]> git.ipfire.org Git - thirdparty/systemd.git/blame - src/nspawn/nspawn-seccomp.c
Merge pull request #14099 from keszybz/machine-ref-unref-fix
[thirdparty/systemd.git] / src / nspawn / nspawn-seccomp.c
CommitLineData
53e1b683 1/* SPDX-License-Identifier: LGPL-2.1+ */
f011b0b8
DH
2
3#include <errno.h>
4#include <linux/netlink.h>
5#include <sys/capability.h>
ca78ad1d 6#include <sys/socket.h>
f011b0b8
DH
7#include <sys/types.h>
8
349cc4a5 9#if HAVE_SECCOMP
f011b0b8
DH
10#include <seccomp.h>
11#endif
12
469830d1 13#include "alloc-util.h"
f011b0b8 14#include "log.h"
469830d1 15#include "nspawn-seccomp.h"
349cc4a5 16#if HAVE_SECCOMP
f011b0b8
DH
17#include "seccomp-util.h"
18#endif
469830d1 19#include "string-util.h"
960e4569 20#include "strv.h"
f011b0b8 21
349cc4a5 22#if HAVE_SECCOMP
f011b0b8 23
469830d1
LP
24static int seccomp_add_default_syscall_filter(
25 scmp_filter_ctx ctx,
26 uint32_t arch,
960e4569
LP
27 uint64_t cap_list_retain,
28 char **syscall_whitelist,
29 char **syscall_blacklist) {
469830d1 30
f011b0b8
DH
31 static const struct {
32 uint64_t capability;
402530d9 33 const char* name;
96bedbe2
LP
34 } whitelist[] = {
35 /* Let's use set names where we can */
44898c53 36 { 0, "@aio" },
96bedbe2 37 { 0, "@basic-io" },
44898c53 38 { 0, "@chown" },
96bedbe2
LP
39 { 0, "@default" },
40 { 0, "@file-system" },
41 { 0, "@io-event" },
42 { 0, "@ipc" },
43 { 0, "@mount" },
44 { 0, "@network-io" },
45 { 0, "@process" },
46 { 0, "@resources" },
47 { 0, "@setuid" },
48 { 0, "@signal" },
44898c53 49 { 0, "@sync" },
96bedbe2
LP
50 { 0, "@timer" },
51
52 /* The following four are sets we optionally enable, in case the caps have been configured for it */
53 { CAP_SYS_TIME, "@clock" },
54 { CAP_SYS_MODULE, "@module" },
55 { CAP_SYS_RAWIO, "@raw-io" },
56 { CAP_IPC_LOCK, "@memlock" },
57
58 /* Plus a good set of additional syscalls which are not part of any of the groups above */
59 { 0, "brk" },
09d3020b 60 { 0, "capget" },
96bedbe2 61 { 0, "capset" },
96bedbe2
LP
62 { 0, "copy_file_range" },
63 { 0, "fadvise64" },
64 { 0, "fadvise64_64" },
96bedbe2 65 { 0, "flock" },
96bedbe2
LP
66 { 0, "get_mempolicy" },
67 { 0, "getcpu" },
68 { 0, "getpriority" },
69 { 0, "getrandom" },
96bedbe2
LP
70 { 0, "ioctl" },
71 { 0, "ioprio_get" },
72 { 0, "kcmp" },
96bedbe2
LP
73 { 0, "madvise" },
74 { 0, "mincore" },
75 { 0, "mprotect" },
76 { 0, "mremap" },
96bedbe2
LP
77 { 0, "name_to_handle_at" },
78 { 0, "oldolduname" },
79 { 0, "olduname" },
80 { 0, "personality" },
96bedbe2
LP
81 { 0, "readahead" },
82 { 0, "readdir" },
83 { 0, "remap_file_pages" },
84 { 0, "sched_get_priority_max" },
85 { 0, "sched_get_priority_min" },
86 { 0, "sched_getaffinity" },
87 { 0, "sched_getattr" },
88 { 0, "sched_getparam" },
89 { 0, "sched_getscheduler" },
90 { 0, "sched_rr_get_interval" },
91 { 0, "sched_yield" },
92 { 0, "seccomp" },
93 { 0, "sendfile" },
94 { 0, "sendfile64" },
95 { 0, "setdomainname" },
96 { 0, "setfsgid" },
97 { 0, "setfsgid32" },
98 { 0, "setfsuid" },
99 { 0, "setfsuid32" },
100 { 0, "sethostname" },
101 { 0, "setpgid" },
102 { 0, "setsid" },
103 { 0, "splice" },
96bedbe2
LP
104 { 0, "sysinfo" },
105 { 0, "tee" },
96bedbe2
LP
106 { 0, "umask" },
107 { 0, "uname" },
108 { 0, "userfaultfd" },
109 { 0, "vmsplice" },
110
111 /* The following individual syscalls are added depending on specified caps */
112 { CAP_SYS_PACCT, "acct" },
113 { CAP_SYS_PTRACE, "process_vm_readv" },
114 { CAP_SYS_PTRACE, "process_vm_writev" },
115 { CAP_SYS_PTRACE, "ptrace" },
116 { CAP_SYS_BOOT, "reboot" },
117 { CAP_SYSLOG, "syslog" },
118 { CAP_SYS_TTY_CONFIG, "vhangup" },
119
120 /*
121 * The following syscalls and groups are knowingly excluded:
122 *
123 * @cpu-emulation
124 * @keyring (NB: keyring is not namespaced!)
125 * @obsolete
9493b168 126 * @pkey
96bedbe2
LP
127 * @swap
128 *
129 * bpf (NB: bpffs is not namespaced!)
130 * fanotify_init
131 * fanotify_mark
132 * kexec_file_load
133 * kexec_load
134 * lookup_dcookie
135 * nfsservctl
136 * open_by_handle_at
137 * perf_event_open
96bedbe2
LP
138 * quotactl
139 */
f011b0b8 140 };
402530d9 141
960e4569 142 char **p;
f47bd097 143 int r;
f011b0b8 144
f47bd097 145 for (size_t i = 0; i < ELEMENTSOF(whitelist); i++) {
96bedbe2 146 if (whitelist[i].capability != 0 && (cap_list_retain & (1ULL << whitelist[i].capability)) == 0)
f011b0b8
DH
147 continue;
148
b54f36c6 149 r = seccomp_add_syscall_filter_item(ctx, whitelist[i].name, SCMP_ACT_ALLOW, syscall_blacklist, false);
402530d9 150 if (r < 0)
7e86bd73 151 return log_error_errno(r, "Failed to add syscall filter item %s: %m", whitelist[i].name);
f011b0b8
DH
152 }
153
96bedbe2 154 STRV_FOREACH(p, syscall_whitelist) {
f47bd097 155 r = seccomp_add_syscall_filter_item(ctx, *p, SCMP_ACT_ALLOW, syscall_blacklist, true);
960e4569 156 if (r < 0)
7e86bd73
ZJS
157 log_warning_errno(r, "Failed to add rule for system call %s on %s, ignoring: %m",
158 *p, seccomp_arch_to_string(arch));
960e4569
LP
159 }
160
7e86bd73 161 return 0;
f011b0b8
DH
162}
163
960e4569 164int setup_seccomp(uint64_t cap_list_retain, char **syscall_whitelist, char **syscall_blacklist) {
469830d1 165 uint32_t arch;
f011b0b8
DH
166 int r;
167
1cec406d 168 if (!is_seccomp_available()) {
960e4569 169 log_debug("SECCOMP features not detected in the kernel, disabling SECCOMP filterering");
1cec406d
FS
170 return 0;
171 }
172
469830d1
LP
173 SECCOMP_FOREACH_LOCAL_ARCH(arch) {
174 _cleanup_(seccomp_releasep) scmp_filter_ctx seccomp = NULL;
469830d1 175
96bedbe2 176 log_debug("Applying whitelist on architecture: %s", seccomp_arch_to_string(arch));
469830d1 177
96bedbe2 178 r = seccomp_init_for_arch(&seccomp, arch, SCMP_ACT_ERRNO(EPERM));
469830d1
LP
179 if (r < 0)
180 return log_error_errno(r, "Failed to allocate seccomp object: %m");
181
96bedbe2
LP
182 r = seccomp_add_default_syscall_filter(seccomp, arch, cap_list_retain, syscall_whitelist, syscall_blacklist);
183 if (r < 0)
184 return r;
185
186 r = seccomp_load(seccomp);
7bc5e0b1 187 if (ERRNO_IS_SECCOMP_FATAL(r))
96bedbe2
LP
188 return log_error_errno(r, "Failed to install seccomp filter: %m");
189 if (r < 0)
190 log_debug_errno(r, "Failed to install filter set for architecture %s, skipping: %m", seccomp_arch_to_string(arch));
191 }
192
193 SECCOMP_FOREACH_LOCAL_ARCH(arch) {
194 _cleanup_(seccomp_releasep) scmp_filter_ctx seccomp = NULL;
195
196 log_debug("Applying NETLINK_AUDIT mask on architecture: %s", seccomp_arch_to_string(arch));
197
198 r = seccomp_init_for_arch(&seccomp, arch, SCMP_ACT_ALLOW);
199 if (r < 0)
200 return log_error_errno(r, "Failed to allocate seccomp object: %m");
469830d1
LP
201
202 /*
203 Audit is broken in containers, much of the userspace audit hookup will fail if running inside a
204 container. We don't care and just turn off creation of audit sockets.
205
206 This will make socket(AF_NETLINK, *, NETLINK_AUDIT) fail with EAFNOSUPPORT which audit userspace uses
207 as indication that audit is disabled in the kernel.
208 */
209
210 r = seccomp_rule_add_exact(
211 seccomp,
212 SCMP_ACT_ERRNO(EAFNOSUPPORT),
213 SCMP_SYS(socket),
214 2,
215 SCMP_A0(SCMP_CMP_EQ, AF_NETLINK),
216 SCMP_A2(SCMP_CMP_EQ, NETLINK_AUDIT));
96bedbe2 217 if (r < 0) {
469830d1 218 log_debug_errno(r, "Failed to add audit seccomp rule, ignoring: %m");
469830d1 219 continue;
96bedbe2 220 }
f011b0b8 221
469830d1 222 r = seccomp_load(seccomp);
7bc5e0b1 223 if (ERRNO_IS_SECCOMP_FATAL(r))
469830d1
LP
224 return log_error_errno(r, "Failed to install seccomp audit filter: %m");
225 if (r < 0)
226 log_debug_errno(r, "Failed to install filter set for architecture %s, skipping: %m", seccomp_arch_to_string(arch));
f011b0b8
DH
227 }
228
469830d1 229 return 0;
f011b0b8
DH
230}
231
232#else
233
960e4569 234int setup_seccomp(uint64_t cap_list_retain, char **syscall_whitelist, char **syscall_blacklist) {
f011b0b8
DH
235 return 0;
236}
237
238#endif