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 <unistd.h>
  25 #include <sys/poll.h>
  26
  27 #include "alloc-util.h"
  28 #include "fd-util.h"
  29 #include "macro.h"
  30 #include "missing.h"
  31 #include "nsflags.h"
  32 #include "process-util.h"
  33 #include "raw-clone.h"
  34 #include "seccomp-util.h"
  35 #include "set.h"
  36 #include "string-util.h"
  37 #include "util.h"
  38 #include "virt.h"
  39
  40 static void test_seccomp_arch_to_string(void) {
  41         uint32_t a, b;
  42         const char *name;
  43
  44         a = seccomp_arch_native();
  45         assert_se(a > 0);
  46         name = seccomp_arch_to_string(a);
  47         assert_se(name);
  48         assert_se(seccomp_arch_from_string(name, &b) >= 0);
  49         assert_se(a == b);
  50 }
  51
  52 static void test_architecture_table(void) {
  53         const char *n, *n2;
  54
  55         NULSTR_FOREACH(n,
  56                        "native\0"
  57                        "x86\0"
  58                        "x86-64\0"
  59                        "x32\0"
  60                        "arm\0"
  61                        "arm64\0"
  62                        "mips\0"
  63                        "mips64\0"
  64                        "mips64-n32\0"
  65                        "mips-le\0"
  66                        "mips64-le\0"
  67                        "mips64-le-n32\0"
  68                        "ppc\0"
  69                        "ppc64\0"
  70                        "ppc64-le\0"
  71                        "s390\0"
  72                        "s390x\0") {
  73                 uint32_t c;
  74
  75                 assert_se(seccomp_arch_from_string(n, &c) >= 0);
  76                 n2 = seccomp_arch_to_string(c);
  77                 log_info("seccomp-arch: %s → 0x%"PRIx32" → %s", n, c, n2);
  78                 assert_se(streq_ptr(n, n2));
  79         }
  80 }
  81
  82 static void test_syscall_filter_set_find(void) {
  83         assert_se(!syscall_filter_set_find(NULL));
  84         assert_se(!syscall_filter_set_find(""));
  85         assert_se(!syscall_filter_set_find("quux"));
  86         assert_se(!syscall_filter_set_find("@quux"));
  87
  88         assert_se(syscall_filter_set_find("@clock") == syscall_filter_sets + SYSCALL_FILTER_SET_CLOCK);
  89         assert_se(syscall_filter_set_find("@default") == syscall_filter_sets + SYSCALL_FILTER_SET_DEFAULT);
  90         assert_se(syscall_filter_set_find("@raw-io") == syscall_filter_sets + SYSCALL_FILTER_SET_RAW_IO);
  91 }
  92
  93 static void test_filter_sets(void) {
  94         unsigned i;
  95         int r;
  96
  97         if (!is_seccomp_available())
  98                 return;
  99         if (geteuid() != 0)
 100                 return;
 101
 102         for (i = 0; i < _SYSCALL_FILTER_SET_MAX; i++) {
 103                 pid_t pid;
 104
 105                 log_info("Testing %s", syscall_filter_sets[i].name);
 106
 107                 pid = fork();
 108                 assert_se(pid >= 0);
 109
 110                 if (pid == 0) { /* Child? */
 111                         int fd;
 112
 113                         if (i == SYSCALL_FILTER_SET_DEFAULT) /* if we look at the default set, whitelist instead of blacklist */
 114                                 r = seccomp_load_syscall_filter_set(SCMP_ACT_ERRNO(EUCLEAN), syscall_filter_sets + i, SCMP_ACT_ALLOW);
 115                         else
 116                                 r = seccomp_load_syscall_filter_set(SCMP_ACT_ALLOW, syscall_filter_sets + i, SCMP_ACT_ERRNO(EUCLEAN));
 117                         if (r < 0)
 118                                 _exit(EXIT_FAILURE);
 119
 120                         /* Test the sycall filter with one random system call */
 121                         fd = eventfd(0, EFD_NONBLOCK|EFD_CLOEXEC);
 122                         if (IN_SET(i, SYSCALL_FILTER_SET_IO_EVENT, SYSCALL_FILTER_SET_DEFAULT))
 123                                 assert_se(fd < 0 && errno == EUCLEAN);
 124                         else {
 125                                 assert_se(fd >= 0);
 126                                 safe_close(fd);
 127                         }
 128
 129                         _exit(EXIT_SUCCESS);
 130                 }
 131
 132                 assert_se(wait_for_terminate_and_warn(syscall_filter_sets[i].name, pid, true) == EXIT_SUCCESS);
 133         }
 134 }
 135
 136 static void test_restrict_namespace(void) {
 137         _cleanup_free_ char *s = NULL;
 138         unsigned long ul;
 139         pid_t pid;
 140
 141         assert_se(namespace_flag_to_string(0) == NULL);
 142         assert_se(streq(namespace_flag_to_string(CLONE_NEWNS), "mnt"));
 143         assert_se(namespace_flag_to_string(CLONE_NEWNS|CLONE_NEWIPC) == NULL);
 144         assert_se(streq(namespace_flag_to_string(CLONE_NEWCGROUP), "cgroup"));
 145
 146         assert_se(namespace_flag_from_string("mnt") == CLONE_NEWNS);
 147         assert_se(namespace_flag_from_string(NULL) == 0);
 148         assert_se(namespace_flag_from_string("") == 0);
 149         assert_se(namespace_flag_from_string("uts") == CLONE_NEWUTS);
 150         assert_se(namespace_flag_from_string(namespace_flag_to_string(CLONE_NEWUTS)) == CLONE_NEWUTS);
 151         assert_se(streq(namespace_flag_to_string(namespace_flag_from_string("ipc")), "ipc"));
 152
 153         assert_se(namespace_flag_from_string_many(NULL, &ul) == 0 && ul == 0);
 154         assert_se(namespace_flag_from_string_many("", &ul) == 0 && ul == 0);
 155         assert_se(namespace_flag_from_string_many("mnt uts ipc", &ul) == 0 && ul == (CLONE_NEWNS|CLONE_NEWUTS|CLONE_NEWIPC));
 156
 157         assert_se(namespace_flag_to_string_many(NAMESPACE_FLAGS_ALL, &s) == 0);
 158         assert_se(streq(s, "cgroup ipc net mnt pid user uts"));
 159         assert_se(namespace_flag_from_string_many(s, &ul) == 0 && ul == NAMESPACE_FLAGS_ALL);
 160
 161         if (!is_seccomp_available())
 162                 return;
 163         if (geteuid() != 0)
 164                 return;
 165
 166         pid = fork();
 167         assert_se(pid >= 0);
 168
 169         if (pid == 0) {
 170
 171                 assert_se(seccomp_restrict_namespaces(CLONE_NEWNS|CLONE_NEWNET) >= 0);
 172
 173                 assert_se(unshare(CLONE_NEWNS) == 0);
 174                 assert_se(unshare(CLONE_NEWNET) == 0);
 175                 assert_se(unshare(CLONE_NEWUTS) == -1);
 176                 assert_se(errno == EPERM);
 177                 assert_se(unshare(CLONE_NEWIPC) == -1);
 178                 assert_se(errno == EPERM);
 179                 assert_se(unshare(CLONE_NEWNET|CLONE_NEWUTS) == -1);
 180                 assert_se(errno == EPERM);
 181
 182                 /* We use fd 0 (stdin) here, which of course will fail with EINVAL on setns(). Except of course our
 183                  * seccomp filter worked, and hits first and makes it return EPERM */
 184                 assert_se(setns(0, CLONE_NEWNS) == -1);
 185                 assert_se(errno == EINVAL);
 186                 assert_se(setns(0, CLONE_NEWNET) == -1);
 187                 assert_se(errno == EINVAL);
 188                 assert_se(setns(0, CLONE_NEWUTS) == -1);
 189                 assert_se(errno == EPERM);
 190                 assert_se(setns(0, CLONE_NEWIPC) == -1);
 191                 assert_se(errno == EPERM);
 192                 assert_se(setns(0, CLONE_NEWNET|CLONE_NEWUTS) == -1);
 193                 assert_se(errno == EPERM);
 194                 assert_se(setns(0, 0) == -1);
 195                 assert_se(errno == EPERM);
 196
 197                 pid = raw_clone(CLONE_NEWNS);
 198                 assert_se(pid >= 0);
 199                 if (pid == 0)
 200                         _exit(EXIT_SUCCESS);
 201                 pid = raw_clone(CLONE_NEWNET);
 202                 assert_se(pid >= 0);
 203                 if (pid == 0)
 204                         _exit(EXIT_SUCCESS);
 205                 pid = raw_clone(CLONE_NEWUTS);
 206                 assert_se(pid < 0);
 207                 assert_se(errno == EPERM);
 208                 pid = raw_clone(CLONE_NEWIPC);
 209                 assert_se(pid < 0);
 210                 assert_se(errno == EPERM);
 211                 pid = raw_clone(CLONE_NEWNET|CLONE_NEWUTS);
 212                 assert_se(pid < 0);
 213                 assert_se(errno == EPERM);
 214
 215                 _exit(EXIT_SUCCESS);
 216         }
 217
 218         assert_se(wait_for_terminate_and_warn("nsseccomp", pid, true) == EXIT_SUCCESS);
 219 }
 220
 221 static void test_protect_sysctl(void) {
 222         pid_t pid;
 223
 224         if (!is_seccomp_available())
 225                 return;
 226         if (geteuid() != 0)
 227                 return;
 228
 229         if (detect_container() > 0) /* in containers _sysctl() is likely missing anyway */
 230                 return;
 231
 232         pid = fork();
 233         assert_se(pid >= 0);
 234
 235         if (pid == 0) {
 236                 assert_se(syscall(__NR__sysctl, NULL) < 0);
 237                 assert_se(errno == EFAULT);
 238
 239                 assert_se(seccomp_protect_sysctl() >= 0);
 240
 241                 assert_se(syscall(__NR__sysctl, 0, 0, 0) < 0);
 242                 assert_se(errno == EPERM);
 243
 244                 _exit(EXIT_SUCCESS);
 245         }
 246
 247         assert_se(wait_for_terminate_and_warn("sysctlseccomp", pid, true) == EXIT_SUCCESS);
 248 }
 249
 250 static void test_restrict_address_families(void) {
 251         pid_t pid;
 252
 253         if (!is_seccomp_available())
 254                 return;
 255         if (geteuid() != 0)
 256                 return;
 257
 258         pid = fork();
 259         assert_se(pid >= 0);
 260
 261         if (pid == 0) {
 262                 int fd;
 263                 Set *s;
 264
 265                 fd = socket(AF_INET, SOCK_DGRAM, 0);
 266                 assert_se(fd >= 0);
 267                 safe_close(fd);
 268
 269                 fd = socket(AF_UNIX, SOCK_DGRAM, 0);
 270                 assert_se(fd >= 0);
 271                 safe_close(fd);
 272
 273                 fd = socket(AF_NETLINK, SOCK_DGRAM, 0);
 274                 assert_se(fd >= 0);
 275                 safe_close(fd);
 276
 277                 assert_se(s = set_new(NULL));
 278                 assert_se(set_put(s, INT_TO_PTR(AF_UNIX)) >= 0);
 279
 280                 assert_se(seccomp_restrict_address_families(s, false) >= 0);
 281
 282                 fd = socket(AF_INET, SOCK_DGRAM, 0);
 283                 assert_se(fd >= 0);
 284                 safe_close(fd);
 285
 286                 assert_se(socket(AF_UNIX, SOCK_DGRAM, 0) < 0);
 287                 assert_se(errno == EAFNOSUPPORT);
 288
 289                 fd = socket(AF_NETLINK, SOCK_DGRAM, 0);
 290                 assert_se(fd >= 0);
 291                 safe_close(fd);
 292
 293                 set_clear(s);
 294
 295                 assert_se(set_put(s, INT_TO_PTR(AF_INET)) >= 0);
 296
 297                 assert_se(seccomp_restrict_address_families(s, true) >= 0);
 298
 299                 fd = socket(AF_INET, SOCK_DGRAM, 0);
 300                 assert_se(fd >= 0);
 301                 safe_close(fd);
 302
 303                 assert_se(socket(AF_UNIX, SOCK_DGRAM, 0) < 0);
 304                 assert_se(errno == EAFNOSUPPORT);
 305
 306                 assert_se(socket(AF_NETLINK, SOCK_DGRAM, 0) < 0);
 307                 assert_se(errno == EAFNOSUPPORT);
 308
 309                 _exit(EXIT_SUCCESS);
 310         }
 311
 312         assert_se(wait_for_terminate_and_warn("socketseccomp", pid, true) == EXIT_SUCCESS);
 313 }
 314
 315 static void test_restrict_realtime(void) {
 316         pid_t pid;
 317
 318         if (!is_seccomp_available())
 319                 return;
 320         if (geteuid() != 0)
 321                 return;
 322
 323         if (detect_container() > 0) /* in containers RT privs are likely missing anyway */
 324                 return;
 325
 326         pid = fork();
 327         assert_se(pid >= 0);
 328
 329         if (pid == 0) {
 330                 assert_se(sched_setscheduler(0, SCHED_FIFO, &(struct sched_param) { .sched_priority = 1 }) >= 0);
 331                 assert_se(sched_setscheduler(0, SCHED_RR, &(struct sched_param) { .sched_priority = 1 }) >= 0);
 332                 assert_se(sched_setscheduler(0, SCHED_IDLE, &(struct sched_param) { .sched_priority = 0 }) >= 0);
 333                 assert_se(sched_setscheduler(0, SCHED_BATCH, &(struct sched_param) { .sched_priority = 0 }) >= 0);
 334                 assert_se(sched_setscheduler(0, SCHED_OTHER, &(struct sched_param) {}) >= 0);
 335
 336                 assert_se(seccomp_restrict_realtime() >= 0);
 337
 338                 assert_se(sched_setscheduler(0, SCHED_IDLE, &(struct sched_param) { .sched_priority = 0 }) >= 0);
 339                 assert_se(sched_setscheduler(0, SCHED_BATCH, &(struct sched_param) { .sched_priority = 0 }) >= 0);
 340                 assert_se(sched_setscheduler(0, SCHED_OTHER, &(struct sched_param) {}) >= 0);
 341
 342                 assert_se(sched_setscheduler(0, SCHED_FIFO, &(struct sched_param) { .sched_priority = 1 }) < 0);
 343                 assert_se(errno == EPERM);
 344                 assert_se(sched_setscheduler(0, SCHED_RR, &(struct sched_param) { .sched_priority = 1 }) < 0);
 345                 assert_se(errno == EPERM);
 346
 347                 _exit(EXIT_SUCCESS);
 348         }
 349
 350         assert_se(wait_for_terminate_and_warn("realtimeseccomp", pid, true) == EXIT_SUCCESS);
 351 }
 352
 353 static void test_memory_deny_write_execute(void) {
 354         pid_t pid;
 355
 356         if (!is_seccomp_available())
 357                 return;
 358         if (geteuid() != 0)
 359                 return;
 360
 361         pid = fork();
 362         assert_se(pid >= 0);
 363
 364         if (pid == 0) {
 365                 void *p;
 366
 367                 p = mmap(NULL, page_size(), PROT_WRITE|PROT_EXEC, MAP_PRIVATE|MAP_ANONYMOUS, -1,0);
 368                 assert_se(p != MAP_FAILED);
 369                 assert_se(munmap(p, page_size()) >= 0);
 370
 371                 seccomp_memory_deny_write_execute();
 372
 373                 p = mmap(NULL, page_size(), PROT_WRITE|PROT_EXEC, MAP_PRIVATE|MAP_ANONYMOUS, -1,0);
 374                 assert_se(p == MAP_FAILED);
 375                 assert_se(errno == EPERM);
 376
 377                 p = mmap(NULL, page_size(), PROT_WRITE|PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1,0);
 378                 assert_se(p != MAP_FAILED);
 379                 assert_se(munmap(p, page_size()) >= 0);
 380
 381                 _exit(EXIT_SUCCESS);
 382         }
 383
 384         assert_se(wait_for_terminate_and_warn("memoryseccomp", pid, true) == EXIT_SUCCESS);
 385 }
 386
 387 static void test_restrict_archs(void) {
 388         pid_t pid;
 389
 390         if (!is_seccomp_available())
 391                 return;
 392         if (geteuid() != 0)
 393                 return;
 394
 395         pid = fork();
 396         assert_se(pid >= 0);
 397
 398         if (pid == 0) {
 399                 _cleanup_set_free_ Set *s = NULL;
 400
 401                 assert_se(access("/", F_OK) >= 0);
 402
 403                 assert_se(s = set_new(NULL));
 404
 405 #ifdef __x86_64__
 406                 assert_se(set_put(s, UINT32_TO_PTR(SCMP_ARCH_X86+1)) >= 0);
 407 #endif
 408                 assert_se(seccomp_restrict_archs(s) >= 0);
 409
 410                 assert_se(access("/", F_OK) >= 0);
 411                 assert_se(seccomp_restrict_archs(NULL) >= 0);
 412
 413                 assert_se(access("/", F_OK) >= 0);
 414
 415                 _exit(EXIT_SUCCESS);
 416         }
 417
 418         assert_se(wait_for_terminate_and_warn("archseccomp", pid, true) == EXIT_SUCCESS);
 419 }
 420
 421 static void test_load_syscall_filter_set_raw(void) {
 422         pid_t pid;
 423
 424         if (!is_seccomp_available())
 425                 return;
 426         if (geteuid() != 0)
 427                 return;
 428
 429         pid = fork();
 430         assert_se(pid >= 0);
 431
 432         if (pid == 0) {
 433                 _cleanup_set_free_ Set *s = NULL;
 434
 435                 assert_se(access("/", F_OK) >= 0);
 436                 assert_se(poll(NULL, 0, 0) == 0);
 437
 438                 assert_se(seccomp_load_syscall_filter_set_raw(SCMP_ACT_ALLOW, NULL, SCMP_ACT_KILL) >= 0);
 439                 assert_se(access("/", F_OK) >= 0);
 440                 assert_se(poll(NULL, 0, 0) == 0);
 441
 442                 assert_se(s = set_new(NULL));
 443                 assert_se(set_put(s, UINT32_TO_PTR(__NR_access + 1)) >= 0);
 444
 445                 assert_se(seccomp_load_syscall_filter_set_raw(SCMP_ACT_ALLOW, s, SCMP_ACT_ERRNO(EUCLEAN)) >= 0);
 446
 447                 assert_se(access("/", F_OK) < 0);
 448                 assert_se(errno == EUCLEAN);
 449
 450                 assert_se(poll(NULL, 0, 0) == 0);
 451
 452                 s = set_free(s);
 453
 454                 assert_se(s = set_new(NULL));
 455                 assert_se(set_put(s, UINT32_TO_PTR(__NR_poll + 1)) >= 0);
 456
 457                 assert_se(seccomp_load_syscall_filter_set_raw(SCMP_ACT_ALLOW, s, SCMP_ACT_ERRNO(EUNATCH)) >= 0);
 458
 459                 assert_se(access("/", F_OK) < 0);
 460                 assert_se(errno == EUCLEAN);
 461
 462                 assert_se(poll(NULL, 0, 0) < 0);
 463                 assert_se(errno == EUNATCH);
 464
 465                 _exit(EXIT_SUCCESS);
 466         }
 467
 468         assert_se(wait_for_terminate_and_warn("syscallrawseccomp", pid, true) == EXIT_SUCCESS);
 469 }
 470
 471 int main(int argc, char *argv[]) {
 472
 473         log_set_max_level(LOG_DEBUG);
 474
 475         test_seccomp_arch_to_string();
 476         test_architecture_table();
 477         test_syscall_filter_set_find();
 478         test_filter_sets();
 479         test_restrict_namespace();
 480         test_protect_sysctl();
 481         test_restrict_address_families();
 482         test_restrict_realtime();
 483         test_memory_deny_write_execute();
 484         test_restrict_archs();
 485         test_load_syscall_filter_set_raw();
 486
 487         return 0;
 488 }