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 __nocf_check 78 #define __nocf_check __attribute__((nocf_check)) 79 #endif 80 81 #ifndef __naked 82 #define __naked __attribute__((__naked__)) 83 #endif 84 85 #ifndef PR_SET_NO_NEW_PRIVS 86 #define PR_SET_NO_NEW_PRIVS 38 87 #define PR_GET_NO_NEW_PRIVS 39 88 #endif 89 90 #ifndef PR_SECCOMP_EXT 91 #define PR_SECCOMP_EXT 43 92 #endif 93 94 #ifndef SECCOMP_EXT_ACT 95 #define SECCOMP_EXT_ACT 1 96 #endif 97 98 #ifndef SECCOMP_EXT_ACT_TSYNC 99 #define SECCOMP_EXT_ACT_TSYNC 1 100 #endif 101 102 #ifndef SECCOMP_MODE_STRICT 103 #define SECCOMP_MODE_STRICT 1 104 #endif 105 106 #ifndef SECCOMP_MODE_FILTER 107 #define SECCOMP_MODE_FILTER 2 108 #endif 109 110 #ifndef SECCOMP_RET_ALLOW 111 struct seccomp_data { 112 int nr; 113 __u32 arch; 114 __u64 instruction_pointer; 115 __u64 args[6]; 116 }; 117 #endif 118 119 #ifndef SECCOMP_RET_KILL_PROCESS 120 #define SECCOMP_RET_KILL_PROCESS 0x80000000U /* kill the process */ 121 #define SECCOMP_RET_KILL_THREAD 0x00000000U /* kill the thread */ 122 #endif 123 #ifndef SECCOMP_RET_KILL 124 #define SECCOMP_RET_KILL SECCOMP_RET_KILL_THREAD 125 #define SECCOMP_RET_TRAP 0x00030000U /* disallow and force a SIGSYS */ 126 #define SECCOMP_RET_ERRNO 0x00050000U /* returns an errno */ 127 #define SECCOMP_RET_TRACE 0x7ff00000U /* pass to a tracer or disallow */ 128 #define SECCOMP_RET_ALLOW 0x7fff0000U /* allow */ 129 #endif 130 #ifndef SECCOMP_RET_LOG 131 #define SECCOMP_RET_LOG 0x7ffc0000U /* allow after logging */ 132 #endif 133 134 #ifndef __NR_seccomp 135 # if defined(__i386__) 136 # define __NR_seccomp 354 137 # elif defined(__x86_64__) 138 # define __NR_seccomp 317 139 # elif defined(__alpha__) 140 # define __NR_seccomp 514 141 # elif defined(__arm__) 142 # define __NR_seccomp 383 143 # elif defined(__aarch64__) 144 # define __NR_seccomp 277 145 # elif defined(__riscv) 146 # define __NR_seccomp 277 147 # elif defined(__csky__) 148 # define __NR_seccomp 277 149 # elif defined(__loongarch__) 150 # define __NR_seccomp 277 151 # elif defined(__hppa__) 152 # define __NR_seccomp 338 153 # elif defined(__powerpc__) 154 # define __NR_seccomp 358 155 # elif defined(__s390__) 156 # define __NR_seccomp 348 157 # elif defined(__xtensa__) 158 # define __NR_seccomp 337 159 # elif defined(__sh__) 160 # define __NR_seccomp 372 161 # elif defined(__mc68000__) 162 # define __NR_seccomp 380 163 # else 164 # warning "seccomp syscall number unknown for this architecture" 165 # define __NR_seccomp 0xffff 166 # endif 167 #endif 168 169 #ifndef __NR_uretprobe 170 # if defined(__x86_64__) 171 # define __NR_uretprobe 335 172 # endif 173 #endif 174 175 #ifndef __NR_uprobe 176 # if defined(__x86_64__) 177 # define __NR_uprobe 336 178 # endif 179 #endif 180 181 #ifndef SECCOMP_SET_MODE_STRICT 182 #define SECCOMP_SET_MODE_STRICT 0 183 #endif 184 185 #ifndef SECCOMP_SET_MODE_FILTER 186 #define SECCOMP_SET_MODE_FILTER 1 187 #endif 188 189 #ifndef SECCOMP_GET_ACTION_AVAIL 190 #define SECCOMP_GET_ACTION_AVAIL 2 191 #endif 192 193 #ifndef SECCOMP_GET_NOTIF_SIZES 194 #define SECCOMP_GET_NOTIF_SIZES 3 195 #endif 196 197 #ifndef SECCOMP_FILTER_FLAG_TSYNC 198 #define SECCOMP_FILTER_FLAG_TSYNC (1UL << 0) 199 #endif 200 201 #ifndef SECCOMP_FILTER_FLAG_LOG 202 #define SECCOMP_FILTER_FLAG_LOG (1UL << 1) 203 #endif 204 205 #ifndef SECCOMP_FILTER_FLAG_SPEC_ALLOW 206 #define SECCOMP_FILTER_FLAG_SPEC_ALLOW (1UL << 2) 207 #endif 208 209 #ifndef PTRACE_SECCOMP_GET_METADATA 210 #define PTRACE_SECCOMP_GET_METADATA 0x420d 211 212 struct seccomp_metadata { 213 __u64 filter_off; /* Input: which filter */ 214 __u64 flags; /* Output: filter's flags */ 215 }; 216 #endif 217 218 #ifndef SECCOMP_FILTER_FLAG_NEW_LISTENER 219 #define SECCOMP_FILTER_FLAG_NEW_LISTENER (1UL << 3) 220 #endif 221 222 #ifndef SECCOMP_RET_USER_NOTIF 223 #define SECCOMP_RET_USER_NOTIF 0x7fc00000U 224 225 #define SECCOMP_IOC_MAGIC '!' 226 #define SECCOMP_IO(nr) _IO(SECCOMP_IOC_MAGIC, nr) 227 #define SECCOMP_IOR(nr, type) _IOR(SECCOMP_IOC_MAGIC, nr, type) 228 #define SECCOMP_IOW(nr, type) _IOW(SECCOMP_IOC_MAGIC, nr, type) 229 #define SECCOMP_IOWR(nr, type) _IOWR(SECCOMP_IOC_MAGIC, nr, type) 230 231 /* Flags for seccomp notification fd ioctl. */ 232 #define SECCOMP_IOCTL_NOTIF_RECV SECCOMP_IOWR(0, struct seccomp_notif) 233 #define SECCOMP_IOCTL_NOTIF_SEND SECCOMP_IOWR(1, \ 234 struct seccomp_notif_resp) 235 #define SECCOMP_IOCTL_NOTIF_ID_VALID SECCOMP_IOW(2, __u64) 236 237 struct seccomp_notif { 238 __u64 id; 239 __u32 pid; 240 __u32 flags; 241 struct seccomp_data data; 242 }; 243 244 struct seccomp_notif_resp { 245 __u64 id; 246 __s64 val; 247 __s32 error; 248 __u32 flags; 249 }; 250 251 struct seccomp_notif_sizes { 252 __u16 seccomp_notif; 253 __u16 seccomp_notif_resp; 254 __u16 seccomp_data; 255 }; 256 #endif 257 258 #ifndef SECCOMP_IOCTL_NOTIF_ADDFD 259 /* On success, the return value is the remote process's added fd number */ 260 #define SECCOMP_IOCTL_NOTIF_ADDFD SECCOMP_IOW(3, \ 261 struct seccomp_notif_addfd) 262 263 /* valid flags for seccomp_notif_addfd */ 264 #define SECCOMP_ADDFD_FLAG_SETFD (1UL << 0) /* Specify remote fd */ 265 266 struct seccomp_notif_addfd { 267 __u64 id; 268 __u32 flags; 269 __u32 srcfd; 270 __u32 newfd; 271 __u32 newfd_flags; 272 }; 273 #endif 274 275 #ifndef SECCOMP_ADDFD_FLAG_SEND 276 #define SECCOMP_ADDFD_FLAG_SEND (1UL << 1) /* Addfd and return it, atomically */ 277 #endif 278 279 struct seccomp_notif_addfd_small { 280 __u64 id; 281 char weird[4]; 282 }; 283 #define SECCOMP_IOCTL_NOTIF_ADDFD_SMALL \ 284 SECCOMP_IOW(3, struct seccomp_notif_addfd_small) 285 286 struct seccomp_notif_addfd_big { 287 union { 288 struct seccomp_notif_addfd addfd; 289 char buf[sizeof(struct seccomp_notif_addfd) + 8]; 290 }; 291 }; 292 #define SECCOMP_IOCTL_NOTIF_ADDFD_BIG \ 293 SECCOMP_IOWR(3, struct seccomp_notif_addfd_big) 294 295 #ifndef PTRACE_EVENTMSG_SYSCALL_ENTRY 296 #define PTRACE_EVENTMSG_SYSCALL_ENTRY 1 297 #define PTRACE_EVENTMSG_SYSCALL_EXIT 2 298 #endif 299 300 #ifndef SECCOMP_USER_NOTIF_FLAG_CONTINUE 301 #define SECCOMP_USER_NOTIF_FLAG_CONTINUE 0x00000001 302 #endif 303 304 #ifndef SECCOMP_FILTER_FLAG_TSYNC_ESRCH 305 #define SECCOMP_FILTER_FLAG_TSYNC_ESRCH (1UL << 4) 306 #endif 307 308 #ifndef SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV 309 #define SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV (1UL << 5) 310 #endif 311 312 #ifndef seccomp 313 int seccomp(unsigned int op, unsigned int flags, void *args) 314 { 315 errno = 0; 316 return syscall(__NR_seccomp, op, flags, args); 317 } 318 #endif 319 320 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 321 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n])) 322 #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 323 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32)) 324 #else 325 #error "wut? Unknown __BYTE_ORDER__?!" 326 #endif 327 328 #define SIBLING_EXIT_UNKILLED 0xbadbeef 329 #define SIBLING_EXIT_FAILURE 0xbadface 330 #define SIBLING_EXIT_NEWPRIVS 0xbadfeed 331 332 static int __filecmp(pid_t pid1, pid_t pid2, int fd1, int fd2) 333 { 334 #ifdef __NR_kcmp 335 errno = 0; 336 return syscall(__NR_kcmp, pid1, pid2, KCMP_FILE, fd1, fd2); 337 #else 338 errno = ENOSYS; 339 return -1; 340 #endif 341 } 342 343 /* Have TH_LOG report actual location filecmp() is used. */ 344 #define filecmp(pid1, pid2, fd1, fd2) ({ \ 345 int _ret; \ 346 \ 347 _ret = __filecmp(pid1, pid2, fd1, fd2); \ 348 if (_ret != 0) { \ 349 if (_ret < 0 && errno == ENOSYS) { \ 350 TH_LOG("kcmp() syscall missing (test is less accurate)");\ 351 _ret = 0; \ 352 } \ 353 } \ 354 _ret; }) 355 356 TEST(kcmp) 357 { 358 int ret; 359 360 ret = __filecmp(getpid(), getpid(), 1, 1); 361 EXPECT_EQ(ret, 0); 362 if (ret != 0 && errno == ENOSYS) 363 SKIP(return, "Kernel does not support kcmp() (missing CONFIG_KCMP?)"); 364 } 365 366 TEST(mode_strict_support) 367 { 368 long ret; 369 370 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL); 371 ASSERT_EQ(0, ret) { 372 TH_LOG("Kernel does not support CONFIG_SECCOMP"); 373 } 374 syscall(__NR_exit, 0); 375 } 376 377 TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL) 378 { 379 long ret; 380 381 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL); 382 ASSERT_EQ(0, ret) { 383 TH_LOG("Kernel does not support CONFIG_SECCOMP"); 384 } 385 syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER, 386 NULL, NULL, NULL); 387 EXPECT_FALSE(true) { 388 TH_LOG("Unreachable!"); 389 } 390 } 391 392 /* Note! This doesn't test no new privs behavior */ 393 TEST(no_new_privs_support) 394 { 395 long ret; 396 397 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 398 EXPECT_EQ(0, ret) { 399 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 400 } 401 } 402 403 /* Tests kernel support by checking for a copy_from_user() fault on NULL. */ 404 TEST(mode_filter_support) 405 { 406 long ret; 407 408 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 409 ASSERT_EQ(0, ret) { 410 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 411 } 412 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL); 413 EXPECT_EQ(-1, ret); 414 EXPECT_EQ(EFAULT, errno) { 415 TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!"); 416 } 417 } 418 419 TEST(mode_filter_without_nnp) 420 { 421 struct sock_filter filter[] = { 422 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 423 }; 424 struct sock_fprog prog = { 425 .len = (unsigned short)ARRAY_SIZE(filter), 426 .filter = filter, 427 }; 428 long ret; 429 cap_t cap = cap_get_proc(); 430 cap_flag_value_t is_cap_sys_admin = 0; 431 432 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0); 433 ASSERT_LE(0, ret) { 434 TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS"); 435 } 436 errno = 0; 437 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 438 /* Succeeds with CAP_SYS_ADMIN, fails without */ 439 cap_get_flag(cap, CAP_SYS_ADMIN, CAP_EFFECTIVE, &is_cap_sys_admin); 440 if (!is_cap_sys_admin) { 441 EXPECT_EQ(-1, ret); 442 EXPECT_EQ(EACCES, errno); 443 } else { 444 EXPECT_EQ(0, ret); 445 } 446 } 447 448 #define MAX_INSNS_PER_PATH 32768 449 450 TEST(filter_size_limits) 451 { 452 int i; 453 int count = BPF_MAXINSNS + 1; 454 struct sock_filter allow[] = { 455 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 456 }; 457 struct sock_filter *filter; 458 struct sock_fprog prog = { }; 459 long ret; 460 461 filter = calloc(count, sizeof(*filter)); 462 ASSERT_NE(NULL, filter); 463 464 for (i = 0; i < count; i++) 465 filter[i] = allow[0]; 466 467 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 468 ASSERT_EQ(0, ret); 469 470 prog.filter = filter; 471 prog.len = count; 472 473 /* Too many filter instructions in a single filter. */ 474 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 475 ASSERT_NE(0, ret) { 476 TH_LOG("Installing %d insn filter was allowed", prog.len); 477 } 478 479 /* One less is okay, though. */ 480 prog.len -= 1; 481 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 482 ASSERT_EQ(0, ret) { 483 TH_LOG("Installing %d insn filter wasn't allowed", prog.len); 484 } 485 } 486 487 TEST(filter_chain_limits) 488 { 489 int i; 490 int count = BPF_MAXINSNS; 491 struct sock_filter allow[] = { 492 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 493 }; 494 struct sock_filter *filter; 495 struct sock_fprog prog = { }; 496 long ret; 497 498 filter = calloc(count, sizeof(*filter)); 499 ASSERT_NE(NULL, filter); 500 501 for (i = 0; i < count; i++) 502 filter[i] = allow[0]; 503 504 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 505 ASSERT_EQ(0, ret); 506 507 prog.filter = filter; 508 prog.len = 1; 509 510 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 511 ASSERT_EQ(0, ret); 512 513 prog.len = count; 514 515 /* Too many total filter instructions. */ 516 for (i = 0; i < MAX_INSNS_PER_PATH; i++) { 517 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 518 if (ret != 0) 519 break; 520 } 521 ASSERT_NE(0, ret) { 522 TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)", 523 i, count, i * (count + 4)); 524 } 525 } 526 527 TEST(mode_filter_cannot_move_to_strict) 528 { 529 struct sock_filter filter[] = { 530 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 531 }; 532 struct sock_fprog prog = { 533 .len = (unsigned short)ARRAY_SIZE(filter), 534 .filter = filter, 535 }; 536 long ret; 537 538 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 539 ASSERT_EQ(0, ret); 540 541 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 542 ASSERT_EQ(0, ret); 543 544 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0); 545 EXPECT_EQ(-1, ret); 546 EXPECT_EQ(EINVAL, errno); 547 } 548 549 550 TEST(mode_filter_get_seccomp) 551 { 552 struct sock_filter filter[] = { 553 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 554 }; 555 struct sock_fprog prog = { 556 .len = (unsigned short)ARRAY_SIZE(filter), 557 .filter = filter, 558 }; 559 long ret; 560 561 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 562 ASSERT_EQ(0, ret); 563 564 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0); 565 EXPECT_EQ(0, ret); 566 567 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 568 ASSERT_EQ(0, ret); 569 570 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0); 571 EXPECT_EQ(2, ret); 572 } 573 574 575 TEST(ALLOW_all) 576 { 577 struct sock_filter filter[] = { 578 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 579 }; 580 struct sock_fprog prog = { 581 .len = (unsigned short)ARRAY_SIZE(filter), 582 .filter = filter, 583 }; 584 long ret; 585 586 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 587 ASSERT_EQ(0, ret); 588 589 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 590 ASSERT_EQ(0, ret); 591 } 592 593 TEST(empty_prog) 594 { 595 struct sock_filter filter[] = { 596 }; 597 struct sock_fprog prog = { 598 .len = (unsigned short)ARRAY_SIZE(filter), 599 .filter = filter, 600 }; 601 long ret; 602 603 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 604 ASSERT_EQ(0, ret); 605 606 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 607 EXPECT_EQ(-1, ret); 608 EXPECT_EQ(EINVAL, errno); 609 } 610 611 TEST(log_all) 612 { 613 struct sock_filter filter[] = { 614 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG), 615 }; 616 struct sock_fprog prog = { 617 .len = (unsigned short)ARRAY_SIZE(filter), 618 .filter = filter, 619 }; 620 long ret; 621 pid_t parent = getppid(); 622 623 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 624 ASSERT_EQ(0, ret); 625 626 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 627 ASSERT_EQ(0, ret); 628 629 /* getppid() should succeed and be logged (no check for logging) */ 630 EXPECT_EQ(parent, syscall(__NR_getppid)); 631 } 632 633 TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS) 634 { 635 struct sock_filter filter[] = { 636 BPF_STMT(BPF_RET|BPF_K, 0x10000000U), 637 }; 638 struct sock_fprog prog = { 639 .len = (unsigned short)ARRAY_SIZE(filter), 640 .filter = filter, 641 }; 642 long ret; 643 644 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 645 ASSERT_EQ(0, ret); 646 647 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 648 ASSERT_EQ(0, ret); 649 EXPECT_EQ(0, syscall(__NR_getpid)) { 650 TH_LOG("getpid() shouldn't ever return"); 651 } 652 } 653 654 /* return code >= 0x80000000 is unused. */ 655 TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS) 656 { 657 struct sock_filter filter[] = { 658 BPF_STMT(BPF_RET|BPF_K, 0x90000000U), 659 }; 660 struct sock_fprog prog = { 661 .len = (unsigned short)ARRAY_SIZE(filter), 662 .filter = filter, 663 }; 664 long ret; 665 666 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 667 ASSERT_EQ(0, ret); 668 669 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 670 ASSERT_EQ(0, ret); 671 EXPECT_EQ(0, syscall(__NR_getpid)) { 672 TH_LOG("getpid() shouldn't ever return"); 673 } 674 } 675 676 TEST_SIGNAL(KILL_all, SIGSYS) 677 { 678 struct sock_filter filter[] = { 679 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 680 }; 681 struct sock_fprog prog = { 682 .len = (unsigned short)ARRAY_SIZE(filter), 683 .filter = filter, 684 }; 685 long ret; 686 687 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 688 ASSERT_EQ(0, ret); 689 690 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 691 ASSERT_EQ(0, ret); 692 } 693 694 TEST_SIGNAL(KILL_one, SIGSYS) 695 { 696 struct sock_filter filter[] = { 697 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 698 offsetof(struct seccomp_data, nr)), 699 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 700 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 701 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 702 }; 703 struct sock_fprog prog = { 704 .len = (unsigned short)ARRAY_SIZE(filter), 705 .filter = filter, 706 }; 707 long ret; 708 pid_t parent = getppid(); 709 710 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 711 ASSERT_EQ(0, ret); 712 713 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 714 ASSERT_EQ(0, ret); 715 716 EXPECT_EQ(parent, syscall(__NR_getppid)); 717 /* getpid() should never return. */ 718 EXPECT_EQ(0, syscall(__NR_getpid)); 719 } 720 721 TEST_SIGNAL(KILL_one_arg_one, SIGSYS) 722 { 723 void *fatal_address; 724 struct sock_filter filter[] = { 725 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 726 offsetof(struct seccomp_data, nr)), 727 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0), 728 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 729 /* Only both with lower 32-bit for now. */ 730 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)), 731 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 732 (unsigned long)&fatal_address, 0, 1), 733 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 734 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 735 }; 736 struct sock_fprog prog = { 737 .len = (unsigned short)ARRAY_SIZE(filter), 738 .filter = filter, 739 }; 740 long ret; 741 pid_t parent = getppid(); 742 struct tms timebuf; 743 clock_t clock = times(&timebuf); 744 745 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 746 ASSERT_EQ(0, ret); 747 748 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 749 ASSERT_EQ(0, ret); 750 751 EXPECT_EQ(parent, syscall(__NR_getppid)); 752 EXPECT_LE(clock, syscall(__NR_times, &timebuf)); 753 /* times() should never return. */ 754 EXPECT_EQ(0, syscall(__NR_times, &fatal_address)); 755 } 756 757 TEST_SIGNAL(KILL_one_arg_six, SIGSYS) 758 { 759 #ifndef __NR_mmap2 760 int sysno = __NR_mmap; 761 #else 762 int sysno = __NR_mmap2; 763 #endif 764 struct sock_filter filter[] = { 765 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 766 offsetof(struct seccomp_data, nr)), 767 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0), 768 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 769 /* Only both with lower 32-bit for now. */ 770 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)), 771 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1), 772 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 773 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 774 }; 775 struct sock_fprog prog = { 776 .len = (unsigned short)ARRAY_SIZE(filter), 777 .filter = filter, 778 }; 779 long ret; 780 pid_t parent = getppid(); 781 int fd; 782 void *map1, *map2; 783 int page_size = sysconf(_SC_PAGESIZE); 784 785 ASSERT_LT(0, page_size); 786 787 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 788 ASSERT_EQ(0, ret); 789 790 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 791 ASSERT_EQ(0, ret); 792 793 fd = open("/dev/zero", O_RDONLY); 794 ASSERT_NE(-1, fd); 795 796 EXPECT_EQ(parent, syscall(__NR_getppid)); 797 map1 = (void *)syscall(sysno, 798 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size); 799 EXPECT_NE(MAP_FAILED, map1); 800 /* mmap2() should never return. */ 801 map2 = (void *)syscall(sysno, 802 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE); 803 EXPECT_EQ(MAP_FAILED, map2); 804 805 /* The test failed, so clean up the resources. */ 806 munmap(map1, page_size); 807 munmap(map2, page_size); 808 close(fd); 809 } 810 811 /* This is a thread task to die via seccomp filter violation. */ 812 void *kill_thread(void *data) 813 { 814 bool die = (bool)data; 815 816 if (die) { 817 syscall(__NR_getpid); 818 return (void *)SIBLING_EXIT_FAILURE; 819 } 820 821 return (void *)SIBLING_EXIT_UNKILLED; 822 } 823 824 enum kill_t { 825 KILL_THREAD, 826 KILL_PROCESS, 827 RET_UNKNOWN 828 }; 829 830 /* Prepare a thread that will kill itself or both of us. */ 831 void kill_thread_or_group(struct __test_metadata *_metadata, 832 enum kill_t kill_how) 833 { 834 pthread_t thread; 835 void *status; 836 /* Kill only when calling __NR_getpid. */ 837 struct sock_filter filter_thread[] = { 838 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 839 offsetof(struct seccomp_data, nr)), 840 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 841 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD), 842 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 843 }; 844 struct sock_fprog prog_thread = { 845 .len = (unsigned short)ARRAY_SIZE(filter_thread), 846 .filter = filter_thread, 847 }; 848 int kill = kill_how == KILL_PROCESS ? SECCOMP_RET_KILL_PROCESS : 0xAAAAAAAA; 849 struct sock_filter filter_process[] = { 850 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 851 offsetof(struct seccomp_data, nr)), 852 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 853 BPF_STMT(BPF_RET|BPF_K, kill), 854 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 855 }; 856 struct sock_fprog prog_process = { 857 .len = (unsigned short)ARRAY_SIZE(filter_process), 858 .filter = filter_process, 859 }; 860 861 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 862 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 863 } 864 865 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, 866 kill_how == KILL_THREAD ? &prog_thread 867 : &prog_process)); 868 869 /* 870 * Add the KILL_THREAD rule again to make sure that the KILL_PROCESS 871 * flag cannot be downgraded by a new filter. 872 */ 873 if (kill_how == KILL_PROCESS) 874 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog_thread)); 875 876 /* Start a thread that will exit immediately. */ 877 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)false)); 878 ASSERT_EQ(0, pthread_join(thread, &status)); 879 ASSERT_EQ(SIBLING_EXIT_UNKILLED, (unsigned long)status); 880 881 /* Start a thread that will die immediately. */ 882 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)true)); 883 ASSERT_EQ(0, pthread_join(thread, &status)); 884 ASSERT_NE(SIBLING_EXIT_FAILURE, (unsigned long)status); 885 886 /* 887 * If we get here, only the spawned thread died. Let the parent know 888 * the whole process didn't die (i.e. this thread, the spawner, 889 * stayed running). 890 */ 891 exit(42); 892 } 893 894 TEST(KILL_thread) 895 { 896 int status; 897 pid_t child_pid; 898 899 child_pid = fork(); 900 ASSERT_LE(0, child_pid); 901 if (child_pid == 0) { 902 kill_thread_or_group(_metadata, KILL_THREAD); 903 _exit(38); 904 } 905 906 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 907 908 /* If only the thread was killed, we'll see exit 42. */ 909 ASSERT_TRUE(WIFEXITED(status)); 910 ASSERT_EQ(42, WEXITSTATUS(status)); 911 } 912 913 TEST(KILL_process) 914 { 915 int status; 916 pid_t child_pid; 917 918 child_pid = fork(); 919 ASSERT_LE(0, child_pid); 920 if (child_pid == 0) { 921 kill_thread_or_group(_metadata, KILL_PROCESS); 922 _exit(38); 923 } 924 925 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 926 927 /* If the entire process was killed, we'll see SIGSYS. */ 928 ASSERT_TRUE(WIFSIGNALED(status)); 929 ASSERT_EQ(SIGSYS, WTERMSIG(status)); 930 } 931 932 TEST(KILL_unknown) 933 { 934 int status; 935 pid_t child_pid; 936 937 child_pid = fork(); 938 ASSERT_LE(0, child_pid); 939 if (child_pid == 0) { 940 kill_thread_or_group(_metadata, RET_UNKNOWN); 941 _exit(38); 942 } 943 944 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 945 946 /* If the entire process was killed, we'll see SIGSYS. */ 947 EXPECT_TRUE(WIFSIGNALED(status)) { 948 TH_LOG("Unknown SECCOMP_RET is only killing the thread?"); 949 } 950 ASSERT_EQ(SIGSYS, WTERMSIG(status)); 951 } 952 953 /* TODO(wad) add 64-bit versus 32-bit arg tests. */ 954 TEST(arg_out_of_range) 955 { 956 struct sock_filter filter[] = { 957 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)), 958 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 959 }; 960 struct sock_fprog prog = { 961 .len = (unsigned short)ARRAY_SIZE(filter), 962 .filter = filter, 963 }; 964 long ret; 965 966 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 967 ASSERT_EQ(0, ret); 968 969 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 970 EXPECT_EQ(-1, ret); 971 EXPECT_EQ(EINVAL, errno); 972 } 973 974 #define ERRNO_FILTER(name, errno) \ 975 struct sock_filter _read_filter_##name[] = { \ 976 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, \ 977 offsetof(struct seccomp_data, nr)), \ 978 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), \ 979 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | errno), \ 980 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), \ 981 }; \ 982 struct sock_fprog prog_##name = { \ 983 .len = (unsigned short)ARRAY_SIZE(_read_filter_##name), \ 984 .filter = _read_filter_##name, \ 985 } 986 987 /* Make sure basic errno values are correctly passed through a filter. */ 988 TEST(ERRNO_valid) 989 { 990 ERRNO_FILTER(valid, E2BIG); 991 long ret; 992 pid_t parent = getppid(); 993 994 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 995 ASSERT_EQ(0, ret); 996 997 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_valid); 998 ASSERT_EQ(0, ret); 999 1000 EXPECT_EQ(parent, syscall(__NR_getppid)); 1001 EXPECT_EQ(-1, read(-1, NULL, 0)); 1002 EXPECT_EQ(E2BIG, errno); 1003 } 1004 1005 /* Make sure an errno of zero is correctly handled by the arch code. */ 1006 TEST(ERRNO_zero) 1007 { 1008 ERRNO_FILTER(zero, 0); 1009 long ret; 1010 pid_t parent = getppid(); 1011 1012 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1013 ASSERT_EQ(0, ret); 1014 1015 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_zero); 1016 ASSERT_EQ(0, ret); 1017 1018 EXPECT_EQ(parent, syscall(__NR_getppid)); 1019 /* "errno" of 0 is ok. */ 1020 EXPECT_EQ(0, read(-1, NULL, 0)); 1021 } 1022 1023 /* 1024 * The SECCOMP_RET_DATA mask is 16 bits wide, but errno is smaller. 1025 * This tests that the errno value gets capped correctly, fixed by 1026 * 580c57f10768 ("seccomp: cap SECCOMP_RET_ERRNO data to MAX_ERRNO"). 1027 */ 1028 TEST(ERRNO_capped) 1029 { 1030 ERRNO_FILTER(capped, 4096); 1031 long ret; 1032 pid_t parent = getppid(); 1033 1034 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1035 ASSERT_EQ(0, ret); 1036 1037 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_capped); 1038 ASSERT_EQ(0, ret); 1039 1040 EXPECT_EQ(parent, syscall(__NR_getppid)); 1041 EXPECT_EQ(-1, read(-1, NULL, 0)); 1042 EXPECT_EQ(4095, errno); 1043 } 1044 1045 /* 1046 * Filters are processed in reverse order: last applied is executed first. 1047 * Since only the SECCOMP_RET_ACTION mask is tested for return values, the 1048 * SECCOMP_RET_DATA mask results will follow the most recently applied 1049 * matching filter return (and not the lowest or highest value). 1050 */ 1051 TEST(ERRNO_order) 1052 { 1053 ERRNO_FILTER(first, 11); 1054 ERRNO_FILTER(second, 13); 1055 ERRNO_FILTER(third, 12); 1056 long ret; 1057 pid_t parent = getppid(); 1058 1059 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1060 ASSERT_EQ(0, ret); 1061 1062 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_first); 1063 ASSERT_EQ(0, ret); 1064 1065 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_second); 1066 ASSERT_EQ(0, ret); 1067 1068 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_third); 1069 ASSERT_EQ(0, ret); 1070 1071 EXPECT_EQ(parent, syscall(__NR_getppid)); 1072 EXPECT_EQ(-1, read(-1, NULL, 0)); 1073 EXPECT_EQ(12, errno); 1074 } 1075 1076 FIXTURE(TRAP) { 1077 struct sock_fprog prog; 1078 }; 1079 1080 FIXTURE_SETUP(TRAP) 1081 { 1082 struct sock_filter filter[] = { 1083 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1084 offsetof(struct seccomp_data, nr)), 1085 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 1086 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP), 1087 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1088 }; 1089 1090 memset(&self->prog, 0, sizeof(self->prog)); 1091 self->prog.filter = malloc(sizeof(filter)); 1092 ASSERT_NE(NULL, self->prog.filter); 1093 memcpy(self->prog.filter, filter, sizeof(filter)); 1094 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 1095 } 1096 1097 FIXTURE_TEARDOWN(TRAP) 1098 { 1099 if (self->prog.filter) 1100 free(self->prog.filter); 1101 } 1102 1103 TEST_F_SIGNAL(TRAP, dfl, SIGSYS) 1104 { 1105 long ret; 1106 1107 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1108 ASSERT_EQ(0, ret); 1109 1110 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 1111 ASSERT_EQ(0, ret); 1112 syscall(__NR_getpid); 1113 } 1114 1115 /* Ensure that SIGSYS overrides SIG_IGN */ 1116 TEST_F_SIGNAL(TRAP, ign, SIGSYS) 1117 { 1118 long ret; 1119 1120 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1121 ASSERT_EQ(0, ret); 1122 1123 signal(SIGSYS, SIG_IGN); 1124 1125 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 1126 ASSERT_EQ(0, ret); 1127 syscall(__NR_getpid); 1128 } 1129 1130 static siginfo_t TRAP_info; 1131 static volatile int TRAP_nr; 1132 static void TRAP_action(int nr, siginfo_t *info, void *void_context) 1133 { 1134 memcpy(&TRAP_info, info, sizeof(TRAP_info)); 1135 TRAP_nr = nr; 1136 } 1137 1138 TEST_F(TRAP, handler) 1139 { 1140 int ret, test; 1141 struct sigaction act; 1142 sigset_t mask; 1143 1144 memset(&act, 0, sizeof(act)); 1145 sigemptyset(&mask); 1146 sigaddset(&mask, SIGSYS); 1147 1148 act.sa_sigaction = &TRAP_action; 1149 act.sa_flags = SA_SIGINFO; 1150 ret = sigaction(SIGSYS, &act, NULL); 1151 ASSERT_EQ(0, ret) { 1152 TH_LOG("sigaction failed"); 1153 } 1154 ret = sigprocmask(SIG_UNBLOCK, &mask, NULL); 1155 ASSERT_EQ(0, ret) { 1156 TH_LOG("sigprocmask failed"); 1157 } 1158 1159 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1160 ASSERT_EQ(0, ret); 1161 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 1162 ASSERT_EQ(0, ret); 1163 TRAP_nr = 0; 1164 memset(&TRAP_info, 0, sizeof(TRAP_info)); 1165 /* Expect the registers to be rolled back. (nr = error) may vary 1166 * based on arch. */ 1167 ret = syscall(__NR_getpid); 1168 /* Silence gcc warning about volatile. */ 1169 test = TRAP_nr; 1170 EXPECT_EQ(SIGSYS, test); 1171 struct local_sigsys { 1172 void *_call_addr; /* calling user insn */ 1173 int _syscall; /* triggering system call number */ 1174 unsigned int _arch; /* AUDIT_ARCH_* of syscall */ 1175 } *sigsys = (struct local_sigsys *) 1176 #ifdef si_syscall 1177 &(TRAP_info.si_call_addr); 1178 #else 1179 &TRAP_info.si_pid; 1180 #endif 1181 EXPECT_EQ(__NR_getpid, sigsys->_syscall); 1182 /* Make sure arch is non-zero. */ 1183 EXPECT_NE(0, sigsys->_arch); 1184 EXPECT_NE(0, (unsigned long)sigsys->_call_addr); 1185 } 1186 1187 FIXTURE(precedence) { 1188 struct sock_fprog allow; 1189 struct sock_fprog log; 1190 struct sock_fprog trace; 1191 struct sock_fprog error; 1192 struct sock_fprog trap; 1193 struct sock_fprog kill; 1194 }; 1195 1196 FIXTURE_SETUP(precedence) 1197 { 1198 struct sock_filter allow_insns[] = { 1199 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1200 }; 1201 struct sock_filter log_insns[] = { 1202 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1203 offsetof(struct seccomp_data, nr)), 1204 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1205 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1206 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG), 1207 }; 1208 struct sock_filter trace_insns[] = { 1209 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1210 offsetof(struct seccomp_data, nr)), 1211 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1212 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1213 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE), 1214 }; 1215 struct sock_filter error_insns[] = { 1216 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1217 offsetof(struct seccomp_data, nr)), 1218 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1219 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1220 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO), 1221 }; 1222 struct sock_filter trap_insns[] = { 1223 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1224 offsetof(struct seccomp_data, nr)), 1225 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1226 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1227 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP), 1228 }; 1229 struct sock_filter kill_insns[] = { 1230 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1231 offsetof(struct seccomp_data, nr)), 1232 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1233 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1234 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 1235 }; 1236 1237 memset(self, 0, sizeof(*self)); 1238 #define FILTER_ALLOC(_x) \ 1239 self->_x.filter = malloc(sizeof(_x##_insns)); \ 1240 ASSERT_NE(NULL, self->_x.filter); \ 1241 memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \ 1242 self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns) 1243 FILTER_ALLOC(allow); 1244 FILTER_ALLOC(log); 1245 FILTER_ALLOC(trace); 1246 FILTER_ALLOC(error); 1247 FILTER_ALLOC(trap); 1248 FILTER_ALLOC(kill); 1249 } 1250 1251 FIXTURE_TEARDOWN(precedence) 1252 { 1253 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter) 1254 FILTER_FREE(allow); 1255 FILTER_FREE(log); 1256 FILTER_FREE(trace); 1257 FILTER_FREE(error); 1258 FILTER_FREE(trap); 1259 FILTER_FREE(kill); 1260 } 1261 1262 TEST_F(precedence, allow_ok) 1263 { 1264 pid_t parent, res = 0; 1265 long ret; 1266 1267 parent = getppid(); 1268 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1269 ASSERT_EQ(0, ret); 1270 1271 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1272 ASSERT_EQ(0, ret); 1273 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1274 ASSERT_EQ(0, ret); 1275 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1276 ASSERT_EQ(0, ret); 1277 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1278 ASSERT_EQ(0, ret); 1279 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1280 ASSERT_EQ(0, ret); 1281 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1282 ASSERT_EQ(0, ret); 1283 /* Should work just fine. */ 1284 res = syscall(__NR_getppid); 1285 EXPECT_EQ(parent, res); 1286 } 1287 1288 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS) 1289 { 1290 pid_t parent, res = 0; 1291 long ret; 1292 1293 parent = getppid(); 1294 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1295 ASSERT_EQ(0, ret); 1296 1297 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1298 ASSERT_EQ(0, ret); 1299 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1300 ASSERT_EQ(0, ret); 1301 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1302 ASSERT_EQ(0, ret); 1303 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1304 ASSERT_EQ(0, ret); 1305 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1306 ASSERT_EQ(0, ret); 1307 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1308 ASSERT_EQ(0, ret); 1309 /* Should work just fine. */ 1310 res = syscall(__NR_getppid); 1311 EXPECT_EQ(parent, res); 1312 /* getpid() should never return. */ 1313 res = syscall(__NR_getpid); 1314 EXPECT_EQ(0, res); 1315 } 1316 1317 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS) 1318 { 1319 pid_t parent; 1320 long ret; 1321 1322 parent = getppid(); 1323 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1324 ASSERT_EQ(0, ret); 1325 1326 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1327 ASSERT_EQ(0, ret); 1328 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1329 ASSERT_EQ(0, ret); 1330 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1331 ASSERT_EQ(0, ret); 1332 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1333 ASSERT_EQ(0, ret); 1334 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1335 ASSERT_EQ(0, ret); 1336 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1337 ASSERT_EQ(0, ret); 1338 /* Should work just fine. */ 1339 EXPECT_EQ(parent, syscall(__NR_getppid)); 1340 /* getpid() should never return. */ 1341 EXPECT_EQ(0, syscall(__NR_getpid)); 1342 } 1343 1344 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS) 1345 { 1346 pid_t parent; 1347 long ret; 1348 1349 parent = getppid(); 1350 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1351 ASSERT_EQ(0, ret); 1352 1353 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1354 ASSERT_EQ(0, ret); 1355 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1356 ASSERT_EQ(0, ret); 1357 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1358 ASSERT_EQ(0, ret); 1359 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1360 ASSERT_EQ(0, ret); 1361 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1362 ASSERT_EQ(0, ret); 1363 /* Should work just fine. */ 1364 EXPECT_EQ(parent, syscall(__NR_getppid)); 1365 /* getpid() should never return. */ 1366 EXPECT_EQ(0, syscall(__NR_getpid)); 1367 } 1368 1369 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS) 1370 { 1371 pid_t parent; 1372 long ret; 1373 1374 parent = getppid(); 1375 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1376 ASSERT_EQ(0, ret); 1377 1378 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1379 ASSERT_EQ(0, ret); 1380 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1381 ASSERT_EQ(0, ret); 1382 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1383 ASSERT_EQ(0, ret); 1384 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1385 ASSERT_EQ(0, ret); 1386 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1387 ASSERT_EQ(0, ret); 1388 /* Should work just fine. */ 1389 EXPECT_EQ(parent, syscall(__NR_getppid)); 1390 /* getpid() should never return. */ 1391 EXPECT_EQ(0, syscall(__NR_getpid)); 1392 } 1393 1394 TEST_F(precedence, errno_is_third) 1395 { 1396 pid_t parent; 1397 long ret; 1398 1399 parent = getppid(); 1400 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1401 ASSERT_EQ(0, ret); 1402 1403 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1404 ASSERT_EQ(0, ret); 1405 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1406 ASSERT_EQ(0, ret); 1407 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1408 ASSERT_EQ(0, ret); 1409 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1410 ASSERT_EQ(0, ret); 1411 /* Should work just fine. */ 1412 EXPECT_EQ(parent, syscall(__NR_getppid)); 1413 EXPECT_EQ(0, syscall(__NR_getpid)); 1414 } 1415 1416 TEST_F(precedence, errno_is_third_in_any_order) 1417 { 1418 pid_t parent; 1419 long ret; 1420 1421 parent = getppid(); 1422 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1423 ASSERT_EQ(0, ret); 1424 1425 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1426 ASSERT_EQ(0, ret); 1427 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1428 ASSERT_EQ(0, ret); 1429 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1430 ASSERT_EQ(0, ret); 1431 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1432 ASSERT_EQ(0, ret); 1433 /* Should work just fine. */ 1434 EXPECT_EQ(parent, syscall(__NR_getppid)); 1435 EXPECT_EQ(0, syscall(__NR_getpid)); 1436 } 1437 1438 TEST_F(precedence, trace_is_fourth) 1439 { 1440 pid_t parent; 1441 long ret; 1442 1443 parent = getppid(); 1444 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1445 ASSERT_EQ(0, ret); 1446 1447 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1448 ASSERT_EQ(0, ret); 1449 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1450 ASSERT_EQ(0, ret); 1451 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1452 ASSERT_EQ(0, ret); 1453 /* Should work just fine. */ 1454 EXPECT_EQ(parent, syscall(__NR_getppid)); 1455 /* No ptracer */ 1456 EXPECT_EQ(-1, syscall(__NR_getpid)); 1457 } 1458 1459 TEST_F(precedence, trace_is_fourth_in_any_order) 1460 { 1461 pid_t parent; 1462 long ret; 1463 1464 parent = getppid(); 1465 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1466 ASSERT_EQ(0, ret); 1467 1468 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1469 ASSERT_EQ(0, ret); 1470 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1471 ASSERT_EQ(0, ret); 1472 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1473 ASSERT_EQ(0, ret); 1474 /* Should work just fine. */ 1475 EXPECT_EQ(parent, syscall(__NR_getppid)); 1476 /* No ptracer */ 1477 EXPECT_EQ(-1, syscall(__NR_getpid)); 1478 } 1479 1480 TEST_F(precedence, log_is_fifth) 1481 { 1482 pid_t mypid, parent; 1483 long ret; 1484 1485 mypid = getpid(); 1486 parent = getppid(); 1487 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1488 ASSERT_EQ(0, ret); 1489 1490 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1491 ASSERT_EQ(0, ret); 1492 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1493 ASSERT_EQ(0, ret); 1494 /* Should work just fine. */ 1495 EXPECT_EQ(parent, syscall(__NR_getppid)); 1496 /* Should also work just fine */ 1497 EXPECT_EQ(mypid, syscall(__NR_getpid)); 1498 } 1499 1500 TEST_F(precedence, log_is_fifth_in_any_order) 1501 { 1502 pid_t mypid, parent; 1503 long ret; 1504 1505 mypid = getpid(); 1506 parent = getppid(); 1507 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1508 ASSERT_EQ(0, ret); 1509 1510 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1511 ASSERT_EQ(0, ret); 1512 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1513 ASSERT_EQ(0, ret); 1514 /* Should work just fine. */ 1515 EXPECT_EQ(parent, syscall(__NR_getppid)); 1516 /* Should also work just fine */ 1517 EXPECT_EQ(mypid, syscall(__NR_getpid)); 1518 } 1519 1520 #ifndef PTRACE_O_TRACESECCOMP 1521 #define PTRACE_O_TRACESECCOMP 0x00000080 1522 #endif 1523 1524 /* Catch the Ubuntu 12.04 value error. */ 1525 #if PTRACE_EVENT_SECCOMP != 7 1526 #undef PTRACE_EVENT_SECCOMP 1527 #endif 1528 1529 #ifndef PTRACE_EVENT_SECCOMP 1530 #define PTRACE_EVENT_SECCOMP 7 1531 #endif 1532 1533 #define PTRACE_EVENT_MASK(status) ((status) >> 16) 1534 bool tracer_running; 1535 void tracer_stop(int sig) 1536 { 1537 tracer_running = false; 1538 } 1539 1540 typedef void tracer_func_t(struct __test_metadata *_metadata, 1541 pid_t tracee, int status, void *args); 1542 1543 void start_tracer(struct __test_metadata *_metadata, int fd, pid_t tracee, 1544 tracer_func_t tracer_func, void *args, bool ptrace_syscall) 1545 { 1546 int ret = -1; 1547 struct sigaction action = { 1548 .sa_handler = tracer_stop, 1549 }; 1550 1551 /* Allow external shutdown. */ 1552 tracer_running = true; 1553 ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL)); 1554 1555 errno = 0; 1556 while (ret == -1 && errno != EINVAL) 1557 ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0); 1558 ASSERT_EQ(0, ret) { 1559 kill(tracee, SIGKILL); 1560 } 1561 /* Wait for attach stop */ 1562 wait(NULL); 1563 1564 ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, ptrace_syscall ? 1565 PTRACE_O_TRACESYSGOOD : 1566 PTRACE_O_TRACESECCOMP); 1567 ASSERT_EQ(0, ret) { 1568 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP"); 1569 kill(tracee, SIGKILL); 1570 } 1571 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT, 1572 tracee, NULL, 0); 1573 ASSERT_EQ(0, ret); 1574 1575 /* Unblock the tracee */ 1576 ASSERT_EQ(1, write(fd, "A", 1)); 1577 ASSERT_EQ(0, close(fd)); 1578 1579 /* Run until we're shut down. Must assert to stop execution. */ 1580 while (tracer_running) { 1581 int status; 1582 1583 if (wait(&status) != tracee) 1584 continue; 1585 1586 if (WIFSIGNALED(status)) { 1587 /* Child caught a fatal signal. */ 1588 return; 1589 } 1590 if (WIFEXITED(status)) { 1591 /* Child exited with code. */ 1592 return; 1593 } 1594 1595 /* Check if we got an expected event. */ 1596 ASSERT_EQ(WIFCONTINUED(status), false); 1597 ASSERT_EQ(WIFSTOPPED(status), true); 1598 ASSERT_EQ(WSTOPSIG(status) & SIGTRAP, SIGTRAP) { 1599 TH_LOG("Unexpected WSTOPSIG: %d", WSTOPSIG(status)); 1600 } 1601 1602 tracer_func(_metadata, tracee, status, args); 1603 1604 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT, 1605 tracee, NULL, 0); 1606 ASSERT_EQ(0, ret); 1607 } 1608 /* Directly report the status of our test harness results. */ 1609 syscall(__NR_exit, _metadata->exit_code); 1610 } 1611 1612 /* Common tracer setup/teardown functions. */ 1613 void cont_handler(int num) 1614 { } 1615 pid_t setup_trace_fixture(struct __test_metadata *_metadata, 1616 tracer_func_t func, void *args, bool ptrace_syscall) 1617 { 1618 char sync; 1619 int pipefd[2]; 1620 pid_t tracer_pid; 1621 pid_t tracee = getpid(); 1622 1623 /* Setup a pipe for clean synchronization. */ 1624 ASSERT_EQ(0, pipe(pipefd)); 1625 1626 /* Fork a child which we'll promote to tracer */ 1627 tracer_pid = fork(); 1628 ASSERT_LE(0, tracer_pid); 1629 signal(SIGALRM, cont_handler); 1630 if (tracer_pid == 0) { 1631 close(pipefd[0]); 1632 start_tracer(_metadata, pipefd[1], tracee, func, args, 1633 ptrace_syscall); 1634 syscall(__NR_exit, 0); 1635 } 1636 close(pipefd[1]); 1637 prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0); 1638 read(pipefd[0], &sync, 1); 1639 close(pipefd[0]); 1640 1641 return tracer_pid; 1642 } 1643 1644 void teardown_trace_fixture(struct __test_metadata *_metadata, 1645 pid_t tracer) 1646 { 1647 if (tracer) { 1648 int status; 1649 ASSERT_EQ(0, kill(tracer, SIGUSR1)); 1650 ASSERT_EQ(tracer, waitpid(tracer, &status, 0)); 1651 } 1652 } 1653 1654 /* "poke" tracer arguments and function. */ 1655 struct tracer_args_poke_t { 1656 unsigned long poke_addr; 1657 }; 1658 1659 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status, 1660 void *args) 1661 { 1662 int ret; 1663 unsigned long msg; 1664 struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args; 1665 1666 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 1667 EXPECT_EQ(0, ret); 1668 /* If this fails, don't try to recover. */ 1669 ASSERT_EQ(0x1001, msg) { 1670 kill(tracee, SIGKILL); 1671 } 1672 /* 1673 * Poke in the message. 1674 * Registers are not touched to try to keep this relatively arch 1675 * agnostic. 1676 */ 1677 ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001); 1678 EXPECT_EQ(0, ret); 1679 } 1680 1681 FIXTURE(TRACE_poke) { 1682 struct sock_fprog prog; 1683 pid_t tracer; 1684 long poked; 1685 struct tracer_args_poke_t tracer_args; 1686 }; 1687 1688 FIXTURE_SETUP(TRACE_poke) 1689 { 1690 struct sock_filter filter[] = { 1691 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1692 offsetof(struct seccomp_data, nr)), 1693 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 1694 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001), 1695 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1696 }; 1697 1698 self->poked = 0; 1699 memset(&self->prog, 0, sizeof(self->prog)); 1700 self->prog.filter = malloc(sizeof(filter)); 1701 ASSERT_NE(NULL, self->prog.filter); 1702 memcpy(self->prog.filter, filter, sizeof(filter)); 1703 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 1704 1705 /* Set up tracer args. */ 1706 self->tracer_args.poke_addr = (unsigned long)&self->poked; 1707 1708 /* Launch tracer. */ 1709 self->tracer = setup_trace_fixture(_metadata, tracer_poke, 1710 &self->tracer_args, false); 1711 } 1712 1713 FIXTURE_TEARDOWN(TRACE_poke) 1714 { 1715 teardown_trace_fixture(_metadata, self->tracer); 1716 if (self->prog.filter) 1717 free(self->prog.filter); 1718 } 1719 1720 TEST_F(TRACE_poke, read_has_side_effects) 1721 { 1722 ssize_t 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 ret = read(-1, NULL, 0); 1732 EXPECT_EQ(-1, ret); 1733 EXPECT_EQ(0x1001, self->poked); 1734 } 1735 1736 TEST_F(TRACE_poke, getpid_runs_normally) 1737 { 1738 long ret; 1739 1740 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1741 ASSERT_EQ(0, ret); 1742 1743 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1744 ASSERT_EQ(0, ret); 1745 1746 EXPECT_EQ(0, self->poked); 1747 EXPECT_NE(0, syscall(__NR_getpid)); 1748 EXPECT_EQ(0, self->poked); 1749 } 1750 1751 #if defined(__x86_64__) 1752 # define ARCH_REGS struct user_regs_struct 1753 # define SYSCALL_NUM(_regs) (_regs).orig_rax 1754 # define SYSCALL_RET(_regs) (_regs).rax 1755 #elif defined(__i386__) 1756 # define ARCH_REGS struct user_regs_struct 1757 # define SYSCALL_NUM(_regs) (_regs).orig_eax 1758 # define SYSCALL_RET(_regs) (_regs).eax 1759 #elif defined(__alpha__) 1760 /* NT_PRSTATUS contains r0-r30, pc and unique, not struct pt_regs. */ 1761 struct alpha_regs { 1762 unsigned long regs[33]; 1763 }; 1764 1765 # define ARCH_REGS struct alpha_regs 1766 /* The kernel keeps the mutable syscall number in r1. */ 1767 # define SYSCALL_NUM(_regs) ((_regs).regs[1]) 1768 # define SYSCALL_RET(_regs) ((_regs).regs[0]) 1769 /* 1770 * Alpha returns positive errno in r0 with the r19/a3 error flag set. 1771 * At a syscall-entry stop, r19 is still the fourth syscall argument. 1772 * Callers must also set the syscall number to -1 to skip the syscall 1773 * before replacing r19 with the error flag. 1774 */ 1775 # define SYSCALL_RET_SET(_regs, _val) \ 1776 do { \ 1777 struct alpha_regs *__regs = &(_regs); \ 1778 long __ret = (_val); \ 1779 \ 1780 __regs->regs[0] = __ret < 0 ? -__ret : __ret; \ 1781 __regs->regs[19] = __ret < 0; \ 1782 } while (0) 1783 #elif defined(__arm__) 1784 # define ARCH_REGS struct pt_regs 1785 # define SYSCALL_NUM(_regs) (_regs).ARM_r7 1786 # ifndef PTRACE_SET_SYSCALL 1787 # define PTRACE_SET_SYSCALL 23 1788 # endif 1789 # define SYSCALL_NUM_SET(_regs, _nr) \ 1790 EXPECT_EQ(0, ptrace(PTRACE_SET_SYSCALL, tracee, NULL, _nr)) 1791 # define SYSCALL_RET(_regs) (_regs).ARM_r0 1792 #elif defined(__aarch64__) 1793 # define ARCH_REGS struct user_pt_regs 1794 # define SYSCALL_NUM(_regs) (_regs).regs[8] 1795 # ifndef NT_ARM_SYSTEM_CALL 1796 # define NT_ARM_SYSTEM_CALL 0x404 1797 # endif 1798 # define SYSCALL_NUM_SET(_regs, _nr) \ 1799 do { \ 1800 struct iovec __v; \ 1801 typeof(_nr) __nr = (_nr); \ 1802 __v.iov_base = &__nr; \ 1803 __v.iov_len = sizeof(__nr); \ 1804 EXPECT_EQ(0, ptrace(PTRACE_SETREGSET, tracee, \ 1805 NT_ARM_SYSTEM_CALL, &__v)); \ 1806 } while (0) 1807 # define SYSCALL_RET(_regs) (_regs).regs[0] 1808 #elif defined(__loongarch__) 1809 # define ARCH_REGS struct user_pt_regs 1810 # define SYSCALL_NUM(_regs) (_regs).regs[11] 1811 # define SYSCALL_RET(_regs) (_regs).regs[4] 1812 #elif defined(__riscv) && __riscv_xlen == 64 1813 # define ARCH_REGS struct user_regs_struct 1814 # define SYSCALL_NUM(_regs) (_regs).a7 1815 # define SYSCALL_RET(_regs) (_regs).a0 1816 #elif defined(__csky__) 1817 # define ARCH_REGS struct pt_regs 1818 # if defined(__CSKYABIV2__) 1819 # define SYSCALL_NUM(_regs) (_regs).regs[3] 1820 # else 1821 # define SYSCALL_NUM(_regs) (_regs).regs[9] 1822 # endif 1823 # define SYSCALL_RET(_regs) (_regs).a0 1824 #elif defined(__hppa__) 1825 # define ARCH_REGS struct user_regs_struct 1826 # define SYSCALL_NUM(_regs) (_regs).gr[20] 1827 # define SYSCALL_RET(_regs) (_regs).gr[28] 1828 #elif defined(__powerpc__) 1829 # define ARCH_REGS struct pt_regs 1830 # define SYSCALL_NUM(_regs) (_regs).gpr[0] 1831 # define SYSCALL_RET(_regs) (_regs).gpr[3] 1832 # define SYSCALL_RET_SET(_regs, _val) \ 1833 do { \ 1834 typeof(_val) _result = (_val); \ 1835 if ((_regs.trap & 0xfff0) == 0x3000) { \ 1836 /* \ 1837 * scv 0 system call uses -ve result \ 1838 * for error, so no need to adjust. \ 1839 */ \ 1840 SYSCALL_RET(_regs) = _result; \ 1841 } else { \ 1842 /* \ 1843 * A syscall error is signaled by the \ 1844 * CR0 SO bit and the code is stored as \ 1845 * a positive value. \ 1846 */ \ 1847 if (_result < 0) { \ 1848 SYSCALL_RET(_regs) = -_result; \ 1849 (_regs).ccr |= 0x10000000; \ 1850 } else { \ 1851 SYSCALL_RET(_regs) = _result; \ 1852 (_regs).ccr &= ~0x10000000; \ 1853 } \ 1854 } \ 1855 } while (0) 1856 # define SYSCALL_RET_SET_ON_PTRACE_EXIT 1857 #elif defined(__s390__) 1858 # define ARCH_REGS s390_regs 1859 # define SYSCALL_NUM(_regs) (_regs).gprs[2] 1860 # define SYSCALL_RET_SET(_regs, _val) \ 1861 TH_LOG("Can't modify syscall return on this architecture") 1862 #elif defined(__mips__) 1863 # include <asm/unistd_nr_n32.h> 1864 # include <asm/unistd_nr_n64.h> 1865 # include <asm/unistd_nr_o32.h> 1866 # define ARCH_REGS struct pt_regs 1867 # define SYSCALL_NUM(_regs) \ 1868 ({ \ 1869 typeof((_regs).regs[2]) _nr; \ 1870 if ((_regs).regs[2] == __NR_O32_Linux) \ 1871 _nr = (_regs).regs[4]; \ 1872 else \ 1873 _nr = (_regs).regs[2]; \ 1874 _nr; \ 1875 }) 1876 # define SYSCALL_NUM_SET(_regs, _nr) \ 1877 do { \ 1878 if ((_regs).regs[2] == __NR_O32_Linux) \ 1879 (_regs).regs[4] = _nr; \ 1880 else \ 1881 (_regs).regs[2] = _nr; \ 1882 } while (0) 1883 # define SYSCALL_RET_SET(_regs, _val) \ 1884 TH_LOG("Can't modify syscall return on this architecture") 1885 #elif defined(__xtensa__) 1886 # define ARCH_REGS struct user_pt_regs 1887 # define SYSCALL_NUM(_regs) (_regs).syscall 1888 /* 1889 * On xtensa syscall return value is in the register 1890 * a2 of the current window which is not fixed. 1891 */ 1892 #define SYSCALL_RET(_regs) (_regs).a[(_regs).windowbase * 4 + 2] 1893 #elif defined(__sh__) 1894 # define ARCH_REGS struct pt_regs 1895 # define SYSCALL_NUM(_regs) (_regs).regs[3] 1896 # define SYSCALL_RET(_regs) (_regs).regs[0] 1897 #elif defined(__mc68000__) 1898 # define ARCH_REGS struct user_regs_struct 1899 # define SYSCALL_NUM(_regs) (_regs).orig_d0 1900 # define SYSCALL_RET(_regs) (_regs).d0 1901 #else 1902 # error "Do not know how to find your architecture's registers and syscalls" 1903 #endif 1904 1905 /* 1906 * Most architectures can change the syscall by just updating the 1907 * associated register. This is the default if not defined above. 1908 */ 1909 #ifndef SYSCALL_NUM_SET 1910 # define SYSCALL_NUM_SET(_regs, _nr) \ 1911 do { \ 1912 SYSCALL_NUM(_regs) = (_nr); \ 1913 } while (0) 1914 #endif 1915 /* 1916 * Most architectures can change the syscall return value by just 1917 * writing to the SYSCALL_RET register. This is the default if not 1918 * defined above. If an architecture cannot set the return value 1919 * (for example when the syscall and return value register is 1920 * shared), report it with TH_LOG() in an arch-specific definition 1921 * of SYSCALL_RET_SET() above, and leave SYSCALL_RET undefined. 1922 */ 1923 #if !defined(SYSCALL_RET) && !defined(SYSCALL_RET_SET) 1924 # error "One of SYSCALL_RET or SYSCALL_RET_SET is needed for this arch" 1925 #endif 1926 #ifndef SYSCALL_RET_SET 1927 # define SYSCALL_RET_SET(_regs, _val) \ 1928 do { \ 1929 SYSCALL_RET(_regs) = (_val); \ 1930 } while (0) 1931 #endif 1932 1933 /* When the syscall return can't be changed, stub out the tests for it. */ 1934 #ifndef SYSCALL_RET 1935 # define EXPECT_SYSCALL_RETURN(val, action) EXPECT_EQ(-1, action) 1936 #else 1937 # define EXPECT_SYSCALL_RETURN(val, action) \ 1938 do { \ 1939 errno = 0; \ 1940 if (val < 0) { \ 1941 EXPECT_EQ(-1, action); \ 1942 EXPECT_EQ(-(val), errno); \ 1943 } else { \ 1944 EXPECT_EQ(val, action); \ 1945 } \ 1946 } while (0) 1947 #endif 1948 1949 /* 1950 * Some architectures (e.g. powerpc) can only set syscall 1951 * return values on syscall exit during ptrace. 1952 */ 1953 const bool ptrace_entry_set_syscall_nr = true; 1954 const bool ptrace_entry_set_syscall_ret = 1955 #ifndef SYSCALL_RET_SET_ON_PTRACE_EXIT 1956 true; 1957 #else 1958 false; 1959 #endif 1960 1961 /* 1962 * Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for 1963 * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux). 1964 */ 1965 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__) || defined(__mc68000__) 1966 # define ARCH_GETREGS(_regs) ptrace(PTRACE_GETREGS, tracee, 0, &(_regs)) 1967 # define ARCH_SETREGS(_regs) ptrace(PTRACE_SETREGS, tracee, 0, &(_regs)) 1968 #else 1969 # define ARCH_GETREGS(_regs) ({ \ 1970 struct iovec __v; \ 1971 __v.iov_base = &(_regs); \ 1972 __v.iov_len = sizeof(_regs); \ 1973 ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &__v); \ 1974 }) 1975 # define ARCH_SETREGS(_regs) ({ \ 1976 struct iovec __v; \ 1977 __v.iov_base = &(_regs); \ 1978 __v.iov_len = sizeof(_regs); \ 1979 ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &__v); \ 1980 }) 1981 #endif 1982 1983 /* Architecture-specific syscall fetching routine. */ 1984 int get_syscall(struct __test_metadata *_metadata, pid_t tracee) 1985 { 1986 ARCH_REGS regs; 1987 1988 EXPECT_EQ(0, ARCH_GETREGS(regs)) { 1989 return -1; 1990 } 1991 1992 return SYSCALL_NUM(regs); 1993 } 1994 1995 /* Architecture-specific syscall changing routine. */ 1996 void __change_syscall(struct __test_metadata *_metadata, 1997 pid_t tracee, long *syscall, long *ret) 1998 { 1999 ARCH_REGS orig, regs; 2000 2001 /* Do not get/set registers if we have nothing to do. */ 2002 if (!syscall && !ret) 2003 return; 2004 2005 EXPECT_EQ(0, ARCH_GETREGS(regs)) { 2006 return; 2007 } 2008 orig = regs; 2009 2010 if (syscall) 2011 SYSCALL_NUM_SET(regs, *syscall); 2012 2013 if (ret) 2014 SYSCALL_RET_SET(regs, *ret); 2015 2016 /* Flush any register changes made. */ 2017 if (memcmp(&orig, ®s, sizeof(orig)) != 0) 2018 EXPECT_EQ(0, ARCH_SETREGS(regs)); 2019 } 2020 2021 /* Change only syscall number. */ 2022 void change_syscall_nr(struct __test_metadata *_metadata, 2023 pid_t tracee, long syscall) 2024 { 2025 __change_syscall(_metadata, tracee, &syscall, NULL); 2026 } 2027 2028 /* Change syscall return value (and set syscall number to -1). */ 2029 void change_syscall_ret(struct __test_metadata *_metadata, 2030 pid_t tracee, long ret) 2031 { 2032 long syscall = -1; 2033 2034 __change_syscall(_metadata, tracee, &syscall, &ret); 2035 } 2036 2037 void tracer_seccomp(struct __test_metadata *_metadata, pid_t tracee, 2038 int status, void *args) 2039 { 2040 int ret; 2041 unsigned long msg; 2042 2043 EXPECT_EQ(PTRACE_EVENT_MASK(status), PTRACE_EVENT_SECCOMP) { 2044 TH_LOG("Unexpected ptrace event: %d", PTRACE_EVENT_MASK(status)); 2045 return; 2046 } 2047 2048 /* Make sure we got the right message. */ 2049 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 2050 EXPECT_EQ(0, ret); 2051 2052 /* Validate and take action on expected syscalls. */ 2053 switch (msg) { 2054 case 0x1002: 2055 /* change getpid to getppid. */ 2056 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee)); 2057 change_syscall_nr(_metadata, tracee, __NR_getppid); 2058 break; 2059 case 0x1003: 2060 /* skip gettid with valid return code. */ 2061 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee)); 2062 change_syscall_ret(_metadata, tracee, 45000); 2063 break; 2064 case 0x1004: 2065 /* skip openat with error. */ 2066 EXPECT_EQ(__NR_openat, get_syscall(_metadata, tracee)); 2067 change_syscall_ret(_metadata, tracee, -ESRCH); 2068 break; 2069 case 0x1005: 2070 /* do nothing (allow getppid) */ 2071 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee)); 2072 break; 2073 default: 2074 EXPECT_EQ(0, msg) { 2075 TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg); 2076 kill(tracee, SIGKILL); 2077 } 2078 } 2079 2080 } 2081 2082 FIXTURE(TRACE_syscall) { 2083 struct sock_fprog prog; 2084 pid_t tracer, mytid, mypid, parent; 2085 long syscall_nr; 2086 }; 2087 2088 void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee, 2089 int status, void *args) 2090 { 2091 int ret; 2092 unsigned long msg; 2093 static bool entry; 2094 long syscall_nr_val, syscall_ret_val; 2095 long *syscall_nr = NULL, *syscall_ret = NULL; 2096 FIXTURE_DATA(TRACE_syscall) *self = args; 2097 2098 EXPECT_EQ(WSTOPSIG(status) & 0x80, 0x80) { 2099 TH_LOG("Unexpected WSTOPSIG: %d", WSTOPSIG(status)); 2100 return; 2101 } 2102 2103 /* 2104 * The traditional way to tell PTRACE_SYSCALL entry/exit 2105 * is by counting. 2106 */ 2107 entry = !entry; 2108 2109 /* Make sure we got an appropriate message. */ 2110 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 2111 EXPECT_EQ(0, ret); 2112 EXPECT_EQ(entry ? PTRACE_EVENTMSG_SYSCALL_ENTRY 2113 : PTRACE_EVENTMSG_SYSCALL_EXIT, msg); 2114 2115 /* 2116 * Some architectures only support setting return values during 2117 * syscall exit under ptrace, and on exit the syscall number may 2118 * no longer be available. Therefore, save the initial sycall 2119 * number here, so it can be examined during both entry and exit 2120 * phases. 2121 */ 2122 if (entry) 2123 self->syscall_nr = get_syscall(_metadata, tracee); 2124 2125 /* 2126 * Depending on the architecture's syscall setting abilities, we 2127 * pick which things to set during this phase (entry or exit). 2128 */ 2129 if (entry == ptrace_entry_set_syscall_nr) 2130 syscall_nr = &syscall_nr_val; 2131 if (entry == ptrace_entry_set_syscall_ret) 2132 syscall_ret = &syscall_ret_val; 2133 2134 /* Now handle the actual rewriting cases. */ 2135 switch (self->syscall_nr) { 2136 case __NR_getpid: 2137 syscall_nr_val = __NR_getppid; 2138 /* Never change syscall return for this case. */ 2139 syscall_ret = NULL; 2140 break; 2141 case __NR_gettid: 2142 syscall_nr_val = -1; 2143 syscall_ret_val = 45000; 2144 break; 2145 case __NR_openat: 2146 syscall_nr_val = -1; 2147 syscall_ret_val = -ESRCH; 2148 break; 2149 default: 2150 /* Unhandled, do nothing. */ 2151 return; 2152 } 2153 2154 __change_syscall(_metadata, tracee, syscall_nr, syscall_ret); 2155 } 2156 2157 FIXTURE_VARIANT(TRACE_syscall) { 2158 /* 2159 * All of the SECCOMP_RET_TRACE behaviors can be tested with either 2160 * SECCOMP_RET_TRACE+PTRACE_CONT or plain ptrace()+PTRACE_SYSCALL. 2161 * This indicates if we should use SECCOMP_RET_TRACE (false), or 2162 * ptrace (true). 2163 */ 2164 bool use_ptrace; 2165 }; 2166 2167 FIXTURE_VARIANT_ADD(TRACE_syscall, ptrace) { 2168 .use_ptrace = true, 2169 }; 2170 2171 FIXTURE_VARIANT_ADD(TRACE_syscall, seccomp) { 2172 .use_ptrace = false, 2173 }; 2174 2175 FIXTURE_SETUP(TRACE_syscall) 2176 { 2177 struct sock_filter filter[] = { 2178 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2179 offsetof(struct seccomp_data, nr)), 2180 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 2181 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002), 2182 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1), 2183 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003), 2184 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_openat, 0, 1), 2185 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004), 2186 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 2187 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1005), 2188 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2189 }; 2190 struct sock_fprog prog = { 2191 .len = (unsigned short)ARRAY_SIZE(filter), 2192 .filter = filter, 2193 }; 2194 long ret; 2195 2196 /* Prepare some testable syscall results. */ 2197 self->mytid = syscall(__NR_gettid); 2198 ASSERT_GT(self->mytid, 0); 2199 ASSERT_NE(self->mytid, 1) { 2200 TH_LOG("Running this test as init is not supported. :)"); 2201 } 2202 2203 self->mypid = getpid(); 2204 ASSERT_GT(self->mypid, 0); 2205 ASSERT_EQ(self->mytid, self->mypid); 2206 2207 self->parent = getppid(); 2208 ASSERT_GT(self->parent, 0); 2209 ASSERT_NE(self->parent, self->mypid); 2210 2211 /* Launch tracer. */ 2212 self->tracer = setup_trace_fixture(_metadata, 2213 variant->use_ptrace ? tracer_ptrace 2214 : tracer_seccomp, 2215 self, variant->use_ptrace); 2216 2217 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 2218 ASSERT_EQ(0, ret); 2219 2220 /* Do not install seccomp rewrite filters, as we'll use ptrace instead. */ 2221 if (variant->use_ptrace) 2222 return; 2223 2224 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2225 ASSERT_EQ(0, ret); 2226 } 2227 2228 FIXTURE_TEARDOWN(TRACE_syscall) 2229 { 2230 teardown_trace_fixture(_metadata, self->tracer); 2231 } 2232 2233 TEST(negative_ENOSYS) 2234 { 2235 #if defined(__arm__) 2236 SKIP(return, "arm32 does not support calling syscall -1"); 2237 #endif 2238 /* 2239 * There should be no difference between an "internal" skip 2240 * and userspace asking for syscall "-1". 2241 */ 2242 errno = 0; 2243 EXPECT_EQ(-1, syscall(-1)); 2244 EXPECT_EQ(errno, ENOSYS); 2245 /* And no difference for "still not valid but not -1". */ 2246 errno = 0; 2247 EXPECT_EQ(-1, syscall(-101)); 2248 EXPECT_EQ(errno, ENOSYS); 2249 } 2250 2251 TEST_F(TRACE_syscall, negative_ENOSYS) 2252 { 2253 negative_ENOSYS(_metadata); 2254 } 2255 2256 TEST_F(TRACE_syscall, syscall_allowed) 2257 { 2258 /* getppid works as expected (no changes). */ 2259 EXPECT_EQ(self->parent, syscall(__NR_getppid)); 2260 EXPECT_NE(self->mypid, syscall(__NR_getppid)); 2261 } 2262 2263 TEST_F(TRACE_syscall, syscall_redirected) 2264 { 2265 /* getpid has been redirected to getppid as expected. */ 2266 EXPECT_EQ(self->parent, syscall(__NR_getpid)); 2267 EXPECT_NE(self->mypid, syscall(__NR_getpid)); 2268 } 2269 2270 TEST_F(TRACE_syscall, syscall_errno) 2271 { 2272 /* Tracer should skip the open syscall, resulting in ESRCH. */ 2273 EXPECT_SYSCALL_RETURN(-ESRCH, syscall(__NR_openat)); 2274 } 2275 2276 TEST_F(TRACE_syscall, syscall_faked) 2277 { 2278 /* Tracer skips the gettid syscall and store altered return value. */ 2279 EXPECT_SYSCALL_RETURN(45000, syscall(__NR_gettid)); 2280 } 2281 2282 TEST_F_SIGNAL(TRACE_syscall, kill_immediate, SIGSYS) 2283 { 2284 struct sock_filter filter[] = { 2285 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2286 offsetof(struct seccomp_data, nr)), 2287 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_mknodat, 0, 1), 2288 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD), 2289 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2290 }; 2291 struct sock_fprog prog = { 2292 .len = (unsigned short)ARRAY_SIZE(filter), 2293 .filter = filter, 2294 }; 2295 long ret; 2296 2297 /* Install "kill on mknodat" filter. */ 2298 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2299 ASSERT_EQ(0, ret); 2300 2301 /* This should immediately die with SIGSYS, regardless of tracer. */ 2302 EXPECT_EQ(-1, syscall(__NR_mknodat, -1, NULL, 0, 0)); 2303 } 2304 2305 TEST_F(TRACE_syscall, skip_after) 2306 { 2307 struct sock_filter filter[] = { 2308 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2309 offsetof(struct seccomp_data, nr)), 2310 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 2311 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM), 2312 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2313 }; 2314 struct sock_fprog prog = { 2315 .len = (unsigned short)ARRAY_SIZE(filter), 2316 .filter = filter, 2317 }; 2318 long ret; 2319 2320 /* Install additional "errno on getppid" filter. */ 2321 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2322 ASSERT_EQ(0, ret); 2323 2324 /* Tracer will redirect getpid to getppid, and we should see EPERM. */ 2325 errno = 0; 2326 EXPECT_EQ(-1, syscall(__NR_getpid)); 2327 EXPECT_EQ(EPERM, errno); 2328 } 2329 2330 TEST_F_SIGNAL(TRACE_syscall, kill_after, SIGSYS) 2331 { 2332 struct sock_filter filter[] = { 2333 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2334 offsetof(struct seccomp_data, nr)), 2335 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 2336 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2337 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2338 }; 2339 struct sock_fprog prog = { 2340 .len = (unsigned short)ARRAY_SIZE(filter), 2341 .filter = filter, 2342 }; 2343 long ret; 2344 2345 /* Install additional "death on getppid" filter. */ 2346 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2347 ASSERT_EQ(0, ret); 2348 2349 /* Tracer will redirect getpid to getppid, and we should die. */ 2350 EXPECT_NE(self->mypid, syscall(__NR_getpid)); 2351 } 2352 2353 TEST(seccomp_syscall) 2354 { 2355 struct sock_filter filter[] = { 2356 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2357 }; 2358 struct sock_fprog prog = { 2359 .len = (unsigned short)ARRAY_SIZE(filter), 2360 .filter = filter, 2361 }; 2362 long ret; 2363 2364 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 2365 ASSERT_EQ(0, ret) { 2366 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2367 } 2368 2369 /* Reject insane operation. */ 2370 ret = seccomp(-1, 0, &prog); 2371 ASSERT_NE(ENOSYS, errno) { 2372 TH_LOG("Kernel does not support seccomp syscall!"); 2373 } 2374 EXPECT_EQ(EINVAL, errno) { 2375 TH_LOG("Did not reject crazy op value!"); 2376 } 2377 2378 /* Reject strict with flags or pointer. */ 2379 ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL); 2380 EXPECT_EQ(EINVAL, errno) { 2381 TH_LOG("Did not reject mode strict with flags!"); 2382 } 2383 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog); 2384 EXPECT_EQ(EINVAL, errno) { 2385 TH_LOG("Did not reject mode strict with uargs!"); 2386 } 2387 2388 /* Reject insane args for filter. */ 2389 ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog); 2390 EXPECT_EQ(EINVAL, errno) { 2391 TH_LOG("Did not reject crazy filter flags!"); 2392 } 2393 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL); 2394 EXPECT_EQ(EFAULT, errno) { 2395 TH_LOG("Did not reject NULL filter!"); 2396 } 2397 2398 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2399 EXPECT_EQ(0, errno) { 2400 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s", 2401 strerror(errno)); 2402 } 2403 } 2404 2405 TEST(seccomp_syscall_mode_lock) 2406 { 2407 struct sock_filter filter[] = { 2408 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2409 }; 2410 struct sock_fprog prog = { 2411 .len = (unsigned short)ARRAY_SIZE(filter), 2412 .filter = filter, 2413 }; 2414 long ret; 2415 2416 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 2417 ASSERT_EQ(0, ret) { 2418 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2419 } 2420 2421 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2422 ASSERT_NE(ENOSYS, errno) { 2423 TH_LOG("Kernel does not support seccomp syscall!"); 2424 } 2425 EXPECT_EQ(0, ret) { 2426 TH_LOG("Could not install filter!"); 2427 } 2428 2429 /* Make sure neither entry point will switch to strict. */ 2430 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0); 2431 EXPECT_EQ(EINVAL, errno) { 2432 TH_LOG("Switched to mode strict!"); 2433 } 2434 2435 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL); 2436 EXPECT_EQ(EINVAL, errno) { 2437 TH_LOG("Switched to mode strict!"); 2438 } 2439 } 2440 2441 /* 2442 * Test detection of known and unknown filter flags. Userspace needs to be able 2443 * to check if a filter flag is supported by the current kernel and a good way 2444 * of doing that is by attempting to enter filter mode, with the flag bit in 2445 * question set, and a NULL pointer for the _args_ parameter. EFAULT indicates 2446 * that the flag is valid and EINVAL indicates that the flag is invalid. 2447 */ 2448 TEST(detect_seccomp_filter_flags) 2449 { 2450 unsigned int flags[] = { SECCOMP_FILTER_FLAG_TSYNC, 2451 SECCOMP_FILTER_FLAG_LOG, 2452 SECCOMP_FILTER_FLAG_SPEC_ALLOW, 2453 SECCOMP_FILTER_FLAG_NEW_LISTENER, 2454 SECCOMP_FILTER_FLAG_TSYNC_ESRCH }; 2455 unsigned int exclusive[] = { 2456 SECCOMP_FILTER_FLAG_TSYNC, 2457 SECCOMP_FILTER_FLAG_NEW_LISTENER }; 2458 unsigned int flag, all_flags, exclusive_mask; 2459 int i; 2460 long ret; 2461 2462 /* Test detection of individual known-good filter flags */ 2463 for (i = 0, all_flags = 0; i < ARRAY_SIZE(flags); i++) { 2464 int bits = 0; 2465 2466 flag = flags[i]; 2467 /* Make sure the flag is a single bit! */ 2468 while (flag) { 2469 if (flag & 0x1) 2470 bits ++; 2471 flag >>= 1; 2472 } 2473 ASSERT_EQ(1, bits); 2474 flag = flags[i]; 2475 2476 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2477 ASSERT_NE(ENOSYS, errno) { 2478 TH_LOG("Kernel does not support seccomp syscall!"); 2479 } 2480 EXPECT_EQ(-1, ret); 2481 EXPECT_EQ(EFAULT, errno) { 2482 TH_LOG("Failed to detect that a known-good filter flag (0x%X) is supported!", 2483 flag); 2484 } 2485 2486 all_flags |= flag; 2487 } 2488 2489 /* 2490 * Test detection of all known-good filter flags combined. But 2491 * for the exclusive flags we need to mask them out and try them 2492 * individually for the "all flags" testing. 2493 */ 2494 exclusive_mask = 0; 2495 for (i = 0; i < ARRAY_SIZE(exclusive); i++) 2496 exclusive_mask |= exclusive[i]; 2497 for (i = 0; i < ARRAY_SIZE(exclusive); i++) { 2498 flag = all_flags & ~exclusive_mask; 2499 flag |= exclusive[i]; 2500 2501 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2502 EXPECT_EQ(-1, ret); 2503 EXPECT_EQ(EFAULT, errno) { 2504 TH_LOG("Failed to detect that all known-good filter flags (0x%X) are supported!", 2505 flag); 2506 } 2507 } 2508 2509 /* Test detection of an unknown filter flags, without exclusives. */ 2510 flag = -1; 2511 flag &= ~exclusive_mask; 2512 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2513 EXPECT_EQ(-1, ret); 2514 EXPECT_EQ(EINVAL, errno) { 2515 TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported!", 2516 flag); 2517 } 2518 2519 /* 2520 * Test detection of an unknown filter flag that may simply need to be 2521 * added to this test 2522 */ 2523 flag = flags[ARRAY_SIZE(flags) - 1] << 1; 2524 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2525 EXPECT_EQ(-1, ret); 2526 EXPECT_EQ(EINVAL, errno) { 2527 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?", 2528 flag); 2529 } 2530 } 2531 2532 TEST(TSYNC_first) 2533 { 2534 struct sock_filter filter[] = { 2535 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2536 }; 2537 struct sock_fprog prog = { 2538 .len = (unsigned short)ARRAY_SIZE(filter), 2539 .filter = filter, 2540 }; 2541 long ret; 2542 2543 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 2544 ASSERT_EQ(0, ret) { 2545 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2546 } 2547 2548 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2549 &prog); 2550 ASSERT_NE(ENOSYS, errno) { 2551 TH_LOG("Kernel does not support seccomp syscall!"); 2552 } 2553 EXPECT_EQ(0, ret) { 2554 TH_LOG("Could not install initial filter with TSYNC!"); 2555 } 2556 } 2557 2558 #define TSYNC_SIBLINGS 2 2559 struct tsync_sibling { 2560 pthread_t tid; 2561 pid_t system_tid; 2562 sem_t *started; 2563 pthread_cond_t *cond; 2564 pthread_mutex_t *mutex; 2565 int diverge; 2566 int num_waits; 2567 struct sock_fprog *prog; 2568 struct __test_metadata *metadata; 2569 }; 2570 2571 /* 2572 * To avoid joining joined threads (which is not allowed by Bionic), 2573 * make sure we both successfully join and clear the tid to skip a 2574 * later join attempt during fixture teardown. Any remaining threads 2575 * will be directly killed during teardown. 2576 */ 2577 #define PTHREAD_JOIN(tid, status) \ 2578 do { \ 2579 int _rc = pthread_join(tid, status); \ 2580 if (_rc) { \ 2581 TH_LOG("pthread_join of tid %u failed: %d\n", \ 2582 (unsigned int)tid, _rc); \ 2583 } else { \ 2584 tid = 0; \ 2585 } \ 2586 } while (0) 2587 2588 FIXTURE(TSYNC) { 2589 struct sock_fprog root_prog, apply_prog; 2590 struct tsync_sibling sibling[TSYNC_SIBLINGS]; 2591 sem_t started; 2592 pthread_cond_t cond; 2593 pthread_mutex_t mutex; 2594 int sibling_count; 2595 }; 2596 2597 FIXTURE_SETUP(TSYNC) 2598 { 2599 struct sock_filter root_filter[] = { 2600 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2601 }; 2602 struct sock_filter apply_filter[] = { 2603 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2604 offsetof(struct seccomp_data, nr)), 2605 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 2606 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2607 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2608 }; 2609 2610 memset(&self->root_prog, 0, sizeof(self->root_prog)); 2611 memset(&self->apply_prog, 0, sizeof(self->apply_prog)); 2612 memset(&self->sibling, 0, sizeof(self->sibling)); 2613 self->root_prog.filter = malloc(sizeof(root_filter)); 2614 ASSERT_NE(NULL, self->root_prog.filter); 2615 memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter)); 2616 self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter); 2617 2618 self->apply_prog.filter = malloc(sizeof(apply_filter)); 2619 ASSERT_NE(NULL, self->apply_prog.filter); 2620 memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter)); 2621 self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter); 2622 2623 self->sibling_count = 0; 2624 pthread_mutex_init(&self->mutex, NULL); 2625 pthread_cond_init(&self->cond, NULL); 2626 sem_init(&self->started, 0, 0); 2627 self->sibling[0].tid = 0; 2628 self->sibling[0].cond = &self->cond; 2629 self->sibling[0].started = &self->started; 2630 self->sibling[0].mutex = &self->mutex; 2631 self->sibling[0].diverge = 0; 2632 self->sibling[0].num_waits = 1; 2633 self->sibling[0].prog = &self->root_prog; 2634 self->sibling[0].metadata = _metadata; 2635 self->sibling[1].tid = 0; 2636 self->sibling[1].cond = &self->cond; 2637 self->sibling[1].started = &self->started; 2638 self->sibling[1].mutex = &self->mutex; 2639 self->sibling[1].diverge = 0; 2640 self->sibling[1].prog = &self->root_prog; 2641 self->sibling[1].num_waits = 1; 2642 self->sibling[1].metadata = _metadata; 2643 } 2644 2645 FIXTURE_TEARDOWN(TSYNC) 2646 { 2647 int sib = 0; 2648 2649 if (self->root_prog.filter) 2650 free(self->root_prog.filter); 2651 if (self->apply_prog.filter) 2652 free(self->apply_prog.filter); 2653 2654 for ( ; sib < self->sibling_count; ++sib) { 2655 struct tsync_sibling *s = &self->sibling[sib]; 2656 2657 if (!s->tid) 2658 continue; 2659 /* 2660 * If a thread is still running, it may be stuck, so hit 2661 * it over the head really hard. 2662 */ 2663 pthread_kill(s->tid, 9); 2664 } 2665 pthread_mutex_destroy(&self->mutex); 2666 pthread_cond_destroy(&self->cond); 2667 sem_destroy(&self->started); 2668 } 2669 2670 void *tsync_sibling(void *data) 2671 { 2672 long ret = 0; 2673 struct tsync_sibling *me = data; 2674 2675 me->system_tid = syscall(__NR_gettid); 2676 2677 pthread_mutex_lock(me->mutex); 2678 if (me->diverge) { 2679 /* Just re-apply the root prog to fork the tree */ 2680 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, 2681 me->prog, 0, 0); 2682 } 2683 sem_post(me->started); 2684 /* Return outside of started so parent notices failures. */ 2685 if (ret) { 2686 pthread_mutex_unlock(me->mutex); 2687 return (void *)SIBLING_EXIT_FAILURE; 2688 } 2689 do { 2690 pthread_cond_wait(me->cond, me->mutex); 2691 me->num_waits = me->num_waits - 1; 2692 } while (me->num_waits); 2693 pthread_mutex_unlock(me->mutex); 2694 2695 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0); 2696 if (!ret) 2697 return (void *)SIBLING_EXIT_NEWPRIVS; 2698 read(-1, NULL, 0); 2699 return (void *)SIBLING_EXIT_UNKILLED; 2700 } 2701 2702 void tsync_start_sibling(struct tsync_sibling *sibling) 2703 { 2704 pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling); 2705 } 2706 2707 TEST_F(TSYNC, siblings_fail_prctl) 2708 { 2709 long ret; 2710 void *status; 2711 struct sock_filter filter[] = { 2712 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2713 offsetof(struct seccomp_data, nr)), 2714 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1), 2715 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL), 2716 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2717 }; 2718 struct sock_fprog prog = { 2719 .len = (unsigned short)ARRAY_SIZE(filter), 2720 .filter = filter, 2721 }; 2722 2723 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2724 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2725 } 2726 2727 /* Check prctl failure detection by requesting sib 0 diverge. */ 2728 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2729 ASSERT_NE(ENOSYS, errno) { 2730 TH_LOG("Kernel does not support seccomp syscall!"); 2731 } 2732 ASSERT_EQ(0, ret) { 2733 TH_LOG("setting filter failed"); 2734 } 2735 2736 self->sibling[0].diverge = 1; 2737 tsync_start_sibling(&self->sibling[0]); 2738 tsync_start_sibling(&self->sibling[1]); 2739 2740 while (self->sibling_count < TSYNC_SIBLINGS) { 2741 sem_wait(&self->started); 2742 self->sibling_count++; 2743 } 2744 2745 /* Signal the threads to clean up*/ 2746 pthread_mutex_lock(&self->mutex); 2747 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2748 TH_LOG("cond broadcast non-zero"); 2749 } 2750 pthread_mutex_unlock(&self->mutex); 2751 2752 /* Ensure diverging sibling failed to call prctl. */ 2753 PTHREAD_JOIN(self->sibling[0].tid, &status); 2754 EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status); 2755 PTHREAD_JOIN(self->sibling[1].tid, &status); 2756 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2757 } 2758 2759 TEST_F(TSYNC, two_siblings_with_ancestor) 2760 { 2761 long ret; 2762 void *status; 2763 2764 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2765 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2766 } 2767 2768 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2769 ASSERT_NE(ENOSYS, errno) { 2770 TH_LOG("Kernel does not support seccomp syscall!"); 2771 } 2772 ASSERT_EQ(0, ret) { 2773 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2774 } 2775 tsync_start_sibling(&self->sibling[0]); 2776 tsync_start_sibling(&self->sibling[1]); 2777 2778 while (self->sibling_count < TSYNC_SIBLINGS) { 2779 sem_wait(&self->started); 2780 self->sibling_count++; 2781 } 2782 2783 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2784 &self->apply_prog); 2785 ASSERT_EQ(0, ret) { 2786 TH_LOG("Could install filter on all threads!"); 2787 } 2788 /* Tell the siblings to test the policy */ 2789 pthread_mutex_lock(&self->mutex); 2790 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2791 TH_LOG("cond broadcast non-zero"); 2792 } 2793 pthread_mutex_unlock(&self->mutex); 2794 /* Ensure they are both killed and don't exit cleanly. */ 2795 PTHREAD_JOIN(self->sibling[0].tid, &status); 2796 EXPECT_EQ(0x0, (long)status); 2797 PTHREAD_JOIN(self->sibling[1].tid, &status); 2798 EXPECT_EQ(0x0, (long)status); 2799 } 2800 2801 TEST_F(TSYNC, two_sibling_want_nnp) 2802 { 2803 void *status; 2804 2805 /* start siblings before any prctl() operations */ 2806 tsync_start_sibling(&self->sibling[0]); 2807 tsync_start_sibling(&self->sibling[1]); 2808 while (self->sibling_count < TSYNC_SIBLINGS) { 2809 sem_wait(&self->started); 2810 self->sibling_count++; 2811 } 2812 2813 /* Tell the siblings to test no 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 upset about lacking nnp. */ 2821 PTHREAD_JOIN(self->sibling[0].tid, &status); 2822 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status); 2823 PTHREAD_JOIN(self->sibling[1].tid, &status); 2824 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status); 2825 } 2826 2827 TEST_F(TSYNC, two_siblings_with_no_filter) 2828 { 2829 long ret; 2830 void *status; 2831 2832 /* start siblings before any prctl() operations */ 2833 tsync_start_sibling(&self->sibling[0]); 2834 tsync_start_sibling(&self->sibling[1]); 2835 while (self->sibling_count < TSYNC_SIBLINGS) { 2836 sem_wait(&self->started); 2837 self->sibling_count++; 2838 } 2839 2840 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2841 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2842 } 2843 2844 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2845 &self->apply_prog); 2846 ASSERT_NE(ENOSYS, errno) { 2847 TH_LOG("Kernel does not support seccomp syscall!"); 2848 } 2849 ASSERT_EQ(0, ret) { 2850 TH_LOG("Could install filter on all threads!"); 2851 } 2852 2853 /* Tell the siblings to test the policy */ 2854 pthread_mutex_lock(&self->mutex); 2855 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2856 TH_LOG("cond broadcast non-zero"); 2857 } 2858 pthread_mutex_unlock(&self->mutex); 2859 2860 /* Ensure they are both killed and don't exit cleanly. */ 2861 PTHREAD_JOIN(self->sibling[0].tid, &status); 2862 EXPECT_EQ(0x0, (long)status); 2863 PTHREAD_JOIN(self->sibling[1].tid, &status); 2864 EXPECT_EQ(0x0, (long)status); 2865 } 2866 2867 TEST_F(TSYNC, two_siblings_with_one_divergence) 2868 { 2869 long ret; 2870 void *status; 2871 2872 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2873 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2874 } 2875 2876 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2877 ASSERT_NE(ENOSYS, errno) { 2878 TH_LOG("Kernel does not support seccomp syscall!"); 2879 } 2880 ASSERT_EQ(0, ret) { 2881 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2882 } 2883 self->sibling[0].diverge = 1; 2884 tsync_start_sibling(&self->sibling[0]); 2885 tsync_start_sibling(&self->sibling[1]); 2886 2887 while (self->sibling_count < TSYNC_SIBLINGS) { 2888 sem_wait(&self->started); 2889 self->sibling_count++; 2890 } 2891 2892 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2893 &self->apply_prog); 2894 ASSERT_EQ(self->sibling[0].system_tid, ret) { 2895 TH_LOG("Did not fail on diverged sibling."); 2896 } 2897 2898 /* Wake the threads */ 2899 pthread_mutex_lock(&self->mutex); 2900 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2901 TH_LOG("cond broadcast non-zero"); 2902 } 2903 pthread_mutex_unlock(&self->mutex); 2904 2905 /* Ensure they are both unkilled. */ 2906 PTHREAD_JOIN(self->sibling[0].tid, &status); 2907 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2908 PTHREAD_JOIN(self->sibling[1].tid, &status); 2909 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2910 } 2911 2912 TEST_F(TSYNC, two_siblings_with_one_divergence_no_tid_in_err) 2913 { 2914 long ret, flags; 2915 void *status; 2916 2917 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2918 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2919 } 2920 2921 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2922 ASSERT_NE(ENOSYS, errno) { 2923 TH_LOG("Kernel does not support seccomp syscall!"); 2924 } 2925 ASSERT_EQ(0, ret) { 2926 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2927 } 2928 self->sibling[0].diverge = 1; 2929 tsync_start_sibling(&self->sibling[0]); 2930 tsync_start_sibling(&self->sibling[1]); 2931 2932 while (self->sibling_count < TSYNC_SIBLINGS) { 2933 sem_wait(&self->started); 2934 self->sibling_count++; 2935 } 2936 2937 flags = SECCOMP_FILTER_FLAG_TSYNC | \ 2938 SECCOMP_FILTER_FLAG_TSYNC_ESRCH; 2939 ret = seccomp(SECCOMP_SET_MODE_FILTER, flags, &self->apply_prog); 2940 ASSERT_EQ(ESRCH, errno) { 2941 TH_LOG("Did not return ESRCH for diverged sibling."); 2942 } 2943 ASSERT_EQ(-1, ret) { 2944 TH_LOG("Did not fail on diverged sibling."); 2945 } 2946 2947 /* Wake the threads */ 2948 pthread_mutex_lock(&self->mutex); 2949 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2950 TH_LOG("cond broadcast non-zero"); 2951 } 2952 pthread_mutex_unlock(&self->mutex); 2953 2954 /* Ensure they are both unkilled. */ 2955 PTHREAD_JOIN(self->sibling[0].tid, &status); 2956 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2957 PTHREAD_JOIN(self->sibling[1].tid, &status); 2958 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2959 } 2960 2961 TEST_F(TSYNC, two_siblings_not_under_filter) 2962 { 2963 long ret, sib; 2964 void *status; 2965 struct timespec delay = { .tv_nsec = 100000000 }; 2966 2967 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2968 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2969 } 2970 2971 /* 2972 * Sibling 0 will have its own seccomp policy 2973 * and Sibling 1 will not be under seccomp at 2974 * all. Sibling 1 will enter seccomp and 0 2975 * will cause failure. 2976 */ 2977 self->sibling[0].diverge = 1; 2978 tsync_start_sibling(&self->sibling[0]); 2979 tsync_start_sibling(&self->sibling[1]); 2980 2981 while (self->sibling_count < TSYNC_SIBLINGS) { 2982 sem_wait(&self->started); 2983 self->sibling_count++; 2984 } 2985 2986 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2987 ASSERT_NE(ENOSYS, errno) { 2988 TH_LOG("Kernel does not support seccomp syscall!"); 2989 } 2990 ASSERT_EQ(0, ret) { 2991 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2992 } 2993 2994 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2995 &self->apply_prog); 2996 ASSERT_EQ(ret, self->sibling[0].system_tid) { 2997 TH_LOG("Did not fail on diverged sibling."); 2998 } 2999 sib = 1; 3000 if (ret == self->sibling[0].system_tid) 3001 sib = 0; 3002 3003 pthread_mutex_lock(&self->mutex); 3004 3005 /* Increment the other siblings num_waits so we can clean up 3006 * the one we just saw. 3007 */ 3008 self->sibling[!sib].num_waits += 1; 3009 3010 /* Signal the thread to clean up*/ 3011 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 3012 TH_LOG("cond broadcast non-zero"); 3013 } 3014 pthread_mutex_unlock(&self->mutex); 3015 PTHREAD_JOIN(self->sibling[sib].tid, &status); 3016 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 3017 /* Poll for actual task death. pthread_join doesn't guarantee it. */ 3018 while (!kill(self->sibling[sib].system_tid, 0)) 3019 nanosleep(&delay, NULL); 3020 /* Switch to the remaining sibling */ 3021 sib = !sib; 3022 3023 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 3024 &self->apply_prog); 3025 ASSERT_EQ(0, ret) { 3026 TH_LOG("Expected the remaining sibling to sync"); 3027 }; 3028 3029 pthread_mutex_lock(&self->mutex); 3030 3031 /* If remaining sibling didn't have a chance to wake up during 3032 * the first broadcast, manually reduce the num_waits now. 3033 */ 3034 if (self->sibling[sib].num_waits > 1) 3035 self->sibling[sib].num_waits = 1; 3036 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 3037 TH_LOG("cond broadcast non-zero"); 3038 } 3039 pthread_mutex_unlock(&self->mutex); 3040 PTHREAD_JOIN(self->sibling[sib].tid, &status); 3041 EXPECT_EQ(0, (long)status); 3042 /* Poll for actual task death. pthread_join doesn't guarantee it. */ 3043 while (!kill(self->sibling[sib].system_tid, 0)) 3044 nanosleep(&delay, NULL); 3045 3046 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 3047 &self->apply_prog); 3048 ASSERT_EQ(0, ret); /* just us chickens */ 3049 } 3050 3051 /* Make sure restarted syscalls are seen directly as "restart_syscall". */ 3052 TEST(syscall_restart) 3053 { 3054 long ret; 3055 unsigned long msg; 3056 pid_t child_pid; 3057 int pipefd[2]; 3058 int status; 3059 siginfo_t info = { }; 3060 struct sock_filter filter[] = { 3061 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 3062 offsetof(struct seccomp_data, nr)), 3063 3064 #ifdef __NR_sigreturn 3065 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 7, 0), 3066 #endif 3067 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 6, 0), 3068 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 5, 0), 3069 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 4, 0), 3070 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 5, 0), 3071 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_clock_nanosleep, 4, 0), 3072 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0), 3073 3074 /* Allow __NR_write for easy logging. */ 3075 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1), 3076 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3077 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 3078 /* The nanosleep jump target. */ 3079 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100), 3080 /* The restart_syscall jump target. */ 3081 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200), 3082 }; 3083 struct sock_fprog prog = { 3084 .len = (unsigned short)ARRAY_SIZE(filter), 3085 .filter = filter, 3086 }; 3087 #if defined(__arm__) 3088 struct utsname utsbuf; 3089 #endif 3090 3091 ASSERT_EQ(0, pipe(pipefd)); 3092 3093 child_pid = fork(); 3094 ASSERT_LE(0, child_pid); 3095 if (child_pid == 0) { 3096 /* Child uses EXPECT not ASSERT to deliver status correctly. */ 3097 char buf = ' '; 3098 struct timespec timeout = { }; 3099 3100 /* Attach parent as tracer and stop. */ 3101 EXPECT_EQ(0, ptrace(PTRACE_TRACEME)); 3102 EXPECT_EQ(0, raise(SIGSTOP)); 3103 3104 EXPECT_EQ(0, close(pipefd[1])); 3105 3106 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 3107 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3108 } 3109 3110 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 3111 EXPECT_EQ(0, ret) { 3112 TH_LOG("Failed to install filter!"); 3113 } 3114 3115 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) { 3116 TH_LOG("Failed to read() sync from parent"); 3117 } 3118 EXPECT_EQ('.', buf) { 3119 TH_LOG("Failed to get sync data from read()"); 3120 } 3121 3122 /* Start nanosleep to be interrupted. */ 3123 timeout.tv_sec = 1; 3124 errno = 0; 3125 EXPECT_EQ(0, nanosleep(&timeout, NULL)) { 3126 TH_LOG("Call to nanosleep() failed (errno %d: %s)", 3127 errno, strerror(errno)); 3128 } 3129 3130 /* Read final sync from parent. */ 3131 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) { 3132 TH_LOG("Failed final read() from parent"); 3133 } 3134 EXPECT_EQ('!', buf) { 3135 TH_LOG("Failed to get final data from read()"); 3136 } 3137 3138 /* Directly report the status of our test harness results. */ 3139 syscall(__NR_exit, _metadata->exit_code); 3140 } 3141 EXPECT_EQ(0, close(pipefd[0])); 3142 3143 /* Attach to child, setup options, and release. */ 3144 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3145 ASSERT_EQ(true, WIFSTOPPED(status)); 3146 ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL, 3147 PTRACE_O_TRACESECCOMP)); 3148 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3149 ASSERT_EQ(1, write(pipefd[1], ".", 1)); 3150 3151 /* Wait for nanosleep() to start. */ 3152 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3153 ASSERT_EQ(true, WIFSTOPPED(status)); 3154 ASSERT_EQ(SIGTRAP, WSTOPSIG(status)); 3155 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16)); 3156 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg)); 3157 ASSERT_EQ(0x100, msg); 3158 ret = get_syscall(_metadata, child_pid); 3159 EXPECT_TRUE(ret == __NR_nanosleep || ret == __NR_clock_nanosleep); 3160 3161 /* Might as well check siginfo for sanity while we're here. */ 3162 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info)); 3163 ASSERT_EQ(SIGTRAP, info.si_signo); 3164 ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code); 3165 EXPECT_EQ(0, info.si_errno); 3166 EXPECT_EQ(getuid(), info.si_uid); 3167 /* Verify signal delivery came from child (seccomp-triggered). */ 3168 EXPECT_EQ(child_pid, info.si_pid); 3169 3170 /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */ 3171 ASSERT_EQ(0, kill(child_pid, SIGSTOP)); 3172 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3173 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3174 ASSERT_EQ(true, WIFSTOPPED(status)); 3175 ASSERT_EQ(SIGSTOP, WSTOPSIG(status)); 3176 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info)); 3177 /* 3178 * There is no siginfo on SIGSTOP any more, so we can't verify 3179 * signal delivery came from parent now (getpid() == info.si_pid). 3180 * https://lkml.kernel.org/r/CAGXu5jJaZAOzP1qFz66tYrtbuywqb+UN2SOA1VLHpCCOiYvYeg@mail.gmail.com 3181 * At least verify the SIGSTOP via PTRACE_GETSIGINFO. 3182 */ 3183 EXPECT_EQ(SIGSTOP, info.si_signo); 3184 3185 /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */ 3186 ASSERT_EQ(0, kill(child_pid, SIGCONT)); 3187 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3188 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3189 ASSERT_EQ(true, WIFSTOPPED(status)); 3190 ASSERT_EQ(SIGCONT, WSTOPSIG(status)); 3191 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3192 3193 /* Wait for restart_syscall() to start. */ 3194 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3195 ASSERT_EQ(true, WIFSTOPPED(status)); 3196 ASSERT_EQ(SIGTRAP, WSTOPSIG(status)); 3197 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16)); 3198 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg)); 3199 3200 ASSERT_EQ(0x200, msg); 3201 ret = get_syscall(_metadata, child_pid); 3202 #if defined(__arm__) 3203 /* 3204 * - native ARM registers do NOT expose true syscall. 3205 * - compat ARM registers on ARM64 DO expose true syscall. 3206 * - values of utsbuf.machine include 'armv8l' or 'armb8b' 3207 * for ARM64 running in compat mode. 3208 */ 3209 ASSERT_EQ(0, uname(&utsbuf)); 3210 if ((strncmp(utsbuf.machine, "arm", 3) == 0) && 3211 (strncmp(utsbuf.machine, "armv8l", 6) != 0) && 3212 (strncmp(utsbuf.machine, "armv8b", 6) != 0)) { 3213 EXPECT_EQ(__NR_nanosleep, ret); 3214 } else 3215 #endif 3216 { 3217 EXPECT_EQ(__NR_restart_syscall, ret); 3218 } 3219 3220 /* Write again to end test. */ 3221 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3222 ASSERT_EQ(1, write(pipefd[1], "!", 1)); 3223 EXPECT_EQ(0, close(pipefd[1])); 3224 3225 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3226 if (WIFSIGNALED(status) || WEXITSTATUS(status)) 3227 _metadata->exit_code = KSFT_FAIL; 3228 } 3229 3230 TEST_SIGNAL(filter_flag_log, SIGSYS) 3231 { 3232 struct sock_filter allow_filter[] = { 3233 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3234 }; 3235 struct sock_filter kill_filter[] = { 3236 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 3237 offsetof(struct seccomp_data, nr)), 3238 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 3239 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 3240 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3241 }; 3242 struct sock_fprog allow_prog = { 3243 .len = (unsigned short)ARRAY_SIZE(allow_filter), 3244 .filter = allow_filter, 3245 }; 3246 struct sock_fprog kill_prog = { 3247 .len = (unsigned short)ARRAY_SIZE(kill_filter), 3248 .filter = kill_filter, 3249 }; 3250 long ret; 3251 pid_t parent = getppid(); 3252 3253 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3254 ASSERT_EQ(0, ret); 3255 3256 /* Verify that the FILTER_FLAG_LOG flag isn't accepted in strict mode */ 3257 ret = seccomp(SECCOMP_SET_MODE_STRICT, SECCOMP_FILTER_FLAG_LOG, 3258 &allow_prog); 3259 ASSERT_NE(ENOSYS, errno) { 3260 TH_LOG("Kernel does not support seccomp syscall!"); 3261 } 3262 EXPECT_NE(0, ret) { 3263 TH_LOG("Kernel accepted FILTER_FLAG_LOG flag in strict mode!"); 3264 } 3265 EXPECT_EQ(EINVAL, errno) { 3266 TH_LOG("Kernel returned unexpected errno for FILTER_FLAG_LOG flag in strict mode!"); 3267 } 3268 3269 /* Verify that a simple, permissive filter can be added with no flags */ 3270 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog); 3271 EXPECT_EQ(0, ret); 3272 3273 /* See if the same filter can be added with the FILTER_FLAG_LOG flag */ 3274 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG, 3275 &allow_prog); 3276 ASSERT_NE(EINVAL, errno) { 3277 TH_LOG("Kernel does not support the FILTER_FLAG_LOG flag!"); 3278 } 3279 EXPECT_EQ(0, ret); 3280 3281 /* Ensure that the kill filter works with the FILTER_FLAG_LOG flag */ 3282 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG, 3283 &kill_prog); 3284 EXPECT_EQ(0, ret); 3285 3286 EXPECT_EQ(parent, syscall(__NR_getppid)); 3287 /* getpid() should never return. */ 3288 EXPECT_EQ(0, syscall(__NR_getpid)); 3289 } 3290 3291 TEST(get_action_avail) 3292 { 3293 __u32 actions[] = { SECCOMP_RET_KILL_THREAD, SECCOMP_RET_TRAP, 3294 SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE, 3295 SECCOMP_RET_LOG, SECCOMP_RET_ALLOW }; 3296 __u32 unknown_action = 0x10000000U; 3297 int i; 3298 long ret; 3299 3300 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[0]); 3301 ASSERT_NE(ENOSYS, errno) { 3302 TH_LOG("Kernel does not support seccomp syscall!"); 3303 } 3304 ASSERT_NE(EINVAL, errno) { 3305 TH_LOG("Kernel does not support SECCOMP_GET_ACTION_AVAIL operation!"); 3306 } 3307 EXPECT_EQ(ret, 0); 3308 3309 for (i = 0; i < ARRAY_SIZE(actions); i++) { 3310 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[i]); 3311 EXPECT_EQ(ret, 0) { 3312 TH_LOG("Expected action (0x%X) not available!", 3313 actions[i]); 3314 } 3315 } 3316 3317 /* Check that an unknown action is handled properly (EOPNOTSUPP) */ 3318 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &unknown_action); 3319 EXPECT_EQ(ret, -1); 3320 EXPECT_EQ(errno, EOPNOTSUPP); 3321 } 3322 3323 TEST(get_metadata) 3324 { 3325 pid_t pid; 3326 int pipefd[2]; 3327 char buf; 3328 struct seccomp_metadata md; 3329 long ret; 3330 3331 /* Only real root can get metadata. */ 3332 if (geteuid()) { 3333 SKIP(return, "get_metadata requires real root"); 3334 return; 3335 } 3336 3337 ASSERT_EQ(0, pipe(pipefd)); 3338 3339 pid = fork(); 3340 ASSERT_GE(pid, 0); 3341 if (pid == 0) { 3342 struct sock_filter filter[] = { 3343 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3344 }; 3345 struct sock_fprog prog = { 3346 .len = (unsigned short)ARRAY_SIZE(filter), 3347 .filter = filter, 3348 }; 3349 3350 /* one with log, one without */ 3351 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 3352 SECCOMP_FILTER_FLAG_LOG, &prog)); 3353 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog)); 3354 3355 EXPECT_EQ(0, close(pipefd[0])); 3356 ASSERT_EQ(1, write(pipefd[1], "1", 1)); 3357 ASSERT_EQ(0, close(pipefd[1])); 3358 3359 while (1) 3360 sleep(100); 3361 } 3362 3363 ASSERT_EQ(0, close(pipefd[1])); 3364 ASSERT_EQ(1, read(pipefd[0], &buf, 1)); 3365 3366 ASSERT_EQ(0, ptrace(PTRACE_ATTACH, pid)); 3367 ASSERT_EQ(pid, waitpid(pid, NULL, 0)); 3368 3369 /* Past here must not use ASSERT or child process is never killed. */ 3370 3371 md.filter_off = 0; 3372 errno = 0; 3373 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md); 3374 EXPECT_EQ(sizeof(md), ret) { 3375 if (errno == EINVAL) 3376 SKIP(goto skip, "Kernel does not support PTRACE_SECCOMP_GET_METADATA (missing CONFIG_CHECKPOINT_RESTORE?)"); 3377 } 3378 3379 EXPECT_EQ(md.flags, SECCOMP_FILTER_FLAG_LOG); 3380 EXPECT_EQ(md.filter_off, 0); 3381 3382 md.filter_off = 1; 3383 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md); 3384 EXPECT_EQ(sizeof(md), ret); 3385 EXPECT_EQ(md.flags, 0); 3386 EXPECT_EQ(md.filter_off, 1); 3387 3388 skip: 3389 ASSERT_EQ(0, kill(pid, SIGKILL)); 3390 } 3391 3392 static int user_notif_syscall(int nr, unsigned int flags) 3393 { 3394 struct sock_filter filter[] = { 3395 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 3396 offsetof(struct seccomp_data, nr)), 3397 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, nr, 0, 1), 3398 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_USER_NOTIF), 3399 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3400 }; 3401 3402 struct sock_fprog prog = { 3403 .len = (unsigned short)ARRAY_SIZE(filter), 3404 .filter = filter, 3405 }; 3406 3407 return seccomp(SECCOMP_SET_MODE_FILTER, flags, &prog); 3408 } 3409 3410 #define USER_NOTIF_MAGIC INT_MAX 3411 TEST(user_notification_basic) 3412 { 3413 pid_t pid; 3414 long ret; 3415 int status, listener; 3416 struct seccomp_notif req = {}; 3417 struct seccomp_notif_resp resp = {}; 3418 struct pollfd pollfd; 3419 3420 struct sock_filter filter[] = { 3421 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3422 }; 3423 struct sock_fprog prog = { 3424 .len = (unsigned short)ARRAY_SIZE(filter), 3425 .filter = filter, 3426 }; 3427 3428 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3429 ASSERT_EQ(0, ret) { 3430 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3431 } 3432 3433 pid = fork(); 3434 ASSERT_GE(pid, 0); 3435 3436 /* Check that we get -ENOSYS with no listener attached */ 3437 if (pid == 0) { 3438 if (user_notif_syscall(__NR_getppid, 0) < 0) 3439 exit(1); 3440 ret = syscall(__NR_getppid); 3441 exit(ret >= 0 || errno != ENOSYS); 3442 } 3443 3444 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3445 EXPECT_EQ(true, WIFEXITED(status)); 3446 EXPECT_EQ(0, WEXITSTATUS(status)); 3447 3448 /* Add some no-op filters for grins. */ 3449 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3450 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3451 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3452 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3453 3454 /* Check that the basic notification machinery works */ 3455 listener = user_notif_syscall(__NR_getppid, 3456 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3457 ASSERT_GE(listener, 0); 3458 3459 /* Installing a second listener in the chain should EBUSY */ 3460 EXPECT_EQ(user_notif_syscall(__NR_getppid, 3461 SECCOMP_FILTER_FLAG_NEW_LISTENER), 3462 -1); 3463 EXPECT_EQ(errno, EBUSY); 3464 3465 pid = fork(); 3466 ASSERT_GE(pid, 0); 3467 3468 if (pid == 0) { 3469 ret = syscall(__NR_getppid); 3470 exit(ret != USER_NOTIF_MAGIC); 3471 } 3472 3473 pollfd.fd = listener; 3474 pollfd.events = POLLIN | POLLOUT; 3475 3476 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3477 EXPECT_EQ(pollfd.revents, POLLIN); 3478 3479 /* Test that we can't pass garbage to the kernel. */ 3480 memset(&req, 0, sizeof(req)); 3481 req.pid = -1; 3482 errno = 0; 3483 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req); 3484 EXPECT_EQ(-1, ret); 3485 EXPECT_EQ(EINVAL, errno); 3486 3487 if (ret) { 3488 req.pid = 0; 3489 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3490 } 3491 3492 pollfd.fd = listener; 3493 pollfd.events = POLLIN | POLLOUT; 3494 3495 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3496 EXPECT_EQ(pollfd.revents, POLLOUT); 3497 3498 EXPECT_EQ(req.data.nr, __NR_getppid); 3499 3500 resp.id = req.id; 3501 resp.error = 0; 3502 resp.val = USER_NOTIF_MAGIC; 3503 3504 /* check that we make sure flags == 0 */ 3505 resp.flags = 1; 3506 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3507 EXPECT_EQ(errno, EINVAL); 3508 3509 resp.flags = 0; 3510 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3511 3512 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3513 EXPECT_EQ(true, WIFEXITED(status)); 3514 EXPECT_EQ(0, WEXITSTATUS(status)); 3515 } 3516 3517 TEST(user_notification_with_tsync) 3518 { 3519 int ret; 3520 unsigned int flags; 3521 3522 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3523 ASSERT_EQ(0, ret) { 3524 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3525 } 3526 3527 /* these were exclusive */ 3528 flags = SECCOMP_FILTER_FLAG_NEW_LISTENER | 3529 SECCOMP_FILTER_FLAG_TSYNC; 3530 ASSERT_EQ(-1, user_notif_syscall(__NR_getppid, flags)); 3531 ASSERT_EQ(EINVAL, errno); 3532 3533 /* but now they're not */ 3534 flags |= SECCOMP_FILTER_FLAG_TSYNC_ESRCH; 3535 ret = user_notif_syscall(__NR_getppid, flags); 3536 close(ret); 3537 ASSERT_LE(0, ret); 3538 } 3539 3540 TEST(user_notification_kill_in_middle) 3541 { 3542 pid_t pid; 3543 long ret; 3544 int listener; 3545 struct seccomp_notif req = {}; 3546 struct seccomp_notif_resp resp = {}; 3547 3548 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3549 ASSERT_EQ(0, ret) { 3550 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3551 } 3552 3553 listener = user_notif_syscall(__NR_getppid, 3554 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3555 ASSERT_GE(listener, 0); 3556 3557 /* 3558 * Check that nothing bad happens when we kill the task in the middle 3559 * of a syscall. 3560 */ 3561 pid = fork(); 3562 ASSERT_GE(pid, 0); 3563 3564 if (pid == 0) { 3565 ret = syscall(__NR_getppid); 3566 exit(ret != USER_NOTIF_MAGIC); 3567 } 3568 3569 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3570 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), 0); 3571 3572 EXPECT_EQ(kill(pid, SIGKILL), 0); 3573 EXPECT_EQ(waitpid(pid, NULL, 0), pid); 3574 3575 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), -1); 3576 3577 resp.id = req.id; 3578 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp); 3579 EXPECT_EQ(ret, -1); 3580 EXPECT_EQ(errno, ENOENT); 3581 } 3582 3583 static int handled = -1; 3584 3585 static void signal_handler(int signal) 3586 { 3587 if (write(handled, "c", 1) != 1) 3588 perror("write from signal"); 3589 } 3590 3591 static void signal_handler_nop(int signal) 3592 { 3593 } 3594 3595 TEST(user_notification_signal) 3596 { 3597 pid_t pid; 3598 long ret; 3599 int status, listener, sk_pair[2]; 3600 struct seccomp_notif req = {}; 3601 struct seccomp_notif_resp resp = {}; 3602 char c; 3603 3604 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3605 ASSERT_EQ(0, ret) { 3606 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3607 } 3608 3609 ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0); 3610 3611 listener = user_notif_syscall(__NR_gettid, 3612 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3613 ASSERT_GE(listener, 0); 3614 3615 pid = fork(); 3616 ASSERT_GE(pid, 0); 3617 3618 if (pid == 0) { 3619 close(sk_pair[0]); 3620 handled = sk_pair[1]; 3621 if (signal(SIGUSR1, signal_handler) == SIG_ERR) { 3622 perror("signal"); 3623 exit(1); 3624 } 3625 /* 3626 * ERESTARTSYS behavior is a bit hard to test, because we need 3627 * to rely on a signal that has not yet been handled. Let's at 3628 * least check that the error code gets propagated through, and 3629 * hope that it doesn't break when there is actually a signal :) 3630 */ 3631 ret = syscall(__NR_gettid); 3632 exit(!(ret == -1 && errno == 512)); 3633 } 3634 3635 close(sk_pair[1]); 3636 3637 memset(&req, 0, sizeof(req)); 3638 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3639 3640 EXPECT_EQ(kill(pid, SIGUSR1), 0); 3641 3642 /* 3643 * Make sure the signal really is delivered, which means we're not 3644 * stuck in the user notification code any more and the notification 3645 * should be dead. 3646 */ 3647 EXPECT_EQ(read(sk_pair[0], &c, 1), 1); 3648 3649 resp.id = req.id; 3650 resp.error = -EPERM; 3651 resp.val = 0; 3652 3653 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3654 EXPECT_EQ(errno, ENOENT); 3655 3656 memset(&req, 0, sizeof(req)); 3657 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3658 3659 resp.id = req.id; 3660 resp.error = -512; /* -ERESTARTSYS */ 3661 resp.val = 0; 3662 3663 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3664 3665 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3666 EXPECT_EQ(true, WIFEXITED(status)); 3667 EXPECT_EQ(0, WEXITSTATUS(status)); 3668 } 3669 3670 TEST(user_notification_closed_listener) 3671 { 3672 pid_t pid; 3673 long ret; 3674 int status, listener; 3675 3676 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3677 ASSERT_EQ(0, ret) { 3678 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3679 } 3680 3681 listener = user_notif_syscall(__NR_getppid, 3682 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3683 ASSERT_GE(listener, 0); 3684 3685 /* 3686 * Check that we get an ENOSYS when the listener is closed. 3687 */ 3688 pid = fork(); 3689 ASSERT_GE(pid, 0); 3690 if (pid == 0) { 3691 close(listener); 3692 ret = syscall(__NR_getppid); 3693 exit(ret != -1 && errno != ENOSYS); 3694 } 3695 3696 close(listener); 3697 3698 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3699 EXPECT_EQ(true, WIFEXITED(status)); 3700 EXPECT_EQ(0, WEXITSTATUS(status)); 3701 } 3702 3703 /* 3704 * Check that a pid in a child namespace still shows up as valid in ours. 3705 */ 3706 TEST(user_notification_child_pid_ns) 3707 { 3708 pid_t pid; 3709 int status, listener; 3710 struct seccomp_notif req = {}; 3711 struct seccomp_notif_resp resp = {}; 3712 3713 ASSERT_EQ(unshare(CLONE_NEWUSER | CLONE_NEWPID), 0) { 3714 if (errno == EINVAL) 3715 SKIP(return, "kernel missing CLONE_NEWUSER support"); 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 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3727 3728 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3729 EXPECT_EQ(req.pid, pid); 3730 3731 resp.id = req.id; 3732 resp.error = 0; 3733 resp.val = USER_NOTIF_MAGIC; 3734 3735 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3736 3737 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3738 EXPECT_EQ(true, WIFEXITED(status)); 3739 EXPECT_EQ(0, WEXITSTATUS(status)); 3740 close(listener); 3741 } 3742 3743 /* 3744 * Check that a pid in a sibling (i.e. unrelated) namespace shows up as 0, i.e. 3745 * invalid. 3746 */ 3747 TEST(user_notification_sibling_pid_ns) 3748 { 3749 pid_t pid, pid2; 3750 int status, listener; 3751 struct seccomp_notif req = {}; 3752 struct seccomp_notif_resp resp = {}; 3753 3754 ASSERT_EQ(prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0), 0) { 3755 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3756 } 3757 3758 listener = user_notif_syscall(__NR_getppid, 3759 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3760 ASSERT_GE(listener, 0); 3761 3762 pid = fork(); 3763 ASSERT_GE(pid, 0); 3764 3765 if (pid == 0) { 3766 ASSERT_EQ(unshare(CLONE_NEWPID), 0) { 3767 if (errno == EPERM) 3768 SKIP(return, "CLONE_NEWPID requires CAP_SYS_ADMIN"); 3769 else if (errno == EINVAL) 3770 SKIP(return, "CLONE_NEWPID is invalid (missing CONFIG_PID_NS?)"); 3771 } 3772 3773 pid2 = fork(); 3774 ASSERT_GE(pid2, 0); 3775 3776 if (pid2 == 0) 3777 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3778 3779 EXPECT_EQ(waitpid(pid2, &status, 0), pid2); 3780 EXPECT_EQ(true, WIFEXITED(status)); 3781 EXPECT_EQ(0, WEXITSTATUS(status)); 3782 exit(WEXITSTATUS(status)); 3783 } 3784 3785 /* Create the sibling ns, and sibling in it. */ 3786 ASSERT_EQ(unshare(CLONE_NEWPID), 0) { 3787 if (errno == EPERM) 3788 SKIP(return, "CLONE_NEWPID requires CAP_SYS_ADMIN"); 3789 else if (errno == EINVAL) 3790 SKIP(return, "CLONE_NEWPID is invalid (missing CONFIG_PID_NS?)"); 3791 } 3792 ASSERT_EQ(errno, 0); 3793 3794 pid2 = fork(); 3795 ASSERT_GE(pid2, 0); 3796 3797 if (pid2 == 0) { 3798 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3799 /* 3800 * The pid should be 0, i.e. the task is in some namespace that 3801 * we can't "see". 3802 */ 3803 EXPECT_EQ(req.pid, 0); 3804 3805 resp.id = req.id; 3806 resp.error = 0; 3807 resp.val = USER_NOTIF_MAGIC; 3808 3809 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3810 exit(0); 3811 } 3812 3813 close(listener); 3814 3815 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3816 EXPECT_EQ(true, WIFEXITED(status)); 3817 EXPECT_EQ(0, WEXITSTATUS(status)); 3818 3819 EXPECT_EQ(waitpid(pid2, &status, 0), pid2); 3820 EXPECT_EQ(true, WIFEXITED(status)); 3821 EXPECT_EQ(0, WEXITSTATUS(status)); 3822 } 3823 3824 TEST(user_notification_fault_recv) 3825 { 3826 pid_t pid; 3827 int status, listener; 3828 struct seccomp_notif req = {}; 3829 struct seccomp_notif_resp resp = {}; 3830 3831 ASSERT_EQ(unshare(CLONE_NEWUSER), 0) { 3832 if (errno == EINVAL) 3833 SKIP(return, "kernel missing CLONE_NEWUSER support"); 3834 } 3835 3836 listener = user_notif_syscall(__NR_getppid, 3837 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3838 ASSERT_GE(listener, 0); 3839 3840 pid = fork(); 3841 ASSERT_GE(pid, 0); 3842 3843 if (pid == 0) 3844 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3845 3846 /* Do a bad recv() */ 3847 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, NULL), -1); 3848 EXPECT_EQ(errno, EFAULT); 3849 3850 /* We should still be able to receive this notification, though. */ 3851 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3852 EXPECT_EQ(req.pid, pid); 3853 3854 resp.id = req.id; 3855 resp.error = 0; 3856 resp.val = USER_NOTIF_MAGIC; 3857 3858 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3859 3860 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3861 EXPECT_EQ(true, WIFEXITED(status)); 3862 EXPECT_EQ(0, WEXITSTATUS(status)); 3863 } 3864 3865 TEST(seccomp_get_notif_sizes) 3866 { 3867 struct seccomp_notif_sizes sizes; 3868 3869 ASSERT_EQ(seccomp(SECCOMP_GET_NOTIF_SIZES, 0, &sizes), 0); 3870 EXPECT_EQ(sizes.seccomp_notif, sizeof(struct seccomp_notif)); 3871 EXPECT_EQ(sizes.seccomp_notif_resp, sizeof(struct seccomp_notif_resp)); 3872 } 3873 3874 TEST(user_notification_continue) 3875 { 3876 pid_t pid; 3877 long ret; 3878 int status, listener; 3879 struct seccomp_notif req = {}; 3880 struct seccomp_notif_resp resp = {}; 3881 struct pollfd pollfd; 3882 3883 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3884 ASSERT_EQ(0, ret) { 3885 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3886 } 3887 3888 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER); 3889 ASSERT_GE(listener, 0); 3890 3891 pid = fork(); 3892 ASSERT_GE(pid, 0); 3893 3894 if (pid == 0) { 3895 int dup_fd, pipe_fds[2]; 3896 pid_t self; 3897 3898 ASSERT_GE(pipe(pipe_fds), 0); 3899 3900 dup_fd = dup(pipe_fds[0]); 3901 ASSERT_GE(dup_fd, 0); 3902 EXPECT_NE(pipe_fds[0], dup_fd); 3903 3904 self = getpid(); 3905 ASSERT_EQ(filecmp(self, self, pipe_fds[0], dup_fd), 0); 3906 exit(0); 3907 } 3908 3909 pollfd.fd = listener; 3910 pollfd.events = POLLIN | POLLOUT; 3911 3912 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3913 EXPECT_EQ(pollfd.revents, POLLIN); 3914 3915 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3916 3917 pollfd.fd = listener; 3918 pollfd.events = POLLIN | POLLOUT; 3919 3920 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3921 EXPECT_EQ(pollfd.revents, POLLOUT); 3922 3923 EXPECT_EQ(req.data.nr, __NR_dup); 3924 3925 resp.id = req.id; 3926 resp.flags = SECCOMP_USER_NOTIF_FLAG_CONTINUE; 3927 3928 /* 3929 * Verify that setting SECCOMP_USER_NOTIF_FLAG_CONTINUE enforces other 3930 * args be set to 0. 3931 */ 3932 resp.error = 0; 3933 resp.val = USER_NOTIF_MAGIC; 3934 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3935 EXPECT_EQ(errno, EINVAL); 3936 3937 resp.error = USER_NOTIF_MAGIC; 3938 resp.val = 0; 3939 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3940 EXPECT_EQ(errno, EINVAL); 3941 3942 resp.error = 0; 3943 resp.val = 0; 3944 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0) { 3945 if (errno == EINVAL) 3946 SKIP(goto skip, "Kernel does not support SECCOMP_USER_NOTIF_FLAG_CONTINUE"); 3947 } 3948 3949 skip: 3950 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3951 EXPECT_EQ(true, WIFEXITED(status)); 3952 EXPECT_EQ(0, WEXITSTATUS(status)) { 3953 if (WEXITSTATUS(status) == 2) { 3954 SKIP(return, "Kernel does not support kcmp() syscall"); 3955 return; 3956 } 3957 } 3958 } 3959 3960 TEST(user_notification_filter_empty) 3961 { 3962 pid_t pid; 3963 long ret; 3964 int status; 3965 struct pollfd pollfd; 3966 struct __clone_args args = { 3967 .flags = CLONE_FILES, 3968 .exit_signal = SIGCHLD, 3969 }; 3970 3971 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3972 ASSERT_EQ(0, ret) { 3973 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3974 } 3975 3976 if (__NR_clone3 < 0) 3977 SKIP(return, "Test not built with clone3 support"); 3978 3979 pid = sys_clone3(&args, sizeof(args)); 3980 ASSERT_GE(pid, 0); 3981 3982 if (pid == 0) { 3983 int listener; 3984 3985 listener = user_notif_syscall(__NR_mknodat, SECCOMP_FILTER_FLAG_NEW_LISTENER); 3986 if (listener < 0) 3987 _exit(EXIT_FAILURE); 3988 3989 if (dup2(listener, 200) != 200) 3990 _exit(EXIT_FAILURE); 3991 3992 close(listener); 3993 3994 _exit(EXIT_SUCCESS); 3995 } 3996 3997 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3998 EXPECT_EQ(true, WIFEXITED(status)); 3999 EXPECT_EQ(0, WEXITSTATUS(status)); 4000 4001 /* 4002 * The seccomp filter has become unused so we should be notified once 4003 * the kernel gets around to cleaning up task struct. 4004 */ 4005 pollfd.fd = 200; 4006 pollfd.events = POLLHUP; 4007 4008 EXPECT_GT(poll(&pollfd, 1, 2000), 0); 4009 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0); 4010 } 4011 4012 TEST(user_ioctl_notification_filter_empty) 4013 { 4014 pid_t pid; 4015 long ret; 4016 int status, p[2]; 4017 struct __clone_args args = { 4018 .flags = CLONE_FILES, 4019 .exit_signal = SIGCHLD, 4020 }; 4021 struct seccomp_notif req = {}; 4022 4023 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4024 ASSERT_EQ(0, ret) { 4025 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4026 } 4027 4028 if (__NR_clone3 < 0) 4029 SKIP(return, "Test not built with clone3 support"); 4030 4031 ASSERT_EQ(0, pipe(p)); 4032 4033 pid = sys_clone3(&args, sizeof(args)); 4034 ASSERT_GE(pid, 0); 4035 4036 if (pid == 0) { 4037 int listener; 4038 4039 listener = user_notif_syscall(__NR_mknodat, SECCOMP_FILTER_FLAG_NEW_LISTENER); 4040 if (listener < 0) 4041 _exit(EXIT_FAILURE); 4042 4043 if (dup2(listener, 200) != 200) 4044 _exit(EXIT_FAILURE); 4045 close(p[1]); 4046 close(listener); 4047 sleep(1); 4048 4049 _exit(EXIT_SUCCESS); 4050 } 4051 if (read(p[0], &status, 1) != 0) 4052 _exit(EXIT_SUCCESS); 4053 close(p[0]); 4054 /* 4055 * The seccomp filter has become unused so we should be notified once 4056 * the kernel gets around to cleaning up task struct. 4057 */ 4058 EXPECT_EQ(ioctl(200, SECCOMP_IOCTL_NOTIF_RECV, &req), -1); 4059 EXPECT_EQ(errno, ENOENT); 4060 4061 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4062 EXPECT_EQ(true, WIFEXITED(status)); 4063 EXPECT_EQ(0, WEXITSTATUS(status)); 4064 } 4065 4066 static void *do_thread(void *data) 4067 { 4068 return NULL; 4069 } 4070 4071 TEST(user_notification_filter_empty_threaded) 4072 { 4073 pid_t pid; 4074 long ret; 4075 int status; 4076 struct pollfd pollfd; 4077 struct __clone_args args = { 4078 .flags = CLONE_FILES, 4079 .exit_signal = SIGCHLD, 4080 }; 4081 4082 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4083 ASSERT_EQ(0, ret) { 4084 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4085 } 4086 4087 if (__NR_clone3 < 0) 4088 SKIP(return, "Test not built with clone3 support"); 4089 4090 pid = sys_clone3(&args, sizeof(args)); 4091 ASSERT_GE(pid, 0); 4092 4093 if (pid == 0) { 4094 pid_t pid1, pid2; 4095 int listener, status; 4096 pthread_t thread; 4097 4098 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER); 4099 if (listener < 0) 4100 _exit(EXIT_FAILURE); 4101 4102 if (dup2(listener, 200) != 200) 4103 _exit(EXIT_FAILURE); 4104 4105 close(listener); 4106 4107 pid1 = fork(); 4108 if (pid1 < 0) 4109 _exit(EXIT_FAILURE); 4110 4111 if (pid1 == 0) 4112 _exit(EXIT_SUCCESS); 4113 4114 pid2 = fork(); 4115 if (pid2 < 0) 4116 _exit(EXIT_FAILURE); 4117 4118 if (pid2 == 0) 4119 _exit(EXIT_SUCCESS); 4120 4121 if (pthread_create(&thread, NULL, do_thread, NULL) || 4122 pthread_join(thread, NULL)) 4123 _exit(EXIT_FAILURE); 4124 4125 if (pthread_create(&thread, NULL, do_thread, NULL) || 4126 pthread_join(thread, NULL)) 4127 _exit(EXIT_FAILURE); 4128 4129 if (waitpid(pid1, &status, 0) != pid1 || !WIFEXITED(status) || 4130 WEXITSTATUS(status)) 4131 _exit(EXIT_FAILURE); 4132 4133 if (waitpid(pid2, &status, 0) != pid2 || !WIFEXITED(status) || 4134 WEXITSTATUS(status)) 4135 _exit(EXIT_FAILURE); 4136 4137 exit(EXIT_SUCCESS); 4138 } 4139 4140 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4141 EXPECT_EQ(true, WIFEXITED(status)); 4142 EXPECT_EQ(0, WEXITSTATUS(status)); 4143 4144 /* 4145 * The seccomp filter has become unused so we should be notified once 4146 * the kernel gets around to cleaning up task struct. 4147 */ 4148 pollfd.fd = 200; 4149 pollfd.events = POLLHUP; 4150 4151 EXPECT_GT(poll(&pollfd, 1, 2000), 0); 4152 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0); 4153 } 4154 4155 4156 int get_next_fd(int prev_fd) 4157 { 4158 for (int i = prev_fd + 1; i < FD_SETSIZE; ++i) { 4159 if (fcntl(i, F_GETFD) == -1) 4160 return i; 4161 } 4162 _exit(EXIT_FAILURE); 4163 } 4164 4165 TEST(user_notification_addfd) 4166 { 4167 pid_t pid; 4168 long ret; 4169 int status, listener, memfd, fd, nextfd; 4170 struct seccomp_notif_addfd addfd = {}; 4171 struct seccomp_notif_addfd_small small = {}; 4172 struct seccomp_notif_addfd_big big = {}; 4173 struct seccomp_notif req = {}; 4174 struct seccomp_notif_resp resp = {}; 4175 /* 100 ms */ 4176 struct timespec delay = { .tv_nsec = 100000000 }; 4177 4178 /* There may be arbitrary already-open fds at test start. */ 4179 memfd = memfd_create("test", 0); 4180 ASSERT_GE(memfd, 0); 4181 nextfd = get_next_fd(memfd); 4182 4183 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4184 ASSERT_EQ(0, ret) { 4185 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4186 } 4187 4188 /* fd: 4 */ 4189 /* Check that the basic notification machinery works */ 4190 listener = user_notif_syscall(__NR_getppid, 4191 SECCOMP_FILTER_FLAG_NEW_LISTENER); 4192 ASSERT_EQ(listener, nextfd); 4193 nextfd = get_next_fd(nextfd); 4194 4195 pid = fork(); 4196 ASSERT_GE(pid, 0); 4197 4198 if (pid == 0) { 4199 /* fds will be added and this value is expected */ 4200 if (syscall(__NR_getppid) != USER_NOTIF_MAGIC) 4201 exit(1); 4202 4203 /* Atomic addfd+send is received here. Check it is a valid fd */ 4204 if (fcntl(syscall(__NR_getppid), F_GETFD) == -1) 4205 exit(1); 4206 4207 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 4208 } 4209 4210 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4211 4212 addfd.srcfd = memfd; 4213 addfd.newfd = 0; 4214 addfd.id = req.id; 4215 addfd.flags = 0x0; 4216 4217 /* Verify bad newfd_flags cannot be set */ 4218 addfd.newfd_flags = ~O_CLOEXEC; 4219 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4220 EXPECT_EQ(errno, EINVAL); 4221 addfd.newfd_flags = O_CLOEXEC; 4222 4223 /* Verify bad flags cannot be set */ 4224 addfd.flags = 0xff; 4225 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4226 EXPECT_EQ(errno, EINVAL); 4227 addfd.flags = 0; 4228 4229 /* Verify that remote_fd cannot be set without setting flags */ 4230 addfd.newfd = 1; 4231 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4232 EXPECT_EQ(errno, EINVAL); 4233 addfd.newfd = 0; 4234 4235 /* Verify small size cannot be set */ 4236 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_SMALL, &small), -1); 4237 EXPECT_EQ(errno, EINVAL); 4238 4239 /* Verify we can't send bits filled in unknown buffer area */ 4240 memset(&big, 0xAA, sizeof(big)); 4241 big.addfd = addfd; 4242 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big), -1); 4243 EXPECT_EQ(errno, E2BIG); 4244 4245 4246 /* Verify we can set an arbitrary remote fd */ 4247 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd); 4248 EXPECT_EQ(fd, nextfd); 4249 nextfd = get_next_fd(nextfd); 4250 EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0); 4251 4252 /* Verify we can set an arbitrary remote fd with large size */ 4253 memset(&big, 0x0, sizeof(big)); 4254 big.addfd = addfd; 4255 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big); 4256 EXPECT_EQ(fd, nextfd); 4257 nextfd = get_next_fd(nextfd); 4258 4259 /* Verify we can set a specific remote fd */ 4260 addfd.newfd = 42; 4261 addfd.flags = SECCOMP_ADDFD_FLAG_SETFD; 4262 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd); 4263 EXPECT_EQ(fd, 42); 4264 EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0); 4265 4266 /* Resume syscall */ 4267 resp.id = req.id; 4268 resp.error = 0; 4269 resp.val = USER_NOTIF_MAGIC; 4270 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4271 4272 /* 4273 * This sets the ID of the ADD FD to the last request plus 1. The 4274 * notification ID increments 1 per notification. 4275 */ 4276 addfd.id = req.id + 1; 4277 4278 /* This spins until the underlying notification is generated */ 4279 while (ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd) != -1 && 4280 errno != -EINPROGRESS) 4281 nanosleep(&delay, NULL); 4282 4283 memset(&req, 0, sizeof(req)); 4284 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4285 ASSERT_EQ(addfd.id, req.id); 4286 4287 /* Verify we can do an atomic addfd and send */ 4288 addfd.newfd = 0; 4289 addfd.flags = SECCOMP_ADDFD_FLAG_SEND; 4290 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd); 4291 /* 4292 * Child has earlier "low" fds and now 42, so we expect the next 4293 * lowest available fd to be assigned here. 4294 */ 4295 EXPECT_EQ(fd, nextfd); 4296 nextfd = get_next_fd(nextfd); 4297 ASSERT_EQ(filecmp(getpid(), pid, memfd, fd), 0); 4298 4299 /* 4300 * This sets the ID of the ADD FD to the last request plus 1. The 4301 * notification ID increments 1 per notification. 4302 */ 4303 addfd.id = req.id + 1; 4304 4305 /* This spins until the underlying notification is generated */ 4306 while (ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd) != -1 && 4307 errno != -EINPROGRESS) 4308 nanosleep(&delay, NULL); 4309 4310 memset(&req, 0, sizeof(req)); 4311 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4312 ASSERT_EQ(addfd.id, req.id); 4313 4314 resp.id = req.id; 4315 resp.error = 0; 4316 resp.val = USER_NOTIF_MAGIC; 4317 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4318 4319 /* Wait for child to finish. */ 4320 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4321 EXPECT_EQ(true, WIFEXITED(status)); 4322 EXPECT_EQ(0, WEXITSTATUS(status)); 4323 4324 close(memfd); 4325 } 4326 4327 TEST(user_notification_addfd_rlimit) 4328 { 4329 pid_t pid; 4330 long ret; 4331 int status, listener, memfd; 4332 struct seccomp_notif_addfd addfd = {}; 4333 struct seccomp_notif req = {}; 4334 struct seccomp_notif_resp resp = {}; 4335 const struct rlimit lim = { 4336 .rlim_cur = 0, 4337 .rlim_max = 0, 4338 }; 4339 4340 memfd = memfd_create("test", 0); 4341 ASSERT_GE(memfd, 0); 4342 4343 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4344 ASSERT_EQ(0, ret) { 4345 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4346 } 4347 4348 /* Check that the basic notification machinery works */ 4349 listener = user_notif_syscall(__NR_getppid, 4350 SECCOMP_FILTER_FLAG_NEW_LISTENER); 4351 ASSERT_GE(listener, 0); 4352 4353 pid = fork(); 4354 ASSERT_GE(pid, 0); 4355 4356 if (pid == 0) 4357 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 4358 4359 4360 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4361 4362 ASSERT_EQ(prlimit(pid, RLIMIT_NOFILE, &lim, NULL), 0); 4363 4364 addfd.srcfd = memfd; 4365 addfd.newfd_flags = O_CLOEXEC; 4366 addfd.newfd = 0; 4367 addfd.id = req.id; 4368 addfd.flags = 0; 4369 4370 /* Should probably spot check /proc/sys/fs/file-nr */ 4371 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4372 EXPECT_EQ(errno, EMFILE); 4373 4374 addfd.flags = SECCOMP_ADDFD_FLAG_SEND; 4375 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4376 EXPECT_EQ(errno, EMFILE); 4377 4378 addfd.newfd = 100; 4379 addfd.flags = SECCOMP_ADDFD_FLAG_SETFD; 4380 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4381 EXPECT_EQ(errno, EBADF); 4382 4383 resp.id = req.id; 4384 resp.error = 0; 4385 resp.val = USER_NOTIF_MAGIC; 4386 4387 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4388 4389 /* Wait for child to finish. */ 4390 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4391 EXPECT_EQ(true, WIFEXITED(status)); 4392 EXPECT_EQ(0, WEXITSTATUS(status)); 4393 4394 close(memfd); 4395 } 4396 4397 #ifndef SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP 4398 #define SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP (1UL << 0) 4399 #define SECCOMP_IOCTL_NOTIF_SET_FLAGS SECCOMP_IOW(4, __u64) 4400 #endif 4401 4402 TEST(user_notification_sync) 4403 { 4404 struct seccomp_notif req = {}; 4405 struct seccomp_notif_resp resp = {}; 4406 int status, listener; 4407 pid_t pid; 4408 long ret; 4409 4410 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4411 ASSERT_EQ(0, ret) { 4412 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4413 } 4414 4415 listener = user_notif_syscall(__NR_getppid, 4416 SECCOMP_FILTER_FLAG_NEW_LISTENER); 4417 ASSERT_GE(listener, 0); 4418 4419 /* Try to set invalid flags. */ 4420 EXPECT_SYSCALL_RETURN(-EINVAL, 4421 ioctl(listener, SECCOMP_IOCTL_NOTIF_SET_FLAGS, 0xffffffff, 0)); 4422 4423 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SET_FLAGS, 4424 SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP, 0), 0); 4425 4426 pid = fork(); 4427 ASSERT_GE(pid, 0); 4428 if (pid == 0) { 4429 ret = syscall(__NR_getppid); 4430 ASSERT_EQ(ret, USER_NOTIF_MAGIC) { 4431 _exit(1); 4432 } 4433 _exit(0); 4434 } 4435 4436 req.pid = 0; 4437 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4438 4439 ASSERT_EQ(req.data.nr, __NR_getppid); 4440 4441 resp.id = req.id; 4442 resp.error = 0; 4443 resp.val = USER_NOTIF_MAGIC; 4444 resp.flags = 0; 4445 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4446 4447 ASSERT_EQ(waitpid(pid, &status, 0), pid); 4448 ASSERT_EQ(status, 0); 4449 } 4450 4451 4452 /* Make sure PTRACE_O_SUSPEND_SECCOMP requires CAP_SYS_ADMIN. */ 4453 FIXTURE(O_SUSPEND_SECCOMP) { 4454 pid_t pid; 4455 }; 4456 4457 FIXTURE_SETUP(O_SUSPEND_SECCOMP) 4458 { 4459 ERRNO_FILTER(block_read, E2BIG); 4460 cap_value_t cap_list[] = { CAP_SYS_ADMIN }; 4461 cap_t caps; 4462 4463 self->pid = 0; 4464 4465 /* make sure we don't have CAP_SYS_ADMIN */ 4466 caps = cap_get_proc(); 4467 ASSERT_NE(NULL, caps); 4468 ASSERT_EQ(0, cap_set_flag(caps, CAP_EFFECTIVE, 1, cap_list, CAP_CLEAR)); 4469 ASSERT_EQ(0, cap_set_proc(caps)); 4470 cap_free(caps); 4471 4472 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)); 4473 ASSERT_EQ(0, prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_block_read)); 4474 4475 self->pid = fork(); 4476 ASSERT_GE(self->pid, 0); 4477 4478 if (self->pid == 0) { 4479 while (1) 4480 pause(); 4481 _exit(127); 4482 } 4483 } 4484 4485 FIXTURE_TEARDOWN(O_SUSPEND_SECCOMP) 4486 { 4487 if (self->pid) 4488 kill(self->pid, SIGKILL); 4489 } 4490 4491 TEST_F(O_SUSPEND_SECCOMP, setoptions) 4492 { 4493 int wstatus; 4494 4495 ASSERT_EQ(0, ptrace(PTRACE_ATTACH, self->pid, NULL, 0)); 4496 ASSERT_EQ(self->pid, wait(&wstatus)); 4497 ASSERT_EQ(-1, ptrace(PTRACE_SETOPTIONS, self->pid, NULL, PTRACE_O_SUSPEND_SECCOMP)); 4498 if (errno == EINVAL) 4499 SKIP(return, "Kernel does not support PTRACE_O_SUSPEND_SECCOMP (missing CONFIG_CHECKPOINT_RESTORE?)"); 4500 ASSERT_EQ(EPERM, errno); 4501 } 4502 4503 TEST_F(O_SUSPEND_SECCOMP, seize) 4504 { 4505 int ret; 4506 4507 ret = ptrace(PTRACE_SEIZE, self->pid, NULL, PTRACE_O_SUSPEND_SECCOMP); 4508 ASSERT_EQ(-1, ret); 4509 if (errno == EINVAL) 4510 SKIP(return, "Kernel does not support PTRACE_O_SUSPEND_SECCOMP (missing CONFIG_CHECKPOINT_RESTORE?)"); 4511 ASSERT_EQ(EPERM, errno); 4512 } 4513 4514 /* 4515 * get_nth - Get the nth, space separated entry in a file. 4516 * 4517 * Returns the length of the read field. 4518 * Throws error if field is zero-lengthed. 4519 */ 4520 static ssize_t get_nth(struct __test_metadata *_metadata, const char *path, 4521 const unsigned int position, char **entry) 4522 { 4523 char *line = NULL; 4524 unsigned int i; 4525 ssize_t nread; 4526 size_t len = 0; 4527 FILE *f; 4528 4529 f = fopen(path, "r"); 4530 ASSERT_NE(f, NULL) { 4531 TH_LOG("Could not open %s: %s", path, strerror(errno)); 4532 } 4533 4534 for (i = 0; i < position; i++) { 4535 nread = getdelim(&line, &len, ' ', f); 4536 ASSERT_GE(nread, 0) { 4537 TH_LOG("Failed to read %d entry in file %s", i, path); 4538 } 4539 } 4540 fclose(f); 4541 4542 ASSERT_GT(nread, 0) { 4543 TH_LOG("Entry in file %s had zero length", path); 4544 } 4545 4546 *entry = line; 4547 return nread - 1; 4548 } 4549 4550 /* For a given PID, get the task state (D, R, etc...) */ 4551 static char get_proc_stat(struct __test_metadata *_metadata, pid_t pid) 4552 { 4553 char proc_path[100] = {0}; 4554 char status; 4555 char *line; 4556 4557 snprintf(proc_path, sizeof(proc_path), "/proc/%d/stat", pid); 4558 ASSERT_EQ(get_nth(_metadata, proc_path, 3, &line), 1); 4559 4560 status = *line; 4561 free(line); 4562 4563 return status; 4564 } 4565 4566 TEST(user_notification_fifo) 4567 { 4568 struct seccomp_notif_resp resp = {}; 4569 struct seccomp_notif req = {}; 4570 int i, status, listener; 4571 pid_t pid, pids[3]; 4572 __u64 baseid; 4573 long ret; 4574 /* 100 ms */ 4575 struct timespec delay = { .tv_nsec = 100000000 }; 4576 4577 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4578 ASSERT_EQ(0, ret) { 4579 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4580 } 4581 4582 /* Setup a listener */ 4583 listener = user_notif_syscall(__NR_getppid, 4584 SECCOMP_FILTER_FLAG_NEW_LISTENER); 4585 ASSERT_GE(listener, 0); 4586 4587 pid = fork(); 4588 ASSERT_GE(pid, 0); 4589 4590 if (pid == 0) { 4591 ret = syscall(__NR_getppid); 4592 exit(ret != USER_NOTIF_MAGIC); 4593 } 4594 4595 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4596 baseid = req.id + 1; 4597 4598 resp.id = req.id; 4599 resp.error = 0; 4600 resp.val = USER_NOTIF_MAGIC; 4601 4602 /* check that we make sure flags == 0 */ 4603 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4604 4605 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4606 EXPECT_EQ(true, WIFEXITED(status)); 4607 EXPECT_EQ(0, WEXITSTATUS(status)); 4608 4609 /* Start children, and generate notifications */ 4610 for (i = 0; i < ARRAY_SIZE(pids); i++) { 4611 pid = fork(); 4612 if (pid == 0) { 4613 ret = syscall(__NR_getppid); 4614 exit(ret != USER_NOTIF_MAGIC); 4615 } 4616 pids[i] = pid; 4617 } 4618 4619 /* This spins until all of the children are sleeping */ 4620 restart_wait: 4621 for (i = 0; i < ARRAY_SIZE(pids); i++) { 4622 if (get_proc_stat(_metadata, pids[i]) != 'S') { 4623 nanosleep(&delay, NULL); 4624 goto restart_wait; 4625 } 4626 } 4627 4628 /* Read the notifications in order (and respond) */ 4629 for (i = 0; i < ARRAY_SIZE(pids); i++) { 4630 memset(&req, 0, sizeof(req)); 4631 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4632 EXPECT_EQ(req.id, baseid + i); 4633 resp.id = req.id; 4634 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4635 } 4636 4637 /* Make sure notifications were received */ 4638 for (i = 0; i < ARRAY_SIZE(pids); i++) { 4639 EXPECT_EQ(waitpid(pids[i], &status, 0), pids[i]); 4640 EXPECT_EQ(true, WIFEXITED(status)); 4641 EXPECT_EQ(0, WEXITSTATUS(status)); 4642 } 4643 } 4644 4645 /* get_proc_syscall - Get the syscall in progress for a given pid 4646 * 4647 * Returns the current syscall number for a given process 4648 * Returns -1 if not in syscall (running or blocked) 4649 */ 4650 static long get_proc_syscall(struct __test_metadata *_metadata, int pid) 4651 { 4652 char proc_path[100] = {0}; 4653 long ret = -1; 4654 ssize_t nread; 4655 char *line; 4656 4657 snprintf(proc_path, sizeof(proc_path), "/proc/%d/syscall", pid); 4658 nread = get_nth(_metadata, proc_path, 1, &line); 4659 ASSERT_GT(nread, 0); 4660 4661 if (!strncmp("running", line, MIN(7, nread))) 4662 ret = strtol(line, NULL, 16); 4663 4664 free(line); 4665 return ret; 4666 } 4667 4668 /* Ensure non-fatal signals prior to receive are unmodified */ 4669 TEST(user_notification_wait_killable_pre_notification) 4670 { 4671 struct sigaction new_action = { 4672 .sa_handler = signal_handler, 4673 }; 4674 int listener, status, sk_pair[2]; 4675 pid_t pid; 4676 long ret; 4677 char c; 4678 /* 100 ms */ 4679 struct timespec delay = { .tv_nsec = 100000000 }; 4680 4681 ASSERT_EQ(sigemptyset(&new_action.sa_mask), 0); 4682 4683 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4684 ASSERT_EQ(0, ret) 4685 { 4686 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4687 } 4688 4689 ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0); 4690 4691 listener = user_notif_syscall( 4692 __NR_getppid, SECCOMP_FILTER_FLAG_NEW_LISTENER | 4693 SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV); 4694 ASSERT_GE(listener, 0); 4695 4696 /* 4697 * Check that we can kill the process with SIGUSR1 prior to receiving 4698 * the notification. SIGUSR1 is wired up to a custom signal handler, 4699 * and make sure it gets called. 4700 */ 4701 pid = fork(); 4702 ASSERT_GE(pid, 0); 4703 4704 if (pid == 0) { 4705 close(sk_pair[0]); 4706 handled = sk_pair[1]; 4707 4708 /* Setup the non-fatal sigaction without SA_RESTART */ 4709 if (sigaction(SIGUSR1, &new_action, NULL)) { 4710 perror("sigaction"); 4711 exit(1); 4712 } 4713 4714 ret = syscall(__NR_getppid); 4715 /* Make sure we got a return from a signal interruption */ 4716 exit(ret != -1 || errno != EINTR); 4717 } 4718 4719 /* 4720 * Make sure we've gotten to the seccomp user notification wait 4721 * from getppid prior to sending any signals 4722 */ 4723 while (get_proc_syscall(_metadata, pid) != __NR_getppid && 4724 get_proc_stat(_metadata, pid) != 'S') 4725 nanosleep(&delay, NULL); 4726 4727 /* Send non-fatal kill signal */ 4728 EXPECT_EQ(kill(pid, SIGUSR1), 0); 4729 4730 /* wait for process to exit (exit checks for EINTR) */ 4731 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4732 EXPECT_EQ(true, WIFEXITED(status)); 4733 EXPECT_EQ(0, WEXITSTATUS(status)); 4734 4735 EXPECT_EQ(read(sk_pair[0], &c, 1), 1); 4736 } 4737 4738 /* Ensure non-fatal signals after receive are blocked */ 4739 TEST(user_notification_wait_killable) 4740 { 4741 struct sigaction new_action = { 4742 .sa_handler = signal_handler, 4743 }; 4744 struct seccomp_notif_resp resp = {}; 4745 struct seccomp_notif req = {}; 4746 int listener, status, sk_pair[2]; 4747 pid_t pid; 4748 long ret; 4749 char c; 4750 /* 100 ms */ 4751 struct timespec delay = { .tv_nsec = 100000000 }; 4752 4753 ASSERT_EQ(sigemptyset(&new_action.sa_mask), 0); 4754 4755 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4756 ASSERT_EQ(0, ret) 4757 { 4758 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4759 } 4760 4761 ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0); 4762 4763 listener = user_notif_syscall( 4764 __NR_getppid, SECCOMP_FILTER_FLAG_NEW_LISTENER | 4765 SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV); 4766 ASSERT_GE(listener, 0); 4767 4768 pid = fork(); 4769 ASSERT_GE(pid, 0); 4770 4771 if (pid == 0) { 4772 close(sk_pair[0]); 4773 handled = sk_pair[1]; 4774 4775 /* Setup the sigaction without SA_RESTART */ 4776 if (sigaction(SIGUSR1, &new_action, NULL)) { 4777 perror("sigaction"); 4778 exit(1); 4779 } 4780 4781 /* Make sure that the syscall is completed (no EINTR) */ 4782 ret = syscall(__NR_getppid); 4783 exit(ret != USER_NOTIF_MAGIC); 4784 } 4785 4786 /* 4787 * Get the notification, to make move the notifying process into a 4788 * non-preemptible (TASK_KILLABLE) state. 4789 */ 4790 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4791 /* Send non-fatal kill signal */ 4792 EXPECT_EQ(kill(pid, SIGUSR1), 0); 4793 4794 /* 4795 * Make sure the task enters moves to TASK_KILLABLE by waiting for 4796 * D (Disk Sleep) state after receiving non-fatal signal. 4797 */ 4798 while (get_proc_stat(_metadata, pid) != 'D') 4799 nanosleep(&delay, NULL); 4800 4801 resp.id = req.id; 4802 resp.val = USER_NOTIF_MAGIC; 4803 /* Make sure the notification is found and able to be replied to */ 4804 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4805 4806 /* 4807 * Make sure that the signal handler does get called once we're back in 4808 * userspace. 4809 */ 4810 EXPECT_EQ(read(sk_pair[0], &c, 1), 1); 4811 /* wait for process to exit (exit checks for USER_NOTIF_MAGIC) */ 4812 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4813 EXPECT_EQ(true, WIFEXITED(status)); 4814 EXPECT_EQ(0, WEXITSTATUS(status)); 4815 } 4816 4817 /* Ensure fatal signals after receive are not blocked */ 4818 TEST(user_notification_wait_killable_fatal) 4819 { 4820 struct seccomp_notif req = {}; 4821 int listener, status; 4822 pid_t pid; 4823 long ret; 4824 /* 100 ms */ 4825 struct timespec delay = { .tv_nsec = 100000000 }; 4826 4827 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4828 ASSERT_EQ(0, ret) 4829 { 4830 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4831 } 4832 4833 listener = user_notif_syscall( 4834 __NR_getppid, SECCOMP_FILTER_FLAG_NEW_LISTENER | 4835 SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV); 4836 ASSERT_GE(listener, 0); 4837 4838 pid = fork(); 4839 ASSERT_GE(pid, 0); 4840 4841 if (pid == 0) { 4842 /* This should never complete as it should get a SIGTERM */ 4843 syscall(__NR_getppid); 4844 exit(1); 4845 } 4846 4847 while (get_proc_stat(_metadata, pid) != 'S') 4848 nanosleep(&delay, NULL); 4849 4850 /* 4851 * Get the notification, to make move the notifying process into a 4852 * non-preemptible (TASK_KILLABLE) state. 4853 */ 4854 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4855 /* Kill the process with a fatal signal */ 4856 EXPECT_EQ(kill(pid, SIGTERM), 0); 4857 4858 /* 4859 * Wait for the process to exit, and make sure the process terminated 4860 * due to the SIGTERM signal. 4861 */ 4862 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4863 EXPECT_EQ(true, WIFSIGNALED(status)); 4864 EXPECT_EQ(SIGTERM, WTERMSIG(status)); 4865 } 4866 4867 /* Ensure signals after the reply do not interrupt */ 4868 TEST(user_notification_wait_killable_after_reply) 4869 { 4870 int i, max_iter = 100000; 4871 int listener, status; 4872 int pipe_fds[2]; 4873 pid_t pid; 4874 long ret; 4875 4876 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4877 ASSERT_EQ(0, ret) 4878 { 4879 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4880 } 4881 4882 listener = user_notif_syscall( 4883 __NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER | 4884 SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV); 4885 ASSERT_GE(listener, 0); 4886 4887 /* 4888 * Used to count invocations. One token is transferred from the child 4889 * to the parent per syscall invocation, the parent tries to take 4890 * one token per successful RECV. If the syscall is restarted after 4891 * RECV the parent will try to get two tokens while the child only 4892 * provided one. 4893 */ 4894 ASSERT_EQ(pipe(pipe_fds), 0); 4895 4896 pid = fork(); 4897 ASSERT_GE(pid, 0); 4898 4899 if (pid == 0) { 4900 struct sigaction new_action = { 4901 .sa_handler = signal_handler_nop, 4902 .sa_flags = SA_RESTART, 4903 }; 4904 struct itimerval timer = { 4905 .it_value = { .tv_usec = 1000 }, 4906 .it_interval = { .tv_usec = 1000 }, 4907 }; 4908 char c = 'a'; 4909 4910 close(pipe_fds[0]); 4911 4912 /* Setup the sigaction with SA_RESTART */ 4913 if (sigaction(SIGALRM, &new_action, NULL)) { 4914 perror("sigaction"); 4915 exit(1); 4916 } 4917 4918 /* 4919 * Kill with SIGALRM repeatedly, to try to hit the race when 4920 * handling the syscall. 4921 */ 4922 if (setitimer(ITIMER_REAL, &timer, NULL) < 0) 4923 perror("setitimer"); 4924 4925 for (i = 0; i < max_iter; ++i) { 4926 int fd; 4927 4928 /* Send one token per iteration to catch repeats. */ 4929 if (write(pipe_fds[1], &c, sizeof(c)) != 1) { 4930 perror("write"); 4931 exit(1); 4932 } 4933 4934 fd = syscall(__NR_dup, 0); 4935 if (fd < 0) { 4936 perror("dup"); 4937 exit(1); 4938 } 4939 close(fd); 4940 } 4941 4942 exit(0); 4943 } 4944 4945 close(pipe_fds[1]); 4946 4947 for (i = 0; i < max_iter; ++i) { 4948 struct seccomp_notif req = {}; 4949 struct seccomp_notif_addfd addfd = {}; 4950 struct pollfd pfd = { 4951 .fd = pipe_fds[0], 4952 .events = POLLIN, 4953 }; 4954 char c; 4955 4956 /* 4957 * Try to receive one token. If it failed, one child syscall 4958 * was restarted after RECV and needed to be handled twice. 4959 */ 4960 ASSERT_EQ(poll(&pfd, 1, 1000), 1) 4961 kill(pid, SIGKILL); 4962 4963 ASSERT_EQ(read(pipe_fds[0], &c, sizeof(c)), 1) 4964 kill(pid, SIGKILL); 4965 4966 /* 4967 * Get the notification, reply to it as fast as possible to test 4968 * whether the child wrongly skips going into the non-preemptible 4969 * (TASK_KILLABLE) state. 4970 */ 4971 do 4972 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req); 4973 while (ret < 0 && errno == ENOENT); /* Accept interruptions before RECV */ 4974 ASSERT_EQ(ret, 0) 4975 kill(pid, SIGKILL); 4976 4977 addfd.id = req.id; 4978 addfd.flags = SECCOMP_ADDFD_FLAG_SEND; 4979 addfd.srcfd = 0; 4980 ASSERT_GE(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), 0) 4981 kill(pid, SIGKILL); 4982 } 4983 4984 /* 4985 * Wait for the process to exit, and make sure the process terminated 4986 * with a zero exit code.. 4987 */ 4988 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4989 EXPECT_EQ(true, WIFEXITED(status)); 4990 EXPECT_EQ(0, WEXITSTATUS(status)); 4991 } 4992 4993 struct tsync_vs_thread_leader_args { 4994 pthread_t leader; 4995 }; 4996 4997 static void *tsync_vs_dead_thread_leader_sibling(void *_args) 4998 { 4999 struct sock_filter allow_filter[] = { 5000 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 5001 }; 5002 struct sock_fprog allow_prog = { 5003 .len = (unsigned short)ARRAY_SIZE(allow_filter), 5004 .filter = allow_filter, 5005 }; 5006 struct tsync_vs_thread_leader_args *args = _args; 5007 void *retval; 5008 long ret; 5009 5010 ret = pthread_join(args->leader, &retval); 5011 if (ret) 5012 exit(1); 5013 if (retval != _args) 5014 exit(2); 5015 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, &allow_prog); 5016 if (ret) 5017 exit(3); 5018 5019 exit(0); 5020 } 5021 5022 /* 5023 * Ensure that a dead thread leader doesn't prevent installing new filters with 5024 * SECCOMP_FILTER_FLAG_TSYNC from other threads. 5025 */ 5026 TEST(tsync_vs_dead_thread_leader) 5027 { 5028 int status; 5029 pid_t pid; 5030 long ret; 5031 5032 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 5033 ASSERT_EQ(0, ret) { 5034 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 5035 } 5036 5037 pid = fork(); 5038 ASSERT_GE(pid, 0); 5039 5040 if (pid == 0) { 5041 struct sock_filter allow_filter[] = { 5042 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 5043 }; 5044 struct sock_fprog allow_prog = { 5045 .len = (unsigned short)ARRAY_SIZE(allow_filter), 5046 .filter = allow_filter, 5047 }; 5048 struct tsync_vs_thread_leader_args *args; 5049 pthread_t sibling; 5050 5051 args = malloc(sizeof(*args)); 5052 ASSERT_NE(NULL, args); 5053 args->leader = pthread_self(); 5054 5055 ret = pthread_create(&sibling, NULL, 5056 tsync_vs_dead_thread_leader_sibling, args); 5057 ASSERT_EQ(0, ret); 5058 5059 /* Install a new filter just to the leader thread. */ 5060 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog); 5061 ASSERT_EQ(0, ret); 5062 pthread_exit(args); 5063 exit(1); 5064 } 5065 5066 EXPECT_EQ(pid, waitpid(pid, &status, 0)); 5067 EXPECT_EQ(0, status); 5068 } 5069 5070 #ifdef __x86_64__ 5071 5072 /* 5073 * We need naked probed_uprobe function. Using __nocf_check 5074 * check to skip possible endbr64 instruction and ignoring 5075 * -Wattributes, otherwise the compilation might fail. 5076 */ 5077 #pragma GCC diagnostic push 5078 #pragma GCC diagnostic ignored "-Wattributes" 5079 5080 __naked __nocf_check noinline int probed_uprobe(void) 5081 { 5082 /* 5083 * Optimized uprobe is possible only on top of nop5 instruction. 5084 */ 5085 asm volatile (" \n" 5086 ".byte 0x0f, 0x1f, 0x44, 0x00, 0x00 \n" 5087 "ret \n" 5088 ); 5089 } 5090 #pragma GCC diagnostic pop 5091 5092 #else 5093 noinline int probed_uprobe(void) 5094 { 5095 return 1; 5096 } 5097 #endif 5098 5099 noinline int probed_uretprobe(void) 5100 { 5101 return 1; 5102 } 5103 5104 static int parse_uint_from_file(const char *file, const char *fmt) 5105 { 5106 int err = -1, ret; 5107 FILE *f; 5108 5109 f = fopen(file, "re"); 5110 if (f) { 5111 err = fscanf(f, fmt, &ret); 5112 fclose(f); 5113 } 5114 return err == 1 ? ret : err; 5115 } 5116 5117 static int determine_uprobe_perf_type(void) 5118 { 5119 const char *file = "/sys/bus/event_source/devices/uprobe/type"; 5120 5121 return parse_uint_from_file(file, "%d\n"); 5122 } 5123 5124 static int determine_uprobe_retprobe_bit(void) 5125 { 5126 const char *file = "/sys/bus/event_source/devices/uprobe/format/retprobe"; 5127 5128 return parse_uint_from_file(file, "config:%d\n"); 5129 } 5130 5131 static ssize_t get_uprobe_offset(const void *addr) 5132 { 5133 size_t start, base, end; 5134 bool found = false; 5135 char buf[256]; 5136 FILE *f; 5137 5138 f = fopen("/proc/self/maps", "r"); 5139 if (!f) 5140 return -1; 5141 5142 while (fscanf(f, "%zx-%zx %s %zx %*[^\n]\n", &start, &end, buf, &base) == 4) { 5143 if (buf[2] == 'x' && (uintptr_t)addr >= start && (uintptr_t)addr < end) { 5144 found = true; 5145 break; 5146 } 5147 } 5148 fclose(f); 5149 return found ? (uintptr_t)addr - start + base : -1; 5150 } 5151 5152 FIXTURE(UPROBE) { 5153 int fd; 5154 }; 5155 5156 FIXTURE_VARIANT(UPROBE) { 5157 /* 5158 * All of the U(RET)PROBE behaviors can be tested with either 5159 * u(ret)probe attached or not 5160 */ 5161 bool attach; 5162 /* 5163 * Test both uprobe and uretprobe. 5164 */ 5165 bool uretprobe; 5166 }; 5167 5168 FIXTURE_VARIANT_ADD(UPROBE, not_attached) { 5169 .attach = false, 5170 .uretprobe = false, 5171 }; 5172 5173 FIXTURE_VARIANT_ADD(UPROBE, uprobe_attached) { 5174 .attach = true, 5175 .uretprobe = false, 5176 }; 5177 5178 FIXTURE_VARIANT_ADD(UPROBE, uretprobe_attached) { 5179 .attach = true, 5180 .uretprobe = true, 5181 }; 5182 5183 FIXTURE_SETUP(UPROBE) 5184 { 5185 const size_t attr_sz = sizeof(struct perf_event_attr); 5186 struct perf_event_attr attr; 5187 ssize_t offset; 5188 int type, bit; 5189 5190 #if !defined(__NR_uprobe) || !defined(__NR_uretprobe) 5191 SKIP(return, "__NR_uprobe ot __NR_uretprobe syscalls not defined"); 5192 #endif 5193 5194 if (!variant->attach) 5195 return; 5196 5197 memset(&attr, 0, attr_sz); 5198 5199 type = determine_uprobe_perf_type(); 5200 ASSERT_GE(type, 0); 5201 5202 if (variant->uretprobe) { 5203 bit = determine_uprobe_retprobe_bit(); 5204 ASSERT_GE(bit, 0); 5205 } 5206 5207 offset = get_uprobe_offset(variant->uretprobe ? (void *)probed_uretprobe 5208 : (void *)probed_uprobe); 5209 ASSERT_GE(offset, 0); 5210 5211 if (variant->uretprobe) 5212 attr.config |= 1 << bit; 5213 attr.size = attr_sz; 5214 attr.type = type; 5215 attr.config1 = ptr_to_u64("/proc/self/exe"); 5216 attr.config2 = offset; 5217 5218 self->fd = syscall(__NR_perf_event_open, &attr, 5219 getpid() /* pid */, -1 /* cpu */, -1 /* group_fd */, 5220 PERF_FLAG_FD_CLOEXEC); 5221 } 5222 5223 FIXTURE_TEARDOWN(UPROBE) 5224 { 5225 /* we could call close(self->fd), but we'd need extra filter for 5226 * that and since we are calling _exit right away.. 5227 */ 5228 } 5229 5230 static int run_probed_with_filter(struct sock_fprog *prog) 5231 { 5232 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) || 5233 seccomp(SECCOMP_SET_MODE_FILTER, 0, prog)) { 5234 return -1; 5235 } 5236 5237 /* 5238 * Uprobe is optimized after first hit, so let's hit twice. 5239 */ 5240 probed_uprobe(); 5241 probed_uprobe(); 5242 5243 probed_uretprobe(); 5244 return 0; 5245 } 5246 5247 TEST_F(UPROBE, uprobe_default_allow) 5248 { 5249 struct sock_filter filter[] = { 5250 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 5251 }; 5252 struct sock_fprog prog = { 5253 .len = (unsigned short)ARRAY_SIZE(filter), 5254 .filter = filter, 5255 }; 5256 5257 ASSERT_EQ(0, run_probed_with_filter(&prog)); 5258 } 5259 5260 TEST_F(UPROBE, uprobe_default_block) 5261 { 5262 struct sock_filter filter[] = { 5263 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 5264 offsetof(struct seccomp_data, nr)), 5265 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit_group, 1, 0), 5266 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 5267 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 5268 }; 5269 struct sock_fprog prog = { 5270 .len = (unsigned short)ARRAY_SIZE(filter), 5271 .filter = filter, 5272 }; 5273 5274 ASSERT_EQ(0, run_probed_with_filter(&prog)); 5275 } 5276 5277 TEST_F(UPROBE, uprobe_block_syscall) 5278 { 5279 struct sock_filter filter[] = { 5280 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 5281 offsetof(struct seccomp_data, nr)), 5282 #ifdef __NR_uprobe 5283 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_uprobe, 1, 2), 5284 #endif 5285 #ifdef __NR_uretprobe 5286 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_uretprobe, 0, 1), 5287 #endif 5288 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 5289 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 5290 }; 5291 struct sock_fprog prog = { 5292 .len = (unsigned short)ARRAY_SIZE(filter), 5293 .filter = filter, 5294 }; 5295 5296 ASSERT_EQ(0, run_probed_with_filter(&prog)); 5297 } 5298 5299 TEST_F(UPROBE, uprobe_default_block_with_syscall) 5300 { 5301 struct sock_filter filter[] = { 5302 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 5303 offsetof(struct seccomp_data, nr)), 5304 #ifdef __NR_uprobe 5305 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_uprobe, 3, 0), 5306 #endif 5307 #ifdef __NR_uretprobe 5308 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_uretprobe, 2, 0), 5309 #endif 5310 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit_group, 1, 0), 5311 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 5312 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 5313 }; 5314 struct sock_fprog prog = { 5315 .len = (unsigned short)ARRAY_SIZE(filter), 5316 .filter = filter, 5317 }; 5318 5319 ASSERT_EQ(0, run_probed_with_filter(&prog)); 5320 } 5321 5322 /* 5323 * TODO: 5324 * - expand NNP testing 5325 * - better arch-specific TRACE and TRAP handlers. 5326 * - endianness checking when appropriate 5327 * - 64-bit arg prodding 5328 * - arch value testing (x86 modes especially) 5329 * - verify that FILTER_FLAG_LOG filters generate log messages 5330 * - verify that RET_LOG generates log messages 5331 */ 5332 5333 TEST_HARNESS_MAIN 5334