]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - tools/testing/selftests/seccomp/seccomp_bpf.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
selftests/seccomp: Allow syscall nr and ret value to be set separately
[mirror_ubuntu-jammy-kernel.git] / tools / testing / selftests / seccomp / seccomp_bpf.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (c) 2012 The Chromium OS Authors. All rights reserved.
4 *
5 * Test code for seccomp bpf.
6 */
7
8 #define _GNU_SOURCE
9 #include <sys/types.h>
10
11 /*
12 * glibc 2.26 and later have SIGSYS in siginfo_t. Before that,
13 * we need to use the kernel's siginfo.h file and trick glibc
14 * into accepting it.
15 */
16 #if !__GLIBC_PREREQ(2, 26)
17 # include <asm/siginfo.h>
18 # define __have_siginfo_t 1
19 # define __have_sigval_t 1
20 # define __have_sigevent_t 1
21 #endif
22
23 #include <errno.h>
24 #include <linux/filter.h>
25 #include <sys/prctl.h>
26 #include <sys/ptrace.h>
27 #include <sys/user.h>
28 #include <linux/prctl.h>
29 #include <linux/ptrace.h>
30 #include <linux/seccomp.h>
31 #include <pthread.h>
32 #include <semaphore.h>
33 #include <signal.h>
34 #include <stddef.h>
35 #include <stdbool.h>
36 #include <string.h>
37 #include <time.h>
38 #include <limits.h>
39 #include <linux/elf.h>
40 #include <sys/uio.h>
41 #include <sys/utsname.h>
42 #include <sys/fcntl.h>
43 #include <sys/mman.h>
44 #include <sys/times.h>
45 #include <sys/socket.h>
46 #include <sys/ioctl.h>
47 #include <linux/kcmp.h>
48 #include <sys/resource.h>
49
50 #include <unistd.h>
51 #include <sys/syscall.h>
52 #include <poll.h>
53
54 #include "../kselftest_harness.h"
55 #include "../clone3/clone3_selftests.h"
56
57 /* Attempt to de-conflict with the selftests tree. */
58 #ifndef SKIP
59 #define SKIP(s, ...) XFAIL(s, ##__VA_ARGS__)
60 #endif
61
62 #ifndef PR_SET_PTRACER
63 # define PR_SET_PTRACER 0x59616d61
64 #endif
65
66 #ifndef PR_SET_NO_NEW_PRIVS
67 #define PR_SET_NO_NEW_PRIVS 38
68 #define PR_GET_NO_NEW_PRIVS 39
69 #endif
70
71 #ifndef PR_SECCOMP_EXT
72 #define PR_SECCOMP_EXT 43
73 #endif
74
75 #ifndef SECCOMP_EXT_ACT
76 #define SECCOMP_EXT_ACT 1
77 #endif
78
79 #ifndef SECCOMP_EXT_ACT_TSYNC
80 #define SECCOMP_EXT_ACT_TSYNC 1
81 #endif
82
83 #ifndef SECCOMP_MODE_STRICT
84 #define SECCOMP_MODE_STRICT 1
85 #endif
86
87 #ifndef SECCOMP_MODE_FILTER
88 #define SECCOMP_MODE_FILTER 2
89 #endif
90
91 #ifndef SECCOMP_RET_ALLOW
92 struct seccomp_data {
93 int nr;
94 __u32 arch;
95 __u64 instruction_pointer;
96 __u64 args[6];
97 };
98 #endif
99
100 #ifndef SECCOMP_RET_KILL_PROCESS
101 #define SECCOMP_RET_KILL_PROCESS 0x80000000U /* kill the process */
102 #define SECCOMP_RET_KILL_THREAD 0x00000000U /* kill the thread */
103 #endif
104 #ifndef SECCOMP_RET_KILL
105 #define SECCOMP_RET_KILL SECCOMP_RET_KILL_THREAD
106 #define SECCOMP_RET_TRAP 0x00030000U /* disallow and force a SIGSYS */
107 #define SECCOMP_RET_ERRNO 0x00050000U /* returns an errno */
108 #define SECCOMP_RET_TRACE 0x7ff00000U /* pass to a tracer or disallow */
109 #define SECCOMP_RET_ALLOW 0x7fff0000U /* allow */
110 #endif
111 #ifndef SECCOMP_RET_LOG
112 #define SECCOMP_RET_LOG 0x7ffc0000U /* allow after logging */
113 #endif
114
115 #ifndef __NR_seccomp
116 # if defined(__i386__)
117 # define __NR_seccomp 354
118 # elif defined(__x86_64__)
119 # define __NR_seccomp 317
120 # elif defined(__arm__)
121 # define __NR_seccomp 383
122 # elif defined(__aarch64__)
123 # define __NR_seccomp 277
124 # elif defined(__riscv)
125 # define __NR_seccomp 277
126 # elif defined(__csky__)
127 # define __NR_seccomp 277
128 # elif defined(__hppa__)
129 # define __NR_seccomp 338
130 # elif defined(__powerpc__)
131 # define __NR_seccomp 358
132 # elif defined(__s390__)
133 # define __NR_seccomp 348
134 # elif defined(__xtensa__)
135 # define __NR_seccomp 337
136 # elif defined(__sh__)
137 # define __NR_seccomp 372
138 # else
139 # warning "seccomp syscall number unknown for this architecture"
140 # define __NR_seccomp 0xffff
141 # endif
142 #endif
143
144 #ifndef SECCOMP_SET_MODE_STRICT
145 #define SECCOMP_SET_MODE_STRICT 0
146 #endif
147
148 #ifndef SECCOMP_SET_MODE_FILTER
149 #define SECCOMP_SET_MODE_FILTER 1
150 #endif
151
152 #ifndef SECCOMP_GET_ACTION_AVAIL
153 #define SECCOMP_GET_ACTION_AVAIL 2
154 #endif
155
156 #ifndef SECCOMP_GET_NOTIF_SIZES
157 #define SECCOMP_GET_NOTIF_SIZES 3
158 #endif
159
160 #ifndef SECCOMP_FILTER_FLAG_TSYNC
161 #define SECCOMP_FILTER_FLAG_TSYNC (1UL << 0)
162 #endif
163
164 #ifndef SECCOMP_FILTER_FLAG_LOG
165 #define SECCOMP_FILTER_FLAG_LOG (1UL << 1)
166 #endif
167
168 #ifndef SECCOMP_FILTER_FLAG_SPEC_ALLOW
169 #define SECCOMP_FILTER_FLAG_SPEC_ALLOW (1UL << 2)
170 #endif
171
172 #ifndef PTRACE_SECCOMP_GET_METADATA
173 #define PTRACE_SECCOMP_GET_METADATA 0x420d
174
175 struct seccomp_metadata {
176 __u64 filter_off; /* Input: which filter */
177 __u64 flags; /* Output: filter's flags */
178 };
179 #endif
180
181 #ifndef SECCOMP_FILTER_FLAG_NEW_LISTENER
182 #define SECCOMP_FILTER_FLAG_NEW_LISTENER (1UL << 3)
183 #endif
184
185 #ifndef SECCOMP_RET_USER_NOTIF
186 #define SECCOMP_RET_USER_NOTIF 0x7fc00000U
187
188 #define SECCOMP_IOC_MAGIC '!'
189 #define SECCOMP_IO(nr) _IO(SECCOMP_IOC_MAGIC, nr)
190 #define SECCOMP_IOR(nr, type) _IOR(SECCOMP_IOC_MAGIC, nr, type)
191 #define SECCOMP_IOW(nr, type) _IOW(SECCOMP_IOC_MAGIC, nr, type)
192 #define SECCOMP_IOWR(nr, type) _IOWR(SECCOMP_IOC_MAGIC, nr, type)
193
194 /* Flags for seccomp notification fd ioctl. */
195 #define SECCOMP_IOCTL_NOTIF_RECV SECCOMP_IOWR(0, struct seccomp_notif)
196 #define SECCOMP_IOCTL_NOTIF_SEND SECCOMP_IOWR(1, \
197 struct seccomp_notif_resp)
198 #define SECCOMP_IOCTL_NOTIF_ID_VALID SECCOMP_IOW(2, __u64)
199
200 struct seccomp_notif {
201 __u64 id;
202 __u32 pid;
203 __u32 flags;
204 struct seccomp_data data;
205 };
206
207 struct seccomp_notif_resp {
208 __u64 id;
209 __s64 val;
210 __s32 error;
211 __u32 flags;
212 };
213
214 struct seccomp_notif_sizes {
215 __u16 seccomp_notif;
216 __u16 seccomp_notif_resp;
217 __u16 seccomp_data;
218 };
219 #endif
220
221 #ifndef SECCOMP_IOCTL_NOTIF_ADDFD
222 /* On success, the return value is the remote process's added fd number */
223 #define SECCOMP_IOCTL_NOTIF_ADDFD SECCOMP_IOW(3, \
224 struct seccomp_notif_addfd)
225
226 /* valid flags for seccomp_notif_addfd */
227 #define SECCOMP_ADDFD_FLAG_SETFD (1UL << 0) /* Specify remote fd */
228
229 struct seccomp_notif_addfd {
230 __u64 id;
231 __u32 flags;
232 __u32 srcfd;
233 __u32 newfd;
234 __u32 newfd_flags;
235 };
236 #endif
237
238 struct seccomp_notif_addfd_small {
239 __u64 id;
240 char weird[4];
241 };
242 #define SECCOMP_IOCTL_NOTIF_ADDFD_SMALL \
243 SECCOMP_IOW(3, struct seccomp_notif_addfd_small)
244
245 struct seccomp_notif_addfd_big {
246 union {
247 struct seccomp_notif_addfd addfd;
248 char buf[sizeof(struct seccomp_notif_addfd) + 8];
249 };
250 };
251 #define SECCOMP_IOCTL_NOTIF_ADDFD_BIG \
252 SECCOMP_IOWR(3, struct seccomp_notif_addfd_big)
253
254 #ifndef PTRACE_EVENTMSG_SYSCALL_ENTRY
255 #define PTRACE_EVENTMSG_SYSCALL_ENTRY 1
256 #define PTRACE_EVENTMSG_SYSCALL_EXIT 2
257 #endif
258
259 #ifndef SECCOMP_USER_NOTIF_FLAG_CONTINUE
260 #define SECCOMP_USER_NOTIF_FLAG_CONTINUE 0x00000001
261 #endif
262
263 #ifndef SECCOMP_FILTER_FLAG_TSYNC_ESRCH
264 #define SECCOMP_FILTER_FLAG_TSYNC_ESRCH (1UL << 4)
265 #endif
266
267 #ifndef seccomp
268 int seccomp(unsigned int op, unsigned int flags, void *args)
269 {
270 errno = 0;
271 return syscall(__NR_seccomp, op, flags, args);
272 }
273 #endif
274
275 #if __BYTE_ORDER == __LITTLE_ENDIAN
276 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]))
277 #elif __BYTE_ORDER == __BIG_ENDIAN
278 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32))
279 #else
280 #error "wut? Unknown __BYTE_ORDER?!"
281 #endif
282
283 #define SIBLING_EXIT_UNKILLED 0xbadbeef
284 #define SIBLING_EXIT_FAILURE 0xbadface
285 #define SIBLING_EXIT_NEWPRIVS 0xbadfeed
286
287 static int __filecmp(pid_t pid1, pid_t pid2, int fd1, int fd2)
288 {
289 #ifdef __NR_kcmp
290 errno = 0;
291 return syscall(__NR_kcmp, pid1, pid2, KCMP_FILE, fd1, fd2);
292 #else
293 errno = ENOSYS;
294 return -1;
295 #endif
296 }
297
298 /* Have TH_LOG report actual location filecmp() is used. */
299 #define filecmp(pid1, pid2, fd1, fd2) ({ \
300 int _ret; \
301 \
302 _ret = __filecmp(pid1, pid2, fd1, fd2); \
303 if (_ret != 0) { \
304 if (_ret < 0 && errno == ENOSYS) { \
305 TH_LOG("kcmp() syscall missing (test is less accurate)");\
306 _ret = 0; \
307 } \
308 } \
309 _ret; })
310
311 TEST(kcmp)
312 {
313 int ret;
314
315 ret = __filecmp(getpid(), getpid(), 1, 1);
316 EXPECT_EQ(ret, 0);
317 if (ret != 0 && errno == ENOSYS)
318 SKIP(return, "Kernel does not support kcmp() (missing CONFIG_CHECKPOINT_RESTORE?)");
319 }
320
321 TEST(mode_strict_support)
322 {
323 long ret;
324
325 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
326 ASSERT_EQ(0, ret) {
327 TH_LOG("Kernel does not support CONFIG_SECCOMP");
328 }
329 syscall(__NR_exit, 0);
330 }
331
332 TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL)
333 {
334 long ret;
335
336 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
337 ASSERT_EQ(0, ret) {
338 TH_LOG("Kernel does not support CONFIG_SECCOMP");
339 }
340 syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
341 NULL, NULL, NULL);
342 EXPECT_FALSE(true) {
343 TH_LOG("Unreachable!");
344 }
345 }
346
347 /* Note! This doesn't test no new privs behavior */
348 TEST(no_new_privs_support)
349 {
350 long ret;
351
352 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
353 EXPECT_EQ(0, ret) {
354 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
355 }
356 }
357
358 /* Tests kernel support by checking for a copy_from_user() fault on NULL. */
359 TEST(mode_filter_support)
360 {
361 long ret;
362
363 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
364 ASSERT_EQ(0, ret) {
365 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
366 }
367 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL);
368 EXPECT_EQ(-1, ret);
369 EXPECT_EQ(EFAULT, errno) {
370 TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!");
371 }
372 }
373
374 TEST(mode_filter_without_nnp)
375 {
376 struct sock_filter filter[] = {
377 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
378 };
379 struct sock_fprog prog = {
380 .len = (unsigned short)ARRAY_SIZE(filter),
381 .filter = filter,
382 };
383 long ret;
384
385 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0);
386 ASSERT_LE(0, ret) {
387 TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS");
388 }
389 errno = 0;
390 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
391 /* Succeeds with CAP_SYS_ADMIN, fails without */
392 /* TODO(wad) check caps not euid */
393 if (geteuid()) {
394 EXPECT_EQ(-1, ret);
395 EXPECT_EQ(EACCES, errno);
396 } else {
397 EXPECT_EQ(0, ret);
398 }
399 }
400
401 #define MAX_INSNS_PER_PATH 32768
402
403 TEST(filter_size_limits)
404 {
405 int i;
406 int count = BPF_MAXINSNS + 1;
407 struct sock_filter allow[] = {
408 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
409 };
410 struct sock_filter *filter;
411 struct sock_fprog prog = { };
412 long ret;
413
414 filter = calloc(count, sizeof(*filter));
415 ASSERT_NE(NULL, filter);
416
417 for (i = 0; i < count; i++)
418 filter[i] = allow[0];
419
420 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
421 ASSERT_EQ(0, ret);
422
423 prog.filter = filter;
424 prog.len = count;
425
426 /* Too many filter instructions in a single filter. */
427 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
428 ASSERT_NE(0, ret) {
429 TH_LOG("Installing %d insn filter was allowed", prog.len);
430 }
431
432 /* One less is okay, though. */
433 prog.len -= 1;
434 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
435 ASSERT_EQ(0, ret) {
436 TH_LOG("Installing %d insn filter wasn't allowed", prog.len);
437 }
438 }
439
440 TEST(filter_chain_limits)
441 {
442 int i;
443 int count = BPF_MAXINSNS;
444 struct sock_filter allow[] = {
445 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
446 };
447 struct sock_filter *filter;
448 struct sock_fprog prog = { };
449 long ret;
450
451 filter = calloc(count, sizeof(*filter));
452 ASSERT_NE(NULL, filter);
453
454 for (i = 0; i < count; i++)
455 filter[i] = allow[0];
456
457 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
458 ASSERT_EQ(0, ret);
459
460 prog.filter = filter;
461 prog.len = 1;
462
463 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
464 ASSERT_EQ(0, ret);
465
466 prog.len = count;
467
468 /* Too many total filter instructions. */
469 for (i = 0; i < MAX_INSNS_PER_PATH; i++) {
470 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
471 if (ret != 0)
472 break;
473 }
474 ASSERT_NE(0, ret) {
475 TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)",
476 i, count, i * (count + 4));
477 }
478 }
479
480 TEST(mode_filter_cannot_move_to_strict)
481 {
482 struct sock_filter filter[] = {
483 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
484 };
485 struct sock_fprog prog = {
486 .len = (unsigned short)ARRAY_SIZE(filter),
487 .filter = filter,
488 };
489 long ret;
490
491 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
492 ASSERT_EQ(0, ret);
493
494 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
495 ASSERT_EQ(0, ret);
496
497 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0);
498 EXPECT_EQ(-1, ret);
499 EXPECT_EQ(EINVAL, errno);
500 }
501
502
503 TEST(mode_filter_get_seccomp)
504 {
505 struct sock_filter filter[] = {
506 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
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_GET_SECCOMP, 0, 0, 0, 0);
518 EXPECT_EQ(0, ret);
519
520 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
521 ASSERT_EQ(0, ret);
522
523 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
524 EXPECT_EQ(2, ret);
525 }
526
527
528 TEST(ALLOW_all)
529 {
530 struct sock_filter filter[] = {
531 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
532 };
533 struct sock_fprog prog = {
534 .len = (unsigned short)ARRAY_SIZE(filter),
535 .filter = filter,
536 };
537 long ret;
538
539 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
540 ASSERT_EQ(0, ret);
541
542 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
543 ASSERT_EQ(0, ret);
544 }
545
546 TEST(empty_prog)
547 {
548 struct sock_filter filter[] = {
549 };
550 struct sock_fprog prog = {
551 .len = (unsigned short)ARRAY_SIZE(filter),
552 .filter = filter,
553 };
554 long ret;
555
556 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
557 ASSERT_EQ(0, ret);
558
559 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
560 EXPECT_EQ(-1, ret);
561 EXPECT_EQ(EINVAL, errno);
562 }
563
564 TEST(log_all)
565 {
566 struct sock_filter filter[] = {
567 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG),
568 };
569 struct sock_fprog prog = {
570 .len = (unsigned short)ARRAY_SIZE(filter),
571 .filter = filter,
572 };
573 long ret;
574 pid_t parent = getppid();
575
576 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
577 ASSERT_EQ(0, ret);
578
579 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
580 ASSERT_EQ(0, ret);
581
582 /* getppid() should succeed and be logged (no check for logging) */
583 EXPECT_EQ(parent, syscall(__NR_getppid));
584 }
585
586 TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS)
587 {
588 struct sock_filter filter[] = {
589 BPF_STMT(BPF_RET|BPF_K, 0x10000000U),
590 };
591 struct sock_fprog prog = {
592 .len = (unsigned short)ARRAY_SIZE(filter),
593 .filter = filter,
594 };
595 long ret;
596
597 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
598 ASSERT_EQ(0, ret);
599
600 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
601 ASSERT_EQ(0, ret);
602 EXPECT_EQ(0, syscall(__NR_getpid)) {
603 TH_LOG("getpid() shouldn't ever return");
604 }
605 }
606
607 /* return code >= 0x80000000 is unused. */
608 TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS)
609 {
610 struct sock_filter filter[] = {
611 BPF_STMT(BPF_RET|BPF_K, 0x90000000U),
612 };
613 struct sock_fprog prog = {
614 .len = (unsigned short)ARRAY_SIZE(filter),
615 .filter = filter,
616 };
617 long ret;
618
619 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
620 ASSERT_EQ(0, ret);
621
622 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
623 ASSERT_EQ(0, ret);
624 EXPECT_EQ(0, syscall(__NR_getpid)) {
625 TH_LOG("getpid() shouldn't ever return");
626 }
627 }
628
629 TEST_SIGNAL(KILL_all, SIGSYS)
630 {
631 struct sock_filter filter[] = {
632 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
633 };
634 struct sock_fprog prog = {
635 .len = (unsigned short)ARRAY_SIZE(filter),
636 .filter = filter,
637 };
638 long ret;
639
640 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
641 ASSERT_EQ(0, ret);
642
643 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
644 ASSERT_EQ(0, ret);
645 }
646
647 TEST_SIGNAL(KILL_one, SIGSYS)
648 {
649 struct sock_filter filter[] = {
650 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
651 offsetof(struct seccomp_data, nr)),
652 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
653 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
654 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
655 };
656 struct sock_fprog prog = {
657 .len = (unsigned short)ARRAY_SIZE(filter),
658 .filter = filter,
659 };
660 long ret;
661 pid_t parent = getppid();
662
663 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
664 ASSERT_EQ(0, ret);
665
666 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
667 ASSERT_EQ(0, ret);
668
669 EXPECT_EQ(parent, syscall(__NR_getppid));
670 /* getpid() should never return. */
671 EXPECT_EQ(0, syscall(__NR_getpid));
672 }
673
674 TEST_SIGNAL(KILL_one_arg_one, SIGSYS)
675 {
676 void *fatal_address;
677 struct sock_filter filter[] = {
678 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
679 offsetof(struct seccomp_data, nr)),
680 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0),
681 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
682 /* Only both with lower 32-bit for now. */
683 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)),
684 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K,
685 (unsigned long)&fatal_address, 0, 1),
686 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
687 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
688 };
689 struct sock_fprog prog = {
690 .len = (unsigned short)ARRAY_SIZE(filter),
691 .filter = filter,
692 };
693 long ret;
694 pid_t parent = getppid();
695 struct tms timebuf;
696 clock_t clock = times(&timebuf);
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
704 EXPECT_EQ(parent, syscall(__NR_getppid));
705 EXPECT_LE(clock, syscall(__NR_times, &timebuf));
706 /* times() should never return. */
707 EXPECT_EQ(0, syscall(__NR_times, &fatal_address));
708 }
709
710 TEST_SIGNAL(KILL_one_arg_six, SIGSYS)
711 {
712 #ifndef __NR_mmap2
713 int sysno = __NR_mmap;
714 #else
715 int sysno = __NR_mmap2;
716 #endif
717 struct sock_filter filter[] = {
718 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
719 offsetof(struct seccomp_data, nr)),
720 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0),
721 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
722 /* Only both with lower 32-bit for now. */
723 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)),
724 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1),
725 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
726 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
727 };
728 struct sock_fprog prog = {
729 .len = (unsigned short)ARRAY_SIZE(filter),
730 .filter = filter,
731 };
732 long ret;
733 pid_t parent = getppid();
734 int fd;
735 void *map1, *map2;
736 int page_size = sysconf(_SC_PAGESIZE);
737
738 ASSERT_LT(0, page_size);
739
740 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
741 ASSERT_EQ(0, ret);
742
743 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
744 ASSERT_EQ(0, ret);
745
746 fd = open("/dev/zero", O_RDONLY);
747 ASSERT_NE(-1, fd);
748
749 EXPECT_EQ(parent, syscall(__NR_getppid));
750 map1 = (void *)syscall(sysno,
751 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size);
752 EXPECT_NE(MAP_FAILED, map1);
753 /* mmap2() should never return. */
754 map2 = (void *)syscall(sysno,
755 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE);
756 EXPECT_EQ(MAP_FAILED, map2);
757
758 /* The test failed, so clean up the resources. */
759 munmap(map1, page_size);
760 munmap(map2, page_size);
761 close(fd);
762 }
763
764 /* This is a thread task to die via seccomp filter violation. */
765 void *kill_thread(void *data)
766 {
767 bool die = (bool)data;
768
769 if (die) {
770 prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
771 return (void *)SIBLING_EXIT_FAILURE;
772 }
773
774 return (void *)SIBLING_EXIT_UNKILLED;
775 }
776
777 enum kill_t {
778 KILL_THREAD,
779 KILL_PROCESS,
780 RET_UNKNOWN
781 };
782
783 /* Prepare a thread that will kill itself or both of us. */
784 void kill_thread_or_group(struct __test_metadata *_metadata,
785 enum kill_t kill_how)
786 {
787 pthread_t thread;
788 void *status;
789 /* Kill only when calling __NR_prctl. */
790 struct sock_filter filter_thread[] = {
791 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
792 offsetof(struct seccomp_data, nr)),
793 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
794 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD),
795 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
796 };
797 struct sock_fprog prog_thread = {
798 .len = (unsigned short)ARRAY_SIZE(filter_thread),
799 .filter = filter_thread,
800 };
801 int kill = kill_how == KILL_PROCESS ? SECCOMP_RET_KILL_PROCESS : 0xAAAAAAAAA;
802 struct sock_filter filter_process[] = {
803 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
804 offsetof(struct seccomp_data, nr)),
805 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
806 BPF_STMT(BPF_RET|BPF_K, kill),
807 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
808 };
809 struct sock_fprog prog_process = {
810 .len = (unsigned short)ARRAY_SIZE(filter_process),
811 .filter = filter_process,
812 };
813
814 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
815 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
816 }
817
818 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0,
819 kill_how == KILL_THREAD ? &prog_thread
820 : &prog_process));
821
822 /*
823 * Add the KILL_THREAD rule again to make sure that the KILL_PROCESS
824 * flag cannot be downgraded by a new filter.
825 */
826 if (kill_how == KILL_PROCESS)
827 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog_thread));
828
829 /* Start a thread that will exit immediately. */
830 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)false));
831 ASSERT_EQ(0, pthread_join(thread, &status));
832 ASSERT_EQ(SIBLING_EXIT_UNKILLED, (unsigned long)status);
833
834 /* Start a thread that will die immediately. */
835 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)true));
836 ASSERT_EQ(0, pthread_join(thread, &status));
837 ASSERT_NE(SIBLING_EXIT_FAILURE, (unsigned long)status);
838
839 /*
840 * If we get here, only the spawned thread died. Let the parent know
841 * the whole process didn't die (i.e. this thread, the spawner,
842 * stayed running).
843 */
844 exit(42);
845 }
846
847 TEST(KILL_thread)
848 {
849 int status;
850 pid_t child_pid;
851
852 child_pid = fork();
853 ASSERT_LE(0, child_pid);
854 if (child_pid == 0) {
855 kill_thread_or_group(_metadata, KILL_THREAD);
856 _exit(38);
857 }
858
859 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
860
861 /* If only the thread was killed, we'll see exit 42. */
862 ASSERT_TRUE(WIFEXITED(status));
863 ASSERT_EQ(42, WEXITSTATUS(status));
864 }
865
866 TEST(KILL_process)
867 {
868 int status;
869 pid_t child_pid;
870
871 child_pid = fork();
872 ASSERT_LE(0, child_pid);
873 if (child_pid == 0) {
874 kill_thread_or_group(_metadata, KILL_PROCESS);
875 _exit(38);
876 }
877
878 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
879
880 /* If the entire process was killed, we'll see SIGSYS. */
881 ASSERT_TRUE(WIFSIGNALED(status));
882 ASSERT_EQ(SIGSYS, WTERMSIG(status));
883 }
884
885 TEST(KILL_unknown)
886 {
887 int status;
888 pid_t child_pid;
889
890 child_pid = fork();
891 ASSERT_LE(0, child_pid);
892 if (child_pid == 0) {
893 kill_thread_or_group(_metadata, RET_UNKNOWN);
894 _exit(38);
895 }
896
897 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
898
899 /* If the entire process was killed, we'll see SIGSYS. */
900 EXPECT_TRUE(WIFSIGNALED(status)) {
901 TH_LOG("Unknown SECCOMP_RET is only killing the thread?");
902 }
903 ASSERT_EQ(SIGSYS, WTERMSIG(status));
904 }
905
906 /* TODO(wad) add 64-bit versus 32-bit arg tests. */
907 TEST(arg_out_of_range)
908 {
909 struct sock_filter filter[] = {
910 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)),
911 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
912 };
913 struct sock_fprog prog = {
914 .len = (unsigned short)ARRAY_SIZE(filter),
915 .filter = filter,
916 };
917 long ret;
918
919 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
920 ASSERT_EQ(0, ret);
921
922 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
923 EXPECT_EQ(-1, ret);
924 EXPECT_EQ(EINVAL, errno);
925 }
926
927 #define ERRNO_FILTER(name, errno) \
928 struct sock_filter _read_filter_##name[] = { \
929 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, \
930 offsetof(struct seccomp_data, nr)), \
931 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), \
932 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | errno), \
933 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), \
934 }; \
935 struct sock_fprog prog_##name = { \
936 .len = (unsigned short)ARRAY_SIZE(_read_filter_##name), \
937 .filter = _read_filter_##name, \
938 }
939
940 /* Make sure basic errno values are correctly passed through a filter. */
941 TEST(ERRNO_valid)
942 {
943 ERRNO_FILTER(valid, E2BIG);
944 long ret;
945 pid_t parent = getppid();
946
947 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
948 ASSERT_EQ(0, ret);
949
950 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_valid);
951 ASSERT_EQ(0, ret);
952
953 EXPECT_EQ(parent, syscall(__NR_getppid));
954 EXPECT_EQ(-1, read(0, NULL, 0));
955 EXPECT_EQ(E2BIG, errno);
956 }
957
958 /* Make sure an errno of zero is correctly handled by the arch code. */
959 TEST(ERRNO_zero)
960 {
961 ERRNO_FILTER(zero, 0);
962 long ret;
963 pid_t parent = getppid();
964
965 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
966 ASSERT_EQ(0, ret);
967
968 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_zero);
969 ASSERT_EQ(0, ret);
970
971 EXPECT_EQ(parent, syscall(__NR_getppid));
972 /* "errno" of 0 is ok. */
973 EXPECT_EQ(0, read(0, NULL, 0));
974 }
975
976 /*
977 * The SECCOMP_RET_DATA mask is 16 bits wide, but errno is smaller.
978 * This tests that the errno value gets capped correctly, fixed by
979 * 580c57f10768 ("seccomp: cap SECCOMP_RET_ERRNO data to MAX_ERRNO").
980 */
981 TEST(ERRNO_capped)
982 {
983 ERRNO_FILTER(capped, 4096);
984 long ret;
985 pid_t parent = getppid();
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, &prog_capped);
991 ASSERT_EQ(0, ret);
992
993 EXPECT_EQ(parent, syscall(__NR_getppid));
994 EXPECT_EQ(-1, read(0, NULL, 0));
995 EXPECT_EQ(4095, errno);
996 }
997
998 /*
999 * Filters are processed in reverse order: last applied is executed first.
1000 * Since only the SECCOMP_RET_ACTION mask is tested for return values, the
1001 * SECCOMP_RET_DATA mask results will follow the most recently applied
1002 * matching filter return (and not the lowest or highest value).
1003 */
1004 TEST(ERRNO_order)
1005 {
1006 ERRNO_FILTER(first, 11);
1007 ERRNO_FILTER(second, 13);
1008 ERRNO_FILTER(third, 12);
1009 long ret;
1010 pid_t parent = getppid();
1011
1012 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1013 ASSERT_EQ(0, ret);
1014
1015 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_first);
1016 ASSERT_EQ(0, ret);
1017
1018 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_second);
1019 ASSERT_EQ(0, ret);
1020
1021 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_third);
1022 ASSERT_EQ(0, ret);
1023
1024 EXPECT_EQ(parent, syscall(__NR_getppid));
1025 EXPECT_EQ(-1, read(0, NULL, 0));
1026 EXPECT_EQ(12, errno);
1027 }
1028
1029 FIXTURE(TRAP) {
1030 struct sock_fprog prog;
1031 };
1032
1033 FIXTURE_SETUP(TRAP)
1034 {
1035 struct sock_filter filter[] = {
1036 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1037 offsetof(struct seccomp_data, nr)),
1038 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
1039 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
1040 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1041 };
1042
1043 memset(&self->prog, 0, sizeof(self->prog));
1044 self->prog.filter = malloc(sizeof(filter));
1045 ASSERT_NE(NULL, self->prog.filter);
1046 memcpy(self->prog.filter, filter, sizeof(filter));
1047 self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1048 }
1049
1050 FIXTURE_TEARDOWN(TRAP)
1051 {
1052 if (self->prog.filter)
1053 free(self->prog.filter);
1054 }
1055
1056 TEST_F_SIGNAL(TRAP, dfl, SIGSYS)
1057 {
1058 long ret;
1059
1060 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1061 ASSERT_EQ(0, ret);
1062
1063 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
1064 ASSERT_EQ(0, ret);
1065 syscall(__NR_getpid);
1066 }
1067
1068 /* Ensure that SIGSYS overrides SIG_IGN */
1069 TEST_F_SIGNAL(TRAP, ign, SIGSYS)
1070 {
1071 long ret;
1072
1073 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1074 ASSERT_EQ(0, ret);
1075
1076 signal(SIGSYS, SIG_IGN);
1077
1078 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
1079 ASSERT_EQ(0, ret);
1080 syscall(__NR_getpid);
1081 }
1082
1083 static siginfo_t TRAP_info;
1084 static volatile int TRAP_nr;
1085 static void TRAP_action(int nr, siginfo_t *info, void *void_context)
1086 {
1087 memcpy(&TRAP_info, info, sizeof(TRAP_info));
1088 TRAP_nr = nr;
1089 }
1090
1091 TEST_F(TRAP, handler)
1092 {
1093 int ret, test;
1094 struct sigaction act;
1095 sigset_t mask;
1096
1097 memset(&act, 0, sizeof(act));
1098 sigemptyset(&mask);
1099 sigaddset(&mask, SIGSYS);
1100
1101 act.sa_sigaction = &TRAP_action;
1102 act.sa_flags = SA_SIGINFO;
1103 ret = sigaction(SIGSYS, &act, NULL);
1104 ASSERT_EQ(0, ret) {
1105 TH_LOG("sigaction failed");
1106 }
1107 ret = sigprocmask(SIG_UNBLOCK, &mask, NULL);
1108 ASSERT_EQ(0, ret) {
1109 TH_LOG("sigprocmask failed");
1110 }
1111
1112 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1113 ASSERT_EQ(0, ret);
1114 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
1115 ASSERT_EQ(0, ret);
1116 TRAP_nr = 0;
1117 memset(&TRAP_info, 0, sizeof(TRAP_info));
1118 /* Expect the registers to be rolled back. (nr = error) may vary
1119 * based on arch. */
1120 ret = syscall(__NR_getpid);
1121 /* Silence gcc warning about volatile. */
1122 test = TRAP_nr;
1123 EXPECT_EQ(SIGSYS, test);
1124 struct local_sigsys {
1125 void *_call_addr; /* calling user insn */
1126 int _syscall; /* triggering system call number */
1127 unsigned int _arch; /* AUDIT_ARCH_* of syscall */
1128 } *sigsys = (struct local_sigsys *)
1129 #ifdef si_syscall
1130 &(TRAP_info.si_call_addr);
1131 #else
1132 &TRAP_info.si_pid;
1133 #endif
1134 EXPECT_EQ(__NR_getpid, sigsys->_syscall);
1135 /* Make sure arch is non-zero. */
1136 EXPECT_NE(0, sigsys->_arch);
1137 EXPECT_NE(0, (unsigned long)sigsys->_call_addr);
1138 }
1139
1140 FIXTURE(precedence) {
1141 struct sock_fprog allow;
1142 struct sock_fprog log;
1143 struct sock_fprog trace;
1144 struct sock_fprog error;
1145 struct sock_fprog trap;
1146 struct sock_fprog kill;
1147 };
1148
1149 FIXTURE_SETUP(precedence)
1150 {
1151 struct sock_filter allow_insns[] = {
1152 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1153 };
1154 struct sock_filter log_insns[] = {
1155 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1156 offsetof(struct seccomp_data, nr)),
1157 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1158 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1159 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG),
1160 };
1161 struct sock_filter trace_insns[] = {
1162 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1163 offsetof(struct seccomp_data, nr)),
1164 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1165 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1166 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE),
1167 };
1168 struct sock_filter error_insns[] = {
1169 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1170 offsetof(struct seccomp_data, nr)),
1171 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1172 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1173 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO),
1174 };
1175 struct sock_filter trap_insns[] = {
1176 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1177 offsetof(struct seccomp_data, nr)),
1178 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1179 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1180 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
1181 };
1182 struct sock_filter kill_insns[] = {
1183 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1184 offsetof(struct seccomp_data, nr)),
1185 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1186 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1187 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1188 };
1189
1190 memset(self, 0, sizeof(*self));
1191 #define FILTER_ALLOC(_x) \
1192 self->_x.filter = malloc(sizeof(_x##_insns)); \
1193 ASSERT_NE(NULL, self->_x.filter); \
1194 memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \
1195 self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns)
1196 FILTER_ALLOC(allow);
1197 FILTER_ALLOC(log);
1198 FILTER_ALLOC(trace);
1199 FILTER_ALLOC(error);
1200 FILTER_ALLOC(trap);
1201 FILTER_ALLOC(kill);
1202 }
1203
1204 FIXTURE_TEARDOWN(precedence)
1205 {
1206 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter)
1207 FILTER_FREE(allow);
1208 FILTER_FREE(log);
1209 FILTER_FREE(trace);
1210 FILTER_FREE(error);
1211 FILTER_FREE(trap);
1212 FILTER_FREE(kill);
1213 }
1214
1215 TEST_F(precedence, allow_ok)
1216 {
1217 pid_t parent, res = 0;
1218 long ret;
1219
1220 parent = getppid();
1221 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1222 ASSERT_EQ(0, ret);
1223
1224 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1225 ASSERT_EQ(0, ret);
1226 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1227 ASSERT_EQ(0, ret);
1228 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1229 ASSERT_EQ(0, ret);
1230 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1231 ASSERT_EQ(0, ret);
1232 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1233 ASSERT_EQ(0, ret);
1234 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1235 ASSERT_EQ(0, ret);
1236 /* Should work just fine. */
1237 res = syscall(__NR_getppid);
1238 EXPECT_EQ(parent, res);
1239 }
1240
1241 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS)
1242 {
1243 pid_t parent, res = 0;
1244 long ret;
1245
1246 parent = getppid();
1247 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1248 ASSERT_EQ(0, ret);
1249
1250 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1251 ASSERT_EQ(0, ret);
1252 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1253 ASSERT_EQ(0, ret);
1254 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1255 ASSERT_EQ(0, ret);
1256 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1257 ASSERT_EQ(0, ret);
1258 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1259 ASSERT_EQ(0, ret);
1260 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1261 ASSERT_EQ(0, ret);
1262 /* Should work just fine. */
1263 res = syscall(__NR_getppid);
1264 EXPECT_EQ(parent, res);
1265 /* getpid() should never return. */
1266 res = syscall(__NR_getpid);
1267 EXPECT_EQ(0, res);
1268 }
1269
1270 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS)
1271 {
1272 pid_t parent;
1273 long ret;
1274
1275 parent = getppid();
1276 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1277 ASSERT_EQ(0, ret);
1278
1279 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1280 ASSERT_EQ(0, ret);
1281 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1282 ASSERT_EQ(0, ret);
1283 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1284 ASSERT_EQ(0, ret);
1285 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1286 ASSERT_EQ(0, ret);
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->trap);
1290 ASSERT_EQ(0, ret);
1291 /* Should work just fine. */
1292 EXPECT_EQ(parent, syscall(__NR_getppid));
1293 /* getpid() should never return. */
1294 EXPECT_EQ(0, syscall(__NR_getpid));
1295 }
1296
1297 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS)
1298 {
1299 pid_t parent;
1300 long ret;
1301
1302 parent = getppid();
1303 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1304 ASSERT_EQ(0, ret);
1305
1306 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1307 ASSERT_EQ(0, ret);
1308 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1309 ASSERT_EQ(0, ret);
1310 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1311 ASSERT_EQ(0, ret);
1312 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1313 ASSERT_EQ(0, ret);
1314 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1315 ASSERT_EQ(0, ret);
1316 /* Should work just fine. */
1317 EXPECT_EQ(parent, syscall(__NR_getppid));
1318 /* getpid() should never return. */
1319 EXPECT_EQ(0, syscall(__NR_getpid));
1320 }
1321
1322 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS)
1323 {
1324 pid_t parent;
1325 long ret;
1326
1327 parent = getppid();
1328 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1329 ASSERT_EQ(0, ret);
1330
1331 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1332 ASSERT_EQ(0, ret);
1333 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1334 ASSERT_EQ(0, ret);
1335 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1336 ASSERT_EQ(0, ret);
1337 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1338 ASSERT_EQ(0, ret);
1339 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1340 ASSERT_EQ(0, ret);
1341 /* Should work just fine. */
1342 EXPECT_EQ(parent, syscall(__NR_getppid));
1343 /* getpid() should never return. */
1344 EXPECT_EQ(0, syscall(__NR_getpid));
1345 }
1346
1347 TEST_F(precedence, errno_is_third)
1348 {
1349 pid_t parent;
1350 long ret;
1351
1352 parent = getppid();
1353 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1354 ASSERT_EQ(0, ret);
1355
1356 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1357 ASSERT_EQ(0, ret);
1358 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1359 ASSERT_EQ(0, ret);
1360 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1361 ASSERT_EQ(0, ret);
1362 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1363 ASSERT_EQ(0, ret);
1364 /* Should work just fine. */
1365 EXPECT_EQ(parent, syscall(__NR_getppid));
1366 EXPECT_EQ(0, syscall(__NR_getpid));
1367 }
1368
1369 TEST_F(precedence, errno_is_third_in_any_order)
1370 {
1371 pid_t parent;
1372 long ret;
1373
1374 parent = getppid();
1375 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1376 ASSERT_EQ(0, ret);
1377
1378 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1379 ASSERT_EQ(0, ret);
1380 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1381 ASSERT_EQ(0, ret);
1382 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1383 ASSERT_EQ(0, ret);
1384 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1385 ASSERT_EQ(0, ret);
1386 /* Should work just fine. */
1387 EXPECT_EQ(parent, syscall(__NR_getppid));
1388 EXPECT_EQ(0, syscall(__NR_getpid));
1389 }
1390
1391 TEST_F(precedence, trace_is_fourth)
1392 {
1393 pid_t parent;
1394 long ret;
1395
1396 parent = getppid();
1397 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1398 ASSERT_EQ(0, ret);
1399
1400 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1401 ASSERT_EQ(0, ret);
1402 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1403 ASSERT_EQ(0, ret);
1404 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1405 ASSERT_EQ(0, ret);
1406 /* Should work just fine. */
1407 EXPECT_EQ(parent, syscall(__NR_getppid));
1408 /* No ptracer */
1409 EXPECT_EQ(-1, syscall(__NR_getpid));
1410 }
1411
1412 TEST_F(precedence, trace_is_fourth_in_any_order)
1413 {
1414 pid_t parent;
1415 long ret;
1416
1417 parent = getppid();
1418 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1419 ASSERT_EQ(0, ret);
1420
1421 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1422 ASSERT_EQ(0, ret);
1423 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1424 ASSERT_EQ(0, ret);
1425 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1426 ASSERT_EQ(0, ret);
1427 /* Should work just fine. */
1428 EXPECT_EQ(parent, syscall(__NR_getppid));
1429 /* No ptracer */
1430 EXPECT_EQ(-1, syscall(__NR_getpid));
1431 }
1432
1433 TEST_F(precedence, log_is_fifth)
1434 {
1435 pid_t mypid, parent;
1436 long ret;
1437
1438 mypid = getpid();
1439 parent = getppid();
1440 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1441 ASSERT_EQ(0, ret);
1442
1443 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1444 ASSERT_EQ(0, ret);
1445 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1446 ASSERT_EQ(0, ret);
1447 /* Should work just fine. */
1448 EXPECT_EQ(parent, syscall(__NR_getppid));
1449 /* Should also work just fine */
1450 EXPECT_EQ(mypid, syscall(__NR_getpid));
1451 }
1452
1453 TEST_F(precedence, log_is_fifth_in_any_order)
1454 {
1455 pid_t mypid, parent;
1456 long ret;
1457
1458 mypid = getpid();
1459 parent = getppid();
1460 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1461 ASSERT_EQ(0, ret);
1462
1463 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1464 ASSERT_EQ(0, ret);
1465 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1466 ASSERT_EQ(0, ret);
1467 /* Should work just fine. */
1468 EXPECT_EQ(parent, syscall(__NR_getppid));
1469 /* Should also work just fine */
1470 EXPECT_EQ(mypid, syscall(__NR_getpid));
1471 }
1472
1473 #ifndef PTRACE_O_TRACESECCOMP
1474 #define PTRACE_O_TRACESECCOMP 0x00000080
1475 #endif
1476
1477 /* Catch the Ubuntu 12.04 value error. */
1478 #if PTRACE_EVENT_SECCOMP != 7
1479 #undef PTRACE_EVENT_SECCOMP
1480 #endif
1481
1482 #ifndef PTRACE_EVENT_SECCOMP
1483 #define PTRACE_EVENT_SECCOMP 7
1484 #endif
1485
1486 #define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP)
1487 bool tracer_running;
1488 void tracer_stop(int sig)
1489 {
1490 tracer_running = false;
1491 }
1492
1493 typedef void tracer_func_t(struct __test_metadata *_metadata,
1494 pid_t tracee, int status, void *args);
1495
1496 void start_tracer(struct __test_metadata *_metadata, int fd, pid_t tracee,
1497 tracer_func_t tracer_func, void *args, bool ptrace_syscall)
1498 {
1499 int ret = -1;
1500 struct sigaction action = {
1501 .sa_handler = tracer_stop,
1502 };
1503
1504 /* Allow external shutdown. */
1505 tracer_running = true;
1506 ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL));
1507
1508 errno = 0;
1509 while (ret == -1 && errno != EINVAL)
1510 ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0);
1511 ASSERT_EQ(0, ret) {
1512 kill(tracee, SIGKILL);
1513 }
1514 /* Wait for attach stop */
1515 wait(NULL);
1516
1517 ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, ptrace_syscall ?
1518 PTRACE_O_TRACESYSGOOD :
1519 PTRACE_O_TRACESECCOMP);
1520 ASSERT_EQ(0, ret) {
1521 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP");
1522 kill(tracee, SIGKILL);
1523 }
1524 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1525 tracee, NULL, 0);
1526 ASSERT_EQ(0, ret);
1527
1528 /* Unblock the tracee */
1529 ASSERT_EQ(1, write(fd, "A", 1));
1530 ASSERT_EQ(0, close(fd));
1531
1532 /* Run until we're shut down. Must assert to stop execution. */
1533 while (tracer_running) {
1534 int status;
1535
1536 if (wait(&status) != tracee)
1537 continue;
1538 if (WIFSIGNALED(status) || WIFEXITED(status))
1539 /* Child is dead. Time to go. */
1540 return;
1541
1542 /* Check if this is a seccomp event. */
1543 ASSERT_EQ(!ptrace_syscall, IS_SECCOMP_EVENT(status));
1544
1545 tracer_func(_metadata, tracee, status, args);
1546
1547 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1548 tracee, NULL, 0);
1549 ASSERT_EQ(0, ret);
1550 }
1551 /* Directly report the status of our test harness results. */
1552 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS : EXIT_FAILURE);
1553 }
1554
1555 /* Common tracer setup/teardown functions. */
1556 void cont_handler(int num)
1557 { }
1558 pid_t setup_trace_fixture(struct __test_metadata *_metadata,
1559 tracer_func_t func, void *args, bool ptrace_syscall)
1560 {
1561 char sync;
1562 int pipefd[2];
1563 pid_t tracer_pid;
1564 pid_t tracee = getpid();
1565
1566 /* Setup a pipe for clean synchronization. */
1567 ASSERT_EQ(0, pipe(pipefd));
1568
1569 /* Fork a child which we'll promote to tracer */
1570 tracer_pid = fork();
1571 ASSERT_LE(0, tracer_pid);
1572 signal(SIGALRM, cont_handler);
1573 if (tracer_pid == 0) {
1574 close(pipefd[0]);
1575 start_tracer(_metadata, pipefd[1], tracee, func, args,
1576 ptrace_syscall);
1577 syscall(__NR_exit, 0);
1578 }
1579 close(pipefd[1]);
1580 prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0);
1581 read(pipefd[0], &sync, 1);
1582 close(pipefd[0]);
1583
1584 return tracer_pid;
1585 }
1586
1587 void teardown_trace_fixture(struct __test_metadata *_metadata,
1588 pid_t tracer)
1589 {
1590 if (tracer) {
1591 int status;
1592 /*
1593 * Extract the exit code from the other process and
1594 * adopt it for ourselves in case its asserts failed.
1595 */
1596 ASSERT_EQ(0, kill(tracer, SIGUSR1));
1597 ASSERT_EQ(tracer, waitpid(tracer, &status, 0));
1598 if (WEXITSTATUS(status))
1599 _metadata->passed = 0;
1600 }
1601 }
1602
1603 /* "poke" tracer arguments and function. */
1604 struct tracer_args_poke_t {
1605 unsigned long poke_addr;
1606 };
1607
1608 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status,
1609 void *args)
1610 {
1611 int ret;
1612 unsigned long msg;
1613 struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args;
1614
1615 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1616 EXPECT_EQ(0, ret);
1617 /* If this fails, don't try to recover. */
1618 ASSERT_EQ(0x1001, msg) {
1619 kill(tracee, SIGKILL);
1620 }
1621 /*
1622 * Poke in the message.
1623 * Registers are not touched to try to keep this relatively arch
1624 * agnostic.
1625 */
1626 ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001);
1627 EXPECT_EQ(0, ret);
1628 }
1629
1630 FIXTURE(TRACE_poke) {
1631 struct sock_fprog prog;
1632 pid_t tracer;
1633 long poked;
1634 struct tracer_args_poke_t tracer_args;
1635 };
1636
1637 FIXTURE_SETUP(TRACE_poke)
1638 {
1639 struct sock_filter filter[] = {
1640 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1641 offsetof(struct seccomp_data, nr)),
1642 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1643 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001),
1644 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1645 };
1646
1647 self->poked = 0;
1648 memset(&self->prog, 0, sizeof(self->prog));
1649 self->prog.filter = malloc(sizeof(filter));
1650 ASSERT_NE(NULL, self->prog.filter);
1651 memcpy(self->prog.filter, filter, sizeof(filter));
1652 self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1653
1654 /* Set up tracer args. */
1655 self->tracer_args.poke_addr = (unsigned long)&self->poked;
1656
1657 /* Launch tracer. */
1658 self->tracer = setup_trace_fixture(_metadata, tracer_poke,
1659 &self->tracer_args, false);
1660 }
1661
1662 FIXTURE_TEARDOWN(TRACE_poke)
1663 {
1664 teardown_trace_fixture(_metadata, self->tracer);
1665 if (self->prog.filter)
1666 free(self->prog.filter);
1667 }
1668
1669 TEST_F(TRACE_poke, read_has_side_effects)
1670 {
1671 ssize_t ret;
1672
1673 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1674 ASSERT_EQ(0, ret);
1675
1676 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1677 ASSERT_EQ(0, ret);
1678
1679 EXPECT_EQ(0, self->poked);
1680 ret = read(-1, NULL, 0);
1681 EXPECT_EQ(-1, ret);
1682 EXPECT_EQ(0x1001, self->poked);
1683 }
1684
1685 TEST_F(TRACE_poke, getpid_runs_normally)
1686 {
1687 long ret;
1688
1689 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1690 ASSERT_EQ(0, ret);
1691
1692 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1693 ASSERT_EQ(0, ret);
1694
1695 EXPECT_EQ(0, self->poked);
1696 EXPECT_NE(0, syscall(__NR_getpid));
1697 EXPECT_EQ(0, self->poked);
1698 }
1699
1700 #if defined(__x86_64__)
1701 # define ARCH_REGS struct user_regs_struct
1702 # define SYSCALL_NUM(_regs) (_regs).orig_rax
1703 # define SYSCALL_RET(_regs) (_regs).rax
1704 #elif defined(__i386__)
1705 # define ARCH_REGS struct user_regs_struct
1706 # define SYSCALL_NUM(_regs) (_regs).orig_eax
1707 # define SYSCALL_RET(_regs) (_regs).eax
1708 #elif defined(__arm__)
1709 # define ARCH_REGS struct pt_regs
1710 # define SYSCALL_NUM(_regs) (_regs).ARM_r7
1711 # ifndef PTRACE_SET_SYSCALL
1712 # define PTRACE_SET_SYSCALL 23
1713 # endif
1714 # define SYSCALL_NUM_SET(_regs, _nr) \
1715 EXPECT_EQ(0, ptrace(PTRACE_SET_SYSCALL, tracee, NULL, _nr))
1716 # define SYSCALL_RET(_regs) (_regs).ARM_r0
1717 #elif defined(__aarch64__)
1718 # define ARCH_REGS struct user_pt_regs
1719 # define SYSCALL_NUM(_regs) (_regs).regs[8]
1720 # ifndef NT_ARM_SYSTEM_CALL
1721 # define NT_ARM_SYSTEM_CALL 0x404
1722 # endif
1723 # define SYSCALL_NUM_SET(_regs, _nr) \
1724 do { \
1725 struct iovec __v; \
1726 typeof(_nr) __nr = (_nr); \
1727 __v.iov_base = &__nr; \
1728 __v.iov_len = sizeof(__nr); \
1729 EXPECT_EQ(0, ptrace(PTRACE_SETREGSET, tracee, \
1730 NT_ARM_SYSTEM_CALL, &__v)); \
1731 } while (0)
1732 # define SYSCALL_RET(_regs) (_regs).regs[0]
1733 #elif defined(__riscv) && __riscv_xlen == 64
1734 # define ARCH_REGS struct user_regs_struct
1735 # define SYSCALL_NUM(_regs) (_regs).a7
1736 # define SYSCALL_RET(_regs) (_regs).a0
1737 #elif defined(__csky__)
1738 # define ARCH_REGS struct pt_regs
1739 # if defined(__CSKYABIV2__)
1740 # define SYSCALL_NUM(_regs) (_regs).regs[3]
1741 # else
1742 # define SYSCALL_NUM(_regs) (_regs).regs[9]
1743 # endif
1744 # define SYSCALL_RET(_regs) (_regs).a0
1745 #elif defined(__hppa__)
1746 # define ARCH_REGS struct user_regs_struct
1747 # define SYSCALL_NUM(_regs) (_regs).gr[20]
1748 # define SYSCALL_RET(_regs) (_regs).gr[28]
1749 #elif defined(__powerpc__)
1750 # define ARCH_REGS struct pt_regs
1751 # define SYSCALL_NUM(_regs) (_regs).gpr[0]
1752 # define SYSCALL_RET(_regs) (_regs).gpr[3]
1753 # define SYSCALL_RET_SET(_regs, _val) \
1754 do { \
1755 typeof(_val) _result = (_val); \
1756 /* \
1757 * A syscall error is signaled by CR0 SO bit \
1758 * and the code is stored as a positive value. \
1759 */ \
1760 if (_result < 0) { \
1761 SYSCALL_RET(_regs) = -result; \
1762 (_regs).ccr |= 0x10000000; \
1763 } else { \
1764 SYSCALL_RET(_regs) = result; \
1765 (_regs).ccr &= ~0x10000000; \
1766 } \
1767 } while (0)
1768 #elif defined(__s390__)
1769 # define ARCH_REGS s390_regs
1770 # define SYSCALL_NUM(_regs) (_regs).gprs[2]
1771 # define SYSCALL_RET_SET(_regs, _val) \
1772 TH_LOG("Can't modify syscall return on this architecture")
1773 #elif defined(__mips__)
1774 # include <asm/unistd_nr_n32.h>
1775 # include <asm/unistd_nr_n64.h>
1776 # include <asm/unistd_nr_o32.h>
1777 # define ARCH_REGS struct pt_regs
1778 # define SYSCALL_NUM(_regs) \
1779 ({ \
1780 typeof((_regs).regs[2]) _nr; \
1781 if ((_regs).regs[2] == __NR_O32_Linux) \
1782 _nr = (_regs).regs[4]; \
1783 else \
1784 _nr = (_regs).regs[2]; \
1785 _nr; \
1786 })
1787 # define SYSCALL_NUM_SET(_regs, _nr) \
1788 do { \
1789 if ((_regs).regs[2] == __NR_O32_Linux) \
1790 (_regs).regs[4] = _nr; \
1791 else \
1792 (_regs).regs[2] = _nr; \
1793 } while (0)
1794 # define SYSCALL_RET_SET(_regs, _val) \
1795 TH_LOG("Can't modify syscall return on this architecture")
1796 #elif defined(__xtensa__)
1797 # define ARCH_REGS struct user_pt_regs
1798 # define SYSCALL_NUM(_regs) (_regs).syscall
1799 /*
1800 * On xtensa syscall return value is in the register
1801 * a2 of the current window which is not fixed.
1802 */
1803 #define SYSCALL_RET(_regs) (_regs).a[(_regs).windowbase * 4 + 2]
1804 #elif defined(__sh__)
1805 # define ARCH_REGS struct pt_regs
1806 # define SYSCALL_NUM(_regs) (_regs).gpr[3]
1807 # define SYSCALL_RET(_regs) (_regs).gpr[0]
1808 #else
1809 # error "Do not know how to find your architecture's registers and syscalls"
1810 #endif
1811
1812 /*
1813 * Most architectures can change the syscall by just updating the
1814 * associated register. This is the default if not defined above.
1815 */
1816 #ifndef SYSCALL_NUM_SET
1817 # define SYSCALL_NUM_SET(_regs, _nr) \
1818 do { \
1819 SYSCALL_NUM(_regs) = (_nr); \
1820 } while (0)
1821 #endif
1822 /*
1823 * Most architectures can change the syscall return value by just
1824 * writing to the SYSCALL_RET register. This is the default if not
1825 * defined above. If an architecture cannot set the return value
1826 * (for example when the syscall and return value register is
1827 * shared), report it with TH_LOG() in an arch-specific definition
1828 * of SYSCALL_RET_SET() above, and leave SYSCALL_RET undefined.
1829 */
1830 #if !defined(SYSCALL_RET) && !defined(SYSCALL_RET_SET)
1831 # error "One of SYSCALL_RET or SYSCALL_RET_SET is needed for this arch"
1832 #endif
1833 #ifndef SYSCALL_RET_SET
1834 # define SYSCALL_RET_SET(_regs, _val) \
1835 do { \
1836 SYSCALL_RET(_regs) = (_val); \
1837 } while (0)
1838 #endif
1839
1840 /* When the syscall return can't be changed, stub out the tests for it. */
1841 #ifndef SYSCALL_RET
1842 # define EXPECT_SYSCALL_RETURN(val, action) EXPECT_EQ(-1, action)
1843 #else
1844 # define EXPECT_SYSCALL_RETURN(val, action) \
1845 do { \
1846 errno = 0; \
1847 if (val < 0) { \
1848 EXPECT_EQ(-1, action); \
1849 EXPECT_EQ(-(val), errno); \
1850 } else { \
1851 EXPECT_EQ(val, action); \
1852 } \
1853 } while (0)
1854 #endif
1855
1856 /*
1857 * Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for
1858 * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux).
1859 */
1860 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__)
1861 # define ARCH_GETREGS(_regs) ptrace(PTRACE_GETREGS, tracee, 0, &(_regs))
1862 # define ARCH_SETREGS(_regs) ptrace(PTRACE_SETREGS, tracee, 0, &(_regs))
1863 #else
1864 # define ARCH_GETREGS(_regs) ({ \
1865 struct iovec __v; \
1866 __v.iov_base = &(_regs); \
1867 __v.iov_len = sizeof(_regs); \
1868 ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &__v); \
1869 })
1870 # define ARCH_SETREGS(_regs) ({ \
1871 struct iovec __v; \
1872 __v.iov_base = &(_regs); \
1873 __v.iov_len = sizeof(_regs); \
1874 ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &__v); \
1875 })
1876 #endif
1877
1878 /* Architecture-specific syscall fetching routine. */
1879 int get_syscall(struct __test_metadata *_metadata, pid_t tracee)
1880 {
1881 ARCH_REGS regs;
1882
1883 EXPECT_EQ(0, ARCH_GETREGS(regs)) {
1884 return -1;
1885 }
1886
1887 return SYSCALL_NUM(regs);
1888 }
1889
1890 /* Architecture-specific syscall changing routine. */
1891 void __change_syscall(struct __test_metadata *_metadata,
1892 pid_t tracee, long *syscall, long *ret)
1893 {
1894 ARCH_REGS orig, regs;
1895
1896 /* Do not get/set registers if we have nothing to do. */
1897 if (!syscall && !ret)
1898 return;
1899
1900 EXPECT_EQ(0, ARCH_GETREGS(regs)) {
1901 return;
1902 }
1903 orig = regs;
1904
1905 if (syscall)
1906 SYSCALL_NUM_SET(regs, *syscall);
1907
1908 if (ret)
1909 SYSCALL_RET_SET(regs, *ret);
1910
1911 /* Flush any register changes made. */
1912 if (memcmp(&orig, &regs, sizeof(orig)) != 0)
1913 EXPECT_EQ(0, ARCH_SETREGS(regs));
1914 }
1915
1916 /* Change only syscall number. */
1917 void change_syscall_nr(struct __test_metadata *_metadata,
1918 pid_t tracee, long syscall)
1919 {
1920 __change_syscall(_metadata, tracee, &syscall, NULL);
1921 }
1922
1923 /* Change syscall return value (and set syscall number to -1). */
1924 void change_syscall_ret(struct __test_metadata *_metadata,
1925 pid_t tracee, long ret)
1926 {
1927 long syscall = -1;
1928
1929 __change_syscall(_metadata, tracee, &syscall, &ret);
1930 }
1931
1932 void tracer_seccomp(struct __test_metadata *_metadata, pid_t tracee,
1933 int status, void *args)
1934 {
1935 int ret;
1936 unsigned long msg;
1937
1938 /* Make sure we got the right message. */
1939 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1940 EXPECT_EQ(0, ret);
1941
1942 /* Validate and take action on expected syscalls. */
1943 switch (msg) {
1944 case 0x1002:
1945 /* change getpid to getppid. */
1946 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee));
1947 change_syscall_nr(_metadata, tracee, __NR_getppid);
1948 break;
1949 case 0x1003:
1950 /* skip gettid with valid return code. */
1951 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee));
1952 change_syscall_ret(_metadata, tracee, 45000);
1953 break;
1954 case 0x1004:
1955 /* skip openat with error. */
1956 EXPECT_EQ(__NR_openat, get_syscall(_metadata, tracee));
1957 change_syscall_ret(_metadata, tracee, -ESRCH);
1958 break;
1959 case 0x1005:
1960 /* do nothing (allow getppid) */
1961 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee));
1962 break;
1963 default:
1964 EXPECT_EQ(0, msg) {
1965 TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg);
1966 kill(tracee, SIGKILL);
1967 }
1968 }
1969
1970 }
1971
1972 FIXTURE(TRACE_syscall) {
1973 struct sock_fprog prog;
1974 pid_t tracer, mytid, mypid, parent;
1975 long syscall_nr;
1976 };
1977
1978 void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee,
1979 int status, void *args)
1980 {
1981 int ret;
1982 unsigned long msg;
1983 static bool entry;
1984 long syscall_nr_val, syscall_ret_val;
1985 long *syscall_nr = NULL, *syscall_ret = NULL;
1986 FIXTURE_DATA(TRACE_syscall) *self = args;
1987
1988 /*
1989 * The traditional way to tell PTRACE_SYSCALL entry/exit
1990 * is by counting.
1991 */
1992 entry = !entry;
1993
1994 /* Make sure we got an appropriate message. */
1995 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1996 EXPECT_EQ(0, ret);
1997 EXPECT_EQ(entry ? PTRACE_EVENTMSG_SYSCALL_ENTRY
1998 : PTRACE_EVENTMSG_SYSCALL_EXIT, msg);
1999
2000 /*
2001 * Some architectures only support setting return values during
2002 * syscall exit under ptrace, and on exit the syscall number may
2003 * no longer be available. Therefore, save the initial sycall
2004 * number here, so it can be examined during both entry and exit
2005 * phases.
2006 */
2007 if (entry)
2008 self->syscall_nr = get_syscall(_metadata, tracee);
2009 else
2010 return;
2011
2012 syscall_nr = &syscall_nr_val;
2013 syscall_ret = &syscall_ret_val;
2014
2015 /* Now handle the actual rewriting cases. */
2016 switch (self->syscall_nr) {
2017 case __NR_getpid:
2018 syscall_nr_val = __NR_getppid;
2019 /* Never change syscall return for this case. */
2020 syscall_ret = NULL;
2021 break;
2022 case __NR_gettid:
2023 syscall_nr_val = -1;
2024 syscall_ret_val = 45000;
2025 break;
2026 case __NR_openat:
2027 syscall_nr_val = -1;
2028 syscall_ret_val = -ESRCH;
2029 break;
2030 default:
2031 /* Unhandled, do nothing. */
2032 return;
2033 }
2034
2035 __change_syscall(_metadata, tracee, syscall_nr, syscall_ret);
2036 }
2037
2038 FIXTURE_VARIANT(TRACE_syscall) {
2039 /*
2040 * All of the SECCOMP_RET_TRACE behaviors can be tested with either
2041 * SECCOMP_RET_TRACE+PTRACE_CONT or plain ptrace()+PTRACE_SYSCALL.
2042 * This indicates if we should use SECCOMP_RET_TRACE (false), or
2043 * ptrace (true).
2044 */
2045 bool use_ptrace;
2046 };
2047
2048 FIXTURE_VARIANT_ADD(TRACE_syscall, ptrace) {
2049 .use_ptrace = true,
2050 };
2051
2052 FIXTURE_VARIANT_ADD(TRACE_syscall, seccomp) {
2053 .use_ptrace = false,
2054 };
2055
2056 FIXTURE_SETUP(TRACE_syscall)
2057 {
2058 struct sock_filter filter[] = {
2059 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2060 offsetof(struct seccomp_data, nr)),
2061 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
2062 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002),
2063 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1),
2064 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003),
2065 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_openat, 0, 1),
2066 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004),
2067 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2068 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1005),
2069 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2070 };
2071 struct sock_fprog prog = {
2072 .len = (unsigned short)ARRAY_SIZE(filter),
2073 .filter = filter,
2074 };
2075 long ret;
2076
2077 /* Prepare some testable syscall results. */
2078 self->mytid = syscall(__NR_gettid);
2079 ASSERT_GT(self->mytid, 0);
2080 ASSERT_NE(self->mytid, 1) {
2081 TH_LOG("Running this test as init is not supported. :)");
2082 }
2083
2084 self->mypid = getpid();
2085 ASSERT_GT(self->mypid, 0);
2086 ASSERT_EQ(self->mytid, self->mypid);
2087
2088 self->parent = getppid();
2089 ASSERT_GT(self->parent, 0);
2090 ASSERT_NE(self->parent, self->mypid);
2091
2092 /* Launch tracer. */
2093 self->tracer = setup_trace_fixture(_metadata,
2094 variant->use_ptrace ? tracer_ptrace
2095 : tracer_seccomp,
2096 self, variant->use_ptrace);
2097
2098 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2099 ASSERT_EQ(0, ret);
2100
2101 if (variant->use_ptrace)
2102 return;
2103
2104 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2105 ASSERT_EQ(0, ret);
2106 }
2107
2108 FIXTURE_TEARDOWN(TRACE_syscall)
2109 {
2110 teardown_trace_fixture(_metadata, self->tracer);
2111 }
2112
2113 TEST(negative_ENOSYS)
2114 {
2115 /*
2116 * There should be no difference between an "internal" skip
2117 * and userspace asking for syscall "-1".
2118 */
2119 errno = 0;
2120 EXPECT_EQ(-1, syscall(-1));
2121 EXPECT_EQ(errno, ENOSYS);
2122 /* And no difference for "still not valid but not -1". */
2123 errno = 0;
2124 EXPECT_EQ(-1, syscall(-101));
2125 EXPECT_EQ(errno, ENOSYS);
2126 }
2127
2128 TEST_F(TRACE_syscall, negative_ENOSYS)
2129 {
2130 negative_ENOSYS(_metadata);
2131 }
2132
2133 TEST_F(TRACE_syscall, syscall_allowed)
2134 {
2135 /* getppid works as expected (no changes). */
2136 EXPECT_EQ(self->parent, syscall(__NR_getppid));
2137 EXPECT_NE(self->mypid, syscall(__NR_getppid));
2138 }
2139
2140 TEST_F(TRACE_syscall, syscall_redirected)
2141 {
2142 /* getpid has been redirected to getppid as expected. */
2143 EXPECT_EQ(self->parent, syscall(__NR_getpid));
2144 EXPECT_NE(self->mypid, syscall(__NR_getpid));
2145 }
2146
2147 TEST_F(TRACE_syscall, syscall_errno)
2148 {
2149 /* Tracer should skip the open syscall, resulting in ESRCH. */
2150 EXPECT_SYSCALL_RETURN(-ESRCH, syscall(__NR_openat));
2151 }
2152
2153 TEST_F(TRACE_syscall, syscall_faked)
2154 {
2155 /* Tracer skips the gettid syscall and store altered return value. */
2156 EXPECT_SYSCALL_RETURN(45000, syscall(__NR_gettid));
2157 }
2158
2159 TEST_F(TRACE_syscall, skip_after)
2160 {
2161 struct sock_filter filter[] = {
2162 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2163 offsetof(struct seccomp_data, nr)),
2164 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2165 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
2166 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2167 };
2168 struct sock_fprog prog = {
2169 .len = (unsigned short)ARRAY_SIZE(filter),
2170 .filter = filter,
2171 };
2172 long ret;
2173
2174 /* Install additional "errno on getppid" filter. */
2175 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2176 ASSERT_EQ(0, ret);
2177
2178 /* Tracer will redirect getpid to getppid, and we should see EPERM. */
2179 errno = 0;
2180 EXPECT_EQ(-1, syscall(__NR_getpid));
2181 EXPECT_EQ(EPERM, errno);
2182 }
2183
2184 TEST_F_SIGNAL(TRACE_syscall, kill_after, SIGSYS)
2185 {
2186 struct sock_filter filter[] = {
2187 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2188 offsetof(struct seccomp_data, nr)),
2189 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2190 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2191 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2192 };
2193 struct sock_fprog prog = {
2194 .len = (unsigned short)ARRAY_SIZE(filter),
2195 .filter = filter,
2196 };
2197 long ret;
2198
2199 /* Install additional "death on getppid" filter. */
2200 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2201 ASSERT_EQ(0, ret);
2202
2203 /* Tracer will redirect getpid to getppid, and we should die. */
2204 EXPECT_NE(self->mypid, syscall(__NR_getpid));
2205 }
2206
2207 TEST(seccomp_syscall)
2208 {
2209 struct sock_filter filter[] = {
2210 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2211 };
2212 struct sock_fprog prog = {
2213 .len = (unsigned short)ARRAY_SIZE(filter),
2214 .filter = filter,
2215 };
2216 long ret;
2217
2218 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2219 ASSERT_EQ(0, ret) {
2220 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2221 }
2222
2223 /* Reject insane operation. */
2224 ret = seccomp(-1, 0, &prog);
2225 ASSERT_NE(ENOSYS, errno) {
2226 TH_LOG("Kernel does not support seccomp syscall!");
2227 }
2228 EXPECT_EQ(EINVAL, errno) {
2229 TH_LOG("Did not reject crazy op value!");
2230 }
2231
2232 /* Reject strict with flags or pointer. */
2233 ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL);
2234 EXPECT_EQ(EINVAL, errno) {
2235 TH_LOG("Did not reject mode strict with flags!");
2236 }
2237 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog);
2238 EXPECT_EQ(EINVAL, errno) {
2239 TH_LOG("Did not reject mode strict with uargs!");
2240 }
2241
2242 /* Reject insane args for filter. */
2243 ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog);
2244 EXPECT_EQ(EINVAL, errno) {
2245 TH_LOG("Did not reject crazy filter flags!");
2246 }
2247 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL);
2248 EXPECT_EQ(EFAULT, errno) {
2249 TH_LOG("Did not reject NULL filter!");
2250 }
2251
2252 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2253 EXPECT_EQ(0, errno) {
2254 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s",
2255 strerror(errno));
2256 }
2257 }
2258
2259 TEST(seccomp_syscall_mode_lock)
2260 {
2261 struct sock_filter filter[] = {
2262 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2263 };
2264 struct sock_fprog prog = {
2265 .len = (unsigned short)ARRAY_SIZE(filter),
2266 .filter = filter,
2267 };
2268 long ret;
2269
2270 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2271 ASSERT_EQ(0, ret) {
2272 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2273 }
2274
2275 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2276 ASSERT_NE(ENOSYS, errno) {
2277 TH_LOG("Kernel does not support seccomp syscall!");
2278 }
2279 EXPECT_EQ(0, ret) {
2280 TH_LOG("Could not install filter!");
2281 }
2282
2283 /* Make sure neither entry point will switch to strict. */
2284 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0);
2285 EXPECT_EQ(EINVAL, errno) {
2286 TH_LOG("Switched to mode strict!");
2287 }
2288
2289 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL);
2290 EXPECT_EQ(EINVAL, errno) {
2291 TH_LOG("Switched to mode strict!");
2292 }
2293 }
2294
2295 /*
2296 * Test detection of known and unknown filter flags. Userspace needs to be able
2297 * to check if a filter flag is supported by the current kernel and a good way
2298 * of doing that is by attempting to enter filter mode, with the flag bit in
2299 * question set, and a NULL pointer for the _args_ parameter. EFAULT indicates
2300 * that the flag is valid and EINVAL indicates that the flag is invalid.
2301 */
2302 TEST(detect_seccomp_filter_flags)
2303 {
2304 unsigned int flags[] = { SECCOMP_FILTER_FLAG_TSYNC,
2305 SECCOMP_FILTER_FLAG_LOG,
2306 SECCOMP_FILTER_FLAG_SPEC_ALLOW,
2307 SECCOMP_FILTER_FLAG_NEW_LISTENER,
2308 SECCOMP_FILTER_FLAG_TSYNC_ESRCH };
2309 unsigned int exclusive[] = {
2310 SECCOMP_FILTER_FLAG_TSYNC,
2311 SECCOMP_FILTER_FLAG_NEW_LISTENER };
2312 unsigned int flag, all_flags, exclusive_mask;
2313 int i;
2314 long ret;
2315
2316 /* Test detection of individual known-good filter flags */
2317 for (i = 0, all_flags = 0; i < ARRAY_SIZE(flags); i++) {
2318 int bits = 0;
2319
2320 flag = flags[i];
2321 /* Make sure the flag is a single bit! */
2322 while (flag) {
2323 if (flag & 0x1)
2324 bits ++;
2325 flag >>= 1;
2326 }
2327 ASSERT_EQ(1, bits);
2328 flag = flags[i];
2329
2330 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2331 ASSERT_NE(ENOSYS, errno) {
2332 TH_LOG("Kernel does not support seccomp syscall!");
2333 }
2334 EXPECT_EQ(-1, ret);
2335 EXPECT_EQ(EFAULT, errno) {
2336 TH_LOG("Failed to detect that a known-good filter flag (0x%X) is supported!",
2337 flag);
2338 }
2339
2340 all_flags |= flag;
2341 }
2342
2343 /*
2344 * Test detection of all known-good filter flags combined. But
2345 * for the exclusive flags we need to mask them out and try them
2346 * individually for the "all flags" testing.
2347 */
2348 exclusive_mask = 0;
2349 for (i = 0; i < ARRAY_SIZE(exclusive); i++)
2350 exclusive_mask |= exclusive[i];
2351 for (i = 0; i < ARRAY_SIZE(exclusive); i++) {
2352 flag = all_flags & ~exclusive_mask;
2353 flag |= exclusive[i];
2354
2355 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2356 EXPECT_EQ(-1, ret);
2357 EXPECT_EQ(EFAULT, errno) {
2358 TH_LOG("Failed to detect that all known-good filter flags (0x%X) are supported!",
2359 flag);
2360 }
2361 }
2362
2363 /* Test detection of an unknown filter flags, without exclusives. */
2364 flag = -1;
2365 flag &= ~exclusive_mask;
2366 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2367 EXPECT_EQ(-1, ret);
2368 EXPECT_EQ(EINVAL, errno) {
2369 TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported!",
2370 flag);
2371 }
2372
2373 /*
2374 * Test detection of an unknown filter flag that may simply need to be
2375 * added to this test
2376 */
2377 flag = flags[ARRAY_SIZE(flags) - 1] << 1;
2378 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2379 EXPECT_EQ(-1, ret);
2380 EXPECT_EQ(EINVAL, errno) {
2381 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?",
2382 flag);
2383 }
2384 }
2385
2386 TEST(TSYNC_first)
2387 {
2388 struct sock_filter filter[] = {
2389 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2390 };
2391 struct sock_fprog prog = {
2392 .len = (unsigned short)ARRAY_SIZE(filter),
2393 .filter = filter,
2394 };
2395 long ret;
2396
2397 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2398 ASSERT_EQ(0, ret) {
2399 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2400 }
2401
2402 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2403 &prog);
2404 ASSERT_NE(ENOSYS, errno) {
2405 TH_LOG("Kernel does not support seccomp syscall!");
2406 }
2407 EXPECT_EQ(0, ret) {
2408 TH_LOG("Could not install initial filter with TSYNC!");
2409 }
2410 }
2411
2412 #define TSYNC_SIBLINGS 2
2413 struct tsync_sibling {
2414 pthread_t tid;
2415 pid_t system_tid;
2416 sem_t *started;
2417 pthread_cond_t *cond;
2418 pthread_mutex_t *mutex;
2419 int diverge;
2420 int num_waits;
2421 struct sock_fprog *prog;
2422 struct __test_metadata *metadata;
2423 };
2424
2425 /*
2426 * To avoid joining joined threads (which is not allowed by Bionic),
2427 * make sure we both successfully join and clear the tid to skip a
2428 * later join attempt during fixture teardown. Any remaining threads
2429 * will be directly killed during teardown.
2430 */
2431 #define PTHREAD_JOIN(tid, status) \
2432 do { \
2433 int _rc = pthread_join(tid, status); \
2434 if (_rc) { \
2435 TH_LOG("pthread_join of tid %u failed: %d\n", \
2436 (unsigned int)tid, _rc); \
2437 } else { \
2438 tid = 0; \
2439 } \
2440 } while (0)
2441
2442 FIXTURE(TSYNC) {
2443 struct sock_fprog root_prog, apply_prog;
2444 struct tsync_sibling sibling[TSYNC_SIBLINGS];
2445 sem_t started;
2446 pthread_cond_t cond;
2447 pthread_mutex_t mutex;
2448 int sibling_count;
2449 };
2450
2451 FIXTURE_SETUP(TSYNC)
2452 {
2453 struct sock_filter root_filter[] = {
2454 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2455 };
2456 struct sock_filter apply_filter[] = {
2457 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2458 offsetof(struct seccomp_data, nr)),
2459 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
2460 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2461 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2462 };
2463
2464 memset(&self->root_prog, 0, sizeof(self->root_prog));
2465 memset(&self->apply_prog, 0, sizeof(self->apply_prog));
2466 memset(&self->sibling, 0, sizeof(self->sibling));
2467 self->root_prog.filter = malloc(sizeof(root_filter));
2468 ASSERT_NE(NULL, self->root_prog.filter);
2469 memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter));
2470 self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter);
2471
2472 self->apply_prog.filter = malloc(sizeof(apply_filter));
2473 ASSERT_NE(NULL, self->apply_prog.filter);
2474 memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter));
2475 self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter);
2476
2477 self->sibling_count = 0;
2478 pthread_mutex_init(&self->mutex, NULL);
2479 pthread_cond_init(&self->cond, NULL);
2480 sem_init(&self->started, 0, 0);
2481 self->sibling[0].tid = 0;
2482 self->sibling[0].cond = &self->cond;
2483 self->sibling[0].started = &self->started;
2484 self->sibling[0].mutex = &self->mutex;
2485 self->sibling[0].diverge = 0;
2486 self->sibling[0].num_waits = 1;
2487 self->sibling[0].prog = &self->root_prog;
2488 self->sibling[0].metadata = _metadata;
2489 self->sibling[1].tid = 0;
2490 self->sibling[1].cond = &self->cond;
2491 self->sibling[1].started = &self->started;
2492 self->sibling[1].mutex = &self->mutex;
2493 self->sibling[1].diverge = 0;
2494 self->sibling[1].prog = &self->root_prog;
2495 self->sibling[1].num_waits = 1;
2496 self->sibling[1].metadata = _metadata;
2497 }
2498
2499 FIXTURE_TEARDOWN(TSYNC)
2500 {
2501 int sib = 0;
2502
2503 if (self->root_prog.filter)
2504 free(self->root_prog.filter);
2505 if (self->apply_prog.filter)
2506 free(self->apply_prog.filter);
2507
2508 for ( ; sib < self->sibling_count; ++sib) {
2509 struct tsync_sibling *s = &self->sibling[sib];
2510
2511 if (!s->tid)
2512 continue;
2513 /*
2514 * If a thread is still running, it may be stuck, so hit
2515 * it over the head really hard.
2516 */
2517 pthread_kill(s->tid, 9);
2518 }
2519 pthread_mutex_destroy(&self->mutex);
2520 pthread_cond_destroy(&self->cond);
2521 sem_destroy(&self->started);
2522 }
2523
2524 void *tsync_sibling(void *data)
2525 {
2526 long ret = 0;
2527 struct tsync_sibling *me = data;
2528
2529 me->system_tid = syscall(__NR_gettid);
2530
2531 pthread_mutex_lock(me->mutex);
2532 if (me->diverge) {
2533 /* Just re-apply the root prog to fork the tree */
2534 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
2535 me->prog, 0, 0);
2536 }
2537 sem_post(me->started);
2538 /* Return outside of started so parent notices failures. */
2539 if (ret) {
2540 pthread_mutex_unlock(me->mutex);
2541 return (void *)SIBLING_EXIT_FAILURE;
2542 }
2543 do {
2544 pthread_cond_wait(me->cond, me->mutex);
2545 me->num_waits = me->num_waits - 1;
2546 } while (me->num_waits);
2547 pthread_mutex_unlock(me->mutex);
2548
2549 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0);
2550 if (!ret)
2551 return (void *)SIBLING_EXIT_NEWPRIVS;
2552 read(0, NULL, 0);
2553 return (void *)SIBLING_EXIT_UNKILLED;
2554 }
2555
2556 void tsync_start_sibling(struct tsync_sibling *sibling)
2557 {
2558 pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling);
2559 }
2560
2561 TEST_F(TSYNC, siblings_fail_prctl)
2562 {
2563 long ret;
2564 void *status;
2565 struct sock_filter filter[] = {
2566 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2567 offsetof(struct seccomp_data, nr)),
2568 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
2569 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL),
2570 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2571 };
2572 struct sock_fprog prog = {
2573 .len = (unsigned short)ARRAY_SIZE(filter),
2574 .filter = filter,
2575 };
2576
2577 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2578 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2579 }
2580
2581 /* Check prctl failure detection by requesting sib 0 diverge. */
2582 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2583 ASSERT_NE(ENOSYS, errno) {
2584 TH_LOG("Kernel does not support seccomp syscall!");
2585 }
2586 ASSERT_EQ(0, ret) {
2587 TH_LOG("setting filter failed");
2588 }
2589
2590 self->sibling[0].diverge = 1;
2591 tsync_start_sibling(&self->sibling[0]);
2592 tsync_start_sibling(&self->sibling[1]);
2593
2594 while (self->sibling_count < TSYNC_SIBLINGS) {
2595 sem_wait(&self->started);
2596 self->sibling_count++;
2597 }
2598
2599 /* Signal the threads to clean up*/
2600 pthread_mutex_lock(&self->mutex);
2601 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2602 TH_LOG("cond broadcast non-zero");
2603 }
2604 pthread_mutex_unlock(&self->mutex);
2605
2606 /* Ensure diverging sibling failed to call prctl. */
2607 PTHREAD_JOIN(self->sibling[0].tid, &status);
2608 EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status);
2609 PTHREAD_JOIN(self->sibling[1].tid, &status);
2610 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2611 }
2612
2613 TEST_F(TSYNC, two_siblings_with_ancestor)
2614 {
2615 long ret;
2616 void *status;
2617
2618 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2619 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2620 }
2621
2622 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2623 ASSERT_NE(ENOSYS, errno) {
2624 TH_LOG("Kernel does not support seccomp syscall!");
2625 }
2626 ASSERT_EQ(0, ret) {
2627 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2628 }
2629 tsync_start_sibling(&self->sibling[0]);
2630 tsync_start_sibling(&self->sibling[1]);
2631
2632 while (self->sibling_count < TSYNC_SIBLINGS) {
2633 sem_wait(&self->started);
2634 self->sibling_count++;
2635 }
2636
2637 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2638 &self->apply_prog);
2639 ASSERT_EQ(0, ret) {
2640 TH_LOG("Could install filter on all threads!");
2641 }
2642 /* Tell the siblings to test the policy */
2643 pthread_mutex_lock(&self->mutex);
2644 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2645 TH_LOG("cond broadcast non-zero");
2646 }
2647 pthread_mutex_unlock(&self->mutex);
2648 /* Ensure they are both killed and don't exit cleanly. */
2649 PTHREAD_JOIN(self->sibling[0].tid, &status);
2650 EXPECT_EQ(0x0, (long)status);
2651 PTHREAD_JOIN(self->sibling[1].tid, &status);
2652 EXPECT_EQ(0x0, (long)status);
2653 }
2654
2655 TEST_F(TSYNC, two_sibling_want_nnp)
2656 {
2657 void *status;
2658
2659 /* start siblings before any prctl() operations */
2660 tsync_start_sibling(&self->sibling[0]);
2661 tsync_start_sibling(&self->sibling[1]);
2662 while (self->sibling_count < TSYNC_SIBLINGS) {
2663 sem_wait(&self->started);
2664 self->sibling_count++;
2665 }
2666
2667 /* Tell the siblings to test no policy */
2668 pthread_mutex_lock(&self->mutex);
2669 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2670 TH_LOG("cond broadcast non-zero");
2671 }
2672 pthread_mutex_unlock(&self->mutex);
2673
2674 /* Ensure they are both upset about lacking nnp. */
2675 PTHREAD_JOIN(self->sibling[0].tid, &status);
2676 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2677 PTHREAD_JOIN(self->sibling[1].tid, &status);
2678 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2679 }
2680
2681 TEST_F(TSYNC, two_siblings_with_no_filter)
2682 {
2683 long ret;
2684 void *status;
2685
2686 /* start siblings before any prctl() operations */
2687 tsync_start_sibling(&self->sibling[0]);
2688 tsync_start_sibling(&self->sibling[1]);
2689 while (self->sibling_count < TSYNC_SIBLINGS) {
2690 sem_wait(&self->started);
2691 self->sibling_count++;
2692 }
2693
2694 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2695 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2696 }
2697
2698 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2699 &self->apply_prog);
2700 ASSERT_NE(ENOSYS, errno) {
2701 TH_LOG("Kernel does not support seccomp syscall!");
2702 }
2703 ASSERT_EQ(0, ret) {
2704 TH_LOG("Could install filter on all threads!");
2705 }
2706
2707 /* Tell the siblings to test the policy */
2708 pthread_mutex_lock(&self->mutex);
2709 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2710 TH_LOG("cond broadcast non-zero");
2711 }
2712 pthread_mutex_unlock(&self->mutex);
2713
2714 /* Ensure they are both killed and don't exit cleanly. */
2715 PTHREAD_JOIN(self->sibling[0].tid, &status);
2716 EXPECT_EQ(0x0, (long)status);
2717 PTHREAD_JOIN(self->sibling[1].tid, &status);
2718 EXPECT_EQ(0x0, (long)status);
2719 }
2720
2721 TEST_F(TSYNC, two_siblings_with_one_divergence)
2722 {
2723 long ret;
2724 void *status;
2725
2726 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2727 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2728 }
2729
2730 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2731 ASSERT_NE(ENOSYS, errno) {
2732 TH_LOG("Kernel does not support seccomp syscall!");
2733 }
2734 ASSERT_EQ(0, ret) {
2735 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2736 }
2737 self->sibling[0].diverge = 1;
2738 tsync_start_sibling(&self->sibling[0]);
2739 tsync_start_sibling(&self->sibling[1]);
2740
2741 while (self->sibling_count < TSYNC_SIBLINGS) {
2742 sem_wait(&self->started);
2743 self->sibling_count++;
2744 }
2745
2746 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2747 &self->apply_prog);
2748 ASSERT_EQ(self->sibling[0].system_tid, ret) {
2749 TH_LOG("Did not fail on diverged sibling.");
2750 }
2751
2752 /* Wake the threads */
2753 pthread_mutex_lock(&self->mutex);
2754 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2755 TH_LOG("cond broadcast non-zero");
2756 }
2757 pthread_mutex_unlock(&self->mutex);
2758
2759 /* Ensure they are both unkilled. */
2760 PTHREAD_JOIN(self->sibling[0].tid, &status);
2761 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2762 PTHREAD_JOIN(self->sibling[1].tid, &status);
2763 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2764 }
2765
2766 TEST_F(TSYNC, two_siblings_with_one_divergence_no_tid_in_err)
2767 {
2768 long ret, flags;
2769 void *status;
2770
2771 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2772 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2773 }
2774
2775 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2776 ASSERT_NE(ENOSYS, errno) {
2777 TH_LOG("Kernel does not support seccomp syscall!");
2778 }
2779 ASSERT_EQ(0, ret) {
2780 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2781 }
2782 self->sibling[0].diverge = 1;
2783 tsync_start_sibling(&self->sibling[0]);
2784 tsync_start_sibling(&self->sibling[1]);
2785
2786 while (self->sibling_count < TSYNC_SIBLINGS) {
2787 sem_wait(&self->started);
2788 self->sibling_count++;
2789 }
2790
2791 flags = SECCOMP_FILTER_FLAG_TSYNC | \
2792 SECCOMP_FILTER_FLAG_TSYNC_ESRCH;
2793 ret = seccomp(SECCOMP_SET_MODE_FILTER, flags, &self->apply_prog);
2794 ASSERT_EQ(ESRCH, errno) {
2795 TH_LOG("Did not return ESRCH for diverged sibling.");
2796 }
2797 ASSERT_EQ(-1, ret) {
2798 TH_LOG("Did not fail on diverged sibling.");
2799 }
2800
2801 /* Wake the threads */
2802 pthread_mutex_lock(&self->mutex);
2803 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2804 TH_LOG("cond broadcast non-zero");
2805 }
2806 pthread_mutex_unlock(&self->mutex);
2807
2808 /* Ensure they are both unkilled. */
2809 PTHREAD_JOIN(self->sibling[0].tid, &status);
2810 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2811 PTHREAD_JOIN(self->sibling[1].tid, &status);
2812 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2813 }
2814
2815 TEST_F(TSYNC, two_siblings_not_under_filter)
2816 {
2817 long ret, sib;
2818 void *status;
2819 struct timespec delay = { .tv_nsec = 100000000 };
2820
2821 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2822 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2823 }
2824
2825 /*
2826 * Sibling 0 will have its own seccomp policy
2827 * and Sibling 1 will not be under seccomp at
2828 * all. Sibling 1 will enter seccomp and 0
2829 * will cause failure.
2830 */
2831 self->sibling[0].diverge = 1;
2832 tsync_start_sibling(&self->sibling[0]);
2833 tsync_start_sibling(&self->sibling[1]);
2834
2835 while (self->sibling_count < TSYNC_SIBLINGS) {
2836 sem_wait(&self->started);
2837 self->sibling_count++;
2838 }
2839
2840 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2841 ASSERT_NE(ENOSYS, errno) {
2842 TH_LOG("Kernel does not support seccomp syscall!");
2843 }
2844 ASSERT_EQ(0, ret) {
2845 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2846 }
2847
2848 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2849 &self->apply_prog);
2850 ASSERT_EQ(ret, self->sibling[0].system_tid) {
2851 TH_LOG("Did not fail on diverged sibling.");
2852 }
2853 sib = 1;
2854 if (ret == self->sibling[0].system_tid)
2855 sib = 0;
2856
2857 pthread_mutex_lock(&self->mutex);
2858
2859 /* Increment the other siblings num_waits so we can clean up
2860 * the one we just saw.
2861 */
2862 self->sibling[!sib].num_waits += 1;
2863
2864 /* Signal the thread to clean up*/
2865 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2866 TH_LOG("cond broadcast non-zero");
2867 }
2868 pthread_mutex_unlock(&self->mutex);
2869 PTHREAD_JOIN(self->sibling[sib].tid, &status);
2870 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2871 /* Poll for actual task death. pthread_join doesn't guarantee it. */
2872 while (!kill(self->sibling[sib].system_tid, 0))
2873 nanosleep(&delay, NULL);
2874 /* Switch to the remaining sibling */
2875 sib = !sib;
2876
2877 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2878 &self->apply_prog);
2879 ASSERT_EQ(0, ret) {
2880 TH_LOG("Expected the remaining sibling to sync");
2881 };
2882
2883 pthread_mutex_lock(&self->mutex);
2884
2885 /* If remaining sibling didn't have a chance to wake up during
2886 * the first broadcast, manually reduce the num_waits now.
2887 */
2888 if (self->sibling[sib].num_waits > 1)
2889 self->sibling[sib].num_waits = 1;
2890 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2891 TH_LOG("cond broadcast non-zero");
2892 }
2893 pthread_mutex_unlock(&self->mutex);
2894 PTHREAD_JOIN(self->sibling[sib].tid, &status);
2895 EXPECT_EQ(0, (long)status);
2896 /* Poll for actual task death. pthread_join doesn't guarantee it. */
2897 while (!kill(self->sibling[sib].system_tid, 0))
2898 nanosleep(&delay, NULL);
2899
2900 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2901 &self->apply_prog);
2902 ASSERT_EQ(0, ret); /* just us chickens */
2903 }
2904
2905 /* Make sure restarted syscalls are seen directly as "restart_syscall". */
2906 TEST(syscall_restart)
2907 {
2908 long ret;
2909 unsigned long msg;
2910 pid_t child_pid;
2911 int pipefd[2];
2912 int status;
2913 siginfo_t info = { };
2914 struct sock_filter filter[] = {
2915 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2916 offsetof(struct seccomp_data, nr)),
2917
2918 #ifdef __NR_sigreturn
2919 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 7, 0),
2920 #endif
2921 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 6, 0),
2922 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 5, 0),
2923 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 4, 0),
2924 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 5, 0),
2925 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_clock_nanosleep, 4, 0),
2926 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0),
2927
2928 /* Allow __NR_write for easy logging. */
2929 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1),
2930 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2931 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2932 /* The nanosleep jump target. */
2933 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100),
2934 /* The restart_syscall jump target. */
2935 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200),
2936 };
2937 struct sock_fprog prog = {
2938 .len = (unsigned short)ARRAY_SIZE(filter),
2939 .filter = filter,
2940 };
2941 #if defined(__arm__)
2942 struct utsname utsbuf;
2943 #endif
2944
2945 ASSERT_EQ(0, pipe(pipefd));
2946
2947 child_pid = fork();
2948 ASSERT_LE(0, child_pid);
2949 if (child_pid == 0) {
2950 /* Child uses EXPECT not ASSERT to deliver status correctly. */
2951 char buf = ' ';
2952 struct timespec timeout = { };
2953
2954 /* Attach parent as tracer and stop. */
2955 EXPECT_EQ(0, ptrace(PTRACE_TRACEME));
2956 EXPECT_EQ(0, raise(SIGSTOP));
2957
2958 EXPECT_EQ(0, close(pipefd[1]));
2959
2960 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2961 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2962 }
2963
2964 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2965 EXPECT_EQ(0, ret) {
2966 TH_LOG("Failed to install filter!");
2967 }
2968
2969 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2970 TH_LOG("Failed to read() sync from parent");
2971 }
2972 EXPECT_EQ('.', buf) {
2973 TH_LOG("Failed to get sync data from read()");
2974 }
2975
2976 /* Start nanosleep to be interrupted. */
2977 timeout.tv_sec = 1;
2978 errno = 0;
2979 EXPECT_EQ(0, nanosleep(&timeout, NULL)) {
2980 TH_LOG("Call to nanosleep() failed (errno %d)", errno);
2981 }
2982
2983 /* Read final sync from parent. */
2984 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2985 TH_LOG("Failed final read() from parent");
2986 }
2987 EXPECT_EQ('!', buf) {
2988 TH_LOG("Failed to get final data from read()");
2989 }
2990
2991 /* Directly report the status of our test harness results. */
2992 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS
2993 : EXIT_FAILURE);
2994 }
2995 EXPECT_EQ(0, close(pipefd[0]));
2996
2997 /* Attach to child, setup options, and release. */
2998 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2999 ASSERT_EQ(true, WIFSTOPPED(status));
3000 ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL,
3001 PTRACE_O_TRACESECCOMP));
3002 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3003 ASSERT_EQ(1, write(pipefd[1], ".", 1));
3004
3005 /* Wait for nanosleep() to start. */
3006 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3007 ASSERT_EQ(true, WIFSTOPPED(status));
3008 ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
3009 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
3010 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
3011 ASSERT_EQ(0x100, msg);
3012 ret = get_syscall(_metadata, child_pid);
3013 EXPECT_TRUE(ret == __NR_nanosleep || ret == __NR_clock_nanosleep);
3014
3015 /* Might as well check siginfo for sanity while we're here. */
3016 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
3017 ASSERT_EQ(SIGTRAP, info.si_signo);
3018 ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code);
3019 EXPECT_EQ(0, info.si_errno);
3020 EXPECT_EQ(getuid(), info.si_uid);
3021 /* Verify signal delivery came from child (seccomp-triggered). */
3022 EXPECT_EQ(child_pid, info.si_pid);
3023
3024 /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */
3025 ASSERT_EQ(0, kill(child_pid, SIGSTOP));
3026 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3027 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3028 ASSERT_EQ(true, WIFSTOPPED(status));
3029 ASSERT_EQ(SIGSTOP, WSTOPSIG(status));
3030 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
3031 /*
3032 * There is no siginfo on SIGSTOP any more, so we can't verify
3033 * signal delivery came from parent now (getpid() == info.si_pid).
3034 * https://lkml.kernel.org/r/CAGXu5jJaZAOzP1qFz66tYrtbuywqb+UN2SOA1VLHpCCOiYvYeg@mail.gmail.com
3035 * At least verify the SIGSTOP via PTRACE_GETSIGINFO.
3036 */
3037 EXPECT_EQ(SIGSTOP, info.si_signo);
3038
3039 /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */
3040 ASSERT_EQ(0, kill(child_pid, SIGCONT));
3041 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3042 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3043 ASSERT_EQ(true, WIFSTOPPED(status));
3044 ASSERT_EQ(SIGCONT, WSTOPSIG(status));
3045 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3046
3047 /* Wait for restart_syscall() to start. */
3048 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3049 ASSERT_EQ(true, WIFSTOPPED(status));
3050 ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
3051 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
3052 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
3053
3054 ASSERT_EQ(0x200, msg);
3055 ret = get_syscall(_metadata, child_pid);
3056 #if defined(__arm__)
3057 /*
3058 * FIXME:
3059 * - native ARM registers do NOT expose true syscall.
3060 * - compat ARM registers on ARM64 DO expose true syscall.
3061 */
3062 ASSERT_EQ(0, uname(&utsbuf));
3063 if (strncmp(utsbuf.machine, "arm", 3) == 0) {
3064 EXPECT_EQ(__NR_nanosleep, ret);
3065 } else
3066 #endif
3067 {
3068 EXPECT_EQ(__NR_restart_syscall, ret);
3069 }
3070
3071 /* Write again to end test. */
3072 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3073 ASSERT_EQ(1, write(pipefd[1], "!", 1));
3074 EXPECT_EQ(0, close(pipefd[1]));
3075
3076 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3077 if (WIFSIGNALED(status) || WEXITSTATUS(status))
3078 _metadata->passed = 0;
3079 }
3080
3081 TEST_SIGNAL(filter_flag_log, SIGSYS)
3082 {
3083 struct sock_filter allow_filter[] = {
3084 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3085 };
3086 struct sock_filter kill_filter[] = {
3087 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
3088 offsetof(struct seccomp_data, nr)),
3089 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
3090 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
3091 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3092 };
3093 struct sock_fprog allow_prog = {
3094 .len = (unsigned short)ARRAY_SIZE(allow_filter),
3095 .filter = allow_filter,
3096 };
3097 struct sock_fprog kill_prog = {
3098 .len = (unsigned short)ARRAY_SIZE(kill_filter),
3099 .filter = kill_filter,
3100 };
3101 long ret;
3102 pid_t parent = getppid();
3103
3104 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3105 ASSERT_EQ(0, ret);
3106
3107 /* Verify that the FILTER_FLAG_LOG flag isn't accepted in strict mode */
3108 ret = seccomp(SECCOMP_SET_MODE_STRICT, SECCOMP_FILTER_FLAG_LOG,
3109 &allow_prog);
3110 ASSERT_NE(ENOSYS, errno) {
3111 TH_LOG("Kernel does not support seccomp syscall!");
3112 }
3113 EXPECT_NE(0, ret) {
3114 TH_LOG("Kernel accepted FILTER_FLAG_LOG flag in strict mode!");
3115 }
3116 EXPECT_EQ(EINVAL, errno) {
3117 TH_LOG("Kernel returned unexpected errno for FILTER_FLAG_LOG flag in strict mode!");
3118 }
3119
3120 /* Verify that a simple, permissive filter can be added with no flags */
3121 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog);
3122 EXPECT_EQ(0, ret);
3123
3124 /* See if the same filter can be added with the FILTER_FLAG_LOG flag */
3125 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
3126 &allow_prog);
3127 ASSERT_NE(EINVAL, errno) {
3128 TH_LOG("Kernel does not support the FILTER_FLAG_LOG flag!");
3129 }
3130 EXPECT_EQ(0, ret);
3131
3132 /* Ensure that the kill filter works with the FILTER_FLAG_LOG flag */
3133 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
3134 &kill_prog);
3135 EXPECT_EQ(0, ret);
3136
3137 EXPECT_EQ(parent, syscall(__NR_getppid));
3138 /* getpid() should never return. */
3139 EXPECT_EQ(0, syscall(__NR_getpid));
3140 }
3141
3142 TEST(get_action_avail)
3143 {
3144 __u32 actions[] = { SECCOMP_RET_KILL_THREAD, SECCOMP_RET_TRAP,
3145 SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE,
3146 SECCOMP_RET_LOG, SECCOMP_RET_ALLOW };
3147 __u32 unknown_action = 0x10000000U;
3148 int i;
3149 long ret;
3150
3151 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[0]);
3152 ASSERT_NE(ENOSYS, errno) {
3153 TH_LOG("Kernel does not support seccomp syscall!");
3154 }
3155 ASSERT_NE(EINVAL, errno) {
3156 TH_LOG("Kernel does not support SECCOMP_GET_ACTION_AVAIL operation!");
3157 }
3158 EXPECT_EQ(ret, 0);
3159
3160 for (i = 0; i < ARRAY_SIZE(actions); i++) {
3161 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[i]);
3162 EXPECT_EQ(ret, 0) {
3163 TH_LOG("Expected action (0x%X) not available!",
3164 actions[i]);
3165 }
3166 }
3167
3168 /* Check that an unknown action is handled properly (EOPNOTSUPP) */
3169 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &unknown_action);
3170 EXPECT_EQ(ret, -1);
3171 EXPECT_EQ(errno, EOPNOTSUPP);
3172 }
3173
3174 TEST(get_metadata)
3175 {
3176 pid_t pid;
3177 int pipefd[2];
3178 char buf;
3179 struct seccomp_metadata md;
3180 long ret;
3181
3182 /* Only real root can get metadata. */
3183 if (geteuid()) {
3184 SKIP(return, "get_metadata requires real root");
3185 return;
3186 }
3187
3188 ASSERT_EQ(0, pipe(pipefd));
3189
3190 pid = fork();
3191 ASSERT_GE(pid, 0);
3192 if (pid == 0) {
3193 struct sock_filter filter[] = {
3194 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3195 };
3196 struct sock_fprog prog = {
3197 .len = (unsigned short)ARRAY_SIZE(filter),
3198 .filter = filter,
3199 };
3200
3201 /* one with log, one without */
3202 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER,
3203 SECCOMP_FILTER_FLAG_LOG, &prog));
3204 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog));
3205
3206 EXPECT_EQ(0, close(pipefd[0]));
3207 ASSERT_EQ(1, write(pipefd[1], "1", 1));
3208 ASSERT_EQ(0, close(pipefd[1]));
3209
3210 while (1)
3211 sleep(100);
3212 }
3213
3214 ASSERT_EQ(0, close(pipefd[1]));
3215 ASSERT_EQ(1, read(pipefd[0], &buf, 1));
3216
3217 ASSERT_EQ(0, ptrace(PTRACE_ATTACH, pid));
3218 ASSERT_EQ(pid, waitpid(pid, NULL, 0));
3219
3220 /* Past here must not use ASSERT or child process is never killed. */
3221
3222 md.filter_off = 0;
3223 errno = 0;
3224 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md);
3225 EXPECT_EQ(sizeof(md), ret) {
3226 if (errno == EINVAL)
3227 SKIP(goto skip, "Kernel does not support PTRACE_SECCOMP_GET_METADATA (missing CONFIG_CHECKPOINT_RESTORE?)");
3228 }
3229
3230 EXPECT_EQ(md.flags, SECCOMP_FILTER_FLAG_LOG);
3231 EXPECT_EQ(md.filter_off, 0);
3232
3233 md.filter_off = 1;
3234 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md);
3235 EXPECT_EQ(sizeof(md), ret);
3236 EXPECT_EQ(md.flags, 0);
3237 EXPECT_EQ(md.filter_off, 1);
3238
3239 skip:
3240 ASSERT_EQ(0, kill(pid, SIGKILL));
3241 }
3242
3243 static int user_notif_syscall(int nr, unsigned int flags)
3244 {
3245 struct sock_filter filter[] = {
3246 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
3247 offsetof(struct seccomp_data, nr)),
3248 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, nr, 0, 1),
3249 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_USER_NOTIF),
3250 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3251 };
3252
3253 struct sock_fprog prog = {
3254 .len = (unsigned short)ARRAY_SIZE(filter),
3255 .filter = filter,
3256 };
3257
3258 return seccomp(SECCOMP_SET_MODE_FILTER, flags, &prog);
3259 }
3260
3261 #define USER_NOTIF_MAGIC INT_MAX
3262 TEST(user_notification_basic)
3263 {
3264 pid_t pid;
3265 long ret;
3266 int status, listener;
3267 struct seccomp_notif req = {};
3268 struct seccomp_notif_resp resp = {};
3269 struct pollfd pollfd;
3270
3271 struct sock_filter filter[] = {
3272 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3273 };
3274 struct sock_fprog prog = {
3275 .len = (unsigned short)ARRAY_SIZE(filter),
3276 .filter = filter,
3277 };
3278
3279 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3280 ASSERT_EQ(0, ret) {
3281 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3282 }
3283
3284 pid = fork();
3285 ASSERT_GE(pid, 0);
3286
3287 /* Check that we get -ENOSYS with no listener attached */
3288 if (pid == 0) {
3289 if (user_notif_syscall(__NR_getppid, 0) < 0)
3290 exit(1);
3291 ret = syscall(__NR_getppid);
3292 exit(ret >= 0 || errno != ENOSYS);
3293 }
3294
3295 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3296 EXPECT_EQ(true, WIFEXITED(status));
3297 EXPECT_EQ(0, WEXITSTATUS(status));
3298
3299 /* Add some no-op filters for grins. */
3300 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3301 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3302 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3303 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3304
3305 /* Check that the basic notification machinery works */
3306 listener = user_notif_syscall(__NR_getppid,
3307 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3308 ASSERT_GE(listener, 0);
3309
3310 /* Installing a second listener in the chain should EBUSY */
3311 EXPECT_EQ(user_notif_syscall(__NR_getppid,
3312 SECCOMP_FILTER_FLAG_NEW_LISTENER),
3313 -1);
3314 EXPECT_EQ(errno, EBUSY);
3315
3316 pid = fork();
3317 ASSERT_GE(pid, 0);
3318
3319 if (pid == 0) {
3320 ret = syscall(__NR_getppid);
3321 exit(ret != USER_NOTIF_MAGIC);
3322 }
3323
3324 pollfd.fd = listener;
3325 pollfd.events = POLLIN | POLLOUT;
3326
3327 EXPECT_GT(poll(&pollfd, 1, -1), 0);
3328 EXPECT_EQ(pollfd.revents, POLLIN);
3329
3330 /* Test that we can't pass garbage to the kernel. */
3331 memset(&req, 0, sizeof(req));
3332 req.pid = -1;
3333 errno = 0;
3334 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req);
3335 EXPECT_EQ(-1, ret);
3336 EXPECT_EQ(EINVAL, errno);
3337
3338 if (ret) {
3339 req.pid = 0;
3340 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3341 }
3342
3343 pollfd.fd = listener;
3344 pollfd.events = POLLIN | POLLOUT;
3345
3346 EXPECT_GT(poll(&pollfd, 1, -1), 0);
3347 EXPECT_EQ(pollfd.revents, POLLOUT);
3348
3349 EXPECT_EQ(req.data.nr, __NR_getppid);
3350
3351 resp.id = req.id;
3352 resp.error = 0;
3353 resp.val = USER_NOTIF_MAGIC;
3354
3355 /* check that we make sure flags == 0 */
3356 resp.flags = 1;
3357 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3358 EXPECT_EQ(errno, EINVAL);
3359
3360 resp.flags = 0;
3361 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3362
3363 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3364 EXPECT_EQ(true, WIFEXITED(status));
3365 EXPECT_EQ(0, WEXITSTATUS(status));
3366 }
3367
3368 TEST(user_notification_with_tsync)
3369 {
3370 int ret;
3371 unsigned int flags;
3372
3373 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3374 ASSERT_EQ(0, ret) {
3375 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3376 }
3377
3378 /* these were exclusive */
3379 flags = SECCOMP_FILTER_FLAG_NEW_LISTENER |
3380 SECCOMP_FILTER_FLAG_TSYNC;
3381 ASSERT_EQ(-1, user_notif_syscall(__NR_getppid, flags));
3382 ASSERT_EQ(EINVAL, errno);
3383
3384 /* but now they're not */
3385 flags |= SECCOMP_FILTER_FLAG_TSYNC_ESRCH;
3386 ret = user_notif_syscall(__NR_getppid, flags);
3387 close(ret);
3388 ASSERT_LE(0, ret);
3389 }
3390
3391 TEST(user_notification_kill_in_middle)
3392 {
3393 pid_t pid;
3394 long ret;
3395 int listener;
3396 struct seccomp_notif req = {};
3397 struct seccomp_notif_resp resp = {};
3398
3399 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3400 ASSERT_EQ(0, ret) {
3401 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3402 }
3403
3404 listener = user_notif_syscall(__NR_getppid,
3405 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3406 ASSERT_GE(listener, 0);
3407
3408 /*
3409 * Check that nothing bad happens when we kill the task in the middle
3410 * of a syscall.
3411 */
3412 pid = fork();
3413 ASSERT_GE(pid, 0);
3414
3415 if (pid == 0) {
3416 ret = syscall(__NR_getppid);
3417 exit(ret != USER_NOTIF_MAGIC);
3418 }
3419
3420 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3421 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), 0);
3422
3423 EXPECT_EQ(kill(pid, SIGKILL), 0);
3424 EXPECT_EQ(waitpid(pid, NULL, 0), pid);
3425
3426 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), -1);
3427
3428 resp.id = req.id;
3429 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp);
3430 EXPECT_EQ(ret, -1);
3431 EXPECT_EQ(errno, ENOENT);
3432 }
3433
3434 static int handled = -1;
3435
3436 static void signal_handler(int signal)
3437 {
3438 if (write(handled, "c", 1) != 1)
3439 perror("write from signal");
3440 }
3441
3442 TEST(user_notification_signal)
3443 {
3444 pid_t pid;
3445 long ret;
3446 int status, listener, sk_pair[2];
3447 struct seccomp_notif req = {};
3448 struct seccomp_notif_resp resp = {};
3449 char c;
3450
3451 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3452 ASSERT_EQ(0, ret) {
3453 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3454 }
3455
3456 ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0);
3457
3458 listener = user_notif_syscall(__NR_gettid,
3459 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3460 ASSERT_GE(listener, 0);
3461
3462 pid = fork();
3463 ASSERT_GE(pid, 0);
3464
3465 if (pid == 0) {
3466 close(sk_pair[0]);
3467 handled = sk_pair[1];
3468 if (signal(SIGUSR1, signal_handler) == SIG_ERR) {
3469 perror("signal");
3470 exit(1);
3471 }
3472 /*
3473 * ERESTARTSYS behavior is a bit hard to test, because we need
3474 * to rely on a signal that has not yet been handled. Let's at
3475 * least check that the error code gets propagated through, and
3476 * hope that it doesn't break when there is actually a signal :)
3477 */
3478 ret = syscall(__NR_gettid);
3479 exit(!(ret == -1 && errno == 512));
3480 }
3481
3482 close(sk_pair[1]);
3483
3484 memset(&req, 0, sizeof(req));
3485 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3486
3487 EXPECT_EQ(kill(pid, SIGUSR1), 0);
3488
3489 /*
3490 * Make sure the signal really is delivered, which means we're not
3491 * stuck in the user notification code any more and the notification
3492 * should be dead.
3493 */
3494 EXPECT_EQ(read(sk_pair[0], &c, 1), 1);
3495
3496 resp.id = req.id;
3497 resp.error = -EPERM;
3498 resp.val = 0;
3499
3500 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3501 EXPECT_EQ(errno, ENOENT);
3502
3503 memset(&req, 0, sizeof(req));
3504 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3505
3506 resp.id = req.id;
3507 resp.error = -512; /* -ERESTARTSYS */
3508 resp.val = 0;
3509
3510 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3511
3512 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3513 EXPECT_EQ(true, WIFEXITED(status));
3514 EXPECT_EQ(0, WEXITSTATUS(status));
3515 }
3516
3517 TEST(user_notification_closed_listener)
3518 {
3519 pid_t pid;
3520 long ret;
3521 int status, listener;
3522
3523 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3524 ASSERT_EQ(0, ret) {
3525 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3526 }
3527
3528 listener = user_notif_syscall(__NR_getppid,
3529 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3530 ASSERT_GE(listener, 0);
3531
3532 /*
3533 * Check that we get an ENOSYS when the listener is closed.
3534 */
3535 pid = fork();
3536 ASSERT_GE(pid, 0);
3537 if (pid == 0) {
3538 close(listener);
3539 ret = syscall(__NR_getppid);
3540 exit(ret != -1 && errno != ENOSYS);
3541 }
3542
3543 close(listener);
3544
3545 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3546 EXPECT_EQ(true, WIFEXITED(status));
3547 EXPECT_EQ(0, WEXITSTATUS(status));
3548 }
3549
3550 /*
3551 * Check that a pid in a child namespace still shows up as valid in ours.
3552 */
3553 TEST(user_notification_child_pid_ns)
3554 {
3555 pid_t pid;
3556 int status, listener;
3557 struct seccomp_notif req = {};
3558 struct seccomp_notif_resp resp = {};
3559
3560 ASSERT_EQ(unshare(CLONE_NEWUSER | CLONE_NEWPID), 0) {
3561 if (errno == EINVAL)
3562 SKIP(return, "kernel missing CLONE_NEWUSER support");
3563 };
3564
3565 listener = user_notif_syscall(__NR_getppid,
3566 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3567 ASSERT_GE(listener, 0);
3568
3569 pid = fork();
3570 ASSERT_GE(pid, 0);
3571
3572 if (pid == 0)
3573 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3574
3575 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3576 EXPECT_EQ(req.pid, pid);
3577
3578 resp.id = req.id;
3579 resp.error = 0;
3580 resp.val = USER_NOTIF_MAGIC;
3581
3582 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3583
3584 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3585 EXPECT_EQ(true, WIFEXITED(status));
3586 EXPECT_EQ(0, WEXITSTATUS(status));
3587 close(listener);
3588 }
3589
3590 /*
3591 * Check that a pid in a sibling (i.e. unrelated) namespace shows up as 0, i.e.
3592 * invalid.
3593 */
3594 TEST(user_notification_sibling_pid_ns)
3595 {
3596 pid_t pid, pid2;
3597 int status, listener;
3598 struct seccomp_notif req = {};
3599 struct seccomp_notif_resp resp = {};
3600
3601 ASSERT_EQ(prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0), 0) {
3602 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3603 }
3604
3605 listener = user_notif_syscall(__NR_getppid,
3606 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3607 ASSERT_GE(listener, 0);
3608
3609 pid = fork();
3610 ASSERT_GE(pid, 0);
3611
3612 if (pid == 0) {
3613 ASSERT_EQ(unshare(CLONE_NEWPID), 0);
3614
3615 pid2 = fork();
3616 ASSERT_GE(pid2, 0);
3617
3618 if (pid2 == 0)
3619 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3620
3621 EXPECT_EQ(waitpid(pid2, &status, 0), pid2);
3622 EXPECT_EQ(true, WIFEXITED(status));
3623 EXPECT_EQ(0, WEXITSTATUS(status));
3624 exit(WEXITSTATUS(status));
3625 }
3626
3627 /* Create the sibling ns, and sibling in it. */
3628 ASSERT_EQ(unshare(CLONE_NEWPID), 0) {
3629 if (errno == EPERM)
3630 SKIP(return, "CLONE_NEWPID requires CAP_SYS_ADMIN");
3631 }
3632 ASSERT_EQ(errno, 0);
3633
3634 pid2 = fork();
3635 ASSERT_GE(pid2, 0);
3636
3637 if (pid2 == 0) {
3638 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3639 /*
3640 * The pid should be 0, i.e. the task is in some namespace that
3641 * we can't "see".
3642 */
3643 EXPECT_EQ(req.pid, 0);
3644
3645 resp.id = req.id;
3646 resp.error = 0;
3647 resp.val = USER_NOTIF_MAGIC;
3648
3649 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3650 exit(0);
3651 }
3652
3653 close(listener);
3654
3655 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3656 EXPECT_EQ(true, WIFEXITED(status));
3657 EXPECT_EQ(0, WEXITSTATUS(status));
3658
3659 EXPECT_EQ(waitpid(pid2, &status, 0), pid2);
3660 EXPECT_EQ(true, WIFEXITED(status));
3661 EXPECT_EQ(0, WEXITSTATUS(status));
3662 }
3663
3664 TEST(user_notification_fault_recv)
3665 {
3666 pid_t pid;
3667 int status, listener;
3668 struct seccomp_notif req = {};
3669 struct seccomp_notif_resp resp = {};
3670
3671 ASSERT_EQ(unshare(CLONE_NEWUSER), 0);
3672
3673 listener = user_notif_syscall(__NR_getppid,
3674 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3675 ASSERT_GE(listener, 0);
3676
3677 pid = fork();
3678 ASSERT_GE(pid, 0);
3679
3680 if (pid == 0)
3681 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3682
3683 /* Do a bad recv() */
3684 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, NULL), -1);
3685 EXPECT_EQ(errno, EFAULT);
3686
3687 /* We should still be able to receive this notification, though. */
3688 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3689 EXPECT_EQ(req.pid, pid);
3690
3691 resp.id = req.id;
3692 resp.error = 0;
3693 resp.val = USER_NOTIF_MAGIC;
3694
3695 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3696
3697 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3698 EXPECT_EQ(true, WIFEXITED(status));
3699 EXPECT_EQ(0, WEXITSTATUS(status));
3700 }
3701
3702 TEST(seccomp_get_notif_sizes)
3703 {
3704 struct seccomp_notif_sizes sizes;
3705
3706 ASSERT_EQ(seccomp(SECCOMP_GET_NOTIF_SIZES, 0, &sizes), 0);
3707 EXPECT_EQ(sizes.seccomp_notif, sizeof(struct seccomp_notif));
3708 EXPECT_EQ(sizes.seccomp_notif_resp, sizeof(struct seccomp_notif_resp));
3709 }
3710
3711 TEST(user_notification_continue)
3712 {
3713 pid_t pid;
3714 long ret;
3715 int status, listener;
3716 struct seccomp_notif req = {};
3717 struct seccomp_notif_resp resp = {};
3718 struct pollfd pollfd;
3719
3720 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3721 ASSERT_EQ(0, ret) {
3722 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3723 }
3724
3725 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER);
3726 ASSERT_GE(listener, 0);
3727
3728 pid = fork();
3729 ASSERT_GE(pid, 0);
3730
3731 if (pid == 0) {
3732 int dup_fd, pipe_fds[2];
3733 pid_t self;
3734
3735 ASSERT_GE(pipe(pipe_fds), 0);
3736
3737 dup_fd = dup(pipe_fds[0]);
3738 ASSERT_GE(dup_fd, 0);
3739 EXPECT_NE(pipe_fds[0], dup_fd);
3740
3741 self = getpid();
3742 ASSERT_EQ(filecmp(self, self, pipe_fds[0], dup_fd), 0);
3743 exit(0);
3744 }
3745
3746 pollfd.fd = listener;
3747 pollfd.events = POLLIN | POLLOUT;
3748
3749 EXPECT_GT(poll(&pollfd, 1, -1), 0);
3750 EXPECT_EQ(pollfd.revents, POLLIN);
3751
3752 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3753
3754 pollfd.fd = listener;
3755 pollfd.events = POLLIN | POLLOUT;
3756
3757 EXPECT_GT(poll(&pollfd, 1, -1), 0);
3758 EXPECT_EQ(pollfd.revents, POLLOUT);
3759
3760 EXPECT_EQ(req.data.nr, __NR_dup);
3761
3762 resp.id = req.id;
3763 resp.flags = SECCOMP_USER_NOTIF_FLAG_CONTINUE;
3764
3765 /*
3766 * Verify that setting SECCOMP_USER_NOTIF_FLAG_CONTINUE enforces other
3767 * args be set to 0.
3768 */
3769 resp.error = 0;
3770 resp.val = USER_NOTIF_MAGIC;
3771 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3772 EXPECT_EQ(errno, EINVAL);
3773
3774 resp.error = USER_NOTIF_MAGIC;
3775 resp.val = 0;
3776 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3777 EXPECT_EQ(errno, EINVAL);
3778
3779 resp.error = 0;
3780 resp.val = 0;
3781 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0) {
3782 if (errno == EINVAL)
3783 SKIP(goto skip, "Kernel does not support SECCOMP_USER_NOTIF_FLAG_CONTINUE");
3784 }
3785
3786 skip:
3787 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3788 EXPECT_EQ(true, WIFEXITED(status));
3789 EXPECT_EQ(0, WEXITSTATUS(status)) {
3790 if (WEXITSTATUS(status) == 2) {
3791 SKIP(return, "Kernel does not support kcmp() syscall");
3792 return;
3793 }
3794 }
3795 }
3796
3797 TEST(user_notification_filter_empty)
3798 {
3799 pid_t pid;
3800 long ret;
3801 int status;
3802 struct pollfd pollfd;
3803 struct clone_args args = {
3804 .flags = CLONE_FILES,
3805 .exit_signal = SIGCHLD,
3806 };
3807
3808 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3809 ASSERT_EQ(0, ret) {
3810 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3811 }
3812
3813 pid = sys_clone3(&args, sizeof(args));
3814 ASSERT_GE(pid, 0);
3815
3816 if (pid == 0) {
3817 int listener;
3818
3819 listener = user_notif_syscall(__NR_mknodat, SECCOMP_FILTER_FLAG_NEW_LISTENER);
3820 if (listener < 0)
3821 _exit(EXIT_FAILURE);
3822
3823 if (dup2(listener, 200) != 200)
3824 _exit(EXIT_FAILURE);
3825
3826 close(listener);
3827
3828 _exit(EXIT_SUCCESS);
3829 }
3830
3831 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3832 EXPECT_EQ(true, WIFEXITED(status));
3833 EXPECT_EQ(0, WEXITSTATUS(status));
3834
3835 /*
3836 * The seccomp filter has become unused so we should be notified once
3837 * the kernel gets around to cleaning up task struct.
3838 */
3839 pollfd.fd = 200;
3840 pollfd.events = POLLHUP;
3841
3842 EXPECT_GT(poll(&pollfd, 1, 2000), 0);
3843 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0);
3844 }
3845
3846 static void *do_thread(void *data)
3847 {
3848 return NULL;
3849 }
3850
3851 TEST(user_notification_filter_empty_threaded)
3852 {
3853 pid_t pid;
3854 long ret;
3855 int status;
3856 struct pollfd pollfd;
3857 struct clone_args args = {
3858 .flags = CLONE_FILES,
3859 .exit_signal = SIGCHLD,
3860 };
3861
3862 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3863 ASSERT_EQ(0, ret) {
3864 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3865 }
3866
3867 pid = sys_clone3(&args, sizeof(args));
3868 ASSERT_GE(pid, 0);
3869
3870 if (pid == 0) {
3871 pid_t pid1, pid2;
3872 int listener, status;
3873 pthread_t thread;
3874
3875 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER);
3876 if (listener < 0)
3877 _exit(EXIT_FAILURE);
3878
3879 if (dup2(listener, 200) != 200)
3880 _exit(EXIT_FAILURE);
3881
3882 close(listener);
3883
3884 pid1 = fork();
3885 if (pid1 < 0)
3886 _exit(EXIT_FAILURE);
3887
3888 if (pid1 == 0)
3889 _exit(EXIT_SUCCESS);
3890
3891 pid2 = fork();
3892 if (pid2 < 0)
3893 _exit(EXIT_FAILURE);
3894
3895 if (pid2 == 0)
3896 _exit(EXIT_SUCCESS);
3897
3898 if (pthread_create(&thread, NULL, do_thread, NULL) ||
3899 pthread_join(thread, NULL))
3900 _exit(EXIT_FAILURE);
3901
3902 if (pthread_create(&thread, NULL, do_thread, NULL) ||
3903 pthread_join(thread, NULL))
3904 _exit(EXIT_FAILURE);
3905
3906 if (waitpid(pid1, &status, 0) != pid1 || !WIFEXITED(status) ||
3907 WEXITSTATUS(status))
3908 _exit(EXIT_FAILURE);
3909
3910 if (waitpid(pid2, &status, 0) != pid2 || !WIFEXITED(status) ||
3911 WEXITSTATUS(status))
3912 _exit(EXIT_FAILURE);
3913
3914 exit(EXIT_SUCCESS);
3915 }
3916
3917 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3918 EXPECT_EQ(true, WIFEXITED(status));
3919 EXPECT_EQ(0, WEXITSTATUS(status));
3920
3921 /*
3922 * The seccomp filter has become unused so we should be notified once
3923 * the kernel gets around to cleaning up task struct.
3924 */
3925 pollfd.fd = 200;
3926 pollfd.events = POLLHUP;
3927
3928 EXPECT_GT(poll(&pollfd, 1, 2000), 0);
3929 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0);
3930 }
3931
3932 TEST(user_notification_addfd)
3933 {
3934 pid_t pid;
3935 long ret;
3936 int status, listener, memfd, fd;
3937 struct seccomp_notif_addfd addfd = {};
3938 struct seccomp_notif_addfd_small small = {};
3939 struct seccomp_notif_addfd_big big = {};
3940 struct seccomp_notif req = {};
3941 struct seccomp_notif_resp resp = {};
3942 /* 100 ms */
3943 struct timespec delay = { .tv_nsec = 100000000 };
3944
3945 memfd = memfd_create("test", 0);
3946 ASSERT_GE(memfd, 0);
3947
3948 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3949 ASSERT_EQ(0, ret) {
3950 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3951 }
3952
3953 /* Check that the basic notification machinery works */
3954 listener = user_notif_syscall(__NR_getppid,
3955 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3956 ASSERT_GE(listener, 0);
3957
3958 pid = fork();
3959 ASSERT_GE(pid, 0);
3960
3961 if (pid == 0) {
3962 if (syscall(__NR_getppid) != USER_NOTIF_MAGIC)
3963 exit(1);
3964 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3965 }
3966
3967 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3968
3969 addfd.srcfd = memfd;
3970 addfd.newfd = 0;
3971 addfd.id = req.id;
3972 addfd.flags = 0x0;
3973
3974 /* Verify bad newfd_flags cannot be set */
3975 addfd.newfd_flags = ~O_CLOEXEC;
3976 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
3977 EXPECT_EQ(errno, EINVAL);
3978 addfd.newfd_flags = O_CLOEXEC;
3979
3980 /* Verify bad flags cannot be set */
3981 addfd.flags = 0xff;
3982 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
3983 EXPECT_EQ(errno, EINVAL);
3984 addfd.flags = 0;
3985
3986 /* Verify that remote_fd cannot be set without setting flags */
3987 addfd.newfd = 1;
3988 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
3989 EXPECT_EQ(errno, EINVAL);
3990 addfd.newfd = 0;
3991
3992 /* Verify small size cannot be set */
3993 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_SMALL, &small), -1);
3994 EXPECT_EQ(errno, EINVAL);
3995
3996 /* Verify we can't send bits filled in unknown buffer area */
3997 memset(&big, 0xAA, sizeof(big));
3998 big.addfd = addfd;
3999 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big), -1);
4000 EXPECT_EQ(errno, E2BIG);
4001
4002
4003 /* Verify we can set an arbitrary remote fd */
4004 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd);
4005 /*
4006 * The child has fds 0(stdin), 1(stdout), 2(stderr), 3(memfd),
4007 * 4(listener), so the newly allocated fd should be 5.
4008 */
4009 EXPECT_EQ(fd, 5);
4010 EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0);
4011
4012 /* Verify we can set an arbitrary remote fd with large size */
4013 memset(&big, 0x0, sizeof(big));
4014 big.addfd = addfd;
4015 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big);
4016 EXPECT_EQ(fd, 6);
4017
4018 /* Verify we can set a specific remote fd */
4019 addfd.newfd = 42;
4020 addfd.flags = SECCOMP_ADDFD_FLAG_SETFD;
4021 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd);
4022 EXPECT_EQ(fd, 42);
4023 EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0);
4024
4025 /* Resume syscall */
4026 resp.id = req.id;
4027 resp.error = 0;
4028 resp.val = USER_NOTIF_MAGIC;
4029 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4030
4031 /*
4032 * This sets the ID of the ADD FD to the last request plus 1. The
4033 * notification ID increments 1 per notification.
4034 */
4035 addfd.id = req.id + 1;
4036
4037 /* This spins until the underlying notification is generated */
4038 while (ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd) != -1 &&
4039 errno != -EINPROGRESS)
4040 nanosleep(&delay, NULL);
4041
4042 memset(&req, 0, sizeof(req));
4043 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4044 ASSERT_EQ(addfd.id, req.id);
4045
4046 resp.id = req.id;
4047 resp.error = 0;
4048 resp.val = USER_NOTIF_MAGIC;
4049 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4050
4051 /* Wait for child to finish. */
4052 EXPECT_EQ(waitpid(pid, &status, 0), pid);
4053 EXPECT_EQ(true, WIFEXITED(status));
4054 EXPECT_EQ(0, WEXITSTATUS(status));
4055
4056 close(memfd);
4057 }
4058
4059 TEST(user_notification_addfd_rlimit)
4060 {
4061 pid_t pid;
4062 long ret;
4063 int status, listener, memfd;
4064 struct seccomp_notif_addfd addfd = {};
4065 struct seccomp_notif req = {};
4066 struct seccomp_notif_resp resp = {};
4067 const struct rlimit lim = {
4068 .rlim_cur = 0,
4069 .rlim_max = 0,
4070 };
4071
4072 memfd = memfd_create("test", 0);
4073 ASSERT_GE(memfd, 0);
4074
4075 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4076 ASSERT_EQ(0, ret) {
4077 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4078 }
4079
4080 /* Check that the basic notification machinery works */
4081 listener = user_notif_syscall(__NR_getppid,
4082 SECCOMP_FILTER_FLAG_NEW_LISTENER);
4083 ASSERT_GE(listener, 0);
4084
4085 pid = fork();
4086 ASSERT_GE(pid, 0);
4087
4088 if (pid == 0)
4089 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
4090
4091
4092 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4093
4094 ASSERT_EQ(prlimit(pid, RLIMIT_NOFILE, &lim, NULL), 0);
4095
4096 addfd.srcfd = memfd;
4097 addfd.newfd_flags = O_CLOEXEC;
4098 addfd.newfd = 0;
4099 addfd.id = req.id;
4100 addfd.flags = 0;
4101
4102 /* Should probably spot check /proc/sys/fs/file-nr */
4103 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4104 EXPECT_EQ(errno, EMFILE);
4105
4106 addfd.newfd = 100;
4107 addfd.flags = SECCOMP_ADDFD_FLAG_SETFD;
4108 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4109 EXPECT_EQ(errno, EBADF);
4110
4111 resp.id = req.id;
4112 resp.error = 0;
4113 resp.val = USER_NOTIF_MAGIC;
4114
4115 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4116
4117 /* Wait for child to finish. */
4118 EXPECT_EQ(waitpid(pid, &status, 0), pid);
4119 EXPECT_EQ(true, WIFEXITED(status));
4120 EXPECT_EQ(0, WEXITSTATUS(status));
4121
4122 close(memfd);
4123 }
4124
4125 /*
4126 * TODO:
4127 * - expand NNP testing
4128 * - better arch-specific TRACE and TRAP handlers.
4129 * - endianness checking when appropriate
4130 * - 64-bit arg prodding
4131 * - arch value testing (x86 modes especially)
4132 * - verify that FILTER_FLAG_LOG filters generate log messages
4133 * - verify that RET_LOG generates log messages
4134 */
4135
4136 TEST_HARNESS_MAIN
Ubuntu Jammy Linux kernel mirror