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
  10 /*
  11  * glibc 2.26 and later have SIGSYS in siginfo_t. Before that,
  12  * we need to use the kernel's siginfo.h file and trick glibc
  13  * into accepting it.
  14  */
  15 #if !__GLIBC_PREREQ(2, 26)
  16 # include <asm/siginfo.h>
  17 # define __have_siginfo_t 1
  18 # define __have_sigval_t 1
  19 # define __have_sigevent_t 1
  20 #endif
  21
  22 #include <errno.h>
  23 #include <linux/filter.h>
  24 #include <sys/prctl.h>
  25 #include <sys/ptrace.h>
  26 #include <sys/user.h>
  27 #include <linux/prctl.h>
  28 #include <linux/ptrace.h>
  29 #include <linux/seccomp.h>
  30 #include <pthread.h>
  31 #include <semaphore.h>
  32 #include <signal.h>
  33 #include <stddef.h>
  34 #include <stdbool.h>
  35 #include <string.h>
  36 #include <time.h>
  37 #include <linux/elf.h>
  38 #include <sys/uio.h>
  39 #include <sys/utsname.h>
  40 #include <sys/fcntl.h>
  41 #include <sys/mman.h>
  42 #include <sys/times.h>
  43
  44 #define _GNU_SOURCE
  45 #include <unistd.h>
  46 #include <sys/syscall.h>
  47
  48 #include "../kselftest_harness.h"
  49
  50 #ifndef PR_SET_PTRACER
  51 # define PR_SET_PTRACER 0x59616d61
  52 #endif
  53
  54 #ifndef PR_SET_NO_NEW_PRIVS
  55 #define PR_SET_NO_NEW_PRIVS 38
  56 #define PR_GET_NO_NEW_PRIVS 39
  57 #endif
  58
  59 #ifndef PR_SECCOMP_EXT
  60 #define PR_SECCOMP_EXT 43
  61 #endif
  62
  63 #ifndef SECCOMP_EXT_ACT
  64 #define SECCOMP_EXT_ACT 1
  65 #endif
  66
  67 #ifndef SECCOMP_EXT_ACT_TSYNC
  68 #define SECCOMP_EXT_ACT_TSYNC 1
  69 #endif
  70
  71 #ifndef SECCOMP_MODE_STRICT
  72 #define SECCOMP_MODE_STRICT 1
  73 #endif
  74
  75 #ifndef SECCOMP_MODE_FILTER
  76 #define SECCOMP_MODE_FILTER 2
  77 #endif
  78
  79 #ifndef SECCOMP_RET_ALLOW
  80 struct seccomp_data {
  81         int nr;
  82         __u32 arch;
  83         __u64 instruction_pointer;
  84         __u64 args[6];
  85 };
  86 #endif
  87
  88 #ifndef SECCOMP_RET_KILL_PROCESS
  89 #define SECCOMP_RET_KILL_PROCESS 0x80000000U /* kill the process */
  90 #define SECCOMP_RET_KILL_THREAD  0x00000000U /* kill the thread */
  91 #endif
  92 #ifndef SECCOMP_RET_KILL
  93 #define SECCOMP_RET_KILL         SECCOMP_RET_KILL_THREAD
  94 #define SECCOMP_RET_TRAP         0x00030000U /* disallow and force a SIGSYS */
  95 #define SECCOMP_RET_ERRNO        0x00050000U /* returns an errno */
  96 #define SECCOMP_RET_TRACE        0x7ff00000U /* pass to a tracer or disallow */
  97 #define SECCOMP_RET_ALLOW        0x7fff0000U /* allow */
  98 #endif
  99 #ifndef SECCOMP_RET_LOG
 100 #define SECCOMP_RET_LOG          0x7ffc0000U /* allow after logging */
 101 #endif
 102
 103 #ifndef __NR_seccomp
 104 # if defined(__i386__)
 105 #  define __NR_seccomp 354
 106 # elif defined(__x86_64__)
 107 #  define __NR_seccomp 317
 108 # elif defined(__arm__)
 109 #  define __NR_seccomp 383
 110 # elif defined(__aarch64__)
 111 #  define __NR_seccomp 277
 112 # elif defined(__hppa__)
 113 #  define __NR_seccomp 338
 114 # elif defined(__powerpc__)
 115 #  define __NR_seccomp 358
 116 # elif defined(__s390__)
 117 #  define __NR_seccomp 348
 118 # else
 119 #  warning "seccomp syscall number unknown for this architecture"
 120 #  define __NR_seccomp 0xffff
 121 # endif
 122 #endif
 123
 124 #ifndef SECCOMP_SET_MODE_STRICT
 125 #define SECCOMP_SET_MODE_STRICT 0
 126 #endif
 127
 128 #ifndef SECCOMP_SET_MODE_FILTER
 129 #define SECCOMP_SET_MODE_FILTER 1
 130 #endif
 131
 132 #ifndef SECCOMP_GET_ACTION_AVAIL
 133 #define SECCOMP_GET_ACTION_AVAIL 2
 134 #endif
 135
 136 #ifndef SECCOMP_FILTER_FLAG_TSYNC
 137 #define SECCOMP_FILTER_FLAG_TSYNC 1
 138 #endif
 139
 140 #ifndef SECCOMP_FILTER_FLAG_LOG
 141 #define SECCOMP_FILTER_FLAG_LOG 2
 142 #endif
 143
 144 #ifndef PTRACE_SECCOMP_GET_METADATA
 145 #define PTRACE_SECCOMP_GET_METADATA     0x420d
 146
 147 struct seccomp_metadata {
 148         __u64 filter_off;       /* Input: which filter */
 149         __u64 flags;             /* Output: filter's flags */
 150 };
 151 #endif
 152
 153 #ifndef seccomp
 154 int seccomp(unsigned int op, unsigned int flags, void *args)
 155 {
 156         errno = 0;
 157         return syscall(__NR_seccomp, op, flags, args);
 158 }
 159 #endif
 160
 161 #if __BYTE_ORDER == __LITTLE_ENDIAN
 162 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]))
 163 #elif __BYTE_ORDER == __BIG_ENDIAN
 164 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]) + sizeof(__u32))
 165 #else
 166 #error "wut? Unknown __BYTE_ORDER?!"
 167 #endif
 168
 169 #define SIBLING_EXIT_UNKILLED   0xbadbeef
 170 #define SIBLING_EXIT_FAILURE    0xbadface
 171 #define SIBLING_EXIT_NEWPRIVS   0xbadfeed
 172
 173 TEST(mode_strict_support)
 174 {
 175         long ret;
 176
 177         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
 178         ASSERT_EQ(0, ret) {
 179                 TH_LOG("Kernel does not support CONFIG_SECCOMP");
 180         }
 181         syscall(__NR_exit, 0);
 182 }
 183
 184 TEST_SIGNAL(mode_strict_cannot_call_prctl, SIGKILL)
 185 {
 186         long ret;
 187
 188         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, NULL, NULL);
 189         ASSERT_EQ(0, ret) {
 190                 TH_LOG("Kernel does not support CONFIG_SECCOMP");
 191         }
 192         syscall(__NR_prctl, PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
 193                 NULL, NULL, NULL);
 194         EXPECT_FALSE(true) {
 195                 TH_LOG("Unreachable!");
 196         }
 197 }
 198
 199 /* Note! This doesn't test no new privs behavior */
 200 TEST(no_new_privs_support)
 201 {
 202         long ret;
 203
 204         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 205         EXPECT_EQ(0, ret) {
 206                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
 207         }
 208 }
 209
 210 /* Tests kernel support by checking for a copy_from_user() fault on NULL. */
 211 TEST(mode_filter_support)
 212 {
 213         long ret;
 214
 215         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
 216         ASSERT_EQ(0, ret) {
 217                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
 218         }
 219         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, NULL, NULL, NULL);
 220         EXPECT_EQ(-1, ret);
 221         EXPECT_EQ(EFAULT, errno) {
 222                 TH_LOG("Kernel does not support CONFIG_SECCOMP_FILTER!");
 223         }
 224 }
 225
 226 TEST(mode_filter_without_nnp)
 227 {
 228         struct sock_filter filter[] = {
 229                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 230         };
 231         struct sock_fprog prog = {
 232                 .len = (unsigned short)ARRAY_SIZE(filter),
 233                 .filter = filter,
 234         };
 235         long ret;
 236
 237         ret = prctl(PR_GET_NO_NEW_PRIVS, 0, NULL, 0, 0);
 238         ASSERT_LE(0, ret) {
 239                 TH_LOG("Expected 0 or unsupported for NO_NEW_PRIVS");
 240         }
 241         errno = 0;
 242         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 243         /* Succeeds with CAP_SYS_ADMIN, fails without */
 244         /* TODO(wad) check caps not euid */
 245         if (geteuid()) {
 246                 EXPECT_EQ(-1, ret);
 247                 EXPECT_EQ(EACCES, errno);
 248         } else {
 249                 EXPECT_EQ(0, ret);
 250         }
 251 }
 252
 253 #define MAX_INSNS_PER_PATH 32768
 254
 255 TEST(filter_size_limits)
 256 {
 257         int i;
 258         int count = BPF_MAXINSNS + 1;
 259         struct sock_filter allow[] = {
 260                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 261         };
 262         struct sock_filter *filter;
 263         struct sock_fprog prog = { };
 264         long ret;
 265
 266         filter = calloc(count, sizeof(*filter));
 267         ASSERT_NE(NULL, filter);
 268
 269         for (i = 0; i < count; i++)
 270                 filter[i] = allow[0];
 271
 272         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 273         ASSERT_EQ(0, ret);
 274
 275         prog.filter = filter;
 276         prog.len = count;
 277
 278         /* Too many filter instructions in a single filter. */
 279         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 280         ASSERT_NE(0, ret) {
 281                 TH_LOG("Installing %d insn filter was allowed", prog.len);
 282         }
 283
 284         /* One less is okay, though. */
 285         prog.len -= 1;
 286         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 287         ASSERT_EQ(0, ret) {
 288                 TH_LOG("Installing %d insn filter wasn't allowed", prog.len);
 289         }
 290 }
 291
 292 TEST(filter_chain_limits)
 293 {
 294         int i;
 295         int count = BPF_MAXINSNS;
 296         struct sock_filter allow[] = {
 297                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 298         };
 299         struct sock_filter *filter;
 300         struct sock_fprog prog = { };
 301         long ret;
 302
 303         filter = calloc(count, sizeof(*filter));
 304         ASSERT_NE(NULL, filter);
 305
 306         for (i = 0; i < count; i++)
 307                 filter[i] = allow[0];
 308
 309         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 310         ASSERT_EQ(0, ret);
 311
 312         prog.filter = filter;
 313         prog.len = 1;
 314
 315         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 316         ASSERT_EQ(0, ret);
 317
 318         prog.len = count;
 319
 320         /* Too many total filter instructions. */
 321         for (i = 0; i < MAX_INSNS_PER_PATH; i++) {
 322                 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 323                 if (ret != 0)
 324                         break;
 325         }
 326         ASSERT_NE(0, ret) {
 327                 TH_LOG("Allowed %d %d-insn filters (total with penalties:%d)",
 328                        i, count, i * (count + 4));
 329         }
 330 }
 331
 332 TEST(mode_filter_cannot_move_to_strict)
 333 {
 334         struct sock_filter filter[] = {
 335                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 336         };
 337         struct sock_fprog prog = {
 338                 .len = (unsigned short)ARRAY_SIZE(filter),
 339                 .filter = filter,
 340         };
 341         long ret;
 342
 343         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 344         ASSERT_EQ(0, ret);
 345
 346         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 347         ASSERT_EQ(0, ret);
 348
 349         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, NULL, 0, 0);
 350         EXPECT_EQ(-1, ret);
 351         EXPECT_EQ(EINVAL, errno);
 352 }
 353
 354
 355 TEST(mode_filter_get_seccomp)
 356 {
 357         struct sock_filter filter[] = {
 358                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 359         };
 360         struct sock_fprog prog = {
 361                 .len = (unsigned short)ARRAY_SIZE(filter),
 362                 .filter = filter,
 363         };
 364         long ret;
 365
 366         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 367         ASSERT_EQ(0, ret);
 368
 369         ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
 370         EXPECT_EQ(0, ret);
 371
 372         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
 373         ASSERT_EQ(0, ret);
 374
 375         ret = prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
 376         EXPECT_EQ(2, ret);
 377 }
 378
 379
 380 TEST(ALLOW_all)
 381 {
 382         struct sock_filter filter[] = {
 383                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 384         };
 385         struct sock_fprog prog = {
 386                 .len = (unsigned short)ARRAY_SIZE(filter),
 387                 .filter = filter,
 388         };
 389         long ret;
 390
 391         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 392         ASSERT_EQ(0, ret);
 393
 394         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 395         ASSERT_EQ(0, ret);
 396 }
 397
 398 TEST(empty_prog)
 399 {
 400         struct sock_filter filter[] = {
 401         };
 402         struct sock_fprog prog = {
 403                 .len = (unsigned short)ARRAY_SIZE(filter),
 404                 .filter = filter,
 405         };
 406         long ret;
 407
 408         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 409         ASSERT_EQ(0, ret);
 410
 411         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 412         EXPECT_EQ(-1, ret);
 413         EXPECT_EQ(EINVAL, errno);
 414 }
 415
 416 TEST(log_all)
 417 {
 418         struct sock_filter filter[] = {
 419                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG),
 420         };
 421         struct sock_fprog prog = {
 422                 .len = (unsigned short)ARRAY_SIZE(filter),
 423                 .filter = filter,
 424         };
 425         long ret;
 426         pid_t parent = getppid();
 427
 428         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 429         ASSERT_EQ(0, ret);
 430
 431         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 432         ASSERT_EQ(0, ret);
 433
 434         /* getppid() should succeed and be logged (no check for logging) */
 435         EXPECT_EQ(parent, syscall(__NR_getppid));
 436 }
 437
 438 TEST_SIGNAL(unknown_ret_is_kill_inside, SIGSYS)
 439 {
 440         struct sock_filter filter[] = {
 441                 BPF_STMT(BPF_RET|BPF_K, 0x10000000U),
 442         };
 443         struct sock_fprog prog = {
 444                 .len = (unsigned short)ARRAY_SIZE(filter),
 445                 .filter = filter,
 446         };
 447         long ret;
 448
 449         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 450         ASSERT_EQ(0, ret);
 451
 452         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 453         ASSERT_EQ(0, ret);
 454         EXPECT_EQ(0, syscall(__NR_getpid)) {
 455                 TH_LOG("getpid() shouldn't ever return");
 456         }
 457 }
 458
 459 /* return code >= 0x80000000 is unused. */
 460 TEST_SIGNAL(unknown_ret_is_kill_above_allow, SIGSYS)
 461 {
 462         struct sock_filter filter[] = {
 463                 BPF_STMT(BPF_RET|BPF_K, 0x90000000U),
 464         };
 465         struct sock_fprog prog = {
 466                 .len = (unsigned short)ARRAY_SIZE(filter),
 467                 .filter = filter,
 468         };
 469         long ret;
 470
 471         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 472         ASSERT_EQ(0, ret);
 473
 474         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 475         ASSERT_EQ(0, ret);
 476         EXPECT_EQ(0, syscall(__NR_getpid)) {
 477                 TH_LOG("getpid() shouldn't ever return");
 478         }
 479 }
 480
 481 TEST_SIGNAL(KILL_all, SIGSYS)
 482 {
 483         struct sock_filter filter[] = {
 484                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
 485         };
 486         struct sock_fprog prog = {
 487                 .len = (unsigned short)ARRAY_SIZE(filter),
 488                 .filter = filter,
 489         };
 490         long ret;
 491
 492         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 493         ASSERT_EQ(0, ret);
 494
 495         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 496         ASSERT_EQ(0, ret);
 497 }
 498
 499 TEST_SIGNAL(KILL_one, SIGSYS)
 500 {
 501         struct sock_filter filter[] = {
 502                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 503                         offsetof(struct seccomp_data, nr)),
 504                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
 505                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
 506                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 507         };
 508         struct sock_fprog prog = {
 509                 .len = (unsigned short)ARRAY_SIZE(filter),
 510                 .filter = filter,
 511         };
 512         long ret;
 513         pid_t parent = getppid();
 514
 515         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 516         ASSERT_EQ(0, ret);
 517
 518         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 519         ASSERT_EQ(0, ret);
 520
 521         EXPECT_EQ(parent, syscall(__NR_getppid));
 522         /* getpid() should never return. */
 523         EXPECT_EQ(0, syscall(__NR_getpid));
 524 }
 525
 526 TEST_SIGNAL(KILL_one_arg_one, SIGSYS)
 527 {
 528         void *fatal_address;
 529         struct sock_filter filter[] = {
 530                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 531                         offsetof(struct seccomp_data, nr)),
 532                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_times, 1, 0),
 533                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 534                 /* Only both with lower 32-bit for now. */
 535                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(0)),
 536                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K,
 537                         (unsigned long)&fatal_address, 0, 1),
 538                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
 539                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 540         };
 541         struct sock_fprog prog = {
 542                 .len = (unsigned short)ARRAY_SIZE(filter),
 543                 .filter = filter,
 544         };
 545         long ret;
 546         pid_t parent = getppid();
 547         struct tms timebuf;
 548         clock_t clock = times(&timebuf);
 549
 550         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 551         ASSERT_EQ(0, ret);
 552
 553         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 554         ASSERT_EQ(0, ret);
 555
 556         EXPECT_EQ(parent, syscall(__NR_getppid));
 557         EXPECT_LE(clock, syscall(__NR_times, &timebuf));
 558         /* times() should never return. */
 559         EXPECT_EQ(0, syscall(__NR_times, &fatal_address));
 560 }
 561
 562 TEST_SIGNAL(KILL_one_arg_six, SIGSYS)
 563 {
 564 #ifndef __NR_mmap2
 565         int sysno = __NR_mmap;
 566 #else
 567         int sysno = __NR_mmap2;
 568 #endif
 569         struct sock_filter filter[] = {
 570                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 571                         offsetof(struct seccomp_data, nr)),
 572                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, sysno, 1, 0),
 573                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 574                 /* Only both with lower 32-bit for now. */
 575                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(5)),
 576                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x0C0FFEE, 0, 1),
 577                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
 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         int fd;
 587         void *map1, *map2;
 588         int page_size = sysconf(_SC_PAGESIZE);
 589
 590         ASSERT_LT(0, page_size);
 591
 592         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 593         ASSERT_EQ(0, ret);
 594
 595         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 596         ASSERT_EQ(0, ret);
 597
 598         fd = open("/dev/zero", O_RDONLY);
 599         ASSERT_NE(-1, fd);
 600
 601         EXPECT_EQ(parent, syscall(__NR_getppid));
 602         map1 = (void *)syscall(sysno,
 603                 NULL, page_size, PROT_READ, MAP_PRIVATE, fd, page_size);
 604         EXPECT_NE(MAP_FAILED, map1);
 605         /* mmap2() should never return. */
 606         map2 = (void *)syscall(sysno,
 607                  NULL, page_size, PROT_READ, MAP_PRIVATE, fd, 0x0C0FFEE);
 608         EXPECT_EQ(MAP_FAILED, map2);
 609
 610         /* The test failed, so clean up the resources. */
 611         munmap(map1, page_size);
 612         munmap(map2, page_size);
 613         close(fd);
 614 }
 615
 616 /* This is a thread task to die via seccomp filter violation. */
 617 void *kill_thread(void *data)
 618 {
 619         bool die = (bool)data;
 620
 621         if (die) {
 622                 prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
 623                 return (void *)SIBLING_EXIT_FAILURE;
 624         }
 625
 626         return (void *)SIBLING_EXIT_UNKILLED;
 627 }
 628
 629 /* Prepare a thread that will kill itself or both of us. */
 630 void kill_thread_or_group(struct __test_metadata *_metadata, bool kill_process)
 631 {
 632         pthread_t thread;
 633         void *status;
 634         /* Kill only when calling __NR_prctl. */
 635         struct sock_filter filter_thread[] = {
 636                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 637                         offsetof(struct seccomp_data, nr)),
 638                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
 639                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_THREAD),
 640                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 641         };
 642         struct sock_fprog prog_thread = {
 643                 .len = (unsigned short)ARRAY_SIZE(filter_thread),
 644                 .filter = filter_thread,
 645         };
 646         struct sock_filter filter_process[] = {
 647                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 648                         offsetof(struct seccomp_data, nr)),
 649                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
 650                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL_PROCESS),
 651                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 652         };
 653         struct sock_fprog prog_process = {
 654                 .len = (unsigned short)ARRAY_SIZE(filter_process),
 655                 .filter = filter_process,
 656         };
 657
 658         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
 659                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
 660         }
 661
 662         ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0,
 663                              kill_process ? &prog_process : &prog_thread));
 664
 665         /*
 666          * Add the KILL_THREAD rule again to make sure that the KILL_PROCESS
 667          * flag cannot be downgraded by a new filter.
 668          */
 669         ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog_thread));
 670
 671         /* Start a thread that will exit immediately. */
 672         ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)false));
 673         ASSERT_EQ(0, pthread_join(thread, &status));
 674         ASSERT_EQ(SIBLING_EXIT_UNKILLED, (unsigned long)status);
 675
 676         /* Start a thread that will die immediately. */
 677         ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)true));
 678         ASSERT_EQ(0, pthread_join(thread, &status));
 679         ASSERT_NE(SIBLING_EXIT_FAILURE, (unsigned long)status);
 680
 681         /*
 682          * If we get here, only the spawned thread died. Let the parent know
 683          * the whole process didn't die (i.e. this thread, the spawner,
 684          * stayed running).
 685          */
 686         exit(42);
 687 }
 688
 689 TEST(KILL_thread)
 690 {
 691         int status;
 692         pid_t child_pid;
 693
 694         child_pid = fork();
 695         ASSERT_LE(0, child_pid);
 696         if (child_pid == 0) {
 697                 kill_thread_or_group(_metadata, false);
 698                 _exit(38);
 699         }
 700
 701         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
 702
 703         /* If only the thread was killed, we'll see exit 42. */
 704         ASSERT_TRUE(WIFEXITED(status));
 705         ASSERT_EQ(42, WEXITSTATUS(status));
 706 }
 707
 708 TEST(KILL_process)
 709 {
 710         int status;
 711         pid_t child_pid;
 712
 713         child_pid = fork();
 714         ASSERT_LE(0, child_pid);
 715         if (child_pid == 0) {
 716                 kill_thread_or_group(_metadata, true);
 717                 _exit(38);
 718         }
 719
 720         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
 721
 722         /* If the entire process was killed, we'll see SIGSYS. */
 723         ASSERT_TRUE(WIFSIGNALED(status));
 724         ASSERT_EQ(SIGSYS, WTERMSIG(status));
 725 }
 726
 727 /* TODO(wad) add 64-bit versus 32-bit arg tests. */
 728 TEST(arg_out_of_range)
 729 {
 730         struct sock_filter filter[] = {
 731                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS, syscall_arg(6)),
 732                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 733         };
 734         struct sock_fprog prog = {
 735                 .len = (unsigned short)ARRAY_SIZE(filter),
 736                 .filter = filter,
 737         };
 738         long ret;
 739
 740         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 741         ASSERT_EQ(0, ret);
 742
 743         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
 744         EXPECT_EQ(-1, ret);
 745         EXPECT_EQ(EINVAL, errno);
 746 }
 747
 748 #define ERRNO_FILTER(name, errno)                                       \
 749         struct sock_filter _read_filter_##name[] = {                    \
 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_read, 0, 1),       \
 753                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | errno),     \
 754                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),             \
 755         };                                                              \
 756         struct sock_fprog prog_##name = {                               \
 757                 .len = (unsigned short)ARRAY_SIZE(_read_filter_##name), \
 758                 .filter = _read_filter_##name,                          \
 759         }
 760
 761 /* Make sure basic errno values are correctly passed through a filter. */
 762 TEST(ERRNO_valid)
 763 {
 764         ERRNO_FILTER(valid, E2BIG);
 765         long ret;
 766         pid_t parent = getppid();
 767
 768         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 769         ASSERT_EQ(0, ret);
 770
 771         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_valid);
 772         ASSERT_EQ(0, ret);
 773
 774         EXPECT_EQ(parent, syscall(__NR_getppid));
 775         EXPECT_EQ(-1, read(0, NULL, 0));
 776         EXPECT_EQ(E2BIG, errno);
 777 }
 778
 779 /* Make sure an errno of zero is correctly handled by the arch code. */
 780 TEST(ERRNO_zero)
 781 {
 782         ERRNO_FILTER(zero, 0);
 783         long ret;
 784         pid_t parent = getppid();
 785
 786         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 787         ASSERT_EQ(0, ret);
 788
 789         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_zero);
 790         ASSERT_EQ(0, ret);
 791
 792         EXPECT_EQ(parent, syscall(__NR_getppid));
 793         /* "errno" of 0 is ok. */
 794         EXPECT_EQ(0, read(0, NULL, 0));
 795 }
 796
 797 /*
 798  * The SECCOMP_RET_DATA mask is 16 bits wide, but errno is smaller.
 799  * This tests that the errno value gets capped correctly, fixed by
 800  * 580c57f10768 ("seccomp: cap SECCOMP_RET_ERRNO data to MAX_ERRNO").
 801  */
 802 TEST(ERRNO_capped)
 803 {
 804         ERRNO_FILTER(capped, 4096);
 805         long ret;
 806         pid_t parent = getppid();
 807
 808         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 809         ASSERT_EQ(0, ret);
 810
 811         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_capped);
 812         ASSERT_EQ(0, ret);
 813
 814         EXPECT_EQ(parent, syscall(__NR_getppid));
 815         EXPECT_EQ(-1, read(0, NULL, 0));
 816         EXPECT_EQ(4095, errno);
 817 }
 818
 819 /*
 820  * Filters are processed in reverse order: last applied is executed first.
 821  * Since only the SECCOMP_RET_ACTION mask is tested for return values, the
 822  * SECCOMP_RET_DATA mask results will follow the most recently applied
 823  * matching filter return (and not the lowest or highest value).
 824  */
 825 TEST(ERRNO_order)
 826 {
 827         ERRNO_FILTER(first,  11);
 828         ERRNO_FILTER(second, 13);
 829         ERRNO_FILTER(third,  12);
 830         long ret;
 831         pid_t parent = getppid();
 832
 833         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 834         ASSERT_EQ(0, ret);
 835
 836         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_first);
 837         ASSERT_EQ(0, ret);
 838
 839         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_second);
 840         ASSERT_EQ(0, ret);
 841
 842         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_third);
 843         ASSERT_EQ(0, ret);
 844
 845         EXPECT_EQ(parent, syscall(__NR_getppid));
 846         EXPECT_EQ(-1, read(0, NULL, 0));
 847         EXPECT_EQ(12, errno);
 848 }
 849
 850 FIXTURE_DATA(TRAP) {
 851         struct sock_fprog prog;
 852 };
 853
 854 FIXTURE_SETUP(TRAP)
 855 {
 856         struct sock_filter filter[] = {
 857                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 858                         offsetof(struct seccomp_data, nr)),
 859                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
 860                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
 861                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 862         };
 863
 864         memset(&self->prog, 0, sizeof(self->prog));
 865         self->prog.filter = malloc(sizeof(filter));
 866         ASSERT_NE(NULL, self->prog.filter);
 867         memcpy(self->prog.filter, filter, sizeof(filter));
 868         self->prog.len = (unsigned short)ARRAY_SIZE(filter);
 869 }
 870
 871 FIXTURE_TEARDOWN(TRAP)
 872 {
 873         if (self->prog.filter)
 874                 free(self->prog.filter);
 875 }
 876
 877 TEST_F_SIGNAL(TRAP, dfl, SIGSYS)
 878 {
 879         long ret;
 880
 881         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 882         ASSERT_EQ(0, ret);
 883
 884         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
 885         ASSERT_EQ(0, ret);
 886         syscall(__NR_getpid);
 887 }
 888
 889 /* Ensure that SIGSYS overrides SIG_IGN */
 890 TEST_F_SIGNAL(TRAP, ign, SIGSYS)
 891 {
 892         long ret;
 893
 894         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 895         ASSERT_EQ(0, ret);
 896
 897         signal(SIGSYS, SIG_IGN);
 898
 899         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
 900         ASSERT_EQ(0, ret);
 901         syscall(__NR_getpid);
 902 }
 903
 904 static siginfo_t TRAP_info;
 905 static volatile int TRAP_nr;
 906 static void TRAP_action(int nr, siginfo_t *info, void *void_context)
 907 {
 908         memcpy(&TRAP_info, info, sizeof(TRAP_info));
 909         TRAP_nr = nr;
 910 }
 911
 912 TEST_F(TRAP, handler)
 913 {
 914         int ret, test;
 915         struct sigaction act;
 916         sigset_t mask;
 917
 918         memset(&act, 0, sizeof(act));
 919         sigemptyset(&mask);
 920         sigaddset(&mask, SIGSYS);
 921
 922         act.sa_sigaction = &TRAP_action;
 923         act.sa_flags = SA_SIGINFO;
 924         ret = sigaction(SIGSYS, &act, NULL);
 925         ASSERT_EQ(0, ret) {
 926                 TH_LOG("sigaction failed");
 927         }
 928         ret = sigprocmask(SIG_UNBLOCK, &mask, NULL);
 929         ASSERT_EQ(0, ret) {
 930                 TH_LOG("sigprocmask failed");
 931         }
 932
 933         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 934         ASSERT_EQ(0, ret);
 935         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog);
 936         ASSERT_EQ(0, ret);
 937         TRAP_nr = 0;
 938         memset(&TRAP_info, 0, sizeof(TRAP_info));
 939         /* Expect the registers to be rolled back. (nr = error) may vary
 940          * based on arch. */
 941         ret = syscall(__NR_getpid);
 942         /* Silence gcc warning about volatile. */
 943         test = TRAP_nr;
 944         EXPECT_EQ(SIGSYS, test);
 945         struct local_sigsys {
 946                 void *_call_addr;       /* calling user insn */
 947                 int _syscall;           /* triggering system call number */
 948                 unsigned int _arch;     /* AUDIT_ARCH_* of syscall */
 949         } *sigsys = (struct local_sigsys *)
 950 #ifdef si_syscall
 951                 &(TRAP_info.si_call_addr);
 952 #else
 953                 &TRAP_info.si_pid;
 954 #endif
 955         EXPECT_EQ(__NR_getpid, sigsys->_syscall);
 956         /* Make sure arch is non-zero. */
 957         EXPECT_NE(0, sigsys->_arch);
 958         EXPECT_NE(0, (unsigned long)sigsys->_call_addr);
 959 }
 960
 961 FIXTURE_DATA(precedence) {
 962         struct sock_fprog allow;
 963         struct sock_fprog log;
 964         struct sock_fprog trace;
 965         struct sock_fprog error;
 966         struct sock_fprog trap;
 967         struct sock_fprog kill;
 968 };
 969
 970 FIXTURE_SETUP(precedence)
 971 {
 972         struct sock_filter allow_insns[] = {
 973                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 974         };
 975         struct sock_filter log_insns[] = {
 976                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 977                         offsetof(struct seccomp_data, nr)),
 978                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
 979                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 980                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_LOG),
 981         };
 982         struct sock_filter trace_insns[] = {
 983                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 984                         offsetof(struct seccomp_data, nr)),
 985                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
 986                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 987                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE),
 988         };
 989         struct sock_filter error_insns[] = {
 990                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 991                         offsetof(struct seccomp_data, nr)),
 992                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
 993                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
 994                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO),
 995         };
 996         struct sock_filter trap_insns[] = {
 997                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
 998                         offsetof(struct seccomp_data, nr)),
 999                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1000                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1001                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRAP),
1002         };
1003         struct sock_filter kill_insns[] = {
1004                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1005                         offsetof(struct seccomp_data, nr)),
1006                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 1, 0),
1007                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1008                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1009         };
1010
1011         memset(self, 0, sizeof(*self));
1012 #define FILTER_ALLOC(_x) \
1013         self->_x.filter = malloc(sizeof(_x##_insns)); \
1014         ASSERT_NE(NULL, self->_x.filter); \
1015         memcpy(self->_x.filter, &_x##_insns, sizeof(_x##_insns)); \
1016         self->_x.len = (unsigned short)ARRAY_SIZE(_x##_insns)
1017         FILTER_ALLOC(allow);
1018         FILTER_ALLOC(log);
1019         FILTER_ALLOC(trace);
1020         FILTER_ALLOC(error);
1021         FILTER_ALLOC(trap);
1022         FILTER_ALLOC(kill);
1023 }
1024
1025 FIXTURE_TEARDOWN(precedence)
1026 {
1027 #define FILTER_FREE(_x) if (self->_x.filter) free(self->_x.filter)
1028         FILTER_FREE(allow);
1029         FILTER_FREE(log);
1030         FILTER_FREE(trace);
1031         FILTER_FREE(error);
1032         FILTER_FREE(trap);
1033         FILTER_FREE(kill);
1034 }
1035
1036 TEST_F(precedence, allow_ok)
1037 {
1038         pid_t parent, res = 0;
1039         long ret;
1040
1041         parent = getppid();
1042         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1043         ASSERT_EQ(0, ret);
1044
1045         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1046         ASSERT_EQ(0, ret);
1047         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1048         ASSERT_EQ(0, ret);
1049         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1050         ASSERT_EQ(0, ret);
1051         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1052         ASSERT_EQ(0, ret);
1053         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1054         ASSERT_EQ(0, ret);
1055         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1056         ASSERT_EQ(0, ret);
1057         /* Should work just fine. */
1058         res = syscall(__NR_getppid);
1059         EXPECT_EQ(parent, res);
1060 }
1061
1062 TEST_F_SIGNAL(precedence, kill_is_highest, SIGSYS)
1063 {
1064         pid_t parent, res = 0;
1065         long ret;
1066
1067         parent = getppid();
1068         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1069         ASSERT_EQ(0, ret);
1070
1071         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1072         ASSERT_EQ(0, ret);
1073         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1074         ASSERT_EQ(0, ret);
1075         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1076         ASSERT_EQ(0, ret);
1077         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1078         ASSERT_EQ(0, ret);
1079         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1080         ASSERT_EQ(0, ret);
1081         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1082         ASSERT_EQ(0, ret);
1083         /* Should work just fine. */
1084         res = syscall(__NR_getppid);
1085         EXPECT_EQ(parent, res);
1086         /* getpid() should never return. */
1087         res = syscall(__NR_getpid);
1088         EXPECT_EQ(0, res);
1089 }
1090
1091 TEST_F_SIGNAL(precedence, kill_is_highest_in_any_order, SIGSYS)
1092 {
1093         pid_t parent;
1094         long ret;
1095
1096         parent = getppid();
1097         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1098         ASSERT_EQ(0, ret);
1099
1100         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1101         ASSERT_EQ(0, ret);
1102         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->kill);
1103         ASSERT_EQ(0, ret);
1104         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1105         ASSERT_EQ(0, ret);
1106         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1107         ASSERT_EQ(0, ret);
1108         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1109         ASSERT_EQ(0, ret);
1110         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1111         ASSERT_EQ(0, ret);
1112         /* Should work just fine. */
1113         EXPECT_EQ(parent, syscall(__NR_getppid));
1114         /* getpid() should never return. */
1115         EXPECT_EQ(0, syscall(__NR_getpid));
1116 }
1117
1118 TEST_F_SIGNAL(precedence, trap_is_second, SIGSYS)
1119 {
1120         pid_t parent;
1121         long ret;
1122
1123         parent = getppid();
1124         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1125         ASSERT_EQ(0, ret);
1126
1127         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1128         ASSERT_EQ(0, ret);
1129         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1130         ASSERT_EQ(0, ret);
1131         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1132         ASSERT_EQ(0, ret);
1133         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1134         ASSERT_EQ(0, ret);
1135         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1136         ASSERT_EQ(0, ret);
1137         /* Should work just fine. */
1138         EXPECT_EQ(parent, syscall(__NR_getppid));
1139         /* getpid() should never return. */
1140         EXPECT_EQ(0, syscall(__NR_getpid));
1141 }
1142
1143 TEST_F_SIGNAL(precedence, trap_is_second_in_any_order, SIGSYS)
1144 {
1145         pid_t parent;
1146         long ret;
1147
1148         parent = getppid();
1149         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1150         ASSERT_EQ(0, ret);
1151
1152         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1153         ASSERT_EQ(0, ret);
1154         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trap);
1155         ASSERT_EQ(0, ret);
1156         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1157         ASSERT_EQ(0, ret);
1158         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1159         ASSERT_EQ(0, ret);
1160         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1161         ASSERT_EQ(0, ret);
1162         /* Should work just fine. */
1163         EXPECT_EQ(parent, syscall(__NR_getppid));
1164         /* getpid() should never return. */
1165         EXPECT_EQ(0, syscall(__NR_getpid));
1166 }
1167
1168 TEST_F(precedence, errno_is_third)
1169 {
1170         pid_t parent;
1171         long ret;
1172
1173         parent = getppid();
1174         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1175         ASSERT_EQ(0, ret);
1176
1177         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1178         ASSERT_EQ(0, ret);
1179         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1180         ASSERT_EQ(0, ret);
1181         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1182         ASSERT_EQ(0, ret);
1183         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1184         ASSERT_EQ(0, ret);
1185         /* Should work just fine. */
1186         EXPECT_EQ(parent, syscall(__NR_getppid));
1187         EXPECT_EQ(0, syscall(__NR_getpid));
1188 }
1189
1190 TEST_F(precedence, errno_is_third_in_any_order)
1191 {
1192         pid_t parent;
1193         long ret;
1194
1195         parent = getppid();
1196         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1197         ASSERT_EQ(0, ret);
1198
1199         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1200         ASSERT_EQ(0, ret);
1201         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->error);
1202         ASSERT_EQ(0, ret);
1203         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1204         ASSERT_EQ(0, ret);
1205         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1206         ASSERT_EQ(0, ret);
1207         /* Should work just fine. */
1208         EXPECT_EQ(parent, syscall(__NR_getppid));
1209         EXPECT_EQ(0, syscall(__NR_getpid));
1210 }
1211
1212 TEST_F(precedence, trace_is_fourth)
1213 {
1214         pid_t parent;
1215         long ret;
1216
1217         parent = getppid();
1218         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1219         ASSERT_EQ(0, ret);
1220
1221         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1222         ASSERT_EQ(0, ret);
1223         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1224         ASSERT_EQ(0, ret);
1225         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1226         ASSERT_EQ(0, ret);
1227         /* Should work just fine. */
1228         EXPECT_EQ(parent, syscall(__NR_getppid));
1229         /* No ptracer */
1230         EXPECT_EQ(-1, syscall(__NR_getpid));
1231 }
1232
1233 TEST_F(precedence, trace_is_fourth_in_any_order)
1234 {
1235         pid_t parent;
1236         long ret;
1237
1238         parent = getppid();
1239         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1240         ASSERT_EQ(0, ret);
1241
1242         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->trace);
1243         ASSERT_EQ(0, ret);
1244         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1245         ASSERT_EQ(0, ret);
1246         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1247         ASSERT_EQ(0, ret);
1248         /* Should work just fine. */
1249         EXPECT_EQ(parent, syscall(__NR_getppid));
1250         /* No ptracer */
1251         EXPECT_EQ(-1, syscall(__NR_getpid));
1252 }
1253
1254 TEST_F(precedence, log_is_fifth)
1255 {
1256         pid_t mypid, parent;
1257         long ret;
1258
1259         mypid = getpid();
1260         parent = getppid();
1261         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1262         ASSERT_EQ(0, ret);
1263
1264         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1265         ASSERT_EQ(0, ret);
1266         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1267         ASSERT_EQ(0, ret);
1268         /* Should work just fine. */
1269         EXPECT_EQ(parent, syscall(__NR_getppid));
1270         /* Should also work just fine */
1271         EXPECT_EQ(mypid, syscall(__NR_getpid));
1272 }
1273
1274 TEST_F(precedence, log_is_fifth_in_any_order)
1275 {
1276         pid_t mypid, parent;
1277         long ret;
1278
1279         mypid = getpid();
1280         parent = getppid();
1281         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1282         ASSERT_EQ(0, ret);
1283
1284         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->log);
1285         ASSERT_EQ(0, ret);
1286         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->allow);
1287         ASSERT_EQ(0, ret);
1288         /* Should work just fine. */
1289         EXPECT_EQ(parent, syscall(__NR_getppid));
1290         /* Should also work just fine */
1291         EXPECT_EQ(mypid, syscall(__NR_getpid));
1292 }
1293
1294 #ifndef PTRACE_O_TRACESECCOMP
1295 #define PTRACE_O_TRACESECCOMP   0x00000080
1296 #endif
1297
1298 /* Catch the Ubuntu 12.04 value error. */
1299 #if PTRACE_EVENT_SECCOMP != 7
1300 #undef PTRACE_EVENT_SECCOMP
1301 #endif
1302
1303 #ifndef PTRACE_EVENT_SECCOMP
1304 #define PTRACE_EVENT_SECCOMP 7
1305 #endif
1306
1307 #define IS_SECCOMP_EVENT(status) ((status >> 16) == PTRACE_EVENT_SECCOMP)
1308 bool tracer_running;
1309 void tracer_stop(int sig)
1310 {
1311         tracer_running = false;
1312 }
1313
1314 typedef void tracer_func_t(struct __test_metadata *_metadata,
1315                            pid_t tracee, int status, void *args);
1316
1317 void start_tracer(struct __test_metadata *_metadata, int fd, pid_t tracee,
1318             tracer_func_t tracer_func, void *args, bool ptrace_syscall)
1319 {
1320         int ret = -1;
1321         struct sigaction action = {
1322                 .sa_handler = tracer_stop,
1323         };
1324
1325         /* Allow external shutdown. */
1326         tracer_running = true;
1327         ASSERT_EQ(0, sigaction(SIGUSR1, &action, NULL));
1328
1329         errno = 0;
1330         while (ret == -1 && errno != EINVAL)
1331                 ret = ptrace(PTRACE_ATTACH, tracee, NULL, 0);
1332         ASSERT_EQ(0, ret) {
1333                 kill(tracee, SIGKILL);
1334         }
1335         /* Wait for attach stop */
1336         wait(NULL);
1337
1338         ret = ptrace(PTRACE_SETOPTIONS, tracee, NULL, ptrace_syscall ?
1339                                                       PTRACE_O_TRACESYSGOOD :
1340                                                       PTRACE_O_TRACESECCOMP);
1341         ASSERT_EQ(0, ret) {
1342                 TH_LOG("Failed to set PTRACE_O_TRACESECCOMP");
1343                 kill(tracee, SIGKILL);
1344         }
1345         ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1346                      tracee, NULL, 0);
1347         ASSERT_EQ(0, ret);
1348
1349         /* Unblock the tracee */
1350         ASSERT_EQ(1, write(fd, "A", 1));
1351         ASSERT_EQ(0, close(fd));
1352
1353         /* Run until we're shut down. Must assert to stop execution. */
1354         while (tracer_running) {
1355                 int status;
1356
1357                 if (wait(&status) != tracee)
1358                         continue;
1359                 if (WIFSIGNALED(status) || WIFEXITED(status))
1360                         /* Child is dead. Time to go. */
1361                         return;
1362
1363                 /* Check if this is a seccomp event. */
1364                 ASSERT_EQ(!ptrace_syscall, IS_SECCOMP_EVENT(status));
1365
1366                 tracer_func(_metadata, tracee, status, args);
1367
1368                 ret = ptrace(ptrace_syscall ? PTRACE_SYSCALL : PTRACE_CONT,
1369                              tracee, NULL, 0);
1370                 ASSERT_EQ(0, ret);
1371         }
1372         /* Directly report the status of our test harness results. */
1373         syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS : EXIT_FAILURE);
1374 }
1375
1376 /* Common tracer setup/teardown functions. */
1377 void cont_handler(int num)
1378 { }
1379 pid_t setup_trace_fixture(struct __test_metadata *_metadata,
1380                           tracer_func_t func, void *args, bool ptrace_syscall)
1381 {
1382         char sync;
1383         int pipefd[2];
1384         pid_t tracer_pid;
1385         pid_t tracee = getpid();
1386
1387         /* Setup a pipe for clean synchronization. */
1388         ASSERT_EQ(0, pipe(pipefd));
1389
1390         /* Fork a child which we'll promote to tracer */
1391         tracer_pid = fork();
1392         ASSERT_LE(0, tracer_pid);
1393         signal(SIGALRM, cont_handler);
1394         if (tracer_pid == 0) {
1395                 close(pipefd[0]);
1396                 start_tracer(_metadata, pipefd[1], tracee, func, args,
1397                              ptrace_syscall);
1398                 syscall(__NR_exit, 0);
1399         }
1400         close(pipefd[1]);
1401         prctl(PR_SET_PTRACER, tracer_pid, 0, 0, 0);
1402         read(pipefd[0], &sync, 1);
1403         close(pipefd[0]);
1404
1405         return tracer_pid;
1406 }
1407 void teardown_trace_fixture(struct __test_metadata *_metadata,
1408                             pid_t tracer)
1409 {
1410         if (tracer) {
1411                 int status;
1412                 /*
1413                  * Extract the exit code from the other process and
1414                  * adopt it for ourselves in case its asserts failed.
1415                  */
1416                 ASSERT_EQ(0, kill(tracer, SIGUSR1));
1417                 ASSERT_EQ(tracer, waitpid(tracer, &status, 0));
1418                 if (WEXITSTATUS(status))
1419                         _metadata->passed = 0;
1420         }
1421 }
1422
1423 /* "poke" tracer arguments and function. */
1424 struct tracer_args_poke_t {
1425         unsigned long poke_addr;
1426 };
1427
1428 void tracer_poke(struct __test_metadata *_metadata, pid_t tracee, int status,
1429                  void *args)
1430 {
1431         int ret;
1432         unsigned long msg;
1433         struct tracer_args_poke_t *info = (struct tracer_args_poke_t *)args;
1434
1435         ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1436         EXPECT_EQ(0, ret);
1437         /* If this fails, don't try to recover. */
1438         ASSERT_EQ(0x1001, msg) {
1439                 kill(tracee, SIGKILL);
1440         }
1441         /*
1442          * Poke in the message.
1443          * Registers are not touched to try to keep this relatively arch
1444          * agnostic.
1445          */
1446         ret = ptrace(PTRACE_POKEDATA, tracee, info->poke_addr, 0x1001);
1447         EXPECT_EQ(0, ret);
1448 }
1449
1450 FIXTURE_DATA(TRACE_poke) {
1451         struct sock_fprog prog;
1452         pid_t tracer;
1453         long poked;
1454         struct tracer_args_poke_t tracer_args;
1455 };
1456
1457 FIXTURE_SETUP(TRACE_poke)
1458 {
1459         struct sock_filter filter[] = {
1460                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1461                         offsetof(struct seccomp_data, nr)),
1462                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
1463                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1001),
1464                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1465         };
1466
1467         self->poked = 0;
1468         memset(&self->prog, 0, sizeof(self->prog));
1469         self->prog.filter = malloc(sizeof(filter));
1470         ASSERT_NE(NULL, self->prog.filter);
1471         memcpy(self->prog.filter, filter, sizeof(filter));
1472         self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1473
1474         /* Set up tracer args. */
1475         self->tracer_args.poke_addr = (unsigned long)&self->poked;
1476
1477         /* Launch tracer. */
1478         self->tracer = setup_trace_fixture(_metadata, tracer_poke,
1479                                            &self->tracer_args, false);
1480 }
1481
1482 FIXTURE_TEARDOWN(TRACE_poke)
1483 {
1484         teardown_trace_fixture(_metadata, self->tracer);
1485         if (self->prog.filter)
1486                 free(self->prog.filter);
1487 }
1488
1489 TEST_F(TRACE_poke, read_has_side_effects)
1490 {
1491         ssize_t ret;
1492
1493         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1494         ASSERT_EQ(0, ret);
1495
1496         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1497         ASSERT_EQ(0, ret);
1498
1499         EXPECT_EQ(0, self->poked);
1500         ret = read(-1, NULL, 0);
1501         EXPECT_EQ(-1, ret);
1502         EXPECT_EQ(0x1001, self->poked);
1503 }
1504
1505 TEST_F(TRACE_poke, getpid_runs_normally)
1506 {
1507         long ret;
1508
1509         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1510         ASSERT_EQ(0, ret);
1511
1512         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1513         ASSERT_EQ(0, ret);
1514
1515         EXPECT_EQ(0, self->poked);
1516         EXPECT_NE(0, syscall(__NR_getpid));
1517         EXPECT_EQ(0, self->poked);
1518 }
1519
1520 #if defined(__x86_64__)
1521 # define ARCH_REGS      struct user_regs_struct
1522 # define SYSCALL_NUM    orig_rax
1523 # define SYSCALL_RET    rax
1524 #elif defined(__i386__)
1525 # define ARCH_REGS      struct user_regs_struct
1526 # define SYSCALL_NUM    orig_eax
1527 # define SYSCALL_RET    eax
1528 #elif defined(__arm__)
1529 # define ARCH_REGS      struct pt_regs
1530 # define SYSCALL_NUM    ARM_r7
1531 # define SYSCALL_RET    ARM_r0
1532 #elif defined(__aarch64__)
1533 # define ARCH_REGS      struct user_pt_regs
1534 # define SYSCALL_NUM    regs[8]
1535 # define SYSCALL_RET    regs[0]
1536 #elif defined(__hppa__)
1537 # define ARCH_REGS      struct user_regs_struct
1538 # define SYSCALL_NUM    gr[20]
1539 # define SYSCALL_RET    gr[28]
1540 #elif defined(__powerpc__)
1541 # define ARCH_REGS      struct pt_regs
1542 # define SYSCALL_NUM    gpr[0]
1543 # define SYSCALL_RET    gpr[3]
1544 #elif defined(__s390__)
1545 # define ARCH_REGS     s390_regs
1546 # define SYSCALL_NUM   gprs[2]
1547 # define SYSCALL_RET   gprs[2]
1548 #elif defined(__mips__)
1549 # define ARCH_REGS      struct pt_regs
1550 # define SYSCALL_NUM    regs[2]
1551 # define SYSCALL_SYSCALL_NUM regs[4]
1552 # define SYSCALL_RET    regs[2]
1553 # define SYSCALL_NUM_RET_SHARE_REG
1554 #else
1555 # error "Do not know how to find your architecture's registers and syscalls"
1556 #endif
1557
1558 /* When the syscall return can't be changed, stub out the tests for it. */
1559 #ifdef SYSCALL_NUM_RET_SHARE_REG
1560 # define EXPECT_SYSCALL_RETURN(val, action)     EXPECT_EQ(-1, action)
1561 #else
1562 # define EXPECT_SYSCALL_RETURN(val, action)     EXPECT_EQ(val, action)
1563 #endif
1564
1565 /* Use PTRACE_GETREGS and PTRACE_SETREGS when available. This is useful for
1566  * architectures without HAVE_ARCH_TRACEHOOK (e.g. User-mode Linux).
1567  */
1568 #if defined(__x86_64__) || defined(__i386__) || defined(__mips__)
1569 #define HAVE_GETREGS
1570 #endif
1571
1572 /* Architecture-specific syscall fetching routine. */
1573 int get_syscall(struct __test_metadata *_metadata, pid_t tracee)
1574 {
1575         ARCH_REGS regs;
1576 #ifdef HAVE_GETREGS
1577         EXPECT_EQ(0, ptrace(PTRACE_GETREGS, tracee, 0, &regs)) {
1578                 TH_LOG("PTRACE_GETREGS failed");
1579                 return -1;
1580         }
1581 #else
1582         struct iovec iov;
1583
1584         iov.iov_base = &regs;
1585         iov.iov_len = sizeof(regs);
1586         EXPECT_EQ(0, ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov)) {
1587                 TH_LOG("PTRACE_GETREGSET failed");
1588                 return -1;
1589         }
1590 #endif
1591
1592 #if defined(__mips__)
1593         if (regs.SYSCALL_NUM == __NR_O32_Linux)
1594                 return regs.SYSCALL_SYSCALL_NUM;
1595 #endif
1596         return regs.SYSCALL_NUM;
1597 }
1598
1599 /* Architecture-specific syscall changing routine. */
1600 void change_syscall(struct __test_metadata *_metadata,
1601                     pid_t tracee, int syscall)
1602 {
1603         int ret;
1604         ARCH_REGS regs;
1605 #ifdef HAVE_GETREGS
1606         ret = ptrace(PTRACE_GETREGS, tracee, 0, &regs);
1607 #else
1608         struct iovec iov;
1609         iov.iov_base = &regs;
1610         iov.iov_len = sizeof(regs);
1611         ret = ptrace(PTRACE_GETREGSET, tracee, NT_PRSTATUS, &iov);
1612 #endif
1613         EXPECT_EQ(0, ret) {}
1614
1615 #if defined(__x86_64__) || defined(__i386__) || defined(__powerpc__) || \
1616     defined(__s390__) || defined(__hppa__)
1617         {
1618                 regs.SYSCALL_NUM = syscall;
1619         }
1620 #elif defined(__mips__)
1621         {
1622                 if (regs.SYSCALL_NUM == __NR_O32_Linux)
1623                         regs.SYSCALL_SYSCALL_NUM = syscall;
1624                 else
1625                         regs.SYSCALL_NUM = syscall;
1626         }
1627
1628 #elif defined(__arm__)
1629 # ifndef PTRACE_SET_SYSCALL
1630 #  define PTRACE_SET_SYSCALL   23
1631 # endif
1632         {
1633                 ret = ptrace(PTRACE_SET_SYSCALL, tracee, NULL, syscall);
1634                 EXPECT_EQ(0, ret);
1635         }
1636
1637 #elif defined(__aarch64__)
1638 # ifndef NT_ARM_SYSTEM_CALL
1639 #  define NT_ARM_SYSTEM_CALL 0x404
1640 # endif
1641         {
1642                 iov.iov_base = &syscall;
1643                 iov.iov_len = sizeof(syscall);
1644                 ret = ptrace(PTRACE_SETREGSET, tracee, NT_ARM_SYSTEM_CALL,
1645                              &iov);
1646                 EXPECT_EQ(0, ret);
1647         }
1648
1649 #else
1650         ASSERT_EQ(1, 0) {
1651                 TH_LOG("How is the syscall changed on this architecture?");
1652         }
1653 #endif
1654
1655         /* If syscall is skipped, change return value. */
1656         if (syscall == -1)
1657 #ifdef SYSCALL_NUM_RET_SHARE_REG
1658                 TH_LOG("Can't modify syscall return on this architecture");
1659 #else
1660                 regs.SYSCALL_RET = EPERM;
1661 #endif
1662
1663 #ifdef HAVE_GETREGS
1664         ret = ptrace(PTRACE_SETREGS, tracee, 0, &regs);
1665 #else
1666         iov.iov_base = &regs;
1667         iov.iov_len = sizeof(regs);
1668         ret = ptrace(PTRACE_SETREGSET, tracee, NT_PRSTATUS, &iov);
1669 #endif
1670         EXPECT_EQ(0, ret);
1671 }
1672
1673 void tracer_syscall(struct __test_metadata *_metadata, pid_t tracee,
1674                     int status, void *args)
1675 {
1676         int ret;
1677         unsigned long msg;
1678
1679         /* Make sure we got the right message. */
1680         ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1681         EXPECT_EQ(0, ret);
1682
1683         /* Validate and take action on expected syscalls. */
1684         switch (msg) {
1685         case 0x1002:
1686                 /* change getpid to getppid. */
1687                 EXPECT_EQ(__NR_getpid, get_syscall(_metadata, tracee));
1688                 change_syscall(_metadata, tracee, __NR_getppid);
1689                 break;
1690         case 0x1003:
1691                 /* skip gettid. */
1692                 EXPECT_EQ(__NR_gettid, get_syscall(_metadata, tracee));
1693                 change_syscall(_metadata, tracee, -1);
1694                 break;
1695         case 0x1004:
1696                 /* do nothing (allow getppid) */
1697                 EXPECT_EQ(__NR_getppid, get_syscall(_metadata, tracee));
1698                 break;
1699         default:
1700                 EXPECT_EQ(0, msg) {
1701                         TH_LOG("Unknown PTRACE_GETEVENTMSG: 0x%lx", msg);
1702                         kill(tracee, SIGKILL);
1703                 }
1704         }
1705
1706 }
1707
1708 void tracer_ptrace(struct __test_metadata *_metadata, pid_t tracee,
1709                    int status, void *args)
1710 {
1711         int ret, nr;
1712         unsigned long msg;
1713         static bool entry;
1714
1715         /* Make sure we got an empty message. */
1716         ret = ptrace(PTRACE_GETEVENTMSG, tracee, NULL, &msg);
1717         EXPECT_EQ(0, ret);
1718         EXPECT_EQ(0, msg);
1719
1720         /* The only way to tell PTRACE_SYSCALL entry/exit is by counting. */
1721         entry = !entry;
1722         if (!entry)
1723                 return;
1724
1725         nr = get_syscall(_metadata, tracee);
1726
1727         if (nr == __NR_getpid)
1728                 change_syscall(_metadata, tracee, __NR_getppid);
1729         if (nr == __NR_openat)
1730                 change_syscall(_metadata, tracee, -1);
1731 }
1732
1733 FIXTURE_DATA(TRACE_syscall) {
1734         struct sock_fprog prog;
1735         pid_t tracer, mytid, mypid, parent;
1736 };
1737
1738 FIXTURE_SETUP(TRACE_syscall)
1739 {
1740         struct sock_filter filter[] = {
1741                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1742                         offsetof(struct seccomp_data, nr)),
1743                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
1744                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1002),
1745                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_gettid, 0, 1),
1746                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1003),
1747                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1748                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE | 0x1004),
1749                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1750         };
1751
1752         memset(&self->prog, 0, sizeof(self->prog));
1753         self->prog.filter = malloc(sizeof(filter));
1754         ASSERT_NE(NULL, self->prog.filter);
1755         memcpy(self->prog.filter, filter, sizeof(filter));
1756         self->prog.len = (unsigned short)ARRAY_SIZE(filter);
1757
1758         /* Prepare some testable syscall results. */
1759         self->mytid = syscall(__NR_gettid);
1760         ASSERT_GT(self->mytid, 0);
1761         ASSERT_NE(self->mytid, 1) {
1762                 TH_LOG("Running this test as init is not supported. :)");
1763         }
1764
1765         self->mypid = getpid();
1766         ASSERT_GT(self->mypid, 0);
1767         ASSERT_EQ(self->mytid, self->mypid);
1768
1769         self->parent = getppid();
1770         ASSERT_GT(self->parent, 0);
1771         ASSERT_NE(self->parent, self->mypid);
1772
1773         /* Launch tracer. */
1774         self->tracer = setup_trace_fixture(_metadata, tracer_syscall, NULL,
1775                                            false);
1776 }
1777
1778 FIXTURE_TEARDOWN(TRACE_syscall)
1779 {
1780         teardown_trace_fixture(_metadata, self->tracer);
1781         if (self->prog.filter)
1782                 free(self->prog.filter);
1783 }
1784
1785 TEST_F(TRACE_syscall, ptrace_syscall_redirected)
1786 {
1787         /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1788         teardown_trace_fixture(_metadata, self->tracer);
1789         self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1790                                            true);
1791
1792         /* Tracer will redirect getpid to getppid. */
1793         EXPECT_NE(self->mypid, syscall(__NR_getpid));
1794 }
1795
1796 TEST_F(TRACE_syscall, ptrace_syscall_dropped)
1797 {
1798         /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1799         teardown_trace_fixture(_metadata, self->tracer);
1800         self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1801                                            true);
1802
1803         /* Tracer should skip the open syscall, resulting in EPERM. */
1804         EXPECT_SYSCALL_RETURN(EPERM, syscall(__NR_openat));
1805 }
1806
1807 TEST_F(TRACE_syscall, syscall_allowed)
1808 {
1809         long ret;
1810
1811         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1812         ASSERT_EQ(0, ret);
1813
1814         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1815         ASSERT_EQ(0, ret);
1816
1817         /* getppid works as expected (no changes). */
1818         EXPECT_EQ(self->parent, syscall(__NR_getppid));
1819         EXPECT_NE(self->mypid, syscall(__NR_getppid));
1820 }
1821
1822 TEST_F(TRACE_syscall, syscall_redirected)
1823 {
1824         long ret;
1825
1826         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1827         ASSERT_EQ(0, ret);
1828
1829         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1830         ASSERT_EQ(0, ret);
1831
1832         /* getpid has been redirected to getppid as expected. */
1833         EXPECT_EQ(self->parent, syscall(__NR_getpid));
1834         EXPECT_NE(self->mypid, syscall(__NR_getpid));
1835 }
1836
1837 TEST_F(TRACE_syscall, syscall_dropped)
1838 {
1839         long ret;
1840
1841         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1842         ASSERT_EQ(0, ret);
1843
1844         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1845         ASSERT_EQ(0, ret);
1846
1847         /* gettid has been skipped and an altered return value stored. */
1848         EXPECT_SYSCALL_RETURN(EPERM, syscall(__NR_gettid));
1849         EXPECT_NE(self->mytid, syscall(__NR_gettid));
1850 }
1851
1852 TEST_F(TRACE_syscall, skip_after_RET_TRACE)
1853 {
1854         struct sock_filter filter[] = {
1855                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1856                         offsetof(struct seccomp_data, nr)),
1857                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1858                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
1859                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1860         };
1861         struct sock_fprog prog = {
1862                 .len = (unsigned short)ARRAY_SIZE(filter),
1863                 .filter = filter,
1864         };
1865         long ret;
1866
1867         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1868         ASSERT_EQ(0, ret);
1869
1870         /* Install fixture filter. */
1871         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1872         ASSERT_EQ(0, ret);
1873
1874         /* Install "errno on getppid" filter. */
1875         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1876         ASSERT_EQ(0, ret);
1877
1878         /* Tracer will redirect getpid to getppid, and we should see EPERM. */
1879         errno = 0;
1880         EXPECT_EQ(-1, syscall(__NR_getpid));
1881         EXPECT_EQ(EPERM, errno);
1882 }
1883
1884 TEST_F_SIGNAL(TRACE_syscall, kill_after_RET_TRACE, SIGSYS)
1885 {
1886         struct sock_filter filter[] = {
1887                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1888                         offsetof(struct seccomp_data, nr)),
1889                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1890                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
1891                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1892         };
1893         struct sock_fprog prog = {
1894                 .len = (unsigned short)ARRAY_SIZE(filter),
1895                 .filter = filter,
1896         };
1897         long ret;
1898
1899         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1900         ASSERT_EQ(0, ret);
1901
1902         /* Install fixture filter. */
1903         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &self->prog, 0, 0);
1904         ASSERT_EQ(0, ret);
1905
1906         /* Install "death on getppid" filter. */
1907         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1908         ASSERT_EQ(0, ret);
1909
1910         /* Tracer will redirect getpid to getppid, and we should die. */
1911         EXPECT_NE(self->mypid, syscall(__NR_getpid));
1912 }
1913
1914 TEST_F(TRACE_syscall, skip_after_ptrace)
1915 {
1916         struct sock_filter filter[] = {
1917                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
1918                         offsetof(struct seccomp_data, nr)),
1919                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getppid, 0, 1),
1920                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EPERM),
1921                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1922         };
1923         struct sock_fprog prog = {
1924                 .len = (unsigned short)ARRAY_SIZE(filter),
1925                 .filter = filter,
1926         };
1927         long ret;
1928
1929         /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1930         teardown_trace_fixture(_metadata, self->tracer);
1931         self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1932                                            true);
1933
1934         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1935         ASSERT_EQ(0, ret);
1936
1937         /* Install "errno on getppid" filter. */
1938         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1939         ASSERT_EQ(0, ret);
1940
1941         /* Tracer will redirect getpid to getppid, and we should see EPERM. */
1942         EXPECT_EQ(-1, syscall(__NR_getpid));
1943         EXPECT_EQ(EPERM, errno);
1944 }
1945
1946 TEST_F_SIGNAL(TRACE_syscall, kill_after_ptrace, SIGSYS)
1947 {
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_getppid, 0, 1),
1952                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
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         long ret;
1960
1961         /* Swap SECCOMP_RET_TRACE tracer for PTRACE_SYSCALL tracer. */
1962         teardown_trace_fixture(_metadata, self->tracer);
1963         self->tracer = setup_trace_fixture(_metadata, tracer_ptrace, NULL,
1964                                            true);
1965
1966         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1967         ASSERT_EQ(0, ret);
1968
1969         /* Install "death on getppid" filter. */
1970         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
1971         ASSERT_EQ(0, ret);
1972
1973         /* Tracer will redirect getpid to getppid, and we should die. */
1974         EXPECT_NE(self->mypid, syscall(__NR_getpid));
1975 }
1976
1977 TEST(seccomp_syscall)
1978 {
1979         struct sock_filter filter[] = {
1980                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
1981         };
1982         struct sock_fprog prog = {
1983                 .len = (unsigned short)ARRAY_SIZE(filter),
1984                 .filter = filter,
1985         };
1986         long ret;
1987
1988         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
1989         ASSERT_EQ(0, ret) {
1990                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
1991         }
1992
1993         /* Reject insane operation. */
1994         ret = seccomp(-1, 0, &prog);
1995         ASSERT_NE(ENOSYS, errno) {
1996                 TH_LOG("Kernel does not support seccomp syscall!");
1997         }
1998         EXPECT_EQ(EINVAL, errno) {
1999                 TH_LOG("Did not reject crazy op value!");
2000         }
2001
2002         /* Reject strict with flags or pointer. */
2003         ret = seccomp(SECCOMP_SET_MODE_STRICT, -1, NULL);
2004         EXPECT_EQ(EINVAL, errno) {
2005                 TH_LOG("Did not reject mode strict with flags!");
2006         }
2007         ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, &prog);
2008         EXPECT_EQ(EINVAL, errno) {
2009                 TH_LOG("Did not reject mode strict with uargs!");
2010         }
2011
2012         /* Reject insane args for filter. */
2013         ret = seccomp(SECCOMP_SET_MODE_FILTER, -1, &prog);
2014         EXPECT_EQ(EINVAL, errno) {
2015                 TH_LOG("Did not reject crazy filter flags!");
2016         }
2017         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, NULL);
2018         EXPECT_EQ(EFAULT, errno) {
2019                 TH_LOG("Did not reject NULL filter!");
2020         }
2021
2022         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2023         EXPECT_EQ(0, errno) {
2024                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER: %s",
2025                         strerror(errno));
2026         }
2027 }
2028
2029 TEST(seccomp_syscall_mode_lock)
2030 {
2031         struct sock_filter filter[] = {
2032                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2033         };
2034         struct sock_fprog prog = {
2035                 .len = (unsigned short)ARRAY_SIZE(filter),
2036                 .filter = filter,
2037         };
2038         long ret;
2039
2040         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2041         ASSERT_EQ(0, ret) {
2042                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2043         }
2044
2045         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2046         ASSERT_NE(ENOSYS, errno) {
2047                 TH_LOG("Kernel does not support seccomp syscall!");
2048         }
2049         EXPECT_EQ(0, ret) {
2050                 TH_LOG("Could not install filter!");
2051         }
2052
2053         /* Make sure neither entry point will switch to strict. */
2054         ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_STRICT, 0, 0, 0);
2055         EXPECT_EQ(EINVAL, errno) {
2056                 TH_LOG("Switched to mode strict!");
2057         }
2058
2059         ret = seccomp(SECCOMP_SET_MODE_STRICT, 0, NULL);
2060         EXPECT_EQ(EINVAL, errno) {
2061                 TH_LOG("Switched to mode strict!");
2062         }
2063 }
2064
2065 /*
2066  * Test detection of known and unknown filter flags. Userspace needs to be able
2067  * to check if a filter flag is supported by the current kernel and a good way
2068  * of doing that is by attempting to enter filter mode, with the flag bit in
2069  * question set, and a NULL pointer for the _args_ parameter. EFAULT indicates
2070  * that the flag is valid and EINVAL indicates that the flag is invalid.
2071  */
2072 TEST(detect_seccomp_filter_flags)
2073 {
2074         unsigned int flags[] = { SECCOMP_FILTER_FLAG_TSYNC,
2075                                  SECCOMP_FILTER_FLAG_LOG };
2076         unsigned int flag, all_flags;
2077         int i;
2078         long ret;
2079
2080         /* Test detection of known-good filter flags */
2081         for (i = 0, all_flags = 0; i < ARRAY_SIZE(flags); i++) {
2082                 flag = flags[i];
2083                 ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2084                 ASSERT_NE(ENOSYS, errno) {
2085                         TH_LOG("Kernel does not support seccomp syscall!");
2086                 }
2087                 EXPECT_EQ(-1, ret);
2088                 EXPECT_EQ(EFAULT, errno) {
2089                         TH_LOG("Failed to detect that a known-good filter flag (0x%X) is supported!",
2090                                flag);
2091                 }
2092
2093                 all_flags |= flag;
2094         }
2095
2096         /* Test detection of all known-good filter flags */
2097         ret = seccomp(SECCOMP_SET_MODE_FILTER, all_flags, NULL);
2098         EXPECT_EQ(-1, ret);
2099         EXPECT_EQ(EFAULT, errno) {
2100                 TH_LOG("Failed to detect that all known-good filter flags (0x%X) are supported!",
2101                        all_flags);
2102         }
2103
2104         /* Test detection of an unknown filter flag */
2105         flag = -1;
2106         ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2107         EXPECT_EQ(-1, ret);
2108         EXPECT_EQ(EINVAL, errno) {
2109                 TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported!",
2110                        flag);
2111         }
2112
2113         /*
2114          * Test detection of an unknown filter flag that may simply need to be
2115          * added to this test
2116          */
2117         flag = flags[ARRAY_SIZE(flags) - 1] << 1;
2118         ret = seccomp(SECCOMP_SET_MODE_FILTER, flag, NULL);
2119         EXPECT_EQ(-1, ret);
2120         EXPECT_EQ(EINVAL, errno) {
2121                 TH_LOG("Failed to detect that an unknown filter flag (0x%X) is unsupported! Does a new flag need to be added to this test?",
2122                        flag);
2123         }
2124 }
2125
2126 TEST(TSYNC_first)
2127 {
2128         struct sock_filter filter[] = {
2129                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2130         };
2131         struct sock_fprog prog = {
2132                 .len = (unsigned short)ARRAY_SIZE(filter),
2133                 .filter = filter,
2134         };
2135         long ret;
2136
2137         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, NULL, 0, 0);
2138         ASSERT_EQ(0, ret) {
2139                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2140         }
2141
2142         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2143                       &prog);
2144         ASSERT_NE(ENOSYS, errno) {
2145                 TH_LOG("Kernel does not support seccomp syscall!");
2146         }
2147         EXPECT_EQ(0, ret) {
2148                 TH_LOG("Could not install initial filter with TSYNC!");
2149         }
2150 }
2151
2152 #define TSYNC_SIBLINGS 2
2153 struct tsync_sibling {
2154         pthread_t tid;
2155         pid_t system_tid;
2156         sem_t *started;
2157         pthread_cond_t *cond;
2158         pthread_mutex_t *mutex;
2159         int diverge;
2160         int num_waits;
2161         struct sock_fprog *prog;
2162         struct __test_metadata *metadata;
2163 };
2164
2165 /*
2166  * To avoid joining joined threads (which is not allowed by Bionic),
2167  * make sure we both successfully join and clear the tid to skip a
2168  * later join attempt during fixture teardown. Any remaining threads
2169  * will be directly killed during teardown.
2170  */
2171 #define PTHREAD_JOIN(tid, status)                                       \
2172         do {                                                            \
2173                 int _rc = pthread_join(tid, status);                    \
2174                 if (_rc) {                                              \
2175                         TH_LOG("pthread_join of tid %u failed: %d\n",   \
2176                                 (unsigned int)tid, _rc);                \
2177                 } else {                                                \
2178                         tid = 0;                                        \
2179                 }                                                       \
2180         } while (0)
2181
2182 FIXTURE_DATA(TSYNC) {
2183         struct sock_fprog root_prog, apply_prog;
2184         struct tsync_sibling sibling[TSYNC_SIBLINGS];
2185         sem_t started;
2186         pthread_cond_t cond;
2187         pthread_mutex_t mutex;
2188         int sibling_count;
2189 };
2190
2191 FIXTURE_SETUP(TSYNC)
2192 {
2193         struct sock_filter root_filter[] = {
2194                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2195         };
2196         struct sock_filter apply_filter[] = {
2197                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2198                         offsetof(struct seccomp_data, nr)),
2199                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
2200                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2201                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2202         };
2203
2204         memset(&self->root_prog, 0, sizeof(self->root_prog));
2205         memset(&self->apply_prog, 0, sizeof(self->apply_prog));
2206         memset(&self->sibling, 0, sizeof(self->sibling));
2207         self->root_prog.filter = malloc(sizeof(root_filter));
2208         ASSERT_NE(NULL, self->root_prog.filter);
2209         memcpy(self->root_prog.filter, &root_filter, sizeof(root_filter));
2210         self->root_prog.len = (unsigned short)ARRAY_SIZE(root_filter);
2211
2212         self->apply_prog.filter = malloc(sizeof(apply_filter));
2213         ASSERT_NE(NULL, self->apply_prog.filter);
2214         memcpy(self->apply_prog.filter, &apply_filter, sizeof(apply_filter));
2215         self->apply_prog.len = (unsigned short)ARRAY_SIZE(apply_filter);
2216
2217         self->sibling_count = 0;
2218         pthread_mutex_init(&self->mutex, NULL);
2219         pthread_cond_init(&self->cond, NULL);
2220         sem_init(&self->started, 0, 0);
2221         self->sibling[0].tid = 0;
2222         self->sibling[0].cond = &self->cond;
2223         self->sibling[0].started = &self->started;
2224         self->sibling[0].mutex = &self->mutex;
2225         self->sibling[0].diverge = 0;
2226         self->sibling[0].num_waits = 1;
2227         self->sibling[0].prog = &self->root_prog;
2228         self->sibling[0].metadata = _metadata;
2229         self->sibling[1].tid = 0;
2230         self->sibling[1].cond = &self->cond;
2231         self->sibling[1].started = &self->started;
2232         self->sibling[1].mutex = &self->mutex;
2233         self->sibling[1].diverge = 0;
2234         self->sibling[1].prog = &self->root_prog;
2235         self->sibling[1].num_waits = 1;
2236         self->sibling[1].metadata = _metadata;
2237 }
2238
2239 FIXTURE_TEARDOWN(TSYNC)
2240 {
2241         int sib = 0;
2242
2243         if (self->root_prog.filter)
2244                 free(self->root_prog.filter);
2245         if (self->apply_prog.filter)
2246                 free(self->apply_prog.filter);
2247
2248         for ( ; sib < self->sibling_count; ++sib) {
2249                 struct tsync_sibling *s = &self->sibling[sib];
2250
2251                 if (!s->tid)
2252                         continue;
2253                 /*
2254                  * If a thread is still running, it may be stuck, so hit
2255                  * it over the head really hard.
2256                  */
2257                 pthread_kill(s->tid, 9);
2258         }
2259         pthread_mutex_destroy(&self->mutex);
2260         pthread_cond_destroy(&self->cond);
2261         sem_destroy(&self->started);
2262 }
2263
2264 void *tsync_sibling(void *data)
2265 {
2266         long ret = 0;
2267         struct tsync_sibling *me = data;
2268
2269         me->system_tid = syscall(__NR_gettid);
2270
2271         pthread_mutex_lock(me->mutex);
2272         if (me->diverge) {
2273                 /* Just re-apply the root prog to fork the tree */
2274                 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER,
2275                                 me->prog, 0, 0);
2276         }
2277         sem_post(me->started);
2278         /* Return outside of started so parent notices failures. */
2279         if (ret) {
2280                 pthread_mutex_unlock(me->mutex);
2281                 return (void *)SIBLING_EXIT_FAILURE;
2282         }
2283         do {
2284                 pthread_cond_wait(me->cond, me->mutex);
2285                 me->num_waits = me->num_waits - 1;
2286         } while (me->num_waits);
2287         pthread_mutex_unlock(me->mutex);
2288
2289         ret = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0);
2290         if (!ret)
2291                 return (void *)SIBLING_EXIT_NEWPRIVS;
2292         read(0, NULL, 0);
2293         return (void *)SIBLING_EXIT_UNKILLED;
2294 }
2295
2296 void tsync_start_sibling(struct tsync_sibling *sibling)
2297 {
2298         pthread_create(&sibling->tid, NULL, tsync_sibling, (void *)sibling);
2299 }
2300
2301 TEST_F(TSYNC, siblings_fail_prctl)
2302 {
2303         long ret;
2304         void *status;
2305         struct sock_filter filter[] = {
2306                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2307                         offsetof(struct seccomp_data, nr)),
2308                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
2309                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | EINVAL),
2310                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2311         };
2312         struct sock_fprog prog = {
2313                 .len = (unsigned short)ARRAY_SIZE(filter),
2314                 .filter = filter,
2315         };
2316
2317         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2318                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2319         }
2320
2321         /* Check prctl failure detection by requesting sib 0 diverge. */
2322         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog);
2323         ASSERT_NE(ENOSYS, errno) {
2324                 TH_LOG("Kernel does not support seccomp syscall!");
2325         }
2326         ASSERT_EQ(0, ret) {
2327                 TH_LOG("setting filter failed");
2328         }
2329
2330         self->sibling[0].diverge = 1;
2331         tsync_start_sibling(&self->sibling[0]);
2332         tsync_start_sibling(&self->sibling[1]);
2333
2334         while (self->sibling_count < TSYNC_SIBLINGS) {
2335                 sem_wait(&self->started);
2336                 self->sibling_count++;
2337         }
2338
2339         /* Signal the threads to clean up*/
2340         pthread_mutex_lock(&self->mutex);
2341         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2342                 TH_LOG("cond broadcast non-zero");
2343         }
2344         pthread_mutex_unlock(&self->mutex);
2345
2346         /* Ensure diverging sibling failed to call prctl. */
2347         PTHREAD_JOIN(self->sibling[0].tid, &status);
2348         EXPECT_EQ(SIBLING_EXIT_FAILURE, (long)status);
2349         PTHREAD_JOIN(self->sibling[1].tid, &status);
2350         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2351 }
2352
2353 TEST_F(TSYNC, two_siblings_with_ancestor)
2354 {
2355         long ret;
2356         void *status;
2357
2358         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2359                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2360         }
2361
2362         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2363         ASSERT_NE(ENOSYS, errno) {
2364                 TH_LOG("Kernel does not support seccomp syscall!");
2365         }
2366         ASSERT_EQ(0, ret) {
2367                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2368         }
2369         tsync_start_sibling(&self->sibling[0]);
2370         tsync_start_sibling(&self->sibling[1]);
2371
2372         while (self->sibling_count < TSYNC_SIBLINGS) {
2373                 sem_wait(&self->started);
2374                 self->sibling_count++;
2375         }
2376
2377         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2378                       &self->apply_prog);
2379         ASSERT_EQ(0, ret) {
2380                 TH_LOG("Could install filter on all threads!");
2381         }
2382         /* Tell the siblings to test the policy */
2383         pthread_mutex_lock(&self->mutex);
2384         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2385                 TH_LOG("cond broadcast non-zero");
2386         }
2387         pthread_mutex_unlock(&self->mutex);
2388         /* Ensure they are both killed and don't exit cleanly. */
2389         PTHREAD_JOIN(self->sibling[0].tid, &status);
2390         EXPECT_EQ(0x0, (long)status);
2391         PTHREAD_JOIN(self->sibling[1].tid, &status);
2392         EXPECT_EQ(0x0, (long)status);
2393 }
2394
2395 TEST_F(TSYNC, two_sibling_want_nnp)
2396 {
2397         void *status;
2398
2399         /* start siblings before any prctl() operations */
2400         tsync_start_sibling(&self->sibling[0]);
2401         tsync_start_sibling(&self->sibling[1]);
2402         while (self->sibling_count < TSYNC_SIBLINGS) {
2403                 sem_wait(&self->started);
2404                 self->sibling_count++;
2405         }
2406
2407         /* Tell the siblings to test no policy */
2408         pthread_mutex_lock(&self->mutex);
2409         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2410                 TH_LOG("cond broadcast non-zero");
2411         }
2412         pthread_mutex_unlock(&self->mutex);
2413
2414         /* Ensure they are both upset about lacking nnp. */
2415         PTHREAD_JOIN(self->sibling[0].tid, &status);
2416         EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2417         PTHREAD_JOIN(self->sibling[1].tid, &status);
2418         EXPECT_EQ(SIBLING_EXIT_NEWPRIVS, (long)status);
2419 }
2420
2421 TEST_F(TSYNC, two_siblings_with_no_filter)
2422 {
2423         long ret;
2424         void *status;
2425
2426         /* start siblings before any prctl() operations */
2427         tsync_start_sibling(&self->sibling[0]);
2428         tsync_start_sibling(&self->sibling[1]);
2429         while (self->sibling_count < TSYNC_SIBLINGS) {
2430                 sem_wait(&self->started);
2431                 self->sibling_count++;
2432         }
2433
2434         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2435                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2436         }
2437
2438         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2439                       &self->apply_prog);
2440         ASSERT_NE(ENOSYS, errno) {
2441                 TH_LOG("Kernel does not support seccomp syscall!");
2442         }
2443         ASSERT_EQ(0, ret) {
2444                 TH_LOG("Could install filter on all threads!");
2445         }
2446
2447         /* Tell the siblings to test the policy */
2448         pthread_mutex_lock(&self->mutex);
2449         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2450                 TH_LOG("cond broadcast non-zero");
2451         }
2452         pthread_mutex_unlock(&self->mutex);
2453
2454         /* Ensure they are both killed and don't exit cleanly. */
2455         PTHREAD_JOIN(self->sibling[0].tid, &status);
2456         EXPECT_EQ(0x0, (long)status);
2457         PTHREAD_JOIN(self->sibling[1].tid, &status);
2458         EXPECT_EQ(0x0, (long)status);
2459 }
2460
2461 TEST_F(TSYNC, two_siblings_with_one_divergence)
2462 {
2463         long ret;
2464         void *status;
2465
2466         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2467                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2468         }
2469
2470         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2471         ASSERT_NE(ENOSYS, errno) {
2472                 TH_LOG("Kernel does not support seccomp syscall!");
2473         }
2474         ASSERT_EQ(0, ret) {
2475                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2476         }
2477         self->sibling[0].diverge = 1;
2478         tsync_start_sibling(&self->sibling[0]);
2479         tsync_start_sibling(&self->sibling[1]);
2480
2481         while (self->sibling_count < TSYNC_SIBLINGS) {
2482                 sem_wait(&self->started);
2483                 self->sibling_count++;
2484         }
2485
2486         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2487                       &self->apply_prog);
2488         ASSERT_EQ(self->sibling[0].system_tid, ret) {
2489                 TH_LOG("Did not fail on diverged sibling.");
2490         }
2491
2492         /* Wake the threads */
2493         pthread_mutex_lock(&self->mutex);
2494         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2495                 TH_LOG("cond broadcast non-zero");
2496         }
2497         pthread_mutex_unlock(&self->mutex);
2498
2499         /* Ensure they are both unkilled. */
2500         PTHREAD_JOIN(self->sibling[0].tid, &status);
2501         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2502         PTHREAD_JOIN(self->sibling[1].tid, &status);
2503         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2504 }
2505
2506 TEST_F(TSYNC, two_siblings_not_under_filter)
2507 {
2508         long ret, sib;
2509         void *status;
2510
2511         ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2512                 TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2513         }
2514
2515         /*
2516          * Sibling 0 will have its own seccomp policy
2517          * and Sibling 1 will not be under seccomp at
2518          * all. Sibling 1 will enter seccomp and 0
2519          * will cause failure.
2520          */
2521         self->sibling[0].diverge = 1;
2522         tsync_start_sibling(&self->sibling[0]);
2523         tsync_start_sibling(&self->sibling[1]);
2524
2525         while (self->sibling_count < TSYNC_SIBLINGS) {
2526                 sem_wait(&self->started);
2527                 self->sibling_count++;
2528         }
2529
2530         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &self->root_prog);
2531         ASSERT_NE(ENOSYS, errno) {
2532                 TH_LOG("Kernel does not support seccomp syscall!");
2533         }
2534         ASSERT_EQ(0, ret) {
2535                 TH_LOG("Kernel does not support SECCOMP_SET_MODE_FILTER!");
2536         }
2537
2538         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2539                       &self->apply_prog);
2540         ASSERT_EQ(ret, self->sibling[0].system_tid) {
2541                 TH_LOG("Did not fail on diverged sibling.");
2542         }
2543         sib = 1;
2544         if (ret == self->sibling[0].system_tid)
2545                 sib = 0;
2546
2547         pthread_mutex_lock(&self->mutex);
2548
2549         /* Increment the other siblings num_waits so we can clean up
2550          * the one we just saw.
2551          */
2552         self->sibling[!sib].num_waits += 1;
2553
2554         /* Signal the thread to clean up*/
2555         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2556                 TH_LOG("cond broadcast non-zero");
2557         }
2558         pthread_mutex_unlock(&self->mutex);
2559         PTHREAD_JOIN(self->sibling[sib].tid, &status);
2560         EXPECT_EQ(SIBLING_EXIT_UNKILLED, (long)status);
2561         /* Poll for actual task death. pthread_join doesn't guarantee it. */
2562         while (!kill(self->sibling[sib].system_tid, 0))
2563                 sleep(0.1);
2564         /* Switch to the remaining sibling */
2565         sib = !sib;
2566
2567         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2568                       &self->apply_prog);
2569         ASSERT_EQ(0, ret) {
2570                 TH_LOG("Expected the remaining sibling to sync");
2571         };
2572
2573         pthread_mutex_lock(&self->mutex);
2574
2575         /* If remaining sibling didn't have a chance to wake up during
2576          * the first broadcast, manually reduce the num_waits now.
2577          */
2578         if (self->sibling[sib].num_waits > 1)
2579                 self->sibling[sib].num_waits = 1;
2580         ASSERT_EQ(0, pthread_cond_broadcast(&self->cond)) {
2581                 TH_LOG("cond broadcast non-zero");
2582         }
2583         pthread_mutex_unlock(&self->mutex);
2584         PTHREAD_JOIN(self->sibling[sib].tid, &status);
2585         EXPECT_EQ(0, (long)status);
2586         /* Poll for actual task death. pthread_join doesn't guarantee it. */
2587         while (!kill(self->sibling[sib].system_tid, 0))
2588                 sleep(0.1);
2589
2590         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_TSYNC,
2591                       &self->apply_prog);
2592         ASSERT_EQ(0, ret);  /* just us chickens */
2593 }
2594
2595 /* Make sure restarted syscalls are seen directly as "restart_syscall". */
2596 TEST(syscall_restart)
2597 {
2598         long ret;
2599         unsigned long msg;
2600         pid_t child_pid;
2601         int pipefd[2];
2602         int status;
2603         siginfo_t info = { };
2604         struct sock_filter filter[] = {
2605                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2606                          offsetof(struct seccomp_data, nr)),
2607
2608 #ifdef __NR_sigreturn
2609                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_sigreturn, 6, 0),
2610 #endif
2611                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 5, 0),
2612                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_exit, 4, 0),
2613                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_rt_sigreturn, 3, 0),
2614                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_nanosleep, 4, 0),
2615                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_restart_syscall, 4, 0),
2616
2617                 /* Allow __NR_write for easy logging. */
2618                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_write, 0, 1),
2619                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2620                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2621                 /* The nanosleep jump target. */
2622                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x100),
2623                 /* The restart_syscall jump target. */
2624                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_TRACE|0x200),
2625         };
2626         struct sock_fprog prog = {
2627                 .len = (unsigned short)ARRAY_SIZE(filter),
2628                 .filter = filter,
2629         };
2630 #if defined(__arm__)
2631         struct utsname utsbuf;
2632 #endif
2633
2634         ASSERT_EQ(0, pipe(pipefd));
2635
2636         child_pid = fork();
2637         ASSERT_LE(0, child_pid);
2638         if (child_pid == 0) {
2639                 /* Child uses EXPECT not ASSERT to deliver status correctly. */
2640                 char buf = ' ';
2641                 struct timespec timeout = { };
2642
2643                 /* Attach parent as tracer and stop. */
2644                 EXPECT_EQ(0, ptrace(PTRACE_TRACEME));
2645                 EXPECT_EQ(0, raise(SIGSTOP));
2646
2647                 EXPECT_EQ(0, close(pipefd[1]));
2648
2649                 EXPECT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
2650                         TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
2651                 }
2652
2653                 ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog, 0, 0);
2654                 EXPECT_EQ(0, ret) {
2655                         TH_LOG("Failed to install filter!");
2656                 }
2657
2658                 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2659                         TH_LOG("Failed to read() sync from parent");
2660                 }
2661                 EXPECT_EQ('.', buf) {
2662                         TH_LOG("Failed to get sync data from read()");
2663                 }
2664
2665                 /* Start nanosleep to be interrupted. */
2666                 timeout.tv_sec = 1;
2667                 errno = 0;
2668                 EXPECT_EQ(0, nanosleep(&timeout, NULL)) {
2669                         TH_LOG("Call to nanosleep() failed (errno %d)", errno);
2670                 }
2671
2672                 /* Read final sync from parent. */
2673                 EXPECT_EQ(1, read(pipefd[0], &buf, 1)) {
2674                         TH_LOG("Failed final read() from parent");
2675                 }
2676                 EXPECT_EQ('!', buf) {
2677                         TH_LOG("Failed to get final data from read()");
2678                 }
2679
2680                 /* Directly report the status of our test harness results. */
2681                 syscall(__NR_exit, _metadata->passed ? EXIT_SUCCESS
2682                                                      : EXIT_FAILURE);
2683         }
2684         EXPECT_EQ(0, close(pipefd[0]));
2685
2686         /* Attach to child, setup options, and release. */
2687         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2688         ASSERT_EQ(true, WIFSTOPPED(status));
2689         ASSERT_EQ(0, ptrace(PTRACE_SETOPTIONS, child_pid, NULL,
2690                             PTRACE_O_TRACESECCOMP));
2691         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2692         ASSERT_EQ(1, write(pipefd[1], ".", 1));
2693
2694         /* Wait for nanosleep() to start. */
2695         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2696         ASSERT_EQ(true, WIFSTOPPED(status));
2697         ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2698         ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2699         ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2700         ASSERT_EQ(0x100, msg);
2701         EXPECT_EQ(__NR_nanosleep, get_syscall(_metadata, child_pid));
2702
2703         /* Might as well check siginfo for sanity while we're here. */
2704         ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2705         ASSERT_EQ(SIGTRAP, info.si_signo);
2706         ASSERT_EQ(SIGTRAP | (PTRACE_EVENT_SECCOMP << 8), info.si_code);
2707         EXPECT_EQ(0, info.si_errno);
2708         EXPECT_EQ(getuid(), info.si_uid);
2709         /* Verify signal delivery came from child (seccomp-triggered). */
2710         EXPECT_EQ(child_pid, info.si_pid);
2711
2712         /* Interrupt nanosleep with SIGSTOP (which we'll need to handle). */
2713         ASSERT_EQ(0, kill(child_pid, SIGSTOP));
2714         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2715         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2716         ASSERT_EQ(true, WIFSTOPPED(status));
2717         ASSERT_EQ(SIGSTOP, WSTOPSIG(status));
2718         /* Verify signal delivery came from parent now. */
2719         ASSERT_EQ(0, ptrace(PTRACE_GETSIGINFO, child_pid, NULL, &info));
2720         EXPECT_EQ(getpid(), info.si_pid);
2721
2722         /* Restart nanosleep with SIGCONT, which triggers restart_syscall. */
2723         ASSERT_EQ(0, kill(child_pid, SIGCONT));
2724         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2725         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2726         ASSERT_EQ(true, WIFSTOPPED(status));
2727         ASSERT_EQ(SIGCONT, WSTOPSIG(status));
2728         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2729
2730         /* Wait for restart_syscall() to start. */
2731         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2732         ASSERT_EQ(true, WIFSTOPPED(status));
2733         ASSERT_EQ(SIGTRAP, WSTOPSIG(status));
2734         ASSERT_EQ(PTRACE_EVENT_SECCOMP, (status >> 16));
2735         ASSERT_EQ(0, ptrace(PTRACE_GETEVENTMSG, child_pid, NULL, &msg));
2736
2737         ASSERT_EQ(0x200, msg);
2738         ret = get_syscall(_metadata, child_pid);
2739 #if defined(__arm__)
2740         /*
2741          * FIXME:
2742          * - native ARM registers do NOT expose true syscall.
2743          * - compat ARM registers on ARM64 DO expose true syscall.
2744          */
2745         ASSERT_EQ(0, uname(&utsbuf));
2746         if (strncmp(utsbuf.machine, "arm", 3) == 0) {
2747                 EXPECT_EQ(__NR_nanosleep, ret);
2748         } else
2749 #endif
2750         {
2751                 EXPECT_EQ(__NR_restart_syscall, ret);
2752         }
2753
2754         /* Write again to end test. */
2755         ASSERT_EQ(0, ptrace(PTRACE_CONT, child_pid, NULL, 0));
2756         ASSERT_EQ(1, write(pipefd[1], "!", 1));
2757         EXPECT_EQ(0, close(pipefd[1]));
2758
2759         ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
2760         if (WIFSIGNALED(status) || WEXITSTATUS(status))
2761                 _metadata->passed = 0;
2762 }
2763
2764 TEST_SIGNAL(filter_flag_log, SIGSYS)
2765 {
2766         struct sock_filter allow_filter[] = {
2767                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2768         };
2769         struct sock_filter kill_filter[] = {
2770                 BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
2771                         offsetof(struct seccomp_data, nr)),
2772                 BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_getpid, 0, 1),
2773                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
2774                 BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2775         };
2776         struct sock_fprog allow_prog = {
2777                 .len = (unsigned short)ARRAY_SIZE(allow_filter),
2778                 .filter = allow_filter,
2779         };
2780         struct sock_fprog kill_prog = {
2781                 .len = (unsigned short)ARRAY_SIZE(kill_filter),
2782                 .filter = kill_filter,
2783         };
2784         long ret;
2785         pid_t parent = getppid();
2786
2787         ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
2788         ASSERT_EQ(0, ret);
2789
2790         /* Verify that the FILTER_FLAG_LOG flag isn't accepted in strict mode */
2791         ret = seccomp(SECCOMP_SET_MODE_STRICT, SECCOMP_FILTER_FLAG_LOG,
2792                       &allow_prog);
2793         ASSERT_NE(ENOSYS, errno) {
2794                 TH_LOG("Kernel does not support seccomp syscall!");
2795         }
2796         EXPECT_NE(0, ret) {
2797                 TH_LOG("Kernel accepted FILTER_FLAG_LOG flag in strict mode!");
2798         }
2799         EXPECT_EQ(EINVAL, errno) {
2800                 TH_LOG("Kernel returned unexpected errno for FILTER_FLAG_LOG flag in strict mode!");
2801         }
2802
2803         /* Verify that a simple, permissive filter can be added with no flags */
2804         ret = seccomp(SECCOMP_SET_MODE_FILTER, 0, &allow_prog);
2805         EXPECT_EQ(0, ret);
2806
2807         /* See if the same filter can be added with the FILTER_FLAG_LOG flag */
2808         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
2809                       &allow_prog);
2810         ASSERT_NE(EINVAL, errno) {
2811                 TH_LOG("Kernel does not support the FILTER_FLAG_LOG flag!");
2812         }
2813         EXPECT_EQ(0, ret);
2814
2815         /* Ensure that the kill filter works with the FILTER_FLAG_LOG flag */
2816         ret = seccomp(SECCOMP_SET_MODE_FILTER, SECCOMP_FILTER_FLAG_LOG,
2817                       &kill_prog);
2818         EXPECT_EQ(0, ret);
2819
2820         EXPECT_EQ(parent, syscall(__NR_getppid));
2821         /* getpid() should never return. */
2822         EXPECT_EQ(0, syscall(__NR_getpid));
2823 }
2824
2825 TEST(get_action_avail)
2826 {
2827         __u32 actions[] = { SECCOMP_RET_KILL_THREAD, SECCOMP_RET_TRAP,
2828                             SECCOMP_RET_ERRNO, SECCOMP_RET_TRACE,
2829                             SECCOMP_RET_LOG,   SECCOMP_RET_ALLOW };
2830         __u32 unknown_action = 0x10000000U;
2831         int i;
2832         long ret;
2833
2834         ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[0]);
2835         ASSERT_NE(ENOSYS, errno) {
2836                 TH_LOG("Kernel does not support seccomp syscall!");
2837         }
2838         ASSERT_NE(EINVAL, errno) {
2839                 TH_LOG("Kernel does not support SECCOMP_GET_ACTION_AVAIL operation!");
2840         }
2841         EXPECT_EQ(ret, 0);
2842
2843         for (i = 0; i < ARRAY_SIZE(actions); i++) {
2844                 ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &actions[i]);
2845                 EXPECT_EQ(ret, 0) {
2846                         TH_LOG("Expected action (0x%X) not available!",
2847                                actions[i]);
2848                 }
2849         }
2850
2851         /* Check that an unknown action is handled properly (EOPNOTSUPP) */
2852         ret = seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &unknown_action);
2853         EXPECT_EQ(ret, -1);
2854         EXPECT_EQ(errno, EOPNOTSUPP);
2855 }
2856
2857 TEST(get_metadata)
2858 {
2859         pid_t pid;
2860         int pipefd[2];
2861         char buf;
2862         struct seccomp_metadata md;
2863
2864         ASSERT_EQ(0, pipe(pipefd));
2865
2866         pid = fork();
2867         ASSERT_GE(pid, 0);
2868         if (pid == 0) {
2869                 struct sock_filter filter[] = {
2870                         BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
2871                 };
2872                 struct sock_fprog prog = {
2873                         .len = (unsigned short)ARRAY_SIZE(filter),
2874                         .filter = filter,
2875                 };
2876
2877                 /* one with log, one without */
2878                 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER,
2879                                      SECCOMP_FILTER_FLAG_LOG, &prog));
2880                 ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, 0, &prog));
2881
2882                 ASSERT_EQ(0, close(pipefd[0]));
2883                 ASSERT_EQ(1, write(pipefd[1], "1", 1));
2884                 ASSERT_EQ(0, close(pipefd[1]));
2885
2886                 while (1)
2887                         sleep(100);
2888         }
2889
2890         ASSERT_EQ(0, close(pipefd[1]));
2891         ASSERT_EQ(1, read(pipefd[0], &buf, 1));
2892
2893         ASSERT_EQ(0, ptrace(PTRACE_ATTACH, pid));
2894         ASSERT_EQ(pid, waitpid(pid, NULL, 0));
2895
2896         md.filter_off = 0;
2897         ASSERT_EQ(sizeof(md), ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md));
2898         EXPECT_EQ(md.flags, SECCOMP_FILTER_FLAG_LOG);
2899         EXPECT_EQ(md.filter_off, 0);
2900
2901         md.filter_off = 1;
2902         ASSERT_EQ(sizeof(md), ptrace(PTRACE_SECCOMP_GET_METADATA, pid, sizeof(md), &md));
2903         EXPECT_EQ(md.flags, 0);
2904         EXPECT_EQ(md.filter_off, 1);
2905
2906         ASSERT_EQ(0, kill(pid, SIGKILL));
2907 }
2908
2909 /*
2910  * TODO:
2911  * - add microbenchmarks
2912  * - expand NNP testing
2913  * - better arch-specific TRACE and TRAP handlers.
2914  * - endianness checking when appropriate
2915  * - 64-bit arg prodding
2916  * - arch value testing (x86 modes especially)
2917  * - verify that FILTER_FLAG_LOG filters generate log messages
2918  * - verify that RET_LOG generates log messages
2919  * - ...
2920  */
2921
2922 TEST_HARNESS_MAIN