1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (c) 2012 The Chromium OS Authors. All rights reserved. 4 * 5 * Test code for seccomp bpf. 6 */ 7 8 #define _GNU_SOURCE 9 #include <sys/types.h> 10 11 /* 12 * glibc 2.26 and later have SIGSYS in siginfo_t. Before that, 13 * we need to use the kernel's siginfo.h file and trick glibc 14 * into accepting it. 15 */ 16 #if !__GLIBC_PREREQ(2, 26) 17 # include <asm/siginfo.h> 18 # define __have_siginfo_t 1 19 # define __have_sigval_t 1 20 # define __have_sigevent_t 1 21 #endif 22 23 #include <errno.h> 24 #include <linux/filter.h> 25 #include <sys/prctl.h> 26 #include <sys/ptrace.h> 27 #include <sys/user.h> 28 #include <linux/prctl.h> 29 #include <linux/ptrace.h> 30 #include <linux/seccomp.h> 31 #include <pthread.h> 32 #include <semaphore.h> 33 #include <signal.h> 34 #include <stddef.h> 35 #include <stdbool.h> 36 #include <string.h> 37 #include <time.h> 38 #include <limits.h> 39 #include <linux/elf.h> 40 #include <sys/uio.h> 41 #include <sys/utsname.h> 42 #include <sys/fcntl.h> 43 #include <sys/mman.h> 44 #include <sys/times.h> 45 #include <sys/socket.h> 46 #include <sys/ioctl.h> 47 #include <linux/kcmp.h> 48 #include <sys/resource.h> 49 50 #include <unistd.h> 51 #include <sys/syscall.h> 52 #include <poll.h> 53 54 #include "../kselftest_harness.h" 55 #include "../clone3/clone3_selftests.h" 56 57 /* Attempt to de-conflict with the selftests tree. */ 58 #ifndef SKIP 59 #define SKIP(s, ...) XFAIL(s, ##__VA_ARGS__) 60 #endif 61 62 #ifndef PR_SET_PTRACER 63 # define PR_SET_PTRACER 0x59616d61 64 #endif 65 66 #ifndef PR_SET_NO_NEW_PRIVS 67 #define PR_SET_NO_NEW_PRIVS 38 68 #define PR_GET_NO_NEW_PRIVS 39 69 #endif 70 71 #ifndef PR_SECCOMP_EXT 72 #define PR_SECCOMP_EXT 43 73 #endif 74 75 #ifndef SECCOMP_EXT_ACT 76 #define SECCOMP_EXT_ACT 1 77 #endif 78 79 #ifndef SECCOMP_EXT_ACT_TSYNC 80 #define SECCOMP_EXT_ACT_TSYNC 1 81 #endif 82 83 #ifndef SECCOMP_MODE_STRICT 84 #define SECCOMP_MODE_STRICT 1 85 #endif 86 87 #ifndef SECCOMP_MODE_FILTER 88 #define SECCOMP_MODE_FILTER 2 89 #endif 90 91 #ifndef SECCOMP_RET_ALLOW 92 struct seccomp_data { 93 int nr; 94 __u32 arch; 95 __u64 instruction_pointer; 96 __u64 args[6]; 97 }; 98 #endif 99 100 #ifndef SECCOMP_RET_KILL_PROCESS 101 #define SECCOMP_RET_KILL_PROCESS 0x80000000U /* kill the process */ 102 #define SECCOMP_RET_KILL_THREAD 0x00000000U /* kill the thread */ 103 #endif 104 #ifndef SECCOMP_RET_KILL 105 #define SECCOMP_RET_KILL SECCOMP_RET_KILL_THREAD 106 #define SECCOMP_RET_TRAP 0x00030000U /* disallow and force a SIGSYS */ 107 #define SECCOMP_RET_ERRNO 0x00050000U /* returns an errno */ 108 #define SECCOMP_RET_TRACE 0x7ff00000U /* pass to a tracer or disallow */ 109 #define SECCOMP_RET_ALLOW 0x7fff0000U /* allow */ 110 #endif 111 #ifndef SECCOMP_RET_LOG 112 #define SECCOMP_RET_LOG 0x7ffc0000U /* allow after logging */ 113 #endif 114 115 #ifndef __NR_seccomp 116 # if defined(__i386__) 117 # define __NR_seccomp 354 118 # elif defined(__x86_64__) 119 # define __NR_seccomp 317 120 # elif defined(__arm__) 121 # define __NR_seccomp 383 122 # elif defined(__aarch64__) 123 # define __NR_seccomp 277 124 # elif defined(__riscv) 125 # define __NR_seccomp 277 126 # elif defined(__hppa__) 127 # define __NR_seccomp 338 128 # elif defined(__powerpc__) 129 # define __NR_seccomp 358 130 # elif defined(__s390__) 131 # define __NR_seccomp 348 132 # elif defined(__xtensa__) 133 # define __NR_seccomp 337 134 # else 135 # warning "seccomp syscall number unknown for this architecture" 136 # define __NR_seccomp 0xffff 137 # endif 138 #endif 139 140 #ifndef SECCOMP_SET_MODE_STRICT 141 #define SECCOMP_SET_MODE_STRICT 0 142 #endif 143 144 #ifndef SECCOMP_SET_MODE_FILTER 145 #define SECCOMP_SET_MODE_FILTER 1 146 #endif 147 148 #ifndef SECCOMP_GET_ACTION_AVAIL 149 #define SECCOMP_GET_ACTION_AVAIL 2 150 #endif 151 152 #ifndef SECCOMP_GET_NOTIF_SIZES 153 #define SECCOMP_GET_NOTIF_SIZES 3 154 #endif 155 156 #ifndef SECCOMP_FILTER_FLAG_TSYNC 157 #define SECCOMP_FILTER_FLAG_TSYNC (1UL << 0) 158 #endif 159 160 #ifndef SECCOMP_FILTER_FLAG_LOG 161 #define SECCOMP_FILTER_FLAG_LOG (1UL << 1) 162 #endif 163 164 #ifndef SECCOMP_FILTER_FLAG_SPEC_ALLOW 165 #define SECCOMP_FILTER_FLAG_SPEC_ALLOW (1UL << 2) 166 #endif 167 168 #ifndef PTRACE_SECCOMP_GET_METADATA 169 #define PTRACE_SECCOMP_GET_METADATA 0x420d 170 171 struct seccomp_metadata { 172 __u64 filter_off; /* Input: which filter */ 173 __u64 flags; /* Output: filter's flags */ 174 }; 175 #endif 176 177 #ifndef SECCOMP_FILTER_FLAG_NEW_LISTENER 178 #define SECCOMP_FILTER_FLAG_NEW_LISTENER (1UL << 3) 179 #endif 180 181 #ifndef SECCOMP_RET_USER_NOTIF 182 #define SECCOMP_RET_USER_NOTIF 0x7fc00000U 183 184 #define SECCOMP_IOC_MAGIC '!' 185 #define SECCOMP_IO(nr) _IO(SECCOMP_IOC_MAGIC, nr) 186 #define SECCOMP_IOR(nr, type) _IOR(SECCOMP_IOC_MAGIC, nr, type) 187 #define SECCOMP_IOW(nr, type) _IOW(SECCOMP_IOC_MAGIC, nr, type) 188 #define SECCOMP_IOWR(nr, type) _IOWR(SECCOMP_IOC_MAGIC, nr, type) 189 190 /* Flags for seccomp notification fd ioctl. */ 191 #define SECCOMP_IOCTL_NOTIF_RECV SECCOMP_IOWR(0, struct seccomp_notif) 192 #define SECCOMP_IOCTL_NOTIF_SEND SECCOMP_IOWR(1, \ 193 struct seccomp_notif_resp) 194 #define SECCOMP_IOCTL_NOTIF_ID_VALID SECCOMP_IOW(2, __u64) 195 196 struct seccomp_notif { 197 __u64 id; 198 __u32 pid; 199 __u32 flags; 200 struct seccomp_data data; 201 }; 202 203 struct seccomp_notif_resp { 204 __u64 id; 205 __s64 val; 206 __s32 error; 207 __u32 flags; 208 }; 209 210 struct seccomp_notif_sizes { 211 __u16 seccomp_notif; 212 __u16 seccomp_notif_resp; 213 __u16 seccomp_data; 214 }; 215 #endif 216 217 #ifndef SECCOMP_IOCTL_NOTIF_ADDFD 218 /* On success, the return value is the remote process's added fd number */ 219 #define SECCOMP_IOCTL_NOTIF_ADDFD SECCOMP_IOW(3, \ 220 struct seccomp_notif_addfd) 221 222 /* valid flags for seccomp_notif_addfd */ 223 #define SECCOMP_ADDFD_FLAG_SETFD (1UL << 0) /* Specify remote fd */ 224 225 struct seccomp_notif_addfd { 226 __u64 id; 227 __u32 flags; 228 __u32 srcfd; 229 __u32 newfd; 230 __u32 newfd_flags; 231 }; 232 #endif 233 234 struct seccomp_notif_addfd_small { 235 __u64 id; 236 char weird[4]; 237 }; 238 #define SECCOMP_IOCTL_NOTIF_ADDFD_SMALL \ 239 SECCOMP_IOW(3, struct seccomp_notif_addfd_small) 240 241 struct seccomp_notif_addfd_big { 242 union { 243 struct seccomp_notif_addfd addfd; 244 char buf[sizeof(struct seccomp_notif_addfd) + 8]; 245 }; 246 }; 247 #define SECCOMP_IOCTL_NOTIF_ADDFD_BIG \ 248 SECCOMP_IOWR(3, struct seccomp_notif_addfd_big) 249 250 #ifndef PTRACE_EVENTMSG_SYSCALL_ENTRY 251 #define PTRACE_EVENTMSG_SYSCALL_ENTRY 1 252 #define PTRACE_EVENTMSG_SYSCALL_EXIT 2 253 #endif 254 255 #ifndef SECCOMP_USER_NOTIF_FLAG_CONTINUE 256 #define SECCOMP_USER_NOTIF_FLAG_CONTINUE 0x00000001 257 #endif 258 259 #ifndef SECCOMP_FILTER_FLAG_TSYNC_ESRCH 260 #define SECCOMP_FILTER_FLAG_TSYNC_ESRCH (1UL << 4) 261 #endif 262 263 #ifndef seccomp 264 int seccomp(unsigned int op, unsigned int flags, void *args) 265 { 266 errno = 0; 267 return syscall(__NR_seccomp, op, flags, args); 268 } 269 #endif 270 271 #if __BYTE_ORDER == __LITTLE_ENDIAN 272 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n])) 273 #elif __BYTE_ORDER == __BIG_ENDIAN 274 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32)) 275 #else 276 #error "wut? Unknown __BYTE_ORDER?!" 277 #endif 278 279 #define SIBLING_EXIT_UNKILLED 0xbadbeef 280 #define SIBLING_EXIT_FAILURE 0xbadface 281 #define SIBLING_EXIT_NEWPRIVS 0xbadfeed 282 283 static int __filecmp(pid_t pid1, pid_t pid2, int fd1, int fd2) 284 { 285 #ifdef __NR_kcmp 286 errno = 0; 287 return syscall(__NR_kcmp, pid1, pid2, KCMP_FILE, fd1, fd2); 288 #else 289 errno = ENOSYS; 290 return -1; 291 #endif 292 } 293 294 /* Have TH_LOG report actual location filecmp() is used. */ 295 #define filecmp(pid1, pid2, fd1, fd2) ({ \ 296 int _ret; \ 297 \ 298 _ret = __filecmp(pid1, pid2, fd1, fd2); \ 299 if (_ret != 0) { \ 300 if (_ret < 0 && errno == ENOSYS) { \ 301 TH_LOG("kcmp() syscall missing (test is less accurate)");\ 302 _ret = 0; \ 303 } \ 304 } \ 305 _ret; }) 306 307 TEST(kcmp) 308 { 309 int ret; 310 311 ret = __filecmp(getpid(), getpid(), 1, 1); 312 EXPECT_EQ(ret, 0); 313 if (ret != 0 && errno == ENOSYS) 314 SKIP(return, "Kernel does not support kcmp() (missing CONFIG_CHECKPOINT_RESTORE?)"); 315 } 316 317 TEST(mode_strict_support) 318 { 319 long ret; 320 321 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL); 322 ASSERT_EQ(0, ret) { 323 TH_LOG("Kernel does not support CONFIG_SECCOMP"); 324 } 325 syscall(__NR_exit, 0); 326 } 327 328 TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL) 329 { 330 long ret; 331 332 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL); 333 ASSERT_EQ(0, ret) { 334 TH_LOG("Kernel does not support CONFIG_SECCOMP"); 335 } 336 syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER, 337 NULL, NULL, NULL); 338 EXPECT_FALSE(true) { 339 TH_LOG("Unreachable!"); 340 } 341 } 342 343 /* Note! This doesn't test no new privs behavior */ 344 TEST(no_new_privs_support) 345 { 346 long ret; 347 348 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 349 EXPECT_EQ(0, ret) { 350 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 351 } 352 } 353 354 /* Tests kernel support by checking for a copy_from_user() fault on NULL. */ 355 TEST(mode_filter_support) 356 { 357 long ret; 358 359 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 360 ASSERT_EQ(0, ret) { 361 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 362 } 363 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL); 364 EXPECT_EQ(-1, ret); 365 EXPECT_EQ(EFAULT, errno) { 366 TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!"); 367 } 368 } 369 370 TEST(mode_filter_without_nnp) 371 { 372 struct sock_filter filter[] = { 373 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 374 }; 375 struct sock_fprog prog = { 376 .len = (unsigned short)ARRAY_SIZE(filter), 377 .filter = filter, 378 }; 379 long ret; 380 381 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0); 382 ASSERT_LE(0, ret) { 383 TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS"); 384 } 385 errno = 0; 386 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 387 /* Succeeds with CAP_SYS_ADMIN, fails without */ 388 /* TODO(wad) check caps not euid */ 389 if (geteuid()) { 390 EXPECT_EQ(-1, ret); 391 EXPECT_EQ(EACCES, errno); 392 } else { 393 EXPECT_EQ(0, ret); 394 } 395 } 396 397 #define MAX_INSNS_PER_PATH 32768 398 399 TEST(filter_size_limits) 400 { 401 int i; 402 int count = BPF_MAXINSNS + 1; 403 struct sock_filter allow[] = { 404 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 405 }; 406 struct sock_filter *filter; 407 struct sock_fprog prog = { }; 408 long ret; 409 410 filter = calloc(count, sizeof(*filter)); 411 ASSERT_NE(NULL, filter); 412 413 for (i = 0; i < count; i++) 414 filter[i] = allow[0]; 415 416 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 417 ASSERT_EQ(0, ret); 418 419 prog.filter = filter; 420 prog.len = count; 421 422 /* Too many filter instructions in a single filter. */ 423 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 424 ASSERT_NE(0, ret) { 425 TH_LOG("Installing %d insn filter was allowed", prog.len); 426 } 427 428 /* One less is okay, though. */ 429 prog.len -= 1; 430 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 431 ASSERT_EQ(0, ret) { 432 TH_LOG("Installing %d insn filter wasn't allowed", prog.len); 433 } 434 } 435 436 TEST(filter_chain_limits) 437 { 438 int i; 439 int count = BPF_MAXINSNS; 440 struct sock_filter allow[] = { 441 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 442 }; 443 struct sock_filter *filter; 444 struct sock_fprog prog = { }; 445 long ret; 446 447 filter = calloc(count, sizeof(*filter)); 448 ASSERT_NE(NULL, filter); 449 450 for (i = 0; i < count; i++) 451 filter[i] = allow[0]; 452 453 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 454 ASSERT_EQ(0, ret); 455 456 prog.filter = filter; 457 prog.len = 1; 458 459 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 460 ASSERT_EQ(0, ret); 461 462 prog.len = count; 463 464 /* Too many total filter instructions. */ 465 for (i = 0; i < MAX_INSNS_PER_PATH; i++) { 466 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 467 if (ret != 0) 468 break; 469 } 470 ASSERT_NE(0, ret) { 471 TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)", 472 i, count, i * (count + 4)); 473 } 474 } 475 476 TEST(mode_filter_cannot_move_to_strict) 477 { 478 struct sock_filter filter[] = { 479 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 480 }; 481 struct sock_fprog prog = { 482 .len = (unsigned short)ARRAY_SIZE(filter), 483 .filter = filter, 484 }; 485 long ret; 486 487 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 488 ASSERT_EQ(0, ret); 489 490 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 491 ASSERT_EQ(0, ret); 492 493 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0); 494 EXPECT_EQ(-1, ret); 495 EXPECT_EQ(EINVAL, errno); 496 } 497 498 499 TEST(mode_filter_get_seccomp) 500 { 501 struct sock_filter filter[] = { 502 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 503 }; 504 struct sock_fprog prog = { 505 .len = (unsigned short)ARRAY_SIZE(filter), 506 .filter = filter, 507 }; 508 long ret; 509 510 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 511 ASSERT_EQ(0, ret); 512 513 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0); 514 EXPECT_EQ(0, ret); 515 516 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 517 ASSERT_EQ(0, ret); 518 519 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0); 520 EXPECT_EQ(2, ret); 521 } 522 523 524 TEST(ALLOW_all) 525 { 526 struct sock_filter filter[] = { 527 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 528 }; 529 struct sock_fprog prog = { 530 .len = (unsigned short)ARRAY_SIZE(filter), 531 .filter = filter, 532 }; 533 long ret; 534 535 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 536 ASSERT_EQ(0, ret); 537 538 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 539 ASSERT_EQ(0, ret); 540 } 541 542 TEST(empty_prog) 543 { 544 struct sock_filter filter[] = { 545 }; 546 struct sock_fprog prog = { 547 .len = (unsigned short)ARRAY_SIZE(filter), 548 .filter = filter, 549 }; 550 long ret; 551 552 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 553 ASSERT_EQ(0, ret); 554 555 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 556 EXPECT_EQ(-1, ret); 557 EXPECT_EQ(EINVAL, errno); 558 } 559 560 TEST(log_all) 561 { 562 struct sock_filter filter[] = { 563 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG), 564 }; 565 struct sock_fprog prog = { 566 .len = (unsigned short)ARRAY_SIZE(filter), 567 .filter = filter, 568 }; 569 long ret; 570 pid_t parent = getppid(); 571 572 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 573 ASSERT_EQ(0, ret); 574 575 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 576 ASSERT_EQ(0, ret); 577 578 /* getppid() should succeed and be logged (no check for logging) */ 579 EXPECT_EQ(parent, syscall(__NR_getppid)); 580 } 581 582 TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS) 583 { 584 struct sock_filter filter[] = { 585 BPF_STMT(BPF_RET|BPF_K, 0x10000000U), 586 }; 587 struct sock_fprog prog = { 588 .len = (unsigned short)ARRAY_SIZE(filter), 589 .filter = filter, 590 }; 591 long ret; 592 593 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 594 ASSERT_EQ(0, ret); 595 596 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 597 ASSERT_EQ(0, ret); 598 EXPECT_EQ(0, syscall(__NR_getpid)) { 599 TH_LOG("getpid() shouldn't ever return"); 600 } 601 } 602 603 /* return code >= 0x80000000 is unused. */ 604 TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS) 605 { 606 struct sock_filter filter[] = { 607 BPF_STMT(BPF_RET|BPF_K, 0x90000000U), 608 }; 609 struct sock_fprog prog = { 610 .len = (unsigned short)ARRAY_SIZE(filter), 611 .filter = filter, 612 }; 613 long ret; 614 615 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 616 ASSERT_EQ(0, ret); 617 618 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 619 ASSERT_EQ(0, ret); 620 EXPECT_EQ(0, syscall(__NR_getpid)) { 621 TH_LOG("getpid() shouldn't ever return"); 622 } 623 } 624 625 TEST_SIGNAL(KILL_all, SIGSYS) 626 { 627 struct sock_filter filter[] = { 628 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 629 }; 630 struct sock_fprog prog = { 631 .len = (unsigned short)ARRAY_SIZE(filter), 632 .filter = filter, 633 }; 634 long ret; 635 636 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 637 ASSERT_EQ(0, ret); 638 639 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 640 ASSERT_EQ(0, ret); 641 } 642 643 TEST_SIGNAL(KILL_one, SIGSYS) 644 { 645 struct sock_filter filter[] = { 646 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 647 offsetof(struct seccomp_data, nr)), 648 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 649 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 650 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 651 }; 652 struct sock_fprog prog = { 653 .len = (unsigned short)ARRAY_SIZE(filter), 654 .filter = filter, 655 }; 656 long ret; 657 pid_t parent = getppid(); 658 659 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 660 ASSERT_EQ(0, ret); 661 662 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 663 ASSERT_EQ(0, ret); 664 665 EXPECT_EQ(parent, syscall(__NR_getppid)); 666 /* getpid() should never return. */ 667 EXPECT_EQ(0, syscall(__NR_getpid)); 668 } 669 670 TEST_SIGNAL(KILL_one_arg_one, SIGSYS) 671 { 672 void *fatal_address; 673 struct sock_filter filter[] = { 674 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 675 offsetof(struct seccomp_data, nr)), 676 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0), 677 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 678 /* Only both with lower 32-bit for now. */ 679 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)), 680 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 681 (unsigned long)&fatal_address, 0, 1), 682 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 683 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 684 }; 685 struct sock_fprog prog = { 686 .len = (unsigned short)ARRAY_SIZE(filter), 687 .filter = filter, 688 }; 689 long ret; 690 pid_t parent = getppid(); 691 struct tms timebuf; 692 clock_t clock = times(&timebuf); 693 694 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 695 ASSERT_EQ(0, ret); 696 697 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 698 ASSERT_EQ(0, ret); 699 700 EXPECT_EQ(parent, syscall(__NR_getppid)); 701 EXPECT_LE(clock, syscall(__NR_times, &timebuf)); 702 /* times() should never return. */ 703 EXPECT_EQ(0, syscall(__NR_times, &fatal_address)); 704 } 705 706 TEST_SIGNAL(KILL_one_arg_six, SIGSYS) 707 { 708 #ifndef __NR_mmap2 709 int sysno = __NR_mmap; 710 #else 711 int sysno = __NR_mmap2; 712 #endif 713 struct sock_filter filter[] = { 714 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 715 offsetof(struct seccomp_data, nr)), 716 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0), 717 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 718 /* Only both with lower 32-bit for now. */ 719 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)), 720 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1), 721 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 722 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 723 }; 724 struct sock_fprog prog = { 725 .len = (unsigned short)ARRAY_SIZE(filter), 726 .filter = filter, 727 }; 728 long ret; 729 pid_t parent = getppid(); 730 int fd; 731 void *map1, *map2; 732 int page_size = sysconf(_SC_PAGESIZE); 733 734 ASSERT_LT(0, page_size); 735 736 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 737 ASSERT_EQ(0, ret); 738 739 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 740 ASSERT_EQ(0, ret); 741 742 fd = open("/dev/zero", O_RDONLY); 743 ASSERT_NE(-1, fd); 744 745 EXPECT_EQ(parent, syscall(__NR_getppid)); 746 map1 = (void *)syscall(sysno, 747 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size); 748 EXPECT_NE(MAP_FAILED, map1); 749 /* mmap2() should never return. */ 750 map2 = (void *)syscall(sysno, 751 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE); 752 EXPECT_EQ(MAP_FAILED, map2); 753 754 /* The test failed, so clean up the resources. */ 755 munmap(map1, page_size); 756 munmap(map2, page_size); 757 close(fd); 758 } 759 760 /* This is a thread task to die via seccomp filter violation. */ 761 void *kill_thread(void *data) 762 { 763 bool die = (bool)data; 764 765 if (die) { 766 prctl(PR_GET_SECCOMP, 0, 0, 0, 0); 767 return (void *)SIBLING_EXIT_FAILURE; 768 } 769 770 return (void *)SIBLING_EXIT_UNKILLED; 771 } 772 773 /* Prepare a thread that will kill itself or both of us. */ 774 void kill_thread_or_group(struct __test_metadata *_metadata, bool kill_process) 775 { 776 pthread_t thread; 777 void *status; 778 /* Kill only when calling __NR_prctl. */ 779 struct sock_filter filter_thread[] = { 780 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 781 offsetof(struct seccomp_data, nr)), 782 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1), 783 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD), 784 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 785 }; 786 struct sock_fprog prog_thread = { 787 .len = (unsigned short)ARRAY_SIZE(filter_thread), 788 .filter = filter_thread, 789 }; 790 struct sock_filter filter_process[] = { 791 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 792 offsetof(struct seccomp_data, nr)), 793 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1), 794 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_PROCESS), 795 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 796 }; 797 struct sock_fprog prog_process = { 798 .len = (unsigned short)ARRAY_SIZE(filter_process), 799 .filter = filter_process, 800 }; 801 802 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 803 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 804 } 805 806 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, 807 kill_process ? &prog_process : &prog_thread)); 808 809 /* 810 * Add the KILL_THREAD rule again to make sure that the KILL_PROCESS 811 * flag cannot be downgraded by a new filter. 812 */ 813 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog_thread)); 814 815 /* Start a thread that will exit immediately. */ 816 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)false)); 817 ASSERT_EQ(0, pthread_join(thread, &status)); 818 ASSERT_EQ(SIBLING_EXIT_UNKILLED, (unsigned long)status); 819 820 /* Start a thread that will die immediately. */ 821 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)true)); 822 ASSERT_EQ(0, pthread_join(thread, &status)); 823 ASSERT_NE(SIBLING_EXIT_FAILURE, (unsigned long)status); 824 825 /* 826 * If we get here, only the spawned thread died. Let the parent know 827 * the whole process didn't die (i.e. this thread, the spawner, 828 * stayed running). 829 */ 830 exit(42); 831 } 832 833 TEST(KILL_thread) 834 { 835 int status; 836 pid_t child_pid; 837 838 child_pid = fork(); 839 ASSERT_LE(0, child_pid); 840 if (child_pid == 0) { 841 kill_thread_or_group(_metadata, false); 842 _exit(38); 843 } 844 845 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 846 847 /* If only the thread was killed, we'll see exit 42. */ 848 ASSERT_TRUE(WIFEXITED(status)); 849 ASSERT_EQ(42, WEXITSTATUS(status)); 850 } 851 852 TEST(KILL_process) 853 { 854 int status; 855 pid_t child_pid; 856 857 child_pid = fork(); 858 ASSERT_LE(0, child_pid); 859 if (child_pid == 0) { 860 kill_thread_or_group(_metadata, true); 861 _exit(38); 862 } 863 864 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 865 866 /* If the entire process was killed, we'll see SIGSYS. */ 867 ASSERT_TRUE(WIFSIGNALED(status)); 868 ASSERT_EQ(SIGSYS, WTERMSIG(status)); 869 } 870 871 /* TODO(wad) add 64-bit versus 32-bit arg tests. */ 872 TEST(arg_out_of_range) 873 { 874 struct sock_filter filter[] = { 875 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)), 876 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 877 }; 878 struct sock_fprog prog = { 879 .len = (unsigned short)ARRAY_SIZE(filter), 880 .filter = filter, 881 }; 882 long ret; 883 884 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 885 ASSERT_EQ(0, ret); 886 887 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 888 EXPECT_EQ(-1, ret); 889 EXPECT_EQ(EINVAL, errno); 890 } 891 892 #define ERRNO_FILTER(name, errno) \ 893 struct sock_filter _read_filter_##name[] = { \ 894 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, \ 895 offsetof(struct seccomp_data, nr)), \ 896 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), \ 897 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | errno), \ 898 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), \ 899 }; \ 900 struct sock_fprog prog_##name = { \ 901 .len = (unsigned short)ARRAY_SIZE(_read_filter_##name), \ 902 .filter = _read_filter_##name, \ 903 } 904 905 /* Make sure basic errno values are correctly passed through a filter. */ 906 TEST(ERRNO_valid) 907 { 908 ERRNO_FILTER(valid, E2BIG); 909 long ret; 910 pid_t parent = getppid(); 911 912 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 913 ASSERT_EQ(0, ret); 914 915 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_valid); 916 ASSERT_EQ(0, ret); 917 918 EXPECT_EQ(parent, syscall(__NR_getppid)); 919 EXPECT_EQ(-1, read(0, NULL, 0)); 920 EXPECT_EQ(E2BIG, errno); 921 } 922 923 /* Make sure an errno of zero is correctly handled by the arch code. */ 924 TEST(ERRNO_zero) 925 { 926 ERRNO_FILTER(zero, 0); 927 long ret; 928 pid_t parent = getppid(); 929 930 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 931 ASSERT_EQ(0, ret); 932 933 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_zero); 934 ASSERT_EQ(0, ret); 935 936 EXPECT_EQ(parent, syscall(__NR_getppid)); 937 /* "errno" of 0 is ok. */ 938 EXPECT_EQ(0, read(0, NULL, 0)); 939 } 940 941 /* 942 * The SECCOMP_RET_DATA mask is 16 bits wide, but errno is smaller. 943 * This tests that the errno value gets capped correctly, fixed by 944 * 580c57f10768 ("seccomp: cap SECCOMP_RET_ERRNO data to MAX_ERRNO"). 945 */ 946 TEST(ERRNO_capped) 947 { 948 ERRNO_FILTER(capped, 4096); 949 long ret; 950 pid_t parent = getppid(); 951 952 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 953 ASSERT_EQ(0, ret); 954 955 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_capped); 956 ASSERT_EQ(0, ret); 957 958 EXPECT_EQ(parent, syscall(__NR_getppid)); 959 EXPECT_EQ(-1, read(0, NULL, 0)); 960 EXPECT_EQ(4095, errno); 961 } 962 963 /* 964 * Filters are processed in reverse order: last applied is executed first. 965 * Since only the SECCOMP_RET_ACTION mask is tested for return values, the 966 * SECCOMP_RET_DATA mask results will follow the most recently applied 967 * matching filter return (and not the lowest or highest value). 968 */ 969 TEST(ERRNO_order) 970 { 971 ERRNO_FILTER(first, 11); 972 ERRNO_FILTER(second, 13); 973 ERRNO_FILTER(third, 12); 974 long ret; 975 pid_t parent = getppid(); 976 977 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 978 ASSERT_EQ(0, ret); 979 980 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_first); 981 ASSERT_EQ(0, ret); 982 983 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_second); 984 ASSERT_EQ(0, ret); 985 986 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_third); 987 ASSERT_EQ(0, ret); 988 989 EXPECT_EQ(parent, syscall(__NR_getppid)); 990 EXPECT_EQ(-1, read(0, NULL, 0)); 991 EXPECT_EQ(12, errno); 992 } 993 994 FIXTURE(TRAP) { 995 struct sock_fprog prog; 996 }; 997 998 FIXTURE_SETUP(TRAP) 999 { 1000 struct sock_filter filter[] = { 1001 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1002 offsetof(struct seccomp_data, nr)), 1003 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 1004 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP), 1005 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1006 }; 1007 1008 memset(&self->prog, 0, sizeof(self->prog)); 1009 self->prog.filter = malloc(sizeof(filter)); 1010 ASSERT_NE(NULL, self->prog.filter); 1011 memcpy(self->prog.filter, filter, sizeof(filter)); 1012 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 1013 } 1014 1015 FIXTURE_TEARDOWN(TRAP) 1016 { 1017 if (self->prog.filter) 1018 free(self->prog.filter); 1019 } 1020 1021 TEST_F_SIGNAL(TRAP, dfl, SIGSYS) 1022 { 1023 long ret; 1024 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->prog); 1029 ASSERT_EQ(0, ret); 1030 syscall(__NR_getpid); 1031 } 1032 1033 /* Ensure that SIGSYS overrides SIG_IGN */ 1034 TEST_F_SIGNAL(TRAP, ign, SIGSYS) 1035 { 1036 long ret; 1037 1038 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1039 ASSERT_EQ(0, ret); 1040 1041 signal(SIGSYS, SIG_IGN); 1042 1043 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 1044 ASSERT_EQ(0, ret); 1045 syscall(__NR_getpid); 1046 } 1047 1048 static siginfo_t TRAP_info; 1049 static volatile int TRAP_nr; 1050 static void TRAP_action(int nr, siginfo_t *info, void *void_context) 1051 { 1052 memcpy(&TRAP_info, info, sizeof(TRAP_info)); 1053 TRAP_nr = nr; 1054 } 1055 1056 TEST_F(TRAP, handler) 1057 { 1058 int ret, test; 1059 struct sigaction act; 1060 sigset_t mask; 1061 1062 memset(&act, 0, sizeof(act)); 1063 sigemptyset(&mask); 1064 sigaddset(&mask, SIGSYS); 1065 1066 act.sa_sigaction = &TRAP_action; 1067 act.sa_flags = SA_SIGINFO; 1068 ret = sigaction(SIGSYS, &act, NULL); 1069 ASSERT_EQ(0, ret) { 1070 TH_LOG("sigaction failed"); 1071 } 1072 ret = sigprocmask(SIG_UNBLOCK, &mask, NULL); 1073 ASSERT_EQ(0, ret) { 1074 TH_LOG("sigprocmask failed"); 1075 } 1076 1077 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1078 ASSERT_EQ(0, ret); 1079 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 1080 ASSERT_EQ(0, ret); 1081 TRAP_nr = 0; 1082 memset(&TRAP_info, 0, sizeof(TRAP_info)); 1083 /* Expect the registers to be rolled back. (nr = error) may vary 1084 * based on arch. */ 1085 ret = syscall(__NR_getpid); 1086 /* Silence gcc warning about volatile. */ 1087 test = TRAP_nr; 1088 EXPECT_EQ(SIGSYS, test); 1089 struct local_sigsys { 1090 void *_call_addr; /* calling user insn */ 1091 int _syscall; /* triggering system call number */ 1092 unsigned int _arch; /* AUDIT_ARCH_* of syscall */ 1093 } *sigsys = (struct local_sigsys *) 1094 #ifdef si_syscall 1095 &(TRAP_info.si_call_addr); 1096 #else 1097 &TRAP_info.si_pid; 1098 #endif 1099 EXPECT_EQ(__NR_getpid, sigsys->_syscall); 1100 /* Make sure arch is non-zero. */ 1101 EXPECT_NE(0, sigsys->_arch); 1102 EXPECT_NE(0, (unsigned long)sigsys->_call_addr); 1103 } 1104 1105 FIXTURE(precedence) { 1106 struct sock_fprog allow; 1107 struct sock_fprog log; 1108 struct sock_fprog trace; 1109 struct sock_fprog error; 1110 struct sock_fprog trap; 1111 struct sock_fprog kill; 1112 }; 1113 1114 FIXTURE_SETUP(precedence) 1115 { 1116 struct sock_filter allow_insns[] = { 1117 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1118 }; 1119 struct sock_filter log_insns[] = { 1120 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1121 offsetof(struct seccomp_data, nr)), 1122 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1123 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1124 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG), 1125 }; 1126 struct sock_filter trace_insns[] = { 1127 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1128 offsetof(struct seccomp_data, nr)), 1129 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1130 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1131 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE), 1132 }; 1133 struct sock_filter error_insns[] = { 1134 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1135 offsetof(struct seccomp_data, nr)), 1136 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1137 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1138 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO), 1139 }; 1140 struct sock_filter trap_insns[] = { 1141 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1142 offsetof(struct seccomp_data, nr)), 1143 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1144 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1145 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP), 1146 }; 1147 struct sock_filter kill_insns[] = { 1148 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1149 offsetof(struct seccomp_data, nr)), 1150 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1151 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1152 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 1153 }; 1154 1155 memset(self, 0, sizeof(*self)); 1156 #define FILTER_ALLOC(_x) \ 1157 self->_x.filter = malloc(sizeof(_x##_insns)); \ 1158 ASSERT_NE(NULL, self->_x.filter); \ 1159 memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \ 1160 self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns) 1161 FILTER_ALLOC(allow); 1162 FILTER_ALLOC(log); 1163 FILTER_ALLOC(trace); 1164 FILTER_ALLOC(error); 1165 FILTER_ALLOC(trap); 1166 FILTER_ALLOC(kill); 1167 } 1168 1169 FIXTURE_TEARDOWN(precedence) 1170 { 1171 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter) 1172 FILTER_FREE(allow); 1173 FILTER_FREE(log); 1174 FILTER_FREE(trace); 1175 FILTER_FREE(error); 1176 FILTER_FREE(trap); 1177 FILTER_FREE(kill); 1178 } 1179 1180 TEST_F(precedence, allow_ok) 1181 { 1182 pid_t parent, res = 0; 1183 long ret; 1184 1185 parent = getppid(); 1186 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1187 ASSERT_EQ(0, ret); 1188 1189 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1190 ASSERT_EQ(0, ret); 1191 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1192 ASSERT_EQ(0, ret); 1193 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1194 ASSERT_EQ(0, ret); 1195 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1196 ASSERT_EQ(0, ret); 1197 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1198 ASSERT_EQ(0, ret); 1199 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1200 ASSERT_EQ(0, ret); 1201 /* Should work just fine. */ 1202 res = syscall(__NR_getppid); 1203 EXPECT_EQ(parent, res); 1204 } 1205 1206 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS) 1207 { 1208 pid_t parent, res = 0; 1209 long ret; 1210 1211 parent = getppid(); 1212 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1213 ASSERT_EQ(0, ret); 1214 1215 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1216 ASSERT_EQ(0, ret); 1217 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1218 ASSERT_EQ(0, ret); 1219 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1220 ASSERT_EQ(0, ret); 1221 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1222 ASSERT_EQ(0, ret); 1223 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1224 ASSERT_EQ(0, ret); 1225 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1226 ASSERT_EQ(0, ret); 1227 /* Should work just fine. */ 1228 res = syscall(__NR_getppid); 1229 EXPECT_EQ(parent, res); 1230 /* getpid() should never return. */ 1231 res = syscall(__NR_getpid); 1232 EXPECT_EQ(0, res); 1233 } 1234 1235 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS) 1236 { 1237 pid_t parent; 1238 long ret; 1239 1240 parent = getppid(); 1241 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1242 ASSERT_EQ(0, ret); 1243 1244 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1245 ASSERT_EQ(0, ret); 1246 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1247 ASSERT_EQ(0, ret); 1248 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1249 ASSERT_EQ(0, ret); 1250 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1251 ASSERT_EQ(0, ret); 1252 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1253 ASSERT_EQ(0, ret); 1254 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1255 ASSERT_EQ(0, ret); 1256 /* Should work just fine. */ 1257 EXPECT_EQ(parent, syscall(__NR_getppid)); 1258 /* getpid() should never return. */ 1259 EXPECT_EQ(0, syscall(__NR_getpid)); 1260 } 1261 1262 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS) 1263 { 1264 pid_t parent; 1265 long ret; 1266 1267 parent = getppid(); 1268 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1269 ASSERT_EQ(0, ret); 1270 1271 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1272 ASSERT_EQ(0, ret); 1273 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1274 ASSERT_EQ(0, ret); 1275 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1276 ASSERT_EQ(0, ret); 1277 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1278 ASSERT_EQ(0, ret); 1279 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1280 ASSERT_EQ(0, ret); 1281 /* Should work just fine. */ 1282 EXPECT_EQ(parent, syscall(__NR_getppid)); 1283 /* getpid() should never return. */ 1284 EXPECT_EQ(0, syscall(__NR_getpid)); 1285 } 1286 1287 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS) 1288 { 1289 pid_t parent; 1290 long ret; 1291 1292 parent = getppid(); 1293 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1294 ASSERT_EQ(0, ret); 1295 1296 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1297 ASSERT_EQ(0, ret); 1298 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1299 ASSERT_EQ(0, ret); 1300 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1301 ASSERT_EQ(0, ret); 1302 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1303 ASSERT_EQ(0, ret); 1304 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1305 ASSERT_EQ(0, ret); 1306 /* Should work just fine. */ 1307 EXPECT_EQ(parent, syscall(__NR_getppid)); 1308 /* getpid() should never return. */ 1309 EXPECT_EQ(0, syscall(__NR_getpid)); 1310 } 1311 1312 TEST_F(precedence, errno_is_third) 1313 { 1314 pid_t parent; 1315 long ret; 1316 1317 parent = getppid(); 1318 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1319 ASSERT_EQ(0, ret); 1320 1321 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1322 ASSERT_EQ(0, ret); 1323 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1324 ASSERT_EQ(0, ret); 1325 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1326 ASSERT_EQ(0, ret); 1327 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1328 ASSERT_EQ(0, ret); 1329 /* Should work just fine. */ 1330 EXPECT_EQ(parent, syscall(__NR_getppid)); 1331 EXPECT_EQ(0, syscall(__NR_getpid)); 1332 } 1333 1334 TEST_F(precedence, errno_is_third_in_any_order) 1335 { 1336 pid_t parent; 1337 long ret; 1338 1339 parent = getppid(); 1340 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1341 ASSERT_EQ(0, ret); 1342 1343 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1344 ASSERT_EQ(0, ret); 1345 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1346 ASSERT_EQ(0, ret); 1347 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1348 ASSERT_EQ(0, ret); 1349 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1350 ASSERT_EQ(0, ret); 1351 /* Should work just fine. */ 1352 EXPECT_EQ(parent, syscall(__NR_getppid)); 1353 EXPECT_EQ(0, syscall(__NR_getpid)); 1354 } 1355 1356 TEST_F(precedence, trace_is_fourth) 1357 { 1358 pid_t parent; 1359 long ret; 1360 1361 parent = getppid(); 1362 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1363 ASSERT_EQ(0, ret); 1364 1365 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1366 ASSERT_EQ(0, ret); 1367 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1368 ASSERT_EQ(0, ret); 1369 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1370 ASSERT_EQ(0, ret); 1371 /* Should work just fine. */ 1372 EXPECT_EQ(parent, syscall(__NR_getppid)); 1373 /* No ptracer */ 1374 EXPECT_EQ(-1, syscall(__NR_getpid)); 1375 } 1376 1377 TEST_F(precedence, trace_is_fourth_in_any_order) 1378 { 1379 pid_t parent; 1380 long ret; 1381 1382 parent = getppid(); 1383 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1384 ASSERT_EQ(0, ret); 1385 1386 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1387 ASSERT_EQ(0, ret); 1388 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1389 ASSERT_EQ(0, ret); 1390 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1391 ASSERT_EQ(0, ret); 1392 /* Should work just fine. */ 1393 EXPECT_EQ(parent, syscall(__NR_getppid)); 1394 /* No ptracer */ 1395 EXPECT_EQ(-1, syscall(__NR_getpid)); 1396 } 1397 1398 TEST_F(precedence, log_is_fifth) 1399 { 1400 pid_t mypid, parent; 1401 long ret; 1402 1403 mypid = getpid(); 1404 parent = getppid(); 1405 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1406 ASSERT_EQ(0, ret); 1407 1408 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1409 ASSERT_EQ(0, ret); 1410 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1411 ASSERT_EQ(0, ret); 1412 /* Should work just fine. */ 1413 EXPECT_EQ(parent, syscall(__NR_getppid)); 1414 /* Should also work just fine */ 1415 EXPECT_EQ(mypid, syscall(__NR_getpid)); 1416 } 1417 1418 TEST_F(precedence, log_is_fifth_in_any_order) 1419 { 1420 pid_t mypid, parent; 1421 long ret; 1422 1423 mypid = getpid(); 1424 parent = getppid(); 1425 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1426 ASSERT_EQ(0, ret); 1427 1428 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1429 ASSERT_EQ(0, ret); 1430 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1431 ASSERT_EQ(0, ret); 1432 /* Should work just fine. */ 1433 EXPECT_EQ(parent, syscall(__NR_getppid)); 1434 /* Should also work just fine */ 1435 EXPECT_EQ(mypid, syscall(__NR_getpid)); 1436 } 1437 1438 #ifndef PTRACE_O_TRACESECCOMP 1439 #define PTRACE_O_TRACESECCOMP 0x00000080 1440 #endif 1441 1442 /* Catch the Ubuntu 12.04 value error. */ 1443 #if PTRACE_EVENT_SECCOMP != 7 1444 #undef PTRACE_EVENT_SECCOMP 1445 #endif 1446 1447 #ifndef PTRACE_EVENT_SECCOMP 1448 #define PTRACE_EVENT_SECCOMP 7 1449 #endif 1450 1451 #define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP) 1452 bool tracer_running; 1453 void tracer_stop(int sig) 1454 { 1455 tracer_running = false; 1456 } 1457 1458 typedef void tracer_func_t(struct __test_metadata *_metadata, 1459 pid_t tracee, int status, void *args); 1460 1461 void start_tracer(struct __test_metadata *_metadata, int fd, pid_t tracee, 1462 tracer_func_t tracer_func, void *args, bool ptrace_syscall) 1463 { 1464 int ret = -1; 1465 struct sigaction action = { 1466 .sa_handler = tracer_stop, 1467 }; 1468 1469 /* Allow external shutdown. */ 1470 tracer_running = true; 1471 ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL)); 1472 1473 errno = 0; 1474 while (ret == -1 && errno != EINVAL) 1475 ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0); 1476 ASSERT_EQ(0, ret) { 1477 kill(tracee, SIGKILL); 1478 } 1479 /* Wait for attach stop */ 1480 wait(NULL); 1481 1482 ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, ptrace_syscall ? 1483 PTRACE_O_TRACESYSGOOD : 1484 PTRACE_O_TRACESECCOMP); 1485 ASSERT_EQ(0, ret) { 1486 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP"); 1487 kill(tracee, SIGKILL); 1488 } 1489 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT, 1490 tracee, NULL, 0); 1491 ASSERT_EQ(0, ret); 1492 1493 /* Unblock the tracee */ 1494 ASSERT_EQ(1, write(fd, "A", 1)); 1495 ASSERT_EQ(0, close(fd)); 1496 1497 /* Run until we're shut down. Must assert to stop execution. */ 1498 while (tracer_running) { 1499 int status; 1500 1501 if (wait(&status) != tracee) 1502 continue; 1503 if (WIFSIGNALED(status) || WIFEXITED(status)) 1504 /* Child is dead. Time to go. */ 1505 return; 1506 1507 /* Check if this is a seccomp event. */ 1508 ASSERT_EQ(!ptrace_syscall, IS_SECCOMP_EVENT(status)); 1509 1510 tracer_func(_metadata, tracee, status, args); 1511 1512 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT, 1513 tracee, NULL, 0); 1514 ASSERT_EQ(0, ret); 1515 } 1516 /* Directly report the status of our test harness results. */ 1517 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS : EXIT_FAILURE); 1518 } 1519 1520 /* Common tracer setup/teardown functions. */ 1521 void cont_handler(int num) 1522 { } 1523 pid_t setup_trace_fixture(struct __test_metadata *_metadata, 1524 tracer_func_t func, void *args, bool ptrace_syscall) 1525 { 1526 char sync; 1527 int pipefd[2]; 1528 pid_t tracer_pid; 1529 pid_t tracee = getpid(); 1530 1531 /* Setup a pipe for clean synchronization. */ 1532 ASSERT_EQ(0, pipe(pipefd)); 1533 1534 /* Fork a child which we'll promote to tracer */ 1535 tracer_pid = fork(); 1536 ASSERT_LE(0, tracer_pid); 1537 signal(SIGALRM, cont_handler); 1538 if (tracer_pid == 0) { 1539 close(pipefd[0]); 1540 start_tracer(_metadata, pipefd[1], tracee, func, args, 1541 ptrace_syscall); 1542 syscall(__NR_exit, 0); 1543 } 1544 close(pipefd[1]); 1545 prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0); 1546 read(pipefd[0], &sync, 1); 1547 close(pipefd[0]); 1548 1549 return tracer_pid; 1550 } 1551 1552 void teardown_trace_fixture(struct __test_metadata *_metadata, 1553 pid_t tracer) 1554 { 1555 if (tracer) { 1556 int status; 1557 /* 1558 * Extract the exit code from the other process and 1559 * adopt it for ourselves in case its asserts failed. 1560 */ 1561 ASSERT_EQ(0, kill(tracer, SIGUSR1)); 1562 ASSERT_EQ(tracer, waitpid(tracer, &status, 0)); 1563 if (WEXITSTATUS(status)) 1564 _metadata->passed = 0; 1565 } 1566 } 1567 1568 /* "poke" tracer arguments and function. */ 1569 struct tracer_args_poke_t { 1570 unsigned long poke_addr; 1571 }; 1572 1573 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status, 1574 void *args) 1575 { 1576 int ret; 1577 unsigned long msg; 1578 struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args; 1579 1580 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 1581 EXPECT_EQ(0, ret); 1582 /* If this fails, don't try to recover. */ 1583 ASSERT_EQ(0x1001, msg) { 1584 kill(tracee, SIGKILL); 1585 } 1586 /* 1587 * Poke in the message. 1588 * Registers are not touched to try to keep this relatively arch 1589 * agnostic. 1590 */ 1591 ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001); 1592 EXPECT_EQ(0, ret); 1593 } 1594 1595 FIXTURE(TRACE_poke) { 1596 struct sock_fprog prog; 1597 pid_t tracer; 1598 long poked; 1599 struct tracer_args_poke_t tracer_args; 1600 }; 1601 1602 FIXTURE_SETUP(TRACE_poke) 1603 { 1604 struct sock_filter filter[] = { 1605 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1606 offsetof(struct seccomp_data, nr)), 1607 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 1608 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001), 1609 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1610 }; 1611 1612 self->poked = 0; 1613 memset(&self->prog, 0, sizeof(self->prog)); 1614 self->prog.filter = malloc(sizeof(filter)); 1615 ASSERT_NE(NULL, self->prog.filter); 1616 memcpy(self->prog.filter, filter, sizeof(filter)); 1617 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 1618 1619 /* Set up tracer args. */ 1620 self->tracer_args.poke_addr = (unsigned long)&self->poked; 1621 1622 /* Launch tracer. */ 1623 self->tracer = setup_trace_fixture(_metadata, tracer_poke, 1624 &self->tracer_args, false); 1625 } 1626 1627 FIXTURE_TEARDOWN(TRACE_poke) 1628 { 1629 teardown_trace_fixture(_metadata, self->tracer); 1630 if (self->prog.filter) 1631 free(self->prog.filter); 1632 } 1633 1634 TEST_F(TRACE_poke, read_has_side_effects) 1635 { 1636 ssize_t ret; 1637 1638 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1639 ASSERT_EQ(0, ret); 1640 1641 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1642 ASSERT_EQ(0, ret); 1643 1644 EXPECT_EQ(0, self->poked); 1645 ret = read(-1, NULL, 0); 1646 EXPECT_EQ(-1, ret); 1647 EXPECT_EQ(0x1001, self->poked); 1648 } 1649 1650 TEST_F(TRACE_poke, getpid_runs_normally) 1651 { 1652 long ret; 1653 1654 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1655 ASSERT_EQ(0, ret); 1656 1657 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1658 ASSERT_EQ(0, ret); 1659 1660 EXPECT_EQ(0, self->poked); 1661 EXPECT_NE(0, syscall(__NR_getpid)); 1662 EXPECT_EQ(0, self->poked); 1663 } 1664 1665 #if defined(__x86_64__) 1666 # define ARCH_REGS struct user_regs_struct 1667 # define SYSCALL_NUM orig_rax 1668 # define SYSCALL_RET rax 1669 #elif defined(__i386__) 1670 # define ARCH_REGS struct user_regs_struct 1671 # define SYSCALL_NUM orig_eax 1672 # define SYSCALL_RET eax 1673 #elif defined(__arm__) 1674 # define ARCH_REGS struct pt_regs 1675 # define SYSCALL_NUM ARM_r7 1676 # define SYSCALL_RET ARM_r0 1677 #elif defined(__aarch64__) 1678 # define ARCH_REGS struct user_pt_regs 1679 # define SYSCALL_NUM regs[8] 1680 # define SYSCALL_RET regs[0] 1681 #elif defined(__riscv) && __riscv_xlen == 64 1682 # define ARCH_REGS struct user_regs_struct 1683 # define SYSCALL_NUM a7 1684 # define SYSCALL_RET a0 1685 #elif defined(__hppa__) 1686 # define ARCH_REGS struct user_regs_struct 1687 # define SYSCALL_NUM gr[20] 1688 # define SYSCALL_RET gr[28] 1689 #elif defined(__powerpc__) 1690 # define ARCH_REGS struct pt_regs 1691 # define SYSCALL_NUM gpr[0] 1692 # define SYSCALL_RET gpr[3] 1693 #elif defined(__s390__) 1694 # define ARCH_REGS s390_regs 1695 # define SYSCALL_NUM gprs[2] 1696 # define SYSCALL_RET gprs[2] 1697 # define SYSCALL_NUM_RET_SHARE_REG 1698 #elif defined(__mips__) 1699 # define ARCH_REGS struct pt_regs 1700 # define SYSCALL_NUM regs[2] 1701 # define SYSCALL_SYSCALL_NUM regs[4] 1702 # define SYSCALL_RET regs[2] 1703 # define SYSCALL_NUM_RET_SHARE_REG 1704 #elif defined(__xtensa__) 1705 # define ARCH_REGS struct user_pt_regs 1706 # define SYSCALL_NUM syscall 1707 /* 1708 * On xtensa syscall return value is in the register 1709 * a2 of the current window which is not fixed. 1710 */ 1711 #define SYSCALL_RET(reg) a[(reg).windowbase * 4 + 2] 1712 #else 1713 # error "Do not know how to find your architecture's registers and syscalls" 1714 #endif 1715 1716 /* When the syscall return can't be changed, stub out the tests for it. */ 1717 #ifdef SYSCALL_NUM_RET_SHARE_REG 1718 # define EXPECT_SYSCALL_RETURN(val, action) EXPECT_EQ(-1, action) 1719 #else 1720 # define EXPECT_SYSCALL_RETURN(val, action) \ 1721 do { \ 1722 errno = 0; \ 1723 if (val < 0) { \ 1724 EXPECT_EQ(-1, action); \ 1725 EXPECT_EQ(-(val), errno); \ 1726 } else { \ 1727 EXPECT_EQ(val, action); \ 1728 } \ 1729 } while (0) 1730 #endif 1731 1732 /* Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for 1733 * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux). 1734 */ 1735 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__) 1736 #define HAVE_GETREGS 1737 #endif 1738 1739 /* Architecture-specific syscall fetching routine. */ 1740 int get_syscall(struct __test_metadata *_metadata, pid_t tracee) 1741 { 1742 ARCH_REGS regs; 1743 #ifdef HAVE_GETREGS 1744 EXPECT_EQ(0, ptrace(PTRACE_GETREGS, tracee, 0, ®s)) { 1745 TH_LOG("PTRACE_GETREGS failed"); 1746 return -1; 1747 } 1748 #else 1749 struct iovec iov; 1750 1751 iov.iov_base = ®s; 1752 iov.iov_len = sizeof(regs); 1753 EXPECT_EQ(0, ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov)) { 1754 TH_LOG("PTRACE_GETREGSET failed"); 1755 return -1; 1756 } 1757 #endif 1758 1759 #if defined(__mips__) 1760 if (regs.SYSCALL_NUM == __NR_O32_Linux) 1761 return regs.SYSCALL_SYSCALL_NUM; 1762 #endif 1763 return regs.SYSCALL_NUM; 1764 } 1765 1766 /* Architecture-specific syscall changing routine. */ 1767 void change_syscall(struct __test_metadata *_metadata, 1768 pid_t tracee, int syscall, int result) 1769 { 1770 int ret; 1771 ARCH_REGS regs; 1772 #ifdef HAVE_GETREGS 1773 ret = ptrace(PTRACE_GETREGS, tracee, 0, ®s); 1774 #else 1775 struct iovec iov; 1776 iov.iov_base = ®s; 1777 iov.iov_len = sizeof(regs); 1778 ret = ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov); 1779 #endif 1780 EXPECT_EQ(0, ret) {} 1781 1782 #if defined(__x86_64__) || defined(__i386__) || defined(__powerpc__) || \ 1783 defined(__s390__) || defined(__hppa__) || defined(__riscv) || \ 1784 defined(__xtensa__) 1785 { 1786 regs.SYSCALL_NUM = syscall; 1787 } 1788 #elif defined(__mips__) 1789 { 1790 if (regs.SYSCALL_NUM == __NR_O32_Linux) 1791 regs.SYSCALL_SYSCALL_NUM = syscall; 1792 else 1793 regs.SYSCALL_NUM = syscall; 1794 } 1795 1796 #elif defined(__arm__) 1797 # ifndef PTRACE_SET_SYSCALL 1798 # define PTRACE_SET_SYSCALL 23 1799 # endif 1800 { 1801 ret = ptrace(PTRACE_SET_SYSCALL, tracee, NULL, syscall); 1802 EXPECT_EQ(0, ret); 1803 } 1804 1805 #elif defined(__aarch64__) 1806 # ifndef NT_ARM_SYSTEM_CALL 1807 # define NT_ARM_SYSTEM_CALL 0x404 1808 # endif 1809 { 1810 iov.iov_base = &syscall; 1811 iov.iov_len = sizeof(syscall); 1812 ret = ptrace(PTRACE_SETREGSET, tracee, NT_ARM_SYSTEM_CALL, 1813 &iov); 1814 EXPECT_EQ(0, ret); 1815 } 1816 1817 #else 1818 ASSERT_EQ(1, 0) { 1819 TH_LOG("How is the syscall changed on this architecture?"); 1820 } 1821 #endif 1822 1823 /* If syscall is skipped, change return value. */ 1824 if (syscall == -1) 1825 #ifdef SYSCALL_NUM_RET_SHARE_REG 1826 TH_LOG("Can't modify syscall return on this architecture"); 1827 1828 #elif defined(__xtensa__) 1829 regs.SYSCALL_RET(regs) = result; 1830 #else 1831 regs.SYSCALL_RET = result; 1832 #endif 1833 1834 #ifdef HAVE_GETREGS 1835 ret = ptrace(PTRACE_SETREGS, tracee, 0, ®s); 1836 #else 1837 iov.iov_base = ®s; 1838 iov.iov_len = sizeof(regs); 1839 ret = ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &iov); 1840 #endif 1841 EXPECT_EQ(0, ret); 1842 } 1843 1844 void tracer_seccomp(struct __test_metadata *_metadata, pid_t tracee, 1845 int status, void *args) 1846 { 1847 int ret; 1848 unsigned long msg; 1849 1850 /* Make sure we got the right message. */ 1851 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 1852 EXPECT_EQ(0, ret); 1853 1854 /* Validate and take action on expected syscalls. */ 1855 switch (msg) { 1856 case 0x1002: 1857 /* change getpid to getppid. */ 1858 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee)); 1859 change_syscall(_metadata, tracee, __NR_getppid, 0); 1860 break; 1861 case 0x1003: 1862 /* skip gettid with valid return code. */ 1863 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee)); 1864 change_syscall(_metadata, tracee, -1, 45000); 1865 break; 1866 case 0x1004: 1867 /* skip openat with error. */ 1868 EXPECT_EQ(__NR_openat, get_syscall(_metadata, tracee)); 1869 change_syscall(_metadata, tracee, -1, -ESRCH); 1870 break; 1871 case 0x1005: 1872 /* do nothing (allow getppid) */ 1873 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee)); 1874 break; 1875 default: 1876 EXPECT_EQ(0, msg) { 1877 TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg); 1878 kill(tracee, SIGKILL); 1879 } 1880 } 1881 1882 } 1883 1884 void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee, 1885 int status, void *args) 1886 { 1887 int ret, nr; 1888 unsigned long msg; 1889 static bool entry; 1890 1891 /* 1892 * The traditional way to tell PTRACE_SYSCALL entry/exit 1893 * is by counting. 1894 */ 1895 entry = !entry; 1896 1897 /* Make sure we got an appropriate message. */ 1898 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 1899 EXPECT_EQ(0, ret); 1900 EXPECT_EQ(entry ? PTRACE_EVENTMSG_SYSCALL_ENTRY 1901 : PTRACE_EVENTMSG_SYSCALL_EXIT, msg); 1902 1903 if (!entry) 1904 return; 1905 1906 nr = get_syscall(_metadata, tracee); 1907 1908 if (nr == __NR_getpid) 1909 change_syscall(_metadata, tracee, __NR_getppid, 0); 1910 if (nr == __NR_gettid) 1911 change_syscall(_metadata, tracee, -1, 45000); 1912 if (nr == __NR_openat) 1913 change_syscall(_metadata, tracee, -1, -ESRCH); 1914 } 1915 1916 FIXTURE(TRACE_syscall) { 1917 struct sock_fprog prog; 1918 pid_t tracer, mytid, mypid, parent; 1919 }; 1920 1921 FIXTURE_VARIANT(TRACE_syscall) { 1922 /* 1923 * All of the SECCOMP_RET_TRACE behaviors can be tested with either 1924 * SECCOMP_RET_TRACE+PTRACE_CONT or plain ptrace()+PTRACE_SYSCALL. 1925 * This indicates if we should use SECCOMP_RET_TRACE (false), or 1926 * ptrace (true). 1927 */ 1928 bool use_ptrace; 1929 }; 1930 1931 FIXTURE_VARIANT_ADD(TRACE_syscall, ptrace) { 1932 .use_ptrace = true, 1933 }; 1934 1935 FIXTURE_VARIANT_ADD(TRACE_syscall, seccomp) { 1936 .use_ptrace = false, 1937 }; 1938 1939 FIXTURE_SETUP(TRACE_syscall) 1940 { 1941 struct sock_filter filter[] = { 1942 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1943 offsetof(struct seccomp_data, nr)), 1944 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 1945 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002), 1946 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1), 1947 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003), 1948 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_openat, 0, 1), 1949 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004), 1950 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 1951 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1005), 1952 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1953 }; 1954 struct sock_fprog prog = { 1955 .len = (unsigned short)ARRAY_SIZE(filter), 1956 .filter = filter, 1957 }; 1958 long ret; 1959 1960 /* Prepare some testable syscall results. */ 1961 self->mytid = syscall(__NR_gettid); 1962 ASSERT_GT(self->mytid, 0); 1963 ASSERT_NE(self->mytid, 1) { 1964 TH_LOG("Running this test as init is not supported. :)"); 1965 } 1966 1967 self->mypid = getpid(); 1968 ASSERT_GT(self->mypid, 0); 1969 ASSERT_EQ(self->mytid, self->mypid); 1970 1971 self->parent = getppid(); 1972 ASSERT_GT(self->parent, 0); 1973 ASSERT_NE(self->parent, self->mypid); 1974 1975 /* Launch tracer. */ 1976 self->tracer = setup_trace_fixture(_metadata, 1977 variant->use_ptrace ? tracer_ptrace 1978 : tracer_seccomp, 1979 NULL, variant->use_ptrace); 1980 1981 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1982 ASSERT_EQ(0, ret); 1983 1984 if (variant->use_ptrace) 1985 return; 1986 1987 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 1988 ASSERT_EQ(0, ret); 1989 } 1990 1991 FIXTURE_TEARDOWN(TRACE_syscall) 1992 { 1993 teardown_trace_fixture(_metadata, self->tracer); 1994 } 1995 1996 TEST(negative_ENOSYS) 1997 { 1998 /* 1999 * There should be no difference between an "internal" skip 2000 * and userspace asking for syscall "-1". 2001 */ 2002 errno = 0; 2003 EXPECT_EQ(-1, syscall(-1)); 2004 EXPECT_EQ(errno, ENOSYS); 2005 /* And no difference for "still not valid but not -1". */ 2006 errno = 0; 2007 EXPECT_EQ(-1, syscall(-101)); 2008 EXPECT_EQ(errno, ENOSYS); 2009 } 2010 2011 TEST_F(TRACE_syscall, negative_ENOSYS) 2012 { 2013 negative_ENOSYS(_metadata); 2014 } 2015 2016 TEST_F(TRACE_syscall, syscall_allowed) 2017 { 2018 /* getppid works as expected (no changes). */ 2019 EXPECT_EQ(self->parent, syscall(__NR_getppid)); 2020 EXPECT_NE(self->mypid, syscall(__NR_getppid)); 2021 } 2022 2023 TEST_F(TRACE_syscall, syscall_redirected) 2024 { 2025 /* getpid has been redirected to getppid as expected. */ 2026 EXPECT_EQ(self->parent, syscall(__NR_getpid)); 2027 EXPECT_NE(self->mypid, syscall(__NR_getpid)); 2028 } 2029 2030 TEST_F(TRACE_syscall, syscall_errno) 2031 { 2032 /* Tracer should skip the open syscall, resulting in ESRCH. */ 2033 EXPECT_SYSCALL_RETURN(-ESRCH, syscall(__NR_openat)); 2034 } 2035 2036 TEST_F(TRACE_syscall, syscall_faked) 2037 { 2038 /* Tracer skips the gettid syscall and store altered return value. */ 2039 EXPECT_SYSCALL_RETURN(45000, syscall(__NR_gettid)); 2040 } 2041 2042 TEST_F(TRACE_syscall, skip_after) 2043 { 2044 struct sock_filter filter[] = { 2045 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2046 offsetof(struct seccomp_data, nr)), 2047 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 2048 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM), 2049 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2050 }; 2051 struct sock_fprog prog = { 2052 .len = (unsigned short)ARRAY_SIZE(filter), 2053 .filter = filter, 2054 }; 2055 long ret; 2056 2057 /* Install additional "errno on getppid" filter. */ 2058 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2059 ASSERT_EQ(0, ret); 2060 2061 /* Tracer will redirect getpid to getppid, and we should see EPERM. */ 2062 errno = 0; 2063 EXPECT_EQ(-1, syscall(__NR_getpid)); 2064 EXPECT_EQ(EPERM, errno); 2065 } 2066 2067 TEST_F_SIGNAL(TRACE_syscall, kill_after, SIGSYS) 2068 { 2069 struct sock_filter filter[] = { 2070 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2071 offsetof(struct seccomp_data, nr)), 2072 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 2073 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2074 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2075 }; 2076 struct sock_fprog prog = { 2077 .len = (unsigned short)ARRAY_SIZE(filter), 2078 .filter = filter, 2079 }; 2080 long ret; 2081 2082 /* Install additional "death on getppid" filter. */ 2083 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2084 ASSERT_EQ(0, ret); 2085 2086 /* Tracer will redirect getpid to getppid, and we should die. */ 2087 EXPECT_NE(self->mypid, syscall(__NR_getpid)); 2088 } 2089 2090 TEST(seccomp_syscall) 2091 { 2092 struct sock_filter filter[] = { 2093 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2094 }; 2095 struct sock_fprog prog = { 2096 .len = (unsigned short)ARRAY_SIZE(filter), 2097 .filter = filter, 2098 }; 2099 long ret; 2100 2101 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 2102 ASSERT_EQ(0, ret) { 2103 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2104 } 2105 2106 /* Reject insane operation. */ 2107 ret = seccomp(-1, 0, &prog); 2108 ASSERT_NE(ENOSYS, errno) { 2109 TH_LOG("Kernel does not support seccomp syscall!"); 2110 } 2111 EXPECT_EQ(EINVAL, errno) { 2112 TH_LOG("Did not reject crazy op value!"); 2113 } 2114 2115 /* Reject strict with flags or pointer. */ 2116 ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL); 2117 EXPECT_EQ(EINVAL, errno) { 2118 TH_LOG("Did not reject mode strict with flags!"); 2119 } 2120 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog); 2121 EXPECT_EQ(EINVAL, errno) { 2122 TH_LOG("Did not reject mode strict with uargs!"); 2123 } 2124 2125 /* Reject insane args for filter. */ 2126 ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog); 2127 EXPECT_EQ(EINVAL, errno) { 2128 TH_LOG("Did not reject crazy filter flags!"); 2129 } 2130 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL); 2131 EXPECT_EQ(EFAULT, errno) { 2132 TH_LOG("Did not reject NULL filter!"); 2133 } 2134 2135 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2136 EXPECT_EQ(0, errno) { 2137 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s", 2138 strerror(errno)); 2139 } 2140 } 2141 2142 TEST(seccomp_syscall_mode_lock) 2143 { 2144 struct sock_filter filter[] = { 2145 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2146 }; 2147 struct sock_fprog prog = { 2148 .len = (unsigned short)ARRAY_SIZE(filter), 2149 .filter = filter, 2150 }; 2151 long ret; 2152 2153 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 2154 ASSERT_EQ(0, ret) { 2155 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2156 } 2157 2158 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2159 ASSERT_NE(ENOSYS, errno) { 2160 TH_LOG("Kernel does not support seccomp syscall!"); 2161 } 2162 EXPECT_EQ(0, ret) { 2163 TH_LOG("Could not install filter!"); 2164 } 2165 2166 /* Make sure neither entry point will switch to strict. */ 2167 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0); 2168 EXPECT_EQ(EINVAL, errno) { 2169 TH_LOG("Switched to mode strict!"); 2170 } 2171 2172 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL); 2173 EXPECT_EQ(EINVAL, errno) { 2174 TH_LOG("Switched to mode strict!"); 2175 } 2176 } 2177 2178 /* 2179 * Test detection of known and unknown filter flags. Userspace needs to be able 2180 * to check if a filter flag is supported by the current kernel and a good way 2181 * of doing that is by attempting to enter filter mode, with the flag bit in 2182 * question set, and a NULL pointer for the _args_ parameter. EFAULT indicates 2183 * that the flag is valid and EINVAL indicates that the flag is invalid. 2184 */ 2185 TEST(detect_seccomp_filter_flags) 2186 { 2187 unsigned int flags[] = { SECCOMP_FILTER_FLAG_TSYNC, 2188 SECCOMP_FILTER_FLAG_LOG, 2189 SECCOMP_FILTER_FLAG_SPEC_ALLOW, 2190 SECCOMP_FILTER_FLAG_NEW_LISTENER, 2191 SECCOMP_FILTER_FLAG_TSYNC_ESRCH }; 2192 unsigned int exclusive[] = { 2193 SECCOMP_FILTER_FLAG_TSYNC, 2194 SECCOMP_FILTER_FLAG_NEW_LISTENER }; 2195 unsigned int flag, all_flags, exclusive_mask; 2196 int i; 2197 long ret; 2198 2199 /* Test detection of individual known-good filter flags */ 2200 for (i = 0, all_flags = 0; i < ARRAY_SIZE(flags); i++) { 2201 int bits = 0; 2202 2203 flag = flags[i]; 2204 /* Make sure the flag is a single bit! */ 2205 while (flag) { 2206 if (flag & 0x1) 2207 bits ++; 2208 flag >>= 1; 2209 } 2210 ASSERT_EQ(1, bits); 2211 flag = flags[i]; 2212 2213 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2214 ASSERT_NE(ENOSYS, errno) { 2215 TH_LOG("Kernel does not support seccomp syscall!"); 2216 } 2217 EXPECT_EQ(-1, ret); 2218 EXPECT_EQ(EFAULT, errno) { 2219 TH_LOG("Failed to detect that a known-good filter flag (0x%X) is supported!", 2220 flag); 2221 } 2222 2223 all_flags |= flag; 2224 } 2225 2226 /* 2227 * Test detection of all known-good filter flags combined. But 2228 * for the exclusive flags we need to mask them out and try them 2229 * individually for the "all flags" testing. 2230 */ 2231 exclusive_mask = 0; 2232 for (i = 0; i < ARRAY_SIZE(exclusive); i++) 2233 exclusive_mask |= exclusive[i]; 2234 for (i = 0; i < ARRAY_SIZE(exclusive); i++) { 2235 flag = all_flags & ~exclusive_mask; 2236 flag |= exclusive[i]; 2237 2238 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2239 EXPECT_EQ(-1, ret); 2240 EXPECT_EQ(EFAULT, errno) { 2241 TH_LOG("Failed to detect that all known-good filter flags (0x%X) are supported!", 2242 flag); 2243 } 2244 } 2245 2246 /* Test detection of an unknown filter flags, without exclusives. */ 2247 flag = -1; 2248 flag &= ~exclusive_mask; 2249 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2250 EXPECT_EQ(-1, ret); 2251 EXPECT_EQ(EINVAL, errno) { 2252 TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported!", 2253 flag); 2254 } 2255 2256 /* 2257 * Test detection of an unknown filter flag that may simply need to be 2258 * added to this test 2259 */ 2260 flag = flags[ARRAY_SIZE(flags) - 1] << 1; 2261 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2262 EXPECT_EQ(-1, ret); 2263 EXPECT_EQ(EINVAL, errno) { 2264 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?", 2265 flag); 2266 } 2267 } 2268 2269 TEST(TSYNC_first) 2270 { 2271 struct sock_filter filter[] = { 2272 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2273 }; 2274 struct sock_fprog prog = { 2275 .len = (unsigned short)ARRAY_SIZE(filter), 2276 .filter = filter, 2277 }; 2278 long ret; 2279 2280 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 2281 ASSERT_EQ(0, ret) { 2282 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2283 } 2284 2285 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2286 &prog); 2287 ASSERT_NE(ENOSYS, errno) { 2288 TH_LOG("Kernel does not support seccomp syscall!"); 2289 } 2290 EXPECT_EQ(0, ret) { 2291 TH_LOG("Could not install initial filter with TSYNC!"); 2292 } 2293 } 2294 2295 #define TSYNC_SIBLINGS 2 2296 struct tsync_sibling { 2297 pthread_t tid; 2298 pid_t system_tid; 2299 sem_t *started; 2300 pthread_cond_t *cond; 2301 pthread_mutex_t *mutex; 2302 int diverge; 2303 int num_waits; 2304 struct sock_fprog *prog; 2305 struct __test_metadata *metadata; 2306 }; 2307 2308 /* 2309 * To avoid joining joined threads (which is not allowed by Bionic), 2310 * make sure we both successfully join and clear the tid to skip a 2311 * later join attempt during fixture teardown. Any remaining threads 2312 * will be directly killed during teardown. 2313 */ 2314 #define PTHREAD_JOIN(tid, status) \ 2315 do { \ 2316 int _rc = pthread_join(tid, status); \ 2317 if (_rc) { \ 2318 TH_LOG("pthread_join of tid %u failed: %d\n", \ 2319 (unsigned int)tid, _rc); \ 2320 } else { \ 2321 tid = 0; \ 2322 } \ 2323 } while (0) 2324 2325 FIXTURE(TSYNC) { 2326 struct sock_fprog root_prog, apply_prog; 2327 struct tsync_sibling sibling[TSYNC_SIBLINGS]; 2328 sem_t started; 2329 pthread_cond_t cond; 2330 pthread_mutex_t mutex; 2331 int sibling_count; 2332 }; 2333 2334 FIXTURE_SETUP(TSYNC) 2335 { 2336 struct sock_filter root_filter[] = { 2337 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2338 }; 2339 struct sock_filter apply_filter[] = { 2340 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2341 offsetof(struct seccomp_data, nr)), 2342 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 2343 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2344 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2345 }; 2346 2347 memset(&self->root_prog, 0, sizeof(self->root_prog)); 2348 memset(&self->apply_prog, 0, sizeof(self->apply_prog)); 2349 memset(&self->sibling, 0, sizeof(self->sibling)); 2350 self->root_prog.filter = malloc(sizeof(root_filter)); 2351 ASSERT_NE(NULL, self->root_prog.filter); 2352 memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter)); 2353 self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter); 2354 2355 self->apply_prog.filter = malloc(sizeof(apply_filter)); 2356 ASSERT_NE(NULL, self->apply_prog.filter); 2357 memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter)); 2358 self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter); 2359 2360 self->sibling_count = 0; 2361 pthread_mutex_init(&self->mutex, NULL); 2362 pthread_cond_init(&self->cond, NULL); 2363 sem_init(&self->started, 0, 0); 2364 self->sibling[0].tid = 0; 2365 self->sibling[0].cond = &self->cond; 2366 self->sibling[0].started = &self->started; 2367 self->sibling[0].mutex = &self->mutex; 2368 self->sibling[0].diverge = 0; 2369 self->sibling[0].num_waits = 1; 2370 self->sibling[0].prog = &self->root_prog; 2371 self->sibling[0].metadata = _metadata; 2372 self->sibling[1].tid = 0; 2373 self->sibling[1].cond = &self->cond; 2374 self->sibling[1].started = &self->started; 2375 self->sibling[1].mutex = &self->mutex; 2376 self->sibling[1].diverge = 0; 2377 self->sibling[1].prog = &self->root_prog; 2378 self->sibling[1].num_waits = 1; 2379 self->sibling[1].metadata = _metadata; 2380 } 2381 2382 FIXTURE_TEARDOWN(TSYNC) 2383 { 2384 int sib = 0; 2385 2386 if (self->root_prog.filter) 2387 free(self->root_prog.filter); 2388 if (self->apply_prog.filter) 2389 free(self->apply_prog.filter); 2390 2391 for ( ; sib < self->sibling_count; ++sib) { 2392 struct tsync_sibling *s = &self->sibling[sib]; 2393 2394 if (!s->tid) 2395 continue; 2396 /* 2397 * If a thread is still running, it may be stuck, so hit 2398 * it over the head really hard. 2399 */ 2400 pthread_kill(s->tid, 9); 2401 } 2402 pthread_mutex_destroy(&self->mutex); 2403 pthread_cond_destroy(&self->cond); 2404 sem_destroy(&self->started); 2405 } 2406 2407 void *tsync_sibling(void *data) 2408 { 2409 long ret = 0; 2410 struct tsync_sibling *me = data; 2411 2412 me->system_tid = syscall(__NR_gettid); 2413 2414 pthread_mutex_lock(me->mutex); 2415 if (me->diverge) { 2416 /* Just re-apply the root prog to fork the tree */ 2417 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, 2418 me->prog, 0, 0); 2419 } 2420 sem_post(me->started); 2421 /* Return outside of started so parent notices failures. */ 2422 if (ret) { 2423 pthread_mutex_unlock(me->mutex); 2424 return (void *)SIBLING_EXIT_FAILURE; 2425 } 2426 do { 2427 pthread_cond_wait(me->cond, me->mutex); 2428 me->num_waits = me->num_waits - 1; 2429 } while (me->num_waits); 2430 pthread_mutex_unlock(me->mutex); 2431 2432 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0); 2433 if (!ret) 2434 return (void *)SIBLING_EXIT_NEWPRIVS; 2435 read(0, NULL, 0); 2436 return (void *)SIBLING_EXIT_UNKILLED; 2437 } 2438 2439 void tsync_start_sibling(struct tsync_sibling *sibling) 2440 { 2441 pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling); 2442 } 2443 2444 TEST_F(TSYNC, siblings_fail_prctl) 2445 { 2446 long ret; 2447 void *status; 2448 struct sock_filter filter[] = { 2449 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2450 offsetof(struct seccomp_data, nr)), 2451 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1), 2452 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL), 2453 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2454 }; 2455 struct sock_fprog prog = { 2456 .len = (unsigned short)ARRAY_SIZE(filter), 2457 .filter = filter, 2458 }; 2459 2460 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2461 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2462 } 2463 2464 /* Check prctl failure detection by requesting sib 0 diverge. */ 2465 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2466 ASSERT_NE(ENOSYS, errno) { 2467 TH_LOG("Kernel does not support seccomp syscall!"); 2468 } 2469 ASSERT_EQ(0, ret) { 2470 TH_LOG("setting filter failed"); 2471 } 2472 2473 self->sibling[0].diverge = 1; 2474 tsync_start_sibling(&self->sibling[0]); 2475 tsync_start_sibling(&self->sibling[1]); 2476 2477 while (self->sibling_count < TSYNC_SIBLINGS) { 2478 sem_wait(&self->started); 2479 self->sibling_count++; 2480 } 2481 2482 /* Signal the threads to clean up*/ 2483 pthread_mutex_lock(&self->mutex); 2484 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2485 TH_LOG("cond broadcast non-zero"); 2486 } 2487 pthread_mutex_unlock(&self->mutex); 2488 2489 /* Ensure diverging sibling failed to call prctl. */ 2490 PTHREAD_JOIN(self->sibling[0].tid, &status); 2491 EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status); 2492 PTHREAD_JOIN(self->sibling[1].tid, &status); 2493 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2494 } 2495 2496 TEST_F(TSYNC, two_siblings_with_ancestor) 2497 { 2498 long ret; 2499 void *status; 2500 2501 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2502 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2503 } 2504 2505 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2506 ASSERT_NE(ENOSYS, errno) { 2507 TH_LOG("Kernel does not support seccomp syscall!"); 2508 } 2509 ASSERT_EQ(0, ret) { 2510 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2511 } 2512 tsync_start_sibling(&self->sibling[0]); 2513 tsync_start_sibling(&self->sibling[1]); 2514 2515 while (self->sibling_count < TSYNC_SIBLINGS) { 2516 sem_wait(&self->started); 2517 self->sibling_count++; 2518 } 2519 2520 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2521 &self->apply_prog); 2522 ASSERT_EQ(0, ret) { 2523 TH_LOG("Could install filter on all threads!"); 2524 } 2525 /* Tell the siblings to test the policy */ 2526 pthread_mutex_lock(&self->mutex); 2527 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2528 TH_LOG("cond broadcast non-zero"); 2529 } 2530 pthread_mutex_unlock(&self->mutex); 2531 /* Ensure they are both killed and don't exit cleanly. */ 2532 PTHREAD_JOIN(self->sibling[0].tid, &status); 2533 EXPECT_EQ(0x0, (long)status); 2534 PTHREAD_JOIN(self->sibling[1].tid, &status); 2535 EXPECT_EQ(0x0, (long)status); 2536 } 2537 2538 TEST_F(TSYNC, two_sibling_want_nnp) 2539 { 2540 void *status; 2541 2542 /* start siblings before any prctl() operations */ 2543 tsync_start_sibling(&self->sibling[0]); 2544 tsync_start_sibling(&self->sibling[1]); 2545 while (self->sibling_count < TSYNC_SIBLINGS) { 2546 sem_wait(&self->started); 2547 self->sibling_count++; 2548 } 2549 2550 /* Tell the siblings to test no policy */ 2551 pthread_mutex_lock(&self->mutex); 2552 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2553 TH_LOG("cond broadcast non-zero"); 2554 } 2555 pthread_mutex_unlock(&self->mutex); 2556 2557 /* Ensure they are both upset about lacking nnp. */ 2558 PTHREAD_JOIN(self->sibling[0].tid, &status); 2559 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status); 2560 PTHREAD_JOIN(self->sibling[1].tid, &status); 2561 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status); 2562 } 2563 2564 TEST_F(TSYNC, two_siblings_with_no_filter) 2565 { 2566 long ret; 2567 void *status; 2568 2569 /* start siblings before any prctl() operations */ 2570 tsync_start_sibling(&self->sibling[0]); 2571 tsync_start_sibling(&self->sibling[1]); 2572 while (self->sibling_count < TSYNC_SIBLINGS) { 2573 sem_wait(&self->started); 2574 self->sibling_count++; 2575 } 2576 2577 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2578 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2579 } 2580 2581 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2582 &self->apply_prog); 2583 ASSERT_NE(ENOSYS, errno) { 2584 TH_LOG("Kernel does not support seccomp syscall!"); 2585 } 2586 ASSERT_EQ(0, ret) { 2587 TH_LOG("Could install filter on all threads!"); 2588 } 2589 2590 /* Tell the siblings to test the policy */ 2591 pthread_mutex_lock(&self->mutex); 2592 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2593 TH_LOG("cond broadcast non-zero"); 2594 } 2595 pthread_mutex_unlock(&self->mutex); 2596 2597 /* Ensure they are both killed and don't exit cleanly. */ 2598 PTHREAD_JOIN(self->sibling[0].tid, &status); 2599 EXPECT_EQ(0x0, (long)status); 2600 PTHREAD_JOIN(self->sibling[1].tid, &status); 2601 EXPECT_EQ(0x0, (long)status); 2602 } 2603 2604 TEST_F(TSYNC, two_siblings_with_one_divergence) 2605 { 2606 long ret; 2607 void *status; 2608 2609 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2610 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2611 } 2612 2613 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2614 ASSERT_NE(ENOSYS, errno) { 2615 TH_LOG("Kernel does not support seccomp syscall!"); 2616 } 2617 ASSERT_EQ(0, ret) { 2618 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2619 } 2620 self->sibling[0].diverge = 1; 2621 tsync_start_sibling(&self->sibling[0]); 2622 tsync_start_sibling(&self->sibling[1]); 2623 2624 while (self->sibling_count < TSYNC_SIBLINGS) { 2625 sem_wait(&self->started); 2626 self->sibling_count++; 2627 } 2628 2629 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2630 &self->apply_prog); 2631 ASSERT_EQ(self->sibling[0].system_tid, ret) { 2632 TH_LOG("Did not fail on diverged sibling."); 2633 } 2634 2635 /* Wake the threads */ 2636 pthread_mutex_lock(&self->mutex); 2637 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2638 TH_LOG("cond broadcast non-zero"); 2639 } 2640 pthread_mutex_unlock(&self->mutex); 2641 2642 /* Ensure they are both unkilled. */ 2643 PTHREAD_JOIN(self->sibling[0].tid, &status); 2644 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2645 PTHREAD_JOIN(self->sibling[1].tid, &status); 2646 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2647 } 2648 2649 TEST_F(TSYNC, two_siblings_with_one_divergence_no_tid_in_err) 2650 { 2651 long ret, flags; 2652 void *status; 2653 2654 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2655 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2656 } 2657 2658 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2659 ASSERT_NE(ENOSYS, errno) { 2660 TH_LOG("Kernel does not support seccomp syscall!"); 2661 } 2662 ASSERT_EQ(0, ret) { 2663 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2664 } 2665 self->sibling[0].diverge = 1; 2666 tsync_start_sibling(&self->sibling[0]); 2667 tsync_start_sibling(&self->sibling[1]); 2668 2669 while (self->sibling_count < TSYNC_SIBLINGS) { 2670 sem_wait(&self->started); 2671 self->sibling_count++; 2672 } 2673 2674 flags = SECCOMP_FILTER_FLAG_TSYNC | \ 2675 SECCOMP_FILTER_FLAG_TSYNC_ESRCH; 2676 ret = seccomp(SECCOMP_SET_MODE_FILTER, flags, &self->apply_prog); 2677 ASSERT_EQ(ESRCH, errno) { 2678 TH_LOG("Did not return ESRCH for diverged sibling."); 2679 } 2680 ASSERT_EQ(-1, ret) { 2681 TH_LOG("Did not fail on diverged sibling."); 2682 } 2683 2684 /* Wake the threads */ 2685 pthread_mutex_lock(&self->mutex); 2686 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2687 TH_LOG("cond broadcast non-zero"); 2688 } 2689 pthread_mutex_unlock(&self->mutex); 2690 2691 /* Ensure they are both unkilled. */ 2692 PTHREAD_JOIN(self->sibling[0].tid, &status); 2693 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2694 PTHREAD_JOIN(self->sibling[1].tid, &status); 2695 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2696 } 2697 2698 TEST_F(TSYNC, two_siblings_not_under_filter) 2699 { 2700 long ret, sib; 2701 void *status; 2702 struct timespec delay = { .tv_nsec = 100000000 }; 2703 2704 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2705 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2706 } 2707 2708 /* 2709 * Sibling 0 will have its own seccomp policy 2710 * and Sibling 1 will not be under seccomp at 2711 * all. Sibling 1 will enter seccomp and 0 2712 * will cause failure. 2713 */ 2714 self->sibling[0].diverge = 1; 2715 tsync_start_sibling(&self->sibling[0]); 2716 tsync_start_sibling(&self->sibling[1]); 2717 2718 while (self->sibling_count < TSYNC_SIBLINGS) { 2719 sem_wait(&self->started); 2720 self->sibling_count++; 2721 } 2722 2723 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2724 ASSERT_NE(ENOSYS, errno) { 2725 TH_LOG("Kernel does not support seccomp syscall!"); 2726 } 2727 ASSERT_EQ(0, ret) { 2728 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2729 } 2730 2731 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2732 &self->apply_prog); 2733 ASSERT_EQ(ret, self->sibling[0].system_tid) { 2734 TH_LOG("Did not fail on diverged sibling."); 2735 } 2736 sib = 1; 2737 if (ret == self->sibling[0].system_tid) 2738 sib = 0; 2739 2740 pthread_mutex_lock(&self->mutex); 2741 2742 /* Increment the other siblings num_waits so we can clean up 2743 * the one we just saw. 2744 */ 2745 self->sibling[!sib].num_waits += 1; 2746 2747 /* Signal the thread to clean up*/ 2748 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2749 TH_LOG("cond broadcast non-zero"); 2750 } 2751 pthread_mutex_unlock(&self->mutex); 2752 PTHREAD_JOIN(self->sibling[sib].tid, &status); 2753 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2754 /* Poll for actual task death. pthread_join doesn't guarantee it. */ 2755 while (!kill(self->sibling[sib].system_tid, 0)) 2756 nanosleep(&delay, NULL); 2757 /* Switch to the remaining sibling */ 2758 sib = !sib; 2759 2760 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2761 &self->apply_prog); 2762 ASSERT_EQ(0, ret) { 2763 TH_LOG("Expected the remaining sibling to sync"); 2764 }; 2765 2766 pthread_mutex_lock(&self->mutex); 2767 2768 /* If remaining sibling didn't have a chance to wake up during 2769 * the first broadcast, manually reduce the num_waits now. 2770 */ 2771 if (self->sibling[sib].num_waits > 1) 2772 self->sibling[sib].num_waits = 1; 2773 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2774 TH_LOG("cond broadcast non-zero"); 2775 } 2776 pthread_mutex_unlock(&self->mutex); 2777 PTHREAD_JOIN(self->sibling[sib].tid, &status); 2778 EXPECT_EQ(0, (long)status); 2779 /* Poll for actual task death. pthread_join doesn't guarantee it. */ 2780 while (!kill(self->sibling[sib].system_tid, 0)) 2781 nanosleep(&delay, NULL); 2782 2783 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2784 &self->apply_prog); 2785 ASSERT_EQ(0, ret); /* just us chickens */ 2786 } 2787 2788 /* Make sure restarted syscalls are seen directly as "restart_syscall". */ 2789 TEST(syscall_restart) 2790 { 2791 long ret; 2792 unsigned long msg; 2793 pid_t child_pid; 2794 int pipefd[2]; 2795 int status; 2796 siginfo_t info = { }; 2797 struct sock_filter filter[] = { 2798 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2799 offsetof(struct seccomp_data, nr)), 2800 2801 #ifdef __NR_sigreturn 2802 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 7, 0), 2803 #endif 2804 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 6, 0), 2805 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 5, 0), 2806 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 4, 0), 2807 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 5, 0), 2808 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_clock_nanosleep, 4, 0), 2809 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0), 2810 2811 /* Allow __NR_write for easy logging. */ 2812 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1), 2813 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2814 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2815 /* The nanosleep jump target. */ 2816 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100), 2817 /* The restart_syscall jump target. */ 2818 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200), 2819 }; 2820 struct sock_fprog prog = { 2821 .len = (unsigned short)ARRAY_SIZE(filter), 2822 .filter = filter, 2823 }; 2824 #if defined(__arm__) 2825 struct utsname utsbuf; 2826 #endif 2827 2828 ASSERT_EQ(0, pipe(pipefd)); 2829 2830 child_pid = fork(); 2831 ASSERT_LE(0, child_pid); 2832 if (child_pid == 0) { 2833 /* Child uses EXPECT not ASSERT to deliver status correctly. */ 2834 char buf = ' '; 2835 struct timespec timeout = { }; 2836 2837 /* Attach parent as tracer and stop. */ 2838 EXPECT_EQ(0, ptrace(PTRACE_TRACEME)); 2839 EXPECT_EQ(0, raise(SIGSTOP)); 2840 2841 EXPECT_EQ(0, close(pipefd[1])); 2842 2843 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2844 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2845 } 2846 2847 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2848 EXPECT_EQ(0, ret) { 2849 TH_LOG("Failed to install filter!"); 2850 } 2851 2852 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) { 2853 TH_LOG("Failed to read() sync from parent"); 2854 } 2855 EXPECT_EQ('.', buf) { 2856 TH_LOG("Failed to get sync data from read()"); 2857 } 2858 2859 /* Start nanosleep to be interrupted. */ 2860 timeout.tv_sec = 1; 2861 errno = 0; 2862 EXPECT_EQ(0, nanosleep(&timeout, NULL)) { 2863 TH_LOG("Call to nanosleep() failed (errno %d)", errno); 2864 } 2865 2866 /* Read final sync from parent. */ 2867 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) { 2868 TH_LOG("Failed final read() from parent"); 2869 } 2870 EXPECT_EQ('!', buf) { 2871 TH_LOG("Failed to get final data from read()"); 2872 } 2873 2874 /* Directly report the status of our test harness results. */ 2875 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS 2876 : EXIT_FAILURE); 2877 } 2878 EXPECT_EQ(0, close(pipefd[0])); 2879 2880 /* Attach to child, setup options, and release. */ 2881 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2882 ASSERT_EQ(true, WIFSTOPPED(status)); 2883 ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL, 2884 PTRACE_O_TRACESECCOMP)); 2885 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 2886 ASSERT_EQ(1, write(pipefd[1], ".", 1)); 2887 2888 /* Wait for nanosleep() to start. */ 2889 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2890 ASSERT_EQ(true, WIFSTOPPED(status)); 2891 ASSERT_EQ(SIGTRAP, WSTOPSIG(status)); 2892 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16)); 2893 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg)); 2894 ASSERT_EQ(0x100, msg); 2895 ret = get_syscall(_metadata, child_pid); 2896 EXPECT_TRUE(ret == __NR_nanosleep || ret == __NR_clock_nanosleep); 2897 2898 /* Might as well check siginfo for sanity while we're here. */ 2899 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info)); 2900 ASSERT_EQ(SIGTRAP, info.si_signo); 2901 ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code); 2902 EXPECT_EQ(0, info.si_errno); 2903 EXPECT_EQ(getuid(), info.si_uid); 2904 /* Verify signal delivery came from child (seccomp-triggered). */ 2905 EXPECT_EQ(child_pid, info.si_pid); 2906 2907 /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */ 2908 ASSERT_EQ(0, kill(child_pid, SIGSTOP)); 2909 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 2910 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2911 ASSERT_EQ(true, WIFSTOPPED(status)); 2912 ASSERT_EQ(SIGSTOP, WSTOPSIG(status)); 2913 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info)); 2914 /* 2915 * There is no siginfo on SIGSTOP any more, so we can't verify 2916 * signal delivery came from parent now (getpid() == info.si_pid). 2917 * https://lkml.kernel.org/r/CAGXu5jJaZAOzP1qFz66tYrtbuywqb+UN2SOA1VLHpCCOiYvYeg@mail.gmail.com 2918 * At least verify the SIGSTOP via PTRACE_GETSIGINFO. 2919 */ 2920 EXPECT_EQ(SIGSTOP, info.si_signo); 2921 2922 /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */ 2923 ASSERT_EQ(0, kill(child_pid, SIGCONT)); 2924 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 2925 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2926 ASSERT_EQ(true, WIFSTOPPED(status)); 2927 ASSERT_EQ(SIGCONT, WSTOPSIG(status)); 2928 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 2929 2930 /* Wait for restart_syscall() to start. */ 2931 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2932 ASSERT_EQ(true, WIFSTOPPED(status)); 2933 ASSERT_EQ(SIGTRAP, WSTOPSIG(status)); 2934 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16)); 2935 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg)); 2936 2937 ASSERT_EQ(0x200, msg); 2938 ret = get_syscall(_metadata, child_pid); 2939 #if defined(__arm__) 2940 /* 2941 * FIXME: 2942 * - native ARM registers do NOT expose true syscall. 2943 * - compat ARM registers on ARM64 DO expose true syscall. 2944 */ 2945 ASSERT_EQ(0, uname(&utsbuf)); 2946 if (strncmp(utsbuf.machine, "arm", 3) == 0) { 2947 EXPECT_EQ(__NR_nanosleep, ret); 2948 } else 2949 #endif 2950 { 2951 EXPECT_EQ(__NR_restart_syscall, ret); 2952 } 2953 2954 /* Write again to end test. */ 2955 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 2956 ASSERT_EQ(1, write(pipefd[1], "!", 1)); 2957 EXPECT_EQ(0, close(pipefd[1])); 2958 2959 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 2960 if (WIFSIGNALED(status) || WEXITSTATUS(status)) 2961 _metadata->passed = 0; 2962 } 2963 2964 TEST_SIGNAL(filter_flag_log, SIGSYS) 2965 { 2966 struct sock_filter allow_filter[] = { 2967 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2968 }; 2969 struct sock_filter kill_filter[] = { 2970 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2971 offsetof(struct seccomp_data, nr)), 2972 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 2973 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2974 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2975 }; 2976 struct sock_fprog allow_prog = { 2977 .len = (unsigned short)ARRAY_SIZE(allow_filter), 2978 .filter = allow_filter, 2979 }; 2980 struct sock_fprog kill_prog = { 2981 .len = (unsigned short)ARRAY_SIZE(kill_filter), 2982 .filter = kill_filter, 2983 }; 2984 long ret; 2985 pid_t parent = getppid(); 2986 2987 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 2988 ASSERT_EQ(0, ret); 2989 2990 /* Verify that the FILTER_FLAG_LOG flag isn't accepted in strict mode */ 2991 ret = seccomp(SECCOMP_SET_MODE_STRICT, SECCOMP_FILTER_FLAG_LOG, 2992 &allow_prog); 2993 ASSERT_NE(ENOSYS, errno) { 2994 TH_LOG("Kernel does not support seccomp syscall!"); 2995 } 2996 EXPECT_NE(0, ret) { 2997 TH_LOG("Kernel accepted FILTER_FLAG_LOG flag in strict mode!"); 2998 } 2999 EXPECT_EQ(EINVAL, errno) { 3000 TH_LOG("Kernel returned unexpected errno for FILTER_FLAG_LOG flag in strict mode!"); 3001 } 3002 3003 /* Verify that a simple, permissive filter can be added with no flags */ 3004 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog); 3005 EXPECT_EQ(0, ret); 3006 3007 /* See if the same filter can be added with the FILTER_FLAG_LOG flag */ 3008 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG, 3009 &allow_prog); 3010 ASSERT_NE(EINVAL, errno) { 3011 TH_LOG("Kernel does not support the FILTER_FLAG_LOG flag!"); 3012 } 3013 EXPECT_EQ(0, ret); 3014 3015 /* Ensure that the kill filter works with the FILTER_FLAG_LOG flag */ 3016 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG, 3017 &kill_prog); 3018 EXPECT_EQ(0, ret); 3019 3020 EXPECT_EQ(parent, syscall(__NR_getppid)); 3021 /* getpid() should never return. */ 3022 EXPECT_EQ(0, syscall(__NR_getpid)); 3023 } 3024 3025 TEST(get_action_avail) 3026 { 3027 __u32 actions[] = { SECCOMP_RET_KILL_THREAD, SECCOMP_RET_TRAP, 3028 SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE, 3029 SECCOMP_RET_LOG, SECCOMP_RET_ALLOW }; 3030 __u32 unknown_action = 0x10000000U; 3031 int i; 3032 long ret; 3033 3034 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[0]); 3035 ASSERT_NE(ENOSYS, errno) { 3036 TH_LOG("Kernel does not support seccomp syscall!"); 3037 } 3038 ASSERT_NE(EINVAL, errno) { 3039 TH_LOG("Kernel does not support SECCOMP_GET_ACTION_AVAIL operation!"); 3040 } 3041 EXPECT_EQ(ret, 0); 3042 3043 for (i = 0; i < ARRAY_SIZE(actions); i++) { 3044 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[i]); 3045 EXPECT_EQ(ret, 0) { 3046 TH_LOG("Expected action (0x%X) not available!", 3047 actions[i]); 3048 } 3049 } 3050 3051 /* Check that an unknown action is handled properly (EOPNOTSUPP) */ 3052 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &unknown_action); 3053 EXPECT_EQ(ret, -1); 3054 EXPECT_EQ(errno, EOPNOTSUPP); 3055 } 3056 3057 TEST(get_metadata) 3058 { 3059 pid_t pid; 3060 int pipefd[2]; 3061 char buf; 3062 struct seccomp_metadata md; 3063 long ret; 3064 3065 /* Only real root can get metadata. */ 3066 if (geteuid()) { 3067 SKIP(return, "get_metadata requires real root"); 3068 return; 3069 } 3070 3071 ASSERT_EQ(0, pipe(pipefd)); 3072 3073 pid = fork(); 3074 ASSERT_GE(pid, 0); 3075 if (pid == 0) { 3076 struct sock_filter filter[] = { 3077 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3078 }; 3079 struct sock_fprog prog = { 3080 .len = (unsigned short)ARRAY_SIZE(filter), 3081 .filter = filter, 3082 }; 3083 3084 /* one with log, one without */ 3085 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 3086 SECCOMP_FILTER_FLAG_LOG, &prog)); 3087 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog)); 3088 3089 EXPECT_EQ(0, close(pipefd[0])); 3090 ASSERT_EQ(1, write(pipefd[1], "1", 1)); 3091 ASSERT_EQ(0, close(pipefd[1])); 3092 3093 while (1) 3094 sleep(100); 3095 } 3096 3097 ASSERT_EQ(0, close(pipefd[1])); 3098 ASSERT_EQ(1, read(pipefd[0], &buf, 1)); 3099 3100 ASSERT_EQ(0, ptrace(PTRACE_ATTACH, pid)); 3101 ASSERT_EQ(pid, waitpid(pid, NULL, 0)); 3102 3103 /* Past here must not use ASSERT or child process is never killed. */ 3104 3105 md.filter_off = 0; 3106 errno = 0; 3107 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md); 3108 EXPECT_EQ(sizeof(md), ret) { 3109 if (errno == EINVAL) 3110 SKIP(goto skip, "Kernel does not support PTRACE_SECCOMP_GET_METADATA (missing CONFIG_CHECKPOINT_RESTORE?)"); 3111 } 3112 3113 EXPECT_EQ(md.flags, SECCOMP_FILTER_FLAG_LOG); 3114 EXPECT_EQ(md.filter_off, 0); 3115 3116 md.filter_off = 1; 3117 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md); 3118 EXPECT_EQ(sizeof(md), ret); 3119 EXPECT_EQ(md.flags, 0); 3120 EXPECT_EQ(md.filter_off, 1); 3121 3122 skip: 3123 ASSERT_EQ(0, kill(pid, SIGKILL)); 3124 } 3125 3126 static int user_notif_syscall(int nr, unsigned int flags) 3127 { 3128 struct sock_filter filter[] = { 3129 BPF_STMT(BPF_LD+BPF_W+BPF_ABS, 3130 offsetof(struct seccomp_data, nr)), 3131 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, nr, 0, 1), 3132 BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_USER_NOTIF), 3133 BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW), 3134 }; 3135 3136 struct sock_fprog prog = { 3137 .len = (unsigned short)ARRAY_SIZE(filter), 3138 .filter = filter, 3139 }; 3140 3141 return seccomp(SECCOMP_SET_MODE_FILTER, flags, &prog); 3142 } 3143 3144 #define USER_NOTIF_MAGIC INT_MAX 3145 TEST(user_notification_basic) 3146 { 3147 pid_t pid; 3148 long ret; 3149 int status, listener; 3150 struct seccomp_notif req = {}; 3151 struct seccomp_notif_resp resp = {}; 3152 struct pollfd pollfd; 3153 3154 struct sock_filter filter[] = { 3155 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3156 }; 3157 struct sock_fprog prog = { 3158 .len = (unsigned short)ARRAY_SIZE(filter), 3159 .filter = filter, 3160 }; 3161 3162 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3163 ASSERT_EQ(0, ret) { 3164 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3165 } 3166 3167 pid = fork(); 3168 ASSERT_GE(pid, 0); 3169 3170 /* Check that we get -ENOSYS with no listener attached */ 3171 if (pid == 0) { 3172 if (user_notif_syscall(__NR_getppid, 0) < 0) 3173 exit(1); 3174 ret = syscall(__NR_getppid); 3175 exit(ret >= 0 || errno != ENOSYS); 3176 } 3177 3178 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3179 EXPECT_EQ(true, WIFEXITED(status)); 3180 EXPECT_EQ(0, WEXITSTATUS(status)); 3181 3182 /* Add some no-op filters for grins. */ 3183 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3184 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3185 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3186 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3187 3188 /* Check that the basic notification machinery works */ 3189 listener = user_notif_syscall(__NR_getppid, 3190 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3191 ASSERT_GE(listener, 0); 3192 3193 /* Installing a second listener in the chain should EBUSY */ 3194 EXPECT_EQ(user_notif_syscall(__NR_getppid, 3195 SECCOMP_FILTER_FLAG_NEW_LISTENER), 3196 -1); 3197 EXPECT_EQ(errno, EBUSY); 3198 3199 pid = fork(); 3200 ASSERT_GE(pid, 0); 3201 3202 if (pid == 0) { 3203 ret = syscall(__NR_getppid); 3204 exit(ret != USER_NOTIF_MAGIC); 3205 } 3206 3207 pollfd.fd = listener; 3208 pollfd.events = POLLIN | POLLOUT; 3209 3210 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3211 EXPECT_EQ(pollfd.revents, POLLIN); 3212 3213 /* Test that we can't pass garbage to the kernel. */ 3214 memset(&req, 0, sizeof(req)); 3215 req.pid = -1; 3216 errno = 0; 3217 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req); 3218 EXPECT_EQ(-1, ret); 3219 EXPECT_EQ(EINVAL, errno); 3220 3221 if (ret) { 3222 req.pid = 0; 3223 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3224 } 3225 3226 pollfd.fd = listener; 3227 pollfd.events = POLLIN | POLLOUT; 3228 3229 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3230 EXPECT_EQ(pollfd.revents, POLLOUT); 3231 3232 EXPECT_EQ(req.data.nr, __NR_getppid); 3233 3234 resp.id = req.id; 3235 resp.error = 0; 3236 resp.val = USER_NOTIF_MAGIC; 3237 3238 /* check that we make sure flags == 0 */ 3239 resp.flags = 1; 3240 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3241 EXPECT_EQ(errno, EINVAL); 3242 3243 resp.flags = 0; 3244 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3245 3246 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3247 EXPECT_EQ(true, WIFEXITED(status)); 3248 EXPECT_EQ(0, WEXITSTATUS(status)); 3249 } 3250 3251 TEST(user_notification_with_tsync) 3252 { 3253 int ret; 3254 unsigned int flags; 3255 3256 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3257 ASSERT_EQ(0, ret) { 3258 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3259 } 3260 3261 /* these were exclusive */ 3262 flags = SECCOMP_FILTER_FLAG_NEW_LISTENER | 3263 SECCOMP_FILTER_FLAG_TSYNC; 3264 ASSERT_EQ(-1, user_notif_syscall(__NR_getppid, flags)); 3265 ASSERT_EQ(EINVAL, errno); 3266 3267 /* but now they're not */ 3268 flags |= SECCOMP_FILTER_FLAG_TSYNC_ESRCH; 3269 ret = user_notif_syscall(__NR_getppid, flags); 3270 close(ret); 3271 ASSERT_LE(0, ret); 3272 } 3273 3274 TEST(user_notification_kill_in_middle) 3275 { 3276 pid_t pid; 3277 long ret; 3278 int listener; 3279 struct seccomp_notif req = {}; 3280 struct seccomp_notif_resp resp = {}; 3281 3282 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3283 ASSERT_EQ(0, ret) { 3284 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3285 } 3286 3287 listener = user_notif_syscall(__NR_getppid, 3288 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3289 ASSERT_GE(listener, 0); 3290 3291 /* 3292 * Check that nothing bad happens when we kill the task in the middle 3293 * of a syscall. 3294 */ 3295 pid = fork(); 3296 ASSERT_GE(pid, 0); 3297 3298 if (pid == 0) { 3299 ret = syscall(__NR_getppid); 3300 exit(ret != USER_NOTIF_MAGIC); 3301 } 3302 3303 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3304 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), 0); 3305 3306 EXPECT_EQ(kill(pid, SIGKILL), 0); 3307 EXPECT_EQ(waitpid(pid, NULL, 0), pid); 3308 3309 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), -1); 3310 3311 resp.id = req.id; 3312 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp); 3313 EXPECT_EQ(ret, -1); 3314 EXPECT_EQ(errno, ENOENT); 3315 } 3316 3317 static int handled = -1; 3318 3319 static void signal_handler(int signal) 3320 { 3321 if (write(handled, "c", 1) != 1) 3322 perror("write from signal"); 3323 } 3324 3325 TEST(user_notification_signal) 3326 { 3327 pid_t pid; 3328 long ret; 3329 int status, listener, sk_pair[2]; 3330 struct seccomp_notif req = {}; 3331 struct seccomp_notif_resp resp = {}; 3332 char c; 3333 3334 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3335 ASSERT_EQ(0, ret) { 3336 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3337 } 3338 3339 ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0); 3340 3341 listener = user_notif_syscall(__NR_gettid, 3342 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3343 ASSERT_GE(listener, 0); 3344 3345 pid = fork(); 3346 ASSERT_GE(pid, 0); 3347 3348 if (pid == 0) { 3349 close(sk_pair[0]); 3350 handled = sk_pair[1]; 3351 if (signal(SIGUSR1, signal_handler) == SIG_ERR) { 3352 perror("signal"); 3353 exit(1); 3354 } 3355 /* 3356 * ERESTARTSYS behavior is a bit hard to test, because we need 3357 * to rely on a signal that has not yet been handled. Let's at 3358 * least check that the error code gets propagated through, and 3359 * hope that it doesn't break when there is actually a signal :) 3360 */ 3361 ret = syscall(__NR_gettid); 3362 exit(!(ret == -1 && errno == 512)); 3363 } 3364 3365 close(sk_pair[1]); 3366 3367 memset(&req, 0, sizeof(req)); 3368 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3369 3370 EXPECT_EQ(kill(pid, SIGUSR1), 0); 3371 3372 /* 3373 * Make sure the signal really is delivered, which means we're not 3374 * stuck in the user notification code any more and the notification 3375 * should be dead. 3376 */ 3377 EXPECT_EQ(read(sk_pair[0], &c, 1), 1); 3378 3379 resp.id = req.id; 3380 resp.error = -EPERM; 3381 resp.val = 0; 3382 3383 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3384 EXPECT_EQ(errno, ENOENT); 3385 3386 memset(&req, 0, sizeof(req)); 3387 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3388 3389 resp.id = req.id; 3390 resp.error = -512; /* -ERESTARTSYS */ 3391 resp.val = 0; 3392 3393 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3394 3395 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3396 EXPECT_EQ(true, WIFEXITED(status)); 3397 EXPECT_EQ(0, WEXITSTATUS(status)); 3398 } 3399 3400 TEST(user_notification_closed_listener) 3401 { 3402 pid_t pid; 3403 long ret; 3404 int status, listener; 3405 3406 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3407 ASSERT_EQ(0, ret) { 3408 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3409 } 3410 3411 listener = user_notif_syscall(__NR_getppid, 3412 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3413 ASSERT_GE(listener, 0); 3414 3415 /* 3416 * Check that we get an ENOSYS when the listener is closed. 3417 */ 3418 pid = fork(); 3419 ASSERT_GE(pid, 0); 3420 if (pid == 0) { 3421 close(listener); 3422 ret = syscall(__NR_getppid); 3423 exit(ret != -1 && errno != ENOSYS); 3424 } 3425 3426 close(listener); 3427 3428 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3429 EXPECT_EQ(true, WIFEXITED(status)); 3430 EXPECT_EQ(0, WEXITSTATUS(status)); 3431 } 3432 3433 /* 3434 * Check that a pid in a child namespace still shows up as valid in ours. 3435 */ 3436 TEST(user_notification_child_pid_ns) 3437 { 3438 pid_t pid; 3439 int status, listener; 3440 struct seccomp_notif req = {}; 3441 struct seccomp_notif_resp resp = {}; 3442 3443 ASSERT_EQ(unshare(CLONE_NEWUSER | CLONE_NEWPID), 0) { 3444 if (errno == EINVAL) 3445 SKIP(return, "kernel missing CLONE_NEWUSER support"); 3446 }; 3447 3448 listener = user_notif_syscall(__NR_getppid, 3449 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3450 ASSERT_GE(listener, 0); 3451 3452 pid = fork(); 3453 ASSERT_GE(pid, 0); 3454 3455 if (pid == 0) 3456 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3457 3458 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3459 EXPECT_EQ(req.pid, pid); 3460 3461 resp.id = req.id; 3462 resp.error = 0; 3463 resp.val = USER_NOTIF_MAGIC; 3464 3465 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3466 3467 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3468 EXPECT_EQ(true, WIFEXITED(status)); 3469 EXPECT_EQ(0, WEXITSTATUS(status)); 3470 close(listener); 3471 } 3472 3473 /* 3474 * Check that a pid in a sibling (i.e. unrelated) namespace shows up as 0, i.e. 3475 * invalid. 3476 */ 3477 TEST(user_notification_sibling_pid_ns) 3478 { 3479 pid_t pid, pid2; 3480 int status, listener; 3481 struct seccomp_notif req = {}; 3482 struct seccomp_notif_resp resp = {}; 3483 3484 ASSERT_EQ(prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0), 0) { 3485 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3486 } 3487 3488 listener = user_notif_syscall(__NR_getppid, 3489 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3490 ASSERT_GE(listener, 0); 3491 3492 pid = fork(); 3493 ASSERT_GE(pid, 0); 3494 3495 if (pid == 0) { 3496 ASSERT_EQ(unshare(CLONE_NEWPID), 0); 3497 3498 pid2 = fork(); 3499 ASSERT_GE(pid2, 0); 3500 3501 if (pid2 == 0) 3502 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3503 3504 EXPECT_EQ(waitpid(pid2, &status, 0), pid2); 3505 EXPECT_EQ(true, WIFEXITED(status)); 3506 EXPECT_EQ(0, WEXITSTATUS(status)); 3507 exit(WEXITSTATUS(status)); 3508 } 3509 3510 /* Create the sibling ns, and sibling in it. */ 3511 ASSERT_EQ(unshare(CLONE_NEWPID), 0) { 3512 if (errno == EPERM) 3513 SKIP(return, "CLONE_NEWPID requires CAP_SYS_ADMIN"); 3514 } 3515 ASSERT_EQ(errno, 0); 3516 3517 pid2 = fork(); 3518 ASSERT_GE(pid2, 0); 3519 3520 if (pid2 == 0) { 3521 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3522 /* 3523 * The pid should be 0, i.e. the task is in some namespace that 3524 * we can't "see". 3525 */ 3526 EXPECT_EQ(req.pid, 0); 3527 3528 resp.id = req.id; 3529 resp.error = 0; 3530 resp.val = USER_NOTIF_MAGIC; 3531 3532 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3533 exit(0); 3534 } 3535 3536 close(listener); 3537 3538 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3539 EXPECT_EQ(true, WIFEXITED(status)); 3540 EXPECT_EQ(0, WEXITSTATUS(status)); 3541 3542 EXPECT_EQ(waitpid(pid2, &status, 0), pid2); 3543 EXPECT_EQ(true, WIFEXITED(status)); 3544 EXPECT_EQ(0, WEXITSTATUS(status)); 3545 } 3546 3547 TEST(user_notification_fault_recv) 3548 { 3549 pid_t pid; 3550 int status, listener; 3551 struct seccomp_notif req = {}; 3552 struct seccomp_notif_resp resp = {}; 3553 3554 ASSERT_EQ(unshare(CLONE_NEWUSER), 0); 3555 3556 listener = user_notif_syscall(__NR_getppid, 3557 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3558 ASSERT_GE(listener, 0); 3559 3560 pid = fork(); 3561 ASSERT_GE(pid, 0); 3562 3563 if (pid == 0) 3564 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3565 3566 /* Do a bad recv() */ 3567 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, NULL), -1); 3568 EXPECT_EQ(errno, EFAULT); 3569 3570 /* We should still be able to receive this notification, though. */ 3571 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3572 EXPECT_EQ(req.pid, pid); 3573 3574 resp.id = req.id; 3575 resp.error = 0; 3576 resp.val = USER_NOTIF_MAGIC; 3577 3578 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3579 3580 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3581 EXPECT_EQ(true, WIFEXITED(status)); 3582 EXPECT_EQ(0, WEXITSTATUS(status)); 3583 } 3584 3585 TEST(seccomp_get_notif_sizes) 3586 { 3587 struct seccomp_notif_sizes sizes; 3588 3589 ASSERT_EQ(seccomp(SECCOMP_GET_NOTIF_SIZES, 0, &sizes), 0); 3590 EXPECT_EQ(sizes.seccomp_notif, sizeof(struct seccomp_notif)); 3591 EXPECT_EQ(sizes.seccomp_notif_resp, sizeof(struct seccomp_notif_resp)); 3592 } 3593 3594 TEST(user_notification_continue) 3595 { 3596 pid_t pid; 3597 long ret; 3598 int status, listener; 3599 struct seccomp_notif req = {}; 3600 struct seccomp_notif_resp resp = {}; 3601 struct pollfd pollfd; 3602 3603 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3604 ASSERT_EQ(0, ret) { 3605 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3606 } 3607 3608 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER); 3609 ASSERT_GE(listener, 0); 3610 3611 pid = fork(); 3612 ASSERT_GE(pid, 0); 3613 3614 if (pid == 0) { 3615 int dup_fd, pipe_fds[2]; 3616 pid_t self; 3617 3618 ASSERT_GE(pipe(pipe_fds), 0); 3619 3620 dup_fd = dup(pipe_fds[0]); 3621 ASSERT_GE(dup_fd, 0); 3622 EXPECT_NE(pipe_fds[0], dup_fd); 3623 3624 self = getpid(); 3625 ASSERT_EQ(filecmp(self, self, pipe_fds[0], dup_fd), 0); 3626 exit(0); 3627 } 3628 3629 pollfd.fd = listener; 3630 pollfd.events = POLLIN | POLLOUT; 3631 3632 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3633 EXPECT_EQ(pollfd.revents, POLLIN); 3634 3635 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3636 3637 pollfd.fd = listener; 3638 pollfd.events = POLLIN | POLLOUT; 3639 3640 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3641 EXPECT_EQ(pollfd.revents, POLLOUT); 3642 3643 EXPECT_EQ(req.data.nr, __NR_dup); 3644 3645 resp.id = req.id; 3646 resp.flags = SECCOMP_USER_NOTIF_FLAG_CONTINUE; 3647 3648 /* 3649 * Verify that setting SECCOMP_USER_NOTIF_FLAG_CONTINUE enforces other 3650 * args be set to 0. 3651 */ 3652 resp.error = 0; 3653 resp.val = USER_NOTIF_MAGIC; 3654 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3655 EXPECT_EQ(errno, EINVAL); 3656 3657 resp.error = USER_NOTIF_MAGIC; 3658 resp.val = 0; 3659 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3660 EXPECT_EQ(errno, EINVAL); 3661 3662 resp.error = 0; 3663 resp.val = 0; 3664 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0) { 3665 if (errno == EINVAL) 3666 SKIP(goto skip, "Kernel does not support SECCOMP_USER_NOTIF_FLAG_CONTINUE"); 3667 } 3668 3669 skip: 3670 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3671 EXPECT_EQ(true, WIFEXITED(status)); 3672 EXPECT_EQ(0, WEXITSTATUS(status)) { 3673 if (WEXITSTATUS(status) == 2) { 3674 SKIP(return, "Kernel does not support kcmp() syscall"); 3675 return; 3676 } 3677 } 3678 } 3679 3680 TEST(user_notification_filter_empty) 3681 { 3682 pid_t pid; 3683 long ret; 3684 int status; 3685 struct pollfd pollfd; 3686 struct clone_args args = { 3687 .flags = CLONE_FILES, 3688 .exit_signal = SIGCHLD, 3689 }; 3690 3691 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3692 ASSERT_EQ(0, ret) { 3693 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3694 } 3695 3696 pid = sys_clone3(&args, sizeof(args)); 3697 ASSERT_GE(pid, 0); 3698 3699 if (pid == 0) { 3700 int listener; 3701 3702 listener = user_notif_syscall(__NR_mknod, SECCOMP_FILTER_FLAG_NEW_LISTENER); 3703 if (listener < 0) 3704 _exit(EXIT_FAILURE); 3705 3706 if (dup2(listener, 200) != 200) 3707 _exit(EXIT_FAILURE); 3708 3709 close(listener); 3710 3711 _exit(EXIT_SUCCESS); 3712 } 3713 3714 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3715 EXPECT_EQ(true, WIFEXITED(status)); 3716 EXPECT_EQ(0, WEXITSTATUS(status)); 3717 3718 /* 3719 * The seccomp filter has become unused so we should be notified once 3720 * the kernel gets around to cleaning up task struct. 3721 */ 3722 pollfd.fd = 200; 3723 pollfd.events = POLLHUP; 3724 3725 EXPECT_GT(poll(&pollfd, 1, 2000), 0); 3726 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0); 3727 } 3728 3729 static void *do_thread(void *data) 3730 { 3731 return NULL; 3732 } 3733 3734 TEST(user_notification_filter_empty_threaded) 3735 { 3736 pid_t pid; 3737 long ret; 3738 int status; 3739 struct pollfd pollfd; 3740 struct clone_args args = { 3741 .flags = CLONE_FILES, 3742 .exit_signal = SIGCHLD, 3743 }; 3744 3745 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3746 ASSERT_EQ(0, ret) { 3747 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3748 } 3749 3750 pid = sys_clone3(&args, sizeof(args)); 3751 ASSERT_GE(pid, 0); 3752 3753 if (pid == 0) { 3754 pid_t pid1, pid2; 3755 int listener, status; 3756 pthread_t thread; 3757 3758 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER); 3759 if (listener < 0) 3760 _exit(EXIT_FAILURE); 3761 3762 if (dup2(listener, 200) != 200) 3763 _exit(EXIT_FAILURE); 3764 3765 close(listener); 3766 3767 pid1 = fork(); 3768 if (pid1 < 0) 3769 _exit(EXIT_FAILURE); 3770 3771 if (pid1 == 0) 3772 _exit(EXIT_SUCCESS); 3773 3774 pid2 = fork(); 3775 if (pid2 < 0) 3776 _exit(EXIT_FAILURE); 3777 3778 if (pid2 == 0) 3779 _exit(EXIT_SUCCESS); 3780 3781 if (pthread_create(&thread, NULL, do_thread, NULL) || 3782 pthread_join(thread, NULL)) 3783 _exit(EXIT_FAILURE); 3784 3785 if (pthread_create(&thread, NULL, do_thread, NULL) || 3786 pthread_join(thread, NULL)) 3787 _exit(EXIT_FAILURE); 3788 3789 if (waitpid(pid1, &status, 0) != pid1 || !WIFEXITED(status) || 3790 WEXITSTATUS(status)) 3791 _exit(EXIT_FAILURE); 3792 3793 if (waitpid(pid2, &status, 0) != pid2 || !WIFEXITED(status) || 3794 WEXITSTATUS(status)) 3795 _exit(EXIT_FAILURE); 3796 3797 exit(EXIT_SUCCESS); 3798 } 3799 3800 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3801 EXPECT_EQ(true, WIFEXITED(status)); 3802 EXPECT_EQ(0, WEXITSTATUS(status)); 3803 3804 /* 3805 * The seccomp filter has become unused so we should be notified once 3806 * the kernel gets around to cleaning up task struct. 3807 */ 3808 pollfd.fd = 200; 3809 pollfd.events = POLLHUP; 3810 3811 EXPECT_GT(poll(&pollfd, 1, 2000), 0); 3812 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0); 3813 } 3814 3815 TEST(user_notification_addfd) 3816 { 3817 pid_t pid; 3818 long ret; 3819 int status, listener, memfd, fd; 3820 struct seccomp_notif_addfd addfd = {}; 3821 struct seccomp_notif_addfd_small small = {}; 3822 struct seccomp_notif_addfd_big big = {}; 3823 struct seccomp_notif req = {}; 3824 struct seccomp_notif_resp resp = {}; 3825 /* 100 ms */ 3826 struct timespec delay = { .tv_nsec = 100000000 }; 3827 3828 memfd = memfd_create("test", 0); 3829 ASSERT_GE(memfd, 0); 3830 3831 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3832 ASSERT_EQ(0, ret) { 3833 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3834 } 3835 3836 /* Check that the basic notification machinery works */ 3837 listener = user_notif_syscall(__NR_getppid, 3838 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3839 ASSERT_GE(listener, 0); 3840 3841 pid = fork(); 3842 ASSERT_GE(pid, 0); 3843 3844 if (pid == 0) { 3845 if (syscall(__NR_getppid) != USER_NOTIF_MAGIC) 3846 exit(1); 3847 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3848 } 3849 3850 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3851 3852 addfd.srcfd = memfd; 3853 addfd.newfd = 0; 3854 addfd.id = req.id; 3855 addfd.flags = 0x0; 3856 3857 /* Verify bad newfd_flags cannot be set */ 3858 addfd.newfd_flags = ~O_CLOEXEC; 3859 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 3860 EXPECT_EQ(errno, EINVAL); 3861 addfd.newfd_flags = O_CLOEXEC; 3862 3863 /* Verify bad flags cannot be set */ 3864 addfd.flags = 0xff; 3865 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 3866 EXPECT_EQ(errno, EINVAL); 3867 addfd.flags = 0; 3868 3869 /* Verify that remote_fd cannot be set without setting flags */ 3870 addfd.newfd = 1; 3871 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 3872 EXPECT_EQ(errno, EINVAL); 3873 addfd.newfd = 0; 3874 3875 /* Verify small size cannot be set */ 3876 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_SMALL, &small), -1); 3877 EXPECT_EQ(errno, EINVAL); 3878 3879 /* Verify we can't send bits filled in unknown buffer area */ 3880 memset(&big, 0xAA, sizeof(big)); 3881 big.addfd = addfd; 3882 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big), -1); 3883 EXPECT_EQ(errno, E2BIG); 3884 3885 3886 /* Verify we can set an arbitrary remote fd */ 3887 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd); 3888 /* 3889 * The child has fds 0(stdin), 1(stdout), 2(stderr), 3(memfd), 3890 * 4(listener), so the newly allocated fd should be 5. 3891 */ 3892 EXPECT_EQ(fd, 5); 3893 EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0); 3894 3895 /* Verify we can set an arbitrary remote fd with large size */ 3896 memset(&big, 0x0, sizeof(big)); 3897 big.addfd = addfd; 3898 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big); 3899 EXPECT_EQ(fd, 6); 3900 3901 /* Verify we can set a specific remote fd */ 3902 addfd.newfd = 42; 3903 addfd.flags = SECCOMP_ADDFD_FLAG_SETFD; 3904 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd); 3905 EXPECT_EQ(fd, 42); 3906 EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0); 3907 3908 /* Resume syscall */ 3909 resp.id = req.id; 3910 resp.error = 0; 3911 resp.val = USER_NOTIF_MAGIC; 3912 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3913 3914 /* 3915 * This sets the ID of the ADD FD to the last request plus 1. The 3916 * notification ID increments 1 per notification. 3917 */ 3918 addfd.id = req.id + 1; 3919 3920 /* This spins until the underlying notification is generated */ 3921 while (ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd) != -1 && 3922 errno != -EINPROGRESS) 3923 nanosleep(&delay, NULL); 3924 3925 memset(&req, 0, sizeof(req)); 3926 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3927 ASSERT_EQ(addfd.id, req.id); 3928 3929 resp.id = req.id; 3930 resp.error = 0; 3931 resp.val = USER_NOTIF_MAGIC; 3932 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3933 3934 /* Wait for child to finish. */ 3935 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3936 EXPECT_EQ(true, WIFEXITED(status)); 3937 EXPECT_EQ(0, WEXITSTATUS(status)); 3938 3939 close(memfd); 3940 } 3941 3942 TEST(user_notification_addfd_rlimit) 3943 { 3944 pid_t pid; 3945 long ret; 3946 int status, listener, memfd; 3947 struct seccomp_notif_addfd addfd = {}; 3948 struct seccomp_notif req = {}; 3949 struct seccomp_notif_resp resp = {}; 3950 const struct rlimit lim = { 3951 .rlim_cur = 0, 3952 .rlim_max = 0, 3953 }; 3954 3955 memfd = memfd_create("test", 0); 3956 ASSERT_GE(memfd, 0); 3957 3958 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3959 ASSERT_EQ(0, ret) { 3960 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3961 } 3962 3963 /* Check that the basic notification machinery works */ 3964 listener = user_notif_syscall(__NR_getppid, 3965 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3966 ASSERT_GE(listener, 0); 3967 3968 pid = fork(); 3969 ASSERT_GE(pid, 0); 3970 3971 if (pid == 0) 3972 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3973 3974 3975 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3976 3977 ASSERT_EQ(prlimit(pid, RLIMIT_NOFILE, &lim, NULL), 0); 3978 3979 addfd.srcfd = memfd; 3980 addfd.newfd_flags = O_CLOEXEC; 3981 addfd.newfd = 0; 3982 addfd.id = req.id; 3983 addfd.flags = 0; 3984 3985 /* Should probably spot check /proc/sys/fs/file-nr */ 3986 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 3987 EXPECT_EQ(errno, EMFILE); 3988 3989 addfd.newfd = 100; 3990 addfd.flags = SECCOMP_ADDFD_FLAG_SETFD; 3991 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 3992 EXPECT_EQ(errno, EBADF); 3993 3994 resp.id = req.id; 3995 resp.error = 0; 3996 resp.val = USER_NOTIF_MAGIC; 3997 3998 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3999 4000 /* Wait for child to finish. */ 4001 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4002 EXPECT_EQ(true, WIFEXITED(status)); 4003 EXPECT_EQ(0, WEXITSTATUS(status)); 4004 4005 close(memfd); 4006 } 4007 4008 /* 4009 * TODO: 4010 * - expand NNP testing 4011 * - better arch-specific TRACE and TRAP handlers. 4012 * - endianness checking when appropriate 4013 * - 64-bit arg prodding 4014 * - arch value testing (x86 modes especially) 4015 * - verify that FILTER_FLAG_LOG filters generate log messages 4016 * - verify that RET_LOG generates log messages 4017 */ 4018 4019 TEST_HARNESS_MAIN 4020