1// SPDX-License-Identifier: GPL-2.0-only 2/* 3 * Copyright (c) 2012 The Chromium OS Authors. All rights reserved. 4 * 5 * Test code for seccomp bpf. 6 */ 7 8#define _GNU_SOURCE 9#include <sys/types.h> 10 11/* 12 * glibc 2.26 and later have SIGSYS in siginfo_t. Before that, 13 * we need to use the kernel's siginfo.h file and trick glibc 14 * into accepting it. 15 */ 16#if !__GLIBC_PREREQ(2, 26) 17# include <asm/siginfo.h> 18# define __have_siginfo_t 1 19# define __have_sigval_t 1 20# define __have_sigevent_t 1 21#endif 22 23#include <errno.h> 24#include <linux/filter.h> 25#include <sys/prctl.h> 26#include <sys/ptrace.h> 27#include <sys/user.h> 28#include <linux/prctl.h> 29#include <linux/ptrace.h> 30#include <linux/seccomp.h> 31#include <pthread.h> 32#include <semaphore.h> 33#include <signal.h> 34#include <stddef.h> 35#include <stdbool.h> 36#include <string.h> 37#include <time.h> 38#include <limits.h> 39#include <linux/elf.h> 40#include <sys/uio.h> 41#include <sys/utsname.h> 42#include <sys/fcntl.h> 43#include <sys/mman.h> 44#include <sys/times.h> 45#include <sys/socket.h> 46#include <sys/ioctl.h> 47#include <linux/kcmp.h> 48#include <sys/resource.h> 49 50#include <unistd.h> 51#include <sys/syscall.h> 52#include <poll.h> 53 54#include "../kselftest_harness.h" 55#include "../clone3/clone3_selftests.h" 56 57/* Attempt to de-conflict with the selftests tree. */ 58#ifndef SKIP 59#define SKIP(s, ...) XFAIL(s, ##__VA_ARGS__) 60#endif 61 62#ifndef PR_SET_PTRACER 63# define PR_SET_PTRACER 0x59616d61 64#endif 65 66#ifndef PR_SET_NO_NEW_PRIVS 67#define PR_SET_NO_NEW_PRIVS 38 68#define PR_GET_NO_NEW_PRIVS 39 69#endif 70 71#ifndef PR_SECCOMP_EXT 72#define PR_SECCOMP_EXT 43 73#endif 74 75#ifndef SECCOMP_EXT_ACT 76#define SECCOMP_EXT_ACT 1 77#endif 78 79#ifndef SECCOMP_EXT_ACT_TSYNC 80#define SECCOMP_EXT_ACT_TSYNC 1 81#endif 82 83#ifndef SECCOMP_MODE_STRICT 84#define SECCOMP_MODE_STRICT 1 85#endif 86 87#ifndef SECCOMP_MODE_FILTER 88#define SECCOMP_MODE_FILTER 2 89#endif 90 91#ifndef SECCOMP_RET_ALLOW 92struct seccomp_data { 93 int nr; 94 __u32 arch; 95 __u64 instruction_pointer; 96 __u64 args[6]; 97}; 98#endif 99 100#ifndef SECCOMP_RET_KILL_PROCESS 101#define SECCOMP_RET_KILL_PROCESS 0x80000000U /* kill the process */ 102#define SECCOMP_RET_KILL_THREAD 0x00000000U /* kill the thread */ 103#endif 104#ifndef SECCOMP_RET_KILL 105#define SECCOMP_RET_KILL SECCOMP_RET_KILL_THREAD 106#define SECCOMP_RET_TRAP 0x00030000U /* disallow and force a SIGSYS */ 107#define SECCOMP_RET_ERRNO 0x00050000U /* returns an errno */ 108#define SECCOMP_RET_TRACE 0x7ff00000U /* pass to a tracer or disallow */ 109#define SECCOMP_RET_ALLOW 0x7fff0000U /* allow */ 110#endif 111#ifndef SECCOMP_RET_LOG 112#define SECCOMP_RET_LOG 0x7ffc0000U /* allow after logging */ 113#endif 114 115#ifndef __NR_seccomp 116# if defined(__i386__) 117# define __NR_seccomp 354 118# elif defined(__x86_64__) 119# define __NR_seccomp 317 120# elif defined(__arm__) 121# define __NR_seccomp 383 122# elif defined(__aarch64__) 123# define __NR_seccomp 277 124# elif defined(__riscv) 125# define __NR_seccomp 277 126# elif defined(__csky__) 127# define __NR_seccomp 277 128# elif defined(__loongarch__) 129# define __NR_seccomp 277 130# elif defined(__hppa__) 131# define __NR_seccomp 338 132# elif defined(__powerpc__) 133# define __NR_seccomp 358 134# elif defined(__s390__) 135# define __NR_seccomp 348 136# elif defined(__xtensa__) 137# define __NR_seccomp 337 138# elif defined(__sh__) 139# define __NR_seccomp 372 140# else 141# warning "seccomp syscall number unknown for this architecture" 142# define __NR_seccomp 0xffff 143# endif 144#endif 145 146#ifndef SECCOMP_SET_MODE_STRICT 147#define SECCOMP_SET_MODE_STRICT 0 148#endif 149 150#ifndef SECCOMP_SET_MODE_FILTER 151#define SECCOMP_SET_MODE_FILTER 1 152#endif 153 154#ifndef SECCOMP_GET_ACTION_AVAIL 155#define SECCOMP_GET_ACTION_AVAIL 2 156#endif 157 158#ifndef SECCOMP_GET_NOTIF_SIZES 159#define SECCOMP_GET_NOTIF_SIZES 3 160#endif 161 162#ifndef SECCOMP_FILTER_FLAG_TSYNC 163#define SECCOMP_FILTER_FLAG_TSYNC (1UL << 0) 164#endif 165 166#ifndef SECCOMP_FILTER_FLAG_LOG 167#define SECCOMP_FILTER_FLAG_LOG (1UL << 1) 168#endif 169 170#ifndef SECCOMP_FILTER_FLAG_SPEC_ALLOW 171#define SECCOMP_FILTER_FLAG_SPEC_ALLOW (1UL << 2) 172#endif 173 174#ifndef PTRACE_SECCOMP_GET_METADATA 175#define PTRACE_SECCOMP_GET_METADATA 0x420d 176 177struct seccomp_metadata { 178 __u64 filter_off; /* Input: which filter */ 179 __u64 flags; /* Output: filter's flags */ 180}; 181#endif 182 183#ifndef SECCOMP_FILTER_FLAG_NEW_LISTENER 184#define SECCOMP_FILTER_FLAG_NEW_LISTENER (1UL << 3) 185#endif 186 187#ifndef SECCOMP_RET_USER_NOTIF 188#define SECCOMP_RET_USER_NOTIF 0x7fc00000U 189 190#define SECCOMP_IOC_MAGIC '!' 191#define SECCOMP_IO(nr) _IO(SECCOMP_IOC_MAGIC, nr) 192#define SECCOMP_IOR(nr, type) _IOR(SECCOMP_IOC_MAGIC, nr, type) 193#define SECCOMP_IOW(nr, type) _IOW(SECCOMP_IOC_MAGIC, nr, type) 194#define SECCOMP_IOWR(nr, type) _IOWR(SECCOMP_IOC_MAGIC, nr, type) 195 196/* Flags for seccomp notification fd ioctl. */ 197#define SECCOMP_IOCTL_NOTIF_RECV SECCOMP_IOWR(0, struct seccomp_notif) 198#define SECCOMP_IOCTL_NOTIF_SEND SECCOMP_IOWR(1, \ 199 struct seccomp_notif_resp) 200#define SECCOMP_IOCTL_NOTIF_ID_VALID SECCOMP_IOW(2, __u64) 201 202struct seccomp_notif { 203 __u64 id; 204 __u32 pid; 205 __u32 flags; 206 struct seccomp_data data; 207}; 208 209struct seccomp_notif_resp { 210 __u64 id; 211 __s64 val; 212 __s32 error; 213 __u32 flags; 214}; 215 216struct seccomp_notif_sizes { 217 __u16 seccomp_notif; 218 __u16 seccomp_notif_resp; 219 __u16 seccomp_data; 220}; 221#endif 222 223#ifndef SECCOMP_IOCTL_NOTIF_ADDFD 224/* On success, the return value is the remote process's added fd number */ 225#define SECCOMP_IOCTL_NOTIF_ADDFD SECCOMP_IOW(3, \ 226 struct seccomp_notif_addfd) 227 228/* valid flags for seccomp_notif_addfd */ 229#define SECCOMP_ADDFD_FLAG_SETFD (1UL << 0) /* Specify remote fd */ 230 231struct seccomp_notif_addfd { 232 __u64 id; 233 __u32 flags; 234 __u32 srcfd; 235 __u32 newfd; 236 __u32 newfd_flags; 237}; 238#endif 239 240struct seccomp_notif_addfd_small { 241 __u64 id; 242 char weird[4]; 243}; 244#define SECCOMP_IOCTL_NOTIF_ADDFD_SMALL \ 245 SECCOMP_IOW(3, struct seccomp_notif_addfd_small) 246 247struct seccomp_notif_addfd_big { 248 union { 249 struct seccomp_notif_addfd addfd; 250 char buf[sizeof(struct seccomp_notif_addfd) + 8]; 251 }; 252}; 253#define SECCOMP_IOCTL_NOTIF_ADDFD_BIG \ 254 SECCOMP_IOWR(3, struct seccomp_notif_addfd_big) 255 256#ifndef PTRACE_EVENTMSG_SYSCALL_ENTRY 257#define PTRACE_EVENTMSG_SYSCALL_ENTRY 1 258#define PTRACE_EVENTMSG_SYSCALL_EXIT 2 259#endif 260 261#ifndef SECCOMP_USER_NOTIF_FLAG_CONTINUE 262#define SECCOMP_USER_NOTIF_FLAG_CONTINUE 0x00000001 263#endif 264 265#ifndef SECCOMP_FILTER_FLAG_TSYNC_ESRCH 266#define SECCOMP_FILTER_FLAG_TSYNC_ESRCH (1UL << 4) 267#endif 268 269#ifndef seccomp 270int seccomp(unsigned int op, unsigned int flags, void *args) 271{ 272 errno = 0; 273 return syscall(__NR_seccomp, op, flags, args); 274} 275#endif 276 277#if __BYTE_ORDER == __LITTLE_ENDIAN 278#define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n])) 279#elif __BYTE_ORDER == __BIG_ENDIAN 280#define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32)) 281#else 282#error "wut? Unknown __BYTE_ORDER?!" 283#endif 284 285#define SIBLING_EXIT_UNKILLED 0xbadbeef 286#define SIBLING_EXIT_FAILURE 0xbadface 287#define SIBLING_EXIT_NEWPRIVS 0xbadfeed 288 289static int __filecmp(pid_t pid1, pid_t pid2, int fd1, int fd2) 290{ 291#ifdef __NR_kcmp 292 errno = 0; 293 return syscall(__NR_kcmp, pid1, pid2, KCMP_FILE, fd1, fd2); 294#else 295 errno = ENOSYS; 296 return -1; 297#endif 298} 299 300/* Have TH_LOG report actual location filecmp() is used. */ 301#define filecmp(pid1, pid2, fd1, fd2) ({ \ 302 int _ret; \ 303 \ 304 _ret = __filecmp(pid1, pid2, fd1, fd2); \ 305 if (_ret != 0) { \ 306 if (_ret < 0 && errno == ENOSYS) { \ 307 TH_LOG("kcmp() syscall missing (test is less accurate)");\ 308 _ret = 0; \ 309 } \ 310 } \ 311 _ret; }) 312 313TEST(kcmp) 314{ 315 int ret; 316 317 ret = __filecmp(getpid(), getpid(), 1, 1); 318 EXPECT_EQ(ret, 0); 319 if (ret != 0 && errno == ENOSYS) 320 SKIP(return, "Kernel does not support kcmp() (missing CONFIG_KCMP?)"); 321} 322 323TEST(mode_strict_support) 324{ 325 long ret; 326 327 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL); 328 ASSERT_EQ(0, ret) { 329 TH_LOG("Kernel does not support CONFIG_SECCOMP"); 330 } 331 syscall(__NR_exit, 0); 332} 333 334TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL) 335{ 336 long ret; 337 338 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL); 339 ASSERT_EQ(0, ret) { 340 TH_LOG("Kernel does not support CONFIG_SECCOMP"); 341 } 342 syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER, 343 NULL, NULL, NULL); 344 EXPECT_FALSE(true) { 345 TH_LOG("Unreachable!"); 346 } 347} 348 349/* Note! This doesn't test no new privs behavior */ 350TEST(no_new_privs_support) 351{ 352 long ret; 353 354 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 355 EXPECT_EQ(0, ret) { 356 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 357 } 358} 359 360/* Tests kernel support by checking for a copy_from_user() fault on NULL. */ 361TEST(mode_filter_support) 362{ 363 long ret; 364 365 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 366 ASSERT_EQ(0, ret) { 367 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 368 } 369 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL); 370 EXPECT_EQ(-1, ret); 371 EXPECT_EQ(EFAULT, errno) { 372 TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!"); 373 } 374} 375 376TEST(mode_filter_without_nnp) 377{ 378 struct sock_filter filter[] = { 379 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 380 }; 381 struct sock_fprog prog = { 382 .len = (unsigned short)ARRAY_SIZE(filter), 383 .filter = filter, 384 }; 385 long ret; 386 387 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0); 388 ASSERT_LE(0, ret) { 389 TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS"); 390 } 391 errno = 0; 392 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 393 /* Succeeds with CAP_SYS_ADMIN, fails without */ 394 /* TODO(wad) check caps not euid */ 395 if (geteuid()) { 396 EXPECT_EQ(-1, ret); 397 EXPECT_EQ(EACCES, errno); 398 } else { 399 EXPECT_EQ(0, ret); 400 } 401} 402 403#define MAX_INSNS_PER_PATH 32768 404 405TEST(filter_size_limits) 406{ 407 int i; 408 int count = BPF_MAXINSNS + 1; 409 struct sock_filter allow[] = { 410 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 411 }; 412 struct sock_filter *filter; 413 struct sock_fprog prog = { }; 414 long ret; 415 416 filter = calloc(count, sizeof(*filter)); 417 ASSERT_NE(NULL, filter); 418 419 for (i = 0; i < count; i++) 420 filter[i] = allow[0]; 421 422 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 423 ASSERT_EQ(0, ret); 424 425 prog.filter = filter; 426 prog.len = count; 427 428 /* Too many filter instructions in a single filter. */ 429 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 430 ASSERT_NE(0, ret) { 431 TH_LOG("Installing %d insn filter was allowed", prog.len); 432 } 433 434 /* One less is okay, though. */ 435 prog.len -= 1; 436 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 437 ASSERT_EQ(0, ret) { 438 TH_LOG("Installing %d insn filter wasn't allowed", prog.len); 439 } 440} 441 442TEST(filter_chain_limits) 443{ 444 int i; 445 int count = BPF_MAXINSNS; 446 struct sock_filter allow[] = { 447 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 448 }; 449 struct sock_filter *filter; 450 struct sock_fprog prog = { }; 451 long ret; 452 453 filter = calloc(count, sizeof(*filter)); 454 ASSERT_NE(NULL, filter); 455 456 for (i = 0; i < count; i++) 457 filter[i] = allow[0]; 458 459 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 460 ASSERT_EQ(0, ret); 461 462 prog.filter = filter; 463 prog.len = 1; 464 465 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 466 ASSERT_EQ(0, ret); 467 468 prog.len = count; 469 470 /* Too many total filter instructions. */ 471 for (i = 0; i < MAX_INSNS_PER_PATH; i++) { 472 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 473 if (ret != 0) 474 break; 475 } 476 ASSERT_NE(0, ret) { 477 TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)", 478 i, count, i * (count + 4)); 479 } 480} 481 482TEST(mode_filter_cannot_move_to_strict) 483{ 484 struct sock_filter filter[] = { 485 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 486 }; 487 struct sock_fprog prog = { 488 .len = (unsigned short)ARRAY_SIZE(filter), 489 .filter = filter, 490 }; 491 long ret; 492 493 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 494 ASSERT_EQ(0, ret); 495 496 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 497 ASSERT_EQ(0, ret); 498 499 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0); 500 EXPECT_EQ(-1, ret); 501 EXPECT_EQ(EINVAL, errno); 502} 503 504 505TEST(mode_filter_get_seccomp) 506{ 507 struct sock_filter filter[] = { 508 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 509 }; 510 struct sock_fprog prog = { 511 .len = (unsigned short)ARRAY_SIZE(filter), 512 .filter = filter, 513 }; 514 long ret; 515 516 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 517 ASSERT_EQ(0, ret); 518 519 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0); 520 EXPECT_EQ(0, ret); 521 522 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 523 ASSERT_EQ(0, ret); 524 525 ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0); 526 EXPECT_EQ(2, ret); 527} 528 529 530TEST(ALLOW_all) 531{ 532 struct sock_filter filter[] = { 533 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 534 }; 535 struct sock_fprog prog = { 536 .len = (unsigned short)ARRAY_SIZE(filter), 537 .filter = filter, 538 }; 539 long ret; 540 541 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 542 ASSERT_EQ(0, ret); 543 544 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 545 ASSERT_EQ(0, ret); 546} 547 548TEST(empty_prog) 549{ 550 struct sock_filter filter[] = { 551 }; 552 struct sock_fprog prog = { 553 .len = (unsigned short)ARRAY_SIZE(filter), 554 .filter = filter, 555 }; 556 long ret; 557 558 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 559 ASSERT_EQ(0, ret); 560 561 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 562 EXPECT_EQ(-1, ret); 563 EXPECT_EQ(EINVAL, errno); 564} 565 566TEST(log_all) 567{ 568 struct sock_filter filter[] = { 569 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG), 570 }; 571 struct sock_fprog prog = { 572 .len = (unsigned short)ARRAY_SIZE(filter), 573 .filter = filter, 574 }; 575 long ret; 576 pid_t parent = getppid(); 577 578 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 579 ASSERT_EQ(0, ret); 580 581 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 582 ASSERT_EQ(0, ret); 583 584 /* getppid() should succeed and be logged (no check for logging) */ 585 EXPECT_EQ(parent, syscall(__NR_getppid)); 586} 587 588TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS) 589{ 590 struct sock_filter filter[] = { 591 BPF_STMT(BPF_RET|BPF_K, 0x10000000U), 592 }; 593 struct sock_fprog prog = { 594 .len = (unsigned short)ARRAY_SIZE(filter), 595 .filter = filter, 596 }; 597 long ret; 598 599 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 600 ASSERT_EQ(0, ret); 601 602 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 603 ASSERT_EQ(0, ret); 604 EXPECT_EQ(0, syscall(__NR_getpid)) { 605 TH_LOG("getpid() shouldn't ever return"); 606 } 607} 608 609/* return code >= 0x80000000 is unused. */ 610TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS) 611{ 612 struct sock_filter filter[] = { 613 BPF_STMT(BPF_RET|BPF_K, 0x90000000U), 614 }; 615 struct sock_fprog prog = { 616 .len = (unsigned short)ARRAY_SIZE(filter), 617 .filter = filter, 618 }; 619 long ret; 620 621 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 622 ASSERT_EQ(0, ret); 623 624 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 625 ASSERT_EQ(0, ret); 626 EXPECT_EQ(0, syscall(__NR_getpid)) { 627 TH_LOG("getpid() shouldn't ever return"); 628 } 629} 630 631TEST_SIGNAL(KILL_all, SIGSYS) 632{ 633 struct sock_filter filter[] = { 634 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 635 }; 636 struct sock_fprog prog = { 637 .len = (unsigned short)ARRAY_SIZE(filter), 638 .filter = filter, 639 }; 640 long ret; 641 642 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 643 ASSERT_EQ(0, ret); 644 645 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 646 ASSERT_EQ(0, ret); 647} 648 649TEST_SIGNAL(KILL_one, SIGSYS) 650{ 651 struct sock_filter filter[] = { 652 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 653 offsetof(struct seccomp_data, nr)), 654 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 655 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 656 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 657 }; 658 struct sock_fprog prog = { 659 .len = (unsigned short)ARRAY_SIZE(filter), 660 .filter = filter, 661 }; 662 long ret; 663 pid_t parent = getppid(); 664 665 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 666 ASSERT_EQ(0, ret); 667 668 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 669 ASSERT_EQ(0, ret); 670 671 EXPECT_EQ(parent, syscall(__NR_getppid)); 672 /* getpid() should never return. */ 673 EXPECT_EQ(0, syscall(__NR_getpid)); 674} 675 676TEST_SIGNAL(KILL_one_arg_one, SIGSYS) 677{ 678 void *fatal_address; 679 struct sock_filter filter[] = { 680 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 681 offsetof(struct seccomp_data, nr)), 682 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0), 683 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 684 /* Only both with lower 32-bit for now. */ 685 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)), 686 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 687 (unsigned long)&fatal_address, 0, 1), 688 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 689 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 690 }; 691 struct sock_fprog prog = { 692 .len = (unsigned short)ARRAY_SIZE(filter), 693 .filter = filter, 694 }; 695 long ret; 696 pid_t parent = getppid(); 697 struct tms timebuf; 698 clock_t clock = times(&timebuf); 699 700 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 701 ASSERT_EQ(0, ret); 702 703 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 704 ASSERT_EQ(0, ret); 705 706 EXPECT_EQ(parent, syscall(__NR_getppid)); 707 EXPECT_LE(clock, syscall(__NR_times, &timebuf)); 708 /* times() should never return. */ 709 EXPECT_EQ(0, syscall(__NR_times, &fatal_address)); 710} 711 712TEST_SIGNAL(KILL_one_arg_six, SIGSYS) 713{ 714#ifndef __NR_mmap2 715 int sysno = __NR_mmap; 716#else 717 int sysno = __NR_mmap2; 718#endif 719 struct sock_filter filter[] = { 720 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 721 offsetof(struct seccomp_data, nr)), 722 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0), 723 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 724 /* Only both with lower 32-bit for now. */ 725 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)), 726 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1), 727 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 728 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 729 }; 730 struct sock_fprog prog = { 731 .len = (unsigned short)ARRAY_SIZE(filter), 732 .filter = filter, 733 }; 734 long ret; 735 pid_t parent = getppid(); 736 int fd; 737 void *map1, *map2; 738 int page_size = sysconf(_SC_PAGESIZE); 739 740 ASSERT_LT(0, page_size); 741 742 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 743 ASSERT_EQ(0, ret); 744 745 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 746 ASSERT_EQ(0, ret); 747 748 fd = open("/dev/zero", O_RDONLY); 749 ASSERT_NE(-1, fd); 750 751 EXPECT_EQ(parent, syscall(__NR_getppid)); 752 map1 = (void *)syscall(sysno, 753 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size); 754 EXPECT_NE(MAP_FAILED, map1); 755 /* mmap2() should never return. */ 756 map2 = (void *)syscall(sysno, 757 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE); 758 EXPECT_EQ(MAP_FAILED, map2); 759 760 /* The test failed, so clean up the resources. */ 761 munmap(map1, page_size); 762 munmap(map2, page_size); 763 close(fd); 764} 765 766/* This is a thread task to die via seccomp filter violation. */ 767void *kill_thread(void *data) 768{ 769 bool die = (bool)data; 770 771 if (die) { 772 prctl(PR_GET_SECCOMP, 0, 0, 0, 0); 773 return (void *)SIBLING_EXIT_FAILURE; 774 } 775 776 return (void *)SIBLING_EXIT_UNKILLED; 777} 778 779enum kill_t { 780 KILL_THREAD, 781 KILL_PROCESS, 782 RET_UNKNOWN 783}; 784 785/* Prepare a thread that will kill itself or both of us. */ 786void kill_thread_or_group(struct __test_metadata *_metadata, 787 enum kill_t kill_how) 788{ 789 pthread_t thread; 790 void *status; 791 /* Kill only when calling __NR_prctl. */ 792 struct sock_filter filter_thread[] = { 793 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 794 offsetof(struct seccomp_data, nr)), 795 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1), 796 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD), 797 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 798 }; 799 struct sock_fprog prog_thread = { 800 .len = (unsigned short)ARRAY_SIZE(filter_thread), 801 .filter = filter_thread, 802 }; 803 int kill = kill_how == KILL_PROCESS ? SECCOMP_RET_KILL_PROCESS : 0xAAAAAAAA; 804 struct sock_filter filter_process[] = { 805 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 806 offsetof(struct seccomp_data, nr)), 807 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1), 808 BPF_STMT(BPF_RET|BPF_K, kill), 809 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 810 }; 811 struct sock_fprog prog_process = { 812 .len = (unsigned short)ARRAY_SIZE(filter_process), 813 .filter = filter_process, 814 }; 815 816 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 817 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 818 } 819 820 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, 821 kill_how == KILL_THREAD ? &prog_thread 822 : &prog_process)); 823 824 /* 825 * Add the KILL_THREAD rule again to make sure that the KILL_PROCESS 826 * flag cannot be downgraded by a new filter. 827 */ 828 if (kill_how == KILL_PROCESS) 829 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog_thread)); 830 831 /* Start a thread that will exit immediately. */ 832 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)false)); 833 ASSERT_EQ(0, pthread_join(thread, &status)); 834 ASSERT_EQ(SIBLING_EXIT_UNKILLED, (unsigned long)status); 835 836 /* Start a thread that will die immediately. */ 837 ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)true)); 838 ASSERT_EQ(0, pthread_join(thread, &status)); 839 ASSERT_NE(SIBLING_EXIT_FAILURE, (unsigned long)status); 840 841 /* 842 * If we get here, only the spawned thread died. Let the parent know 843 * the whole process didn't die (i.e. this thread, the spawner, 844 * stayed running). 845 */ 846 exit(42); 847} 848 849TEST(KILL_thread) 850{ 851 int status; 852 pid_t child_pid; 853 854 child_pid = fork(); 855 ASSERT_LE(0, child_pid); 856 if (child_pid == 0) { 857 kill_thread_or_group(_metadata, KILL_THREAD); 858 _exit(38); 859 } 860 861 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 862 863 /* If only the thread was killed, we'll see exit 42. */ 864 ASSERT_TRUE(WIFEXITED(status)); 865 ASSERT_EQ(42, WEXITSTATUS(status)); 866} 867 868TEST(KILL_process) 869{ 870 int status; 871 pid_t child_pid; 872 873 child_pid = fork(); 874 ASSERT_LE(0, child_pid); 875 if (child_pid == 0) { 876 kill_thread_or_group(_metadata, KILL_PROCESS); 877 _exit(38); 878 } 879 880 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 881 882 /* If the entire process was killed, we'll see SIGSYS. */ 883 ASSERT_TRUE(WIFSIGNALED(status)); 884 ASSERT_EQ(SIGSYS, WTERMSIG(status)); 885} 886 887TEST(KILL_unknown) 888{ 889 int status; 890 pid_t child_pid; 891 892 child_pid = fork(); 893 ASSERT_LE(0, child_pid); 894 if (child_pid == 0) { 895 kill_thread_or_group(_metadata, RET_UNKNOWN); 896 _exit(38); 897 } 898 899 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 900 901 /* If the entire process was killed, we'll see SIGSYS. */ 902 EXPECT_TRUE(WIFSIGNALED(status)) { 903 TH_LOG("Unknown SECCOMP_RET is only killing the thread?"); 904 } 905 ASSERT_EQ(SIGSYS, WTERMSIG(status)); 906} 907 908/* TODO(wad) add 64-bit versus 32-bit arg tests. */ 909TEST(arg_out_of_range) 910{ 911 struct sock_filter filter[] = { 912 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)), 913 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 914 }; 915 struct sock_fprog prog = { 916 .len = (unsigned short)ARRAY_SIZE(filter), 917 .filter = filter, 918 }; 919 long ret; 920 921 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 922 ASSERT_EQ(0, ret); 923 924 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog); 925 EXPECT_EQ(-1, ret); 926 EXPECT_EQ(EINVAL, errno); 927} 928 929#define ERRNO_FILTER(name, errno) \ 930 struct sock_filter _read_filter_##name[] = { \ 931 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, \ 932 offsetof(struct seccomp_data, nr)), \ 933 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), \ 934 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | errno), \ 935 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), \ 936 }; \ 937 struct sock_fprog prog_##name = { \ 938 .len = (unsigned short)ARRAY_SIZE(_read_filter_##name), \ 939 .filter = _read_filter_##name, \ 940 } 941 942/* Make sure basic errno values are correctly passed through a filter. */ 943TEST(ERRNO_valid) 944{ 945 ERRNO_FILTER(valid, E2BIG); 946 long ret; 947 pid_t parent = getppid(); 948 949 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 950 ASSERT_EQ(0, ret); 951 952 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_valid); 953 ASSERT_EQ(0, ret); 954 955 EXPECT_EQ(parent, syscall(__NR_getppid)); 956 EXPECT_EQ(-1, read(-1, NULL, 0)); 957 EXPECT_EQ(E2BIG, errno); 958} 959 960/* Make sure an errno of zero is correctly handled by the arch code. */ 961TEST(ERRNO_zero) 962{ 963 ERRNO_FILTER(zero, 0); 964 long ret; 965 pid_t parent = getppid(); 966 967 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 968 ASSERT_EQ(0, ret); 969 970 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_zero); 971 ASSERT_EQ(0, ret); 972 973 EXPECT_EQ(parent, syscall(__NR_getppid)); 974 /* "errno" of 0 is ok. */ 975 EXPECT_EQ(0, read(-1, NULL, 0)); 976} 977 978/* 979 * The SECCOMP_RET_DATA mask is 16 bits wide, but errno is smaller. 980 * This tests that the errno value gets capped correctly, fixed by 981 * 580c57f10768 ("seccomp: cap SECCOMP_RET_ERRNO data to MAX_ERRNO"). 982 */ 983TEST(ERRNO_capped) 984{ 985 ERRNO_FILTER(capped, 4096); 986 long ret; 987 pid_t parent = getppid(); 988 989 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 990 ASSERT_EQ(0, ret); 991 992 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_capped); 993 ASSERT_EQ(0, ret); 994 995 EXPECT_EQ(parent, syscall(__NR_getppid)); 996 EXPECT_EQ(-1, read(-1, NULL, 0)); 997 EXPECT_EQ(4095, errno); 998} 999 1000/* 1001 * Filters are processed in reverse order: last applied is executed first. 1002 * Since only the SECCOMP_RET_ACTION mask is tested for return values, the 1003 * SECCOMP_RET_DATA mask results will follow the most recently applied 1004 * matching filter return (and not the lowest or highest value). 1005 */ 1006TEST(ERRNO_order) 1007{ 1008 ERRNO_FILTER(first, 11); 1009 ERRNO_FILTER(second, 13); 1010 ERRNO_FILTER(third, 12); 1011 long ret; 1012 pid_t parent = getppid(); 1013 1014 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1015 ASSERT_EQ(0, ret); 1016 1017 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_first); 1018 ASSERT_EQ(0, ret); 1019 1020 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_second); 1021 ASSERT_EQ(0, ret); 1022 1023 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_third); 1024 ASSERT_EQ(0, ret); 1025 1026 EXPECT_EQ(parent, syscall(__NR_getppid)); 1027 EXPECT_EQ(-1, read(-1, NULL, 0)); 1028 EXPECT_EQ(12, errno); 1029} 1030 1031FIXTURE(TRAP) { 1032 struct sock_fprog prog; 1033}; 1034 1035FIXTURE_SETUP(TRAP) 1036{ 1037 struct sock_filter filter[] = { 1038 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1039 offsetof(struct seccomp_data, nr)), 1040 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 1041 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP), 1042 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1043 }; 1044 1045 memset(&self->prog, 0, sizeof(self->prog)); 1046 self->prog.filter = malloc(sizeof(filter)); 1047 ASSERT_NE(NULL, self->prog.filter); 1048 memcpy(self->prog.filter, filter, sizeof(filter)); 1049 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 1050} 1051 1052FIXTURE_TEARDOWN(TRAP) 1053{ 1054 if (self->prog.filter) 1055 free(self->prog.filter); 1056} 1057 1058TEST_F_SIGNAL(TRAP, dfl, SIGSYS) 1059{ 1060 long ret; 1061 1062 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1063 ASSERT_EQ(0, ret); 1064 1065 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 1066 ASSERT_EQ(0, ret); 1067 syscall(__NR_getpid); 1068} 1069 1070/* Ensure that SIGSYS overrides SIG_IGN */ 1071TEST_F_SIGNAL(TRAP, ign, SIGSYS) 1072{ 1073 long ret; 1074 1075 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1076 ASSERT_EQ(0, ret); 1077 1078 signal(SIGSYS, SIG_IGN); 1079 1080 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 1081 ASSERT_EQ(0, ret); 1082 syscall(__NR_getpid); 1083} 1084 1085static siginfo_t TRAP_info; 1086static volatile int TRAP_nr; 1087static void TRAP_action(int nr, siginfo_t *info, void *void_context) 1088{ 1089 memcpy(&TRAP_info, info, sizeof(TRAP_info)); 1090 TRAP_nr = nr; 1091} 1092 1093TEST_F(TRAP, handler) 1094{ 1095 int ret, test; 1096 struct sigaction act; 1097 sigset_t mask; 1098 1099 memset(&act, 0, sizeof(act)); 1100 sigemptyset(&mask); 1101 sigaddset(&mask, SIGSYS); 1102 1103 act.sa_sigaction = &TRAP_action; 1104 act.sa_flags = SA_SIGINFO; 1105 ret = sigaction(SIGSYS, &act, NULL); 1106 ASSERT_EQ(0, ret) { 1107 TH_LOG("sigaction failed"); 1108 } 1109 ret = sigprocmask(SIG_UNBLOCK, &mask, NULL); 1110 ASSERT_EQ(0, ret) { 1111 TH_LOG("sigprocmask failed"); 1112 } 1113 1114 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1115 ASSERT_EQ(0, ret); 1116 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog); 1117 ASSERT_EQ(0, ret); 1118 TRAP_nr = 0; 1119 memset(&TRAP_info, 0, sizeof(TRAP_info)); 1120 /* Expect the registers to be rolled back. (nr = error) may vary 1121 * based on arch. */ 1122 ret = syscall(__NR_getpid); 1123 /* Silence gcc warning about volatile. */ 1124 test = TRAP_nr; 1125 EXPECT_EQ(SIGSYS, test); 1126 struct local_sigsys { 1127 void *_call_addr; /* calling user insn */ 1128 int _syscall; /* triggering system call number */ 1129 unsigned int _arch; /* AUDIT_ARCH_* of syscall */ 1130 } *sigsys = (struct local_sigsys *) 1131#ifdef si_syscall 1132 &(TRAP_info.si_call_addr); 1133#else 1134 &TRAP_info.si_pid; 1135#endif 1136 EXPECT_EQ(__NR_getpid, sigsys->_syscall); 1137 /* Make sure arch is non-zero. */ 1138 EXPECT_NE(0, sigsys->_arch); 1139 EXPECT_NE(0, (unsigned long)sigsys->_call_addr); 1140} 1141 1142FIXTURE(precedence) { 1143 struct sock_fprog allow; 1144 struct sock_fprog log; 1145 struct sock_fprog trace; 1146 struct sock_fprog error; 1147 struct sock_fprog trap; 1148 struct sock_fprog kill; 1149}; 1150 1151FIXTURE_SETUP(precedence) 1152{ 1153 struct sock_filter allow_insns[] = { 1154 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1155 }; 1156 struct sock_filter log_insns[] = { 1157 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1158 offsetof(struct seccomp_data, nr)), 1159 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1160 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1161 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG), 1162 }; 1163 struct sock_filter trace_insns[] = { 1164 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1165 offsetof(struct seccomp_data, nr)), 1166 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1167 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1168 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE), 1169 }; 1170 struct sock_filter error_insns[] = { 1171 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1172 offsetof(struct seccomp_data, nr)), 1173 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1174 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1175 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO), 1176 }; 1177 struct sock_filter trap_insns[] = { 1178 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1179 offsetof(struct seccomp_data, nr)), 1180 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1181 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1182 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP), 1183 }; 1184 struct sock_filter kill_insns[] = { 1185 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1186 offsetof(struct seccomp_data, nr)), 1187 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0), 1188 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1189 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 1190 }; 1191 1192 memset(self, 0, sizeof(*self)); 1193#define FILTER_ALLOC(_x) \ 1194 self->_x.filter = malloc(sizeof(_x##_insns)); \ 1195 ASSERT_NE(NULL, self->_x.filter); \ 1196 memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \ 1197 self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns) 1198 FILTER_ALLOC(allow); 1199 FILTER_ALLOC(log); 1200 FILTER_ALLOC(trace); 1201 FILTER_ALLOC(error); 1202 FILTER_ALLOC(trap); 1203 FILTER_ALLOC(kill); 1204} 1205 1206FIXTURE_TEARDOWN(precedence) 1207{ 1208#define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter) 1209 FILTER_FREE(allow); 1210 FILTER_FREE(log); 1211 FILTER_FREE(trace); 1212 FILTER_FREE(error); 1213 FILTER_FREE(trap); 1214 FILTER_FREE(kill); 1215} 1216 1217TEST_F(precedence, allow_ok) 1218{ 1219 pid_t parent, res = 0; 1220 long ret; 1221 1222 parent = getppid(); 1223 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1224 ASSERT_EQ(0, ret); 1225 1226 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1227 ASSERT_EQ(0, ret); 1228 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1229 ASSERT_EQ(0, ret); 1230 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1231 ASSERT_EQ(0, ret); 1232 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1233 ASSERT_EQ(0, ret); 1234 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1235 ASSERT_EQ(0, ret); 1236 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1237 ASSERT_EQ(0, ret); 1238 /* Should work just fine. */ 1239 res = syscall(__NR_getppid); 1240 EXPECT_EQ(parent, res); 1241} 1242 1243TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS) 1244{ 1245 pid_t parent, res = 0; 1246 long ret; 1247 1248 parent = getppid(); 1249 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1250 ASSERT_EQ(0, ret); 1251 1252 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1253 ASSERT_EQ(0, ret); 1254 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1255 ASSERT_EQ(0, ret); 1256 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1257 ASSERT_EQ(0, ret); 1258 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1259 ASSERT_EQ(0, ret); 1260 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1261 ASSERT_EQ(0, ret); 1262 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1263 ASSERT_EQ(0, ret); 1264 /* Should work just fine. */ 1265 res = syscall(__NR_getppid); 1266 EXPECT_EQ(parent, res); 1267 /* getpid() should never return. */ 1268 res = syscall(__NR_getpid); 1269 EXPECT_EQ(0, res); 1270} 1271 1272TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS) 1273{ 1274 pid_t parent; 1275 long ret; 1276 1277 parent = getppid(); 1278 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1279 ASSERT_EQ(0, ret); 1280 1281 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1282 ASSERT_EQ(0, ret); 1283 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill); 1284 ASSERT_EQ(0, ret); 1285 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1286 ASSERT_EQ(0, ret); 1287 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1288 ASSERT_EQ(0, ret); 1289 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1290 ASSERT_EQ(0, ret); 1291 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1292 ASSERT_EQ(0, ret); 1293 /* Should work just fine. */ 1294 EXPECT_EQ(parent, syscall(__NR_getppid)); 1295 /* getpid() should never return. */ 1296 EXPECT_EQ(0, syscall(__NR_getpid)); 1297} 1298 1299TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS) 1300{ 1301 pid_t parent; 1302 long ret; 1303 1304 parent = getppid(); 1305 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1306 ASSERT_EQ(0, ret); 1307 1308 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1309 ASSERT_EQ(0, ret); 1310 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1311 ASSERT_EQ(0, ret); 1312 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1313 ASSERT_EQ(0, ret); 1314 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1315 ASSERT_EQ(0, ret); 1316 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1317 ASSERT_EQ(0, ret); 1318 /* Should work just fine. */ 1319 EXPECT_EQ(parent, syscall(__NR_getppid)); 1320 /* getpid() should never return. */ 1321 EXPECT_EQ(0, syscall(__NR_getpid)); 1322} 1323 1324TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS) 1325{ 1326 pid_t parent; 1327 long ret; 1328 1329 parent = getppid(); 1330 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1331 ASSERT_EQ(0, ret); 1332 1333 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1334 ASSERT_EQ(0, ret); 1335 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap); 1336 ASSERT_EQ(0, ret); 1337 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1338 ASSERT_EQ(0, ret); 1339 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1340 ASSERT_EQ(0, ret); 1341 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1342 ASSERT_EQ(0, ret); 1343 /* Should work just fine. */ 1344 EXPECT_EQ(parent, syscall(__NR_getppid)); 1345 /* getpid() should never return. */ 1346 EXPECT_EQ(0, syscall(__NR_getpid)); 1347} 1348 1349TEST_F(precedence, errno_is_third) 1350{ 1351 pid_t parent; 1352 long ret; 1353 1354 parent = getppid(); 1355 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1356 ASSERT_EQ(0, ret); 1357 1358 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1359 ASSERT_EQ(0, ret); 1360 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1361 ASSERT_EQ(0, ret); 1362 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1363 ASSERT_EQ(0, ret); 1364 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 1365 ASSERT_EQ(0, ret); 1366 /* Should work just fine. */ 1367 EXPECT_EQ(parent, syscall(__NR_getppid)); 1368 EXPECT_EQ(0, syscall(__NR_getpid)); 1369} 1370 1371TEST_F(precedence, errno_is_third_in_any_order) 1372{ 1373 pid_t parent; 1374 long ret; 1375 1376 parent = getppid(); 1377 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1378 ASSERT_EQ(0, ret); 1379 1380 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1381 ASSERT_EQ(0, ret); 1382 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error); 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->allow); 1387 ASSERT_EQ(0, ret); 1388 /* Should work just fine. */ 1389 EXPECT_EQ(parent, syscall(__NR_getppid)); 1390 EXPECT_EQ(0, syscall(__NR_getpid)); 1391} 1392 1393TEST_F(precedence, trace_is_fourth) 1394{ 1395 pid_t parent; 1396 long ret; 1397 1398 parent = getppid(); 1399 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1400 ASSERT_EQ(0, ret); 1401 1402 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1403 ASSERT_EQ(0, ret); 1404 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1405 ASSERT_EQ(0, ret); 1406 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1407 ASSERT_EQ(0, ret); 1408 /* Should work just fine. */ 1409 EXPECT_EQ(parent, syscall(__NR_getppid)); 1410 /* No ptracer */ 1411 EXPECT_EQ(-1, syscall(__NR_getpid)); 1412} 1413 1414TEST_F(precedence, trace_is_fourth_in_any_order) 1415{ 1416 pid_t parent; 1417 long ret; 1418 1419 parent = getppid(); 1420 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1421 ASSERT_EQ(0, ret); 1422 1423 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace); 1424 ASSERT_EQ(0, ret); 1425 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1426 ASSERT_EQ(0, ret); 1427 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1428 ASSERT_EQ(0, ret); 1429 /* Should work just fine. */ 1430 EXPECT_EQ(parent, syscall(__NR_getppid)); 1431 /* No ptracer */ 1432 EXPECT_EQ(-1, syscall(__NR_getpid)); 1433} 1434 1435TEST_F(precedence, log_is_fifth) 1436{ 1437 pid_t mypid, parent; 1438 long ret; 1439 1440 mypid = getpid(); 1441 parent = getppid(); 1442 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1443 ASSERT_EQ(0, ret); 1444 1445 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1446 ASSERT_EQ(0, ret); 1447 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1448 ASSERT_EQ(0, ret); 1449 /* Should work just fine. */ 1450 EXPECT_EQ(parent, syscall(__NR_getppid)); 1451 /* Should also work just fine */ 1452 EXPECT_EQ(mypid, syscall(__NR_getpid)); 1453} 1454 1455TEST_F(precedence, log_is_fifth_in_any_order) 1456{ 1457 pid_t mypid, parent; 1458 long ret; 1459 1460 mypid = getpid(); 1461 parent = getppid(); 1462 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1463 ASSERT_EQ(0, ret); 1464 1465 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log); 1466 ASSERT_EQ(0, ret); 1467 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow); 1468 ASSERT_EQ(0, ret); 1469 /* Should work just fine. */ 1470 EXPECT_EQ(parent, syscall(__NR_getppid)); 1471 /* Should also work just fine */ 1472 EXPECT_EQ(mypid, syscall(__NR_getpid)); 1473} 1474 1475#ifndef PTRACE_O_TRACESECCOMP 1476#define PTRACE_O_TRACESECCOMP 0x00000080 1477#endif 1478 1479/* Catch the Ubuntu 12.04 value error. */ 1480#if PTRACE_EVENT_SECCOMP != 7 1481#undef PTRACE_EVENT_SECCOMP 1482#endif 1483 1484#ifndef PTRACE_EVENT_SECCOMP 1485#define PTRACE_EVENT_SECCOMP 7 1486#endif 1487 1488#define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP) 1489bool tracer_running; 1490void tracer_stop(int sig) 1491{ 1492 tracer_running = false; 1493} 1494 1495typedef void tracer_func_t(struct __test_metadata *_metadata, 1496 pid_t tracee, int status, void *args); 1497 1498void start_tracer(struct __test_metadata *_metadata, int fd, pid_t tracee, 1499 tracer_func_t tracer_func, void *args, bool ptrace_syscall) 1500{ 1501 int ret = -1; 1502 struct sigaction action = { 1503 .sa_handler = tracer_stop, 1504 }; 1505 1506 /* Allow external shutdown. */ 1507 tracer_running = true; 1508 ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL)); 1509 1510 errno = 0; 1511 while (ret == -1 && errno != EINVAL) 1512 ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0); 1513 ASSERT_EQ(0, ret) { 1514 kill(tracee, SIGKILL); 1515 } 1516 /* Wait for attach stop */ 1517 wait(NULL); 1518 1519 ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, ptrace_syscall ? 1520 PTRACE_O_TRACESYSGOOD : 1521 PTRACE_O_TRACESECCOMP); 1522 ASSERT_EQ(0, ret) { 1523 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP"); 1524 kill(tracee, SIGKILL); 1525 } 1526 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT, 1527 tracee, NULL, 0); 1528 ASSERT_EQ(0, ret); 1529 1530 /* Unblock the tracee */ 1531 ASSERT_EQ(1, write(fd, "A", 1)); 1532 ASSERT_EQ(0, close(fd)); 1533 1534 /* Run until we're shut down. Must assert to stop execution. */ 1535 while (tracer_running) { 1536 int status; 1537 1538 if (wait(&status) != tracee) 1539 continue; 1540 if (WIFSIGNALED(status) || WIFEXITED(status)) 1541 /* Child is dead. Time to go. */ 1542 return; 1543 1544 /* Check if this is a seccomp event. */ 1545 ASSERT_EQ(!ptrace_syscall, IS_SECCOMP_EVENT(status)); 1546 1547 tracer_func(_metadata, tracee, status, args); 1548 1549 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT, 1550 tracee, NULL, 0); 1551 ASSERT_EQ(0, ret); 1552 } 1553 /* Directly report the status of our test harness results. */ 1554 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS : EXIT_FAILURE); 1555} 1556 1557/* Common tracer setup/teardown functions. */ 1558void cont_handler(int num) 1559{ } 1560pid_t setup_trace_fixture(struct __test_metadata *_metadata, 1561 tracer_func_t func, void *args, bool ptrace_syscall) 1562{ 1563 char sync; 1564 int pipefd[2]; 1565 pid_t tracer_pid; 1566 pid_t tracee = getpid(); 1567 1568 /* Setup a pipe for clean synchronization. */ 1569 ASSERT_EQ(0, pipe(pipefd)); 1570 1571 /* Fork a child which we'll promote to tracer */ 1572 tracer_pid = fork(); 1573 ASSERT_LE(0, tracer_pid); 1574 signal(SIGALRM, cont_handler); 1575 if (tracer_pid == 0) { 1576 close(pipefd[0]); 1577 start_tracer(_metadata, pipefd[1], tracee, func, args, 1578 ptrace_syscall); 1579 syscall(__NR_exit, 0); 1580 } 1581 close(pipefd[1]); 1582 prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0); 1583 read(pipefd[0], &sync, 1); 1584 close(pipefd[0]); 1585 1586 return tracer_pid; 1587} 1588 1589void teardown_trace_fixture(struct __test_metadata *_metadata, 1590 pid_t tracer) 1591{ 1592 if (tracer) { 1593 int status; 1594 /* 1595 * Extract the exit code from the other process and 1596 * adopt it for ourselves in case its asserts failed. 1597 */ 1598 ASSERT_EQ(0, kill(tracer, SIGUSR1)); 1599 ASSERT_EQ(tracer, waitpid(tracer, &status, 0)); 1600 if (WEXITSTATUS(status)) 1601 _metadata->passed = 0; 1602 } 1603} 1604 1605/* "poke" tracer arguments and function. */ 1606struct tracer_args_poke_t { 1607 unsigned long poke_addr; 1608}; 1609 1610void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status, 1611 void *args) 1612{ 1613 int ret; 1614 unsigned long msg; 1615 struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args; 1616 1617 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 1618 EXPECT_EQ(0, ret); 1619 /* If this fails, don't try to recover. */ 1620 ASSERT_EQ(0x1001, msg) { 1621 kill(tracee, SIGKILL); 1622 } 1623 /* 1624 * Poke in the message. 1625 * Registers are not touched to try to keep this relatively arch 1626 * agnostic. 1627 */ 1628 ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001); 1629 EXPECT_EQ(0, ret); 1630} 1631 1632FIXTURE(TRACE_poke) { 1633 struct sock_fprog prog; 1634 pid_t tracer; 1635 long poked; 1636 struct tracer_args_poke_t tracer_args; 1637}; 1638 1639FIXTURE_SETUP(TRACE_poke) 1640{ 1641 struct sock_filter filter[] = { 1642 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 1643 offsetof(struct seccomp_data, nr)), 1644 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 1645 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001), 1646 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 1647 }; 1648 1649 self->poked = 0; 1650 memset(&self->prog, 0, sizeof(self->prog)); 1651 self->prog.filter = malloc(sizeof(filter)); 1652 ASSERT_NE(NULL, self->prog.filter); 1653 memcpy(self->prog.filter, filter, sizeof(filter)); 1654 self->prog.len = (unsigned short)ARRAY_SIZE(filter); 1655 1656 /* Set up tracer args. */ 1657 self->tracer_args.poke_addr = (unsigned long)&self->poked; 1658 1659 /* Launch tracer. */ 1660 self->tracer = setup_trace_fixture(_metadata, tracer_poke, 1661 &self->tracer_args, false); 1662} 1663 1664FIXTURE_TEARDOWN(TRACE_poke) 1665{ 1666 teardown_trace_fixture(_metadata, self->tracer); 1667 if (self->prog.filter) 1668 free(self->prog.filter); 1669} 1670 1671TEST_F(TRACE_poke, read_has_side_effects) 1672{ 1673 ssize_t ret; 1674 1675 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1676 ASSERT_EQ(0, ret); 1677 1678 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1679 ASSERT_EQ(0, ret); 1680 1681 EXPECT_EQ(0, self->poked); 1682 ret = read(-1, NULL, 0); 1683 EXPECT_EQ(-1, ret); 1684 EXPECT_EQ(0x1001, self->poked); 1685} 1686 1687TEST_F(TRACE_poke, getpid_runs_normally) 1688{ 1689 long ret; 1690 1691 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 1692 ASSERT_EQ(0, ret); 1693 1694 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0); 1695 ASSERT_EQ(0, ret); 1696 1697 EXPECT_EQ(0, self->poked); 1698 EXPECT_NE(0, syscall(__NR_getpid)); 1699 EXPECT_EQ(0, self->poked); 1700} 1701 1702#if defined(__x86_64__) 1703# define ARCH_REGS struct user_regs_struct 1704# define SYSCALL_NUM(_regs) (_regs).orig_rax 1705# define SYSCALL_RET(_regs) (_regs).rax 1706#elif defined(__i386__) 1707# define ARCH_REGS struct user_regs_struct 1708# define SYSCALL_NUM(_regs) (_regs).orig_eax 1709# define SYSCALL_RET(_regs) (_regs).eax 1710#elif defined(__arm__) 1711# define ARCH_REGS struct pt_regs 1712# define SYSCALL_NUM(_regs) (_regs).ARM_r7 1713# ifndef PTRACE_SET_SYSCALL 1714# define PTRACE_SET_SYSCALL 23 1715# endif 1716# define SYSCALL_NUM_SET(_regs, _nr) \ 1717 EXPECT_EQ(0, ptrace(PTRACE_SET_SYSCALL, tracee, NULL, _nr)) 1718# define SYSCALL_RET(_regs) (_regs).ARM_r0 1719#elif defined(__aarch64__) 1720# define ARCH_REGS struct user_pt_regs 1721# define SYSCALL_NUM(_regs) (_regs).regs[8] 1722# ifndef NT_ARM_SYSTEM_CALL 1723# define NT_ARM_SYSTEM_CALL 0x404 1724# endif 1725# define SYSCALL_NUM_SET(_regs, _nr) \ 1726 do { \ 1727 struct iovec __v; \ 1728 typeof(_nr) __nr = (_nr); \ 1729 __v.iov_base = &__nr; \ 1730 __v.iov_len = sizeof(__nr); \ 1731 EXPECT_EQ(0, ptrace(PTRACE_SETREGSET, tracee, \ 1732 NT_ARM_SYSTEM_CALL, &__v)); \ 1733 } while (0) 1734# define SYSCALL_RET(_regs) (_regs).regs[0] 1735#elif defined(__loongarch__) 1736# define ARCH_REGS struct user_pt_regs 1737# define SYSCALL_NUM(_regs) (_regs).regs[11] 1738# define SYSCALL_RET(_regs) (_regs).regs[4] 1739#elif defined(__riscv) && __riscv_xlen == 64 1740# define ARCH_REGS struct user_regs_struct 1741# define SYSCALL_NUM(_regs) (_regs).a7 1742# define SYSCALL_RET(_regs) (_regs).a0 1743#elif defined(__csky__) 1744# define ARCH_REGS struct pt_regs 1745# if defined(__CSKYABIV2__) 1746# define SYSCALL_NUM(_regs) (_regs).regs[3] 1747# else 1748# define SYSCALL_NUM(_regs) (_regs).regs[9] 1749# endif 1750# define SYSCALL_RET(_regs) (_regs).a0 1751#elif defined(__hppa__) 1752# define ARCH_REGS struct user_regs_struct 1753# define SYSCALL_NUM(_regs) (_regs).gr[20] 1754# define SYSCALL_RET(_regs) (_regs).gr[28] 1755#elif defined(__powerpc__) 1756# define ARCH_REGS struct pt_regs 1757# define SYSCALL_NUM(_regs) (_regs).gpr[0] 1758# define SYSCALL_RET(_regs) (_regs).gpr[3] 1759# define SYSCALL_RET_SET(_regs, _val) \ 1760 do { \ 1761 typeof(_val) _result = (_val); \ 1762 if ((_regs.trap & 0xfff0) == 0x3000) { \ 1763 /* \ 1764 * scv 0 system call uses -ve result \ 1765 * for error, so no need to adjust. \ 1766 */ \ 1767 SYSCALL_RET(_regs) = _result; \ 1768 } else { \ 1769 /* \ 1770 * A syscall error is signaled by the \ 1771 * CR0 SO bit and the code is stored as \ 1772 * a positive value. \ 1773 */ \ 1774 if (_result < 0) { \ 1775 SYSCALL_RET(_regs) = -_result; \ 1776 (_regs).ccr |= 0x10000000; \ 1777 } else { \ 1778 SYSCALL_RET(_regs) = _result; \ 1779 (_regs).ccr &= ~0x10000000; \ 1780 } \ 1781 } \ 1782 } while (0) 1783# define SYSCALL_RET_SET_ON_PTRACE_EXIT 1784#elif defined(__s390__) 1785# define ARCH_REGS s390_regs 1786# define SYSCALL_NUM(_regs) (_regs).gprs[2] 1787# define SYSCALL_RET_SET(_regs, _val) \ 1788 TH_LOG("Can't modify syscall return on this architecture") 1789#elif defined(__mips__) 1790# include <asm/unistd_nr_n32.h> 1791# include <asm/unistd_nr_n64.h> 1792# include <asm/unistd_nr_o32.h> 1793# define ARCH_REGS struct pt_regs 1794# define SYSCALL_NUM(_regs) \ 1795 ({ \ 1796 typeof((_regs).regs[2]) _nr; \ 1797 if ((_regs).regs[2] == __NR_O32_Linux) \ 1798 _nr = (_regs).regs[4]; \ 1799 else \ 1800 _nr = (_regs).regs[2]; \ 1801 _nr; \ 1802 }) 1803# define SYSCALL_NUM_SET(_regs, _nr) \ 1804 do { \ 1805 if ((_regs).regs[2] == __NR_O32_Linux) \ 1806 (_regs).regs[4] = _nr; \ 1807 else \ 1808 (_regs).regs[2] = _nr; \ 1809 } while (0) 1810# define SYSCALL_RET_SET(_regs, _val) \ 1811 TH_LOG("Can't modify syscall return on this architecture") 1812#elif defined(__xtensa__) 1813# define ARCH_REGS struct user_pt_regs 1814# define SYSCALL_NUM(_regs) (_regs).syscall 1815/* 1816 * On xtensa syscall return value is in the register 1817 * a2 of the current window which is not fixed. 1818 */ 1819#define SYSCALL_RET(_regs) (_regs).a[(_regs).windowbase * 4 + 2] 1820#elif defined(__sh__) 1821# define ARCH_REGS struct pt_regs 1822# define SYSCALL_NUM(_regs) (_regs).regs[3] 1823# define SYSCALL_RET(_regs) (_regs).regs[0] 1824#else 1825# error "Do not know how to find your architecture's registers and syscalls" 1826#endif 1827 1828/* 1829 * Most architectures can change the syscall by just updating the 1830 * associated register. This is the default if not defined above. 1831 */ 1832#ifndef SYSCALL_NUM_SET 1833# define SYSCALL_NUM_SET(_regs, _nr) \ 1834 do { \ 1835 SYSCALL_NUM(_regs) = (_nr); \ 1836 } while (0) 1837#endif 1838/* 1839 * Most architectures can change the syscall return value by just 1840 * writing to the SYSCALL_RET register. This is the default if not 1841 * defined above. If an architecture cannot set the return value 1842 * (for example when the syscall and return value register is 1843 * shared), report it with TH_LOG() in an arch-specific definition 1844 * of SYSCALL_RET_SET() above, and leave SYSCALL_RET undefined. 1845 */ 1846#if !defined(SYSCALL_RET) && !defined(SYSCALL_RET_SET) 1847# error "One of SYSCALL_RET or SYSCALL_RET_SET is needed for this arch" 1848#endif 1849#ifndef SYSCALL_RET_SET 1850# define SYSCALL_RET_SET(_regs, _val) \ 1851 do { \ 1852 SYSCALL_RET(_regs) = (_val); \ 1853 } while (0) 1854#endif 1855 1856/* When the syscall return can't be changed, stub out the tests for it. */ 1857#ifndef SYSCALL_RET 1858# define EXPECT_SYSCALL_RETURN(val, action) EXPECT_EQ(-1, action) 1859#else 1860# define EXPECT_SYSCALL_RETURN(val, action) \ 1861 do { \ 1862 errno = 0; \ 1863 if (val < 0) { \ 1864 EXPECT_EQ(-1, action); \ 1865 EXPECT_EQ(-(val), errno); \ 1866 } else { \ 1867 EXPECT_EQ(val, action); \ 1868 } \ 1869 } while (0) 1870#endif 1871 1872/* 1873 * Some architectures (e.g. powerpc) can only set syscall 1874 * return values on syscall exit during ptrace. 1875 */ 1876const bool ptrace_entry_set_syscall_nr = true; 1877const bool ptrace_entry_set_syscall_ret = 1878#ifndef SYSCALL_RET_SET_ON_PTRACE_EXIT 1879 true; 1880#else 1881 false; 1882#endif 1883 1884/* 1885 * Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for 1886 * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux). 1887 */ 1888#if defined(__x86_64__) || defined(__i386__) || defined(__mips__) 1889# define ARCH_GETREGS(_regs) ptrace(PTRACE_GETREGS, tracee, 0, &(_regs)) 1890# define ARCH_SETREGS(_regs) ptrace(PTRACE_SETREGS, tracee, 0, &(_regs)) 1891#else 1892# define ARCH_GETREGS(_regs) ({ \ 1893 struct iovec __v; \ 1894 __v.iov_base = &(_regs); \ 1895 __v.iov_len = sizeof(_regs); \ 1896 ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &__v); \ 1897 }) 1898# define ARCH_SETREGS(_regs) ({ \ 1899 struct iovec __v; \ 1900 __v.iov_base = &(_regs); \ 1901 __v.iov_len = sizeof(_regs); \ 1902 ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &__v); \ 1903 }) 1904#endif 1905 1906/* Architecture-specific syscall fetching routine. */ 1907int get_syscall(struct __test_metadata *_metadata, pid_t tracee) 1908{ 1909 ARCH_REGS regs; 1910 1911 EXPECT_EQ(0, ARCH_GETREGS(regs)) { 1912 return -1; 1913 } 1914 1915 return SYSCALL_NUM(regs); 1916} 1917 1918/* Architecture-specific syscall changing routine. */ 1919void __change_syscall(struct __test_metadata *_metadata, 1920 pid_t tracee, long *syscall, long *ret) 1921{ 1922 ARCH_REGS orig, regs; 1923 1924 /* Do not get/set registers if we have nothing to do. */ 1925 if (!syscall && !ret) 1926 return; 1927 1928 EXPECT_EQ(0, ARCH_GETREGS(regs)) { 1929 return; 1930 } 1931 orig = regs; 1932 1933 if (syscall) 1934 SYSCALL_NUM_SET(regs, *syscall); 1935 1936 if (ret) 1937 SYSCALL_RET_SET(regs, *ret); 1938 1939 /* Flush any register changes made. */ 1940 if (memcmp(&orig, ®s, sizeof(orig)) != 0) 1941 EXPECT_EQ(0, ARCH_SETREGS(regs)); 1942} 1943 1944/* Change only syscall number. */ 1945void change_syscall_nr(struct __test_metadata *_metadata, 1946 pid_t tracee, long syscall) 1947{ 1948 __change_syscall(_metadata, tracee, &syscall, NULL); 1949} 1950 1951/* Change syscall return value (and set syscall number to -1). */ 1952void change_syscall_ret(struct __test_metadata *_metadata, 1953 pid_t tracee, long ret) 1954{ 1955 long syscall = -1; 1956 1957 __change_syscall(_metadata, tracee, &syscall, &ret); 1958} 1959 1960void tracer_seccomp(struct __test_metadata *_metadata, pid_t tracee, 1961 int status, void *args) 1962{ 1963 int ret; 1964 unsigned long msg; 1965 1966 /* Make sure we got the right message. */ 1967 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 1968 EXPECT_EQ(0, ret); 1969 1970 /* Validate and take action on expected syscalls. */ 1971 switch (msg) { 1972 case 0x1002: 1973 /* change getpid to getppid. */ 1974 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee)); 1975 change_syscall_nr(_metadata, tracee, __NR_getppid); 1976 break; 1977 case 0x1003: 1978 /* skip gettid with valid return code. */ 1979 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee)); 1980 change_syscall_ret(_metadata, tracee, 45000); 1981 break; 1982 case 0x1004: 1983 /* skip openat with error. */ 1984 EXPECT_EQ(__NR_openat, get_syscall(_metadata, tracee)); 1985 change_syscall_ret(_metadata, tracee, -ESRCH); 1986 break; 1987 case 0x1005: 1988 /* do nothing (allow getppid) */ 1989 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee)); 1990 break; 1991 default: 1992 EXPECT_EQ(0, msg) { 1993 TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg); 1994 kill(tracee, SIGKILL); 1995 } 1996 } 1997 1998} 1999 2000FIXTURE(TRACE_syscall) { 2001 struct sock_fprog prog; 2002 pid_t tracer, mytid, mypid, parent; 2003 long syscall_nr; 2004}; 2005 2006void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee, 2007 int status, void *args) 2008{ 2009 int ret; 2010 unsigned long msg; 2011 static bool entry; 2012 long syscall_nr_val, syscall_ret_val; 2013 long *syscall_nr = NULL, *syscall_ret = NULL; 2014 FIXTURE_DATA(TRACE_syscall) *self = args; 2015 2016 /* 2017 * The traditional way to tell PTRACE_SYSCALL entry/exit 2018 * is by counting. 2019 */ 2020 entry = !entry; 2021 2022 /* Make sure we got an appropriate message. */ 2023 ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg); 2024 EXPECT_EQ(0, ret); 2025 EXPECT_EQ(entry ? PTRACE_EVENTMSG_SYSCALL_ENTRY 2026 : PTRACE_EVENTMSG_SYSCALL_EXIT, msg); 2027 2028 /* 2029 * Some architectures only support setting return values during 2030 * syscall exit under ptrace, and on exit the syscall number may 2031 * no longer be available. Therefore, save the initial sycall 2032 * number here, so it can be examined during both entry and exit 2033 * phases. 2034 */ 2035 if (entry) 2036 self->syscall_nr = get_syscall(_metadata, tracee); 2037 2038 /* 2039 * Depending on the architecture's syscall setting abilities, we 2040 * pick which things to set during this phase (entry or exit). 2041 */ 2042 if (entry == ptrace_entry_set_syscall_nr) 2043 syscall_nr = &syscall_nr_val; 2044 if (entry == ptrace_entry_set_syscall_ret) 2045 syscall_ret = &syscall_ret_val; 2046 2047 /* Now handle the actual rewriting cases. */ 2048 switch (self->syscall_nr) { 2049 case __NR_getpid: 2050 syscall_nr_val = __NR_getppid; 2051 /* Never change syscall return for this case. */ 2052 syscall_ret = NULL; 2053 break; 2054 case __NR_gettid: 2055 syscall_nr_val = -1; 2056 syscall_ret_val = 45000; 2057 break; 2058 case __NR_openat: 2059 syscall_nr_val = -1; 2060 syscall_ret_val = -ESRCH; 2061 break; 2062 default: 2063 /* Unhandled, do nothing. */ 2064 return; 2065 } 2066 2067 __change_syscall(_metadata, tracee, syscall_nr, syscall_ret); 2068} 2069 2070FIXTURE_VARIANT(TRACE_syscall) { 2071 /* 2072 * All of the SECCOMP_RET_TRACE behaviors can be tested with either 2073 * SECCOMP_RET_TRACE+PTRACE_CONT or plain ptrace()+PTRACE_SYSCALL. 2074 * This indicates if we should use SECCOMP_RET_TRACE (false), or 2075 * ptrace (true). 2076 */ 2077 bool use_ptrace; 2078}; 2079 2080FIXTURE_VARIANT_ADD(TRACE_syscall, ptrace) { 2081 .use_ptrace = true, 2082}; 2083 2084FIXTURE_VARIANT_ADD(TRACE_syscall, seccomp) { 2085 .use_ptrace = false, 2086}; 2087 2088FIXTURE_SETUP(TRACE_syscall) 2089{ 2090 struct sock_filter filter[] = { 2091 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2092 offsetof(struct seccomp_data, nr)), 2093 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 2094 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002), 2095 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1), 2096 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003), 2097 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_openat, 0, 1), 2098 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004), 2099 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 2100 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1005), 2101 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2102 }; 2103 struct sock_fprog prog = { 2104 .len = (unsigned short)ARRAY_SIZE(filter), 2105 .filter = filter, 2106 }; 2107 long ret; 2108 2109 /* Prepare some testable syscall results. */ 2110 self->mytid = syscall(__NR_gettid); 2111 ASSERT_GT(self->mytid, 0); 2112 ASSERT_NE(self->mytid, 1) { 2113 TH_LOG("Running this test as init is not supported. :)"); 2114 } 2115 2116 self->mypid = getpid(); 2117 ASSERT_GT(self->mypid, 0); 2118 ASSERT_EQ(self->mytid, self->mypid); 2119 2120 self->parent = getppid(); 2121 ASSERT_GT(self->parent, 0); 2122 ASSERT_NE(self->parent, self->mypid); 2123 2124 /* Launch tracer. */ 2125 self->tracer = setup_trace_fixture(_metadata, 2126 variant->use_ptrace ? tracer_ptrace 2127 : tracer_seccomp, 2128 self, variant->use_ptrace); 2129 2130 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 2131 ASSERT_EQ(0, ret); 2132 2133 if (variant->use_ptrace) 2134 return; 2135 2136 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2137 ASSERT_EQ(0, ret); 2138} 2139 2140FIXTURE_TEARDOWN(TRACE_syscall) 2141{ 2142 teardown_trace_fixture(_metadata, self->tracer); 2143} 2144 2145TEST(negative_ENOSYS) 2146{ 2147 /* 2148 * There should be no difference between an "internal" skip 2149 * and userspace asking for syscall "-1". 2150 */ 2151 errno = 0; 2152 EXPECT_EQ(-1, syscall(-1)); 2153 EXPECT_EQ(errno, ENOSYS); 2154 /* And no difference for "still not valid but not -1". */ 2155 errno = 0; 2156 EXPECT_EQ(-1, syscall(-101)); 2157 EXPECT_EQ(errno, ENOSYS); 2158} 2159 2160TEST_F(TRACE_syscall, negative_ENOSYS) 2161{ 2162 negative_ENOSYS(_metadata); 2163} 2164 2165TEST_F(TRACE_syscall, syscall_allowed) 2166{ 2167 /* getppid works as expected (no changes). */ 2168 EXPECT_EQ(self->parent, syscall(__NR_getppid)); 2169 EXPECT_NE(self->mypid, syscall(__NR_getppid)); 2170} 2171 2172TEST_F(TRACE_syscall, syscall_redirected) 2173{ 2174 /* getpid has been redirected to getppid as expected. */ 2175 EXPECT_EQ(self->parent, syscall(__NR_getpid)); 2176 EXPECT_NE(self->mypid, syscall(__NR_getpid)); 2177} 2178 2179TEST_F(TRACE_syscall, syscall_errno) 2180{ 2181 /* Tracer should skip the open syscall, resulting in ESRCH. */ 2182 EXPECT_SYSCALL_RETURN(-ESRCH, syscall(__NR_openat)); 2183} 2184 2185TEST_F(TRACE_syscall, syscall_faked) 2186{ 2187 /* Tracer skips the gettid syscall and store altered return value. */ 2188 EXPECT_SYSCALL_RETURN(45000, syscall(__NR_gettid)); 2189} 2190 2191TEST_F(TRACE_syscall, skip_after) 2192{ 2193 struct sock_filter filter[] = { 2194 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2195 offsetof(struct seccomp_data, nr)), 2196 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 2197 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM), 2198 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2199 }; 2200 struct sock_fprog prog = { 2201 .len = (unsigned short)ARRAY_SIZE(filter), 2202 .filter = filter, 2203 }; 2204 long ret; 2205 2206 /* Install additional "errno on getppid" filter. */ 2207 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2208 ASSERT_EQ(0, ret); 2209 2210 /* Tracer will redirect getpid to getppid, and we should see EPERM. */ 2211 errno = 0; 2212 EXPECT_EQ(-1, syscall(__NR_getpid)); 2213 EXPECT_EQ(EPERM, errno); 2214} 2215 2216TEST_F_SIGNAL(TRACE_syscall, kill_after, SIGSYS) 2217{ 2218 struct sock_filter filter[] = { 2219 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2220 offsetof(struct seccomp_data, nr)), 2221 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1), 2222 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2223 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2224 }; 2225 struct sock_fprog prog = { 2226 .len = (unsigned short)ARRAY_SIZE(filter), 2227 .filter = filter, 2228 }; 2229 long ret; 2230 2231 /* Install additional "death on getppid" filter. */ 2232 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2233 ASSERT_EQ(0, ret); 2234 2235 /* Tracer will redirect getpid to getppid, and we should die. */ 2236 EXPECT_NE(self->mypid, syscall(__NR_getpid)); 2237} 2238 2239TEST(seccomp_syscall) 2240{ 2241 struct sock_filter filter[] = { 2242 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2243 }; 2244 struct sock_fprog prog = { 2245 .len = (unsigned short)ARRAY_SIZE(filter), 2246 .filter = filter, 2247 }; 2248 long ret; 2249 2250 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 2251 ASSERT_EQ(0, ret) { 2252 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2253 } 2254 2255 /* Reject insane operation. */ 2256 ret = seccomp(-1, 0, &prog); 2257 ASSERT_NE(ENOSYS, errno) { 2258 TH_LOG("Kernel does not support seccomp syscall!"); 2259 } 2260 EXPECT_EQ(EINVAL, errno) { 2261 TH_LOG("Did not reject crazy op value!"); 2262 } 2263 2264 /* Reject strict with flags or pointer. */ 2265 ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL); 2266 EXPECT_EQ(EINVAL, errno) { 2267 TH_LOG("Did not reject mode strict with flags!"); 2268 } 2269 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog); 2270 EXPECT_EQ(EINVAL, errno) { 2271 TH_LOG("Did not reject mode strict with uargs!"); 2272 } 2273 2274 /* Reject insane args for filter. */ 2275 ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog); 2276 EXPECT_EQ(EINVAL, errno) { 2277 TH_LOG("Did not reject crazy filter flags!"); 2278 } 2279 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL); 2280 EXPECT_EQ(EFAULT, errno) { 2281 TH_LOG("Did not reject NULL filter!"); 2282 } 2283 2284 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2285 EXPECT_EQ(0, errno) { 2286 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s", 2287 strerror(errno)); 2288 } 2289} 2290 2291TEST(seccomp_syscall_mode_lock) 2292{ 2293 struct sock_filter filter[] = { 2294 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2295 }; 2296 struct sock_fprog prog = { 2297 .len = (unsigned short)ARRAY_SIZE(filter), 2298 .filter = filter, 2299 }; 2300 long ret; 2301 2302 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 2303 ASSERT_EQ(0, ret) { 2304 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2305 } 2306 2307 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2308 ASSERT_NE(ENOSYS, errno) { 2309 TH_LOG("Kernel does not support seccomp syscall!"); 2310 } 2311 EXPECT_EQ(0, ret) { 2312 TH_LOG("Could not install filter!"); 2313 } 2314 2315 /* Make sure neither entry point will switch to strict. */ 2316 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0); 2317 EXPECT_EQ(EINVAL, errno) { 2318 TH_LOG("Switched to mode strict!"); 2319 } 2320 2321 ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL); 2322 EXPECT_EQ(EINVAL, errno) { 2323 TH_LOG("Switched to mode strict!"); 2324 } 2325} 2326 2327/* 2328 * Test detection of known and unknown filter flags. Userspace needs to be able 2329 * to check if a filter flag is supported by the current kernel and a good way 2330 * of doing that is by attempting to enter filter mode, with the flag bit in 2331 * question set, and a NULL pointer for the _args_ parameter. EFAULT indicates 2332 * that the flag is valid and EINVAL indicates that the flag is invalid. 2333 */ 2334TEST(detect_seccomp_filter_flags) 2335{ 2336 unsigned int flags[] = { SECCOMP_FILTER_FLAG_TSYNC, 2337 SECCOMP_FILTER_FLAG_LOG, 2338 SECCOMP_FILTER_FLAG_SPEC_ALLOW, 2339 SECCOMP_FILTER_FLAG_NEW_LISTENER, 2340 SECCOMP_FILTER_FLAG_TSYNC_ESRCH }; 2341 unsigned int exclusive[] = { 2342 SECCOMP_FILTER_FLAG_TSYNC, 2343 SECCOMP_FILTER_FLAG_NEW_LISTENER }; 2344 unsigned int flag, all_flags, exclusive_mask; 2345 int i; 2346 long ret; 2347 2348 /* Test detection of individual known-good filter flags */ 2349 for (i = 0, all_flags = 0; i < ARRAY_SIZE(flags); i++) { 2350 int bits = 0; 2351 2352 flag = flags[i]; 2353 /* Make sure the flag is a single bit! */ 2354 while (flag) { 2355 if (flag & 0x1) 2356 bits ++; 2357 flag >>= 1; 2358 } 2359 ASSERT_EQ(1, bits); 2360 flag = flags[i]; 2361 2362 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2363 ASSERT_NE(ENOSYS, errno) { 2364 TH_LOG("Kernel does not support seccomp syscall!"); 2365 } 2366 EXPECT_EQ(-1, ret); 2367 EXPECT_EQ(EFAULT, errno) { 2368 TH_LOG("Failed to detect that a known-good filter flag (0x%X) is supported!", 2369 flag); 2370 } 2371 2372 all_flags |= flag; 2373 } 2374 2375 /* 2376 * Test detection of all known-good filter flags combined. But 2377 * for the exclusive flags we need to mask them out and try them 2378 * individually for the "all flags" testing. 2379 */ 2380 exclusive_mask = 0; 2381 for (i = 0; i < ARRAY_SIZE(exclusive); i++) 2382 exclusive_mask |= exclusive[i]; 2383 for (i = 0; i < ARRAY_SIZE(exclusive); i++) { 2384 flag = all_flags & ~exclusive_mask; 2385 flag |= exclusive[i]; 2386 2387 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2388 EXPECT_EQ(-1, ret); 2389 EXPECT_EQ(EFAULT, errno) { 2390 TH_LOG("Failed to detect that all known-good filter flags (0x%X) are supported!", 2391 flag); 2392 } 2393 } 2394 2395 /* Test detection of an unknown filter flags, without exclusives. */ 2396 flag = -1; 2397 flag &= ~exclusive_mask; 2398 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2399 EXPECT_EQ(-1, ret); 2400 EXPECT_EQ(EINVAL, errno) { 2401 TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported!", 2402 flag); 2403 } 2404 2405 /* 2406 * Test detection of an unknown filter flag that may simply need to be 2407 * added to this test 2408 */ 2409 flag = flags[ARRAY_SIZE(flags) - 1] << 1; 2410 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL); 2411 EXPECT_EQ(-1, ret); 2412 EXPECT_EQ(EINVAL, errno) { 2413 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?", 2414 flag); 2415 } 2416} 2417 2418TEST(TSYNC_first) 2419{ 2420 struct sock_filter filter[] = { 2421 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2422 }; 2423 struct sock_fprog prog = { 2424 .len = (unsigned short)ARRAY_SIZE(filter), 2425 .filter = filter, 2426 }; 2427 long ret; 2428 2429 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0); 2430 ASSERT_EQ(0, ret) { 2431 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2432 } 2433 2434 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2435 &prog); 2436 ASSERT_NE(ENOSYS, errno) { 2437 TH_LOG("Kernel does not support seccomp syscall!"); 2438 } 2439 EXPECT_EQ(0, ret) { 2440 TH_LOG("Could not install initial filter with TSYNC!"); 2441 } 2442} 2443 2444#define TSYNC_SIBLINGS 2 2445struct tsync_sibling { 2446 pthread_t tid; 2447 pid_t system_tid; 2448 sem_t *started; 2449 pthread_cond_t *cond; 2450 pthread_mutex_t *mutex; 2451 int diverge; 2452 int num_waits; 2453 struct sock_fprog *prog; 2454 struct __test_metadata *metadata; 2455}; 2456 2457/* 2458 * To avoid joining joined threads (which is not allowed by Bionic), 2459 * make sure we both successfully join and clear the tid to skip a 2460 * later join attempt during fixture teardown. Any remaining threads 2461 * will be directly killed during teardown. 2462 */ 2463#define PTHREAD_JOIN(tid, status) \ 2464 do { \ 2465 int _rc = pthread_join(tid, status); \ 2466 if (_rc) { \ 2467 TH_LOG("pthread_join of tid %u failed: %d\n", \ 2468 (unsigned int)tid, _rc); \ 2469 } else { \ 2470 tid = 0; \ 2471 } \ 2472 } while (0) 2473 2474FIXTURE(TSYNC) { 2475 struct sock_fprog root_prog, apply_prog; 2476 struct tsync_sibling sibling[TSYNC_SIBLINGS]; 2477 sem_t started; 2478 pthread_cond_t cond; 2479 pthread_mutex_t mutex; 2480 int sibling_count; 2481}; 2482 2483FIXTURE_SETUP(TSYNC) 2484{ 2485 struct sock_filter root_filter[] = { 2486 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2487 }; 2488 struct sock_filter apply_filter[] = { 2489 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2490 offsetof(struct seccomp_data, nr)), 2491 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1), 2492 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2493 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2494 }; 2495 2496 memset(&self->root_prog, 0, sizeof(self->root_prog)); 2497 memset(&self->apply_prog, 0, sizeof(self->apply_prog)); 2498 memset(&self->sibling, 0, sizeof(self->sibling)); 2499 self->root_prog.filter = malloc(sizeof(root_filter)); 2500 ASSERT_NE(NULL, self->root_prog.filter); 2501 memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter)); 2502 self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter); 2503 2504 self->apply_prog.filter = malloc(sizeof(apply_filter)); 2505 ASSERT_NE(NULL, self->apply_prog.filter); 2506 memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter)); 2507 self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter); 2508 2509 self->sibling_count = 0; 2510 pthread_mutex_init(&self->mutex, NULL); 2511 pthread_cond_init(&self->cond, NULL); 2512 sem_init(&self->started, 0, 0); 2513 self->sibling[0].tid = 0; 2514 self->sibling[0].cond = &self->cond; 2515 self->sibling[0].started = &self->started; 2516 self->sibling[0].mutex = &self->mutex; 2517 self->sibling[0].diverge = 0; 2518 self->sibling[0].num_waits = 1; 2519 self->sibling[0].prog = &self->root_prog; 2520 self->sibling[0].metadata = _metadata; 2521 self->sibling[1].tid = 0; 2522 self->sibling[1].cond = &self->cond; 2523 self->sibling[1].started = &self->started; 2524 self->sibling[1].mutex = &self->mutex; 2525 self->sibling[1].diverge = 0; 2526 self->sibling[1].prog = &self->root_prog; 2527 self->sibling[1].num_waits = 1; 2528 self->sibling[1].metadata = _metadata; 2529} 2530 2531FIXTURE_TEARDOWN(TSYNC) 2532{ 2533 int sib = 0; 2534 2535 if (self->root_prog.filter) 2536 free(self->root_prog.filter); 2537 if (self->apply_prog.filter) 2538 free(self->apply_prog.filter); 2539 2540 for ( ; sib < self->sibling_count; ++sib) { 2541 struct tsync_sibling *s = &self->sibling[sib]; 2542 2543 if (!s->tid) 2544 continue; 2545 /* 2546 * If a thread is still running, it may be stuck, so hit 2547 * it over the head really hard. 2548 */ 2549 pthread_kill(s->tid, 9); 2550 } 2551 pthread_mutex_destroy(&self->mutex); 2552 pthread_cond_destroy(&self->cond); 2553 sem_destroy(&self->started); 2554} 2555 2556void *tsync_sibling(void *data) 2557{ 2558 long ret = 0; 2559 struct tsync_sibling *me = data; 2560 2561 me->system_tid = syscall(__NR_gettid); 2562 2563 pthread_mutex_lock(me->mutex); 2564 if (me->diverge) { 2565 /* Just re-apply the root prog to fork the tree */ 2566 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, 2567 me->prog, 0, 0); 2568 } 2569 sem_post(me->started); 2570 /* Return outside of started so parent notices failures. */ 2571 if (ret) { 2572 pthread_mutex_unlock(me->mutex); 2573 return (void *)SIBLING_EXIT_FAILURE; 2574 } 2575 do { 2576 pthread_cond_wait(me->cond, me->mutex); 2577 me->num_waits = me->num_waits - 1; 2578 } while (me->num_waits); 2579 pthread_mutex_unlock(me->mutex); 2580 2581 ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0); 2582 if (!ret) 2583 return (void *)SIBLING_EXIT_NEWPRIVS; 2584 read(-1, NULL, 0); 2585 return (void *)SIBLING_EXIT_UNKILLED; 2586} 2587 2588void tsync_start_sibling(struct tsync_sibling *sibling) 2589{ 2590 pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling); 2591} 2592 2593TEST_F(TSYNC, siblings_fail_prctl) 2594{ 2595 long ret; 2596 void *status; 2597 struct sock_filter filter[] = { 2598 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2599 offsetof(struct seccomp_data, nr)), 2600 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1), 2601 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL), 2602 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2603 }; 2604 struct sock_fprog prog = { 2605 .len = (unsigned short)ARRAY_SIZE(filter), 2606 .filter = filter, 2607 }; 2608 2609 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2610 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2611 } 2612 2613 /* Check prctl failure detection by requesting sib 0 diverge. */ 2614 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog); 2615 ASSERT_NE(ENOSYS, errno) { 2616 TH_LOG("Kernel does not support seccomp syscall!"); 2617 } 2618 ASSERT_EQ(0, ret) { 2619 TH_LOG("setting filter failed"); 2620 } 2621 2622 self->sibling[0].diverge = 1; 2623 tsync_start_sibling(&self->sibling[0]); 2624 tsync_start_sibling(&self->sibling[1]); 2625 2626 while (self->sibling_count < TSYNC_SIBLINGS) { 2627 sem_wait(&self->started); 2628 self->sibling_count++; 2629 } 2630 2631 /* Signal the threads to clean up*/ 2632 pthread_mutex_lock(&self->mutex); 2633 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2634 TH_LOG("cond broadcast non-zero"); 2635 } 2636 pthread_mutex_unlock(&self->mutex); 2637 2638 /* Ensure diverging sibling failed to call prctl. */ 2639 PTHREAD_JOIN(self->sibling[0].tid, &status); 2640 EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status); 2641 PTHREAD_JOIN(self->sibling[1].tid, &status); 2642 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2643} 2644 2645TEST_F(TSYNC, two_siblings_with_ancestor) 2646{ 2647 long ret; 2648 void *status; 2649 2650 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2651 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2652 } 2653 2654 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2655 ASSERT_NE(ENOSYS, errno) { 2656 TH_LOG("Kernel does not support seccomp syscall!"); 2657 } 2658 ASSERT_EQ(0, ret) { 2659 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2660 } 2661 tsync_start_sibling(&self->sibling[0]); 2662 tsync_start_sibling(&self->sibling[1]); 2663 2664 while (self->sibling_count < TSYNC_SIBLINGS) { 2665 sem_wait(&self->started); 2666 self->sibling_count++; 2667 } 2668 2669 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2670 &self->apply_prog); 2671 ASSERT_EQ(0, ret) { 2672 TH_LOG("Could install filter on all threads!"); 2673 } 2674 /* Tell the siblings to test the policy */ 2675 pthread_mutex_lock(&self->mutex); 2676 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2677 TH_LOG("cond broadcast non-zero"); 2678 } 2679 pthread_mutex_unlock(&self->mutex); 2680 /* Ensure they are both killed and don't exit cleanly. */ 2681 PTHREAD_JOIN(self->sibling[0].tid, &status); 2682 EXPECT_EQ(0x0, (long)status); 2683 PTHREAD_JOIN(self->sibling[1].tid, &status); 2684 EXPECT_EQ(0x0, (long)status); 2685} 2686 2687TEST_F(TSYNC, two_sibling_want_nnp) 2688{ 2689 void *status; 2690 2691 /* start siblings before any prctl() operations */ 2692 tsync_start_sibling(&self->sibling[0]); 2693 tsync_start_sibling(&self->sibling[1]); 2694 while (self->sibling_count < TSYNC_SIBLINGS) { 2695 sem_wait(&self->started); 2696 self->sibling_count++; 2697 } 2698 2699 /* Tell the siblings to test no policy */ 2700 pthread_mutex_lock(&self->mutex); 2701 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2702 TH_LOG("cond broadcast non-zero"); 2703 } 2704 pthread_mutex_unlock(&self->mutex); 2705 2706 /* Ensure they are both upset about lacking nnp. */ 2707 PTHREAD_JOIN(self->sibling[0].tid, &status); 2708 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status); 2709 PTHREAD_JOIN(self->sibling[1].tid, &status); 2710 EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status); 2711} 2712 2713TEST_F(TSYNC, two_siblings_with_no_filter) 2714{ 2715 long ret; 2716 void *status; 2717 2718 /* start siblings before any prctl() operations */ 2719 tsync_start_sibling(&self->sibling[0]); 2720 tsync_start_sibling(&self->sibling[1]); 2721 while (self->sibling_count < TSYNC_SIBLINGS) { 2722 sem_wait(&self->started); 2723 self->sibling_count++; 2724 } 2725 2726 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2727 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2728 } 2729 2730 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2731 &self->apply_prog); 2732 ASSERT_NE(ENOSYS, errno) { 2733 TH_LOG("Kernel does not support seccomp syscall!"); 2734 } 2735 ASSERT_EQ(0, ret) { 2736 TH_LOG("Could install filter on all threads!"); 2737 } 2738 2739 /* Tell the siblings to test the policy */ 2740 pthread_mutex_lock(&self->mutex); 2741 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2742 TH_LOG("cond broadcast non-zero"); 2743 } 2744 pthread_mutex_unlock(&self->mutex); 2745 2746 /* Ensure they are both killed and don't exit cleanly. */ 2747 PTHREAD_JOIN(self->sibling[0].tid, &status); 2748 EXPECT_EQ(0x0, (long)status); 2749 PTHREAD_JOIN(self->sibling[1].tid, &status); 2750 EXPECT_EQ(0x0, (long)status); 2751} 2752 2753TEST_F(TSYNC, two_siblings_with_one_divergence) 2754{ 2755 long ret; 2756 void *status; 2757 2758 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2759 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2760 } 2761 2762 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2763 ASSERT_NE(ENOSYS, errno) { 2764 TH_LOG("Kernel does not support seccomp syscall!"); 2765 } 2766 ASSERT_EQ(0, ret) { 2767 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2768 } 2769 self->sibling[0].diverge = 1; 2770 tsync_start_sibling(&self->sibling[0]); 2771 tsync_start_sibling(&self->sibling[1]); 2772 2773 while (self->sibling_count < TSYNC_SIBLINGS) { 2774 sem_wait(&self->started); 2775 self->sibling_count++; 2776 } 2777 2778 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2779 &self->apply_prog); 2780 ASSERT_EQ(self->sibling[0].system_tid, ret) { 2781 TH_LOG("Did not fail on diverged sibling."); 2782 } 2783 2784 /* Wake the threads */ 2785 pthread_mutex_lock(&self->mutex); 2786 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2787 TH_LOG("cond broadcast non-zero"); 2788 } 2789 pthread_mutex_unlock(&self->mutex); 2790 2791 /* Ensure they are both unkilled. */ 2792 PTHREAD_JOIN(self->sibling[0].tid, &status); 2793 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2794 PTHREAD_JOIN(self->sibling[1].tid, &status); 2795 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2796} 2797 2798TEST_F(TSYNC, two_siblings_with_one_divergence_no_tid_in_err) 2799{ 2800 long ret, flags; 2801 void *status; 2802 2803 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2804 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2805 } 2806 2807 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2808 ASSERT_NE(ENOSYS, errno) { 2809 TH_LOG("Kernel does not support seccomp syscall!"); 2810 } 2811 ASSERT_EQ(0, ret) { 2812 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2813 } 2814 self->sibling[0].diverge = 1; 2815 tsync_start_sibling(&self->sibling[0]); 2816 tsync_start_sibling(&self->sibling[1]); 2817 2818 while (self->sibling_count < TSYNC_SIBLINGS) { 2819 sem_wait(&self->started); 2820 self->sibling_count++; 2821 } 2822 2823 flags = SECCOMP_FILTER_FLAG_TSYNC | \ 2824 SECCOMP_FILTER_FLAG_TSYNC_ESRCH; 2825 ret = seccomp(SECCOMP_SET_MODE_FILTER, flags, &self->apply_prog); 2826 ASSERT_EQ(ESRCH, errno) { 2827 TH_LOG("Did not return ESRCH for diverged sibling."); 2828 } 2829 ASSERT_EQ(-1, ret) { 2830 TH_LOG("Did not fail on diverged sibling."); 2831 } 2832 2833 /* Wake the threads */ 2834 pthread_mutex_lock(&self->mutex); 2835 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2836 TH_LOG("cond broadcast non-zero"); 2837 } 2838 pthread_mutex_unlock(&self->mutex); 2839 2840 /* Ensure they are both unkilled. */ 2841 PTHREAD_JOIN(self->sibling[0].tid, &status); 2842 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2843 PTHREAD_JOIN(self->sibling[1].tid, &status); 2844 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2845} 2846 2847TEST_F(TSYNC, two_siblings_not_under_filter) 2848{ 2849 long ret, sib; 2850 void *status; 2851 struct timespec delay = { .tv_nsec = 100000000 }; 2852 2853 ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2854 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2855 } 2856 2857 /* 2858 * Sibling 0 will have its own seccomp policy 2859 * and Sibling 1 will not be under seccomp at 2860 * all. Sibling 1 will enter seccomp and 0 2861 * will cause failure. 2862 */ 2863 self->sibling[0].diverge = 1; 2864 tsync_start_sibling(&self->sibling[0]); 2865 tsync_start_sibling(&self->sibling[1]); 2866 2867 while (self->sibling_count < TSYNC_SIBLINGS) { 2868 sem_wait(&self->started); 2869 self->sibling_count++; 2870 } 2871 2872 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog); 2873 ASSERT_NE(ENOSYS, errno) { 2874 TH_LOG("Kernel does not support seccomp syscall!"); 2875 } 2876 ASSERT_EQ(0, ret) { 2877 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!"); 2878 } 2879 2880 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2881 &self->apply_prog); 2882 ASSERT_EQ(ret, self->sibling[0].system_tid) { 2883 TH_LOG("Did not fail on diverged sibling."); 2884 } 2885 sib = 1; 2886 if (ret == self->sibling[0].system_tid) 2887 sib = 0; 2888 2889 pthread_mutex_lock(&self->mutex); 2890 2891 /* Increment the other siblings num_waits so we can clean up 2892 * the one we just saw. 2893 */ 2894 self->sibling[!sib].num_waits += 1; 2895 2896 /* Signal the thread to clean up*/ 2897 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2898 TH_LOG("cond broadcast non-zero"); 2899 } 2900 pthread_mutex_unlock(&self->mutex); 2901 PTHREAD_JOIN(self->sibling[sib].tid, &status); 2902 EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status); 2903 /* Poll for actual task death. pthread_join doesn't guarantee it. */ 2904 while (!kill(self->sibling[sib].system_tid, 0)) 2905 nanosleep(&delay, NULL); 2906 /* Switch to the remaining sibling */ 2907 sib = !sib; 2908 2909 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2910 &self->apply_prog); 2911 ASSERT_EQ(0, ret) { 2912 TH_LOG("Expected the remaining sibling to sync"); 2913 }; 2914 2915 pthread_mutex_lock(&self->mutex); 2916 2917 /* If remaining sibling didn't have a chance to wake up during 2918 * the first broadcast, manually reduce the num_waits now. 2919 */ 2920 if (self->sibling[sib].num_waits > 1) 2921 self->sibling[sib].num_waits = 1; 2922 ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) { 2923 TH_LOG("cond broadcast non-zero"); 2924 } 2925 pthread_mutex_unlock(&self->mutex); 2926 PTHREAD_JOIN(self->sibling[sib].tid, &status); 2927 EXPECT_EQ(0, (long)status); 2928 /* Poll for actual task death. pthread_join doesn't guarantee it. */ 2929 while (!kill(self->sibling[sib].system_tid, 0)) 2930 nanosleep(&delay, NULL); 2931 2932 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC, 2933 &self->apply_prog); 2934 ASSERT_EQ(0, ret); /* just us chickens */ 2935} 2936 2937/* Make sure restarted syscalls are seen directly as "restart_syscall". */ 2938TEST(syscall_restart) 2939{ 2940 long ret; 2941 unsigned long msg; 2942 pid_t child_pid; 2943 int pipefd[2]; 2944 int status; 2945 siginfo_t info = { }; 2946 struct sock_filter filter[] = { 2947 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2948 offsetof(struct seccomp_data, nr)), 2949 2950#ifdef __NR_sigreturn 2951 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 7, 0), 2952#endif 2953 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 6, 0), 2954 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 5, 0), 2955 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 4, 0), 2956 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 5, 0), 2957 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_clock_nanosleep, 4, 0), 2958 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0), 2959 2960 /* Allow __NR_write for easy logging. */ 2961 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1), 2962 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 2963 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 2964 /* The nanosleep jump target. */ 2965 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100), 2966 /* The restart_syscall jump target. */ 2967 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200), 2968 }; 2969 struct sock_fprog prog = { 2970 .len = (unsigned short)ARRAY_SIZE(filter), 2971 .filter = filter, 2972 }; 2973#if defined(__arm__) 2974 struct utsname utsbuf; 2975#endif 2976 2977 ASSERT_EQ(0, pipe(pipefd)); 2978 2979 child_pid = fork(); 2980 ASSERT_LE(0, child_pid); 2981 if (child_pid == 0) { 2982 /* Child uses EXPECT not ASSERT to deliver status correctly. */ 2983 char buf = ' '; 2984 struct timespec timeout = { }; 2985 2986 /* Attach parent as tracer and stop. */ 2987 EXPECT_EQ(0, ptrace(PTRACE_TRACEME)); 2988 EXPECT_EQ(0, raise(SIGSTOP)); 2989 2990 EXPECT_EQ(0, close(pipefd[1])); 2991 2992 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) { 2993 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 2994 } 2995 2996 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0); 2997 EXPECT_EQ(0, ret) { 2998 TH_LOG("Failed to install filter!"); 2999 } 3000 3001 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) { 3002 TH_LOG("Failed to read() sync from parent"); 3003 } 3004 EXPECT_EQ('.', buf) { 3005 TH_LOG("Failed to get sync data from read()"); 3006 } 3007 3008 /* Start nanosleep to be interrupted. */ 3009 timeout.tv_sec = 1; 3010 errno = 0; 3011 EXPECT_EQ(0, nanosleep(&timeout, NULL)) { 3012 TH_LOG("Call to nanosleep() failed (errno %d)", errno); 3013 } 3014 3015 /* Read final sync from parent. */ 3016 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) { 3017 TH_LOG("Failed final read() from parent"); 3018 } 3019 EXPECT_EQ('!', buf) { 3020 TH_LOG("Failed to get final data from read()"); 3021 } 3022 3023 /* Directly report the status of our test harness results. */ 3024 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS 3025 : EXIT_FAILURE); 3026 } 3027 EXPECT_EQ(0, close(pipefd[0])); 3028 3029 /* Attach to child, setup options, and release. */ 3030 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3031 ASSERT_EQ(true, WIFSTOPPED(status)); 3032 ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL, 3033 PTRACE_O_TRACESECCOMP)); 3034 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3035 ASSERT_EQ(1, write(pipefd[1], ".", 1)); 3036 3037 /* Wait for nanosleep() to start. */ 3038 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3039 ASSERT_EQ(true, WIFSTOPPED(status)); 3040 ASSERT_EQ(SIGTRAP, WSTOPSIG(status)); 3041 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16)); 3042 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg)); 3043 ASSERT_EQ(0x100, msg); 3044 ret = get_syscall(_metadata, child_pid); 3045 EXPECT_TRUE(ret == __NR_nanosleep || ret == __NR_clock_nanosleep); 3046 3047 /* Might as well check siginfo for sanity while we're here. */ 3048 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info)); 3049 ASSERT_EQ(SIGTRAP, info.si_signo); 3050 ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code); 3051 EXPECT_EQ(0, info.si_errno); 3052 EXPECT_EQ(getuid(), info.si_uid); 3053 /* Verify signal delivery came from child (seccomp-triggered). */ 3054 EXPECT_EQ(child_pid, info.si_pid); 3055 3056 /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */ 3057 ASSERT_EQ(0, kill(child_pid, SIGSTOP)); 3058 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3059 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3060 ASSERT_EQ(true, WIFSTOPPED(status)); 3061 ASSERT_EQ(SIGSTOP, WSTOPSIG(status)); 3062 ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info)); 3063 /* 3064 * There is no siginfo on SIGSTOP any more, so we can't verify 3065 * signal delivery came from parent now (getpid() == info.si_pid). 3066 * https://lkml.kernel.org/r/CAGXu5jJaZAOzP1qFz66tYrtbuywqb+UN2SOA1VLHpCCOiYvYeg@mail.gmail.com 3067 * At least verify the SIGSTOP via PTRACE_GETSIGINFO. 3068 */ 3069 EXPECT_EQ(SIGSTOP, info.si_signo); 3070 3071 /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */ 3072 ASSERT_EQ(0, kill(child_pid, SIGCONT)); 3073 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3074 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3075 ASSERT_EQ(true, WIFSTOPPED(status)); 3076 ASSERT_EQ(SIGCONT, WSTOPSIG(status)); 3077 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3078 3079 /* Wait for restart_syscall() to start. */ 3080 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3081 ASSERT_EQ(true, WIFSTOPPED(status)); 3082 ASSERT_EQ(SIGTRAP, WSTOPSIG(status)); 3083 ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16)); 3084 ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg)); 3085 3086 ASSERT_EQ(0x200, msg); 3087 ret = get_syscall(_metadata, child_pid); 3088#if defined(__arm__) 3089 /* 3090 * FIXME: 3091 * - native ARM registers do NOT expose true syscall. 3092 * - compat ARM registers on ARM64 DO expose true syscall. 3093 */ 3094 ASSERT_EQ(0, uname(&utsbuf)); 3095 if (strncmp(utsbuf.machine, "arm", 3) == 0) { 3096 EXPECT_EQ(__NR_nanosleep, ret); 3097 } else 3098#endif 3099 { 3100 EXPECT_EQ(__NR_restart_syscall, ret); 3101 } 3102 3103 /* Write again to end test. */ 3104 ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0)); 3105 ASSERT_EQ(1, write(pipefd[1], "!", 1)); 3106 EXPECT_EQ(0, close(pipefd[1])); 3107 3108 ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0)); 3109 if (WIFSIGNALED(status) || WEXITSTATUS(status)) 3110 _metadata->passed = 0; 3111} 3112 3113TEST_SIGNAL(filter_flag_log, SIGSYS) 3114{ 3115 struct sock_filter allow_filter[] = { 3116 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3117 }; 3118 struct sock_filter kill_filter[] = { 3119 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 3120 offsetof(struct seccomp_data, nr)), 3121 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1), 3122 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL), 3123 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3124 }; 3125 struct sock_fprog allow_prog = { 3126 .len = (unsigned short)ARRAY_SIZE(allow_filter), 3127 .filter = allow_filter, 3128 }; 3129 struct sock_fprog kill_prog = { 3130 .len = (unsigned short)ARRAY_SIZE(kill_filter), 3131 .filter = kill_filter, 3132 }; 3133 long ret; 3134 pid_t parent = getppid(); 3135 3136 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3137 ASSERT_EQ(0, ret); 3138 3139 /* Verify that the FILTER_FLAG_LOG flag isn't accepted in strict mode */ 3140 ret = seccomp(SECCOMP_SET_MODE_STRICT, SECCOMP_FILTER_FLAG_LOG, 3141 &allow_prog); 3142 ASSERT_NE(ENOSYS, errno) { 3143 TH_LOG("Kernel does not support seccomp syscall!"); 3144 } 3145 EXPECT_NE(0, ret) { 3146 TH_LOG("Kernel accepted FILTER_FLAG_LOG flag in strict mode!"); 3147 } 3148 EXPECT_EQ(EINVAL, errno) { 3149 TH_LOG("Kernel returned unexpected errno for FILTER_FLAG_LOG flag in strict mode!"); 3150 } 3151 3152 /* Verify that a simple, permissive filter can be added with no flags */ 3153 ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog); 3154 EXPECT_EQ(0, ret); 3155 3156 /* See if the same filter can be added with the FILTER_FLAG_LOG flag */ 3157 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG, 3158 &allow_prog); 3159 ASSERT_NE(EINVAL, errno) { 3160 TH_LOG("Kernel does not support the FILTER_FLAG_LOG flag!"); 3161 } 3162 EXPECT_EQ(0, ret); 3163 3164 /* Ensure that the kill filter works with the FILTER_FLAG_LOG flag */ 3165 ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG, 3166 &kill_prog); 3167 EXPECT_EQ(0, ret); 3168 3169 EXPECT_EQ(parent, syscall(__NR_getppid)); 3170 /* getpid() should never return. */ 3171 EXPECT_EQ(0, syscall(__NR_getpid)); 3172} 3173 3174TEST(get_action_avail) 3175{ 3176 __u32 actions[] = { SECCOMP_RET_KILL_THREAD, SECCOMP_RET_TRAP, 3177 SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE, 3178 SECCOMP_RET_LOG, SECCOMP_RET_ALLOW }; 3179 __u32 unknown_action = 0x10000000U; 3180 int i; 3181 long ret; 3182 3183 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[0]); 3184 ASSERT_NE(ENOSYS, errno) { 3185 TH_LOG("Kernel does not support seccomp syscall!"); 3186 } 3187 ASSERT_NE(EINVAL, errno) { 3188 TH_LOG("Kernel does not support SECCOMP_GET_ACTION_AVAIL operation!"); 3189 } 3190 EXPECT_EQ(ret, 0); 3191 3192 for (i = 0; i < ARRAY_SIZE(actions); i++) { 3193 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[i]); 3194 EXPECT_EQ(ret, 0) { 3195 TH_LOG("Expected action (0x%X) not available!", 3196 actions[i]); 3197 } 3198 } 3199 3200 /* Check that an unknown action is handled properly (EOPNOTSUPP) */ 3201 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &unknown_action); 3202 EXPECT_EQ(ret, -1); 3203 EXPECT_EQ(errno, EOPNOTSUPP); 3204} 3205 3206TEST(get_metadata) 3207{ 3208 pid_t pid; 3209 int pipefd[2]; 3210 char buf; 3211 struct seccomp_metadata md; 3212 long ret; 3213 3214 /* Only real root can get metadata. */ 3215 if (geteuid()) { 3216 SKIP(return, "get_metadata requires real root"); 3217 return; 3218 } 3219 3220 ASSERT_EQ(0, pipe(pipefd)); 3221 3222 pid = fork(); 3223 ASSERT_GE(pid, 0); 3224 if (pid == 0) { 3225 struct sock_filter filter[] = { 3226 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3227 }; 3228 struct sock_fprog prog = { 3229 .len = (unsigned short)ARRAY_SIZE(filter), 3230 .filter = filter, 3231 }; 3232 3233 /* one with log, one without */ 3234 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 3235 SECCOMP_FILTER_FLAG_LOG, &prog)); 3236 EXPECT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog)); 3237 3238 EXPECT_EQ(0, close(pipefd[0])); 3239 ASSERT_EQ(1, write(pipefd[1], "1", 1)); 3240 ASSERT_EQ(0, close(pipefd[1])); 3241 3242 while (1) 3243 sleep(100); 3244 } 3245 3246 ASSERT_EQ(0, close(pipefd[1])); 3247 ASSERT_EQ(1, read(pipefd[0], &buf, 1)); 3248 3249 ASSERT_EQ(0, ptrace(PTRACE_ATTACH, pid)); 3250 ASSERT_EQ(pid, waitpid(pid, NULL, 0)); 3251 3252 /* Past here must not use ASSERT or child process is never killed. */ 3253 3254 md.filter_off = 0; 3255 errno = 0; 3256 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md); 3257 EXPECT_EQ(sizeof(md), ret) { 3258 if (errno == EINVAL) 3259 SKIP(goto skip, "Kernel does not support PTRACE_SECCOMP_GET_METADATA (missing CONFIG_CHECKPOINT_RESTORE?)"); 3260 } 3261 3262 EXPECT_EQ(md.flags, SECCOMP_FILTER_FLAG_LOG); 3263 EXPECT_EQ(md.filter_off, 0); 3264 3265 md.filter_off = 1; 3266 ret = ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md); 3267 EXPECT_EQ(sizeof(md), ret); 3268 EXPECT_EQ(md.flags, 0); 3269 EXPECT_EQ(md.filter_off, 1); 3270 3271skip: 3272 ASSERT_EQ(0, kill(pid, SIGKILL)); 3273} 3274 3275static int user_notif_syscall(int nr, unsigned int flags) 3276{ 3277 struct sock_filter filter[] = { 3278 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 3279 offsetof(struct seccomp_data, nr)), 3280 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, nr, 0, 1), 3281 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_USER_NOTIF), 3282 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3283 }; 3284 3285 struct sock_fprog prog = { 3286 .len = (unsigned short)ARRAY_SIZE(filter), 3287 .filter = filter, 3288 }; 3289 3290 return seccomp(SECCOMP_SET_MODE_FILTER, flags, &prog); 3291} 3292 3293#define USER_NOTIF_MAGIC INT_MAX 3294TEST(user_notification_basic) 3295{ 3296 pid_t pid; 3297 long ret; 3298 int status, listener; 3299 struct seccomp_notif req = {}; 3300 struct seccomp_notif_resp resp = {}; 3301 struct pollfd pollfd; 3302 3303 struct sock_filter filter[] = { 3304 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW), 3305 }; 3306 struct sock_fprog prog = { 3307 .len = (unsigned short)ARRAY_SIZE(filter), 3308 .filter = filter, 3309 }; 3310 3311 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3312 ASSERT_EQ(0, ret) { 3313 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3314 } 3315 3316 pid = fork(); 3317 ASSERT_GE(pid, 0); 3318 3319 /* Check that we get -ENOSYS with no listener attached */ 3320 if (pid == 0) { 3321 if (user_notif_syscall(__NR_getppid, 0) < 0) 3322 exit(1); 3323 ret = syscall(__NR_getppid); 3324 exit(ret >= 0 || errno != ENOSYS); 3325 } 3326 3327 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3328 EXPECT_EQ(true, WIFEXITED(status)); 3329 EXPECT_EQ(0, WEXITSTATUS(status)); 3330 3331 /* Add some no-op filters for grins. */ 3332 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3333 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3334 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3335 EXPECT_EQ(seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog), 0); 3336 3337 /* Check that the basic notification machinery works */ 3338 listener = user_notif_syscall(__NR_getppid, 3339 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3340 ASSERT_GE(listener, 0); 3341 3342 /* Installing a second listener in the chain should EBUSY */ 3343 EXPECT_EQ(user_notif_syscall(__NR_getppid, 3344 SECCOMP_FILTER_FLAG_NEW_LISTENER), 3345 -1); 3346 EXPECT_EQ(errno, EBUSY); 3347 3348 pid = fork(); 3349 ASSERT_GE(pid, 0); 3350 3351 if (pid == 0) { 3352 ret = syscall(__NR_getppid); 3353 exit(ret != USER_NOTIF_MAGIC); 3354 } 3355 3356 pollfd.fd = listener; 3357 pollfd.events = POLLIN | POLLOUT; 3358 3359 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3360 EXPECT_EQ(pollfd.revents, POLLIN); 3361 3362 /* Test that we can't pass garbage to the kernel. */ 3363 memset(&req, 0, sizeof(req)); 3364 req.pid = -1; 3365 errno = 0; 3366 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req); 3367 EXPECT_EQ(-1, ret); 3368 EXPECT_EQ(EINVAL, errno); 3369 3370 if (ret) { 3371 req.pid = 0; 3372 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3373 } 3374 3375 pollfd.fd = listener; 3376 pollfd.events = POLLIN | POLLOUT; 3377 3378 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3379 EXPECT_EQ(pollfd.revents, POLLOUT); 3380 3381 EXPECT_EQ(req.data.nr, __NR_getppid); 3382 3383 resp.id = req.id; 3384 resp.error = 0; 3385 resp.val = USER_NOTIF_MAGIC; 3386 3387 /* check that we make sure flags == 0 */ 3388 resp.flags = 1; 3389 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3390 EXPECT_EQ(errno, EINVAL); 3391 3392 resp.flags = 0; 3393 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3394 3395 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3396 EXPECT_EQ(true, WIFEXITED(status)); 3397 EXPECT_EQ(0, WEXITSTATUS(status)); 3398} 3399 3400TEST(user_notification_with_tsync) 3401{ 3402 int ret; 3403 unsigned int flags; 3404 3405 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3406 ASSERT_EQ(0, ret) { 3407 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3408 } 3409 3410 /* these were exclusive */ 3411 flags = SECCOMP_FILTER_FLAG_NEW_LISTENER | 3412 SECCOMP_FILTER_FLAG_TSYNC; 3413 ASSERT_EQ(-1, user_notif_syscall(__NR_getppid, flags)); 3414 ASSERT_EQ(EINVAL, errno); 3415 3416 /* but now they're not */ 3417 flags |= SECCOMP_FILTER_FLAG_TSYNC_ESRCH; 3418 ret = user_notif_syscall(__NR_getppid, flags); 3419 close(ret); 3420 ASSERT_LE(0, ret); 3421} 3422 3423TEST(user_notification_kill_in_middle) 3424{ 3425 pid_t pid; 3426 long ret; 3427 int listener; 3428 struct seccomp_notif req = {}; 3429 struct seccomp_notif_resp resp = {}; 3430 3431 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3432 ASSERT_EQ(0, ret) { 3433 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3434 } 3435 3436 listener = user_notif_syscall(__NR_getppid, 3437 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3438 ASSERT_GE(listener, 0); 3439 3440 /* 3441 * Check that nothing bad happens when we kill the task in the middle 3442 * of a syscall. 3443 */ 3444 pid = fork(); 3445 ASSERT_GE(pid, 0); 3446 3447 if (pid == 0) { 3448 ret = syscall(__NR_getppid); 3449 exit(ret != USER_NOTIF_MAGIC); 3450 } 3451 3452 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3453 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), 0); 3454 3455 EXPECT_EQ(kill(pid, SIGKILL), 0); 3456 EXPECT_EQ(waitpid(pid, NULL, 0), pid); 3457 3458 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ID_VALID, &req.id), -1); 3459 3460 resp.id = req.id; 3461 ret = ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp); 3462 EXPECT_EQ(ret, -1); 3463 EXPECT_EQ(errno, ENOENT); 3464} 3465 3466static int handled = -1; 3467 3468static void signal_handler(int signal) 3469{ 3470 if (write(handled, "c", 1) != 1) 3471 perror("write from signal"); 3472} 3473 3474TEST(user_notification_signal) 3475{ 3476 pid_t pid; 3477 long ret; 3478 int status, listener, sk_pair[2]; 3479 struct seccomp_notif req = {}; 3480 struct seccomp_notif_resp resp = {}; 3481 char c; 3482 3483 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3484 ASSERT_EQ(0, ret) { 3485 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3486 } 3487 3488 ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0); 3489 3490 listener = user_notif_syscall(__NR_gettid, 3491 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3492 ASSERT_GE(listener, 0); 3493 3494 pid = fork(); 3495 ASSERT_GE(pid, 0); 3496 3497 if (pid == 0) { 3498 close(sk_pair[0]); 3499 handled = sk_pair[1]; 3500 if (signal(SIGUSR1, signal_handler) == SIG_ERR) { 3501 perror("signal"); 3502 exit(1); 3503 } 3504 /* 3505 * ERESTARTSYS behavior is a bit hard to test, because we need 3506 * to rely on a signal that has not yet been handled. Let's at 3507 * least check that the error code gets propagated through, and 3508 * hope that it doesn't break when there is actually a signal :) 3509 */ 3510 ret = syscall(__NR_gettid); 3511 exit(!(ret == -1 && errno == 512)); 3512 } 3513 3514 close(sk_pair[1]); 3515 3516 memset(&req, 0, sizeof(req)); 3517 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3518 3519 EXPECT_EQ(kill(pid, SIGUSR1), 0); 3520 3521 /* 3522 * Make sure the signal really is delivered, which means we're not 3523 * stuck in the user notification code any more and the notification 3524 * should be dead. 3525 */ 3526 EXPECT_EQ(read(sk_pair[0], &c, 1), 1); 3527 3528 resp.id = req.id; 3529 resp.error = -EPERM; 3530 resp.val = 0; 3531 3532 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3533 EXPECT_EQ(errno, ENOENT); 3534 3535 memset(&req, 0, sizeof(req)); 3536 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3537 3538 resp.id = req.id; 3539 resp.error = -512; /* -ERESTARTSYS */ 3540 resp.val = 0; 3541 3542 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3543 3544 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3545 EXPECT_EQ(true, WIFEXITED(status)); 3546 EXPECT_EQ(0, WEXITSTATUS(status)); 3547} 3548 3549TEST(user_notification_closed_listener) 3550{ 3551 pid_t pid; 3552 long ret; 3553 int status, listener; 3554 3555 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3556 ASSERT_EQ(0, ret) { 3557 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3558 } 3559 3560 listener = user_notif_syscall(__NR_getppid, 3561 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3562 ASSERT_GE(listener, 0); 3563 3564 /* 3565 * Check that we get an ENOSYS when the listener is closed. 3566 */ 3567 pid = fork(); 3568 ASSERT_GE(pid, 0); 3569 if (pid == 0) { 3570 close(listener); 3571 ret = syscall(__NR_getppid); 3572 exit(ret != -1 && errno != ENOSYS); 3573 } 3574 3575 close(listener); 3576 3577 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3578 EXPECT_EQ(true, WIFEXITED(status)); 3579 EXPECT_EQ(0, WEXITSTATUS(status)); 3580} 3581 3582/* 3583 * Check that a pid in a child namespace still shows up as valid in ours. 3584 */ 3585TEST(user_notification_child_pid_ns) 3586{ 3587 pid_t pid; 3588 int status, listener; 3589 struct seccomp_notif req = {}; 3590 struct seccomp_notif_resp resp = {}; 3591 3592 ASSERT_EQ(unshare(CLONE_NEWUSER | CLONE_NEWPID), 0) { 3593 if (errno == EINVAL) 3594 SKIP(return, "kernel missing CLONE_NEWUSER support"); 3595 }; 3596 3597 listener = user_notif_syscall(__NR_getppid, 3598 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3599 ASSERT_GE(listener, 0); 3600 3601 pid = fork(); 3602 ASSERT_GE(pid, 0); 3603 3604 if (pid == 0) 3605 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3606 3607 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3608 EXPECT_EQ(req.pid, pid); 3609 3610 resp.id = req.id; 3611 resp.error = 0; 3612 resp.val = USER_NOTIF_MAGIC; 3613 3614 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3615 3616 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3617 EXPECT_EQ(true, WIFEXITED(status)); 3618 EXPECT_EQ(0, WEXITSTATUS(status)); 3619 close(listener); 3620} 3621 3622/* 3623 * Check that a pid in a sibling (i.e. unrelated) namespace shows up as 0, i.e. 3624 * invalid. 3625 */ 3626TEST(user_notification_sibling_pid_ns) 3627{ 3628 pid_t pid, pid2; 3629 int status, listener; 3630 struct seccomp_notif req = {}; 3631 struct seccomp_notif_resp resp = {}; 3632 3633 ASSERT_EQ(prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0), 0) { 3634 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3635 } 3636 3637 listener = user_notif_syscall(__NR_getppid, 3638 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3639 ASSERT_GE(listener, 0); 3640 3641 pid = fork(); 3642 ASSERT_GE(pid, 0); 3643 3644 if (pid == 0) { 3645 ASSERT_EQ(unshare(CLONE_NEWPID), 0); 3646 3647 pid2 = fork(); 3648 ASSERT_GE(pid2, 0); 3649 3650 if (pid2 == 0) 3651 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3652 3653 EXPECT_EQ(waitpid(pid2, &status, 0), pid2); 3654 EXPECT_EQ(true, WIFEXITED(status)); 3655 EXPECT_EQ(0, WEXITSTATUS(status)); 3656 exit(WEXITSTATUS(status)); 3657 } 3658 3659 /* Create the sibling ns, and sibling in it. */ 3660 ASSERT_EQ(unshare(CLONE_NEWPID), 0) { 3661 if (errno == EPERM) 3662 SKIP(return, "CLONE_NEWPID requires CAP_SYS_ADMIN"); 3663 } 3664 ASSERT_EQ(errno, 0); 3665 3666 pid2 = fork(); 3667 ASSERT_GE(pid2, 0); 3668 3669 if (pid2 == 0) { 3670 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3671 /* 3672 * The pid should be 0, i.e. the task is in some namespace that 3673 * we can't "see". 3674 */ 3675 EXPECT_EQ(req.pid, 0); 3676 3677 resp.id = req.id; 3678 resp.error = 0; 3679 resp.val = USER_NOTIF_MAGIC; 3680 3681 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3682 exit(0); 3683 } 3684 3685 close(listener); 3686 3687 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3688 EXPECT_EQ(true, WIFEXITED(status)); 3689 EXPECT_EQ(0, WEXITSTATUS(status)); 3690 3691 EXPECT_EQ(waitpid(pid2, &status, 0), pid2); 3692 EXPECT_EQ(true, WIFEXITED(status)); 3693 EXPECT_EQ(0, WEXITSTATUS(status)); 3694} 3695 3696TEST(user_notification_fault_recv) 3697{ 3698 pid_t pid; 3699 int status, listener; 3700 struct seccomp_notif req = {}; 3701 struct seccomp_notif_resp resp = {}; 3702 3703 ASSERT_EQ(unshare(CLONE_NEWUSER), 0); 3704 3705 listener = user_notif_syscall(__NR_getppid, 3706 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3707 ASSERT_GE(listener, 0); 3708 3709 pid = fork(); 3710 ASSERT_GE(pid, 0); 3711 3712 if (pid == 0) 3713 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3714 3715 /* Do a bad recv() */ 3716 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, NULL), -1); 3717 EXPECT_EQ(errno, EFAULT); 3718 3719 /* We should still be able to receive this notification, though. */ 3720 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3721 EXPECT_EQ(req.pid, pid); 3722 3723 resp.id = req.id; 3724 resp.error = 0; 3725 resp.val = USER_NOTIF_MAGIC; 3726 3727 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 3728 3729 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3730 EXPECT_EQ(true, WIFEXITED(status)); 3731 EXPECT_EQ(0, WEXITSTATUS(status)); 3732} 3733 3734TEST(seccomp_get_notif_sizes) 3735{ 3736 struct seccomp_notif_sizes sizes; 3737 3738 ASSERT_EQ(seccomp(SECCOMP_GET_NOTIF_SIZES, 0, &sizes), 0); 3739 EXPECT_EQ(sizes.seccomp_notif, sizeof(struct seccomp_notif)); 3740 EXPECT_EQ(sizes.seccomp_notif_resp, sizeof(struct seccomp_notif_resp)); 3741} 3742 3743TEST(user_notification_continue) 3744{ 3745 pid_t pid; 3746 long ret; 3747 int status, listener; 3748 struct seccomp_notif req = {}; 3749 struct seccomp_notif_resp resp = {}; 3750 struct pollfd pollfd; 3751 3752 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3753 ASSERT_EQ(0, ret) { 3754 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3755 } 3756 3757 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER); 3758 ASSERT_GE(listener, 0); 3759 3760 pid = fork(); 3761 ASSERT_GE(pid, 0); 3762 3763 if (pid == 0) { 3764 int dup_fd, pipe_fds[2]; 3765 pid_t self; 3766 3767 ASSERT_GE(pipe(pipe_fds), 0); 3768 3769 dup_fd = dup(pipe_fds[0]); 3770 ASSERT_GE(dup_fd, 0); 3771 EXPECT_NE(pipe_fds[0], dup_fd); 3772 3773 self = getpid(); 3774 ASSERT_EQ(filecmp(self, self, pipe_fds[0], dup_fd), 0); 3775 exit(0); 3776 } 3777 3778 pollfd.fd = listener; 3779 pollfd.events = POLLIN | POLLOUT; 3780 3781 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3782 EXPECT_EQ(pollfd.revents, POLLIN); 3783 3784 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 3785 3786 pollfd.fd = listener; 3787 pollfd.events = POLLIN | POLLOUT; 3788 3789 EXPECT_GT(poll(&pollfd, 1, -1), 0); 3790 EXPECT_EQ(pollfd.revents, POLLOUT); 3791 3792 EXPECT_EQ(req.data.nr, __NR_dup); 3793 3794 resp.id = req.id; 3795 resp.flags = SECCOMP_USER_NOTIF_FLAG_CONTINUE; 3796 3797 /* 3798 * Verify that setting SECCOMP_USER_NOTIF_FLAG_CONTINUE enforces other 3799 * args be set to 0. 3800 */ 3801 resp.error = 0; 3802 resp.val = USER_NOTIF_MAGIC; 3803 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3804 EXPECT_EQ(errno, EINVAL); 3805 3806 resp.error = USER_NOTIF_MAGIC; 3807 resp.val = 0; 3808 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), -1); 3809 EXPECT_EQ(errno, EINVAL); 3810 3811 resp.error = 0; 3812 resp.val = 0; 3813 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0) { 3814 if (errno == EINVAL) 3815 SKIP(goto skip, "Kernel does not support SECCOMP_USER_NOTIF_FLAG_CONTINUE"); 3816 } 3817 3818skip: 3819 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3820 EXPECT_EQ(true, WIFEXITED(status)); 3821 EXPECT_EQ(0, WEXITSTATUS(status)) { 3822 if (WEXITSTATUS(status) == 2) { 3823 SKIP(return, "Kernel does not support kcmp() syscall"); 3824 return; 3825 } 3826 } 3827} 3828 3829TEST(user_notification_filter_empty) 3830{ 3831 pid_t pid; 3832 long ret; 3833 int status; 3834 struct pollfd pollfd; 3835 struct __clone_args args = { 3836 .flags = CLONE_FILES, 3837 .exit_signal = SIGCHLD, 3838 }; 3839 3840 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3841 ASSERT_EQ(0, ret) { 3842 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3843 } 3844 3845 pid = sys_clone3(&args, sizeof(args)); 3846 ASSERT_GE(pid, 0); 3847 3848 if (pid == 0) { 3849 int listener; 3850 3851 listener = user_notif_syscall(__NR_mknodat, SECCOMP_FILTER_FLAG_NEW_LISTENER); 3852 if (listener < 0) 3853 _exit(EXIT_FAILURE); 3854 3855 if (dup2(listener, 200) != 200) 3856 _exit(EXIT_FAILURE); 3857 3858 close(listener); 3859 3860 _exit(EXIT_SUCCESS); 3861 } 3862 3863 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3864 EXPECT_EQ(true, WIFEXITED(status)); 3865 EXPECT_EQ(0, WEXITSTATUS(status)); 3866 3867 /* 3868 * The seccomp filter has become unused so we should be notified once 3869 * the kernel gets around to cleaning up task struct. 3870 */ 3871 pollfd.fd = 200; 3872 pollfd.events = POLLHUP; 3873 3874 EXPECT_GT(poll(&pollfd, 1, 2000), 0); 3875 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0); 3876} 3877 3878static void *do_thread(void *data) 3879{ 3880 return NULL; 3881} 3882 3883TEST(user_notification_filter_empty_threaded) 3884{ 3885 pid_t pid; 3886 long ret; 3887 int status; 3888 struct pollfd pollfd; 3889 struct __clone_args args = { 3890 .flags = CLONE_FILES, 3891 .exit_signal = SIGCHLD, 3892 }; 3893 3894 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3895 ASSERT_EQ(0, ret) { 3896 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3897 } 3898 3899 pid = sys_clone3(&args, sizeof(args)); 3900 ASSERT_GE(pid, 0); 3901 3902 if (pid == 0) { 3903 pid_t pid1, pid2; 3904 int listener, status; 3905 pthread_t thread; 3906 3907 listener = user_notif_syscall(__NR_dup, SECCOMP_FILTER_FLAG_NEW_LISTENER); 3908 if (listener < 0) 3909 _exit(EXIT_FAILURE); 3910 3911 if (dup2(listener, 200) != 200) 3912 _exit(EXIT_FAILURE); 3913 3914 close(listener); 3915 3916 pid1 = fork(); 3917 if (pid1 < 0) 3918 _exit(EXIT_FAILURE); 3919 3920 if (pid1 == 0) 3921 _exit(EXIT_SUCCESS); 3922 3923 pid2 = fork(); 3924 if (pid2 < 0) 3925 _exit(EXIT_FAILURE); 3926 3927 if (pid2 == 0) 3928 _exit(EXIT_SUCCESS); 3929 3930 if (pthread_create(&thread, NULL, do_thread, NULL) || 3931 pthread_join(thread, NULL)) 3932 _exit(EXIT_FAILURE); 3933 3934 if (pthread_create(&thread, NULL, do_thread, NULL) || 3935 pthread_join(thread, NULL)) 3936 _exit(EXIT_FAILURE); 3937 3938 if (waitpid(pid1, &status, 0) != pid1 || !WIFEXITED(status) || 3939 WEXITSTATUS(status)) 3940 _exit(EXIT_FAILURE); 3941 3942 if (waitpid(pid2, &status, 0) != pid2 || !WIFEXITED(status) || 3943 WEXITSTATUS(status)) 3944 _exit(EXIT_FAILURE); 3945 3946 exit(EXIT_SUCCESS); 3947 } 3948 3949 EXPECT_EQ(waitpid(pid, &status, 0), pid); 3950 EXPECT_EQ(true, WIFEXITED(status)); 3951 EXPECT_EQ(0, WEXITSTATUS(status)); 3952 3953 /* 3954 * The seccomp filter has become unused so we should be notified once 3955 * the kernel gets around to cleaning up task struct. 3956 */ 3957 pollfd.fd = 200; 3958 pollfd.events = POLLHUP; 3959 3960 EXPECT_GT(poll(&pollfd, 1, 2000), 0); 3961 EXPECT_GT((pollfd.revents & POLLHUP) ?: 0, 0); 3962} 3963 3964TEST(user_notification_addfd) 3965{ 3966 pid_t pid; 3967 long ret; 3968 int status, listener, memfd, fd; 3969 struct seccomp_notif_addfd addfd = {}; 3970 struct seccomp_notif_addfd_small small = {}; 3971 struct seccomp_notif_addfd_big big = {}; 3972 struct seccomp_notif req = {}; 3973 struct seccomp_notif_resp resp = {}; 3974 /* 100 ms */ 3975 struct timespec delay = { .tv_nsec = 100000000 }; 3976 3977 memfd = memfd_create("test", 0); 3978 ASSERT_GE(memfd, 0); 3979 3980 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 3981 ASSERT_EQ(0, ret) { 3982 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 3983 } 3984 3985 /* Check that the basic notification machinery works */ 3986 listener = user_notif_syscall(__NR_getppid, 3987 SECCOMP_FILTER_FLAG_NEW_LISTENER); 3988 ASSERT_GE(listener, 0); 3989 3990 pid = fork(); 3991 ASSERT_GE(pid, 0); 3992 3993 if (pid == 0) { 3994 if (syscall(__NR_getppid) != USER_NOTIF_MAGIC) 3995 exit(1); 3996 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 3997 } 3998 3999 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4000 4001 addfd.srcfd = memfd; 4002 addfd.newfd = 0; 4003 addfd.id = req.id; 4004 addfd.flags = 0x0; 4005 4006 /* Verify bad newfd_flags cannot be set */ 4007 addfd.newfd_flags = ~O_CLOEXEC; 4008 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4009 EXPECT_EQ(errno, EINVAL); 4010 addfd.newfd_flags = O_CLOEXEC; 4011 4012 /* Verify bad flags cannot be set */ 4013 addfd.flags = 0xff; 4014 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4015 EXPECT_EQ(errno, EINVAL); 4016 addfd.flags = 0; 4017 4018 /* Verify that remote_fd cannot be set without setting flags */ 4019 addfd.newfd = 1; 4020 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4021 EXPECT_EQ(errno, EINVAL); 4022 addfd.newfd = 0; 4023 4024 /* Verify small size cannot be set */ 4025 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_SMALL, &small), -1); 4026 EXPECT_EQ(errno, EINVAL); 4027 4028 /* Verify we can't send bits filled in unknown buffer area */ 4029 memset(&big, 0xAA, sizeof(big)); 4030 big.addfd = addfd; 4031 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big), -1); 4032 EXPECT_EQ(errno, E2BIG); 4033 4034 4035 /* Verify we can set an arbitrary remote fd */ 4036 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd); 4037 /* 4038 * The child has fds 0(stdin), 1(stdout), 2(stderr), 3(memfd), 4039 * 4(listener), so the newly allocated fd should be 5. 4040 */ 4041 EXPECT_EQ(fd, 5); 4042 EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0); 4043 4044 /* Verify we can set an arbitrary remote fd with large size */ 4045 memset(&big, 0x0, sizeof(big)); 4046 big.addfd = addfd; 4047 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD_BIG, &big); 4048 EXPECT_EQ(fd, 6); 4049 4050 /* Verify we can set a specific remote fd */ 4051 addfd.newfd = 42; 4052 addfd.flags = SECCOMP_ADDFD_FLAG_SETFD; 4053 fd = ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd); 4054 EXPECT_EQ(fd, 42); 4055 EXPECT_EQ(filecmp(getpid(), pid, memfd, fd), 0); 4056 4057 /* Resume syscall */ 4058 resp.id = req.id; 4059 resp.error = 0; 4060 resp.val = USER_NOTIF_MAGIC; 4061 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4062 4063 /* 4064 * This sets the ID of the ADD FD to the last request plus 1. The 4065 * notification ID increments 1 per notification. 4066 */ 4067 addfd.id = req.id + 1; 4068 4069 /* This spins until the underlying notification is generated */ 4070 while (ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd) != -1 && 4071 errno != -EINPROGRESS) 4072 nanosleep(&delay, NULL); 4073 4074 memset(&req, 0, sizeof(req)); 4075 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4076 ASSERT_EQ(addfd.id, req.id); 4077 4078 resp.id = req.id; 4079 resp.error = 0; 4080 resp.val = USER_NOTIF_MAGIC; 4081 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4082 4083 /* Wait for child to finish. */ 4084 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4085 EXPECT_EQ(true, WIFEXITED(status)); 4086 EXPECT_EQ(0, WEXITSTATUS(status)); 4087 4088 close(memfd); 4089} 4090 4091TEST(user_notification_addfd_rlimit) 4092{ 4093 pid_t pid; 4094 long ret; 4095 int status, listener, memfd; 4096 struct seccomp_notif_addfd addfd = {}; 4097 struct seccomp_notif req = {}; 4098 struct seccomp_notif_resp resp = {}; 4099 const struct rlimit lim = { 4100 .rlim_cur = 0, 4101 .rlim_max = 0, 4102 }; 4103 4104 memfd = memfd_create("test", 0); 4105 ASSERT_GE(memfd, 0); 4106 4107 ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0); 4108 ASSERT_EQ(0, ret) { 4109 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!"); 4110 } 4111 4112 /* Check that the basic notification machinery works */ 4113 listener = user_notif_syscall(__NR_getppid, 4114 SECCOMP_FILTER_FLAG_NEW_LISTENER); 4115 ASSERT_GE(listener, 0); 4116 4117 pid = fork(); 4118 ASSERT_GE(pid, 0); 4119 4120 if (pid == 0) 4121 exit(syscall(__NR_getppid) != USER_NOTIF_MAGIC); 4122 4123 4124 ASSERT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0); 4125 4126 ASSERT_EQ(prlimit(pid, RLIMIT_NOFILE, &lim, NULL), 0); 4127 4128 addfd.srcfd = memfd; 4129 addfd.newfd_flags = O_CLOEXEC; 4130 addfd.newfd = 0; 4131 addfd.id = req.id; 4132 addfd.flags = 0; 4133 4134 /* Should probably spot check /proc/sys/fs/file-nr */ 4135 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4136 EXPECT_EQ(errno, EMFILE); 4137 4138 addfd.newfd = 100; 4139 addfd.flags = SECCOMP_ADDFD_FLAG_SETFD; 4140 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_ADDFD, &addfd), -1); 4141 EXPECT_EQ(errno, EBADF); 4142 4143 resp.id = req.id; 4144 resp.error = 0; 4145 resp.val = USER_NOTIF_MAGIC; 4146 4147 EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_SEND, &resp), 0); 4148 4149 /* Wait for child to finish. */ 4150 EXPECT_EQ(waitpid(pid, &status, 0), pid); 4151 EXPECT_EQ(true, WIFEXITED(status)); 4152 EXPECT_EQ(0, WEXITSTATUS(status)); 4153 4154 close(memfd); 4155} 4156 4157/* 4158 * TODO: 4159 * - expand NNP testing 4160 * - better arch-specific TRACE and TRAP handlers. 4161 * - endianness checking when appropriate 4162 * - 64-bit arg prodding 4163 * - arch value testing (x86 modes especially) 4164 * - verify that FILTER_FLAG_LOG filters generate log messages 4165 * - verify that RET_LOG generates log messages 4166 */ 4167 4168TEST_HARNESS_MAIN 4169