xref: /openbmc/linux/net/bpf/test_run.c (revision c4a7b9b5)
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_member1 {
555 	int a;
556 };
557 
558 struct prog_test_member {
559 	struct prog_test_member1 m;
560 	int c;
561 };
562 
563 struct prog_test_ref_kfunc {
564 	int a;
565 	int b;
566 	struct prog_test_member memb;
567 	struct prog_test_ref_kfunc *next;
568 	refcount_t cnt;
569 };
570 
571 static struct prog_test_ref_kfunc prog_test_struct = {
572 	.a = 42,
573 	.b = 108,
574 	.next = &prog_test_struct,
575 	.cnt = REFCOUNT_INIT(1),
576 };
577 
578 noinline struct prog_test_ref_kfunc *
579 bpf_kfunc_call_test_acquire(unsigned long *scalar_ptr)
580 {
581 	refcount_inc(&prog_test_struct.cnt);
582 	return &prog_test_struct;
583 }
584 
585 noinline struct prog_test_member *
586 bpf_kfunc_call_memb_acquire(void)
587 {
588 	WARN_ON_ONCE(1);
589 	return NULL;
590 }
591 
592 noinline void bpf_kfunc_call_test_release(struct prog_test_ref_kfunc *p)
593 {
594 	if (!p)
595 		return;
596 
597 	refcount_dec(&p->cnt);
598 }
599 
600 noinline void bpf_kfunc_call_memb_release(struct prog_test_member *p)
601 {
602 }
603 
604 noinline void bpf_kfunc_call_memb1_release(struct prog_test_member1 *p)
605 {
606 	WARN_ON_ONCE(1);
607 }
608 
609 static int *__bpf_kfunc_call_test_get_mem(struct prog_test_ref_kfunc *p, const int size)
610 {
611 	if (size > 2 * sizeof(int))
612 		return NULL;
613 
614 	return (int *)p;
615 }
616 
617 noinline int *bpf_kfunc_call_test_get_rdwr_mem(struct prog_test_ref_kfunc *p, const int rdwr_buf_size)
618 {
619 	return __bpf_kfunc_call_test_get_mem(p, rdwr_buf_size);
620 }
621 
622 noinline int *bpf_kfunc_call_test_get_rdonly_mem(struct prog_test_ref_kfunc *p, const int rdonly_buf_size)
623 {
624 	return __bpf_kfunc_call_test_get_mem(p, rdonly_buf_size);
625 }
626 
627 /* the next 2 ones can't be really used for testing expect to ensure
628  * that the verifier rejects the call.
629  * Acquire functions must return struct pointers, so these ones are
630  * failing.
631  */
632 noinline int *bpf_kfunc_call_test_acq_rdonly_mem(struct prog_test_ref_kfunc *p, const int rdonly_buf_size)
633 {
634 	return __bpf_kfunc_call_test_get_mem(p, rdonly_buf_size);
635 }
636 
637 noinline void bpf_kfunc_call_int_mem_release(int *p)
638 {
639 }
640 
641 noinline struct prog_test_ref_kfunc *
642 bpf_kfunc_call_test_kptr_get(struct prog_test_ref_kfunc **pp, int a, int b)
643 {
644 	struct prog_test_ref_kfunc *p = READ_ONCE(*pp);
645 
646 	if (!p)
647 		return NULL;
648 	refcount_inc(&p->cnt);
649 	return p;
650 }
651 
652 struct prog_test_pass1 {
653 	int x0;
654 	struct {
655 		int x1;
656 		struct {
657 			int x2;
658 			struct {
659 				int x3;
660 			};
661 		};
662 	};
663 };
664 
665 struct prog_test_pass2 {
666 	int len;
667 	short arr1[4];
668 	struct {
669 		char arr2[4];
670 		unsigned long arr3[8];
671 	} x;
672 };
673 
674 struct prog_test_fail1 {
675 	void *p;
676 	int x;
677 };
678 
679 struct prog_test_fail2 {
680 	int x8;
681 	struct prog_test_pass1 x;
682 };
683 
684 struct prog_test_fail3 {
685 	int len;
686 	char arr1[2];
687 	char arr2[];
688 };
689 
690 noinline void bpf_kfunc_call_test_pass_ctx(struct __sk_buff *skb)
691 {
692 }
693 
694 noinline void bpf_kfunc_call_test_pass1(struct prog_test_pass1 *p)
695 {
696 }
697 
698 noinline void bpf_kfunc_call_test_pass2(struct prog_test_pass2 *p)
699 {
700 }
701 
702 noinline void bpf_kfunc_call_test_fail1(struct prog_test_fail1 *p)
703 {
704 }
705 
706 noinline void bpf_kfunc_call_test_fail2(struct prog_test_fail2 *p)
707 {
708 }
709 
710 noinline void bpf_kfunc_call_test_fail3(struct prog_test_fail3 *p)
711 {
712 }
713 
714 noinline void bpf_kfunc_call_test_mem_len_pass1(void *mem, int mem__sz)
715 {
716 }
717 
718 noinline void bpf_kfunc_call_test_mem_len_fail1(void *mem, int len)
719 {
720 }
721 
722 noinline void bpf_kfunc_call_test_mem_len_fail2(u64 *mem, int len)
723 {
724 }
725 
726 noinline void bpf_kfunc_call_test_ref(struct prog_test_ref_kfunc *p)
727 {
728 }
729 
730 noinline void bpf_kfunc_call_test_destructive(void)
731 {
732 }
733 
734 __diag_pop();
735 
736 ALLOW_ERROR_INJECTION(bpf_modify_return_test, ERRNO);
737 
738 BTF_SET8_START(test_sk_check_kfunc_ids)
739 BTF_ID_FLAGS(func, bpf_kfunc_call_test1)
740 BTF_ID_FLAGS(func, bpf_kfunc_call_test2)
741 BTF_ID_FLAGS(func, bpf_kfunc_call_test3)
742 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acquire, KF_ACQUIRE | KF_RET_NULL)
743 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_acquire, KF_ACQUIRE | KF_RET_NULL)
744 BTF_ID_FLAGS(func, bpf_kfunc_call_test_release, KF_RELEASE)
745 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_release, KF_RELEASE)
746 BTF_ID_FLAGS(func, bpf_kfunc_call_memb1_release, KF_RELEASE)
747 BTF_ID_FLAGS(func, bpf_kfunc_call_test_get_rdwr_mem, KF_RET_NULL)
748 BTF_ID_FLAGS(func, bpf_kfunc_call_test_get_rdonly_mem, KF_RET_NULL)
749 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acq_rdonly_mem, KF_ACQUIRE | KF_RET_NULL)
750 BTF_ID_FLAGS(func, bpf_kfunc_call_int_mem_release, KF_RELEASE)
751 BTF_ID_FLAGS(func, bpf_kfunc_call_test_kptr_get, KF_ACQUIRE | KF_RET_NULL | KF_KPTR_GET)
752 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass_ctx)
753 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass1)
754 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass2)
755 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail1)
756 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail2)
757 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail3)
758 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_pass1)
759 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail1)
760 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail2)
761 BTF_ID_FLAGS(func, bpf_kfunc_call_test_ref, KF_TRUSTED_ARGS)
762 BTF_ID_FLAGS(func, bpf_kfunc_call_test_destructive, KF_DESTRUCTIVE)
763 BTF_SET8_END(test_sk_check_kfunc_ids)
764 
765 static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
766 			   u32 size, u32 headroom, u32 tailroom)
767 {
768 	void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
769 	void *data;
770 
771 	if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
772 		return ERR_PTR(-EINVAL);
773 
774 	if (user_size > size)
775 		return ERR_PTR(-EMSGSIZE);
776 
777 	data = kzalloc(size + headroom + tailroom, GFP_USER);
778 	if (!data)
779 		return ERR_PTR(-ENOMEM);
780 
781 	if (copy_from_user(data + headroom, data_in, user_size)) {
782 		kfree(data);
783 		return ERR_PTR(-EFAULT);
784 	}
785 
786 	return data;
787 }
788 
789 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
790 			      const union bpf_attr *kattr,
791 			      union bpf_attr __user *uattr)
792 {
793 	struct bpf_fentry_test_t arg = {};
794 	u16 side_effect = 0, ret = 0;
795 	int b = 2, err = -EFAULT;
796 	u32 retval = 0;
797 
798 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
799 		return -EINVAL;
800 
801 	switch (prog->expected_attach_type) {
802 	case BPF_TRACE_FENTRY:
803 	case BPF_TRACE_FEXIT:
804 		if (bpf_fentry_test1(1) != 2 ||
805 		    bpf_fentry_test2(2, 3) != 5 ||
806 		    bpf_fentry_test3(4, 5, 6) != 15 ||
807 		    bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
808 		    bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
809 		    bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
810 		    bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
811 		    bpf_fentry_test8(&arg) != 0)
812 			goto out;
813 		break;
814 	case BPF_MODIFY_RETURN:
815 		ret = bpf_modify_return_test(1, &b);
816 		if (b != 2)
817 			side_effect = 1;
818 		break;
819 	default:
820 		goto out;
821 	}
822 
823 	retval = ((u32)side_effect << 16) | ret;
824 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
825 		goto out;
826 
827 	err = 0;
828 out:
829 	trace_bpf_test_finish(&err);
830 	return err;
831 }
832 
833 struct bpf_raw_tp_test_run_info {
834 	struct bpf_prog *prog;
835 	void *ctx;
836 	u32 retval;
837 };
838 
839 static void
840 __bpf_prog_test_run_raw_tp(void *data)
841 {
842 	struct bpf_raw_tp_test_run_info *info = data;
843 
844 	rcu_read_lock();
845 	info->retval = bpf_prog_run(info->prog, info->ctx);
846 	rcu_read_unlock();
847 }
848 
849 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
850 			     const union bpf_attr *kattr,
851 			     union bpf_attr __user *uattr)
852 {
853 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
854 	__u32 ctx_size_in = kattr->test.ctx_size_in;
855 	struct bpf_raw_tp_test_run_info info;
856 	int cpu = kattr->test.cpu, err = 0;
857 	int current_cpu;
858 
859 	/* doesn't support data_in/out, ctx_out, duration, or repeat */
860 	if (kattr->test.data_in || kattr->test.data_out ||
861 	    kattr->test.ctx_out || kattr->test.duration ||
862 	    kattr->test.repeat || kattr->test.batch_size)
863 		return -EINVAL;
864 
865 	if (ctx_size_in < prog->aux->max_ctx_offset ||
866 	    ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
867 		return -EINVAL;
868 
869 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
870 		return -EINVAL;
871 
872 	if (ctx_size_in) {
873 		info.ctx = memdup_user(ctx_in, ctx_size_in);
874 		if (IS_ERR(info.ctx))
875 			return PTR_ERR(info.ctx);
876 	} else {
877 		info.ctx = NULL;
878 	}
879 
880 	info.prog = prog;
881 
882 	current_cpu = get_cpu();
883 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
884 	    cpu == current_cpu) {
885 		__bpf_prog_test_run_raw_tp(&info);
886 	} else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
887 		/* smp_call_function_single() also checks cpu_online()
888 		 * after csd_lock(). However, since cpu is from user
889 		 * space, let's do an extra quick check to filter out
890 		 * invalid value before smp_call_function_single().
891 		 */
892 		err = -ENXIO;
893 	} else {
894 		err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
895 					       &info, 1);
896 	}
897 	put_cpu();
898 
899 	if (!err &&
900 	    copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
901 		err = -EFAULT;
902 
903 	kfree(info.ctx);
904 	return err;
905 }
906 
907 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
908 {
909 	void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
910 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
911 	u32 size = kattr->test.ctx_size_in;
912 	void *data;
913 	int err;
914 
915 	if (!data_in && !data_out)
916 		return NULL;
917 
918 	data = kzalloc(max_size, GFP_USER);
919 	if (!data)
920 		return ERR_PTR(-ENOMEM);
921 
922 	if (data_in) {
923 		err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
924 		if (err) {
925 			kfree(data);
926 			return ERR_PTR(err);
927 		}
928 
929 		size = min_t(u32, max_size, size);
930 		if (copy_from_user(data, data_in, size)) {
931 			kfree(data);
932 			return ERR_PTR(-EFAULT);
933 		}
934 	}
935 	return data;
936 }
937 
938 static int bpf_ctx_finish(const union bpf_attr *kattr,
939 			  union bpf_attr __user *uattr, const void *data,
940 			  u32 size)
941 {
942 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
943 	int err = -EFAULT;
944 	u32 copy_size = size;
945 
946 	if (!data || !data_out)
947 		return 0;
948 
949 	if (copy_size > kattr->test.ctx_size_out) {
950 		copy_size = kattr->test.ctx_size_out;
951 		err = -ENOSPC;
952 	}
953 
954 	if (copy_to_user(data_out, data, copy_size))
955 		goto out;
956 	if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
957 		goto out;
958 	if (err != -ENOSPC)
959 		err = 0;
960 out:
961 	return err;
962 }
963 
964 /**
965  * range_is_zero - test whether buffer is initialized
966  * @buf: buffer to check
967  * @from: check from this position
968  * @to: check up until (excluding) this position
969  *
970  * This function returns true if the there is a non-zero byte
971  * in the buf in the range [from,to).
972  */
973 static inline bool range_is_zero(void *buf, size_t from, size_t to)
974 {
975 	return !memchr_inv((u8 *)buf + from, 0, to - from);
976 }
977 
978 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
979 {
980 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
981 
982 	if (!skb->len)
983 		return -EINVAL;
984 
985 	if (!__skb)
986 		return 0;
987 
988 	/* make sure the fields we don't use are zeroed */
989 	if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
990 		return -EINVAL;
991 
992 	/* mark is allowed */
993 
994 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
995 			   offsetof(struct __sk_buff, priority)))
996 		return -EINVAL;
997 
998 	/* priority is allowed */
999 	/* ingress_ifindex is allowed */
1000 	/* ifindex is allowed */
1001 
1002 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
1003 			   offsetof(struct __sk_buff, cb)))
1004 		return -EINVAL;
1005 
1006 	/* cb is allowed */
1007 
1008 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
1009 			   offsetof(struct __sk_buff, tstamp)))
1010 		return -EINVAL;
1011 
1012 	/* tstamp is allowed */
1013 	/* wire_len is allowed */
1014 	/* gso_segs is allowed */
1015 
1016 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
1017 			   offsetof(struct __sk_buff, gso_size)))
1018 		return -EINVAL;
1019 
1020 	/* gso_size is allowed */
1021 
1022 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
1023 			   offsetof(struct __sk_buff, hwtstamp)))
1024 		return -EINVAL;
1025 
1026 	/* hwtstamp is allowed */
1027 
1028 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp),
1029 			   sizeof(struct __sk_buff)))
1030 		return -EINVAL;
1031 
1032 	skb->mark = __skb->mark;
1033 	skb->priority = __skb->priority;
1034 	skb->skb_iif = __skb->ingress_ifindex;
1035 	skb->tstamp = __skb->tstamp;
1036 	memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
1037 
1038 	if (__skb->wire_len == 0) {
1039 		cb->pkt_len = skb->len;
1040 	} else {
1041 		if (__skb->wire_len < skb->len ||
1042 		    __skb->wire_len > GSO_LEGACY_MAX_SIZE)
1043 			return -EINVAL;
1044 		cb->pkt_len = __skb->wire_len;
1045 	}
1046 
1047 	if (__skb->gso_segs > GSO_MAX_SEGS)
1048 		return -EINVAL;
1049 	skb_shinfo(skb)->gso_segs = __skb->gso_segs;
1050 	skb_shinfo(skb)->gso_size = __skb->gso_size;
1051 	skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp;
1052 
1053 	return 0;
1054 }
1055 
1056 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
1057 {
1058 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
1059 
1060 	if (!__skb)
1061 		return;
1062 
1063 	__skb->mark = skb->mark;
1064 	__skb->priority = skb->priority;
1065 	__skb->ingress_ifindex = skb->skb_iif;
1066 	__skb->ifindex = skb->dev->ifindex;
1067 	__skb->tstamp = skb->tstamp;
1068 	memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
1069 	__skb->wire_len = cb->pkt_len;
1070 	__skb->gso_segs = skb_shinfo(skb)->gso_segs;
1071 	__skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp;
1072 }
1073 
1074 static struct proto bpf_dummy_proto = {
1075 	.name   = "bpf_dummy",
1076 	.owner  = THIS_MODULE,
1077 	.obj_size = sizeof(struct sock),
1078 };
1079 
1080 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
1081 			  union bpf_attr __user *uattr)
1082 {
1083 	bool is_l2 = false, is_direct_pkt_access = false;
1084 	struct net *net = current->nsproxy->net_ns;
1085 	struct net_device *dev = net->loopback_dev;
1086 	u32 size = kattr->test.data_size_in;
1087 	u32 repeat = kattr->test.repeat;
1088 	struct __sk_buff *ctx = NULL;
1089 	u32 retval, duration;
1090 	int hh_len = ETH_HLEN;
1091 	struct sk_buff *skb;
1092 	struct sock *sk;
1093 	void *data;
1094 	int ret;
1095 
1096 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1097 		return -EINVAL;
1098 
1099 	data = bpf_test_init(kattr, kattr->test.data_size_in,
1100 			     size, NET_SKB_PAD + NET_IP_ALIGN,
1101 			     SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1102 	if (IS_ERR(data))
1103 		return PTR_ERR(data);
1104 
1105 	ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
1106 	if (IS_ERR(ctx)) {
1107 		kfree(data);
1108 		return PTR_ERR(ctx);
1109 	}
1110 
1111 	switch (prog->type) {
1112 	case BPF_PROG_TYPE_SCHED_CLS:
1113 	case BPF_PROG_TYPE_SCHED_ACT:
1114 		is_l2 = true;
1115 		fallthrough;
1116 	case BPF_PROG_TYPE_LWT_IN:
1117 	case BPF_PROG_TYPE_LWT_OUT:
1118 	case BPF_PROG_TYPE_LWT_XMIT:
1119 		is_direct_pkt_access = true;
1120 		break;
1121 	default:
1122 		break;
1123 	}
1124 
1125 	sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
1126 	if (!sk) {
1127 		kfree(data);
1128 		kfree(ctx);
1129 		return -ENOMEM;
1130 	}
1131 	sock_init_data(NULL, sk);
1132 
1133 	skb = build_skb(data, 0);
1134 	if (!skb) {
1135 		kfree(data);
1136 		kfree(ctx);
1137 		sk_free(sk);
1138 		return -ENOMEM;
1139 	}
1140 	skb->sk = sk;
1141 
1142 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1143 	__skb_put(skb, size);
1144 	if (ctx && ctx->ifindex > 1) {
1145 		dev = dev_get_by_index(net, ctx->ifindex);
1146 		if (!dev) {
1147 			ret = -ENODEV;
1148 			goto out;
1149 		}
1150 	}
1151 	skb->protocol = eth_type_trans(skb, dev);
1152 	skb_reset_network_header(skb);
1153 
1154 	switch (skb->protocol) {
1155 	case htons(ETH_P_IP):
1156 		sk->sk_family = AF_INET;
1157 		if (sizeof(struct iphdr) <= skb_headlen(skb)) {
1158 			sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
1159 			sk->sk_daddr = ip_hdr(skb)->daddr;
1160 		}
1161 		break;
1162 #if IS_ENABLED(CONFIG_IPV6)
1163 	case htons(ETH_P_IPV6):
1164 		sk->sk_family = AF_INET6;
1165 		if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
1166 			sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
1167 			sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
1168 		}
1169 		break;
1170 #endif
1171 	default:
1172 		break;
1173 	}
1174 
1175 	if (is_l2)
1176 		__skb_push(skb, hh_len);
1177 	if (is_direct_pkt_access)
1178 		bpf_compute_data_pointers(skb);
1179 	ret = convert___skb_to_skb(skb, ctx);
1180 	if (ret)
1181 		goto out;
1182 	ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
1183 	if (ret)
1184 		goto out;
1185 	if (!is_l2) {
1186 		if (skb_headroom(skb) < hh_len) {
1187 			int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
1188 
1189 			if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
1190 				ret = -ENOMEM;
1191 				goto out;
1192 			}
1193 		}
1194 		memset(__skb_push(skb, hh_len), 0, hh_len);
1195 	}
1196 	convert_skb_to___skb(skb, ctx);
1197 
1198 	size = skb->len;
1199 	/* bpf program can never convert linear skb to non-linear */
1200 	if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
1201 		size = skb_headlen(skb);
1202 	ret = bpf_test_finish(kattr, uattr, skb->data, NULL, size, retval,
1203 			      duration);
1204 	if (!ret)
1205 		ret = bpf_ctx_finish(kattr, uattr, ctx,
1206 				     sizeof(struct __sk_buff));
1207 out:
1208 	if (dev && dev != net->loopback_dev)
1209 		dev_put(dev);
1210 	kfree_skb(skb);
1211 	sk_free(sk);
1212 	kfree(ctx);
1213 	return ret;
1214 }
1215 
1216 static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp)
1217 {
1218 	unsigned int ingress_ifindex, rx_queue_index;
1219 	struct netdev_rx_queue *rxqueue;
1220 	struct net_device *device;
1221 
1222 	if (!xdp_md)
1223 		return 0;
1224 
1225 	if (xdp_md->egress_ifindex != 0)
1226 		return -EINVAL;
1227 
1228 	ingress_ifindex = xdp_md->ingress_ifindex;
1229 	rx_queue_index = xdp_md->rx_queue_index;
1230 
1231 	if (!ingress_ifindex && rx_queue_index)
1232 		return -EINVAL;
1233 
1234 	if (ingress_ifindex) {
1235 		device = dev_get_by_index(current->nsproxy->net_ns,
1236 					  ingress_ifindex);
1237 		if (!device)
1238 			return -ENODEV;
1239 
1240 		if (rx_queue_index >= device->real_num_rx_queues)
1241 			goto free_dev;
1242 
1243 		rxqueue = __netif_get_rx_queue(device, rx_queue_index);
1244 
1245 		if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq))
1246 			goto free_dev;
1247 
1248 		xdp->rxq = &rxqueue->xdp_rxq;
1249 		/* The device is now tracked in the xdp->rxq for later
1250 		 * dev_put()
1251 		 */
1252 	}
1253 
1254 	xdp->data = xdp->data_meta + xdp_md->data;
1255 	return 0;
1256 
1257 free_dev:
1258 	dev_put(device);
1259 	return -EINVAL;
1260 }
1261 
1262 static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md)
1263 {
1264 	if (!xdp_md)
1265 		return;
1266 
1267 	xdp_md->data = xdp->data - xdp->data_meta;
1268 	xdp_md->data_end = xdp->data_end - xdp->data_meta;
1269 
1270 	if (xdp_md->ingress_ifindex)
1271 		dev_put(xdp->rxq->dev);
1272 }
1273 
1274 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
1275 			  union bpf_attr __user *uattr)
1276 {
1277 	bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES);
1278 	u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1279 	u32 batch_size = kattr->test.batch_size;
1280 	u32 retval = 0, duration, max_data_sz;
1281 	u32 size = kattr->test.data_size_in;
1282 	u32 headroom = XDP_PACKET_HEADROOM;
1283 	u32 repeat = kattr->test.repeat;
1284 	struct netdev_rx_queue *rxqueue;
1285 	struct skb_shared_info *sinfo;
1286 	struct xdp_buff xdp = {};
1287 	int i, ret = -EINVAL;
1288 	struct xdp_md *ctx;
1289 	void *data;
1290 
1291 	if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
1292 	    prog->expected_attach_type == BPF_XDP_CPUMAP)
1293 		return -EINVAL;
1294 
1295 	if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES)
1296 		return -EINVAL;
1297 
1298 	if (do_live) {
1299 		if (!batch_size)
1300 			batch_size = NAPI_POLL_WEIGHT;
1301 		else if (batch_size > TEST_XDP_MAX_BATCH)
1302 			return -E2BIG;
1303 
1304 		headroom += sizeof(struct xdp_page_head);
1305 	} else if (batch_size) {
1306 		return -EINVAL;
1307 	}
1308 
1309 	ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md));
1310 	if (IS_ERR(ctx))
1311 		return PTR_ERR(ctx);
1312 
1313 	if (ctx) {
1314 		/* There can't be user provided data before the meta data */
1315 		if (ctx->data_meta || ctx->data_end != size ||
1316 		    ctx->data > ctx->data_end ||
1317 		    unlikely(xdp_metalen_invalid(ctx->data)) ||
1318 		    (do_live && (kattr->test.data_out || kattr->test.ctx_out)))
1319 			goto free_ctx;
1320 		/* Meta data is allocated from the headroom */
1321 		headroom -= ctx->data;
1322 	}
1323 
1324 	max_data_sz = 4096 - headroom - tailroom;
1325 	if (size > max_data_sz) {
1326 		/* disallow live data mode for jumbo frames */
1327 		if (do_live)
1328 			goto free_ctx;
1329 		size = max_data_sz;
1330 	}
1331 
1332 	data = bpf_test_init(kattr, size, max_data_sz, headroom, tailroom);
1333 	if (IS_ERR(data)) {
1334 		ret = PTR_ERR(data);
1335 		goto free_ctx;
1336 	}
1337 
1338 	rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
1339 	rxqueue->xdp_rxq.frag_size = headroom + max_data_sz + tailroom;
1340 	xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq);
1341 	xdp_prepare_buff(&xdp, data, headroom, size, true);
1342 	sinfo = xdp_get_shared_info_from_buff(&xdp);
1343 
1344 	ret = xdp_convert_md_to_buff(ctx, &xdp);
1345 	if (ret)
1346 		goto free_data;
1347 
1348 	if (unlikely(kattr->test.data_size_in > size)) {
1349 		void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
1350 
1351 		while (size < kattr->test.data_size_in) {
1352 			struct page *page;
1353 			skb_frag_t *frag;
1354 			u32 data_len;
1355 
1356 			if (sinfo->nr_frags == MAX_SKB_FRAGS) {
1357 				ret = -ENOMEM;
1358 				goto out;
1359 			}
1360 
1361 			page = alloc_page(GFP_KERNEL);
1362 			if (!page) {
1363 				ret = -ENOMEM;
1364 				goto out;
1365 			}
1366 
1367 			frag = &sinfo->frags[sinfo->nr_frags++];
1368 			__skb_frag_set_page(frag, page);
1369 
1370 			data_len = min_t(u32, kattr->test.data_size_in - size,
1371 					 PAGE_SIZE);
1372 			skb_frag_size_set(frag, data_len);
1373 
1374 			if (copy_from_user(page_address(page), data_in + size,
1375 					   data_len)) {
1376 				ret = -EFAULT;
1377 				goto out;
1378 			}
1379 			sinfo->xdp_frags_size += data_len;
1380 			size += data_len;
1381 		}
1382 		xdp_buff_set_frags_flag(&xdp);
1383 	}
1384 
1385 	if (repeat > 1)
1386 		bpf_prog_change_xdp(NULL, prog);
1387 
1388 	if (do_live)
1389 		ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration);
1390 	else
1391 		ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
1392 	/* We convert the xdp_buff back to an xdp_md before checking the return
1393 	 * code so the reference count of any held netdevice will be decremented
1394 	 * even if the test run failed.
1395 	 */
1396 	xdp_convert_buff_to_md(&xdp, ctx);
1397 	if (ret)
1398 		goto out;
1399 
1400 	size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size;
1401 	ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size,
1402 			      retval, duration);
1403 	if (!ret)
1404 		ret = bpf_ctx_finish(kattr, uattr, ctx,
1405 				     sizeof(struct xdp_md));
1406 
1407 out:
1408 	if (repeat > 1)
1409 		bpf_prog_change_xdp(prog, NULL);
1410 free_data:
1411 	for (i = 0; i < sinfo->nr_frags; i++)
1412 		__free_page(skb_frag_page(&sinfo->frags[i]));
1413 	kfree(data);
1414 free_ctx:
1415 	kfree(ctx);
1416 	return ret;
1417 }
1418 
1419 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
1420 {
1421 	/* make sure the fields we don't use are zeroed */
1422 	if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
1423 		return -EINVAL;
1424 
1425 	/* flags is allowed */
1426 
1427 	if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
1428 			   sizeof(struct bpf_flow_keys)))
1429 		return -EINVAL;
1430 
1431 	return 0;
1432 }
1433 
1434 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1435 				     const union bpf_attr *kattr,
1436 				     union bpf_attr __user *uattr)
1437 {
1438 	struct bpf_test_timer t = { NO_PREEMPT };
1439 	u32 size = kattr->test.data_size_in;
1440 	struct bpf_flow_dissector ctx = {};
1441 	u32 repeat = kattr->test.repeat;
1442 	struct bpf_flow_keys *user_ctx;
1443 	struct bpf_flow_keys flow_keys;
1444 	const struct ethhdr *eth;
1445 	unsigned int flags = 0;
1446 	u32 retval, duration;
1447 	void *data;
1448 	int ret;
1449 
1450 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1451 		return -EINVAL;
1452 
1453 	if (size < ETH_HLEN)
1454 		return -EINVAL;
1455 
1456 	data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0);
1457 	if (IS_ERR(data))
1458 		return PTR_ERR(data);
1459 
1460 	eth = (struct ethhdr *)data;
1461 
1462 	if (!repeat)
1463 		repeat = 1;
1464 
1465 	user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
1466 	if (IS_ERR(user_ctx)) {
1467 		kfree(data);
1468 		return PTR_ERR(user_ctx);
1469 	}
1470 	if (user_ctx) {
1471 		ret = verify_user_bpf_flow_keys(user_ctx);
1472 		if (ret)
1473 			goto out;
1474 		flags = user_ctx->flags;
1475 	}
1476 
1477 	ctx.flow_keys = &flow_keys;
1478 	ctx.data = data;
1479 	ctx.data_end = (__u8 *)data + size;
1480 
1481 	bpf_test_timer_enter(&t);
1482 	do {
1483 		retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
1484 					  size, flags);
1485 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1486 	bpf_test_timer_leave(&t);
1487 
1488 	if (ret < 0)
1489 		goto out;
1490 
1491 	ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL,
1492 			      sizeof(flow_keys), retval, duration);
1493 	if (!ret)
1494 		ret = bpf_ctx_finish(kattr, uattr, user_ctx,
1495 				     sizeof(struct bpf_flow_keys));
1496 
1497 out:
1498 	kfree(user_ctx);
1499 	kfree(data);
1500 	return ret;
1501 }
1502 
1503 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
1504 				union bpf_attr __user *uattr)
1505 {
1506 	struct bpf_test_timer t = { NO_PREEMPT };
1507 	struct bpf_prog_array *progs = NULL;
1508 	struct bpf_sk_lookup_kern ctx = {};
1509 	u32 repeat = kattr->test.repeat;
1510 	struct bpf_sk_lookup *user_ctx;
1511 	u32 retval, duration;
1512 	int ret = -EINVAL;
1513 
1514 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1515 		return -EINVAL;
1516 
1517 	if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
1518 	    kattr->test.data_size_out)
1519 		return -EINVAL;
1520 
1521 	if (!repeat)
1522 		repeat = 1;
1523 
1524 	user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
1525 	if (IS_ERR(user_ctx))
1526 		return PTR_ERR(user_ctx);
1527 
1528 	if (!user_ctx)
1529 		return -EINVAL;
1530 
1531 	if (user_ctx->sk)
1532 		goto out;
1533 
1534 	if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
1535 		goto out;
1536 
1537 	if (user_ctx->local_port > U16_MAX) {
1538 		ret = -ERANGE;
1539 		goto out;
1540 	}
1541 
1542 	ctx.family = (u16)user_ctx->family;
1543 	ctx.protocol = (u16)user_ctx->protocol;
1544 	ctx.dport = (u16)user_ctx->local_port;
1545 	ctx.sport = user_ctx->remote_port;
1546 
1547 	switch (ctx.family) {
1548 	case AF_INET:
1549 		ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
1550 		ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
1551 		break;
1552 
1553 #if IS_ENABLED(CONFIG_IPV6)
1554 	case AF_INET6:
1555 		ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
1556 		ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
1557 		break;
1558 #endif
1559 
1560 	default:
1561 		ret = -EAFNOSUPPORT;
1562 		goto out;
1563 	}
1564 
1565 	progs = bpf_prog_array_alloc(1, GFP_KERNEL);
1566 	if (!progs) {
1567 		ret = -ENOMEM;
1568 		goto out;
1569 	}
1570 
1571 	progs->items[0].prog = prog;
1572 
1573 	bpf_test_timer_enter(&t);
1574 	do {
1575 		ctx.selected_sk = NULL;
1576 		retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run);
1577 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1578 	bpf_test_timer_leave(&t);
1579 
1580 	if (ret < 0)
1581 		goto out;
1582 
1583 	user_ctx->cookie = 0;
1584 	if (ctx.selected_sk) {
1585 		if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
1586 			ret = -EOPNOTSUPP;
1587 			goto out;
1588 		}
1589 
1590 		user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
1591 	}
1592 
1593 	ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1594 	if (!ret)
1595 		ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
1596 
1597 out:
1598 	bpf_prog_array_free(progs);
1599 	kfree(user_ctx);
1600 	return ret;
1601 }
1602 
1603 int bpf_prog_test_run_syscall(struct bpf_prog *prog,
1604 			      const union bpf_attr *kattr,
1605 			      union bpf_attr __user *uattr)
1606 {
1607 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
1608 	__u32 ctx_size_in = kattr->test.ctx_size_in;
1609 	void *ctx = NULL;
1610 	u32 retval;
1611 	int err = 0;
1612 
1613 	/* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
1614 	if (kattr->test.data_in || kattr->test.data_out ||
1615 	    kattr->test.ctx_out || kattr->test.duration ||
1616 	    kattr->test.repeat || kattr->test.flags ||
1617 	    kattr->test.batch_size)
1618 		return -EINVAL;
1619 
1620 	if (ctx_size_in < prog->aux->max_ctx_offset ||
1621 	    ctx_size_in > U16_MAX)
1622 		return -EINVAL;
1623 
1624 	if (ctx_size_in) {
1625 		ctx = memdup_user(ctx_in, ctx_size_in);
1626 		if (IS_ERR(ctx))
1627 			return PTR_ERR(ctx);
1628 	}
1629 
1630 	rcu_read_lock_trace();
1631 	retval = bpf_prog_run_pin_on_cpu(prog, ctx);
1632 	rcu_read_unlock_trace();
1633 
1634 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
1635 		err = -EFAULT;
1636 		goto out;
1637 	}
1638 	if (ctx_size_in)
1639 		if (copy_to_user(ctx_in, ctx, ctx_size_in))
1640 			err = -EFAULT;
1641 out:
1642 	kfree(ctx);
1643 	return err;
1644 }
1645 
1646 static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = {
1647 	.owner = THIS_MODULE,
1648 	.set   = &test_sk_check_kfunc_ids,
1649 };
1650 
1651 BTF_ID_LIST(bpf_prog_test_dtor_kfunc_ids)
1652 BTF_ID(struct, prog_test_ref_kfunc)
1653 BTF_ID(func, bpf_kfunc_call_test_release)
1654 BTF_ID(struct, prog_test_member)
1655 BTF_ID(func, bpf_kfunc_call_memb_release)
1656 
1657 static int __init bpf_prog_test_run_init(void)
1658 {
1659 	const struct btf_id_dtor_kfunc bpf_prog_test_dtor_kfunc[] = {
1660 		{
1661 		  .btf_id       = bpf_prog_test_dtor_kfunc_ids[0],
1662 		  .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[1]
1663 		},
1664 		{
1665 		  .btf_id	= bpf_prog_test_dtor_kfunc_ids[2],
1666 		  .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[3],
1667 		},
1668 	};
1669 	int ret;
1670 
1671 	ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set);
1672 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_prog_test_kfunc_set);
1673 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SYSCALL, &bpf_prog_test_kfunc_set);
1674 	return ret ?: register_btf_id_dtor_kfuncs(bpf_prog_test_dtor_kfunc,
1675 						  ARRAY_SIZE(bpf_prog_test_dtor_kfunc),
1676 						  THIS_MODULE);
1677 }
1678 late_initcall(bpf_prog_test_run_init);
1679