[PATCH 2/2] virtiofsd: move to a new pid namespace

[Stefan Hajnoczi]

virtiofsd needs access to /proc/self/fd.  Let's move to a new pid
namespace so that a compromised process cannot see another other
processes running on the system.

One wrinkle in this approach: unshare(CLONE_NEWPID) affects *child*
processes and not the current process.  Therefore we need to fork the
pid 1 process that will actually run virtiofsd and leave a parent in
waitpid(2).  This is not the same thing as daemonization and parent
processes should not notice a difference.

Signed-off-by: Stefan Hajnoczi <address@hidden>
---
 contrib/virtiofsd/passthrough_ll.c | 95 ++++++++++++++++++++++--------
 1 file changed, 72 insertions(+), 23 deletions(-)

diff --git a/contrib/virtiofsd/passthrough_ll.c 
b/contrib/virtiofsd/passthrough_ll.c
index c27ff7d800..b6ee9b2e90 100644
--- a/contrib/virtiofsd/passthrough_ll.c
+++ b/contrib/virtiofsd/passthrough_ll.c
@@ -56,9 +56,12 @@
 #include <sys/xattr.h>
 #include <sys/mman.h>
 #include <sys/socket.h>
+#include <sys/types.h>
 #include <sys/un.h>
+#include <sys/wait.h>
 #include <sys/capability.h>
 
+
 #include "ireg.h"
 #include <sys/mount.h>
 #include <sys/resource.h>
@@ -2749,6 +2752,72 @@ static void setup_net_namespace(void)
        }
 }
 
+/*
+ * Move to a new pid namespace to prevent access to other processes if this
+ * process is compromised.
+ */
+static void setup_pid_namespace(void)
+{
+       pid_t child;
+
+       /*
+        * Create a new pid namespace for *child* processes.  We'll have to
+        * fork in order to enter the new pid namespace.  A new mount namespace
+        * is also needed so that we can remount /proc for the new pid
+        * namespace.
+        */
+       if (unshare(CLONE_NEWPID | CLONE_NEWNS) != 0) {
+               fuse_log(FUSE_LOG_ERR, "unshare(CLONE_NEWPID | CLONE_NEWNS): 
%m\n");
+               exit(1);
+       }
+
+       child = fork();
+       if (child < 0) {
+               fuse_log(FUSE_LOG_ERR, "fork() failed: %m\n");
+               exit(1);
+       }
+       if (child > 0) {
+               pid_t waited;
+               int wstatus;
+
+               /* The parent waits for the child */
+               do {
+                       waited = waitpid(child, &wstatus, 0);
+               } while (waited < 0 && errno == EINTR);
+
+               if (WIFEXITED(wstatus)) {
+                       exit(WEXITSTATUS(wstatus));
+               }
+
+               exit(1);
+       }
+
+       /*
+        * If the mounts have shared propagation then we want to opt out so our
+        * mount changes don't affect the parent mount namespace.
+        */
+       if (mount(NULL, "/", NULL, MS_REC|MS_SLAVE, NULL) < 0) {
+               fuse_log(FUSE_LOG_ERR, "mount(/, MS_REC|MS_SLAVE): %m\n");
+               exit(1);
+       }
+
+       /* The child must remount /proc to use the new pid namespace */
+       if (mount("proc", "/proc", "proc",
+                 MS_NODEV | MS_NOEXEC | MS_NOSUID | MS_RELATIME, NULL) < 0) {
+               fuse_log(FUSE_LOG_ERR, "mount(/proc): %m\n");
+               exit(1);
+       }
+}
+
+static void setup_proc_self_fd(struct lo_data *lo)
+{
+       lo->proc_self_fd = open("/proc/self/fd", O_PATH);
+       if (lo->proc_self_fd == -1) {
+               fuse_log(FUSE_LOG_ERR, "open(/proc/self/fd, O_PATH): %m\n");
+               exit(1);
+       }
+}
+
 /* This magic is based on lxc's lxc_pivot_root() */
 static void setup_pivot_root(const char *source)
 {
@@ -2803,20 +2872,10 @@ static void setup_pivot_root(const char *source)
 
 /*
  * Make the source directory our root so symlinks cannot escape and no other
- * files are accessible.
+ * files are accessible.  Assumes unshare(CLONE_NEWNS) was already called.
  */
 static void setup_mount_namespace(const char *source)
 {
-       if (unshare(CLONE_NEWNS) != 0) {
-               fuse_log(FUSE_LOG_ERR, "unshare(CLONE_NEWNS): %m\n");
-               exit(1);
-       }
-
-       if (mount(NULL, "/", NULL, MS_REC|MS_SLAVE, NULL) < 0) {
-               fuse_log(FUSE_LOG_ERR, "mount(/, MS_REC|MS_PRIVATE): %m\n");
-               exit(1);
-       }
-
        if (mount(source, source, NULL, MS_BIND, NULL) < 0) {
                fuse_log(FUSE_LOG_ERR, "mount(%s, %s, MS_BIND): %m\n", source, 
source);
                exit(1);
@@ -2831,6 +2890,8 @@ static void setup_mount_namespace(const char *source)
  */
 static void setup_sandbox(struct lo_data *lo, bool enable_syslog)
 {
+       setup_pid_namespace();
+       setup_proc_self_fd(lo);
        setup_net_namespace();
        setup_mount_namespace(lo->source);
        setup_seccomp(enable_syslog);
@@ -2860,15 +2921,6 @@ static void setup_root(struct lo_data *lo, struct 
lo_inode *root)
        g_atomic_int_set(&root->refcount, 2);
 }
 
-static void setup_proc_self_fd(struct lo_data *lo)
-{
-       lo->proc_self_fd = open("/proc/self/fd", O_PATH);
-       if (lo->proc_self_fd == -1) {
-               fuse_log(FUSE_LOG_ERR, "open(/proc/self/fd, O_PATH): %m\n");
-               exit(1);
-       }
-}
-
 /* Raise the maximum number of open file descriptors to the system limit */
 static void setup_nofile_rlimit(void)
 {
@@ -3110,9 +3162,6 @@ int main(int argc, char *argv[])
                get_shared(&lo, &lo.root);
        }
 
-       /* Must be after daemonize to get the right /proc/self/fd */
-       setup_proc_self_fd(&lo);
-
        setup_sandbox(&lo, opts.syslog);
 
        setup_root(&lo, &lo.root);
-- 
2.21.0