]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - tools/testing/selftests/seccomp/seccomp_bpf.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:


2e58549b2f0211ac11747d35b71c05145e81fbe7
[mirror_ubuntu-zesty-kernel.git] / tools / testing / selftests / seccomp / seccomp_bpf.c
1 /*
2 * Copyright (c) 2012 The Chromium OS Authors. All rights reserved.
3 * Use of this source code is governed by the GPLv2 license.
4 *
5 * Test code for seccomp bpf.
6 */
7
8 #include <sys/types.h>
9 #include <asm/siginfo.h>
10 #define __have_siginfo_t 1
11 #define __have_sigval_t 1
12 #define __have_sigevent_t 1
13
14 #include <errno.h>
15 #include <linux/filter.h>
16 #include <sys/prctl.h>
17 #include <sys/ptrace.h>
18 #include <sys/user.h>
19 #include <linux/prctl.h>
20 #include <linux/ptrace.h>
21 #include <linux/seccomp.h>
22 #include <pthread.h>
23 #include <semaphore.h>
24 #include <signal.h>
25 #include <stddef.h>
26 #include <stdbool.h>
27 #include <string.h>
28 #include <time.h>
29 #include <linux/elf.h>
30 #include <sys/uio.h>
31 #include <sys/utsname.h>
32 #include <sys/fcntl.h>
33 #include <sys/mman.h>
34 #include <sys/times.h>
35
36 #define _GNU_SOURCE
37 #include <unistd.h>
38 #include <sys/syscall.h>
39
40 #include "test_harness.h"
41
42 #ifndef PR_SET_PTRACER
43 # define PR_SET_PTRACER 0x59616d61
44 #endif
45
46 #ifndef PR_SET_NO_NEW_PRIVS
47 #define PR_SET_NO_NEW_PRIVS 38
48 #define PR_GET_NO_NEW_PRIVS 39
49 #endif
50
51 #ifndef PR_SECCOMP_EXT
52 #define PR_SECCOMP_EXT 43
53 #endif
54
55 #ifndef SECCOMP_EXT_ACT
56 #define SECCOMP_EXT_ACT 1
57 #endif
58
59 #ifndef SECCOMP_EXT_ACT_TSYNC
60 #define SECCOMP_EXT_ACT_TSYNC 1
61 #endif
62
63 #ifndef SECCOMP_MODE_STRICT
64 #define SECCOMP_MODE_STRICT 1
65 #endif
66
67 #ifndef SECCOMP_MODE_FILTER
68 #define SECCOMP_MODE_FILTER 2
69 #endif
70
71 #ifndef SECCOMP_RET_KILL
72 #define SECCOMP_RET_KILL 0x00000000U /* kill the task immediately */
73 #define SECCOMP_RET_TRAP 0x00030000U /* disallow and force a SIGSYS */
74 #define SECCOMP_RET_ERRNO 0x00050000U /* returns an errno */
75 #define SECCOMP_RET_TRACE 0x7ff00000U /* pass to a tracer or disallow */
76 #define SECCOMP_RET_ALLOW 0x7fff0000U /* allow */
77
78 /* Masks for the return value sections. */
79 #define SECCOMP_RET_ACTION 0x7fff0000U
80 #define SECCOMP_RET_DATA 0x0000ffffU
81
82 struct seccomp_data {
83 int nr;
84 __u32 arch;
85 __u64 instruction_pointer;
86 __u64 args[6];
87 };
88 #endif
89
90 #if __BYTE_ORDER == __LITTLE_ENDIAN
91 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]))
92 #elif __BYTE_ORDER == __BIG_ENDIAN
93 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32))
94 #else
95 #error "wut? Unknown __BYTE_ORDER?!"
96 #endif
97
98 #define SIBLING_EXIT_UNKILLED 0xbadbeef
99 #define SIBLING_EXIT_FAILURE 0xbadface
100 #define SIBLING_EXIT_NEWPRIVS 0xbadfeed
101
102 TEST(mode_strict_support)
103 {
104 long ret;
105
106 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
107 ASSERT_EQ(0, ret) {
108 TH_LOG("Kernel does not support CONFIG_SECCOMP");
109 }
110 syscall(__NR_exit, 1);
111 }
112
113 TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL)
114 {
115 long ret;
116
117 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
118 ASSERT_EQ(0, ret) {
119 TH_LOG("Kernel does not support CONFIG_SECCOMP");
120 }
121 syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
122 NULL, NULL, NULL);
123 EXPECT_FALSE(true) {
124 TH_LOG("Unreachable!");
125 }
126 }
127
128 /* Note! This doesn't test no new privs behavior */
129 TEST(no_new_privs_support)
130 {
131 long ret;
132
133 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
134 EXPECT_EQ(0, ret) {
135 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
136 }
137 }
138
139 /* Tests kernel support by checking for a copy_from_user() fault on * NULL. */
140 TEST(mode_filter_support)
141 {
142 long ret;
143
144 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
145 ASSERT_EQ(0, ret) {
146 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
147 }
148 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL);
149 EXPECT_EQ(-1, ret);
150 EXPECT_EQ(EFAULT, errno) {
151 TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!");
152 }
153 }
154
155 TEST(mode_filter_without_nnp)
156 {
157 struct sock_filter filter[] = {
158 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
159 };
160 struct sock_fprog prog = {
161 .len = (unsigned short)ARRAY_SIZE(filter),
162 .filter = filter,
163 };
164 long ret;
165
166 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0);
167 ASSERT_LE(0, ret) {
168 TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS");
169 }
170 errno = 0;
171 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
172 /* Succeeds with CAP_SYS_ADMIN, fails without */
173 /* TODO(wad) check caps not euid */
174 if (geteuid()) {
175 EXPECT_EQ(-1, ret);
176 EXPECT_EQ(EACCES, errno);
177 } else {
178 EXPECT_EQ(0, ret);
179 }
180 }
181
182 #define MAX_INSNS_PER_PATH 32768
183
184 TEST(filter_size_limits)
185 {
186 int i;
187 int count = BPF_MAXINSNS + 1;
188 struct sock_filter allow[] = {
189 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
190 };
191 struct sock_filter *filter;
192 struct sock_fprog prog = { };
193 long ret;
194
195 filter = calloc(count, sizeof(*filter));
196 ASSERT_NE(NULL, filter);
197
198 for (i = 0; i < count; i++)
199 filter[i] = allow[0];
200
201 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
202 ASSERT_EQ(0, ret);
203
204 prog.filter = filter;
205 prog.len = count;
206
207 /* Too many filter instructions in a single filter. */
208 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
209 ASSERT_NE(0, ret) {
210 TH_LOG("Installing %d insn filter was allowed", prog.len);
211 }
212
213 /* One less is okay, though. */
214 prog.len -= 1;
215 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
216 ASSERT_EQ(0, ret) {
217 TH_LOG("Installing %d insn filter wasn't allowed", prog.len);
218 }
219 }
220
221 TEST(filter_chain_limits)
222 {
223 int i;
224 int count = BPF_MAXINSNS;
225 struct sock_filter allow[] = {
226 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
227 };
228 struct sock_filter *filter;
229 struct sock_fprog prog = { };
230 long ret;
231
232 filter = calloc(count, sizeof(*filter));
233 ASSERT_NE(NULL, filter);
234
235 for (i = 0; i < count; i++)
236 filter[i] = allow[0];
237
238 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
239 ASSERT_EQ(0, ret);
240
241 prog.filter = filter;
242 prog.len = 1;
243
244 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
245 ASSERT_EQ(0, ret);
246
247 prog.len = count;
248
249 /* Too many total filter instructions. */
250 for (i = 0; i < MAX_INSNS_PER_PATH; i++) {
251 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
252 if (ret != 0)
253 break;
254 }
255 ASSERT_NE(0, ret) {
256 TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)",
257 i, count, i * (count + 4));
258 }
259 }
260
261 TEST(mode_filter_cannot_move_to_strict)
262 {
263 struct sock_filter filter[] = {
264 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
265 };
266 struct sock_fprog prog = {
267 .len = (unsigned short)ARRAY_SIZE(filter),
268 .filter = filter,
269 };
270 long ret;
271
272 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
273 ASSERT_EQ(0, ret);
274
275 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
276 ASSERT_EQ(0, ret);
277
278 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0);
279 EXPECT_EQ(-1, ret);
280 EXPECT_EQ(EINVAL, errno);
281 }
282
283
284 TEST(mode_filter_get_seccomp)
285 {
286 struct sock_filter filter[] = {
287 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
288 };
289 struct sock_fprog prog = {
290 .len = (unsigned short)ARRAY_SIZE(filter),
291 .filter = filter,
292 };
293 long ret;
294
295 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
296 ASSERT_EQ(0, ret);
297
298 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
299 EXPECT_EQ(0, ret);
300
301 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
302 ASSERT_EQ(0, ret);
303
304 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
305 EXPECT_EQ(2, ret);
306 }
307
308
309 TEST(ALLOW_all)
310 {
311 struct sock_filter filter[] = {
312 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
313 };
314 struct sock_fprog prog = {
315 .len = (unsigned short)ARRAY_SIZE(filter),
316 .filter = filter,
317 };
318 long ret;
319
320 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
321 ASSERT_EQ(0, ret);
322
323 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
324 ASSERT_EQ(0, ret);
325 }
326
327 TEST(empty_prog)
328 {
329 struct sock_filter filter[] = {
330 };
331 struct sock_fprog prog = {
332 .len = (unsigned short)ARRAY_SIZE(filter),
333 .filter = filter,
334 };
335 long ret;
336
337 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
338 ASSERT_EQ(0, ret);
339
340 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
341 EXPECT_EQ(-1, ret);
342 EXPECT_EQ(EINVAL, errno);
343 }
344
345 TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS)
346 {
347 struct sock_filter filter[] = {
348 BPF_STMT(BPF_RET|BPF_K, 0x10000000U),
349 };
350 struct sock_fprog prog = {
351 .len = (unsigned short)ARRAY_SIZE(filter),
352 .filter = filter,
353 };
354 long ret;
355
356 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
357 ASSERT_EQ(0, ret);
358
359 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
360 ASSERT_EQ(0, ret);
361 EXPECT_EQ(0, syscall(__NR_getpid)) {
362 TH_LOG("getpid() shouldn't ever return");
363 }
364 }
365
366 /* return code >= 0x80000000 is unused. */
367 TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS)
368 {
369 struct sock_filter filter[] = {
370 BPF_STMT(BPF_RET|BPF_K, 0x90000000U),
371 };
372 struct sock_fprog prog = {
373 .len = (unsigned short)ARRAY_SIZE(filter),
374 .filter = filter,
375 };
376 long ret;
377
378 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
379 ASSERT_EQ(0, ret);
380
381 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
382 ASSERT_EQ(0, ret);
383 EXPECT_EQ(0, syscall(__NR_getpid)) {
384 TH_LOG("getpid() shouldn't ever return");
385 }
386 }
387
388 TEST_SIGNAL(KILL_all, SIGSYS)
389 {
390 struct sock_filter filter[] = {
391 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
392 };
393 struct sock_fprog prog = {
394 .len = (unsigned short)ARRAY_SIZE(filter),
395 .filter = filter,
396 };
397 long ret;
398
399 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
400 ASSERT_EQ(0, ret);
401
402 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
403 ASSERT_EQ(0, ret);
404 }
405
406 TEST_SIGNAL(KILL_one, SIGSYS)
407 {
408 struct sock_filter filter[] = {
409 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
410 offsetof(struct seccomp_data, nr)),
411 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
412 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
413 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
414 };
415 struct sock_fprog prog = {
416 .len = (unsigned short)ARRAY_SIZE(filter),
417 .filter = filter,
418 };
419 long ret;
420 pid_t parent = getppid();
421
422 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
423 ASSERT_EQ(0, ret);
424
425 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
426 ASSERT_EQ(0, ret);
427
428 EXPECT_EQ(parent, syscall(__NR_getppid));
429 /* getpid() should never return. */
430 EXPECT_EQ(0, syscall(__NR_getpid));
431 }
432
433 TEST_SIGNAL(KILL_one_arg_one, SIGSYS)
434 {
435 void *fatal_address;
436 struct sock_filter filter[] = {
437 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
438 offsetof(struct seccomp_data, nr)),
439 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0),
440 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
441 /* Only both with lower 32-bit for now. */
442 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)),
443 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K,
444 (unsigned long)&fatal_address, 0, 1),
445 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
446 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
447 };
448 struct sock_fprog prog = {
449 .len = (unsigned short)ARRAY_SIZE(filter),
450 .filter = filter,
451 };
452 long ret;
453 pid_t parent = getppid();
454 struct tms timebuf;
455 clock_t clock = times(&timebuf);
456
457 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
458 ASSERT_EQ(0, ret);
459
460 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
461 ASSERT_EQ(0, ret);
462
463 EXPECT_EQ(parent, syscall(__NR_getppid));
464 EXPECT_LE(clock, syscall(__NR_times, &timebuf));
465 /* times() should never return. */
466 EXPECT_EQ(0, syscall(__NR_times, &fatal_address));
467 }
468
469 TEST_SIGNAL(KILL_one_arg_six, SIGSYS)
470 {
471 #ifndef __NR_mmap2
472 int sysno = __NR_mmap;
473 #else
474 int sysno = __NR_mmap2;
475 #endif
476 struct sock_filter filter[] = {
477 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
478 offsetof(struct seccomp_data, nr)),
479 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0),
480 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
481 /* Only both with lower 32-bit for now. */
482 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)),
483 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1),
484 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
485 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
486 };
487 struct sock_fprog prog = {
488 .len = (unsigned short)ARRAY_SIZE(filter),
489 .filter = filter,
490 };
491 long ret;
492 pid_t parent = getppid();
493 int fd;
494 void *map1, *map2;
495 int page_size = sysconf(_SC_PAGESIZE);
496
497 ASSERT_LT(0, page_size);
498
499 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
500 ASSERT_EQ(0, ret);
501
502 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
503 ASSERT_EQ(0, ret);
504
505 fd = open("/dev/zero", O_RDONLY);
506 ASSERT_NE(-1, fd);
507
508 EXPECT_EQ(parent, syscall(__NR_getppid));
509 map1 = (void *)syscall(sysno,
510 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size);
511 EXPECT_NE(MAP_FAILED, map1);
512 /* mmap2() should never return. */
513 map2 = (void *)syscall(sysno,
514 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE);
515 EXPECT_EQ(MAP_FAILED, map2);
516
517 /* The test failed, so clean up the resources. */
518 munmap(map1, page_size);
519 munmap(map2, page_size);
520 close(fd);
521 }
522
523 /* TODO(wad) add 64-bit versus 32-bit arg tests. */
524 TEST(arg_out_of_range)
525 {
526 struct sock_filter filter[] = {
527 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)),
528 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
529 };
530 struct sock_fprog prog = {
531 .len = (unsigned short)ARRAY_SIZE(filter),
532 .filter = filter,
533 };
534 long ret;
535
536 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
537 ASSERT_EQ(0, ret);
538
539 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
540 EXPECT_EQ(-1, ret);
541 EXPECT_EQ(EINVAL, errno);
542 }
543
544 TEST(ERRNO_valid)
545 {
546 struct sock_filter filter[] = {
547 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
548 offsetof(struct seccomp_data, nr)),
549 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
550 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | E2BIG),
551 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
552 };
553 struct sock_fprog prog = {
554 .len = (unsigned short)ARRAY_SIZE(filter),
555 .filter = filter,
556 };
557 long ret;
558 pid_t parent = getppid();
559
560 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
561 ASSERT_EQ(0, ret);
562
563 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
564 ASSERT_EQ(0, ret);
565
566 EXPECT_EQ(parent, syscall(__NR_getppid));
567 EXPECT_EQ(-1, read(0, NULL, 0));
568 EXPECT_EQ(E2BIG, errno);
569 }
570
571 TEST(ERRNO_zero)
572 {
573 struct sock_filter filter[] = {
574 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
575 offsetof(struct seccomp_data, nr)),
576 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
577 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | 0),
578 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
579 };
580 struct sock_fprog prog = {
581 .len = (unsigned short)ARRAY_SIZE(filter),
582 .filter = filter,
583 };
584 long ret;
585 pid_t parent = getppid();
586
587 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
588 ASSERT_EQ(0, ret);
589
590 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
591 ASSERT_EQ(0, ret);
592
593 EXPECT_EQ(parent, syscall(__NR_getppid));
594 /* "errno" of 0 is ok. */
595 EXPECT_EQ(0, read(0, NULL, 0));
596 }
597
598 TEST(ERRNO_capped)
599 {
600 struct sock_filter filter[] = {
601 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
602 offsetof(struct seccomp_data, nr)),
603 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
604 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | 4096),
605 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
606 };
607 struct sock_fprog prog = {
608 .len = (unsigned short)ARRAY_SIZE(filter),
609 .filter = filter,
610 };
611 long ret;
612 pid_t parent = getppid();
613
614 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
615 ASSERT_EQ(0, ret);
616
617 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
618 ASSERT_EQ(0, ret);
619
620 EXPECT_EQ(parent, syscall(__NR_getppid));
621 EXPECT_EQ(-1, read(0, NULL, 0));
622 EXPECT_EQ(4095, errno);
623 }
624
625 FIXTURE_DATA(TRAP) {
626 struct sock_fprog prog;
627 };
628
629 FIXTURE_SETUP(TRAP)
630 {
631 struct sock_filter filter[] = {
632 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
633 offsetof(struct seccomp_data, nr)),
634 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
635 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
636 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
637 };
638
639 memset(&self->prog, 0, sizeof(self->prog));
640 self->prog.filter = malloc(sizeof(filter));
641 ASSERT_NE(NULL, self->prog.filter);
642 memcpy(self->prog.filter, filter, sizeof(filter));
643 self->prog.len = (unsigned short)ARRAY_SIZE(filter);
644 }
645
646 FIXTURE_TEARDOWN(TRAP)
647 {
648 if (self->prog.filter)
649 free(self->prog.filter);
650 }
651
652 TEST_F_SIGNAL(TRAP, dfl, SIGSYS)
653 {
654 long ret;
655
656 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
657 ASSERT_EQ(0, ret);
658
659 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
660 ASSERT_EQ(0, ret);
661 syscall(__NR_getpid);
662 }
663
664 /* Ensure that SIGSYS overrides SIG_IGN */
665 TEST_F_SIGNAL(TRAP, ign, SIGSYS)
666 {
667 long ret;
668
669 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
670 ASSERT_EQ(0, ret);
671
672 signal(SIGSYS, SIG_IGN);
673
674 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
675 ASSERT_EQ(0, ret);
676 syscall(__NR_getpid);
677 }
678
679 static struct siginfo TRAP_info;
680 static volatile int TRAP_nr;
681 static void TRAP_action(int nr, siginfo_t *info, void *void_context)
682 {
683 memcpy(&TRAP_info, info, sizeof(TRAP_info));
684 TRAP_nr = nr;
685 }
686
687 TEST_F(TRAP, handler)
688 {
689 int ret, test;
690 struct sigaction act;
691 sigset_t mask;
692
693 memset(&act, 0, sizeof(act));
694 sigemptyset(&mask);
695 sigaddset(&mask, SIGSYS);
696
697 act.sa_sigaction = &TRAP_action;
698 act.sa_flags = SA_SIGINFO;
699 ret = sigaction(SIGSYS, &act, NULL);
700 ASSERT_EQ(0, ret) {
701 TH_LOG("sigaction failed");
702 }
703 ret = sigprocmask(SIG_UNBLOCK, &mask, NULL);
704 ASSERT_EQ(0, ret) {
705 TH_LOG("sigprocmask failed");
706 }
707
708 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
709 ASSERT_EQ(0, ret);
710 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
711 ASSERT_EQ(0, ret);
712 TRAP_nr = 0;
713 memset(&TRAP_info, 0, sizeof(TRAP_info));
714 /* Expect the registers to be rolled back. (nr = error) may vary
715 * based on arch. */
716 ret = syscall(__NR_getpid);
717 /* Silence gcc warning about volatile. */
718 test = TRAP_nr;
719 EXPECT_EQ(SIGSYS, test);
720 struct local_sigsys {
721 void *_call_addr; /* calling user insn */
722 int _syscall; /* triggering system call number */
723 unsigned int _arch; /* AUDIT_ARCH_* of syscall */
724 } *sigsys = (struct local_sigsys *)
725 #ifdef si_syscall
726 &(TRAP_info.si_call_addr);
727 #else
728 &TRAP_info.si_pid;
729 #endif
730 EXPECT_EQ(__NR_getpid, sigsys->_syscall);
731 /* Make sure arch is non-zero. */
732 EXPECT_NE(0, sigsys->_arch);
733 EXPECT_NE(0, (unsigned long)sigsys->_call_addr);
734 }
735
736 FIXTURE_DATA(precedence) {
737 struct sock_fprog allow;
738 struct sock_fprog trace;
739 struct sock_fprog error;
740 struct sock_fprog trap;
741 struct sock_fprog kill;
742 };
743
744 FIXTURE_SETUP(precedence)
745 {
746 struct sock_filter allow_insns[] = {
747 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
748 };
749 struct sock_filter trace_insns[] = {
750 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
751 offsetof(struct seccomp_data, nr)),
752 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
753 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
754 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE),
755 };
756 struct sock_filter error_insns[] = {
757 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
758 offsetof(struct seccomp_data, nr)),
759 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
760 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
761 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO),
762 };
763 struct sock_filter trap_insns[] = {
764 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
765 offsetof(struct seccomp_data, nr)),
766 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
767 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
768 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
769 };
770 struct sock_filter kill_insns[] = {
771 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
772 offsetof(struct seccomp_data, nr)),
773 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
774 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
775 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
776 };
777
778 memset(self, 0, sizeof(*self));
779 #define FILTER_ALLOC(_x) \
780 self->_x.filter = malloc(sizeof(_x##_insns)); \
781 ASSERT_NE(NULL, self->_x.filter); \
782 memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \
783 self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns)
784 FILTER_ALLOC(allow);
785 FILTER_ALLOC(trace);
786 FILTER_ALLOC(error);
787 FILTER_ALLOC(trap);
788 FILTER_ALLOC(kill);
789 }
790
791 FIXTURE_TEARDOWN(precedence)
792 {
793 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter)
794 FILTER_FREE(allow);
795 FILTER_FREE(trace);
796 FILTER_FREE(error);
797 FILTER_FREE(trap);
798 FILTER_FREE(kill);
799 }
800
801 TEST_F(precedence, allow_ok)
802 {
803 pid_t parent, res = 0;
804 long ret;
805
806 parent = getppid();
807 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
808 ASSERT_EQ(0, ret);
809
810 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
811 ASSERT_EQ(0, ret);
812 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
813 ASSERT_EQ(0, ret);
814 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
815 ASSERT_EQ(0, ret);
816 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
817 ASSERT_EQ(0, ret);
818 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
819 ASSERT_EQ(0, ret);
820 /* Should work just fine. */
821 res = syscall(__NR_getppid);
822 EXPECT_EQ(parent, res);
823 }
824
825 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS)
826 {
827 pid_t parent, res = 0;
828 long ret;
829
830 parent = getppid();
831 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
832 ASSERT_EQ(0, ret);
833
834 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
835 ASSERT_EQ(0, ret);
836 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
837 ASSERT_EQ(0, ret);
838 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
839 ASSERT_EQ(0, ret);
840 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
841 ASSERT_EQ(0, ret);
842 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
843 ASSERT_EQ(0, ret);
844 /* Should work just fine. */
845 res = syscall(__NR_getppid);
846 EXPECT_EQ(parent, res);
847 /* getpid() should never return. */
848 res = syscall(__NR_getpid);
849 EXPECT_EQ(0, res);
850 }
851
852 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS)
853 {
854 pid_t parent;
855 long ret;
856
857 parent = getppid();
858 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
859 ASSERT_EQ(0, ret);
860
861 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
862 ASSERT_EQ(0, ret);
863 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
864 ASSERT_EQ(0, ret);
865 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
866 ASSERT_EQ(0, ret);
867 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
868 ASSERT_EQ(0, ret);
869 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
870 ASSERT_EQ(0, ret);
871 /* Should work just fine. */
872 EXPECT_EQ(parent, syscall(__NR_getppid));
873 /* getpid() should never return. */
874 EXPECT_EQ(0, syscall(__NR_getpid));
875 }
876
877 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS)
878 {
879 pid_t parent;
880 long ret;
881
882 parent = getppid();
883 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
884 ASSERT_EQ(0, ret);
885
886 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
887 ASSERT_EQ(0, ret);
888 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
889 ASSERT_EQ(0, ret);
890 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
891 ASSERT_EQ(0, ret);
892 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
893 ASSERT_EQ(0, ret);
894 /* Should work just fine. */
895 EXPECT_EQ(parent, syscall(__NR_getppid));
896 /* getpid() should never return. */
897 EXPECT_EQ(0, syscall(__NR_getpid));
898 }
899
900 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS)
901 {
902 pid_t parent;
903 long ret;
904
905 parent = getppid();
906 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
907 ASSERT_EQ(0, ret);
908
909 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
910 ASSERT_EQ(0, ret);
911 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
912 ASSERT_EQ(0, ret);
913 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
914 ASSERT_EQ(0, ret);
915 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
916 ASSERT_EQ(0, ret);
917 /* Should work just fine. */
918 EXPECT_EQ(parent, syscall(__NR_getppid));
919 /* getpid() should never return. */
920 EXPECT_EQ(0, syscall(__NR_getpid));
921 }
922
923 TEST_F(precedence, errno_is_third)
924 {
925 pid_t parent;
926 long ret;
927
928 parent = getppid();
929 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
930 ASSERT_EQ(0, ret);
931
932 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
933 ASSERT_EQ(0, ret);
934 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
935 ASSERT_EQ(0, ret);
936 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
937 ASSERT_EQ(0, ret);
938 /* Should work just fine. */
939 EXPECT_EQ(parent, syscall(__NR_getppid));
940 EXPECT_EQ(0, syscall(__NR_getpid));
941 }
942
943 TEST_F(precedence, errno_is_third_in_any_order)
944 {
945 pid_t parent;
946 long ret;
947
948 parent = getppid();
949 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
950 ASSERT_EQ(0, ret);
951
952 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
953 ASSERT_EQ(0, ret);
954 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
955 ASSERT_EQ(0, ret);
956 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
957 ASSERT_EQ(0, ret);
958 /* Should work just fine. */
959 EXPECT_EQ(parent, syscall(__NR_getppid));
960 EXPECT_EQ(0, syscall(__NR_getpid));
961 }
962
963 TEST_F(precedence, trace_is_fourth)
964 {
965 pid_t parent;
966 long ret;
967
968 parent = getppid();
969 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
970 ASSERT_EQ(0, ret);
971
972 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
973 ASSERT_EQ(0, ret);
974 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
975 ASSERT_EQ(0, ret);
976 /* Should work just fine. */
977 EXPECT_EQ(parent, syscall(__NR_getppid));
978 /* No ptracer */
979 EXPECT_EQ(-1, syscall(__NR_getpid));
980 }
981
982 TEST_F(precedence, trace_is_fourth_in_any_order)
983 {
984 pid_t parent;
985 long ret;
986
987 parent = getppid();
988 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
989 ASSERT_EQ(0, ret);
990
991 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
992 ASSERT_EQ(0, ret);
993 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
994 ASSERT_EQ(0, ret);
995 /* Should work just fine. */
996 EXPECT_EQ(parent, syscall(__NR_getppid));
997 /* No ptracer */
998 EXPECT_EQ(-1, syscall(__NR_getpid));
999 }
1000
1001 #ifndef PTRACE_O_TRACESECCOMP
1002 #define PTRACE_O_TRACESECCOMP 0x00000080
1003 #endif
1004
1005 /* Catch the Ubuntu 12.04 value error. */
1006 #if PTRACE_EVENT_SECCOMP != 7
1007 #undef PTRACE_EVENT_SECCOMP
1008 #endif
1009
1010 #ifndef PTRACE_EVENT_SECCOMP
1011 #define PTRACE_EVENT_SECCOMP 7
1012 #endif
1013
1014 #define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP)
1015 bool tracer_running;
1016 void tracer_stop(int sig)
1017 {
1018 tracer_running = false;
1019 }
1020
1021 typedef void tracer_func_t(struct __test_metadata *_metadata,
1022 pid_t tracee, int status, void *args);
1023
1024 void tracer(struct __test_metadata *_metadata, int fd, pid_t tracee,
1025 tracer_func_t tracer_func, void *args)
1026 {
1027 int ret = -1;
1028 struct sigaction action = {
1029 .sa_handler = tracer_stop,
1030 };
1031
1032 /* Allow external shutdown. */
1033 tracer_running = true;
1034 ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL));
1035
1036 errno = 0;
1037 while (ret == -1 && errno != EINVAL)
1038 ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0);
1039 ASSERT_EQ(0, ret) {
1040 kill(tracee, SIGKILL);
1041 }
1042 /* Wait for attach stop */
1043 wait(NULL);
1044
1045 ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, PTRACE_O_TRACESECCOMP);
1046 ASSERT_EQ(0, ret) {
1047 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP");
1048 kill(tracee, SIGKILL);
1049 }
1050 ptrace(PTRACE_CONT, tracee, NULL, 0);
1051
1052 /* Unblock the tracee */
1053 ASSERT_EQ(1, write(fd, "A", 1));
1054 ASSERT_EQ(0, close(fd));
1055
1056 /* Run until we're shut down. Must assert to stop execution. */
1057 while (tracer_running) {
1058 int status;
1059
1060 if (wait(&status) != tracee)
1061 continue;
1062 if (WIFSIGNALED(status) || WIFEXITED(status))
1063 /* Child is dead. Time to go. */
1064 return;
1065
1066 /* Make sure this is a seccomp event. */
1067 ASSERT_EQ(true, IS_SECCOMP_EVENT(status));
1068
1069 tracer_func(_metadata, tracee, status, args);
1070
1071 ret = ptrace(PTRACE_CONT, tracee, NULL, NULL);
1072 ASSERT_EQ(0, ret);
1073 }
1074 /* Directly report the status of our test harness results. */
1075 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS : EXIT_FAILURE);
1076 }
1077
1078 /* Common tracer setup/teardown functions. */
1079 void cont_handler(int num)
1080 { }
1081 pid_t setup_trace_fixture(struct __test_metadata *_metadata,
1082 tracer_func_t func, void *args)
1083 {
1084 char sync;
1085 int pipefd[2];
1086 pid_t tracer_pid;
1087 pid_t tracee = getpid();
1088
1089 /* Setup a pipe for clean synchronization. */
1090 ASSERT_EQ(0, pipe(pipefd));
1091
1092 /* Fork a child which we'll promote to tracer */
1093 tracer_pid = fork();
1094 ASSERT_LE(0, tracer_pid);
1095 signal(SIGALRM, cont_handler);
1096 if (tracer_pid == 0) {
1097 close(pipefd[0]);
1098 tracer(_metadata, pipefd[1], tracee, func, args);
1099 syscall(__NR_exit, 0);
1100 }
1101 close(pipefd[1]);
1102 prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0);
1103 read(pipefd[0], &sync, 1);
1104 close(pipefd[0]);
1105
1106 return tracer_pid;
1107 }
1108 void teardown_trace_fixture(struct __test_metadata *_metadata,
1109 pid_t tracer)
1110 {
1111 if (tracer) {
1112 int status;
1113 /*
1114 * Extract the exit code from the other process and
1115 * adopt it for ourselves in case its asserts failed.
1116 */
1117 ASSERT_EQ(0, kill(tracer, SIGUSR1));
1118 ASSERT_EQ(tracer, waitpid(tracer, &status, 0));
1119 if (WEXITSTATUS(status))
1120 _metadata->passed = 0;
1121 }
1122 }
1123
1124 /* "poke" tracer arguments and function. */
1125 struct tracer_args_poke_t {
1126 unsigned long poke_addr;
1127 };
1128
1129 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status,
1130 void *args)
1131 {
1132 int ret;
1133 unsigned long msg;
1134 struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args;
1135
1136 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1137 EXPECT_EQ(0, ret);
1138 /* If this fails, don't try to recover. */
1139 ASSERT_EQ(0x1001, msg) {
1140 kill(tracee, SIGKILL);
1141 }
1142 /*
1143 * Poke in the message.
1144 * Registers are not touched to try to keep this relatively arch
1145 * agnostic.
1146 */
1147 ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001);
1148 EXPECT_EQ(0, ret);
1149 }
1150
1151 FIXTURE_DATA(TRACE_poke) {
1152 struct sock_fprog prog;
1153 pid_t tracer;
1154 long poked;
1155 struct tracer_args_poke_t tracer_args;
1156 };
1157
1158 FIXTURE_SETUP(TRACE_poke)
1159 {
1160 struct sock_filter filter[] = {
1161 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1162 offsetof(struct seccomp_data, nr)),
1163 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1164 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001),
1165 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1166 };
1167
1168 self->poked = 0;
1169 memset(&self->prog, 0, sizeof(self->prog));
1170 self->prog.filter = malloc(sizeof(filter));
1171 ASSERT_NE(NULL, self->prog.filter);
1172 memcpy(self->prog.filter, filter, sizeof(filter));
1173 self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1174
1175 /* Set up tracer args. */
1176 self->tracer_args.poke_addr = (unsigned long)&self->poked;
1177
1178 /* Launch tracer. */
1179 self->tracer = setup_trace_fixture(_metadata, tracer_poke,
1180 &self->tracer_args);
1181 }
1182
1183 FIXTURE_TEARDOWN(TRACE_poke)
1184 {
1185 teardown_trace_fixture(_metadata, self->tracer);
1186 if (self->prog.filter)
1187 free(self->prog.filter);
1188 }
1189
1190 TEST_F(TRACE_poke, read_has_side_effects)
1191 {
1192 ssize_t ret;
1193
1194 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1195 ASSERT_EQ(0, ret);
1196
1197 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1198 ASSERT_EQ(0, ret);
1199
1200 EXPECT_EQ(0, self->poked);
1201 ret = read(-1, NULL, 0);
1202 EXPECT_EQ(-1, ret);
1203 EXPECT_EQ(0x1001, self->poked);
1204 }
1205
1206 TEST_F(TRACE_poke, getpid_runs_normally)
1207 {
1208 long ret;
1209
1210 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1211 ASSERT_EQ(0, ret);
1212
1213 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1214 ASSERT_EQ(0, ret);
1215
1216 EXPECT_EQ(0, self->poked);
1217 EXPECT_NE(0, syscall(__NR_getpid));
1218 EXPECT_EQ(0, self->poked);
1219 }
1220
1221 #if defined(__x86_64__)
1222 # define ARCH_REGS struct user_regs_struct
1223 # define SYSCALL_NUM orig_rax
1224 # define SYSCALL_RET rax
1225 #elif defined(__i386__)
1226 # define ARCH_REGS struct user_regs_struct
1227 # define SYSCALL_NUM orig_eax
1228 # define SYSCALL_RET eax
1229 #elif defined(__arm__)
1230 # define ARCH_REGS struct pt_regs
1231 # define SYSCALL_NUM ARM_r7
1232 # define SYSCALL_RET ARM_r0
1233 #elif defined(__aarch64__)
1234 # define ARCH_REGS struct user_pt_regs
1235 # define SYSCALL_NUM regs[8]
1236 # define SYSCALL_RET regs[0]
1237 #elif defined(__hppa__)
1238 # define ARCH_REGS struct user_regs_struct
1239 # define SYSCALL_NUM gr[20]
1240 # define SYSCALL_RET gr[28]
1241 #elif defined(__powerpc__)
1242 # define ARCH_REGS struct pt_regs
1243 # define SYSCALL_NUM gpr[0]
1244 # define SYSCALL_RET gpr[3]
1245 #elif defined(__s390__)
1246 # define ARCH_REGS s390_regs
1247 # define SYSCALL_NUM gprs[2]
1248 # define SYSCALL_RET gprs[2]
1249 #elif defined(__mips__)
1250 # define ARCH_REGS struct pt_regs
1251 # define SYSCALL_NUM regs[2]
1252 # define SYSCALL_SYSCALL_NUM regs[4]
1253 # define SYSCALL_RET regs[2]
1254 # define SYSCALL_NUM_RET_SHARE_REG
1255 #else
1256 # error "Do not know how to find your architecture's registers and syscalls"
1257 #endif
1258
1259 /* Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for
1260 * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux).
1261 */
1262 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__)
1263 #define HAVE_GETREGS
1264 #endif
1265
1266 /* Architecture-specific syscall fetching routine. */
1267 int get_syscall(struct __test_metadata *_metadata, pid_t tracee)
1268 {
1269 ARCH_REGS regs;
1270 #ifdef HAVE_GETREGS
1271 EXPECT_EQ(0, ptrace(PTRACE_GETREGS, tracee, 0, &regs)) {
1272 TH_LOG("PTRACE_GETREGS failed");
1273 return -1;
1274 }
1275 #else
1276 struct iovec iov;
1277
1278 iov.iov_base = &regs;
1279 iov.iov_len = sizeof(regs);
1280 EXPECT_EQ(0, ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov)) {
1281 TH_LOG("PTRACE_GETREGSET failed");
1282 return -1;
1283 }
1284 #endif
1285
1286 #if defined(__mips__)
1287 if (regs.SYSCALL_NUM == __NR_O32_Linux)
1288 return regs.SYSCALL_SYSCALL_NUM;
1289 #endif
1290 return regs.SYSCALL_NUM;
1291 }
1292
1293 /* Architecture-specific syscall changing routine. */
1294 void change_syscall(struct __test_metadata *_metadata,
1295 pid_t tracee, int syscall)
1296 {
1297 int ret;
1298 ARCH_REGS regs;
1299 #ifdef HAVE_GETREGS
1300 ret = ptrace(PTRACE_GETREGS, tracee, 0, &regs);
1301 #else
1302 struct iovec iov;
1303 iov.iov_base = &regs;
1304 iov.iov_len = sizeof(regs);
1305 ret = ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov);
1306 #endif
1307 EXPECT_EQ(0, ret);
1308
1309 #if defined(__x86_64__) || defined(__i386__) || defined(__powerpc__) || \
1310 defined(__s390__) || defined(__hppa__)
1311 {
1312 regs.SYSCALL_NUM = syscall;
1313 }
1314 #elif defined(__mips__)
1315 {
1316 if (regs.SYSCALL_NUM == __NR_O32_Linux)
1317 regs.SYSCALL_SYSCALL_NUM = syscall;
1318 else
1319 regs.SYSCALL_NUM = syscall;
1320 }
1321
1322 #elif defined(__arm__)
1323 # ifndef PTRACE_SET_SYSCALL
1324 # define PTRACE_SET_SYSCALL 23
1325 # endif
1326 {
1327 ret = ptrace(PTRACE_SET_SYSCALL, tracee, NULL, syscall);
1328 EXPECT_EQ(0, ret);
1329 }
1330
1331 #elif defined(__aarch64__)
1332 # ifndef NT_ARM_SYSTEM_CALL
1333 # define NT_ARM_SYSTEM_CALL 0x404
1334 # endif
1335 {
1336 iov.iov_base = &syscall;
1337 iov.iov_len = sizeof(syscall);
1338 ret = ptrace(PTRACE_SETREGSET, tracee, NT_ARM_SYSTEM_CALL,
1339 &iov);
1340 EXPECT_EQ(0, ret);
1341 }
1342
1343 #else
1344 ASSERT_EQ(1, 0) {
1345 TH_LOG("How is the syscall changed on this architecture?");
1346 }
1347 #endif
1348
1349 /* If syscall is skipped, change return value. */
1350 if (syscall == -1)
1351 #ifdef SYSCALL_NUM_RET_SHARE_REG
1352 TH_LOG("Can't modify syscall return on this architecture");
1353 #else
1354 regs.SYSCALL_RET = 1;
1355 #endif
1356
1357 #ifdef HAVE_GETREGS
1358 ret = ptrace(PTRACE_SETREGS, tracee, 0, &regs);
1359 #else
1360 iov.iov_base = &regs;
1361 iov.iov_len = sizeof(regs);
1362 ret = ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &iov);
1363 #endif
1364 EXPECT_EQ(0, ret);
1365 }
1366
1367 void tracer_syscall(struct __test_metadata *_metadata, pid_t tracee,
1368 int status, void *args)
1369 {
1370 int ret;
1371 unsigned long msg;
1372
1373 /* Make sure we got the right message. */
1374 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1375 EXPECT_EQ(0, ret);
1376
1377 /* Validate and take action on expected syscalls. */
1378 switch (msg) {
1379 case 0x1002:
1380 /* change getpid to getppid. */
1381 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee));
1382 change_syscall(_metadata, tracee, __NR_getppid);
1383 break;
1384 case 0x1003:
1385 /* skip gettid. */
1386 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee));
1387 change_syscall(_metadata, tracee, -1);
1388 break;
1389 case 0x1004:
1390 /* do nothing (allow getppid) */
1391 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee));
1392 break;
1393 default:
1394 EXPECT_EQ(0, msg) {
1395 TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg);
1396 kill(tracee, SIGKILL);
1397 }
1398 }
1399
1400 }
1401
1402 FIXTURE_DATA(TRACE_syscall) {
1403 struct sock_fprog prog;
1404 pid_t tracer, mytid, mypid, parent;
1405 };
1406
1407 FIXTURE_SETUP(TRACE_syscall)
1408 {
1409 struct sock_filter filter[] = {
1410 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1411 offsetof(struct seccomp_data, nr)),
1412 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
1413 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002),
1414 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1),
1415 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003),
1416 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1417 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004),
1418 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1419 };
1420
1421 memset(&self->prog, 0, sizeof(self->prog));
1422 self->prog.filter = malloc(sizeof(filter));
1423 ASSERT_NE(NULL, self->prog.filter);
1424 memcpy(self->prog.filter, filter, sizeof(filter));
1425 self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1426
1427 /* Prepare some testable syscall results. */
1428 self->mytid = syscall(__NR_gettid);
1429 ASSERT_GT(self->mytid, 0);
1430 ASSERT_NE(self->mytid, 1) {
1431 TH_LOG("Running this test as init is not supported. :)");
1432 }
1433
1434 self->mypid = getpid();
1435 ASSERT_GT(self->mypid, 0);
1436 ASSERT_EQ(self->mytid, self->mypid);
1437
1438 self->parent = getppid();
1439 ASSERT_GT(self->parent, 0);
1440 ASSERT_NE(self->parent, self->mypid);
1441
1442 /* Launch tracer. */
1443 self->tracer = setup_trace_fixture(_metadata, tracer_syscall, NULL);
1444 }
1445
1446 FIXTURE_TEARDOWN(TRACE_syscall)
1447 {
1448 teardown_trace_fixture(_metadata, self->tracer);
1449 if (self->prog.filter)
1450 free(self->prog.filter);
1451 }
1452
1453 TEST_F(TRACE_syscall, syscall_allowed)
1454 {
1455 long ret;
1456
1457 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1458 ASSERT_EQ(0, ret);
1459
1460 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1461 ASSERT_EQ(0, ret);
1462
1463 /* getppid works as expected (no changes). */
1464 EXPECT_EQ(self->parent, syscall(__NR_getppid));
1465 EXPECT_NE(self->mypid, syscall(__NR_getppid));
1466 }
1467
1468 TEST_F(TRACE_syscall, syscall_redirected)
1469 {
1470 long ret;
1471
1472 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1473 ASSERT_EQ(0, ret);
1474
1475 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1476 ASSERT_EQ(0, ret);
1477
1478 /* getpid has been redirected to getppid as expected. */
1479 EXPECT_EQ(self->parent, syscall(__NR_getpid));
1480 EXPECT_NE(self->mypid, syscall(__NR_getpid));
1481 }
1482
1483 TEST_F(TRACE_syscall, syscall_dropped)
1484 {
1485 long ret;
1486
1487 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1488 ASSERT_EQ(0, ret);
1489
1490 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1491 ASSERT_EQ(0, ret);
1492
1493 #ifdef SYSCALL_NUM_RET_SHARE_REG
1494 /* gettid has been skipped */
1495 EXPECT_EQ(-1, syscall(__NR_gettid));
1496 #else
1497 /* gettid has been skipped and an altered return value stored. */
1498 EXPECT_EQ(1, syscall(__NR_gettid));
1499 #endif
1500 EXPECT_NE(self->mytid, syscall(__NR_gettid));
1501 }
1502
1503 #ifndef __NR_seccomp
1504 # if defined(__i386__)
1505 # define __NR_seccomp 354
1506 # elif defined(__x86_64__)
1507 # define __NR_seccomp 317
1508 # elif defined(__arm__)
1509 # define __NR_seccomp 383
1510 # elif defined(__aarch64__)
1511 # define __NR_seccomp 277
1512 # elif defined(__hppa__)
1513 # define __NR_seccomp 338
1514 # elif defined(__powerpc__)
1515 # define __NR_seccomp 358
1516 # elif defined(__s390__)
1517 # define __NR_seccomp 348
1518 # else
1519 # warning "seccomp syscall number unknown for this architecture"
1520 # define __NR_seccomp 0xffff
1521 # endif
1522 #endif
1523
1524 #ifndef SECCOMP_SET_MODE_STRICT
1525 #define SECCOMP_SET_MODE_STRICT 0
1526 #endif
1527
1528 #ifndef SECCOMP_SET_MODE_FILTER
1529 #define SECCOMP_SET_MODE_FILTER 1
1530 #endif
1531
1532 #ifndef SECCOMP_FILTER_FLAG_TSYNC
1533 #define SECCOMP_FILTER_FLAG_TSYNC 1
1534 #endif
1535
1536 #ifndef seccomp
1537 int seccomp(unsigned int op, unsigned int flags, void *args)
1538 {
1539 errno = 0;
1540 return syscall(__NR_seccomp, op, flags, args);
1541 }
1542 #endif
1543
1544 TEST(seccomp_syscall)
1545 {
1546 struct sock_filter filter[] = {
1547 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1548 };
1549 struct sock_fprog prog = {
1550 .len = (unsigned short)ARRAY_SIZE(filter),
1551 .filter = filter,
1552 };
1553 long ret;
1554
1555 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1556 ASSERT_EQ(0, ret) {
1557 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1558 }
1559
1560 /* Reject insane operation. */
1561 ret = seccomp(-1, 0, &prog);
1562 ASSERT_NE(ENOSYS, errno) {
1563 TH_LOG("Kernel does not support seccomp syscall!");
1564 }
1565 EXPECT_EQ(EINVAL, errno) {
1566 TH_LOG("Did not reject crazy op value!");
1567 }
1568
1569 /* Reject strict with flags or pointer. */
1570 ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL);
1571 EXPECT_EQ(EINVAL, errno) {
1572 TH_LOG("Did not reject mode strict with flags!");
1573 }
1574 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog);
1575 EXPECT_EQ(EINVAL, errno) {
1576 TH_LOG("Did not reject mode strict with uargs!");
1577 }
1578
1579 /* Reject insane args for filter. */
1580 ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog);
1581 EXPECT_EQ(EINVAL, errno) {
1582 TH_LOG("Did not reject crazy filter flags!");
1583 }
1584 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL);
1585 EXPECT_EQ(EFAULT, errno) {
1586 TH_LOG("Did not reject NULL filter!");
1587 }
1588
1589 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
1590 EXPECT_EQ(0, errno) {
1591 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s",
1592 strerror(errno));
1593 }
1594 }
1595
1596 TEST(seccomp_syscall_mode_lock)
1597 {
1598 struct sock_filter filter[] = {
1599 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1600 };
1601 struct sock_fprog prog = {
1602 .len = (unsigned short)ARRAY_SIZE(filter),
1603 .filter = filter,
1604 };
1605 long ret;
1606
1607 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
1608 ASSERT_EQ(0, ret) {
1609 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1610 }
1611
1612 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
1613 ASSERT_NE(ENOSYS, errno) {
1614 TH_LOG("Kernel does not support seccomp syscall!");
1615 }
1616 EXPECT_EQ(0, ret) {
1617 TH_LOG("Could not install filter!");
1618 }
1619
1620 /* Make sure neither entry point will switch to strict. */
1621 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0);
1622 EXPECT_EQ(EINVAL, errno) {
1623 TH_LOG("Switched to mode strict!");
1624 }
1625
1626 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL);
1627 EXPECT_EQ(EINVAL, errno) {
1628 TH_LOG("Switched to mode strict!");
1629 }
1630 }
1631
1632 TEST(TSYNC_first)
1633 {
1634 struct sock_filter filter[] = {
1635 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1636 };
1637 struct sock_fprog prog = {
1638 .len = (unsigned short)ARRAY_SIZE(filter),
1639 .filter = filter,
1640 };
1641 long ret;
1642
1643 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
1644 ASSERT_EQ(0, ret) {
1645 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1646 }
1647
1648 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
1649 &prog);
1650 ASSERT_NE(ENOSYS, errno) {
1651 TH_LOG("Kernel does not support seccomp syscall!");
1652 }
1653 EXPECT_EQ(0, ret) {
1654 TH_LOG("Could not install initial filter with TSYNC!");
1655 }
1656 }
1657
1658 #define TSYNC_SIBLINGS 2
1659 struct tsync_sibling {
1660 pthread_t tid;
1661 pid_t system_tid;
1662 sem_t *started;
1663 pthread_cond_t *cond;
1664 pthread_mutex_t *mutex;
1665 int diverge;
1666 int num_waits;
1667 struct sock_fprog *prog;
1668 struct __test_metadata *metadata;
1669 };
1670
1671 FIXTURE_DATA(TSYNC) {
1672 struct sock_fprog root_prog, apply_prog;
1673 struct tsync_sibling sibling[TSYNC_SIBLINGS];
1674 sem_t started;
1675 pthread_cond_t cond;
1676 pthread_mutex_t mutex;
1677 int sibling_count;
1678 };
1679
1680 FIXTURE_SETUP(TSYNC)
1681 {
1682 struct sock_filter root_filter[] = {
1683 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1684 };
1685 struct sock_filter apply_filter[] = {
1686 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1687 offsetof(struct seccomp_data, nr)),
1688 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1689 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1690 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1691 };
1692
1693 memset(&self->root_prog, 0, sizeof(self->root_prog));
1694 memset(&self->apply_prog, 0, sizeof(self->apply_prog));
1695 memset(&self->sibling, 0, sizeof(self->sibling));
1696 self->root_prog.filter = malloc(sizeof(root_filter));
1697 ASSERT_NE(NULL, self->root_prog.filter);
1698 memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter));
1699 self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter);
1700
1701 self->apply_prog.filter = malloc(sizeof(apply_filter));
1702 ASSERT_NE(NULL, self->apply_prog.filter);
1703 memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter));
1704 self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter);
1705
1706 self->sibling_count = 0;
1707 pthread_mutex_init(&self->mutex, NULL);
1708 pthread_cond_init(&self->cond, NULL);
1709 sem_init(&self->started, 0, 0);
1710 self->sibling[0].tid = 0;
1711 self->sibling[0].cond = &self->cond;
1712 self->sibling[0].started = &self->started;
1713 self->sibling[0].mutex = &self->mutex;
1714 self->sibling[0].diverge = 0;
1715 self->sibling[0].num_waits = 1;
1716 self->sibling[0].prog = &self->root_prog;
1717 self->sibling[0].metadata = _metadata;
1718 self->sibling[1].tid = 0;
1719 self->sibling[1].cond = &self->cond;
1720 self->sibling[1].started = &self->started;
1721 self->sibling[1].mutex = &self->mutex;
1722 self->sibling[1].diverge = 0;
1723 self->sibling[1].prog = &self->root_prog;
1724 self->sibling[1].num_waits = 1;
1725 self->sibling[1].metadata = _metadata;
1726 }
1727
1728 FIXTURE_TEARDOWN(TSYNC)
1729 {
1730 int sib = 0;
1731
1732 if (self->root_prog.filter)
1733 free(self->root_prog.filter);
1734 if (self->apply_prog.filter)
1735 free(self->apply_prog.filter);
1736
1737 for ( ; sib < self->sibling_count; ++sib) {
1738 struct tsync_sibling *s = &self->sibling[sib];
1739 void *status;
1740
1741 if (!s->tid)
1742 continue;
1743 if (pthread_kill(s->tid, 0)) {
1744 pthread_cancel(s->tid);
1745 pthread_join(s->tid, &status);
1746 }
1747 }
1748 pthread_mutex_destroy(&self->mutex);
1749 pthread_cond_destroy(&self->cond);
1750 sem_destroy(&self->started);
1751 }
1752
1753 void *tsync_sibling(void *data)
1754 {
1755 long ret = 0;
1756 struct tsync_sibling *me = data;
1757
1758 me->system_tid = syscall(__NR_gettid);
1759
1760 pthread_mutex_lock(me->mutex);
1761 if (me->diverge) {
1762 /* Just re-apply the root prog to fork the tree */
1763 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
1764 me->prog, 0, 0);
1765 }
1766 sem_post(me->started);
1767 /* Return outside of started so parent notices failures. */
1768 if (ret) {
1769 pthread_mutex_unlock(me->mutex);
1770 return (void *)SIBLING_EXIT_FAILURE;
1771 }
1772 do {
1773 pthread_cond_wait(me->cond, me->mutex);
1774 me->num_waits = me->num_waits - 1;
1775 } while (me->num_waits);
1776 pthread_mutex_unlock(me->mutex);
1777
1778 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0);
1779 if (!ret)
1780 return (void *)SIBLING_EXIT_NEWPRIVS;
1781 read(0, NULL, 0);
1782 return (void *)SIBLING_EXIT_UNKILLED;
1783 }
1784
1785 void tsync_start_sibling(struct tsync_sibling *sibling)
1786 {
1787 pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling);
1788 }
1789
1790 TEST_F(TSYNC, siblings_fail_prctl)
1791 {
1792 long ret;
1793 void *status;
1794 struct sock_filter filter[] = {
1795 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1796 offsetof(struct seccomp_data, nr)),
1797 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
1798 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL),
1799 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1800 };
1801 struct sock_fprog prog = {
1802 .len = (unsigned short)ARRAY_SIZE(filter),
1803 .filter = filter,
1804 };
1805
1806 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
1807 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1808 }
1809
1810 /* Check prctl failure detection by requesting sib 0 diverge. */
1811 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
1812 ASSERT_NE(ENOSYS, errno) {
1813 TH_LOG("Kernel does not support seccomp syscall!");
1814 }
1815 ASSERT_EQ(0, ret) {
1816 TH_LOG("setting filter failed");
1817 }
1818
1819 self->sibling[0].diverge = 1;
1820 tsync_start_sibling(&self->sibling[0]);
1821 tsync_start_sibling(&self->sibling[1]);
1822
1823 while (self->sibling_count < TSYNC_SIBLINGS) {
1824 sem_wait(&self->started);
1825 self->sibling_count++;
1826 }
1827
1828 /* Signal the threads to clean up*/
1829 pthread_mutex_lock(&self->mutex);
1830 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
1831 TH_LOG("cond broadcast non-zero");
1832 }
1833 pthread_mutex_unlock(&self->mutex);
1834
1835 /* Ensure diverging sibling failed to call prctl. */
1836 pthread_join(self->sibling[0].tid, &status);
1837 EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status);
1838 pthread_join(self->sibling[1].tid, &status);
1839 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
1840 }
1841
1842 TEST_F(TSYNC, two_siblings_with_ancestor)
1843 {
1844 long ret;
1845 void *status;
1846
1847 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
1848 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1849 }
1850
1851 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
1852 ASSERT_NE(ENOSYS, errno) {
1853 TH_LOG("Kernel does not support seccomp syscall!");
1854 }
1855 ASSERT_EQ(0, ret) {
1856 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
1857 }
1858 tsync_start_sibling(&self->sibling[0]);
1859 tsync_start_sibling(&self->sibling[1]);
1860
1861 while (self->sibling_count < TSYNC_SIBLINGS) {
1862 sem_wait(&self->started);
1863 self->sibling_count++;
1864 }
1865
1866 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
1867 &self->apply_prog);
1868 ASSERT_EQ(0, ret) {
1869 TH_LOG("Could install filter on all threads!");
1870 }
1871 /* Tell the siblings to test the policy */
1872 pthread_mutex_lock(&self->mutex);
1873 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
1874 TH_LOG("cond broadcast non-zero");
1875 }
1876 pthread_mutex_unlock(&self->mutex);
1877 /* Ensure they are both killed and don't exit cleanly. */
1878 pthread_join(self->sibling[0].tid, &status);
1879 EXPECT_EQ(0x0, (long)status);
1880 pthread_join(self->sibling[1].tid, &status);
1881 EXPECT_EQ(0x0, (long)status);
1882 }
1883
1884 TEST_F(TSYNC, two_sibling_want_nnp)
1885 {
1886 void *status;
1887
1888 /* start siblings before any prctl() operations */
1889 tsync_start_sibling(&self->sibling[0]);
1890 tsync_start_sibling(&self->sibling[1]);
1891 while (self->sibling_count < TSYNC_SIBLINGS) {
1892 sem_wait(&self->started);
1893 self->sibling_count++;
1894 }
1895
1896 /* Tell the siblings to test no policy */
1897 pthread_mutex_lock(&self->mutex);
1898 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
1899 TH_LOG("cond broadcast non-zero");
1900 }
1901 pthread_mutex_unlock(&self->mutex);
1902
1903 /* Ensure they are both upset about lacking nnp. */
1904 pthread_join(self->sibling[0].tid, &status);
1905 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
1906 pthread_join(self->sibling[1].tid, &status);
1907 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
1908 }
1909
1910 TEST_F(TSYNC, two_siblings_with_no_filter)
1911 {
1912 long ret;
1913 void *status;
1914
1915 /* start siblings before any prctl() operations */
1916 tsync_start_sibling(&self->sibling[0]);
1917 tsync_start_sibling(&self->sibling[1]);
1918 while (self->sibling_count < TSYNC_SIBLINGS) {
1919 sem_wait(&self->started);
1920 self->sibling_count++;
1921 }
1922
1923 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
1924 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1925 }
1926
1927 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
1928 &self->apply_prog);
1929 ASSERT_NE(ENOSYS, errno) {
1930 TH_LOG("Kernel does not support seccomp syscall!");
1931 }
1932 ASSERT_EQ(0, ret) {
1933 TH_LOG("Could install filter on all threads!");
1934 }
1935
1936 /* Tell the siblings to test the policy */
1937 pthread_mutex_lock(&self->mutex);
1938 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
1939 TH_LOG("cond broadcast non-zero");
1940 }
1941 pthread_mutex_unlock(&self->mutex);
1942
1943 /* Ensure they are both killed and don't exit cleanly. */
1944 pthread_join(self->sibling[0].tid, &status);
1945 EXPECT_EQ(0x0, (long)status);
1946 pthread_join(self->sibling[1].tid, &status);
1947 EXPECT_EQ(0x0, (long)status);
1948 }
1949
1950 TEST_F(TSYNC, two_siblings_with_one_divergence)
1951 {
1952 long ret;
1953 void *status;
1954
1955 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
1956 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1957 }
1958
1959 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
1960 ASSERT_NE(ENOSYS, errno) {
1961 TH_LOG("Kernel does not support seccomp syscall!");
1962 }
1963 ASSERT_EQ(0, ret) {
1964 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
1965 }
1966 self->sibling[0].diverge = 1;
1967 tsync_start_sibling(&self->sibling[0]);
1968 tsync_start_sibling(&self->sibling[1]);
1969
1970 while (self->sibling_count < TSYNC_SIBLINGS) {
1971 sem_wait(&self->started);
1972 self->sibling_count++;
1973 }
1974
1975 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
1976 &self->apply_prog);
1977 ASSERT_EQ(self->sibling[0].system_tid, ret) {
1978 TH_LOG("Did not fail on diverged sibling.");
1979 }
1980
1981 /* Wake the threads */
1982 pthread_mutex_lock(&self->mutex);
1983 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
1984 TH_LOG("cond broadcast non-zero");
1985 }
1986 pthread_mutex_unlock(&self->mutex);
1987
1988 /* Ensure they are both unkilled. */
1989 pthread_join(self->sibling[0].tid, &status);
1990 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
1991 pthread_join(self->sibling[1].tid, &status);
1992 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
1993 }
1994
1995 TEST_F(TSYNC, two_siblings_not_under_filter)
1996 {
1997 long ret, sib;
1998 void *status;
1999
2000 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2001 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2002 }
2003
2004 /*
2005 * Sibling 0 will have its own seccomp policy
2006 * and Sibling 1 will not be under seccomp at
2007 * all. Sibling 1 will enter seccomp and 0
2008 * will cause failure.
2009 */
2010 self->sibling[0].diverge = 1;
2011 tsync_start_sibling(&self->sibling[0]);
2012 tsync_start_sibling(&self->sibling[1]);
2013
2014 while (self->sibling_count < TSYNC_SIBLINGS) {
2015 sem_wait(&self->started);
2016 self->sibling_count++;
2017 }
2018
2019 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2020 ASSERT_NE(ENOSYS, errno) {
2021 TH_LOG("Kernel does not support seccomp syscall!");
2022 }
2023 ASSERT_EQ(0, ret) {
2024 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2025 }
2026
2027 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2028 &self->apply_prog);
2029 ASSERT_EQ(ret, self->sibling[0].system_tid) {
2030 TH_LOG("Did not fail on diverged sibling.");
2031 }
2032 sib = 1;
2033 if (ret == self->sibling[0].system_tid)
2034 sib = 0;
2035
2036 pthread_mutex_lock(&self->mutex);
2037
2038 /* Increment the other siblings num_waits so we can clean up
2039 * the one we just saw.
2040 */
2041 self->sibling[!sib].num_waits += 1;
2042
2043 /* Signal the thread to clean up*/
2044 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2045 TH_LOG("cond broadcast non-zero");
2046 }
2047 pthread_mutex_unlock(&self->mutex);
2048 pthread_join(self->sibling[sib].tid, &status);
2049 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2050 /* Poll for actual task death. pthread_join doesn't guarantee it. */
2051 while (!kill(self->sibling[sib].system_tid, 0))
2052 sleep(0.1);
2053 /* Switch to the remaining sibling */
2054 sib = !sib;
2055
2056 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2057 &self->apply_prog);
2058 ASSERT_EQ(0, ret) {
2059 TH_LOG("Expected the remaining sibling to sync");
2060 };
2061
2062 pthread_mutex_lock(&self->mutex);
2063
2064 /* If remaining sibling didn't have a chance to wake up during
2065 * the first broadcast, manually reduce the num_waits now.
2066 */
2067 if (self->sibling[sib].num_waits > 1)
2068 self->sibling[sib].num_waits = 1;
2069 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2070 TH_LOG("cond broadcast non-zero");
2071 }
2072 pthread_mutex_unlock(&self->mutex);
2073 pthread_join(self->sibling[sib].tid, &status);
2074 EXPECT_EQ(0, (long)status);
2075 /* Poll for actual task death. pthread_join doesn't guarantee it. */
2076 while (!kill(self->sibling[sib].system_tid, 0))
2077 sleep(0.1);
2078
2079 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2080 &self->apply_prog);
2081 ASSERT_EQ(0, ret); /* just us chickens */
2082 }
2083
2084 /* Make sure restarted syscalls are seen directly as "restart_syscall". */
2085 TEST(syscall_restart)
2086 {
2087 long ret;
2088 unsigned long msg;
2089 pid_t child_pid;
2090 int pipefd[2];
2091 int status;
2092 siginfo_t info = { };
2093 struct sock_filter filter[] = {
2094 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2095 offsetof(struct seccomp_data, nr)),
2096
2097 #ifdef __NR_sigreturn
2098 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 6, 0),
2099 #endif
2100 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 5, 0),
2101 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 4, 0),
2102 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 3, 0),
2103 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 4, 0),
2104 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0),
2105
2106 /* Allow __NR_write for easy logging. */
2107 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1),
2108 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2109 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2110 /* The nanosleep jump target. */
2111 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100),
2112 /* The restart_syscall jump target. */
2113 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200),
2114 };
2115 struct sock_fprog prog = {
2116 .len = (unsigned short)ARRAY_SIZE(filter),
2117 .filter = filter,
2118 };
2119 #if defined(__arm__)
2120 struct utsname utsbuf;
2121 #endif
2122
2123 ASSERT_EQ(0, pipe(pipefd));
2124
2125 child_pid = fork();
2126 ASSERT_LE(0, child_pid);
2127 if (child_pid == 0) {
2128 /* Child uses EXPECT not ASSERT to deliver status correctly. */
2129 char buf = ' ';
2130 struct timespec timeout = { };
2131
2132 /* Attach parent as tracer and stop. */
2133 EXPECT_EQ(0, ptrace(PTRACE_TRACEME));
2134 EXPECT_EQ(0, raise(SIGSTOP));
2135
2136 EXPECT_EQ(0, close(pipefd[1]));
2137
2138 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2139 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2140 }
2141
2142 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2143 EXPECT_EQ(0, ret) {
2144 TH_LOG("Failed to install filter!");
2145 }
2146
2147 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2148 TH_LOG("Failed to read() sync from parent");
2149 }
2150 EXPECT_EQ('.', buf) {
2151 TH_LOG("Failed to get sync data from read()");
2152 }
2153
2154 /* Start nanosleep to be interrupted. */
2155 timeout.tv_sec = 1;
2156 errno = 0;
2157 EXPECT_EQ(0, nanosleep(&timeout, NULL)) {
2158 TH_LOG("Call to nanosleep() failed (errno %d)", errno);
2159 }
2160
2161 /* Read final sync from parent. */
2162 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2163 TH_LOG("Failed final read() from parent");
2164 }
2165 EXPECT_EQ('!', buf) {
2166 TH_LOG("Failed to get final data from read()");
2167 }
2168
2169 /* Directly report the status of our test harness results. */
2170 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS
2171 : EXIT_FAILURE);
2172 }
2173 EXPECT_EQ(0, close(pipefd[0]));
2174
2175 /* Attach to child, setup options, and release. */
2176 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2177 ASSERT_EQ(true, WIFSTOPPED(status));
2178 ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL,
2179 PTRACE_O_TRACESECCOMP));
2180 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2181 ASSERT_EQ(1, write(pipefd[1], ".", 1));
2182
2183 /* Wait for nanosleep() to start. */
2184 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2185 ASSERT_EQ(true, WIFSTOPPED(status));
2186 ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2187 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2188 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2189 ASSERT_EQ(0x100, msg);
2190 EXPECT_EQ(__NR_nanosleep, get_syscall(_metadata, child_pid));
2191
2192 /* Might as well check siginfo for sanity while we're here. */
2193 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2194 ASSERT_EQ(SIGTRAP, info.si_signo);
2195 ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code);
2196 EXPECT_EQ(0, info.si_errno);
2197 EXPECT_EQ(getuid(), info.si_uid);
2198 /* Verify signal delivery came from child (seccomp-triggered). */
2199 EXPECT_EQ(child_pid, info.si_pid);
2200
2201 /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */
2202 ASSERT_EQ(0, kill(child_pid, SIGSTOP));
2203 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2204 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2205 ASSERT_EQ(true, WIFSTOPPED(status));
2206 ASSERT_EQ(SIGSTOP, WSTOPSIG(status));
2207 /* Verify signal delivery came from parent now. */
2208 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2209 EXPECT_EQ(getpid(), info.si_pid);
2210
2211 /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */
2212 ASSERT_EQ(0, kill(child_pid, SIGCONT));
2213 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2214 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2215 ASSERT_EQ(true, WIFSTOPPED(status));
2216 ASSERT_EQ(SIGCONT, WSTOPSIG(status));
2217 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2218
2219 /* Wait for restart_syscall() to start. */
2220 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2221 ASSERT_EQ(true, WIFSTOPPED(status));
2222 ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2223 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2224 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2225
2226 ASSERT_EQ(0x200, msg);
2227 ret = get_syscall(_metadata, child_pid);
2228 #if defined(__arm__)
2229 /*
2230 * FIXME:
2231 * - native ARM registers do NOT expose true syscall.
2232 * - compat ARM registers on ARM64 DO expose true syscall.
2233 */
2234 ASSERT_EQ(0, uname(&utsbuf));
2235 if (strncmp(utsbuf.machine, "arm", 3) == 0) {
2236 EXPECT_EQ(__NR_nanosleep, ret);
2237 } else
2238 #endif
2239 {
2240 EXPECT_EQ(__NR_restart_syscall, ret);
2241 }
2242
2243 /* Write again to end test. */
2244 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2245 ASSERT_EQ(1, write(pipefd[1], "!", 1));
2246 EXPECT_EQ(0, close(pipefd[1]));
2247
2248 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2249 if (WIFSIGNALED(status) || WEXITSTATUS(status))
2250 _metadata->passed = 0;
2251 }
2252
2253 /*
2254 * TODO:
2255 * - add microbenchmarks
2256 * - expand NNP testing
2257 * - better arch-specific TRACE and TRAP handlers.
2258 * - endianness checking when appropriate
2259 * - 64-bit arg prodding
2260 * - arch value testing (x86 modes especially)
2261 * - ...
2262 */
2263
2264 TEST_HARNESS_MAIN
ubuntu-zesty-kernel mirror