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/time.h> 28 #include <sys/user.h> 29 #include <linux/prctl.h> 30 #include <linux/ptrace.h> 31 #include <linux/seccomp.h> 32 #include <pthread.h> 33 #include <semaphore.h> 34 #include <signal.h> 35 #include <stddef.h> 36 #include <stdbool.h> 37 #include <string.h> 38 #include <time.h> 39 #include <limits.h> 40 #include <linux/elf.h> 41 #include <sys/uio.h> 42 #include <sys/utsname.h> 43 #include <sys/fcntl.h> 44 #include <sys/mman.h> 45 #include <sys/times.h> 46 #include <sys/socket.h> 47 #include <sys/ioctl.h> 48 #include <linux/kcmp.h> 49 #include <sys/resource.h> 50 #include <sys/capability.h> 51 #include <linux/perf_event.h> 52 53 #include <unistd.h> 54 #include <sys/syscall.h> 55 #include <poll.h> 56 57 #include "../kselftest_harness.h" 58 #include "../clone3/clone3_selftests.h" 59 60 /* Attempt to de-conflict with the selftests tree. */ 61 #ifndef SKIP 62 #define SKIP(s, ...) XFAIL(s, ##__VA_ARGS__) 63 #endif 64 65 #ifndef MIN 66 #define MIN(X, Y) ((X) < (Y) ? (X) : (Y)) 67 #endif 68 69 #ifndef PR_SET_PTRACER 70 # define PR_SET_PTRACER 0x59616d61 71 #endif 72 73 #ifndef noinline 74 #define noinline __attribute__((noinline)) 75 #endif 76 77 #ifndef PR_SET_NO_NEW_PRIVS 78 #define PR_SET_NO_NEW_PRIVS 38 79 #define PR_GET_NO_NEW_PRIVS 39 80 #endif 81 82 #ifndef PR_SECCOMP_EXT 83 #define PR_SECCOMP_EXT 43 84 #endif 85 86 #ifndef SECCOMP_EXT_ACT 87 #define SECCOMP_EXT_ACT 1 88 #endif 89 90 #ifndef SECCOMP_EXT_ACT_TSYNC 91 #define SECCOMP_EXT_ACT_TSYNC 1 92 #endif 93 94 #ifndef SECCOMP_MODE_STRICT 95 #define SECCOMP_MODE_STRICT 1 96 #endif 97 98 #ifndef SECCOMP_MODE_FILTER 99 #define SECCOMP_MODE_FILTER 2 100 #endif 101 102 #ifndef SECCOMP_RET_ALLOW 103 struct seccomp_data { 104 int nr; 105 __u32 arch; 106 __u64 instruction_pointer; 107 __u64 args[6]; 108 }; 109 #endif 110 111 #ifndef SECCOMP_RET_KILL_PROCESS 112 #define SECCOMP_RET_KILL_PROCESS 0x80000000U /* kill the process */ 113 #define SECCOMP_RET_KILL_THREAD 0x00000000U /* kill the thread */ 114 #endif 115 #ifndef SECCOMP_RET_KILL 116 #define SECCOMP_RET_KILL SECCOMP_RET_KILL_THREAD 117 #define SECCOMP_RET_TRAP 0x00030000U /* disallow and force a SIGSYS */ 118 #define SECCOMP_RET_ERRNO 0x00050000U /* returns an errno */ 119 #define SECCOMP_RET_TRACE 0x7ff00000U /* pass to a tracer or disallow */ 120 #define SECCOMP_RET_ALLOW 0x7fff0000U /* allow */ 121 #endif 122 #ifndef SECCOMP_RET_LOG 123 #define SECCOMP_RET_LOG 0x7ffc0000U /* allow after logging */ 124 #endif 125 126 #ifndef __NR_seccomp 127 # if defined(__i386__) 128 # define __NR_seccomp 354 129 # elif defined(__x86_64__) 130 # define __NR_seccomp 317 131 # elif defined(__arm__) 132 # define __NR_seccomp 383 133 # elif defined(__aarch64__) 134 # define __NR_seccomp 277 135 # elif defined(__riscv) 136 # define __NR_seccomp 277 137 # elif defined(__csky__) 138 # define __NR_seccomp 277 139 # elif defined(__loongarch__) 140 # define __NR_seccomp 277 141 # elif defined(__hppa__) 142 # define __NR_seccomp 338 143 # elif defined(__powerpc__) 144 # define __NR_seccomp 358 145 # elif defined(__s390__) 146 # define __NR_seccomp 348 147 # elif defined(__xtensa__) 148 # define __NR_seccomp 337 149 # elif defined(__sh__) 150 # define __NR_seccomp 372 151 # elif defined(__mc68000__) 152 # define __NR_seccomp 380 153 # else 154 # warning "seccomp syscall number unknown for this architecture" 155 # define __NR_seccomp 0xffff 156 # endif 157 #endif 158 159 #ifndef __NR_uretprobe 160 # if defined(__x86_64__) 161 # define __NR_uretprobe 335 162 # endif 163 #endif 164 165 #ifndef SECCOMP_SET_MODE_STRICT 166 #define SECCOMP_SET_MODE_STRICT 0 167 #endif 168 169 #ifndef SECCOMP_SET_MODE_FILTER 170 #define SECCOMP_SET_MODE_FILTER 1 171 #endif 172 173 #ifndef SECCOMP_GET_ACTION_AVAIL 174 #define SECCOMP_GET_ACTION_AVAIL 2 175 #endif 176 177 #ifndef SECCOMP_GET_NOTIF_SIZES 178 #define SECCOMP_GET_NOTIF_SIZES 3 179 #endif 180 181 #ifndef SECCOMP_FILTER_FLAG_TSYNC 182 #define SECCOMP_FILTER_FLAG_TSYNC (1UL << 0) 183 #endif 184 185 #ifndef SECCOMP_FILTER_FLAG_LOG 186 #define SECCOMP_FILTER_FLAG_LOG (1UL << 1) 187 #endif 188 189 #ifndef SECCOMP_FILTER_FLAG_SPEC_ALLOW 190 #define SECCOMP_FILTER_FLAG_SPEC_ALLOW (1UL << 2) 191 #endif 192 193 #ifndef PTRACE_SECCOMP_GET_METADATA 194 #define PTRACE_SECCOMP_GET_METADATA 0x420d 195 196 struct seccomp_metadata { 197 __u64 filter_off; /* Input: which filter */ 198 __u64 flags; /* Output: filter's flags */ 199 }; 200 #endif 201 202 #ifndef SECCOMP_FILTER_FLAG_NEW_LISTENER 203 #define SECCOMP_FILTER_FLAG_NEW_LISTENER (1UL << 3) 204 #endif 205 206 #ifndef SECCOMP_RET_USER_NOTIF 207 #define SECCOMP_RET_USER_NOTIF 0x7fc00000U 208 209 #define SECCOMP_IOC_MAGIC '!' 210 #define SECCOMP_IO(nr) _IO(SECCOMP_IOC_MAGIC, nr) 211 #define SECCOMP_IOR(nr, type) _IOR(SECCOMP_IOC_MAGIC, nr, type) 212 #define SECCOMP_IOW(nr, type) _IOW(SECCOMP_IOC_MAGIC, nr, type) 213 #define SECCOMP_IOWR(nr, type) _IOWR(SECCOMP_IOC_MAGIC, nr, type) 214 215 /* Flags for seccomp notification fd ioctl. */ 216 #define SECCOMP_IOCTL_NOTIF_RECV SECCOMP_IOWR(0, struct seccomp_notif) 217 #define SECCOMP_IOCTL_NOTIF_SEND SECCOMP_IOWR(1, \ 218 struct seccomp_notif_resp) 219 #define SECCOMP_IOCTL_NOTIF_ID_VALID SECCOMP_IOW(2, __u64) 220 221 struct seccomp_notif { 222 __u64 id; 223 __u32 pid; 224 __u32 flags; 225 struct seccomp_data data; 226 }; 227 228 struct seccomp_notif_resp { 229 __u64 id; 230 __s64 val; 231 __s32 error; 232 __u32 flags; 233 }; 234 235 struct seccomp_notif_sizes { 236 __u16 seccomp_notif; 237 __u16 seccomp_notif_resp; 238 __u16 seccomp_data; 239 }; 240 #endif 241 242 #ifndef SECCOMP_IOCTL_NOTIF_ADDFD 243 /* On success, the return value is the remote process's added fd number */ 244 #define SECCOMP_IOCTL_NOTIF_ADDFD SECCOMP_IOW(3, \ 245 struct seccomp_notif_addfd) 246 247 /* valid flags for seccomp_notif_addfd */ 248 #define SECCOMP_ADDFD_FLAG_SETFD (1UL << 0) /* Specify remote fd */ 249 250 struct seccomp_notif_addfd { 251 __u64 id; 252 __u32 flags; 253 __u32 srcfd; 254 __u32 newfd; 255 __u32 newfd_flags; 256 }; 257 #endif 258 259 #ifndef SECCOMP_ADDFD_FLAG_SEND 260 #define SECCOMP_ADDFD_FLAG_SEND (1UL << 1) /* Addfd and return it, atomically */ 261 #endif 262 263 struct seccomp_notif_addfd_small { 264 __u64 id; 265 char weird[4]; 266 }; 267 #define SECCOMP_IOCTL_NOTIF_ADDFD_SMALL \ 268 SECCOMP_IOW(3, struct seccomp_notif_addfd_small) 269 270 struct seccomp_notif_addfd_big { 271 union { 272 struct seccomp_notif_addfd addfd; 273 char buf[sizeof(struct seccomp_notif_addfd) + 8]; 274 }; 275 }; 276 #define SECCOMP_IOCTL_NOTIF_ADDFD_BIG \ 277 SECCOMP_IOWR(3, struct seccomp_notif_addfd_big) 278 279 #ifndef PTRACE_EVENTMSG_SYSCALL_ENTRY 280 #define PTRACE_EVENTMSG_SYSCALL_ENTRY 1 281 #define PTRACE_EVENTMSG_SYSCALL_EXIT 2 282 #endif 283 284 #ifndef SECCOMP_USER_NOTIF_FLAG_CONTINUE 285 #define SECCOMP_USER_NOTIF_FLAG_CONTINUE 0x00000001 286 #endif 287 288 #ifndef SECCOMP_FILTER_FLAG_TSYNC_ESRCH 289 #define SECCOMP_FILTER_FLAG_TSYNC_ESRCH (1UL << 4) 290 #endif 291 292 #ifndef SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV 293 #define SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV (1UL << 5) 294 #endif 295 296 #ifndef seccomp 297 int seccomp(unsigned int op, unsigned int flags, void *args) 298 { 299 errno = 0; 300 return syscall(__NR_seccomp, op, flags, args); 301 } 302 #endif 303 304 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 305 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n])) 306 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 307 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32)) 308 #else 309 #error "wut? Unknown __BYTE_ORDER__?!" 310 #endif 311 312 #define SIBLING_EXIT_UNKILLED 0xbadbeef 313 #define SIBLING_EXIT_FAILURE 0xbadface 314 #define SIBLING_EXIT_NEWPRIVS 0xbadfeed 315 316 static int __filecmp(pid_t pid1, pid_t pid2, int fd1, int fd2) 317 { 318 #ifdef __NR_kcmp 319 errno = 0; 320 return syscall(__NR_kcmp, pid1, pid2, KCMP_FILE, fd1, fd2); 321 #else 322 errno = ENOSYS; 323 return -1; 324 #endif 325 } 326 327 /* Have TH_LOG report actual location filecmp() is used. */ 328 #define filecmp(pid1, pid2, fd1, fd2) ({ \ 329 int _ret; \ 330 \ 331 _ret = __filecmp(pid1, pid2, fd1, fd2); \ 332 if (_ret != 0) { \ 333 if (_ret < 0 && errno == ENOSYS) { \ 334 TH_LOG("kcmp() syscall missing (test is less accurate)");\ 335 _ret = 0; \ 336 } \ 337 } \ 338 _ret; }) 339 340 TEST(kcmp) 341 { 342 int ret; 343 344 ret = __filecmp(getpid(), getpid(), 1, 1); 345 EXPECT_EQ(ret, 0); 346 if (ret != 0 && errno == ENOSYS) 347 SKIP(return, "Kernel does not support kcmp() (missing CONFIG_KCMP?)"); 348 } 349 350 TEST(mode_strict_support) 351 { 352 long ret; 353 354 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL); 355 ASSERT_EQ(0, ret) { 356 TH_LOG("Kernel does not support CONFIG_SECCOMP"); 357 } 358 syscall(__NR_exit, 0); 359 } 360 361 TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL) 362 { 363 long ret; 364 365 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL); 366 ASSERT_EQ(0, ret) { 367 TH_LOG("Kernel does not support CONFIG_SECCOMP"); 368 } 369 syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER, 370 NULL, NULL, NULL); 371 EXPECT_FALSE(true) { 372 TH_LOG("Unreachable!"); 373 } 374 } 375 376 /* Note! This doesn't test no new privs behavior */ 377 TEST(no_new_privs_support) 378 { 379 long ret; 380 381 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 382 EXPECT_EQ(0, ret) { 383 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 384 } 385 } 386 387 /* Tests kernel support by checking for a copy_from_user() fault on NULL. */ 388 TEST(mode_filter_support) 389 { 390 long ret; 391 392 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 393 ASSERT_EQ(0, ret) { 394 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 395 } 396 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL); 397 EXPECT_EQ(-1, ret); 398 EXPECT_EQ(EFAULT, errno) { 399 TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!"); 400 } 401 } 402 403 TEST(mode_filter_without_nnp) 404 { 405 struct sock_filter filter[] = { 406 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 407 }; 408 struct sock_fprog prog = { 409 .len = (unsigned short)ARRAY_SIZE(filter), 410 .filter = filter, 411 }; 412 long ret; 413 cap_t cap = cap_get_proc(); 414 cap_flag_value_t is_cap_sys_admin = 0; 415 416 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0); 417 ASSERT_LE(0, ret) { 418 TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS"); 419 } 420 errno = 0; 421 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 422 /* Succeeds with CAP_SYS_ADMIN, fails without */ 423 cap_get_flag(cap, CAP_SYS_ADMIN, CAP_EFFECTIVE, &is_cap_sys_admin); 424 if (!is_cap_sys_admin) { 425 EXPECT_EQ(-1, ret); 426 EXPECT_EQ(EACCES, errno); 427 } else { 428 EXPECT_EQ(0, ret); 429 } 430 } 431 432 #define MAX_INSNS_PER_PATH 32768 433 434 TEST(filter_size_limits) 435 { 436 int i; 437 int count = BPF_MAXINSNS + 1; 438 struct sock_filter allow[] = { 439 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 440 }; 441 struct sock_filter *filter; 442 struct sock_fprog prog = { }; 443 long ret; 444 445 filter = calloc(count, sizeof(*filter)); 446 ASSERT_NE(NULL, filter); 447 448 for (i = 0; i < count; i++) 449 filter[i] = allow[0]; 450 451 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 452 ASSERT_EQ(0, ret); 453 454 prog.filter = filter; 455 prog.len = count; 456 457 /* Too many filter instructions in a single filter. */ 458 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 459 ASSERT_NE(0, ret) { 460 TH_LOG("Installing %d insn filter was allowed", prog.len); 461 } 462 463 /* One less is okay, though. */ 464 prog.len -= 1; 465 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 466 ASSERT_EQ(0, ret) { 467 TH_LOG("Installing %d insn filter wasn't allowed", prog.len); 468 } 469 } 470 471 TEST(filter_chain_limits) 472 { 473 int i; 474 int count = BPF_MAXINSNS; 475 struct sock_filter allow[] = { 476 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 477 }; 478 struct sock_filter *filter; 479 struct sock_fprog prog = { }; 480 long ret; 481 482 filter = calloc(count, sizeof(*filter)); 483 ASSERT_NE(NULL, filter); 484 485 for (i = 0; i < count; i++) 486 filter[i] = allow[0]; 487 488 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 489 ASSERT_EQ(0, ret); 490 491 prog.filter = filter; 492 prog.len = 1; 493 494 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 495 ASSERT_EQ(0, ret); 496 497 prog.len = count; 498 499 /* Too many total filter instructions. */ 500 for (i = 0; i < MAX_INSNS_PER_PATH; i++) { 501 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 502 if (ret != 0) 503 break; 504 } 505 ASSERT_NE(0, ret) { 506 TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)", 507 i, count, i * (count + 4)); 508 } 509 } 510 511 TEST(mode_filter_cannot_move_to_strict) 512 { 513 struct sock_filter filter[] = { 514 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 515 }; 516 struct sock_fprog prog = { 517 .len = (unsigned short)ARRAY_SIZE(filter), 518 .filter = filter, 519 }; 520 long ret; 521 522 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 523 ASSERT_EQ(0, ret); 524 525 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 526 ASSERT_EQ(0, ret); 527 528 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0); 529 EXPECT_EQ(-1, ret); 530 EXPECT_EQ(EINVAL, errno); 531 } 532 533 534 TEST(mode_filter_get_seccomp) 535 { 536 struct sock_filter filter[] = { 537 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 538 }; 539 struct sock_fprog prog = { 540 .len = (unsigned short)ARRAY_SIZE(filter), 541 .filter = filter, 542 }; 543 long ret; 544 545 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 546 ASSERT_EQ(0, ret); 547 548 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0); 549 EXPECT_EQ(0, ret); 550 551 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 552 ASSERT_EQ(0, ret); 553 554 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0); 555 EXPECT_EQ(2, ret); 556 } 557 558 559 TEST(ALLOW_all) 560 { 561 struct sock_filter filter[] = { 562 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 563 }; 564 struct sock_fprog prog = { 565 .len = (unsigned short)ARRAY_SIZE(filter), 566 .filter = filter, 567 }; 568 long ret; 569 570 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 571 ASSERT_EQ(0, ret); 572 573 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 574 ASSERT_EQ(0, ret); 575 } 576 577 TEST(empty_prog) 578 { 579 struct sock_filter filter[] = { 580 }; 581 struct sock_fprog prog = { 582 .len = (unsigned short)ARRAY_SIZE(filter), 583 .filter = filter, 584 }; 585 long ret; 586 587 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 588 ASSERT_EQ(0, ret); 589 590 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 591 EXPECT_EQ(-1, ret); 592 EXPECT_EQ(EINVAL, errno); 593 } 594 595 TEST(log_all) 596 { 597 struct sock_filter filter[] = { 598 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG), 599 }; 600 struct sock_fprog prog = { 601 .len = (unsigned short)ARRAY_SIZE(filter), 602 .filter = filter, 603 }; 604 long ret; 605 pid_t parent = getppid(); 606 607 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 608 ASSERT_EQ(0, ret); 609 610 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 611 ASSERT_EQ(0, ret); 612 613 /* getppid() should succeed and be logged (no check for logging) */ 614 EXPECT_EQ(parent, syscall(__NR_getppid)); 615 } 616 617 TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS) 618 { 619 struct sock_filter filter[] = { 620 BPF_STMT(BPF_RET|BPF_K, 0x10000000U), 621 }; 622 struct sock_fprog prog = { 623 .len = (unsigned short)ARRAY_SIZE(filter), 624 .filter = filter, 625 }; 626 long ret; 627 628 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 629 ASSERT_EQ(0, ret); 630 631 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 632 ASSERT_EQ(0, ret); 633 EXPECT_EQ(0, syscall(__NR_getpid)) { 634 TH_LOG("getpid() shouldn't ever return"); 635 } 636 } 637 638 /* return code >= 0x80000000 is unused. */ 639 TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS) 640 { 641 struct sock_filter filter[] = { 642 BPF_STMT(BPF_RET|BPF_K, 0x90000000U), 643 }; 644 struct sock_fprog prog = { 645 .len = (unsigned short)ARRAY_SIZE(filter), 646 .filter = filter, 647 }; 648 long ret; 649 650 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 651 ASSERT_EQ(0, ret); 652 653 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 654 ASSERT_EQ(0, ret); 655 EXPECT_EQ(0, syscall(__NR_getpid)) { 656 TH_LOG("getpid() shouldn't ever return"); 657 } 658 } 659 660 TEST_SIGNAL(KILL_all, SIGSYS) 661 { 662 struct sock_filter filter[] = { 663 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 664 }; 665 struct sock_fprog prog = { 666 .len = (unsigned short)ARRAY_SIZE(filter), 667 .filter = filter, 668 }; 669 long ret; 670 671 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 672 ASSERT_EQ(0, ret); 673 674 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 675 ASSERT_EQ(0, ret); 676 } 677 678 TEST_SIGNAL(KILL_one, SIGSYS) 679 { 680 struct sock_filter filter[] = { 681 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 682 offsetof(struct seccomp_data, nr)), 683 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 684 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 685 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 686 }; 687 struct sock_fprog prog = { 688 .len = (unsigned short)ARRAY_SIZE(filter), 689 .filter = filter, 690 }; 691 long ret; 692 pid_t parent = getppid(); 693 694 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 695 ASSERT_EQ(0, ret); 696 697 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 698 ASSERT_EQ(0, ret); 699 700 EXPECT_EQ(parent, syscall(__NR_getppid)); 701 /* getpid() should never return. */ 702 EXPECT_EQ(0, syscall(__NR_getpid)); 703 } 704 705 TEST_SIGNAL(KILL_one_arg_one, SIGSYS) 706 { 707 void *fatal_address; 708 struct sock_filter filter[] = { 709 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 710 offsetof(struct seccomp_data, nr)), 711 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0), 712 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 713 /* Only both with lower 32-bit for now. */ 714 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)), 715 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 716 (unsigned long)&fatal_address, 0, 1), 717 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 718 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 719 }; 720 struct sock_fprog prog = { 721 .len = (unsigned short)ARRAY_SIZE(filter), 722 .filter = filter, 723 }; 724 long ret; 725 pid_t parent = getppid(); 726 struct tms timebuf; 727 clock_t clock = times(&timebuf); 728 729 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 730 ASSERT_EQ(0, ret); 731 732 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 733 ASSERT_EQ(0, ret); 734 735 EXPECT_EQ(parent, syscall(__NR_getppid)); 736 EXPECT_LE(clock, syscall(__NR_times, &timebuf)); 737 /* times() should never return. */ 738 EXPECT_EQ(0, syscall(__NR_times, &fatal_address)); 739 } 740 741 TEST_SIGNAL(KILL_one_arg_six, SIGSYS) 742 { 743 #ifndef __NR_mmap2 744 int sysno = __NR_mmap; 745 #else 746 int sysno = __NR_mmap2; 747 #endif 748 struct sock_filter filter[] = { 749 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 750 offsetof(struct seccomp_data, nr)), 751 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0), 752 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 753 /* Only both with lower 32-bit for now. */ 754 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)), 755 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1), 756 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 757 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 758 }; 759 struct sock_fprog prog = { 760 .len = (unsigned short)ARRAY_SIZE(filter), 761 .filter = filter, 762 }; 763 long ret; 764 pid_t parent = getppid(); 765 int fd; 766 void *map1, *map2; 767 int page_size = sysconf(_SC_PAGESIZE); 768 769 ASSERT_LT(0, page_size); 770 771 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 772 ASSERT_EQ(0, ret); 773 774 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 775 ASSERT_EQ(0, ret); 776 777 fd = open("/dev/zero", O_RDONLY); 778 ASSERT_NE(-1, fd); 779 780 EXPECT_EQ(parent, syscall(__NR_getppid)); 781 map1 = (void *)syscall(sysno, 782 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size); 783 EXPECT_NE(MAP_FAILED, map1); 784 /* mmap2() should never return. */ 785 map2 = (void *)syscall(sysno, 786 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE); 787 EXPECT_EQ(MAP_FAILED, map2); 788 789 /* The test failed, so clean up the resources. */ 790 munmap(map1, page_size); 791 munmap(map2, page_size); 792 close(fd); 793 } 794 795 /* This is a thread task to die via seccomp filter violation. */ 796 void *kill_thread(void *data) 797 { 798 bool die = (bool)data; 799 800 if (die) { 801 syscall(__NR_getpid); 802 return (void *)SIBLING_EXIT_FAILURE; 803 } 804 805 return (void *)SIBLING_EXIT_UNKILLED; 806 } 807 808 enum kill_t { 809 KILL_THREAD, 810 KILL_PROCESS, 811 RET_UNKNOWN 812 }; 813 814 /* Prepare a thread that will kill itself or both of us. */ 815 void kill_thread_or_group(struct __test_metadata *_metadata, 816 enum kill_t kill_how) 817 { 818 pthread_t thread; 819 void *status; 820 /* Kill only when calling __NR_getpid. */ 821 struct sock_filter filter_thread[] = { 822 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 823 offsetof(struct seccomp_data, nr)), 824 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 825 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD), 826 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 827 }; 828 struct sock_fprog prog_thread = { 829 .len = (unsigned short)ARRAY_SIZE(filter_thread), 830 .filter = filter_thread, 831 }; 832 int kill = kill_how == KILL_PROCESS ? SECCOMP_RET_KILL_PROCESS : 0xAAAAAAAA; 833 struct sock_filter filter_process[] = { 834 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 835 offsetof(struct seccomp_data, nr)), 836 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 837 BPF_STMT(BPF_RET|BPF_K, kill), 838 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 839 }; 840 struct sock_fprog prog_process = { 841 .len = (unsigned short)ARRAY_SIZE(filter_process), 842 .filter = filter_process, 843 }; 844 845 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 846 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 847 } 848 849 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, 850 kill_how == KILL_THREAD ? &prog_thread 851 : &prog_process)); 852 853 /* 854 * Add the KILL_THREAD rule again to make sure that the KILL_PROCESS 855 * flag cannot be downgraded by a new filter. 856 */ 857 if (kill_how == KILL_PROCESS) 858 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog_thread)); 859 860 /* Start a thread that will exit immediately. */ 861 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)false)); 862 ASSERT_EQ(0, pthread_join(thread, &status)); 863 ASSERT_EQ(SIBLING_EXIT_UNKILLED, (unsigned long)status); 864 865 /* Start a thread that will die immediately. */ 866 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)true)); 867 ASSERT_EQ(0, pthread_join(thread, &status)); 868 ASSERT_NE(SIBLING_EXIT_FAILURE, (unsigned long)status); 869 870 /* 871 * If we get here, only the spawned thread died. Let the parent know 872 * the whole process didn't die (i.e. this thread, the spawner, 873 * stayed running). 874 */ 875 exit(42); 876 } 877 878 TEST(KILL_thread) 879 { 880 int status; 881 pid_t child_pid; 882 883 child_pid = fork(); 884 ASSERT_LE(0, child_pid); 885 if (child_pid == 0) { 886 kill_thread_or_group(_metadata, KILL_THREAD); 887 _exit(38); 888 } 889 890 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 891 892 /* If only the thread was killed, we'll see exit 42. */ 893 ASSERT_TRUE(WIFEXITED(status)); 894 ASSERT_EQ(42, WEXITSTATUS(status)); 895 } 896 897 TEST(KILL_process) 898 { 899 int status; 900 pid_t child_pid; 901 902 child_pid = fork(); 903 ASSERT_LE(0, child_pid); 904 if (child_pid == 0) { 905 kill_thread_or_group(_metadata, KILL_PROCESS); 906 _exit(38); 907 } 908 909 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 910 911 /* If the entire process was killed, we'll see SIGSYS. */ 912 ASSERT_TRUE(WIFSIGNALED(status)); 913 ASSERT_EQ(SIGSYS, WTERMSIG(status)); 914 } 915 916 TEST(KILL_unknown) 917 { 918 int status; 919 pid_t child_pid; 920 921 child_pid = fork(); 922 ASSERT_LE(0, child_pid); 923 if (child_pid == 0) { 924 kill_thread_or_group(_metadata, RET_UNKNOWN); 925 _exit(38); 926 } 927 928 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 929 930 /* If the entire process was killed, we'll see SIGSYS. */ 931 EXPECT_TRUE(WIFSIGNALED(status)) { 932 TH_LOG("Unknown SECCOMP_RET is only killing the thread?"); 933 } 934 ASSERT_EQ(SIGSYS, WTERMSIG(status)); 935 } 936 937 /* TODO(wad) add 64-bit versus 32-bit arg tests. */ 938 TEST(arg_out_of_range) 939 { 940 struct sock_filter filter[] = { 941 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)), 942 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 943 }; 944 struct sock_fprog prog = { 945 .len = (unsigned short)ARRAY_SIZE(filter), 946 .filter = filter, 947 }; 948 long ret; 949 950 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 951 ASSERT_EQ(0, ret); 952 953 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 954 EXPECT_EQ(-1, ret); 955 EXPECT_EQ(EINVAL, errno); 956 } 957 958 #define ERRNO_FILTER(name, errno) \ 959 struct sock_filter _read_filter_##name[] = { \ 960 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, \ 961 offsetof(struct seccomp_data, nr)), \ 962 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), \ 963 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | errno), \ 964 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), \ 965 }; \ 966 struct sock_fprog prog_##name = { \ 967 .len = (unsigned short)ARRAY_SIZE(_read_filter_##name), \ 968 .filter = _read_filter_##name, \ 969 } 970 971 /* Make sure basic errno values are correctly passed through a filter. */ 972 TEST(ERRNO_valid) 973 { 974 ERRNO_FILTER(valid, E2BIG); 975 long ret; 976 pid_t parent = getppid(); 977 978 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 979 ASSERT_EQ(0, ret); 980 981 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_valid); 982 ASSERT_EQ(0, ret); 983 984 EXPECT_EQ(parent, syscall(__NR_getppid)); 985 EXPECT_EQ(-1, read(-1, NULL, 0)); 986 EXPECT_EQ(E2BIG, errno); 987 } 988 989 /* Make sure an errno of zero is correctly handled by the arch code. */ 990 TEST(ERRNO_zero) 991 { 992 ERRNO_FILTER(zero, 0); 993 long ret; 994 pid_t parent = getppid(); 995 996 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 997 ASSERT_EQ(0, ret); 998 999 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_zero); 1000 ASSERT_EQ(0, ret); 1001 1002 EXPECT_EQ(parent, syscall(__NR_getppid)); 1003 /* "errno" of 0 is ok. */ 1004 EXPECT_EQ(0, read(-1, NULL, 0)); 1005 } 1006 1007 /* 1008 * The SECCOMP_RET_DATA mask is 16 bits wide, but errno is smaller. 1009 * This tests that the errno value gets capped correctly, fixed by 1010 * 580c57f10768 ("seccomp: cap SECCOMP_RET_ERRNO data to MAX_ERRNO"). 1011 */ 1012 TEST(ERRNO_capped) 1013 { 1014 ERRNO_FILTER(capped, 4096); 1015 long ret; 1016 pid_t parent = getppid(); 1017 1018 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1019 ASSERT_EQ(0, ret); 1020 1021 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_capped); 1022 ASSERT_EQ(0, ret); 1023 1024 EXPECT_EQ(parent, syscall(__NR_getppid)); 1025 EXPECT_EQ(-1, read(-1, NULL, 0)); 1026 EXPECT_EQ(4095, errno); 1027 } 1028 1029 /* 1030 * Filters are processed in reverse order: last applied is executed first. 1031 * Since only the SECCOMP_RET_ACTION mask is tested for return values, the 1032 * SECCOMP_RET_DATA mask results will follow the most recently applied 1033 * matching filter return (and not the lowest or highest value). 1034 */ 1035 TEST(ERRNO_order) 1036 { 1037 ERRNO_FILTER(first, 11); 1038 ERRNO_FILTER(second, 13); 1039 ERRNO_FILTER(third, 12); 1040 long ret; 1041 pid_t parent = getppid(); 1042 1043 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1044 ASSERT_EQ(0, ret); 1045 1046 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_first); 1047 ASSERT_EQ(0, ret); 1048 1049 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_second); 1050 ASSERT_EQ(0, ret); 1051 1052 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_third); 1053 ASSERT_EQ(0, ret); 1054 1055 EXPECT_EQ(parent, syscall(__NR_getppid)); 1056 EXPECT_EQ(-1, read(-1, NULL, 0)); 1057 EXPECT_EQ(12, errno); 1058 } 1059 1060 FIXTURE(TRAP) { 1061 struct sock_fprog prog; 1062 }; 1063 1064 FIXTURE_SETUP(TRAP) 1065 { 1066 struct sock_filter filter[] = { 1067 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1068 offsetof(struct seccomp_data, nr)), 1069 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 1070 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP), 1071 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1072 }; 1073 1074 memset(&self->prog, 0, sizeof(self->prog)); 1075 self->prog.filter = malloc(sizeof(filter)); 1076 ASSERT_NE(NULL, self->prog.filter); 1077 memcpy(self->prog.filter, filter, sizeof(filter)); 1078 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 1079 } 1080 1081 FIXTURE_TEARDOWN(TRAP) 1082 { 1083 if (self->prog.filter) 1084 free(self->prog.filter); 1085 } 1086 1087 TEST_F_SIGNAL(TRAP, dfl, SIGSYS) 1088 { 1089 long ret; 1090 1091 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1092 ASSERT_EQ(0, ret); 1093 1094 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 1095 ASSERT_EQ(0, ret); 1096 syscall(__NR_getpid); 1097 } 1098 1099 /* Ensure that SIGSYS overrides SIG_IGN */ 1100 TEST_F_SIGNAL(TRAP, ign, SIGSYS) 1101 { 1102 long ret; 1103 1104 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1105 ASSERT_EQ(0, ret); 1106 1107 signal(SIGSYS, SIG_IGN); 1108 1109 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 1110 ASSERT_EQ(0, ret); 1111 syscall(__NR_getpid); 1112 } 1113 1114 static siginfo_t TRAP_info; 1115 static volatile int TRAP_nr; 1116 static void TRAP_action(int nr, siginfo_t *info, void *void_context) 1117 { 1118 memcpy(&TRAP_info, info, sizeof(TRAP_info)); 1119 TRAP_nr = nr; 1120 } 1121 1122 TEST_F(TRAP, handler) 1123 { 1124 int ret, test; 1125 struct sigaction act; 1126 sigset_t mask; 1127 1128 memset(&act, 0, sizeof(act)); 1129 sigemptyset(&mask); 1130 sigaddset(&mask, SIGSYS); 1131 1132 act.sa_sigaction = &TRAP_action; 1133 act.sa_flags = SA_SIGINFO; 1134 ret = sigaction(SIGSYS, &act, NULL); 1135 ASSERT_EQ(0, ret) { 1136 TH_LOG("sigaction failed"); 1137 } 1138 ret = sigprocmask(SIG_UNBLOCK, &mask, NULL); 1139 ASSERT_EQ(0, ret) { 1140 TH_LOG("sigprocmask failed"); 1141 } 1142 1143 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1144 ASSERT_EQ(0, ret); 1145 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 1146 ASSERT_EQ(0, ret); 1147 TRAP_nr = 0; 1148 memset(&TRAP_info, 0, sizeof(TRAP_info)); 1149 /* Expect the registers to be rolled back. (nr = error) may vary 1150 * based on arch. */ 1151 ret = syscall(__NR_getpid); 1152 /* Silence gcc warning about volatile. */ 1153 test = TRAP_nr; 1154 EXPECT_EQ(SIGSYS, test); 1155 struct local_sigsys { 1156 void *_call_addr; /* calling user insn */ 1157 int _syscall; /* triggering system call number */ 1158 unsigned int _arch; /* AUDIT_ARCH_* of syscall */ 1159 } *sigsys = (struct local_sigsys *) 1160 #ifdef si_syscall 1161 &(TRAP_info.si_call_addr); 1162 #else 1163 &TRAP_info.si_pid; 1164 #endif 1165 EXPECT_EQ(__NR_getpid, sigsys->_syscall); 1166 /* Make sure arch is non-zero. */ 1167 EXPECT_NE(0, sigsys->_arch); 1168 EXPECT_NE(0, (unsigned long)sigsys->_call_addr); 1169 } 1170 1171 FIXTURE(precedence) { 1172 struct sock_fprog allow; 1173 struct sock_fprog log; 1174 struct sock_fprog trace; 1175 struct sock_fprog error; 1176 struct sock_fprog trap; 1177 struct sock_fprog kill; 1178 }; 1179 1180 FIXTURE_SETUP(precedence) 1181 { 1182 struct sock_filter allow_insns[] = { 1183 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1184 }; 1185 struct sock_filter log_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_LOG), 1191 }; 1192 struct sock_filter trace_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_TRACE), 1198 }; 1199 struct sock_filter error_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_ERRNO), 1205 }; 1206 struct sock_filter trap_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_TRAP), 1212 }; 1213 struct sock_filter kill_insns[] = { 1214 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1215 offsetof(struct seccomp_data, nr)), 1216 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1217 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1218 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 1219 }; 1220 1221 memset(self, 0, sizeof(*self)); 1222 #define FILTER_ALLOC(_x) \ 1223 self->_x.filter = malloc(sizeof(_x##_insns)); \ 1224 ASSERT_NE(NULL, self->_x.filter); \ 1225 memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \ 1226 self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns) 1227 FILTER_ALLOC(allow); 1228 FILTER_ALLOC(log); 1229 FILTER_ALLOC(trace); 1230 FILTER_ALLOC(error); 1231 FILTER_ALLOC(trap); 1232 FILTER_ALLOC(kill); 1233 } 1234 1235 FIXTURE_TEARDOWN(precedence) 1236 { 1237 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter) 1238 FILTER_FREE(allow); 1239 FILTER_FREE(log); 1240 FILTER_FREE(trace); 1241 FILTER_FREE(error); 1242 FILTER_FREE(trap); 1243 FILTER_FREE(kill); 1244 } 1245 1246 TEST_F(precedence, allow_ok) 1247 { 1248 pid_t parent, res = 0; 1249 long ret; 1250 1251 parent = getppid(); 1252 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1253 ASSERT_EQ(0, ret); 1254 1255 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1256 ASSERT_EQ(0, ret); 1257 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1258 ASSERT_EQ(0, ret); 1259 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1260 ASSERT_EQ(0, ret); 1261 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1262 ASSERT_EQ(0, ret); 1263 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1264 ASSERT_EQ(0, ret); 1265 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1266 ASSERT_EQ(0, ret); 1267 /* Should work just fine. */ 1268 res = syscall(__NR_getppid); 1269 EXPECT_EQ(parent, res); 1270 } 1271 1272 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS) 1273 { 1274 pid_t parent, res = 0; 1275 long ret; 1276 1277 parent = getppid(); 1278 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1279 ASSERT_EQ(0, ret); 1280 1281 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1282 ASSERT_EQ(0, ret); 1283 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1284 ASSERT_EQ(0, ret); 1285 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1286 ASSERT_EQ(0, ret); 1287 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1288 ASSERT_EQ(0, ret); 1289 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1290 ASSERT_EQ(0, ret); 1291 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1292 ASSERT_EQ(0, ret); 1293 /* Should work just fine. */ 1294 res = syscall(__NR_getppid); 1295 EXPECT_EQ(parent, res); 1296 /* getpid() should never return. */ 1297 res = syscall(__NR_getpid); 1298 EXPECT_EQ(0, res); 1299 } 1300 1301 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS) 1302 { 1303 pid_t parent; 1304 long ret; 1305 1306 parent = getppid(); 1307 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1308 ASSERT_EQ(0, ret); 1309 1310 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1311 ASSERT_EQ(0, ret); 1312 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1313 ASSERT_EQ(0, ret); 1314 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1315 ASSERT_EQ(0, ret); 1316 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1317 ASSERT_EQ(0, ret); 1318 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1319 ASSERT_EQ(0, ret); 1320 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1321 ASSERT_EQ(0, ret); 1322 /* Should work just fine. */ 1323 EXPECT_EQ(parent, syscall(__NR_getppid)); 1324 /* getpid() should never return. */ 1325 EXPECT_EQ(0, syscall(__NR_getpid)); 1326 } 1327 1328 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS) 1329 { 1330 pid_t parent; 1331 long ret; 1332 1333 parent = getppid(); 1334 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1335 ASSERT_EQ(0, ret); 1336 1337 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1338 ASSERT_EQ(0, ret); 1339 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1340 ASSERT_EQ(0, ret); 1341 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1342 ASSERT_EQ(0, ret); 1343 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1344 ASSERT_EQ(0, ret); 1345 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1346 ASSERT_EQ(0, ret); 1347 /* Should work just fine. */ 1348 EXPECT_EQ(parent, syscall(__NR_getppid)); 1349 /* getpid() should never return. */ 1350 EXPECT_EQ(0, syscall(__NR_getpid)); 1351 } 1352 1353 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS) 1354 { 1355 pid_t parent; 1356 long ret; 1357 1358 parent = getppid(); 1359 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1360 ASSERT_EQ(0, ret); 1361 1362 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1363 ASSERT_EQ(0, ret); 1364 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1365 ASSERT_EQ(0, ret); 1366 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1367 ASSERT_EQ(0, ret); 1368 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1369 ASSERT_EQ(0, ret); 1370 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1371 ASSERT_EQ(0, ret); 1372 /* Should work just fine. */ 1373 EXPECT_EQ(parent, syscall(__NR_getppid)); 1374 /* getpid() should never return. */ 1375 EXPECT_EQ(0, syscall(__NR_getpid)); 1376 } 1377 1378 TEST_F(precedence, errno_is_third) 1379 { 1380 pid_t parent; 1381 long ret; 1382 1383 parent = getppid(); 1384 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1385 ASSERT_EQ(0, ret); 1386 1387 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1388 ASSERT_EQ(0, ret); 1389 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1390 ASSERT_EQ(0, ret); 1391 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1392 ASSERT_EQ(0, ret); 1393 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1394 ASSERT_EQ(0, ret); 1395 /* Should work just fine. */ 1396 EXPECT_EQ(parent, syscall(__NR_getppid)); 1397 EXPECT_EQ(0, syscall(__NR_getpid)); 1398 } 1399 1400 TEST_F(precedence, errno_is_third_in_any_order) 1401 { 1402 pid_t parent; 1403 long ret; 1404 1405 parent = getppid(); 1406 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1407 ASSERT_EQ(0, ret); 1408 1409 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1410 ASSERT_EQ(0, ret); 1411 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1412 ASSERT_EQ(0, ret); 1413 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1414 ASSERT_EQ(0, ret); 1415 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1416 ASSERT_EQ(0, ret); 1417 /* Should work just fine. */ 1418 EXPECT_EQ(parent, syscall(__NR_getppid)); 1419 EXPECT_EQ(0, syscall(__NR_getpid)); 1420 } 1421 1422 TEST_F(precedence, trace_is_fourth) 1423 { 1424 pid_t parent; 1425 long ret; 1426 1427 parent = getppid(); 1428 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1429 ASSERT_EQ(0, ret); 1430 1431 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1432 ASSERT_EQ(0, ret); 1433 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1434 ASSERT_EQ(0, ret); 1435 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1436 ASSERT_EQ(0, ret); 1437 /* Should work just fine. */ 1438 EXPECT_EQ(parent, syscall(__NR_getppid)); 1439 /* No ptracer */ 1440 EXPECT_EQ(-1, syscall(__NR_getpid)); 1441 } 1442 1443 TEST_F(precedence, trace_is_fourth_in_any_order) 1444 { 1445 pid_t parent; 1446 long ret; 1447 1448 parent = getppid(); 1449 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1450 ASSERT_EQ(0, ret); 1451 1452 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1453 ASSERT_EQ(0, ret); 1454 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1455 ASSERT_EQ(0, ret); 1456 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1457 ASSERT_EQ(0, ret); 1458 /* Should work just fine. */ 1459 EXPECT_EQ(parent, syscall(__NR_getppid)); 1460 /* No ptracer */ 1461 EXPECT_EQ(-1, syscall(__NR_getpid)); 1462 } 1463 1464 TEST_F(precedence, log_is_fifth) 1465 { 1466 pid_t mypid, parent; 1467 long ret; 1468 1469 mypid = getpid(); 1470 parent = getppid(); 1471 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1472 ASSERT_EQ(0, ret); 1473 1474 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1475 ASSERT_EQ(0, ret); 1476 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1477 ASSERT_EQ(0, ret); 1478 /* Should work just fine. */ 1479 EXPECT_EQ(parent, syscall(__NR_getppid)); 1480 /* Should also work just fine */ 1481 EXPECT_EQ(mypid, syscall(__NR_getpid)); 1482 } 1483 1484 TEST_F(precedence, log_is_fifth_in_any_order) 1485 { 1486 pid_t mypid, parent; 1487 long ret; 1488 1489 mypid = getpid(); 1490 parent = getppid(); 1491 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1492 ASSERT_EQ(0, ret); 1493 1494 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1495 ASSERT_EQ(0, ret); 1496 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1497 ASSERT_EQ(0, ret); 1498 /* Should work just fine. */ 1499 EXPECT_EQ(parent, syscall(__NR_getppid)); 1500 /* Should also work just fine */ 1501 EXPECT_EQ(mypid, syscall(__NR_getpid)); 1502 } 1503 1504 #ifndef PTRACE_O_TRACESECCOMP 1505 #define PTRACE_O_TRACESECCOMP 0x00000080 1506 #endif 1507 1508 /* Catch the Ubuntu 12.04 value error. */ 1509 #if PTRACE_EVENT_SECCOMP != 7 1510 #undef PTRACE_EVENT_SECCOMP 1511 #endif 1512 1513 #ifndef PTRACE_EVENT_SECCOMP 1514 #define PTRACE_EVENT_SECCOMP 7 1515 #endif 1516 1517 #define PTRACE_EVENT_MASK(status) ((status) >> 16) 1518 bool tracer_running; 1519 void tracer_stop(int sig) 1520 { 1521 tracer_running = false; 1522 } 1523 1524 typedef void tracer_func_t(struct __test_metadata *_metadata, 1525 pid_t tracee, int status, void *args); 1526 1527 void start_tracer(struct __test_metadata *_metadata, int fd, pid_t tracee, 1528 tracer_func_t tracer_func, void *args, bool ptrace_syscall) 1529 { 1530 int ret = -1; 1531 struct sigaction action = { 1532 .sa_handler = tracer_stop, 1533 }; 1534 1535 /* Allow external shutdown. */ 1536 tracer_running = true; 1537 ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL)); 1538 1539 errno = 0; 1540 while (ret == -1 && errno != EINVAL) 1541 ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0); 1542 ASSERT_EQ(0, ret) { 1543 kill(tracee, SIGKILL); 1544 } 1545 /* Wait for attach stop */ 1546 wait(NULL); 1547 1548 ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, ptrace_syscall ? 1549 PTRACE_O_TRACESYSGOOD : 1550 PTRACE_O_TRACESECCOMP); 1551 ASSERT_EQ(0, ret) { 1552 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP"); 1553 kill(tracee, SIGKILL); 1554 } 1555 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT, 1556 tracee, NULL, 0); 1557 ASSERT_EQ(0, ret); 1558 1559 /* Unblock the tracee */ 1560 ASSERT_EQ(1, write(fd, "A", 1)); 1561 ASSERT_EQ(0, close(fd)); 1562 1563 /* Run until we're shut down. Must assert to stop execution. */ 1564 while (tracer_running) { 1565 int status; 1566 1567 if (wait(&status) != tracee) 1568 continue; 1569 1570 if (WIFSIGNALED(status)) { 1571 /* Child caught a fatal signal. */ 1572 return; 1573 } 1574 if (WIFEXITED(status)) { 1575 /* Child exited with code. */ 1576 return; 1577 } 1578 1579 /* Check if we got an expected event. */ 1580 ASSERT_EQ(WIFCONTINUED(status), false); 1581 ASSERT_EQ(WIFSTOPPED(status), true); 1582 ASSERT_EQ(WSTOPSIG(status) & SIGTRAP, SIGTRAP) { 1583 TH_LOG("Unexpected WSTOPSIG: %d", WSTOPSIG(status)); 1584 } 1585 1586 tracer_func(_metadata, tracee, status, args); 1587 1588 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT, 1589 tracee, NULL, 0); 1590 ASSERT_EQ(0, ret); 1591 } 1592 /* Directly report the status of our test harness results. */ 1593 syscall(__NR_exit, _metadata->exit_code); 1594 } 1595 1596 /* Common tracer setup/teardown functions. */ 1597 void cont_handler(int num) 1598 { } 1599 pid_t setup_trace_fixture(struct __test_metadata *_metadata, 1600 tracer_func_t func, void *args, bool ptrace_syscall) 1601 { 1602 char sync; 1603 int pipefd[2]; 1604 pid_t tracer_pid; 1605 pid_t tracee = getpid(); 1606 1607 /* Setup a pipe for clean synchronization. */ 1608 ASSERT_EQ(0, pipe(pipefd)); 1609 1610 /* Fork a child which we'll promote to tracer */ 1611 tracer_pid = fork(); 1612 ASSERT_LE(0, tracer_pid); 1613 signal(SIGALRM, cont_handler); 1614 if (tracer_pid == 0) { 1615 close(pipefd[0]); 1616 start_tracer(_metadata, pipefd[1], tracee, func, args, 1617 ptrace_syscall); 1618 syscall(__NR_exit, 0); 1619 } 1620 close(pipefd[1]); 1621 prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0); 1622 read(pipefd[0], &sync, 1); 1623 close(pipefd[0]); 1624 1625 return tracer_pid; 1626 } 1627 1628 void teardown_trace_fixture(struct __test_metadata *_metadata, 1629 pid_t tracer) 1630 { 1631 if (tracer) { 1632 int status; 1633 ASSERT_EQ(0, kill(tracer, SIGUSR1)); 1634 ASSERT_EQ(tracer, waitpid(tracer, &status, 0)); 1635 } 1636 } 1637 1638 /* "poke" tracer arguments and function. */ 1639 struct tracer_args_poke_t { 1640 unsigned long poke_addr; 1641 }; 1642 1643 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status, 1644 void *args) 1645 { 1646 int ret; 1647 unsigned long msg; 1648 struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args; 1649 1650 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 1651 EXPECT_EQ(0, ret); 1652 /* If this fails, don't try to recover. */ 1653 ASSERT_EQ(0x1001, msg) { 1654 kill(tracee, SIGKILL); 1655 } 1656 /* 1657 * Poke in the message. 1658 * Registers are not touched to try to keep this relatively arch 1659 * agnostic. 1660 */ 1661 ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001); 1662 EXPECT_EQ(0, ret); 1663 } 1664 1665 FIXTURE(TRACE_poke) { 1666 struct sock_fprog prog; 1667 pid_t tracer; 1668 long poked; 1669 struct tracer_args_poke_t tracer_args; 1670 }; 1671 1672 FIXTURE_SETUP(TRACE_poke) 1673 { 1674 struct sock_filter filter[] = { 1675 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1676 offsetof(struct seccomp_data, nr)), 1677 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 1678 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001), 1679 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1680 }; 1681 1682 self->poked = 0; 1683 memset(&self->prog, 0, sizeof(self->prog)); 1684 self->prog.filter = malloc(sizeof(filter)); 1685 ASSERT_NE(NULL, self->prog.filter); 1686 memcpy(self->prog.filter, filter, sizeof(filter)); 1687 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 1688 1689 /* Set up tracer args. */ 1690 self->tracer_args.poke_addr = (unsigned long)&self->poked; 1691 1692 /* Launch tracer. */ 1693 self->tracer = setup_trace_fixture(_metadata, tracer_poke, 1694 &self->tracer_args, false); 1695 } 1696 1697 FIXTURE_TEARDOWN(TRACE_poke) 1698 { 1699 teardown_trace_fixture(_metadata, self->tracer); 1700 if (self->prog.filter) 1701 free(self->prog.filter); 1702 } 1703 1704 TEST_F(TRACE_poke, read_has_side_effects) 1705 { 1706 ssize_t ret; 1707 1708 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1709 ASSERT_EQ(0, ret); 1710 1711 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1712 ASSERT_EQ(0, ret); 1713 1714 EXPECT_EQ(0, self->poked); 1715 ret = read(-1, NULL, 0); 1716 EXPECT_EQ(-1, ret); 1717 EXPECT_EQ(0x1001, self->poked); 1718 } 1719 1720 TEST_F(TRACE_poke, getpid_runs_normally) 1721 { 1722 long ret; 1723 1724 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1725 ASSERT_EQ(0, ret); 1726 1727 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1728 ASSERT_EQ(0, ret); 1729 1730 EXPECT_EQ(0, self->poked); 1731 EXPECT_NE(0, syscall(__NR_getpid)); 1732 EXPECT_EQ(0, self->poked); 1733 } 1734 1735 #if defined(__x86_64__) 1736 # define ARCH_REGS struct user_regs_struct 1737 # define SYSCALL_NUM(_regs) (_regs).orig_rax 1738 # define SYSCALL_RET(_regs) (_regs).rax 1739 #elif defined(__i386__) 1740 # define ARCH_REGS struct user_regs_struct 1741 # define SYSCALL_NUM(_regs) (_regs).orig_eax 1742 # define SYSCALL_RET(_regs) (_regs).eax 1743 #elif defined(__arm__) 1744 # define ARCH_REGS struct pt_regs 1745 # define SYSCALL_NUM(_regs) (_regs).ARM_r7 1746 # ifndef PTRACE_SET_SYSCALL 1747 # define PTRACE_SET_SYSCALL 23 1748 # endif 1749 # define SYSCALL_NUM_SET(_regs, _nr) \ 1750 EXPECT_EQ(0, ptrace(PTRACE_SET_SYSCALL, tracee, NULL, _nr)) 1751 # define SYSCALL_RET(_regs) (_regs).ARM_r0 1752 #elif defined(__aarch64__) 1753 # define ARCH_REGS struct user_pt_regs 1754 # define SYSCALL_NUM(_regs) (_regs).regs[8] 1755 # ifndef NT_ARM_SYSTEM_CALL 1756 # define NT_ARM_SYSTEM_CALL 0x404 1757 # endif 1758 # define SYSCALL_NUM_SET(_regs, _nr) \ 1759 do { \ 1760 struct iovec __v; \ 1761 typeof(_nr) __nr = (_nr); \ 1762 __v.iov_base = &__nr; \ 1763 __v.iov_len = sizeof(__nr); \ 1764 EXPECT_EQ(0, ptrace(PTRACE_SETREGSET, tracee, \ 1765 NT_ARM_SYSTEM_CALL, &__v)); \ 1766 } while (0) 1767 # define SYSCALL_RET(_regs) (_regs).regs[0] 1768 #elif defined(__loongarch__) 1769 # define ARCH_REGS struct user_pt_regs 1770 # define SYSCALL_NUM(_regs) (_regs).regs[11] 1771 # define SYSCALL_RET(_regs) (_regs).regs[4] 1772 #elif defined(__riscv) && __riscv_xlen == 64 1773 # define ARCH_REGS struct user_regs_struct 1774 # define SYSCALL_NUM(_regs) (_regs).a7 1775 # define SYSCALL_RET(_regs) (_regs).a0 1776 #elif defined(__csky__) 1777 # define ARCH_REGS struct pt_regs 1778 # if defined(__CSKYABIV2__) 1779 # define SYSCALL_NUM(_regs) (_regs).regs[3] 1780 # else 1781 # define SYSCALL_NUM(_regs) (_regs).regs[9] 1782 # endif 1783 # define SYSCALL_RET(_regs) (_regs).a0 1784 #elif defined(__hppa__) 1785 # define ARCH_REGS struct user_regs_struct 1786 # define SYSCALL_NUM(_regs) (_regs).gr[20] 1787 # define SYSCALL_RET(_regs) (_regs).gr[28] 1788 #elif defined(__powerpc__) 1789 # define ARCH_REGS struct pt_regs 1790 # define SYSCALL_NUM(_regs) (_regs).gpr[0] 1791 # define SYSCALL_RET(_regs) (_regs).gpr[3] 1792 # define SYSCALL_RET_SET(_regs, _val) \ 1793 do { \ 1794 typeof(_val) _result = (_val); \ 1795 if ((_regs.trap & 0xfff0) == 0x3000) { \ 1796 /* \ 1797 * scv 0 system call uses -ve result \ 1798 * for error, so no need to adjust. \ 1799 */ \ 1800 SYSCALL_RET(_regs) = _result; \ 1801 } else { \ 1802 /* \ 1803 * A syscall error is signaled by the \ 1804 * CR0 SO bit and the code is stored as \ 1805 * a positive value. \ 1806 */ \ 1807 if (_result < 0) { \ 1808 SYSCALL_RET(_regs) = -_result; \ 1809 (_regs).ccr |= 0x10000000; \ 1810 } else { \ 1811 SYSCALL_RET(_regs) = _result; \ 1812 (_regs).ccr &= ~0x10000000; \ 1813 } \ 1814 } \ 1815 } while (0) 1816 # define SYSCALL_RET_SET_ON_PTRACE_EXIT 1817 #elif defined(__s390__) 1818 # define ARCH_REGS s390_regs 1819 # define SYSCALL_NUM(_regs) (_regs).gprs[2] 1820 # define SYSCALL_RET_SET(_regs, _val) \ 1821 TH_LOG("Can't modify syscall return on this architecture") 1822 #elif defined(__mips__) 1823 # include <asm/unistd_nr_n32.h> 1824 # include <asm/unistd_nr_n64.h> 1825 # include <asm/unistd_nr_o32.h> 1826 # define ARCH_REGS struct pt_regs 1827 # define SYSCALL_NUM(_regs) \ 1828 ({ \ 1829 typeof((_regs).regs[2]) _nr; \ 1830 if ((_regs).regs[2] == __NR_O32_Linux) \ 1831 _nr = (_regs).regs[4]; \ 1832 else \ 1833 _nr = (_regs).regs[2]; \ 1834 _nr; \ 1835 }) 1836 # define SYSCALL_NUM_SET(_regs, _nr) \ 1837 do { \ 1838 if ((_regs).regs[2] == __NR_O32_Linux) \ 1839 (_regs).regs[4] = _nr; \ 1840 else \ 1841 (_regs).regs[2] = _nr; \ 1842 } while (0) 1843 # define SYSCALL_RET_SET(_regs, _val) \ 1844 TH_LOG("Can't modify syscall return on this architecture") 1845 #elif defined(__xtensa__) 1846 # define ARCH_REGS struct user_pt_regs 1847 # define SYSCALL_NUM(_regs) (_regs).syscall 1848 /* 1849 * On xtensa syscall return value is in the register 1850 * a2 of the current window which is not fixed. 1851 */ 1852 #define SYSCALL_RET(_regs) (_regs).a[(_regs).windowbase * 4 + 2] 1853 #elif defined(__sh__) 1854 # define ARCH_REGS struct pt_regs 1855 # define SYSCALL_NUM(_regs) (_regs).regs[3] 1856 # define SYSCALL_RET(_regs) (_regs).regs[0] 1857 #elif defined(__mc68000__) 1858 # define ARCH_REGS struct user_regs_struct 1859 # define SYSCALL_NUM(_regs) (_regs).orig_d0 1860 # define SYSCALL_RET(_regs) (_regs).d0 1861 #else 1862 # error "Do not know how to find your architecture's registers and syscalls" 1863 #endif 1864 1865 /* 1866 * Most architectures can change the syscall by just updating the 1867 * associated register. This is the default if not defined above. 1868 */ 1869 #ifndef SYSCALL_NUM_SET 1870 # define SYSCALL_NUM_SET(_regs, _nr) \ 1871 do { \ 1872 SYSCALL_NUM(_regs) = (_nr); \ 1873 } while (0) 1874 #endif 1875 /* 1876 * Most architectures can change the syscall return value by just 1877 * writing to the SYSCALL_RET register. This is the default if not 1878 * defined above. If an architecture cannot set the return value 1879 * (for example when the syscall and return value register is 1880 * shared), report it with TH_LOG() in an arch-specific definition 1881 * of SYSCALL_RET_SET() above, and leave SYSCALL_RET undefined. 1882 */ 1883 #if !defined(SYSCALL_RET) && !defined(SYSCALL_RET_SET) 1884 # error "One of SYSCALL_RET or SYSCALL_RET_SET is needed for this arch" 1885 #endif 1886 #ifndef SYSCALL_RET_SET 1887 # define SYSCALL_RET_SET(_regs, _val) \ 1888 do { \ 1889 SYSCALL_RET(_regs) = (_val); \ 1890 } while (0) 1891 #endif 1892 1893 /* When the syscall return can't be changed, stub out the tests for it. */ 1894 #ifndef SYSCALL_RET 1895 # define EXPECT_SYSCALL_RETURN(val, action) EXPECT_EQ(-1, action) 1896 #else 1897 # define EXPECT_SYSCALL_RETURN(val, action) \ 1898 do { \ 1899 errno = 0; \ 1900 if (val < 0) { \ 1901 EXPECT_EQ(-1, action); \ 1902 EXPECT_EQ(-(val), errno); \ 1903 } else { \ 1904 EXPECT_EQ(val, action); \ 1905 } \ 1906 } while (0) 1907 #endif 1908 1909 /* 1910 * Some architectures (e.g. powerpc) can only set syscall 1911 * return values on syscall exit during ptrace. 1912 */ 1913 const bool ptrace_entry_set_syscall_nr = true; 1914 const bool ptrace_entry_set_syscall_ret = 1915 #ifndef SYSCALL_RET_SET_ON_PTRACE_EXIT 1916 true; 1917 #else 1918 false; 1919 #endif 1920 1921 /* 1922 * Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for 1923 * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux). 1924 */ 1925 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__) || defined(__mc68000__) 1926 # define ARCH_GETREGS(_regs) ptrace(PTRACE_GETREGS, tracee, 0, &(_regs)) 1927 # define ARCH_SETREGS(_regs) ptrace(PTRACE_SETREGS, tracee, 0, &(_regs)) 1928 #else 1929 # define ARCH_GETREGS(_regs) ({ \ 1930 struct iovec __v; \ 1931 __v.iov_base = &(_regs); \ 1932 __v.iov_len = sizeof(_regs); \ 1933 ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &__v); \ 1934 }) 1935 # define ARCH_SETREGS(_regs) ({ \ 1936 struct iovec __v; \ 1937 __v.iov_base = &(_regs); \ 1938 __v.iov_len = sizeof(_regs); \ 1939 ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &__v); \ 1940 }) 1941 #endif 1942 1943 /* Architecture-specific syscall fetching routine. */ 1944 int get_syscall(struct __test_metadata *_metadata, pid_t tracee) 1945 { 1946 ARCH_REGS regs; 1947 1948 EXPECT_EQ(0, ARCH_GETREGS(regs)) { 1949 return -1; 1950 } 1951 1952 return SYSCALL_NUM(regs); 1953 } 1954 1955 /* Architecture-specific syscall changing routine. */ 1956 void __change_syscall(struct __test_metadata *_metadata, 1957 pid_t tracee, long *syscall, long *ret) 1958 { 1959 ARCH_REGS orig, regs; 1960 1961 /* Do not get/set registers if we have nothing to do. */ 1962 if (!syscall && !ret) 1963 return; 1964 1965 EXPECT_EQ(0, ARCH_GETREGS(regs)) { 1966 return; 1967 } 1968 orig = regs; 1969 1970 if (syscall) 1971 SYSCALL_NUM_SET(regs, *syscall); 1972 1973 if (ret) 1974 SYSCALL_RET_SET(regs, *ret); 1975 1976 /* Flush any register changes made. */ 1977 if (memcmp(&orig, ®s, sizeof(orig)) != 0) 1978 EXPECT_EQ(0, ARCH_SETREGS(regs)); 1979 } 1980 1981 /* Change only syscall number. */ 1982 void change_syscall_nr(struct __test_metadata *_metadata, 1983 pid_t tracee, long syscall) 1984 { 1985 __change_syscall(_metadata, tracee, &syscall, NULL); 1986 } 1987 1988 /* Change syscall return value (and set syscall number to -1). */ 1989 void change_syscall_ret(struct __test_metadata *_metadata, 1990 pid_t tracee, long ret) 1991 { 1992 long syscall = -1; 1993 1994 __change_syscall(_metadata, tracee, &syscall, &ret); 1995 } 1996 1997 void tracer_seccomp(struct __test_metadata *_metadata, pid_t tracee, 1998 int status, void *args) 1999 { 2000 int ret; 2001 unsigned long msg; 2002 2003 EXPECT_EQ(PTRACE_EVENT_MASK(status), PTRACE_EVENT_SECCOMP) { 2004 TH_LOG("Unexpected ptrace event: %d", PTRACE_EVENT_MASK(status)); 2005 return; 2006 } 2007 2008 /* Make sure we got the right message. */ 2009 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 2010 EXPECT_EQ(0, ret); 2011 2012 /* Validate and take action on expected syscalls. */ 2013 switch (msg) { 2014 case 0x1002: 2015 /* change getpid to getppid. */ 2016 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee)); 2017 change_syscall_nr(_metadata, tracee, __NR_getppid); 2018 break; 2019 case 0x1003: 2020 /* skip gettid with valid return code. */ 2021 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee)); 2022 change_syscall_ret(_metadata, tracee, 45000); 2023 break; 2024 case 0x1004: 2025 /* skip openat with error. */ 2026 EXPECT_EQ(__NR_openat, get_syscall(_metadata, tracee)); 2027 change_syscall_ret(_metadata, tracee, -ESRCH); 2028 break; 2029 case 0x1005: 2030 /* do nothing (allow getppid) */ 2031 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee)); 2032 break; 2033 default: 2034 EXPECT_EQ(0, msg) { 2035 TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg); 2036 kill(tracee, SIGKILL); 2037 } 2038 } 2039 2040 } 2041 2042 FIXTURE(TRACE_syscall) { 2043 struct sock_fprog prog; 2044 pid_t tracer, mytid, mypid, parent; 2045 long syscall_nr; 2046 }; 2047 2048 void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee, 2049 int status, void *args) 2050 { 2051 int ret; 2052 unsigned long msg; 2053 static bool entry; 2054 long syscall_nr_val, syscall_ret_val; 2055 long *syscall_nr = NULL, *syscall_ret = NULL; 2056 FIXTURE_DATA(TRACE_syscall) *self = args; 2057 2058 EXPECT_EQ(WSTOPSIG(status) & 0x80, 0x80) { 2059 TH_LOG("Unexpected WSTOPSIG: %d", WSTOPSIG(status)); 2060 return; 2061 } 2062 2063 /* 2064 * The traditional way to tell PTRACE_SYSCALL entry/exit 2065 * is by counting. 2066 */ 2067 entry = !entry; 2068 2069 /* Make sure we got an appropriate message. */ 2070 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 2071 EXPECT_EQ(0, ret); 2072 EXPECT_EQ(entry ? PTRACE_EVENTMSG_SYSCALL_ENTRY 2073 : PTRACE_EVENTMSG_SYSCALL_EXIT, msg); 2074 2075 /* 2076 * Some architectures only support setting return values during 2077 * syscall exit under ptrace, and on exit the syscall number may 2078 * no longer be available. Therefore, save the initial sycall 2079 * number here, so it can be examined during both entry and exit 2080 * phases. 2081 */ 2082 if (entry) 2083 self->syscall_nr = get_syscall(_metadata, tracee); 2084 2085 /* 2086 * Depending on the architecture's syscall setting abilities, we 2087 * pick which things to set during this phase (entry or exit). 2088 */ 2089 if (entry == ptrace_entry_set_syscall_nr) 2090 syscall_nr = &syscall_nr_val; 2091 if (entry == ptrace_entry_set_syscall_ret) 2092 syscall_ret = &syscall_ret_val; 2093 2094 /* Now handle the actual rewriting cases. */ 2095 switch (self->syscall_nr) { 2096 case __NR_getpid: 2097 syscall_nr_val = __NR_getppid; 2098 /* Never change syscall return for this case. */ 2099 syscall_ret = NULL; 2100 break; 2101 case __NR_gettid: 2102 syscall_nr_val = -1; 2103 syscall_ret_val = 45000; 2104 break; 2105 case __NR_openat: 2106 syscall_nr_val = -1; 2107 syscall_ret_val = -ESRCH; 2108 break; 2109 default: 2110 /* Unhandled, do nothing. */ 2111 return; 2112 } 2113 2114 __change_syscall(_metadata, tracee, syscall_nr, syscall_ret); 2115 } 2116 2117 FIXTURE_VARIANT(TRACE_syscall) { 2118 /* 2119 * All of the SECCOMP_RET_TRACE behaviors can be tested with either 2120 * SECCOMP_RET_TRACE+PTRACE_CONT or plain ptrace()+PTRACE_SYSCALL. 2121 * This indicates if we should use SECCOMP_RET_TRACE (false), or 2122 * ptrace (true). 2123 */ 2124 bool use_ptrace; 2125 }; 2126 2127 FIXTURE_VARIANT_ADD(TRACE_syscall, ptrace) { 2128 .use_ptrace = true, 2129 }; 2130 2131 FIXTURE_VARIANT_ADD(TRACE_syscall, seccomp) { 2132 .use_ptrace = false, 2133 }; 2134 2135 FIXTURE_SETUP(TRACE_syscall) 2136 { 2137 struct sock_filter filter[] = { 2138 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2139 offsetof(struct seccomp_data, nr)), 2140 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 2141 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002), 2142 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1), 2143 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003), 2144 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_openat, 0, 1), 2145 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004), 2146 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 2147 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1005), 2148 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2149 }; 2150 struct sock_fprog prog = { 2151 .len = (unsigned short)ARRAY_SIZE(filter), 2152 .filter = filter, 2153 }; 2154 long ret; 2155 2156 /* Prepare some testable syscall results. */ 2157 self->mytid = syscall(__NR_gettid); 2158 ASSERT_GT(self->mytid, 0); 2159 ASSERT_NE(self->mytid, 1) { 2160 TH_LOG("Running this test as init is not supported. :)"); 2161 } 2162 2163 self->mypid = getpid(); 2164 ASSERT_GT(self->mypid, 0); 2165 ASSERT_EQ(self->mytid, self->mypid); 2166 2167 self->parent = getppid(); 2168 ASSERT_GT(self->parent, 0); 2169 ASSERT_NE(self->parent, self->mypid); 2170 2171 /* Launch tracer. */ 2172 self->tracer = setup_trace_fixture(_metadata, 2173 variant->use_ptrace ? tracer_ptrace 2174 : tracer_seccomp, 2175 self, variant->use_ptrace); 2176 2177 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 2178 ASSERT_EQ(0, ret); 2179 2180 /* Do not install seccomp rewrite filters, as we'll use ptrace instead. */ 2181 if (variant->use_ptrace) 2182 return; 2183 2184 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2185 ASSERT_EQ(0, ret); 2186 } 2187 2188 FIXTURE_TEARDOWN(TRACE_syscall) 2189 { 2190 teardown_trace_fixture(_metadata, self->tracer); 2191 } 2192 2193 TEST(negative_ENOSYS) 2194 { 2195 #if defined(__arm__) 2196 SKIP(return, "arm32 does not support calling syscall -1"); 2197 #endif 2198 /* 2199 * There should be no difference between an "internal" skip 2200 * and userspace asking for syscall "-1". 2201 */ 2202 errno = 0; 2203 EXPECT_EQ(-1, syscall(-1)); 2204 EXPECT_EQ(errno, ENOSYS); 2205 /* And no difference for "still not valid but not -1". */ 2206 errno = 0; 2207 EXPECT_EQ(-1, syscall(-101)); 2208 EXPECT_EQ(errno, ENOSYS); 2209 } 2210 2211 TEST_F(TRACE_syscall, negative_ENOSYS) 2212 { 2213 negative_ENOSYS(_metadata); 2214 } 2215 2216 TEST_F(TRACE_syscall, syscall_allowed) 2217 { 2218 /* getppid works as expected (no changes). */ 2219 EXPECT_EQ(self->parent, syscall(__NR_getppid)); 2220 EXPECT_NE(self->mypid, syscall(__NR_getppid)); 2221 } 2222 2223 TEST_F(TRACE_syscall, syscall_redirected) 2224 { 2225 /* getpid has been redirected to getppid as expected. */ 2226 EXPECT_EQ(self->parent, syscall(__NR_getpid)); 2227 EXPECT_NE(self->mypid, syscall(__NR_getpid)); 2228 } 2229 2230 TEST_F(TRACE_syscall, syscall_errno) 2231 { 2232 /* Tracer should skip the open syscall, resulting in ESRCH. */ 2233 EXPECT_SYSCALL_RETURN(-ESRCH, syscall(__NR_openat)); 2234 } 2235 2236 TEST_F(TRACE_syscall, syscall_faked) 2237 { 2238 /* Tracer skips the gettid syscall and store altered return value. */ 2239 EXPECT_SYSCALL_RETURN(45000, syscall(__NR_gettid)); 2240 } 2241 2242 TEST_F_SIGNAL(TRACE_syscall, kill_immediate, SIGSYS) 2243 { 2244 struct sock_filter filter[] = { 2245 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2246 offsetof(struct seccomp_data, nr)), 2247 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_mknodat, 0, 1), 2248 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD), 2249 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2250 }; 2251 struct sock_fprog prog = { 2252 .len = (unsigned short)ARRAY_SIZE(filter), 2253 .filter = filter, 2254 }; 2255 long ret; 2256 2257 /* Install "kill on mknodat" filter. */ 2258 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2259 ASSERT_EQ(0, ret); 2260 2261 /* This should immediately die with SIGSYS, regardless of tracer. */ 2262 EXPECT_EQ(-1, syscall(__NR_mknodat, -1, NULL, 0, 0)); 2263 } 2264 2265 TEST_F(TRACE_syscall, skip_after) 2266 { 2267 struct sock_filter filter[] = { 2268 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2269 offsetof(struct seccomp_data, nr)), 2270 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 2271 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM), 2272 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2273 }; 2274 struct sock_fprog prog = { 2275 .len = (unsigned short)ARRAY_SIZE(filter), 2276 .filter = filter, 2277 }; 2278 long ret; 2279 2280 /* Install additional "errno on getppid" filter. */ 2281 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2282 ASSERT_EQ(0, ret); 2283 2284 /* Tracer will redirect getpid to getppid, and we should see EPERM. */ 2285 errno = 0; 2286 EXPECT_EQ(-1, syscall(__NR_getpid)); 2287 EXPECT_EQ(EPERM, errno); 2288 } 2289 2290 TEST_F_SIGNAL(TRACE_syscall, kill_after, SIGSYS) 2291 { 2292 struct sock_filter filter[] = { 2293 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2294 offsetof(struct seccomp_data, nr)), 2295 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 2296 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2297 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2298 }; 2299 struct sock_fprog prog = { 2300 .len = (unsigned short)ARRAY_SIZE(filter), 2301 .filter = filter, 2302 }; 2303 long ret; 2304 2305 /* Install additional "death on getppid" filter. */ 2306 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2307 ASSERT_EQ(0, ret); 2308 2309 /* Tracer will redirect getpid to getppid, and we should die. */ 2310 EXPECT_NE(self->mypid, syscall(__NR_getpid)); 2311 } 2312 2313 TEST(seccomp_syscall) 2314 { 2315 struct sock_filter filter[] = { 2316 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2317 }; 2318 struct sock_fprog prog = { 2319 .len = (unsigned short)ARRAY_SIZE(filter), 2320 .filter = filter, 2321 }; 2322 long ret; 2323 2324 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 2325 ASSERT_EQ(0, ret) { 2326 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2327 } 2328 2329 /* Reject insane operation. */ 2330 ret = seccomp(-1, 0, &prog); 2331 ASSERT_NE(ENOSYS, errno) { 2332 TH_LOG("Kernel does not support seccomp syscall!"); 2333 } 2334 EXPECT_EQ(EINVAL, errno) { 2335 TH_LOG("Did not reject crazy op value!"); 2336 } 2337 2338 /* Reject strict with flags or pointer. */ 2339 ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL); 2340 EXPECT_EQ(EINVAL, errno) { 2341 TH_LOG("Did not reject mode strict with flags!"); 2342 } 2343 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog); 2344 EXPECT_EQ(EINVAL, errno) { 2345 TH_LOG("Did not reject mode strict with uargs!"); 2346 } 2347 2348 /* Reject insane args for filter. */ 2349 ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog); 2350 EXPECT_EQ(EINVAL, errno) { 2351 TH_LOG("Did not reject crazy filter flags!"); 2352 } 2353 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL); 2354 EXPECT_EQ(EFAULT, errno) { 2355 TH_LOG("Did not reject NULL filter!"); 2356 } 2357 2358 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2359 EXPECT_EQ(0, errno) { 2360 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s", 2361 strerror(errno)); 2362 } 2363 } 2364 2365 TEST(seccomp_syscall_mode_lock) 2366 { 2367 struct sock_filter filter[] = { 2368 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2369 }; 2370 struct sock_fprog prog = { 2371 .len = (unsigned short)ARRAY_SIZE(filter), 2372 .filter = filter, 2373 }; 2374 long ret; 2375 2376 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 2377 ASSERT_EQ(0, ret) { 2378 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2379 } 2380 2381 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2382 ASSERT_NE(ENOSYS, errno) { 2383 TH_LOG("Kernel does not support seccomp syscall!"); 2384 } 2385 EXPECT_EQ(0, ret) { 2386 TH_LOG("Could not install filter!"); 2387 } 2388 2389 /* Make sure neither entry point will switch to strict. */ 2390 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0); 2391 EXPECT_EQ(EINVAL, errno) { 2392 TH_LOG("Switched to mode strict!"); 2393 } 2394 2395 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL); 2396 EXPECT_EQ(EINVAL, errno) { 2397 TH_LOG("Switched to mode strict!"); 2398 } 2399 } 2400 2401 /* 2402 * Test detection of known and unknown filter flags. Userspace needs to be able 2403 * to check if a filter flag is supported by the current kernel and a good way 2404 * of doing that is by attempting to enter filter mode, with the flag bit in 2405 * question set, and a NULL pointer for the _args_ parameter. EFAULT indicates 2406 * that the flag is valid and EINVAL indicates that the flag is invalid. 2407 */ 2408 TEST(detect_seccomp_filter_flags) 2409 { 2410 unsigned int flags[] = { SECCOMP_FILTER_FLAG_TSYNC, 2411 SECCOMP_FILTER_FLAG_LOG, 2412 SECCOMP_FILTER_FLAG_SPEC_ALLOW, 2413 SECCOMP_FILTER_FLAG_NEW_LISTENER, 2414 SECCOMP_FILTER_FLAG_TSYNC_ESRCH }; 2415 unsigned int exclusive[] = { 2416 SECCOMP_FILTER_FLAG_TSYNC, 2417 SECCOMP_FILTER_FLAG_NEW_LISTENER }; 2418 unsigned int flag, all_flags, exclusive_mask; 2419 int i; 2420 long ret; 2421 2422 /* Test detection of individual known-good filter flags */ 2423 for (i = 0, all_flags = 0; i < ARRAY_SIZE(flags); i++) { 2424 int bits = 0; 2425 2426 flag = flags[i]; 2427 /* Make sure the flag is a single bit! */ 2428 while (flag) { 2429 if (flag & 0x1) 2430 bits ++; 2431 flag >>= 1; 2432 } 2433 ASSERT_EQ(1, bits); 2434 flag = flags[i]; 2435 2436 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2437 ASSERT_NE(ENOSYS, errno) { 2438 TH_LOG("Kernel does not support seccomp syscall!"); 2439 } 2440 EXPECT_EQ(-1, ret); 2441 EXPECT_EQ(EFAULT, errno) { 2442 TH_LOG("Failed to detect that a known-good filter flag (0x%X) is supported!", 2443 flag); 2444 } 2445 2446 all_flags |= flag; 2447 } 2448 2449 /* 2450 * Test detection of all known-good filter flags combined. But 2451 * for the exclusive flags we need to mask them out and try them 2452 * individually for the "all flags" testing. 2453 */ 2454 exclusive_mask = 0; 2455 for (i = 0; i < ARRAY_SIZE(exclusive); i++) 2456 exclusive_mask |= exclusive[i]; 2457 for (i = 0; i < ARRAY_SIZE(exclusive); i++) { 2458 flag = all_flags & ~exclusive_mask; 2459 flag |= exclusive[i]; 2460 2461 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2462 EXPECT_EQ(-1, ret); 2463 EXPECT_EQ(EFAULT, errno) { 2464 TH_LOG("Failed to detect that all known-good filter flags (0x%X) are supported!", 2465 flag); 2466 } 2467 } 2468 2469 /* Test detection of an unknown filter flags, without exclusives. */ 2470 flag = -1; 2471 flag &= ~exclusive_mask; 2472 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2473 EXPECT_EQ(-1, ret); 2474 EXPECT_EQ(EINVAL, errno) { 2475 TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported!", 2476 flag); 2477 } 2478 2479 /* 2480 * Test detection of an unknown filter flag that may simply need to be 2481 * added to this test 2482 */ 2483 flag = flags[ARRAY_SIZE(flags) - 1] << 1; 2484 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2485 EXPECT_EQ(-1, ret); 2486 EXPECT_EQ(EINVAL, errno) { 2487 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?", 2488 flag); 2489 } 2490 } 2491 2492 TEST(TSYNC_first) 2493 { 2494 struct sock_filter filter[] = { 2495 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2496 }; 2497 struct sock_fprog prog = { 2498 .len = (unsigned short)ARRAY_SIZE(filter), 2499 .filter = filter, 2500 }; 2501 long ret; 2502 2503 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 2504 ASSERT_EQ(0, ret) { 2505 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2506 } 2507 2508 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2509 &prog); 2510 ASSERT_NE(ENOSYS, errno) { 2511 TH_LOG("Kernel does not support seccomp syscall!"); 2512 } 2513 EXPECT_EQ(0, ret) { 2514 TH_LOG("Could not install initial filter with TSYNC!"); 2515 } 2516 } 2517 2518 #define TSYNC_SIBLINGS 2 2519 struct tsync_sibling { 2520 pthread_t tid; 2521 pid_t system_tid; 2522 sem_t *started; 2523 pthread_cond_t *cond; 2524 pthread_mutex_t *mutex; 2525 int diverge; 2526 int num_waits; 2527 struct sock_fprog *prog; 2528 struct __test_metadata *metadata; 2529 }; 2530 2531 /* 2532 * To avoid joining joined threads (which is not allowed by Bionic), 2533 * make sure we both successfully join and clear the tid to skip a 2534 * later join attempt during fixture teardown. Any remaining threads 2535 * will be directly killed during teardown. 2536 */ 2537 #define PTHREAD_JOIN(tid, status) \ 2538 do { \ 2539 int _rc = pthread_join(tid, status); \ 2540 if (_rc) { \ 2541 TH_LOG("pthread_join of tid %u failed: %d\n", \ 2542 (unsigned int)tid, _rc); \ 2543 } else { \ 2544 tid = 0; \ 2545 } \ 2546 } while (0) 2547 2548 FIXTURE(TSYNC) { 2549 struct sock_fprog root_prog, apply_prog; 2550 struct tsync_sibling sibling[TSYNC_SIBLINGS]; 2551 sem_t started; 2552 pthread_cond_t cond; 2553 pthread_mutex_t mutex; 2554 int sibling_count; 2555 }; 2556 2557 FIXTURE_SETUP(TSYNC) 2558 { 2559 struct sock_filter root_filter[] = { 2560 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2561 }; 2562 struct sock_filter apply_filter[] = { 2563 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2564 offsetof(struct seccomp_data, nr)), 2565 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 2566 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2567 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2568 }; 2569 2570 memset(&self->root_prog, 0, sizeof(self->root_prog)); 2571 memset(&self->apply_prog, 0, sizeof(self->apply_prog)); 2572 memset(&self->sibling, 0, sizeof(self->sibling)); 2573 self->root_prog.filter = malloc(sizeof(root_filter)); 2574 ASSERT_NE(NULL, self->root_prog.filter); 2575 memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter)); 2576 self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter); 2577 2578 self->apply_prog.filter = malloc(sizeof(apply_filter)); 2579 ASSERT_NE(NULL, self->apply_prog.filter); 2580 memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter)); 2581 self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter); 2582 2583 self->sibling_count = 0; 2584 pthread_mutex_init(&self->mutex, NULL); 2585 pthread_cond_init(&self->cond, NULL); 2586 sem_init(&self->started, 0, 0); 2587 self->sibling[0].tid = 0; 2588 self->sibling[0].cond = &self->cond; 2589 self->sibling[0].started = &self->started; 2590 self->sibling[0].mutex = &self->mutex; 2591 self->sibling[0].diverge = 0; 2592 self->sibling[0].num_waits = 1; 2593 self->sibling[0].prog = &self->root_prog; 2594 self->sibling[0].metadata = _metadata; 2595 self->sibling[1].tid = 0; 2596 self->sibling[1].cond = &self->cond; 2597 self->sibling[1].started = &self->started; 2598 self->sibling[1].mutex = &self->mutex; 2599 self->sibling[1].diverge = 0; 2600 self->sibling[1].prog = &self->root_prog; 2601 self->sibling[1].num_waits = 1; 2602 self->sibling[1].metadata = _metadata; 2603 } 2604 2605 FIXTURE_TEARDOWN(TSYNC) 2606 { 2607 int sib = 0; 2608 2609 if (self->root_prog.filter) 2610 free(self->root_prog.filter); 2611 if (self->apply_prog.filter) 2612 free(self->apply_prog.filter); 2613 2614 for ( ; sib < self->sibling_count; ++sib) { 2615 struct tsync_sibling *s = &self->sibling[sib]; 2616 2617 if (!s->tid) 2618 continue; 2619 /* 2620 * If a thread is still running, it may be stuck, so hit 2621 * it over the head really hard. 2622 */ 2623 pthread_kill(s->tid, 9); 2624 } 2625 pthread_mutex_destroy(&self->mutex); 2626 pthread_cond_destroy(&self->cond); 2627 sem_destroy(&self->started); 2628 } 2629 2630 void *tsync_sibling(void *data) 2631 { 2632 long ret = 0; 2633 struct tsync_sibling *me = data; 2634 2635 me->system_tid = syscall(__NR_gettid); 2636 2637 pthread_mutex_lock(me->mutex); 2638 if (me->diverge) { 2639 /* Just re-apply the root prog to fork the tree */ 2640 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, 2641 me->prog, 0, 0); 2642 } 2643 sem_post(me->started); 2644 /* Return outside of started so parent notices failures. */ 2645 if (ret) { 2646 pthread_mutex_unlock(me->mutex); 2647 return (void *)SIBLING_EXIT_FAILURE; 2648 } 2649 do { 2650 pthread_cond_wait(me->cond, me->mutex); 2651 me->num_waits = me->num_waits - 1; 2652 } while (me->num_waits); 2653 pthread_mutex_unlock(me->mutex); 2654 2655 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0); 2656 if (!ret) 2657 return (void *)SIBLING_EXIT_NEWPRIVS; 2658 read(-1, NULL, 0); 2659 return (void *)SIBLING_EXIT_UNKILLED; 2660 } 2661 2662 void tsync_start_sibling(struct tsync_sibling *sibling) 2663 { 2664 pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling); 2665 } 2666 2667 TEST_F(TSYNC, siblings_fail_prctl) 2668 { 2669 long ret; 2670 void *status; 2671 struct sock_filter filter[] = { 2672 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2673 offsetof(struct seccomp_data, nr)), 2674 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1), 2675 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL), 2676 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2677 }; 2678 struct sock_fprog prog = { 2679 .len = (unsigned short)ARRAY_SIZE(filter), 2680 .filter = filter, 2681 }; 2682 2683 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2684 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2685 } 2686 2687 /* Check prctl failure detection by requesting sib 0 diverge. */ 2688 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2689 ASSERT_NE(ENOSYS, errno) { 2690 TH_LOG("Kernel does not support seccomp syscall!"); 2691 } 2692 ASSERT_EQ(0, ret) { 2693 TH_LOG("setting filter failed"); 2694 } 2695 2696 self->sibling[0].diverge = 1; 2697 tsync_start_sibling(&self->sibling[0]); 2698 tsync_start_sibling(&self->sibling[1]); 2699 2700 while (self->sibling_count < TSYNC_SIBLINGS) { 2701 sem_wait(&self->started); 2702 self->sibling_count++; 2703 } 2704 2705 /* Signal the threads to clean up*/ 2706 pthread_mutex_lock(&self->mutex); 2707 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2708 TH_LOG("cond broadcast non-zero"); 2709 } 2710 pthread_mutex_unlock(&self->mutex); 2711 2712 /* Ensure diverging sibling failed to call prctl. */ 2713 PTHREAD_JOIN(self->sibling[0].tid, &status); 2714 EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status); 2715 PTHREAD_JOIN(self->sibling[1].tid, &status); 2716 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2717 } 2718 2719 TEST_F(TSYNC, two_siblings_with_ancestor) 2720 { 2721 long ret; 2722 void *status; 2723 2724 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2725 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2726 } 2727 2728 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2729 ASSERT_NE(ENOSYS, errno) { 2730 TH_LOG("Kernel does not support seccomp syscall!"); 2731 } 2732 ASSERT_EQ(0, ret) { 2733 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2734 } 2735 tsync_start_sibling(&self->sibling[0]); 2736 tsync_start_sibling(&self->sibling[1]); 2737 2738 while (self->sibling_count < TSYNC_SIBLINGS) { 2739 sem_wait(&self->started); 2740 self->sibling_count++; 2741 } 2742 2743 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2744 &self->apply_prog); 2745 ASSERT_EQ(0, ret) { 2746 TH_LOG("Could install filter on all threads!"); 2747 } 2748 /* Tell the siblings to test the policy */ 2749 pthread_mutex_lock(&self->mutex); 2750 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2751 TH_LOG("cond broadcast non-zero"); 2752 } 2753 pthread_mutex_unlock(&self->mutex); 2754 /* Ensure they are both killed and don't exit cleanly. */ 2755 PTHREAD_JOIN(self->sibling[0].tid, &status); 2756 EXPECT_EQ(0x0, (long)status); 2757 PTHREAD_JOIN(self->sibling[1].tid, &status); 2758 EXPECT_EQ(0x0, (long)status); 2759 } 2760 2761 TEST_F(TSYNC, two_sibling_want_nnp) 2762 { 2763 void *status; 2764 2765 /* start siblings before any prctl() operations */ 2766 tsync_start_sibling(&self->sibling[0]); 2767 tsync_start_sibling(&self->sibling[1]); 2768 while (self->sibling_count < TSYNC_SIBLINGS) { 2769 sem_wait(&self->started); 2770 self->sibling_count++; 2771 } 2772 2773 /* Tell the siblings to test no policy */ 2774 pthread_mutex_lock(&self->mutex); 2775 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2776 TH_LOG("cond broadcast non-zero"); 2777 } 2778 pthread_mutex_unlock(&self->mutex); 2779 2780 /* Ensure they are both upset about lacking nnp. */ 2781 PTHREAD_JOIN(self->sibling[0].tid, &status); 2782 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status); 2783 PTHREAD_JOIN(self->sibling[1].tid, &status); 2784 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status); 2785 } 2786 2787 TEST_F(TSYNC, two_siblings_with_no_filter) 2788 { 2789 long ret; 2790 void *status; 2791 2792 /* start siblings before any prctl() operations */ 2793 tsync_start_sibling(&self->sibling[0]); 2794 tsync_start_sibling(&self->sibling[1]); 2795 while (self->sibling_count < TSYNC_SIBLINGS) { 2796 sem_wait(&self->started); 2797 self->sibling_count++; 2798 } 2799 2800 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2801 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2802 } 2803 2804 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2805 &self->apply_prog); 2806 ASSERT_NE(ENOSYS, errno) { 2807 TH_LOG("Kernel does not support seccomp syscall!"); 2808 } 2809 ASSERT_EQ(0, ret) { 2810 TH_LOG("Could install filter on all threads!"); 2811 } 2812 2813 /* Tell the siblings to test the policy */ 2814 pthread_mutex_lock(&self->mutex); 2815 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2816 TH_LOG("cond broadcast non-zero"); 2817 } 2818 pthread_mutex_unlock(&self->mutex); 2819 2820 /* Ensure they are both killed and don't exit cleanly. */ 2821 PTHREAD_JOIN(self->sibling[0].tid, &status); 2822 EXPECT_EQ(0x0, (long)status); 2823 PTHREAD_JOIN(self->sibling[1].tid, &status); 2824 EXPECT_EQ(0x0, (long)status); 2825 } 2826 2827 TEST_F(TSYNC, two_siblings_with_one_divergence) 2828 { 2829 long ret; 2830 void *status; 2831 2832 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2833 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2834 } 2835 2836 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2837 ASSERT_NE(ENOSYS, errno) { 2838 TH_LOG("Kernel does not support seccomp syscall!"); 2839 } 2840 ASSERT_EQ(0, ret) { 2841 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2842 } 2843 self->sibling[0].diverge = 1; 2844 tsync_start_sibling(&self->sibling[0]); 2845 tsync_start_sibling(&self->sibling[1]); 2846 2847 while (self->sibling_count < TSYNC_SIBLINGS) { 2848 sem_wait(&self->started); 2849 self->sibling_count++; 2850 } 2851 2852 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2853 &self->apply_prog); 2854 ASSERT_EQ(self->sibling[0].system_tid, ret) { 2855 TH_LOG("Did not fail on diverged sibling."); 2856 } 2857 2858 /* Wake the threads */ 2859 pthread_mutex_lock(&self->mutex); 2860 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2861 TH_LOG("cond broadcast non-zero"); 2862 } 2863 pthread_mutex_unlock(&self->mutex); 2864 2865 /* Ensure they are both unkilled. */ 2866 PTHREAD_JOIN(self->sibling[0].tid, &status); 2867 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2868 PTHREAD_JOIN(self->sibling[1].tid, &status); 2869 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2870 } 2871 2872 TEST_F(TSYNC, two_siblings_with_one_divergence_no_tid_in_err) 2873 { 2874 long ret, flags; 2875 void *status; 2876 2877 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2878 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2879 } 2880 2881 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2882 ASSERT_NE(ENOSYS, errno) { 2883 TH_LOG("Kernel does not support seccomp syscall!"); 2884 } 2885 ASSERT_EQ(0, ret) { 2886 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2887 } 2888 self->sibling[0].diverge = 1; 2889 tsync_start_sibling(&self->sibling[0]); 2890 tsync_start_sibling(&self->sibling[1]); 2891 2892 while (self->sibling_count < TSYNC_SIBLINGS) { 2893 sem_wait(&self->started); 2894 self->sibling_count++; 2895 } 2896 2897 flags = SECCOMP_FILTER_FLAG_TSYNC | \ 2898 SECCOMP_FILTER_FLAG_TSYNC_ESRCH; 2899 ret = seccomp(SECCOMP_SET_MODE_FILTER, flags, &self->apply_prog); 2900 ASSERT_EQ(ESRCH, errno) { 2901 TH_LOG("Did not return ESRCH for diverged sibling."); 2902 } 2903 ASSERT_EQ(-1, ret) { 2904 TH_LOG("Did not fail on diverged sibling."); 2905 } 2906 2907 /* Wake the threads */ 2908 pthread_mutex_lock(&self->mutex); 2909 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2910 TH_LOG("cond broadcast non-zero"); 2911 } 2912 pthread_mutex_unlock(&self->mutex); 2913 2914 /* Ensure they are both unkilled. */ 2915 PTHREAD_JOIN(self->sibling[0].tid, &status); 2916 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2917 PTHREAD_JOIN(self->sibling[1].tid, &status); 2918 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2919 } 2920 2921 TEST_F(TSYNC, two_siblings_not_under_filter) 2922 { 2923 long ret, sib; 2924 void *status; 2925 struct timespec delay = { .tv_nsec = 100000000 }; 2926 2927 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2928 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2929 } 2930 2931 /* 2932 * Sibling 0 will have its own seccomp policy 2933 * and Sibling 1 will not be under seccomp at 2934 * all. Sibling 1 will enter seccomp and 0 2935 * will cause failure. 2936 */ 2937 self->sibling[0].diverge = 1; 2938 tsync_start_sibling(&self->sibling[0]); 2939 tsync_start_sibling(&self->sibling[1]); 2940 2941 while (self->sibling_count < TSYNC_SIBLINGS) { 2942 sem_wait(&self->started); 2943 self->sibling_count++; 2944 } 2945 2946 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2947 ASSERT_NE(ENOSYS, errno) { 2948 TH_LOG("Kernel does not support seccomp syscall!"); 2949 } 2950 ASSERT_EQ(0, ret) { 2951 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2952 } 2953 2954 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2955 &self->apply_prog); 2956 ASSERT_EQ(ret, self->sibling[0].system_tid) { 2957 TH_LOG("Did not fail on diverged sibling."); 2958 } 2959 sib = 1; 2960 if (ret == self->sibling[0].system_tid) 2961 sib = 0; 2962 2963 pthread_mutex_lock(&self->mutex); 2964 2965 /* Increment the other siblings num_waits so we can clean up 2966 * the one we just saw. 2967 */ 2968 self->sibling[!sib].num_waits += 1; 2969 2970 /* Signal the thread to clean up*/ 2971 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2972 TH_LOG("cond broadcast non-zero"); 2973 } 2974 pthread_mutex_unlock(&self->mutex); 2975 PTHREAD_JOIN(self->sibling[sib].tid, &status); 2976 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2977 /* Poll for actual task death. pthread_join doesn't guarantee it. */ 2978 while (!kill(self->sibling[sib].system_tid, 0)) 2979 nanosleep(&delay, NULL); 2980 /* Switch to the remaining sibling */ 2981 sib = !sib; 2982 2983 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2984 &self->apply_prog); 2985 ASSERT_EQ(0, ret) { 2986 TH_LOG("Expected the remaining sibling to sync"); 2987 }; 2988 2989 pthread_mutex_lock(&self->mutex); 2990 2991 /* If remaining sibling didn't have a chance to wake up during 2992 * the first broadcast, manually reduce the num_waits now. 2993 */ 2994 if (self->sibling[sib].num_waits > 1) 2995 self->sibling[sib].num_waits = 1; 2996 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2997 TH_LOG("cond broadcast non-zero"); 2998 } 2999 pthread_mutex_unlock(&self->mutex); 3000 PTHREAD_JOIN(self->sibling[sib].tid, &status); 3001 EXPECT_EQ(0, (long)status); 3002 /* Poll for actual task death. pthread_join doesn't guarantee it. */ 3003 while (!kill(self->sibling[sib].system_tid, 0)) 3004 nanosleep(&delay, NULL); 3005 3006 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 3007 &self->apply_prog); 3008 ASSERT_EQ(0, ret); /* just us chickens */ 3009 } 3010 3011 /* Make sure restarted syscalls are seen directly as "restart_syscall". */ 3012 TEST(syscall_restart) 3013 { 3014 long ret; 3015 unsigned long msg; 3016 pid_t child_pid; 3017 int pipefd[2]; 3018 int status; 3019 siginfo_t info = { }; 3020 struct sock_filter filter[] = { 3021 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 3022 offsetof(struct seccomp_data, nr)), 3023 3024 #ifdef __NR_sigreturn 3025 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 7, 0), 3026 #endif 3027 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 6, 0), 3028 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 5, 0), 3029 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 4, 0), 3030 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 5, 0), 3031 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_clock_nanosleep, 4, 0), 3032 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0), 3033 3034 /* Allow __NR_write for easy logging. */ 3035 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1), 3036 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3037 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 3038 /* The nanosleep jump target. */ 3039 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100), 3040 /* The restart_syscall jump target. */ 3041 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200), 3042 }; 3043 struct sock_fprog prog = { 3044 .len = (unsigned short)ARRAY_SIZE(filter), 3045 .filter = filter, 3046 }; 3047 #if defined(__arm__) 3048 struct utsname utsbuf; 3049 #endif 3050 3051 ASSERT_EQ(0, pipe(pipefd)); 3052 3053 child_pid = fork(); 3054 ASSERT_LE(0, child_pid); 3055 if (child_pid == 0) { 3056 /* Child uses EXPECT not ASSERT to deliver status correctly. */ 3057 char buf = ' '; 3058 struct timespec timeout = { }; 3059 3060 /* Attach parent as tracer and stop. */ 3061 EXPECT_EQ(0, ptrace(PTRACE_TRACEME)); 3062 EXPECT_EQ(0, raise(SIGSTOP)); 3063 3064 EXPECT_EQ(0, close(pipefd[1])); 3065 3066 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 3067 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3068 } 3069 3070 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 3071 EXPECT_EQ(0, ret) { 3072 TH_LOG("Failed to install filter!"); 3073 } 3074 3075 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) { 3076 TH_LOG("Failed to read() sync from parent"); 3077 } 3078 EXPECT_EQ('.', buf) { 3079 TH_LOG("Failed to get sync data from read()"); 3080 } 3081 3082 /* Start nanosleep to be interrupted. */ 3083 timeout.tv_sec = 1; 3084 errno = 0; 3085 EXPECT_EQ(0, nanosleep(&timeout, NULL)) { 3086 TH_LOG("Call to nanosleep() failed (errno %d: %s)", 3087 errno, strerror(errno)); 3088 } 3089 3090 /* Read final sync from parent. */ 3091 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) { 3092 TH_LOG("Failed final read() from parent"); 3093 } 3094 EXPECT_EQ('!', buf) { 3095 TH_LOG("Failed to get final data from read()"); 3096 } 3097 3098 /* Directly report the status of our test harness results. */ 3099 syscall(__NR_exit, _metadata->exit_code); 3100 } 3101 EXPECT_EQ(0, close(pipefd[0])); 3102 3103 /* Attach to child, setup options, and release. */ 3104 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3105 ASSERT_EQ(true, WIFSTOPPED(status)); 3106 ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL, 3107 PTRACE_O_TRACESECCOMP)); 3108 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3109 ASSERT_EQ(1, write(pipefd[1], ".", 1)); 3110 3111 /* Wait for nanosleep() to start. */ 3112 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3113 ASSERT_EQ(true, WIFSTOPPED(status)); 3114 ASSERT_EQ(SIGTRAP, WSTOPSIG(status)); 3115 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16)); 3116 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg)); 3117 ASSERT_EQ(0x100, msg); 3118 ret = get_syscall(_metadata, child_pid); 3119 EXPECT_TRUE(ret == __NR_nanosleep || ret == __NR_clock_nanosleep); 3120 3121 /* Might as well check siginfo for sanity while we're here. */ 3122 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info)); 3123 ASSERT_EQ(SIGTRAP, info.si_signo); 3124 ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code); 3125 EXPECT_EQ(0, info.si_errno); 3126 EXPECT_EQ(getuid(), info.si_uid); 3127 /* Verify signal delivery came from child (seccomp-triggered). */ 3128 EXPECT_EQ(child_pid, info.si_pid); 3129 3130 /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */ 3131 ASSERT_EQ(0, kill(child_pid, SIGSTOP)); 3132 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3133 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3134 ASSERT_EQ(true, WIFSTOPPED(status)); 3135 ASSERT_EQ(SIGSTOP, WSTOPSIG(status)); 3136 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info)); 3137 /* 3138 * There is no siginfo on SIGSTOP any more, so we can't verify 3139 * signal delivery came from parent now (getpid() == info.si_pid). 3140 * https://lkml.kernel.org/r/CAGXu5jJaZAOzP1qFz66tYrtbuywqb+UN2SOA1VLHpCCOiYvYeg@mail.gmail.com 3141 * At least verify the SIGSTOP via PTRACE_GETSIGINFO. 3142 */ 3143 EXPECT_EQ(SIGSTOP, info.si_signo); 3144 3145 /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */ 3146 ASSERT_EQ(0, kill(child_pid, SIGCONT)); 3147 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3148 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3149 ASSERT_EQ(true, WIFSTOPPED(status)); 3150 ASSERT_EQ(SIGCONT, WSTOPSIG(status)); 3151 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3152 3153 /* Wait for restart_syscall() to start. */ 3154 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3155 ASSERT_EQ(true, WIFSTOPPED(status)); 3156 ASSERT_EQ(SIGTRAP, WSTOPSIG(status)); 3157 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16)); 3158 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg)); 3159 3160 ASSERT_EQ(0x200, msg); 3161 ret = get_syscall(_metadata, child_pid); 3162 #if defined(__arm__) 3163 /* 3164 * - native ARM registers do NOT expose true syscall. 3165 * - compat ARM registers on ARM64 DO expose true syscall. 3166 * - values of utsbuf.machine include 'armv8l' or 'armb8b' 3167 * for ARM64 running in compat mode. 3168 */ 3169 ASSERT_EQ(0, uname(&utsbuf)); 3170 if ((strncmp(utsbuf.machine, "arm", 3) == 0) && 3171 (strncmp(utsbuf.machine, "armv8l", 6) != 0) && 3172 (strncmp(utsbuf.machine, "armv8b", 6) != 0)) { 3173 EXPECT_EQ(__NR_nanosleep, ret); 3174 } else 3175 #endif 3176 { 3177 EXPECT_EQ(__NR_restart_syscall, ret); 3178 } 3179 3180 /* Write again to end test. */ 3181 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3182 ASSERT_EQ(1, write(pipefd[1], "!", 1)); 3183 EXPECT_EQ(0, close(pipefd[1])); 3184 3185 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3186 if (WIFSIGNALED(status) || WEXITSTATUS(status)) 3187 _metadata->exit_code = KSFT_FAIL; 3188 } 3189 3190 TEST_SIGNAL(filter_flag_log, SIGSYS) 3191 { 3192 struct sock_filter allow_filter[] = { 3193 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3194 }; 3195 struct sock_filter kill_filter[] = { 3196 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 3197 offsetof(struct seccomp_data, nr)), 3198 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 3199 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 3200 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3201 }; 3202 struct sock_fprog allow_prog = { 3203 .len = (unsigned short)ARRAY_SIZE(allow_filter), 3204 .filter = allow_filter, 3205 }; 3206 struct sock_fprog kill_prog = { 3207 .len = (unsigned short)ARRAY_SIZE(kill_filter), 3208 .filter = kill_filter, 3209 }; 3210 long ret; 3211 pid_t parent = getppid(); 3212 3213 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3214 ASSERT_EQ(0, ret); 3215 3216 /* Verify that the FILTER_FLAG_LOG flag isn't accepted in strict mode */ 3217 ret = seccomp(SECCOMP_SET_MODE_STRICT, SECCOMP_FILTER_FLAG_LOG, 3218 &allow_prog); 3219 ASSERT_NE(ENOSYS, errno) { 3220 TH_LOG("Kernel does not support seccomp syscall!"); 3221 } 3222 EXPECT_NE(0, ret) { 3223 TH_LOG("Kernel accepted FILTER_FLAG_LOG flag in strict mode!"); 3224 } 3225 EXPECT_EQ(EINVAL, errno) { 3226 TH_LOG("Kernel returned unexpected errno for FILTER_FLAG_LOG flag in strict mode!"); 3227 } 3228 3229 /* Verify that a simple, permissive filter can be added with no flags */ 3230 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog); 3231 EXPECT_EQ(0, ret); 3232 3233 /* See if the same filter can be added with the FILTER_FLAG_LOG flag */ 3234 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG, 3235 &allow_prog); 3236 ASSERT_NE(EINVAL, errno) { 3237 TH_LOG("Kernel does not support the FILTER_FLAG_LOG flag!"); 3238 } 3239 EXPECT_EQ(0, ret); 3240 3241 /* Ensure that the kill filter works with the FILTER_FLAG_LOG flag */ 3242 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG, 3243 &kill_prog); 3244 EXPECT_EQ(0, ret); 3245 3246 EXPECT_EQ(parent, syscall(__NR_getppid)); 3247 /* getpid() should never return. */ 3248 EXPECT_EQ(0, syscall(__NR_getpid)); 3249 } 3250 3251 TEST(get_action_avail) 3252 { 3253 __u32 actions[] = { SECCOMP_RET_KILL_THREAD, SECCOMP_RET_TRAP, 3254 SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE, 3255 SECCOMP_RET_LOG, SECCOMP_RET_ALLOW }; 3256 __u32 unknown_action = 0x10000000U; 3257 int i; 3258 long ret; 3259 3260 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[0]); 3261 ASSERT_NE(ENOSYS, errno) { 3262 TH_LOG("Kernel does not support seccomp syscall!"); 3263 } 3264 ASSERT_NE(EINVAL, errno) { 3265 TH_LOG("Kernel does not support SECCOMP_GET_ACTION_AVAIL operation!"); 3266 } 3267 EXPECT_EQ(ret, 0); 3268 3269 for (i = 0; i < ARRAY_SIZE(actions); i++) { 3270 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[i]); 3271 EXPECT_EQ(ret, 0) { 3272 TH_LOG("Expected action (0x%X) not available!", 3273 actions[i]); 3274 } 3275 } 3276 3277 /* Check that an unknown action is handled properly (EOPNOTSUPP) */ 3278 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &unknown_action); 3279 EXPECT_EQ(ret, -1); 3280 EXPECT_EQ(errno, EOPNOTSUPP); 3281 } 3282 3283 TEST(get_metadata) 3284 { 3285 pid_t pid; 3286 int pipefd[2]; 3287 char buf; 3288 struct seccomp_metadata md; 3289 long ret; 3290 3291 /* Only real root can get metadata. */ 3292 if (geteuid()) { 3293 SKIP(return, "get_metadata requires real root"); 3294 return; 3295 } 3296 3297 ASSERT_EQ(0, pipe(pipefd)); 3298 3299 pid = fork(); 3300 ASSERT_GE(pid, 0); 3301 if (pid == 0) { 3302 struct sock_filter filter[] = { 3303 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3304 }; 3305 struct sock_fprog prog = { 3306 .len = (unsigned short)ARRAY_SIZE(filter), 3307 .filter = filter, 3308 }; 3309 3310 /* one with log, one without */ 3311 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 3312 SECCOMP_FILTER_FLAG_LOG, &prog)); 3313 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog)); 3314 3315 EXPECT_EQ(0, close(pipefd[0])); 3316 ASSERT_EQ(1, write(pipefd[1], "1", 1)); 3317 ASSERT_EQ(0, close(pipefd[1])); 3318 3319 while (1) 3320 sleep(100); 3321 } 3322 3323 ASSERT_EQ(0, close(pipefd[1])); 3324 ASSERT_EQ(1, read(pipefd[0], &buf, 1)); 3325 3326 ASSERT_EQ(0, ptrace(PTRACE_ATTACH, pid)); 3327 ASSERT_EQ(pid, waitpid(pid, NULL, 0)); 3328 3329 /* Past here must not use ASSERT or child process is never killed. */ 3330 3331 md.filter_off = 0; 3332 errno = 0; 3333 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md); 3334 EXPECT_EQ(sizeof(md), ret) { 3335 if (errno == EINVAL) 3336 SKIP(goto skip, "Kernel does not support PTRACE_SECCOMP_GET_METADATA (missing CONFIG_CHECKPOINT_RESTORE?)"); 3337 } 3338 3339 EXPECT_EQ(md.flags, SECCOMP_FILTER_FLAG_LOG); 3340 EXPECT_EQ(md.filter_off, 0); 3341 3342 md.filter_off = 1; 3343 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md); 3344 EXPECT_EQ(sizeof(md), ret); 3345 EXPECT_EQ(md.flags, 0); 3346 EXPECT_EQ(md.filter_off, 1); 3347 3348 skip: 3349 ASSERT_EQ(0, kill(pid, SIGKILL)); 3350 } 3351 3352 static int user_notif_syscall(int nr, unsigned int flags) 3353 { 3354 struct sock_filter filter[] = { 3355 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 3356 offsetof(struct seccomp_data, nr)), 3357 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, nr, 0, 1), 3358 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_USER_NOTIF), 3359 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3360 }; 3361 3362 struct sock_fprog prog = { 3363 .len = (unsigned short)ARRAY_SIZE(filter), 3364 .filter = filter, 3365 }; 3366 3367 return seccomp(SECCOMP_SET_MODE_FILTER, flags, &prog); 3368 } 3369 3370 #define USER_NOTIF_MAGIC INT_MAX 3371 TEST(user_notification_basic) 3372 { 3373 pid_t pid; 3374 long ret; 3375 int status, listener; 3376 struct seccomp_notif req = {}; 3377 struct seccomp_notif_resp resp = {}; 3378 struct pollfd pollfd; 3379 3380 struct sock_filter filter[] = { 3381 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3382 }; 3383 struct sock_fprog prog = { 3384 .len = (unsigned short)ARRAY_SIZE(filter), 3385 .filter = filter, 3386 }; 3387 3388 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3389 ASSERT_EQ(0, ret) { 3390 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3391 } 3392 3393 pid = fork(); 3394 ASSERT_GE(pid, 0); 3395 3396 /* Check that we get -ENOSYS with no listener attached */ 3397 if (pid == 0) { 3398 if (user_notif_syscall(__NR_getppid, 0) < 0) 3399 exit(1); 3400 ret = syscall(__NR_getppid); 3401 exit(ret >= 0 || errno != ENOSYS); 3402 } 3403 3404 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3405 EXPECT_EQ(true, WIFEXITED(status)); 3406 EXPECT_EQ(0, WEXITSTATUS(status)); 3407 3408 /* Add some no-op filters for grins. */ 3409 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3410 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3411 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3412 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3413 3414 /* Check that the basic notification machinery works */ 3415 listener = user_notif_syscall(__NR_getppid, 3416 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3417 ASSERT_GE(listener, 0); 3418 3419 /* Installing a second listener in the chain should EBUSY */ 3420 EXPECT_EQ(user_notif_syscall(__NR_getppid, 3421 SECCOMP_FILTER_FLAG_NEW_LISTENER), 3422 -1); 3423 EXPECT_EQ(errno, EBUSY); 3424 3425 pid = fork(); 3426 ASSERT_GE(pid, 0); 3427 3428 if (pid == 0) { 3429 ret = syscall(__NR_getppid); 3430 exit(ret != USER_NOTIF_MAGIC); 3431 } 3432 3433 pollfd.fd = listener; 3434 pollfd.events = POLLIN | POLLOUT; 3435 3436 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3437 EXPECT_EQ(pollfd.revents, POLLIN); 3438 3439 /* Test that we can't pass garbage to the kernel. */ 3440 memset(&req, 0, sizeof(req)); 3441 req.pid = -1; 3442 errno = 0; 3443 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req); 3444 EXPECT_EQ(-1, ret); 3445 EXPECT_EQ(EINVAL, errno); 3446 3447 if (ret) { 3448 req.pid = 0; 3449 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3450 } 3451 3452 pollfd.fd = listener; 3453 pollfd.events = POLLIN | POLLOUT; 3454 3455 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3456 EXPECT_EQ(pollfd.revents, POLLOUT); 3457 3458 EXPECT_EQ(req.data.nr, __NR_getppid); 3459 3460 resp.id = req.id; 3461 resp.error = 0; 3462 resp.val = USER_NOTIF_MAGIC; 3463 3464 /* check that we make sure flags == 0 */ 3465 resp.flags = 1; 3466 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3467 EXPECT_EQ(errno, EINVAL); 3468 3469 resp.flags = 0; 3470 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3471 3472 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3473 EXPECT_EQ(true, WIFEXITED(status)); 3474 EXPECT_EQ(0, WEXITSTATUS(status)); 3475 } 3476 3477 TEST(user_notification_with_tsync) 3478 { 3479 int ret; 3480 unsigned int flags; 3481 3482 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3483 ASSERT_EQ(0, ret) { 3484 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3485 } 3486 3487 /* these were exclusive */ 3488 flags = SECCOMP_FILTER_FLAG_NEW_LISTENER | 3489 SECCOMP_FILTER_FLAG_TSYNC; 3490 ASSERT_EQ(-1, user_notif_syscall(__NR_getppid, flags)); 3491 ASSERT_EQ(EINVAL, errno); 3492 3493 /* but now they're not */ 3494 flags |= SECCOMP_FILTER_FLAG_TSYNC_ESRCH; 3495 ret = user_notif_syscall(__NR_getppid, flags); 3496 close(ret); 3497 ASSERT_LE(0, ret); 3498 } 3499 3500 TEST(user_notification_kill_in_middle) 3501 { 3502 pid_t pid; 3503 long ret; 3504 int listener; 3505 struct seccomp_notif req = {}; 3506 struct seccomp_notif_resp resp = {}; 3507 3508 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3509 ASSERT_EQ(0, ret) { 3510 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3511 } 3512 3513 listener = user_notif_syscall(__NR_getppid, 3514 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3515 ASSERT_GE(listener, 0); 3516 3517 /* 3518 * Check that nothing bad happens when we kill the task in the middle 3519 * of a syscall. 3520 */ 3521 pid = fork(); 3522 ASSERT_GE(pid, 0); 3523 3524 if (pid == 0) { 3525 ret = syscall(__NR_getppid); 3526 exit(ret != USER_NOTIF_MAGIC); 3527 } 3528 3529 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3530 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), 0); 3531 3532 EXPECT_EQ(kill(pid, SIGKILL), 0); 3533 EXPECT_EQ(waitpid(pid, NULL, 0), pid); 3534 3535 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), -1); 3536 3537 resp.id = req.id; 3538 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp); 3539 EXPECT_EQ(ret, -1); 3540 EXPECT_EQ(errno, ENOENT); 3541 } 3542 3543 static int handled = -1; 3544 3545 static void signal_handler(int signal) 3546 { 3547 if (write(handled, "c", 1) != 1) 3548 perror("write from signal"); 3549 } 3550 3551 static void signal_handler_nop(int signal) 3552 { 3553 } 3554 3555 TEST(user_notification_signal) 3556 { 3557 pid_t pid; 3558 long ret; 3559 int status, listener, sk_pair[2]; 3560 struct seccomp_notif req = {}; 3561 struct seccomp_notif_resp resp = {}; 3562 char c; 3563 3564 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3565 ASSERT_EQ(0, ret) { 3566 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3567 } 3568 3569 ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0); 3570 3571 listener = user_notif_syscall(__NR_gettid, 3572 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3573 ASSERT_GE(listener, 0); 3574 3575 pid = fork(); 3576 ASSERT_GE(pid, 0); 3577 3578 if (pid == 0) { 3579 close(sk_pair[0]); 3580 handled = sk_pair[1]; 3581 if (signal(SIGUSR1, signal_handler) == SIG_ERR) { 3582 perror("signal"); 3583 exit(1); 3584 } 3585 /* 3586 * ERESTARTSYS behavior is a bit hard to test, because we need 3587 * to rely on a signal that has not yet been handled. Let's at 3588 * least check that the error code gets propagated through, and 3589 * hope that it doesn't break when there is actually a signal :) 3590 */ 3591 ret = syscall(__NR_gettid); 3592 exit(!(ret == -1 && errno == 512)); 3593 } 3594 3595 close(sk_pair[1]); 3596 3597 memset(&req, 0, sizeof(req)); 3598 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3599 3600 EXPECT_EQ(kill(pid, SIGUSR1), 0); 3601 3602 /* 3603 * Make sure the signal really is delivered, which means we're not 3604 * stuck in the user notification code any more and the notification 3605 * should be dead. 3606 */ 3607 EXPECT_EQ(read(sk_pair[0], &c, 1), 1); 3608 3609 resp.id = req.id; 3610 resp.error = -EPERM; 3611 resp.val = 0; 3612 3613 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3614 EXPECT_EQ(errno, ENOENT); 3615 3616 memset(&req, 0, sizeof(req)); 3617 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3618 3619 resp.id = req.id; 3620 resp.error = -512; /* -ERESTARTSYS */ 3621 resp.val = 0; 3622 3623 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3624 3625 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3626 EXPECT_EQ(true, WIFEXITED(status)); 3627 EXPECT_EQ(0, WEXITSTATUS(status)); 3628 } 3629 3630 TEST(user_notification_closed_listener) 3631 { 3632 pid_t pid; 3633 long ret; 3634 int status, listener; 3635 3636 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3637 ASSERT_EQ(0, ret) { 3638 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3639 } 3640 3641 listener = user_notif_syscall(__NR_getppid, 3642 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3643 ASSERT_GE(listener, 0); 3644 3645 /* 3646 * Check that we get an ENOSYS when the listener is closed. 3647 */ 3648 pid = fork(); 3649 ASSERT_GE(pid, 0); 3650 if (pid == 0) { 3651 close(listener); 3652 ret = syscall(__NR_getppid); 3653 exit(ret != -1 && errno != ENOSYS); 3654 } 3655 3656 close(listener); 3657 3658 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3659 EXPECT_EQ(true, WIFEXITED(status)); 3660 EXPECT_EQ(0, WEXITSTATUS(status)); 3661 } 3662 3663 /* 3664 * Check that a pid in a child namespace still shows up as valid in ours. 3665 */ 3666 TEST(user_notification_child_pid_ns) 3667 { 3668 pid_t pid; 3669 int status, listener; 3670 struct seccomp_notif req = {}; 3671 struct seccomp_notif_resp resp = {}; 3672 3673 ASSERT_EQ(unshare(CLONE_NEWUSER | CLONE_NEWPID), 0) { 3674 if (errno == EINVAL) 3675 SKIP(return, "kernel missing CLONE_NEWUSER support"); 3676 }; 3677 3678 listener = user_notif_syscall(__NR_getppid, 3679 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3680 ASSERT_GE(listener, 0); 3681 3682 pid = fork(); 3683 ASSERT_GE(pid, 0); 3684 3685 if (pid == 0) 3686 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3687 3688 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3689 EXPECT_EQ(req.pid, pid); 3690 3691 resp.id = req.id; 3692 resp.error = 0; 3693 resp.val = USER_NOTIF_MAGIC; 3694 3695 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3696 3697 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3698 EXPECT_EQ(true, WIFEXITED(status)); 3699 EXPECT_EQ(0, WEXITSTATUS(status)); 3700 close(listener); 3701 } 3702 3703 /* 3704 * Check that a pid in a sibling (i.e. unrelated) namespace shows up as 0, i.e. 3705 * invalid. 3706 */ 3707 TEST(user_notification_sibling_pid_ns) 3708 { 3709 pid_t pid, pid2; 3710 int status, listener; 3711 struct seccomp_notif req = {}; 3712 struct seccomp_notif_resp resp = {}; 3713 3714 ASSERT_EQ(prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0), 0) { 3715 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3716 } 3717 3718 listener = user_notif_syscall(__NR_getppid, 3719 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3720 ASSERT_GE(listener, 0); 3721 3722 pid = fork(); 3723 ASSERT_GE(pid, 0); 3724 3725 if (pid == 0) { 3726 ASSERT_EQ(unshare(CLONE_NEWPID), 0) { 3727 if (errno == EPERM) 3728 SKIP(return, "CLONE_NEWPID requires CAP_SYS_ADMIN"); 3729 else if (errno == EINVAL) 3730 SKIP(return, "CLONE_NEWPID is invalid (missing CONFIG_PID_NS?)"); 3731 } 3732 3733 pid2 = fork(); 3734 ASSERT_GE(pid2, 0); 3735 3736 if (pid2 == 0) 3737 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3738 3739 EXPECT_EQ(waitpid(pid2, &status, 0), pid2); 3740 EXPECT_EQ(true, WIFEXITED(status)); 3741 EXPECT_EQ(0, WEXITSTATUS(status)); 3742 exit(WEXITSTATUS(status)); 3743 } 3744 3745 /* Create the sibling ns, and sibling in it. */ 3746 ASSERT_EQ(unshare(CLONE_NEWPID), 0) { 3747 if (errno == EPERM) 3748 SKIP(return, "CLONE_NEWPID requires CAP_SYS_ADMIN"); 3749 else if (errno == EINVAL) 3750 SKIP(return, "CLONE_NEWPID is invalid (missing CONFIG_PID_NS?)"); 3751 } 3752 ASSERT_EQ(errno, 0); 3753 3754 pid2 = fork(); 3755 ASSERT_GE(pid2, 0); 3756 3757 if (pid2 == 0) { 3758 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3759 /* 3760 * The pid should be 0, i.e. the task is in some namespace that 3761 * we can't "see". 3762 */ 3763 EXPECT_EQ(req.pid, 0); 3764 3765 resp.id = req.id; 3766 resp.error = 0; 3767 resp.val = USER_NOTIF_MAGIC; 3768 3769 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3770 exit(0); 3771 } 3772 3773 close(listener); 3774 3775 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3776 EXPECT_EQ(true, WIFEXITED(status)); 3777 EXPECT_EQ(0, WEXITSTATUS(status)); 3778 3779 EXPECT_EQ(waitpid(pid2, &status, 0), pid2); 3780 EXPECT_EQ(true, WIFEXITED(status)); 3781 EXPECT_EQ(0, WEXITSTATUS(status)); 3782 } 3783 3784 TEST(user_notification_fault_recv) 3785 { 3786 pid_t pid; 3787 int status, listener; 3788 struct seccomp_notif req = {}; 3789 struct seccomp_notif_resp resp = {}; 3790 3791 ASSERT_EQ(unshare(CLONE_NEWUSER), 0) { 3792 if (errno == EINVAL) 3793 SKIP(return, "kernel missing CLONE_NEWUSER support"); 3794 } 3795 3796 listener = user_notif_syscall(__NR_getppid, 3797 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3798 ASSERT_GE(listener, 0); 3799 3800 pid = fork(); 3801 ASSERT_GE(pid, 0); 3802 3803 if (pid == 0) 3804 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3805 3806 /* Do a bad recv() */ 3807 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, NULL), -1); 3808 EXPECT_EQ(errno, EFAULT); 3809 3810 /* We should still be able to receive this notification, though. */ 3811 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3812 EXPECT_EQ(req.pid, pid); 3813 3814 resp.id = req.id; 3815 resp.error = 0; 3816 resp.val = USER_NOTIF_MAGIC; 3817 3818 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3819 3820 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3821 EXPECT_EQ(true, WIFEXITED(status)); 3822 EXPECT_EQ(0, WEXITSTATUS(status)); 3823 } 3824 3825 TEST(seccomp_get_notif_sizes) 3826 { 3827 struct seccomp_notif_sizes sizes; 3828 3829 ASSERT_EQ(seccomp(SECCOMP_GET_NOTIF_SIZES, 0, &sizes), 0); 3830 EXPECT_EQ(sizes.seccomp_notif, sizeof(struct seccomp_notif)); 3831 EXPECT_EQ(sizes.seccomp_notif_resp, sizeof(struct seccomp_notif_resp)); 3832 } 3833 3834 TEST(user_notification_continue) 3835 { 3836 pid_t pid; 3837 long ret; 3838 int status, listener; 3839 struct seccomp_notif req = {}; 3840 struct seccomp_notif_resp resp = {}; 3841 struct pollfd pollfd; 3842 3843 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3844 ASSERT_EQ(0, ret) { 3845 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3846 } 3847 3848 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER); 3849 ASSERT_GE(listener, 0); 3850 3851 pid = fork(); 3852 ASSERT_GE(pid, 0); 3853 3854 if (pid == 0) { 3855 int dup_fd, pipe_fds[2]; 3856 pid_t self; 3857 3858 ASSERT_GE(pipe(pipe_fds), 0); 3859 3860 dup_fd = dup(pipe_fds[0]); 3861 ASSERT_GE(dup_fd, 0); 3862 EXPECT_NE(pipe_fds[0], dup_fd); 3863 3864 self = getpid(); 3865 ASSERT_EQ(filecmp(self, self, pipe_fds[0], dup_fd), 0); 3866 exit(0); 3867 } 3868 3869 pollfd.fd = listener; 3870 pollfd.events = POLLIN | POLLOUT; 3871 3872 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3873 EXPECT_EQ(pollfd.revents, POLLIN); 3874 3875 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3876 3877 pollfd.fd = listener; 3878 pollfd.events = POLLIN | POLLOUT; 3879 3880 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3881 EXPECT_EQ(pollfd.revents, POLLOUT); 3882 3883 EXPECT_EQ(req.data.nr, __NR_dup); 3884 3885 resp.id = req.id; 3886 resp.flags = SECCOMP_USER_NOTIF_FLAG_CONTINUE; 3887 3888 /* 3889 * Verify that setting SECCOMP_USER_NOTIF_FLAG_CONTINUE enforces other 3890 * args be set to 0. 3891 */ 3892 resp.error = 0; 3893 resp.val = USER_NOTIF_MAGIC; 3894 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3895 EXPECT_EQ(errno, EINVAL); 3896 3897 resp.error = USER_NOTIF_MAGIC; 3898 resp.val = 0; 3899 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3900 EXPECT_EQ(errno, EINVAL); 3901 3902 resp.error = 0; 3903 resp.val = 0; 3904 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0) { 3905 if (errno == EINVAL) 3906 SKIP(goto skip, "Kernel does not support SECCOMP_USER_NOTIF_FLAG_CONTINUE"); 3907 } 3908 3909 skip: 3910 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3911 EXPECT_EQ(true, WIFEXITED(status)); 3912 EXPECT_EQ(0, WEXITSTATUS(status)) { 3913 if (WEXITSTATUS(status) == 2) { 3914 SKIP(return, "Kernel does not support kcmp() syscall"); 3915 return; 3916 } 3917 } 3918 } 3919 3920 TEST(user_notification_filter_empty) 3921 { 3922 pid_t pid; 3923 long ret; 3924 int status; 3925 struct pollfd pollfd; 3926 struct __clone_args args = { 3927 .flags = CLONE_FILES, 3928 .exit_signal = SIGCHLD, 3929 }; 3930 3931 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3932 ASSERT_EQ(0, ret) { 3933 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3934 } 3935 3936 if (__NR_clone3 < 0) 3937 SKIP(return, "Test not built with clone3 support"); 3938 3939 pid = sys_clone3(&args, sizeof(args)); 3940 ASSERT_GE(pid, 0); 3941 3942 if (pid == 0) { 3943 int listener; 3944 3945 listener = user_notif_syscall(__NR_mknodat, SECCOMP_FILTER_FLAG_NEW_LISTENER); 3946 if (listener < 0) 3947 _exit(EXIT_FAILURE); 3948 3949 if (dup2(listener, 200) != 200) 3950 _exit(EXIT_FAILURE); 3951 3952 close(listener); 3953 3954 _exit(EXIT_SUCCESS); 3955 } 3956 3957 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3958 EXPECT_EQ(true, WIFEXITED(status)); 3959 EXPECT_EQ(0, WEXITSTATUS(status)); 3960 3961 /* 3962 * The seccomp filter has become unused so we should be notified once 3963 * the kernel gets around to cleaning up task struct. 3964 */ 3965 pollfd.fd = 200; 3966 pollfd.events = POLLHUP; 3967 3968 EXPECT_GT(poll(&pollfd, 1, 2000), 0); 3969 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0); 3970 } 3971 3972 TEST(user_ioctl_notification_filter_empty) 3973 { 3974 pid_t pid; 3975 long ret; 3976 int status, p[2]; 3977 struct __clone_args args = { 3978 .flags = CLONE_FILES, 3979 .exit_signal = SIGCHLD, 3980 }; 3981 struct seccomp_notif req = {}; 3982 3983 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3984 ASSERT_EQ(0, ret) { 3985 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3986 } 3987 3988 if (__NR_clone3 < 0) 3989 SKIP(return, "Test not built with clone3 support"); 3990 3991 ASSERT_EQ(0, pipe(p)); 3992 3993 pid = sys_clone3(&args, sizeof(args)); 3994 ASSERT_GE(pid, 0); 3995 3996 if (pid == 0) { 3997 int listener; 3998 3999 listener = user_notif_syscall(__NR_mknodat, SECCOMP_FILTER_FLAG_NEW_LISTENER); 4000 if (listener < 0) 4001 _exit(EXIT_FAILURE); 4002 4003 if (dup2(listener, 200) != 200) 4004 _exit(EXIT_FAILURE); 4005 close(p[1]); 4006 close(listener); 4007 sleep(1); 4008 4009 _exit(EXIT_SUCCESS); 4010 } 4011 if (read(p[0], &status, 1) != 0) 4012 _exit(EXIT_SUCCESS); 4013 close(p[0]); 4014 /* 4015 * The seccomp filter has become unused so we should be notified once 4016 * the kernel gets around to cleaning up task struct. 4017 */ 4018 EXPECT_EQ(ioctl(200, SECCOMP_IOCTL_NOTIF_RECV, &req), -1); 4019 EXPECT_EQ(errno, ENOENT); 4020 4021 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4022 EXPECT_EQ(true, WIFEXITED(status)); 4023 EXPECT_EQ(0, WEXITSTATUS(status)); 4024 } 4025 4026 static void *do_thread(void *data) 4027 { 4028 return NULL; 4029 } 4030 4031 TEST(user_notification_filter_empty_threaded) 4032 { 4033 pid_t pid; 4034 long ret; 4035 int status; 4036 struct pollfd pollfd; 4037 struct __clone_args args = { 4038 .flags = CLONE_FILES, 4039 .exit_signal = SIGCHLD, 4040 }; 4041 4042 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4043 ASSERT_EQ(0, ret) { 4044 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4045 } 4046 4047 if (__NR_clone3 < 0) 4048 SKIP(return, "Test not built with clone3 support"); 4049 4050 pid = sys_clone3(&args, sizeof(args)); 4051 ASSERT_GE(pid, 0); 4052 4053 if (pid == 0) { 4054 pid_t pid1, pid2; 4055 int listener, status; 4056 pthread_t thread; 4057 4058 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER); 4059 if (listener < 0) 4060 _exit(EXIT_FAILURE); 4061 4062 if (dup2(listener, 200) != 200) 4063 _exit(EXIT_FAILURE); 4064 4065 close(listener); 4066 4067 pid1 = fork(); 4068 if (pid1 < 0) 4069 _exit(EXIT_FAILURE); 4070 4071 if (pid1 == 0) 4072 _exit(EXIT_SUCCESS); 4073 4074 pid2 = fork(); 4075 if (pid2 < 0) 4076 _exit(EXIT_FAILURE); 4077 4078 if (pid2 == 0) 4079 _exit(EXIT_SUCCESS); 4080 4081 if (pthread_create(&thread, NULL, do_thread, NULL) || 4082 pthread_join(thread, NULL)) 4083 _exit(EXIT_FAILURE); 4084 4085 if (pthread_create(&thread, NULL, do_thread, NULL) || 4086 pthread_join(thread, NULL)) 4087 _exit(EXIT_FAILURE); 4088 4089 if (waitpid(pid1, &status, 0) != pid1 || !WIFEXITED(status) || 4090 WEXITSTATUS(status)) 4091 _exit(EXIT_FAILURE); 4092 4093 if (waitpid(pid2, &status, 0) != pid2 || !WIFEXITED(status) || 4094 WEXITSTATUS(status)) 4095 _exit(EXIT_FAILURE); 4096 4097 exit(EXIT_SUCCESS); 4098 } 4099 4100 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4101 EXPECT_EQ(true, WIFEXITED(status)); 4102 EXPECT_EQ(0, WEXITSTATUS(status)); 4103 4104 /* 4105 * The seccomp filter has become unused so we should be notified once 4106 * the kernel gets around to cleaning up task struct. 4107 */ 4108 pollfd.fd = 200; 4109 pollfd.events = POLLHUP; 4110 4111 EXPECT_GT(poll(&pollfd, 1, 2000), 0); 4112 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0); 4113 } 4114 4115 4116 int get_next_fd(int prev_fd) 4117 { 4118 for (int i = prev_fd + 1; i < FD_SETSIZE; ++i) { 4119 if (fcntl(i, F_GETFD) == -1) 4120 return i; 4121 } 4122 _exit(EXIT_FAILURE); 4123 } 4124 4125 TEST(user_notification_addfd) 4126 { 4127 pid_t pid; 4128 long ret; 4129 int status, listener, memfd, fd, nextfd; 4130 struct seccomp_notif_addfd addfd = {}; 4131 struct seccomp_notif_addfd_small small = {}; 4132 struct seccomp_notif_addfd_big big = {}; 4133 struct seccomp_notif req = {}; 4134 struct seccomp_notif_resp resp = {}; 4135 /* 100 ms */ 4136 struct timespec delay = { .tv_nsec = 100000000 }; 4137 4138 /* There may be arbitrary already-open fds at test start. */ 4139 memfd = memfd_create("test", 0); 4140 ASSERT_GE(memfd, 0); 4141 nextfd = get_next_fd(memfd); 4142 4143 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4144 ASSERT_EQ(0, ret) { 4145 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4146 } 4147 4148 /* fd: 4 */ 4149 /* Check that the basic notification machinery works */ 4150 listener = user_notif_syscall(__NR_getppid, 4151 SECCOMP_FILTER_FLAG_NEW_LISTENER); 4152 ASSERT_EQ(listener, nextfd); 4153 nextfd = get_next_fd(nextfd); 4154 4155 pid = fork(); 4156 ASSERT_GE(pid, 0); 4157 4158 if (pid == 0) { 4159 /* fds will be added and this value is expected */ 4160 if (syscall(__NR_getppid) != USER_NOTIF_MAGIC) 4161 exit(1); 4162 4163 /* Atomic addfd+send is received here. Check it is a valid fd */ 4164 if (fcntl(syscall(__NR_getppid), F_GETFD) == -1) 4165 exit(1); 4166 4167 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 4168 } 4169 4170 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4171 4172 addfd.srcfd = memfd; 4173 addfd.newfd = 0; 4174 addfd.id = req.id; 4175 addfd.flags = 0x0; 4176 4177 /* Verify bad newfd_flags cannot be set */ 4178 addfd.newfd_flags = ~O_CLOEXEC; 4179 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4180 EXPECT_EQ(errno, EINVAL); 4181 addfd.newfd_flags = O_CLOEXEC; 4182 4183 /* Verify bad flags cannot be set */ 4184 addfd.flags = 0xff; 4185 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4186 EXPECT_EQ(errno, EINVAL); 4187 addfd.flags = 0; 4188 4189 /* Verify that remote_fd cannot be set without setting flags */ 4190 addfd.newfd = 1; 4191 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4192 EXPECT_EQ(errno, EINVAL); 4193 addfd.newfd = 0; 4194 4195 /* Verify small size cannot be set */ 4196 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_SMALL, &small), -1); 4197 EXPECT_EQ(errno, EINVAL); 4198 4199 /* Verify we can't send bits filled in unknown buffer area */ 4200 memset(&big, 0xAA, sizeof(big)); 4201 big.addfd = addfd; 4202 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big), -1); 4203 EXPECT_EQ(errno, E2BIG); 4204 4205 4206 /* Verify we can set an arbitrary remote fd */ 4207 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd); 4208 EXPECT_EQ(fd, nextfd); 4209 nextfd = get_next_fd(nextfd); 4210 EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0); 4211 4212 /* Verify we can set an arbitrary remote fd with large size */ 4213 memset(&big, 0x0, sizeof(big)); 4214 big.addfd = addfd; 4215 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big); 4216 EXPECT_EQ(fd, nextfd); 4217 nextfd = get_next_fd(nextfd); 4218 4219 /* Verify we can set a specific remote fd */ 4220 addfd.newfd = 42; 4221 addfd.flags = SECCOMP_ADDFD_FLAG_SETFD; 4222 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd); 4223 EXPECT_EQ(fd, 42); 4224 EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0); 4225 4226 /* Resume syscall */ 4227 resp.id = req.id; 4228 resp.error = 0; 4229 resp.val = USER_NOTIF_MAGIC; 4230 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4231 4232 /* 4233 * This sets the ID of the ADD FD to the last request plus 1. The 4234 * notification ID increments 1 per notification. 4235 */ 4236 addfd.id = req.id + 1; 4237 4238 /* This spins until the underlying notification is generated */ 4239 while (ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd) != -1 && 4240 errno != -EINPROGRESS) 4241 nanosleep(&delay, NULL); 4242 4243 memset(&req, 0, sizeof(req)); 4244 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4245 ASSERT_EQ(addfd.id, req.id); 4246 4247 /* Verify we can do an atomic addfd and send */ 4248 addfd.newfd = 0; 4249 addfd.flags = SECCOMP_ADDFD_FLAG_SEND; 4250 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd); 4251 /* 4252 * Child has earlier "low" fds and now 42, so we expect the next 4253 * lowest available fd to be assigned here. 4254 */ 4255 EXPECT_EQ(fd, nextfd); 4256 nextfd = get_next_fd(nextfd); 4257 ASSERT_EQ(filecmp(getpid(), pid, memfd, fd), 0); 4258 4259 /* 4260 * This sets the ID of the ADD FD to the last request plus 1. The 4261 * notification ID increments 1 per notification. 4262 */ 4263 addfd.id = req.id + 1; 4264 4265 /* This spins until the underlying notification is generated */ 4266 while (ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd) != -1 && 4267 errno != -EINPROGRESS) 4268 nanosleep(&delay, NULL); 4269 4270 memset(&req, 0, sizeof(req)); 4271 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4272 ASSERT_EQ(addfd.id, req.id); 4273 4274 resp.id = req.id; 4275 resp.error = 0; 4276 resp.val = USER_NOTIF_MAGIC; 4277 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4278 4279 /* Wait for child to finish. */ 4280 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4281 EXPECT_EQ(true, WIFEXITED(status)); 4282 EXPECT_EQ(0, WEXITSTATUS(status)); 4283 4284 close(memfd); 4285 } 4286 4287 TEST(user_notification_addfd_rlimit) 4288 { 4289 pid_t pid; 4290 long ret; 4291 int status, listener, memfd; 4292 struct seccomp_notif_addfd addfd = {}; 4293 struct seccomp_notif req = {}; 4294 struct seccomp_notif_resp resp = {}; 4295 const struct rlimit lim = { 4296 .rlim_cur = 0, 4297 .rlim_max = 0, 4298 }; 4299 4300 memfd = memfd_create("test", 0); 4301 ASSERT_GE(memfd, 0); 4302 4303 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4304 ASSERT_EQ(0, ret) { 4305 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4306 } 4307 4308 /* Check that the basic notification machinery works */ 4309 listener = user_notif_syscall(__NR_getppid, 4310 SECCOMP_FILTER_FLAG_NEW_LISTENER); 4311 ASSERT_GE(listener, 0); 4312 4313 pid = fork(); 4314 ASSERT_GE(pid, 0); 4315 4316 if (pid == 0) 4317 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 4318 4319 4320 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4321 4322 ASSERT_EQ(prlimit(pid, RLIMIT_NOFILE, &lim, NULL), 0); 4323 4324 addfd.srcfd = memfd; 4325 addfd.newfd_flags = O_CLOEXEC; 4326 addfd.newfd = 0; 4327 addfd.id = req.id; 4328 addfd.flags = 0; 4329 4330 /* Should probably spot check /proc/sys/fs/file-nr */ 4331 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4332 EXPECT_EQ(errno, EMFILE); 4333 4334 addfd.flags = SECCOMP_ADDFD_FLAG_SEND; 4335 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4336 EXPECT_EQ(errno, EMFILE); 4337 4338 addfd.newfd = 100; 4339 addfd.flags = SECCOMP_ADDFD_FLAG_SETFD; 4340 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4341 EXPECT_EQ(errno, EBADF); 4342 4343 resp.id = req.id; 4344 resp.error = 0; 4345 resp.val = USER_NOTIF_MAGIC; 4346 4347 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4348 4349 /* Wait for child to finish. */ 4350 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4351 EXPECT_EQ(true, WIFEXITED(status)); 4352 EXPECT_EQ(0, WEXITSTATUS(status)); 4353 4354 close(memfd); 4355 } 4356 4357 #ifndef SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP 4358 #define SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP (1UL << 0) 4359 #define SECCOMP_IOCTL_NOTIF_SET_FLAGS SECCOMP_IOW(4, __u64) 4360 #endif 4361 4362 TEST(user_notification_sync) 4363 { 4364 struct seccomp_notif req = {}; 4365 struct seccomp_notif_resp resp = {}; 4366 int status, listener; 4367 pid_t pid; 4368 long ret; 4369 4370 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4371 ASSERT_EQ(0, ret) { 4372 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4373 } 4374 4375 listener = user_notif_syscall(__NR_getppid, 4376 SECCOMP_FILTER_FLAG_NEW_LISTENER); 4377 ASSERT_GE(listener, 0); 4378 4379 /* Try to set invalid flags. */ 4380 EXPECT_SYSCALL_RETURN(-EINVAL, 4381 ioctl(listener, SECCOMP_IOCTL_NOTIF_SET_FLAGS, 0xffffffff, 0)); 4382 4383 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SET_FLAGS, 4384 SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP, 0), 0); 4385 4386 pid = fork(); 4387 ASSERT_GE(pid, 0); 4388 if (pid == 0) { 4389 ret = syscall(__NR_getppid); 4390 ASSERT_EQ(ret, USER_NOTIF_MAGIC) { 4391 _exit(1); 4392 } 4393 _exit(0); 4394 } 4395 4396 req.pid = 0; 4397 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4398 4399 ASSERT_EQ(req.data.nr, __NR_getppid); 4400 4401 resp.id = req.id; 4402 resp.error = 0; 4403 resp.val = USER_NOTIF_MAGIC; 4404 resp.flags = 0; 4405 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4406 4407 ASSERT_EQ(waitpid(pid, &status, 0), pid); 4408 ASSERT_EQ(status, 0); 4409 } 4410 4411 4412 /* Make sure PTRACE_O_SUSPEND_SECCOMP requires CAP_SYS_ADMIN. */ 4413 FIXTURE(O_SUSPEND_SECCOMP) { 4414 pid_t pid; 4415 }; 4416 4417 FIXTURE_SETUP(O_SUSPEND_SECCOMP) 4418 { 4419 ERRNO_FILTER(block_read, E2BIG); 4420 cap_value_t cap_list[] = { CAP_SYS_ADMIN }; 4421 cap_t caps; 4422 4423 self->pid = 0; 4424 4425 /* make sure we don't have CAP_SYS_ADMIN */ 4426 caps = cap_get_proc(); 4427 ASSERT_NE(NULL, caps); 4428 ASSERT_EQ(0, cap_set_flag(caps, CAP_EFFECTIVE, 1, cap_list, CAP_CLEAR)); 4429 ASSERT_EQ(0, cap_set_proc(caps)); 4430 cap_free(caps); 4431 4432 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)); 4433 ASSERT_EQ(0, prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_block_read)); 4434 4435 self->pid = fork(); 4436 ASSERT_GE(self->pid, 0); 4437 4438 if (self->pid == 0) { 4439 while (1) 4440 pause(); 4441 _exit(127); 4442 } 4443 } 4444 4445 FIXTURE_TEARDOWN(O_SUSPEND_SECCOMP) 4446 { 4447 if (self->pid) 4448 kill(self->pid, SIGKILL); 4449 } 4450 4451 TEST_F(O_SUSPEND_SECCOMP, setoptions) 4452 { 4453 int wstatus; 4454 4455 ASSERT_EQ(0, ptrace(PTRACE_ATTACH, self->pid, NULL, 0)); 4456 ASSERT_EQ(self->pid, wait(&wstatus)); 4457 ASSERT_EQ(-1, ptrace(PTRACE_SETOPTIONS, self->pid, NULL, PTRACE_O_SUSPEND_SECCOMP)); 4458 if (errno == EINVAL) 4459 SKIP(return, "Kernel does not support PTRACE_O_SUSPEND_SECCOMP (missing CONFIG_CHECKPOINT_RESTORE?)"); 4460 ASSERT_EQ(EPERM, errno); 4461 } 4462 4463 TEST_F(O_SUSPEND_SECCOMP, seize) 4464 { 4465 int ret; 4466 4467 ret = ptrace(PTRACE_SEIZE, self->pid, NULL, PTRACE_O_SUSPEND_SECCOMP); 4468 ASSERT_EQ(-1, ret); 4469 if (errno == EINVAL) 4470 SKIP(return, "Kernel does not support PTRACE_O_SUSPEND_SECCOMP (missing CONFIG_CHECKPOINT_RESTORE?)"); 4471 ASSERT_EQ(EPERM, errno); 4472 } 4473 4474 /* 4475 * get_nth - Get the nth, space separated entry in a file. 4476 * 4477 * Returns the length of the read field. 4478 * Throws error if field is zero-lengthed. 4479 */ 4480 static ssize_t get_nth(struct __test_metadata *_metadata, const char *path, 4481 const unsigned int position, char **entry) 4482 { 4483 char *line = NULL; 4484 unsigned int i; 4485 ssize_t nread; 4486 size_t len = 0; 4487 FILE *f; 4488 4489 f = fopen(path, "r"); 4490 ASSERT_NE(f, NULL) { 4491 TH_LOG("Could not open %s: %s", path, strerror(errno)); 4492 } 4493 4494 for (i = 0; i < position; i++) { 4495 nread = getdelim(&line, &len, ' ', f); 4496 ASSERT_GE(nread, 0) { 4497 TH_LOG("Failed to read %d entry in file %s", i, path); 4498 } 4499 } 4500 fclose(f); 4501 4502 ASSERT_GT(nread, 0) { 4503 TH_LOG("Entry in file %s had zero length", path); 4504 } 4505 4506 *entry = line; 4507 return nread - 1; 4508 } 4509 4510 /* For a given PID, get the task state (D, R, etc...) */ 4511 static char get_proc_stat(struct __test_metadata *_metadata, pid_t pid) 4512 { 4513 char proc_path[100] = {0}; 4514 char status; 4515 char *line; 4516 4517 snprintf(proc_path, sizeof(proc_path), "/proc/%d/stat", pid); 4518 ASSERT_EQ(get_nth(_metadata, proc_path, 3, &line), 1); 4519 4520 status = *line; 4521 free(line); 4522 4523 return status; 4524 } 4525 4526 TEST(user_notification_fifo) 4527 { 4528 struct seccomp_notif_resp resp = {}; 4529 struct seccomp_notif req = {}; 4530 int i, status, listener; 4531 pid_t pid, pids[3]; 4532 __u64 baseid; 4533 long ret; 4534 /* 100 ms */ 4535 struct timespec delay = { .tv_nsec = 100000000 }; 4536 4537 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4538 ASSERT_EQ(0, ret) { 4539 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4540 } 4541 4542 /* Setup a listener */ 4543 listener = user_notif_syscall(__NR_getppid, 4544 SECCOMP_FILTER_FLAG_NEW_LISTENER); 4545 ASSERT_GE(listener, 0); 4546 4547 pid = fork(); 4548 ASSERT_GE(pid, 0); 4549 4550 if (pid == 0) { 4551 ret = syscall(__NR_getppid); 4552 exit(ret != USER_NOTIF_MAGIC); 4553 } 4554 4555 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4556 baseid = req.id + 1; 4557 4558 resp.id = req.id; 4559 resp.error = 0; 4560 resp.val = USER_NOTIF_MAGIC; 4561 4562 /* check that we make sure flags == 0 */ 4563 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4564 4565 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4566 EXPECT_EQ(true, WIFEXITED(status)); 4567 EXPECT_EQ(0, WEXITSTATUS(status)); 4568 4569 /* Start children, and generate notifications */ 4570 for (i = 0; i < ARRAY_SIZE(pids); i++) { 4571 pid = fork(); 4572 if (pid == 0) { 4573 ret = syscall(__NR_getppid); 4574 exit(ret != USER_NOTIF_MAGIC); 4575 } 4576 pids[i] = pid; 4577 } 4578 4579 /* This spins until all of the children are sleeping */ 4580 restart_wait: 4581 for (i = 0; i < ARRAY_SIZE(pids); i++) { 4582 if (get_proc_stat(_metadata, pids[i]) != 'S') { 4583 nanosleep(&delay, NULL); 4584 goto restart_wait; 4585 } 4586 } 4587 4588 /* Read the notifications in order (and respond) */ 4589 for (i = 0; i < ARRAY_SIZE(pids); i++) { 4590 memset(&req, 0, sizeof(req)); 4591 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4592 EXPECT_EQ(req.id, baseid + i); 4593 resp.id = req.id; 4594 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4595 } 4596 4597 /* Make sure notifications were received */ 4598 for (i = 0; i < ARRAY_SIZE(pids); i++) { 4599 EXPECT_EQ(waitpid(pids[i], &status, 0), pids[i]); 4600 EXPECT_EQ(true, WIFEXITED(status)); 4601 EXPECT_EQ(0, WEXITSTATUS(status)); 4602 } 4603 } 4604 4605 /* get_proc_syscall - Get the syscall in progress for a given pid 4606 * 4607 * Returns the current syscall number for a given process 4608 * Returns -1 if not in syscall (running or blocked) 4609 */ 4610 static long get_proc_syscall(struct __test_metadata *_metadata, int pid) 4611 { 4612 char proc_path[100] = {0}; 4613 long ret = -1; 4614 ssize_t nread; 4615 char *line; 4616 4617 snprintf(proc_path, sizeof(proc_path), "/proc/%d/syscall", pid); 4618 nread = get_nth(_metadata, proc_path, 1, &line); 4619 ASSERT_GT(nread, 0); 4620 4621 if (!strncmp("running", line, MIN(7, nread))) 4622 ret = strtol(line, NULL, 16); 4623 4624 free(line); 4625 return ret; 4626 } 4627 4628 /* Ensure non-fatal signals prior to receive are unmodified */ 4629 TEST(user_notification_wait_killable_pre_notification) 4630 { 4631 struct sigaction new_action = { 4632 .sa_handler = signal_handler, 4633 }; 4634 int listener, status, sk_pair[2]; 4635 pid_t pid; 4636 long ret; 4637 char c; 4638 /* 100 ms */ 4639 struct timespec delay = { .tv_nsec = 100000000 }; 4640 4641 ASSERT_EQ(sigemptyset(&new_action.sa_mask), 0); 4642 4643 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4644 ASSERT_EQ(0, ret) 4645 { 4646 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4647 } 4648 4649 ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0); 4650 4651 listener = user_notif_syscall( 4652 __NR_getppid, SECCOMP_FILTER_FLAG_NEW_LISTENER | 4653 SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV); 4654 ASSERT_GE(listener, 0); 4655 4656 /* 4657 * Check that we can kill the process with SIGUSR1 prior to receiving 4658 * the notification. SIGUSR1 is wired up to a custom signal handler, 4659 * and make sure it gets called. 4660 */ 4661 pid = fork(); 4662 ASSERT_GE(pid, 0); 4663 4664 if (pid == 0) { 4665 close(sk_pair[0]); 4666 handled = sk_pair[1]; 4667 4668 /* Setup the non-fatal sigaction without SA_RESTART */ 4669 if (sigaction(SIGUSR1, &new_action, NULL)) { 4670 perror("sigaction"); 4671 exit(1); 4672 } 4673 4674 ret = syscall(__NR_getppid); 4675 /* Make sure we got a return from a signal interruption */ 4676 exit(ret != -1 || errno != EINTR); 4677 } 4678 4679 /* 4680 * Make sure we've gotten to the seccomp user notification wait 4681 * from getppid prior to sending any signals 4682 */ 4683 while (get_proc_syscall(_metadata, pid) != __NR_getppid && 4684 get_proc_stat(_metadata, pid) != 'S') 4685 nanosleep(&delay, NULL); 4686 4687 /* Send non-fatal kill signal */ 4688 EXPECT_EQ(kill(pid, SIGUSR1), 0); 4689 4690 /* wait for process to exit (exit checks for EINTR) */ 4691 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4692 EXPECT_EQ(true, WIFEXITED(status)); 4693 EXPECT_EQ(0, WEXITSTATUS(status)); 4694 4695 EXPECT_EQ(read(sk_pair[0], &c, 1), 1); 4696 } 4697 4698 /* Ensure non-fatal signals after receive are blocked */ 4699 TEST(user_notification_wait_killable) 4700 { 4701 struct sigaction new_action = { 4702 .sa_handler = signal_handler, 4703 }; 4704 struct seccomp_notif_resp resp = {}; 4705 struct seccomp_notif req = {}; 4706 int listener, status, sk_pair[2]; 4707 pid_t pid; 4708 long ret; 4709 char c; 4710 /* 100 ms */ 4711 struct timespec delay = { .tv_nsec = 100000000 }; 4712 4713 ASSERT_EQ(sigemptyset(&new_action.sa_mask), 0); 4714 4715 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4716 ASSERT_EQ(0, ret) 4717 { 4718 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4719 } 4720 4721 ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0); 4722 4723 listener = user_notif_syscall( 4724 __NR_getppid, SECCOMP_FILTER_FLAG_NEW_LISTENER | 4725 SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV); 4726 ASSERT_GE(listener, 0); 4727 4728 pid = fork(); 4729 ASSERT_GE(pid, 0); 4730 4731 if (pid == 0) { 4732 close(sk_pair[0]); 4733 handled = sk_pair[1]; 4734 4735 /* Setup the sigaction without SA_RESTART */ 4736 if (sigaction(SIGUSR1, &new_action, NULL)) { 4737 perror("sigaction"); 4738 exit(1); 4739 } 4740 4741 /* Make sure that the syscall is completed (no EINTR) */ 4742 ret = syscall(__NR_getppid); 4743 exit(ret != USER_NOTIF_MAGIC); 4744 } 4745 4746 /* 4747 * Get the notification, to make move the notifying process into a 4748 * non-preemptible (TASK_KILLABLE) state. 4749 */ 4750 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4751 /* Send non-fatal kill signal */ 4752 EXPECT_EQ(kill(pid, SIGUSR1), 0); 4753 4754 /* 4755 * Make sure the task enters moves to TASK_KILLABLE by waiting for 4756 * D (Disk Sleep) state after receiving non-fatal signal. 4757 */ 4758 while (get_proc_stat(_metadata, pid) != 'D') 4759 nanosleep(&delay, NULL); 4760 4761 resp.id = req.id; 4762 resp.val = USER_NOTIF_MAGIC; 4763 /* Make sure the notification is found and able to be replied to */ 4764 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4765 4766 /* 4767 * Make sure that the signal handler does get called once we're back in 4768 * userspace. 4769 */ 4770 EXPECT_EQ(read(sk_pair[0], &c, 1), 1); 4771 /* wait for process to exit (exit checks for USER_NOTIF_MAGIC) */ 4772 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4773 EXPECT_EQ(true, WIFEXITED(status)); 4774 EXPECT_EQ(0, WEXITSTATUS(status)); 4775 } 4776 4777 /* Ensure fatal signals after receive are not blocked */ 4778 TEST(user_notification_wait_killable_fatal) 4779 { 4780 struct seccomp_notif req = {}; 4781 int listener, status; 4782 pid_t pid; 4783 long ret; 4784 /* 100 ms */ 4785 struct timespec delay = { .tv_nsec = 100000000 }; 4786 4787 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4788 ASSERT_EQ(0, ret) 4789 { 4790 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4791 } 4792 4793 listener = user_notif_syscall( 4794 __NR_getppid, SECCOMP_FILTER_FLAG_NEW_LISTENER | 4795 SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV); 4796 ASSERT_GE(listener, 0); 4797 4798 pid = fork(); 4799 ASSERT_GE(pid, 0); 4800 4801 if (pid == 0) { 4802 /* This should never complete as it should get a SIGTERM */ 4803 syscall(__NR_getppid); 4804 exit(1); 4805 } 4806 4807 while (get_proc_stat(_metadata, pid) != 'S') 4808 nanosleep(&delay, NULL); 4809 4810 /* 4811 * Get the notification, to make move the notifying process into a 4812 * non-preemptible (TASK_KILLABLE) state. 4813 */ 4814 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4815 /* Kill the process with a fatal signal */ 4816 EXPECT_EQ(kill(pid, SIGTERM), 0); 4817 4818 /* 4819 * Wait for the process to exit, and make sure the process terminated 4820 * due to the SIGTERM signal. 4821 */ 4822 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4823 EXPECT_EQ(true, WIFSIGNALED(status)); 4824 EXPECT_EQ(SIGTERM, WTERMSIG(status)); 4825 } 4826 4827 /* Ensure signals after the reply do not interrupt */ 4828 TEST(user_notification_wait_killable_after_reply) 4829 { 4830 int i, max_iter = 100000; 4831 int listener, status; 4832 int pipe_fds[2]; 4833 pid_t pid; 4834 long ret; 4835 4836 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4837 ASSERT_EQ(0, ret) 4838 { 4839 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4840 } 4841 4842 listener = user_notif_syscall( 4843 __NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER | 4844 SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV); 4845 ASSERT_GE(listener, 0); 4846 4847 /* 4848 * Used to count invocations. One token is transferred from the child 4849 * to the parent per syscall invocation, the parent tries to take 4850 * one token per successful RECV. If the syscall is restarted after 4851 * RECV the parent will try to get two tokens while the child only 4852 * provided one. 4853 */ 4854 ASSERT_EQ(pipe(pipe_fds), 0); 4855 4856 pid = fork(); 4857 ASSERT_GE(pid, 0); 4858 4859 if (pid == 0) { 4860 struct sigaction new_action = { 4861 .sa_handler = signal_handler_nop, 4862 .sa_flags = SA_RESTART, 4863 }; 4864 struct itimerval timer = { 4865 .it_value = { .tv_usec = 1000 }, 4866 .it_interval = { .tv_usec = 1000 }, 4867 }; 4868 char c = 'a'; 4869 4870 close(pipe_fds[0]); 4871 4872 /* Setup the sigaction with SA_RESTART */ 4873 if (sigaction(SIGALRM, &new_action, NULL)) { 4874 perror("sigaction"); 4875 exit(1); 4876 } 4877 4878 /* 4879 * Kill with SIGALRM repeatedly, to try to hit the race when 4880 * handling the syscall. 4881 */ 4882 if (setitimer(ITIMER_REAL, &timer, NULL) < 0) 4883 perror("setitimer"); 4884 4885 for (i = 0; i < max_iter; ++i) { 4886 int fd; 4887 4888 /* Send one token per iteration to catch repeats. */ 4889 if (write(pipe_fds[1], &c, sizeof(c)) != 1) { 4890 perror("write"); 4891 exit(1); 4892 } 4893 4894 fd = syscall(__NR_dup, 0); 4895 if (fd < 0) { 4896 perror("dup"); 4897 exit(1); 4898 } 4899 close(fd); 4900 } 4901 4902 exit(0); 4903 } 4904 4905 close(pipe_fds[1]); 4906 4907 for (i = 0; i < max_iter; ++i) { 4908 struct seccomp_notif req = {}; 4909 struct seccomp_notif_addfd addfd = {}; 4910 struct pollfd pfd = { 4911 .fd = pipe_fds[0], 4912 .events = POLLIN, 4913 }; 4914 char c; 4915 4916 /* 4917 * Try to receive one token. If it failed, one child syscall 4918 * was restarted after RECV and needed to be handled twice. 4919 */ 4920 ASSERT_EQ(poll(&pfd, 1, 1000), 1) 4921 kill(pid, SIGKILL); 4922 4923 ASSERT_EQ(read(pipe_fds[0], &c, sizeof(c)), 1) 4924 kill(pid, SIGKILL); 4925 4926 /* 4927 * Get the notification, reply to it as fast as possible to test 4928 * whether the child wrongly skips going into the non-preemptible 4929 * (TASK_KILLABLE) state. 4930 */ 4931 do 4932 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req); 4933 while (ret < 0 && errno == ENOENT); /* Accept interruptions before RECV */ 4934 ASSERT_EQ(ret, 0) 4935 kill(pid, SIGKILL); 4936 4937 addfd.id = req.id; 4938 addfd.flags = SECCOMP_ADDFD_FLAG_SEND; 4939 addfd.srcfd = 0; 4940 ASSERT_GE(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), 0) 4941 kill(pid, SIGKILL); 4942 } 4943 4944 /* 4945 * Wait for the process to exit, and make sure the process terminated 4946 * with a zero exit code.. 4947 */ 4948 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4949 EXPECT_EQ(true, WIFEXITED(status)); 4950 EXPECT_EQ(0, WEXITSTATUS(status)); 4951 } 4952 4953 struct tsync_vs_thread_leader_args { 4954 pthread_t leader; 4955 }; 4956 4957 static void *tsync_vs_dead_thread_leader_sibling(void *_args) 4958 { 4959 struct sock_filter allow_filter[] = { 4960 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 4961 }; 4962 struct sock_fprog allow_prog = { 4963 .len = (unsigned short)ARRAY_SIZE(allow_filter), 4964 .filter = allow_filter, 4965 }; 4966 struct tsync_vs_thread_leader_args *args = _args; 4967 void *retval; 4968 long ret; 4969 4970 ret = pthread_join(args->leader, &retval); 4971 if (ret) 4972 exit(1); 4973 if (retval != _args) 4974 exit(2); 4975 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, &allow_prog); 4976 if (ret) 4977 exit(3); 4978 4979 exit(0); 4980 } 4981 4982 /* 4983 * Ensure that a dead thread leader doesn't prevent installing new filters with 4984 * SECCOMP_FILTER_FLAG_TSYNC from other threads. 4985 */ 4986 TEST(tsync_vs_dead_thread_leader) 4987 { 4988 int status; 4989 pid_t pid; 4990 long ret; 4991 4992 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4993 ASSERT_EQ(0, ret) { 4994 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4995 } 4996 4997 pid = fork(); 4998 ASSERT_GE(pid, 0); 4999 5000 if (pid == 0) { 5001 struct sock_filter allow_filter[] = { 5002 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 5003 }; 5004 struct sock_fprog allow_prog = { 5005 .len = (unsigned short)ARRAY_SIZE(allow_filter), 5006 .filter = allow_filter, 5007 }; 5008 struct tsync_vs_thread_leader_args *args; 5009 pthread_t sibling; 5010 5011 args = malloc(sizeof(*args)); 5012 ASSERT_NE(NULL, args); 5013 args->leader = pthread_self(); 5014 5015 ret = pthread_create(&sibling, NULL, 5016 tsync_vs_dead_thread_leader_sibling, args); 5017 ASSERT_EQ(0, ret); 5018 5019 /* Install a new filter just to the leader thread. */ 5020 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog); 5021 ASSERT_EQ(0, ret); 5022 pthread_exit(args); 5023 exit(1); 5024 } 5025 5026 EXPECT_EQ(pid, waitpid(pid, &status, 0)); 5027 EXPECT_EQ(0, status); 5028 } 5029 5030 noinline int probed(void) 5031 { 5032 return 1; 5033 } 5034 5035 static int parse_uint_from_file(const char *file, const char *fmt) 5036 { 5037 int err = -1, ret; 5038 FILE *f; 5039 5040 f = fopen(file, "re"); 5041 if (f) { 5042 err = fscanf(f, fmt, &ret); 5043 fclose(f); 5044 } 5045 return err == 1 ? ret : err; 5046 } 5047 5048 static int determine_uprobe_perf_type(void) 5049 { 5050 const char *file = "/sys/bus/event_source/devices/uprobe/type"; 5051 5052 return parse_uint_from_file(file, "%d\n"); 5053 } 5054 5055 static int determine_uprobe_retprobe_bit(void) 5056 { 5057 const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe"; 5058 5059 return parse_uint_from_file(file, "config:%d\n"); 5060 } 5061 5062 static ssize_t get_uprobe_offset(const void *addr) 5063 { 5064 size_t start, base, end; 5065 bool found = false; 5066 char buf[256]; 5067 FILE *f; 5068 5069 f = fopen("/proc/self/maps", "r"); 5070 if (!f) 5071 return -1; 5072 5073 while (fscanf(f, "%zx-%zx %s %zx %*[^\n]\n", &start, &end, buf, &base) == 4) { 5074 if (buf[2] == 'x' && (uintptr_t)addr >= start && (uintptr_t)addr < end) { 5075 found = true; 5076 break; 5077 } 5078 } 5079 fclose(f); 5080 return found ? (uintptr_t)addr - start + base : -1; 5081 } 5082 5083 FIXTURE(URETPROBE) { 5084 int fd; 5085 }; 5086 5087 FIXTURE_VARIANT(URETPROBE) { 5088 /* 5089 * All of the URETPROBE behaviors can be tested with either 5090 * uretprobe attached or not 5091 */ 5092 bool attach; 5093 }; 5094 5095 FIXTURE_VARIANT_ADD(URETPROBE, attached) { 5096 .attach = true, 5097 }; 5098 5099 FIXTURE_VARIANT_ADD(URETPROBE, not_attached) { 5100 .attach = false, 5101 }; 5102 5103 FIXTURE_SETUP(URETPROBE) 5104 { 5105 const size_t attr_sz = sizeof(struct perf_event_attr); 5106 struct perf_event_attr attr; 5107 ssize_t offset; 5108 int type, bit; 5109 5110 #ifndef __NR_uretprobe 5111 SKIP(return, "__NR_uretprobe syscall not defined"); 5112 #endif 5113 5114 if (!variant->attach) 5115 return; 5116 5117 memset(&attr, 0, attr_sz); 5118 5119 type = determine_uprobe_perf_type(); 5120 ASSERT_GE(type, 0); 5121 bit = determine_uprobe_retprobe_bit(); 5122 ASSERT_GE(bit, 0); 5123 offset = get_uprobe_offset(probed); 5124 ASSERT_GE(offset, 0); 5125 5126 attr.config |= 1 << bit; 5127 attr.size = attr_sz; 5128 attr.type = type; 5129 attr.config1 = ptr_to_u64("/proc/self/exe"); 5130 attr.config2 = offset; 5131 5132 self->fd = syscall(__NR_perf_event_open, &attr, 5133 getpid() /* pid */, -1 /* cpu */, -1 /* group_fd */, 5134 PERF_FLAG_FD_CLOEXEC); 5135 } 5136 5137 FIXTURE_TEARDOWN(URETPROBE) 5138 { 5139 /* we could call close(self->fd), but we'd need extra filter for 5140 * that and since we are calling _exit right away.. 5141 */ 5142 } 5143 5144 static int run_probed_with_filter(struct sock_fprog *prog) 5145 { 5146 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) || 5147 seccomp(SECCOMP_SET_MODE_FILTER, 0, prog)) { 5148 return -1; 5149 } 5150 5151 probed(); 5152 return 0; 5153 } 5154 5155 TEST_F(URETPROBE, uretprobe_default_allow) 5156 { 5157 struct sock_filter filter[] = { 5158 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 5159 }; 5160 struct sock_fprog prog = { 5161 .len = (unsigned short)ARRAY_SIZE(filter), 5162 .filter = filter, 5163 }; 5164 5165 ASSERT_EQ(0, run_probed_with_filter(&prog)); 5166 } 5167 5168 TEST_F(URETPROBE, uretprobe_default_block) 5169 { 5170 struct sock_filter filter[] = { 5171 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 5172 offsetof(struct seccomp_data, nr)), 5173 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit_group, 1, 0), 5174 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 5175 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 5176 }; 5177 struct sock_fprog prog = { 5178 .len = (unsigned short)ARRAY_SIZE(filter), 5179 .filter = filter, 5180 }; 5181 5182 ASSERT_EQ(0, run_probed_with_filter(&prog)); 5183 } 5184 5185 TEST_F(URETPROBE, uretprobe_block_uretprobe_syscall) 5186 { 5187 struct sock_filter filter[] = { 5188 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 5189 offsetof(struct seccomp_data, nr)), 5190 #ifdef __NR_uretprobe 5191 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_uretprobe, 0, 1), 5192 #endif 5193 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 5194 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 5195 }; 5196 struct sock_fprog prog = { 5197 .len = (unsigned short)ARRAY_SIZE(filter), 5198 .filter = filter, 5199 }; 5200 5201 ASSERT_EQ(0, run_probed_with_filter(&prog)); 5202 } 5203 5204 TEST_F(URETPROBE, uretprobe_default_block_with_uretprobe_syscall) 5205 { 5206 struct sock_filter filter[] = { 5207 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 5208 offsetof(struct seccomp_data, nr)), 5209 #ifdef __NR_uretprobe 5210 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_uretprobe, 2, 0), 5211 #endif 5212 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit_group, 1, 0), 5213 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 5214 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 5215 }; 5216 struct sock_fprog prog = { 5217 .len = (unsigned short)ARRAY_SIZE(filter), 5218 .filter = filter, 5219 }; 5220 5221 ASSERT_EQ(0, run_probed_with_filter(&prog)); 5222 } 5223 5224 /* 5225 * TODO: 5226 * - expand NNP testing 5227 * - better arch-specific TRACE and TRAP handlers. 5228 * - endianness checking when appropriate 5229 * - 64-bit arg prodding 5230 * - arch value testing (x86 modes especially) 5231 * - verify that FILTER_FLAG_LOG filters generate log messages 5232 * - verify that RET_LOG generates log messages 5233 */ 5234 5235 TEST_HARNESS_MAIN 5236