]>
Commit | Line | Data |
---|---|---|
2f668be7 EO |
1 | /* |
2 | * QEMU seccomp mode 2 support with libseccomp | |
3 | * | |
4 | * Copyright IBM, Corp. 2012 | |
5 | * | |
6 | * Authors: | |
7 | * Eduardo Otubo <[email protected]> | |
8 | * | |
9 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
10 | * the COPYING file in the top-level directory. | |
11 | * | |
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. | |
14 | */ | |
d38ea87a | 15 | #include "qemu/osdep.h" |
9d0fdecb YMZ |
16 | #include "qemu/config-file.h" |
17 | #include "qemu/option.h" | |
18 | #include "qemu/module.h" | |
19 | #include "qemu/error-report.h" | |
20 | #include <sys/prctl.h> | |
2f668be7 | 21 | #include <seccomp.h> |
9c17d615 | 22 | #include "sysemu/seccomp.h" |
bda08a57 | 23 | #include <linux/seccomp.h> |
2f668be7 | 24 | |
81bed73b JH |
25 | /* For some architectures (notably ARM) cacheflush is not supported until |
26 | * libseccomp 2.2.3, but configure enforces that we are using a more recent | |
27 | * version on those hosts, so it is OK for this check to be less strict. | |
28 | */ | |
47d2067a AJ |
29 | #if SCMP_VER_MAJOR >= 3 |
30 | #define HAVE_CACHEFLUSH | |
81bed73b | 31 | #elif SCMP_VER_MAJOR == 2 && SCMP_VER_MINOR >= 2 |
47d2067a AJ |
32 | #define HAVE_CACHEFLUSH |
33 | #endif | |
34 | ||
2f668be7 EO |
35 | struct QemuSeccompSyscall { |
36 | int32_t num; | |
1bd6152a | 37 | uint8_t set; |
056de1e8 MAL |
38 | uint8_t narg; |
39 | const struct scmp_arg_cmp *arg_cmp; | |
40 | }; | |
41 | ||
42 | const struct scmp_arg_cmp sched_setscheduler_arg[] = { | |
43 | SCMP_A1(SCMP_CMP_NE, SCHED_IDLE) | |
2f668be7 EO |
44 | }; |
45 | ||
1bd6152a EO |
46 | static const struct QemuSeccompSyscall blacklist[] = { |
47 | /* default set of syscalls to blacklist */ | |
48 | { SCMP_SYS(reboot), QEMU_SECCOMP_SET_DEFAULT }, | |
49 | { SCMP_SYS(swapon), QEMU_SECCOMP_SET_DEFAULT }, | |
50 | { SCMP_SYS(swapoff), QEMU_SECCOMP_SET_DEFAULT }, | |
51 | { SCMP_SYS(syslog), QEMU_SECCOMP_SET_DEFAULT }, | |
52 | { SCMP_SYS(mount), QEMU_SECCOMP_SET_DEFAULT }, | |
53 | { SCMP_SYS(umount), QEMU_SECCOMP_SET_DEFAULT }, | |
54 | { SCMP_SYS(kexec_load), QEMU_SECCOMP_SET_DEFAULT }, | |
55 | { SCMP_SYS(afs_syscall), QEMU_SECCOMP_SET_DEFAULT }, | |
56 | { SCMP_SYS(break), QEMU_SECCOMP_SET_DEFAULT }, | |
57 | { SCMP_SYS(ftime), QEMU_SECCOMP_SET_DEFAULT }, | |
58 | { SCMP_SYS(getpmsg), QEMU_SECCOMP_SET_DEFAULT }, | |
59 | { SCMP_SYS(gtty), QEMU_SECCOMP_SET_DEFAULT }, | |
60 | { SCMP_SYS(lock), QEMU_SECCOMP_SET_DEFAULT }, | |
61 | { SCMP_SYS(mpx), QEMU_SECCOMP_SET_DEFAULT }, | |
62 | { SCMP_SYS(prof), QEMU_SECCOMP_SET_DEFAULT }, | |
63 | { SCMP_SYS(profil), QEMU_SECCOMP_SET_DEFAULT }, | |
64 | { SCMP_SYS(putpmsg), QEMU_SECCOMP_SET_DEFAULT }, | |
65 | { SCMP_SYS(security), QEMU_SECCOMP_SET_DEFAULT }, | |
66 | { SCMP_SYS(stty), QEMU_SECCOMP_SET_DEFAULT }, | |
67 | { SCMP_SYS(tuxcall), QEMU_SECCOMP_SET_DEFAULT }, | |
68 | { SCMP_SYS(ulimit), QEMU_SECCOMP_SET_DEFAULT }, | |
69 | { SCMP_SYS(vserver), QEMU_SECCOMP_SET_DEFAULT }, | |
2b716fa6 EO |
70 | /* obsolete */ |
71 | { SCMP_SYS(readdir), QEMU_SECCOMP_SET_OBSOLETE }, | |
72 | { SCMP_SYS(_sysctl), QEMU_SECCOMP_SET_OBSOLETE }, | |
73 | { SCMP_SYS(bdflush), QEMU_SECCOMP_SET_OBSOLETE }, | |
74 | { SCMP_SYS(create_module), QEMU_SECCOMP_SET_OBSOLETE }, | |
75 | { SCMP_SYS(get_kernel_syms), QEMU_SECCOMP_SET_OBSOLETE }, | |
76 | { SCMP_SYS(query_module), QEMU_SECCOMP_SET_OBSOLETE }, | |
77 | { SCMP_SYS(sgetmask), QEMU_SECCOMP_SET_OBSOLETE }, | |
78 | { SCMP_SYS(ssetmask), QEMU_SECCOMP_SET_OBSOLETE }, | |
79 | { SCMP_SYS(sysfs), QEMU_SECCOMP_SET_OBSOLETE }, | |
80 | { SCMP_SYS(uselib), QEMU_SECCOMP_SET_OBSOLETE }, | |
81 | { SCMP_SYS(ustat), QEMU_SECCOMP_SET_OBSOLETE }, | |
73a1e647 EO |
82 | /* privileged */ |
83 | { SCMP_SYS(setuid), QEMU_SECCOMP_SET_PRIVILEGED }, | |
84 | { SCMP_SYS(setgid), QEMU_SECCOMP_SET_PRIVILEGED }, | |
85 | { SCMP_SYS(setpgid), QEMU_SECCOMP_SET_PRIVILEGED }, | |
86 | { SCMP_SYS(setsid), QEMU_SECCOMP_SET_PRIVILEGED }, | |
87 | { SCMP_SYS(setreuid), QEMU_SECCOMP_SET_PRIVILEGED }, | |
88 | { SCMP_SYS(setregid), QEMU_SECCOMP_SET_PRIVILEGED }, | |
89 | { SCMP_SYS(setresuid), QEMU_SECCOMP_SET_PRIVILEGED }, | |
90 | { SCMP_SYS(setresgid), QEMU_SECCOMP_SET_PRIVILEGED }, | |
91 | { SCMP_SYS(setfsuid), QEMU_SECCOMP_SET_PRIVILEGED }, | |
92 | { SCMP_SYS(setfsgid), QEMU_SECCOMP_SET_PRIVILEGED }, | |
995a226f EO |
93 | /* spawn */ |
94 | { SCMP_SYS(fork), QEMU_SECCOMP_SET_SPAWN }, | |
95 | { SCMP_SYS(vfork), QEMU_SECCOMP_SET_SPAWN }, | |
96 | { SCMP_SYS(execve), QEMU_SECCOMP_SET_SPAWN }, | |
24f8cdc5 EO |
97 | /* resource control */ |
98 | { SCMP_SYS(getpriority), QEMU_SECCOMP_SET_RESOURCECTL }, | |
99 | { SCMP_SYS(setpriority), QEMU_SECCOMP_SET_RESOURCECTL }, | |
100 | { SCMP_SYS(sched_setparam), QEMU_SECCOMP_SET_RESOURCECTL }, | |
101 | { SCMP_SYS(sched_getparam), QEMU_SECCOMP_SET_RESOURCECTL }, | |
056de1e8 MAL |
102 | { SCMP_SYS(sched_setscheduler), QEMU_SECCOMP_SET_RESOURCECTL, |
103 | ARRAY_SIZE(sched_setscheduler_arg), sched_setscheduler_arg }, | |
24f8cdc5 EO |
104 | { SCMP_SYS(sched_getscheduler), QEMU_SECCOMP_SET_RESOURCECTL }, |
105 | { SCMP_SYS(sched_setaffinity), QEMU_SECCOMP_SET_RESOURCECTL }, | |
106 | { SCMP_SYS(sched_getaffinity), QEMU_SECCOMP_SET_RESOURCECTL }, | |
107 | { SCMP_SYS(sched_get_priority_max), QEMU_SECCOMP_SET_RESOURCECTL }, | |
108 | { SCMP_SYS(sched_get_priority_min), QEMU_SECCOMP_SET_RESOURCECTL }, | |
2f668be7 EO |
109 | }; |
110 | ||
bda08a57 MAL |
111 | static inline __attribute__((unused)) int |
112 | qemu_seccomp(unsigned int operation, unsigned int flags, void *args) | |
113 | { | |
114 | #ifdef __NR_seccomp | |
115 | return syscall(__NR_seccomp, operation, flags, args); | |
116 | #else | |
117 | errno = ENOSYS; | |
118 | return -1; | |
119 | #endif | |
120 | } | |
121 | ||
122 | static uint32_t qemu_seccomp_get_kill_action(void) | |
123 | { | |
124 | #if defined(SECCOMP_GET_ACTION_AVAIL) && defined(SCMP_ACT_KILL_PROCESS) && \ | |
125 | defined(SECCOMP_RET_KILL_PROCESS) | |
126 | { | |
127 | uint32_t action = SECCOMP_RET_KILL_PROCESS; | |
128 | ||
129 | if (qemu_seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &action) == 0) { | |
130 | return SCMP_ACT_KILL_PROCESS; | |
131 | } | |
132 | } | |
133 | #endif | |
134 | ||
135 | return SCMP_ACT_TRAP; | |
136 | } | |
137 | ||
2b716fa6 | 138 | |
9d0fdecb | 139 | static int seccomp_start(uint32_t seccomp_opts) |
2f668be7 EO |
140 | { |
141 | int rc = 0; | |
142 | unsigned int i = 0; | |
143 | scmp_filter_ctx ctx; | |
bda08a57 | 144 | uint32_t action = qemu_seccomp_get_kill_action(); |
2f668be7 | 145 | |
1bd6152a | 146 | ctx = seccomp_init(SCMP_ACT_ALLOW); |
2f668be7 | 147 | if (ctx == NULL) { |
2a13f991 | 148 | rc = -1; |
2f668be7 EO |
149 | goto seccomp_return; |
150 | } | |
151 | ||
1bd6152a | 152 | for (i = 0; i < ARRAY_SIZE(blacklist); i++) { |
2b716fa6 EO |
153 | if (!(seccomp_opts & blacklist[i].set)) { |
154 | continue; | |
155 | } | |
156 | ||
bda08a57 | 157 | rc = seccomp_rule_add_array(ctx, action, blacklist[i].num, |
056de1e8 | 158 | blacklist[i].narg, blacklist[i].arg_cmp); |
2f668be7 EO |
159 | if (rc < 0) { |
160 | goto seccomp_return; | |
161 | } | |
162 | } | |
163 | ||
164 | rc = seccomp_load(ctx); | |
165 | ||
166 | seccomp_return: | |
167 | seccomp_release(ctx); | |
168 | return rc; | |
169 | } | |
9d0fdecb YMZ |
170 | |
171 | #ifdef CONFIG_SECCOMP | |
172 | int parse_sandbox(void *opaque, QemuOpts *opts, Error **errp) | |
173 | { | |
174 | if (qemu_opt_get_bool(opts, "enable", false)) { | |
175 | uint32_t seccomp_opts = QEMU_SECCOMP_SET_DEFAULT | |
176 | | QEMU_SECCOMP_SET_OBSOLETE; | |
177 | const char *value = NULL; | |
178 | ||
179 | value = qemu_opt_get(opts, "obsolete"); | |
180 | if (value) { | |
181 | if (g_str_equal(value, "allow")) { | |
182 | seccomp_opts &= ~QEMU_SECCOMP_SET_OBSOLETE; | |
183 | } else if (g_str_equal(value, "deny")) { | |
184 | /* this is the default option, this if is here | |
185 | * to provide a little bit of consistency for | |
186 | * the command line */ | |
187 | } else { | |
188 | error_report("invalid argument for obsolete"); | |
189 | return -1; | |
190 | } | |
191 | } | |
192 | ||
193 | value = qemu_opt_get(opts, "elevateprivileges"); | |
194 | if (value) { | |
195 | if (g_str_equal(value, "deny")) { | |
196 | seccomp_opts |= QEMU_SECCOMP_SET_PRIVILEGED; | |
197 | } else if (g_str_equal(value, "children")) { | |
198 | seccomp_opts |= QEMU_SECCOMP_SET_PRIVILEGED; | |
199 | ||
200 | /* calling prctl directly because we're | |
201 | * not sure if host has CAP_SYS_ADMIN set*/ | |
202 | if (prctl(PR_SET_NO_NEW_PRIVS, 1)) { | |
203 | error_report("failed to set no_new_privs " | |
204 | "aborting"); | |
205 | return -1; | |
206 | } | |
207 | } else if (g_str_equal(value, "allow")) { | |
208 | /* default value */ | |
209 | } else { | |
210 | error_report("invalid argument for elevateprivileges"); | |
211 | return -1; | |
212 | } | |
213 | } | |
214 | ||
215 | value = qemu_opt_get(opts, "spawn"); | |
216 | if (value) { | |
217 | if (g_str_equal(value, "deny")) { | |
218 | seccomp_opts |= QEMU_SECCOMP_SET_SPAWN; | |
219 | } else if (g_str_equal(value, "allow")) { | |
220 | /* default value */ | |
221 | } else { | |
222 | error_report("invalid argument for spawn"); | |
223 | return -1; | |
224 | } | |
225 | } | |
226 | ||
227 | value = qemu_opt_get(opts, "resourcecontrol"); | |
228 | if (value) { | |
229 | if (g_str_equal(value, "deny")) { | |
230 | seccomp_opts |= QEMU_SECCOMP_SET_RESOURCECTL; | |
231 | } else if (g_str_equal(value, "allow")) { | |
232 | /* default value */ | |
233 | } else { | |
234 | error_report("invalid argument for resourcecontrol"); | |
235 | return -1; | |
236 | } | |
237 | } | |
238 | ||
239 | if (seccomp_start(seccomp_opts) < 0) { | |
240 | error_report("failed to install seccomp syscall filter " | |
241 | "in the kernel"); | |
242 | return -1; | |
243 | } | |
244 | } | |
245 | ||
246 | return 0; | |
247 | } | |
248 | ||
249 | static QemuOptsList qemu_sandbox_opts = { | |
250 | .name = "sandbox", | |
251 | .implied_opt_name = "enable", | |
252 | .head = QTAILQ_HEAD_INITIALIZER(qemu_sandbox_opts.head), | |
253 | .desc = { | |
254 | { | |
255 | .name = "enable", | |
256 | .type = QEMU_OPT_BOOL, | |
257 | }, | |
258 | { | |
259 | .name = "obsolete", | |
260 | .type = QEMU_OPT_STRING, | |
261 | }, | |
262 | { | |
263 | .name = "elevateprivileges", | |
264 | .type = QEMU_OPT_STRING, | |
265 | }, | |
266 | { | |
267 | .name = "spawn", | |
268 | .type = QEMU_OPT_STRING, | |
269 | }, | |
270 | { | |
271 | .name = "resourcecontrol", | |
272 | .type = QEMU_OPT_STRING, | |
273 | }, | |
274 | { /* end of list */ } | |
275 | }, | |
276 | }; | |
277 | ||
278 | static void seccomp_register(void) | |
279 | { | |
280 | qemu_add_opts(&qemu_sandbox_opts); | |
281 | } | |
282 | opts_init(seccomp_register); | |
283 | #endif |