2 * Copyright (C) 2015 John Crispin <blogic@openwrt.org>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU Lesser General Public License version 2.1
6 * as published by the Free Software Foundation
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
15 #include <sys/mount.h>
16 #include <sys/prctl.h>
18 #include <sys/types.h>
30 #include <linux/limits.h>
33 #include "capabilities.h"
39 #include <libubox/uloop.h>
42 #define STACK_SIZE (1024 * 1024)
43 #define OPT_ARGS "S:C:n:h:r:w:d:psulocU:G:NR:fFO:T:E"
68 extern int pivot_root(const char *new_root, const char *put_old);
72 static char child_stack[STACK_SIZE];
74 static int mkdir_p(char *dir, mode_t mask)
76 char *l = strrchr(dir, '/');
84 if (mkdir_p(dir, mask))
89 ret = mkdir(dir, mask);
90 if (ret && errno == EEXIST)
94 ERROR("mkdir(%s, %d) failed: %m\n", dir, mask);
99 static int _mount_bind(const char *root, const char *path, const char *target, int readonly, int strict, int error)
104 int remount_flags = MS_BIND | MS_REMOUNT;
106 if (stat(path, &s)) {
107 ERROR("stat(%s) failed: %m\n", path);
111 snprintf(new, sizeof(new), "%s%s", root, target?target:path);
113 if (S_ISDIR(s.st_mode)) {
116 mkdir_p(dirname(new), 0755);
117 snprintf(new, sizeof(new), "%s%s", root, target?target:path);
118 fd = creat(new, 0644);
120 ERROR("creat(%s) failed: %m\n", new);
126 if (mount(path, new, NULL, MS_BIND, NULL)) {
127 ERROR("failed to mount -B %s %s: %m\n", path, new);
132 remount_flags |= MS_RDONLY;
135 remount_flags |= MS_NOEXEC | MS_NOSUID | MS_NODEV;
137 if ((strict || readonly) && mount(NULL, new, NULL, remount_flags, NULL)) {
138 ERROR("failed to remount (%s%s%s) %s: %m\n", readonly?"ro":"rw",
139 (readonly && strict)?", ":"", strict?"strict":"", new);
143 DEBUG("mount -B %s %s (%s%s%s)\n", path, new,
144 readonly?"ro":"rw", (readonly && strict)?", ":"", strict?"strict":"");
149 int mount_bind(const char *root, const char *path, int readonly, int error) {
150 return _mount_bind(root, path, NULL, readonly, 0, error);
153 static int mount_overlay(char *jail_root, char *overlaydir) {
154 char *upperdir, *workdir, *optsstr;
155 const char mountoptsformat[] = "lowerdir=%s,upperdir=%s,workdir=%s";
158 if (asprintf(&upperdir, "%s%s", overlaydir, "/upper") < 0)
161 if (asprintf(&workdir, "%s%s", overlaydir, "/work") < 0)
164 if (asprintf(&optsstr, mountoptsformat, jail_root, upperdir, workdir) < 0)
167 if (mkdir_p(upperdir, 0755) || mkdir_p(workdir, 0755))
170 DEBUG("mount -t overlay %s %s (%s)\n", jail_root, jail_root, optsstr);
172 if (mount(jail_root, jail_root, "overlay", MS_NOATIME, optsstr))
187 static int build_jail_fs(void)
189 char jail_root[] = "/tmp/ujail-XXXXXX";
190 char tmpovdir[] = "/tmp/ujail-overlay-XXXXXX";
191 char tmpdevdir[] = "/tmp/ujail-XXXXXX/dev";
192 char *overlaydir = NULL;
194 if (mkdtemp(jail_root) == NULL) {
195 ERROR("mkdtemp(%s) failed: %m\n", jail_root);
199 /* oldroot can't be MS_SHARED else pivot_root() fails */
200 if (mount("none", "/", NULL, MS_REC|MS_PRIVATE, NULL)) {
201 ERROR("private mount failed %m\n");
206 if (mount(opts.extroot, jail_root, NULL, MS_BIND, NULL)) {
207 ERROR("extroot mount failed %m\n");
211 if (mount("tmpfs", jail_root, "tmpfs", MS_NOATIME, "mode=0755")) {
212 ERROR("tmpfs mount failed %m\n");
217 if (opts.tmpoverlaysize) {
218 char mountoptsstr[] = "mode=0755,size=XXXXXXXX";
220 snprintf(mountoptsstr, sizeof(mountoptsstr),
221 "mode=0755,size=%s", opts.tmpoverlaysize);
222 if (mkdtemp(tmpovdir) == NULL) {
223 ERROR("mkdtemp(%s) failed: %m\n", jail_root);
226 if (mount("tmpfs", tmpovdir, "tmpfs", MS_NOATIME,
228 ERROR("failed to mount tmpfs for overlay (size=%s)\n", opts.tmpoverlaysize);
231 overlaydir = tmpovdir;
235 overlaydir = opts.overlaydir;
238 mount_overlay(jail_root, overlaydir);
240 if (chdir(jail_root)) {
241 ERROR("chdir(%s) (jail_root) failed: %m\n", jail_root);
245 snprintf(tmpdevdir, sizeof(tmpdevdir), "%s/dev", jail_root);
246 mkdir_p(tmpdevdir, 0755);
247 if (mount(NULL, tmpdevdir, "tmpfs", MS_NOATIME | MS_NOEXEC | MS_NOSUID, "size=1M"))
250 if (mount_all(jail_root)) {
251 ERROR("mount_all() failed\n");
255 if (opts.namespace & CLONE_NEWNET) {
256 char hostdir[PATH_MAX], jailetc[PATH_MAX], jaillink[PATH_MAX];
258 snprintf(hostdir, PATH_MAX, "/tmp/resolv.conf-%s.d", opts.name);
259 mkdir_p(hostdir, 0755);
260 _mount_bind(jail_root, hostdir, "/tmp/resolv.conf.d", 1, 1, -1);
261 snprintf(jailetc, PATH_MAX, "%s/etc", jail_root);
262 mkdir_p(jailetc, 0755);
263 snprintf(jaillink, PATH_MAX, "%s/etc/resolv.conf", jail_root);
266 symlink("../tmp/resolv.conf.d/resolv.conf.auto", jaillink);
269 char dirbuf[sizeof(jail_root) + 4];
270 snprintf(dirbuf, sizeof(dirbuf), "%s/old", jail_root);
273 if (pivot_root(jail_root, dirbuf) == -1) {
274 ERROR("pivot_root(%s, %s) failed: %m\n", jail_root, dirbuf);
278 ERROR("chdir(/) (after pivot_root) failed: %m\n");
282 snprintf(dirbuf, sizeof(dirbuf), "/old%s", jail_root);
283 umount2(dirbuf, MNT_DETACH);
285 if (opts.tmpoverlaysize) {
286 char tmpdirbuf[sizeof(tmpovdir) + 4];
287 snprintf(tmpdirbuf, sizeof(tmpdirbuf), "/old%s", tmpovdir);
288 umount2(tmpdirbuf, MNT_DETACH);
292 umount2("/old", MNT_DETACH);
296 mkdir("/proc", 0755);
297 mount("proc", "/proc", "proc", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
299 * make /proc/sys read-only while keeping read-write to
300 * /proc/sys/net if CLONE_NEWNET is set.
302 if (opts.namespace & CLONE_NEWNET)
303 mount("/proc/sys/net", "/proc/self/net", NULL, MS_BIND, 0);
305 mount("/proc/sys", "/proc/sys", NULL, MS_BIND, 0);
306 mount(NULL, "/proc/sys", NULL, MS_REMOUNT | MS_RDONLY, 0);
307 mount(NULL, "/proc", NULL, MS_REMOUNT, 0);
309 if (opts.namespace & CLONE_NEWNET)
310 mount("/proc/self/net", "/proc/sys/net", NULL, MS_MOVE, 0);
314 mount("sysfs", "/sys", "sysfs", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID | MS_RDONLY, 0);
317 mount(NULL, "/", NULL, MS_RDONLY | MS_REMOUNT, 0);
322 static int write_uid_gid_map(pid_t child_pid, bool gidmap, int id)
326 const char *map_format = "%d %d %d\n";
327 if (snprintf(map_path, sizeof(map_path), "/proc/%d/%s",
328 child_pid, gidmap?"gid_map":"uid_map") < 0)
331 if ((map_file = open(map_path, O_WRONLY)) == -1)
334 if (dprintf(map_file, map_format, 0, id, 1) == -1) {
343 static int write_setgroups(pid_t child_pid, bool allow)
346 char setgroups_path[64];
348 if (snprintf(setgroups_path, sizeof(setgroups_path), "/proc/%d/setgroups",
353 if ((setgroups_file = open(setgroups_path, O_WRONLY)) == -1) {
357 if (dprintf(setgroups_file, allow?"allow":"deny") == -1) {
358 close(setgroups_file);
362 close(setgroups_file);
366 static void get_jail_user(int *user, int *user_gid, int *gr_gid)
368 struct passwd *p = NULL;
369 struct group *g = NULL;
372 p = getpwnam(opts.user);
374 ERROR("failed to get uid/gid for user %s: %d (%s)\n",
375 opts.user, errno, strerror(errno));
379 *user_gid = p->pw_gid;
386 g = getgrnam(opts.group);
388 ERROR("failed to get gid for group %s: %m\n", opts.group);
397 static void set_jail_user(int pw_uid, int user_gid, int gr_gid)
399 if ((user_gid != -1) && initgroups(opts.user, user_gid)) {
400 ERROR("failed to initgroups() for user %s: %m\n", opts.user);
404 if ((gr_gid != -1) && setregid(gr_gid, gr_gid)) {
405 ERROR("failed to set group id %d: %m\n", gr_gid);
409 if ((pw_uid != -1) && setreuid(pw_uid, pw_uid)) {
410 ERROR("failed to set user id %d: %m\n", pw_uid);
416 static char** build_envp(const char *seccomp)
418 static char *envp[MAX_ENVP];
419 static char preload_var[PATH_MAX];
420 static char seccomp_var[PATH_MAX];
421 static char debug_var[] = "LD_DEBUG=all";
422 static char container_var[] = "container=ujail";
423 const char *preload_lib = find_lib("libpreload-seccomp.so");
426 if (seccomp && !preload_lib) {
427 ERROR("failed to add preload-lib to env\n");
431 snprintf(seccomp_var, sizeof(seccomp_var), "SECCOMP_FILE=%s", seccomp);
432 envp[count++] = seccomp_var;
433 snprintf(preload_var, sizeof(preload_var), "LD_PRELOAD=%s", preload_lib);
434 envp[count++] = preload_var;
437 envp[count++] = container_var;
440 envp[count++] = debug_var;
445 static void usage(void)
447 fprintf(stderr, "ujail <options> -- <binary> <params ...>\n");
448 fprintf(stderr, " -d <num>\tshow debug log (increase num to increase verbosity)\n");
449 fprintf(stderr, " -S <file>\tseccomp filter config\n");
450 fprintf(stderr, " -C <file>\tcapabilities drop config\n");
451 fprintf(stderr, " -c\t\tset PR_SET_NO_NEW_PRIVS\n");
452 fprintf(stderr, " -n <name>\tthe name of the jail\n");
453 fprintf(stderr, "namespace jail options:\n");
454 fprintf(stderr, " -h <hostname>\tchange the hostname of the jail\n");
455 fprintf(stderr, " -N\t\tjail has network namespace\n");
456 fprintf(stderr, " -f\t\tjail has user namespace\n");
457 fprintf(stderr, " -F\t\tjail has cgroups namespace\n");
458 fprintf(stderr, " -r <file>\treadonly files that should be staged\n");
459 fprintf(stderr, " -w <file>\twriteable files that should be staged\n");
460 fprintf(stderr, " -p\t\tjail has /proc\n");
461 fprintf(stderr, " -s\t\tjail has /sys\n");
462 fprintf(stderr, " -l\t\tjail has /dev/log\n");
463 fprintf(stderr, " -u\t\tjail has a ubus socket\n");
464 fprintf(stderr, " -U <name>\tuser to run jailed process\n");
465 fprintf(stderr, " -G <name>\tgroup to run jailed process\n");
466 fprintf(stderr, " -o\t\tremont jail root (/) read only\n");
467 fprintf(stderr, " -R <dir>\texternal jail rootfs (system container)\n");
468 fprintf(stderr, " -O <dir>\tdirectory for r/w overlayfs\n");
469 fprintf(stderr, " -T <size>\tuse tmpfs r/w overlayfs with <size>\n");
470 fprintf(stderr, " -E\t\tfail if jail cannot be setup\n");
471 fprintf(stderr, "\nWarning: by default root inside the jail is the same\n\
472 and he has the same powers as root outside the jail,\n\
473 thus he can escape the jail and/or break stuff.\n\
474 Please use seccomp/capabilities (-S/-C) to restrict his powers\n\n\
475 If you use none of the namespace jail options,\n\
476 ujail will not use namespace/build a jail,\n\
477 and will only drop capabilities/apply seccomp filter.\n\n");
480 static int exec_jail(void *pipes_ptr)
482 int *pipes = (int*)pipes_ptr;
484 int pw_uid, pw_gid, gr_gid;
491 if (write(pipes[1], buf, 1) < 1) {
492 ERROR("can't write to parent\n");
495 if (read(pipes[2], buf, 1) < 1) {
496 ERROR("can't read from parent\n");
500 ERROR("parent had an error, child exiting\n");
507 if (opts.namespace & CLONE_NEWUSER) {
516 // if (setgroups(0, NULL) < 0) {
517 // ERROR("setgroups\n");
518 // exit(EXIT_FAILURE);
522 if (opts.namespace && opts.hostname && strlen(opts.hostname) > 0
523 && sethostname(opts.hostname, strlen(opts.hostname))) {
524 ERROR("sethostname(%s) failed: %m\n", opts.hostname);
528 if ((opts.namespace & CLONE_NEWNS) && build_jail_fs()) {
529 ERROR("failed to build jail fs\n");
533 if (opts.capabilities && drop_capabilities(opts.capabilities))
536 if (opts.no_new_privs && prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
537 ERROR("prctl(PR_SET_NO_NEW_PRIVS) failed: %m\n");
541 if (!(opts.namespace & CLONE_NEWUSER)) {
542 get_jail_user(&pw_uid, &pw_gid, &gr_gid);
543 set_jail_user(pw_uid, pw_gid, gr_gid);
546 char **envp = build_envp(opts.seccomp);
550 INFO("exec-ing %s\n", *opts.jail_argv);
551 execve(*opts.jail_argv, opts.jail_argv, envp);
552 /* we get there only if execve fails */
553 ERROR("failed to execve %s: %m\n", *opts.jail_argv);
557 static int jail_running = 1;
558 static int jail_return_code = 0;
560 static void jail_process_timeout_cb(struct uloop_timeout *t);
561 static struct uloop_timeout jail_process_timeout = {
562 .cb = jail_process_timeout_cb,
565 static void jail_process_handler(struct uloop_process *c, int ret)
567 uloop_timeout_cancel(&jail_process_timeout);
568 if (WIFEXITED(ret)) {
569 jail_return_code = WEXITSTATUS(ret);
570 INFO("jail (%d) exited with exit: %d\n", c->pid, jail_return_code);
572 jail_return_code = WTERMSIG(ret);
573 INFO("jail (%d) exited with signal: %d\n", c->pid, jail_return_code);
579 static struct uloop_process jail_process = {
580 .cb = jail_process_handler,
583 static void jail_process_timeout_cb(struct uloop_timeout *t)
585 DEBUG("jail process failed to stop, sending SIGKILL\n");
586 kill(jail_process.pid, SIGKILL);
589 static void jail_handle_signal(int signo)
591 DEBUG("forwarding signal %d to the jailed process\n", signo);
592 kill(jail_process.pid, signo);
595 static int netns_open_pid(const pid_t target_ns)
597 char pid_net_path[PATH_MAX];
599 snprintf(pid_net_path, sizeof(pid_net_path), "/proc/%u/ns/net", target_ns);
601 return open(pid_net_path, O_RDONLY);
604 static void netns_updown(pid_t pid, bool start)
606 struct ubus_context *ctx = ubus_connect(NULL);
607 static struct blob_buf req;
613 blob_buf_init(&req, 0);
614 blobmsg_add_string(&req, "jail", opts.name);
615 blobmsg_add_u32(&req, "pid", pid);
616 blobmsg_add_u8(&req, "start", start);
618 if (ubus_lookup_id(ctx, "network", &id) ||
619 ubus_invoke(ctx, id, "netns_updown", req.head, NULL, NULL, 3000))
620 INFO("ubus request failed\n");
626 int main(int argc, char **argv)
629 uid_t uid = getuid();
630 char log[] = "/dev/log";
631 char ubus[] = "/var/run/ubus.sock";
638 ERROR("not root, aborting: %m\n");
644 init_library_search();
646 while ((ch = getopt(argc, argv, OPT_ARGS)) != -1) {
649 debug = atoi(optarg);
652 opts.namespace |= CLONE_NEWNS;
656 opts.namespace |= CLONE_NEWNS;
660 opts.namespace |= CLONE_NEWUSER;
663 opts.namespace |= CLONE_NEWCGROUP;
666 opts.extroot = optarg;
669 opts.namespace |= CLONE_NEWNS;
673 opts.seccomp = optarg;
674 add_mount(optarg, 1, -1);
677 opts.capabilities = optarg;
680 opts.no_new_privs = 1;
686 opts.namespace |= CLONE_NEWNET;
689 opts.namespace |= CLONE_NEWUTS;
690 opts.hostname = optarg;
693 opts.namespace |= CLONE_NEWNS;
694 add_path_and_deps(optarg, 1, 0, 0);
697 opts.namespace |= CLONE_NEWNS;
698 add_path_and_deps(optarg, 0, 0, 0);
701 opts.namespace |= CLONE_NEWNS;
702 add_mount(ubus, 0, -1);
705 opts.namespace |= CLONE_NEWNS;
706 add_mount(log, 0, -1);
715 opts.overlaydir = optarg;
718 opts.tmpoverlaysize = optarg;
721 opts.require_jail = 1;
727 opts.namespace |= CLONE_NEWIPC | CLONE_NEWPID;
729 if (opts.tmpoverlaysize && strlen(opts.tmpoverlaysize) > 8) {
730 ERROR("size parameter too long: \"%s\"\n", opts.tmpoverlaysize);
734 /* no <binary> param found */
735 if (argc - optind < 1) {
739 if (!(opts.namespace||opts.capabilities||opts.seccomp)) {
740 ERROR("Not using namespaces, capabilities or seccomp !!!\n\n");
744 DEBUG("Using namespaces(0x%08x), capabilities(%d), seccomp(%d)\n",
746 opts.capabilities != 0,
749 opts.jail_argv = &argv[optind];
751 get_jail_user(&opts.pw_uid, &opts.pw_gid, &opts.gr_gid);
754 if (opts.namespace && add_path_and_deps(*opts.jail_argv, 1, -1, 0)) {
755 ERROR("failed to load dependencies\n");
760 if (opts.namespace && opts.seccomp && add_path_and_deps("libpreload-seccomp.so", 1, -1, 1)) {
761 ERROR("failed to load libpreload-seccomp.so\n");
762 if (opts.require_jail)
767 prctl(PR_SET_NAME, opts.name, NULL, NULL, NULL);
771 sigfillset(&sigmask);
772 for (i = 0; i < _NSIG; i++) {
773 struct sigaction s = { 0 };
775 if (!sigismember(&sigmask, i))
777 if ((i == SIGCHLD) || (i == SIGPIPE))
780 s.sa_handler = jail_handle_signal;
781 sigaction(i, &s, NULL);
784 if (opts.namespace) {
785 if (opts.namespace & CLONE_NEWNS) {
786 add_mount("/dev/full", 0, -1);
787 add_mount("/dev/null", 0, -1);
788 add_mount("/dev/random", 0, -1);
789 add_mount("/dev/urandom", 0, -1);
790 add_mount("/dev/tty", 0, -1);
791 add_mount("/dev/zero", 0, -1);
792 add_mount("/dev/console", 0, -1);
794 if (!opts.extroot && (opts.user || opts.group)) {
795 add_mount("/etc/passwd", 0, -1);
796 add_mount("/etc/group", 0, -1);
799 #if defined(__GLIBC__)
801 add_mount("/etc/nsswitch.conf", 0, -1);
804 if (!(opts.namespace & CLONE_NEWNET)) {
805 add_mount("/etc/resolv.conf", 0, -1);
809 if (pipe(&pipes[0]) < 0 || pipe(&pipes[2]) < 0)
812 jail_process.pid = clone(exec_jail, child_stack + STACK_SIZE, SIGCHLD | opts.namespace, &pipes);
814 jail_process.pid = fork();
817 if (jail_process.pid > 0) {
822 if (read(pipes[0], sig_buf, 1) < 1) {
823 ERROR("can't read from child\n");
827 if (opts.namespace & CLONE_NEWUSER) {
828 bool has_gr = (opts.gr_gid != -1);
829 if (write_setgroups(jail_process.pid, false)) {
830 ERROR("can't write setgroups\n");
833 if (opts.pw_uid != -1) {
834 write_uid_gid_map(jail_process.pid, 0, opts.pw_uid);
835 write_uid_gid_map(jail_process.pid, 1, has_gr?opts.gr_gid:opts.pw_gid);
837 write_uid_gid_map(jail_process.pid, 0, 65534);
838 write_uid_gid_map(jail_process.pid, 1, has_gr?opts.gr_gid:65534);
842 if (opts.namespace & CLONE_NEWNET) {
843 netns_fd = netns_open_pid(jail_process.pid);
844 netns_updown(jail_process.pid, true);
848 if (write(pipes[3], sig_buf, 1) < 0) {
849 ERROR("can't write to child\n");
853 uloop_process_add(&jail_process);
856 DEBUG("uloop interrupted, killing jail process\n");
857 kill(jail_process.pid, SIGTERM);
858 uloop_timeout_set(&jail_process_timeout, 1000);
862 if (opts.namespace & CLONE_NEWNET) {
863 setns(netns_fd, CLONE_NEWNET);
864 netns_updown(getpid(), false);
867 return jail_return_code;
868 } else if (jail_process.pid == 0) {
869 /* fork child process */
870 return exec_jail(NULL);
872 ERROR("failed to clone/fork: %m\n");