src/test/test-seccomp.c

   1 /***
   2   This file is part of systemd.
   3
   4   Copyright 2016 Lennart Poettering
   5
   6   systemd is free software; you can redistribute it and/or modify it
   7   under the terms of the GNU Lesser General Public License as published by
   8   the Free Software Foundation; either version 2.1 of the License, or
   9   (at your option) any later version.
  10
  11   systemd is distributed in the hope that it will be useful, but
  12   WITHOUT ANY WARRANTY; without even the implied warranty of
  13   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14   Lesser General Public License for more details.
  15
  16   You should have received a copy of the GNU Lesser General Public License
  17   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  18 ***/
  19
  20 #include <sched.h>
  21 #include <stdlib.h>
  22 #include <sys/eventfd.h>
  23 #include <sys/mman.h>
  24 #include <sys/poll.h>
  25 #include <sys/shm.h>
  26 #include <sys/types.h>
  27 #include <unistd.h>
  28
  29 #include "alloc-util.h"
  30 #include "fd-util.h"
  31 #include "macro.h"
  32 #include "missing.h"
  33 #include "nsflags.h"
  34 #include "process-util.h"
  35 #include "raw-clone.h"
  36 #include "seccomp-util.h"
  37 #include "set.h"
  38 #include "string-util.h"
  39 #include "util.h"
  40 #include "virt.h"
  41
  42 #if SCMP_SYS(socket) < 0 || defined(__i386__) || defined(__s390x__) || defined(__s390__)
  43 /* On these archs, socket() is implemented via the socketcall() syscall multiplexer,
  44  * and we can't restrict it hence via seccomp. */
  45 #  define SECCOMP_RESTRICT_ADDRESS_FAMILIES_BROKEN 1
  46 #else
  47 #  define SECCOMP_RESTRICT_ADDRESS_FAMILIES_BROKEN 0
  48 #endif
  49
  50
  51 static void test_seccomp_arch_to_string(void) {
  52         uint32_t a, b;
  53         const char *name;
  54
  55         a = seccomp_arch_native();
  56         assert_se(a > 0);
  57         name = seccomp_arch_to_string(a);
  58         assert_se(name);
  59         assert_se(seccomp_arch_from_string(name, &b) >= 0);
  60         assert_se(a == b);
  61 }
  62
  63 static void test_architecture_table(void) {
  64         const char *n, *n2;
  65
  66         NULSTR_FOREACH(n,
  67                        "native\0"
  68                        "x86\0"
  69                        "x86-64\0"
  70                        "x32\0"
  71                        "arm\0"
  72                        "arm64\0"
  73                        "mips\0"
  74                        "mips64\0"
  75                        "mips64-n32\0"
  76                        "mips-le\0"
  77                        "mips64-le\0"
  78                        "mips64-le-n32\0"
  79                        "ppc\0"
  80                        "ppc64\0"
  81                        "ppc64-le\0"
  82                        "s390\0"
  83                        "s390x\0") {
  84                 uint32_t c;
  85
  86                 assert_se(seccomp_arch_from_string(n, &c) >= 0);
  87                 n2 = seccomp_arch_to_string(c);
  88                 log_info("seccomp-arch: %s → 0x%"PRIx32" → %s", n, c, n2);
  89                 assert_se(streq_ptr(n, n2));
  90         }
  91 }
  92
  93 static void test_syscall_filter_set_find(void) {
  94         assert_se(!syscall_filter_set_find(NULL));
  95         assert_se(!syscall_filter_set_find(""));
  96         assert_se(!syscall_filter_set_find("quux"));
  97         assert_se(!syscall_filter_set_find("@quux"));
  98
  99         assert_se(syscall_filter_set_find("@clock") == syscall_filter_sets + SYSCALL_FILTER_SET_CLOCK);
 100         assert_se(syscall_filter_set_find("@default") == syscall_filter_sets + SYSCALL_FILTER_SET_DEFAULT);
 101         assert_se(syscall_filter_set_find("@raw-io") == syscall_filter_sets + SYSCALL_FILTER_SET_RAW_IO);
 102 }
 103
 104 static void test_filter_sets(void) {
 105         unsigned i;
 106         int r;
 107
 108         if (!is_seccomp_available())
 109                 return;
 110         if (geteuid() != 0)
 111                 return;
 112
 113         for (i = 0; i < _SYSCALL_FILTER_SET_MAX; i++) {
 114                 pid_t pid;
 115
 116                 log_info("Testing %s", syscall_filter_sets[i].name);
 117
 118                 pid = fork();
 119                 assert_se(pid >= 0);
 120
 121                 if (pid == 0) { /* Child? */
 122                         int fd;
 123
 124                         if (i == SYSCALL_FILTER_SET_DEFAULT) /* if we look at the default set, whitelist instead of blacklist */
 125                                 r = seccomp_load_syscall_filter_set(SCMP_ACT_ERRNO(EUCLEAN), syscall_filter_sets + i, SCMP_ACT_ALLOW);
 126                         else
 127                                 r = seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW, syscall_filter_sets + i, SCMP_ACT_ERRNO(EUCLEAN));
 128                         if (r < 0)
 129                                 _exit(EXIT_FAILURE);
 130
 131                         /* Test the sycall filter with one random system call */
 132                         fd = eventfd(0, EFD_NONBLOCK|EFD_CLOEXEC);
 133                         if (IN_SET(i, SYSCALL_FILTER_SET_IO_EVENT, SYSCALL_FILTER_SET_DEFAULT))
 134                                 assert_se(fd < 0 && errno == EUCLEAN);
 135                         else {
 136                                 assert_se(fd >= 0);
 137                                 safe_close(fd);
 138                         }
 139
 140                         _exit(EXIT_SUCCESS);
 141                 }
 142
 143                 assert_se(wait_for_terminate_and_warn(syscall_filter_sets[i].name, pid, true) == EXIT_SUCCESS);
 144         }
 145 }
 146
 147 static void test_restrict_namespace(void) {
 148         _cleanup_free_ char *s = NULL;
 149         unsigned long ul;
 150         pid_t pid;
 151
 152         assert_se(namespace_flag_to_string(0) == NULL);
 153         assert_se(streq(namespace_flag_to_string(CLONE_NEWNS), "mnt"));
 154         assert_se(namespace_flag_to_string(CLONE_NEWNS|CLONE_NEWIPC) == NULL);
 155         assert_se(streq(namespace_flag_to_string(CLONE_NEWCGROUP), "cgroup"));
 156
 157         assert_se(namespace_flag_from_string("mnt") == CLONE_NEWNS);
 158         assert_se(namespace_flag_from_string(NULL) == 0);
 159         assert_se(namespace_flag_from_string("") == 0);
 160         assert_se(namespace_flag_from_string("uts") == CLONE_NEWUTS);
 161         assert_se(namespace_flag_from_string(namespace_flag_to_string(CLONE_NEWUTS)) == CLONE_NEWUTS);
 162         assert_se(streq(namespace_flag_to_string(namespace_flag_from_string("ipc")), "ipc"));
 163
 164         assert_se(namespace_flag_from_string_many(NULL, &ul) == 0 && ul == 0);
 165         assert_se(namespace_flag_from_string_many("", &ul) == 0 && ul == 0);
 166         assert_se(namespace_flag_from_string_many("mnt uts ipc", &ul) == 0 && ul == (CLONE_NEWNS|CLONE_NEWUTS|CLONE_NEWIPC));
 167
 168         assert_se(namespace_flag_to_string_many(NAMESPACE_FLAGS_ALL, &s) == 0);
 169         assert_se(streq(s, "cgroup ipc net mnt pid user uts"));
 170         assert_se(namespace_flag_from_string_many(s, &ul) == 0 && ul == NAMESPACE_FLAGS_ALL);
 171
 172         if (!is_seccomp_available())
 173                 return;
 174         if (geteuid() != 0)
 175                 return;
 176
 177         pid = fork();
 178         assert_se(pid >= 0);
 179
 180         if (pid == 0) {
 181
 182                 assert_se(seccomp_restrict_namespaces(CLONE_NEWNS|CLONE_NEWNET) >= 0);
 183
 184                 assert_se(unshare(CLONE_NEWNS) == 0);
 185                 assert_se(unshare(CLONE_NEWNET) == 0);
 186                 assert_se(unshare(CLONE_NEWUTS) == -1);
 187                 assert_se(errno == EPERM);
 188                 assert_se(unshare(CLONE_NEWIPC) == -1);
 189                 assert_se(errno == EPERM);
 190                 assert_se(unshare(CLONE_NEWNET|CLONE_NEWUTS) == -1);
 191                 assert_se(errno == EPERM);
 192
 193                 /* We use fd 0 (stdin) here, which of course will fail with EINVAL on setns(). Except of course our
 194                  * seccomp filter worked, and hits first and makes it return EPERM */
 195                 assert_se(setns(0, CLONE_NEWNS) == -1);
 196                 assert_se(errno == EINVAL);
 197                 assert_se(setns(0, CLONE_NEWNET) == -1);
 198                 assert_se(errno == EINVAL);
 199                 assert_se(setns(0, CLONE_NEWUTS) == -1);
 200                 assert_se(errno == EPERM);
 201                 assert_se(setns(0, CLONE_NEWIPC) == -1);
 202                 assert_se(errno == EPERM);
 203                 assert_se(setns(0, CLONE_NEWNET|CLONE_NEWUTS) == -1);
 204                 assert_se(errno == EPERM);
 205                 assert_se(setns(0, 0) == -1);
 206                 assert_se(errno == EPERM);
 207
 208                 pid = raw_clone(CLONE_NEWNS);
 209                 assert_se(pid >= 0);
 210                 if (pid == 0)
 211                         _exit(EXIT_SUCCESS);
 212                 pid = raw_clone(CLONE_NEWNET);
 213                 assert_se(pid >= 0);
 214                 if (pid == 0)
 215                         _exit(EXIT_SUCCESS);
 216                 pid = raw_clone(CLONE_NEWUTS);
 217                 assert_se(pid < 0);
 218                 assert_se(errno == EPERM);
 219                 pid = raw_clone(CLONE_NEWIPC);
 220                 assert_se(pid < 0);
 221                 assert_se(errno == EPERM);
 222                 pid = raw_clone(CLONE_NEWNET|CLONE_NEWUTS);
 223                 assert_se(pid < 0);
 224                 assert_se(errno == EPERM);
 225
 226                 _exit(EXIT_SUCCESS);
 227         }
 228
 229         assert_se(wait_for_terminate_and_warn("nsseccomp", pid, true) == EXIT_SUCCESS);
 230 }
 231
 232 static void test_protect_sysctl(void) {
 233         pid_t pid;
 234
 235         if (!is_seccomp_available())
 236                 return;
 237         if (geteuid() != 0)
 238                 return;
 239
 240         if (detect_container() > 0) /* in containers _sysctl() is likely missing anyway */
 241                 return;
 242
 243         pid = fork();
 244         assert_se(pid >= 0);
 245
 246         if (pid == 0) {
 247 #if __NR__sysctl > 0
 248                 assert_se(syscall(__NR__sysctl, NULL) < 0);
 249                 assert_se(errno == EFAULT);
 250 #endif
 251
 252                 assert_se(seccomp_protect_sysctl() >= 0);
 253
 254 #if __NR__sysctl > 0
 255                 assert_se(syscall(__NR__sysctl, 0, 0, 0) < 0);
 256                 assert_se(errno == EPERM);
 257 #endif
 258
 259                 _exit(EXIT_SUCCESS);
 260         }
 261
 262         assert_se(wait_for_terminate_and_warn("sysctlseccomp", pid, true) == EXIT_SUCCESS);
 263 }
 264
 265 static void test_restrict_address_families(void) {
 266         pid_t pid;
 267
 268         if (!is_seccomp_available())
 269                 return;
 270         if (geteuid() != 0)
 271                 return;
 272
 273         pid = fork();
 274         assert_se(pid >= 0);
 275
 276         if (pid == 0) {
 277                 int fd;
 278                 Set *s;
 279
 280                 fd = socket(AF_INET, SOCK_DGRAM, 0);
 281                 assert_se(fd >= 0);
 282                 safe_close(fd);
 283
 284                 fd = socket(AF_UNIX, SOCK_DGRAM, 0);
 285                 assert_se(fd >= 0);
 286                 safe_close(fd);
 287
 288                 fd = socket(AF_NETLINK, SOCK_DGRAM, 0);
 289                 assert_se(fd >= 0);
 290                 safe_close(fd);
 291
 292                 assert_se(s = set_new(NULL));
 293                 assert_se(set_put(s, INT_TO_PTR(AF_UNIX)) >= 0);
 294
 295                 assert_se(seccomp_restrict_address_families(s, false) >= 0);
 296
 297                 fd = socket(AF_INET, SOCK_DGRAM, 0);
 298                 assert_se(fd >= 0);
 299                 safe_close(fd);
 300
 301                 fd = socket(AF_UNIX, SOCK_DGRAM, 0);
 302 #if SECCOMP_RESTRICT_ADDRESS_FAMILIES_BROKEN
 303                 assert_se(fd >= 0);
 304                 safe_close(fd);
 305 #else
 306                 assert_se(fd < 0);
 307                 assert_se(errno == EAFNOSUPPORT);
 308 #endif
 309
 310                 fd = socket(AF_NETLINK, SOCK_DGRAM, 0);
 311                 assert_se(fd >= 0);
 312                 safe_close(fd);
 313
 314                 set_clear(s);
 315
 316                 assert_se(set_put(s, INT_TO_PTR(AF_INET)) >= 0);
 317
 318                 assert_se(seccomp_restrict_address_families(s, true) >= 0);
 319
 320                 fd = socket(AF_INET, SOCK_DGRAM, 0);
 321                 assert_se(fd >= 0);
 322                 safe_close(fd);
 323
 324                 fd = socket(AF_UNIX, SOCK_DGRAM, 0);
 325 #if SECCOMP_RESTRICT_ADDRESS_FAMILIES_BROKEN
 326                 assert_se(fd >= 0);
 327                 safe_close(fd);
 328 #else
 329                 assert_se(fd < 0);
 330                 assert_se(errno == EAFNOSUPPORT);
 331 #endif
 332
 333                 fd = socket(AF_NETLINK, SOCK_DGRAM, 0);
 334 #if SECCOMP_RESTRICT_ADDRESS_FAMILIES_BROKEN
 335                 assert_se(fd >= 0);
 336                 safe_close(fd);
 337 #else
 338                 assert_se(fd < 0);
 339                 assert_se(errno == EAFNOSUPPORT);
 340 #endif
 341
 342                 _exit(EXIT_SUCCESS);
 343         }
 344
 345         assert_se(wait_for_terminate_and_warn("socketseccomp", pid, true) == EXIT_SUCCESS);
 346 }
 347
 348 static void test_restrict_realtime(void) {
 349         pid_t pid;
 350
 351         if (!is_seccomp_available())
 352                 return;
 353         if (geteuid() != 0)
 354                 return;
 355
 356         if (detect_container() > 0) /* in containers RT privs are likely missing anyway */
 357                 return;
 358
 359         pid = fork();
 360         assert_se(pid >= 0);
 361
 362         if (pid == 0) {
 363                 assert_se(sched_setscheduler(0, SCHED_FIFO, &(struct sched_param) { .sched_priority = 1 }) >= 0);
 364                 assert_se(sched_setscheduler(0, SCHED_RR, &(struct sched_param) { .sched_priority = 1 }) >= 0);
 365                 assert_se(sched_setscheduler(0, SCHED_IDLE, &(struct sched_param) { .sched_priority = 0 }) >= 0);
 366                 assert_se(sched_setscheduler(0, SCHED_BATCH, &(struct sched_param) { .sched_priority = 0 }) >= 0);
 367                 assert_se(sched_setscheduler(0, SCHED_OTHER, &(struct sched_param) {}) >= 0);
 368
 369                 assert_se(seccomp_restrict_realtime() >= 0);
 370
 371                 assert_se(sched_setscheduler(0, SCHED_IDLE, &(struct sched_param) { .sched_priority = 0 }) >= 0);
 372                 assert_se(sched_setscheduler(0, SCHED_BATCH, &(struct sched_param) { .sched_priority = 0 }) >= 0);
 373                 assert_se(sched_setscheduler(0, SCHED_OTHER, &(struct sched_param) {}) >= 0);
 374
 375                 assert_se(sched_setscheduler(0, SCHED_FIFO, &(struct sched_param) { .sched_priority = 1 }) < 0);
 376                 assert_se(errno == EPERM);
 377                 assert_se(sched_setscheduler(0, SCHED_RR, &(struct sched_param) { .sched_priority = 1 }) < 0);
 378                 assert_se(errno == EPERM);
 379
 380                 _exit(EXIT_SUCCESS);
 381         }
 382
 383         assert_se(wait_for_terminate_and_warn("realtimeseccomp", pid, true) == EXIT_SUCCESS);
 384 }
 385
 386 static void test_memory_deny_write_execute_mmap(void) {
 387         pid_t pid;
 388
 389         if (!is_seccomp_available())
 390                 return;
 391         if (geteuid() != 0)
 392                 return;
 393
 394         pid = fork();
 395         assert_se(pid >= 0);
 396
 397         if (pid == 0) {
 398                 void *p;
 399
 400                 p = mmap(NULL, page_size(), PROT_WRITE|PROT_EXEC, MAP_PRIVATE|MAP_ANONYMOUS, -1,0);
 401                 assert_se(p != MAP_FAILED);
 402                 assert_se(munmap(p, page_size()) >= 0);
 403
 404                 p = mmap(NULL, page_size(), PROT_WRITE|PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1,0);
 405                 assert_se(p != MAP_FAILED);
 406                 assert_se(munmap(p, page_size()) >= 0);
 407
 408                 assert_se(seccomp_memory_deny_write_execute() >= 0);
 409
 410                 p = mmap(NULL, page_size(), PROT_WRITE|PROT_EXEC, MAP_PRIVATE|MAP_ANONYMOUS, -1,0);
 411 #if defined(__x86_64__) || defined(__i386__) || defined(__powerpc64__) || defined(__arm__) || defined(__aarch64__)
 412                 assert_se(p == MAP_FAILED);
 413                 assert_se(errno == EPERM);
 414 #else /* unknown architectures */
 415                 assert_se(p != MAP_FAILED);
 416                 assert_se(munmap(p, page_size()) >= 0);
 417 #endif
 418
 419                 p = mmap(NULL, page_size(), PROT_WRITE|PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1,0);
 420                 assert_se(p != MAP_FAILED);
 421                 assert_se(munmap(p, page_size()) >= 0);
 422
 423                 _exit(EXIT_SUCCESS);
 424         }
 425
 426         assert_se(wait_for_terminate_and_warn("memoryseccomp-mmap", pid, true) == EXIT_SUCCESS);
 427 }
 428
 429 static void test_memory_deny_write_execute_shmat(void) {
 430         int shmid;
 431         pid_t pid;
 432
 433         if (!is_seccomp_available())
 434                 return;
 435         if (geteuid() != 0)
 436                 return;
 437
 438         shmid = shmget(IPC_PRIVATE, page_size(), 0);
 439         assert_se(shmid >= 0);
 440
 441         pid = fork();
 442         assert_se(pid >= 0);
 443
 444         if (pid == 0) {
 445                 void *p;
 446
 447                 p = shmat(shmid, NULL, 0);
 448                 assert_se(p != MAP_FAILED);
 449                 assert_se(shmdt(p) == 0);
 450
 451                 p = shmat(shmid, NULL, SHM_EXEC);
 452                 assert_se(p != MAP_FAILED);
 453                 assert_se(shmdt(p) == 0);
 454
 455                 assert_se(seccomp_memory_deny_write_execute() >= 0);
 456
 457                 p = shmat(shmid, NULL, SHM_EXEC);
 458 #if defined(__x86_64__) || defined(__arm__) || defined(__aarch64__)
 459                 assert_se(p == MAP_FAILED);
 460                 assert_se(errno == EPERM);
 461 #else /* __i386__, __powerpc64__, and "unknown" architectures */
 462                 assert_se(p != MAP_FAILED);
 463                 assert_se(shmdt(p) == 0);
 464 #endif
 465
 466                 p = shmat(shmid, NULL, 0);
 467                 assert_se(p != MAP_FAILED);
 468                 assert_se(shmdt(p) == 0);
 469
 470                 _exit(EXIT_SUCCESS);
 471         }
 472
 473         assert_se(wait_for_terminate_and_warn("memoryseccomp-shmat", pid, true) == EXIT_SUCCESS);
 474 }
 475
 476 static void test_restrict_archs(void) {
 477         pid_t pid;
 478
 479         if (!is_seccomp_available())
 480                 return;
 481         if (geteuid() != 0)
 482                 return;
 483
 484         pid = fork();
 485         assert_se(pid >= 0);
 486
 487         if (pid == 0) {
 488                 _cleanup_set_free_ Set *s = NULL;
 489
 490                 assert_se(access("/", F_OK) >= 0);
 491
 492                 assert_se(s = set_new(NULL));
 493
 494 #ifdef __x86_64__
 495                 assert_se(set_put(s, UINT32_TO_PTR(SCMP_ARCH_X86+1)) >= 0);
 496 #endif
 497                 assert_se(seccomp_restrict_archs(s) >= 0);
 498
 499                 assert_se(access("/", F_OK) >= 0);
 500                 assert_se(seccomp_restrict_archs(NULL) >= 0);
 501
 502                 assert_se(access("/", F_OK) >= 0);
 503
 504                 _exit(EXIT_SUCCESS);
 505         }
 506
 507         assert_se(wait_for_terminate_and_warn("archseccomp", pid, true) == EXIT_SUCCESS);
 508 }
 509
 510 static void test_load_syscall_filter_set_raw(void) {
 511         pid_t pid;
 512
 513         if (!is_seccomp_available())
 514                 return;
 515         if (geteuid() != 0)
 516                 return;
 517
 518         pid = fork();
 519         assert_se(pid >= 0);
 520
 521         if (pid == 0) {
 522                 _cleanup_set_free_ Set *s = NULL;
 523
 524                 assert_se(access("/", F_OK) >= 0);
 525                 assert_se(poll(NULL, 0, 0) == 0);
 526
 527                 assert_se(seccomp_load_syscall_filter_set_raw(SCMP_ACT_ALLOW, NULL, SCMP_ACT_KILL) >= 0);
 528                 assert_se(access("/", F_OK) >= 0);
 529                 assert_se(poll(NULL, 0, 0) == 0);
 530
 531                 assert_se(s = set_new(NULL));
 532                 assert_se(set_put(s, UINT32_TO_PTR(__NR_access + 1)) >= 0);
 533
 534                 assert_se(seccomp_load_syscall_filter_set_raw(SCMP_ACT_ALLOW, s, SCMP_ACT_ERRNO(EUCLEAN)) >= 0);
 535
 536                 assert_se(access("/", F_OK) < 0);
 537                 assert_se(errno == EUCLEAN);
 538
 539                 assert_se(poll(NULL, 0, 0) == 0);
 540
 541                 s = set_free(s);
 542
 543                 assert_se(s = set_new(NULL));
 544                 assert_se(set_put(s, UINT32_TO_PTR(__NR_poll + 1)) >= 0);
 545
 546                 assert_se(seccomp_load_syscall_filter_set_raw(SCMP_ACT_ALLOW, s, SCMP_ACT_ERRNO(EUNATCH)) >= 0);
 547
 548                 assert_se(access("/", F_OK) < 0);
 549                 assert_se(errno == EUCLEAN);
 550
 551                 assert_se(poll(NULL, 0, 0) < 0);
 552                 assert_se(errno == EUNATCH);
 553
 554                 _exit(EXIT_SUCCESS);
 555         }
 556
 557         assert_se(wait_for_terminate_and_warn("syscallrawseccomp", pid, true) == EXIT_SUCCESS);
 558 }
 559
 560 int main(int argc, char *argv[]) {
 561
 562         log_set_max_level(LOG_DEBUG);
 563
 564         test_seccomp_arch_to_string();
 565         test_architecture_table();
 566         test_syscall_filter_set_find();
 567         test_filter_sets();
 568         test_restrict_namespace();
 569         test_protect_sysctl();
 570         test_restrict_address_families();
 571         test_restrict_realtime();
 572         test_memory_deny_write_execute_mmap();
 573         test_memory_deny_write_execute_shmat();
 574         test_restrict_archs();
 575         test_load_syscall_filter_set_raw();
 576
 577         return 0;
 578 }