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 #include <sys/capability.h> 50 #include <linux/perf_event.h> 51 52 #include <unistd.h> 53 #include <sys/syscall.h> 54 #include <poll.h> 55 56 #include "../kselftest_harness.h" 57 #include "../clone3/clone3_selftests.h" 58 59 /* Attempt to de-conflict with the selftests tree. */ 60 #ifndef SKIP 61 #define SKIP(s, ...) XFAIL(s, ##__VA_ARGS__) 62 #endif 63 64 #ifndef MIN 65 #define MIN(X, Y) ((X) < (Y) ? (X) : (Y)) 66 #endif 67 68 #ifndef PR_SET_PTRACER 69 # define PR_SET_PTRACER 0x59616d61 70 #endif 71 72 #ifndef noinline 73 #define noinline __attribute__((noinline)) 74 #endif 75 76 #ifndef PR_SET_NO_NEW_PRIVS 77 #define PR_SET_NO_NEW_PRIVS 38 78 #define PR_GET_NO_NEW_PRIVS 39 79 #endif 80 81 #ifndef PR_SECCOMP_EXT 82 #define PR_SECCOMP_EXT 43 83 #endif 84 85 #ifndef SECCOMP_EXT_ACT 86 #define SECCOMP_EXT_ACT 1 87 #endif 88 89 #ifndef SECCOMP_EXT_ACT_TSYNC 90 #define SECCOMP_EXT_ACT_TSYNC 1 91 #endif 92 93 #ifndef SECCOMP_MODE_STRICT 94 #define SECCOMP_MODE_STRICT 1 95 #endif 96 97 #ifndef SECCOMP_MODE_FILTER 98 #define SECCOMP_MODE_FILTER 2 99 #endif 100 101 #ifndef SECCOMP_RET_ALLOW 102 struct seccomp_data { 103 int nr; 104 __u32 arch; 105 __u64 instruction_pointer; 106 __u64 args[6]; 107 }; 108 #endif 109 110 #ifndef SECCOMP_RET_KILL_PROCESS 111 #define SECCOMP_RET_KILL_PROCESS 0x80000000U /* kill the process */ 112 #define SECCOMP_RET_KILL_THREAD 0x00000000U /* kill the thread */ 113 #endif 114 #ifndef SECCOMP_RET_KILL 115 #define SECCOMP_RET_KILL SECCOMP_RET_KILL_THREAD 116 #define SECCOMP_RET_TRAP 0x00030000U /* disallow and force a SIGSYS */ 117 #define SECCOMP_RET_ERRNO 0x00050000U /* returns an errno */ 118 #define SECCOMP_RET_TRACE 0x7ff00000U /* pass to a tracer or disallow */ 119 #define SECCOMP_RET_ALLOW 0x7fff0000U /* allow */ 120 #endif 121 #ifndef SECCOMP_RET_LOG 122 #define SECCOMP_RET_LOG 0x7ffc0000U /* allow after logging */ 123 #endif 124 125 #ifndef __NR_seccomp 126 # if defined(__i386__) 127 # define __NR_seccomp 354 128 # elif defined(__x86_64__) 129 # define __NR_seccomp 317 130 # elif defined(__arm__) 131 # define __NR_seccomp 383 132 # elif defined(__aarch64__) 133 # define __NR_seccomp 277 134 # elif defined(__riscv) 135 # define __NR_seccomp 277 136 # elif defined(__csky__) 137 # define __NR_seccomp 277 138 # elif defined(__loongarch__) 139 # define __NR_seccomp 277 140 # elif defined(__hppa__) 141 # define __NR_seccomp 338 142 # elif defined(__powerpc__) 143 # define __NR_seccomp 358 144 # elif defined(__s390__) 145 # define __NR_seccomp 348 146 # elif defined(__xtensa__) 147 # define __NR_seccomp 337 148 # elif defined(__sh__) 149 # define __NR_seccomp 372 150 # elif defined(__mc68000__) 151 # define __NR_seccomp 380 152 # else 153 # warning "seccomp syscall number unknown for this architecture" 154 # define __NR_seccomp 0xffff 155 # endif 156 #endif 157 158 #ifndef SECCOMP_SET_MODE_STRICT 159 #define SECCOMP_SET_MODE_STRICT 0 160 #endif 161 162 #ifndef SECCOMP_SET_MODE_FILTER 163 #define SECCOMP_SET_MODE_FILTER 1 164 #endif 165 166 #ifndef SECCOMP_GET_ACTION_AVAIL 167 #define SECCOMP_GET_ACTION_AVAIL 2 168 #endif 169 170 #ifndef SECCOMP_GET_NOTIF_SIZES 171 #define SECCOMP_GET_NOTIF_SIZES 3 172 #endif 173 174 #ifndef SECCOMP_FILTER_FLAG_TSYNC 175 #define SECCOMP_FILTER_FLAG_TSYNC (1UL << 0) 176 #endif 177 178 #ifndef SECCOMP_FILTER_FLAG_LOG 179 #define SECCOMP_FILTER_FLAG_LOG (1UL << 1) 180 #endif 181 182 #ifndef SECCOMP_FILTER_FLAG_SPEC_ALLOW 183 #define SECCOMP_FILTER_FLAG_SPEC_ALLOW (1UL << 2) 184 #endif 185 186 #ifndef PTRACE_SECCOMP_GET_METADATA 187 #define PTRACE_SECCOMP_GET_METADATA 0x420d 188 189 struct seccomp_metadata { 190 __u64 filter_off; /* Input: which filter */ 191 __u64 flags; /* Output: filter's flags */ 192 }; 193 #endif 194 195 #ifndef SECCOMP_FILTER_FLAG_NEW_LISTENER 196 #define SECCOMP_FILTER_FLAG_NEW_LISTENER (1UL << 3) 197 #endif 198 199 #ifndef SECCOMP_RET_USER_NOTIF 200 #define SECCOMP_RET_USER_NOTIF 0x7fc00000U 201 202 #define SECCOMP_IOC_MAGIC '!' 203 #define SECCOMP_IO(nr) _IO(SECCOMP_IOC_MAGIC, nr) 204 #define SECCOMP_IOR(nr, type) _IOR(SECCOMP_IOC_MAGIC, nr, type) 205 #define SECCOMP_IOW(nr, type) _IOW(SECCOMP_IOC_MAGIC, nr, type) 206 #define SECCOMP_IOWR(nr, type) _IOWR(SECCOMP_IOC_MAGIC, nr, type) 207 208 /* Flags for seccomp notification fd ioctl. */ 209 #define SECCOMP_IOCTL_NOTIF_RECV SECCOMP_IOWR(0, struct seccomp_notif) 210 #define SECCOMP_IOCTL_NOTIF_SEND SECCOMP_IOWR(1, \ 211 struct seccomp_notif_resp) 212 #define SECCOMP_IOCTL_NOTIF_ID_VALID SECCOMP_IOW(2, __u64) 213 214 struct seccomp_notif { 215 __u64 id; 216 __u32 pid; 217 __u32 flags; 218 struct seccomp_data data; 219 }; 220 221 struct seccomp_notif_resp { 222 __u64 id; 223 __s64 val; 224 __s32 error; 225 __u32 flags; 226 }; 227 228 struct seccomp_notif_sizes { 229 __u16 seccomp_notif; 230 __u16 seccomp_notif_resp; 231 __u16 seccomp_data; 232 }; 233 #endif 234 235 #ifndef SECCOMP_IOCTL_NOTIF_ADDFD 236 /* On success, the return value is the remote process's added fd number */ 237 #define SECCOMP_IOCTL_NOTIF_ADDFD SECCOMP_IOW(3, \ 238 struct seccomp_notif_addfd) 239 240 /* valid flags for seccomp_notif_addfd */ 241 #define SECCOMP_ADDFD_FLAG_SETFD (1UL << 0) /* Specify remote fd */ 242 243 struct seccomp_notif_addfd { 244 __u64 id; 245 __u32 flags; 246 __u32 srcfd; 247 __u32 newfd; 248 __u32 newfd_flags; 249 }; 250 #endif 251 252 #ifndef SECCOMP_ADDFD_FLAG_SEND 253 #define SECCOMP_ADDFD_FLAG_SEND (1UL << 1) /* Addfd and return it, atomically */ 254 #endif 255 256 struct seccomp_notif_addfd_small { 257 __u64 id; 258 char weird[4]; 259 }; 260 #define SECCOMP_IOCTL_NOTIF_ADDFD_SMALL \ 261 SECCOMP_IOW(3, struct seccomp_notif_addfd_small) 262 263 struct seccomp_notif_addfd_big { 264 union { 265 struct seccomp_notif_addfd addfd; 266 char buf[sizeof(struct seccomp_notif_addfd) + 8]; 267 }; 268 }; 269 #define SECCOMP_IOCTL_NOTIF_ADDFD_BIG \ 270 SECCOMP_IOWR(3, struct seccomp_notif_addfd_big) 271 272 #ifndef PTRACE_EVENTMSG_SYSCALL_ENTRY 273 #define PTRACE_EVENTMSG_SYSCALL_ENTRY 1 274 #define PTRACE_EVENTMSG_SYSCALL_EXIT 2 275 #endif 276 277 #ifndef SECCOMP_USER_NOTIF_FLAG_CONTINUE 278 #define SECCOMP_USER_NOTIF_FLAG_CONTINUE 0x00000001 279 #endif 280 281 #ifndef SECCOMP_FILTER_FLAG_TSYNC_ESRCH 282 #define SECCOMP_FILTER_FLAG_TSYNC_ESRCH (1UL << 4) 283 #endif 284 285 #ifndef SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV 286 #define SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV (1UL << 5) 287 #endif 288 289 #ifndef seccomp 290 int seccomp(unsigned int op, unsigned int flags, void *args) 291 { 292 errno = 0; 293 return syscall(__NR_seccomp, op, flags, args); 294 } 295 #endif 296 297 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 298 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n])) 299 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 300 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32)) 301 #else 302 #error "wut? Unknown __BYTE_ORDER__?!" 303 #endif 304 305 #define SIBLING_EXIT_UNKILLED 0xbadbeef 306 #define SIBLING_EXIT_FAILURE 0xbadface 307 #define SIBLING_EXIT_NEWPRIVS 0xbadfeed 308 309 static int __filecmp(pid_t pid1, pid_t pid2, int fd1, int fd2) 310 { 311 #ifdef __NR_kcmp 312 errno = 0; 313 return syscall(__NR_kcmp, pid1, pid2, KCMP_FILE, fd1, fd2); 314 #else 315 errno = ENOSYS; 316 return -1; 317 #endif 318 } 319 320 /* Have TH_LOG report actual location filecmp() is used. */ 321 #define filecmp(pid1, pid2, fd1, fd2) ({ \ 322 int _ret; \ 323 \ 324 _ret = __filecmp(pid1, pid2, fd1, fd2); \ 325 if (_ret != 0) { \ 326 if (_ret < 0 && errno == ENOSYS) { \ 327 TH_LOG("kcmp() syscall missing (test is less accurate)");\ 328 _ret = 0; \ 329 } \ 330 } \ 331 _ret; }) 332 333 TEST(kcmp) 334 { 335 int ret; 336 337 ret = __filecmp(getpid(), getpid(), 1, 1); 338 EXPECT_EQ(ret, 0); 339 if (ret != 0 && errno == ENOSYS) 340 SKIP(return, "Kernel does not support kcmp() (missing CONFIG_KCMP?)"); 341 } 342 343 TEST(mode_strict_support) 344 { 345 long ret; 346 347 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL); 348 ASSERT_EQ(0, ret) { 349 TH_LOG("Kernel does not support CONFIG_SECCOMP"); 350 } 351 syscall(__NR_exit, 0); 352 } 353 354 TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL) 355 { 356 long ret; 357 358 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL); 359 ASSERT_EQ(0, ret) { 360 TH_LOG("Kernel does not support CONFIG_SECCOMP"); 361 } 362 syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER, 363 NULL, NULL, NULL); 364 EXPECT_FALSE(true) { 365 TH_LOG("Unreachable!"); 366 } 367 } 368 369 /* Note! This doesn't test no new privs behavior */ 370 TEST(no_new_privs_support) 371 { 372 long ret; 373 374 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 375 EXPECT_EQ(0, ret) { 376 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 377 } 378 } 379 380 /* Tests kernel support by checking for a copy_from_user() fault on NULL. */ 381 TEST(mode_filter_support) 382 { 383 long ret; 384 385 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 386 ASSERT_EQ(0, ret) { 387 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 388 } 389 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL); 390 EXPECT_EQ(-1, ret); 391 EXPECT_EQ(EFAULT, errno) { 392 TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!"); 393 } 394 } 395 396 TEST(mode_filter_without_nnp) 397 { 398 struct sock_filter filter[] = { 399 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 400 }; 401 struct sock_fprog prog = { 402 .len = (unsigned short)ARRAY_SIZE(filter), 403 .filter = filter, 404 }; 405 long ret; 406 cap_t cap = cap_get_proc(); 407 cap_flag_value_t is_cap_sys_admin = 0; 408 409 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0); 410 ASSERT_LE(0, ret) { 411 TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS"); 412 } 413 errno = 0; 414 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 415 /* Succeeds with CAP_SYS_ADMIN, fails without */ 416 cap_get_flag(cap, CAP_SYS_ADMIN, CAP_EFFECTIVE, &is_cap_sys_admin); 417 if (!is_cap_sys_admin) { 418 EXPECT_EQ(-1, ret); 419 EXPECT_EQ(EACCES, errno); 420 } else { 421 EXPECT_EQ(0, ret); 422 } 423 } 424 425 #define MAX_INSNS_PER_PATH 32768 426 427 TEST(filter_size_limits) 428 { 429 int i; 430 int count = BPF_MAXINSNS + 1; 431 struct sock_filter allow[] = { 432 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 433 }; 434 struct sock_filter *filter; 435 struct sock_fprog prog = { }; 436 long ret; 437 438 filter = calloc(count, sizeof(*filter)); 439 ASSERT_NE(NULL, filter); 440 441 for (i = 0; i < count; i++) 442 filter[i] = allow[0]; 443 444 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 445 ASSERT_EQ(0, ret); 446 447 prog.filter = filter; 448 prog.len = count; 449 450 /* Too many filter instructions in a single filter. */ 451 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 452 ASSERT_NE(0, ret) { 453 TH_LOG("Installing %d insn filter was allowed", prog.len); 454 } 455 456 /* One less is okay, though. */ 457 prog.len -= 1; 458 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 459 ASSERT_EQ(0, ret) { 460 TH_LOG("Installing %d insn filter wasn't allowed", prog.len); 461 } 462 } 463 464 TEST(filter_chain_limits) 465 { 466 int i; 467 int count = BPF_MAXINSNS; 468 struct sock_filter allow[] = { 469 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 470 }; 471 struct sock_filter *filter; 472 struct sock_fprog prog = { }; 473 long ret; 474 475 filter = calloc(count, sizeof(*filter)); 476 ASSERT_NE(NULL, filter); 477 478 for (i = 0; i < count; i++) 479 filter[i] = allow[0]; 480 481 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 482 ASSERT_EQ(0, ret); 483 484 prog.filter = filter; 485 prog.len = 1; 486 487 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 488 ASSERT_EQ(0, ret); 489 490 prog.len = count; 491 492 /* Too many total filter instructions. */ 493 for (i = 0; i < MAX_INSNS_PER_PATH; i++) { 494 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 495 if (ret != 0) 496 break; 497 } 498 ASSERT_NE(0, ret) { 499 TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)", 500 i, count, i * (count + 4)); 501 } 502 } 503 504 TEST(mode_filter_cannot_move_to_strict) 505 { 506 struct sock_filter filter[] = { 507 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 508 }; 509 struct sock_fprog prog = { 510 .len = (unsigned short)ARRAY_SIZE(filter), 511 .filter = filter, 512 }; 513 long ret; 514 515 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 516 ASSERT_EQ(0, ret); 517 518 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 519 ASSERT_EQ(0, ret); 520 521 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0); 522 EXPECT_EQ(-1, ret); 523 EXPECT_EQ(EINVAL, errno); 524 } 525 526 527 TEST(mode_filter_get_seccomp) 528 { 529 struct sock_filter filter[] = { 530 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 531 }; 532 struct sock_fprog prog = { 533 .len = (unsigned short)ARRAY_SIZE(filter), 534 .filter = filter, 535 }; 536 long ret; 537 538 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 539 ASSERT_EQ(0, ret); 540 541 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0); 542 EXPECT_EQ(0, ret); 543 544 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 545 ASSERT_EQ(0, ret); 546 547 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0); 548 EXPECT_EQ(2, ret); 549 } 550 551 552 TEST(ALLOW_all) 553 { 554 struct sock_filter filter[] = { 555 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 556 }; 557 struct sock_fprog prog = { 558 .len = (unsigned short)ARRAY_SIZE(filter), 559 .filter = filter, 560 }; 561 long ret; 562 563 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 564 ASSERT_EQ(0, ret); 565 566 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 567 ASSERT_EQ(0, ret); 568 } 569 570 TEST(empty_prog) 571 { 572 struct sock_filter filter[] = { 573 }; 574 struct sock_fprog prog = { 575 .len = (unsigned short)ARRAY_SIZE(filter), 576 .filter = filter, 577 }; 578 long ret; 579 580 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 581 ASSERT_EQ(0, ret); 582 583 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 584 EXPECT_EQ(-1, ret); 585 EXPECT_EQ(EINVAL, errno); 586 } 587 588 TEST(log_all) 589 { 590 struct sock_filter filter[] = { 591 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG), 592 }; 593 struct sock_fprog prog = { 594 .len = (unsigned short)ARRAY_SIZE(filter), 595 .filter = filter, 596 }; 597 long ret; 598 pid_t parent = getppid(); 599 600 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 601 ASSERT_EQ(0, ret); 602 603 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 604 ASSERT_EQ(0, ret); 605 606 /* getppid() should succeed and be logged (no check for logging) */ 607 EXPECT_EQ(parent, syscall(__NR_getppid)); 608 } 609 610 TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS) 611 { 612 struct sock_filter filter[] = { 613 BPF_STMT(BPF_RET|BPF_K, 0x10000000U), 614 }; 615 struct sock_fprog prog = { 616 .len = (unsigned short)ARRAY_SIZE(filter), 617 .filter = filter, 618 }; 619 long ret; 620 621 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 622 ASSERT_EQ(0, ret); 623 624 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 625 ASSERT_EQ(0, ret); 626 EXPECT_EQ(0, syscall(__NR_getpid)) { 627 TH_LOG("getpid() shouldn't ever return"); 628 } 629 } 630 631 /* return code >= 0x80000000 is unused. */ 632 TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS) 633 { 634 struct sock_filter filter[] = { 635 BPF_STMT(BPF_RET|BPF_K, 0x90000000U), 636 }; 637 struct sock_fprog prog = { 638 .len = (unsigned short)ARRAY_SIZE(filter), 639 .filter = filter, 640 }; 641 long ret; 642 643 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 644 ASSERT_EQ(0, ret); 645 646 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 647 ASSERT_EQ(0, ret); 648 EXPECT_EQ(0, syscall(__NR_getpid)) { 649 TH_LOG("getpid() shouldn't ever return"); 650 } 651 } 652 653 TEST_SIGNAL(KILL_all, SIGSYS) 654 { 655 struct sock_filter filter[] = { 656 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 657 }; 658 struct sock_fprog prog = { 659 .len = (unsigned short)ARRAY_SIZE(filter), 660 .filter = filter, 661 }; 662 long ret; 663 664 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 665 ASSERT_EQ(0, ret); 666 667 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 668 ASSERT_EQ(0, ret); 669 } 670 671 TEST_SIGNAL(KILL_one, SIGSYS) 672 { 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_getpid, 0, 1), 677 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 678 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 679 }; 680 struct sock_fprog prog = { 681 .len = (unsigned short)ARRAY_SIZE(filter), 682 .filter = filter, 683 }; 684 long ret; 685 pid_t parent = getppid(); 686 687 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 688 ASSERT_EQ(0, ret); 689 690 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 691 ASSERT_EQ(0, ret); 692 693 EXPECT_EQ(parent, syscall(__NR_getppid)); 694 /* getpid() should never return. */ 695 EXPECT_EQ(0, syscall(__NR_getpid)); 696 } 697 698 TEST_SIGNAL(KILL_one_arg_one, SIGSYS) 699 { 700 void *fatal_address; 701 struct sock_filter filter[] = { 702 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 703 offsetof(struct seccomp_data, nr)), 704 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0), 705 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 706 /* Only both with lower 32-bit for now. */ 707 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)), 708 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 709 (unsigned long)&fatal_address, 0, 1), 710 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 711 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 712 }; 713 struct sock_fprog prog = { 714 .len = (unsigned short)ARRAY_SIZE(filter), 715 .filter = filter, 716 }; 717 long ret; 718 pid_t parent = getppid(); 719 struct tms timebuf; 720 clock_t clock = times(&timebuf); 721 722 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 723 ASSERT_EQ(0, ret); 724 725 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 726 ASSERT_EQ(0, ret); 727 728 EXPECT_EQ(parent, syscall(__NR_getppid)); 729 EXPECT_LE(clock, syscall(__NR_times, &timebuf)); 730 /* times() should never return. */ 731 EXPECT_EQ(0, syscall(__NR_times, &fatal_address)); 732 } 733 734 TEST_SIGNAL(KILL_one_arg_six, SIGSYS) 735 { 736 #ifndef __NR_mmap2 737 int sysno = __NR_mmap; 738 #else 739 int sysno = __NR_mmap2; 740 #endif 741 struct sock_filter filter[] = { 742 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 743 offsetof(struct seccomp_data, nr)), 744 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0), 745 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 746 /* Only both with lower 32-bit for now. */ 747 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)), 748 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1), 749 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 750 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 751 }; 752 struct sock_fprog prog = { 753 .len = (unsigned short)ARRAY_SIZE(filter), 754 .filter = filter, 755 }; 756 long ret; 757 pid_t parent = getppid(); 758 int fd; 759 void *map1, *map2; 760 int page_size = sysconf(_SC_PAGESIZE); 761 762 ASSERT_LT(0, page_size); 763 764 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 765 ASSERT_EQ(0, ret); 766 767 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 768 ASSERT_EQ(0, ret); 769 770 fd = open("/dev/zero", O_RDONLY); 771 ASSERT_NE(-1, fd); 772 773 EXPECT_EQ(parent, syscall(__NR_getppid)); 774 map1 = (void *)syscall(sysno, 775 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size); 776 EXPECT_NE(MAP_FAILED, map1); 777 /* mmap2() should never return. */ 778 map2 = (void *)syscall(sysno, 779 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE); 780 EXPECT_EQ(MAP_FAILED, map2); 781 782 /* The test failed, so clean up the resources. */ 783 munmap(map1, page_size); 784 munmap(map2, page_size); 785 close(fd); 786 } 787 788 /* This is a thread task to die via seccomp filter violation. */ 789 void *kill_thread(void *data) 790 { 791 bool die = (bool)data; 792 793 if (die) { 794 syscall(__NR_getpid); 795 return (void *)SIBLING_EXIT_FAILURE; 796 } 797 798 return (void *)SIBLING_EXIT_UNKILLED; 799 } 800 801 enum kill_t { 802 KILL_THREAD, 803 KILL_PROCESS, 804 RET_UNKNOWN 805 }; 806 807 /* Prepare a thread that will kill itself or both of us. */ 808 void kill_thread_or_group(struct __test_metadata *_metadata, 809 enum kill_t kill_how) 810 { 811 pthread_t thread; 812 void *status; 813 /* Kill only when calling __NR_getpid. */ 814 struct sock_filter filter_thread[] = { 815 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 816 offsetof(struct seccomp_data, nr)), 817 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 818 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD), 819 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 820 }; 821 struct sock_fprog prog_thread = { 822 .len = (unsigned short)ARRAY_SIZE(filter_thread), 823 .filter = filter_thread, 824 }; 825 int kill = kill_how == KILL_PROCESS ? SECCOMP_RET_KILL_PROCESS : 0xAAAAAAAA; 826 struct sock_filter filter_process[] = { 827 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 828 offsetof(struct seccomp_data, nr)), 829 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 830 BPF_STMT(BPF_RET|BPF_K, kill), 831 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 832 }; 833 struct sock_fprog prog_process = { 834 .len = (unsigned short)ARRAY_SIZE(filter_process), 835 .filter = filter_process, 836 }; 837 838 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 839 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 840 } 841 842 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, 843 kill_how == KILL_THREAD ? &prog_thread 844 : &prog_process)); 845 846 /* 847 * Add the KILL_THREAD rule again to make sure that the KILL_PROCESS 848 * flag cannot be downgraded by a new filter. 849 */ 850 if (kill_how == KILL_PROCESS) 851 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog_thread)); 852 853 /* Start a thread that will exit immediately. */ 854 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)false)); 855 ASSERT_EQ(0, pthread_join(thread, &status)); 856 ASSERT_EQ(SIBLING_EXIT_UNKILLED, (unsigned long)status); 857 858 /* Start a thread that will die immediately. */ 859 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)true)); 860 ASSERT_EQ(0, pthread_join(thread, &status)); 861 ASSERT_NE(SIBLING_EXIT_FAILURE, (unsigned long)status); 862 863 /* 864 * If we get here, only the spawned thread died. Let the parent know 865 * the whole process didn't die (i.e. this thread, the spawner, 866 * stayed running). 867 */ 868 exit(42); 869 } 870 871 TEST(KILL_thread) 872 { 873 int status; 874 pid_t child_pid; 875 876 child_pid = fork(); 877 ASSERT_LE(0, child_pid); 878 if (child_pid == 0) { 879 kill_thread_or_group(_metadata, KILL_THREAD); 880 _exit(38); 881 } 882 883 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 884 885 /* If only the thread was killed, we'll see exit 42. */ 886 ASSERT_TRUE(WIFEXITED(status)); 887 ASSERT_EQ(42, WEXITSTATUS(status)); 888 } 889 890 TEST(KILL_process) 891 { 892 int status; 893 pid_t child_pid; 894 895 child_pid = fork(); 896 ASSERT_LE(0, child_pid); 897 if (child_pid == 0) { 898 kill_thread_or_group(_metadata, KILL_PROCESS); 899 _exit(38); 900 } 901 902 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 903 904 /* If the entire process was killed, we'll see SIGSYS. */ 905 ASSERT_TRUE(WIFSIGNALED(status)); 906 ASSERT_EQ(SIGSYS, WTERMSIG(status)); 907 } 908 909 TEST(KILL_unknown) 910 { 911 int status; 912 pid_t child_pid; 913 914 child_pid = fork(); 915 ASSERT_LE(0, child_pid); 916 if (child_pid == 0) { 917 kill_thread_or_group(_metadata, RET_UNKNOWN); 918 _exit(38); 919 } 920 921 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 922 923 /* If the entire process was killed, we'll see SIGSYS. */ 924 EXPECT_TRUE(WIFSIGNALED(status)) { 925 TH_LOG("Unknown SECCOMP_RET is only killing the thread?"); 926 } 927 ASSERT_EQ(SIGSYS, WTERMSIG(status)); 928 } 929 930 /* TODO(wad) add 64-bit versus 32-bit arg tests. */ 931 TEST(arg_out_of_range) 932 { 933 struct sock_filter filter[] = { 934 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)), 935 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 936 }; 937 struct sock_fprog prog = { 938 .len = (unsigned short)ARRAY_SIZE(filter), 939 .filter = filter, 940 }; 941 long ret; 942 943 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 944 ASSERT_EQ(0, ret); 945 946 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 947 EXPECT_EQ(-1, ret); 948 EXPECT_EQ(EINVAL, errno); 949 } 950 951 #define ERRNO_FILTER(name, errno) \ 952 struct sock_filter _read_filter_##name[] = { \ 953 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, \ 954 offsetof(struct seccomp_data, nr)), \ 955 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), \ 956 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | errno), \ 957 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), \ 958 }; \ 959 struct sock_fprog prog_##name = { \ 960 .len = (unsigned short)ARRAY_SIZE(_read_filter_##name), \ 961 .filter = _read_filter_##name, \ 962 } 963 964 /* Make sure basic errno values are correctly passed through a filter. */ 965 TEST(ERRNO_valid) 966 { 967 ERRNO_FILTER(valid, E2BIG); 968 long ret; 969 pid_t parent = getppid(); 970 971 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 972 ASSERT_EQ(0, ret); 973 974 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_valid); 975 ASSERT_EQ(0, ret); 976 977 EXPECT_EQ(parent, syscall(__NR_getppid)); 978 EXPECT_EQ(-1, read(-1, NULL, 0)); 979 EXPECT_EQ(E2BIG, errno); 980 } 981 982 /* Make sure an errno of zero is correctly handled by the arch code. */ 983 TEST(ERRNO_zero) 984 { 985 ERRNO_FILTER(zero, 0); 986 long ret; 987 pid_t parent = getppid(); 988 989 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 990 ASSERT_EQ(0, ret); 991 992 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_zero); 993 ASSERT_EQ(0, ret); 994 995 EXPECT_EQ(parent, syscall(__NR_getppid)); 996 /* "errno" of 0 is ok. */ 997 EXPECT_EQ(0, read(-1, NULL, 0)); 998 } 999 1000 /* 1001 * The SECCOMP_RET_DATA mask is 16 bits wide, but errno is smaller. 1002 * This tests that the errno value gets capped correctly, fixed by 1003 * 580c57f10768 ("seccomp: cap SECCOMP_RET_ERRNO data to MAX_ERRNO"). 1004 */ 1005 TEST(ERRNO_capped) 1006 { 1007 ERRNO_FILTER(capped, 4096); 1008 long ret; 1009 pid_t parent = getppid(); 1010 1011 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1012 ASSERT_EQ(0, ret); 1013 1014 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_capped); 1015 ASSERT_EQ(0, ret); 1016 1017 EXPECT_EQ(parent, syscall(__NR_getppid)); 1018 EXPECT_EQ(-1, read(-1, NULL, 0)); 1019 EXPECT_EQ(4095, errno); 1020 } 1021 1022 /* 1023 * Filters are processed in reverse order: last applied is executed first. 1024 * Since only the SECCOMP_RET_ACTION mask is tested for return values, the 1025 * SECCOMP_RET_DATA mask results will follow the most recently applied 1026 * matching filter return (and not the lowest or highest value). 1027 */ 1028 TEST(ERRNO_order) 1029 { 1030 ERRNO_FILTER(first, 11); 1031 ERRNO_FILTER(second, 13); 1032 ERRNO_FILTER(third, 12); 1033 long ret; 1034 pid_t parent = getppid(); 1035 1036 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1037 ASSERT_EQ(0, ret); 1038 1039 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_first); 1040 ASSERT_EQ(0, ret); 1041 1042 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_second); 1043 ASSERT_EQ(0, ret); 1044 1045 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_third); 1046 ASSERT_EQ(0, ret); 1047 1048 EXPECT_EQ(parent, syscall(__NR_getppid)); 1049 EXPECT_EQ(-1, read(-1, NULL, 0)); 1050 EXPECT_EQ(12, errno); 1051 } 1052 1053 FIXTURE(TRAP) { 1054 struct sock_fprog prog; 1055 }; 1056 1057 FIXTURE_SETUP(TRAP) 1058 { 1059 struct sock_filter filter[] = { 1060 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1061 offsetof(struct seccomp_data, nr)), 1062 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 1063 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP), 1064 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1065 }; 1066 1067 memset(&self->prog, 0, sizeof(self->prog)); 1068 self->prog.filter = malloc(sizeof(filter)); 1069 ASSERT_NE(NULL, self->prog.filter); 1070 memcpy(self->prog.filter, filter, sizeof(filter)); 1071 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 1072 } 1073 1074 FIXTURE_TEARDOWN(TRAP) 1075 { 1076 if (self->prog.filter) 1077 free(self->prog.filter); 1078 } 1079 1080 TEST_F_SIGNAL(TRAP, dfl, SIGSYS) 1081 { 1082 long ret; 1083 1084 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1085 ASSERT_EQ(0, ret); 1086 1087 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 1088 ASSERT_EQ(0, ret); 1089 syscall(__NR_getpid); 1090 } 1091 1092 /* Ensure that SIGSYS overrides SIG_IGN */ 1093 TEST_F_SIGNAL(TRAP, ign, SIGSYS) 1094 { 1095 long ret; 1096 1097 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1098 ASSERT_EQ(0, ret); 1099 1100 signal(SIGSYS, SIG_IGN); 1101 1102 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 1103 ASSERT_EQ(0, ret); 1104 syscall(__NR_getpid); 1105 } 1106 1107 static siginfo_t TRAP_info; 1108 static volatile int TRAP_nr; 1109 static void TRAP_action(int nr, siginfo_t *info, void *void_context) 1110 { 1111 memcpy(&TRAP_info, info, sizeof(TRAP_info)); 1112 TRAP_nr = nr; 1113 } 1114 1115 TEST_F(TRAP, handler) 1116 { 1117 int ret, test; 1118 struct sigaction act; 1119 sigset_t mask; 1120 1121 memset(&act, 0, sizeof(act)); 1122 sigemptyset(&mask); 1123 sigaddset(&mask, SIGSYS); 1124 1125 act.sa_sigaction = &TRAP_action; 1126 act.sa_flags = SA_SIGINFO; 1127 ret = sigaction(SIGSYS, &act, NULL); 1128 ASSERT_EQ(0, ret) { 1129 TH_LOG("sigaction failed"); 1130 } 1131 ret = sigprocmask(SIG_UNBLOCK, &mask, NULL); 1132 ASSERT_EQ(0, ret) { 1133 TH_LOG("sigprocmask failed"); 1134 } 1135 1136 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1137 ASSERT_EQ(0, ret); 1138 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 1139 ASSERT_EQ(0, ret); 1140 TRAP_nr = 0; 1141 memset(&TRAP_info, 0, sizeof(TRAP_info)); 1142 /* Expect the registers to be rolled back. (nr = error) may vary 1143 * based on arch. */ 1144 ret = syscall(__NR_getpid); 1145 /* Silence gcc warning about volatile. */ 1146 test = TRAP_nr; 1147 EXPECT_EQ(SIGSYS, test); 1148 struct local_sigsys { 1149 void *_call_addr; /* calling user insn */ 1150 int _syscall; /* triggering system call number */ 1151 unsigned int _arch; /* AUDIT_ARCH_* of syscall */ 1152 } *sigsys = (struct local_sigsys *) 1153 #ifdef si_syscall 1154 &(TRAP_info.si_call_addr); 1155 #else 1156 &TRAP_info.si_pid; 1157 #endif 1158 EXPECT_EQ(__NR_getpid, sigsys->_syscall); 1159 /* Make sure arch is non-zero. */ 1160 EXPECT_NE(0, sigsys->_arch); 1161 EXPECT_NE(0, (unsigned long)sigsys->_call_addr); 1162 } 1163 1164 FIXTURE(precedence) { 1165 struct sock_fprog allow; 1166 struct sock_fprog log; 1167 struct sock_fprog trace; 1168 struct sock_fprog error; 1169 struct sock_fprog trap; 1170 struct sock_fprog kill; 1171 }; 1172 1173 FIXTURE_SETUP(precedence) 1174 { 1175 struct sock_filter allow_insns[] = { 1176 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1177 }; 1178 struct sock_filter log_insns[] = { 1179 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1180 offsetof(struct seccomp_data, nr)), 1181 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1182 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1183 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG), 1184 }; 1185 struct sock_filter trace_insns[] = { 1186 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1187 offsetof(struct seccomp_data, nr)), 1188 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1189 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1190 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE), 1191 }; 1192 struct sock_filter error_insns[] = { 1193 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1194 offsetof(struct seccomp_data, nr)), 1195 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1196 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1197 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO), 1198 }; 1199 struct sock_filter trap_insns[] = { 1200 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1201 offsetof(struct seccomp_data, nr)), 1202 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1203 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1204 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP), 1205 }; 1206 struct sock_filter kill_insns[] = { 1207 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1208 offsetof(struct seccomp_data, nr)), 1209 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1210 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1211 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 1212 }; 1213 1214 memset(self, 0, sizeof(*self)); 1215 #define FILTER_ALLOC(_x) \ 1216 self->_x.filter = malloc(sizeof(_x##_insns)); \ 1217 ASSERT_NE(NULL, self->_x.filter); \ 1218 memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \ 1219 self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns) 1220 FILTER_ALLOC(allow); 1221 FILTER_ALLOC(log); 1222 FILTER_ALLOC(trace); 1223 FILTER_ALLOC(error); 1224 FILTER_ALLOC(trap); 1225 FILTER_ALLOC(kill); 1226 } 1227 1228 FIXTURE_TEARDOWN(precedence) 1229 { 1230 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter) 1231 FILTER_FREE(allow); 1232 FILTER_FREE(log); 1233 FILTER_FREE(trace); 1234 FILTER_FREE(error); 1235 FILTER_FREE(trap); 1236 FILTER_FREE(kill); 1237 } 1238 1239 TEST_F(precedence, allow_ok) 1240 { 1241 pid_t parent, res = 0; 1242 long ret; 1243 1244 parent = getppid(); 1245 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1246 ASSERT_EQ(0, ret); 1247 1248 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 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->error); 1255 ASSERT_EQ(0, ret); 1256 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1257 ASSERT_EQ(0, ret); 1258 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1259 ASSERT_EQ(0, ret); 1260 /* Should work just fine. */ 1261 res = syscall(__NR_getppid); 1262 EXPECT_EQ(parent, res); 1263 } 1264 1265 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS) 1266 { 1267 pid_t parent, res = 0; 1268 long ret; 1269 1270 parent = getppid(); 1271 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1272 ASSERT_EQ(0, ret); 1273 1274 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1275 ASSERT_EQ(0, ret); 1276 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1277 ASSERT_EQ(0, ret); 1278 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1279 ASSERT_EQ(0, ret); 1280 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1281 ASSERT_EQ(0, ret); 1282 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1283 ASSERT_EQ(0, ret); 1284 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1285 ASSERT_EQ(0, ret); 1286 /* Should work just fine. */ 1287 res = syscall(__NR_getppid); 1288 EXPECT_EQ(parent, res); 1289 /* getpid() should never return. */ 1290 res = syscall(__NR_getpid); 1291 EXPECT_EQ(0, res); 1292 } 1293 1294 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS) 1295 { 1296 pid_t parent; 1297 long ret; 1298 1299 parent = getppid(); 1300 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1301 ASSERT_EQ(0, ret); 1302 1303 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1304 ASSERT_EQ(0, ret); 1305 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1306 ASSERT_EQ(0, ret); 1307 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1308 ASSERT_EQ(0, ret); 1309 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1310 ASSERT_EQ(0, ret); 1311 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1312 ASSERT_EQ(0, ret); 1313 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1314 ASSERT_EQ(0, ret); 1315 /* Should work just fine. */ 1316 EXPECT_EQ(parent, syscall(__NR_getppid)); 1317 /* getpid() should never return. */ 1318 EXPECT_EQ(0, syscall(__NR_getpid)); 1319 } 1320 1321 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS) 1322 { 1323 pid_t parent; 1324 long ret; 1325 1326 parent = getppid(); 1327 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1328 ASSERT_EQ(0, ret); 1329 1330 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1331 ASSERT_EQ(0, ret); 1332 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1333 ASSERT_EQ(0, ret); 1334 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1335 ASSERT_EQ(0, ret); 1336 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1337 ASSERT_EQ(0, ret); 1338 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1339 ASSERT_EQ(0, ret); 1340 /* Should work just fine. */ 1341 EXPECT_EQ(parent, syscall(__NR_getppid)); 1342 /* getpid() should never return. */ 1343 EXPECT_EQ(0, syscall(__NR_getpid)); 1344 } 1345 1346 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS) 1347 { 1348 pid_t parent; 1349 long ret; 1350 1351 parent = getppid(); 1352 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1353 ASSERT_EQ(0, ret); 1354 1355 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1356 ASSERT_EQ(0, ret); 1357 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1358 ASSERT_EQ(0, ret); 1359 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1360 ASSERT_EQ(0, ret); 1361 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1362 ASSERT_EQ(0, ret); 1363 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1364 ASSERT_EQ(0, ret); 1365 /* Should work just fine. */ 1366 EXPECT_EQ(parent, syscall(__NR_getppid)); 1367 /* getpid() should never return. */ 1368 EXPECT_EQ(0, syscall(__NR_getpid)); 1369 } 1370 1371 TEST_F(precedence, errno_is_third) 1372 { 1373 pid_t parent; 1374 long ret; 1375 1376 parent = getppid(); 1377 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1378 ASSERT_EQ(0, ret); 1379 1380 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1381 ASSERT_EQ(0, ret); 1382 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1383 ASSERT_EQ(0, ret); 1384 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1385 ASSERT_EQ(0, ret); 1386 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1387 ASSERT_EQ(0, ret); 1388 /* Should work just fine. */ 1389 EXPECT_EQ(parent, syscall(__NR_getppid)); 1390 EXPECT_EQ(0, syscall(__NR_getpid)); 1391 } 1392 1393 TEST_F(precedence, errno_is_third_in_any_order) 1394 { 1395 pid_t parent; 1396 long ret; 1397 1398 parent = getppid(); 1399 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1400 ASSERT_EQ(0, ret); 1401 1402 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1403 ASSERT_EQ(0, ret); 1404 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1405 ASSERT_EQ(0, ret); 1406 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1407 ASSERT_EQ(0, ret); 1408 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1409 ASSERT_EQ(0, ret); 1410 /* Should work just fine. */ 1411 EXPECT_EQ(parent, syscall(__NR_getppid)); 1412 EXPECT_EQ(0, syscall(__NR_getpid)); 1413 } 1414 1415 TEST_F(precedence, trace_is_fourth) 1416 { 1417 pid_t parent; 1418 long ret; 1419 1420 parent = getppid(); 1421 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1422 ASSERT_EQ(0, ret); 1423 1424 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1425 ASSERT_EQ(0, ret); 1426 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1427 ASSERT_EQ(0, ret); 1428 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1429 ASSERT_EQ(0, ret); 1430 /* Should work just fine. */ 1431 EXPECT_EQ(parent, syscall(__NR_getppid)); 1432 /* No ptracer */ 1433 EXPECT_EQ(-1, syscall(__NR_getpid)); 1434 } 1435 1436 TEST_F(precedence, trace_is_fourth_in_any_order) 1437 { 1438 pid_t parent; 1439 long ret; 1440 1441 parent = getppid(); 1442 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1443 ASSERT_EQ(0, ret); 1444 1445 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1446 ASSERT_EQ(0, ret); 1447 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1448 ASSERT_EQ(0, ret); 1449 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1450 ASSERT_EQ(0, ret); 1451 /* Should work just fine. */ 1452 EXPECT_EQ(parent, syscall(__NR_getppid)); 1453 /* No ptracer */ 1454 EXPECT_EQ(-1, syscall(__NR_getpid)); 1455 } 1456 1457 TEST_F(precedence, log_is_fifth) 1458 { 1459 pid_t mypid, parent; 1460 long ret; 1461 1462 mypid = getpid(); 1463 parent = getppid(); 1464 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1465 ASSERT_EQ(0, ret); 1466 1467 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1468 ASSERT_EQ(0, ret); 1469 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1470 ASSERT_EQ(0, ret); 1471 /* Should work just fine. */ 1472 EXPECT_EQ(parent, syscall(__NR_getppid)); 1473 /* Should also work just fine */ 1474 EXPECT_EQ(mypid, syscall(__NR_getpid)); 1475 } 1476 1477 TEST_F(precedence, log_is_fifth_in_any_order) 1478 { 1479 pid_t mypid, parent; 1480 long ret; 1481 1482 mypid = getpid(); 1483 parent = getppid(); 1484 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1485 ASSERT_EQ(0, ret); 1486 1487 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1488 ASSERT_EQ(0, ret); 1489 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1490 ASSERT_EQ(0, ret); 1491 /* Should work just fine. */ 1492 EXPECT_EQ(parent, syscall(__NR_getppid)); 1493 /* Should also work just fine */ 1494 EXPECT_EQ(mypid, syscall(__NR_getpid)); 1495 } 1496 1497 #ifndef PTRACE_O_TRACESECCOMP 1498 #define PTRACE_O_TRACESECCOMP 0x00000080 1499 #endif 1500 1501 /* Catch the Ubuntu 12.04 value error. */ 1502 #if PTRACE_EVENT_SECCOMP != 7 1503 #undef PTRACE_EVENT_SECCOMP 1504 #endif 1505 1506 #ifndef PTRACE_EVENT_SECCOMP 1507 #define PTRACE_EVENT_SECCOMP 7 1508 #endif 1509 1510 #define PTRACE_EVENT_MASK(status) ((status) >> 16) 1511 bool tracer_running; 1512 void tracer_stop(int sig) 1513 { 1514 tracer_running = false; 1515 } 1516 1517 typedef void tracer_func_t(struct __test_metadata *_metadata, 1518 pid_t tracee, int status, void *args); 1519 1520 void start_tracer(struct __test_metadata *_metadata, int fd, pid_t tracee, 1521 tracer_func_t tracer_func, void *args, bool ptrace_syscall) 1522 { 1523 int ret = -1; 1524 struct sigaction action = { 1525 .sa_handler = tracer_stop, 1526 }; 1527 1528 /* Allow external shutdown. */ 1529 tracer_running = true; 1530 ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL)); 1531 1532 errno = 0; 1533 while (ret == -1 && errno != EINVAL) 1534 ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0); 1535 ASSERT_EQ(0, ret) { 1536 kill(tracee, SIGKILL); 1537 } 1538 /* Wait for attach stop */ 1539 wait(NULL); 1540 1541 ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, ptrace_syscall ? 1542 PTRACE_O_TRACESYSGOOD : 1543 PTRACE_O_TRACESECCOMP); 1544 ASSERT_EQ(0, ret) { 1545 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP"); 1546 kill(tracee, SIGKILL); 1547 } 1548 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT, 1549 tracee, NULL, 0); 1550 ASSERT_EQ(0, ret); 1551 1552 /* Unblock the tracee */ 1553 ASSERT_EQ(1, write(fd, "A", 1)); 1554 ASSERT_EQ(0, close(fd)); 1555 1556 /* Run until we're shut down. Must assert to stop execution. */ 1557 while (tracer_running) { 1558 int status; 1559 1560 if (wait(&status) != tracee) 1561 continue; 1562 1563 if (WIFSIGNALED(status)) { 1564 /* Child caught a fatal signal. */ 1565 return; 1566 } 1567 if (WIFEXITED(status)) { 1568 /* Child exited with code. */ 1569 return; 1570 } 1571 1572 /* Check if we got an expected event. */ 1573 ASSERT_EQ(WIFCONTINUED(status), false); 1574 ASSERT_EQ(WIFSTOPPED(status), true); 1575 ASSERT_EQ(WSTOPSIG(status) & SIGTRAP, SIGTRAP) { 1576 TH_LOG("Unexpected WSTOPSIG: %d", WSTOPSIG(status)); 1577 } 1578 1579 tracer_func(_metadata, tracee, status, args); 1580 1581 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT, 1582 tracee, NULL, 0); 1583 ASSERT_EQ(0, ret); 1584 } 1585 /* Directly report the status of our test harness results. */ 1586 syscall(__NR_exit, _metadata->exit_code); 1587 } 1588 1589 /* Common tracer setup/teardown functions. */ 1590 void cont_handler(int num) 1591 { } 1592 pid_t setup_trace_fixture(struct __test_metadata *_metadata, 1593 tracer_func_t func, void *args, bool ptrace_syscall) 1594 { 1595 char sync; 1596 int pipefd[2]; 1597 pid_t tracer_pid; 1598 pid_t tracee = getpid(); 1599 1600 /* Setup a pipe for clean synchronization. */ 1601 ASSERT_EQ(0, pipe(pipefd)); 1602 1603 /* Fork a child which we'll promote to tracer */ 1604 tracer_pid = fork(); 1605 ASSERT_LE(0, tracer_pid); 1606 signal(SIGALRM, cont_handler); 1607 if (tracer_pid == 0) { 1608 close(pipefd[0]); 1609 start_tracer(_metadata, pipefd[1], tracee, func, args, 1610 ptrace_syscall); 1611 syscall(__NR_exit, 0); 1612 } 1613 close(pipefd[1]); 1614 prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0); 1615 read(pipefd[0], &sync, 1); 1616 close(pipefd[0]); 1617 1618 return tracer_pid; 1619 } 1620 1621 void teardown_trace_fixture(struct __test_metadata *_metadata, 1622 pid_t tracer) 1623 { 1624 if (tracer) { 1625 int status; 1626 /* 1627 * Extract the exit code from the other process and 1628 * adopt it for ourselves in case its asserts failed. 1629 */ 1630 ASSERT_EQ(0, kill(tracer, SIGUSR1)); 1631 ASSERT_EQ(tracer, waitpid(tracer, &status, 0)); 1632 if (WEXITSTATUS(status)) 1633 _metadata->exit_code = KSFT_FAIL; 1634 } 1635 } 1636 1637 /* "poke" tracer arguments and function. */ 1638 struct tracer_args_poke_t { 1639 unsigned long poke_addr; 1640 }; 1641 1642 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status, 1643 void *args) 1644 { 1645 int ret; 1646 unsigned long msg; 1647 struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args; 1648 1649 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 1650 EXPECT_EQ(0, ret); 1651 /* If this fails, don't try to recover. */ 1652 ASSERT_EQ(0x1001, msg) { 1653 kill(tracee, SIGKILL); 1654 } 1655 /* 1656 * Poke in the message. 1657 * Registers are not touched to try to keep this relatively arch 1658 * agnostic. 1659 */ 1660 ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001); 1661 EXPECT_EQ(0, ret); 1662 } 1663 1664 FIXTURE(TRACE_poke) { 1665 struct sock_fprog prog; 1666 pid_t tracer; 1667 long poked; 1668 struct tracer_args_poke_t tracer_args; 1669 }; 1670 1671 FIXTURE_SETUP(TRACE_poke) 1672 { 1673 struct sock_filter filter[] = { 1674 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1675 offsetof(struct seccomp_data, nr)), 1676 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 1677 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001), 1678 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1679 }; 1680 1681 self->poked = 0; 1682 memset(&self->prog, 0, sizeof(self->prog)); 1683 self->prog.filter = malloc(sizeof(filter)); 1684 ASSERT_NE(NULL, self->prog.filter); 1685 memcpy(self->prog.filter, filter, sizeof(filter)); 1686 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 1687 1688 /* Set up tracer args. */ 1689 self->tracer_args.poke_addr = (unsigned long)&self->poked; 1690 1691 /* Launch tracer. */ 1692 self->tracer = setup_trace_fixture(_metadata, tracer_poke, 1693 &self->tracer_args, false); 1694 } 1695 1696 FIXTURE_TEARDOWN(TRACE_poke) 1697 { 1698 teardown_trace_fixture(_metadata, self->tracer); 1699 if (self->prog.filter) 1700 free(self->prog.filter); 1701 } 1702 1703 TEST_F(TRACE_poke, read_has_side_effects) 1704 { 1705 ssize_t ret; 1706 1707 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1708 ASSERT_EQ(0, ret); 1709 1710 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1711 ASSERT_EQ(0, ret); 1712 1713 EXPECT_EQ(0, self->poked); 1714 ret = read(-1, NULL, 0); 1715 EXPECT_EQ(-1, ret); 1716 EXPECT_EQ(0x1001, self->poked); 1717 } 1718 1719 TEST_F(TRACE_poke, getpid_runs_normally) 1720 { 1721 long ret; 1722 1723 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1724 ASSERT_EQ(0, ret); 1725 1726 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1727 ASSERT_EQ(0, ret); 1728 1729 EXPECT_EQ(0, self->poked); 1730 EXPECT_NE(0, syscall(__NR_getpid)); 1731 EXPECT_EQ(0, self->poked); 1732 } 1733 1734 #if defined(__x86_64__) 1735 # define ARCH_REGS struct user_regs_struct 1736 # define SYSCALL_NUM(_regs) (_regs).orig_rax 1737 # define SYSCALL_RET(_regs) (_regs).rax 1738 #elif defined(__i386__) 1739 # define ARCH_REGS struct user_regs_struct 1740 # define SYSCALL_NUM(_regs) (_regs).orig_eax 1741 # define SYSCALL_RET(_regs) (_regs).eax 1742 #elif defined(__arm__) 1743 # define ARCH_REGS struct pt_regs 1744 # define SYSCALL_NUM(_regs) (_regs).ARM_r7 1745 # ifndef PTRACE_SET_SYSCALL 1746 # define PTRACE_SET_SYSCALL 23 1747 # endif 1748 # define SYSCALL_NUM_SET(_regs, _nr) \ 1749 EXPECT_EQ(0, ptrace(PTRACE_SET_SYSCALL, tracee, NULL, _nr)) 1750 # define SYSCALL_RET(_regs) (_regs).ARM_r0 1751 #elif defined(__aarch64__) 1752 # define ARCH_REGS struct user_pt_regs 1753 # define SYSCALL_NUM(_regs) (_regs).regs[8] 1754 # ifndef NT_ARM_SYSTEM_CALL 1755 # define NT_ARM_SYSTEM_CALL 0x404 1756 # endif 1757 # define SYSCALL_NUM_SET(_regs, _nr) \ 1758 do { \ 1759 struct iovec __v; \ 1760 typeof(_nr) __nr = (_nr); \ 1761 __v.iov_base = &__nr; \ 1762 __v.iov_len = sizeof(__nr); \ 1763 EXPECT_EQ(0, ptrace(PTRACE_SETREGSET, tracee, \ 1764 NT_ARM_SYSTEM_CALL, &__v)); \ 1765 } while (0) 1766 # define SYSCALL_RET(_regs) (_regs).regs[0] 1767 #elif defined(__loongarch__) 1768 # define ARCH_REGS struct user_pt_regs 1769 # define SYSCALL_NUM(_regs) (_regs).regs[11] 1770 # define SYSCALL_RET(_regs) (_regs).regs[4] 1771 #elif defined(__riscv) && __riscv_xlen == 64 1772 # define ARCH_REGS struct user_regs_struct 1773 # define SYSCALL_NUM(_regs) (_regs).a7 1774 # define SYSCALL_RET(_regs) (_regs).a0 1775 #elif defined(__csky__) 1776 # define ARCH_REGS struct pt_regs 1777 # if defined(__CSKYABIV2__) 1778 # define SYSCALL_NUM(_regs) (_regs).regs[3] 1779 # else 1780 # define SYSCALL_NUM(_regs) (_regs).regs[9] 1781 # endif 1782 # define SYSCALL_RET(_regs) (_regs).a0 1783 #elif defined(__hppa__) 1784 # define ARCH_REGS struct user_regs_struct 1785 # define SYSCALL_NUM(_regs) (_regs).gr[20] 1786 # define SYSCALL_RET(_regs) (_regs).gr[28] 1787 #elif defined(__powerpc__) 1788 # define ARCH_REGS struct pt_regs 1789 # define SYSCALL_NUM(_regs) (_regs).gpr[0] 1790 # define SYSCALL_RET(_regs) (_regs).gpr[3] 1791 # define SYSCALL_RET_SET(_regs, _val) \ 1792 do { \ 1793 typeof(_val) _result = (_val); \ 1794 if ((_regs.trap & 0xfff0) == 0x3000) { \ 1795 /* \ 1796 * scv 0 system call uses -ve result \ 1797 * for error, so no need to adjust. \ 1798 */ \ 1799 SYSCALL_RET(_regs) = _result; \ 1800 } else { \ 1801 /* \ 1802 * A syscall error is signaled by the \ 1803 * CR0 SO bit and the code is stored as \ 1804 * a positive value. \ 1805 */ \ 1806 if (_result < 0) { \ 1807 SYSCALL_RET(_regs) = -_result; \ 1808 (_regs).ccr |= 0x10000000; \ 1809 } else { \ 1810 SYSCALL_RET(_regs) = _result; \ 1811 (_regs).ccr &= ~0x10000000; \ 1812 } \ 1813 } \ 1814 } while (0) 1815 # define SYSCALL_RET_SET_ON_PTRACE_EXIT 1816 #elif defined(__s390__) 1817 # define ARCH_REGS s390_regs 1818 # define SYSCALL_NUM(_regs) (_regs).gprs[2] 1819 # define SYSCALL_RET_SET(_regs, _val) \ 1820 TH_LOG("Can't modify syscall return on this architecture") 1821 #elif defined(__mips__) 1822 # include <asm/unistd_nr_n32.h> 1823 # include <asm/unistd_nr_n64.h> 1824 # include <asm/unistd_nr_o32.h> 1825 # define ARCH_REGS struct pt_regs 1826 # define SYSCALL_NUM(_regs) \ 1827 ({ \ 1828 typeof((_regs).regs[2]) _nr; \ 1829 if ((_regs).regs[2] == __NR_O32_Linux) \ 1830 _nr = (_regs).regs[4]; \ 1831 else \ 1832 _nr = (_regs).regs[2]; \ 1833 _nr; \ 1834 }) 1835 # define SYSCALL_NUM_SET(_regs, _nr) \ 1836 do { \ 1837 if ((_regs).regs[2] == __NR_O32_Linux) \ 1838 (_regs).regs[4] = _nr; \ 1839 else \ 1840 (_regs).regs[2] = _nr; \ 1841 } while (0) 1842 # define SYSCALL_RET_SET(_regs, _val) \ 1843 TH_LOG("Can't modify syscall return on this architecture") 1844 #elif defined(__xtensa__) 1845 # define ARCH_REGS struct user_pt_regs 1846 # define SYSCALL_NUM(_regs) (_regs).syscall 1847 /* 1848 * On xtensa syscall return value is in the register 1849 * a2 of the current window which is not fixed. 1850 */ 1851 #define SYSCALL_RET(_regs) (_regs).a[(_regs).windowbase * 4 + 2] 1852 #elif defined(__sh__) 1853 # define ARCH_REGS struct pt_regs 1854 # define SYSCALL_NUM(_regs) (_regs).regs[3] 1855 # define SYSCALL_RET(_regs) (_regs).regs[0] 1856 #elif defined(__mc68000__) 1857 # define ARCH_REGS struct user_regs_struct 1858 # define SYSCALL_NUM(_regs) (_regs).orig_d0 1859 # define SYSCALL_RET(_regs) (_regs).d0 1860 #else 1861 # error "Do not know how to find your architecture's registers and syscalls" 1862 #endif 1863 1864 /* 1865 * Most architectures can change the syscall by just updating the 1866 * associated register. This is the default if not defined above. 1867 */ 1868 #ifndef SYSCALL_NUM_SET 1869 # define SYSCALL_NUM_SET(_regs, _nr) \ 1870 do { \ 1871 SYSCALL_NUM(_regs) = (_nr); \ 1872 } while (0) 1873 #endif 1874 /* 1875 * Most architectures can change the syscall return value by just 1876 * writing to the SYSCALL_RET register. This is the default if not 1877 * defined above. If an architecture cannot set the return value 1878 * (for example when the syscall and return value register is 1879 * shared), report it with TH_LOG() in an arch-specific definition 1880 * of SYSCALL_RET_SET() above, and leave SYSCALL_RET undefined. 1881 */ 1882 #if !defined(SYSCALL_RET) && !defined(SYSCALL_RET_SET) 1883 # error "One of SYSCALL_RET or SYSCALL_RET_SET is needed for this arch" 1884 #endif 1885 #ifndef SYSCALL_RET_SET 1886 # define SYSCALL_RET_SET(_regs, _val) \ 1887 do { \ 1888 SYSCALL_RET(_regs) = (_val); \ 1889 } while (0) 1890 #endif 1891 1892 /* When the syscall return can't be changed, stub out the tests for it. */ 1893 #ifndef SYSCALL_RET 1894 # define EXPECT_SYSCALL_RETURN(val, action) EXPECT_EQ(-1, action) 1895 #else 1896 # define EXPECT_SYSCALL_RETURN(val, action) \ 1897 do { \ 1898 errno = 0; \ 1899 if (val < 0) { \ 1900 EXPECT_EQ(-1, action); \ 1901 EXPECT_EQ(-(val), errno); \ 1902 } else { \ 1903 EXPECT_EQ(val, action); \ 1904 } \ 1905 } while (0) 1906 #endif 1907 1908 /* 1909 * Some architectures (e.g. powerpc) can only set syscall 1910 * return values on syscall exit during ptrace. 1911 */ 1912 const bool ptrace_entry_set_syscall_nr = true; 1913 const bool ptrace_entry_set_syscall_ret = 1914 #ifndef SYSCALL_RET_SET_ON_PTRACE_EXIT 1915 true; 1916 #else 1917 false; 1918 #endif 1919 1920 /* 1921 * Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for 1922 * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux). 1923 */ 1924 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__) || defined(__mc68000__) 1925 # define ARCH_GETREGS(_regs) ptrace(PTRACE_GETREGS, tracee, 0, &(_regs)) 1926 # define ARCH_SETREGS(_regs) ptrace(PTRACE_SETREGS, tracee, 0, &(_regs)) 1927 #else 1928 # define ARCH_GETREGS(_regs) ({ \ 1929 struct iovec __v; \ 1930 __v.iov_base = &(_regs); \ 1931 __v.iov_len = sizeof(_regs); \ 1932 ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &__v); \ 1933 }) 1934 # define ARCH_SETREGS(_regs) ({ \ 1935 struct iovec __v; \ 1936 __v.iov_base = &(_regs); \ 1937 __v.iov_len = sizeof(_regs); \ 1938 ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &__v); \ 1939 }) 1940 #endif 1941 1942 /* Architecture-specific syscall fetching routine. */ 1943 int get_syscall(struct __test_metadata *_metadata, pid_t tracee) 1944 { 1945 ARCH_REGS regs; 1946 1947 EXPECT_EQ(0, ARCH_GETREGS(regs)) { 1948 return -1; 1949 } 1950 1951 return SYSCALL_NUM(regs); 1952 } 1953 1954 /* Architecture-specific syscall changing routine. */ 1955 void __change_syscall(struct __test_metadata *_metadata, 1956 pid_t tracee, long *syscall, long *ret) 1957 { 1958 ARCH_REGS orig, regs; 1959 1960 /* Do not get/set registers if we have nothing to do. */ 1961 if (!syscall && !ret) 1962 return; 1963 1964 EXPECT_EQ(0, ARCH_GETREGS(regs)) { 1965 return; 1966 } 1967 orig = regs; 1968 1969 if (syscall) 1970 SYSCALL_NUM_SET(regs, *syscall); 1971 1972 if (ret) 1973 SYSCALL_RET_SET(regs, *ret); 1974 1975 /* Flush any register changes made. */ 1976 if (memcmp(&orig, ®s, sizeof(orig)) != 0) 1977 EXPECT_EQ(0, ARCH_SETREGS(regs)); 1978 } 1979 1980 /* Change only syscall number. */ 1981 void change_syscall_nr(struct __test_metadata *_metadata, 1982 pid_t tracee, long syscall) 1983 { 1984 __change_syscall(_metadata, tracee, &syscall, NULL); 1985 } 1986 1987 /* Change syscall return value (and set syscall number to -1). */ 1988 void change_syscall_ret(struct __test_metadata *_metadata, 1989 pid_t tracee, long ret) 1990 { 1991 long syscall = -1; 1992 1993 __change_syscall(_metadata, tracee, &syscall, &ret); 1994 } 1995 1996 void tracer_seccomp(struct __test_metadata *_metadata, pid_t tracee, 1997 int status, void *args) 1998 { 1999 int ret; 2000 unsigned long msg; 2001 2002 EXPECT_EQ(PTRACE_EVENT_MASK(status), PTRACE_EVENT_SECCOMP) { 2003 TH_LOG("Unexpected ptrace event: %d", PTRACE_EVENT_MASK(status)); 2004 return; 2005 } 2006 2007 /* Make sure we got the right message. */ 2008 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 2009 EXPECT_EQ(0, ret); 2010 2011 /* Validate and take action on expected syscalls. */ 2012 switch (msg) { 2013 case 0x1002: 2014 /* change getpid to getppid. */ 2015 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee)); 2016 change_syscall_nr(_metadata, tracee, __NR_getppid); 2017 break; 2018 case 0x1003: 2019 /* skip gettid with valid return code. */ 2020 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee)); 2021 change_syscall_ret(_metadata, tracee, 45000); 2022 break; 2023 case 0x1004: 2024 /* skip openat with error. */ 2025 EXPECT_EQ(__NR_openat, get_syscall(_metadata, tracee)); 2026 change_syscall_ret(_metadata, tracee, -ESRCH); 2027 break; 2028 case 0x1005: 2029 /* do nothing (allow getppid) */ 2030 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee)); 2031 break; 2032 default: 2033 EXPECT_EQ(0, msg) { 2034 TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg); 2035 kill(tracee, SIGKILL); 2036 } 2037 } 2038 2039 } 2040 2041 FIXTURE(TRACE_syscall) { 2042 struct sock_fprog prog; 2043 pid_t tracer, mytid, mypid, parent; 2044 long syscall_nr; 2045 }; 2046 2047 void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee, 2048 int status, void *args) 2049 { 2050 int ret; 2051 unsigned long msg; 2052 static bool entry; 2053 long syscall_nr_val, syscall_ret_val; 2054 long *syscall_nr = NULL, *syscall_ret = NULL; 2055 FIXTURE_DATA(TRACE_syscall) *self = args; 2056 2057 EXPECT_EQ(WSTOPSIG(status) & 0x80, 0x80) { 2058 TH_LOG("Unexpected WSTOPSIG: %d", WSTOPSIG(status)); 2059 return; 2060 } 2061 2062 /* 2063 * The traditional way to tell PTRACE_SYSCALL entry/exit 2064 * is by counting. 2065 */ 2066 entry = !entry; 2067 2068 /* Make sure we got an appropriate message. */ 2069 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 2070 EXPECT_EQ(0, ret); 2071 EXPECT_EQ(entry ? PTRACE_EVENTMSG_SYSCALL_ENTRY 2072 : PTRACE_EVENTMSG_SYSCALL_EXIT, msg); 2073 2074 /* 2075 * Some architectures only support setting return values during 2076 * syscall exit under ptrace, and on exit the syscall number may 2077 * no longer be available. Therefore, save the initial sycall 2078 * number here, so it can be examined during both entry and exit 2079 * phases. 2080 */ 2081 if (entry) 2082 self->syscall_nr = get_syscall(_metadata, tracee); 2083 2084 /* 2085 * Depending on the architecture's syscall setting abilities, we 2086 * pick which things to set during this phase (entry or exit). 2087 */ 2088 if (entry == ptrace_entry_set_syscall_nr) 2089 syscall_nr = &syscall_nr_val; 2090 if (entry == ptrace_entry_set_syscall_ret) 2091 syscall_ret = &syscall_ret_val; 2092 2093 /* Now handle the actual rewriting cases. */ 2094 switch (self->syscall_nr) { 2095 case __NR_getpid: 2096 syscall_nr_val = __NR_getppid; 2097 /* Never change syscall return for this case. */ 2098 syscall_ret = NULL; 2099 break; 2100 case __NR_gettid: 2101 syscall_nr_val = -1; 2102 syscall_ret_val = 45000; 2103 break; 2104 case __NR_openat: 2105 syscall_nr_val = -1; 2106 syscall_ret_val = -ESRCH; 2107 break; 2108 default: 2109 /* Unhandled, do nothing. */ 2110 return; 2111 } 2112 2113 __change_syscall(_metadata, tracee, syscall_nr, syscall_ret); 2114 } 2115 2116 FIXTURE_VARIANT(TRACE_syscall) { 2117 /* 2118 * All of the SECCOMP_RET_TRACE behaviors can be tested with either 2119 * SECCOMP_RET_TRACE+PTRACE_CONT or plain ptrace()+PTRACE_SYSCALL. 2120 * This indicates if we should use SECCOMP_RET_TRACE (false), or 2121 * ptrace (true). 2122 */ 2123 bool use_ptrace; 2124 }; 2125 2126 FIXTURE_VARIANT_ADD(TRACE_syscall, ptrace) { 2127 .use_ptrace = true, 2128 }; 2129 2130 FIXTURE_VARIANT_ADD(TRACE_syscall, seccomp) { 2131 .use_ptrace = false, 2132 }; 2133 2134 FIXTURE_SETUP(TRACE_syscall) 2135 { 2136 struct sock_filter filter[] = { 2137 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2138 offsetof(struct seccomp_data, nr)), 2139 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 2140 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002), 2141 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1), 2142 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003), 2143 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_openat, 0, 1), 2144 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004), 2145 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 2146 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1005), 2147 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2148 }; 2149 struct sock_fprog prog = { 2150 .len = (unsigned short)ARRAY_SIZE(filter), 2151 .filter = filter, 2152 }; 2153 long ret; 2154 2155 /* Prepare some testable syscall results. */ 2156 self->mytid = syscall(__NR_gettid); 2157 ASSERT_GT(self->mytid, 0); 2158 ASSERT_NE(self->mytid, 1) { 2159 TH_LOG("Running this test as init is not supported. :)"); 2160 } 2161 2162 self->mypid = getpid(); 2163 ASSERT_GT(self->mypid, 0); 2164 ASSERT_EQ(self->mytid, self->mypid); 2165 2166 self->parent = getppid(); 2167 ASSERT_GT(self->parent, 0); 2168 ASSERT_NE(self->parent, self->mypid); 2169 2170 /* Launch tracer. */ 2171 self->tracer = setup_trace_fixture(_metadata, 2172 variant->use_ptrace ? tracer_ptrace 2173 : tracer_seccomp, 2174 self, variant->use_ptrace); 2175 2176 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 2177 ASSERT_EQ(0, ret); 2178 2179 /* Do not install seccomp rewrite filters, as we'll use ptrace instead. */ 2180 if (variant->use_ptrace) 2181 return; 2182 2183 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2184 ASSERT_EQ(0, ret); 2185 } 2186 2187 FIXTURE_TEARDOWN(TRACE_syscall) 2188 { 2189 teardown_trace_fixture(_metadata, self->tracer); 2190 } 2191 2192 TEST(negative_ENOSYS) 2193 { 2194 #if defined(__arm__) 2195 SKIP(return, "arm32 does not support calling syscall -1"); 2196 #endif 2197 /* 2198 * There should be no difference between an "internal" skip 2199 * and userspace asking for syscall "-1". 2200 */ 2201 errno = 0; 2202 EXPECT_EQ(-1, syscall(-1)); 2203 EXPECT_EQ(errno, ENOSYS); 2204 /* And no difference for "still not valid but not -1". */ 2205 errno = 0; 2206 EXPECT_EQ(-1, syscall(-101)); 2207 EXPECT_EQ(errno, ENOSYS); 2208 } 2209 2210 TEST_F(TRACE_syscall, negative_ENOSYS) 2211 { 2212 negative_ENOSYS(_metadata); 2213 } 2214 2215 TEST_F(TRACE_syscall, syscall_allowed) 2216 { 2217 /* getppid works as expected (no changes). */ 2218 EXPECT_EQ(self->parent, syscall(__NR_getppid)); 2219 EXPECT_NE(self->mypid, syscall(__NR_getppid)); 2220 } 2221 2222 TEST_F(TRACE_syscall, syscall_redirected) 2223 { 2224 /* getpid has been redirected to getppid as expected. */ 2225 EXPECT_EQ(self->parent, syscall(__NR_getpid)); 2226 EXPECT_NE(self->mypid, syscall(__NR_getpid)); 2227 } 2228 2229 TEST_F(TRACE_syscall, syscall_errno) 2230 { 2231 /* Tracer should skip the open syscall, resulting in ESRCH. */ 2232 EXPECT_SYSCALL_RETURN(-ESRCH, syscall(__NR_openat)); 2233 } 2234 2235 TEST_F(TRACE_syscall, syscall_faked) 2236 { 2237 /* Tracer skips the gettid syscall and store altered return value. */ 2238 EXPECT_SYSCALL_RETURN(45000, syscall(__NR_gettid)); 2239 } 2240 2241 TEST_F_SIGNAL(TRACE_syscall, kill_immediate, SIGSYS) 2242 { 2243 struct sock_filter filter[] = { 2244 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2245 offsetof(struct seccomp_data, nr)), 2246 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_mknodat, 0, 1), 2247 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD), 2248 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2249 }; 2250 struct sock_fprog prog = { 2251 .len = (unsigned short)ARRAY_SIZE(filter), 2252 .filter = filter, 2253 }; 2254 long ret; 2255 2256 /* Install "kill on mknodat" filter. */ 2257 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2258 ASSERT_EQ(0, ret); 2259 2260 /* This should immediately die with SIGSYS, regardless of tracer. */ 2261 EXPECT_EQ(-1, syscall(__NR_mknodat, -1, NULL, 0, 0)); 2262 } 2263 2264 TEST_F(TRACE_syscall, skip_after) 2265 { 2266 struct sock_filter filter[] = { 2267 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2268 offsetof(struct seccomp_data, nr)), 2269 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 2270 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM), 2271 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2272 }; 2273 struct sock_fprog prog = { 2274 .len = (unsigned short)ARRAY_SIZE(filter), 2275 .filter = filter, 2276 }; 2277 long ret; 2278 2279 /* Install additional "errno on getppid" filter. */ 2280 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2281 ASSERT_EQ(0, ret); 2282 2283 /* Tracer will redirect getpid to getppid, and we should see EPERM. */ 2284 errno = 0; 2285 EXPECT_EQ(-1, syscall(__NR_getpid)); 2286 EXPECT_EQ(EPERM, errno); 2287 } 2288 2289 TEST_F_SIGNAL(TRACE_syscall, kill_after, SIGSYS) 2290 { 2291 struct sock_filter filter[] = { 2292 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2293 offsetof(struct seccomp_data, nr)), 2294 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 2295 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2296 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2297 }; 2298 struct sock_fprog prog = { 2299 .len = (unsigned short)ARRAY_SIZE(filter), 2300 .filter = filter, 2301 }; 2302 long ret; 2303 2304 /* Install additional "death on getppid" filter. */ 2305 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2306 ASSERT_EQ(0, ret); 2307 2308 /* Tracer will redirect getpid to getppid, and we should die. */ 2309 EXPECT_NE(self->mypid, syscall(__NR_getpid)); 2310 } 2311 2312 TEST(seccomp_syscall) 2313 { 2314 struct sock_filter filter[] = { 2315 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2316 }; 2317 struct sock_fprog prog = { 2318 .len = (unsigned short)ARRAY_SIZE(filter), 2319 .filter = filter, 2320 }; 2321 long ret; 2322 2323 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 2324 ASSERT_EQ(0, ret) { 2325 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2326 } 2327 2328 /* Reject insane operation. */ 2329 ret = seccomp(-1, 0, &prog); 2330 ASSERT_NE(ENOSYS, errno) { 2331 TH_LOG("Kernel does not support seccomp syscall!"); 2332 } 2333 EXPECT_EQ(EINVAL, errno) { 2334 TH_LOG("Did not reject crazy op value!"); 2335 } 2336 2337 /* Reject strict with flags or pointer. */ 2338 ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL); 2339 EXPECT_EQ(EINVAL, errno) { 2340 TH_LOG("Did not reject mode strict with flags!"); 2341 } 2342 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog); 2343 EXPECT_EQ(EINVAL, errno) { 2344 TH_LOG("Did not reject mode strict with uargs!"); 2345 } 2346 2347 /* Reject insane args for filter. */ 2348 ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog); 2349 EXPECT_EQ(EINVAL, errno) { 2350 TH_LOG("Did not reject crazy filter flags!"); 2351 } 2352 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL); 2353 EXPECT_EQ(EFAULT, errno) { 2354 TH_LOG("Did not reject NULL filter!"); 2355 } 2356 2357 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2358 EXPECT_EQ(0, errno) { 2359 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s", 2360 strerror(errno)); 2361 } 2362 } 2363 2364 TEST(seccomp_syscall_mode_lock) 2365 { 2366 struct sock_filter filter[] = { 2367 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2368 }; 2369 struct sock_fprog prog = { 2370 .len = (unsigned short)ARRAY_SIZE(filter), 2371 .filter = filter, 2372 }; 2373 long ret; 2374 2375 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 2376 ASSERT_EQ(0, ret) { 2377 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2378 } 2379 2380 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2381 ASSERT_NE(ENOSYS, errno) { 2382 TH_LOG("Kernel does not support seccomp syscall!"); 2383 } 2384 EXPECT_EQ(0, ret) { 2385 TH_LOG("Could not install filter!"); 2386 } 2387 2388 /* Make sure neither entry point will switch to strict. */ 2389 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0); 2390 EXPECT_EQ(EINVAL, errno) { 2391 TH_LOG("Switched to mode strict!"); 2392 } 2393 2394 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL); 2395 EXPECT_EQ(EINVAL, errno) { 2396 TH_LOG("Switched to mode strict!"); 2397 } 2398 } 2399 2400 /* 2401 * Test detection of known and unknown filter flags. Userspace needs to be able 2402 * to check if a filter flag is supported by the current kernel and a good way 2403 * of doing that is by attempting to enter filter mode, with the flag bit in 2404 * question set, and a NULL pointer for the _args_ parameter. EFAULT indicates 2405 * that the flag is valid and EINVAL indicates that the flag is invalid. 2406 */ 2407 TEST(detect_seccomp_filter_flags) 2408 { 2409 unsigned int flags[] = { SECCOMP_FILTER_FLAG_TSYNC, 2410 SECCOMP_FILTER_FLAG_LOG, 2411 SECCOMP_FILTER_FLAG_SPEC_ALLOW, 2412 SECCOMP_FILTER_FLAG_NEW_LISTENER, 2413 SECCOMP_FILTER_FLAG_TSYNC_ESRCH }; 2414 unsigned int exclusive[] = { 2415 SECCOMP_FILTER_FLAG_TSYNC, 2416 SECCOMP_FILTER_FLAG_NEW_LISTENER }; 2417 unsigned int flag, all_flags, exclusive_mask; 2418 int i; 2419 long ret; 2420 2421 /* Test detection of individual known-good filter flags */ 2422 for (i = 0, all_flags = 0; i < ARRAY_SIZE(flags); i++) { 2423 int bits = 0; 2424 2425 flag = flags[i]; 2426 /* Make sure the flag is a single bit! */ 2427 while (flag) { 2428 if (flag & 0x1) 2429 bits ++; 2430 flag >>= 1; 2431 } 2432 ASSERT_EQ(1, bits); 2433 flag = flags[i]; 2434 2435 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2436 ASSERT_NE(ENOSYS, errno) { 2437 TH_LOG("Kernel does not support seccomp syscall!"); 2438 } 2439 EXPECT_EQ(-1, ret); 2440 EXPECT_EQ(EFAULT, errno) { 2441 TH_LOG("Failed to detect that a known-good filter flag (0x%X) is supported!", 2442 flag); 2443 } 2444 2445 all_flags |= flag; 2446 } 2447 2448 /* 2449 * Test detection of all known-good filter flags combined. But 2450 * for the exclusive flags we need to mask them out and try them 2451 * individually for the "all flags" testing. 2452 */ 2453 exclusive_mask = 0; 2454 for (i = 0; i < ARRAY_SIZE(exclusive); i++) 2455 exclusive_mask |= exclusive[i]; 2456 for (i = 0; i < ARRAY_SIZE(exclusive); i++) { 2457 flag = all_flags & ~exclusive_mask; 2458 flag |= exclusive[i]; 2459 2460 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2461 EXPECT_EQ(-1, ret); 2462 EXPECT_EQ(EFAULT, errno) { 2463 TH_LOG("Failed to detect that all known-good filter flags (0x%X) are supported!", 2464 flag); 2465 } 2466 } 2467 2468 /* Test detection of an unknown filter flags, without exclusives. */ 2469 flag = -1; 2470 flag &= ~exclusive_mask; 2471 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2472 EXPECT_EQ(-1, ret); 2473 EXPECT_EQ(EINVAL, errno) { 2474 TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported!", 2475 flag); 2476 } 2477 2478 /* 2479 * Test detection of an unknown filter flag that may simply need to be 2480 * added to this test 2481 */ 2482 flag = flags[ARRAY_SIZE(flags) - 1] << 1; 2483 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2484 EXPECT_EQ(-1, ret); 2485 EXPECT_EQ(EINVAL, errno) { 2486 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?", 2487 flag); 2488 } 2489 } 2490 2491 TEST(TSYNC_first) 2492 { 2493 struct sock_filter filter[] = { 2494 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2495 }; 2496 struct sock_fprog prog = { 2497 .len = (unsigned short)ARRAY_SIZE(filter), 2498 .filter = filter, 2499 }; 2500 long ret; 2501 2502 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 2503 ASSERT_EQ(0, ret) { 2504 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2505 } 2506 2507 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2508 &prog); 2509 ASSERT_NE(ENOSYS, errno) { 2510 TH_LOG("Kernel does not support seccomp syscall!"); 2511 } 2512 EXPECT_EQ(0, ret) { 2513 TH_LOG("Could not install initial filter with TSYNC!"); 2514 } 2515 } 2516 2517 #define TSYNC_SIBLINGS 2 2518 struct tsync_sibling { 2519 pthread_t tid; 2520 pid_t system_tid; 2521 sem_t *started; 2522 pthread_cond_t *cond; 2523 pthread_mutex_t *mutex; 2524 int diverge; 2525 int num_waits; 2526 struct sock_fprog *prog; 2527 struct __test_metadata *metadata; 2528 }; 2529 2530 /* 2531 * To avoid joining joined threads (which is not allowed by Bionic), 2532 * make sure we both successfully join and clear the tid to skip a 2533 * later join attempt during fixture teardown. Any remaining threads 2534 * will be directly killed during teardown. 2535 */ 2536 #define PTHREAD_JOIN(tid, status) \ 2537 do { \ 2538 int _rc = pthread_join(tid, status); \ 2539 if (_rc) { \ 2540 TH_LOG("pthread_join of tid %u failed: %d\n", \ 2541 (unsigned int)tid, _rc); \ 2542 } else { \ 2543 tid = 0; \ 2544 } \ 2545 } while (0) 2546 2547 FIXTURE(TSYNC) { 2548 struct sock_fprog root_prog, apply_prog; 2549 struct tsync_sibling sibling[TSYNC_SIBLINGS]; 2550 sem_t started; 2551 pthread_cond_t cond; 2552 pthread_mutex_t mutex; 2553 int sibling_count; 2554 }; 2555 2556 FIXTURE_SETUP(TSYNC) 2557 { 2558 struct sock_filter root_filter[] = { 2559 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2560 }; 2561 struct sock_filter apply_filter[] = { 2562 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2563 offsetof(struct seccomp_data, nr)), 2564 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 2565 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2566 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2567 }; 2568 2569 memset(&self->root_prog, 0, sizeof(self->root_prog)); 2570 memset(&self->apply_prog, 0, sizeof(self->apply_prog)); 2571 memset(&self->sibling, 0, sizeof(self->sibling)); 2572 self->root_prog.filter = malloc(sizeof(root_filter)); 2573 ASSERT_NE(NULL, self->root_prog.filter); 2574 memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter)); 2575 self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter); 2576 2577 self->apply_prog.filter = malloc(sizeof(apply_filter)); 2578 ASSERT_NE(NULL, self->apply_prog.filter); 2579 memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter)); 2580 self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter); 2581 2582 self->sibling_count = 0; 2583 pthread_mutex_init(&self->mutex, NULL); 2584 pthread_cond_init(&self->cond, NULL); 2585 sem_init(&self->started, 0, 0); 2586 self->sibling[0].tid = 0; 2587 self->sibling[0].cond = &self->cond; 2588 self->sibling[0].started = &self->started; 2589 self->sibling[0].mutex = &self->mutex; 2590 self->sibling[0].diverge = 0; 2591 self->sibling[0].num_waits = 1; 2592 self->sibling[0].prog = &self->root_prog; 2593 self->sibling[0].metadata = _metadata; 2594 self->sibling[1].tid = 0; 2595 self->sibling[1].cond = &self->cond; 2596 self->sibling[1].started = &self->started; 2597 self->sibling[1].mutex = &self->mutex; 2598 self->sibling[1].diverge = 0; 2599 self->sibling[1].prog = &self->root_prog; 2600 self->sibling[1].num_waits = 1; 2601 self->sibling[1].metadata = _metadata; 2602 } 2603 2604 FIXTURE_TEARDOWN(TSYNC) 2605 { 2606 int sib = 0; 2607 2608 if (self->root_prog.filter) 2609 free(self->root_prog.filter); 2610 if (self->apply_prog.filter) 2611 free(self->apply_prog.filter); 2612 2613 for ( ; sib < self->sibling_count; ++sib) { 2614 struct tsync_sibling *s = &self->sibling[sib]; 2615 2616 if (!s->tid) 2617 continue; 2618 /* 2619 * If a thread is still running, it may be stuck, so hit 2620 * it over the head really hard. 2621 */ 2622 pthread_kill(s->tid, 9); 2623 } 2624 pthread_mutex_destroy(&self->mutex); 2625 pthread_cond_destroy(&self->cond); 2626 sem_destroy(&self->started); 2627 } 2628 2629 void *tsync_sibling(void *data) 2630 { 2631 long ret = 0; 2632 struct tsync_sibling *me = data; 2633 2634 me->system_tid = syscall(__NR_gettid); 2635 2636 pthread_mutex_lock(me->mutex); 2637 if (me->diverge) { 2638 /* Just re-apply the root prog to fork the tree */ 2639 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, 2640 me->prog, 0, 0); 2641 } 2642 sem_post(me->started); 2643 /* Return outside of started so parent notices failures. */ 2644 if (ret) { 2645 pthread_mutex_unlock(me->mutex); 2646 return (void *)SIBLING_EXIT_FAILURE; 2647 } 2648 do { 2649 pthread_cond_wait(me->cond, me->mutex); 2650 me->num_waits = me->num_waits - 1; 2651 } while (me->num_waits); 2652 pthread_mutex_unlock(me->mutex); 2653 2654 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0); 2655 if (!ret) 2656 return (void *)SIBLING_EXIT_NEWPRIVS; 2657 read(-1, NULL, 0); 2658 return (void *)SIBLING_EXIT_UNKILLED; 2659 } 2660 2661 void tsync_start_sibling(struct tsync_sibling *sibling) 2662 { 2663 pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling); 2664 } 2665 2666 TEST_F(TSYNC, siblings_fail_prctl) 2667 { 2668 long ret; 2669 void *status; 2670 struct sock_filter filter[] = { 2671 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2672 offsetof(struct seccomp_data, nr)), 2673 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1), 2674 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL), 2675 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2676 }; 2677 struct sock_fprog prog = { 2678 .len = (unsigned short)ARRAY_SIZE(filter), 2679 .filter = filter, 2680 }; 2681 2682 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2683 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2684 } 2685 2686 /* Check prctl failure detection by requesting sib 0 diverge. */ 2687 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2688 ASSERT_NE(ENOSYS, errno) { 2689 TH_LOG("Kernel does not support seccomp syscall!"); 2690 } 2691 ASSERT_EQ(0, ret) { 2692 TH_LOG("setting filter failed"); 2693 } 2694 2695 self->sibling[0].diverge = 1; 2696 tsync_start_sibling(&self->sibling[0]); 2697 tsync_start_sibling(&self->sibling[1]); 2698 2699 while (self->sibling_count < TSYNC_SIBLINGS) { 2700 sem_wait(&self->started); 2701 self->sibling_count++; 2702 } 2703 2704 /* Signal the threads to clean up*/ 2705 pthread_mutex_lock(&self->mutex); 2706 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2707 TH_LOG("cond broadcast non-zero"); 2708 } 2709 pthread_mutex_unlock(&self->mutex); 2710 2711 /* Ensure diverging sibling failed to call prctl. */ 2712 PTHREAD_JOIN(self->sibling[0].tid, &status); 2713 EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status); 2714 PTHREAD_JOIN(self->sibling[1].tid, &status); 2715 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2716 } 2717 2718 TEST_F(TSYNC, two_siblings_with_ancestor) 2719 { 2720 long ret; 2721 void *status; 2722 2723 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2724 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2725 } 2726 2727 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2728 ASSERT_NE(ENOSYS, errno) { 2729 TH_LOG("Kernel does not support seccomp syscall!"); 2730 } 2731 ASSERT_EQ(0, ret) { 2732 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2733 } 2734 tsync_start_sibling(&self->sibling[0]); 2735 tsync_start_sibling(&self->sibling[1]); 2736 2737 while (self->sibling_count < TSYNC_SIBLINGS) { 2738 sem_wait(&self->started); 2739 self->sibling_count++; 2740 } 2741 2742 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2743 &self->apply_prog); 2744 ASSERT_EQ(0, ret) { 2745 TH_LOG("Could install filter on all threads!"); 2746 } 2747 /* Tell the siblings to test the policy */ 2748 pthread_mutex_lock(&self->mutex); 2749 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2750 TH_LOG("cond broadcast non-zero"); 2751 } 2752 pthread_mutex_unlock(&self->mutex); 2753 /* Ensure they are both killed and don't exit cleanly. */ 2754 PTHREAD_JOIN(self->sibling[0].tid, &status); 2755 EXPECT_EQ(0x0, (long)status); 2756 PTHREAD_JOIN(self->sibling[1].tid, &status); 2757 EXPECT_EQ(0x0, (long)status); 2758 } 2759 2760 TEST_F(TSYNC, two_sibling_want_nnp) 2761 { 2762 void *status; 2763 2764 /* start siblings before any prctl() operations */ 2765 tsync_start_sibling(&self->sibling[0]); 2766 tsync_start_sibling(&self->sibling[1]); 2767 while (self->sibling_count < TSYNC_SIBLINGS) { 2768 sem_wait(&self->started); 2769 self->sibling_count++; 2770 } 2771 2772 /* Tell the siblings to test no policy */ 2773 pthread_mutex_lock(&self->mutex); 2774 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2775 TH_LOG("cond broadcast non-zero"); 2776 } 2777 pthread_mutex_unlock(&self->mutex); 2778 2779 /* Ensure they are both upset about lacking nnp. */ 2780 PTHREAD_JOIN(self->sibling[0].tid, &status); 2781 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status); 2782 PTHREAD_JOIN(self->sibling[1].tid, &status); 2783 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status); 2784 } 2785 2786 TEST_F(TSYNC, two_siblings_with_no_filter) 2787 { 2788 long ret; 2789 void *status; 2790 2791 /* start siblings before any prctl() operations */ 2792 tsync_start_sibling(&self->sibling[0]); 2793 tsync_start_sibling(&self->sibling[1]); 2794 while (self->sibling_count < TSYNC_SIBLINGS) { 2795 sem_wait(&self->started); 2796 self->sibling_count++; 2797 } 2798 2799 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2800 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2801 } 2802 2803 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2804 &self->apply_prog); 2805 ASSERT_NE(ENOSYS, errno) { 2806 TH_LOG("Kernel does not support seccomp syscall!"); 2807 } 2808 ASSERT_EQ(0, ret) { 2809 TH_LOG("Could install filter on all threads!"); 2810 } 2811 2812 /* Tell the siblings to test the policy */ 2813 pthread_mutex_lock(&self->mutex); 2814 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2815 TH_LOG("cond broadcast non-zero"); 2816 } 2817 pthread_mutex_unlock(&self->mutex); 2818 2819 /* Ensure they are both killed and don't exit cleanly. */ 2820 PTHREAD_JOIN(self->sibling[0].tid, &status); 2821 EXPECT_EQ(0x0, (long)status); 2822 PTHREAD_JOIN(self->sibling[1].tid, &status); 2823 EXPECT_EQ(0x0, (long)status); 2824 } 2825 2826 TEST_F(TSYNC, two_siblings_with_one_divergence) 2827 { 2828 long ret; 2829 void *status; 2830 2831 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2832 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2833 } 2834 2835 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2836 ASSERT_NE(ENOSYS, errno) { 2837 TH_LOG("Kernel does not support seccomp syscall!"); 2838 } 2839 ASSERT_EQ(0, ret) { 2840 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2841 } 2842 self->sibling[0].diverge = 1; 2843 tsync_start_sibling(&self->sibling[0]); 2844 tsync_start_sibling(&self->sibling[1]); 2845 2846 while (self->sibling_count < TSYNC_SIBLINGS) { 2847 sem_wait(&self->started); 2848 self->sibling_count++; 2849 } 2850 2851 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2852 &self->apply_prog); 2853 ASSERT_EQ(self->sibling[0].system_tid, ret) { 2854 TH_LOG("Did not fail on diverged sibling."); 2855 } 2856 2857 /* Wake the threads */ 2858 pthread_mutex_lock(&self->mutex); 2859 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2860 TH_LOG("cond broadcast non-zero"); 2861 } 2862 pthread_mutex_unlock(&self->mutex); 2863 2864 /* Ensure they are both unkilled. */ 2865 PTHREAD_JOIN(self->sibling[0].tid, &status); 2866 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2867 PTHREAD_JOIN(self->sibling[1].tid, &status); 2868 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2869 } 2870 2871 TEST_F(TSYNC, two_siblings_with_one_divergence_no_tid_in_err) 2872 { 2873 long ret, flags; 2874 void *status; 2875 2876 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2877 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2878 } 2879 2880 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2881 ASSERT_NE(ENOSYS, errno) { 2882 TH_LOG("Kernel does not support seccomp syscall!"); 2883 } 2884 ASSERT_EQ(0, ret) { 2885 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2886 } 2887 self->sibling[0].diverge = 1; 2888 tsync_start_sibling(&self->sibling[0]); 2889 tsync_start_sibling(&self->sibling[1]); 2890 2891 while (self->sibling_count < TSYNC_SIBLINGS) { 2892 sem_wait(&self->started); 2893 self->sibling_count++; 2894 } 2895 2896 flags = SECCOMP_FILTER_FLAG_TSYNC | \ 2897 SECCOMP_FILTER_FLAG_TSYNC_ESRCH; 2898 ret = seccomp(SECCOMP_SET_MODE_FILTER, flags, &self->apply_prog); 2899 ASSERT_EQ(ESRCH, errno) { 2900 TH_LOG("Did not return ESRCH for diverged sibling."); 2901 } 2902 ASSERT_EQ(-1, ret) { 2903 TH_LOG("Did not fail on diverged sibling."); 2904 } 2905 2906 /* Wake the threads */ 2907 pthread_mutex_lock(&self->mutex); 2908 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2909 TH_LOG("cond broadcast non-zero"); 2910 } 2911 pthread_mutex_unlock(&self->mutex); 2912 2913 /* Ensure they are both unkilled. */ 2914 PTHREAD_JOIN(self->sibling[0].tid, &status); 2915 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2916 PTHREAD_JOIN(self->sibling[1].tid, &status); 2917 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2918 } 2919 2920 TEST_F(TSYNC, two_siblings_not_under_filter) 2921 { 2922 long ret, sib; 2923 void *status; 2924 struct timespec delay = { .tv_nsec = 100000000 }; 2925 2926 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2927 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2928 } 2929 2930 /* 2931 * Sibling 0 will have its own seccomp policy 2932 * and Sibling 1 will not be under seccomp at 2933 * all. Sibling 1 will enter seccomp and 0 2934 * will cause failure. 2935 */ 2936 self->sibling[0].diverge = 1; 2937 tsync_start_sibling(&self->sibling[0]); 2938 tsync_start_sibling(&self->sibling[1]); 2939 2940 while (self->sibling_count < TSYNC_SIBLINGS) { 2941 sem_wait(&self->started); 2942 self->sibling_count++; 2943 } 2944 2945 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2946 ASSERT_NE(ENOSYS, errno) { 2947 TH_LOG("Kernel does not support seccomp syscall!"); 2948 } 2949 ASSERT_EQ(0, ret) { 2950 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2951 } 2952 2953 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2954 &self->apply_prog); 2955 ASSERT_EQ(ret, self->sibling[0].system_tid) { 2956 TH_LOG("Did not fail on diverged sibling."); 2957 } 2958 sib = 1; 2959 if (ret == self->sibling[0].system_tid) 2960 sib = 0; 2961 2962 pthread_mutex_lock(&self->mutex); 2963 2964 /* Increment the other siblings num_waits so we can clean up 2965 * the one we just saw. 2966 */ 2967 self->sibling[!sib].num_waits += 1; 2968 2969 /* Signal the thread to clean up*/ 2970 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2971 TH_LOG("cond broadcast non-zero"); 2972 } 2973 pthread_mutex_unlock(&self->mutex); 2974 PTHREAD_JOIN(self->sibling[sib].tid, &status); 2975 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2976 /* Poll for actual task death. pthread_join doesn't guarantee it. */ 2977 while (!kill(self->sibling[sib].system_tid, 0)) 2978 nanosleep(&delay, NULL); 2979 /* Switch to the remaining sibling */ 2980 sib = !sib; 2981 2982 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2983 &self->apply_prog); 2984 ASSERT_EQ(0, ret) { 2985 TH_LOG("Expected the remaining sibling to sync"); 2986 }; 2987 2988 pthread_mutex_lock(&self->mutex); 2989 2990 /* If remaining sibling didn't have a chance to wake up during 2991 * the first broadcast, manually reduce the num_waits now. 2992 */ 2993 if (self->sibling[sib].num_waits > 1) 2994 self->sibling[sib].num_waits = 1; 2995 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2996 TH_LOG("cond broadcast non-zero"); 2997 } 2998 pthread_mutex_unlock(&self->mutex); 2999 PTHREAD_JOIN(self->sibling[sib].tid, &status); 3000 EXPECT_EQ(0, (long)status); 3001 /* Poll for actual task death. pthread_join doesn't guarantee it. */ 3002 while (!kill(self->sibling[sib].system_tid, 0)) 3003 nanosleep(&delay, NULL); 3004 3005 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 3006 &self->apply_prog); 3007 ASSERT_EQ(0, ret); /* just us chickens */ 3008 } 3009 3010 /* Make sure restarted syscalls are seen directly as "restart_syscall". */ 3011 TEST(syscall_restart) 3012 { 3013 long ret; 3014 unsigned long msg; 3015 pid_t child_pid; 3016 int pipefd[2]; 3017 int status; 3018 siginfo_t info = { }; 3019 struct sock_filter filter[] = { 3020 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 3021 offsetof(struct seccomp_data, nr)), 3022 3023 #ifdef __NR_sigreturn 3024 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 7, 0), 3025 #endif 3026 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 6, 0), 3027 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 5, 0), 3028 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 4, 0), 3029 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 5, 0), 3030 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_clock_nanosleep, 4, 0), 3031 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0), 3032 3033 /* Allow __NR_write for easy logging. */ 3034 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1), 3035 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3036 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 3037 /* The nanosleep jump target. */ 3038 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100), 3039 /* The restart_syscall jump target. */ 3040 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200), 3041 }; 3042 struct sock_fprog prog = { 3043 .len = (unsigned short)ARRAY_SIZE(filter), 3044 .filter = filter, 3045 }; 3046 #if defined(__arm__) 3047 struct utsname utsbuf; 3048 #endif 3049 3050 ASSERT_EQ(0, pipe(pipefd)); 3051 3052 child_pid = fork(); 3053 ASSERT_LE(0, child_pid); 3054 if (child_pid == 0) { 3055 /* Child uses EXPECT not ASSERT to deliver status correctly. */ 3056 char buf = ' '; 3057 struct timespec timeout = { }; 3058 3059 /* Attach parent as tracer and stop. */ 3060 EXPECT_EQ(0, ptrace(PTRACE_TRACEME)); 3061 EXPECT_EQ(0, raise(SIGSTOP)); 3062 3063 EXPECT_EQ(0, close(pipefd[1])); 3064 3065 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 3066 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3067 } 3068 3069 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 3070 EXPECT_EQ(0, ret) { 3071 TH_LOG("Failed to install filter!"); 3072 } 3073 3074 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) { 3075 TH_LOG("Failed to read() sync from parent"); 3076 } 3077 EXPECT_EQ('.', buf) { 3078 TH_LOG("Failed to get sync data from read()"); 3079 } 3080 3081 /* Start nanosleep to be interrupted. */ 3082 timeout.tv_sec = 1; 3083 errno = 0; 3084 EXPECT_EQ(0, nanosleep(&timeout, NULL)) { 3085 TH_LOG("Call to nanosleep() failed (errno %d: %s)", 3086 errno, strerror(errno)); 3087 } 3088 3089 /* Read final sync from parent. */ 3090 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) { 3091 TH_LOG("Failed final read() from parent"); 3092 } 3093 EXPECT_EQ('!', buf) { 3094 TH_LOG("Failed to get final data from read()"); 3095 } 3096 3097 /* Directly report the status of our test harness results. */ 3098 syscall(__NR_exit, _metadata->exit_code); 3099 } 3100 EXPECT_EQ(0, close(pipefd[0])); 3101 3102 /* Attach to child, setup options, and release. */ 3103 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3104 ASSERT_EQ(true, WIFSTOPPED(status)); 3105 ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL, 3106 PTRACE_O_TRACESECCOMP)); 3107 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3108 ASSERT_EQ(1, write(pipefd[1], ".", 1)); 3109 3110 /* Wait for nanosleep() to start. */ 3111 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3112 ASSERT_EQ(true, WIFSTOPPED(status)); 3113 ASSERT_EQ(SIGTRAP, WSTOPSIG(status)); 3114 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16)); 3115 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg)); 3116 ASSERT_EQ(0x100, msg); 3117 ret = get_syscall(_metadata, child_pid); 3118 EXPECT_TRUE(ret == __NR_nanosleep || ret == __NR_clock_nanosleep); 3119 3120 /* Might as well check siginfo for sanity while we're here. */ 3121 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info)); 3122 ASSERT_EQ(SIGTRAP, info.si_signo); 3123 ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code); 3124 EXPECT_EQ(0, info.si_errno); 3125 EXPECT_EQ(getuid(), info.si_uid); 3126 /* Verify signal delivery came from child (seccomp-triggered). */ 3127 EXPECT_EQ(child_pid, info.si_pid); 3128 3129 /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */ 3130 ASSERT_EQ(0, kill(child_pid, SIGSTOP)); 3131 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3132 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3133 ASSERT_EQ(true, WIFSTOPPED(status)); 3134 ASSERT_EQ(SIGSTOP, WSTOPSIG(status)); 3135 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info)); 3136 /* 3137 * There is no siginfo on SIGSTOP any more, so we can't verify 3138 * signal delivery came from parent now (getpid() == info.si_pid). 3139 * https://lkml.kernel.org/r/CAGXu5jJaZAOzP1qFz66tYrtbuywqb+UN2SOA1VLHpCCOiYvYeg@mail.gmail.com 3140 * At least verify the SIGSTOP via PTRACE_GETSIGINFO. 3141 */ 3142 EXPECT_EQ(SIGSTOP, info.si_signo); 3143 3144 /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */ 3145 ASSERT_EQ(0, kill(child_pid, SIGCONT)); 3146 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3147 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3148 ASSERT_EQ(true, WIFSTOPPED(status)); 3149 ASSERT_EQ(SIGCONT, WSTOPSIG(status)); 3150 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3151 3152 /* Wait for restart_syscall() to start. */ 3153 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3154 ASSERT_EQ(true, WIFSTOPPED(status)); 3155 ASSERT_EQ(SIGTRAP, WSTOPSIG(status)); 3156 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16)); 3157 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg)); 3158 3159 ASSERT_EQ(0x200, msg); 3160 ret = get_syscall(_metadata, child_pid); 3161 #if defined(__arm__) 3162 /* 3163 * FIXME: 3164 * - native ARM registers do NOT expose true syscall. 3165 * - compat ARM registers on ARM64 DO expose true syscall. 3166 */ 3167 ASSERT_EQ(0, uname(&utsbuf)); 3168 if (strncmp(utsbuf.machine, "arm", 3) == 0) { 3169 EXPECT_EQ(__NR_nanosleep, ret); 3170 } else 3171 #endif 3172 { 3173 EXPECT_EQ(__NR_restart_syscall, ret); 3174 } 3175 3176 /* Write again to end test. */ 3177 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3178 ASSERT_EQ(1, write(pipefd[1], "!", 1)); 3179 EXPECT_EQ(0, close(pipefd[1])); 3180 3181 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3182 if (WIFSIGNALED(status) || WEXITSTATUS(status)) 3183 _metadata->exit_code = KSFT_FAIL; 3184 } 3185 3186 TEST_SIGNAL(filter_flag_log, SIGSYS) 3187 { 3188 struct sock_filter allow_filter[] = { 3189 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3190 }; 3191 struct sock_filter kill_filter[] = { 3192 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 3193 offsetof(struct seccomp_data, nr)), 3194 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 3195 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 3196 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3197 }; 3198 struct sock_fprog allow_prog = { 3199 .len = (unsigned short)ARRAY_SIZE(allow_filter), 3200 .filter = allow_filter, 3201 }; 3202 struct sock_fprog kill_prog = { 3203 .len = (unsigned short)ARRAY_SIZE(kill_filter), 3204 .filter = kill_filter, 3205 }; 3206 long ret; 3207 pid_t parent = getppid(); 3208 3209 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3210 ASSERT_EQ(0, ret); 3211 3212 /* Verify that the FILTER_FLAG_LOG flag isn't accepted in strict mode */ 3213 ret = seccomp(SECCOMP_SET_MODE_STRICT, SECCOMP_FILTER_FLAG_LOG, 3214 &allow_prog); 3215 ASSERT_NE(ENOSYS, errno) { 3216 TH_LOG("Kernel does not support seccomp syscall!"); 3217 } 3218 EXPECT_NE(0, ret) { 3219 TH_LOG("Kernel accepted FILTER_FLAG_LOG flag in strict mode!"); 3220 } 3221 EXPECT_EQ(EINVAL, errno) { 3222 TH_LOG("Kernel returned unexpected errno for FILTER_FLAG_LOG flag in strict mode!"); 3223 } 3224 3225 /* Verify that a simple, permissive filter can be added with no flags */ 3226 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog); 3227 EXPECT_EQ(0, ret); 3228 3229 /* See if the same filter can be added with the FILTER_FLAG_LOG flag */ 3230 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG, 3231 &allow_prog); 3232 ASSERT_NE(EINVAL, errno) { 3233 TH_LOG("Kernel does not support the FILTER_FLAG_LOG flag!"); 3234 } 3235 EXPECT_EQ(0, ret); 3236 3237 /* Ensure that the kill filter works with the FILTER_FLAG_LOG flag */ 3238 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG, 3239 &kill_prog); 3240 EXPECT_EQ(0, ret); 3241 3242 EXPECT_EQ(parent, syscall(__NR_getppid)); 3243 /* getpid() should never return. */ 3244 EXPECT_EQ(0, syscall(__NR_getpid)); 3245 } 3246 3247 TEST(get_action_avail) 3248 { 3249 __u32 actions[] = { SECCOMP_RET_KILL_THREAD, SECCOMP_RET_TRAP, 3250 SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE, 3251 SECCOMP_RET_LOG, SECCOMP_RET_ALLOW }; 3252 __u32 unknown_action = 0x10000000U; 3253 int i; 3254 long ret; 3255 3256 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[0]); 3257 ASSERT_NE(ENOSYS, errno) { 3258 TH_LOG("Kernel does not support seccomp syscall!"); 3259 } 3260 ASSERT_NE(EINVAL, errno) { 3261 TH_LOG("Kernel does not support SECCOMP_GET_ACTION_AVAIL operation!"); 3262 } 3263 EXPECT_EQ(ret, 0); 3264 3265 for (i = 0; i < ARRAY_SIZE(actions); i++) { 3266 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[i]); 3267 EXPECT_EQ(ret, 0) { 3268 TH_LOG("Expected action (0x%X) not available!", 3269 actions[i]); 3270 } 3271 } 3272 3273 /* Check that an unknown action is handled properly (EOPNOTSUPP) */ 3274 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &unknown_action); 3275 EXPECT_EQ(ret, -1); 3276 EXPECT_EQ(errno, EOPNOTSUPP); 3277 } 3278 3279 TEST(get_metadata) 3280 { 3281 pid_t pid; 3282 int pipefd[2]; 3283 char buf; 3284 struct seccomp_metadata md; 3285 long ret; 3286 3287 /* Only real root can get metadata. */ 3288 if (geteuid()) { 3289 SKIP(return, "get_metadata requires real root"); 3290 return; 3291 } 3292 3293 ASSERT_EQ(0, pipe(pipefd)); 3294 3295 pid = fork(); 3296 ASSERT_GE(pid, 0); 3297 if (pid == 0) { 3298 struct sock_filter filter[] = { 3299 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3300 }; 3301 struct sock_fprog prog = { 3302 .len = (unsigned short)ARRAY_SIZE(filter), 3303 .filter = filter, 3304 }; 3305 3306 /* one with log, one without */ 3307 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 3308 SECCOMP_FILTER_FLAG_LOG, &prog)); 3309 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog)); 3310 3311 EXPECT_EQ(0, close(pipefd[0])); 3312 ASSERT_EQ(1, write(pipefd[1], "1", 1)); 3313 ASSERT_EQ(0, close(pipefd[1])); 3314 3315 while (1) 3316 sleep(100); 3317 } 3318 3319 ASSERT_EQ(0, close(pipefd[1])); 3320 ASSERT_EQ(1, read(pipefd[0], &buf, 1)); 3321 3322 ASSERT_EQ(0, ptrace(PTRACE_ATTACH, pid)); 3323 ASSERT_EQ(pid, waitpid(pid, NULL, 0)); 3324 3325 /* Past here must not use ASSERT or child process is never killed. */ 3326 3327 md.filter_off = 0; 3328 errno = 0; 3329 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md); 3330 EXPECT_EQ(sizeof(md), ret) { 3331 if (errno == EINVAL) 3332 SKIP(goto skip, "Kernel does not support PTRACE_SECCOMP_GET_METADATA (missing CONFIG_CHECKPOINT_RESTORE?)"); 3333 } 3334 3335 EXPECT_EQ(md.flags, SECCOMP_FILTER_FLAG_LOG); 3336 EXPECT_EQ(md.filter_off, 0); 3337 3338 md.filter_off = 1; 3339 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md); 3340 EXPECT_EQ(sizeof(md), ret); 3341 EXPECT_EQ(md.flags, 0); 3342 EXPECT_EQ(md.filter_off, 1); 3343 3344 skip: 3345 ASSERT_EQ(0, kill(pid, SIGKILL)); 3346 } 3347 3348 static int user_notif_syscall(int nr, unsigned int flags) 3349 { 3350 struct sock_filter filter[] = { 3351 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 3352 offsetof(struct seccomp_data, nr)), 3353 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, nr, 0, 1), 3354 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_USER_NOTIF), 3355 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3356 }; 3357 3358 struct sock_fprog prog = { 3359 .len = (unsigned short)ARRAY_SIZE(filter), 3360 .filter = filter, 3361 }; 3362 3363 return seccomp(SECCOMP_SET_MODE_FILTER, flags, &prog); 3364 } 3365 3366 #define USER_NOTIF_MAGIC INT_MAX 3367 TEST(user_notification_basic) 3368 { 3369 pid_t pid; 3370 long ret; 3371 int status, listener; 3372 struct seccomp_notif req = {}; 3373 struct seccomp_notif_resp resp = {}; 3374 struct pollfd pollfd; 3375 3376 struct sock_filter filter[] = { 3377 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3378 }; 3379 struct sock_fprog prog = { 3380 .len = (unsigned short)ARRAY_SIZE(filter), 3381 .filter = filter, 3382 }; 3383 3384 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3385 ASSERT_EQ(0, ret) { 3386 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3387 } 3388 3389 pid = fork(); 3390 ASSERT_GE(pid, 0); 3391 3392 /* Check that we get -ENOSYS with no listener attached */ 3393 if (pid == 0) { 3394 if (user_notif_syscall(__NR_getppid, 0) < 0) 3395 exit(1); 3396 ret = syscall(__NR_getppid); 3397 exit(ret >= 0 || errno != ENOSYS); 3398 } 3399 3400 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3401 EXPECT_EQ(true, WIFEXITED(status)); 3402 EXPECT_EQ(0, WEXITSTATUS(status)); 3403 3404 /* Add some no-op filters for grins. */ 3405 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3406 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3407 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3408 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3409 3410 /* Check that the basic notification machinery works */ 3411 listener = user_notif_syscall(__NR_getppid, 3412 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3413 ASSERT_GE(listener, 0); 3414 3415 /* Installing a second listener in the chain should EBUSY */ 3416 EXPECT_EQ(user_notif_syscall(__NR_getppid, 3417 SECCOMP_FILTER_FLAG_NEW_LISTENER), 3418 -1); 3419 EXPECT_EQ(errno, EBUSY); 3420 3421 pid = fork(); 3422 ASSERT_GE(pid, 0); 3423 3424 if (pid == 0) { 3425 ret = syscall(__NR_getppid); 3426 exit(ret != USER_NOTIF_MAGIC); 3427 } 3428 3429 pollfd.fd = listener; 3430 pollfd.events = POLLIN | POLLOUT; 3431 3432 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3433 EXPECT_EQ(pollfd.revents, POLLIN); 3434 3435 /* Test that we can't pass garbage to the kernel. */ 3436 memset(&req, 0, sizeof(req)); 3437 req.pid = -1; 3438 errno = 0; 3439 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req); 3440 EXPECT_EQ(-1, ret); 3441 EXPECT_EQ(EINVAL, errno); 3442 3443 if (ret) { 3444 req.pid = 0; 3445 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3446 } 3447 3448 pollfd.fd = listener; 3449 pollfd.events = POLLIN | POLLOUT; 3450 3451 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3452 EXPECT_EQ(pollfd.revents, POLLOUT); 3453 3454 EXPECT_EQ(req.data.nr, __NR_getppid); 3455 3456 resp.id = req.id; 3457 resp.error = 0; 3458 resp.val = USER_NOTIF_MAGIC; 3459 3460 /* check that we make sure flags == 0 */ 3461 resp.flags = 1; 3462 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3463 EXPECT_EQ(errno, EINVAL); 3464 3465 resp.flags = 0; 3466 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3467 3468 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3469 EXPECT_EQ(true, WIFEXITED(status)); 3470 EXPECT_EQ(0, WEXITSTATUS(status)); 3471 } 3472 3473 TEST(user_notification_with_tsync) 3474 { 3475 int ret; 3476 unsigned int flags; 3477 3478 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3479 ASSERT_EQ(0, ret) { 3480 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3481 } 3482 3483 /* these were exclusive */ 3484 flags = SECCOMP_FILTER_FLAG_NEW_LISTENER | 3485 SECCOMP_FILTER_FLAG_TSYNC; 3486 ASSERT_EQ(-1, user_notif_syscall(__NR_getppid, flags)); 3487 ASSERT_EQ(EINVAL, errno); 3488 3489 /* but now they're not */ 3490 flags |= SECCOMP_FILTER_FLAG_TSYNC_ESRCH; 3491 ret = user_notif_syscall(__NR_getppid, flags); 3492 close(ret); 3493 ASSERT_LE(0, ret); 3494 } 3495 3496 TEST(user_notification_kill_in_middle) 3497 { 3498 pid_t pid; 3499 long ret; 3500 int listener; 3501 struct seccomp_notif req = {}; 3502 struct seccomp_notif_resp resp = {}; 3503 3504 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3505 ASSERT_EQ(0, ret) { 3506 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3507 } 3508 3509 listener = user_notif_syscall(__NR_getppid, 3510 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3511 ASSERT_GE(listener, 0); 3512 3513 /* 3514 * Check that nothing bad happens when we kill the task in the middle 3515 * of a syscall. 3516 */ 3517 pid = fork(); 3518 ASSERT_GE(pid, 0); 3519 3520 if (pid == 0) { 3521 ret = syscall(__NR_getppid); 3522 exit(ret != USER_NOTIF_MAGIC); 3523 } 3524 3525 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3526 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), 0); 3527 3528 EXPECT_EQ(kill(pid, SIGKILL), 0); 3529 EXPECT_EQ(waitpid(pid, NULL, 0), pid); 3530 3531 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), -1); 3532 3533 resp.id = req.id; 3534 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp); 3535 EXPECT_EQ(ret, -1); 3536 EXPECT_EQ(errno, ENOENT); 3537 } 3538 3539 static int handled = -1; 3540 3541 static void signal_handler(int signal) 3542 { 3543 if (write(handled, "c", 1) != 1) 3544 perror("write from signal"); 3545 } 3546 3547 TEST(user_notification_signal) 3548 { 3549 pid_t pid; 3550 long ret; 3551 int status, listener, sk_pair[2]; 3552 struct seccomp_notif req = {}; 3553 struct seccomp_notif_resp resp = {}; 3554 char c; 3555 3556 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3557 ASSERT_EQ(0, ret) { 3558 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3559 } 3560 3561 ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0); 3562 3563 listener = user_notif_syscall(__NR_gettid, 3564 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3565 ASSERT_GE(listener, 0); 3566 3567 pid = fork(); 3568 ASSERT_GE(pid, 0); 3569 3570 if (pid == 0) { 3571 close(sk_pair[0]); 3572 handled = sk_pair[1]; 3573 if (signal(SIGUSR1, signal_handler) == SIG_ERR) { 3574 perror("signal"); 3575 exit(1); 3576 } 3577 /* 3578 * ERESTARTSYS behavior is a bit hard to test, because we need 3579 * to rely on a signal that has not yet been handled. Let's at 3580 * least check that the error code gets propagated through, and 3581 * hope that it doesn't break when there is actually a signal :) 3582 */ 3583 ret = syscall(__NR_gettid); 3584 exit(!(ret == -1 && errno == 512)); 3585 } 3586 3587 close(sk_pair[1]); 3588 3589 memset(&req, 0, sizeof(req)); 3590 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3591 3592 EXPECT_EQ(kill(pid, SIGUSR1), 0); 3593 3594 /* 3595 * Make sure the signal really is delivered, which means we're not 3596 * stuck in the user notification code any more and the notification 3597 * should be dead. 3598 */ 3599 EXPECT_EQ(read(sk_pair[0], &c, 1), 1); 3600 3601 resp.id = req.id; 3602 resp.error = -EPERM; 3603 resp.val = 0; 3604 3605 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3606 EXPECT_EQ(errno, ENOENT); 3607 3608 memset(&req, 0, sizeof(req)); 3609 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3610 3611 resp.id = req.id; 3612 resp.error = -512; /* -ERESTARTSYS */ 3613 resp.val = 0; 3614 3615 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3616 3617 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3618 EXPECT_EQ(true, WIFEXITED(status)); 3619 EXPECT_EQ(0, WEXITSTATUS(status)); 3620 } 3621 3622 TEST(user_notification_closed_listener) 3623 { 3624 pid_t pid; 3625 long ret; 3626 int status, listener; 3627 3628 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3629 ASSERT_EQ(0, ret) { 3630 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3631 } 3632 3633 listener = user_notif_syscall(__NR_getppid, 3634 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3635 ASSERT_GE(listener, 0); 3636 3637 /* 3638 * Check that we get an ENOSYS when the listener is closed. 3639 */ 3640 pid = fork(); 3641 ASSERT_GE(pid, 0); 3642 if (pid == 0) { 3643 close(listener); 3644 ret = syscall(__NR_getppid); 3645 exit(ret != -1 && errno != ENOSYS); 3646 } 3647 3648 close(listener); 3649 3650 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3651 EXPECT_EQ(true, WIFEXITED(status)); 3652 EXPECT_EQ(0, WEXITSTATUS(status)); 3653 } 3654 3655 /* 3656 * Check that a pid in a child namespace still shows up as valid in ours. 3657 */ 3658 TEST(user_notification_child_pid_ns) 3659 { 3660 pid_t pid; 3661 int status, listener; 3662 struct seccomp_notif req = {}; 3663 struct seccomp_notif_resp resp = {}; 3664 3665 ASSERT_EQ(unshare(CLONE_NEWUSER | CLONE_NEWPID), 0) { 3666 if (errno == EINVAL) 3667 SKIP(return, "kernel missing CLONE_NEWUSER support"); 3668 }; 3669 3670 listener = user_notif_syscall(__NR_getppid, 3671 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3672 ASSERT_GE(listener, 0); 3673 3674 pid = fork(); 3675 ASSERT_GE(pid, 0); 3676 3677 if (pid == 0) 3678 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3679 3680 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3681 EXPECT_EQ(req.pid, pid); 3682 3683 resp.id = req.id; 3684 resp.error = 0; 3685 resp.val = USER_NOTIF_MAGIC; 3686 3687 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3688 3689 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3690 EXPECT_EQ(true, WIFEXITED(status)); 3691 EXPECT_EQ(0, WEXITSTATUS(status)); 3692 close(listener); 3693 } 3694 3695 /* 3696 * Check that a pid in a sibling (i.e. unrelated) namespace shows up as 0, i.e. 3697 * invalid. 3698 */ 3699 TEST(user_notification_sibling_pid_ns) 3700 { 3701 pid_t pid, pid2; 3702 int status, listener; 3703 struct seccomp_notif req = {}; 3704 struct seccomp_notif_resp resp = {}; 3705 3706 ASSERT_EQ(prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0), 0) { 3707 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3708 } 3709 3710 listener = user_notif_syscall(__NR_getppid, 3711 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3712 ASSERT_GE(listener, 0); 3713 3714 pid = fork(); 3715 ASSERT_GE(pid, 0); 3716 3717 if (pid == 0) { 3718 ASSERT_EQ(unshare(CLONE_NEWPID), 0) { 3719 if (errno == EPERM) 3720 SKIP(return, "CLONE_NEWPID requires CAP_SYS_ADMIN"); 3721 else if (errno == EINVAL) 3722 SKIP(return, "CLONE_NEWPID is invalid (missing CONFIG_PID_NS?)"); 3723 } 3724 3725 pid2 = fork(); 3726 ASSERT_GE(pid2, 0); 3727 3728 if (pid2 == 0) 3729 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3730 3731 EXPECT_EQ(waitpid(pid2, &status, 0), pid2); 3732 EXPECT_EQ(true, WIFEXITED(status)); 3733 EXPECT_EQ(0, WEXITSTATUS(status)); 3734 exit(WEXITSTATUS(status)); 3735 } 3736 3737 /* Create the sibling ns, and sibling in it. */ 3738 ASSERT_EQ(unshare(CLONE_NEWPID), 0) { 3739 if (errno == EPERM) 3740 SKIP(return, "CLONE_NEWPID requires CAP_SYS_ADMIN"); 3741 else if (errno == EINVAL) 3742 SKIP(return, "CLONE_NEWPID is invalid (missing CONFIG_PID_NS?)"); 3743 } 3744 ASSERT_EQ(errno, 0); 3745 3746 pid2 = fork(); 3747 ASSERT_GE(pid2, 0); 3748 3749 if (pid2 == 0) { 3750 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3751 /* 3752 * The pid should be 0, i.e. the task is in some namespace that 3753 * we can't "see". 3754 */ 3755 EXPECT_EQ(req.pid, 0); 3756 3757 resp.id = req.id; 3758 resp.error = 0; 3759 resp.val = USER_NOTIF_MAGIC; 3760 3761 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3762 exit(0); 3763 } 3764 3765 close(listener); 3766 3767 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3768 EXPECT_EQ(true, WIFEXITED(status)); 3769 EXPECT_EQ(0, WEXITSTATUS(status)); 3770 3771 EXPECT_EQ(waitpid(pid2, &status, 0), pid2); 3772 EXPECT_EQ(true, WIFEXITED(status)); 3773 EXPECT_EQ(0, WEXITSTATUS(status)); 3774 } 3775 3776 TEST(user_notification_fault_recv) 3777 { 3778 pid_t pid; 3779 int status, listener; 3780 struct seccomp_notif req = {}; 3781 struct seccomp_notif_resp resp = {}; 3782 3783 ASSERT_EQ(unshare(CLONE_NEWUSER), 0) { 3784 if (errno == EINVAL) 3785 SKIP(return, "kernel missing CLONE_NEWUSER support"); 3786 } 3787 3788 listener = user_notif_syscall(__NR_getppid, 3789 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3790 ASSERT_GE(listener, 0); 3791 3792 pid = fork(); 3793 ASSERT_GE(pid, 0); 3794 3795 if (pid == 0) 3796 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3797 3798 /* Do a bad recv() */ 3799 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, NULL), -1); 3800 EXPECT_EQ(errno, EFAULT); 3801 3802 /* We should still be able to receive this notification, though. */ 3803 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3804 EXPECT_EQ(req.pid, pid); 3805 3806 resp.id = req.id; 3807 resp.error = 0; 3808 resp.val = USER_NOTIF_MAGIC; 3809 3810 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3811 3812 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3813 EXPECT_EQ(true, WIFEXITED(status)); 3814 EXPECT_EQ(0, WEXITSTATUS(status)); 3815 } 3816 3817 TEST(seccomp_get_notif_sizes) 3818 { 3819 struct seccomp_notif_sizes sizes; 3820 3821 ASSERT_EQ(seccomp(SECCOMP_GET_NOTIF_SIZES, 0, &sizes), 0); 3822 EXPECT_EQ(sizes.seccomp_notif, sizeof(struct seccomp_notif)); 3823 EXPECT_EQ(sizes.seccomp_notif_resp, sizeof(struct seccomp_notif_resp)); 3824 } 3825 3826 TEST(user_notification_continue) 3827 { 3828 pid_t pid; 3829 long ret; 3830 int status, listener; 3831 struct seccomp_notif req = {}; 3832 struct seccomp_notif_resp resp = {}; 3833 struct pollfd pollfd; 3834 3835 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3836 ASSERT_EQ(0, ret) { 3837 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3838 } 3839 3840 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER); 3841 ASSERT_GE(listener, 0); 3842 3843 pid = fork(); 3844 ASSERT_GE(pid, 0); 3845 3846 if (pid == 0) { 3847 int dup_fd, pipe_fds[2]; 3848 pid_t self; 3849 3850 ASSERT_GE(pipe(pipe_fds), 0); 3851 3852 dup_fd = dup(pipe_fds[0]); 3853 ASSERT_GE(dup_fd, 0); 3854 EXPECT_NE(pipe_fds[0], dup_fd); 3855 3856 self = getpid(); 3857 ASSERT_EQ(filecmp(self, self, pipe_fds[0], dup_fd), 0); 3858 exit(0); 3859 } 3860 3861 pollfd.fd = listener; 3862 pollfd.events = POLLIN | POLLOUT; 3863 3864 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3865 EXPECT_EQ(pollfd.revents, POLLIN); 3866 3867 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3868 3869 pollfd.fd = listener; 3870 pollfd.events = POLLIN | POLLOUT; 3871 3872 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3873 EXPECT_EQ(pollfd.revents, POLLOUT); 3874 3875 EXPECT_EQ(req.data.nr, __NR_dup); 3876 3877 resp.id = req.id; 3878 resp.flags = SECCOMP_USER_NOTIF_FLAG_CONTINUE; 3879 3880 /* 3881 * Verify that setting SECCOMP_USER_NOTIF_FLAG_CONTINUE enforces other 3882 * args be set to 0. 3883 */ 3884 resp.error = 0; 3885 resp.val = USER_NOTIF_MAGIC; 3886 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3887 EXPECT_EQ(errno, EINVAL); 3888 3889 resp.error = USER_NOTIF_MAGIC; 3890 resp.val = 0; 3891 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3892 EXPECT_EQ(errno, EINVAL); 3893 3894 resp.error = 0; 3895 resp.val = 0; 3896 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0) { 3897 if (errno == EINVAL) 3898 SKIP(goto skip, "Kernel does not support SECCOMP_USER_NOTIF_FLAG_CONTINUE"); 3899 } 3900 3901 skip: 3902 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3903 EXPECT_EQ(true, WIFEXITED(status)); 3904 EXPECT_EQ(0, WEXITSTATUS(status)) { 3905 if (WEXITSTATUS(status) == 2) { 3906 SKIP(return, "Kernel does not support kcmp() syscall"); 3907 return; 3908 } 3909 } 3910 } 3911 3912 TEST(user_notification_filter_empty) 3913 { 3914 pid_t pid; 3915 long ret; 3916 int status; 3917 struct pollfd pollfd; 3918 struct __clone_args args = { 3919 .flags = CLONE_FILES, 3920 .exit_signal = SIGCHLD, 3921 }; 3922 3923 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3924 ASSERT_EQ(0, ret) { 3925 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3926 } 3927 3928 if (__NR_clone3 < 0) 3929 SKIP(return, "Test not built with clone3 support"); 3930 3931 pid = sys_clone3(&args, sizeof(args)); 3932 ASSERT_GE(pid, 0); 3933 3934 if (pid == 0) { 3935 int listener; 3936 3937 listener = user_notif_syscall(__NR_mknodat, SECCOMP_FILTER_FLAG_NEW_LISTENER); 3938 if (listener < 0) 3939 _exit(EXIT_FAILURE); 3940 3941 if (dup2(listener, 200) != 200) 3942 _exit(EXIT_FAILURE); 3943 3944 close(listener); 3945 3946 _exit(EXIT_SUCCESS); 3947 } 3948 3949 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3950 EXPECT_EQ(true, WIFEXITED(status)); 3951 EXPECT_EQ(0, WEXITSTATUS(status)); 3952 3953 /* 3954 * The seccomp filter has become unused so we should be notified once 3955 * the kernel gets around to cleaning up task struct. 3956 */ 3957 pollfd.fd = 200; 3958 pollfd.events = POLLHUP; 3959 3960 EXPECT_GT(poll(&pollfd, 1, 2000), 0); 3961 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0); 3962 } 3963 3964 TEST(user_ioctl_notification_filter_empty) 3965 { 3966 pid_t pid; 3967 long ret; 3968 int status, p[2]; 3969 struct __clone_args args = { 3970 .flags = CLONE_FILES, 3971 .exit_signal = SIGCHLD, 3972 }; 3973 struct seccomp_notif req = {}; 3974 3975 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3976 ASSERT_EQ(0, ret) { 3977 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3978 } 3979 3980 if (__NR_clone3 < 0) 3981 SKIP(return, "Test not built with clone3 support"); 3982 3983 ASSERT_EQ(0, pipe(p)); 3984 3985 pid = sys_clone3(&args, sizeof(args)); 3986 ASSERT_GE(pid, 0); 3987 3988 if (pid == 0) { 3989 int listener; 3990 3991 listener = user_notif_syscall(__NR_mknodat, SECCOMP_FILTER_FLAG_NEW_LISTENER); 3992 if (listener < 0) 3993 _exit(EXIT_FAILURE); 3994 3995 if (dup2(listener, 200) != 200) 3996 _exit(EXIT_FAILURE); 3997 close(p[1]); 3998 close(listener); 3999 sleep(1); 4000 4001 _exit(EXIT_SUCCESS); 4002 } 4003 if (read(p[0], &status, 1) != 0) 4004 _exit(EXIT_SUCCESS); 4005 close(p[0]); 4006 /* 4007 * The seccomp filter has become unused so we should be notified once 4008 * the kernel gets around to cleaning up task struct. 4009 */ 4010 EXPECT_EQ(ioctl(200, SECCOMP_IOCTL_NOTIF_RECV, &req), -1); 4011 EXPECT_EQ(errno, ENOENT); 4012 4013 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4014 EXPECT_EQ(true, WIFEXITED(status)); 4015 EXPECT_EQ(0, WEXITSTATUS(status)); 4016 } 4017 4018 static void *do_thread(void *data) 4019 { 4020 return NULL; 4021 } 4022 4023 TEST(user_notification_filter_empty_threaded) 4024 { 4025 pid_t pid; 4026 long ret; 4027 int status; 4028 struct pollfd pollfd; 4029 struct __clone_args args = { 4030 .flags = CLONE_FILES, 4031 .exit_signal = SIGCHLD, 4032 }; 4033 4034 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4035 ASSERT_EQ(0, ret) { 4036 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4037 } 4038 4039 if (__NR_clone3 < 0) 4040 SKIP(return, "Test not built with clone3 support"); 4041 4042 pid = sys_clone3(&args, sizeof(args)); 4043 ASSERT_GE(pid, 0); 4044 4045 if (pid == 0) { 4046 pid_t pid1, pid2; 4047 int listener, status; 4048 pthread_t thread; 4049 4050 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER); 4051 if (listener < 0) 4052 _exit(EXIT_FAILURE); 4053 4054 if (dup2(listener, 200) != 200) 4055 _exit(EXIT_FAILURE); 4056 4057 close(listener); 4058 4059 pid1 = fork(); 4060 if (pid1 < 0) 4061 _exit(EXIT_FAILURE); 4062 4063 if (pid1 == 0) 4064 _exit(EXIT_SUCCESS); 4065 4066 pid2 = fork(); 4067 if (pid2 < 0) 4068 _exit(EXIT_FAILURE); 4069 4070 if (pid2 == 0) 4071 _exit(EXIT_SUCCESS); 4072 4073 if (pthread_create(&thread, NULL, do_thread, NULL) || 4074 pthread_join(thread, NULL)) 4075 _exit(EXIT_FAILURE); 4076 4077 if (pthread_create(&thread, NULL, do_thread, NULL) || 4078 pthread_join(thread, NULL)) 4079 _exit(EXIT_FAILURE); 4080 4081 if (waitpid(pid1, &status, 0) != pid1 || !WIFEXITED(status) || 4082 WEXITSTATUS(status)) 4083 _exit(EXIT_FAILURE); 4084 4085 if (waitpid(pid2, &status, 0) != pid2 || !WIFEXITED(status) || 4086 WEXITSTATUS(status)) 4087 _exit(EXIT_FAILURE); 4088 4089 exit(EXIT_SUCCESS); 4090 } 4091 4092 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4093 EXPECT_EQ(true, WIFEXITED(status)); 4094 EXPECT_EQ(0, WEXITSTATUS(status)); 4095 4096 /* 4097 * The seccomp filter has become unused so we should be notified once 4098 * the kernel gets around to cleaning up task struct. 4099 */ 4100 pollfd.fd = 200; 4101 pollfd.events = POLLHUP; 4102 4103 EXPECT_GT(poll(&pollfd, 1, 2000), 0); 4104 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0); 4105 } 4106 4107 4108 int get_next_fd(int prev_fd) 4109 { 4110 for (int i = prev_fd + 1; i < FD_SETSIZE; ++i) { 4111 if (fcntl(i, F_GETFD) == -1) 4112 return i; 4113 } 4114 _exit(EXIT_FAILURE); 4115 } 4116 4117 TEST(user_notification_addfd) 4118 { 4119 pid_t pid; 4120 long ret; 4121 int status, listener, memfd, fd, nextfd; 4122 struct seccomp_notif_addfd addfd = {}; 4123 struct seccomp_notif_addfd_small small = {}; 4124 struct seccomp_notif_addfd_big big = {}; 4125 struct seccomp_notif req = {}; 4126 struct seccomp_notif_resp resp = {}; 4127 /* 100 ms */ 4128 struct timespec delay = { .tv_nsec = 100000000 }; 4129 4130 /* There may be arbitrary already-open fds at test start. */ 4131 memfd = memfd_create("test", 0); 4132 ASSERT_GE(memfd, 0); 4133 nextfd = get_next_fd(memfd); 4134 4135 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4136 ASSERT_EQ(0, ret) { 4137 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4138 } 4139 4140 /* fd: 4 */ 4141 /* Check that the basic notification machinery works */ 4142 listener = user_notif_syscall(__NR_getppid, 4143 SECCOMP_FILTER_FLAG_NEW_LISTENER); 4144 ASSERT_EQ(listener, nextfd); 4145 nextfd = get_next_fd(nextfd); 4146 4147 pid = fork(); 4148 ASSERT_GE(pid, 0); 4149 4150 if (pid == 0) { 4151 /* fds will be added and this value is expected */ 4152 if (syscall(__NR_getppid) != USER_NOTIF_MAGIC) 4153 exit(1); 4154 4155 /* Atomic addfd+send is received here. Check it is a valid fd */ 4156 if (fcntl(syscall(__NR_getppid), F_GETFD) == -1) 4157 exit(1); 4158 4159 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 4160 } 4161 4162 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4163 4164 addfd.srcfd = memfd; 4165 addfd.newfd = 0; 4166 addfd.id = req.id; 4167 addfd.flags = 0x0; 4168 4169 /* Verify bad newfd_flags cannot be set */ 4170 addfd.newfd_flags = ~O_CLOEXEC; 4171 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4172 EXPECT_EQ(errno, EINVAL); 4173 addfd.newfd_flags = O_CLOEXEC; 4174 4175 /* Verify bad flags cannot be set */ 4176 addfd.flags = 0xff; 4177 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4178 EXPECT_EQ(errno, EINVAL); 4179 addfd.flags = 0; 4180 4181 /* Verify that remote_fd cannot be set without setting flags */ 4182 addfd.newfd = 1; 4183 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4184 EXPECT_EQ(errno, EINVAL); 4185 addfd.newfd = 0; 4186 4187 /* Verify small size cannot be set */ 4188 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_SMALL, &small), -1); 4189 EXPECT_EQ(errno, EINVAL); 4190 4191 /* Verify we can't send bits filled in unknown buffer area */ 4192 memset(&big, 0xAA, sizeof(big)); 4193 big.addfd = addfd; 4194 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big), -1); 4195 EXPECT_EQ(errno, E2BIG); 4196 4197 4198 /* Verify we can set an arbitrary remote fd */ 4199 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd); 4200 EXPECT_EQ(fd, nextfd); 4201 nextfd = get_next_fd(nextfd); 4202 EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0); 4203 4204 /* Verify we can set an arbitrary remote fd with large size */ 4205 memset(&big, 0x0, sizeof(big)); 4206 big.addfd = addfd; 4207 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big); 4208 EXPECT_EQ(fd, nextfd); 4209 nextfd = get_next_fd(nextfd); 4210 4211 /* Verify we can set a specific remote fd */ 4212 addfd.newfd = 42; 4213 addfd.flags = SECCOMP_ADDFD_FLAG_SETFD; 4214 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd); 4215 EXPECT_EQ(fd, 42); 4216 EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0); 4217 4218 /* Resume syscall */ 4219 resp.id = req.id; 4220 resp.error = 0; 4221 resp.val = USER_NOTIF_MAGIC; 4222 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4223 4224 /* 4225 * This sets the ID of the ADD FD to the last request plus 1. The 4226 * notification ID increments 1 per notification. 4227 */ 4228 addfd.id = req.id + 1; 4229 4230 /* This spins until the underlying notification is generated */ 4231 while (ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd) != -1 && 4232 errno != -EINPROGRESS) 4233 nanosleep(&delay, NULL); 4234 4235 memset(&req, 0, sizeof(req)); 4236 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4237 ASSERT_EQ(addfd.id, req.id); 4238 4239 /* Verify we can do an atomic addfd and send */ 4240 addfd.newfd = 0; 4241 addfd.flags = SECCOMP_ADDFD_FLAG_SEND; 4242 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd); 4243 /* 4244 * Child has earlier "low" fds and now 42, so we expect the next 4245 * lowest available fd to be assigned here. 4246 */ 4247 EXPECT_EQ(fd, nextfd); 4248 nextfd = get_next_fd(nextfd); 4249 ASSERT_EQ(filecmp(getpid(), pid, memfd, fd), 0); 4250 4251 /* 4252 * This sets the ID of the ADD FD to the last request plus 1. The 4253 * notification ID increments 1 per notification. 4254 */ 4255 addfd.id = req.id + 1; 4256 4257 /* This spins until the underlying notification is generated */ 4258 while (ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd) != -1 && 4259 errno != -EINPROGRESS) 4260 nanosleep(&delay, NULL); 4261 4262 memset(&req, 0, sizeof(req)); 4263 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4264 ASSERT_EQ(addfd.id, req.id); 4265 4266 resp.id = req.id; 4267 resp.error = 0; 4268 resp.val = USER_NOTIF_MAGIC; 4269 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4270 4271 /* Wait for child to finish. */ 4272 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4273 EXPECT_EQ(true, WIFEXITED(status)); 4274 EXPECT_EQ(0, WEXITSTATUS(status)); 4275 4276 close(memfd); 4277 } 4278 4279 TEST(user_notification_addfd_rlimit) 4280 { 4281 pid_t pid; 4282 long ret; 4283 int status, listener, memfd; 4284 struct seccomp_notif_addfd addfd = {}; 4285 struct seccomp_notif req = {}; 4286 struct seccomp_notif_resp resp = {}; 4287 const struct rlimit lim = { 4288 .rlim_cur = 0, 4289 .rlim_max = 0, 4290 }; 4291 4292 memfd = memfd_create("test", 0); 4293 ASSERT_GE(memfd, 0); 4294 4295 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4296 ASSERT_EQ(0, ret) { 4297 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4298 } 4299 4300 /* Check that the basic notification machinery works */ 4301 listener = user_notif_syscall(__NR_getppid, 4302 SECCOMP_FILTER_FLAG_NEW_LISTENER); 4303 ASSERT_GE(listener, 0); 4304 4305 pid = fork(); 4306 ASSERT_GE(pid, 0); 4307 4308 if (pid == 0) 4309 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 4310 4311 4312 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4313 4314 ASSERT_EQ(prlimit(pid, RLIMIT_NOFILE, &lim, NULL), 0); 4315 4316 addfd.srcfd = memfd; 4317 addfd.newfd_flags = O_CLOEXEC; 4318 addfd.newfd = 0; 4319 addfd.id = req.id; 4320 addfd.flags = 0; 4321 4322 /* Should probably spot check /proc/sys/fs/file-nr */ 4323 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4324 EXPECT_EQ(errno, EMFILE); 4325 4326 addfd.flags = SECCOMP_ADDFD_FLAG_SEND; 4327 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4328 EXPECT_EQ(errno, EMFILE); 4329 4330 addfd.newfd = 100; 4331 addfd.flags = SECCOMP_ADDFD_FLAG_SETFD; 4332 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4333 EXPECT_EQ(errno, EBADF); 4334 4335 resp.id = req.id; 4336 resp.error = 0; 4337 resp.val = USER_NOTIF_MAGIC; 4338 4339 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4340 4341 /* Wait for child to finish. */ 4342 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4343 EXPECT_EQ(true, WIFEXITED(status)); 4344 EXPECT_EQ(0, WEXITSTATUS(status)); 4345 4346 close(memfd); 4347 } 4348 4349 #ifndef SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP 4350 #define SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP (1UL << 0) 4351 #define SECCOMP_IOCTL_NOTIF_SET_FLAGS SECCOMP_IOW(4, __u64) 4352 #endif 4353 4354 TEST(user_notification_sync) 4355 { 4356 struct seccomp_notif req = {}; 4357 struct seccomp_notif_resp resp = {}; 4358 int status, listener; 4359 pid_t pid; 4360 long ret; 4361 4362 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4363 ASSERT_EQ(0, ret) { 4364 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4365 } 4366 4367 listener = user_notif_syscall(__NR_getppid, 4368 SECCOMP_FILTER_FLAG_NEW_LISTENER); 4369 ASSERT_GE(listener, 0); 4370 4371 /* Try to set invalid flags. */ 4372 EXPECT_SYSCALL_RETURN(-EINVAL, 4373 ioctl(listener, SECCOMP_IOCTL_NOTIF_SET_FLAGS, 0xffffffff, 0)); 4374 4375 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SET_FLAGS, 4376 SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP, 0), 0); 4377 4378 pid = fork(); 4379 ASSERT_GE(pid, 0); 4380 if (pid == 0) { 4381 ret = syscall(__NR_getppid); 4382 ASSERT_EQ(ret, USER_NOTIF_MAGIC) { 4383 _exit(1); 4384 } 4385 _exit(0); 4386 } 4387 4388 req.pid = 0; 4389 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4390 4391 ASSERT_EQ(req.data.nr, __NR_getppid); 4392 4393 resp.id = req.id; 4394 resp.error = 0; 4395 resp.val = USER_NOTIF_MAGIC; 4396 resp.flags = 0; 4397 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4398 4399 ASSERT_EQ(waitpid(pid, &status, 0), pid); 4400 ASSERT_EQ(status, 0); 4401 } 4402 4403 4404 /* Make sure PTRACE_O_SUSPEND_SECCOMP requires CAP_SYS_ADMIN. */ 4405 FIXTURE(O_SUSPEND_SECCOMP) { 4406 pid_t pid; 4407 }; 4408 4409 FIXTURE_SETUP(O_SUSPEND_SECCOMP) 4410 { 4411 ERRNO_FILTER(block_read, E2BIG); 4412 cap_value_t cap_list[] = { CAP_SYS_ADMIN }; 4413 cap_t caps; 4414 4415 self->pid = 0; 4416 4417 /* make sure we don't have CAP_SYS_ADMIN */ 4418 caps = cap_get_proc(); 4419 ASSERT_NE(NULL, caps); 4420 ASSERT_EQ(0, cap_set_flag(caps, CAP_EFFECTIVE, 1, cap_list, CAP_CLEAR)); 4421 ASSERT_EQ(0, cap_set_proc(caps)); 4422 cap_free(caps); 4423 4424 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)); 4425 ASSERT_EQ(0, prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_block_read)); 4426 4427 self->pid = fork(); 4428 ASSERT_GE(self->pid, 0); 4429 4430 if (self->pid == 0) { 4431 while (1) 4432 pause(); 4433 _exit(127); 4434 } 4435 } 4436 4437 FIXTURE_TEARDOWN(O_SUSPEND_SECCOMP) 4438 { 4439 if (self->pid) 4440 kill(self->pid, SIGKILL); 4441 } 4442 4443 TEST_F(O_SUSPEND_SECCOMP, setoptions) 4444 { 4445 int wstatus; 4446 4447 ASSERT_EQ(0, ptrace(PTRACE_ATTACH, self->pid, NULL, 0)); 4448 ASSERT_EQ(self->pid, wait(&wstatus)); 4449 ASSERT_EQ(-1, ptrace(PTRACE_SETOPTIONS, self->pid, NULL, PTRACE_O_SUSPEND_SECCOMP)); 4450 if (errno == EINVAL) 4451 SKIP(return, "Kernel does not support PTRACE_O_SUSPEND_SECCOMP (missing CONFIG_CHECKPOINT_RESTORE?)"); 4452 ASSERT_EQ(EPERM, errno); 4453 } 4454 4455 TEST_F(O_SUSPEND_SECCOMP, seize) 4456 { 4457 int ret; 4458 4459 ret = ptrace(PTRACE_SEIZE, self->pid, NULL, PTRACE_O_SUSPEND_SECCOMP); 4460 ASSERT_EQ(-1, ret); 4461 if (errno == EINVAL) 4462 SKIP(return, "Kernel does not support PTRACE_O_SUSPEND_SECCOMP (missing CONFIG_CHECKPOINT_RESTORE?)"); 4463 ASSERT_EQ(EPERM, errno); 4464 } 4465 4466 /* 4467 * get_nth - Get the nth, space separated entry in a file. 4468 * 4469 * Returns the length of the read field. 4470 * Throws error if field is zero-lengthed. 4471 */ 4472 static ssize_t get_nth(struct __test_metadata *_metadata, const char *path, 4473 const unsigned int position, char **entry) 4474 { 4475 char *line = NULL; 4476 unsigned int i; 4477 ssize_t nread; 4478 size_t len = 0; 4479 FILE *f; 4480 4481 f = fopen(path, "r"); 4482 ASSERT_NE(f, NULL) { 4483 TH_LOG("Could not open %s: %s", path, strerror(errno)); 4484 } 4485 4486 for (i = 0; i < position; i++) { 4487 nread = getdelim(&line, &len, ' ', f); 4488 ASSERT_GE(nread, 0) { 4489 TH_LOG("Failed to read %d entry in file %s", i, path); 4490 } 4491 } 4492 fclose(f); 4493 4494 ASSERT_GT(nread, 0) { 4495 TH_LOG("Entry in file %s had zero length", path); 4496 } 4497 4498 *entry = line; 4499 return nread - 1; 4500 } 4501 4502 /* For a given PID, get the task state (D, R, etc...) */ 4503 static char get_proc_stat(struct __test_metadata *_metadata, pid_t pid) 4504 { 4505 char proc_path[100] = {0}; 4506 char status; 4507 char *line; 4508 4509 snprintf(proc_path, sizeof(proc_path), "/proc/%d/stat", pid); 4510 ASSERT_EQ(get_nth(_metadata, proc_path, 3, &line), 1); 4511 4512 status = *line; 4513 free(line); 4514 4515 return status; 4516 } 4517 4518 TEST(user_notification_fifo) 4519 { 4520 struct seccomp_notif_resp resp = {}; 4521 struct seccomp_notif req = {}; 4522 int i, status, listener; 4523 pid_t pid, pids[3]; 4524 __u64 baseid; 4525 long ret; 4526 /* 100 ms */ 4527 struct timespec delay = { .tv_nsec = 100000000 }; 4528 4529 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4530 ASSERT_EQ(0, ret) { 4531 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4532 } 4533 4534 /* Setup a listener */ 4535 listener = user_notif_syscall(__NR_getppid, 4536 SECCOMP_FILTER_FLAG_NEW_LISTENER); 4537 ASSERT_GE(listener, 0); 4538 4539 pid = fork(); 4540 ASSERT_GE(pid, 0); 4541 4542 if (pid == 0) { 4543 ret = syscall(__NR_getppid); 4544 exit(ret != USER_NOTIF_MAGIC); 4545 } 4546 4547 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4548 baseid = req.id + 1; 4549 4550 resp.id = req.id; 4551 resp.error = 0; 4552 resp.val = USER_NOTIF_MAGIC; 4553 4554 /* check that we make sure flags == 0 */ 4555 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4556 4557 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4558 EXPECT_EQ(true, WIFEXITED(status)); 4559 EXPECT_EQ(0, WEXITSTATUS(status)); 4560 4561 /* Start children, and generate notifications */ 4562 for (i = 0; i < ARRAY_SIZE(pids); i++) { 4563 pid = fork(); 4564 if (pid == 0) { 4565 ret = syscall(__NR_getppid); 4566 exit(ret != USER_NOTIF_MAGIC); 4567 } 4568 pids[i] = pid; 4569 } 4570 4571 /* This spins until all of the children are sleeping */ 4572 restart_wait: 4573 for (i = 0; i < ARRAY_SIZE(pids); i++) { 4574 if (get_proc_stat(_metadata, pids[i]) != 'S') { 4575 nanosleep(&delay, NULL); 4576 goto restart_wait; 4577 } 4578 } 4579 4580 /* Read the notifications in order (and respond) */ 4581 for (i = 0; i < ARRAY_SIZE(pids); i++) { 4582 memset(&req, 0, sizeof(req)); 4583 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4584 EXPECT_EQ(req.id, baseid + i); 4585 resp.id = req.id; 4586 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4587 } 4588 4589 /* Make sure notifications were received */ 4590 for (i = 0; i < ARRAY_SIZE(pids); i++) { 4591 EXPECT_EQ(waitpid(pids[i], &status, 0), pids[i]); 4592 EXPECT_EQ(true, WIFEXITED(status)); 4593 EXPECT_EQ(0, WEXITSTATUS(status)); 4594 } 4595 } 4596 4597 /* get_proc_syscall - Get the syscall in progress for a given pid 4598 * 4599 * Returns the current syscall number for a given process 4600 * Returns -1 if not in syscall (running or blocked) 4601 */ 4602 static long get_proc_syscall(struct __test_metadata *_metadata, int pid) 4603 { 4604 char proc_path[100] = {0}; 4605 long ret = -1; 4606 ssize_t nread; 4607 char *line; 4608 4609 snprintf(proc_path, sizeof(proc_path), "/proc/%d/syscall", pid); 4610 nread = get_nth(_metadata, proc_path, 1, &line); 4611 ASSERT_GT(nread, 0); 4612 4613 if (!strncmp("running", line, MIN(7, nread))) 4614 ret = strtol(line, NULL, 16); 4615 4616 free(line); 4617 return ret; 4618 } 4619 4620 /* Ensure non-fatal signals prior to receive are unmodified */ 4621 TEST(user_notification_wait_killable_pre_notification) 4622 { 4623 struct sigaction new_action = { 4624 .sa_handler = signal_handler, 4625 }; 4626 int listener, status, sk_pair[2]; 4627 pid_t pid; 4628 long ret; 4629 char c; 4630 /* 100 ms */ 4631 struct timespec delay = { .tv_nsec = 100000000 }; 4632 4633 ASSERT_EQ(sigemptyset(&new_action.sa_mask), 0); 4634 4635 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4636 ASSERT_EQ(0, ret) 4637 { 4638 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4639 } 4640 4641 ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0); 4642 4643 listener = user_notif_syscall( 4644 __NR_getppid, SECCOMP_FILTER_FLAG_NEW_LISTENER | 4645 SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV); 4646 ASSERT_GE(listener, 0); 4647 4648 /* 4649 * Check that we can kill the process with SIGUSR1 prior to receiving 4650 * the notification. SIGUSR1 is wired up to a custom signal handler, 4651 * and make sure it gets called. 4652 */ 4653 pid = fork(); 4654 ASSERT_GE(pid, 0); 4655 4656 if (pid == 0) { 4657 close(sk_pair[0]); 4658 handled = sk_pair[1]; 4659 4660 /* Setup the non-fatal sigaction without SA_RESTART */ 4661 if (sigaction(SIGUSR1, &new_action, NULL)) { 4662 perror("sigaction"); 4663 exit(1); 4664 } 4665 4666 ret = syscall(__NR_getppid); 4667 /* Make sure we got a return from a signal interruption */ 4668 exit(ret != -1 || errno != EINTR); 4669 } 4670 4671 /* 4672 * Make sure we've gotten to the seccomp user notification wait 4673 * from getppid prior to sending any signals 4674 */ 4675 while (get_proc_syscall(_metadata, pid) != __NR_getppid && 4676 get_proc_stat(_metadata, pid) != 'S') 4677 nanosleep(&delay, NULL); 4678 4679 /* Send non-fatal kill signal */ 4680 EXPECT_EQ(kill(pid, SIGUSR1), 0); 4681 4682 /* wait for process to exit (exit checks for EINTR) */ 4683 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4684 EXPECT_EQ(true, WIFEXITED(status)); 4685 EXPECT_EQ(0, WEXITSTATUS(status)); 4686 4687 EXPECT_EQ(read(sk_pair[0], &c, 1), 1); 4688 } 4689 4690 /* Ensure non-fatal signals after receive are blocked */ 4691 TEST(user_notification_wait_killable) 4692 { 4693 struct sigaction new_action = { 4694 .sa_handler = signal_handler, 4695 }; 4696 struct seccomp_notif_resp resp = {}; 4697 struct seccomp_notif req = {}; 4698 int listener, status, sk_pair[2]; 4699 pid_t pid; 4700 long ret; 4701 char c; 4702 /* 100 ms */ 4703 struct timespec delay = { .tv_nsec = 100000000 }; 4704 4705 ASSERT_EQ(sigemptyset(&new_action.sa_mask), 0); 4706 4707 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4708 ASSERT_EQ(0, ret) 4709 { 4710 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4711 } 4712 4713 ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0); 4714 4715 listener = user_notif_syscall( 4716 __NR_getppid, SECCOMP_FILTER_FLAG_NEW_LISTENER | 4717 SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV); 4718 ASSERT_GE(listener, 0); 4719 4720 pid = fork(); 4721 ASSERT_GE(pid, 0); 4722 4723 if (pid == 0) { 4724 close(sk_pair[0]); 4725 handled = sk_pair[1]; 4726 4727 /* Setup the sigaction without SA_RESTART */ 4728 if (sigaction(SIGUSR1, &new_action, NULL)) { 4729 perror("sigaction"); 4730 exit(1); 4731 } 4732 4733 /* Make sure that the syscall is completed (no EINTR) */ 4734 ret = syscall(__NR_getppid); 4735 exit(ret != USER_NOTIF_MAGIC); 4736 } 4737 4738 /* 4739 * Get the notification, to make move the notifying process into a 4740 * non-preemptible (TASK_KILLABLE) state. 4741 */ 4742 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4743 /* Send non-fatal kill signal */ 4744 EXPECT_EQ(kill(pid, SIGUSR1), 0); 4745 4746 /* 4747 * Make sure the task enters moves to TASK_KILLABLE by waiting for 4748 * D (Disk Sleep) state after receiving non-fatal signal. 4749 */ 4750 while (get_proc_stat(_metadata, pid) != 'D') 4751 nanosleep(&delay, NULL); 4752 4753 resp.id = req.id; 4754 resp.val = USER_NOTIF_MAGIC; 4755 /* Make sure the notification is found and able to be replied to */ 4756 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4757 4758 /* 4759 * Make sure that the signal handler does get called once we're back in 4760 * userspace. 4761 */ 4762 EXPECT_EQ(read(sk_pair[0], &c, 1), 1); 4763 /* wait for process to exit (exit checks for USER_NOTIF_MAGIC) */ 4764 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4765 EXPECT_EQ(true, WIFEXITED(status)); 4766 EXPECT_EQ(0, WEXITSTATUS(status)); 4767 } 4768 4769 /* Ensure fatal signals after receive are not blocked */ 4770 TEST(user_notification_wait_killable_fatal) 4771 { 4772 struct seccomp_notif req = {}; 4773 int listener, status; 4774 pid_t pid; 4775 long ret; 4776 /* 100 ms */ 4777 struct timespec delay = { .tv_nsec = 100000000 }; 4778 4779 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4780 ASSERT_EQ(0, ret) 4781 { 4782 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4783 } 4784 4785 listener = user_notif_syscall( 4786 __NR_getppid, SECCOMP_FILTER_FLAG_NEW_LISTENER | 4787 SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV); 4788 ASSERT_GE(listener, 0); 4789 4790 pid = fork(); 4791 ASSERT_GE(pid, 0); 4792 4793 if (pid == 0) { 4794 /* This should never complete as it should get a SIGTERM */ 4795 syscall(__NR_getppid); 4796 exit(1); 4797 } 4798 4799 while (get_proc_stat(_metadata, pid) != 'S') 4800 nanosleep(&delay, NULL); 4801 4802 /* 4803 * Get the notification, to make move the notifying process into a 4804 * non-preemptible (TASK_KILLABLE) state. 4805 */ 4806 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4807 /* Kill the process with a fatal signal */ 4808 EXPECT_EQ(kill(pid, SIGTERM), 0); 4809 4810 /* 4811 * Wait for the process to exit, and make sure the process terminated 4812 * due to the SIGTERM signal. 4813 */ 4814 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4815 EXPECT_EQ(true, WIFSIGNALED(status)); 4816 EXPECT_EQ(SIGTERM, WTERMSIG(status)); 4817 } 4818 4819 struct tsync_vs_thread_leader_args { 4820 pthread_t leader; 4821 }; 4822 4823 static void *tsync_vs_dead_thread_leader_sibling(void *_args) 4824 { 4825 struct sock_filter allow_filter[] = { 4826 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 4827 }; 4828 struct sock_fprog allow_prog = { 4829 .len = (unsigned short)ARRAY_SIZE(allow_filter), 4830 .filter = allow_filter, 4831 }; 4832 struct tsync_vs_thread_leader_args *args = _args; 4833 void *retval; 4834 long ret; 4835 4836 ret = pthread_join(args->leader, &retval); 4837 if (ret) 4838 exit(1); 4839 if (retval != _args) 4840 exit(2); 4841 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, &allow_prog); 4842 if (ret) 4843 exit(3); 4844 4845 exit(0); 4846 } 4847 4848 /* 4849 * Ensure that a dead thread leader doesn't prevent installing new filters with 4850 * SECCOMP_FILTER_FLAG_TSYNC from other threads. 4851 */ 4852 TEST(tsync_vs_dead_thread_leader) 4853 { 4854 int status; 4855 pid_t pid; 4856 long ret; 4857 4858 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4859 ASSERT_EQ(0, ret) { 4860 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4861 } 4862 4863 pid = fork(); 4864 ASSERT_GE(pid, 0); 4865 4866 if (pid == 0) { 4867 struct sock_filter allow_filter[] = { 4868 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 4869 }; 4870 struct sock_fprog allow_prog = { 4871 .len = (unsigned short)ARRAY_SIZE(allow_filter), 4872 .filter = allow_filter, 4873 }; 4874 struct tsync_vs_thread_leader_args *args; 4875 pthread_t sibling; 4876 4877 args = malloc(sizeof(*args)); 4878 ASSERT_NE(NULL, args); 4879 args->leader = pthread_self(); 4880 4881 ret = pthread_create(&sibling, NULL, 4882 tsync_vs_dead_thread_leader_sibling, args); 4883 ASSERT_EQ(0, ret); 4884 4885 /* Install a new filter just to the leader thread. */ 4886 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog); 4887 ASSERT_EQ(0, ret); 4888 pthread_exit(args); 4889 exit(1); 4890 } 4891 4892 EXPECT_EQ(pid, waitpid(pid, &status, 0)); 4893 EXPECT_EQ(0, status); 4894 } 4895 4896 noinline int probed(void) 4897 { 4898 return 1; 4899 } 4900 4901 static int parse_uint_from_file(const char *file, const char *fmt) 4902 { 4903 int err = -1, ret; 4904 FILE *f; 4905 4906 f = fopen(file, "re"); 4907 if (f) { 4908 err = fscanf(f, fmt, &ret); 4909 fclose(f); 4910 } 4911 return err == 1 ? ret : err; 4912 } 4913 4914 static int determine_uprobe_perf_type(void) 4915 { 4916 const char *file = "/sys/bus/event_source/devices/uprobe/type"; 4917 4918 return parse_uint_from_file(file, "%d\n"); 4919 } 4920 4921 static int determine_uprobe_retprobe_bit(void) 4922 { 4923 const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe"; 4924 4925 return parse_uint_from_file(file, "config:%d\n"); 4926 } 4927 4928 static ssize_t get_uprobe_offset(const void *addr) 4929 { 4930 size_t start, base, end; 4931 bool found = false; 4932 char buf[256]; 4933 FILE *f; 4934 4935 f = fopen("/proc/self/maps", "r"); 4936 if (!f) 4937 return -1; 4938 4939 while (fscanf(f, "%zx-%zx %s %zx %*[^\n]\n", &start, &end, buf, &base) == 4) { 4940 if (buf[2] == 'x' && (uintptr_t)addr >= start && (uintptr_t)addr < end) { 4941 found = true; 4942 break; 4943 } 4944 } 4945 fclose(f); 4946 return found ? (uintptr_t)addr - start + base : -1; 4947 } 4948 4949 FIXTURE(URETPROBE) { 4950 int fd; 4951 }; 4952 4953 FIXTURE_VARIANT(URETPROBE) { 4954 /* 4955 * All of the URETPROBE behaviors can be tested with either 4956 * uretprobe attached or not 4957 */ 4958 bool attach; 4959 }; 4960 4961 FIXTURE_VARIANT_ADD(URETPROBE, attached) { 4962 .attach = true, 4963 }; 4964 4965 FIXTURE_VARIANT_ADD(URETPROBE, not_attached) { 4966 .attach = false, 4967 }; 4968 4969 FIXTURE_SETUP(URETPROBE) 4970 { 4971 const size_t attr_sz = sizeof(struct perf_event_attr); 4972 struct perf_event_attr attr; 4973 ssize_t offset; 4974 int type, bit; 4975 4976 #ifndef __NR_uretprobe 4977 SKIP(return, "__NR_uretprobe syscall not defined"); 4978 #endif 4979 4980 if (!variant->attach) 4981 return; 4982 4983 memset(&attr, 0, attr_sz); 4984 4985 type = determine_uprobe_perf_type(); 4986 ASSERT_GE(type, 0); 4987 bit = determine_uprobe_retprobe_bit(); 4988 ASSERT_GE(bit, 0); 4989 offset = get_uprobe_offset(probed); 4990 ASSERT_GE(offset, 0); 4991 4992 attr.config |= 1 << bit; 4993 attr.size = attr_sz; 4994 attr.type = type; 4995 attr.config1 = ptr_to_u64("/proc/self/exe"); 4996 attr.config2 = offset; 4997 4998 self->fd = syscall(__NR_perf_event_open, &attr, 4999 getpid() /* pid */, -1 /* cpu */, -1 /* group_fd */, 5000 PERF_FLAG_FD_CLOEXEC); 5001 } 5002 5003 FIXTURE_TEARDOWN(URETPROBE) 5004 { 5005 /* we could call close(self->fd), but we'd need extra filter for 5006 * that and since we are calling _exit right away.. 5007 */ 5008 } 5009 5010 static int run_probed_with_filter(struct sock_fprog *prog) 5011 { 5012 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) || 5013 seccomp(SECCOMP_SET_MODE_FILTER, 0, prog)) { 5014 return -1; 5015 } 5016 5017 probed(); 5018 return 0; 5019 } 5020 5021 TEST_F(URETPROBE, uretprobe_default_allow) 5022 { 5023 struct sock_filter filter[] = { 5024 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 5025 }; 5026 struct sock_fprog prog = { 5027 .len = (unsigned short)ARRAY_SIZE(filter), 5028 .filter = filter, 5029 }; 5030 5031 ASSERT_EQ(0, run_probed_with_filter(&prog)); 5032 } 5033 5034 TEST_F(URETPROBE, uretprobe_default_block) 5035 { 5036 struct sock_filter filter[] = { 5037 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 5038 offsetof(struct seccomp_data, nr)), 5039 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit_group, 1, 0), 5040 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 5041 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 5042 }; 5043 struct sock_fprog prog = { 5044 .len = (unsigned short)ARRAY_SIZE(filter), 5045 .filter = filter, 5046 }; 5047 5048 ASSERT_EQ(0, run_probed_with_filter(&prog)); 5049 } 5050 5051 TEST_F(URETPROBE, uretprobe_block_uretprobe_syscall) 5052 { 5053 struct sock_filter filter[] = { 5054 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 5055 offsetof(struct seccomp_data, nr)), 5056 #ifdef __NR_uretprobe 5057 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_uretprobe, 0, 1), 5058 #endif 5059 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 5060 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 5061 }; 5062 struct sock_fprog prog = { 5063 .len = (unsigned short)ARRAY_SIZE(filter), 5064 .filter = filter, 5065 }; 5066 5067 ASSERT_EQ(0, run_probed_with_filter(&prog)); 5068 } 5069 5070 TEST_F(URETPROBE, uretprobe_default_block_with_uretprobe_syscall) 5071 { 5072 struct sock_filter filter[] = { 5073 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 5074 offsetof(struct seccomp_data, nr)), 5075 #ifdef __NR_uretprobe 5076 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_uretprobe, 2, 0), 5077 #endif 5078 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit_group, 1, 0), 5079 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 5080 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 5081 }; 5082 struct sock_fprog prog = { 5083 .len = (unsigned short)ARRAY_SIZE(filter), 5084 .filter = filter, 5085 }; 5086 5087 ASSERT_EQ(0, run_probed_with_filter(&prog)); 5088 } 5089 5090 /* 5091 * TODO: 5092 * - expand NNP testing 5093 * - better arch-specific TRACE and TRAP handlers. 5094 * - endianness checking when appropriate 5095 * - 64-bit arg prodding 5096 * - arch value testing (x86 modes especially) 5097 * - verify that FILTER_FLAG_LOG filters generate log messages 5098 * - verify that RET_LOG generates log messages 5099 */ 5100 5101 TEST_HARNESS_MAIN 5102