int sandbox(void* sandbox_arg) {
// Get rid of unused parameter warning
(void)sandbox_arg;
pid_t child_pid = getpid();
if (arg_debug)
printf("Initializing child process\n");
// close each end of the unused pipes
close(parent_to_child_fds[1]);
close(child_to_parent_fds[0]);
// wait for parent to do base setup
wait_for_other(parent_to_child_fds[0]);
if (arg_debug && child_pid == 1)
printf("PID namespace installed\n");
//****************************
// set hostname
//****************************
if (cfg.hostname) {
if (sethostname(cfg.hostname, strlen(cfg.hostname)) < 0)
errExit("sethostname");
}
//****************************
// mount namespace
//****************************
// mount events are not forwarded between the host the sandbox
if (mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL) < 0) {
chk_chroot();
}
//****************************
// log sandbox data
//****************************
if (cfg.name)
fs_logger2("sandbox name:", cfg.name);
fs_logger2int("sandbox pid:", (int) sandbox_pid);
if (cfg.chrootdir)
fs_logger("sandbox filesystem: chroot");
else if (arg_overlay)
fs_logger("sandbox filesystem: overlay");
else
fs_logger("sandbox filesystem: local");
fs_logger("install mount namespace");
//****************************
// netfilter etc.
//****************************
if (arg_netfilter && any_bridge_configured()) { // assuming by default the client filter
netfilter(arg_netfilter_file);
}
if (arg_netfilter6 && any_bridge_configured()) { // assuming by default the client filter
netfilter6(arg_netfilter6_file);
}
// load IBUS env variables
if (arg_nonetwork || any_bridge_configured() || any_interface_configured()) {
// do nothing - there are problems with ibus version 1.5.11
}
else
env_ibus_load();
// grab a copy of cp command
fs_build_cp_command();
// trace pre-install
if (arg_trace || arg_tracelog)
fs_trace_preload();
//****************************
// configure filesystem
//****************************
#ifdef HAVE_SECCOMP
int enforce_seccomp = 0;
#endif
#ifdef HAVE_CHROOT
if (cfg.chrootdir) {
fs_chroot(cfg.chrootdir);
// redo cp command
fs_build_cp_command();
// force caps and seccomp if not started as root
if (getuid() != 0) {
// force default seccomp inside the chroot, no keep or drop list
// the list build on top of the default drop list is kept intact
arg_seccomp = 1;
#ifdef HAVE_SECCOMP
enforce_seccomp = 1;
#endif
if (cfg.seccomp_list_drop) {
free(cfg.seccomp_list_drop);
cfg.seccomp_list_drop = NULL;
}
if (cfg.seccomp_list_keep) {
free(cfg.seccomp_list_keep);
cfg.seccomp_list_keep = NULL;
}
// disable all capabilities
if (arg_caps_default_filter || arg_caps_list)
fprintf(stderr, "Warning: all capabilities disabled for a regular user during chroot\n");
arg_caps_drop_all = 1;
// drop all supplementary groups; /etc/group file inside chroot
// is controlled by a regular usr
arg_nogroups = 1;
if (!arg_quiet)
printf("Dropping all Linux capabilities and enforcing default seccomp filter\n");
}
else
arg_seccomp = 1;
//****************************
// trace pre-install, this time inside chroot
//****************************
if (arg_trace || arg_tracelog)
fs_trace_preload();
}
else
#endif
if (arg_overlay)
fs_overlayfs();
else
fs_basic_fs();
//****************************
// set hostname in /etc/hostname
//****************************
if (cfg.hostname) {
fs_hostname(cfg.hostname);
}
//****************************
// private mode
//****************************
if (arg_private) {
if (cfg.home_private) // --private=
fs_private_homedir();
else // --private
fs_private();
}
if (arg_private_dev)
fs_private_dev();
if (arg_private_etc) {
fs_private_etc_list();
// create /etc/ld.so.preload file again
if (arg_trace || arg_tracelog)
fs_trace_preload();
}
if (arg_private_bin)
fs_private_bin_list();
if (arg_private_tmp)
fs_private_tmp();
//****************************
// apply the profile file
//****************************
if (cfg.profile) {
// apply all whitelist commands ...
fs_whitelist();
// ... followed by blacklist commands
fs_blacklist();
}
//****************************
// install trace
//****************************
if (arg_trace || arg_tracelog)
fs_trace();
//****************************
// update /proc, /dev, /boot directorymy
//****************************
fs_proc_sys_dev_boot();
//****************************
// --nosound and fix for pulseaudio 7.0
//****************************
if (arg_nosound)
pulseaudio_disable();
else
pulseaudio_init();
//****************************
// networking
//****************************
if (arg_nonetwork) {
net_if_up("lo");
if (arg_debug)
printf("Network namespace enabled, only loopback interface available\n");
}
else if (any_bridge_configured() || any_interface_configured()) {
// configure lo and eth0...eth3
net_if_up("lo");
if (mac_not_zero(cfg.bridge0.macsandbox))
net_config_mac(cfg.bridge0.devsandbox, cfg.bridge0.macsandbox);
sandbox_if_up(&cfg.bridge0);
if (mac_not_zero(cfg.bridge1.macsandbox))
net_config_mac(cfg.bridge1.devsandbox, cfg.bridge1.macsandbox);
sandbox_if_up(&cfg.bridge1);
if (mac_not_zero(cfg.bridge2.macsandbox))
net_config_mac(cfg.bridge2.devsandbox, cfg.bridge2.macsandbox);
sandbox_if_up(&cfg.bridge2);
if (mac_not_zero(cfg.bridge3.macsandbox))
net_config_mac(cfg.bridge3.devsandbox, cfg.bridge3.macsandbox);
sandbox_if_up(&cfg.bridge3);
// add a default route
if (cfg.defaultgw) {
// set the default route
if (net_add_route(0, 0, cfg.defaultgw))
fprintf(stderr, "Warning: cannot configure default route\n");
}
// enable interfaces
if (cfg.interface0.configured && cfg.interface0.ip) {
if (arg_debug)
printf("Configuring %d.%d.%d.%d address on interface %s\n", PRINT_IP(cfg.interface0.ip), cfg.interface0.dev);
net_if_ip(cfg.interface0.dev, cfg.interface0.ip, cfg.interface0.mask, cfg.interface0.mtu);
net_if_up(cfg.interface0.dev);
}
if (cfg.interface1.configured && cfg.interface1.ip) {
if (arg_debug)
printf("Configuring %d.%d.%d.%d address on interface %s\n", PRINT_IP(cfg.interface1.ip), cfg.interface1.dev);
net_if_ip(cfg.interface1.dev, cfg.interface1.ip, cfg.interface1.mask, cfg.interface1.mtu);
net_if_up(cfg.interface1.dev);
}
if (cfg.interface2.configured && cfg.interface2.ip) {
if (arg_debug)
printf("Configuring %d.%d.%d.%d address on interface %s\n", PRINT_IP(cfg.interface2.ip), cfg.interface2.dev);
net_if_ip(cfg.interface2.dev, cfg.interface2.ip, cfg.interface2.mask, cfg.interface2.mtu);
net_if_up(cfg.interface2.dev);
}
if (cfg.interface3.configured && cfg.interface3.ip) {
if (arg_debug)
printf("Configuring %d.%d.%d.%d address on interface %s\n", PRINT_IP(cfg.interface3.ip), cfg.interface3.dev);
net_if_ip(cfg.interface3.dev, cfg.interface3.ip, cfg.interface3.mask, cfg.interface3.mtu);
net_if_up(cfg.interface3.dev);
}
if (arg_debug)
printf("Network namespace enabled\n");
}
// if any dns server is configured, it is time to set it now
fs_resolvconf();
fs_logger_print();
fs_logger_change_owner();
// print network configuration
if (!arg_quiet) {
if (any_bridge_configured() || any_interface_configured() || cfg.defaultgw || cfg.dns1) {
printf("\n");
if (any_bridge_configured() || any_interface_configured())
net_ifprint();
if (cfg.defaultgw != 0)
printf("Default gateway %d.%d.%d.%d\n", PRINT_IP(cfg.defaultgw));
if (cfg.dns1 != 0)
printf("DNS server %d.%d.%d.%d\n", PRINT_IP(cfg.dns1));
if (cfg.dns2 != 0)
printf("DNS server %d.%d.%d.%d\n", PRINT_IP(cfg.dns2));
if (cfg.dns3 != 0)
printf("DNS server %d.%d.%d.%d\n", PRINT_IP(cfg.dns3));
printf("\n");
}
}
fs_delete_cp_command();
//****************************
// set application environment
//****************************
prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0); // kill the child in case the parent died
int cwd = 0;
if (cfg.cwd) {
if (chdir(cfg.cwd) == 0)
cwd = 1;
}
if (!cwd) {
if (chdir("/") < 0)
errExit("chdir");
if (cfg.homedir) {
struct stat s;
if (stat(cfg.homedir, &s) == 0) {
/* coverity[toctou] */
if (chdir(cfg.homedir) < 0)
errExit("chdir");
}
}
}
// set environment
env_defaults();
// set user-supplied environment variables
env_apply();
// set nice
if (arg_nice) {
errno = 0;
int rv = nice(cfg.nice);
(void) rv;
if (errno) {
fprintf(stderr, "Warning: cannot set nice value\n");
errno = 0;
}
}
// clean /tmp/.X11-unix sockets
fs_x11();
//****************************
// set security filters
//****************************
// set capabilities
if (!arg_noroot)
set_caps();
// set rlimits
set_rlimits();
// set seccomp
#ifdef HAVE_SECCOMP
// install protocol filter
if (cfg.protocol) {
protocol_filter(); // install filter
protocol_filter_save(); // save filter in PROTOCOL_CFG
}
// if a keep list is available, disregard the drop list
if (arg_seccomp == 1) {
if (cfg.seccomp_list_keep)
seccomp_filter_keep();
else if (cfg.seccomp_list_errno)
seccomp_filter_errno();
else
seccomp_filter_drop(enforce_seccomp);
}
#endif
// set cpu affinity
if (cfg.cpus) {
save_cpu(); // save cpu affinity mask to CPU_CFG file
set_cpu_affinity();
}
// save cgroup in CGROUP_CFG file
if (cfg.cgroup)
save_cgroup();
//****************************************
// drop privileges or create a new user namespace
//****************************************
save_nogroups();
if (arg_noroot) {
int rv = unshare(CLONE_NEWUSER);
if (rv == -1) {
fprintf(stderr, "Error: cannot mount a new user namespace\n");
perror("unshare");
drop_privs(arg_nogroups);
}
}
else
drop_privs(arg_nogroups);
// notify parent that new user namespace has been created so a proper
// UID/GID map can be setup
notify_other(child_to_parent_fds[1]);
close(child_to_parent_fds[1]);
// wait for parent to finish setting up a proper UID/GID map
wait_for_other(parent_to_child_fds[0]);
close(parent_to_child_fds[0]);
// somehow, the new user namespace resets capabilities;
// we need to do them again
if (arg_noroot) {
set_caps();
if (arg_debug)
printf("noroot user namespace installed\n");
}
//****************************************
// fork the application and monitor it
//****************************************
pid_t app_pid = fork();
if (app_pid == -1)
errExit("fork");
if (app_pid == 0) {
prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0); // kill the child in case the parent died
start_application(); // start app
}
monitor_application(app_pid); // monitor application
return 0;
}
void join(pid_t pid, int argc, char **argv, int index) {
EUID_ASSERT();
char *homedir = cfg.homedir;
extract_command(argc, argv, index);
signal (SIGTERM, signal_handler);
// if the pid is that of a firejail process, use the pid of the first child process
EUID_ROOT();
char *comm = pid_proc_comm(pid);
EUID_USER();
if (comm) {
if (strcmp(comm, "firejail") == 0) {
pid_t child;
if (find_child(pid, &child) == 0) {
pid = child;
printf("Switching to pid %u, the first child process inside the sandbox\n", (unsigned) pid);
}
}
free(comm);
}
// check privileges for non-root users
uid_t uid = getuid();
if (uid != 0) {
uid_t sandbox_uid = pid_get_uid(pid);
if (uid != sandbox_uid) {
fprintf(stderr, "Error: permission is denied to join a sandbox created by a different user.\n");
exit(1);
}
}
EUID_ROOT();
// in user mode set caps seccomp, cpu, cgroup, etc
if (getuid() != 0) {
extract_caps_seccomp(pid);
extract_cpu(pid);
extract_cgroup(pid);
extract_nogroups(pid);
extract_user_namespace(pid);
}
// set cgroup
if (cfg.cgroup) // not available for uid 0
set_cgroup(cfg.cgroup);
// join namespaces
if (arg_join_network) {
if (join_namespace(pid, "net"))
exit(1);
}
else if (arg_join_filesystem) {
if (join_namespace(pid, "mnt"))
exit(1);
}
else {
if (join_namespace(pid, "ipc") ||
join_namespace(pid, "net") ||
join_namespace(pid, "pid") ||
join_namespace(pid, "uts") ||
join_namespace(pid, "mnt"))
exit(1);
}
pid_t child = fork();
if (child < 0)
errExit("fork");
if (child == 0) {
// chroot into /proc/PID/root directory
char *rootdir;
if (asprintf(&rootdir, "/proc/%d/root", pid) == -1)
errExit("asprintf");
int rv;
if (!arg_join_network) {
rv = chroot(rootdir); // this will fail for processes in sandboxes not started with --chroot option
if (rv == 0)
printf("changing root to %s\n", rootdir);
}
prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0); // kill the child in case the parent died
if (chdir("/") < 0)
errExit("chdir");
if (homedir) {
struct stat s;
if (stat(homedir, &s) == 0) {
/* coverity[toctou] */
if (chdir(homedir) < 0)
errExit("chdir");
}
}
// set cpu affinity
if (cfg.cpus) // not available for uid 0
set_cpu_affinity();
// set caps filter
if (apply_caps == 1) // not available for uid 0
caps_set(caps);
#ifdef HAVE_SECCOMP
// read cfg.protocol from file
if (getuid() != 0)
protocol_filter_load(RUN_PROTOCOL_CFG);
if (cfg.protocol) { // not available for uid 0
seccomp_load(RUN_SECCOMP_PROTOCOL); // install filter
}
// set seccomp filter
if (apply_seccomp == 1) // not available for uid 0
seccomp_load(RUN_SECCOMP_CFG);
#endif
// mount user namespace or drop privileges
if (arg_noroot) { // not available for uid 0
if (arg_debug)
printf("Joining user namespace\n");
if (join_namespace(1, "user"))
exit(1);
// user namespace resets capabilities
// set caps filter
if (apply_caps == 1) // not available for uid 0
caps_set(caps);
}
else
drop_privs(arg_nogroups); // nogroups not available for uid 0
// set nice
if (arg_nice) {
errno = 0;
int rv = nice(cfg.nice);
(void) rv;
if (errno) {
fprintf(stderr, "Warning: cannot set nice value\n");
errno = 0;
}
}
env_defaults();
if (cfg.command_line == NULL) {
assert(cfg.shell);
cfg.command_line = cfg.shell;
cfg.window_title = cfg.shell;
}
int cwd = 0;
if (cfg.cwd) {
if (chdir(cfg.cwd) == 0)
cwd = 1;
}
if (!cwd) {
if (chdir("/") < 0)
errExit("chdir");
if (cfg.homedir) {
struct stat s;
if (stat(cfg.homedir, &s) == 0) {
/* coverity[toctou] */
if (chdir(cfg.homedir) < 0)
errExit("chdir");
}
}
}
start_application();
// it will never get here!!!
}
// wait for the child to finish
waitpid(child, NULL, 0);
flush_stdin();
exit(0);
}
