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:"
  44
  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)
  52
  53 static struct {
  54         char *name;
  55         char *hostname;
  56         char **jail_argv;
  57         char *seccomp;
  58         char *capabilities;
  59         char *user;
  60         char *group;
  61         char *extroot;
  62         int no_new_privs;
  63         int namespace;
  64         int procfs;
  65         int ronly;
  66         int sysfs;
  67 } opts;
  68
  69 extern int pivot_root(const char *new_root, const char *put_old);
  70
  71 int debug = 0;
  72
  73 static char child_stack[STACK_SIZE];
  74
  75 static int mkdir_p(char *dir, mode_t mask)
  76 {
  77         char *l = strrchr(dir, '/');
  78         int ret;
  79
  80         if (!l)
  81                 return 0;
  82
  83         *l = '\0';
  84
  85         if (mkdir_p(dir, mask))
  86                 return -1;
  87
  88         *l = '/';
  89
  90         ret = mkdir(dir, mask);
  91         if (ret && errno == EEXIST)
  92                 return 0;
  93
  94         if (ret)
  95                 ERROR("mkdir(%s, %d) failed: %m\n", dir, mask);
  96
  97         return ret;
  98 }
  99
 100 static int _mount_bind(const char *root, const char *path, const char *target, int readonly, int strict, int error)
 101 {
 102         struct stat s;
 103         char new[PATH_MAX];
 104         int fd;
 105         int remount_flags = MS_BIND | MS_REMOUNT;
 106
 107         if (stat(path, &s)) {
 108                 ERROR("stat(%s) failed: %m\n", path);
 109                 return error;
 110         }
 111
 112         snprintf(new, sizeof(new), "%s%s", root, target?target:path);
 113
 114         if (S_ISDIR(s.st_mode)) {
 115                 mkdir_p(new, 0755);
 116         } else {
 117                 mkdir_p(dirname(new), 0755);
 118                 snprintf(new, sizeof(new), "%s%s", root, target?target:path);
 119                 fd = creat(new, 0644);
 120                 if (fd == -1) {
 121                         ERROR("creat(%s) failed: %m\n", new);
 122                         return -1;
 123                 }
 124                 close(fd);
 125         }
 126
 127         if (mount(path, new, NULL, MS_BIND, NULL)) {
 128                 ERROR("failed to mount -B %s %s: %m\n", path, new);
 129                 return -1;
 130         }
 131
 132         if (readonly)
 133                 remount_flags |= MS_RDONLY;
 134
 135         if (strict)
 136                 remount_flags |= MS_NOEXEC | MS_NOSUID | MS_NODEV;
 137
 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);
 141                 return -1;
 142         }
 143
 144         DEBUG("mount -B %s %s (%s%s%s)\n", path, new,
 145               readonly?"ro":"rw", (readonly && strict)?", ":"", strict?"strict":"");
 146
 147         return 0;
 148 }
 149
 150 int mount_bind(const char *root, const char *path, int readonly, int error) {
 151         return _mount_bind(root, path, NULL, readonly, 0, error);
 152 }
 153
 154 static int build_jail_fs(void)
 155 {
 156         char jail_root[] = "/tmp/ujail-XXXXXX";
 157         if (mkdtemp(jail_root) == NULL) {
 158                 ERROR("mkdtemp(%s) failed: %m\n", jail_root);
 159                 return -1;
 160         }
 161
 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");
 165                 return -1;
 166         }
 167
 168         if (opts.extroot) {
 169                 if (mount(opts.extroot, jail_root, NULL, MS_BIND | MS_REC, NULL)) {
 170                         ERROR("extroot mount failed %m\n");
 171                         return -1;
 172                 }
 173         } else {
 174                 if (mount("tmpfs", jail_root, "tmpfs", MS_NOATIME, "mode=0755")) {
 175                         ERROR("tmpfs mount failed %m\n");
 176                         return -1;
 177                 }
 178         }
 179
 180         if (chdir(jail_root)) {
 181                 ERROR("chdir(%s) (jail_root) failed: %m\n", jail_root);
 182                 return -1;
 183         }
 184
 185         if (mount_all(jail_root)) {
 186                 ERROR("mount_all() failed\n");
 187                 return -1;
 188         }
 189
 190         if (opts.namespace & NAMESPACE_NET) {
 191                 char hostdir[PATH_MAX], jailetc[PATH_MAX], jaillink[PATH_MAX];
 192
 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);
 200         }
 201
 202         char dirbuf[sizeof(jail_root) + 4];
 203         snprintf(dirbuf, sizeof(dirbuf), "%s/old", jail_root);
 204         mkdir(dirbuf, 0755);
 205
 206         if (pivot_root(jail_root, dirbuf) == -1) {
 207                 ERROR("pivot_root(%s, %s) failed: %m\n", jail_root, dirbuf);
 208                 return -1;
 209         }
 210         if (chdir("/")) {
 211                 ERROR("chdir(/) (after pivot_root) failed: %m\n");
 212                 return -1;
 213         }
 214
 215         snprintf(dirbuf, sizeof(dirbuf), "/old%s", jail_root);
 216         rmdir(dirbuf);
 217         umount2("/old", MNT_DETACH);
 218         rmdir("/old");
 219
 220         if (opts.procfs) {
 221                 mkdir("/proc", 0755);
 222                 mount("proc", "/proc", "proc", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
 223         }
 224         if (opts.sysfs) {
 225                 mkdir("/sys", 0755);
 226                 mount("sysfs", "/sys", "sysfs", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
 227         }
 228         if (opts.ronly)
 229                 mount(NULL, "/", NULL, MS_RDONLY | MS_REMOUNT, 0);
 230
 231         return 0;
 232 }
 233
 234 #define MAX_ENVP        8
 235 static char** build_envp(const char *seccomp)
 236 {
 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");
 243         int count = 0;
 244
 245         if (seccomp && !preload_lib) {
 246                 ERROR("failed to add preload-lib to env\n");
 247                 return NULL;
 248         }
 249         if (seccomp) {
 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;
 254         }
 255
 256         if (is_extroot)
 257                 envp[count++] = container_var;
 258
 259         if (debug > 1)
 260                 envp[count++] = debug_var;
 261
 262         return envp;
 263 }
 264
 265 static void usage(void)
 266 {
 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");
 293 }
 294
 295 static int exec_jail(void *_notused)
 296 {
 297         struct passwd *p = NULL;
 298         struct group *g = NULL;
 299
 300         if (opts.capabilities && drop_capabilities(opts.capabilities))
 301                 exit(EXIT_FAILURE);
 302
 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");
 305                 exit(EXIT_FAILURE);
 306         }
 307
 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);
 311                 exit(EXIT_FAILURE);
 312         }
 313
 314         if (opts.namespace && build_jail_fs()) {
 315                 ERROR("failed to build jail fs\n");
 316                 exit(EXIT_FAILURE);
 317         }
 318
 319         if (opts.user) {
 320                 p = getpwnam(opts.user);
 321                 if (!p) {
 322                         ERROR("failed to get uid/gid for user %s: %d (%s)\n",
 323                               opts.user, errno, strerror(errno));
 324                         exit(EXIT_FAILURE);
 325                 }
 326         }
 327
 328         if (opts.group) {
 329                 g = getgrnam(opts.group);
 330                 if (!g) {
 331                         ERROR("failed to get gid for group %s: %m\n", opts.group);
 332                         exit(EXIT_FAILURE);
 333                 }
 334         }
 335
 336         if (p && p->pw_gid && initgroups(opts.user, p->pw_gid)) {
 337                 ERROR("failed to initgroups() for user %s: %m\n", opts.user);
 338                 exit(EXIT_FAILURE);
 339         }
 340
 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);
 343                 exit(EXIT_FAILURE);
 344         }
 345
 346         if (p && p->pw_uid && setuid(p->pw_uid)) {
 347                 ERROR("failed to set user id %d: %m\n", p->pw_uid);
 348                 exit(EXIT_FAILURE);
 349         }
 350
 351
 352         char **envp = build_envp(opts.seccomp);
 353         if (!envp)
 354                 exit(EXIT_FAILURE);
 355
 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);
 360         exit(EXIT_FAILURE);
 361 }
 362
 363 static int jail_running = 1;
 364 static int jail_return_code = 0;
 365
 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,
 369 };
 370
 371 static void jail_process_handler(struct uloop_process *c, int ret)
 372 {
 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);
 377         } else {
 378                 jail_return_code = WTERMSIG(ret);
 379                 INFO("jail (%d) exited with signal: %d\n", c->pid, jail_return_code);
 380         }
 381         jail_running = 0;
 382         uloop_end();
 383 }
 384
 385 static struct uloop_process jail_process = {
 386         .cb = jail_process_handler,
 387 };
 388
 389 static void jail_process_timeout_cb(struct uloop_timeout *t)
 390 {
 391         DEBUG("jail process failed to stop, sending SIGKILL\n");
 392         kill(jail_process.pid, SIGKILL);
 393 }
 394
 395 static void jail_handle_signal(int signo)
 396 {
 397         DEBUG("forwarding signal %d to the jailed process\n", signo);
 398         kill(jail_process.pid, signo);
 399 }
 400
 401 static void netns_updown(bool start)
 402 {
 403         struct ubus_context *ctx = ubus_connect(NULL);
 404         static struct blob_buf req;
 405         uint32_t id;
 406         pid_t pid = getpid();
 407
 408         if (!ctx)
 409                 return;
 410
 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);
 415
 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");
 419
 420         blob_buf_free(&req);
 421         ubus_free(ctx);
 422 }
 423
 424 int main(int argc, char **argv)
 425 {
 426         sigset_t sigmask;
 427         uid_t uid = getuid();
 428         char log[] = "/dev/log";
 429         char ubus[] = "/var/run/ubus.sock";
 430         int ch, i;
 431
 432         if (uid) {
 433                 ERROR("not root, aborting: %m\n");
 434                 return EXIT_FAILURE;
 435         }
 436
 437         umask(022);
 438         mount_list_init();
 439         init_library_search();
 440
 441         while ((ch = getopt(argc, argv, OPT_ARGS)) != -1) {
 442                 switch (ch) {
 443                 case 'd':
 444                         debug = atoi(optarg);
 445                         break;
 446                 case 'p':
 447                         opts.namespace |= NAMESPACE_MOUNT;
 448                         opts.procfs = 1;
 449                         break;
 450                 case 'o':
 451                         opts.namespace |= NAMESPACE_MOUNT;
 452                         opts.ronly = 1;
 453                         break;
 454                 case 'R':
 455                         opts.namespace |= NAMESPACE_MOUNT | NAMESPACE_UTS;
 456                         opts.extroot = optarg;
 457                         break;
 458                 case 's':
 459                         opts.namespace |= NAMESPACE_MOUNT;
 460                         opts.sysfs = 1;
 461                         break;
 462                 case 'S':
 463                         opts.seccomp = optarg;
 464                         add_mount(optarg, 1, -1);
 465                         break;
 466                 case 'C':
 467                         opts.capabilities = optarg;
 468                         break;
 469                 case 'c':
 470                         opts.no_new_privs = 1;
 471                         break;
 472                 case 'n':
 473                         opts.name = optarg;
 474                         break;
 475                 case 'N':
 476                         opts.namespace |= NAMESPACE_NET;
 477                         break;
 478                 case 'h':
 479                         opts.hostname = optarg;
 480                         break;
 481                 case 'r':
 482                         opts.namespace |= NAMESPACE_MOUNT;
 483                         add_path_and_deps(optarg, 1, 0, 0);
 484                         break;
 485                 case 'w':
 486                         opts.namespace |= NAMESPACE_MOUNT;
 487                         add_path_and_deps(optarg, 0, 0, 0);
 488                         break;
 489                 case 'u':
 490                         opts.namespace |= NAMESPACE_MOUNT;
 491                         add_mount(ubus, 0, -1);
 492                         break;
 493                 case 'l':
 494                         opts.namespace |= NAMESPACE_MOUNT;
 495                         add_mount(log, 0, -1);
 496                         break;
 497                 case 'U':
 498                         opts.user = optarg;
 499                         break;
 500                 case 'G':
 501                         opts.group = optarg;
 502                         break;
 503                 }
 504         }
 505
 506         /* no <binary> param found */
 507         if (argc - optind < 1) {
 508                 usage();
 509                 return EXIT_FAILURE;
 510         }
 511         if (!(opts.namespace||opts.capabilities||opts.seccomp)) {
 512                 ERROR("Not using namespaces, capabilities or seccomp !!!\n\n");
 513                 usage();
 514                 return EXIT_FAILURE;
 515         }
 516         DEBUG("Using namespaces(%d), capabilities(%d), seccomp(%d)\n",
 517                 opts.namespace,
 518                 opts.capabilities != 0,
 519                 opts.seccomp != 0);
 520
 521         opts.jail_argv = &argv[optind];
 522
 523         if (!opts.extroot) {
 524                 if (opts.namespace && add_path_and_deps(*opts.jail_argv, 1, -1, 0)) {
 525                         ERROR("failed to load dependencies\n");
 526                         return -1;
 527                 }
 528
 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");
 531                         return -1;
 532                 }
 533         }
 534
 535         if (opts.name)
 536                 prctl(PR_SET_NAME, opts.name, NULL, NULL, NULL);
 537
 538         uloop_init();
 539
 540         sigfillset(&sigmask);
 541         for (i = 0; i < _NSIG; i++) {
 542                 struct sigaction s = { 0 };
 543
 544                 if (!sigismember(&sigmask, i))
 545                         continue;
 546                 if ((i == SIGCHLD) || (i == SIGPIPE))
 547                         continue;
 548
 549                 s.sa_handler = jail_handle_signal;
 550                 sigaction(i, &s, NULL);
 551         }
 552
 553         if (opts.namespace) {
 554                 int flags = SIGCHLD | CLONE_NEWPID | CLONE_NEWIPC;
 555
 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);
 565
 566                         if (opts.user || opts.group) {
 567                                 add_mount("/etc/passwd", 0, -1);
 568                                 add_mount("/etc/group", 0, -1);
 569                         }
 570
 571                         if (!(opts.namespace & CLONE_NEWNET)) {
 572                                 add_mount("/etc/resolv.conf", 0, -1);
 573                         }
 574                 }
 575
 576                 if (opts.hostname)
 577                         flags |= CLONE_NEWUTS;
 578
 579                 if (opts.namespace & NAMESPACE_NET) {
 580                         unshare(CLONE_NEWNET);
 581                         netns_updown(true);
 582                 };
 583
 584                 jail_process.pid = clone(exec_jail, child_stack + STACK_SIZE, flags, NULL);
 585         } else {
 586                 jail_process.pid = fork();
 587         }
 588
 589         if (jail_process.pid > 0) {
 590                 /* parent process */
 591                 uloop_process_add(&jail_process);
 592                 uloop_run();
 593                 if (jail_running) {
 594                         DEBUG("uloop interrupted, killing jail process\n");
 595                         kill(jail_process.pid, SIGTERM);
 596                         uloop_timeout_set(&jail_process_timeout, 1000);
 597                         uloop_run();
 598                 }
 599                 uloop_done();
 600                 if (opts.namespace & NAMESPACE_NET)
 601                         netns_updown(false);
 602
 603                 return jail_return_code;
 604         } else if (jail_process.pid == 0) {
 605                 /* fork child process */
 606                 return exec_jail(NULL);
 607         } else {
 608                 ERROR("failed to clone/fork: %m\n");
 609                 return EXIT_FAILURE;
 610         }
 611 }