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 tmpdevdir
[] = "/tmp/ujail-XXXXXX/dev";
191 char *overlaydir
= NULL
;
193 if (mkdtemp(jail_root
) == NULL
) {
194 ERROR("mkdtemp(%s) failed: %m\n", jail_root
);
198 /* oldroot can't be MS_SHARED else pivot_root() fails */
199 if (mount("none", "/", NULL
, MS_REC
|MS_PRIVATE
, NULL
)) {
200 ERROR("private mount failed %m\n");
205 if (mount(opts
.extroot
, jail_root
, NULL
, MS_BIND
, NULL
)) {
206 ERROR("extroot mount failed %m\n");
210 if (mount("tmpfs", jail_root
, "tmpfs", MS_NOATIME
, "mode=0755")) {
211 ERROR("tmpfs mount failed %m\n");
216 if (opts
.tmpoverlaysize
) {
217 char mountoptsstr
[] = "mode=0755,size=XXXXXXXX";
219 snprintf(mountoptsstr
, sizeof(mountoptsstr
),
220 "mode=0755,size=%s", opts
.tmpoverlaysize
);
221 if (mkdtemp(tmpovdir
) == NULL
) {
222 ERROR("mkdtemp(%s) failed: %m\n", jail_root
);
225 if (mount("tmpfs", tmpovdir
, "tmpfs", MS_NOATIME
,
227 ERROR("failed to mount tmpfs for overlay (size=%s)\n", opts
.tmpoverlaysize
);
230 overlaydir
= tmpovdir
;
234 overlaydir
= opts
.overlaydir
;
237 mount_overlay(jail_root
, overlaydir
);
239 if (chdir(jail_root
)) {
240 ERROR("chdir(%s) (jail_root) failed: %m\n", jail_root
);
244 snprintf(tmpdevdir
, sizeof(tmpdevdir
), "%s/dev", jail_root
);
245 mkdir_p(tmpdevdir
, 0755);
246 if (mount(NULL
, tmpdevdir
, "tmpfs", MS_NOATIME
| MS_NOEXEC
| MS_NOSUID
, "size=1M"))
249 if (mount_all(jail_root
)) {
250 ERROR("mount_all() failed\n");
254 if (opts
.namespace & CLONE_NEWNET
) {
255 char hostdir
[PATH_MAX
], jailetc
[PATH_MAX
], jaillink
[PATH_MAX
];
257 snprintf(hostdir
, PATH_MAX
, "/tmp/resolv.conf-%s.d", opts
.name
);
258 mkdir_p(hostdir
, 0755);
259 _mount_bind(jail_root
, hostdir
, "/tmp/resolv.conf.d", 1, 1, -1);
260 snprintf(jailetc
, PATH_MAX
, "%s/etc", jail_root
);
261 mkdir_p(jailetc
, 0755);
262 snprintf(jaillink
, PATH_MAX
, "%s/etc/resolv.conf", jail_root
);
265 symlink("../tmp/resolv.conf.d/resolv.conf.auto", jaillink
);
268 char dirbuf
[sizeof(jail_root
) + 4];
269 snprintf(dirbuf
, sizeof(dirbuf
), "%s/old", jail_root
);
272 if (pivot_root(jail_root
, dirbuf
) == -1) {
273 ERROR("pivot_root(%s, %s) failed: %m\n", jail_root
, dirbuf
);
277 ERROR("chdir(/) (after pivot_root) failed: %m\n");
281 snprintf(dirbuf
, sizeof(dirbuf
), "/old%s", jail_root
);
282 umount2(dirbuf
, MNT_DETACH
);
284 if (opts
.tmpoverlaysize
) {
285 char tmpdirbuf
[sizeof(tmpovdir
) + 4];
286 snprintf(tmpdirbuf
, sizeof(tmpdirbuf
), "/old%s", tmpovdir
);
287 umount2(tmpdirbuf
, MNT_DETACH
);
291 umount2("/old", MNT_DETACH
);
295 mkdir("/proc", 0755);
296 mount("proc", "/proc", "proc", MS_NOATIME
| MS_NODEV
| MS_NOEXEC
| MS_NOSUID
, 0);
298 * make /proc/sys read-only while keeping read-write to
299 * /proc/sys/net if CLONE_NEWNET is set.
301 if (opts
.namespace & CLONE_NEWNET
)
302 mount("/proc/sys/net", "/proc/self/net", NULL
, MS_BIND
, 0);
304 mount("/proc/sys", "/proc/sys", NULL
, MS_BIND
, 0);
305 mount(NULL
, "/proc/sys", NULL
, MS_REMOUNT
| MS_RDONLY
, 0);
306 mount(NULL
, "/proc", NULL
, MS_REMOUNT
, 0);
308 if (opts
.namespace & CLONE_NEWNET
)
309 mount("/proc/self/net", "/proc/sys/net", NULL
, MS_MOVE
, 0);
313 mount("sysfs", "/sys", "sysfs", MS_NOATIME
| MS_NODEV
| MS_NOEXEC
| MS_NOSUID
| MS_RDONLY
, 0);
316 mount(NULL
, "/", NULL
, MS_RDONLY
| MS_REMOUNT
, 0);
321 static int write_uid_gid_map(pid_t child_pid
, bool gidmap
, int id
)
325 const char *map_format
= "%d %d %d\n";
326 if (snprintf(map_path
, sizeof(map_path
), "/proc/%d/%s",
327 child_pid
, gidmap
?"gid_map":"uid_map") < 0)
330 if ((map_file
= open(map_path
, O_WRONLY
)) == -1)
333 if (dprintf(map_file
, map_format
, 0, id
, 1) == -1) {
342 static int write_setgroups(pid_t child_pid
, bool allow
)
345 char setgroups_path
[64];
347 if (snprintf(setgroups_path
, sizeof(setgroups_path
), "/proc/%d/setgroups",
352 if ((setgroups_file
= open(setgroups_path
, O_WRONLY
)) == -1) {
356 if (dprintf(setgroups_file
, allow
?"allow":"deny") == -1) {
357 close(setgroups_file
);
361 close(setgroups_file
);
365 static void get_jail_user(int *user
, int *user_gid
, int *gr_gid
)
367 struct passwd
*p
= NULL
;
368 struct group
*g
= NULL
;
371 p
= getpwnam(opts
.user
);
373 ERROR("failed to get uid/gid for user %s: %d (%s)\n",
374 opts
.user
, errno
, strerror(errno
));
378 *user_gid
= p
->pw_gid
;
385 g
= getgrnam(opts
.group
);
387 ERROR("failed to get gid for group %s: %m\n", opts
.group
);
396 static void set_jail_user(int pw_uid
, int user_gid
, int gr_gid
)
398 if ((user_gid
!= -1) && initgroups(opts
.user
, user_gid
)) {
399 ERROR("failed to initgroups() for user %s: %m\n", opts
.user
);
403 if ((gr_gid
!= -1) && setregid(gr_gid
, gr_gid
)) {
404 ERROR("failed to set group id %d: %m\n", gr_gid
);
408 if ((pw_uid
!= -1) && setreuid(pw_uid
, pw_uid
)) {
409 ERROR("failed to set user id %d: %m\n", pw_uid
);
415 static char** build_envp(const char *seccomp
)
417 static char *envp
[MAX_ENVP
];
418 static char preload_var
[PATH_MAX
];
419 static char seccomp_var
[PATH_MAX
];
420 static char debug_var
[] = "LD_DEBUG=all";
421 static char container_var
[] = "container=ujail";
422 const char *preload_lib
= find_lib("libpreload-seccomp.so");
425 if (seccomp
&& !preload_lib
) {
426 ERROR("failed to add preload-lib to env\n");
430 snprintf(seccomp_var
, sizeof(seccomp_var
), "SECCOMP_FILE=%s", seccomp
);
431 envp
[count
++] = seccomp_var
;
432 snprintf(preload_var
, sizeof(preload_var
), "LD_PRELOAD=%s", preload_lib
);
433 envp
[count
++] = preload_var
;
436 envp
[count
++] = container_var
;
439 envp
[count
++] = debug_var
;
444 static void usage(void)
446 fprintf(stderr
, "ujail <options> -- <binary> <params ...>\n");
447 fprintf(stderr
, " -d <num>\tshow debug log (increase num to increase verbosity)\n");
448 fprintf(stderr
, " -S <file>\tseccomp filter config\n");
449 fprintf(stderr
, " -C <file>\tcapabilities drop config\n");
450 fprintf(stderr
, " -c\t\tset PR_SET_NO_NEW_PRIVS\n");
451 fprintf(stderr
, " -n <name>\tthe name of the jail\n");
452 fprintf(stderr
, "namespace jail options:\n");
453 fprintf(stderr
, " -h <hostname>\tchange the hostname of the jail\n");
454 fprintf(stderr
, " -N\t\tjail has network namespace\n");
455 fprintf(stderr
, " -f\t\tjail has user namespace\n");
456 fprintf(stderr
, " -F\t\tjail has cgroups namespace\n");
457 fprintf(stderr
, " -r <file>\treadonly files that should be staged\n");
458 fprintf(stderr
, " -w <file>\twriteable files that should be staged\n");
459 fprintf(stderr
, " -p\t\tjail has /proc\n");
460 fprintf(stderr
, " -s\t\tjail has /sys\n");
461 fprintf(stderr
, " -l\t\tjail has /dev/log\n");
462 fprintf(stderr
, " -u\t\tjail has a ubus socket\n");
463 fprintf(stderr
, " -U <name>\tuser to run jailed process\n");
464 fprintf(stderr
, " -G <name>\tgroup to run jailed process\n");
465 fprintf(stderr
, " -o\t\tremont jail root (/) read only\n");
466 fprintf(stderr
, " -R <dir>\texternal jail rootfs (system container)\n");
467 fprintf(stderr
, " -O <dir>\tdirectory for r/w overlayfs\n");
468 fprintf(stderr
, " -T <size>\tuse tmpfs r/w overlayfs with <size>\n");
469 fprintf(stderr
, "\nWarning: by default root inside the jail is the same\n\
470 and he has the same powers as root outside the jail,\n\
471 thus he can escape the jail and/or break stuff.\n\
472 Please use seccomp/capabilities (-S/-C) to restrict his powers\n\n\
473 If you use none of the namespace jail options,\n\
474 ujail will not use namespace/build a jail,\n\
475 and will only drop capabilities/apply seccomp filter.\n\n");
478 static int exec_jail(void *pipes_ptr
)
480 int *pipes
= (int*)pipes_ptr
;
482 int pw_uid
, pw_gid
, gr_gid
;
489 if (write(pipes
[1], buf
, 1) < 1) {
490 ERROR("can't write to parent\n");
493 if (read(pipes
[2], buf
, 1) < 1) {
494 ERROR("can't read from parent\n");
498 ERROR("parent had an error, child exiting\n");
505 if (opts
.namespace & CLONE_NEWUSER
) {
514 // if (setgroups(0, NULL) < 0) {
515 // ERROR("setgroups\n");
516 // exit(EXIT_FAILURE);
520 if (opts
.namespace && opts
.hostname
&& strlen(opts
.hostname
) > 0
521 && sethostname(opts
.hostname
, strlen(opts
.hostname
))) {
522 ERROR("sethostname(%s) failed: %m\n", opts
.hostname
);
526 if ((opts
.namespace & CLONE_NEWNS
) && build_jail_fs()) {
527 ERROR("failed to build jail fs\n");
531 if (opts
.capabilities
&& drop_capabilities(opts
.capabilities
))
534 if (opts
.no_new_privs
&& prctl(PR_SET_NO_NEW_PRIVS
, 1, 0, 0, 0)) {
535 ERROR("prctl(PR_SET_NO_NEW_PRIVS) failed: %m\n");
539 if (!(opts
.namespace & CLONE_NEWUSER
)) {
540 get_jail_user(&pw_uid
, &pw_gid
, &gr_gid
);
541 set_jail_user(pw_uid
, pw_gid
, gr_gid
);
544 char **envp
= build_envp(opts
.seccomp
);
548 INFO("exec-ing %s\n", *opts
.jail_argv
);
549 execve(*opts
.jail_argv
, opts
.jail_argv
, envp
);
550 /* we get there only if execve fails */
551 ERROR("failed to execve %s: %m\n", *opts
.jail_argv
);
555 static int jail_running
= 1;
556 static int jail_return_code
= 0;
558 static void jail_process_timeout_cb(struct uloop_timeout
*t
);
559 static struct uloop_timeout jail_process_timeout
= {
560 .cb
= jail_process_timeout_cb
,
563 static void jail_process_handler(struct uloop_process
*c
, int ret
)
565 uloop_timeout_cancel(&jail_process_timeout
);
566 if (WIFEXITED(ret
)) {
567 jail_return_code
= WEXITSTATUS(ret
);
568 INFO("jail (%d) exited with exit: %d\n", c
->pid
, jail_return_code
);
570 jail_return_code
= WTERMSIG(ret
);
571 INFO("jail (%d) exited with signal: %d\n", c
->pid
, jail_return_code
);
577 static struct uloop_process jail_process
= {
578 .cb
= jail_process_handler
,
581 static void jail_process_timeout_cb(struct uloop_timeout
*t
)
583 DEBUG("jail process failed to stop, sending SIGKILL\n");
584 kill(jail_process
.pid
, SIGKILL
);
587 static void jail_handle_signal(int signo
)
589 DEBUG("forwarding signal %d to the jailed process\n", signo
);
590 kill(jail_process
.pid
, signo
);
593 static int netns_open_pid(const pid_t target_ns
)
595 char pid_net_path
[PATH_MAX
];
597 snprintf(pid_net_path
, sizeof(pid_net_path
), "/proc/%u/ns/net", target_ns
);
599 return open(pid_net_path
, O_RDONLY
);
602 static void netns_updown(pid_t pid
, bool start
)
604 struct ubus_context
*ctx
= ubus_connect(NULL
);
605 static struct blob_buf req
;
611 blob_buf_init(&req
, 0);
612 blobmsg_add_string(&req
, "jail", opts
.name
);
613 blobmsg_add_u32(&req
, "pid", pid
);
614 blobmsg_add_u8(&req
, "start", start
);
616 if (ubus_lookup_id(ctx
, "network", &id
) ||
617 ubus_invoke(ctx
, id
, "netns_updown", req
.head
, NULL
, NULL
, 3000))
618 INFO("ubus request failed\n");
624 int main(int argc
, char **argv
)
627 uid_t uid
= getuid();
628 char log
[] = "/dev/log";
629 char ubus
[] = "/var/run/ubus.sock";
636 ERROR("not root, aborting: %m\n");
642 init_library_search();
644 while ((ch
= getopt(argc
, argv
, OPT_ARGS
)) != -1) {
647 debug
= atoi(optarg
);
650 opts
.namespace |= CLONE_NEWNS
;
654 opts
.namespace |= CLONE_NEWNS
;
658 opts
.namespace |= CLONE_NEWUSER
;
661 opts
.namespace |= CLONE_NEWCGROUP
;
664 opts
.extroot
= optarg
;
667 opts
.namespace |= CLONE_NEWNS
;
671 opts
.seccomp
= optarg
;
672 add_mount(optarg
, 1, -1);
675 opts
.capabilities
= optarg
;
678 opts
.no_new_privs
= 1;
684 opts
.namespace |= CLONE_NEWNET
;
687 opts
.namespace |= CLONE_NEWUTS
;
688 opts
.hostname
= optarg
;
691 opts
.namespace |= CLONE_NEWNS
;
692 add_path_and_deps(optarg
, 1, 0, 0);
695 opts
.namespace |= CLONE_NEWNS
;
696 add_path_and_deps(optarg
, 0, 0, 0);
699 opts
.namespace |= CLONE_NEWNS
;
700 add_mount(ubus
, 0, -1);
703 opts
.namespace |= CLONE_NEWNS
;
704 add_mount(log
, 0, -1);
713 opts
.overlaydir
= optarg
;
716 opts
.tmpoverlaysize
= optarg
;
722 opts
.namespace |= CLONE_NEWIPC
| CLONE_NEWPID
;
724 if (opts
.tmpoverlaysize
&& strlen(opts
.tmpoverlaysize
) > 8) {
725 ERROR("size parameter too long: \"%s\"\n", opts
.tmpoverlaysize
);
729 /* no <binary> param found */
730 if (argc
- optind
< 1) {
734 if (!(opts
.namespace||opts
.capabilities
||opts
.seccomp
)) {
735 ERROR("Not using namespaces, capabilities or seccomp !!!\n\n");
739 DEBUG("Using namespaces(0x%08x), capabilities(%d), seccomp(%d)\n",
741 opts
.capabilities
!= 0,
744 opts
.jail_argv
= &argv
[optind
];
746 get_jail_user(&opts
.pw_uid
, &opts
.pw_gid
, &opts
.gr_gid
);
749 if (opts
.namespace && add_path_and_deps(*opts
.jail_argv
, 1, -1, 0)) {
750 ERROR("failed to load dependencies\n");
755 if (opts
.namespace && opts
.seccomp
&& add_path_and_deps("libpreload-seccomp.so", 1, -1, 1)) {
756 ERROR("failed to load libpreload-seccomp.so\n");
761 prctl(PR_SET_NAME
, opts
.name
, NULL
, NULL
, NULL
);
765 sigfillset(&sigmask
);
766 for (i
= 0; i
< _NSIG
; i
++) {
767 struct sigaction s
= { 0 };
769 if (!sigismember(&sigmask
, i
))
771 if ((i
== SIGCHLD
) || (i
== SIGPIPE
))
774 s
.sa_handler
= jail_handle_signal
;
775 sigaction(i
, &s
, NULL
);
778 if (opts
.namespace) {
779 if (opts
.namespace & CLONE_NEWNS
) {
780 add_mount("/dev/full", 0, -1);
781 add_mount("/dev/null", 0, -1);
782 add_mount("/dev/random", 0, -1);
783 add_mount("/dev/urandom", 0, -1);
784 add_mount("/dev/tty", 0, -1);
785 add_mount("/dev/zero", 0, -1);
786 add_mount("/dev/console", 0, -1);
788 if (!opts
.extroot
&& (opts
.user
|| opts
.group
)) {
789 add_mount("/etc/passwd", 0, -1);
790 add_mount("/etc/group", 0, -1);
793 #if defined(__GLIBC__)
795 add_mount("/etc/nsswitch.conf", 0, -1);
798 if (!(opts
.namespace & CLONE_NEWNET
)) {
799 add_mount("/etc/resolv.conf", 0, -1);
803 if (pipe(&pipes
[0]) < 0 || pipe(&pipes
[2]) < 0)
806 jail_process
.pid
= clone(exec_jail
, child_stack
+ STACK_SIZE
, SIGCHLD
| opts
.namespace, &pipes
);
808 jail_process
.pid
= fork();
811 if (jail_process
.pid
> 0) {
816 if (read(pipes
[0], sig_buf
, 1) < 1) {
817 ERROR("can't read from child\n");
821 if (opts
.namespace & CLONE_NEWUSER
) {
822 bool has_gr
= (opts
.gr_gid
!= -1);
823 if (write_setgroups(jail_process
.pid
, false)) {
824 ERROR("can't write setgroups\n");
827 if (opts
.pw_uid
!= -1) {
828 write_uid_gid_map(jail_process
.pid
, 0, opts
.pw_uid
);
829 write_uid_gid_map(jail_process
.pid
, 1, has_gr
?opts
.gr_gid
:opts
.pw_gid
);
831 write_uid_gid_map(jail_process
.pid
, 0, 65534);
832 write_uid_gid_map(jail_process
.pid
, 1, has_gr
?opts
.gr_gid
:65534);
836 if (opts
.namespace & CLONE_NEWNET
) {
837 netns_fd
= netns_open_pid(jail_process
.pid
);
838 netns_updown(jail_process
.pid
, true);
842 if (write(pipes
[3], sig_buf
, 1) < 0) {
843 ERROR("can't write to child\n");
847 uloop_process_add(&jail_process
);
850 DEBUG("uloop interrupted, killing jail process\n");
851 kill(jail_process
.pid
, SIGTERM
);
852 uloop_timeout_set(&jail_process_timeout
, 1000);
856 if (opts
.namespace & CLONE_NEWNET
) {
857 setns(netns_fd
, CLONE_NEWNET
);
858 netns_updown(getpid(), false);
861 return jail_return_code
;
862 } else if (jail_process
.pid
== 0) {
863 /* fork child process */
864 return exec_jail(NULL
);
866 ERROR("failed to clone/fork: %m\n");