qemu-seccomp.c

   1 /*
   2  * QEMU seccomp mode 2 support with libseccomp
   3  *
   4  * Copyright IBM, Corp. 2012
   5  *
   6  * Authors:
   7  *  Eduardo Otubo    <eotubo@br.ibm.com>
   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  */
  15
  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>
  25
  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.
  29  */
  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
  35
  36 struct QemuSeccompSyscall {
  37     int32_t num;
  38     uint8_t set;
  39     uint8_t narg;
  40     const struct scmp_arg_cmp *arg_cmp;
  41 };
  42
  43 const struct scmp_arg_cmp sched_setscheduler_arg[] = {
  44     SCMP_A1(SCMP_CMP_NE, SCHED_IDLE)
  45 };
  46
  47 static const struct QemuSeccompSyscall blacklist[] = {
  48     /* default set of syscalls to blacklist */
  49     { SCMP_SYS(reboot),                 QEMU_SECCOMP_SET_DEFAULT },
  50     { SCMP_SYS(swapon),                 QEMU_SECCOMP_SET_DEFAULT },
  51     { SCMP_SYS(swapoff),                QEMU_SECCOMP_SET_DEFAULT },
  52     { SCMP_SYS(syslog),                 QEMU_SECCOMP_SET_DEFAULT },
  53     { SCMP_SYS(mount),                  QEMU_SECCOMP_SET_DEFAULT },
  54     { SCMP_SYS(umount),                 QEMU_SECCOMP_SET_DEFAULT },
  55     { SCMP_SYS(kexec_load),             QEMU_SECCOMP_SET_DEFAULT },
  56     { SCMP_SYS(afs_syscall),            QEMU_SECCOMP_SET_DEFAULT },
  57     { SCMP_SYS(break),                  QEMU_SECCOMP_SET_DEFAULT },
  58     { SCMP_SYS(ftime),                  QEMU_SECCOMP_SET_DEFAULT },
  59     { SCMP_SYS(getpmsg),                QEMU_SECCOMP_SET_DEFAULT },
  60     { SCMP_SYS(gtty),                   QEMU_SECCOMP_SET_DEFAULT },
  61     { SCMP_SYS(lock),                   QEMU_SECCOMP_SET_DEFAULT },
  62     { SCMP_SYS(mpx),                    QEMU_SECCOMP_SET_DEFAULT },
  63     { SCMP_SYS(prof),                   QEMU_SECCOMP_SET_DEFAULT },
  64     { SCMP_SYS(profil),                 QEMU_SECCOMP_SET_DEFAULT },
  65     { SCMP_SYS(putpmsg),                QEMU_SECCOMP_SET_DEFAULT },
  66     { SCMP_SYS(security),               QEMU_SECCOMP_SET_DEFAULT },
  67     { SCMP_SYS(stty),                   QEMU_SECCOMP_SET_DEFAULT },
  68     { SCMP_SYS(tuxcall),                QEMU_SECCOMP_SET_DEFAULT },
  69     { SCMP_SYS(ulimit),                 QEMU_SECCOMP_SET_DEFAULT },
  70     { SCMP_SYS(vserver),                QEMU_SECCOMP_SET_DEFAULT },
  71     /* obsolete */
  72     { SCMP_SYS(readdir),                QEMU_SECCOMP_SET_OBSOLETE },
  73     { SCMP_SYS(_sysctl),                QEMU_SECCOMP_SET_OBSOLETE },
  74     { SCMP_SYS(bdflush),                QEMU_SECCOMP_SET_OBSOLETE },
  75     { SCMP_SYS(create_module),          QEMU_SECCOMP_SET_OBSOLETE },
  76     { SCMP_SYS(get_kernel_syms),        QEMU_SECCOMP_SET_OBSOLETE },
  77     { SCMP_SYS(query_module),           QEMU_SECCOMP_SET_OBSOLETE },
  78     { SCMP_SYS(sgetmask),               QEMU_SECCOMP_SET_OBSOLETE },
  79     { SCMP_SYS(ssetmask),               QEMU_SECCOMP_SET_OBSOLETE },
  80     { SCMP_SYS(sysfs),                  QEMU_SECCOMP_SET_OBSOLETE },
  81     { SCMP_SYS(uselib),                 QEMU_SECCOMP_SET_OBSOLETE },
  82     { SCMP_SYS(ustat),                  QEMU_SECCOMP_SET_OBSOLETE },
  83     /* privileged */
  84     { SCMP_SYS(setuid),                 QEMU_SECCOMP_SET_PRIVILEGED },
  85     { SCMP_SYS(setgid),                 QEMU_SECCOMP_SET_PRIVILEGED },
  86     { SCMP_SYS(setpgid),                QEMU_SECCOMP_SET_PRIVILEGED },
  87     { SCMP_SYS(setsid),                 QEMU_SECCOMP_SET_PRIVILEGED },
  88     { SCMP_SYS(setreuid),               QEMU_SECCOMP_SET_PRIVILEGED },
  89     { SCMP_SYS(setregid),               QEMU_SECCOMP_SET_PRIVILEGED },
  90     { SCMP_SYS(setresuid),              QEMU_SECCOMP_SET_PRIVILEGED },
  91     { SCMP_SYS(setresgid),              QEMU_SECCOMP_SET_PRIVILEGED },
  92     { SCMP_SYS(setfsuid),               QEMU_SECCOMP_SET_PRIVILEGED },
  93     { SCMP_SYS(setfsgid),               QEMU_SECCOMP_SET_PRIVILEGED },
  94     /* spawn */
  95     { SCMP_SYS(fork),                   QEMU_SECCOMP_SET_SPAWN },
  96     { SCMP_SYS(vfork),                  QEMU_SECCOMP_SET_SPAWN },
  97     { SCMP_SYS(execve),                 QEMU_SECCOMP_SET_SPAWN },
  98     /* resource control */
  99     { SCMP_SYS(getpriority),            QEMU_SECCOMP_SET_RESOURCECTL },
 100     { SCMP_SYS(setpriority),            QEMU_SECCOMP_SET_RESOURCECTL },
 101     { SCMP_SYS(sched_setparam),         QEMU_SECCOMP_SET_RESOURCECTL },
 102     { SCMP_SYS(sched_getparam),         QEMU_SECCOMP_SET_RESOURCECTL },
 103     { SCMP_SYS(sched_setscheduler),     QEMU_SECCOMP_SET_RESOURCECTL,
 104       ARRAY_SIZE(sched_setscheduler_arg), sched_setscheduler_arg },
 105     { SCMP_SYS(sched_getscheduler),     QEMU_SECCOMP_SET_RESOURCECTL },
 106     { SCMP_SYS(sched_setaffinity),      QEMU_SECCOMP_SET_RESOURCECTL },
 107     { SCMP_SYS(sched_getaffinity),      QEMU_SECCOMP_SET_RESOURCECTL },
 108     { SCMP_SYS(sched_get_priority_max), QEMU_SECCOMP_SET_RESOURCECTL },
 109     { SCMP_SYS(sched_get_priority_min), QEMU_SECCOMP_SET_RESOURCECTL },
 110 };
 111
 112 static inline __attribute__((unused)) int
 113 qemu_seccomp(unsigned int operation, unsigned int flags, void *args)
 114 {
 115 #ifdef __NR_seccomp
 116     return syscall(__NR_seccomp, operation, flags, args);
 117 #else
 118     errno = ENOSYS;
 119     return -1;
 120 #endif
 121 }
 122
 123 static uint32_t qemu_seccomp_get_kill_action(void)
 124 {
 125 #if defined(SECCOMP_GET_ACTION_AVAIL) && defined(SCMP_ACT_KILL_PROCESS) && \
 126     defined(SECCOMP_RET_KILL_PROCESS)
 127     {
 128         uint32_t action = SECCOMP_RET_KILL_PROCESS;
 129
 130         if (qemu_seccomp(SECCOMP_GET_ACTION_AVAIL, 0, &action) == 0) {
 131             return SCMP_ACT_KILL_PROCESS;
 132         }
 133     }
 134 #endif
 135
 136     return SCMP_ACT_TRAP;
 137 }
 138
 139
 140 static int seccomp_start(uint32_t seccomp_opts)
 141 {
 142     int rc = 0;
 143     unsigned int i = 0;
 144     scmp_filter_ctx ctx;
 145     uint32_t action = qemu_seccomp_get_kill_action();
 146
 147     ctx = seccomp_init(SCMP_ACT_ALLOW);
 148     if (ctx == NULL) {
 149         rc = -1;
 150         goto seccomp_return;
 151     }
 152
 153     rc = seccomp_attr_set(ctx, SCMP_FLTATR_CTL_TSYNC, 1);
 154     if (rc != 0) {
 155         goto seccomp_return;
 156     }
 157
 158     for (i = 0; i < ARRAY_SIZE(blacklist); i++) {
 159         if (!(seccomp_opts & blacklist[i].set)) {
 160             continue;
 161         }
 162
 163         rc = seccomp_rule_add_array(ctx, action, blacklist[i].num,
 164                                     blacklist[i].narg, blacklist[i].arg_cmp);
 165         if (rc < 0) {
 166             goto seccomp_return;
 167         }
 168     }
 169
 170     rc = seccomp_load(ctx);
 171
 172   seccomp_return:
 173     seccomp_release(ctx);
 174     return rc;
 175 }
 176
 177 #ifdef CONFIG_SECCOMP
 178 int parse_sandbox(void *opaque, QemuOpts *opts, Error **errp)
 179 {
 180     if (qemu_opt_get_bool(opts, "enable", false)) {
 181         uint32_t seccomp_opts = QEMU_SECCOMP_SET_DEFAULT
 182                 | QEMU_SECCOMP_SET_OBSOLETE;
 183         const char *value = NULL;
 184
 185         value = qemu_opt_get(opts, "obsolete");
 186         if (value) {
 187             if (g_str_equal(value, "allow")) {
 188                 seccomp_opts &= ~QEMU_SECCOMP_SET_OBSOLETE;
 189             } else if (g_str_equal(value, "deny")) {
 190                 /* this is the default option, this if is here
 191                  * to provide a little bit of consistency for
 192                  * the command line */
 193             } else {
 194                 error_setg(errp, "invalid argument for obsolete");
 195                 return -1;
 196             }
 197         }
 198
 199         value = qemu_opt_get(opts, "elevateprivileges");
 200         if (value) {
 201             if (g_str_equal(value, "deny")) {
 202                 seccomp_opts |= QEMU_SECCOMP_SET_PRIVILEGED;
 203             } else if (g_str_equal(value, "children")) {
 204                 seccomp_opts |= QEMU_SECCOMP_SET_PRIVILEGED;
 205
 206                 /* calling prctl directly because we're
 207                  * not sure if host has CAP_SYS_ADMIN set*/
 208                 if (prctl(PR_SET_NO_NEW_PRIVS, 1)) {
 209                     error_setg(errp, "failed to set no_new_privs aborting");
 210                     return -1;
 211                 }
 212             } else if (g_str_equal(value, "allow")) {
 213                 /* default value */
 214             } else {
 215                 error_setg(errp, "invalid argument for elevateprivileges");
 216                 return -1;
 217             }
 218         }
 219
 220         value = qemu_opt_get(opts, "spawn");
 221         if (value) {
 222             if (g_str_equal(value, "deny")) {
 223                 seccomp_opts |= QEMU_SECCOMP_SET_SPAWN;
 224             } else if (g_str_equal(value, "allow")) {
 225                 /* default value */
 226             } else {
 227                 error_setg(errp, "invalid argument for spawn");
 228                 return -1;
 229             }
 230         }
 231
 232         value = qemu_opt_get(opts, "resourcecontrol");
 233         if (value) {
 234             if (g_str_equal(value, "deny")) {
 235                 seccomp_opts |= QEMU_SECCOMP_SET_RESOURCECTL;
 236             } else if (g_str_equal(value, "allow")) {
 237                 /* default value */
 238             } else {
 239                 error_setg(errp, "invalid argument for resourcecontrol");
 240                 return -1;
 241             }
 242         }
 243
 244         if (seccomp_start(seccomp_opts) < 0) {
 245             error_setg(errp, "failed to install seccomp syscall filter "
 246                        "in the kernel");
 247             return -1;
 248         }
 249     }
 250
 251     return 0;
 252 }
 253
 254 static QemuOptsList qemu_sandbox_opts = {
 255     .name = "sandbox",
 256     .implied_opt_name = "enable",
 257     .head = QTAILQ_HEAD_INITIALIZER(qemu_sandbox_opts.head),
 258     .desc = {
 259         {
 260             .name = "enable",
 261             .type = QEMU_OPT_BOOL,
 262         },
 263         {
 264             .name = "obsolete",
 265             .type = QEMU_OPT_STRING,
 266         },
 267         {
 268             .name = "elevateprivileges",
 269             .type = QEMU_OPT_STRING,
 270         },
 271         {
 272             .name = "spawn",
 273             .type = QEMU_OPT_STRING,
 274         },
 275         {
 276             .name = "resourcecontrol",
 277             .type = QEMU_OPT_STRING,
 278         },
 279         { /* end of list */ }
 280     },
 281 };
 282
 283 static void seccomp_register(void)
 284 {
 285     bool add = false;
 286
 287     /* FIXME: use seccomp_api_get() >= 2 check when released */
 288
 289 #if defined(SECCOMP_FILTER_FLAG_TSYNC)
 290     int check;
 291
 292     /* check host TSYNC capability, it returns errno == ENOSYS if unavailable */
 293     check = qemu_seccomp(SECCOMP_SET_MODE_FILTER,
 294                          SECCOMP_FILTER_FLAG_TSYNC, NULL);
 295     if (check < 0 && errno == EFAULT) {
 296         add = true;
 297     }
 298 #endif
 299
 300     if (add) {
 301         qemu_add_opts(&qemu_sandbox_opts);
 302     }
 303 }
 304 opts_init(seccomp_register);
 305 #endif