]> git.ipfire.org Git - thirdparty/systemd.git/blob - src/nspawn/nspawn-seccomp.c
tree-wide: remove Lennart's copyright lines
[thirdparty/systemd.git] / src / nspawn / nspawn-seccomp.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #include <errno.h>
4 #include <linux/netlink.h>
5 #include <sys/capability.h>
6 #include <sys/types.h>
7
8 #if HAVE_SECCOMP
9 #include <seccomp.h>
10 #endif
11
12 #include "alloc-util.h"
13 #include "log.h"
14 #include "nspawn-seccomp.h"
15 #if HAVE_SECCOMP
16 #include "seccomp-util.h"
17 #endif
18 #include "string-util.h"
19 #include "strv.h"
20
21 #if HAVE_SECCOMP
22
23 static int seccomp_add_default_syscall_filter(
24 scmp_filter_ctx ctx,
25 uint32_t arch,
26 uint64_t cap_list_retain,
27 char **syscall_whitelist,
28 char **syscall_blacklist) {
29
30 static const struct {
31 uint64_t capability;
32 const char* name;
33 } whitelist[] = {
34 /* Let's use set names where we can */
35 { 0, "@aio" },
36 { 0, "@basic-io" },
37 { 0, "@chown" },
38 { 0, "@default" },
39 { 0, "@file-system" },
40 { 0, "@io-event" },
41 { 0, "@ipc" },
42 { 0, "@mount" },
43 { 0, "@network-io" },
44 { 0, "@process" },
45 { 0, "@resources" },
46 { 0, "@setuid" },
47 { 0, "@signal" },
48 { 0, "@sync" },
49 { 0, "@timer" },
50
51 /* The following four are sets we optionally enable, in case the caps have been configured for it */
52 { CAP_SYS_TIME, "@clock" },
53 { CAP_SYS_MODULE, "@module" },
54 { CAP_SYS_RAWIO, "@raw-io" },
55 { CAP_IPC_LOCK, "@memlock" },
56
57 /* Plus a good set of additional syscalls which are not part of any of the groups above */
58 { 0, "brk" },
59 { 0, "capget" },
60 { 0, "capset" },
61 { 0, "copy_file_range" },
62 { 0, "fadvise64" },
63 { 0, "fadvise64_64" },
64 { 0, "flock" },
65 { 0, "get_mempolicy" },
66 { 0, "getcpu" },
67 { 0, "getpriority" },
68 { 0, "getrandom" },
69 { 0, "ioctl" },
70 { 0, "ioprio_get" },
71 { 0, "kcmp" },
72 { 0, "madvise" },
73 { 0, "mincore" },
74 { 0, "mprotect" },
75 { 0, "mremap" },
76 { 0, "name_to_handle_at" },
77 { 0, "oldolduname" },
78 { 0, "olduname" },
79 { 0, "personality" },
80 { 0, "readahead" },
81 { 0, "readdir" },
82 { 0, "remap_file_pages" },
83 { 0, "sched_get_priority_max" },
84 { 0, "sched_get_priority_min" },
85 { 0, "sched_getaffinity" },
86 { 0, "sched_getattr" },
87 { 0, "sched_getparam" },
88 { 0, "sched_getscheduler" },
89 { 0, "sched_rr_get_interval" },
90 { 0, "sched_yield" },
91 { 0, "seccomp" },
92 { 0, "sendfile" },
93 { 0, "sendfile64" },
94 { 0, "setdomainname" },
95 { 0, "setfsgid" },
96 { 0, "setfsgid32" },
97 { 0, "setfsuid" },
98 { 0, "setfsuid32" },
99 { 0, "sethostname" },
100 { 0, "setpgid" },
101 { 0, "setsid" },
102 { 0, "splice" },
103 { 0, "sysinfo" },
104 { 0, "tee" },
105 { 0, "umask" },
106 { 0, "uname" },
107 { 0, "userfaultfd" },
108 { 0, "vmsplice" },
109
110 /* The following individual syscalls are added depending on specified caps */
111 { CAP_SYS_PACCT, "acct" },
112 { CAP_SYS_PTRACE, "process_vm_readv" },
113 { CAP_SYS_PTRACE, "process_vm_writev" },
114 { CAP_SYS_PTRACE, "ptrace" },
115 { CAP_SYS_BOOT, "reboot" },
116 { CAP_SYSLOG, "syslog" },
117 { CAP_SYS_TTY_CONFIG, "vhangup" },
118
119 /*
120 * The following syscalls and groups are knowingly excluded:
121 *
122 * @cpu-emulation
123 * @keyring (NB: keyring is not namespaced!)
124 * @obsolete
125 * @swap
126 *
127 * bpf (NB: bpffs is not namespaced!)
128 * fanotify_init
129 * fanotify_mark
130 * kexec_file_load
131 * kexec_load
132 * lookup_dcookie
133 * nfsservctl
134 * open_by_handle_at
135 * perf_event_open
136 * pkey_alloc
137 * pkey_free
138 * pkey_mprotect
139 * quotactl
140 */
141 };
142
143 int r, c = 0;
144 size_t i;
145 char **p;
146
147 for (i = 0; i < ELEMENTSOF(whitelist); i++) {
148 if (whitelist[i].capability != 0 && (cap_list_retain & (1ULL << whitelist[i].capability)) == 0)
149 continue;
150
151 r = seccomp_add_syscall_filter_item(ctx, whitelist[i].name, SCMP_ACT_ALLOW, syscall_blacklist);
152 if (r < 0)
153 /* If the system call is not known on this architecture, then that's fine, let's ignore it */
154 log_debug_errno(r, "Failed to add rule for system call %s on %s, ignoring: %m", whitelist[i].name, seccomp_arch_to_string(arch));
155 else
156 c++;
157 }
158
159 STRV_FOREACH(p, syscall_whitelist) {
160 r = seccomp_add_syscall_filter_item(ctx, *p, SCMP_ACT_ALLOW, syscall_blacklist);
161 if (r < 0)
162 log_debug_errno(r, "Failed to add rule for system call %s on %s, ignoring: %m", *p, seccomp_arch_to_string(arch));
163 else
164 c++;
165 }
166
167 return c;
168 }
169
170 int setup_seccomp(uint64_t cap_list_retain, char **syscall_whitelist, char **syscall_blacklist) {
171 uint32_t arch;
172 int r;
173
174 if (!is_seccomp_available()) {
175 log_debug("SECCOMP features not detected in the kernel, disabling SECCOMP filterering");
176 return 0;
177 }
178
179 SECCOMP_FOREACH_LOCAL_ARCH(arch) {
180 _cleanup_(seccomp_releasep) scmp_filter_ctx seccomp = NULL;
181
182 log_debug("Applying whitelist on architecture: %s", seccomp_arch_to_string(arch));
183
184 r = seccomp_init_for_arch(&seccomp, arch, SCMP_ACT_ERRNO(EPERM));
185 if (r < 0)
186 return log_error_errno(r, "Failed to allocate seccomp object: %m");
187
188 r = seccomp_add_default_syscall_filter(seccomp, arch, cap_list_retain, syscall_whitelist, syscall_blacklist);
189 if (r < 0)
190 return r;
191
192 r = seccomp_load(seccomp);
193 if (IN_SET(r, -EPERM, -EACCES))
194 return log_error_errno(r, "Failed to install seccomp filter: %m");
195 if (r < 0)
196 log_debug_errno(r, "Failed to install filter set for architecture %s, skipping: %m", seccomp_arch_to_string(arch));
197 }
198
199 SECCOMP_FOREACH_LOCAL_ARCH(arch) {
200 _cleanup_(seccomp_releasep) scmp_filter_ctx seccomp = NULL;
201
202 log_debug("Applying NETLINK_AUDIT mask on architecture: %s", seccomp_arch_to_string(arch));
203
204 r = seccomp_init_for_arch(&seccomp, arch, SCMP_ACT_ALLOW);
205 if (r < 0)
206 return log_error_errno(r, "Failed to allocate seccomp object: %m");
207
208 /*
209 Audit is broken in containers, much of the userspace audit hookup will fail if running inside a
210 container. We don't care and just turn off creation of audit sockets.
211
212 This will make socket(AF_NETLINK, *, NETLINK_AUDIT) fail with EAFNOSUPPORT which audit userspace uses
213 as indication that audit is disabled in the kernel.
214 */
215
216 r = seccomp_rule_add_exact(
217 seccomp,
218 SCMP_ACT_ERRNO(EAFNOSUPPORT),
219 SCMP_SYS(socket),
220 2,
221 SCMP_A0(SCMP_CMP_EQ, AF_NETLINK),
222 SCMP_A2(SCMP_CMP_EQ, NETLINK_AUDIT));
223 if (r < 0) {
224 log_debug_errno(r, "Failed to add audit seccomp rule, ignoring: %m");
225 continue;
226 }
227
228 r = seccomp_load(seccomp);
229 if (IN_SET(r, -EPERM, -EACCES))
230 return log_error_errno(r, "Failed to install seccomp audit filter: %m");
231 if (r < 0)
232 log_debug_errno(r, "Failed to install filter set for architecture %s, skipping: %m", seccomp_arch_to_string(arch));
233 }
234
235 return 0;
236 }
237
238 #else
239
240 int setup_seccomp(uint64_t cap_list_retain, char **syscall_whitelist, char **syscall_blacklist) {
241 return 0;
242 }
243
244 #endif