xref: /linux/net/bpf/test_run.c (revision 05a06be722896e51f65dbbb6a3610f85a8353d6b)
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 frame_was_changed(const struct xdp_page_head *head)
212 {
213 	/* xdp_scrub_frame() zeroes the data pointer, flags is the last field,
214 	 * i.e. has the highest chances to be overwritten. If those two are
215 	 * untouched, it's most likely safe to skip the context reset.
216 	 */
217 	return head->frm.data != head->orig_ctx.data ||
218 	       head->frm.flags != head->orig_ctx.flags;
219 }
220 
221 static bool ctx_was_changed(struct xdp_page_head *head)
222 {
223 	return head->orig_ctx.data != head->ctx.data ||
224 		head->orig_ctx.data_meta != head->ctx.data_meta ||
225 		head->orig_ctx.data_end != head->ctx.data_end;
226 }
227 
228 static void reset_ctx(struct xdp_page_head *head)
229 {
230 	if (likely(!frame_was_changed(head) && !ctx_was_changed(head)))
231 		return;
232 
233 	head->ctx.data = head->orig_ctx.data;
234 	head->ctx.data_meta = head->orig_ctx.data_meta;
235 	head->ctx.data_end = head->orig_ctx.data_end;
236 	xdp_update_frame_from_buff(&head->ctx, &head->frm);
237 }
238 
239 static int xdp_recv_frames(struct xdp_frame **frames, int nframes,
240 			   struct sk_buff **skbs,
241 			   struct net_device *dev)
242 {
243 	gfp_t gfp = __GFP_ZERO | GFP_ATOMIC;
244 	int i, n;
245 	LIST_HEAD(list);
246 
247 	n = kmem_cache_alloc_bulk(skbuff_cache, gfp, nframes, (void **)skbs);
248 	if (unlikely(n == 0)) {
249 		for (i = 0; i < nframes; i++)
250 			xdp_return_frame(frames[i]);
251 		return -ENOMEM;
252 	}
253 
254 	for (i = 0; i < nframes; i++) {
255 		struct xdp_frame *xdpf = frames[i];
256 		struct sk_buff *skb = skbs[i];
257 
258 		skb = __xdp_build_skb_from_frame(xdpf, skb, dev);
259 		if (!skb) {
260 			xdp_return_frame(xdpf);
261 			continue;
262 		}
263 
264 		list_add_tail(&skb->list, &list);
265 	}
266 	netif_receive_skb_list(&list);
267 
268 	return 0;
269 }
270 
271 static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
272 			      u32 repeat)
273 {
274 	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
275 	int err = 0, act, ret, i, nframes = 0, batch_sz;
276 	struct xdp_frame **frames = xdp->frames;
277 	struct xdp_page_head *head;
278 	struct xdp_frame *frm;
279 	bool redirect = false;
280 	struct xdp_buff *ctx;
281 	struct page *page;
282 
283 	batch_sz = min_t(u32, repeat, xdp->batch_size);
284 
285 	local_bh_disable();
286 	xdp_set_return_frame_no_direct();
287 
288 	for (i = 0; i < batch_sz; i++) {
289 		page = page_pool_dev_alloc_pages(xdp->pp);
290 		if (!page) {
291 			err = -ENOMEM;
292 			goto out;
293 		}
294 
295 		head = phys_to_virt(page_to_phys(page));
296 		reset_ctx(head);
297 		ctx = &head->ctx;
298 		frm = &head->frm;
299 		xdp->frame_cnt++;
300 
301 		act = bpf_prog_run_xdp(prog, ctx);
302 
303 		/* if program changed pkt bounds we need to update the xdp_frame */
304 		if (unlikely(ctx_was_changed(head))) {
305 			ret = xdp_update_frame_from_buff(ctx, frm);
306 			if (ret) {
307 				xdp_return_buff(ctx);
308 				continue;
309 			}
310 		}
311 
312 		switch (act) {
313 		case XDP_TX:
314 			/* we can't do a real XDP_TX since we're not in the
315 			 * driver, so turn it into a REDIRECT back to the same
316 			 * index
317 			 */
318 			ri->tgt_index = xdp->dev->ifindex;
319 			ri->map_id = INT_MAX;
320 			ri->map_type = BPF_MAP_TYPE_UNSPEC;
321 			fallthrough;
322 		case XDP_REDIRECT:
323 			redirect = true;
324 			ret = xdp_do_redirect_frame(xdp->dev, ctx, frm, prog);
325 			if (ret)
326 				xdp_return_buff(ctx);
327 			break;
328 		case XDP_PASS:
329 			frames[nframes++] = frm;
330 			break;
331 		default:
332 			bpf_warn_invalid_xdp_action(NULL, prog, act);
333 			fallthrough;
334 		case XDP_DROP:
335 			xdp_return_buff(ctx);
336 			break;
337 		}
338 	}
339 
340 out:
341 	if (redirect)
342 		xdp_do_flush();
343 	if (nframes) {
344 		ret = xdp_recv_frames(frames, nframes, xdp->skbs, xdp->dev);
345 		if (ret)
346 			err = ret;
347 	}
348 
349 	xdp_clear_return_frame_no_direct();
350 	local_bh_enable();
351 	return err;
352 }
353 
354 static int bpf_test_run_xdp_live(struct bpf_prog *prog, struct xdp_buff *ctx,
355 				 u32 repeat, u32 batch_size, u32 *time)
356 
357 {
358 	struct xdp_test_data xdp = { .batch_size = batch_size };
359 	struct bpf_test_timer t = { .mode = NO_MIGRATE };
360 	int ret;
361 
362 	if (!repeat)
363 		repeat = 1;
364 
365 	ret = xdp_test_run_setup(&xdp, ctx);
366 	if (ret)
367 		return ret;
368 
369 	bpf_test_timer_enter(&t);
370 	do {
371 		xdp.frame_cnt = 0;
372 		ret = xdp_test_run_batch(&xdp, prog, repeat - t.i);
373 		if (unlikely(ret < 0))
374 			break;
375 	} while (bpf_test_timer_continue(&t, xdp.frame_cnt, repeat, &ret, time));
376 	bpf_test_timer_leave(&t);
377 
378 	xdp_test_run_teardown(&xdp);
379 	return ret;
380 }
381 
382 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
383 			u32 *retval, u32 *time, bool xdp)
384 {
385 	struct bpf_prog_array_item item = {.prog = prog};
386 	struct bpf_run_ctx *old_ctx;
387 	struct bpf_cg_run_ctx run_ctx;
388 	struct bpf_test_timer t = { NO_MIGRATE };
389 	enum bpf_cgroup_storage_type stype;
390 	int ret;
391 
392 	for_each_cgroup_storage_type(stype) {
393 		item.cgroup_storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
394 		if (IS_ERR(item.cgroup_storage[stype])) {
395 			item.cgroup_storage[stype] = NULL;
396 			for_each_cgroup_storage_type(stype)
397 				bpf_cgroup_storage_free(item.cgroup_storage[stype]);
398 			return -ENOMEM;
399 		}
400 	}
401 
402 	if (!repeat)
403 		repeat = 1;
404 
405 	bpf_test_timer_enter(&t);
406 	old_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
407 	do {
408 		run_ctx.prog_item = &item;
409 		local_bh_disable();
410 		if (xdp)
411 			*retval = bpf_prog_run_xdp(prog, ctx);
412 		else
413 			*retval = bpf_prog_run(prog, ctx);
414 		local_bh_enable();
415 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, time));
416 	bpf_reset_run_ctx(old_ctx);
417 	bpf_test_timer_leave(&t);
418 
419 	for_each_cgroup_storage_type(stype)
420 		bpf_cgroup_storage_free(item.cgroup_storage[stype]);
421 
422 	return ret;
423 }
424 
425 static int bpf_test_finish(const union bpf_attr *kattr,
426 			   union bpf_attr __user *uattr, const void *data,
427 			   struct skb_shared_info *sinfo, u32 size,
428 			   u32 retval, u32 duration)
429 {
430 	void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
431 	int err = -EFAULT;
432 	u32 copy_size = size;
433 
434 	/* Clamp copy if the user has provided a size hint, but copy the full
435 	 * buffer if not to retain old behaviour.
436 	 */
437 	if (kattr->test.data_size_out &&
438 	    copy_size > kattr->test.data_size_out) {
439 		copy_size = kattr->test.data_size_out;
440 		err = -ENOSPC;
441 	}
442 
443 	if (data_out) {
444 		int len = sinfo ? copy_size - sinfo->xdp_frags_size : copy_size;
445 
446 		if (len < 0) {
447 			err = -ENOSPC;
448 			goto out;
449 		}
450 
451 		if (copy_to_user(data_out, data, len))
452 			goto out;
453 
454 		if (sinfo) {
455 			int i, offset = len;
456 			u32 data_len;
457 
458 			for (i = 0; i < sinfo->nr_frags; i++) {
459 				skb_frag_t *frag = &sinfo->frags[i];
460 
461 				if (offset >= copy_size) {
462 					err = -ENOSPC;
463 					break;
464 				}
465 
466 				data_len = min_t(u32, copy_size - offset,
467 						 skb_frag_size(frag));
468 
469 				if (copy_to_user(data_out + offset,
470 						 skb_frag_address(frag),
471 						 data_len))
472 					goto out;
473 
474 				offset += data_len;
475 			}
476 		}
477 	}
478 
479 	if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
480 		goto out;
481 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
482 		goto out;
483 	if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
484 		goto out;
485 	if (err != -ENOSPC)
486 		err = 0;
487 out:
488 	trace_bpf_test_finish(&err);
489 	return err;
490 }
491 
492 /* Integer types of various sizes and pointer combinations cover variety of
493  * architecture dependent calling conventions. 7+ can be supported in the
494  * future.
495  */
496 __diag_push();
497 __diag_ignore_all("-Wmissing-prototypes",
498 		  "Global functions as their definitions will be in vmlinux BTF");
499 __bpf_kfunc int bpf_fentry_test1(int a)
500 {
501 	return a + 1;
502 }
503 EXPORT_SYMBOL_GPL(bpf_fentry_test1);
504 
505 int noinline bpf_fentry_test2(int a, u64 b)
506 {
507 	return a + b;
508 }
509 
510 int noinline bpf_fentry_test3(char a, int b, u64 c)
511 {
512 	return a + b + c;
513 }
514 
515 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
516 {
517 	return (long)a + b + c + d;
518 }
519 
520 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
521 {
522 	return a + (long)b + c + d + e;
523 }
524 
525 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
526 {
527 	return a + (long)b + c + d + (long)e + f;
528 }
529 
530 struct bpf_fentry_test_t {
531 	struct bpf_fentry_test_t *a;
532 };
533 
534 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
535 {
536 	return (long)arg;
537 }
538 
539 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
540 {
541 	return (long)arg->a;
542 }
543 
544 __bpf_kfunc int bpf_modify_return_test(int a, int *b)
545 {
546 	*b += 1;
547 	return a + *b;
548 }
549 
550 __bpf_kfunc u64 bpf_kfunc_call_test1(struct sock *sk, u32 a, u64 b, u32 c, u64 d)
551 {
552 	return a + b + c + d;
553 }
554 
555 __bpf_kfunc int bpf_kfunc_call_test2(struct sock *sk, u32 a, u32 b)
556 {
557 	return a + b;
558 }
559 
560 __bpf_kfunc struct sock *bpf_kfunc_call_test3(struct sock *sk)
561 {
562 	return sk;
563 }
564 
565 long noinline bpf_kfunc_call_test4(signed char a, short b, int c, long d)
566 {
567 	/* Provoke the compiler to assume that the caller has sign-extended a,
568 	 * b and c on platforms where this is required (e.g. s390x).
569 	 */
570 	return (long)a + (long)b + (long)c + d;
571 }
572 
573 int noinline bpf_fentry_shadow_test(int a)
574 {
575 	return a + 1;
576 }
577 
578 struct prog_test_member1 {
579 	int a;
580 };
581 
582 struct prog_test_member {
583 	struct prog_test_member1 m;
584 	int c;
585 };
586 
587 struct prog_test_ref_kfunc {
588 	int a;
589 	int b;
590 	struct prog_test_member memb;
591 	struct prog_test_ref_kfunc *next;
592 	refcount_t cnt;
593 };
594 
595 static struct prog_test_ref_kfunc prog_test_struct = {
596 	.a = 42,
597 	.b = 108,
598 	.next = &prog_test_struct,
599 	.cnt = REFCOUNT_INIT(1),
600 };
601 
602 __bpf_kfunc struct prog_test_ref_kfunc *
603 bpf_kfunc_call_test_acquire(unsigned long *scalar_ptr)
604 {
605 	refcount_inc(&prog_test_struct.cnt);
606 	return &prog_test_struct;
607 }
608 
609 __bpf_kfunc void bpf_kfunc_call_test_offset(struct prog_test_ref_kfunc *p)
610 {
611 	WARN_ON_ONCE(1);
612 }
613 
614 __bpf_kfunc struct prog_test_member *
615 bpf_kfunc_call_memb_acquire(void)
616 {
617 	WARN_ON_ONCE(1);
618 	return NULL;
619 }
620 
621 __bpf_kfunc void bpf_kfunc_call_test_release(struct prog_test_ref_kfunc *p)
622 {
623 	refcount_dec(&p->cnt);
624 }
625 
626 __bpf_kfunc void bpf_kfunc_call_memb_release(struct prog_test_member *p)
627 {
628 }
629 
630 __bpf_kfunc void bpf_kfunc_call_memb1_release(struct prog_test_member1 *p)
631 {
632 	WARN_ON_ONCE(1);
633 }
634 
635 static int *__bpf_kfunc_call_test_get_mem(struct prog_test_ref_kfunc *p, const int size)
636 {
637 	if (size > 2 * sizeof(int))
638 		return NULL;
639 
640 	return (int *)p;
641 }
642 
643 __bpf_kfunc int *bpf_kfunc_call_test_get_rdwr_mem(struct prog_test_ref_kfunc *p,
644 						  const int rdwr_buf_size)
645 {
646 	return __bpf_kfunc_call_test_get_mem(p, rdwr_buf_size);
647 }
648 
649 __bpf_kfunc int *bpf_kfunc_call_test_get_rdonly_mem(struct prog_test_ref_kfunc *p,
650 						    const int rdonly_buf_size)
651 {
652 	return __bpf_kfunc_call_test_get_mem(p, rdonly_buf_size);
653 }
654 
655 /* the next 2 ones can't be really used for testing expect to ensure
656  * that the verifier rejects the call.
657  * Acquire functions must return struct pointers, so these ones are
658  * failing.
659  */
660 __bpf_kfunc int *bpf_kfunc_call_test_acq_rdonly_mem(struct prog_test_ref_kfunc *p,
661 						    const int rdonly_buf_size)
662 {
663 	return __bpf_kfunc_call_test_get_mem(p, rdonly_buf_size);
664 }
665 
666 __bpf_kfunc void bpf_kfunc_call_int_mem_release(int *p)
667 {
668 }
669 
670 __bpf_kfunc struct prog_test_ref_kfunc *
671 bpf_kfunc_call_test_kptr_get(struct prog_test_ref_kfunc **pp, int a, int b)
672 {
673 	struct prog_test_ref_kfunc *p = READ_ONCE(*pp);
674 
675 	if (!p)
676 		return NULL;
677 	refcount_inc(&p->cnt);
678 	return p;
679 }
680 
681 struct prog_test_pass1 {
682 	int x0;
683 	struct {
684 		int x1;
685 		struct {
686 			int x2;
687 			struct {
688 				int x3;
689 			};
690 		};
691 	};
692 };
693 
694 struct prog_test_pass2 {
695 	int len;
696 	short arr1[4];
697 	struct {
698 		char arr2[4];
699 		unsigned long arr3[8];
700 	} x;
701 };
702 
703 struct prog_test_fail1 {
704 	void *p;
705 	int x;
706 };
707 
708 struct prog_test_fail2 {
709 	int x8;
710 	struct prog_test_pass1 x;
711 };
712 
713 struct prog_test_fail3 {
714 	int len;
715 	char arr1[2];
716 	char arr2[];
717 };
718 
719 __bpf_kfunc void bpf_kfunc_call_test_pass_ctx(struct __sk_buff *skb)
720 {
721 }
722 
723 __bpf_kfunc void bpf_kfunc_call_test_pass1(struct prog_test_pass1 *p)
724 {
725 }
726 
727 __bpf_kfunc void bpf_kfunc_call_test_pass2(struct prog_test_pass2 *p)
728 {
729 }
730 
731 __bpf_kfunc void bpf_kfunc_call_test_fail1(struct prog_test_fail1 *p)
732 {
733 }
734 
735 __bpf_kfunc void bpf_kfunc_call_test_fail2(struct prog_test_fail2 *p)
736 {
737 }
738 
739 __bpf_kfunc void bpf_kfunc_call_test_fail3(struct prog_test_fail3 *p)
740 {
741 }
742 
743 __bpf_kfunc void bpf_kfunc_call_test_mem_len_pass1(void *mem, int mem__sz)
744 {
745 }
746 
747 __bpf_kfunc void bpf_kfunc_call_test_mem_len_fail1(void *mem, int len)
748 {
749 }
750 
751 __bpf_kfunc void bpf_kfunc_call_test_mem_len_fail2(u64 *mem, int len)
752 {
753 }
754 
755 __bpf_kfunc void bpf_kfunc_call_test_ref(struct prog_test_ref_kfunc *p)
756 {
757 	/* p != NULL, but p->cnt could be 0 */
758 }
759 
760 __bpf_kfunc void bpf_kfunc_call_test_destructive(void)
761 {
762 }
763 
764 __bpf_kfunc static u32 bpf_kfunc_call_test_static_unused_arg(u32 arg, u32 unused)
765 {
766 	return arg;
767 }
768 
769 __diag_pop();
770 
771 BTF_SET8_START(bpf_test_modify_return_ids)
772 BTF_ID_FLAGS(func, bpf_modify_return_test)
773 BTF_ID_FLAGS(func, bpf_fentry_test1, KF_SLEEPABLE)
774 BTF_SET8_END(bpf_test_modify_return_ids)
775 
776 static const struct btf_kfunc_id_set bpf_test_modify_return_set = {
777 	.owner = THIS_MODULE,
778 	.set   = &bpf_test_modify_return_ids,
779 };
780 
781 BTF_SET8_START(test_sk_check_kfunc_ids)
782 BTF_ID_FLAGS(func, bpf_kfunc_call_test1)
783 BTF_ID_FLAGS(func, bpf_kfunc_call_test2)
784 BTF_ID_FLAGS(func, bpf_kfunc_call_test3)
785 BTF_ID_FLAGS(func, bpf_kfunc_call_test4)
786 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acquire, KF_ACQUIRE | KF_RET_NULL)
787 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_acquire, KF_ACQUIRE | KF_RET_NULL)
788 BTF_ID_FLAGS(func, bpf_kfunc_call_test_release, KF_RELEASE)
789 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_release, KF_RELEASE)
790 BTF_ID_FLAGS(func, bpf_kfunc_call_memb1_release, KF_RELEASE)
791 BTF_ID_FLAGS(func, bpf_kfunc_call_test_get_rdwr_mem, KF_RET_NULL)
792 BTF_ID_FLAGS(func, bpf_kfunc_call_test_get_rdonly_mem, KF_RET_NULL)
793 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acq_rdonly_mem, KF_ACQUIRE | KF_RET_NULL)
794 BTF_ID_FLAGS(func, bpf_kfunc_call_int_mem_release, KF_RELEASE)
795 BTF_ID_FLAGS(func, bpf_kfunc_call_test_kptr_get, KF_ACQUIRE | KF_RET_NULL | KF_KPTR_GET)
796 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass_ctx)
797 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass1)
798 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass2)
799 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail1)
800 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail2)
801 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail3)
802 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_pass1)
803 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail1)
804 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail2)
805 BTF_ID_FLAGS(func, bpf_kfunc_call_test_ref, KF_TRUSTED_ARGS | KF_RCU)
806 BTF_ID_FLAGS(func, bpf_kfunc_call_test_destructive, KF_DESTRUCTIVE)
807 BTF_ID_FLAGS(func, bpf_kfunc_call_test_static_unused_arg)
808 BTF_ID_FLAGS(func, bpf_kfunc_call_test_offset)
809 BTF_SET8_END(test_sk_check_kfunc_ids)
810 
811 static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
812 			   u32 size, u32 headroom, u32 tailroom)
813 {
814 	void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
815 	void *data;
816 
817 	if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
818 		return ERR_PTR(-EINVAL);
819 
820 	if (user_size > size)
821 		return ERR_PTR(-EMSGSIZE);
822 
823 	size = SKB_DATA_ALIGN(size);
824 	data = kzalloc(size + headroom + tailroom, GFP_USER);
825 	if (!data)
826 		return ERR_PTR(-ENOMEM);
827 
828 	if (copy_from_user(data + headroom, data_in, user_size)) {
829 		kfree(data);
830 		return ERR_PTR(-EFAULT);
831 	}
832 
833 	return data;
834 }
835 
836 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
837 			      const union bpf_attr *kattr,
838 			      union bpf_attr __user *uattr)
839 {
840 	struct bpf_fentry_test_t arg = {};
841 	u16 side_effect = 0, ret = 0;
842 	int b = 2, err = -EFAULT;
843 	u32 retval = 0;
844 
845 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
846 		return -EINVAL;
847 
848 	switch (prog->expected_attach_type) {
849 	case BPF_TRACE_FENTRY:
850 	case BPF_TRACE_FEXIT:
851 		if (bpf_fentry_test1(1) != 2 ||
852 		    bpf_fentry_test2(2, 3) != 5 ||
853 		    bpf_fentry_test3(4, 5, 6) != 15 ||
854 		    bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
855 		    bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
856 		    bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
857 		    bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
858 		    bpf_fentry_test8(&arg) != 0)
859 			goto out;
860 		break;
861 	case BPF_MODIFY_RETURN:
862 		ret = bpf_modify_return_test(1, &b);
863 		if (b != 2)
864 			side_effect = 1;
865 		break;
866 	default:
867 		goto out;
868 	}
869 
870 	retval = ((u32)side_effect << 16) | ret;
871 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
872 		goto out;
873 
874 	err = 0;
875 out:
876 	trace_bpf_test_finish(&err);
877 	return err;
878 }
879 
880 struct bpf_raw_tp_test_run_info {
881 	struct bpf_prog *prog;
882 	void *ctx;
883 	u32 retval;
884 };
885 
886 static void
887 __bpf_prog_test_run_raw_tp(void *data)
888 {
889 	struct bpf_raw_tp_test_run_info *info = data;
890 
891 	rcu_read_lock();
892 	info->retval = bpf_prog_run(info->prog, info->ctx);
893 	rcu_read_unlock();
894 }
895 
896 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
897 			     const union bpf_attr *kattr,
898 			     union bpf_attr __user *uattr)
899 {
900 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
901 	__u32 ctx_size_in = kattr->test.ctx_size_in;
902 	struct bpf_raw_tp_test_run_info info;
903 	int cpu = kattr->test.cpu, err = 0;
904 	int current_cpu;
905 
906 	/* doesn't support data_in/out, ctx_out, duration, or repeat */
907 	if (kattr->test.data_in || kattr->test.data_out ||
908 	    kattr->test.ctx_out || kattr->test.duration ||
909 	    kattr->test.repeat || kattr->test.batch_size)
910 		return -EINVAL;
911 
912 	if (ctx_size_in < prog->aux->max_ctx_offset ||
913 	    ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
914 		return -EINVAL;
915 
916 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
917 		return -EINVAL;
918 
919 	if (ctx_size_in) {
920 		info.ctx = memdup_user(ctx_in, ctx_size_in);
921 		if (IS_ERR(info.ctx))
922 			return PTR_ERR(info.ctx);
923 	} else {
924 		info.ctx = NULL;
925 	}
926 
927 	info.prog = prog;
928 
929 	current_cpu = get_cpu();
930 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
931 	    cpu == current_cpu) {
932 		__bpf_prog_test_run_raw_tp(&info);
933 	} else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
934 		/* smp_call_function_single() also checks cpu_online()
935 		 * after csd_lock(). However, since cpu is from user
936 		 * space, let's do an extra quick check to filter out
937 		 * invalid value before smp_call_function_single().
938 		 */
939 		err = -ENXIO;
940 	} else {
941 		err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
942 					       &info, 1);
943 	}
944 	put_cpu();
945 
946 	if (!err &&
947 	    copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
948 		err = -EFAULT;
949 
950 	kfree(info.ctx);
951 	return err;
952 }
953 
954 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
955 {
956 	void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
957 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
958 	u32 size = kattr->test.ctx_size_in;
959 	void *data;
960 	int err;
961 
962 	if (!data_in && !data_out)
963 		return NULL;
964 
965 	data = kzalloc(max_size, GFP_USER);
966 	if (!data)
967 		return ERR_PTR(-ENOMEM);
968 
969 	if (data_in) {
970 		err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
971 		if (err) {
972 			kfree(data);
973 			return ERR_PTR(err);
974 		}
975 
976 		size = min_t(u32, max_size, size);
977 		if (copy_from_user(data, data_in, size)) {
978 			kfree(data);
979 			return ERR_PTR(-EFAULT);
980 		}
981 	}
982 	return data;
983 }
984 
985 static int bpf_ctx_finish(const union bpf_attr *kattr,
986 			  union bpf_attr __user *uattr, const void *data,
987 			  u32 size)
988 {
989 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
990 	int err = -EFAULT;
991 	u32 copy_size = size;
992 
993 	if (!data || !data_out)
994 		return 0;
995 
996 	if (copy_size > kattr->test.ctx_size_out) {
997 		copy_size = kattr->test.ctx_size_out;
998 		err = -ENOSPC;
999 	}
1000 
1001 	if (copy_to_user(data_out, data, copy_size))
1002 		goto out;
1003 	if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
1004 		goto out;
1005 	if (err != -ENOSPC)
1006 		err = 0;
1007 out:
1008 	return err;
1009 }
1010 
1011 /**
1012  * range_is_zero - test whether buffer is initialized
1013  * @buf: buffer to check
1014  * @from: check from this position
1015  * @to: check up until (excluding) this position
1016  *
1017  * This function returns true if the there is a non-zero byte
1018  * in the buf in the range [from,to).
1019  */
1020 static inline bool range_is_zero(void *buf, size_t from, size_t to)
1021 {
1022 	return !memchr_inv((u8 *)buf + from, 0, to - from);
1023 }
1024 
1025 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
1026 {
1027 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
1028 
1029 	if (!__skb)
1030 		return 0;
1031 
1032 	/* make sure the fields we don't use are zeroed */
1033 	if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
1034 		return -EINVAL;
1035 
1036 	/* mark is allowed */
1037 
1038 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
1039 			   offsetof(struct __sk_buff, priority)))
1040 		return -EINVAL;
1041 
1042 	/* priority is allowed */
1043 	/* ingress_ifindex is allowed */
1044 	/* ifindex is allowed */
1045 
1046 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
1047 			   offsetof(struct __sk_buff, cb)))
1048 		return -EINVAL;
1049 
1050 	/* cb is allowed */
1051 
1052 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
1053 			   offsetof(struct __sk_buff, tstamp)))
1054 		return -EINVAL;
1055 
1056 	/* tstamp is allowed */
1057 	/* wire_len is allowed */
1058 	/* gso_segs is allowed */
1059 
1060 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
1061 			   offsetof(struct __sk_buff, gso_size)))
1062 		return -EINVAL;
1063 
1064 	/* gso_size is allowed */
1065 
1066 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
1067 			   offsetof(struct __sk_buff, hwtstamp)))
1068 		return -EINVAL;
1069 
1070 	/* hwtstamp is allowed */
1071 
1072 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp),
1073 			   sizeof(struct __sk_buff)))
1074 		return -EINVAL;
1075 
1076 	skb->mark = __skb->mark;
1077 	skb->priority = __skb->priority;
1078 	skb->skb_iif = __skb->ingress_ifindex;
1079 	skb->tstamp = __skb->tstamp;
1080 	memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
1081 
1082 	if (__skb->wire_len == 0) {
1083 		cb->pkt_len = skb->len;
1084 	} else {
1085 		if (__skb->wire_len < skb->len ||
1086 		    __skb->wire_len > GSO_LEGACY_MAX_SIZE)
1087 			return -EINVAL;
1088 		cb->pkt_len = __skb->wire_len;
1089 	}
1090 
1091 	if (__skb->gso_segs > GSO_MAX_SEGS)
1092 		return -EINVAL;
1093 	skb_shinfo(skb)->gso_segs = __skb->gso_segs;
1094 	skb_shinfo(skb)->gso_size = __skb->gso_size;
1095 	skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp;
1096 
1097 	return 0;
1098 }
1099 
1100 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
1101 {
1102 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
1103 
1104 	if (!__skb)
1105 		return;
1106 
1107 	__skb->mark = skb->mark;
1108 	__skb->priority = skb->priority;
1109 	__skb->ingress_ifindex = skb->skb_iif;
1110 	__skb->ifindex = skb->dev->ifindex;
1111 	__skb->tstamp = skb->tstamp;
1112 	memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
1113 	__skb->wire_len = cb->pkt_len;
1114 	__skb->gso_segs = skb_shinfo(skb)->gso_segs;
1115 	__skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp;
1116 }
1117 
1118 static struct proto bpf_dummy_proto = {
1119 	.name   = "bpf_dummy",
1120 	.owner  = THIS_MODULE,
1121 	.obj_size = sizeof(struct sock),
1122 };
1123 
1124 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
1125 			  union bpf_attr __user *uattr)
1126 {
1127 	bool is_l2 = false, is_direct_pkt_access = false;
1128 	struct net *net = current->nsproxy->net_ns;
1129 	struct net_device *dev = net->loopback_dev;
1130 	u32 size = kattr->test.data_size_in;
1131 	u32 repeat = kattr->test.repeat;
1132 	struct __sk_buff *ctx = NULL;
1133 	u32 retval, duration;
1134 	int hh_len = ETH_HLEN;
1135 	struct sk_buff *skb;
1136 	struct sock *sk;
1137 	void *data;
1138 	int ret;
1139 
1140 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1141 		return -EINVAL;
1142 
1143 	data = bpf_test_init(kattr, kattr->test.data_size_in,
1144 			     size, NET_SKB_PAD + NET_IP_ALIGN,
1145 			     SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1146 	if (IS_ERR(data))
1147 		return PTR_ERR(data);
1148 
1149 	ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
1150 	if (IS_ERR(ctx)) {
1151 		kfree(data);
1152 		return PTR_ERR(ctx);
1153 	}
1154 
1155 	switch (prog->type) {
1156 	case BPF_PROG_TYPE_SCHED_CLS:
1157 	case BPF_PROG_TYPE_SCHED_ACT:
1158 		is_l2 = true;
1159 		fallthrough;
1160 	case BPF_PROG_TYPE_LWT_IN:
1161 	case BPF_PROG_TYPE_LWT_OUT:
1162 	case BPF_PROG_TYPE_LWT_XMIT:
1163 		is_direct_pkt_access = true;
1164 		break;
1165 	default:
1166 		break;
1167 	}
1168 
1169 	sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
1170 	if (!sk) {
1171 		kfree(data);
1172 		kfree(ctx);
1173 		return -ENOMEM;
1174 	}
1175 	sock_init_data(NULL, sk);
1176 
1177 	skb = slab_build_skb(data);
1178 	if (!skb) {
1179 		kfree(data);
1180 		kfree(ctx);
1181 		sk_free(sk);
1182 		return -ENOMEM;
1183 	}
1184 	skb->sk = sk;
1185 
1186 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1187 	__skb_put(skb, size);
1188 	if (ctx && ctx->ifindex > 1) {
1189 		dev = dev_get_by_index(net, ctx->ifindex);
1190 		if (!dev) {
1191 			ret = -ENODEV;
1192 			goto out;
1193 		}
1194 	}
1195 	skb->protocol = eth_type_trans(skb, dev);
1196 	skb_reset_network_header(skb);
1197 
1198 	switch (skb->protocol) {
1199 	case htons(ETH_P_IP):
1200 		sk->sk_family = AF_INET;
1201 		if (sizeof(struct iphdr) <= skb_headlen(skb)) {
1202 			sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
1203 			sk->sk_daddr = ip_hdr(skb)->daddr;
1204 		}
1205 		break;
1206 #if IS_ENABLED(CONFIG_IPV6)
1207 	case htons(ETH_P_IPV6):
1208 		sk->sk_family = AF_INET6;
1209 		if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
1210 			sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
1211 			sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
1212 		}
1213 		break;
1214 #endif
1215 	default:
1216 		break;
1217 	}
1218 
1219 	if (is_l2)
1220 		__skb_push(skb, hh_len);
1221 	if (is_direct_pkt_access)
1222 		bpf_compute_data_pointers(skb);
1223 	ret = convert___skb_to_skb(skb, ctx);
1224 	if (ret)
1225 		goto out;
1226 	ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
1227 	if (ret)
1228 		goto out;
1229 	if (!is_l2) {
1230 		if (skb_headroom(skb) < hh_len) {
1231 			int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
1232 
1233 			if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
1234 				ret = -ENOMEM;
1235 				goto out;
1236 			}
1237 		}
1238 		memset(__skb_push(skb, hh_len), 0, hh_len);
1239 	}
1240 	convert_skb_to___skb(skb, ctx);
1241 
1242 	size = skb->len;
1243 	/* bpf program can never convert linear skb to non-linear */
1244 	if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
1245 		size = skb_headlen(skb);
1246 	ret = bpf_test_finish(kattr, uattr, skb->data, NULL, size, retval,
1247 			      duration);
1248 	if (!ret)
1249 		ret = bpf_ctx_finish(kattr, uattr, ctx,
1250 				     sizeof(struct __sk_buff));
1251 out:
1252 	if (dev && dev != net->loopback_dev)
1253 		dev_put(dev);
1254 	kfree_skb(skb);
1255 	sk_free(sk);
1256 	kfree(ctx);
1257 	return ret;
1258 }
1259 
1260 static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp)
1261 {
1262 	unsigned int ingress_ifindex, rx_queue_index;
1263 	struct netdev_rx_queue *rxqueue;
1264 	struct net_device *device;
1265 
1266 	if (!xdp_md)
1267 		return 0;
1268 
1269 	if (xdp_md->egress_ifindex != 0)
1270 		return -EINVAL;
1271 
1272 	ingress_ifindex = xdp_md->ingress_ifindex;
1273 	rx_queue_index = xdp_md->rx_queue_index;
1274 
1275 	if (!ingress_ifindex && rx_queue_index)
1276 		return -EINVAL;
1277 
1278 	if (ingress_ifindex) {
1279 		device = dev_get_by_index(current->nsproxy->net_ns,
1280 					  ingress_ifindex);
1281 		if (!device)
1282 			return -ENODEV;
1283 
1284 		if (rx_queue_index >= device->real_num_rx_queues)
1285 			goto free_dev;
1286 
1287 		rxqueue = __netif_get_rx_queue(device, rx_queue_index);
1288 
1289 		if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq))
1290 			goto free_dev;
1291 
1292 		xdp->rxq = &rxqueue->xdp_rxq;
1293 		/* The device is now tracked in the xdp->rxq for later
1294 		 * dev_put()
1295 		 */
1296 	}
1297 
1298 	xdp->data = xdp->data_meta + xdp_md->data;
1299 	return 0;
1300 
1301 free_dev:
1302 	dev_put(device);
1303 	return -EINVAL;
1304 }
1305 
1306 static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md)
1307 {
1308 	if (!xdp_md)
1309 		return;
1310 
1311 	xdp_md->data = xdp->data - xdp->data_meta;
1312 	xdp_md->data_end = xdp->data_end - xdp->data_meta;
1313 
1314 	if (xdp_md->ingress_ifindex)
1315 		dev_put(xdp->rxq->dev);
1316 }
1317 
1318 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
1319 			  union bpf_attr __user *uattr)
1320 {
1321 	bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES);
1322 	u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1323 	u32 batch_size = kattr->test.batch_size;
1324 	u32 retval = 0, duration, max_data_sz;
1325 	u32 size = kattr->test.data_size_in;
1326 	u32 headroom = XDP_PACKET_HEADROOM;
1327 	u32 repeat = kattr->test.repeat;
1328 	struct netdev_rx_queue *rxqueue;
1329 	struct skb_shared_info *sinfo;
1330 	struct xdp_buff xdp = {};
1331 	int i, ret = -EINVAL;
1332 	struct xdp_md *ctx;
1333 	void *data;
1334 
1335 	if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
1336 	    prog->expected_attach_type == BPF_XDP_CPUMAP)
1337 		return -EINVAL;
1338 
1339 	if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES)
1340 		return -EINVAL;
1341 
1342 	if (bpf_prog_is_dev_bound(prog->aux))
1343 		return -EINVAL;
1344 
1345 	if (do_live) {
1346 		if (!batch_size)
1347 			batch_size = NAPI_POLL_WEIGHT;
1348 		else if (batch_size > TEST_XDP_MAX_BATCH)
1349 			return -E2BIG;
1350 
1351 		headroom += sizeof(struct xdp_page_head);
1352 	} else if (batch_size) {
1353 		return -EINVAL;
1354 	}
1355 
1356 	ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md));
1357 	if (IS_ERR(ctx))
1358 		return PTR_ERR(ctx);
1359 
1360 	if (ctx) {
1361 		/* There can't be user provided data before the meta data */
1362 		if (ctx->data_meta || ctx->data_end != size ||
1363 		    ctx->data > ctx->data_end ||
1364 		    unlikely(xdp_metalen_invalid(ctx->data)) ||
1365 		    (do_live && (kattr->test.data_out || kattr->test.ctx_out)))
1366 			goto free_ctx;
1367 		/* Meta data is allocated from the headroom */
1368 		headroom -= ctx->data;
1369 	}
1370 
1371 	max_data_sz = 4096 - headroom - tailroom;
1372 	if (size > max_data_sz) {
1373 		/* disallow live data mode for jumbo frames */
1374 		if (do_live)
1375 			goto free_ctx;
1376 		size = max_data_sz;
1377 	}
1378 
1379 	data = bpf_test_init(kattr, size, max_data_sz, headroom, tailroom);
1380 	if (IS_ERR(data)) {
1381 		ret = PTR_ERR(data);
1382 		goto free_ctx;
1383 	}
1384 
1385 	rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
1386 	rxqueue->xdp_rxq.frag_size = headroom + max_data_sz + tailroom;
1387 	xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq);
1388 	xdp_prepare_buff(&xdp, data, headroom, size, true);
1389 	sinfo = xdp_get_shared_info_from_buff(&xdp);
1390 
1391 	ret = xdp_convert_md_to_buff(ctx, &xdp);
1392 	if (ret)
1393 		goto free_data;
1394 
1395 	if (unlikely(kattr->test.data_size_in > size)) {
1396 		void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
1397 
1398 		while (size < kattr->test.data_size_in) {
1399 			struct page *page;
1400 			skb_frag_t *frag;
1401 			u32 data_len;
1402 
1403 			if (sinfo->nr_frags == MAX_SKB_FRAGS) {
1404 				ret = -ENOMEM;
1405 				goto out;
1406 			}
1407 
1408 			page = alloc_page(GFP_KERNEL);
1409 			if (!page) {
1410 				ret = -ENOMEM;
1411 				goto out;
1412 			}
1413 
1414 			frag = &sinfo->frags[sinfo->nr_frags++];
1415 			__skb_frag_set_page(frag, page);
1416 
1417 			data_len = min_t(u32, kattr->test.data_size_in - size,
1418 					 PAGE_SIZE);
1419 			skb_frag_size_set(frag, data_len);
1420 
1421 			if (copy_from_user(page_address(page), data_in + size,
1422 					   data_len)) {
1423 				ret = -EFAULT;
1424 				goto out;
1425 			}
1426 			sinfo->xdp_frags_size += data_len;
1427 			size += data_len;
1428 		}
1429 		xdp_buff_set_frags_flag(&xdp);
1430 	}
1431 
1432 	if (repeat > 1)
1433 		bpf_prog_change_xdp(NULL, prog);
1434 
1435 	if (do_live)
1436 		ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration);
1437 	else
1438 		ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
1439 	/* We convert the xdp_buff back to an xdp_md before checking the return
1440 	 * code so the reference count of any held netdevice will be decremented
1441 	 * even if the test run failed.
1442 	 */
1443 	xdp_convert_buff_to_md(&xdp, ctx);
1444 	if (ret)
1445 		goto out;
1446 
1447 	size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size;
1448 	ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size,
1449 			      retval, duration);
1450 	if (!ret)
1451 		ret = bpf_ctx_finish(kattr, uattr, ctx,
1452 				     sizeof(struct xdp_md));
1453 
1454 out:
1455 	if (repeat > 1)
1456 		bpf_prog_change_xdp(prog, NULL);
1457 free_data:
1458 	for (i = 0; i < sinfo->nr_frags; i++)
1459 		__free_page(skb_frag_page(&sinfo->frags[i]));
1460 	kfree(data);
1461 free_ctx:
1462 	kfree(ctx);
1463 	return ret;
1464 }
1465 
1466 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
1467 {
1468 	/* make sure the fields we don't use are zeroed */
1469 	if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
1470 		return -EINVAL;
1471 
1472 	/* flags is allowed */
1473 
1474 	if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
1475 			   sizeof(struct bpf_flow_keys)))
1476 		return -EINVAL;
1477 
1478 	return 0;
1479 }
1480 
1481 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1482 				     const union bpf_attr *kattr,
1483 				     union bpf_attr __user *uattr)
1484 {
1485 	struct bpf_test_timer t = { NO_PREEMPT };
1486 	u32 size = kattr->test.data_size_in;
1487 	struct bpf_flow_dissector ctx = {};
1488 	u32 repeat = kattr->test.repeat;
1489 	struct bpf_flow_keys *user_ctx;
1490 	struct bpf_flow_keys flow_keys;
1491 	const struct ethhdr *eth;
1492 	unsigned int flags = 0;
1493 	u32 retval, duration;
1494 	void *data;
1495 	int ret;
1496 
1497 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1498 		return -EINVAL;
1499 
1500 	if (size < ETH_HLEN)
1501 		return -EINVAL;
1502 
1503 	data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0);
1504 	if (IS_ERR(data))
1505 		return PTR_ERR(data);
1506 
1507 	eth = (struct ethhdr *)data;
1508 
1509 	if (!repeat)
1510 		repeat = 1;
1511 
1512 	user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
1513 	if (IS_ERR(user_ctx)) {
1514 		kfree(data);
1515 		return PTR_ERR(user_ctx);
1516 	}
1517 	if (user_ctx) {
1518 		ret = verify_user_bpf_flow_keys(user_ctx);
1519 		if (ret)
1520 			goto out;
1521 		flags = user_ctx->flags;
1522 	}
1523 
1524 	ctx.flow_keys = &flow_keys;
1525 	ctx.data = data;
1526 	ctx.data_end = (__u8 *)data + size;
1527 
1528 	bpf_test_timer_enter(&t);
1529 	do {
1530 		retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
1531 					  size, flags);
1532 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1533 	bpf_test_timer_leave(&t);
1534 
1535 	if (ret < 0)
1536 		goto out;
1537 
1538 	ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL,
1539 			      sizeof(flow_keys), retval, duration);
1540 	if (!ret)
1541 		ret = bpf_ctx_finish(kattr, uattr, user_ctx,
1542 				     sizeof(struct bpf_flow_keys));
1543 
1544 out:
1545 	kfree(user_ctx);
1546 	kfree(data);
1547 	return ret;
1548 }
1549 
1550 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
1551 				union bpf_attr __user *uattr)
1552 {
1553 	struct bpf_test_timer t = { NO_PREEMPT };
1554 	struct bpf_prog_array *progs = NULL;
1555 	struct bpf_sk_lookup_kern ctx = {};
1556 	u32 repeat = kattr->test.repeat;
1557 	struct bpf_sk_lookup *user_ctx;
1558 	u32 retval, duration;
1559 	int ret = -EINVAL;
1560 
1561 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1562 		return -EINVAL;
1563 
1564 	if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
1565 	    kattr->test.data_size_out)
1566 		return -EINVAL;
1567 
1568 	if (!repeat)
1569 		repeat = 1;
1570 
1571 	user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
1572 	if (IS_ERR(user_ctx))
1573 		return PTR_ERR(user_ctx);
1574 
1575 	if (!user_ctx)
1576 		return -EINVAL;
1577 
1578 	if (user_ctx->sk)
1579 		goto out;
1580 
1581 	if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
1582 		goto out;
1583 
1584 	if (user_ctx->local_port > U16_MAX) {
1585 		ret = -ERANGE;
1586 		goto out;
1587 	}
1588 
1589 	ctx.family = (u16)user_ctx->family;
1590 	ctx.protocol = (u16)user_ctx->protocol;
1591 	ctx.dport = (u16)user_ctx->local_port;
1592 	ctx.sport = user_ctx->remote_port;
1593 
1594 	switch (ctx.family) {
1595 	case AF_INET:
1596 		ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
1597 		ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
1598 		break;
1599 
1600 #if IS_ENABLED(CONFIG_IPV6)
1601 	case AF_INET6:
1602 		ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
1603 		ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
1604 		break;
1605 #endif
1606 
1607 	default:
1608 		ret = -EAFNOSUPPORT;
1609 		goto out;
1610 	}
1611 
1612 	progs = bpf_prog_array_alloc(1, GFP_KERNEL);
1613 	if (!progs) {
1614 		ret = -ENOMEM;
1615 		goto out;
1616 	}
1617 
1618 	progs->items[0].prog = prog;
1619 
1620 	bpf_test_timer_enter(&t);
1621 	do {
1622 		ctx.selected_sk = NULL;
1623 		retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run);
1624 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1625 	bpf_test_timer_leave(&t);
1626 
1627 	if (ret < 0)
1628 		goto out;
1629 
1630 	user_ctx->cookie = 0;
1631 	if (ctx.selected_sk) {
1632 		if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
1633 			ret = -EOPNOTSUPP;
1634 			goto out;
1635 		}
1636 
1637 		user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
1638 	}
1639 
1640 	ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1641 	if (!ret)
1642 		ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
1643 
1644 out:
1645 	bpf_prog_array_free(progs);
1646 	kfree(user_ctx);
1647 	return ret;
1648 }
1649 
1650 int bpf_prog_test_run_syscall(struct bpf_prog *prog,
1651 			      const union bpf_attr *kattr,
1652 			      union bpf_attr __user *uattr)
1653 {
1654 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
1655 	__u32 ctx_size_in = kattr->test.ctx_size_in;
1656 	void *ctx = NULL;
1657 	u32 retval;
1658 	int err = 0;
1659 
1660 	/* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
1661 	if (kattr->test.data_in || kattr->test.data_out ||
1662 	    kattr->test.ctx_out || kattr->test.duration ||
1663 	    kattr->test.repeat || kattr->test.flags ||
1664 	    kattr->test.batch_size)
1665 		return -EINVAL;
1666 
1667 	if (ctx_size_in < prog->aux->max_ctx_offset ||
1668 	    ctx_size_in > U16_MAX)
1669 		return -EINVAL;
1670 
1671 	if (ctx_size_in) {
1672 		ctx = memdup_user(ctx_in, ctx_size_in);
1673 		if (IS_ERR(ctx))
1674 			return PTR_ERR(ctx);
1675 	}
1676 
1677 	rcu_read_lock_trace();
1678 	retval = bpf_prog_run_pin_on_cpu(prog, ctx);
1679 	rcu_read_unlock_trace();
1680 
1681 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
1682 		err = -EFAULT;
1683 		goto out;
1684 	}
1685 	if (ctx_size_in)
1686 		if (copy_to_user(ctx_in, ctx, ctx_size_in))
1687 			err = -EFAULT;
1688 out:
1689 	kfree(ctx);
1690 	return err;
1691 }
1692 
1693 static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = {
1694 	.owner = THIS_MODULE,
1695 	.set   = &test_sk_check_kfunc_ids,
1696 };
1697 
1698 BTF_ID_LIST(bpf_prog_test_dtor_kfunc_ids)
1699 BTF_ID(struct, prog_test_ref_kfunc)
1700 BTF_ID(func, bpf_kfunc_call_test_release)
1701 BTF_ID(struct, prog_test_member)
1702 BTF_ID(func, bpf_kfunc_call_memb_release)
1703 
1704 static int __init bpf_prog_test_run_init(void)
1705 {
1706 	const struct btf_id_dtor_kfunc bpf_prog_test_dtor_kfunc[] = {
1707 		{
1708 		  .btf_id       = bpf_prog_test_dtor_kfunc_ids[0],
1709 		  .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[1]
1710 		},
1711 		{
1712 		  .btf_id	= bpf_prog_test_dtor_kfunc_ids[2],
1713 		  .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[3],
1714 		},
1715 	};
1716 	int ret;
1717 
1718 	ret = register_btf_fmodret_id_set(&bpf_test_modify_return_set);
1719 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set);
1720 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_prog_test_kfunc_set);
1721 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SYSCALL, &bpf_prog_test_kfunc_set);
1722 	return ret ?: register_btf_id_dtor_kfuncs(bpf_prog_test_dtor_kfunc,
1723 						  ARRAY_SIZE(bpf_prog_test_dtor_kfunc),
1724 						  THIS_MODULE);
1725 }
1726 late_initcall(bpf_prog_test_run_init);
1727