xref: /openbmc/linux/net/bpf/test_run.c (revision 586cb1d6)
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 	size = SKB_DATA_ALIGN(size);
778 	data = kzalloc(size + headroom + tailroom, GFP_USER);
779 	if (!data)
780 		return ERR_PTR(-ENOMEM);
781 
782 	if (copy_from_user(data + headroom, data_in, user_size)) {
783 		kfree(data);
784 		return ERR_PTR(-EFAULT);
785 	}
786 
787 	return data;
788 }
789 
790 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
791 			      const union bpf_attr *kattr,
792 			      union bpf_attr __user *uattr)
793 {
794 	struct bpf_fentry_test_t arg = {};
795 	u16 side_effect = 0, ret = 0;
796 	int b = 2, err = -EFAULT;
797 	u32 retval = 0;
798 
799 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
800 		return -EINVAL;
801 
802 	switch (prog->expected_attach_type) {
803 	case BPF_TRACE_FENTRY:
804 	case BPF_TRACE_FEXIT:
805 		if (bpf_fentry_test1(1) != 2 ||
806 		    bpf_fentry_test2(2, 3) != 5 ||
807 		    bpf_fentry_test3(4, 5, 6) != 15 ||
808 		    bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
809 		    bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
810 		    bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
811 		    bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
812 		    bpf_fentry_test8(&arg) != 0)
813 			goto out;
814 		break;
815 	case BPF_MODIFY_RETURN:
816 		ret = bpf_modify_return_test(1, &b);
817 		if (b != 2)
818 			side_effect = 1;
819 		break;
820 	default:
821 		goto out;
822 	}
823 
824 	retval = ((u32)side_effect << 16) | ret;
825 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
826 		goto out;
827 
828 	err = 0;
829 out:
830 	trace_bpf_test_finish(&err);
831 	return err;
832 }
833 
834 struct bpf_raw_tp_test_run_info {
835 	struct bpf_prog *prog;
836 	void *ctx;
837 	u32 retval;
838 };
839 
840 static void
841 __bpf_prog_test_run_raw_tp(void *data)
842 {
843 	struct bpf_raw_tp_test_run_info *info = data;
844 
845 	rcu_read_lock();
846 	info->retval = bpf_prog_run(info->prog, info->ctx);
847 	rcu_read_unlock();
848 }
849 
850 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
851 			     const union bpf_attr *kattr,
852 			     union bpf_attr __user *uattr)
853 {
854 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
855 	__u32 ctx_size_in = kattr->test.ctx_size_in;
856 	struct bpf_raw_tp_test_run_info info;
857 	int cpu = kattr->test.cpu, err = 0;
858 	int current_cpu;
859 
860 	/* doesn't support data_in/out, ctx_out, duration, or repeat */
861 	if (kattr->test.data_in || kattr->test.data_out ||
862 	    kattr->test.ctx_out || kattr->test.duration ||
863 	    kattr->test.repeat || kattr->test.batch_size)
864 		return -EINVAL;
865 
866 	if (ctx_size_in < prog->aux->max_ctx_offset ||
867 	    ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
868 		return -EINVAL;
869 
870 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
871 		return -EINVAL;
872 
873 	if (ctx_size_in) {
874 		info.ctx = memdup_user(ctx_in, ctx_size_in);
875 		if (IS_ERR(info.ctx))
876 			return PTR_ERR(info.ctx);
877 	} else {
878 		info.ctx = NULL;
879 	}
880 
881 	info.prog = prog;
882 
883 	current_cpu = get_cpu();
884 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
885 	    cpu == current_cpu) {
886 		__bpf_prog_test_run_raw_tp(&info);
887 	} else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
888 		/* smp_call_function_single() also checks cpu_online()
889 		 * after csd_lock(). However, since cpu is from user
890 		 * space, let's do an extra quick check to filter out
891 		 * invalid value before smp_call_function_single().
892 		 */
893 		err = -ENXIO;
894 	} else {
895 		err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
896 					       &info, 1);
897 	}
898 	put_cpu();
899 
900 	if (!err &&
901 	    copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
902 		err = -EFAULT;
903 
904 	kfree(info.ctx);
905 	return err;
906 }
907 
908 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
909 {
910 	void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
911 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
912 	u32 size = kattr->test.ctx_size_in;
913 	void *data;
914 	int err;
915 
916 	if (!data_in && !data_out)
917 		return NULL;
918 
919 	data = kzalloc(max_size, GFP_USER);
920 	if (!data)
921 		return ERR_PTR(-ENOMEM);
922 
923 	if (data_in) {
924 		err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
925 		if (err) {
926 			kfree(data);
927 			return ERR_PTR(err);
928 		}
929 
930 		size = min_t(u32, max_size, size);
931 		if (copy_from_user(data, data_in, size)) {
932 			kfree(data);
933 			return ERR_PTR(-EFAULT);
934 		}
935 	}
936 	return data;
937 }
938 
939 static int bpf_ctx_finish(const union bpf_attr *kattr,
940 			  union bpf_attr __user *uattr, const void *data,
941 			  u32 size)
942 {
943 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
944 	int err = -EFAULT;
945 	u32 copy_size = size;
946 
947 	if (!data || !data_out)
948 		return 0;
949 
950 	if (copy_size > kattr->test.ctx_size_out) {
951 		copy_size = kattr->test.ctx_size_out;
952 		err = -ENOSPC;
953 	}
954 
955 	if (copy_to_user(data_out, data, copy_size))
956 		goto out;
957 	if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
958 		goto out;
959 	if (err != -ENOSPC)
960 		err = 0;
961 out:
962 	return err;
963 }
964 
965 /**
966  * range_is_zero - test whether buffer is initialized
967  * @buf: buffer to check
968  * @from: check from this position
969  * @to: check up until (excluding) this position
970  *
971  * This function returns true if the there is a non-zero byte
972  * in the buf in the range [from,to).
973  */
974 static inline bool range_is_zero(void *buf, size_t from, size_t to)
975 {
976 	return !memchr_inv((u8 *)buf + from, 0, to - from);
977 }
978 
979 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
980 {
981 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
982 
983 	if (!skb->len)
984 		return -EINVAL;
985 
986 	if (!__skb)
987 		return 0;
988 
989 	/* make sure the fields we don't use are zeroed */
990 	if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
991 		return -EINVAL;
992 
993 	/* mark is allowed */
994 
995 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
996 			   offsetof(struct __sk_buff, priority)))
997 		return -EINVAL;
998 
999 	/* priority is allowed */
1000 	/* ingress_ifindex is allowed */
1001 	/* ifindex is allowed */
1002 
1003 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
1004 			   offsetof(struct __sk_buff, cb)))
1005 		return -EINVAL;
1006 
1007 	/* cb is allowed */
1008 
1009 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
1010 			   offsetof(struct __sk_buff, tstamp)))
1011 		return -EINVAL;
1012 
1013 	/* tstamp is allowed */
1014 	/* wire_len is allowed */
1015 	/* gso_segs is allowed */
1016 
1017 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
1018 			   offsetof(struct __sk_buff, gso_size)))
1019 		return -EINVAL;
1020 
1021 	/* gso_size is allowed */
1022 
1023 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
1024 			   offsetof(struct __sk_buff, hwtstamp)))
1025 		return -EINVAL;
1026 
1027 	/* hwtstamp is allowed */
1028 
1029 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp),
1030 			   sizeof(struct __sk_buff)))
1031 		return -EINVAL;
1032 
1033 	skb->mark = __skb->mark;
1034 	skb->priority = __skb->priority;
1035 	skb->skb_iif = __skb->ingress_ifindex;
1036 	skb->tstamp = __skb->tstamp;
1037 	memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
1038 
1039 	if (__skb->wire_len == 0) {
1040 		cb->pkt_len = skb->len;
1041 	} else {
1042 		if (__skb->wire_len < skb->len ||
1043 		    __skb->wire_len > GSO_LEGACY_MAX_SIZE)
1044 			return -EINVAL;
1045 		cb->pkt_len = __skb->wire_len;
1046 	}
1047 
1048 	if (__skb->gso_segs > GSO_MAX_SEGS)
1049 		return -EINVAL;
1050 	skb_shinfo(skb)->gso_segs = __skb->gso_segs;
1051 	skb_shinfo(skb)->gso_size = __skb->gso_size;
1052 	skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp;
1053 
1054 	return 0;
1055 }
1056 
1057 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
1058 {
1059 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
1060 
1061 	if (!__skb)
1062 		return;
1063 
1064 	__skb->mark = skb->mark;
1065 	__skb->priority = skb->priority;
1066 	__skb->ingress_ifindex = skb->skb_iif;
1067 	__skb->ifindex = skb->dev->ifindex;
1068 	__skb->tstamp = skb->tstamp;
1069 	memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
1070 	__skb->wire_len = cb->pkt_len;
1071 	__skb->gso_segs = skb_shinfo(skb)->gso_segs;
1072 	__skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp;
1073 }
1074 
1075 static struct proto bpf_dummy_proto = {
1076 	.name   = "bpf_dummy",
1077 	.owner  = THIS_MODULE,
1078 	.obj_size = sizeof(struct sock),
1079 };
1080 
1081 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
1082 			  union bpf_attr __user *uattr)
1083 {
1084 	bool is_l2 = false, is_direct_pkt_access = false;
1085 	struct net *net = current->nsproxy->net_ns;
1086 	struct net_device *dev = net->loopback_dev;
1087 	u32 size = kattr->test.data_size_in;
1088 	u32 repeat = kattr->test.repeat;
1089 	struct __sk_buff *ctx = NULL;
1090 	u32 retval, duration;
1091 	int hh_len = ETH_HLEN;
1092 	struct sk_buff *skb;
1093 	struct sock *sk;
1094 	void *data;
1095 	int ret;
1096 
1097 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1098 		return -EINVAL;
1099 
1100 	data = bpf_test_init(kattr, kattr->test.data_size_in,
1101 			     size, NET_SKB_PAD + NET_IP_ALIGN,
1102 			     SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1103 	if (IS_ERR(data))
1104 		return PTR_ERR(data);
1105 
1106 	ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
1107 	if (IS_ERR(ctx)) {
1108 		kfree(data);
1109 		return PTR_ERR(ctx);
1110 	}
1111 
1112 	switch (prog->type) {
1113 	case BPF_PROG_TYPE_SCHED_CLS:
1114 	case BPF_PROG_TYPE_SCHED_ACT:
1115 		is_l2 = true;
1116 		fallthrough;
1117 	case BPF_PROG_TYPE_LWT_IN:
1118 	case BPF_PROG_TYPE_LWT_OUT:
1119 	case BPF_PROG_TYPE_LWT_XMIT:
1120 		is_direct_pkt_access = true;
1121 		break;
1122 	default:
1123 		break;
1124 	}
1125 
1126 	sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
1127 	if (!sk) {
1128 		kfree(data);
1129 		kfree(ctx);
1130 		return -ENOMEM;
1131 	}
1132 	sock_init_data(NULL, sk);
1133 
1134 	skb = build_skb(data, 0);
1135 	if (!skb) {
1136 		kfree(data);
1137 		kfree(ctx);
1138 		sk_free(sk);
1139 		return -ENOMEM;
1140 	}
1141 	skb->sk = sk;
1142 
1143 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1144 	__skb_put(skb, size);
1145 	if (ctx && ctx->ifindex > 1) {
1146 		dev = dev_get_by_index(net, ctx->ifindex);
1147 		if (!dev) {
1148 			ret = -ENODEV;
1149 			goto out;
1150 		}
1151 	}
1152 	skb->protocol = eth_type_trans(skb, dev);
1153 	skb_reset_network_header(skb);
1154 
1155 	switch (skb->protocol) {
1156 	case htons(ETH_P_IP):
1157 		sk->sk_family = AF_INET;
1158 		if (sizeof(struct iphdr) <= skb_headlen(skb)) {
1159 			sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
1160 			sk->sk_daddr = ip_hdr(skb)->daddr;
1161 		}
1162 		break;
1163 #if IS_ENABLED(CONFIG_IPV6)
1164 	case htons(ETH_P_IPV6):
1165 		sk->sk_family = AF_INET6;
1166 		if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
1167 			sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
1168 			sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
1169 		}
1170 		break;
1171 #endif
1172 	default:
1173 		break;
1174 	}
1175 
1176 	if (is_l2)
1177 		__skb_push(skb, hh_len);
1178 	if (is_direct_pkt_access)
1179 		bpf_compute_data_pointers(skb);
1180 	ret = convert___skb_to_skb(skb, ctx);
1181 	if (ret)
1182 		goto out;
1183 	ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
1184 	if (ret)
1185 		goto out;
1186 	if (!is_l2) {
1187 		if (skb_headroom(skb) < hh_len) {
1188 			int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
1189 
1190 			if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
1191 				ret = -ENOMEM;
1192 				goto out;
1193 			}
1194 		}
1195 		memset(__skb_push(skb, hh_len), 0, hh_len);
1196 	}
1197 	convert_skb_to___skb(skb, ctx);
1198 
1199 	size = skb->len;
1200 	/* bpf program can never convert linear skb to non-linear */
1201 	if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
1202 		size = skb_headlen(skb);
1203 	ret = bpf_test_finish(kattr, uattr, skb->data, NULL, size, retval,
1204 			      duration);
1205 	if (!ret)
1206 		ret = bpf_ctx_finish(kattr, uattr, ctx,
1207 				     sizeof(struct __sk_buff));
1208 out:
1209 	if (dev && dev != net->loopback_dev)
1210 		dev_put(dev);
1211 	kfree_skb(skb);
1212 	sk_free(sk);
1213 	kfree(ctx);
1214 	return ret;
1215 }
1216 
1217 static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp)
1218 {
1219 	unsigned int ingress_ifindex, rx_queue_index;
1220 	struct netdev_rx_queue *rxqueue;
1221 	struct net_device *device;
1222 
1223 	if (!xdp_md)
1224 		return 0;
1225 
1226 	if (xdp_md->egress_ifindex != 0)
1227 		return -EINVAL;
1228 
1229 	ingress_ifindex = xdp_md->ingress_ifindex;
1230 	rx_queue_index = xdp_md->rx_queue_index;
1231 
1232 	if (!ingress_ifindex && rx_queue_index)
1233 		return -EINVAL;
1234 
1235 	if (ingress_ifindex) {
1236 		device = dev_get_by_index(current->nsproxy->net_ns,
1237 					  ingress_ifindex);
1238 		if (!device)
1239 			return -ENODEV;
1240 
1241 		if (rx_queue_index >= device->real_num_rx_queues)
1242 			goto free_dev;
1243 
1244 		rxqueue = __netif_get_rx_queue(device, rx_queue_index);
1245 
1246 		if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq))
1247 			goto free_dev;
1248 
1249 		xdp->rxq = &rxqueue->xdp_rxq;
1250 		/* The device is now tracked in the xdp->rxq for later
1251 		 * dev_put()
1252 		 */
1253 	}
1254 
1255 	xdp->data = xdp->data_meta + xdp_md->data;
1256 	return 0;
1257 
1258 free_dev:
1259 	dev_put(device);
1260 	return -EINVAL;
1261 }
1262 
1263 static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md)
1264 {
1265 	if (!xdp_md)
1266 		return;
1267 
1268 	xdp_md->data = xdp->data - xdp->data_meta;
1269 	xdp_md->data_end = xdp->data_end - xdp->data_meta;
1270 
1271 	if (xdp_md->ingress_ifindex)
1272 		dev_put(xdp->rxq->dev);
1273 }
1274 
1275 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
1276 			  union bpf_attr __user *uattr)
1277 {
1278 	bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES);
1279 	u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1280 	u32 batch_size = kattr->test.batch_size;
1281 	u32 retval = 0, duration, max_data_sz;
1282 	u32 size = kattr->test.data_size_in;
1283 	u32 headroom = XDP_PACKET_HEADROOM;
1284 	u32 repeat = kattr->test.repeat;
1285 	struct netdev_rx_queue *rxqueue;
1286 	struct skb_shared_info *sinfo;
1287 	struct xdp_buff xdp = {};
1288 	int i, ret = -EINVAL;
1289 	struct xdp_md *ctx;
1290 	void *data;
1291 
1292 	if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
1293 	    prog->expected_attach_type == BPF_XDP_CPUMAP)
1294 		return -EINVAL;
1295 
1296 	if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES)
1297 		return -EINVAL;
1298 
1299 	if (do_live) {
1300 		if (!batch_size)
1301 			batch_size = NAPI_POLL_WEIGHT;
1302 		else if (batch_size > TEST_XDP_MAX_BATCH)
1303 			return -E2BIG;
1304 
1305 		headroom += sizeof(struct xdp_page_head);
1306 	} else if (batch_size) {
1307 		return -EINVAL;
1308 	}
1309 
1310 	ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md));
1311 	if (IS_ERR(ctx))
1312 		return PTR_ERR(ctx);
1313 
1314 	if (ctx) {
1315 		/* There can't be user provided data before the meta data */
1316 		if (ctx->data_meta || ctx->data_end != size ||
1317 		    ctx->data > ctx->data_end ||
1318 		    unlikely(xdp_metalen_invalid(ctx->data)) ||
1319 		    (do_live && (kattr->test.data_out || kattr->test.ctx_out)))
1320 			goto free_ctx;
1321 		/* Meta data is allocated from the headroom */
1322 		headroom -= ctx->data;
1323 	}
1324 
1325 	max_data_sz = 4096 - headroom - tailroom;
1326 	if (size > max_data_sz) {
1327 		/* disallow live data mode for jumbo frames */
1328 		if (do_live)
1329 			goto free_ctx;
1330 		size = max_data_sz;
1331 	}
1332 
1333 	data = bpf_test_init(kattr, size, max_data_sz, headroom, tailroom);
1334 	if (IS_ERR(data)) {
1335 		ret = PTR_ERR(data);
1336 		goto free_ctx;
1337 	}
1338 
1339 	rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
1340 	rxqueue->xdp_rxq.frag_size = headroom + max_data_sz + tailroom;
1341 	xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq);
1342 	xdp_prepare_buff(&xdp, data, headroom, size, true);
1343 	sinfo = xdp_get_shared_info_from_buff(&xdp);
1344 
1345 	ret = xdp_convert_md_to_buff(ctx, &xdp);
1346 	if (ret)
1347 		goto free_data;
1348 
1349 	if (unlikely(kattr->test.data_size_in > size)) {
1350 		void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
1351 
1352 		while (size < kattr->test.data_size_in) {
1353 			struct page *page;
1354 			skb_frag_t *frag;
1355 			u32 data_len;
1356 
1357 			if (sinfo->nr_frags == MAX_SKB_FRAGS) {
1358 				ret = -ENOMEM;
1359 				goto out;
1360 			}
1361 
1362 			page = alloc_page(GFP_KERNEL);
1363 			if (!page) {
1364 				ret = -ENOMEM;
1365 				goto out;
1366 			}
1367 
1368 			frag = &sinfo->frags[sinfo->nr_frags++];
1369 			__skb_frag_set_page(frag, page);
1370 
1371 			data_len = min_t(u32, kattr->test.data_size_in - size,
1372 					 PAGE_SIZE);
1373 			skb_frag_size_set(frag, data_len);
1374 
1375 			if (copy_from_user(page_address(page), data_in + size,
1376 					   data_len)) {
1377 				ret = -EFAULT;
1378 				goto out;
1379 			}
1380 			sinfo->xdp_frags_size += data_len;
1381 			size += data_len;
1382 		}
1383 		xdp_buff_set_frags_flag(&xdp);
1384 	}
1385 
1386 	if (repeat > 1)
1387 		bpf_prog_change_xdp(NULL, prog);
1388 
1389 	if (do_live)
1390 		ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration);
1391 	else
1392 		ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
1393 	/* We convert the xdp_buff back to an xdp_md before checking the return
1394 	 * code so the reference count of any held netdevice will be decremented
1395 	 * even if the test run failed.
1396 	 */
1397 	xdp_convert_buff_to_md(&xdp, ctx);
1398 	if (ret)
1399 		goto out;
1400 
1401 	size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size;
1402 	ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size,
1403 			      retval, duration);
1404 	if (!ret)
1405 		ret = bpf_ctx_finish(kattr, uattr, ctx,
1406 				     sizeof(struct xdp_md));
1407 
1408 out:
1409 	if (repeat > 1)
1410 		bpf_prog_change_xdp(prog, NULL);
1411 free_data:
1412 	for (i = 0; i < sinfo->nr_frags; i++)
1413 		__free_page(skb_frag_page(&sinfo->frags[i]));
1414 	kfree(data);
1415 free_ctx:
1416 	kfree(ctx);
1417 	return ret;
1418 }
1419 
1420 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
1421 {
1422 	/* make sure the fields we don't use are zeroed */
1423 	if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
1424 		return -EINVAL;
1425 
1426 	/* flags is allowed */
1427 
1428 	if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
1429 			   sizeof(struct bpf_flow_keys)))
1430 		return -EINVAL;
1431 
1432 	return 0;
1433 }
1434 
1435 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1436 				     const union bpf_attr *kattr,
1437 				     union bpf_attr __user *uattr)
1438 {
1439 	struct bpf_test_timer t = { NO_PREEMPT };
1440 	u32 size = kattr->test.data_size_in;
1441 	struct bpf_flow_dissector ctx = {};
1442 	u32 repeat = kattr->test.repeat;
1443 	struct bpf_flow_keys *user_ctx;
1444 	struct bpf_flow_keys flow_keys;
1445 	const struct ethhdr *eth;
1446 	unsigned int flags = 0;
1447 	u32 retval, duration;
1448 	void *data;
1449 	int ret;
1450 
1451 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1452 		return -EINVAL;
1453 
1454 	if (size < ETH_HLEN)
1455 		return -EINVAL;
1456 
1457 	data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0);
1458 	if (IS_ERR(data))
1459 		return PTR_ERR(data);
1460 
1461 	eth = (struct ethhdr *)data;
1462 
1463 	if (!repeat)
1464 		repeat = 1;
1465 
1466 	user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
1467 	if (IS_ERR(user_ctx)) {
1468 		kfree(data);
1469 		return PTR_ERR(user_ctx);
1470 	}
1471 	if (user_ctx) {
1472 		ret = verify_user_bpf_flow_keys(user_ctx);
1473 		if (ret)
1474 			goto out;
1475 		flags = user_ctx->flags;
1476 	}
1477 
1478 	ctx.flow_keys = &flow_keys;
1479 	ctx.data = data;
1480 	ctx.data_end = (__u8 *)data + size;
1481 
1482 	bpf_test_timer_enter(&t);
1483 	do {
1484 		retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
1485 					  size, flags);
1486 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1487 	bpf_test_timer_leave(&t);
1488 
1489 	if (ret < 0)
1490 		goto out;
1491 
1492 	ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL,
1493 			      sizeof(flow_keys), retval, duration);
1494 	if (!ret)
1495 		ret = bpf_ctx_finish(kattr, uattr, user_ctx,
1496 				     sizeof(struct bpf_flow_keys));
1497 
1498 out:
1499 	kfree(user_ctx);
1500 	kfree(data);
1501 	return ret;
1502 }
1503 
1504 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
1505 				union bpf_attr __user *uattr)
1506 {
1507 	struct bpf_test_timer t = { NO_PREEMPT };
1508 	struct bpf_prog_array *progs = NULL;
1509 	struct bpf_sk_lookup_kern ctx = {};
1510 	u32 repeat = kattr->test.repeat;
1511 	struct bpf_sk_lookup *user_ctx;
1512 	u32 retval, duration;
1513 	int ret = -EINVAL;
1514 
1515 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1516 		return -EINVAL;
1517 
1518 	if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
1519 	    kattr->test.data_size_out)
1520 		return -EINVAL;
1521 
1522 	if (!repeat)
1523 		repeat = 1;
1524 
1525 	user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
1526 	if (IS_ERR(user_ctx))
1527 		return PTR_ERR(user_ctx);
1528 
1529 	if (!user_ctx)
1530 		return -EINVAL;
1531 
1532 	if (user_ctx->sk)
1533 		goto out;
1534 
1535 	if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
1536 		goto out;
1537 
1538 	if (user_ctx->local_port > U16_MAX) {
1539 		ret = -ERANGE;
1540 		goto out;
1541 	}
1542 
1543 	ctx.family = (u16)user_ctx->family;
1544 	ctx.protocol = (u16)user_ctx->protocol;
1545 	ctx.dport = (u16)user_ctx->local_port;
1546 	ctx.sport = user_ctx->remote_port;
1547 
1548 	switch (ctx.family) {
1549 	case AF_INET:
1550 		ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
1551 		ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
1552 		break;
1553 
1554 #if IS_ENABLED(CONFIG_IPV6)
1555 	case AF_INET6:
1556 		ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
1557 		ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
1558 		break;
1559 #endif
1560 
1561 	default:
1562 		ret = -EAFNOSUPPORT;
1563 		goto out;
1564 	}
1565 
1566 	progs = bpf_prog_array_alloc(1, GFP_KERNEL);
1567 	if (!progs) {
1568 		ret = -ENOMEM;
1569 		goto out;
1570 	}
1571 
1572 	progs->items[0].prog = prog;
1573 
1574 	bpf_test_timer_enter(&t);
1575 	do {
1576 		ctx.selected_sk = NULL;
1577 		retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run);
1578 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1579 	bpf_test_timer_leave(&t);
1580 
1581 	if (ret < 0)
1582 		goto out;
1583 
1584 	user_ctx->cookie = 0;
1585 	if (ctx.selected_sk) {
1586 		if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
1587 			ret = -EOPNOTSUPP;
1588 			goto out;
1589 		}
1590 
1591 		user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
1592 	}
1593 
1594 	ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1595 	if (!ret)
1596 		ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
1597 
1598 out:
1599 	bpf_prog_array_free(progs);
1600 	kfree(user_ctx);
1601 	return ret;
1602 }
1603 
1604 int bpf_prog_test_run_syscall(struct bpf_prog *prog,
1605 			      const union bpf_attr *kattr,
1606 			      union bpf_attr __user *uattr)
1607 {
1608 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
1609 	__u32 ctx_size_in = kattr->test.ctx_size_in;
1610 	void *ctx = NULL;
1611 	u32 retval;
1612 	int err = 0;
1613 
1614 	/* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
1615 	if (kattr->test.data_in || kattr->test.data_out ||
1616 	    kattr->test.ctx_out || kattr->test.duration ||
1617 	    kattr->test.repeat || kattr->test.flags ||
1618 	    kattr->test.batch_size)
1619 		return -EINVAL;
1620 
1621 	if (ctx_size_in < prog->aux->max_ctx_offset ||
1622 	    ctx_size_in > U16_MAX)
1623 		return -EINVAL;
1624 
1625 	if (ctx_size_in) {
1626 		ctx = memdup_user(ctx_in, ctx_size_in);
1627 		if (IS_ERR(ctx))
1628 			return PTR_ERR(ctx);
1629 	}
1630 
1631 	rcu_read_lock_trace();
1632 	retval = bpf_prog_run_pin_on_cpu(prog, ctx);
1633 	rcu_read_unlock_trace();
1634 
1635 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
1636 		err = -EFAULT;
1637 		goto out;
1638 	}
1639 	if (ctx_size_in)
1640 		if (copy_to_user(ctx_in, ctx, ctx_size_in))
1641 			err = -EFAULT;
1642 out:
1643 	kfree(ctx);
1644 	return err;
1645 }
1646 
1647 static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = {
1648 	.owner = THIS_MODULE,
1649 	.set   = &test_sk_check_kfunc_ids,
1650 };
1651 
1652 BTF_ID_LIST(bpf_prog_test_dtor_kfunc_ids)
1653 BTF_ID(struct, prog_test_ref_kfunc)
1654 BTF_ID(func, bpf_kfunc_call_test_release)
1655 BTF_ID(struct, prog_test_member)
1656 BTF_ID(func, bpf_kfunc_call_memb_release)
1657 
1658 static int __init bpf_prog_test_run_init(void)
1659 {
1660 	const struct btf_id_dtor_kfunc bpf_prog_test_dtor_kfunc[] = {
1661 		{
1662 		  .btf_id       = bpf_prog_test_dtor_kfunc_ids[0],
1663 		  .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[1]
1664 		},
1665 		{
1666 		  .btf_id	= bpf_prog_test_dtor_kfunc_ids[2],
1667 		  .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[3],
1668 		},
1669 	};
1670 	int ret;
1671 
1672 	ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set);
1673 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_prog_test_kfunc_set);
1674 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SYSCALL, &bpf_prog_test_kfunc_set);
1675 	return ret ?: register_btf_id_dtor_kfuncs(bpf_prog_test_dtor_kfunc,
1676 						  ARRAY_SIZE(bpf_prog_test_dtor_kfunc),
1677 						  THIS_MODULE);
1678 }
1679 late_initcall(bpf_prog_test_run_init);
1680