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