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 "../kselftest_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_ALLOW 72 struct seccomp_data { 73 int nr; 74 __u32 arch; 75 __u64 instruction_pointer; 76 __u64 args[6]; 77 }; 78 #endif 79 80 #ifndef SECCOMP_RET_KILL_PROCESS 81 #define SECCOMP_RET_KILL_PROCESS 0x80000000U /* kill the process */ 82 #define SECCOMP_RET_KILL_THREAD 0x00000000U /* kill the thread */ 83 #endif 84 #ifndef SECCOMP_RET_KILL 85 #define SECCOMP_RET_KILL SECCOMP_RET_KILL_THREAD 86 #define SECCOMP_RET_TRAP 0x00030000U /* disallow and force a SIGSYS */ 87 #define SECCOMP_RET_ERRNO 0x00050000U /* returns an errno */ 88 #define SECCOMP_RET_TRACE 0x7ff00000U /* pass to a tracer or disallow */ 89 #define SECCOMP_RET_ALLOW 0x7fff0000U /* allow */ 90 #endif 91 #ifndef SECCOMP_RET_LOG 92 #define SECCOMP_RET_LOG 0x7ffc0000U /* allow after logging */ 93 #endif 94 95 #ifndef __NR_seccomp 96 # if defined(__i386__) 97 # define __NR_seccomp 354 98 # elif defined(__x86_64__) 99 # define __NR_seccomp 317 100 # elif defined(__arm__) 101 # define __NR_seccomp 383 102 # elif defined(__aarch64__) 103 # define __NR_seccomp 277 104 # elif defined(__hppa__) 105 # define __NR_seccomp 338 106 # elif defined(__powerpc__) 107 # define __NR_seccomp 358 108 # elif defined(__s390__) 109 # define __NR_seccomp 348 110 # else 111 # warning "seccomp syscall number unknown for this architecture" 112 # define __NR_seccomp 0xffff 113 # endif 114 #endif 115 116 #ifndef SECCOMP_SET_MODE_STRICT 117 #define SECCOMP_SET_MODE_STRICT 0 118 #endif 119 120 #ifndef SECCOMP_SET_MODE_FILTER 121 #define SECCOMP_SET_MODE_FILTER 1 122 #endif 123 124 #ifndef SECCOMP_GET_ACTION_AVAIL 125 #define SECCOMP_GET_ACTION_AVAIL 2 126 #endif 127 128 #ifndef SECCOMP_FILTER_FLAG_TSYNC 129 #define SECCOMP_FILTER_FLAG_TSYNC 1 130 #endif 131 132 #ifndef SECCOMP_FILTER_FLAG_LOG 133 #define SECCOMP_FILTER_FLAG_LOG 2 134 #endif 135 136 #ifndef seccomp 137 int seccomp(unsigned int op, unsigned int flags, void *args) 138 { 139 errno = 0; 140 return syscall(__NR_seccomp, op, flags, args); 141 } 142 #endif 143 144 #if __BYTE_ORDER == __LITTLE_ENDIAN 145 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n])) 146 #elif __BYTE_ORDER == __BIG_ENDIAN 147 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32)) 148 #else 149 #error "wut? Unknown __BYTE_ORDER?!" 150 #endif 151 152 #define SIBLING_EXIT_UNKILLED 0xbadbeef 153 #define SIBLING_EXIT_FAILURE 0xbadface 154 #define SIBLING_EXIT_NEWPRIVS 0xbadfeed 155 156 TEST(mode_strict_support) 157 { 158 long ret; 159 160 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL); 161 ASSERT_EQ(0, ret) { 162 TH_LOG("Kernel does not support CONFIG_SECCOMP"); 163 } 164 syscall(__NR_exit, 0); 165 } 166 167 TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL) 168 { 169 long ret; 170 171 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL); 172 ASSERT_EQ(0, ret) { 173 TH_LOG("Kernel does not support CONFIG_SECCOMP"); 174 } 175 syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER, 176 NULL, NULL, NULL); 177 EXPECT_FALSE(true) { 178 TH_LOG("Unreachable!"); 179 } 180 } 181 182 /* Note! This doesn't test no new privs behavior */ 183 TEST(no_new_privs_support) 184 { 185 long ret; 186 187 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 188 EXPECT_EQ(0, ret) { 189 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 190 } 191 } 192 193 /* Tests kernel support by checking for a copy_from_user() fault on NULL. */ 194 TEST(mode_filter_support) 195 { 196 long ret; 197 198 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 199 ASSERT_EQ(0, ret) { 200 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 201 } 202 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL); 203 EXPECT_EQ(-1, ret); 204 EXPECT_EQ(EFAULT, errno) { 205 TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!"); 206 } 207 } 208 209 TEST(mode_filter_without_nnp) 210 { 211 struct sock_filter filter[] = { 212 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 213 }; 214 struct sock_fprog prog = { 215 .len = (unsigned short)ARRAY_SIZE(filter), 216 .filter = filter, 217 }; 218 long ret; 219 220 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0); 221 ASSERT_LE(0, ret) { 222 TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS"); 223 } 224 errno = 0; 225 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 226 /* Succeeds with CAP_SYS_ADMIN, fails without */ 227 /* TODO(wad) check caps not euid */ 228 if (geteuid()) { 229 EXPECT_EQ(-1, ret); 230 EXPECT_EQ(EACCES, errno); 231 } else { 232 EXPECT_EQ(0, ret); 233 } 234 } 235 236 #define MAX_INSNS_PER_PATH 32768 237 238 TEST(filter_size_limits) 239 { 240 int i; 241 int count = BPF_MAXINSNS + 1; 242 struct sock_filter allow[] = { 243 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 244 }; 245 struct sock_filter *filter; 246 struct sock_fprog prog = { }; 247 long ret; 248 249 filter = calloc(count, sizeof(*filter)); 250 ASSERT_NE(NULL, filter); 251 252 for (i = 0; i < count; i++) 253 filter[i] = allow[0]; 254 255 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 256 ASSERT_EQ(0, ret); 257 258 prog.filter = filter; 259 prog.len = count; 260 261 /* Too many filter instructions in a single filter. */ 262 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 263 ASSERT_NE(0, ret) { 264 TH_LOG("Installing %d insn filter was allowed", prog.len); 265 } 266 267 /* One less is okay, though. */ 268 prog.len -= 1; 269 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 270 ASSERT_EQ(0, ret) { 271 TH_LOG("Installing %d insn filter wasn't allowed", prog.len); 272 } 273 } 274 275 TEST(filter_chain_limits) 276 { 277 int i; 278 int count = BPF_MAXINSNS; 279 struct sock_filter allow[] = { 280 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 281 }; 282 struct sock_filter *filter; 283 struct sock_fprog prog = { }; 284 long ret; 285 286 filter = calloc(count, sizeof(*filter)); 287 ASSERT_NE(NULL, filter); 288 289 for (i = 0; i < count; i++) 290 filter[i] = allow[0]; 291 292 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 293 ASSERT_EQ(0, ret); 294 295 prog.filter = filter; 296 prog.len = 1; 297 298 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 299 ASSERT_EQ(0, ret); 300 301 prog.len = count; 302 303 /* Too many total filter instructions. */ 304 for (i = 0; i < MAX_INSNS_PER_PATH; i++) { 305 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 306 if (ret != 0) 307 break; 308 } 309 ASSERT_NE(0, ret) { 310 TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)", 311 i, count, i * (count + 4)); 312 } 313 } 314 315 TEST(mode_filter_cannot_move_to_strict) 316 { 317 struct sock_filter filter[] = { 318 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 319 }; 320 struct sock_fprog prog = { 321 .len = (unsigned short)ARRAY_SIZE(filter), 322 .filter = filter, 323 }; 324 long ret; 325 326 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 327 ASSERT_EQ(0, ret); 328 329 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 330 ASSERT_EQ(0, ret); 331 332 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0); 333 EXPECT_EQ(-1, ret); 334 EXPECT_EQ(EINVAL, errno); 335 } 336 337 338 TEST(mode_filter_get_seccomp) 339 { 340 struct sock_filter filter[] = { 341 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 342 }; 343 struct sock_fprog prog = { 344 .len = (unsigned short)ARRAY_SIZE(filter), 345 .filter = filter, 346 }; 347 long ret; 348 349 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 350 ASSERT_EQ(0, ret); 351 352 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0); 353 EXPECT_EQ(0, ret); 354 355 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 356 ASSERT_EQ(0, ret); 357 358 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0); 359 EXPECT_EQ(2, ret); 360 } 361 362 363 TEST(ALLOW_all) 364 { 365 struct sock_filter filter[] = { 366 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 367 }; 368 struct sock_fprog prog = { 369 .len = (unsigned short)ARRAY_SIZE(filter), 370 .filter = filter, 371 }; 372 long ret; 373 374 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 375 ASSERT_EQ(0, ret); 376 377 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 378 ASSERT_EQ(0, ret); 379 } 380 381 TEST(empty_prog) 382 { 383 struct sock_filter filter[] = { 384 }; 385 struct sock_fprog prog = { 386 .len = (unsigned short)ARRAY_SIZE(filter), 387 .filter = filter, 388 }; 389 long ret; 390 391 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 392 ASSERT_EQ(0, ret); 393 394 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 395 EXPECT_EQ(-1, ret); 396 EXPECT_EQ(EINVAL, errno); 397 } 398 399 TEST(log_all) 400 { 401 struct sock_filter filter[] = { 402 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG), 403 }; 404 struct sock_fprog prog = { 405 .len = (unsigned short)ARRAY_SIZE(filter), 406 .filter = filter, 407 }; 408 long ret; 409 pid_t parent = getppid(); 410 411 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 412 ASSERT_EQ(0, ret); 413 414 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 415 ASSERT_EQ(0, ret); 416 417 /* getppid() should succeed and be logged (no check for logging) */ 418 EXPECT_EQ(parent, syscall(__NR_getppid)); 419 } 420 421 TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS) 422 { 423 struct sock_filter filter[] = { 424 BPF_STMT(BPF_RET|BPF_K, 0x10000000U), 425 }; 426 struct sock_fprog prog = { 427 .len = (unsigned short)ARRAY_SIZE(filter), 428 .filter = filter, 429 }; 430 long ret; 431 432 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 433 ASSERT_EQ(0, ret); 434 435 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 436 ASSERT_EQ(0, ret); 437 EXPECT_EQ(0, syscall(__NR_getpid)) { 438 TH_LOG("getpid() shouldn't ever return"); 439 } 440 } 441 442 /* return code >= 0x80000000 is unused. */ 443 TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS) 444 { 445 struct sock_filter filter[] = { 446 BPF_STMT(BPF_RET|BPF_K, 0x90000000U), 447 }; 448 struct sock_fprog prog = { 449 .len = (unsigned short)ARRAY_SIZE(filter), 450 .filter = filter, 451 }; 452 long ret; 453 454 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 455 ASSERT_EQ(0, ret); 456 457 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 458 ASSERT_EQ(0, ret); 459 EXPECT_EQ(0, syscall(__NR_getpid)) { 460 TH_LOG("getpid() shouldn't ever return"); 461 } 462 } 463 464 TEST_SIGNAL(KILL_all, SIGSYS) 465 { 466 struct sock_filter filter[] = { 467 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 468 }; 469 struct sock_fprog prog = { 470 .len = (unsigned short)ARRAY_SIZE(filter), 471 .filter = filter, 472 }; 473 long ret; 474 475 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 476 ASSERT_EQ(0, ret); 477 478 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 479 ASSERT_EQ(0, ret); 480 } 481 482 TEST_SIGNAL(KILL_one, SIGSYS) 483 { 484 struct sock_filter filter[] = { 485 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 486 offsetof(struct seccomp_data, nr)), 487 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 488 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 489 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 490 }; 491 struct sock_fprog prog = { 492 .len = (unsigned short)ARRAY_SIZE(filter), 493 .filter = filter, 494 }; 495 long ret; 496 pid_t parent = getppid(); 497 498 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 499 ASSERT_EQ(0, ret); 500 501 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 502 ASSERT_EQ(0, ret); 503 504 EXPECT_EQ(parent, syscall(__NR_getppid)); 505 /* getpid() should never return. */ 506 EXPECT_EQ(0, syscall(__NR_getpid)); 507 } 508 509 TEST_SIGNAL(KILL_one_arg_one, SIGSYS) 510 { 511 void *fatal_address; 512 struct sock_filter filter[] = { 513 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 514 offsetof(struct seccomp_data, nr)), 515 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0), 516 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 517 /* Only both with lower 32-bit for now. */ 518 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)), 519 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 520 (unsigned long)&fatal_address, 0, 1), 521 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 522 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 523 }; 524 struct sock_fprog prog = { 525 .len = (unsigned short)ARRAY_SIZE(filter), 526 .filter = filter, 527 }; 528 long ret; 529 pid_t parent = getppid(); 530 struct tms timebuf; 531 clock_t clock = times(&timebuf); 532 533 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 534 ASSERT_EQ(0, ret); 535 536 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 537 ASSERT_EQ(0, ret); 538 539 EXPECT_EQ(parent, syscall(__NR_getppid)); 540 EXPECT_LE(clock, syscall(__NR_times, &timebuf)); 541 /* times() should never return. */ 542 EXPECT_EQ(0, syscall(__NR_times, &fatal_address)); 543 } 544 545 TEST_SIGNAL(KILL_one_arg_six, SIGSYS) 546 { 547 #ifndef __NR_mmap2 548 int sysno = __NR_mmap; 549 #else 550 int sysno = __NR_mmap2; 551 #endif 552 struct sock_filter filter[] = { 553 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 554 offsetof(struct seccomp_data, nr)), 555 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0), 556 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 557 /* Only both with lower 32-bit for now. */ 558 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)), 559 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1), 560 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 561 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 562 }; 563 struct sock_fprog prog = { 564 .len = (unsigned short)ARRAY_SIZE(filter), 565 .filter = filter, 566 }; 567 long ret; 568 pid_t parent = getppid(); 569 int fd; 570 void *map1, *map2; 571 int page_size = sysconf(_SC_PAGESIZE); 572 573 ASSERT_LT(0, page_size); 574 575 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 576 ASSERT_EQ(0, ret); 577 578 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 579 ASSERT_EQ(0, ret); 580 581 fd = open("/dev/zero", O_RDONLY); 582 ASSERT_NE(-1, fd); 583 584 EXPECT_EQ(parent, syscall(__NR_getppid)); 585 map1 = (void *)syscall(sysno, 586 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size); 587 EXPECT_NE(MAP_FAILED, map1); 588 /* mmap2() should never return. */ 589 map2 = (void *)syscall(sysno, 590 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE); 591 EXPECT_EQ(MAP_FAILED, map2); 592 593 /* The test failed, so clean up the resources. */ 594 munmap(map1, page_size); 595 munmap(map2, page_size); 596 close(fd); 597 } 598 599 /* This is a thread task to die via seccomp filter violation. */ 600 void *kill_thread(void *data) 601 { 602 bool die = (bool)data; 603 604 if (die) { 605 prctl(PR_GET_SECCOMP, 0, 0, 0, 0); 606 return (void *)SIBLING_EXIT_FAILURE; 607 } 608 609 return (void *)SIBLING_EXIT_UNKILLED; 610 } 611 612 /* Prepare a thread that will kill itself or both of us. */ 613 void kill_thread_or_group(struct __test_metadata *_metadata, bool kill_process) 614 { 615 pthread_t thread; 616 void *status; 617 /* Kill only when calling __NR_prctl. */ 618 struct sock_filter filter_thread[] = { 619 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 620 offsetof(struct seccomp_data, nr)), 621 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1), 622 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD), 623 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 624 }; 625 struct sock_fprog prog_thread = { 626 .len = (unsigned short)ARRAY_SIZE(filter_thread), 627 .filter = filter_thread, 628 }; 629 struct sock_filter filter_process[] = { 630 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 631 offsetof(struct seccomp_data, nr)), 632 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1), 633 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_PROCESS), 634 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 635 }; 636 struct sock_fprog prog_process = { 637 .len = (unsigned short)ARRAY_SIZE(filter_process), 638 .filter = filter_process, 639 }; 640 641 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 642 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 643 } 644 645 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, 646 kill_process ? &prog_process : &prog_thread)); 647 648 /* 649 * Add the KILL_THREAD rule again to make sure that the KILL_PROCESS 650 * flag cannot be downgraded by a new filter. 651 */ 652 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog_thread)); 653 654 /* Start a thread that will exit immediately. */ 655 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)false)); 656 ASSERT_EQ(0, pthread_join(thread, &status)); 657 ASSERT_EQ(SIBLING_EXIT_UNKILLED, (unsigned long)status); 658 659 /* Start a thread that will die immediately. */ 660 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)true)); 661 ASSERT_EQ(0, pthread_join(thread, &status)); 662 ASSERT_NE(SIBLING_EXIT_FAILURE, (unsigned long)status); 663 664 /* 665 * If we get here, only the spawned thread died. Let the parent know 666 * the whole process didn't die (i.e. this thread, the spawner, 667 * stayed running). 668 */ 669 exit(42); 670 } 671 672 TEST(KILL_thread) 673 { 674 int status; 675 pid_t child_pid; 676 677 child_pid = fork(); 678 ASSERT_LE(0, child_pid); 679 if (child_pid == 0) { 680 kill_thread_or_group(_metadata, false); 681 _exit(38); 682 } 683 684 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 685 686 /* If only the thread was killed, we'll see exit 42. */ 687 ASSERT_TRUE(WIFEXITED(status)); 688 ASSERT_EQ(42, WEXITSTATUS(status)); 689 } 690 691 TEST(KILL_process) 692 { 693 int status; 694 pid_t child_pid; 695 696 child_pid = fork(); 697 ASSERT_LE(0, child_pid); 698 if (child_pid == 0) { 699 kill_thread_or_group(_metadata, true); 700 _exit(38); 701 } 702 703 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 704 705 /* If the entire process was killed, we'll see SIGSYS. */ 706 ASSERT_TRUE(WIFSIGNALED(status)); 707 ASSERT_EQ(SIGSYS, WTERMSIG(status)); 708 } 709 710 /* TODO(wad) add 64-bit versus 32-bit arg tests. */ 711 TEST(arg_out_of_range) 712 { 713 struct sock_filter filter[] = { 714 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)), 715 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 716 }; 717 struct sock_fprog prog = { 718 .len = (unsigned short)ARRAY_SIZE(filter), 719 .filter = filter, 720 }; 721 long ret; 722 723 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 724 ASSERT_EQ(0, ret); 725 726 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 727 EXPECT_EQ(-1, ret); 728 EXPECT_EQ(EINVAL, errno); 729 } 730 731 #define ERRNO_FILTER(name, errno) \ 732 struct sock_filter _read_filter_##name[] = { \ 733 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, \ 734 offsetof(struct seccomp_data, nr)), \ 735 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), \ 736 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | errno), \ 737 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), \ 738 }; \ 739 struct sock_fprog prog_##name = { \ 740 .len = (unsigned short)ARRAY_SIZE(_read_filter_##name), \ 741 .filter = _read_filter_##name, \ 742 } 743 744 /* Make sure basic errno values are correctly passed through a filter. */ 745 TEST(ERRNO_valid) 746 { 747 ERRNO_FILTER(valid, E2BIG); 748 long ret; 749 pid_t parent = getppid(); 750 751 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 752 ASSERT_EQ(0, ret); 753 754 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_valid); 755 ASSERT_EQ(0, ret); 756 757 EXPECT_EQ(parent, syscall(__NR_getppid)); 758 EXPECT_EQ(-1, read(0, NULL, 0)); 759 EXPECT_EQ(E2BIG, errno); 760 } 761 762 /* Make sure an errno of zero is correctly handled by the arch code. */ 763 TEST(ERRNO_zero) 764 { 765 ERRNO_FILTER(zero, 0); 766 long ret; 767 pid_t parent = getppid(); 768 769 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 770 ASSERT_EQ(0, ret); 771 772 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_zero); 773 ASSERT_EQ(0, ret); 774 775 EXPECT_EQ(parent, syscall(__NR_getppid)); 776 /* "errno" of 0 is ok. */ 777 EXPECT_EQ(0, read(0, NULL, 0)); 778 } 779 780 /* 781 * The SECCOMP_RET_DATA mask is 16 bits wide, but errno is smaller. 782 * This tests that the errno value gets capped correctly, fixed by 783 * 580c57f10768 ("seccomp: cap SECCOMP_RET_ERRNO data to MAX_ERRNO"). 784 */ 785 TEST(ERRNO_capped) 786 { 787 ERRNO_FILTER(capped, 4096); 788 long ret; 789 pid_t parent = getppid(); 790 791 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 792 ASSERT_EQ(0, ret); 793 794 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_capped); 795 ASSERT_EQ(0, ret); 796 797 EXPECT_EQ(parent, syscall(__NR_getppid)); 798 EXPECT_EQ(-1, read(0, NULL, 0)); 799 EXPECT_EQ(4095, errno); 800 } 801 802 /* 803 * Filters are processed in reverse order: last applied is executed first. 804 * Since only the SECCOMP_RET_ACTION mask is tested for return values, the 805 * SECCOMP_RET_DATA mask results will follow the most recently applied 806 * matching filter return (and not the lowest or highest value). 807 */ 808 TEST(ERRNO_order) 809 { 810 ERRNO_FILTER(first, 11); 811 ERRNO_FILTER(second, 13); 812 ERRNO_FILTER(third, 12); 813 long ret; 814 pid_t parent = getppid(); 815 816 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 817 ASSERT_EQ(0, ret); 818 819 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_first); 820 ASSERT_EQ(0, ret); 821 822 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_second); 823 ASSERT_EQ(0, ret); 824 825 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_third); 826 ASSERT_EQ(0, ret); 827 828 EXPECT_EQ(parent, syscall(__NR_getppid)); 829 EXPECT_EQ(-1, read(0, NULL, 0)); 830 EXPECT_EQ(12, errno); 831 } 832 833 FIXTURE_DATA(TRAP) { 834 struct sock_fprog prog; 835 }; 836 837 FIXTURE_SETUP(TRAP) 838 { 839 struct sock_filter filter[] = { 840 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 841 offsetof(struct seccomp_data, nr)), 842 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 843 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP), 844 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 845 }; 846 847 memset(&self->prog, 0, sizeof(self->prog)); 848 self->prog.filter = malloc(sizeof(filter)); 849 ASSERT_NE(NULL, self->prog.filter); 850 memcpy(self->prog.filter, filter, sizeof(filter)); 851 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 852 } 853 854 FIXTURE_TEARDOWN(TRAP) 855 { 856 if (self->prog.filter) 857 free(self->prog.filter); 858 } 859 860 TEST_F_SIGNAL(TRAP, dfl, SIGSYS) 861 { 862 long ret; 863 864 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 865 ASSERT_EQ(0, ret); 866 867 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 868 ASSERT_EQ(0, ret); 869 syscall(__NR_getpid); 870 } 871 872 /* Ensure that SIGSYS overrides SIG_IGN */ 873 TEST_F_SIGNAL(TRAP, ign, SIGSYS) 874 { 875 long ret; 876 877 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 878 ASSERT_EQ(0, ret); 879 880 signal(SIGSYS, SIG_IGN); 881 882 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 883 ASSERT_EQ(0, ret); 884 syscall(__NR_getpid); 885 } 886 887 static struct siginfo TRAP_info; 888 static volatile int TRAP_nr; 889 static void TRAP_action(int nr, siginfo_t *info, void *void_context) 890 { 891 memcpy(&TRAP_info, info, sizeof(TRAP_info)); 892 TRAP_nr = nr; 893 } 894 895 TEST_F(TRAP, handler) 896 { 897 int ret, test; 898 struct sigaction act; 899 sigset_t mask; 900 901 memset(&act, 0, sizeof(act)); 902 sigemptyset(&mask); 903 sigaddset(&mask, SIGSYS); 904 905 act.sa_sigaction = &TRAP_action; 906 act.sa_flags = SA_SIGINFO; 907 ret = sigaction(SIGSYS, &act, NULL); 908 ASSERT_EQ(0, ret) { 909 TH_LOG("sigaction failed"); 910 } 911 ret = sigprocmask(SIG_UNBLOCK, &mask, NULL); 912 ASSERT_EQ(0, ret) { 913 TH_LOG("sigprocmask failed"); 914 } 915 916 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 917 ASSERT_EQ(0, ret); 918 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 919 ASSERT_EQ(0, ret); 920 TRAP_nr = 0; 921 memset(&TRAP_info, 0, sizeof(TRAP_info)); 922 /* Expect the registers to be rolled back. (nr = error) may vary 923 * based on arch. */ 924 ret = syscall(__NR_getpid); 925 /* Silence gcc warning about volatile. */ 926 test = TRAP_nr; 927 EXPECT_EQ(SIGSYS, test); 928 struct local_sigsys { 929 void *_call_addr; /* calling user insn */ 930 int _syscall; /* triggering system call number */ 931 unsigned int _arch; /* AUDIT_ARCH_* of syscall */ 932 } *sigsys = (struct local_sigsys *) 933 #ifdef si_syscall 934 &(TRAP_info.si_call_addr); 935 #else 936 &TRAP_info.si_pid; 937 #endif 938 EXPECT_EQ(__NR_getpid, sigsys->_syscall); 939 /* Make sure arch is non-zero. */ 940 EXPECT_NE(0, sigsys->_arch); 941 EXPECT_NE(0, (unsigned long)sigsys->_call_addr); 942 } 943 944 FIXTURE_DATA(precedence) { 945 struct sock_fprog allow; 946 struct sock_fprog log; 947 struct sock_fprog trace; 948 struct sock_fprog error; 949 struct sock_fprog trap; 950 struct sock_fprog kill; 951 }; 952 953 FIXTURE_SETUP(precedence) 954 { 955 struct sock_filter allow_insns[] = { 956 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 957 }; 958 struct sock_filter log_insns[] = { 959 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 960 offsetof(struct seccomp_data, nr)), 961 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 962 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 963 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG), 964 }; 965 struct sock_filter trace_insns[] = { 966 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 967 offsetof(struct seccomp_data, nr)), 968 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 969 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 970 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE), 971 }; 972 struct sock_filter error_insns[] = { 973 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 974 offsetof(struct seccomp_data, nr)), 975 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 976 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 977 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO), 978 }; 979 struct sock_filter trap_insns[] = { 980 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 981 offsetof(struct seccomp_data, nr)), 982 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 983 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 984 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP), 985 }; 986 struct sock_filter kill_insns[] = { 987 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 988 offsetof(struct seccomp_data, nr)), 989 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 990 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 991 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 992 }; 993 994 memset(self, 0, sizeof(*self)); 995 #define FILTER_ALLOC(_x) \ 996 self->_x.filter = malloc(sizeof(_x##_insns)); \ 997 ASSERT_NE(NULL, self->_x.filter); \ 998 memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \ 999 self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns) 1000 FILTER_ALLOC(allow); 1001 FILTER_ALLOC(log); 1002 FILTER_ALLOC(trace); 1003 FILTER_ALLOC(error); 1004 FILTER_ALLOC(trap); 1005 FILTER_ALLOC(kill); 1006 } 1007 1008 FIXTURE_TEARDOWN(precedence) 1009 { 1010 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter) 1011 FILTER_FREE(allow); 1012 FILTER_FREE(log); 1013 FILTER_FREE(trace); 1014 FILTER_FREE(error); 1015 FILTER_FREE(trap); 1016 FILTER_FREE(kill); 1017 } 1018 1019 TEST_F(precedence, allow_ok) 1020 { 1021 pid_t parent, res = 0; 1022 long ret; 1023 1024 parent = getppid(); 1025 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1026 ASSERT_EQ(0, ret); 1027 1028 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1029 ASSERT_EQ(0, ret); 1030 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1031 ASSERT_EQ(0, ret); 1032 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1033 ASSERT_EQ(0, ret); 1034 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1035 ASSERT_EQ(0, ret); 1036 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1037 ASSERT_EQ(0, ret); 1038 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1039 ASSERT_EQ(0, ret); 1040 /* Should work just fine. */ 1041 res = syscall(__NR_getppid); 1042 EXPECT_EQ(parent, res); 1043 } 1044 1045 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS) 1046 { 1047 pid_t parent, res = 0; 1048 long ret; 1049 1050 parent = getppid(); 1051 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1052 ASSERT_EQ(0, ret); 1053 1054 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1055 ASSERT_EQ(0, ret); 1056 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1057 ASSERT_EQ(0, ret); 1058 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1059 ASSERT_EQ(0, ret); 1060 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1061 ASSERT_EQ(0, ret); 1062 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1063 ASSERT_EQ(0, ret); 1064 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1065 ASSERT_EQ(0, ret); 1066 /* Should work just fine. */ 1067 res = syscall(__NR_getppid); 1068 EXPECT_EQ(parent, res); 1069 /* getpid() should never return. */ 1070 res = syscall(__NR_getpid); 1071 EXPECT_EQ(0, res); 1072 } 1073 1074 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS) 1075 { 1076 pid_t parent; 1077 long ret; 1078 1079 parent = getppid(); 1080 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1081 ASSERT_EQ(0, ret); 1082 1083 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1084 ASSERT_EQ(0, ret); 1085 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1086 ASSERT_EQ(0, ret); 1087 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1088 ASSERT_EQ(0, ret); 1089 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1090 ASSERT_EQ(0, ret); 1091 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1092 ASSERT_EQ(0, ret); 1093 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1094 ASSERT_EQ(0, ret); 1095 /* Should work just fine. */ 1096 EXPECT_EQ(parent, syscall(__NR_getppid)); 1097 /* getpid() should never return. */ 1098 EXPECT_EQ(0, syscall(__NR_getpid)); 1099 } 1100 1101 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS) 1102 { 1103 pid_t parent; 1104 long ret; 1105 1106 parent = getppid(); 1107 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1108 ASSERT_EQ(0, ret); 1109 1110 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1111 ASSERT_EQ(0, ret); 1112 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1113 ASSERT_EQ(0, ret); 1114 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1115 ASSERT_EQ(0, ret); 1116 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1117 ASSERT_EQ(0, ret); 1118 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1119 ASSERT_EQ(0, ret); 1120 /* Should work just fine. */ 1121 EXPECT_EQ(parent, syscall(__NR_getppid)); 1122 /* getpid() should never return. */ 1123 EXPECT_EQ(0, syscall(__NR_getpid)); 1124 } 1125 1126 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS) 1127 { 1128 pid_t parent; 1129 long ret; 1130 1131 parent = getppid(); 1132 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1133 ASSERT_EQ(0, ret); 1134 1135 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1136 ASSERT_EQ(0, ret); 1137 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1138 ASSERT_EQ(0, ret); 1139 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1140 ASSERT_EQ(0, ret); 1141 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1142 ASSERT_EQ(0, ret); 1143 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1144 ASSERT_EQ(0, ret); 1145 /* Should work just fine. */ 1146 EXPECT_EQ(parent, syscall(__NR_getppid)); 1147 /* getpid() should never return. */ 1148 EXPECT_EQ(0, syscall(__NR_getpid)); 1149 } 1150 1151 TEST_F(precedence, errno_is_third) 1152 { 1153 pid_t parent; 1154 long ret; 1155 1156 parent = getppid(); 1157 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1158 ASSERT_EQ(0, ret); 1159 1160 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1161 ASSERT_EQ(0, ret); 1162 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1163 ASSERT_EQ(0, ret); 1164 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1165 ASSERT_EQ(0, ret); 1166 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1167 ASSERT_EQ(0, ret); 1168 /* Should work just fine. */ 1169 EXPECT_EQ(parent, syscall(__NR_getppid)); 1170 EXPECT_EQ(0, syscall(__NR_getpid)); 1171 } 1172 1173 TEST_F(precedence, errno_is_third_in_any_order) 1174 { 1175 pid_t parent; 1176 long ret; 1177 1178 parent = getppid(); 1179 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1180 ASSERT_EQ(0, ret); 1181 1182 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1183 ASSERT_EQ(0, ret); 1184 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1185 ASSERT_EQ(0, ret); 1186 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1187 ASSERT_EQ(0, ret); 1188 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1189 ASSERT_EQ(0, ret); 1190 /* Should work just fine. */ 1191 EXPECT_EQ(parent, syscall(__NR_getppid)); 1192 EXPECT_EQ(0, syscall(__NR_getpid)); 1193 } 1194 1195 TEST_F(precedence, trace_is_fourth) 1196 { 1197 pid_t parent; 1198 long ret; 1199 1200 parent = getppid(); 1201 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1202 ASSERT_EQ(0, ret); 1203 1204 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1205 ASSERT_EQ(0, ret); 1206 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1207 ASSERT_EQ(0, ret); 1208 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1209 ASSERT_EQ(0, ret); 1210 /* Should work just fine. */ 1211 EXPECT_EQ(parent, syscall(__NR_getppid)); 1212 /* No ptracer */ 1213 EXPECT_EQ(-1, syscall(__NR_getpid)); 1214 } 1215 1216 TEST_F(precedence, trace_is_fourth_in_any_order) 1217 { 1218 pid_t parent; 1219 long ret; 1220 1221 parent = getppid(); 1222 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1223 ASSERT_EQ(0, ret); 1224 1225 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1226 ASSERT_EQ(0, ret); 1227 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1228 ASSERT_EQ(0, ret); 1229 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1230 ASSERT_EQ(0, ret); 1231 /* Should work just fine. */ 1232 EXPECT_EQ(parent, syscall(__NR_getppid)); 1233 /* No ptracer */ 1234 EXPECT_EQ(-1, syscall(__NR_getpid)); 1235 } 1236 1237 TEST_F(precedence, log_is_fifth) 1238 { 1239 pid_t mypid, parent; 1240 long ret; 1241 1242 mypid = getpid(); 1243 parent = getppid(); 1244 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1245 ASSERT_EQ(0, ret); 1246 1247 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1248 ASSERT_EQ(0, ret); 1249 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1250 ASSERT_EQ(0, ret); 1251 /* Should work just fine. */ 1252 EXPECT_EQ(parent, syscall(__NR_getppid)); 1253 /* Should also work just fine */ 1254 EXPECT_EQ(mypid, syscall(__NR_getpid)); 1255 } 1256 1257 TEST_F(precedence, log_is_fifth_in_any_order) 1258 { 1259 pid_t mypid, parent; 1260 long ret; 1261 1262 mypid = getpid(); 1263 parent = getppid(); 1264 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1265 ASSERT_EQ(0, ret); 1266 1267 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1268 ASSERT_EQ(0, ret); 1269 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1270 ASSERT_EQ(0, ret); 1271 /* Should work just fine. */ 1272 EXPECT_EQ(parent, syscall(__NR_getppid)); 1273 /* Should also work just fine */ 1274 EXPECT_EQ(mypid, syscall(__NR_getpid)); 1275 } 1276 1277 #ifndef PTRACE_O_TRACESECCOMP 1278 #define PTRACE_O_TRACESECCOMP 0x00000080 1279 #endif 1280 1281 /* Catch the Ubuntu 12.04 value error. */ 1282 #if PTRACE_EVENT_SECCOMP != 7 1283 #undef PTRACE_EVENT_SECCOMP 1284 #endif 1285 1286 #ifndef PTRACE_EVENT_SECCOMP 1287 #define PTRACE_EVENT_SECCOMP 7 1288 #endif 1289 1290 #define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP) 1291 bool tracer_running; 1292 void tracer_stop(int sig) 1293 { 1294 tracer_running = false; 1295 } 1296 1297 typedef void tracer_func_t(struct __test_metadata *_metadata, 1298 pid_t tracee, int status, void *args); 1299 1300 void start_tracer(struct __test_metadata *_metadata, int fd, pid_t tracee, 1301 tracer_func_t tracer_func, void *args, bool ptrace_syscall) 1302 { 1303 int ret = -1; 1304 struct sigaction action = { 1305 .sa_handler = tracer_stop, 1306 }; 1307 1308 /* Allow external shutdown. */ 1309 tracer_running = true; 1310 ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL)); 1311 1312 errno = 0; 1313 while (ret == -1 && errno != EINVAL) 1314 ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0); 1315 ASSERT_EQ(0, ret) { 1316 kill(tracee, SIGKILL); 1317 } 1318 /* Wait for attach stop */ 1319 wait(NULL); 1320 1321 ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, ptrace_syscall ? 1322 PTRACE_O_TRACESYSGOOD : 1323 PTRACE_O_TRACESECCOMP); 1324 ASSERT_EQ(0, ret) { 1325 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP"); 1326 kill(tracee, SIGKILL); 1327 } 1328 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT, 1329 tracee, NULL, 0); 1330 ASSERT_EQ(0, ret); 1331 1332 /* Unblock the tracee */ 1333 ASSERT_EQ(1, write(fd, "A", 1)); 1334 ASSERT_EQ(0, close(fd)); 1335 1336 /* Run until we're shut down. Must assert to stop execution. */ 1337 while (tracer_running) { 1338 int status; 1339 1340 if (wait(&status) != tracee) 1341 continue; 1342 if (WIFSIGNALED(status) || WIFEXITED(status)) 1343 /* Child is dead. Time to go. */ 1344 return; 1345 1346 /* Check if this is a seccomp event. */ 1347 ASSERT_EQ(!ptrace_syscall, IS_SECCOMP_EVENT(status)); 1348 1349 tracer_func(_metadata, tracee, status, args); 1350 1351 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT, 1352 tracee, NULL, 0); 1353 ASSERT_EQ(0, ret); 1354 } 1355 /* Directly report the status of our test harness results. */ 1356 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS : EXIT_FAILURE); 1357 } 1358 1359 /* Common tracer setup/teardown functions. */ 1360 void cont_handler(int num) 1361 { } 1362 pid_t setup_trace_fixture(struct __test_metadata *_metadata, 1363 tracer_func_t func, void *args, bool ptrace_syscall) 1364 { 1365 char sync; 1366 int pipefd[2]; 1367 pid_t tracer_pid; 1368 pid_t tracee = getpid(); 1369 1370 /* Setup a pipe for clean synchronization. */ 1371 ASSERT_EQ(0, pipe(pipefd)); 1372 1373 /* Fork a child which we'll promote to tracer */ 1374 tracer_pid = fork(); 1375 ASSERT_LE(0, tracer_pid); 1376 signal(SIGALRM, cont_handler); 1377 if (tracer_pid == 0) { 1378 close(pipefd[0]); 1379 start_tracer(_metadata, pipefd[1], tracee, func, args, 1380 ptrace_syscall); 1381 syscall(__NR_exit, 0); 1382 } 1383 close(pipefd[1]); 1384 prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0); 1385 read(pipefd[0], &sync, 1); 1386 close(pipefd[0]); 1387 1388 return tracer_pid; 1389 } 1390 void teardown_trace_fixture(struct __test_metadata *_metadata, 1391 pid_t tracer) 1392 { 1393 if (tracer) { 1394 int status; 1395 /* 1396 * Extract the exit code from the other process and 1397 * adopt it for ourselves in case its asserts failed. 1398 */ 1399 ASSERT_EQ(0, kill(tracer, SIGUSR1)); 1400 ASSERT_EQ(tracer, waitpid(tracer, &status, 0)); 1401 if (WEXITSTATUS(status)) 1402 _metadata->passed = 0; 1403 } 1404 } 1405 1406 /* "poke" tracer arguments and function. */ 1407 struct tracer_args_poke_t { 1408 unsigned long poke_addr; 1409 }; 1410 1411 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status, 1412 void *args) 1413 { 1414 int ret; 1415 unsigned long msg; 1416 struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args; 1417 1418 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 1419 EXPECT_EQ(0, ret); 1420 /* If this fails, don't try to recover. */ 1421 ASSERT_EQ(0x1001, msg) { 1422 kill(tracee, SIGKILL); 1423 } 1424 /* 1425 * Poke in the message. 1426 * Registers are not touched to try to keep this relatively arch 1427 * agnostic. 1428 */ 1429 ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001); 1430 EXPECT_EQ(0, ret); 1431 } 1432 1433 FIXTURE_DATA(TRACE_poke) { 1434 struct sock_fprog prog; 1435 pid_t tracer; 1436 long poked; 1437 struct tracer_args_poke_t tracer_args; 1438 }; 1439 1440 FIXTURE_SETUP(TRACE_poke) 1441 { 1442 struct sock_filter filter[] = { 1443 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1444 offsetof(struct seccomp_data, nr)), 1445 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 1446 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001), 1447 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1448 }; 1449 1450 self->poked = 0; 1451 memset(&self->prog, 0, sizeof(self->prog)); 1452 self->prog.filter = malloc(sizeof(filter)); 1453 ASSERT_NE(NULL, self->prog.filter); 1454 memcpy(self->prog.filter, filter, sizeof(filter)); 1455 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 1456 1457 /* Set up tracer args. */ 1458 self->tracer_args.poke_addr = (unsigned long)&self->poked; 1459 1460 /* Launch tracer. */ 1461 self->tracer = setup_trace_fixture(_metadata, tracer_poke, 1462 &self->tracer_args, false); 1463 } 1464 1465 FIXTURE_TEARDOWN(TRACE_poke) 1466 { 1467 teardown_trace_fixture(_metadata, self->tracer); 1468 if (self->prog.filter) 1469 free(self->prog.filter); 1470 } 1471 1472 TEST_F(TRACE_poke, read_has_side_effects) 1473 { 1474 ssize_t ret; 1475 1476 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1477 ASSERT_EQ(0, ret); 1478 1479 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1480 ASSERT_EQ(0, ret); 1481 1482 EXPECT_EQ(0, self->poked); 1483 ret = read(-1, NULL, 0); 1484 EXPECT_EQ(-1, ret); 1485 EXPECT_EQ(0x1001, self->poked); 1486 } 1487 1488 TEST_F(TRACE_poke, getpid_runs_normally) 1489 { 1490 long ret; 1491 1492 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1493 ASSERT_EQ(0, ret); 1494 1495 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1496 ASSERT_EQ(0, ret); 1497 1498 EXPECT_EQ(0, self->poked); 1499 EXPECT_NE(0, syscall(__NR_getpid)); 1500 EXPECT_EQ(0, self->poked); 1501 } 1502 1503 #if defined(__x86_64__) 1504 # define ARCH_REGS struct user_regs_struct 1505 # define SYSCALL_NUM orig_rax 1506 # define SYSCALL_RET rax 1507 #elif defined(__i386__) 1508 # define ARCH_REGS struct user_regs_struct 1509 # define SYSCALL_NUM orig_eax 1510 # define SYSCALL_RET eax 1511 #elif defined(__arm__) 1512 # define ARCH_REGS struct pt_regs 1513 # define SYSCALL_NUM ARM_r7 1514 # define SYSCALL_RET ARM_r0 1515 #elif defined(__aarch64__) 1516 # define ARCH_REGS struct user_pt_regs 1517 # define SYSCALL_NUM regs[8] 1518 # define SYSCALL_RET regs[0] 1519 #elif defined(__hppa__) 1520 # define ARCH_REGS struct user_regs_struct 1521 # define SYSCALL_NUM gr[20] 1522 # define SYSCALL_RET gr[28] 1523 #elif defined(__powerpc__) 1524 # define ARCH_REGS struct pt_regs 1525 # define SYSCALL_NUM gpr[0] 1526 # define SYSCALL_RET gpr[3] 1527 #elif defined(__s390__) 1528 # define ARCH_REGS s390_regs 1529 # define SYSCALL_NUM gprs[2] 1530 # define SYSCALL_RET gprs[2] 1531 #elif defined(__mips__) 1532 # define ARCH_REGS struct pt_regs 1533 # define SYSCALL_NUM regs[2] 1534 # define SYSCALL_SYSCALL_NUM regs[4] 1535 # define SYSCALL_RET regs[2] 1536 # define SYSCALL_NUM_RET_SHARE_REG 1537 #else 1538 # error "Do not know how to find your architecture's registers and syscalls" 1539 #endif 1540 1541 /* When the syscall return can't be changed, stub out the tests for it. */ 1542 #ifdef SYSCALL_NUM_RET_SHARE_REG 1543 # define EXPECT_SYSCALL_RETURN(val, action) EXPECT_EQ(-1, action) 1544 #else 1545 # define EXPECT_SYSCALL_RETURN(val, action) EXPECT_EQ(val, action) 1546 #endif 1547 1548 /* Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for 1549 * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux). 1550 */ 1551 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__) 1552 #define HAVE_GETREGS 1553 #endif 1554 1555 /* Architecture-specific syscall fetching routine. */ 1556 int get_syscall(struct __test_metadata *_metadata, pid_t tracee) 1557 { 1558 ARCH_REGS regs; 1559 #ifdef HAVE_GETREGS 1560 EXPECT_EQ(0, ptrace(PTRACE_GETREGS, tracee, 0, ®s)) { 1561 TH_LOG("PTRACE_GETREGS failed"); 1562 return -1; 1563 } 1564 #else 1565 struct iovec iov; 1566 1567 iov.iov_base = ®s; 1568 iov.iov_len = sizeof(regs); 1569 EXPECT_EQ(0, ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov)) { 1570 TH_LOG("PTRACE_GETREGSET failed"); 1571 return -1; 1572 } 1573 #endif 1574 1575 #if defined(__mips__) 1576 if (regs.SYSCALL_NUM == __NR_O32_Linux) 1577 return regs.SYSCALL_SYSCALL_NUM; 1578 #endif 1579 return regs.SYSCALL_NUM; 1580 } 1581 1582 /* Architecture-specific syscall changing routine. */ 1583 void change_syscall(struct __test_metadata *_metadata, 1584 pid_t tracee, int syscall) 1585 { 1586 int ret; 1587 ARCH_REGS regs; 1588 #ifdef HAVE_GETREGS 1589 ret = ptrace(PTRACE_GETREGS, tracee, 0, ®s); 1590 #else 1591 struct iovec iov; 1592 iov.iov_base = ®s; 1593 iov.iov_len = sizeof(regs); 1594 ret = ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov); 1595 #endif 1596 EXPECT_EQ(0, ret) {} 1597 1598 #if defined(__x86_64__) || defined(__i386__) || defined(__powerpc__) || \ 1599 defined(__s390__) || defined(__hppa__) 1600 { 1601 regs.SYSCALL_NUM = syscall; 1602 } 1603 #elif defined(__mips__) 1604 { 1605 if (regs.SYSCALL_NUM == __NR_O32_Linux) 1606 regs.SYSCALL_SYSCALL_NUM = syscall; 1607 else 1608 regs.SYSCALL_NUM = syscall; 1609 } 1610 1611 #elif defined(__arm__) 1612 # ifndef PTRACE_SET_SYSCALL 1613 # define PTRACE_SET_SYSCALL 23 1614 # endif 1615 { 1616 ret = ptrace(PTRACE_SET_SYSCALL, tracee, NULL, syscall); 1617 EXPECT_EQ(0, ret); 1618 } 1619 1620 #elif defined(__aarch64__) 1621 # ifndef NT_ARM_SYSTEM_CALL 1622 # define NT_ARM_SYSTEM_CALL 0x404 1623 # endif 1624 { 1625 iov.iov_base = &syscall; 1626 iov.iov_len = sizeof(syscall); 1627 ret = ptrace(PTRACE_SETREGSET, tracee, NT_ARM_SYSTEM_CALL, 1628 &iov); 1629 EXPECT_EQ(0, ret); 1630 } 1631 1632 #else 1633 ASSERT_EQ(1, 0) { 1634 TH_LOG("How is the syscall changed on this architecture?"); 1635 } 1636 #endif 1637 1638 /* If syscall is skipped, change return value. */ 1639 if (syscall == -1) 1640 #ifdef SYSCALL_NUM_RET_SHARE_REG 1641 TH_LOG("Can't modify syscall return on this architecture"); 1642 #else 1643 regs.SYSCALL_RET = EPERM; 1644 #endif 1645 1646 #ifdef HAVE_GETREGS 1647 ret = ptrace(PTRACE_SETREGS, tracee, 0, ®s); 1648 #else 1649 iov.iov_base = ®s; 1650 iov.iov_len = sizeof(regs); 1651 ret = ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &iov); 1652 #endif 1653 EXPECT_EQ(0, ret); 1654 } 1655 1656 void tracer_syscall(struct __test_metadata *_metadata, pid_t tracee, 1657 int status, void *args) 1658 { 1659 int ret; 1660 unsigned long msg; 1661 1662 /* Make sure we got the right message. */ 1663 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 1664 EXPECT_EQ(0, ret); 1665 1666 /* Validate and take action on expected syscalls. */ 1667 switch (msg) { 1668 case 0x1002: 1669 /* change getpid to getppid. */ 1670 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee)); 1671 change_syscall(_metadata, tracee, __NR_getppid); 1672 break; 1673 case 0x1003: 1674 /* skip gettid. */ 1675 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee)); 1676 change_syscall(_metadata, tracee, -1); 1677 break; 1678 case 0x1004: 1679 /* do nothing (allow getppid) */ 1680 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee)); 1681 break; 1682 default: 1683 EXPECT_EQ(0, msg) { 1684 TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg); 1685 kill(tracee, SIGKILL); 1686 } 1687 } 1688 1689 } 1690 1691 void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee, 1692 int status, void *args) 1693 { 1694 int ret, nr; 1695 unsigned long msg; 1696 static bool entry; 1697 1698 /* Make sure we got an empty message. */ 1699 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 1700 EXPECT_EQ(0, ret); 1701 EXPECT_EQ(0, msg); 1702 1703 /* The only way to tell PTRACE_SYSCALL entry/exit is by counting. */ 1704 entry = !entry; 1705 if (!entry) 1706 return; 1707 1708 nr = get_syscall(_metadata, tracee); 1709 1710 if (nr == __NR_getpid) 1711 change_syscall(_metadata, tracee, __NR_getppid); 1712 if (nr == __NR_open) 1713 change_syscall(_metadata, tracee, -1); 1714 } 1715 1716 FIXTURE_DATA(TRACE_syscall) { 1717 struct sock_fprog prog; 1718 pid_t tracer, mytid, mypid, parent; 1719 }; 1720 1721 FIXTURE_SETUP(TRACE_syscall) 1722 { 1723 struct sock_filter filter[] = { 1724 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1725 offsetof(struct seccomp_data, nr)), 1726 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 1727 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002), 1728 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1), 1729 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003), 1730 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 1731 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004), 1732 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1733 }; 1734 1735 memset(&self->prog, 0, sizeof(self->prog)); 1736 self->prog.filter = malloc(sizeof(filter)); 1737 ASSERT_NE(NULL, self->prog.filter); 1738 memcpy(self->prog.filter, filter, sizeof(filter)); 1739 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 1740 1741 /* Prepare some testable syscall results. */ 1742 self->mytid = syscall(__NR_gettid); 1743 ASSERT_GT(self->mytid, 0); 1744 ASSERT_NE(self->mytid, 1) { 1745 TH_LOG("Running this test as init is not supported. :)"); 1746 } 1747 1748 self->mypid = getpid(); 1749 ASSERT_GT(self->mypid, 0); 1750 ASSERT_EQ(self->mytid, self->mypid); 1751 1752 self->parent = getppid(); 1753 ASSERT_GT(self->parent, 0); 1754 ASSERT_NE(self->parent, self->mypid); 1755 1756 /* Launch tracer. */ 1757 self->tracer = setup_trace_fixture(_metadata, tracer_syscall, NULL, 1758 false); 1759 } 1760 1761 FIXTURE_TEARDOWN(TRACE_syscall) 1762 { 1763 teardown_trace_fixture(_metadata, self->tracer); 1764 if (self->prog.filter) 1765 free(self->prog.filter); 1766 } 1767 1768 TEST_F(TRACE_syscall, ptrace_syscall_redirected) 1769 { 1770 /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */ 1771 teardown_trace_fixture(_metadata, self->tracer); 1772 self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL, 1773 true); 1774 1775 /* Tracer will redirect getpid to getppid. */ 1776 EXPECT_NE(self->mypid, syscall(__NR_getpid)); 1777 } 1778 1779 TEST_F(TRACE_syscall, ptrace_syscall_dropped) 1780 { 1781 /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */ 1782 teardown_trace_fixture(_metadata, self->tracer); 1783 self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL, 1784 true); 1785 1786 /* Tracer should skip the open syscall, resulting in EPERM. */ 1787 EXPECT_SYSCALL_RETURN(EPERM, syscall(__NR_open)); 1788 } 1789 1790 TEST_F(TRACE_syscall, syscall_allowed) 1791 { 1792 long ret; 1793 1794 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1795 ASSERT_EQ(0, ret); 1796 1797 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1798 ASSERT_EQ(0, ret); 1799 1800 /* getppid works as expected (no changes). */ 1801 EXPECT_EQ(self->parent, syscall(__NR_getppid)); 1802 EXPECT_NE(self->mypid, syscall(__NR_getppid)); 1803 } 1804 1805 TEST_F(TRACE_syscall, syscall_redirected) 1806 { 1807 long ret; 1808 1809 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1810 ASSERT_EQ(0, ret); 1811 1812 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1813 ASSERT_EQ(0, ret); 1814 1815 /* getpid has been redirected to getppid as expected. */ 1816 EXPECT_EQ(self->parent, syscall(__NR_getpid)); 1817 EXPECT_NE(self->mypid, syscall(__NR_getpid)); 1818 } 1819 1820 TEST_F(TRACE_syscall, syscall_dropped) 1821 { 1822 long ret; 1823 1824 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1825 ASSERT_EQ(0, ret); 1826 1827 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1828 ASSERT_EQ(0, ret); 1829 1830 /* gettid has been skipped and an altered return value stored. */ 1831 EXPECT_SYSCALL_RETURN(EPERM, syscall(__NR_gettid)); 1832 EXPECT_NE(self->mytid, syscall(__NR_gettid)); 1833 } 1834 1835 TEST_F(TRACE_syscall, skip_after_RET_TRACE) 1836 { 1837 struct sock_filter filter[] = { 1838 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1839 offsetof(struct seccomp_data, nr)), 1840 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 1841 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM), 1842 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1843 }; 1844 struct sock_fprog prog = { 1845 .len = (unsigned short)ARRAY_SIZE(filter), 1846 .filter = filter, 1847 }; 1848 long ret; 1849 1850 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1851 ASSERT_EQ(0, ret); 1852 1853 /* Install fixture filter. */ 1854 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1855 ASSERT_EQ(0, ret); 1856 1857 /* Install "errno on getppid" filter. */ 1858 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 1859 ASSERT_EQ(0, ret); 1860 1861 /* Tracer will redirect getpid to getppid, and we should see EPERM. */ 1862 errno = 0; 1863 EXPECT_EQ(-1, syscall(__NR_getpid)); 1864 EXPECT_EQ(EPERM, errno); 1865 } 1866 1867 TEST_F_SIGNAL(TRACE_syscall, kill_after_RET_TRACE, SIGSYS) 1868 { 1869 struct sock_filter filter[] = { 1870 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1871 offsetof(struct seccomp_data, nr)), 1872 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 1873 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 1874 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1875 }; 1876 struct sock_fprog prog = { 1877 .len = (unsigned short)ARRAY_SIZE(filter), 1878 .filter = filter, 1879 }; 1880 long ret; 1881 1882 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1883 ASSERT_EQ(0, ret); 1884 1885 /* Install fixture filter. */ 1886 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1887 ASSERT_EQ(0, ret); 1888 1889 /* Install "death on getppid" filter. */ 1890 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 1891 ASSERT_EQ(0, ret); 1892 1893 /* Tracer will redirect getpid to getppid, and we should die. */ 1894 EXPECT_NE(self->mypid, syscall(__NR_getpid)); 1895 } 1896 1897 TEST_F(TRACE_syscall, skip_after_ptrace) 1898 { 1899 struct sock_filter filter[] = { 1900 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1901 offsetof(struct seccomp_data, nr)), 1902 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 1903 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM), 1904 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1905 }; 1906 struct sock_fprog prog = { 1907 .len = (unsigned short)ARRAY_SIZE(filter), 1908 .filter = filter, 1909 }; 1910 long ret; 1911 1912 /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */ 1913 teardown_trace_fixture(_metadata, self->tracer); 1914 self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL, 1915 true); 1916 1917 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1918 ASSERT_EQ(0, ret); 1919 1920 /* Install "errno on getppid" filter. */ 1921 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 1922 ASSERT_EQ(0, ret); 1923 1924 /* Tracer will redirect getpid to getppid, and we should see EPERM. */ 1925 EXPECT_EQ(-1, syscall(__NR_getpid)); 1926 EXPECT_EQ(EPERM, errno); 1927 } 1928 1929 TEST_F_SIGNAL(TRACE_syscall, kill_after_ptrace, SIGSYS) 1930 { 1931 struct sock_filter filter[] = { 1932 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1933 offsetof(struct seccomp_data, nr)), 1934 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 1935 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 1936 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1937 }; 1938 struct sock_fprog prog = { 1939 .len = (unsigned short)ARRAY_SIZE(filter), 1940 .filter = filter, 1941 }; 1942 long ret; 1943 1944 /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */ 1945 teardown_trace_fixture(_metadata, self->tracer); 1946 self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL, 1947 true); 1948 1949 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1950 ASSERT_EQ(0, ret); 1951 1952 /* Install "death on getppid" filter. */ 1953 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 1954 ASSERT_EQ(0, ret); 1955 1956 /* Tracer will redirect getpid to getppid, and we should die. */ 1957 EXPECT_NE(self->mypid, syscall(__NR_getpid)); 1958 } 1959 1960 TEST(seccomp_syscall) 1961 { 1962 struct sock_filter filter[] = { 1963 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1964 }; 1965 struct sock_fprog prog = { 1966 .len = (unsigned short)ARRAY_SIZE(filter), 1967 .filter = filter, 1968 }; 1969 long ret; 1970 1971 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1972 ASSERT_EQ(0, ret) { 1973 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 1974 } 1975 1976 /* Reject insane operation. */ 1977 ret = seccomp(-1, 0, &prog); 1978 ASSERT_NE(ENOSYS, errno) { 1979 TH_LOG("Kernel does not support seccomp syscall!"); 1980 } 1981 EXPECT_EQ(EINVAL, errno) { 1982 TH_LOG("Did not reject crazy op value!"); 1983 } 1984 1985 /* Reject strict with flags or pointer. */ 1986 ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL); 1987 EXPECT_EQ(EINVAL, errno) { 1988 TH_LOG("Did not reject mode strict with flags!"); 1989 } 1990 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog); 1991 EXPECT_EQ(EINVAL, errno) { 1992 TH_LOG("Did not reject mode strict with uargs!"); 1993 } 1994 1995 /* Reject insane args for filter. */ 1996 ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog); 1997 EXPECT_EQ(EINVAL, errno) { 1998 TH_LOG("Did not reject crazy filter flags!"); 1999 } 2000 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL); 2001 EXPECT_EQ(EFAULT, errno) { 2002 TH_LOG("Did not reject NULL filter!"); 2003 } 2004 2005 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2006 EXPECT_EQ(0, errno) { 2007 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s", 2008 strerror(errno)); 2009 } 2010 } 2011 2012 TEST(seccomp_syscall_mode_lock) 2013 { 2014 struct sock_filter filter[] = { 2015 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2016 }; 2017 struct sock_fprog prog = { 2018 .len = (unsigned short)ARRAY_SIZE(filter), 2019 .filter = filter, 2020 }; 2021 long ret; 2022 2023 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 2024 ASSERT_EQ(0, ret) { 2025 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2026 } 2027 2028 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2029 ASSERT_NE(ENOSYS, errno) { 2030 TH_LOG("Kernel does not support seccomp syscall!"); 2031 } 2032 EXPECT_EQ(0, ret) { 2033 TH_LOG("Could not install filter!"); 2034 } 2035 2036 /* Make sure neither entry point will switch to strict. */ 2037 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0); 2038 EXPECT_EQ(EINVAL, errno) { 2039 TH_LOG("Switched to mode strict!"); 2040 } 2041 2042 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL); 2043 EXPECT_EQ(EINVAL, errno) { 2044 TH_LOG("Switched to mode strict!"); 2045 } 2046 } 2047 2048 /* 2049 * Test detection of known and unknown filter flags. Userspace needs to be able 2050 * to check if a filter flag is supported by the current kernel and a good way 2051 * of doing that is by attempting to enter filter mode, with the flag bit in 2052 * question set, and a NULL pointer for the _args_ parameter. EFAULT indicates 2053 * that the flag is valid and EINVAL indicates that the flag is invalid. 2054 */ 2055 TEST(detect_seccomp_filter_flags) 2056 { 2057 unsigned int flags[] = { SECCOMP_FILTER_FLAG_TSYNC, 2058 SECCOMP_FILTER_FLAG_LOG }; 2059 unsigned int flag, all_flags; 2060 int i; 2061 long ret; 2062 2063 /* Test detection of known-good filter flags */ 2064 for (i = 0, all_flags = 0; i < ARRAY_SIZE(flags); i++) { 2065 flag = flags[i]; 2066 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2067 ASSERT_NE(ENOSYS, errno) { 2068 TH_LOG("Kernel does not support seccomp syscall!"); 2069 } 2070 EXPECT_EQ(-1, ret); 2071 EXPECT_EQ(EFAULT, errno) { 2072 TH_LOG("Failed to detect that a known-good filter flag (0x%X) is supported!", 2073 flag); 2074 } 2075 2076 all_flags |= flag; 2077 } 2078 2079 /* Test detection of all known-good filter flags */ 2080 ret = seccomp(SECCOMP_SET_MODE_FILTER, all_flags, NULL); 2081 EXPECT_EQ(-1, ret); 2082 EXPECT_EQ(EFAULT, errno) { 2083 TH_LOG("Failed to detect that all known-good filter flags (0x%X) are supported!", 2084 all_flags); 2085 } 2086 2087 /* Test detection of an unknown filter flag */ 2088 flag = -1; 2089 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2090 EXPECT_EQ(-1, ret); 2091 EXPECT_EQ(EINVAL, errno) { 2092 TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported!", 2093 flag); 2094 } 2095 2096 /* 2097 * Test detection of an unknown filter flag that may simply need to be 2098 * added to this test 2099 */ 2100 flag = flags[ARRAY_SIZE(flags) - 1] << 1; 2101 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2102 EXPECT_EQ(-1, ret); 2103 EXPECT_EQ(EINVAL, errno) { 2104 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?", 2105 flag); 2106 } 2107 } 2108 2109 TEST(TSYNC_first) 2110 { 2111 struct sock_filter filter[] = { 2112 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2113 }; 2114 struct sock_fprog prog = { 2115 .len = (unsigned short)ARRAY_SIZE(filter), 2116 .filter = filter, 2117 }; 2118 long ret; 2119 2120 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 2121 ASSERT_EQ(0, ret) { 2122 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2123 } 2124 2125 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2126 &prog); 2127 ASSERT_NE(ENOSYS, errno) { 2128 TH_LOG("Kernel does not support seccomp syscall!"); 2129 } 2130 EXPECT_EQ(0, ret) { 2131 TH_LOG("Could not install initial filter with TSYNC!"); 2132 } 2133 } 2134 2135 #define TSYNC_SIBLINGS 2 2136 struct tsync_sibling { 2137 pthread_t tid; 2138 pid_t system_tid; 2139 sem_t *started; 2140 pthread_cond_t *cond; 2141 pthread_mutex_t *mutex; 2142 int diverge; 2143 int num_waits; 2144 struct sock_fprog *prog; 2145 struct __test_metadata *metadata; 2146 }; 2147 2148 /* 2149 * To avoid joining joined threads (which is not allowed by Bionic), 2150 * make sure we both successfully join and clear the tid to skip a 2151 * later join attempt during fixture teardown. Any remaining threads 2152 * will be directly killed during teardown. 2153 */ 2154 #define PTHREAD_JOIN(tid, status) \ 2155 do { \ 2156 int _rc = pthread_join(tid, status); \ 2157 if (_rc) { \ 2158 TH_LOG("pthread_join of tid %u failed: %d\n", \ 2159 (unsigned int)tid, _rc); \ 2160 } else { \ 2161 tid = 0; \ 2162 } \ 2163 } while (0) 2164 2165 FIXTURE_DATA(TSYNC) { 2166 struct sock_fprog root_prog, apply_prog; 2167 struct tsync_sibling sibling[TSYNC_SIBLINGS]; 2168 sem_t started; 2169 pthread_cond_t cond; 2170 pthread_mutex_t mutex; 2171 int sibling_count; 2172 }; 2173 2174 FIXTURE_SETUP(TSYNC) 2175 { 2176 struct sock_filter root_filter[] = { 2177 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2178 }; 2179 struct sock_filter apply_filter[] = { 2180 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2181 offsetof(struct seccomp_data, nr)), 2182 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 2183 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2184 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2185 }; 2186 2187 memset(&self->root_prog, 0, sizeof(self->root_prog)); 2188 memset(&self->apply_prog, 0, sizeof(self->apply_prog)); 2189 memset(&self->sibling, 0, sizeof(self->sibling)); 2190 self->root_prog.filter = malloc(sizeof(root_filter)); 2191 ASSERT_NE(NULL, self->root_prog.filter); 2192 memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter)); 2193 self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter); 2194 2195 self->apply_prog.filter = malloc(sizeof(apply_filter)); 2196 ASSERT_NE(NULL, self->apply_prog.filter); 2197 memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter)); 2198 self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter); 2199 2200 self->sibling_count = 0; 2201 pthread_mutex_init(&self->mutex, NULL); 2202 pthread_cond_init(&self->cond, NULL); 2203 sem_init(&self->started, 0, 0); 2204 self->sibling[0].tid = 0; 2205 self->sibling[0].cond = &self->cond; 2206 self->sibling[0].started = &self->started; 2207 self->sibling[0].mutex = &self->mutex; 2208 self->sibling[0].diverge = 0; 2209 self->sibling[0].num_waits = 1; 2210 self->sibling[0].prog = &self->root_prog; 2211 self->sibling[0].metadata = _metadata; 2212 self->sibling[1].tid = 0; 2213 self->sibling[1].cond = &self->cond; 2214 self->sibling[1].started = &self->started; 2215 self->sibling[1].mutex = &self->mutex; 2216 self->sibling[1].diverge = 0; 2217 self->sibling[1].prog = &self->root_prog; 2218 self->sibling[1].num_waits = 1; 2219 self->sibling[1].metadata = _metadata; 2220 } 2221 2222 FIXTURE_TEARDOWN(TSYNC) 2223 { 2224 int sib = 0; 2225 2226 if (self->root_prog.filter) 2227 free(self->root_prog.filter); 2228 if (self->apply_prog.filter) 2229 free(self->apply_prog.filter); 2230 2231 for ( ; sib < self->sibling_count; ++sib) { 2232 struct tsync_sibling *s = &self->sibling[sib]; 2233 2234 if (!s->tid) 2235 continue; 2236 /* 2237 * If a thread is still running, it may be stuck, so hit 2238 * it over the head really hard. 2239 */ 2240 pthread_kill(s->tid, 9); 2241 } 2242 pthread_mutex_destroy(&self->mutex); 2243 pthread_cond_destroy(&self->cond); 2244 sem_destroy(&self->started); 2245 } 2246 2247 void *tsync_sibling(void *data) 2248 { 2249 long ret = 0; 2250 struct tsync_sibling *me = data; 2251 2252 me->system_tid = syscall(__NR_gettid); 2253 2254 pthread_mutex_lock(me->mutex); 2255 if (me->diverge) { 2256 /* Just re-apply the root prog to fork the tree */ 2257 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, 2258 me->prog, 0, 0); 2259 } 2260 sem_post(me->started); 2261 /* Return outside of started so parent notices failures. */ 2262 if (ret) { 2263 pthread_mutex_unlock(me->mutex); 2264 return (void *)SIBLING_EXIT_FAILURE; 2265 } 2266 do { 2267 pthread_cond_wait(me->cond, me->mutex); 2268 me->num_waits = me->num_waits - 1; 2269 } while (me->num_waits); 2270 pthread_mutex_unlock(me->mutex); 2271 2272 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0); 2273 if (!ret) 2274 return (void *)SIBLING_EXIT_NEWPRIVS; 2275 read(0, NULL, 0); 2276 return (void *)SIBLING_EXIT_UNKILLED; 2277 } 2278 2279 void tsync_start_sibling(struct tsync_sibling *sibling) 2280 { 2281 pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling); 2282 } 2283 2284 TEST_F(TSYNC, siblings_fail_prctl) 2285 { 2286 long ret; 2287 void *status; 2288 struct sock_filter filter[] = { 2289 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2290 offsetof(struct seccomp_data, nr)), 2291 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1), 2292 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL), 2293 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2294 }; 2295 struct sock_fprog prog = { 2296 .len = (unsigned short)ARRAY_SIZE(filter), 2297 .filter = filter, 2298 }; 2299 2300 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2301 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2302 } 2303 2304 /* Check prctl failure detection by requesting sib 0 diverge. */ 2305 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2306 ASSERT_NE(ENOSYS, errno) { 2307 TH_LOG("Kernel does not support seccomp syscall!"); 2308 } 2309 ASSERT_EQ(0, ret) { 2310 TH_LOG("setting filter failed"); 2311 } 2312 2313 self->sibling[0].diverge = 1; 2314 tsync_start_sibling(&self->sibling[0]); 2315 tsync_start_sibling(&self->sibling[1]); 2316 2317 while (self->sibling_count < TSYNC_SIBLINGS) { 2318 sem_wait(&self->started); 2319 self->sibling_count++; 2320 } 2321 2322 /* Signal the threads to clean up*/ 2323 pthread_mutex_lock(&self->mutex); 2324 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2325 TH_LOG("cond broadcast non-zero"); 2326 } 2327 pthread_mutex_unlock(&self->mutex); 2328 2329 /* Ensure diverging sibling failed to call prctl. */ 2330 PTHREAD_JOIN(self->sibling[0].tid, &status); 2331 EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status); 2332 PTHREAD_JOIN(self->sibling[1].tid, &status); 2333 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2334 } 2335 2336 TEST_F(TSYNC, two_siblings_with_ancestor) 2337 { 2338 long ret; 2339 void *status; 2340 2341 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2342 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2343 } 2344 2345 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2346 ASSERT_NE(ENOSYS, errno) { 2347 TH_LOG("Kernel does not support seccomp syscall!"); 2348 } 2349 ASSERT_EQ(0, ret) { 2350 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2351 } 2352 tsync_start_sibling(&self->sibling[0]); 2353 tsync_start_sibling(&self->sibling[1]); 2354 2355 while (self->sibling_count < TSYNC_SIBLINGS) { 2356 sem_wait(&self->started); 2357 self->sibling_count++; 2358 } 2359 2360 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2361 &self->apply_prog); 2362 ASSERT_EQ(0, ret) { 2363 TH_LOG("Could install filter on all threads!"); 2364 } 2365 /* Tell the siblings to test the policy */ 2366 pthread_mutex_lock(&self->mutex); 2367 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2368 TH_LOG("cond broadcast non-zero"); 2369 } 2370 pthread_mutex_unlock(&self->mutex); 2371 /* Ensure they are both killed and don't exit cleanly. */ 2372 PTHREAD_JOIN(self->sibling[0].tid, &status); 2373 EXPECT_EQ(0x0, (long)status); 2374 PTHREAD_JOIN(self->sibling[1].tid, &status); 2375 EXPECT_EQ(0x0, (long)status); 2376 } 2377 2378 TEST_F(TSYNC, two_sibling_want_nnp) 2379 { 2380 void *status; 2381 2382 /* start siblings before any prctl() operations */ 2383 tsync_start_sibling(&self->sibling[0]); 2384 tsync_start_sibling(&self->sibling[1]); 2385 while (self->sibling_count < TSYNC_SIBLINGS) { 2386 sem_wait(&self->started); 2387 self->sibling_count++; 2388 } 2389 2390 /* Tell the siblings to test no policy */ 2391 pthread_mutex_lock(&self->mutex); 2392 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2393 TH_LOG("cond broadcast non-zero"); 2394 } 2395 pthread_mutex_unlock(&self->mutex); 2396 2397 /* Ensure they are both upset about lacking nnp. */ 2398 PTHREAD_JOIN(self->sibling[0].tid, &status); 2399 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status); 2400 PTHREAD_JOIN(self->sibling[1].tid, &status); 2401 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status); 2402 } 2403 2404 TEST_F(TSYNC, two_siblings_with_no_filter) 2405 { 2406 long ret; 2407 void *status; 2408 2409 /* start siblings before any prctl() operations */ 2410 tsync_start_sibling(&self->sibling[0]); 2411 tsync_start_sibling(&self->sibling[1]); 2412 while (self->sibling_count < TSYNC_SIBLINGS) { 2413 sem_wait(&self->started); 2414 self->sibling_count++; 2415 } 2416 2417 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2418 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2419 } 2420 2421 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2422 &self->apply_prog); 2423 ASSERT_NE(ENOSYS, errno) { 2424 TH_LOG("Kernel does not support seccomp syscall!"); 2425 } 2426 ASSERT_EQ(0, ret) { 2427 TH_LOG("Could install filter on all threads!"); 2428 } 2429 2430 /* Tell the siblings to test the policy */ 2431 pthread_mutex_lock(&self->mutex); 2432 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2433 TH_LOG("cond broadcast non-zero"); 2434 } 2435 pthread_mutex_unlock(&self->mutex); 2436 2437 /* Ensure they are both killed and don't exit cleanly. */ 2438 PTHREAD_JOIN(self->sibling[0].tid, &status); 2439 EXPECT_EQ(0x0, (long)status); 2440 PTHREAD_JOIN(self->sibling[1].tid, &status); 2441 EXPECT_EQ(0x0, (long)status); 2442 } 2443 2444 TEST_F(TSYNC, two_siblings_with_one_divergence) 2445 { 2446 long ret; 2447 void *status; 2448 2449 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2450 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2451 } 2452 2453 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2454 ASSERT_NE(ENOSYS, errno) { 2455 TH_LOG("Kernel does not support seccomp syscall!"); 2456 } 2457 ASSERT_EQ(0, ret) { 2458 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2459 } 2460 self->sibling[0].diverge = 1; 2461 tsync_start_sibling(&self->sibling[0]); 2462 tsync_start_sibling(&self->sibling[1]); 2463 2464 while (self->sibling_count < TSYNC_SIBLINGS) { 2465 sem_wait(&self->started); 2466 self->sibling_count++; 2467 } 2468 2469 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2470 &self->apply_prog); 2471 ASSERT_EQ(self->sibling[0].system_tid, ret) { 2472 TH_LOG("Did not fail on diverged sibling."); 2473 } 2474 2475 /* Wake the threads */ 2476 pthread_mutex_lock(&self->mutex); 2477 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2478 TH_LOG("cond broadcast non-zero"); 2479 } 2480 pthread_mutex_unlock(&self->mutex); 2481 2482 /* Ensure they are both unkilled. */ 2483 PTHREAD_JOIN(self->sibling[0].tid, &status); 2484 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2485 PTHREAD_JOIN(self->sibling[1].tid, &status); 2486 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2487 } 2488 2489 TEST_F(TSYNC, two_siblings_not_under_filter) 2490 { 2491 long ret, sib; 2492 void *status; 2493 2494 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2495 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2496 } 2497 2498 /* 2499 * Sibling 0 will have its own seccomp policy 2500 * and Sibling 1 will not be under seccomp at 2501 * all. Sibling 1 will enter seccomp and 0 2502 * will cause failure. 2503 */ 2504 self->sibling[0].diverge = 1; 2505 tsync_start_sibling(&self->sibling[0]); 2506 tsync_start_sibling(&self->sibling[1]); 2507 2508 while (self->sibling_count < TSYNC_SIBLINGS) { 2509 sem_wait(&self->started); 2510 self->sibling_count++; 2511 } 2512 2513 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2514 ASSERT_NE(ENOSYS, errno) { 2515 TH_LOG("Kernel does not support seccomp syscall!"); 2516 } 2517 ASSERT_EQ(0, ret) { 2518 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2519 } 2520 2521 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2522 &self->apply_prog); 2523 ASSERT_EQ(ret, self->sibling[0].system_tid) { 2524 TH_LOG("Did not fail on diverged sibling."); 2525 } 2526 sib = 1; 2527 if (ret == self->sibling[0].system_tid) 2528 sib = 0; 2529 2530 pthread_mutex_lock(&self->mutex); 2531 2532 /* Increment the other siblings num_waits so we can clean up 2533 * the one we just saw. 2534 */ 2535 self->sibling[!sib].num_waits += 1; 2536 2537 /* Signal the thread to clean up*/ 2538 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2539 TH_LOG("cond broadcast non-zero"); 2540 } 2541 pthread_mutex_unlock(&self->mutex); 2542 PTHREAD_JOIN(self->sibling[sib].tid, &status); 2543 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2544 /* Poll for actual task death. pthread_join doesn't guarantee it. */ 2545 while (!kill(self->sibling[sib].system_tid, 0)) 2546 sleep(0.1); 2547 /* Switch to the remaining sibling */ 2548 sib = !sib; 2549 2550 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2551 &self->apply_prog); 2552 ASSERT_EQ(0, ret) { 2553 TH_LOG("Expected the remaining sibling to sync"); 2554 }; 2555 2556 pthread_mutex_lock(&self->mutex); 2557 2558 /* If remaining sibling didn't have a chance to wake up during 2559 * the first broadcast, manually reduce the num_waits now. 2560 */ 2561 if (self->sibling[sib].num_waits > 1) 2562 self->sibling[sib].num_waits = 1; 2563 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2564 TH_LOG("cond broadcast non-zero"); 2565 } 2566 pthread_mutex_unlock(&self->mutex); 2567 PTHREAD_JOIN(self->sibling[sib].tid, &status); 2568 EXPECT_EQ(0, (long)status); 2569 /* Poll for actual task death. pthread_join doesn't guarantee it. */ 2570 while (!kill(self->sibling[sib].system_tid, 0)) 2571 sleep(0.1); 2572 2573 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2574 &self->apply_prog); 2575 ASSERT_EQ(0, ret); /* just us chickens */ 2576 } 2577 2578 /* Make sure restarted syscalls are seen directly as "restart_syscall". */ 2579 TEST(syscall_restart) 2580 { 2581 long ret; 2582 unsigned long msg; 2583 pid_t child_pid; 2584 int pipefd[2]; 2585 int status; 2586 siginfo_t info = { }; 2587 struct sock_filter filter[] = { 2588 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2589 offsetof(struct seccomp_data, nr)), 2590 2591 #ifdef __NR_sigreturn 2592 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 6, 0), 2593 #endif 2594 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 5, 0), 2595 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 4, 0), 2596 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 3, 0), 2597 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 4, 0), 2598 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0), 2599 2600 /* Allow __NR_write for easy logging. */ 2601 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1), 2602 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2603 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2604 /* The nanosleep jump target. */ 2605 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100), 2606 /* The restart_syscall jump target. */ 2607 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200), 2608 }; 2609 struct sock_fprog prog = { 2610 .len = (unsigned short)ARRAY_SIZE(filter), 2611 .filter = filter, 2612 }; 2613 #if defined(__arm__) 2614 struct utsname utsbuf; 2615 #endif 2616 2617 ASSERT_EQ(0, pipe(pipefd)); 2618 2619 child_pid = fork(); 2620 ASSERT_LE(0, child_pid); 2621 if (child_pid == 0) { 2622 /* Child uses EXPECT not ASSERT to deliver status correctly. */ 2623 char buf = ' '; 2624 struct timespec timeout = { }; 2625 2626 /* Attach parent as tracer and stop. */ 2627 EXPECT_EQ(0, ptrace(PTRACE_TRACEME)); 2628 EXPECT_EQ(0, raise(SIGSTOP)); 2629 2630 EXPECT_EQ(0, close(pipefd[1])); 2631 2632 EXPECT_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 = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2637 EXPECT_EQ(0, ret) { 2638 TH_LOG("Failed to install filter!"); 2639 } 2640 2641 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) { 2642 TH_LOG("Failed to read() sync from parent"); 2643 } 2644 EXPECT_EQ('.', buf) { 2645 TH_LOG("Failed to get sync data from read()"); 2646 } 2647 2648 /* Start nanosleep to be interrupted. */ 2649 timeout.tv_sec = 1; 2650 errno = 0; 2651 EXPECT_EQ(0, nanosleep(&timeout, NULL)) { 2652 TH_LOG("Call to nanosleep() failed (errno %d)", errno); 2653 } 2654 2655 /* Read final sync from parent. */ 2656 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) { 2657 TH_LOG("Failed final read() from parent"); 2658 } 2659 EXPECT_EQ('!', buf) { 2660 TH_LOG("Failed to get final data from read()"); 2661 } 2662 2663 /* Directly report the status of our test harness results. */ 2664 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS 2665 : EXIT_FAILURE); 2666 } 2667 EXPECT_EQ(0, close(pipefd[0])); 2668 2669 /* Attach to child, setup options, and release. */ 2670 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2671 ASSERT_EQ(true, WIFSTOPPED(status)); 2672 ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL, 2673 PTRACE_O_TRACESECCOMP)); 2674 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 2675 ASSERT_EQ(1, write(pipefd[1], ".", 1)); 2676 2677 /* Wait for nanosleep() to start. */ 2678 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2679 ASSERT_EQ(true, WIFSTOPPED(status)); 2680 ASSERT_EQ(SIGTRAP, WSTOPSIG(status)); 2681 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16)); 2682 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg)); 2683 ASSERT_EQ(0x100, msg); 2684 EXPECT_EQ(__NR_nanosleep, get_syscall(_metadata, child_pid)); 2685 2686 /* Might as well check siginfo for sanity while we're here. */ 2687 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info)); 2688 ASSERT_EQ(SIGTRAP, info.si_signo); 2689 ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code); 2690 EXPECT_EQ(0, info.si_errno); 2691 EXPECT_EQ(getuid(), info.si_uid); 2692 /* Verify signal delivery came from child (seccomp-triggered). */ 2693 EXPECT_EQ(child_pid, info.si_pid); 2694 2695 /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */ 2696 ASSERT_EQ(0, kill(child_pid, SIGSTOP)); 2697 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 2698 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2699 ASSERT_EQ(true, WIFSTOPPED(status)); 2700 ASSERT_EQ(SIGSTOP, WSTOPSIG(status)); 2701 /* Verify signal delivery came from parent now. */ 2702 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info)); 2703 EXPECT_EQ(getpid(), info.si_pid); 2704 2705 /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */ 2706 ASSERT_EQ(0, kill(child_pid, SIGCONT)); 2707 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 2708 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2709 ASSERT_EQ(true, WIFSTOPPED(status)); 2710 ASSERT_EQ(SIGCONT, WSTOPSIG(status)); 2711 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 2712 2713 /* Wait for restart_syscall() to start. */ 2714 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2715 ASSERT_EQ(true, WIFSTOPPED(status)); 2716 ASSERT_EQ(SIGTRAP, WSTOPSIG(status)); 2717 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16)); 2718 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg)); 2719 2720 ASSERT_EQ(0x200, msg); 2721 ret = get_syscall(_metadata, child_pid); 2722 #if defined(__arm__) 2723 /* 2724 * FIXME: 2725 * - native ARM registers do NOT expose true syscall. 2726 * - compat ARM registers on ARM64 DO expose true syscall. 2727 */ 2728 ASSERT_EQ(0, uname(&utsbuf)); 2729 if (strncmp(utsbuf.machine, "arm", 3) == 0) { 2730 EXPECT_EQ(__NR_nanosleep, ret); 2731 } else 2732 #endif 2733 { 2734 EXPECT_EQ(__NR_restart_syscall, ret); 2735 } 2736 2737 /* Write again to end test. */ 2738 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 2739 ASSERT_EQ(1, write(pipefd[1], "!", 1)); 2740 EXPECT_EQ(0, close(pipefd[1])); 2741 2742 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2743 if (WIFSIGNALED(status) || WEXITSTATUS(status)) 2744 _metadata->passed = 0; 2745 } 2746 2747 TEST_SIGNAL(filter_flag_log, SIGSYS) 2748 { 2749 struct sock_filter allow_filter[] = { 2750 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2751 }; 2752 struct sock_filter kill_filter[] = { 2753 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2754 offsetof(struct seccomp_data, nr)), 2755 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 2756 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2757 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2758 }; 2759 struct sock_fprog allow_prog = { 2760 .len = (unsigned short)ARRAY_SIZE(allow_filter), 2761 .filter = allow_filter, 2762 }; 2763 struct sock_fprog kill_prog = { 2764 .len = (unsigned short)ARRAY_SIZE(kill_filter), 2765 .filter = kill_filter, 2766 }; 2767 long ret; 2768 pid_t parent = getppid(); 2769 2770 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 2771 ASSERT_EQ(0, ret); 2772 2773 /* Verify that the FILTER_FLAG_LOG flag isn't accepted in strict mode */ 2774 ret = seccomp(SECCOMP_SET_MODE_STRICT, SECCOMP_FILTER_FLAG_LOG, 2775 &allow_prog); 2776 ASSERT_NE(ENOSYS, errno) { 2777 TH_LOG("Kernel does not support seccomp syscall!"); 2778 } 2779 EXPECT_NE(0, ret) { 2780 TH_LOG("Kernel accepted FILTER_FLAG_LOG flag in strict mode!"); 2781 } 2782 EXPECT_EQ(EINVAL, errno) { 2783 TH_LOG("Kernel returned unexpected errno for FILTER_FLAG_LOG flag in strict mode!"); 2784 } 2785 2786 /* Verify that a simple, permissive filter can be added with no flags */ 2787 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog); 2788 EXPECT_EQ(0, ret); 2789 2790 /* See if the same filter can be added with the FILTER_FLAG_LOG flag */ 2791 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG, 2792 &allow_prog); 2793 ASSERT_NE(EINVAL, errno) { 2794 TH_LOG("Kernel does not support the FILTER_FLAG_LOG flag!"); 2795 } 2796 EXPECT_EQ(0, ret); 2797 2798 /* Ensure that the kill filter works with the FILTER_FLAG_LOG flag */ 2799 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG, 2800 &kill_prog); 2801 EXPECT_EQ(0, ret); 2802 2803 EXPECT_EQ(parent, syscall(__NR_getppid)); 2804 /* getpid() should never return. */ 2805 EXPECT_EQ(0, syscall(__NR_getpid)); 2806 } 2807 2808 TEST(get_action_avail) 2809 { 2810 __u32 actions[] = { SECCOMP_RET_KILL_THREAD, SECCOMP_RET_TRAP, 2811 SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE, 2812 SECCOMP_RET_LOG, SECCOMP_RET_ALLOW }; 2813 __u32 unknown_action = 0x10000000U; 2814 int i; 2815 long ret; 2816 2817 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[0]); 2818 ASSERT_NE(ENOSYS, errno) { 2819 TH_LOG("Kernel does not support seccomp syscall!"); 2820 } 2821 ASSERT_NE(EINVAL, errno) { 2822 TH_LOG("Kernel does not support SECCOMP_GET_ACTION_AVAIL operation!"); 2823 } 2824 EXPECT_EQ(ret, 0); 2825 2826 for (i = 0; i < ARRAY_SIZE(actions); i++) { 2827 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[i]); 2828 EXPECT_EQ(ret, 0) { 2829 TH_LOG("Expected action (0x%X) not available!", 2830 actions[i]); 2831 } 2832 } 2833 2834 /* Check that an unknown action is handled properly (EOPNOTSUPP) */ 2835 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &unknown_action); 2836 EXPECT_EQ(ret, -1); 2837 EXPECT_EQ(errno, EOPNOTSUPP); 2838 } 2839 2840 /* 2841 * TODO: 2842 * - add microbenchmarks 2843 * - expand NNP testing 2844 * - better arch-specific TRACE and TRAP handlers. 2845 * - endianness checking when appropriate 2846 * - 64-bit arg prodding 2847 * - arch value testing (x86 modes especially) 2848 * - verify that FILTER_FLAG_LOG filters generate log messages 2849 * - verify that RET_LOG generates log messages 2850 * - ... 2851 */ 2852 2853 TEST_HARNESS_MAIN 2854