sys_linux: add support for seccomp filters

The Linux secure computing (seccomp) facility allows a process to
install a filter in the kernel that will allow only specific system
calls to be made. The process is killed when trying to make other system
calls. This is useful to reduce the kernel attack surface and possibly
prevent kernel exploits when the process is compromised.

Use the libseccomp library to add rules and load the filter into the
kernel. Keep a list of system calls that are always allowed after
chronyd is initialized. Restrict arguments that may be passed to the
socket(), setsockopt(), fcntl(), and ioctl() system calls. Arguments
to socketcall(), which is used on some architectures as a multiplexer
instead of separate socket system calls, are not restricted for now.
The mailonchange directive is not allowed as it calls sendmail.

Calls made by the libraries that chronyd is using have to be covered
too. It's difficult to determine which system calls they need as it may
change after an upgrade and it may depend on their configuration (e.g.
resolver in libc). There are also differences between architectures. It
can all break very easily and is therefore disabled by default. It can
be enabled with the new -F option.

This is based on a patch from Andrew Griffiths <agriffit@redhat.com>.

chrony.texi.in

@@ -1029,6 +1029,11 @@ switch after start in order to drop root privileges.  It overrides the
 @code{user} directive (default @code{@DEFAULT_USER@}).  It may be set to a
 non-root user only when @code{chronyd} is compiled with support for Linux
 capabilities (libcap).
+@item -F <level>
+This option configures a system call filter when @code{chronyd} is compiled with
+support for the Linux secure computing (seccomp) facility. In level 1 the
+process is killed when a forbidden system call is made, in level -1 the SYSSIG
+signal is thrown instead and in level 0 the filter is disabled (default 0).
 @item -q
 When run in this mode, @code{chronyd} will set the system clock once
 and exit.  It will not detach from the terminal.

chronyd.8.in

@@ -111,6 +111,12 @@ directive (default \fB@DEFAULT_USER@\fR).  It may be set to a non-root user
 only when \fBchronyd\fR is compiled with support for Linux capabilities
 (libcap).
 .TP
+\fB\-F\fR \fIlevel\fR
+This option configures a system call filter when \fBchronyd\fR is compiled with
+support for the Linux secure computing (seccomp) facility. In level 1 the
+process is killed when a forbidden system call is made, in level -1 the SYSSIG
+signal is thrown instead and in level 0 the filter is disabled (default 0).
+.TP
 .B \-q
 When run in this mode, chronyd will set the system clock once
 and exit.  It will not detach from the terminal.

configure

@@ -115,6 +115,7 @@ For better control, use the options below.
   --disable-rtc          Don't include RTC even on Linux
   --disable-privdrop     Disable support for dropping root privileges
   --without-libcap       Don't use libcap even if it is available
+  --without-seccomp      Don't use seccomp even if it is available
   --disable-asyncdns     Disable asynchronous name resolving
   --disable-forcednsretry Don't retry on permanent DNS error
   --with-ntp-era=SECONDS Specify earliest assumed NTP time in seconds
@@ -217,6 +218,8 @@ try_rtc=0
 feat_droproot=1
 try_libcap=-1
 try_clockctl=0
+feat_scfilter=1
+try_seccomp=-1
 readline_lib=""
 readline_inc=""
 ncurses_lib=""
@@ -322,6 +325,9 @@ do
     --disable-asyncdns)
       feat_asyncdns=0
     ;;
+    --without-seccomp)
+      try_seccomp=0
+    ;;
     --disable-forcednsretry)
       feat_forcednsretry=0
     ;;
@@ -387,6 +393,7 @@ case $SYSTEM in
         EXTRA_OBJECTS="sys_generic.o sys_linux.o wrap_adjtimex.o"
         [ $try_libcap != "0" ] && try_libcap=1
         try_rtc=1
+        [ $try_seccomp != "0" ] && try_seccomp=1
         try_setsched=1
         try_lockmem=1
         try_phc=1
@@ -602,6 +609,14 @@ then
   add_def FEAT_PRIVDROP
 fi
 
+if [ $feat_scfilter = "1" ] && [ $try_seccomp = "1" ] && \
+  test_code seccomp 'seccomp.h' '' '-lseccomp' \
+    'seccomp_init(SCMP_ACT_KILL);'
+then
+  add_def FEAT_SCFILTER
+  EXTRA_LIBS="$EXTRA_LIBS -lseccomp"
+fi
+
 if [ $feat_rtc = "1" ] && [ $try_rtc = "1" ] && \
   test_code '<linux/rtc.h>' 'sys/ioctl.h linux/rtc.h' '' '' \
     'ioctl(1, RTC_UIE_ON&RTC_UIE_OFF&RTC_RD_TIME&RTC_SET_TIME, 0&RTC_UF);'
@@ -792,7 +807,7 @@ add_def MAIL_PROGRAM "\"$mail_program\""
 
 common_features="`get_features ASYNCDNS IPV6 SECHASH DEBUG`"
 chronyc_features="`get_features READLINE`"
-chronyd_features="`get_features CMDMON NTP REFCLOCK RTC PRIVDROP`"
+chronyd_features="`get_features CMDMON NTP REFCLOCK RTC PRIVDROP SCFILTER`"
 add_def CHRONYC_FEATURES "\"$chronyc_features $common_features\""
 add_def CHRONYD_FEATURES "\"$chronyd_features $common_features\""
 echo "Features : $chronyd_features $chronyc_features $common_features"

main.c

@@ -354,7 +354,7 @@ int main
   int debug = 0, nofork = 0, address_family = IPADDR_UNSPEC;
   int do_init_rtc = 0, restarted = 0;
   int other_pid;
-  int lock_memory = 0, sched_priority = 0;
+  int scfilter_level = 0, lock_memory = 0, sched_priority = 0;
   int system_log = 1;
   int config_args = 0;
 
@@ -384,6 +384,10 @@ int main
       } else {
         user = *argv;
       }
+    } else if (!strcmp("-F", *argv)) {
+      ++argv, --argc;
+      if (argc == 0 || sscanf(*argv, "%d", &scfilter_level) != 1)
+        LOG_FATAL(LOGF_Main, "Bad syscall filter level");
     } else if (!strcmp("-s", *argv)) {
       do_init_rtc = 1;
     } else if (!strcmp("-v", *argv) || !strcmp("--version",*argv)) {
@@ -521,6 +525,9 @@ int main
 
   CAM_OpenUnixSocket();
 
+  if (scfilter_level)
+    SYS_EnableSystemCallFilter(scfilter_level);
+
   if (ref_mode == REF_ModeNormal && CNF_GetInitSources() > 0) {
     ref_mode = REF_ModeInitStepSlew;
   }

sys.c

@@ -122,6 +122,17 @@ void SYS_DropRoot(uid_t uid, gid_t gid)
 
 /* ================================================== */
 
+void SYS_EnableSystemCallFilter(int level)
+{
+#if defined(LINUX) && defined(FEAT_SCFILTER)
+  SYS_Linux_EnableSystemCallFilter(level);
+#else
+  LOG_FATAL(LOGF_Sys, "system call filter not supported");
+#endif
+}
+
+/* ================================================== */
+
 void SYS_SetScheduler(int SchedPriority)
 {
 #if defined(LINUX) && defined(HAVE_SCHED_SETSCHEDULER)

sys.h

@@ -38,6 +38,10 @@ extern void SYS_Finalise(void);
 /* Drop root privileges to the specified user and group */
 extern void SYS_DropRoot(uid_t uid, gid_t gid);
 
+/* Enable a system call filter to allow only system calls
+   which chronyd normally needs after initialization */
+extern void SYS_EnableSystemCallFilter(int level);
+
 extern void SYS_SetScheduler(int SchedPriority);
 extern void SYS_LockMemory(void);
 

sys_linux.c

@@ -48,6 +48,20 @@
 #include <grp.h>
 #endif
 
+#ifdef FEAT_SCFILTER
+#include <sys/prctl.h>
+#include <seccomp.h>
+#ifdef FEAT_PHC
+#include <linux/ptp_clock.h>
+#endif
+#ifdef FEAT_PPS
+#include <linux/pps.h>
+#endif
+#ifdef FEAT_RTC
+#include <linux/rtc.h>
+#endif
+#endif
+
 #include "sys_generic.h"
 #include "sys_linux.h"
 #include "conf.h"
@@ -412,6 +426,143 @@ SYS_Linux_DropRoot(uid_t uid, gid_t gid)
 
 /* ================================================== */
 
+#ifdef FEAT_SCFILTER
+static
+void check_seccomp_applicability(void)
+{
+  int mail_enabled;
+  double mail_threshold;
+  char *mail_user;
+
+  CNF_GetMailOnChange(&mail_enabled, &mail_threshold, &mail_user);
+  if (mail_enabled)
+    LOG_FATAL(LOGF_SysLinux, "mailonchange directive cannot be used with -F enabled");
+}
+
+/* ================================================== */
+
+void
+SYS_Linux_EnableSystemCallFilter(int level)
+{
+  const int syscalls[] = {
+    /* Clock */
+    SCMP_SYS(adjtimex), SCMP_SYS(gettimeofday), SCMP_SYS(settimeofday),
+    SCMP_SYS(time),
+    /* Process */
+    SCMP_SYS(clone), SCMP_SYS(exit), SCMP_SYS(exit_group),
+    SCMP_SYS(rt_sigreturn), SCMP_SYS(sigreturn),
+    /* Memory */
+    SCMP_SYS(brk), SCMP_SYS(madvise), SCMP_SYS(mmap), SCMP_SYS(mmap2),
+    SCMP_SYS(mprotect), SCMP_SYS(munmap), SCMP_SYS(shmdt),
+    /* Filesystem */
+    SCMP_SYS(chmod), SCMP_SYS(chown), SCMP_SYS(chown32), SCMP_SYS(fstat),
+    SCMP_SYS(fstat64), SCMP_SYS(lseek), SCMP_SYS(rename), SCMP_SYS(stat),
+    SCMP_SYS(stat64), SCMP_SYS(unlink),
+    /* Socket */
+    SCMP_SYS(bind), SCMP_SYS(connect), SCMP_SYS(getsockname),
+    SCMP_SYS(recvfrom), SCMP_SYS(recvmsg), SCMP_SYS(sendmmsg),
+    SCMP_SYS(sendmsg), SCMP_SYS(sendto),
+    /* TODO: check socketcall arguments */
+    SCMP_SYS(socketcall),
+    /* General I/O */
+    SCMP_SYS(_newselect), SCMP_SYS(close), SCMP_SYS(open), SCMP_SYS(pipe),
+    SCMP_SYS(poll), SCMP_SYS(read), SCMP_SYS(futex), SCMP_SYS(select),
+    SCMP_SYS(set_robust_list), SCMP_SYS(write),
+  };
+
+  const int socket_domains[] = {
+    AF_NETLINK, AF_UNIX, AF_INET,
+#ifdef FEAT_IPV6
+    AF_INET6,
+#endif
+  };
+
+  const static int socket_options[][2] = {
+    { SOL_IP, IP_PKTINFO },
+#ifdef FEAT_IPV6
+    { SOL_IPV6, IPV6_V6ONLY }, { SOL_IPV6, IPV6_RECVPKTINFO },
+#endif
+    { SOL_SOCKET, SO_BROADCAST }, { SOL_SOCKET, SO_REUSEADDR },
+    { SOL_SOCKET, SO_TIMESTAMP },
+  };
+
+  const static int fcntls[] = { F_GETFD, F_SETFD };
+
+  const static unsigned long ioctls[] = {
+    FIONREAD,
+#ifdef FEAT_PPS
+    PTP_SYS_OFFSET,
+#endif
+#ifdef FEAT_PPS
+    PPS_FETCH,
+#endif
+#ifdef FEAT_RTC
+    RTC_RD_TIME, RTC_SET_TIME, RTC_UIE_ON, RTC_UIE_OFF,
+#endif
+  };
+
+  scmp_filter_ctx *ctx;
+  int i;
+
+  /* Check if the chronyd configuration is supported */
+  check_seccomp_applicability();
+
+  ctx = seccomp_init(level > 0 ? SCMP_ACT_KILL : SCMP_ACT_TRAP);
+  if (ctx == NULL)
+      LOG_FATAL(LOGF_SysLinux, "Failed to initialize seccomp");
+
+  /* Add system calls that are always allowed */
+  for (i = 0; i < (sizeof (syscalls) / sizeof (*syscalls)); i++) {
+    if (seccomp_rule_add(ctx, SCMP_ACT_ALLOW, syscalls[i], 0) < 0)
+      goto add_failed;
+  }
+
+  /* Allow sockets to be created only in selected domains */
+  for (i = 0; i < sizeof (socket_domains) / sizeof (*socket_domains); i++) {
+    if (seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(socket), 1,
+                         SCMP_A0(SCMP_CMP_EQ, socket_domains[i])) < 0)
+      goto add_failed;
+  }
+
+  /* Allow setting only selected sockets options */
+  for (i = 0; i < sizeof (socket_options) / sizeof (*socket_options); i++) {
+    if (seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(setsockopt), 3,
+                         SCMP_A1(SCMP_CMP_EQ, socket_options[i][0]),
+                         SCMP_A2(SCMP_CMP_EQ, socket_options[i][1]),
+                         SCMP_A4(SCMP_CMP_LE, sizeof (int))) < 0)
+      goto add_failed;
+  }
+
+  /* Allow only selected fcntl calls */
+  for (i = 0; i < sizeof (fcntls) / sizeof (*fcntls); i++) {
+    if (seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(fcntl), 1,
+                         SCMP_A1(SCMP_CMP_EQ, fcntls[i])) < 0 ||
+        seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(fcntl64), 1,
+                         SCMP_A1(SCMP_CMP_EQ, fcntls[i])) < 0)
+      goto add_failed;
+  }
+
+  /* Allow only selected ioctls */
+  for (i = 0; i < sizeof (ioctls) / sizeof (*ioctls); i++) {
+    if (seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(ioctl), 1,
+                         SCMP_A1(SCMP_CMP_EQ, ioctls[i])) < 0)
+      goto add_failed;
+  }
+
+  if (seccomp_load(ctx) < 0)
+      LOG(LOGS_INFO, LOGF_SysLinux, "Failed to load seccomp rules");
+
+  LOG(LOGS_INFO, LOGF_SysLinux, "Loaded seccomp filter");
+  seccomp_release(ctx);
+  return;
+
+add_failed:
+  LOG_FATAL(LOGF_SysLinux, "Failed to add seccomp rules");
+}
+#endif
+
+/* ================================================== */
+
 #if defined(HAVE_SCHED_SETSCHEDULER)
   /* Install SCHED_FIFO real-time scheduler with specified priority */
 void SYS_Linux_SetScheduler(int SchedPriority)

sys_linux.h

@@ -33,6 +33,8 @@ extern void SYS_Linux_Finalise(void);
 
 extern void SYS_Linux_DropRoot(uid_t uid, gid_t gid);
 
+extern void SYS_Linux_EnableSystemCallFilter(int level);
+
 extern void SYS_Linux_MemLockAll(int LockAll);
 
 extern void SYS_Linux_SetScheduler(int SchedPriority);