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
  24 /* For some architectures (notably ARM) cacheflush is not supported until
  25  * libseccomp 2.2.3, but configure enforces that we are using a more recent
  26  * version on those hosts, so it is OK for this check to be less strict.
  27  */
  28 #if SCMP_VER_MAJOR >= 3
  29   #define HAVE_CACHEFLUSH
  30 #elif SCMP_VER_MAJOR == 2 && SCMP_VER_MINOR >= 2
  31   #define HAVE_CACHEFLUSH
  32 #endif
  33
  34 struct QemuSeccompSyscall {
  35     int32_t num;
  36     uint8_t set;
  37     uint8_t narg;
  38     const struct scmp_arg_cmp *arg_cmp;
  39 };
  40
  41 const struct scmp_arg_cmp sched_setscheduler_arg[] = {
  42     SCMP_A1(SCMP_CMP_NE, SCHED_IDLE)
  43 };
  44
  45 static const struct QemuSeccompSyscall blacklist[] = {
  46     /* default set of syscalls to blacklist */
  47     { SCMP_SYS(reboot),                 QEMU_SECCOMP_SET_DEFAULT },
  48     { SCMP_SYS(swapon),                 QEMU_SECCOMP_SET_DEFAULT },
  49     { SCMP_SYS(swapoff),                QEMU_SECCOMP_SET_DEFAULT },
  50     { SCMP_SYS(syslog),                 QEMU_SECCOMP_SET_DEFAULT },
  51     { SCMP_SYS(mount),                  QEMU_SECCOMP_SET_DEFAULT },
  52     { SCMP_SYS(umount),                 QEMU_SECCOMP_SET_DEFAULT },
  53     { SCMP_SYS(kexec_load),             QEMU_SECCOMP_SET_DEFAULT },
  54     { SCMP_SYS(afs_syscall),            QEMU_SECCOMP_SET_DEFAULT },
  55     { SCMP_SYS(break),                  QEMU_SECCOMP_SET_DEFAULT },
  56     { SCMP_SYS(ftime),                  QEMU_SECCOMP_SET_DEFAULT },
  57     { SCMP_SYS(getpmsg),                QEMU_SECCOMP_SET_DEFAULT },
  58     { SCMP_SYS(gtty),                   QEMU_SECCOMP_SET_DEFAULT },
  59     { SCMP_SYS(lock),                   QEMU_SECCOMP_SET_DEFAULT },
  60     { SCMP_SYS(mpx),                    QEMU_SECCOMP_SET_DEFAULT },
  61     { SCMP_SYS(prof),                   QEMU_SECCOMP_SET_DEFAULT },
  62     { SCMP_SYS(profil),                 QEMU_SECCOMP_SET_DEFAULT },
  63     { SCMP_SYS(putpmsg),                QEMU_SECCOMP_SET_DEFAULT },
  64     { SCMP_SYS(security),               QEMU_SECCOMP_SET_DEFAULT },
  65     { SCMP_SYS(stty),                   QEMU_SECCOMP_SET_DEFAULT },
  66     { SCMP_SYS(tuxcall),                QEMU_SECCOMP_SET_DEFAULT },
  67     { SCMP_SYS(ulimit),                 QEMU_SECCOMP_SET_DEFAULT },
  68     { SCMP_SYS(vserver),                QEMU_SECCOMP_SET_DEFAULT },
  69     /* obsolete */
  70     { SCMP_SYS(readdir),                QEMU_SECCOMP_SET_OBSOLETE },
  71     { SCMP_SYS(_sysctl),                QEMU_SECCOMP_SET_OBSOLETE },
  72     { SCMP_SYS(bdflush),                QEMU_SECCOMP_SET_OBSOLETE },
  73     { SCMP_SYS(create_module),          QEMU_SECCOMP_SET_OBSOLETE },
  74     { SCMP_SYS(get_kernel_syms),        QEMU_SECCOMP_SET_OBSOLETE },
  75     { SCMP_SYS(query_module),           QEMU_SECCOMP_SET_OBSOLETE },
  76     { SCMP_SYS(sgetmask),               QEMU_SECCOMP_SET_OBSOLETE },
  77     { SCMP_SYS(ssetmask),               QEMU_SECCOMP_SET_OBSOLETE },
  78     { SCMP_SYS(sysfs),                  QEMU_SECCOMP_SET_OBSOLETE },
  79     { SCMP_SYS(uselib),                 QEMU_SECCOMP_SET_OBSOLETE },
  80     { SCMP_SYS(ustat),                  QEMU_SECCOMP_SET_OBSOLETE },
  81     /* privileged */
  82     { SCMP_SYS(setuid),                 QEMU_SECCOMP_SET_PRIVILEGED },
  83     { SCMP_SYS(setgid),                 QEMU_SECCOMP_SET_PRIVILEGED },
  84     { SCMP_SYS(setpgid),                QEMU_SECCOMP_SET_PRIVILEGED },
  85     { SCMP_SYS(setsid),                 QEMU_SECCOMP_SET_PRIVILEGED },
  86     { SCMP_SYS(setreuid),               QEMU_SECCOMP_SET_PRIVILEGED },
  87     { SCMP_SYS(setregid),               QEMU_SECCOMP_SET_PRIVILEGED },
  88     { SCMP_SYS(setresuid),              QEMU_SECCOMP_SET_PRIVILEGED },
  89     { SCMP_SYS(setresgid),              QEMU_SECCOMP_SET_PRIVILEGED },
  90     { SCMP_SYS(setfsuid),               QEMU_SECCOMP_SET_PRIVILEGED },
  91     { SCMP_SYS(setfsgid),               QEMU_SECCOMP_SET_PRIVILEGED },
  92     /* spawn */
  93     { SCMP_SYS(fork),                   QEMU_SECCOMP_SET_SPAWN },
  94     { SCMP_SYS(vfork),                  QEMU_SECCOMP_SET_SPAWN },
  95     { SCMP_SYS(execve),                 QEMU_SECCOMP_SET_SPAWN },
  96     /* resource control */
  97     { SCMP_SYS(getpriority),            QEMU_SECCOMP_SET_RESOURCECTL },
  98     { SCMP_SYS(setpriority),            QEMU_SECCOMP_SET_RESOURCECTL },
  99     { SCMP_SYS(sched_setparam),         QEMU_SECCOMP_SET_RESOURCECTL },
 100     { SCMP_SYS(sched_getparam),         QEMU_SECCOMP_SET_RESOURCECTL },
 101     { SCMP_SYS(sched_setscheduler),     QEMU_SECCOMP_SET_RESOURCECTL,
 102       ARRAY_SIZE(sched_setscheduler_arg), sched_setscheduler_arg },
 103     { SCMP_SYS(sched_getscheduler),     QEMU_SECCOMP_SET_RESOURCECTL },
 104     { SCMP_SYS(sched_setaffinity),      QEMU_SECCOMP_SET_RESOURCECTL },
 105     { SCMP_SYS(sched_getaffinity),      QEMU_SECCOMP_SET_RESOURCECTL },
 106     { SCMP_SYS(sched_get_priority_max), QEMU_SECCOMP_SET_RESOURCECTL },
 107     { SCMP_SYS(sched_get_priority_min), QEMU_SECCOMP_SET_RESOURCECTL },
 108 };
 109
 110
 111 static int seccomp_start(uint32_t seccomp_opts)
 112 {
 113     int rc = 0;
 114     unsigned int i = 0;
 115     scmp_filter_ctx ctx;
 116
 117     ctx = seccomp_init(SCMP_ACT_ALLOW);
 118     if (ctx == NULL) {
 119         rc = -1;
 120         goto seccomp_return;
 121     }
 122
 123     for (i = 0; i < ARRAY_SIZE(blacklist); i++) {
 124         if (!(seccomp_opts & blacklist[i].set)) {
 125             continue;
 126         }
 127
 128         rc = seccomp_rule_add_array(ctx, SCMP_ACT_KILL, blacklist[i].num,
 129                                     blacklist[i].narg, blacklist[i].arg_cmp);
 130         if (rc < 0) {
 131             goto seccomp_return;
 132         }
 133     }
 134
 135     rc = seccomp_load(ctx);
 136
 137   seccomp_return:
 138     seccomp_release(ctx);
 139     return rc;
 140 }
 141
 142 #ifdef CONFIG_SECCOMP
 143 int parse_sandbox(void *opaque, QemuOpts *opts, Error **errp)
 144 {
 145     if (qemu_opt_get_bool(opts, "enable", false)) {
 146         uint32_t seccomp_opts = QEMU_SECCOMP_SET_DEFAULT
 147                 | QEMU_SECCOMP_SET_OBSOLETE;
 148         const char *value = NULL;
 149
 150         value = qemu_opt_get(opts, "obsolete");
 151         if (value) {
 152             if (g_str_equal(value, "allow")) {
 153                 seccomp_opts &= ~QEMU_SECCOMP_SET_OBSOLETE;
 154             } else if (g_str_equal(value, "deny")) {
 155                 /* this is the default option, this if is here
 156                  * to provide a little bit of consistency for
 157                  * the command line */
 158             } else {
 159                 error_report("invalid argument for obsolete");
 160                 return -1;
 161             }
 162         }
 163
 164         value = qemu_opt_get(opts, "elevateprivileges");
 165         if (value) {
 166             if (g_str_equal(value, "deny")) {
 167                 seccomp_opts |= QEMU_SECCOMP_SET_PRIVILEGED;
 168             } else if (g_str_equal(value, "children")) {
 169                 seccomp_opts |= QEMU_SECCOMP_SET_PRIVILEGED;
 170
 171                 /* calling prctl directly because we're
 172                  * not sure if host has CAP_SYS_ADMIN set*/
 173                 if (prctl(PR_SET_NO_NEW_PRIVS, 1)) {
 174                     error_report("failed to set no_new_privs "
 175                                  "aborting");
 176                     return -1;
 177                 }
 178             } else if (g_str_equal(value, "allow")) {
 179                 /* default value */
 180             } else {
 181                 error_report("invalid argument for elevateprivileges");
 182                 return -1;
 183             }
 184         }
 185
 186         value = qemu_opt_get(opts, "spawn");
 187         if (value) {
 188             if (g_str_equal(value, "deny")) {
 189                 seccomp_opts |= QEMU_SECCOMP_SET_SPAWN;
 190             } else if (g_str_equal(value, "allow")) {
 191                 /* default value */
 192             } else {
 193                 error_report("invalid argument for spawn");
 194                 return -1;
 195             }
 196         }
 197
 198         value = qemu_opt_get(opts, "resourcecontrol");
 199         if (value) {
 200             if (g_str_equal(value, "deny")) {
 201                 seccomp_opts |= QEMU_SECCOMP_SET_RESOURCECTL;
 202             } else if (g_str_equal(value, "allow")) {
 203                 /* default value */
 204             } else {
 205                 error_report("invalid argument for resourcecontrol");
 206                 return -1;
 207             }
 208         }
 209
 210         if (seccomp_start(seccomp_opts) < 0) {
 211             error_report("failed to install seccomp syscall filter "
 212                          "in the kernel");
 213             return -1;
 214         }
 215     }
 216
 217     return 0;
 218 }
 219
 220 static QemuOptsList qemu_sandbox_opts = {
 221     .name = "sandbox",
 222     .implied_opt_name = "enable",
 223     .head = QTAILQ_HEAD_INITIALIZER(qemu_sandbox_opts.head),
 224     .desc = {
 225         {
 226             .name = "enable",
 227             .type = QEMU_OPT_BOOL,
 228         },
 229         {
 230             .name = "obsolete",
 231             .type = QEMU_OPT_STRING,
 232         },
 233         {
 234             .name = "elevateprivileges",
 235             .type = QEMU_OPT_STRING,
 236         },
 237         {
 238             .name = "spawn",
 239             .type = QEMU_OPT_STRING,
 240         },
 241         {
 242             .name = "resourcecontrol",
 243             .type = QEMU_OPT_STRING,
 244         },
 245         { /* end of list */ }
 246     },
 247 };
 248
 249 static void seccomp_register(void)
 250 {
 251     qemu_add_opts(&qemu_sandbox_opts);
 252 }
 253 opts_init(seccomp_register);
 254 #endif