• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2012 The Chromium OS Authors. All rights reserved.
3  * Use of this source code is governed by the GPLv2 license.
4  *
5  * Test code for seccomp bpf.
6  */
7 
8 #include <asm/siginfo.h>
9 #define __have_siginfo_t 1
10 #define __have_sigval_t 1
11 #define __have_sigevent_t 1
12 
13 #include <errno.h>
14 #include <linux/filter.h>
15 #include <sys/prctl.h>
16 #include <sys/ptrace.h>
17 #include <sys/types.h>
18 #include <sys/user.h>
19 #include <linux/prctl.h>
20 #include <linux/ptrace.h>
21 #include <linux/seccomp.h>
22 #include <pthread.h>
23 #include <semaphore.h>
24 #include <signal.h>
25 #include <stddef.h>
26 #include <stdbool.h>
27 #include <string.h>
28 #include <time.h>
29 #include <linux/elf.h>
30 #include <sys/uio.h>
31 #include <sys/utsname.h>
32 #include <sys/fcntl.h>
33 #include <sys/mman.h>
34 #include <sys/times.h>
35 
36 #define _GNU_SOURCE
37 #include <unistd.h>
38 #include <sys/syscall.h>
39 
40 #include "test_harness.h"
41 
42 #ifndef PR_SET_PTRACER
43 # define PR_SET_PTRACER 0x59616d61
44 #endif
45 
46 #ifndef PR_SET_NO_NEW_PRIVS
47 #define PR_SET_NO_NEW_PRIVS 38
48 #define PR_GET_NO_NEW_PRIVS 39
49 #endif
50 
51 #ifndef PR_SECCOMP_EXT
52 #define PR_SECCOMP_EXT 43
53 #endif
54 
55 #ifndef SECCOMP_EXT_ACT
56 #define SECCOMP_EXT_ACT 1
57 #endif
58 
59 #ifndef SECCOMP_EXT_ACT_TSYNC
60 #define SECCOMP_EXT_ACT_TSYNC 1
61 #endif
62 
63 #ifndef SECCOMP_MODE_STRICT
64 #define SECCOMP_MODE_STRICT 1
65 #endif
66 
67 #ifndef SECCOMP_MODE_FILTER
68 #define SECCOMP_MODE_FILTER 2
69 #endif
70 
71 #ifndef SECCOMP_RET_KILL
72 #define SECCOMP_RET_KILL        0x00000000U /* kill the task immediately */
73 #define SECCOMP_RET_TRAP        0x00030000U /* disallow and force a SIGSYS */
74 #define SECCOMP_RET_ERRNO       0x00050000U /* returns an errno */
75 #define SECCOMP_RET_TRACE       0x7ff00000U /* pass to a tracer or disallow */
76 #define SECCOMP_RET_ALLOW       0x7fff0000U /* allow */
77 
78 /* Masks for the return value sections. */
79 #define SECCOMP_RET_ACTION      0x7fff0000U
80 #define SECCOMP_RET_DATA        0x0000ffffU
81 
82 struct seccomp_data {
83 	int nr;
84 	__u32 arch;
85 	__u64 instruction_pointer;
86 	__u64 args[6];
87 };
88 #endif
89 
90 #if __BYTE_ORDER == __LITTLE_ENDIAN
91 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]))
92 #elif __BYTE_ORDER == __BIG_ENDIAN
93 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32))
94 #else
95 #error "wut? Unknown __BYTE_ORDER?!"
96 #endif
97 
98 #define SIBLING_EXIT_UNKILLED	0xbadbeef
99 #define SIBLING_EXIT_FAILURE	0xbadface
100 #define SIBLING_EXIT_NEWPRIVS	0xbadfeed
101 
TEST(mode_strict_support)102 TEST(mode_strict_support)
103 {
104 	long ret;
105 
106 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
107 	ASSERT_EQ(0, ret) {
108 		TH_LOG("Kernel does not support CONFIG_SECCOMP");
109 	}
110 	syscall(__NR_exit, 1);
111 }
112 
TEST_SIGNAL(mode_strict_cannot_call_prctl,SIGKILL)113 TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL)
114 {
115 	long ret;
116 
117 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
118 	ASSERT_EQ(0, ret) {
119 		TH_LOG("Kernel does not support CONFIG_SECCOMP");
120 	}
121 	syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
122 		NULL, NULL, NULL);
123 	EXPECT_FALSE(true) {
124 		TH_LOG("Unreachable!");
125 	}
126 }
127 
128 /* Note! This doesn't test no new privs behavior */
TEST(no_new_privs_support)129 TEST(no_new_privs_support)
130 {
131 	long ret;
132 
133 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
134 	EXPECT_EQ(0, ret) {
135 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
136 	}
137 }
138 
139 /* Tests kernel support by checking for a copy_from_user() fault on * NULL. */
TEST(mode_filter_support)140 TEST(mode_filter_support)
141 {
142 	long ret;
143 
144 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
145 	ASSERT_EQ(0, ret) {
146 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
147 	}
148 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL);
149 	EXPECT_EQ(-1, ret);
150 	EXPECT_EQ(EFAULT, errno) {
151 		TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!");
152 	}
153 }
154 
TEST(mode_filter_without_nnp)155 TEST(mode_filter_without_nnp)
156 {
157 	struct sock_filter filter[] = {
158 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
159 	};
160 	struct sock_fprog prog = {
161 		.len = (unsigned short)ARRAY_SIZE(filter),
162 		.filter = filter,
163 	};
164 	long ret;
165 
166 	ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0);
167 	ASSERT_LE(0, ret) {
168 		TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS");
169 	}
170 	errno = 0;
171 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
172 	/* Succeeds with CAP_SYS_ADMIN, fails without */
173 	/* TODO(wad) check caps not euid */
174 	if (geteuid()) {
175 		EXPECT_EQ(-1, ret);
176 		EXPECT_EQ(EACCES, errno);
177 	} else {
178 		EXPECT_EQ(0, ret);
179 	}
180 }
181 
182 #define MAX_INSNS_PER_PATH 32768
183 
TEST(filter_size_limits)184 TEST(filter_size_limits)
185 {
186 	int i;
187 	int count = BPF_MAXINSNS + 1;
188 	struct sock_filter allow[] = {
189 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
190 	};
191 	struct sock_filter *filter;
192 	struct sock_fprog prog = { };
193 	long ret;
194 
195 	filter = calloc(count, sizeof(*filter));
196 	ASSERT_NE(NULL, filter);
197 
198 	for (i = 0; i < count; i++)
199 		filter[i] = allow[0];
200 
201 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
202 	ASSERT_EQ(0, ret);
203 
204 	prog.filter = filter;
205 	prog.len = count;
206 
207 	/* Too many filter instructions in a single filter. */
208 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
209 	ASSERT_NE(0, ret) {
210 		TH_LOG("Installing %d insn filter was allowed", prog.len);
211 	}
212 
213 	/* One less is okay, though. */
214 	prog.len -= 1;
215 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
216 	ASSERT_EQ(0, ret) {
217 		TH_LOG("Installing %d insn filter wasn't allowed", prog.len);
218 	}
219 }
220 
TEST(filter_chain_limits)221 TEST(filter_chain_limits)
222 {
223 	int i;
224 	int count = BPF_MAXINSNS;
225 	struct sock_filter allow[] = {
226 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
227 	};
228 	struct sock_filter *filter;
229 	struct sock_fprog prog = { };
230 	long ret;
231 
232 	filter = calloc(count, sizeof(*filter));
233 	ASSERT_NE(NULL, filter);
234 
235 	for (i = 0; i < count; i++)
236 		filter[i] = allow[0];
237 
238 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
239 	ASSERT_EQ(0, ret);
240 
241 	prog.filter = filter;
242 	prog.len = 1;
243 
244 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
245 	ASSERT_EQ(0, ret);
246 
247 	prog.len = count;
248 
249 	/* Too many total filter instructions. */
250 	for (i = 0; i < MAX_INSNS_PER_PATH; i++) {
251 		ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
252 		if (ret != 0)
253 			break;
254 	}
255 	ASSERT_NE(0, ret) {
256 		TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)",
257 		       i, count, i * (count + 4));
258 	}
259 }
260 
TEST(mode_filter_cannot_move_to_strict)261 TEST(mode_filter_cannot_move_to_strict)
262 {
263 	struct sock_filter filter[] = {
264 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
265 	};
266 	struct sock_fprog prog = {
267 		.len = (unsigned short)ARRAY_SIZE(filter),
268 		.filter = filter,
269 	};
270 	long ret;
271 
272 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
273 	ASSERT_EQ(0, ret);
274 
275 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
276 	ASSERT_EQ(0, ret);
277 
278 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0);
279 	EXPECT_EQ(-1, ret);
280 	EXPECT_EQ(EINVAL, errno);
281 }
282 
283 
TEST(mode_filter_get_seccomp)284 TEST(mode_filter_get_seccomp)
285 {
286 	struct sock_filter filter[] = {
287 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
288 	};
289 	struct sock_fprog prog = {
290 		.len = (unsigned short)ARRAY_SIZE(filter),
291 		.filter = filter,
292 	};
293 	long ret;
294 
295 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
296 	ASSERT_EQ(0, ret);
297 
298 	ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
299 	EXPECT_EQ(0, ret);
300 
301 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
302 	ASSERT_EQ(0, ret);
303 
304 	ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
305 	EXPECT_EQ(2, ret);
306 }
307 
308 
TEST(ALLOW_all)309 TEST(ALLOW_all)
310 {
311 	struct sock_filter filter[] = {
312 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
313 	};
314 	struct sock_fprog prog = {
315 		.len = (unsigned short)ARRAY_SIZE(filter),
316 		.filter = filter,
317 	};
318 	long ret;
319 
320 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
321 	ASSERT_EQ(0, ret);
322 
323 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
324 	ASSERT_EQ(0, ret);
325 }
326 
TEST(empty_prog)327 TEST(empty_prog)
328 {
329 	struct sock_filter filter[] = {
330 	};
331 	struct sock_fprog prog = {
332 		.len = (unsigned short)ARRAY_SIZE(filter),
333 		.filter = filter,
334 	};
335 	long ret;
336 
337 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
338 	ASSERT_EQ(0, ret);
339 
340 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
341 	EXPECT_EQ(-1, ret);
342 	EXPECT_EQ(EINVAL, errno);
343 }
344 
TEST_SIGNAL(unknown_ret_is_kill_inside,SIGSYS)345 TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS)
346 {
347 	struct sock_filter filter[] = {
348 		BPF_STMT(BPF_RET|BPF_K, 0x10000000U),
349 	};
350 	struct sock_fprog prog = {
351 		.len = (unsigned short)ARRAY_SIZE(filter),
352 		.filter = filter,
353 	};
354 	long ret;
355 
356 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
357 	ASSERT_EQ(0, ret);
358 
359 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
360 	ASSERT_EQ(0, ret);
361 	EXPECT_EQ(0, syscall(__NR_getpid)) {
362 		TH_LOG("getpid() shouldn't ever return");
363 	}
364 }
365 
366 /* return code >= 0x80000000 is unused. */
TEST_SIGNAL(unknown_ret_is_kill_above_allow,SIGSYS)367 TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS)
368 {
369 	struct sock_filter filter[] = {
370 		BPF_STMT(BPF_RET|BPF_K, 0x90000000U),
371 	};
372 	struct sock_fprog prog = {
373 		.len = (unsigned short)ARRAY_SIZE(filter),
374 		.filter = filter,
375 	};
376 	long ret;
377 
378 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
379 	ASSERT_EQ(0, ret);
380 
381 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
382 	ASSERT_EQ(0, ret);
383 	EXPECT_EQ(0, syscall(__NR_getpid)) {
384 		TH_LOG("getpid() shouldn't ever return");
385 	}
386 }
387 
TEST_SIGNAL(KILL_all,SIGSYS)388 TEST_SIGNAL(KILL_all, SIGSYS)
389 {
390 	struct sock_filter filter[] = {
391 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
392 	};
393 	struct sock_fprog prog = {
394 		.len = (unsigned short)ARRAY_SIZE(filter),
395 		.filter = filter,
396 	};
397 	long ret;
398 
399 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
400 	ASSERT_EQ(0, ret);
401 
402 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
403 	ASSERT_EQ(0, ret);
404 }
405 
TEST_SIGNAL(KILL_one,SIGSYS)406 TEST_SIGNAL(KILL_one, SIGSYS)
407 {
408 	struct sock_filter filter[] = {
409 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
410 			offsetof(struct seccomp_data, nr)),
411 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
412 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
413 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
414 	};
415 	struct sock_fprog prog = {
416 		.len = (unsigned short)ARRAY_SIZE(filter),
417 		.filter = filter,
418 	};
419 	long ret;
420 	pid_t parent = getppid();
421 
422 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
423 	ASSERT_EQ(0, ret);
424 
425 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
426 	ASSERT_EQ(0, ret);
427 
428 	EXPECT_EQ(parent, syscall(__NR_getppid));
429 	/* getpid() should never return. */
430 	EXPECT_EQ(0, syscall(__NR_getpid));
431 }
432 
TEST_SIGNAL(KILL_one_arg_one,SIGSYS)433 TEST_SIGNAL(KILL_one_arg_one, SIGSYS)
434 {
435 	void *fatal_address;
436 	struct sock_filter filter[] = {
437 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
438 			offsetof(struct seccomp_data, nr)),
439 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0),
440 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
441 		/* Only both with lower 32-bit for now. */
442 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)),
443 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K,
444 			(unsigned long)&fatal_address, 0, 1),
445 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
446 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
447 	};
448 	struct sock_fprog prog = {
449 		.len = (unsigned short)ARRAY_SIZE(filter),
450 		.filter = filter,
451 	};
452 	long ret;
453 	pid_t parent = getppid();
454 	struct tms timebuf;
455 	clock_t clock = times(&timebuf);
456 
457 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
458 	ASSERT_EQ(0, ret);
459 
460 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
461 	ASSERT_EQ(0, ret);
462 
463 	EXPECT_EQ(parent, syscall(__NR_getppid));
464 	EXPECT_LE(clock, syscall(__NR_times, &timebuf));
465 	/* times() should never return. */
466 	EXPECT_EQ(0, syscall(__NR_times, &fatal_address));
467 }
468 
TEST_SIGNAL(KILL_one_arg_six,SIGSYS)469 TEST_SIGNAL(KILL_one_arg_six, SIGSYS)
470 {
471 #ifndef __NR_mmap2
472 	int sysno = __NR_mmap;
473 #else
474 	int sysno = __NR_mmap2;
475 #endif
476 	struct sock_filter filter[] = {
477 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
478 			offsetof(struct seccomp_data, nr)),
479 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0),
480 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
481 		/* Only both with lower 32-bit for now. */
482 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)),
483 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1),
484 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
485 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
486 	};
487 	struct sock_fprog prog = {
488 		.len = (unsigned short)ARRAY_SIZE(filter),
489 		.filter = filter,
490 	};
491 	long ret;
492 	pid_t parent = getppid();
493 	int fd;
494 	void *map1, *map2;
495 	int page_size = sysconf(_SC_PAGESIZE);
496 
497 	ASSERT_LT(0, page_size);
498 
499 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
500 	ASSERT_EQ(0, ret);
501 
502 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
503 	ASSERT_EQ(0, ret);
504 
505 	fd = open("/dev/zero", O_RDONLY);
506 	ASSERT_NE(-1, fd);
507 
508 	EXPECT_EQ(parent, syscall(__NR_getppid));
509 	map1 = (void *)syscall(sysno,
510 		NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size);
511 	EXPECT_NE(MAP_FAILED, map1);
512 	/* mmap2() should never return. */
513 	map2 = (void *)syscall(sysno,
514 		 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE);
515 	EXPECT_EQ(MAP_FAILED, map2);
516 
517 	/* The test failed, so clean up the resources. */
518 	munmap(map1, page_size);
519 	munmap(map2, page_size);
520 	close(fd);
521 }
522 
523 /* TODO(wad) add 64-bit versus 32-bit arg tests. */
TEST(arg_out_of_range)524 TEST(arg_out_of_range)
525 {
526 	struct sock_filter filter[] = {
527 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)),
528 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
529 	};
530 	struct sock_fprog prog = {
531 		.len = (unsigned short)ARRAY_SIZE(filter),
532 		.filter = filter,
533 	};
534 	long ret;
535 
536 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
537 	ASSERT_EQ(0, ret);
538 
539 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
540 	EXPECT_EQ(-1, ret);
541 	EXPECT_EQ(EINVAL, errno);
542 }
543 
TEST(ERRNO_valid)544 TEST(ERRNO_valid)
545 {
546 	struct sock_filter filter[] = {
547 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
548 			offsetof(struct seccomp_data, nr)),
549 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
550 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | E2BIG),
551 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
552 	};
553 	struct sock_fprog prog = {
554 		.len = (unsigned short)ARRAY_SIZE(filter),
555 		.filter = filter,
556 	};
557 	long ret;
558 	pid_t parent = getppid();
559 
560 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
561 	ASSERT_EQ(0, ret);
562 
563 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
564 	ASSERT_EQ(0, ret);
565 
566 	EXPECT_EQ(parent, syscall(__NR_getppid));
567 	EXPECT_EQ(-1, read(0, NULL, 0));
568 	EXPECT_EQ(E2BIG, errno);
569 }
570 
TEST(ERRNO_zero)571 TEST(ERRNO_zero)
572 {
573 	struct sock_filter filter[] = {
574 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
575 			offsetof(struct seccomp_data, nr)),
576 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
577 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | 0),
578 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
579 	};
580 	struct sock_fprog prog = {
581 		.len = (unsigned short)ARRAY_SIZE(filter),
582 		.filter = filter,
583 	};
584 	long ret;
585 	pid_t parent = getppid();
586 
587 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
588 	ASSERT_EQ(0, ret);
589 
590 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
591 	ASSERT_EQ(0, ret);
592 
593 	EXPECT_EQ(parent, syscall(__NR_getppid));
594 	/* "errno" of 0 is ok. */
595 	EXPECT_EQ(0, read(0, NULL, 0));
596 }
597 
TEST(ERRNO_capped)598 TEST(ERRNO_capped)
599 {
600 	struct sock_filter filter[] = {
601 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
602 			offsetof(struct seccomp_data, nr)),
603 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
604 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | 4096),
605 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
606 	};
607 	struct sock_fprog prog = {
608 		.len = (unsigned short)ARRAY_SIZE(filter),
609 		.filter = filter,
610 	};
611 	long ret;
612 	pid_t parent = getppid();
613 
614 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
615 	ASSERT_EQ(0, ret);
616 
617 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
618 	ASSERT_EQ(0, ret);
619 
620 	EXPECT_EQ(parent, syscall(__NR_getppid));
621 	EXPECT_EQ(-1, read(0, NULL, 0));
622 	EXPECT_EQ(4095, errno);
623 }
624 
FIXTURE_DATA(TRAP)625 FIXTURE_DATA(TRAP) {
626 	struct sock_fprog prog;
627 };
628 
FIXTURE_SETUP(TRAP)629 FIXTURE_SETUP(TRAP)
630 {
631 	struct sock_filter filter[] = {
632 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
633 			offsetof(struct seccomp_data, nr)),
634 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
635 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
636 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
637 	};
638 
639 	memset(&self->prog, 0, sizeof(self->prog));
640 	self->prog.filter = malloc(sizeof(filter));
641 	ASSERT_NE(NULL, self->prog.filter);
642 	memcpy(self->prog.filter, filter, sizeof(filter));
643 	self->prog.len = (unsigned short)ARRAY_SIZE(filter);
644 }
645 
FIXTURE_TEARDOWN(TRAP)646 FIXTURE_TEARDOWN(TRAP)
647 {
648 	if (self->prog.filter)
649 		free(self->prog.filter);
650 }
651 
TEST_F_SIGNAL(TRAP,dfl,SIGSYS)652 TEST_F_SIGNAL(TRAP, dfl, SIGSYS)
653 {
654 	long ret;
655 
656 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
657 	ASSERT_EQ(0, ret);
658 
659 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
660 	ASSERT_EQ(0, ret);
661 	syscall(__NR_getpid);
662 }
663 
664 /* Ensure that SIGSYS overrides SIG_IGN */
TEST_F_SIGNAL(TRAP,ign,SIGSYS)665 TEST_F_SIGNAL(TRAP, ign, SIGSYS)
666 {
667 	long ret;
668 
669 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
670 	ASSERT_EQ(0, ret);
671 
672 	signal(SIGSYS, SIG_IGN);
673 
674 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
675 	ASSERT_EQ(0, ret);
676 	syscall(__NR_getpid);
677 }
678 
679 static struct siginfo TRAP_info;
680 static volatile int TRAP_nr;
TRAP_action(int nr,siginfo_t * info,void * void_context)681 static void TRAP_action(int nr, siginfo_t *info, void *void_context)
682 {
683 	memcpy(&TRAP_info, info, sizeof(TRAP_info));
684 	TRAP_nr = nr;
685 }
686 
TEST_F(TRAP,handler)687 TEST_F(TRAP, handler)
688 {
689 	int ret, test;
690 	struct sigaction act;
691 	sigset_t mask;
692 
693 	memset(&act, 0, sizeof(act));
694 	sigemptyset(&mask);
695 	sigaddset(&mask, SIGSYS);
696 
697 	act.sa_sigaction = &TRAP_action;
698 	act.sa_flags = SA_SIGINFO;
699 	ret = sigaction(SIGSYS, &act, NULL);
700 	ASSERT_EQ(0, ret) {
701 		TH_LOG("sigaction failed");
702 	}
703 	ret = sigprocmask(SIG_UNBLOCK, &mask, NULL);
704 	ASSERT_EQ(0, ret) {
705 		TH_LOG("sigprocmask failed");
706 	}
707 
708 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
709 	ASSERT_EQ(0, ret);
710 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
711 	ASSERT_EQ(0, ret);
712 	TRAP_nr = 0;
713 	memset(&TRAP_info, 0, sizeof(TRAP_info));
714 	/* Expect the registers to be rolled back. (nr = error) may vary
715 	 * based on arch. */
716 	ret = syscall(__NR_getpid);
717 	/* Silence gcc warning about volatile. */
718 	test = TRAP_nr;
719 	EXPECT_EQ(SIGSYS, test);
720 	struct local_sigsys {
721 		void *_call_addr;	/* calling user insn */
722 		int _syscall;		/* triggering system call number */
723 		unsigned int _arch;	/* AUDIT_ARCH_* of syscall */
724 	} *sigsys = (struct local_sigsys *)
725 #ifdef si_syscall
726 		&(TRAP_info.si_call_addr);
727 #else
728 		&TRAP_info.si_pid;
729 #endif
730 	EXPECT_EQ(__NR_getpid, sigsys->_syscall);
731 	/* Make sure arch is non-zero. */
732 	EXPECT_NE(0, sigsys->_arch);
733 	EXPECT_NE(0, (unsigned long)sigsys->_call_addr);
734 }
735 
FIXTURE_DATA(precedence)736 FIXTURE_DATA(precedence) {
737 	struct sock_fprog allow;
738 	struct sock_fprog trace;
739 	struct sock_fprog error;
740 	struct sock_fprog trap;
741 	struct sock_fprog kill;
742 };
743 
FIXTURE_SETUP(precedence)744 FIXTURE_SETUP(precedence)
745 {
746 	struct sock_filter allow_insns[] = {
747 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
748 	};
749 	struct sock_filter trace_insns[] = {
750 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
751 			offsetof(struct seccomp_data, nr)),
752 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
753 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
754 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE),
755 	};
756 	struct sock_filter error_insns[] = {
757 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
758 			offsetof(struct seccomp_data, nr)),
759 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
760 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
761 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO),
762 	};
763 	struct sock_filter trap_insns[] = {
764 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
765 			offsetof(struct seccomp_data, nr)),
766 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
767 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
768 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
769 	};
770 	struct sock_filter kill_insns[] = {
771 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
772 			offsetof(struct seccomp_data, nr)),
773 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
774 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
775 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
776 	};
777 
778 	memset(self, 0, sizeof(*self));
779 #define FILTER_ALLOC(_x) \
780 	self->_x.filter = malloc(sizeof(_x##_insns)); \
781 	ASSERT_NE(NULL, self->_x.filter); \
782 	memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \
783 	self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns)
784 	FILTER_ALLOC(allow);
785 	FILTER_ALLOC(trace);
786 	FILTER_ALLOC(error);
787 	FILTER_ALLOC(trap);
788 	FILTER_ALLOC(kill);
789 }
790 
FIXTURE_TEARDOWN(precedence)791 FIXTURE_TEARDOWN(precedence)
792 {
793 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter)
794 	FILTER_FREE(allow);
795 	FILTER_FREE(trace);
796 	FILTER_FREE(error);
797 	FILTER_FREE(trap);
798 	FILTER_FREE(kill);
799 }
800 
TEST_F(precedence,allow_ok)801 TEST_F(precedence, allow_ok)
802 {
803 	pid_t parent, res = 0;
804 	long ret;
805 
806 	parent = getppid();
807 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
808 	ASSERT_EQ(0, ret);
809 
810 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
811 	ASSERT_EQ(0, ret);
812 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
813 	ASSERT_EQ(0, ret);
814 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
815 	ASSERT_EQ(0, ret);
816 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
817 	ASSERT_EQ(0, ret);
818 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
819 	ASSERT_EQ(0, ret);
820 	/* Should work just fine. */
821 	res = syscall(__NR_getppid);
822 	EXPECT_EQ(parent, res);
823 }
824 
TEST_F_SIGNAL(precedence,kill_is_highest,SIGSYS)825 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS)
826 {
827 	pid_t parent, res = 0;
828 	long ret;
829 
830 	parent = getppid();
831 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
832 	ASSERT_EQ(0, ret);
833 
834 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
835 	ASSERT_EQ(0, ret);
836 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
837 	ASSERT_EQ(0, ret);
838 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
839 	ASSERT_EQ(0, ret);
840 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
841 	ASSERT_EQ(0, ret);
842 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
843 	ASSERT_EQ(0, ret);
844 	/* Should work just fine. */
845 	res = syscall(__NR_getppid);
846 	EXPECT_EQ(parent, res);
847 	/* getpid() should never return. */
848 	res = syscall(__NR_getpid);
849 	EXPECT_EQ(0, res);
850 }
851 
TEST_F_SIGNAL(precedence,kill_is_highest_in_any_order,SIGSYS)852 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS)
853 {
854 	pid_t parent;
855 	long ret;
856 
857 	parent = getppid();
858 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
859 	ASSERT_EQ(0, ret);
860 
861 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
862 	ASSERT_EQ(0, ret);
863 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
864 	ASSERT_EQ(0, ret);
865 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
866 	ASSERT_EQ(0, ret);
867 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
868 	ASSERT_EQ(0, ret);
869 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
870 	ASSERT_EQ(0, ret);
871 	/* Should work just fine. */
872 	EXPECT_EQ(parent, syscall(__NR_getppid));
873 	/* getpid() should never return. */
874 	EXPECT_EQ(0, syscall(__NR_getpid));
875 }
876 
TEST_F_SIGNAL(precedence,trap_is_second,SIGSYS)877 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS)
878 {
879 	pid_t parent;
880 	long ret;
881 
882 	parent = getppid();
883 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
884 	ASSERT_EQ(0, ret);
885 
886 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
887 	ASSERT_EQ(0, ret);
888 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
889 	ASSERT_EQ(0, ret);
890 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
891 	ASSERT_EQ(0, ret);
892 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
893 	ASSERT_EQ(0, ret);
894 	/* Should work just fine. */
895 	EXPECT_EQ(parent, syscall(__NR_getppid));
896 	/* getpid() should never return. */
897 	EXPECT_EQ(0, syscall(__NR_getpid));
898 }
899 
TEST_F_SIGNAL(precedence,trap_is_second_in_any_order,SIGSYS)900 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS)
901 {
902 	pid_t parent;
903 	long ret;
904 
905 	parent = getppid();
906 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
907 	ASSERT_EQ(0, ret);
908 
909 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
910 	ASSERT_EQ(0, ret);
911 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
912 	ASSERT_EQ(0, ret);
913 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
914 	ASSERT_EQ(0, ret);
915 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
916 	ASSERT_EQ(0, ret);
917 	/* Should work just fine. */
918 	EXPECT_EQ(parent, syscall(__NR_getppid));
919 	/* getpid() should never return. */
920 	EXPECT_EQ(0, syscall(__NR_getpid));
921 }
922 
TEST_F(precedence,errno_is_third)923 TEST_F(precedence, errno_is_third)
924 {
925 	pid_t parent;
926 	long ret;
927 
928 	parent = getppid();
929 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
930 	ASSERT_EQ(0, ret);
931 
932 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
933 	ASSERT_EQ(0, ret);
934 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
935 	ASSERT_EQ(0, ret);
936 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
937 	ASSERT_EQ(0, ret);
938 	/* Should work just fine. */
939 	EXPECT_EQ(parent, syscall(__NR_getppid));
940 	EXPECT_EQ(0, syscall(__NR_getpid));
941 }
942 
TEST_F(precedence,errno_is_third_in_any_order)943 TEST_F(precedence, errno_is_third_in_any_order)
944 {
945 	pid_t parent;
946 	long ret;
947 
948 	parent = getppid();
949 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
950 	ASSERT_EQ(0, ret);
951 
952 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
953 	ASSERT_EQ(0, ret);
954 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
955 	ASSERT_EQ(0, ret);
956 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
957 	ASSERT_EQ(0, ret);
958 	/* Should work just fine. */
959 	EXPECT_EQ(parent, syscall(__NR_getppid));
960 	EXPECT_EQ(0, syscall(__NR_getpid));
961 }
962 
TEST_F(precedence,trace_is_fourth)963 TEST_F(precedence, trace_is_fourth)
964 {
965 	pid_t parent;
966 	long ret;
967 
968 	parent = getppid();
969 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
970 	ASSERT_EQ(0, ret);
971 
972 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
973 	ASSERT_EQ(0, ret);
974 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
975 	ASSERT_EQ(0, ret);
976 	/* Should work just fine. */
977 	EXPECT_EQ(parent, syscall(__NR_getppid));
978 	/* No ptracer */
979 	EXPECT_EQ(-1, syscall(__NR_getpid));
980 }
981 
TEST_F(precedence,trace_is_fourth_in_any_order)982 TEST_F(precedence, trace_is_fourth_in_any_order)
983 {
984 	pid_t parent;
985 	long ret;
986 
987 	parent = getppid();
988 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
989 	ASSERT_EQ(0, ret);
990 
991 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
992 	ASSERT_EQ(0, ret);
993 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
994 	ASSERT_EQ(0, ret);
995 	/* Should work just fine. */
996 	EXPECT_EQ(parent, syscall(__NR_getppid));
997 	/* No ptracer */
998 	EXPECT_EQ(-1, syscall(__NR_getpid));
999 }
1000 
1001 #ifndef PTRACE_O_TRACESECCOMP
1002 #define PTRACE_O_TRACESECCOMP	0x00000080
1003 #endif
1004 
1005 /* Catch the Ubuntu 12.04 value error. */
1006 #if PTRACE_EVENT_SECCOMP != 7
1007 #undef PTRACE_EVENT_SECCOMP
1008 #endif
1009 
1010 #ifndef PTRACE_EVENT_SECCOMP
1011 #define PTRACE_EVENT_SECCOMP 7
1012 #endif
1013 
1014 #define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP)
1015 bool tracer_running;
tracer_stop(int sig)1016 void tracer_stop(int sig)
1017 {
1018 	tracer_running = false;
1019 }
1020 
1021 typedef void tracer_func_t(struct __test_metadata *_metadata,
1022 			   pid_t tracee, int status, void *args);
1023 
tracer(struct __test_metadata * _metadata,int fd,pid_t tracee,tracer_func_t tracer_func,void * args)1024 void tracer(struct __test_metadata *_metadata, int fd, pid_t tracee,
1025 	    tracer_func_t tracer_func, void *args)
1026 {
1027 	int ret = -1;
1028 	struct sigaction action = {
1029 		.sa_handler = tracer_stop,
1030 	};
1031 
1032 	/* Allow external shutdown. */
1033 	tracer_running = true;
1034 	ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL));
1035 
1036 	errno = 0;
1037 	while (ret == -1 && errno != EINVAL)
1038 		ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0);
1039 	ASSERT_EQ(0, ret) {
1040 		kill(tracee, SIGKILL);
1041 	}
1042 	/* Wait for attach stop */
1043 	wait(NULL);
1044 
1045 	ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, PTRACE_O_TRACESECCOMP);
1046 	ASSERT_EQ(0, ret) {
1047 		TH_LOG("Failed to set PTRACE_O_TRACESECCOMP");
1048 		kill(tracee, SIGKILL);
1049 	}
1050 	ptrace(PTRACE_CONT, tracee, NULL, 0);
1051 
1052 	/* Unblock the tracee */
1053 	ASSERT_EQ(1, write(fd, "A", 1));
1054 	ASSERT_EQ(0, close(fd));
1055 
1056 	/* Run until we're shut down. Must assert to stop execution. */
1057 	while (tracer_running) {
1058 		int status;
1059 
1060 		if (wait(&status) != tracee)
1061 			continue;
1062 		if (WIFSIGNALED(status) || WIFEXITED(status))
1063 			/* Child is dead. Time to go. */
1064 			return;
1065 
1066 		/* Make sure this is a seccomp event. */
1067 		ASSERT_EQ(true, IS_SECCOMP_EVENT(status));
1068 
1069 		tracer_func(_metadata, tracee, status, args);
1070 
1071 		ret = ptrace(PTRACE_CONT, tracee, NULL, NULL);
1072 		ASSERT_EQ(0, ret);
1073 	}
1074 	/* Directly report the status of our test harness results. */
1075 	syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS : EXIT_FAILURE);
1076 }
1077 
1078 /* Common tracer setup/teardown functions. */
cont_handler(int num)1079 void cont_handler(int num)
1080 { }
setup_trace_fixture(struct __test_metadata * _metadata,tracer_func_t func,void * args)1081 pid_t setup_trace_fixture(struct __test_metadata *_metadata,
1082 			  tracer_func_t func, void *args)
1083 {
1084 	char sync;
1085 	int pipefd[2];
1086 	pid_t tracer_pid;
1087 	pid_t tracee = getpid();
1088 
1089 	/* Setup a pipe for clean synchronization. */
1090 	ASSERT_EQ(0, pipe(pipefd));
1091 
1092 	/* Fork a child which we'll promote to tracer */
1093 	tracer_pid = fork();
1094 	ASSERT_LE(0, tracer_pid);
1095 	signal(SIGALRM, cont_handler);
1096 	if (tracer_pid == 0) {
1097 		close(pipefd[0]);
1098 		tracer(_metadata, pipefd[1], tracee, func, args);
1099 		syscall(__NR_exit, 0);
1100 	}
1101 	close(pipefd[1]);
1102 	prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0);
1103 	read(pipefd[0], &sync, 1);
1104 	close(pipefd[0]);
1105 
1106 	return tracer_pid;
1107 }
teardown_trace_fixture(struct __test_metadata * _metadata,pid_t tracer)1108 void teardown_trace_fixture(struct __test_metadata *_metadata,
1109 			    pid_t tracer)
1110 {
1111 	if (tracer) {
1112 		int status;
1113 		/*
1114 		 * Extract the exit code from the other process and
1115 		 * adopt it for ourselves in case its asserts failed.
1116 		 */
1117 		ASSERT_EQ(0, kill(tracer, SIGUSR1));
1118 		ASSERT_EQ(tracer, waitpid(tracer, &status, 0));
1119 		if (WEXITSTATUS(status))
1120 			_metadata->passed = 0;
1121 	}
1122 }
1123 
1124 /* "poke" tracer arguments and function. */
1125 struct tracer_args_poke_t {
1126 	unsigned long poke_addr;
1127 };
1128 
tracer_poke(struct __test_metadata * _metadata,pid_t tracee,int status,void * args)1129 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status,
1130 		 void *args)
1131 {
1132 	int ret;
1133 	unsigned long msg;
1134 	struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args;
1135 
1136 	ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1137 	EXPECT_EQ(0, ret);
1138 	/* If this fails, don't try to recover. */
1139 	ASSERT_EQ(0x1001, msg) {
1140 		kill(tracee, SIGKILL);
1141 	}
1142 	/*
1143 	 * Poke in the message.
1144 	 * Registers are not touched to try to keep this relatively arch
1145 	 * agnostic.
1146 	 */
1147 	ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001);
1148 	EXPECT_EQ(0, ret);
1149 }
1150 
FIXTURE_DATA(TRACE_poke)1151 FIXTURE_DATA(TRACE_poke) {
1152 	struct sock_fprog prog;
1153 	pid_t tracer;
1154 	long poked;
1155 	struct tracer_args_poke_t tracer_args;
1156 };
1157 
FIXTURE_SETUP(TRACE_poke)1158 FIXTURE_SETUP(TRACE_poke)
1159 {
1160 	struct sock_filter filter[] = {
1161 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1162 			offsetof(struct seccomp_data, nr)),
1163 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1164 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001),
1165 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1166 	};
1167 
1168 	self->poked = 0;
1169 	memset(&self->prog, 0, sizeof(self->prog));
1170 	self->prog.filter = malloc(sizeof(filter));
1171 	ASSERT_NE(NULL, self->prog.filter);
1172 	memcpy(self->prog.filter, filter, sizeof(filter));
1173 	self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1174 
1175 	/* Set up tracer args. */
1176 	self->tracer_args.poke_addr = (unsigned long)&self->poked;
1177 
1178 	/* Launch tracer. */
1179 	self->tracer = setup_trace_fixture(_metadata, tracer_poke,
1180 					   &self->tracer_args);
1181 }
1182 
FIXTURE_TEARDOWN(TRACE_poke)1183 FIXTURE_TEARDOWN(TRACE_poke)
1184 {
1185 	teardown_trace_fixture(_metadata, self->tracer);
1186 	if (self->prog.filter)
1187 		free(self->prog.filter);
1188 }
1189 
TEST_F(TRACE_poke,read_has_side_effects)1190 TEST_F(TRACE_poke, read_has_side_effects)
1191 {
1192 	ssize_t ret;
1193 
1194 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1195 	ASSERT_EQ(0, ret);
1196 
1197 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1198 	ASSERT_EQ(0, ret);
1199 
1200 	EXPECT_EQ(0, self->poked);
1201 	ret = read(-1, NULL, 0);
1202 	EXPECT_EQ(-1, ret);
1203 	EXPECT_EQ(0x1001, self->poked);
1204 }
1205 
TEST_F(TRACE_poke,getpid_runs_normally)1206 TEST_F(TRACE_poke, getpid_runs_normally)
1207 {
1208 	long ret;
1209 
1210 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1211 	ASSERT_EQ(0, ret);
1212 
1213 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1214 	ASSERT_EQ(0, ret);
1215 
1216 	EXPECT_EQ(0, self->poked);
1217 	EXPECT_NE(0, syscall(__NR_getpid));
1218 	EXPECT_EQ(0, self->poked);
1219 }
1220 
1221 #if defined(__x86_64__)
1222 # define ARCH_REGS	struct user_regs_struct
1223 # define SYSCALL_NUM	orig_rax
1224 # define SYSCALL_RET	rax
1225 #elif defined(__i386__)
1226 # define ARCH_REGS	struct user_regs_struct
1227 # define SYSCALL_NUM	orig_eax
1228 # define SYSCALL_RET	eax
1229 #elif defined(__arm__)
1230 # define ARCH_REGS	struct pt_regs
1231 # define SYSCALL_NUM	ARM_r7
1232 # define SYSCALL_RET	ARM_r0
1233 #elif defined(__aarch64__)
1234 # define ARCH_REGS	struct user_pt_regs
1235 # define SYSCALL_NUM	regs[8]
1236 # define SYSCALL_RET	regs[0]
1237 #elif defined(__powerpc__)
1238 # define ARCH_REGS	struct pt_regs
1239 # define SYSCALL_NUM	gpr[0]
1240 # define SYSCALL_RET	gpr[3]
1241 #elif defined(__s390__)
1242 # define ARCH_REGS     s390_regs
1243 # define SYSCALL_NUM   gprs[2]
1244 # define SYSCALL_RET   gprs[2]
1245 #else
1246 # error "Do not know how to find your architecture's registers and syscalls"
1247 #endif
1248 
1249 /* Architecture-specific syscall fetching routine. */
get_syscall(struct __test_metadata * _metadata,pid_t tracee)1250 int get_syscall(struct __test_metadata *_metadata, pid_t tracee)
1251 {
1252 	struct iovec iov;
1253 	ARCH_REGS regs;
1254 
1255 	iov.iov_base = &regs;
1256 	iov.iov_len = sizeof(regs);
1257 	EXPECT_EQ(0, ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov)) {
1258 		TH_LOG("PTRACE_GETREGSET failed");
1259 		return -1;
1260 	}
1261 
1262 	return regs.SYSCALL_NUM;
1263 }
1264 
1265 /* Architecture-specific syscall changing routine. */
change_syscall(struct __test_metadata * _metadata,pid_t tracee,int syscall)1266 void change_syscall(struct __test_metadata *_metadata,
1267 		    pid_t tracee, int syscall)
1268 {
1269 	struct iovec iov;
1270 	int ret;
1271 	ARCH_REGS regs;
1272 
1273 	iov.iov_base = &regs;
1274 	iov.iov_len = sizeof(regs);
1275 	ret = ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov);
1276 	EXPECT_EQ(0, ret);
1277 
1278 #if defined(__x86_64__) || defined(__i386__) || defined(__powerpc__) || \
1279     defined(__s390__)
1280 	{
1281 		regs.SYSCALL_NUM = syscall;
1282 	}
1283 
1284 #elif defined(__arm__)
1285 # ifndef PTRACE_SET_SYSCALL
1286 #  define PTRACE_SET_SYSCALL   23
1287 # endif
1288 	{
1289 		ret = ptrace(PTRACE_SET_SYSCALL, tracee, NULL, syscall);
1290 		EXPECT_EQ(0, ret);
1291 	}
1292 
1293 #elif defined(__aarch64__)
1294 # ifndef NT_ARM_SYSTEM_CALL
1295 #  define NT_ARM_SYSTEM_CALL 0x404
1296 # endif
1297 	{
1298 		iov.iov_base = &syscall;
1299 		iov.iov_len = sizeof(syscall);
1300 		ret = ptrace(PTRACE_SETREGSET, tracee, NT_ARM_SYSTEM_CALL,
1301 			     &iov);
1302 		EXPECT_EQ(0, ret);
1303 	}
1304 
1305 #else
1306 	ASSERT_EQ(1, 0) {
1307 		TH_LOG("How is the syscall changed on this architecture?");
1308 	}
1309 #endif
1310 
1311 	/* If syscall is skipped, change return value. */
1312 	if (syscall == -1)
1313 		regs.SYSCALL_RET = 1;
1314 
1315 	iov.iov_base = &regs;
1316 	iov.iov_len = sizeof(regs);
1317 	ret = ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &iov);
1318 	EXPECT_EQ(0, ret);
1319 }
1320 
tracer_syscall(struct __test_metadata * _metadata,pid_t tracee,int status,void * args)1321 void tracer_syscall(struct __test_metadata *_metadata, pid_t tracee,
1322 		    int status, void *args)
1323 {
1324 	int ret;
1325 	unsigned long msg;
1326 
1327 	/* Make sure we got the right message. */
1328 	ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1329 	EXPECT_EQ(0, ret);
1330 
1331 	/* Validate and take action on expected syscalls. */
1332 	switch (msg) {
1333 	case 0x1002:
1334 		/* change getpid to getppid. */
1335 		EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee));
1336 		change_syscall(_metadata, tracee, __NR_getppid);
1337 		break;
1338 	case 0x1003:
1339 		/* skip gettid. */
1340 		EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee));
1341 		change_syscall(_metadata, tracee, -1);
1342 		break;
1343 	case 0x1004:
1344 		/* do nothing (allow getppid) */
1345 		EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee));
1346 		break;
1347 	default:
1348 		EXPECT_EQ(0, msg) {
1349 			TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg);
1350 			kill(tracee, SIGKILL);
1351 		}
1352 	}
1353 
1354 }
1355 
FIXTURE_DATA(TRACE_syscall)1356 FIXTURE_DATA(TRACE_syscall) {
1357 	struct sock_fprog prog;
1358 	pid_t tracer, mytid, mypid, parent;
1359 };
1360 
FIXTURE_SETUP(TRACE_syscall)1361 FIXTURE_SETUP(TRACE_syscall)
1362 {
1363 	struct sock_filter filter[] = {
1364 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1365 			offsetof(struct seccomp_data, nr)),
1366 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
1367 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002),
1368 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1),
1369 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003),
1370 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1371 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004),
1372 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1373 	};
1374 
1375 	memset(&self->prog, 0, sizeof(self->prog));
1376 	self->prog.filter = malloc(sizeof(filter));
1377 	ASSERT_NE(NULL, self->prog.filter);
1378 	memcpy(self->prog.filter, filter, sizeof(filter));
1379 	self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1380 
1381 	/* Prepare some testable syscall results. */
1382 	self->mytid = syscall(__NR_gettid);
1383 	ASSERT_GT(self->mytid, 0);
1384 	ASSERT_NE(self->mytid, 1) {
1385 		TH_LOG("Running this test as init is not supported. :)");
1386 	}
1387 
1388 	self->mypid = getpid();
1389 	ASSERT_GT(self->mypid, 0);
1390 	ASSERT_EQ(self->mytid, self->mypid);
1391 
1392 	self->parent = getppid();
1393 	ASSERT_GT(self->parent, 0);
1394 	ASSERT_NE(self->parent, self->mypid);
1395 
1396 	/* Launch tracer. */
1397 	self->tracer = setup_trace_fixture(_metadata, tracer_syscall, NULL);
1398 }
1399 
FIXTURE_TEARDOWN(TRACE_syscall)1400 FIXTURE_TEARDOWN(TRACE_syscall)
1401 {
1402 	teardown_trace_fixture(_metadata, self->tracer);
1403 	if (self->prog.filter)
1404 		free(self->prog.filter);
1405 }
1406 
TEST_F(TRACE_syscall,syscall_allowed)1407 TEST_F(TRACE_syscall, syscall_allowed)
1408 {
1409 	long ret;
1410 
1411 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1412 	ASSERT_EQ(0, ret);
1413 
1414 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1415 	ASSERT_EQ(0, ret);
1416 
1417 	/* getppid works as expected (no changes). */
1418 	EXPECT_EQ(self->parent, syscall(__NR_getppid));
1419 	EXPECT_NE(self->mypid, syscall(__NR_getppid));
1420 }
1421 
TEST_F(TRACE_syscall,syscall_redirected)1422 TEST_F(TRACE_syscall, syscall_redirected)
1423 {
1424 	long ret;
1425 
1426 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1427 	ASSERT_EQ(0, ret);
1428 
1429 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1430 	ASSERT_EQ(0, ret);
1431 
1432 	/* getpid has been redirected to getppid as expected. */
1433 	EXPECT_EQ(self->parent, syscall(__NR_getpid));
1434 	EXPECT_NE(self->mypid, syscall(__NR_getpid));
1435 }
1436 
TEST_F(TRACE_syscall,syscall_dropped)1437 TEST_F(TRACE_syscall, syscall_dropped)
1438 {
1439 	long ret;
1440 
1441 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1442 	ASSERT_EQ(0, ret);
1443 
1444 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1445 	ASSERT_EQ(0, ret);
1446 
1447 	/* gettid has been skipped and an altered return value stored. */
1448 	EXPECT_EQ(1, syscall(__NR_gettid));
1449 	EXPECT_NE(self->mytid, syscall(__NR_gettid));
1450 }
1451 
1452 #ifndef __NR_seccomp
1453 # if defined(__i386__)
1454 #  define __NR_seccomp 354
1455 # elif defined(__x86_64__)
1456 #  define __NR_seccomp 317
1457 # elif defined(__arm__)
1458 #  define __NR_seccomp 383
1459 # elif defined(__aarch64__)
1460 #  define __NR_seccomp 277
1461 # elif defined(__powerpc__)
1462 #  define __NR_seccomp 358
1463 # elif defined(__s390__)
1464 #  define __NR_seccomp 348
1465 # else
1466 #  warning "seccomp syscall number unknown for this architecture"
1467 #  define __NR_seccomp 0xffff
1468 # endif
1469 #endif
1470 
1471 #ifndef SECCOMP_SET_MODE_STRICT
1472 #define SECCOMP_SET_MODE_STRICT 0
1473 #endif
1474 
1475 #ifndef SECCOMP_SET_MODE_FILTER
1476 #define SECCOMP_SET_MODE_FILTER 1
1477 #endif
1478 
1479 #ifndef SECCOMP_FILTER_FLAG_TSYNC
1480 #define SECCOMP_FILTER_FLAG_TSYNC (1UL << 0)
1481 #endif
1482 
1483 #ifndef SECCOMP_FILTER_FLAG_SPEC_ALLOW
1484 #define SECCOMP_FILTER_FLAG_SPEC_ALLOW (1UL << 2)
1485 #endif
1486 
1487 #ifndef seccomp
seccomp(unsigned int op,unsigned int flags,void * args)1488 int seccomp(unsigned int op, unsigned int flags, void *args)
1489 {
1490 	errno = 0;
1491 	return syscall(__NR_seccomp, op, flags, args);
1492 }
1493 #endif
1494 
TEST(seccomp_syscall)1495 TEST(seccomp_syscall)
1496 {
1497 	struct sock_filter filter[] = {
1498 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1499 	};
1500 	struct sock_fprog prog = {
1501 		.len = (unsigned short)ARRAY_SIZE(filter),
1502 		.filter = filter,
1503 	};
1504 	long ret;
1505 
1506 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1507 	ASSERT_EQ(0, ret) {
1508 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1509 	}
1510 
1511 	/* Reject insane operation. */
1512 	ret = seccomp(-1, 0, &prog);
1513 	ASSERT_NE(ENOSYS, errno) {
1514 		TH_LOG("Kernel does not support seccomp syscall!");
1515 	}
1516 	EXPECT_EQ(EINVAL, errno) {
1517 		TH_LOG("Did not reject crazy op value!");
1518 	}
1519 
1520 	/* Reject strict with flags or pointer. */
1521 	ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL);
1522 	EXPECT_EQ(EINVAL, errno) {
1523 		TH_LOG("Did not reject mode strict with flags!");
1524 	}
1525 	ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog);
1526 	EXPECT_EQ(EINVAL, errno) {
1527 		TH_LOG("Did not reject mode strict with uargs!");
1528 	}
1529 
1530 	/* Reject insane args for filter. */
1531 	ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog);
1532 	EXPECT_EQ(EINVAL, errno) {
1533 		TH_LOG("Did not reject crazy filter flags!");
1534 	}
1535 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL);
1536 	EXPECT_EQ(EFAULT, errno) {
1537 		TH_LOG("Did not reject NULL filter!");
1538 	}
1539 
1540 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
1541 	EXPECT_EQ(0, errno) {
1542 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s",
1543 			strerror(errno));
1544 	}
1545 }
1546 
TEST(seccomp_syscall_mode_lock)1547 TEST(seccomp_syscall_mode_lock)
1548 {
1549 	struct sock_filter filter[] = {
1550 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1551 	};
1552 	struct sock_fprog prog = {
1553 		.len = (unsigned short)ARRAY_SIZE(filter),
1554 		.filter = filter,
1555 	};
1556 	long ret;
1557 
1558 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
1559 	ASSERT_EQ(0, ret) {
1560 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1561 	}
1562 
1563 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
1564 	ASSERT_NE(ENOSYS, errno) {
1565 		TH_LOG("Kernel does not support seccomp syscall!");
1566 	}
1567 	EXPECT_EQ(0, ret) {
1568 		TH_LOG("Could not install filter!");
1569 	}
1570 
1571 	/* Make sure neither entry point will switch to strict. */
1572 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0);
1573 	EXPECT_EQ(EINVAL, errno) {
1574 		TH_LOG("Switched to mode strict!");
1575 	}
1576 
1577 	ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL);
1578 	EXPECT_EQ(EINVAL, errno) {
1579 		TH_LOG("Switched to mode strict!");
1580 	}
1581 }
1582 
1583 /*
1584  * Test detection of known and unknown filter flags. Userspace needs to be able
1585  * to check if a filter flag is supported by the current kernel and a good way
1586  * of doing that is by attempting to enter filter mode, with the flag bit in
1587  * question set, and a NULL pointer for the _args_ parameter. EFAULT indicates
1588  * that the flag is valid and EINVAL indicates that the flag is invalid.
1589  */
TEST(detect_seccomp_filter_flags)1590 TEST(detect_seccomp_filter_flags)
1591 {
1592 	unsigned int flags[] = { SECCOMP_FILTER_FLAG_TSYNC,
1593 				 SECCOMP_FILTER_FLAG_SPEC_ALLOW };
1594 	unsigned int flag, all_flags;
1595 	int i;
1596 	long ret;
1597 
1598 	/* Test detection of known-good filter flags */
1599 	for (i = 0, all_flags = 0; i < ARRAY_SIZE(flags); i++) {
1600 		int bits = 0;
1601 
1602 		flag = flags[i];
1603 		/* Make sure the flag is a single bit! */
1604 		while (flag) {
1605 			if (flag & 0x1)
1606 				bits ++;
1607 			flag >>= 1;
1608 		}
1609 		ASSERT_EQ(1, bits);
1610 		flag = flags[i];
1611 
1612 		ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
1613 		ASSERT_NE(ENOSYS, errno) {
1614 			TH_LOG("Kernel does not support seccomp syscall!");
1615 		}
1616 		EXPECT_EQ(-1, ret);
1617 		EXPECT_EQ(EFAULT, errno) {
1618 			TH_LOG("Failed to detect that a known-good filter flag (0x%X) is supported!",
1619 			       flag);
1620 		}
1621 
1622 		all_flags |= flag;
1623 	}
1624 
1625 	/* Test detection of all known-good filter flags */
1626 	ret = seccomp(SECCOMP_SET_MODE_FILTER, all_flags, NULL);
1627 	EXPECT_EQ(-1, ret);
1628 	EXPECT_EQ(EFAULT, errno) {
1629 		TH_LOG("Failed to detect that all known-good filter flags (0x%X) are supported!",
1630 		       all_flags);
1631 	}
1632 
1633 	/* Test detection of an unknown filter flag */
1634 	flag = -1;
1635 	ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
1636 	EXPECT_EQ(-1, ret);
1637 	EXPECT_EQ(EINVAL, errno) {
1638 		TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported!",
1639 		       flag);
1640 	}
1641 
1642 	/*
1643 	 * Test detection of an unknown filter flag that may simply need to be
1644 	 * added to this test
1645 	 */
1646 	flag = flags[ARRAY_SIZE(flags) - 1] << 1;
1647 	ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
1648 	EXPECT_EQ(-1, ret);
1649 	EXPECT_EQ(EINVAL, errno) {
1650 		TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported! Does a new flag need to be added to this test?",
1651 		       flag);
1652 	}
1653 }
1654 
TEST(TSYNC_first)1655 TEST(TSYNC_first)
1656 {
1657 	struct sock_filter filter[] = {
1658 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1659 	};
1660 	struct sock_fprog prog = {
1661 		.len = (unsigned short)ARRAY_SIZE(filter),
1662 		.filter = filter,
1663 	};
1664 	long ret;
1665 
1666 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
1667 	ASSERT_EQ(0, ret) {
1668 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1669 	}
1670 
1671 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
1672 		      &prog);
1673 	ASSERT_NE(ENOSYS, errno) {
1674 		TH_LOG("Kernel does not support seccomp syscall!");
1675 	}
1676 	EXPECT_EQ(0, ret) {
1677 		TH_LOG("Could not install initial filter with TSYNC!");
1678 	}
1679 }
1680 
1681 #define TSYNC_SIBLINGS 2
1682 struct tsync_sibling {
1683 	pthread_t tid;
1684 	pid_t system_tid;
1685 	sem_t *started;
1686 	pthread_cond_t *cond;
1687 	pthread_mutex_t *mutex;
1688 	int diverge;
1689 	int num_waits;
1690 	struct sock_fprog *prog;
1691 	struct __test_metadata *metadata;
1692 };
1693 
FIXTURE_DATA(TSYNC)1694 FIXTURE_DATA(TSYNC) {
1695 	struct sock_fprog root_prog, apply_prog;
1696 	struct tsync_sibling sibling[TSYNC_SIBLINGS];
1697 	sem_t started;
1698 	pthread_cond_t cond;
1699 	pthread_mutex_t mutex;
1700 	int sibling_count;
1701 };
1702 
FIXTURE_SETUP(TSYNC)1703 FIXTURE_SETUP(TSYNC)
1704 {
1705 	struct sock_filter root_filter[] = {
1706 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1707 	};
1708 	struct sock_filter apply_filter[] = {
1709 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1710 			offsetof(struct seccomp_data, nr)),
1711 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1712 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1713 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1714 	};
1715 
1716 	memset(&self->root_prog, 0, sizeof(self->root_prog));
1717 	memset(&self->apply_prog, 0, sizeof(self->apply_prog));
1718 	memset(&self->sibling, 0, sizeof(self->sibling));
1719 	self->root_prog.filter = malloc(sizeof(root_filter));
1720 	ASSERT_NE(NULL, self->root_prog.filter);
1721 	memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter));
1722 	self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter);
1723 
1724 	self->apply_prog.filter = malloc(sizeof(apply_filter));
1725 	ASSERT_NE(NULL, self->apply_prog.filter);
1726 	memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter));
1727 	self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter);
1728 
1729 	self->sibling_count = 0;
1730 	pthread_mutex_init(&self->mutex, NULL);
1731 	pthread_cond_init(&self->cond, NULL);
1732 	sem_init(&self->started, 0, 0);
1733 	self->sibling[0].tid = 0;
1734 	self->sibling[0].cond = &self->cond;
1735 	self->sibling[0].started = &self->started;
1736 	self->sibling[0].mutex = &self->mutex;
1737 	self->sibling[0].diverge = 0;
1738 	self->sibling[0].num_waits = 1;
1739 	self->sibling[0].prog = &self->root_prog;
1740 	self->sibling[0].metadata = _metadata;
1741 	self->sibling[1].tid = 0;
1742 	self->sibling[1].cond = &self->cond;
1743 	self->sibling[1].started = &self->started;
1744 	self->sibling[1].mutex = &self->mutex;
1745 	self->sibling[1].diverge = 0;
1746 	self->sibling[1].prog = &self->root_prog;
1747 	self->sibling[1].num_waits = 1;
1748 	self->sibling[1].metadata = _metadata;
1749 }
1750 
FIXTURE_TEARDOWN(TSYNC)1751 FIXTURE_TEARDOWN(TSYNC)
1752 {
1753 	int sib = 0;
1754 
1755 	if (self->root_prog.filter)
1756 		free(self->root_prog.filter);
1757 	if (self->apply_prog.filter)
1758 		free(self->apply_prog.filter);
1759 
1760 	for ( ; sib < self->sibling_count; ++sib) {
1761 		struct tsync_sibling *s = &self->sibling[sib];
1762 		void *status;
1763 
1764 		if (!s->tid)
1765 			continue;
1766 		if (pthread_kill(s->tid, 0)) {
1767 			pthread_cancel(s->tid);
1768 			pthread_join(s->tid, &status);
1769 		}
1770 	}
1771 	pthread_mutex_destroy(&self->mutex);
1772 	pthread_cond_destroy(&self->cond);
1773 	sem_destroy(&self->started);
1774 }
1775 
tsync_sibling(void * data)1776 void *tsync_sibling(void *data)
1777 {
1778 	long ret = 0;
1779 	struct tsync_sibling *me = data;
1780 
1781 	me->system_tid = syscall(__NR_gettid);
1782 
1783 	pthread_mutex_lock(me->mutex);
1784 	if (me->diverge) {
1785 		/* Just re-apply the root prog to fork the tree */
1786 		ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
1787 				me->prog, 0, 0);
1788 	}
1789 	sem_post(me->started);
1790 	/* Return outside of started so parent notices failures. */
1791 	if (ret) {
1792 		pthread_mutex_unlock(me->mutex);
1793 		return (void *)SIBLING_EXIT_FAILURE;
1794 	}
1795 	do {
1796 		pthread_cond_wait(me->cond, me->mutex);
1797 		me->num_waits = me->num_waits - 1;
1798 	} while (me->num_waits);
1799 	pthread_mutex_unlock(me->mutex);
1800 
1801 	ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0);
1802 	if (!ret)
1803 		return (void *)SIBLING_EXIT_NEWPRIVS;
1804 	read(0, NULL, 0);
1805 	return (void *)SIBLING_EXIT_UNKILLED;
1806 }
1807 
tsync_start_sibling(struct tsync_sibling * sibling)1808 void tsync_start_sibling(struct tsync_sibling *sibling)
1809 {
1810 	pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling);
1811 }
1812 
TEST_F(TSYNC,siblings_fail_prctl)1813 TEST_F(TSYNC, siblings_fail_prctl)
1814 {
1815 	long ret;
1816 	void *status;
1817 	struct sock_filter filter[] = {
1818 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1819 			offsetof(struct seccomp_data, nr)),
1820 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
1821 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL),
1822 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1823 	};
1824 	struct sock_fprog prog = {
1825 		.len = (unsigned short)ARRAY_SIZE(filter),
1826 		.filter = filter,
1827 	};
1828 
1829 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
1830 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1831 	}
1832 
1833 	/* Check prctl failure detection by requesting sib 0 diverge. */
1834 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
1835 	ASSERT_NE(ENOSYS, errno) {
1836 		TH_LOG("Kernel does not support seccomp syscall!");
1837 	}
1838 	ASSERT_EQ(0, ret) {
1839 		TH_LOG("setting filter failed");
1840 	}
1841 
1842 	self->sibling[0].diverge = 1;
1843 	tsync_start_sibling(&self->sibling[0]);
1844 	tsync_start_sibling(&self->sibling[1]);
1845 
1846 	while (self->sibling_count < TSYNC_SIBLINGS) {
1847 		sem_wait(&self->started);
1848 		self->sibling_count++;
1849 	}
1850 
1851 	/* Signal the threads to clean up*/
1852 	pthread_mutex_lock(&self->mutex);
1853 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
1854 		TH_LOG("cond broadcast non-zero");
1855 	}
1856 	pthread_mutex_unlock(&self->mutex);
1857 
1858 	/* Ensure diverging sibling failed to call prctl. */
1859 	pthread_join(self->sibling[0].tid, &status);
1860 	EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status);
1861 	pthread_join(self->sibling[1].tid, &status);
1862 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
1863 }
1864 
TEST_F(TSYNC,two_siblings_with_ancestor)1865 TEST_F(TSYNC, two_siblings_with_ancestor)
1866 {
1867 	long ret;
1868 	void *status;
1869 
1870 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
1871 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1872 	}
1873 
1874 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
1875 	ASSERT_NE(ENOSYS, errno) {
1876 		TH_LOG("Kernel does not support seccomp syscall!");
1877 	}
1878 	ASSERT_EQ(0, ret) {
1879 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
1880 	}
1881 	tsync_start_sibling(&self->sibling[0]);
1882 	tsync_start_sibling(&self->sibling[1]);
1883 
1884 	while (self->sibling_count < TSYNC_SIBLINGS) {
1885 		sem_wait(&self->started);
1886 		self->sibling_count++;
1887 	}
1888 
1889 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
1890 		      &self->apply_prog);
1891 	ASSERT_EQ(0, ret) {
1892 		TH_LOG("Could install filter on all threads!");
1893 	}
1894 	/* Tell the siblings to test the policy */
1895 	pthread_mutex_lock(&self->mutex);
1896 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
1897 		TH_LOG("cond broadcast non-zero");
1898 	}
1899 	pthread_mutex_unlock(&self->mutex);
1900 	/* Ensure they are both killed and don't exit cleanly. */
1901 	pthread_join(self->sibling[0].tid, &status);
1902 	EXPECT_EQ(0x0, (long)status);
1903 	pthread_join(self->sibling[1].tid, &status);
1904 	EXPECT_EQ(0x0, (long)status);
1905 }
1906 
TEST_F(TSYNC,two_sibling_want_nnp)1907 TEST_F(TSYNC, two_sibling_want_nnp)
1908 {
1909 	void *status;
1910 
1911 	/* start siblings before any prctl() operations */
1912 	tsync_start_sibling(&self->sibling[0]);
1913 	tsync_start_sibling(&self->sibling[1]);
1914 	while (self->sibling_count < TSYNC_SIBLINGS) {
1915 		sem_wait(&self->started);
1916 		self->sibling_count++;
1917 	}
1918 
1919 	/* Tell the siblings to test no policy */
1920 	pthread_mutex_lock(&self->mutex);
1921 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
1922 		TH_LOG("cond broadcast non-zero");
1923 	}
1924 	pthread_mutex_unlock(&self->mutex);
1925 
1926 	/* Ensure they are both upset about lacking nnp. */
1927 	pthread_join(self->sibling[0].tid, &status);
1928 	EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
1929 	pthread_join(self->sibling[1].tid, &status);
1930 	EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
1931 }
1932 
TEST_F(TSYNC,two_siblings_with_no_filter)1933 TEST_F(TSYNC, two_siblings_with_no_filter)
1934 {
1935 	long ret;
1936 	void *status;
1937 
1938 	/* start siblings before any prctl() operations */
1939 	tsync_start_sibling(&self->sibling[0]);
1940 	tsync_start_sibling(&self->sibling[1]);
1941 	while (self->sibling_count < TSYNC_SIBLINGS) {
1942 		sem_wait(&self->started);
1943 		self->sibling_count++;
1944 	}
1945 
1946 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
1947 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1948 	}
1949 
1950 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
1951 		      &self->apply_prog);
1952 	ASSERT_NE(ENOSYS, errno) {
1953 		TH_LOG("Kernel does not support seccomp syscall!");
1954 	}
1955 	ASSERT_EQ(0, ret) {
1956 		TH_LOG("Could install filter on all threads!");
1957 	}
1958 
1959 	/* Tell the siblings to test the policy */
1960 	pthread_mutex_lock(&self->mutex);
1961 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
1962 		TH_LOG("cond broadcast non-zero");
1963 	}
1964 	pthread_mutex_unlock(&self->mutex);
1965 
1966 	/* Ensure they are both killed and don't exit cleanly. */
1967 	pthread_join(self->sibling[0].tid, &status);
1968 	EXPECT_EQ(0x0, (long)status);
1969 	pthread_join(self->sibling[1].tid, &status);
1970 	EXPECT_EQ(0x0, (long)status);
1971 }
1972 
TEST_F(TSYNC,two_siblings_with_one_divergence)1973 TEST_F(TSYNC, two_siblings_with_one_divergence)
1974 {
1975 	long ret;
1976 	void *status;
1977 
1978 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
1979 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1980 	}
1981 
1982 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
1983 	ASSERT_NE(ENOSYS, errno) {
1984 		TH_LOG("Kernel does not support seccomp syscall!");
1985 	}
1986 	ASSERT_EQ(0, ret) {
1987 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
1988 	}
1989 	self->sibling[0].diverge = 1;
1990 	tsync_start_sibling(&self->sibling[0]);
1991 	tsync_start_sibling(&self->sibling[1]);
1992 
1993 	while (self->sibling_count < TSYNC_SIBLINGS) {
1994 		sem_wait(&self->started);
1995 		self->sibling_count++;
1996 	}
1997 
1998 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
1999 		      &self->apply_prog);
2000 	ASSERT_EQ(self->sibling[0].system_tid, ret) {
2001 		TH_LOG("Did not fail on diverged sibling.");
2002 	}
2003 
2004 	/* Wake the threads */
2005 	pthread_mutex_lock(&self->mutex);
2006 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2007 		TH_LOG("cond broadcast non-zero");
2008 	}
2009 	pthread_mutex_unlock(&self->mutex);
2010 
2011 	/* Ensure they are both unkilled. */
2012 	pthread_join(self->sibling[0].tid, &status);
2013 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2014 	pthread_join(self->sibling[1].tid, &status);
2015 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2016 }
2017 
TEST_F(TSYNC,two_siblings_not_under_filter)2018 TEST_F(TSYNC, two_siblings_not_under_filter)
2019 {
2020 	long ret, sib;
2021 	void *status;
2022 
2023 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2024 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2025 	}
2026 
2027 	/*
2028 	 * Sibling 0 will have its own seccomp policy
2029 	 * and Sibling 1 will not be under seccomp at
2030 	 * all. Sibling 1 will enter seccomp and 0
2031 	 * will cause failure.
2032 	 */
2033 	self->sibling[0].diverge = 1;
2034 	tsync_start_sibling(&self->sibling[0]);
2035 	tsync_start_sibling(&self->sibling[1]);
2036 
2037 	while (self->sibling_count < TSYNC_SIBLINGS) {
2038 		sem_wait(&self->started);
2039 		self->sibling_count++;
2040 	}
2041 
2042 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2043 	ASSERT_NE(ENOSYS, errno) {
2044 		TH_LOG("Kernel does not support seccomp syscall!");
2045 	}
2046 	ASSERT_EQ(0, ret) {
2047 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2048 	}
2049 
2050 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2051 		      &self->apply_prog);
2052 	ASSERT_EQ(ret, self->sibling[0].system_tid) {
2053 		TH_LOG("Did not fail on diverged sibling.");
2054 	}
2055 	sib = 1;
2056 	if (ret == self->sibling[0].system_tid)
2057 		sib = 0;
2058 
2059 	pthread_mutex_lock(&self->mutex);
2060 
2061 	/* Increment the other siblings num_waits so we can clean up
2062 	 * the one we just saw.
2063 	 */
2064 	self->sibling[!sib].num_waits += 1;
2065 
2066 	/* Signal the thread to clean up*/
2067 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2068 		TH_LOG("cond broadcast non-zero");
2069 	}
2070 	pthread_mutex_unlock(&self->mutex);
2071 	pthread_join(self->sibling[sib].tid, &status);
2072 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2073 	/* Poll for actual task death. pthread_join doesn't guarantee it. */
2074 	while (!kill(self->sibling[sib].system_tid, 0))
2075 		sleep(0.1);
2076 	/* Switch to the remaining sibling */
2077 	sib = !sib;
2078 
2079 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2080 		      &self->apply_prog);
2081 	ASSERT_EQ(0, ret) {
2082 		TH_LOG("Expected the remaining sibling to sync");
2083 	};
2084 
2085 	pthread_mutex_lock(&self->mutex);
2086 
2087 	/* If remaining sibling didn't have a chance to wake up during
2088 	 * the first broadcast, manually reduce the num_waits now.
2089 	 */
2090 	if (self->sibling[sib].num_waits > 1)
2091 		self->sibling[sib].num_waits = 1;
2092 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2093 		TH_LOG("cond broadcast non-zero");
2094 	}
2095 	pthread_mutex_unlock(&self->mutex);
2096 	pthread_join(self->sibling[sib].tid, &status);
2097 	EXPECT_EQ(0, (long)status);
2098 	/* Poll for actual task death. pthread_join doesn't guarantee it. */
2099 	while (!kill(self->sibling[sib].system_tid, 0))
2100 		sleep(0.1);
2101 
2102 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2103 		      &self->apply_prog);
2104 	ASSERT_EQ(0, ret);  /* just us chickens */
2105 }
2106 
2107 /* Make sure restarted syscalls are seen directly as "restart_syscall". */
TEST(syscall_restart)2108 TEST(syscall_restart)
2109 {
2110 	long ret;
2111 	unsigned long msg;
2112 	pid_t child_pid;
2113 	int pipefd[2];
2114 	int status;
2115 	siginfo_t info = { };
2116 	struct sock_filter filter[] = {
2117 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2118 			 offsetof(struct seccomp_data, nr)),
2119 
2120 #ifdef __NR_sigreturn
2121 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 6, 0),
2122 #endif
2123 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 5, 0),
2124 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 4, 0),
2125 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 3, 0),
2126 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 4, 0),
2127 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0),
2128 
2129 		/* Allow __NR_write for easy logging. */
2130 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1),
2131 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2132 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2133 		/* The nanosleep jump target. */
2134 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100),
2135 		/* The restart_syscall jump target. */
2136 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200),
2137 	};
2138 	struct sock_fprog prog = {
2139 		.len = (unsigned short)ARRAY_SIZE(filter),
2140 		.filter = filter,
2141 	};
2142 #if defined(__arm__)
2143 	struct utsname utsbuf;
2144 #endif
2145 
2146 	ASSERT_EQ(0, pipe(pipefd));
2147 
2148 	child_pid = fork();
2149 	ASSERT_LE(0, child_pid);
2150 	if (child_pid == 0) {
2151 		/* Child uses EXPECT not ASSERT to deliver status correctly. */
2152 		char buf = ' ';
2153 		struct timespec timeout = { };
2154 
2155 		/* Attach parent as tracer and stop. */
2156 		EXPECT_EQ(0, ptrace(PTRACE_TRACEME));
2157 		EXPECT_EQ(0, raise(SIGSTOP));
2158 
2159 		EXPECT_EQ(0, close(pipefd[1]));
2160 
2161 		EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2162 			TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2163 		}
2164 
2165 		ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2166 		EXPECT_EQ(0, ret) {
2167 			TH_LOG("Failed to install filter!");
2168 		}
2169 
2170 		EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2171 			TH_LOG("Failed to read() sync from parent");
2172 		}
2173 		EXPECT_EQ('.', buf) {
2174 			TH_LOG("Failed to get sync data from read()");
2175 		}
2176 
2177 		/* Start nanosleep to be interrupted. */
2178 		timeout.tv_sec = 1;
2179 		errno = 0;
2180 		EXPECT_EQ(0, nanosleep(&timeout, NULL)) {
2181 			TH_LOG("Call to nanosleep() failed (errno %d)", errno);
2182 		}
2183 
2184 		/* Read final sync from parent. */
2185 		EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2186 			TH_LOG("Failed final read() from parent");
2187 		}
2188 		EXPECT_EQ('!', buf) {
2189 			TH_LOG("Failed to get final data from read()");
2190 		}
2191 
2192 		/* Directly report the status of our test harness results. */
2193 		syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS
2194 						     : EXIT_FAILURE);
2195 	}
2196 	EXPECT_EQ(0, close(pipefd[0]));
2197 
2198 	/* Attach to child, setup options, and release. */
2199 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2200 	ASSERT_EQ(true, WIFSTOPPED(status));
2201 	ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL,
2202 			    PTRACE_O_TRACESECCOMP));
2203 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2204 	ASSERT_EQ(1, write(pipefd[1], ".", 1));
2205 
2206 	/* Wait for nanosleep() to start. */
2207 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2208 	ASSERT_EQ(true, WIFSTOPPED(status));
2209 	ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2210 	ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2211 	ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2212 	ASSERT_EQ(0x100, msg);
2213 	EXPECT_EQ(__NR_nanosleep, get_syscall(_metadata, child_pid));
2214 
2215 	/* Might as well check siginfo for sanity while we're here. */
2216 	ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2217 	ASSERT_EQ(SIGTRAP, info.si_signo);
2218 	ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code);
2219 	EXPECT_EQ(0, info.si_errno);
2220 	EXPECT_EQ(getuid(), info.si_uid);
2221 	/* Verify signal delivery came from child (seccomp-triggered). */
2222 	EXPECT_EQ(child_pid, info.si_pid);
2223 
2224 	/* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */
2225 	ASSERT_EQ(0, kill(child_pid, SIGSTOP));
2226 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2227 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2228 	ASSERT_EQ(true, WIFSTOPPED(status));
2229 	ASSERT_EQ(SIGSTOP, WSTOPSIG(status));
2230 	/* Verify signal delivery came from parent now. */
2231 	ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2232 	EXPECT_EQ(getpid(), info.si_pid);
2233 
2234 	/* Restart nanosleep with SIGCONT, which triggers restart_syscall. */
2235 	ASSERT_EQ(0, kill(child_pid, SIGCONT));
2236 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2237 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2238 	ASSERT_EQ(true, WIFSTOPPED(status));
2239 	ASSERT_EQ(SIGCONT, WSTOPSIG(status));
2240 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2241 
2242 	/* Wait for restart_syscall() to start. */
2243 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2244 	ASSERT_EQ(true, WIFSTOPPED(status));
2245 	ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2246 	ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2247 	ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2248 
2249 	ASSERT_EQ(0x200, msg);
2250 	ret = get_syscall(_metadata, child_pid);
2251 #if defined(__arm__)
2252 	/*
2253 	 * FIXME:
2254 	 * - native ARM registers do NOT expose true syscall.
2255 	 * - compat ARM registers on ARM64 DO expose true syscall.
2256 	 */
2257 	ASSERT_EQ(0, uname(&utsbuf));
2258 	if (strncmp(utsbuf.machine, "arm", 3) == 0) {
2259 		EXPECT_EQ(__NR_nanosleep, ret);
2260 	} else
2261 #endif
2262 	{
2263 		EXPECT_EQ(__NR_restart_syscall, ret);
2264 	}
2265 
2266 	/* Write again to end test. */
2267 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2268 	ASSERT_EQ(1, write(pipefd[1], "!", 1));
2269 	EXPECT_EQ(0, close(pipefd[1]));
2270 
2271 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2272 	if (WIFSIGNALED(status) || WEXITSTATUS(status))
2273 		_metadata->passed = 0;
2274 }
2275 
2276 /*
2277  * TODO:
2278  * - add microbenchmarks
2279  * - expand NNP testing
2280  * - better arch-specific TRACE and TRAP handlers.
2281  * - endianness checking when appropriate
2282  * - 64-bit arg prodding
2283  * - arch value testing (x86 modes especially)
2284  * - ...
2285  */
2286 
2287 TEST_HARNESS_MAIN
2288