tools/testing/selftests/seccomp/seccomp_bpf.c

   1 /*
   2  * Copyright (c) 2012 The Chromium OS Authors. All rights reserved.
   3  * Use of this source code is governed by the GPLv2 license.
   4  *
   5  * Test code for seccomp bpf.
   6  */
   7
   8 #include <sys/types.h>
   9 #include <asm/siginfo.h>
  10 #define __have_siginfo_t 1
  11 #define __have_sigval_t 1
  12 #define __have_sigevent_t 1
  13
  14 #include <errno.h>
  15 #include <linux/filter.h>
  16 #include <sys/prctl.h>
  17 #include <sys/ptrace.h>
  18 #include <sys/user.h>
  19 #include <linux/prctl.h>
  20 #include <linux/ptrace.h>
  21 #include <linux/seccomp.h>
  22 #include <pthread.h>
  23 #include <semaphore.h>
  24 #include <signal.h>
  25 #include <stddef.h>
  26 #include <stdbool.h>
  27 #include <string.h>
  28 #include <time.h>
  29 #include <linux/elf.h>
  30 #include <sys/uio.h>
  31 #include <sys/utsname.h>
  32 #include <sys/fcntl.h>
  33 #include <sys/mman.h>
  34 #include <sys/times.h>
  35
  36 #define _GNU_SOURCE
  37 #include <unistd.h>
  38 #include <sys/syscall.h>
  39
  40 #include "test_harness.h"
  41
  42 #ifndef PR_SET_PTRACER
  43 # define PR_SET_PTRACER 0x59616d61
  44 #endif
  45
  46 #ifndef PR_SET_NO_NEW_PRIVS
  47 #define PR_SET_NO_NEW_PRIVS 38
  48 #define PR_GET_NO_NEW_PRIVS 39
  49 #endif
  50
  51 #ifndef PR_SECCOMP_EXT
  52 #define PR_SECCOMP_EXT 43
  53 #endif
  54
  55 #ifndef SECCOMP_EXT_ACT
  56 #define SECCOMP_EXT_ACT 1
  57 #endif
  58
  59 #ifndef SECCOMP_EXT_ACT_TSYNC
  60 #define SECCOMP_EXT_ACT_TSYNC 1
  61 #endif
  62
  63 #ifndef SECCOMP_MODE_STRICT
  64 #define SECCOMP_MODE_STRICT 1
  65 #endif
  66
  67 #ifndef SECCOMP_MODE_FILTER
  68 #define SECCOMP_MODE_FILTER 2
  69 #endif
  70
  71 #ifndef SECCOMP_RET_KILL
  72 #define SECCOMP_RET_KILL        0x00000000U /* kill the task immediately */
  73 #define SECCOMP_RET_TRAP        0x00030000U /* disallow and force a SIGSYS */
  74 #define SECCOMP_RET_ERRNO       0x00050000U /* returns an errno */
  75 #define SECCOMP_RET_TRACE       0x7ff00000U /* pass to a tracer or disallow */
  76 #define SECCOMP_RET_ALLOW       0x7fff0000U /* allow */
  77
  78 /* Masks for the return value sections. */
  79 #define SECCOMP_RET_ACTION      0x7fff0000U
  80 #define SECCOMP_RET_DATA        0x0000ffffU
  81
  82 struct seccomp_data {
  83         int nr;
  84         __u32 arch;
  85         __u64 instruction_pointer;
  86         __u64 args[6];
  87 };
  88 #endif
  89
  90 #if __BYTE_ORDER == __LITTLE_ENDIAN
  91 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]))
  92 #elif __BYTE_ORDER == __BIG_ENDIAN
  93 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32))
  94 #else
  95 #error "wut? Unknown __BYTE_ORDER?!"
  96 #endif
  97
  98 #define SIBLING_EXIT_UNKILLED   0xbadbeef
  99 #define SIBLING_EXIT_FAILURE    0xbadface
 100 #define SIBLING_EXIT_NEWPRIVS   0xbadfeed
 101
 102 TEST(mode_strict_support)
 103 {
 104         long ret;
 105
 106         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
 107         ASSERT_EQ(0, ret) {
 108                 TH_LOG("Kernel does not support CONFIG_SECCOMP");
 109         }
 110         syscall(__NR_exit, 1);
 111 }
 112
 113 TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL)
 114 {
 115         long ret;
 116
 117         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
 118         ASSERT_EQ(0, ret) {
 119                 TH_LOG("Kernel does not support CONFIG_SECCOMP");
 120         }
 121         syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
 122                 NULL, NULL, NULL);
 123         EXPECT_FALSE(true) {
 124                 TH_LOG("Unreachable!");
 125         }
 126 }
 127
 128 /* Note! This doesn't test no new privs behavior */
 129 TEST(no_new_privs_support)
 130 {
 131         long ret;
 132
 133         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 134         EXPECT_EQ(0, ret) {
 135                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
 136         }
 137 }
 138
 139 /* Tests kernel support by checking for a copy_from_user() fault on * NULL. */
 140 TEST(mode_filter_support)
 141 {
 142         long ret;
 143
 144         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
 145         ASSERT_EQ(0, ret) {
 146                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
 147         }
 148         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL);
 149         EXPECT_EQ(-1, ret);
 150         EXPECT_EQ(EFAULT, errno) {
 151                 TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!");
 152         }
 153 }
 154
 155 TEST(mode_filter_without_nnp)
 156 {
 157         struct sock_filter filter[] = {
 158                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 159         };
 160         struct sock_fprog prog = {
 161                 .len = (unsigned short)ARRAY_SIZE(filter),
 162                 .filter = filter,
 163         };
 164         long ret;
 165
 166         ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0);
 167         ASSERT_LE(0, ret) {
 168                 TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS");
 169         }
 170         errno = 0;
 171         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 172         /* Succeeds with CAP_SYS_ADMIN, fails without */
 173         /* TODO(wad) check caps not euid */
 174         if (geteuid()) {
 175                 EXPECT_EQ(-1, ret);
 176                 EXPECT_EQ(EACCES, errno);
 177         } else {
 178                 EXPECT_EQ(0, ret);
 179         }
 180 }
 181
 182 #define MAX_INSNS_PER_PATH 32768
 183
 184 TEST(filter_size_limits)
 185 {
 186         int i;
 187         int count = BPF_MAXINSNS + 1;
 188         struct sock_filter allow[] = {
 189                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 190         };
 191         struct sock_filter *filter;
 192         struct sock_fprog prog = { };
 193         long ret;
 194
 195         filter = calloc(count, sizeof(*filter));
 196         ASSERT_NE(NULL, filter);
 197
 198         for (i = 0; i < count; i++)
 199                 filter[i] = allow[0];
 200
 201         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 202         ASSERT_EQ(0, ret);
 203
 204         prog.filter = filter;
 205         prog.len = count;
 206
 207         /* Too many filter instructions in a single filter. */
 208         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 209         ASSERT_NE(0, ret) {
 210                 TH_LOG("Installing %d insn filter was allowed", prog.len);
 211         }
 212
 213         /* One less is okay, though. */
 214         prog.len -= 1;
 215         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 216         ASSERT_EQ(0, ret) {
 217                 TH_LOG("Installing %d insn filter wasn't allowed", prog.len);
 218         }
 219 }
 220
 221 TEST(filter_chain_limits)
 222 {
 223         int i;
 224         int count = BPF_MAXINSNS;
 225         struct sock_filter allow[] = {
 226                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 227         };
 228         struct sock_filter *filter;
 229         struct sock_fprog prog = { };
 230         long ret;
 231
 232         filter = calloc(count, sizeof(*filter));
 233         ASSERT_NE(NULL, filter);
 234
 235         for (i = 0; i < count; i++)
 236                 filter[i] = allow[0];
 237
 238         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 239         ASSERT_EQ(0, ret);
 240
 241         prog.filter = filter;
 242         prog.len = 1;
 243
 244         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 245         ASSERT_EQ(0, ret);
 246
 247         prog.len = count;
 248
 249         /* Too many total filter instructions. */
 250         for (i = 0; i < MAX_INSNS_PER_PATH; i++) {
 251                 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 252                 if (ret != 0)
 253                         break;
 254         }
 255         ASSERT_NE(0, ret) {
 256                 TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)",
 257                        i, count, i * (count + 4));
 258         }
 259 }
 260
 261 TEST(mode_filter_cannot_move_to_strict)
 262 {
 263         struct sock_filter filter[] = {
 264                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 265         };
 266         struct sock_fprog prog = {
 267                 .len = (unsigned short)ARRAY_SIZE(filter),
 268                 .filter = filter,
 269         };
 270         long ret;
 271
 272         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 273         ASSERT_EQ(0, ret);
 274
 275         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 276         ASSERT_EQ(0, ret);
 277
 278         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0);
 279         EXPECT_EQ(-1, ret);
 280         EXPECT_EQ(EINVAL, errno);
 281 }
 282
 283
 284 TEST(mode_filter_get_seccomp)
 285 {
 286         struct sock_filter filter[] = {
 287                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 288         };
 289         struct sock_fprog prog = {
 290                 .len = (unsigned short)ARRAY_SIZE(filter),
 291                 .filter = filter,
 292         };
 293         long ret;
 294
 295         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 296         ASSERT_EQ(0, ret);
 297
 298         ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
 299         EXPECT_EQ(0, ret);
 300
 301         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 302         ASSERT_EQ(0, ret);
 303
 304         ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
 305         EXPECT_EQ(2, ret);
 306 }
 307
 308
 309 TEST(ALLOW_all)
 310 {
 311         struct sock_filter filter[] = {
 312                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 313         };
 314         struct sock_fprog prog = {
 315                 .len = (unsigned short)ARRAY_SIZE(filter),
 316                 .filter = filter,
 317         };
 318         long ret;
 319
 320         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 321         ASSERT_EQ(0, ret);
 322
 323         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 324         ASSERT_EQ(0, ret);
 325 }
 326
 327 TEST(empty_prog)
 328 {
 329         struct sock_filter filter[] = {
 330         };
 331         struct sock_fprog prog = {
 332                 .len = (unsigned short)ARRAY_SIZE(filter),
 333                 .filter = filter,
 334         };
 335         long ret;
 336
 337         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 338         ASSERT_EQ(0, ret);
 339
 340         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 341         EXPECT_EQ(-1, ret);
 342         EXPECT_EQ(EINVAL, errno);
 343 }
 344
 345 TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS)
 346 {
 347         struct sock_filter filter[] = {
 348                 BPF_STMT(BPF_RET|BPF_K, 0x10000000U),
 349         };
 350         struct sock_fprog prog = {
 351                 .len = (unsigned short)ARRAY_SIZE(filter),
 352                 .filter = filter,
 353         };
 354         long ret;
 355
 356         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 357         ASSERT_EQ(0, ret);
 358
 359         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 360         ASSERT_EQ(0, ret);
 361         EXPECT_EQ(0, syscall(__NR_getpid)) {
 362                 TH_LOG("getpid() shouldn't ever return");
 363         }
 364 }
 365
 366 /* return code >= 0x80000000 is unused. */
 367 TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS)
 368 {
 369         struct sock_filter filter[] = {
 370                 BPF_STMT(BPF_RET|BPF_K, 0x90000000U),
 371         };
 372         struct sock_fprog prog = {
 373                 .len = (unsigned short)ARRAY_SIZE(filter),
 374                 .filter = filter,
 375         };
 376         long ret;
 377
 378         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 379         ASSERT_EQ(0, ret);
 380
 381         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 382         ASSERT_EQ(0, ret);
 383         EXPECT_EQ(0, syscall(__NR_getpid)) {
 384                 TH_LOG("getpid() shouldn't ever return");
 385         }
 386 }
 387
 388 TEST_SIGNAL(KILL_all, SIGSYS)
 389 {
 390         struct sock_filter filter[] = {
 391                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
 392         };
 393         struct sock_fprog prog = {
 394                 .len = (unsigned short)ARRAY_SIZE(filter),
 395                 .filter = filter,
 396         };
 397         long ret;
 398
 399         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 400         ASSERT_EQ(0, ret);
 401
 402         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 403         ASSERT_EQ(0, ret);
 404 }
 405
 406 TEST_SIGNAL(KILL_one, SIGSYS)
 407 {
 408         struct sock_filter filter[] = {
 409                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 410                         offsetof(struct seccomp_data, nr)),
 411                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
 412                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
 413                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 414         };
 415         struct sock_fprog prog = {
 416                 .len = (unsigned short)ARRAY_SIZE(filter),
 417                 .filter = filter,
 418         };
 419         long ret;
 420         pid_t parent = getppid();
 421
 422         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 423         ASSERT_EQ(0, ret);
 424
 425         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 426         ASSERT_EQ(0, ret);
 427
 428         EXPECT_EQ(parent, syscall(__NR_getppid));
 429         /* getpid() should never return. */
 430         EXPECT_EQ(0, syscall(__NR_getpid));
 431 }
 432
 433 TEST_SIGNAL(KILL_one_arg_one, SIGSYS)
 434 {
 435         void *fatal_address;
 436         struct sock_filter filter[] = {
 437                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 438                         offsetof(struct seccomp_data, nr)),
 439                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0),
 440                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 441                 /* Only both with lower 32-bit for now. */
 442                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)),
 443                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K,
 444                         (unsigned long)&fatal_address, 0, 1),
 445                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
 446                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 447         };
 448         struct sock_fprog prog = {
 449                 .len = (unsigned short)ARRAY_SIZE(filter),
 450                 .filter = filter,
 451         };
 452         long ret;
 453         pid_t parent = getppid();
 454         struct tms timebuf;
 455         clock_t clock = times(&timebuf);
 456
 457         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 458         ASSERT_EQ(0, ret);
 459
 460         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 461         ASSERT_EQ(0, ret);
 462
 463         EXPECT_EQ(parent, syscall(__NR_getppid));
 464         EXPECT_LE(clock, syscall(__NR_times, &timebuf));
 465         /* times() should never return. */
 466         EXPECT_EQ(0, syscall(__NR_times, &fatal_address));
 467 }
 468
 469 TEST_SIGNAL(KILL_one_arg_six, SIGSYS)
 470 {
 471 #ifndef __NR_mmap2
 472         int sysno = __NR_mmap;
 473 #else
 474         int sysno = __NR_mmap2;
 475 #endif
 476         struct sock_filter filter[] = {
 477                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 478                         offsetof(struct seccomp_data, nr)),
 479                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0),
 480                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 481                 /* Only both with lower 32-bit for now. */
 482                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)),
 483                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1),
 484                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
 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         pid_t parent = getppid();
 493         int fd;
 494         void *map1, *map2;
 495         int page_size = sysconf(_SC_PAGESIZE);
 496
 497         ASSERT_LT(0, page_size);
 498
 499         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 500         ASSERT_EQ(0, ret);
 501
 502         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 503         ASSERT_EQ(0, ret);
 504
 505         fd = open("/dev/zero", O_RDONLY);
 506         ASSERT_NE(-1, fd);
 507
 508         EXPECT_EQ(parent, syscall(__NR_getppid));
 509         map1 = (void *)syscall(sysno,
 510                 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size);
 511         EXPECT_NE(MAP_FAILED, map1);
 512         /* mmap2() should never return. */
 513         map2 = (void *)syscall(sysno,
 514                  NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE);
 515         EXPECT_EQ(MAP_FAILED, map2);
 516
 517         /* The test failed, so clean up the resources. */
 518         munmap(map1, page_size);
 519         munmap(map2, page_size);
 520         close(fd);
 521 }
 522
 523 /* TODO(wad) add 64-bit versus 32-bit arg tests. */
 524 TEST(arg_out_of_range)
 525 {
 526         struct sock_filter filter[] = {
 527                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)),
 528                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 529         };
 530         struct sock_fprog prog = {
 531                 .len = (unsigned short)ARRAY_SIZE(filter),
 532                 .filter = filter,
 533         };
 534         long ret;
 535
 536         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 537         ASSERT_EQ(0, ret);
 538
 539         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 540         EXPECT_EQ(-1, ret);
 541         EXPECT_EQ(EINVAL, errno);
 542 }
 543
 544 TEST(ERRNO_valid)
 545 {
 546         struct sock_filter filter[] = {
 547                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 548                         offsetof(struct seccomp_data, nr)),
 549                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
 550                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | E2BIG),
 551                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 552         };
 553         struct sock_fprog prog = {
 554                 .len = (unsigned short)ARRAY_SIZE(filter),
 555                 .filter = filter,
 556         };
 557         long ret;
 558         pid_t parent = getppid();
 559
 560         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 561         ASSERT_EQ(0, ret);
 562
 563         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 564         ASSERT_EQ(0, ret);
 565
 566         EXPECT_EQ(parent, syscall(__NR_getppid));
 567         EXPECT_EQ(-1, read(0, NULL, 0));
 568         EXPECT_EQ(E2BIG, errno);
 569 }
 570
 571 TEST(ERRNO_zero)
 572 {
 573         struct sock_filter filter[] = {
 574                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 575                         offsetof(struct seccomp_data, nr)),
 576                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
 577                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | 0),
 578                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 579         };
 580         struct sock_fprog prog = {
 581                 .len = (unsigned short)ARRAY_SIZE(filter),
 582                 .filter = filter,
 583         };
 584         long ret;
 585         pid_t parent = getppid();
 586
 587         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 588         ASSERT_EQ(0, ret);
 589
 590         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 591         ASSERT_EQ(0, ret);
 592
 593         EXPECT_EQ(parent, syscall(__NR_getppid));
 594         /* "errno" of 0 is ok. */
 595         EXPECT_EQ(0, read(0, NULL, 0));
 596 }
 597
 598 TEST(ERRNO_capped)
 599 {
 600         struct sock_filter filter[] = {
 601                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 602                         offsetof(struct seccomp_data, nr)),
 603                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
 604                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | 4096),
 605                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 606         };
 607         struct sock_fprog prog = {
 608                 .len = (unsigned short)ARRAY_SIZE(filter),
 609                 .filter = filter,
 610         };
 611         long ret;
 612         pid_t parent = getppid();
 613
 614         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 615         ASSERT_EQ(0, ret);
 616
 617         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 618         ASSERT_EQ(0, ret);
 619
 620         EXPECT_EQ(parent, syscall(__NR_getppid));
 621         EXPECT_EQ(-1, read(0, NULL, 0));
 622         EXPECT_EQ(4095, errno);
 623 }
 624
 625 FIXTURE_DATA(TRAP) {
 626         struct sock_fprog prog;
 627 };
 628
 629 FIXTURE_SETUP(TRAP)
 630 {
 631         struct sock_filter filter[] = {
 632                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 633                         offsetof(struct seccomp_data, nr)),
 634                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
 635                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
 636                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 637         };
 638
 639         memset(&self->prog, 0, sizeof(self->prog));
 640         self->prog.filter = malloc(sizeof(filter));
 641         ASSERT_NE(NULL, self->prog.filter);
 642         memcpy(self->prog.filter, filter, sizeof(filter));
 643         self->prog.len = (unsigned short)ARRAY_SIZE(filter);
 644 }
 645
 646 FIXTURE_TEARDOWN(TRAP)
 647 {
 648         if (self->prog.filter)
 649                 free(self->prog.filter);
 650 }
 651
 652 TEST_F_SIGNAL(TRAP, dfl, SIGSYS)
 653 {
 654         long ret;
 655
 656         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 657         ASSERT_EQ(0, ret);
 658
 659         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
 660         ASSERT_EQ(0, ret);
 661         syscall(__NR_getpid);
 662 }
 663
 664 /* Ensure that SIGSYS overrides SIG_IGN */
 665 TEST_F_SIGNAL(TRAP, ign, SIGSYS)
 666 {
 667         long ret;
 668
 669         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 670         ASSERT_EQ(0, ret);
 671
 672         signal(SIGSYS, SIG_IGN);
 673
 674         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
 675         ASSERT_EQ(0, ret);
 676         syscall(__NR_getpid);
 677 }
 678
 679 static struct siginfo TRAP_info;
 680 static volatile int TRAP_nr;
 681 static void TRAP_action(int nr, siginfo_t *info, void *void_context)
 682 {
 683         memcpy(&TRAP_info, info, sizeof(TRAP_info));
 684         TRAP_nr = nr;
 685 }
 686
 687 TEST_F(TRAP, handler)
 688 {
 689         int ret, test;
 690         struct sigaction act;
 691         sigset_t mask;
 692
 693         memset(&act, 0, sizeof(act));
 694         sigemptyset(&mask);
 695         sigaddset(&mask, SIGSYS);
 696
 697         act.sa_sigaction = &TRAP_action;
 698         act.sa_flags = SA_SIGINFO;
 699         ret = sigaction(SIGSYS, &act, NULL);
 700         ASSERT_EQ(0, ret) {
 701                 TH_LOG("sigaction failed");
 702         }
 703         ret = sigprocmask(SIG_UNBLOCK, &mask, NULL);
 704         ASSERT_EQ(0, ret) {
 705                 TH_LOG("sigprocmask failed");
 706         }
 707
 708         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 709         ASSERT_EQ(0, ret);
 710         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
 711         ASSERT_EQ(0, ret);
 712         TRAP_nr = 0;
 713         memset(&TRAP_info, 0, sizeof(TRAP_info));
 714         /* Expect the registers to be rolled back. (nr = error) may vary
 715          * based on arch. */
 716         ret = syscall(__NR_getpid);
 717         /* Silence gcc warning about volatile. */
 718         test = TRAP_nr;
 719         EXPECT_EQ(SIGSYS, test);
 720         struct local_sigsys {
 721                 void *_call_addr;       /* calling user insn */
 722                 int _syscall;           /* triggering system call number */
 723                 unsigned int _arch;     /* AUDIT_ARCH_* of syscall */
 724         } *sigsys = (struct local_sigsys *)
 725 #ifdef si_syscall
 726                 &(TRAP_info.si_call_addr);
 727 #else
 728                 &TRAP_info.si_pid;
 729 #endif
 730         EXPECT_EQ(__NR_getpid, sigsys->_syscall);
 731         /* Make sure arch is non-zero. */
 732         EXPECT_NE(0, sigsys->_arch);
 733         EXPECT_NE(0, (unsigned long)sigsys->_call_addr);
 734 }
 735
 736 FIXTURE_DATA(precedence) {
 737         struct sock_fprog allow;
 738         struct sock_fprog trace;
 739         struct sock_fprog error;
 740         struct sock_fprog trap;
 741         struct sock_fprog kill;
 742 };
 743
 744 FIXTURE_SETUP(precedence)
 745 {
 746         struct sock_filter allow_insns[] = {
 747                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 748         };
 749         struct sock_filter trace_insns[] = {
 750                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 751                         offsetof(struct seccomp_data, nr)),
 752                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
 753                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 754                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE),
 755         };
 756         struct sock_filter error_insns[] = {
 757                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 758                         offsetof(struct seccomp_data, nr)),
 759                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
 760                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 761                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO),
 762         };
 763         struct sock_filter trap_insns[] = {
 764                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 765                         offsetof(struct seccomp_data, nr)),
 766                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
 767                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 768                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
 769         };
 770         struct sock_filter kill_insns[] = {
 771                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 772                         offsetof(struct seccomp_data, nr)),
 773                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
 774                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 775                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
 776         };
 777
 778         memset(self, 0, sizeof(*self));
 779 #define FILTER_ALLOC(_x) \
 780         self->_x.filter = malloc(sizeof(_x##_insns)); \
 781         ASSERT_NE(NULL, self->_x.filter); \
 782         memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \
 783         self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns)
 784         FILTER_ALLOC(allow);
 785         FILTER_ALLOC(trace);
 786         FILTER_ALLOC(error);
 787         FILTER_ALLOC(trap);
 788         FILTER_ALLOC(kill);
 789 }
 790
 791 FIXTURE_TEARDOWN(precedence)
 792 {
 793 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter)
 794         FILTER_FREE(allow);
 795         FILTER_FREE(trace);
 796         FILTER_FREE(error);
 797         FILTER_FREE(trap);
 798         FILTER_FREE(kill);
 799 }
 800
 801 TEST_F(precedence, allow_ok)
 802 {
 803         pid_t parent, res = 0;
 804         long ret;
 805
 806         parent = getppid();
 807         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 808         ASSERT_EQ(0, ret);
 809
 810         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 811         ASSERT_EQ(0, ret);
 812         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 813         ASSERT_EQ(0, ret);
 814         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 815         ASSERT_EQ(0, ret);
 816         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
 817         ASSERT_EQ(0, ret);
 818         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
 819         ASSERT_EQ(0, ret);
 820         /* Should work just fine. */
 821         res = syscall(__NR_getppid);
 822         EXPECT_EQ(parent, res);
 823 }
 824
 825 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS)
 826 {
 827         pid_t parent, res = 0;
 828         long ret;
 829
 830         parent = getppid();
 831         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 832         ASSERT_EQ(0, ret);
 833
 834         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 835         ASSERT_EQ(0, ret);
 836         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 837         ASSERT_EQ(0, ret);
 838         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 839         ASSERT_EQ(0, ret);
 840         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
 841         ASSERT_EQ(0, ret);
 842         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
 843         ASSERT_EQ(0, ret);
 844         /* Should work just fine. */
 845         res = syscall(__NR_getppid);
 846         EXPECT_EQ(parent, res);
 847         /* getpid() should never return. */
 848         res = syscall(__NR_getpid);
 849         EXPECT_EQ(0, res);
 850 }
 851
 852 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS)
 853 {
 854         pid_t parent;
 855         long ret;
 856
 857         parent = getppid();
 858         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 859         ASSERT_EQ(0, ret);
 860
 861         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 862         ASSERT_EQ(0, ret);
 863         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
 864         ASSERT_EQ(0, ret);
 865         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 866         ASSERT_EQ(0, ret);
 867         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 868         ASSERT_EQ(0, ret);
 869         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
 870         ASSERT_EQ(0, ret);
 871         /* Should work just fine. */
 872         EXPECT_EQ(parent, syscall(__NR_getppid));
 873         /* getpid() should never return. */
 874         EXPECT_EQ(0, syscall(__NR_getpid));
 875 }
 876
 877 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS)
 878 {
 879         pid_t parent;
 880         long ret;
 881
 882         parent = getppid();
 883         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 884         ASSERT_EQ(0, ret);
 885
 886         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 887         ASSERT_EQ(0, ret);
 888         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 889         ASSERT_EQ(0, ret);
 890         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 891         ASSERT_EQ(0, ret);
 892         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
 893         ASSERT_EQ(0, ret);
 894         /* Should work just fine. */
 895         EXPECT_EQ(parent, syscall(__NR_getppid));
 896         /* getpid() should never return. */
 897         EXPECT_EQ(0, syscall(__NR_getpid));
 898 }
 899
 900 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS)
 901 {
 902         pid_t parent;
 903         long ret;
 904
 905         parent = getppid();
 906         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 907         ASSERT_EQ(0, ret);
 908
 909         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 910         ASSERT_EQ(0, ret);
 911         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
 912         ASSERT_EQ(0, ret);
 913         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 914         ASSERT_EQ(0, ret);
 915         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 916         ASSERT_EQ(0, ret);
 917         /* Should work just fine. */
 918         EXPECT_EQ(parent, syscall(__NR_getppid));
 919         /* getpid() should never return. */
 920         EXPECT_EQ(0, syscall(__NR_getpid));
 921 }
 922
 923 TEST_F(precedence, errno_is_third)
 924 {
 925         pid_t parent;
 926         long ret;
 927
 928         parent = getppid();
 929         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 930         ASSERT_EQ(0, ret);
 931
 932         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 933         ASSERT_EQ(0, ret);
 934         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 935         ASSERT_EQ(0, ret);
 936         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 937         ASSERT_EQ(0, ret);
 938         /* Should work just fine. */
 939         EXPECT_EQ(parent, syscall(__NR_getppid));
 940         EXPECT_EQ(0, syscall(__NR_getpid));
 941 }
 942
 943 TEST_F(precedence, errno_is_third_in_any_order)
 944 {
 945         pid_t parent;
 946         long ret;
 947
 948         parent = getppid();
 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, &self->error);
 953         ASSERT_EQ(0, ret);
 954         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 955         ASSERT_EQ(0, ret);
 956         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 957         ASSERT_EQ(0, ret);
 958         /* Should work just fine. */
 959         EXPECT_EQ(parent, syscall(__NR_getppid));
 960         EXPECT_EQ(0, syscall(__NR_getpid));
 961 }
 962
 963 TEST_F(precedence, trace_is_fourth)
 964 {
 965         pid_t parent;
 966         long ret;
 967
 968         parent = getppid();
 969         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 970         ASSERT_EQ(0, ret);
 971
 972         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 973         ASSERT_EQ(0, ret);
 974         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 975         ASSERT_EQ(0, ret);
 976         /* Should work just fine. */
 977         EXPECT_EQ(parent, syscall(__NR_getppid));
 978         /* No ptracer */
 979         EXPECT_EQ(-1, syscall(__NR_getpid));
 980 }
 981
 982 TEST_F(precedence, trace_is_fourth_in_any_order)
 983 {
 984         pid_t parent;
 985         long ret;
 986
 987         parent = getppid();
 988         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 989         ASSERT_EQ(0, ret);
 990
 991         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 992         ASSERT_EQ(0, ret);
 993         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 994         ASSERT_EQ(0, ret);
 995         /* Should work just fine. */
 996         EXPECT_EQ(parent, syscall(__NR_getppid));
 997         /* No ptracer */
 998         EXPECT_EQ(-1, syscall(__NR_getpid));
 999 }
1000
1001 #ifndef PTRACE_O_TRACESECCOMP
1002 #define PTRACE_O_TRACESECCOMP   0x00000080
1003 #endif
1004
1005 /* Catch the Ubuntu 12.04 value error. */
1006 #if PTRACE_EVENT_SECCOMP != 7
1007 #undef PTRACE_EVENT_SECCOMP
1008 #endif
1009
1010 #ifndef PTRACE_EVENT_SECCOMP
1011 #define PTRACE_EVENT_SECCOMP 7
1012 #endif
1013
1014 #define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP)
1015 bool tracer_running;
1016 void tracer_stop(int sig)
1017 {
1018         tracer_running = false;
1019 }
1020
1021 typedef void tracer_func_t(struct __test_metadata *_metadata,
1022                            pid_t tracee, int status, void *args);
1023
1024 void tracer(struct __test_metadata *_metadata, int fd, pid_t tracee,
1025             tracer_func_t tracer_func, void *args)
1026 {
1027         int ret = -1;
1028         struct sigaction action = {
1029                 .sa_handler = tracer_stop,
1030         };
1031
1032         /* Allow external shutdown. */
1033         tracer_running = true;
1034         ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL));
1035
1036         errno = 0;
1037         while (ret == -1 && errno != EINVAL)
1038                 ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0);
1039         ASSERT_EQ(0, ret) {
1040                 kill(tracee, SIGKILL);
1041         }
1042         /* Wait for attach stop */
1043         wait(NULL);
1044
1045         ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, PTRACE_O_TRACESECCOMP);
1046         ASSERT_EQ(0, ret) {
1047                 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP");
1048                 kill(tracee, SIGKILL);
1049         }
1050         ptrace(PTRACE_CONT, tracee, NULL, 0);
1051
1052         /* Unblock the tracee */
1053         ASSERT_EQ(1, write(fd, "A", 1));
1054         ASSERT_EQ(0, close(fd));
1055
1056         /* Run until we're shut down. Must assert to stop execution. */
1057         while (tracer_running) {
1058                 int status;
1059
1060                 if (wait(&status) != tracee)
1061                         continue;
1062                 if (WIFSIGNALED(status) || WIFEXITED(status))
1063                         /* Child is dead. Time to go. */
1064                         return;
1065
1066                 /* Make sure this is a seccomp event. */
1067                 ASSERT_EQ(true, IS_SECCOMP_EVENT(status));
1068
1069                 tracer_func(_metadata, tracee, status, args);
1070
1071                 ret = ptrace(PTRACE_CONT, tracee, NULL, NULL);
1072                 ASSERT_EQ(0, ret);
1073         }
1074         /* Directly report the status of our test harness results. */
1075         syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS : EXIT_FAILURE);
1076 }
1077
1078 /* Common tracer setup/teardown functions. */
1079 void cont_handler(int num)
1080 { }
1081 pid_t setup_trace_fixture(struct __test_metadata *_metadata,
1082                           tracer_func_t func, void *args)
1083 {
1084         char sync;
1085         int pipefd[2];
1086         pid_t tracer_pid;
1087         pid_t tracee = getpid();
1088
1089         /* Setup a pipe for clean synchronization. */
1090         ASSERT_EQ(0, pipe(pipefd));
1091
1092         /* Fork a child which we'll promote to tracer */
1093         tracer_pid = fork();
1094         ASSERT_LE(0, tracer_pid);
1095         signal(SIGALRM, cont_handler);
1096         if (tracer_pid == 0) {
1097                 close(pipefd[0]);
1098                 tracer(_metadata, pipefd[1], tracee, func, args);
1099                 syscall(__NR_exit, 0);
1100         }
1101         close(pipefd[1]);
1102         prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0);
1103         read(pipefd[0], &sync, 1);
1104         close(pipefd[0]);
1105
1106         return tracer_pid;
1107 }
1108 void teardown_trace_fixture(struct __test_metadata *_metadata,
1109                             pid_t tracer)
1110 {
1111         if (tracer) {
1112                 int status;
1113                 /*
1114                  * Extract the exit code from the other process and
1115                  * adopt it for ourselves in case its asserts failed.
1116                  */
1117                 ASSERT_EQ(0, kill(tracer, SIGUSR1));
1118                 ASSERT_EQ(tracer, waitpid(tracer, &status, 0));
1119                 if (WEXITSTATUS(status))
1120                         _metadata->passed = 0;
1121         }
1122 }
1123
1124 /* "poke" tracer arguments and function. */
1125 struct tracer_args_poke_t {
1126         unsigned long poke_addr;
1127 };
1128
1129 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status,
1130                  void *args)
1131 {
1132         int ret;
1133         unsigned long msg;
1134         struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args;
1135
1136         ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1137         EXPECT_EQ(0, ret);
1138         /* If this fails, don't try to recover. */
1139         ASSERT_EQ(0x1001, msg) {
1140                 kill(tracee, SIGKILL);
1141         }
1142         /*
1143          * Poke in the message.
1144          * Registers are not touched to try to keep this relatively arch
1145          * agnostic.
1146          */
1147         ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001);
1148         EXPECT_EQ(0, ret);
1149 }
1150
1151 FIXTURE_DATA(TRACE_poke) {
1152         struct sock_fprog prog;
1153         pid_t tracer;
1154         long poked;
1155         struct tracer_args_poke_t tracer_args;
1156 };
1157
1158 FIXTURE_SETUP(TRACE_poke)
1159 {
1160         struct sock_filter filter[] = {
1161                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1162                         offsetof(struct seccomp_data, nr)),
1163                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1164                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001),
1165                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1166         };
1167
1168         self->poked = 0;
1169         memset(&self->prog, 0, sizeof(self->prog));
1170         self->prog.filter = malloc(sizeof(filter));
1171         ASSERT_NE(NULL, self->prog.filter);
1172         memcpy(self->prog.filter, filter, sizeof(filter));
1173         self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1174
1175         /* Set up tracer args. */
1176         self->tracer_args.poke_addr = (unsigned long)&self->poked;
1177
1178         /* Launch tracer. */
1179         self->tracer = setup_trace_fixture(_metadata, tracer_poke,
1180                                            &self->tracer_args);
1181 }
1182
1183 FIXTURE_TEARDOWN(TRACE_poke)
1184 {
1185         teardown_trace_fixture(_metadata, self->tracer);
1186         if (self->prog.filter)
1187                 free(self->prog.filter);
1188 }
1189
1190 TEST_F(TRACE_poke, read_has_side_effects)
1191 {
1192         ssize_t ret;
1193
1194         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1195         ASSERT_EQ(0, ret);
1196
1197         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1198         ASSERT_EQ(0, ret);
1199
1200         EXPECT_EQ(0, self->poked);
1201         ret = read(-1, NULL, 0);
1202         EXPECT_EQ(-1, ret);
1203         EXPECT_EQ(0x1001, self->poked);
1204 }
1205
1206 TEST_F(TRACE_poke, getpid_runs_normally)
1207 {
1208         long ret;
1209
1210         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1211         ASSERT_EQ(0, ret);
1212
1213         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1214         ASSERT_EQ(0, ret);
1215
1216         EXPECT_EQ(0, self->poked);
1217         EXPECT_NE(0, syscall(__NR_getpid));
1218         EXPECT_EQ(0, self->poked);
1219 }
1220
1221 #if defined(__x86_64__)
1222 # define ARCH_REGS      struct user_regs_struct
1223 # define SYSCALL_NUM    orig_rax
1224 # define SYSCALL_RET    rax
1225 #elif defined(__i386__)
1226 # define ARCH_REGS      struct user_regs_struct
1227 # define SYSCALL_NUM    orig_eax
1228 # define SYSCALL_RET    eax
1229 #elif defined(__arm__)
1230 # define ARCH_REGS      struct pt_regs
1231 # define SYSCALL_NUM    ARM_r7
1232 # define SYSCALL_RET    ARM_r0
1233 #elif defined(__aarch64__)
1234 # define ARCH_REGS      struct user_pt_regs
1235 # define SYSCALL_NUM    regs[8]
1236 # define SYSCALL_RET    regs[0]
1237 #elif defined(__powerpc__)
1238 # define ARCH_REGS      struct pt_regs
1239 # define SYSCALL_NUM    gpr[0]
1240 # define SYSCALL_RET    gpr[3]
1241 #elif defined(__s390__)
1242 # define ARCH_REGS     s390_regs
1243 # define SYSCALL_NUM   gprs[2]
1244 # define SYSCALL_RET   gprs[2]
1245 #elif defined(__mips__)
1246 # define ARCH_REGS      struct pt_regs
1247 # define SYSCALL_NUM    regs[2]
1248 # define SYSCALL_SYSCALL_NUM regs[4]
1249 # define SYSCALL_RET    regs[2]
1250 # define SYSCALL_NUM_RET_SHARE_REG
1251 #else
1252 # error "Do not know how to find your architecture's registers and syscalls"
1253 #endif
1254
1255 /* Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for
1256  * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux).
1257  */
1258 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__)
1259 #define HAVE_GETREGS
1260 #endif
1261
1262 /* Architecture-specific syscall fetching routine. */
1263 int get_syscall(struct __test_metadata *_metadata, pid_t tracee)
1264 {
1265         ARCH_REGS regs;
1266 #ifdef HAVE_GETREGS
1267         EXPECT_EQ(0, ptrace(PTRACE_GETREGS, tracee, 0, &regs)) {
1268                 TH_LOG("PTRACE_GETREGS failed");
1269                 return -1;
1270         }
1271 #else
1272         struct iovec iov;
1273
1274         iov.iov_base = &regs;
1275         iov.iov_len = sizeof(regs);
1276         EXPECT_EQ(0, ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov)) {
1277                 TH_LOG("PTRACE_GETREGSET failed");
1278                 return -1;
1279         }
1280 #endif
1281
1282 #if defined(__mips__)
1283         if (regs.SYSCALL_NUM == __NR_O32_Linux)
1284                 return regs.SYSCALL_SYSCALL_NUM;
1285 #endif
1286         return regs.SYSCALL_NUM;
1287 }
1288
1289 /* Architecture-specific syscall changing routine. */
1290 void change_syscall(struct __test_metadata *_metadata,
1291                     pid_t tracee, int syscall)
1292 {
1293         int ret;
1294         ARCH_REGS regs;
1295 #ifdef HAVE_GETREGS
1296         ret = ptrace(PTRACE_GETREGS, tracee, 0, &regs);
1297 #else
1298         struct iovec iov;
1299         iov.iov_base = &regs;
1300         iov.iov_len = sizeof(regs);
1301         ret = ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov);
1302 #endif
1303         EXPECT_EQ(0, ret);
1304
1305 #if defined(__x86_64__) || defined(__i386__) || defined(__powerpc__) || \
1306     defined(__s390__)
1307         {
1308                 regs.SYSCALL_NUM = syscall;
1309         }
1310 #elif defined(__mips__)
1311         {
1312                 if (regs.SYSCALL_NUM == __NR_O32_Linux)
1313                         regs.SYSCALL_SYSCALL_NUM = syscall;
1314                 else
1315                         regs.SYSCALL_NUM = syscall;
1316         }
1317
1318 #elif defined(__arm__)
1319 # ifndef PTRACE_SET_SYSCALL
1320 #  define PTRACE_SET_SYSCALL   23
1321 # endif
1322         {
1323                 ret = ptrace(PTRACE_SET_SYSCALL, tracee, NULL, syscall);
1324                 EXPECT_EQ(0, ret);
1325         }
1326
1327 #elif defined(__aarch64__)
1328 # ifndef NT_ARM_SYSTEM_CALL
1329 #  define NT_ARM_SYSTEM_CALL 0x404
1330 # endif
1331         {
1332                 iov.iov_base = &syscall;
1333                 iov.iov_len = sizeof(syscall);
1334                 ret = ptrace(PTRACE_SETREGSET, tracee, NT_ARM_SYSTEM_CALL,
1335                              &iov);
1336                 EXPECT_EQ(0, ret);
1337         }
1338
1339 #else
1340         ASSERT_EQ(1, 0) {
1341                 TH_LOG("How is the syscall changed on this architecture?");
1342         }
1343 #endif
1344
1345         /* If syscall is skipped, change return value. */
1346         if (syscall == -1)
1347 #ifdef SYSCALL_NUM_RET_SHARE_REG
1348                 TH_LOG("Can't modify syscall return on this architecture");
1349 #else
1350                 regs.SYSCALL_RET = 1;
1351 #endif
1352
1353 #ifdef HAVE_GETREGS
1354         ret = ptrace(PTRACE_SETREGS, tracee, 0, &regs);
1355 #else
1356         iov.iov_base = &regs;
1357         iov.iov_len = sizeof(regs);
1358         ret = ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &iov);
1359 #endif
1360         EXPECT_EQ(0, ret);
1361 }
1362
1363 void tracer_syscall(struct __test_metadata *_metadata, pid_t tracee,
1364                     int status, void *args)
1365 {
1366         int ret;
1367         unsigned long msg;
1368
1369         /* Make sure we got the right message. */
1370         ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1371         EXPECT_EQ(0, ret);
1372
1373         /* Validate and take action on expected syscalls. */
1374         switch (msg) {
1375         case 0x1002:
1376                 /* change getpid to getppid. */
1377                 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee));
1378                 change_syscall(_metadata, tracee, __NR_getppid);
1379                 break;
1380         case 0x1003:
1381                 /* skip gettid. */
1382                 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee));
1383                 change_syscall(_metadata, tracee, -1);
1384                 break;
1385         case 0x1004:
1386                 /* do nothing (allow getppid) */
1387                 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee));
1388                 break;
1389         default:
1390                 EXPECT_EQ(0, msg) {
1391                         TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg);
1392                         kill(tracee, SIGKILL);
1393                 }
1394         }
1395
1396 }
1397
1398 FIXTURE_DATA(TRACE_syscall) {
1399         struct sock_fprog prog;
1400         pid_t tracer, mytid, mypid, parent;
1401 };
1402
1403 FIXTURE_SETUP(TRACE_syscall)
1404 {
1405         struct sock_filter filter[] = {
1406                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1407                         offsetof(struct seccomp_data, nr)),
1408                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
1409                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002),
1410                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1),
1411                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003),
1412                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1413                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004),
1414                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1415         };
1416
1417         memset(&self->prog, 0, sizeof(self->prog));
1418         self->prog.filter = malloc(sizeof(filter));
1419         ASSERT_NE(NULL, self->prog.filter);
1420         memcpy(self->prog.filter, filter, sizeof(filter));
1421         self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1422
1423         /* Prepare some testable syscall results. */
1424         self->mytid = syscall(__NR_gettid);
1425         ASSERT_GT(self->mytid, 0);
1426         ASSERT_NE(self->mytid, 1) {
1427                 TH_LOG("Running this test as init is not supported. :)");
1428         }
1429
1430         self->mypid = getpid();
1431         ASSERT_GT(self->mypid, 0);
1432         ASSERT_EQ(self->mytid, self->mypid);
1433
1434         self->parent = getppid();
1435         ASSERT_GT(self->parent, 0);
1436         ASSERT_NE(self->parent, self->mypid);
1437
1438         /* Launch tracer. */
1439         self->tracer = setup_trace_fixture(_metadata, tracer_syscall, NULL);
1440 }
1441
1442 FIXTURE_TEARDOWN(TRACE_syscall)
1443 {
1444         teardown_trace_fixture(_metadata, self->tracer);
1445         if (self->prog.filter)
1446                 free(self->prog.filter);
1447 }
1448
1449 TEST_F(TRACE_syscall, syscall_allowed)
1450 {
1451         long ret;
1452
1453         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1454         ASSERT_EQ(0, ret);
1455
1456         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1457         ASSERT_EQ(0, ret);
1458
1459         /* getppid works as expected (no changes). */
1460         EXPECT_EQ(self->parent, syscall(__NR_getppid));
1461         EXPECT_NE(self->mypid, syscall(__NR_getppid));
1462 }
1463
1464 TEST_F(TRACE_syscall, syscall_redirected)
1465 {
1466         long ret;
1467
1468         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1469         ASSERT_EQ(0, ret);
1470
1471         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1472         ASSERT_EQ(0, ret);
1473
1474         /* getpid has been redirected to getppid as expected. */
1475         EXPECT_EQ(self->parent, syscall(__NR_getpid));
1476         EXPECT_NE(self->mypid, syscall(__NR_getpid));
1477 }
1478
1479 TEST_F(TRACE_syscall, syscall_dropped)
1480 {
1481         long ret;
1482
1483         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1484         ASSERT_EQ(0, ret);
1485
1486         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1487         ASSERT_EQ(0, ret);
1488
1489 #ifdef SYSCALL_NUM_RET_SHARE_REG
1490         /* gettid has been skipped */
1491         EXPECT_EQ(-1, syscall(__NR_gettid));
1492 #else
1493         /* gettid has been skipped and an altered return value stored. */
1494         EXPECT_EQ(1, syscall(__NR_gettid));
1495 #endif
1496         EXPECT_NE(self->mytid, syscall(__NR_gettid));
1497 }
1498
1499 #ifndef __NR_seccomp
1500 # if defined(__i386__)
1501 #  define __NR_seccomp 354
1502 # elif defined(__x86_64__)
1503 #  define __NR_seccomp 317
1504 # elif defined(__arm__)
1505 #  define __NR_seccomp 383
1506 # elif defined(__aarch64__)
1507 #  define __NR_seccomp 277
1508 # elif defined(__powerpc__)
1509 #  define __NR_seccomp 358
1510 # elif defined(__s390__)
1511 #  define __NR_seccomp 348
1512 # else
1513 #  warning "seccomp syscall number unknown for this architecture"
1514 #  define __NR_seccomp 0xffff
1515 # endif
1516 #endif
1517
1518 #ifndef SECCOMP_SET_MODE_STRICT
1519 #define SECCOMP_SET_MODE_STRICT 0
1520 #endif
1521
1522 #ifndef SECCOMP_SET_MODE_FILTER
1523 #define SECCOMP_SET_MODE_FILTER 1
1524 #endif
1525
1526 #ifndef SECCOMP_FILTER_FLAG_TSYNC
1527 #define SECCOMP_FILTER_FLAG_TSYNC 1
1528 #endif
1529
1530 #ifndef seccomp
1531 int seccomp(unsigned int op, unsigned int flags, void *args)
1532 {
1533         errno = 0;
1534         return syscall(__NR_seccomp, op, flags, args);
1535 }
1536 #endif
1537
1538 TEST(seccomp_syscall)
1539 {
1540         struct sock_filter filter[] = {
1541                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1542         };
1543         struct sock_fprog prog = {
1544                 .len = (unsigned short)ARRAY_SIZE(filter),
1545                 .filter = filter,
1546         };
1547         long ret;
1548
1549         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1550         ASSERT_EQ(0, ret) {
1551                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1552         }
1553
1554         /* Reject insane operation. */
1555         ret = seccomp(-1, 0, &prog);
1556         ASSERT_NE(ENOSYS, errno) {
1557                 TH_LOG("Kernel does not support seccomp syscall!");
1558         }
1559         EXPECT_EQ(EINVAL, errno) {
1560                 TH_LOG("Did not reject crazy op value!");
1561         }
1562
1563         /* Reject strict with flags or pointer. */
1564         ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL);
1565         EXPECT_EQ(EINVAL, errno) {
1566                 TH_LOG("Did not reject mode strict with flags!");
1567         }
1568         ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog);
1569         EXPECT_EQ(EINVAL, errno) {
1570                 TH_LOG("Did not reject mode strict with uargs!");
1571         }
1572
1573         /* Reject insane args for filter. */
1574         ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog);
1575         EXPECT_EQ(EINVAL, errno) {
1576                 TH_LOG("Did not reject crazy filter flags!");
1577         }
1578         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL);
1579         EXPECT_EQ(EFAULT, errno) {
1580                 TH_LOG("Did not reject NULL filter!");
1581         }
1582
1583         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
1584         EXPECT_EQ(0, errno) {
1585                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s",
1586                         strerror(errno));
1587         }
1588 }
1589
1590 TEST(seccomp_syscall_mode_lock)
1591 {
1592         struct sock_filter filter[] = {
1593                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1594         };
1595         struct sock_fprog prog = {
1596                 .len = (unsigned short)ARRAY_SIZE(filter),
1597                 .filter = filter,
1598         };
1599         long ret;
1600
1601         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
1602         ASSERT_EQ(0, ret) {
1603                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1604         }
1605
1606         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
1607         ASSERT_NE(ENOSYS, errno) {
1608                 TH_LOG("Kernel does not support seccomp syscall!");
1609         }
1610         EXPECT_EQ(0, ret) {
1611                 TH_LOG("Could not install filter!");
1612         }
1613
1614         /* Make sure neither entry point will switch to strict. */
1615         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0);
1616         EXPECT_EQ(EINVAL, errno) {
1617                 TH_LOG("Switched to mode strict!");
1618         }
1619
1620         ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL);
1621         EXPECT_EQ(EINVAL, errno) {
1622                 TH_LOG("Switched to mode strict!");
1623         }
1624 }
1625
1626 TEST(TSYNC_first)
1627 {
1628         struct sock_filter filter[] = {
1629                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1630         };
1631         struct sock_fprog prog = {
1632                 .len = (unsigned short)ARRAY_SIZE(filter),
1633                 .filter = filter,
1634         };
1635         long ret;
1636
1637         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
1638         ASSERT_EQ(0, ret) {
1639                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1640         }
1641
1642         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
1643                       &prog);
1644         ASSERT_NE(ENOSYS, errno) {
1645                 TH_LOG("Kernel does not support seccomp syscall!");
1646         }
1647         EXPECT_EQ(0, ret) {
1648                 TH_LOG("Could not install initial filter with TSYNC!");
1649         }
1650 }
1651
1652 #define TSYNC_SIBLINGS 2
1653 struct tsync_sibling {
1654         pthread_t tid;
1655         pid_t system_tid;
1656         sem_t *started;
1657         pthread_cond_t *cond;
1658         pthread_mutex_t *mutex;
1659         int diverge;
1660         int num_waits;
1661         struct sock_fprog *prog;
1662         struct __test_metadata *metadata;
1663 };
1664
1665 FIXTURE_DATA(TSYNC) {
1666         struct sock_fprog root_prog, apply_prog;
1667         struct tsync_sibling sibling[TSYNC_SIBLINGS];
1668         sem_t started;
1669         pthread_cond_t cond;
1670         pthread_mutex_t mutex;
1671         int sibling_count;
1672 };
1673
1674 FIXTURE_SETUP(TSYNC)
1675 {
1676         struct sock_filter root_filter[] = {
1677                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1678         };
1679         struct sock_filter apply_filter[] = {
1680                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1681                         offsetof(struct seccomp_data, nr)),
1682                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1683                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1684                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1685         };
1686
1687         memset(&self->root_prog, 0, sizeof(self->root_prog));
1688         memset(&self->apply_prog, 0, sizeof(self->apply_prog));
1689         memset(&self->sibling, 0, sizeof(self->sibling));
1690         self->root_prog.filter = malloc(sizeof(root_filter));
1691         ASSERT_NE(NULL, self->root_prog.filter);
1692         memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter));
1693         self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter);
1694
1695         self->apply_prog.filter = malloc(sizeof(apply_filter));
1696         ASSERT_NE(NULL, self->apply_prog.filter);
1697         memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter));
1698         self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter);
1699
1700         self->sibling_count = 0;
1701         pthread_mutex_init(&self->mutex, NULL);
1702         pthread_cond_init(&self->cond, NULL);
1703         sem_init(&self->started, 0, 0);
1704         self->sibling[0].tid = 0;
1705         self->sibling[0].cond = &self->cond;
1706         self->sibling[0].started = &self->started;
1707         self->sibling[0].mutex = &self->mutex;
1708         self->sibling[0].diverge = 0;
1709         self->sibling[0].num_waits = 1;
1710         self->sibling[0].prog = &self->root_prog;
1711         self->sibling[0].metadata = _metadata;
1712         self->sibling[1].tid = 0;
1713         self->sibling[1].cond = &self->cond;
1714         self->sibling[1].started = &self->started;
1715         self->sibling[1].mutex = &self->mutex;
1716         self->sibling[1].diverge = 0;
1717         self->sibling[1].prog = &self->root_prog;
1718         self->sibling[1].num_waits = 1;
1719         self->sibling[1].metadata = _metadata;
1720 }
1721
1722 FIXTURE_TEARDOWN(TSYNC)
1723 {
1724         int sib = 0;
1725
1726         if (self->root_prog.filter)
1727                 free(self->root_prog.filter);
1728         if (self->apply_prog.filter)
1729                 free(self->apply_prog.filter);
1730
1731         for ( ; sib < self->sibling_count; ++sib) {
1732                 struct tsync_sibling *s = &self->sibling[sib];
1733                 void *status;
1734
1735                 if (!s->tid)
1736                         continue;
1737                 if (pthread_kill(s->tid, 0)) {
1738                         pthread_cancel(s->tid);
1739                         pthread_join(s->tid, &status);
1740                 }
1741         }
1742         pthread_mutex_destroy(&self->mutex);
1743         pthread_cond_destroy(&self->cond);
1744         sem_destroy(&self->started);
1745 }
1746
1747 void *tsync_sibling(void *data)
1748 {
1749         long ret = 0;
1750         struct tsync_sibling *me = data;
1751
1752         me->system_tid = syscall(__NR_gettid);
1753
1754         pthread_mutex_lock(me->mutex);
1755         if (me->diverge) {
1756                 /* Just re-apply the root prog to fork the tree */
1757                 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
1758                                 me->prog, 0, 0);
1759         }
1760         sem_post(me->started);
1761         /* Return outside of started so parent notices failures. */
1762         if (ret) {
1763                 pthread_mutex_unlock(me->mutex);
1764                 return (void *)SIBLING_EXIT_FAILURE;
1765         }
1766         do {
1767                 pthread_cond_wait(me->cond, me->mutex);
1768                 me->num_waits = me->num_waits - 1;
1769         } while (me->num_waits);
1770         pthread_mutex_unlock(me->mutex);
1771
1772         ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0);
1773         if (!ret)
1774                 return (void *)SIBLING_EXIT_NEWPRIVS;
1775         read(0, NULL, 0);
1776         return (void *)SIBLING_EXIT_UNKILLED;
1777 }
1778
1779 void tsync_start_sibling(struct tsync_sibling *sibling)
1780 {
1781         pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling);
1782 }
1783
1784 TEST_F(TSYNC, siblings_fail_prctl)
1785 {
1786         long ret;
1787         void *status;
1788         struct sock_filter filter[] = {
1789                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1790                         offsetof(struct seccomp_data, nr)),
1791                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
1792                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL),
1793                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1794         };
1795         struct sock_fprog prog = {
1796                 .len = (unsigned short)ARRAY_SIZE(filter),
1797                 .filter = filter,
1798         };
1799
1800         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
1801                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1802         }
1803
1804         /* Check prctl failure detection by requesting sib 0 diverge. */
1805         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
1806         ASSERT_NE(ENOSYS, errno) {
1807                 TH_LOG("Kernel does not support seccomp syscall!");
1808         }
1809         ASSERT_EQ(0, ret) {
1810                 TH_LOG("setting filter failed");
1811         }
1812
1813         self->sibling[0].diverge = 1;
1814         tsync_start_sibling(&self->sibling[0]);
1815         tsync_start_sibling(&self->sibling[1]);
1816
1817         while (self->sibling_count < TSYNC_SIBLINGS) {
1818                 sem_wait(&self->started);
1819                 self->sibling_count++;
1820         }
1821
1822         /* Signal the threads to clean up*/
1823         pthread_mutex_lock(&self->mutex);
1824         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
1825                 TH_LOG("cond broadcast non-zero");
1826         }
1827         pthread_mutex_unlock(&self->mutex);
1828
1829         /* Ensure diverging sibling failed to call prctl. */
1830         pthread_join(self->sibling[0].tid, &status);
1831         EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status);
1832         pthread_join(self->sibling[1].tid, &status);
1833         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
1834 }
1835
1836 TEST_F(TSYNC, two_siblings_with_ancestor)
1837 {
1838         long ret;
1839         void *status;
1840
1841         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
1842                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1843         }
1844
1845         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
1846         ASSERT_NE(ENOSYS, errno) {
1847                 TH_LOG("Kernel does not support seccomp syscall!");
1848         }
1849         ASSERT_EQ(0, ret) {
1850                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
1851         }
1852         tsync_start_sibling(&self->sibling[0]);
1853         tsync_start_sibling(&self->sibling[1]);
1854
1855         while (self->sibling_count < TSYNC_SIBLINGS) {
1856                 sem_wait(&self->started);
1857                 self->sibling_count++;
1858         }
1859
1860         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
1861                       &self->apply_prog);
1862         ASSERT_EQ(0, ret) {
1863                 TH_LOG("Could install filter on all threads!");
1864         }
1865         /* Tell the siblings to test the policy */
1866         pthread_mutex_lock(&self->mutex);
1867         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
1868                 TH_LOG("cond broadcast non-zero");
1869         }
1870         pthread_mutex_unlock(&self->mutex);
1871         /* Ensure they are both killed and don't exit cleanly. */
1872         pthread_join(self->sibling[0].tid, &status);
1873         EXPECT_EQ(0x0, (long)status);
1874         pthread_join(self->sibling[1].tid, &status);
1875         EXPECT_EQ(0x0, (long)status);
1876 }
1877
1878 TEST_F(TSYNC, two_sibling_want_nnp)
1879 {
1880         void *status;
1881
1882         /* start siblings before any prctl() operations */
1883         tsync_start_sibling(&self->sibling[0]);
1884         tsync_start_sibling(&self->sibling[1]);
1885         while (self->sibling_count < TSYNC_SIBLINGS) {
1886                 sem_wait(&self->started);
1887                 self->sibling_count++;
1888         }
1889
1890         /* Tell the siblings to test no policy */
1891         pthread_mutex_lock(&self->mutex);
1892         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
1893                 TH_LOG("cond broadcast non-zero");
1894         }
1895         pthread_mutex_unlock(&self->mutex);
1896
1897         /* Ensure they are both upset about lacking nnp. */
1898         pthread_join(self->sibling[0].tid, &status);
1899         EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
1900         pthread_join(self->sibling[1].tid, &status);
1901         EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
1902 }
1903
1904 TEST_F(TSYNC, two_siblings_with_no_filter)
1905 {
1906         long ret;
1907         void *status;
1908
1909         /* start siblings before any prctl() operations */
1910         tsync_start_sibling(&self->sibling[0]);
1911         tsync_start_sibling(&self->sibling[1]);
1912         while (self->sibling_count < TSYNC_SIBLINGS) {
1913                 sem_wait(&self->started);
1914                 self->sibling_count++;
1915         }
1916
1917         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
1918                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1919         }
1920
1921         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
1922                       &self->apply_prog);
1923         ASSERT_NE(ENOSYS, errno) {
1924                 TH_LOG("Kernel does not support seccomp syscall!");
1925         }
1926         ASSERT_EQ(0, ret) {
1927                 TH_LOG("Could install filter on all threads!");
1928         }
1929
1930         /* Tell the siblings to test the policy */
1931         pthread_mutex_lock(&self->mutex);
1932         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
1933                 TH_LOG("cond broadcast non-zero");
1934         }
1935         pthread_mutex_unlock(&self->mutex);
1936
1937         /* Ensure they are both killed and don't exit cleanly. */
1938         pthread_join(self->sibling[0].tid, &status);
1939         EXPECT_EQ(0x0, (long)status);
1940         pthread_join(self->sibling[1].tid, &status);
1941         EXPECT_EQ(0x0, (long)status);
1942 }
1943
1944 TEST_F(TSYNC, two_siblings_with_one_divergence)
1945 {
1946         long ret;
1947         void *status;
1948
1949         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
1950                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1951         }
1952
1953         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
1954         ASSERT_NE(ENOSYS, errno) {
1955                 TH_LOG("Kernel does not support seccomp syscall!");
1956         }
1957         ASSERT_EQ(0, ret) {
1958                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
1959         }
1960         self->sibling[0].diverge = 1;
1961         tsync_start_sibling(&self->sibling[0]);
1962         tsync_start_sibling(&self->sibling[1]);
1963
1964         while (self->sibling_count < TSYNC_SIBLINGS) {
1965                 sem_wait(&self->started);
1966                 self->sibling_count++;
1967         }
1968
1969         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
1970                       &self->apply_prog);
1971         ASSERT_EQ(self->sibling[0].system_tid, ret) {
1972                 TH_LOG("Did not fail on diverged sibling.");
1973         }
1974
1975         /* Wake the threads */
1976         pthread_mutex_lock(&self->mutex);
1977         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
1978                 TH_LOG("cond broadcast non-zero");
1979         }
1980         pthread_mutex_unlock(&self->mutex);
1981
1982         /* Ensure they are both unkilled. */
1983         pthread_join(self->sibling[0].tid, &status);
1984         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
1985         pthread_join(self->sibling[1].tid, &status);
1986         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
1987 }
1988
1989 TEST_F(TSYNC, two_siblings_not_under_filter)
1990 {
1991         long ret, sib;
1992         void *status;
1993
1994         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
1995                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1996         }
1997
1998         /*
1999          * Sibling 0 will have its own seccomp policy
2000          * and Sibling 1 will not be under seccomp at
2001          * all. Sibling 1 will enter seccomp and 0
2002          * will cause failure.
2003          */
2004         self->sibling[0].diverge = 1;
2005         tsync_start_sibling(&self->sibling[0]);
2006         tsync_start_sibling(&self->sibling[1]);
2007
2008         while (self->sibling_count < TSYNC_SIBLINGS) {
2009                 sem_wait(&self->started);
2010                 self->sibling_count++;
2011         }
2012
2013         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2014         ASSERT_NE(ENOSYS, errno) {
2015                 TH_LOG("Kernel does not support seccomp syscall!");
2016         }
2017         ASSERT_EQ(0, ret) {
2018                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2019         }
2020
2021         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2022                       &self->apply_prog);
2023         ASSERT_EQ(ret, self->sibling[0].system_tid) {
2024                 TH_LOG("Did not fail on diverged sibling.");
2025         }
2026         sib = 1;
2027         if (ret == self->sibling[0].system_tid)
2028                 sib = 0;
2029
2030         pthread_mutex_lock(&self->mutex);
2031
2032         /* Increment the other siblings num_waits so we can clean up
2033          * the one we just saw.
2034          */
2035         self->sibling[!sib].num_waits += 1;
2036
2037         /* Signal the thread to clean up*/
2038         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2039                 TH_LOG("cond broadcast non-zero");
2040         }
2041         pthread_mutex_unlock(&self->mutex);
2042         pthread_join(self->sibling[sib].tid, &status);
2043         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2044         /* Poll for actual task death. pthread_join doesn't guarantee it. */
2045         while (!kill(self->sibling[sib].system_tid, 0))
2046                 sleep(0.1);
2047         /* Switch to the remaining sibling */
2048         sib = !sib;
2049
2050         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2051                       &self->apply_prog);
2052         ASSERT_EQ(0, ret) {
2053                 TH_LOG("Expected the remaining sibling to sync");
2054         };
2055
2056         pthread_mutex_lock(&self->mutex);
2057
2058         /* If remaining sibling didn't have a chance to wake up during
2059          * the first broadcast, manually reduce the num_waits now.
2060          */
2061         if (self->sibling[sib].num_waits > 1)
2062                 self->sibling[sib].num_waits = 1;
2063         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2064                 TH_LOG("cond broadcast non-zero");
2065         }
2066         pthread_mutex_unlock(&self->mutex);
2067         pthread_join(self->sibling[sib].tid, &status);
2068         EXPECT_EQ(0, (long)status);
2069         /* Poll for actual task death. pthread_join doesn't guarantee it. */
2070         while (!kill(self->sibling[sib].system_tid, 0))
2071                 sleep(0.1);
2072
2073         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2074                       &self->apply_prog);
2075         ASSERT_EQ(0, ret);  /* just us chickens */
2076 }
2077
2078 /* Make sure restarted syscalls are seen directly as "restart_syscall". */
2079 TEST(syscall_restart)
2080 {
2081         long ret;
2082         unsigned long msg;
2083         pid_t child_pid;
2084         int pipefd[2];
2085         int status;
2086         siginfo_t info = { };
2087         struct sock_filter filter[] = {
2088                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2089                          offsetof(struct seccomp_data, nr)),
2090
2091 #ifdef __NR_sigreturn
2092                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 6, 0),
2093 #endif
2094                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 5, 0),
2095                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 4, 0),
2096                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 3, 0),
2097                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 4, 0),
2098                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0),
2099
2100                 /* Allow __NR_write for easy logging. */
2101                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1),
2102                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2103                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2104                 /* The nanosleep jump target. */
2105                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100),
2106                 /* The restart_syscall jump target. */
2107                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200),
2108         };
2109         struct sock_fprog prog = {
2110                 .len = (unsigned short)ARRAY_SIZE(filter),
2111                 .filter = filter,
2112         };
2113 #if defined(__arm__)
2114         struct utsname utsbuf;
2115 #endif
2116
2117         ASSERT_EQ(0, pipe(pipefd));
2118
2119         child_pid = fork();
2120         ASSERT_LE(0, child_pid);
2121         if (child_pid == 0) {
2122                 /* Child uses EXPECT not ASSERT to deliver status correctly. */
2123                 char buf = ' ';
2124                 struct timespec timeout = { };
2125
2126                 /* Attach parent as tracer and stop. */
2127                 EXPECT_EQ(0, ptrace(PTRACE_TRACEME));
2128                 EXPECT_EQ(0, raise(SIGSTOP));
2129
2130                 EXPECT_EQ(0, close(pipefd[1]));
2131
2132                 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2133                         TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2134                 }
2135
2136                 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2137                 EXPECT_EQ(0, ret) {
2138                         TH_LOG("Failed to install filter!");
2139                 }
2140
2141                 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2142                         TH_LOG("Failed to read() sync from parent");
2143                 }
2144                 EXPECT_EQ('.', buf) {
2145                         TH_LOG("Failed to get sync data from read()");
2146                 }
2147
2148                 /* Start nanosleep to be interrupted. */
2149                 timeout.tv_sec = 1;
2150                 errno = 0;
2151                 EXPECT_EQ(0, nanosleep(&timeout, NULL)) {
2152                         TH_LOG("Call to nanosleep() failed (errno %d)", errno);
2153                 }
2154
2155                 /* Read final sync from parent. */
2156                 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2157                         TH_LOG("Failed final read() from parent");
2158                 }
2159                 EXPECT_EQ('!', buf) {
2160                         TH_LOG("Failed to get final data from read()");
2161                 }
2162
2163                 /* Directly report the status of our test harness results. */
2164                 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS
2165                                                      : EXIT_FAILURE);
2166         }
2167         EXPECT_EQ(0, close(pipefd[0]));
2168
2169         /* Attach to child, setup options, and release. */
2170         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2171         ASSERT_EQ(true, WIFSTOPPED(status));
2172         ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL,
2173                             PTRACE_O_TRACESECCOMP));
2174         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2175         ASSERT_EQ(1, write(pipefd[1], ".", 1));
2176
2177         /* Wait for nanosleep() to start. */
2178         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2179         ASSERT_EQ(true, WIFSTOPPED(status));
2180         ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2181         ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2182         ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2183         ASSERT_EQ(0x100, msg);
2184         EXPECT_EQ(__NR_nanosleep, get_syscall(_metadata, child_pid));
2185
2186         /* Might as well check siginfo for sanity while we're here. */
2187         ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2188         ASSERT_EQ(SIGTRAP, info.si_signo);
2189         ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code);
2190         EXPECT_EQ(0, info.si_errno);
2191         EXPECT_EQ(getuid(), info.si_uid);
2192         /* Verify signal delivery came from child (seccomp-triggered). */
2193         EXPECT_EQ(child_pid, info.si_pid);
2194
2195         /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */
2196         ASSERT_EQ(0, kill(child_pid, SIGSTOP));
2197         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2198         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2199         ASSERT_EQ(true, WIFSTOPPED(status));
2200         ASSERT_EQ(SIGSTOP, WSTOPSIG(status));
2201         /* Verify signal delivery came from parent now. */
2202         ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2203         EXPECT_EQ(getpid(), info.si_pid);
2204
2205         /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */
2206         ASSERT_EQ(0, kill(child_pid, SIGCONT));
2207         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2208         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2209         ASSERT_EQ(true, WIFSTOPPED(status));
2210         ASSERT_EQ(SIGCONT, WSTOPSIG(status));
2211         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2212
2213         /* Wait for restart_syscall() to start. */
2214         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2215         ASSERT_EQ(true, WIFSTOPPED(status));
2216         ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2217         ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2218         ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2219
2220         ASSERT_EQ(0x200, msg);
2221         ret = get_syscall(_metadata, child_pid);
2222 #if defined(__arm__)
2223         /*
2224          * FIXME:
2225          * - native ARM registers do NOT expose true syscall.
2226          * - compat ARM registers on ARM64 DO expose true syscall.
2227          */
2228         ASSERT_EQ(0, uname(&utsbuf));
2229         if (strncmp(utsbuf.machine, "arm", 3) == 0) {
2230                 EXPECT_EQ(__NR_nanosleep, ret);
2231         } else
2232 #endif
2233         {
2234                 EXPECT_EQ(__NR_restart_syscall, ret);
2235         }
2236
2237         /* Write again to end test. */
2238         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2239         ASSERT_EQ(1, write(pipefd[1], "!", 1));
2240         EXPECT_EQ(0, close(pipefd[1]));
2241
2242         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2243         if (WIFSIGNALED(status) || WEXITSTATUS(status))
2244                 _metadata->passed = 0;
2245 }
2246
2247 /*
2248  * TODO:
2249  * - add microbenchmarks
2250  * - expand NNP testing
2251  * - better arch-specific TRACE and TRAP handlers.
2252  * - endianness checking when appropriate
2253  * - 64-bit arg prodding
2254  * - arch value testing (x86 modes especially)
2255  * - ...
2256  */
2257
2258 TEST_HARNESS_MAIN