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.
8 #include <asm/siginfo.h>
9 #define __have_siginfo_t 1
10 #define __have_sigval_t 1
11 #define __have_sigevent_t 1
14 #include <linux/filter.h>
15 #include <sys/prctl.h>
16 #include <sys/ptrace.h>
17 #include <sys/types.h>
19 #include <linux/prctl.h>
20 #include <linux/ptrace.h>
21 #include <linux/seccomp.h>
23 #include <semaphore.h>
29 #include <linux/elf.h>
31 #include <sys/utsname.h>
32 #include <sys/fcntl.h>
34 #include <sys/times.h>
38 #include <sys/syscall.h>
40 #include "test_harness.h"
42 #ifndef PR_SET_PTRACER
43 # define PR_SET_PTRACER 0x59616d61
46 #ifndef PR_SET_NO_NEW_PRIVS
47 #define PR_SET_NO_NEW_PRIVS 38
48 #define PR_GET_NO_NEW_PRIVS 39
51 #ifndef PR_SECCOMP_EXT
52 #define PR_SECCOMP_EXT 43
55 #ifndef SECCOMP_EXT_ACT
56 #define SECCOMP_EXT_ACT 1
59 #ifndef SECCOMP_EXT_ACT_TSYNC
60 #define SECCOMP_EXT_ACT_TSYNC 1
63 #ifndef SECCOMP_MODE_STRICT
64 #define SECCOMP_MODE_STRICT 1
67 #ifndef SECCOMP_MODE_FILTER
68 #define SECCOMP_MODE_FILTER 2
71 #ifndef SECCOMP_RET_KILL
72 #define SECCOMP_RET_KILL 0x00000000U /* kill the task immediately */
73 #define SECCOMP_RET_TRAP 0x00030000U /* disallow and force a SIGSYS */
74 #define SECCOMP_RET_ERRNO 0x00050000U /* returns an errno */
75 #define SECCOMP_RET_TRACE 0x7ff00000U /* pass to a tracer or disallow */
76 #define SECCOMP_RET_ALLOW 0x7fff0000U /* allow */
78 /* Masks for the return value sections. */
79 #define SECCOMP_RET_ACTION 0x7fff0000U
80 #define SECCOMP_RET_DATA 0x0000ffffU
85 __u64 instruction_pointer
;
90 #if __BYTE_ORDER == __LITTLE_ENDIAN
91 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]))
92 #elif __BYTE_ORDER == __BIG_ENDIAN
93 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32))
95 #error "wut? Unknown __BYTE_ORDER?!"
98 #define SIBLING_EXIT_UNKILLED 0xbadbeef
99 #define SIBLING_EXIT_FAILURE 0xbadface
100 #define SIBLING_EXIT_NEWPRIVS 0xbadfeed
102 TEST(mode_strict_support
)
106 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_STRICT
, NULL
, NULL
, NULL
);
108 TH_LOG("Kernel does not support CONFIG_SECCOMP");
110 syscall(__NR_exit
, 1);
113 TEST_SIGNAL(mode_strict_cannot_call_prctl
, SIGKILL
)
117 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_STRICT
, NULL
, NULL
, NULL
);
119 TH_LOG("Kernel does not support CONFIG_SECCOMP");
121 syscall(__NR_prctl
, PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
,
124 TH_LOG("Unreachable!");
128 /* Note! This doesn't test no new privs behavior */
129 TEST(no_new_privs_support
)
133 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
135 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
139 /* Tests kernel support by checking for a copy_from_user() fault on * NULL. */
140 TEST(mode_filter_support
)
144 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, NULL
, 0, 0);
146 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
148 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, NULL
, NULL
, NULL
);
150 EXPECT_EQ(EFAULT
, errno
) {
151 TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!");
155 TEST(mode_filter_without_nnp
)
157 struct sock_filter filter
[] = {
158 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
160 struct sock_fprog prog
= {
161 .len
= (unsigned short)ARRAY_SIZE(filter
),
166 ret
= prctl(PR_GET_NO_NEW_PRIVS
, 0, NULL
, 0, 0);
168 TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS");
171 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
, 0, 0);
172 /* Succeeds with CAP_SYS_ADMIN, fails without */
173 /* TODO(wad) check caps not euid */
176 EXPECT_EQ(EACCES
, errno
);
182 #define MAX_INSNS_PER_PATH 32768
184 TEST(filter_size_limits
)
187 int count
= BPF_MAXINSNS
+ 1;
188 struct sock_filter allow
[] = {
189 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
191 struct sock_filter
*filter
;
192 struct sock_fprog prog
= { };
195 filter
= calloc(count
, sizeof(*filter
));
196 ASSERT_NE(NULL
, filter
);
198 for (i
= 0; i
< count
; i
++)
199 filter
[i
] = allow
[0];
201 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
204 prog
.filter
= filter
;
207 /* Too many filter instructions in a single filter. */
208 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
, 0, 0);
210 TH_LOG("Installing %d insn filter was allowed", prog
.len
);
213 /* One less is okay, though. */
215 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
, 0, 0);
217 TH_LOG("Installing %d insn filter wasn't allowed", prog
.len
);
221 TEST(filter_chain_limits
)
224 int count
= BPF_MAXINSNS
;
225 struct sock_filter allow
[] = {
226 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
228 struct sock_filter
*filter
;
229 struct sock_fprog prog
= { };
232 filter
= calloc(count
, sizeof(*filter
));
233 ASSERT_NE(NULL
, filter
);
235 for (i
= 0; i
< count
; i
++)
236 filter
[i
] = allow
[0];
238 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
241 prog
.filter
= filter
;
244 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
, 0, 0);
249 /* Too many total filter instructions. */
250 for (i
= 0; i
< MAX_INSNS_PER_PATH
; i
++) {
251 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
, 0, 0);
256 TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)",
257 i
, count
, i
* (count
+ 4));
261 TEST(mode_filter_cannot_move_to_strict
)
263 struct sock_filter filter
[] = {
264 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
266 struct sock_fprog prog
= {
267 .len
= (unsigned short)ARRAY_SIZE(filter
),
272 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
275 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
, 0, 0);
278 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_STRICT
, NULL
, 0, 0);
280 EXPECT_EQ(EINVAL
, errno
);
284 TEST(mode_filter_get_seccomp
)
286 struct sock_filter filter
[] = {
287 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
289 struct sock_fprog prog
= {
290 .len
= (unsigned short)ARRAY_SIZE(filter
),
295 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
298 ret
= prctl(PR_GET_SECCOMP
, 0, 0, 0, 0);
301 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
, 0, 0);
304 ret
= prctl(PR_GET_SECCOMP
, 0, 0, 0, 0);
311 struct sock_filter filter
[] = {
312 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
314 struct sock_fprog prog
= {
315 .len
= (unsigned short)ARRAY_SIZE(filter
),
320 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
323 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
);
329 struct sock_filter filter
[] = {
331 struct sock_fprog prog
= {
332 .len
= (unsigned short)ARRAY_SIZE(filter
),
337 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
340 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
);
342 EXPECT_EQ(EINVAL
, errno
);
345 TEST_SIGNAL(unknown_ret_is_kill_inside
, SIGSYS
)
347 struct sock_filter filter
[] = {
348 BPF_STMT(BPF_RET
|BPF_K
, 0x10000000U
),
350 struct sock_fprog prog
= {
351 .len
= (unsigned short)ARRAY_SIZE(filter
),
356 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
359 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
);
361 EXPECT_EQ(0, syscall(__NR_getpid
)) {
362 TH_LOG("getpid() shouldn't ever return");
366 /* return code >= 0x80000000 is unused. */
367 TEST_SIGNAL(unknown_ret_is_kill_above_allow
, SIGSYS
)
369 struct sock_filter filter
[] = {
370 BPF_STMT(BPF_RET
|BPF_K
, 0x90000000U
),
372 struct sock_fprog prog
= {
373 .len
= (unsigned short)ARRAY_SIZE(filter
),
378 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
381 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
);
383 EXPECT_EQ(0, syscall(__NR_getpid
)) {
384 TH_LOG("getpid() shouldn't ever return");
388 TEST_SIGNAL(KILL_all
, SIGSYS
)
390 struct sock_filter filter
[] = {
391 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_KILL
),
393 struct sock_fprog prog
= {
394 .len
= (unsigned short)ARRAY_SIZE(filter
),
399 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
402 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
);
406 TEST_SIGNAL(KILL_one
, SIGSYS
)
408 struct sock_filter filter
[] = {
409 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
410 offsetof(struct seccomp_data
, nr
)),
411 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_getpid
, 0, 1),
412 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_KILL
),
413 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
415 struct sock_fprog prog
= {
416 .len
= (unsigned short)ARRAY_SIZE(filter
),
420 pid_t parent
= getppid();
422 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
425 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
);
428 EXPECT_EQ(parent
, syscall(__NR_getppid
));
429 /* getpid() should never return. */
430 EXPECT_EQ(0, syscall(__NR_getpid
));
433 TEST_SIGNAL(KILL_one_arg_one
, SIGSYS
)
436 struct sock_filter filter
[] = {
437 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
438 offsetof(struct seccomp_data
, nr
)),
439 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_times
, 1, 0),
440 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
441 /* Only both with lower 32-bit for now. */
442 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
, syscall_arg(0)),
443 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
,
444 (unsigned long)&fatal_address
, 0, 1),
445 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_KILL
),
446 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
448 struct sock_fprog prog
= {
449 .len
= (unsigned short)ARRAY_SIZE(filter
),
453 pid_t parent
= getppid();
455 clock_t clock
= times(&timebuf
);
457 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
460 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
);
463 EXPECT_EQ(parent
, syscall(__NR_getppid
));
464 EXPECT_LE(clock
, syscall(__NR_times
, &timebuf
));
465 /* times() should never return. */
466 EXPECT_EQ(0, syscall(__NR_times
, &fatal_address
));
469 TEST_SIGNAL(KILL_one_arg_six
, SIGSYS
)
472 int sysno
= __NR_mmap
;
474 int sysno
= __NR_mmap2
;
476 struct sock_filter filter
[] = {
477 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
478 offsetof(struct seccomp_data
, nr
)),
479 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, sysno
, 1, 0),
480 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
481 /* Only both with lower 32-bit for now. */
482 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
, syscall_arg(5)),
483 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, 0x0C0FFEE, 0, 1),
484 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_KILL
),
485 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
487 struct sock_fprog prog
= {
488 .len
= (unsigned short)ARRAY_SIZE(filter
),
492 pid_t parent
= getppid();
495 int page_size
= sysconf(_SC_PAGESIZE
);
497 ASSERT_LT(0, page_size
);
499 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
502 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
);
505 fd
= open("/dev/zero", O_RDONLY
);
508 EXPECT_EQ(parent
, syscall(__NR_getppid
));
509 map1
= (void *)syscall(sysno
,
510 NULL
, page_size
, PROT_READ
, MAP_PRIVATE
, fd
, page_size
);
511 EXPECT_NE(MAP_FAILED
, map1
);
512 /* mmap2() should never return. */
513 map2
= (void *)syscall(sysno
,
514 NULL
, page_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0x0C0FFEE);
515 EXPECT_EQ(MAP_FAILED
, map2
);
517 /* The test failed, so clean up the resources. */
518 munmap(map1
, page_size
);
519 munmap(map2
, page_size
);
523 /* TODO(wad) add 64-bit versus 32-bit arg tests. */
524 TEST(arg_out_of_range
)
526 struct sock_filter filter
[] = {
527 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
, syscall_arg(6)),
528 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
530 struct sock_fprog prog
= {
531 .len
= (unsigned short)ARRAY_SIZE(filter
),
536 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
539 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
);
541 EXPECT_EQ(EINVAL
, errno
);
546 struct sock_filter filter
[] = {
547 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
548 offsetof(struct seccomp_data
, nr
)),
549 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_read
, 0, 1),
550 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ERRNO
| E2BIG
),
551 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
553 struct sock_fprog prog
= {
554 .len
= (unsigned short)ARRAY_SIZE(filter
),
558 pid_t parent
= getppid();
560 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
563 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
);
566 EXPECT_EQ(parent
, syscall(__NR_getppid
));
567 EXPECT_EQ(-1, read(0, NULL
, 0));
568 EXPECT_EQ(E2BIG
, errno
);
573 struct sock_filter filter
[] = {
574 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
575 offsetof(struct seccomp_data
, nr
)),
576 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_read
, 0, 1),
577 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ERRNO
| 0),
578 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
580 struct sock_fprog prog
= {
581 .len
= (unsigned short)ARRAY_SIZE(filter
),
585 pid_t parent
= getppid();
587 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
590 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
);
593 EXPECT_EQ(parent
, syscall(__NR_getppid
));
594 /* "errno" of 0 is ok. */
595 EXPECT_EQ(0, read(0, NULL
, 0));
600 struct sock_filter filter
[] = {
601 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
602 offsetof(struct seccomp_data
, nr
)),
603 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_read
, 0, 1),
604 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ERRNO
| 4096),
605 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
607 struct sock_fprog prog
= {
608 .len
= (unsigned short)ARRAY_SIZE(filter
),
612 pid_t parent
= getppid();
614 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
617 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
);
620 EXPECT_EQ(parent
, syscall(__NR_getppid
));
621 EXPECT_EQ(-1, read(0, NULL
, 0));
622 EXPECT_EQ(4095, errno
);
626 struct sock_fprog prog
;
631 struct sock_filter filter
[] = {
632 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
633 offsetof(struct seccomp_data
, nr
)),
634 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_getpid
, 0, 1),
635 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_TRAP
),
636 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
639 memset(&self
->prog
, 0, sizeof(self
->prog
));
640 self
->prog
.filter
= malloc(sizeof(filter
));
641 ASSERT_NE(NULL
, self
->prog
.filter
);
642 memcpy(self
->prog
.filter
, filter
, sizeof(filter
));
643 self
->prog
.len
= (unsigned short)ARRAY_SIZE(filter
);
646 FIXTURE_TEARDOWN(TRAP
)
648 if (self
->prog
.filter
)
649 free(self
->prog
.filter
);
652 TEST_F_SIGNAL(TRAP
, dfl
, SIGSYS
)
656 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
659 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->prog
);
661 syscall(__NR_getpid
);
664 /* Ensure that SIGSYS overrides SIG_IGN */
665 TEST_F_SIGNAL(TRAP
, ign
, SIGSYS
)
669 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
672 signal(SIGSYS
, SIG_IGN
);
674 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->prog
);
676 syscall(__NR_getpid
);
679 static struct siginfo TRAP_info
;
680 static volatile int TRAP_nr
;
681 static void TRAP_action(int nr
, siginfo_t
*info
, void *void_context
)
683 memcpy(&TRAP_info
, info
, sizeof(TRAP_info
));
687 TEST_F(TRAP
, handler
)
690 struct sigaction act
;
693 memset(&act
, 0, sizeof(act
));
695 sigaddset(&mask
, SIGSYS
);
697 act
.sa_sigaction
= &TRAP_action
;
698 act
.sa_flags
= SA_SIGINFO
;
699 ret
= sigaction(SIGSYS
, &act
, NULL
);
701 TH_LOG("sigaction failed");
703 ret
= sigprocmask(SIG_UNBLOCK
, &mask
, NULL
);
705 TH_LOG("sigprocmask failed");
708 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
710 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->prog
);
713 memset(&TRAP_info
, 0, sizeof(TRAP_info
));
714 /* Expect the registers to be rolled back. (nr = error) may vary
716 ret
= syscall(__NR_getpid
);
717 /* Silence gcc warning about volatile. */
719 EXPECT_EQ(SIGSYS
, test
);
720 struct local_sigsys
{
721 void *_call_addr
; /* calling user insn */
722 int _syscall
; /* triggering system call number */
723 unsigned int _arch
; /* AUDIT_ARCH_* of syscall */
724 } *sigsys
= (struct local_sigsys
*)
726 &(TRAP_info
.si_call_addr
);
730 EXPECT_EQ(__NR_getpid
, sigsys
->_syscall
);
731 /* Make sure arch is non-zero. */
732 EXPECT_NE(0, sigsys
->_arch
);
733 EXPECT_NE(0, (unsigned long)sigsys
->_call_addr
);
736 FIXTURE_DATA(precedence
) {
737 struct sock_fprog allow
;
738 struct sock_fprog trace
;
739 struct sock_fprog error
;
740 struct sock_fprog trap
;
741 struct sock_fprog kill
;
744 FIXTURE_SETUP(precedence
)
746 struct sock_filter allow_insns
[] = {
747 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
749 struct sock_filter trace_insns
[] = {
750 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
751 offsetof(struct seccomp_data
, nr
)),
752 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_getpid
, 1, 0),
753 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
754 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_TRACE
),
756 struct sock_filter error_insns
[] = {
757 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
758 offsetof(struct seccomp_data
, nr
)),
759 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_getpid
, 1, 0),
760 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
761 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ERRNO
),
763 struct sock_filter trap_insns
[] = {
764 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
765 offsetof(struct seccomp_data
, nr
)),
766 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_getpid
, 1, 0),
767 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
768 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_TRAP
),
770 struct sock_filter kill_insns
[] = {
771 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
772 offsetof(struct seccomp_data
, nr
)),
773 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_getpid
, 1, 0),
774 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
775 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_KILL
),
778 memset(self
, 0, sizeof(*self
));
779 #define FILTER_ALLOC(_x) \
780 self->_x.filter = malloc(sizeof(_x##_insns)); \
781 ASSERT_NE(NULL, self->_x.filter); \
782 memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \
783 self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns)
791 FIXTURE_TEARDOWN(precedence
)
793 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter)
801 TEST_F(precedence
, allow_ok
)
803 pid_t parent
, res
= 0;
807 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
810 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->allow
);
812 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->trace
);
814 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->error
);
816 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->trap
);
818 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->kill
);
820 /* Should work just fine. */
821 res
= syscall(__NR_getppid
);
822 EXPECT_EQ(parent
, res
);
825 TEST_F_SIGNAL(precedence
, kill_is_highest
, SIGSYS
)
827 pid_t parent
, res
= 0;
831 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
834 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->allow
);
836 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->trace
);
838 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->error
);
840 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->trap
);
842 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->kill
);
844 /* Should work just fine. */
845 res
= syscall(__NR_getppid
);
846 EXPECT_EQ(parent
, res
);
847 /* getpid() should never return. */
848 res
= syscall(__NR_getpid
);
852 TEST_F_SIGNAL(precedence
, kill_is_highest_in_any_order
, SIGSYS
)
858 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
861 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->allow
);
863 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->kill
);
865 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->error
);
867 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->trace
);
869 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->trap
);
871 /* Should work just fine. */
872 EXPECT_EQ(parent
, syscall(__NR_getppid
));
873 /* getpid() should never return. */
874 EXPECT_EQ(0, syscall(__NR_getpid
));
877 TEST_F_SIGNAL(precedence
, trap_is_second
, SIGSYS
)
883 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
886 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->allow
);
888 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->trace
);
890 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->error
);
892 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->trap
);
894 /* Should work just fine. */
895 EXPECT_EQ(parent
, syscall(__NR_getppid
));
896 /* getpid() should never return. */
897 EXPECT_EQ(0, syscall(__NR_getpid
));
900 TEST_F_SIGNAL(precedence
, trap_is_second_in_any_order
, SIGSYS
)
906 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
909 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->allow
);
911 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->trap
);
913 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->trace
);
915 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->error
);
917 /* Should work just fine. */
918 EXPECT_EQ(parent
, syscall(__NR_getppid
));
919 /* getpid() should never return. */
920 EXPECT_EQ(0, syscall(__NR_getpid
));
923 TEST_F(precedence
, errno_is_third
)
929 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
932 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->allow
);
934 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->trace
);
936 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->error
);
938 /* Should work just fine. */
939 EXPECT_EQ(parent
, syscall(__NR_getppid
));
940 EXPECT_EQ(0, syscall(__NR_getpid
));
943 TEST_F(precedence
, errno_is_third_in_any_order
)
949 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
952 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->error
);
954 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->trace
);
956 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->allow
);
958 /* Should work just fine. */
959 EXPECT_EQ(parent
, syscall(__NR_getppid
));
960 EXPECT_EQ(0, syscall(__NR_getpid
));
963 TEST_F(precedence
, trace_is_fourth
)
969 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
972 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->allow
);
974 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->trace
);
976 /* Should work just fine. */
977 EXPECT_EQ(parent
, syscall(__NR_getppid
));
979 EXPECT_EQ(-1, syscall(__NR_getpid
));
982 TEST_F(precedence
, trace_is_fourth_in_any_order
)
988 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
991 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->trace
);
993 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->allow
);
995 /* Should work just fine. */
996 EXPECT_EQ(parent
, syscall(__NR_getppid
));
998 EXPECT_EQ(-1, syscall(__NR_getpid
));
1001 #ifndef PTRACE_O_TRACESECCOMP
1002 #define PTRACE_O_TRACESECCOMP 0x00000080
1005 /* Catch the Ubuntu 12.04 value error. */
1006 #if PTRACE_EVENT_SECCOMP != 7
1007 #undef PTRACE_EVENT_SECCOMP
1010 #ifndef PTRACE_EVENT_SECCOMP
1011 #define PTRACE_EVENT_SECCOMP 7
1014 #define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP)
1015 bool tracer_running
;
1016 void tracer_stop(int sig
)
1018 tracer_running
= false;
1021 typedef void tracer_func_t(struct __test_metadata
*_metadata
,
1022 pid_t tracee
, int status
, void *args
);
1024 void tracer(struct __test_metadata
*_metadata
, int fd
, pid_t tracee
,
1025 tracer_func_t tracer_func
, void *args
)
1028 struct sigaction action
= {
1029 .sa_handler
= tracer_stop
,
1032 /* Allow external shutdown. */
1033 tracer_running
= true;
1034 ASSERT_EQ(0, sigaction(SIGUSR1
, &action
, NULL
));
1037 while (ret
== -1 && errno
!= EINVAL
)
1038 ret
= ptrace(PTRACE_ATTACH
, tracee
, NULL
, 0);
1040 kill(tracee
, SIGKILL
);
1042 /* Wait for attach stop */
1045 ret
= ptrace(PTRACE_SETOPTIONS
, tracee
, NULL
, PTRACE_O_TRACESECCOMP
);
1047 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP");
1048 kill(tracee
, SIGKILL
);
1050 ptrace(PTRACE_CONT
, tracee
, NULL
, 0);
1052 /* Unblock the tracee */
1053 ASSERT_EQ(1, write(fd
, "A", 1));
1054 ASSERT_EQ(0, close(fd
));
1056 /* Run until we're shut down. Must assert to stop execution. */
1057 while (tracer_running
) {
1060 if (wait(&status
) != tracee
)
1062 if (WIFSIGNALED(status
) || WIFEXITED(status
))
1063 /* Child is dead. Time to go. */
1066 /* Make sure this is a seccomp event. */
1067 ASSERT_EQ(true, IS_SECCOMP_EVENT(status
));
1069 tracer_func(_metadata
, tracee
, status
, args
);
1071 ret
= ptrace(PTRACE_CONT
, tracee
, NULL
, NULL
);
1074 /* Directly report the status of our test harness results. */
1075 syscall(__NR_exit
, _metadata
->passed
? EXIT_SUCCESS
: EXIT_FAILURE
);
1078 /* Common tracer setup/teardown functions. */
1079 void cont_handler(int num
)
1081 pid_t
setup_trace_fixture(struct __test_metadata
*_metadata
,
1082 tracer_func_t func
, void *args
)
1087 pid_t tracee
= getpid();
1089 /* Setup a pipe for clean synchronization. */
1090 ASSERT_EQ(0, pipe(pipefd
));
1092 /* Fork a child which we'll promote to tracer */
1093 tracer_pid
= fork();
1094 ASSERT_LE(0, tracer_pid
);
1095 signal(SIGALRM
, cont_handler
);
1096 if (tracer_pid
== 0) {
1098 tracer(_metadata
, pipefd
[1], tracee
, func
, args
);
1099 syscall(__NR_exit
, 0);
1102 prctl(PR_SET_PTRACER
, tracer_pid
, 0, 0, 0);
1103 read(pipefd
[0], &sync
, 1);
1108 void teardown_trace_fixture(struct __test_metadata
*_metadata
,
1114 * Extract the exit code from the other process and
1115 * adopt it for ourselves in case its asserts failed.
1117 ASSERT_EQ(0, kill(tracer
, SIGUSR1
));
1118 ASSERT_EQ(tracer
, waitpid(tracer
, &status
, 0));
1119 if (WEXITSTATUS(status
))
1120 _metadata
->passed
= 0;
1124 /* "poke" tracer arguments and function. */
1125 struct tracer_args_poke_t
{
1126 unsigned long poke_addr
;
1129 void tracer_poke(struct __test_metadata
*_metadata
, pid_t tracee
, int status
,
1134 struct tracer_args_poke_t
*info
= (struct tracer_args_poke_t
*)args
;
1136 ret
= ptrace(PTRACE_GETEVENTMSG
, tracee
, NULL
, &msg
);
1138 /* If this fails, don't try to recover. */
1139 ASSERT_EQ(0x1001, msg
) {
1140 kill(tracee
, SIGKILL
);
1143 * Poke in the message.
1144 * Registers are not touched to try to keep this relatively arch
1147 ret
= ptrace(PTRACE_POKEDATA
, tracee
, info
->poke_addr
, 0x1001);
1151 FIXTURE_DATA(TRACE_poke
) {
1152 struct sock_fprog prog
;
1155 struct tracer_args_poke_t tracer_args
;
1158 FIXTURE_SETUP(TRACE_poke
)
1160 struct sock_filter filter
[] = {
1161 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
1162 offsetof(struct seccomp_data
, nr
)),
1163 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_read
, 0, 1),
1164 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_TRACE
| 0x1001),
1165 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
1169 memset(&self
->prog
, 0, sizeof(self
->prog
));
1170 self
->prog
.filter
= malloc(sizeof(filter
));
1171 ASSERT_NE(NULL
, self
->prog
.filter
);
1172 memcpy(self
->prog
.filter
, filter
, sizeof(filter
));
1173 self
->prog
.len
= (unsigned short)ARRAY_SIZE(filter
);
1175 /* Set up tracer args. */
1176 self
->tracer_args
.poke_addr
= (unsigned long)&self
->poked
;
1178 /* Launch tracer. */
1179 self
->tracer
= setup_trace_fixture(_metadata
, tracer_poke
,
1180 &self
->tracer_args
);
1183 FIXTURE_TEARDOWN(TRACE_poke
)
1185 teardown_trace_fixture(_metadata
, self
->tracer
);
1186 if (self
->prog
.filter
)
1187 free(self
->prog
.filter
);
1190 TEST_F(TRACE_poke
, read_has_side_effects
)
1194 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
1197 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->prog
, 0, 0);
1200 EXPECT_EQ(0, self
->poked
);
1201 ret
= read(-1, NULL
, 0);
1203 EXPECT_EQ(0x1001, self
->poked
);
1206 TEST_F(TRACE_poke
, getpid_runs_normally
)
1210 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
1213 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->prog
, 0, 0);
1216 EXPECT_EQ(0, self
->poked
);
1217 EXPECT_NE(0, syscall(__NR_getpid
));
1218 EXPECT_EQ(0, self
->poked
);
1221 #if defined(__x86_64__)
1222 # define ARCH_REGS struct user_regs_struct
1223 # define SYSCALL_NUM orig_rax
1224 # define SYSCALL_RET rax
1225 #elif defined(__i386__)
1226 # define ARCH_REGS struct user_regs_struct
1227 # define SYSCALL_NUM orig_eax
1228 # define SYSCALL_RET eax
1229 #elif defined(__arm__)
1230 # define ARCH_REGS struct pt_regs
1231 # define SYSCALL_NUM ARM_r7
1232 # define SYSCALL_RET ARM_r0
1233 #elif defined(__aarch64__)
1234 # define ARCH_REGS struct user_pt_regs
1235 # define SYSCALL_NUM regs[8]
1236 # define SYSCALL_RET regs[0]
1237 #elif defined(__powerpc__)
1238 # define ARCH_REGS struct pt_regs
1239 # define SYSCALL_NUM gpr[0]
1240 # define SYSCALL_RET gpr[3]
1241 #elif defined(__s390__)
1242 # define ARCH_REGS s390_regs
1243 # define SYSCALL_NUM gprs[2]
1244 # define SYSCALL_RET gprs[2]
1246 # error "Do not know how to find your architecture's registers and syscalls"
1249 /* Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for
1250 * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux).
1252 #if defined(__x86_64__) || defined(__i386__)
1253 #define HAVE_GETREGS
1256 /* Architecture-specific syscall fetching routine. */
1257 int get_syscall(struct __test_metadata
*_metadata
, pid_t tracee
)
1261 EXPECT_EQ(0, ptrace(PTRACE_GETREGS
, tracee
, 0, ®s
)) {
1262 TH_LOG("PTRACE_GETREGS failed");
1268 iov
.iov_base
= ®s
;
1269 iov
.iov_len
= sizeof(regs
);
1270 EXPECT_EQ(0, ptrace(PTRACE_GETREGSET
, tracee
, NT_PRSTATUS
, &iov
)) {
1271 TH_LOG("PTRACE_GETREGSET failed");
1276 return regs
.SYSCALL_NUM
;
1279 /* Architecture-specific syscall changing routine. */
1280 void change_syscall(struct __test_metadata
*_metadata
,
1281 pid_t tracee
, int syscall
)
1286 ret
= ptrace(PTRACE_GETREGS
, tracee
, 0, ®s
);
1289 iov
.iov_base
= ®s
;
1290 iov
.iov_len
= sizeof(regs
);
1291 ret
= ptrace(PTRACE_GETREGSET
, tracee
, NT_PRSTATUS
, &iov
);
1295 #if defined(__x86_64__) || defined(__i386__) || defined(__powerpc__) || \
1298 regs
.SYSCALL_NUM
= syscall
;
1301 #elif defined(__arm__)
1302 # ifndef PTRACE_SET_SYSCALL
1303 # define PTRACE_SET_SYSCALL 23
1306 ret
= ptrace(PTRACE_SET_SYSCALL
, tracee
, NULL
, syscall
);
1310 #elif defined(__aarch64__)
1311 # ifndef NT_ARM_SYSTEM_CALL
1312 # define NT_ARM_SYSTEM_CALL 0x404
1315 iov
.iov_base
= &syscall
;
1316 iov
.iov_len
= sizeof(syscall
);
1317 ret
= ptrace(PTRACE_SETREGSET
, tracee
, NT_ARM_SYSTEM_CALL
,
1324 TH_LOG("How is the syscall changed on this architecture?");
1328 /* If syscall is skipped, change return value. */
1330 regs
.SYSCALL_RET
= 1;
1333 ret
= ptrace(PTRACE_SETREGS
, tracee
, 0, ®s
);
1335 iov
.iov_base
= ®s
;
1336 iov
.iov_len
= sizeof(regs
);
1337 ret
= ptrace(PTRACE_SETREGSET
, tracee
, NT_PRSTATUS
, &iov
);
1342 void tracer_syscall(struct __test_metadata
*_metadata
, pid_t tracee
,
1343 int status
, void *args
)
1348 /* Make sure we got the right message. */
1349 ret
= ptrace(PTRACE_GETEVENTMSG
, tracee
, NULL
, &msg
);
1352 /* Validate and take action on expected syscalls. */
1355 /* change getpid to getppid. */
1356 EXPECT_EQ(__NR_getpid
, get_syscall(_metadata
, tracee
));
1357 change_syscall(_metadata
, tracee
, __NR_getppid
);
1361 EXPECT_EQ(__NR_gettid
, get_syscall(_metadata
, tracee
));
1362 change_syscall(_metadata
, tracee
, -1);
1365 /* do nothing (allow getppid) */
1366 EXPECT_EQ(__NR_getppid
, get_syscall(_metadata
, tracee
));
1370 TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg
);
1371 kill(tracee
, SIGKILL
);
1377 FIXTURE_DATA(TRACE_syscall
) {
1378 struct sock_fprog prog
;
1379 pid_t tracer
, mytid
, mypid
, parent
;
1382 FIXTURE_SETUP(TRACE_syscall
)
1384 struct sock_filter filter
[] = {
1385 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
1386 offsetof(struct seccomp_data
, nr
)),
1387 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_getpid
, 0, 1),
1388 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_TRACE
| 0x1002),
1389 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_gettid
, 0, 1),
1390 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_TRACE
| 0x1003),
1391 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_getppid
, 0, 1),
1392 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_TRACE
| 0x1004),
1393 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
1396 memset(&self
->prog
, 0, sizeof(self
->prog
));
1397 self
->prog
.filter
= malloc(sizeof(filter
));
1398 ASSERT_NE(NULL
, self
->prog
.filter
);
1399 memcpy(self
->prog
.filter
, filter
, sizeof(filter
));
1400 self
->prog
.len
= (unsigned short)ARRAY_SIZE(filter
);
1402 /* Prepare some testable syscall results. */
1403 self
->mytid
= syscall(__NR_gettid
);
1404 ASSERT_GT(self
->mytid
, 0);
1405 ASSERT_NE(self
->mytid
, 1) {
1406 TH_LOG("Running this test as init is not supported. :)");
1409 self
->mypid
= getpid();
1410 ASSERT_GT(self
->mypid
, 0);
1411 ASSERT_EQ(self
->mytid
, self
->mypid
);
1413 self
->parent
= getppid();
1414 ASSERT_GT(self
->parent
, 0);
1415 ASSERT_NE(self
->parent
, self
->mypid
);
1417 /* Launch tracer. */
1418 self
->tracer
= setup_trace_fixture(_metadata
, tracer_syscall
, NULL
);
1421 FIXTURE_TEARDOWN(TRACE_syscall
)
1423 teardown_trace_fixture(_metadata
, self
->tracer
);
1424 if (self
->prog
.filter
)
1425 free(self
->prog
.filter
);
1428 TEST_F(TRACE_syscall
, syscall_allowed
)
1432 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
1435 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->prog
, 0, 0);
1438 /* getppid works as expected (no changes). */
1439 EXPECT_EQ(self
->parent
, syscall(__NR_getppid
));
1440 EXPECT_NE(self
->mypid
, syscall(__NR_getppid
));
1443 TEST_F(TRACE_syscall
, syscall_redirected
)
1447 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
1450 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->prog
, 0, 0);
1453 /* getpid has been redirected to getppid as expected. */
1454 EXPECT_EQ(self
->parent
, syscall(__NR_getpid
));
1455 EXPECT_NE(self
->mypid
, syscall(__NR_getpid
));
1458 TEST_F(TRACE_syscall
, syscall_dropped
)
1462 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
1465 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &self
->prog
, 0, 0);
1468 /* gettid has been skipped and an altered return value stored. */
1469 EXPECT_EQ(1, syscall(__NR_gettid
));
1470 EXPECT_NE(self
->mytid
, syscall(__NR_gettid
));
1473 #ifndef __NR_seccomp
1474 # if defined(__i386__)
1475 # define __NR_seccomp 354
1476 # elif defined(__x86_64__)
1477 # define __NR_seccomp 317
1478 # elif defined(__arm__)
1479 # define __NR_seccomp 383
1480 # elif defined(__aarch64__)
1481 # define __NR_seccomp 277
1482 # elif defined(__powerpc__)
1483 # define __NR_seccomp 358
1484 # elif defined(__s390__)
1485 # define __NR_seccomp 348
1487 # warning "seccomp syscall number unknown for this architecture"
1488 # define __NR_seccomp 0xffff
1492 #ifndef SECCOMP_SET_MODE_STRICT
1493 #define SECCOMP_SET_MODE_STRICT 0
1496 #ifndef SECCOMP_SET_MODE_FILTER
1497 #define SECCOMP_SET_MODE_FILTER 1
1500 #ifndef SECCOMP_FILTER_FLAG_TSYNC
1501 #define SECCOMP_FILTER_FLAG_TSYNC 1
1505 int seccomp(unsigned int op
, unsigned int flags
, void *args
)
1508 return syscall(__NR_seccomp
, op
, flags
, args
);
1512 TEST(seccomp_syscall
)
1514 struct sock_filter filter
[] = {
1515 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
1517 struct sock_fprog prog
= {
1518 .len
= (unsigned short)ARRAY_SIZE(filter
),
1523 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0);
1525 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1528 /* Reject insane operation. */
1529 ret
= seccomp(-1, 0, &prog
);
1530 ASSERT_NE(ENOSYS
, errno
) {
1531 TH_LOG("Kernel does not support seccomp syscall!");
1533 EXPECT_EQ(EINVAL
, errno
) {
1534 TH_LOG("Did not reject crazy op value!");
1537 /* Reject strict with flags or pointer. */
1538 ret
= seccomp(SECCOMP_SET_MODE_STRICT
, -1, NULL
);
1539 EXPECT_EQ(EINVAL
, errno
) {
1540 TH_LOG("Did not reject mode strict with flags!");
1542 ret
= seccomp(SECCOMP_SET_MODE_STRICT
, 0, &prog
);
1543 EXPECT_EQ(EINVAL
, errno
) {
1544 TH_LOG("Did not reject mode strict with uargs!");
1547 /* Reject insane args for filter. */
1548 ret
= seccomp(SECCOMP_SET_MODE_FILTER
, -1, &prog
);
1549 EXPECT_EQ(EINVAL
, errno
) {
1550 TH_LOG("Did not reject crazy filter flags!");
1552 ret
= seccomp(SECCOMP_SET_MODE_FILTER
, 0, NULL
);
1553 EXPECT_EQ(EFAULT
, errno
) {
1554 TH_LOG("Did not reject NULL filter!");
1557 ret
= seccomp(SECCOMP_SET_MODE_FILTER
, 0, &prog
);
1558 EXPECT_EQ(0, errno
) {
1559 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s",
1564 TEST(seccomp_syscall_mode_lock
)
1566 struct sock_filter filter
[] = {
1567 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
1569 struct sock_fprog prog
= {
1570 .len
= (unsigned short)ARRAY_SIZE(filter
),
1575 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, NULL
, 0, 0);
1577 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1580 ret
= seccomp(SECCOMP_SET_MODE_FILTER
, 0, &prog
);
1581 ASSERT_NE(ENOSYS
, errno
) {
1582 TH_LOG("Kernel does not support seccomp syscall!");
1585 TH_LOG("Could not install filter!");
1588 /* Make sure neither entry point will switch to strict. */
1589 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_STRICT
, 0, 0, 0);
1590 EXPECT_EQ(EINVAL
, errno
) {
1591 TH_LOG("Switched to mode strict!");
1594 ret
= seccomp(SECCOMP_SET_MODE_STRICT
, 0, NULL
);
1595 EXPECT_EQ(EINVAL
, errno
) {
1596 TH_LOG("Switched to mode strict!");
1602 struct sock_filter filter
[] = {
1603 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
1605 struct sock_fprog prog
= {
1606 .len
= (unsigned short)ARRAY_SIZE(filter
),
1611 ret
= prctl(PR_SET_NO_NEW_PRIVS
, 1, NULL
, 0, 0);
1613 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1616 ret
= seccomp(SECCOMP_SET_MODE_FILTER
, SECCOMP_FILTER_FLAG_TSYNC
,
1618 ASSERT_NE(ENOSYS
, errno
) {
1619 TH_LOG("Kernel does not support seccomp syscall!");
1622 TH_LOG("Could not install initial filter with TSYNC!");
1626 #define TSYNC_SIBLINGS 2
1627 struct tsync_sibling
{
1631 pthread_cond_t
*cond
;
1632 pthread_mutex_t
*mutex
;
1635 struct sock_fprog
*prog
;
1636 struct __test_metadata
*metadata
;
1639 FIXTURE_DATA(TSYNC
) {
1640 struct sock_fprog root_prog
, apply_prog
;
1641 struct tsync_sibling sibling
[TSYNC_SIBLINGS
];
1643 pthread_cond_t cond
;
1644 pthread_mutex_t mutex
;
1648 FIXTURE_SETUP(TSYNC
)
1650 struct sock_filter root_filter
[] = {
1651 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
1653 struct sock_filter apply_filter
[] = {
1654 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
1655 offsetof(struct seccomp_data
, nr
)),
1656 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_read
, 0, 1),
1657 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_KILL
),
1658 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
1661 memset(&self
->root_prog
, 0, sizeof(self
->root_prog
));
1662 memset(&self
->apply_prog
, 0, sizeof(self
->apply_prog
));
1663 memset(&self
->sibling
, 0, sizeof(self
->sibling
));
1664 self
->root_prog
.filter
= malloc(sizeof(root_filter
));
1665 ASSERT_NE(NULL
, self
->root_prog
.filter
);
1666 memcpy(self
->root_prog
.filter
, &root_filter
, sizeof(root_filter
));
1667 self
->root_prog
.len
= (unsigned short)ARRAY_SIZE(root_filter
);
1669 self
->apply_prog
.filter
= malloc(sizeof(apply_filter
));
1670 ASSERT_NE(NULL
, self
->apply_prog
.filter
);
1671 memcpy(self
->apply_prog
.filter
, &apply_filter
, sizeof(apply_filter
));
1672 self
->apply_prog
.len
= (unsigned short)ARRAY_SIZE(apply_filter
);
1674 self
->sibling_count
= 0;
1675 pthread_mutex_init(&self
->mutex
, NULL
);
1676 pthread_cond_init(&self
->cond
, NULL
);
1677 sem_init(&self
->started
, 0, 0);
1678 self
->sibling
[0].tid
= 0;
1679 self
->sibling
[0].cond
= &self
->cond
;
1680 self
->sibling
[0].started
= &self
->started
;
1681 self
->sibling
[0].mutex
= &self
->mutex
;
1682 self
->sibling
[0].diverge
= 0;
1683 self
->sibling
[0].num_waits
= 1;
1684 self
->sibling
[0].prog
= &self
->root_prog
;
1685 self
->sibling
[0].metadata
= _metadata
;
1686 self
->sibling
[1].tid
= 0;
1687 self
->sibling
[1].cond
= &self
->cond
;
1688 self
->sibling
[1].started
= &self
->started
;
1689 self
->sibling
[1].mutex
= &self
->mutex
;
1690 self
->sibling
[1].diverge
= 0;
1691 self
->sibling
[1].prog
= &self
->root_prog
;
1692 self
->sibling
[1].num_waits
= 1;
1693 self
->sibling
[1].metadata
= _metadata
;
1696 FIXTURE_TEARDOWN(TSYNC
)
1700 if (self
->root_prog
.filter
)
1701 free(self
->root_prog
.filter
);
1702 if (self
->apply_prog
.filter
)
1703 free(self
->apply_prog
.filter
);
1705 for ( ; sib
< self
->sibling_count
; ++sib
) {
1706 struct tsync_sibling
*s
= &self
->sibling
[sib
];
1711 if (pthread_kill(s
->tid
, 0)) {
1712 pthread_cancel(s
->tid
);
1713 pthread_join(s
->tid
, &status
);
1716 pthread_mutex_destroy(&self
->mutex
);
1717 pthread_cond_destroy(&self
->cond
);
1718 sem_destroy(&self
->started
);
1721 void *tsync_sibling(void *data
)
1724 struct tsync_sibling
*me
= data
;
1726 me
->system_tid
= syscall(__NR_gettid
);
1728 pthread_mutex_lock(me
->mutex
);
1730 /* Just re-apply the root prog to fork the tree */
1731 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
,
1734 sem_post(me
->started
);
1735 /* Return outside of started so parent notices failures. */
1737 pthread_mutex_unlock(me
->mutex
);
1738 return (void *)SIBLING_EXIT_FAILURE
;
1741 pthread_cond_wait(me
->cond
, me
->mutex
);
1742 me
->num_waits
= me
->num_waits
- 1;
1743 } while (me
->num_waits
);
1744 pthread_mutex_unlock(me
->mutex
);
1746 ret
= prctl(PR_GET_NO_NEW_PRIVS
, 0, 0, 0, 0);
1748 return (void *)SIBLING_EXIT_NEWPRIVS
;
1750 return (void *)SIBLING_EXIT_UNKILLED
;
1753 void tsync_start_sibling(struct tsync_sibling
*sibling
)
1755 pthread_create(&sibling
->tid
, NULL
, tsync_sibling
, (void *)sibling
);
1758 TEST_F(TSYNC
, siblings_fail_prctl
)
1762 struct sock_filter filter
[] = {
1763 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
1764 offsetof(struct seccomp_data
, nr
)),
1765 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_prctl
, 0, 1),
1766 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ERRNO
| EINVAL
),
1767 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
1769 struct sock_fprog prog
= {
1770 .len
= (unsigned short)ARRAY_SIZE(filter
),
1774 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0)) {
1775 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1778 /* Check prctl failure detection by requesting sib 0 diverge. */
1779 ret
= seccomp(SECCOMP_SET_MODE_FILTER
, 0, &prog
);
1780 ASSERT_NE(ENOSYS
, errno
) {
1781 TH_LOG("Kernel does not support seccomp syscall!");
1784 TH_LOG("setting filter failed");
1787 self
->sibling
[0].diverge
= 1;
1788 tsync_start_sibling(&self
->sibling
[0]);
1789 tsync_start_sibling(&self
->sibling
[1]);
1791 while (self
->sibling_count
< TSYNC_SIBLINGS
) {
1792 sem_wait(&self
->started
);
1793 self
->sibling_count
++;
1796 /* Signal the threads to clean up*/
1797 pthread_mutex_lock(&self
->mutex
);
1798 ASSERT_EQ(0, pthread_cond_broadcast(&self
->cond
)) {
1799 TH_LOG("cond broadcast non-zero");
1801 pthread_mutex_unlock(&self
->mutex
);
1803 /* Ensure diverging sibling failed to call prctl. */
1804 pthread_join(self
->sibling
[0].tid
, &status
);
1805 EXPECT_EQ(SIBLING_EXIT_FAILURE
, (long)status
);
1806 pthread_join(self
->sibling
[1].tid
, &status
);
1807 EXPECT_EQ(SIBLING_EXIT_UNKILLED
, (long)status
);
1810 TEST_F(TSYNC
, two_siblings_with_ancestor
)
1815 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0)) {
1816 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1819 ret
= seccomp(SECCOMP_SET_MODE_FILTER
, 0, &self
->root_prog
);
1820 ASSERT_NE(ENOSYS
, errno
) {
1821 TH_LOG("Kernel does not support seccomp syscall!");
1824 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
1826 tsync_start_sibling(&self
->sibling
[0]);
1827 tsync_start_sibling(&self
->sibling
[1]);
1829 while (self
->sibling_count
< TSYNC_SIBLINGS
) {
1830 sem_wait(&self
->started
);
1831 self
->sibling_count
++;
1834 ret
= seccomp(SECCOMP_SET_MODE_FILTER
, SECCOMP_FILTER_FLAG_TSYNC
,
1837 TH_LOG("Could install filter on all threads!");
1839 /* Tell the siblings to test the policy */
1840 pthread_mutex_lock(&self
->mutex
);
1841 ASSERT_EQ(0, pthread_cond_broadcast(&self
->cond
)) {
1842 TH_LOG("cond broadcast non-zero");
1844 pthread_mutex_unlock(&self
->mutex
);
1845 /* Ensure they are both killed and don't exit cleanly. */
1846 pthread_join(self
->sibling
[0].tid
, &status
);
1847 EXPECT_EQ(0x0, (long)status
);
1848 pthread_join(self
->sibling
[1].tid
, &status
);
1849 EXPECT_EQ(0x0, (long)status
);
1852 TEST_F(TSYNC
, two_sibling_want_nnp
)
1856 /* start siblings before any prctl() operations */
1857 tsync_start_sibling(&self
->sibling
[0]);
1858 tsync_start_sibling(&self
->sibling
[1]);
1859 while (self
->sibling_count
< TSYNC_SIBLINGS
) {
1860 sem_wait(&self
->started
);
1861 self
->sibling_count
++;
1864 /* Tell the siblings to test no policy */
1865 pthread_mutex_lock(&self
->mutex
);
1866 ASSERT_EQ(0, pthread_cond_broadcast(&self
->cond
)) {
1867 TH_LOG("cond broadcast non-zero");
1869 pthread_mutex_unlock(&self
->mutex
);
1871 /* Ensure they are both upset about lacking nnp. */
1872 pthread_join(self
->sibling
[0].tid
, &status
);
1873 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS
, (long)status
);
1874 pthread_join(self
->sibling
[1].tid
, &status
);
1875 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS
, (long)status
);
1878 TEST_F(TSYNC
, two_siblings_with_no_filter
)
1883 /* start siblings before any prctl() operations */
1884 tsync_start_sibling(&self
->sibling
[0]);
1885 tsync_start_sibling(&self
->sibling
[1]);
1886 while (self
->sibling_count
< TSYNC_SIBLINGS
) {
1887 sem_wait(&self
->started
);
1888 self
->sibling_count
++;
1891 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0)) {
1892 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1895 ret
= seccomp(SECCOMP_SET_MODE_FILTER
, SECCOMP_FILTER_FLAG_TSYNC
,
1897 ASSERT_NE(ENOSYS
, errno
) {
1898 TH_LOG("Kernel does not support seccomp syscall!");
1901 TH_LOG("Could install filter on all threads!");
1904 /* Tell the siblings to test the policy */
1905 pthread_mutex_lock(&self
->mutex
);
1906 ASSERT_EQ(0, pthread_cond_broadcast(&self
->cond
)) {
1907 TH_LOG("cond broadcast non-zero");
1909 pthread_mutex_unlock(&self
->mutex
);
1911 /* Ensure they are both killed and don't exit cleanly. */
1912 pthread_join(self
->sibling
[0].tid
, &status
);
1913 EXPECT_EQ(0x0, (long)status
);
1914 pthread_join(self
->sibling
[1].tid
, &status
);
1915 EXPECT_EQ(0x0, (long)status
);
1918 TEST_F(TSYNC
, two_siblings_with_one_divergence
)
1923 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0)) {
1924 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1927 ret
= seccomp(SECCOMP_SET_MODE_FILTER
, 0, &self
->root_prog
);
1928 ASSERT_NE(ENOSYS
, errno
) {
1929 TH_LOG("Kernel does not support seccomp syscall!");
1932 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
1934 self
->sibling
[0].diverge
= 1;
1935 tsync_start_sibling(&self
->sibling
[0]);
1936 tsync_start_sibling(&self
->sibling
[1]);
1938 while (self
->sibling_count
< TSYNC_SIBLINGS
) {
1939 sem_wait(&self
->started
);
1940 self
->sibling_count
++;
1943 ret
= seccomp(SECCOMP_SET_MODE_FILTER
, SECCOMP_FILTER_FLAG_TSYNC
,
1945 ASSERT_EQ(self
->sibling
[0].system_tid
, ret
) {
1946 TH_LOG("Did not fail on diverged sibling.");
1949 /* Wake the threads */
1950 pthread_mutex_lock(&self
->mutex
);
1951 ASSERT_EQ(0, pthread_cond_broadcast(&self
->cond
)) {
1952 TH_LOG("cond broadcast non-zero");
1954 pthread_mutex_unlock(&self
->mutex
);
1956 /* Ensure they are both unkilled. */
1957 pthread_join(self
->sibling
[0].tid
, &status
);
1958 EXPECT_EQ(SIBLING_EXIT_UNKILLED
, (long)status
);
1959 pthread_join(self
->sibling
[1].tid
, &status
);
1960 EXPECT_EQ(SIBLING_EXIT_UNKILLED
, (long)status
);
1963 TEST_F(TSYNC
, two_siblings_not_under_filter
)
1968 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0)) {
1969 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1973 * Sibling 0 will have its own seccomp policy
1974 * and Sibling 1 will not be under seccomp at
1975 * all. Sibling 1 will enter seccomp and 0
1976 * will cause failure.
1978 self
->sibling
[0].diverge
= 1;
1979 tsync_start_sibling(&self
->sibling
[0]);
1980 tsync_start_sibling(&self
->sibling
[1]);
1982 while (self
->sibling_count
< TSYNC_SIBLINGS
) {
1983 sem_wait(&self
->started
);
1984 self
->sibling_count
++;
1987 ret
= seccomp(SECCOMP_SET_MODE_FILTER
, 0, &self
->root_prog
);
1988 ASSERT_NE(ENOSYS
, errno
) {
1989 TH_LOG("Kernel does not support seccomp syscall!");
1992 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
1995 ret
= seccomp(SECCOMP_SET_MODE_FILTER
, SECCOMP_FILTER_FLAG_TSYNC
,
1997 ASSERT_EQ(ret
, self
->sibling
[0].system_tid
) {
1998 TH_LOG("Did not fail on diverged sibling.");
2001 if (ret
== self
->sibling
[0].system_tid
)
2004 pthread_mutex_lock(&self
->mutex
);
2006 /* Increment the other siblings num_waits so we can clean up
2007 * the one we just saw.
2009 self
->sibling
[!sib
].num_waits
+= 1;
2011 /* Signal the thread to clean up*/
2012 ASSERT_EQ(0, pthread_cond_broadcast(&self
->cond
)) {
2013 TH_LOG("cond broadcast non-zero");
2015 pthread_mutex_unlock(&self
->mutex
);
2016 pthread_join(self
->sibling
[sib
].tid
, &status
);
2017 EXPECT_EQ(SIBLING_EXIT_UNKILLED
, (long)status
);
2018 /* Poll for actual task death. pthread_join doesn't guarantee it. */
2019 while (!kill(self
->sibling
[sib
].system_tid
, 0))
2021 /* Switch to the remaining sibling */
2024 ret
= seccomp(SECCOMP_SET_MODE_FILTER
, SECCOMP_FILTER_FLAG_TSYNC
,
2027 TH_LOG("Expected the remaining sibling to sync");
2030 pthread_mutex_lock(&self
->mutex
);
2032 /* If remaining sibling didn't have a chance to wake up during
2033 * the first broadcast, manually reduce the num_waits now.
2035 if (self
->sibling
[sib
].num_waits
> 1)
2036 self
->sibling
[sib
].num_waits
= 1;
2037 ASSERT_EQ(0, pthread_cond_broadcast(&self
->cond
)) {
2038 TH_LOG("cond broadcast non-zero");
2040 pthread_mutex_unlock(&self
->mutex
);
2041 pthread_join(self
->sibling
[sib
].tid
, &status
);
2042 EXPECT_EQ(0, (long)status
);
2043 /* Poll for actual task death. pthread_join doesn't guarantee it. */
2044 while (!kill(self
->sibling
[sib
].system_tid
, 0))
2047 ret
= seccomp(SECCOMP_SET_MODE_FILTER
, SECCOMP_FILTER_FLAG_TSYNC
,
2049 ASSERT_EQ(0, ret
); /* just us chickens */
2052 /* Make sure restarted syscalls are seen directly as "restart_syscall". */
2053 TEST(syscall_restart
)
2060 siginfo_t info
= { };
2061 struct sock_filter filter
[] = {
2062 BPF_STMT(BPF_LD
|BPF_W
|BPF_ABS
,
2063 offsetof(struct seccomp_data
, nr
)),
2065 #ifdef __NR_sigreturn
2066 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_sigreturn
, 6, 0),
2068 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_read
, 5, 0),
2069 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_exit
, 4, 0),
2070 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_rt_sigreturn
, 3, 0),
2071 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_nanosleep
, 4, 0),
2072 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_restart_syscall
, 4, 0),
2074 /* Allow __NR_write for easy logging. */
2075 BPF_JUMP(BPF_JMP
|BPF_JEQ
|BPF_K
, __NR_write
, 0, 1),
2076 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_ALLOW
),
2077 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_KILL
),
2078 /* The nanosleep jump target. */
2079 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_TRACE
|0x100),
2080 /* The restart_syscall jump target. */
2081 BPF_STMT(BPF_RET
|BPF_K
, SECCOMP_RET_TRACE
|0x200),
2083 struct sock_fprog prog
= {
2084 .len
= (unsigned short)ARRAY_SIZE(filter
),
2087 #if defined(__arm__)
2088 struct utsname utsbuf
;
2091 ASSERT_EQ(0, pipe(pipefd
));
2094 ASSERT_LE(0, child_pid
);
2095 if (child_pid
== 0) {
2096 /* Child uses EXPECT not ASSERT to deliver status correctly. */
2098 struct timespec timeout
= { };
2100 /* Attach parent as tracer and stop. */
2101 EXPECT_EQ(0, ptrace(PTRACE_TRACEME
));
2102 EXPECT_EQ(0, raise(SIGSTOP
));
2104 EXPECT_EQ(0, close(pipefd
[1]));
2106 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0)) {
2107 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2110 ret
= prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, &prog
, 0, 0);
2112 TH_LOG("Failed to install filter!");
2115 EXPECT_EQ(1, read(pipefd
[0], &buf
, 1)) {
2116 TH_LOG("Failed to read() sync from parent");
2118 EXPECT_EQ('.', buf
) {
2119 TH_LOG("Failed to get sync data from read()");
2122 /* Start nanosleep to be interrupted. */
2125 EXPECT_EQ(0, nanosleep(&timeout
, NULL
)) {
2126 TH_LOG("Call to nanosleep() failed (errno %d)", errno
);
2129 /* Read final sync from parent. */
2130 EXPECT_EQ(1, read(pipefd
[0], &buf
, 1)) {
2131 TH_LOG("Failed final read() from parent");
2133 EXPECT_EQ('!', buf
) {
2134 TH_LOG("Failed to get final data from read()");
2137 /* Directly report the status of our test harness results. */
2138 syscall(__NR_exit
, _metadata
->passed
? EXIT_SUCCESS
2141 EXPECT_EQ(0, close(pipefd
[0]));
2143 /* Attach to child, setup options, and release. */
2144 ASSERT_EQ(child_pid
, waitpid(child_pid
, &status
, 0));
2145 ASSERT_EQ(true, WIFSTOPPED(status
));
2146 ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS
, child_pid
, NULL
,
2147 PTRACE_O_TRACESECCOMP
));
2148 ASSERT_EQ(0, ptrace(PTRACE_CONT
, child_pid
, NULL
, 0));
2149 ASSERT_EQ(1, write(pipefd
[1], ".", 1));
2151 /* Wait for nanosleep() to start. */
2152 ASSERT_EQ(child_pid
, waitpid(child_pid
, &status
, 0));
2153 ASSERT_EQ(true, WIFSTOPPED(status
));
2154 ASSERT_EQ(SIGTRAP
, WSTOPSIG(status
));
2155 ASSERT_EQ(PTRACE_EVENT_SECCOMP
, (status
>> 16));
2156 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG
, child_pid
, NULL
, &msg
));
2157 ASSERT_EQ(0x100, msg
);
2158 EXPECT_EQ(__NR_nanosleep
, get_syscall(_metadata
, child_pid
));
2160 /* Might as well check siginfo for sanity while we're here. */
2161 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO
, child_pid
, NULL
, &info
));
2162 ASSERT_EQ(SIGTRAP
, info
.si_signo
);
2163 ASSERT_EQ(SIGTRAP
| (PTRACE_EVENT_SECCOMP
<< 8), info
.si_code
);
2164 EXPECT_EQ(0, info
.si_errno
);
2165 EXPECT_EQ(getuid(), info
.si_uid
);
2166 /* Verify signal delivery came from child (seccomp-triggered). */
2167 EXPECT_EQ(child_pid
, info
.si_pid
);
2169 /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */
2170 ASSERT_EQ(0, kill(child_pid
, SIGSTOP
));
2171 ASSERT_EQ(0, ptrace(PTRACE_CONT
, child_pid
, NULL
, 0));
2172 ASSERT_EQ(child_pid
, waitpid(child_pid
, &status
, 0));
2173 ASSERT_EQ(true, WIFSTOPPED(status
));
2174 ASSERT_EQ(SIGSTOP
, WSTOPSIG(status
));
2175 /* Verify signal delivery came from parent now. */
2176 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO
, child_pid
, NULL
, &info
));
2177 EXPECT_EQ(getpid(), info
.si_pid
);
2179 /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */
2180 ASSERT_EQ(0, kill(child_pid
, SIGCONT
));
2181 ASSERT_EQ(0, ptrace(PTRACE_CONT
, child_pid
, NULL
, 0));
2182 ASSERT_EQ(child_pid
, waitpid(child_pid
, &status
, 0));
2183 ASSERT_EQ(true, WIFSTOPPED(status
));
2184 ASSERT_EQ(SIGCONT
, WSTOPSIG(status
));
2185 ASSERT_EQ(0, ptrace(PTRACE_CONT
, child_pid
, NULL
, 0));
2187 /* Wait for restart_syscall() to start. */
2188 ASSERT_EQ(child_pid
, waitpid(child_pid
, &status
, 0));
2189 ASSERT_EQ(true, WIFSTOPPED(status
));
2190 ASSERT_EQ(SIGTRAP
, WSTOPSIG(status
));
2191 ASSERT_EQ(PTRACE_EVENT_SECCOMP
, (status
>> 16));
2192 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG
, child_pid
, NULL
, &msg
));
2194 ASSERT_EQ(0x200, msg
);
2195 ret
= get_syscall(_metadata
, child_pid
);
2196 #if defined(__arm__)
2199 * - native ARM registers do NOT expose true syscall.
2200 * - compat ARM registers on ARM64 DO expose true syscall.
2202 ASSERT_EQ(0, uname(&utsbuf
));
2203 if (strncmp(utsbuf
.machine
, "arm", 3) == 0) {
2204 EXPECT_EQ(__NR_nanosleep
, ret
);
2208 EXPECT_EQ(__NR_restart_syscall
, ret
);
2211 /* Write again to end test. */
2212 ASSERT_EQ(0, ptrace(PTRACE_CONT
, child_pid
, NULL
, 0));
2213 ASSERT_EQ(1, write(pipefd
[1], "!", 1));
2214 EXPECT_EQ(0, close(pipefd
[1]));
2216 ASSERT_EQ(child_pid
, waitpid(child_pid
, &status
, 0));
2217 if (WIFSIGNALED(status
) || WEXITSTATUS(status
))
2218 _metadata
->passed
= 0;
2223 * - add microbenchmarks
2224 * - expand NNP testing
2225 * - better arch-specific TRACE and TRAP handlers.
2226 * - endianness checking when appropriate
2227 * - 64-bit arg prodding
2228 * - arch value testing (x86 modes especially)