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:"
45 #define NAMESPACE_MOUNT (1U << 0)
46 #define NAMESPACE_IPC (1U << 1)
47 #define NAMESPACE_NET (1U << 2)
48 #define NAMESPACE_PID (1U << 3)
49 #define NAMESPACE_USER (1U << 4)
50 #define NAMESPACE_UTS (1U << 5)
51 #define NAMESPACE_CGROUP (1U << 6)
69 extern int pivot_root(const char *new_root, const char *put_old);
73 static char child_stack[STACK_SIZE];
75 static int mkdir_p(char *dir, mode_t mask)
77 char *l = strrchr(dir, '/');
85 if (mkdir_p(dir, mask))
90 ret = mkdir(dir, mask);
91 if (ret && errno == EEXIST)
95 ERROR("mkdir(%s, %d) failed: %m\n", dir, mask);
100 static int _mount_bind(const char *root, const char *path, const char *target, int readonly, int strict, int error)
105 int remount_flags = MS_BIND | MS_REMOUNT;
107 if (stat(path, &s)) {
108 ERROR("stat(%s) failed: %m\n", path);
112 snprintf(new, sizeof(new), "%s%s", root, target?target:path);
114 if (S_ISDIR(s.st_mode)) {
117 mkdir_p(dirname(new), 0755);
118 snprintf(new, sizeof(new), "%s%s", root, target?target:path);
119 fd = creat(new, 0644);
121 ERROR("creat(%s) failed: %m\n", new);
127 if (mount(path, new, NULL, MS_BIND, NULL)) {
128 ERROR("failed to mount -B %s %s: %m\n", path, new);
133 remount_flags |= MS_RDONLY;
136 remount_flags |= MS_NOEXEC | MS_NOSUID | MS_NODEV;
138 if ((strict || readonly) && mount(NULL, new, NULL, remount_flags, NULL)) {
139 ERROR("failed to remount (%s%s%s) %s: %m\n", readonly?"ro":"rw",
140 (readonly && strict)?", ":"", strict?"strict":"", new);
144 DEBUG("mount -B %s %s (%s%s%s)\n", path, new,
145 readonly?"ro":"rw", (readonly && strict)?", ":"", strict?"strict":"");
150 int mount_bind(const char *root, const char *path, int readonly, int error) {
151 return _mount_bind(root, path, NULL, readonly, 0, error);
154 static int build_jail_fs(void)
156 char jail_root[] = "/tmp/ujail-XXXXXX";
157 if (mkdtemp(jail_root) == NULL) {
158 ERROR("mkdtemp(%s) failed: %m\n", jail_root);
162 /* oldroot can't be MS_SHARED else pivot_root() fails */
163 if (mount("none", "/", NULL, MS_REC|MS_PRIVATE, NULL)) {
164 ERROR("private mount failed %m\n");
169 if (mount(opts.extroot, jail_root, NULL, MS_BIND | MS_REC, NULL)) {
170 ERROR("extroot mount failed %m\n");
174 if (mount("tmpfs", jail_root, "tmpfs", MS_NOATIME, "mode=0755")) {
175 ERROR("tmpfs mount failed %m\n");
180 if (chdir(jail_root)) {
181 ERROR("chdir(%s) (jail_root) failed: %m\n", jail_root);
185 if (mount_all(jail_root)) {
186 ERROR("mount_all() failed\n");
190 if (opts.namespace & NAMESPACE_NET) {
191 char hostdir[PATH_MAX], jailetc[PATH_MAX], jaillink[PATH_MAX];
193 snprintf(hostdir, PATH_MAX, "/tmp/resolv.conf-%s.d", opts.name);
194 mkdir_p(hostdir, 0755);
195 _mount_bind(jail_root, hostdir, "/tmp/resolv.conf.d", 1, 1, -1);
196 snprintf(jailetc, PATH_MAX, "%s/etc", jail_root);
197 mkdir_p(jailetc, 0755);
198 snprintf(jaillink, PATH_MAX, "%s/etc/resolv.conf", jail_root);
199 symlink("../tmp/resolv.conf.d/resolv.conf.auto", jaillink);
202 char dirbuf[sizeof(jail_root) + 4];
203 snprintf(dirbuf, sizeof(dirbuf), "%s/old", jail_root);
206 if (pivot_root(jail_root, dirbuf) == -1) {
207 ERROR("pivot_root(%s, %s) failed: %m\n", jail_root, dirbuf);
211 ERROR("chdir(/) (after pivot_root) failed: %m\n");
215 snprintf(dirbuf, sizeof(dirbuf), "/old%s", jail_root);
217 umount2("/old", MNT_DETACH);
221 mkdir("/proc", 0755);
222 mount("proc", "/proc", "proc", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
226 mount("sysfs", "/sys", "sysfs", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
229 mount(NULL, "/", NULL, MS_RDONLY | MS_REMOUNT, 0);
235 static char** build_envp(const char *seccomp)
237 static char *envp[MAX_ENVP];
238 static char preload_var[PATH_MAX];
239 static char seccomp_var[PATH_MAX];
240 static char debug_var[] = "LD_DEBUG=all";
241 static char container_var[] = "container=ujail";
242 const char *preload_lib = find_lib("libpreload-seccomp.so");
245 if (seccomp && !preload_lib) {
246 ERROR("failed to add preload-lib to env\n");
250 snprintf(seccomp_var, sizeof(seccomp_var), "SECCOMP_FILE=%s", seccomp);
251 envp[count++] = seccomp_var;
252 snprintf(preload_var, sizeof(preload_var), "LD_PRELOAD=%s", preload_lib);
253 envp[count++] = preload_var;
257 envp[count++] = container_var;
260 envp[count++] = debug_var;
265 static void usage(void)
267 fprintf(stderr, "ujail <options> -- <binary> <params ...>\n");
268 fprintf(stderr, " -d <num>\tshow debug log (increase num to increase verbosity)\n");
269 fprintf(stderr, " -S <file>\tseccomp filter config\n");
270 fprintf(stderr, " -C <file>\tcapabilities drop config\n");
271 fprintf(stderr, " -c\t\tset PR_SET_NO_NEW_PRIVS\n");
272 fprintf(stderr, " -n <name>\tthe name of the jail\n");
273 fprintf(stderr, "namespace jail options:\n");
274 fprintf(stderr, " -h <hostname>\tchange the hostname of the jail\n");
275 fprintf(stderr, " -N\t\tjail has network namespace\n");
276 fprintf(stderr, " -r <file>\treadonly files that should be staged\n");
277 fprintf(stderr, " -w <file>\twriteable files that should be staged\n");
278 fprintf(stderr, " -p\t\tjail has /proc\n");
279 fprintf(stderr, " -s\t\tjail has /sys\n");
280 fprintf(stderr, " -l\t\tjail has /dev/log\n");
281 fprintf(stderr, " -u\t\tjail has a ubus socket\n");
282 fprintf(stderr, " -U <name>\tuser to run jailed process\n");
283 fprintf(stderr, " -G <name>\tgroup to run jailed process\n");
284 fprintf(stderr, " -o\t\tremont jail root (/) read only\n");
285 fprintf(stderr, " -R <dir>\texternal jail rootfs (system container)\n");
286 fprintf(stderr, "\nWarning: by default root inside the jail is the same\n\
287 and he has the same powers as root outside the jail,\n\
288 thus he can escape the jail and/or break stuff.\n\
289 Please use seccomp/capabilities (-S/-C) to restrict his powers\n\n\
290 If you use none of the namespace jail options,\n\
291 ujail will not use namespace/build a jail,\n\
292 and will only drop capabilities/apply seccomp filter.\n\n");
295 static int exec_jail(void *_notused)
297 struct passwd *p = NULL;
298 struct group *g = NULL;
300 if (opts.capabilities && drop_capabilities(opts.capabilities))
303 if (opts.no_new_privs && prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
304 ERROR("prctl(PR_SET_NO_NEW_PRIVS) failed: %m\n");
308 if (opts.namespace && opts.hostname && strlen(opts.hostname) > 0
309 && sethostname(opts.hostname, strlen(opts.hostname))) {
310 ERROR("sethostname(%s) failed: %m\n", opts.hostname);
314 if (opts.namespace && build_jail_fs()) {
315 ERROR("failed to build jail fs\n");
320 p = getpwnam(opts.user);
322 ERROR("failed to get uid/gid for user %s: %d (%s)\n",
323 opts.user, errno, strerror(errno));
329 g = getgrnam(opts.group);
331 ERROR("failed to get gid for group %s: %m\n", opts.group);
336 if (p && p->pw_gid && initgroups(opts.user, p->pw_gid)) {
337 ERROR("failed to initgroups() for user %s: %m\n", opts.user);
341 if (g && g->gr_gid && setgid(g->gr_gid)) {
342 ERROR("failed to set group id %d: %m\n", g?g->gr_gid:p->pw_gid);
346 if (p && p->pw_uid && setuid(p->pw_uid)) {
347 ERROR("failed to set user id %d: %m\n", p->pw_uid);
352 char **envp = build_envp(opts.seccomp);
356 INFO("exec-ing %s\n", *opts.jail_argv);
357 execve(*opts.jail_argv, opts.jail_argv, envp);
358 /* we get there only if execve fails */
359 ERROR("failed to execve %s: %m\n", *opts.jail_argv);
363 static int jail_running = 1;
364 static int jail_return_code = 0;
366 static void jail_process_timeout_cb(struct uloop_timeout *t);
367 static struct uloop_timeout jail_process_timeout = {
368 .cb = jail_process_timeout_cb,
371 static void jail_process_handler(struct uloop_process *c, int ret)
373 uloop_timeout_cancel(&jail_process_timeout);
374 if (WIFEXITED(ret)) {
375 jail_return_code = WEXITSTATUS(ret);
376 INFO("jail (%d) exited with exit: %d\n", c->pid, jail_return_code);
378 jail_return_code = WTERMSIG(ret);
379 INFO("jail (%d) exited with signal: %d\n", c->pid, jail_return_code);
385 static struct uloop_process jail_process = {
386 .cb = jail_process_handler,
389 static void jail_process_timeout_cb(struct uloop_timeout *t)
391 DEBUG("jail process failed to stop, sending SIGKILL\n");
392 kill(jail_process.pid, SIGKILL);
395 static void jail_handle_signal(int signo)
397 DEBUG("forwarding signal %d to the jailed process\n", signo);
398 kill(jail_process.pid, signo);
401 static void netns_updown(bool start)
403 struct ubus_context *ctx = ubus_connect(NULL);
404 static struct blob_buf req;
406 pid_t pid = getpid();
411 blob_buf_init(&req, 0);
412 blobmsg_add_string(&req, "jail", opts.name);
413 blobmsg_add_u32(&req, "pid", pid);
414 blobmsg_add_u8(&req, "start", start);
416 if (ubus_lookup_id(ctx, "network", &id) ||
417 ubus_invoke(ctx, id, "netns_updown", req.head, NULL, NULL, 3000))
418 INFO("ubus request failed\n");
424 int main(int argc, char **argv)
427 uid_t uid = getuid();
428 char log[] = "/dev/log";
429 char ubus[] = "/var/run/ubus.sock";
433 ERROR("not root, aborting: %m\n");
439 init_library_search();
441 while ((ch = getopt(argc, argv, OPT_ARGS)) != -1) {
444 debug = atoi(optarg);
447 opts.namespace |= NAMESPACE_MOUNT;
451 opts.namespace |= NAMESPACE_MOUNT;
455 opts.namespace |= NAMESPACE_MOUNT | NAMESPACE_UTS;
456 opts.extroot = optarg;
459 opts.namespace |= NAMESPACE_MOUNT;
463 opts.seccomp = optarg;
464 add_mount(optarg, 1, -1);
467 opts.capabilities = optarg;
470 opts.no_new_privs = 1;
476 opts.namespace |= NAMESPACE_NET;
479 opts.hostname = optarg;
482 opts.namespace |= NAMESPACE_MOUNT;
483 add_path_and_deps(optarg, 1, 0, 0);
486 opts.namespace |= NAMESPACE_MOUNT;
487 add_path_and_deps(optarg, 0, 0, 0);
490 opts.namespace |= NAMESPACE_MOUNT;
491 add_mount(ubus, 0, -1);
494 opts.namespace |= NAMESPACE_MOUNT;
495 add_mount(log, 0, -1);
506 /* no <binary> param found */
507 if (argc - optind < 1) {
511 if (!(opts.namespace||opts.capabilities||opts.seccomp)) {
512 ERROR("Not using namespaces, capabilities or seccomp !!!\n\n");
516 DEBUG("Using namespaces(%d), capabilities(%d), seccomp(%d)\n",
518 opts.capabilities != 0,
521 opts.jail_argv = &argv[optind];
524 if (opts.namespace && add_path_and_deps(*opts.jail_argv, 1, -1, 0)) {
525 ERROR("failed to load dependencies\n");
529 if (opts.namespace && opts.seccomp && add_path_and_deps("libpreload-seccomp.so", 1, -1, 1)) {
530 ERROR("failed to load libpreload-seccomp.so\n");
536 prctl(PR_SET_NAME, opts.name, NULL, NULL, NULL);
540 sigfillset(&sigmask);
541 for (i = 0; i < _NSIG; i++) {
542 struct sigaction s = { 0 };
544 if (!sigismember(&sigmask, i))
546 if ((i == SIGCHLD) || (i == SIGPIPE))
549 s.sa_handler = jail_handle_signal;
550 sigaction(i, &s, NULL);
553 if (opts.namespace) {
554 int flags = SIGCHLD | CLONE_NEWPID | CLONE_NEWIPC;
556 if (opts.namespace & NAMESPACE_MOUNT) {
557 flags |= CLONE_NEWNS;
558 add_mount("/dev/full", 0, -1);
559 add_mount("/dev/null", 0, -1);
560 add_mount("/dev/random", 0, -1);
561 add_mount("/dev/urandom", 0, -1);
562 add_mount("/dev/tty", 0, -1);
563 add_mount("/dev/zero", 0, -1);
564 add_mount("/dev/console", 0, -1);
566 if (opts.user || opts.group) {
567 add_mount("/etc/passwd", 0, -1);
568 add_mount("/etc/group", 0, -1);
571 if (!(opts.namespace & CLONE_NEWNET)) {
572 add_mount("/etc/resolv.conf", 0, -1);
577 flags |= CLONE_NEWUTS;
579 if (opts.namespace & NAMESPACE_NET) {
580 unshare(CLONE_NEWNET);
584 jail_process.pid = clone(exec_jail, child_stack + STACK_SIZE, flags, NULL);
586 jail_process.pid = fork();
589 if (jail_process.pid > 0) {
591 uloop_process_add(&jail_process);
594 DEBUG("uloop interrupted, killing jail process\n");
595 kill(jail_process.pid, SIGTERM);
596 uloop_timeout_set(&jail_process_timeout, 1000);
600 if (opts.namespace & NAMESPACE_NET)
603 return jail_return_code;
604 } else if (jail_process.pid == 0) {
605 /* fork child process */
606 return exec_jail(NULL);
608 ERROR("failed to clone/fork: %m\n");