]> 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: Provide generic syscall setting macro
[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 # define SYSCALL_RET(_regs) (_regs).ARM_r0
1712 #elif defined(__aarch64__)
1713 # define ARCH_REGS struct user_pt_regs
1714 # define SYSCALL_NUM(_regs) (_regs).regs[8]
1715 # define SYSCALL_RET(_regs) (_regs).regs[0]
1716 #elif defined(__riscv) && __riscv_xlen == 64
1717 # define ARCH_REGS struct user_regs_struct
1718 # define SYSCALL_NUM(_regs) (_regs).a7
1719 # define SYSCALL_RET(_regs) (_regs).a0
1720 #elif defined(__csky__)
1721 # define ARCH_REGS struct pt_regs
1722 # if defined(__CSKYABIV2__)
1723 # define SYSCALL_NUM(_regs) (_regs).regs[3]
1724 # else
1725 # define SYSCALL_NUM(_regs) (_regs).regs[9]
1726 # endif
1727 # define SYSCALL_RET(_regs) (_regs).a0
1728 #elif defined(__hppa__)
1729 # define ARCH_REGS struct user_regs_struct
1730 # define SYSCALL_NUM(_regs) (_regs).gr[20]
1731 # define SYSCALL_RET(_regs) (_regs).gr[28]
1732 #elif defined(__powerpc__)
1733 # define ARCH_REGS struct pt_regs
1734 # define SYSCALL_NUM(_regs) (_regs).gpr[0]
1735 # define SYSCALL_RET(_regs) (_regs).gpr[3]
1736 #elif defined(__s390__)
1737 # define ARCH_REGS s390_regs
1738 # define SYSCALL_NUM(_regs) (_regs).gprs[2]
1739 # define SYSCALL_RET(_regs) (_regs).gprs[2]
1740 # define SYSCALL_NUM_RET_SHARE_REG
1741 #elif defined(__mips__)
1742 # define ARCH_REGS struct pt_regs
1743 # define SYSCALL_NUM(_regs) (_regs).regs[2]
1744 # define SYSCALL_SYSCALL_NUM regs[4]
1745 # define SYSCALL_RET(_regs) (_regs).regs[2]
1746 # define SYSCALL_NUM_RET_SHARE_REG
1747 #elif defined(__xtensa__)
1748 # define ARCH_REGS struct user_pt_regs
1749 # define SYSCALL_NUM(_regs) (_regs).syscall
1750 /*
1751 * On xtensa syscall return value is in the register
1752 * a2 of the current window which is not fixed.
1753 */
1754 #define SYSCALL_RET(_regs) (_regs).a[(_regs).windowbase * 4 + 2]
1755 #elif defined(__sh__)
1756 # define ARCH_REGS struct pt_regs
1757 # define SYSCALL_NUM(_regs) (_regs).gpr[3]
1758 # define SYSCALL_RET(_regs) (_regs).gpr[0]
1759 #else
1760 # error "Do not know how to find your architecture's registers and syscalls"
1761 #endif
1762
1763 /*
1764 * Most architectures can change the syscall by just updating the
1765 * associated register. This is the default if not defined above.
1766 */
1767 #ifndef SYSCALL_NUM_SET
1768 # define SYSCALL_NUM_SET(_regs, _nr) \
1769 do { \
1770 SYSCALL_NUM(_regs) = (_nr); \
1771 } while (0)
1772 #endif
1773
1774 /* When the syscall return can't be changed, stub out the tests for it. */
1775 #ifdef SYSCALL_NUM_RET_SHARE_REG
1776 # define EXPECT_SYSCALL_RETURN(val, action) EXPECT_EQ(-1, action)
1777 #else
1778 # define EXPECT_SYSCALL_RETURN(val, action) \
1779 do { \
1780 errno = 0; \
1781 if (val < 0) { \
1782 EXPECT_EQ(-1, action); \
1783 EXPECT_EQ(-(val), errno); \
1784 } else { \
1785 EXPECT_EQ(val, action); \
1786 } \
1787 } while (0)
1788 #endif
1789
1790 /* Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for
1791 * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux).
1792 */
1793 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__)
1794 #define HAVE_GETREGS
1795 #endif
1796
1797 /* Architecture-specific syscall fetching routine. */
1798 int get_syscall(struct __test_metadata *_metadata, pid_t tracee)
1799 {
1800 ARCH_REGS regs;
1801 #ifdef HAVE_GETREGS
1802 EXPECT_EQ(0, ptrace(PTRACE_GETREGS, tracee, 0, &regs)) {
1803 TH_LOG("PTRACE_GETREGS failed");
1804 return -1;
1805 }
1806 #else
1807 struct iovec iov;
1808
1809 iov.iov_base = &regs;
1810 iov.iov_len = sizeof(regs);
1811 EXPECT_EQ(0, ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov)) {
1812 TH_LOG("PTRACE_GETREGSET failed");
1813 return -1;
1814 }
1815 #endif
1816
1817 #if defined(__mips__)
1818 if (SYSCALL_NUM(regs) == __NR_O32_Linux)
1819 return regs.SYSCALL_SYSCALL_NUM;
1820 #endif
1821 return SYSCALL_NUM(regs);
1822 }
1823
1824 /* Architecture-specific syscall changing routine. */
1825 void change_syscall(struct __test_metadata *_metadata,
1826 pid_t tracee, int syscall, int result)
1827 {
1828 int ret;
1829 ARCH_REGS regs;
1830 #ifdef HAVE_GETREGS
1831 ret = ptrace(PTRACE_GETREGS, tracee, 0, &regs);
1832 #else
1833 struct iovec iov;
1834 iov.iov_base = &regs;
1835 iov.iov_len = sizeof(regs);
1836 ret = ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov);
1837 #endif
1838 EXPECT_EQ(0, ret) {}
1839
1840 #if defined(__x86_64__) || defined(__i386__) || defined(__powerpc__) || \
1841 defined(__s390__) || defined(__hppa__) || defined(__riscv) || \
1842 defined(__xtensa__) || defined(__csky__) || defined(__sh__)
1843 {
1844 SYSCALL_NUM_SET(regs, syscall);
1845 }
1846 #elif defined(__mips__)
1847 {
1848 if (SYSCALL_NUM(regs) == __NR_O32_Linux)
1849 regs.SYSCALL_SYSCALL_NUM = syscall;
1850 else
1851 SYSCALL_NUM_SET(regs, syscall);
1852 }
1853
1854 #elif defined(__arm__)
1855 # ifndef PTRACE_SET_SYSCALL
1856 # define PTRACE_SET_SYSCALL 23
1857 # endif
1858 {
1859 ret = ptrace(PTRACE_SET_SYSCALL, tracee, NULL, syscall);
1860 EXPECT_EQ(0, ret);
1861 }
1862
1863 #elif defined(__aarch64__)
1864 # ifndef NT_ARM_SYSTEM_CALL
1865 # define NT_ARM_SYSTEM_CALL 0x404
1866 # endif
1867 {
1868 iov.iov_base = &syscall;
1869 iov.iov_len = sizeof(syscall);
1870 ret = ptrace(PTRACE_SETREGSET, tracee, NT_ARM_SYSTEM_CALL,
1871 &iov);
1872 EXPECT_EQ(0, ret);
1873 }
1874
1875 #else
1876 ASSERT_EQ(1, 0) {
1877 TH_LOG("How is the syscall changed on this architecture?");
1878 }
1879 #endif
1880
1881 /* If syscall is skipped, change return value. */
1882 if (syscall == -1)
1883 #ifdef SYSCALL_NUM_RET_SHARE_REG
1884 TH_LOG("Can't modify syscall return on this architecture");
1885 #else
1886 SYSCALL_RET(regs) = result;
1887 #endif
1888
1889 #ifdef HAVE_GETREGS
1890 ret = ptrace(PTRACE_SETREGS, tracee, 0, &regs);
1891 #else
1892 iov.iov_base = &regs;
1893 iov.iov_len = sizeof(regs);
1894 ret = ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &iov);
1895 #endif
1896 EXPECT_EQ(0, ret);
1897 }
1898
1899 void tracer_seccomp(struct __test_metadata *_metadata, pid_t tracee,
1900 int status, void *args)
1901 {
1902 int ret;
1903 unsigned long msg;
1904
1905 /* Make sure we got the right message. */
1906 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1907 EXPECT_EQ(0, ret);
1908
1909 /* Validate and take action on expected syscalls. */
1910 switch (msg) {
1911 case 0x1002:
1912 /* change getpid to getppid. */
1913 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee));
1914 change_syscall(_metadata, tracee, __NR_getppid, 0);
1915 break;
1916 case 0x1003:
1917 /* skip gettid with valid return code. */
1918 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee));
1919 change_syscall(_metadata, tracee, -1, 45000);
1920 break;
1921 case 0x1004:
1922 /* skip openat with error. */
1923 EXPECT_EQ(__NR_openat, get_syscall(_metadata, tracee));
1924 change_syscall(_metadata, tracee, -1, -ESRCH);
1925 break;
1926 case 0x1005:
1927 /* do nothing (allow getppid) */
1928 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee));
1929 break;
1930 default:
1931 EXPECT_EQ(0, msg) {
1932 TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg);
1933 kill(tracee, SIGKILL);
1934 }
1935 }
1936
1937 }
1938
1939 void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee,
1940 int status, void *args)
1941 {
1942 int ret, nr;
1943 unsigned long msg;
1944 static bool entry;
1945
1946 /*
1947 * The traditional way to tell PTRACE_SYSCALL entry/exit
1948 * is by counting.
1949 */
1950 entry = !entry;
1951
1952 /* Make sure we got an appropriate message. */
1953 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1954 EXPECT_EQ(0, ret);
1955 EXPECT_EQ(entry ? PTRACE_EVENTMSG_SYSCALL_ENTRY
1956 : PTRACE_EVENTMSG_SYSCALL_EXIT, msg);
1957
1958 if (!entry)
1959 return;
1960
1961 nr = get_syscall(_metadata, tracee);
1962
1963 if (nr == __NR_getpid)
1964 change_syscall(_metadata, tracee, __NR_getppid, 0);
1965 if (nr == __NR_gettid)
1966 change_syscall(_metadata, tracee, -1, 45000);
1967 if (nr == __NR_openat)
1968 change_syscall(_metadata, tracee, -1, -ESRCH);
1969 }
1970
1971 FIXTURE(TRACE_syscall) {
1972 struct sock_fprog prog;
1973 pid_t tracer, mytid, mypid, parent;
1974 };
1975
1976 FIXTURE_VARIANT(TRACE_syscall) {
1977 /*
1978 * All of the SECCOMP_RET_TRACE behaviors can be tested with either
1979 * SECCOMP_RET_TRACE+PTRACE_CONT or plain ptrace()+PTRACE_SYSCALL.
1980 * This indicates if we should use SECCOMP_RET_TRACE (false), or
1981 * ptrace (true).
1982 */
1983 bool use_ptrace;
1984 };
1985
1986 FIXTURE_VARIANT_ADD(TRACE_syscall, ptrace) {
1987 .use_ptrace = true,
1988 };
1989
1990 FIXTURE_VARIANT_ADD(TRACE_syscall, seccomp) {
1991 .use_ptrace = false,
1992 };
1993
1994 FIXTURE_SETUP(TRACE_syscall)
1995 {
1996 struct sock_filter filter[] = {
1997 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1998 offsetof(struct seccomp_data, nr)),
1999 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
2000 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002),
2001 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1),
2002 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003),
2003 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_openat, 0, 1),
2004 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004),
2005 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2006 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1005),
2007 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2008 };
2009 struct sock_fprog prog = {
2010 .len = (unsigned short)ARRAY_SIZE(filter),
2011 .filter = filter,
2012 };
2013 long ret;
2014
2015 /* Prepare some testable syscall results. */
2016 self->mytid = syscall(__NR_gettid);
2017 ASSERT_GT(self->mytid, 0);
2018 ASSERT_NE(self->mytid, 1) {
2019 TH_LOG("Running this test as init is not supported. :)");
2020 }
2021
2022 self->mypid = getpid();
2023 ASSERT_GT(self->mypid, 0);
2024 ASSERT_EQ(self->mytid, self->mypid);
2025
2026 self->parent = getppid();
2027 ASSERT_GT(self->parent, 0);
2028 ASSERT_NE(self->parent, self->mypid);
2029
2030 /* Launch tracer. */
2031 self->tracer = setup_trace_fixture(_metadata,
2032 variant->use_ptrace ? tracer_ptrace
2033 : tracer_seccomp,
2034 NULL, variant->use_ptrace);
2035
2036 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2037 ASSERT_EQ(0, ret);
2038
2039 if (variant->use_ptrace)
2040 return;
2041
2042 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2043 ASSERT_EQ(0, ret);
2044 }
2045
2046 FIXTURE_TEARDOWN(TRACE_syscall)
2047 {
2048 teardown_trace_fixture(_metadata, self->tracer);
2049 }
2050
2051 TEST(negative_ENOSYS)
2052 {
2053 /*
2054 * There should be no difference between an "internal" skip
2055 * and userspace asking for syscall "-1".
2056 */
2057 errno = 0;
2058 EXPECT_EQ(-1, syscall(-1));
2059 EXPECT_EQ(errno, ENOSYS);
2060 /* And no difference for "still not valid but not -1". */
2061 errno = 0;
2062 EXPECT_EQ(-1, syscall(-101));
2063 EXPECT_EQ(errno, ENOSYS);
2064 }
2065
2066 TEST_F(TRACE_syscall, negative_ENOSYS)
2067 {
2068 negative_ENOSYS(_metadata);
2069 }
2070
2071 TEST_F(TRACE_syscall, syscall_allowed)
2072 {
2073 /* getppid works as expected (no changes). */
2074 EXPECT_EQ(self->parent, syscall(__NR_getppid));
2075 EXPECT_NE(self->mypid, syscall(__NR_getppid));
2076 }
2077
2078 TEST_F(TRACE_syscall, syscall_redirected)
2079 {
2080 /* getpid has been redirected to getppid as expected. */
2081 EXPECT_EQ(self->parent, syscall(__NR_getpid));
2082 EXPECT_NE(self->mypid, syscall(__NR_getpid));
2083 }
2084
2085 TEST_F(TRACE_syscall, syscall_errno)
2086 {
2087 /* Tracer should skip the open syscall, resulting in ESRCH. */
2088 EXPECT_SYSCALL_RETURN(-ESRCH, syscall(__NR_openat));
2089 }
2090
2091 TEST_F(TRACE_syscall, syscall_faked)
2092 {
2093 /* Tracer skips the gettid syscall and store altered return value. */
2094 EXPECT_SYSCALL_RETURN(45000, syscall(__NR_gettid));
2095 }
2096
2097 TEST_F(TRACE_syscall, skip_after)
2098 {
2099 struct sock_filter filter[] = {
2100 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2101 offsetof(struct seccomp_data, nr)),
2102 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2103 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
2104 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2105 };
2106 struct sock_fprog prog = {
2107 .len = (unsigned short)ARRAY_SIZE(filter),
2108 .filter = filter,
2109 };
2110 long ret;
2111
2112 /* Install additional "errno on getppid" filter. */
2113 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2114 ASSERT_EQ(0, ret);
2115
2116 /* Tracer will redirect getpid to getppid, and we should see EPERM. */
2117 errno = 0;
2118 EXPECT_EQ(-1, syscall(__NR_getpid));
2119 EXPECT_EQ(EPERM, errno);
2120 }
2121
2122 TEST_F_SIGNAL(TRACE_syscall, kill_after, SIGSYS)
2123 {
2124 struct sock_filter filter[] = {
2125 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2126 offsetof(struct seccomp_data, nr)),
2127 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
2128 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2129 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2130 };
2131 struct sock_fprog prog = {
2132 .len = (unsigned short)ARRAY_SIZE(filter),
2133 .filter = filter,
2134 };
2135 long ret;
2136
2137 /* Install additional "death on getppid" filter. */
2138 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2139 ASSERT_EQ(0, ret);
2140
2141 /* Tracer will redirect getpid to getppid, and we should die. */
2142 EXPECT_NE(self->mypid, syscall(__NR_getpid));
2143 }
2144
2145 TEST(seccomp_syscall)
2146 {
2147 struct sock_filter filter[] = {
2148 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2149 };
2150 struct sock_fprog prog = {
2151 .len = (unsigned short)ARRAY_SIZE(filter),
2152 .filter = filter,
2153 };
2154 long ret;
2155
2156 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2157 ASSERT_EQ(0, ret) {
2158 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2159 }
2160
2161 /* Reject insane operation. */
2162 ret = seccomp(-1, 0, &prog);
2163 ASSERT_NE(ENOSYS, errno) {
2164 TH_LOG("Kernel does not support seccomp syscall!");
2165 }
2166 EXPECT_EQ(EINVAL, errno) {
2167 TH_LOG("Did not reject crazy op value!");
2168 }
2169
2170 /* Reject strict with flags or pointer. */
2171 ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL);
2172 EXPECT_EQ(EINVAL, errno) {
2173 TH_LOG("Did not reject mode strict with flags!");
2174 }
2175 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog);
2176 EXPECT_EQ(EINVAL, errno) {
2177 TH_LOG("Did not reject mode strict with uargs!");
2178 }
2179
2180 /* Reject insane args for filter. */
2181 ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog);
2182 EXPECT_EQ(EINVAL, errno) {
2183 TH_LOG("Did not reject crazy filter flags!");
2184 }
2185 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL);
2186 EXPECT_EQ(EFAULT, errno) {
2187 TH_LOG("Did not reject NULL filter!");
2188 }
2189
2190 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2191 EXPECT_EQ(0, errno) {
2192 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s",
2193 strerror(errno));
2194 }
2195 }
2196
2197 TEST(seccomp_syscall_mode_lock)
2198 {
2199 struct sock_filter filter[] = {
2200 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2201 };
2202 struct sock_fprog prog = {
2203 .len = (unsigned short)ARRAY_SIZE(filter),
2204 .filter = filter,
2205 };
2206 long ret;
2207
2208 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2209 ASSERT_EQ(0, ret) {
2210 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2211 }
2212
2213 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2214 ASSERT_NE(ENOSYS, errno) {
2215 TH_LOG("Kernel does not support seccomp syscall!");
2216 }
2217 EXPECT_EQ(0, ret) {
2218 TH_LOG("Could not install filter!");
2219 }
2220
2221 /* Make sure neither entry point will switch to strict. */
2222 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0);
2223 EXPECT_EQ(EINVAL, errno) {
2224 TH_LOG("Switched to mode strict!");
2225 }
2226
2227 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL);
2228 EXPECT_EQ(EINVAL, errno) {
2229 TH_LOG("Switched to mode strict!");
2230 }
2231 }
2232
2233 /*
2234 * Test detection of known and unknown filter flags. Userspace needs to be able
2235 * to check if a filter flag is supported by the current kernel and a good way
2236 * of doing that is by attempting to enter filter mode, with the flag bit in
2237 * question set, and a NULL pointer for the _args_ parameter. EFAULT indicates
2238 * that the flag is valid and EINVAL indicates that the flag is invalid.
2239 */
2240 TEST(detect_seccomp_filter_flags)
2241 {
2242 unsigned int flags[] = { SECCOMP_FILTER_FLAG_TSYNC,
2243 SECCOMP_FILTER_FLAG_LOG,
2244 SECCOMP_FILTER_FLAG_SPEC_ALLOW,
2245 SECCOMP_FILTER_FLAG_NEW_LISTENER,
2246 SECCOMP_FILTER_FLAG_TSYNC_ESRCH };
2247 unsigned int exclusive[] = {
2248 SECCOMP_FILTER_FLAG_TSYNC,
2249 SECCOMP_FILTER_FLAG_NEW_LISTENER };
2250 unsigned int flag, all_flags, exclusive_mask;
2251 int i;
2252 long ret;
2253
2254 /* Test detection of individual known-good filter flags */
2255 for (i = 0, all_flags = 0; i < ARRAY_SIZE(flags); i++) {
2256 int bits = 0;
2257
2258 flag = flags[i];
2259 /* Make sure the flag is a single bit! */
2260 while (flag) {
2261 if (flag & 0x1)
2262 bits ++;
2263 flag >>= 1;
2264 }
2265 ASSERT_EQ(1, bits);
2266 flag = flags[i];
2267
2268 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2269 ASSERT_NE(ENOSYS, errno) {
2270 TH_LOG("Kernel does not support seccomp syscall!");
2271 }
2272 EXPECT_EQ(-1, ret);
2273 EXPECT_EQ(EFAULT, errno) {
2274 TH_LOG("Failed to detect that a known-good filter flag (0x%X) is supported!",
2275 flag);
2276 }
2277
2278 all_flags |= flag;
2279 }
2280
2281 /*
2282 * Test detection of all known-good filter flags combined. But
2283 * for the exclusive flags we need to mask them out and try them
2284 * individually for the "all flags" testing.
2285 */
2286 exclusive_mask = 0;
2287 for (i = 0; i < ARRAY_SIZE(exclusive); i++)
2288 exclusive_mask |= exclusive[i];
2289 for (i = 0; i < ARRAY_SIZE(exclusive); i++) {
2290 flag = all_flags & ~exclusive_mask;
2291 flag |= exclusive[i];
2292
2293 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2294 EXPECT_EQ(-1, ret);
2295 EXPECT_EQ(EFAULT, errno) {
2296 TH_LOG("Failed to detect that all known-good filter flags (0x%X) are supported!",
2297 flag);
2298 }
2299 }
2300
2301 /* Test detection of an unknown filter flags, without exclusives. */
2302 flag = -1;
2303 flag &= ~exclusive_mask;
2304 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2305 EXPECT_EQ(-1, ret);
2306 EXPECT_EQ(EINVAL, errno) {
2307 TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported!",
2308 flag);
2309 }
2310
2311 /*
2312 * Test detection of an unknown filter flag that may simply need to be
2313 * added to this test
2314 */
2315 flag = flags[ARRAY_SIZE(flags) - 1] << 1;
2316 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2317 EXPECT_EQ(-1, ret);
2318 EXPECT_EQ(EINVAL, errno) {
2319 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?",
2320 flag);
2321 }
2322 }
2323
2324 TEST(TSYNC_first)
2325 {
2326 struct sock_filter filter[] = {
2327 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2328 };
2329 struct sock_fprog prog = {
2330 .len = (unsigned short)ARRAY_SIZE(filter),
2331 .filter = filter,
2332 };
2333 long ret;
2334
2335 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2336 ASSERT_EQ(0, ret) {
2337 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2338 }
2339
2340 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2341 &prog);
2342 ASSERT_NE(ENOSYS, errno) {
2343 TH_LOG("Kernel does not support seccomp syscall!");
2344 }
2345 EXPECT_EQ(0, ret) {
2346 TH_LOG("Could not install initial filter with TSYNC!");
2347 }
2348 }
2349
2350 #define TSYNC_SIBLINGS 2
2351 struct tsync_sibling {
2352 pthread_t tid;
2353 pid_t system_tid;
2354 sem_t *started;
2355 pthread_cond_t *cond;
2356 pthread_mutex_t *mutex;
2357 int diverge;
2358 int num_waits;
2359 struct sock_fprog *prog;
2360 struct __test_metadata *metadata;
2361 };
2362
2363 /*
2364 * To avoid joining joined threads (which is not allowed by Bionic),
2365 * make sure we both successfully join and clear the tid to skip a
2366 * later join attempt during fixture teardown. Any remaining threads
2367 * will be directly killed during teardown.
2368 */
2369 #define PTHREAD_JOIN(tid, status) \
2370 do { \
2371 int _rc = pthread_join(tid, status); \
2372 if (_rc) { \
2373 TH_LOG("pthread_join of tid %u failed: %d\n", \
2374 (unsigned int)tid, _rc); \
2375 } else { \
2376 tid = 0; \
2377 } \
2378 } while (0)
2379
2380 FIXTURE(TSYNC) {
2381 struct sock_fprog root_prog, apply_prog;
2382 struct tsync_sibling sibling[TSYNC_SIBLINGS];
2383 sem_t started;
2384 pthread_cond_t cond;
2385 pthread_mutex_t mutex;
2386 int sibling_count;
2387 };
2388
2389 FIXTURE_SETUP(TSYNC)
2390 {
2391 struct sock_filter root_filter[] = {
2392 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2393 };
2394 struct sock_filter apply_filter[] = {
2395 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2396 offsetof(struct seccomp_data, nr)),
2397 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
2398 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2399 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2400 };
2401
2402 memset(&self->root_prog, 0, sizeof(self->root_prog));
2403 memset(&self->apply_prog, 0, sizeof(self->apply_prog));
2404 memset(&self->sibling, 0, sizeof(self->sibling));
2405 self->root_prog.filter = malloc(sizeof(root_filter));
2406 ASSERT_NE(NULL, self->root_prog.filter);
2407 memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter));
2408 self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter);
2409
2410 self->apply_prog.filter = malloc(sizeof(apply_filter));
2411 ASSERT_NE(NULL, self->apply_prog.filter);
2412 memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter));
2413 self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter);
2414
2415 self->sibling_count = 0;
2416 pthread_mutex_init(&self->mutex, NULL);
2417 pthread_cond_init(&self->cond, NULL);
2418 sem_init(&self->started, 0, 0);
2419 self->sibling[0].tid = 0;
2420 self->sibling[0].cond = &self->cond;
2421 self->sibling[0].started = &self->started;
2422 self->sibling[0].mutex = &self->mutex;
2423 self->sibling[0].diverge = 0;
2424 self->sibling[0].num_waits = 1;
2425 self->sibling[0].prog = &self->root_prog;
2426 self->sibling[0].metadata = _metadata;
2427 self->sibling[1].tid = 0;
2428 self->sibling[1].cond = &self->cond;
2429 self->sibling[1].started = &self->started;
2430 self->sibling[1].mutex = &self->mutex;
2431 self->sibling[1].diverge = 0;
2432 self->sibling[1].prog = &self->root_prog;
2433 self->sibling[1].num_waits = 1;
2434 self->sibling[1].metadata = _metadata;
2435 }
2436
2437 FIXTURE_TEARDOWN(TSYNC)
2438 {
2439 int sib = 0;
2440
2441 if (self->root_prog.filter)
2442 free(self->root_prog.filter);
2443 if (self->apply_prog.filter)
2444 free(self->apply_prog.filter);
2445
2446 for ( ; sib < self->sibling_count; ++sib) {
2447 struct tsync_sibling *s = &self->sibling[sib];
2448
2449 if (!s->tid)
2450 continue;
2451 /*
2452 * If a thread is still running, it may be stuck, so hit
2453 * it over the head really hard.
2454 */
2455 pthread_kill(s->tid, 9);
2456 }
2457 pthread_mutex_destroy(&self->mutex);
2458 pthread_cond_destroy(&self->cond);
2459 sem_destroy(&self->started);
2460 }
2461
2462 void *tsync_sibling(void *data)
2463 {
2464 long ret = 0;
2465 struct tsync_sibling *me = data;
2466
2467 me->system_tid = syscall(__NR_gettid);
2468
2469 pthread_mutex_lock(me->mutex);
2470 if (me->diverge) {
2471 /* Just re-apply the root prog to fork the tree */
2472 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
2473 me->prog, 0, 0);
2474 }
2475 sem_post(me->started);
2476 /* Return outside of started so parent notices failures. */
2477 if (ret) {
2478 pthread_mutex_unlock(me->mutex);
2479 return (void *)SIBLING_EXIT_FAILURE;
2480 }
2481 do {
2482 pthread_cond_wait(me->cond, me->mutex);
2483 me->num_waits = me->num_waits - 1;
2484 } while (me->num_waits);
2485 pthread_mutex_unlock(me->mutex);
2486
2487 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0);
2488 if (!ret)
2489 return (void *)SIBLING_EXIT_NEWPRIVS;
2490 read(0, NULL, 0);
2491 return (void *)SIBLING_EXIT_UNKILLED;
2492 }
2493
2494 void tsync_start_sibling(struct tsync_sibling *sibling)
2495 {
2496 pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling);
2497 }
2498
2499 TEST_F(TSYNC, siblings_fail_prctl)
2500 {
2501 long ret;
2502 void *status;
2503 struct sock_filter filter[] = {
2504 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2505 offsetof(struct seccomp_data, nr)),
2506 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
2507 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL),
2508 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2509 };
2510 struct sock_fprog prog = {
2511 .len = (unsigned short)ARRAY_SIZE(filter),
2512 .filter = filter,
2513 };
2514
2515 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2516 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2517 }
2518
2519 /* Check prctl failure detection by requesting sib 0 diverge. */
2520 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2521 ASSERT_NE(ENOSYS, errno) {
2522 TH_LOG("Kernel does not support seccomp syscall!");
2523 }
2524 ASSERT_EQ(0, ret) {
2525 TH_LOG("setting filter failed");
2526 }
2527
2528 self->sibling[0].diverge = 1;
2529 tsync_start_sibling(&self->sibling[0]);
2530 tsync_start_sibling(&self->sibling[1]);
2531
2532 while (self->sibling_count < TSYNC_SIBLINGS) {
2533 sem_wait(&self->started);
2534 self->sibling_count++;
2535 }
2536
2537 /* Signal the threads to clean up*/
2538 pthread_mutex_lock(&self->mutex);
2539 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2540 TH_LOG("cond broadcast non-zero");
2541 }
2542 pthread_mutex_unlock(&self->mutex);
2543
2544 /* Ensure diverging sibling failed to call prctl. */
2545 PTHREAD_JOIN(self->sibling[0].tid, &status);
2546 EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status);
2547 PTHREAD_JOIN(self->sibling[1].tid, &status);
2548 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2549 }
2550
2551 TEST_F(TSYNC, two_siblings_with_ancestor)
2552 {
2553 long ret;
2554 void *status;
2555
2556 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2557 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2558 }
2559
2560 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2561 ASSERT_NE(ENOSYS, errno) {
2562 TH_LOG("Kernel does not support seccomp syscall!");
2563 }
2564 ASSERT_EQ(0, ret) {
2565 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2566 }
2567 tsync_start_sibling(&self->sibling[0]);
2568 tsync_start_sibling(&self->sibling[1]);
2569
2570 while (self->sibling_count < TSYNC_SIBLINGS) {
2571 sem_wait(&self->started);
2572 self->sibling_count++;
2573 }
2574
2575 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2576 &self->apply_prog);
2577 ASSERT_EQ(0, ret) {
2578 TH_LOG("Could install filter on all threads!");
2579 }
2580 /* Tell the siblings to test the policy */
2581 pthread_mutex_lock(&self->mutex);
2582 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2583 TH_LOG("cond broadcast non-zero");
2584 }
2585 pthread_mutex_unlock(&self->mutex);
2586 /* Ensure they are both killed and don't exit cleanly. */
2587 PTHREAD_JOIN(self->sibling[0].tid, &status);
2588 EXPECT_EQ(0x0, (long)status);
2589 PTHREAD_JOIN(self->sibling[1].tid, &status);
2590 EXPECT_EQ(0x0, (long)status);
2591 }
2592
2593 TEST_F(TSYNC, two_sibling_want_nnp)
2594 {
2595 void *status;
2596
2597 /* start siblings before any prctl() operations */
2598 tsync_start_sibling(&self->sibling[0]);
2599 tsync_start_sibling(&self->sibling[1]);
2600 while (self->sibling_count < TSYNC_SIBLINGS) {
2601 sem_wait(&self->started);
2602 self->sibling_count++;
2603 }
2604
2605 /* Tell the siblings to test no policy */
2606 pthread_mutex_lock(&self->mutex);
2607 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2608 TH_LOG("cond broadcast non-zero");
2609 }
2610 pthread_mutex_unlock(&self->mutex);
2611
2612 /* Ensure they are both upset about lacking nnp. */
2613 PTHREAD_JOIN(self->sibling[0].tid, &status);
2614 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2615 PTHREAD_JOIN(self->sibling[1].tid, &status);
2616 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2617 }
2618
2619 TEST_F(TSYNC, two_siblings_with_no_filter)
2620 {
2621 long ret;
2622 void *status;
2623
2624 /* start siblings before any prctl() operations */
2625 tsync_start_sibling(&self->sibling[0]);
2626 tsync_start_sibling(&self->sibling[1]);
2627 while (self->sibling_count < TSYNC_SIBLINGS) {
2628 sem_wait(&self->started);
2629 self->sibling_count++;
2630 }
2631
2632 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2633 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2634 }
2635
2636 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2637 &self->apply_prog);
2638 ASSERT_NE(ENOSYS, errno) {
2639 TH_LOG("Kernel does not support seccomp syscall!");
2640 }
2641 ASSERT_EQ(0, ret) {
2642 TH_LOG("Could install filter on all threads!");
2643 }
2644
2645 /* Tell the siblings to test the policy */
2646 pthread_mutex_lock(&self->mutex);
2647 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2648 TH_LOG("cond broadcast non-zero");
2649 }
2650 pthread_mutex_unlock(&self->mutex);
2651
2652 /* Ensure they are both killed and don't exit cleanly. */
2653 PTHREAD_JOIN(self->sibling[0].tid, &status);
2654 EXPECT_EQ(0x0, (long)status);
2655 PTHREAD_JOIN(self->sibling[1].tid, &status);
2656 EXPECT_EQ(0x0, (long)status);
2657 }
2658
2659 TEST_F(TSYNC, two_siblings_with_one_divergence)
2660 {
2661 long ret;
2662 void *status;
2663
2664 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2665 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2666 }
2667
2668 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2669 ASSERT_NE(ENOSYS, errno) {
2670 TH_LOG("Kernel does not support seccomp syscall!");
2671 }
2672 ASSERT_EQ(0, ret) {
2673 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2674 }
2675 self->sibling[0].diverge = 1;
2676 tsync_start_sibling(&self->sibling[0]);
2677 tsync_start_sibling(&self->sibling[1]);
2678
2679 while (self->sibling_count < TSYNC_SIBLINGS) {
2680 sem_wait(&self->started);
2681 self->sibling_count++;
2682 }
2683
2684 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2685 &self->apply_prog);
2686 ASSERT_EQ(self->sibling[0].system_tid, ret) {
2687 TH_LOG("Did not fail on diverged sibling.");
2688 }
2689
2690 /* Wake the threads */
2691 pthread_mutex_lock(&self->mutex);
2692 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2693 TH_LOG("cond broadcast non-zero");
2694 }
2695 pthread_mutex_unlock(&self->mutex);
2696
2697 /* Ensure they are both unkilled. */
2698 PTHREAD_JOIN(self->sibling[0].tid, &status);
2699 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2700 PTHREAD_JOIN(self->sibling[1].tid, &status);
2701 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2702 }
2703
2704 TEST_F(TSYNC, two_siblings_with_one_divergence_no_tid_in_err)
2705 {
2706 long ret, flags;
2707 void *status;
2708
2709 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2710 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2711 }
2712
2713 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2714 ASSERT_NE(ENOSYS, errno) {
2715 TH_LOG("Kernel does not support seccomp syscall!");
2716 }
2717 ASSERT_EQ(0, ret) {
2718 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2719 }
2720 self->sibling[0].diverge = 1;
2721 tsync_start_sibling(&self->sibling[0]);
2722 tsync_start_sibling(&self->sibling[1]);
2723
2724 while (self->sibling_count < TSYNC_SIBLINGS) {
2725 sem_wait(&self->started);
2726 self->sibling_count++;
2727 }
2728
2729 flags = SECCOMP_FILTER_FLAG_TSYNC | \
2730 SECCOMP_FILTER_FLAG_TSYNC_ESRCH;
2731 ret = seccomp(SECCOMP_SET_MODE_FILTER, flags, &self->apply_prog);
2732 ASSERT_EQ(ESRCH, errno) {
2733 TH_LOG("Did not return ESRCH for diverged sibling.");
2734 }
2735 ASSERT_EQ(-1, ret) {
2736 TH_LOG("Did not fail on diverged sibling.");
2737 }
2738
2739 /* Wake the threads */
2740 pthread_mutex_lock(&self->mutex);
2741 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2742 TH_LOG("cond broadcast non-zero");
2743 }
2744 pthread_mutex_unlock(&self->mutex);
2745
2746 /* Ensure they are both unkilled. */
2747 PTHREAD_JOIN(self->sibling[0].tid, &status);
2748 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2749 PTHREAD_JOIN(self->sibling[1].tid, &status);
2750 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2751 }
2752
2753 TEST_F(TSYNC, two_siblings_not_under_filter)
2754 {
2755 long ret, sib;
2756 void *status;
2757 struct timespec delay = { .tv_nsec = 100000000 };
2758
2759 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2760 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2761 }
2762
2763 /*
2764 * Sibling 0 will have its own seccomp policy
2765 * and Sibling 1 will not be under seccomp at
2766 * all. Sibling 1 will enter seccomp and 0
2767 * will cause failure.
2768 */
2769 self->sibling[0].diverge = 1;
2770 tsync_start_sibling(&self->sibling[0]);
2771 tsync_start_sibling(&self->sibling[1]);
2772
2773 while (self->sibling_count < TSYNC_SIBLINGS) {
2774 sem_wait(&self->started);
2775 self->sibling_count++;
2776 }
2777
2778 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2779 ASSERT_NE(ENOSYS, errno) {
2780 TH_LOG("Kernel does not support seccomp syscall!");
2781 }
2782 ASSERT_EQ(0, ret) {
2783 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2784 }
2785
2786 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2787 &self->apply_prog);
2788 ASSERT_EQ(ret, self->sibling[0].system_tid) {
2789 TH_LOG("Did not fail on diverged sibling.");
2790 }
2791 sib = 1;
2792 if (ret == self->sibling[0].system_tid)
2793 sib = 0;
2794
2795 pthread_mutex_lock(&self->mutex);
2796
2797 /* Increment the other siblings num_waits so we can clean up
2798 * the one we just saw.
2799 */
2800 self->sibling[!sib].num_waits += 1;
2801
2802 /* Signal the thread to clean up*/
2803 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2804 TH_LOG("cond broadcast non-zero");
2805 }
2806 pthread_mutex_unlock(&self->mutex);
2807 PTHREAD_JOIN(self->sibling[sib].tid, &status);
2808 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2809 /* Poll for actual task death. pthread_join doesn't guarantee it. */
2810 while (!kill(self->sibling[sib].system_tid, 0))
2811 nanosleep(&delay, NULL);
2812 /* Switch to the remaining sibling */
2813 sib = !sib;
2814
2815 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2816 &self->apply_prog);
2817 ASSERT_EQ(0, ret) {
2818 TH_LOG("Expected the remaining sibling to sync");
2819 };
2820
2821 pthread_mutex_lock(&self->mutex);
2822
2823 /* If remaining sibling didn't have a chance to wake up during
2824 * the first broadcast, manually reduce the num_waits now.
2825 */
2826 if (self->sibling[sib].num_waits > 1)
2827 self->sibling[sib].num_waits = 1;
2828 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2829 TH_LOG("cond broadcast non-zero");
2830 }
2831 pthread_mutex_unlock(&self->mutex);
2832 PTHREAD_JOIN(self->sibling[sib].tid, &status);
2833 EXPECT_EQ(0, (long)status);
2834 /* Poll for actual task death. pthread_join doesn't guarantee it. */
2835 while (!kill(self->sibling[sib].system_tid, 0))
2836 nanosleep(&delay, NULL);
2837
2838 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2839 &self->apply_prog);
2840 ASSERT_EQ(0, ret); /* just us chickens */
2841 }
2842
2843 /* Make sure restarted syscalls are seen directly as "restart_syscall". */
2844 TEST(syscall_restart)
2845 {
2846 long ret;
2847 unsigned long msg;
2848 pid_t child_pid;
2849 int pipefd[2];
2850 int status;
2851 siginfo_t info = { };
2852 struct sock_filter filter[] = {
2853 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2854 offsetof(struct seccomp_data, nr)),
2855
2856 #ifdef __NR_sigreturn
2857 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 7, 0),
2858 #endif
2859 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 6, 0),
2860 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 5, 0),
2861 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 4, 0),
2862 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 5, 0),
2863 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_clock_nanosleep, 4, 0),
2864 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0),
2865
2866 /* Allow __NR_write for easy logging. */
2867 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1),
2868 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2869 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2870 /* The nanosleep jump target. */
2871 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100),
2872 /* The restart_syscall jump target. */
2873 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200),
2874 };
2875 struct sock_fprog prog = {
2876 .len = (unsigned short)ARRAY_SIZE(filter),
2877 .filter = filter,
2878 };
2879 #if defined(__arm__)
2880 struct utsname utsbuf;
2881 #endif
2882
2883 ASSERT_EQ(0, pipe(pipefd));
2884
2885 child_pid = fork();
2886 ASSERT_LE(0, child_pid);
2887 if (child_pid == 0) {
2888 /* Child uses EXPECT not ASSERT to deliver status correctly. */
2889 char buf = ' ';
2890 struct timespec timeout = { };
2891
2892 /* Attach parent as tracer and stop. */
2893 EXPECT_EQ(0, ptrace(PTRACE_TRACEME));
2894 EXPECT_EQ(0, raise(SIGSTOP));
2895
2896 EXPECT_EQ(0, close(pipefd[1]));
2897
2898 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2899 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2900 }
2901
2902 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2903 EXPECT_EQ(0, ret) {
2904 TH_LOG("Failed to install filter!");
2905 }
2906
2907 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2908 TH_LOG("Failed to read() sync from parent");
2909 }
2910 EXPECT_EQ('.', buf) {
2911 TH_LOG("Failed to get sync data from read()");
2912 }
2913
2914 /* Start nanosleep to be interrupted. */
2915 timeout.tv_sec = 1;
2916 errno = 0;
2917 EXPECT_EQ(0, nanosleep(&timeout, NULL)) {
2918 TH_LOG("Call to nanosleep() failed (errno %d)", errno);
2919 }
2920
2921 /* Read final sync from parent. */
2922 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2923 TH_LOG("Failed final read() from parent");
2924 }
2925 EXPECT_EQ('!', buf) {
2926 TH_LOG("Failed to get final data from read()");
2927 }
2928
2929 /* Directly report the status of our test harness results. */
2930 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS
2931 : EXIT_FAILURE);
2932 }
2933 EXPECT_EQ(0, close(pipefd[0]));
2934
2935 /* Attach to child, setup options, and release. */
2936 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2937 ASSERT_EQ(true, WIFSTOPPED(status));
2938 ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL,
2939 PTRACE_O_TRACESECCOMP));
2940 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2941 ASSERT_EQ(1, write(pipefd[1], ".", 1));
2942
2943 /* Wait for nanosleep() to start. */
2944 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2945 ASSERT_EQ(true, WIFSTOPPED(status));
2946 ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2947 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2948 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2949 ASSERT_EQ(0x100, msg);
2950 ret = get_syscall(_metadata, child_pid);
2951 EXPECT_TRUE(ret == __NR_nanosleep || ret == __NR_clock_nanosleep);
2952
2953 /* Might as well check siginfo for sanity while we're here. */
2954 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2955 ASSERT_EQ(SIGTRAP, info.si_signo);
2956 ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code);
2957 EXPECT_EQ(0, info.si_errno);
2958 EXPECT_EQ(getuid(), info.si_uid);
2959 /* Verify signal delivery came from child (seccomp-triggered). */
2960 EXPECT_EQ(child_pid, info.si_pid);
2961
2962 /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */
2963 ASSERT_EQ(0, kill(child_pid, SIGSTOP));
2964 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2965 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2966 ASSERT_EQ(true, WIFSTOPPED(status));
2967 ASSERT_EQ(SIGSTOP, WSTOPSIG(status));
2968 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2969 /*
2970 * There is no siginfo on SIGSTOP any more, so we can't verify
2971 * signal delivery came from parent now (getpid() == info.si_pid).
2972 * https://lkml.kernel.org/r/CAGXu5jJaZAOzP1qFz66tYrtbuywqb+UN2SOA1VLHpCCOiYvYeg@mail.gmail.com
2973 * At least verify the SIGSTOP via PTRACE_GETSIGINFO.
2974 */
2975 EXPECT_EQ(SIGSTOP, info.si_signo);
2976
2977 /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */
2978 ASSERT_EQ(0, kill(child_pid, SIGCONT));
2979 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2980 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2981 ASSERT_EQ(true, WIFSTOPPED(status));
2982 ASSERT_EQ(SIGCONT, WSTOPSIG(status));
2983 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2984
2985 /* Wait for restart_syscall() to start. */
2986 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2987 ASSERT_EQ(true, WIFSTOPPED(status));
2988 ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2989 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2990 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2991
2992 ASSERT_EQ(0x200, msg);
2993 ret = get_syscall(_metadata, child_pid);
2994 #if defined(__arm__)
2995 /*
2996 * FIXME:
2997 * - native ARM registers do NOT expose true syscall.
2998 * - compat ARM registers on ARM64 DO expose true syscall.
2999 */
3000 ASSERT_EQ(0, uname(&utsbuf));
3001 if (strncmp(utsbuf.machine, "arm", 3) == 0) {
3002 EXPECT_EQ(__NR_nanosleep, ret);
3003 } else
3004 #endif
3005 {
3006 EXPECT_EQ(__NR_restart_syscall, ret);
3007 }
3008
3009 /* Write again to end test. */
3010 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
3011 ASSERT_EQ(1, write(pipefd[1], "!", 1));
3012 EXPECT_EQ(0, close(pipefd[1]));
3013
3014 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
3015 if (WIFSIGNALED(status) || WEXITSTATUS(status))
3016 _metadata->passed = 0;
3017 }
3018
3019 TEST_SIGNAL(filter_flag_log, SIGSYS)
3020 {
3021 struct sock_filter allow_filter[] = {
3022 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3023 };
3024 struct sock_filter kill_filter[] = {
3025 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
3026 offsetof(struct seccomp_data, nr)),
3027 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
3028 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
3029 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3030 };
3031 struct sock_fprog allow_prog = {
3032 .len = (unsigned short)ARRAY_SIZE(allow_filter),
3033 .filter = allow_filter,
3034 };
3035 struct sock_fprog kill_prog = {
3036 .len = (unsigned short)ARRAY_SIZE(kill_filter),
3037 .filter = kill_filter,
3038 };
3039 long ret;
3040 pid_t parent = getppid();
3041
3042 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3043 ASSERT_EQ(0, ret);
3044
3045 /* Verify that the FILTER_FLAG_LOG flag isn't accepted in strict mode */
3046 ret = seccomp(SECCOMP_SET_MODE_STRICT, SECCOMP_FILTER_FLAG_LOG,
3047 &allow_prog);
3048 ASSERT_NE(ENOSYS, errno) {
3049 TH_LOG("Kernel does not support seccomp syscall!");
3050 }
3051 EXPECT_NE(0, ret) {
3052 TH_LOG("Kernel accepted FILTER_FLAG_LOG flag in strict mode!");
3053 }
3054 EXPECT_EQ(EINVAL, errno) {
3055 TH_LOG("Kernel returned unexpected errno for FILTER_FLAG_LOG flag in strict mode!");
3056 }
3057
3058 /* Verify that a simple, permissive filter can be added with no flags */
3059 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog);
3060 EXPECT_EQ(0, ret);
3061
3062 /* See if the same filter can be added with the FILTER_FLAG_LOG flag */
3063 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
3064 &allow_prog);
3065 ASSERT_NE(EINVAL, errno) {
3066 TH_LOG("Kernel does not support the FILTER_FLAG_LOG flag!");
3067 }
3068 EXPECT_EQ(0, ret);
3069
3070 /* Ensure that the kill filter works with the FILTER_FLAG_LOG flag */
3071 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
3072 &kill_prog);
3073 EXPECT_EQ(0, ret);
3074
3075 EXPECT_EQ(parent, syscall(__NR_getppid));
3076 /* getpid() should never return. */
3077 EXPECT_EQ(0, syscall(__NR_getpid));
3078 }
3079
3080 TEST(get_action_avail)
3081 {
3082 __u32 actions[] = { SECCOMP_RET_KILL_THREAD, SECCOMP_RET_TRAP,
3083 SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE,
3084 SECCOMP_RET_LOG, SECCOMP_RET_ALLOW };
3085 __u32 unknown_action = 0x10000000U;
3086 int i;
3087 long ret;
3088
3089 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[0]);
3090 ASSERT_NE(ENOSYS, errno) {
3091 TH_LOG("Kernel does not support seccomp syscall!");
3092 }
3093 ASSERT_NE(EINVAL, errno) {
3094 TH_LOG("Kernel does not support SECCOMP_GET_ACTION_AVAIL operation!");
3095 }
3096 EXPECT_EQ(ret, 0);
3097
3098 for (i = 0; i < ARRAY_SIZE(actions); i++) {
3099 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[i]);
3100 EXPECT_EQ(ret, 0) {
3101 TH_LOG("Expected action (0x%X) not available!",
3102 actions[i]);
3103 }
3104 }
3105
3106 /* Check that an unknown action is handled properly (EOPNOTSUPP) */
3107 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &unknown_action);
3108 EXPECT_EQ(ret, -1);
3109 EXPECT_EQ(errno, EOPNOTSUPP);
3110 }
3111
3112 TEST(get_metadata)
3113 {
3114 pid_t pid;
3115 int pipefd[2];
3116 char buf;
3117 struct seccomp_metadata md;
3118 long ret;
3119
3120 /* Only real root can get metadata. */
3121 if (geteuid()) {
3122 SKIP(return, "get_metadata requires real root");
3123 return;
3124 }
3125
3126 ASSERT_EQ(0, pipe(pipefd));
3127
3128 pid = fork();
3129 ASSERT_GE(pid, 0);
3130 if (pid == 0) {
3131 struct sock_filter filter[] = {
3132 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3133 };
3134 struct sock_fprog prog = {
3135 .len = (unsigned short)ARRAY_SIZE(filter),
3136 .filter = filter,
3137 };
3138
3139 /* one with log, one without */
3140 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER,
3141 SECCOMP_FILTER_FLAG_LOG, &prog));
3142 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog));
3143
3144 EXPECT_EQ(0, close(pipefd[0]));
3145 ASSERT_EQ(1, write(pipefd[1], "1", 1));
3146 ASSERT_EQ(0, close(pipefd[1]));
3147
3148 while (1)
3149 sleep(100);
3150 }
3151
3152 ASSERT_EQ(0, close(pipefd[1]));
3153 ASSERT_EQ(1, read(pipefd[0], &buf, 1));
3154
3155 ASSERT_EQ(0, ptrace(PTRACE_ATTACH, pid));
3156 ASSERT_EQ(pid, waitpid(pid, NULL, 0));
3157
3158 /* Past here must not use ASSERT or child process is never killed. */
3159
3160 md.filter_off = 0;
3161 errno = 0;
3162 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md);
3163 EXPECT_EQ(sizeof(md), ret) {
3164 if (errno == EINVAL)
3165 SKIP(goto skip, "Kernel does not support PTRACE_SECCOMP_GET_METADATA (missing CONFIG_CHECKPOINT_RESTORE?)");
3166 }
3167
3168 EXPECT_EQ(md.flags, SECCOMP_FILTER_FLAG_LOG);
3169 EXPECT_EQ(md.filter_off, 0);
3170
3171 md.filter_off = 1;
3172 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md);
3173 EXPECT_EQ(sizeof(md), ret);
3174 EXPECT_EQ(md.flags, 0);
3175 EXPECT_EQ(md.filter_off, 1);
3176
3177 skip:
3178 ASSERT_EQ(0, kill(pid, SIGKILL));
3179 }
3180
3181 static int user_notif_syscall(int nr, unsigned int flags)
3182 {
3183 struct sock_filter filter[] = {
3184 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
3185 offsetof(struct seccomp_data, nr)),
3186 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, nr, 0, 1),
3187 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_USER_NOTIF),
3188 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3189 };
3190
3191 struct sock_fprog prog = {
3192 .len = (unsigned short)ARRAY_SIZE(filter),
3193 .filter = filter,
3194 };
3195
3196 return seccomp(SECCOMP_SET_MODE_FILTER, flags, &prog);
3197 }
3198
3199 #define USER_NOTIF_MAGIC INT_MAX
3200 TEST(user_notification_basic)
3201 {
3202 pid_t pid;
3203 long ret;
3204 int status, listener;
3205 struct seccomp_notif req = {};
3206 struct seccomp_notif_resp resp = {};
3207 struct pollfd pollfd;
3208
3209 struct sock_filter filter[] = {
3210 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
3211 };
3212 struct sock_fprog prog = {
3213 .len = (unsigned short)ARRAY_SIZE(filter),
3214 .filter = filter,
3215 };
3216
3217 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3218 ASSERT_EQ(0, ret) {
3219 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3220 }
3221
3222 pid = fork();
3223 ASSERT_GE(pid, 0);
3224
3225 /* Check that we get -ENOSYS with no listener attached */
3226 if (pid == 0) {
3227 if (user_notif_syscall(__NR_getppid, 0) < 0)
3228 exit(1);
3229 ret = syscall(__NR_getppid);
3230 exit(ret >= 0 || errno != ENOSYS);
3231 }
3232
3233 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3234 EXPECT_EQ(true, WIFEXITED(status));
3235 EXPECT_EQ(0, WEXITSTATUS(status));
3236
3237 /* Add some no-op filters for grins. */
3238 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3239 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3240 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3241 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0);
3242
3243 /* Check that the basic notification machinery works */
3244 listener = user_notif_syscall(__NR_getppid,
3245 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3246 ASSERT_GE(listener, 0);
3247
3248 /* Installing a second listener in the chain should EBUSY */
3249 EXPECT_EQ(user_notif_syscall(__NR_getppid,
3250 SECCOMP_FILTER_FLAG_NEW_LISTENER),
3251 -1);
3252 EXPECT_EQ(errno, EBUSY);
3253
3254 pid = fork();
3255 ASSERT_GE(pid, 0);
3256
3257 if (pid == 0) {
3258 ret = syscall(__NR_getppid);
3259 exit(ret != USER_NOTIF_MAGIC);
3260 }
3261
3262 pollfd.fd = listener;
3263 pollfd.events = POLLIN | POLLOUT;
3264
3265 EXPECT_GT(poll(&pollfd, 1, -1), 0);
3266 EXPECT_EQ(pollfd.revents, POLLIN);
3267
3268 /* Test that we can't pass garbage to the kernel. */
3269 memset(&req, 0, sizeof(req));
3270 req.pid = -1;
3271 errno = 0;
3272 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req);
3273 EXPECT_EQ(-1, ret);
3274 EXPECT_EQ(EINVAL, errno);
3275
3276 if (ret) {
3277 req.pid = 0;
3278 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3279 }
3280
3281 pollfd.fd = listener;
3282 pollfd.events = POLLIN | POLLOUT;
3283
3284 EXPECT_GT(poll(&pollfd, 1, -1), 0);
3285 EXPECT_EQ(pollfd.revents, POLLOUT);
3286
3287 EXPECT_EQ(req.data.nr, __NR_getppid);
3288
3289 resp.id = req.id;
3290 resp.error = 0;
3291 resp.val = USER_NOTIF_MAGIC;
3292
3293 /* check that we make sure flags == 0 */
3294 resp.flags = 1;
3295 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3296 EXPECT_EQ(errno, EINVAL);
3297
3298 resp.flags = 0;
3299 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3300
3301 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3302 EXPECT_EQ(true, WIFEXITED(status));
3303 EXPECT_EQ(0, WEXITSTATUS(status));
3304 }
3305
3306 TEST(user_notification_with_tsync)
3307 {
3308 int ret;
3309 unsigned int flags;
3310
3311 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3312 ASSERT_EQ(0, ret) {
3313 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3314 }
3315
3316 /* these were exclusive */
3317 flags = SECCOMP_FILTER_FLAG_NEW_LISTENER |
3318 SECCOMP_FILTER_FLAG_TSYNC;
3319 ASSERT_EQ(-1, user_notif_syscall(__NR_getppid, flags));
3320 ASSERT_EQ(EINVAL, errno);
3321
3322 /* but now they're not */
3323 flags |= SECCOMP_FILTER_FLAG_TSYNC_ESRCH;
3324 ret = user_notif_syscall(__NR_getppid, flags);
3325 close(ret);
3326 ASSERT_LE(0, ret);
3327 }
3328
3329 TEST(user_notification_kill_in_middle)
3330 {
3331 pid_t pid;
3332 long ret;
3333 int listener;
3334 struct seccomp_notif req = {};
3335 struct seccomp_notif_resp resp = {};
3336
3337 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3338 ASSERT_EQ(0, ret) {
3339 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3340 }
3341
3342 listener = user_notif_syscall(__NR_getppid,
3343 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3344 ASSERT_GE(listener, 0);
3345
3346 /*
3347 * Check that nothing bad happens when we kill the task in the middle
3348 * of a syscall.
3349 */
3350 pid = fork();
3351 ASSERT_GE(pid, 0);
3352
3353 if (pid == 0) {
3354 ret = syscall(__NR_getppid);
3355 exit(ret != USER_NOTIF_MAGIC);
3356 }
3357
3358 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3359 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), 0);
3360
3361 EXPECT_EQ(kill(pid, SIGKILL), 0);
3362 EXPECT_EQ(waitpid(pid, NULL, 0), pid);
3363
3364 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), -1);
3365
3366 resp.id = req.id;
3367 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp);
3368 EXPECT_EQ(ret, -1);
3369 EXPECT_EQ(errno, ENOENT);
3370 }
3371
3372 static int handled = -1;
3373
3374 static void signal_handler(int signal)
3375 {
3376 if (write(handled, "c", 1) != 1)
3377 perror("write from signal");
3378 }
3379
3380 TEST(user_notification_signal)
3381 {
3382 pid_t pid;
3383 long ret;
3384 int status, listener, sk_pair[2];
3385 struct seccomp_notif req = {};
3386 struct seccomp_notif_resp resp = {};
3387 char c;
3388
3389 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3390 ASSERT_EQ(0, ret) {
3391 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3392 }
3393
3394 ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0);
3395
3396 listener = user_notif_syscall(__NR_gettid,
3397 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3398 ASSERT_GE(listener, 0);
3399
3400 pid = fork();
3401 ASSERT_GE(pid, 0);
3402
3403 if (pid == 0) {
3404 close(sk_pair[0]);
3405 handled = sk_pair[1];
3406 if (signal(SIGUSR1, signal_handler) == SIG_ERR) {
3407 perror("signal");
3408 exit(1);
3409 }
3410 /*
3411 * ERESTARTSYS behavior is a bit hard to test, because we need
3412 * to rely on a signal that has not yet been handled. Let's at
3413 * least check that the error code gets propagated through, and
3414 * hope that it doesn't break when there is actually a signal :)
3415 */
3416 ret = syscall(__NR_gettid);
3417 exit(!(ret == -1 && errno == 512));
3418 }
3419
3420 close(sk_pair[1]);
3421
3422 memset(&req, 0, sizeof(req));
3423 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3424
3425 EXPECT_EQ(kill(pid, SIGUSR1), 0);
3426
3427 /*
3428 * Make sure the signal really is delivered, which means we're not
3429 * stuck in the user notification code any more and the notification
3430 * should be dead.
3431 */
3432 EXPECT_EQ(read(sk_pair[0], &c, 1), 1);
3433
3434 resp.id = req.id;
3435 resp.error = -EPERM;
3436 resp.val = 0;
3437
3438 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3439 EXPECT_EQ(errno, ENOENT);
3440
3441 memset(&req, 0, sizeof(req));
3442 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3443
3444 resp.id = req.id;
3445 resp.error = -512; /* -ERESTARTSYS */
3446 resp.val = 0;
3447
3448 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3449
3450 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3451 EXPECT_EQ(true, WIFEXITED(status));
3452 EXPECT_EQ(0, WEXITSTATUS(status));
3453 }
3454
3455 TEST(user_notification_closed_listener)
3456 {
3457 pid_t pid;
3458 long ret;
3459 int status, listener;
3460
3461 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3462 ASSERT_EQ(0, ret) {
3463 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3464 }
3465
3466 listener = user_notif_syscall(__NR_getppid,
3467 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3468 ASSERT_GE(listener, 0);
3469
3470 /*
3471 * Check that we get an ENOSYS when the listener is closed.
3472 */
3473 pid = fork();
3474 ASSERT_GE(pid, 0);
3475 if (pid == 0) {
3476 close(listener);
3477 ret = syscall(__NR_getppid);
3478 exit(ret != -1 && errno != ENOSYS);
3479 }
3480
3481 close(listener);
3482
3483 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3484 EXPECT_EQ(true, WIFEXITED(status));
3485 EXPECT_EQ(0, WEXITSTATUS(status));
3486 }
3487
3488 /*
3489 * Check that a pid in a child namespace still shows up as valid in ours.
3490 */
3491 TEST(user_notification_child_pid_ns)
3492 {
3493 pid_t pid;
3494 int status, listener;
3495 struct seccomp_notif req = {};
3496 struct seccomp_notif_resp resp = {};
3497
3498 ASSERT_EQ(unshare(CLONE_NEWUSER | CLONE_NEWPID), 0) {
3499 if (errno == EINVAL)
3500 SKIP(return, "kernel missing CLONE_NEWUSER support");
3501 };
3502
3503 listener = user_notif_syscall(__NR_getppid,
3504 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3505 ASSERT_GE(listener, 0);
3506
3507 pid = fork();
3508 ASSERT_GE(pid, 0);
3509
3510 if (pid == 0)
3511 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3512
3513 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3514 EXPECT_EQ(req.pid, pid);
3515
3516 resp.id = req.id;
3517 resp.error = 0;
3518 resp.val = USER_NOTIF_MAGIC;
3519
3520 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3521
3522 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3523 EXPECT_EQ(true, WIFEXITED(status));
3524 EXPECT_EQ(0, WEXITSTATUS(status));
3525 close(listener);
3526 }
3527
3528 /*
3529 * Check that a pid in a sibling (i.e. unrelated) namespace shows up as 0, i.e.
3530 * invalid.
3531 */
3532 TEST(user_notification_sibling_pid_ns)
3533 {
3534 pid_t pid, pid2;
3535 int status, listener;
3536 struct seccomp_notif req = {};
3537 struct seccomp_notif_resp resp = {};
3538
3539 ASSERT_EQ(prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0), 0) {
3540 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3541 }
3542
3543 listener = user_notif_syscall(__NR_getppid,
3544 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3545 ASSERT_GE(listener, 0);
3546
3547 pid = fork();
3548 ASSERT_GE(pid, 0);
3549
3550 if (pid == 0) {
3551 ASSERT_EQ(unshare(CLONE_NEWPID), 0);
3552
3553 pid2 = fork();
3554 ASSERT_GE(pid2, 0);
3555
3556 if (pid2 == 0)
3557 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3558
3559 EXPECT_EQ(waitpid(pid2, &status, 0), pid2);
3560 EXPECT_EQ(true, WIFEXITED(status));
3561 EXPECT_EQ(0, WEXITSTATUS(status));
3562 exit(WEXITSTATUS(status));
3563 }
3564
3565 /* Create the sibling ns, and sibling in it. */
3566 ASSERT_EQ(unshare(CLONE_NEWPID), 0) {
3567 if (errno == EPERM)
3568 SKIP(return, "CLONE_NEWPID requires CAP_SYS_ADMIN");
3569 }
3570 ASSERT_EQ(errno, 0);
3571
3572 pid2 = fork();
3573 ASSERT_GE(pid2, 0);
3574
3575 if (pid2 == 0) {
3576 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3577 /*
3578 * The pid should be 0, i.e. the task is in some namespace that
3579 * we can't "see".
3580 */
3581 EXPECT_EQ(req.pid, 0);
3582
3583 resp.id = req.id;
3584 resp.error = 0;
3585 resp.val = USER_NOTIF_MAGIC;
3586
3587 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3588 exit(0);
3589 }
3590
3591 close(listener);
3592
3593 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3594 EXPECT_EQ(true, WIFEXITED(status));
3595 EXPECT_EQ(0, WEXITSTATUS(status));
3596
3597 EXPECT_EQ(waitpid(pid2, &status, 0), pid2);
3598 EXPECT_EQ(true, WIFEXITED(status));
3599 EXPECT_EQ(0, WEXITSTATUS(status));
3600 }
3601
3602 TEST(user_notification_fault_recv)
3603 {
3604 pid_t pid;
3605 int status, listener;
3606 struct seccomp_notif req = {};
3607 struct seccomp_notif_resp resp = {};
3608
3609 ASSERT_EQ(unshare(CLONE_NEWUSER), 0);
3610
3611 listener = user_notif_syscall(__NR_getppid,
3612 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3613 ASSERT_GE(listener, 0);
3614
3615 pid = fork();
3616 ASSERT_GE(pid, 0);
3617
3618 if (pid == 0)
3619 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3620
3621 /* Do a bad recv() */
3622 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, NULL), -1);
3623 EXPECT_EQ(errno, EFAULT);
3624
3625 /* We should still be able to receive this notification, though. */
3626 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3627 EXPECT_EQ(req.pid, pid);
3628
3629 resp.id = req.id;
3630 resp.error = 0;
3631 resp.val = USER_NOTIF_MAGIC;
3632
3633 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3634
3635 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3636 EXPECT_EQ(true, WIFEXITED(status));
3637 EXPECT_EQ(0, WEXITSTATUS(status));
3638 }
3639
3640 TEST(seccomp_get_notif_sizes)
3641 {
3642 struct seccomp_notif_sizes sizes;
3643
3644 ASSERT_EQ(seccomp(SECCOMP_GET_NOTIF_SIZES, 0, &sizes), 0);
3645 EXPECT_EQ(sizes.seccomp_notif, sizeof(struct seccomp_notif));
3646 EXPECT_EQ(sizes.seccomp_notif_resp, sizeof(struct seccomp_notif_resp));
3647 }
3648
3649 TEST(user_notification_continue)
3650 {
3651 pid_t pid;
3652 long ret;
3653 int status, listener;
3654 struct seccomp_notif req = {};
3655 struct seccomp_notif_resp resp = {};
3656 struct pollfd pollfd;
3657
3658 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3659 ASSERT_EQ(0, ret) {
3660 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3661 }
3662
3663 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER);
3664 ASSERT_GE(listener, 0);
3665
3666 pid = fork();
3667 ASSERT_GE(pid, 0);
3668
3669 if (pid == 0) {
3670 int dup_fd, pipe_fds[2];
3671 pid_t self;
3672
3673 ASSERT_GE(pipe(pipe_fds), 0);
3674
3675 dup_fd = dup(pipe_fds[0]);
3676 ASSERT_GE(dup_fd, 0);
3677 EXPECT_NE(pipe_fds[0], dup_fd);
3678
3679 self = getpid();
3680 ASSERT_EQ(filecmp(self, self, pipe_fds[0], dup_fd), 0);
3681 exit(0);
3682 }
3683
3684 pollfd.fd = listener;
3685 pollfd.events = POLLIN | POLLOUT;
3686
3687 EXPECT_GT(poll(&pollfd, 1, -1), 0);
3688 EXPECT_EQ(pollfd.revents, POLLIN);
3689
3690 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3691
3692 pollfd.fd = listener;
3693 pollfd.events = POLLIN | POLLOUT;
3694
3695 EXPECT_GT(poll(&pollfd, 1, -1), 0);
3696 EXPECT_EQ(pollfd.revents, POLLOUT);
3697
3698 EXPECT_EQ(req.data.nr, __NR_dup);
3699
3700 resp.id = req.id;
3701 resp.flags = SECCOMP_USER_NOTIF_FLAG_CONTINUE;
3702
3703 /*
3704 * Verify that setting SECCOMP_USER_NOTIF_FLAG_CONTINUE enforces other
3705 * args be set to 0.
3706 */
3707 resp.error = 0;
3708 resp.val = USER_NOTIF_MAGIC;
3709 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3710 EXPECT_EQ(errno, EINVAL);
3711
3712 resp.error = USER_NOTIF_MAGIC;
3713 resp.val = 0;
3714 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1);
3715 EXPECT_EQ(errno, EINVAL);
3716
3717 resp.error = 0;
3718 resp.val = 0;
3719 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0) {
3720 if (errno == EINVAL)
3721 SKIP(goto skip, "Kernel does not support SECCOMP_USER_NOTIF_FLAG_CONTINUE");
3722 }
3723
3724 skip:
3725 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3726 EXPECT_EQ(true, WIFEXITED(status));
3727 EXPECT_EQ(0, WEXITSTATUS(status)) {
3728 if (WEXITSTATUS(status) == 2) {
3729 SKIP(return, "Kernel does not support kcmp() syscall");
3730 return;
3731 }
3732 }
3733 }
3734
3735 TEST(user_notification_filter_empty)
3736 {
3737 pid_t pid;
3738 long ret;
3739 int status;
3740 struct pollfd pollfd;
3741 struct clone_args args = {
3742 .flags = CLONE_FILES,
3743 .exit_signal = SIGCHLD,
3744 };
3745
3746 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3747 ASSERT_EQ(0, ret) {
3748 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3749 }
3750
3751 pid = sys_clone3(&args, sizeof(args));
3752 ASSERT_GE(pid, 0);
3753
3754 if (pid == 0) {
3755 int listener;
3756
3757 listener = user_notif_syscall(__NR_mknodat, SECCOMP_FILTER_FLAG_NEW_LISTENER);
3758 if (listener < 0)
3759 _exit(EXIT_FAILURE);
3760
3761 if (dup2(listener, 200) != 200)
3762 _exit(EXIT_FAILURE);
3763
3764 close(listener);
3765
3766 _exit(EXIT_SUCCESS);
3767 }
3768
3769 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3770 EXPECT_EQ(true, WIFEXITED(status));
3771 EXPECT_EQ(0, WEXITSTATUS(status));
3772
3773 /*
3774 * The seccomp filter has become unused so we should be notified once
3775 * the kernel gets around to cleaning up task struct.
3776 */
3777 pollfd.fd = 200;
3778 pollfd.events = POLLHUP;
3779
3780 EXPECT_GT(poll(&pollfd, 1, 2000), 0);
3781 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0);
3782 }
3783
3784 static void *do_thread(void *data)
3785 {
3786 return NULL;
3787 }
3788
3789 TEST(user_notification_filter_empty_threaded)
3790 {
3791 pid_t pid;
3792 long ret;
3793 int status;
3794 struct pollfd pollfd;
3795 struct clone_args args = {
3796 .flags = CLONE_FILES,
3797 .exit_signal = SIGCHLD,
3798 };
3799
3800 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3801 ASSERT_EQ(0, ret) {
3802 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3803 }
3804
3805 pid = sys_clone3(&args, sizeof(args));
3806 ASSERT_GE(pid, 0);
3807
3808 if (pid == 0) {
3809 pid_t pid1, pid2;
3810 int listener, status;
3811 pthread_t thread;
3812
3813 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER);
3814 if (listener < 0)
3815 _exit(EXIT_FAILURE);
3816
3817 if (dup2(listener, 200) != 200)
3818 _exit(EXIT_FAILURE);
3819
3820 close(listener);
3821
3822 pid1 = fork();
3823 if (pid1 < 0)
3824 _exit(EXIT_FAILURE);
3825
3826 if (pid1 == 0)
3827 _exit(EXIT_SUCCESS);
3828
3829 pid2 = fork();
3830 if (pid2 < 0)
3831 _exit(EXIT_FAILURE);
3832
3833 if (pid2 == 0)
3834 _exit(EXIT_SUCCESS);
3835
3836 if (pthread_create(&thread, NULL, do_thread, NULL) ||
3837 pthread_join(thread, NULL))
3838 _exit(EXIT_FAILURE);
3839
3840 if (pthread_create(&thread, NULL, do_thread, NULL) ||
3841 pthread_join(thread, NULL))
3842 _exit(EXIT_FAILURE);
3843
3844 if (waitpid(pid1, &status, 0) != pid1 || !WIFEXITED(status) ||
3845 WEXITSTATUS(status))
3846 _exit(EXIT_FAILURE);
3847
3848 if (waitpid(pid2, &status, 0) != pid2 || !WIFEXITED(status) ||
3849 WEXITSTATUS(status))
3850 _exit(EXIT_FAILURE);
3851
3852 exit(EXIT_SUCCESS);
3853 }
3854
3855 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3856 EXPECT_EQ(true, WIFEXITED(status));
3857 EXPECT_EQ(0, WEXITSTATUS(status));
3858
3859 /*
3860 * The seccomp filter has become unused so we should be notified once
3861 * the kernel gets around to cleaning up task struct.
3862 */
3863 pollfd.fd = 200;
3864 pollfd.events = POLLHUP;
3865
3866 EXPECT_GT(poll(&pollfd, 1, 2000), 0);
3867 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0);
3868 }
3869
3870 TEST(user_notification_addfd)
3871 {
3872 pid_t pid;
3873 long ret;
3874 int status, listener, memfd, fd;
3875 struct seccomp_notif_addfd addfd = {};
3876 struct seccomp_notif_addfd_small small = {};
3877 struct seccomp_notif_addfd_big big = {};
3878 struct seccomp_notif req = {};
3879 struct seccomp_notif_resp resp = {};
3880 /* 100 ms */
3881 struct timespec delay = { .tv_nsec = 100000000 };
3882
3883 memfd = memfd_create("test", 0);
3884 ASSERT_GE(memfd, 0);
3885
3886 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
3887 ASSERT_EQ(0, ret) {
3888 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
3889 }
3890
3891 /* Check that the basic notification machinery works */
3892 listener = user_notif_syscall(__NR_getppid,
3893 SECCOMP_FILTER_FLAG_NEW_LISTENER);
3894 ASSERT_GE(listener, 0);
3895
3896 pid = fork();
3897 ASSERT_GE(pid, 0);
3898
3899 if (pid == 0) {
3900 if (syscall(__NR_getppid) != USER_NOTIF_MAGIC)
3901 exit(1);
3902 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
3903 }
3904
3905 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3906
3907 addfd.srcfd = memfd;
3908 addfd.newfd = 0;
3909 addfd.id = req.id;
3910 addfd.flags = 0x0;
3911
3912 /* Verify bad newfd_flags cannot be set */
3913 addfd.newfd_flags = ~O_CLOEXEC;
3914 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
3915 EXPECT_EQ(errno, EINVAL);
3916 addfd.newfd_flags = O_CLOEXEC;
3917
3918 /* Verify bad flags cannot be set */
3919 addfd.flags = 0xff;
3920 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
3921 EXPECT_EQ(errno, EINVAL);
3922 addfd.flags = 0;
3923
3924 /* Verify that remote_fd cannot be set without setting flags */
3925 addfd.newfd = 1;
3926 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
3927 EXPECT_EQ(errno, EINVAL);
3928 addfd.newfd = 0;
3929
3930 /* Verify small size cannot be set */
3931 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_SMALL, &small), -1);
3932 EXPECT_EQ(errno, EINVAL);
3933
3934 /* Verify we can't send bits filled in unknown buffer area */
3935 memset(&big, 0xAA, sizeof(big));
3936 big.addfd = addfd;
3937 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big), -1);
3938 EXPECT_EQ(errno, E2BIG);
3939
3940
3941 /* Verify we can set an arbitrary remote fd */
3942 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd);
3943 /*
3944 * The child has fds 0(stdin), 1(stdout), 2(stderr), 3(memfd),
3945 * 4(listener), so the newly allocated fd should be 5.
3946 */
3947 EXPECT_EQ(fd, 5);
3948 EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0);
3949
3950 /* Verify we can set an arbitrary remote fd with large size */
3951 memset(&big, 0x0, sizeof(big));
3952 big.addfd = addfd;
3953 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big);
3954 EXPECT_EQ(fd, 6);
3955
3956 /* Verify we can set a specific remote fd */
3957 addfd.newfd = 42;
3958 addfd.flags = SECCOMP_ADDFD_FLAG_SETFD;
3959 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd);
3960 EXPECT_EQ(fd, 42);
3961 EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0);
3962
3963 /* Resume syscall */
3964 resp.id = req.id;
3965 resp.error = 0;
3966 resp.val = USER_NOTIF_MAGIC;
3967 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3968
3969 /*
3970 * This sets the ID of the ADD FD to the last request plus 1. The
3971 * notification ID increments 1 per notification.
3972 */
3973 addfd.id = req.id + 1;
3974
3975 /* This spins until the underlying notification is generated */
3976 while (ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd) != -1 &&
3977 errno != -EINPROGRESS)
3978 nanosleep(&delay, NULL);
3979
3980 memset(&req, 0, sizeof(req));
3981 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
3982 ASSERT_EQ(addfd.id, req.id);
3983
3984 resp.id = req.id;
3985 resp.error = 0;
3986 resp.val = USER_NOTIF_MAGIC;
3987 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
3988
3989 /* Wait for child to finish. */
3990 EXPECT_EQ(waitpid(pid, &status, 0), pid);
3991 EXPECT_EQ(true, WIFEXITED(status));
3992 EXPECT_EQ(0, WEXITSTATUS(status));
3993
3994 close(memfd);
3995 }
3996
3997 TEST(user_notification_addfd_rlimit)
3998 {
3999 pid_t pid;
4000 long ret;
4001 int status, listener, memfd;
4002 struct seccomp_notif_addfd addfd = {};
4003 struct seccomp_notif req = {};
4004 struct seccomp_notif_resp resp = {};
4005 const struct rlimit lim = {
4006 .rlim_cur = 0,
4007 .rlim_max = 0,
4008 };
4009
4010 memfd = memfd_create("test", 0);
4011 ASSERT_GE(memfd, 0);
4012
4013 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
4014 ASSERT_EQ(0, ret) {
4015 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
4016 }
4017
4018 /* Check that the basic notification machinery works */
4019 listener = user_notif_syscall(__NR_getppid,
4020 SECCOMP_FILTER_FLAG_NEW_LISTENER);
4021 ASSERT_GE(listener, 0);
4022
4023 pid = fork();
4024 ASSERT_GE(pid, 0);
4025
4026 if (pid == 0)
4027 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC);
4028
4029
4030 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
4031
4032 ASSERT_EQ(prlimit(pid, RLIMIT_NOFILE, &lim, NULL), 0);
4033
4034 addfd.srcfd = memfd;
4035 addfd.newfd_flags = O_CLOEXEC;
4036 addfd.newfd = 0;
4037 addfd.id = req.id;
4038 addfd.flags = 0;
4039
4040 /* Should probably spot check /proc/sys/fs/file-nr */
4041 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4042 EXPECT_EQ(errno, EMFILE);
4043
4044 addfd.newfd = 100;
4045 addfd.flags = SECCOMP_ADDFD_FLAG_SETFD;
4046 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1);
4047 EXPECT_EQ(errno, EBADF);
4048
4049 resp.id = req.id;
4050 resp.error = 0;
4051 resp.val = USER_NOTIF_MAGIC;
4052
4053 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0);
4054
4055 /* Wait for child to finish. */
4056 EXPECT_EQ(waitpid(pid, &status, 0), pid);
4057 EXPECT_EQ(true, WIFEXITED(status));
4058 EXPECT_EQ(0, WEXITSTATUS(status));
4059
4060 close(memfd);
4061 }
4062
4063 /*
4064 * TODO:
4065 * - expand NNP testing
4066 * - better arch-specific TRACE and TRAP handlers.
4067 * - endianness checking when appropriate
4068 * - 64-bit arg prodding
4069 * - arch value testing (x86 modes especially)
4070 * - verify that FILTER_FLAG_LOG filters generate log messages
4071 * - verify that RET_LOG generates log messages
4072 */
4073
4074 TEST_HARNESS_MAIN
Ubuntu Jammy Linux kernel mirror