1 /* SPDX-License-Identifier: LGPL-2.1+ */
5 #include <linux/seccomp.h>
14 #include "alloc-util.h"
15 #include "errno-list.h"
18 #include "nulstr-util.h"
19 #include "process-util.h"
20 #include "seccomp-util.h"
22 #include "string-util.h"
25 const uint32_t seccomp_local_archs
[] = {
27 /* Note: always list the native arch we are compiled as last, so that users can blacklist seccomp(), but our own calls to it still succeed */
29 #if defined(__x86_64__) && defined(__ILP32__)
32 SCMP_ARCH_X32
, /* native */
33 #elif defined(__x86_64__) && !defined(__ILP32__)
36 SCMP_ARCH_X86_64
, /* native */
37 #elif defined(__i386__)
39 #elif defined(__aarch64__)
41 SCMP_ARCH_AARCH64
, /* native */
42 #elif defined(__arm__)
44 #elif defined(__mips__) && __BYTE_ORDER == __BIG_ENDIAN && _MIPS_SIM == _MIPS_SIM_ABI32
46 SCMP_ARCH_MIPS
, /* native */
47 #elif defined(__mips__) && __BYTE_ORDER == __LITTLE_ENDIAN && _MIPS_SIM == _MIPS_SIM_ABI32
49 SCMP_ARCH_MIPSEL
, /* native */
50 #elif defined(__mips__) && __BYTE_ORDER == __BIG_ENDIAN && _MIPS_SIM == _MIPS_SIM_ABI64
53 SCMP_ARCH_MIPSEL64N32
,
56 SCMP_ARCH_MIPS64
, /* native */
57 #elif defined(__mips__) && __BYTE_ORDER == __LITTLE_ENDIAN && _MIPS_SIM == _MIPS_SIM_ABI64
61 SCMP_ARCH_MIPSEL64N32
,
63 SCMP_ARCH_MIPSEL64
, /* native */
64 #elif defined(__mips__) && __BYTE_ORDER == __BIG_ENDIAN && _MIPS_SIM == _MIPS_SIM_NABI32
69 SCMP_ARCH_MIPSEL64N32
,
70 SCMP_ARCH_MIPS64N32
, /* native */
71 #elif defined(__mips__) && __BYTE_ORDER == __LITTLE_ENDIAN && _MIPS_SIM == _MIPS_SIM_NABI32
77 SCMP_ARCH_MIPSEL64N32
, /* native */
78 #elif defined(__powerpc64__) && __BYTE_ORDER == __BIG_ENDIAN
81 SCMP_ARCH_PPC64
, /* native */
82 #elif defined(__powerpc64__) && __BYTE_ORDER == __LITTLE_ENDIAN
85 SCMP_ARCH_PPC64LE
, /* native */
86 #elif defined(__powerpc__)
88 #elif defined(__s390x__)
90 SCMP_ARCH_S390X
, /* native */
91 #elif defined(__s390__)
97 const char* seccomp_arch_to_string(uint32_t c
) {
98 /* Maintain order used in <seccomp.h>.
100 * Names used here should be the same as those used for ConditionArchitecture=,
101 * except for "subarchitectures" like x32. */
104 case SCMP_ARCH_NATIVE
:
108 case SCMP_ARCH_X86_64
:
114 case SCMP_ARCH_AARCH64
:
118 case SCMP_ARCH_MIPS64
:
120 case SCMP_ARCH_MIPS64N32
:
122 case SCMP_ARCH_MIPSEL
:
124 case SCMP_ARCH_MIPSEL64
:
126 case SCMP_ARCH_MIPSEL64N32
:
127 return "mips64-le-n32";
130 case SCMP_ARCH_PPC64
:
132 case SCMP_ARCH_PPC64LE
:
136 case SCMP_ARCH_S390X
:
143 int seccomp_arch_from_string(const char *n
, uint32_t *ret
) {
149 if (streq(n
, "native"))
150 *ret
= SCMP_ARCH_NATIVE
;
151 else if (streq(n
, "x86"))
152 *ret
= SCMP_ARCH_X86
;
153 else if (streq(n
, "x86-64"))
154 *ret
= SCMP_ARCH_X86_64
;
155 else if (streq(n
, "x32"))
156 *ret
= SCMP_ARCH_X32
;
157 else if (streq(n
, "arm"))
158 *ret
= SCMP_ARCH_ARM
;
159 else if (streq(n
, "arm64"))
160 *ret
= SCMP_ARCH_AARCH64
;
161 else if (streq(n
, "mips"))
162 *ret
= SCMP_ARCH_MIPS
;
163 else if (streq(n
, "mips64"))
164 *ret
= SCMP_ARCH_MIPS64
;
165 else if (streq(n
, "mips64-n32"))
166 *ret
= SCMP_ARCH_MIPS64N32
;
167 else if (streq(n
, "mips-le"))
168 *ret
= SCMP_ARCH_MIPSEL
;
169 else if (streq(n
, "mips64-le"))
170 *ret
= SCMP_ARCH_MIPSEL64
;
171 else if (streq(n
, "mips64-le-n32"))
172 *ret
= SCMP_ARCH_MIPSEL64N32
;
173 else if (streq(n
, "ppc"))
174 *ret
= SCMP_ARCH_PPC
;
175 else if (streq(n
, "ppc64"))
176 *ret
= SCMP_ARCH_PPC64
;
177 else if (streq(n
, "ppc64-le"))
178 *ret
= SCMP_ARCH_PPC64LE
;
179 else if (streq(n
, "s390"))
180 *ret
= SCMP_ARCH_S390
;
181 else if (streq(n
, "s390x"))
182 *ret
= SCMP_ARCH_S390X
;
189 int seccomp_init_for_arch(scmp_filter_ctx
*ret
, uint32_t arch
, uint32_t default_action
) {
190 scmp_filter_ctx seccomp
;
193 /* Much like seccomp_init(), but initializes the filter for one specific architecture only, without affecting
194 * any others. Also, turns off the NNP fiddling. */
196 seccomp
= seccomp_init(default_action
);
200 if (arch
!= SCMP_ARCH_NATIVE
&&
201 arch
!= seccomp_arch_native()) {
203 r
= seccomp_arch_remove(seccomp
, seccomp_arch_native());
207 r
= seccomp_arch_add(seccomp
, arch
);
211 assert(seccomp_arch_exist(seccomp
, arch
) >= 0);
212 assert(seccomp_arch_exist(seccomp
, SCMP_ARCH_NATIVE
) == -EEXIST
);
213 assert(seccomp_arch_exist(seccomp
, seccomp_arch_native()) == -EEXIST
);
215 assert(seccomp_arch_exist(seccomp
, SCMP_ARCH_NATIVE
) >= 0);
216 assert(seccomp_arch_exist(seccomp
, seccomp_arch_native()) >= 0);
219 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_ACT_BADARCH
, SCMP_ACT_ALLOW
);
223 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
231 seccomp_release(seccomp
);
235 static bool is_basic_seccomp_available(void) {
236 return prctl(PR_GET_SECCOMP
, 0, 0, 0, 0) >= 0;
239 static bool is_seccomp_filter_available(void) {
240 return prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, NULL
, 0, 0) < 0 &&
244 bool is_seccomp_available(void) {
245 static int cached_enabled
= -1;
247 if (cached_enabled
< 0)
249 is_basic_seccomp_available() &&
250 is_seccomp_filter_available();
252 return cached_enabled
;
255 const SyscallFilterSet syscall_filter_sets
[_SYSCALL_FILTER_SET_MAX
] = {
256 [SYSCALL_FILTER_SET_DEFAULT
] = {
258 .help
= "System calls that are always permitted",
285 "getrlimit\0" /* make sure processes can query stack size and such */
307 [SYSCALL_FILTER_SET_AIO
] = {
309 .help
= "Asynchronous IO",
318 [SYSCALL_FILTER_SET_BASIC_IO
] = {
339 [SYSCALL_FILTER_SET_CHOWN
] = {
341 .help
= "Change ownership of files and directories",
351 [SYSCALL_FILTER_SET_CLOCK
] = {
353 .help
= "Change the system time",
361 [SYSCALL_FILTER_SET_CPU_EMULATION
] = {
362 .name
= "@cpu-emulation",
363 .help
= "System calls for CPU emulation functionality",
371 [SYSCALL_FILTER_SET_DEBUG
] = {
373 .help
= "Debugging, performance monitoring and tracing functionality",
379 #ifdef __NR_s390_runtime_instr
380 "s390_runtime_instr\0"
382 "sys_debug_setcontext\0"
384 [SYSCALL_FILTER_SET_FILE_SYSTEM
] = {
385 .name
= "@file-system",
386 .help
= "File system operations",
416 "inotify_add_watch\0"
467 [SYSCALL_FILTER_SET_IO_EVENT
] = {
469 .help
= "Event loop system calls",
486 [SYSCALL_FILTER_SET_IPC
] = {
488 .help
= "SysV IPC, POSIX Message Queues or other IPC",
505 "process_vm_writev\0"
515 [SYSCALL_FILTER_SET_KEYRING
] = {
517 .help
= "Kernel keyring access",
523 [SYSCALL_FILTER_SET_MEMLOCK
] = {
525 .help
= "Memory locking control",
533 [SYSCALL_FILTER_SET_MODULE
] = {
535 .help
= "Loading and unloading of kernel modules",
541 [SYSCALL_FILTER_SET_MOUNT
] = {
543 .help
= "Mounting and unmounting of file systems",
551 [SYSCALL_FILTER_SET_NETWORK_IO
] = {
552 .name
= "@network-io",
553 .help
= "Network or Unix socket IO, should not be needed if not network facing",
577 [SYSCALL_FILTER_SET_OBSOLETE
] = {
578 /* some unknown even to libseccomp */
580 .help
= "Unusual, obsolete or unimplemented system calls",
609 [SYSCALL_FILTER_SET_PRIVILEGED
] = {
610 .name
= "@privileged",
611 .help
= "All system calls which need super-user capabilities",
626 "open_by_handle_at\0"
639 "setuid\0" /* We list the explicit system calls here, as @setuid also includes setgid() which is not necessarily privileged */
643 [SYSCALL_FILTER_SET_PROCESS
] = {
645 .help
= "Process control, execution, namespaceing operations",
648 "capget\0" /* Able to query arbitrary processes */
656 "rt_tgsigqueueinfo\0"
658 "swapcontext\0" /* Some archs e.g. powerpc32 are using it to do userspace context switches */
668 [SYSCALL_FILTER_SET_RAW_IO
] = {
670 .help
= "Raw I/O port access",
677 #ifdef __NR_s390_pci_mmio_read
678 "s390_pci_mmio_read\0"
680 #ifdef __NR_s390_pci_mmio_write
681 "s390_pci_mmio_write\0"
684 [SYSCALL_FILTER_SET_REBOOT
] = {
686 .help
= "Reboot and reboot preparation/kexec",
692 [SYSCALL_FILTER_SET_RESOURCES
] = {
693 .name
= "@resources",
694 .help
= "Alter resource settings",
701 "sched_setaffinity\0"
704 "sched_setscheduler\0"
709 [SYSCALL_FILTER_SET_SETUID
] = {
711 .help
= "Operations for changing user/group credentials",
728 [SYSCALL_FILTER_SET_SIGNAL
] = {
730 .help
= "Process signal handling",
746 [SYSCALL_FILTER_SET_SWAP
] = {
748 .help
= "Enable/disable swap devices",
753 [SYSCALL_FILTER_SET_SYNC
] = {
755 .help
= "Synchronize files and memory to storage",
764 [SYSCALL_FILTER_SET_SYSTEM_SERVICE
] = {
765 .name
= "@system-service",
766 .help
= "General system service operations",
801 "name_to_handle_at\0"
808 "sched_get_priority_max\0"
809 "sched_get_priority_min\0"
810 "sched_getaffinity\0"
813 "sched_getscheduler\0"
814 "sched_rr_get_interval\0"
832 [SYSCALL_FILTER_SET_TIMER
] = {
834 .help
= "Schedule operations by time",
851 const SyscallFilterSet
*syscall_filter_set_find(const char *name
) {
854 if (isempty(name
) || name
[0] != '@')
857 for (i
= 0; i
< _SYSCALL_FILTER_SET_MAX
; i
++)
858 if (streq(syscall_filter_sets
[i
].name
, name
))
859 return syscall_filter_sets
+ i
;
864 static int seccomp_add_syscall_filter_set(scmp_filter_ctx seccomp
, const SyscallFilterSet
*set
, uint32_t action
, char **exclude
, bool log_missing
);
866 int seccomp_add_syscall_filter_item(scmp_filter_ctx
*seccomp
, const char *name
, uint32_t action
, char **exclude
, bool log_missing
) {
870 if (strv_contains(exclude
, name
))
873 if (name
[0] == '@') {
874 const SyscallFilterSet
*other
;
876 other
= syscall_filter_set_find(name
);
878 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
879 "Filter set %s is not known!",
882 return seccomp_add_syscall_filter_set(seccomp
, other
, action
, exclude
, log_missing
);
887 id
= seccomp_syscall_resolve_name(name
);
888 if (id
== __NR_SCMP_ERROR
) {
890 log_debug("System call %s is not known, ignoring.", name
);
894 r
= seccomp_rule_add_exact(seccomp
, action
, id
, 0);
896 /* If the system call is not known on this architecture, then that's fine, let's ignore it */
897 bool ignore
= r
== -EDOM
;
899 if (!ignore
|| log_missing
)
900 log_debug_errno(r
, "Failed to add rule for system call %s() / %d%s: %m",
901 name
, id
, ignore
? ", ignoring" : "");
910 static int seccomp_add_syscall_filter_set(
911 scmp_filter_ctx seccomp
,
912 const SyscallFilterSet
*set
,
923 NULSTR_FOREACH(sys
, set
->value
) {
924 r
= seccomp_add_syscall_filter_item(seccomp
, sys
, action
, exclude
, log_missing
);
932 int seccomp_load_syscall_filter_set(uint32_t default_action
, const SyscallFilterSet
*set
, uint32_t action
, bool log_missing
) {
938 /* The one-stop solution: allocate a seccomp object, add the specified filter to it, and apply it. Once for
939 * each local arch. */
941 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
942 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
944 log_debug("Operating on architecture: %s", seccomp_arch_to_string(arch
));
946 r
= seccomp_init_for_arch(&seccomp
, arch
, default_action
);
950 r
= seccomp_add_syscall_filter_set(seccomp
, set
, action
, NULL
, log_missing
);
952 return log_debug_errno(r
, "Failed to add filter set: %m");
954 r
= seccomp_load(seccomp
);
955 if (IN_SET(r
, -EPERM
, -EACCES
))
958 log_debug_errno(r
, "Failed to install filter set for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
964 int seccomp_load_syscall_filter_set_raw(uint32_t default_action
, Hashmap
* set
, uint32_t action
, bool log_missing
) {
968 /* Similar to seccomp_load_syscall_filter_set(), but takes a raw Set* of syscalls, instead of a
969 * SyscallFilterSet* table. */
971 if (hashmap_isempty(set
) && default_action
== SCMP_ACT_ALLOW
)
974 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
975 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
977 void *syscall_id
, *val
;
979 log_debug("Operating on architecture: %s", seccomp_arch_to_string(arch
));
981 r
= seccomp_init_for_arch(&seccomp
, arch
, default_action
);
985 HASHMAP_FOREACH_KEY(val
, syscall_id
, set
, i
) {
987 int id
= PTR_TO_INT(syscall_id
) - 1;
988 int error
= PTR_TO_INT(val
);
990 if (action
!= SCMP_ACT_ALLOW
&& error
>= 0)
991 a
= SCMP_ACT_ERRNO(error
);
993 r
= seccomp_rule_add_exact(seccomp
, a
, id
, 0);
995 /* If the system call is not known on this architecture, then that's fine, let's ignore it */
996 _cleanup_free_
char *n
= NULL
;
999 n
= seccomp_syscall_resolve_num_arch(SCMP_ARCH_NATIVE
, id
);
1000 ignore
= r
== -EDOM
;
1001 if (!ignore
|| log_missing
)
1002 log_debug_errno(r
, "Failed to add rule for system call %s() / %d%s: %m",
1003 strna(n
), id
, ignore
? ", ignoring" : "");
1009 r
= seccomp_load(seccomp
);
1010 if (IN_SET(r
, -EPERM
, -EACCES
))
1013 log_debug_errno(r
, "Failed to install filter set for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1019 int seccomp_parse_syscall_filter_full(
1023 SeccompParseFlags flags
,
1025 const char *filename
,
1033 if (name
[0] == '@') {
1034 const SyscallFilterSet
*set
;
1037 set
= syscall_filter_set_find(name
);
1039 if (!(flags
& SECCOMP_PARSE_PERMISSIVE
))
1042 log_syntax(unit
, flags
& SECCOMP_PARSE_LOG
? LOG_WARNING
: LOG_DEBUG
, filename
, line
, 0,
1043 "Unknown system call group, ignoring: %s", name
);
1047 NULSTR_FOREACH(i
, set
->value
) {
1048 /* Call ourselves again, for the group to parse. Note that we downgrade logging here (i.e. take
1049 * away the SECCOMP_PARSE_LOG flag) since any issues in the group table are our own problem,
1050 * not a problem in user configuration data and we shouldn't pretend otherwise by complaining
1052 r
= seccomp_parse_syscall_filter_full(i
, errno_num
, filter
, flags
&~ SECCOMP_PARSE_LOG
, unit
, filename
, line
);
1059 id
= seccomp_syscall_resolve_name(name
);
1060 if (id
== __NR_SCMP_ERROR
) {
1061 if (!(flags
& SECCOMP_PARSE_PERMISSIVE
))
1064 log_syntax(unit
, flags
& SECCOMP_PARSE_LOG
? LOG_WARNING
: LOG_DEBUG
, filename
, line
, 0,
1065 "Failed to parse system call, ignoring: %s", name
);
1069 /* If we previously wanted to forbid a syscall and now
1070 * we want to allow it, then remove it from the list. */
1071 if (!(flags
& SECCOMP_PARSE_INVERT
) == !!(flags
& SECCOMP_PARSE_WHITELIST
)) {
1072 r
= hashmap_put(filter
, INT_TO_PTR(id
+ 1), INT_TO_PTR(errno_num
));
1076 return flags
& SECCOMP_PARSE_LOG
? log_oom() : -ENOMEM
;
1078 assert_se(hashmap_update(filter
, INT_TO_PTR(id
+ 1), INT_TO_PTR(errno_num
)) == 0);
1084 (void) hashmap_remove(filter
, INT_TO_PTR(id
+ 1));
1090 int seccomp_restrict_namespaces(unsigned long retain
) {
1094 if (DEBUG_LOGGING
) {
1095 _cleanup_free_
char *s
= NULL
;
1097 (void) namespace_flags_to_string(retain
, &s
);
1098 log_debug("Restricting namespace to: %s.", strna(s
));
1102 if ((retain
& NAMESPACE_FLAGS_ALL
) == NAMESPACE_FLAGS_ALL
)
1105 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
1106 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
1109 log_debug("Operating on architecture: %s", seccomp_arch_to_string(arch
));
1111 r
= seccomp_init_for_arch(&seccomp
, arch
, SCMP_ACT_ALLOW
);
1115 if ((retain
& NAMESPACE_FLAGS_ALL
) == 0)
1116 /* If every single kind of namespace shall be prohibited, then let's block the whole setns() syscall
1118 r
= seccomp_rule_add_exact(
1120 SCMP_ACT_ERRNO(EPERM
),
1124 /* Otherwise, block only the invocations with the appropriate flags in the loop below, but also the
1125 * special invocation with a zero flags argument, right here. */
1126 r
= seccomp_rule_add_exact(
1128 SCMP_ACT_ERRNO(EPERM
),
1131 SCMP_A1(SCMP_CMP_EQ
, 0));
1133 log_debug_errno(r
, "Failed to add setns() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1137 for (i
= 0; namespace_flag_map
[i
].name
; i
++) {
1140 f
= namespace_flag_map
[i
].flag
;
1141 if ((retain
& f
) == f
) {
1142 log_debug("Permitting %s.", namespace_flag_map
[i
].name
);
1146 log_debug("Blocking %s.", namespace_flag_map
[i
].name
);
1148 r
= seccomp_rule_add_exact(
1150 SCMP_ACT_ERRNO(EPERM
),
1153 SCMP_A0(SCMP_CMP_MASKED_EQ
, f
, f
));
1155 log_debug_errno(r
, "Failed to add unshare() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1159 /* On s390/s390x the first two parameters to clone are switched */
1160 if (!IN_SET(arch
, SCMP_ARCH_S390
, SCMP_ARCH_S390X
))
1161 r
= seccomp_rule_add_exact(
1163 SCMP_ACT_ERRNO(EPERM
),
1166 SCMP_A0(SCMP_CMP_MASKED_EQ
, f
, f
));
1168 r
= seccomp_rule_add_exact(
1170 SCMP_ACT_ERRNO(EPERM
),
1173 SCMP_A1(SCMP_CMP_MASKED_EQ
, f
, f
));
1175 log_debug_errno(r
, "Failed to add clone() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1179 if ((retain
& NAMESPACE_FLAGS_ALL
) != 0) {
1180 r
= seccomp_rule_add_exact(
1182 SCMP_ACT_ERRNO(EPERM
),
1185 SCMP_A1(SCMP_CMP_MASKED_EQ
, f
, f
));
1187 log_debug_errno(r
, "Failed to add setns() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1195 r
= seccomp_load(seccomp
);
1196 if (IN_SET(r
, -EPERM
, -EACCES
))
1199 log_debug_errno(r
, "Failed to install namespace restriction rules for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1205 int seccomp_protect_sysctl(void) {
1209 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
1210 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
1212 log_debug("Operating on architecture: %s", seccomp_arch_to_string(arch
));
1214 if (IN_SET(arch
, SCMP_ARCH_X32
, SCMP_ARCH_AARCH64
))
1215 /* No _sysctl syscall */
1218 r
= seccomp_init_for_arch(&seccomp
, arch
, SCMP_ACT_ALLOW
);
1222 r
= seccomp_rule_add_exact(
1224 SCMP_ACT_ERRNO(EPERM
),
1228 log_debug_errno(r
, "Failed to add _sysctl() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1232 r
= seccomp_load(seccomp
);
1233 if (IN_SET(r
, -EPERM
, -EACCES
))
1236 log_debug_errno(r
, "Failed to install sysctl protection rules for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1242 int seccomp_restrict_address_families(Set
*address_families
, bool whitelist
) {
1246 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
1247 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
1251 log_debug("Operating on architecture: %s", seccomp_arch_to_string(arch
));
1255 case SCMP_ARCH_X86_64
:
1258 case SCMP_ARCH_AARCH64
:
1260 case SCMP_ARCH_PPC64
:
1261 case SCMP_ARCH_PPC64LE
:
1262 case SCMP_ARCH_MIPSEL64N32
:
1263 case SCMP_ARCH_MIPS64N32
:
1264 case SCMP_ARCH_MIPSEL64
:
1265 case SCMP_ARCH_MIPS64
:
1266 /* These we know we support (i.e. are the ones that do not use socketcall()) */
1270 case SCMP_ARCH_S390
:
1271 case SCMP_ARCH_S390X
:
1273 case SCMP_ARCH_MIPSEL
:
1274 case SCMP_ARCH_MIPS
:
1276 /* These we either know we don't support (i.e. are the ones that do use socketcall()), or we
1285 r
= seccomp_init_for_arch(&seccomp
, arch
, SCMP_ACT_ALLOW
);
1290 int af
, first
= 0, last
= 0;
1293 /* If this is a whitelist, we first block the address families that are out of range and then
1294 * everything that is not in the set. First, we find the lowest and highest address family in
1297 SET_FOREACH(afp
, address_families
, i
) {
1298 af
= PTR_TO_INT(afp
);
1300 if (af
<= 0 || af
>= af_max())
1303 if (first
== 0 || af
< first
)
1306 if (last
== 0 || af
> last
)
1310 assert((first
== 0) == (last
== 0));
1314 /* No entries in the valid range, block everything */
1315 r
= seccomp_rule_add_exact(
1317 SCMP_ACT_ERRNO(EAFNOSUPPORT
),
1321 log_debug_errno(r
, "Failed to add socket() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1327 /* Block everything below the first entry */
1328 r
= seccomp_rule_add_exact(
1330 SCMP_ACT_ERRNO(EAFNOSUPPORT
),
1333 SCMP_A0(SCMP_CMP_LT
, first
));
1335 log_debug_errno(r
, "Failed to add socket() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1339 /* Block everything above the last entry */
1340 r
= seccomp_rule_add_exact(
1342 SCMP_ACT_ERRNO(EAFNOSUPPORT
),
1345 SCMP_A0(SCMP_CMP_GT
, last
));
1347 log_debug_errno(r
, "Failed to add socket() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1351 /* Block everything between the first and last entry */
1352 for (af
= 1; af
< af_max(); af
++) {
1354 if (set_contains(address_families
, INT_TO_PTR(af
)))
1357 r
= seccomp_rule_add_exact(
1359 SCMP_ACT_ERRNO(EAFNOSUPPORT
),
1362 SCMP_A0(SCMP_CMP_EQ
, af
));
1367 log_debug_errno(r
, "Failed to add socket() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1375 /* If this is a blacklist, then generate one rule for
1376 * each address family that are then combined in OR
1379 SET_FOREACH(af
, address_families
, i
) {
1381 r
= seccomp_rule_add_exact(
1383 SCMP_ACT_ERRNO(EAFNOSUPPORT
),
1386 SCMP_A0(SCMP_CMP_EQ
, PTR_TO_INT(af
)));
1391 log_debug_errno(r
, "Failed to add socket() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1396 r
= seccomp_load(seccomp
);
1397 if (IN_SET(r
, -EPERM
, -EACCES
))
1400 log_debug_errno(r
, "Failed to install socket family rules for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1406 int seccomp_restrict_realtime(void) {
1407 static const int permitted_policies
[] = {
1413 int r
, max_policy
= 0;
1417 /* Determine the highest policy constant we want to allow */
1418 for (i
= 0; i
< ELEMENTSOF(permitted_policies
); i
++)
1419 if (permitted_policies
[i
] > max_policy
)
1420 max_policy
= permitted_policies
[i
];
1422 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
1423 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
1426 log_debug("Operating on architecture: %s", seccomp_arch_to_string(arch
));
1428 r
= seccomp_init_for_arch(&seccomp
, arch
, SCMP_ACT_ALLOW
);
1432 /* Go through all policies with lower values than that, and block them -- unless they appear in the
1434 for (p
= 0; p
< max_policy
; p
++) {
1437 /* Check if this is in the whitelist. */
1438 for (i
= 0; i
< ELEMENTSOF(permitted_policies
); i
++)
1439 if (permitted_policies
[i
] == p
) {
1447 /* Deny this policy */
1448 r
= seccomp_rule_add_exact(
1450 SCMP_ACT_ERRNO(EPERM
),
1451 SCMP_SYS(sched_setscheduler
),
1453 SCMP_A1(SCMP_CMP_EQ
, p
));
1455 log_debug_errno(r
, "Failed to add scheduler rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1460 /* Blacklist all other policies, i.e. the ones with higher values. Note that all comparisons are
1461 * unsigned here, hence no need no check for < 0 values. */
1462 r
= seccomp_rule_add_exact(
1464 SCMP_ACT_ERRNO(EPERM
),
1465 SCMP_SYS(sched_setscheduler
),
1467 SCMP_A1(SCMP_CMP_GT
, max_policy
));
1469 log_debug_errno(r
, "Failed to add scheduler rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1473 r
= seccomp_load(seccomp
);
1474 if (IN_SET(r
, -EPERM
, -EACCES
))
1477 log_debug_errno(r
, "Failed to install realtime protection rules for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1483 static int add_seccomp_syscall_filter(scmp_filter_ctx seccomp
,
1487 const struct scmp_arg_cmp arg
) {
1490 r
= seccomp_rule_add_exact(seccomp
, SCMP_ACT_ERRNO(EPERM
), nr
, arg_cnt
, arg
);
1492 _cleanup_free_
char *n
= NULL
;
1494 n
= seccomp_syscall_resolve_num_arch(arch
, nr
);
1495 log_debug_errno(r
, "Failed to add %s() rule for architecture %s, skipping: %m",
1497 seccomp_arch_to_string(arch
));
1503 /* For known architectures, check that syscalls are indeed defined or not. */
1504 #if defined(__x86_64__) || defined(__arm__) || defined(__aarch64__)
1505 assert_cc(SCMP_SYS(shmget
) > 0);
1506 assert_cc(SCMP_SYS(shmat
) > 0);
1507 assert_cc(SCMP_SYS(shmdt
) > 0);
1510 int seccomp_memory_deny_write_execute(void) {
1514 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
1515 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
1516 int filter_syscall
= 0, block_syscall
= 0, shmat_syscall
= 0;
1518 log_debug("Operating on architecture: %s", seccomp_arch_to_string(arch
));
1523 filter_syscall
= SCMP_SYS(mmap2
);
1524 block_syscall
= SCMP_SYS(mmap
);
1525 shmat_syscall
= SCMP_SYS(shmat
);
1529 case SCMP_ARCH_PPC64
:
1530 case SCMP_ARCH_PPC64LE
:
1531 filter_syscall
= SCMP_SYS(mmap
);
1533 /* Note that shmat() isn't available, and the call is multiplexed through ipc().
1534 * We ignore that here, which means there's still a way to get writable/executable
1535 * memory, if an IPC key is mapped like this. That's a pity, but no total loss. */
1540 filter_syscall
= SCMP_SYS(mmap2
); /* arm has only mmap2 */
1541 shmat_syscall
= SCMP_SYS(shmat
);
1544 case SCMP_ARCH_X86_64
:
1546 case SCMP_ARCH_AARCH64
:
1547 filter_syscall
= SCMP_SYS(mmap
); /* amd64, x32, and arm64 have only mmap */
1548 shmat_syscall
= SCMP_SYS(shmat
);
1551 /* Please add more definitions here, if you port systemd to other architectures! */
1553 #if !defined(__i386__) && !defined(__x86_64__) && !defined(__powerpc__) && !defined(__powerpc64__) && !defined(__arm__) && !defined(__aarch64__)
1554 #warning "Consider adding the right mmap() syscall definitions here!"
1558 /* Can't filter mmap() on this arch, then skip it */
1559 if (filter_syscall
== 0)
1562 r
= seccomp_init_for_arch(&seccomp
, arch
, SCMP_ACT_ALLOW
);
1566 r
= add_seccomp_syscall_filter(seccomp
, arch
, filter_syscall
,
1568 SCMP_A2(SCMP_CMP_MASKED_EQ
, PROT_EXEC
|PROT_WRITE
, PROT_EXEC
|PROT_WRITE
));
1572 if (block_syscall
!= 0) {
1573 r
= add_seccomp_syscall_filter(seccomp
, arch
, block_syscall
, 0, (const struct scmp_arg_cmp
){} );
1578 r
= add_seccomp_syscall_filter(seccomp
, arch
, SCMP_SYS(mprotect
),
1580 SCMP_A2(SCMP_CMP_MASKED_EQ
, PROT_EXEC
, PROT_EXEC
));
1584 #ifdef __NR_pkey_mprotect
1585 r
= add_seccomp_syscall_filter(seccomp
, arch
, SCMP_SYS(pkey_mprotect
),
1587 SCMP_A2(SCMP_CMP_MASKED_EQ
, PROT_EXEC
, PROT_EXEC
));
1592 if (shmat_syscall
> 0) {
1593 r
= add_seccomp_syscall_filter(seccomp
, arch
, SCMP_SYS(shmat
),
1595 SCMP_A2(SCMP_CMP_MASKED_EQ
, SHM_EXEC
, SHM_EXEC
));
1600 r
= seccomp_load(seccomp
);
1601 if (IN_SET(r
, -EPERM
, -EACCES
))
1604 log_debug_errno(r
, "Failed to install MemoryDenyWriteExecute= rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1610 int seccomp_restrict_archs(Set
*archs
) {
1611 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
1616 /* This installs a filter with no rules, but that restricts the system call architectures to the specified
1619 * There are some qualifications. However the most important use is to stop processes from bypassing
1620 * system call restrictions, in case they used a broader (multiplexing) syscall which is only available
1621 * in a non-native architecture. There are no holes in this use case, at least so far. */
1623 /* Note libseccomp includes our "native" (current) architecture in the filter by default.
1624 * We do not remove it. For example, our callers expect to be able to call execve() afterwards
1625 * to run a program with the restrictions applied. */
1626 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1630 SET_FOREACH(id
, archs
, i
) {
1631 r
= seccomp_arch_add(seccomp
, PTR_TO_UINT32(id
) - 1);
1632 if (r
< 0 && r
!= -EEXIST
)
1636 /* The vdso for x32 assumes that x86-64 syscalls are available. Let's allow them, since x32
1637 * x32 syscalls should basically match x86-64 for everything except the pointer type.
1638 * The important thing is that you can block the old 32-bit x86 syscalls.
1639 * https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=850047 */
1641 if (seccomp_arch_native() == SCMP_ARCH_X32
||
1642 set_contains(archs
, UINT32_TO_PTR(SCMP_ARCH_X32
+ 1))) {
1644 r
= seccomp_arch_add(seccomp
, SCMP_ARCH_X86_64
);
1645 if (r
< 0 && r
!= -EEXIST
)
1649 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1653 r
= seccomp_load(seccomp
);
1654 if (IN_SET(r
, -EPERM
, -EACCES
))
1657 log_debug_errno(r
, "Failed to restrict system call architectures, skipping: %m");
1662 int parse_syscall_archs(char **l
, Set
**archs
) {
1663 _cleanup_set_free_ Set
*_archs
;
1670 r
= set_ensure_allocated(&_archs
, NULL
);
1674 STRV_FOREACH(s
, l
) {
1677 r
= seccomp_arch_from_string(*s
, &a
);
1681 r
= set_put(_archs
, UINT32_TO_PTR(a
+ 1));
1686 *archs
= TAKE_PTR(_archs
);
1691 int seccomp_filter_set_add(Hashmap
*filter
, bool add
, const SyscallFilterSet
*set
) {
1697 NULSTR_FOREACH(i
, set
->value
) {
1700 const SyscallFilterSet
*more
;
1702 more
= syscall_filter_set_find(i
);
1706 r
= seccomp_filter_set_add(filter
, add
, more
);
1712 id
= seccomp_syscall_resolve_name(i
);
1713 if (id
== __NR_SCMP_ERROR
) {
1714 log_debug("Couldn't resolve system call, ignoring: %s", i
);
1719 r
= hashmap_put(filter
, INT_TO_PTR(id
+ 1), INT_TO_PTR(-1));
1723 (void) hashmap_remove(filter
, INT_TO_PTR(id
+ 1));
1730 int seccomp_lock_personality(unsigned long personality
) {
1734 if (personality
>= PERSONALITY_INVALID
)
1737 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
1738 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
1740 r
= seccomp_init_for_arch(&seccomp
, arch
, SCMP_ACT_ALLOW
);
1744 r
= seccomp_rule_add_exact(
1746 SCMP_ACT_ERRNO(EPERM
),
1747 SCMP_SYS(personality
),
1749 SCMP_A0(SCMP_CMP_NE
, personality
));
1751 log_debug_errno(r
, "Failed to add scheduler rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1755 r
= seccomp_load(seccomp
);
1756 if (IN_SET(r
, -EPERM
, -EACCES
))
1759 log_debug_errno(r
, "Failed to enable personality lock for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1765 int seccomp_protect_hostname(void) {
1769 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
1770 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
1772 r
= seccomp_init_for_arch(&seccomp
, arch
, SCMP_ACT_ALLOW
);
1776 r
= seccomp_rule_add_exact(
1778 SCMP_ACT_ERRNO(EPERM
),
1779 SCMP_SYS(sethostname
),
1782 log_debug_errno(r
, "Failed to add sethostname() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1786 r
= seccomp_rule_add_exact(
1788 SCMP_ACT_ERRNO(EPERM
),
1789 SCMP_SYS(setdomainname
),
1792 log_debug_errno(r
, "Failed to add setdomainname() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1796 r
= seccomp_load(seccomp
);
1797 if (IN_SET(r
, -EPERM
, -EACCES
))
1800 log_debug_errno(r
, "Failed to apply hostname restrictions for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1806 int seccomp_restrict_suid_sgid(void) {
1810 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
1811 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
1813 r
= seccomp_init_for_arch(&seccomp
, arch
, SCMP_ACT_ALLOW
);
1817 /* Checks the mode_t parameter of the following system calls:
1819 * → chmod() + fchmod() + fchmodat()
1820 * → open() + creat() + openat()
1821 * → mkdir() + mkdirat()
1822 * → mknod() + mknodat()
1825 for (unsigned bit
= 0; bit
< 2; bit
++) {
1826 /* Block S_ISUID in the first iteration, S_ISGID in the second */
1827 mode_t m
= bit
== 0 ? S_ISUID
: S_ISGID
;
1829 r
= seccomp_rule_add_exact(
1831 SCMP_ACT_ERRNO(EPERM
),
1834 SCMP_A1(SCMP_CMP_MASKED_EQ
, m
, m
));
1838 r
= seccomp_rule_add_exact(
1840 SCMP_ACT_ERRNO(EPERM
),
1843 SCMP_A1(SCMP_CMP_MASKED_EQ
, m
, m
));
1847 r
= seccomp_rule_add_exact(
1849 SCMP_ACT_ERRNO(EPERM
),
1852 SCMP_A2(SCMP_CMP_MASKED_EQ
, m
, m
));
1856 r
= seccomp_rule_add_exact(
1858 SCMP_ACT_ERRNO(EPERM
),
1861 SCMP_A1(SCMP_CMP_MASKED_EQ
, m
, m
));
1865 r
= seccomp_rule_add_exact(
1867 SCMP_ACT_ERRNO(EPERM
),
1870 SCMP_A2(SCMP_CMP_MASKED_EQ
, m
, m
));
1874 r
= seccomp_rule_add_exact(
1876 SCMP_ACT_ERRNO(EPERM
),
1879 SCMP_A1(SCMP_CMP_MASKED_EQ
, m
, m
));
1883 r
= seccomp_rule_add_exact(
1885 SCMP_ACT_ERRNO(EPERM
),
1888 SCMP_A2(SCMP_CMP_MASKED_EQ
, m
, m
));
1892 r
= seccomp_rule_add_exact(
1894 SCMP_ACT_ERRNO(EPERM
),
1897 SCMP_A1(SCMP_CMP_MASKED_EQ
, O_CREAT
, O_CREAT
),
1898 SCMP_A2(SCMP_CMP_MASKED_EQ
, m
, m
));
1902 r
= seccomp_rule_add_exact(
1904 SCMP_ACT_ERRNO(EPERM
),
1907 SCMP_A2(SCMP_CMP_MASKED_EQ
, O_CREAT
, O_CREAT
),
1908 SCMP_A3(SCMP_CMP_MASKED_EQ
, m
, m
));
1912 r
= seccomp_rule_add_exact(
1914 SCMP_ACT_ERRNO(EPERM
),
1917 SCMP_A1(SCMP_CMP_MASKED_EQ
, m
, m
));
1922 log_debug_errno(r
, "Failed to add suid/sgid rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1926 r
= seccomp_load(seccomp
);
1927 if (IN_SET(r
, -EPERM
, -EACCES
))
1930 log_debug_errno(r
, "Failed to apply suid/sgid restrictions for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));