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 #include "qemu/osdep.h"
  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>
  21 #include <seccomp.h>
  22 #include "sysemu/seccomp.h"
  23 #include <linux/seccomp.h>
  24
  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  */
  29 #if SCMP_VER_MAJOR >= 3
  30   #define HAVE_CACHEFLUSH
  31 #elif SCMP_VER_MAJOR == 2 && SCMP_VER_MINOR >= 2
  32   #define HAVE_CACHEFLUSH
  33 #endif
  34
  35 struct QemuSeccompSyscall {
  36     int32_t num;
  37     uint8_t set;
  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)
  44 };
  45
  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 },
  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 },
  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 },
  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 },
  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 },
 102     { SCMP_SYS(sched_setscheduler),     QEMU_SECCOMP_SET_RESOURCECTL,
 103       ARRAY_SIZE(sched_setscheduler_arg), sched_setscheduler_arg },
 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 },
 109 };
 110
 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
 138
 139 static int seccomp_start(uint32_t seccomp_opts)
 140 {
 141     int rc = 0;
 142     unsigned int i = 0;
 143     scmp_filter_ctx ctx;
 144     uint32_t action = qemu_seccomp_get_kill_action();
 145
 146     ctx = seccomp_init(SCMP_ACT_ALLOW);
 147     if (ctx == NULL) {
 148         rc = -1;
 149         goto seccomp_return;
 150     }
 151
 152     rc = seccomp_attr_set(ctx, SCMP_FLTATR_CTL_TSYNC, 1);
 153     if (rc != 0) {
 154         goto seccomp_return;
 155     }
 156
 157     for (i = 0; i < ARRAY_SIZE(blacklist); i++) {
 158         if (!(seccomp_opts & blacklist[i].set)) {
 159             continue;
 160         }
 161
 162         rc = seccomp_rule_add_array(ctx, action, blacklist[i].num,
 163                                     blacklist[i].narg, blacklist[i].arg_cmp);
 164         if (rc < 0) {
 165             goto seccomp_return;
 166         }
 167     }
 168
 169     rc = seccomp_load(ctx);
 170
 171   seccomp_return:
 172     seccomp_release(ctx);
 173     return rc;
 174 }
 175
 176 #ifdef CONFIG_SECCOMP
 177 int parse_sandbox(void *opaque, QemuOpts *opts, Error **errp)
 178 {
 179     if (qemu_opt_get_bool(opts, "enable", false)) {
 180         uint32_t seccomp_opts = QEMU_SECCOMP_SET_DEFAULT
 181                 | QEMU_SECCOMP_SET_OBSOLETE;
 182         const char *value = NULL;
 183
 184         value = qemu_opt_get(opts, "obsolete");
 185         if (value) {
 186             if (g_str_equal(value, "allow")) {
 187                 seccomp_opts &= ~QEMU_SECCOMP_SET_OBSOLETE;
 188             } else if (g_str_equal(value, "deny")) {
 189                 /* this is the default option, this if is here
 190                  * to provide a little bit of consistency for
 191                  * the command line */
 192             } else {
 193                 error_report("invalid argument for obsolete");
 194                 return -1;
 195             }
 196         }
 197
 198         value = qemu_opt_get(opts, "elevateprivileges");
 199         if (value) {
 200             if (g_str_equal(value, "deny")) {
 201                 seccomp_opts |= QEMU_SECCOMP_SET_PRIVILEGED;
 202             } else if (g_str_equal(value, "children")) {
 203                 seccomp_opts |= QEMU_SECCOMP_SET_PRIVILEGED;
 204
 205                 /* calling prctl directly because we're
 206                  * not sure if host has CAP_SYS_ADMIN set*/
 207                 if (prctl(PR_SET_NO_NEW_PRIVS, 1)) {
 208                     error_report("failed to set no_new_privs "
 209                                  "aborting");
 210                     return -1;
 211                 }
 212             } else if (g_str_equal(value, "allow")) {
 213                 /* default value */
 214             } else {
 215                 error_report("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_report("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_report("invalid argument for resourcecontrol");
 240                 return -1;
 241             }
 242         }
 243
 244         if (seccomp_start(seccomp_opts) < 0) {
 245             error_report("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