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 "../kselftest_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 #define ERRNO_FILTER(name, errno)                                       \
 545         struct sock_filter _read_filter_##name[] = {                    \
 546                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,                          \
 547                         offsetof(struct seccomp_data, nr)),             \
 548                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),       \
 549                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | errno),     \
 550                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),             \
 551         };                                                              \
 552         struct sock_fprog prog_##name = {                               \
 553                 .len = (unsigned short)ARRAY_SIZE(_read_filter_##name), \
 554                 .filter = _read_filter_##name,                          \
 555         }
 556
 557 /* Make sure basic errno values are correctly passed through a filter. */
 558 TEST(ERRNO_valid)
 559 {
 560         ERRNO_FILTER(valid, E2BIG);
 561         long ret;
 562         pid_t parent = getppid();
 563
 564         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 565         ASSERT_EQ(0, ret);
 566
 567         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_valid);
 568         ASSERT_EQ(0, ret);
 569
 570         EXPECT_EQ(parent, syscall(__NR_getppid));
 571         EXPECT_EQ(-1, read(0, NULL, 0));
 572         EXPECT_EQ(E2BIG, errno);
 573 }
 574
 575 /* Make sure an errno of zero is correctly handled by the arch code. */
 576 TEST(ERRNO_zero)
 577 {
 578         ERRNO_FILTER(zero, 0);
 579         long ret;
 580         pid_t parent = getppid();
 581
 582         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 583         ASSERT_EQ(0, ret);
 584
 585         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_zero);
 586         ASSERT_EQ(0, ret);
 587
 588         EXPECT_EQ(parent, syscall(__NR_getppid));
 589         /* "errno" of 0 is ok. */
 590         EXPECT_EQ(0, read(0, NULL, 0));
 591 }
 592
 593 /*
 594  * The SECCOMP_RET_DATA mask is 16 bits wide, but errno is smaller.
 595  * This tests that the errno value gets capped correctly, fixed by
 596  * 580c57f10768 ("seccomp: cap SECCOMP_RET_ERRNO data to MAX_ERRNO").
 597  */
 598 TEST(ERRNO_capped)
 599 {
 600         ERRNO_FILTER(capped, 4096);
 601         long ret;
 602         pid_t parent = getppid();
 603
 604         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 605         ASSERT_EQ(0, ret);
 606
 607         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_capped);
 608         ASSERT_EQ(0, ret);
 609
 610         EXPECT_EQ(parent, syscall(__NR_getppid));
 611         EXPECT_EQ(-1, read(0, NULL, 0));
 612         EXPECT_EQ(4095, errno);
 613 }
 614
 615 /*
 616  * Filters are processed in reverse order: last applied is executed first.
 617  * Since only the SECCOMP_RET_ACTION mask is tested for return values, the
 618  * SECCOMP_RET_DATA mask results will follow the most recently applied
 619  * matching filter return (and not the lowest or highest value).
 620  */
 621 TEST(ERRNO_order)
 622 {
 623         ERRNO_FILTER(first,  11);
 624         ERRNO_FILTER(second, 13);
 625         ERRNO_FILTER(third,  12);
 626         long ret;
 627         pid_t parent = getppid();
 628
 629         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 630         ASSERT_EQ(0, ret);
 631
 632         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_first);
 633         ASSERT_EQ(0, ret);
 634
 635         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_second);
 636         ASSERT_EQ(0, ret);
 637
 638         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_third);
 639         ASSERT_EQ(0, ret);
 640
 641         EXPECT_EQ(parent, syscall(__NR_getppid));
 642         EXPECT_EQ(-1, read(0, NULL, 0));
 643         EXPECT_EQ(12, errno);
 644 }
 645
 646 FIXTURE_DATA(TRAP) {
 647         struct sock_fprog prog;
 648 };
 649
 650 FIXTURE_SETUP(TRAP)
 651 {
 652         struct sock_filter filter[] = {
 653                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 654                         offsetof(struct seccomp_data, nr)),
 655                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
 656                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
 657                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 658         };
 659
 660         memset(&self->prog, 0, sizeof(self->prog));
 661         self->prog.filter = malloc(sizeof(filter));
 662         ASSERT_NE(NULL, self->prog.filter);
 663         memcpy(self->prog.filter, filter, sizeof(filter));
 664         self->prog.len = (unsigned short)ARRAY_SIZE(filter);
 665 }
 666
 667 FIXTURE_TEARDOWN(TRAP)
 668 {
 669         if (self->prog.filter)
 670                 free(self->prog.filter);
 671 }
 672
 673 TEST_F_SIGNAL(TRAP, dfl, SIGSYS)
 674 {
 675         long ret;
 676
 677         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 678         ASSERT_EQ(0, ret);
 679
 680         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
 681         ASSERT_EQ(0, ret);
 682         syscall(__NR_getpid);
 683 }
 684
 685 /* Ensure that SIGSYS overrides SIG_IGN */
 686 TEST_F_SIGNAL(TRAP, ign, SIGSYS)
 687 {
 688         long ret;
 689
 690         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 691         ASSERT_EQ(0, ret);
 692
 693         signal(SIGSYS, SIG_IGN);
 694
 695         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
 696         ASSERT_EQ(0, ret);
 697         syscall(__NR_getpid);
 698 }
 699
 700 static struct siginfo TRAP_info;
 701 static volatile int TRAP_nr;
 702 static void TRAP_action(int nr, siginfo_t *info, void *void_context)
 703 {
 704         memcpy(&TRAP_info, info, sizeof(TRAP_info));
 705         TRAP_nr = nr;
 706 }
 707
 708 TEST_F(TRAP, handler)
 709 {
 710         int ret, test;
 711         struct sigaction act;
 712         sigset_t mask;
 713
 714         memset(&act, 0, sizeof(act));
 715         sigemptyset(&mask);
 716         sigaddset(&mask, SIGSYS);
 717
 718         act.sa_sigaction = &TRAP_action;
 719         act.sa_flags = SA_SIGINFO;
 720         ret = sigaction(SIGSYS, &act, NULL);
 721         ASSERT_EQ(0, ret) {
 722                 TH_LOG("sigaction failed");
 723         }
 724         ret = sigprocmask(SIG_UNBLOCK, &mask, NULL);
 725         ASSERT_EQ(0, ret) {
 726                 TH_LOG("sigprocmask failed");
 727         }
 728
 729         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 730         ASSERT_EQ(0, ret);
 731         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
 732         ASSERT_EQ(0, ret);
 733         TRAP_nr = 0;
 734         memset(&TRAP_info, 0, sizeof(TRAP_info));
 735         /* Expect the registers to be rolled back. (nr = error) may vary
 736          * based on arch. */
 737         ret = syscall(__NR_getpid);
 738         /* Silence gcc warning about volatile. */
 739         test = TRAP_nr;
 740         EXPECT_EQ(SIGSYS, test);
 741         struct local_sigsys {
 742                 void *_call_addr;       /* calling user insn */
 743                 int _syscall;           /* triggering system call number */
 744                 unsigned int _arch;     /* AUDIT_ARCH_* of syscall */
 745         } *sigsys = (struct local_sigsys *)
 746 #ifdef si_syscall
 747                 &(TRAP_info.si_call_addr);
 748 #else
 749                 &TRAP_info.si_pid;
 750 #endif
 751         EXPECT_EQ(__NR_getpid, sigsys->_syscall);
 752         /* Make sure arch is non-zero. */
 753         EXPECT_NE(0, sigsys->_arch);
 754         EXPECT_NE(0, (unsigned long)sigsys->_call_addr);
 755 }
 756
 757 FIXTURE_DATA(precedence) {
 758         struct sock_fprog allow;
 759         struct sock_fprog trace;
 760         struct sock_fprog error;
 761         struct sock_fprog trap;
 762         struct sock_fprog kill;
 763 };
 764
 765 FIXTURE_SETUP(precedence)
 766 {
 767         struct sock_filter allow_insns[] = {
 768                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 769         };
 770         struct sock_filter trace_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_TRACE),
 776         };
 777         struct sock_filter error_insns[] = {
 778                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 779                         offsetof(struct seccomp_data, nr)),
 780                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
 781                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 782                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO),
 783         };
 784         struct sock_filter trap_insns[] = {
 785                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 786                         offsetof(struct seccomp_data, nr)),
 787                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
 788                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 789                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
 790         };
 791         struct sock_filter kill_insns[] = {
 792                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 793                         offsetof(struct seccomp_data, nr)),
 794                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
 795                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 796                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
 797         };
 798
 799         memset(self, 0, sizeof(*self));
 800 #define FILTER_ALLOC(_x) \
 801         self->_x.filter = malloc(sizeof(_x##_insns)); \
 802         ASSERT_NE(NULL, self->_x.filter); \
 803         memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \
 804         self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns)
 805         FILTER_ALLOC(allow);
 806         FILTER_ALLOC(trace);
 807         FILTER_ALLOC(error);
 808         FILTER_ALLOC(trap);
 809         FILTER_ALLOC(kill);
 810 }
 811
 812 FIXTURE_TEARDOWN(precedence)
 813 {
 814 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter)
 815         FILTER_FREE(allow);
 816         FILTER_FREE(trace);
 817         FILTER_FREE(error);
 818         FILTER_FREE(trap);
 819         FILTER_FREE(kill);
 820 }
 821
 822 TEST_F(precedence, allow_ok)
 823 {
 824         pid_t parent, res = 0;
 825         long ret;
 826
 827         parent = getppid();
 828         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 829         ASSERT_EQ(0, ret);
 830
 831         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 832         ASSERT_EQ(0, ret);
 833         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 834         ASSERT_EQ(0, ret);
 835         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 836         ASSERT_EQ(0, ret);
 837         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
 838         ASSERT_EQ(0, ret);
 839         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
 840         ASSERT_EQ(0, ret);
 841         /* Should work just fine. */
 842         res = syscall(__NR_getppid);
 843         EXPECT_EQ(parent, res);
 844 }
 845
 846 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS)
 847 {
 848         pid_t parent, res = 0;
 849         long ret;
 850
 851         parent = getppid();
 852         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 853         ASSERT_EQ(0, ret);
 854
 855         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 856         ASSERT_EQ(0, ret);
 857         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 858         ASSERT_EQ(0, ret);
 859         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 860         ASSERT_EQ(0, ret);
 861         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
 862         ASSERT_EQ(0, ret);
 863         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
 864         ASSERT_EQ(0, ret);
 865         /* Should work just fine. */
 866         res = syscall(__NR_getppid);
 867         EXPECT_EQ(parent, res);
 868         /* getpid() should never return. */
 869         res = syscall(__NR_getpid);
 870         EXPECT_EQ(0, res);
 871 }
 872
 873 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS)
 874 {
 875         pid_t parent;
 876         long ret;
 877
 878         parent = getppid();
 879         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 880         ASSERT_EQ(0, ret);
 881
 882         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 883         ASSERT_EQ(0, ret);
 884         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
 885         ASSERT_EQ(0, ret);
 886         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 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->trap);
 891         ASSERT_EQ(0, ret);
 892         /* Should work just fine. */
 893         EXPECT_EQ(parent, syscall(__NR_getppid));
 894         /* getpid() should never return. */
 895         EXPECT_EQ(0, syscall(__NR_getpid));
 896 }
 897
 898 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS)
 899 {
 900         pid_t parent;
 901         long ret;
 902
 903         parent = getppid();
 904         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 905         ASSERT_EQ(0, ret);
 906
 907         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 908         ASSERT_EQ(0, ret);
 909         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 910         ASSERT_EQ(0, ret);
 911         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 912         ASSERT_EQ(0, ret);
 913         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
 914         ASSERT_EQ(0, ret);
 915         /* Should work just fine. */
 916         EXPECT_EQ(parent, syscall(__NR_getppid));
 917         /* getpid() should never return. */
 918         EXPECT_EQ(0, syscall(__NR_getpid));
 919 }
 920
 921 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS)
 922 {
 923         pid_t parent;
 924         long ret;
 925
 926         parent = getppid();
 927         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 928         ASSERT_EQ(0, ret);
 929
 930         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 931         ASSERT_EQ(0, ret);
 932         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
 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         /* getpid() should never return. */
 941         EXPECT_EQ(0, syscall(__NR_getpid));
 942 }
 943
 944 TEST_F(precedence, errno_is_third)
 945 {
 946         pid_t parent;
 947         long ret;
 948
 949         parent = getppid();
 950         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 951         ASSERT_EQ(0, ret);
 952
 953         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 954         ASSERT_EQ(0, ret);
 955         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 956         ASSERT_EQ(0, ret);
 957         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 958         ASSERT_EQ(0, ret);
 959         /* Should work just fine. */
 960         EXPECT_EQ(parent, syscall(__NR_getppid));
 961         EXPECT_EQ(0, syscall(__NR_getpid));
 962 }
 963
 964 TEST_F(precedence, errno_is_third_in_any_order)
 965 {
 966         pid_t parent;
 967         long ret;
 968
 969         parent = getppid();
 970         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 971         ASSERT_EQ(0, ret);
 972
 973         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 974         ASSERT_EQ(0, ret);
 975         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 976         ASSERT_EQ(0, ret);
 977         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 978         ASSERT_EQ(0, ret);
 979         /* Should work just fine. */
 980         EXPECT_EQ(parent, syscall(__NR_getppid));
 981         EXPECT_EQ(0, syscall(__NR_getpid));
 982 }
 983
 984 TEST_F(precedence, trace_is_fourth)
 985 {
 986         pid_t parent;
 987         long ret;
 988
 989         parent = getppid();
 990         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 991         ASSERT_EQ(0, ret);
 992
 993         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 994         ASSERT_EQ(0, ret);
 995         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 996         ASSERT_EQ(0, ret);
 997         /* Should work just fine. */
 998         EXPECT_EQ(parent, syscall(__NR_getppid));
 999         /* No ptracer */
1000         EXPECT_EQ(-1, syscall(__NR_getpid));
1001 }
1002
1003 TEST_F(precedence, trace_is_fourth_in_any_order)
1004 {
1005         pid_t parent;
1006         long ret;
1007
1008         parent = getppid();
1009         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1010         ASSERT_EQ(0, ret);
1011
1012         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1013         ASSERT_EQ(0, ret);
1014         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1015         ASSERT_EQ(0, ret);
1016         /* Should work just fine. */
1017         EXPECT_EQ(parent, syscall(__NR_getppid));
1018         /* No ptracer */
1019         EXPECT_EQ(-1, syscall(__NR_getpid));
1020 }
1021
1022 #ifndef PTRACE_O_TRACESECCOMP
1023 #define PTRACE_O_TRACESECCOMP   0x00000080
1024 #endif
1025
1026 /* Catch the Ubuntu 12.04 value error. */
1027 #if PTRACE_EVENT_SECCOMP != 7
1028 #undef PTRACE_EVENT_SECCOMP
1029 #endif
1030
1031 #ifndef PTRACE_EVENT_SECCOMP
1032 #define PTRACE_EVENT_SECCOMP 7
1033 #endif
1034
1035 #define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP)
1036 bool tracer_running;
1037 void tracer_stop(int sig)
1038 {
1039         tracer_running = false;
1040 }
1041
1042 typedef void tracer_func_t(struct __test_metadata *_metadata,
1043                            pid_t tracee, int status, void *args);
1044
1045 void start_tracer(struct __test_metadata *_metadata, int fd, pid_t tracee,
1046             tracer_func_t tracer_func, void *args, bool ptrace_syscall)
1047 {
1048         int ret = -1;
1049         struct sigaction action = {
1050                 .sa_handler = tracer_stop,
1051         };
1052
1053         /* Allow external shutdown. */
1054         tracer_running = true;
1055         ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL));
1056
1057         errno = 0;
1058         while (ret == -1 && errno != EINVAL)
1059                 ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0);
1060         ASSERT_EQ(0, ret) {
1061                 kill(tracee, SIGKILL);
1062         }
1063         /* Wait for attach stop */
1064         wait(NULL);
1065
1066         ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, ptrace_syscall ?
1067                                                       PTRACE_O_TRACESYSGOOD :
1068                                                       PTRACE_O_TRACESECCOMP);
1069         ASSERT_EQ(0, ret) {
1070                 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP");
1071                 kill(tracee, SIGKILL);
1072         }
1073         ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1074                      tracee, NULL, 0);
1075         ASSERT_EQ(0, ret);
1076
1077         /* Unblock the tracee */
1078         ASSERT_EQ(1, write(fd, "A", 1));
1079         ASSERT_EQ(0, close(fd));
1080
1081         /* Run until we're shut down. Must assert to stop execution. */
1082         while (tracer_running) {
1083                 int status;
1084
1085                 if (wait(&status) != tracee)
1086                         continue;
1087                 if (WIFSIGNALED(status) || WIFEXITED(status))
1088                         /* Child is dead. Time to go. */
1089                         return;
1090
1091                 /* Check if this is a seccomp event. */
1092                 ASSERT_EQ(!ptrace_syscall, IS_SECCOMP_EVENT(status));
1093
1094                 tracer_func(_metadata, tracee, status, args);
1095
1096                 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1097                              tracee, NULL, 0);
1098                 ASSERT_EQ(0, ret);
1099         }
1100         /* Directly report the status of our test harness results. */
1101         syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS : EXIT_FAILURE);
1102 }
1103
1104 /* Common tracer setup/teardown functions. */
1105 void cont_handler(int num)
1106 { }
1107 pid_t setup_trace_fixture(struct __test_metadata *_metadata,
1108                           tracer_func_t func, void *args, bool ptrace_syscall)
1109 {
1110         char sync;
1111         int pipefd[2];
1112         pid_t tracer_pid;
1113         pid_t tracee = getpid();
1114
1115         /* Setup a pipe for clean synchronization. */
1116         ASSERT_EQ(0, pipe(pipefd));
1117
1118         /* Fork a child which we'll promote to tracer */
1119         tracer_pid = fork();
1120         ASSERT_LE(0, tracer_pid);
1121         signal(SIGALRM, cont_handler);
1122         if (tracer_pid == 0) {
1123                 close(pipefd[0]);
1124                 start_tracer(_metadata, pipefd[1], tracee, func, args,
1125                              ptrace_syscall);
1126                 syscall(__NR_exit, 0);
1127         }
1128         close(pipefd[1]);
1129         prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0);
1130         read(pipefd[0], &sync, 1);
1131         close(pipefd[0]);
1132
1133         return tracer_pid;
1134 }
1135 void teardown_trace_fixture(struct __test_metadata *_metadata,
1136                             pid_t tracer)
1137 {
1138         if (tracer) {
1139                 int status;
1140                 /*
1141                  * Extract the exit code from the other process and
1142                  * adopt it for ourselves in case its asserts failed.
1143                  */
1144                 ASSERT_EQ(0, kill(tracer, SIGUSR1));
1145                 ASSERT_EQ(tracer, waitpid(tracer, &status, 0));
1146                 if (WEXITSTATUS(status))
1147                         _metadata->passed = 0;
1148         }
1149 }
1150
1151 /* "poke" tracer arguments and function. */
1152 struct tracer_args_poke_t {
1153         unsigned long poke_addr;
1154 };
1155
1156 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status,
1157                  void *args)
1158 {
1159         int ret;
1160         unsigned long msg;
1161         struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args;
1162
1163         ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1164         EXPECT_EQ(0, ret);
1165         /* If this fails, don't try to recover. */
1166         ASSERT_EQ(0x1001, msg) {
1167                 kill(tracee, SIGKILL);
1168         }
1169         /*
1170          * Poke in the message.
1171          * Registers are not touched to try to keep this relatively arch
1172          * agnostic.
1173          */
1174         ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001);
1175         EXPECT_EQ(0, ret);
1176 }
1177
1178 FIXTURE_DATA(TRACE_poke) {
1179         struct sock_fprog prog;
1180         pid_t tracer;
1181         long poked;
1182         struct tracer_args_poke_t tracer_args;
1183 };
1184
1185 FIXTURE_SETUP(TRACE_poke)
1186 {
1187         struct sock_filter filter[] = {
1188                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1189                         offsetof(struct seccomp_data, nr)),
1190                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1191                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001),
1192                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1193         };
1194
1195         self->poked = 0;
1196         memset(&self->prog, 0, sizeof(self->prog));
1197         self->prog.filter = malloc(sizeof(filter));
1198         ASSERT_NE(NULL, self->prog.filter);
1199         memcpy(self->prog.filter, filter, sizeof(filter));
1200         self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1201
1202         /* Set up tracer args. */
1203         self->tracer_args.poke_addr = (unsigned long)&self->poked;
1204
1205         /* Launch tracer. */
1206         self->tracer = setup_trace_fixture(_metadata, tracer_poke,
1207                                            &self->tracer_args, false);
1208 }
1209
1210 FIXTURE_TEARDOWN(TRACE_poke)
1211 {
1212         teardown_trace_fixture(_metadata, self->tracer);
1213         if (self->prog.filter)
1214                 free(self->prog.filter);
1215 }
1216
1217 TEST_F(TRACE_poke, read_has_side_effects)
1218 {
1219         ssize_t ret;
1220
1221         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1222         ASSERT_EQ(0, ret);
1223
1224         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1225         ASSERT_EQ(0, ret);
1226
1227         EXPECT_EQ(0, self->poked);
1228         ret = read(-1, NULL, 0);
1229         EXPECT_EQ(-1, ret);
1230         EXPECT_EQ(0x1001, self->poked);
1231 }
1232
1233 TEST_F(TRACE_poke, getpid_runs_normally)
1234 {
1235         long ret;
1236
1237         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1238         ASSERT_EQ(0, ret);
1239
1240         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1241         ASSERT_EQ(0, ret);
1242
1243         EXPECT_EQ(0, self->poked);
1244         EXPECT_NE(0, syscall(__NR_getpid));
1245         EXPECT_EQ(0, self->poked);
1246 }
1247
1248 #if defined(__x86_64__)
1249 # define ARCH_REGS      struct user_regs_struct
1250 # define SYSCALL_NUM    orig_rax
1251 # define SYSCALL_RET    rax
1252 #elif defined(__i386__)
1253 # define ARCH_REGS      struct user_regs_struct
1254 # define SYSCALL_NUM    orig_eax
1255 # define SYSCALL_RET    eax
1256 #elif defined(__arm__)
1257 # define ARCH_REGS      struct pt_regs
1258 # define SYSCALL_NUM    ARM_r7
1259 # define SYSCALL_RET    ARM_r0
1260 #elif defined(__aarch64__)
1261 # define ARCH_REGS      struct user_pt_regs
1262 # define SYSCALL_NUM    regs[8]
1263 # define SYSCALL_RET    regs[0]
1264 #elif defined(__hppa__)
1265 # define ARCH_REGS      struct user_regs_struct
1266 # define SYSCALL_NUM    gr[20]
1267 # define SYSCALL_RET    gr[28]
1268 #elif defined(__powerpc__)
1269 # define ARCH_REGS      struct pt_regs
1270 # define SYSCALL_NUM    gpr[0]
1271 # define SYSCALL_RET    gpr[3]
1272 #elif defined(__s390__)
1273 # define ARCH_REGS     s390_regs
1274 # define SYSCALL_NUM   gprs[2]
1275 # define SYSCALL_RET   gprs[2]
1276 #elif defined(__mips__)
1277 # define ARCH_REGS      struct pt_regs
1278 # define SYSCALL_NUM    regs[2]
1279 # define SYSCALL_SYSCALL_NUM regs[4]
1280 # define SYSCALL_RET    regs[2]
1281 # define SYSCALL_NUM_RET_SHARE_REG
1282 #else
1283 # error "Do not know how to find your architecture's registers and syscalls"
1284 #endif
1285
1286 /* When the syscall return can't be changed, stub out the tests for it. */
1287 #ifdef SYSCALL_NUM_RET_SHARE_REG
1288 # define EXPECT_SYSCALL_RETURN(val, action)     EXPECT_EQ(-1, action)
1289 #else
1290 # define EXPECT_SYSCALL_RETURN(val, action)     EXPECT_EQ(val, action)
1291 #endif
1292
1293 /* Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for
1294  * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux).
1295  */
1296 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__)
1297 #define HAVE_GETREGS
1298 #endif
1299
1300 /* Architecture-specific syscall fetching routine. */
1301 int get_syscall(struct __test_metadata *_metadata, pid_t tracee)
1302 {
1303         ARCH_REGS regs;
1304 #ifdef HAVE_GETREGS
1305         EXPECT_EQ(0, ptrace(PTRACE_GETREGS, tracee, 0, &regs)) {
1306                 TH_LOG("PTRACE_GETREGS failed");
1307                 return -1;
1308         }
1309 #else
1310         struct iovec iov;
1311
1312         iov.iov_base = &regs;
1313         iov.iov_len = sizeof(regs);
1314         EXPECT_EQ(0, ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov)) {
1315                 TH_LOG("PTRACE_GETREGSET failed");
1316                 return -1;
1317         }
1318 #endif
1319
1320 #if defined(__mips__)
1321         if (regs.SYSCALL_NUM == __NR_O32_Linux)
1322                 return regs.SYSCALL_SYSCALL_NUM;
1323 #endif
1324         return regs.SYSCALL_NUM;
1325 }
1326
1327 /* Architecture-specific syscall changing routine. */
1328 void change_syscall(struct __test_metadata *_metadata,
1329                     pid_t tracee, int syscall)
1330 {
1331         int ret;
1332         ARCH_REGS regs;
1333 #ifdef HAVE_GETREGS
1334         ret = ptrace(PTRACE_GETREGS, tracee, 0, &regs);
1335 #else
1336         struct iovec iov;
1337         iov.iov_base = &regs;
1338         iov.iov_len = sizeof(regs);
1339         ret = ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov);
1340 #endif
1341         EXPECT_EQ(0, ret) {}
1342
1343 #if defined(__x86_64__) || defined(__i386__) || defined(__powerpc__) || \
1344     defined(__s390__) || defined(__hppa__)
1345         {
1346                 regs.SYSCALL_NUM = syscall;
1347         }
1348 #elif defined(__mips__)
1349         {
1350                 if (regs.SYSCALL_NUM == __NR_O32_Linux)
1351                         regs.SYSCALL_SYSCALL_NUM = syscall;
1352                 else
1353                         regs.SYSCALL_NUM = syscall;
1354         }
1355
1356 #elif defined(__arm__)
1357 # ifndef PTRACE_SET_SYSCALL
1358 #  define PTRACE_SET_SYSCALL   23
1359 # endif
1360         {
1361                 ret = ptrace(PTRACE_SET_SYSCALL, tracee, NULL, syscall);
1362                 EXPECT_EQ(0, ret);
1363         }
1364
1365 #elif defined(__aarch64__)
1366 # ifndef NT_ARM_SYSTEM_CALL
1367 #  define NT_ARM_SYSTEM_CALL 0x404
1368 # endif
1369         {
1370                 iov.iov_base = &syscall;
1371                 iov.iov_len = sizeof(syscall);
1372                 ret = ptrace(PTRACE_SETREGSET, tracee, NT_ARM_SYSTEM_CALL,
1373                              &iov);
1374                 EXPECT_EQ(0, ret);
1375         }
1376
1377 #else
1378         ASSERT_EQ(1, 0) {
1379                 TH_LOG("How is the syscall changed on this architecture?");
1380         }
1381 #endif
1382
1383         /* If syscall is skipped, change return value. */
1384         if (syscall == -1)
1385 #ifdef SYSCALL_NUM_RET_SHARE_REG
1386                 TH_LOG("Can't modify syscall return on this architecture");
1387 #else
1388                 regs.SYSCALL_RET = EPERM;
1389 #endif
1390
1391 #ifdef HAVE_GETREGS
1392         ret = ptrace(PTRACE_SETREGS, tracee, 0, &regs);
1393 #else
1394         iov.iov_base = &regs;
1395         iov.iov_len = sizeof(regs);
1396         ret = ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &iov);
1397 #endif
1398         EXPECT_EQ(0, ret);
1399 }
1400
1401 void tracer_syscall(struct __test_metadata *_metadata, pid_t tracee,
1402                     int status, void *args)
1403 {
1404         int ret;
1405         unsigned long msg;
1406
1407         /* Make sure we got the right message. */
1408         ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1409         EXPECT_EQ(0, ret);
1410
1411         /* Validate and take action on expected syscalls. */
1412         switch (msg) {
1413         case 0x1002:
1414                 /* change getpid to getppid. */
1415                 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee));
1416                 change_syscall(_metadata, tracee, __NR_getppid);
1417                 break;
1418         case 0x1003:
1419                 /* skip gettid. */
1420                 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee));
1421                 change_syscall(_metadata, tracee, -1);
1422                 break;
1423         case 0x1004:
1424                 /* do nothing (allow getppid) */
1425                 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee));
1426                 break;
1427         default:
1428                 EXPECT_EQ(0, msg) {
1429                         TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg);
1430                         kill(tracee, SIGKILL);
1431                 }
1432         }
1433
1434 }
1435
1436 void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee,
1437                    int status, void *args)
1438 {
1439         int ret, nr;
1440         unsigned long msg;
1441         static bool entry;
1442
1443         /* Make sure we got an empty message. */
1444         ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1445         EXPECT_EQ(0, ret);
1446         EXPECT_EQ(0, msg);
1447
1448         /* The only way to tell PTRACE_SYSCALL entry/exit is by counting. */
1449         entry = !entry;
1450         if (!entry)
1451                 return;
1452
1453         nr = get_syscall(_metadata, tracee);
1454
1455         if (nr == __NR_getpid)
1456                 change_syscall(_metadata, tracee, __NR_getppid);
1457         if (nr == __NR_open)
1458                 change_syscall(_metadata, tracee, -1);
1459 }
1460
1461 FIXTURE_DATA(TRACE_syscall) {
1462         struct sock_fprog prog;
1463         pid_t tracer, mytid, mypid, parent;
1464 };
1465
1466 FIXTURE_SETUP(TRACE_syscall)
1467 {
1468         struct sock_filter filter[] = {
1469                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1470                         offsetof(struct seccomp_data, nr)),
1471                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
1472                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002),
1473                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1),
1474                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003),
1475                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1476                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004),
1477                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1478         };
1479
1480         memset(&self->prog, 0, sizeof(self->prog));
1481         self->prog.filter = malloc(sizeof(filter));
1482         ASSERT_NE(NULL, self->prog.filter);
1483         memcpy(self->prog.filter, filter, sizeof(filter));
1484         self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1485
1486         /* Prepare some testable syscall results. */
1487         self->mytid = syscall(__NR_gettid);
1488         ASSERT_GT(self->mytid, 0);
1489         ASSERT_NE(self->mytid, 1) {
1490                 TH_LOG("Running this test as init is not supported. :)");
1491         }
1492
1493         self->mypid = getpid();
1494         ASSERT_GT(self->mypid, 0);
1495         ASSERT_EQ(self->mytid, self->mypid);
1496
1497         self->parent = getppid();
1498         ASSERT_GT(self->parent, 0);
1499         ASSERT_NE(self->parent, self->mypid);
1500
1501         /* Launch tracer. */
1502         self->tracer = setup_trace_fixture(_metadata, tracer_syscall, NULL,
1503                                            false);
1504 }
1505
1506 FIXTURE_TEARDOWN(TRACE_syscall)
1507 {
1508         teardown_trace_fixture(_metadata, self->tracer);
1509         if (self->prog.filter)
1510                 free(self->prog.filter);
1511 }
1512
1513 TEST_F(TRACE_syscall, ptrace_syscall_redirected)
1514 {
1515         /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1516         teardown_trace_fixture(_metadata, self->tracer);
1517         self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1518                                            true);
1519
1520         /* Tracer will redirect getpid to getppid. */
1521         EXPECT_NE(self->mypid, syscall(__NR_getpid));
1522 }
1523
1524 TEST_F(TRACE_syscall, ptrace_syscall_dropped)
1525 {
1526         /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1527         teardown_trace_fixture(_metadata, self->tracer);
1528         self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1529                                            true);
1530
1531         /* Tracer should skip the open syscall, resulting in EPERM. */
1532         EXPECT_SYSCALL_RETURN(EPERM, syscall(__NR_open));
1533 }
1534
1535 TEST_F(TRACE_syscall, syscall_allowed)
1536 {
1537         long ret;
1538
1539         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1540         ASSERT_EQ(0, ret);
1541
1542         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1543         ASSERT_EQ(0, ret);
1544
1545         /* getppid works as expected (no changes). */
1546         EXPECT_EQ(self->parent, syscall(__NR_getppid));
1547         EXPECT_NE(self->mypid, syscall(__NR_getppid));
1548 }
1549
1550 TEST_F(TRACE_syscall, syscall_redirected)
1551 {
1552         long ret;
1553
1554         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1555         ASSERT_EQ(0, ret);
1556
1557         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1558         ASSERT_EQ(0, ret);
1559
1560         /* getpid has been redirected to getppid as expected. */
1561         EXPECT_EQ(self->parent, syscall(__NR_getpid));
1562         EXPECT_NE(self->mypid, syscall(__NR_getpid));
1563 }
1564
1565 TEST_F(TRACE_syscall, syscall_dropped)
1566 {
1567         long ret;
1568
1569         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1570         ASSERT_EQ(0, ret);
1571
1572         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1573         ASSERT_EQ(0, ret);
1574
1575         /* gettid has been skipped and an altered return value stored. */
1576         EXPECT_SYSCALL_RETURN(EPERM, syscall(__NR_gettid));
1577         EXPECT_NE(self->mytid, syscall(__NR_gettid));
1578 }
1579
1580 TEST_F(TRACE_syscall, skip_after_RET_TRACE)
1581 {
1582         struct sock_filter filter[] = {
1583                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1584                         offsetof(struct seccomp_data, nr)),
1585                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1586                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
1587                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1588         };
1589         struct sock_fprog prog = {
1590                 .len = (unsigned short)ARRAY_SIZE(filter),
1591                 .filter = filter,
1592         };
1593         long ret;
1594
1595         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1596         ASSERT_EQ(0, ret);
1597
1598         /* Install fixture filter. */
1599         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1600         ASSERT_EQ(0, ret);
1601
1602         /* Install "errno on getppid" filter. */
1603         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1604         ASSERT_EQ(0, ret);
1605
1606         /* Tracer will redirect getpid to getppid, and we should see EPERM. */
1607         errno = 0;
1608         EXPECT_EQ(-1, syscall(__NR_getpid));
1609         EXPECT_EQ(EPERM, errno);
1610 }
1611
1612 TEST_F_SIGNAL(TRACE_syscall, kill_after_RET_TRACE, SIGSYS)
1613 {
1614         struct sock_filter filter[] = {
1615                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1616                         offsetof(struct seccomp_data, nr)),
1617                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1618                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1619                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1620         };
1621         struct sock_fprog prog = {
1622                 .len = (unsigned short)ARRAY_SIZE(filter),
1623                 .filter = filter,
1624         };
1625         long ret;
1626
1627         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1628         ASSERT_EQ(0, ret);
1629
1630         /* Install fixture filter. */
1631         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1632         ASSERT_EQ(0, ret);
1633
1634         /* Install "death on getppid" filter. */
1635         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1636         ASSERT_EQ(0, ret);
1637
1638         /* Tracer will redirect getpid to getppid, and we should die. */
1639         EXPECT_NE(self->mypid, syscall(__NR_getpid));
1640 }
1641
1642 TEST_F(TRACE_syscall, skip_after_ptrace)
1643 {
1644         struct sock_filter filter[] = {
1645                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1646                         offsetof(struct seccomp_data, nr)),
1647                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1648                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
1649                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1650         };
1651         struct sock_fprog prog = {
1652                 .len = (unsigned short)ARRAY_SIZE(filter),
1653                 .filter = filter,
1654         };
1655         long ret;
1656
1657         /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1658         teardown_trace_fixture(_metadata, self->tracer);
1659         self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1660                                            true);
1661
1662         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1663         ASSERT_EQ(0, ret);
1664
1665         /* Install "errno on getppid" filter. */
1666         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1667         ASSERT_EQ(0, ret);
1668
1669         /* Tracer will redirect getpid to getppid, and we should see EPERM. */
1670         EXPECT_EQ(-1, syscall(__NR_getpid));
1671         EXPECT_EQ(EPERM, errno);
1672 }
1673
1674 TEST_F_SIGNAL(TRACE_syscall, kill_after_ptrace, SIGSYS)
1675 {
1676         struct sock_filter filter[] = {
1677                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1678                         offsetof(struct seccomp_data, nr)),
1679                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1680                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1681                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1682         };
1683         struct sock_fprog prog = {
1684                 .len = (unsigned short)ARRAY_SIZE(filter),
1685                 .filter = filter,
1686         };
1687         long ret;
1688
1689         /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1690         teardown_trace_fixture(_metadata, self->tracer);
1691         self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1692                                            true);
1693
1694         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1695         ASSERT_EQ(0, ret);
1696
1697         /* Install "death on getppid" filter. */
1698         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1699         ASSERT_EQ(0, ret);
1700
1701         /* Tracer will redirect getpid to getppid, and we should die. */
1702         EXPECT_NE(self->mypid, syscall(__NR_getpid));
1703 }
1704
1705 #ifndef __NR_seccomp
1706 # if defined(__i386__)
1707 #  define __NR_seccomp 354
1708 # elif defined(__x86_64__)
1709 #  define __NR_seccomp 317
1710 # elif defined(__arm__)
1711 #  define __NR_seccomp 383
1712 # elif defined(__aarch64__)
1713 #  define __NR_seccomp 277
1714 # elif defined(__hppa__)
1715 #  define __NR_seccomp 338
1716 # elif defined(__powerpc__)
1717 #  define __NR_seccomp 358
1718 # elif defined(__s390__)
1719 #  define __NR_seccomp 348
1720 # else
1721 #  warning "seccomp syscall number unknown for this architecture"
1722 #  define __NR_seccomp 0xffff
1723 # endif
1724 #endif
1725
1726 #ifndef SECCOMP_SET_MODE_STRICT
1727 #define SECCOMP_SET_MODE_STRICT 0
1728 #endif
1729
1730 #ifndef SECCOMP_SET_MODE_FILTER
1731 #define SECCOMP_SET_MODE_FILTER 1
1732 #endif
1733
1734 #ifndef SECCOMP_GET_ACTION_AVAIL
1735 #define SECCOMP_GET_ACTION_AVAIL 2
1736 #endif
1737
1738 #ifndef SECCOMP_FILTER_FLAG_TSYNC
1739 #define SECCOMP_FILTER_FLAG_TSYNC 1
1740 #endif
1741
1742 #ifndef seccomp
1743 int seccomp(unsigned int op, unsigned int flags, void *args)
1744 {
1745         errno = 0;
1746         return syscall(__NR_seccomp, op, flags, args);
1747 }
1748 #endif
1749
1750 TEST(seccomp_syscall)
1751 {
1752         struct sock_filter filter[] = {
1753                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1754         };
1755         struct sock_fprog prog = {
1756                 .len = (unsigned short)ARRAY_SIZE(filter),
1757                 .filter = filter,
1758         };
1759         long ret;
1760
1761         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1762         ASSERT_EQ(0, ret) {
1763                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1764         }
1765
1766         /* Reject insane operation. */
1767         ret = seccomp(-1, 0, &prog);
1768         ASSERT_NE(ENOSYS, errno) {
1769                 TH_LOG("Kernel does not support seccomp syscall!");
1770         }
1771         EXPECT_EQ(EINVAL, errno) {
1772                 TH_LOG("Did not reject crazy op value!");
1773         }
1774
1775         /* Reject strict with flags or pointer. */
1776         ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL);
1777         EXPECT_EQ(EINVAL, errno) {
1778                 TH_LOG("Did not reject mode strict with flags!");
1779         }
1780         ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog);
1781         EXPECT_EQ(EINVAL, errno) {
1782                 TH_LOG("Did not reject mode strict with uargs!");
1783         }
1784
1785         /* Reject insane args for filter. */
1786         ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog);
1787         EXPECT_EQ(EINVAL, errno) {
1788                 TH_LOG("Did not reject crazy filter flags!");
1789         }
1790         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL);
1791         EXPECT_EQ(EFAULT, errno) {
1792                 TH_LOG("Did not reject NULL filter!");
1793         }
1794
1795         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
1796         EXPECT_EQ(0, errno) {
1797                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s",
1798                         strerror(errno));
1799         }
1800 }
1801
1802 TEST(seccomp_syscall_mode_lock)
1803 {
1804         struct sock_filter filter[] = {
1805                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1806         };
1807         struct sock_fprog prog = {
1808                 .len = (unsigned short)ARRAY_SIZE(filter),
1809                 .filter = filter,
1810         };
1811         long ret;
1812
1813         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
1814         ASSERT_EQ(0, ret) {
1815                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1816         }
1817
1818         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
1819         ASSERT_NE(ENOSYS, errno) {
1820                 TH_LOG("Kernel does not support seccomp syscall!");
1821         }
1822         EXPECT_EQ(0, ret) {
1823                 TH_LOG("Could not install filter!");
1824         }
1825
1826         /* Make sure neither entry point will switch to strict. */
1827         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0);
1828         EXPECT_EQ(EINVAL, errno) {
1829                 TH_LOG("Switched to mode strict!");
1830         }
1831
1832         ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL);
1833         EXPECT_EQ(EINVAL, errno) {
1834                 TH_LOG("Switched to mode strict!");
1835         }
1836 }
1837
1838 TEST(TSYNC_first)
1839 {
1840         struct sock_filter filter[] = {
1841                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1842         };
1843         struct sock_fprog prog = {
1844                 .len = (unsigned short)ARRAY_SIZE(filter),
1845                 .filter = filter,
1846         };
1847         long ret;
1848
1849         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
1850         ASSERT_EQ(0, ret) {
1851                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1852         }
1853
1854         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
1855                       &prog);
1856         ASSERT_NE(ENOSYS, errno) {
1857                 TH_LOG("Kernel does not support seccomp syscall!");
1858         }
1859         EXPECT_EQ(0, ret) {
1860                 TH_LOG("Could not install initial filter with TSYNC!");
1861         }
1862 }
1863
1864 #define TSYNC_SIBLINGS 2
1865 struct tsync_sibling {
1866         pthread_t tid;
1867         pid_t system_tid;
1868         sem_t *started;
1869         pthread_cond_t *cond;
1870         pthread_mutex_t *mutex;
1871         int diverge;
1872         int num_waits;
1873         struct sock_fprog *prog;
1874         struct __test_metadata *metadata;
1875 };
1876
1877 /*
1878  * To avoid joining joined threads (which is not allowed by Bionic),
1879  * make sure we both successfully join and clear the tid to skip a
1880  * later join attempt during fixture teardown. Any remaining threads
1881  * will be directly killed during teardown.
1882  */
1883 #define PTHREAD_JOIN(tid, status)                                       \
1884         do {                                                            \
1885                 int _rc = pthread_join(tid, status);                    \
1886                 if (_rc) {                                              \
1887                         TH_LOG("pthread_join of tid %u failed: %d\n",   \
1888                                 (unsigned int)tid, _rc);                \
1889                 } else {                                                \
1890                         tid = 0;                                        \
1891                 }                                                       \
1892         } while (0)
1893
1894 FIXTURE_DATA(TSYNC) {
1895         struct sock_fprog root_prog, apply_prog;
1896         struct tsync_sibling sibling[TSYNC_SIBLINGS];
1897         sem_t started;
1898         pthread_cond_t cond;
1899         pthread_mutex_t mutex;
1900         int sibling_count;
1901 };
1902
1903 FIXTURE_SETUP(TSYNC)
1904 {
1905         struct sock_filter root_filter[] = {
1906                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1907         };
1908         struct sock_filter apply_filter[] = {
1909                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1910                         offsetof(struct seccomp_data, nr)),
1911                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1912                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1913                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1914         };
1915
1916         memset(&self->root_prog, 0, sizeof(self->root_prog));
1917         memset(&self->apply_prog, 0, sizeof(self->apply_prog));
1918         memset(&self->sibling, 0, sizeof(self->sibling));
1919         self->root_prog.filter = malloc(sizeof(root_filter));
1920         ASSERT_NE(NULL, self->root_prog.filter);
1921         memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter));
1922         self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter);
1923
1924         self->apply_prog.filter = malloc(sizeof(apply_filter));
1925         ASSERT_NE(NULL, self->apply_prog.filter);
1926         memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter));
1927         self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter);
1928
1929         self->sibling_count = 0;
1930         pthread_mutex_init(&self->mutex, NULL);
1931         pthread_cond_init(&self->cond, NULL);
1932         sem_init(&self->started, 0, 0);
1933         self->sibling[0].tid = 0;
1934         self->sibling[0].cond = &self->cond;
1935         self->sibling[0].started = &self->started;
1936         self->sibling[0].mutex = &self->mutex;
1937         self->sibling[0].diverge = 0;
1938         self->sibling[0].num_waits = 1;
1939         self->sibling[0].prog = &self->root_prog;
1940         self->sibling[0].metadata = _metadata;
1941         self->sibling[1].tid = 0;
1942         self->sibling[1].cond = &self->cond;
1943         self->sibling[1].started = &self->started;
1944         self->sibling[1].mutex = &self->mutex;
1945         self->sibling[1].diverge = 0;
1946         self->sibling[1].prog = &self->root_prog;
1947         self->sibling[1].num_waits = 1;
1948         self->sibling[1].metadata = _metadata;
1949 }
1950
1951 FIXTURE_TEARDOWN(TSYNC)
1952 {
1953         int sib = 0;
1954
1955         if (self->root_prog.filter)
1956                 free(self->root_prog.filter);
1957         if (self->apply_prog.filter)
1958                 free(self->apply_prog.filter);
1959
1960         for ( ; sib < self->sibling_count; ++sib) {
1961                 struct tsync_sibling *s = &self->sibling[sib];
1962
1963                 if (!s->tid)
1964                         continue;
1965                 /*
1966                  * If a thread is still running, it may be stuck, so hit
1967                  * it over the head really hard.
1968                  */
1969                 pthread_kill(s->tid, 9);
1970         }
1971         pthread_mutex_destroy(&self->mutex);
1972         pthread_cond_destroy(&self->cond);
1973         sem_destroy(&self->started);
1974 }
1975
1976 void *tsync_sibling(void *data)
1977 {
1978         long ret = 0;
1979         struct tsync_sibling *me = data;
1980
1981         me->system_tid = syscall(__NR_gettid);
1982
1983         pthread_mutex_lock(me->mutex);
1984         if (me->diverge) {
1985                 /* Just re-apply the root prog to fork the tree */
1986                 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
1987                                 me->prog, 0, 0);
1988         }
1989         sem_post(me->started);
1990         /* Return outside of started so parent notices failures. */
1991         if (ret) {
1992                 pthread_mutex_unlock(me->mutex);
1993                 return (void *)SIBLING_EXIT_FAILURE;
1994         }
1995         do {
1996                 pthread_cond_wait(me->cond, me->mutex);
1997                 me->num_waits = me->num_waits - 1;
1998         } while (me->num_waits);
1999         pthread_mutex_unlock(me->mutex);
2000
2001         ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0);
2002         if (!ret)
2003                 return (void *)SIBLING_EXIT_NEWPRIVS;
2004         read(0, NULL, 0);
2005         return (void *)SIBLING_EXIT_UNKILLED;
2006 }
2007
2008 void tsync_start_sibling(struct tsync_sibling *sibling)
2009 {
2010         pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling);
2011 }
2012
2013 TEST_F(TSYNC, siblings_fail_prctl)
2014 {
2015         long ret;
2016         void *status;
2017         struct sock_filter filter[] = {
2018                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2019                         offsetof(struct seccomp_data, nr)),
2020                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
2021                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL),
2022                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2023         };
2024         struct sock_fprog prog = {
2025                 .len = (unsigned short)ARRAY_SIZE(filter),
2026                 .filter = filter,
2027         };
2028
2029         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2030                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2031         }
2032
2033         /* Check prctl failure detection by requesting sib 0 diverge. */
2034         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2035         ASSERT_NE(ENOSYS, errno) {
2036                 TH_LOG("Kernel does not support seccomp syscall!");
2037         }
2038         ASSERT_EQ(0, ret) {
2039                 TH_LOG("setting filter failed");
2040         }
2041
2042         self->sibling[0].diverge = 1;
2043         tsync_start_sibling(&self->sibling[0]);
2044         tsync_start_sibling(&self->sibling[1]);
2045
2046         while (self->sibling_count < TSYNC_SIBLINGS) {
2047                 sem_wait(&self->started);
2048                 self->sibling_count++;
2049         }
2050
2051         /* Signal the threads to clean up*/
2052         pthread_mutex_lock(&self->mutex);
2053         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2054                 TH_LOG("cond broadcast non-zero");
2055         }
2056         pthread_mutex_unlock(&self->mutex);
2057
2058         /* Ensure diverging sibling failed to call prctl. */
2059         PTHREAD_JOIN(self->sibling[0].tid, &status);
2060         EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status);
2061         PTHREAD_JOIN(self->sibling[1].tid, &status);
2062         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2063 }
2064
2065 TEST_F(TSYNC, two_siblings_with_ancestor)
2066 {
2067         long ret;
2068         void *status;
2069
2070         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2071                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2072         }
2073
2074         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2075         ASSERT_NE(ENOSYS, errno) {
2076                 TH_LOG("Kernel does not support seccomp syscall!");
2077         }
2078         ASSERT_EQ(0, ret) {
2079                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2080         }
2081         tsync_start_sibling(&self->sibling[0]);
2082         tsync_start_sibling(&self->sibling[1]);
2083
2084         while (self->sibling_count < TSYNC_SIBLINGS) {
2085                 sem_wait(&self->started);
2086                 self->sibling_count++;
2087         }
2088
2089         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2090                       &self->apply_prog);
2091         ASSERT_EQ(0, ret) {
2092                 TH_LOG("Could install filter on all threads!");
2093         }
2094         /* Tell the siblings to test the policy */
2095         pthread_mutex_lock(&self->mutex);
2096         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2097                 TH_LOG("cond broadcast non-zero");
2098         }
2099         pthread_mutex_unlock(&self->mutex);
2100         /* Ensure they are both killed and don't exit cleanly. */
2101         PTHREAD_JOIN(self->sibling[0].tid, &status);
2102         EXPECT_EQ(0x0, (long)status);
2103         PTHREAD_JOIN(self->sibling[1].tid, &status);
2104         EXPECT_EQ(0x0, (long)status);
2105 }
2106
2107 TEST_F(TSYNC, two_sibling_want_nnp)
2108 {
2109         void *status;
2110
2111         /* start siblings before any prctl() operations */
2112         tsync_start_sibling(&self->sibling[0]);
2113         tsync_start_sibling(&self->sibling[1]);
2114         while (self->sibling_count < TSYNC_SIBLINGS) {
2115                 sem_wait(&self->started);
2116                 self->sibling_count++;
2117         }
2118
2119         /* Tell the siblings to test no policy */
2120         pthread_mutex_lock(&self->mutex);
2121         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2122                 TH_LOG("cond broadcast non-zero");
2123         }
2124         pthread_mutex_unlock(&self->mutex);
2125
2126         /* Ensure they are both upset about lacking nnp. */
2127         PTHREAD_JOIN(self->sibling[0].tid, &status);
2128         EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2129         PTHREAD_JOIN(self->sibling[1].tid, &status);
2130         EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2131 }
2132
2133 TEST_F(TSYNC, two_siblings_with_no_filter)
2134 {
2135         long ret;
2136         void *status;
2137
2138         /* start siblings before any prctl() operations */
2139         tsync_start_sibling(&self->sibling[0]);
2140         tsync_start_sibling(&self->sibling[1]);
2141         while (self->sibling_count < TSYNC_SIBLINGS) {
2142                 sem_wait(&self->started);
2143                 self->sibling_count++;
2144         }
2145
2146         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2147                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2148         }
2149
2150         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2151                       &self->apply_prog);
2152         ASSERT_NE(ENOSYS, errno) {
2153                 TH_LOG("Kernel does not support seccomp syscall!");
2154         }
2155         ASSERT_EQ(0, ret) {
2156                 TH_LOG("Could install filter on all threads!");
2157         }
2158
2159         /* Tell the siblings to test the policy */
2160         pthread_mutex_lock(&self->mutex);
2161         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2162                 TH_LOG("cond broadcast non-zero");
2163         }
2164         pthread_mutex_unlock(&self->mutex);
2165
2166         /* Ensure they are both killed and don't exit cleanly. */
2167         PTHREAD_JOIN(self->sibling[0].tid, &status);
2168         EXPECT_EQ(0x0, (long)status);
2169         PTHREAD_JOIN(self->sibling[1].tid, &status);
2170         EXPECT_EQ(0x0, (long)status);
2171 }
2172
2173 TEST_F(TSYNC, two_siblings_with_one_divergence)
2174 {
2175         long ret;
2176         void *status;
2177
2178         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2179                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2180         }
2181
2182         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2183         ASSERT_NE(ENOSYS, errno) {
2184                 TH_LOG("Kernel does not support seccomp syscall!");
2185         }
2186         ASSERT_EQ(0, ret) {
2187                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2188         }
2189         self->sibling[0].diverge = 1;
2190         tsync_start_sibling(&self->sibling[0]);
2191         tsync_start_sibling(&self->sibling[1]);
2192
2193         while (self->sibling_count < TSYNC_SIBLINGS) {
2194                 sem_wait(&self->started);
2195                 self->sibling_count++;
2196         }
2197
2198         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2199                       &self->apply_prog);
2200         ASSERT_EQ(self->sibling[0].system_tid, ret) {
2201                 TH_LOG("Did not fail on diverged sibling.");
2202         }
2203
2204         /* Wake the threads */
2205         pthread_mutex_lock(&self->mutex);
2206         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2207                 TH_LOG("cond broadcast non-zero");
2208         }
2209         pthread_mutex_unlock(&self->mutex);
2210
2211         /* Ensure they are both unkilled. */
2212         PTHREAD_JOIN(self->sibling[0].tid, &status);
2213         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2214         PTHREAD_JOIN(self->sibling[1].tid, &status);
2215         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2216 }
2217
2218 TEST_F(TSYNC, two_siblings_not_under_filter)
2219 {
2220         long ret, sib;
2221         void *status;
2222
2223         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2224                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2225         }
2226
2227         /*
2228          * Sibling 0 will have its own seccomp policy
2229          * and Sibling 1 will not be under seccomp at
2230          * all. Sibling 1 will enter seccomp and 0
2231          * will cause failure.
2232          */
2233         self->sibling[0].diverge = 1;
2234         tsync_start_sibling(&self->sibling[0]);
2235         tsync_start_sibling(&self->sibling[1]);
2236
2237         while (self->sibling_count < TSYNC_SIBLINGS) {
2238                 sem_wait(&self->started);
2239                 self->sibling_count++;
2240         }
2241
2242         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2243         ASSERT_NE(ENOSYS, errno) {
2244                 TH_LOG("Kernel does not support seccomp syscall!");
2245         }
2246         ASSERT_EQ(0, ret) {
2247                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2248         }
2249
2250         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2251                       &self->apply_prog);
2252         ASSERT_EQ(ret, self->sibling[0].system_tid) {
2253                 TH_LOG("Did not fail on diverged sibling.");
2254         }
2255         sib = 1;
2256         if (ret == self->sibling[0].system_tid)
2257                 sib = 0;
2258
2259         pthread_mutex_lock(&self->mutex);
2260
2261         /* Increment the other siblings num_waits so we can clean up
2262          * the one we just saw.
2263          */
2264         self->sibling[!sib].num_waits += 1;
2265
2266         /* Signal the thread to clean up*/
2267         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2268                 TH_LOG("cond broadcast non-zero");
2269         }
2270         pthread_mutex_unlock(&self->mutex);
2271         PTHREAD_JOIN(self->sibling[sib].tid, &status);
2272         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2273         /* Poll for actual task death. pthread_join doesn't guarantee it. */
2274         while (!kill(self->sibling[sib].system_tid, 0))
2275                 sleep(0.1);
2276         /* Switch to the remaining sibling */
2277         sib = !sib;
2278
2279         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2280                       &self->apply_prog);
2281         ASSERT_EQ(0, ret) {
2282                 TH_LOG("Expected the remaining sibling to sync");
2283         };
2284
2285         pthread_mutex_lock(&self->mutex);
2286
2287         /* If remaining sibling didn't have a chance to wake up during
2288          * the first broadcast, manually reduce the num_waits now.
2289          */
2290         if (self->sibling[sib].num_waits > 1)
2291                 self->sibling[sib].num_waits = 1;
2292         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2293                 TH_LOG("cond broadcast non-zero");
2294         }
2295         pthread_mutex_unlock(&self->mutex);
2296         PTHREAD_JOIN(self->sibling[sib].tid, &status);
2297         EXPECT_EQ(0, (long)status);
2298         /* Poll for actual task death. pthread_join doesn't guarantee it. */
2299         while (!kill(self->sibling[sib].system_tid, 0))
2300                 sleep(0.1);
2301
2302         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2303                       &self->apply_prog);
2304         ASSERT_EQ(0, ret);  /* just us chickens */
2305 }
2306
2307 /* Make sure restarted syscalls are seen directly as "restart_syscall". */
2308 TEST(syscall_restart)
2309 {
2310         long ret;
2311         unsigned long msg;
2312         pid_t child_pid;
2313         int pipefd[2];
2314         int status;
2315         siginfo_t info = { };
2316         struct sock_filter filter[] = {
2317                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2318                          offsetof(struct seccomp_data, nr)),
2319
2320 #ifdef __NR_sigreturn
2321                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 6, 0),
2322 #endif
2323                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 5, 0),
2324                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 4, 0),
2325                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 3, 0),
2326                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 4, 0),
2327                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0),
2328
2329                 /* Allow __NR_write for easy logging. */
2330                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1),
2331                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2332                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2333                 /* The nanosleep jump target. */
2334                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100),
2335                 /* The restart_syscall jump target. */
2336                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200),
2337         };
2338         struct sock_fprog prog = {
2339                 .len = (unsigned short)ARRAY_SIZE(filter),
2340                 .filter = filter,
2341         };
2342 #if defined(__arm__)
2343         struct utsname utsbuf;
2344 #endif
2345
2346         ASSERT_EQ(0, pipe(pipefd));
2347
2348         child_pid = fork();
2349         ASSERT_LE(0, child_pid);
2350         if (child_pid == 0) {
2351                 /* Child uses EXPECT not ASSERT to deliver status correctly. */
2352                 char buf = ' ';
2353                 struct timespec timeout = { };
2354
2355                 /* Attach parent as tracer and stop. */
2356                 EXPECT_EQ(0, ptrace(PTRACE_TRACEME));
2357                 EXPECT_EQ(0, raise(SIGSTOP));
2358
2359                 EXPECT_EQ(0, close(pipefd[1]));
2360
2361                 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2362                         TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2363                 }
2364
2365                 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2366                 EXPECT_EQ(0, ret) {
2367                         TH_LOG("Failed to install filter!");
2368                 }
2369
2370                 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2371                         TH_LOG("Failed to read() sync from parent");
2372                 }
2373                 EXPECT_EQ('.', buf) {
2374                         TH_LOG("Failed to get sync data from read()");
2375                 }
2376
2377                 /* Start nanosleep to be interrupted. */
2378                 timeout.tv_sec = 1;
2379                 errno = 0;
2380                 EXPECT_EQ(0, nanosleep(&timeout, NULL)) {
2381                         TH_LOG("Call to nanosleep() failed (errno %d)", errno);
2382                 }
2383
2384                 /* Read final sync from parent. */
2385                 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2386                         TH_LOG("Failed final read() from parent");
2387                 }
2388                 EXPECT_EQ('!', buf) {
2389                         TH_LOG("Failed to get final data from read()");
2390                 }
2391
2392                 /* Directly report the status of our test harness results. */
2393                 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS
2394                                                      : EXIT_FAILURE);
2395         }
2396         EXPECT_EQ(0, close(pipefd[0]));
2397
2398         /* Attach to child, setup options, and release. */
2399         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2400         ASSERT_EQ(true, WIFSTOPPED(status));
2401         ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL,
2402                             PTRACE_O_TRACESECCOMP));
2403         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2404         ASSERT_EQ(1, write(pipefd[1], ".", 1));
2405
2406         /* Wait for nanosleep() to start. */
2407         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2408         ASSERT_EQ(true, WIFSTOPPED(status));
2409         ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2410         ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2411         ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2412         ASSERT_EQ(0x100, msg);
2413         EXPECT_EQ(__NR_nanosleep, get_syscall(_metadata, child_pid));
2414
2415         /* Might as well check siginfo for sanity while we're here. */
2416         ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2417         ASSERT_EQ(SIGTRAP, info.si_signo);
2418         ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code);
2419         EXPECT_EQ(0, info.si_errno);
2420         EXPECT_EQ(getuid(), info.si_uid);
2421         /* Verify signal delivery came from child (seccomp-triggered). */
2422         EXPECT_EQ(child_pid, info.si_pid);
2423
2424         /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */
2425         ASSERT_EQ(0, kill(child_pid, SIGSTOP));
2426         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2427         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2428         ASSERT_EQ(true, WIFSTOPPED(status));
2429         ASSERT_EQ(SIGSTOP, WSTOPSIG(status));
2430         /* Verify signal delivery came from parent now. */
2431         ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2432         EXPECT_EQ(getpid(), info.si_pid);
2433
2434         /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */
2435         ASSERT_EQ(0, kill(child_pid, SIGCONT));
2436         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2437         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2438         ASSERT_EQ(true, WIFSTOPPED(status));
2439         ASSERT_EQ(SIGCONT, WSTOPSIG(status));
2440         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2441
2442         /* Wait for restart_syscall() to start. */
2443         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2444         ASSERT_EQ(true, WIFSTOPPED(status));
2445         ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2446         ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2447         ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2448
2449         ASSERT_EQ(0x200, msg);
2450         ret = get_syscall(_metadata, child_pid);
2451 #if defined(__arm__)
2452         /*
2453          * FIXME:
2454          * - native ARM registers do NOT expose true syscall.
2455          * - compat ARM registers on ARM64 DO expose true syscall.
2456          */
2457         ASSERT_EQ(0, uname(&utsbuf));
2458         if (strncmp(utsbuf.machine, "arm", 3) == 0) {
2459                 EXPECT_EQ(__NR_nanosleep, ret);
2460         } else
2461 #endif
2462         {
2463                 EXPECT_EQ(__NR_restart_syscall, ret);
2464         }
2465
2466         /* Write again to end test. */
2467         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2468         ASSERT_EQ(1, write(pipefd[1], "!", 1));
2469         EXPECT_EQ(0, close(pipefd[1]));
2470
2471         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2472         if (WIFSIGNALED(status) || WEXITSTATUS(status))
2473                 _metadata->passed = 0;
2474 }
2475
2476 TEST(get_action_avail)
2477 {
2478         __u32 actions[] = { SECCOMP_RET_KILL,  SECCOMP_RET_TRAP,
2479                             SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE,
2480                             SECCOMP_RET_ALLOW };
2481         __u32 unknown_action = 0x10000000U;
2482         int i;
2483         long ret;
2484
2485         ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[0]);
2486         ASSERT_NE(ENOSYS, errno) {
2487                 TH_LOG("Kernel does not support seccomp syscall!");
2488         }
2489         ASSERT_NE(EINVAL, errno) {
2490                 TH_LOG("Kernel does not support SECCOMP_GET_ACTION_AVAIL operation!");
2491         }
2492         EXPECT_EQ(ret, 0);
2493
2494         for (i = 0; i < ARRAY_SIZE(actions); i++) {
2495                 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[i]);
2496                 EXPECT_EQ(ret, 0) {
2497                         TH_LOG("Expected action (0x%X) not available!",
2498                                actions[i]);
2499                 }
2500         }
2501
2502         /* Check that an unknown action is handled properly (EOPNOTSUPP) */
2503         ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &unknown_action);
2504         EXPECT_EQ(ret, -1);
2505         EXPECT_EQ(errno, EOPNOTSUPP);
2506 }
2507
2508 /*
2509  * TODO:
2510  * - add microbenchmarks
2511  * - expand NNP testing
2512  * - better arch-specific TRACE and TRAP handlers.
2513  * - endianness checking when appropriate
2514  * - 64-bit arg prodding
2515  * - arch value testing (x86 modes especially)
2516  * - ...
2517  */
2518
2519 TEST_HARNESS_MAIN