configure: do not limit Hypervisor.framework test to Darwin
[qemu.git] / qemu-seccomp.c
blob8325ecb766e0f3fb27ff85a84a301a66a0a7a0fd
1 /*
2 * QEMU seccomp mode 2 support with libseccomp
4 * Copyright IBM, Corp. 2012
6 * Authors:
7 * Eduardo Otubo <eotubo@br.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
16 #include "qemu/osdep.h"
17 #include "qapi/error.h"
18 #include "qemu/config-file.h"
19 #include "qemu/option.h"
20 #include "qemu/module.h"
21 #include <sys/prctl.h>
22 #include <seccomp.h>
23 #include "sysemu/seccomp.h"
24 #include <linux/seccomp.h>
26 /* For some architectures (notably ARM) cacheflush is not supported until
27 * libseccomp 2.2.3, but configure enforces that we are using a more recent
28 * version on those hosts, so it is OK for this check to be less strict.
30 #if SCMP_VER_MAJOR >= 3
31 #define HAVE_CACHEFLUSH
32 #elif SCMP_VER_MAJOR == 2 && SCMP_VER_MINOR >= 2
33 #define HAVE_CACHEFLUSH
34 #endif
36 struct QemuSeccompSyscall {
37 int32_t num;
38 uint8_t set;
39 uint8_t narg;
40 const struct scmp_arg_cmp *arg_cmp;
43 const struct scmp_arg_cmp sched_setscheduler_arg[] = {
44 /* was SCMP_A1(SCMP_CMP_NE, SCHED_IDLE), but expanded due to GCC 4.x bug */
45 { .arg = 1, .op = SCMP_CMP_NE, .datum_a = SCHED_IDLE }
48 static const struct QemuSeccompSyscall blacklist[] = {
49 /* default set of syscalls to blacklist */
50 { SCMP_SYS(reboot), QEMU_SECCOMP_SET_DEFAULT },
51 { SCMP_SYS(swapon), QEMU_SECCOMP_SET_DEFAULT },
52 { SCMP_SYS(swapoff), QEMU_SECCOMP_SET_DEFAULT },
53 { SCMP_SYS(syslog), QEMU_SECCOMP_SET_DEFAULT },
54 { SCMP_SYS(mount), QEMU_SECCOMP_SET_DEFAULT },
55 { SCMP_SYS(umount), QEMU_SECCOMP_SET_DEFAULT },
56 { SCMP_SYS(kexec_load), QEMU_SECCOMP_SET_DEFAULT },
57 { SCMP_SYS(afs_syscall), QEMU_SECCOMP_SET_DEFAULT },
58 { SCMP_SYS(break), QEMU_SECCOMP_SET_DEFAULT },
59 { SCMP_SYS(ftime), QEMU_SECCOMP_SET_DEFAULT },
60 { SCMP_SYS(getpmsg), QEMU_SECCOMP_SET_DEFAULT },
61 { SCMP_SYS(gtty), QEMU_SECCOMP_SET_DEFAULT },
62 { SCMP_SYS(lock), QEMU_SECCOMP_SET_DEFAULT },
63 { SCMP_SYS(mpx), QEMU_SECCOMP_SET_DEFAULT },
64 { SCMP_SYS(prof), QEMU_SECCOMP_SET_DEFAULT },
65 { SCMP_SYS(profil), QEMU_SECCOMP_SET_DEFAULT },
66 { SCMP_SYS(putpmsg), QEMU_SECCOMP_SET_DEFAULT },
67 { SCMP_SYS(security), QEMU_SECCOMP_SET_DEFAULT },
68 { SCMP_SYS(stty), QEMU_SECCOMP_SET_DEFAULT },
69 { SCMP_SYS(tuxcall), QEMU_SECCOMP_SET_DEFAULT },
70 { SCMP_SYS(ulimit), QEMU_SECCOMP_SET_DEFAULT },
71 { SCMP_SYS(vserver), QEMU_SECCOMP_SET_DEFAULT },
72 /* obsolete */
73 { SCMP_SYS(readdir), QEMU_SECCOMP_SET_OBSOLETE },
74 { SCMP_SYS(_sysctl), QEMU_SECCOMP_SET_OBSOLETE },
75 { SCMP_SYS(bdflush), QEMU_SECCOMP_SET_OBSOLETE },
76 { SCMP_SYS(create_module), QEMU_SECCOMP_SET_OBSOLETE },
77 { SCMP_SYS(get_kernel_syms), QEMU_SECCOMP_SET_OBSOLETE },
78 { SCMP_SYS(query_module), QEMU_SECCOMP_SET_OBSOLETE },
79 { SCMP_SYS(sgetmask), QEMU_SECCOMP_SET_OBSOLETE },
80 { SCMP_SYS(ssetmask), QEMU_SECCOMP_SET_OBSOLETE },
81 { SCMP_SYS(sysfs), QEMU_SECCOMP_SET_OBSOLETE },
82 { SCMP_SYS(uselib), QEMU_SECCOMP_SET_OBSOLETE },
83 { SCMP_SYS(ustat), QEMU_SECCOMP_SET_OBSOLETE },
84 /* privileged */
85 { SCMP_SYS(setuid), QEMU_SECCOMP_SET_PRIVILEGED },
86 { SCMP_SYS(setgid), QEMU_SECCOMP_SET_PRIVILEGED },
87 { SCMP_SYS(setpgid), QEMU_SECCOMP_SET_PRIVILEGED },
88 { SCMP_SYS(setsid), QEMU_SECCOMP_SET_PRIVILEGED },
89 { SCMP_SYS(setreuid), QEMU_SECCOMP_SET_PRIVILEGED },
90 { SCMP_SYS(setregid), QEMU_SECCOMP_SET_PRIVILEGED },
91 { SCMP_SYS(setresuid), QEMU_SECCOMP_SET_PRIVILEGED },
92 { SCMP_SYS(setresgid), QEMU_SECCOMP_SET_PRIVILEGED },
93 { SCMP_SYS(setfsuid), QEMU_SECCOMP_SET_PRIVILEGED },
94 { SCMP_SYS(setfsgid), QEMU_SECCOMP_SET_PRIVILEGED },
95 /* spawn */
96 { SCMP_SYS(fork), QEMU_SECCOMP_SET_SPAWN },
97 { SCMP_SYS(vfork), QEMU_SECCOMP_SET_SPAWN },
98 { SCMP_SYS(execve), QEMU_SECCOMP_SET_SPAWN },
99 /* resource control */
100 { SCMP_SYS(getpriority), QEMU_SECCOMP_SET_RESOURCECTL },
101 { SCMP_SYS(setpriority), QEMU_SECCOMP_SET_RESOURCECTL },
102 { SCMP_SYS(sched_setparam), QEMU_SECCOMP_SET_RESOURCECTL },
103 { SCMP_SYS(sched_getparam), QEMU_SECCOMP_SET_RESOURCECTL },
104 { SCMP_SYS(sched_setscheduler), QEMU_SECCOMP_SET_RESOURCECTL,
105 ARRAY_SIZE(sched_setscheduler_arg), sched_setscheduler_arg },
106 { SCMP_SYS(sched_getscheduler), QEMU_SECCOMP_SET_RESOURCECTL },
107 { SCMP_SYS(sched_setaffinity), QEMU_SECCOMP_SET_RESOURCECTL },
108 { SCMP_SYS(sched_getaffinity), QEMU_SECCOMP_SET_RESOURCECTL },
109 { SCMP_SYS(sched_get_priority_max), QEMU_SECCOMP_SET_RESOURCECTL },
110 { SCMP_SYS(sched_get_priority_min), QEMU_SECCOMP_SET_RESOURCECTL },
113 static inline __attribute__((unused)) int
114 qemu_seccomp(unsigned int operation, unsigned int flags, void *args)
116 #ifdef __NR_seccomp
117 return syscall(__NR_seccomp, operation, flags, args);
118 #else
119 errno = ENOSYS;
120 return -1;
121 #endif
124 static uint32_t qemu_seccomp_get_action(int set)
126 switch (set) {
127 case QEMU_SECCOMP_SET_DEFAULT:
128 case QEMU_SECCOMP_SET_OBSOLETE:
129 case QEMU_SECCOMP_SET_PRIVILEGED:
130 case QEMU_SECCOMP_SET_SPAWN: {
131 #if defined(SECCOMP_GET_ACTION_AVAIL) && defined(SCMP_ACT_KILL_PROCESS) && \
132 defined(SECCOMP_RET_KILL_PROCESS)
133 static int kill_process = -1;
134 if (kill_process == -1) {
135 uint32_t action = SECCOMP_RET_KILL_PROCESS;
137 if (qemu_seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &action) == 0) {
138 kill_process = 1;
139 } else {
140 kill_process = 0;
143 if (kill_process == 1) {
144 return SCMP_ACT_KILL_PROCESS;
146 #endif
147 return SCMP_ACT_TRAP;
150 case QEMU_SECCOMP_SET_RESOURCECTL:
151 return SCMP_ACT_ERRNO(EPERM);
153 default:
154 g_assert_not_reached();
159 static int seccomp_start(uint32_t seccomp_opts, Error **errp)
161 int rc = -1;
162 unsigned int i = 0;
163 scmp_filter_ctx ctx;
165 ctx = seccomp_init(SCMP_ACT_ALLOW);
166 if (ctx == NULL) {
167 error_setg(errp, "failed to initialize seccomp context");
168 goto seccomp_return;
171 rc = seccomp_attr_set(ctx, SCMP_FLTATR_CTL_TSYNC, 1);
172 if (rc != 0) {
173 error_setg_errno(errp, -rc,
174 "failed to set seccomp thread synchronization");
175 goto seccomp_return;
178 for (i = 0; i < ARRAY_SIZE(blacklist); i++) {
179 uint32_t action;
180 if (!(seccomp_opts & blacklist[i].set)) {
181 continue;
184 action = qemu_seccomp_get_action(blacklist[i].set);
185 rc = seccomp_rule_add_array(ctx, action, blacklist[i].num,
186 blacklist[i].narg, blacklist[i].arg_cmp);
187 if (rc < 0) {
188 error_setg_errno(errp, -rc,
189 "failed to add seccomp blacklist rules");
190 goto seccomp_return;
194 rc = seccomp_load(ctx);
195 if (rc < 0) {
196 error_setg_errno(errp, -rc,
197 "failed to load seccomp syscall filter in kernel");
200 seccomp_return:
201 seccomp_release(ctx);
202 return rc < 0 ? -1 : 0;
205 #ifdef CONFIG_SECCOMP
206 int parse_sandbox(void *opaque, QemuOpts *opts, Error **errp)
208 if (qemu_opt_get_bool(opts, "enable", false)) {
209 uint32_t seccomp_opts = QEMU_SECCOMP_SET_DEFAULT
210 | QEMU_SECCOMP_SET_OBSOLETE;
211 const char *value = NULL;
213 value = qemu_opt_get(opts, "obsolete");
214 if (value) {
215 if (g_str_equal(value, "allow")) {
216 seccomp_opts &= ~QEMU_SECCOMP_SET_OBSOLETE;
217 } else if (g_str_equal(value, "deny")) {
218 /* this is the default option, this if is here
219 * to provide a little bit of consistency for
220 * the command line */
221 } else {
222 error_setg(errp, "invalid argument for obsolete");
223 return -1;
227 value = qemu_opt_get(opts, "elevateprivileges");
228 if (value) {
229 if (g_str_equal(value, "deny")) {
230 seccomp_opts |= QEMU_SECCOMP_SET_PRIVILEGED;
231 } else if (g_str_equal(value, "children")) {
232 seccomp_opts |= QEMU_SECCOMP_SET_PRIVILEGED;
234 /* calling prctl directly because we're
235 * not sure if host has CAP_SYS_ADMIN set*/
236 if (prctl(PR_SET_NO_NEW_PRIVS, 1)) {
237 error_setg(errp, "failed to set no_new_privs aborting");
238 return -1;
240 } else if (g_str_equal(value, "allow")) {
241 /* default value */
242 } else {
243 error_setg(errp, "invalid argument for elevateprivileges");
244 return -1;
248 value = qemu_opt_get(opts, "spawn");
249 if (value) {
250 if (g_str_equal(value, "deny")) {
251 seccomp_opts |= QEMU_SECCOMP_SET_SPAWN;
252 } else if (g_str_equal(value, "allow")) {
253 /* default value */
254 } else {
255 error_setg(errp, "invalid argument for spawn");
256 return -1;
260 value = qemu_opt_get(opts, "resourcecontrol");
261 if (value) {
262 if (g_str_equal(value, "deny")) {
263 seccomp_opts |= QEMU_SECCOMP_SET_RESOURCECTL;
264 } else if (g_str_equal(value, "allow")) {
265 /* default value */
266 } else {
267 error_setg(errp, "invalid argument for resourcecontrol");
268 return -1;
272 if (seccomp_start(seccomp_opts, errp) < 0) {
273 return -1;
277 return 0;
280 static QemuOptsList qemu_sandbox_opts = {
281 .name = "sandbox",
282 .implied_opt_name = "enable",
283 .head = QTAILQ_HEAD_INITIALIZER(qemu_sandbox_opts.head),
284 .desc = {
286 .name = "enable",
287 .type = QEMU_OPT_BOOL,
290 .name = "obsolete",
291 .type = QEMU_OPT_STRING,
294 .name = "elevateprivileges",
295 .type = QEMU_OPT_STRING,
298 .name = "spawn",
299 .type = QEMU_OPT_STRING,
302 .name = "resourcecontrol",
303 .type = QEMU_OPT_STRING,
305 { /* end of list */ }
309 static void seccomp_register(void)
311 bool add = false;
313 /* FIXME: use seccomp_api_get() >= 2 check when released */
315 #if defined(SECCOMP_FILTER_FLAG_TSYNC)
316 int check;
318 /* check host TSYNC capability, it returns errno == ENOSYS if unavailable */
319 check = qemu_seccomp(SECCOMP_SET_MODE_FILTER,
320 SECCOMP_FILTER_FLAG_TSYNC, NULL);
321 if (check < 0 && errno == EFAULT) {
322 add = true;
324 #endif
326 if (add) {
327 qemu_add_opts(&qemu_sandbox_opts);
330 opts_init(seccomp_register);
331 #endif