jail/jail.c

   1 /*
   2  * Copyright (C) 2015 John Crispin <blogic@openwrt.org>
   3  *
   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
   7  *
   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.
  12  */
  13
  14 #define _GNU_SOURCE
  15 #include <sys/mount.h>
  16 #include <sys/prctl.h>
  17 #include <sys/wait.h>
  18 #include <sys/types.h>
  19
  20 #include <stdlib.h>
  21 #include <unistd.h>
  22 #include <errno.h>
  23 #include <pwd.h>
  24 #include <grp.h>
  25 #include <string.h>
  26 #include <sys/stat.h>
  27 #include <fcntl.h>
  28 #include <libgen.h>
  29 #include <sched.h>
  30 #include <linux/limits.h>
  31 #include <signal.h>
  32
  33 #include "capabilities.h"
  34 #include "elf.h"
  35 #include "fs.h"
  36 #include "jail.h"
  37 #include "log.h"
  38
  39 #include <libubox/uloop.h>
  40 #include <libubus.h>
  41
  42 #define STACK_SIZE      (1024 * 1024)
  43 #define OPT_ARGS        "S:C:n:h:r:w:d:psulocU:G:NR:fFO:T:E"
  44
  45 static struct {
  46         char *name;
  47         char *hostname;
  48         char **jail_argv;
  49         char *seccomp;
  50         char *capabilities;
  51         char *user;
  52         char *group;
  53         char *extroot;
  54         char *overlaydir;
  55         char *tmpoverlaysize;
  56         int no_new_privs;
  57         int namespace;
  58         int procfs;
  59         int ronly;
  60         int sysfs;
  61         int pw_uid;
  62         int pw_gid;
  63         int gr_gid;
  64         int require_jail;
  65 } opts;
  66
  67
  68 extern int pivot_root(const char *new_root, const char *put_old);
  69
  70 int debug = 0;
  71
  72 static char child_stack[STACK_SIZE];
  73
  74 static int mkdir_p(char *dir, mode_t mask)
  75 {
  76         char *l = strrchr(dir, '/');
  77         int ret;
  78
  79         if (!l)
  80                 return 0;
  81
  82         *l = '\0';
  83
  84         if (mkdir_p(dir, mask))
  85                 return -1;
  86
  87         *l = '/';
  88
  89         ret = mkdir(dir, mask);
  90         if (ret && errno == EEXIST)
  91                 return 0;
  92
  93         if (ret)
  94                 ERROR("mkdir(%s, %d) failed: %m\n", dir, mask);
  95
  96         return ret;
  97 }
  98
  99 static int _mount_bind(const char *root, const char *path, const char *target, int readonly, int strict, int error)
 100 {
 101         struct stat s;
 102         char new[PATH_MAX];
 103         int fd;
 104         int remount_flags = MS_BIND | MS_REMOUNT;
 105
 106         if (stat(path, &s)) {
 107                 ERROR("stat(%s) failed: %m\n", path);
 108                 return error;
 109         }
 110
 111         snprintf(new, sizeof(new), "%s%s", root, target?target:path);
 112
 113         if (S_ISDIR(s.st_mode)) {
 114                 mkdir_p(new, 0755);
 115         } else {
 116                 mkdir_p(dirname(new), 0755);
 117                 snprintf(new, sizeof(new), "%s%s", root, target?target:path);
 118                 fd = creat(new, 0644);
 119                 if (fd == -1) {
 120                         ERROR("creat(%s) failed: %m\n", new);
 121                         return -1;
 122                 }
 123                 close(fd);
 124         }
 125
 126         if (mount(path, new, NULL, MS_BIND, NULL)) {
 127                 ERROR("failed to mount -B %s %s: %m\n", path, new);
 128                 return -1;
 129         }
 130
 131         if (readonly)
 132                 remount_flags |= MS_RDONLY;
 133
 134         if (strict)
 135                 remount_flags |= MS_NOEXEC | MS_NOSUID | MS_NODEV;
 136
 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);
 140                 return -1;
 141         }
 142
 143         DEBUG("mount -B %s %s (%s%s%s)\n", path, new,
 144               readonly?"ro":"rw", (readonly && strict)?", ":"", strict?"strict":"");
 145
 146         return 0;
 147 }
 148
 149 int mount_bind(const char *root, const char *path, int readonly, int error) {
 150         return _mount_bind(root, path, NULL, readonly, 0, error);
 151 }
 152
 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";
 156         int ret = -1;
 157
 158         if (asprintf(&upperdir, "%s%s", overlaydir, "/upper") < 0)
 159                 goto out;
 160
 161         if (asprintf(&workdir, "%s%s", overlaydir, "/work") < 0)
 162                 goto upper_printf;
 163
 164         if (asprintf(&optsstr, mountoptsformat, jail_root, upperdir, workdir) < 0)
 165                 goto work_printf;
 166
 167         if (mkdir_p(upperdir, 0755) || mkdir_p(workdir, 0755))
 168                 goto opts_printf;
 169
 170         DEBUG("mount -t overlay %s %s (%s)\n", jail_root, jail_root, optsstr);
 171
 172         if (mount(jail_root, jail_root, "overlay", MS_NOATIME, optsstr))
 173                 goto opts_printf;
 174
 175         ret = 0;
 176
 177 opts_printf:
 178         free(optsstr);
 179 work_printf:
 180         free(workdir);
 181 upper_printf:
 182         free(upperdir);
 183 out:
 184         return ret;
 185 }
 186
 187 static int build_jail_fs(void)
 188 {
 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;
 193
 194         if (mkdtemp(jail_root) == NULL) {
 195                 ERROR("mkdtemp(%s) failed: %m\n", jail_root);
 196                 return -1;
 197         }
 198
 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");
 202                 return -1;
 203         }
 204
 205         if (opts.extroot) {
 206                 if (mount(opts.extroot, jail_root, NULL, MS_BIND, NULL)) {
 207                         ERROR("extroot mount failed %m\n");
 208                         return -1;
 209                 }
 210         } else {
 211                 if (mount("tmpfs", jail_root, "tmpfs", MS_NOATIME, "mode=0755")) {
 212                         ERROR("tmpfs mount failed %m\n");
 213                         return -1;
 214                 }
 215         }
 216
 217         if (opts.tmpoverlaysize) {
 218                 char mountoptsstr[] = "mode=0755,size=XXXXXXXX";
 219
 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);
 224                         return -1;
 225                 }
 226                 if (mount("tmpfs", tmpovdir, "tmpfs", MS_NOATIME,
 227                           mountoptsstr)) {
 228                         ERROR("failed to mount tmpfs for overlay (size=%s)\n", opts.tmpoverlaysize);
 229                         return -1;
 230                 }
 231                 overlaydir = tmpovdir;
 232         }
 233
 234         if (opts.overlaydir)
 235                 overlaydir = opts.overlaydir;
 236
 237         if (overlaydir)
 238                 mount_overlay(jail_root, overlaydir);
 239
 240         if (chdir(jail_root)) {
 241                 ERROR("chdir(%s) (jail_root) failed: %m\n", jail_root);
 242                 return -1;
 243         }
 244
 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"))
 248                 return -1;
 249
 250         if (mount_all(jail_root)) {
 251                 ERROR("mount_all() failed\n");
 252                 return -1;
 253         }
 254
 255         if (opts.namespace & CLONE_NEWNET) {
 256                 char hostdir[PATH_MAX], jailetc[PATH_MAX], jaillink[PATH_MAX];
 257
 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);
 264                 if (overlaydir)
 265                         unlink(jaillink);
 266                 symlink("../tmp/resolv.conf.d/resolv.conf.auto", jaillink);
 267         }
 268
 269         char dirbuf[sizeof(jail_root) + 4];
 270         snprintf(dirbuf, sizeof(dirbuf), "%s/old", jail_root);
 271         mkdir(dirbuf, 0755);
 272
 273         if (pivot_root(jail_root, dirbuf) == -1) {
 274                 ERROR("pivot_root(%s, %s) failed: %m\n", jail_root, dirbuf);
 275                 return -1;
 276         }
 277         if (chdir("/")) {
 278                 ERROR("chdir(/) (after pivot_root) failed: %m\n");
 279                 return -1;
 280         }
 281
 282         snprintf(dirbuf, sizeof(dirbuf), "/old%s", jail_root);
 283         umount2(dirbuf, MNT_DETACH);
 284         rmdir(dirbuf);
 285         if (opts.tmpoverlaysize) {
 286                 char tmpdirbuf[sizeof(tmpovdir) + 4];
 287                 snprintf(tmpdirbuf, sizeof(tmpdirbuf), "/old%s", tmpovdir);
 288                 umount2(tmpdirbuf, MNT_DETACH);
 289                 rmdir(tmpdirbuf);
 290         }
 291
 292         umount2("/old", MNT_DETACH);
 293         rmdir("/old");
 294
 295         if (opts.procfs) {
 296                 mkdir("/proc", 0755);
 297                 mount("proc", "/proc", "proc", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
 298                 /*
 299                  * make /proc/sys read-only while keeping read-write to
 300                  * /proc/sys/net if CLONE_NEWNET is set.
 301                  */
 302                 if (opts.namespace & CLONE_NEWNET)
 303                         mount("/proc/sys/net", "/proc/self/net", NULL, MS_BIND, 0);
 304
 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);
 308
 309                 if (opts.namespace & CLONE_NEWNET)
 310                         mount("/proc/self/net", "/proc/sys/net", NULL, MS_MOVE, 0);
 311         }
 312         if (opts.sysfs) {
 313                 mkdir("/sys", 0755);
 314                 mount("sysfs", "/sys", "sysfs", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID | MS_RDONLY, 0);
 315         }
 316         if (opts.ronly)
 317                 mount(NULL, "/", NULL, MS_RDONLY | MS_REMOUNT, 0);
 318
 319         return 0;
 320 }
 321
 322 static int write_uid_gid_map(pid_t child_pid, bool gidmap, int id)
 323 {
 324         int map_file;
 325         char map_path[64];
 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)
 329                 return -1;
 330
 331         if ((map_file = open(map_path, O_WRONLY)) == -1)
 332                 return -1;
 333
 334         if (dprintf(map_file, map_format, 0, id, 1) == -1) {
 335                 close(map_file);
 336                 return -1;
 337         }
 338
 339         close(map_file);
 340         return 0;
 341 }
 342
 343 static int write_setgroups(pid_t child_pid, bool allow)
 344 {
 345         int setgroups_file;
 346         char setgroups_path[64];
 347
 348         if (snprintf(setgroups_path, sizeof(setgroups_path), "/proc/%d/setgroups",
 349                 child_pid) < 0) {
 350                 return -1;
 351         }
 352
 353         if ((setgroups_file = open(setgroups_path, O_WRONLY)) == -1) {
 354                 return -1;
 355         }
 356
 357         if (dprintf(setgroups_file, allow?"allow":"deny") == -1) {
 358                 close(setgroups_file);
 359                 return -1;
 360         }
 361
 362         close(setgroups_file);
 363         return 0;
 364 }
 365
 366 static void get_jail_user(int *user, int *user_gid, int *gr_gid)
 367 {
 368         struct passwd *p = NULL;
 369         struct group *g = NULL;
 370
 371         if (opts.user) {
 372                 p = getpwnam(opts.user);
 373                 if (!p) {
 374                         ERROR("failed to get uid/gid for user %s: %d (%s)\n",
 375                               opts.user, errno, strerror(errno));
 376                         exit(EXIT_FAILURE);
 377                 }
 378                 *user = p->pw_uid;
 379                 *user_gid = p->pw_gid;
 380         } else {
 381                 *user = -1;
 382                 *user_gid = -1;
 383         }
 384
 385         if (opts.group) {
 386                 g = getgrnam(opts.group);
 387                 if (!g) {
 388                         ERROR("failed to get gid for group %s: %m\n", opts.group);
 389                         exit(EXIT_FAILURE);
 390                 }
 391                 *gr_gid = g->gr_gid;
 392         } else {
 393                 *gr_gid = -1;
 394         }
 395 };
 396
 397 static void set_jail_user(int pw_uid, int user_gid, int gr_gid)
 398 {
 399         if ((user_gid != -1) && initgroups(opts.user, user_gid)) {
 400                 ERROR("failed to initgroups() for user %s: %m\n", opts.user);
 401                 exit(EXIT_FAILURE);
 402         }
 403
 404         if ((gr_gid != -1) && setregid(gr_gid, gr_gid)) {
 405                 ERROR("failed to set group id %d: %m\n", gr_gid);
 406                 exit(EXIT_FAILURE);
 407         }
 408
 409         if ((pw_uid != -1) && setreuid(pw_uid, pw_uid)) {
 410                 ERROR("failed to set user id %d: %m\n", pw_uid);
 411                 exit(EXIT_FAILURE);
 412         }
 413 }
 414
 415 #define MAX_ENVP        8
 416 static char** build_envp(const char *seccomp)
 417 {
 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");
 424         int count = 0;
 425
 426         if (seccomp && !preload_lib) {
 427                 ERROR("failed to add preload-lib to env\n");
 428                 return NULL;
 429         }
 430         if (seccomp) {
 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;
 435         }
 436
 437         envp[count++] = container_var;
 438
 439         if (debug > 1)
 440                 envp[count++] = debug_var;
 441
 442         return envp;
 443 }
 444
 445 static void usage(void)
 446 {
 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");
 478 }
 479
 480 static int exec_jail(void *pipes_ptr)
 481 {
 482         int *pipes = (int*)pipes_ptr;
 483         char buf[1];
 484         int pw_uid, pw_gid, gr_gid;
 485
 486         close(pipes[0]);
 487         close(pipes[3]);
 488
 489
 490         buf[0] = 'i';
 491         if (write(pipes[1], buf, 1) < 1) {
 492                 ERROR("can't write to parent\n");
 493                 exit(EXIT_FAILURE);
 494         }
 495         if (read(pipes[2], buf, 1) < 1) {
 496                 ERROR("can't read from parent\n");
 497                 exit(EXIT_FAILURE);
 498         }
 499         if (buf[0] != 'O') {
 500                 ERROR("parent had an error, child exiting\n");
 501                 exit(EXIT_FAILURE);
 502         }
 503
 504         close(pipes[1]);
 505         close(pipes[2]);
 506
 507         if (opts.namespace & CLONE_NEWUSER) {
 508                 if (setgid(0) < 0) {
 509                         ERROR("setgid\n");
 510                         exit(EXIT_FAILURE);
 511                 }
 512                 if (setuid(0) < 0) {
 513                         ERROR("setuid\n");
 514                         exit(EXIT_FAILURE);
 515                 }
 516 //              if (setgroups(0, NULL) < 0) {
 517 //                      ERROR("setgroups\n");
 518 //                      exit(EXIT_FAILURE);
 519 //              }
 520         }
 521
 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);
 525                 exit(EXIT_FAILURE);
 526         }
 527
 528         if ((opts.namespace & CLONE_NEWNS) && build_jail_fs()) {
 529                 ERROR("failed to build jail fs\n");
 530                 exit(EXIT_FAILURE);
 531         }
 532
 533         if (opts.capabilities && drop_capabilities(opts.capabilities))
 534                 exit(EXIT_FAILURE);
 535
 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");
 538                 exit(EXIT_FAILURE);
 539         }
 540
 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);
 544         }
 545
 546         char **envp = build_envp(opts.seccomp);
 547         if (!envp)
 548                 exit(EXIT_FAILURE);
 549
 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);
 554         exit(EXIT_FAILURE);
 555 }
 556
 557 static int jail_running = 1;
 558 static int jail_return_code = 0;
 559
 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,
 563 };
 564
 565 static void jail_process_handler(struct uloop_process *c, int ret)
 566 {
 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);
 571         } else {
 572                 jail_return_code = WTERMSIG(ret);
 573                 INFO("jail (%d) exited with signal: %d\n", c->pid, jail_return_code);
 574         }
 575         jail_running = 0;
 576         uloop_end();
 577 }
 578
 579 static struct uloop_process jail_process = {
 580         .cb = jail_process_handler,
 581 };
 582
 583 static void jail_process_timeout_cb(struct uloop_timeout *t)
 584 {
 585         DEBUG("jail process failed to stop, sending SIGKILL\n");
 586         kill(jail_process.pid, SIGKILL);
 587 }
 588
 589 static void jail_handle_signal(int signo)
 590 {
 591         DEBUG("forwarding signal %d to the jailed process\n", signo);
 592         kill(jail_process.pid, signo);
 593 }
 594
 595 static int netns_open_pid(const pid_t target_ns)
 596 {
 597         char pid_net_path[PATH_MAX];
 598
 599         snprintf(pid_net_path, sizeof(pid_net_path), "/proc/%u/ns/net", target_ns);
 600
 601         return open(pid_net_path, O_RDONLY);
 602 }
 603
 604 static void netns_updown(pid_t pid, bool start)
 605 {
 606         struct ubus_context *ctx = ubus_connect(NULL);
 607         static struct blob_buf req;
 608         uint32_t id;
 609
 610         if (!ctx)
 611                 return;
 612
 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);
 617
 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");
 621
 622         blob_buf_free(&req);
 623         ubus_free(ctx);
 624 }
 625
 626 int main(int argc, char **argv)
 627 {
 628         sigset_t sigmask;
 629         uid_t uid = getuid();
 630         char log[] = "/dev/log";
 631         char ubus[] = "/var/run/ubus.sock";
 632         int ch, i;
 633         int pipes[4];
 634         char sig_buf[1];
 635         int netns_fd;
 636
 637         if (uid) {
 638                 ERROR("not root, aborting: %m\n");
 639                 return EXIT_FAILURE;
 640         }
 641
 642         umask(022);
 643         mount_list_init();
 644         init_library_search();
 645
 646         while ((ch = getopt(argc, argv, OPT_ARGS)) != -1) {
 647                 switch (ch) {
 648                 case 'd':
 649                         debug = atoi(optarg);
 650                         break;
 651                 case 'p':
 652                         opts.namespace |= CLONE_NEWNS;
 653                         opts.procfs = 1;
 654                         break;
 655                 case 'o':
 656                         opts.namespace |= CLONE_NEWNS;
 657                         opts.ronly = 1;
 658                         break;
 659                 case 'f':
 660                         opts.namespace |= CLONE_NEWUSER;
 661                         break;
 662                 case 'F':
 663                         opts.namespace |= CLONE_NEWCGROUP;
 664                         break;
 665                 case 'R':
 666                         opts.extroot = optarg;
 667                         break;
 668                 case 's':
 669                         opts.namespace |= CLONE_NEWNS;
 670                         opts.sysfs = 1;
 671                         break;
 672                 case 'S':
 673                         opts.seccomp = optarg;
 674                         add_mount(optarg, 1, -1);
 675                         break;
 676                 case 'C':
 677                         opts.capabilities = optarg;
 678                         break;
 679                 case 'c':
 680                         opts.no_new_privs = 1;
 681                         break;
 682                 case 'n':
 683                         opts.name = optarg;
 684                         break;
 685                 case 'N':
 686                         opts.namespace |= CLONE_NEWNET;
 687                         break;
 688                 case 'h':
 689                         opts.namespace |= CLONE_NEWUTS;
 690                         opts.hostname = optarg;
 691                         break;
 692                 case 'r':
 693                         opts.namespace |= CLONE_NEWNS;
 694                         add_path_and_deps(optarg, 1, 0, 0);
 695                         break;
 696                 case 'w':
 697                         opts.namespace |= CLONE_NEWNS;
 698                         add_path_and_deps(optarg, 0, 0, 0);
 699                         break;
 700                 case 'u':
 701                         opts.namespace |= CLONE_NEWNS;
 702                         add_mount(ubus, 0, -1);
 703                         break;
 704                 case 'l':
 705                         opts.namespace |= CLONE_NEWNS;
 706                         add_mount(log, 0, -1);
 707                         break;
 708                 case 'U':
 709                         opts.user = optarg;
 710                         break;
 711                 case 'G':
 712                         opts.group = optarg;
 713                         break;
 714                 case 'O':
 715                         opts.overlaydir = optarg;
 716                         break;
 717                 case 'T':
 718                         opts.tmpoverlaysize = optarg;
 719                         break;
 720                 case 'E':
 721                         opts.require_jail = 1;
 722                         break;
 723                 }
 724         }
 725
 726         if (opts.namespace)
 727                 opts.namespace |= CLONE_NEWIPC | CLONE_NEWPID;
 728
 729         if (opts.tmpoverlaysize && strlen(opts.tmpoverlaysize) > 8) {
 730                 ERROR("size parameter too long: \"%s\"\n", opts.tmpoverlaysize);
 731                 return -1;
 732         }
 733
 734         /* no <binary> param found */
 735         if (argc - optind < 1) {
 736                 usage();
 737                 return EXIT_FAILURE;
 738         }
 739         if (!(opts.namespace||opts.capabilities||opts.seccomp)) {
 740                 ERROR("Not using namespaces, capabilities or seccomp !!!\n\n");
 741                 usage();
 742                 return EXIT_FAILURE;
 743         }
 744         DEBUG("Using namespaces(0x%08x), capabilities(%d), seccomp(%d)\n",
 745                 opts.namespace,
 746                 opts.capabilities != 0,
 747                 opts.seccomp != 0);
 748
 749         opts.jail_argv = &argv[optind];
 750
 751         get_jail_user(&opts.pw_uid, &opts.pw_gid, &opts.gr_gid);
 752
 753         if (!opts.extroot) {
 754                 if (opts.namespace && add_path_and_deps(*opts.jail_argv, 1, -1, 0)) {
 755                         ERROR("failed to load dependencies\n");
 756                         return -1;
 757                 }
 758         }
 759
 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                 opts.seccomp = 0;
 763                 if (opts.require_jail)
 764                         return -1;
 765         }
 766
 767         if (opts.name)
 768                 prctl(PR_SET_NAME, opts.name, NULL, NULL, NULL);
 769
 770         uloop_init();
 771
 772         sigfillset(&sigmask);
 773         for (i = 0; i < _NSIG; i++) {
 774                 struct sigaction s = { 0 };
 775
 776                 if (!sigismember(&sigmask, i))
 777                         continue;
 778                 if ((i == SIGCHLD) || (i == SIGPIPE))
 779                         continue;
 780
 781                 s.sa_handler = jail_handle_signal;
 782                 sigaction(i, &s, NULL);
 783         }
 784
 785         if (opts.namespace) {
 786                 if (opts.namespace & CLONE_NEWNS) {
 787                         add_mount("/dev/full", 0, -1);
 788                         add_mount("/dev/null", 0, -1);
 789                         add_mount("/dev/random", 0, -1);
 790                         add_mount("/dev/urandom", 0, -1);
 791                         add_mount("/dev/tty", 0, -1);
 792                         add_mount("/dev/zero", 0, -1);
 793                         add_mount("/dev/console", 0, -1);
 794
 795                         if (!opts.extroot && (opts.user || opts.group)) {
 796                                 add_mount("/etc/passwd", 0, -1);
 797                                 add_mount("/etc/group", 0, -1);
 798                         }
 799
 800 #if defined(__GLIBC__)
 801                         if (!opts.extroot)
 802                                 add_mount("/etc/nsswitch.conf", 0, -1);
 803 #endif
 804
 805                         if (!(opts.namespace & CLONE_NEWNET)) {
 806                                 add_mount("/etc/resolv.conf", 0, -1);
 807                         }
 808                 }
 809
 810                 if (pipe(&pipes[0]) < 0 || pipe(&pipes[2]) < 0)
 811                         return -1;
 812
 813                 jail_process.pid = clone(exec_jail, child_stack + STACK_SIZE, SIGCHLD | opts.namespace, &pipes);
 814         } else {
 815                 jail_process.pid = fork();
 816         }
 817
 818         if (jail_process.pid > 0) {
 819                 seteuid(0);
 820                 /* parent process */
 821                 close(pipes[1]);
 822                 close(pipes[2]);
 823                 if (read(pipes[0], sig_buf, 1) < 1) {
 824                         ERROR("can't read from child\n");
 825                         return -1;
 826                 }
 827                 close(pipes[0]);
 828                 if (opts.namespace & CLONE_NEWUSER) {
 829                         bool has_gr = (opts.gr_gid != -1);
 830                         if (write_setgroups(jail_process.pid, false)) {
 831                                 ERROR("can't write setgroups\n");
 832                                 return -1;
 833                         }
 834                         if (opts.pw_uid != -1) {
 835                                 write_uid_gid_map(jail_process.pid, 0, opts.pw_uid);
 836                                 write_uid_gid_map(jail_process.pid, 1, has_gr?opts.gr_gid:opts.pw_gid);
 837                         } else {
 838                                 write_uid_gid_map(jail_process.pid, 0, 65534);
 839                                 write_uid_gid_map(jail_process.pid, 1, has_gr?opts.gr_gid:65534);
 840                         }
 841                 }
 842
 843                 if (opts.namespace & CLONE_NEWNET) {
 844                         netns_fd = netns_open_pid(jail_process.pid);
 845                         netns_updown(jail_process.pid, true);
 846                 }
 847
 848                 sig_buf[0] = 'O';
 849                 if (write(pipes[3], sig_buf, 1) < 0) {
 850                         ERROR("can't write to child\n");
 851                         return -1;
 852                 }
 853                 close(pipes[3]);
 854                 uloop_process_add(&jail_process);
 855                 uloop_run();
 856                 if (jail_running) {
 857                         DEBUG("uloop interrupted, killing jail process\n");
 858                         kill(jail_process.pid, SIGTERM);
 859                         uloop_timeout_set(&jail_process_timeout, 1000);
 860                         uloop_run();
 861                 }
 862                 uloop_done();
 863                 if (opts.namespace & CLONE_NEWNET) {
 864                         setns(netns_fd, CLONE_NEWNET);
 865                         netns_updown(getpid(), false);
 866                         close(netns_fd);
 867                 }
 868                 return jail_return_code;
 869         } else if (jail_process.pid == 0) {
 870                 /* fork child process */
 871                 return exec_jail(NULL);
 872         } else {
 873                 ERROR("failed to clone/fork: %m\n");
 874                 return EXIT_FAILURE;
 875         }
 876 }