1 // SPDX-License-Identifier: GPL-2.0
2
3 /* eBPF example program:
4 *
5 * - Creates arraymap in kernel with 4 bytes keys and 8 byte values
6 *
7 * - Loads eBPF program
8 *
9 * The eBPF program accesses the map passed in to store two pieces of
10 * information. The number of invocations of the program, which maps
11 * to the number of packets received, is stored to key 0. Key 1 is
12 * incremented on each iteration by the number of bytes stored in
13 * the skb. The program also stores the number of received bytes
14 * in the cgroup storage.
15 *
16 * - Attaches the new program to a cgroup using BPF_PROG_ATTACH
17 *
18 * - Every second, reads map[0] and map[1] to see how many bytes and
19 * packets were seen on any socket of tasks in the given cgroup.
20 */
21
22 #define _GNU_SOURCE
23
24 #include <stdio.h>
25 #include <stdlib.h>
26 #include <assert.h>
27 #include <sys/resource.h>
28 #include <sys/time.h>
29 #include <unistd.h>
30 #include <linux/filter.h>
31
32 #include <linux/bpf.h>
33 #include <bpf/bpf.h>
34
35 #include "bpf_util.h"
36 #include "bpf_rlimit.h"
37 #include "cgroup_helpers.h"
38
39 #define FOO "/foo"
40 #define BAR "/foo/bar/"
41 #define PING_CMD "ping -q -c1 -w1 127.0.0.1 > /dev/null"
42
43 char bpf_log_buf[BPF_LOG_BUF_SIZE];
44
45 #ifdef DEBUG
46 #define debug(args...) printf(args)
47 #else
48 #define debug(args...)
49 #endif
50
prog_load(int verdict)51 static int prog_load(int verdict)
52 {
53 int ret;
54 struct bpf_insn prog[] = {
55 BPF_MOV64_IMM(BPF_REG_0, verdict), /* r0 = verdict */
56 BPF_EXIT_INSN(),
57 };
58 size_t insns_cnt = sizeof(prog) / sizeof(struct bpf_insn);
59
60 ret = bpf_load_program(BPF_PROG_TYPE_CGROUP_SKB,
61 prog, insns_cnt, "GPL", 0,
62 bpf_log_buf, BPF_LOG_BUF_SIZE);
63
64 if (ret < 0) {
65 log_err("Loading program");
66 printf("Output from verifier:\n%s\n-------\n", bpf_log_buf);
67 return 0;
68 }
69 return ret;
70 }
71
test_foo_bar(void)72 static int test_foo_bar(void)
73 {
74 int drop_prog, allow_prog, foo = 0, bar = 0, rc = 0;
75
76 allow_prog = prog_load(1);
77 if (!allow_prog)
78 goto err;
79
80 drop_prog = prog_load(0);
81 if (!drop_prog)
82 goto err;
83
84 if (setup_cgroup_environment())
85 goto err;
86
87 /* Create cgroup /foo, get fd, and join it */
88 foo = create_and_get_cgroup(FOO);
89 if (foo < 0)
90 goto err;
91
92 if (join_cgroup(FOO))
93 goto err;
94
95 if (bpf_prog_attach(drop_prog, foo, BPF_CGROUP_INET_EGRESS,
96 BPF_F_ALLOW_OVERRIDE)) {
97 log_err("Attaching prog to /foo");
98 goto err;
99 }
100
101 debug("Attached DROP prog. This ping in cgroup /foo should fail...\n");
102 assert(system(PING_CMD) != 0);
103
104 /* Create cgroup /foo/bar, get fd, and join it */
105 bar = create_and_get_cgroup(BAR);
106 if (bar < 0)
107 goto err;
108
109 if (join_cgroup(BAR))
110 goto err;
111
112 debug("Attached DROP prog. This ping in cgroup /foo/bar should fail...\n");
113 assert(system(PING_CMD) != 0);
114
115 if (bpf_prog_attach(allow_prog, bar, BPF_CGROUP_INET_EGRESS,
116 BPF_F_ALLOW_OVERRIDE)) {
117 log_err("Attaching prog to /foo/bar");
118 goto err;
119 }
120
121 debug("Attached PASS prog. This ping in cgroup /foo/bar should pass...\n");
122 assert(system(PING_CMD) == 0);
123
124 if (bpf_prog_detach(bar, BPF_CGROUP_INET_EGRESS)) {
125 log_err("Detaching program from /foo/bar");
126 goto err;
127 }
128
129 debug("Detached PASS from /foo/bar while DROP is attached to /foo.\n"
130 "This ping in cgroup /foo/bar should fail...\n");
131 assert(system(PING_CMD) != 0);
132
133 if (bpf_prog_attach(allow_prog, bar, BPF_CGROUP_INET_EGRESS,
134 BPF_F_ALLOW_OVERRIDE)) {
135 log_err("Attaching prog to /foo/bar");
136 goto err;
137 }
138
139 if (bpf_prog_detach(foo, BPF_CGROUP_INET_EGRESS)) {
140 log_err("Detaching program from /foo");
141 goto err;
142 }
143
144 debug("Attached PASS from /foo/bar and detached DROP from /foo.\n"
145 "This ping in cgroup /foo/bar should pass...\n");
146 assert(system(PING_CMD) == 0);
147
148 if (bpf_prog_attach(allow_prog, bar, BPF_CGROUP_INET_EGRESS,
149 BPF_F_ALLOW_OVERRIDE)) {
150 log_err("Attaching prog to /foo/bar");
151 goto err;
152 }
153
154 if (!bpf_prog_attach(allow_prog, bar, BPF_CGROUP_INET_EGRESS, 0)) {
155 errno = 0;
156 log_err("Unexpected success attaching prog to /foo/bar");
157 goto err;
158 }
159
160 if (bpf_prog_detach(bar, BPF_CGROUP_INET_EGRESS)) {
161 log_err("Detaching program from /foo/bar");
162 goto err;
163 }
164
165 if (!bpf_prog_detach(foo, BPF_CGROUP_INET_EGRESS)) {
166 errno = 0;
167 log_err("Unexpected success in double detach from /foo");
168 goto err;
169 }
170
171 if (bpf_prog_attach(allow_prog, foo, BPF_CGROUP_INET_EGRESS, 0)) {
172 log_err("Attaching non-overridable prog to /foo");
173 goto err;
174 }
175
176 if (!bpf_prog_attach(allow_prog, bar, BPF_CGROUP_INET_EGRESS, 0)) {
177 errno = 0;
178 log_err("Unexpected success attaching non-overridable prog to /foo/bar");
179 goto err;
180 }
181
182 if (!bpf_prog_attach(allow_prog, bar, BPF_CGROUP_INET_EGRESS,
183 BPF_F_ALLOW_OVERRIDE)) {
184 errno = 0;
185 log_err("Unexpected success attaching overridable prog to /foo/bar");
186 goto err;
187 }
188
189 if (!bpf_prog_attach(allow_prog, foo, BPF_CGROUP_INET_EGRESS,
190 BPF_F_ALLOW_OVERRIDE)) {
191 errno = 0;
192 log_err("Unexpected success attaching overridable prog to /foo");
193 goto err;
194 }
195
196 if (bpf_prog_attach(drop_prog, foo, BPF_CGROUP_INET_EGRESS, 0)) {
197 log_err("Attaching different non-overridable prog to /foo");
198 goto err;
199 }
200
201 goto out;
202
203 err:
204 rc = 1;
205
206 out:
207 close(foo);
208 close(bar);
209 cleanup_cgroup_environment();
210 if (!rc)
211 printf("#override:PASS\n");
212 else
213 printf("#override:FAIL\n");
214 return rc;
215 }
216
217 static int map_fd = -1;
218
prog_load_cnt(int verdict,int val)219 static int prog_load_cnt(int verdict, int val)
220 {
221 int cgroup_storage_fd, percpu_cgroup_storage_fd;
222
223 if (map_fd < 0)
224 map_fd = bpf_create_map(BPF_MAP_TYPE_ARRAY, 4, 8, 1, 0);
225 if (map_fd < 0) {
226 printf("failed to create map '%s'\n", strerror(errno));
227 return -1;
228 }
229
230 cgroup_storage_fd = bpf_create_map(BPF_MAP_TYPE_CGROUP_STORAGE,
231 sizeof(struct bpf_cgroup_storage_key), 8, 0, 0);
232 if (cgroup_storage_fd < 0) {
233 printf("failed to create map '%s'\n", strerror(errno));
234 return -1;
235 }
236
237 percpu_cgroup_storage_fd = bpf_create_map(
238 BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE,
239 sizeof(struct bpf_cgroup_storage_key), 8, 0, 0);
240 if (percpu_cgroup_storage_fd < 0) {
241 printf("failed to create map '%s'\n", strerror(errno));
242 return -1;
243 }
244
245 struct bpf_insn prog[] = {
246 BPF_MOV32_IMM(BPF_REG_0, 0),
247 BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_0, -4), /* *(u32 *)(fp - 4) = r0 */
248 BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
249 BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), /* r2 = fp - 4 */
250 BPF_LD_MAP_FD(BPF_REG_1, map_fd),
251 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem),
252 BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2),
253 BPF_MOV64_IMM(BPF_REG_1, val), /* r1 = 1 */
254 BPF_RAW_INSN(BPF_STX | BPF_XADD | BPF_DW, BPF_REG_0, BPF_REG_1, 0, 0), /* xadd r0 += r1 */
255
256 BPF_LD_MAP_FD(BPF_REG_1, cgroup_storage_fd),
257 BPF_MOV64_IMM(BPF_REG_2, 0),
258 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_get_local_storage),
259 BPF_MOV64_IMM(BPF_REG_1, val),
260 BPF_RAW_INSN(BPF_STX | BPF_XADD | BPF_W, BPF_REG_0, BPF_REG_1, 0, 0),
261
262 BPF_LD_MAP_FD(BPF_REG_1, percpu_cgroup_storage_fd),
263 BPF_MOV64_IMM(BPF_REG_2, 0),
264 BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_get_local_storage),
265 BPF_LDX_MEM(BPF_W, BPF_REG_3, BPF_REG_0, 0),
266 BPF_ALU64_IMM(BPF_ADD, BPF_REG_3, 0x1),
267 BPF_STX_MEM(BPF_W, BPF_REG_0, BPF_REG_3, 0),
268
269 BPF_MOV64_IMM(BPF_REG_0, verdict), /* r0 = verdict */
270 BPF_EXIT_INSN(),
271 };
272 size_t insns_cnt = sizeof(prog) / sizeof(struct bpf_insn);
273 int ret;
274
275 ret = bpf_load_program(BPF_PROG_TYPE_CGROUP_SKB,
276 prog, insns_cnt, "GPL", 0,
277 bpf_log_buf, BPF_LOG_BUF_SIZE);
278
279 if (ret < 0) {
280 log_err("Loading program");
281 printf("Output from verifier:\n%s\n-------\n", bpf_log_buf);
282 return 0;
283 }
284 close(cgroup_storage_fd);
285 return ret;
286 }
287
288
test_multiprog(void)289 static int test_multiprog(void)
290 {
291 __u32 prog_ids[4], prog_cnt = 0, attach_flags, saved_prog_id;
292 int cg1 = 0, cg2 = 0, cg3 = 0, cg4 = 0, cg5 = 0, key = 0;
293 int drop_prog, allow_prog[6] = {}, rc = 0;
294 unsigned long long value;
295 int i = 0;
296
297 for (i = 0; i < 6; i++) {
298 allow_prog[i] = prog_load_cnt(1, 1 << i);
299 if (!allow_prog[i])
300 goto err;
301 }
302 drop_prog = prog_load_cnt(0, 1);
303 if (!drop_prog)
304 goto err;
305
306 if (setup_cgroup_environment())
307 goto err;
308
309 cg1 = create_and_get_cgroup("/cg1");
310 if (cg1 < 0)
311 goto err;
312 cg2 = create_and_get_cgroup("/cg1/cg2");
313 if (cg2 < 0)
314 goto err;
315 cg3 = create_and_get_cgroup("/cg1/cg2/cg3");
316 if (cg3 < 0)
317 goto err;
318 cg4 = create_and_get_cgroup("/cg1/cg2/cg3/cg4");
319 if (cg4 < 0)
320 goto err;
321 cg5 = create_and_get_cgroup("/cg1/cg2/cg3/cg4/cg5");
322 if (cg5 < 0)
323 goto err;
324
325 if (join_cgroup("/cg1/cg2/cg3/cg4/cg5"))
326 goto err;
327
328 if (bpf_prog_attach(allow_prog[0], cg1, BPF_CGROUP_INET_EGRESS,
329 BPF_F_ALLOW_MULTI)) {
330 log_err("Attaching prog to cg1");
331 goto err;
332 }
333 if (!bpf_prog_attach(allow_prog[0], cg1, BPF_CGROUP_INET_EGRESS,
334 BPF_F_ALLOW_MULTI)) {
335 log_err("Unexpected success attaching the same prog to cg1");
336 goto err;
337 }
338 if (bpf_prog_attach(allow_prog[1], cg1, BPF_CGROUP_INET_EGRESS,
339 BPF_F_ALLOW_MULTI)) {
340 log_err("Attaching prog2 to cg1");
341 goto err;
342 }
343 if (bpf_prog_attach(allow_prog[2], cg2, BPF_CGROUP_INET_EGRESS,
344 BPF_F_ALLOW_OVERRIDE)) {
345 log_err("Attaching prog to cg2");
346 goto err;
347 }
348 if (bpf_prog_attach(allow_prog[3], cg3, BPF_CGROUP_INET_EGRESS,
349 BPF_F_ALLOW_MULTI)) {
350 log_err("Attaching prog to cg3");
351 goto err;
352 }
353 if (bpf_prog_attach(allow_prog[4], cg4, BPF_CGROUP_INET_EGRESS,
354 BPF_F_ALLOW_OVERRIDE)) {
355 log_err("Attaching prog to cg4");
356 goto err;
357 }
358 if (bpf_prog_attach(allow_prog[5], cg5, BPF_CGROUP_INET_EGRESS, 0)) {
359 log_err("Attaching prog to cg5");
360 goto err;
361 }
362 assert(system(PING_CMD) == 0);
363 assert(bpf_map_lookup_elem(map_fd, &key, &value) == 0);
364 assert(value == 1 + 2 + 8 + 32);
365
366 /* query the number of effective progs in cg5 */
367 assert(bpf_prog_query(cg5, BPF_CGROUP_INET_EGRESS, BPF_F_QUERY_EFFECTIVE,
368 NULL, NULL, &prog_cnt) == 0);
369 assert(prog_cnt == 4);
370 /* retrieve prog_ids of effective progs in cg5 */
371 assert(bpf_prog_query(cg5, BPF_CGROUP_INET_EGRESS, BPF_F_QUERY_EFFECTIVE,
372 &attach_flags, prog_ids, &prog_cnt) == 0);
373 assert(prog_cnt == 4);
374 assert(attach_flags == 0);
375 saved_prog_id = prog_ids[0];
376 /* check enospc handling */
377 prog_ids[0] = 0;
378 prog_cnt = 2;
379 assert(bpf_prog_query(cg5, BPF_CGROUP_INET_EGRESS, BPF_F_QUERY_EFFECTIVE,
380 &attach_flags, prog_ids, &prog_cnt) == -1 &&
381 errno == ENOSPC);
382 assert(prog_cnt == 4);
383 /* check that prog_ids are returned even when buffer is too small */
384 assert(prog_ids[0] == saved_prog_id);
385 /* retrieve prog_id of single attached prog in cg5 */
386 prog_ids[0] = 0;
387 assert(bpf_prog_query(cg5, BPF_CGROUP_INET_EGRESS, 0,
388 NULL, prog_ids, &prog_cnt) == 0);
389 assert(prog_cnt == 1);
390 assert(prog_ids[0] == saved_prog_id);
391
392 /* detach bottom program and ping again */
393 if (bpf_prog_detach2(-1, cg5, BPF_CGROUP_INET_EGRESS)) {
394 log_err("Detaching prog from cg5");
395 goto err;
396 }
397 value = 0;
398 assert(bpf_map_update_elem(map_fd, &key, &value, 0) == 0);
399 assert(system(PING_CMD) == 0);
400 assert(bpf_map_lookup_elem(map_fd, &key, &value) == 0);
401 assert(value == 1 + 2 + 8 + 16);
402
403 /* detach 3rd from bottom program and ping again */
404 errno = 0;
405 if (!bpf_prog_detach2(0, cg3, BPF_CGROUP_INET_EGRESS)) {
406 log_err("Unexpected success on detach from cg3");
407 goto err;
408 }
409 if (bpf_prog_detach2(allow_prog[3], cg3, BPF_CGROUP_INET_EGRESS)) {
410 log_err("Detaching from cg3");
411 goto err;
412 }
413 value = 0;
414 assert(bpf_map_update_elem(map_fd, &key, &value, 0) == 0);
415 assert(system(PING_CMD) == 0);
416 assert(bpf_map_lookup_elem(map_fd, &key, &value) == 0);
417 assert(value == 1 + 2 + 16);
418
419 /* detach 2nd from bottom program and ping again */
420 if (bpf_prog_detach2(-1, cg4, BPF_CGROUP_INET_EGRESS)) {
421 log_err("Detaching prog from cg4");
422 goto err;
423 }
424 value = 0;
425 assert(bpf_map_update_elem(map_fd, &key, &value, 0) == 0);
426 assert(system(PING_CMD) == 0);
427 assert(bpf_map_lookup_elem(map_fd, &key, &value) == 0);
428 assert(value == 1 + 2 + 4);
429
430 prog_cnt = 4;
431 assert(bpf_prog_query(cg5, BPF_CGROUP_INET_EGRESS, BPF_F_QUERY_EFFECTIVE,
432 &attach_flags, prog_ids, &prog_cnt) == 0);
433 assert(prog_cnt == 3);
434 assert(attach_flags == 0);
435 assert(bpf_prog_query(cg5, BPF_CGROUP_INET_EGRESS, 0,
436 NULL, prog_ids, &prog_cnt) == 0);
437 assert(prog_cnt == 0);
438 goto out;
439 err:
440 rc = 1;
441
442 out:
443 for (i = 0; i < 6; i++)
444 if (allow_prog[i] > 0)
445 close(allow_prog[i]);
446 close(cg1);
447 close(cg2);
448 close(cg3);
449 close(cg4);
450 close(cg5);
451 cleanup_cgroup_environment();
452 if (!rc)
453 printf("#multi:PASS\n");
454 else
455 printf("#multi:FAIL\n");
456 return rc;
457 }
458
test_autodetach(void)459 static int test_autodetach(void)
460 {
461 __u32 prog_cnt = 4, attach_flags;
462 int allow_prog[2] = {0};
463 __u32 prog_ids[2] = {0};
464 int cg = 0, i, rc = -1;
465 void *ptr = NULL;
466 int attempts;
467
468 for (i = 0; i < ARRAY_SIZE(allow_prog); i++) {
469 allow_prog[i] = prog_load_cnt(1, 1 << i);
470 if (!allow_prog[i])
471 goto err;
472 }
473
474 if (setup_cgroup_environment())
475 goto err;
476
477 /* create a cgroup, attach two programs and remember their ids */
478 cg = create_and_get_cgroup("/cg_autodetach");
479 if (cg < 0)
480 goto err;
481
482 if (join_cgroup("/cg_autodetach"))
483 goto err;
484
485 for (i = 0; i < ARRAY_SIZE(allow_prog); i++) {
486 if (bpf_prog_attach(allow_prog[i], cg, BPF_CGROUP_INET_EGRESS,
487 BPF_F_ALLOW_MULTI)) {
488 log_err("Attaching prog[%d] to cg:egress", i);
489 goto err;
490 }
491 }
492
493 /* make sure that programs are attached and run some traffic */
494 assert(bpf_prog_query(cg, BPF_CGROUP_INET_EGRESS, 0, &attach_flags,
495 prog_ids, &prog_cnt) == 0);
496 assert(system(PING_CMD) == 0);
497
498 /* allocate some memory (4Mb) to pin the original cgroup */
499 ptr = malloc(4 * (1 << 20));
500 if (!ptr)
501 goto err;
502
503 /* close programs and cgroup fd */
504 for (i = 0; i < ARRAY_SIZE(allow_prog); i++) {
505 close(allow_prog[i]);
506 allow_prog[i] = 0;
507 }
508
509 close(cg);
510 cg = 0;
511
512 /* leave the cgroup and remove it. don't detach programs */
513 cleanup_cgroup_environment();
514
515 /* wait for the asynchronous auto-detachment.
516 * wait for no more than 5 sec and give up.
517 */
518 for (i = 0; i < ARRAY_SIZE(prog_ids); i++) {
519 for (attempts = 5; attempts >= 0; attempts--) {
520 int fd = bpf_prog_get_fd_by_id(prog_ids[i]);
521
522 if (fd < 0)
523 break;
524
525 /* don't leave the fd open */
526 close(fd);
527
528 if (!attempts)
529 goto err;
530
531 sleep(1);
532 }
533 }
534
535 rc = 0;
536 err:
537 for (i = 0; i < ARRAY_SIZE(allow_prog); i++)
538 if (allow_prog[i] > 0)
539 close(allow_prog[i]);
540 if (cg)
541 close(cg);
542 free(ptr);
543 cleanup_cgroup_environment();
544 if (!rc)
545 printf("#autodetach:PASS\n");
546 else
547 printf("#autodetach:FAIL\n");
548 return rc;
549 }
550
main(void)551 int main(void)
552 {
553 int (*tests[])(void) = {
554 test_foo_bar,
555 test_multiprog,
556 test_autodetach,
557 };
558 int errors = 0;
559 int i;
560
561 for (i = 0; i < ARRAY_SIZE(tests); i++)
562 if (tests[i]())
563 errors++;
564
565 if (errors)
566 printf("test_cgroup_attach:FAIL\n");
567 else
568 printf("test_cgroup_attach:PASS\n");
569
570 return errors ? EXIT_FAILURE : EXIT_SUCCESS;
571 }
572