2 This file is part of systemd.
4 Copyright 2014 Lennart Poettering
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.
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.
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/>.
21 #include <linux/seccomp.h>
25 #include <sys/prctl.h>
29 #include "alloc-util.h"
32 #include "process-util.h"
33 #include "seccomp-util.h"
35 #include "string-util.h"
38 #include "errno-list.h"
40 const uint32_t seccomp_local_archs
[] = {
42 /* 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 */
44 #if defined(__x86_64__) && defined(__ILP32__)
47 SCMP_ARCH_X32
, /* native */
48 #elif defined(__x86_64__) && !defined(__ILP32__)
51 SCMP_ARCH_X86_64
, /* native */
52 #elif defined(__i386__)
54 #elif defined(__aarch64__)
56 SCMP_ARCH_AARCH64
, /* native */
57 #elif defined(__arm__)
59 #elif defined(__mips__) && __BYTE_ORDER == __BIG_ENDIAN && _MIPS_SIM == _MIPS_SIM_ABI32
61 SCMP_ARCH_MIPS
, /* native */
62 #elif defined(__mips__) && __BYTE_ORDER == __LITTLE_ENDIAN && _MIPS_SIM == _MIPS_SIM_ABI32
64 SCMP_ARCH_MIPSEL
, /* native */
65 #elif defined(__mips__) && __BYTE_ORDER == __BIG_ENDIAN && _MIPS_SIM == _MIPS_SIM_ABI64
68 SCMP_ARCH_MIPSEL64N32
,
71 SCMP_ARCH_MIPS64
, /* native */
72 #elif defined(__mips__) && __BYTE_ORDER == __LITTLE_ENDIAN && _MIPS_SIM == _MIPS_SIM_ABI64
76 SCMP_ARCH_MIPSEL64N32
,
78 SCMP_ARCH_MIPSEL64
, /* native */
79 #elif defined(__mips__) && __BYTE_ORDER == __BIG_ENDIAN && _MIPS_SIM == _MIPS_SIM_NABI32
84 SCMP_ARCH_MIPSEL64N32
,
85 SCMP_ARCH_MIPS64N32
, /* native */
86 #elif defined(__mips__) && __BYTE_ORDER == __LITTLE_ENDIAN && _MIPS_SIM == _MIPS_SIM_NABI32
92 SCMP_ARCH_MIPSEL64N32
, /* native */
93 #elif defined(__powerpc64__) && __BYTE_ORDER == __BIG_ENDIAN
96 SCMP_ARCH_PPC64
, /* native */
97 #elif defined(__powerpc64__) && __BYTE_ORDER == __LITTLE_ENDIAN
100 SCMP_ARCH_PPC64LE
, /* native */
101 #elif defined(__powerpc__)
103 #elif defined(__s390x__)
105 SCMP_ARCH_S390X
, /* native */
106 #elif defined(__s390__)
112 const char* seccomp_arch_to_string(uint32_t c
) {
113 /* Maintain order used in <seccomp.h>.
115 * Names used here should be the same as those used for ConditionArchitecture=,
116 * except for "subarchitectures" like x32. */
119 case SCMP_ARCH_NATIVE
:
123 case SCMP_ARCH_X86_64
:
129 case SCMP_ARCH_AARCH64
:
133 case SCMP_ARCH_MIPS64
:
135 case SCMP_ARCH_MIPS64N32
:
137 case SCMP_ARCH_MIPSEL
:
139 case SCMP_ARCH_MIPSEL64
:
141 case SCMP_ARCH_MIPSEL64N32
:
142 return "mips64-le-n32";
145 case SCMP_ARCH_PPC64
:
147 case SCMP_ARCH_PPC64LE
:
151 case SCMP_ARCH_S390X
:
158 int seccomp_arch_from_string(const char *n
, uint32_t *ret
) {
164 if (streq(n
, "native"))
165 *ret
= SCMP_ARCH_NATIVE
;
166 else if (streq(n
, "x86"))
167 *ret
= SCMP_ARCH_X86
;
168 else if (streq(n
, "x86-64"))
169 *ret
= SCMP_ARCH_X86_64
;
170 else if (streq(n
, "x32"))
171 *ret
= SCMP_ARCH_X32
;
172 else if (streq(n
, "arm"))
173 *ret
= SCMP_ARCH_ARM
;
174 else if (streq(n
, "arm64"))
175 *ret
= SCMP_ARCH_AARCH64
;
176 else if (streq(n
, "mips"))
177 *ret
= SCMP_ARCH_MIPS
;
178 else if (streq(n
, "mips64"))
179 *ret
= SCMP_ARCH_MIPS64
;
180 else if (streq(n
, "mips64-n32"))
181 *ret
= SCMP_ARCH_MIPS64N32
;
182 else if (streq(n
, "mips-le"))
183 *ret
= SCMP_ARCH_MIPSEL
;
184 else if (streq(n
, "mips64-le"))
185 *ret
= SCMP_ARCH_MIPSEL64
;
186 else if (streq(n
, "mips64-le-n32"))
187 *ret
= SCMP_ARCH_MIPSEL64N32
;
188 else if (streq(n
, "ppc"))
189 *ret
= SCMP_ARCH_PPC
;
190 else if (streq(n
, "ppc64"))
191 *ret
= SCMP_ARCH_PPC64
;
192 else if (streq(n
, "ppc64-le"))
193 *ret
= SCMP_ARCH_PPC64LE
;
194 else if (streq(n
, "s390"))
195 *ret
= SCMP_ARCH_S390
;
196 else if (streq(n
, "s390x"))
197 *ret
= SCMP_ARCH_S390X
;
204 int seccomp_init_for_arch(scmp_filter_ctx
*ret
, uint32_t arch
, uint32_t default_action
) {
205 scmp_filter_ctx seccomp
;
208 /* Much like seccomp_init(), but initializes the filter for one specific architecture only, without affecting
209 * any others. Also, turns off the NNP fiddling. */
211 seccomp
= seccomp_init(default_action
);
215 if (arch
!= SCMP_ARCH_NATIVE
&&
216 arch
!= seccomp_arch_native()) {
218 r
= seccomp_arch_remove(seccomp
, seccomp_arch_native());
222 r
= seccomp_arch_add(seccomp
, arch
);
226 assert(seccomp_arch_exist(seccomp
, arch
) >= 0);
227 assert(seccomp_arch_exist(seccomp
, SCMP_ARCH_NATIVE
) == -EEXIST
);
228 assert(seccomp_arch_exist(seccomp
, seccomp_arch_native()) == -EEXIST
);
230 assert(seccomp_arch_exist(seccomp
, SCMP_ARCH_NATIVE
) >= 0);
231 assert(seccomp_arch_exist(seccomp
, seccomp_arch_native()) >= 0);
234 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_ACT_BADARCH
, SCMP_ACT_ALLOW
);
238 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
246 seccomp_release(seccomp
);
250 static bool is_basic_seccomp_available(void) {
251 return prctl(PR_GET_SECCOMP
, 0, 0, 0, 0) >= 0;
254 static bool is_seccomp_filter_available(void) {
255 return prctl(PR_SET_SECCOMP
, SECCOMP_MODE_FILTER
, NULL
, 0, 0) < 0 &&
259 bool is_seccomp_available(void) {
260 static int cached_enabled
= -1;
262 if (cached_enabled
< 0)
264 is_basic_seccomp_available() &&
265 is_seccomp_filter_available();
267 return cached_enabled
;
270 const SyscallFilterSet syscall_filter_sets
[_SYSCALL_FILTER_SET_MAX
] = {
271 [SYSCALL_FILTER_SET_DEFAULT
] = {
273 .help
= "System calls that are always permitted",
300 "getrlimit\0" /* make sure processes can query stack size and such */
320 [SYSCALL_FILTER_SET_BASIC_IO
] = {
339 [SYSCALL_FILTER_SET_CLOCK
] = {
341 .help
= "Change the system time",
349 [SYSCALL_FILTER_SET_CPU_EMULATION
] = {
350 .name
= "@cpu-emulation",
351 .help
= "System calls for CPU emulation functionality",
359 [SYSCALL_FILTER_SET_DEBUG
] = {
361 .help
= "Debugging, performance monitoring and tracing functionality",
366 "process_vm_writev\0"
369 #ifdef __NR_s390_runtime_instr
370 "s390_runtime_instr\0"
372 "sys_debug_setcontext\0"
374 [SYSCALL_FILTER_SET_FILE_SYSTEM
] = {
375 .name
= "@file-system",
376 .help
= "File system operations",
406 "inotify_add_watch\0"
457 [SYSCALL_FILTER_SET_IO_EVENT
] = {
459 .help
= "Event loop system calls",
476 [SYSCALL_FILTER_SET_IPC
] = {
478 .help
= "SysV IPC, POSIX Message Queues or other IPC",
495 "process_vm_writev\0"
505 [SYSCALL_FILTER_SET_KEYRING
] = {
507 .help
= "Kernel keyring access",
513 [SYSCALL_FILTER_SET_MEMLOCK
] = {
515 .help
= "Memory locking control",
523 [SYSCALL_FILTER_SET_MODULE
] = {
525 .help
= "Loading and unloading of kernel modules",
531 [SYSCALL_FILTER_SET_MOUNT
] = {
533 .help
= "Mounting and unmounting of file systems",
541 [SYSCALL_FILTER_SET_NETWORK_IO
] = {
542 .name
= "@network-io",
543 .help
= "Network or Unix socket IO, should not be needed if not network facing",
567 [SYSCALL_FILTER_SET_OBSOLETE
] = {
568 /* some unknown even to libseccomp */
570 .help
= "Unusual, obsolete or unimplemented system calls",
599 [SYSCALL_FILTER_SET_PRIVILEGED
] = {
600 .name
= "@privileged",
601 .help
= "All system calls which need super-user capabilities",
640 [SYSCALL_FILTER_SET_PROCESS
] = {
642 .help
= "Process control, execution, namespaceing operations",
645 "capget\0" /* Able to query arbitrary processes */
653 "rt_tgsigqueueinfo\0"
664 [SYSCALL_FILTER_SET_RAW_IO
] = {
666 .help
= "Raw I/O port access",
673 #ifdef __NR_s390_pci_mmio_read
674 "s390_pci_mmio_read\0"
676 #ifdef __NR_s390_pci_mmio_write
677 "s390_pci_mmio_write\0"
680 [SYSCALL_FILTER_SET_REBOOT
] = {
682 .help
= "Reboot and reboot preparation/kexec",
688 [SYSCALL_FILTER_SET_RESOURCES
] = {
689 .name
= "@resources",
690 .help
= "Alter resource settings",
697 "sched_setaffinity\0"
700 "sched_setscheduler\0"
705 [SYSCALL_FILTER_SET_SETUID
] = {
707 .help
= "Operations for changing user/group credentials",
724 [SYSCALL_FILTER_SET_SIGNAL
] = {
726 .help
= "Process signal handling",
742 [SYSCALL_FILTER_SET_SWAP
] = {
744 .help
= "Enable/disable swap devices",
749 [SYSCALL_FILTER_SET_TIMER
] = {
751 .help
= "Schedule operations by time",
768 const SyscallFilterSet
*syscall_filter_set_find(const char *name
) {
771 if (isempty(name
) || name
[0] != '@')
774 for (i
= 0; i
< _SYSCALL_FILTER_SET_MAX
; i
++)
775 if (streq(syscall_filter_sets
[i
].name
, name
))
776 return syscall_filter_sets
+ i
;
781 static int seccomp_add_syscall_filter_set(scmp_filter_ctx seccomp
, const SyscallFilterSet
*set
, uint32_t action
, char **exclude
);
783 int seccomp_add_syscall_filter_item(scmp_filter_ctx
*seccomp
, const char *name
, uint32_t action
, char **exclude
) {
789 if (strv_contains(exclude
, name
))
792 if (name
[0] == '@') {
793 const SyscallFilterSet
*other
;
795 other
= syscall_filter_set_find(name
);
797 log_debug("Filter set %s is not known!", name
);
801 r
= seccomp_add_syscall_filter_set(seccomp
, other
, action
, exclude
);
807 id
= seccomp_syscall_resolve_name(name
);
808 if (id
== __NR_SCMP_ERROR
) {
809 log_debug("System call %s is not known, ignoring.", name
);
813 r
= seccomp_rule_add_exact(seccomp
, action
, id
, 0);
815 /* If the system call is not known on this architecture, then that's fine, let's ignore it */
816 log_debug_errno(r
, "Failed to add rule for system call %s() / %d, ignoring: %m", name
, id
);
822 static int seccomp_add_syscall_filter_set(
823 scmp_filter_ctx seccomp
,
824 const SyscallFilterSet
*set
,
834 NULSTR_FOREACH(sys
, set
->value
) {
835 r
= seccomp_add_syscall_filter_item(seccomp
, sys
, action
, exclude
);
843 int seccomp_load_syscall_filter_set(uint32_t default_action
, const SyscallFilterSet
*set
, uint32_t action
) {
849 /* The one-stop solution: allocate a seccomp object, add the specified filter to it, and apply it. Once for
850 * earch local arch. */
852 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
853 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
855 log_debug("Operating on architecture: %s", seccomp_arch_to_string(arch
));
857 r
= seccomp_init_for_arch(&seccomp
, arch
, default_action
);
861 r
= seccomp_add_syscall_filter_set(seccomp
, set
, action
, NULL
);
863 log_debug_errno(r
, "Failed to add filter set, ignoring: %m");
867 r
= seccomp_load(seccomp
);
868 if (IN_SET(r
, -EPERM
, -EACCES
))
871 log_debug_errno(r
, "Failed to install filter set for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
877 int seccomp_load_syscall_filter_set_raw(uint32_t default_action
, Set
* set
, uint32_t action
) {
881 /* Similar to seccomp_load_syscall_filter_set(), but takes a raw Set* of syscalls, instead of a
882 * SyscallFilterSet* table. */
884 if (set_isempty(set
) && default_action
== SCMP_ACT_ALLOW
)
887 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
888 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
892 log_debug("Operating on architecture: %s", seccomp_arch_to_string(arch
));
894 r
= seccomp_init_for_arch(&seccomp
, arch
, default_action
);
898 SET_FOREACH(id
, set
, i
) {
899 r
= seccomp_rule_add_exact(seccomp
, action
, PTR_TO_INT(id
) - 1, 0);
901 /* If the system call is not known on this architecture, then that's fine, let's ignore it */
902 _cleanup_free_
char *n
= NULL
;
904 n
= seccomp_syscall_resolve_num_arch(arch
, PTR_TO_INT(id
) - 1);
905 log_debug_errno(r
, "Failed to add rule for system call %s() / %d, ignoring: %m", strna(n
), PTR_TO_INT(id
) - 1);
909 r
= seccomp_load(seccomp
);
910 if (IN_SET(r
, -EPERM
, -EACCES
))
913 log_debug_errno(r
, "Failed to install filter set for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
919 int seccomp_restrict_namespaces(unsigned long retain
) {
923 if (log_get_max_level() >= LOG_DEBUG
) {
924 _cleanup_free_
char *s
= NULL
;
926 (void) namespace_flag_to_string_many(retain
, &s
);
927 log_debug("Restricting namespace to: %s.", strna(s
));
931 if ((retain
& NAMESPACE_FLAGS_ALL
) == NAMESPACE_FLAGS_ALL
)
934 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
935 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
938 log_debug("Operating on architecture: %s", seccomp_arch_to_string(arch
));
940 r
= seccomp_init_for_arch(&seccomp
, arch
, SCMP_ACT_ALLOW
);
944 if ((retain
& NAMESPACE_FLAGS_ALL
) == 0)
945 /* If every single kind of namespace shall be prohibited, then let's block the whole setns() syscall
947 r
= seccomp_rule_add_exact(
949 SCMP_ACT_ERRNO(EPERM
),
953 /* Otherwise, block only the invocations with the appropriate flags in the loop below, but also the
954 * special invocation with a zero flags argument, right here. */
955 r
= seccomp_rule_add_exact(
957 SCMP_ACT_ERRNO(EPERM
),
960 SCMP_A1(SCMP_CMP_EQ
, 0));
962 log_debug_errno(r
, "Failed to add setns() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
966 for (i
= 0; namespace_flag_map
[i
].name
; i
++) {
969 f
= namespace_flag_map
[i
].flag
;
970 if ((retain
& f
) == f
) {
971 log_debug("Permitting %s.", namespace_flag_map
[i
].name
);
975 log_debug("Blocking %s.", namespace_flag_map
[i
].name
);
977 r
= seccomp_rule_add_exact(
979 SCMP_ACT_ERRNO(EPERM
),
982 SCMP_A0(SCMP_CMP_MASKED_EQ
, f
, f
));
984 log_debug_errno(r
, "Failed to add unshare() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
988 /* On s390/s390x the first two parameters to clone are switched */
989 if (!IN_SET(arch
, SCMP_ARCH_S390
, SCMP_ARCH_S390X
))
990 r
= seccomp_rule_add_exact(
992 SCMP_ACT_ERRNO(EPERM
),
995 SCMP_A0(SCMP_CMP_MASKED_EQ
, f
, f
));
997 r
= seccomp_rule_add_exact(
999 SCMP_ACT_ERRNO(EPERM
),
1002 SCMP_A1(SCMP_CMP_MASKED_EQ
, f
, f
));
1004 log_debug_errno(r
, "Failed to add clone() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1008 if ((retain
& NAMESPACE_FLAGS_ALL
) != 0) {
1009 r
= seccomp_rule_add_exact(
1011 SCMP_ACT_ERRNO(EPERM
),
1014 SCMP_A1(SCMP_CMP_MASKED_EQ
, f
, f
));
1016 log_debug_errno(r
, "Failed to add setns() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1024 r
= seccomp_load(seccomp
);
1025 if (IN_SET(r
, -EPERM
, -EACCES
))
1028 log_debug_errno(r
, "Failed to install namespace restriction rules for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1034 int seccomp_protect_sysctl(void) {
1038 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
1039 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
1041 log_debug("Operating on architecture: %s", seccomp_arch_to_string(arch
));
1043 if (IN_SET(arch
, SCMP_ARCH_X32
, SCMP_ARCH_AARCH64
))
1044 /* No _sysctl syscall */
1047 r
= seccomp_init_for_arch(&seccomp
, arch
, SCMP_ACT_ALLOW
);
1051 r
= seccomp_rule_add_exact(
1053 SCMP_ACT_ERRNO(EPERM
),
1057 log_debug_errno(r
, "Failed to add _sysctl() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1061 r
= seccomp_load(seccomp
);
1062 if (IN_SET(r
, -EPERM
, -EACCES
))
1065 log_debug_errno(r
, "Failed to install sysctl protection rules for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1071 int seccomp_restrict_address_families(Set
*address_families
, bool whitelist
) {
1075 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
1076 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
1080 log_debug("Operating on architecture: %s", seccomp_arch_to_string(arch
));
1084 case SCMP_ARCH_X86_64
:
1087 case SCMP_ARCH_AARCH64
:
1088 case SCMP_ARCH_PPC64
:
1089 case SCMP_ARCH_PPC64LE
:
1090 /* These we know we support (i.e. are the ones that do not use socketcall()) */
1094 case SCMP_ARCH_S390
:
1095 case SCMP_ARCH_S390X
:
1099 /* These we either know we don't support (i.e. are the ones that do use socketcall()), or we
1108 r
= seccomp_init_for_arch(&seccomp
, arch
, SCMP_ACT_ALLOW
);
1113 int af
, first
= 0, last
= 0;
1116 /* If this is a whitelist, we first block the address families that are out of range and then
1117 * everything that is not in the set. First, we find the lowest and highest address family in
1120 SET_FOREACH(afp
, address_families
, i
) {
1121 af
= PTR_TO_INT(afp
);
1123 if (af
<= 0 || af
>= af_max())
1126 if (first
== 0 || af
< first
)
1129 if (last
== 0 || af
> last
)
1133 assert((first
== 0) == (last
== 0));
1137 /* No entries in the valid range, block everything */
1138 r
= seccomp_rule_add_exact(
1140 SCMP_ACT_ERRNO(EAFNOSUPPORT
),
1144 log_debug_errno(r
, "Failed to add socket() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1150 /* Block everything below the first entry */
1151 r
= seccomp_rule_add_exact(
1153 SCMP_ACT_ERRNO(EAFNOSUPPORT
),
1156 SCMP_A0(SCMP_CMP_LT
, first
));
1158 log_debug_errno(r
, "Failed to add socket() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1162 /* Block everything above the last entry */
1163 r
= seccomp_rule_add_exact(
1165 SCMP_ACT_ERRNO(EAFNOSUPPORT
),
1168 SCMP_A0(SCMP_CMP_GT
, last
));
1170 log_debug_errno(r
, "Failed to add socket() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1174 /* Block everything between the first and last entry */
1175 for (af
= 1; af
< af_max(); af
++) {
1177 if (set_contains(address_families
, INT_TO_PTR(af
)))
1180 r
= seccomp_rule_add_exact(
1182 SCMP_ACT_ERRNO(EAFNOSUPPORT
),
1185 SCMP_A0(SCMP_CMP_EQ
, af
));
1190 log_debug_errno(r
, "Failed to add socket() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1198 /* If this is a blacklist, then generate one rule for
1199 * each address family that are then combined in OR
1202 SET_FOREACH(af
, address_families
, i
) {
1204 r
= seccomp_rule_add_exact(
1206 SCMP_ACT_ERRNO(EAFNOSUPPORT
),
1209 SCMP_A0(SCMP_CMP_EQ
, PTR_TO_INT(af
)));
1214 log_debug_errno(r
, "Failed to add socket() rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1219 r
= seccomp_load(seccomp
);
1220 if (IN_SET(r
, -EPERM
, -EACCES
))
1223 log_debug_errno(r
, "Failed to install socket family rules for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1229 int seccomp_restrict_realtime(void) {
1230 static const int permitted_policies
[] = {
1236 int r
, max_policy
= 0;
1240 /* Determine the highest policy constant we want to allow */
1241 for (i
= 0; i
< ELEMENTSOF(permitted_policies
); i
++)
1242 if (permitted_policies
[i
] > max_policy
)
1243 max_policy
= permitted_policies
[i
];
1245 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
1246 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
1249 log_debug("Operating on architecture: %s", seccomp_arch_to_string(arch
));
1251 r
= seccomp_init_for_arch(&seccomp
, arch
, SCMP_ACT_ALLOW
);
1255 /* Go through all policies with lower values than that, and block them -- unless they appear in the
1257 for (p
= 0; p
< max_policy
; p
++) {
1260 /* Check if this is in the whitelist. */
1261 for (i
= 0; i
< ELEMENTSOF(permitted_policies
); i
++)
1262 if (permitted_policies
[i
] == p
) {
1270 /* Deny this policy */
1271 r
= seccomp_rule_add_exact(
1273 SCMP_ACT_ERRNO(EPERM
),
1274 SCMP_SYS(sched_setscheduler
),
1276 SCMP_A1(SCMP_CMP_EQ
, p
));
1278 log_debug_errno(r
, "Failed to add scheduler rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1283 /* Blacklist all other policies, i.e. the ones with higher values. Note that all comparisons are
1284 * unsigned here, hence no need no check for < 0 values. */
1285 r
= seccomp_rule_add_exact(
1287 SCMP_ACT_ERRNO(EPERM
),
1288 SCMP_SYS(sched_setscheduler
),
1290 SCMP_A1(SCMP_CMP_GT
, max_policy
));
1292 log_debug_errno(r
, "Failed to add scheduler rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1296 r
= seccomp_load(seccomp
);
1297 if (IN_SET(r
, -EPERM
, -EACCES
))
1300 log_debug_errno(r
, "Failed to install realtime protection rules for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1306 static int add_seccomp_syscall_filter(scmp_filter_ctx seccomp
,
1309 unsigned int arg_cnt
,
1310 const struct scmp_arg_cmp arg
) {
1313 r
= seccomp_rule_add_exact(seccomp
, SCMP_ACT_ERRNO(EPERM
), nr
, arg_cnt
, arg
);
1315 _cleanup_free_
char *n
= NULL
;
1317 n
= seccomp_syscall_resolve_num_arch(arch
, nr
);
1318 log_debug_errno(r
, "Failed to add %s() rule for architecture %s, skipping: %m",
1320 seccomp_arch_to_string(arch
));
1326 /* For known architectures, check that syscalls are indeed defined or not. */
1327 #if defined(__x86_64__) || defined(__arm__) || defined(__aarch64__)
1328 assert_cc(SCMP_SYS(shmget
) > 0);
1329 assert_cc(SCMP_SYS(shmat
) > 0);
1330 assert_cc(SCMP_SYS(shmdt
) > 0);
1331 #elif defined(__i386__) || defined(__powerpc64__)
1332 assert_cc(SCMP_SYS(shmget
) < 0);
1333 assert_cc(SCMP_SYS(shmat
) < 0);
1334 assert_cc(SCMP_SYS(shmdt
) < 0);
1337 int seccomp_memory_deny_write_execute(void) {
1342 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
1343 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
1344 int filter_syscall
= 0, block_syscall
= 0, shmat_syscall
= 0;
1346 log_debug("Operating on architecture: %s", seccomp_arch_to_string(arch
));
1351 filter_syscall
= SCMP_SYS(mmap2
);
1352 block_syscall
= SCMP_SYS(mmap
);
1355 case SCMP_ARCH_PPC64
:
1356 case SCMP_ARCH_PPC64LE
:
1357 filter_syscall
= SCMP_SYS(mmap
);
1359 /* Note that shmat() isn't available, and the call is multiplexed through ipc().
1360 * We ignore that here, which means there's still a way to get writable/executable
1361 * memory, if an IPC key is mapped like this. That's a pity, but no total loss. */
1366 filter_syscall
= SCMP_SYS(mmap2
); /* arm has only mmap2 */
1367 shmat_syscall
= SCMP_SYS(shmat
);
1370 case SCMP_ARCH_X86_64
:
1372 case SCMP_ARCH_AARCH64
:
1373 filter_syscall
= SCMP_SYS(mmap
); /* amd64, x32, and arm64 have only mmap */
1374 shmat_syscall
= SCMP_SYS(shmat
);
1377 /* Please add more definitions here, if you port systemd to other architectures! */
1379 #if !defined(__i386__) && !defined(__x86_64__) && !defined(__powerpc64__) && !defined(__arm__) && !defined(__aarch64__)
1380 #warning "Consider adding the right mmap() syscall definitions here!"
1384 /* Can't filter mmap() on this arch, then skip it */
1385 if (filter_syscall
== 0)
1388 r
= seccomp_init_for_arch(&seccomp
, arch
, SCMP_ACT_ALLOW
);
1392 r
= add_seccomp_syscall_filter(seccomp
, arch
, filter_syscall
,
1394 SCMP_A2(SCMP_CMP_MASKED_EQ
, PROT_EXEC
|PROT_WRITE
, PROT_EXEC
|PROT_WRITE
));
1398 if (block_syscall
!= 0) {
1399 r
= add_seccomp_syscall_filter(seccomp
, arch
, block_syscall
, 0, (const struct scmp_arg_cmp
){} );
1404 r
= add_seccomp_syscall_filter(seccomp
, arch
, SCMP_SYS(mprotect
),
1406 SCMP_A2(SCMP_CMP_MASKED_EQ
, PROT_EXEC
, PROT_EXEC
));
1410 if (shmat_syscall
!= 0) {
1411 r
= add_seccomp_syscall_filter(seccomp
, arch
, SCMP_SYS(shmat
),
1413 SCMP_A2(SCMP_CMP_MASKED_EQ
, SHM_EXEC
, SHM_EXEC
));
1418 r
= seccomp_load(seccomp
);
1419 if (IN_SET(r
, -EPERM
, -EACCES
))
1422 log_debug_errno(r
, "Failed to install MemoryDenyWriteExecute= rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1428 int seccomp_restrict_archs(Set
*archs
) {
1429 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
1434 /* This installs a filter with no rules, but that restricts the system call architectures to the specified
1437 seccomp
= seccomp_init(SCMP_ACT_ALLOW
);
1441 SET_FOREACH(id
, archs
, i
) {
1442 r
= seccomp_arch_add(seccomp
, PTR_TO_UINT32(id
) - 1);
1449 r
= seccomp_attr_set(seccomp
, SCMP_FLTATR_CTL_NNP
, 0);
1453 r
= seccomp_load(seccomp
);
1454 if (IN_SET(r
, -EPERM
, -EACCES
))
1457 log_debug_errno(r
, "Failed to restrict system call architectures, skipping: %m");
1462 int parse_syscall_archs(char **l
, Set
**archs
) {
1463 _cleanup_set_free_ Set
*_archs
;
1470 r
= set_ensure_allocated(&_archs
, NULL
);
1474 STRV_FOREACH(s
, l
) {
1477 r
= seccomp_arch_from_string(*s
, &a
);
1481 r
= set_put(_archs
, UINT32_TO_PTR(a
+ 1));
1492 int seccomp_filter_set_add(Set
*filter
, bool add
, const SyscallFilterSet
*set
) {
1498 NULSTR_FOREACH(i
, set
->value
) {
1501 const SyscallFilterSet
*more
;
1503 more
= syscall_filter_set_find(i
);
1507 r
= seccomp_filter_set_add(filter
, add
, more
);
1513 id
= seccomp_syscall_resolve_name(i
);
1514 if (id
== __NR_SCMP_ERROR
) {
1515 log_debug("Couldn't resolve system call, ignoring: %s", i
);
1520 r
= set_put(filter
, INT_TO_PTR(id
+ 1));
1524 (void) set_remove(filter
, INT_TO_PTR(id
+ 1));
1531 int seccomp_lock_personality(unsigned long personality
) {
1535 if (personality
>= PERSONALITY_INVALID
)
1538 SECCOMP_FOREACH_LOCAL_ARCH(arch
) {
1539 _cleanup_(seccomp_releasep
) scmp_filter_ctx seccomp
= NULL
;
1541 r
= seccomp_init_for_arch(&seccomp
, arch
, SCMP_ACT_ALLOW
);
1545 r
= seccomp_rule_add_exact(
1547 SCMP_ACT_ERRNO(EPERM
),
1548 SCMP_SYS(personality
),
1550 SCMP_A0(SCMP_CMP_NE
, personality
));
1552 log_debug_errno(r
, "Failed to add scheduler rule for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));
1556 r
= seccomp_load(seccomp
);
1557 if (IN_SET(r
, -EPERM
, -EACCES
))
1560 log_debug_errno(r
, "Failed to enable personality lock for architecture %s, skipping: %m", seccomp_arch_to_string(arch
));