tools/testing/selftests/seccomp/seccomp_bpf.c

   1 /*
   2  * Copyright (c) 2012 The Chromium OS Authors. All rights reserved.
   3  * Use of this source code is governed by the GPLv2 license.
   4  *
   5  * Test code for seccomp bpf.
   6  */
   7
   8 #include <sys/types.h>
   9 #include <asm/siginfo.h>
  10 #define __have_siginfo_t 1
  11 #define __have_sigval_t 1
  12 #define __have_sigevent_t 1
  13
  14 #include <errno.h>
  15 #include <linux/filter.h>
  16 #include <sys/prctl.h>
  17 #include <sys/ptrace.h>
  18 #include <sys/user.h>
  19 #include <linux/prctl.h>
  20 #include <linux/ptrace.h>
  21 #include <linux/seccomp.h>
  22 #include <pthread.h>
  23 #include <semaphore.h>
  24 #include <signal.h>
  25 #include <stddef.h>
  26 #include <stdbool.h>
  27 #include <string.h>
  28 #include <time.h>
  29 #include <linux/elf.h>
  30 #include <sys/uio.h>
  31 #include <sys/utsname.h>
  32 #include <sys/fcntl.h>
  33 #include <sys/mman.h>
  34 #include <sys/times.h>
  35
  36 #define _GNU_SOURCE
  37 #include <unistd.h>
  38 #include <sys/syscall.h>
  39
  40 #include "test_harness.h"
  41
  42 #ifndef PR_SET_PTRACER
  43 # define PR_SET_PTRACER 0x59616d61
  44 #endif
  45
  46 #ifndef PR_SET_NO_NEW_PRIVS
  47 #define PR_SET_NO_NEW_PRIVS 38
  48 #define PR_GET_NO_NEW_PRIVS 39
  49 #endif
  50
  51 #ifndef PR_SECCOMP_EXT
  52 #define PR_SECCOMP_EXT 43
  53 #endif
  54
  55 #ifndef SECCOMP_EXT_ACT
  56 #define SECCOMP_EXT_ACT 1
  57 #endif
  58
  59 #ifndef SECCOMP_EXT_ACT_TSYNC
  60 #define SECCOMP_EXT_ACT_TSYNC 1
  61 #endif
  62
  63 #ifndef SECCOMP_MODE_STRICT
  64 #define SECCOMP_MODE_STRICT 1
  65 #endif
  66
  67 #ifndef SECCOMP_MODE_FILTER
  68 #define SECCOMP_MODE_FILTER 2
  69 #endif
  70
  71 #ifndef SECCOMP_RET_KILL
  72 #define SECCOMP_RET_KILL        0x00000000U /* kill the task immediately */
  73 #define SECCOMP_RET_TRAP        0x00030000U /* disallow and force a SIGSYS */
  74 #define SECCOMP_RET_ERRNO       0x00050000U /* returns an errno */
  75 #define SECCOMP_RET_TRACE       0x7ff00000U /* pass to a tracer or disallow */
  76 #define SECCOMP_RET_ALLOW       0x7fff0000U /* allow */
  77
  78 /* Masks for the return value sections. */
  79 #define SECCOMP_RET_ACTION      0x7fff0000U
  80 #define SECCOMP_RET_DATA        0x0000ffffU
  81
  82 struct seccomp_data {
  83         int nr;
  84         __u32 arch;
  85         __u64 instruction_pointer;
  86         __u64 args[6];
  87 };
  88 #endif
  89
  90 #if __BYTE_ORDER == __LITTLE_ENDIAN
  91 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]))
  92 #elif __BYTE_ORDER == __BIG_ENDIAN
  93 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32))
  94 #else
  95 #error "wut? Unknown __BYTE_ORDER?!"
  96 #endif
  97
  98 #define SIBLING_EXIT_UNKILLED   0xbadbeef
  99 #define SIBLING_EXIT_FAILURE    0xbadface
 100 #define SIBLING_EXIT_NEWPRIVS   0xbadfeed
 101
 102 TEST(mode_strict_support)
 103 {
 104         long ret;
 105
 106         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
 107         ASSERT_EQ(0, ret) {
 108                 TH_LOG("Kernel does not support CONFIG_SECCOMP");
 109         }
 110         syscall(__NR_exit, 1);
 111 }
 112
 113 TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL)
 114 {
 115         long ret;
 116
 117         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
 118         ASSERT_EQ(0, ret) {
 119                 TH_LOG("Kernel does not support CONFIG_SECCOMP");
 120         }
 121         syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
 122                 NULL, NULL, NULL);
 123         EXPECT_FALSE(true) {
 124                 TH_LOG("Unreachable!");
 125         }
 126 }
 127
 128 /* Note! This doesn't test no new privs behavior */
 129 TEST(no_new_privs_support)
 130 {
 131         long ret;
 132
 133         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 134         EXPECT_EQ(0, ret) {
 135                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
 136         }
 137 }
 138
 139 /* Tests kernel support by checking for a copy_from_user() fault on * NULL. */
 140 TEST(mode_filter_support)
 141 {
 142         long ret;
 143
 144         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
 145         ASSERT_EQ(0, ret) {
 146                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
 147         }
 148         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL);
 149         EXPECT_EQ(-1, ret);
 150         EXPECT_EQ(EFAULT, errno) {
 151                 TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!");
 152         }
 153 }
 154
 155 TEST(mode_filter_without_nnp)
 156 {
 157         struct sock_filter filter[] = {
 158                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 159         };
 160         struct sock_fprog prog = {
 161                 .len = (unsigned short)ARRAY_SIZE(filter),
 162                 .filter = filter,
 163         };
 164         long ret;
 165
 166         ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0);
 167         ASSERT_LE(0, ret) {
 168                 TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS");
 169         }
 170         errno = 0;
 171         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 172         /* Succeeds with CAP_SYS_ADMIN, fails without */
 173         /* TODO(wad) check caps not euid */
 174         if (geteuid()) {
 175                 EXPECT_EQ(-1, ret);
 176                 EXPECT_EQ(EACCES, errno);
 177         } else {
 178                 EXPECT_EQ(0, ret);
 179         }
 180 }
 181
 182 #define MAX_INSNS_PER_PATH 32768
 183
 184 TEST(filter_size_limits)
 185 {
 186         int i;
 187         int count = BPF_MAXINSNS + 1;
 188         struct sock_filter allow[] = {
 189                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 190         };
 191         struct sock_filter *filter;
 192         struct sock_fprog prog = { };
 193         long ret;
 194
 195         filter = calloc(count, sizeof(*filter));
 196         ASSERT_NE(NULL, filter);
 197
 198         for (i = 0; i < count; i++)
 199                 filter[i] = allow[0];
 200
 201         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 202         ASSERT_EQ(0, ret);
 203
 204         prog.filter = filter;
 205         prog.len = count;
 206
 207         /* Too many filter instructions in a single filter. */
 208         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 209         ASSERT_NE(0, ret) {
 210                 TH_LOG("Installing %d insn filter was allowed", prog.len);
 211         }
 212
 213         /* One less is okay, though. */
 214         prog.len -= 1;
 215         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 216         ASSERT_EQ(0, ret) {
 217                 TH_LOG("Installing %d insn filter wasn't allowed", prog.len);
 218         }
 219 }
 220
 221 TEST(filter_chain_limits)
 222 {
 223         int i;
 224         int count = BPF_MAXINSNS;
 225         struct sock_filter allow[] = {
 226                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 227         };
 228         struct sock_filter *filter;
 229         struct sock_fprog prog = { };
 230         long ret;
 231
 232         filter = calloc(count, sizeof(*filter));
 233         ASSERT_NE(NULL, filter);
 234
 235         for (i = 0; i < count; i++)
 236                 filter[i] = allow[0];
 237
 238         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 239         ASSERT_EQ(0, ret);
 240
 241         prog.filter = filter;
 242         prog.len = 1;
 243
 244         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 245         ASSERT_EQ(0, ret);
 246
 247         prog.len = count;
 248
 249         /* Too many total filter instructions. */
 250         for (i = 0; i < MAX_INSNS_PER_PATH; i++) {
 251                 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 252                 if (ret != 0)
 253                         break;
 254         }
 255         ASSERT_NE(0, ret) {
 256                 TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)",
 257                        i, count, i * (count + 4));
 258         }
 259 }
 260
 261 TEST(mode_filter_cannot_move_to_strict)
 262 {
 263         struct sock_filter filter[] = {
 264                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 265         };
 266         struct sock_fprog prog = {
 267                 .len = (unsigned short)ARRAY_SIZE(filter),
 268                 .filter = filter,
 269         };
 270         long ret;
 271
 272         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 273         ASSERT_EQ(0, ret);
 274
 275         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 276         ASSERT_EQ(0, ret);
 277
 278         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0);
 279         EXPECT_EQ(-1, ret);
 280         EXPECT_EQ(EINVAL, errno);
 281 }
 282
 283
 284 TEST(mode_filter_get_seccomp)
 285 {
 286         struct sock_filter filter[] = {
 287                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 288         };
 289         struct sock_fprog prog = {
 290                 .len = (unsigned short)ARRAY_SIZE(filter),
 291                 .filter = filter,
 292         };
 293         long ret;
 294
 295         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 296         ASSERT_EQ(0, ret);
 297
 298         ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
 299         EXPECT_EQ(0, ret);
 300
 301         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 302         ASSERT_EQ(0, ret);
 303
 304         ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
 305         EXPECT_EQ(2, ret);
 306 }
 307
 308
 309 TEST(ALLOW_all)
 310 {
 311         struct sock_filter filter[] = {
 312                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 313         };
 314         struct sock_fprog prog = {
 315                 .len = (unsigned short)ARRAY_SIZE(filter),
 316                 .filter = filter,
 317         };
 318         long ret;
 319
 320         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 321         ASSERT_EQ(0, ret);
 322
 323         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 324         ASSERT_EQ(0, ret);
 325 }
 326
 327 TEST(empty_prog)
 328 {
 329         struct sock_filter filter[] = {
 330         };
 331         struct sock_fprog prog = {
 332                 .len = (unsigned short)ARRAY_SIZE(filter),
 333                 .filter = filter,
 334         };
 335         long ret;
 336
 337         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 338         ASSERT_EQ(0, ret);
 339
 340         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 341         EXPECT_EQ(-1, ret);
 342         EXPECT_EQ(EINVAL, errno);
 343 }
 344
 345 TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS)
 346 {
 347         struct sock_filter filter[] = {
 348                 BPF_STMT(BPF_RET|BPF_K, 0x10000000U),
 349         };
 350         struct sock_fprog prog = {
 351                 .len = (unsigned short)ARRAY_SIZE(filter),
 352                 .filter = filter,
 353         };
 354         long ret;
 355
 356         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 357         ASSERT_EQ(0, ret);
 358
 359         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 360         ASSERT_EQ(0, ret);
 361         EXPECT_EQ(0, syscall(__NR_getpid)) {
 362                 TH_LOG("getpid() shouldn't ever return");
 363         }
 364 }
 365
 366 /* return code >= 0x80000000 is unused. */
 367 TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS)
 368 {
 369         struct sock_filter filter[] = {
 370                 BPF_STMT(BPF_RET|BPF_K, 0x90000000U),
 371         };
 372         struct sock_fprog prog = {
 373                 .len = (unsigned short)ARRAY_SIZE(filter),
 374                 .filter = filter,
 375         };
 376         long ret;
 377
 378         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 379         ASSERT_EQ(0, ret);
 380
 381         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 382         ASSERT_EQ(0, ret);
 383         EXPECT_EQ(0, syscall(__NR_getpid)) {
 384                 TH_LOG("getpid() shouldn't ever return");
 385         }
 386 }
 387
 388 TEST_SIGNAL(KILL_all, SIGSYS)
 389 {
 390         struct sock_filter filter[] = {
 391                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
 392         };
 393         struct sock_fprog prog = {
 394                 .len = (unsigned short)ARRAY_SIZE(filter),
 395                 .filter = filter,
 396         };
 397         long ret;
 398
 399         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 400         ASSERT_EQ(0, ret);
 401
 402         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 403         ASSERT_EQ(0, ret);
 404 }
 405
 406 TEST_SIGNAL(KILL_one, SIGSYS)
 407 {
 408         struct sock_filter filter[] = {
 409                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 410                         offsetof(struct seccomp_data, nr)),
 411                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
 412                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
 413                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 414         };
 415         struct sock_fprog prog = {
 416                 .len = (unsigned short)ARRAY_SIZE(filter),
 417                 .filter = filter,
 418         };
 419         long ret;
 420         pid_t parent = getppid();
 421
 422         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 423         ASSERT_EQ(0, ret);
 424
 425         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 426         ASSERT_EQ(0, ret);
 427
 428         EXPECT_EQ(parent, syscall(__NR_getppid));
 429         /* getpid() should never return. */
 430         EXPECT_EQ(0, syscall(__NR_getpid));
 431 }
 432
 433 TEST_SIGNAL(KILL_one_arg_one, SIGSYS)
 434 {
 435         void *fatal_address;
 436         struct sock_filter filter[] = {
 437                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 438                         offsetof(struct seccomp_data, nr)),
 439                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0),
 440                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 441                 /* Only both with lower 32-bit for now. */
 442                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)),
 443                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K,
 444                         (unsigned long)&fatal_address, 0, 1),
 445                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
 446                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 447         };
 448         struct sock_fprog prog = {
 449                 .len = (unsigned short)ARRAY_SIZE(filter),
 450                 .filter = filter,
 451         };
 452         long ret;
 453         pid_t parent = getppid();
 454         struct tms timebuf;
 455         clock_t clock = times(&timebuf);
 456
 457         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 458         ASSERT_EQ(0, ret);
 459
 460         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 461         ASSERT_EQ(0, ret);
 462
 463         EXPECT_EQ(parent, syscall(__NR_getppid));
 464         EXPECT_LE(clock, syscall(__NR_times, &timebuf));
 465         /* times() should never return. */
 466         EXPECT_EQ(0, syscall(__NR_times, &fatal_address));
 467 }
 468
 469 TEST_SIGNAL(KILL_one_arg_six, SIGSYS)
 470 {
 471 #ifndef __NR_mmap2
 472         int sysno = __NR_mmap;
 473 #else
 474         int sysno = __NR_mmap2;
 475 #endif
 476         struct sock_filter filter[] = {
 477                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 478                         offsetof(struct seccomp_data, nr)),
 479                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0),
 480                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 481                 /* Only both with lower 32-bit for now. */
 482                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)),
 483                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1),
 484                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
 485                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 486         };
 487         struct sock_fprog prog = {
 488                 .len = (unsigned short)ARRAY_SIZE(filter),
 489                 .filter = filter,
 490         };
 491         long ret;
 492         pid_t parent = getppid();
 493         int fd;
 494         void *map1, *map2;
 495         int page_size = sysconf(_SC_PAGESIZE);
 496
 497         ASSERT_LT(0, page_size);
 498
 499         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 500         ASSERT_EQ(0, ret);
 501
 502         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 503         ASSERT_EQ(0, ret);
 504
 505         fd = open("/dev/zero", O_RDONLY);
 506         ASSERT_NE(-1, fd);
 507
 508         EXPECT_EQ(parent, syscall(__NR_getppid));
 509         map1 = (void *)syscall(sysno,
 510                 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size);
 511         EXPECT_NE(MAP_FAILED, map1);
 512         /* mmap2() should never return. */
 513         map2 = (void *)syscall(sysno,
 514                  NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE);
 515         EXPECT_EQ(MAP_FAILED, map2);
 516
 517         /* The test failed, so clean up the resources. */
 518         munmap(map1, page_size);
 519         munmap(map2, page_size);
 520         close(fd);
 521 }
 522
 523 /* TODO(wad) add 64-bit versus 32-bit arg tests. */
 524 TEST(arg_out_of_range)
 525 {
 526         struct sock_filter filter[] = {
 527                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)),
 528                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 529         };
 530         struct sock_fprog prog = {
 531                 .len = (unsigned short)ARRAY_SIZE(filter),
 532                 .filter = filter,
 533         };
 534         long ret;
 535
 536         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 537         ASSERT_EQ(0, ret);
 538
 539         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 540         EXPECT_EQ(-1, ret);
 541         EXPECT_EQ(EINVAL, errno);
 542 }
 543
 544 TEST(ERRNO_valid)
 545 {
 546         struct sock_filter filter[] = {
 547                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 548                         offsetof(struct seccomp_data, nr)),
 549                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
 550                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | E2BIG),
 551                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 552         };
 553         struct sock_fprog prog = {
 554                 .len = (unsigned short)ARRAY_SIZE(filter),
 555                 .filter = filter,
 556         };
 557         long ret;
 558         pid_t parent = getppid();
 559
 560         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 561         ASSERT_EQ(0, ret);
 562
 563         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 564         ASSERT_EQ(0, ret);
 565
 566         EXPECT_EQ(parent, syscall(__NR_getppid));
 567         EXPECT_EQ(-1, read(0, NULL, 0));
 568         EXPECT_EQ(E2BIG, errno);
 569 }
 570
 571 TEST(ERRNO_zero)
 572 {
 573         struct sock_filter filter[] = {
 574                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 575                         offsetof(struct seccomp_data, nr)),
 576                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
 577                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | 0),
 578                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 579         };
 580         struct sock_fprog prog = {
 581                 .len = (unsigned short)ARRAY_SIZE(filter),
 582                 .filter = filter,
 583         };
 584         long ret;
 585         pid_t parent = getppid();
 586
 587         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 588         ASSERT_EQ(0, ret);
 589
 590         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 591         ASSERT_EQ(0, ret);
 592
 593         EXPECT_EQ(parent, syscall(__NR_getppid));
 594         /* "errno" of 0 is ok. */
 595         EXPECT_EQ(0, read(0, NULL, 0));
 596 }
 597
 598 TEST(ERRNO_capped)
 599 {
 600         struct sock_filter filter[] = {
 601                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 602                         offsetof(struct seccomp_data, nr)),
 603                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
 604                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | 4096),
 605                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 606         };
 607         struct sock_fprog prog = {
 608                 .len = (unsigned short)ARRAY_SIZE(filter),
 609                 .filter = filter,
 610         };
 611         long ret;
 612         pid_t parent = getppid();
 613
 614         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 615         ASSERT_EQ(0, ret);
 616
 617         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 618         ASSERT_EQ(0, ret);
 619
 620         EXPECT_EQ(parent, syscall(__NR_getppid));
 621         EXPECT_EQ(-1, read(0, NULL, 0));
 622         EXPECT_EQ(4095, errno);
 623 }
 624
 625 FIXTURE_DATA(TRAP) {
 626         struct sock_fprog prog;
 627 };
 628
 629 FIXTURE_SETUP(TRAP)
 630 {
 631         struct sock_filter filter[] = {
 632                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 633                         offsetof(struct seccomp_data, nr)),
 634                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
 635                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
 636                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 637         };
 638
 639         memset(&self->prog, 0, sizeof(self->prog));
 640         self->prog.filter = malloc(sizeof(filter));
 641         ASSERT_NE(NULL, self->prog.filter);
 642         memcpy(self->prog.filter, filter, sizeof(filter));
 643         self->prog.len = (unsigned short)ARRAY_SIZE(filter);
 644 }
 645
 646 FIXTURE_TEARDOWN(TRAP)
 647 {
 648         if (self->prog.filter)
 649                 free(self->prog.filter);
 650 }
 651
 652 TEST_F_SIGNAL(TRAP, dfl, SIGSYS)
 653 {
 654         long ret;
 655
 656         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 657         ASSERT_EQ(0, ret);
 658
 659         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
 660         ASSERT_EQ(0, ret);
 661         syscall(__NR_getpid);
 662 }
 663
 664 /* Ensure that SIGSYS overrides SIG_IGN */
 665 TEST_F_SIGNAL(TRAP, ign, SIGSYS)
 666 {
 667         long ret;
 668
 669         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 670         ASSERT_EQ(0, ret);
 671
 672         signal(SIGSYS, SIG_IGN);
 673
 674         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
 675         ASSERT_EQ(0, ret);
 676         syscall(__NR_getpid);
 677 }
 678
 679 static struct siginfo TRAP_info;
 680 static volatile int TRAP_nr;
 681 static void TRAP_action(int nr, siginfo_t *info, void *void_context)
 682 {
 683         memcpy(&TRAP_info, info, sizeof(TRAP_info));
 684         TRAP_nr = nr;
 685 }
 686
 687 TEST_F(TRAP, handler)
 688 {
 689         int ret, test;
 690         struct sigaction act;
 691         sigset_t mask;
 692
 693         memset(&act, 0, sizeof(act));
 694         sigemptyset(&mask);
 695         sigaddset(&mask, SIGSYS);
 696
 697         act.sa_sigaction = &TRAP_action;
 698         act.sa_flags = SA_SIGINFO;
 699         ret = sigaction(SIGSYS, &act, NULL);
 700         ASSERT_EQ(0, ret) {
 701                 TH_LOG("sigaction failed");
 702         }
 703         ret = sigprocmask(SIG_UNBLOCK, &mask, NULL);
 704         ASSERT_EQ(0, ret) {
 705                 TH_LOG("sigprocmask failed");
 706         }
 707
 708         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 709         ASSERT_EQ(0, ret);
 710         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
 711         ASSERT_EQ(0, ret);
 712         TRAP_nr = 0;
 713         memset(&TRAP_info, 0, sizeof(TRAP_info));
 714         /* Expect the registers to be rolled back. (nr = error) may vary
 715          * based on arch. */
 716         ret = syscall(__NR_getpid);
 717         /* Silence gcc warning about volatile. */
 718         test = TRAP_nr;
 719         EXPECT_EQ(SIGSYS, test);
 720         struct local_sigsys {
 721                 void *_call_addr;       /* calling user insn */
 722                 int _syscall;           /* triggering system call number */
 723                 unsigned int _arch;     /* AUDIT_ARCH_* of syscall */
 724         } *sigsys = (struct local_sigsys *)
 725 #ifdef si_syscall
 726                 &(TRAP_info.si_call_addr);
 727 #else
 728                 &TRAP_info.si_pid;
 729 #endif
 730         EXPECT_EQ(__NR_getpid, sigsys->_syscall);
 731         /* Make sure arch is non-zero. */
 732         EXPECT_NE(0, sigsys->_arch);
 733         EXPECT_NE(0, (unsigned long)sigsys->_call_addr);
 734 }
 735
 736 FIXTURE_DATA(precedence) {
 737         struct sock_fprog allow;
 738         struct sock_fprog trace;
 739         struct sock_fprog error;
 740         struct sock_fprog trap;
 741         struct sock_fprog kill;
 742 };
 743
 744 FIXTURE_SETUP(precedence)
 745 {
 746         struct sock_filter allow_insns[] = {
 747                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 748         };
 749         struct sock_filter trace_insns[] = {
 750                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 751                         offsetof(struct seccomp_data, nr)),
 752                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
 753                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 754                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE),
 755         };
 756         struct sock_filter error_insns[] = {
 757                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 758                         offsetof(struct seccomp_data, nr)),
 759                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
 760                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 761                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO),
 762         };
 763         struct sock_filter trap_insns[] = {
 764                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 765                         offsetof(struct seccomp_data, nr)),
 766                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
 767                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 768                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
 769         };
 770         struct sock_filter kill_insns[] = {
 771                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 772                         offsetof(struct seccomp_data, nr)),
 773                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
 774                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 775                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
 776         };
 777
 778         memset(self, 0, sizeof(*self));
 779 #define FILTER_ALLOC(_x) \
 780         self->_x.filter = malloc(sizeof(_x##_insns)); \
 781         ASSERT_NE(NULL, self->_x.filter); \
 782         memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \
 783         self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns)
 784         FILTER_ALLOC(allow);
 785         FILTER_ALLOC(trace);
 786         FILTER_ALLOC(error);
 787         FILTER_ALLOC(trap);
 788         FILTER_ALLOC(kill);
 789 }
 790
 791 FIXTURE_TEARDOWN(precedence)
 792 {
 793 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter)
 794         FILTER_FREE(allow);
 795         FILTER_FREE(trace);
 796         FILTER_FREE(error);
 797         FILTER_FREE(trap);
 798         FILTER_FREE(kill);
 799 }
 800
 801 TEST_F(precedence, allow_ok)
 802 {
 803         pid_t parent, res = 0;
 804         long ret;
 805
 806         parent = getppid();
 807         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 808         ASSERT_EQ(0, ret);
 809
 810         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 811         ASSERT_EQ(0, ret);
 812         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 813         ASSERT_EQ(0, ret);
 814         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 815         ASSERT_EQ(0, ret);
 816         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
 817         ASSERT_EQ(0, ret);
 818         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
 819         ASSERT_EQ(0, ret);
 820         /* Should work just fine. */
 821         res = syscall(__NR_getppid);
 822         EXPECT_EQ(parent, res);
 823 }
 824
 825 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS)
 826 {
 827         pid_t parent, res = 0;
 828         long ret;
 829
 830         parent = getppid();
 831         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 832         ASSERT_EQ(0, ret);
 833
 834         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 835         ASSERT_EQ(0, ret);
 836         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 837         ASSERT_EQ(0, ret);
 838         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 839         ASSERT_EQ(0, ret);
 840         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
 841         ASSERT_EQ(0, ret);
 842         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
 843         ASSERT_EQ(0, ret);
 844         /* Should work just fine. */
 845         res = syscall(__NR_getppid);
 846         EXPECT_EQ(parent, res);
 847         /* getpid() should never return. */
 848         res = syscall(__NR_getpid);
 849         EXPECT_EQ(0, res);
 850 }
 851
 852 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS)
 853 {
 854         pid_t parent;
 855         long ret;
 856
 857         parent = getppid();
 858         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 859         ASSERT_EQ(0, ret);
 860
 861         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 862         ASSERT_EQ(0, ret);
 863         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
 864         ASSERT_EQ(0, ret);
 865         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 866         ASSERT_EQ(0, ret);
 867         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 868         ASSERT_EQ(0, ret);
 869         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
 870         ASSERT_EQ(0, ret);
 871         /* Should work just fine. */
 872         EXPECT_EQ(parent, syscall(__NR_getppid));
 873         /* getpid() should never return. */
 874         EXPECT_EQ(0, syscall(__NR_getpid));
 875 }
 876
 877 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS)
 878 {
 879         pid_t parent;
 880         long ret;
 881
 882         parent = getppid();
 883         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 884         ASSERT_EQ(0, ret);
 885
 886         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 887         ASSERT_EQ(0, ret);
 888         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 889         ASSERT_EQ(0, ret);
 890         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 891         ASSERT_EQ(0, ret);
 892         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
 893         ASSERT_EQ(0, ret);
 894         /* Should work just fine. */
 895         EXPECT_EQ(parent, syscall(__NR_getppid));
 896         /* getpid() should never return. */
 897         EXPECT_EQ(0, syscall(__NR_getpid));
 898 }
 899
 900 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS)
 901 {
 902         pid_t parent;
 903         long ret;
 904
 905         parent = getppid();
 906         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 907         ASSERT_EQ(0, ret);
 908
 909         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 910         ASSERT_EQ(0, ret);
 911         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
 912         ASSERT_EQ(0, ret);
 913         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 914         ASSERT_EQ(0, ret);
 915         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 916         ASSERT_EQ(0, ret);
 917         /* Should work just fine. */
 918         EXPECT_EQ(parent, syscall(__NR_getppid));
 919         /* getpid() should never return. */
 920         EXPECT_EQ(0, syscall(__NR_getpid));
 921 }
 922
 923 TEST_F(precedence, errno_is_third)
 924 {
 925         pid_t parent;
 926         long ret;
 927
 928         parent = getppid();
 929         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 930         ASSERT_EQ(0, ret);
 931
 932         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 933         ASSERT_EQ(0, ret);
 934         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 935         ASSERT_EQ(0, ret);
 936         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 937         ASSERT_EQ(0, ret);
 938         /* Should work just fine. */
 939         EXPECT_EQ(parent, syscall(__NR_getppid));
 940         EXPECT_EQ(0, syscall(__NR_getpid));
 941 }
 942
 943 TEST_F(precedence, errno_is_third_in_any_order)
 944 {
 945         pid_t parent;
 946         long ret;
 947
 948         parent = getppid();
 949         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 950         ASSERT_EQ(0, ret);
 951
 952         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
 953         ASSERT_EQ(0, ret);
 954         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 955         ASSERT_EQ(0, ret);
 956         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 957         ASSERT_EQ(0, ret);
 958         /* Should work just fine. */
 959         EXPECT_EQ(parent, syscall(__NR_getppid));
 960         EXPECT_EQ(0, syscall(__NR_getpid));
 961 }
 962
 963 TEST_F(precedence, trace_is_fourth)
 964 {
 965         pid_t parent;
 966         long ret;
 967
 968         parent = getppid();
 969         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 970         ASSERT_EQ(0, ret);
 971
 972         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 973         ASSERT_EQ(0, ret);
 974         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 975         ASSERT_EQ(0, ret);
 976         /* Should work just fine. */
 977         EXPECT_EQ(parent, syscall(__NR_getppid));
 978         /* No ptracer */
 979         EXPECT_EQ(-1, syscall(__NR_getpid));
 980 }
 981
 982 TEST_F(precedence, trace_is_fourth_in_any_order)
 983 {
 984         pid_t parent;
 985         long ret;
 986
 987         parent = getppid();
 988         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 989         ASSERT_EQ(0, ret);
 990
 991         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
 992         ASSERT_EQ(0, ret);
 993         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
 994         ASSERT_EQ(0, ret);
 995         /* Should work just fine. */
 996         EXPECT_EQ(parent, syscall(__NR_getppid));
 997         /* No ptracer */
 998         EXPECT_EQ(-1, syscall(__NR_getpid));
 999 }
1000
1001 #ifndef PTRACE_O_TRACESECCOMP
1002 #define PTRACE_O_TRACESECCOMP   0x00000080
1003 #endif
1004
1005 /* Catch the Ubuntu 12.04 value error. */
1006 #if PTRACE_EVENT_SECCOMP != 7
1007 #undef PTRACE_EVENT_SECCOMP
1008 #endif
1009
1010 #ifndef PTRACE_EVENT_SECCOMP
1011 #define PTRACE_EVENT_SECCOMP 7
1012 #endif
1013
1014 #define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP)
1015 bool tracer_running;
1016 void tracer_stop(int sig)
1017 {
1018         tracer_running = false;
1019 }
1020
1021 typedef void tracer_func_t(struct __test_metadata *_metadata,
1022                            pid_t tracee, int status, void *args);
1023
1024 void start_tracer(struct __test_metadata *_metadata, int fd, pid_t tracee,
1025             tracer_func_t tracer_func, void *args, bool ptrace_syscall)
1026 {
1027         int ret = -1;
1028         struct sigaction action = {
1029                 .sa_handler = tracer_stop,
1030         };
1031
1032         /* Allow external shutdown. */
1033         tracer_running = true;
1034         ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL));
1035
1036         errno = 0;
1037         while (ret == -1 && errno != EINVAL)
1038                 ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0);
1039         ASSERT_EQ(0, ret) {
1040                 kill(tracee, SIGKILL);
1041         }
1042         /* Wait for attach stop */
1043         wait(NULL);
1044
1045         ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, ptrace_syscall ?
1046                                                       PTRACE_O_TRACESYSGOOD :
1047                                                       PTRACE_O_TRACESECCOMP);
1048         ASSERT_EQ(0, ret) {
1049                 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP");
1050                 kill(tracee, SIGKILL);
1051         }
1052         ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1053                      tracee, NULL, 0);
1054         ASSERT_EQ(0, ret);
1055
1056         /* Unblock the tracee */
1057         ASSERT_EQ(1, write(fd, "A", 1));
1058         ASSERT_EQ(0, close(fd));
1059
1060         /* Run until we're shut down. Must assert to stop execution. */
1061         while (tracer_running) {
1062                 int status;
1063
1064                 if (wait(&status) != tracee)
1065                         continue;
1066                 if (WIFSIGNALED(status) || WIFEXITED(status))
1067                         /* Child is dead. Time to go. */
1068                         return;
1069
1070                 /* Check if this is a seccomp event. */
1071                 ASSERT_EQ(!ptrace_syscall, IS_SECCOMP_EVENT(status));
1072
1073                 tracer_func(_metadata, tracee, status, args);
1074
1075                 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1076                              tracee, NULL, 0);
1077                 ASSERT_EQ(0, ret);
1078         }
1079         /* Directly report the status of our test harness results. */
1080         syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS : EXIT_FAILURE);
1081 }
1082
1083 /* Common tracer setup/teardown functions. */
1084 void cont_handler(int num)
1085 { }
1086 pid_t setup_trace_fixture(struct __test_metadata *_metadata,
1087                           tracer_func_t func, void *args, bool ptrace_syscall)
1088 {
1089         char sync;
1090         int pipefd[2];
1091         pid_t tracer_pid;
1092         pid_t tracee = getpid();
1093
1094         /* Setup a pipe for clean synchronization. */
1095         ASSERT_EQ(0, pipe(pipefd));
1096
1097         /* Fork a child which we'll promote to tracer */
1098         tracer_pid = fork();
1099         ASSERT_LE(0, tracer_pid);
1100         signal(SIGALRM, cont_handler);
1101         if (tracer_pid == 0) {
1102                 close(pipefd[0]);
1103                 start_tracer(_metadata, pipefd[1], tracee, func, args,
1104                              ptrace_syscall);
1105                 syscall(__NR_exit, 0);
1106         }
1107         close(pipefd[1]);
1108         prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0);
1109         read(pipefd[0], &sync, 1);
1110         close(pipefd[0]);
1111
1112         return tracer_pid;
1113 }
1114 void teardown_trace_fixture(struct __test_metadata *_metadata,
1115                             pid_t tracer)
1116 {
1117         if (tracer) {
1118                 int status;
1119                 /*
1120                  * Extract the exit code from the other process and
1121                  * adopt it for ourselves in case its asserts failed.
1122                  */
1123                 ASSERT_EQ(0, kill(tracer, SIGUSR1));
1124                 ASSERT_EQ(tracer, waitpid(tracer, &status, 0));
1125                 if (WEXITSTATUS(status))
1126                         _metadata->passed = 0;
1127         }
1128 }
1129
1130 /* "poke" tracer arguments and function. */
1131 struct tracer_args_poke_t {
1132         unsigned long poke_addr;
1133 };
1134
1135 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status,
1136                  void *args)
1137 {
1138         int ret;
1139         unsigned long msg;
1140         struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args;
1141
1142         ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1143         EXPECT_EQ(0, ret);
1144         /* If this fails, don't try to recover. */
1145         ASSERT_EQ(0x1001, msg) {
1146                 kill(tracee, SIGKILL);
1147         }
1148         /*
1149          * Poke in the message.
1150          * Registers are not touched to try to keep this relatively arch
1151          * agnostic.
1152          */
1153         ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001);
1154         EXPECT_EQ(0, ret);
1155 }
1156
1157 FIXTURE_DATA(TRACE_poke) {
1158         struct sock_fprog prog;
1159         pid_t tracer;
1160         long poked;
1161         struct tracer_args_poke_t tracer_args;
1162 };
1163
1164 FIXTURE_SETUP(TRACE_poke)
1165 {
1166         struct sock_filter filter[] = {
1167                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1168                         offsetof(struct seccomp_data, nr)),
1169                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1170                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001),
1171                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1172         };
1173
1174         self->poked = 0;
1175         memset(&self->prog, 0, sizeof(self->prog));
1176         self->prog.filter = malloc(sizeof(filter));
1177         ASSERT_NE(NULL, self->prog.filter);
1178         memcpy(self->prog.filter, filter, sizeof(filter));
1179         self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1180
1181         /* Set up tracer args. */
1182         self->tracer_args.poke_addr = (unsigned long)&self->poked;
1183
1184         /* Launch tracer. */
1185         self->tracer = setup_trace_fixture(_metadata, tracer_poke,
1186                                            &self->tracer_args, false);
1187 }
1188
1189 FIXTURE_TEARDOWN(TRACE_poke)
1190 {
1191         teardown_trace_fixture(_metadata, self->tracer);
1192         if (self->prog.filter)
1193                 free(self->prog.filter);
1194 }
1195
1196 TEST_F(TRACE_poke, read_has_side_effects)
1197 {
1198         ssize_t ret;
1199
1200         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1201         ASSERT_EQ(0, ret);
1202
1203         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1204         ASSERT_EQ(0, ret);
1205
1206         EXPECT_EQ(0, self->poked);
1207         ret = read(-1, NULL, 0);
1208         EXPECT_EQ(-1, ret);
1209         EXPECT_EQ(0x1001, self->poked);
1210 }
1211
1212 TEST_F(TRACE_poke, getpid_runs_normally)
1213 {
1214         long ret;
1215
1216         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1217         ASSERT_EQ(0, ret);
1218
1219         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1220         ASSERT_EQ(0, ret);
1221
1222         EXPECT_EQ(0, self->poked);
1223         EXPECT_NE(0, syscall(__NR_getpid));
1224         EXPECT_EQ(0, self->poked);
1225 }
1226
1227 #if defined(__x86_64__)
1228 # define ARCH_REGS      struct user_regs_struct
1229 # define SYSCALL_NUM    orig_rax
1230 # define SYSCALL_RET    rax
1231 #elif defined(__i386__)
1232 # define ARCH_REGS      struct user_regs_struct
1233 # define SYSCALL_NUM    orig_eax
1234 # define SYSCALL_RET    eax
1235 #elif defined(__arm__)
1236 # define ARCH_REGS      struct pt_regs
1237 # define SYSCALL_NUM    ARM_r7
1238 # define SYSCALL_RET    ARM_r0
1239 #elif defined(__aarch64__)
1240 # define ARCH_REGS      struct user_pt_regs
1241 # define SYSCALL_NUM    regs[8]
1242 # define SYSCALL_RET    regs[0]
1243 #elif defined(__hppa__)
1244 # define ARCH_REGS      struct user_regs_struct
1245 # define SYSCALL_NUM    gr[20]
1246 # define SYSCALL_RET    gr[28]
1247 #elif defined(__powerpc__)
1248 # define ARCH_REGS      struct pt_regs
1249 # define SYSCALL_NUM    gpr[0]
1250 # define SYSCALL_RET    gpr[3]
1251 #elif defined(__s390__)
1252 # define ARCH_REGS     s390_regs
1253 # define SYSCALL_NUM   gprs[2]
1254 # define SYSCALL_RET   gprs[2]
1255 #elif defined(__mips__)
1256 # define ARCH_REGS      struct pt_regs
1257 # define SYSCALL_NUM    regs[2]
1258 # define SYSCALL_SYSCALL_NUM regs[4]
1259 # define SYSCALL_RET    regs[2]
1260 # define SYSCALL_NUM_RET_SHARE_REG
1261 #else
1262 # error "Do not know how to find your architecture's registers and syscalls"
1263 #endif
1264
1265 /* Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for
1266  * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux).
1267  */
1268 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__)
1269 #define HAVE_GETREGS
1270 #endif
1271
1272 /* Architecture-specific syscall fetching routine. */
1273 int get_syscall(struct __test_metadata *_metadata, pid_t tracee)
1274 {
1275         ARCH_REGS regs;
1276 #ifdef HAVE_GETREGS
1277         EXPECT_EQ(0, ptrace(PTRACE_GETREGS, tracee, 0, &regs)) {
1278                 TH_LOG("PTRACE_GETREGS failed");
1279                 return -1;
1280         }
1281 #else
1282         struct iovec iov;
1283
1284         iov.iov_base = &regs;
1285         iov.iov_len = sizeof(regs);
1286         EXPECT_EQ(0, ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov)) {
1287                 TH_LOG("PTRACE_GETREGSET failed");
1288                 return -1;
1289         }
1290 #endif
1291
1292 #if defined(__mips__)
1293         if (regs.SYSCALL_NUM == __NR_O32_Linux)
1294                 return regs.SYSCALL_SYSCALL_NUM;
1295 #endif
1296         return regs.SYSCALL_NUM;
1297 }
1298
1299 /* Architecture-specific syscall changing routine. */
1300 void change_syscall(struct __test_metadata *_metadata,
1301                     pid_t tracee, int syscall)
1302 {
1303         int ret;
1304         ARCH_REGS regs;
1305 #ifdef HAVE_GETREGS
1306         ret = ptrace(PTRACE_GETREGS, tracee, 0, &regs);
1307 #else
1308         struct iovec iov;
1309         iov.iov_base = &regs;
1310         iov.iov_len = sizeof(regs);
1311         ret = ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov);
1312 #endif
1313         EXPECT_EQ(0, ret);
1314
1315 #if defined(__x86_64__) || defined(__i386__) || defined(__powerpc__) || \
1316     defined(__s390__) || defined(__hppa__)
1317         {
1318                 regs.SYSCALL_NUM = syscall;
1319         }
1320 #elif defined(__mips__)
1321         {
1322                 if (regs.SYSCALL_NUM == __NR_O32_Linux)
1323                         regs.SYSCALL_SYSCALL_NUM = syscall;
1324                 else
1325                         regs.SYSCALL_NUM = syscall;
1326         }
1327
1328 #elif defined(__arm__)
1329 # ifndef PTRACE_SET_SYSCALL
1330 #  define PTRACE_SET_SYSCALL   23
1331 # endif
1332         {
1333                 ret = ptrace(PTRACE_SET_SYSCALL, tracee, NULL, syscall);
1334                 EXPECT_EQ(0, ret);
1335         }
1336
1337 #elif defined(__aarch64__)
1338 # ifndef NT_ARM_SYSTEM_CALL
1339 #  define NT_ARM_SYSTEM_CALL 0x404
1340 # endif
1341         {
1342                 iov.iov_base = &syscall;
1343                 iov.iov_len = sizeof(syscall);
1344                 ret = ptrace(PTRACE_SETREGSET, tracee, NT_ARM_SYSTEM_CALL,
1345                              &iov);
1346                 EXPECT_EQ(0, ret);
1347         }
1348
1349 #else
1350         ASSERT_EQ(1, 0) {
1351                 TH_LOG("How is the syscall changed on this architecture?");
1352         }
1353 #endif
1354
1355         /* If syscall is skipped, change return value. */
1356         if (syscall == -1)
1357 #ifdef SYSCALL_NUM_RET_SHARE_REG
1358                 TH_LOG("Can't modify syscall return on this architecture");
1359 #else
1360                 regs.SYSCALL_RET = 1;
1361 #endif
1362
1363 #ifdef HAVE_GETREGS
1364         ret = ptrace(PTRACE_SETREGS, tracee, 0, &regs);
1365 #else
1366         iov.iov_base = &regs;
1367         iov.iov_len = sizeof(regs);
1368         ret = ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &iov);
1369 #endif
1370         EXPECT_EQ(0, ret);
1371 }
1372
1373 void tracer_syscall(struct __test_metadata *_metadata, pid_t tracee,
1374                     int status, void *args)
1375 {
1376         int ret;
1377         unsigned long msg;
1378
1379         /* Make sure we got the right message. */
1380         ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1381         EXPECT_EQ(0, ret);
1382
1383         /* Validate and take action on expected syscalls. */
1384         switch (msg) {
1385         case 0x1002:
1386                 /* change getpid to getppid. */
1387                 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee));
1388                 change_syscall(_metadata, tracee, __NR_getppid);
1389                 break;
1390         case 0x1003:
1391                 /* skip gettid. */
1392                 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee));
1393                 change_syscall(_metadata, tracee, -1);
1394                 break;
1395         case 0x1004:
1396                 /* do nothing (allow getppid) */
1397                 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee));
1398                 break;
1399         default:
1400                 EXPECT_EQ(0, msg) {
1401                         TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg);
1402                         kill(tracee, SIGKILL);
1403                 }
1404         }
1405
1406 }
1407
1408 void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee,
1409                    int status, void *args)
1410 {
1411         int ret, nr;
1412         unsigned long msg;
1413         static bool entry;
1414
1415         /* Make sure we got an empty message. */
1416         ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1417         EXPECT_EQ(0, ret);
1418         EXPECT_EQ(0, msg);
1419
1420         /* The only way to tell PTRACE_SYSCALL entry/exit is by counting. */
1421         entry = !entry;
1422         if (!entry)
1423                 return;
1424
1425         nr = get_syscall(_metadata, tracee);
1426
1427         if (nr == __NR_getpid)
1428                 change_syscall(_metadata, tracee, __NR_getppid);
1429 }
1430
1431 FIXTURE_DATA(TRACE_syscall) {
1432         struct sock_fprog prog;
1433         pid_t tracer, mytid, mypid, parent;
1434 };
1435
1436 FIXTURE_SETUP(TRACE_syscall)
1437 {
1438         struct sock_filter filter[] = {
1439                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1440                         offsetof(struct seccomp_data, nr)),
1441                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
1442                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002),
1443                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1),
1444                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003),
1445                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1446                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004),
1447                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1448         };
1449
1450         memset(&self->prog, 0, sizeof(self->prog));
1451         self->prog.filter = malloc(sizeof(filter));
1452         ASSERT_NE(NULL, self->prog.filter);
1453         memcpy(self->prog.filter, filter, sizeof(filter));
1454         self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1455
1456         /* Prepare some testable syscall results. */
1457         self->mytid = syscall(__NR_gettid);
1458         ASSERT_GT(self->mytid, 0);
1459         ASSERT_NE(self->mytid, 1) {
1460                 TH_LOG("Running this test as init is not supported. :)");
1461         }
1462
1463         self->mypid = getpid();
1464         ASSERT_GT(self->mypid, 0);
1465         ASSERT_EQ(self->mytid, self->mypid);
1466
1467         self->parent = getppid();
1468         ASSERT_GT(self->parent, 0);
1469         ASSERT_NE(self->parent, self->mypid);
1470
1471         /* Launch tracer. */
1472         self->tracer = setup_trace_fixture(_metadata, tracer_syscall, NULL,
1473                                            false);
1474 }
1475
1476 FIXTURE_TEARDOWN(TRACE_syscall)
1477 {
1478         teardown_trace_fixture(_metadata, self->tracer);
1479         if (self->prog.filter)
1480                 free(self->prog.filter);
1481 }
1482
1483 TEST_F(TRACE_syscall, syscall_allowed)
1484 {
1485         long ret;
1486
1487         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1488         ASSERT_EQ(0, ret);
1489
1490         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1491         ASSERT_EQ(0, ret);
1492
1493         /* getppid works as expected (no changes). */
1494         EXPECT_EQ(self->parent, syscall(__NR_getppid));
1495         EXPECT_NE(self->mypid, syscall(__NR_getppid));
1496 }
1497
1498 TEST_F(TRACE_syscall, syscall_redirected)
1499 {
1500         long ret;
1501
1502         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1503         ASSERT_EQ(0, ret);
1504
1505         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1506         ASSERT_EQ(0, ret);
1507
1508         /* getpid has been redirected to getppid as expected. */
1509         EXPECT_EQ(self->parent, syscall(__NR_getpid));
1510         EXPECT_NE(self->mypid, syscall(__NR_getpid));
1511 }
1512
1513 TEST_F(TRACE_syscall, syscall_dropped)
1514 {
1515         long ret;
1516
1517         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1518         ASSERT_EQ(0, ret);
1519
1520         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1521         ASSERT_EQ(0, ret);
1522
1523 #ifdef SYSCALL_NUM_RET_SHARE_REG
1524         /* gettid has been skipped */
1525         EXPECT_EQ(-1, syscall(__NR_gettid));
1526 #else
1527         /* gettid has been skipped and an altered return value stored. */
1528         EXPECT_EQ(1, syscall(__NR_gettid));
1529 #endif
1530         EXPECT_NE(self->mytid, syscall(__NR_gettid));
1531 }
1532
1533 TEST_F(TRACE_syscall, skip_after_RET_TRACE)
1534 {
1535         struct sock_filter filter[] = {
1536                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1537                         offsetof(struct seccomp_data, nr)),
1538                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1539                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
1540                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1541         };
1542         struct sock_fprog prog = {
1543                 .len = (unsigned short)ARRAY_SIZE(filter),
1544                 .filter = filter,
1545         };
1546         long ret;
1547
1548         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1549         ASSERT_EQ(0, ret);
1550
1551         /* Install fixture filter. */
1552         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1553         ASSERT_EQ(0, ret);
1554
1555         /* Install "errno on getppid" filter. */
1556         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1557         ASSERT_EQ(0, ret);
1558
1559         /* Tracer will redirect getpid to getppid, and we should see EPERM. */
1560         EXPECT_EQ(-1, syscall(__NR_getpid));
1561         EXPECT_EQ(EPERM, errno);
1562 }
1563
1564 TEST_F_SIGNAL(TRACE_syscall, kill_after_RET_TRACE, SIGSYS)
1565 {
1566         struct sock_filter filter[] = {
1567                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1568                         offsetof(struct seccomp_data, nr)),
1569                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1570                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1571                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1572         };
1573         struct sock_fprog prog = {
1574                 .len = (unsigned short)ARRAY_SIZE(filter),
1575                 .filter = filter,
1576         };
1577         long ret;
1578
1579         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1580         ASSERT_EQ(0, ret);
1581
1582         /* Install fixture filter. */
1583         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1584         ASSERT_EQ(0, ret);
1585
1586         /* Install "death on getppid" filter. */
1587         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1588         ASSERT_EQ(0, ret);
1589
1590         /* Tracer will redirect getpid to getppid, and we should die. */
1591         EXPECT_NE(self->mypid, syscall(__NR_getpid));
1592 }
1593
1594 TEST_F(TRACE_syscall, skip_after_ptrace)
1595 {
1596         struct sock_filter filter[] = {
1597                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1598                         offsetof(struct seccomp_data, nr)),
1599                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1600                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
1601                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1602         };
1603         struct sock_fprog prog = {
1604                 .len = (unsigned short)ARRAY_SIZE(filter),
1605                 .filter = filter,
1606         };
1607         long ret;
1608
1609         /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1610         teardown_trace_fixture(_metadata, self->tracer);
1611         self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1612                                            true);
1613
1614         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1615         ASSERT_EQ(0, ret);
1616
1617         /* Install "errno on getppid" filter. */
1618         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1619         ASSERT_EQ(0, ret);
1620
1621         /* Tracer will redirect getpid to getppid, and we should see EPERM. */
1622         EXPECT_EQ(-1, syscall(__NR_getpid));
1623         EXPECT_EQ(EPERM, errno);
1624 }
1625
1626 TEST_F_SIGNAL(TRACE_syscall, kill_after_ptrace, SIGSYS)
1627 {
1628         struct sock_filter filter[] = {
1629                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1630                         offsetof(struct seccomp_data, nr)),
1631                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1632                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1633                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1634         };
1635         struct sock_fprog prog = {
1636                 .len = (unsigned short)ARRAY_SIZE(filter),
1637                 .filter = filter,
1638         };
1639         long ret;
1640
1641         /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1642         teardown_trace_fixture(_metadata, self->tracer);
1643         self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1644                                            true);
1645
1646         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1647         ASSERT_EQ(0, ret);
1648
1649         /* Install "death on getppid" filter. */
1650         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1651         ASSERT_EQ(0, ret);
1652
1653         /* Tracer will redirect getpid to getppid, and we should die. */
1654         EXPECT_NE(self->mypid, syscall(__NR_getpid));
1655 }
1656
1657 #ifndef __NR_seccomp
1658 # if defined(__i386__)
1659 #  define __NR_seccomp 354
1660 # elif defined(__x86_64__)
1661 #  define __NR_seccomp 317
1662 # elif defined(__arm__)
1663 #  define __NR_seccomp 383
1664 # elif defined(__aarch64__)
1665 #  define __NR_seccomp 277
1666 # elif defined(__hppa__)
1667 #  define __NR_seccomp 338
1668 # elif defined(__powerpc__)
1669 #  define __NR_seccomp 358
1670 # elif defined(__s390__)
1671 #  define __NR_seccomp 348
1672 # else
1673 #  warning "seccomp syscall number unknown for this architecture"
1674 #  define __NR_seccomp 0xffff
1675 # endif
1676 #endif
1677
1678 #ifndef SECCOMP_SET_MODE_STRICT
1679 #define SECCOMP_SET_MODE_STRICT 0
1680 #endif
1681
1682 #ifndef SECCOMP_SET_MODE_FILTER
1683 #define SECCOMP_SET_MODE_FILTER 1
1684 #endif
1685
1686 #ifndef SECCOMP_FILTER_FLAG_TSYNC
1687 #define SECCOMP_FILTER_FLAG_TSYNC 1
1688 #endif
1689
1690 #ifndef seccomp
1691 int seccomp(unsigned int op, unsigned int flags, void *args)
1692 {
1693         errno = 0;
1694         return syscall(__NR_seccomp, op, flags, args);
1695 }
1696 #endif
1697
1698 TEST(seccomp_syscall)
1699 {
1700         struct sock_filter filter[] = {
1701                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1702         };
1703         struct sock_fprog prog = {
1704                 .len = (unsigned short)ARRAY_SIZE(filter),
1705                 .filter = filter,
1706         };
1707         long ret;
1708
1709         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1710         ASSERT_EQ(0, ret) {
1711                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1712         }
1713
1714         /* Reject insane operation. */
1715         ret = seccomp(-1, 0, &prog);
1716         ASSERT_NE(ENOSYS, errno) {
1717                 TH_LOG("Kernel does not support seccomp syscall!");
1718         }
1719         EXPECT_EQ(EINVAL, errno) {
1720                 TH_LOG("Did not reject crazy op value!");
1721         }
1722
1723         /* Reject strict with flags or pointer. */
1724         ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL);
1725         EXPECT_EQ(EINVAL, errno) {
1726                 TH_LOG("Did not reject mode strict with flags!");
1727         }
1728         ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog);
1729         EXPECT_EQ(EINVAL, errno) {
1730                 TH_LOG("Did not reject mode strict with uargs!");
1731         }
1732
1733         /* Reject insane args for filter. */
1734         ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog);
1735         EXPECT_EQ(EINVAL, errno) {
1736                 TH_LOG("Did not reject crazy filter flags!");
1737         }
1738         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL);
1739         EXPECT_EQ(EFAULT, errno) {
1740                 TH_LOG("Did not reject NULL filter!");
1741         }
1742
1743         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
1744         EXPECT_EQ(0, errno) {
1745                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s",
1746                         strerror(errno));
1747         }
1748 }
1749
1750 TEST(seccomp_syscall_mode_lock)
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, NULL, 0, 0);
1762         ASSERT_EQ(0, ret) {
1763                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1764         }
1765
1766         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
1767         ASSERT_NE(ENOSYS, errno) {
1768                 TH_LOG("Kernel does not support seccomp syscall!");
1769         }
1770         EXPECT_EQ(0, ret) {
1771                 TH_LOG("Could not install filter!");
1772         }
1773
1774         /* Make sure neither entry point will switch to strict. */
1775         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0);
1776         EXPECT_EQ(EINVAL, errno) {
1777                 TH_LOG("Switched to mode strict!");
1778         }
1779
1780         ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL);
1781         EXPECT_EQ(EINVAL, errno) {
1782                 TH_LOG("Switched to mode strict!");
1783         }
1784 }
1785
1786 TEST(TSYNC_first)
1787 {
1788         struct sock_filter filter[] = {
1789                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1790         };
1791         struct sock_fprog prog = {
1792                 .len = (unsigned short)ARRAY_SIZE(filter),
1793                 .filter = filter,
1794         };
1795         long ret;
1796
1797         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
1798         ASSERT_EQ(0, ret) {
1799                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1800         }
1801
1802         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
1803                       &prog);
1804         ASSERT_NE(ENOSYS, errno) {
1805                 TH_LOG("Kernel does not support seccomp syscall!");
1806         }
1807         EXPECT_EQ(0, ret) {
1808                 TH_LOG("Could not install initial filter with TSYNC!");
1809         }
1810 }
1811
1812 #define TSYNC_SIBLINGS 2
1813 struct tsync_sibling {
1814         pthread_t tid;
1815         pid_t system_tid;
1816         sem_t *started;
1817         pthread_cond_t *cond;
1818         pthread_mutex_t *mutex;
1819         int diverge;
1820         int num_waits;
1821         struct sock_fprog *prog;
1822         struct __test_metadata *metadata;
1823 };
1824
1825 FIXTURE_DATA(TSYNC) {
1826         struct sock_fprog root_prog, apply_prog;
1827         struct tsync_sibling sibling[TSYNC_SIBLINGS];
1828         sem_t started;
1829         pthread_cond_t cond;
1830         pthread_mutex_t mutex;
1831         int sibling_count;
1832 };
1833
1834 FIXTURE_SETUP(TSYNC)
1835 {
1836         struct sock_filter root_filter[] = {
1837                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1838         };
1839         struct sock_filter apply_filter[] = {
1840                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1841                         offsetof(struct seccomp_data, nr)),
1842                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1843                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1844                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1845         };
1846
1847         memset(&self->root_prog, 0, sizeof(self->root_prog));
1848         memset(&self->apply_prog, 0, sizeof(self->apply_prog));
1849         memset(&self->sibling, 0, sizeof(self->sibling));
1850         self->root_prog.filter = malloc(sizeof(root_filter));
1851         ASSERT_NE(NULL, self->root_prog.filter);
1852         memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter));
1853         self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter);
1854
1855         self->apply_prog.filter = malloc(sizeof(apply_filter));
1856         ASSERT_NE(NULL, self->apply_prog.filter);
1857         memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter));
1858         self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter);
1859
1860         self->sibling_count = 0;
1861         pthread_mutex_init(&self->mutex, NULL);
1862         pthread_cond_init(&self->cond, NULL);
1863         sem_init(&self->started, 0, 0);
1864         self->sibling[0].tid = 0;
1865         self->sibling[0].cond = &self->cond;
1866         self->sibling[0].started = &self->started;
1867         self->sibling[0].mutex = &self->mutex;
1868         self->sibling[0].diverge = 0;
1869         self->sibling[0].num_waits = 1;
1870         self->sibling[0].prog = &self->root_prog;
1871         self->sibling[0].metadata = _metadata;
1872         self->sibling[1].tid = 0;
1873         self->sibling[1].cond = &self->cond;
1874         self->sibling[1].started = &self->started;
1875         self->sibling[1].mutex = &self->mutex;
1876         self->sibling[1].diverge = 0;
1877         self->sibling[1].prog = &self->root_prog;
1878         self->sibling[1].num_waits = 1;
1879         self->sibling[1].metadata = _metadata;
1880 }
1881
1882 FIXTURE_TEARDOWN(TSYNC)
1883 {
1884         int sib = 0;
1885
1886         if (self->root_prog.filter)
1887                 free(self->root_prog.filter);
1888         if (self->apply_prog.filter)
1889                 free(self->apply_prog.filter);
1890
1891         for ( ; sib < self->sibling_count; ++sib) {
1892                 struct tsync_sibling *s = &self->sibling[sib];
1893                 void *status;
1894
1895                 if (!s->tid)
1896                         continue;
1897                 if (pthread_kill(s->tid, 0)) {
1898                         pthread_cancel(s->tid);
1899                         pthread_join(s->tid, &status);
1900                 }
1901         }
1902         pthread_mutex_destroy(&self->mutex);
1903         pthread_cond_destroy(&self->cond);
1904         sem_destroy(&self->started);
1905 }
1906
1907 void *tsync_sibling(void *data)
1908 {
1909         long ret = 0;
1910         struct tsync_sibling *me = data;
1911
1912         me->system_tid = syscall(__NR_gettid);
1913
1914         pthread_mutex_lock(me->mutex);
1915         if (me->diverge) {
1916                 /* Just re-apply the root prog to fork the tree */
1917                 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
1918                                 me->prog, 0, 0);
1919         }
1920         sem_post(me->started);
1921         /* Return outside of started so parent notices failures. */
1922         if (ret) {
1923                 pthread_mutex_unlock(me->mutex);
1924                 return (void *)SIBLING_EXIT_FAILURE;
1925         }
1926         do {
1927                 pthread_cond_wait(me->cond, me->mutex);
1928                 me->num_waits = me->num_waits - 1;
1929         } while (me->num_waits);
1930         pthread_mutex_unlock(me->mutex);
1931
1932         ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0);
1933         if (!ret)
1934                 return (void *)SIBLING_EXIT_NEWPRIVS;
1935         read(0, NULL, 0);
1936         return (void *)SIBLING_EXIT_UNKILLED;
1937 }
1938
1939 void tsync_start_sibling(struct tsync_sibling *sibling)
1940 {
1941         pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling);
1942 }
1943
1944 TEST_F(TSYNC, siblings_fail_prctl)
1945 {
1946         long ret;
1947         void *status;
1948         struct sock_filter filter[] = {
1949                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1950                         offsetof(struct seccomp_data, nr)),
1951                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
1952                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL),
1953                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1954         };
1955         struct sock_fprog prog = {
1956                 .len = (unsigned short)ARRAY_SIZE(filter),
1957                 .filter = filter,
1958         };
1959
1960         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
1961                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1962         }
1963
1964         /* Check prctl failure detection by requesting sib 0 diverge. */
1965         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
1966         ASSERT_NE(ENOSYS, errno) {
1967                 TH_LOG("Kernel does not support seccomp syscall!");
1968         }
1969         ASSERT_EQ(0, ret) {
1970                 TH_LOG("setting filter failed");
1971         }
1972
1973         self->sibling[0].diverge = 1;
1974         tsync_start_sibling(&self->sibling[0]);
1975         tsync_start_sibling(&self->sibling[1]);
1976
1977         while (self->sibling_count < TSYNC_SIBLINGS) {
1978                 sem_wait(&self->started);
1979                 self->sibling_count++;
1980         }
1981
1982         /* Signal the threads to clean up*/
1983         pthread_mutex_lock(&self->mutex);
1984         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
1985                 TH_LOG("cond broadcast non-zero");
1986         }
1987         pthread_mutex_unlock(&self->mutex);
1988
1989         /* Ensure diverging sibling failed to call prctl. */
1990         pthread_join(self->sibling[0].tid, &status);
1991         EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status);
1992         pthread_join(self->sibling[1].tid, &status);
1993         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
1994 }
1995
1996 TEST_F(TSYNC, two_siblings_with_ancestor)
1997 {
1998         long ret;
1999         void *status;
2000
2001         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2002                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2003         }
2004
2005         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2006         ASSERT_NE(ENOSYS, errno) {
2007                 TH_LOG("Kernel does not support seccomp syscall!");
2008         }
2009         ASSERT_EQ(0, ret) {
2010                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2011         }
2012         tsync_start_sibling(&self->sibling[0]);
2013         tsync_start_sibling(&self->sibling[1]);
2014
2015         while (self->sibling_count < TSYNC_SIBLINGS) {
2016                 sem_wait(&self->started);
2017                 self->sibling_count++;
2018         }
2019
2020         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2021                       &self->apply_prog);
2022         ASSERT_EQ(0, ret) {
2023                 TH_LOG("Could install filter on all threads!");
2024         }
2025         /* Tell the siblings to test the policy */
2026         pthread_mutex_lock(&self->mutex);
2027         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2028                 TH_LOG("cond broadcast non-zero");
2029         }
2030         pthread_mutex_unlock(&self->mutex);
2031         /* Ensure they are both killed and don't exit cleanly. */
2032         pthread_join(self->sibling[0].tid, &status);
2033         EXPECT_EQ(0x0, (long)status);
2034         pthread_join(self->sibling[1].tid, &status);
2035         EXPECT_EQ(0x0, (long)status);
2036 }
2037
2038 TEST_F(TSYNC, two_sibling_want_nnp)
2039 {
2040         void *status;
2041
2042         /* start siblings before any prctl() operations */
2043         tsync_start_sibling(&self->sibling[0]);
2044         tsync_start_sibling(&self->sibling[1]);
2045         while (self->sibling_count < TSYNC_SIBLINGS) {
2046                 sem_wait(&self->started);
2047                 self->sibling_count++;
2048         }
2049
2050         /* Tell the siblings to test no policy */
2051         pthread_mutex_lock(&self->mutex);
2052         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2053                 TH_LOG("cond broadcast non-zero");
2054         }
2055         pthread_mutex_unlock(&self->mutex);
2056
2057         /* Ensure they are both upset about lacking nnp. */
2058         pthread_join(self->sibling[0].tid, &status);
2059         EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2060         pthread_join(self->sibling[1].tid, &status);
2061         EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2062 }
2063
2064 TEST_F(TSYNC, two_siblings_with_no_filter)
2065 {
2066         long ret;
2067         void *status;
2068
2069         /* start siblings before any prctl() operations */
2070         tsync_start_sibling(&self->sibling[0]);
2071         tsync_start_sibling(&self->sibling[1]);
2072         while (self->sibling_count < TSYNC_SIBLINGS) {
2073                 sem_wait(&self->started);
2074                 self->sibling_count++;
2075         }
2076
2077         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2078                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2079         }
2080
2081         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2082                       &self->apply_prog);
2083         ASSERT_NE(ENOSYS, errno) {
2084                 TH_LOG("Kernel does not support seccomp syscall!");
2085         }
2086         ASSERT_EQ(0, ret) {
2087                 TH_LOG("Could install filter on all threads!");
2088         }
2089
2090         /* Tell the siblings to test the policy */
2091         pthread_mutex_lock(&self->mutex);
2092         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2093                 TH_LOG("cond broadcast non-zero");
2094         }
2095         pthread_mutex_unlock(&self->mutex);
2096
2097         /* Ensure they are both killed and don't exit cleanly. */
2098         pthread_join(self->sibling[0].tid, &status);
2099         EXPECT_EQ(0x0, (long)status);
2100         pthread_join(self->sibling[1].tid, &status);
2101         EXPECT_EQ(0x0, (long)status);
2102 }
2103
2104 TEST_F(TSYNC, two_siblings_with_one_divergence)
2105 {
2106         long ret;
2107         void *status;
2108
2109         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2110                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2111         }
2112
2113         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2114         ASSERT_NE(ENOSYS, errno) {
2115                 TH_LOG("Kernel does not support seccomp syscall!");
2116         }
2117         ASSERT_EQ(0, ret) {
2118                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2119         }
2120         self->sibling[0].diverge = 1;
2121         tsync_start_sibling(&self->sibling[0]);
2122         tsync_start_sibling(&self->sibling[1]);
2123
2124         while (self->sibling_count < TSYNC_SIBLINGS) {
2125                 sem_wait(&self->started);
2126                 self->sibling_count++;
2127         }
2128
2129         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2130                       &self->apply_prog);
2131         ASSERT_EQ(self->sibling[0].system_tid, ret) {
2132                 TH_LOG("Did not fail on diverged sibling.");
2133         }
2134
2135         /* Wake the threads */
2136         pthread_mutex_lock(&self->mutex);
2137         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2138                 TH_LOG("cond broadcast non-zero");
2139         }
2140         pthread_mutex_unlock(&self->mutex);
2141
2142         /* Ensure they are both unkilled. */
2143         pthread_join(self->sibling[0].tid, &status);
2144         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2145         pthread_join(self->sibling[1].tid, &status);
2146         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2147 }
2148
2149 TEST_F(TSYNC, two_siblings_not_under_filter)
2150 {
2151         long ret, sib;
2152         void *status;
2153
2154         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2155                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2156         }
2157
2158         /*
2159          * Sibling 0 will have its own seccomp policy
2160          * and Sibling 1 will not be under seccomp at
2161          * all. Sibling 1 will enter seccomp and 0
2162          * will cause failure.
2163          */
2164         self->sibling[0].diverge = 1;
2165         tsync_start_sibling(&self->sibling[0]);
2166         tsync_start_sibling(&self->sibling[1]);
2167
2168         while (self->sibling_count < TSYNC_SIBLINGS) {
2169                 sem_wait(&self->started);
2170                 self->sibling_count++;
2171         }
2172
2173         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2174         ASSERT_NE(ENOSYS, errno) {
2175                 TH_LOG("Kernel does not support seccomp syscall!");
2176         }
2177         ASSERT_EQ(0, ret) {
2178                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2179         }
2180
2181         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2182                       &self->apply_prog);
2183         ASSERT_EQ(ret, self->sibling[0].system_tid) {
2184                 TH_LOG("Did not fail on diverged sibling.");
2185         }
2186         sib = 1;
2187         if (ret == self->sibling[0].system_tid)
2188                 sib = 0;
2189
2190         pthread_mutex_lock(&self->mutex);
2191
2192         /* Increment the other siblings num_waits so we can clean up
2193          * the one we just saw.
2194          */
2195         self->sibling[!sib].num_waits += 1;
2196
2197         /* Signal the thread to clean up*/
2198         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2199                 TH_LOG("cond broadcast non-zero");
2200         }
2201         pthread_mutex_unlock(&self->mutex);
2202         pthread_join(self->sibling[sib].tid, &status);
2203         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2204         /* Poll for actual task death. pthread_join doesn't guarantee it. */
2205         while (!kill(self->sibling[sib].system_tid, 0))
2206                 sleep(0.1);
2207         /* Switch to the remaining sibling */
2208         sib = !sib;
2209
2210         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2211                       &self->apply_prog);
2212         ASSERT_EQ(0, ret) {
2213                 TH_LOG("Expected the remaining sibling to sync");
2214         };
2215
2216         pthread_mutex_lock(&self->mutex);
2217
2218         /* If remaining sibling didn't have a chance to wake up during
2219          * the first broadcast, manually reduce the num_waits now.
2220          */
2221         if (self->sibling[sib].num_waits > 1)
2222                 self->sibling[sib].num_waits = 1;
2223         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2224                 TH_LOG("cond broadcast non-zero");
2225         }
2226         pthread_mutex_unlock(&self->mutex);
2227         pthread_join(self->sibling[sib].tid, &status);
2228         EXPECT_EQ(0, (long)status);
2229         /* Poll for actual task death. pthread_join doesn't guarantee it. */
2230         while (!kill(self->sibling[sib].system_tid, 0))
2231                 sleep(0.1);
2232
2233         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2234                       &self->apply_prog);
2235         ASSERT_EQ(0, ret);  /* just us chickens */
2236 }
2237
2238 /* Make sure restarted syscalls are seen directly as "restart_syscall". */
2239 TEST(syscall_restart)
2240 {
2241         long ret;
2242         unsigned long msg;
2243         pid_t child_pid;
2244         int pipefd[2];
2245         int status;
2246         siginfo_t info = { };
2247         struct sock_filter filter[] = {
2248                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2249                          offsetof(struct seccomp_data, nr)),
2250
2251 #ifdef __NR_sigreturn
2252                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 6, 0),
2253 #endif
2254                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 5, 0),
2255                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 4, 0),
2256                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 3, 0),
2257                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 4, 0),
2258                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0),
2259
2260                 /* Allow __NR_write for easy logging. */
2261                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1),
2262                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2263                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2264                 /* The nanosleep jump target. */
2265                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100),
2266                 /* The restart_syscall jump target. */
2267                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200),
2268         };
2269         struct sock_fprog prog = {
2270                 .len = (unsigned short)ARRAY_SIZE(filter),
2271                 .filter = filter,
2272         };
2273 #if defined(__arm__)
2274         struct utsname utsbuf;
2275 #endif
2276
2277         ASSERT_EQ(0, pipe(pipefd));
2278
2279         child_pid = fork();
2280         ASSERT_LE(0, child_pid);
2281         if (child_pid == 0) {
2282                 /* Child uses EXPECT not ASSERT to deliver status correctly. */
2283                 char buf = ' ';
2284                 struct timespec timeout = { };
2285
2286                 /* Attach parent as tracer and stop. */
2287                 EXPECT_EQ(0, ptrace(PTRACE_TRACEME));
2288                 EXPECT_EQ(0, raise(SIGSTOP));
2289
2290                 EXPECT_EQ(0, close(pipefd[1]));
2291
2292                 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2293                         TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2294                 }
2295
2296                 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2297                 EXPECT_EQ(0, ret) {
2298                         TH_LOG("Failed to install filter!");
2299                 }
2300
2301                 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2302                         TH_LOG("Failed to read() sync from parent");
2303                 }
2304                 EXPECT_EQ('.', buf) {
2305                         TH_LOG("Failed to get sync data from read()");
2306                 }
2307
2308                 /* Start nanosleep to be interrupted. */
2309                 timeout.tv_sec = 1;
2310                 errno = 0;
2311                 EXPECT_EQ(0, nanosleep(&timeout, NULL)) {
2312                         TH_LOG("Call to nanosleep() failed (errno %d)", errno);
2313                 }
2314
2315                 /* Read final sync from parent. */
2316                 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2317                         TH_LOG("Failed final read() from parent");
2318                 }
2319                 EXPECT_EQ('!', buf) {
2320                         TH_LOG("Failed to get final data from read()");
2321                 }
2322
2323                 /* Directly report the status of our test harness results. */
2324                 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS
2325                                                      : EXIT_FAILURE);
2326         }
2327         EXPECT_EQ(0, close(pipefd[0]));
2328
2329         /* Attach to child, setup options, and release. */
2330         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2331         ASSERT_EQ(true, WIFSTOPPED(status));
2332         ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL,
2333                             PTRACE_O_TRACESECCOMP));
2334         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2335         ASSERT_EQ(1, write(pipefd[1], ".", 1));
2336
2337         /* Wait for nanosleep() to start. */
2338         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2339         ASSERT_EQ(true, WIFSTOPPED(status));
2340         ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2341         ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2342         ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2343         ASSERT_EQ(0x100, msg);
2344         EXPECT_EQ(__NR_nanosleep, get_syscall(_metadata, child_pid));
2345
2346         /* Might as well check siginfo for sanity while we're here. */
2347         ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2348         ASSERT_EQ(SIGTRAP, info.si_signo);
2349         ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code);
2350         EXPECT_EQ(0, info.si_errno);
2351         EXPECT_EQ(getuid(), info.si_uid);
2352         /* Verify signal delivery came from child (seccomp-triggered). */
2353         EXPECT_EQ(child_pid, info.si_pid);
2354
2355         /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */
2356         ASSERT_EQ(0, kill(child_pid, SIGSTOP));
2357         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2358         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2359         ASSERT_EQ(true, WIFSTOPPED(status));
2360         ASSERT_EQ(SIGSTOP, WSTOPSIG(status));
2361         /* Verify signal delivery came from parent now. */
2362         ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2363         EXPECT_EQ(getpid(), info.si_pid);
2364
2365         /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */
2366         ASSERT_EQ(0, kill(child_pid, SIGCONT));
2367         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2368         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2369         ASSERT_EQ(true, WIFSTOPPED(status));
2370         ASSERT_EQ(SIGCONT, WSTOPSIG(status));
2371         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2372
2373         /* Wait for restart_syscall() to start. */
2374         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2375         ASSERT_EQ(true, WIFSTOPPED(status));
2376         ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2377         ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2378         ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2379
2380         ASSERT_EQ(0x200, msg);
2381         ret = get_syscall(_metadata, child_pid);
2382 #if defined(__arm__)
2383         /*
2384          * FIXME:
2385          * - native ARM registers do NOT expose true syscall.
2386          * - compat ARM registers on ARM64 DO expose true syscall.
2387          */
2388         ASSERT_EQ(0, uname(&utsbuf));
2389         if (strncmp(utsbuf.machine, "arm", 3) == 0) {
2390                 EXPECT_EQ(__NR_nanosleep, ret);
2391         } else
2392 #endif
2393         {
2394                 EXPECT_EQ(__NR_restart_syscall, ret);
2395         }
2396
2397         /* Write again to end test. */
2398         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2399         ASSERT_EQ(1, write(pipefd[1], "!", 1));
2400         EXPECT_EQ(0, close(pipefd[1]));
2401
2402         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2403         if (WIFSIGNALED(status) || WEXITSTATUS(status))
2404                 _metadata->passed = 0;
2405 }
2406
2407 /*
2408  * TODO:
2409  * - add microbenchmarks
2410  * - expand NNP testing
2411  * - better arch-specific TRACE and TRAP handlers.
2412  * - endianness checking when appropriate
2413  * - 64-bit arg prodding
2414  * - arch value testing (x86 modes especially)
2415  * - ...
2416  */
2417
2418 TEST_HARNESS_MAIN