xref: /openbmc/linux/net/bpf/test_run.c (revision 0cb4228f)
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 
493 int noinline bpf_fentry_test2(int a, u64 b)
494 {
495 	return a + b;
496 }
497 
498 int noinline bpf_fentry_test3(char a, int b, u64 c)
499 {
500 	return a + b + c;
501 }
502 
503 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
504 {
505 	return (long)a + b + c + d;
506 }
507 
508 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
509 {
510 	return a + (long)b + c + d + e;
511 }
512 
513 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
514 {
515 	return a + (long)b + c + d + (long)e + f;
516 }
517 
518 struct bpf_fentry_test_t {
519 	struct bpf_fentry_test_t *a;
520 };
521 
522 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
523 {
524 	return (long)arg;
525 }
526 
527 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
528 {
529 	return (long)arg->a;
530 }
531 
532 int noinline bpf_modify_return_test(int a, int *b)
533 {
534 	*b += 1;
535 	return a + *b;
536 }
537 
538 u64 noinline bpf_kfunc_call_test1(struct sock *sk, u32 a, u64 b, u32 c, u64 d)
539 {
540 	return a + b + c + d;
541 }
542 
543 int noinline bpf_kfunc_call_test2(struct sock *sk, u32 a, u32 b)
544 {
545 	return a + b;
546 }
547 
548 struct sock * noinline bpf_kfunc_call_test3(struct sock *sk)
549 {
550 	return sk;
551 }
552 
553 struct prog_test_member1 {
554 	int a;
555 };
556 
557 struct prog_test_member {
558 	struct prog_test_member1 m;
559 	int c;
560 };
561 
562 struct prog_test_ref_kfunc {
563 	int a;
564 	int b;
565 	struct prog_test_member memb;
566 	struct prog_test_ref_kfunc *next;
567 	refcount_t cnt;
568 };
569 
570 static struct prog_test_ref_kfunc prog_test_struct = {
571 	.a = 42,
572 	.b = 108,
573 	.next = &prog_test_struct,
574 	.cnt = REFCOUNT_INIT(1),
575 };
576 
577 noinline struct prog_test_ref_kfunc *
578 bpf_kfunc_call_test_acquire(unsigned long *scalar_ptr)
579 {
580 	refcount_inc(&prog_test_struct.cnt);
581 	return &prog_test_struct;
582 }
583 
584 noinline struct prog_test_member *
585 bpf_kfunc_call_memb_acquire(void)
586 {
587 	WARN_ON_ONCE(1);
588 	return NULL;
589 }
590 
591 noinline void bpf_kfunc_call_test_release(struct prog_test_ref_kfunc *p)
592 {
593 	if (!p)
594 		return;
595 
596 	refcount_dec(&p->cnt);
597 }
598 
599 noinline void bpf_kfunc_call_memb_release(struct prog_test_member *p)
600 {
601 }
602 
603 noinline void bpf_kfunc_call_memb1_release(struct prog_test_member1 *p)
604 {
605 	WARN_ON_ONCE(1);
606 }
607 
608 static int *__bpf_kfunc_call_test_get_mem(struct prog_test_ref_kfunc *p, const int size)
609 {
610 	if (size > 2 * sizeof(int))
611 		return NULL;
612 
613 	return (int *)p;
614 }
615 
616 noinline int *bpf_kfunc_call_test_get_rdwr_mem(struct prog_test_ref_kfunc *p, const int rdwr_buf_size)
617 {
618 	return __bpf_kfunc_call_test_get_mem(p, rdwr_buf_size);
619 }
620 
621 noinline int *bpf_kfunc_call_test_get_rdonly_mem(struct prog_test_ref_kfunc *p, const int rdonly_buf_size)
622 {
623 	return __bpf_kfunc_call_test_get_mem(p, rdonly_buf_size);
624 }
625 
626 /* the next 2 ones can't be really used for testing expect to ensure
627  * that the verifier rejects the call.
628  * Acquire functions must return struct pointers, so these ones are
629  * failing.
630  */
631 noinline int *bpf_kfunc_call_test_acq_rdonly_mem(struct prog_test_ref_kfunc *p, const int rdonly_buf_size)
632 {
633 	return __bpf_kfunc_call_test_get_mem(p, rdonly_buf_size);
634 }
635 
636 noinline void bpf_kfunc_call_int_mem_release(int *p)
637 {
638 }
639 
640 noinline struct prog_test_ref_kfunc *
641 bpf_kfunc_call_test_kptr_get(struct prog_test_ref_kfunc **pp, int a, int b)
642 {
643 	struct prog_test_ref_kfunc *p = READ_ONCE(*pp);
644 
645 	if (!p)
646 		return NULL;
647 	refcount_inc(&p->cnt);
648 	return p;
649 }
650 
651 struct prog_test_pass1 {
652 	int x0;
653 	struct {
654 		int x1;
655 		struct {
656 			int x2;
657 			struct {
658 				int x3;
659 			};
660 		};
661 	};
662 };
663 
664 struct prog_test_pass2 {
665 	int len;
666 	short arr1[4];
667 	struct {
668 		char arr2[4];
669 		unsigned long arr3[8];
670 	} x;
671 };
672 
673 struct prog_test_fail1 {
674 	void *p;
675 	int x;
676 };
677 
678 struct prog_test_fail2 {
679 	int x8;
680 	struct prog_test_pass1 x;
681 };
682 
683 struct prog_test_fail3 {
684 	int len;
685 	char arr1[2];
686 	char arr2[];
687 };
688 
689 noinline void bpf_kfunc_call_test_pass_ctx(struct __sk_buff *skb)
690 {
691 }
692 
693 noinline void bpf_kfunc_call_test_pass1(struct prog_test_pass1 *p)
694 {
695 }
696 
697 noinline void bpf_kfunc_call_test_pass2(struct prog_test_pass2 *p)
698 {
699 }
700 
701 noinline void bpf_kfunc_call_test_fail1(struct prog_test_fail1 *p)
702 {
703 }
704 
705 noinline void bpf_kfunc_call_test_fail2(struct prog_test_fail2 *p)
706 {
707 }
708 
709 noinline void bpf_kfunc_call_test_fail3(struct prog_test_fail3 *p)
710 {
711 }
712 
713 noinline void bpf_kfunc_call_test_mem_len_pass1(void *mem, int mem__sz)
714 {
715 }
716 
717 noinline void bpf_kfunc_call_test_mem_len_fail1(void *mem, int len)
718 {
719 }
720 
721 noinline void bpf_kfunc_call_test_mem_len_fail2(u64 *mem, int len)
722 {
723 }
724 
725 noinline void bpf_kfunc_call_test_ref(struct prog_test_ref_kfunc *p)
726 {
727 }
728 
729 noinline void bpf_kfunc_call_test_destructive(void)
730 {
731 }
732 
733 __diag_pop();
734 
735 BTF_SET8_START(bpf_test_modify_return_ids)
736 BTF_ID_FLAGS(func, bpf_modify_return_test)
737 BTF_ID_FLAGS(func, bpf_fentry_test1, KF_SLEEPABLE)
738 BTF_SET8_END(bpf_test_modify_return_ids)
739 
740 static const struct btf_kfunc_id_set bpf_test_modify_return_set = {
741 	.owner = THIS_MODULE,
742 	.set   = &bpf_test_modify_return_ids,
743 };
744 
745 BTF_SET8_START(test_sk_check_kfunc_ids)
746 BTF_ID_FLAGS(func, bpf_kfunc_call_test1)
747 BTF_ID_FLAGS(func, bpf_kfunc_call_test2)
748 BTF_ID_FLAGS(func, bpf_kfunc_call_test3)
749 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acquire, KF_ACQUIRE | KF_RET_NULL)
750 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_acquire, KF_ACQUIRE | KF_RET_NULL)
751 BTF_ID_FLAGS(func, bpf_kfunc_call_test_release, KF_RELEASE)
752 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_release, KF_RELEASE)
753 BTF_ID_FLAGS(func, bpf_kfunc_call_memb1_release, KF_RELEASE)
754 BTF_ID_FLAGS(func, bpf_kfunc_call_test_get_rdwr_mem, KF_RET_NULL)
755 BTF_ID_FLAGS(func, bpf_kfunc_call_test_get_rdonly_mem, KF_RET_NULL)
756 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acq_rdonly_mem, KF_ACQUIRE | KF_RET_NULL)
757 BTF_ID_FLAGS(func, bpf_kfunc_call_int_mem_release, KF_RELEASE)
758 BTF_ID_FLAGS(func, bpf_kfunc_call_test_kptr_get, KF_ACQUIRE | KF_RET_NULL | KF_KPTR_GET)
759 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass_ctx)
760 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass1)
761 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass2)
762 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail1)
763 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail2)
764 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail3)
765 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_pass1)
766 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail1)
767 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail2)
768 BTF_ID_FLAGS(func, bpf_kfunc_call_test_ref, KF_TRUSTED_ARGS)
769 BTF_ID_FLAGS(func, bpf_kfunc_call_test_destructive, KF_DESTRUCTIVE)
770 BTF_SET8_END(test_sk_check_kfunc_ids)
771 
772 static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
773 			   u32 size, u32 headroom, u32 tailroom)
774 {
775 	void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
776 	void *data;
777 
778 	if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
779 		return ERR_PTR(-EINVAL);
780 
781 	if (user_size > size)
782 		return ERR_PTR(-EMSGSIZE);
783 
784 	size = SKB_DATA_ALIGN(size);
785 	data = kzalloc(size + headroom + tailroom, GFP_USER);
786 	if (!data)
787 		return ERR_PTR(-ENOMEM);
788 
789 	if (copy_from_user(data + headroom, data_in, user_size)) {
790 		kfree(data);
791 		return ERR_PTR(-EFAULT);
792 	}
793 
794 	return data;
795 }
796 
797 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
798 			      const union bpf_attr *kattr,
799 			      union bpf_attr __user *uattr)
800 {
801 	struct bpf_fentry_test_t arg = {};
802 	u16 side_effect = 0, ret = 0;
803 	int b = 2, err = -EFAULT;
804 	u32 retval = 0;
805 
806 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
807 		return -EINVAL;
808 
809 	switch (prog->expected_attach_type) {
810 	case BPF_TRACE_FENTRY:
811 	case BPF_TRACE_FEXIT:
812 		if (bpf_fentry_test1(1) != 2 ||
813 		    bpf_fentry_test2(2, 3) != 5 ||
814 		    bpf_fentry_test3(4, 5, 6) != 15 ||
815 		    bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
816 		    bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
817 		    bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
818 		    bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
819 		    bpf_fentry_test8(&arg) != 0)
820 			goto out;
821 		break;
822 	case BPF_MODIFY_RETURN:
823 		ret = bpf_modify_return_test(1, &b);
824 		if (b != 2)
825 			side_effect = 1;
826 		break;
827 	default:
828 		goto out;
829 	}
830 
831 	retval = ((u32)side_effect << 16) | ret;
832 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
833 		goto out;
834 
835 	err = 0;
836 out:
837 	trace_bpf_test_finish(&err);
838 	return err;
839 }
840 
841 struct bpf_raw_tp_test_run_info {
842 	struct bpf_prog *prog;
843 	void *ctx;
844 	u32 retval;
845 };
846 
847 static void
848 __bpf_prog_test_run_raw_tp(void *data)
849 {
850 	struct bpf_raw_tp_test_run_info *info = data;
851 
852 	rcu_read_lock();
853 	info->retval = bpf_prog_run(info->prog, info->ctx);
854 	rcu_read_unlock();
855 }
856 
857 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
858 			     const union bpf_attr *kattr,
859 			     union bpf_attr __user *uattr)
860 {
861 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
862 	__u32 ctx_size_in = kattr->test.ctx_size_in;
863 	struct bpf_raw_tp_test_run_info info;
864 	int cpu = kattr->test.cpu, err = 0;
865 	int current_cpu;
866 
867 	/* doesn't support data_in/out, ctx_out, duration, or repeat */
868 	if (kattr->test.data_in || kattr->test.data_out ||
869 	    kattr->test.ctx_out || kattr->test.duration ||
870 	    kattr->test.repeat || kattr->test.batch_size)
871 		return -EINVAL;
872 
873 	if (ctx_size_in < prog->aux->max_ctx_offset ||
874 	    ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
875 		return -EINVAL;
876 
877 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
878 		return -EINVAL;
879 
880 	if (ctx_size_in) {
881 		info.ctx = memdup_user(ctx_in, ctx_size_in);
882 		if (IS_ERR(info.ctx))
883 			return PTR_ERR(info.ctx);
884 	} else {
885 		info.ctx = NULL;
886 	}
887 
888 	info.prog = prog;
889 
890 	current_cpu = get_cpu();
891 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
892 	    cpu == current_cpu) {
893 		__bpf_prog_test_run_raw_tp(&info);
894 	} else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
895 		/* smp_call_function_single() also checks cpu_online()
896 		 * after csd_lock(). However, since cpu is from user
897 		 * space, let's do an extra quick check to filter out
898 		 * invalid value before smp_call_function_single().
899 		 */
900 		err = -ENXIO;
901 	} else {
902 		err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
903 					       &info, 1);
904 	}
905 	put_cpu();
906 
907 	if (!err &&
908 	    copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
909 		err = -EFAULT;
910 
911 	kfree(info.ctx);
912 	return err;
913 }
914 
915 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
916 {
917 	void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
918 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
919 	u32 size = kattr->test.ctx_size_in;
920 	void *data;
921 	int err;
922 
923 	if (!data_in && !data_out)
924 		return NULL;
925 
926 	data = kzalloc(max_size, GFP_USER);
927 	if (!data)
928 		return ERR_PTR(-ENOMEM);
929 
930 	if (data_in) {
931 		err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
932 		if (err) {
933 			kfree(data);
934 			return ERR_PTR(err);
935 		}
936 
937 		size = min_t(u32, max_size, size);
938 		if (copy_from_user(data, data_in, size)) {
939 			kfree(data);
940 			return ERR_PTR(-EFAULT);
941 		}
942 	}
943 	return data;
944 }
945 
946 static int bpf_ctx_finish(const union bpf_attr *kattr,
947 			  union bpf_attr __user *uattr, const void *data,
948 			  u32 size)
949 {
950 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
951 	int err = -EFAULT;
952 	u32 copy_size = size;
953 
954 	if (!data || !data_out)
955 		return 0;
956 
957 	if (copy_size > kattr->test.ctx_size_out) {
958 		copy_size = kattr->test.ctx_size_out;
959 		err = -ENOSPC;
960 	}
961 
962 	if (copy_to_user(data_out, data, copy_size))
963 		goto out;
964 	if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
965 		goto out;
966 	if (err != -ENOSPC)
967 		err = 0;
968 out:
969 	return err;
970 }
971 
972 /**
973  * range_is_zero - test whether buffer is initialized
974  * @buf: buffer to check
975  * @from: check from this position
976  * @to: check up until (excluding) this position
977  *
978  * This function returns true if the there is a non-zero byte
979  * in the buf in the range [from,to).
980  */
981 static inline bool range_is_zero(void *buf, size_t from, size_t to)
982 {
983 	return !memchr_inv((u8 *)buf + from, 0, to - from);
984 }
985 
986 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
987 {
988 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
989 
990 	if (!__skb)
991 		return 0;
992 
993 	/* make sure the fields we don't use are zeroed */
994 	if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
995 		return -EINVAL;
996 
997 	/* mark is allowed */
998 
999 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
1000 			   offsetof(struct __sk_buff, priority)))
1001 		return -EINVAL;
1002 
1003 	/* priority is allowed */
1004 	/* ingress_ifindex is allowed */
1005 	/* ifindex is allowed */
1006 
1007 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
1008 			   offsetof(struct __sk_buff, cb)))
1009 		return -EINVAL;
1010 
1011 	/* cb is allowed */
1012 
1013 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
1014 			   offsetof(struct __sk_buff, tstamp)))
1015 		return -EINVAL;
1016 
1017 	/* tstamp is allowed */
1018 	/* wire_len is allowed */
1019 	/* gso_segs is allowed */
1020 
1021 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
1022 			   offsetof(struct __sk_buff, gso_size)))
1023 		return -EINVAL;
1024 
1025 	/* gso_size is allowed */
1026 
1027 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
1028 			   offsetof(struct __sk_buff, hwtstamp)))
1029 		return -EINVAL;
1030 
1031 	/* hwtstamp is allowed */
1032 
1033 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp),
1034 			   sizeof(struct __sk_buff)))
1035 		return -EINVAL;
1036 
1037 	skb->mark = __skb->mark;
1038 	skb->priority = __skb->priority;
1039 	skb->skb_iif = __skb->ingress_ifindex;
1040 	skb->tstamp = __skb->tstamp;
1041 	memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
1042 
1043 	if (__skb->wire_len == 0) {
1044 		cb->pkt_len = skb->len;
1045 	} else {
1046 		if (__skb->wire_len < skb->len ||
1047 		    __skb->wire_len > GSO_LEGACY_MAX_SIZE)
1048 			return -EINVAL;
1049 		cb->pkt_len = __skb->wire_len;
1050 	}
1051 
1052 	if (__skb->gso_segs > GSO_MAX_SEGS)
1053 		return -EINVAL;
1054 	skb_shinfo(skb)->gso_segs = __skb->gso_segs;
1055 	skb_shinfo(skb)->gso_size = __skb->gso_size;
1056 	skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp;
1057 
1058 	return 0;
1059 }
1060 
1061 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
1062 {
1063 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
1064 
1065 	if (!__skb)
1066 		return;
1067 
1068 	__skb->mark = skb->mark;
1069 	__skb->priority = skb->priority;
1070 	__skb->ingress_ifindex = skb->skb_iif;
1071 	__skb->ifindex = skb->dev->ifindex;
1072 	__skb->tstamp = skb->tstamp;
1073 	memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
1074 	__skb->wire_len = cb->pkt_len;
1075 	__skb->gso_segs = skb_shinfo(skb)->gso_segs;
1076 	__skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp;
1077 }
1078 
1079 static struct proto bpf_dummy_proto = {
1080 	.name   = "bpf_dummy",
1081 	.owner  = THIS_MODULE,
1082 	.obj_size = sizeof(struct sock),
1083 };
1084 
1085 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
1086 			  union bpf_attr __user *uattr)
1087 {
1088 	bool is_l2 = false, is_direct_pkt_access = false;
1089 	struct net *net = current->nsproxy->net_ns;
1090 	struct net_device *dev = net->loopback_dev;
1091 	u32 size = kattr->test.data_size_in;
1092 	u32 repeat = kattr->test.repeat;
1093 	struct __sk_buff *ctx = NULL;
1094 	u32 retval, duration;
1095 	int hh_len = ETH_HLEN;
1096 	struct sk_buff *skb;
1097 	struct sock *sk;
1098 	void *data;
1099 	int ret;
1100 
1101 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1102 		return -EINVAL;
1103 
1104 	data = bpf_test_init(kattr, kattr->test.data_size_in,
1105 			     size, NET_SKB_PAD + NET_IP_ALIGN,
1106 			     SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1107 	if (IS_ERR(data))
1108 		return PTR_ERR(data);
1109 
1110 	ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
1111 	if (IS_ERR(ctx)) {
1112 		kfree(data);
1113 		return PTR_ERR(ctx);
1114 	}
1115 
1116 	switch (prog->type) {
1117 	case BPF_PROG_TYPE_SCHED_CLS:
1118 	case BPF_PROG_TYPE_SCHED_ACT:
1119 		is_l2 = true;
1120 		fallthrough;
1121 	case BPF_PROG_TYPE_LWT_IN:
1122 	case BPF_PROG_TYPE_LWT_OUT:
1123 	case BPF_PROG_TYPE_LWT_XMIT:
1124 		is_direct_pkt_access = true;
1125 		break;
1126 	default:
1127 		break;
1128 	}
1129 
1130 	sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
1131 	if (!sk) {
1132 		kfree(data);
1133 		kfree(ctx);
1134 		return -ENOMEM;
1135 	}
1136 	sock_init_data(NULL, sk);
1137 
1138 	skb = slab_build_skb(data);
1139 	if (!skb) {
1140 		kfree(data);
1141 		kfree(ctx);
1142 		sk_free(sk);
1143 		return -ENOMEM;
1144 	}
1145 	skb->sk = sk;
1146 
1147 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1148 	__skb_put(skb, size);
1149 	if (ctx && ctx->ifindex > 1) {
1150 		dev = dev_get_by_index(net, ctx->ifindex);
1151 		if (!dev) {
1152 			ret = -ENODEV;
1153 			goto out;
1154 		}
1155 	}
1156 	skb->protocol = eth_type_trans(skb, dev);
1157 	skb_reset_network_header(skb);
1158 
1159 	switch (skb->protocol) {
1160 	case htons(ETH_P_IP):
1161 		sk->sk_family = AF_INET;
1162 		if (sizeof(struct iphdr) <= skb_headlen(skb)) {
1163 			sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
1164 			sk->sk_daddr = ip_hdr(skb)->daddr;
1165 		}
1166 		break;
1167 #if IS_ENABLED(CONFIG_IPV6)
1168 	case htons(ETH_P_IPV6):
1169 		sk->sk_family = AF_INET6;
1170 		if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
1171 			sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
1172 			sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
1173 		}
1174 		break;
1175 #endif
1176 	default:
1177 		break;
1178 	}
1179 
1180 	if (is_l2)
1181 		__skb_push(skb, hh_len);
1182 	if (is_direct_pkt_access)
1183 		bpf_compute_data_pointers(skb);
1184 	ret = convert___skb_to_skb(skb, ctx);
1185 	if (ret)
1186 		goto out;
1187 	ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
1188 	if (ret)
1189 		goto out;
1190 	if (!is_l2) {
1191 		if (skb_headroom(skb) < hh_len) {
1192 			int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
1193 
1194 			if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
1195 				ret = -ENOMEM;
1196 				goto out;
1197 			}
1198 		}
1199 		memset(__skb_push(skb, hh_len), 0, hh_len);
1200 	}
1201 	convert_skb_to___skb(skb, ctx);
1202 
1203 	size = skb->len;
1204 	/* bpf program can never convert linear skb to non-linear */
1205 	if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
1206 		size = skb_headlen(skb);
1207 	ret = bpf_test_finish(kattr, uattr, skb->data, NULL, size, retval,
1208 			      duration);
1209 	if (!ret)
1210 		ret = bpf_ctx_finish(kattr, uattr, ctx,
1211 				     sizeof(struct __sk_buff));
1212 out:
1213 	if (dev && dev != net->loopback_dev)
1214 		dev_put(dev);
1215 	kfree_skb(skb);
1216 	sk_free(sk);
1217 	kfree(ctx);
1218 	return ret;
1219 }
1220 
1221 static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp)
1222 {
1223 	unsigned int ingress_ifindex, rx_queue_index;
1224 	struct netdev_rx_queue *rxqueue;
1225 	struct net_device *device;
1226 
1227 	if (!xdp_md)
1228 		return 0;
1229 
1230 	if (xdp_md->egress_ifindex != 0)
1231 		return -EINVAL;
1232 
1233 	ingress_ifindex = xdp_md->ingress_ifindex;
1234 	rx_queue_index = xdp_md->rx_queue_index;
1235 
1236 	if (!ingress_ifindex && rx_queue_index)
1237 		return -EINVAL;
1238 
1239 	if (ingress_ifindex) {
1240 		device = dev_get_by_index(current->nsproxy->net_ns,
1241 					  ingress_ifindex);
1242 		if (!device)
1243 			return -ENODEV;
1244 
1245 		if (rx_queue_index >= device->real_num_rx_queues)
1246 			goto free_dev;
1247 
1248 		rxqueue = __netif_get_rx_queue(device, rx_queue_index);
1249 
1250 		if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq))
1251 			goto free_dev;
1252 
1253 		xdp->rxq = &rxqueue->xdp_rxq;
1254 		/* The device is now tracked in the xdp->rxq for later
1255 		 * dev_put()
1256 		 */
1257 	}
1258 
1259 	xdp->data = xdp->data_meta + xdp_md->data;
1260 	return 0;
1261 
1262 free_dev:
1263 	dev_put(device);
1264 	return -EINVAL;
1265 }
1266 
1267 static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md)
1268 {
1269 	if (!xdp_md)
1270 		return;
1271 
1272 	xdp_md->data = xdp->data - xdp->data_meta;
1273 	xdp_md->data_end = xdp->data_end - xdp->data_meta;
1274 
1275 	if (xdp_md->ingress_ifindex)
1276 		dev_put(xdp->rxq->dev);
1277 }
1278 
1279 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
1280 			  union bpf_attr __user *uattr)
1281 {
1282 	bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES);
1283 	u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1284 	u32 batch_size = kattr->test.batch_size;
1285 	u32 retval = 0, duration, max_data_sz;
1286 	u32 size = kattr->test.data_size_in;
1287 	u32 headroom = XDP_PACKET_HEADROOM;
1288 	u32 repeat = kattr->test.repeat;
1289 	struct netdev_rx_queue *rxqueue;
1290 	struct skb_shared_info *sinfo;
1291 	struct xdp_buff xdp = {};
1292 	int i, ret = -EINVAL;
1293 	struct xdp_md *ctx;
1294 	void *data;
1295 
1296 	if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
1297 	    prog->expected_attach_type == BPF_XDP_CPUMAP)
1298 		return -EINVAL;
1299 
1300 	if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES)
1301 		return -EINVAL;
1302 
1303 	if (do_live) {
1304 		if (!batch_size)
1305 			batch_size = NAPI_POLL_WEIGHT;
1306 		else if (batch_size > TEST_XDP_MAX_BATCH)
1307 			return -E2BIG;
1308 
1309 		headroom += sizeof(struct xdp_page_head);
1310 	} else if (batch_size) {
1311 		return -EINVAL;
1312 	}
1313 
1314 	ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md));
1315 	if (IS_ERR(ctx))
1316 		return PTR_ERR(ctx);
1317 
1318 	if (ctx) {
1319 		/* There can't be user provided data before the meta data */
1320 		if (ctx->data_meta || ctx->data_end != size ||
1321 		    ctx->data > ctx->data_end ||
1322 		    unlikely(xdp_metalen_invalid(ctx->data)) ||
1323 		    (do_live && (kattr->test.data_out || kattr->test.ctx_out)))
1324 			goto free_ctx;
1325 		/* Meta data is allocated from the headroom */
1326 		headroom -= ctx->data;
1327 	}
1328 
1329 	max_data_sz = 4096 - headroom - tailroom;
1330 	if (size > max_data_sz) {
1331 		/* disallow live data mode for jumbo frames */
1332 		if (do_live)
1333 			goto free_ctx;
1334 		size = max_data_sz;
1335 	}
1336 
1337 	data = bpf_test_init(kattr, size, max_data_sz, headroom, tailroom);
1338 	if (IS_ERR(data)) {
1339 		ret = PTR_ERR(data);
1340 		goto free_ctx;
1341 	}
1342 
1343 	rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
1344 	rxqueue->xdp_rxq.frag_size = headroom + max_data_sz + tailroom;
1345 	xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq);
1346 	xdp_prepare_buff(&xdp, data, headroom, size, true);
1347 	sinfo = xdp_get_shared_info_from_buff(&xdp);
1348 
1349 	ret = xdp_convert_md_to_buff(ctx, &xdp);
1350 	if (ret)
1351 		goto free_data;
1352 
1353 	if (unlikely(kattr->test.data_size_in > size)) {
1354 		void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
1355 
1356 		while (size < kattr->test.data_size_in) {
1357 			struct page *page;
1358 			skb_frag_t *frag;
1359 			u32 data_len;
1360 
1361 			if (sinfo->nr_frags == MAX_SKB_FRAGS) {
1362 				ret = -ENOMEM;
1363 				goto out;
1364 			}
1365 
1366 			page = alloc_page(GFP_KERNEL);
1367 			if (!page) {
1368 				ret = -ENOMEM;
1369 				goto out;
1370 			}
1371 
1372 			frag = &sinfo->frags[sinfo->nr_frags++];
1373 			__skb_frag_set_page(frag, page);
1374 
1375 			data_len = min_t(u32, kattr->test.data_size_in - size,
1376 					 PAGE_SIZE);
1377 			skb_frag_size_set(frag, data_len);
1378 
1379 			if (copy_from_user(page_address(page), data_in + size,
1380 					   data_len)) {
1381 				ret = -EFAULT;
1382 				goto out;
1383 			}
1384 			sinfo->xdp_frags_size += data_len;
1385 			size += data_len;
1386 		}
1387 		xdp_buff_set_frags_flag(&xdp);
1388 	}
1389 
1390 	if (repeat > 1)
1391 		bpf_prog_change_xdp(NULL, prog);
1392 
1393 	if (do_live)
1394 		ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration);
1395 	else
1396 		ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
1397 	/* We convert the xdp_buff back to an xdp_md before checking the return
1398 	 * code so the reference count of any held netdevice will be decremented
1399 	 * even if the test run failed.
1400 	 */
1401 	xdp_convert_buff_to_md(&xdp, ctx);
1402 	if (ret)
1403 		goto out;
1404 
1405 	size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size;
1406 	ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size,
1407 			      retval, duration);
1408 	if (!ret)
1409 		ret = bpf_ctx_finish(kattr, uattr, ctx,
1410 				     sizeof(struct xdp_md));
1411 
1412 out:
1413 	if (repeat > 1)
1414 		bpf_prog_change_xdp(prog, NULL);
1415 free_data:
1416 	for (i = 0; i < sinfo->nr_frags; i++)
1417 		__free_page(skb_frag_page(&sinfo->frags[i]));
1418 	kfree(data);
1419 free_ctx:
1420 	kfree(ctx);
1421 	return ret;
1422 }
1423 
1424 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
1425 {
1426 	/* make sure the fields we don't use are zeroed */
1427 	if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
1428 		return -EINVAL;
1429 
1430 	/* flags is allowed */
1431 
1432 	if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
1433 			   sizeof(struct bpf_flow_keys)))
1434 		return -EINVAL;
1435 
1436 	return 0;
1437 }
1438 
1439 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1440 				     const union bpf_attr *kattr,
1441 				     union bpf_attr __user *uattr)
1442 {
1443 	struct bpf_test_timer t = { NO_PREEMPT };
1444 	u32 size = kattr->test.data_size_in;
1445 	struct bpf_flow_dissector ctx = {};
1446 	u32 repeat = kattr->test.repeat;
1447 	struct bpf_flow_keys *user_ctx;
1448 	struct bpf_flow_keys flow_keys;
1449 	const struct ethhdr *eth;
1450 	unsigned int flags = 0;
1451 	u32 retval, duration;
1452 	void *data;
1453 	int ret;
1454 
1455 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1456 		return -EINVAL;
1457 
1458 	if (size < ETH_HLEN)
1459 		return -EINVAL;
1460 
1461 	data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0);
1462 	if (IS_ERR(data))
1463 		return PTR_ERR(data);
1464 
1465 	eth = (struct ethhdr *)data;
1466 
1467 	if (!repeat)
1468 		repeat = 1;
1469 
1470 	user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
1471 	if (IS_ERR(user_ctx)) {
1472 		kfree(data);
1473 		return PTR_ERR(user_ctx);
1474 	}
1475 	if (user_ctx) {
1476 		ret = verify_user_bpf_flow_keys(user_ctx);
1477 		if (ret)
1478 			goto out;
1479 		flags = user_ctx->flags;
1480 	}
1481 
1482 	ctx.flow_keys = &flow_keys;
1483 	ctx.data = data;
1484 	ctx.data_end = (__u8 *)data + size;
1485 
1486 	bpf_test_timer_enter(&t);
1487 	do {
1488 		retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
1489 					  size, flags);
1490 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1491 	bpf_test_timer_leave(&t);
1492 
1493 	if (ret < 0)
1494 		goto out;
1495 
1496 	ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL,
1497 			      sizeof(flow_keys), retval, duration);
1498 	if (!ret)
1499 		ret = bpf_ctx_finish(kattr, uattr, user_ctx,
1500 				     sizeof(struct bpf_flow_keys));
1501 
1502 out:
1503 	kfree(user_ctx);
1504 	kfree(data);
1505 	return ret;
1506 }
1507 
1508 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
1509 				union bpf_attr __user *uattr)
1510 {
1511 	struct bpf_test_timer t = { NO_PREEMPT };
1512 	struct bpf_prog_array *progs = NULL;
1513 	struct bpf_sk_lookup_kern ctx = {};
1514 	u32 repeat = kattr->test.repeat;
1515 	struct bpf_sk_lookup *user_ctx;
1516 	u32 retval, duration;
1517 	int ret = -EINVAL;
1518 
1519 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1520 		return -EINVAL;
1521 
1522 	if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
1523 	    kattr->test.data_size_out)
1524 		return -EINVAL;
1525 
1526 	if (!repeat)
1527 		repeat = 1;
1528 
1529 	user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
1530 	if (IS_ERR(user_ctx))
1531 		return PTR_ERR(user_ctx);
1532 
1533 	if (!user_ctx)
1534 		return -EINVAL;
1535 
1536 	if (user_ctx->sk)
1537 		goto out;
1538 
1539 	if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
1540 		goto out;
1541 
1542 	if (user_ctx->local_port > U16_MAX) {
1543 		ret = -ERANGE;
1544 		goto out;
1545 	}
1546 
1547 	ctx.family = (u16)user_ctx->family;
1548 	ctx.protocol = (u16)user_ctx->protocol;
1549 	ctx.dport = (u16)user_ctx->local_port;
1550 	ctx.sport = user_ctx->remote_port;
1551 
1552 	switch (ctx.family) {
1553 	case AF_INET:
1554 		ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
1555 		ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
1556 		break;
1557 
1558 #if IS_ENABLED(CONFIG_IPV6)
1559 	case AF_INET6:
1560 		ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
1561 		ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
1562 		break;
1563 #endif
1564 
1565 	default:
1566 		ret = -EAFNOSUPPORT;
1567 		goto out;
1568 	}
1569 
1570 	progs = bpf_prog_array_alloc(1, GFP_KERNEL);
1571 	if (!progs) {
1572 		ret = -ENOMEM;
1573 		goto out;
1574 	}
1575 
1576 	progs->items[0].prog = prog;
1577 
1578 	bpf_test_timer_enter(&t);
1579 	do {
1580 		ctx.selected_sk = NULL;
1581 		retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run);
1582 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1583 	bpf_test_timer_leave(&t);
1584 
1585 	if (ret < 0)
1586 		goto out;
1587 
1588 	user_ctx->cookie = 0;
1589 	if (ctx.selected_sk) {
1590 		if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
1591 			ret = -EOPNOTSUPP;
1592 			goto out;
1593 		}
1594 
1595 		user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
1596 	}
1597 
1598 	ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1599 	if (!ret)
1600 		ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
1601 
1602 out:
1603 	bpf_prog_array_free(progs);
1604 	kfree(user_ctx);
1605 	return ret;
1606 }
1607 
1608 int bpf_prog_test_run_syscall(struct bpf_prog *prog,
1609 			      const union bpf_attr *kattr,
1610 			      union bpf_attr __user *uattr)
1611 {
1612 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
1613 	__u32 ctx_size_in = kattr->test.ctx_size_in;
1614 	void *ctx = NULL;
1615 	u32 retval;
1616 	int err = 0;
1617 
1618 	/* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
1619 	if (kattr->test.data_in || kattr->test.data_out ||
1620 	    kattr->test.ctx_out || kattr->test.duration ||
1621 	    kattr->test.repeat || kattr->test.flags ||
1622 	    kattr->test.batch_size)
1623 		return -EINVAL;
1624 
1625 	if (ctx_size_in < prog->aux->max_ctx_offset ||
1626 	    ctx_size_in > U16_MAX)
1627 		return -EINVAL;
1628 
1629 	if (ctx_size_in) {
1630 		ctx = memdup_user(ctx_in, ctx_size_in);
1631 		if (IS_ERR(ctx))
1632 			return PTR_ERR(ctx);
1633 	}
1634 
1635 	rcu_read_lock_trace();
1636 	retval = bpf_prog_run_pin_on_cpu(prog, ctx);
1637 	rcu_read_unlock_trace();
1638 
1639 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
1640 		err = -EFAULT;
1641 		goto out;
1642 	}
1643 	if (ctx_size_in)
1644 		if (copy_to_user(ctx_in, ctx, ctx_size_in))
1645 			err = -EFAULT;
1646 out:
1647 	kfree(ctx);
1648 	return err;
1649 }
1650 
1651 static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = {
1652 	.owner = THIS_MODULE,
1653 	.set   = &test_sk_check_kfunc_ids,
1654 };
1655 
1656 BTF_ID_LIST(bpf_prog_test_dtor_kfunc_ids)
1657 BTF_ID(struct, prog_test_ref_kfunc)
1658 BTF_ID(func, bpf_kfunc_call_test_release)
1659 BTF_ID(struct, prog_test_member)
1660 BTF_ID(func, bpf_kfunc_call_memb_release)
1661 
1662 static int __init bpf_prog_test_run_init(void)
1663 {
1664 	const struct btf_id_dtor_kfunc bpf_prog_test_dtor_kfunc[] = {
1665 		{
1666 		  .btf_id       = bpf_prog_test_dtor_kfunc_ids[0],
1667 		  .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[1]
1668 		},
1669 		{
1670 		  .btf_id	= bpf_prog_test_dtor_kfunc_ids[2],
1671 		  .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[3],
1672 		},
1673 	};
1674 	int ret;
1675 
1676 	ret = register_btf_fmodret_id_set(&bpf_test_modify_return_set);
1677 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set);
1678 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_prog_test_kfunc_set);
1679 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SYSCALL, &bpf_prog_test_kfunc_set);
1680 	return ret ?: register_btf_id_dtor_kfuncs(bpf_prog_test_dtor_kfunc,
1681 						  ARRAY_SIZE(bpf_prog_test_dtor_kfunc),
1682 						  THIS_MODULE);
1683 }
1684 late_initcall(bpf_prog_test_run_init);
1685