[PATCH 19/25] virtiofsd: add seccomp whitelist

[Dr. David Alan Gilbert (git)]

From: Stefan Hajnoczi <address@hidden>

Only allow system calls that are needed by virtiofsd.  All other system
calls cause SIGSYS to be directed at the thread.

Restricting system calls reduces the kernel attack surface and limits
what the process can do when compromised.

Signed-off-by: Stefan Hajnoczi <address@hidden>
with additional entries by:
Signed-off-by: Ganesh Maharaj Mahalingam <address@hidden>
Signed-off-by: Masayoshi Mizuma <address@hidden>
Signed-off-by: Misono Tomohiro <address@hidden>
Signed-off-by: piaojun <address@hidden>
Signed-off-by: Vivek Goyal <address@hidden>
Signed-off-by: Eric Ren <address@hidden>
---
 contrib/virtiofsd/Makefile.objs    |   5 +-
 contrib/virtiofsd/passthrough_ll.c |   2 +
 contrib/virtiofsd/seccomp.c        | 132 +++++++++++++++++++++++++++++
 contrib/virtiofsd/seccomp.h        |  14 +++
 4 files changed, 152 insertions(+), 1 deletion(-)
 create mode 100644 contrib/virtiofsd/seccomp.c
 create mode 100644 contrib/virtiofsd/seccomp.h

diff --git a/contrib/virtiofsd/Makefile.objs b/contrib/virtiofsd/Makefile.objs
index 67be16332c..941b19f18e 100644
--- a/contrib/virtiofsd/Makefile.objs
+++ b/contrib/virtiofsd/Makefile.objs
@@ -6,5 +6,8 @@ virtiofsd-obj-y = buffer.o \
                   fuse_signals.o \
                   fuse_virtio.o \
                   helper.o \
-                  passthrough_ll.o
+                  passthrough_ll.o \
+                  seccomp.o
 
+seccomp.o-cflags := $(SECCOMP_CFLAGS)
+seccomp.o-libs := $(SECCOMP_LIBS)
diff --git a/contrib/virtiofsd/passthrough_ll.c 
b/contrib/virtiofsd/passthrough_ll.c
index c027db64e6..93873bf6f4 100644
--- a/contrib/virtiofsd/passthrough_ll.c
+++ b/contrib/virtiofsd/passthrough_ll.c
@@ -56,6 +56,7 @@
 #include <sys/wait.h>
 
 #include "passthrough_helpers.h"
+#include "seccomp.h"
 
 #define HAVE_POSIX_FALLOCATE 1
 
@@ -1979,6 +1980,7 @@ static void setup_sandbox(struct lo_data *lo)
        setup_proc_self_fd(lo);
        setup_net_namespace();
        setup_mount_namespace(lo->source);
+       setup_seccomp();
 }
 
 int main(int argc, char *argv[])
diff --git a/contrib/virtiofsd/seccomp.c b/contrib/virtiofsd/seccomp.c
new file mode 100644
index 0000000000..df1390d6be
--- /dev/null
+++ b/contrib/virtiofsd/seccomp.c
@@ -0,0 +1,132 @@
+/*
+ * Seccomp sandboxing for virtiofsd
+ *
+ * Copyright (C) 2019 Red Hat, Inc.
+ *
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ */
+
+#include "seccomp.h"
+#include "fuse_i.h"
+#include "fuse_log.h"
+#include <errno.h>
+#include <glib.h>
+#include <seccomp.h>
+#include <stdlib.h>
+
+static const int syscall_whitelist[] = {
+    /* TODO ireg sem*() syscalls */
+    SCMP_SYS(brk),
+    SCMP_SYS(capget), /* For CAP_FSETID */
+    SCMP_SYS(capset),
+    SCMP_SYS(clock_gettime),
+    SCMP_SYS(clone),
+    SCMP_SYS(close),
+    SCMP_SYS(copy_file_range),
+    SCMP_SYS(dup),
+    SCMP_SYS(eventfd2),
+    SCMP_SYS(exit),
+    SCMP_SYS(exit_group),
+    SCMP_SYS(fallocate),
+    SCMP_SYS(fchmodat),
+    SCMP_SYS(fchownat),
+    SCMP_SYS(fcntl),
+    SCMP_SYS(fdatasync),
+    SCMP_SYS(fgetxattr),
+    SCMP_SYS(flistxattr),
+    SCMP_SYS(flock),
+    SCMP_SYS(fremovexattr),
+    SCMP_SYS(fsetxattr),
+    SCMP_SYS(fstat),
+    SCMP_SYS(fstatfs),
+    SCMP_SYS(fsync),
+    SCMP_SYS(ftruncate),
+    SCMP_SYS(futex),
+    SCMP_SYS(getdents),
+    SCMP_SYS(getdents64),
+    SCMP_SYS(getegid),
+    SCMP_SYS(geteuid),
+    SCMP_SYS(getpid),
+    SCMP_SYS(gettid),
+    SCMP_SYS(gettimeofday),
+    SCMP_SYS(linkat),
+    SCMP_SYS(lseek),
+    SCMP_SYS(madvise),
+    SCMP_SYS(mkdirat),
+    SCMP_SYS(mknodat),
+    SCMP_SYS(mmap),
+    SCMP_SYS(mprotect),
+    SCMP_SYS(mremap),
+    SCMP_SYS(munmap),
+    SCMP_SYS(newfstatat),
+    SCMP_SYS(open),
+    SCMP_SYS(openat),
+    SCMP_SYS(ppoll),
+    SCMP_SYS(prctl), /* TODO restrict to just PR_SET_NAME? */
+    SCMP_SYS(preadv),
+    SCMP_SYS(pread64),
+    SCMP_SYS(pwritev),
+    SCMP_SYS(pwrite64),
+    SCMP_SYS(read),
+    SCMP_SYS(readlinkat),
+    SCMP_SYS(recvmsg),
+    SCMP_SYS(renameat),
+    SCMP_SYS(renameat2),
+    SCMP_SYS(rt_sigaction),
+    SCMP_SYS(rt_sigprocmask),
+    SCMP_SYS(rt_sigreturn),
+    SCMP_SYS(sendmsg),
+    SCMP_SYS(setresgid),
+    SCMP_SYS(setresuid),
+    SCMP_SYS(set_robust_list),
+    SCMP_SYS(symlinkat),
+    SCMP_SYS(time), /* Rarely needed, except on static builds */
+    SCMP_SYS(tgkill),
+    SCMP_SYS(unlinkat),
+    SCMP_SYS(utimensat),
+    SCMP_SYS(write),
+    SCMP_SYS(writev),
+};
+
+void setup_seccomp(void)
+{
+    scmp_filter_ctx ctx;
+    size_t i;
+
+#ifdef SCMP_ACT_KILL_PROCESS
+    ctx = seccomp_init(SCMP_ACT_KILL_PROCESS);
+    /* Handle a newer libseccomp but an older kernel */
+    if (!ctx && errno == EOPNOTSUPP) {
+        ctx = seccomp_init(SCMP_ACT_KILL);
+    }
+#else
+    ctx = seccomp_init(SCMP_ACT_KILL);
+#endif
+    if (!ctx) {
+        fuse_log(FUSE_LOG_ERR, "seccomp_init() failed\n");
+        exit(1);
+    }
+
+    for (i = 0; i < G_N_ELEMENTS(syscall_whitelist); i++) {
+        if (seccomp_rule_add(ctx, SCMP_ACT_ALLOW,
+                             syscall_whitelist[i], 0) != 0) {
+            fuse_log(FUSE_LOG_ERR, "seccomp_rule_add syscall %d",
+                     syscall_whitelist[i]);
+            exit(1);
+        }
+    }
+
+    /* libvhost-user calls this for post-copy migration, we don't need it */
+    if (seccomp_rule_add(ctx, SCMP_ACT_ERRNO(ENOSYS),
+                         SCMP_SYS(userfaultfd), 0) != 0) {
+        fuse_log(FUSE_LOG_ERR, "seccomp_rule_add userfaultfd failed\n");
+        exit(1);
+    }
+
+    if (seccomp_load(ctx) < 0) {
+        fuse_log(FUSE_LOG_ERR, "seccomp_load() failed\n");
+        exit(1);
+    }
+
+    seccomp_release(ctx);
+}
diff --git a/contrib/virtiofsd/seccomp.h b/contrib/virtiofsd/seccomp.h
new file mode 100644
index 0000000000..86bce72652
--- /dev/null
+++ b/contrib/virtiofsd/seccomp.h
@@ -0,0 +1,14 @@
+/*
+ * Seccomp sandboxing for virtiofsd
+ *
+ * Copyright (C) 2019 Red Hat, Inc.
+ *
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ */
+
+#ifndef VIRTIOFSD_SECCOMP_H
+#define VIRTIOFSD_SECCOMP_H
+
+void setup_seccomp(void);
+
+#endif /* VIRTIOFSD_SECCOMP_H */
-- 
2.23.0