projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
seccomp: Fix ioctl number for SECCOMP_IOCTL_NOTIF_ID_VALID
[linux-block.git] / tools / testing / selftests / seccomp / seccomp_bpf.c
CommitLineData
e500db3f 1// SPDX-License-Identifier: GPL-2.0-only
c99ee51a
KC		 2/*
3 * Copyright (c) 2012 The Chromium OS Authors. All rights reserved.
c99ee51a
KC		 4 *
5 * Test code for seccomp bpf.
6 */
7
6a21cc50 8#define _GNU_SOURCE
0ce105bf 9#include <sys/types.h>
10859f38
KC		 10
11/*
12 * glibc 2.26 and later have SIGSYS in siginfo_t. Before that,
13 * we need to use the kernel's siginfo.h file and trick glibc
14 * into accepting it.
15 */
16#if !__GLIBC_PREREQ(2, 26)
17# include <asm/siginfo.h>
18# define __have_siginfo_t 1
19# define __have_sigval_t 1
20# define __have_sigevent_t 1
21#endif
c99ee51a
KC		 22
23#include <errno.h>
24#include <linux/filter.h>
25#include <sys/prctl.h>
26#include <sys/ptrace.h>
27#include <sys/user.h>
28#include <linux/prctl.h>
29#include <linux/ptrace.h>
30#include <linux/seccomp.h>
c99ee51a
KC		 31#include <pthread.h>
32#include <semaphore.h>
33#include <signal.h>
34#include <stddef.h>
35#include <stdbool.h>
36#include <string.h>
256d0afb 37#include <time.h>
223e660b 38#include <limits.h>
c99ee51a
KC		 39#include <linux/elf.h>
40#include <sys/uio.h>
256d0afb 41#include <sys/utsname.h>
fd88d16c
RS		 42#include <sys/fcntl.h>
43#include <sys/mman.h>
44#include <sys/times.h>
6a21cc50
TA		 45#include <sys/socket.h>
46#include <sys/ioctl.h>
0eebfed2 47#include <linux/kcmp.h>
c99ee51a 48
c99ee51a
KC		 49#include <unistd.h>
50#include <sys/syscall.h>
6a21cc50 51#include <poll.h>
c99ee51a 52
0b40808a 53#include "../kselftest_harness.h"
ad568218 54#include "../clone3/clone3_selftests.h"
c99ee51a 55
8b1bc88c
KC		 56/* Attempt to de-conflict with the selftests tree. */
57#ifndef SKIP
58#define SKIP(s, ...) XFAIL(s, ##__VA_ARGS__)
59#endif
60
c99ee51a
KC		 61#ifndef PR_SET_PTRACER
62# define PR_SET_PTRACER 0x59616d61
63#endif
64
65#ifndef PR_SET_NO_NEW_PRIVS
66#define PR_SET_NO_NEW_PRIVS 38
67#define PR_GET_NO_NEW_PRIVS 39
68#endif
69
70#ifndef PR_SECCOMP_EXT
71#define PR_SECCOMP_EXT 43
72#endif
73
74#ifndef SECCOMP_EXT_ACT
75#define SECCOMP_EXT_ACT 1
76#endif
77
78#ifndef SECCOMP_EXT_ACT_TSYNC
79#define SECCOMP_EXT_ACT_TSYNC 1
80#endif
81
82#ifndef SECCOMP_MODE_STRICT
83#define SECCOMP_MODE_STRICT 1
84#endif
85
86#ifndef SECCOMP_MODE_FILTER
87#define SECCOMP_MODE_FILTER 2
88#endif
89
f3e1821d 90#ifndef SECCOMP_RET_ALLOW
c99ee51a
KC		 91struct seccomp_data {
92 int nr;
93 __u32 arch;
94 __u64 instruction_pointer;
95 __u64 args[6];
96};
97#endif
98
f3e1821d
KC		 99#ifndef SECCOMP_RET_KILL_PROCESS
100#define SECCOMP_RET_KILL_PROCESS 0x80000000U /* kill the process */
fd76875c
KC		 101#define SECCOMP_RET_KILL_THREAD 0x00000000U /* kill the thread */
102#endif
c99ee51a 103#ifndef SECCOMP_RET_KILL
fd76875c
KC		 104#define SECCOMP_RET_KILL SECCOMP_RET_KILL_THREAD
105#define SECCOMP_RET_TRAP 0x00030000U /* disallow and force a SIGSYS */
106#define SECCOMP_RET_ERRNO 0x00050000U /* returns an errno */
107#define SECCOMP_RET_TRACE 0x7ff00000U /* pass to a tracer or disallow */
108#define SECCOMP_RET_ALLOW 0x7fff0000U /* allow */
59f5cf44
TH		 109#endif
110#ifndef SECCOMP_RET_LOG
fd76875c 111#define SECCOMP_RET_LOG 0x7ffc0000U /* allow after logging */
59f5cf44 112#endif
c99ee51a 113
f3e1821d
KC		 114#ifndef __NR_seccomp
115# if defined(__i386__)
116# define __NR_seccomp 354
117# elif defined(__x86_64__)
118# define __NR_seccomp 317
119# elif defined(__arm__)
120# define __NR_seccomp 383
121# elif defined(__aarch64__)
122# define __NR_seccomp 277
5340627e
DA		 123# elif defined(__riscv)
124# define __NR_seccomp 277
f3e1821d
KC		 125# elif defined(__hppa__)
126# define __NR_seccomp 338
127# elif defined(__powerpc__)
128# define __NR_seccomp 358
129# elif defined(__s390__)
130# define __NR_seccomp 348
131# else
132# warning "seccomp syscall number unknown for this architecture"
133# define __NR_seccomp 0xffff
134# endif
135#endif
c99ee51a 136
f3e1821d
KC		 137#ifndef SECCOMP_SET_MODE_STRICT
138#define SECCOMP_SET_MODE_STRICT 0
139#endif
140
141#ifndef SECCOMP_SET_MODE_FILTER
142#define SECCOMP_SET_MODE_FILTER 1
143#endif
144
145#ifndef SECCOMP_GET_ACTION_AVAIL
146#define SECCOMP_GET_ACTION_AVAIL 2
147#endif
148
6a21cc50
TA		 149#ifndef SECCOMP_GET_NOTIF_SIZES
150#define SECCOMP_GET_NOTIF_SIZES 3
151#endif
152
f3e1821d 153#ifndef SECCOMP_FILTER_FLAG_TSYNC
00a02d0c 154#define SECCOMP_FILTER_FLAG_TSYNC (1UL << 0)
f3e1821d
KC		 155#endif
156
157#ifndef SECCOMP_FILTER_FLAG_LOG
00a02d0c
KC		 158#define SECCOMP_FILTER_FLAG_LOG (1UL << 1)
159#endif
160
161#ifndef SECCOMP_FILTER_FLAG_SPEC_ALLOW
162#define SECCOMP_FILTER_FLAG_SPEC_ALLOW (1UL << 2)
f3e1821d
KC		 163#endif
164
d057dc4e
TA		 165#ifndef PTRACE_SECCOMP_GET_METADATA
166#define PTRACE_SECCOMP_GET_METADATA 0x420d
167
168struct seccomp_metadata {
169 __u64 filter_off; /* Input: which filter */
170 __u64 flags; /* Output: filter's flags */
171};
172#endif
173
6a21cc50
TA		 174#ifndef SECCOMP_FILTER_FLAG_NEW_LISTENER
175#define SECCOMP_FILTER_FLAG_NEW_LISTENER (1UL << 3)
176
177#define SECCOMP_RET_USER_NOTIF 0x7fc00000U
178
179#define SECCOMP_IOC_MAGIC '!'
180#define SECCOMP_IO(nr) _IO(SECCOMP_IOC_MAGIC, nr)
181#define SECCOMP_IOR(nr, type) _IOR(SECCOMP_IOC_MAGIC, nr, type)
182#define SECCOMP_IOW(nr, type) _IOW(SECCOMP_IOC_MAGIC, nr, type)
183#define SECCOMP_IOWR(nr, type) _IOWR(SECCOMP_IOC_MAGIC, nr, type)
184
185/* Flags for seccomp notification fd ioctl. */
186#define SECCOMP_IOCTL_NOTIF_RECV SECCOMP_IOWR(0, struct seccomp_notif)
187#define SECCOMP_IOCTL_NOTIF_SEND SECCOMP_IOWR(1, \
188 struct seccomp_notif_resp)
47e33c05 189#define SECCOMP_IOCTL_NOTIF_ID_VALID SECCOMP_IOW(2, __u64)
6a21cc50
TA		 190
191struct seccomp_notif {
192 __u64 id;
193 __u32 pid;
194 __u32 flags;
195 struct seccomp_data data;
196};
197
198struct seccomp_notif_resp {
199 __u64 id;
200 __s64 val;
201 __s32 error;
202 __u32 flags;
203};
204
205struct seccomp_notif_sizes {
206 __u16 seccomp_notif;
207 __u16 seccomp_notif_resp;
208 __u16 seccomp_data;
209};
210#endif
211
88282297
TA		 212#ifndef PTRACE_EVENTMSG_SYSCALL_ENTRY
213#define PTRACE_EVENTMSG_SYSCALL_ENTRY 1
214#define PTRACE_EVENTMSG_SYSCALL_EXIT 2
215#endif
216
2aa8d8d0
CB		 217#ifndef SECCOMP_USER_NOTIF_FLAG_CONTINUE
218#define SECCOMP_USER_NOTIF_FLAG_CONTINUE 0x00000001
219#endif
220
51891498
TA		 221#ifndef SECCOMP_FILTER_FLAG_TSYNC_ESRCH
222#define SECCOMP_FILTER_FLAG_TSYNC_ESRCH (1UL << 4)
223#endif
224
f3e1821d
KC		 225#ifndef seccomp
226int seccomp(unsigned int op, unsigned int flags, void *args)
227{
228 errno = 0;
229 return syscall(__NR_seccomp, op, flags, args);
230}
c99ee51a
KC		 231#endif
232
c385d0db 233#if __BYTE_ORDER == __LITTLE_ENDIAN
c99ee51a 234#define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]))
c385d0db
ME		 235#elif __BYTE_ORDER == __BIG_ENDIAN
236#define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32))
237#else
238#error "wut? Unknown __BYTE_ORDER?!"
239#endif
c99ee51a
KC		 240
241#define SIBLING_EXIT_UNKILLED 0xbadbeef
242#define SIBLING_EXIT_FAILURE 0xbadface
243#define SIBLING_EXIT_NEWPRIVS 0xbadfeed
244
cf8918db
KC		 245static int __filecmp(pid_t pid1, pid_t pid2, int fd1, int fd2)
246{
247#ifdef __NR_kcmp
248 errno = 0;
249 return syscall(__NR_kcmp, pid1, pid2, KCMP_FILE, fd1, fd2);
250#else
251 errno = ENOSYS;
252 return -1;
253#endif
254}
255
256/* Have TH_LOG report actual location filecmp() is used. */
257#define filecmp(pid1, pid2, fd1, fd2) ({ \
258 int _ret; \
259 \
260 _ret = __filecmp(pid1, pid2, fd1, fd2); \
261 if (_ret != 0) { \
262 if (_ret < 0 && errno == ENOSYS) { \
263 TH_LOG("kcmp() syscall missing (test is less accurate)");\
264 _ret = 0; \
265 } \
266 } \
267 _ret; })
268
269TEST(kcmp)
270{
271 int ret;
272
273 ret = __filecmp(getpid(), getpid(), 1, 1);
274 EXPECT_EQ(ret, 0);
275 if (ret != 0 && errno == ENOSYS)
276 SKIP(return, "Kernel does not support kcmp() (missing CONFIG_CHECKPOINT_RESTORE?)");
277}
278
c99ee51a
KC		 279TEST(mode_strict_support)
280{
281 long ret;
282
283 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
284 ASSERT_EQ(0, ret) {
285 TH_LOG("Kernel does not support CONFIG_SECCOMP");
286 }
369130b6 287 syscall(__NR_exit, 0);
c99ee51a
KC		 288}
289
290TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL)
291{
292 long ret;
293
294 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
295 ASSERT_EQ(0, ret) {
296 TH_LOG("Kernel does not support CONFIG_SECCOMP");
297 }
298 syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
299 NULL, NULL, NULL);
300 EXPECT_FALSE(true) {
301 TH_LOG("Unreachable!");
302 }
303}
304
305/* Note! This doesn't test no new privs behavior */
306TEST(no_new_privs_support)
307{
308 long ret;
309
310 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
311 EXPECT_EQ(0, ret) {
312 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
313 }
314}
315
f3f6e306 316/* Tests kernel support by checking for a copy_from_user() fault on NULL. */
c99ee51a
KC		 317TEST(mode_filter_support)
318{
319 long ret;
320
321 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
322 ASSERT_EQ(0, ret) {
323 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
324 }
325 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL);
326 EXPECT_EQ(-1, ret);
327 EXPECT_EQ(EFAULT, errno) {
328 TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!");
329 }
330}
331
332TEST(mode_filter_without_nnp)
333{
334 struct sock_filter filter[] = {
335 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
336 };
337 struct sock_fprog prog = {
338 .len = (unsigned short)ARRAY_SIZE(filter),
339 .filter = filter,
340 };
341 long ret;
342
343 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0);
344 ASSERT_LE(0, ret) {
345 TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS");
346 }
347 errno = 0;
348 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
349 /* Succeeds with CAP_SYS_ADMIN, fails without */
350 /* TODO(wad) check caps not euid */
351 if (geteuid()) {
352 EXPECT_EQ(-1, ret);
353 EXPECT_EQ(EACCES, errno);
354 } else {
355 EXPECT_EQ(0, ret);
356 }
357}
358
359#define MAX_INSNS_PER_PATH 32768
360
361TEST(filter_size_limits)
362{
363 int i;
364 int count = BPF_MAXINSNS + 1;
365 struct sock_filter allow[] = {
366 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
367 };
368 struct sock_filter *filter;
369 struct sock_fprog prog = { };
370 long ret;
371
372 filter = calloc(count, sizeof(*filter));
373 ASSERT_NE(NULL, filter);
374
375 for (i = 0; i < count; i++)
376 filter[i] = allow[0];
377
378 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
379 ASSERT_EQ(0, ret);
380
381 prog.filter = filter;
382 prog.len = count;
383
384 /* Too many filter instructions in a single filter. */
385 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
386 ASSERT_NE(0, ret) {
387 TH_LOG("Installing %d insn filter was allowed", prog.len);
388 }
389
390 /* One less is okay, though. */
391 prog.len -= 1;
392 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
393 ASSERT_EQ(0, ret) {
394 TH_LOG("Installing %d insn filter wasn't allowed", prog.len);
395 }
396}
397
398TEST(filter_chain_limits)
399{
400 int i;
401 int count = BPF_MAXINSNS;
402 struct sock_filter allow[] = {
403 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
404 };
405 struct sock_filter *filter;
406 struct sock_fprog prog = { };
407 long ret;
408
409 filter = calloc(count, sizeof(*filter));
410 ASSERT_NE(NULL, filter);
411
412 for (i = 0; i < count; i++)
413 filter[i] = allow[0];
414
415 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
416 ASSERT_EQ(0, ret);
417
418 prog.filter = filter;
419 prog.len = 1;
420
421 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
422 ASSERT_EQ(0, ret);
423
424 prog.len = count;
425
426 /* Too many total filter instructions. */
427 for (i = 0; i < MAX_INSNS_PER_PATH; i++) {
428 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
429 if (ret != 0)
430 break;
431 }
432 ASSERT_NE(0, ret) {
433 TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)",
434 i, count, i * (count + 4));
435 }
436}
437
438TEST(mode_filter_cannot_move_to_strict)
439{
440 struct sock_filter filter[] = {
441 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
442 };
443 struct sock_fprog prog = {
444 .len = (unsigned short)ARRAY_SIZE(filter),
445 .filter = filter,
446 };
447 long ret;
448
449 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
450 ASSERT_EQ(0, ret);
451
452 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
453 ASSERT_EQ(0, ret);
454
455 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0);
456 EXPECT_EQ(-1, ret);
457 EXPECT_EQ(EINVAL, errno);
458}
459
460
461TEST(mode_filter_get_seccomp)
462{
463 struct sock_filter filter[] = {
464 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
465 };
466 struct sock_fprog prog = {
467 .len = (unsigned short)ARRAY_SIZE(filter),
468 .filter = filter,
469 };
470 long ret;
471
472 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
473 ASSERT_EQ(0, ret);
474
475 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
476 EXPECT_EQ(0, ret);
477
478 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
479 ASSERT_EQ(0, ret);
480
481 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
482 EXPECT_EQ(2, ret);
483}
484
485
486TEST(ALLOW_all)
487{
488 struct sock_filter filter[] = {
489 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
490 };
491 struct sock_fprog prog = {
492 .len = (unsigned short)ARRAY_SIZE(filter),
493 .filter = filter,
494 };
495 long ret;
496
497 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
498 ASSERT_EQ(0, ret);
499
500 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
501 ASSERT_EQ(0, ret);
502}
503
504TEST(empty_prog)
505{
506 struct sock_filter filter[] = {
507 };
508 struct sock_fprog prog = {
509 .len = (unsigned short)ARRAY_SIZE(filter),
510 .filter = filter,
511 };
512 long ret;
513
514 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
515 ASSERT_EQ(0, ret);
516
517 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
518 EXPECT_EQ(-1, ret);
519 EXPECT_EQ(EINVAL, errno);
520}
521
59f5cf44
TH		 522TEST(log_all)
523{
524 struct sock_filter filter[] = {
525 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG),
526 };
527 struct sock_fprog prog = {
528 .len = (unsigned short)ARRAY_SIZE(filter),
529 .filter = filter,
530 };
531 long ret;
532 pid_t parent = getppid();
533
534 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
535 ASSERT_EQ(0, ret);
536
537 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
538 ASSERT_EQ(0, ret);
539
540 /* getppid() should succeed and be logged (no check for logging) */
541 EXPECT_EQ(parent, syscall(__NR_getppid));
542}
543
c99ee51a
KC		 544TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS)
545{
546 struct sock_filter filter[] = {
547 BPF_STMT(BPF_RET|BPF_K, 0x10000000U),
548 };
549 struct sock_fprog prog = {
550 .len = (unsigned short)ARRAY_SIZE(filter),
551 .filter = filter,
552 };
553 long ret;
554
555 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
556 ASSERT_EQ(0, ret);
557
558 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
559 ASSERT_EQ(0, ret);
560 EXPECT_EQ(0, syscall(__NR_getpid)) {
561 TH_LOG("getpid() shouldn't ever return");
562 }
563}
564
565/* return code >= 0x80000000 is unused. */
566TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS)
567{
568 struct sock_filter filter[] = {
569 BPF_STMT(BPF_RET|BPF_K, 0x90000000U),
570 };
571 struct sock_fprog prog = {
572 .len = (unsigned short)ARRAY_SIZE(filter),
573 .filter = filter,
574 };
575 long ret;
576
577 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
578 ASSERT_EQ(0, ret);
579
580 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
581 ASSERT_EQ(0, ret);
582 EXPECT_EQ(0, syscall(__NR_getpid)) {
583 TH_LOG("getpid() shouldn't ever return");
584 }
585}
586
587TEST_SIGNAL(KILL_all, SIGSYS)
588{
589 struct sock_filter filter[] = {
590 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
591 };
592 struct sock_fprog prog = {
593 .len = (unsigned short)ARRAY_SIZE(filter),
594 .filter = filter,
595 };
596 long ret;
597
598 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
599 ASSERT_EQ(0, ret);
600
601 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
602 ASSERT_EQ(0, ret);
603}
604
605TEST_SIGNAL(KILL_one, SIGSYS)
606{
607 struct sock_filter filter[] = {
608 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
609 offsetof(struct seccomp_data, nr)),
610 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
611 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
612 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
613 };
614 struct sock_fprog prog = {
615 .len = (unsigned short)ARRAY_SIZE(filter),
616 .filter = filter,
617 };
618 long ret;
619 pid_t parent = getppid();
620
621 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
622 ASSERT_EQ(0, ret);
623
624 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
625 ASSERT_EQ(0, ret);
626
627 EXPECT_EQ(parent, syscall(__NR_getppid));
628 /* getpid() should never return. */
629 EXPECT_EQ(0, syscall(__NR_getpid));
630}
631
632TEST_SIGNAL(KILL_one_arg_one, SIGSYS)
633{
fd88d16c 634 void *fatal_address;
c99ee51a
KC		 635 struct sock_filter filter[] = {
636 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
637 offsetof(struct seccomp_data, nr)),
fd88d16c 638 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0),
c99ee51a
KC		 639 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
640 /* Only both with lower 32-bit for now. */
641 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)),
fd88d16c
RS		 642 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K,
643 (unsigned long)&fatal_address, 0, 1),
c99ee51a
KC		 644 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
645 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
646 };
647 struct sock_fprog prog = {
648 .len = (unsigned short)ARRAY_SIZE(filter),
649 .filter = filter,
650 };
651 long ret;
652 pid_t parent = getppid();
fd88d16c
RS		 653 struct tms timebuf;
654 clock_t clock = times(&timebuf);
c99ee51a
KC		 655
656 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
657 ASSERT_EQ(0, ret);
658
659 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
660 ASSERT_EQ(0, ret);
661
662 EXPECT_EQ(parent, syscall(__NR_getppid));
fd88d16c
RS		 663 EXPECT_LE(clock, syscall(__NR_times, &timebuf));
664 /* times() should never return. */
665 EXPECT_EQ(0, syscall(__NR_times, &fatal_address));
c99ee51a
KC		 666}
667
668TEST_SIGNAL(KILL_one_arg_six, SIGSYS)
669{
fd88d16c
RS		 670#ifndef __NR_mmap2
671 int sysno = __NR_mmap;
672#else
673 int sysno = __NR_mmap2;
674#endif
c99ee51a
KC		 675 struct sock_filter filter[] = {
676 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
677 offsetof(struct seccomp_data, nr)),
fd88d16c 678 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0),
c99ee51a
KC		 679 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
680 /* Only both with lower 32-bit for now. */
681 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)),
682 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1),
683 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
684 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
685 };
686 struct sock_fprog prog = {
687 .len = (unsigned short)ARRAY_SIZE(filter),
688 .filter = filter,
689 };
690 long ret;
691 pid_t parent = getppid();
fd88d16c
RS		 692 int fd;
693 void *map1, *map2;
2ce47b44
BJZ		 694 int page_size = sysconf(_SC_PAGESIZE);
695
696 ASSERT_LT(0, page_size);
c99ee51a
KC		 697
698 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
699 ASSERT_EQ(0, ret);
700
701 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
702 ASSERT_EQ(0, ret);
703
fd88d16c
RS		 704 fd = open("/dev/zero", O_RDONLY);
705 ASSERT_NE(-1, fd);
706
c99ee51a 707 EXPECT_EQ(parent, syscall(__NR_getppid));
fd88d16c 708 map1 = (void *)syscall(sysno,
2ce47b44 709 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size);
fd88d16c
RS		 710 EXPECT_NE(MAP_FAILED, map1);
711 /* mmap2() should never return. */
712 map2 = (void *)syscall(sysno,
2ce47b44 713 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE);
fd88d16c
RS		 714 EXPECT_EQ(MAP_FAILED, map2);
715
716 /* The test failed, so clean up the resources. */
2ce47b44
BJZ		 717 munmap(map1, page_size);
718 munmap(map2, page_size);
fd88d16c 719 close(fd);
c99ee51a
KC		 720}
721
f3e1821d
KC		 722/* This is a thread task to die via seccomp filter violation. */
723void *kill_thread(void *data)
724{
725 bool die = (bool)data;
726
727 if (die) {
728 prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
729 return (void *)SIBLING_EXIT_FAILURE;
730 }
731
732 return (void *)SIBLING_EXIT_UNKILLED;
733}
734
735/* Prepare a thread that will kill itself or both of us. */
736void kill_thread_or_group(struct __test_metadata *_metadata, bool kill_process)
737{
738 pthread_t thread;
739 void *status;
740 /* Kill only when calling __NR_prctl. */
741 struct sock_filter filter_thread[] = {
742 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
743 offsetof(struct seccomp_data, nr)),
744 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
745 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD),
746 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
747 };
748 struct sock_fprog prog_thread = {
749 .len = (unsigned short)ARRAY_SIZE(filter_thread),
750 .filter = filter_thread,
751 };
752 struct sock_filter filter_process[] = {
753 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
754 offsetof(struct seccomp_data, nr)),
755 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
756 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_PROCESS),
757 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
758 };
759 struct sock_fprog prog_process = {
760 .len = (unsigned short)ARRAY_SIZE(filter_process),
761 .filter = filter_process,
762 };
763
764 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
765 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
766 }
767
768 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0,
769 kill_process ? &prog_process : &prog_thread));
770
771 /*
772 * Add the KILL_THREAD rule again to make sure that the KILL_PROCESS
773 * flag cannot be downgraded by a new filter.
774 */
775 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog_thread));
776
777 /* Start a thread that will exit immediately. */
778 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)false));
779 ASSERT_EQ(0, pthread_join(thread, &status));
780 ASSERT_EQ(SIBLING_EXIT_UNKILLED, (unsigned long)status);
781
782 /* Start a thread that will die immediately. */
783 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)true));
784 ASSERT_EQ(0, pthread_join(thread, &status));
785 ASSERT_NE(SIBLING_EXIT_FAILURE, (unsigned long)status);
786
787 /*
788 * If we get here, only the spawned thread died. Let the parent know
789 * the whole process didn't die (i.e. this thread, the spawner,
790 * stayed running).
791 */
792 exit(42);
793}
794
795TEST(KILL_thread)
796{
797 int status;
798 pid_t child_pid;
799
800 child_pid = fork();
801 ASSERT_LE(0, child_pid);
802 if (child_pid == 0) {
803 kill_thread_or_group(_metadata, false);
804 _exit(38);
805 }
806
807 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
808
809 /* If only the thread was killed, we'll see exit 42. */
810 ASSERT_TRUE(WIFEXITED(status));
811 ASSERT_EQ(42, WEXITSTATUS(status));
812}
813
814TEST(KILL_process)
815{
816 int status;
817 pid_t child_pid;
818
819 child_pid = fork();
820 ASSERT_LE(0, child_pid);
821 if (child_pid == 0) {
822 kill_thread_or_group(_metadata, true);
823 _exit(38);
824 }
825
826 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
827
828 /* If the entire process was killed, we'll see SIGSYS. */
829 ASSERT_TRUE(WIFSIGNALED(status));
830 ASSERT_EQ(SIGSYS, WTERMSIG(status));
831}
832
c99ee51a
KC		 833/* TODO(wad) add 64-bit versus 32-bit arg tests. */
834TEST(arg_out_of_range)
835{
836 struct sock_filter filter[] = {
837 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)),
838 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
839 };
840 struct sock_fprog prog = {
841 .len = (unsigned short)ARRAY_SIZE(filter),
842 .filter = filter,
843 };
844 long ret;
845
846 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
847 ASSERT_EQ(0, ret);
848
849 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
850 EXPECT_EQ(-1, ret);
851 EXPECT_EQ(EINVAL, errno);
852}
853
f3f6e306
KC		 854#define ERRNO_FILTER(name, errno) \
855 struct sock_filter _read_filter_##name[] = { \
856 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, \
857 offsetof(struct seccomp_data, nr)), \
858 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), \
859 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | errno), \
860 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), \
861 }; \
862 struct sock_fprog prog_##name = { \
863 .len = (unsigned short)ARRAY_SIZE(_read_filter_##name), \
864 .filter = _read_filter_##name, \
865 }
866
867/* Make sure basic errno values are correctly passed through a filter. */
c99ee51a
KC		 868TEST(ERRNO_valid)
869{
f3f6e306 870 ERRNO_FILTER(valid, E2BIG);
c99ee51a
KC		 871 long ret;
872 pid_t parent = getppid();
873
874 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
875 ASSERT_EQ(0, ret);
876
f3f6e306 877 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_valid);
c99ee51a
KC		 878 ASSERT_EQ(0, ret);
879
880 EXPECT_EQ(parent, syscall(__NR_getppid));
881 EXPECT_EQ(-1, read(0, NULL, 0));
882 EXPECT_EQ(E2BIG, errno);
883}
884
f3f6e306 885/* Make sure an errno of zero is correctly handled by the arch code. */
c99ee51a
KC		 886TEST(ERRNO_zero)
887{
f3f6e306 888 ERRNO_FILTER(zero, 0);
c99ee51a
KC		 889 long ret;
890 pid_t parent = getppid();
891
892 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
893 ASSERT_EQ(0, ret);
894
f3f6e306 895 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_zero);
c99ee51a
KC		 896 ASSERT_EQ(0, ret);
897
898 EXPECT_EQ(parent, syscall(__NR_getppid));
899 /* "errno" of 0 is ok. */
900 EXPECT_EQ(0, read(0, NULL, 0));
901}
902
f3f6e306
KC		 903/*
904 * The SECCOMP_RET_DATA mask is 16 bits wide, but errno is smaller.
905 * This tests that the errno value gets capped correctly, fixed by
906 * 580c57f10768 ("seccomp: cap SECCOMP_RET_ERRNO data to MAX_ERRNO").
907 */
c99ee51a
KC		 908TEST(ERRNO_capped)
909{
f3f6e306 910 ERRNO_FILTER(capped, 4096);
c99ee51a
KC		 911 long ret;
912 pid_t parent = getppid();
913
914 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
915 ASSERT_EQ(0, ret);
916
f3f6e306 917 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_capped);
c99ee51a
KC		 918 ASSERT_EQ(0, ret);
919
920 EXPECT_EQ(parent, syscall(__NR_getppid));
921 EXPECT_EQ(-1, read(0, NULL, 0));
922 EXPECT_EQ(4095, errno);
923}
924
f3f6e306
KC		 925/*
926 * Filters are processed in reverse order: last applied is executed first.
927 * Since only the SECCOMP_RET_ACTION mask is tested for return values, the
928 * SECCOMP_RET_DATA mask results will follow the most recently applied
929 * matching filter return (and not the lowest or highest value).
930 */
931TEST(ERRNO_order)
932{
933 ERRNO_FILTER(first, 11);
934 ERRNO_FILTER(second, 13);
935 ERRNO_FILTER(third, 12);
936 long ret;
937 pid_t parent = getppid();
938
939 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
940 ASSERT_EQ(0, ret);
941
942 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_first);
943 ASSERT_EQ(0, ret);
944
945 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_second);
946 ASSERT_EQ(0, ret);
947
948 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_third);
949 ASSERT_EQ(0, ret);
950
951 EXPECT_EQ(parent, syscall(__NR_getppid));
952 EXPECT_EQ(-1, read(0, NULL, 0));
953 EXPECT_EQ(12, errno);
954}
955
1ae81d78 956FIXTURE(TRAP) {
c99ee51a
KC		 957 struct sock_fprog prog;
958};
959
960FIXTURE_SETUP(TRAP)
961{
962 struct sock_filter filter[] = {
963 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
964 offsetof(struct seccomp_data, nr)),
965 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
966 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
967 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
968 };
969
970 memset(&self->prog, 0, sizeof(self->prog));
971 self->prog.filter = malloc(sizeof(filter));
972 ASSERT_NE(NULL, self->prog.filter);
973 memcpy(self->prog.filter, filter, sizeof(filter));
974 self->prog.len = (unsigned short)ARRAY_SIZE(filter);
975}
976
977FIXTURE_TEARDOWN(TRAP)
978{
979 if (self->prog.filter)
980 free(self->prog.filter);
981}
982
983TEST_F_SIGNAL(TRAP, dfl, SIGSYS)
984{
985 long ret;
986
987 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
988 ASSERT_EQ(0, ret);
989
990 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
991 ASSERT_EQ(0, ret);
992 syscall(__NR_getpid);
993}
994
995/* Ensure that SIGSYS overrides SIG_IGN */
996TEST_F_SIGNAL(TRAP, ign, SIGSYS)
997{
998 long ret;
999
1000 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1001 ASSERT_EQ(0, ret);
1002
1003 signal(SIGSYS, SIG_IGN);
1004
1005 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
1006 ASSERT_EQ(0, ret);
1007 syscall(__NR_getpid);
1008}
1009
10859f38 1010static siginfo_t TRAP_info;
c99ee51a
KC		 1011static volatile int TRAP_nr;
1012static void TRAP_action(int nr, siginfo_t *info, void *void_context)
1013{
1014 memcpy(&TRAP_info, info, sizeof(TRAP_info));
1015 TRAP_nr = nr;
1016}
1017
1018TEST_F(TRAP, handler)
1019{
1020 int ret, test;
1021 struct sigaction act;
1022 sigset_t mask;
1023
1024 memset(&act, 0, sizeof(act));
1025 sigemptyset(&mask);
1026 sigaddset(&mask, SIGSYS);
1027
1028 act.sa_sigaction = &TRAP_action;
1029 act.sa_flags = SA_SIGINFO;
1030 ret = sigaction(SIGSYS, &act, NULL);
1031 ASSERT_EQ(0, ret) {
1032 TH_LOG("sigaction failed");
1033 }
1034 ret = sigprocmask(SIG_UNBLOCK, &mask, NULL);
1035 ASSERT_EQ(0, ret) {
1036 TH_LOG("sigprocmask failed");
1037 }
1038
1039 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1040 ASSERT_EQ(0, ret);
1041 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
1042 ASSERT_EQ(0, ret);
1043 TRAP_nr = 0;
1044 memset(&TRAP_info, 0, sizeof(TRAP_info));
1045 /* Expect the registers to be rolled back. (nr = error) may vary
1046 * based on arch. */
1047 ret = syscall(__NR_getpid);
1048 /* Silence gcc warning about volatile. */
1049 test = TRAP_nr;
1050 EXPECT_EQ(SIGSYS, test);
1051 struct local_sigsys {
1052 void *_call_addr; /* calling user insn */
1053 int _syscall; /* triggering system call number */
1054 unsigned int _arch; /* AUDIT_ARCH_* of syscall */
1055 } *sigsys = (struct local_sigsys *)
1056#ifdef si_syscall
1057 &(TRAP_info.si_call_addr);
1058#else
1059 &TRAP_info.si_pid;
1060#endif
1061 EXPECT_EQ(__NR_getpid, sigsys->_syscall);
1062 /* Make sure arch is non-zero. */
1063 EXPECT_NE(0, sigsys->_arch);
1064 EXPECT_NE(0, (unsigned long)sigsys->_call_addr);
1065}
1066
1ae81d78 1067FIXTURE(precedence) {
c99ee51a 1068 struct sock_fprog allow;
59f5cf44 1069 struct sock_fprog log;
c99ee51a
KC		 1070 struct sock_fprog trace;
1071 struct sock_fprog error;
1072 struct sock_fprog trap;
1073 struct sock_fprog kill;
1074};
1075
1076FIXTURE_SETUP(precedence)
1077{
1078 struct sock_filter allow_insns[] = {
1079 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1080 };
59f5cf44
TH		 1081 struct sock_filter log_insns[] = {
1082 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1083 offsetof(struct seccomp_data, nr)),
1084 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1085 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1086 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG),
1087 };
c99ee51a
KC		 1088 struct sock_filter trace_insns[] = {
1089 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1090 offsetof(struct seccomp_data, nr)),
1091 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1092 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1093 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE),
1094 };
1095 struct sock_filter error_insns[] = {
1096 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1097 offsetof(struct seccomp_data, nr)),
1098 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1099 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1100 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO),
1101 };
1102 struct sock_filter trap_insns[] = {
1103 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1104 offsetof(struct seccomp_data, nr)),
1105 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1106 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1107 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
1108 };
1109 struct sock_filter kill_insns[] = {
1110 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1111 offsetof(struct seccomp_data, nr)),
1112 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1113 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1114 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1115 };
1116
1117 memset(self, 0, sizeof(*self));
1118#define FILTER_ALLOC(_x) \
1119 self->_x.filter = malloc(sizeof(_x##_insns)); \
1120 ASSERT_NE(NULL, self->_x.filter); \
1121 memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \
1122 self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns)
1123 FILTER_ALLOC(allow);
59f5cf44 1124 FILTER_ALLOC(log);
c99ee51a
KC		 1125 FILTER_ALLOC(trace);
1126 FILTER_ALLOC(error);
1127 FILTER_ALLOC(trap);
1128 FILTER_ALLOC(kill);
1129}
1130
1131FIXTURE_TEARDOWN(precedence)
1132{
1133#define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter)
1134 FILTER_FREE(allow);
59f5cf44 1135 FILTER_FREE(log);
c99ee51a
KC		 1136 FILTER_FREE(trace);
1137 FILTER_FREE(error);
1138 FILTER_FREE(trap);
1139 FILTER_FREE(kill);
1140}
1141
1142TEST_F(precedence, allow_ok)
1143{
1144 pid_t parent, res = 0;
1145 long ret;
1146
1147 parent = getppid();
1148 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1149 ASSERT_EQ(0, ret);
1150
1151 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1152 ASSERT_EQ(0, ret);
59f5cf44
TH		 1153 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1154 ASSERT_EQ(0, ret);
c99ee51a
KC		 1155 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1156 ASSERT_EQ(0, ret);
1157 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1158 ASSERT_EQ(0, ret);
1159 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1160 ASSERT_EQ(0, ret);
1161 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1162 ASSERT_EQ(0, ret);
1163 /* Should work just fine. */
1164 res = syscall(__NR_getppid);
1165 EXPECT_EQ(parent, res);
1166}
1167
1168TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS)
1169{
1170 pid_t parent, res = 0;
1171 long ret;
1172
1173 parent = getppid();
1174 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1175 ASSERT_EQ(0, ret);
1176
1177 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1178 ASSERT_EQ(0, ret);
59f5cf44
TH		 1179 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1180 ASSERT_EQ(0, ret);
c99ee51a
KC		 1181 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1182 ASSERT_EQ(0, ret);
1183 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1184 ASSERT_EQ(0, ret);
1185 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1186 ASSERT_EQ(0, ret);
1187 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1188 ASSERT_EQ(0, ret);
1189 /* Should work just fine. */
1190 res = syscall(__NR_getppid);
1191 EXPECT_EQ(parent, res);
1192 /* getpid() should never return. */
1193 res = syscall(__NR_getpid);
1194 EXPECT_EQ(0, res);
1195}
1196
1197TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS)
1198{
1199 pid_t parent;
1200 long ret;
1201
1202 parent = getppid();
1203 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1204 ASSERT_EQ(0, ret);
1205
1206 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1207 ASSERT_EQ(0, ret);
1208 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1209 ASSERT_EQ(0, ret);
1210 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1211 ASSERT_EQ(0, ret);
59f5cf44
TH		 1212 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1213 ASSERT_EQ(0, ret);
c99ee51a
KC		 1214 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1215 ASSERT_EQ(0, ret);
1216 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1217 ASSERT_EQ(0, ret);
1218 /* Should work just fine. */
1219 EXPECT_EQ(parent, syscall(__NR_getppid));
1220 /* getpid() should never return. */
1221 EXPECT_EQ(0, syscall(__NR_getpid));
1222}
1223
1224TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS)
1225{
1226 pid_t parent;
1227 long ret;
1228
1229 parent = getppid();
1230 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1231 ASSERT_EQ(0, ret);
1232
1233 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1234 ASSERT_EQ(0, ret);
59f5cf44
TH		 1235 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1236 ASSERT_EQ(0, ret);
c99ee51a
KC		 1237 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1238 ASSERT_EQ(0, ret);
1239 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1240 ASSERT_EQ(0, ret);
1241 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1242 ASSERT_EQ(0, ret);
1243 /* Should work just fine. */
1244 EXPECT_EQ(parent, syscall(__NR_getppid));
1245 /* getpid() should never return. */
1246 EXPECT_EQ(0, syscall(__NR_getpid));
1247}
1248
1249TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS)
1250{
1251 pid_t parent;
1252 long ret;
1253
1254 parent = getppid();
1255 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1256 ASSERT_EQ(0, ret);
1257
1258 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1259 ASSERT_EQ(0, ret);
1260 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1261 ASSERT_EQ(0, ret);
59f5cf44
TH		 1262 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1263 ASSERT_EQ(0, ret);
c99ee51a
KC		 1264 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1265 ASSERT_EQ(0, ret);
1266 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1267 ASSERT_EQ(0, ret);
1268 /* Should work just fine. */
1269 EXPECT_EQ(parent, syscall(__NR_getppid));
1270 /* getpid() should never return. */
1271 EXPECT_EQ(0, syscall(__NR_getpid));
1272}
1273
1274TEST_F(precedence, errno_is_third)
1275{
1276 pid_t parent;
1277 long ret;
1278
1279 parent = getppid();
1280 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1281 ASSERT_EQ(0, ret);
1282
1283 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1284 ASSERT_EQ(0, ret);
59f5cf44
TH		 1285 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1286 ASSERT_EQ(0, ret);
c99ee51a
KC		 1287 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1288 ASSERT_EQ(0, ret);
1289 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1290 ASSERT_EQ(0, ret);
1291 /* Should work just fine. */
1292 EXPECT_EQ(parent, syscall(__NR_getppid));
1293 EXPECT_EQ(0, syscall(__NR_getpid));
1294}
1295
1296TEST_F(precedence, errno_is_third_in_any_order)
1297{
1298 pid_t parent;
1299 long ret;
1300
1301 parent = getppid();
1302 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1303 ASSERT_EQ(0, ret);
1304
59f5cf44
TH		 1305 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1306 ASSERT_EQ(0, ret);
c99ee51a
KC		 1307 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1308 ASSERT_EQ(0, ret);
1309 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1310 ASSERT_EQ(0, ret);
1311 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1312 ASSERT_EQ(0, ret);
1313 /* Should work just fine. */
1314 EXPECT_EQ(parent, syscall(__NR_getppid));
1315 EXPECT_EQ(0, syscall(__NR_getpid));
1316}
1317
1318TEST_F(precedence, trace_is_fourth)
1319{
1320 pid_t parent;
1321 long ret;
1322
1323 parent = getppid();
1324 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1325 ASSERT_EQ(0, ret);
1326
1327 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1328 ASSERT_EQ(0, ret);
59f5cf44
TH		 1329 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1330 ASSERT_EQ(0, ret);
c99ee51a
KC		 1331 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1332 ASSERT_EQ(0, ret);
1333 /* Should work just fine. */
1334 EXPECT_EQ(parent, syscall(__NR_getppid));
1335 /* No ptracer */
1336 EXPECT_EQ(-1, syscall(__NR_getpid));
1337}
1338
1339TEST_F(precedence, trace_is_fourth_in_any_order)
1340{
1341 pid_t parent;
1342 long ret;
1343
1344 parent = getppid();
1345 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1346 ASSERT_EQ(0, ret);
1347
1348 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1349 ASSERT_EQ(0, ret);
1350 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1351 ASSERT_EQ(0, ret);
59f5cf44
TH		 1352 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1353 ASSERT_EQ(0, ret);
c99ee51a
KC		 1354 /* Should work just fine. */
1355 EXPECT_EQ(parent, syscall(__NR_getppid));
1356 /* No ptracer */
1357 EXPECT_EQ(-1, syscall(__NR_getpid));
1358}
1359
59f5cf44
TH		 1360TEST_F(precedence, log_is_fifth)
1361{
1362 pid_t mypid, parent;
1363 long ret;
1364
1365 mypid = getpid();
1366 parent = getppid();
1367 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1368 ASSERT_EQ(0, ret);
1369
1370 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1371 ASSERT_EQ(0, ret);
1372 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1373 ASSERT_EQ(0, ret);
1374 /* Should work just fine. */
1375 EXPECT_EQ(parent, syscall(__NR_getppid));
1376 /* Should also work just fine */
1377 EXPECT_EQ(mypid, syscall(__NR_getpid));
1378}
1379
1380TEST_F(precedence, log_is_fifth_in_any_order)
1381{
1382 pid_t mypid, parent;
1383 long ret;
1384
1385 mypid = getpid();
1386 parent = getppid();
1387 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1388 ASSERT_EQ(0, ret);
1389
1390 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1391 ASSERT_EQ(0, ret);
1392 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1393 ASSERT_EQ(0, ret);
1394 /* Should work just fine. */
1395 EXPECT_EQ(parent, syscall(__NR_getppid));
1396 /* Should also work just fine */
1397 EXPECT_EQ(mypid, syscall(__NR_getpid));
1398}
1399
c99ee51a
KC		 1400#ifndef PTRACE_O_TRACESECCOMP
1401#define PTRACE_O_TRACESECCOMP 0x00000080
1402#endif
1403
1404/* Catch the Ubuntu 12.04 value error. */
1405#if PTRACE_EVENT_SECCOMP != 7
1406#undef PTRACE_EVENT_SECCOMP
1407#endif
1408
1409#ifndef PTRACE_EVENT_SECCOMP
1410#define PTRACE_EVENT_SECCOMP 7
1411#endif
1412
1413#define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP)
1414bool tracer_running;
1415void tracer_stop(int sig)
1416{
1417 tracer_running = false;
1418}
1419
1420typedef void tracer_func_t(struct __test_metadata *_metadata,
1421 pid_t tracee, int status, void *args);
1422
58d0a862
KC		 1423void start_tracer(struct __test_metadata *_metadata, int fd, pid_t tracee,
1424 tracer_func_t tracer_func, void *args, bool ptrace_syscall)
c99ee51a
KC		 1425{
1426 int ret = -1;
1427 struct sigaction action = {
1428 .sa_handler = tracer_stop,
1429 };
1430
1431 /* Allow external shutdown. */
1432 tracer_running = true;
1433 ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL));
1434
1435 errno = 0;
1436 while (ret == -1 && errno != EINVAL)
1437 ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0);
1438 ASSERT_EQ(0, ret) {
1439 kill(tracee, SIGKILL);
1440 }
1441 /* Wait for attach stop */
1442 wait(NULL);
1443
58d0a862
KC		 1444 ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, ptrace_syscall ?
1445 PTRACE_O_TRACESYSGOOD :
1446 PTRACE_O_TRACESECCOMP);
c99ee51a
KC		 1447 ASSERT_EQ(0, ret) {
1448 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP");
1449 kill(tracee, SIGKILL);
1450 }
58d0a862
KC		 1451 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1452 tracee, NULL, 0);
1453 ASSERT_EQ(0, ret);
c99ee51a
KC		 1454
1455 /* Unblock the tracee */
1456 ASSERT_EQ(1, write(fd, "A", 1));
1457 ASSERT_EQ(0, close(fd));
1458
1459 /* Run until we're shut down. Must assert to stop execution. */
1460 while (tracer_running) {
1461 int status;
1462
1463 if (wait(&status) != tracee)
1464 continue;
1465 if (WIFSIGNALED(status) || WIFEXITED(status))
1466 /* Child is dead. Time to go. */
1467 return;
1468
58d0a862
KC		 1469 /* Check if this is a seccomp event. */
1470 ASSERT_EQ(!ptrace_syscall, IS_SECCOMP_EVENT(status));
c99ee51a
KC		 1471
1472 tracer_func(_metadata, tracee, status, args);
1473
58d0a862
KC		 1474 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1475 tracee, NULL, 0);
c99ee51a
KC		 1476 ASSERT_EQ(0, ret);
1477 }
1478 /* Directly report the status of our test harness results. */
1479 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS : EXIT_FAILURE);
1480}
1481
1482/* Common tracer setup/teardown functions. */
1483void cont_handler(int num)
1484{ }
1485pid_t setup_trace_fixture(struct __test_metadata *_metadata,
58d0a862 1486 tracer_func_t func, void *args, bool ptrace_syscall)
c99ee51a
KC		 1487{
1488 char sync;
1489 int pipefd[2];
1490 pid_t tracer_pid;
1491 pid_t tracee = getpid();
1492
1493 /* Setup a pipe for clean synchronization. */
1494 ASSERT_EQ(0, pipe(pipefd));
1495
1496 /* Fork a child which we'll promote to tracer */
1497 tracer_pid = fork();
1498 ASSERT_LE(0, tracer_pid);
1499 signal(SIGALRM, cont_handler);
1500 if (tracer_pid == 0) {
1501 close(pipefd[0]);
58d0a862
KC		 1502 start_tracer(_metadata, pipefd[1], tracee, func, args,
1503 ptrace_syscall);
c99ee51a
KC		 1504 syscall(__NR_exit, 0);
1505 }
1506 close(pipefd[1]);
1507 prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0);
1508 read(pipefd[0], &sync, 1);
1509 close(pipefd[0]);
1510
1511 return tracer_pid;
1512}
1513void teardown_trace_fixture(struct __test_metadata *_metadata,
1514 pid_t tracer)
1515{
1516 if (tracer) {
1517 int status;
1518 /*
1519 * Extract the exit code from the other process and
1520 * adopt it for ourselves in case its asserts failed.
1521 */
1522 ASSERT_EQ(0, kill(tracer, SIGUSR1));
1523 ASSERT_EQ(tracer, waitpid(tracer, &status, 0));
1524 if (WEXITSTATUS(status))
1525 _metadata->passed = 0;
1526 }
1527}
1528
1529/* "poke" tracer arguments and function. */
1530struct tracer_args_poke_t {
1531 unsigned long poke_addr;
1532};
1533
1534void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status,
1535 void *args)
1536{
1537 int ret;
1538 unsigned long msg;
1539 struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args;
1540
1541 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1542 EXPECT_EQ(0, ret);
1543 /* If this fails, don't try to recover. */
1544 ASSERT_EQ(0x1001, msg) {
1545 kill(tracee, SIGKILL);
1546 }
1547 /*
1548 * Poke in the message.
1549 * Registers are not touched to try to keep this relatively arch
1550 * agnostic.
1551 */
1552 ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001);
1553 EXPECT_EQ(0, ret);
1554}
1555
1ae81d78 1556FIXTURE(TRACE_poke) {
c99ee51a
KC		 1557 struct sock_fprog prog;
1558 pid_t tracer;
1559 long poked;
1560 struct tracer_args_poke_t tracer_args;
1561};
1562
1563FIXTURE_SETUP(TRACE_poke)
1564{
1565 struct sock_filter filter[] = {
1566 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1567 offsetof(struct seccomp_data, nr)),
1568 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1569 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001),
1570 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1571 };
1572
1573 self->poked = 0;
1574 memset(&self->prog, 0, sizeof(self->prog));
1575 self->prog.filter = malloc(sizeof(filter));
1576 ASSERT_NE(NULL, self->prog.filter);
1577 memcpy(self->prog.filter, filter, sizeof(filter));
1578 self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1579
1580 /* Set up tracer args. */
1581 self->tracer_args.poke_addr = (unsigned long)&self->poked;
1582
1583 /* Launch tracer. */
1584 self->tracer = setup_trace_fixture(_metadata, tracer_poke,
58d0a862 1585 &self->tracer_args, false);
c99ee51a
KC		 1586}
1587
1588FIXTURE_TEARDOWN(TRACE_poke)
1589{
1590 teardown_trace_fixture(_metadata, self->tracer);
1591 if (self->prog.filter)
1592 free(self->prog.filter);
1593}
1594
1595TEST_F(TRACE_poke, read_has_side_effects)
1596{
1597 ssize_t ret;
1598
1599 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1600 ASSERT_EQ(0, ret);
1601
1602 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1603 ASSERT_EQ(0, ret);
1604
1605 EXPECT_EQ(0, self->poked);
1606 ret = read(-1, NULL, 0);
1607 EXPECT_EQ(-1, ret);
1608 EXPECT_EQ(0x1001, self->poked);
1609}
1610
1611TEST_F(TRACE_poke, getpid_runs_normally)
1612{
1613 long ret;
1614
1615 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1616 ASSERT_EQ(0, ret);
1617
1618 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1619 ASSERT_EQ(0, ret);
1620
1621 EXPECT_EQ(0, self->poked);
1622 EXPECT_NE(0, syscall(__NR_getpid));
1623 EXPECT_EQ(0, self->poked);
1624}
1625
1626#if defined(__x86_64__)
1627# define ARCH_REGS struct user_regs_struct
1628# define SYSCALL_NUM orig_rax
1629# define SYSCALL_RET rax
1630#elif defined(__i386__)
1631# define ARCH_REGS struct user_regs_struct
1632# define SYSCALL_NUM orig_eax
1633# define SYSCALL_RET eax
1634#elif defined(__arm__)
1635# define ARCH_REGS struct pt_regs
1636# define SYSCALL_NUM ARM_r7
1637# define SYSCALL_RET ARM_r0
1638#elif defined(__aarch64__)
1639# define ARCH_REGS struct user_pt_regs
1640# define SYSCALL_NUM regs[8]
1641# define SYSCALL_RET regs[0]
5340627e
DA		 1642#elif defined(__riscv) && __riscv_xlen == 64
1643# define ARCH_REGS struct user_regs_struct
1644# define SYSCALL_NUM a7
1645# define SYSCALL_RET a0
64e2a42b
HD		 1646#elif defined(__hppa__)
1647# define ARCH_REGS struct user_regs_struct
1648# define SYSCALL_NUM gr[20]
1649# define SYSCALL_RET gr[28]
5d83c2b3
ME		 1650#elif defined(__powerpc__)
1651# define ARCH_REGS struct pt_regs
1652# define SYSCALL_NUM gpr[0]
1653# define SYSCALL_RET gpr[3]
b623c4da
KC		 1654#elif defined(__s390__)
1655# define ARCH_REGS s390_regs
1656# define SYSCALL_NUM gprs[2]
1657# define SYSCALL_RET gprs[2]
0ce105bf
MR		 1658#elif defined(__mips__)
1659# define ARCH_REGS struct pt_regs
1660# define SYSCALL_NUM regs[2]
1661# define SYSCALL_SYSCALL_NUM regs[4]
1662# define SYSCALL_RET regs[2]
1663# define SYSCALL_NUM_RET_SHARE_REG
c99ee51a
KC		 1664#else
1665# error "Do not know how to find your architecture's registers and syscalls"
1666#endif
1667
a33b2d03
KC		 1668/* When the syscall return can't be changed, stub out the tests for it. */
1669#ifdef SYSCALL_NUM_RET_SHARE_REG
1670# define EXPECT_SYSCALL_RETURN(val, action) EXPECT_EQ(-1, action)
1671#else
ed5f1326
KC		 1672# define EXPECT_SYSCALL_RETURN(val, action) \
1673 do { \
1674 errno = 0; \
1675 if (val < 0) { \
1676 EXPECT_EQ(-1, action); \
1677 EXPECT_EQ(-(val), errno); \
1678 } else { \
1679 EXPECT_EQ(val, action); \
1680 } \
1681 } while (0)
a33b2d03
KC		 1682#endif
1683
4a0b8807
MS		 1684/* Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for
1685 * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux).
1686 */
0ce105bf 1687#if defined(__x86_64__) || defined(__i386__) || defined(__mips__)
4a0b8807
MS		 1688#define HAVE_GETREGS
1689#endif
1690
c99ee51a
KC		 1691/* Architecture-specific syscall fetching routine. */
1692int get_syscall(struct __test_metadata *_metadata, pid_t tracee)
1693{
c99ee51a 1694 ARCH_REGS regs;
4a0b8807
MS		 1695#ifdef HAVE_GETREGS
1696 EXPECT_EQ(0, ptrace(PTRACE_GETREGS, tracee, 0, &regs)) {
1697 TH_LOG("PTRACE_GETREGS failed");
1698 return -1;
1699 }
1700#else
1701 struct iovec iov;
c99ee51a
KC		 1702
1703 iov.iov_base = &regs;
1704 iov.iov_len = sizeof(regs);
1705 EXPECT_EQ(0, ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov)) {
1706 TH_LOG("PTRACE_GETREGSET failed");
1707 return -1;
1708 }
4a0b8807 1709#endif
c99ee51a 1710
0ce105bf
MR		 1711#if defined(__mips__)
1712 if (regs.SYSCALL_NUM == __NR_O32_Linux)
1713 return regs.SYSCALL_SYSCALL_NUM;
1714#endif
c99ee51a
KC		 1715 return regs.SYSCALL_NUM;
1716}
1717
1718/* Architecture-specific syscall changing routine. */
1719void change_syscall(struct __test_metadata *_metadata,
ed5f1326 1720 pid_t tracee, int syscall, int result)
c99ee51a 1721{
c99ee51a
KC		 1722 int ret;
1723 ARCH_REGS regs;
4a0b8807
MS		 1724#ifdef HAVE_GETREGS
1725 ret = ptrace(PTRACE_GETREGS, tracee, 0, &regs);
1726#else
1727 struct iovec iov;
c99ee51a
KC		 1728 iov.iov_base = &regs;
1729 iov.iov_len = sizeof(regs);
1730 ret = ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov);
4a0b8807 1731#endif
34a048cc 1732 EXPECT_EQ(0, ret) {}
c99ee51a 1733
256d0afb 1734#if defined(__x86_64__) || defined(__i386__) || defined(__powerpc__) || \
5340627e 1735 defined(__s390__) || defined(__hppa__) || defined(__riscv)
c99ee51a
KC		 1736 {
1737 regs.SYSCALL_NUM = syscall;
1738 }
0ce105bf
MR		 1739#elif defined(__mips__)
1740 {
1741 if (regs.SYSCALL_NUM == __NR_O32_Linux)
1742 regs.SYSCALL_SYSCALL_NUM = syscall;
1743 else
1744 regs.SYSCALL_NUM = syscall;
1745 }
c99ee51a
KC		 1746
1747#elif defined(__arm__)
1748# ifndef PTRACE_SET_SYSCALL
1749# define PTRACE_SET_SYSCALL 23
1750# endif
1751 {
1752 ret = ptrace(PTRACE_SET_SYSCALL, tracee, NULL, syscall);
1753 EXPECT_EQ(0, ret);
1754 }
1755
256d0afb
KC		 1756#elif defined(__aarch64__)
1757# ifndef NT_ARM_SYSTEM_CALL
1758# define NT_ARM_SYSTEM_CALL 0x404
1759# endif
1760 {
1761 iov.iov_base = &syscall;
1762 iov.iov_len = sizeof(syscall);
1763 ret = ptrace(PTRACE_SETREGSET, tracee, NT_ARM_SYSTEM_CALL,
1764 &iov);
1765 EXPECT_EQ(0, ret);
1766 }
1767
c99ee51a
KC		 1768#else
1769 ASSERT_EQ(1, 0) {
1770 TH_LOG("How is the syscall changed on this architecture?");
1771 }
1772#endif
1773
1774 /* If syscall is skipped, change return value. */
1775 if (syscall == -1)
0ce105bf
MR		 1776#ifdef SYSCALL_NUM_RET_SHARE_REG
1777 TH_LOG("Can't modify syscall return on this architecture");
1778#else
ed5f1326 1779 regs.SYSCALL_RET = result;
0ce105bf 1780#endif
c99ee51a 1781
4a0b8807
MS		 1782#ifdef HAVE_GETREGS
1783 ret = ptrace(PTRACE_SETREGS, tracee, 0, &regs);
1784#else
256d0afb
KC		 1785 iov.iov_base = &regs;
1786 iov.iov_len = sizeof(regs);
c99ee51a 1787 ret = ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &iov);
4a0b8807 1788#endif
c99ee51a
KC		 1789 EXPECT_EQ(0, ret);
1790}
1791
1792void tracer_syscall(struct __test_metadata *_metadata, pid_t tracee,
1793 int status, void *args)
1794{
1795 int ret;
1796 unsigned long msg;
1797
1798 /* Make sure we got the right message. */
1799 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1800 EXPECT_EQ(0, ret);
1801
b623c4da 1802 /* Validate and take action on expected syscalls. */
c99ee51a
KC		 1803 switch (msg) {
1804 case 0x1002:
1805 /* change getpid to getppid. */
b623c4da 1806 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee));
ed5f1326 1807 change_syscall(_metadata, tracee, __NR_getppid, 0);
c99ee51a
KC		 1808 break;
1809 case 0x1003:
ed5f1326 1810 /* skip gettid with valid return code. */
b623c4da 1811 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee));
ed5f1326 1812 change_syscall(_metadata, tracee, -1, 45000);
c99ee51a
KC		 1813 break;
1814 case 0x1004:
ed5f1326
KC		 1815 /* skip openat with error. */
1816 EXPECT_EQ(__NR_openat, get_syscall(_metadata, tracee));
1817 change_syscall(_metadata, tracee, -1, -ESRCH);
1818 break;
1819 case 0x1005:
c99ee51a 1820 /* do nothing (allow getppid) */
b623c4da 1821 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee));
c99ee51a
KC		 1822 break;
1823 default:
1824 EXPECT_EQ(0, msg) {
1825 TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg);
1826 kill(tracee, SIGKILL);
1827 }
1828 }
1829
1830}
1831
58d0a862
KC		 1832void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee,
1833 int status, void *args)
1834{
1835 int ret, nr;
1836 unsigned long msg;
1837 static bool entry;
1838
201766a2
EK		 1839 /*
1840 * The traditional way to tell PTRACE_SYSCALL entry/exit
1841 * is by counting.
1842 */
1843 entry = !entry;
1844
1845 /* Make sure we got an appropriate message. */
58d0a862
KC		 1846 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1847 EXPECT_EQ(0, ret);
201766a2
EK		 1848 EXPECT_EQ(entry ? PTRACE_EVENTMSG_SYSCALL_ENTRY
1849 : PTRACE_EVENTMSG_SYSCALL_EXIT, msg);
58d0a862 1850
58d0a862
KC		 1851 if (!entry)
1852 return;
1853
1854 nr = get_syscall(_metadata, tracee);
1855
1856 if (nr == __NR_getpid)
ed5f1326
KC		 1857 change_syscall(_metadata, tracee, __NR_getppid, 0);
1858 if (nr == __NR_gettid)
1859 change_syscall(_metadata, tracee, -1, 45000);
912ec316 1860 if (nr == __NR_openat)
ed5f1326 1861 change_syscall(_metadata, tracee, -1, -ESRCH);
58d0a862
KC		 1862}
1863
1ae81d78 1864FIXTURE(TRACE_syscall) {
c99ee51a
KC		 1865 struct sock_fprog prog;
1866 pid_t tracer, mytid, mypid, parent;
1867};
1868
1869FIXTURE_SETUP(TRACE_syscall)
1870{
1871 struct sock_filter filter[] = {
1872 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1873 offsetof(struct seccomp_data, nr)),
1874 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
1875 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002),
1876 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1),
1877 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003),
ed5f1326 1878 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_openat, 0, 1),
c99ee51a 1879 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004),
ed5f1326
KC		 1880 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1881 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1005),
c99ee51a
KC		 1882 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1883 };
1884
1885 memset(&self->prog, 0, sizeof(self->prog));
1886 self->prog.filter = malloc(sizeof(filter));
1887 ASSERT_NE(NULL, self->prog.filter);
1888 memcpy(self->prog.filter, filter, sizeof(filter));
1889 self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1890
1891 /* Prepare some testable syscall results. */
1892 self->mytid = syscall(__NR_gettid);
1893 ASSERT_GT(self->mytid, 0);
1894 ASSERT_NE(self->mytid, 1) {
1895 TH_LOG("Running this test as init is not supported. :)");
1896 }
1897
1898 self->mypid = getpid();
1899 ASSERT_GT(self->mypid, 0);
1900 ASSERT_EQ(self->mytid, self->mypid);
1901
1902 self->parent = getppid();
1903 ASSERT_GT(self->parent, 0);
1904 ASSERT_NE(self->parent, self->mypid);
1905
1906 /* Launch tracer. */
58d0a862
KC		 1907 self->tracer = setup_trace_fixture(_metadata, tracer_syscall, NULL,
1908 false);
c99ee51a
KC		 1909}
1910
1911FIXTURE_TEARDOWN(TRACE_syscall)
1912{
1913 teardown_trace_fixture(_metadata, self->tracer);
1914 if (self->prog.filter)
1915 free(self->prog.filter);
1916}
1917
a33b2d03
KC		 1918TEST_F(TRACE_syscall, ptrace_syscall_redirected)
1919{
1920 /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1921 teardown_trace_fixture(_metadata, self->tracer);
1922 self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1923 true);
1924
1925 /* Tracer will redirect getpid to getppid. */
1926 EXPECT_NE(self->mypid, syscall(__NR_getpid));
1927}
1928
ed5f1326
KC		 1929TEST_F(TRACE_syscall, ptrace_syscall_errno)
1930{
1931 /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1932 teardown_trace_fixture(_metadata, self->tracer);
1933 self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1934 true);
1935
1936 /* Tracer should skip the open syscall, resulting in ESRCH. */
1937 EXPECT_SYSCALL_RETURN(-ESRCH, syscall(__NR_openat));
1938}
1939
1940TEST_F(TRACE_syscall, ptrace_syscall_faked)
a33b2d03
KC		 1941{
1942 /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1943 teardown_trace_fixture(_metadata, self->tracer);
1944 self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1945 true);
1946
ed5f1326
KC		 1947 /* Tracer should skip the gettid syscall, resulting fake pid. */
1948 EXPECT_SYSCALL_RETURN(45000, syscall(__NR_gettid));
a33b2d03
KC		 1949}
1950
c99ee51a
KC		 1951TEST_F(TRACE_syscall, syscall_allowed)
1952{
1953 long ret;
1954
1955 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1956 ASSERT_EQ(0, ret);
1957
1958 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1959 ASSERT_EQ(0, ret);
1960
1961 /* getppid works as expected (no changes). */
1962 EXPECT_EQ(self->parent, syscall(__NR_getppid));
1963 EXPECT_NE(self->mypid, syscall(__NR_getppid));
1964}
1965
1966TEST_F(TRACE_syscall, syscall_redirected)
1967{
1968 long ret;
1969
1970 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1971 ASSERT_EQ(0, ret);
1972
1973 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1974 ASSERT_EQ(0, ret);
1975
1976 /* getpid has been redirected to getppid as expected. */
1977 EXPECT_EQ(self->parent, syscall(__NR_getpid));
1978 EXPECT_NE(self->mypid, syscall(__NR_getpid));
1979}
1980
ed5f1326
KC		 1981TEST_F(TRACE_syscall, syscall_errno)
1982{
1983 long ret;
1984
1985 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1986 ASSERT_EQ(0, ret);
1987
1988 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1989 ASSERT_EQ(0, ret);
1990
1991 /* openat has been skipped and an errno return. */
1992 EXPECT_SYSCALL_RETURN(-ESRCH, syscall(__NR_openat));
1993}
1994
1995TEST_F(TRACE_syscall, syscall_faked)
c99ee51a
KC		 1996{
1997 long ret;
1998
1999 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2000 ASSERT_EQ(0, ret);
2001
2002 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
2003 ASSERT_EQ(0, ret);
2004
2005 /* gettid has been skipped and an altered return value stored. */
ed5f1326 2006 EXPECT_SYSCALL_RETURN(45000, syscall(__NR_gettid));
c99ee51a
KC		 2007}
2008
58d0a862
KC		 2009TEST_F(TRACE_syscall, skip_after_RET_TRACE)
2010{
2011 struct sock_filter filter[] = {
2012 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2013 offsetof(struct seccomp_data, nr)),
2014 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2015 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
2016 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2017 };
2018 struct sock_fprog prog = {
2019 .len = (unsigned short)ARRAY_SIZE(filter),
2020 .filter = filter,
2021 };
2022 long ret;
2023
2024 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2025 ASSERT_EQ(0, ret);
2026
2027 /* Install fixture filter. */
2028 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
2029 ASSERT_EQ(0, ret);
2030
2031 /* Install "errno on getppid" filter. */
2032 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2033 ASSERT_EQ(0, ret);
2034
2035 /* Tracer will redirect getpid to getppid, and we should see EPERM. */
a33b2d03 2036 errno = 0;
58d0a862
KC		 2037 EXPECT_EQ(-1, syscall(__NR_getpid));
2038 EXPECT_EQ(EPERM, errno);
2039}
2040
2041TEST_F_SIGNAL(TRACE_syscall, kill_after_RET_TRACE, SIGSYS)
2042{
2043 struct sock_filter filter[] = {
2044 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2045 offsetof(struct seccomp_data, nr)),
2046 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2047 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2048 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2049 };
2050 struct sock_fprog prog = {
2051 .len = (unsigned short)ARRAY_SIZE(filter),
2052 .filter = filter,
2053 };
2054 long ret;
2055
2056 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2057 ASSERT_EQ(0, ret);
2058
2059 /* Install fixture filter. */
2060 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
2061 ASSERT_EQ(0, ret);
2062
2063 /* Install "death on getppid" filter. */
2064 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2065 ASSERT_EQ(0, ret);
2066
2067 /* Tracer will redirect getpid to getppid, and we should die. */
2068 EXPECT_NE(self->mypid, syscall(__NR_getpid));
2069}
2070
2071TEST_F(TRACE_syscall, skip_after_ptrace)
2072{
2073 struct sock_filter filter[] = {
2074 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2075 offsetof(struct seccomp_data, nr)),
2076 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2077 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
2078 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2079 };
2080 struct sock_fprog prog = {
2081 .len = (unsigned short)ARRAY_SIZE(filter),
2082 .filter = filter,
2083 };
2084 long ret;
2085
2086 /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
2087 teardown_trace_fixture(_metadata, self->tracer);
2088 self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
2089 true);
2090
2091 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2092 ASSERT_EQ(0, ret);
2093
2094 /* Install "errno on getppid" filter. */
2095 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2096 ASSERT_EQ(0, ret);
2097
2098 /* Tracer will redirect getpid to getppid, and we should see EPERM. */
2099 EXPECT_EQ(-1, syscall(__NR_getpid));
2100 EXPECT_EQ(EPERM, errno);
2101}
2102
2103TEST_F_SIGNAL(TRACE_syscall, kill_after_ptrace, SIGSYS)
2104{
2105 struct sock_filter filter[] = {
2106 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2107 offsetof(struct seccomp_data, nr)),
2108 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2109 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2110 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2111 };
2112 struct sock_fprog prog = {
2113 .len = (unsigned short)ARRAY_SIZE(filter),
2114 .filter = filter,
2115 };
2116 long ret;
2117
2118 /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
2119 teardown_trace_fixture(_metadata, self->tracer);
2120 self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
2121 true);
2122
2123 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2124 ASSERT_EQ(0, ret);
2125
2126 /* Install "death on getppid" filter. */
2127 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2128 ASSERT_EQ(0, ret);
2129
2130 /* Tracer will redirect getpid to getppid, and we should die. */
2131 EXPECT_NE(self->mypid, syscall(__NR_getpid));
2132}
2133
c99ee51a
KC		 2134TEST(seccomp_syscall)
2135{
2136 struct sock_filter filter[] = {
2137 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2138 };
2139 struct sock_fprog prog = {
2140 .len = (unsigned short)ARRAY_SIZE(filter),
2141 .filter = filter,
2142 };
2143 long ret;
2144
2145 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2146 ASSERT_EQ(0, ret) {
2147 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2148 }
2149
2150 /* Reject insane operation. */
2151 ret = seccomp(-1, 0, &prog);
b623c4da
KC		 2152 ASSERT_NE(ENOSYS, errno) {
2153 TH_LOG("Kernel does not support seccomp syscall!");
2154 }
c99ee51a
KC		 2155 EXPECT_EQ(EINVAL, errno) {
2156 TH_LOG("Did not reject crazy op value!");
2157 }
2158
2159 /* Reject strict with flags or pointer. */
2160 ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL);
2161 EXPECT_EQ(EINVAL, errno) {
2162 TH_LOG("Did not reject mode strict with flags!");
2163 }
2164 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog);
2165 EXPECT_EQ(EINVAL, errno) {
2166 TH_LOG("Did not reject mode strict with uargs!");
2167 }
2168
2169 /* Reject insane args for filter. */
2170 ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog);
2171 EXPECT_EQ(EINVAL, errno) {
2172 TH_LOG("Did not reject crazy filter flags!");
2173 }
2174 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL);
2175 EXPECT_EQ(EFAULT, errno) {
2176 TH_LOG("Did not reject NULL filter!");
2177 }
2178
2179 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2180 EXPECT_EQ(0, errno) {
2181 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s",
2182 strerror(errno));
2183 }
2184}
2185
2186TEST(seccomp_syscall_mode_lock)
2187{
2188 struct sock_filter filter[] = {
2189 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2190 };
2191 struct sock_fprog prog = {
2192 .len = (unsigned short)ARRAY_SIZE(filter),
2193 .filter = filter,
2194 };
2195 long ret;
2196
2197 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2198 ASSERT_EQ(0, ret) {
2199 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2200 }
2201
2202 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
b623c4da
KC		 2203 ASSERT_NE(ENOSYS, errno) {
2204 TH_LOG("Kernel does not support seccomp syscall!");
2205 }
c99ee51a
KC		 2206 EXPECT_EQ(0, ret) {
2207 TH_LOG("Could not install filter!");
2208 }
2209
2210 /* Make sure neither entry point will switch to strict. */
2211 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0);
2212 EXPECT_EQ(EINVAL, errno) {
2213 TH_LOG("Switched to mode strict!");
2214 }
2215
2216 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL);
2217 EXPECT_EQ(EINVAL, errno) {
2218 TH_LOG("Switched to mode strict!");
2219 }
2220}
2221
2b7ea5b5
TH		 2222/*
2223 * Test detection of known and unknown filter flags. Userspace needs to be able
2224 * to check if a filter flag is supported by the current kernel and a good way
2225 * of doing that is by attempting to enter filter mode, with the flag bit in
2226 * question set, and a NULL pointer for the _args_ parameter. EFAULT indicates
2227 * that the flag is valid and EINVAL indicates that the flag is invalid.
2228 */
2229TEST(detect_seccomp_filter_flags)
2230{
e66a3997 2231 unsigned int flags[] = { SECCOMP_FILTER_FLAG_TSYNC,
00a02d0c 2232 SECCOMP_FILTER_FLAG_LOG,
6a21cc50 2233 SECCOMP_FILTER_FLAG_SPEC_ALLOW,
51891498
TA		 2234 SECCOMP_FILTER_FLAG_NEW_LISTENER,
2235 SECCOMP_FILTER_FLAG_TSYNC_ESRCH };
4ee07767
KC		 2236 unsigned int exclusive[] = {
2237 SECCOMP_FILTER_FLAG_TSYNC,
2238 SECCOMP_FILTER_FLAG_NEW_LISTENER };
2239 unsigned int flag, all_flags, exclusive_mask;
2b7ea5b5
TH		 2240 int i;
2241 long ret;
2242
4ee07767 2243 /* Test detection of individual known-good filter flags */
2b7ea5b5 2244 for (i = 0, all_flags = 0; i < ARRAY_SIZE(flags); i++) {
00a02d0c
KC		 2245 int bits = 0;
2246
2b7ea5b5 2247 flag = flags[i];
00a02d0c
KC		 2248 /* Make sure the flag is a single bit! */
2249 while (flag) {
2250 if (flag & 0x1)
2251 bits ++;
2252 flag >>= 1;
2253 }
2254 ASSERT_EQ(1, bits);
2255 flag = flags[i];
2256
2b7ea5b5
TH		 2257 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2258 ASSERT_NE(ENOSYS, errno) {
2259 TH_LOG("Kernel does not support seccomp syscall!");
2260 }
2261 EXPECT_EQ(-1, ret);
2262 EXPECT_EQ(EFAULT, errno) {
2263 TH_LOG("Failed to detect that a known-good filter flag (0x%X) is supported!",
2264 flag);
2265 }
2266
2267 all_flags |= flag;
2268 }
2269
4ee07767
KC		 2270 /*
2271 * Test detection of all known-good filter flags combined. But
2272 * for the exclusive flags we need to mask them out and try them
2273 * individually for the "all flags" testing.
2274 */
2275 exclusive_mask = 0;
2276 for (i = 0; i < ARRAY_SIZE(exclusive); i++)
2277 exclusive_mask |= exclusive[i];
2278 for (i = 0; i < ARRAY_SIZE(exclusive); i++) {
2279 flag = all_flags & ~exclusive_mask;
2280 flag |= exclusive[i];
2281
2282 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2283 EXPECT_EQ(-1, ret);
2284 EXPECT_EQ(EFAULT, errno) {
2285 TH_LOG("Failed to detect that all known-good filter flags (0x%X) are supported!",
2286 flag);
2287 }
2b7ea5b5
TH		 2288 }
2289
4ee07767 2290 /* Test detection of an unknown filter flags, without exclusives. */
2b7ea5b5 2291 flag = -1;
4ee07767 2292 flag &= ~exclusive_mask;
2b7ea5b5
TH		 2293 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2294 EXPECT_EQ(-1, ret);
2295 EXPECT_EQ(EINVAL, errno) {
2296 TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported!",
2297 flag);
2298 }
2299
2300 /*
2301 * Test detection of an unknown filter flag that may simply need to be
2302 * added to this test
2303 */
2304 flag = flags[ARRAY_SIZE(flags) - 1] << 1;
2305 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2306 EXPECT_EQ(-1, ret);
2307 EXPECT_EQ(EINVAL, errno) {
2308 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?",
2309 flag);
2310 }
2311}
2312
c99ee51a
KC		 2313TEST(TSYNC_first)
2314{
2315 struct sock_filter filter[] = {
2316 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2317 };
2318 struct sock_fprog prog = {
2319 .len = (unsigned short)ARRAY_SIZE(filter),
2320 .filter = filter,
2321 };
2322 long ret;
2323
2324 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2325 ASSERT_EQ(0, ret) {
2326 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2327 }
2328
6c045d07 2329 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
c99ee51a 2330 &prog);
b623c4da
KC		 2331 ASSERT_NE(ENOSYS, errno) {
2332 TH_LOG("Kernel does not support seccomp syscall!");
2333 }
c99ee51a
KC		 2334 EXPECT_EQ(0, ret) {
2335 TH_LOG("Could not install initial filter with TSYNC!");
2336 }
2337}
2338
2339#define TSYNC_SIBLINGS 2
2340struct tsync_sibling {
2341 pthread_t tid;
2342 pid_t system_tid;
2343 sem_t *started;
2344 pthread_cond_t *cond;
2345 pthread_mutex_t *mutex;
2346 int diverge;
2347 int num_waits;
2348 struct sock_fprog *prog;
2349 struct __test_metadata *metadata;
2350};
2351
93bd70e3
KC		 2352/*
2353 * To avoid joining joined threads (which is not allowed by Bionic),
2354 * make sure we both successfully join and clear the tid to skip a
2355 * later join attempt during fixture teardown. Any remaining threads
2356 * will be directly killed during teardown.
2357 */
2358#define PTHREAD_JOIN(tid, status) \
2359 do { \
2360 int _rc = pthread_join(tid, status); \
2361 if (_rc) { \
2362 TH_LOG("pthread_join of tid %u failed: %d\n", \
2363 (unsigned int)tid, _rc); \
2364 } else { \
2365 tid = 0; \
2366 } \
2367 } while (0)
2368
1ae81d78 2369FIXTURE(TSYNC) {
c99ee51a
KC		 2370 struct sock_fprog root_prog, apply_prog;
2371 struct tsync_sibling sibling[TSYNC_SIBLINGS];
2372 sem_t started;
2373 pthread_cond_t cond;
2374 pthread_mutex_t mutex;
2375 int sibling_count;
2376};
2377
2378FIXTURE_SETUP(TSYNC)
2379{
2380 struct sock_filter root_filter[] = {
2381 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2382 };
2383 struct sock_filter apply_filter[] = {
2384 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2385 offsetof(struct seccomp_data, nr)),
2386 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
2387 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2388 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2389 };
2390
2391 memset(&self->root_prog, 0, sizeof(self->root_prog));
2392 memset(&self->apply_prog, 0, sizeof(self->apply_prog));
2393 memset(&self->sibling, 0, sizeof(self->sibling));
2394 self->root_prog.filter = malloc(sizeof(root_filter));
2395 ASSERT_NE(NULL, self->root_prog.filter);
2396 memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter));
2397 self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter);
2398
2399 self->apply_prog.filter = malloc(sizeof(apply_filter));
2400 ASSERT_NE(NULL, self->apply_prog.filter);
2401 memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter));
2402 self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter);
2403
2404 self->sibling_count = 0;
2405 pthread_mutex_init(&self->mutex, NULL);
2406 pthread_cond_init(&self->cond, NULL);
2407 sem_init(&self->started, 0, 0);
2408 self->sibling[0].tid = 0;
2409 self->sibling[0].cond = &self->cond;
2410 self->sibling[0].started = &self->started;
2411 self->sibling[0].mutex = &self->mutex;
2412 self->sibling[0].diverge = 0;
2413 self->sibling[0].num_waits = 1;
2414 self->sibling[0].prog = &self->root_prog;
2415 self->sibling[0].metadata = _metadata;
2416 self->sibling[1].tid = 0;
2417 self->sibling[1].cond = &self->cond;
2418 self->sibling[1].started = &self->started;
2419 self->sibling[1].mutex = &self->mutex;
2420 self->sibling[1].diverge = 0;
2421 self->sibling[1].prog = &self->root_prog;
2422 self->sibling[1].num_waits = 1;
2423 self->sibling[1].metadata = _metadata;
2424}
2425
2426FIXTURE_TEARDOWN(TSYNC)
2427{
2428 int sib = 0;
2429
2430 if (self->root_prog.filter)
2431 free(self->root_prog.filter);
2432 if (self->apply_prog.filter)
2433 free(self->apply_prog.filter);
2434
2435 for ( ; sib < self->sibling_count; ++sib) {
2436 struct tsync_sibling *s = &self->sibling[sib];
c99ee51a
KC		 2437
2438 if (!s->tid)
2439 continue;
93bd70e3
KC		 2440 /*
2441 * If a thread is still running, it may be stuck, so hit
2442 * it over the head really hard.
2443 */
2444 pthread_kill(s->tid, 9);
c99ee51a
KC		 2445 }
2446 pthread_mutex_destroy(&self->mutex);
2447 pthread_cond_destroy(&self->cond);
2448 sem_destroy(&self->started);
2449}
2450
2451void *tsync_sibling(void *data)
2452{
2453 long ret = 0;
2454 struct tsync_sibling *me = data;
2455
2456 me->system_tid = syscall(__NR_gettid);
2457
2458 pthread_mutex_lock(me->mutex);
2459 if (me->diverge) {
2460 /* Just re-apply the root prog to fork the tree */
2461 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
2462 me->prog, 0, 0);
2463 }
2464 sem_post(me->started);
2465 /* Return outside of started so parent notices failures. */
2466 if (ret) {
2467 pthread_mutex_unlock(me->mutex);
2468 return (void *)SIBLING_EXIT_FAILURE;
2469 }
2470 do {
2471 pthread_cond_wait(me->cond, me->mutex);
2472 me->num_waits = me->num_waits - 1;
2473 } while (me->num_waits);
2474 pthread_mutex_unlock(me->mutex);
2475
2476 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0);
2477 if (!ret)
2478 return (void *)SIBLING_EXIT_NEWPRIVS;
2479 read(0, NULL, 0);
2480 return (void *)SIBLING_EXIT_UNKILLED;
2481}
2482
2483void tsync_start_sibling(struct tsync_sibling *sibling)
2484{
2485 pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling);
2486}
2487
2488TEST_F(TSYNC, siblings_fail_prctl)
2489{
2490 long ret;
2491 void *status;
2492 struct sock_filter filter[] = {
2493 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2494 offsetof(struct seccomp_data, nr)),
2495 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
2496 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL),
2497 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2498 };
2499 struct sock_fprog prog = {
2500 .len = (unsigned short)ARRAY_SIZE(filter),
2501 .filter = filter,
2502 };
2503
2504 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2505 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2506 }
2507
2508 /* Check prctl failure detection by requesting sib 0 diverge. */
2509 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
b623c4da
KC		 2510 ASSERT_NE(ENOSYS, errno) {
2511 TH_LOG("Kernel does not support seccomp syscall!");
2512 }
c99ee51a
KC		 2513 ASSERT_EQ(0, ret) {
2514 TH_LOG("setting filter failed");
2515 }
2516
2517 self->sibling[0].diverge = 1;
2518 tsync_start_sibling(&self->sibling[0]);
2519 tsync_start_sibling(&self->sibling[1]);
2520
2521 while (self->sibling_count < TSYNC_SIBLINGS) {
2522 sem_wait(&self->started);
2523 self->sibling_count++;
2524 }
2525
2526 /* Signal the threads to clean up*/
2527 pthread_mutex_lock(&self->mutex);
2528 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2529 TH_LOG("cond broadcast non-zero");
2530 }
2531 pthread_mutex_unlock(&self->mutex);
2532
2533 /* Ensure diverging sibling failed to call prctl. */
93bd70e3 2534 PTHREAD_JOIN(self->sibling[0].tid, &status);
c99ee51a 2535 EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status);
93bd70e3 2536 PTHREAD_JOIN(self->sibling[1].tid, &status);
c99ee51a
KC		 2537 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2538}
2539
2540TEST_F(TSYNC, two_siblings_with_ancestor)
2541{
2542 long ret;
2543 void *status;
2544
2545 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2546 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2547 }
2548
2549 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
b623c4da
KC		 2550 ASSERT_NE(ENOSYS, errno) {
2551 TH_LOG("Kernel does not support seccomp syscall!");
2552 }
c99ee51a
KC		 2553 ASSERT_EQ(0, ret) {
2554 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2555 }
2556 tsync_start_sibling(&self->sibling[0]);
2557 tsync_start_sibling(&self->sibling[1]);
2558
2559 while (self->sibling_count < TSYNC_SIBLINGS) {
2560 sem_wait(&self->started);
2561 self->sibling_count++;
2562 }
2563
6c045d07 2564 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
c99ee51a
KC		 2565 &self->apply_prog);
2566 ASSERT_EQ(0, ret) {
2567 TH_LOG("Could install filter on all threads!");
2568 }
2569 /* Tell the siblings to test the policy */
2570 pthread_mutex_lock(&self->mutex);
2571 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2572 TH_LOG("cond broadcast non-zero");
2573 }
2574 pthread_mutex_unlock(&self->mutex);
2575 /* Ensure they are both killed and don't exit cleanly. */
93bd70e3 2576 PTHREAD_JOIN(self->sibling[0].tid, &status);
c99ee51a 2577 EXPECT_EQ(0x0, (long)status);
93bd70e3 2578 PTHREAD_JOIN(self->sibling[1].tid, &status);
c99ee51a
KC		 2579 EXPECT_EQ(0x0, (long)status);
2580}
2581
2582TEST_F(TSYNC, two_sibling_want_nnp)
2583{
2584 void *status;
2585
2586 /* start siblings before any prctl() operations */
2587 tsync_start_sibling(&self->sibling[0]);
2588 tsync_start_sibling(&self->sibling[1]);
2589 while (self->sibling_count < TSYNC_SIBLINGS) {
2590 sem_wait(&self->started);
2591 self->sibling_count++;
2592 }
2593
2594 /* Tell the siblings to test no policy */
2595 pthread_mutex_lock(&self->mutex);
2596 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2597 TH_LOG("cond broadcast non-zero");
2598 }
2599 pthread_mutex_unlock(&self->mutex);
2600
2601 /* Ensure they are both upset about lacking nnp. */
93bd70e3 2602 PTHREAD_JOIN(self->sibling[0].tid, &status);
c99ee51a 2603 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
93bd70e3 2604 PTHREAD_JOIN(self->sibling[1].tid, &status);
c99ee51a
KC		 2605 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2606}
2607
2608TEST_F(TSYNC, two_siblings_with_no_filter)
2609{
2610 long ret;
2611 void *status;
2612
2613 /* start siblings before any prctl() operations */
2614 tsync_start_sibling(&self->sibling[0]);
2615 tsync_start_sibling(&self->sibling[1]);
2616 while (self->sibling_count < TSYNC_SIBLINGS) {
2617 sem_wait(&self->started);
2618 self->sibling_count++;
2619 }
2620
2621 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2622 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2623 }
2624
6c045d07 2625 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
c99ee51a 2626 &self->apply_prog);
b623c4da
KC		 2627 ASSERT_NE(ENOSYS, errno) {
2628 TH_LOG("Kernel does not support seccomp syscall!");
2629 }
c99ee51a
KC		 2630 ASSERT_EQ(0, ret) {
2631 TH_LOG("Could install filter on all threads!");
2632 }
2633
2634 /* Tell the siblings to test the policy */
2635 pthread_mutex_lock(&self->mutex);
2636 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2637 TH_LOG("cond broadcast non-zero");
2638 }
2639 pthread_mutex_unlock(&self->mutex);
2640
2641 /* Ensure they are both killed and don't exit cleanly. */
93bd70e3 2642 PTHREAD_JOIN(self->sibling[0].tid, &status);
c99ee51a 2643 EXPECT_EQ(0x0, (long)status);
93bd70e3 2644 PTHREAD_JOIN(self->sibling[1].tid, &status);
c99ee51a
KC		 2645 EXPECT_EQ(0x0, (long)status);
2646}
2647
2648TEST_F(TSYNC, two_siblings_with_one_divergence)
2649{
2650 long ret;
2651 void *status;
2652
2653 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2654 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2655 }
2656
2657 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
b623c4da
KC		 2658 ASSERT_NE(ENOSYS, errno) {
2659 TH_LOG("Kernel does not support seccomp syscall!");
2660 }
c99ee51a
KC		 2661 ASSERT_EQ(0, ret) {
2662 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2663 }
2664 self->sibling[0].diverge = 1;
2665 tsync_start_sibling(&self->sibling[0]);
2666 tsync_start_sibling(&self->sibling[1]);
2667
2668 while (self->sibling_count < TSYNC_SIBLINGS) {
2669 sem_wait(&self->started);
2670 self->sibling_count++;
2671 }
2672
6c045d07 2673 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
c99ee51a
KC		 2674 &self->apply_prog);
2675 ASSERT_EQ(self->sibling[0].system_tid, ret) {
2676 TH_LOG("Did not fail on diverged sibling.");
2677 }
2678
2679 /* Wake the threads */
2680 pthread_mutex_lock(&self->mutex);
2681 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2682 TH_LOG("cond broadcast non-zero");
2683 }
2684 pthread_mutex_unlock(&self->mutex);
2685
2686 /* Ensure they are both unkilled. */
93bd70e3 2687 PTHREAD_JOIN(self->sibling[0].tid, &status);
c99ee51a 2688 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
93bd70e3 2689 PTHREAD_JOIN(self->sibling[1].tid, &status);
c99ee51a
KC		 2690 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2691}
2692
51891498
TA		 2693TEST_F(TSYNC, two_siblings_with_one_divergence_no_tid_in_err)
2694{
2695 long ret, flags;
2696 void *status;
2697
2698 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2699 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2700 }
2701
2702 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2703 ASSERT_NE(ENOSYS, errno) {
2704 TH_LOG("Kernel does not support seccomp syscall!");
2705 }
2706 ASSERT_EQ(0, ret) {
2707 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2708 }
2709 self->sibling[0].diverge = 1;
2710 tsync_start_sibling(&self->sibling[0]);
2711 tsync_start_sibling(&self->sibling[1]);
2712
2713 while (self->sibling_count < TSYNC_SIBLINGS) {
2714 sem_wait(&self->started);
2715 self->sibling_count++;
2716 }
2717
2718 flags = SECCOMP_FILTER_FLAG_TSYNC | \
2719 SECCOMP_FILTER_FLAG_TSYNC_ESRCH;
2720 ret = seccomp(SECCOMP_SET_MODE_FILTER, flags, &self->apply_prog);
2721 ASSERT_EQ(ESRCH, errno) {
2722 TH_LOG("Did not return ESRCH for diverged sibling.");
2723 }
2724 ASSERT_EQ(-1, ret) {
2725 TH_LOG("Did not fail on diverged sibling.");
2726 }
2727
2728 /* Wake the threads */
2729 pthread_mutex_lock(&self->mutex);
2730 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2731 TH_LOG("cond broadcast non-zero");
2732 }
2733 pthread_mutex_unlock(&self->mutex);
2734
2735 /* Ensure they are both unkilled. */
2736 PTHREAD_JOIN(self->sibling[0].tid, &status);
2737 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2738 PTHREAD_JOIN(self->sibling[1].tid, &status);
2739 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2740}
2741
c99ee51a
KC		 2742TEST_F(TSYNC, two_siblings_not_under_filter)
2743{
2744 long ret, sib;
2745 void *status;
ed492c2a 2746 struct timespec delay = { .tv_nsec = 100000000 };
c99ee51a
KC		 2747
2748 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2749 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2750 }
2751
2752 /*
2753 * Sibling 0 will have its own seccomp policy
2754 * and Sibling 1 will not be under seccomp at
2755 * all. Sibling 1 will enter seccomp and 0
2756 * will cause failure.
2757 */
2758 self->sibling[0].diverge = 1;
2759 tsync_start_sibling(&self->sibling[0]);
2760 tsync_start_sibling(&self->sibling[1]);
2761
2762 while (self->sibling_count < TSYNC_SIBLINGS) {
2763 sem_wait(&self->started);
2764 self->sibling_count++;
2765 }
2766
2767 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
b623c4da
KC		 2768 ASSERT_NE(ENOSYS, errno) {
2769 TH_LOG("Kernel does not support seccomp syscall!");
2770 }
c99ee51a
KC		 2771 ASSERT_EQ(0, ret) {
2772 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2773 }
2774
6c045d07 2775 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
c99ee51a
KC		 2776 &self->apply_prog);
2777 ASSERT_EQ(ret, self->sibling[0].system_tid) {
2778 TH_LOG("Did not fail on diverged sibling.");
2779 }
2780 sib = 1;
2781 if (ret == self->sibling[0].system_tid)
2782 sib = 0;
2783
2784 pthread_mutex_lock(&self->mutex);
2785
2786 /* Increment the other siblings num_waits so we can clean up
2787 * the one we just saw.
2788 */
2789 self->sibling[!sib].num_waits += 1;
2790
2791 /* Signal the thread to clean up*/
2792 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2793 TH_LOG("cond broadcast non-zero");
2794 }
2795 pthread_mutex_unlock(&self->mutex);
93bd70e3 2796 PTHREAD_JOIN(self->sibling[sib].tid, &status);
c99ee51a
KC		 2797 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2798 /* Poll for actual task death. pthread_join doesn't guarantee it. */
2799 while (!kill(self->sibling[sib].system_tid, 0))
ed492c2a 2800 nanosleep(&delay, NULL);
c99ee51a
KC		 2801 /* Switch to the remaining sibling */
2802 sib = !sib;
2803
6c045d07 2804 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
c99ee51a
KC		 2805 &self->apply_prog);
2806 ASSERT_EQ(0, ret) {
2807 TH_LOG("Expected the remaining sibling to sync");
2808 };
2809
2810 pthread_mutex_lock(&self->mutex);
2811
2812 /* If remaining sibling didn't have a chance to wake up during
2813 * the first broadcast, manually reduce the num_waits now.
2814 */
2815 if (self->sibling[sib].num_waits > 1)
2816 self->sibling[sib].num_waits = 1;
2817 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2818 TH_LOG("cond broadcast non-zero");
2819 }
2820 pthread_mutex_unlock(&self->mutex);
93bd70e3 2821 PTHREAD_JOIN(self->sibling[sib].tid, &status);
c99ee51a
KC		 2822 EXPECT_EQ(0, (long)status);
2823 /* Poll for actual task death. pthread_join doesn't guarantee it. */
2824 while (!kill(self->sibling[sib].system_tid, 0))
ed492c2a 2825 nanosleep(&delay, NULL);
c99ee51a 2826
6c045d07 2827 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
c99ee51a
KC		 2828 &self->apply_prog);
2829 ASSERT_EQ(0, ret); /* just us chickens */
2830}
2831
2832/* Make sure restarted syscalls are seen directly as "restart_syscall". */
2833TEST(syscall_restart)
2834{
2835 long ret;
2836 unsigned long msg;
2837 pid_t child_pid;
2838 int pipefd[2];
2839 int status;
2840 siginfo_t info = { };
2841 struct sock_filter filter[] = {
2842 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2843 offsetof(struct seccomp_data, nr)),
2844
2845#ifdef __NR_sigreturn
d42b8dbe 2846 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 7, 0),
c99ee51a 2847#endif
d42b8dbe
TLSC		 2848 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 6, 0),
2849 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 5, 0),
2850 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 4, 0),
2851 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 5, 0),
2852 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_clock_nanosleep, 4, 0),
c99ee51a
KC		 2853 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0),
2854
2855 /* Allow __NR_write for easy logging. */
2856 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1),
2857 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2858 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
256d0afb
KC		 2859 /* The nanosleep jump target. */
2860 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100),
2861 /* The restart_syscall jump target. */
2862 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200),
c99ee51a
KC		 2863 };
2864 struct sock_fprog prog = {
2865 .len = (unsigned short)ARRAY_SIZE(filter),
2866 .filter = filter,
2867 };
256d0afb
KC		 2868#if defined(__arm__)
2869 struct utsname utsbuf;
2870#endif
c99ee51a
KC		 2871
2872 ASSERT_EQ(0, pipe(pipefd));
2873
2874 child_pid = fork();
2875 ASSERT_LE(0, child_pid);
2876 if (child_pid == 0) {
2877 /* Child uses EXPECT not ASSERT to deliver status correctly. */
2878 char buf = ' ';
256d0afb 2879 struct timespec timeout = { };
c99ee51a
KC		 2880
2881 /* Attach parent as tracer and stop. */
2882 EXPECT_EQ(0, ptrace(PTRACE_TRACEME));
2883 EXPECT_EQ(0, raise(SIGSTOP));
2884
2885 EXPECT_EQ(0, close(pipefd[1]));
2886
2887 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2888 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2889 }
2890
2891 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2892 EXPECT_EQ(0, ret) {
2893 TH_LOG("Failed to install filter!");
2894 }
2895
2896 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2897 TH_LOG("Failed to read() sync from parent");
2898 }
2899 EXPECT_EQ('.', buf) {
2900 TH_LOG("Failed to get sync data from read()");
2901 }
2902
256d0afb
KC		 2903 /* Start nanosleep to be interrupted. */
2904 timeout.tv_sec = 1;
c99ee51a 2905 errno = 0;
256d0afb
KC		 2906 EXPECT_EQ(0, nanosleep(&timeout, NULL)) {
2907 TH_LOG("Call to nanosleep() failed (errno %d)", errno);
c99ee51a
KC		 2908 }
2909
2910 /* Read final sync from parent. */
2911 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2912 TH_LOG("Failed final read() from parent");
2913 }
2914 EXPECT_EQ('!', buf) {
2915 TH_LOG("Failed to get final data from read()");
2916 }
2917
2918 /* Directly report the status of our test harness results. */
2919 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS
2920 : EXIT_FAILURE);
2921 }
2922 EXPECT_EQ(0, close(pipefd[0]));
2923
2924 /* Attach to child, setup options, and release. */
2925 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2926 ASSERT_EQ(true, WIFSTOPPED(status));
2927 ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL,
2928 PTRACE_O_TRACESECCOMP));
2929 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2930 ASSERT_EQ(1, write(pipefd[1], ".", 1));
2931
256d0afb 2932 /* Wait for nanosleep() to start. */
c99ee51a
KC		 2933 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2934 ASSERT_EQ(true, WIFSTOPPED(status));
2935 ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2936 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2937 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2938 ASSERT_EQ(0x100, msg);
d42b8dbe
TLSC		 2939 ret = get_syscall(_metadata, child_pid);
2940 EXPECT_TRUE(ret == __NR_nanosleep || ret == __NR_clock_nanosleep);
c99ee51a
KC		 2941
2942 /* Might as well check siginfo for sanity while we're here. */
2943 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2944 ASSERT_EQ(SIGTRAP, info.si_signo);
2945 ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code);
2946 EXPECT_EQ(0, info.si_errno);
2947 EXPECT_EQ(getuid(), info.si_uid);
2948 /* Verify signal delivery came from child (seccomp-triggered). */
2949 EXPECT_EQ(child_pid, info.si_pid);
2950
256d0afb 2951 /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */
c99ee51a
KC		 2952 ASSERT_EQ(0, kill(child_pid, SIGSTOP));
2953 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2954 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2955 ASSERT_EQ(true, WIFSTOPPED(status));
2956 ASSERT_EQ(SIGSTOP, WSTOPSIG(status));
c99ee51a 2957 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2bd61abe
KC		 2958 /*
2959 * There is no siginfo on SIGSTOP any more, so we can't verify
2960 * signal delivery came from parent now (getpid() == info.si_pid).
2961 * https://lkml.kernel.org/r/CAGXu5jJaZAOzP1qFz66tYrtbuywqb+UN2SOA1VLHpCCOiYvYeg@mail.gmail.com
2962 * At least verify the SIGSTOP via PTRACE_GETSIGINFO.
2963 */
2964 EXPECT_EQ(SIGSTOP, info.si_signo);
c99ee51a 2965
256d0afb 2966 /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */
c99ee51a
KC		 2967 ASSERT_EQ(0, kill(child_pid, SIGCONT));
2968 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2969 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2970 ASSERT_EQ(true, WIFSTOPPED(status));
2971 ASSERT_EQ(SIGCONT, WSTOPSIG(status));
2972 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2973
2974 /* Wait for restart_syscall() to start. */
2975 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2976 ASSERT_EQ(true, WIFSTOPPED(status));
2977 ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2978 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2979 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
256d0afb 2980
c99ee51a
KC		 2981 ASSERT_EQ(0x200, msg);
2982 ret = get_syscall(_metadata, child_pid);
2983#if defined(__arm__)
256d0afb
KC		 2984 /*
2985 * FIXME:
2986 * - native ARM registers do NOT expose true syscall.
2987 * - compat ARM registers on ARM64 DO expose true syscall.
2988 */
2989 ASSERT_EQ(0, uname(&utsbuf));
2990 if (strncmp(utsbuf.machine, "arm", 3) == 0) {
2991 EXPECT_EQ(__NR_nanosleep, ret);
2992 } else
c99ee51a 2993#endif
256d0afb
KC		 2994 {
2995 EXPECT_EQ(__NR_restart_syscall, ret);
2996 }
c99ee51a 2997
256d0afb 2998 /* Write again to end test. */
c99ee51a
KC		 2999 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3000 ASSERT_EQ(1, write(pipefd[1], "!", 1));
3001 EXPECT_EQ(0, close(pipefd[1]));
3002
3003 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3004 if (WIFSIGNALED(status) || WEXITSTATUS(status))
3005 _metadata->passed = 0;
3006}
3007
e66a3997
TH		 3008TEST_SIGNAL(filter_flag_log, SIGSYS)
3009{
3010 struct sock_filter allow_filter[] = {
3011 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3012 };
3013 struct sock_filter kill_filter[] = {
3014 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
3015 offsetof(struct seccomp_data, nr)),
3016 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
3017 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
3018 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3019 };
3020 struct sock_fprog allow_prog = {
3021 .len = (unsigned short)ARRAY_SIZE(allow_filter),
3022 .filter = allow_filter,
3023 };
3024 struct sock_fprog kill_prog = {
3025 .len = (unsigned short)ARRAY_SIZE(kill_filter),
3026 .filter = kill_filter,
3027 };
3028 long ret;
3029 pid_t parent = getppid();
3030
3031 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3032 ASSERT_EQ(0, ret);
3033
3034 /* Verify that the FILTER_FLAG_LOG flag isn't accepted in strict mode */
3035 ret = seccomp(SECCOMP_SET_MODE_STRICT, SECCOMP_FILTER_FLAG_LOG,
3036 &allow_prog);
3037 ASSERT_NE(ENOSYS, errno) {
3038 TH_LOG("Kernel does not support seccomp syscall!");
3039 }
3040 EXPECT_NE(0, ret) {
3041 TH_LOG("Kernel accepted FILTER_FLAG_LOG flag in strict mode!");
3042 }
3043 EXPECT_EQ(EINVAL, errno) {
3044 TH_LOG("Kernel returned unexpected errno for FILTER_FLAG_LOG flag in strict mode!");
3045 }
3046
3047 /* Verify that a simple, permissive filter can be added with no flags */
3048 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog);
3049 EXPECT_EQ(0, ret);
3050
3051 /* See if the same filter can be added with the FILTER_FLAG_LOG flag */
3052 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
3053 &allow_prog);
3054 ASSERT_NE(EINVAL, errno) {
3055 TH_LOG("Kernel does not support the FILTER_FLAG_LOG flag!");
3056 }
3057 EXPECT_EQ(0, ret);
3058
3059 /* Ensure that the kill filter works with the FILTER_FLAG_LOG flag */
3060 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
3061 &kill_prog);
3062 EXPECT_EQ(0, ret);
3063
3064 EXPECT_EQ(parent, syscall(__NR_getppid));
3065 /* getpid() should never return. */
3066 EXPECT_EQ(0, syscall(__NR_getpid));
3067}
3068
d612b1fd
TH		 3069TEST(get_action_avail)
3070{
fd76875c 3071 __u32 actions[] = { SECCOMP_RET_KILL_THREAD, SECCOMP_RET_TRAP,
d612b1fd 3072 SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE,
59f5cf44 3073 SECCOMP_RET_LOG, SECCOMP_RET_ALLOW };
d612b1fd
TH		 3074 __u32 unknown_action = 0x10000000U;
3075 int i;
3076 long ret;
3077
3078 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[0]);
3079 ASSERT_NE(ENOSYS, errno) {
3080 TH_LOG("Kernel does not support seccomp syscall!");
3081 }
3082 ASSERT_NE(EINVAL, errno) {
3083 TH_LOG("Kernel does not support SECCOMP_GET_ACTION_AVAIL operation!");
3084 }
3085 EXPECT_EQ(ret, 0);
3086
3087 for (i = 0; i < ARRAY_SIZE(actions); i++) {
3088 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[i]);
3089 EXPECT_EQ(ret, 0) {
3090 TH_LOG("Expected action (0x%X) not available!",
3091 actions[i]);
3092 }
3093 }
3094
3095 /* Check that an unknown action is handled properly (EOPNOTSUPP) */
3096 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &unknown_action);
3097 EXPECT_EQ(ret, -1);
3098 EXPECT_EQ(errno, EOPNOTSUPP);
3099}
3100
d057dc4e
TA		 3101TEST(get_metadata)
3102{
3103 pid_t pid;
3104 int pipefd[2];
3105 char buf;
3106 struct seccomp_metadata md;
6c3b6d50 3107 long ret;
d057dc4e 3108
3aa415dd
TA		 3109 /* Only real root can get metadata. */
3110 if (geteuid()) {
8b1bc88c 3111 SKIP(return, "get_metadata requires real root");
3aa415dd
TA		 3112 return;
3113 }
3114
d057dc4e
TA		 3115 ASSERT_EQ(0, pipe(pipefd));
3116
3117 pid = fork();
3118 ASSERT_GE(pid, 0);
3119 if (pid == 0) {
3120 struct sock_filter filter[] = {
3121 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3122 };
3123 struct sock_fprog prog = {
3124 .len = (unsigned short)ARRAY_SIZE(filter),
3125 .filter = filter,
3126 };
3127
3128 /* one with log, one without */
fb024a07 3129 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER,
d057dc4e 3130 SECCOMP_FILTER_FLAG_LOG, &prog));
fb024a07 3131 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog));
d057dc4e 3132
fb024a07 3133 EXPECT_EQ(0, close(pipefd[0]));
d057dc4e
TA		 3134 ASSERT_EQ(1, write(pipefd[1], "1", 1));
3135 ASSERT_EQ(0, close(pipefd[1]));
3136
3137 while (1)
3138 sleep(100);
3139 }
3140
3141 ASSERT_EQ(0, close(pipefd[1]));
3142 ASSERT_EQ(1, read(pipefd[0], &buf, 1));
3143
3144 ASSERT_EQ(0, ptrace(PTRACE_ATTACH, pid));
3145 ASSERT_EQ(pid, waitpid(pid, NULL, 0));
3146
6c3b6d50
KC		 3147 /* Past here must not use ASSERT or child process is never killed. */
3148
d057dc4e 3149 md.filter_off = 0;
6c3b6d50
KC		 3150 errno = 0;
3151 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md);
3152 EXPECT_EQ(sizeof(md), ret) {
3153 if (errno == EINVAL)
8b1bc88c 3154 SKIP(goto skip, "Kernel does not support PTRACE_SECCOMP_GET_METADATA (missing CONFIG_CHECKPOINT_RESTORE?)");
6c3b6d50
KC		 3155 }
3156
d057dc4e
TA		 3157 EXPECT_EQ(md.flags, SECCOMP_FILTER_FLAG_LOG);
3158 EXPECT_EQ(md.filter_off, 0);
3159
3160 md.filter_off = 1;
6c3b6d50
KC		 3161 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md);
3162 EXPECT_EQ(sizeof(md), ret);
d057dc4e
TA		 3163 EXPECT_EQ(md.flags, 0);
3164 EXPECT_EQ(md.filter_off, 1);
3165
6c3b6d50 3166skip:
d057dc4e
TA		 3167 ASSERT_EQ(0, kill(pid, SIGKILL));
3168}
3169
279ed890 3170static int user_notif_syscall(int nr, unsigned int flags)
6a21cc50
TA		 3171{
3172 struct sock_filter filter[] = {
3173 BPF_STMT(BPF_LD+BPF_W+BPF_ABS,
3174 offsetof(struct seccomp_data, nr)),
3175 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, nr, 0, 1),
3176 BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_USER_NOTIF),
3177 BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW),
3178 };
3179
3180 struct sock_fprog prog = {
3181 .len = (unsigned short)ARRAY_SIZE(filter),
3182 .filter = filter,
3183 };
3184
3185 return seccomp(SECCOMP_SET_MODE_FILTER, flags, &prog);
3186}
3187
223e660b 3188#define USER_NOTIF_MAGIC INT_MAX
6a21cc50
TA		 3189TEST(user_notification_basic)
3190{
3191 pid_t pid;
3192 long ret;
3193 int status, listener;
3194 struct seccomp_notif req = {};
3195 struct seccomp_notif_resp resp = {};
3196 struct pollfd pollfd;
3197
3198 struct sock_filter filter[] = {
3199 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3200 };
3201 struct sock_fprog prog = {
3202 .len = (unsigned short)ARRAY_SIZE(filter),
3203 .filter = filter,
3204 };
3205
c7140706
TA		 3206 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3207 ASSERT_EQ(0, ret) {
3208 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3209 }
3210
6a21cc50
TA		 3211 pid = fork();
3212 ASSERT_GE(pid, 0);
3213
3214 /* Check that we get -ENOSYS with no listener attached */
3215 if (pid == 0) {
279ed890 3216 if (user_notif_syscall(__NR_getppid, 0) < 0)
6a21cc50 3217 exit(1);
9dd3fcb0 3218 ret = syscall(__NR_getppid);
6a21cc50
TA		 3219 exit(ret >= 0 || errno != ENOSYS);
3220 }
3221
3222 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3223 EXPECT_EQ(true, WIFEXITED(status));
3224 EXPECT_EQ(0, WEXITSTATUS(status));
3225
0b54b443 3226 /* Add some no-op filters for grins. */
6a21cc50
TA		 3227 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3228 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3229 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3230 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3231
3232 /* Check that the basic notification machinery works */
279ed890
KC		 3233 listener = user_notif_syscall(__NR_getppid,
3234 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3d244c19 3235 ASSERT_GE(listener, 0);
6a21cc50
TA		 3236
3237 /* Installing a second listener in the chain should EBUSY */
279ed890
KC		 3238 EXPECT_EQ(user_notif_syscall(__NR_getppid,
3239 SECCOMP_FILTER_FLAG_NEW_LISTENER),
6a21cc50
TA		 3240 -1);
3241 EXPECT_EQ(errno, EBUSY);
3242
3243 pid = fork();
3244 ASSERT_GE(pid, 0);
3245
3246 if (pid == 0) {
9dd3fcb0 3247 ret = syscall(__NR_getppid);
6a21cc50
TA		 3248 exit(ret != USER_NOTIF_MAGIC);
3249 }
3250
3251 pollfd.fd = listener;
3252 pollfd.events = POLLIN | POLLOUT;
3253
3254 EXPECT_GT(poll(&pollfd, 1, -1), 0);
3255 EXPECT_EQ(pollfd.revents, POLLIN);
3256
e4ab5ccc
SD		 3257 /* Test that we can't pass garbage to the kernel. */
3258 memset(&req, 0, sizeof(req));
3259 req.pid = -1;
3260 errno = 0;
3261 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req);
3262 EXPECT_EQ(-1, ret);
3263 EXPECT_EQ(EINVAL, errno);
3264
3265 if (ret) {
3266 req.pid = 0;
3267 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3268 }
6a21cc50
TA		 3269
3270 pollfd.fd = listener;
3271 pollfd.events = POLLIN | POLLOUT;
3272
3273 EXPECT_GT(poll(&pollfd, 1, -1), 0);
3274 EXPECT_EQ(pollfd.revents, POLLOUT);
3275
9dd3fcb0 3276 EXPECT_EQ(req.data.nr, __NR_getppid);
6a21cc50
TA		 3277
3278 resp.id = req.id;
3279 resp.error = 0;
3280 resp.val = USER_NOTIF_MAGIC;
3281
3282 /* check that we make sure flags == 0 */
3283 resp.flags = 1;
3284 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3285 EXPECT_EQ(errno, EINVAL);
3286
3287 resp.flags = 0;
3288 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3289
3290 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3291 EXPECT_EQ(true, WIFEXITED(status));
3292 EXPECT_EQ(0, WEXITSTATUS(status));
3293}
3294
51891498
TA		 3295TEST(user_notification_with_tsync)
3296{
3297 int ret;
3298 unsigned int flags;
3299
e4d05028
KC		 3300 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3301 ASSERT_EQ(0, ret) {
3302 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3303 }
3304
51891498
TA		 3305 /* these were exclusive */
3306 flags = SECCOMP_FILTER_FLAG_NEW_LISTENER |
3307 SECCOMP_FILTER_FLAG_TSYNC;
279ed890 3308 ASSERT_EQ(-1, user_notif_syscall(__NR_getppid, flags));
51891498
TA		 3309 ASSERT_EQ(EINVAL, errno);
3310
3311 /* but now they're not */
3312 flags |= SECCOMP_FILTER_FLAG_TSYNC_ESRCH;
279ed890 3313 ret = user_notif_syscall(__NR_getppid, flags);
51891498
TA		 3314 close(ret);
3315 ASSERT_LE(0, ret);
3316}
3317
6a21cc50
TA		 3318TEST(user_notification_kill_in_middle)
3319{
3320 pid_t pid;
3321 long ret;
3322 int listener;
3323 struct seccomp_notif req = {};
3324 struct seccomp_notif_resp resp = {};
3325
c7140706
TA		 3326 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3327 ASSERT_EQ(0, ret) {
3328 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3329 }
3330
279ed890
KC		 3331 listener = user_notif_syscall(__NR_getppid,
3332 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3d244c19 3333 ASSERT_GE(listener, 0);
6a21cc50
TA		 3334
3335 /*
3336 * Check that nothing bad happens when we kill the task in the middle
3337 * of a syscall.
3338 */
3339 pid = fork();
3340 ASSERT_GE(pid, 0);
3341
3342 if (pid == 0) {
9dd3fcb0 3343 ret = syscall(__NR_getppid);
6a21cc50
TA		 3344 exit(ret != USER_NOTIF_MAGIC);
3345 }
3346
3347 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3348 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), 0);
3349
3350 EXPECT_EQ(kill(pid, SIGKILL), 0);
3351 EXPECT_EQ(waitpid(pid, NULL, 0), pid);
3352
3353 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), -1);
3354
3355 resp.id = req.id;
3356 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp);
3357 EXPECT_EQ(ret, -1);
3358 EXPECT_EQ(errno, ENOENT);
3359}
3360
3361static int handled = -1;
3362
3363static void signal_handler(int signal)
3364{
3365 if (write(handled, "c", 1) != 1)
3366 perror("write from signal");
3367}
3368
3369TEST(user_notification_signal)
3370{
3371 pid_t pid;
3372 long ret;
3373 int status, listener, sk_pair[2];
3374 struct seccomp_notif req = {};
3375 struct seccomp_notif_resp resp = {};
3376 char c;
3377
c7140706
TA		 3378 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3379 ASSERT_EQ(0, ret) {
3380 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3381 }
3382
6a21cc50
TA		 3383 ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0);
3384
279ed890
KC		 3385 listener = user_notif_syscall(__NR_gettid,
3386 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3d244c19 3387 ASSERT_GE(listener, 0);
6a21cc50
TA		 3388
3389 pid = fork();
3390 ASSERT_GE(pid, 0);
3391
3392 if (pid == 0) {
3393 close(sk_pair[0]);
3394 handled = sk_pair[1];
3395 if (signal(SIGUSR1, signal_handler) == SIG_ERR) {
3396 perror("signal");
3397 exit(1);
3398 }
3399 /*
3400 * ERESTARTSYS behavior is a bit hard to test, because we need
3401 * to rely on a signal that has not yet been handled. Let's at
3402 * least check that the error code gets propagated through, and
3403 * hope that it doesn't break when there is actually a signal :)
3404 */
3405 ret = syscall(__NR_gettid);
3406 exit(!(ret == -1 && errno == 512));
3407 }
3408
3409 close(sk_pair[1]);
3410
88c13f8b 3411 memset(&req, 0, sizeof(req));
6a21cc50
TA		 3412 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3413
3414 EXPECT_EQ(kill(pid, SIGUSR1), 0);
3415
3416 /*
3417 * Make sure the signal really is delivered, which means we're not
3418 * stuck in the user notification code any more and the notification
3419 * should be dead.
3420 */
3421 EXPECT_EQ(read(sk_pair[0], &c, 1), 1);
3422
3423 resp.id = req.id;
3424 resp.error = -EPERM;
3425 resp.val = 0;
3426
3427 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3428 EXPECT_EQ(errno, ENOENT);
3429
88c13f8b 3430 memset(&req, 0, sizeof(req));
6a21cc50
TA		 3431 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3432
3433 resp.id = req.id;
3434 resp.error = -512; /* -ERESTARTSYS */
3435 resp.val = 0;
3436
3437 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3438
3439 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3440 EXPECT_EQ(true, WIFEXITED(status));
3441 EXPECT_EQ(0, WEXITSTATUS(status));
3442}
3443
3444TEST(user_notification_closed_listener)
3445{
3446 pid_t pid;
3447 long ret;
3448 int status, listener;
3449
c7140706
TA		 3450 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3451 ASSERT_EQ(0, ret) {
3452 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3453 }
3454
279ed890
KC		 3455 listener = user_notif_syscall(__NR_getppid,
3456 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3d244c19 3457 ASSERT_GE(listener, 0);
6a21cc50
TA		 3458
3459 /*
3460 * Check that we get an ENOSYS when the listener is closed.
3461 */
3462 pid = fork();
3463 ASSERT_GE(pid, 0);
3464 if (pid == 0) {
3465 close(listener);
9dd3fcb0 3466 ret = syscall(__NR_getppid);
6a21cc50
TA		 3467 exit(ret != -1 && errno != ENOSYS);
3468 }
3469
3470 close(listener);
3471
3472 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3473 EXPECT_EQ(true, WIFEXITED(status));
3474 EXPECT_EQ(0, WEXITSTATUS(status));
3475}
3476
3477/*
3478 * Check that a pid in a child namespace still shows up as valid in ours.
3479 */
3480TEST(user_notification_child_pid_ns)
3481{
3482 pid_t pid;
3483 int status, listener;
3484 struct seccomp_notif req = {};
3485 struct seccomp_notif_resp resp = {};
3486
d7d2e5bb
KC		 3487 ASSERT_EQ(unshare(CLONE_NEWUSER | CLONE_NEWPID), 0) {
3488 if (errno == EINVAL)
3489 SKIP(return, "kernel missing CLONE_NEWUSER support");
3490 };
6a21cc50 3491
279ed890
KC		 3492 listener = user_notif_syscall(__NR_getppid,
3493 SECCOMP_FILTER_FLAG_NEW_LISTENER);
6a21cc50
TA		 3494 ASSERT_GE(listener, 0);
3495
3496 pid = fork();
3497 ASSERT_GE(pid, 0);
3498
3499 if (pid == 0)
9dd3fcb0 3500 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
6a21cc50
TA		 3501
3502 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3503 EXPECT_EQ(req.pid, pid);
3504
3505 resp.id = req.id;
3506 resp.error = 0;
3507 resp.val = USER_NOTIF_MAGIC;
3508
3509 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3510
3511 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3512 EXPECT_EQ(true, WIFEXITED(status));
3513 EXPECT_EQ(0, WEXITSTATUS(status));
3514 close(listener);
3515}
3516
3517/*
3518 * Check that a pid in a sibling (i.e. unrelated) namespace shows up as 0, i.e.
3519 * invalid.
3520 */
3521TEST(user_notification_sibling_pid_ns)
3522{
3523 pid_t pid, pid2;
3524 int status, listener;
3525 struct seccomp_notif req = {};
3526 struct seccomp_notif_resp resp = {};
3527
c7140706
TA		 3528 ASSERT_EQ(prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0), 0) {
3529 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3530 }
3531
279ed890
KC		 3532 listener = user_notif_syscall(__NR_getppid,
3533 SECCOMP_FILTER_FLAG_NEW_LISTENER);
6a21cc50
TA		 3534 ASSERT_GE(listener, 0);
3535
3536 pid = fork();
3537 ASSERT_GE(pid, 0);
3538
3539 if (pid == 0) {
3540 ASSERT_EQ(unshare(CLONE_NEWPID), 0);
3541
3542 pid2 = fork();
3543 ASSERT_GE(pid2, 0);
3544
3545 if (pid2 == 0)
9dd3fcb0 3546 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
6a21cc50
TA		 3547
3548 EXPECT_EQ(waitpid(pid2, &status, 0), pid2);
3549 EXPECT_EQ(true, WIFEXITED(status));
3550 EXPECT_EQ(0, WEXITSTATUS(status));
3551 exit(WEXITSTATUS(status));
3552 }
3553
3554 /* Create the sibling ns, and sibling in it. */
d7d2e5bb
KC		 3555 ASSERT_EQ(unshare(CLONE_NEWPID), 0) {
3556 if (errno == EPERM)
3557 SKIP(return, "CLONE_NEWPID requires CAP_SYS_ADMIN");
3558 }
9dd3fcb0 3559 ASSERT_EQ(errno, 0);
6a21cc50
TA		 3560
3561 pid2 = fork();
9dd3fcb0 3562 ASSERT_GE(pid2, 0);
6a21cc50
TA		 3563
3564 if (pid2 == 0) {
3565 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3566 /*
3567 * The pid should be 0, i.e. the task is in some namespace that
3568 * we can't "see".
3569 */
9dd3fcb0 3570 EXPECT_EQ(req.pid, 0);
6a21cc50
TA		 3571
3572 resp.id = req.id;
3573 resp.error = 0;
3574 resp.val = USER_NOTIF_MAGIC;
3575
3576 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3577 exit(0);
3578 }
3579
3580 close(listener);
3581
3582 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3583 EXPECT_EQ(true, WIFEXITED(status));
3584 EXPECT_EQ(0, WEXITSTATUS(status));
3585
3586 EXPECT_EQ(waitpid(pid2, &status, 0), pid2);
3587 EXPECT_EQ(true, WIFEXITED(status));
3588 EXPECT_EQ(0, WEXITSTATUS(status));
3589}
3590
3591TEST(user_notification_fault_recv)
3592{
3593 pid_t pid;
3594 int status, listener;
3595 struct seccomp_notif req = {};
3596 struct seccomp_notif_resp resp = {};
3597
30d53a58
TA		 3598 ASSERT_EQ(unshare(CLONE_NEWUSER), 0);
3599
279ed890
KC		 3600 listener = user_notif_syscall(__NR_getppid,
3601 SECCOMP_FILTER_FLAG_NEW_LISTENER);
6a21cc50
TA		 3602 ASSERT_GE(listener, 0);
3603
3604 pid = fork();
3605 ASSERT_GE(pid, 0);
3606
3607 if (pid == 0)
9dd3fcb0 3608 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
6a21cc50
TA		 3609
3610 /* Do a bad recv() */
3611 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, NULL), -1);
3612 EXPECT_EQ(errno, EFAULT);
3613
3614 /* We should still be able to receive this notification, though. */
3615 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3616 EXPECT_EQ(req.pid, pid);
3617
3618 resp.id = req.id;
3619 resp.error = 0;
3620 resp.val = USER_NOTIF_MAGIC;
3621
3622 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3623
3624 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3625 EXPECT_EQ(true, WIFEXITED(status));
3626 EXPECT_EQ(0, WEXITSTATUS(status));
3627}
3628
3629TEST(seccomp_get_notif_sizes)
3630{
3631 struct seccomp_notif_sizes sizes;
3632
3d244c19 3633 ASSERT_EQ(seccomp(SECCOMP_GET_NOTIF_SIZES, 0, &sizes), 0);
6a21cc50
TA		 3634 EXPECT_EQ(sizes.seccomp_notif, sizeof(struct seccomp_notif));
3635 EXPECT_EQ(sizes.seccomp_notif_resp, sizeof(struct seccomp_notif_resp));
3636}
3637
0eebfed2
CB		 3638TEST(user_notification_continue)
3639{
3640 pid_t pid;
3641 long ret;
3642 int status, listener;
3643 struct seccomp_notif req = {};
3644 struct seccomp_notif_resp resp = {};
3645 struct pollfd pollfd;
3646
3647 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3648 ASSERT_EQ(0, ret) {
3649 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3650 }
3651
279ed890 3652 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER);
0eebfed2
CB		 3653 ASSERT_GE(listener, 0);
3654
3655 pid = fork();
3656 ASSERT_GE(pid, 0);
3657
3658 if (pid == 0) {
3659 int dup_fd, pipe_fds[2];
3660 pid_t self;
3661
cf8918db 3662 ASSERT_GE(pipe(pipe_fds), 0);
0eebfed2
CB		 3663
3664 dup_fd = dup(pipe_fds[0]);
cf8918db
KC		 3665 ASSERT_GE(dup_fd, 0);
3666 EXPECT_NE(pipe_fds[0], dup_fd);
0eebfed2
CB		 3667
3668 self = getpid();
cf8918db 3669 ASSERT_EQ(filecmp(self, self, pipe_fds[0], dup_fd), 0);
0eebfed2
CB		 3670 exit(0);
3671 }
3672
3673 pollfd.fd = listener;
3674 pollfd.events = POLLIN | POLLOUT;
3675
3676 EXPECT_GT(poll(&pollfd, 1, -1), 0);
3677 EXPECT_EQ(pollfd.revents, POLLIN);
3678
3679 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3680
3681 pollfd.fd = listener;
3682 pollfd.events = POLLIN | POLLOUT;
3683
3684 EXPECT_GT(poll(&pollfd, 1, -1), 0);
3685 EXPECT_EQ(pollfd.revents, POLLOUT);
3686
3687 EXPECT_EQ(req.data.nr, __NR_dup);
3688
3689 resp.id = req.id;
3690 resp.flags = SECCOMP_USER_NOTIF_FLAG_CONTINUE;
3691
3692 /*
3693 * Verify that setting SECCOMP_USER_NOTIF_FLAG_CONTINUE enforces other
3694 * args be set to 0.
3695 */
3696 resp.error = 0;
3697 resp.val = USER_NOTIF_MAGIC;
3698 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3699 EXPECT_EQ(errno, EINVAL);
3700
3701 resp.error = USER_NOTIF_MAGIC;
3702 resp.val = 0;
3703 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3704 EXPECT_EQ(errno, EINVAL);
3705
3706 resp.error = 0;
3707 resp.val = 0;
3708 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0) {
3709 if (errno == EINVAL)
8b1bc88c 3710 SKIP(goto skip, "Kernel does not support SECCOMP_USER_NOTIF_FLAG_CONTINUE");
0eebfed2
CB		 3711 }
3712
3713skip:
3714 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3715 EXPECT_EQ(true, WIFEXITED(status));
3716 EXPECT_EQ(0, WEXITSTATUS(status)) {
3717 if (WEXITSTATUS(status) == 2) {
8b1bc88c 3718 SKIP(return, "Kernel does not support kcmp() syscall");
0eebfed2
CB		 3719 return;
3720 }
3721 }
3722}
3723
ad568218
CB		 3724TEST(user_notification_filter_empty)
3725{
3726 pid_t pid;
3727 long ret;
3728 int status;
3729 struct pollfd pollfd;
3730 struct clone_args args = {
3731 .flags = CLONE_FILES,
3732 .exit_signal = SIGCHLD,
3733 };
3734
3735 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3736 ASSERT_EQ(0, ret) {
3737 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3738 }
3739
3740 pid = sys_clone3(&args, sizeof(args));
3741 ASSERT_GE(pid, 0);
3742
3743 if (pid == 0) {
3744 int listener;
3745
279ed890 3746 listener = user_notif_syscall(__NR_mknod, SECCOMP_FILTER_FLAG_NEW_LISTENER);
ad568218
CB		 3747 if (listener < 0)
3748 _exit(EXIT_FAILURE);
3749
3750 if (dup2(listener, 200) != 200)
3751 _exit(EXIT_FAILURE);
3752
3753 close(listener);
3754
3755 _exit(EXIT_SUCCESS);
3756 }
3757
3758 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3759 EXPECT_EQ(true, WIFEXITED(status));
3760 EXPECT_EQ(0, WEXITSTATUS(status));
3761
3762 /*
3763 * The seccomp filter has become unused so we should be notified once
3764 * the kernel gets around to cleaning up task struct.
3765 */
3766 pollfd.fd = 200;
3767 pollfd.events = POLLHUP;
3768
3769 EXPECT_GT(poll(&pollfd, 1, 2000), 0);
3770 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0);
3771}
3772
3773static void *do_thread(void *data)
3774{
3775 return NULL;
3776}
3777
3778TEST(user_notification_filter_empty_threaded)
3779{
3780 pid_t pid;
3781 long ret;
3782 int status;
3783 struct pollfd pollfd;
3784 struct clone_args args = {
3785 .flags = CLONE_FILES,
3786 .exit_signal = SIGCHLD,
3787 };
3788
3789 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3790 ASSERT_EQ(0, ret) {
3791 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3792 }
3793
3794 pid = sys_clone3(&args, sizeof(args));
3795 ASSERT_GE(pid, 0);
3796
3797 if (pid == 0) {
3798 pid_t pid1, pid2;
3799 int listener, status;
3800 pthread_t thread;
3801
279ed890 3802 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER);
ad568218
CB		 3803 if (listener < 0)
3804 _exit(EXIT_FAILURE);
3805
3806 if (dup2(listener, 200) != 200)
3807 _exit(EXIT_FAILURE);
3808
3809 close(listener);
3810
3811 pid1 = fork();
3812 if (pid1 < 0)
3813 _exit(EXIT_FAILURE);
3814
3815 if (pid1 == 0)
3816 _exit(EXIT_SUCCESS);
3817
3818 pid2 = fork();
3819 if (pid2 < 0)
3820 _exit(EXIT_FAILURE);
3821
3822 if (pid2 == 0)
3823 _exit(EXIT_SUCCESS);
3824
3825 if (pthread_create(&thread, NULL, do_thread, NULL) ||
3826 pthread_join(thread, NULL))
3827 _exit(EXIT_FAILURE);
3828
3829 if (pthread_create(&thread, NULL, do_thread, NULL) ||
3830 pthread_join(thread, NULL))
3831 _exit(EXIT_FAILURE);
3832
3833 if (waitpid(pid1, &status, 0) != pid1 || !WIFEXITED(status) ||
3834 WEXITSTATUS(status))
3835 _exit(EXIT_FAILURE);
3836
3837 if (waitpid(pid2, &status, 0) != pid2 || !WIFEXITED(status) ||
3838 WEXITSTATUS(status))
3839 _exit(EXIT_FAILURE);
3840
3841 exit(EXIT_SUCCESS);
3842 }
3843
3844 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3845 EXPECT_EQ(true, WIFEXITED(status));
3846 EXPECT_EQ(0, WEXITSTATUS(status));
3847
3848 /*
3849 * The seccomp filter has become unused so we should be notified once
3850 * the kernel gets around to cleaning up task struct.
3851 */
3852 pollfd.fd = 200;
3853 pollfd.events = POLLHUP;
3854
3855 EXPECT_GT(poll(&pollfd, 1, 2000), 0);
3856 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0);
3857}
3858
c99ee51a
KC		 3859/*
3860 * TODO:
c99ee51a
KC		 3861 * - expand NNP testing
3862 * - better arch-specific TRACE and TRAP handlers.
3863 * - endianness checking when appropriate
3864 * - 64-bit arg prodding
3865 * - arch value testing (x86 modes especially)
e66a3997 3866 * - verify that FILTER_FLAG_LOG filters generate log messages
59f5cf44 3867 * - verify that RET_LOG generates log messages
c99ee51a
KC		 3868 */
3869
3870TEST_HARNESS_MAIN
Linux 6.x block layer and io_uring tree(s)