1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2012 The Chromium OS Authors. All rights reserved.
4  *
5  * Test code for seccomp bpf.
6  */
7 
8 #define _GNU_SOURCE
9 #include <sys/types.h>
10 
11 /*
12  * glibc 2.26 and later have SIGSYS in siginfo_t. Before that,
13  * we need to use the kernel's siginfo.h file and trick glibc
14  * into accepting it.
15  */
16 #if !__GLIBC_PREREQ(2, 26)
17 # include <asm/siginfo.h>
18 # define __have_siginfo_t 1
19 # define __have_sigval_t 1
20 # define __have_sigevent_t 1
21 #endif
22 
23 #include <errno.h>
24 #include <linux/filter.h>
25 #include <sys/prctl.h>
26 #include <sys/ptrace.h>
27 #include <sys/user.h>
28 #include <linux/prctl.h>
29 #include <linux/ptrace.h>
30 #include <linux/seccomp.h>
31 #include <pthread.h>
32 #include <semaphore.h>
33 #include <signal.h>
34 #include <stddef.h>
35 #include <stdbool.h>
36 #include <string.h>
37 #include <time.h>
38 #include <limits.h>
39 #include <linux/elf.h>
40 #include <sys/uio.h>
41 #include <sys/utsname.h>
42 #include <sys/fcntl.h>
43 #include <sys/mman.h>
44 #include <sys/times.h>
45 #include <sys/socket.h>
46 #include <sys/ioctl.h>
47 #include <linux/kcmp.h>
48 #include <sys/resource.h>
49 #include <sys/capability.h>
50 
51 #include <unistd.h>
52 #include <sys/syscall.h>
53 #include <poll.h>
54 
55 #include "../kselftest_harness.h"
56 #include "../clone3/clone3_selftests.h"
57 
58 /* Attempt to de-conflict with the selftests tree. */
59 #ifndef SKIP
60 #define SKIP(s, ...)	XFAIL(s, ##__VA_ARGS__)
61 #endif
62 
63 #define MIN(X, Y) ((X) < (Y) ? (X) : (Y))
64 
65 #ifndef PR_SET_PTRACER
66 # define PR_SET_PTRACER 0x59616d61
67 #endif
68 
69 #ifndef PR_SET_NO_NEW_PRIVS
70 #define PR_SET_NO_NEW_PRIVS 38
71 #define PR_GET_NO_NEW_PRIVS 39
72 #endif
73 
74 #ifndef PR_SECCOMP_EXT
75 #define PR_SECCOMP_EXT 43
76 #endif
77 
78 #ifndef SECCOMP_EXT_ACT
79 #define SECCOMP_EXT_ACT 1
80 #endif
81 
82 #ifndef SECCOMP_EXT_ACT_TSYNC
83 #define SECCOMP_EXT_ACT_TSYNC 1
84 #endif
85 
86 #ifndef SECCOMP_MODE_STRICT
87 #define SECCOMP_MODE_STRICT 1
88 #endif
89 
90 #ifndef SECCOMP_MODE_FILTER
91 #define SECCOMP_MODE_FILTER 2
92 #endif
93 
94 #ifndef SECCOMP_RET_ALLOW
95 struct seccomp_data {
96 	int nr;
97 	__u32 arch;
98 	__u64 instruction_pointer;
99 	__u64 args[6];
100 };
101 #endif
102 
103 #ifndef SECCOMP_RET_KILL_PROCESS
104 #define SECCOMP_RET_KILL_PROCESS 0x80000000U /* kill the process */
105 #define SECCOMP_RET_KILL_THREAD	 0x00000000U /* kill the thread */
106 #endif
107 #ifndef SECCOMP_RET_KILL
108 #define SECCOMP_RET_KILL	 SECCOMP_RET_KILL_THREAD
109 #define SECCOMP_RET_TRAP	 0x00030000U /* disallow and force a SIGSYS */
110 #define SECCOMP_RET_ERRNO	 0x00050000U /* returns an errno */
111 #define SECCOMP_RET_TRACE	 0x7ff00000U /* pass to a tracer or disallow */
112 #define SECCOMP_RET_ALLOW	 0x7fff0000U /* allow */
113 #endif
114 #ifndef SECCOMP_RET_LOG
115 #define SECCOMP_RET_LOG		 0x7ffc0000U /* allow after logging */
116 #endif
117 
118 #ifndef __NR_seccomp
119 # if defined(__i386__)
120 #  define __NR_seccomp 354
121 # elif defined(__x86_64__)
122 #  define __NR_seccomp 317
123 # elif defined(__arm__)
124 #  define __NR_seccomp 383
125 # elif defined(__aarch64__)
126 #  define __NR_seccomp 277
127 # elif defined(__riscv)
128 #  define __NR_seccomp 277
129 # elif defined(__csky__)
130 #  define __NR_seccomp 277
131 # elif defined(__hppa__)
132 #  define __NR_seccomp 338
133 # elif defined(__powerpc__)
134 #  define __NR_seccomp 358
135 # elif defined(__s390__)
136 #  define __NR_seccomp 348
137 # elif defined(__xtensa__)
138 #  define __NR_seccomp 337
139 # elif defined(__sh__)
140 #  define __NR_seccomp 372
141 # else
142 #  warning "seccomp syscall number unknown for this architecture"
143 #  define __NR_seccomp 0xffff
144 # endif
145 #endif
146 
147 #ifndef SECCOMP_SET_MODE_STRICT
148 #define SECCOMP_SET_MODE_STRICT 0
149 #endif
150 
151 #ifndef SECCOMP_SET_MODE_FILTER
152 #define SECCOMP_SET_MODE_FILTER 1
153 #endif
154 
155 #ifndef SECCOMP_GET_ACTION_AVAIL
156 #define SECCOMP_GET_ACTION_AVAIL 2
157 #endif
158 
159 #ifndef SECCOMP_GET_NOTIF_SIZES
160 #define SECCOMP_GET_NOTIF_SIZES 3
161 #endif
162 
163 #ifndef SECCOMP_FILTER_FLAG_TSYNC
164 #define SECCOMP_FILTER_FLAG_TSYNC (1UL << 0)
165 #endif
166 
167 #ifndef SECCOMP_FILTER_FLAG_LOG
168 #define SECCOMP_FILTER_FLAG_LOG (1UL << 1)
169 #endif
170 
171 #ifndef SECCOMP_FILTER_FLAG_SPEC_ALLOW
172 #define SECCOMP_FILTER_FLAG_SPEC_ALLOW (1UL << 2)
173 #endif
174 
175 #ifndef PTRACE_SECCOMP_GET_METADATA
176 #define PTRACE_SECCOMP_GET_METADATA	0x420d
177 
178 struct seccomp_metadata {
179 	__u64 filter_off;       /* Input: which filter */
180 	__u64 flags;             /* Output: filter's flags */
181 };
182 #endif
183 
184 #ifndef SECCOMP_FILTER_FLAG_NEW_LISTENER
185 #define SECCOMP_FILTER_FLAG_NEW_LISTENER	(1UL << 3)
186 #endif
187 
188 #ifndef SECCOMP_RET_USER_NOTIF
189 #define SECCOMP_RET_USER_NOTIF 0x7fc00000U
190 
191 #define SECCOMP_IOC_MAGIC		'!'
192 #define SECCOMP_IO(nr)			_IO(SECCOMP_IOC_MAGIC, nr)
193 #define SECCOMP_IOR(nr, type)		_IOR(SECCOMP_IOC_MAGIC, nr, type)
194 #define SECCOMP_IOW(nr, type)		_IOW(SECCOMP_IOC_MAGIC, nr, type)
195 #define SECCOMP_IOWR(nr, type)		_IOWR(SECCOMP_IOC_MAGIC, nr, type)
196 
197 /* Flags for seccomp notification fd ioctl. */
198 #define SECCOMP_IOCTL_NOTIF_RECV	SECCOMP_IOWR(0, struct seccomp_notif)
199 #define SECCOMP_IOCTL_NOTIF_SEND	SECCOMP_IOWR(1,	\
200 						struct seccomp_notif_resp)
201 #define SECCOMP_IOCTL_NOTIF_ID_VALID	SECCOMP_IOW(2, __u64)
202 
203 struct seccomp_notif {
204 	__u64 id;
205 	__u32 pid;
206 	__u32 flags;
207 	struct seccomp_data data;
208 };
209 
210 struct seccomp_notif_resp {
211 	__u64 id;
212 	__s64 val;
213 	__s32 error;
214 	__u32 flags;
215 };
216 
217 struct seccomp_notif_sizes {
218 	__u16 seccomp_notif;
219 	__u16 seccomp_notif_resp;
220 	__u16 seccomp_data;
221 };
222 #endif
223 
224 #ifndef SECCOMP_IOCTL_NOTIF_ADDFD
225 /* On success, the return value is the remote process's added fd number */
226 #define SECCOMP_IOCTL_NOTIF_ADDFD	SECCOMP_IOW(3,	\
227 						struct seccomp_notif_addfd)
228 
229 /* valid flags for seccomp_notif_addfd */
230 #define SECCOMP_ADDFD_FLAG_SETFD	(1UL << 0) /* Specify remote fd */
231 
232 struct seccomp_notif_addfd {
233 	__u64 id;
234 	__u32 flags;
235 	__u32 srcfd;
236 	__u32 newfd;
237 	__u32 newfd_flags;
238 };
239 #endif
240 
241 #ifndef SECCOMP_ADDFD_FLAG_SEND
242 #define SECCOMP_ADDFD_FLAG_SEND	(1UL << 1) /* Addfd and return it, atomically */
243 #endif
244 
245 struct seccomp_notif_addfd_small {
246 	__u64 id;
247 	char weird[4];
248 };
249 #define SECCOMP_IOCTL_NOTIF_ADDFD_SMALL	\
250 	SECCOMP_IOW(3, struct seccomp_notif_addfd_small)
251 
252 struct seccomp_notif_addfd_big {
253 	union {
254 		struct seccomp_notif_addfd addfd;
255 		char buf[sizeof(struct seccomp_notif_addfd) + 8];
256 	};
257 };
258 #define SECCOMP_IOCTL_NOTIF_ADDFD_BIG	\
259 	SECCOMP_IOWR(3, struct seccomp_notif_addfd_big)
260 
261 #ifndef PTRACE_EVENTMSG_SYSCALL_ENTRY
262 #define PTRACE_EVENTMSG_SYSCALL_ENTRY	1
263 #define PTRACE_EVENTMSG_SYSCALL_EXIT	2
264 #endif
265 
266 #ifndef SECCOMP_USER_NOTIF_FLAG_CONTINUE
267 #define SECCOMP_USER_NOTIF_FLAG_CONTINUE 0x00000001
268 #endif
269 
270 #ifndef SECCOMP_FILTER_FLAG_TSYNC_ESRCH
271 #define SECCOMP_FILTER_FLAG_TSYNC_ESRCH (1UL << 4)
272 #endif
273 
274 #ifndef SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV
275 #define SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV (1UL << 5)
276 #endif
277 
278 #ifndef seccomp
279 int seccomp(unsigned int op, unsigned int flags, void *args)
280 {
281 	errno = 0;
282 	return syscall(__NR_seccomp, op, flags, args);
283 }
284 #endif
285 
286 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
287 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]))
288 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
289 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32))
290 #else
291 #error "wut? Unknown __BYTE_ORDER__?!"
292 #endif
293 
294 #define SIBLING_EXIT_UNKILLED	0xbadbeef
295 #define SIBLING_EXIT_FAILURE	0xbadface
296 #define SIBLING_EXIT_NEWPRIVS	0xbadfeed
297 
298 static int __filecmp(pid_t pid1, pid_t pid2, int fd1, int fd2)
299 {
300 #ifdef __NR_kcmp
301 	errno = 0;
302 	return syscall(__NR_kcmp, pid1, pid2, KCMP_FILE, fd1, fd2);
303 #else
304 	errno = ENOSYS;
305 	return -1;
306 #endif
307 }
308 
309 /* Have TH_LOG report actual location filecmp() is used. */
310 #define filecmp(pid1, pid2, fd1, fd2)	({		\
311 	int _ret;					\
312 							\
313 	_ret = __filecmp(pid1, pid2, fd1, fd2);		\
314 	if (_ret != 0) {				\
315 		if (_ret < 0 && errno == ENOSYS) {	\
316 			TH_LOG("kcmp() syscall missing (test is less accurate)");\
317 			_ret = 0;			\
318 		}					\
319 	}						\
320 	_ret; })
321 
322 TEST(kcmp)
323 {
324 	int ret;
325 
326 	ret = __filecmp(getpid(), getpid(), 1, 1);
327 	EXPECT_EQ(ret, 0);
328 	if (ret != 0 && errno == ENOSYS)
329 		SKIP(return, "Kernel does not support kcmp() (missing CONFIG_KCMP?)");
330 }
331 
332 TEST(mode_strict_support)
333 {
334 	long ret;
335 
336 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
337 	ASSERT_EQ(0, ret) {
338 		TH_LOG("Kernel does not support CONFIG_SECCOMP");
339 	}
340 	syscall(__NR_exit, 0);
341 }
342 
343 TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL)
344 {
345 	long ret;
346 
347 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
348 	ASSERT_EQ(0, ret) {
349 		TH_LOG("Kernel does not support CONFIG_SECCOMP");
350 	}
351 	syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
352 		NULL, NULL, NULL);
353 	EXPECT_FALSE(true) {
354 		TH_LOG("Unreachable!");
355 	}
356 }
357 
358 /* Note! This doesn't test no new privs behavior */
359 TEST(no_new_privs_support)
360 {
361 	long ret;
362 
363 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
364 	EXPECT_EQ(0, ret) {
365 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
366 	}
367 }
368 
369 /* Tests kernel support by checking for a copy_from_user() fault on NULL. */
370 TEST(mode_filter_support)
371 {
372 	long ret;
373 
374 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
375 	ASSERT_EQ(0, ret) {
376 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
377 	}
378 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL);
379 	EXPECT_EQ(-1, ret);
380 	EXPECT_EQ(EFAULT, errno) {
381 		TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!");
382 	}
383 }
384 
385 TEST(mode_filter_without_nnp)
386 {
387 	struct sock_filter filter[] = {
388 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
389 	};
390 	struct sock_fprog prog = {
391 		.len = (unsigned short)ARRAY_SIZE(filter),
392 		.filter = filter,
393 	};
394 	long ret;
395 
396 	ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0);
397 	ASSERT_LE(0, ret) {
398 		TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS");
399 	}
400 	errno = 0;
401 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
402 	/* Succeeds with CAP_SYS_ADMIN, fails without */
403 	/* TODO(wad) check caps not euid */
404 	if (geteuid()) {
405 		EXPECT_EQ(-1, ret);
406 		EXPECT_EQ(EACCES, errno);
407 	} else {
408 		EXPECT_EQ(0, ret);
409 	}
410 }
411 
412 #define MAX_INSNS_PER_PATH 32768
413 
414 TEST(filter_size_limits)
415 {
416 	int i;
417 	int count = BPF_MAXINSNS + 1;
418 	struct sock_filter allow[] = {
419 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
420 	};
421 	struct sock_filter *filter;
422 	struct sock_fprog prog = { };
423 	long ret;
424 
425 	filter = calloc(count, sizeof(*filter));
426 	ASSERT_NE(NULL, filter);
427 
428 	for (i = 0; i < count; i++)
429 		filter[i] = allow[0];
430 
431 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
432 	ASSERT_EQ(0, ret);
433 
434 	prog.filter = filter;
435 	prog.len = count;
436 
437 	/* Too many filter instructions in a single filter. */
438 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
439 	ASSERT_NE(0, ret) {
440 		TH_LOG("Installing %d insn filter was allowed", prog.len);
441 	}
442 
443 	/* One less is okay, though. */
444 	prog.len -= 1;
445 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
446 	ASSERT_EQ(0, ret) {
447 		TH_LOG("Installing %d insn filter wasn't allowed", prog.len);
448 	}
449 }
450 
451 TEST(filter_chain_limits)
452 {
453 	int i;
454 	int count = BPF_MAXINSNS;
455 	struct sock_filter allow[] = {
456 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
457 	};
458 	struct sock_filter *filter;
459 	struct sock_fprog prog = { };
460 	long ret;
461 
462 	filter = calloc(count, sizeof(*filter));
463 	ASSERT_NE(NULL, filter);
464 
465 	for (i = 0; i < count; i++)
466 		filter[i] = allow[0];
467 
468 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
469 	ASSERT_EQ(0, ret);
470 
471 	prog.filter = filter;
472 	prog.len = 1;
473 
474 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
475 	ASSERT_EQ(0, ret);
476 
477 	prog.len = count;
478 
479 	/* Too many total filter instructions. */
480 	for (i = 0; i < MAX_INSNS_PER_PATH; i++) {
481 		ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
482 		if (ret != 0)
483 			break;
484 	}
485 	ASSERT_NE(0, ret) {
486 		TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)",
487 		       i, count, i * (count + 4));
488 	}
489 }
490 
491 TEST(mode_filter_cannot_move_to_strict)
492 {
493 	struct sock_filter filter[] = {
494 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
495 	};
496 	struct sock_fprog prog = {
497 		.len = (unsigned short)ARRAY_SIZE(filter),
498 		.filter = filter,
499 	};
500 	long ret;
501 
502 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
503 	ASSERT_EQ(0, ret);
504 
505 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
506 	ASSERT_EQ(0, ret);
507 
508 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0);
509 	EXPECT_EQ(-1, ret);
510 	EXPECT_EQ(EINVAL, errno);
511 }
512 
513 
514 TEST(mode_filter_get_seccomp)
515 {
516 	struct sock_filter filter[] = {
517 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
518 	};
519 	struct sock_fprog prog = {
520 		.len = (unsigned short)ARRAY_SIZE(filter),
521 		.filter = filter,
522 	};
523 	long ret;
524 
525 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
526 	ASSERT_EQ(0, ret);
527 
528 	ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
529 	EXPECT_EQ(0, ret);
530 
531 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
532 	ASSERT_EQ(0, ret);
533 
534 	ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
535 	EXPECT_EQ(2, ret);
536 }
537 
538 
539 TEST(ALLOW_all)
540 {
541 	struct sock_filter filter[] = {
542 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
543 	};
544 	struct sock_fprog prog = {
545 		.len = (unsigned short)ARRAY_SIZE(filter),
546 		.filter = filter,
547 	};
548 	long ret;
549 
550 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
551 	ASSERT_EQ(0, ret);
552 
553 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
554 	ASSERT_EQ(0, ret);
555 }
556 
557 TEST(empty_prog)
558 {
559 	struct sock_filter filter[] = {
560 	};
561 	struct sock_fprog prog = {
562 		.len = (unsigned short)ARRAY_SIZE(filter),
563 		.filter = filter,
564 	};
565 	long ret;
566 
567 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
568 	ASSERT_EQ(0, ret);
569 
570 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
571 	EXPECT_EQ(-1, ret);
572 	EXPECT_EQ(EINVAL, errno);
573 }
574 
575 TEST(log_all)
576 {
577 	struct sock_filter filter[] = {
578 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG),
579 	};
580 	struct sock_fprog prog = {
581 		.len = (unsigned short)ARRAY_SIZE(filter),
582 		.filter = filter,
583 	};
584 	long ret;
585 	pid_t parent = getppid();
586 
587 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
588 	ASSERT_EQ(0, ret);
589 
590 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
591 	ASSERT_EQ(0, ret);
592 
593 	/* getppid() should succeed and be logged (no check for logging) */
594 	EXPECT_EQ(parent, syscall(__NR_getppid));
595 }
596 
597 TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS)
598 {
599 	struct sock_filter filter[] = {
600 		BPF_STMT(BPF_RET|BPF_K, 0x10000000U),
601 	};
602 	struct sock_fprog prog = {
603 		.len = (unsigned short)ARRAY_SIZE(filter),
604 		.filter = filter,
605 	};
606 	long ret;
607 
608 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
609 	ASSERT_EQ(0, ret);
610 
611 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
612 	ASSERT_EQ(0, ret);
613 	EXPECT_EQ(0, syscall(__NR_getpid)) {
614 		TH_LOG("getpid() shouldn't ever return");
615 	}
616 }
617 
618 /* return code >= 0x80000000 is unused. */
619 TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS)
620 {
621 	struct sock_filter filter[] = {
622 		BPF_STMT(BPF_RET|BPF_K, 0x90000000U),
623 	};
624 	struct sock_fprog prog = {
625 		.len = (unsigned short)ARRAY_SIZE(filter),
626 		.filter = filter,
627 	};
628 	long ret;
629 
630 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
631 	ASSERT_EQ(0, ret);
632 
633 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
634 	ASSERT_EQ(0, ret);
635 	EXPECT_EQ(0, syscall(__NR_getpid)) {
636 		TH_LOG("getpid() shouldn't ever return");
637 	}
638 }
639 
640 TEST_SIGNAL(KILL_all, SIGSYS)
641 {
642 	struct sock_filter filter[] = {
643 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
644 	};
645 	struct sock_fprog prog = {
646 		.len = (unsigned short)ARRAY_SIZE(filter),
647 		.filter = filter,
648 	};
649 	long ret;
650 
651 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
652 	ASSERT_EQ(0, ret);
653 
654 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
655 	ASSERT_EQ(0, ret);
656 }
657 
658 TEST_SIGNAL(KILL_one, SIGSYS)
659 {
660 	struct sock_filter filter[] = {
661 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
662 			offsetof(struct seccomp_data, nr)),
663 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
664 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
665 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
666 	};
667 	struct sock_fprog prog = {
668 		.len = (unsigned short)ARRAY_SIZE(filter),
669 		.filter = filter,
670 	};
671 	long ret;
672 	pid_t parent = getppid();
673 
674 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
675 	ASSERT_EQ(0, ret);
676 
677 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
678 	ASSERT_EQ(0, ret);
679 
680 	EXPECT_EQ(parent, syscall(__NR_getppid));
681 	/* getpid() should never return. */
682 	EXPECT_EQ(0, syscall(__NR_getpid));
683 }
684 
685 TEST_SIGNAL(KILL_one_arg_one, SIGSYS)
686 {
687 	void *fatal_address;
688 	struct sock_filter filter[] = {
689 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
690 			offsetof(struct seccomp_data, nr)),
691 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0),
692 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
693 		/* Only both with lower 32-bit for now. */
694 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)),
695 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K,
696 			(unsigned long)&fatal_address, 0, 1),
697 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
698 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
699 	};
700 	struct sock_fprog prog = {
701 		.len = (unsigned short)ARRAY_SIZE(filter),
702 		.filter = filter,
703 	};
704 	long ret;
705 	pid_t parent = getppid();
706 	struct tms timebuf;
707 	clock_t clock = times(&timebuf);
708 
709 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
710 	ASSERT_EQ(0, ret);
711 
712 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
713 	ASSERT_EQ(0, ret);
714 
715 	EXPECT_EQ(parent, syscall(__NR_getppid));
716 	EXPECT_LE(clock, syscall(__NR_times, &timebuf));
717 	/* times() should never return. */
718 	EXPECT_EQ(0, syscall(__NR_times, &fatal_address));
719 }
720 
721 TEST_SIGNAL(KILL_one_arg_six, SIGSYS)
722 {
723 #ifndef __NR_mmap2
724 	int sysno = __NR_mmap;
725 #else
726 	int sysno = __NR_mmap2;
727 #endif
728 	struct sock_filter filter[] = {
729 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
730 			offsetof(struct seccomp_data, nr)),
731 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0),
732 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
733 		/* Only both with lower 32-bit for now. */
734 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)),
735 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1),
736 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
737 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
738 	};
739 	struct sock_fprog prog = {
740 		.len = (unsigned short)ARRAY_SIZE(filter),
741 		.filter = filter,
742 	};
743 	long ret;
744 	pid_t parent = getppid();
745 	int fd;
746 	void *map1, *map2;
747 	int page_size = sysconf(_SC_PAGESIZE);
748 
749 	ASSERT_LT(0, page_size);
750 
751 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
752 	ASSERT_EQ(0, ret);
753 
754 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
755 	ASSERT_EQ(0, ret);
756 
757 	fd = open("/dev/zero", O_RDONLY);
758 	ASSERT_NE(-1, fd);
759 
760 	EXPECT_EQ(parent, syscall(__NR_getppid));
761 	map1 = (void *)syscall(sysno,
762 		NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size);
763 	EXPECT_NE(MAP_FAILED, map1);
764 	/* mmap2() should never return. */
765 	map2 = (void *)syscall(sysno,
766 		 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE);
767 	EXPECT_EQ(MAP_FAILED, map2);
768 
769 	/* The test failed, so clean up the resources. */
770 	munmap(map1, page_size);
771 	munmap(map2, page_size);
772 	close(fd);
773 }
774 
775 /* This is a thread task to die via seccomp filter violation. */
776 void *kill_thread(void *data)
777 {
778 	bool die = (bool)data;
779 
780 	if (die) {
781 		prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
782 		return (void *)SIBLING_EXIT_FAILURE;
783 	}
784 
785 	return (void *)SIBLING_EXIT_UNKILLED;
786 }
787 
788 enum kill_t {
789 	KILL_THREAD,
790 	KILL_PROCESS,
791 	RET_UNKNOWN
792 };
793 
794 /* Prepare a thread that will kill itself or both of us. */
795 void kill_thread_or_group(struct __test_metadata *_metadata,
796 			  enum kill_t kill_how)
797 {
798 	pthread_t thread;
799 	void *status;
800 	/* Kill only when calling __NR_prctl. */
801 	struct sock_filter filter_thread[] = {
802 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
803 			offsetof(struct seccomp_data, nr)),
804 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
805 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD),
806 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
807 	};
808 	struct sock_fprog prog_thread = {
809 		.len = (unsigned short)ARRAY_SIZE(filter_thread),
810 		.filter = filter_thread,
811 	};
812 	int kill = kill_how == KILL_PROCESS ? SECCOMP_RET_KILL_PROCESS : 0xAAAAAAAA;
813 	struct sock_filter filter_process[] = {
814 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
815 			offsetof(struct seccomp_data, nr)),
816 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
817 		BPF_STMT(BPF_RET|BPF_K, kill),
818 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
819 	};
820 	struct sock_fprog prog_process = {
821 		.len = (unsigned short)ARRAY_SIZE(filter_process),
822 		.filter = filter_process,
823 	};
824 
825 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
826 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
827 	}
828 
829 	ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0,
830 			     kill_how == KILL_THREAD ? &prog_thread
831 						     : &prog_process));
832 
833 	/*
834 	 * Add the KILL_THREAD rule again to make sure that the KILL_PROCESS
835 	 * flag cannot be downgraded by a new filter.
836 	 */
837 	if (kill_how == KILL_PROCESS)
838 		ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog_thread));
839 
840 	/* Start a thread that will exit immediately. */
841 	ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)false));
842 	ASSERT_EQ(0, pthread_join(thread, &status));
843 	ASSERT_EQ(SIBLING_EXIT_UNKILLED, (unsigned long)status);
844 
845 	/* Start a thread that will die immediately. */
846 	ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)true));
847 	ASSERT_EQ(0, pthread_join(thread, &status));
848 	ASSERT_NE(SIBLING_EXIT_FAILURE, (unsigned long)status);
849 
850 	/*
851 	 * If we get here, only the spawned thread died. Let the parent know
852 	 * the whole process didn't die (i.e. this thread, the spawner,
853 	 * stayed running).
854 	 */
855 	exit(42);
856 }
857 
858 TEST(KILL_thread)
859 {
860 	int status;
861 	pid_t child_pid;
862 
863 	child_pid = fork();
864 	ASSERT_LE(0, child_pid);
865 	if (child_pid == 0) {
866 		kill_thread_or_group(_metadata, KILL_THREAD);
867 		_exit(38);
868 	}
869 
870 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
871 
872 	/* If only the thread was killed, we'll see exit 42. */
873 	ASSERT_TRUE(WIFEXITED(status));
874 	ASSERT_EQ(42, WEXITSTATUS(status));
875 }
876 
877 TEST(KILL_process)
878 {
879 	int status;
880 	pid_t child_pid;
881 
882 	child_pid = fork();
883 	ASSERT_LE(0, child_pid);
884 	if (child_pid == 0) {
885 		kill_thread_or_group(_metadata, KILL_PROCESS);
886 		_exit(38);
887 	}
888 
889 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
890 
891 	/* If the entire process was killed, we'll see SIGSYS. */
892 	ASSERT_TRUE(WIFSIGNALED(status));
893 	ASSERT_EQ(SIGSYS, WTERMSIG(status));
894 }
895 
896 TEST(KILL_unknown)
897 {
898 	int status;
899 	pid_t child_pid;
900 
901 	child_pid = fork();
902 	ASSERT_LE(0, child_pid);
903 	if (child_pid == 0) {
904 		kill_thread_or_group(_metadata, RET_UNKNOWN);
905 		_exit(38);
906 	}
907 
908 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
909 
910 	/* If the entire process was killed, we'll see SIGSYS. */
911 	EXPECT_TRUE(WIFSIGNALED(status)) {
912 		TH_LOG("Unknown SECCOMP_RET is only killing the thread?");
913 	}
914 	ASSERT_EQ(SIGSYS, WTERMSIG(status));
915 }
916 
917 /* TODO(wad) add 64-bit versus 32-bit arg tests. */
918 TEST(arg_out_of_range)
919 {
920 	struct sock_filter filter[] = {
921 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)),
922 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
923 	};
924 	struct sock_fprog prog = {
925 		.len = (unsigned short)ARRAY_SIZE(filter),
926 		.filter = filter,
927 	};
928 	long ret;
929 
930 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
931 	ASSERT_EQ(0, ret);
932 
933 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
934 	EXPECT_EQ(-1, ret);
935 	EXPECT_EQ(EINVAL, errno);
936 }
937 
938 #define ERRNO_FILTER(name, errno)					\
939 	struct sock_filter _read_filter_##name[] = {			\
940 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,				\
941 			offsetof(struct seccomp_data, nr)),		\
942 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),	\
943 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | errno),	\
944 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),		\
945 	};								\
946 	struct sock_fprog prog_##name = {				\
947 		.len = (unsigned short)ARRAY_SIZE(_read_filter_##name),	\
948 		.filter = _read_filter_##name,				\
949 	}
950 
951 /* Make sure basic errno values are correctly passed through a filter. */
952 TEST(ERRNO_valid)
953 {
954 	ERRNO_FILTER(valid, E2BIG);
955 	long ret;
956 	pid_t parent = getppid();
957 
958 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
959 	ASSERT_EQ(0, ret);
960 
961 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_valid);
962 	ASSERT_EQ(0, ret);
963 
964 	EXPECT_EQ(parent, syscall(__NR_getppid));
965 	EXPECT_EQ(-1, read(-1, NULL, 0));
966 	EXPECT_EQ(E2BIG, errno);
967 }
968 
969 /* Make sure an errno of zero is correctly handled by the arch code. */
970 TEST(ERRNO_zero)
971 {
972 	ERRNO_FILTER(zero, 0);
973 	long ret;
974 	pid_t parent = getppid();
975 
976 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
977 	ASSERT_EQ(0, ret);
978 
979 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_zero);
980 	ASSERT_EQ(0, ret);
981 
982 	EXPECT_EQ(parent, syscall(__NR_getppid));
983 	/* "errno" of 0 is ok. */
984 	EXPECT_EQ(0, read(-1, NULL, 0));
985 }
986 
987 /*
988  * The SECCOMP_RET_DATA mask is 16 bits wide, but errno is smaller.
989  * This tests that the errno value gets capped correctly, fixed by
990  * 580c57f10768 ("seccomp: cap SECCOMP_RET_ERRNO data to MAX_ERRNO").
991  */
992 TEST(ERRNO_capped)
993 {
994 	ERRNO_FILTER(capped, 4096);
995 	long ret;
996 	pid_t parent = getppid();
997 
998 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
999 	ASSERT_EQ(0, ret);
1000 
1001 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_capped);
1002 	ASSERT_EQ(0, ret);
1003 
1004 	EXPECT_EQ(parent, syscall(__NR_getppid));
1005 	EXPECT_EQ(-1, read(-1, NULL, 0));
1006 	EXPECT_EQ(4095, errno);
1007 }
1008 
1009 /*
1010  * Filters are processed in reverse order: last applied is executed first.
1011  * Since only the SECCOMP_RET_ACTION mask is tested for return values, the
1012  * SECCOMP_RET_DATA mask results will follow the most recently applied
1013  * matching filter return (and not the lowest or highest value).
1014  */
1015 TEST(ERRNO_order)
1016 {
1017 	ERRNO_FILTER(first,  11);
1018 	ERRNO_FILTER(second, 13);
1019 	ERRNO_FILTER(third,  12);
1020 	long ret;
1021 	pid_t parent = getppid();
1022 
1023 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1024 	ASSERT_EQ(0, ret);
1025 
1026 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_first);
1027 	ASSERT_EQ(0, ret);
1028 
1029 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_second);
1030 	ASSERT_EQ(0, ret);
1031 
1032 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_third);
1033 	ASSERT_EQ(0, ret);
1034 
1035 	EXPECT_EQ(parent, syscall(__NR_getppid));
1036 	EXPECT_EQ(-1, read(-1, NULL, 0));
1037 	EXPECT_EQ(12, errno);
1038 }
1039 
1040 FIXTURE(TRAP) {
1041 	struct sock_fprog prog;
1042 };
1043 
1044 FIXTURE_SETUP(TRAP)
1045 {
1046 	struct sock_filter filter[] = {
1047 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1048 			offsetof(struct seccomp_data, nr)),
1049 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
1050 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
1051 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1052 	};
1053 
1054 	memset(&self->prog, 0, sizeof(self->prog));
1055 	self->prog.filter = malloc(sizeof(filter));
1056 	ASSERT_NE(NULL, self->prog.filter);
1057 	memcpy(self->prog.filter, filter, sizeof(filter));
1058 	self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1059 }
1060 
1061 FIXTURE_TEARDOWN(TRAP)
1062 {
1063 	if (self->prog.filter)
1064 		free(self->prog.filter);
1065 }
1066 
1067 TEST_F_SIGNAL(TRAP, dfl, SIGSYS)
1068 {
1069 	long ret;
1070 
1071 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1072 	ASSERT_EQ(0, ret);
1073 
1074 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
1075 	ASSERT_EQ(0, ret);
1076 	syscall(__NR_getpid);
1077 }
1078 
1079 /* Ensure that SIGSYS overrides SIG_IGN */
1080 TEST_F_SIGNAL(TRAP, ign, SIGSYS)
1081 {
1082 	long ret;
1083 
1084 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1085 	ASSERT_EQ(0, ret);
1086 
1087 	signal(SIGSYS, SIG_IGN);
1088 
1089 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
1090 	ASSERT_EQ(0, ret);
1091 	syscall(__NR_getpid);
1092 }
1093 
1094 static siginfo_t TRAP_info;
1095 static volatile int TRAP_nr;
1096 static void TRAP_action(int nr, siginfo_t *info, void *void_context)
1097 {
1098 	memcpy(&TRAP_info, info, sizeof(TRAP_info));
1099 	TRAP_nr = nr;
1100 }
1101 
1102 TEST_F(TRAP, handler)
1103 {
1104 	int ret, test;
1105 	struct sigaction act;
1106 	sigset_t mask;
1107 
1108 	memset(&act, 0, sizeof(act));
1109 	sigemptyset(&mask);
1110 	sigaddset(&mask, SIGSYS);
1111 
1112 	act.sa_sigaction = &TRAP_action;
1113 	act.sa_flags = SA_SIGINFO;
1114 	ret = sigaction(SIGSYS, &act, NULL);
1115 	ASSERT_EQ(0, ret) {
1116 		TH_LOG("sigaction failed");
1117 	}
1118 	ret = sigprocmask(SIG_UNBLOCK, &mask, NULL);
1119 	ASSERT_EQ(0, ret) {
1120 		TH_LOG("sigprocmask failed");
1121 	}
1122 
1123 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1124 	ASSERT_EQ(0, ret);
1125 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
1126 	ASSERT_EQ(0, ret);
1127 	TRAP_nr = 0;
1128 	memset(&TRAP_info, 0, sizeof(TRAP_info));
1129 	/* Expect the registers to be rolled back. (nr = error) may vary
1130 	 * based on arch. */
1131 	ret = syscall(__NR_getpid);
1132 	/* Silence gcc warning about volatile. */
1133 	test = TRAP_nr;
1134 	EXPECT_EQ(SIGSYS, test);
1135 	struct local_sigsys {
1136 		void *_call_addr;	/* calling user insn */
1137 		int _syscall;		/* triggering system call number */
1138 		unsigned int _arch;	/* AUDIT_ARCH_* of syscall */
1139 	} *sigsys = (struct local_sigsys *)
1140 #ifdef si_syscall
1141 		&(TRAP_info.si_call_addr);
1142 #else
1143 		&TRAP_info.si_pid;
1144 #endif
1145 	EXPECT_EQ(__NR_getpid, sigsys->_syscall);
1146 	/* Make sure arch is non-zero. */
1147 	EXPECT_NE(0, sigsys->_arch);
1148 	EXPECT_NE(0, (unsigned long)sigsys->_call_addr);
1149 }
1150 
1151 FIXTURE(precedence) {
1152 	struct sock_fprog allow;
1153 	struct sock_fprog log;
1154 	struct sock_fprog trace;
1155 	struct sock_fprog error;
1156 	struct sock_fprog trap;
1157 	struct sock_fprog kill;
1158 };
1159 
1160 FIXTURE_SETUP(precedence)
1161 {
1162 	struct sock_filter allow_insns[] = {
1163 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1164 	};
1165 	struct sock_filter log_insns[] = {
1166 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1167 			offsetof(struct seccomp_data, nr)),
1168 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1169 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1170 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG),
1171 	};
1172 	struct sock_filter trace_insns[] = {
1173 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1174 			offsetof(struct seccomp_data, nr)),
1175 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1176 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1177 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE),
1178 	};
1179 	struct sock_filter error_insns[] = {
1180 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1181 			offsetof(struct seccomp_data, nr)),
1182 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1183 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1184 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO),
1185 	};
1186 	struct sock_filter trap_insns[] = {
1187 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1188 			offsetof(struct seccomp_data, nr)),
1189 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1190 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1191 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
1192 	};
1193 	struct sock_filter kill_insns[] = {
1194 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1195 			offsetof(struct seccomp_data, nr)),
1196 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1197 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1198 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1199 	};
1200 
1201 	memset(self, 0, sizeof(*self));
1202 #define FILTER_ALLOC(_x) \
1203 	self->_x.filter = malloc(sizeof(_x##_insns)); \
1204 	ASSERT_NE(NULL, self->_x.filter); \
1205 	memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \
1206 	self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns)
1207 	FILTER_ALLOC(allow);
1208 	FILTER_ALLOC(log);
1209 	FILTER_ALLOC(trace);
1210 	FILTER_ALLOC(error);
1211 	FILTER_ALLOC(trap);
1212 	FILTER_ALLOC(kill);
1213 }
1214 
1215 FIXTURE_TEARDOWN(precedence)
1216 {
1217 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter)
1218 	FILTER_FREE(allow);
1219 	FILTER_FREE(log);
1220 	FILTER_FREE(trace);
1221 	FILTER_FREE(error);
1222 	FILTER_FREE(trap);
1223 	FILTER_FREE(kill);
1224 }
1225 
1226 TEST_F(precedence, allow_ok)
1227 {
1228 	pid_t parent, res = 0;
1229 	long ret;
1230 
1231 	parent = getppid();
1232 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1233 	ASSERT_EQ(0, ret);
1234 
1235 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1236 	ASSERT_EQ(0, ret);
1237 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1238 	ASSERT_EQ(0, ret);
1239 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1240 	ASSERT_EQ(0, ret);
1241 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1242 	ASSERT_EQ(0, ret);
1243 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1244 	ASSERT_EQ(0, ret);
1245 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1246 	ASSERT_EQ(0, ret);
1247 	/* Should work just fine. */
1248 	res = syscall(__NR_getppid);
1249 	EXPECT_EQ(parent, res);
1250 }
1251 
1252 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS)
1253 {
1254 	pid_t parent, res = 0;
1255 	long ret;
1256 
1257 	parent = getppid();
1258 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1259 	ASSERT_EQ(0, ret);
1260 
1261 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1262 	ASSERT_EQ(0, ret);
1263 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1264 	ASSERT_EQ(0, ret);
1265 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1266 	ASSERT_EQ(0, ret);
1267 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1268 	ASSERT_EQ(0, ret);
1269 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1270 	ASSERT_EQ(0, ret);
1271 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1272 	ASSERT_EQ(0, ret);
1273 	/* Should work just fine. */
1274 	res = syscall(__NR_getppid);
1275 	EXPECT_EQ(parent, res);
1276 	/* getpid() should never return. */
1277 	res = syscall(__NR_getpid);
1278 	EXPECT_EQ(0, res);
1279 }
1280 
1281 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS)
1282 {
1283 	pid_t parent;
1284 	long ret;
1285 
1286 	parent = getppid();
1287 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1288 	ASSERT_EQ(0, ret);
1289 
1290 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1291 	ASSERT_EQ(0, ret);
1292 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1293 	ASSERT_EQ(0, ret);
1294 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1295 	ASSERT_EQ(0, ret);
1296 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1297 	ASSERT_EQ(0, ret);
1298 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1299 	ASSERT_EQ(0, ret);
1300 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1301 	ASSERT_EQ(0, ret);
1302 	/* Should work just fine. */
1303 	EXPECT_EQ(parent, syscall(__NR_getppid));
1304 	/* getpid() should never return. */
1305 	EXPECT_EQ(0, syscall(__NR_getpid));
1306 }
1307 
1308 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS)
1309 {
1310 	pid_t parent;
1311 	long ret;
1312 
1313 	parent = getppid();
1314 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1315 	ASSERT_EQ(0, ret);
1316 
1317 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1318 	ASSERT_EQ(0, ret);
1319 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1320 	ASSERT_EQ(0, ret);
1321 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1322 	ASSERT_EQ(0, ret);
1323 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1324 	ASSERT_EQ(0, ret);
1325 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1326 	ASSERT_EQ(0, ret);
1327 	/* Should work just fine. */
1328 	EXPECT_EQ(parent, syscall(__NR_getppid));
1329 	/* getpid() should never return. */
1330 	EXPECT_EQ(0, syscall(__NR_getpid));
1331 }
1332 
1333 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS)
1334 {
1335 	pid_t parent;
1336 	long ret;
1337 
1338 	parent = getppid();
1339 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1340 	ASSERT_EQ(0, ret);
1341 
1342 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1343 	ASSERT_EQ(0, ret);
1344 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1345 	ASSERT_EQ(0, ret);
1346 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1347 	ASSERT_EQ(0, ret);
1348 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1349 	ASSERT_EQ(0, ret);
1350 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1351 	ASSERT_EQ(0, ret);
1352 	/* Should work just fine. */
1353 	EXPECT_EQ(parent, syscall(__NR_getppid));
1354 	/* getpid() should never return. */
1355 	EXPECT_EQ(0, syscall(__NR_getpid));
1356 }
1357 
1358 TEST_F(precedence, errno_is_third)
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 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1374 	ASSERT_EQ(0, ret);
1375 	/* Should work just fine. */
1376 	EXPECT_EQ(parent, syscall(__NR_getppid));
1377 	EXPECT_EQ(0, syscall(__NR_getpid));
1378 }
1379 
1380 TEST_F(precedence, errno_is_third_in_any_order)
1381 {
1382 	pid_t parent;
1383 	long ret;
1384 
1385 	parent = getppid();
1386 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1387 	ASSERT_EQ(0, ret);
1388 
1389 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1390 	ASSERT_EQ(0, ret);
1391 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1392 	ASSERT_EQ(0, ret);
1393 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1394 	ASSERT_EQ(0, ret);
1395 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1396 	ASSERT_EQ(0, ret);
1397 	/* Should work just fine. */
1398 	EXPECT_EQ(parent, syscall(__NR_getppid));
1399 	EXPECT_EQ(0, syscall(__NR_getpid));
1400 }
1401 
1402 TEST_F(precedence, trace_is_fourth)
1403 {
1404 	pid_t parent;
1405 	long ret;
1406 
1407 	parent = getppid();
1408 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1409 	ASSERT_EQ(0, ret);
1410 
1411 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1412 	ASSERT_EQ(0, ret);
1413 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1414 	ASSERT_EQ(0, ret);
1415 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1416 	ASSERT_EQ(0, ret);
1417 	/* Should work just fine. */
1418 	EXPECT_EQ(parent, syscall(__NR_getppid));
1419 	/* No ptracer */
1420 	EXPECT_EQ(-1, syscall(__NR_getpid));
1421 }
1422 
1423 TEST_F(precedence, trace_is_fourth_in_any_order)
1424 {
1425 	pid_t parent;
1426 	long ret;
1427 
1428 	parent = getppid();
1429 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1430 	ASSERT_EQ(0, ret);
1431 
1432 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1433 	ASSERT_EQ(0, ret);
1434 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1435 	ASSERT_EQ(0, ret);
1436 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1437 	ASSERT_EQ(0, ret);
1438 	/* Should work just fine. */
1439 	EXPECT_EQ(parent, syscall(__NR_getppid));
1440 	/* No ptracer */
1441 	EXPECT_EQ(-1, syscall(__NR_getpid));
1442 }
1443 
1444 TEST_F(precedence, log_is_fifth)
1445 {
1446 	pid_t mypid, parent;
1447 	long ret;
1448 
1449 	mypid = getpid();
1450 	parent = getppid();
1451 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1452 	ASSERT_EQ(0, ret);
1453 
1454 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1455 	ASSERT_EQ(0, ret);
1456 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1457 	ASSERT_EQ(0, ret);
1458 	/* Should work just fine. */
1459 	EXPECT_EQ(parent, syscall(__NR_getppid));
1460 	/* Should also work just fine */
1461 	EXPECT_EQ(mypid, syscall(__NR_getpid));
1462 }
1463 
1464 TEST_F(precedence, log_is_fifth_in_any_order)
1465 {
1466 	pid_t mypid, parent;
1467 	long ret;
1468 
1469 	mypid = getpid();
1470 	parent = getppid();
1471 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1472 	ASSERT_EQ(0, ret);
1473 
1474 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1475 	ASSERT_EQ(0, ret);
1476 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1477 	ASSERT_EQ(0, ret);
1478 	/* Should work just fine. */
1479 	EXPECT_EQ(parent, syscall(__NR_getppid));
1480 	/* Should also work just fine */
1481 	EXPECT_EQ(mypid, syscall(__NR_getpid));
1482 }
1483 
1484 #ifndef PTRACE_O_TRACESECCOMP
1485 #define PTRACE_O_TRACESECCOMP	0x00000080
1486 #endif
1487 
1488 /* Catch the Ubuntu 12.04 value error. */
1489 #if PTRACE_EVENT_SECCOMP != 7
1490 #undef PTRACE_EVENT_SECCOMP
1491 #endif
1492 
1493 #ifndef PTRACE_EVENT_SECCOMP
1494 #define PTRACE_EVENT_SECCOMP 7
1495 #endif
1496 
1497 #define PTRACE_EVENT_MASK(status) ((status) >> 16)
1498 bool tracer_running;
1499 void tracer_stop(int sig)
1500 {
1501 	tracer_running = false;
1502 }
1503 
1504 typedef void tracer_func_t(struct __test_metadata *_metadata,
1505 			   pid_t tracee, int status, void *args);
1506 
1507 void start_tracer(struct __test_metadata *_metadata, int fd, pid_t tracee,
1508 	    tracer_func_t tracer_func, void *args, bool ptrace_syscall)
1509 {
1510 	int ret = -1;
1511 	struct sigaction action = {
1512 		.sa_handler = tracer_stop,
1513 	};
1514 
1515 	/* Allow external shutdown. */
1516 	tracer_running = true;
1517 	ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL));
1518 
1519 	errno = 0;
1520 	while (ret == -1 && errno != EINVAL)
1521 		ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0);
1522 	ASSERT_EQ(0, ret) {
1523 		kill(tracee, SIGKILL);
1524 	}
1525 	/* Wait for attach stop */
1526 	wait(NULL);
1527 
1528 	ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, ptrace_syscall ?
1529 						      PTRACE_O_TRACESYSGOOD :
1530 						      PTRACE_O_TRACESECCOMP);
1531 	ASSERT_EQ(0, ret) {
1532 		TH_LOG("Failed to set PTRACE_O_TRACESECCOMP");
1533 		kill(tracee, SIGKILL);
1534 	}
1535 	ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1536 		     tracee, NULL, 0);
1537 	ASSERT_EQ(0, ret);
1538 
1539 	/* Unblock the tracee */
1540 	ASSERT_EQ(1, write(fd, "A", 1));
1541 	ASSERT_EQ(0, close(fd));
1542 
1543 	/* Run until we're shut down. Must assert to stop execution. */
1544 	while (tracer_running) {
1545 		int status;
1546 
1547 		if (wait(&status) != tracee)
1548 			continue;
1549 
1550 		if (WIFSIGNALED(status)) {
1551 			/* Child caught a fatal signal. */
1552 			return;
1553 		}
1554 		if (WIFEXITED(status)) {
1555 			/* Child exited with code. */
1556 			return;
1557 		}
1558 
1559 		/* Check if we got an expected event. */
1560 		ASSERT_EQ(WIFCONTINUED(status), false);
1561 		ASSERT_EQ(WIFSTOPPED(status), true);
1562 		ASSERT_EQ(WSTOPSIG(status) & SIGTRAP, SIGTRAP) {
1563 			TH_LOG("Unexpected WSTOPSIG: %d", WSTOPSIG(status));
1564 		}
1565 
1566 		tracer_func(_metadata, tracee, status, args);
1567 
1568 		ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1569 			     tracee, NULL, 0);
1570 		ASSERT_EQ(0, ret);
1571 	}
1572 	/* Directly report the status of our test harness results. */
1573 	syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS : EXIT_FAILURE);
1574 }
1575 
1576 /* Common tracer setup/teardown functions. */
1577 void cont_handler(int num)
1578 { }
1579 pid_t setup_trace_fixture(struct __test_metadata *_metadata,
1580 			  tracer_func_t func, void *args, bool ptrace_syscall)
1581 {
1582 	char sync;
1583 	int pipefd[2];
1584 	pid_t tracer_pid;
1585 	pid_t tracee = getpid();
1586 
1587 	/* Setup a pipe for clean synchronization. */
1588 	ASSERT_EQ(0, pipe(pipefd));
1589 
1590 	/* Fork a child which we'll promote to tracer */
1591 	tracer_pid = fork();
1592 	ASSERT_LE(0, tracer_pid);
1593 	signal(SIGALRM, cont_handler);
1594 	if (tracer_pid == 0) {
1595 		close(pipefd[0]);
1596 		start_tracer(_metadata, pipefd[1], tracee, func, args,
1597 			     ptrace_syscall);
1598 		syscall(__NR_exit, 0);
1599 	}
1600 	close(pipefd[1]);
1601 	prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0);
1602 	read(pipefd[0], &sync, 1);
1603 	close(pipefd[0]);
1604 
1605 	return tracer_pid;
1606 }
1607 
1608 void teardown_trace_fixture(struct __test_metadata *_metadata,
1609 			    pid_t tracer)
1610 {
1611 	if (tracer) {
1612 		int status;
1613 		/*
1614 		 * Extract the exit code from the other process and
1615 		 * adopt it for ourselves in case its asserts failed.
1616 		 */
1617 		ASSERT_EQ(0, kill(tracer, SIGUSR1));
1618 		ASSERT_EQ(tracer, waitpid(tracer, &status, 0));
1619 		if (WEXITSTATUS(status))
1620 			_metadata->passed = 0;
1621 	}
1622 }
1623 
1624 /* "poke" tracer arguments and function. */
1625 struct tracer_args_poke_t {
1626 	unsigned long poke_addr;
1627 };
1628 
1629 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status,
1630 		 void *args)
1631 {
1632 	int ret;
1633 	unsigned long msg;
1634 	struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args;
1635 
1636 	ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1637 	EXPECT_EQ(0, ret);
1638 	/* If this fails, don't try to recover. */
1639 	ASSERT_EQ(0x1001, msg) {
1640 		kill(tracee, SIGKILL);
1641 	}
1642 	/*
1643 	 * Poke in the message.
1644 	 * Registers are not touched to try to keep this relatively arch
1645 	 * agnostic.
1646 	 */
1647 	ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001);
1648 	EXPECT_EQ(0, ret);
1649 }
1650 
1651 FIXTURE(TRACE_poke) {
1652 	struct sock_fprog prog;
1653 	pid_t tracer;
1654 	long poked;
1655 	struct tracer_args_poke_t tracer_args;
1656 };
1657 
1658 FIXTURE_SETUP(TRACE_poke)
1659 {
1660 	struct sock_filter filter[] = {
1661 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1662 			offsetof(struct seccomp_data, nr)),
1663 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1664 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001),
1665 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1666 	};
1667 
1668 	self->poked = 0;
1669 	memset(&self->prog, 0, sizeof(self->prog));
1670 	self->prog.filter = malloc(sizeof(filter));
1671 	ASSERT_NE(NULL, self->prog.filter);
1672 	memcpy(self->prog.filter, filter, sizeof(filter));
1673 	self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1674 
1675 	/* Set up tracer args. */
1676 	self->tracer_args.poke_addr = (unsigned long)&self->poked;
1677 
1678 	/* Launch tracer. */
1679 	self->tracer = setup_trace_fixture(_metadata, tracer_poke,
1680 					   &self->tracer_args, false);
1681 }
1682 
1683 FIXTURE_TEARDOWN(TRACE_poke)
1684 {
1685 	teardown_trace_fixture(_metadata, self->tracer);
1686 	if (self->prog.filter)
1687 		free(self->prog.filter);
1688 }
1689 
1690 TEST_F(TRACE_poke, read_has_side_effects)
1691 {
1692 	ssize_t ret;
1693 
1694 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1695 	ASSERT_EQ(0, ret);
1696 
1697 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1698 	ASSERT_EQ(0, ret);
1699 
1700 	EXPECT_EQ(0, self->poked);
1701 	ret = read(-1, NULL, 0);
1702 	EXPECT_EQ(-1, ret);
1703 	EXPECT_EQ(0x1001, self->poked);
1704 }
1705 
1706 TEST_F(TRACE_poke, getpid_runs_normally)
1707 {
1708 	long ret;
1709 
1710 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1711 	ASSERT_EQ(0, ret);
1712 
1713 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1714 	ASSERT_EQ(0, ret);
1715 
1716 	EXPECT_EQ(0, self->poked);
1717 	EXPECT_NE(0, syscall(__NR_getpid));
1718 	EXPECT_EQ(0, self->poked);
1719 }
1720 
1721 #if defined(__x86_64__)
1722 # define ARCH_REGS		struct user_regs_struct
1723 # define SYSCALL_NUM(_regs)	(_regs).orig_rax
1724 # define SYSCALL_RET(_regs)	(_regs).rax
1725 #elif defined(__i386__)
1726 # define ARCH_REGS		struct user_regs_struct
1727 # define SYSCALL_NUM(_regs)	(_regs).orig_eax
1728 # define SYSCALL_RET(_regs)	(_regs).eax
1729 #elif defined(__arm__)
1730 # define ARCH_REGS		struct pt_regs
1731 # define SYSCALL_NUM(_regs)	(_regs).ARM_r7
1732 # ifndef PTRACE_SET_SYSCALL
1733 #  define PTRACE_SET_SYSCALL   23
1734 # endif
1735 # define SYSCALL_NUM_SET(_regs, _nr)	\
1736 		EXPECT_EQ(0, ptrace(PTRACE_SET_SYSCALL, tracee, NULL, _nr))
1737 # define SYSCALL_RET(_regs)	(_regs).ARM_r0
1738 #elif defined(__aarch64__)
1739 # define ARCH_REGS		struct user_pt_regs
1740 # define SYSCALL_NUM(_regs)	(_regs).regs[8]
1741 # ifndef NT_ARM_SYSTEM_CALL
1742 #  define NT_ARM_SYSTEM_CALL 0x404
1743 # endif
1744 # define SYSCALL_NUM_SET(_regs, _nr)				\
1745 	do {							\
1746 		struct iovec __v;				\
1747 		typeof(_nr) __nr = (_nr);			\
1748 		__v.iov_base = &__nr;				\
1749 		__v.iov_len = sizeof(__nr);			\
1750 		EXPECT_EQ(0, ptrace(PTRACE_SETREGSET, tracee,	\
1751 				    NT_ARM_SYSTEM_CALL, &__v));	\
1752 	} while (0)
1753 # define SYSCALL_RET(_regs)	(_regs).regs[0]
1754 #elif defined(__riscv) && __riscv_xlen == 64
1755 # define ARCH_REGS		struct user_regs_struct
1756 # define SYSCALL_NUM(_regs)	(_regs).a7
1757 # define SYSCALL_RET(_regs)	(_regs).a0
1758 #elif defined(__csky__)
1759 # define ARCH_REGS		struct pt_regs
1760 #  if defined(__CSKYABIV2__)
1761 #   define SYSCALL_NUM(_regs)	(_regs).regs[3]
1762 #  else
1763 #   define SYSCALL_NUM(_regs)	(_regs).regs[9]
1764 #  endif
1765 # define SYSCALL_RET(_regs)	(_regs).a0
1766 #elif defined(__hppa__)
1767 # define ARCH_REGS		struct user_regs_struct
1768 # define SYSCALL_NUM(_regs)	(_regs).gr[20]
1769 # define SYSCALL_RET(_regs)	(_regs).gr[28]
1770 #elif defined(__powerpc__)
1771 # define ARCH_REGS		struct pt_regs
1772 # define SYSCALL_NUM(_regs)	(_regs).gpr[0]
1773 # define SYSCALL_RET(_regs)	(_regs).gpr[3]
1774 # define SYSCALL_RET_SET(_regs, _val)				\
1775 	do {							\
1776 		typeof(_val) _result = (_val);			\
1777 		if ((_regs.trap & 0xfff0) == 0x3000) {		\
1778 			/*					\
1779 			 * scv 0 system call uses -ve result	\
1780 			 * for error, so no need to adjust.	\
1781 			 */					\
1782 			SYSCALL_RET(_regs) = _result;		\
1783 		} else {					\
1784 			/*					\
1785 			 * A syscall error is signaled by the	\
1786 			 * CR0 SO bit and the code is stored as	\
1787 			 * a positive value.			\
1788 			 */					\
1789 			if (_result < 0) {			\
1790 				SYSCALL_RET(_regs) = -_result;	\
1791 				(_regs).ccr |= 0x10000000;	\
1792 			} else {				\
1793 				SYSCALL_RET(_regs) = _result;	\
1794 				(_regs).ccr &= ~0x10000000;	\
1795 			}					\
1796 		}						\
1797 	} while (0)
1798 # define SYSCALL_RET_SET_ON_PTRACE_EXIT
1799 #elif defined(__s390__)
1800 # define ARCH_REGS		s390_regs
1801 # define SYSCALL_NUM(_regs)	(_regs).gprs[2]
1802 # define SYSCALL_RET_SET(_regs, _val)			\
1803 		TH_LOG("Can't modify syscall return on this architecture")
1804 #elif defined(__mips__)
1805 # include <asm/unistd_nr_n32.h>
1806 # include <asm/unistd_nr_n64.h>
1807 # include <asm/unistd_nr_o32.h>
1808 # define ARCH_REGS		struct pt_regs
1809 # define SYSCALL_NUM(_regs)				\
1810 	({						\
1811 		typeof((_regs).regs[2]) _nr;		\
1812 		if ((_regs).regs[2] == __NR_O32_Linux)	\
1813 			_nr = (_regs).regs[4];		\
1814 		else					\
1815 			_nr = (_regs).regs[2];		\
1816 		_nr;					\
1817 	})
1818 # define SYSCALL_NUM_SET(_regs, _nr)			\
1819 	do {						\
1820 		if ((_regs).regs[2] == __NR_O32_Linux)	\
1821 			(_regs).regs[4] = _nr;		\
1822 		else					\
1823 			(_regs).regs[2] = _nr;		\
1824 	} while (0)
1825 # define SYSCALL_RET_SET(_regs, _val)			\
1826 		TH_LOG("Can't modify syscall return on this architecture")
1827 #elif defined(__xtensa__)
1828 # define ARCH_REGS		struct user_pt_regs
1829 # define SYSCALL_NUM(_regs)	(_regs).syscall
1830 /*
1831  * On xtensa syscall return value is in the register
1832  * a2 of the current window which is not fixed.
1833  */
1834 #define SYSCALL_RET(_regs)	(_regs).a[(_regs).windowbase * 4 + 2]
1835 #elif defined(__sh__)
1836 # define ARCH_REGS		struct pt_regs
1837 # define SYSCALL_NUM(_regs)	(_regs).regs[3]
1838 # define SYSCALL_RET(_regs)	(_regs).regs[0]
1839 #else
1840 # error "Do not know how to find your architecture's registers and syscalls"
1841 #endif
1842 
1843 /*
1844  * Most architectures can change the syscall by just updating the
1845  * associated register. This is the default if not defined above.
1846  */
1847 #ifndef SYSCALL_NUM_SET
1848 # define SYSCALL_NUM_SET(_regs, _nr)		\
1849 	do {					\
1850 		SYSCALL_NUM(_regs) = (_nr);	\
1851 	} while (0)
1852 #endif
1853 /*
1854  * Most architectures can change the syscall return value by just
1855  * writing to the SYSCALL_RET register. This is the default if not
1856  * defined above. If an architecture cannot set the return value
1857  * (for example when the syscall and return value register is
1858  * shared), report it with TH_LOG() in an arch-specific definition
1859  * of SYSCALL_RET_SET() above, and leave SYSCALL_RET undefined.
1860  */
1861 #if !defined(SYSCALL_RET) && !defined(SYSCALL_RET_SET)
1862 # error "One of SYSCALL_RET or SYSCALL_RET_SET is needed for this arch"
1863 #endif
1864 #ifndef SYSCALL_RET_SET
1865 # define SYSCALL_RET_SET(_regs, _val)		\
1866 	do {					\
1867 		SYSCALL_RET(_regs) = (_val);	\
1868 	} while (0)
1869 #endif
1870 
1871 /* When the syscall return can't be changed, stub out the tests for it. */
1872 #ifndef SYSCALL_RET
1873 # define EXPECT_SYSCALL_RETURN(val, action)	EXPECT_EQ(-1, action)
1874 #else
1875 # define EXPECT_SYSCALL_RETURN(val, action)		\
1876 	do {						\
1877 		errno = 0;				\
1878 		if (val < 0) {				\
1879 			EXPECT_EQ(-1, action);		\
1880 			EXPECT_EQ(-(val), errno);	\
1881 		} else {				\
1882 			EXPECT_EQ(val, action);		\
1883 		}					\
1884 	} while (0)
1885 #endif
1886 
1887 /*
1888  * Some architectures (e.g. powerpc) can only set syscall
1889  * return values on syscall exit during ptrace.
1890  */
1891 const bool ptrace_entry_set_syscall_nr = true;
1892 const bool ptrace_entry_set_syscall_ret =
1893 #ifndef SYSCALL_RET_SET_ON_PTRACE_EXIT
1894 	true;
1895 #else
1896 	false;
1897 #endif
1898 
1899 /*
1900  * Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for
1901  * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux).
1902  */
1903 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__)
1904 # define ARCH_GETREGS(_regs)	ptrace(PTRACE_GETREGS, tracee, 0, &(_regs))
1905 # define ARCH_SETREGS(_regs)	ptrace(PTRACE_SETREGS, tracee, 0, &(_regs))
1906 #else
1907 # define ARCH_GETREGS(_regs)	({					\
1908 		struct iovec __v;					\
1909 		__v.iov_base = &(_regs);				\
1910 		__v.iov_len = sizeof(_regs);				\
1911 		ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &__v);	\
1912 	})
1913 # define ARCH_SETREGS(_regs)	({					\
1914 		struct iovec __v;					\
1915 		__v.iov_base = &(_regs);				\
1916 		__v.iov_len = sizeof(_regs);				\
1917 		ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &__v);	\
1918 	})
1919 #endif
1920 
1921 /* Architecture-specific syscall fetching routine. */
1922 int get_syscall(struct __test_metadata *_metadata, pid_t tracee)
1923 {
1924 	ARCH_REGS regs;
1925 
1926 	EXPECT_EQ(0, ARCH_GETREGS(regs)) {
1927 		return -1;
1928 	}
1929 
1930 	return SYSCALL_NUM(regs);
1931 }
1932 
1933 /* Architecture-specific syscall changing routine. */
1934 void __change_syscall(struct __test_metadata *_metadata,
1935 		    pid_t tracee, long *syscall, long *ret)
1936 {
1937 	ARCH_REGS orig, regs;
1938 
1939 	/* Do not get/set registers if we have nothing to do. */
1940 	if (!syscall && !ret)
1941 		return;
1942 
1943 	EXPECT_EQ(0, ARCH_GETREGS(regs)) {
1944 		return;
1945 	}
1946 	orig = regs;
1947 
1948 	if (syscall)
1949 		SYSCALL_NUM_SET(regs, *syscall);
1950 
1951 	if (ret)
1952 		SYSCALL_RET_SET(regs, *ret);
1953 
1954 	/* Flush any register changes made. */
1955 	if (memcmp(&orig, &regs, sizeof(orig)) != 0)
1956 		EXPECT_EQ(0, ARCH_SETREGS(regs));
1957 }
1958 
1959 /* Change only syscall number. */
1960 void change_syscall_nr(struct __test_metadata *_metadata,
1961 		       pid_t tracee, long syscall)
1962 {
1963 	__change_syscall(_metadata, tracee, &syscall, NULL);
1964 }
1965 
1966 /* Change syscall return value (and set syscall number to -1). */
1967 void change_syscall_ret(struct __test_metadata *_metadata,
1968 			pid_t tracee, long ret)
1969 {
1970 	long syscall = -1;
1971 
1972 	__change_syscall(_metadata, tracee, &syscall, &ret);
1973 }
1974 
1975 void tracer_seccomp(struct __test_metadata *_metadata, pid_t tracee,
1976 		    int status, void *args)
1977 {
1978 	int ret;
1979 	unsigned long msg;
1980 
1981 	EXPECT_EQ(PTRACE_EVENT_MASK(status), PTRACE_EVENT_SECCOMP) {
1982 		TH_LOG("Unexpected ptrace event: %d", PTRACE_EVENT_MASK(status));
1983 		return;
1984 	}
1985 
1986 	/* Make sure we got the right message. */
1987 	ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1988 	EXPECT_EQ(0, ret);
1989 
1990 	/* Validate and take action on expected syscalls. */
1991 	switch (msg) {
1992 	case 0x1002:
1993 		/* change getpid to getppid. */
1994 		EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee));
1995 		change_syscall_nr(_metadata, tracee, __NR_getppid);
1996 		break;
1997 	case 0x1003:
1998 		/* skip gettid with valid return code. */
1999 		EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee));
2000 		change_syscall_ret(_metadata, tracee, 45000);
2001 		break;
2002 	case 0x1004:
2003 		/* skip openat with error. */
2004 		EXPECT_EQ(__NR_openat, get_syscall(_metadata, tracee));
2005 		change_syscall_ret(_metadata, tracee, -ESRCH);
2006 		break;
2007 	case 0x1005:
2008 		/* do nothing (allow getppid) */
2009 		EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee));
2010 		break;
2011 	default:
2012 		EXPECT_EQ(0, msg) {
2013 			TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg);
2014 			kill(tracee, SIGKILL);
2015 		}
2016 	}
2017 
2018 }
2019 
2020 FIXTURE(TRACE_syscall) {
2021 	struct sock_fprog prog;
2022 	pid_t tracer, mytid, mypid, parent;
2023 	long syscall_nr;
2024 };
2025 
2026 void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee,
2027 		   int status, void *args)
2028 {
2029 	int ret;
2030 	unsigned long msg;
2031 	static bool entry;
2032 	long syscall_nr_val, syscall_ret_val;
2033 	long *syscall_nr = NULL, *syscall_ret = NULL;
2034 	FIXTURE_DATA(TRACE_syscall) *self = args;
2035 
2036 	EXPECT_EQ(WSTOPSIG(status) & 0x80, 0x80) {
2037 		TH_LOG("Unexpected WSTOPSIG: %d", WSTOPSIG(status));
2038 		return;
2039 	}
2040 
2041 	/*
2042 	 * The traditional way to tell PTRACE_SYSCALL entry/exit
2043 	 * is by counting.
2044 	 */
2045 	entry = !entry;
2046 
2047 	/* Make sure we got an appropriate message. */
2048 	ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
2049 	EXPECT_EQ(0, ret);
2050 	EXPECT_EQ(entry ? PTRACE_EVENTMSG_SYSCALL_ENTRY
2051 			: PTRACE_EVENTMSG_SYSCALL_EXIT, msg);
2052 
2053 	/*
2054 	 * Some architectures only support setting return values during
2055 	 * syscall exit under ptrace, and on exit the syscall number may
2056 	 * no longer be available. Therefore, save the initial sycall
2057 	 * number here, so it can be examined during both entry and exit
2058 	 * phases.
2059 	 */
2060 	if (entry)
2061 		self->syscall_nr = get_syscall(_metadata, tracee);
2062 
2063 	/*
2064 	 * Depending on the architecture's syscall setting abilities, we
2065 	 * pick which things to set during this phase (entry or exit).
2066 	 */
2067 	if (entry == ptrace_entry_set_syscall_nr)
2068 		syscall_nr = &syscall_nr_val;
2069 	if (entry == ptrace_entry_set_syscall_ret)
2070 		syscall_ret = &syscall_ret_val;
2071 
2072 	/* Now handle the actual rewriting cases. */
2073 	switch (self->syscall_nr) {
2074 	case __NR_getpid:
2075 		syscall_nr_val = __NR_getppid;
2076 		/* Never change syscall return for this case. */
2077 		syscall_ret = NULL;
2078 		break;
2079 	case __NR_gettid:
2080 		syscall_nr_val = -1;
2081 		syscall_ret_val = 45000;
2082 		break;
2083 	case __NR_openat:
2084 		syscall_nr_val = -1;
2085 		syscall_ret_val = -ESRCH;
2086 		break;
2087 	default:
2088 		/* Unhandled, do nothing. */
2089 		return;
2090 	}
2091 
2092 	__change_syscall(_metadata, tracee, syscall_nr, syscall_ret);
2093 }
2094 
2095 FIXTURE_VARIANT(TRACE_syscall) {
2096 	/*
2097 	 * All of the SECCOMP_RET_TRACE behaviors can be tested with either
2098 	 * SECCOMP_RET_TRACE+PTRACE_CONT or plain ptrace()+PTRACE_SYSCALL.
2099 	 * This indicates if we should use SECCOMP_RET_TRACE (false), or
2100 	 * ptrace (true).
2101 	 */
2102 	bool use_ptrace;
2103 };
2104 
2105 FIXTURE_VARIANT_ADD(TRACE_syscall, ptrace) {
2106 	.use_ptrace = true,
2107 };
2108 
2109 FIXTURE_VARIANT_ADD(TRACE_syscall, seccomp) {
2110 	.use_ptrace = false,
2111 };
2112 
2113 FIXTURE_SETUP(TRACE_syscall)
2114 {
2115 	struct sock_filter filter[] = {
2116 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2117 			offsetof(struct seccomp_data, nr)),
2118 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
2119 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002),
2120 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1),
2121 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003),
2122 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_openat, 0, 1),
2123 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004),
2124 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2125 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1005),
2126 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2127 	};
2128 	struct sock_fprog prog = {
2129 		.len = (unsigned short)ARRAY_SIZE(filter),
2130 		.filter = filter,
2131 	};
2132 	long ret;
2133 
2134 	/* Prepare some testable syscall results. */
2135 	self->mytid = syscall(__NR_gettid);
2136 	ASSERT_GT(self->mytid, 0);
2137 	ASSERT_NE(self->mytid, 1) {
2138 		TH_LOG("Running this test as init is not supported. :)");
2139 	}
2140 
2141 	self->mypid = getpid();
2142 	ASSERT_GT(self->mypid, 0);
2143 	ASSERT_EQ(self->mytid, self->mypid);
2144 
2145 	self->parent = getppid();
2146 	ASSERT_GT(self->parent, 0);
2147 	ASSERT_NE(self->parent, self->mypid);
2148 
2149 	/* Launch tracer. */
2150 	self->tracer = setup_trace_fixture(_metadata,
2151 					   variant->use_ptrace ? tracer_ptrace
2152 							       : tracer_seccomp,
2153 					   self, variant->use_ptrace);
2154 
2155 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2156 	ASSERT_EQ(0, ret);
2157 
2158 	/* Do not install seccomp rewrite filters, as we'll use ptrace instead. */
2159 	if (variant->use_ptrace)
2160 		return;
2161 
2162 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2163 	ASSERT_EQ(0, ret);
2164 }
2165 
2166 FIXTURE_TEARDOWN(TRACE_syscall)
2167 {
2168 	teardown_trace_fixture(_metadata, self->tracer);
2169 }
2170 
2171 TEST(negative_ENOSYS)
2172 {
2173 	/*
2174 	 * There should be no difference between an "internal" skip
2175 	 * and userspace asking for syscall "-1".
2176 	 */
2177 	errno = 0;
2178 	EXPECT_EQ(-1, syscall(-1));
2179 	EXPECT_EQ(errno, ENOSYS);
2180 	/* And no difference for "still not valid but not -1". */
2181 	errno = 0;
2182 	EXPECT_EQ(-1, syscall(-101));
2183 	EXPECT_EQ(errno, ENOSYS);
2184 }
2185 
2186 TEST_F(TRACE_syscall, negative_ENOSYS)
2187 {
2188 	negative_ENOSYS(_metadata);
2189 }
2190 
2191 TEST_F(TRACE_syscall, syscall_allowed)
2192 {
2193 	/* getppid works as expected (no changes). */
2194 	EXPECT_EQ(self->parent, syscall(__NR_getppid));
2195 	EXPECT_NE(self->mypid, syscall(__NR_getppid));
2196 }
2197 
2198 TEST_F(TRACE_syscall, syscall_redirected)
2199 {
2200 	/* getpid has been redirected to getppid as expected. */
2201 	EXPECT_EQ(self->parent, syscall(__NR_getpid));
2202 	EXPECT_NE(self->mypid, syscall(__NR_getpid));
2203 }
2204 
2205 TEST_F(TRACE_syscall, syscall_errno)
2206 {
2207 	/* Tracer should skip the open syscall, resulting in ESRCH. */
2208 	EXPECT_SYSCALL_RETURN(-ESRCH, syscall(__NR_openat));
2209 }
2210 
2211 TEST_F(TRACE_syscall, syscall_faked)
2212 {
2213 	/* Tracer skips the gettid syscall and store altered return value. */
2214 	EXPECT_SYSCALL_RETURN(45000, syscall(__NR_gettid));
2215 }
2216 
2217 TEST_F_SIGNAL(TRACE_syscall, kill_immediate, SIGSYS)
2218 {
2219 	struct sock_filter filter[] = {
2220 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2221 			offsetof(struct seccomp_data, nr)),
2222 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_mknodat, 0, 1),
2223 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD),
2224 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2225 	};
2226 	struct sock_fprog prog = {
2227 		.len = (unsigned short)ARRAY_SIZE(filter),
2228 		.filter = filter,
2229 	};
2230 	long ret;
2231 
2232 	/* Install "kill on mknodat" filter. */
2233 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2234 	ASSERT_EQ(0, ret);
2235 
2236 	/* This should immediately die with SIGSYS, regardless of tracer. */
2237 	EXPECT_EQ(-1, syscall(__NR_mknodat, -1, NULL, 0, 0));
2238 }
2239 
2240 TEST_F(TRACE_syscall, skip_after)
2241 {
2242 	struct sock_filter filter[] = {
2243 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2244 			offsetof(struct seccomp_data, nr)),
2245 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2246 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
2247 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2248 	};
2249 	struct sock_fprog prog = {
2250 		.len = (unsigned short)ARRAY_SIZE(filter),
2251 		.filter = filter,
2252 	};
2253 	long ret;
2254 
2255 	/* Install additional "errno on getppid" filter. */
2256 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2257 	ASSERT_EQ(0, ret);
2258 
2259 	/* Tracer will redirect getpid to getppid, and we should see EPERM. */
2260 	errno = 0;
2261 	EXPECT_EQ(-1, syscall(__NR_getpid));
2262 	EXPECT_EQ(EPERM, errno);
2263 }
2264 
2265 TEST_F_SIGNAL(TRACE_syscall, kill_after, SIGSYS)
2266 {
2267 	struct sock_filter filter[] = {
2268 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2269 			offsetof(struct seccomp_data, nr)),
2270 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2271 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2272 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2273 	};
2274 	struct sock_fprog prog = {
2275 		.len = (unsigned short)ARRAY_SIZE(filter),
2276 		.filter = filter,
2277 	};
2278 	long ret;
2279 
2280 	/* Install additional "death on getppid" filter. */
2281 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2282 	ASSERT_EQ(0, ret);
2283 
2284 	/* Tracer will redirect getpid to getppid, and we should die. */
2285 	EXPECT_NE(self->mypid, syscall(__NR_getpid));
2286 }
2287 
2288 TEST(seccomp_syscall)
2289 {
2290 	struct sock_filter filter[] = {
2291 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2292 	};
2293 	struct sock_fprog prog = {
2294 		.len = (unsigned short)ARRAY_SIZE(filter),
2295 		.filter = filter,
2296 	};
2297 	long ret;
2298 
2299 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2300 	ASSERT_EQ(0, ret) {
2301 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2302 	}
2303 
2304 	/* Reject insane operation. */
2305 	ret = seccomp(-1, 0, &prog);
2306 	ASSERT_NE(ENOSYS, errno) {
2307 		TH_LOG("Kernel does not support seccomp syscall!");
2308 	}
2309 	EXPECT_EQ(EINVAL, errno) {
2310 		TH_LOG("Did not reject crazy op value!");
2311 	}
2312 
2313 	/* Reject strict with flags or pointer. */
2314 	ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL);
2315 	EXPECT_EQ(EINVAL, errno) {
2316 		TH_LOG("Did not reject mode strict with flags!");
2317 	}
2318 	ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog);
2319 	EXPECT_EQ(EINVAL, errno) {
2320 		TH_LOG("Did not reject mode strict with uargs!");
2321 	}
2322 
2323 	/* Reject insane args for filter. */
2324 	ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog);
2325 	EXPECT_EQ(EINVAL, errno) {
2326 		TH_LOG("Did not reject crazy filter flags!");
2327 	}
2328 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL);
2329 	EXPECT_EQ(EFAULT, errno) {
2330 		TH_LOG("Did not reject NULL filter!");
2331 	}
2332 
2333 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2334 	EXPECT_EQ(0, errno) {
2335 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s",
2336 			strerror(errno));
2337 	}
2338 }
2339 
2340 TEST(seccomp_syscall_mode_lock)
2341 {
2342 	struct sock_filter filter[] = {
2343 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2344 	};
2345 	struct sock_fprog prog = {
2346 		.len = (unsigned short)ARRAY_SIZE(filter),
2347 		.filter = filter,
2348 	};
2349 	long ret;
2350 
2351 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2352 	ASSERT_EQ(0, ret) {
2353 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2354 	}
2355 
2356 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2357 	ASSERT_NE(ENOSYS, errno) {
2358 		TH_LOG("Kernel does not support seccomp syscall!");
2359 	}
2360 	EXPECT_EQ(0, ret) {
2361 		TH_LOG("Could not install filter!");
2362 	}
2363 
2364 	/* Make sure neither entry point will switch to strict. */
2365 	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0);
2366 	EXPECT_EQ(EINVAL, errno) {
2367 		TH_LOG("Switched to mode strict!");
2368 	}
2369 
2370 	ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL);
2371 	EXPECT_EQ(EINVAL, errno) {
2372 		TH_LOG("Switched to mode strict!");
2373 	}
2374 }
2375 
2376 /*
2377  * Test detection of known and unknown filter flags. Userspace needs to be able
2378  * to check if a filter flag is supported by the current kernel and a good way
2379  * of doing that is by attempting to enter filter mode, with the flag bit in
2380  * question set, and a NULL pointer for the _args_ parameter. EFAULT indicates
2381  * that the flag is valid and EINVAL indicates that the flag is invalid.
2382  */
2383 TEST(detect_seccomp_filter_flags)
2384 {
2385 	unsigned int flags[] = { SECCOMP_FILTER_FLAG_TSYNC,
2386 				 SECCOMP_FILTER_FLAG_LOG,
2387 				 SECCOMP_FILTER_FLAG_SPEC_ALLOW,
2388 				 SECCOMP_FILTER_FLAG_NEW_LISTENER,
2389 				 SECCOMP_FILTER_FLAG_TSYNC_ESRCH };
2390 	unsigned int exclusive[] = {
2391 				SECCOMP_FILTER_FLAG_TSYNC,
2392 				SECCOMP_FILTER_FLAG_NEW_LISTENER };
2393 	unsigned int flag, all_flags, exclusive_mask;
2394 	int i;
2395 	long ret;
2396 
2397 	/* Test detection of individual known-good filter flags */
2398 	for (i = 0, all_flags = 0; i < ARRAY_SIZE(flags); i++) {
2399 		int bits = 0;
2400 
2401 		flag = flags[i];
2402 		/* Make sure the flag is a single bit! */
2403 		while (flag) {
2404 			if (flag & 0x1)
2405 				bits ++;
2406 			flag >>= 1;
2407 		}
2408 		ASSERT_EQ(1, bits);
2409 		flag = flags[i];
2410 
2411 		ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2412 		ASSERT_NE(ENOSYS, errno) {
2413 			TH_LOG("Kernel does not support seccomp syscall!");
2414 		}
2415 		EXPECT_EQ(-1, ret);
2416 		EXPECT_EQ(EFAULT, errno) {
2417 			TH_LOG("Failed to detect that a known-good filter flag (0x%X) is supported!",
2418 			       flag);
2419 		}
2420 
2421 		all_flags |= flag;
2422 	}
2423 
2424 	/*
2425 	 * Test detection of all known-good filter flags combined. But
2426 	 * for the exclusive flags we need to mask them out and try them
2427 	 * individually for the "all flags" testing.
2428 	 */
2429 	exclusive_mask = 0;
2430 	for (i = 0; i < ARRAY_SIZE(exclusive); i++)
2431 		exclusive_mask |= exclusive[i];
2432 	for (i = 0; i < ARRAY_SIZE(exclusive); i++) {
2433 		flag = all_flags & ~exclusive_mask;
2434 		flag |= exclusive[i];
2435 
2436 		ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2437 		EXPECT_EQ(-1, ret);
2438 		EXPECT_EQ(EFAULT, errno) {
2439 			TH_LOG("Failed to detect that all known-good filter flags (0x%X) are supported!",
2440 			       flag);
2441 		}
2442 	}
2443 
2444 	/* Test detection of an unknown filter flags, without exclusives. */
2445 	flag = -1;
2446 	flag &= ~exclusive_mask;
2447 	ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2448 	EXPECT_EQ(-1, ret);
2449 	EXPECT_EQ(EINVAL, errno) {
2450 		TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported!",
2451 		       flag);
2452 	}
2453 
2454 	/*
2455 	 * Test detection of an unknown filter flag that may simply need to be
2456 	 * added to this test
2457 	 */
2458 	flag = flags[ARRAY_SIZE(flags) - 1] << 1;
2459 	ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2460 	EXPECT_EQ(-1, ret);
2461 	EXPECT_EQ(EINVAL, errno) {
2462 		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?",
2463 		       flag);
2464 	}
2465 }
2466 
2467 TEST(TSYNC_first)
2468 {
2469 	struct sock_filter filter[] = {
2470 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2471 	};
2472 	struct sock_fprog prog = {
2473 		.len = (unsigned short)ARRAY_SIZE(filter),
2474 		.filter = filter,
2475 	};
2476 	long ret;
2477 
2478 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2479 	ASSERT_EQ(0, ret) {
2480 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2481 	}
2482 
2483 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2484 		      &prog);
2485 	ASSERT_NE(ENOSYS, errno) {
2486 		TH_LOG("Kernel does not support seccomp syscall!");
2487 	}
2488 	EXPECT_EQ(0, ret) {
2489 		TH_LOG("Could not install initial filter with TSYNC!");
2490 	}
2491 }
2492 
2493 #define TSYNC_SIBLINGS 2
2494 struct tsync_sibling {
2495 	pthread_t tid;
2496 	pid_t system_tid;
2497 	sem_t *started;
2498 	pthread_cond_t *cond;
2499 	pthread_mutex_t *mutex;
2500 	int diverge;
2501 	int num_waits;
2502 	struct sock_fprog *prog;
2503 	struct __test_metadata *metadata;
2504 };
2505 
2506 /*
2507  * To avoid joining joined threads (which is not allowed by Bionic),
2508  * make sure we both successfully join and clear the tid to skip a
2509  * later join attempt during fixture teardown. Any remaining threads
2510  * will be directly killed during teardown.
2511  */
2512 #define PTHREAD_JOIN(tid, status)					\
2513 	do {								\
2514 		int _rc = pthread_join(tid, status);			\
2515 		if (_rc) {						\
2516 			TH_LOG("pthread_join of tid %u failed: %d\n",	\
2517 				(unsigned int)tid, _rc);		\
2518 		} else {						\
2519 			tid = 0;					\
2520 		}							\
2521 	} while (0)
2522 
2523 FIXTURE(TSYNC) {
2524 	struct sock_fprog root_prog, apply_prog;
2525 	struct tsync_sibling sibling[TSYNC_SIBLINGS];
2526 	sem_t started;
2527 	pthread_cond_t cond;
2528 	pthread_mutex_t mutex;
2529 	int sibling_count;
2530 };
2531 
2532 FIXTURE_SETUP(TSYNC)
2533 {
2534 	struct sock_filter root_filter[] = {
2535 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2536 	};
2537 	struct sock_filter apply_filter[] = {
2538 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2539 			offsetof(struct seccomp_data, nr)),
2540 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
2541 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2542 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2543 	};
2544 
2545 	memset(&self->root_prog, 0, sizeof(self->root_prog));
2546 	memset(&self->apply_prog, 0, sizeof(self->apply_prog));
2547 	memset(&self->sibling, 0, sizeof(self->sibling));
2548 	self->root_prog.filter = malloc(sizeof(root_filter));
2549 	ASSERT_NE(NULL, self->root_prog.filter);
2550 	memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter));
2551 	self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter);
2552 
2553 	self->apply_prog.filter = malloc(sizeof(apply_filter));
2554 	ASSERT_NE(NULL, self->apply_prog.filter);
2555 	memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter));
2556 	self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter);
2557 
2558 	self->sibling_count = 0;
2559 	pthread_mutex_init(&self->mutex, NULL);
2560 	pthread_cond_init(&self->cond, NULL);
2561 	sem_init(&self->started, 0, 0);
2562 	self->sibling[0].tid = 0;
2563 	self->sibling[0].cond = &self->cond;
2564 	self->sibling[0].started = &self->started;
2565 	self->sibling[0].mutex = &self->mutex;
2566 	self->sibling[0].diverge = 0;
2567 	self->sibling[0].num_waits = 1;
2568 	self->sibling[0].prog = &self->root_prog;
2569 	self->sibling[0].metadata = _metadata;
2570 	self->sibling[1].tid = 0;
2571 	self->sibling[1].cond = &self->cond;
2572 	self->sibling[1].started = &self->started;
2573 	self->sibling[1].mutex = &self->mutex;
2574 	self->sibling[1].diverge = 0;
2575 	self->sibling[1].prog = &self->root_prog;
2576 	self->sibling[1].num_waits = 1;
2577 	self->sibling[1].metadata = _metadata;
2578 }
2579 
2580 FIXTURE_TEARDOWN(TSYNC)
2581 {
2582 	int sib = 0;
2583 
2584 	if (self->root_prog.filter)
2585 		free(self->root_prog.filter);
2586 	if (self->apply_prog.filter)
2587 		free(self->apply_prog.filter);
2588 
2589 	for ( ; sib < self->sibling_count; ++sib) {
2590 		struct tsync_sibling *s = &self->sibling[sib];
2591 
2592 		if (!s->tid)
2593 			continue;
2594 		/*
2595 		 * If a thread is still running, it may be stuck, so hit
2596 		 * it over the head really hard.
2597 		 */
2598 		pthread_kill(s->tid, 9);
2599 	}
2600 	pthread_mutex_destroy(&self->mutex);
2601 	pthread_cond_destroy(&self->cond);
2602 	sem_destroy(&self->started);
2603 }
2604 
2605 void *tsync_sibling(void *data)
2606 {
2607 	long ret = 0;
2608 	struct tsync_sibling *me = data;
2609 
2610 	me->system_tid = syscall(__NR_gettid);
2611 
2612 	pthread_mutex_lock(me->mutex);
2613 	if (me->diverge) {
2614 		/* Just re-apply the root prog to fork the tree */
2615 		ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
2616 				me->prog, 0, 0);
2617 	}
2618 	sem_post(me->started);
2619 	/* Return outside of started so parent notices failures. */
2620 	if (ret) {
2621 		pthread_mutex_unlock(me->mutex);
2622 		return (void *)SIBLING_EXIT_FAILURE;
2623 	}
2624 	do {
2625 		pthread_cond_wait(me->cond, me->mutex);
2626 		me->num_waits = me->num_waits - 1;
2627 	} while (me->num_waits);
2628 	pthread_mutex_unlock(me->mutex);
2629 
2630 	ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0);
2631 	if (!ret)
2632 		return (void *)SIBLING_EXIT_NEWPRIVS;
2633 	read(-1, NULL, 0);
2634 	return (void *)SIBLING_EXIT_UNKILLED;
2635 }
2636 
2637 void tsync_start_sibling(struct tsync_sibling *sibling)
2638 {
2639 	pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling);
2640 }
2641 
2642 TEST_F(TSYNC, siblings_fail_prctl)
2643 {
2644 	long ret;
2645 	void *status;
2646 	struct sock_filter filter[] = {
2647 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2648 			offsetof(struct seccomp_data, nr)),
2649 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
2650 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL),
2651 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2652 	};
2653 	struct sock_fprog prog = {
2654 		.len = (unsigned short)ARRAY_SIZE(filter),
2655 		.filter = filter,
2656 	};
2657 
2658 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2659 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2660 	}
2661 
2662 	/* Check prctl failure detection by requesting sib 0 diverge. */
2663 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2664 	ASSERT_NE(ENOSYS, errno) {
2665 		TH_LOG("Kernel does not support seccomp syscall!");
2666 	}
2667 	ASSERT_EQ(0, ret) {
2668 		TH_LOG("setting filter failed");
2669 	}
2670 
2671 	self->sibling[0].diverge = 1;
2672 	tsync_start_sibling(&self->sibling[0]);
2673 	tsync_start_sibling(&self->sibling[1]);
2674 
2675 	while (self->sibling_count < TSYNC_SIBLINGS) {
2676 		sem_wait(&self->started);
2677 		self->sibling_count++;
2678 	}
2679 
2680 	/* Signal the threads to clean up*/
2681 	pthread_mutex_lock(&self->mutex);
2682 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2683 		TH_LOG("cond broadcast non-zero");
2684 	}
2685 	pthread_mutex_unlock(&self->mutex);
2686 
2687 	/* Ensure diverging sibling failed to call prctl. */
2688 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2689 	EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status);
2690 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2691 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2692 }
2693 
2694 TEST_F(TSYNC, two_siblings_with_ancestor)
2695 {
2696 	long ret;
2697 	void *status;
2698 
2699 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2700 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2701 	}
2702 
2703 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2704 	ASSERT_NE(ENOSYS, errno) {
2705 		TH_LOG("Kernel does not support seccomp syscall!");
2706 	}
2707 	ASSERT_EQ(0, ret) {
2708 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2709 	}
2710 	tsync_start_sibling(&self->sibling[0]);
2711 	tsync_start_sibling(&self->sibling[1]);
2712 
2713 	while (self->sibling_count < TSYNC_SIBLINGS) {
2714 		sem_wait(&self->started);
2715 		self->sibling_count++;
2716 	}
2717 
2718 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2719 		      &self->apply_prog);
2720 	ASSERT_EQ(0, ret) {
2721 		TH_LOG("Could install filter on all threads!");
2722 	}
2723 	/* Tell the siblings to test the policy */
2724 	pthread_mutex_lock(&self->mutex);
2725 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2726 		TH_LOG("cond broadcast non-zero");
2727 	}
2728 	pthread_mutex_unlock(&self->mutex);
2729 	/* Ensure they are both killed and don't exit cleanly. */
2730 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2731 	EXPECT_EQ(0x0, (long)status);
2732 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2733 	EXPECT_EQ(0x0, (long)status);
2734 }
2735 
2736 TEST_F(TSYNC, two_sibling_want_nnp)
2737 {
2738 	void *status;
2739 
2740 	/* start siblings before any prctl() operations */
2741 	tsync_start_sibling(&self->sibling[0]);
2742 	tsync_start_sibling(&self->sibling[1]);
2743 	while (self->sibling_count < TSYNC_SIBLINGS) {
2744 		sem_wait(&self->started);
2745 		self->sibling_count++;
2746 	}
2747 
2748 	/* Tell the siblings to test no policy */
2749 	pthread_mutex_lock(&self->mutex);
2750 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2751 		TH_LOG("cond broadcast non-zero");
2752 	}
2753 	pthread_mutex_unlock(&self->mutex);
2754 
2755 	/* Ensure they are both upset about lacking nnp. */
2756 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2757 	EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2758 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2759 	EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2760 }
2761 
2762 TEST_F(TSYNC, two_siblings_with_no_filter)
2763 {
2764 	long ret;
2765 	void *status;
2766 
2767 	/* start siblings before any prctl() operations */
2768 	tsync_start_sibling(&self->sibling[0]);
2769 	tsync_start_sibling(&self->sibling[1]);
2770 	while (self->sibling_count < TSYNC_SIBLINGS) {
2771 		sem_wait(&self->started);
2772 		self->sibling_count++;
2773 	}
2774 
2775 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2776 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2777 	}
2778 
2779 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2780 		      &self->apply_prog);
2781 	ASSERT_NE(ENOSYS, errno) {
2782 		TH_LOG("Kernel does not support seccomp syscall!");
2783 	}
2784 	ASSERT_EQ(0, ret) {
2785 		TH_LOG("Could install filter on all threads!");
2786 	}
2787 
2788 	/* Tell the siblings to test the policy */
2789 	pthread_mutex_lock(&self->mutex);
2790 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2791 		TH_LOG("cond broadcast non-zero");
2792 	}
2793 	pthread_mutex_unlock(&self->mutex);
2794 
2795 	/* Ensure they are both killed and don't exit cleanly. */
2796 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2797 	EXPECT_EQ(0x0, (long)status);
2798 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2799 	EXPECT_EQ(0x0, (long)status);
2800 }
2801 
2802 TEST_F(TSYNC, two_siblings_with_one_divergence)
2803 {
2804 	long ret;
2805 	void *status;
2806 
2807 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2808 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2809 	}
2810 
2811 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2812 	ASSERT_NE(ENOSYS, errno) {
2813 		TH_LOG("Kernel does not support seccomp syscall!");
2814 	}
2815 	ASSERT_EQ(0, ret) {
2816 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2817 	}
2818 	self->sibling[0].diverge = 1;
2819 	tsync_start_sibling(&self->sibling[0]);
2820 	tsync_start_sibling(&self->sibling[1]);
2821 
2822 	while (self->sibling_count < TSYNC_SIBLINGS) {
2823 		sem_wait(&self->started);
2824 		self->sibling_count++;
2825 	}
2826 
2827 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2828 		      &self->apply_prog);
2829 	ASSERT_EQ(self->sibling[0].system_tid, ret) {
2830 		TH_LOG("Did not fail on diverged sibling.");
2831 	}
2832 
2833 	/* Wake the threads */
2834 	pthread_mutex_lock(&self->mutex);
2835 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2836 		TH_LOG("cond broadcast non-zero");
2837 	}
2838 	pthread_mutex_unlock(&self->mutex);
2839 
2840 	/* Ensure they are both unkilled. */
2841 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2842 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2843 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2844 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2845 }
2846 
2847 TEST_F(TSYNC, two_siblings_with_one_divergence_no_tid_in_err)
2848 {
2849 	long ret, flags;
2850 	void *status;
2851 
2852 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2853 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2854 	}
2855 
2856 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2857 	ASSERT_NE(ENOSYS, errno) {
2858 		TH_LOG("Kernel does not support seccomp syscall!");
2859 	}
2860 	ASSERT_EQ(0, ret) {
2861 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2862 	}
2863 	self->sibling[0].diverge = 1;
2864 	tsync_start_sibling(&self->sibling[0]);
2865 	tsync_start_sibling(&self->sibling[1]);
2866 
2867 	while (self->sibling_count < TSYNC_SIBLINGS) {
2868 		sem_wait(&self->started);
2869 		self->sibling_count++;
2870 	}
2871 
2872 	flags = SECCOMP_FILTER_FLAG_TSYNC | \
2873 		SECCOMP_FILTER_FLAG_TSYNC_ESRCH;
2874 	ret = seccomp(SECCOMP_SET_MODE_FILTER, flags, &self->apply_prog);
2875 	ASSERT_EQ(ESRCH, errno) {
2876 		TH_LOG("Did not return ESRCH for diverged sibling.");
2877 	}
2878 	ASSERT_EQ(-1, ret) {
2879 		TH_LOG("Did not fail on diverged sibling.");
2880 	}
2881 
2882 	/* Wake the threads */
2883 	pthread_mutex_lock(&self->mutex);
2884 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2885 		TH_LOG("cond broadcast non-zero");
2886 	}
2887 	pthread_mutex_unlock(&self->mutex);
2888 
2889 	/* Ensure they are both unkilled. */
2890 	PTHREAD_JOIN(self->sibling[0].tid, &status);
2891 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2892 	PTHREAD_JOIN(self->sibling[1].tid, &status);
2893 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2894 }
2895 
2896 TEST_F(TSYNC, two_siblings_not_under_filter)
2897 {
2898 	long ret, sib;
2899 	void *status;
2900 	struct timespec delay = { .tv_nsec = 100000000 };
2901 
2902 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2903 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2904 	}
2905 
2906 	/*
2907 	 * Sibling 0 will have its own seccomp policy
2908 	 * and Sibling 1 will not be under seccomp at
2909 	 * all. Sibling 1 will enter seccomp and 0
2910 	 * will cause failure.
2911 	 */
2912 	self->sibling[0].diverge = 1;
2913 	tsync_start_sibling(&self->sibling[0]);
2914 	tsync_start_sibling(&self->sibling[1]);
2915 
2916 	while (self->sibling_count < TSYNC_SIBLINGS) {
2917 		sem_wait(&self->started);
2918 		self->sibling_count++;
2919 	}
2920 
2921 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2922 	ASSERT_NE(ENOSYS, errno) {
2923 		TH_LOG("Kernel does not support seccomp syscall!");
2924 	}
2925 	ASSERT_EQ(0, ret) {
2926 		TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2927 	}
2928 
2929 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2930 		      &self->apply_prog);
2931 	ASSERT_EQ(ret, self->sibling[0].system_tid) {
2932 		TH_LOG("Did not fail on diverged sibling.");
2933 	}
2934 	sib = 1;
2935 	if (ret == self->sibling[0].system_tid)
2936 		sib = 0;
2937 
2938 	pthread_mutex_lock(&self->mutex);
2939 
2940 	/* Increment the other siblings num_waits so we can clean up
2941 	 * the one we just saw.
2942 	 */
2943 	self->sibling[!sib].num_waits += 1;
2944 
2945 	/* Signal the thread to clean up*/
2946 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2947 		TH_LOG("cond broadcast non-zero");
2948 	}
2949 	pthread_mutex_unlock(&self->mutex);
2950 	PTHREAD_JOIN(self->sibling[sib].tid, &status);
2951 	EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2952 	/* Poll for actual task death. pthread_join doesn't guarantee it. */
2953 	while (!kill(self->sibling[sib].system_tid, 0))
2954 		nanosleep(&delay, NULL);
2955 	/* Switch to the remaining sibling */
2956 	sib = !sib;
2957 
2958 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2959 		      &self->apply_prog);
2960 	ASSERT_EQ(0, ret) {
2961 		TH_LOG("Expected the remaining sibling to sync");
2962 	};
2963 
2964 	pthread_mutex_lock(&self->mutex);
2965 
2966 	/* If remaining sibling didn't have a chance to wake up during
2967 	 * the first broadcast, manually reduce the num_waits now.
2968 	 */
2969 	if (self->sibling[sib].num_waits > 1)
2970 		self->sibling[sib].num_waits = 1;
2971 	ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2972 		TH_LOG("cond broadcast non-zero");
2973 	}
2974 	pthread_mutex_unlock(&self->mutex);
2975 	PTHREAD_JOIN(self->sibling[sib].tid, &status);
2976 	EXPECT_EQ(0, (long)status);
2977 	/* Poll for actual task death. pthread_join doesn't guarantee it. */
2978 	while (!kill(self->sibling[sib].system_tid, 0))
2979 		nanosleep(&delay, NULL);
2980 
2981 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2982 		      &self->apply_prog);
2983 	ASSERT_EQ(0, ret);  /* just us chickens */
2984 }
2985 
2986 /* Make sure restarted syscalls are seen directly as "restart_syscall". */
2987 TEST(syscall_restart)
2988 {
2989 	long ret;
2990 	unsigned long msg;
2991 	pid_t child_pid;
2992 	int pipefd[2];
2993 	int status;
2994 	siginfo_t info = { };
2995 	struct sock_filter filter[] = {
2996 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2997 			 offsetof(struct seccomp_data, nr)),
2998 
2999 #ifdef __NR_sigreturn
3000 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 7, 0),
3001 #endif
3002 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 6, 0),
3003 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 5, 0),
3004 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 4, 0),
3005 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 5, 0),
3006 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_clock_nanosleep, 4, 0),
3007 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0),
3008 
3009 		/* Allow __NR_write for easy logging. */
3010 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1),
3011 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3012 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
3013 		/* The nanosleep jump target. */
3014 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100),
3015 		/* The restart_syscall jump target. */
3016 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200),
3017 	};
3018 	struct sock_fprog prog = {
3019 		.len = (unsigned short)ARRAY_SIZE(filter),
3020 		.filter = filter,
3021 	};
3022 #if defined(__arm__)
3023 	struct utsname utsbuf;
3024 #endif
3025 
3026 	ASSERT_EQ(0, pipe(pipefd));
3027 
3028 	child_pid = fork();
3029 	ASSERT_LE(0, child_pid);
3030 	if (child_pid == 0) {
3031 		/* Child uses EXPECT not ASSERT to deliver status correctly. */
3032 		char buf = ' ';
3033 		struct timespec timeout = { };
3034 
3035 		/* Attach parent as tracer and stop. */
3036 		EXPECT_EQ(0, ptrace(PTRACE_TRACEME));
3037 		EXPECT_EQ(0, raise(SIGSTOP));
3038 
3039 		EXPECT_EQ(0, close(pipefd[1]));
3040 
3041 		EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
3042 			TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3043 		}
3044 
3045 		ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
3046 		EXPECT_EQ(0, ret) {
3047 			TH_LOG("Failed to install filter!");
3048 		}
3049 
3050 		EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
3051 			TH_LOG("Failed to read() sync from parent");
3052 		}
3053 		EXPECT_EQ('.', buf) {
3054 			TH_LOG("Failed to get sync data from read()");
3055 		}
3056 
3057 		/* Start nanosleep to be interrupted. */
3058 		timeout.tv_sec = 1;
3059 		errno = 0;
3060 		EXPECT_EQ(0, nanosleep(&timeout, NULL)) {
3061 			TH_LOG("Call to nanosleep() failed (errno %d)", errno);
3062 		}
3063 
3064 		/* Read final sync from parent. */
3065 		EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
3066 			TH_LOG("Failed final read() from parent");
3067 		}
3068 		EXPECT_EQ('!', buf) {
3069 			TH_LOG("Failed to get final data from read()");
3070 		}
3071 
3072 		/* Directly report the status of our test harness results. */
3073 		syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS
3074 						     : EXIT_FAILURE);
3075 	}
3076 	EXPECT_EQ(0, close(pipefd[0]));
3077 
3078 	/* Attach to child, setup options, and release. */
3079 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3080 	ASSERT_EQ(true, WIFSTOPPED(status));
3081 	ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL,
3082 			    PTRACE_O_TRACESECCOMP));
3083 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3084 	ASSERT_EQ(1, write(pipefd[1], ".", 1));
3085 
3086 	/* Wait for nanosleep() to start. */
3087 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3088 	ASSERT_EQ(true, WIFSTOPPED(status));
3089 	ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
3090 	ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
3091 	ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
3092 	ASSERT_EQ(0x100, msg);
3093 	ret = get_syscall(_metadata, child_pid);
3094 	EXPECT_TRUE(ret == __NR_nanosleep || ret == __NR_clock_nanosleep);
3095 
3096 	/* Might as well check siginfo for sanity while we're here. */
3097 	ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
3098 	ASSERT_EQ(SIGTRAP, info.si_signo);
3099 	ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code);
3100 	EXPECT_EQ(0, info.si_errno);
3101 	EXPECT_EQ(getuid(), info.si_uid);
3102 	/* Verify signal delivery came from child (seccomp-triggered). */
3103 	EXPECT_EQ(child_pid, info.si_pid);
3104 
3105 	/* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */
3106 	ASSERT_EQ(0, kill(child_pid, SIGSTOP));
3107 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3108 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3109 	ASSERT_EQ(true, WIFSTOPPED(status));
3110 	ASSERT_EQ(SIGSTOP, WSTOPSIG(status));
3111 	ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
3112 	/*
3113 	 * There is no siginfo on SIGSTOP any more, so we can't verify
3114 	 * signal delivery came from parent now (getpid() == info.si_pid).
3115 	 * https://lkml.kernel.org/r/CAGXu5jJaZAOzP1qFz66tYrtbuywqb+UN2SOA1VLHpCCOiYvYeg@mail.gmail.com
3116 	 * At least verify the SIGSTOP via PTRACE_GETSIGINFO.
3117 	 */
3118 	EXPECT_EQ(SIGSTOP, info.si_signo);
3119 
3120 	/* Restart nanosleep with SIGCONT, which triggers restart_syscall. */
3121 	ASSERT_EQ(0, kill(child_pid, SIGCONT));
3122 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3123 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3124 	ASSERT_EQ(true, WIFSTOPPED(status));
3125 	ASSERT_EQ(SIGCONT, WSTOPSIG(status));
3126 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3127 
3128 	/* Wait for restart_syscall() to start. */
3129 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3130 	ASSERT_EQ(true, WIFSTOPPED(status));
3131 	ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
3132 	ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
3133 	ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
3134 
3135 	ASSERT_EQ(0x200, msg);
3136 	ret = get_syscall(_metadata, child_pid);
3137 #if defined(__arm__)
3138 	/*
3139 	 * FIXME:
3140 	 * - native ARM registers do NOT expose true syscall.
3141 	 * - compat ARM registers on ARM64 DO expose true syscall.
3142 	 */
3143 	ASSERT_EQ(0, uname(&utsbuf));
3144 	if (strncmp(utsbuf.machine, "arm", 3) == 0) {
3145 		EXPECT_EQ(__NR_nanosleep, ret);
3146 	} else
3147 #endif
3148 	{
3149 		EXPECT_EQ(__NR_restart_syscall, ret);
3150 	}
3151 
3152 	/* Write again to end test. */
3153 	ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3154 	ASSERT_EQ(1, write(pipefd[1], "!", 1));
3155 	EXPECT_EQ(0, close(pipefd[1]));
3156 
3157 	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3158 	if (WIFSIGNALED(status) || WEXITSTATUS(status))
3159 		_metadata->passed = 0;
3160 }
3161 
3162 TEST_SIGNAL(filter_flag_log, SIGSYS)
3163 {
3164 	struct sock_filter allow_filter[] = {
3165 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3166 	};
3167 	struct sock_filter kill_filter[] = {
3168 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
3169 			offsetof(struct seccomp_data, nr)),
3170 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
3171 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
3172 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3173 	};
3174 	struct sock_fprog allow_prog = {
3175 		.len = (unsigned short)ARRAY_SIZE(allow_filter),
3176 		.filter = allow_filter,
3177 	};
3178 	struct sock_fprog kill_prog = {
3179 		.len = (unsigned short)ARRAY_SIZE(kill_filter),
3180 		.filter = kill_filter,
3181 	};
3182 	long ret;
3183 	pid_t parent = getppid();
3184 
3185 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3186 	ASSERT_EQ(0, ret);
3187 
3188 	/* Verify that the FILTER_FLAG_LOG flag isn't accepted in strict mode */
3189 	ret = seccomp(SECCOMP_SET_MODE_STRICT, SECCOMP_FILTER_FLAG_LOG,
3190 		      &allow_prog);
3191 	ASSERT_NE(ENOSYS, errno) {
3192 		TH_LOG("Kernel does not support seccomp syscall!");
3193 	}
3194 	EXPECT_NE(0, ret) {
3195 		TH_LOG("Kernel accepted FILTER_FLAG_LOG flag in strict mode!");
3196 	}
3197 	EXPECT_EQ(EINVAL, errno) {
3198 		TH_LOG("Kernel returned unexpected errno for FILTER_FLAG_LOG flag in strict mode!");
3199 	}
3200 
3201 	/* Verify that a simple, permissive filter can be added with no flags */
3202 	ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog);
3203 	EXPECT_EQ(0, ret);
3204 
3205 	/* See if the same filter can be added with the FILTER_FLAG_LOG flag */
3206 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
3207 		      &allow_prog);
3208 	ASSERT_NE(EINVAL, errno) {
3209 		TH_LOG("Kernel does not support the FILTER_FLAG_LOG flag!");
3210 	}
3211 	EXPECT_EQ(0, ret);
3212 
3213 	/* Ensure that the kill filter works with the FILTER_FLAG_LOG flag */
3214 	ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
3215 		      &kill_prog);
3216 	EXPECT_EQ(0, ret);
3217 
3218 	EXPECT_EQ(parent, syscall(__NR_getppid));
3219 	/* getpid() should never return. */
3220 	EXPECT_EQ(0, syscall(__NR_getpid));
3221 }
3222 
3223 TEST(get_action_avail)
3224 {
3225 	__u32 actions[] = { SECCOMP_RET_KILL_THREAD, SECCOMP_RET_TRAP,
3226 			    SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE,
3227 			    SECCOMP_RET_LOG,   SECCOMP_RET_ALLOW };
3228 	__u32 unknown_action = 0x10000000U;
3229 	int i;
3230 	long ret;
3231 
3232 	ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[0]);
3233 	ASSERT_NE(ENOSYS, errno) {
3234 		TH_LOG("Kernel does not support seccomp syscall!");
3235 	}
3236 	ASSERT_NE(EINVAL, errno) {
3237 		TH_LOG("Kernel does not support SECCOMP_GET_ACTION_AVAIL operation!");
3238 	}
3239 	EXPECT_EQ(ret, 0);
3240 
3241 	for (i = 0; i < ARRAY_SIZE(actions); i++) {
3242 		ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[i]);
3243 		EXPECT_EQ(ret, 0) {
3244 			TH_LOG("Expected action (0x%X) not available!",
3245 			       actions[i]);
3246 		}
3247 	}
3248 
3249 	/* Check that an unknown action is handled properly (EOPNOTSUPP) */
3250 	ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &unknown_action);
3251 	EXPECT_EQ(ret, -1);
3252 	EXPECT_EQ(errno, EOPNOTSUPP);
3253 }
3254 
3255 TEST(get_metadata)
3256 {
3257 	pid_t pid;
3258 	int pipefd[2];
3259 	char buf;
3260 	struct seccomp_metadata md;
3261 	long ret;
3262 
3263 	/* Only real root can get metadata. */
3264 	if (geteuid()) {
3265 		SKIP(return, "get_metadata requires real root");
3266 		return;
3267 	}
3268 
3269 	ASSERT_EQ(0, pipe(pipefd));
3270 
3271 	pid = fork();
3272 	ASSERT_GE(pid, 0);
3273 	if (pid == 0) {
3274 		struct sock_filter filter[] = {
3275 			BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3276 		};
3277 		struct sock_fprog prog = {
3278 			.len = (unsigned short)ARRAY_SIZE(filter),
3279 			.filter = filter,
3280 		};
3281 
3282 		/* one with log, one without */
3283 		EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER,
3284 				     SECCOMP_FILTER_FLAG_LOG, &prog));
3285 		EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog));
3286 
3287 		EXPECT_EQ(0, close(pipefd[0]));
3288 		ASSERT_EQ(1, write(pipefd[1], "1", 1));
3289 		ASSERT_EQ(0, close(pipefd[1]));
3290 
3291 		while (1)
3292 			sleep(100);
3293 	}
3294 
3295 	ASSERT_EQ(0, close(pipefd[1]));
3296 	ASSERT_EQ(1, read(pipefd[0], &buf, 1));
3297 
3298 	ASSERT_EQ(0, ptrace(PTRACE_ATTACH, pid));
3299 	ASSERT_EQ(pid, waitpid(pid, NULL, 0));
3300 
3301 	/* Past here must not use ASSERT or child process is never killed. */
3302 
3303 	md.filter_off = 0;
3304 	errno = 0;
3305 	ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md);
3306 	EXPECT_EQ(sizeof(md), ret) {
3307 		if (errno == EINVAL)
3308 			SKIP(goto skip, "Kernel does not support PTRACE_SECCOMP_GET_METADATA (missing CONFIG_CHECKPOINT_RESTORE?)");
3309 	}
3310 
3311 	EXPECT_EQ(md.flags, SECCOMP_FILTER_FLAG_LOG);
3312 	EXPECT_EQ(md.filter_off, 0);
3313 
3314 	md.filter_off = 1;
3315 	ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md);
3316 	EXPECT_EQ(sizeof(md), ret);
3317 	EXPECT_EQ(md.flags, 0);
3318 	EXPECT_EQ(md.filter_off, 1);
3319 
3320 skip:
3321 	ASSERT_EQ(0, kill(pid, SIGKILL));
3322 }
3323 
3324 static int user_notif_syscall(int nr, unsigned int flags)
3325 {
3326 	struct sock_filter filter[] = {
3327 		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
3328 			offsetof(struct seccomp_data, nr)),
3329 		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, nr, 0, 1),
3330 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_USER_NOTIF),
3331 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3332 	};
3333 
3334 	struct sock_fprog prog = {
3335 		.len = (unsigned short)ARRAY_SIZE(filter),
3336 		.filter = filter,
3337 	};
3338 
3339 	return seccomp(SECCOMP_SET_MODE_FILTER, flags, &prog);
3340 }
3341 
3342 #define USER_NOTIF_MAGIC INT_MAX
3343 TEST(user_notification_basic)
3344 {
3345 	pid_t pid;
3346 	long ret;
3347 	int status, listener;
3348 	struct seccomp_notif req = {};
3349 	struct seccomp_notif_resp resp = {};
3350 	struct pollfd pollfd;
3351 
3352 	struct sock_filter filter[] = {
3353 		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3354 	};
3355 	struct sock_fprog prog = {
3356 		.len = (unsigned short)ARRAY_SIZE(filter),
3357 		.filter = filter,
3358 	};
3359 
3360 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3361 	ASSERT_EQ(0, ret) {
3362 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3363 	}
3364 
3365 	pid = fork();
3366 	ASSERT_GE(pid, 0);
3367 
3368 	/* Check that we get -ENOSYS with no listener attached */
3369 	if (pid == 0) {
3370 		if (user_notif_syscall(__NR_getppid, 0) < 0)
3371 			exit(1);
3372 		ret = syscall(__NR_getppid);
3373 		exit(ret >= 0 || errno != ENOSYS);
3374 	}
3375 
3376 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3377 	EXPECT_EQ(true, WIFEXITED(status));
3378 	EXPECT_EQ(0, WEXITSTATUS(status));
3379 
3380 	/* Add some no-op filters for grins. */
3381 	EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3382 	EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3383 	EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3384 	EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3385 
3386 	/* Check that the basic notification machinery works */
3387 	listener = user_notif_syscall(__NR_getppid,
3388 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
3389 	ASSERT_GE(listener, 0);
3390 
3391 	/* Installing a second listener in the chain should EBUSY */
3392 	EXPECT_EQ(user_notif_syscall(__NR_getppid,
3393 				     SECCOMP_FILTER_FLAG_NEW_LISTENER),
3394 		  -1);
3395 	EXPECT_EQ(errno, EBUSY);
3396 
3397 	pid = fork();
3398 	ASSERT_GE(pid, 0);
3399 
3400 	if (pid == 0) {
3401 		ret = syscall(__NR_getppid);
3402 		exit(ret != USER_NOTIF_MAGIC);
3403 	}
3404 
3405 	pollfd.fd = listener;
3406 	pollfd.events = POLLIN | POLLOUT;
3407 
3408 	EXPECT_GT(poll(&pollfd, 1, -1), 0);
3409 	EXPECT_EQ(pollfd.revents, POLLIN);
3410 
3411 	/* Test that we can't pass garbage to the kernel. */
3412 	memset(&req, 0, sizeof(req));
3413 	req.pid = -1;
3414 	errno = 0;
3415 	ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req);
3416 	EXPECT_EQ(-1, ret);
3417 	EXPECT_EQ(EINVAL, errno);
3418 
3419 	if (ret) {
3420 		req.pid = 0;
3421 		EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3422 	}
3423 
3424 	pollfd.fd = listener;
3425 	pollfd.events = POLLIN | POLLOUT;
3426 
3427 	EXPECT_GT(poll(&pollfd, 1, -1), 0);
3428 	EXPECT_EQ(pollfd.revents, POLLOUT);
3429 
3430 	EXPECT_EQ(req.data.nr,  __NR_getppid);
3431 
3432 	resp.id = req.id;
3433 	resp.error = 0;
3434 	resp.val = USER_NOTIF_MAGIC;
3435 
3436 	/* check that we make sure flags == 0 */
3437 	resp.flags = 1;
3438 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3439 	EXPECT_EQ(errno, EINVAL);
3440 
3441 	resp.flags = 0;
3442 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3443 
3444 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3445 	EXPECT_EQ(true, WIFEXITED(status));
3446 	EXPECT_EQ(0, WEXITSTATUS(status));
3447 }
3448 
3449 TEST(user_notification_with_tsync)
3450 {
3451 	int ret;
3452 	unsigned int flags;
3453 
3454 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3455 	ASSERT_EQ(0, ret) {
3456 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3457 	}
3458 
3459 	/* these were exclusive */
3460 	flags = SECCOMP_FILTER_FLAG_NEW_LISTENER |
3461 		SECCOMP_FILTER_FLAG_TSYNC;
3462 	ASSERT_EQ(-1, user_notif_syscall(__NR_getppid, flags));
3463 	ASSERT_EQ(EINVAL, errno);
3464 
3465 	/* but now they're not */
3466 	flags |= SECCOMP_FILTER_FLAG_TSYNC_ESRCH;
3467 	ret = user_notif_syscall(__NR_getppid, flags);
3468 	close(ret);
3469 	ASSERT_LE(0, ret);
3470 }
3471 
3472 TEST(user_notification_kill_in_middle)
3473 {
3474 	pid_t pid;
3475 	long ret;
3476 	int listener;
3477 	struct seccomp_notif req = {};
3478 	struct seccomp_notif_resp resp = {};
3479 
3480 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3481 	ASSERT_EQ(0, ret) {
3482 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3483 	}
3484 
3485 	listener = user_notif_syscall(__NR_getppid,
3486 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
3487 	ASSERT_GE(listener, 0);
3488 
3489 	/*
3490 	 * Check that nothing bad happens when we kill the task in the middle
3491 	 * of a syscall.
3492 	 */
3493 	pid = fork();
3494 	ASSERT_GE(pid, 0);
3495 
3496 	if (pid == 0) {
3497 		ret = syscall(__NR_getppid);
3498 		exit(ret != USER_NOTIF_MAGIC);
3499 	}
3500 
3501 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3502 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), 0);
3503 
3504 	EXPECT_EQ(kill(pid, SIGKILL), 0);
3505 	EXPECT_EQ(waitpid(pid, NULL, 0), pid);
3506 
3507 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), -1);
3508 
3509 	resp.id = req.id;
3510 	ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp);
3511 	EXPECT_EQ(ret, -1);
3512 	EXPECT_EQ(errno, ENOENT);
3513 }
3514 
3515 static int handled = -1;
3516 
3517 static void signal_handler(int signal)
3518 {
3519 	if (write(handled, "c", 1) != 1)
3520 		perror("write from signal");
3521 }
3522 
3523 TEST(user_notification_signal)
3524 {
3525 	pid_t pid;
3526 	long ret;
3527 	int status, listener, sk_pair[2];
3528 	struct seccomp_notif req = {};
3529 	struct seccomp_notif_resp resp = {};
3530 	char c;
3531 
3532 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3533 	ASSERT_EQ(0, ret) {
3534 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3535 	}
3536 
3537 	ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0);
3538 
3539 	listener = user_notif_syscall(__NR_gettid,
3540 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
3541 	ASSERT_GE(listener, 0);
3542 
3543 	pid = fork();
3544 	ASSERT_GE(pid, 0);
3545 
3546 	if (pid == 0) {
3547 		close(sk_pair[0]);
3548 		handled = sk_pair[1];
3549 		if (signal(SIGUSR1, signal_handler) == SIG_ERR) {
3550 			perror("signal");
3551 			exit(1);
3552 		}
3553 		/*
3554 		 * ERESTARTSYS behavior is a bit hard to test, because we need
3555 		 * to rely on a signal that has not yet been handled. Let's at
3556 		 * least check that the error code gets propagated through, and
3557 		 * hope that it doesn't break when there is actually a signal :)
3558 		 */
3559 		ret = syscall(__NR_gettid);
3560 		exit(!(ret == -1 && errno == 512));
3561 	}
3562 
3563 	close(sk_pair[1]);
3564 
3565 	memset(&req, 0, sizeof(req));
3566 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3567 
3568 	EXPECT_EQ(kill(pid, SIGUSR1), 0);
3569 
3570 	/*
3571 	 * Make sure the signal really is delivered, which means we're not
3572 	 * stuck in the user notification code any more and the notification
3573 	 * should be dead.
3574 	 */
3575 	EXPECT_EQ(read(sk_pair[0], &c, 1), 1);
3576 
3577 	resp.id = req.id;
3578 	resp.error = -EPERM;
3579 	resp.val = 0;
3580 
3581 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3582 	EXPECT_EQ(errno, ENOENT);
3583 
3584 	memset(&req, 0, sizeof(req));
3585 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3586 
3587 	resp.id = req.id;
3588 	resp.error = -512; /* -ERESTARTSYS */
3589 	resp.val = 0;
3590 
3591 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3592 
3593 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3594 	EXPECT_EQ(true, WIFEXITED(status));
3595 	EXPECT_EQ(0, WEXITSTATUS(status));
3596 }
3597 
3598 TEST(user_notification_closed_listener)
3599 {
3600 	pid_t pid;
3601 	long ret;
3602 	int status, listener;
3603 
3604 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3605 	ASSERT_EQ(0, ret) {
3606 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3607 	}
3608 
3609 	listener = user_notif_syscall(__NR_getppid,
3610 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
3611 	ASSERT_GE(listener, 0);
3612 
3613 	/*
3614 	 * Check that we get an ENOSYS when the listener is closed.
3615 	 */
3616 	pid = fork();
3617 	ASSERT_GE(pid, 0);
3618 	if (pid == 0) {
3619 		close(listener);
3620 		ret = syscall(__NR_getppid);
3621 		exit(ret != -1 && errno != ENOSYS);
3622 	}
3623 
3624 	close(listener);
3625 
3626 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3627 	EXPECT_EQ(true, WIFEXITED(status));
3628 	EXPECT_EQ(0, WEXITSTATUS(status));
3629 }
3630 
3631 /*
3632  * Check that a pid in a child namespace still shows up as valid in ours.
3633  */
3634 TEST(user_notification_child_pid_ns)
3635 {
3636 	pid_t pid;
3637 	int status, listener;
3638 	struct seccomp_notif req = {};
3639 	struct seccomp_notif_resp resp = {};
3640 
3641 	ASSERT_EQ(unshare(CLONE_NEWUSER | CLONE_NEWPID), 0) {
3642 		if (errno == EINVAL)
3643 			SKIP(return, "kernel missing CLONE_NEWUSER support");
3644 	};
3645 
3646 	listener = user_notif_syscall(__NR_getppid,
3647 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
3648 	ASSERT_GE(listener, 0);
3649 
3650 	pid = fork();
3651 	ASSERT_GE(pid, 0);
3652 
3653 	if (pid == 0)
3654 		exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3655 
3656 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3657 	EXPECT_EQ(req.pid, pid);
3658 
3659 	resp.id = req.id;
3660 	resp.error = 0;
3661 	resp.val = USER_NOTIF_MAGIC;
3662 
3663 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3664 
3665 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3666 	EXPECT_EQ(true, WIFEXITED(status));
3667 	EXPECT_EQ(0, WEXITSTATUS(status));
3668 	close(listener);
3669 }
3670 
3671 /*
3672  * Check that a pid in a sibling (i.e. unrelated) namespace shows up as 0, i.e.
3673  * invalid.
3674  */
3675 TEST(user_notification_sibling_pid_ns)
3676 {
3677 	pid_t pid, pid2;
3678 	int status, listener;
3679 	struct seccomp_notif req = {};
3680 	struct seccomp_notif_resp resp = {};
3681 
3682 	ASSERT_EQ(prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0), 0) {
3683 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3684 	}
3685 
3686 	listener = user_notif_syscall(__NR_getppid,
3687 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
3688 	ASSERT_GE(listener, 0);
3689 
3690 	pid = fork();
3691 	ASSERT_GE(pid, 0);
3692 
3693 	if (pid == 0) {
3694 		ASSERT_EQ(unshare(CLONE_NEWPID), 0);
3695 
3696 		pid2 = fork();
3697 		ASSERT_GE(pid2, 0);
3698 
3699 		if (pid2 == 0)
3700 			exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3701 
3702 		EXPECT_EQ(waitpid(pid2, &status, 0), pid2);
3703 		EXPECT_EQ(true, WIFEXITED(status));
3704 		EXPECT_EQ(0, WEXITSTATUS(status));
3705 		exit(WEXITSTATUS(status));
3706 	}
3707 
3708 	/* Create the sibling ns, and sibling in it. */
3709 	ASSERT_EQ(unshare(CLONE_NEWPID), 0) {
3710 		if (errno == EPERM)
3711 			SKIP(return, "CLONE_NEWPID requires CAP_SYS_ADMIN");
3712 	}
3713 	ASSERT_EQ(errno, 0);
3714 
3715 	pid2 = fork();
3716 	ASSERT_GE(pid2, 0);
3717 
3718 	if (pid2 == 0) {
3719 		ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3720 		/*
3721 		 * The pid should be 0, i.e. the task is in some namespace that
3722 		 * we can't "see".
3723 		 */
3724 		EXPECT_EQ(req.pid, 0);
3725 
3726 		resp.id = req.id;
3727 		resp.error = 0;
3728 		resp.val = USER_NOTIF_MAGIC;
3729 
3730 		ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3731 		exit(0);
3732 	}
3733 
3734 	close(listener);
3735 
3736 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3737 	EXPECT_EQ(true, WIFEXITED(status));
3738 	EXPECT_EQ(0, WEXITSTATUS(status));
3739 
3740 	EXPECT_EQ(waitpid(pid2, &status, 0), pid2);
3741 	EXPECT_EQ(true, WIFEXITED(status));
3742 	EXPECT_EQ(0, WEXITSTATUS(status));
3743 }
3744 
3745 TEST(user_notification_fault_recv)
3746 {
3747 	pid_t pid;
3748 	int status, listener;
3749 	struct seccomp_notif req = {};
3750 	struct seccomp_notif_resp resp = {};
3751 
3752 	ASSERT_EQ(unshare(CLONE_NEWUSER), 0) {
3753 		if (errno == EINVAL)
3754 			SKIP(return, "kernel missing CLONE_NEWUSER support");
3755 	}
3756 
3757 	listener = user_notif_syscall(__NR_getppid,
3758 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
3759 	ASSERT_GE(listener, 0);
3760 
3761 	pid = fork();
3762 	ASSERT_GE(pid, 0);
3763 
3764 	if (pid == 0)
3765 		exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3766 
3767 	/* Do a bad recv() */
3768 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, NULL), -1);
3769 	EXPECT_EQ(errno, EFAULT);
3770 
3771 	/* We should still be able to receive this notification, though. */
3772 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3773 	EXPECT_EQ(req.pid, pid);
3774 
3775 	resp.id = req.id;
3776 	resp.error = 0;
3777 	resp.val = USER_NOTIF_MAGIC;
3778 
3779 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3780 
3781 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3782 	EXPECT_EQ(true, WIFEXITED(status));
3783 	EXPECT_EQ(0, WEXITSTATUS(status));
3784 }
3785 
3786 TEST(seccomp_get_notif_sizes)
3787 {
3788 	struct seccomp_notif_sizes sizes;
3789 
3790 	ASSERT_EQ(seccomp(SECCOMP_GET_NOTIF_SIZES, 0, &sizes), 0);
3791 	EXPECT_EQ(sizes.seccomp_notif, sizeof(struct seccomp_notif));
3792 	EXPECT_EQ(sizes.seccomp_notif_resp, sizeof(struct seccomp_notif_resp));
3793 }
3794 
3795 TEST(user_notification_continue)
3796 {
3797 	pid_t pid;
3798 	long ret;
3799 	int status, listener;
3800 	struct seccomp_notif req = {};
3801 	struct seccomp_notif_resp resp = {};
3802 	struct pollfd pollfd;
3803 
3804 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3805 	ASSERT_EQ(0, ret) {
3806 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3807 	}
3808 
3809 	listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER);
3810 	ASSERT_GE(listener, 0);
3811 
3812 	pid = fork();
3813 	ASSERT_GE(pid, 0);
3814 
3815 	if (pid == 0) {
3816 		int dup_fd, pipe_fds[2];
3817 		pid_t self;
3818 
3819 		ASSERT_GE(pipe(pipe_fds), 0);
3820 
3821 		dup_fd = dup(pipe_fds[0]);
3822 		ASSERT_GE(dup_fd, 0);
3823 		EXPECT_NE(pipe_fds[0], dup_fd);
3824 
3825 		self = getpid();
3826 		ASSERT_EQ(filecmp(self, self, pipe_fds[0], dup_fd), 0);
3827 		exit(0);
3828 	}
3829 
3830 	pollfd.fd = listener;
3831 	pollfd.events = POLLIN | POLLOUT;
3832 
3833 	EXPECT_GT(poll(&pollfd, 1, -1), 0);
3834 	EXPECT_EQ(pollfd.revents, POLLIN);
3835 
3836 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3837 
3838 	pollfd.fd = listener;
3839 	pollfd.events = POLLIN | POLLOUT;
3840 
3841 	EXPECT_GT(poll(&pollfd, 1, -1), 0);
3842 	EXPECT_EQ(pollfd.revents, POLLOUT);
3843 
3844 	EXPECT_EQ(req.data.nr, __NR_dup);
3845 
3846 	resp.id = req.id;
3847 	resp.flags = SECCOMP_USER_NOTIF_FLAG_CONTINUE;
3848 
3849 	/*
3850 	 * Verify that setting SECCOMP_USER_NOTIF_FLAG_CONTINUE enforces other
3851 	 * args be set to 0.
3852 	 */
3853 	resp.error = 0;
3854 	resp.val = USER_NOTIF_MAGIC;
3855 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3856 	EXPECT_EQ(errno, EINVAL);
3857 
3858 	resp.error = USER_NOTIF_MAGIC;
3859 	resp.val = 0;
3860 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3861 	EXPECT_EQ(errno, EINVAL);
3862 
3863 	resp.error = 0;
3864 	resp.val = 0;
3865 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0) {
3866 		if (errno == EINVAL)
3867 			SKIP(goto skip, "Kernel does not support SECCOMP_USER_NOTIF_FLAG_CONTINUE");
3868 	}
3869 
3870 skip:
3871 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3872 	EXPECT_EQ(true, WIFEXITED(status));
3873 	EXPECT_EQ(0, WEXITSTATUS(status)) {
3874 		if (WEXITSTATUS(status) == 2) {
3875 			SKIP(return, "Kernel does not support kcmp() syscall");
3876 			return;
3877 		}
3878 	}
3879 }
3880 
3881 TEST(user_notification_filter_empty)
3882 {
3883 	pid_t pid;
3884 	long ret;
3885 	int status;
3886 	struct pollfd pollfd;
3887 	struct __clone_args args = {
3888 		.flags = CLONE_FILES,
3889 		.exit_signal = SIGCHLD,
3890 	};
3891 
3892 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3893 	ASSERT_EQ(0, ret) {
3894 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3895 	}
3896 
3897 	pid = sys_clone3(&args, sizeof(args));
3898 	ASSERT_GE(pid, 0);
3899 
3900 	if (pid == 0) {
3901 		int listener;
3902 
3903 		listener = user_notif_syscall(__NR_mknodat, SECCOMP_FILTER_FLAG_NEW_LISTENER);
3904 		if (listener < 0)
3905 			_exit(EXIT_FAILURE);
3906 
3907 		if (dup2(listener, 200) != 200)
3908 			_exit(EXIT_FAILURE);
3909 
3910 		close(listener);
3911 
3912 		_exit(EXIT_SUCCESS);
3913 	}
3914 
3915 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
3916 	EXPECT_EQ(true, WIFEXITED(status));
3917 	EXPECT_EQ(0, WEXITSTATUS(status));
3918 
3919 	/*
3920 	 * The seccomp filter has become unused so we should be notified once
3921 	 * the kernel gets around to cleaning up task struct.
3922 	 */
3923 	pollfd.fd = 200;
3924 	pollfd.events = POLLHUP;
3925 
3926 	EXPECT_GT(poll(&pollfd, 1, 2000), 0);
3927 	EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0);
3928 }
3929 
3930 static void *do_thread(void *data)
3931 {
3932 	return NULL;
3933 }
3934 
3935 TEST(user_notification_filter_empty_threaded)
3936 {
3937 	pid_t pid;
3938 	long ret;
3939 	int status;
3940 	struct pollfd pollfd;
3941 	struct __clone_args args = {
3942 		.flags = CLONE_FILES,
3943 		.exit_signal = SIGCHLD,
3944 	};
3945 
3946 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3947 	ASSERT_EQ(0, ret) {
3948 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3949 	}
3950 
3951 	pid = sys_clone3(&args, sizeof(args));
3952 	ASSERT_GE(pid, 0);
3953 
3954 	if (pid == 0) {
3955 		pid_t pid1, pid2;
3956 		int listener, status;
3957 		pthread_t thread;
3958 
3959 		listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER);
3960 		if (listener < 0)
3961 			_exit(EXIT_FAILURE);
3962 
3963 		if (dup2(listener, 200) != 200)
3964 			_exit(EXIT_FAILURE);
3965 
3966 		close(listener);
3967 
3968 		pid1 = fork();
3969 		if (pid1 < 0)
3970 			_exit(EXIT_FAILURE);
3971 
3972 		if (pid1 == 0)
3973 			_exit(EXIT_SUCCESS);
3974 
3975 		pid2 = fork();
3976 		if (pid2 < 0)
3977 			_exit(EXIT_FAILURE);
3978 
3979 		if (pid2 == 0)
3980 			_exit(EXIT_SUCCESS);
3981 
3982 		if (pthread_create(&thread, NULL, do_thread, NULL) ||
3983 		    pthread_join(thread, NULL))
3984 			_exit(EXIT_FAILURE);
3985 
3986 		if (pthread_create(&thread, NULL, do_thread, NULL) ||
3987 		    pthread_join(thread, NULL))
3988 			_exit(EXIT_FAILURE);
3989 
3990 		if (waitpid(pid1, &status, 0) != pid1 || !WIFEXITED(status) ||
3991 		    WEXITSTATUS(status))
3992 			_exit(EXIT_FAILURE);
3993 
3994 		if (waitpid(pid2, &status, 0) != pid2 || !WIFEXITED(status) ||
3995 		    WEXITSTATUS(status))
3996 			_exit(EXIT_FAILURE);
3997 
3998 		exit(EXIT_SUCCESS);
3999 	}
4000 
4001 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
4002 	EXPECT_EQ(true, WIFEXITED(status));
4003 	EXPECT_EQ(0, WEXITSTATUS(status));
4004 
4005 	/*
4006 	 * The seccomp filter has become unused so we should be notified once
4007 	 * the kernel gets around to cleaning up task struct.
4008 	 */
4009 	pollfd.fd = 200;
4010 	pollfd.events = POLLHUP;
4011 
4012 	EXPECT_GT(poll(&pollfd, 1, 2000), 0);
4013 	EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0);
4014 }
4015 
4016 TEST(user_notification_addfd)
4017 {
4018 	pid_t pid;
4019 	long ret;
4020 	int status, listener, memfd, fd, nextfd;
4021 	struct seccomp_notif_addfd addfd = {};
4022 	struct seccomp_notif_addfd_small small = {};
4023 	struct seccomp_notif_addfd_big big = {};
4024 	struct seccomp_notif req = {};
4025 	struct seccomp_notif_resp resp = {};
4026 	/* 100 ms */
4027 	struct timespec delay = { .tv_nsec = 100000000 };
4028 
4029 	/* There may be arbitrary already-open fds at test start. */
4030 	memfd = memfd_create("test", 0);
4031 	ASSERT_GE(memfd, 0);
4032 	nextfd = memfd + 1;
4033 
4034 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4035 	ASSERT_EQ(0, ret) {
4036 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4037 	}
4038 
4039 	/* fd: 4 */
4040 	/* Check that the basic notification machinery works */
4041 	listener = user_notif_syscall(__NR_getppid,
4042 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
4043 	ASSERT_EQ(listener, nextfd++);
4044 
4045 	pid = fork();
4046 	ASSERT_GE(pid, 0);
4047 
4048 	if (pid == 0) {
4049 		/* fds will be added and this value is expected */
4050 		if (syscall(__NR_getppid) != USER_NOTIF_MAGIC)
4051 			exit(1);
4052 
4053 		/* Atomic addfd+send is received here. Check it is a valid fd */
4054 		if (fcntl(syscall(__NR_getppid), F_GETFD) == -1)
4055 			exit(1);
4056 
4057 		exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
4058 	}
4059 
4060 	ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4061 
4062 	addfd.srcfd = memfd;
4063 	addfd.newfd = 0;
4064 	addfd.id = req.id;
4065 	addfd.flags = 0x0;
4066 
4067 	/* Verify bad newfd_flags cannot be set */
4068 	addfd.newfd_flags = ~O_CLOEXEC;
4069 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4070 	EXPECT_EQ(errno, EINVAL);
4071 	addfd.newfd_flags = O_CLOEXEC;
4072 
4073 	/* Verify bad flags cannot be set */
4074 	addfd.flags = 0xff;
4075 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4076 	EXPECT_EQ(errno, EINVAL);
4077 	addfd.flags = 0;
4078 
4079 	/* Verify that remote_fd cannot be set without setting flags */
4080 	addfd.newfd = 1;
4081 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4082 	EXPECT_EQ(errno, EINVAL);
4083 	addfd.newfd = 0;
4084 
4085 	/* Verify small size cannot be set */
4086 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_SMALL, &small), -1);
4087 	EXPECT_EQ(errno, EINVAL);
4088 
4089 	/* Verify we can't send bits filled in unknown buffer area */
4090 	memset(&big, 0xAA, sizeof(big));
4091 	big.addfd = addfd;
4092 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big), -1);
4093 	EXPECT_EQ(errno, E2BIG);
4094 
4095 
4096 	/* Verify we can set an arbitrary remote fd */
4097 	fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd);
4098 	EXPECT_EQ(fd, nextfd++);
4099 	EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0);
4100 
4101 	/* Verify we can set an arbitrary remote fd with large size */
4102 	memset(&big, 0x0, sizeof(big));
4103 	big.addfd = addfd;
4104 	fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big);
4105 	EXPECT_EQ(fd, nextfd++);
4106 
4107 	/* Verify we can set a specific remote fd */
4108 	addfd.newfd = 42;
4109 	addfd.flags = SECCOMP_ADDFD_FLAG_SETFD;
4110 	fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd);
4111 	EXPECT_EQ(fd, 42);
4112 	EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0);
4113 
4114 	/* Resume syscall */
4115 	resp.id = req.id;
4116 	resp.error = 0;
4117 	resp.val = USER_NOTIF_MAGIC;
4118 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4119 
4120 	/*
4121 	 * This sets the ID of the ADD FD to the last request plus 1. The
4122 	 * notification ID increments 1 per notification.
4123 	 */
4124 	addfd.id = req.id + 1;
4125 
4126 	/* This spins until the underlying notification is generated */
4127 	while (ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd) != -1 &&
4128 	       errno != -EINPROGRESS)
4129 		nanosleep(&delay, NULL);
4130 
4131 	memset(&req, 0, sizeof(req));
4132 	ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4133 	ASSERT_EQ(addfd.id, req.id);
4134 
4135 	/* Verify we can do an atomic addfd and send */
4136 	addfd.newfd = 0;
4137 	addfd.flags = SECCOMP_ADDFD_FLAG_SEND;
4138 	fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd);
4139 	/*
4140 	 * Child has earlier "low" fds and now 42, so we expect the next
4141 	 * lowest available fd to be assigned here.
4142 	 */
4143 	EXPECT_EQ(fd, nextfd++);
4144 	ASSERT_EQ(filecmp(getpid(), pid, memfd, fd), 0);
4145 
4146 	/*
4147 	 * This sets the ID of the ADD FD to the last request plus 1. The
4148 	 * notification ID increments 1 per notification.
4149 	 */
4150 	addfd.id = req.id + 1;
4151 
4152 	/* This spins until the underlying notification is generated */
4153 	while (ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd) != -1 &&
4154 	       errno != -EINPROGRESS)
4155 		nanosleep(&delay, NULL);
4156 
4157 	memset(&req, 0, sizeof(req));
4158 	ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4159 	ASSERT_EQ(addfd.id, req.id);
4160 
4161 	resp.id = req.id;
4162 	resp.error = 0;
4163 	resp.val = USER_NOTIF_MAGIC;
4164 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4165 
4166 	/* Wait for child to finish. */
4167 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
4168 	EXPECT_EQ(true, WIFEXITED(status));
4169 	EXPECT_EQ(0, WEXITSTATUS(status));
4170 
4171 	close(memfd);
4172 }
4173 
4174 TEST(user_notification_addfd_rlimit)
4175 {
4176 	pid_t pid;
4177 	long ret;
4178 	int status, listener, memfd;
4179 	struct seccomp_notif_addfd addfd = {};
4180 	struct seccomp_notif req = {};
4181 	struct seccomp_notif_resp resp = {};
4182 	const struct rlimit lim = {
4183 		.rlim_cur	= 0,
4184 		.rlim_max	= 0,
4185 	};
4186 
4187 	memfd = memfd_create("test", 0);
4188 	ASSERT_GE(memfd, 0);
4189 
4190 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4191 	ASSERT_EQ(0, ret) {
4192 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4193 	}
4194 
4195 	/* Check that the basic notification machinery works */
4196 	listener = user_notif_syscall(__NR_getppid,
4197 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
4198 	ASSERT_GE(listener, 0);
4199 
4200 	pid = fork();
4201 	ASSERT_GE(pid, 0);
4202 
4203 	if (pid == 0)
4204 		exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
4205 
4206 
4207 	ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4208 
4209 	ASSERT_EQ(prlimit(pid, RLIMIT_NOFILE, &lim, NULL), 0);
4210 
4211 	addfd.srcfd = memfd;
4212 	addfd.newfd_flags = O_CLOEXEC;
4213 	addfd.newfd = 0;
4214 	addfd.id = req.id;
4215 	addfd.flags = 0;
4216 
4217 	/* Should probably spot check /proc/sys/fs/file-nr */
4218 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4219 	EXPECT_EQ(errno, EMFILE);
4220 
4221 	addfd.flags = SECCOMP_ADDFD_FLAG_SEND;
4222 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4223 	EXPECT_EQ(errno, EMFILE);
4224 
4225 	addfd.newfd = 100;
4226 	addfd.flags = SECCOMP_ADDFD_FLAG_SETFD;
4227 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4228 	EXPECT_EQ(errno, EBADF);
4229 
4230 	resp.id = req.id;
4231 	resp.error = 0;
4232 	resp.val = USER_NOTIF_MAGIC;
4233 
4234 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4235 
4236 	/* Wait for child to finish. */
4237 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
4238 	EXPECT_EQ(true, WIFEXITED(status));
4239 	EXPECT_EQ(0, WEXITSTATUS(status));
4240 
4241 	close(memfd);
4242 }
4243 
4244 /* Make sure PTRACE_O_SUSPEND_SECCOMP requires CAP_SYS_ADMIN. */
4245 FIXTURE(O_SUSPEND_SECCOMP) {
4246 	pid_t pid;
4247 };
4248 
4249 FIXTURE_SETUP(O_SUSPEND_SECCOMP)
4250 {
4251 	ERRNO_FILTER(block_read, E2BIG);
4252 	cap_value_t cap_list[] = { CAP_SYS_ADMIN };
4253 	cap_t caps;
4254 
4255 	self->pid = 0;
4256 
4257 	/* make sure we don't have CAP_SYS_ADMIN */
4258 	caps = cap_get_proc();
4259 	ASSERT_NE(NULL, caps);
4260 	ASSERT_EQ(0, cap_set_flag(caps, CAP_EFFECTIVE, 1, cap_list, CAP_CLEAR));
4261 	ASSERT_EQ(0, cap_set_proc(caps));
4262 	cap_free(caps);
4263 
4264 	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0));
4265 	ASSERT_EQ(0, prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_block_read));
4266 
4267 	self->pid = fork();
4268 	ASSERT_GE(self->pid, 0);
4269 
4270 	if (self->pid == 0) {
4271 		while (1)
4272 			pause();
4273 		_exit(127);
4274 	}
4275 }
4276 
4277 FIXTURE_TEARDOWN(O_SUSPEND_SECCOMP)
4278 {
4279 	if (self->pid)
4280 		kill(self->pid, SIGKILL);
4281 }
4282 
4283 TEST_F(O_SUSPEND_SECCOMP, setoptions)
4284 {
4285 	int wstatus;
4286 
4287 	ASSERT_EQ(0, ptrace(PTRACE_ATTACH, self->pid, NULL, 0));
4288 	ASSERT_EQ(self->pid, wait(&wstatus));
4289 	ASSERT_EQ(-1, ptrace(PTRACE_SETOPTIONS, self->pid, NULL, PTRACE_O_SUSPEND_SECCOMP));
4290 	if (errno == EINVAL)
4291 		SKIP(return, "Kernel does not support PTRACE_O_SUSPEND_SECCOMP (missing CONFIG_CHECKPOINT_RESTORE?)");
4292 	ASSERT_EQ(EPERM, errno);
4293 }
4294 
4295 TEST_F(O_SUSPEND_SECCOMP, seize)
4296 {
4297 	int ret;
4298 
4299 	ret = ptrace(PTRACE_SEIZE, self->pid, NULL, PTRACE_O_SUSPEND_SECCOMP);
4300 	ASSERT_EQ(-1, ret);
4301 	if (errno == EINVAL)
4302 		SKIP(return, "Kernel does not support PTRACE_O_SUSPEND_SECCOMP (missing CONFIG_CHECKPOINT_RESTORE?)");
4303 	ASSERT_EQ(EPERM, errno);
4304 }
4305 
4306 /*
4307  * get_nth - Get the nth, space separated entry in a file.
4308  *
4309  * Returns the length of the read field.
4310  * Throws error if field is zero-lengthed.
4311  */
4312 static ssize_t get_nth(struct __test_metadata *_metadata, const char *path,
4313 		     const unsigned int position, char **entry)
4314 {
4315 	char *line = NULL;
4316 	unsigned int i;
4317 	ssize_t nread;
4318 	size_t len = 0;
4319 	FILE *f;
4320 
4321 	f = fopen(path, "r");
4322 	ASSERT_NE(f, NULL) {
4323 		TH_LOG("Could not open %s: %s", path, strerror(errno));
4324 	}
4325 
4326 	for (i = 0; i < position; i++) {
4327 		nread = getdelim(&line, &len, ' ', f);
4328 		ASSERT_GE(nread, 0) {
4329 			TH_LOG("Failed to read %d entry in file %s", i, path);
4330 		}
4331 	}
4332 	fclose(f);
4333 
4334 	ASSERT_GT(nread, 0) {
4335 		TH_LOG("Entry in file %s had zero length", path);
4336 	}
4337 
4338 	*entry = line;
4339 	return nread - 1;
4340 }
4341 
4342 /* For a given PID, get the task state (D, R, etc...) */
4343 static char get_proc_stat(struct __test_metadata *_metadata, pid_t pid)
4344 {
4345 	char proc_path[100] = {0};
4346 	char status;
4347 	char *line;
4348 
4349 	snprintf(proc_path, sizeof(proc_path), "/proc/%d/stat", pid);
4350 	ASSERT_EQ(get_nth(_metadata, proc_path, 3, &line), 1);
4351 
4352 	status = *line;
4353 	free(line);
4354 
4355 	return status;
4356 }
4357 
4358 TEST(user_notification_fifo)
4359 {
4360 	struct seccomp_notif_resp resp = {};
4361 	struct seccomp_notif req = {};
4362 	int i, status, listener;
4363 	pid_t pid, pids[3];
4364 	__u64 baseid;
4365 	long ret;
4366 	/* 100 ms */
4367 	struct timespec delay = { .tv_nsec = 100000000 };
4368 
4369 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4370 	ASSERT_EQ(0, ret) {
4371 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4372 	}
4373 
4374 	/* Setup a listener */
4375 	listener = user_notif_syscall(__NR_getppid,
4376 				      SECCOMP_FILTER_FLAG_NEW_LISTENER);
4377 	ASSERT_GE(listener, 0);
4378 
4379 	pid = fork();
4380 	ASSERT_GE(pid, 0);
4381 
4382 	if (pid == 0) {
4383 		ret = syscall(__NR_getppid);
4384 		exit(ret != USER_NOTIF_MAGIC);
4385 	}
4386 
4387 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4388 	baseid = req.id + 1;
4389 
4390 	resp.id = req.id;
4391 	resp.error = 0;
4392 	resp.val = USER_NOTIF_MAGIC;
4393 
4394 	/* check that we make sure flags == 0 */
4395 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4396 
4397 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
4398 	EXPECT_EQ(true, WIFEXITED(status));
4399 	EXPECT_EQ(0, WEXITSTATUS(status));
4400 
4401 	/* Start children, and generate notifications */
4402 	for (i = 0; i < ARRAY_SIZE(pids); i++) {
4403 		pid = fork();
4404 		if (pid == 0) {
4405 			ret = syscall(__NR_getppid);
4406 			exit(ret != USER_NOTIF_MAGIC);
4407 		}
4408 		pids[i] = pid;
4409 	}
4410 
4411 	/* This spins until all of the children are sleeping */
4412 restart_wait:
4413 	for (i = 0; i < ARRAY_SIZE(pids); i++) {
4414 		if (get_proc_stat(_metadata, pids[i]) != 'S') {
4415 			nanosleep(&delay, NULL);
4416 			goto restart_wait;
4417 		}
4418 	}
4419 
4420 	/* Read the notifications in order (and respond) */
4421 	for (i = 0; i < ARRAY_SIZE(pids); i++) {
4422 		memset(&req, 0, sizeof(req));
4423 		EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4424 		EXPECT_EQ(req.id, baseid + i);
4425 		resp.id = req.id;
4426 		EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4427 	}
4428 
4429 	/* Make sure notifications were received */
4430 	for (i = 0; i < ARRAY_SIZE(pids); i++) {
4431 		EXPECT_EQ(waitpid(pids[i], &status, 0), pids[i]);
4432 		EXPECT_EQ(true, WIFEXITED(status));
4433 		EXPECT_EQ(0, WEXITSTATUS(status));
4434 	}
4435 }
4436 
4437 /* get_proc_syscall - Get the syscall in progress for a given pid
4438  *
4439  * Returns the current syscall number for a given process
4440  * Returns -1 if not in syscall (running or blocked)
4441  */
4442 static long get_proc_syscall(struct __test_metadata *_metadata, int pid)
4443 {
4444 	char proc_path[100] = {0};
4445 	long ret = -1;
4446 	ssize_t nread;
4447 	char *line;
4448 
4449 	snprintf(proc_path, sizeof(proc_path), "/proc/%d/syscall", pid);
4450 	nread = get_nth(_metadata, proc_path, 1, &line);
4451 	ASSERT_GT(nread, 0);
4452 
4453 	if (!strncmp("running", line, MIN(7, nread)))
4454 		ret = strtol(line, NULL, 16);
4455 
4456 	free(line);
4457 	return ret;
4458 }
4459 
4460 /* Ensure non-fatal signals prior to receive are unmodified */
4461 TEST(user_notification_wait_killable_pre_notification)
4462 {
4463 	struct sigaction new_action = {
4464 		.sa_handler = signal_handler,
4465 	};
4466 	int listener, status, sk_pair[2];
4467 	pid_t pid;
4468 	long ret;
4469 	char c;
4470 	/* 100 ms */
4471 	struct timespec delay = { .tv_nsec = 100000000 };
4472 
4473 	ASSERT_EQ(sigemptyset(&new_action.sa_mask), 0);
4474 
4475 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4476 	ASSERT_EQ(0, ret)
4477 	{
4478 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4479 	}
4480 
4481 	ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0);
4482 
4483 	listener = user_notif_syscall(
4484 		__NR_getppid, SECCOMP_FILTER_FLAG_NEW_LISTENER |
4485 				      SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV);
4486 	ASSERT_GE(listener, 0);
4487 
4488 	/*
4489 	 * Check that we can kill the process with SIGUSR1 prior to receiving
4490 	 * the notification. SIGUSR1 is wired up to a custom signal handler,
4491 	 * and make sure it gets called.
4492 	 */
4493 	pid = fork();
4494 	ASSERT_GE(pid, 0);
4495 
4496 	if (pid == 0) {
4497 		close(sk_pair[0]);
4498 		handled = sk_pair[1];
4499 
4500 		/* Setup the non-fatal sigaction without SA_RESTART */
4501 		if (sigaction(SIGUSR1, &new_action, NULL)) {
4502 			perror("sigaction");
4503 			exit(1);
4504 		}
4505 
4506 		ret = syscall(__NR_getppid);
4507 		/* Make sure we got a return from a signal interruption */
4508 		exit(ret != -1 || errno != EINTR);
4509 	}
4510 
4511 	/*
4512 	 * Make sure we've gotten to the seccomp user notification wait
4513 	 * from getppid prior to sending any signals
4514 	 */
4515 	while (get_proc_syscall(_metadata, pid) != __NR_getppid &&
4516 	       get_proc_stat(_metadata, pid) != 'S')
4517 		nanosleep(&delay, NULL);
4518 
4519 	/* Send non-fatal kill signal */
4520 	EXPECT_EQ(kill(pid, SIGUSR1), 0);
4521 
4522 	/* wait for process to exit (exit checks for EINTR) */
4523 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
4524 	EXPECT_EQ(true, WIFEXITED(status));
4525 	EXPECT_EQ(0, WEXITSTATUS(status));
4526 
4527 	EXPECT_EQ(read(sk_pair[0], &c, 1), 1);
4528 }
4529 
4530 /* Ensure non-fatal signals after receive are blocked */
4531 TEST(user_notification_wait_killable)
4532 {
4533 	struct sigaction new_action = {
4534 		.sa_handler = signal_handler,
4535 	};
4536 	struct seccomp_notif_resp resp = {};
4537 	struct seccomp_notif req = {};
4538 	int listener, status, sk_pair[2];
4539 	pid_t pid;
4540 	long ret;
4541 	char c;
4542 	/* 100 ms */
4543 	struct timespec delay = { .tv_nsec = 100000000 };
4544 
4545 	ASSERT_EQ(sigemptyset(&new_action.sa_mask), 0);
4546 
4547 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4548 	ASSERT_EQ(0, ret)
4549 	{
4550 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4551 	}
4552 
4553 	ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0);
4554 
4555 	listener = user_notif_syscall(
4556 		__NR_getppid, SECCOMP_FILTER_FLAG_NEW_LISTENER |
4557 				      SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV);
4558 	ASSERT_GE(listener, 0);
4559 
4560 	pid = fork();
4561 	ASSERT_GE(pid, 0);
4562 
4563 	if (pid == 0) {
4564 		close(sk_pair[0]);
4565 		handled = sk_pair[1];
4566 
4567 		/* Setup the sigaction without SA_RESTART */
4568 		if (sigaction(SIGUSR1, &new_action, NULL)) {
4569 			perror("sigaction");
4570 			exit(1);
4571 		}
4572 
4573 		/* Make sure that the syscall is completed (no EINTR) */
4574 		ret = syscall(__NR_getppid);
4575 		exit(ret != USER_NOTIF_MAGIC);
4576 	}
4577 
4578 	/*
4579 	 * Get the notification, to make move the notifying process into a
4580 	 * non-preemptible (TASK_KILLABLE) state.
4581 	 */
4582 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4583 	/* Send non-fatal kill signal */
4584 	EXPECT_EQ(kill(pid, SIGUSR1), 0);
4585 
4586 	/*
4587 	 * Make sure the task enters moves to TASK_KILLABLE by waiting for
4588 	 * D (Disk Sleep) state after receiving non-fatal signal.
4589 	 */
4590 	while (get_proc_stat(_metadata, pid) != 'D')
4591 		nanosleep(&delay, NULL);
4592 
4593 	resp.id = req.id;
4594 	resp.val = USER_NOTIF_MAGIC;
4595 	/* Make sure the notification is found and able to be replied to */
4596 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4597 
4598 	/*
4599 	 * Make sure that the signal handler does get called once we're back in
4600 	 * userspace.
4601 	 */
4602 	EXPECT_EQ(read(sk_pair[0], &c, 1), 1);
4603 	/* wait for process to exit (exit checks for USER_NOTIF_MAGIC) */
4604 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
4605 	EXPECT_EQ(true, WIFEXITED(status));
4606 	EXPECT_EQ(0, WEXITSTATUS(status));
4607 }
4608 
4609 /* Ensure fatal signals after receive are not blocked */
4610 TEST(user_notification_wait_killable_fatal)
4611 {
4612 	struct seccomp_notif req = {};
4613 	int listener, status;
4614 	pid_t pid;
4615 	long ret;
4616 	/* 100 ms */
4617 	struct timespec delay = { .tv_nsec = 100000000 };
4618 
4619 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4620 	ASSERT_EQ(0, ret)
4621 	{
4622 		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4623 	}
4624 
4625 	listener = user_notif_syscall(
4626 		__NR_getppid, SECCOMP_FILTER_FLAG_NEW_LISTENER |
4627 				      SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV);
4628 	ASSERT_GE(listener, 0);
4629 
4630 	pid = fork();
4631 	ASSERT_GE(pid, 0);
4632 
4633 	if (pid == 0) {
4634 		/* This should never complete as it should get a SIGTERM */
4635 		syscall(__NR_getppid);
4636 		exit(1);
4637 	}
4638 
4639 	while (get_proc_stat(_metadata, pid) != 'S')
4640 		nanosleep(&delay, NULL);
4641 
4642 	/*
4643 	 * Get the notification, to make move the notifying process into a
4644 	 * non-preemptible (TASK_KILLABLE) state.
4645 	 */
4646 	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4647 	/* Kill the process with a fatal signal */
4648 	EXPECT_EQ(kill(pid, SIGTERM), 0);
4649 
4650 	/*
4651 	 * Wait for the process to exit, and make sure the process terminated
4652 	 * due to the SIGTERM signal.
4653 	 */
4654 	EXPECT_EQ(waitpid(pid, &status, 0), pid);
4655 	EXPECT_EQ(true, WIFSIGNALED(status));
4656 	EXPECT_EQ(SIGTERM, WTERMSIG(status));
4657 }
4658 
4659 /*
4660  * TODO:
4661  * - expand NNP testing
4662  * - better arch-specific TRACE and TRAP handlers.
4663  * - endianness checking when appropriate
4664  * - 64-bit arg prodding
4665  * - arch value testing (x86 modes especially)
4666  * - verify that FILTER_FLAG_LOG filters generate log messages
4667  * - verify that RET_LOG generates log messages
4668  */
4669 
4670 TEST_HARNESS_MAIN
4671