2 * Copyright (c) 2012 The Chromium OS Authors. All rights reserved.
3 * Use of this source code is governed by the GPLv2 license.
5 * Test code for seccomp bpf.
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
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
24 #include <linux/filter.h>
25 #include <sys/prctl.h>
26 #include <sys/ptrace.h>
28 #include <linux/prctl.h>
29 #include <linux/ptrace.h>
30 #include <linux/seccomp.h>
32 #include <semaphore.h>
38 #include <linux/elf.h>
40 #include <sys/utsname.h>
41 #include <sys/fcntl.h>
43 #include <sys/times.h>
44 #include <sys/socket.h>
45 #include <sys/ioctl.h>
48 #include <sys/syscall.h>
51 #include "../kselftest_harness.h"
53 #ifndef PR_SET_PTRACER
54 # define PR_SET_PTRACER 0x59616d61
57 #ifndef PR_SET_NO_NEW_PRIVS
58 #define PR_SET_NO_NEW_PRIVS 38
59 #define PR_GET_NO_NEW_PRIVS 39
62 #ifndef PR_SECCOMP_EXT
63 #define PR_SECCOMP_EXT 43
66 #ifndef SECCOMP_EXT_ACT
67 #define SECCOMP_EXT_ACT 1
70 #ifndef SECCOMP_EXT_ACT_TSYNC
71 #define SECCOMP_EXT_ACT_TSYNC 1
74 #ifndef SECCOMP_MODE_STRICT
75 #define SECCOMP_MODE_STRICT 1
78 #ifndef SECCOMP_MODE_FILTER
79 #define SECCOMP_MODE_FILTER 2
82 #ifndef SECCOMP_RET_ALLOW
86 __u64 instruction_pointer;
91 #ifndef SECCOMP_RET_KILL_PROCESS
92 #define SECCOMP_RET_KILL_PROCESS 0x80000000U /* kill the process */
93 #define SECCOMP_RET_KILL_THREAD 0x00000000U /* kill the thread */
95 #ifndef SECCOMP_RET_KILL
96 #define SECCOMP_RET_KILL SECCOMP_RET_KILL_THREAD
97 #define SECCOMP_RET_TRAP 0x00030000U /* disallow and force a SIGSYS */
98 #define SECCOMP_RET_ERRNO 0x00050000U /* returns an errno */
99 #define SECCOMP_RET_TRACE 0x7ff00000U /* pass to a tracer or disallow */
100 #define SECCOMP_RET_ALLOW 0x7fff0000U /* allow */
102 #ifndef SECCOMP_RET_LOG
103 #define SECCOMP_RET_LOG 0x7ffc0000U /* allow after logging */
107 # if defined(__i386__)
108 # define __NR_seccomp 354
109 # elif defined(__x86_64__)
110 # define __NR_seccomp 317
111 # elif defined(__arm__)
112 # define __NR_seccomp 383
113 # elif defined(__aarch64__)
114 # define __NR_seccomp 277
115 # elif defined(__hppa__)
116 # define __NR_seccomp 338
117 # elif defined(__powerpc__)
118 # define __NR_seccomp 358
119 # elif defined(__s390__)
120 # define __NR_seccomp 348
122 # warning "seccomp syscall number unknown for this architecture"
123 # define __NR_seccomp 0xffff
127 #ifndef SECCOMP_SET_MODE_STRICT
128 #define SECCOMP_SET_MODE_STRICT 0
131 #ifndef SECCOMP_SET_MODE_FILTER
132 #define SECCOMP_SET_MODE_FILTER 1
135 #ifndef SECCOMP_GET_ACTION_AVAIL
136 #define SECCOMP_GET_ACTION_AVAIL 2
139 #ifndef SECCOMP_GET_NOTIF_SIZES
140 #define SECCOMP_GET_NOTIF_SIZES 3
143 #ifndef SECCOMP_FILTER_FLAG_TSYNC
144 #define SECCOMP_FILTER_FLAG_TSYNC (1UL << 0)
147 #ifndef SECCOMP_FILTER_FLAG_LOG
148 #define SECCOMP_FILTER_FLAG_LOG (1UL << 1)
151 #ifndef SECCOMP_FILTER_FLAG_SPEC_ALLOW
152 #define SECCOMP_FILTER_FLAG_SPEC_ALLOW (1UL << 2)
155 #ifndef PTRACE_SECCOMP_GET_METADATA
156 #define PTRACE_SECCOMP_GET_METADATA 0x420d
158 struct seccomp_metadata {
159 __u64 filter_off; /* Input: which filter */
160 __u64 flags; /* Output: filter's flags */
164 #ifndef SECCOMP_FILTER_FLAG_NEW_LISTENER
165 #define SECCOMP_FILTER_FLAG_NEW_LISTENER (1UL << 3)
167 #define SECCOMP_RET_USER_NOTIF 0x7fc00000U
169 #define SECCOMP_IOC_MAGIC '!'
170 #define SECCOMP_IO(nr) _IO(SECCOMP_IOC_MAGIC, nr)
171 #define SECCOMP_IOR(nr, type) _IOR(SECCOMP_IOC_MAGIC, nr, type)
172 #define SECCOMP_IOW(nr, type) _IOW(SECCOMP_IOC_MAGIC, nr, type)
173 #define SECCOMP_IOWR(nr, type) _IOWR(SECCOMP_IOC_MAGIC, nr, type)
175 /* Flags for seccomp notification fd ioctl. */
176 #define SECCOMP_IOCTL_NOTIF_RECV SECCOMP_IOWR(0, struct seccomp_notif)
177 #define SECCOMP_IOCTL_NOTIF_SEND SECCOMP_IOWR(1, \
178 struct seccomp_notif_resp)
179 #define SECCOMP_IOCTL_NOTIF_ID_VALID SECCOMP_IOR(2, __u64)
181 struct seccomp_notif {
185 struct seccomp_data data;
188 struct seccomp_notif_resp {
195 struct seccomp_notif_sizes {
197 __u16 seccomp_notif_resp;
203 int seccomp(unsigned int op, unsigned int flags, void *args)
206 return syscall(__NR_seccomp, op, flags, args);
210 #if __BYTE_ORDER == __LITTLE_ENDIAN
211 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]))
212 #elif __BYTE_ORDER == __BIG_ENDIAN
213 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32))
215 #error "wut? Unknown __BYTE_ORDER?!"
218 #define SIBLING_EXIT_UNKILLED 0xbadbeef
219 #define SIBLING_EXIT_FAILURE 0xbadface
220 #define SIBLING_EXIT_NEWPRIVS 0xbadfeed
222 TEST(mode_strict_support)
226 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
228 TH_LOG("Kernel does not support CONFIG_SECCOMP");
230 syscall(__NR_exit, 0);
233 TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL)
237 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
239 TH_LOG("Kernel does not support CONFIG_SECCOMP");
241 syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
244 TH_LOG("Unreachable!");
248 /* Note! This doesn't test no new privs behavior */
249 TEST(no_new_privs_support)
253 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
255 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
259 /* Tests kernel support by checking for a copy_from_user() fault on NULL. */
260 TEST(mode_filter_support)
264 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
266 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
268 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL);
270 EXPECT_EQ(EFAULT, errno) {
271 TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!");
275 TEST(mode_filter_without_nnp)
277 struct sock_filter filter[] = {
278 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
280 struct sock_fprog prog = {
281 .len = (unsigned short)ARRAY_SIZE(filter),
286 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0);
288 TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS");
291 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
292 /* Succeeds with CAP_SYS_ADMIN, fails without */
293 /* TODO(wad) check caps not euid */
296 EXPECT_EQ(EACCES, errno);
302 #define MAX_INSNS_PER_PATH 32768
304 TEST(filter_size_limits)
307 int count = BPF_MAXINSNS + 1;
308 struct sock_filter allow[] = {
309 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
311 struct sock_filter *filter;
312 struct sock_fprog prog = { };
315 filter = calloc(count, sizeof(*filter));
316 ASSERT_NE(NULL, filter);
318 for (i = 0; i < count; i++)
319 filter[i] = allow[0];
321 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
324 prog.filter = filter;
327 /* Too many filter instructions in a single filter. */
328 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
330 TH_LOG("Installing %d insn filter was allowed", prog.len);
333 /* One less is okay, though. */
335 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
337 TH_LOG("Installing %d insn filter wasn't allowed", prog.len);
341 TEST(filter_chain_limits)
344 int count = BPF_MAXINSNS;
345 struct sock_filter allow[] = {
346 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
348 struct sock_filter *filter;
349 struct sock_fprog prog = { };
352 filter = calloc(count, sizeof(*filter));
353 ASSERT_NE(NULL, filter);
355 for (i = 0; i < count; i++)
356 filter[i] = allow[0];
358 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
361 prog.filter = filter;
364 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
369 /* Too many total filter instructions. */
370 for (i = 0; i < MAX_INSNS_PER_PATH; i++) {
371 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
376 TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)",
377 i, count, i * (count + 4));
381 TEST(mode_filter_cannot_move_to_strict)
383 struct sock_filter filter[] = {
384 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
386 struct sock_fprog prog = {
387 .len = (unsigned short)ARRAY_SIZE(filter),
392 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
395 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
398 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0);
400 EXPECT_EQ(EINVAL, errno);
404 TEST(mode_filter_get_seccomp)
406 struct sock_filter filter[] = {
407 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
409 struct sock_fprog prog = {
410 .len = (unsigned short)ARRAY_SIZE(filter),
415 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
418 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
421 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
424 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
431 struct sock_filter filter[] = {
432 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
434 struct sock_fprog prog = {
435 .len = (unsigned short)ARRAY_SIZE(filter),
440 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
443 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
449 struct sock_filter filter[] = {
451 struct sock_fprog prog = {
452 .len = (unsigned short)ARRAY_SIZE(filter),
457 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
460 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
462 EXPECT_EQ(EINVAL, errno);
467 struct sock_filter filter[] = {
468 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG),
470 struct sock_fprog prog = {
471 .len = (unsigned short)ARRAY_SIZE(filter),
475 pid_t parent = getppid();
477 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
480 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
483 /* getppid() should succeed and be logged (no check for logging) */
484 EXPECT_EQ(parent, syscall(__NR_getppid));
487 TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS)
489 struct sock_filter filter[] = {
490 BPF_STMT(BPF_RET|BPF_K, 0x10000000U),
492 struct sock_fprog prog = {
493 .len = (unsigned short)ARRAY_SIZE(filter),
498 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
501 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
503 EXPECT_EQ(0, syscall(__NR_getpid)) {
504 TH_LOG("getpid() shouldn't ever return");
508 /* return code >= 0x80000000 is unused. */
509 TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS)
511 struct sock_filter filter[] = {
512 BPF_STMT(BPF_RET|BPF_K, 0x90000000U),
514 struct sock_fprog prog = {
515 .len = (unsigned short)ARRAY_SIZE(filter),
520 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
523 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
525 EXPECT_EQ(0, syscall(__NR_getpid)) {
526 TH_LOG("getpid() shouldn't ever return");
530 TEST_SIGNAL(KILL_all, SIGSYS)
532 struct sock_filter filter[] = {
533 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
535 struct sock_fprog prog = {
536 .len = (unsigned short)ARRAY_SIZE(filter),
541 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
544 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
548 TEST_SIGNAL(KILL_one, SIGSYS)
550 struct sock_filter filter[] = {
551 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
552 offsetof(struct seccomp_data, nr)),
553 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
554 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
555 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
557 struct sock_fprog prog = {
558 .len = (unsigned short)ARRAY_SIZE(filter),
562 pid_t parent = getppid();
564 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
567 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
570 EXPECT_EQ(parent, syscall(__NR_getppid));
571 /* getpid() should never return. */
572 EXPECT_EQ(0, syscall(__NR_getpid));
575 TEST_SIGNAL(KILL_one_arg_one, SIGSYS)
578 struct sock_filter filter[] = {
579 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
580 offsetof(struct seccomp_data, nr)),
581 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0),
582 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
583 /* Only both with lower 32-bit for now. */
584 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)),
585 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K,
586 (unsigned long)&fatal_address, 0, 1),
587 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
588 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
590 struct sock_fprog prog = {
591 .len = (unsigned short)ARRAY_SIZE(filter),
595 pid_t parent = getppid();
597 clock_t clock = times(&timebuf);
599 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
602 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
605 EXPECT_EQ(parent, syscall(__NR_getppid));
606 EXPECT_LE(clock, syscall(__NR_times, &timebuf));
607 /* times() should never return. */
608 EXPECT_EQ(0, syscall(__NR_times, &fatal_address));
611 TEST_SIGNAL(KILL_one_arg_six, SIGSYS)
614 int sysno = __NR_mmap;
616 int sysno = __NR_mmap2;
618 struct sock_filter filter[] = {
619 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
620 offsetof(struct seccomp_data, nr)),
621 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0),
622 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
623 /* Only both with lower 32-bit for now. */
624 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)),
625 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1),
626 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
627 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
629 struct sock_fprog prog = {
630 .len = (unsigned short)ARRAY_SIZE(filter),
634 pid_t parent = getppid();
637 int page_size = sysconf(_SC_PAGESIZE);
639 ASSERT_LT(0, page_size);
641 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
644 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
647 fd = open("/dev/zero", O_RDONLY);
650 EXPECT_EQ(parent, syscall(__NR_getppid));
651 map1 = (void *)syscall(sysno,
652 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size);
653 EXPECT_NE(MAP_FAILED, map1);
654 /* mmap2() should never return. */
655 map2 = (void *)syscall(sysno,
656 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE);
657 EXPECT_EQ(MAP_FAILED, map2);
659 /* The test failed, so clean up the resources. */
660 munmap(map1, page_size);
661 munmap(map2, page_size);
665 /* This is a thread task to die via seccomp filter violation. */
666 void *kill_thread(void *data)
668 bool die = (bool)data;
671 prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
672 return (void *)SIBLING_EXIT_FAILURE;
675 return (void *)SIBLING_EXIT_UNKILLED;
678 /* Prepare a thread that will kill itself or both of us. */
679 void kill_thread_or_group(struct __test_metadata *_metadata, bool kill_process)
683 /* Kill only when calling __NR_prctl. */
684 struct sock_filter filter_thread[] = {
685 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
686 offsetof(struct seccomp_data, nr)),
687 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
688 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD),
689 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
691 struct sock_fprog prog_thread = {
692 .len = (unsigned short)ARRAY_SIZE(filter_thread),
693 .filter = filter_thread,
695 struct sock_filter filter_process[] = {
696 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
697 offsetof(struct seccomp_data, nr)),
698 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
699 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_PROCESS),
700 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
702 struct sock_fprog prog_process = {
703 .len = (unsigned short)ARRAY_SIZE(filter_process),
704 .filter = filter_process,
707 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
708 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
711 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0,
712 kill_process ? &prog_process : &prog_thread));
715 * Add the KILL_THREAD rule again to make sure that the KILL_PROCESS
716 * flag cannot be downgraded by a new filter.
718 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog_thread));
720 /* Start a thread that will exit immediately. */
721 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)false));
722 ASSERT_EQ(0, pthread_join(thread, &status));
723 ASSERT_EQ(SIBLING_EXIT_UNKILLED, (unsigned long)status);
725 /* Start a thread that will die immediately. */
726 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)true));
727 ASSERT_EQ(0, pthread_join(thread, &status));
728 ASSERT_NE(SIBLING_EXIT_FAILURE, (unsigned long)status);
731 * If we get here, only the spawned thread died. Let the parent know
732 * the whole process didn't die (i.e. this thread, the spawner,
744 ASSERT_LE(0, child_pid);
745 if (child_pid == 0) {
746 kill_thread_or_group(_metadata, false);
750 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
752 /* If only the thread was killed, we'll see exit 42. */
753 ASSERT_TRUE(WIFEXITED(status));
754 ASSERT_EQ(42, WEXITSTATUS(status));
763 ASSERT_LE(0, child_pid);
764 if (child_pid == 0) {
765 kill_thread_or_group(_metadata, true);
769 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
771 /* If the entire process was killed, we'll see SIGSYS. */
772 ASSERT_TRUE(WIFSIGNALED(status));
773 ASSERT_EQ(SIGSYS, WTERMSIG(status));
776 /* TODO(wad) add 64-bit versus 32-bit arg tests. */
777 TEST(arg_out_of_range)
779 struct sock_filter filter[] = {
780 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)),
781 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
783 struct sock_fprog prog = {
784 .len = (unsigned short)ARRAY_SIZE(filter),
789 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
792 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
794 EXPECT_EQ(EINVAL, errno);
797 #define ERRNO_FILTER(name, errno) \
798 struct sock_filter _read_filter_##name[] = { \
799 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, \
800 offsetof(struct seccomp_data, nr)), \
801 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), \
802 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | errno), \
803 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), \
805 struct sock_fprog prog_##name = { \
806 .len = (unsigned short)ARRAY_SIZE(_read_filter_##name), \
807 .filter = _read_filter_##name, \
810 /* Make sure basic errno values are correctly passed through a filter. */
813 ERRNO_FILTER(valid, E2BIG);
815 pid_t parent = getppid();
817 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
820 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_valid);
823 EXPECT_EQ(parent, syscall(__NR_getppid));
824 EXPECT_EQ(-1, read(0, NULL, 0));
825 EXPECT_EQ(E2BIG, errno);
828 /* Make sure an errno of zero is correctly handled by the arch code. */
831 ERRNO_FILTER(zero, 0);
833 pid_t parent = getppid();
835 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
838 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_zero);
841 EXPECT_EQ(parent, syscall(__NR_getppid));
842 /* "errno" of 0 is ok. */
843 EXPECT_EQ(0, read(0, NULL, 0));
847 * The SECCOMP_RET_DATA mask is 16 bits wide, but errno is smaller.
848 * This tests that the errno value gets capped correctly, fixed by
849 * 580c57f10768 ("seccomp: cap SECCOMP_RET_ERRNO data to MAX_ERRNO").
853 ERRNO_FILTER(capped, 4096);
855 pid_t parent = getppid();
857 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
860 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_capped);
863 EXPECT_EQ(parent, syscall(__NR_getppid));
864 EXPECT_EQ(-1, read(0, NULL, 0));
865 EXPECT_EQ(4095, errno);
869 * Filters are processed in reverse order: last applied is executed first.
870 * Since only the SECCOMP_RET_ACTION mask is tested for return values, the
871 * SECCOMP_RET_DATA mask results will follow the most recently applied
872 * matching filter return (and not the lowest or highest value).
876 ERRNO_FILTER(first, 11);
877 ERRNO_FILTER(second, 13);
878 ERRNO_FILTER(third, 12);
880 pid_t parent = getppid();
882 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
885 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_first);
888 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_second);
891 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_third);
894 EXPECT_EQ(parent, syscall(__NR_getppid));
895 EXPECT_EQ(-1, read(0, NULL, 0));
896 EXPECT_EQ(12, errno);
900 struct sock_fprog prog;
905 struct sock_filter filter[] = {
906 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
907 offsetof(struct seccomp_data, nr)),
908 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
909 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
910 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
913 memset(&self->prog, 0, sizeof(self->prog));
914 self->prog.filter = malloc(sizeof(filter));
915 ASSERT_NE(NULL, self->prog.filter);
916 memcpy(self->prog.filter, filter, sizeof(filter));
917 self->prog.len = (unsigned short)ARRAY_SIZE(filter);
920 FIXTURE_TEARDOWN(TRAP)
922 if (self->prog.filter)
923 free(self->prog.filter);
926 TEST_F_SIGNAL(TRAP, dfl, SIGSYS)
930 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
933 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
935 syscall(__NR_getpid);
938 /* Ensure that SIGSYS overrides SIG_IGN */
939 TEST_F_SIGNAL(TRAP, ign, SIGSYS)
943 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
946 signal(SIGSYS, SIG_IGN);
948 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
950 syscall(__NR_getpid);
953 static siginfo_t TRAP_info;
954 static volatile int TRAP_nr;
955 static void TRAP_action(int nr, siginfo_t *info, void *void_context)
957 memcpy(&TRAP_info, info, sizeof(TRAP_info));
961 TEST_F(TRAP, handler)
964 struct sigaction act;
967 memset(&act, 0, sizeof(act));
969 sigaddset(&mask, SIGSYS);
971 act.sa_sigaction = &TRAP_action;
972 act.sa_flags = SA_SIGINFO;
973 ret = sigaction(SIGSYS, &act, NULL);
975 TH_LOG("sigaction failed");
977 ret = sigprocmask(SIG_UNBLOCK, &mask, NULL);
979 TH_LOG("sigprocmask failed");
982 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
984 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
987 memset(&TRAP_info, 0, sizeof(TRAP_info));
988 /* Expect the registers to be rolled back. (nr = error) may vary
990 ret = syscall(__NR_getpid);
991 /* Silence gcc warning about volatile. */
993 EXPECT_EQ(SIGSYS, test);
994 struct local_sigsys {
995 void *_call_addr; /* calling user insn */
996 int _syscall; /* triggering system call number */
997 unsigned int _arch; /* AUDIT_ARCH_* of syscall */
998 } *sigsys = (struct local_sigsys *)
1000 &(TRAP_info.si_call_addr);
1004 EXPECT_EQ(__NR_getpid, sigsys->_syscall);
1005 /* Make sure arch is non-zero. */
1006 EXPECT_NE(0, sigsys->_arch);
1007 EXPECT_NE(0, (unsigned long)sigsys->_call_addr);
1010 FIXTURE_DATA(precedence) {
1011 struct sock_fprog allow;
1012 struct sock_fprog log;
1013 struct sock_fprog trace;
1014 struct sock_fprog error;
1015 struct sock_fprog trap;
1016 struct sock_fprog kill;
1019 FIXTURE_SETUP(precedence)
1021 struct sock_filter allow_insns[] = {
1022 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1024 struct sock_filter log_insns[] = {
1025 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1026 offsetof(struct seccomp_data, nr)),
1027 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1028 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1029 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG),
1031 struct sock_filter trace_insns[] = {
1032 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1033 offsetof(struct seccomp_data, nr)),
1034 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1035 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1036 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE),
1038 struct sock_filter error_insns[] = {
1039 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1040 offsetof(struct seccomp_data, nr)),
1041 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1042 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1043 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO),
1045 struct sock_filter trap_insns[] = {
1046 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1047 offsetof(struct seccomp_data, nr)),
1048 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1049 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1050 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
1052 struct sock_filter kill_insns[] = {
1053 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1054 offsetof(struct seccomp_data, nr)),
1055 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1056 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1057 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1060 memset(self, 0, sizeof(*self));
1061 #define FILTER_ALLOC(_x) \
1062 self->_x.filter = malloc(sizeof(_x##_insns)); \
1063 ASSERT_NE(NULL, self->_x.filter); \
1064 memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \
1065 self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns)
1066 FILTER_ALLOC(allow);
1068 FILTER_ALLOC(trace);
1069 FILTER_ALLOC(error);
1074 FIXTURE_TEARDOWN(precedence)
1076 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter)
1085 TEST_F(precedence, allow_ok)
1087 pid_t parent, res = 0;
1091 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1094 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1096 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1098 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1100 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1102 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1104 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1106 /* Should work just fine. */
1107 res = syscall(__NR_getppid);
1108 EXPECT_EQ(parent, res);
1111 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS)
1113 pid_t parent, res = 0;
1117 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1120 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1122 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1124 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1126 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1128 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1130 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1132 /* Should work just fine. */
1133 res = syscall(__NR_getppid);
1134 EXPECT_EQ(parent, res);
1135 /* getpid() should never return. */
1136 res = syscall(__NR_getpid);
1140 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS)
1146 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1149 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1151 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1153 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1155 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1157 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1159 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1161 /* Should work just fine. */
1162 EXPECT_EQ(parent, syscall(__NR_getppid));
1163 /* getpid() should never return. */
1164 EXPECT_EQ(0, syscall(__NR_getpid));
1167 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS)
1173 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1176 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1178 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1180 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1182 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1184 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1186 /* Should work just fine. */
1187 EXPECT_EQ(parent, syscall(__NR_getppid));
1188 /* getpid() should never return. */
1189 EXPECT_EQ(0, syscall(__NR_getpid));
1192 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS)
1198 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1201 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1203 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1205 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1207 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1209 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1211 /* Should work just fine. */
1212 EXPECT_EQ(parent, syscall(__NR_getppid));
1213 /* getpid() should never return. */
1214 EXPECT_EQ(0, syscall(__NR_getpid));
1217 TEST_F(precedence, errno_is_third)
1223 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1226 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1228 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1230 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1232 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1234 /* Should work just fine. */
1235 EXPECT_EQ(parent, syscall(__NR_getppid));
1236 EXPECT_EQ(0, syscall(__NR_getpid));
1239 TEST_F(precedence, errno_is_third_in_any_order)
1245 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1248 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1250 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1252 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1254 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1256 /* Should work just fine. */
1257 EXPECT_EQ(parent, syscall(__NR_getppid));
1258 EXPECT_EQ(0, syscall(__NR_getpid));
1261 TEST_F(precedence, trace_is_fourth)
1267 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1270 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1272 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1274 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1276 /* Should work just fine. */
1277 EXPECT_EQ(parent, syscall(__NR_getppid));
1279 EXPECT_EQ(-1, syscall(__NR_getpid));
1282 TEST_F(precedence, trace_is_fourth_in_any_order)
1288 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1291 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1293 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1295 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1297 /* Should work just fine. */
1298 EXPECT_EQ(parent, syscall(__NR_getppid));
1300 EXPECT_EQ(-1, syscall(__NR_getpid));
1303 TEST_F(precedence, log_is_fifth)
1305 pid_t mypid, parent;
1310 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1313 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1315 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1317 /* Should work just fine. */
1318 EXPECT_EQ(parent, syscall(__NR_getppid));
1319 /* Should also work just fine */
1320 EXPECT_EQ(mypid, syscall(__NR_getpid));
1323 TEST_F(precedence, log_is_fifth_in_any_order)
1325 pid_t mypid, parent;
1330 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1333 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1335 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1337 /* Should work just fine. */
1338 EXPECT_EQ(parent, syscall(__NR_getppid));
1339 /* Should also work just fine */
1340 EXPECT_EQ(mypid, syscall(__NR_getpid));
1343 #ifndef PTRACE_O_TRACESECCOMP
1344 #define PTRACE_O_TRACESECCOMP 0x00000080
1347 /* Catch the Ubuntu 12.04 value error. */
1348 #if PTRACE_EVENT_SECCOMP != 7
1349 #undef PTRACE_EVENT_SECCOMP
1352 #ifndef PTRACE_EVENT_SECCOMP
1353 #define PTRACE_EVENT_SECCOMP 7
1356 #define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP)
1357 bool tracer_running;
1358 void tracer_stop(int sig)
1360 tracer_running = false;
1363 typedef void tracer_func_t(struct __test_metadata *_metadata,
1364 pid_t tracee, int status, void *args);
1366 void start_tracer(struct __test_metadata *_metadata, int fd, pid_t tracee,
1367 tracer_func_t tracer_func, void *args, bool ptrace_syscall)
1370 struct sigaction action = {
1371 .sa_handler = tracer_stop,
1374 /* Allow external shutdown. */
1375 tracer_running = true;
1376 ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL));
1379 while (ret == -1 && errno != EINVAL)
1380 ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0);
1382 kill(tracee, SIGKILL);
1384 /* Wait for attach stop */
1387 ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, ptrace_syscall ?
1388 PTRACE_O_TRACESYSGOOD :
1389 PTRACE_O_TRACESECCOMP);
1391 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP");
1392 kill(tracee, SIGKILL);
1394 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1398 /* Unblock the tracee */
1399 ASSERT_EQ(1, write(fd, "A", 1));
1400 ASSERT_EQ(0, close(fd));
1402 /* Run until we're shut down. Must assert to stop execution. */
1403 while (tracer_running) {
1406 if (wait(&status) != tracee)
1408 if (WIFSIGNALED(status) || WIFEXITED(status))
1409 /* Child is dead. Time to go. */
1412 /* Check if this is a seccomp event. */
1413 ASSERT_EQ(!ptrace_syscall, IS_SECCOMP_EVENT(status));
1415 tracer_func(_metadata, tracee, status, args);
1417 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1421 /* Directly report the status of our test harness results. */
1422 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS : EXIT_FAILURE);
1425 /* Common tracer setup/teardown functions. */
1426 void cont_handler(int num)
1428 pid_t setup_trace_fixture(struct __test_metadata *_metadata,
1429 tracer_func_t func, void *args, bool ptrace_syscall)
1434 pid_t tracee = getpid();
1436 /* Setup a pipe for clean synchronization. */
1437 ASSERT_EQ(0, pipe(pipefd));
1439 /* Fork a child which we'll promote to tracer */
1440 tracer_pid = fork();
1441 ASSERT_LE(0, tracer_pid);
1442 signal(SIGALRM, cont_handler);
1443 if (tracer_pid == 0) {
1445 start_tracer(_metadata, pipefd[1], tracee, func, args,
1447 syscall(__NR_exit, 0);
1450 prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0);
1451 read(pipefd[0], &sync, 1);
1456 void teardown_trace_fixture(struct __test_metadata *_metadata,
1462 * Extract the exit code from the other process and
1463 * adopt it for ourselves in case its asserts failed.
1465 ASSERT_EQ(0, kill(tracer, SIGUSR1));
1466 ASSERT_EQ(tracer, waitpid(tracer, &status, 0));
1467 if (WEXITSTATUS(status))
1468 _metadata->passed = 0;
1472 /* "poke" tracer arguments and function. */
1473 struct tracer_args_poke_t {
1474 unsigned long poke_addr;
1477 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status,
1482 struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args;
1484 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1486 /* If this fails, don't try to recover. */
1487 ASSERT_EQ(0x1001, msg) {
1488 kill(tracee, SIGKILL);
1491 * Poke in the message.
1492 * Registers are not touched to try to keep this relatively arch
1495 ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001);
1499 FIXTURE_DATA(TRACE_poke) {
1500 struct sock_fprog prog;
1503 struct tracer_args_poke_t tracer_args;
1506 FIXTURE_SETUP(TRACE_poke)
1508 struct sock_filter filter[] = {
1509 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1510 offsetof(struct seccomp_data, nr)),
1511 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1512 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001),
1513 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1517 memset(&self->prog, 0, sizeof(self->prog));
1518 self->prog.filter = malloc(sizeof(filter));
1519 ASSERT_NE(NULL, self->prog.filter);
1520 memcpy(self->prog.filter, filter, sizeof(filter));
1521 self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1523 /* Set up tracer args. */
1524 self->tracer_args.poke_addr = (unsigned long)&self->poked;
1526 /* Launch tracer. */
1527 self->tracer = setup_trace_fixture(_metadata, tracer_poke,
1528 &self->tracer_args, false);
1531 FIXTURE_TEARDOWN(TRACE_poke)
1533 teardown_trace_fixture(_metadata, self->tracer);
1534 if (self->prog.filter)
1535 free(self->prog.filter);
1538 TEST_F(TRACE_poke, read_has_side_effects)
1542 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1545 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1548 EXPECT_EQ(0, self->poked);
1549 ret = read(-1, NULL, 0);
1551 EXPECT_EQ(0x1001, self->poked);
1554 TEST_F(TRACE_poke, getpid_runs_normally)
1558 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1561 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1564 EXPECT_EQ(0, self->poked);
1565 EXPECT_NE(0, syscall(__NR_getpid));
1566 EXPECT_EQ(0, self->poked);
1569 #if defined(__x86_64__)
1570 # define ARCH_REGS struct user_regs_struct
1571 # define SYSCALL_NUM orig_rax
1572 # define SYSCALL_RET rax
1573 #elif defined(__i386__)
1574 # define ARCH_REGS struct user_regs_struct
1575 # define SYSCALL_NUM orig_eax
1576 # define SYSCALL_RET eax
1577 #elif defined(__arm__)
1578 # define ARCH_REGS struct pt_regs
1579 # define SYSCALL_NUM ARM_r7
1580 # define SYSCALL_RET ARM_r0
1581 #elif defined(__aarch64__)
1582 # define ARCH_REGS struct user_pt_regs
1583 # define SYSCALL_NUM regs[8]
1584 # define SYSCALL_RET regs[0]
1585 #elif defined(__hppa__)
1586 # define ARCH_REGS struct user_regs_struct
1587 # define SYSCALL_NUM gr[20]
1588 # define SYSCALL_RET gr[28]
1589 #elif defined(__powerpc__)
1590 # define ARCH_REGS struct pt_regs
1591 # define SYSCALL_NUM gpr[0]
1592 # define SYSCALL_RET gpr[3]
1593 #elif defined(__s390__)
1594 # define ARCH_REGS s390_regs
1595 # define SYSCALL_NUM gprs[2]
1596 # define SYSCALL_RET gprs[2]
1597 #elif defined(__mips__)
1598 # define ARCH_REGS struct pt_regs
1599 # define SYSCALL_NUM regs[2]
1600 # define SYSCALL_SYSCALL_NUM regs[4]
1601 # define SYSCALL_RET regs[2]
1602 # define SYSCALL_NUM_RET_SHARE_REG
1604 # error "Do not know how to find your architecture's registers and syscalls"
1607 /* When the syscall return can't be changed, stub out the tests for it. */
1608 #ifdef SYSCALL_NUM_RET_SHARE_REG
1609 # define EXPECT_SYSCALL_RETURN(val, action) EXPECT_EQ(-1, action)
1611 # define EXPECT_SYSCALL_RETURN(val, action) EXPECT_EQ(val, action)
1614 /* Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for
1615 * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux).
1617 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__)
1618 #define HAVE_GETREGS
1621 /* Architecture-specific syscall fetching routine. */
1622 int get_syscall(struct __test_metadata *_metadata, pid_t tracee)
1626 EXPECT_EQ(0, ptrace(PTRACE_GETREGS, tracee, 0, ®s)) {
1627 TH_LOG("PTRACE_GETREGS failed");
1633 iov.iov_base = ®s;
1634 iov.iov_len = sizeof(regs);
1635 EXPECT_EQ(0, ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov)) {
1636 TH_LOG("PTRACE_GETREGSET failed");
1641 #if defined(__mips__)
1642 if (regs.SYSCALL_NUM == __NR_O32_Linux)
1643 return regs.SYSCALL_SYSCALL_NUM;
1645 return regs.SYSCALL_NUM;
1648 /* Architecture-specific syscall changing routine. */
1649 void change_syscall(struct __test_metadata *_metadata,
1650 pid_t tracee, int syscall)
1655 ret = ptrace(PTRACE_GETREGS, tracee, 0, ®s);
1658 iov.iov_base = ®s;
1659 iov.iov_len = sizeof(regs);
1660 ret = ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov);
1662 EXPECT_EQ(0, ret) {}
1664 #if defined(__x86_64__) || defined(__i386__) || defined(__powerpc__) || \
1665 defined(__s390__) || defined(__hppa__)
1667 regs.SYSCALL_NUM = syscall;
1669 #elif defined(__mips__)
1671 if (regs.SYSCALL_NUM == __NR_O32_Linux)
1672 regs.SYSCALL_SYSCALL_NUM = syscall;
1674 regs.SYSCALL_NUM = syscall;
1677 #elif defined(__arm__)
1678 # ifndef PTRACE_SET_SYSCALL
1679 # define PTRACE_SET_SYSCALL 23
1682 ret = ptrace(PTRACE_SET_SYSCALL, tracee, NULL, syscall);
1686 #elif defined(__aarch64__)
1687 # ifndef NT_ARM_SYSTEM_CALL
1688 # define NT_ARM_SYSTEM_CALL 0x404
1691 iov.iov_base = &syscall;
1692 iov.iov_len = sizeof(syscall);
1693 ret = ptrace(PTRACE_SETREGSET, tracee, NT_ARM_SYSTEM_CALL,
1700 TH_LOG("How is the syscall changed on this architecture?");
1704 /* If syscall is skipped, change return value. */
1706 #ifdef SYSCALL_NUM_RET_SHARE_REG
1707 TH_LOG("Can't modify syscall return on this architecture");
1709 regs.SYSCALL_RET = EPERM;
1713 ret = ptrace(PTRACE_SETREGS, tracee, 0, ®s);
1715 iov.iov_base = ®s;
1716 iov.iov_len = sizeof(regs);
1717 ret = ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &iov);
1722 void tracer_syscall(struct __test_metadata *_metadata, pid_t tracee,
1723 int status, void *args)
1728 /* Make sure we got the right message. */
1729 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1732 /* Validate and take action on expected syscalls. */
1735 /* change getpid to getppid. */
1736 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee));
1737 change_syscall(_metadata, tracee, __NR_getppid);
1741 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee));
1742 change_syscall(_metadata, tracee, -1);
1745 /* do nothing (allow getppid) */
1746 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee));
1750 TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg);
1751 kill(tracee, SIGKILL);
1757 void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee,
1758 int status, void *args)
1764 /* Make sure we got an empty message. */
1765 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1769 /* The only way to tell PTRACE_SYSCALL entry/exit is by counting. */
1774 nr = get_syscall(_metadata, tracee);
1776 if (nr == __NR_getpid)
1777 change_syscall(_metadata, tracee, __NR_getppid);
1778 if (nr == __NR_openat)
1779 change_syscall(_metadata, tracee, -1);
1782 FIXTURE_DATA(TRACE_syscall) {
1783 struct sock_fprog prog;
1784 pid_t tracer, mytid, mypid, parent;
1787 FIXTURE_SETUP(TRACE_syscall)
1789 struct sock_filter filter[] = {
1790 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1791 offsetof(struct seccomp_data, nr)),
1792 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
1793 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002),
1794 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1),
1795 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003),
1796 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1797 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004),
1798 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1801 memset(&self->prog, 0, sizeof(self->prog));
1802 self->prog.filter = malloc(sizeof(filter));
1803 ASSERT_NE(NULL, self->prog.filter);
1804 memcpy(self->prog.filter, filter, sizeof(filter));
1805 self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1807 /* Prepare some testable syscall results. */
1808 self->mytid = syscall(__NR_gettid);
1809 ASSERT_GT(self->mytid, 0);
1810 ASSERT_NE(self->mytid, 1) {
1811 TH_LOG("Running this test as init is not supported. :)");
1814 self->mypid = getpid();
1815 ASSERT_GT(self->mypid, 0);
1816 ASSERT_EQ(self->mytid, self->mypid);
1818 self->parent = getppid();
1819 ASSERT_GT(self->parent, 0);
1820 ASSERT_NE(self->parent, self->mypid);
1822 /* Launch tracer. */
1823 self->tracer = setup_trace_fixture(_metadata, tracer_syscall, NULL,
1827 FIXTURE_TEARDOWN(TRACE_syscall)
1829 teardown_trace_fixture(_metadata, self->tracer);
1830 if (self->prog.filter)
1831 free(self->prog.filter);
1834 TEST_F(TRACE_syscall, ptrace_syscall_redirected)
1836 /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1837 teardown_trace_fixture(_metadata, self->tracer);
1838 self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1841 /* Tracer will redirect getpid to getppid. */
1842 EXPECT_NE(self->mypid, syscall(__NR_getpid));
1845 TEST_F(TRACE_syscall, ptrace_syscall_dropped)
1847 /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1848 teardown_trace_fixture(_metadata, self->tracer);
1849 self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1852 /* Tracer should skip the open syscall, resulting in EPERM. */
1853 EXPECT_SYSCALL_RETURN(EPERM, syscall(__NR_openat));
1856 TEST_F(TRACE_syscall, syscall_allowed)
1860 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1863 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1866 /* getppid works as expected (no changes). */
1867 EXPECT_EQ(self->parent, syscall(__NR_getppid));
1868 EXPECT_NE(self->mypid, syscall(__NR_getppid));
1871 TEST_F(TRACE_syscall, syscall_redirected)
1875 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1878 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1881 /* getpid has been redirected to getppid as expected. */
1882 EXPECT_EQ(self->parent, syscall(__NR_getpid));
1883 EXPECT_NE(self->mypid, syscall(__NR_getpid));
1886 TEST_F(TRACE_syscall, syscall_dropped)
1890 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1893 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1896 /* gettid has been skipped and an altered return value stored. */
1897 EXPECT_SYSCALL_RETURN(EPERM, syscall(__NR_gettid));
1898 EXPECT_NE(self->mytid, syscall(__NR_gettid));
1901 TEST_F(TRACE_syscall, skip_after_RET_TRACE)
1903 struct sock_filter filter[] = {
1904 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1905 offsetof(struct seccomp_data, nr)),
1906 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1907 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
1908 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1910 struct sock_fprog prog = {
1911 .len = (unsigned short)ARRAY_SIZE(filter),
1916 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1919 /* Install fixture filter. */
1920 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1923 /* Install "errno on getppid" filter. */
1924 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1927 /* Tracer will redirect getpid to getppid, and we should see EPERM. */
1929 EXPECT_EQ(-1, syscall(__NR_getpid));
1930 EXPECT_EQ(EPERM, errno);
1933 TEST_F_SIGNAL(TRACE_syscall, kill_after_RET_TRACE, SIGSYS)
1935 struct sock_filter filter[] = {
1936 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1937 offsetof(struct seccomp_data, nr)),
1938 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1939 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1940 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1942 struct sock_fprog prog = {
1943 .len = (unsigned short)ARRAY_SIZE(filter),
1948 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1951 /* Install fixture filter. */
1952 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1955 /* Install "death on getppid" filter. */
1956 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1959 /* Tracer will redirect getpid to getppid, and we should die. */
1960 EXPECT_NE(self->mypid, syscall(__NR_getpid));
1963 TEST_F(TRACE_syscall, skip_after_ptrace)
1965 struct sock_filter filter[] = {
1966 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1967 offsetof(struct seccomp_data, nr)),
1968 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1969 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
1970 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1972 struct sock_fprog prog = {
1973 .len = (unsigned short)ARRAY_SIZE(filter),
1978 /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1979 teardown_trace_fixture(_metadata, self->tracer);
1980 self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1983 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1986 /* Install "errno on getppid" filter. */
1987 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1990 /* Tracer will redirect getpid to getppid, and we should see EPERM. */
1991 EXPECT_EQ(-1, syscall(__NR_getpid));
1992 EXPECT_EQ(EPERM, errno);
1995 TEST_F_SIGNAL(TRACE_syscall, kill_after_ptrace, SIGSYS)
1997 struct sock_filter filter[] = {
1998 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1999 offsetof(struct seccomp_data, nr)),
2000 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2001 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2002 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2004 struct sock_fprog prog = {
2005 .len = (unsigned short)ARRAY_SIZE(filter),
2010 /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
2011 teardown_trace_fixture(_metadata, self->tracer);
2012 self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
2015 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2018 /* Install "death on getppid" filter. */
2019 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2022 /* Tracer will redirect getpid to getppid, and we should die. */
2023 EXPECT_NE(self->mypid, syscall(__NR_getpid));
2026 TEST(seccomp_syscall)
2028 struct sock_filter filter[] = {
2029 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2031 struct sock_fprog prog = {
2032 .len = (unsigned short)ARRAY_SIZE(filter),
2037 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2039 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2042 /* Reject insane operation. */
2043 ret = seccomp(-1, 0, &prog);
2044 ASSERT_NE(ENOSYS, errno) {
2045 TH_LOG("Kernel does not support seccomp syscall!");
2047 EXPECT_EQ(EINVAL, errno) {
2048 TH_LOG("Did not reject crazy op value!");
2051 /* Reject strict with flags or pointer. */
2052 ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL);
2053 EXPECT_EQ(EINVAL, errno) {
2054 TH_LOG("Did not reject mode strict with flags!");
2056 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog);
2057 EXPECT_EQ(EINVAL, errno) {
2058 TH_LOG("Did not reject mode strict with uargs!");
2061 /* Reject insane args for filter. */
2062 ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog);
2063 EXPECT_EQ(EINVAL, errno) {
2064 TH_LOG("Did not reject crazy filter flags!");
2066 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL);
2067 EXPECT_EQ(EFAULT, errno) {
2068 TH_LOG("Did not reject NULL filter!");
2071 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2072 EXPECT_EQ(0, errno) {
2073 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s",
2078 TEST(seccomp_syscall_mode_lock)
2080 struct sock_filter filter[] = {
2081 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2083 struct sock_fprog prog = {
2084 .len = (unsigned short)ARRAY_SIZE(filter),
2089 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2091 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2094 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2095 ASSERT_NE(ENOSYS, errno) {
2096 TH_LOG("Kernel does not support seccomp syscall!");
2099 TH_LOG("Could not install filter!");
2102 /* Make sure neither entry point will switch to strict. */
2103 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0);
2104 EXPECT_EQ(EINVAL, errno) {
2105 TH_LOG("Switched to mode strict!");
2108 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL);
2109 EXPECT_EQ(EINVAL, errno) {
2110 TH_LOG("Switched to mode strict!");
2115 * Test detection of known and unknown filter flags. Userspace needs to be able
2116 * to check if a filter flag is supported by the current kernel and a good way
2117 * of doing that is by attempting to enter filter mode, with the flag bit in
2118 * question set, and a NULL pointer for the _args_ parameter. EFAULT indicates
2119 * that the flag is valid and EINVAL indicates that the flag is invalid.
2121 TEST(detect_seccomp_filter_flags)
2123 unsigned int flags[] = { SECCOMP_FILTER_FLAG_TSYNC,
2124 SECCOMP_FILTER_FLAG_LOG,
2125 SECCOMP_FILTER_FLAG_SPEC_ALLOW,
2126 SECCOMP_FILTER_FLAG_NEW_LISTENER };
2127 unsigned int flag, all_flags;
2131 /* Test detection of known-good filter flags */
2132 for (i = 0, all_flags = 0; i < ARRAY_SIZE(flags); i++) {
2136 /* Make sure the flag is a single bit! */
2145 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2146 ASSERT_NE(ENOSYS, errno) {
2147 TH_LOG("Kernel does not support seccomp syscall!");
2150 EXPECT_EQ(EFAULT, errno) {
2151 TH_LOG("Failed to detect that a known-good filter flag (0x%X) is supported!",
2158 /* Test detection of all known-good filter flags */
2159 ret = seccomp(SECCOMP_SET_MODE_FILTER, all_flags, NULL);
2161 EXPECT_EQ(EFAULT, errno) {
2162 TH_LOG("Failed to detect that all known-good filter flags (0x%X) are supported!",
2166 /* Test detection of an unknown filter flag */
2168 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2170 EXPECT_EQ(EINVAL, errno) {
2171 TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported!",
2176 * Test detection of an unknown filter flag that may simply need to be
2177 * added to this test
2179 flag = flags[ARRAY_SIZE(flags) - 1] << 1;
2180 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2182 EXPECT_EQ(EINVAL, errno) {
2183 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?",
2190 struct sock_filter filter[] = {
2191 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2193 struct sock_fprog prog = {
2194 .len = (unsigned short)ARRAY_SIZE(filter),
2199 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2201 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2204 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2206 ASSERT_NE(ENOSYS, errno) {
2207 TH_LOG("Kernel does not support seccomp syscall!");
2210 TH_LOG("Could not install initial filter with TSYNC!");
2214 #define TSYNC_SIBLINGS 2
2215 struct tsync_sibling {
2219 pthread_cond_t *cond;
2220 pthread_mutex_t *mutex;
2223 struct sock_fprog *prog;
2224 struct __test_metadata *metadata;
2228 * To avoid joining joined threads (which is not allowed by Bionic),
2229 * make sure we both successfully join and clear the tid to skip a
2230 * later join attempt during fixture teardown. Any remaining threads
2231 * will be directly killed during teardown.
2233 #define PTHREAD_JOIN(tid, status) \
2235 int _rc = pthread_join(tid, status); \
2237 TH_LOG("pthread_join of tid %u failed: %d\n", \
2238 (unsigned int)tid, _rc); \
2244 FIXTURE_DATA(TSYNC) {
2245 struct sock_fprog root_prog, apply_prog;
2246 struct tsync_sibling sibling[TSYNC_SIBLINGS];
2248 pthread_cond_t cond;
2249 pthread_mutex_t mutex;
2253 FIXTURE_SETUP(TSYNC)
2255 struct sock_filter root_filter[] = {
2256 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2258 struct sock_filter apply_filter[] = {
2259 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2260 offsetof(struct seccomp_data, nr)),
2261 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
2262 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2263 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2266 memset(&self->root_prog, 0, sizeof(self->root_prog));
2267 memset(&self->apply_prog, 0, sizeof(self->apply_prog));
2268 memset(&self->sibling, 0, sizeof(self->sibling));
2269 self->root_prog.filter = malloc(sizeof(root_filter));
2270 ASSERT_NE(NULL, self->root_prog.filter);
2271 memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter));
2272 self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter);
2274 self->apply_prog.filter = malloc(sizeof(apply_filter));
2275 ASSERT_NE(NULL, self->apply_prog.filter);
2276 memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter));
2277 self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter);
2279 self->sibling_count = 0;
2280 pthread_mutex_init(&self->mutex, NULL);
2281 pthread_cond_init(&self->cond, NULL);
2282 sem_init(&self->started, 0, 0);
2283 self->sibling[0].tid = 0;
2284 self->sibling[0].cond = &self->cond;
2285 self->sibling[0].started = &self->started;
2286 self->sibling[0].mutex = &self->mutex;
2287 self->sibling[0].diverge = 0;
2288 self->sibling[0].num_waits = 1;
2289 self->sibling[0].prog = &self->root_prog;
2290 self->sibling[0].metadata = _metadata;
2291 self->sibling[1].tid = 0;
2292 self->sibling[1].cond = &self->cond;
2293 self->sibling[1].started = &self->started;
2294 self->sibling[1].mutex = &self->mutex;
2295 self->sibling[1].diverge = 0;
2296 self->sibling[1].prog = &self->root_prog;
2297 self->sibling[1].num_waits = 1;
2298 self->sibling[1].metadata = _metadata;
2301 FIXTURE_TEARDOWN(TSYNC)
2305 if (self->root_prog.filter)
2306 free(self->root_prog.filter);
2307 if (self->apply_prog.filter)
2308 free(self->apply_prog.filter);
2310 for ( ; sib < self->sibling_count; ++sib) {
2311 struct tsync_sibling *s = &self->sibling[sib];
2316 * If a thread is still running, it may be stuck, so hit
2317 * it over the head really hard.
2319 pthread_kill(s->tid, 9);
2321 pthread_mutex_destroy(&self->mutex);
2322 pthread_cond_destroy(&self->cond);
2323 sem_destroy(&self->started);
2326 void *tsync_sibling(void *data)
2329 struct tsync_sibling *me = data;
2331 me->system_tid = syscall(__NR_gettid);
2333 pthread_mutex_lock(me->mutex);
2335 /* Just re-apply the root prog to fork the tree */
2336 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
2339 sem_post(me->started);
2340 /* Return outside of started so parent notices failures. */
2342 pthread_mutex_unlock(me->mutex);
2343 return (void *)SIBLING_EXIT_FAILURE;
2346 pthread_cond_wait(me->cond, me->mutex);
2347 me->num_waits = me->num_waits - 1;
2348 } while (me->num_waits);
2349 pthread_mutex_unlock(me->mutex);
2351 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0);
2353 return (void *)SIBLING_EXIT_NEWPRIVS;
2355 return (void *)SIBLING_EXIT_UNKILLED;
2358 void tsync_start_sibling(struct tsync_sibling *sibling)
2360 pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling);
2363 TEST_F(TSYNC, siblings_fail_prctl)
2367 struct sock_filter filter[] = {
2368 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2369 offsetof(struct seccomp_data, nr)),
2370 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
2371 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL),
2372 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2374 struct sock_fprog prog = {
2375 .len = (unsigned short)ARRAY_SIZE(filter),
2379 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2380 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2383 /* Check prctl failure detection by requesting sib 0 diverge. */
2384 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2385 ASSERT_NE(ENOSYS, errno) {
2386 TH_LOG("Kernel does not support seccomp syscall!");
2389 TH_LOG("setting filter failed");
2392 self->sibling[0].diverge = 1;
2393 tsync_start_sibling(&self->sibling[0]);
2394 tsync_start_sibling(&self->sibling[1]);
2396 while (self->sibling_count < TSYNC_SIBLINGS) {
2397 sem_wait(&self->started);
2398 self->sibling_count++;
2401 /* Signal the threads to clean up*/
2402 pthread_mutex_lock(&self->mutex);
2403 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2404 TH_LOG("cond broadcast non-zero");
2406 pthread_mutex_unlock(&self->mutex);
2408 /* Ensure diverging sibling failed to call prctl. */
2409 PTHREAD_JOIN(self->sibling[0].tid, &status);
2410 EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status);
2411 PTHREAD_JOIN(self->sibling[1].tid, &status);
2412 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2415 TEST_F(TSYNC, two_siblings_with_ancestor)
2420 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2421 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2424 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2425 ASSERT_NE(ENOSYS, errno) {
2426 TH_LOG("Kernel does not support seccomp syscall!");
2429 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2431 tsync_start_sibling(&self->sibling[0]);
2432 tsync_start_sibling(&self->sibling[1]);
2434 while (self->sibling_count < TSYNC_SIBLINGS) {
2435 sem_wait(&self->started);
2436 self->sibling_count++;
2439 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2442 TH_LOG("Could install filter on all threads!");
2444 /* Tell the siblings to test the policy */
2445 pthread_mutex_lock(&self->mutex);
2446 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2447 TH_LOG("cond broadcast non-zero");
2449 pthread_mutex_unlock(&self->mutex);
2450 /* Ensure they are both killed and don't exit cleanly. */
2451 PTHREAD_JOIN(self->sibling[0].tid, &status);
2452 EXPECT_EQ(0x0, (long)status);
2453 PTHREAD_JOIN(self->sibling[1].tid, &status);
2454 EXPECT_EQ(0x0, (long)status);
2457 TEST_F(TSYNC, two_sibling_want_nnp)
2461 /* start siblings before any prctl() operations */
2462 tsync_start_sibling(&self->sibling[0]);
2463 tsync_start_sibling(&self->sibling[1]);
2464 while (self->sibling_count < TSYNC_SIBLINGS) {
2465 sem_wait(&self->started);
2466 self->sibling_count++;
2469 /* Tell the siblings to test no policy */
2470 pthread_mutex_lock(&self->mutex);
2471 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2472 TH_LOG("cond broadcast non-zero");
2474 pthread_mutex_unlock(&self->mutex);
2476 /* Ensure they are both upset about lacking nnp. */
2477 PTHREAD_JOIN(self->sibling[0].tid, &status);
2478 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2479 PTHREAD_JOIN(self->sibling[1].tid, &status);
2480 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2483 TEST_F(TSYNC, two_siblings_with_no_filter)
2488 /* start siblings before any prctl() operations */
2489 tsync_start_sibling(&self->sibling[0]);
2490 tsync_start_sibling(&self->sibling[1]);
2491 while (self->sibling_count < TSYNC_SIBLINGS) {
2492 sem_wait(&self->started);
2493 self->sibling_count++;
2496 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2497 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2500 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2502 ASSERT_NE(ENOSYS, errno) {
2503 TH_LOG("Kernel does not support seccomp syscall!");
2506 TH_LOG("Could install filter on all threads!");
2509 /* Tell the siblings to test the policy */
2510 pthread_mutex_lock(&self->mutex);
2511 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2512 TH_LOG("cond broadcast non-zero");
2514 pthread_mutex_unlock(&self->mutex);
2516 /* Ensure they are both killed and don't exit cleanly. */
2517 PTHREAD_JOIN(self->sibling[0].tid, &status);
2518 EXPECT_EQ(0x0, (long)status);
2519 PTHREAD_JOIN(self->sibling[1].tid, &status);
2520 EXPECT_EQ(0x0, (long)status);
2523 TEST_F(TSYNC, two_siblings_with_one_divergence)
2528 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2529 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2532 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2533 ASSERT_NE(ENOSYS, errno) {
2534 TH_LOG("Kernel does not support seccomp syscall!");
2537 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2539 self->sibling[0].diverge = 1;
2540 tsync_start_sibling(&self->sibling[0]);
2541 tsync_start_sibling(&self->sibling[1]);
2543 while (self->sibling_count < TSYNC_SIBLINGS) {
2544 sem_wait(&self->started);
2545 self->sibling_count++;
2548 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2550 ASSERT_EQ(self->sibling[0].system_tid, ret) {
2551 TH_LOG("Did not fail on diverged sibling.");
2554 /* Wake the threads */
2555 pthread_mutex_lock(&self->mutex);
2556 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2557 TH_LOG("cond broadcast non-zero");
2559 pthread_mutex_unlock(&self->mutex);
2561 /* Ensure they are both unkilled. */
2562 PTHREAD_JOIN(self->sibling[0].tid, &status);
2563 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2564 PTHREAD_JOIN(self->sibling[1].tid, &status);
2565 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2568 TEST_F(TSYNC, two_siblings_not_under_filter)
2573 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2574 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2578 * Sibling 0 will have its own seccomp policy
2579 * and Sibling 1 will not be under seccomp at
2580 * all. Sibling 1 will enter seccomp and 0
2581 * will cause failure.
2583 self->sibling[0].diverge = 1;
2584 tsync_start_sibling(&self->sibling[0]);
2585 tsync_start_sibling(&self->sibling[1]);
2587 while (self->sibling_count < TSYNC_SIBLINGS) {
2588 sem_wait(&self->started);
2589 self->sibling_count++;
2592 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2593 ASSERT_NE(ENOSYS, errno) {
2594 TH_LOG("Kernel does not support seccomp syscall!");
2597 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2600 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2602 ASSERT_EQ(ret, self->sibling[0].system_tid) {
2603 TH_LOG("Did not fail on diverged sibling.");
2606 if (ret == self->sibling[0].system_tid)
2609 pthread_mutex_lock(&self->mutex);
2611 /* Increment the other siblings num_waits so we can clean up
2612 * the one we just saw.
2614 self->sibling[!sib].num_waits += 1;
2616 /* Signal the thread to clean up*/
2617 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2618 TH_LOG("cond broadcast non-zero");
2620 pthread_mutex_unlock(&self->mutex);
2621 PTHREAD_JOIN(self->sibling[sib].tid, &status);
2622 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2623 /* Poll for actual task death. pthread_join doesn't guarantee it. */
2624 while (!kill(self->sibling[sib].system_tid, 0))
2626 /* Switch to the remaining sibling */
2629 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2632 TH_LOG("Expected the remaining sibling to sync");
2635 pthread_mutex_lock(&self->mutex);
2637 /* If remaining sibling didn't have a chance to wake up during
2638 * the first broadcast, manually reduce the num_waits now.
2640 if (self->sibling[sib].num_waits > 1)
2641 self->sibling[sib].num_waits = 1;
2642 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2643 TH_LOG("cond broadcast non-zero");
2645 pthread_mutex_unlock(&self->mutex);
2646 PTHREAD_JOIN(self->sibling[sib].tid, &status);
2647 EXPECT_EQ(0, (long)status);
2648 /* Poll for actual task death. pthread_join doesn't guarantee it. */
2649 while (!kill(self->sibling[sib].system_tid, 0))
2652 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2654 ASSERT_EQ(0, ret); /* just us chickens */
2657 /* Make sure restarted syscalls are seen directly as "restart_syscall". */
2658 TEST(syscall_restart)
2665 siginfo_t info = { };
2666 struct sock_filter filter[] = {
2667 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2668 offsetof(struct seccomp_data, nr)),
2670 #ifdef __NR_sigreturn
2671 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 6, 0),
2673 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 5, 0),
2674 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 4, 0),
2675 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 3, 0),
2676 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 4, 0),
2677 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0),
2679 /* Allow __NR_write for easy logging. */
2680 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1),
2681 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2682 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2683 /* The nanosleep jump target. */
2684 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100),
2685 /* The restart_syscall jump target. */
2686 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200),
2688 struct sock_fprog prog = {
2689 .len = (unsigned short)ARRAY_SIZE(filter),
2692 #if defined(__arm__)
2693 struct utsname utsbuf;
2696 ASSERT_EQ(0, pipe(pipefd));
2699 ASSERT_LE(0, child_pid);
2700 if (child_pid == 0) {
2701 /* Child uses EXPECT not ASSERT to deliver status correctly. */
2703 struct timespec timeout = { };
2705 /* Attach parent as tracer and stop. */
2706 EXPECT_EQ(0, ptrace(PTRACE_TRACEME));
2707 EXPECT_EQ(0, raise(SIGSTOP));
2709 EXPECT_EQ(0, close(pipefd[1]));
2711 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2712 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2715 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2717 TH_LOG("Failed to install filter!");
2720 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2721 TH_LOG("Failed to read() sync from parent");
2723 EXPECT_EQ('.', buf) {
2724 TH_LOG("Failed to get sync data from read()");
2727 /* Start nanosleep to be interrupted. */
2730 EXPECT_EQ(0, nanosleep(&timeout, NULL)) {
2731 TH_LOG("Call to nanosleep() failed (errno %d)", errno);
2734 /* Read final sync from parent. */
2735 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2736 TH_LOG("Failed final read() from parent");
2738 EXPECT_EQ('!', buf) {
2739 TH_LOG("Failed to get final data from read()");
2742 /* Directly report the status of our test harness results. */
2743 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS
2746 EXPECT_EQ(0, close(pipefd[0]));
2748 /* Attach to child, setup options, and release. */
2749 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2750 ASSERT_EQ(true, WIFSTOPPED(status));
2751 ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL,
2752 PTRACE_O_TRACESECCOMP));
2753 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2754 ASSERT_EQ(1, write(pipefd[1], ".", 1));
2756 /* Wait for nanosleep() to start. */
2757 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2758 ASSERT_EQ(true, WIFSTOPPED(status));
2759 ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2760 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2761 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2762 ASSERT_EQ(0x100, msg);
2763 EXPECT_EQ(__NR_nanosleep, get_syscall(_metadata, child_pid));
2765 /* Might as well check siginfo for sanity while we're here. */
2766 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2767 ASSERT_EQ(SIGTRAP, info.si_signo);
2768 ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code);
2769 EXPECT_EQ(0, info.si_errno);
2770 EXPECT_EQ(getuid(), info.si_uid);
2771 /* Verify signal delivery came from child (seccomp-triggered). */
2772 EXPECT_EQ(child_pid, info.si_pid);
2774 /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */
2775 ASSERT_EQ(0, kill(child_pid, SIGSTOP));
2776 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2777 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2778 ASSERT_EQ(true, WIFSTOPPED(status));
2779 ASSERT_EQ(SIGSTOP, WSTOPSIG(status));
2780 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2782 * There is no siginfo on SIGSTOP any more, so we can't verify
2783 * signal delivery came from parent now (getpid() == info.si_pid).
2784 * https://lkml.kernel.org/r/CAGXu5jJaZAOzP1qFz66tYrtbuywqb+UN2SOA1VLHpCCOiYvYeg@mail.gmail.com
2785 * At least verify the SIGSTOP via PTRACE_GETSIGINFO.
2787 EXPECT_EQ(SIGSTOP, info.si_signo);
2789 /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */
2790 ASSERT_EQ(0, kill(child_pid, SIGCONT));
2791 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2792 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2793 ASSERT_EQ(true, WIFSTOPPED(status));
2794 ASSERT_EQ(SIGCONT, WSTOPSIG(status));
2795 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2797 /* Wait for restart_syscall() to start. */
2798 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2799 ASSERT_EQ(true, WIFSTOPPED(status));
2800 ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2801 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2802 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2804 ASSERT_EQ(0x200, msg);
2805 ret = get_syscall(_metadata, child_pid);
2806 #if defined(__arm__)
2809 * - native ARM registers do NOT expose true syscall.
2810 * - compat ARM registers on ARM64 DO expose true syscall.
2812 ASSERT_EQ(0, uname(&utsbuf));
2813 if (strncmp(utsbuf.machine, "arm", 3) == 0) {
2814 EXPECT_EQ(__NR_nanosleep, ret);
2818 EXPECT_EQ(__NR_restart_syscall, ret);
2821 /* Write again to end test. */
2822 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2823 ASSERT_EQ(1, write(pipefd[1], "!", 1));
2824 EXPECT_EQ(0, close(pipefd[1]));
2826 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2827 if (WIFSIGNALED(status) || WEXITSTATUS(status))
2828 _metadata->passed = 0;
2831 TEST_SIGNAL(filter_flag_log, SIGSYS)
2833 struct sock_filter allow_filter[] = {
2834 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2836 struct sock_filter kill_filter[] = {
2837 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2838 offsetof(struct seccomp_data, nr)),
2839 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
2840 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2841 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2843 struct sock_fprog allow_prog = {
2844 .len = (unsigned short)ARRAY_SIZE(allow_filter),
2845 .filter = allow_filter,
2847 struct sock_fprog kill_prog = {
2848 .len = (unsigned short)ARRAY_SIZE(kill_filter),
2849 .filter = kill_filter,
2852 pid_t parent = getppid();
2854 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2857 /* Verify that the FILTER_FLAG_LOG flag isn't accepted in strict mode */
2858 ret = seccomp(SECCOMP_SET_MODE_STRICT, SECCOMP_FILTER_FLAG_LOG,
2860 ASSERT_NE(ENOSYS, errno) {
2861 TH_LOG("Kernel does not support seccomp syscall!");
2864 TH_LOG("Kernel accepted FILTER_FLAG_LOG flag in strict mode!");
2866 EXPECT_EQ(EINVAL, errno) {
2867 TH_LOG("Kernel returned unexpected errno for FILTER_FLAG_LOG flag in strict mode!");
2870 /* Verify that a simple, permissive filter can be added with no flags */
2871 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog);
2874 /* See if the same filter can be added with the FILTER_FLAG_LOG flag */
2875 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
2877 ASSERT_NE(EINVAL, errno) {
2878 TH_LOG("Kernel does not support the FILTER_FLAG_LOG flag!");
2882 /* Ensure that the kill filter works with the FILTER_FLAG_LOG flag */
2883 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
2887 EXPECT_EQ(parent, syscall(__NR_getppid));
2888 /* getpid() should never return. */
2889 EXPECT_EQ(0, syscall(__NR_getpid));
2892 TEST(get_action_avail)
2894 __u32 actions[] = { SECCOMP_RET_KILL_THREAD, SECCOMP_RET_TRAP,
2895 SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE,
2896 SECCOMP_RET_LOG, SECCOMP_RET_ALLOW };
2897 __u32 unknown_action = 0x10000000U;
2901 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[0]);
2902 ASSERT_NE(ENOSYS, errno) {
2903 TH_LOG("Kernel does not support seccomp syscall!");
2905 ASSERT_NE(EINVAL, errno) {
2906 TH_LOG("Kernel does not support SECCOMP_GET_ACTION_AVAIL operation!");
2910 for (i = 0; i < ARRAY_SIZE(actions); i++) {
2911 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[i]);
2913 TH_LOG("Expected action (0x%X) not available!",
2918 /* Check that an unknown action is handled properly (EOPNOTSUPP) */
2919 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &unknown_action);
2921 EXPECT_EQ(errno, EOPNOTSUPP);
2929 struct seccomp_metadata md;
2932 ASSERT_EQ(0, pipe(pipefd));
2937 struct sock_filter filter[] = {
2938 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2940 struct sock_fprog prog = {
2941 .len = (unsigned short)ARRAY_SIZE(filter),
2945 /* one with log, one without */
2946 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER,
2947 SECCOMP_FILTER_FLAG_LOG, &prog));
2948 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog));
2950 ASSERT_EQ(0, close(pipefd[0]));
2951 ASSERT_EQ(1, write(pipefd[1], "1", 1));
2952 ASSERT_EQ(0, close(pipefd[1]));
2958 ASSERT_EQ(0, close(pipefd[1]));
2959 ASSERT_EQ(1, read(pipefd[0], &buf, 1));
2961 ASSERT_EQ(0, ptrace(PTRACE_ATTACH, pid));
2962 ASSERT_EQ(pid, waitpid(pid, NULL, 0));
2964 /* Past here must not use ASSERT or child process is never killed. */
2968 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md);
2969 EXPECT_EQ(sizeof(md), ret) {
2970 if (errno == EINVAL)
2971 XFAIL(goto skip, "Kernel does not support PTRACE_SECCOMP_GET_METADATA (missing CONFIG_CHECKPOINT_RESTORE?)");
2974 EXPECT_EQ(md.flags, SECCOMP_FILTER_FLAG_LOG);
2975 EXPECT_EQ(md.filter_off, 0);
2978 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md);
2979 EXPECT_EQ(sizeof(md), ret);
2980 EXPECT_EQ(md.flags, 0);
2981 EXPECT_EQ(md.filter_off, 1);
2984 ASSERT_EQ(0, kill(pid, SIGKILL));
2987 static int user_trap_syscall(int nr, unsigned int flags)
2989 struct sock_filter filter[] = {
2990 BPF_STMT(BPF_LD+BPF_W+BPF_ABS,
2991 offsetof(struct seccomp_data, nr)),
2992 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, nr, 0, 1),
2993 BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_USER_NOTIF),
2994 BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW),
2997 struct sock_fprog prog = {
2998 .len = (unsigned short)ARRAY_SIZE(filter),
3002 return seccomp(SECCOMP_SET_MODE_FILTER, flags, &prog);
3005 #define USER_NOTIF_MAGIC 116983961184613L
3006 TEST(user_notification_basic)
3010 int status, listener;
3011 struct seccomp_notif req = {};
3012 struct seccomp_notif_resp resp = {};
3013 struct pollfd pollfd;
3015 struct sock_filter filter[] = {
3016 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3018 struct sock_fprog prog = {
3019 .len = (unsigned short)ARRAY_SIZE(filter),
3026 /* Check that we get -ENOSYS with no listener attached */
3028 if (user_trap_syscall(__NR_getpid, 0) < 0)
3030 ret = syscall(__NR_getpid);
3031 exit(ret >= 0 || errno != ENOSYS);
3034 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3035 EXPECT_EQ(true, WIFEXITED(status));
3036 EXPECT_EQ(0, WEXITSTATUS(status));
3038 /* Add some no-op filters so for grins. */
3039 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3040 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3041 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3042 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3044 /* Check that the basic notification machinery works */
3045 listener = user_trap_syscall(__NR_getpid,
3046 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3047 EXPECT_GE(listener, 0);
3049 /* Installing a second listener in the chain should EBUSY */
3050 EXPECT_EQ(user_trap_syscall(__NR_getpid,
3051 SECCOMP_FILTER_FLAG_NEW_LISTENER),
3053 EXPECT_EQ(errno, EBUSY);
3059 ret = syscall(__NR_getpid);
3060 exit(ret != USER_NOTIF_MAGIC);
3063 pollfd.fd = listener;
3064 pollfd.events = POLLIN | POLLOUT;
3066 EXPECT_GT(poll(&pollfd, 1, -1), 0);
3067 EXPECT_EQ(pollfd.revents, POLLIN);
3069 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3071 pollfd.fd = listener;
3072 pollfd.events = POLLIN | POLLOUT;
3074 EXPECT_GT(poll(&pollfd, 1, -1), 0);
3075 EXPECT_EQ(pollfd.revents, POLLOUT);
3077 EXPECT_EQ(req.data.nr, __NR_getpid);
3081 resp.val = USER_NOTIF_MAGIC;
3083 /* check that we make sure flags == 0 */
3085 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3086 EXPECT_EQ(errno, EINVAL);
3089 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3091 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3092 EXPECT_EQ(true, WIFEXITED(status));
3093 EXPECT_EQ(0, WEXITSTATUS(status));
3096 TEST(user_notification_kill_in_middle)
3101 struct seccomp_notif req = {};
3102 struct seccomp_notif_resp resp = {};
3104 listener = user_trap_syscall(__NR_getpid,
3105 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3106 EXPECT_GE(listener, 0);
3109 * Check that nothing bad happens when we kill the task in the middle
3116 ret = syscall(__NR_getpid);
3117 exit(ret != USER_NOTIF_MAGIC);
3120 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3121 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), 0);
3123 EXPECT_EQ(kill(pid, SIGKILL), 0);
3124 EXPECT_EQ(waitpid(pid, NULL, 0), pid);
3126 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), -1);
3129 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp);
3131 EXPECT_EQ(errno, ENOENT);
3134 static int handled = -1;
3136 static void signal_handler(int signal)
3138 if (write(handled, "c", 1) != 1)
3139 perror("write from signal");
3142 TEST(user_notification_signal)
3146 int status, listener, sk_pair[2];
3147 struct seccomp_notif req = {};
3148 struct seccomp_notif_resp resp = {};
3151 ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0);
3153 listener = user_trap_syscall(__NR_gettid,
3154 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3155 EXPECT_GE(listener, 0);
3162 handled = sk_pair[1];
3163 if (signal(SIGUSR1, signal_handler) == SIG_ERR) {
3168 * ERESTARTSYS behavior is a bit hard to test, because we need
3169 * to rely on a signal that has not yet been handled. Let's at
3170 * least check that the error code gets propagated through, and
3171 * hope that it doesn't break when there is actually a signal :)
3173 ret = syscall(__NR_gettid);
3174 exit(!(ret == -1 && errno == 512));
3179 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3181 EXPECT_EQ(kill(pid, SIGUSR1), 0);
3184 * Make sure the signal really is delivered, which means we're not
3185 * stuck in the user notification code any more and the notification
3188 EXPECT_EQ(read(sk_pair[0], &c, 1), 1);
3191 resp.error = -EPERM;
3194 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3195 EXPECT_EQ(errno, ENOENT);
3197 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3200 resp.error = -512; /* -ERESTARTSYS */
3203 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3205 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3206 EXPECT_EQ(true, WIFEXITED(status));
3207 EXPECT_EQ(0, WEXITSTATUS(status));
3210 TEST(user_notification_closed_listener)
3214 int status, listener;
3216 listener = user_trap_syscall(__NR_getpid,
3217 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3218 EXPECT_GE(listener, 0);
3221 * Check that we get an ENOSYS when the listener is closed.
3227 ret = syscall(__NR_getpid);
3228 exit(ret != -1 && errno != ENOSYS);
3233 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3234 EXPECT_EQ(true, WIFEXITED(status));
3235 EXPECT_EQ(0, WEXITSTATUS(status));
3239 * Check that a pid in a child namespace still shows up as valid in ours.
3241 TEST(user_notification_child_pid_ns)
3244 int status, listener;
3245 struct seccomp_notif req = {};
3246 struct seccomp_notif_resp resp = {};
3248 ASSERT_EQ(unshare(CLONE_NEWPID), 0);
3250 listener = user_trap_syscall(__NR_getpid, SECCOMP_FILTER_FLAG_NEW_LISTENER);
3251 ASSERT_GE(listener, 0);
3257 exit(syscall(__NR_getpid) != USER_NOTIF_MAGIC);
3259 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3260 EXPECT_EQ(req.pid, pid);
3264 resp.val = USER_NOTIF_MAGIC;
3266 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3268 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3269 EXPECT_EQ(true, WIFEXITED(status));
3270 EXPECT_EQ(0, WEXITSTATUS(status));
3275 * Check that a pid in a sibling (i.e. unrelated) namespace shows up as 0, i.e.
3278 TEST(user_notification_sibling_pid_ns)
3281 int status, listener;
3282 struct seccomp_notif req = {};
3283 struct seccomp_notif_resp resp = {};
3285 listener = user_trap_syscall(__NR_getpid, SECCOMP_FILTER_FLAG_NEW_LISTENER);
3286 ASSERT_GE(listener, 0);
3292 ASSERT_EQ(unshare(CLONE_NEWPID), 0);
3298 exit(syscall(__NR_getpid) != USER_NOTIF_MAGIC);
3300 EXPECT_EQ(waitpid(pid2, &status, 0), pid2);
3301 EXPECT_EQ(true, WIFEXITED(status));
3302 EXPECT_EQ(0, WEXITSTATUS(status));
3303 exit(WEXITSTATUS(status));
3306 /* Create the sibling ns, and sibling in it. */
3307 EXPECT_EQ(unshare(CLONE_NEWPID), 0);
3308 EXPECT_EQ(errno, 0);
3314 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3316 * The pid should be 0, i.e. the task is in some namespace that
3319 ASSERT_EQ(req.pid, 0);
3323 resp.val = USER_NOTIF_MAGIC;
3325 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3331 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3332 EXPECT_EQ(true, WIFEXITED(status));
3333 EXPECT_EQ(0, WEXITSTATUS(status));
3335 EXPECT_EQ(waitpid(pid2, &status, 0), pid2);
3336 EXPECT_EQ(true, WIFEXITED(status));
3337 EXPECT_EQ(0, WEXITSTATUS(status));
3340 TEST(user_notification_fault_recv)
3343 int status, listener;
3344 struct seccomp_notif req = {};
3345 struct seccomp_notif_resp resp = {};
3347 listener = user_trap_syscall(__NR_getpid, SECCOMP_FILTER_FLAG_NEW_LISTENER);
3348 ASSERT_GE(listener, 0);
3354 exit(syscall(__NR_getpid) != USER_NOTIF_MAGIC);
3356 /* Do a bad recv() */
3357 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, NULL), -1);
3358 EXPECT_EQ(errno, EFAULT);
3360 /* We should still be able to receive this notification, though. */
3361 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3362 EXPECT_EQ(req.pid, pid);
3366 resp.val = USER_NOTIF_MAGIC;
3368 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3370 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3371 EXPECT_EQ(true, WIFEXITED(status));
3372 EXPECT_EQ(0, WEXITSTATUS(status));
3375 TEST(seccomp_get_notif_sizes)
3377 struct seccomp_notif_sizes sizes;
3379 EXPECT_EQ(seccomp(SECCOMP_GET_NOTIF_SIZES, 0, &sizes), 0);
3380 EXPECT_EQ(sizes.seccomp_notif, sizeof(struct seccomp_notif));
3381 EXPECT_EQ(sizes.seccomp_notif_resp, sizeof(struct seccomp_notif_resp));
3386 * - add microbenchmarks
3387 * - expand NNP testing
3388 * - better arch-specific TRACE and TRAP handlers.
3389 * - endianness checking when appropriate
3390 * - 64-bit arg prodding
3391 * - arch value testing (x86 modes especially)
3392 * - verify that FILTER_FLAG_LOG filters generate log messages
3393 * - verify that RET_LOG generates log messages