1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Randomized tests for eBPF longest-prefix-match maps
4  *
5  * This program runs randomized tests against the lpm-bpf-map. It implements a
6  * "Trivial Longest Prefix Match" (tlpm) based on simple, linear, singly linked
7  * lists. The implementation should be pretty straightforward.
8  *
9  * Based on tlpm, this inserts randomized data into bpf-lpm-maps and verifies
10  * the trie-based bpf-map implementation behaves the same way as tlpm.
11  */
12 
13 #include <assert.h>
14 #include <errno.h>
15 #include <inttypes.h>
16 #include <linux/bpf.h>
17 #include <stdio.h>
18 #include <stdlib.h>
19 #include <string.h>
20 #include <time.h>
21 #include <unistd.h>
22 #include <arpa/inet.h>
23 #include <sys/time.h>
24 #include <sys/resource.h>
25 
26 #include <bpf/bpf.h>
27 #include "bpf_util.h"
28 
29 struct tlpm_node {
30 	struct tlpm_node *next;
31 	size_t n_bits;
32 	uint8_t key[];
33 };
34 
35 static struct tlpm_node *tlpm_match(struct tlpm_node *list,
36 				    const uint8_t *key,
37 				    size_t n_bits);
38 
39 static struct tlpm_node *tlpm_add(struct tlpm_node *list,
40 				  const uint8_t *key,
41 				  size_t n_bits)
42 {
43 	struct tlpm_node *node;
44 	size_t n;
45 
46 	n = (n_bits + 7) / 8;
47 
48 	/* 'overwrite' an equivalent entry if one already exists */
49 	node = tlpm_match(list, key, n_bits);
50 	if (node && node->n_bits == n_bits) {
51 		memcpy(node->key, key, n);
52 		return list;
53 	}
54 
55 	/* add new entry with @key/@n_bits to @list and return new head */
56 
57 	node = malloc(sizeof(*node) + n);
58 	assert(node);
59 
60 	node->next = list;
61 	node->n_bits = n_bits;
62 	memcpy(node->key, key, n);
63 
64 	return node;
65 }
66 
67 static void tlpm_clear(struct tlpm_node *list)
68 {
69 	struct tlpm_node *node;
70 
71 	/* free all entries in @list */
72 
73 	while ((node = list)) {
74 		list = list->next;
75 		free(node);
76 	}
77 }
78 
79 static struct tlpm_node *tlpm_match(struct tlpm_node *list,
80 				    const uint8_t *key,
81 				    size_t n_bits)
82 {
83 	struct tlpm_node *best = NULL;
84 	size_t i;
85 
86 	/* Perform longest prefix-match on @key/@n_bits. That is, iterate all
87 	 * entries and match each prefix against @key. Remember the "best"
88 	 * entry we find (i.e., the longest prefix that matches) and return it
89 	 * to the caller when done.
90 	 */
91 
92 	for ( ; list; list = list->next) {
93 		for (i = 0; i < n_bits && i < list->n_bits; ++i) {
94 			if ((key[i / 8] & (1 << (7 - i % 8))) !=
95 			    (list->key[i / 8] & (1 << (7 - i % 8))))
96 				break;
97 		}
98 
99 		if (i >= list->n_bits) {
100 			if (!best || i > best->n_bits)
101 				best = list;
102 		}
103 	}
104 
105 	return best;
106 }
107 
108 static struct tlpm_node *tlpm_delete(struct tlpm_node *list,
109 				     const uint8_t *key,
110 				     size_t n_bits)
111 {
112 	struct tlpm_node *best = tlpm_match(list, key, n_bits);
113 	struct tlpm_node *node;
114 
115 	if (!best || best->n_bits != n_bits)
116 		return list;
117 
118 	if (best == list) {
119 		node = best->next;
120 		free(best);
121 		return node;
122 	}
123 
124 	for (node = list; node; node = node->next) {
125 		if (node->next == best) {
126 			node->next = best->next;
127 			free(best);
128 			return list;
129 		}
130 	}
131 	/* should never get here */
132 	assert(0);
133 	return list;
134 }
135 
136 static void test_lpm_basic(void)
137 {
138 	struct tlpm_node *list = NULL, *t1, *t2;
139 
140 	/* very basic, static tests to verify tlpm works as expected */
141 
142 	assert(!tlpm_match(list, (uint8_t[]){ 0xff }, 8));
143 
144 	t1 = list = tlpm_add(list, (uint8_t[]){ 0xff }, 8);
145 	assert(t1 == tlpm_match(list, (uint8_t[]){ 0xff }, 8));
146 	assert(t1 == tlpm_match(list, (uint8_t[]){ 0xff, 0xff }, 16));
147 	assert(t1 == tlpm_match(list, (uint8_t[]){ 0xff, 0x00 }, 16));
148 	assert(!tlpm_match(list, (uint8_t[]){ 0x7f }, 8));
149 	assert(!tlpm_match(list, (uint8_t[]){ 0xfe }, 8));
150 	assert(!tlpm_match(list, (uint8_t[]){ 0xff }, 7));
151 
152 	t2 = list = tlpm_add(list, (uint8_t[]){ 0xff, 0xff }, 16);
153 	assert(t1 == tlpm_match(list, (uint8_t[]){ 0xff }, 8));
154 	assert(t2 == tlpm_match(list, (uint8_t[]){ 0xff, 0xff }, 16));
155 	assert(t1 == tlpm_match(list, (uint8_t[]){ 0xff, 0xff }, 15));
156 	assert(!tlpm_match(list, (uint8_t[]){ 0x7f, 0xff }, 16));
157 
158 	list = tlpm_delete(list, (uint8_t[]){ 0xff, 0xff }, 16);
159 	assert(t1 == tlpm_match(list, (uint8_t[]){ 0xff }, 8));
160 	assert(t1 == tlpm_match(list, (uint8_t[]){ 0xff, 0xff }, 16));
161 
162 	list = tlpm_delete(list, (uint8_t[]){ 0xff }, 8);
163 	assert(!tlpm_match(list, (uint8_t[]){ 0xff }, 8));
164 
165 	tlpm_clear(list);
166 }
167 
168 static void test_lpm_order(void)
169 {
170 	struct tlpm_node *t1, *t2, *l1 = NULL, *l2 = NULL;
171 	size_t i, j;
172 
173 	/* Verify the tlpm implementation works correctly regardless of the
174 	 * order of entries. Insert a random set of entries into @l1, and copy
175 	 * the same data in reverse order into @l2. Then verify a lookup of
176 	 * random keys will yield the same result in both sets.
177 	 */
178 
179 	for (i = 0; i < (1 << 12); ++i)
180 		l1 = tlpm_add(l1, (uint8_t[]){
181 					rand() % 0xff,
182 					rand() % 0xff,
183 				}, rand() % 16 + 1);
184 
185 	for (t1 = l1; t1; t1 = t1->next)
186 		l2 = tlpm_add(l2, t1->key, t1->n_bits);
187 
188 	for (i = 0; i < (1 << 8); ++i) {
189 		uint8_t key[] = { rand() % 0xff, rand() % 0xff };
190 
191 		t1 = tlpm_match(l1, key, 16);
192 		t2 = tlpm_match(l2, key, 16);
193 
194 		assert(!t1 == !t2);
195 		if (t1) {
196 			assert(t1->n_bits == t2->n_bits);
197 			for (j = 0; j < t1->n_bits; ++j)
198 				assert((t1->key[j / 8] & (1 << (7 - j % 8))) ==
199 				       (t2->key[j / 8] & (1 << (7 - j % 8))));
200 		}
201 	}
202 
203 	tlpm_clear(l1);
204 	tlpm_clear(l2);
205 }
206 
207 static void test_lpm_map(int keysize)
208 {
209 	size_t i, j, n_matches, n_matches_after_delete, n_nodes, n_lookups;
210 	struct tlpm_node *t, *list = NULL;
211 	struct bpf_lpm_trie_key *key;
212 	uint8_t *data, *value;
213 	int r, map;
214 
215 	/* Compare behavior of tlpm vs. bpf-lpm. Create a randomized set of
216 	 * prefixes and insert it into both tlpm and bpf-lpm. Then run some
217 	 * randomized lookups and verify both maps return the same result.
218 	 */
219 
220 	n_matches = 0;
221 	n_matches_after_delete = 0;
222 	n_nodes = 1 << 8;
223 	n_lookups = 1 << 16;
224 
225 	data = alloca(keysize);
226 	memset(data, 0, keysize);
227 
228 	value = alloca(keysize + 1);
229 	memset(value, 0, keysize + 1);
230 
231 	key = alloca(sizeof(*key) + keysize);
232 	memset(key, 0, sizeof(*key) + keysize);
233 
234 	map = bpf_create_map(BPF_MAP_TYPE_LPM_TRIE,
235 			     sizeof(*key) + keysize,
236 			     keysize + 1,
237 			     4096,
238 			     BPF_F_NO_PREALLOC);
239 	assert(map >= 0);
240 
241 	for (i = 0; i < n_nodes; ++i) {
242 		for (j = 0; j < keysize; ++j)
243 			value[j] = rand() & 0xff;
244 		value[keysize] = rand() % (8 * keysize + 1);
245 
246 		list = tlpm_add(list, value, value[keysize]);
247 
248 		key->prefixlen = value[keysize];
249 		memcpy(key->data, value, keysize);
250 		r = bpf_map_update_elem(map, key, value, 0);
251 		assert(!r);
252 	}
253 
254 	for (i = 0; i < n_lookups; ++i) {
255 		for (j = 0; j < keysize; ++j)
256 			data[j] = rand() & 0xff;
257 
258 		t = tlpm_match(list, data, 8 * keysize);
259 
260 		key->prefixlen = 8 * keysize;
261 		memcpy(key->data, data, keysize);
262 		r = bpf_map_lookup_elem(map, key, value);
263 		assert(!r || errno == ENOENT);
264 		assert(!t == !!r);
265 
266 		if (t) {
267 			++n_matches;
268 			assert(t->n_bits == value[keysize]);
269 			for (j = 0; j < t->n_bits; ++j)
270 				assert((t->key[j / 8] & (1 << (7 - j % 8))) ==
271 				       (value[j / 8] & (1 << (7 - j % 8))));
272 		}
273 	}
274 
275 	/* Remove the first half of the elements in the tlpm and the
276 	 * corresponding nodes from the bpf-lpm.  Then run the same
277 	 * large number of random lookups in both and make sure they match.
278 	 * Note: we need to count the number of nodes actually inserted
279 	 * since there may have been duplicates.
280 	 */
281 	for (i = 0, t = list; t; i++, t = t->next)
282 		;
283 	for (j = 0; j < i / 2; ++j) {
284 		key->prefixlen = list->n_bits;
285 		memcpy(key->data, list->key, keysize);
286 		r = bpf_map_delete_elem(map, key);
287 		assert(!r);
288 		list = tlpm_delete(list, list->key, list->n_bits);
289 		assert(list);
290 	}
291 	for (i = 0; i < n_lookups; ++i) {
292 		for (j = 0; j < keysize; ++j)
293 			data[j] = rand() & 0xff;
294 
295 		t = tlpm_match(list, data, 8 * keysize);
296 
297 		key->prefixlen = 8 * keysize;
298 		memcpy(key->data, data, keysize);
299 		r = bpf_map_lookup_elem(map, key, value);
300 		assert(!r || errno == ENOENT);
301 		assert(!t == !!r);
302 
303 		if (t) {
304 			++n_matches_after_delete;
305 			assert(t->n_bits == value[keysize]);
306 			for (j = 0; j < t->n_bits; ++j)
307 				assert((t->key[j / 8] & (1 << (7 - j % 8))) ==
308 				       (value[j / 8] & (1 << (7 - j % 8))));
309 		}
310 	}
311 
312 	close(map);
313 	tlpm_clear(list);
314 
315 	/* With 255 random nodes in the map, we are pretty likely to match
316 	 * something on every lookup. For statistics, use this:
317 	 *
318 	 *     printf("          nodes: %zu\n"
319 	 *            "        lookups: %zu\n"
320 	 *            "        matches: %zu\n"
321 	 *            "matches(delete): %zu\n",
322 	 *            n_nodes, n_lookups, n_matches, n_matches_after_delete);
323 	 */
324 }
325 
326 /* Test the implementation with some 'real world' examples */
327 
328 static void test_lpm_ipaddr(void)
329 {
330 	struct bpf_lpm_trie_key *key_ipv4;
331 	struct bpf_lpm_trie_key *key_ipv6;
332 	size_t key_size_ipv4;
333 	size_t key_size_ipv6;
334 	int map_fd_ipv4;
335 	int map_fd_ipv6;
336 	__u64 value;
337 
338 	key_size_ipv4 = sizeof(*key_ipv4) + sizeof(__u32);
339 	key_size_ipv6 = sizeof(*key_ipv6) + sizeof(__u32) * 4;
340 	key_ipv4 = alloca(key_size_ipv4);
341 	key_ipv6 = alloca(key_size_ipv6);
342 
343 	map_fd_ipv4 = bpf_create_map(BPF_MAP_TYPE_LPM_TRIE,
344 				     key_size_ipv4, sizeof(value),
345 				     100, BPF_F_NO_PREALLOC);
346 	assert(map_fd_ipv4 >= 0);
347 
348 	map_fd_ipv6 = bpf_create_map(BPF_MAP_TYPE_LPM_TRIE,
349 				     key_size_ipv6, sizeof(value),
350 				     100, BPF_F_NO_PREALLOC);
351 	assert(map_fd_ipv6 >= 0);
352 
353 	/* Fill data some IPv4 and IPv6 address ranges */
354 	value = 1;
355 	key_ipv4->prefixlen = 16;
356 	inet_pton(AF_INET, "192.168.0.0", key_ipv4->data);
357 	assert(bpf_map_update_elem(map_fd_ipv4, key_ipv4, &value, 0) == 0);
358 
359 	value = 2;
360 	key_ipv4->prefixlen = 24;
361 	inet_pton(AF_INET, "192.168.0.0", key_ipv4->data);
362 	assert(bpf_map_update_elem(map_fd_ipv4, key_ipv4, &value, 0) == 0);
363 
364 	value = 3;
365 	key_ipv4->prefixlen = 24;
366 	inet_pton(AF_INET, "192.168.128.0", key_ipv4->data);
367 	assert(bpf_map_update_elem(map_fd_ipv4, key_ipv4, &value, 0) == 0);
368 
369 	value = 5;
370 	key_ipv4->prefixlen = 24;
371 	inet_pton(AF_INET, "192.168.1.0", key_ipv4->data);
372 	assert(bpf_map_update_elem(map_fd_ipv4, key_ipv4, &value, 0) == 0);
373 
374 	value = 4;
375 	key_ipv4->prefixlen = 23;
376 	inet_pton(AF_INET, "192.168.0.0", key_ipv4->data);
377 	assert(bpf_map_update_elem(map_fd_ipv4, key_ipv4, &value, 0) == 0);
378 
379 	value = 0xdeadbeef;
380 	key_ipv6->prefixlen = 64;
381 	inet_pton(AF_INET6, "2a00:1450:4001:814::200e", key_ipv6->data);
382 	assert(bpf_map_update_elem(map_fd_ipv6, key_ipv6, &value, 0) == 0);
383 
384 	/* Set tprefixlen to maximum for lookups */
385 	key_ipv4->prefixlen = 32;
386 	key_ipv6->prefixlen = 128;
387 
388 	/* Test some lookups that should come back with a value */
389 	inet_pton(AF_INET, "192.168.128.23", key_ipv4->data);
390 	assert(bpf_map_lookup_elem(map_fd_ipv4, key_ipv4, &value) == 0);
391 	assert(value == 3);
392 
393 	inet_pton(AF_INET, "192.168.0.1", key_ipv4->data);
394 	assert(bpf_map_lookup_elem(map_fd_ipv4, key_ipv4, &value) == 0);
395 	assert(value == 2);
396 
397 	inet_pton(AF_INET6, "2a00:1450:4001:814::", key_ipv6->data);
398 	assert(bpf_map_lookup_elem(map_fd_ipv6, key_ipv6, &value) == 0);
399 	assert(value == 0xdeadbeef);
400 
401 	inet_pton(AF_INET6, "2a00:1450:4001:814::1", key_ipv6->data);
402 	assert(bpf_map_lookup_elem(map_fd_ipv6, key_ipv6, &value) == 0);
403 	assert(value == 0xdeadbeef);
404 
405 	/* Test some lookups that should not match any entry */
406 	inet_pton(AF_INET, "10.0.0.1", key_ipv4->data);
407 	assert(bpf_map_lookup_elem(map_fd_ipv4, key_ipv4, &value) == -1 &&
408 	       errno == ENOENT);
409 
410 	inet_pton(AF_INET, "11.11.11.11", key_ipv4->data);
411 	assert(bpf_map_lookup_elem(map_fd_ipv4, key_ipv4, &value) == -1 &&
412 	       errno == ENOENT);
413 
414 	inet_pton(AF_INET6, "2a00:ffff::", key_ipv6->data);
415 	assert(bpf_map_lookup_elem(map_fd_ipv6, key_ipv6, &value) == -1 &&
416 	       errno == ENOENT);
417 
418 	close(map_fd_ipv4);
419 	close(map_fd_ipv6);
420 }
421 
422 static void test_lpm_delete(void)
423 {
424 	struct bpf_lpm_trie_key *key;
425 	size_t key_size;
426 	int map_fd;
427 	__u64 value;
428 
429 	key_size = sizeof(*key) + sizeof(__u32);
430 	key = alloca(key_size);
431 
432 	map_fd = bpf_create_map(BPF_MAP_TYPE_LPM_TRIE,
433 				key_size, sizeof(value),
434 				100, BPF_F_NO_PREALLOC);
435 	assert(map_fd >= 0);
436 
437 	/* Add nodes:
438 	 * 192.168.0.0/16   (1)
439 	 * 192.168.0.0/24   (2)
440 	 * 192.168.128.0/24 (3)
441 	 * 192.168.1.0/24   (4)
442 	 *
443 	 *         (1)
444 	 *        /   \
445          *     (IM)    (3)
446 	 *    /   \
447          *   (2)  (4)
448 	 */
449 	value = 1;
450 	key->prefixlen = 16;
451 	inet_pton(AF_INET, "192.168.0.0", key->data);
452 	assert(bpf_map_update_elem(map_fd, key, &value, 0) == 0);
453 
454 	value = 2;
455 	key->prefixlen = 24;
456 	inet_pton(AF_INET, "192.168.0.0", key->data);
457 	assert(bpf_map_update_elem(map_fd, key, &value, 0) == 0);
458 
459 	value = 3;
460 	key->prefixlen = 24;
461 	inet_pton(AF_INET, "192.168.128.0", key->data);
462 	assert(bpf_map_update_elem(map_fd, key, &value, 0) == 0);
463 
464 	value = 4;
465 	key->prefixlen = 24;
466 	inet_pton(AF_INET, "192.168.1.0", key->data);
467 	assert(bpf_map_update_elem(map_fd, key, &value, 0) == 0);
468 
469 	/* remove non-existent node */
470 	key->prefixlen = 32;
471 	inet_pton(AF_INET, "10.0.0.1", key->data);
472 	assert(bpf_map_lookup_elem(map_fd, key, &value) == -1 &&
473 		errno == ENOENT);
474 
475 	/* assert initial lookup */
476 	key->prefixlen = 32;
477 	inet_pton(AF_INET, "192.168.0.1", key->data);
478 	assert(bpf_map_lookup_elem(map_fd, key, &value) == 0);
479 	assert(value == 2);
480 
481 	/* remove leaf node */
482 	key->prefixlen = 24;
483 	inet_pton(AF_INET, "192.168.0.0", key->data);
484 	assert(bpf_map_delete_elem(map_fd, key) == 0);
485 
486 	key->prefixlen = 32;
487 	inet_pton(AF_INET, "192.168.0.1", key->data);
488 	assert(bpf_map_lookup_elem(map_fd, key, &value) == 0);
489 	assert(value == 1);
490 
491 	/* remove leaf (and intermediary) node */
492 	key->prefixlen = 24;
493 	inet_pton(AF_INET, "192.168.1.0", key->data);
494 	assert(bpf_map_delete_elem(map_fd, key) == 0);
495 
496 	key->prefixlen = 32;
497 	inet_pton(AF_INET, "192.168.1.1", key->data);
498 	assert(bpf_map_lookup_elem(map_fd, key, &value) == 0);
499 	assert(value == 1);
500 
501 	/* remove root node */
502 	key->prefixlen = 16;
503 	inet_pton(AF_INET, "192.168.0.0", key->data);
504 	assert(bpf_map_delete_elem(map_fd, key) == 0);
505 
506 	key->prefixlen = 32;
507 	inet_pton(AF_INET, "192.168.128.1", key->data);
508 	assert(bpf_map_lookup_elem(map_fd, key, &value) == 0);
509 	assert(value == 3);
510 
511 	/* remove last node */
512 	key->prefixlen = 24;
513 	inet_pton(AF_INET, "192.168.128.0", key->data);
514 	assert(bpf_map_delete_elem(map_fd, key) == 0);
515 
516 	key->prefixlen = 32;
517 	inet_pton(AF_INET, "192.168.128.1", key->data);
518 	assert(bpf_map_lookup_elem(map_fd, key, &value) == -1 &&
519 		errno == ENOENT);
520 
521 	close(map_fd);
522 }
523 
524 int main(void)
525 {
526 	struct rlimit limit  = { RLIM_INFINITY, RLIM_INFINITY };
527 	int i, ret;
528 
529 	/* we want predictable, pseudo random tests */
530 	srand(0xf00ba1);
531 
532 	/* allow unlimited locked memory */
533 	ret = setrlimit(RLIMIT_MEMLOCK, &limit);
534 	if (ret < 0)
535 		perror("Unable to lift memlock rlimit");
536 
537 	test_lpm_basic();
538 	test_lpm_order();
539 
540 	/* Test with 8, 16, 24, 32, ... 128 bit prefix length */
541 	for (i = 1; i <= 16; ++i)
542 		test_lpm_map(i);
543 
544 	test_lpm_ipaddr();
545 
546 	test_lpm_delete();
547 
548 	printf("test_lpm: OK\n");
549 	return 0;
550 }
551