1 /* 2 * Copyright (c) 2012 Will Drewry <wad@dataspill.org> 3 * 4 * Permission to use, copy, modify, and distribute this software for any 5 * purpose with or without fee is hereby granted, provided that the above 6 * copyright notice and this permission notice appear in all copies. 7 * 8 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 9 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 10 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 11 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 12 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 13 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 14 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 15 */ 16 17 /* 18 * Uncomment the SANDBOX_SECCOMP_FILTER_DEBUG macro below to help diagnose 19 * filter breakage during development. *Do not* use this in production, 20 * as it relies on making library calls that are unsafe in signal context. 21 * 22 * Instead, live systems the auditctl(8) may be used to monitor failures. 23 * E.g. 24 * auditctl -a task,always -F uid=<privsep uid> 25 */ 26 /* #define SANDBOX_SECCOMP_FILTER_DEBUG 1 */ 27 28 /* XXX it should be possible to do logging via the log socket safely */ 29 30 #ifdef SANDBOX_SECCOMP_FILTER_DEBUG 31 /* Use the kernel headers in case of an older toolchain. */ 32 # include <asm/siginfo.h> 33 # define __have_siginfo_t 1 34 # define __have_sigval_t 1 35 # define __have_sigevent_t 1 36 #endif /* SANDBOX_SECCOMP_FILTER_DEBUG */ 37 38 #include "includes.h" 39 40 #ifdef SANDBOX_SECCOMP_FILTER 41 42 #include <sys/types.h> 43 #include <sys/resource.h> 44 #include <sys/prctl.h> 45 #include <sys/mman.h> 46 #include <sys/syscall.h> 47 48 #include <linux/net.h> 49 #include <linux/audit.h> 50 #include <linux/filter.h> 51 #include <linux/seccomp.h> 52 #include <elf.h> 53 54 #include <asm/unistd.h> 55 #ifdef __s390__ 56 #include <asm/zcrypt.h> 57 #endif 58 59 #include <errno.h> 60 #include <signal.h> 61 #include <stdarg.h> 62 #include <stddef.h> /* for offsetof */ 63 #include <stdio.h> 64 #include <stdlib.h> 65 #include <string.h> 66 #include <unistd.h> 67 68 #include "log.h" 69 #include "ssh-sandbox.h" 70 #include "xmalloc.h" 71 72 /* Linux seccomp_filter sandbox */ 73 #define SECCOMP_FILTER_FAIL SECCOMP_RET_KILL 74 75 /* Use a signal handler to emit violations when debugging */ 76 #ifdef SANDBOX_SECCOMP_FILTER_DEBUG 77 # undef SECCOMP_FILTER_FAIL 78 # define SECCOMP_FILTER_FAIL SECCOMP_RET_TRAP 79 #endif /* SANDBOX_SECCOMP_FILTER_DEBUG */ 80 81 #if __BYTE_ORDER == __LITTLE_ENDIAN 82 # define ARG_LO_OFFSET 0 83 # define ARG_HI_OFFSET sizeof(uint32_t) 84 #elif __BYTE_ORDER == __BIG_ENDIAN 85 # define ARG_LO_OFFSET sizeof(uint32_t) 86 # define ARG_HI_OFFSET 0 87 #else 88 #error "Unknown endianness" 89 #endif 90 91 /* Simple helpers to avoid manual errors (but larger BPF programs). */ 92 #define SC_DENY(_nr, _errno) \ 93 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, (_nr), 0, 1), \ 94 BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ERRNO|(_errno)) 95 #define SC_ALLOW(_nr) \ 96 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, (_nr), 0, 1), \ 97 BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW) 98 #define SC_ALLOW_ARG(_nr, _arg_nr, _arg_val) \ 99 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, (_nr), 0, 6), \ 100 /* load and test syscall argument, low word */ \ 101 BPF_STMT(BPF_LD+BPF_W+BPF_ABS, \ 102 offsetof(struct seccomp_data, args[(_arg_nr)]) + ARG_LO_OFFSET), \ 103 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, \ 104 ((_arg_val) & 0xFFFFFFFF), 0, 3), \ 105 /* load and test syscall argument, high word */ \ 106 BPF_STMT(BPF_LD+BPF_W+BPF_ABS, \ 107 offsetof(struct seccomp_data, args[(_arg_nr)]) + ARG_HI_OFFSET), \ 108 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, \ 109 (((uint32_t)((uint64_t)(_arg_val) >> 32)) & 0xFFFFFFFF), 0, 1), \ 110 BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW), \ 111 /* reload syscall number; all rules expect it in accumulator */ \ 112 BPF_STMT(BPF_LD+BPF_W+BPF_ABS, \ 113 offsetof(struct seccomp_data, nr)) 114 /* Allow if syscall argument contains only values in mask */ 115 #define SC_ALLOW_ARG_MASK(_nr, _arg_nr, _arg_mask) \ 116 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, (_nr), 0, 8), \ 117 /* load, mask and test syscall argument, low word */ \ 118 BPF_STMT(BPF_LD+BPF_W+BPF_ABS, \ 119 offsetof(struct seccomp_data, args[(_arg_nr)]) + ARG_LO_OFFSET), \ 120 BPF_STMT(BPF_ALU+BPF_AND+BPF_K, ~((_arg_mask) & 0xFFFFFFFF)), \ 121 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, 0, 0, 4), \ 122 /* load, mask and test syscall argument, high word */ \ 123 BPF_STMT(BPF_LD+BPF_W+BPF_ABS, \ 124 offsetof(struct seccomp_data, args[(_arg_nr)]) + ARG_HI_OFFSET), \ 125 BPF_STMT(BPF_ALU+BPF_AND+BPF_K, \ 126 ~(((uint32_t)((uint64_t)(_arg_mask) >> 32)) & 0xFFFFFFFF)), \ 127 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, 0, 0, 1), \ 128 BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW), \ 129 /* reload syscall number; all rules expect it in accumulator */ \ 130 BPF_STMT(BPF_LD+BPF_W+BPF_ABS, \ 131 offsetof(struct seccomp_data, nr)) 132 133 /* Syscall filtering set for preauth. */ 134 static const struct sock_filter preauth_insns[] = { 135 /* Ensure the syscall arch convention is as expected. */ 136 BPF_STMT(BPF_LD+BPF_W+BPF_ABS, 137 offsetof(struct seccomp_data, arch)), 138 BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, SECCOMP_AUDIT_ARCH, 1, 0), 139 BPF_STMT(BPF_RET+BPF_K, SECCOMP_FILTER_FAIL), 140 /* Load the syscall number for checking. */ 141 BPF_STMT(BPF_LD+BPF_W+BPF_ABS, 142 offsetof(struct seccomp_data, nr)), 143 144 /* Syscalls to non-fatally deny */ 145 #ifdef __NR_lstat 146 SC_DENY(__NR_lstat, EACCES), 147 #endif 148 #ifdef __NR_lstat64 149 SC_DENY(__NR_lstat64, EACCES), 150 #endif 151 #ifdef __NR_fstat 152 SC_DENY(__NR_fstat, EACCES), 153 #endif 154 #ifdef __NR_fstat64 155 SC_DENY(__NR_fstat64, EACCES), 156 #endif 157 #ifdef __NR_fstatat64 158 SC_DENY(__NR_fstatat64, EACCES), 159 #endif 160 #ifdef __NR_open 161 SC_DENY(__NR_open, EACCES), 162 #endif 163 #ifdef __NR_openat 164 SC_DENY(__NR_openat, EACCES), 165 #endif 166 #ifdef __NR_newfstatat 167 SC_DENY(__NR_newfstatat, EACCES), 168 #endif 169 #ifdef __NR_stat 170 SC_DENY(__NR_stat, EACCES), 171 #endif 172 #ifdef __NR_stat64 173 SC_DENY(__NR_stat64, EACCES), 174 #endif 175 #ifdef __NR_shmget 176 SC_DENY(__NR_shmget, EACCES), 177 #endif 178 #ifdef __NR_shmat 179 SC_DENY(__NR_shmat, EACCES), 180 #endif 181 #ifdef __NR_shmdt 182 SC_DENY(__NR_shmdt, EACCES), 183 #endif 184 #ifdef __NR_ipc 185 SC_DENY(__NR_ipc, EACCES), 186 #endif 187 #ifdef __NR_statx 188 SC_DENY(__NR_statx, EACCES), 189 #endif 190 191 /* Syscalls to permit */ 192 #ifdef __NR_brk 193 SC_ALLOW(__NR_brk), 194 #endif 195 #ifdef __NR_clock_gettime 196 SC_ALLOW(__NR_clock_gettime), 197 #endif 198 #ifdef __NR_clock_gettime64 199 SC_ALLOW(__NR_clock_gettime64), 200 #endif 201 #ifdef __NR_close 202 SC_ALLOW(__NR_close), 203 #endif 204 #ifdef __NR_exit 205 SC_ALLOW(__NR_exit), 206 #endif 207 #ifdef __NR_exit_group 208 SC_ALLOW(__NR_exit_group), 209 #endif 210 #ifdef __NR_futex 211 SC_ALLOW(__NR_futex), 212 #endif 213 #ifdef __NR_futex_time64 214 SC_ALLOW(__NR_futex_time64), 215 #endif 216 #ifdef __NR_geteuid 217 SC_ALLOW(__NR_geteuid), 218 #endif 219 #ifdef __NR_geteuid32 220 SC_ALLOW(__NR_geteuid32), 221 #endif 222 #ifdef __NR_getpgid 223 SC_ALLOW(__NR_getpgid), 224 #endif 225 #ifdef __NR_getpid 226 SC_ALLOW(__NR_getpid), 227 #endif 228 #ifdef __NR_getrandom 229 SC_ALLOW(__NR_getrandom), 230 #endif 231 #ifdef __NR_gettid 232 SC_ALLOW(__NR_gettid), 233 #endif 234 #ifdef __NR_gettimeofday 235 SC_ALLOW(__NR_gettimeofday), 236 #endif 237 #ifdef __NR_getuid 238 SC_ALLOW(__NR_getuid), 239 #endif 240 #ifdef __NR_getuid32 241 SC_ALLOW(__NR_getuid32), 242 #endif 243 #ifdef __NR_madvise 244 SC_ALLOW(__NR_madvise), 245 #endif 246 #ifdef __NR_mmap 247 SC_ALLOW_ARG_MASK(__NR_mmap, 2, PROT_READ|PROT_WRITE|PROT_NONE), 248 #endif 249 #ifdef __NR_mmap2 250 SC_ALLOW_ARG_MASK(__NR_mmap2, 2, PROT_READ|PROT_WRITE|PROT_NONE), 251 #endif 252 #ifdef __NR_mprotect 253 SC_ALLOW_ARG_MASK(__NR_mprotect, 2, PROT_READ|PROT_WRITE|PROT_NONE), 254 #endif 255 #ifdef __NR_mremap 256 SC_ALLOW(__NR_mremap), 257 #endif 258 #ifdef __NR_munmap 259 SC_ALLOW(__NR_munmap), 260 #endif 261 #ifdef __NR_nanosleep 262 SC_ALLOW(__NR_nanosleep), 263 #endif 264 #ifdef __NR_clock_nanosleep 265 SC_ALLOW(__NR_clock_nanosleep), 266 #endif 267 #ifdef __NR_clock_nanosleep_time64 268 SC_ALLOW(__NR_clock_nanosleep_time64), 269 #endif 270 #ifdef __NR_clock_gettime64 271 SC_ALLOW(__NR_clock_gettime64), 272 #endif 273 #ifdef __NR__newselect 274 SC_ALLOW(__NR__newselect), 275 #endif 276 #ifdef __NR_ppoll 277 SC_ALLOW(__NR_ppoll), 278 #endif 279 #ifdef __NR_ppoll_time64 280 SC_ALLOW(__NR_ppoll_time64), 281 #endif 282 #ifdef __NR_poll 283 SC_ALLOW(__NR_poll), 284 #endif 285 #ifdef __NR_pselect6 286 SC_ALLOW(__NR_pselect6), 287 #endif 288 #ifdef __NR_pselect6_time64 289 SC_ALLOW(__NR_pselect6_time64), 290 #endif 291 #ifdef __NR_read 292 SC_ALLOW(__NR_read), 293 #endif 294 #ifdef __NR_rt_sigprocmask 295 SC_ALLOW(__NR_rt_sigprocmask), 296 #endif 297 #ifdef __NR_select 298 SC_ALLOW(__NR_select), 299 #endif 300 #ifdef __NR_shutdown 301 SC_ALLOW(__NR_shutdown), 302 #endif 303 #ifdef __NR_sigprocmask 304 SC_ALLOW(__NR_sigprocmask), 305 #endif 306 #ifdef __NR_time 307 SC_ALLOW(__NR_time), 308 #endif 309 #ifdef __NR_write 310 SC_ALLOW(__NR_write), 311 #endif 312 #ifdef __NR_socketcall 313 SC_ALLOW_ARG(__NR_socketcall, 0, SYS_SHUTDOWN), 314 SC_DENY(__NR_socketcall, EACCES), 315 #endif 316 #if defined(__NR_ioctl) && defined(__s390__) 317 /* Allow ioctls for ICA crypto card on s390 */ 318 SC_ALLOW_ARG(__NR_ioctl, 1, Z90STAT_STATUS_MASK), 319 SC_ALLOW_ARG(__NR_ioctl, 1, ICARSAMODEXPO), 320 SC_ALLOW_ARG(__NR_ioctl, 1, ICARSACRT), 321 SC_ALLOW_ARG(__NR_ioctl, 1, ZSECSENDCPRB), 322 /* Allow ioctls for EP11 crypto card on s390 */ 323 SC_ALLOW_ARG(__NR_ioctl, 1, ZSENDEP11CPRB), 324 #endif 325 #if defined(__x86_64__) && defined(__ILP32__) && defined(__X32_SYSCALL_BIT) 326 /* 327 * On Linux x32, the clock_gettime VDSO falls back to the 328 * x86-64 syscall under some circumstances, e.g. 329 * https://bugs.debian.org/849923 330 */ 331 SC_ALLOW(__NR_clock_gettime & ~__X32_SYSCALL_BIT), 332 #endif 333 334 /* Default deny */ 335 BPF_STMT(BPF_RET+BPF_K, SECCOMP_FILTER_FAIL), 336 }; 337 338 static const struct sock_fprog preauth_program = { 339 .len = (unsigned short)(sizeof(preauth_insns)/sizeof(preauth_insns[0])), 340 .filter = (struct sock_filter *)preauth_insns, 341 }; 342 343 struct ssh_sandbox { 344 pid_t child_pid; 345 }; 346 347 struct ssh_sandbox * 348 ssh_sandbox_init(struct monitor *monitor) 349 { 350 struct ssh_sandbox *box; 351 352 /* 353 * Strictly, we don't need to maintain any state here but we need 354 * to return non-NULL to satisfy the API. 355 */ 356 debug3("%s: preparing seccomp filter sandbox", __func__); 357 box = xcalloc(1, sizeof(*box)); 358 box->child_pid = 0; 359 360 return box; 361 } 362 363 #ifdef SANDBOX_SECCOMP_FILTER_DEBUG 364 extern struct monitor *pmonitor; 365 void mm_log_handler(LogLevel level, const char *msg, void *ctx); 366 367 static void 368 ssh_sandbox_violation(int signum, siginfo_t *info, void *void_context) 369 { 370 char msg[256]; 371 372 snprintf(msg, sizeof(msg), 373 "%s: unexpected system call (arch:0x%x,syscall:%d @ %p)", 374 __func__, info->si_arch, info->si_syscall, info->si_call_addr); 375 mm_log_handler(SYSLOG_LEVEL_FATAL, msg, pmonitor); 376 _exit(1); 377 } 378 379 static void 380 ssh_sandbox_child_debugging(void) 381 { 382 struct sigaction act; 383 sigset_t mask; 384 385 debug3("%s: installing SIGSYS handler", __func__); 386 memset(&act, 0, sizeof(act)); 387 sigemptyset(&mask); 388 sigaddset(&mask, SIGSYS); 389 390 act.sa_sigaction = &ssh_sandbox_violation; 391 act.sa_flags = SA_SIGINFO; 392 if (sigaction(SIGSYS, &act, NULL) == -1) 393 fatal("%s: sigaction(SIGSYS): %s", __func__, strerror(errno)); 394 if (sigprocmask(SIG_UNBLOCK, &mask, NULL) == -1) 395 fatal("%s: sigprocmask(SIGSYS): %s", 396 __func__, strerror(errno)); 397 } 398 #endif /* SANDBOX_SECCOMP_FILTER_DEBUG */ 399 400 void 401 ssh_sandbox_child(struct ssh_sandbox *box) 402 { 403 struct rlimit rl_zero, rl_one = {.rlim_cur = 1, .rlim_max = 1}; 404 int nnp_failed = 0; 405 406 /* Set rlimits for completeness if possible. */ 407 rl_zero.rlim_cur = rl_zero.rlim_max = 0; 408 if (setrlimit(RLIMIT_FSIZE, &rl_zero) == -1) 409 fatal("%s: setrlimit(RLIMIT_FSIZE, { 0, 0 }): %s", 410 __func__, strerror(errno)); 411 /* 412 * Cannot use zero for nfds, because poll(2) will fail with 413 * errno=EINVAL if npfds>RLIMIT_NOFILE. 414 */ 415 if (setrlimit(RLIMIT_NOFILE, &rl_one) == -1) 416 fatal("%s: setrlimit(RLIMIT_NOFILE, { 0, 0 }): %s", 417 __func__, strerror(errno)); 418 if (setrlimit(RLIMIT_NPROC, &rl_zero) == -1) 419 fatal("%s: setrlimit(RLIMIT_NPROC, { 0, 0 }): %s", 420 __func__, strerror(errno)); 421 422 #ifdef SANDBOX_SECCOMP_FILTER_DEBUG 423 ssh_sandbox_child_debugging(); 424 #endif /* SANDBOX_SECCOMP_FILTER_DEBUG */ 425 426 debug3("%s: setting PR_SET_NO_NEW_PRIVS", __func__); 427 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) == -1) { 428 debug("%s: prctl(PR_SET_NO_NEW_PRIVS): %s", 429 __func__, strerror(errno)); 430 nnp_failed = 1; 431 } 432 debug3("%s: attaching seccomp filter program", __func__); 433 if (prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &preauth_program) == -1) 434 debug("%s: prctl(PR_SET_SECCOMP): %s", 435 __func__, strerror(errno)); 436 else if (nnp_failed) 437 fatal("%s: SECCOMP_MODE_FILTER activated but " 438 "PR_SET_NO_NEW_PRIVS failed", __func__); 439 } 440 441 void 442 ssh_sandbox_parent_finish(struct ssh_sandbox *box) 443 { 444 free(box); 445 debug3("%s: finished", __func__); 446 } 447 448 void 449 ssh_sandbox_parent_preauth(struct ssh_sandbox *box, pid_t child_pid) 450 { 451 box->child_pid = child_pid; 452 } 453 454 #endif /* SANDBOX_SECCOMP_FILTER */ 455