xref: /linux/tools/testing/selftests/seccomp/seccomp_bpf.c (revision 56d8b784c56588cd40f98e4b1d4f6e29e3cb02b8)
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 <limits.h>
39 #include <linux/elf.h>
40 #include <sys/uio.h>
41 #include <sys/utsname.h>
42 #include <sys/fcntl.h>
43 #include <sys/mman.h>
44 #include <sys/times.h>
45 #include <sys/socket.h>
46 #include <sys/ioctl.h>
47 #include <linux/kcmp.h>
48 #include <sys/resource.h>
49 #include <sys/capability.h>
50 
51 #include <unistd.h>
52 #include <sys/syscall.h>
53 #include <poll.h>
54 
55 #include "../kselftest_harness.h"
56 #include "../clone3/clone3_selftests.h"
57 
58 /* Attempt to de-conflict with the selftests tree. */
59 #ifndef SKIP
60 #define SKIP(s, ...)	XFAIL(s, ##__VA_ARGS__)
61 #endif
62 
63 #ifndef MIN
64 #define MIN(X, Y) ((X) < (Y) ? (X) : (Y))
65 #endif
66 
67 #ifndef PR_SET_PTRACER
68 # define PR_SET_PTRACER 0x59616d61
69 #endif
70 
71 #ifndef PR_SET_NO_NEW_PRIVS
72 #define PR_SET_NO_NEW_PRIVS 38
73 #define PR_GET_NO_NEW_PRIVS 39
74 #endif
75 
76 #ifndef PR_SECCOMP_EXT
77 #define PR_SECCOMP_EXT 43
78 #endif
79 
80 #ifndef SECCOMP_EXT_ACT
81 #define SECCOMP_EXT_ACT 1
82 #endif
83 
84 #ifndef SECCOMP_EXT_ACT_TSYNC
85 #define SECCOMP_EXT_ACT_TSYNC 1
86 #endif
87 
88 #ifndef SECCOMP_MODE_STRICT
89 #define SECCOMP_MODE_STRICT 1
90 #endif
91 
92 #ifndef SECCOMP_MODE_FILTER
93 #define SECCOMP_MODE_FILTER 2
94 #endif
95 
96 #ifndef SECCOMP_RET_ALLOW
97 struct seccomp_data {
98 	int nr;
99 	__u32 arch;
100 	__u64 instruction_pointer;
101 	__u64 args[6];
102 };
103 #endif
104 
105 #ifndef SECCOMP_RET_KILL_PROCESS
106 #define SECCOMP_RET_KILL_PROCESS 0x80000000U /* kill the process */
107 #define SECCOMP_RET_KILL_THREAD	 0x00000000U /* kill the thread */
108 #endif
109 #ifndef SECCOMP_RET_KILL
110 #define SECCOMP_RET_KILL	 SECCOMP_RET_KILL_THREAD
111 #define SECCOMP_RET_TRAP	 0x00030000U /* disallow and force a SIGSYS */
112 #define SECCOMP_RET_ERRNO	 0x00050000U /* returns an errno */
113 #define SECCOMP_RET_TRACE	 0x7ff00000U /* pass to a tracer or disallow */
114 #define SECCOMP_RET_ALLOW	 0x7fff0000U /* allow */
115 #endif
116 #ifndef SECCOMP_RET_LOG
117 #define SECCOMP_RET_LOG		 0x7ffc0000U /* allow after logging */
118 #endif
119 
120 #ifndef __NR_seccomp
121 # if defined(__i386__)
122 #  define __NR_seccomp 354
123 # elif defined(__x86_64__)
124 #  define __NR_seccomp 317
125 # elif defined(__arm__)
126 #  define __NR_seccomp 383
127 # elif defined(__aarch64__)
128 #  define __NR_seccomp 277
129 # elif defined(__riscv)
130 #  define __NR_seccomp 277
131 # elif defined(__csky__)
132 #  define __NR_seccomp 277
133 # elif defined(__loongarch__)
134 #  define __NR_seccomp 277
135 # elif defined(__hppa__)
136 #  define __NR_seccomp 338
137 # elif defined(__powerpc__)
138 #  define __NR_seccomp 358
139 # elif defined(__s390__)
140 #  define __NR_seccomp 348
141 # elif defined(__xtensa__)
142 #  define __NR_seccomp 337
143 # elif defined(__sh__)
144 #  define __NR_seccomp 372
145 # elif defined(__mc68000__)
146 #  define __NR_seccomp 380
147 # else
148 #  warning "seccomp syscall number unknown for this architecture"
149 #  define __NR_seccomp 0xffff
150 # endif
151 #endif
152 
153 #ifndef SECCOMP_SET_MODE_STRICT
154 #define SECCOMP_SET_MODE_STRICT 0
155 #endif
156 
157 #ifndef SECCOMP_SET_MODE_FILTER
158 #define SECCOMP_SET_MODE_FILTER 1
159 #endif
160 
161 #ifndef SECCOMP_GET_ACTION_AVAIL
162 #define SECCOMP_GET_ACTION_AVAIL 2
163 #endif
164 
165 #ifndef SECCOMP_GET_NOTIF_SIZES
166 #define SECCOMP_GET_NOTIF_SIZES 3
167 #endif
168 
169 #ifndef SECCOMP_FILTER_FLAG_TSYNC
170 #define SECCOMP_FILTER_FLAG_TSYNC (1UL << 0)
171 #endif
172 
173 #ifndef SECCOMP_FILTER_FLAG_LOG
174 #define SECCOMP_FILTER_FLAG_LOG (1UL << 1)
175 #endif
176 
177 #ifndef SECCOMP_FILTER_FLAG_SPEC_ALLOW
178 #define SECCOMP_FILTER_FLAG_SPEC_ALLOW (1UL << 2)
179 #endif
180 
181 #ifndef PTRACE_SECCOMP_GET_METADATA
182 #define PTRACE_SECCOMP_GET_METADATA	0x420d
183 
184 struct seccomp_metadata {
185 	__u64 filter_off;       /* Input: which filter */
186 	__u64 flags;             /* Output: filter's flags */
187 };
188 #endif
189 
190 #ifndef SECCOMP_FILTER_FLAG_NEW_LISTENER
191 #define SECCOMP_FILTER_FLAG_NEW_LISTENER	(1UL << 3)
192 #endif
193 
194 #ifndef SECCOMP_RET_USER_NOTIF
195 #define SECCOMP_RET_USER_NOTIF 0x7fc00000U
196 
197 #define SECCOMP_IOC_MAGIC		'!'
198 #define SECCOMP_IO(nr)			_IO(SECCOMP_IOC_MAGIC, nr)
199 #define SECCOMP_IOR(nr, type)		_IOR(SECCOMP_IOC_MAGIC, nr, type)
200 #define SECCOMP_IOW(nr, type)		_IOW(SECCOMP_IOC_MAGIC, nr, type)
201 #define SECCOMP_IOWR(nr, type)		_IOWR(SECCOMP_IOC_MAGIC, nr, type)
202 
203 /* Flags for seccomp notification fd ioctl. */
204 #define SECCOMP_IOCTL_NOTIF_RECV	SECCOMP_IOWR(0, struct seccomp_notif)
205 #define SECCOMP_IOCTL_NOTIF_SEND	SECCOMP_IOWR(1,	\
206 						struct seccomp_notif_resp)
207 #define SECCOMP_IOCTL_NOTIF_ID_VALID	SECCOMP_IOW(2, __u64)
208 
209 struct seccomp_notif {
210 	__u64 id;
211 	__u32 pid;
212 	__u32 flags;
213 	struct seccomp_data data;
214 };
215 
216 struct seccomp_notif_resp {
217 	__u64 id;
218 	__s64 val;
219 	__s32 error;
220 	__u32 flags;
221 };
222 
223 struct seccomp_notif_sizes {
224 	__u16 seccomp_notif;
225 	__u16 seccomp_notif_resp;
226 	__u16 seccomp_data;
227 };
228 #endif
229 
230 #ifndef SECCOMP_IOCTL_NOTIF_ADDFD
231 /* On success, the return value is the remote process's added fd number */
232 #define SECCOMP_IOCTL_NOTIF_ADDFD	SECCOMP_IOW(3,	\
233 						struct seccomp_notif_addfd)
234 
235 /* valid flags for seccomp_notif_addfd */
236 #define SECCOMP_ADDFD_FLAG_SETFD	(1UL << 0) /* Specify remote fd */
237 
238 struct seccomp_notif_addfd {
239 	__u64 id;
240 	__u32 flags;
241 	__u32 srcfd;
242 	__u32 newfd;
243 	__u32 newfd_flags;
244 };
245 #endif
246 
247 #ifndef SECCOMP_ADDFD_FLAG_SEND
248 #define SECCOMP_ADDFD_FLAG_SEND	(1UL << 1) /* Addfd and return it, atomically */
249 #endif
250 
251 struct seccomp_notif_addfd_small {
252 	__u64 id;
253 	char weird[4];
254 };
255 #define SECCOMP_IOCTL_NOTIF_ADDFD_SMALL	\
256 	SECCOMP_IOW(3, struct seccomp_notif_addfd_small)
257 
258 struct seccomp_notif_addfd_big {
259 	union {
260 		struct seccomp_notif_addfd addfd;
261 		char buf[sizeof(struct seccomp_notif_addfd) + 8];
262 	};
263 };
264 #define SECCOMP_IOCTL_NOTIF_ADDFD_BIG	\
265 	SECCOMP_IOWR(3, struct seccomp_notif_addfd_big)
266 
267 #ifndef PTRACE_EVENTMSG_SYSCALL_ENTRY
268 #define PTRACE_EVENTMSG_SYSCALL_ENTRY	1
269 #define PTRACE_EVENTMSG_SYSCALL_EXIT	2
270 #endif
271 
272 #ifndef SECCOMP_USER_NOTIF_FLAG_CONTINUE
273 #define SECCOMP_USER_NOTIF_FLAG_CONTINUE 0x00000001
274 #endif
275 
276 #ifndef SECCOMP_FILTER_FLAG_TSYNC_ESRCH
277 #define SECCOMP_FILTER_FLAG_TSYNC_ESRCH (1UL << 4)
278 #endif
279 
280 #ifndef SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV
281 #define SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV (1UL << 5)
282 #endif
283 
284 #ifndef seccomp
285 int seccomp(unsigned int op, unsigned int flags, void *args)
286 {
287 	errno = 0;
288 	return syscall(__NR_seccomp, op, flags, args);
289 }
290 #endif
291 
292 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
293 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]))
294 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
295 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32))
296 #else
297 #error "wut? Unknown __BYTE_ORDER__?!"
298 #endif
299 
300 #define SIBLING_EXIT_UNKILLED	0xbadbeef
301 #define SIBLING_EXIT_FAILURE	0xbadface
302 #define SIBLING_EXIT_NEWPRIVS	0xbadfeed
303 
304 static int __filecmp(pid_t pid1, pid_t pid2, int fd1, int fd2)
305 {
306 #ifdef __NR_kcmp
307 	errno = 0;
308 	return syscall(__NR_kcmp, pid1, pid2, KCMP_FILE, fd1, fd2);
309 #else
310 	errno = ENOSYS;
311 	return -1;
312 #endif
313 }
314 
315 /* Have TH_LOG report actual location filecmp() is used. */
316 #define filecmp(pid1, pid2, fd1, fd2)	({		\
317 	int _ret;					\
318 							\
319 	_ret = __filecmp(pid1, pid2, fd1, fd2);		\
320 	if (_ret != 0) {				\
321 		if (_ret < 0 && errno == ENOSYS) {	\
322 			TH_LOG("kcmp() syscall missing (test is less accurate)");\
323 			_ret = 0;			\
324 		}					\
325 	}						\
326 	_ret; })
327 
328 TEST(kcmp)
329 {
330 	int ret;
331 
332 	ret = __filecmp(getpid(), getpid(), 1, 1);
333 	EXPECT_EQ(ret, 0);
334 	if (ret != 0 && errno == ENOSYS)
335 		SKIP(return, "Kernel does not support kcmp() (missing CONFIG_KCMP?)");
336 }
337 
338 TEST(mode_strict_support)
339 {
340 	long ret;
341 
342 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
343 	ASSERT_EQ(0, ret) {
344 		TH_LOG("Kernel does not support CONFIG_SECCOMP");
345 	}
346 	syscall(__NR_exit, 0);
347 }
348 
349 TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL)
350 {
351 	long ret;
352 
353 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
354 	ASSERT_EQ(0, ret) {
355 		TH_LOG("Kernel does not support CONFIG_SECCOMP");
356 	}
357 	syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
358 		NULL, NULL, NULL);
359 	EXPECT_FALSE(true) {
360 		TH_LOG("Unreachable!");
361 	}
362 }
363 
364 /* Note! This doesn't test no new privs behavior */
365 TEST(no_new_privs_support)
366 {
367 	long ret;
368 
369 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
370 	EXPECT_EQ(0, ret) {
371 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
372 	}
373 }
374 
375 /* Tests kernel support by checking for a copy_from_user() fault on NULL. */
376 TEST(mode_filter_support)
377 {
378 	long ret;
379 
380 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
381 	ASSERT_EQ(0, ret) {
382 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
383 	}
384 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL);
385 	EXPECT_EQ(-1, ret);
386 	EXPECT_EQ(EFAULT, errno) {
387 		TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!");
388 	}
389 }
390 
391 TEST(mode_filter_without_nnp)
392 {
393 	struct sock_filter filter[] = {
394 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
395 	};
396 	struct sock_fprog prog = {
397 		.len = (unsigned short)ARRAY_SIZE(filter),
398 		.filter = filter,
399 	};
400 	long ret;
401 	cap_t cap = cap_get_proc();
402 	cap_flag_value_t is_cap_sys_admin = 0;
403 
404 	ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0);
405 	ASSERT_LE(0, ret) {
406 		TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS");
407 	}
408 	errno = 0;
409 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
410 	/* Succeeds with CAP_SYS_ADMIN, fails without */
411 	cap_get_flag(cap, CAP_SYS_ADMIN, CAP_EFFECTIVE, &is_cap_sys_admin);
412 	if (!is_cap_sys_admin) {
413 		EXPECT_EQ(-1, ret);
414 		EXPECT_EQ(EACCES, errno);
415 	} else {
416 		EXPECT_EQ(0, ret);
417 	}
418 }
419 
420 #define MAX_INSNS_PER_PATH 32768
421 
422 TEST(filter_size_limits)
423 {
424 	int i;
425 	int count = BPF_MAXINSNS + 1;
426 	struct sock_filter allow[] = {
427 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
428 	};
429 	struct sock_filter *filter;
430 	struct sock_fprog prog = { };
431 	long ret;
432 
433 	filter = calloc(count, sizeof(*filter));
434 	ASSERT_NE(NULL, filter);
435 
436 	for (i = 0; i < count; i++)
437 		filter[i] = allow[0];
438 
439 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
440 	ASSERT_EQ(0, ret);
441 
442 	prog.filter = filter;
443 	prog.len = count;
444 
445 	/* Too many filter instructions in a single filter. */
446 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
447 	ASSERT_NE(0, ret) {
448 		TH_LOG("Installing %d insn filter was allowed", prog.len);
449 	}
450 
451 	/* One less is okay, though. */
452 	prog.len -= 1;
453 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
454 	ASSERT_EQ(0, ret) {
455 		TH_LOG("Installing %d insn filter wasn't allowed", prog.len);
456 	}
457 }
458 
459 TEST(filter_chain_limits)
460 {
461 	int i;
462 	int count = BPF_MAXINSNS;
463 	struct sock_filter allow[] = {
464 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
465 	};
466 	struct sock_filter *filter;
467 	struct sock_fprog prog = { };
468 	long ret;
469 
470 	filter = calloc(count, sizeof(*filter));
471 	ASSERT_NE(NULL, filter);
472 
473 	for (i = 0; i < count; i++)
474 		filter[i] = allow[0];
475 
476 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
477 	ASSERT_EQ(0, ret);
478 
479 	prog.filter = filter;
480 	prog.len = 1;
481 
482 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
483 	ASSERT_EQ(0, ret);
484 
485 	prog.len = count;
486 
487 	/* Too many total filter instructions. */
488 	for (i = 0; i < MAX_INSNS_PER_PATH; i++) {
489 		ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
490 		if (ret != 0)
491 			break;
492 	}
493 	ASSERT_NE(0, ret) {
494 		TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)",
495 		       i, count, i * (count + 4));
496 	}
497 }
498 
499 TEST(mode_filter_cannot_move_to_strict)
500 {
501 	struct sock_filter filter[] = {
502 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
503 	};
504 	struct sock_fprog prog = {
505 		.len = (unsigned short)ARRAY_SIZE(filter),
506 		.filter = filter,
507 	};
508 	long ret;
509 
510 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
511 	ASSERT_EQ(0, ret);
512 
513 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
514 	ASSERT_EQ(0, ret);
515 
516 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0);
517 	EXPECT_EQ(-1, ret);
518 	EXPECT_EQ(EINVAL, errno);
519 }
520 
521 
522 TEST(mode_filter_get_seccomp)
523 {
524 	struct sock_filter filter[] = {
525 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
526 	};
527 	struct sock_fprog prog = {
528 		.len = (unsigned short)ARRAY_SIZE(filter),
529 		.filter = filter,
530 	};
531 	long ret;
532 
533 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
534 	ASSERT_EQ(0, ret);
535 
536 	ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
537 	EXPECT_EQ(0, ret);
538 
539 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
540 	ASSERT_EQ(0, ret);
541 
542 	ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
543 	EXPECT_EQ(2, ret);
544 }
545 
546 
547 TEST(ALLOW_all)
548 {
549 	struct sock_filter filter[] = {
550 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
551 	};
552 	struct sock_fprog prog = {
553 		.len = (unsigned short)ARRAY_SIZE(filter),
554 		.filter = filter,
555 	};
556 	long ret;
557 
558 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
559 	ASSERT_EQ(0, ret);
560 
561 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
562 	ASSERT_EQ(0, ret);
563 }
564 
565 TEST(empty_prog)
566 {
567 	struct sock_filter filter[] = {
568 	};
569 	struct sock_fprog prog = {
570 		.len = (unsigned short)ARRAY_SIZE(filter),
571 		.filter = filter,
572 	};
573 	long ret;
574 
575 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
576 	ASSERT_EQ(0, ret);
577 
578 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
579 	EXPECT_EQ(-1, ret);
580 	EXPECT_EQ(EINVAL, errno);
581 }
582 
583 TEST(log_all)
584 {
585 	struct sock_filter filter[] = {
586 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG),
587 	};
588 	struct sock_fprog prog = {
589 		.len = (unsigned short)ARRAY_SIZE(filter),
590 		.filter = filter,
591 	};
592 	long ret;
593 	pid_t parent = getppid();
594 
595 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
596 	ASSERT_EQ(0, ret);
597 
598 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
599 	ASSERT_EQ(0, ret);
600 
601 	/* getppid() should succeed and be logged (no check for logging) */
602 	EXPECT_EQ(parent, syscall(__NR_getppid));
603 }
604 
605 TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS)
606 {
607 	struct sock_filter filter[] = {
608 		BPF_STMT(BPF_RET|BPF_K, 0x10000000U),
609 	};
610 	struct sock_fprog prog = {
611 		.len = (unsigned short)ARRAY_SIZE(filter),
612 		.filter = filter,
613 	};
614 	long ret;
615 
616 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
617 	ASSERT_EQ(0, ret);
618 
619 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
620 	ASSERT_EQ(0, ret);
621 	EXPECT_EQ(0, syscall(__NR_getpid)) {
622 		TH_LOG("getpid() shouldn't ever return");
623 	}
624 }
625 
626 /* return code >= 0x80000000 is unused. */
627 TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS)
628 {
629 	struct sock_filter filter[] = {
630 		BPF_STMT(BPF_RET|BPF_K, 0x90000000U),
631 	};
632 	struct sock_fprog prog = {
633 		.len = (unsigned short)ARRAY_SIZE(filter),
634 		.filter = filter,
635 	};
636 	long ret;
637 
638 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
639 	ASSERT_EQ(0, ret);
640 
641 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
642 	ASSERT_EQ(0, ret);
643 	EXPECT_EQ(0, syscall(__NR_getpid)) {
644 		TH_LOG("getpid() shouldn't ever return");
645 	}
646 }
647 
648 TEST_SIGNAL(KILL_all, SIGSYS)
649 {
650 	struct sock_filter filter[] = {
651 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
652 	};
653 	struct sock_fprog prog = {
654 		.len = (unsigned short)ARRAY_SIZE(filter),
655 		.filter = filter,
656 	};
657 	long ret;
658 
659 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
660 	ASSERT_EQ(0, ret);
661 
662 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
663 	ASSERT_EQ(0, ret);
664 }
665 
666 TEST_SIGNAL(KILL_one, SIGSYS)
667 {
668 	struct sock_filter filter[] = {
669 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
670 			offsetof(struct seccomp_data, nr)),
671 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
672 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
673 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
674 	};
675 	struct sock_fprog prog = {
676 		.len = (unsigned short)ARRAY_SIZE(filter),
677 		.filter = filter,
678 	};
679 	long ret;
680 	pid_t parent = getppid();
681 
682 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
683 	ASSERT_EQ(0, ret);
684 
685 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
686 	ASSERT_EQ(0, ret);
687 
688 	EXPECT_EQ(parent, syscall(__NR_getppid));
689 	/* getpid() should never return. */
690 	EXPECT_EQ(0, syscall(__NR_getpid));
691 }
692 
693 TEST_SIGNAL(KILL_one_arg_one, SIGSYS)
694 {
695 	void *fatal_address;
696 	struct sock_filter filter[] = {
697 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
698 			offsetof(struct seccomp_data, nr)),
699 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0),
700 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
701 		/* Only both with lower 32-bit for now. */
702 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)),
703 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K,
704 			(unsigned long)&fatal_address, 0, 1),
705 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
706 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
707 	};
708 	struct sock_fprog prog = {
709 		.len = (unsigned short)ARRAY_SIZE(filter),
710 		.filter = filter,
711 	};
712 	long ret;
713 	pid_t parent = getppid();
714 	struct tms timebuf;
715 	clock_t clock = times(&timebuf);
716 
717 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
718 	ASSERT_EQ(0, ret);
719 
720 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
721 	ASSERT_EQ(0, ret);
722 
723 	EXPECT_EQ(parent, syscall(__NR_getppid));
724 	EXPECT_LE(clock, syscall(__NR_times, &timebuf));
725 	/* times() should never return. */
726 	EXPECT_EQ(0, syscall(__NR_times, &fatal_address));
727 }
728 
729 TEST_SIGNAL(KILL_one_arg_six, SIGSYS)
730 {
731 #ifndef __NR_mmap2
732 	int sysno = __NR_mmap;
733 #else
734 	int sysno = __NR_mmap2;
735 #endif
736 	struct sock_filter filter[] = {
737 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
738 			offsetof(struct seccomp_data, nr)),
739 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0),
740 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
741 		/* Only both with lower 32-bit for now. */
742 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)),
743 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1),
744 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
745 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
746 	};
747 	struct sock_fprog prog = {
748 		.len = (unsigned short)ARRAY_SIZE(filter),
749 		.filter = filter,
750 	};
751 	long ret;
752 	pid_t parent = getppid();
753 	int fd;
754 	void *map1, *map2;
755 	int page_size = sysconf(_SC_PAGESIZE);
756 
757 	ASSERT_LT(0, page_size);
758 
759 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
760 	ASSERT_EQ(0, ret);
761 
762 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
763 	ASSERT_EQ(0, ret);
764 
765 	fd = open("/dev/zero", O_RDONLY);
766 	ASSERT_NE(-1, fd);
767 
768 	EXPECT_EQ(parent, syscall(__NR_getppid));
769 	map1 = (void *)syscall(sysno,
770 		NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size);
771 	EXPECT_NE(MAP_FAILED, map1);
772 	/* mmap2() should never return. */
773 	map2 = (void *)syscall(sysno,
774 		 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE);
775 	EXPECT_EQ(MAP_FAILED, map2);
776 
777 	/* The test failed, so clean up the resources. */
778 	munmap(map1, page_size);
779 	munmap(map2, page_size);
780 	close(fd);
781 }
782 
783 /* This is a thread task to die via seccomp filter violation. */
784 void *kill_thread(void *data)
785 {
786 	bool die = (bool)data;
787 
788 	if (die) {
789 		syscall(__NR_getpid);
790 		return (void *)SIBLING_EXIT_FAILURE;
791 	}
792 
793 	return (void *)SIBLING_EXIT_UNKILLED;
794 }
795 
796 enum kill_t {
797 	KILL_THREAD,
798 	KILL_PROCESS,
799 	RET_UNKNOWN
800 };
801 
802 /* Prepare a thread that will kill itself or both of us. */
803 void kill_thread_or_group(struct __test_metadata *_metadata,
804 			  enum kill_t kill_how)
805 {
806 	pthread_t thread;
807 	void *status;
808 	/* Kill only when calling __NR_getpid. */
809 	struct sock_filter filter_thread[] = {
810 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
811 			offsetof(struct seccomp_data, nr)),
812 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
813 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD),
814 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
815 	};
816 	struct sock_fprog prog_thread = {
817 		.len = (unsigned short)ARRAY_SIZE(filter_thread),
818 		.filter = filter_thread,
819 	};
820 	int kill = kill_how == KILL_PROCESS ? SECCOMP_RET_KILL_PROCESS : 0xAAAAAAAA;
821 	struct sock_filter filter_process[] = {
822 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
823 			offsetof(struct seccomp_data, nr)),
824 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
825 		BPF_STMT(BPF_RET|BPF_K, kill),
826 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
827 	};
828 	struct sock_fprog prog_process = {
829 		.len = (unsigned short)ARRAY_SIZE(filter_process),
830 		.filter = filter_process,
831 	};
832 
833 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
834 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
835 	}
836 
837 	ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0,
838 			     kill_how == KILL_THREAD ? &prog_thread
839 						     : &prog_process));
840 
841 	/*
842 	 * Add the KILL_THREAD rule again to make sure that the KILL_PROCESS
843 	 * flag cannot be downgraded by a new filter.
844 	 */
845 	if (kill_how == KILL_PROCESS)
846 		ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog_thread));
847 
848 	/* Start a thread that will exit immediately. */
849 	ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)false));
850 	ASSERT_EQ(0, pthread_join(thread, &status));
851 	ASSERT_EQ(SIBLING_EXIT_UNKILLED, (unsigned long)status);
852 
853 	/* Start a thread that will die immediately. */
854 	ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)true));
855 	ASSERT_EQ(0, pthread_join(thread, &status));
856 	ASSERT_NE(SIBLING_EXIT_FAILURE, (unsigned long)status);
857 
858 	/*
859 	 * If we get here, only the spawned thread died. Let the parent know
860 	 * the whole process didn't die (i.e. this thread, the spawner,
861 	 * stayed running).
862 	 */
863 	exit(42);
864 }
865 
866 TEST(KILL_thread)
867 {
868 	int status;
869 	pid_t child_pid;
870 
871 	child_pid = fork();
872 	ASSERT_LE(0, child_pid);
873 	if (child_pid == 0) {
874 		kill_thread_or_group(_metadata, KILL_THREAD);
875 		_exit(38);
876 	}
877 
878 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
879 
880 	/* If only the thread was killed, we'll see exit 42. */
881 	ASSERT_TRUE(WIFEXITED(status));
882 	ASSERT_EQ(42, WEXITSTATUS(status));
883 }
884 
885 TEST(KILL_process)
886 {
887 	int status;
888 	pid_t child_pid;
889 
890 	child_pid = fork();
891 	ASSERT_LE(0, child_pid);
892 	if (child_pid == 0) {
893 		kill_thread_or_group(_metadata, KILL_PROCESS);
894 		_exit(38);
895 	}
896 
897 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
898 
899 	/* If the entire process was killed, we'll see SIGSYS. */
900 	ASSERT_TRUE(WIFSIGNALED(status));
901 	ASSERT_EQ(SIGSYS, WTERMSIG(status));
902 }
903 
904 TEST(KILL_unknown)
905 {
906 	int status;
907 	pid_t child_pid;
908 
909 	child_pid = fork();
910 	ASSERT_LE(0, child_pid);
911 	if (child_pid == 0) {
912 		kill_thread_or_group(_metadata, RET_UNKNOWN);
913 		_exit(38);
914 	}
915 
916 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
917 
918 	/* If the entire process was killed, we'll see SIGSYS. */
919 	EXPECT_TRUE(WIFSIGNALED(status)) {
920 		TH_LOG("Unknown SECCOMP_RET is only killing the thread?");
921 	}
922 	ASSERT_EQ(SIGSYS, WTERMSIG(status));
923 }
924 
925 /* TODO(wad) add 64-bit versus 32-bit arg tests. */
926 TEST(arg_out_of_range)
927 {
928 	struct sock_filter filter[] = {
929 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)),
930 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
931 	};
932 	struct sock_fprog prog = {
933 		.len = (unsigned short)ARRAY_SIZE(filter),
934 		.filter = filter,
935 	};
936 	long ret;
937 
938 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
939 	ASSERT_EQ(0, ret);
940 
941 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
942 	EXPECT_EQ(-1, ret);
943 	EXPECT_EQ(EINVAL, errno);
944 }
945 
946 #define ERRNO_FILTER(name, errno)					\
947 	struct sock_filter _read_filter_##name[] = {			\
948 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,				\
949 			offsetof(struct seccomp_data, nr)),		\
950 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),	\
951 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | errno),	\
952 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),		\
953 	};								\
954 	struct sock_fprog prog_##name = {				\
955 		.len = (unsigned short)ARRAY_SIZE(_read_filter_##name),	\
956 		.filter = _read_filter_##name,				\
957 	}
958 
959 /* Make sure basic errno values are correctly passed through a filter. */
960 TEST(ERRNO_valid)
961 {
962 	ERRNO_FILTER(valid, E2BIG);
963 	long ret;
964 	pid_t parent = getppid();
965 
966 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
967 	ASSERT_EQ(0, ret);
968 
969 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_valid);
970 	ASSERT_EQ(0, ret);
971 
972 	EXPECT_EQ(parent, syscall(__NR_getppid));
973 	EXPECT_EQ(-1, read(-1, NULL, 0));
974 	EXPECT_EQ(E2BIG, errno);
975 }
976 
977 /* Make sure an errno of zero is correctly handled by the arch code. */
978 TEST(ERRNO_zero)
979 {
980 	ERRNO_FILTER(zero, 0);
981 	long ret;
982 	pid_t parent = getppid();
983 
984 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
985 	ASSERT_EQ(0, ret);
986 
987 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_zero);
988 	ASSERT_EQ(0, ret);
989 
990 	EXPECT_EQ(parent, syscall(__NR_getppid));
991 	/* "errno" of 0 is ok. */
992 	EXPECT_EQ(0, read(-1, NULL, 0));
993 }
994 
995 /*
996  * The SECCOMP_RET_DATA mask is 16 bits wide, but errno is smaller.
997  * This tests that the errno value gets capped correctly, fixed by
998  * 580c57f10768 ("seccomp: cap SECCOMP_RET_ERRNO data to MAX_ERRNO").
999  */
1000 TEST(ERRNO_capped)
1001 {
1002 	ERRNO_FILTER(capped, 4096);
1003 	long ret;
1004 	pid_t parent = getppid();
1005 
1006 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1007 	ASSERT_EQ(0, ret);
1008 
1009 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_capped);
1010 	ASSERT_EQ(0, ret);
1011 
1012 	EXPECT_EQ(parent, syscall(__NR_getppid));
1013 	EXPECT_EQ(-1, read(-1, NULL, 0));
1014 	EXPECT_EQ(4095, errno);
1015 }
1016 
1017 /*
1018  * Filters are processed in reverse order: last applied is executed first.
1019  * Since only the SECCOMP_RET_ACTION mask is tested for return values, the
1020  * SECCOMP_RET_DATA mask results will follow the most recently applied
1021  * matching filter return (and not the lowest or highest value).
1022  */
1023 TEST(ERRNO_order)
1024 {
1025 	ERRNO_FILTER(first,  11);
1026 	ERRNO_FILTER(second, 13);
1027 	ERRNO_FILTER(third,  12);
1028 	long ret;
1029 	pid_t parent = getppid();
1030 
1031 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1032 	ASSERT_EQ(0, ret);
1033 
1034 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_first);
1035 	ASSERT_EQ(0, ret);
1036 
1037 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_second);
1038 	ASSERT_EQ(0, ret);
1039 
1040 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_third);
1041 	ASSERT_EQ(0, ret);
1042 
1043 	EXPECT_EQ(parent, syscall(__NR_getppid));
1044 	EXPECT_EQ(-1, read(-1, NULL, 0));
1045 	EXPECT_EQ(12, errno);
1046 }
1047 
1048 FIXTURE(TRAP) {
1049 	struct sock_fprog prog;
1050 };
1051 
1052 FIXTURE_SETUP(TRAP)
1053 {
1054 	struct sock_filter filter[] = {
1055 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1056 			offsetof(struct seccomp_data, nr)),
1057 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
1058 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
1059 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1060 	};
1061 
1062 	memset(&self->prog, 0, sizeof(self->prog));
1063 	self->prog.filter = malloc(sizeof(filter));
1064 	ASSERT_NE(NULL, self->prog.filter);
1065 	memcpy(self->prog.filter, filter, sizeof(filter));
1066 	self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1067 }
1068 
1069 FIXTURE_TEARDOWN(TRAP)
1070 {
1071 	if (self->prog.filter)
1072 		free(self->prog.filter);
1073 }
1074 
1075 TEST_F_SIGNAL(TRAP, dfl, SIGSYS)
1076 {
1077 	long ret;
1078 
1079 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1080 	ASSERT_EQ(0, ret);
1081 
1082 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
1083 	ASSERT_EQ(0, ret);
1084 	syscall(__NR_getpid);
1085 }
1086 
1087 /* Ensure that SIGSYS overrides SIG_IGN */
1088 TEST_F_SIGNAL(TRAP, ign, SIGSYS)
1089 {
1090 	long ret;
1091 
1092 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1093 	ASSERT_EQ(0, ret);
1094 
1095 	signal(SIGSYS, SIG_IGN);
1096 
1097 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
1098 	ASSERT_EQ(0, ret);
1099 	syscall(__NR_getpid);
1100 }
1101 
1102 static siginfo_t TRAP_info;
1103 static volatile int TRAP_nr;
1104 static void TRAP_action(int nr, siginfo_t *info, void *void_context)
1105 {
1106 	memcpy(&TRAP_info, info, sizeof(TRAP_info));
1107 	TRAP_nr = nr;
1108 }
1109 
1110 TEST_F(TRAP, handler)
1111 {
1112 	int ret, test;
1113 	struct sigaction act;
1114 	sigset_t mask;
1115 
1116 	memset(&act, 0, sizeof(act));
1117 	sigemptyset(&mask);
1118 	sigaddset(&mask, SIGSYS);
1119 
1120 	act.sa_sigaction = &TRAP_action;
1121 	act.sa_flags = SA_SIGINFO;
1122 	ret = sigaction(SIGSYS, &act, NULL);
1123 	ASSERT_EQ(0, ret) {
1124 		TH_LOG("sigaction failed");
1125 	}
1126 	ret = sigprocmask(SIG_UNBLOCK, &mask, NULL);
1127 	ASSERT_EQ(0, ret) {
1128 		TH_LOG("sigprocmask failed");
1129 	}
1130 
1131 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1132 	ASSERT_EQ(0, ret);
1133 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
1134 	ASSERT_EQ(0, ret);
1135 	TRAP_nr = 0;
1136 	memset(&TRAP_info, 0, sizeof(TRAP_info));
1137 	/* Expect the registers to be rolled back. (nr = error) may vary
1138 	 * based on arch. */
1139 	ret = syscall(__NR_getpid);
1140 	/* Silence gcc warning about volatile. */
1141 	test = TRAP_nr;
1142 	EXPECT_EQ(SIGSYS, test);
1143 	struct local_sigsys {
1144 		void *_call_addr;	/* calling user insn */
1145 		int _syscall;		/* triggering system call number */
1146 		unsigned int _arch;	/* AUDIT_ARCH_* of syscall */
1147 	} *sigsys = (struct local_sigsys *)
1148 #ifdef si_syscall
1149 		&(TRAP_info.si_call_addr);
1150 #else
1151 		&TRAP_info.si_pid;
1152 #endif
1153 	EXPECT_EQ(__NR_getpid, sigsys->_syscall);
1154 	/* Make sure arch is non-zero. */
1155 	EXPECT_NE(0, sigsys->_arch);
1156 	EXPECT_NE(0, (unsigned long)sigsys->_call_addr);
1157 }
1158 
1159 FIXTURE(precedence) {
1160 	struct sock_fprog allow;
1161 	struct sock_fprog log;
1162 	struct sock_fprog trace;
1163 	struct sock_fprog error;
1164 	struct sock_fprog trap;
1165 	struct sock_fprog kill;
1166 };
1167 
1168 FIXTURE_SETUP(precedence)
1169 {
1170 	struct sock_filter allow_insns[] = {
1171 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1172 	};
1173 	struct sock_filter log_insns[] = {
1174 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1175 			offsetof(struct seccomp_data, nr)),
1176 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1177 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1178 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG),
1179 	};
1180 	struct sock_filter trace_insns[] = {
1181 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1182 			offsetof(struct seccomp_data, nr)),
1183 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1184 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1185 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE),
1186 	};
1187 	struct sock_filter error_insns[] = {
1188 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1189 			offsetof(struct seccomp_data, nr)),
1190 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1191 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1192 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO),
1193 	};
1194 	struct sock_filter trap_insns[] = {
1195 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1196 			offsetof(struct seccomp_data, nr)),
1197 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1198 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1199 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
1200 	};
1201 	struct sock_filter kill_insns[] = {
1202 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1203 			offsetof(struct seccomp_data, nr)),
1204 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1205 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1206 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1207 	};
1208 
1209 	memset(self, 0, sizeof(*self));
1210 #define FILTER_ALLOC(_x) \
1211 	self->_x.filter = malloc(sizeof(_x##_insns)); \
1212 	ASSERT_NE(NULL, self->_x.filter); \
1213 	memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \
1214 	self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns)
1215 	FILTER_ALLOC(allow);
1216 	FILTER_ALLOC(log);
1217 	FILTER_ALLOC(trace);
1218 	FILTER_ALLOC(error);
1219 	FILTER_ALLOC(trap);
1220 	FILTER_ALLOC(kill);
1221 }
1222 
1223 FIXTURE_TEARDOWN(precedence)
1224 {
1225 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter)
1226 	FILTER_FREE(allow);
1227 	FILTER_FREE(log);
1228 	FILTER_FREE(trace);
1229 	FILTER_FREE(error);
1230 	FILTER_FREE(trap);
1231 	FILTER_FREE(kill);
1232 }
1233 
1234 TEST_F(precedence, allow_ok)
1235 {
1236 	pid_t parent, res = 0;
1237 	long ret;
1238 
1239 	parent = getppid();
1240 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1241 	ASSERT_EQ(0, ret);
1242 
1243 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1244 	ASSERT_EQ(0, ret);
1245 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1246 	ASSERT_EQ(0, ret);
1247 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1248 	ASSERT_EQ(0, ret);
1249 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1250 	ASSERT_EQ(0, ret);
1251 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1252 	ASSERT_EQ(0, ret);
1253 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1254 	ASSERT_EQ(0, ret);
1255 	/* Should work just fine. */
1256 	res = syscall(__NR_getppid);
1257 	EXPECT_EQ(parent, res);
1258 }
1259 
1260 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS)
1261 {
1262 	pid_t parent, res = 0;
1263 	long ret;
1264 
1265 	parent = getppid();
1266 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1267 	ASSERT_EQ(0, ret);
1268 
1269 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1270 	ASSERT_EQ(0, ret);
1271 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1272 	ASSERT_EQ(0, ret);
1273 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1274 	ASSERT_EQ(0, ret);
1275 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1276 	ASSERT_EQ(0, ret);
1277 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1278 	ASSERT_EQ(0, ret);
1279 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1280 	ASSERT_EQ(0, ret);
1281 	/* Should work just fine. */
1282 	res = syscall(__NR_getppid);
1283 	EXPECT_EQ(parent, res);
1284 	/* getpid() should never return. */
1285 	res = syscall(__NR_getpid);
1286 	EXPECT_EQ(0, res);
1287 }
1288 
1289 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS)
1290 {
1291 	pid_t parent;
1292 	long ret;
1293 
1294 	parent = getppid();
1295 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1296 	ASSERT_EQ(0, ret);
1297 
1298 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1299 	ASSERT_EQ(0, ret);
1300 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1301 	ASSERT_EQ(0, ret);
1302 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1303 	ASSERT_EQ(0, ret);
1304 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1305 	ASSERT_EQ(0, ret);
1306 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1307 	ASSERT_EQ(0, ret);
1308 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1309 	ASSERT_EQ(0, ret);
1310 	/* Should work just fine. */
1311 	EXPECT_EQ(parent, syscall(__NR_getppid));
1312 	/* getpid() should never return. */
1313 	EXPECT_EQ(0, syscall(__NR_getpid));
1314 }
1315 
1316 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS)
1317 {
1318 	pid_t parent;
1319 	long ret;
1320 
1321 	parent = getppid();
1322 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1323 	ASSERT_EQ(0, ret);
1324 
1325 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1326 	ASSERT_EQ(0, ret);
1327 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1328 	ASSERT_EQ(0, ret);
1329 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1330 	ASSERT_EQ(0, ret);
1331 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1332 	ASSERT_EQ(0, ret);
1333 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1334 	ASSERT_EQ(0, ret);
1335 	/* Should work just fine. */
1336 	EXPECT_EQ(parent, syscall(__NR_getppid));
1337 	/* getpid() should never return. */
1338 	EXPECT_EQ(0, syscall(__NR_getpid));
1339 }
1340 
1341 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS)
1342 {
1343 	pid_t parent;
1344 	long ret;
1345 
1346 	parent = getppid();
1347 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1348 	ASSERT_EQ(0, ret);
1349 
1350 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1351 	ASSERT_EQ(0, ret);
1352 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1353 	ASSERT_EQ(0, ret);
1354 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1355 	ASSERT_EQ(0, ret);
1356 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1357 	ASSERT_EQ(0, ret);
1358 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1359 	ASSERT_EQ(0, ret);
1360 	/* Should work just fine. */
1361 	EXPECT_EQ(parent, syscall(__NR_getppid));
1362 	/* getpid() should never return. */
1363 	EXPECT_EQ(0, syscall(__NR_getpid));
1364 }
1365 
1366 TEST_F(precedence, errno_is_third)
1367 {
1368 	pid_t parent;
1369 	long ret;
1370 
1371 	parent = getppid();
1372 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1373 	ASSERT_EQ(0, ret);
1374 
1375 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1376 	ASSERT_EQ(0, ret);
1377 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1378 	ASSERT_EQ(0, ret);
1379 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1380 	ASSERT_EQ(0, ret);
1381 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1382 	ASSERT_EQ(0, ret);
1383 	/* Should work just fine. */
1384 	EXPECT_EQ(parent, syscall(__NR_getppid));
1385 	EXPECT_EQ(0, syscall(__NR_getpid));
1386 }
1387 
1388 TEST_F(precedence, errno_is_third_in_any_order)
1389 {
1390 	pid_t parent;
1391 	long ret;
1392 
1393 	parent = getppid();
1394 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1395 	ASSERT_EQ(0, ret);
1396 
1397 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1398 	ASSERT_EQ(0, ret);
1399 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1400 	ASSERT_EQ(0, ret);
1401 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1402 	ASSERT_EQ(0, ret);
1403 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1404 	ASSERT_EQ(0, ret);
1405 	/* Should work just fine. */
1406 	EXPECT_EQ(parent, syscall(__NR_getppid));
1407 	EXPECT_EQ(0, syscall(__NR_getpid));
1408 }
1409 
1410 TEST_F(precedence, trace_is_fourth)
1411 {
1412 	pid_t parent;
1413 	long ret;
1414 
1415 	parent = getppid();
1416 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1417 	ASSERT_EQ(0, ret);
1418 
1419 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1420 	ASSERT_EQ(0, ret);
1421 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1422 	ASSERT_EQ(0, ret);
1423 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1424 	ASSERT_EQ(0, ret);
1425 	/* Should work just fine. */
1426 	EXPECT_EQ(parent, syscall(__NR_getppid));
1427 	/* No ptracer */
1428 	EXPECT_EQ(-1, syscall(__NR_getpid));
1429 }
1430 
1431 TEST_F(precedence, trace_is_fourth_in_any_order)
1432 {
1433 	pid_t parent;
1434 	long ret;
1435 
1436 	parent = getppid();
1437 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1438 	ASSERT_EQ(0, ret);
1439 
1440 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1441 	ASSERT_EQ(0, ret);
1442 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1443 	ASSERT_EQ(0, ret);
1444 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1445 	ASSERT_EQ(0, ret);
1446 	/* Should work just fine. */
1447 	EXPECT_EQ(parent, syscall(__NR_getppid));
1448 	/* No ptracer */
1449 	EXPECT_EQ(-1, syscall(__NR_getpid));
1450 }
1451 
1452 TEST_F(precedence, log_is_fifth)
1453 {
1454 	pid_t mypid, parent;
1455 	long ret;
1456 
1457 	mypid = getpid();
1458 	parent = getppid();
1459 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1460 	ASSERT_EQ(0, ret);
1461 
1462 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1463 	ASSERT_EQ(0, ret);
1464 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1465 	ASSERT_EQ(0, ret);
1466 	/* Should work just fine. */
1467 	EXPECT_EQ(parent, syscall(__NR_getppid));
1468 	/* Should also work just fine */
1469 	EXPECT_EQ(mypid, syscall(__NR_getpid));
1470 }
1471 
1472 TEST_F(precedence, log_is_fifth_in_any_order)
1473 {
1474 	pid_t mypid, parent;
1475 	long ret;
1476 
1477 	mypid = getpid();
1478 	parent = getppid();
1479 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1480 	ASSERT_EQ(0, ret);
1481 
1482 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1483 	ASSERT_EQ(0, ret);
1484 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1485 	ASSERT_EQ(0, ret);
1486 	/* Should work just fine. */
1487 	EXPECT_EQ(parent, syscall(__NR_getppid));
1488 	/* Should also work just fine */
1489 	EXPECT_EQ(mypid, syscall(__NR_getpid));
1490 }
1491 
1492 #ifndef PTRACE_O_TRACESECCOMP
1493 #define PTRACE_O_TRACESECCOMP	0x00000080
1494 #endif
1495 
1496 /* Catch the Ubuntu 12.04 value error. */
1497 #if PTRACE_EVENT_SECCOMP != 7
1498 #undef PTRACE_EVENT_SECCOMP
1499 #endif
1500 
1501 #ifndef PTRACE_EVENT_SECCOMP
1502 #define PTRACE_EVENT_SECCOMP 7
1503 #endif
1504 
1505 #define PTRACE_EVENT_MASK(status) ((status) >> 16)
1506 bool tracer_running;
1507 void tracer_stop(int sig)
1508 {
1509 	tracer_running = false;
1510 }
1511 
1512 typedef void tracer_func_t(struct __test_metadata *_metadata,
1513 			   pid_t tracee, int status, void *args);
1514 
1515 void start_tracer(struct __test_metadata *_metadata, int fd, pid_t tracee,
1516 	    tracer_func_t tracer_func, void *args, bool ptrace_syscall)
1517 {
1518 	int ret = -1;
1519 	struct sigaction action = {
1520 		.sa_handler = tracer_stop,
1521 	};
1522 
1523 	/* Allow external shutdown. */
1524 	tracer_running = true;
1525 	ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL));
1526 
1527 	errno = 0;
1528 	while (ret == -1 && errno != EINVAL)
1529 		ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0);
1530 	ASSERT_EQ(0, ret) {
1531 		kill(tracee, SIGKILL);
1532 	}
1533 	/* Wait for attach stop */
1534 	wait(NULL);
1535 
1536 	ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, ptrace_syscall ?
1537 						      PTRACE_O_TRACESYSGOOD :
1538 						      PTRACE_O_TRACESECCOMP);
1539 	ASSERT_EQ(0, ret) {
1540 		TH_LOG("Failed to set PTRACE_O_TRACESECCOMP");
1541 		kill(tracee, SIGKILL);
1542 	}
1543 	ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1544 		     tracee, NULL, 0);
1545 	ASSERT_EQ(0, ret);
1546 
1547 	/* Unblock the tracee */
1548 	ASSERT_EQ(1, write(fd, "A", 1));
1549 	ASSERT_EQ(0, close(fd));
1550 
1551 	/* Run until we're shut down. Must assert to stop execution. */
1552 	while (tracer_running) {
1553 		int status;
1554 
1555 		if (wait(&status) != tracee)
1556 			continue;
1557 
1558 		if (WIFSIGNALED(status)) {
1559 			/* Child caught a fatal signal. */
1560 			return;
1561 		}
1562 		if (WIFEXITED(status)) {
1563 			/* Child exited with code. */
1564 			return;
1565 		}
1566 
1567 		/* Check if we got an expected event. */
1568 		ASSERT_EQ(WIFCONTINUED(status), false);
1569 		ASSERT_EQ(WIFSTOPPED(status), true);
1570 		ASSERT_EQ(WSTOPSIG(status) & SIGTRAP, SIGTRAP) {
1571 			TH_LOG("Unexpected WSTOPSIG: %d", WSTOPSIG(status));
1572 		}
1573 
1574 		tracer_func(_metadata, tracee, status, args);
1575 
1576 		ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1577 			     tracee, NULL, 0);
1578 		ASSERT_EQ(0, ret);
1579 	}
1580 	/* Directly report the status of our test harness results. */
1581 	syscall(__NR_exit, _metadata->exit_code);
1582 }
1583 
1584 /* Common tracer setup/teardown functions. */
1585 void cont_handler(int num)
1586 { }
1587 pid_t setup_trace_fixture(struct __test_metadata *_metadata,
1588 			  tracer_func_t func, void *args, bool ptrace_syscall)
1589 {
1590 	char sync;
1591 	int pipefd[2];
1592 	pid_t tracer_pid;
1593 	pid_t tracee = getpid();
1594 
1595 	/* Setup a pipe for clean synchronization. */
1596 	ASSERT_EQ(0, pipe(pipefd));
1597 
1598 	/* Fork a child which we'll promote to tracer */
1599 	tracer_pid = fork();
1600 	ASSERT_LE(0, tracer_pid);
1601 	signal(SIGALRM, cont_handler);
1602 	if (tracer_pid == 0) {
1603 		close(pipefd[0]);
1604 		start_tracer(_metadata, pipefd[1], tracee, func, args,
1605 			     ptrace_syscall);
1606 		syscall(__NR_exit, 0);
1607 	}
1608 	close(pipefd[1]);
1609 	prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0);
1610 	read(pipefd[0], &sync, 1);
1611 	close(pipefd[0]);
1612 
1613 	return tracer_pid;
1614 }
1615 
1616 void teardown_trace_fixture(struct __test_metadata *_metadata,
1617 			    pid_t tracer)
1618 {
1619 	if (tracer) {
1620 		int status;
1621 		/*
1622 		 * Extract the exit code from the other process and
1623 		 * adopt it for ourselves in case its asserts failed.
1624 		 */
1625 		ASSERT_EQ(0, kill(tracer, SIGUSR1));
1626 		ASSERT_EQ(tracer, waitpid(tracer, &status, 0));
1627 		if (WEXITSTATUS(status))
1628 			_metadata->exit_code = KSFT_FAIL;
1629 	}
1630 }
1631 
1632 /* "poke" tracer arguments and function. */
1633 struct tracer_args_poke_t {
1634 	unsigned long poke_addr;
1635 };
1636 
1637 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status,
1638 		 void *args)
1639 {
1640 	int ret;
1641 	unsigned long msg;
1642 	struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args;
1643 
1644 	ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1645 	EXPECT_EQ(0, ret);
1646 	/* If this fails, don't try to recover. */
1647 	ASSERT_EQ(0x1001, msg) {
1648 		kill(tracee, SIGKILL);
1649 	}
1650 	/*
1651 	 * Poke in the message.
1652 	 * Registers are not touched to try to keep this relatively arch
1653 	 * agnostic.
1654 	 */
1655 	ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001);
1656 	EXPECT_EQ(0, ret);
1657 }
1658 
1659 FIXTURE(TRACE_poke) {
1660 	struct sock_fprog prog;
1661 	pid_t tracer;
1662 	long poked;
1663 	struct tracer_args_poke_t tracer_args;
1664 };
1665 
1666 FIXTURE_SETUP(TRACE_poke)
1667 {
1668 	struct sock_filter filter[] = {
1669 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1670 			offsetof(struct seccomp_data, nr)),
1671 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1672 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001),
1673 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1674 	};
1675 
1676 	self->poked = 0;
1677 	memset(&self->prog, 0, sizeof(self->prog));
1678 	self->prog.filter = malloc(sizeof(filter));
1679 	ASSERT_NE(NULL, self->prog.filter);
1680 	memcpy(self->prog.filter, filter, sizeof(filter));
1681 	self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1682 
1683 	/* Set up tracer args. */
1684 	self->tracer_args.poke_addr = (unsigned long)&self->poked;
1685 
1686 	/* Launch tracer. */
1687 	self->tracer = setup_trace_fixture(_metadata, tracer_poke,
1688 					   &self->tracer_args, false);
1689 }
1690 
1691 FIXTURE_TEARDOWN(TRACE_poke)
1692 {
1693 	teardown_trace_fixture(_metadata, self->tracer);
1694 	if (self->prog.filter)
1695 		free(self->prog.filter);
1696 }
1697 
1698 TEST_F(TRACE_poke, read_has_side_effects)
1699 {
1700 	ssize_t ret;
1701 
1702 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1703 	ASSERT_EQ(0, ret);
1704 
1705 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1706 	ASSERT_EQ(0, ret);
1707 
1708 	EXPECT_EQ(0, self->poked);
1709 	ret = read(-1, NULL, 0);
1710 	EXPECT_EQ(-1, ret);
1711 	EXPECT_EQ(0x1001, self->poked);
1712 }
1713 
1714 TEST_F(TRACE_poke, getpid_runs_normally)
1715 {
1716 	long ret;
1717 
1718 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1719 	ASSERT_EQ(0, ret);
1720 
1721 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1722 	ASSERT_EQ(0, ret);
1723 
1724 	EXPECT_EQ(0, self->poked);
1725 	EXPECT_NE(0, syscall(__NR_getpid));
1726 	EXPECT_EQ(0, self->poked);
1727 }
1728 
1729 #if defined(__x86_64__)
1730 # define ARCH_REGS		struct user_regs_struct
1731 # define SYSCALL_NUM(_regs)	(_regs).orig_rax
1732 # define SYSCALL_RET(_regs)	(_regs).rax
1733 #elif defined(__i386__)
1734 # define ARCH_REGS		struct user_regs_struct
1735 # define SYSCALL_NUM(_regs)	(_regs).orig_eax
1736 # define SYSCALL_RET(_regs)	(_regs).eax
1737 #elif defined(__arm__)
1738 # define ARCH_REGS		struct pt_regs
1739 # define SYSCALL_NUM(_regs)	(_regs).ARM_r7
1740 # ifndef PTRACE_SET_SYSCALL
1741 #  define PTRACE_SET_SYSCALL   23
1742 # endif
1743 # define SYSCALL_NUM_SET(_regs, _nr)	\
1744 		EXPECT_EQ(0, ptrace(PTRACE_SET_SYSCALL, tracee, NULL, _nr))
1745 # define SYSCALL_RET(_regs)	(_regs).ARM_r0
1746 #elif defined(__aarch64__)
1747 # define ARCH_REGS		struct user_pt_regs
1748 # define SYSCALL_NUM(_regs)	(_regs).regs[8]
1749 # ifndef NT_ARM_SYSTEM_CALL
1750 #  define NT_ARM_SYSTEM_CALL 0x404
1751 # endif
1752 # define SYSCALL_NUM_SET(_regs, _nr)				\
1753 	do {							\
1754 		struct iovec __v;				\
1755 		typeof(_nr) __nr = (_nr);			\
1756 		__v.iov_base = &__nr;				\
1757 		__v.iov_len = sizeof(__nr);			\
1758 		EXPECT_EQ(0, ptrace(PTRACE_SETREGSET, tracee,	\
1759 				    NT_ARM_SYSTEM_CALL, &__v));	\
1760 	} while (0)
1761 # define SYSCALL_RET(_regs)	(_regs).regs[0]
1762 #elif defined(__loongarch__)
1763 # define ARCH_REGS		struct user_pt_regs
1764 # define SYSCALL_NUM(_regs)	(_regs).regs[11]
1765 # define SYSCALL_RET(_regs)	(_regs).regs[4]
1766 #elif defined(__riscv) && __riscv_xlen == 64
1767 # define ARCH_REGS		struct user_regs_struct
1768 # define SYSCALL_NUM(_regs)	(_regs).a7
1769 # define SYSCALL_RET(_regs)	(_regs).a0
1770 #elif defined(__csky__)
1771 # define ARCH_REGS		struct pt_regs
1772 #  if defined(__CSKYABIV2__)
1773 #   define SYSCALL_NUM(_regs)	(_regs).regs[3]
1774 #  else
1775 #   define SYSCALL_NUM(_regs)	(_regs).regs[9]
1776 #  endif
1777 # define SYSCALL_RET(_regs)	(_regs).a0
1778 #elif defined(__hppa__)
1779 # define ARCH_REGS		struct user_regs_struct
1780 # define SYSCALL_NUM(_regs)	(_regs).gr[20]
1781 # define SYSCALL_RET(_regs)	(_regs).gr[28]
1782 #elif defined(__powerpc__)
1783 # define ARCH_REGS		struct pt_regs
1784 # define SYSCALL_NUM(_regs)	(_regs).gpr[0]
1785 # define SYSCALL_RET(_regs)	(_regs).gpr[3]
1786 # define SYSCALL_RET_SET(_regs, _val)				\
1787 	do {							\
1788 		typeof(_val) _result = (_val);			\
1789 		if ((_regs.trap & 0xfff0) == 0x3000) {		\
1790 			/*					\
1791 			 * scv 0 system call uses -ve result	\
1792 			 * for error, so no need to adjust.	\
1793 			 */					\
1794 			SYSCALL_RET(_regs) = _result;		\
1795 		} else {					\
1796 			/*					\
1797 			 * A syscall error is signaled by the	\
1798 			 * CR0 SO bit and the code is stored as	\
1799 			 * a positive value.			\
1800 			 */					\
1801 			if (_result < 0) {			\
1802 				SYSCALL_RET(_regs) = -_result;	\
1803 				(_regs).ccr |= 0x10000000;	\
1804 			} else {				\
1805 				SYSCALL_RET(_regs) = _result;	\
1806 				(_regs).ccr &= ~0x10000000;	\
1807 			}					\
1808 		}						\
1809 	} while (0)
1810 # define SYSCALL_RET_SET_ON_PTRACE_EXIT
1811 #elif defined(__s390__)
1812 # define ARCH_REGS		s390_regs
1813 # define SYSCALL_NUM(_regs)	(_regs).gprs[2]
1814 # define SYSCALL_RET_SET(_regs, _val)			\
1815 		TH_LOG("Can't modify syscall return on this architecture")
1816 #elif defined(__mips__)
1817 # include <asm/unistd_nr_n32.h>
1818 # include <asm/unistd_nr_n64.h>
1819 # include <asm/unistd_nr_o32.h>
1820 # define ARCH_REGS		struct pt_regs
1821 # define SYSCALL_NUM(_regs)				\
1822 	({						\
1823 		typeof((_regs).regs[2]) _nr;		\
1824 		if ((_regs).regs[2] == __NR_O32_Linux)	\
1825 			_nr = (_regs).regs[4];		\
1826 		else					\
1827 			_nr = (_regs).regs[2];		\
1828 		_nr;					\
1829 	})
1830 # define SYSCALL_NUM_SET(_regs, _nr)			\
1831 	do {						\
1832 		if ((_regs).regs[2] == __NR_O32_Linux)	\
1833 			(_regs).regs[4] = _nr;		\
1834 		else					\
1835 			(_regs).regs[2] = _nr;		\
1836 	} while (0)
1837 # define SYSCALL_RET_SET(_regs, _val)			\
1838 		TH_LOG("Can't modify syscall return on this architecture")
1839 #elif defined(__xtensa__)
1840 # define ARCH_REGS		struct user_pt_regs
1841 # define SYSCALL_NUM(_regs)	(_regs).syscall
1842 /*
1843  * On xtensa syscall return value is in the register
1844  * a2 of the current window which is not fixed.
1845  */
1846 #define SYSCALL_RET(_regs)	(_regs).a[(_regs).windowbase * 4 + 2]
1847 #elif defined(__sh__)
1848 # define ARCH_REGS		struct pt_regs
1849 # define SYSCALL_NUM(_regs)	(_regs).regs[3]
1850 # define SYSCALL_RET(_regs)	(_regs).regs[0]
1851 #elif defined(__mc68000__)
1852 # define ARCH_REGS		struct user_regs_struct
1853 # define SYSCALL_NUM(_regs)	(_regs).orig_d0
1854 # define SYSCALL_RET(_regs)	(_regs).d0
1855 #else
1856 # error "Do not know how to find your architecture's registers and syscalls"
1857 #endif
1858 
1859 /*
1860  * Most architectures can change the syscall by just updating the
1861  * associated register. This is the default if not defined above.
1862  */
1863 #ifndef SYSCALL_NUM_SET
1864 # define SYSCALL_NUM_SET(_regs, _nr)		\
1865 	do {					\
1866 		SYSCALL_NUM(_regs) = (_nr);	\
1867 	} while (0)
1868 #endif
1869 /*
1870  * Most architectures can change the syscall return value by just
1871  * writing to the SYSCALL_RET register. This is the default if not
1872  * defined above. If an architecture cannot set the return value
1873  * (for example when the syscall and return value register is
1874  * shared), report it with TH_LOG() in an arch-specific definition
1875  * of SYSCALL_RET_SET() above, and leave SYSCALL_RET undefined.
1876  */
1877 #if !defined(SYSCALL_RET) && !defined(SYSCALL_RET_SET)
1878 # error "One of SYSCALL_RET or SYSCALL_RET_SET is needed for this arch"
1879 #endif
1880 #ifndef SYSCALL_RET_SET
1881 # define SYSCALL_RET_SET(_regs, _val)		\
1882 	do {					\
1883 		SYSCALL_RET(_regs) = (_val);	\
1884 	} while (0)
1885 #endif
1886 
1887 /* When the syscall return can't be changed, stub out the tests for it. */
1888 #ifndef SYSCALL_RET
1889 # define EXPECT_SYSCALL_RETURN(val, action)	EXPECT_EQ(-1, action)
1890 #else
1891 # define EXPECT_SYSCALL_RETURN(val, action)		\
1892 	do {						\
1893 		errno = 0;				\
1894 		if (val < 0) {				\
1895 			EXPECT_EQ(-1, action);		\
1896 			EXPECT_EQ(-(val), errno);	\
1897 		} else {				\
1898 			EXPECT_EQ(val, action);		\
1899 		}					\
1900 	} while (0)
1901 #endif
1902 
1903 /*
1904  * Some architectures (e.g. powerpc) can only set syscall
1905  * return values on syscall exit during ptrace.
1906  */
1907 const bool ptrace_entry_set_syscall_nr = true;
1908 const bool ptrace_entry_set_syscall_ret =
1909 #ifndef SYSCALL_RET_SET_ON_PTRACE_EXIT
1910 	true;
1911 #else
1912 	false;
1913 #endif
1914 
1915 /*
1916  * Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for
1917  * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux).
1918  */
1919 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__) || defined(__mc68000__)
1920 # define ARCH_GETREGS(_regs)	ptrace(PTRACE_GETREGS, tracee, 0, &(_regs))
1921 # define ARCH_SETREGS(_regs)	ptrace(PTRACE_SETREGS, tracee, 0, &(_regs))
1922 #else
1923 # define ARCH_GETREGS(_regs)	({					\
1924 		struct iovec __v;					\
1925 		__v.iov_base = &(_regs);				\
1926 		__v.iov_len = sizeof(_regs);				\
1927 		ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &__v);	\
1928 	})
1929 # define ARCH_SETREGS(_regs)	({					\
1930 		struct iovec __v;					\
1931 		__v.iov_base = &(_regs);				\
1932 		__v.iov_len = sizeof(_regs);				\
1933 		ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &__v);	\
1934 	})
1935 #endif
1936 
1937 /* Architecture-specific syscall fetching routine. */
1938 int get_syscall(struct __test_metadata *_metadata, pid_t tracee)
1939 {
1940 	ARCH_REGS regs;
1941 
1942 	EXPECT_EQ(0, ARCH_GETREGS(regs)) {
1943 		return -1;
1944 	}
1945 
1946 	return SYSCALL_NUM(regs);
1947 }
1948 
1949 /* Architecture-specific syscall changing routine. */
1950 void __change_syscall(struct __test_metadata *_metadata,
1951 		    pid_t tracee, long *syscall, long *ret)
1952 {
1953 	ARCH_REGS orig, regs;
1954 
1955 	/* Do not get/set registers if we have nothing to do. */
1956 	if (!syscall && !ret)
1957 		return;
1958 
1959 	EXPECT_EQ(0, ARCH_GETREGS(regs)) {
1960 		return;
1961 	}
1962 	orig = regs;
1963 
1964 	if (syscall)
1965 		SYSCALL_NUM_SET(regs, *syscall);
1966 
1967 	if (ret)
1968 		SYSCALL_RET_SET(regs, *ret);
1969 
1970 	/* Flush any register changes made. */
1971 	if (memcmp(&orig, &regs, sizeof(orig)) != 0)
1972 		EXPECT_EQ(0, ARCH_SETREGS(regs));
1973 }
1974 
1975 /* Change only syscall number. */
1976 void change_syscall_nr(struct __test_metadata *_metadata,
1977 		       pid_t tracee, long syscall)
1978 {
1979 	__change_syscall(_metadata, tracee, &syscall, NULL);
1980 }
1981 
1982 /* Change syscall return value (and set syscall number to -1). */
1983 void change_syscall_ret(struct __test_metadata *_metadata,
1984 			pid_t tracee, long ret)
1985 {
1986 	long syscall = -1;
1987 
1988 	__change_syscall(_metadata, tracee, &syscall, &ret);
1989 }
1990 
1991 void tracer_seccomp(struct __test_metadata *_metadata, pid_t tracee,
1992 		    int status, void *args)
1993 {
1994 	int ret;
1995 	unsigned long msg;
1996 
1997 	EXPECT_EQ(PTRACE_EVENT_MASK(status), PTRACE_EVENT_SECCOMP) {
1998 		TH_LOG("Unexpected ptrace event: %d", PTRACE_EVENT_MASK(status));
1999 		return;
2000 	}
2001 
2002 	/* Make sure we got the right message. */
2003 	ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
2004 	EXPECT_EQ(0, ret);
2005 
2006 	/* Validate and take action on expected syscalls. */
2007 	switch (msg) {
2008 	case 0x1002:
2009 		/* change getpid to getppid. */
2010 		EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee));
2011 		change_syscall_nr(_metadata, tracee, __NR_getppid);
2012 		break;
2013 	case 0x1003:
2014 		/* skip gettid with valid return code. */
2015 		EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee));
2016 		change_syscall_ret(_metadata, tracee, 45000);
2017 		break;
2018 	case 0x1004:
2019 		/* skip openat with error. */
2020 		EXPECT_EQ(__NR_openat, get_syscall(_metadata, tracee));
2021 		change_syscall_ret(_metadata, tracee, -ESRCH);
2022 		break;
2023 	case 0x1005:
2024 		/* do nothing (allow getppid) */
2025 		EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee));
2026 		break;
2027 	default:
2028 		EXPECT_EQ(0, msg) {
2029 			TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg);
2030 			kill(tracee, SIGKILL);
2031 		}
2032 	}
2033 
2034 }
2035 
2036 FIXTURE(TRACE_syscall) {
2037 	struct sock_fprog prog;
2038 	pid_t tracer, mytid, mypid, parent;
2039 	long syscall_nr;
2040 };
2041 
2042 void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee,
2043 		   int status, void *args)
2044 {
2045 	int ret;
2046 	unsigned long msg;
2047 	static bool entry;
2048 	long syscall_nr_val, syscall_ret_val;
2049 	long *syscall_nr = NULL, *syscall_ret = NULL;
2050 	FIXTURE_DATA(TRACE_syscall) *self = args;
2051 
2052 	EXPECT_EQ(WSTOPSIG(status) & 0x80, 0x80) {
2053 		TH_LOG("Unexpected WSTOPSIG: %d", WSTOPSIG(status));
2054 		return;
2055 	}
2056 
2057 	/*
2058 	 * The traditional way to tell PTRACE_SYSCALL entry/exit
2059 	 * is by counting.
2060 	 */
2061 	entry = !entry;
2062 
2063 	/* Make sure we got an appropriate message. */
2064 	ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
2065 	EXPECT_EQ(0, ret);
2066 	EXPECT_EQ(entry ? PTRACE_EVENTMSG_SYSCALL_ENTRY
2067 			: PTRACE_EVENTMSG_SYSCALL_EXIT, msg);
2068 
2069 	/*
2070 	 * Some architectures only support setting return values during
2071 	 * syscall exit under ptrace, and on exit the syscall number may
2072 	 * no longer be available. Therefore, save the initial sycall
2073 	 * number here, so it can be examined during both entry and exit
2074 	 * phases.
2075 	 */
2076 	if (entry)
2077 		self->syscall_nr = get_syscall(_metadata, tracee);
2078 
2079 	/*
2080 	 * Depending on the architecture's syscall setting abilities, we
2081 	 * pick which things to set during this phase (entry or exit).
2082 	 */
2083 	if (entry == ptrace_entry_set_syscall_nr)
2084 		syscall_nr = &syscall_nr_val;
2085 	if (entry == ptrace_entry_set_syscall_ret)
2086 		syscall_ret = &syscall_ret_val;
2087 
2088 	/* Now handle the actual rewriting cases. */
2089 	switch (self->syscall_nr) {
2090 	case __NR_getpid:
2091 		syscall_nr_val = __NR_getppid;
2092 		/* Never change syscall return for this case. */
2093 		syscall_ret = NULL;
2094 		break;
2095 	case __NR_gettid:
2096 		syscall_nr_val = -1;
2097 		syscall_ret_val = 45000;
2098 		break;
2099 	case __NR_openat:
2100 		syscall_nr_val = -1;
2101 		syscall_ret_val = -ESRCH;
2102 		break;
2103 	default:
2104 		/* Unhandled, do nothing. */
2105 		return;
2106 	}
2107 
2108 	__change_syscall(_metadata, tracee, syscall_nr, syscall_ret);
2109 }
2110 
2111 FIXTURE_VARIANT(TRACE_syscall) {
2112 	/*
2113 	 * All of the SECCOMP_RET_TRACE behaviors can be tested with either
2114 	 * SECCOMP_RET_TRACE+PTRACE_CONT or plain ptrace()+PTRACE_SYSCALL.
2115 	 * This indicates if we should use SECCOMP_RET_TRACE (false), or
2116 	 * ptrace (true).
2117 	 */
2118 	bool use_ptrace;
2119 };
2120 
2121 FIXTURE_VARIANT_ADD(TRACE_syscall, ptrace) {
2122 	.use_ptrace = true,
2123 };
2124 
2125 FIXTURE_VARIANT_ADD(TRACE_syscall, seccomp) {
2126 	.use_ptrace = false,
2127 };
2128 
2129 FIXTURE_SETUP(TRACE_syscall)
2130 {
2131 	struct sock_filter filter[] = {
2132 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2133 			offsetof(struct seccomp_data, nr)),
2134 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
2135 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002),
2136 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1),
2137 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003),
2138 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_openat, 0, 1),
2139 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004),
2140 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2141 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1005),
2142 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2143 	};
2144 	struct sock_fprog prog = {
2145 		.len = (unsigned short)ARRAY_SIZE(filter),
2146 		.filter = filter,
2147 	};
2148 	long ret;
2149 
2150 	/* Prepare some testable syscall results. */
2151 	self->mytid = syscall(__NR_gettid);
2152 	ASSERT_GT(self->mytid, 0);
2153 	ASSERT_NE(self->mytid, 1) {
2154 		TH_LOG("Running this test as init is not supported. :)");
2155 	}
2156 
2157 	self->mypid = getpid();
2158 	ASSERT_GT(self->mypid, 0);
2159 	ASSERT_EQ(self->mytid, self->mypid);
2160 
2161 	self->parent = getppid();
2162 	ASSERT_GT(self->parent, 0);
2163 	ASSERT_NE(self->parent, self->mypid);
2164 
2165 	/* Launch tracer. */
2166 	self->tracer = setup_trace_fixture(_metadata,
2167 					   variant->use_ptrace ? tracer_ptrace
2168 							       : tracer_seccomp,
2169 					   self, variant->use_ptrace);
2170 
2171 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2172 	ASSERT_EQ(0, ret);
2173 
2174 	/* Do not install seccomp rewrite filters, as we'll use ptrace instead. */
2175 	if (variant->use_ptrace)
2176 		return;
2177 
2178 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2179 	ASSERT_EQ(0, ret);
2180 }
2181 
2182 FIXTURE_TEARDOWN(TRACE_syscall)
2183 {
2184 	teardown_trace_fixture(_metadata, self->tracer);
2185 }
2186 
2187 TEST(negative_ENOSYS)
2188 {
2189 #if defined(__arm__)
2190 	SKIP(return, "arm32 does not support calling syscall -1");
2191 #endif
2192 	/*
2193 	 * There should be no difference between an "internal" skip
2194 	 * and userspace asking for syscall "-1".
2195 	 */
2196 	errno = 0;
2197 	EXPECT_EQ(-1, syscall(-1));
2198 	EXPECT_EQ(errno, ENOSYS);
2199 	/* And no difference for "still not valid but not -1". */
2200 	errno = 0;
2201 	EXPECT_EQ(-1, syscall(-101));
2202 	EXPECT_EQ(errno, ENOSYS);
2203 }
2204 
2205 TEST_F(TRACE_syscall, negative_ENOSYS)
2206 {
2207 	negative_ENOSYS(_metadata);
2208 }
2209 
2210 TEST_F(TRACE_syscall, syscall_allowed)
2211 {
2212 	/* getppid works as expected (no changes). */
2213 	EXPECT_EQ(self->parent, syscall(__NR_getppid));
2214 	EXPECT_NE(self->mypid, syscall(__NR_getppid));
2215 }
2216 
2217 TEST_F(TRACE_syscall, syscall_redirected)
2218 {
2219 	/* getpid has been redirected to getppid as expected. */
2220 	EXPECT_EQ(self->parent, syscall(__NR_getpid));
2221 	EXPECT_NE(self->mypid, syscall(__NR_getpid));
2222 }
2223 
2224 TEST_F(TRACE_syscall, syscall_errno)
2225 {
2226 	/* Tracer should skip the open syscall, resulting in ESRCH. */
2227 	EXPECT_SYSCALL_RETURN(-ESRCH, syscall(__NR_openat));
2228 }
2229 
2230 TEST_F(TRACE_syscall, syscall_faked)
2231 {
2232 	/* Tracer skips the gettid syscall and store altered return value. */
2233 	EXPECT_SYSCALL_RETURN(45000, syscall(__NR_gettid));
2234 }
2235 
2236 TEST_F_SIGNAL(TRACE_syscall, kill_immediate, SIGSYS)
2237 {
2238 	struct sock_filter filter[] = {
2239 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2240 			offsetof(struct seccomp_data, nr)),
2241 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_mknodat, 0, 1),
2242 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD),
2243 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2244 	};
2245 	struct sock_fprog prog = {
2246 		.len = (unsigned short)ARRAY_SIZE(filter),
2247 		.filter = filter,
2248 	};
2249 	long ret;
2250 
2251 	/* Install "kill on mknodat" filter. */
2252 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2253 	ASSERT_EQ(0, ret);
2254 
2255 	/* This should immediately die with SIGSYS, regardless of tracer. */
2256 	EXPECT_EQ(-1, syscall(__NR_mknodat, -1, NULL, 0, 0));
2257 }
2258 
2259 TEST_F(TRACE_syscall, skip_after)
2260 {
2261 	struct sock_filter filter[] = {
2262 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2263 			offsetof(struct seccomp_data, nr)),
2264 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2265 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
2266 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2267 	};
2268 	struct sock_fprog prog = {
2269 		.len = (unsigned short)ARRAY_SIZE(filter),
2270 		.filter = filter,
2271 	};
2272 	long ret;
2273 
2274 	/* Install additional "errno on getppid" filter. */
2275 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2276 	ASSERT_EQ(0, ret);
2277 
2278 	/* Tracer will redirect getpid to getppid, and we should see EPERM. */
2279 	errno = 0;
2280 	EXPECT_EQ(-1, syscall(__NR_getpid));
2281 	EXPECT_EQ(EPERM, errno);
2282 }
2283 
2284 TEST_F_SIGNAL(TRACE_syscall, kill_after, SIGSYS)
2285 {
2286 	struct sock_filter filter[] = {
2287 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2288 			offsetof(struct seccomp_data, nr)),
2289 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2290 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2291 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2292 	};
2293 	struct sock_fprog prog = {
2294 		.len = (unsigned short)ARRAY_SIZE(filter),
2295 		.filter = filter,
2296 	};
2297 	long ret;
2298 
2299 	/* Install additional "death on getppid" filter. */
2300 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2301 	ASSERT_EQ(0, ret);
2302 
2303 	/* Tracer will redirect getpid to getppid, and we should die. */
2304 	EXPECT_NE(self->mypid, syscall(__NR_getpid));
2305 }
2306 
2307 TEST(seccomp_syscall)
2308 {
2309 	struct sock_filter filter[] = {
2310 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2311 	};
2312 	struct sock_fprog prog = {
2313 		.len = (unsigned short)ARRAY_SIZE(filter),
2314 		.filter = filter,
2315 	};
2316 	long ret;
2317 
2318 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2319 	ASSERT_EQ(0, ret) {
2320 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2321 	}
2322 
2323 	/* Reject insane operation. */
2324 	ret = seccomp(-1, 0, &prog);
2325 	ASSERT_NE(ENOSYS, errno) {
2326 		TH_LOG("Kernel does not support seccomp syscall!");
2327 	}
2328 	EXPECT_EQ(EINVAL, errno) {
2329 		TH_LOG("Did not reject crazy op value!");
2330 	}
2331 
2332 	/* Reject strict with flags or pointer. */
2333 	ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL);
2334 	EXPECT_EQ(EINVAL, errno) {
2335 		TH_LOG("Did not reject mode strict with flags!");
2336 	}
2337 	ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog);
2338 	EXPECT_EQ(EINVAL, errno) {
2339 		TH_LOG("Did not reject mode strict with uargs!");
2340 	}
2341 
2342 	/* Reject insane args for filter. */
2343 	ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog);
2344 	EXPECT_EQ(EINVAL, errno) {
2345 		TH_LOG("Did not reject crazy filter flags!");
2346 	}
2347 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL);
2348 	EXPECT_EQ(EFAULT, errno) {
2349 		TH_LOG("Did not reject NULL filter!");
2350 	}
2351 
2352 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2353 	EXPECT_EQ(0, errno) {
2354 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s",
2355 			strerror(errno));
2356 	}
2357 }
2358 
2359 TEST(seccomp_syscall_mode_lock)
2360 {
2361 	struct sock_filter filter[] = {
2362 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2363 	};
2364 	struct sock_fprog prog = {
2365 		.len = (unsigned short)ARRAY_SIZE(filter),
2366 		.filter = filter,
2367 	};
2368 	long ret;
2369 
2370 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2371 	ASSERT_EQ(0, ret) {
2372 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2373 	}
2374 
2375 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2376 	ASSERT_NE(ENOSYS, errno) {
2377 		TH_LOG("Kernel does not support seccomp syscall!");
2378 	}
2379 	EXPECT_EQ(0, ret) {
2380 		TH_LOG("Could not install filter!");
2381 	}
2382 
2383 	/* Make sure neither entry point will switch to strict. */
2384 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0);
2385 	EXPECT_EQ(EINVAL, errno) {
2386 		TH_LOG("Switched to mode strict!");
2387 	}
2388 
2389 	ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL);
2390 	EXPECT_EQ(EINVAL, errno) {
2391 		TH_LOG("Switched to mode strict!");
2392 	}
2393 }
2394 
2395 /*
2396  * Test detection of known and unknown filter flags. Userspace needs to be able
2397  * to check if a filter flag is supported by the current kernel and a good way
2398  * of doing that is by attempting to enter filter mode, with the flag bit in
2399  * question set, and a NULL pointer for the _args_ parameter. EFAULT indicates
2400  * that the flag is valid and EINVAL indicates that the flag is invalid.
2401  */
2402 TEST(detect_seccomp_filter_flags)
2403 {
2404 	unsigned int flags[] = { SECCOMP_FILTER_FLAG_TSYNC,
2405 				 SECCOMP_FILTER_FLAG_LOG,
2406 				 SECCOMP_FILTER_FLAG_SPEC_ALLOW,
2407 				 SECCOMP_FILTER_FLAG_NEW_LISTENER,
2408 				 SECCOMP_FILTER_FLAG_TSYNC_ESRCH };
2409 	unsigned int exclusive[] = {
2410 				SECCOMP_FILTER_FLAG_TSYNC,
2411 				SECCOMP_FILTER_FLAG_NEW_LISTENER };
2412 	unsigned int flag, all_flags, exclusive_mask;
2413 	int i;
2414 	long ret;
2415 
2416 	/* Test detection of individual known-good filter flags */
2417 	for (i = 0, all_flags = 0; i < ARRAY_SIZE(flags); i++) {
2418 		int bits = 0;
2419 
2420 		flag = flags[i];
2421 		/* Make sure the flag is a single bit! */
2422 		while (flag) {
2423 			if (flag & 0x1)
2424 				bits ++;
2425 			flag >>= 1;
2426 		}
2427 		ASSERT_EQ(1, bits);
2428 		flag = flags[i];
2429 
2430 		ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2431 		ASSERT_NE(ENOSYS, errno) {
2432 			TH_LOG("Kernel does not support seccomp syscall!");
2433 		}
2434 		EXPECT_EQ(-1, ret);
2435 		EXPECT_EQ(EFAULT, errno) {
2436 			TH_LOG("Failed to detect that a known-good filter flag (0x%X) is supported!",
2437 			       flag);
2438 		}
2439 
2440 		all_flags |= flag;
2441 	}
2442 
2443 	/*
2444 	 * Test detection of all known-good filter flags combined. But
2445 	 * for the exclusive flags we need to mask them out and try them
2446 	 * individually for the "all flags" testing.
2447 	 */
2448 	exclusive_mask = 0;
2449 	for (i = 0; i < ARRAY_SIZE(exclusive); i++)
2450 		exclusive_mask |= exclusive[i];
2451 	for (i = 0; i < ARRAY_SIZE(exclusive); i++) {
2452 		flag = all_flags & ~exclusive_mask;
2453 		flag |= exclusive[i];
2454 
2455 		ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2456 		EXPECT_EQ(-1, ret);
2457 		EXPECT_EQ(EFAULT, errno) {
2458 			TH_LOG("Failed to detect that all known-good filter flags (0x%X) are supported!",
2459 			       flag);
2460 		}
2461 	}
2462 
2463 	/* Test detection of an unknown filter flags, without exclusives. */
2464 	flag = -1;
2465 	flag &= ~exclusive_mask;
2466 	ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2467 	EXPECT_EQ(-1, ret);
2468 	EXPECT_EQ(EINVAL, errno) {
2469 		TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported!",
2470 		       flag);
2471 	}
2472 
2473 	/*
2474 	 * Test detection of an unknown filter flag that may simply need to be
2475 	 * added to this test
2476 	 */
2477 	flag = flags[ARRAY_SIZE(flags) - 1] << 1;
2478 	ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2479 	EXPECT_EQ(-1, ret);
2480 	EXPECT_EQ(EINVAL, errno) {
2481 		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?",
2482 		       flag);
2483 	}
2484 }
2485 
2486 TEST(TSYNC_first)
2487 {
2488 	struct sock_filter filter[] = {
2489 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2490 	};
2491 	struct sock_fprog prog = {
2492 		.len = (unsigned short)ARRAY_SIZE(filter),
2493 		.filter = filter,
2494 	};
2495 	long ret;
2496 
2497 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2498 	ASSERT_EQ(0, ret) {
2499 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2500 	}
2501 
2502 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2503 		      &prog);
2504 	ASSERT_NE(ENOSYS, errno) {
2505 		TH_LOG("Kernel does not support seccomp syscall!");
2506 	}
2507 	EXPECT_EQ(0, ret) {
2508 		TH_LOG("Could not install initial filter with TSYNC!");
2509 	}
2510 }
2511 
2512 #define TSYNC_SIBLINGS 2
2513 struct tsync_sibling {
2514 	pthread_t tid;
2515 	pid_t system_tid;
2516 	sem_t *started;
2517 	pthread_cond_t *cond;
2518 	pthread_mutex_t *mutex;
2519 	int diverge;
2520 	int num_waits;
2521 	struct sock_fprog *prog;
2522 	struct __test_metadata *metadata;
2523 };
2524 
2525 /*
2526  * To avoid joining joined threads (which is not allowed by Bionic),
2527  * make sure we both successfully join and clear the tid to skip a
2528  * later join attempt during fixture teardown. Any remaining threads
2529  * will be directly killed during teardown.
2530  */
2531 #define PTHREAD_JOIN(tid, status)					\
2532 	do {								\
2533 		int _rc = pthread_join(tid, status);			\
2534 		if (_rc) {						\
2535 			TH_LOG("pthread_join of tid %u failed: %d\n",	\
2536 				(unsigned int)tid, _rc);		\
2537 		} else {						\
2538 			tid = 0;					\
2539 		}							\
2540 	} while (0)
2541 
2542 FIXTURE(TSYNC) {
2543 	struct sock_fprog root_prog, apply_prog;
2544 	struct tsync_sibling sibling[TSYNC_SIBLINGS];
2545 	sem_t started;
2546 	pthread_cond_t cond;
2547 	pthread_mutex_t mutex;
2548 	int sibling_count;
2549 };
2550 
2551 FIXTURE_SETUP(TSYNC)
2552 {
2553 	struct sock_filter root_filter[] = {
2554 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2555 	};
2556 	struct sock_filter apply_filter[] = {
2557 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2558 			offsetof(struct seccomp_data, nr)),
2559 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
2560 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2561 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2562 	};
2563 
2564 	memset(&self->root_prog, 0, sizeof(self->root_prog));
2565 	memset(&self->apply_prog, 0, sizeof(self->apply_prog));
2566 	memset(&self->sibling, 0, sizeof(self->sibling));
2567 	self->root_prog.filter = malloc(sizeof(root_filter));
2568 	ASSERT_NE(NULL, self->root_prog.filter);
2569 	memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter));
2570 	self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter);
2571 
2572 	self->apply_prog.filter = malloc(sizeof(apply_filter));
2573 	ASSERT_NE(NULL, self->apply_prog.filter);
2574 	memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter));
2575 	self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter);
2576 
2577 	self->sibling_count = 0;
2578 	pthread_mutex_init(&self->mutex, NULL);
2579 	pthread_cond_init(&self->cond, NULL);
2580 	sem_init(&self->started, 0, 0);
2581 	self->sibling[0].tid = 0;
2582 	self->sibling[0].cond = &self->cond;
2583 	self->sibling[0].started = &self->started;
2584 	self->sibling[0].mutex = &self->mutex;
2585 	self->sibling[0].diverge = 0;
2586 	self->sibling[0].num_waits = 1;
2587 	self->sibling[0].prog = &self->root_prog;
2588 	self->sibling[0].metadata = _metadata;
2589 	self->sibling[1].tid = 0;
2590 	self->sibling[1].cond = &self->cond;
2591 	self->sibling[1].started = &self->started;
2592 	self->sibling[1].mutex = &self->mutex;
2593 	self->sibling[1].diverge = 0;
2594 	self->sibling[1].prog = &self->root_prog;
2595 	self->sibling[1].num_waits = 1;
2596 	self->sibling[1].metadata = _metadata;
2597 }
2598 
2599 FIXTURE_TEARDOWN(TSYNC)
2600 {
2601 	int sib = 0;
2602 
2603 	if (self->root_prog.filter)
2604 		free(self->root_prog.filter);
2605 	if (self->apply_prog.filter)
2606 		free(self->apply_prog.filter);
2607 
2608 	for ( ; sib < self->sibling_count; ++sib) {
2609 		struct tsync_sibling *s = &self->sibling[sib];
2610 
2611 		if (!s->tid)
2612 			continue;
2613 		/*
2614 		 * If a thread is still running, it may be stuck, so hit
2615 		 * it over the head really hard.
2616 		 */
2617 		pthread_kill(s->tid, 9);
2618 	}
2619 	pthread_mutex_destroy(&self->mutex);
2620 	pthread_cond_destroy(&self->cond);
2621 	sem_destroy(&self->started);
2622 }
2623 
2624 void *tsync_sibling(void *data)
2625 {
2626 	long ret = 0;
2627 	struct tsync_sibling *me = data;
2628 
2629 	me->system_tid = syscall(__NR_gettid);
2630 
2631 	pthread_mutex_lock(me->mutex);
2632 	if (me->diverge) {
2633 		/* Just re-apply the root prog to fork the tree */
2634 		ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
2635 				me->prog, 0, 0);
2636 	}
2637 	sem_post(me->started);
2638 	/* Return outside of started so parent notices failures. */
2639 	if (ret) {
2640 		pthread_mutex_unlock(me->mutex);
2641 		return (void *)SIBLING_EXIT_FAILURE;
2642 	}
2643 	do {
2644 		pthread_cond_wait(me->cond, me->mutex);
2645 		me->num_waits = me->num_waits - 1;
2646 	} while (me->num_waits);
2647 	pthread_mutex_unlock(me->mutex);
2648 
2649 	ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0);
2650 	if (!ret)
2651 		return (void *)SIBLING_EXIT_NEWPRIVS;
2652 	read(-1, NULL, 0);
2653 	return (void *)SIBLING_EXIT_UNKILLED;
2654 }
2655 
2656 void tsync_start_sibling(struct tsync_sibling *sibling)
2657 {
2658 	pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling);
2659 }
2660 
2661 TEST_F(TSYNC, siblings_fail_prctl)
2662 {
2663 	long ret;
2664 	void *status;
2665 	struct sock_filter filter[] = {
2666 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2667 			offsetof(struct seccomp_data, nr)),
2668 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
2669 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL),
2670 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2671 	};
2672 	struct sock_fprog prog = {
2673 		.len = (unsigned short)ARRAY_SIZE(filter),
2674 		.filter = filter,
2675 	};
2676 
2677 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2678 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2679 	}
2680 
2681 	/* Check prctl failure detection by requesting sib 0 diverge. */
2682 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2683 	ASSERT_NE(ENOSYS, errno) {
2684 		TH_LOG("Kernel does not support seccomp syscall!");
2685 	}
2686 	ASSERT_EQ(0, ret) {
2687 		TH_LOG("setting filter failed");
2688 	}
2689 
2690 	self->sibling[0].diverge = 1;
2691 	tsync_start_sibling(&self->sibling[0]);
2692 	tsync_start_sibling(&self->sibling[1]);
2693 
2694 	while (self->sibling_count < TSYNC_SIBLINGS) {
2695 		sem_wait(&self->started);
2696 		self->sibling_count++;
2697 	}
2698 
2699 	/* Signal the threads to clean up*/
2700 	pthread_mutex_lock(&self->mutex);
2701 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2702 		TH_LOG("cond broadcast non-zero");
2703 	}
2704 	pthread_mutex_unlock(&self->mutex);
2705 
2706 	/* Ensure diverging sibling failed to call prctl. */
2707 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2708 	EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status);
2709 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2710 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2711 }
2712 
2713 TEST_F(TSYNC, two_siblings_with_ancestor)
2714 {
2715 	long ret;
2716 	void *status;
2717 
2718 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2719 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2720 	}
2721 
2722 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2723 	ASSERT_NE(ENOSYS, errno) {
2724 		TH_LOG("Kernel does not support seccomp syscall!");
2725 	}
2726 	ASSERT_EQ(0, ret) {
2727 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2728 	}
2729 	tsync_start_sibling(&self->sibling[0]);
2730 	tsync_start_sibling(&self->sibling[1]);
2731 
2732 	while (self->sibling_count < TSYNC_SIBLINGS) {
2733 		sem_wait(&self->started);
2734 		self->sibling_count++;
2735 	}
2736 
2737 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2738 		      &self->apply_prog);
2739 	ASSERT_EQ(0, ret) {
2740 		TH_LOG("Could install filter on all threads!");
2741 	}
2742 	/* Tell the siblings to test the policy */
2743 	pthread_mutex_lock(&self->mutex);
2744 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2745 		TH_LOG("cond broadcast non-zero");
2746 	}
2747 	pthread_mutex_unlock(&self->mutex);
2748 	/* Ensure they are both killed and don't exit cleanly. */
2749 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2750 	EXPECT_EQ(0x0, (long)status);
2751 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2752 	EXPECT_EQ(0x0, (long)status);
2753 }
2754 
2755 TEST_F(TSYNC, two_sibling_want_nnp)
2756 {
2757 	void *status;
2758 
2759 	/* start siblings before any prctl() operations */
2760 	tsync_start_sibling(&self->sibling[0]);
2761 	tsync_start_sibling(&self->sibling[1]);
2762 	while (self->sibling_count < TSYNC_SIBLINGS) {
2763 		sem_wait(&self->started);
2764 		self->sibling_count++;
2765 	}
2766 
2767 	/* Tell the siblings to test no policy */
2768 	pthread_mutex_lock(&self->mutex);
2769 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2770 		TH_LOG("cond broadcast non-zero");
2771 	}
2772 	pthread_mutex_unlock(&self->mutex);
2773 
2774 	/* Ensure they are both upset about lacking nnp. */
2775 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2776 	EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2777 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2778 	EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2779 }
2780 
2781 TEST_F(TSYNC, two_siblings_with_no_filter)
2782 {
2783 	long ret;
2784 	void *status;
2785 
2786 	/* start siblings before any prctl() operations */
2787 	tsync_start_sibling(&self->sibling[0]);
2788 	tsync_start_sibling(&self->sibling[1]);
2789 	while (self->sibling_count < TSYNC_SIBLINGS) {
2790 		sem_wait(&self->started);
2791 		self->sibling_count++;
2792 	}
2793 
2794 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2795 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2796 	}
2797 
2798 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2799 		      &self->apply_prog);
2800 	ASSERT_NE(ENOSYS, errno) {
2801 		TH_LOG("Kernel does not support seccomp syscall!");
2802 	}
2803 	ASSERT_EQ(0, ret) {
2804 		TH_LOG("Could install filter on all threads!");
2805 	}
2806 
2807 	/* Tell the siblings to test the policy */
2808 	pthread_mutex_lock(&self->mutex);
2809 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2810 		TH_LOG("cond broadcast non-zero");
2811 	}
2812 	pthread_mutex_unlock(&self->mutex);
2813 
2814 	/* Ensure they are both killed and don't exit cleanly. */
2815 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2816 	EXPECT_EQ(0x0, (long)status);
2817 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2818 	EXPECT_EQ(0x0, (long)status);
2819 }
2820 
2821 TEST_F(TSYNC, two_siblings_with_one_divergence)
2822 {
2823 	long ret;
2824 	void *status;
2825 
2826 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2827 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2828 	}
2829 
2830 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2831 	ASSERT_NE(ENOSYS, errno) {
2832 		TH_LOG("Kernel does not support seccomp syscall!");
2833 	}
2834 	ASSERT_EQ(0, ret) {
2835 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2836 	}
2837 	self->sibling[0].diverge = 1;
2838 	tsync_start_sibling(&self->sibling[0]);
2839 	tsync_start_sibling(&self->sibling[1]);
2840 
2841 	while (self->sibling_count < TSYNC_SIBLINGS) {
2842 		sem_wait(&self->started);
2843 		self->sibling_count++;
2844 	}
2845 
2846 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2847 		      &self->apply_prog);
2848 	ASSERT_EQ(self->sibling[0].system_tid, ret) {
2849 		TH_LOG("Did not fail on diverged sibling.");
2850 	}
2851 
2852 	/* Wake the threads */
2853 	pthread_mutex_lock(&self->mutex);
2854 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2855 		TH_LOG("cond broadcast non-zero");
2856 	}
2857 	pthread_mutex_unlock(&self->mutex);
2858 
2859 	/* Ensure they are both unkilled. */
2860 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2861 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2862 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2863 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2864 }
2865 
2866 TEST_F(TSYNC, two_siblings_with_one_divergence_no_tid_in_err)
2867 {
2868 	long ret, flags;
2869 	void *status;
2870 
2871 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2872 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2873 	}
2874 
2875 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2876 	ASSERT_NE(ENOSYS, errno) {
2877 		TH_LOG("Kernel does not support seccomp syscall!");
2878 	}
2879 	ASSERT_EQ(0, ret) {
2880 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2881 	}
2882 	self->sibling[0].diverge = 1;
2883 	tsync_start_sibling(&self->sibling[0]);
2884 	tsync_start_sibling(&self->sibling[1]);
2885 
2886 	while (self->sibling_count < TSYNC_SIBLINGS) {
2887 		sem_wait(&self->started);
2888 		self->sibling_count++;
2889 	}
2890 
2891 	flags = SECCOMP_FILTER_FLAG_TSYNC | \
2892 		SECCOMP_FILTER_FLAG_TSYNC_ESRCH;
2893 	ret = seccomp(SECCOMP_SET_MODE_FILTER, flags, &self->apply_prog);
2894 	ASSERT_EQ(ESRCH, errno) {
2895 		TH_LOG("Did not return ESRCH for diverged sibling.");
2896 	}
2897 	ASSERT_EQ(-1, ret) {
2898 		TH_LOG("Did not fail on diverged sibling.");
2899 	}
2900 
2901 	/* Wake the threads */
2902 	pthread_mutex_lock(&self->mutex);
2903 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2904 		TH_LOG("cond broadcast non-zero");
2905 	}
2906 	pthread_mutex_unlock(&self->mutex);
2907 
2908 	/* Ensure they are both unkilled. */
2909 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2910 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2911 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2912 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2913 }
2914 
2915 TEST_F(TSYNC, two_siblings_not_under_filter)
2916 {
2917 	long ret, sib;
2918 	void *status;
2919 	struct timespec delay = { .tv_nsec = 100000000 };
2920 
2921 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2922 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2923 	}
2924 
2925 	/*
2926 	 * Sibling 0 will have its own seccomp policy
2927 	 * and Sibling 1 will not be under seccomp at
2928 	 * all. Sibling 1 will enter seccomp and 0
2929 	 * will cause failure.
2930 	 */
2931 	self->sibling[0].diverge = 1;
2932 	tsync_start_sibling(&self->sibling[0]);
2933 	tsync_start_sibling(&self->sibling[1]);
2934 
2935 	while (self->sibling_count < TSYNC_SIBLINGS) {
2936 		sem_wait(&self->started);
2937 		self->sibling_count++;
2938 	}
2939 
2940 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2941 	ASSERT_NE(ENOSYS, errno) {
2942 		TH_LOG("Kernel does not support seccomp syscall!");
2943 	}
2944 	ASSERT_EQ(0, ret) {
2945 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2946 	}
2947 
2948 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2949 		      &self->apply_prog);
2950 	ASSERT_EQ(ret, self->sibling[0].system_tid) {
2951 		TH_LOG("Did not fail on diverged sibling.");
2952 	}
2953 	sib = 1;
2954 	if (ret == self->sibling[0].system_tid)
2955 		sib = 0;
2956 
2957 	pthread_mutex_lock(&self->mutex);
2958 
2959 	/* Increment the other siblings num_waits so we can clean up
2960 	 * the one we just saw.
2961 	 */
2962 	self->sibling[!sib].num_waits += 1;
2963 
2964 	/* Signal the thread to clean up*/
2965 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2966 		TH_LOG("cond broadcast non-zero");
2967 	}
2968 	pthread_mutex_unlock(&self->mutex);
2969 	PTHREAD_JOIN(self->sibling[sib].tid, &status);
2970 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2971 	/* Poll for actual task death. pthread_join doesn't guarantee it. */
2972 	while (!kill(self->sibling[sib].system_tid, 0))
2973 		nanosleep(&delay, NULL);
2974 	/* Switch to the remaining sibling */
2975 	sib = !sib;
2976 
2977 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2978 		      &self->apply_prog);
2979 	ASSERT_EQ(0, ret) {
2980 		TH_LOG("Expected the remaining sibling to sync");
2981 	};
2982 
2983 	pthread_mutex_lock(&self->mutex);
2984 
2985 	/* If remaining sibling didn't have a chance to wake up during
2986 	 * the first broadcast, manually reduce the num_waits now.
2987 	 */
2988 	if (self->sibling[sib].num_waits > 1)
2989 		self->sibling[sib].num_waits = 1;
2990 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2991 		TH_LOG("cond broadcast non-zero");
2992 	}
2993 	pthread_mutex_unlock(&self->mutex);
2994 	PTHREAD_JOIN(self->sibling[sib].tid, &status);
2995 	EXPECT_EQ(0, (long)status);
2996 	/* Poll for actual task death. pthread_join doesn't guarantee it. */
2997 	while (!kill(self->sibling[sib].system_tid, 0))
2998 		nanosleep(&delay, NULL);
2999 
3000 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
3001 		      &self->apply_prog);
3002 	ASSERT_EQ(0, ret);  /* just us chickens */
3003 }
3004 
3005 /* Make sure restarted syscalls are seen directly as "restart_syscall". */
3006 TEST(syscall_restart)
3007 {
3008 	long ret;
3009 	unsigned long msg;
3010 	pid_t child_pid;
3011 	int pipefd[2];
3012 	int status;
3013 	siginfo_t info = { };
3014 	struct sock_filter filter[] = {
3015 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
3016 			 offsetof(struct seccomp_data, nr)),
3017 
3018 #ifdef __NR_sigreturn
3019 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 7, 0),
3020 #endif
3021 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 6, 0),
3022 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 5, 0),
3023 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 4, 0),
3024 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 5, 0),
3025 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_clock_nanosleep, 4, 0),
3026 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0),
3027 
3028 		/* Allow __NR_write for easy logging. */
3029 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1),
3030 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3031 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
3032 		/* The nanosleep jump target. */
3033 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100),
3034 		/* The restart_syscall jump target. */
3035 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200),
3036 	};
3037 	struct sock_fprog prog = {
3038 		.len = (unsigned short)ARRAY_SIZE(filter),
3039 		.filter = filter,
3040 	};
3041 #if defined(__arm__)
3042 	struct utsname utsbuf;
3043 #endif
3044 
3045 	ASSERT_EQ(0, pipe(pipefd));
3046 
3047 	child_pid = fork();
3048 	ASSERT_LE(0, child_pid);
3049 	if (child_pid == 0) {
3050 		/* Child uses EXPECT not ASSERT to deliver status correctly. */
3051 		char buf = ' ';
3052 		struct timespec timeout = { };
3053 
3054 		/* Attach parent as tracer and stop. */
3055 		EXPECT_EQ(0, ptrace(PTRACE_TRACEME));
3056 		EXPECT_EQ(0, raise(SIGSTOP));
3057 
3058 		EXPECT_EQ(0, close(pipefd[1]));
3059 
3060 		EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
3061 			TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3062 		}
3063 
3064 		ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
3065 		EXPECT_EQ(0, ret) {
3066 			TH_LOG("Failed to install filter!");
3067 		}
3068 
3069 		EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
3070 			TH_LOG("Failed to read() sync from parent");
3071 		}
3072 		EXPECT_EQ('.', buf) {
3073 			TH_LOG("Failed to get sync data from read()");
3074 		}
3075 
3076 		/* Start nanosleep to be interrupted. */
3077 		timeout.tv_sec = 1;
3078 		errno = 0;
3079 		EXPECT_EQ(0, nanosleep(&timeout, NULL)) {
3080 			TH_LOG("Call to nanosleep() failed (errno %d: %s)",
3081 				errno, strerror(errno));
3082 		}
3083 
3084 		/* Read final sync from parent. */
3085 		EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
3086 			TH_LOG("Failed final read() from parent");
3087 		}
3088 		EXPECT_EQ('!', buf) {
3089 			TH_LOG("Failed to get final data from read()");
3090 		}
3091 
3092 		/* Directly report the status of our test harness results. */
3093 		syscall(__NR_exit, _metadata->exit_code);
3094 	}
3095 	EXPECT_EQ(0, close(pipefd[0]));
3096 
3097 	/* Attach to child, setup options, and release. */
3098 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3099 	ASSERT_EQ(true, WIFSTOPPED(status));
3100 	ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL,
3101 			    PTRACE_O_TRACESECCOMP));
3102 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3103 	ASSERT_EQ(1, write(pipefd[1], ".", 1));
3104 
3105 	/* Wait for nanosleep() to start. */
3106 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3107 	ASSERT_EQ(true, WIFSTOPPED(status));
3108 	ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
3109 	ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
3110 	ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
3111 	ASSERT_EQ(0x100, msg);
3112 	ret = get_syscall(_metadata, child_pid);
3113 	EXPECT_TRUE(ret == __NR_nanosleep || ret == __NR_clock_nanosleep);
3114 
3115 	/* Might as well check siginfo for sanity while we're here. */
3116 	ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
3117 	ASSERT_EQ(SIGTRAP, info.si_signo);
3118 	ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code);
3119 	EXPECT_EQ(0, info.si_errno);
3120 	EXPECT_EQ(getuid(), info.si_uid);
3121 	/* Verify signal delivery came from child (seccomp-triggered). */
3122 	EXPECT_EQ(child_pid, info.si_pid);
3123 
3124 	/* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */
3125 	ASSERT_EQ(0, kill(child_pid, SIGSTOP));
3126 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3127 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3128 	ASSERT_EQ(true, WIFSTOPPED(status));
3129 	ASSERT_EQ(SIGSTOP, WSTOPSIG(status));
3130 	ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
3131 	/*
3132 	 * There is no siginfo on SIGSTOP any more, so we can't verify
3133 	 * signal delivery came from parent now (getpid() == info.si_pid).
3134 	 * https://lkml.kernel.org/r/CAGXu5jJaZAOzP1qFz66tYrtbuywqb+UN2SOA1VLHpCCOiYvYeg@mail.gmail.com
3135 	 * At least verify the SIGSTOP via PTRACE_GETSIGINFO.
3136 	 */
3137 	EXPECT_EQ(SIGSTOP, info.si_signo);
3138 
3139 	/* Restart nanosleep with SIGCONT, which triggers restart_syscall. */
3140 	ASSERT_EQ(0, kill(child_pid, SIGCONT));
3141 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3142 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3143 	ASSERT_EQ(true, WIFSTOPPED(status));
3144 	ASSERT_EQ(SIGCONT, WSTOPSIG(status));
3145 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3146 
3147 	/* Wait for restart_syscall() to start. */
3148 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3149 	ASSERT_EQ(true, WIFSTOPPED(status));
3150 	ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
3151 	ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
3152 	ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
3153 
3154 	ASSERT_EQ(0x200, msg);
3155 	ret = get_syscall(_metadata, child_pid);
3156 #if defined(__arm__)
3157 	/*
3158 	 * FIXME:
3159 	 * - native ARM registers do NOT expose true syscall.
3160 	 * - compat ARM registers on ARM64 DO expose true syscall.
3161 	 */
3162 	ASSERT_EQ(0, uname(&utsbuf));
3163 	if (strncmp(utsbuf.machine, "arm", 3) == 0) {
3164 		EXPECT_EQ(__NR_nanosleep, ret);
3165 	} else
3166 #endif
3167 	{
3168 		EXPECT_EQ(__NR_restart_syscall, ret);
3169 	}
3170 
3171 	/* Write again to end test. */
3172 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3173 	ASSERT_EQ(1, write(pipefd[1], "!", 1));
3174 	EXPECT_EQ(0, close(pipefd[1]));
3175 
3176 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3177 	if (WIFSIGNALED(status) || WEXITSTATUS(status))
3178 		_metadata->exit_code = KSFT_FAIL;
3179 }
3180 
3181 TEST_SIGNAL(filter_flag_log, SIGSYS)
3182 {
3183 	struct sock_filter allow_filter[] = {
3184 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3185 	};
3186 	struct sock_filter kill_filter[] = {
3187 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
3188 			offsetof(struct seccomp_data, nr)),
3189 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
3190 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
3191 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3192 	};
3193 	struct sock_fprog allow_prog = {
3194 		.len = (unsigned short)ARRAY_SIZE(allow_filter),
3195 		.filter = allow_filter,
3196 	};
3197 	struct sock_fprog kill_prog = {
3198 		.len = (unsigned short)ARRAY_SIZE(kill_filter),
3199 		.filter = kill_filter,
3200 	};
3201 	long ret;
3202 	pid_t parent = getppid();
3203 
3204 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3205 	ASSERT_EQ(0, ret);
3206 
3207 	/* Verify that the FILTER_FLAG_LOG flag isn't accepted in strict mode */
3208 	ret = seccomp(SECCOMP_SET_MODE_STRICT, SECCOMP_FILTER_FLAG_LOG,
3209 		      &allow_prog);
3210 	ASSERT_NE(ENOSYS, errno) {
3211 		TH_LOG("Kernel does not support seccomp syscall!");
3212 	}
3213 	EXPECT_NE(0, ret) {
3214 		TH_LOG("Kernel accepted FILTER_FLAG_LOG flag in strict mode!");
3215 	}
3216 	EXPECT_EQ(EINVAL, errno) {
3217 		TH_LOG("Kernel returned unexpected errno for FILTER_FLAG_LOG flag in strict mode!");
3218 	}
3219 
3220 	/* Verify that a simple, permissive filter can be added with no flags */
3221 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog);
3222 	EXPECT_EQ(0, ret);
3223 
3224 	/* See if the same filter can be added with the FILTER_FLAG_LOG flag */
3225 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
3226 		      &allow_prog);
3227 	ASSERT_NE(EINVAL, errno) {
3228 		TH_LOG("Kernel does not support the FILTER_FLAG_LOG flag!");
3229 	}
3230 	EXPECT_EQ(0, ret);
3231 
3232 	/* Ensure that the kill filter works with the FILTER_FLAG_LOG flag */
3233 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
3234 		      &kill_prog);
3235 	EXPECT_EQ(0, ret);
3236 
3237 	EXPECT_EQ(parent, syscall(__NR_getppid));
3238 	/* getpid() should never return. */
3239 	EXPECT_EQ(0, syscall(__NR_getpid));
3240 }
3241 
3242 TEST(get_action_avail)
3243 {
3244 	__u32 actions[] = { SECCOMP_RET_KILL_THREAD, SECCOMP_RET_TRAP,
3245 			    SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE,
3246 			    SECCOMP_RET_LOG,   SECCOMP_RET_ALLOW };
3247 	__u32 unknown_action = 0x10000000U;
3248 	int i;
3249 	long ret;
3250 
3251 	ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[0]);
3252 	ASSERT_NE(ENOSYS, errno) {
3253 		TH_LOG("Kernel does not support seccomp syscall!");
3254 	}
3255 	ASSERT_NE(EINVAL, errno) {
3256 		TH_LOG("Kernel does not support SECCOMP_GET_ACTION_AVAIL operation!");
3257 	}
3258 	EXPECT_EQ(ret, 0);
3259 
3260 	for (i = 0; i < ARRAY_SIZE(actions); i++) {
3261 		ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[i]);
3262 		EXPECT_EQ(ret, 0) {
3263 			TH_LOG("Expected action (0x%X) not available!",
3264 			       actions[i]);
3265 		}
3266 	}
3267 
3268 	/* Check that an unknown action is handled properly (EOPNOTSUPP) */
3269 	ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &unknown_action);
3270 	EXPECT_EQ(ret, -1);
3271 	EXPECT_EQ(errno, EOPNOTSUPP);
3272 }
3273 
3274 TEST(get_metadata)
3275 {
3276 	pid_t pid;
3277 	int pipefd[2];
3278 	char buf;
3279 	struct seccomp_metadata md;
3280 	long ret;
3281 
3282 	/* Only real root can get metadata. */
3283 	if (geteuid()) {
3284 		SKIP(return, "get_metadata requires real root");
3285 		return;
3286 	}
3287 
3288 	ASSERT_EQ(0, pipe(pipefd));
3289 
3290 	pid = fork();
3291 	ASSERT_GE(pid, 0);
3292 	if (pid == 0) {
3293 		struct sock_filter filter[] = {
3294 			BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3295 		};
3296 		struct sock_fprog prog = {
3297 			.len = (unsigned short)ARRAY_SIZE(filter),
3298 			.filter = filter,
3299 		};
3300 
3301 		/* one with log, one without */
3302 		EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER,
3303 				     SECCOMP_FILTER_FLAG_LOG, &prog));
3304 		EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog));
3305 
3306 		EXPECT_EQ(0, close(pipefd[0]));
3307 		ASSERT_EQ(1, write(pipefd[1], "1", 1));
3308 		ASSERT_EQ(0, close(pipefd[1]));
3309 
3310 		while (1)
3311 			sleep(100);
3312 	}
3313 
3314 	ASSERT_EQ(0, close(pipefd[1]));
3315 	ASSERT_EQ(1, read(pipefd[0], &buf, 1));
3316 
3317 	ASSERT_EQ(0, ptrace(PTRACE_ATTACH, pid));
3318 	ASSERT_EQ(pid, waitpid(pid, NULL, 0));
3319 
3320 	/* Past here must not use ASSERT or child process is never killed. */
3321 
3322 	md.filter_off = 0;
3323 	errno = 0;
3324 	ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md);
3325 	EXPECT_EQ(sizeof(md), ret) {
3326 		if (errno == EINVAL)
3327 			SKIP(goto skip, "Kernel does not support PTRACE_SECCOMP_GET_METADATA (missing CONFIG_CHECKPOINT_RESTORE?)");
3328 	}
3329 
3330 	EXPECT_EQ(md.flags, SECCOMP_FILTER_FLAG_LOG);
3331 	EXPECT_EQ(md.filter_off, 0);
3332 
3333 	md.filter_off = 1;
3334 	ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md);
3335 	EXPECT_EQ(sizeof(md), ret);
3336 	EXPECT_EQ(md.flags, 0);
3337 	EXPECT_EQ(md.filter_off, 1);
3338 
3339 skip:
3340 	ASSERT_EQ(0, kill(pid, SIGKILL));
3341 }
3342 
3343 static int user_notif_syscall(int nr, unsigned int flags)
3344 {
3345 	struct sock_filter filter[] = {
3346 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
3347 			offsetof(struct seccomp_data, nr)),
3348 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, nr, 0, 1),
3349 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_USER_NOTIF),
3350 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3351 	};
3352 
3353 	struct sock_fprog prog = {
3354 		.len = (unsigned short)ARRAY_SIZE(filter),
3355 		.filter = filter,
3356 	};
3357 
3358 	return seccomp(SECCOMP_SET_MODE_FILTER, flags, &prog);
3359 }
3360 
3361 #define USER_NOTIF_MAGIC INT_MAX
3362 TEST(user_notification_basic)
3363 {
3364 	pid_t pid;
3365 	long ret;
3366 	int status, listener;
3367 	struct seccomp_notif req = {};
3368 	struct seccomp_notif_resp resp = {};
3369 	struct pollfd pollfd;
3370 
3371 	struct sock_filter filter[] = {
3372 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3373 	};
3374 	struct sock_fprog prog = {
3375 		.len = (unsigned short)ARRAY_SIZE(filter),
3376 		.filter = filter,
3377 	};
3378 
3379 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3380 	ASSERT_EQ(0, ret) {
3381 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3382 	}
3383 
3384 	pid = fork();
3385 	ASSERT_GE(pid, 0);
3386 
3387 	/* Check that we get -ENOSYS with no listener attached */
3388 	if (pid == 0) {
3389 		if (user_notif_syscall(__NR_getppid, 0) < 0)
3390 			exit(1);
3391 		ret = syscall(__NR_getppid);
3392 		exit(ret >= 0 || errno != ENOSYS);
3393 	}
3394 
3395 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3396 	EXPECT_EQ(true, WIFEXITED(status));
3397 	EXPECT_EQ(0, WEXITSTATUS(status));
3398 
3399 	/* Add some no-op filters for grins. */
3400 	EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3401 	EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3402 	EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3403 	EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3404 
3405 	/* Check that the basic notification machinery works */
3406 	listener = user_notif_syscall(__NR_getppid,
3407 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
3408 	ASSERT_GE(listener, 0);
3409 
3410 	/* Installing a second listener in the chain should EBUSY */
3411 	EXPECT_EQ(user_notif_syscall(__NR_getppid,
3412 				     SECCOMP_FILTER_FLAG_NEW_LISTENER),
3413 		  -1);
3414 	EXPECT_EQ(errno, EBUSY);
3415 
3416 	pid = fork();
3417 	ASSERT_GE(pid, 0);
3418 
3419 	if (pid == 0) {
3420 		ret = syscall(__NR_getppid);
3421 		exit(ret != USER_NOTIF_MAGIC);
3422 	}
3423 
3424 	pollfd.fd = listener;
3425 	pollfd.events = POLLIN | POLLOUT;
3426 
3427 	EXPECT_GT(poll(&pollfd, 1, -1), 0);
3428 	EXPECT_EQ(pollfd.revents, POLLIN);
3429 
3430 	/* Test that we can't pass garbage to the kernel. */
3431 	memset(&req, 0, sizeof(req));
3432 	req.pid = -1;
3433 	errno = 0;
3434 	ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req);
3435 	EXPECT_EQ(-1, ret);
3436 	EXPECT_EQ(EINVAL, errno);
3437 
3438 	if (ret) {
3439 		req.pid = 0;
3440 		EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3441 	}
3442 
3443 	pollfd.fd = listener;
3444 	pollfd.events = POLLIN | POLLOUT;
3445 
3446 	EXPECT_GT(poll(&pollfd, 1, -1), 0);
3447 	EXPECT_EQ(pollfd.revents, POLLOUT);
3448 
3449 	EXPECT_EQ(req.data.nr,  __NR_getppid);
3450 
3451 	resp.id = req.id;
3452 	resp.error = 0;
3453 	resp.val = USER_NOTIF_MAGIC;
3454 
3455 	/* check that we make sure flags == 0 */
3456 	resp.flags = 1;
3457 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3458 	EXPECT_EQ(errno, EINVAL);
3459 
3460 	resp.flags = 0;
3461 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3462 
3463 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3464 	EXPECT_EQ(true, WIFEXITED(status));
3465 	EXPECT_EQ(0, WEXITSTATUS(status));
3466 }
3467 
3468 TEST(user_notification_with_tsync)
3469 {
3470 	int ret;
3471 	unsigned int flags;
3472 
3473 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3474 	ASSERT_EQ(0, ret) {
3475 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3476 	}
3477 
3478 	/* these were exclusive */
3479 	flags = SECCOMP_FILTER_FLAG_NEW_LISTENER |
3480 		SECCOMP_FILTER_FLAG_TSYNC;
3481 	ASSERT_EQ(-1, user_notif_syscall(__NR_getppid, flags));
3482 	ASSERT_EQ(EINVAL, errno);
3483 
3484 	/* but now they're not */
3485 	flags |= SECCOMP_FILTER_FLAG_TSYNC_ESRCH;
3486 	ret = user_notif_syscall(__NR_getppid, flags);
3487 	close(ret);
3488 	ASSERT_LE(0, ret);
3489 }
3490 
3491 TEST(user_notification_kill_in_middle)
3492 {
3493 	pid_t pid;
3494 	long ret;
3495 	int listener;
3496 	struct seccomp_notif req = {};
3497 	struct seccomp_notif_resp resp = {};
3498 
3499 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3500 	ASSERT_EQ(0, ret) {
3501 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3502 	}
3503 
3504 	listener = user_notif_syscall(__NR_getppid,
3505 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
3506 	ASSERT_GE(listener, 0);
3507 
3508 	/*
3509 	 * Check that nothing bad happens when we kill the task in the middle
3510 	 * of a syscall.
3511 	 */
3512 	pid = fork();
3513 	ASSERT_GE(pid, 0);
3514 
3515 	if (pid == 0) {
3516 		ret = syscall(__NR_getppid);
3517 		exit(ret != USER_NOTIF_MAGIC);
3518 	}
3519 
3520 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3521 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), 0);
3522 
3523 	EXPECT_EQ(kill(pid, SIGKILL), 0);
3524 	EXPECT_EQ(waitpid(pid, NULL, 0), pid);
3525 
3526 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), -1);
3527 
3528 	resp.id = req.id;
3529 	ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp);
3530 	EXPECT_EQ(ret, -1);
3531 	EXPECT_EQ(errno, ENOENT);
3532 }
3533 
3534 static int handled = -1;
3535 
3536 static void signal_handler(int signal)
3537 {
3538 	if (write(handled, "c", 1) != 1)
3539 		perror("write from signal");
3540 }
3541 
3542 TEST(user_notification_signal)
3543 {
3544 	pid_t pid;
3545 	long ret;
3546 	int status, listener, sk_pair[2];
3547 	struct seccomp_notif req = {};
3548 	struct seccomp_notif_resp resp = {};
3549 	char c;
3550 
3551 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3552 	ASSERT_EQ(0, ret) {
3553 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3554 	}
3555 
3556 	ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0);
3557 
3558 	listener = user_notif_syscall(__NR_gettid,
3559 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
3560 	ASSERT_GE(listener, 0);
3561 
3562 	pid = fork();
3563 	ASSERT_GE(pid, 0);
3564 
3565 	if (pid == 0) {
3566 		close(sk_pair[0]);
3567 		handled = sk_pair[1];
3568 		if (signal(SIGUSR1, signal_handler) == SIG_ERR) {
3569 			perror("signal");
3570 			exit(1);
3571 		}
3572 		/*
3573 		 * ERESTARTSYS behavior is a bit hard to test, because we need
3574 		 * to rely on a signal that has not yet been handled. Let's at
3575 		 * least check that the error code gets propagated through, and
3576 		 * hope that it doesn't break when there is actually a signal :)
3577 		 */
3578 		ret = syscall(__NR_gettid);
3579 		exit(!(ret == -1 && errno == 512));
3580 	}
3581 
3582 	close(sk_pair[1]);
3583 
3584 	memset(&req, 0, sizeof(req));
3585 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3586 
3587 	EXPECT_EQ(kill(pid, SIGUSR1), 0);
3588 
3589 	/*
3590 	 * Make sure the signal really is delivered, which means we're not
3591 	 * stuck in the user notification code any more and the notification
3592 	 * should be dead.
3593 	 */
3594 	EXPECT_EQ(read(sk_pair[0], &c, 1), 1);
3595 
3596 	resp.id = req.id;
3597 	resp.error = -EPERM;
3598 	resp.val = 0;
3599 
3600 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3601 	EXPECT_EQ(errno, ENOENT);
3602 
3603 	memset(&req, 0, sizeof(req));
3604 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3605 
3606 	resp.id = req.id;
3607 	resp.error = -512; /* -ERESTARTSYS */
3608 	resp.val = 0;
3609 
3610 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3611 
3612 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3613 	EXPECT_EQ(true, WIFEXITED(status));
3614 	EXPECT_EQ(0, WEXITSTATUS(status));
3615 }
3616 
3617 TEST(user_notification_closed_listener)
3618 {
3619 	pid_t pid;
3620 	long ret;
3621 	int status, listener;
3622 
3623 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3624 	ASSERT_EQ(0, ret) {
3625 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3626 	}
3627 
3628 	listener = user_notif_syscall(__NR_getppid,
3629 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
3630 	ASSERT_GE(listener, 0);
3631 
3632 	/*
3633 	 * Check that we get an ENOSYS when the listener is closed.
3634 	 */
3635 	pid = fork();
3636 	ASSERT_GE(pid, 0);
3637 	if (pid == 0) {
3638 		close(listener);
3639 		ret = syscall(__NR_getppid);
3640 		exit(ret != -1 && errno != ENOSYS);
3641 	}
3642 
3643 	close(listener);
3644 
3645 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3646 	EXPECT_EQ(true, WIFEXITED(status));
3647 	EXPECT_EQ(0, WEXITSTATUS(status));
3648 }
3649 
3650 /*
3651  * Check that a pid in a child namespace still shows up as valid in ours.
3652  */
3653 TEST(user_notification_child_pid_ns)
3654 {
3655 	pid_t pid;
3656 	int status, listener;
3657 	struct seccomp_notif req = {};
3658 	struct seccomp_notif_resp resp = {};
3659 
3660 	ASSERT_EQ(unshare(CLONE_NEWUSER | CLONE_NEWPID), 0) {
3661 		if (errno == EINVAL)
3662 			SKIP(return, "kernel missing CLONE_NEWUSER support");
3663 	};
3664 
3665 	listener = user_notif_syscall(__NR_getppid,
3666 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
3667 	ASSERT_GE(listener, 0);
3668 
3669 	pid = fork();
3670 	ASSERT_GE(pid, 0);
3671 
3672 	if (pid == 0)
3673 		exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3674 
3675 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3676 	EXPECT_EQ(req.pid, pid);
3677 
3678 	resp.id = req.id;
3679 	resp.error = 0;
3680 	resp.val = USER_NOTIF_MAGIC;
3681 
3682 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3683 
3684 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3685 	EXPECT_EQ(true, WIFEXITED(status));
3686 	EXPECT_EQ(0, WEXITSTATUS(status));
3687 	close(listener);
3688 }
3689 
3690 /*
3691  * Check that a pid in a sibling (i.e. unrelated) namespace shows up as 0, i.e.
3692  * invalid.
3693  */
3694 TEST(user_notification_sibling_pid_ns)
3695 {
3696 	pid_t pid, pid2;
3697 	int status, listener;
3698 	struct seccomp_notif req = {};
3699 	struct seccomp_notif_resp resp = {};
3700 
3701 	ASSERT_EQ(prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0), 0) {
3702 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3703 	}
3704 
3705 	listener = user_notif_syscall(__NR_getppid,
3706 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
3707 	ASSERT_GE(listener, 0);
3708 
3709 	pid = fork();
3710 	ASSERT_GE(pid, 0);
3711 
3712 	if (pid == 0) {
3713 		ASSERT_EQ(unshare(CLONE_NEWPID), 0) {
3714 			if (errno == EPERM)
3715 				SKIP(return, "CLONE_NEWPID requires CAP_SYS_ADMIN");
3716 			else if (errno == EINVAL)
3717 				SKIP(return, "CLONE_NEWPID is invalid (missing CONFIG_PID_NS?)");
3718 		}
3719 
3720 		pid2 = fork();
3721 		ASSERT_GE(pid2, 0);
3722 
3723 		if (pid2 == 0)
3724 			exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3725 
3726 		EXPECT_EQ(waitpid(pid2, &status, 0), pid2);
3727 		EXPECT_EQ(true, WIFEXITED(status));
3728 		EXPECT_EQ(0, WEXITSTATUS(status));
3729 		exit(WEXITSTATUS(status));
3730 	}
3731 
3732 	/* Create the sibling ns, and sibling in it. */
3733 	ASSERT_EQ(unshare(CLONE_NEWPID), 0) {
3734 		if (errno == EPERM)
3735 			SKIP(return, "CLONE_NEWPID requires CAP_SYS_ADMIN");
3736 		else if (errno == EINVAL)
3737 			SKIP(return, "CLONE_NEWPID is invalid (missing CONFIG_PID_NS?)");
3738 	}
3739 	ASSERT_EQ(errno, 0);
3740 
3741 	pid2 = fork();
3742 	ASSERT_GE(pid2, 0);
3743 
3744 	if (pid2 == 0) {
3745 		ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3746 		/*
3747 		 * The pid should be 0, i.e. the task is in some namespace that
3748 		 * we can't "see".
3749 		 */
3750 		EXPECT_EQ(req.pid, 0);
3751 
3752 		resp.id = req.id;
3753 		resp.error = 0;
3754 		resp.val = USER_NOTIF_MAGIC;
3755 
3756 		ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3757 		exit(0);
3758 	}
3759 
3760 	close(listener);
3761 
3762 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3763 	EXPECT_EQ(true, WIFEXITED(status));
3764 	EXPECT_EQ(0, WEXITSTATUS(status));
3765 
3766 	EXPECT_EQ(waitpid(pid2, &status, 0), pid2);
3767 	EXPECT_EQ(true, WIFEXITED(status));
3768 	EXPECT_EQ(0, WEXITSTATUS(status));
3769 }
3770 
3771 TEST(user_notification_fault_recv)
3772 {
3773 	pid_t pid;
3774 	int status, listener;
3775 	struct seccomp_notif req = {};
3776 	struct seccomp_notif_resp resp = {};
3777 
3778 	ASSERT_EQ(unshare(CLONE_NEWUSER), 0) {
3779 		if (errno == EINVAL)
3780 			SKIP(return, "kernel missing CLONE_NEWUSER support");
3781 	}
3782 
3783 	listener = user_notif_syscall(__NR_getppid,
3784 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
3785 	ASSERT_GE(listener, 0);
3786 
3787 	pid = fork();
3788 	ASSERT_GE(pid, 0);
3789 
3790 	if (pid == 0)
3791 		exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3792 
3793 	/* Do a bad recv() */
3794 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, NULL), -1);
3795 	EXPECT_EQ(errno, EFAULT);
3796 
3797 	/* We should still be able to receive this notification, though. */
3798 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3799 	EXPECT_EQ(req.pid, pid);
3800 
3801 	resp.id = req.id;
3802 	resp.error = 0;
3803 	resp.val = USER_NOTIF_MAGIC;
3804 
3805 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3806 
3807 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3808 	EXPECT_EQ(true, WIFEXITED(status));
3809 	EXPECT_EQ(0, WEXITSTATUS(status));
3810 }
3811 
3812 TEST(seccomp_get_notif_sizes)
3813 {
3814 	struct seccomp_notif_sizes sizes;
3815 
3816 	ASSERT_EQ(seccomp(SECCOMP_GET_NOTIF_SIZES, 0, &sizes), 0);
3817 	EXPECT_EQ(sizes.seccomp_notif, sizeof(struct seccomp_notif));
3818 	EXPECT_EQ(sizes.seccomp_notif_resp, sizeof(struct seccomp_notif_resp));
3819 }
3820 
3821 TEST(user_notification_continue)
3822 {
3823 	pid_t pid;
3824 	long ret;
3825 	int status, listener;
3826 	struct seccomp_notif req = {};
3827 	struct seccomp_notif_resp resp = {};
3828 	struct pollfd pollfd;
3829 
3830 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3831 	ASSERT_EQ(0, ret) {
3832 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3833 	}
3834 
3835 	listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER);
3836 	ASSERT_GE(listener, 0);
3837 
3838 	pid = fork();
3839 	ASSERT_GE(pid, 0);
3840 
3841 	if (pid == 0) {
3842 		int dup_fd, pipe_fds[2];
3843 		pid_t self;
3844 
3845 		ASSERT_GE(pipe(pipe_fds), 0);
3846 
3847 		dup_fd = dup(pipe_fds[0]);
3848 		ASSERT_GE(dup_fd, 0);
3849 		EXPECT_NE(pipe_fds[0], dup_fd);
3850 
3851 		self = getpid();
3852 		ASSERT_EQ(filecmp(self, self, pipe_fds[0], dup_fd), 0);
3853 		exit(0);
3854 	}
3855 
3856 	pollfd.fd = listener;
3857 	pollfd.events = POLLIN | POLLOUT;
3858 
3859 	EXPECT_GT(poll(&pollfd, 1, -1), 0);
3860 	EXPECT_EQ(pollfd.revents, POLLIN);
3861 
3862 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3863 
3864 	pollfd.fd = listener;
3865 	pollfd.events = POLLIN | POLLOUT;
3866 
3867 	EXPECT_GT(poll(&pollfd, 1, -1), 0);
3868 	EXPECT_EQ(pollfd.revents, POLLOUT);
3869 
3870 	EXPECT_EQ(req.data.nr, __NR_dup);
3871 
3872 	resp.id = req.id;
3873 	resp.flags = SECCOMP_USER_NOTIF_FLAG_CONTINUE;
3874 
3875 	/*
3876 	 * Verify that setting SECCOMP_USER_NOTIF_FLAG_CONTINUE enforces other
3877 	 * args be set to 0.
3878 	 */
3879 	resp.error = 0;
3880 	resp.val = USER_NOTIF_MAGIC;
3881 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3882 	EXPECT_EQ(errno, EINVAL);
3883 
3884 	resp.error = USER_NOTIF_MAGIC;
3885 	resp.val = 0;
3886 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3887 	EXPECT_EQ(errno, EINVAL);
3888 
3889 	resp.error = 0;
3890 	resp.val = 0;
3891 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0) {
3892 		if (errno == EINVAL)
3893 			SKIP(goto skip, "Kernel does not support SECCOMP_USER_NOTIF_FLAG_CONTINUE");
3894 	}
3895 
3896 skip:
3897 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3898 	EXPECT_EQ(true, WIFEXITED(status));
3899 	EXPECT_EQ(0, WEXITSTATUS(status)) {
3900 		if (WEXITSTATUS(status) == 2) {
3901 			SKIP(return, "Kernel does not support kcmp() syscall");
3902 			return;
3903 		}
3904 	}
3905 }
3906 
3907 TEST(user_notification_filter_empty)
3908 {
3909 	pid_t pid;
3910 	long ret;
3911 	int status;
3912 	struct pollfd pollfd;
3913 	struct __clone_args args = {
3914 		.flags = CLONE_FILES,
3915 		.exit_signal = SIGCHLD,
3916 	};
3917 
3918 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3919 	ASSERT_EQ(0, ret) {
3920 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3921 	}
3922 
3923 	if (__NR_clone3 < 0)
3924 		SKIP(return, "Test not built with clone3 support");
3925 
3926 	pid = sys_clone3(&args, sizeof(args));
3927 	ASSERT_GE(pid, 0);
3928 
3929 	if (pid == 0) {
3930 		int listener;
3931 
3932 		listener = user_notif_syscall(__NR_mknodat, SECCOMP_FILTER_FLAG_NEW_LISTENER);
3933 		if (listener < 0)
3934 			_exit(EXIT_FAILURE);
3935 
3936 		if (dup2(listener, 200) != 200)
3937 			_exit(EXIT_FAILURE);
3938 
3939 		close(listener);
3940 
3941 		_exit(EXIT_SUCCESS);
3942 	}
3943 
3944 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3945 	EXPECT_EQ(true, WIFEXITED(status));
3946 	EXPECT_EQ(0, WEXITSTATUS(status));
3947 
3948 	/*
3949 	 * The seccomp filter has become unused so we should be notified once
3950 	 * the kernel gets around to cleaning up task struct.
3951 	 */
3952 	pollfd.fd = 200;
3953 	pollfd.events = POLLHUP;
3954 
3955 	EXPECT_GT(poll(&pollfd, 1, 2000), 0);
3956 	EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0);
3957 }
3958 
3959 TEST(user_ioctl_notification_filter_empty)
3960 {
3961 	pid_t pid;
3962 	long ret;
3963 	int status, p[2];
3964 	struct __clone_args args = {
3965 		.flags = CLONE_FILES,
3966 		.exit_signal = SIGCHLD,
3967 	};
3968 	struct seccomp_notif req = {};
3969 
3970 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3971 	ASSERT_EQ(0, ret) {
3972 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3973 	}
3974 
3975 	if (__NR_clone3 < 0)
3976 		SKIP(return, "Test not built with clone3 support");
3977 
3978 	ASSERT_EQ(0, pipe(p));
3979 
3980 	pid = sys_clone3(&args, sizeof(args));
3981 	ASSERT_GE(pid, 0);
3982 
3983 	if (pid == 0) {
3984 		int listener;
3985 
3986 		listener = user_notif_syscall(__NR_mknodat, SECCOMP_FILTER_FLAG_NEW_LISTENER);
3987 		if (listener < 0)
3988 			_exit(EXIT_FAILURE);
3989 
3990 		if (dup2(listener, 200) != 200)
3991 			_exit(EXIT_FAILURE);
3992 		close(p[1]);
3993 		close(listener);
3994 		sleep(1);
3995 
3996 		_exit(EXIT_SUCCESS);
3997 	}
3998 	if (read(p[0], &status, 1) != 0)
3999 		_exit(EXIT_SUCCESS);
4000 	close(p[0]);
4001 	/*
4002 	 * The seccomp filter has become unused so we should be notified once
4003 	 * the kernel gets around to cleaning up task struct.
4004 	 */
4005 	EXPECT_EQ(ioctl(200, SECCOMP_IOCTL_NOTIF_RECV, &req), -1);
4006 	EXPECT_EQ(errno, ENOENT);
4007 
4008 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
4009 	EXPECT_EQ(true, WIFEXITED(status));
4010 	EXPECT_EQ(0, WEXITSTATUS(status));
4011 }
4012 
4013 static void *do_thread(void *data)
4014 {
4015 	return NULL;
4016 }
4017 
4018 TEST(user_notification_filter_empty_threaded)
4019 {
4020 	pid_t pid;
4021 	long ret;
4022 	int status;
4023 	struct pollfd pollfd;
4024 	struct __clone_args args = {
4025 		.flags = CLONE_FILES,
4026 		.exit_signal = SIGCHLD,
4027 	};
4028 
4029 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4030 	ASSERT_EQ(0, ret) {
4031 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4032 	}
4033 
4034 	if (__NR_clone3 < 0)
4035 		SKIP(return, "Test not built with clone3 support");
4036 
4037 	pid = sys_clone3(&args, sizeof(args));
4038 	ASSERT_GE(pid, 0);
4039 
4040 	if (pid == 0) {
4041 		pid_t pid1, pid2;
4042 		int listener, status;
4043 		pthread_t thread;
4044 
4045 		listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER);
4046 		if (listener < 0)
4047 			_exit(EXIT_FAILURE);
4048 
4049 		if (dup2(listener, 200) != 200)
4050 			_exit(EXIT_FAILURE);
4051 
4052 		close(listener);
4053 
4054 		pid1 = fork();
4055 		if (pid1 < 0)
4056 			_exit(EXIT_FAILURE);
4057 
4058 		if (pid1 == 0)
4059 			_exit(EXIT_SUCCESS);
4060 
4061 		pid2 = fork();
4062 		if (pid2 < 0)
4063 			_exit(EXIT_FAILURE);
4064 
4065 		if (pid2 == 0)
4066 			_exit(EXIT_SUCCESS);
4067 
4068 		if (pthread_create(&thread, NULL, do_thread, NULL) ||
4069 		    pthread_join(thread, NULL))
4070 			_exit(EXIT_FAILURE);
4071 
4072 		if (pthread_create(&thread, NULL, do_thread, NULL) ||
4073 		    pthread_join(thread, NULL))
4074 			_exit(EXIT_FAILURE);
4075 
4076 		if (waitpid(pid1, &status, 0) != pid1 || !WIFEXITED(status) ||
4077 		    WEXITSTATUS(status))
4078 			_exit(EXIT_FAILURE);
4079 
4080 		if (waitpid(pid2, &status, 0) != pid2 || !WIFEXITED(status) ||
4081 		    WEXITSTATUS(status))
4082 			_exit(EXIT_FAILURE);
4083 
4084 		exit(EXIT_SUCCESS);
4085 	}
4086 
4087 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
4088 	EXPECT_EQ(true, WIFEXITED(status));
4089 	EXPECT_EQ(0, WEXITSTATUS(status));
4090 
4091 	/*
4092 	 * The seccomp filter has become unused so we should be notified once
4093 	 * the kernel gets around to cleaning up task struct.
4094 	 */
4095 	pollfd.fd = 200;
4096 	pollfd.events = POLLHUP;
4097 
4098 	EXPECT_GT(poll(&pollfd, 1, 2000), 0);
4099 	EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0);
4100 }
4101 
4102 
4103 int get_next_fd(int prev_fd)
4104 {
4105 	for (int i = prev_fd + 1; i < FD_SETSIZE; ++i) {
4106 		if (fcntl(i, F_GETFD) == -1)
4107 			return i;
4108 	}
4109 	_exit(EXIT_FAILURE);
4110 }
4111 
4112 TEST(user_notification_addfd)
4113 {
4114 	pid_t pid;
4115 	long ret;
4116 	int status, listener, memfd, fd, nextfd;
4117 	struct seccomp_notif_addfd addfd = {};
4118 	struct seccomp_notif_addfd_small small = {};
4119 	struct seccomp_notif_addfd_big big = {};
4120 	struct seccomp_notif req = {};
4121 	struct seccomp_notif_resp resp = {};
4122 	/* 100 ms */
4123 	struct timespec delay = { .tv_nsec = 100000000 };
4124 
4125 	/* There may be arbitrary already-open fds at test start. */
4126 	memfd = memfd_create("test", 0);
4127 	ASSERT_GE(memfd, 0);
4128 	nextfd = get_next_fd(memfd);
4129 
4130 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4131 	ASSERT_EQ(0, ret) {
4132 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4133 	}
4134 
4135 	/* fd: 4 */
4136 	/* Check that the basic notification machinery works */
4137 	listener = user_notif_syscall(__NR_getppid,
4138 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
4139 	ASSERT_EQ(listener, nextfd);
4140 	nextfd = get_next_fd(nextfd);
4141 
4142 	pid = fork();
4143 	ASSERT_GE(pid, 0);
4144 
4145 	if (pid == 0) {
4146 		/* fds will be added and this value is expected */
4147 		if (syscall(__NR_getppid) != USER_NOTIF_MAGIC)
4148 			exit(1);
4149 
4150 		/* Atomic addfd+send is received here. Check it is a valid fd */
4151 		if (fcntl(syscall(__NR_getppid), F_GETFD) == -1)
4152 			exit(1);
4153 
4154 		exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
4155 	}
4156 
4157 	ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4158 
4159 	addfd.srcfd = memfd;
4160 	addfd.newfd = 0;
4161 	addfd.id = req.id;
4162 	addfd.flags = 0x0;
4163 
4164 	/* Verify bad newfd_flags cannot be set */
4165 	addfd.newfd_flags = ~O_CLOEXEC;
4166 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4167 	EXPECT_EQ(errno, EINVAL);
4168 	addfd.newfd_flags = O_CLOEXEC;
4169 
4170 	/* Verify bad flags cannot be set */
4171 	addfd.flags = 0xff;
4172 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4173 	EXPECT_EQ(errno, EINVAL);
4174 	addfd.flags = 0;
4175 
4176 	/* Verify that remote_fd cannot be set without setting flags */
4177 	addfd.newfd = 1;
4178 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4179 	EXPECT_EQ(errno, EINVAL);
4180 	addfd.newfd = 0;
4181 
4182 	/* Verify small size cannot be set */
4183 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_SMALL, &small), -1);
4184 	EXPECT_EQ(errno, EINVAL);
4185 
4186 	/* Verify we can't send bits filled in unknown buffer area */
4187 	memset(&big, 0xAA, sizeof(big));
4188 	big.addfd = addfd;
4189 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big), -1);
4190 	EXPECT_EQ(errno, E2BIG);
4191 
4192 
4193 	/* Verify we can set an arbitrary remote fd */
4194 	fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd);
4195 	EXPECT_EQ(fd, nextfd);
4196 	nextfd = get_next_fd(nextfd);
4197 	EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0);
4198 
4199 	/* Verify we can set an arbitrary remote fd with large size */
4200 	memset(&big, 0x0, sizeof(big));
4201 	big.addfd = addfd;
4202 	fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big);
4203 	EXPECT_EQ(fd, nextfd);
4204 	nextfd = get_next_fd(nextfd);
4205 
4206 	/* Verify we can set a specific remote fd */
4207 	addfd.newfd = 42;
4208 	addfd.flags = SECCOMP_ADDFD_FLAG_SETFD;
4209 	fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd);
4210 	EXPECT_EQ(fd, 42);
4211 	EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0);
4212 
4213 	/* Resume syscall */
4214 	resp.id = req.id;
4215 	resp.error = 0;
4216 	resp.val = USER_NOTIF_MAGIC;
4217 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4218 
4219 	/*
4220 	 * This sets the ID of the ADD FD to the last request plus 1. The
4221 	 * notification ID increments 1 per notification.
4222 	 */
4223 	addfd.id = req.id + 1;
4224 
4225 	/* This spins until the underlying notification is generated */
4226 	while (ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd) != -1 &&
4227 	       errno != -EINPROGRESS)
4228 		nanosleep(&delay, NULL);
4229 
4230 	memset(&req, 0, sizeof(req));
4231 	ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4232 	ASSERT_EQ(addfd.id, req.id);
4233 
4234 	/* Verify we can do an atomic addfd and send */
4235 	addfd.newfd = 0;
4236 	addfd.flags = SECCOMP_ADDFD_FLAG_SEND;
4237 	fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd);
4238 	/*
4239 	 * Child has earlier "low" fds and now 42, so we expect the next
4240 	 * lowest available fd to be assigned here.
4241 	 */
4242 	EXPECT_EQ(fd, nextfd);
4243 	nextfd = get_next_fd(nextfd);
4244 	ASSERT_EQ(filecmp(getpid(), pid, memfd, fd), 0);
4245 
4246 	/*
4247 	 * This sets the ID of the ADD FD to the last request plus 1. The
4248 	 * notification ID increments 1 per notification.
4249 	 */
4250 	addfd.id = req.id + 1;
4251 
4252 	/* This spins until the underlying notification is generated */
4253 	while (ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd) != -1 &&
4254 	       errno != -EINPROGRESS)
4255 		nanosleep(&delay, NULL);
4256 
4257 	memset(&req, 0, sizeof(req));
4258 	ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4259 	ASSERT_EQ(addfd.id, req.id);
4260 
4261 	resp.id = req.id;
4262 	resp.error = 0;
4263 	resp.val = USER_NOTIF_MAGIC;
4264 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4265 
4266 	/* Wait for child to finish. */
4267 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
4268 	EXPECT_EQ(true, WIFEXITED(status));
4269 	EXPECT_EQ(0, WEXITSTATUS(status));
4270 
4271 	close(memfd);
4272 }
4273 
4274 TEST(user_notification_addfd_rlimit)
4275 {
4276 	pid_t pid;
4277 	long ret;
4278 	int status, listener, memfd;
4279 	struct seccomp_notif_addfd addfd = {};
4280 	struct seccomp_notif req = {};
4281 	struct seccomp_notif_resp resp = {};
4282 	const struct rlimit lim = {
4283 		.rlim_cur	= 0,
4284 		.rlim_max	= 0,
4285 	};
4286 
4287 	memfd = memfd_create("test", 0);
4288 	ASSERT_GE(memfd, 0);
4289 
4290 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4291 	ASSERT_EQ(0, ret) {
4292 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4293 	}
4294 
4295 	/* Check that the basic notification machinery works */
4296 	listener = user_notif_syscall(__NR_getppid,
4297 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
4298 	ASSERT_GE(listener, 0);
4299 
4300 	pid = fork();
4301 	ASSERT_GE(pid, 0);
4302 
4303 	if (pid == 0)
4304 		exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
4305 
4306 
4307 	ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4308 
4309 	ASSERT_EQ(prlimit(pid, RLIMIT_NOFILE, &lim, NULL), 0);
4310 
4311 	addfd.srcfd = memfd;
4312 	addfd.newfd_flags = O_CLOEXEC;
4313 	addfd.newfd = 0;
4314 	addfd.id = req.id;
4315 	addfd.flags = 0;
4316 
4317 	/* Should probably spot check /proc/sys/fs/file-nr */
4318 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4319 	EXPECT_EQ(errno, EMFILE);
4320 
4321 	addfd.flags = SECCOMP_ADDFD_FLAG_SEND;
4322 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4323 	EXPECT_EQ(errno, EMFILE);
4324 
4325 	addfd.newfd = 100;
4326 	addfd.flags = SECCOMP_ADDFD_FLAG_SETFD;
4327 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4328 	EXPECT_EQ(errno, EBADF);
4329 
4330 	resp.id = req.id;
4331 	resp.error = 0;
4332 	resp.val = USER_NOTIF_MAGIC;
4333 
4334 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4335 
4336 	/* Wait for child to finish. */
4337 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
4338 	EXPECT_EQ(true, WIFEXITED(status));
4339 	EXPECT_EQ(0, WEXITSTATUS(status));
4340 
4341 	close(memfd);
4342 }
4343 
4344 #ifndef SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP
4345 #define SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP (1UL << 0)
4346 #define SECCOMP_IOCTL_NOTIF_SET_FLAGS  SECCOMP_IOW(4, __u64)
4347 #endif
4348 
4349 TEST(user_notification_sync)
4350 {
4351 	struct seccomp_notif req = {};
4352 	struct seccomp_notif_resp resp = {};
4353 	int status, listener;
4354 	pid_t pid;
4355 	long ret;
4356 
4357 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4358 	ASSERT_EQ(0, ret) {
4359 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4360 	}
4361 
4362 	listener = user_notif_syscall(__NR_getppid,
4363 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
4364 	ASSERT_GE(listener, 0);
4365 
4366 	/* Try to set invalid flags. */
4367 	EXPECT_SYSCALL_RETURN(-EINVAL,
4368 		ioctl(listener, SECCOMP_IOCTL_NOTIF_SET_FLAGS, 0xffffffff, 0));
4369 
4370 	ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SET_FLAGS,
4371 			SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP, 0), 0);
4372 
4373 	pid = fork();
4374 	ASSERT_GE(pid, 0);
4375 	if (pid == 0) {
4376 		ret = syscall(__NR_getppid);
4377 		ASSERT_EQ(ret, USER_NOTIF_MAGIC) {
4378 			_exit(1);
4379 		}
4380 		_exit(0);
4381 	}
4382 
4383 	req.pid = 0;
4384 	ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4385 
4386 	ASSERT_EQ(req.data.nr,  __NR_getppid);
4387 
4388 	resp.id = req.id;
4389 	resp.error = 0;
4390 	resp.val = USER_NOTIF_MAGIC;
4391 	resp.flags = 0;
4392 	ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4393 
4394 	ASSERT_EQ(waitpid(pid, &status, 0), pid);
4395 	ASSERT_EQ(status, 0);
4396 }
4397 
4398 
4399 /* Make sure PTRACE_O_SUSPEND_SECCOMP requires CAP_SYS_ADMIN. */
4400 FIXTURE(O_SUSPEND_SECCOMP) {
4401 	pid_t pid;
4402 };
4403 
4404 FIXTURE_SETUP(O_SUSPEND_SECCOMP)
4405 {
4406 	ERRNO_FILTER(block_read, E2BIG);
4407 	cap_value_t cap_list[] = { CAP_SYS_ADMIN };
4408 	cap_t caps;
4409 
4410 	self->pid = 0;
4411 
4412 	/* make sure we don't have CAP_SYS_ADMIN */
4413 	caps = cap_get_proc();
4414 	ASSERT_NE(NULL, caps);
4415 	ASSERT_EQ(0, cap_set_flag(caps, CAP_EFFECTIVE, 1, cap_list, CAP_CLEAR));
4416 	ASSERT_EQ(0, cap_set_proc(caps));
4417 	cap_free(caps);
4418 
4419 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0));
4420 	ASSERT_EQ(0, prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_block_read));
4421 
4422 	self->pid = fork();
4423 	ASSERT_GE(self->pid, 0);
4424 
4425 	if (self->pid == 0) {
4426 		while (1)
4427 			pause();
4428 		_exit(127);
4429 	}
4430 }
4431 
4432 FIXTURE_TEARDOWN(O_SUSPEND_SECCOMP)
4433 {
4434 	if (self->pid)
4435 		kill(self->pid, SIGKILL);
4436 }
4437 
4438 TEST_F(O_SUSPEND_SECCOMP, setoptions)
4439 {
4440 	int wstatus;
4441 
4442 	ASSERT_EQ(0, ptrace(PTRACE_ATTACH, self->pid, NULL, 0));
4443 	ASSERT_EQ(self->pid, wait(&wstatus));
4444 	ASSERT_EQ(-1, ptrace(PTRACE_SETOPTIONS, self->pid, NULL, PTRACE_O_SUSPEND_SECCOMP));
4445 	if (errno == EINVAL)
4446 		SKIP(return, "Kernel does not support PTRACE_O_SUSPEND_SECCOMP (missing CONFIG_CHECKPOINT_RESTORE?)");
4447 	ASSERT_EQ(EPERM, errno);
4448 }
4449 
4450 TEST_F(O_SUSPEND_SECCOMP, seize)
4451 {
4452 	int ret;
4453 
4454 	ret = ptrace(PTRACE_SEIZE, self->pid, NULL, PTRACE_O_SUSPEND_SECCOMP);
4455 	ASSERT_EQ(-1, ret);
4456 	if (errno == EINVAL)
4457 		SKIP(return, "Kernel does not support PTRACE_O_SUSPEND_SECCOMP (missing CONFIG_CHECKPOINT_RESTORE?)");
4458 	ASSERT_EQ(EPERM, errno);
4459 }
4460 
4461 /*
4462  * get_nth - Get the nth, space separated entry in a file.
4463  *
4464  * Returns the length of the read field.
4465  * Throws error if field is zero-lengthed.
4466  */
4467 static ssize_t get_nth(struct __test_metadata *_metadata, const char *path,
4468 		     const unsigned int position, char **entry)
4469 {
4470 	char *line = NULL;
4471 	unsigned int i;
4472 	ssize_t nread;
4473 	size_t len = 0;
4474 	FILE *f;
4475 
4476 	f = fopen(path, "r");
4477 	ASSERT_NE(f, NULL) {
4478 		TH_LOG("Could not open %s: %s", path, strerror(errno));
4479 	}
4480 
4481 	for (i = 0; i < position; i++) {
4482 		nread = getdelim(&line, &len, ' ', f);
4483 		ASSERT_GE(nread, 0) {
4484 			TH_LOG("Failed to read %d entry in file %s", i, path);
4485 		}
4486 	}
4487 	fclose(f);
4488 
4489 	ASSERT_GT(nread, 0) {
4490 		TH_LOG("Entry in file %s had zero length", path);
4491 	}
4492 
4493 	*entry = line;
4494 	return nread - 1;
4495 }
4496 
4497 /* For a given PID, get the task state (D, R, etc...) */
4498 static char get_proc_stat(struct __test_metadata *_metadata, pid_t pid)
4499 {
4500 	char proc_path[100] = {0};
4501 	char status;
4502 	char *line;
4503 
4504 	snprintf(proc_path, sizeof(proc_path), "/proc/%d/stat", pid);
4505 	ASSERT_EQ(get_nth(_metadata, proc_path, 3, &line), 1);
4506 
4507 	status = *line;
4508 	free(line);
4509 
4510 	return status;
4511 }
4512 
4513 TEST(user_notification_fifo)
4514 {
4515 	struct seccomp_notif_resp resp = {};
4516 	struct seccomp_notif req = {};
4517 	int i, status, listener;
4518 	pid_t pid, pids[3];
4519 	__u64 baseid;
4520 	long ret;
4521 	/* 100 ms */
4522 	struct timespec delay = { .tv_nsec = 100000000 };
4523 
4524 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4525 	ASSERT_EQ(0, ret) {
4526 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4527 	}
4528 
4529 	/* Setup a listener */
4530 	listener = user_notif_syscall(__NR_getppid,
4531 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
4532 	ASSERT_GE(listener, 0);
4533 
4534 	pid = fork();
4535 	ASSERT_GE(pid, 0);
4536 
4537 	if (pid == 0) {
4538 		ret = syscall(__NR_getppid);
4539 		exit(ret != USER_NOTIF_MAGIC);
4540 	}
4541 
4542 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4543 	baseid = req.id + 1;
4544 
4545 	resp.id = req.id;
4546 	resp.error = 0;
4547 	resp.val = USER_NOTIF_MAGIC;
4548 
4549 	/* check that we make sure flags == 0 */
4550 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4551 
4552 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
4553 	EXPECT_EQ(true, WIFEXITED(status));
4554 	EXPECT_EQ(0, WEXITSTATUS(status));
4555 
4556 	/* Start children, and generate notifications */
4557 	for (i = 0; i < ARRAY_SIZE(pids); i++) {
4558 		pid = fork();
4559 		if (pid == 0) {
4560 			ret = syscall(__NR_getppid);
4561 			exit(ret != USER_NOTIF_MAGIC);
4562 		}
4563 		pids[i] = pid;
4564 	}
4565 
4566 	/* This spins until all of the children are sleeping */
4567 restart_wait:
4568 	for (i = 0; i < ARRAY_SIZE(pids); i++) {
4569 		if (get_proc_stat(_metadata, pids[i]) != 'S') {
4570 			nanosleep(&delay, NULL);
4571 			goto restart_wait;
4572 		}
4573 	}
4574 
4575 	/* Read the notifications in order (and respond) */
4576 	for (i = 0; i < ARRAY_SIZE(pids); i++) {
4577 		memset(&req, 0, sizeof(req));
4578 		EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4579 		EXPECT_EQ(req.id, baseid + i);
4580 		resp.id = req.id;
4581 		EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4582 	}
4583 
4584 	/* Make sure notifications were received */
4585 	for (i = 0; i < ARRAY_SIZE(pids); i++) {
4586 		EXPECT_EQ(waitpid(pids[i], &status, 0), pids[i]);
4587 		EXPECT_EQ(true, WIFEXITED(status));
4588 		EXPECT_EQ(0, WEXITSTATUS(status));
4589 	}
4590 }
4591 
4592 /* get_proc_syscall - Get the syscall in progress for a given pid
4593  *
4594  * Returns the current syscall number for a given process
4595  * Returns -1 if not in syscall (running or blocked)
4596  */
4597 static long get_proc_syscall(struct __test_metadata *_metadata, int pid)
4598 {
4599 	char proc_path[100] = {0};
4600 	long ret = -1;
4601 	ssize_t nread;
4602 	char *line;
4603 
4604 	snprintf(proc_path, sizeof(proc_path), "/proc/%d/syscall", pid);
4605 	nread = get_nth(_metadata, proc_path, 1, &line);
4606 	ASSERT_GT(nread, 0);
4607 
4608 	if (!strncmp("running", line, MIN(7, nread)))
4609 		ret = strtol(line, NULL, 16);
4610 
4611 	free(line);
4612 	return ret;
4613 }
4614 
4615 /* Ensure non-fatal signals prior to receive are unmodified */
4616 TEST(user_notification_wait_killable_pre_notification)
4617 {
4618 	struct sigaction new_action = {
4619 		.sa_handler = signal_handler,
4620 	};
4621 	int listener, status, sk_pair[2];
4622 	pid_t pid;
4623 	long ret;
4624 	char c;
4625 	/* 100 ms */
4626 	struct timespec delay = { .tv_nsec = 100000000 };
4627 
4628 	ASSERT_EQ(sigemptyset(&new_action.sa_mask), 0);
4629 
4630 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4631 	ASSERT_EQ(0, ret)
4632 	{
4633 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4634 	}
4635 
4636 	ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0);
4637 
4638 	listener = user_notif_syscall(
4639 		__NR_getppid, SECCOMP_FILTER_FLAG_NEW_LISTENER |
4640 				      SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV);
4641 	ASSERT_GE(listener, 0);
4642 
4643 	/*
4644 	 * Check that we can kill the process with SIGUSR1 prior to receiving
4645 	 * the notification. SIGUSR1 is wired up to a custom signal handler,
4646 	 * and make sure it gets called.
4647 	 */
4648 	pid = fork();
4649 	ASSERT_GE(pid, 0);
4650 
4651 	if (pid == 0) {
4652 		close(sk_pair[0]);
4653 		handled = sk_pair[1];
4654 
4655 		/* Setup the non-fatal sigaction without SA_RESTART */
4656 		if (sigaction(SIGUSR1, &new_action, NULL)) {
4657 			perror("sigaction");
4658 			exit(1);
4659 		}
4660 
4661 		ret = syscall(__NR_getppid);
4662 		/* Make sure we got a return from a signal interruption */
4663 		exit(ret != -1 || errno != EINTR);
4664 	}
4665 
4666 	/*
4667 	 * Make sure we've gotten to the seccomp user notification wait
4668 	 * from getppid prior to sending any signals
4669 	 */
4670 	while (get_proc_syscall(_metadata, pid) != __NR_getppid &&
4671 	       get_proc_stat(_metadata, pid) != 'S')
4672 		nanosleep(&delay, NULL);
4673 
4674 	/* Send non-fatal kill signal */
4675 	EXPECT_EQ(kill(pid, SIGUSR1), 0);
4676 
4677 	/* wait for process to exit (exit checks for EINTR) */
4678 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
4679 	EXPECT_EQ(true, WIFEXITED(status));
4680 	EXPECT_EQ(0, WEXITSTATUS(status));
4681 
4682 	EXPECT_EQ(read(sk_pair[0], &c, 1), 1);
4683 }
4684 
4685 /* Ensure non-fatal signals after receive are blocked */
4686 TEST(user_notification_wait_killable)
4687 {
4688 	struct sigaction new_action = {
4689 		.sa_handler = signal_handler,
4690 	};
4691 	struct seccomp_notif_resp resp = {};
4692 	struct seccomp_notif req = {};
4693 	int listener, status, sk_pair[2];
4694 	pid_t pid;
4695 	long ret;
4696 	char c;
4697 	/* 100 ms */
4698 	struct timespec delay = { .tv_nsec = 100000000 };
4699 
4700 	ASSERT_EQ(sigemptyset(&new_action.sa_mask), 0);
4701 
4702 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4703 	ASSERT_EQ(0, ret)
4704 	{
4705 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4706 	}
4707 
4708 	ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0);
4709 
4710 	listener = user_notif_syscall(
4711 		__NR_getppid, SECCOMP_FILTER_FLAG_NEW_LISTENER |
4712 				      SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV);
4713 	ASSERT_GE(listener, 0);
4714 
4715 	pid = fork();
4716 	ASSERT_GE(pid, 0);
4717 
4718 	if (pid == 0) {
4719 		close(sk_pair[0]);
4720 		handled = sk_pair[1];
4721 
4722 		/* Setup the sigaction without SA_RESTART */
4723 		if (sigaction(SIGUSR1, &new_action, NULL)) {
4724 			perror("sigaction");
4725 			exit(1);
4726 		}
4727 
4728 		/* Make sure that the syscall is completed (no EINTR) */
4729 		ret = syscall(__NR_getppid);
4730 		exit(ret != USER_NOTIF_MAGIC);
4731 	}
4732 
4733 	/*
4734 	 * Get the notification, to make move the notifying process into a
4735 	 * non-preemptible (TASK_KILLABLE) state.
4736 	 */
4737 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4738 	/* Send non-fatal kill signal */
4739 	EXPECT_EQ(kill(pid, SIGUSR1), 0);
4740 
4741 	/*
4742 	 * Make sure the task enters moves to TASK_KILLABLE by waiting for
4743 	 * D (Disk Sleep) state after receiving non-fatal signal.
4744 	 */
4745 	while (get_proc_stat(_metadata, pid) != 'D')
4746 		nanosleep(&delay, NULL);
4747 
4748 	resp.id = req.id;
4749 	resp.val = USER_NOTIF_MAGIC;
4750 	/* Make sure the notification is found and able to be replied to */
4751 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4752 
4753 	/*
4754 	 * Make sure that the signal handler does get called once we're back in
4755 	 * userspace.
4756 	 */
4757 	EXPECT_EQ(read(sk_pair[0], &c, 1), 1);
4758 	/* wait for process to exit (exit checks for USER_NOTIF_MAGIC) */
4759 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
4760 	EXPECT_EQ(true, WIFEXITED(status));
4761 	EXPECT_EQ(0, WEXITSTATUS(status));
4762 }
4763 
4764 /* Ensure fatal signals after receive are not blocked */
4765 TEST(user_notification_wait_killable_fatal)
4766 {
4767 	struct seccomp_notif req = {};
4768 	int listener, status;
4769 	pid_t pid;
4770 	long ret;
4771 	/* 100 ms */
4772 	struct timespec delay = { .tv_nsec = 100000000 };
4773 
4774 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4775 	ASSERT_EQ(0, ret)
4776 	{
4777 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4778 	}
4779 
4780 	listener = user_notif_syscall(
4781 		__NR_getppid, SECCOMP_FILTER_FLAG_NEW_LISTENER |
4782 				      SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV);
4783 	ASSERT_GE(listener, 0);
4784 
4785 	pid = fork();
4786 	ASSERT_GE(pid, 0);
4787 
4788 	if (pid == 0) {
4789 		/* This should never complete as it should get a SIGTERM */
4790 		syscall(__NR_getppid);
4791 		exit(1);
4792 	}
4793 
4794 	while (get_proc_stat(_metadata, pid) != 'S')
4795 		nanosleep(&delay, NULL);
4796 
4797 	/*
4798 	 * Get the notification, to make move the notifying process into a
4799 	 * non-preemptible (TASK_KILLABLE) state.
4800 	 */
4801 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4802 	/* Kill the process with a fatal signal */
4803 	EXPECT_EQ(kill(pid, SIGTERM), 0);
4804 
4805 	/*
4806 	 * Wait for the process to exit, and make sure the process terminated
4807 	 * due to the SIGTERM signal.
4808 	 */
4809 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
4810 	EXPECT_EQ(true, WIFSIGNALED(status));
4811 	EXPECT_EQ(SIGTERM, WTERMSIG(status));
4812 }
4813 
4814 struct tsync_vs_thread_leader_args {
4815 	pthread_t leader;
4816 };
4817 
4818 static void *tsync_vs_dead_thread_leader_sibling(void *_args)
4819 {
4820 	struct sock_filter allow_filter[] = {
4821 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
4822 	};
4823 	struct sock_fprog allow_prog = {
4824 		.len = (unsigned short)ARRAY_SIZE(allow_filter),
4825 		.filter = allow_filter,
4826 	};
4827 	struct tsync_vs_thread_leader_args *args = _args;
4828 	void *retval;
4829 	long ret;
4830 
4831 	ret = pthread_join(args->leader, &retval);
4832 	if (ret)
4833 		exit(1);
4834 	if (retval != _args)
4835 		exit(2);
4836 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, &allow_prog);
4837 	if (ret)
4838 		exit(3);
4839 
4840 	exit(0);
4841 }
4842 
4843 /*
4844  * Ensure that a dead thread leader doesn't prevent installing new filters with
4845  * SECCOMP_FILTER_FLAG_TSYNC from other threads.
4846  */
4847 TEST(tsync_vs_dead_thread_leader)
4848 {
4849 	int status;
4850 	pid_t pid;
4851 	long ret;
4852 
4853 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4854 	ASSERT_EQ(0, ret) {
4855 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4856 	}
4857 
4858 	pid = fork();
4859 	ASSERT_GE(pid, 0);
4860 
4861 	if (pid == 0) {
4862 		struct sock_filter allow_filter[] = {
4863 			BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
4864 		};
4865 		struct sock_fprog allow_prog = {
4866 			.len = (unsigned short)ARRAY_SIZE(allow_filter),
4867 			.filter = allow_filter,
4868 		};
4869 		struct  tsync_vs_thread_leader_args *args;
4870 		pthread_t sibling;
4871 
4872 		args = malloc(sizeof(*args));
4873 		ASSERT_NE(NULL, args);
4874 		args->leader = pthread_self();
4875 
4876 		ret = pthread_create(&sibling, NULL,
4877 				     tsync_vs_dead_thread_leader_sibling, args);
4878 		ASSERT_EQ(0, ret);
4879 
4880 		/* Install a new filter just to the leader thread. */
4881 		ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog);
4882 		ASSERT_EQ(0, ret);
4883 		pthread_exit(args);
4884 		exit(1);
4885 	}
4886 
4887 	EXPECT_EQ(pid, waitpid(pid, &status, 0));
4888 	EXPECT_EQ(0, status);
4889 }
4890 
4891 /*
4892  * TODO:
4893  * - expand NNP testing
4894  * - better arch-specific TRACE and TRAP handlers.
4895  * - endianness checking when appropriate
4896  * - 64-bit arg prodding
4897  * - arch value testing (x86 modes especially)
4898  * - verify that FILTER_FLAG_LOG filters generate log messages
4899  * - verify that RET_LOG generates log messages
4900  */
4901 
4902 TEST_HARNESS_MAIN
4903