xref: /linux/tools/testing/selftests/seccomp/seccomp_bpf.c (revision b8d312aa075f33282565467662c4628dae0a2aff)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2012 The Chromium OS Authors. All rights reserved.
4  *
5  * Test code for seccomp bpf.
6  */
7 
8 #define _GNU_SOURCE
9 #include <sys/types.h>
10 
11 /*
12  * glibc 2.26 and later have SIGSYS in siginfo_t. Before that,
13  * we need to use the kernel's siginfo.h file and trick glibc
14  * into accepting it.
15  */
16 #if !__GLIBC_PREREQ(2, 26)
17 # include <asm/siginfo.h>
18 # define __have_siginfo_t 1
19 # define __have_sigval_t 1
20 # define __have_sigevent_t 1
21 #endif
22 
23 #include <errno.h>
24 #include <linux/filter.h>
25 #include <sys/prctl.h>
26 #include <sys/ptrace.h>
27 #include <sys/user.h>
28 #include <linux/prctl.h>
29 #include <linux/ptrace.h>
30 #include <linux/seccomp.h>
31 #include <pthread.h>
32 #include <semaphore.h>
33 #include <signal.h>
34 #include <stddef.h>
35 #include <stdbool.h>
36 #include <string.h>
37 #include <time.h>
38 #include <linux/elf.h>
39 #include <sys/uio.h>
40 #include <sys/utsname.h>
41 #include <sys/fcntl.h>
42 #include <sys/mman.h>
43 #include <sys/times.h>
44 #include <sys/socket.h>
45 #include <sys/ioctl.h>
46 
47 #include <unistd.h>
48 #include <sys/syscall.h>
49 #include <poll.h>
50 
51 #include "../kselftest_harness.h"
52 
53 #ifndef PR_SET_PTRACER
54 # define PR_SET_PTRACER 0x59616d61
55 #endif
56 
57 #ifndef PR_SET_NO_NEW_PRIVS
58 #define PR_SET_NO_NEW_PRIVS 38
59 #define PR_GET_NO_NEW_PRIVS 39
60 #endif
61 
62 #ifndef PR_SECCOMP_EXT
63 #define PR_SECCOMP_EXT 43
64 #endif
65 
66 #ifndef SECCOMP_EXT_ACT
67 #define SECCOMP_EXT_ACT 1
68 #endif
69 
70 #ifndef SECCOMP_EXT_ACT_TSYNC
71 #define SECCOMP_EXT_ACT_TSYNC 1
72 #endif
73 
74 #ifndef SECCOMP_MODE_STRICT
75 #define SECCOMP_MODE_STRICT 1
76 #endif
77 
78 #ifndef SECCOMP_MODE_FILTER
79 #define SECCOMP_MODE_FILTER 2
80 #endif
81 
82 #ifndef SECCOMP_RET_ALLOW
83 struct seccomp_data {
84 	int nr;
85 	__u32 arch;
86 	__u64 instruction_pointer;
87 	__u64 args[6];
88 };
89 #endif
90 
91 #ifndef SECCOMP_RET_KILL_PROCESS
92 #define SECCOMP_RET_KILL_PROCESS 0x80000000U /* kill the process */
93 #define SECCOMP_RET_KILL_THREAD	 0x00000000U /* kill the thread */
94 #endif
95 #ifndef SECCOMP_RET_KILL
96 #define SECCOMP_RET_KILL	 SECCOMP_RET_KILL_THREAD
97 #define SECCOMP_RET_TRAP	 0x00030000U /* disallow and force a SIGSYS */
98 #define SECCOMP_RET_ERRNO	 0x00050000U /* returns an errno */
99 #define SECCOMP_RET_TRACE	 0x7ff00000U /* pass to a tracer or disallow */
100 #define SECCOMP_RET_ALLOW	 0x7fff0000U /* allow */
101 #endif
102 #ifndef SECCOMP_RET_LOG
103 #define SECCOMP_RET_LOG		 0x7ffc0000U /* allow after logging */
104 #endif
105 
106 #ifndef __NR_seccomp
107 # if defined(__i386__)
108 #  define __NR_seccomp 354
109 # elif defined(__x86_64__)
110 #  define __NR_seccomp 317
111 # elif defined(__arm__)
112 #  define __NR_seccomp 383
113 # elif defined(__aarch64__)
114 #  define __NR_seccomp 277
115 # elif defined(__hppa__)
116 #  define __NR_seccomp 338
117 # elif defined(__powerpc__)
118 #  define __NR_seccomp 358
119 # elif defined(__s390__)
120 #  define __NR_seccomp 348
121 # else
122 #  warning "seccomp syscall number unknown for this architecture"
123 #  define __NR_seccomp 0xffff
124 # endif
125 #endif
126 
127 #ifndef SECCOMP_SET_MODE_STRICT
128 #define SECCOMP_SET_MODE_STRICT 0
129 #endif
130 
131 #ifndef SECCOMP_SET_MODE_FILTER
132 #define SECCOMP_SET_MODE_FILTER 1
133 #endif
134 
135 #ifndef SECCOMP_GET_ACTION_AVAIL
136 #define SECCOMP_GET_ACTION_AVAIL 2
137 #endif
138 
139 #ifndef SECCOMP_GET_NOTIF_SIZES
140 #define SECCOMP_GET_NOTIF_SIZES 3
141 #endif
142 
143 #ifndef SECCOMP_FILTER_FLAG_TSYNC
144 #define SECCOMP_FILTER_FLAG_TSYNC (1UL << 0)
145 #endif
146 
147 #ifndef SECCOMP_FILTER_FLAG_LOG
148 #define SECCOMP_FILTER_FLAG_LOG (1UL << 1)
149 #endif
150 
151 #ifndef SECCOMP_FILTER_FLAG_SPEC_ALLOW
152 #define SECCOMP_FILTER_FLAG_SPEC_ALLOW (1UL << 2)
153 #endif
154 
155 #ifndef PTRACE_SECCOMP_GET_METADATA
156 #define PTRACE_SECCOMP_GET_METADATA	0x420d
157 
158 struct seccomp_metadata {
159 	__u64 filter_off;       /* Input: which filter */
160 	__u64 flags;             /* Output: filter's flags */
161 };
162 #endif
163 
164 #ifndef SECCOMP_FILTER_FLAG_NEW_LISTENER
165 #define SECCOMP_FILTER_FLAG_NEW_LISTENER	(1UL << 3)
166 
167 #define SECCOMP_RET_USER_NOTIF 0x7fc00000U
168 
169 #define SECCOMP_IOC_MAGIC		'!'
170 #define SECCOMP_IO(nr)			_IO(SECCOMP_IOC_MAGIC, nr)
171 #define SECCOMP_IOR(nr, type)		_IOR(SECCOMP_IOC_MAGIC, nr, type)
172 #define SECCOMP_IOW(nr, type)		_IOW(SECCOMP_IOC_MAGIC, nr, type)
173 #define SECCOMP_IOWR(nr, type)		_IOWR(SECCOMP_IOC_MAGIC, nr, type)
174 
175 /* Flags for seccomp notification fd ioctl. */
176 #define SECCOMP_IOCTL_NOTIF_RECV	SECCOMP_IOWR(0, struct seccomp_notif)
177 #define SECCOMP_IOCTL_NOTIF_SEND	SECCOMP_IOWR(1,	\
178 						struct seccomp_notif_resp)
179 #define SECCOMP_IOCTL_NOTIF_ID_VALID	SECCOMP_IOR(2, __u64)
180 
181 struct seccomp_notif {
182 	__u64 id;
183 	__u32 pid;
184 	__u32 flags;
185 	struct seccomp_data data;
186 };
187 
188 struct seccomp_notif_resp {
189 	__u64 id;
190 	__s64 val;
191 	__s32 error;
192 	__u32 flags;
193 };
194 
195 struct seccomp_notif_sizes {
196 	__u16 seccomp_notif;
197 	__u16 seccomp_notif_resp;
198 	__u16 seccomp_data;
199 };
200 #endif
201 
202 #ifndef seccomp
203 int seccomp(unsigned int op, unsigned int flags, void *args)
204 {
205 	errno = 0;
206 	return syscall(__NR_seccomp, op, flags, args);
207 }
208 #endif
209 
210 #if __BYTE_ORDER == __LITTLE_ENDIAN
211 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]))
212 #elif __BYTE_ORDER == __BIG_ENDIAN
213 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32))
214 #else
215 #error "wut? Unknown __BYTE_ORDER?!"
216 #endif
217 
218 #define SIBLING_EXIT_UNKILLED	0xbadbeef
219 #define SIBLING_EXIT_FAILURE	0xbadface
220 #define SIBLING_EXIT_NEWPRIVS	0xbadfeed
221 
222 TEST(mode_strict_support)
223 {
224 	long ret;
225 
226 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
227 	ASSERT_EQ(0, ret) {
228 		TH_LOG("Kernel does not support CONFIG_SECCOMP");
229 	}
230 	syscall(__NR_exit, 0);
231 }
232 
233 TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL)
234 {
235 	long ret;
236 
237 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
238 	ASSERT_EQ(0, ret) {
239 		TH_LOG("Kernel does not support CONFIG_SECCOMP");
240 	}
241 	syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
242 		NULL, NULL, NULL);
243 	EXPECT_FALSE(true) {
244 		TH_LOG("Unreachable!");
245 	}
246 }
247 
248 /* Note! This doesn't test no new privs behavior */
249 TEST(no_new_privs_support)
250 {
251 	long ret;
252 
253 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
254 	EXPECT_EQ(0, ret) {
255 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
256 	}
257 }
258 
259 /* Tests kernel support by checking for a copy_from_user() fault on NULL. */
260 TEST(mode_filter_support)
261 {
262 	long ret;
263 
264 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
265 	ASSERT_EQ(0, ret) {
266 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
267 	}
268 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL);
269 	EXPECT_EQ(-1, ret);
270 	EXPECT_EQ(EFAULT, errno) {
271 		TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!");
272 	}
273 }
274 
275 TEST(mode_filter_without_nnp)
276 {
277 	struct sock_filter filter[] = {
278 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
279 	};
280 	struct sock_fprog prog = {
281 		.len = (unsigned short)ARRAY_SIZE(filter),
282 		.filter = filter,
283 	};
284 	long ret;
285 
286 	ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0);
287 	ASSERT_LE(0, ret) {
288 		TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS");
289 	}
290 	errno = 0;
291 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
292 	/* Succeeds with CAP_SYS_ADMIN, fails without */
293 	/* TODO(wad) check caps not euid */
294 	if (geteuid()) {
295 		EXPECT_EQ(-1, ret);
296 		EXPECT_EQ(EACCES, errno);
297 	} else {
298 		EXPECT_EQ(0, ret);
299 	}
300 }
301 
302 #define MAX_INSNS_PER_PATH 32768
303 
304 TEST(filter_size_limits)
305 {
306 	int i;
307 	int count = BPF_MAXINSNS + 1;
308 	struct sock_filter allow[] = {
309 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
310 	};
311 	struct sock_filter *filter;
312 	struct sock_fprog prog = { };
313 	long ret;
314 
315 	filter = calloc(count, sizeof(*filter));
316 	ASSERT_NE(NULL, filter);
317 
318 	for (i = 0; i < count; i++)
319 		filter[i] = allow[0];
320 
321 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
322 	ASSERT_EQ(0, ret);
323 
324 	prog.filter = filter;
325 	prog.len = count;
326 
327 	/* Too many filter instructions in a single filter. */
328 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
329 	ASSERT_NE(0, ret) {
330 		TH_LOG("Installing %d insn filter was allowed", prog.len);
331 	}
332 
333 	/* One less is okay, though. */
334 	prog.len -= 1;
335 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
336 	ASSERT_EQ(0, ret) {
337 		TH_LOG("Installing %d insn filter wasn't allowed", prog.len);
338 	}
339 }
340 
341 TEST(filter_chain_limits)
342 {
343 	int i;
344 	int count = BPF_MAXINSNS;
345 	struct sock_filter allow[] = {
346 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
347 	};
348 	struct sock_filter *filter;
349 	struct sock_fprog prog = { };
350 	long ret;
351 
352 	filter = calloc(count, sizeof(*filter));
353 	ASSERT_NE(NULL, filter);
354 
355 	for (i = 0; i < count; i++)
356 		filter[i] = allow[0];
357 
358 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
359 	ASSERT_EQ(0, ret);
360 
361 	prog.filter = filter;
362 	prog.len = 1;
363 
364 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
365 	ASSERT_EQ(0, ret);
366 
367 	prog.len = count;
368 
369 	/* Too many total filter instructions. */
370 	for (i = 0; i < MAX_INSNS_PER_PATH; i++) {
371 		ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
372 		if (ret != 0)
373 			break;
374 	}
375 	ASSERT_NE(0, ret) {
376 		TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)",
377 		       i, count, i * (count + 4));
378 	}
379 }
380 
381 TEST(mode_filter_cannot_move_to_strict)
382 {
383 	struct sock_filter filter[] = {
384 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
385 	};
386 	struct sock_fprog prog = {
387 		.len = (unsigned short)ARRAY_SIZE(filter),
388 		.filter = filter,
389 	};
390 	long ret;
391 
392 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
393 	ASSERT_EQ(0, ret);
394 
395 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
396 	ASSERT_EQ(0, ret);
397 
398 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0);
399 	EXPECT_EQ(-1, ret);
400 	EXPECT_EQ(EINVAL, errno);
401 }
402 
403 
404 TEST(mode_filter_get_seccomp)
405 {
406 	struct sock_filter filter[] = {
407 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
408 	};
409 	struct sock_fprog prog = {
410 		.len = (unsigned short)ARRAY_SIZE(filter),
411 		.filter = filter,
412 	};
413 	long ret;
414 
415 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
416 	ASSERT_EQ(0, ret);
417 
418 	ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
419 	EXPECT_EQ(0, ret);
420 
421 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
422 	ASSERT_EQ(0, ret);
423 
424 	ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
425 	EXPECT_EQ(2, ret);
426 }
427 
428 
429 TEST(ALLOW_all)
430 {
431 	struct sock_filter filter[] = {
432 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
433 	};
434 	struct sock_fprog prog = {
435 		.len = (unsigned short)ARRAY_SIZE(filter),
436 		.filter = filter,
437 	};
438 	long ret;
439 
440 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
441 	ASSERT_EQ(0, ret);
442 
443 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
444 	ASSERT_EQ(0, ret);
445 }
446 
447 TEST(empty_prog)
448 {
449 	struct sock_filter filter[] = {
450 	};
451 	struct sock_fprog prog = {
452 		.len = (unsigned short)ARRAY_SIZE(filter),
453 		.filter = filter,
454 	};
455 	long ret;
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 	EXPECT_EQ(-1, ret);
462 	EXPECT_EQ(EINVAL, errno);
463 }
464 
465 TEST(log_all)
466 {
467 	struct sock_filter filter[] = {
468 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG),
469 	};
470 	struct sock_fprog prog = {
471 		.len = (unsigned short)ARRAY_SIZE(filter),
472 		.filter = filter,
473 	};
474 	long ret;
475 	pid_t parent = getppid();
476 
477 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
478 	ASSERT_EQ(0, ret);
479 
480 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
481 	ASSERT_EQ(0, ret);
482 
483 	/* getppid() should succeed and be logged (no check for logging) */
484 	EXPECT_EQ(parent, syscall(__NR_getppid));
485 }
486 
487 TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS)
488 {
489 	struct sock_filter filter[] = {
490 		BPF_STMT(BPF_RET|BPF_K, 0x10000000U),
491 	};
492 	struct sock_fprog prog = {
493 		.len = (unsigned short)ARRAY_SIZE(filter),
494 		.filter = filter,
495 	};
496 	long ret;
497 
498 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
499 	ASSERT_EQ(0, ret);
500 
501 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
502 	ASSERT_EQ(0, ret);
503 	EXPECT_EQ(0, syscall(__NR_getpid)) {
504 		TH_LOG("getpid() shouldn't ever return");
505 	}
506 }
507 
508 /* return code >= 0x80000000 is unused. */
509 TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS)
510 {
511 	struct sock_filter filter[] = {
512 		BPF_STMT(BPF_RET|BPF_K, 0x90000000U),
513 	};
514 	struct sock_fprog prog = {
515 		.len = (unsigned short)ARRAY_SIZE(filter),
516 		.filter = filter,
517 	};
518 	long ret;
519 
520 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
521 	ASSERT_EQ(0, ret);
522 
523 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
524 	ASSERT_EQ(0, ret);
525 	EXPECT_EQ(0, syscall(__NR_getpid)) {
526 		TH_LOG("getpid() shouldn't ever return");
527 	}
528 }
529 
530 TEST_SIGNAL(KILL_all, SIGSYS)
531 {
532 	struct sock_filter filter[] = {
533 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
534 	};
535 	struct sock_fprog prog = {
536 		.len = (unsigned short)ARRAY_SIZE(filter),
537 		.filter = filter,
538 	};
539 	long ret;
540 
541 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
542 	ASSERT_EQ(0, ret);
543 
544 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
545 	ASSERT_EQ(0, ret);
546 }
547 
548 TEST_SIGNAL(KILL_one, SIGSYS)
549 {
550 	struct sock_filter filter[] = {
551 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
552 			offsetof(struct seccomp_data, nr)),
553 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
554 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
555 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
556 	};
557 	struct sock_fprog prog = {
558 		.len = (unsigned short)ARRAY_SIZE(filter),
559 		.filter = filter,
560 	};
561 	long ret;
562 	pid_t parent = getppid();
563 
564 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
565 	ASSERT_EQ(0, ret);
566 
567 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
568 	ASSERT_EQ(0, ret);
569 
570 	EXPECT_EQ(parent, syscall(__NR_getppid));
571 	/* getpid() should never return. */
572 	EXPECT_EQ(0, syscall(__NR_getpid));
573 }
574 
575 TEST_SIGNAL(KILL_one_arg_one, SIGSYS)
576 {
577 	void *fatal_address;
578 	struct sock_filter filter[] = {
579 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
580 			offsetof(struct seccomp_data, nr)),
581 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0),
582 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
583 		/* Only both with lower 32-bit for now. */
584 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)),
585 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K,
586 			(unsigned long)&fatal_address, 0, 1),
587 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
588 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
589 	};
590 	struct sock_fprog prog = {
591 		.len = (unsigned short)ARRAY_SIZE(filter),
592 		.filter = filter,
593 	};
594 	long ret;
595 	pid_t parent = getppid();
596 	struct tms timebuf;
597 	clock_t clock = times(&timebuf);
598 
599 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
600 	ASSERT_EQ(0, ret);
601 
602 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
603 	ASSERT_EQ(0, ret);
604 
605 	EXPECT_EQ(parent, syscall(__NR_getppid));
606 	EXPECT_LE(clock, syscall(__NR_times, &timebuf));
607 	/* times() should never return. */
608 	EXPECT_EQ(0, syscall(__NR_times, &fatal_address));
609 }
610 
611 TEST_SIGNAL(KILL_one_arg_six, SIGSYS)
612 {
613 #ifndef __NR_mmap2
614 	int sysno = __NR_mmap;
615 #else
616 	int sysno = __NR_mmap2;
617 #endif
618 	struct sock_filter filter[] = {
619 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
620 			offsetof(struct seccomp_data, nr)),
621 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0),
622 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
623 		/* Only both with lower 32-bit for now. */
624 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)),
625 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1),
626 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
627 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
628 	};
629 	struct sock_fprog prog = {
630 		.len = (unsigned short)ARRAY_SIZE(filter),
631 		.filter = filter,
632 	};
633 	long ret;
634 	pid_t parent = getppid();
635 	int fd;
636 	void *map1, *map2;
637 	int page_size = sysconf(_SC_PAGESIZE);
638 
639 	ASSERT_LT(0, page_size);
640 
641 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
642 	ASSERT_EQ(0, ret);
643 
644 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
645 	ASSERT_EQ(0, ret);
646 
647 	fd = open("/dev/zero", O_RDONLY);
648 	ASSERT_NE(-1, fd);
649 
650 	EXPECT_EQ(parent, syscall(__NR_getppid));
651 	map1 = (void *)syscall(sysno,
652 		NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size);
653 	EXPECT_NE(MAP_FAILED, map1);
654 	/* mmap2() should never return. */
655 	map2 = (void *)syscall(sysno,
656 		 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE);
657 	EXPECT_EQ(MAP_FAILED, map2);
658 
659 	/* The test failed, so clean up the resources. */
660 	munmap(map1, page_size);
661 	munmap(map2, page_size);
662 	close(fd);
663 }
664 
665 /* This is a thread task to die via seccomp filter violation. */
666 void *kill_thread(void *data)
667 {
668 	bool die = (bool)data;
669 
670 	if (die) {
671 		prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
672 		return (void *)SIBLING_EXIT_FAILURE;
673 	}
674 
675 	return (void *)SIBLING_EXIT_UNKILLED;
676 }
677 
678 /* Prepare a thread that will kill itself or both of us. */
679 void kill_thread_or_group(struct __test_metadata *_metadata, bool kill_process)
680 {
681 	pthread_t thread;
682 	void *status;
683 	/* Kill only when calling __NR_prctl. */
684 	struct sock_filter filter_thread[] = {
685 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
686 			offsetof(struct seccomp_data, nr)),
687 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
688 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD),
689 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
690 	};
691 	struct sock_fprog prog_thread = {
692 		.len = (unsigned short)ARRAY_SIZE(filter_thread),
693 		.filter = filter_thread,
694 	};
695 	struct sock_filter filter_process[] = {
696 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
697 			offsetof(struct seccomp_data, nr)),
698 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
699 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_PROCESS),
700 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
701 	};
702 	struct sock_fprog prog_process = {
703 		.len = (unsigned short)ARRAY_SIZE(filter_process),
704 		.filter = filter_process,
705 	};
706 
707 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
708 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
709 	}
710 
711 	ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0,
712 			     kill_process ? &prog_process : &prog_thread));
713 
714 	/*
715 	 * Add the KILL_THREAD rule again to make sure that the KILL_PROCESS
716 	 * flag cannot be downgraded by a new filter.
717 	 */
718 	ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog_thread));
719 
720 	/* Start a thread that will exit immediately. */
721 	ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)false));
722 	ASSERT_EQ(0, pthread_join(thread, &status));
723 	ASSERT_EQ(SIBLING_EXIT_UNKILLED, (unsigned long)status);
724 
725 	/* Start a thread that will die immediately. */
726 	ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)true));
727 	ASSERT_EQ(0, pthread_join(thread, &status));
728 	ASSERT_NE(SIBLING_EXIT_FAILURE, (unsigned long)status);
729 
730 	/*
731 	 * If we get here, only the spawned thread died. Let the parent know
732 	 * the whole process didn't die (i.e. this thread, the spawner,
733 	 * stayed running).
734 	 */
735 	exit(42);
736 }
737 
738 TEST(KILL_thread)
739 {
740 	int status;
741 	pid_t child_pid;
742 
743 	child_pid = fork();
744 	ASSERT_LE(0, child_pid);
745 	if (child_pid == 0) {
746 		kill_thread_or_group(_metadata, false);
747 		_exit(38);
748 	}
749 
750 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
751 
752 	/* If only the thread was killed, we'll see exit 42. */
753 	ASSERT_TRUE(WIFEXITED(status));
754 	ASSERT_EQ(42, WEXITSTATUS(status));
755 }
756 
757 TEST(KILL_process)
758 {
759 	int status;
760 	pid_t child_pid;
761 
762 	child_pid = fork();
763 	ASSERT_LE(0, child_pid);
764 	if (child_pid == 0) {
765 		kill_thread_or_group(_metadata, true);
766 		_exit(38);
767 	}
768 
769 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
770 
771 	/* If the entire process was killed, we'll see SIGSYS. */
772 	ASSERT_TRUE(WIFSIGNALED(status));
773 	ASSERT_EQ(SIGSYS, WTERMSIG(status));
774 }
775 
776 /* TODO(wad) add 64-bit versus 32-bit arg tests. */
777 TEST(arg_out_of_range)
778 {
779 	struct sock_filter filter[] = {
780 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)),
781 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
782 	};
783 	struct sock_fprog prog = {
784 		.len = (unsigned short)ARRAY_SIZE(filter),
785 		.filter = filter,
786 	};
787 	long ret;
788 
789 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
790 	ASSERT_EQ(0, ret);
791 
792 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
793 	EXPECT_EQ(-1, ret);
794 	EXPECT_EQ(EINVAL, errno);
795 }
796 
797 #define ERRNO_FILTER(name, errno)					\
798 	struct sock_filter _read_filter_##name[] = {			\
799 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,				\
800 			offsetof(struct seccomp_data, nr)),		\
801 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),	\
802 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | errno),	\
803 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),		\
804 	};								\
805 	struct sock_fprog prog_##name = {				\
806 		.len = (unsigned short)ARRAY_SIZE(_read_filter_##name),	\
807 		.filter = _read_filter_##name,				\
808 	}
809 
810 /* Make sure basic errno values are correctly passed through a filter. */
811 TEST(ERRNO_valid)
812 {
813 	ERRNO_FILTER(valid, E2BIG);
814 	long ret;
815 	pid_t parent = getppid();
816 
817 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
818 	ASSERT_EQ(0, ret);
819 
820 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_valid);
821 	ASSERT_EQ(0, ret);
822 
823 	EXPECT_EQ(parent, syscall(__NR_getppid));
824 	EXPECT_EQ(-1, read(0, NULL, 0));
825 	EXPECT_EQ(E2BIG, errno);
826 }
827 
828 /* Make sure an errno of zero is correctly handled by the arch code. */
829 TEST(ERRNO_zero)
830 {
831 	ERRNO_FILTER(zero, 0);
832 	long ret;
833 	pid_t parent = getppid();
834 
835 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
836 	ASSERT_EQ(0, ret);
837 
838 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_zero);
839 	ASSERT_EQ(0, ret);
840 
841 	EXPECT_EQ(parent, syscall(__NR_getppid));
842 	/* "errno" of 0 is ok. */
843 	EXPECT_EQ(0, read(0, NULL, 0));
844 }
845 
846 /*
847  * The SECCOMP_RET_DATA mask is 16 bits wide, but errno is smaller.
848  * This tests that the errno value gets capped correctly, fixed by
849  * 580c57f10768 ("seccomp: cap SECCOMP_RET_ERRNO data to MAX_ERRNO").
850  */
851 TEST(ERRNO_capped)
852 {
853 	ERRNO_FILTER(capped, 4096);
854 	long ret;
855 	pid_t parent = getppid();
856 
857 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
858 	ASSERT_EQ(0, ret);
859 
860 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_capped);
861 	ASSERT_EQ(0, ret);
862 
863 	EXPECT_EQ(parent, syscall(__NR_getppid));
864 	EXPECT_EQ(-1, read(0, NULL, 0));
865 	EXPECT_EQ(4095, errno);
866 }
867 
868 /*
869  * Filters are processed in reverse order: last applied is executed first.
870  * Since only the SECCOMP_RET_ACTION mask is tested for return values, the
871  * SECCOMP_RET_DATA mask results will follow the most recently applied
872  * matching filter return (and not the lowest or highest value).
873  */
874 TEST(ERRNO_order)
875 {
876 	ERRNO_FILTER(first,  11);
877 	ERRNO_FILTER(second, 13);
878 	ERRNO_FILTER(third,  12);
879 	long ret;
880 	pid_t parent = getppid();
881 
882 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
883 	ASSERT_EQ(0, ret);
884 
885 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_first);
886 	ASSERT_EQ(0, ret);
887 
888 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_second);
889 	ASSERT_EQ(0, ret);
890 
891 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_third);
892 	ASSERT_EQ(0, ret);
893 
894 	EXPECT_EQ(parent, syscall(__NR_getppid));
895 	EXPECT_EQ(-1, read(0, NULL, 0));
896 	EXPECT_EQ(12, errno);
897 }
898 
899 FIXTURE_DATA(TRAP) {
900 	struct sock_fprog prog;
901 };
902 
903 FIXTURE_SETUP(TRAP)
904 {
905 	struct sock_filter filter[] = {
906 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
907 			offsetof(struct seccomp_data, nr)),
908 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
909 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
910 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
911 	};
912 
913 	memset(&self->prog, 0, sizeof(self->prog));
914 	self->prog.filter = malloc(sizeof(filter));
915 	ASSERT_NE(NULL, self->prog.filter);
916 	memcpy(self->prog.filter, filter, sizeof(filter));
917 	self->prog.len = (unsigned short)ARRAY_SIZE(filter);
918 }
919 
920 FIXTURE_TEARDOWN(TRAP)
921 {
922 	if (self->prog.filter)
923 		free(self->prog.filter);
924 }
925 
926 TEST_F_SIGNAL(TRAP, dfl, SIGSYS)
927 {
928 	long ret;
929 
930 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
931 	ASSERT_EQ(0, ret);
932 
933 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
934 	ASSERT_EQ(0, ret);
935 	syscall(__NR_getpid);
936 }
937 
938 /* Ensure that SIGSYS overrides SIG_IGN */
939 TEST_F_SIGNAL(TRAP, ign, SIGSYS)
940 {
941 	long ret;
942 
943 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
944 	ASSERT_EQ(0, ret);
945 
946 	signal(SIGSYS, SIG_IGN);
947 
948 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
949 	ASSERT_EQ(0, ret);
950 	syscall(__NR_getpid);
951 }
952 
953 static siginfo_t TRAP_info;
954 static volatile int TRAP_nr;
955 static void TRAP_action(int nr, siginfo_t *info, void *void_context)
956 {
957 	memcpy(&TRAP_info, info, sizeof(TRAP_info));
958 	TRAP_nr = nr;
959 }
960 
961 TEST_F(TRAP, handler)
962 {
963 	int ret, test;
964 	struct sigaction act;
965 	sigset_t mask;
966 
967 	memset(&act, 0, sizeof(act));
968 	sigemptyset(&mask);
969 	sigaddset(&mask, SIGSYS);
970 
971 	act.sa_sigaction = &TRAP_action;
972 	act.sa_flags = SA_SIGINFO;
973 	ret = sigaction(SIGSYS, &act, NULL);
974 	ASSERT_EQ(0, ret) {
975 		TH_LOG("sigaction failed");
976 	}
977 	ret = sigprocmask(SIG_UNBLOCK, &mask, NULL);
978 	ASSERT_EQ(0, ret) {
979 		TH_LOG("sigprocmask failed");
980 	}
981 
982 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
983 	ASSERT_EQ(0, ret);
984 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
985 	ASSERT_EQ(0, ret);
986 	TRAP_nr = 0;
987 	memset(&TRAP_info, 0, sizeof(TRAP_info));
988 	/* Expect the registers to be rolled back. (nr = error) may vary
989 	 * based on arch. */
990 	ret = syscall(__NR_getpid);
991 	/* Silence gcc warning about volatile. */
992 	test = TRAP_nr;
993 	EXPECT_EQ(SIGSYS, test);
994 	struct local_sigsys {
995 		void *_call_addr;	/* calling user insn */
996 		int _syscall;		/* triggering system call number */
997 		unsigned int _arch;	/* AUDIT_ARCH_* of syscall */
998 	} *sigsys = (struct local_sigsys *)
999 #ifdef si_syscall
1000 		&(TRAP_info.si_call_addr);
1001 #else
1002 		&TRAP_info.si_pid;
1003 #endif
1004 	EXPECT_EQ(__NR_getpid, sigsys->_syscall);
1005 	/* Make sure arch is non-zero. */
1006 	EXPECT_NE(0, sigsys->_arch);
1007 	EXPECT_NE(0, (unsigned long)sigsys->_call_addr);
1008 }
1009 
1010 FIXTURE_DATA(precedence) {
1011 	struct sock_fprog allow;
1012 	struct sock_fprog log;
1013 	struct sock_fprog trace;
1014 	struct sock_fprog error;
1015 	struct sock_fprog trap;
1016 	struct sock_fprog kill;
1017 };
1018 
1019 FIXTURE_SETUP(precedence)
1020 {
1021 	struct sock_filter allow_insns[] = {
1022 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1023 	};
1024 	struct sock_filter log_insns[] = {
1025 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1026 			offsetof(struct seccomp_data, nr)),
1027 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1028 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1029 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG),
1030 	};
1031 	struct sock_filter trace_insns[] = {
1032 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1033 			offsetof(struct seccomp_data, nr)),
1034 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1035 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1036 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE),
1037 	};
1038 	struct sock_filter error_insns[] = {
1039 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1040 			offsetof(struct seccomp_data, nr)),
1041 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1042 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1043 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO),
1044 	};
1045 	struct sock_filter trap_insns[] = {
1046 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1047 			offsetof(struct seccomp_data, nr)),
1048 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1049 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1050 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
1051 	};
1052 	struct sock_filter kill_insns[] = {
1053 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1054 			offsetof(struct seccomp_data, nr)),
1055 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1056 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1057 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1058 	};
1059 
1060 	memset(self, 0, sizeof(*self));
1061 #define FILTER_ALLOC(_x) \
1062 	self->_x.filter = malloc(sizeof(_x##_insns)); \
1063 	ASSERT_NE(NULL, self->_x.filter); \
1064 	memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \
1065 	self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns)
1066 	FILTER_ALLOC(allow);
1067 	FILTER_ALLOC(log);
1068 	FILTER_ALLOC(trace);
1069 	FILTER_ALLOC(error);
1070 	FILTER_ALLOC(trap);
1071 	FILTER_ALLOC(kill);
1072 }
1073 
1074 FIXTURE_TEARDOWN(precedence)
1075 {
1076 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter)
1077 	FILTER_FREE(allow);
1078 	FILTER_FREE(log);
1079 	FILTER_FREE(trace);
1080 	FILTER_FREE(error);
1081 	FILTER_FREE(trap);
1082 	FILTER_FREE(kill);
1083 }
1084 
1085 TEST_F(precedence, allow_ok)
1086 {
1087 	pid_t parent, res = 0;
1088 	long ret;
1089 
1090 	parent = getppid();
1091 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1092 	ASSERT_EQ(0, ret);
1093 
1094 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1095 	ASSERT_EQ(0, ret);
1096 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1097 	ASSERT_EQ(0, ret);
1098 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1099 	ASSERT_EQ(0, ret);
1100 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1101 	ASSERT_EQ(0, ret);
1102 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1103 	ASSERT_EQ(0, ret);
1104 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1105 	ASSERT_EQ(0, ret);
1106 	/* Should work just fine. */
1107 	res = syscall(__NR_getppid);
1108 	EXPECT_EQ(parent, res);
1109 }
1110 
1111 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS)
1112 {
1113 	pid_t parent, res = 0;
1114 	long ret;
1115 
1116 	parent = getppid();
1117 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1118 	ASSERT_EQ(0, ret);
1119 
1120 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1121 	ASSERT_EQ(0, ret);
1122 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1123 	ASSERT_EQ(0, ret);
1124 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1125 	ASSERT_EQ(0, ret);
1126 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1127 	ASSERT_EQ(0, ret);
1128 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1129 	ASSERT_EQ(0, ret);
1130 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1131 	ASSERT_EQ(0, ret);
1132 	/* Should work just fine. */
1133 	res = syscall(__NR_getppid);
1134 	EXPECT_EQ(parent, res);
1135 	/* getpid() should never return. */
1136 	res = syscall(__NR_getpid);
1137 	EXPECT_EQ(0, res);
1138 }
1139 
1140 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS)
1141 {
1142 	pid_t parent;
1143 	long ret;
1144 
1145 	parent = getppid();
1146 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1147 	ASSERT_EQ(0, ret);
1148 
1149 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1150 	ASSERT_EQ(0, ret);
1151 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1152 	ASSERT_EQ(0, ret);
1153 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1154 	ASSERT_EQ(0, ret);
1155 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1156 	ASSERT_EQ(0, ret);
1157 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1158 	ASSERT_EQ(0, ret);
1159 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1160 	ASSERT_EQ(0, ret);
1161 	/* Should work just fine. */
1162 	EXPECT_EQ(parent, syscall(__NR_getppid));
1163 	/* getpid() should never return. */
1164 	EXPECT_EQ(0, syscall(__NR_getpid));
1165 }
1166 
1167 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS)
1168 {
1169 	pid_t parent;
1170 	long ret;
1171 
1172 	parent = getppid();
1173 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1174 	ASSERT_EQ(0, ret);
1175 
1176 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1177 	ASSERT_EQ(0, ret);
1178 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1179 	ASSERT_EQ(0, ret);
1180 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1181 	ASSERT_EQ(0, ret);
1182 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1183 	ASSERT_EQ(0, ret);
1184 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1185 	ASSERT_EQ(0, ret);
1186 	/* Should work just fine. */
1187 	EXPECT_EQ(parent, syscall(__NR_getppid));
1188 	/* getpid() should never return. */
1189 	EXPECT_EQ(0, syscall(__NR_getpid));
1190 }
1191 
1192 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS)
1193 {
1194 	pid_t parent;
1195 	long ret;
1196 
1197 	parent = getppid();
1198 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1199 	ASSERT_EQ(0, ret);
1200 
1201 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1202 	ASSERT_EQ(0, ret);
1203 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1204 	ASSERT_EQ(0, ret);
1205 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1206 	ASSERT_EQ(0, ret);
1207 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1208 	ASSERT_EQ(0, ret);
1209 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1210 	ASSERT_EQ(0, ret);
1211 	/* Should work just fine. */
1212 	EXPECT_EQ(parent, syscall(__NR_getppid));
1213 	/* getpid() should never return. */
1214 	EXPECT_EQ(0, syscall(__NR_getpid));
1215 }
1216 
1217 TEST_F(precedence, errno_is_third)
1218 {
1219 	pid_t parent;
1220 	long ret;
1221 
1222 	parent = getppid();
1223 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1224 	ASSERT_EQ(0, ret);
1225 
1226 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1227 	ASSERT_EQ(0, ret);
1228 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1229 	ASSERT_EQ(0, ret);
1230 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1231 	ASSERT_EQ(0, ret);
1232 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1233 	ASSERT_EQ(0, ret);
1234 	/* Should work just fine. */
1235 	EXPECT_EQ(parent, syscall(__NR_getppid));
1236 	EXPECT_EQ(0, syscall(__NR_getpid));
1237 }
1238 
1239 TEST_F(precedence, errno_is_third_in_any_order)
1240 {
1241 	pid_t parent;
1242 	long ret;
1243 
1244 	parent = getppid();
1245 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1246 	ASSERT_EQ(0, ret);
1247 
1248 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1249 	ASSERT_EQ(0, ret);
1250 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1251 	ASSERT_EQ(0, ret);
1252 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1253 	ASSERT_EQ(0, ret);
1254 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1255 	ASSERT_EQ(0, ret);
1256 	/* Should work just fine. */
1257 	EXPECT_EQ(parent, syscall(__NR_getppid));
1258 	EXPECT_EQ(0, syscall(__NR_getpid));
1259 }
1260 
1261 TEST_F(precedence, trace_is_fourth)
1262 {
1263 	pid_t parent;
1264 	long ret;
1265 
1266 	parent = getppid();
1267 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1268 	ASSERT_EQ(0, ret);
1269 
1270 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1271 	ASSERT_EQ(0, ret);
1272 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1273 	ASSERT_EQ(0, ret);
1274 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1275 	ASSERT_EQ(0, ret);
1276 	/* Should work just fine. */
1277 	EXPECT_EQ(parent, syscall(__NR_getppid));
1278 	/* No ptracer */
1279 	EXPECT_EQ(-1, syscall(__NR_getpid));
1280 }
1281 
1282 TEST_F(precedence, trace_is_fourth_in_any_order)
1283 {
1284 	pid_t parent;
1285 	long ret;
1286 
1287 	parent = getppid();
1288 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1289 	ASSERT_EQ(0, ret);
1290 
1291 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1292 	ASSERT_EQ(0, ret);
1293 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1294 	ASSERT_EQ(0, ret);
1295 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1296 	ASSERT_EQ(0, ret);
1297 	/* Should work just fine. */
1298 	EXPECT_EQ(parent, syscall(__NR_getppid));
1299 	/* No ptracer */
1300 	EXPECT_EQ(-1, syscall(__NR_getpid));
1301 }
1302 
1303 TEST_F(precedence, log_is_fifth)
1304 {
1305 	pid_t mypid, parent;
1306 	long ret;
1307 
1308 	mypid = getpid();
1309 	parent = getppid();
1310 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1311 	ASSERT_EQ(0, ret);
1312 
1313 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1314 	ASSERT_EQ(0, ret);
1315 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1316 	ASSERT_EQ(0, ret);
1317 	/* Should work just fine. */
1318 	EXPECT_EQ(parent, syscall(__NR_getppid));
1319 	/* Should also work just fine */
1320 	EXPECT_EQ(mypid, syscall(__NR_getpid));
1321 }
1322 
1323 TEST_F(precedence, log_is_fifth_in_any_order)
1324 {
1325 	pid_t mypid, parent;
1326 	long ret;
1327 
1328 	mypid = getpid();
1329 	parent = getppid();
1330 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1331 	ASSERT_EQ(0, ret);
1332 
1333 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1334 	ASSERT_EQ(0, ret);
1335 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1336 	ASSERT_EQ(0, ret);
1337 	/* Should work just fine. */
1338 	EXPECT_EQ(parent, syscall(__NR_getppid));
1339 	/* Should also work just fine */
1340 	EXPECT_EQ(mypid, syscall(__NR_getpid));
1341 }
1342 
1343 #ifndef PTRACE_O_TRACESECCOMP
1344 #define PTRACE_O_TRACESECCOMP	0x00000080
1345 #endif
1346 
1347 /* Catch the Ubuntu 12.04 value error. */
1348 #if PTRACE_EVENT_SECCOMP != 7
1349 #undef PTRACE_EVENT_SECCOMP
1350 #endif
1351 
1352 #ifndef PTRACE_EVENT_SECCOMP
1353 #define PTRACE_EVENT_SECCOMP 7
1354 #endif
1355 
1356 #define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP)
1357 bool tracer_running;
1358 void tracer_stop(int sig)
1359 {
1360 	tracer_running = false;
1361 }
1362 
1363 typedef void tracer_func_t(struct __test_metadata *_metadata,
1364 			   pid_t tracee, int status, void *args);
1365 
1366 void start_tracer(struct __test_metadata *_metadata, int fd, pid_t tracee,
1367 	    tracer_func_t tracer_func, void *args, bool ptrace_syscall)
1368 {
1369 	int ret = -1;
1370 	struct sigaction action = {
1371 		.sa_handler = tracer_stop,
1372 	};
1373 
1374 	/* Allow external shutdown. */
1375 	tracer_running = true;
1376 	ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL));
1377 
1378 	errno = 0;
1379 	while (ret == -1 && errno != EINVAL)
1380 		ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0);
1381 	ASSERT_EQ(0, ret) {
1382 		kill(tracee, SIGKILL);
1383 	}
1384 	/* Wait for attach stop */
1385 	wait(NULL);
1386 
1387 	ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, ptrace_syscall ?
1388 						      PTRACE_O_TRACESYSGOOD :
1389 						      PTRACE_O_TRACESECCOMP);
1390 	ASSERT_EQ(0, ret) {
1391 		TH_LOG("Failed to set PTRACE_O_TRACESECCOMP");
1392 		kill(tracee, SIGKILL);
1393 	}
1394 	ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1395 		     tracee, NULL, 0);
1396 	ASSERT_EQ(0, ret);
1397 
1398 	/* Unblock the tracee */
1399 	ASSERT_EQ(1, write(fd, "A", 1));
1400 	ASSERT_EQ(0, close(fd));
1401 
1402 	/* Run until we're shut down. Must assert to stop execution. */
1403 	while (tracer_running) {
1404 		int status;
1405 
1406 		if (wait(&status) != tracee)
1407 			continue;
1408 		if (WIFSIGNALED(status) || WIFEXITED(status))
1409 			/* Child is dead. Time to go. */
1410 			return;
1411 
1412 		/* Check if this is a seccomp event. */
1413 		ASSERT_EQ(!ptrace_syscall, IS_SECCOMP_EVENT(status));
1414 
1415 		tracer_func(_metadata, tracee, status, args);
1416 
1417 		ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1418 			     tracee, NULL, 0);
1419 		ASSERT_EQ(0, ret);
1420 	}
1421 	/* Directly report the status of our test harness results. */
1422 	syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS : EXIT_FAILURE);
1423 }
1424 
1425 /* Common tracer setup/teardown functions. */
1426 void cont_handler(int num)
1427 { }
1428 pid_t setup_trace_fixture(struct __test_metadata *_metadata,
1429 			  tracer_func_t func, void *args, bool ptrace_syscall)
1430 {
1431 	char sync;
1432 	int pipefd[2];
1433 	pid_t tracer_pid;
1434 	pid_t tracee = getpid();
1435 
1436 	/* Setup a pipe for clean synchronization. */
1437 	ASSERT_EQ(0, pipe(pipefd));
1438 
1439 	/* Fork a child which we'll promote to tracer */
1440 	tracer_pid = fork();
1441 	ASSERT_LE(0, tracer_pid);
1442 	signal(SIGALRM, cont_handler);
1443 	if (tracer_pid == 0) {
1444 		close(pipefd[0]);
1445 		start_tracer(_metadata, pipefd[1], tracee, func, args,
1446 			     ptrace_syscall);
1447 		syscall(__NR_exit, 0);
1448 	}
1449 	close(pipefd[1]);
1450 	prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0);
1451 	read(pipefd[0], &sync, 1);
1452 	close(pipefd[0]);
1453 
1454 	return tracer_pid;
1455 }
1456 void teardown_trace_fixture(struct __test_metadata *_metadata,
1457 			    pid_t tracer)
1458 {
1459 	if (tracer) {
1460 		int status;
1461 		/*
1462 		 * Extract the exit code from the other process and
1463 		 * adopt it for ourselves in case its asserts failed.
1464 		 */
1465 		ASSERT_EQ(0, kill(tracer, SIGUSR1));
1466 		ASSERT_EQ(tracer, waitpid(tracer, &status, 0));
1467 		if (WEXITSTATUS(status))
1468 			_metadata->passed = 0;
1469 	}
1470 }
1471 
1472 /* "poke" tracer arguments and function. */
1473 struct tracer_args_poke_t {
1474 	unsigned long poke_addr;
1475 };
1476 
1477 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status,
1478 		 void *args)
1479 {
1480 	int ret;
1481 	unsigned long msg;
1482 	struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args;
1483 
1484 	ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1485 	EXPECT_EQ(0, ret);
1486 	/* If this fails, don't try to recover. */
1487 	ASSERT_EQ(0x1001, msg) {
1488 		kill(tracee, SIGKILL);
1489 	}
1490 	/*
1491 	 * Poke in the message.
1492 	 * Registers are not touched to try to keep this relatively arch
1493 	 * agnostic.
1494 	 */
1495 	ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001);
1496 	EXPECT_EQ(0, ret);
1497 }
1498 
1499 FIXTURE_DATA(TRACE_poke) {
1500 	struct sock_fprog prog;
1501 	pid_t tracer;
1502 	long poked;
1503 	struct tracer_args_poke_t tracer_args;
1504 };
1505 
1506 FIXTURE_SETUP(TRACE_poke)
1507 {
1508 	struct sock_filter filter[] = {
1509 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1510 			offsetof(struct seccomp_data, nr)),
1511 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1512 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001),
1513 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1514 	};
1515 
1516 	self->poked = 0;
1517 	memset(&self->prog, 0, sizeof(self->prog));
1518 	self->prog.filter = malloc(sizeof(filter));
1519 	ASSERT_NE(NULL, self->prog.filter);
1520 	memcpy(self->prog.filter, filter, sizeof(filter));
1521 	self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1522 
1523 	/* Set up tracer args. */
1524 	self->tracer_args.poke_addr = (unsigned long)&self->poked;
1525 
1526 	/* Launch tracer. */
1527 	self->tracer = setup_trace_fixture(_metadata, tracer_poke,
1528 					   &self->tracer_args, false);
1529 }
1530 
1531 FIXTURE_TEARDOWN(TRACE_poke)
1532 {
1533 	teardown_trace_fixture(_metadata, self->tracer);
1534 	if (self->prog.filter)
1535 		free(self->prog.filter);
1536 }
1537 
1538 TEST_F(TRACE_poke, read_has_side_effects)
1539 {
1540 	ssize_t ret;
1541 
1542 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1543 	ASSERT_EQ(0, ret);
1544 
1545 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1546 	ASSERT_EQ(0, ret);
1547 
1548 	EXPECT_EQ(0, self->poked);
1549 	ret = read(-1, NULL, 0);
1550 	EXPECT_EQ(-1, ret);
1551 	EXPECT_EQ(0x1001, self->poked);
1552 }
1553 
1554 TEST_F(TRACE_poke, getpid_runs_normally)
1555 {
1556 	long ret;
1557 
1558 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1559 	ASSERT_EQ(0, ret);
1560 
1561 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1562 	ASSERT_EQ(0, ret);
1563 
1564 	EXPECT_EQ(0, self->poked);
1565 	EXPECT_NE(0, syscall(__NR_getpid));
1566 	EXPECT_EQ(0, self->poked);
1567 }
1568 
1569 #if defined(__x86_64__)
1570 # define ARCH_REGS	struct user_regs_struct
1571 # define SYSCALL_NUM	orig_rax
1572 # define SYSCALL_RET	rax
1573 #elif defined(__i386__)
1574 # define ARCH_REGS	struct user_regs_struct
1575 # define SYSCALL_NUM	orig_eax
1576 # define SYSCALL_RET	eax
1577 #elif defined(__arm__)
1578 # define ARCH_REGS	struct pt_regs
1579 # define SYSCALL_NUM	ARM_r7
1580 # define SYSCALL_RET	ARM_r0
1581 #elif defined(__aarch64__)
1582 # define ARCH_REGS	struct user_pt_regs
1583 # define SYSCALL_NUM	regs[8]
1584 # define SYSCALL_RET	regs[0]
1585 #elif defined(__hppa__)
1586 # define ARCH_REGS	struct user_regs_struct
1587 # define SYSCALL_NUM	gr[20]
1588 # define SYSCALL_RET	gr[28]
1589 #elif defined(__powerpc__)
1590 # define ARCH_REGS	struct pt_regs
1591 # define SYSCALL_NUM	gpr[0]
1592 # define SYSCALL_RET	gpr[3]
1593 #elif defined(__s390__)
1594 # define ARCH_REGS     s390_regs
1595 # define SYSCALL_NUM   gprs[2]
1596 # define SYSCALL_RET   gprs[2]
1597 #elif defined(__mips__)
1598 # define ARCH_REGS	struct pt_regs
1599 # define SYSCALL_NUM	regs[2]
1600 # define SYSCALL_SYSCALL_NUM regs[4]
1601 # define SYSCALL_RET	regs[2]
1602 # define SYSCALL_NUM_RET_SHARE_REG
1603 #else
1604 # error "Do not know how to find your architecture's registers and syscalls"
1605 #endif
1606 
1607 /* When the syscall return can't be changed, stub out the tests for it. */
1608 #ifdef SYSCALL_NUM_RET_SHARE_REG
1609 # define EXPECT_SYSCALL_RETURN(val, action)	EXPECT_EQ(-1, action)
1610 #else
1611 # define EXPECT_SYSCALL_RETURN(val, action)		\
1612 	do {						\
1613 		errno = 0;				\
1614 		if (val < 0) {				\
1615 			EXPECT_EQ(-1, action);		\
1616 			EXPECT_EQ(-(val), errno);	\
1617 		} else {				\
1618 			EXPECT_EQ(val, action);		\
1619 		}					\
1620 	} while (0)
1621 #endif
1622 
1623 /* Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for
1624  * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux).
1625  */
1626 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__)
1627 #define HAVE_GETREGS
1628 #endif
1629 
1630 /* Architecture-specific syscall fetching routine. */
1631 int get_syscall(struct __test_metadata *_metadata, pid_t tracee)
1632 {
1633 	ARCH_REGS regs;
1634 #ifdef HAVE_GETREGS
1635 	EXPECT_EQ(0, ptrace(PTRACE_GETREGS, tracee, 0, &regs)) {
1636 		TH_LOG("PTRACE_GETREGS failed");
1637 		return -1;
1638 	}
1639 #else
1640 	struct iovec iov;
1641 
1642 	iov.iov_base = &regs;
1643 	iov.iov_len = sizeof(regs);
1644 	EXPECT_EQ(0, ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov)) {
1645 		TH_LOG("PTRACE_GETREGSET failed");
1646 		return -1;
1647 	}
1648 #endif
1649 
1650 #if defined(__mips__)
1651 	if (regs.SYSCALL_NUM == __NR_O32_Linux)
1652 		return regs.SYSCALL_SYSCALL_NUM;
1653 #endif
1654 	return regs.SYSCALL_NUM;
1655 }
1656 
1657 /* Architecture-specific syscall changing routine. */
1658 void change_syscall(struct __test_metadata *_metadata,
1659 		    pid_t tracee, int syscall, int result)
1660 {
1661 	int ret;
1662 	ARCH_REGS regs;
1663 #ifdef HAVE_GETREGS
1664 	ret = ptrace(PTRACE_GETREGS, tracee, 0, &regs);
1665 #else
1666 	struct iovec iov;
1667 	iov.iov_base = &regs;
1668 	iov.iov_len = sizeof(regs);
1669 	ret = ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov);
1670 #endif
1671 	EXPECT_EQ(0, ret) {}
1672 
1673 #if defined(__x86_64__) || defined(__i386__) || defined(__powerpc__) || \
1674     defined(__s390__) || defined(__hppa__)
1675 	{
1676 		regs.SYSCALL_NUM = syscall;
1677 	}
1678 #elif defined(__mips__)
1679 	{
1680 		if (regs.SYSCALL_NUM == __NR_O32_Linux)
1681 			regs.SYSCALL_SYSCALL_NUM = syscall;
1682 		else
1683 			regs.SYSCALL_NUM = syscall;
1684 	}
1685 
1686 #elif defined(__arm__)
1687 # ifndef PTRACE_SET_SYSCALL
1688 #  define PTRACE_SET_SYSCALL   23
1689 # endif
1690 	{
1691 		ret = ptrace(PTRACE_SET_SYSCALL, tracee, NULL, syscall);
1692 		EXPECT_EQ(0, ret);
1693 	}
1694 
1695 #elif defined(__aarch64__)
1696 # ifndef NT_ARM_SYSTEM_CALL
1697 #  define NT_ARM_SYSTEM_CALL 0x404
1698 # endif
1699 	{
1700 		iov.iov_base = &syscall;
1701 		iov.iov_len = sizeof(syscall);
1702 		ret = ptrace(PTRACE_SETREGSET, tracee, NT_ARM_SYSTEM_CALL,
1703 			     &iov);
1704 		EXPECT_EQ(0, ret);
1705 	}
1706 
1707 #else
1708 	ASSERT_EQ(1, 0) {
1709 		TH_LOG("How is the syscall changed on this architecture?");
1710 	}
1711 #endif
1712 
1713 	/* If syscall is skipped, change return value. */
1714 	if (syscall == -1)
1715 #ifdef SYSCALL_NUM_RET_SHARE_REG
1716 		TH_LOG("Can't modify syscall return on this architecture");
1717 #else
1718 		regs.SYSCALL_RET = result;
1719 #endif
1720 
1721 #ifdef HAVE_GETREGS
1722 	ret = ptrace(PTRACE_SETREGS, tracee, 0, &regs);
1723 #else
1724 	iov.iov_base = &regs;
1725 	iov.iov_len = sizeof(regs);
1726 	ret = ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &iov);
1727 #endif
1728 	EXPECT_EQ(0, ret);
1729 }
1730 
1731 void tracer_syscall(struct __test_metadata *_metadata, pid_t tracee,
1732 		    int status, void *args)
1733 {
1734 	int ret;
1735 	unsigned long msg;
1736 
1737 	/* Make sure we got the right message. */
1738 	ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1739 	EXPECT_EQ(0, ret);
1740 
1741 	/* Validate and take action on expected syscalls. */
1742 	switch (msg) {
1743 	case 0x1002:
1744 		/* change getpid to getppid. */
1745 		EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee));
1746 		change_syscall(_metadata, tracee, __NR_getppid, 0);
1747 		break;
1748 	case 0x1003:
1749 		/* skip gettid with valid return code. */
1750 		EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee));
1751 		change_syscall(_metadata, tracee, -1, 45000);
1752 		break;
1753 	case 0x1004:
1754 		/* skip openat with error. */
1755 		EXPECT_EQ(__NR_openat, get_syscall(_metadata, tracee));
1756 		change_syscall(_metadata, tracee, -1, -ESRCH);
1757 		break;
1758 	case 0x1005:
1759 		/* do nothing (allow getppid) */
1760 		EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee));
1761 		break;
1762 	default:
1763 		EXPECT_EQ(0, msg) {
1764 			TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg);
1765 			kill(tracee, SIGKILL);
1766 		}
1767 	}
1768 
1769 }
1770 
1771 void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee,
1772 		   int status, void *args)
1773 {
1774 	int ret, nr;
1775 	unsigned long msg;
1776 	static bool entry;
1777 
1778 	/*
1779 	 * The traditional way to tell PTRACE_SYSCALL entry/exit
1780 	 * is by counting.
1781 	 */
1782 	entry = !entry;
1783 
1784 	/* Make sure we got an appropriate message. */
1785 	ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1786 	EXPECT_EQ(0, ret);
1787 	EXPECT_EQ(entry ? PTRACE_EVENTMSG_SYSCALL_ENTRY
1788 			: PTRACE_EVENTMSG_SYSCALL_EXIT, msg);
1789 
1790 	if (!entry)
1791 		return;
1792 
1793 	nr = get_syscall(_metadata, tracee);
1794 
1795 	if (nr == __NR_getpid)
1796 		change_syscall(_metadata, tracee, __NR_getppid, 0);
1797 	if (nr == __NR_gettid)
1798 		change_syscall(_metadata, tracee, -1, 45000);
1799 	if (nr == __NR_openat)
1800 		change_syscall(_metadata, tracee, -1, -ESRCH);
1801 }
1802 
1803 FIXTURE_DATA(TRACE_syscall) {
1804 	struct sock_fprog prog;
1805 	pid_t tracer, mytid, mypid, parent;
1806 };
1807 
1808 FIXTURE_SETUP(TRACE_syscall)
1809 {
1810 	struct sock_filter filter[] = {
1811 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1812 			offsetof(struct seccomp_data, nr)),
1813 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
1814 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002),
1815 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1),
1816 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003),
1817 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_openat, 0, 1),
1818 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004),
1819 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1820 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1005),
1821 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1822 	};
1823 
1824 	memset(&self->prog, 0, sizeof(self->prog));
1825 	self->prog.filter = malloc(sizeof(filter));
1826 	ASSERT_NE(NULL, self->prog.filter);
1827 	memcpy(self->prog.filter, filter, sizeof(filter));
1828 	self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1829 
1830 	/* Prepare some testable syscall results. */
1831 	self->mytid = syscall(__NR_gettid);
1832 	ASSERT_GT(self->mytid, 0);
1833 	ASSERT_NE(self->mytid, 1) {
1834 		TH_LOG("Running this test as init is not supported. :)");
1835 	}
1836 
1837 	self->mypid = getpid();
1838 	ASSERT_GT(self->mypid, 0);
1839 	ASSERT_EQ(self->mytid, self->mypid);
1840 
1841 	self->parent = getppid();
1842 	ASSERT_GT(self->parent, 0);
1843 	ASSERT_NE(self->parent, self->mypid);
1844 
1845 	/* Launch tracer. */
1846 	self->tracer = setup_trace_fixture(_metadata, tracer_syscall, NULL,
1847 					   false);
1848 }
1849 
1850 FIXTURE_TEARDOWN(TRACE_syscall)
1851 {
1852 	teardown_trace_fixture(_metadata, self->tracer);
1853 	if (self->prog.filter)
1854 		free(self->prog.filter);
1855 }
1856 
1857 TEST_F(TRACE_syscall, ptrace_syscall_redirected)
1858 {
1859 	/* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1860 	teardown_trace_fixture(_metadata, self->tracer);
1861 	self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1862 					   true);
1863 
1864 	/* Tracer will redirect getpid to getppid. */
1865 	EXPECT_NE(self->mypid, syscall(__NR_getpid));
1866 }
1867 
1868 TEST_F(TRACE_syscall, ptrace_syscall_errno)
1869 {
1870 	/* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1871 	teardown_trace_fixture(_metadata, self->tracer);
1872 	self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1873 					   true);
1874 
1875 	/* Tracer should skip the open syscall, resulting in ESRCH. */
1876 	EXPECT_SYSCALL_RETURN(-ESRCH, syscall(__NR_openat));
1877 }
1878 
1879 TEST_F(TRACE_syscall, ptrace_syscall_faked)
1880 {
1881 	/* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1882 	teardown_trace_fixture(_metadata, self->tracer);
1883 	self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1884 					   true);
1885 
1886 	/* Tracer should skip the gettid syscall, resulting fake pid. */
1887 	EXPECT_SYSCALL_RETURN(45000, syscall(__NR_gettid));
1888 }
1889 
1890 TEST_F(TRACE_syscall, syscall_allowed)
1891 {
1892 	long ret;
1893 
1894 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1895 	ASSERT_EQ(0, ret);
1896 
1897 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1898 	ASSERT_EQ(0, ret);
1899 
1900 	/* getppid works as expected (no changes). */
1901 	EXPECT_EQ(self->parent, syscall(__NR_getppid));
1902 	EXPECT_NE(self->mypid, syscall(__NR_getppid));
1903 }
1904 
1905 TEST_F(TRACE_syscall, syscall_redirected)
1906 {
1907 	long ret;
1908 
1909 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1910 	ASSERT_EQ(0, ret);
1911 
1912 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1913 	ASSERT_EQ(0, ret);
1914 
1915 	/* getpid has been redirected to getppid as expected. */
1916 	EXPECT_EQ(self->parent, syscall(__NR_getpid));
1917 	EXPECT_NE(self->mypid, syscall(__NR_getpid));
1918 }
1919 
1920 TEST_F(TRACE_syscall, syscall_errno)
1921 {
1922 	long ret;
1923 
1924 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1925 	ASSERT_EQ(0, ret);
1926 
1927 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1928 	ASSERT_EQ(0, ret);
1929 
1930 	/* openat has been skipped and an errno return. */
1931 	EXPECT_SYSCALL_RETURN(-ESRCH, syscall(__NR_openat));
1932 }
1933 
1934 TEST_F(TRACE_syscall, syscall_faked)
1935 {
1936 	long ret;
1937 
1938 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1939 	ASSERT_EQ(0, ret);
1940 
1941 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1942 	ASSERT_EQ(0, ret);
1943 
1944 	/* gettid has been skipped and an altered return value stored. */
1945 	EXPECT_SYSCALL_RETURN(45000, syscall(__NR_gettid));
1946 }
1947 
1948 TEST_F(TRACE_syscall, skip_after_RET_TRACE)
1949 {
1950 	struct sock_filter filter[] = {
1951 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1952 			offsetof(struct seccomp_data, nr)),
1953 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1954 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
1955 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1956 	};
1957 	struct sock_fprog prog = {
1958 		.len = (unsigned short)ARRAY_SIZE(filter),
1959 		.filter = filter,
1960 	};
1961 	long ret;
1962 
1963 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1964 	ASSERT_EQ(0, ret);
1965 
1966 	/* Install fixture filter. */
1967 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1968 	ASSERT_EQ(0, ret);
1969 
1970 	/* Install "errno on getppid" filter. */
1971 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1972 	ASSERT_EQ(0, ret);
1973 
1974 	/* Tracer will redirect getpid to getppid, and we should see EPERM. */
1975 	errno = 0;
1976 	EXPECT_EQ(-1, syscall(__NR_getpid));
1977 	EXPECT_EQ(EPERM, errno);
1978 }
1979 
1980 TEST_F_SIGNAL(TRACE_syscall, kill_after_RET_TRACE, SIGSYS)
1981 {
1982 	struct sock_filter filter[] = {
1983 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1984 			offsetof(struct seccomp_data, nr)),
1985 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1986 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1987 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1988 	};
1989 	struct sock_fprog prog = {
1990 		.len = (unsigned short)ARRAY_SIZE(filter),
1991 		.filter = filter,
1992 	};
1993 	long ret;
1994 
1995 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1996 	ASSERT_EQ(0, ret);
1997 
1998 	/* Install fixture filter. */
1999 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
2000 	ASSERT_EQ(0, ret);
2001 
2002 	/* Install "death on getppid" filter. */
2003 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2004 	ASSERT_EQ(0, ret);
2005 
2006 	/* Tracer will redirect getpid to getppid, and we should die. */
2007 	EXPECT_NE(self->mypid, syscall(__NR_getpid));
2008 }
2009 
2010 TEST_F(TRACE_syscall, skip_after_ptrace)
2011 {
2012 	struct sock_filter filter[] = {
2013 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2014 			offsetof(struct seccomp_data, nr)),
2015 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2016 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
2017 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2018 	};
2019 	struct sock_fprog prog = {
2020 		.len = (unsigned short)ARRAY_SIZE(filter),
2021 		.filter = filter,
2022 	};
2023 	long ret;
2024 
2025 	/* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
2026 	teardown_trace_fixture(_metadata, self->tracer);
2027 	self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
2028 					   true);
2029 
2030 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2031 	ASSERT_EQ(0, ret);
2032 
2033 	/* Install "errno on getppid" filter. */
2034 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2035 	ASSERT_EQ(0, ret);
2036 
2037 	/* Tracer will redirect getpid to getppid, and we should see EPERM. */
2038 	EXPECT_EQ(-1, syscall(__NR_getpid));
2039 	EXPECT_EQ(EPERM, errno);
2040 }
2041 
2042 TEST_F_SIGNAL(TRACE_syscall, kill_after_ptrace, SIGSYS)
2043 {
2044 	struct sock_filter filter[] = {
2045 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2046 			offsetof(struct seccomp_data, nr)),
2047 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2048 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2049 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2050 	};
2051 	struct sock_fprog prog = {
2052 		.len = (unsigned short)ARRAY_SIZE(filter),
2053 		.filter = filter,
2054 	};
2055 	long ret;
2056 
2057 	/* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
2058 	teardown_trace_fixture(_metadata, self->tracer);
2059 	self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
2060 					   true);
2061 
2062 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2063 	ASSERT_EQ(0, ret);
2064 
2065 	/* Install "death on getppid" filter. */
2066 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2067 	ASSERT_EQ(0, ret);
2068 
2069 	/* Tracer will redirect getpid to getppid, and we should die. */
2070 	EXPECT_NE(self->mypid, syscall(__NR_getpid));
2071 }
2072 
2073 TEST(seccomp_syscall)
2074 {
2075 	struct sock_filter filter[] = {
2076 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2077 	};
2078 	struct sock_fprog prog = {
2079 		.len = (unsigned short)ARRAY_SIZE(filter),
2080 		.filter = filter,
2081 	};
2082 	long ret;
2083 
2084 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2085 	ASSERT_EQ(0, ret) {
2086 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2087 	}
2088 
2089 	/* Reject insane operation. */
2090 	ret = seccomp(-1, 0, &prog);
2091 	ASSERT_NE(ENOSYS, errno) {
2092 		TH_LOG("Kernel does not support seccomp syscall!");
2093 	}
2094 	EXPECT_EQ(EINVAL, errno) {
2095 		TH_LOG("Did not reject crazy op value!");
2096 	}
2097 
2098 	/* Reject strict with flags or pointer. */
2099 	ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL);
2100 	EXPECT_EQ(EINVAL, errno) {
2101 		TH_LOG("Did not reject mode strict with flags!");
2102 	}
2103 	ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog);
2104 	EXPECT_EQ(EINVAL, errno) {
2105 		TH_LOG("Did not reject mode strict with uargs!");
2106 	}
2107 
2108 	/* Reject insane args for filter. */
2109 	ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog);
2110 	EXPECT_EQ(EINVAL, errno) {
2111 		TH_LOG("Did not reject crazy filter flags!");
2112 	}
2113 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL);
2114 	EXPECT_EQ(EFAULT, errno) {
2115 		TH_LOG("Did not reject NULL filter!");
2116 	}
2117 
2118 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2119 	EXPECT_EQ(0, errno) {
2120 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s",
2121 			strerror(errno));
2122 	}
2123 }
2124 
2125 TEST(seccomp_syscall_mode_lock)
2126 {
2127 	struct sock_filter filter[] = {
2128 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2129 	};
2130 	struct sock_fprog prog = {
2131 		.len = (unsigned short)ARRAY_SIZE(filter),
2132 		.filter = filter,
2133 	};
2134 	long ret;
2135 
2136 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2137 	ASSERT_EQ(0, ret) {
2138 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2139 	}
2140 
2141 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2142 	ASSERT_NE(ENOSYS, errno) {
2143 		TH_LOG("Kernel does not support seccomp syscall!");
2144 	}
2145 	EXPECT_EQ(0, ret) {
2146 		TH_LOG("Could not install filter!");
2147 	}
2148 
2149 	/* Make sure neither entry point will switch to strict. */
2150 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0);
2151 	EXPECT_EQ(EINVAL, errno) {
2152 		TH_LOG("Switched to mode strict!");
2153 	}
2154 
2155 	ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL);
2156 	EXPECT_EQ(EINVAL, errno) {
2157 		TH_LOG("Switched to mode strict!");
2158 	}
2159 }
2160 
2161 /*
2162  * Test detection of known and unknown filter flags. Userspace needs to be able
2163  * to check if a filter flag is supported by the current kernel and a good way
2164  * of doing that is by attempting to enter filter mode, with the flag bit in
2165  * question set, and a NULL pointer for the _args_ parameter. EFAULT indicates
2166  * that the flag is valid and EINVAL indicates that the flag is invalid.
2167  */
2168 TEST(detect_seccomp_filter_flags)
2169 {
2170 	unsigned int flags[] = { SECCOMP_FILTER_FLAG_TSYNC,
2171 				 SECCOMP_FILTER_FLAG_LOG,
2172 				 SECCOMP_FILTER_FLAG_SPEC_ALLOW,
2173 				 SECCOMP_FILTER_FLAG_NEW_LISTENER };
2174 	unsigned int exclusive[] = {
2175 				SECCOMP_FILTER_FLAG_TSYNC,
2176 				SECCOMP_FILTER_FLAG_NEW_LISTENER };
2177 	unsigned int flag, all_flags, exclusive_mask;
2178 	int i;
2179 	long ret;
2180 
2181 	/* Test detection of individual known-good filter flags */
2182 	for (i = 0, all_flags = 0; i < ARRAY_SIZE(flags); i++) {
2183 		int bits = 0;
2184 
2185 		flag = flags[i];
2186 		/* Make sure the flag is a single bit! */
2187 		while (flag) {
2188 			if (flag & 0x1)
2189 				bits ++;
2190 			flag >>= 1;
2191 		}
2192 		ASSERT_EQ(1, bits);
2193 		flag = flags[i];
2194 
2195 		ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2196 		ASSERT_NE(ENOSYS, errno) {
2197 			TH_LOG("Kernel does not support seccomp syscall!");
2198 		}
2199 		EXPECT_EQ(-1, ret);
2200 		EXPECT_EQ(EFAULT, errno) {
2201 			TH_LOG("Failed to detect that a known-good filter flag (0x%X) is supported!",
2202 			       flag);
2203 		}
2204 
2205 		all_flags |= flag;
2206 	}
2207 
2208 	/*
2209 	 * Test detection of all known-good filter flags combined. But
2210 	 * for the exclusive flags we need to mask them out and try them
2211 	 * individually for the "all flags" testing.
2212 	 */
2213 	exclusive_mask = 0;
2214 	for (i = 0; i < ARRAY_SIZE(exclusive); i++)
2215 		exclusive_mask |= exclusive[i];
2216 	for (i = 0; i < ARRAY_SIZE(exclusive); i++) {
2217 		flag = all_flags & ~exclusive_mask;
2218 		flag |= exclusive[i];
2219 
2220 		ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2221 		EXPECT_EQ(-1, ret);
2222 		EXPECT_EQ(EFAULT, errno) {
2223 			TH_LOG("Failed to detect that all known-good filter flags (0x%X) are supported!",
2224 			       flag);
2225 		}
2226 	}
2227 
2228 	/* Test detection of an unknown filter flags, without exclusives. */
2229 	flag = -1;
2230 	flag &= ~exclusive_mask;
2231 	ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2232 	EXPECT_EQ(-1, ret);
2233 	EXPECT_EQ(EINVAL, errno) {
2234 		TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported!",
2235 		       flag);
2236 	}
2237 
2238 	/*
2239 	 * Test detection of an unknown filter flag that may simply need to be
2240 	 * added to this test
2241 	 */
2242 	flag = flags[ARRAY_SIZE(flags) - 1] << 1;
2243 	ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2244 	EXPECT_EQ(-1, ret);
2245 	EXPECT_EQ(EINVAL, errno) {
2246 		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?",
2247 		       flag);
2248 	}
2249 }
2250 
2251 TEST(TSYNC_first)
2252 {
2253 	struct sock_filter filter[] = {
2254 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2255 	};
2256 	struct sock_fprog prog = {
2257 		.len = (unsigned short)ARRAY_SIZE(filter),
2258 		.filter = filter,
2259 	};
2260 	long ret;
2261 
2262 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2263 	ASSERT_EQ(0, ret) {
2264 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2265 	}
2266 
2267 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2268 		      &prog);
2269 	ASSERT_NE(ENOSYS, errno) {
2270 		TH_LOG("Kernel does not support seccomp syscall!");
2271 	}
2272 	EXPECT_EQ(0, ret) {
2273 		TH_LOG("Could not install initial filter with TSYNC!");
2274 	}
2275 }
2276 
2277 #define TSYNC_SIBLINGS 2
2278 struct tsync_sibling {
2279 	pthread_t tid;
2280 	pid_t system_tid;
2281 	sem_t *started;
2282 	pthread_cond_t *cond;
2283 	pthread_mutex_t *mutex;
2284 	int diverge;
2285 	int num_waits;
2286 	struct sock_fprog *prog;
2287 	struct __test_metadata *metadata;
2288 };
2289 
2290 /*
2291  * To avoid joining joined threads (which is not allowed by Bionic),
2292  * make sure we both successfully join and clear the tid to skip a
2293  * later join attempt during fixture teardown. Any remaining threads
2294  * will be directly killed during teardown.
2295  */
2296 #define PTHREAD_JOIN(tid, status)					\
2297 	do {								\
2298 		int _rc = pthread_join(tid, status);			\
2299 		if (_rc) {						\
2300 			TH_LOG("pthread_join of tid %u failed: %d\n",	\
2301 				(unsigned int)tid, _rc);		\
2302 		} else {						\
2303 			tid = 0;					\
2304 		}							\
2305 	} while (0)
2306 
2307 FIXTURE_DATA(TSYNC) {
2308 	struct sock_fprog root_prog, apply_prog;
2309 	struct tsync_sibling sibling[TSYNC_SIBLINGS];
2310 	sem_t started;
2311 	pthread_cond_t cond;
2312 	pthread_mutex_t mutex;
2313 	int sibling_count;
2314 };
2315 
2316 FIXTURE_SETUP(TSYNC)
2317 {
2318 	struct sock_filter root_filter[] = {
2319 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2320 	};
2321 	struct sock_filter apply_filter[] = {
2322 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2323 			offsetof(struct seccomp_data, nr)),
2324 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
2325 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2326 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2327 	};
2328 
2329 	memset(&self->root_prog, 0, sizeof(self->root_prog));
2330 	memset(&self->apply_prog, 0, sizeof(self->apply_prog));
2331 	memset(&self->sibling, 0, sizeof(self->sibling));
2332 	self->root_prog.filter = malloc(sizeof(root_filter));
2333 	ASSERT_NE(NULL, self->root_prog.filter);
2334 	memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter));
2335 	self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter);
2336 
2337 	self->apply_prog.filter = malloc(sizeof(apply_filter));
2338 	ASSERT_NE(NULL, self->apply_prog.filter);
2339 	memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter));
2340 	self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter);
2341 
2342 	self->sibling_count = 0;
2343 	pthread_mutex_init(&self->mutex, NULL);
2344 	pthread_cond_init(&self->cond, NULL);
2345 	sem_init(&self->started, 0, 0);
2346 	self->sibling[0].tid = 0;
2347 	self->sibling[0].cond = &self->cond;
2348 	self->sibling[0].started = &self->started;
2349 	self->sibling[0].mutex = &self->mutex;
2350 	self->sibling[0].diverge = 0;
2351 	self->sibling[0].num_waits = 1;
2352 	self->sibling[0].prog = &self->root_prog;
2353 	self->sibling[0].metadata = _metadata;
2354 	self->sibling[1].tid = 0;
2355 	self->sibling[1].cond = &self->cond;
2356 	self->sibling[1].started = &self->started;
2357 	self->sibling[1].mutex = &self->mutex;
2358 	self->sibling[1].diverge = 0;
2359 	self->sibling[1].prog = &self->root_prog;
2360 	self->sibling[1].num_waits = 1;
2361 	self->sibling[1].metadata = _metadata;
2362 }
2363 
2364 FIXTURE_TEARDOWN(TSYNC)
2365 {
2366 	int sib = 0;
2367 
2368 	if (self->root_prog.filter)
2369 		free(self->root_prog.filter);
2370 	if (self->apply_prog.filter)
2371 		free(self->apply_prog.filter);
2372 
2373 	for ( ; sib < self->sibling_count; ++sib) {
2374 		struct tsync_sibling *s = &self->sibling[sib];
2375 
2376 		if (!s->tid)
2377 			continue;
2378 		/*
2379 		 * If a thread is still running, it may be stuck, so hit
2380 		 * it over the head really hard.
2381 		 */
2382 		pthread_kill(s->tid, 9);
2383 	}
2384 	pthread_mutex_destroy(&self->mutex);
2385 	pthread_cond_destroy(&self->cond);
2386 	sem_destroy(&self->started);
2387 }
2388 
2389 void *tsync_sibling(void *data)
2390 {
2391 	long ret = 0;
2392 	struct tsync_sibling *me = data;
2393 
2394 	me->system_tid = syscall(__NR_gettid);
2395 
2396 	pthread_mutex_lock(me->mutex);
2397 	if (me->diverge) {
2398 		/* Just re-apply the root prog to fork the tree */
2399 		ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
2400 				me->prog, 0, 0);
2401 	}
2402 	sem_post(me->started);
2403 	/* Return outside of started so parent notices failures. */
2404 	if (ret) {
2405 		pthread_mutex_unlock(me->mutex);
2406 		return (void *)SIBLING_EXIT_FAILURE;
2407 	}
2408 	do {
2409 		pthread_cond_wait(me->cond, me->mutex);
2410 		me->num_waits = me->num_waits - 1;
2411 	} while (me->num_waits);
2412 	pthread_mutex_unlock(me->mutex);
2413 
2414 	ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0);
2415 	if (!ret)
2416 		return (void *)SIBLING_EXIT_NEWPRIVS;
2417 	read(0, NULL, 0);
2418 	return (void *)SIBLING_EXIT_UNKILLED;
2419 }
2420 
2421 void tsync_start_sibling(struct tsync_sibling *sibling)
2422 {
2423 	pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling);
2424 }
2425 
2426 TEST_F(TSYNC, siblings_fail_prctl)
2427 {
2428 	long ret;
2429 	void *status;
2430 	struct sock_filter filter[] = {
2431 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2432 			offsetof(struct seccomp_data, nr)),
2433 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
2434 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL),
2435 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2436 	};
2437 	struct sock_fprog prog = {
2438 		.len = (unsigned short)ARRAY_SIZE(filter),
2439 		.filter = filter,
2440 	};
2441 
2442 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2443 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2444 	}
2445 
2446 	/* Check prctl failure detection by requesting sib 0 diverge. */
2447 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2448 	ASSERT_NE(ENOSYS, errno) {
2449 		TH_LOG("Kernel does not support seccomp syscall!");
2450 	}
2451 	ASSERT_EQ(0, ret) {
2452 		TH_LOG("setting filter failed");
2453 	}
2454 
2455 	self->sibling[0].diverge = 1;
2456 	tsync_start_sibling(&self->sibling[0]);
2457 	tsync_start_sibling(&self->sibling[1]);
2458 
2459 	while (self->sibling_count < TSYNC_SIBLINGS) {
2460 		sem_wait(&self->started);
2461 		self->sibling_count++;
2462 	}
2463 
2464 	/* Signal the threads to clean up*/
2465 	pthread_mutex_lock(&self->mutex);
2466 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2467 		TH_LOG("cond broadcast non-zero");
2468 	}
2469 	pthread_mutex_unlock(&self->mutex);
2470 
2471 	/* Ensure diverging sibling failed to call prctl. */
2472 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2473 	EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status);
2474 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2475 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2476 }
2477 
2478 TEST_F(TSYNC, two_siblings_with_ancestor)
2479 {
2480 	long ret;
2481 	void *status;
2482 
2483 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2484 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2485 	}
2486 
2487 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2488 	ASSERT_NE(ENOSYS, errno) {
2489 		TH_LOG("Kernel does not support seccomp syscall!");
2490 	}
2491 	ASSERT_EQ(0, ret) {
2492 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2493 	}
2494 	tsync_start_sibling(&self->sibling[0]);
2495 	tsync_start_sibling(&self->sibling[1]);
2496 
2497 	while (self->sibling_count < TSYNC_SIBLINGS) {
2498 		sem_wait(&self->started);
2499 		self->sibling_count++;
2500 	}
2501 
2502 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2503 		      &self->apply_prog);
2504 	ASSERT_EQ(0, ret) {
2505 		TH_LOG("Could install filter on all threads!");
2506 	}
2507 	/* Tell the siblings to test the policy */
2508 	pthread_mutex_lock(&self->mutex);
2509 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2510 		TH_LOG("cond broadcast non-zero");
2511 	}
2512 	pthread_mutex_unlock(&self->mutex);
2513 	/* Ensure they are both killed and don't exit cleanly. */
2514 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2515 	EXPECT_EQ(0x0, (long)status);
2516 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2517 	EXPECT_EQ(0x0, (long)status);
2518 }
2519 
2520 TEST_F(TSYNC, two_sibling_want_nnp)
2521 {
2522 	void *status;
2523 
2524 	/* start siblings before any prctl() operations */
2525 	tsync_start_sibling(&self->sibling[0]);
2526 	tsync_start_sibling(&self->sibling[1]);
2527 	while (self->sibling_count < TSYNC_SIBLINGS) {
2528 		sem_wait(&self->started);
2529 		self->sibling_count++;
2530 	}
2531 
2532 	/* Tell the siblings to test no policy */
2533 	pthread_mutex_lock(&self->mutex);
2534 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2535 		TH_LOG("cond broadcast non-zero");
2536 	}
2537 	pthread_mutex_unlock(&self->mutex);
2538 
2539 	/* Ensure they are both upset about lacking nnp. */
2540 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2541 	EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2542 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2543 	EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2544 }
2545 
2546 TEST_F(TSYNC, two_siblings_with_no_filter)
2547 {
2548 	long ret;
2549 	void *status;
2550 
2551 	/* start siblings before any prctl() operations */
2552 	tsync_start_sibling(&self->sibling[0]);
2553 	tsync_start_sibling(&self->sibling[1]);
2554 	while (self->sibling_count < TSYNC_SIBLINGS) {
2555 		sem_wait(&self->started);
2556 		self->sibling_count++;
2557 	}
2558 
2559 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2560 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2561 	}
2562 
2563 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2564 		      &self->apply_prog);
2565 	ASSERT_NE(ENOSYS, errno) {
2566 		TH_LOG("Kernel does not support seccomp syscall!");
2567 	}
2568 	ASSERT_EQ(0, ret) {
2569 		TH_LOG("Could install filter on all threads!");
2570 	}
2571 
2572 	/* Tell the siblings to test the policy */
2573 	pthread_mutex_lock(&self->mutex);
2574 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2575 		TH_LOG("cond broadcast non-zero");
2576 	}
2577 	pthread_mutex_unlock(&self->mutex);
2578 
2579 	/* Ensure they are both killed and don't exit cleanly. */
2580 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2581 	EXPECT_EQ(0x0, (long)status);
2582 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2583 	EXPECT_EQ(0x0, (long)status);
2584 }
2585 
2586 TEST_F(TSYNC, two_siblings_with_one_divergence)
2587 {
2588 	long ret;
2589 	void *status;
2590 
2591 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2592 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2593 	}
2594 
2595 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2596 	ASSERT_NE(ENOSYS, errno) {
2597 		TH_LOG("Kernel does not support seccomp syscall!");
2598 	}
2599 	ASSERT_EQ(0, ret) {
2600 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2601 	}
2602 	self->sibling[0].diverge = 1;
2603 	tsync_start_sibling(&self->sibling[0]);
2604 	tsync_start_sibling(&self->sibling[1]);
2605 
2606 	while (self->sibling_count < TSYNC_SIBLINGS) {
2607 		sem_wait(&self->started);
2608 		self->sibling_count++;
2609 	}
2610 
2611 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2612 		      &self->apply_prog);
2613 	ASSERT_EQ(self->sibling[0].system_tid, ret) {
2614 		TH_LOG("Did not fail on diverged sibling.");
2615 	}
2616 
2617 	/* Wake the threads */
2618 	pthread_mutex_lock(&self->mutex);
2619 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2620 		TH_LOG("cond broadcast non-zero");
2621 	}
2622 	pthread_mutex_unlock(&self->mutex);
2623 
2624 	/* Ensure they are both unkilled. */
2625 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2626 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2627 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2628 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2629 }
2630 
2631 TEST_F(TSYNC, two_siblings_not_under_filter)
2632 {
2633 	long ret, sib;
2634 	void *status;
2635 	struct timespec delay = { .tv_nsec = 100000000 };
2636 
2637 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2638 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2639 	}
2640 
2641 	/*
2642 	 * Sibling 0 will have its own seccomp policy
2643 	 * and Sibling 1 will not be under seccomp at
2644 	 * all. Sibling 1 will enter seccomp and 0
2645 	 * will cause failure.
2646 	 */
2647 	self->sibling[0].diverge = 1;
2648 	tsync_start_sibling(&self->sibling[0]);
2649 	tsync_start_sibling(&self->sibling[1]);
2650 
2651 	while (self->sibling_count < TSYNC_SIBLINGS) {
2652 		sem_wait(&self->started);
2653 		self->sibling_count++;
2654 	}
2655 
2656 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2657 	ASSERT_NE(ENOSYS, errno) {
2658 		TH_LOG("Kernel does not support seccomp syscall!");
2659 	}
2660 	ASSERT_EQ(0, ret) {
2661 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2662 	}
2663 
2664 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2665 		      &self->apply_prog);
2666 	ASSERT_EQ(ret, self->sibling[0].system_tid) {
2667 		TH_LOG("Did not fail on diverged sibling.");
2668 	}
2669 	sib = 1;
2670 	if (ret == self->sibling[0].system_tid)
2671 		sib = 0;
2672 
2673 	pthread_mutex_lock(&self->mutex);
2674 
2675 	/* Increment the other siblings num_waits so we can clean up
2676 	 * the one we just saw.
2677 	 */
2678 	self->sibling[!sib].num_waits += 1;
2679 
2680 	/* Signal the thread to clean up*/
2681 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2682 		TH_LOG("cond broadcast non-zero");
2683 	}
2684 	pthread_mutex_unlock(&self->mutex);
2685 	PTHREAD_JOIN(self->sibling[sib].tid, &status);
2686 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2687 	/* Poll for actual task death. pthread_join doesn't guarantee it. */
2688 	while (!kill(self->sibling[sib].system_tid, 0))
2689 		nanosleep(&delay, NULL);
2690 	/* Switch to the remaining sibling */
2691 	sib = !sib;
2692 
2693 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2694 		      &self->apply_prog);
2695 	ASSERT_EQ(0, ret) {
2696 		TH_LOG("Expected the remaining sibling to sync");
2697 	};
2698 
2699 	pthread_mutex_lock(&self->mutex);
2700 
2701 	/* If remaining sibling didn't have a chance to wake up during
2702 	 * the first broadcast, manually reduce the num_waits now.
2703 	 */
2704 	if (self->sibling[sib].num_waits > 1)
2705 		self->sibling[sib].num_waits = 1;
2706 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2707 		TH_LOG("cond broadcast non-zero");
2708 	}
2709 	pthread_mutex_unlock(&self->mutex);
2710 	PTHREAD_JOIN(self->sibling[sib].tid, &status);
2711 	EXPECT_EQ(0, (long)status);
2712 	/* Poll for actual task death. pthread_join doesn't guarantee it. */
2713 	while (!kill(self->sibling[sib].system_tid, 0))
2714 		nanosleep(&delay, NULL);
2715 
2716 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2717 		      &self->apply_prog);
2718 	ASSERT_EQ(0, ret);  /* just us chickens */
2719 }
2720 
2721 /* Make sure restarted syscalls are seen directly as "restart_syscall". */
2722 TEST(syscall_restart)
2723 {
2724 	long ret;
2725 	unsigned long msg;
2726 	pid_t child_pid;
2727 	int pipefd[2];
2728 	int status;
2729 	siginfo_t info = { };
2730 	struct sock_filter filter[] = {
2731 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2732 			 offsetof(struct seccomp_data, nr)),
2733 
2734 #ifdef __NR_sigreturn
2735 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 6, 0),
2736 #endif
2737 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 5, 0),
2738 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 4, 0),
2739 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 3, 0),
2740 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 4, 0),
2741 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0),
2742 
2743 		/* Allow __NR_write for easy logging. */
2744 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1),
2745 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2746 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2747 		/* The nanosleep jump target. */
2748 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100),
2749 		/* The restart_syscall jump target. */
2750 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200),
2751 	};
2752 	struct sock_fprog prog = {
2753 		.len = (unsigned short)ARRAY_SIZE(filter),
2754 		.filter = filter,
2755 	};
2756 #if defined(__arm__)
2757 	struct utsname utsbuf;
2758 #endif
2759 
2760 	ASSERT_EQ(0, pipe(pipefd));
2761 
2762 	child_pid = fork();
2763 	ASSERT_LE(0, child_pid);
2764 	if (child_pid == 0) {
2765 		/* Child uses EXPECT not ASSERT to deliver status correctly. */
2766 		char buf = ' ';
2767 		struct timespec timeout = { };
2768 
2769 		/* Attach parent as tracer and stop. */
2770 		EXPECT_EQ(0, ptrace(PTRACE_TRACEME));
2771 		EXPECT_EQ(0, raise(SIGSTOP));
2772 
2773 		EXPECT_EQ(0, close(pipefd[1]));
2774 
2775 		EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2776 			TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2777 		}
2778 
2779 		ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2780 		EXPECT_EQ(0, ret) {
2781 			TH_LOG("Failed to install filter!");
2782 		}
2783 
2784 		EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2785 			TH_LOG("Failed to read() sync from parent");
2786 		}
2787 		EXPECT_EQ('.', buf) {
2788 			TH_LOG("Failed to get sync data from read()");
2789 		}
2790 
2791 		/* Start nanosleep to be interrupted. */
2792 		timeout.tv_sec = 1;
2793 		errno = 0;
2794 		EXPECT_EQ(0, nanosleep(&timeout, NULL)) {
2795 			TH_LOG("Call to nanosleep() failed (errno %d)", errno);
2796 		}
2797 
2798 		/* Read final sync from parent. */
2799 		EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2800 			TH_LOG("Failed final read() from parent");
2801 		}
2802 		EXPECT_EQ('!', buf) {
2803 			TH_LOG("Failed to get final data from read()");
2804 		}
2805 
2806 		/* Directly report the status of our test harness results. */
2807 		syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS
2808 						     : EXIT_FAILURE);
2809 	}
2810 	EXPECT_EQ(0, close(pipefd[0]));
2811 
2812 	/* Attach to child, setup options, and release. */
2813 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2814 	ASSERT_EQ(true, WIFSTOPPED(status));
2815 	ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL,
2816 			    PTRACE_O_TRACESECCOMP));
2817 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2818 	ASSERT_EQ(1, write(pipefd[1], ".", 1));
2819 
2820 	/* Wait for nanosleep() to start. */
2821 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2822 	ASSERT_EQ(true, WIFSTOPPED(status));
2823 	ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2824 	ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2825 	ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2826 	ASSERT_EQ(0x100, msg);
2827 	EXPECT_EQ(__NR_nanosleep, get_syscall(_metadata, child_pid));
2828 
2829 	/* Might as well check siginfo for sanity while we're here. */
2830 	ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2831 	ASSERT_EQ(SIGTRAP, info.si_signo);
2832 	ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code);
2833 	EXPECT_EQ(0, info.si_errno);
2834 	EXPECT_EQ(getuid(), info.si_uid);
2835 	/* Verify signal delivery came from child (seccomp-triggered). */
2836 	EXPECT_EQ(child_pid, info.si_pid);
2837 
2838 	/* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */
2839 	ASSERT_EQ(0, kill(child_pid, SIGSTOP));
2840 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2841 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2842 	ASSERT_EQ(true, WIFSTOPPED(status));
2843 	ASSERT_EQ(SIGSTOP, WSTOPSIG(status));
2844 	ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2845 	/*
2846 	 * There is no siginfo on SIGSTOP any more, so we can't verify
2847 	 * signal delivery came from parent now (getpid() == info.si_pid).
2848 	 * https://lkml.kernel.org/r/CAGXu5jJaZAOzP1qFz66tYrtbuywqb+UN2SOA1VLHpCCOiYvYeg@mail.gmail.com
2849 	 * At least verify the SIGSTOP via PTRACE_GETSIGINFO.
2850 	 */
2851 	EXPECT_EQ(SIGSTOP, info.si_signo);
2852 
2853 	/* Restart nanosleep with SIGCONT, which triggers restart_syscall. */
2854 	ASSERT_EQ(0, kill(child_pid, SIGCONT));
2855 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2856 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2857 	ASSERT_EQ(true, WIFSTOPPED(status));
2858 	ASSERT_EQ(SIGCONT, WSTOPSIG(status));
2859 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2860 
2861 	/* Wait for restart_syscall() to start. */
2862 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2863 	ASSERT_EQ(true, WIFSTOPPED(status));
2864 	ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2865 	ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2866 	ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2867 
2868 	ASSERT_EQ(0x200, msg);
2869 	ret = get_syscall(_metadata, child_pid);
2870 #if defined(__arm__)
2871 	/*
2872 	 * FIXME:
2873 	 * - native ARM registers do NOT expose true syscall.
2874 	 * - compat ARM registers on ARM64 DO expose true syscall.
2875 	 */
2876 	ASSERT_EQ(0, uname(&utsbuf));
2877 	if (strncmp(utsbuf.machine, "arm", 3) == 0) {
2878 		EXPECT_EQ(__NR_nanosleep, ret);
2879 	} else
2880 #endif
2881 	{
2882 		EXPECT_EQ(__NR_restart_syscall, ret);
2883 	}
2884 
2885 	/* Write again to end test. */
2886 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2887 	ASSERT_EQ(1, write(pipefd[1], "!", 1));
2888 	EXPECT_EQ(0, close(pipefd[1]));
2889 
2890 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2891 	if (WIFSIGNALED(status) || WEXITSTATUS(status))
2892 		_metadata->passed = 0;
2893 }
2894 
2895 TEST_SIGNAL(filter_flag_log, SIGSYS)
2896 {
2897 	struct sock_filter allow_filter[] = {
2898 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2899 	};
2900 	struct sock_filter kill_filter[] = {
2901 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2902 			offsetof(struct seccomp_data, nr)),
2903 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
2904 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2905 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2906 	};
2907 	struct sock_fprog allow_prog = {
2908 		.len = (unsigned short)ARRAY_SIZE(allow_filter),
2909 		.filter = allow_filter,
2910 	};
2911 	struct sock_fprog kill_prog = {
2912 		.len = (unsigned short)ARRAY_SIZE(kill_filter),
2913 		.filter = kill_filter,
2914 	};
2915 	long ret;
2916 	pid_t parent = getppid();
2917 
2918 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2919 	ASSERT_EQ(0, ret);
2920 
2921 	/* Verify that the FILTER_FLAG_LOG flag isn't accepted in strict mode */
2922 	ret = seccomp(SECCOMP_SET_MODE_STRICT, SECCOMP_FILTER_FLAG_LOG,
2923 		      &allow_prog);
2924 	ASSERT_NE(ENOSYS, errno) {
2925 		TH_LOG("Kernel does not support seccomp syscall!");
2926 	}
2927 	EXPECT_NE(0, ret) {
2928 		TH_LOG("Kernel accepted FILTER_FLAG_LOG flag in strict mode!");
2929 	}
2930 	EXPECT_EQ(EINVAL, errno) {
2931 		TH_LOG("Kernel returned unexpected errno for FILTER_FLAG_LOG flag in strict mode!");
2932 	}
2933 
2934 	/* Verify that a simple, permissive filter can be added with no flags */
2935 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog);
2936 	EXPECT_EQ(0, ret);
2937 
2938 	/* See if the same filter can be added with the FILTER_FLAG_LOG flag */
2939 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
2940 		      &allow_prog);
2941 	ASSERT_NE(EINVAL, errno) {
2942 		TH_LOG("Kernel does not support the FILTER_FLAG_LOG flag!");
2943 	}
2944 	EXPECT_EQ(0, ret);
2945 
2946 	/* Ensure that the kill filter works with the FILTER_FLAG_LOG flag */
2947 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
2948 		      &kill_prog);
2949 	EXPECT_EQ(0, ret);
2950 
2951 	EXPECT_EQ(parent, syscall(__NR_getppid));
2952 	/* getpid() should never return. */
2953 	EXPECT_EQ(0, syscall(__NR_getpid));
2954 }
2955 
2956 TEST(get_action_avail)
2957 {
2958 	__u32 actions[] = { SECCOMP_RET_KILL_THREAD, SECCOMP_RET_TRAP,
2959 			    SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE,
2960 			    SECCOMP_RET_LOG,   SECCOMP_RET_ALLOW };
2961 	__u32 unknown_action = 0x10000000U;
2962 	int i;
2963 	long ret;
2964 
2965 	ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[0]);
2966 	ASSERT_NE(ENOSYS, errno) {
2967 		TH_LOG("Kernel does not support seccomp syscall!");
2968 	}
2969 	ASSERT_NE(EINVAL, errno) {
2970 		TH_LOG("Kernel does not support SECCOMP_GET_ACTION_AVAIL operation!");
2971 	}
2972 	EXPECT_EQ(ret, 0);
2973 
2974 	for (i = 0; i < ARRAY_SIZE(actions); i++) {
2975 		ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[i]);
2976 		EXPECT_EQ(ret, 0) {
2977 			TH_LOG("Expected action (0x%X) not available!",
2978 			       actions[i]);
2979 		}
2980 	}
2981 
2982 	/* Check that an unknown action is handled properly (EOPNOTSUPP) */
2983 	ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &unknown_action);
2984 	EXPECT_EQ(ret, -1);
2985 	EXPECT_EQ(errno, EOPNOTSUPP);
2986 }
2987 
2988 TEST(get_metadata)
2989 {
2990 	pid_t pid;
2991 	int pipefd[2];
2992 	char buf;
2993 	struct seccomp_metadata md;
2994 	long ret;
2995 
2996 	/* Only real root can get metadata. */
2997 	if (geteuid()) {
2998 		XFAIL(return, "get_metadata requires real root");
2999 		return;
3000 	}
3001 
3002 	ASSERT_EQ(0, pipe(pipefd));
3003 
3004 	pid = fork();
3005 	ASSERT_GE(pid, 0);
3006 	if (pid == 0) {
3007 		struct sock_filter filter[] = {
3008 			BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3009 		};
3010 		struct sock_fprog prog = {
3011 			.len = (unsigned short)ARRAY_SIZE(filter),
3012 			.filter = filter,
3013 		};
3014 
3015 		/* one with log, one without */
3016 		EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER,
3017 				     SECCOMP_FILTER_FLAG_LOG, &prog));
3018 		EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog));
3019 
3020 		EXPECT_EQ(0, close(pipefd[0]));
3021 		ASSERT_EQ(1, write(pipefd[1], "1", 1));
3022 		ASSERT_EQ(0, close(pipefd[1]));
3023 
3024 		while (1)
3025 			sleep(100);
3026 	}
3027 
3028 	ASSERT_EQ(0, close(pipefd[1]));
3029 	ASSERT_EQ(1, read(pipefd[0], &buf, 1));
3030 
3031 	ASSERT_EQ(0, ptrace(PTRACE_ATTACH, pid));
3032 	ASSERT_EQ(pid, waitpid(pid, NULL, 0));
3033 
3034 	/* Past here must not use ASSERT or child process is never killed. */
3035 
3036 	md.filter_off = 0;
3037 	errno = 0;
3038 	ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md);
3039 	EXPECT_EQ(sizeof(md), ret) {
3040 		if (errno == EINVAL)
3041 			XFAIL(goto skip, "Kernel does not support PTRACE_SECCOMP_GET_METADATA (missing CONFIG_CHECKPOINT_RESTORE?)");
3042 	}
3043 
3044 	EXPECT_EQ(md.flags, SECCOMP_FILTER_FLAG_LOG);
3045 	EXPECT_EQ(md.filter_off, 0);
3046 
3047 	md.filter_off = 1;
3048 	ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md);
3049 	EXPECT_EQ(sizeof(md), ret);
3050 	EXPECT_EQ(md.flags, 0);
3051 	EXPECT_EQ(md.filter_off, 1);
3052 
3053 skip:
3054 	ASSERT_EQ(0, kill(pid, SIGKILL));
3055 }
3056 
3057 static int user_trap_syscall(int nr, unsigned int flags)
3058 {
3059 	struct sock_filter filter[] = {
3060 		BPF_STMT(BPF_LD+BPF_W+BPF_ABS,
3061 			offsetof(struct seccomp_data, nr)),
3062 		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, nr, 0, 1),
3063 		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_USER_NOTIF),
3064 		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW),
3065 	};
3066 
3067 	struct sock_fprog prog = {
3068 		.len = (unsigned short)ARRAY_SIZE(filter),
3069 		.filter = filter,
3070 	};
3071 
3072 	return seccomp(SECCOMP_SET_MODE_FILTER, flags, &prog);
3073 }
3074 
3075 #define USER_NOTIF_MAGIC 116983961184613L
3076 TEST(user_notification_basic)
3077 {
3078 	pid_t pid;
3079 	long ret;
3080 	int status, listener;
3081 	struct seccomp_notif req = {};
3082 	struct seccomp_notif_resp resp = {};
3083 	struct pollfd pollfd;
3084 
3085 	struct sock_filter filter[] = {
3086 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3087 	};
3088 	struct sock_fprog prog = {
3089 		.len = (unsigned short)ARRAY_SIZE(filter),
3090 		.filter = filter,
3091 	};
3092 
3093 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3094 	ASSERT_EQ(0, ret) {
3095 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3096 	}
3097 
3098 	pid = fork();
3099 	ASSERT_GE(pid, 0);
3100 
3101 	/* Check that we get -ENOSYS with no listener attached */
3102 	if (pid == 0) {
3103 		if (user_trap_syscall(__NR_getppid, 0) < 0)
3104 			exit(1);
3105 		ret = syscall(__NR_getppid);
3106 		exit(ret >= 0 || errno != ENOSYS);
3107 	}
3108 
3109 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3110 	EXPECT_EQ(true, WIFEXITED(status));
3111 	EXPECT_EQ(0, WEXITSTATUS(status));
3112 
3113 	/* Add some no-op filters for grins. */
3114 	EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3115 	EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3116 	EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3117 	EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3118 
3119 	/* Check that the basic notification machinery works */
3120 	listener = user_trap_syscall(__NR_getppid,
3121 				     SECCOMP_FILTER_FLAG_NEW_LISTENER);
3122 	ASSERT_GE(listener, 0);
3123 
3124 	/* Installing a second listener in the chain should EBUSY */
3125 	EXPECT_EQ(user_trap_syscall(__NR_getppid,
3126 				    SECCOMP_FILTER_FLAG_NEW_LISTENER),
3127 		  -1);
3128 	EXPECT_EQ(errno, EBUSY);
3129 
3130 	pid = fork();
3131 	ASSERT_GE(pid, 0);
3132 
3133 	if (pid == 0) {
3134 		ret = syscall(__NR_getppid);
3135 		exit(ret != USER_NOTIF_MAGIC);
3136 	}
3137 
3138 	pollfd.fd = listener;
3139 	pollfd.events = POLLIN | POLLOUT;
3140 
3141 	EXPECT_GT(poll(&pollfd, 1, -1), 0);
3142 	EXPECT_EQ(pollfd.revents, POLLIN);
3143 
3144 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3145 
3146 	pollfd.fd = listener;
3147 	pollfd.events = POLLIN | POLLOUT;
3148 
3149 	EXPECT_GT(poll(&pollfd, 1, -1), 0);
3150 	EXPECT_EQ(pollfd.revents, POLLOUT);
3151 
3152 	EXPECT_EQ(req.data.nr,  __NR_getppid);
3153 
3154 	resp.id = req.id;
3155 	resp.error = 0;
3156 	resp.val = USER_NOTIF_MAGIC;
3157 
3158 	/* check that we make sure flags == 0 */
3159 	resp.flags = 1;
3160 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3161 	EXPECT_EQ(errno, EINVAL);
3162 
3163 	resp.flags = 0;
3164 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3165 
3166 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3167 	EXPECT_EQ(true, WIFEXITED(status));
3168 	EXPECT_EQ(0, WEXITSTATUS(status));
3169 }
3170 
3171 TEST(user_notification_kill_in_middle)
3172 {
3173 	pid_t pid;
3174 	long ret;
3175 	int listener;
3176 	struct seccomp_notif req = {};
3177 	struct seccomp_notif_resp resp = {};
3178 
3179 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3180 	ASSERT_EQ(0, ret) {
3181 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3182 	}
3183 
3184 	listener = user_trap_syscall(__NR_getppid,
3185 				     SECCOMP_FILTER_FLAG_NEW_LISTENER);
3186 	ASSERT_GE(listener, 0);
3187 
3188 	/*
3189 	 * Check that nothing bad happens when we kill the task in the middle
3190 	 * of a syscall.
3191 	 */
3192 	pid = fork();
3193 	ASSERT_GE(pid, 0);
3194 
3195 	if (pid == 0) {
3196 		ret = syscall(__NR_getppid);
3197 		exit(ret != USER_NOTIF_MAGIC);
3198 	}
3199 
3200 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3201 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), 0);
3202 
3203 	EXPECT_EQ(kill(pid, SIGKILL), 0);
3204 	EXPECT_EQ(waitpid(pid, NULL, 0), pid);
3205 
3206 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), -1);
3207 
3208 	resp.id = req.id;
3209 	ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp);
3210 	EXPECT_EQ(ret, -1);
3211 	EXPECT_EQ(errno, ENOENT);
3212 }
3213 
3214 static int handled = -1;
3215 
3216 static void signal_handler(int signal)
3217 {
3218 	if (write(handled, "c", 1) != 1)
3219 		perror("write from signal");
3220 }
3221 
3222 TEST(user_notification_signal)
3223 {
3224 	pid_t pid;
3225 	long ret;
3226 	int status, listener, sk_pair[2];
3227 	struct seccomp_notif req = {};
3228 	struct seccomp_notif_resp resp = {};
3229 	char c;
3230 
3231 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3232 	ASSERT_EQ(0, ret) {
3233 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3234 	}
3235 
3236 	ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0);
3237 
3238 	listener = user_trap_syscall(__NR_gettid,
3239 				     SECCOMP_FILTER_FLAG_NEW_LISTENER);
3240 	ASSERT_GE(listener, 0);
3241 
3242 	pid = fork();
3243 	ASSERT_GE(pid, 0);
3244 
3245 	if (pid == 0) {
3246 		close(sk_pair[0]);
3247 		handled = sk_pair[1];
3248 		if (signal(SIGUSR1, signal_handler) == SIG_ERR) {
3249 			perror("signal");
3250 			exit(1);
3251 		}
3252 		/*
3253 		 * ERESTARTSYS behavior is a bit hard to test, because we need
3254 		 * to rely on a signal that has not yet been handled. Let's at
3255 		 * least check that the error code gets propagated through, and
3256 		 * hope that it doesn't break when there is actually a signal :)
3257 		 */
3258 		ret = syscall(__NR_gettid);
3259 		exit(!(ret == -1 && errno == 512));
3260 	}
3261 
3262 	close(sk_pair[1]);
3263 
3264 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3265 
3266 	EXPECT_EQ(kill(pid, SIGUSR1), 0);
3267 
3268 	/*
3269 	 * Make sure the signal really is delivered, which means we're not
3270 	 * stuck in the user notification code any more and the notification
3271 	 * should be dead.
3272 	 */
3273 	EXPECT_EQ(read(sk_pair[0], &c, 1), 1);
3274 
3275 	resp.id = req.id;
3276 	resp.error = -EPERM;
3277 	resp.val = 0;
3278 
3279 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3280 	EXPECT_EQ(errno, ENOENT);
3281 
3282 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3283 
3284 	resp.id = req.id;
3285 	resp.error = -512; /* -ERESTARTSYS */
3286 	resp.val = 0;
3287 
3288 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3289 
3290 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3291 	EXPECT_EQ(true, WIFEXITED(status));
3292 	EXPECT_EQ(0, WEXITSTATUS(status));
3293 }
3294 
3295 TEST(user_notification_closed_listener)
3296 {
3297 	pid_t pid;
3298 	long ret;
3299 	int status, listener;
3300 
3301 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3302 	ASSERT_EQ(0, ret) {
3303 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3304 	}
3305 
3306 	listener = user_trap_syscall(__NR_getppid,
3307 				     SECCOMP_FILTER_FLAG_NEW_LISTENER);
3308 	ASSERT_GE(listener, 0);
3309 
3310 	/*
3311 	 * Check that we get an ENOSYS when the listener is closed.
3312 	 */
3313 	pid = fork();
3314 	ASSERT_GE(pid, 0);
3315 	if (pid == 0) {
3316 		close(listener);
3317 		ret = syscall(__NR_getppid);
3318 		exit(ret != -1 && errno != ENOSYS);
3319 	}
3320 
3321 	close(listener);
3322 
3323 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3324 	EXPECT_EQ(true, WIFEXITED(status));
3325 	EXPECT_EQ(0, WEXITSTATUS(status));
3326 }
3327 
3328 /*
3329  * Check that a pid in a child namespace still shows up as valid in ours.
3330  */
3331 TEST(user_notification_child_pid_ns)
3332 {
3333 	pid_t pid;
3334 	int status, listener;
3335 	struct seccomp_notif req = {};
3336 	struct seccomp_notif_resp resp = {};
3337 
3338 	ASSERT_EQ(unshare(CLONE_NEWUSER | CLONE_NEWPID), 0);
3339 
3340 	listener = user_trap_syscall(__NR_getppid,
3341 				     SECCOMP_FILTER_FLAG_NEW_LISTENER);
3342 	ASSERT_GE(listener, 0);
3343 
3344 	pid = fork();
3345 	ASSERT_GE(pid, 0);
3346 
3347 	if (pid == 0)
3348 		exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3349 
3350 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3351 	EXPECT_EQ(req.pid, pid);
3352 
3353 	resp.id = req.id;
3354 	resp.error = 0;
3355 	resp.val = USER_NOTIF_MAGIC;
3356 
3357 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3358 
3359 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3360 	EXPECT_EQ(true, WIFEXITED(status));
3361 	EXPECT_EQ(0, WEXITSTATUS(status));
3362 	close(listener);
3363 }
3364 
3365 /*
3366  * Check that a pid in a sibling (i.e. unrelated) namespace shows up as 0, i.e.
3367  * invalid.
3368  */
3369 TEST(user_notification_sibling_pid_ns)
3370 {
3371 	pid_t pid, pid2;
3372 	int status, listener;
3373 	struct seccomp_notif req = {};
3374 	struct seccomp_notif_resp resp = {};
3375 
3376 	ASSERT_EQ(prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0), 0) {
3377 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3378 	}
3379 
3380 	listener = user_trap_syscall(__NR_getppid,
3381 				     SECCOMP_FILTER_FLAG_NEW_LISTENER);
3382 	ASSERT_GE(listener, 0);
3383 
3384 	pid = fork();
3385 	ASSERT_GE(pid, 0);
3386 
3387 	if (pid == 0) {
3388 		ASSERT_EQ(unshare(CLONE_NEWPID), 0);
3389 
3390 		pid2 = fork();
3391 		ASSERT_GE(pid2, 0);
3392 
3393 		if (pid2 == 0)
3394 			exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3395 
3396 		EXPECT_EQ(waitpid(pid2, &status, 0), pid2);
3397 		EXPECT_EQ(true, WIFEXITED(status));
3398 		EXPECT_EQ(0, WEXITSTATUS(status));
3399 		exit(WEXITSTATUS(status));
3400 	}
3401 
3402 	/* Create the sibling ns, and sibling in it. */
3403 	ASSERT_EQ(unshare(CLONE_NEWPID), 0);
3404 	ASSERT_EQ(errno, 0);
3405 
3406 	pid2 = fork();
3407 	ASSERT_GE(pid2, 0);
3408 
3409 	if (pid2 == 0) {
3410 		ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3411 		/*
3412 		 * The pid should be 0, i.e. the task is in some namespace that
3413 		 * we can't "see".
3414 		 */
3415 		EXPECT_EQ(req.pid, 0);
3416 
3417 		resp.id = req.id;
3418 		resp.error = 0;
3419 		resp.val = USER_NOTIF_MAGIC;
3420 
3421 		ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3422 		exit(0);
3423 	}
3424 
3425 	close(listener);
3426 
3427 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3428 	EXPECT_EQ(true, WIFEXITED(status));
3429 	EXPECT_EQ(0, WEXITSTATUS(status));
3430 
3431 	EXPECT_EQ(waitpid(pid2, &status, 0), pid2);
3432 	EXPECT_EQ(true, WIFEXITED(status));
3433 	EXPECT_EQ(0, WEXITSTATUS(status));
3434 }
3435 
3436 TEST(user_notification_fault_recv)
3437 {
3438 	pid_t pid;
3439 	int status, listener;
3440 	struct seccomp_notif req = {};
3441 	struct seccomp_notif_resp resp = {};
3442 
3443 	ASSERT_EQ(unshare(CLONE_NEWUSER), 0);
3444 
3445 	listener = user_trap_syscall(__NR_getppid,
3446 				     SECCOMP_FILTER_FLAG_NEW_LISTENER);
3447 	ASSERT_GE(listener, 0);
3448 
3449 	pid = fork();
3450 	ASSERT_GE(pid, 0);
3451 
3452 	if (pid == 0)
3453 		exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3454 
3455 	/* Do a bad recv() */
3456 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, NULL), -1);
3457 	EXPECT_EQ(errno, EFAULT);
3458 
3459 	/* We should still be able to receive this notification, though. */
3460 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3461 	EXPECT_EQ(req.pid, pid);
3462 
3463 	resp.id = req.id;
3464 	resp.error = 0;
3465 	resp.val = USER_NOTIF_MAGIC;
3466 
3467 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3468 
3469 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3470 	EXPECT_EQ(true, WIFEXITED(status));
3471 	EXPECT_EQ(0, WEXITSTATUS(status));
3472 }
3473 
3474 TEST(seccomp_get_notif_sizes)
3475 {
3476 	struct seccomp_notif_sizes sizes;
3477 
3478 	ASSERT_EQ(seccomp(SECCOMP_GET_NOTIF_SIZES, 0, &sizes), 0);
3479 	EXPECT_EQ(sizes.seccomp_notif, sizeof(struct seccomp_notif));
3480 	EXPECT_EQ(sizes.seccomp_notif_resp, sizeof(struct seccomp_notif_resp));
3481 }
3482 
3483 /*
3484  * TODO:
3485  * - add microbenchmarks
3486  * - expand NNP testing
3487  * - better arch-specific TRACE and TRAP handlers.
3488  * - endianness checking when appropriate
3489  * - 64-bit arg prodding
3490  * - arch value testing (x86 modes especially)
3491  * - verify that FILTER_FLAG_LOG filters generate log messages
3492  * - verify that RET_LOG generates log messages
3493  * - ...
3494  */
3495 
3496 TEST_HARNESS_MAIN
3497