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 <asm/siginfo.h> 9 #define __have_siginfo_t 1 10 #define __have_sigval_t 1 11 #define __have_sigevent_t 1 12 13 #include <errno.h> 14 #include <linux/filter.h> 15 #include <sys/prctl.h> 16 #include <sys/ptrace.h> 17 #include <sys/types.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; 495 496 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 497 ASSERT_EQ(0, ret); 498 499 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 500 ASSERT_EQ(0, ret); 501 502 fd = open("/dev/zero", O_RDONLY); 503 ASSERT_NE(-1, fd); 504 505 EXPECT_EQ(parent, syscall(__NR_getppid)); 506 map1 = (void *)syscall(sysno, 507 NULL, PAGE_SIZE, PROT_READ, MAP_PRIVATE, fd, PAGE_SIZE); 508 EXPECT_NE(MAP_FAILED, map1); 509 /* mmap2() should never return. */ 510 map2 = (void *)syscall(sysno, 511 NULL, PAGE_SIZE, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE); 512 EXPECT_EQ(MAP_FAILED, map2); 513 514 /* The test failed, so clean up the resources. */ 515 munmap(map1, PAGE_SIZE); 516 munmap(map2, PAGE_SIZE); 517 close(fd); 518 } 519 520 /* TODO(wad) add 64-bit versus 32-bit arg tests. */ 521 TEST(arg_out_of_range) 522 { 523 struct sock_filter filter[] = { 524 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)), 525 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 526 }; 527 struct sock_fprog prog = { 528 .len = (unsigned short)ARRAY_SIZE(filter), 529 .filter = filter, 530 }; 531 long ret; 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 EXPECT_EQ(-1, ret); 538 EXPECT_EQ(EINVAL, errno); 539 } 540 541 TEST(ERRNO_valid) 542 { 543 struct sock_filter filter[] = { 544 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 545 offsetof(struct seccomp_data, nr)), 546 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 547 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | E2BIG), 548 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 549 }; 550 struct sock_fprog prog = { 551 .len = (unsigned short)ARRAY_SIZE(filter), 552 .filter = filter, 553 }; 554 long ret; 555 pid_t parent = getppid(); 556 557 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 558 ASSERT_EQ(0, ret); 559 560 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 561 ASSERT_EQ(0, ret); 562 563 EXPECT_EQ(parent, syscall(__NR_getppid)); 564 EXPECT_EQ(-1, read(0, NULL, 0)); 565 EXPECT_EQ(E2BIG, errno); 566 } 567 568 TEST(ERRNO_zero) 569 { 570 struct sock_filter filter[] = { 571 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 572 offsetof(struct seccomp_data, nr)), 573 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 574 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | 0), 575 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 576 }; 577 struct sock_fprog prog = { 578 .len = (unsigned short)ARRAY_SIZE(filter), 579 .filter = filter, 580 }; 581 long ret; 582 pid_t parent = getppid(); 583 584 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 585 ASSERT_EQ(0, ret); 586 587 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 588 ASSERT_EQ(0, ret); 589 590 EXPECT_EQ(parent, syscall(__NR_getppid)); 591 /* "errno" of 0 is ok. */ 592 EXPECT_EQ(0, read(0, NULL, 0)); 593 } 594 595 TEST(ERRNO_capped) 596 { 597 struct sock_filter filter[] = { 598 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 599 offsetof(struct seccomp_data, nr)), 600 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 601 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | 4096), 602 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 603 }; 604 struct sock_fprog prog = { 605 .len = (unsigned short)ARRAY_SIZE(filter), 606 .filter = filter, 607 }; 608 long ret; 609 pid_t parent = getppid(); 610 611 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 612 ASSERT_EQ(0, ret); 613 614 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 615 ASSERT_EQ(0, ret); 616 617 EXPECT_EQ(parent, syscall(__NR_getppid)); 618 EXPECT_EQ(-1, read(0, NULL, 0)); 619 EXPECT_EQ(4095, errno); 620 } 621 622 FIXTURE_DATA(TRAP) { 623 struct sock_fprog prog; 624 }; 625 626 FIXTURE_SETUP(TRAP) 627 { 628 struct sock_filter filter[] = { 629 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 630 offsetof(struct seccomp_data, nr)), 631 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 632 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP), 633 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 634 }; 635 636 memset(&self->prog, 0, sizeof(self->prog)); 637 self->prog.filter = malloc(sizeof(filter)); 638 ASSERT_NE(NULL, self->prog.filter); 639 memcpy(self->prog.filter, filter, sizeof(filter)); 640 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 641 } 642 643 FIXTURE_TEARDOWN(TRAP) 644 { 645 if (self->prog.filter) 646 free(self->prog.filter); 647 } 648 649 TEST_F_SIGNAL(TRAP, dfl, SIGSYS) 650 { 651 long ret; 652 653 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 654 ASSERT_EQ(0, ret); 655 656 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 657 ASSERT_EQ(0, ret); 658 syscall(__NR_getpid); 659 } 660 661 /* Ensure that SIGSYS overrides SIG_IGN */ 662 TEST_F_SIGNAL(TRAP, ign, SIGSYS) 663 { 664 long ret; 665 666 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 667 ASSERT_EQ(0, ret); 668 669 signal(SIGSYS, SIG_IGN); 670 671 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 672 ASSERT_EQ(0, ret); 673 syscall(__NR_getpid); 674 } 675 676 static struct siginfo TRAP_info; 677 static volatile int TRAP_nr; 678 static void TRAP_action(int nr, siginfo_t *info, void *void_context) 679 { 680 memcpy(&TRAP_info, info, sizeof(TRAP_info)); 681 TRAP_nr = nr; 682 } 683 684 TEST_F(TRAP, handler) 685 { 686 int ret, test; 687 struct sigaction act; 688 sigset_t mask; 689 690 memset(&act, 0, sizeof(act)); 691 sigemptyset(&mask); 692 sigaddset(&mask, SIGSYS); 693 694 act.sa_sigaction = &TRAP_action; 695 act.sa_flags = SA_SIGINFO; 696 ret = sigaction(SIGSYS, &act, NULL); 697 ASSERT_EQ(0, ret) { 698 TH_LOG("sigaction failed"); 699 } 700 ret = sigprocmask(SIG_UNBLOCK, &mask, NULL); 701 ASSERT_EQ(0, ret) { 702 TH_LOG("sigprocmask failed"); 703 } 704 705 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 706 ASSERT_EQ(0, ret); 707 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 708 ASSERT_EQ(0, ret); 709 TRAP_nr = 0; 710 memset(&TRAP_info, 0, sizeof(TRAP_info)); 711 /* Expect the registers to be rolled back. (nr = error) may vary 712 * based on arch. */ 713 ret = syscall(__NR_getpid); 714 /* Silence gcc warning about volatile. */ 715 test = TRAP_nr; 716 EXPECT_EQ(SIGSYS, test); 717 struct local_sigsys { 718 void *_call_addr; /* calling user insn */ 719 int _syscall; /* triggering system call number */ 720 unsigned int _arch; /* AUDIT_ARCH_* of syscall */ 721 } *sigsys = (struct local_sigsys *) 722 #ifdef si_syscall 723 &(TRAP_info.si_call_addr); 724 #else 725 &TRAP_info.si_pid; 726 #endif 727 EXPECT_EQ(__NR_getpid, sigsys->_syscall); 728 /* Make sure arch is non-zero. */ 729 EXPECT_NE(0, sigsys->_arch); 730 EXPECT_NE(0, (unsigned long)sigsys->_call_addr); 731 } 732 733 FIXTURE_DATA(precedence) { 734 struct sock_fprog allow; 735 struct sock_fprog trace; 736 struct sock_fprog error; 737 struct sock_fprog trap; 738 struct sock_fprog kill; 739 }; 740 741 FIXTURE_SETUP(precedence) 742 { 743 struct sock_filter allow_insns[] = { 744 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 745 }; 746 struct sock_filter trace_insns[] = { 747 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 748 offsetof(struct seccomp_data, nr)), 749 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 750 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 751 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE), 752 }; 753 struct sock_filter error_insns[] = { 754 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 755 offsetof(struct seccomp_data, nr)), 756 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 757 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 758 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO), 759 }; 760 struct sock_filter trap_insns[] = { 761 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 762 offsetof(struct seccomp_data, nr)), 763 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 764 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 765 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP), 766 }; 767 struct sock_filter kill_insns[] = { 768 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 769 offsetof(struct seccomp_data, nr)), 770 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 771 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 772 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 773 }; 774 775 memset(self, 0, sizeof(*self)); 776 #define FILTER_ALLOC(_x) \ 777 self->_x.filter = malloc(sizeof(_x##_insns)); \ 778 ASSERT_NE(NULL, self->_x.filter); \ 779 memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \ 780 self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns) 781 FILTER_ALLOC(allow); 782 FILTER_ALLOC(trace); 783 FILTER_ALLOC(error); 784 FILTER_ALLOC(trap); 785 FILTER_ALLOC(kill); 786 } 787 788 FIXTURE_TEARDOWN(precedence) 789 { 790 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter) 791 FILTER_FREE(allow); 792 FILTER_FREE(trace); 793 FILTER_FREE(error); 794 FILTER_FREE(trap); 795 FILTER_FREE(kill); 796 } 797 798 TEST_F(precedence, allow_ok) 799 { 800 pid_t parent, res = 0; 801 long ret; 802 803 parent = getppid(); 804 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 805 ASSERT_EQ(0, ret); 806 807 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 808 ASSERT_EQ(0, ret); 809 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 810 ASSERT_EQ(0, ret); 811 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 812 ASSERT_EQ(0, ret); 813 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 814 ASSERT_EQ(0, ret); 815 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 816 ASSERT_EQ(0, ret); 817 /* Should work just fine. */ 818 res = syscall(__NR_getppid); 819 EXPECT_EQ(parent, res); 820 } 821 822 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS) 823 { 824 pid_t parent, res = 0; 825 long ret; 826 827 parent = getppid(); 828 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 829 ASSERT_EQ(0, ret); 830 831 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 832 ASSERT_EQ(0, ret); 833 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 834 ASSERT_EQ(0, ret); 835 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 836 ASSERT_EQ(0, ret); 837 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 838 ASSERT_EQ(0, ret); 839 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 840 ASSERT_EQ(0, ret); 841 /* Should work just fine. */ 842 res = syscall(__NR_getppid); 843 EXPECT_EQ(parent, res); 844 /* getpid() should never return. */ 845 res = syscall(__NR_getpid); 846 EXPECT_EQ(0, res); 847 } 848 849 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS) 850 { 851 pid_t parent; 852 long ret; 853 854 parent = getppid(); 855 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 856 ASSERT_EQ(0, ret); 857 858 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 859 ASSERT_EQ(0, ret); 860 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 861 ASSERT_EQ(0, ret); 862 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 863 ASSERT_EQ(0, ret); 864 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 865 ASSERT_EQ(0, ret); 866 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 867 ASSERT_EQ(0, ret); 868 /* Should work just fine. */ 869 EXPECT_EQ(parent, syscall(__NR_getppid)); 870 /* getpid() should never return. */ 871 EXPECT_EQ(0, syscall(__NR_getpid)); 872 } 873 874 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS) 875 { 876 pid_t parent; 877 long ret; 878 879 parent = getppid(); 880 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 881 ASSERT_EQ(0, ret); 882 883 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 884 ASSERT_EQ(0, ret); 885 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 886 ASSERT_EQ(0, ret); 887 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 888 ASSERT_EQ(0, ret); 889 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 890 ASSERT_EQ(0, ret); 891 /* Should work just fine. */ 892 EXPECT_EQ(parent, syscall(__NR_getppid)); 893 /* getpid() should never return. */ 894 EXPECT_EQ(0, syscall(__NR_getpid)); 895 } 896 897 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS) 898 { 899 pid_t parent; 900 long ret; 901 902 parent = getppid(); 903 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 904 ASSERT_EQ(0, ret); 905 906 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 907 ASSERT_EQ(0, ret); 908 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 909 ASSERT_EQ(0, ret); 910 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 911 ASSERT_EQ(0, ret); 912 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 913 ASSERT_EQ(0, ret); 914 /* Should work just fine. */ 915 EXPECT_EQ(parent, syscall(__NR_getppid)); 916 /* getpid() should never return. */ 917 EXPECT_EQ(0, syscall(__NR_getpid)); 918 } 919 920 TEST_F(precedence, errno_is_third) 921 { 922 pid_t parent; 923 long ret; 924 925 parent = getppid(); 926 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 927 ASSERT_EQ(0, ret); 928 929 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 930 ASSERT_EQ(0, ret); 931 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 932 ASSERT_EQ(0, ret); 933 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 934 ASSERT_EQ(0, ret); 935 /* Should work just fine. */ 936 EXPECT_EQ(parent, syscall(__NR_getppid)); 937 EXPECT_EQ(0, syscall(__NR_getpid)); 938 } 939 940 TEST_F(precedence, errno_is_third_in_any_order) 941 { 942 pid_t parent; 943 long ret; 944 945 parent = getppid(); 946 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 947 ASSERT_EQ(0, ret); 948 949 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 950 ASSERT_EQ(0, ret); 951 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 952 ASSERT_EQ(0, ret); 953 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 954 ASSERT_EQ(0, ret); 955 /* Should work just fine. */ 956 EXPECT_EQ(parent, syscall(__NR_getppid)); 957 EXPECT_EQ(0, syscall(__NR_getpid)); 958 } 959 960 TEST_F(precedence, trace_is_fourth) 961 { 962 pid_t parent; 963 long ret; 964 965 parent = getppid(); 966 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 967 ASSERT_EQ(0, ret); 968 969 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 970 ASSERT_EQ(0, ret); 971 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 972 ASSERT_EQ(0, ret); 973 /* Should work just fine. */ 974 EXPECT_EQ(parent, syscall(__NR_getppid)); 975 /* No ptracer */ 976 EXPECT_EQ(-1, syscall(__NR_getpid)); 977 } 978 979 TEST_F(precedence, trace_is_fourth_in_any_order) 980 { 981 pid_t parent; 982 long ret; 983 984 parent = getppid(); 985 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 986 ASSERT_EQ(0, ret); 987 988 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 989 ASSERT_EQ(0, ret); 990 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 991 ASSERT_EQ(0, ret); 992 /* Should work just fine. */ 993 EXPECT_EQ(parent, syscall(__NR_getppid)); 994 /* No ptracer */ 995 EXPECT_EQ(-1, syscall(__NR_getpid)); 996 } 997 998 #ifndef PTRACE_O_TRACESECCOMP 999 #define PTRACE_O_TRACESECCOMP 0x00000080 1000 #endif 1001 1002 /* Catch the Ubuntu 12.04 value error. */ 1003 #if PTRACE_EVENT_SECCOMP != 7 1004 #undef PTRACE_EVENT_SECCOMP 1005 #endif 1006 1007 #ifndef PTRACE_EVENT_SECCOMP 1008 #define PTRACE_EVENT_SECCOMP 7 1009 #endif 1010 1011 #define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP) 1012 bool tracer_running; 1013 void tracer_stop(int sig) 1014 { 1015 tracer_running = false; 1016 } 1017 1018 typedef void tracer_func_t(struct __test_metadata *_metadata, 1019 pid_t tracee, int status, void *args); 1020 1021 void tracer(struct __test_metadata *_metadata, int fd, pid_t tracee, 1022 tracer_func_t tracer_func, void *args) 1023 { 1024 int ret = -1; 1025 struct sigaction action = { 1026 .sa_handler = tracer_stop, 1027 }; 1028 1029 /* Allow external shutdown. */ 1030 tracer_running = true; 1031 ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL)); 1032 1033 errno = 0; 1034 while (ret == -1 && errno != EINVAL) 1035 ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0); 1036 ASSERT_EQ(0, ret) { 1037 kill(tracee, SIGKILL); 1038 } 1039 /* Wait for attach stop */ 1040 wait(NULL); 1041 1042 ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, PTRACE_O_TRACESECCOMP); 1043 ASSERT_EQ(0, ret) { 1044 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP"); 1045 kill(tracee, SIGKILL); 1046 } 1047 ptrace(PTRACE_CONT, tracee, NULL, 0); 1048 1049 /* Unblock the tracee */ 1050 ASSERT_EQ(1, write(fd, "A", 1)); 1051 ASSERT_EQ(0, close(fd)); 1052 1053 /* Run until we're shut down. Must assert to stop execution. */ 1054 while (tracer_running) { 1055 int status; 1056 1057 if (wait(&status) != tracee) 1058 continue; 1059 if (WIFSIGNALED(status) || WIFEXITED(status)) 1060 /* Child is dead. Time to go. */ 1061 return; 1062 1063 /* Make sure this is a seccomp event. */ 1064 ASSERT_EQ(true, IS_SECCOMP_EVENT(status)); 1065 1066 tracer_func(_metadata, tracee, status, args); 1067 1068 ret = ptrace(PTRACE_CONT, tracee, NULL, NULL); 1069 ASSERT_EQ(0, ret); 1070 } 1071 /* Directly report the status of our test harness results. */ 1072 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS : EXIT_FAILURE); 1073 } 1074 1075 /* Common tracer setup/teardown functions. */ 1076 void cont_handler(int num) 1077 { } 1078 pid_t setup_trace_fixture(struct __test_metadata *_metadata, 1079 tracer_func_t func, void *args) 1080 { 1081 char sync; 1082 int pipefd[2]; 1083 pid_t tracer_pid; 1084 pid_t tracee = getpid(); 1085 1086 /* Setup a pipe for clean synchronization. */ 1087 ASSERT_EQ(0, pipe(pipefd)); 1088 1089 /* Fork a child which we'll promote to tracer */ 1090 tracer_pid = fork(); 1091 ASSERT_LE(0, tracer_pid); 1092 signal(SIGALRM, cont_handler); 1093 if (tracer_pid == 0) { 1094 close(pipefd[0]); 1095 tracer(_metadata, pipefd[1], tracee, func, args); 1096 syscall(__NR_exit, 0); 1097 } 1098 close(pipefd[1]); 1099 prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0); 1100 read(pipefd[0], &sync, 1); 1101 close(pipefd[0]); 1102 1103 return tracer_pid; 1104 } 1105 void teardown_trace_fixture(struct __test_metadata *_metadata, 1106 pid_t tracer) 1107 { 1108 if (tracer) { 1109 int status; 1110 /* 1111 * Extract the exit code from the other process and 1112 * adopt it for ourselves in case its asserts failed. 1113 */ 1114 ASSERT_EQ(0, kill(tracer, SIGUSR1)); 1115 ASSERT_EQ(tracer, waitpid(tracer, &status, 0)); 1116 if (WEXITSTATUS(status)) 1117 _metadata->passed = 0; 1118 } 1119 } 1120 1121 /* "poke" tracer arguments and function. */ 1122 struct tracer_args_poke_t { 1123 unsigned long poke_addr; 1124 }; 1125 1126 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status, 1127 void *args) 1128 { 1129 int ret; 1130 unsigned long msg; 1131 struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args; 1132 1133 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 1134 EXPECT_EQ(0, ret); 1135 /* If this fails, don't try to recover. */ 1136 ASSERT_EQ(0x1001, msg) { 1137 kill(tracee, SIGKILL); 1138 } 1139 /* 1140 * Poke in the message. 1141 * Registers are not touched to try to keep this relatively arch 1142 * agnostic. 1143 */ 1144 ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001); 1145 EXPECT_EQ(0, ret); 1146 } 1147 1148 FIXTURE_DATA(TRACE_poke) { 1149 struct sock_fprog prog; 1150 pid_t tracer; 1151 long poked; 1152 struct tracer_args_poke_t tracer_args; 1153 }; 1154 1155 FIXTURE_SETUP(TRACE_poke) 1156 { 1157 struct sock_filter filter[] = { 1158 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1159 offsetof(struct seccomp_data, nr)), 1160 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 1161 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001), 1162 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1163 }; 1164 1165 self->poked = 0; 1166 memset(&self->prog, 0, sizeof(self->prog)); 1167 self->prog.filter = malloc(sizeof(filter)); 1168 ASSERT_NE(NULL, self->prog.filter); 1169 memcpy(self->prog.filter, filter, sizeof(filter)); 1170 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 1171 1172 /* Set up tracer args. */ 1173 self->tracer_args.poke_addr = (unsigned long)&self->poked; 1174 1175 /* Launch tracer. */ 1176 self->tracer = setup_trace_fixture(_metadata, tracer_poke, 1177 &self->tracer_args); 1178 } 1179 1180 FIXTURE_TEARDOWN(TRACE_poke) 1181 { 1182 teardown_trace_fixture(_metadata, self->tracer); 1183 if (self->prog.filter) 1184 free(self->prog.filter); 1185 } 1186 1187 TEST_F(TRACE_poke, read_has_side_effects) 1188 { 1189 ssize_t ret; 1190 1191 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1192 ASSERT_EQ(0, ret); 1193 1194 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1195 ASSERT_EQ(0, ret); 1196 1197 EXPECT_EQ(0, self->poked); 1198 ret = read(-1, NULL, 0); 1199 EXPECT_EQ(-1, ret); 1200 EXPECT_EQ(0x1001, self->poked); 1201 } 1202 1203 TEST_F(TRACE_poke, getpid_runs_normally) 1204 { 1205 long ret; 1206 1207 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1208 ASSERT_EQ(0, ret); 1209 1210 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1211 ASSERT_EQ(0, ret); 1212 1213 EXPECT_EQ(0, self->poked); 1214 EXPECT_NE(0, syscall(__NR_getpid)); 1215 EXPECT_EQ(0, self->poked); 1216 } 1217 1218 #if defined(__x86_64__) 1219 # define ARCH_REGS struct user_regs_struct 1220 # define SYSCALL_NUM orig_rax 1221 # define SYSCALL_RET rax 1222 #elif defined(__i386__) 1223 # define ARCH_REGS struct user_regs_struct 1224 # define SYSCALL_NUM orig_eax 1225 # define SYSCALL_RET eax 1226 #elif defined(__arm__) 1227 # define ARCH_REGS struct pt_regs 1228 # define SYSCALL_NUM ARM_r7 1229 # define SYSCALL_RET ARM_r0 1230 #elif defined(__aarch64__) 1231 # define ARCH_REGS struct user_pt_regs 1232 # define SYSCALL_NUM regs[8] 1233 # define SYSCALL_RET regs[0] 1234 #elif defined(__powerpc__) 1235 # define ARCH_REGS struct pt_regs 1236 # define SYSCALL_NUM gpr[0] 1237 # define SYSCALL_RET gpr[3] 1238 #elif defined(__s390__) 1239 # define ARCH_REGS s390_regs 1240 # define SYSCALL_NUM gprs[2] 1241 # define SYSCALL_RET gprs[2] 1242 #else 1243 # error "Do not know how to find your architecture's registers and syscalls" 1244 #endif 1245 1246 /* Architecture-specific syscall fetching routine. */ 1247 int get_syscall(struct __test_metadata *_metadata, pid_t tracee) 1248 { 1249 struct iovec iov; 1250 ARCH_REGS regs; 1251 1252 iov.iov_base = ®s; 1253 iov.iov_len = sizeof(regs); 1254 EXPECT_EQ(0, ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov)) { 1255 TH_LOG("PTRACE_GETREGSET failed"); 1256 return -1; 1257 } 1258 1259 return regs.SYSCALL_NUM; 1260 } 1261 1262 /* Architecture-specific syscall changing routine. */ 1263 void change_syscall(struct __test_metadata *_metadata, 1264 pid_t tracee, int syscall) 1265 { 1266 struct iovec iov; 1267 int ret; 1268 ARCH_REGS regs; 1269 1270 iov.iov_base = ®s; 1271 iov.iov_len = sizeof(regs); 1272 ret = ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov); 1273 EXPECT_EQ(0, ret); 1274 1275 #if defined(__x86_64__) || defined(__i386__) || defined(__powerpc__) || \ 1276 defined(__s390__) 1277 { 1278 regs.SYSCALL_NUM = syscall; 1279 } 1280 1281 #elif defined(__arm__) 1282 # ifndef PTRACE_SET_SYSCALL 1283 # define PTRACE_SET_SYSCALL 23 1284 # endif 1285 { 1286 ret = ptrace(PTRACE_SET_SYSCALL, tracee, NULL, syscall); 1287 EXPECT_EQ(0, ret); 1288 } 1289 1290 #elif defined(__aarch64__) 1291 # ifndef NT_ARM_SYSTEM_CALL 1292 # define NT_ARM_SYSTEM_CALL 0x404 1293 # endif 1294 { 1295 iov.iov_base = &syscall; 1296 iov.iov_len = sizeof(syscall); 1297 ret = ptrace(PTRACE_SETREGSET, tracee, NT_ARM_SYSTEM_CALL, 1298 &iov); 1299 EXPECT_EQ(0, ret); 1300 } 1301 1302 #else 1303 ASSERT_EQ(1, 0) { 1304 TH_LOG("How is the syscall changed on this architecture?"); 1305 } 1306 #endif 1307 1308 /* If syscall is skipped, change return value. */ 1309 if (syscall == -1) 1310 regs.SYSCALL_RET = 1; 1311 1312 iov.iov_base = ®s; 1313 iov.iov_len = sizeof(regs); 1314 ret = ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &iov); 1315 EXPECT_EQ(0, ret); 1316 } 1317 1318 void tracer_syscall(struct __test_metadata *_metadata, pid_t tracee, 1319 int status, void *args) 1320 { 1321 int ret; 1322 unsigned long msg; 1323 1324 /* Make sure we got the right message. */ 1325 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 1326 EXPECT_EQ(0, ret); 1327 1328 /* Validate and take action on expected syscalls. */ 1329 switch (msg) { 1330 case 0x1002: 1331 /* change getpid to getppid. */ 1332 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee)); 1333 change_syscall(_metadata, tracee, __NR_getppid); 1334 break; 1335 case 0x1003: 1336 /* skip gettid. */ 1337 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee)); 1338 change_syscall(_metadata, tracee, -1); 1339 break; 1340 case 0x1004: 1341 /* do nothing (allow getppid) */ 1342 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee)); 1343 break; 1344 default: 1345 EXPECT_EQ(0, msg) { 1346 TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg); 1347 kill(tracee, SIGKILL); 1348 } 1349 } 1350 1351 } 1352 1353 FIXTURE_DATA(TRACE_syscall) { 1354 struct sock_fprog prog; 1355 pid_t tracer, mytid, mypid, parent; 1356 }; 1357 1358 FIXTURE_SETUP(TRACE_syscall) 1359 { 1360 struct sock_filter filter[] = { 1361 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1362 offsetof(struct seccomp_data, nr)), 1363 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 1364 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002), 1365 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1), 1366 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003), 1367 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 1368 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004), 1369 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1370 }; 1371 1372 memset(&self->prog, 0, sizeof(self->prog)); 1373 self->prog.filter = malloc(sizeof(filter)); 1374 ASSERT_NE(NULL, self->prog.filter); 1375 memcpy(self->prog.filter, filter, sizeof(filter)); 1376 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 1377 1378 /* Prepare some testable syscall results. */ 1379 self->mytid = syscall(__NR_gettid); 1380 ASSERT_GT(self->mytid, 0); 1381 ASSERT_NE(self->mytid, 1) { 1382 TH_LOG("Running this test as init is not supported. :)"); 1383 } 1384 1385 self->mypid = getpid(); 1386 ASSERT_GT(self->mypid, 0); 1387 ASSERT_EQ(self->mytid, self->mypid); 1388 1389 self->parent = getppid(); 1390 ASSERT_GT(self->parent, 0); 1391 ASSERT_NE(self->parent, self->mypid); 1392 1393 /* Launch tracer. */ 1394 self->tracer = setup_trace_fixture(_metadata, tracer_syscall, NULL); 1395 } 1396 1397 FIXTURE_TEARDOWN(TRACE_syscall) 1398 { 1399 teardown_trace_fixture(_metadata, self->tracer); 1400 if (self->prog.filter) 1401 free(self->prog.filter); 1402 } 1403 1404 TEST_F(TRACE_syscall, syscall_allowed) 1405 { 1406 long ret; 1407 1408 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1409 ASSERT_EQ(0, ret); 1410 1411 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1412 ASSERT_EQ(0, ret); 1413 1414 /* getppid works as expected (no changes). */ 1415 EXPECT_EQ(self->parent, syscall(__NR_getppid)); 1416 EXPECT_NE(self->mypid, syscall(__NR_getppid)); 1417 } 1418 1419 TEST_F(TRACE_syscall, syscall_redirected) 1420 { 1421 long ret; 1422 1423 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1424 ASSERT_EQ(0, ret); 1425 1426 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1427 ASSERT_EQ(0, ret); 1428 1429 /* getpid has been redirected to getppid as expected. */ 1430 EXPECT_EQ(self->parent, syscall(__NR_getpid)); 1431 EXPECT_NE(self->mypid, syscall(__NR_getpid)); 1432 } 1433 1434 TEST_F(TRACE_syscall, syscall_dropped) 1435 { 1436 long ret; 1437 1438 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1439 ASSERT_EQ(0, ret); 1440 1441 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1442 ASSERT_EQ(0, ret); 1443 1444 /* gettid has been skipped and an altered return value stored. */ 1445 EXPECT_EQ(1, syscall(__NR_gettid)); 1446 EXPECT_NE(self->mytid, syscall(__NR_gettid)); 1447 } 1448 1449 #ifndef __NR_seccomp 1450 # if defined(__i386__) 1451 # define __NR_seccomp 354 1452 # elif defined(__x86_64__) 1453 # define __NR_seccomp 317 1454 # elif defined(__arm__) 1455 # define __NR_seccomp 383 1456 # elif defined(__aarch64__) 1457 # define __NR_seccomp 277 1458 # elif defined(__powerpc__) 1459 # define __NR_seccomp 358 1460 # elif defined(__s390__) 1461 # define __NR_seccomp 348 1462 # else 1463 # warning "seccomp syscall number unknown for this architecture" 1464 # define __NR_seccomp 0xffff 1465 # endif 1466 #endif 1467 1468 #ifndef SECCOMP_SET_MODE_STRICT 1469 #define SECCOMP_SET_MODE_STRICT 0 1470 #endif 1471 1472 #ifndef SECCOMP_SET_MODE_FILTER 1473 #define SECCOMP_SET_MODE_FILTER 1 1474 #endif 1475 1476 #ifndef SECCOMP_FLAG_FILTER_TSYNC 1477 #define SECCOMP_FLAG_FILTER_TSYNC 1 1478 #endif 1479 1480 #ifndef seccomp 1481 int seccomp(unsigned int op, unsigned int flags, struct sock_fprog *filter) 1482 { 1483 errno = 0; 1484 return syscall(__NR_seccomp, op, flags, filter); 1485 } 1486 #endif 1487 1488 TEST(seccomp_syscall) 1489 { 1490 struct sock_filter filter[] = { 1491 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1492 }; 1493 struct sock_fprog prog = { 1494 .len = (unsigned short)ARRAY_SIZE(filter), 1495 .filter = filter, 1496 }; 1497 long ret; 1498 1499 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1500 ASSERT_EQ(0, ret) { 1501 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 1502 } 1503 1504 /* Reject insane operation. */ 1505 ret = seccomp(-1, 0, &prog); 1506 ASSERT_NE(ENOSYS, errno) { 1507 TH_LOG("Kernel does not support seccomp syscall!"); 1508 } 1509 EXPECT_EQ(EINVAL, errno) { 1510 TH_LOG("Did not reject crazy op value!"); 1511 } 1512 1513 /* Reject strict with flags or pointer. */ 1514 ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL); 1515 EXPECT_EQ(EINVAL, errno) { 1516 TH_LOG("Did not reject mode strict with flags!"); 1517 } 1518 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog); 1519 EXPECT_EQ(EINVAL, errno) { 1520 TH_LOG("Did not reject mode strict with uargs!"); 1521 } 1522 1523 /* Reject insane args for filter. */ 1524 ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog); 1525 EXPECT_EQ(EINVAL, errno) { 1526 TH_LOG("Did not reject crazy filter flags!"); 1527 } 1528 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL); 1529 EXPECT_EQ(EFAULT, errno) { 1530 TH_LOG("Did not reject NULL filter!"); 1531 } 1532 1533 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 1534 EXPECT_EQ(0, errno) { 1535 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s", 1536 strerror(errno)); 1537 } 1538 } 1539 1540 TEST(seccomp_syscall_mode_lock) 1541 { 1542 struct sock_filter filter[] = { 1543 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1544 }; 1545 struct sock_fprog prog = { 1546 .len = (unsigned short)ARRAY_SIZE(filter), 1547 .filter = filter, 1548 }; 1549 long ret; 1550 1551 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 1552 ASSERT_EQ(0, ret) { 1553 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 1554 } 1555 1556 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 1557 ASSERT_NE(ENOSYS, errno) { 1558 TH_LOG("Kernel does not support seccomp syscall!"); 1559 } 1560 EXPECT_EQ(0, ret) { 1561 TH_LOG("Could not install filter!"); 1562 } 1563 1564 /* Make sure neither entry point will switch to strict. */ 1565 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0); 1566 EXPECT_EQ(EINVAL, errno) { 1567 TH_LOG("Switched to mode strict!"); 1568 } 1569 1570 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL); 1571 EXPECT_EQ(EINVAL, errno) { 1572 TH_LOG("Switched to mode strict!"); 1573 } 1574 } 1575 1576 TEST(TSYNC_first) 1577 { 1578 struct sock_filter filter[] = { 1579 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1580 }; 1581 struct sock_fprog prog = { 1582 .len = (unsigned short)ARRAY_SIZE(filter), 1583 .filter = filter, 1584 }; 1585 long ret; 1586 1587 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 1588 ASSERT_EQ(0, ret) { 1589 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 1590 } 1591 1592 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FLAG_FILTER_TSYNC, 1593 &prog); 1594 ASSERT_NE(ENOSYS, errno) { 1595 TH_LOG("Kernel does not support seccomp syscall!"); 1596 } 1597 EXPECT_EQ(0, ret) { 1598 TH_LOG("Could not install initial filter with TSYNC!"); 1599 } 1600 } 1601 1602 #define TSYNC_SIBLINGS 2 1603 struct tsync_sibling { 1604 pthread_t tid; 1605 pid_t system_tid; 1606 sem_t *started; 1607 pthread_cond_t *cond; 1608 pthread_mutex_t *mutex; 1609 int diverge; 1610 int num_waits; 1611 struct sock_fprog *prog; 1612 struct __test_metadata *metadata; 1613 }; 1614 1615 FIXTURE_DATA(TSYNC) { 1616 struct sock_fprog root_prog, apply_prog; 1617 struct tsync_sibling sibling[TSYNC_SIBLINGS]; 1618 sem_t started; 1619 pthread_cond_t cond; 1620 pthread_mutex_t mutex; 1621 int sibling_count; 1622 }; 1623 1624 FIXTURE_SETUP(TSYNC) 1625 { 1626 struct sock_filter root_filter[] = { 1627 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1628 }; 1629 struct sock_filter apply_filter[] = { 1630 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1631 offsetof(struct seccomp_data, nr)), 1632 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 1633 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 1634 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1635 }; 1636 1637 memset(&self->root_prog, 0, sizeof(self->root_prog)); 1638 memset(&self->apply_prog, 0, sizeof(self->apply_prog)); 1639 memset(&self->sibling, 0, sizeof(self->sibling)); 1640 self->root_prog.filter = malloc(sizeof(root_filter)); 1641 ASSERT_NE(NULL, self->root_prog.filter); 1642 memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter)); 1643 self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter); 1644 1645 self->apply_prog.filter = malloc(sizeof(apply_filter)); 1646 ASSERT_NE(NULL, self->apply_prog.filter); 1647 memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter)); 1648 self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter); 1649 1650 self->sibling_count = 0; 1651 pthread_mutex_init(&self->mutex, NULL); 1652 pthread_cond_init(&self->cond, NULL); 1653 sem_init(&self->started, 0, 0); 1654 self->sibling[0].tid = 0; 1655 self->sibling[0].cond = &self->cond; 1656 self->sibling[0].started = &self->started; 1657 self->sibling[0].mutex = &self->mutex; 1658 self->sibling[0].diverge = 0; 1659 self->sibling[0].num_waits = 1; 1660 self->sibling[0].prog = &self->root_prog; 1661 self->sibling[0].metadata = _metadata; 1662 self->sibling[1].tid = 0; 1663 self->sibling[1].cond = &self->cond; 1664 self->sibling[1].started = &self->started; 1665 self->sibling[1].mutex = &self->mutex; 1666 self->sibling[1].diverge = 0; 1667 self->sibling[1].prog = &self->root_prog; 1668 self->sibling[1].num_waits = 1; 1669 self->sibling[1].metadata = _metadata; 1670 } 1671 1672 FIXTURE_TEARDOWN(TSYNC) 1673 { 1674 int sib = 0; 1675 1676 if (self->root_prog.filter) 1677 free(self->root_prog.filter); 1678 if (self->apply_prog.filter) 1679 free(self->apply_prog.filter); 1680 1681 for ( ; sib < self->sibling_count; ++sib) { 1682 struct tsync_sibling *s = &self->sibling[sib]; 1683 void *status; 1684 1685 if (!s->tid) 1686 continue; 1687 if (pthread_kill(s->tid, 0)) { 1688 pthread_cancel(s->tid); 1689 pthread_join(s->tid, &status); 1690 } 1691 } 1692 pthread_mutex_destroy(&self->mutex); 1693 pthread_cond_destroy(&self->cond); 1694 sem_destroy(&self->started); 1695 } 1696 1697 void *tsync_sibling(void *data) 1698 { 1699 long ret = 0; 1700 struct tsync_sibling *me = data; 1701 1702 me->system_tid = syscall(__NR_gettid); 1703 1704 pthread_mutex_lock(me->mutex); 1705 if (me->diverge) { 1706 /* Just re-apply the root prog to fork the tree */ 1707 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, 1708 me->prog, 0, 0); 1709 } 1710 sem_post(me->started); 1711 /* Return outside of started so parent notices failures. */ 1712 if (ret) { 1713 pthread_mutex_unlock(me->mutex); 1714 return (void *)SIBLING_EXIT_FAILURE; 1715 } 1716 do { 1717 pthread_cond_wait(me->cond, me->mutex); 1718 me->num_waits = me->num_waits - 1; 1719 } while (me->num_waits); 1720 pthread_mutex_unlock(me->mutex); 1721 1722 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0); 1723 if (!ret) 1724 return (void *)SIBLING_EXIT_NEWPRIVS; 1725 read(0, NULL, 0); 1726 return (void *)SIBLING_EXIT_UNKILLED; 1727 } 1728 1729 void tsync_start_sibling(struct tsync_sibling *sibling) 1730 { 1731 pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling); 1732 } 1733 1734 TEST_F(TSYNC, siblings_fail_prctl) 1735 { 1736 long ret; 1737 void *status; 1738 struct sock_filter filter[] = { 1739 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1740 offsetof(struct seccomp_data, nr)), 1741 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1), 1742 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL), 1743 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1744 }; 1745 struct sock_fprog prog = { 1746 .len = (unsigned short)ARRAY_SIZE(filter), 1747 .filter = filter, 1748 }; 1749 1750 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 1751 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 1752 } 1753 1754 /* Check prctl failure detection by requesting sib 0 diverge. */ 1755 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 1756 ASSERT_NE(ENOSYS, errno) { 1757 TH_LOG("Kernel does not support seccomp syscall!"); 1758 } 1759 ASSERT_EQ(0, ret) { 1760 TH_LOG("setting filter failed"); 1761 } 1762 1763 self->sibling[0].diverge = 1; 1764 tsync_start_sibling(&self->sibling[0]); 1765 tsync_start_sibling(&self->sibling[1]); 1766 1767 while (self->sibling_count < TSYNC_SIBLINGS) { 1768 sem_wait(&self->started); 1769 self->sibling_count++; 1770 } 1771 1772 /* Signal the threads to clean up*/ 1773 pthread_mutex_lock(&self->mutex); 1774 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 1775 TH_LOG("cond broadcast non-zero"); 1776 } 1777 pthread_mutex_unlock(&self->mutex); 1778 1779 /* Ensure diverging sibling failed to call prctl. */ 1780 pthread_join(self->sibling[0].tid, &status); 1781 EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status); 1782 pthread_join(self->sibling[1].tid, &status); 1783 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 1784 } 1785 1786 TEST_F(TSYNC, two_siblings_with_ancestor) 1787 { 1788 long ret; 1789 void *status; 1790 1791 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 1792 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 1793 } 1794 1795 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 1796 ASSERT_NE(ENOSYS, errno) { 1797 TH_LOG("Kernel does not support seccomp syscall!"); 1798 } 1799 ASSERT_EQ(0, ret) { 1800 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 1801 } 1802 tsync_start_sibling(&self->sibling[0]); 1803 tsync_start_sibling(&self->sibling[1]); 1804 1805 while (self->sibling_count < TSYNC_SIBLINGS) { 1806 sem_wait(&self->started); 1807 self->sibling_count++; 1808 } 1809 1810 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FLAG_FILTER_TSYNC, 1811 &self->apply_prog); 1812 ASSERT_EQ(0, ret) { 1813 TH_LOG("Could install filter on all threads!"); 1814 } 1815 /* Tell the siblings to test the policy */ 1816 pthread_mutex_lock(&self->mutex); 1817 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 1818 TH_LOG("cond broadcast non-zero"); 1819 } 1820 pthread_mutex_unlock(&self->mutex); 1821 /* Ensure they are both killed and don't exit cleanly. */ 1822 pthread_join(self->sibling[0].tid, &status); 1823 EXPECT_EQ(0x0, (long)status); 1824 pthread_join(self->sibling[1].tid, &status); 1825 EXPECT_EQ(0x0, (long)status); 1826 } 1827 1828 TEST_F(TSYNC, two_sibling_want_nnp) 1829 { 1830 void *status; 1831 1832 /* start siblings before any prctl() operations */ 1833 tsync_start_sibling(&self->sibling[0]); 1834 tsync_start_sibling(&self->sibling[1]); 1835 while (self->sibling_count < TSYNC_SIBLINGS) { 1836 sem_wait(&self->started); 1837 self->sibling_count++; 1838 } 1839 1840 /* Tell the siblings to test no policy */ 1841 pthread_mutex_lock(&self->mutex); 1842 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 1843 TH_LOG("cond broadcast non-zero"); 1844 } 1845 pthread_mutex_unlock(&self->mutex); 1846 1847 /* Ensure they are both upset about lacking nnp. */ 1848 pthread_join(self->sibling[0].tid, &status); 1849 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status); 1850 pthread_join(self->sibling[1].tid, &status); 1851 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status); 1852 } 1853 1854 TEST_F(TSYNC, two_siblings_with_no_filter) 1855 { 1856 long ret; 1857 void *status; 1858 1859 /* start siblings before any prctl() operations */ 1860 tsync_start_sibling(&self->sibling[0]); 1861 tsync_start_sibling(&self->sibling[1]); 1862 while (self->sibling_count < TSYNC_SIBLINGS) { 1863 sem_wait(&self->started); 1864 self->sibling_count++; 1865 } 1866 1867 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 1868 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 1869 } 1870 1871 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FLAG_FILTER_TSYNC, 1872 &self->apply_prog); 1873 ASSERT_NE(ENOSYS, errno) { 1874 TH_LOG("Kernel does not support seccomp syscall!"); 1875 } 1876 ASSERT_EQ(0, ret) { 1877 TH_LOG("Could install filter on all threads!"); 1878 } 1879 1880 /* Tell the siblings to test the policy */ 1881 pthread_mutex_lock(&self->mutex); 1882 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 1883 TH_LOG("cond broadcast non-zero"); 1884 } 1885 pthread_mutex_unlock(&self->mutex); 1886 1887 /* Ensure they are both killed and don't exit cleanly. */ 1888 pthread_join(self->sibling[0].tid, &status); 1889 EXPECT_EQ(0x0, (long)status); 1890 pthread_join(self->sibling[1].tid, &status); 1891 EXPECT_EQ(0x0, (long)status); 1892 } 1893 1894 TEST_F(TSYNC, two_siblings_with_one_divergence) 1895 { 1896 long ret; 1897 void *status; 1898 1899 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 1900 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 1901 } 1902 1903 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 1904 ASSERT_NE(ENOSYS, errno) { 1905 TH_LOG("Kernel does not support seccomp syscall!"); 1906 } 1907 ASSERT_EQ(0, ret) { 1908 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 1909 } 1910 self->sibling[0].diverge = 1; 1911 tsync_start_sibling(&self->sibling[0]); 1912 tsync_start_sibling(&self->sibling[1]); 1913 1914 while (self->sibling_count < TSYNC_SIBLINGS) { 1915 sem_wait(&self->started); 1916 self->sibling_count++; 1917 } 1918 1919 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FLAG_FILTER_TSYNC, 1920 &self->apply_prog); 1921 ASSERT_EQ(self->sibling[0].system_tid, ret) { 1922 TH_LOG("Did not fail on diverged sibling."); 1923 } 1924 1925 /* Wake the threads */ 1926 pthread_mutex_lock(&self->mutex); 1927 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 1928 TH_LOG("cond broadcast non-zero"); 1929 } 1930 pthread_mutex_unlock(&self->mutex); 1931 1932 /* Ensure they are both unkilled. */ 1933 pthread_join(self->sibling[0].tid, &status); 1934 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 1935 pthread_join(self->sibling[1].tid, &status); 1936 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 1937 } 1938 1939 TEST_F(TSYNC, two_siblings_not_under_filter) 1940 { 1941 long ret, sib; 1942 void *status; 1943 1944 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 1945 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 1946 } 1947 1948 /* 1949 * Sibling 0 will have its own seccomp policy 1950 * and Sibling 1 will not be under seccomp at 1951 * all. Sibling 1 will enter seccomp and 0 1952 * will cause failure. 1953 */ 1954 self->sibling[0].diverge = 1; 1955 tsync_start_sibling(&self->sibling[0]); 1956 tsync_start_sibling(&self->sibling[1]); 1957 1958 while (self->sibling_count < TSYNC_SIBLINGS) { 1959 sem_wait(&self->started); 1960 self->sibling_count++; 1961 } 1962 1963 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 1964 ASSERT_NE(ENOSYS, errno) { 1965 TH_LOG("Kernel does not support seccomp syscall!"); 1966 } 1967 ASSERT_EQ(0, ret) { 1968 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 1969 } 1970 1971 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FLAG_FILTER_TSYNC, 1972 &self->apply_prog); 1973 ASSERT_EQ(ret, self->sibling[0].system_tid) { 1974 TH_LOG("Did not fail on diverged sibling."); 1975 } 1976 sib = 1; 1977 if (ret == self->sibling[0].system_tid) 1978 sib = 0; 1979 1980 pthread_mutex_lock(&self->mutex); 1981 1982 /* Increment the other siblings num_waits so we can clean up 1983 * the one we just saw. 1984 */ 1985 self->sibling[!sib].num_waits += 1; 1986 1987 /* Signal the thread to clean up*/ 1988 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 1989 TH_LOG("cond broadcast non-zero"); 1990 } 1991 pthread_mutex_unlock(&self->mutex); 1992 pthread_join(self->sibling[sib].tid, &status); 1993 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 1994 /* Poll for actual task death. pthread_join doesn't guarantee it. */ 1995 while (!kill(self->sibling[sib].system_tid, 0)) 1996 sleep(0.1); 1997 /* Switch to the remaining sibling */ 1998 sib = !sib; 1999 2000 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FLAG_FILTER_TSYNC, 2001 &self->apply_prog); 2002 ASSERT_EQ(0, ret) { 2003 TH_LOG("Expected the remaining sibling to sync"); 2004 }; 2005 2006 pthread_mutex_lock(&self->mutex); 2007 2008 /* If remaining sibling didn't have a chance to wake up during 2009 * the first broadcast, manually reduce the num_waits now. 2010 */ 2011 if (self->sibling[sib].num_waits > 1) 2012 self->sibling[sib].num_waits = 1; 2013 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2014 TH_LOG("cond broadcast non-zero"); 2015 } 2016 pthread_mutex_unlock(&self->mutex); 2017 pthread_join(self->sibling[sib].tid, &status); 2018 EXPECT_EQ(0, (long)status); 2019 /* Poll for actual task death. pthread_join doesn't guarantee it. */ 2020 while (!kill(self->sibling[sib].system_tid, 0)) 2021 sleep(0.1); 2022 2023 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FLAG_FILTER_TSYNC, 2024 &self->apply_prog); 2025 ASSERT_EQ(0, ret); /* just us chickens */ 2026 } 2027 2028 /* Make sure restarted syscalls are seen directly as "restart_syscall". */ 2029 TEST(syscall_restart) 2030 { 2031 long ret; 2032 unsigned long msg; 2033 pid_t child_pid; 2034 int pipefd[2]; 2035 int status; 2036 siginfo_t info = { }; 2037 struct sock_filter filter[] = { 2038 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2039 offsetof(struct seccomp_data, nr)), 2040 2041 #ifdef __NR_sigreturn 2042 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 6, 0), 2043 #endif 2044 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 5, 0), 2045 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 4, 0), 2046 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 3, 0), 2047 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 4, 0), 2048 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0), 2049 2050 /* Allow __NR_write for easy logging. */ 2051 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1), 2052 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2053 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2054 /* The nanosleep jump target. */ 2055 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100), 2056 /* The restart_syscall jump target. */ 2057 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200), 2058 }; 2059 struct sock_fprog prog = { 2060 .len = (unsigned short)ARRAY_SIZE(filter), 2061 .filter = filter, 2062 }; 2063 #if defined(__arm__) 2064 struct utsname utsbuf; 2065 #endif 2066 2067 ASSERT_EQ(0, pipe(pipefd)); 2068 2069 child_pid = fork(); 2070 ASSERT_LE(0, child_pid); 2071 if (child_pid == 0) { 2072 /* Child uses EXPECT not ASSERT to deliver status correctly. */ 2073 char buf = ' '; 2074 struct timespec timeout = { }; 2075 2076 /* Attach parent as tracer and stop. */ 2077 EXPECT_EQ(0, ptrace(PTRACE_TRACEME)); 2078 EXPECT_EQ(0, raise(SIGSTOP)); 2079 2080 EXPECT_EQ(0, close(pipefd[1])); 2081 2082 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2083 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2084 } 2085 2086 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2087 EXPECT_EQ(0, ret) { 2088 TH_LOG("Failed to install filter!"); 2089 } 2090 2091 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) { 2092 TH_LOG("Failed to read() sync from parent"); 2093 } 2094 EXPECT_EQ('.', buf) { 2095 TH_LOG("Failed to get sync data from read()"); 2096 } 2097 2098 /* Start nanosleep to be interrupted. */ 2099 timeout.tv_sec = 1; 2100 errno = 0; 2101 EXPECT_EQ(0, nanosleep(&timeout, NULL)) { 2102 TH_LOG("Call to nanosleep() failed (errno %d)", errno); 2103 } 2104 2105 /* Read final sync from parent. */ 2106 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) { 2107 TH_LOG("Failed final read() from parent"); 2108 } 2109 EXPECT_EQ('!', buf) { 2110 TH_LOG("Failed to get final data from read()"); 2111 } 2112 2113 /* Directly report the status of our test harness results. */ 2114 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS 2115 : EXIT_FAILURE); 2116 } 2117 EXPECT_EQ(0, close(pipefd[0])); 2118 2119 /* Attach to child, setup options, and release. */ 2120 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2121 ASSERT_EQ(true, WIFSTOPPED(status)); 2122 ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL, 2123 PTRACE_O_TRACESECCOMP)); 2124 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 2125 ASSERT_EQ(1, write(pipefd[1], ".", 1)); 2126 2127 /* Wait for nanosleep() to start. */ 2128 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2129 ASSERT_EQ(true, WIFSTOPPED(status)); 2130 ASSERT_EQ(SIGTRAP, WSTOPSIG(status)); 2131 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16)); 2132 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg)); 2133 ASSERT_EQ(0x100, msg); 2134 EXPECT_EQ(__NR_nanosleep, get_syscall(_metadata, child_pid)); 2135 2136 /* Might as well check siginfo for sanity while we're here. */ 2137 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info)); 2138 ASSERT_EQ(SIGTRAP, info.si_signo); 2139 ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code); 2140 EXPECT_EQ(0, info.si_errno); 2141 EXPECT_EQ(getuid(), info.si_uid); 2142 /* Verify signal delivery came from child (seccomp-triggered). */ 2143 EXPECT_EQ(child_pid, info.si_pid); 2144 2145 /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */ 2146 ASSERT_EQ(0, kill(child_pid, SIGSTOP)); 2147 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 2148 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2149 ASSERT_EQ(true, WIFSTOPPED(status)); 2150 ASSERT_EQ(SIGSTOP, WSTOPSIG(status)); 2151 /* Verify signal delivery came from parent now. */ 2152 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info)); 2153 EXPECT_EQ(getpid(), info.si_pid); 2154 2155 /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */ 2156 ASSERT_EQ(0, kill(child_pid, SIGCONT)); 2157 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 2158 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2159 ASSERT_EQ(true, WIFSTOPPED(status)); 2160 ASSERT_EQ(SIGCONT, WSTOPSIG(status)); 2161 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 2162 2163 /* Wait for restart_syscall() to start. */ 2164 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2165 ASSERT_EQ(true, WIFSTOPPED(status)); 2166 ASSERT_EQ(SIGTRAP, WSTOPSIG(status)); 2167 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16)); 2168 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg)); 2169 2170 ASSERT_EQ(0x200, msg); 2171 ret = get_syscall(_metadata, child_pid); 2172 #if defined(__arm__) 2173 /* 2174 * FIXME: 2175 * - native ARM registers do NOT expose true syscall. 2176 * - compat ARM registers on ARM64 DO expose true syscall. 2177 */ 2178 ASSERT_EQ(0, uname(&utsbuf)); 2179 if (strncmp(utsbuf.machine, "arm", 3) == 0) { 2180 EXPECT_EQ(__NR_nanosleep, ret); 2181 } else 2182 #endif 2183 { 2184 EXPECT_EQ(__NR_restart_syscall, ret); 2185 } 2186 2187 /* Write again to end test. */ 2188 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 2189 ASSERT_EQ(1, write(pipefd[1], "!", 1)); 2190 EXPECT_EQ(0, close(pipefd[1])); 2191 2192 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2193 if (WIFSIGNALED(status) || WEXITSTATUS(status)) 2194 _metadata->passed = 0; 2195 } 2196 2197 /* 2198 * TODO: 2199 * - add microbenchmarks 2200 * - expand NNP testing 2201 * - better arch-specific TRACE and TRAP handlers. 2202 * - endianness checking when appropriate 2203 * - 64-bit arg prodding 2204 * - arch value testing (x86 modes especially) 2205 * - ... 2206 */ 2207 2208 TEST_HARNESS_MAIN 2209