b + c + d;
507 }
508 
509 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
510 {
511 	return a + (long)b + c + d + e;
512 }
513 
514 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
515 {
516 	return a + (long)b + c + d + (long)e + f;
517 }
518 
519 struct bpf_fentry_test_t {
520 	struct bpf_fentry_test_t *a;
521 };
522 
523 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
524 {
525 	return (long)arg;
526 }
527 
528 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
529 {
530 	return (long)arg->a;
531 }
532 
533 int noinline bpf_modify_return_test(int a, int *b)
534 {
535 	*b += 1;
536 	return a + *b;
537 }
538 
539 u64 noinline bpf_kfunc_call_test1(struct sock *sk, u32 a, u64 b, u32 c, u64 d)
540 {
541 	return a + b + c + d;
542 }
543 
544 int noinline bpf_kfunc_call_test2(struct sock *sk, u32 a, u32 b)
545 {
546 	return a + b;
547 }
548 
549 struct sock * noinline bpf_kfunc_call_test3(struct sock *sk)
550 {
551 	return sk;
552 }
553 
554 struct prog_test_member {
555 	u64 c;
556 };
557 
558 struct prog_test_ref_kfunc {
559 	int a;
560 	int b;
561 	struct prog_test_member memb;
562 	struct prog_test_ref_kfunc *next;
563 };
564 
565 static struct prog_test_ref_kfunc prog_test_struct = {
566 	.a = 42,
567 	.b = 108,
568 	.next = &prog_test_struct,
569 };
570 
571 noinline struct prog_test_ref_kfunc *
572 bpf_kfunc_call_test_acquire(unsigned long *scalar_ptr)
573 {
574 	/* randomly return NULL */
575 	if (get_jiffies_64() % 2)
576 		return NULL;
577 	return &prog_test_struct;
578 }
579 
580 noinline void bpf_kfunc_call_test_release(struct prog_test_ref_kfunc *p)
581 {
582 }
583 
584 noinline void bpf_kfunc_call_memb_release(struct prog_test_member *p)
585 {
586 }
587 
588 struct prog_test_pass1 {
589 	int x0;
590 	struct {
591 		int x1;
592 		struct {
593 			int x2;
594 			struct {
595 				int x3;
596 			};
597 		};
598 	};
599 };
600 
601 struct prog_test_pass2 {
602 	int len;
603 	short arr1[4];
604 	struct {
605 		char arr2[4];
606 		unsigned long arr3[8];
607 	} x;
608 };
609 
610 struct prog_test_fail1 {
611 	void *p;
612 	int x;
613 };
614 
615 struct prog_test_fail2 {
616 	int x8;
617 	struct prog_test_pass1 x;
618 };
619 
620 struct prog_test_fail3 {
621 	int len;
622 	char arr1[2];
623 	char arr2[];
624 };
625 
626 noinline void bpf_kfunc_call_test_pass_ctx(struct __sk_buff *skb)
627 {
628 }
629 
630 noinline void bpf_kfunc_call_test_pass1(struct prog_test_pass1 *p)
631 {
632 }
633 
634 noinline void bpf_kfunc_call_test_pass2(struct prog_test_pass2 *p)
635 {
636 }
637 
638 noinline void bpf_kfunc_call_test_fail1(struct prog_test_fail1 *p)
639 {
640 }
641 
642 noinline void bpf_kfunc_call_test_fail2(struct prog_test_fail2 *p)
643 {
644 }
645 
646 noinline void bpf_kfunc_call_test_fail3(struct prog_test_fail3 *p)
647 {
648 }
649 
650 noinline void bpf_kfunc_call_test_mem_len_pass1(void *mem, int mem__sz)
651 {
652 }
653 
654 noinline void bpf_kfunc_call_test_mem_len_fail1(void *mem, int len)
655 {
656 }
657 
658 noinline void bpf_kfunc_call_test_mem_len_fail2(u64 *mem, int len)
659 {
660 }
661 
662 __diag_pop();
663 
664 ALLOW_ERROR_INJECTION(bpf_modify_return_test, ERRNO);
665 
666 BTF_SET_START(test_sk_check_kfunc_ids)
667 BTF_ID(func, bpf_kfunc_call_test1)
668 BTF_ID(func, bpf_kfunc_call_test2)
669 BTF_ID(func, bpf_kfunc_call_test3)
670 BTF_ID(func, bpf_kfunc_call_test_acquire)
671 BTF_ID(func, bpf_kfunc_call_test_release)
672 BTF_ID(func, bpf_kfunc_call_memb_release)
673 BTF_ID(func, bpf_kfunc_call_test_pass_ctx)
674 BTF_ID(func, bpf_kfunc_call_test_pass1)
675 BTF_ID(func, bpf_kfunc_call_test_pass2)
676 BTF_ID(func, bpf_kfunc_call_test_fail1)
677 BTF_ID(func, bpf_kfunc_call_test_fail2)
678 BTF_ID(func, bpf_kfunc_call_test_fail3)
679 BTF_ID(func, bpf_kfunc_call_test_mem_len_pass1)
680 BTF_ID(func, bpf_kfunc_call_test_mem_len_fail1)
681 BTF_ID(func, bpf_kfunc_call_test_mem_len_fail2)
682 BTF_SET_END(test_sk_check_kfunc_ids)
683 
684 BTF_SET_START(test_sk_acquire_kfunc_ids)
685 BTF_ID(func, bpf_kfunc_call_test_acquire)
686 BTF_SET_END(test_sk_acquire_kfunc_ids)
687 
688 BTF_SET_START(test_sk_release_kfunc_ids)
689 BTF_ID(func, bpf_kfunc_call_test_release)
690 BTF_ID(func, bpf_kfunc_call_memb_release)
691 BTF_SET_END(test_sk_release_kfunc_ids)
692 
693 BTF_SET_START(test_sk_ret_null_kfunc_ids)
694 BTF_ID(func, bpf_kfunc_call_test_acquire)
695 BTF_SET_END(test_sk_ret_null_kfunc_ids)
696 
697 static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
698 			   u32 size, u32 headroom, u32 tailroom)
699 {
700 	void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
701 	void *data;
702 
703 	if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
704 		return ERR_PTR(-EINVAL);
705 
706 	if (user_size > size)
707 		return ERR_PTR(-EMSGSIZE);
708 
709 	data = kzalloc(size + headroom + tailroom, GFP_USER);
710 	if (!data)
711 		return ERR_PTR(-ENOMEM);
712 
713 	if (copy_from_user(data + headroom, data_in, user_size)) {
714 		kfree(data);
715 		return ERR_PTR(-EFAULT);
716 	}
717 
718 	return data;
719 }
720 
721 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
722 			      const union bpf_attr *kattr,
723 			      union bpf_attr __user *uattr)
724 {
725 	struct bpf_fentry_test_t arg = {};
726 	u16 side_effect = 0, ret = 0;
727 	int b = 2, err = -EFAULT;
728 	u32 retval = 0;
729 
730 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
731 		return -EINVAL;
732 
733 	switch (prog->expected_attach_type) {
734 	case BPF_TRACE_FENTRY:
735 	case BPF_TRACE_FEXIT:
736 		if (bpf_fentry_test1(1) != 2 ||
737 		    bpf_fentry_test2(2, 3) != 5 ||
738 		    bpf_fentry_test3(4, 5, 6) != 15 ||
739 		    bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
740 		    bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
741 		    bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
742 		    bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
743 		    bpf_fentry_test8(&arg) != 0)
744 			goto out;
745 		break;
746 	case BPF_MODIFY_RETURN:
747 		ret = bpf_modify_return_test(1, &b);
748 		if (b != 2)
749 			side_effect = 1;
750 		break;
751 	default:
752 		goto out;
753 	}
754 
755 	retval = ((u32)side_effect << 16) | ret;
756 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
757 		goto out;
758 
759 	err = 0;
760 out:
761 	trace_bpf_test_finish(&err);
762 	return err;
763 }
764 
765 struct bpf_raw_tp_test_run_info {
766 	struct bpf_prog *prog;
767 	void *ctx;
768 	u32 retval;
769 };
770 
771 static void
772 __bpf_prog_test_run_raw_tp(void *data)
773 {
774 	struct bpf_raw_tp_test_run_info *info = data;
775 
776 	rcu_read_lock();
777 	info->retval = bpf_prog_run(info->prog, info->ctx);
778 	rcu_read_unlock();
779 }
780 
781 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
782 			     const union bpf_attr *kattr,
783 			     union bpf_attr __user *uattr)
784 {
785 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
786 	__u32 ctx_size_in = kattr->test.ctx_size_in;
787 	struct bpf_raw_tp_test_run_info info;
788 	int cpu = kattr->test.cpu, err = 0;
789 	int current_cpu;
790 
791 	/* doesn't support data_in/out, ctx_out, duration, or repeat */
792 	if (kattr->test.data_in || kattr->test.data_out ||
793 	    kattr->test.ctx_out || kattr->test.duration ||
794 	    kattr->test.repeat || kattr->test.batch_size)
795 		return -EINVAL;
796 
797 	if (ctx_size_in < prog->aux->max_ctx_offset ||
798 	    ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
799 		return -EINVAL;
800 
801 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
802 		return -EINVAL;
803 
804 	if (ctx_size_in) {
805 		info.ctx = memdup_user(ctx_in, ctx_size_in);
806 		if (IS_ERR(info.ctx))
807 			return PTR_ERR(info.ctx);
808 	} else {
809 		info.ctx = NULL;
810 	}
811 
812 	info.prog = prog;
813 
814 	current_cpu = get_cpu();
815 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
816 	    cpu == current_cpu) {
817 		__bpf_prog_test_run_raw_tp(&info);
818 	} else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
819 		/* smp_call_function_single() also checks cpu_online()
820 		 * after csd_lock(). However, since cpu is from user
821 		 * space, let's do an extra quick check to filter out
822 		 * invalid value before smp_call_function_single().
823 		 */
824 		err = -ENXIO;
825 	} else {
826 		err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
827 					       &info, 1);
828 	}
829 	put_cpu();
830 
831 	if (!err &&
832 	    copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
833 		err = -EFAULT;
834 
835 	kfree(info.ctx);
836 	return err;
837 }
838 
839 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
840 {
841 	void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
842 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
843 	u32 size = kattr->test.ctx_size_in;
844 	void *data;
845 	int err;
846 
847 	if (!data_in && !data_out)
848 		return NULL;
849 
850 	data = kzalloc(max_size, GFP_USER);
851 	if (!data)
852 		return ERR_PTR(-ENOMEM);
853 
854 	if (data_in) {
855 		err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
856 		if (err) {
857 			kfree(data);
858 			return ERR_PTR(err);
859 		}
860 
861 		size = min_t(u32, max_size, size);
862 		if (copy_from_user(data, data_in, size)) {
863 			kfree(data);
864 			return ERR_PTR(-EFAULT);
865 		}
866 	}
867 	return data;
868 }
869 
870 static int bpf_ctx_finish(const union bpf_attr *kattr,
871 			  union bpf_attr __user *uattr, const void *data,
872 			  u32 size)
873 {
874 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
875 	int err = -EFAULT;
876 	u32 copy_size = size;
877 
878 	if (!data || !data_out)
879 		return 0;
880 
881 	if (copy_size > kattr->test.ctx_size_out) {
882 		copy_size = kattr->test.ctx_size_out;
883 		err = -ENOSPC;
884 	}
885 
886 	if (copy_to_user(data_out, data, copy_size))
887 		goto out;
888 	if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
889 		goto out;
890 	if (err != -ENOSPC)
891 		err = 0;
892 out:
893 	return err;
894 }
895 
896 /**
897  * range_is_zero - test whether buffer is initialized
898  * @buf: buffer to check
899  * @from: check from this position
900  * @to: check up until (excluding) this position
901  *
902  * This function returns true if the there is a non-zero byte
903  * in the buf in the range [from,to).
904  */
905 static inline bool range_is_zero(void *buf, size_t from, size_t to)
906 {
907 	return !memchr_inv((u8 *)buf + from, 0, to - from);
908 }
909 
910 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
911 {
912 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
913 
914 	if (!__skb)
915 		return 0;
916 
917 	/* make sure the fields we don't use are zeroed */
918 	if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
919 		return -EINVAL;
920 
921 	/* mark is allowed */
922 
923 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
924 			   offsetof(struct __sk_buff, priority)))
925 		return -EINVAL;
926 
927 	/* priority is allowed */
928 	/* ingress_ifindex is allowed */
929 	/* ifindex is allowed */
930 
931 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
932 			   offsetof(struct __sk_buff, cb)))
933 		return -EINVAL;
934 
935 	/* cb is allowed */
936 
937 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
938 			   offsetof(struct __sk_buff, tstamp)))
939 		return -EINVAL;
940 
941 	/* tstamp is allowed */
942 	/* wire_len is allowed */
943 	/* gso_segs is allowed */
944 
945 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
946 			   offsetof(struct __sk_buff, gso_size)))
947 		return -EINVAL;
948 
949 	/* gso_size is allowed */
950 
951 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
952 			   offsetof(struct __sk_buff, hwtstamp)))
953 		return -EINVAL;
954 
955 	/* hwtstamp is allowed */
956 
957 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp),
958 			   sizeof(struct __sk_buff)))
959 		return -EINVAL;
960 
961 	skb->mark = __skb->mark;
962 	skb->priority = __skb->priority;
963 	skb->skb_iif = __skb->ingress_ifindex;
964 	skb->tstamp = __skb->tstamp;
965 	memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
966 
967 	if (__skb->wire_len == 0) {
968 		cb->pkt_len = skb->len;
969 	} else {
970 		if (__skb->wire_len < skb->len ||
971 		    __skb->wire_len > GSO_MAX_SIZE)
972 			return -EINVAL;
973 		cb->pkt_len = __skb->wire_len;
974 	}
975 
976 	if (__skb->gso_segs > GSO_MAX_SEGS)
977 		return -EINVAL;
978 	skb_shinfo(skb)->gso_segs = __skb->gso_segs;
979 	skb_shinfo(skb)->gso_size = __skb->gso_size;
980 	skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp;
981 
982 	return 0;
983 }
984 
985 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
986 {
987 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
988 
989 	if (!__skb)
990 		return;
991 
992 	__skb->mark = skb->mark;
993 	__skb->priority = skb->priority;
994 	__skb->ingress_ifindex = skb->skb_iif;
995 	__skb->ifindex = skb->dev->ifindex;
996 	__skb->tstamp = skb->tstamp;
997 	memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
998 	__skb->wire_len = cb->pkt_len;
999 	__skb->gso_segs = skb_shinfo(skb)->gso_segs;
1000 	__skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp;
1001 }
1002 
1003 static struct proto bpf_dummy_proto = {
1004 	.name   = "bpf_dummy",
1005 	.owner  = THIS_MODULE,
1006 	.obj_size = sizeof(struct sock),
1007 };
1008 
1009 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
1010 			  union bpf_attr __user *uattr)
1011 {
1012 	bool is_l2 = false, is_direct_pkt_access = false;
1013 	struct net *net = current->nsproxy->net_ns;
1014 	struct net_device *dev = net->loopback_dev;
1015 	u32 size = kattr->test.data_size_in;
1016 	u32 repeat = kattr->test.repeat;
1017 	struct __sk_buff *ctx = NULL;
1018 	u32 retval, duration;
1019 	int hh_len = ETH_HLEN;
1020 	struct sk_buff *skb;
1021 	struct sock *sk;
1022 	void *data;
1023 	int ret;
1024 
1025 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1026 		return -EINVAL;
1027 
1028 	data = bpf_test_init(kattr, kattr->test.data_size_in,
1029 			     size, NET_SKB_PAD + NET_IP_ALIGN,
1030 			     SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1031 	if (IS_ERR(data))
1032 		return PTR_ERR(data);
1033 
1034 	ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
1035 	if (IS_ERR(ctx)) {
1036 		kfree(data);
1037 		return PTR_ERR(ctx);
1038 	}
1039 
1040 	switch (prog->type) {
1041 	case BPF_PROG_TYPE_SCHED_CLS:
1042 	case BPF_PROG_TYPE_SCHED_ACT:
1043 		is_l2 = true;
1044 		fallthrough;
1045 	case BPF_PROG_TYPE_LWT_IN:
1046 	case BPF_PROG_TYPE_LWT_OUT:
1047 	case BPF_PROG_TYPE_LWT_XMIT:
1048 		is_direct_pkt_access = true;
1049 		break;
1050 	default:
1051 		break;
1052 	}
1053 
1054 	sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
1055 	if (!sk) {
1056 		kfree(data);
1057 		kfree(ctx);
1058 		return -ENOMEM;
1059 	}
1060 	sock_init_data(NULL, sk);
1061 
1062 	skb = build_skb(data, 0);
1063 	if (!skb) {
1064 		kfree(data);
1065 		kfree(ctx);
1066 		sk_free(sk);
1067 		return -ENOMEM;
1068 	}
1069 	skb->sk = sk;
1070 
1071 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1072 	__skb_put(skb, size);
1073 	if (ctx && ctx->ifindex > 1) {
1074 		dev = dev_get_by_index(net, ctx->ifindex);
1075 		if (!dev) {
1076 			ret = -ENODEV;
1077 			goto out;
1078 		}
1079 	}
1080 	skb->protocol = eth_type_trans(skb, dev);
1081 	skb_reset_network_header(skb);
1082 
1083 	switch (skb->protocol) {
1084 	case htons(ETH_P_IP):
1085 		sk->sk_family = AF_INET;
1086 		if (sizeof(struct iphdr) <= skb_headlen(skb)) {
1087 			sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
1088 			sk->sk_daddr = ip_hdr(skb)->daddr;
1089 		}
1090 		break;
1091 #if IS_ENABLED(CONFIG_IPV6)
1092 	case htons(ETH_P_IPV6):
1093 		sk->sk_family = AF_INET6;
1094 		if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
1095 			sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
1096 			sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
1097 		}
1098 		break;
1099 #endif
1100 	default:
1101 		break;
1102 	}
1103 
1104 	if (is_l2)
1105 		__skb_push(skb, hh_len);
1106 	if (is_direct_pkt_access)
1107 		bpf_compute_data_pointers(skb);
1108 	ret = convert___skb_to_skb(skb, ctx);
1109 	if (ret)
1110 		goto out;
1111 	ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
1112 	if (ret)
1113 		goto out;
1114 	if (!is_l2) {
1115 		if (skb_headroom(skb) < hh_len) {
1116 			int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
1117 
1118 			if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
1119 				ret = -ENOMEM;
1120 				goto out;
1121 			}
1122 		}
1123 		memset(__skb_push(skb, hh_len), 0, hh_len);
1124 	}
1125 	convert_skb_to___skb(skb, ctx);
1126 
1127 	size = skb->len;
1128 	/* bpf program can never convert linear skb to non-linear */
1129 	if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
1130 		size = skb_headlen(skb);
1131 	ret = bpf_test_finish(kattr, uattr, skb->data, NULL, size, retval,
1132 			      duration);
1133 	if (!ret)
1134 		ret = bpf_ctx_finish(kattr, uattr, ctx,
1135 				     sizeof(struct __sk_buff));
1136 out:
1137 	if (dev && dev != net->loopback_dev)
1138 		dev_put(dev);
1139 	kfree_skb(skb);
1140 	sk_free(sk);
1141 	kfree(ctx);
1142 	return ret;
1143 }
1144 
1145 static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp)
1146 {
1147 	unsigned int ingress_ifindex, rx_queue_index;
1148 	struct netdev_rx_queue *rxqueue;
1149 	struct net_device *device;
1150 
1151 	if (!xdp_md)
1152 		return 0;
1153 
1154 	if (xdp_md->egress_ifindex != 0)
1155 		return -EINVAL;
1156 
1157 	ingress_ifindex = xdp_md->ingress_ifindex;
1158 	rx_queue_index = xdp_md->rx_queue_index;
1159 
1160 	if (!ingress_ifindex && rx_queue_index)
1161 		return -EINVAL;
1162 
1163 	if (ingress_ifindex) {
1164 		device = dev_get_by_index(current->nsproxy->net_ns,
1165 					  ingress_ifindex);
1166 		if (!device)
1167 			return -ENODEV;
1168 
1169 		if (rx_queue_index >= device->real_num_rx_queues)
1170 			goto free_dev;
1171 
1172 		rxqueue = __netif_get_rx_queue(device, rx_queue_index);
1173 
1174 		if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq))
1175 			goto free_dev;
1176 
1177 		xdp->rxq = &rxqueue->xdp_rxq;
1178 		/* The device is now tracked in the xdp->rxq for later
1179 		 * dev_put()
1180 		 */
1181 	}
1182 
1183 	xdp->data = xdp->data_meta + xdp_md->data;
1184 	return 0;
1185 
1186 free_dev:
1187 	dev_put(device);
1188 	return -EINVAL;
1189 }
1190 
1191 static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md)
1192 {
1193 	if (!xdp_md)
1194 		return;
1195 
1196 	xdp_md->data = xdp->data - xdp->data_meta;
1197 	xdp_md->data_end = xdp->data_end - xdp->data_meta;
1198 
1199 	if (xdp_md->ingress_ifindex)
1200 		dev_put(xdp->rxq->dev);
1201 }
1202 
1203 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
1204 			  union bpf_attr __user *uattr)
1205 {
1206 	bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES);
1207 	u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1208 	u32 batch_size = kattr->test.batch_size;
1209 	u32 retval = 0, duration, max_data_sz;
1210 	u32 size = kattr->test.data_size_in;
1211 	u32 headroom = XDP_PACKET_HEADROOM;
1212 	u32 repeat = kattr->test.repeat;
1213 	struct netdev_rx_queue *rxqueue;
1214 	struct skb_shared_info *sinfo;
1215 	struct xdp_buff xdp = {};
1216 	int i, ret = -EINVAL;
1217 	struct xdp_md *ctx;
1218 	void *data;
1219 
1220 	if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
1221 	    prog->expected_attach_type == BPF_XDP_CPUMAP)
1222 		return -EINVAL;
1223 
1224 	if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES)
1225 		return -EINVAL;
1226 
1227 	if (do_live) {
1228 		if (!batch_size)
1229 			batch_size = NAPI_POLL_WEIGHT;
1230 		else if (batch_size > TEST_XDP_MAX_BATCH)
1231 			return -E2BIG;
1232 
1233 		headroom += sizeof(struct xdp_page_head);
1234 	} else if (batch_size) {
1235 		return -EINVAL;
1236 	}
1237 
1238 	ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md));
1239 	if (IS_ERR(ctx))
1240 		return PTR_ERR(ctx);
1241 
1242 	if (ctx) {
1243 		/* There can't be user provided data before the meta data */
1244 		if (ctx->data_meta || ctx->data_end != size ||
1245 		    ctx->data > ctx->data_end ||
1246 		    unlikely(xdp_metalen_invalid(ctx->data)) ||
1247 		    (do_live && (kattr->test.data_out || kattr->test.ctx_out)))
1248 			goto free_ctx;
1249 		/* Meta data is allocated from the headroom */
1250 		headroom -= ctx->data;
1251 	}
1252 
1253 	max_data_sz = 4096 - headroom - tailroom;
1254 	if (size > max_data_sz) {
1255 		/* disallow live data mode for jumbo frames */
1256 		if (do_live)
1257 			goto free_ctx;
1258 		size = max_data_sz;
1259 	}
1260 
1261 	data = bpf_test_init(kattr, size, max_data_sz, headroom, tailroom);
1262 	if (IS_ERR(data)) {
1263 		ret = PTR_ERR(data);
1264 		goto free_ctx;
1265 	}
1266 
1267 	rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
1268 	rxqueue->xdp_rxq.frag_size = headroom + max_data_sz + tailroom;
1269 	xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq);
1270 	xdp_prepare_buff(&xdp, data, headroom, size, true);
1271 	sinfo = xdp_get_shared_info_from_buff(&xdp);
1272 
1273 	ret = xdp_convert_md_to_buff(ctx, &xdp);
1274 	if (ret)
1275 		goto free_data;
1276 
1277 	if (unlikely(kattr->test.data_size_in > size)) {
1278 		void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
1279 
1280 		while (size < kattr->test.data_size_in) {
1281 			struct page *page;
1282 			skb_frag_t *frag;
1283 			u32 data_len;
1284 
1285 			if (sinfo->nr_frags == MAX_SKB_FRAGS) {
1286 				ret = -ENOMEM;
1287 				goto out;
1288 			}
1289 
1290 			page = alloc_page(GFP_KERNEL);
1291 			if (!page) {
1292 				ret = -ENOMEM;
1293 				goto out;
1294 			}
1295 
1296 			frag = &sinfo->frags[sinfo->nr_frags++];
1297 			__skb_frag_set_page(frag, page);
1298 
1299 			data_len = min_t(u32, kattr->test.data_size_in - size,
1300 					 PAGE_SIZE);
1301 			skb_frag_size_set(frag, data_len);
1302 
1303 			if (copy_from_user(page_address(page), data_in + size,
1304 					   data_len)) {
1305 				ret = -EFAULT;
1306 				goto out;
1307 			}
1308 			sinfo->xdp_frags_size += data_len;
1309 			size += data_len;
1310 		}
1311 		xdp_buff_set_frags_flag(&xdp);
1312 	}
1313 
1314 	if (repeat > 1)
1315 		bpf_prog_change_xdp(NULL, prog);
1316 
1317 	if (do_live)
1318 		ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration);
1319 	else
1320 		ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
1321 	/* We convert the xdp_buff back to an xdp_md before checking the return
1322 	 * code so the reference count of any held netdevice will be decremented
1323 	 * even if the test run failed.
1324 	 */
1325 	xdp_convert_buff_to_md(&xdp, ctx);
1326 	if (ret)
1327 		goto out;
1328 
1329 	size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size;
1330 	ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size,
1331 			      retval, duration);
1332 	if (!ret)
1333 		ret = bpf_ctx_finish(kattr, uattr, ctx,
1334 				     sizeof(struct xdp_md));
1335 
1336 out:
1337 	if (repeat > 1)
1338 		bpf_prog_change_xdp(prog, NULL);
1339 free_data:
1340 	for (i = 0; i < sinfo->nr_frags; i++)
1341 		__free_page(skb_frag_page(&sinfo->frags[i]));
1342 	kfree(data);
1343 free_ctx:
1344 	kfree(ctx);
1345 	return ret;
1346 }
1347 
1348 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
1349 {
1350 	/* make sure the fields we don't use are zeroed */
1351 	if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
1352 		return -EINVAL;
1353 
1354 	/* flags is allowed */
1355 
1356 	if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
1357 			   sizeof(struct bpf_flow_keys)))
1358 		return -EINVAL;
1359 
1360 	return 0;
1361 }
1362 
1363 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1364 				     const union bpf_attr *kattr,
1365 				     union bpf_attr __user *uattr)
1366 {
1367 	struct bpf_test_timer t = { NO_PREEMPT };
1368 	u32 size = kattr->test.data_size_in;
1369 	struct bpf_flow_dissector ctx = {};
1370 	u32 repeat = kattr->test.repeat;
1371 	struct bpf_flow_keys *user_ctx;
1372 	struct bpf_flow_keys flow_keys;
1373 	const struct ethhdr *eth;
1374 	unsigned int flags = 0;
1375 	u32 retval, duration;
1376 	void *data;
1377 	int ret;
1378 
1379 	if (prog->type != BPF_PROG_TYPE_FLOW_DISSECTOR)
1380 		return -EINVAL;
1381 
1382 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1383 		return -EINVAL;
1384 
1385 	if (size < ETH_HLEN)
1386 		return -EINVAL;
1387 
1388 	data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0);
1389 	if (IS_ERR(data))
1390 		return PTR_ERR(data);
1391 
1392 	eth = (struct ethhdr *)data;
1393 
1394 	if (!repeat)
1395 		repeat = 1;
1396 
1397 	user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
1398 	if (IS_ERR(user_ctx)) {
1399 		kfree(data);
1400 		return PTR_ERR(user_ctx);
1401 	}
1402 	if (user_ctx) {
1403 		ret = verify_user_bpf_flow_keys(user_ctx);
1404 		if (ret)
1405 			goto out;
1406 		flags = user_ctx->flags;
1407 	}
1408 
1409 	ctx.flow_keys = &flow_keys;
1410 	ctx.data = data;
1411 	ctx.data_end = (__u8 *)data + size;
1412 
1413 	bpf_test_timer_enter(&t);
1414 	do {
1415 		retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
1416 					  size, flags);
1417 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1418 	bpf_test_timer_leave(&t);
1419 
1420 	if (ret < 0)
1421 		goto out;
1422 
1423 	ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL,
1424 			      sizeof(flow_keys), retval, duration);
1425 	if (!ret)
1426 		ret = bpf_ctx_finish(kattr, uattr, user_ctx,
1427 				     sizeof(struct bpf_flow_keys));
1428 
1429 out:
1430 	kfree(user_ctx);
1431 	kfree(data);
1432 	return ret;
1433 }
1434 
1435 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
1436 				union bpf_attr __user *uattr)
1437 {
1438 	struct bpf_test_timer t = { NO_PREEMPT };
1439 	struct bpf_prog_array *progs = NULL;
1440 	struct bpf_sk_lookup_kern ctx = {};
1441 	u32 repeat = kattr->test.repeat;
1442 	struct bpf_sk_lookup *user_ctx;
1443 	u32 retval, duration;
1444 	int ret = -EINVAL;
1445 
1446 	if (prog->type != BPF_PROG_TYPE_SK_LOOKUP)
1447 		return -EINVAL;
1448 
1449 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1450 		return -EINVAL;
1451 
1452 	if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
1453 	    kattr->test.data_size_out)
1454 		return -EINVAL;
1455 
1456 	if (!repeat)
1457 		repeat = 1;
1458 
1459 	user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
1460 	if (IS_ERR(user_ctx))
1461 		return PTR_ERR(user_ctx);
1462 
1463 	if (!user_ctx)
1464 		return -EINVAL;
1465 
1466 	if (user_ctx->sk)
1467 		goto out;
1468 
1469 	if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
1470 		goto out;
1471 
1472 	if (user_ctx->local_port > U16_MAX) {
1473 		ret = -ERANGE;
1474 		goto out;
1475 	}
1476 
1477 	ctx.family = (u16)user_ctx->family;
1478 	ctx.protocol = (u16)user_ctx->protocol;
1479 	ctx.dport = (u16)user_ctx->local_port;
1480 	ctx.sport = user_ctx->remote_port;
1481 
1482 	switch (ctx.family) {
1483 	case AF_INET:
1484 		ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
1485 		ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
1486 		break;
1487 
1488 #if IS_ENABLED(CONFIG_IPV6)
1489 	case AF_INET6:
1490 		ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
1491 		ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
1492 		break;
1493 #endif
1494 
1495 	default:
1496 		ret = -EAFNOSUPPORT;
1497 		goto out;
1498 	}
1499 
1500 	progs = bpf_prog_array_alloc(1, GFP_KERNEL);
1501 	if (!progs) {
1502 		ret = -ENOMEM;
1503 		goto out;
1504 	}
1505 
1506 	progs->items[0].prog = prog;
1507 
1508 	bpf_test_timer_enter(&t);
1509 	do {
1510 		ctx.selected_sk = NULL;
1511 		retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run);
1512 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1513 	bpf_test_timer_leave(&t);
1514 
1515 	if (ret < 0)
1516 		goto out;
1517 
1518 	user_ctx->cookie = 0;
1519 	if (ctx.selected_sk) {
1520 		if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
1521 			ret = -EOPNOTSUPP;
1522 			goto out;
1523 		}
1524 
1525 		user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
1526 	}
1527 
1528 	ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1529 	if (!ret)
1530 		ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
1531 
1532 out:
1533 	bpf_prog_array_free(progs);
1534 	kfree(user_ctx);
1535 	return ret;
1536 }
1537 
1538 int bpf_prog_test_run_syscall(struct bpf_prog *prog,
1539 			      const union bpf_attr *kattr,
1540 			      union bpf_attr __user *uattr)
1541 {
1542 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
1543 	__u32 ctx_size_in = kattr->test.ctx_size_in;
1544 	void *ctx = NULL;
1545 	u32 retval;
1546 	int err = 0;
1547 
1548 	/* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
1549 	if (kattr->test.data_in || kattr->test.data_out ||
1550 	    kattr->test.ctx_out || kattr->test.duration ||
1551 	    kattr->test.repeat || kattr->test.flags ||
1552 	    kattr->test.batch_size)
1553 		return -EINVAL;
1554 
1555 	if (ctx_size_in < prog->aux->max_ctx_offset ||
1556 	    ctx_size_in > U16_MAX)
1557 		return -EINVAL;
1558 
1559 	if (ctx_size_in) {
1560 		ctx = memdup_user(ctx_in, ctx_size_in);
1561 		if (IS_ERR(ctx))
1562 			return PTR_ERR(ctx);
1563 	}
1564 
1565 	rcu_read_lock_trace();
1566 	retval = bpf_prog_run_pin_on_cpu(prog, ctx);
1567 	rcu_read_unlock_trace();
1568 
1569 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
1570 		err = -EFAULT;
1571 		goto out;
1572 	}
1573 	if (ctx_size_in)
1574 		if (copy_to_user(ctx_in, ctx, ctx_size_in))
1575 			err = -EFAULT;
1576 out:
1577 	kfree(ctx);
1578 	return err;
1579 }
1580 
1581 static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = {
1582 	.owner        = THIS_MODULE,
1583 	.check_set    = &test_sk_check_kfunc_ids,
1584 	.acquire_set  = &test_sk_acquire_kfunc_ids,
1585 	.release_set  = &test_sk_release_kfunc_ids,
1586 	.ret_null_set = &test_sk_ret_null_kfunc_ids,
1587 };
1588 
1589 static int __init bpf_prog_test_run_init(void)
1590 {
1591 	return register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set);
1592 }
1593 late_initcall(bpf_prog_test_run_init);
1594