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:"
67 extern int pivot_root(const char *new_root
, const char *put_old
);
71 static char child_stack
[STACK_SIZE
];
73 static int mkdir_p(char *dir
, mode_t mask
)
75 char *l
= strrchr(dir
, '/');
83 if (mkdir_p(dir
, mask
))
88 ret
= mkdir(dir
, mask
);
89 if (ret
&& errno
== EEXIST
)
93 ERROR("mkdir(%s, %d) failed: %m\n", dir
, mask
);
98 static int _mount_bind(const char *root
, const char *path
, const char *target
, int readonly
, int strict
, int error
)
103 int remount_flags
= MS_BIND
| MS_REMOUNT
;
105 if (stat(path
, &s
)) {
106 ERROR("stat(%s) failed: %m\n", path
);
110 snprintf(new, sizeof(new), "%s%s", root
, target
?target
:path
);
112 if (S_ISDIR(s
.st_mode
)) {
115 mkdir_p(dirname(new), 0755);
116 snprintf(new, sizeof(new), "%s%s", root
, target
?target
:path
);
117 fd
= creat(new, 0644);
119 ERROR("creat(%s) failed: %m\n", new);
125 if (mount(path
, new, NULL
, MS_BIND
, NULL
)) {
126 ERROR("failed to mount -B %s %s: %m\n", path
, new);
131 remount_flags
|= MS_RDONLY
;
134 remount_flags
|= MS_NOEXEC
| MS_NOSUID
| MS_NODEV
;
136 if ((strict
|| readonly
) && mount(NULL
, new, NULL
, remount_flags
, NULL
)) {
137 ERROR("failed to remount (%s%s%s) %s: %m\n", readonly
?"ro":"rw",
138 (readonly
&& strict
)?", ":"", strict
?"strict":"", new);
142 DEBUG("mount -B %s %s (%s%s%s)\n", path
, new,
143 readonly
?"ro":"rw", (readonly
&& strict
)?", ":"", strict
?"strict":"");
148 int mount_bind(const char *root
, const char *path
, int readonly
, int error
) {
149 return _mount_bind(root
, path
, NULL
, readonly
, 0, error
);
152 static int mount_overlay(char *jail_root
, char *overlaydir
) {
153 char *upperdir
, *workdir
, *optsstr
;
154 const char mountoptsformat
[] = "lowerdir=%s,upperdir=%s,workdir=%s";
157 if (asprintf(&upperdir
, "%s%s", overlaydir
, "/upper") < 0)
160 if (asprintf(&workdir
, "%s%s", overlaydir
, "/work") < 0)
163 if (asprintf(&optsstr
, mountoptsformat
, jail_root
, upperdir
, workdir
) < 0)
166 if (mkdir_p(upperdir
, 0755) || mkdir_p(workdir
, 0755))
169 DEBUG("mount -t overlay %s %s (%s)\n", jail_root
, jail_root
, optsstr
);
171 if (mount(jail_root
, jail_root
, "overlay", MS_NOATIME
, optsstr
))
186 static int build_jail_fs(void)
188 char jail_root
[] = "/tmp/ujail-XXXXXX";
189 char tmpovdir
[] = "/tmp/ujail-overlay-XXXXXX";
190 char *overlaydir
= NULL
;
192 if (mkdtemp(jail_root
) == NULL
) {
193 ERROR("mkdtemp(%s) failed: %m\n", jail_root
);
197 /* oldroot can't be MS_SHARED else pivot_root() fails */
198 if (mount("none", "/", NULL
, MS_REC
|MS_PRIVATE
, NULL
)) {
199 ERROR("private mount failed %m\n");
204 if (mount(opts
.extroot
, jail_root
, NULL
, MS_BIND
, NULL
)) {
205 ERROR("extroot mount failed %m\n");
209 if (mount("tmpfs", jail_root
, "tmpfs", MS_NOATIME
, "mode=0755")) {
210 ERROR("tmpfs mount failed %m\n");
215 if (opts
.tmpoverlaysize
) {
216 char mountoptsstr
[] = "mode=0755,size=XXXXXXXX";
218 snprintf(mountoptsstr
, sizeof(mountoptsstr
),
219 "mode=0755,size=%s", opts
.tmpoverlaysize
);
220 if (mkdtemp(tmpovdir
) == NULL
) {
221 ERROR("mkdtemp(%s) failed: %m\n", jail_root
);
224 if (mount("tmpfs", tmpovdir
, "tmpfs", MS_NOATIME
,
226 ERROR("failed to mount tmpfs for overlay (size=%s)\n", opts
.tmpoverlaysize
);
229 overlaydir
= tmpovdir
;
233 overlaydir
= opts
.overlaydir
;
236 mount_overlay(jail_root
, overlaydir
);
238 if (chdir(jail_root
)) {
239 ERROR("chdir(%s) (jail_root) failed: %m\n", jail_root
);
243 if (mount_all(jail_root
)) {
244 ERROR("mount_all() failed\n");
248 if (opts
.namespace & CLONE_NEWNET
) {
249 char hostdir
[PATH_MAX
], jailetc
[PATH_MAX
], jaillink
[PATH_MAX
];
251 snprintf(hostdir
, PATH_MAX
, "/tmp/resolv.conf-%s.d", opts
.name
);
252 mkdir_p(hostdir
, 0755);
253 _mount_bind(jail_root
, hostdir
, "/tmp/resolv.conf.d", 1, 1, -1);
254 snprintf(jailetc
, PATH_MAX
, "%s/etc", jail_root
);
255 mkdir_p(jailetc
, 0755);
256 snprintf(jaillink
, PATH_MAX
, "%s/etc/resolv.conf", jail_root
);
257 symlink("../tmp/resolv.conf.d/resolv.conf.auto", jaillink
);
260 char dirbuf
[sizeof(jail_root
) + 4];
261 snprintf(dirbuf
, sizeof(dirbuf
), "%s/old", jail_root
);
264 if (pivot_root(jail_root
, dirbuf
) == -1) {
265 ERROR("pivot_root(%s, %s) failed: %m\n", jail_root
, dirbuf
);
269 ERROR("chdir(/) (after pivot_root) failed: %m\n");
273 snprintf(dirbuf
, sizeof(dirbuf
), "/old%s", jail_root
);
274 umount2(dirbuf
, MNT_DETACH
);
276 if (opts
.tmpoverlaysize
) {
277 char tmpdirbuf
[sizeof(tmpovdir
) + 4];
278 snprintf(tmpdirbuf
, sizeof(tmpdirbuf
), "/old%s", tmpovdir
);
279 umount2(tmpdirbuf
, MNT_DETACH
);
283 umount2("/old", MNT_DETACH
);
287 mkdir("/proc", 0755);
288 mount("proc", "/proc", "proc", MS_NOATIME
| MS_NODEV
| MS_NOEXEC
| MS_NOSUID
, 0);
290 * make /proc/sys read-only while keeping read-write to
291 * /proc/sys/net if CLONE_NEWNET is set.
293 if (opts
.namespace & CLONE_NEWNET
)
294 mount("/proc/sys/net", "/proc/self/net", NULL
, MS_BIND
, 0);
296 mount("/proc/sys", "/proc/sys", NULL
, MS_BIND
, 0);
297 mount(NULL
, "/proc/sys", NULL
, MS_REMOUNT
| MS_RDONLY
, 0);
298 mount(NULL
, "/proc", NULL
, MS_REMOUNT
, 0);
300 if (opts
.namespace & CLONE_NEWNET
)
301 mount("/proc/self/net", "/proc/sys/net", NULL
, MS_MOVE
, 0);
305 mount("sysfs", "/sys", "sysfs", MS_NOATIME
| MS_NODEV
| MS_NOEXEC
| MS_NOSUID
, 0);
308 mount(NULL
, "/", NULL
, MS_RDONLY
| MS_REMOUNT
, 0);
313 static int write_uid_gid_map(pid_t child_pid
, bool gidmap
, int id
)
317 const char *map_format
= "%d %d %d\n";
318 if (snprintf(map_path
, sizeof(map_path
), "/proc/%d/%s",
319 child_pid
, gidmap
?"gid_map":"uid_map") < 0)
322 if ((map_file
= open(map_path
, O_WRONLY
)) == -1)
325 if (dprintf(map_file
, map_format
, 0, id
, 1) == -1) {
334 static int write_setgroups(pid_t child_pid
, bool allow
)
337 char setgroups_path
[64];
339 if (snprintf(setgroups_path
, sizeof(setgroups_path
), "/proc/%d/setgroups",
344 if ((setgroups_file
= open(setgroups_path
, O_WRONLY
)) == -1) {
348 if (dprintf(setgroups_file
, allow
?"allow":"deny") == -1) {
349 close(setgroups_file
);
353 close(setgroups_file
);
357 static void get_jail_user(int *user
, int *user_gid
, int *gr_gid
)
359 struct passwd
*p
= NULL
;
360 struct group
*g
= NULL
;
363 p
= getpwnam(opts
.user
);
365 ERROR("failed to get uid/gid for user %s: %d (%s)\n",
366 opts
.user
, errno
, strerror(errno
));
370 *user_gid
= p
->pw_gid
;
377 g
= getgrnam(opts
.group
);
379 ERROR("failed to get gid for group %s: %m\n", opts
.group
);
388 static void set_jail_user(int pw_uid
, int user_gid
, int gr_gid
)
390 if ((user_gid
!= -1) && initgroups(opts
.user
, user_gid
)) {
391 ERROR("failed to initgroups() for user %s: %m\n", opts
.user
);
395 if ((gr_gid
!= -1) && setregid(gr_gid
, gr_gid
)) {
396 ERROR("failed to set group id %d: %m\n", gr_gid
);
400 if ((pw_uid
!= -1) && setreuid(pw_uid
, pw_uid
)) {
401 ERROR("failed to set user id %d: %m\n", pw_uid
);
407 static char** build_envp(const char *seccomp
)
409 static char *envp
[MAX_ENVP
];
410 static char preload_var
[PATH_MAX
];
411 static char seccomp_var
[PATH_MAX
];
412 static char debug_var
[] = "LD_DEBUG=all";
413 static char container_var
[] = "container=ujail";
414 const char *preload_lib
= find_lib("libpreload-seccomp.so");
417 if (seccomp
&& !preload_lib
) {
418 ERROR("failed to add preload-lib to env\n");
422 snprintf(seccomp_var
, sizeof(seccomp_var
), "SECCOMP_FILE=%s", seccomp
);
423 envp
[count
++] = seccomp_var
;
424 snprintf(preload_var
, sizeof(preload_var
), "LD_PRELOAD=%s", preload_lib
);
425 envp
[count
++] = preload_var
;
428 envp
[count
++] = container_var
;
431 envp
[count
++] = debug_var
;
436 static void usage(void)
438 fprintf(stderr
, "ujail <options> -- <binary> <params ...>\n");
439 fprintf(stderr
, " -d <num>\tshow debug log (increase num to increase verbosity)\n");
440 fprintf(stderr
, " -S <file>\tseccomp filter config\n");
441 fprintf(stderr
, " -C <file>\tcapabilities drop config\n");
442 fprintf(stderr
, " -c\t\tset PR_SET_NO_NEW_PRIVS\n");
443 fprintf(stderr
, " -n <name>\tthe name of the jail\n");
444 fprintf(stderr
, "namespace jail options:\n");
445 fprintf(stderr
, " -h <hostname>\tchange the hostname of the jail\n");
446 fprintf(stderr
, " -N\t\tjail has network namespace\n");
447 fprintf(stderr
, " -f\t\tjail has user namespace\n");
448 fprintf(stderr
, " -F\t\tjail has cgroups namespace\n");
449 fprintf(stderr
, " -r <file>\treadonly files that should be staged\n");
450 fprintf(stderr
, " -w <file>\twriteable files that should be staged\n");
451 fprintf(stderr
, " -p\t\tjail has /proc\n");
452 fprintf(stderr
, " -s\t\tjail has /sys\n");
453 fprintf(stderr
, " -l\t\tjail has /dev/log\n");
454 fprintf(stderr
, " -u\t\tjail has a ubus socket\n");
455 fprintf(stderr
, " -U <name>\tuser to run jailed process\n");
456 fprintf(stderr
, " -G <name>\tgroup to run jailed process\n");
457 fprintf(stderr
, " -o\t\tremont jail root (/) read only\n");
458 fprintf(stderr
, " -R <dir>\texternal jail rootfs (system container)\n");
459 fprintf(stderr
, " -O <dir>\tdirectory for r/w overlayfs\n");
460 fprintf(stderr
, " -T <size>\tuse tmpfs r/w overlayfs with <size>\n");
461 fprintf(stderr
, "\nWarning: by default root inside the jail is the same\n\
462 and he has the same powers as root outside the jail,\n\
463 thus he can escape the jail and/or break stuff.\n\
464 Please use seccomp/capabilities (-S/-C) to restrict his powers\n\n\
465 If you use none of the namespace jail options,\n\
466 ujail will not use namespace/build a jail,\n\
467 and will only drop capabilities/apply seccomp filter.\n\n");
470 static int exec_jail(void *pipes_ptr
)
472 int *pipes
= (int*)pipes_ptr
;
474 int pw_uid
, pw_gid
, gr_gid
;
481 if (write(pipes
[1], buf
, 1) < 1) {
482 ERROR("can't write to parent\n");
485 if (read(pipes
[2], buf
, 1) < 1) {
486 ERROR("can't read from parent\n");
490 ERROR("parent had an error, child exiting\n");
497 if (opts
.namespace & CLONE_NEWUSER
) {
506 // if (setgroups(0, NULL) < 0) {
507 // ERROR("setgroups\n");
508 // exit(EXIT_FAILURE);
512 if (opts
.namespace && opts
.hostname
&& strlen(opts
.hostname
) > 0
513 && sethostname(opts
.hostname
, strlen(opts
.hostname
))) {
514 ERROR("sethostname(%s) failed: %m\n", opts
.hostname
);
518 if ((opts
.namespace & CLONE_NEWNS
) && build_jail_fs()) {
519 ERROR("failed to build jail fs\n");
523 if (opts
.capabilities
&& drop_capabilities(opts
.capabilities
))
526 if (opts
.no_new_privs
&& prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0)) {
527 ERROR("prctl(PR_SET_NO_NEW_PRIVS) failed: %m\n");
531 if (!(opts
.namespace & CLONE_NEWUSER
)) {
532 get_jail_user(&pw_uid
, &pw_gid
, &gr_gid
);
533 set_jail_user(pw_uid
, pw_gid
, gr_gid
);
536 char **envp
= build_envp(opts
.seccomp
);
540 INFO("exec-ing %s\n", *opts
.jail_argv
);
541 execve(*opts
.jail_argv
, opts
.jail_argv
, envp
);
542 /* we get there only if execve fails */
543 ERROR("failed to execve %s: %m\n", *opts
.jail_argv
);
547 static int jail_running
= 1;
548 static int jail_return_code
= 0;
550 static void jail_process_timeout_cb(struct uloop_timeout
*t
);
551 static struct uloop_timeout jail_process_timeout
= {
552 .cb
= jail_process_timeout_cb
,
555 static void jail_process_handler(struct uloop_process
*c
, int ret
)
557 uloop_timeout_cancel(&jail_process_timeout
);
558 if (WIFEXITED(ret
)) {
559 jail_return_code
= WEXITSTATUS(ret
);
560 INFO("jail (%d) exited with exit: %d\n", c
->pid
, jail_return_code
);
562 jail_return_code
= WTERMSIG(ret
);
563 INFO("jail (%d) exited with signal: %d\n", c
->pid
, jail_return_code
);
569 static struct uloop_process jail_process
= {
570 .cb
= jail_process_handler
,
573 static void jail_process_timeout_cb(struct uloop_timeout
*t
)
575 DEBUG("jail process failed to stop, sending SIGKILL\n");
576 kill(jail_process
.pid
, SIGKILL
);
579 static void jail_handle_signal(int signo
)
581 DEBUG("forwarding signal %d to the jailed process\n", signo
);
582 kill(jail_process
.pid
, signo
);
585 static int netns_open_pid(const pid_t target_ns
)
587 char pid_net_path
[PATH_MAX
];
589 snprintf(pid_net_path
, sizeof(pid_net_path
), "/proc/%u/ns/net", target_ns
);
591 return open(pid_net_path
, O_RDONLY
);
594 static void netns_updown(pid_t pid
, bool start
)
596 struct ubus_context
*ctx
= ubus_connect(NULL
);
597 static struct blob_buf req
;
603 blob_buf_init(&req
, 0);
604 blobmsg_add_string(&req
, "jail", opts
.name
);
605 blobmsg_add_u32(&req
, "pid", pid
);
606 blobmsg_add_u8(&req
, "start", start
);
608 if (ubus_lookup_id(ctx
, "network", &id
) ||
609 ubus_invoke(ctx
, id
, "netns_updown", req
.head
, NULL
, NULL
, 3000))
610 INFO("ubus request failed\n");
616 int main(int argc
, char **argv
)
619 uid_t uid
= getuid();
620 char log
[] = "/dev/log";
621 char ubus
[] = "/var/run/ubus.sock";
628 ERROR("not root, aborting: %m\n");
634 init_library_search();
636 while ((ch
= getopt(argc
, argv
, OPT_ARGS
)) != -1) {
639 debug
= atoi(optarg
);
642 opts
.namespace |= CLONE_NEWNS
;
646 opts
.namespace |= CLONE_NEWNS
;
650 opts
.namespace |= CLONE_NEWUSER
;
653 opts
.namespace |= CLONE_NEWCGROUP
;
656 opts
.extroot
= optarg
;
659 opts
.namespace |= CLONE_NEWNS
;
663 opts
.seccomp
= optarg
;
664 add_mount(optarg
, 1, -1);
667 opts
.capabilities
= optarg
;
670 opts
.no_new_privs
= 1;
676 opts
.namespace |= CLONE_NEWNET
;
679 opts
.namespace |= CLONE_NEWUTS
;
680 opts
.hostname
= optarg
;
683 opts
.namespace |= CLONE_NEWNS
;
684 add_path_and_deps(optarg
, 1, 0, 0);
687 opts
.namespace |= CLONE_NEWNS
;
688 add_path_and_deps(optarg
, 0, 0, 0);
691 opts
.namespace |= CLONE_NEWNS
;
692 add_mount(ubus
, 0, -1);
695 opts
.namespace |= CLONE_NEWNS
;
696 add_mount(log
, 0, -1);
705 opts
.overlaydir
= optarg
;
708 opts
.tmpoverlaysize
= optarg
;
714 opts
.namespace |= CLONE_NEWIPC
| CLONE_NEWPID
;
716 if (opts
.tmpoverlaysize
&& strlen(opts
.tmpoverlaysize
) > 8) {
717 ERROR("size parameter too long: \"%s\"\n", opts
.tmpoverlaysize
);
721 /* no <binary> param found */
722 if (argc
- optind
< 1) {
726 if (!(opts
.namespace||opts
.capabilities
||opts
.seccomp
)) {
727 ERROR("Not using namespaces, capabilities or seccomp !!!\n\n");
731 DEBUG("Using namespaces(0x%08x), capabilities(%d), seccomp(%d)\n",
733 opts
.capabilities
!= 0,
736 opts
.jail_argv
= &argv
[optind
];
738 get_jail_user(&opts
.pw_uid
, &opts
.pw_gid
, &opts
.gr_gid
);
741 if (opts
.namespace && add_path_and_deps(*opts
.jail_argv
, 1, -1, 0)) {
742 ERROR("failed to load dependencies\n");
747 if (opts
.namespace && opts
.seccomp
&& add_path_and_deps("libpreload-seccomp.so", 1, -1, 1)) {
748 ERROR("failed to load libpreload-seccomp.so\n");
753 prctl(PR_SET_NAME
, opts
.name
, NULL
, NULL
, NULL
);
757 sigfillset(&sigmask
);
758 for (i
= 0; i
< _NSIG
; i
++) {
759 struct sigaction s
= { 0 };
761 if (!sigismember(&sigmask
, i
))
763 if ((i
== SIGCHLD
) || (i
== SIGPIPE
))
766 s
.sa_handler
= jail_handle_signal
;
767 sigaction(i
, &s
, NULL
);
770 if (opts
.namespace) {
771 if (opts
.namespace & CLONE_NEWNS
) {
772 add_mount("/dev/full", 0, -1);
773 add_mount("/dev/null", 0, -1);
774 add_mount("/dev/random", 0, -1);
775 add_mount("/dev/urandom", 0, -1);
776 add_mount("/dev/tty", 0, -1);
777 add_mount("/dev/zero", 0, -1);
778 add_mount("/dev/console", 0, -1);
780 if (!opts
.extroot
&& (opts
.user
|| opts
.group
)) {
781 add_mount("/etc/passwd", 0, -1);
782 add_mount("/etc/group", 0, -1);
785 if (!(opts
.namespace & CLONE_NEWNET
)) {
786 add_mount("/etc/resolv.conf", 0, -1);
790 if (pipe(&pipes
[0]) < 0 || pipe(&pipes
[2]) < 0)
793 jail_process
.pid
= clone(exec_jail
, child_stack
+ STACK_SIZE
, SIGCHLD
| opts
.namespace, &pipes
);
795 jail_process
.pid
= fork();
798 if (jail_process
.pid
> 0) {
803 if (read(pipes
[0], sig_buf
, 1) < 1) {
804 ERROR("can't read from child\n");
808 if (opts
.namespace & CLONE_NEWUSER
) {
809 bool has_gr
= (opts
.gr_gid
!= -1);
810 if (write_setgroups(jail_process
.pid
, false)) {
811 ERROR("can't write setgroups\n");
814 if (opts
.pw_uid
!= -1) {
815 write_uid_gid_map(jail_process
.pid
, 0, opts
.pw_uid
);
816 write_uid_gid_map(jail_process
.pid
, 1, has_gr
?opts
.gr_gid
:opts
.pw_gid
);
818 write_uid_gid_map(jail_process
.pid
, 0, 65534);
819 write_uid_gid_map(jail_process
.pid
, 1, has_gr
?opts
.gr_gid
:65534);
823 if (opts
.namespace & CLONE_NEWNET
) {
824 netns_fd
= netns_open_pid(jail_process
.pid
);
825 netns_updown(jail_process
.pid
, true);
829 if (write(pipes
[3], sig_buf
, 1) < 0) {
830 ERROR("can't write to child\n");
834 uloop_process_add(&jail_process
);
837 DEBUG("uloop interrupted, killing jail process\n");
838 kill(jail_process
.pid
, SIGTERM
);
839 uloop_timeout_set(&jail_process_timeout
, 1000);
843 if (opts
.namespace & CLONE_NEWNET
) {
844 setns(netns_fd
, CLONE_NEWNET
);
845 netns_updown(getpid(), false);
848 return jail_return_code
;
849 } else if (jail_process
.pid
== 0) {
850 /* fork child process */
851 return exec_jail(NULL
);
853 ERROR("failed to clone/fork: %m\n");