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:N"
  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         int no_new_privs;
  62         int namespace;
  63         int procfs;
  64         int ronly;
  65         int sysfs;
  66 } opts;
  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 build_jail_fs(void)
 154 {
 155         char jail_root[] = "/tmp/ujail-XXXXXX";
 156         if (mkdtemp(jail_root) == NULL) {
 157                 ERROR("mkdtemp(%s) failed: %m\n", jail_root);
 158                 return -1;
 159         }
 160
 161         /* oldroot can't be MS_SHARED else pivot_root() fails */
 162         if (mount("none", "/", NULL, MS_REC|MS_PRIVATE, NULL)) {
 163                 ERROR("private mount failed %m\n");
 164                 return -1;
 165         }
 166
 167         if (mount("tmpfs", jail_root, "tmpfs", MS_NOATIME, "mode=0755")) {
 168                 ERROR("tmpfs mount failed %m\n");
 169                 return -1;
 170         }
 171
 172         if (chdir(jail_root)) {
 173                 ERROR("chdir(%s) (jail_root) failed: %m\n", jail_root);
 174                 return -1;
 175         }
 176
 177         if (mount_all(jail_root)) {
 178                 ERROR("mount_all() failed\n");
 179                 return -1;
 180         }
 181
 182         if (opts.namespace & NAMESPACE_NET) {
 183                 char hostdir[PATH_MAX], jailetc[PATH_MAX], jaillink[PATH_MAX];
 184
 185                 snprintf(hostdir, PATH_MAX, "/tmp/resolv.conf-%s.d", opts.name);
 186                 mkdir_p(hostdir, 0755);
 187                 _mount_bind(jail_root, hostdir, "/tmp/resolv.conf.d", 1, 1, -1);
 188                 snprintf(jailetc, PATH_MAX, "%s/etc", jail_root);
 189                 mkdir_p(jailetc, 0755);
 190                 snprintf(jaillink, PATH_MAX, "%s/etc/resolv.conf", jail_root);
 191                 symlink("../tmp/resolv.conf.d/resolv.conf.auto", jaillink);
 192         }
 193
 194         char dirbuf[sizeof(jail_root) + 4];
 195         snprintf(dirbuf, sizeof(dirbuf), "%s/old", jail_root);
 196         mkdir(dirbuf, 0755);
 197
 198         if (pivot_root(jail_root, dirbuf) == -1) {
 199                 ERROR("pivot_root(%s, %s) failed: %m\n", jail_root, dirbuf);
 200                 return -1;
 201         }
 202         if (chdir("/")) {
 203                 ERROR("chdir(/) (after pivot_root) failed: %m\n");
 204                 return -1;
 205         }
 206
 207         snprintf(dirbuf, sizeof(dirbuf), "/old%s", jail_root);
 208         rmdir(dirbuf);
 209         umount2("/old", MNT_DETACH);
 210         rmdir("/old");
 211
 212         if (opts.procfs) {
 213                 mkdir("/proc", 0755);
 214                 mount("proc", "/proc", "proc", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
 215         }
 216         if (opts.sysfs) {
 217                 mkdir("/sys", 0755);
 218                 mount("sysfs", "/sys", "sysfs", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
 219         }
 220         if (opts.ronly)
 221                 mount(NULL, "/", NULL, MS_RDONLY | MS_REMOUNT, 0);
 222
 223         return 0;
 224 }
 225
 226 #define MAX_ENVP        8
 227 static char** build_envp(const char *seccomp)
 228 {
 229         static char *envp[MAX_ENVP];
 230         static char preload_var[PATH_MAX];
 231         static char seccomp_var[PATH_MAX];
 232         static char debug_var[] = "LD_DEBUG=all";
 233         const char *preload_lib = find_lib("libpreload-seccomp.so");
 234         int count = 0;
 235
 236         if (seccomp && !preload_lib) {
 237                 ERROR("failed to add preload-lib to env\n");
 238                 return NULL;
 239         }
 240         if (seccomp) {
 241                 snprintf(seccomp_var, sizeof(seccomp_var), "SECCOMP_FILE=%s", seccomp);
 242                 envp[count++] = seccomp_var;
 243                 snprintf(preload_var, sizeof(preload_var), "LD_PRELOAD=%s", preload_lib);
 244                 envp[count++] = preload_var;
 245         }
 246         if (debug > 1)
 247                 envp[count++] = debug_var;
 248
 249         return envp;
 250 }
 251
 252 static void usage(void)
 253 {
 254         fprintf(stderr, "ujail <options> -- <binary> <params ...>\n");
 255         fprintf(stderr, "  -d <num>\tshow debug log (increase num to increase verbosity)\n");
 256         fprintf(stderr, "  -S <file>\tseccomp filter config\n");
 257         fprintf(stderr, "  -C <file>\tcapabilities drop config\n");
 258         fprintf(stderr, "  -c\t\tset PR_SET_NO_NEW_PRIVS\n");
 259         fprintf(stderr, "  -n <name>\tthe name of the jail\n");
 260         fprintf(stderr, "namespace jail options:\n");
 261         fprintf(stderr, "  -h <hostname>\tchange the hostname of the jail\n");
 262         fprintf(stderr, "  -N\t\tjail has network namespace\n");
 263         fprintf(stderr, "  -r <file>\treadonly files that should be staged\n");
 264         fprintf(stderr, "  -w <file>\twriteable files that should be staged\n");
 265         fprintf(stderr, "  -p\t\tjail has /proc\n");
 266         fprintf(stderr, "  -s\t\tjail has /sys\n");
 267         fprintf(stderr, "  -l\t\tjail has /dev/log\n");
 268         fprintf(stderr, "  -u\t\tjail has a ubus socket\n");
 269         fprintf(stderr, "  -U <name>\tuser to run jailed process\n");
 270         fprintf(stderr, "  -G <name>\tgroup to run jailed process\n");
 271         fprintf(stderr, "  -o\t\tremont jail root (/) read only\n");
 272         fprintf(stderr, "\nWarning: by default root inside the jail is the same\n\
 273 and he has the same powers as root outside the jail,\n\
 274 thus he can escape the jail and/or break stuff.\n\
 275 Please use seccomp/capabilities (-S/-C) to restrict his powers\n\n\
 276 If you use none of the namespace jail options,\n\
 277 ujail will not use namespace/build a jail,\n\
 278 and will only drop capabilities/apply seccomp filter.\n\n");
 279 }
 280
 281 static int exec_jail(void *_notused)
 282 {
 283         struct passwd *p = NULL;
 284         struct group *g = NULL;
 285
 286         if (opts.capabilities && drop_capabilities(opts.capabilities))
 287                 exit(EXIT_FAILURE);
 288
 289         if (opts.no_new_privs && prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
 290                 ERROR("prctl(PR_SET_NO_NEW_PRIVS) failed: %m\n");
 291                 exit(EXIT_FAILURE);
 292         }
 293
 294         if (opts.namespace && opts.hostname && strlen(opts.hostname) > 0
 295                         && sethostname(opts.hostname, strlen(opts.hostname))) {
 296                 ERROR("sethostname(%s) failed: %m\n", opts.hostname);
 297                 exit(EXIT_FAILURE);
 298         }
 299
 300         if (opts.namespace && build_jail_fs()) {
 301                 ERROR("failed to build jail fs\n");
 302                 exit(EXIT_FAILURE);
 303         }
 304
 305         if (opts.user) {
 306                 p = getpwnam(opts.user);
 307                 if (!p) {
 308                         ERROR("failed to get uid/gid for user %s: %d (%s)\n",
 309                               opts.user, errno, strerror(errno));
 310                         exit(EXIT_FAILURE);
 311                 }
 312         }
 313
 314         if (opts.group) {
 315                 g = getgrnam(opts.group);
 316                 if (!g) {
 317                         ERROR("failed to get gid for group %s: %m\n", opts.group);
 318                         exit(EXIT_FAILURE);
 319                 }
 320         }
 321
 322         if (p && p->pw_gid && initgroups(opts.user, p->pw_gid)) {
 323                 ERROR("failed to initgroups() for user %s: %m\n", opts.user);
 324                 exit(EXIT_FAILURE);
 325         }
 326
 327         if (g && g->gr_gid && setgid(g->gr_gid)) {
 328                 ERROR("failed to set group id %d: %m\n", g?g->gr_gid:p->pw_gid);
 329                 exit(EXIT_FAILURE);
 330         }
 331
 332         if (p && p->pw_uid && setuid(p->pw_uid)) {
 333                 ERROR("failed to set user id %d: %m\n", p->pw_uid);
 334                 exit(EXIT_FAILURE);
 335         }
 336
 337
 338         char **envp = build_envp(opts.seccomp);
 339         if (!envp)
 340                 exit(EXIT_FAILURE);
 341
 342         INFO("exec-ing %s\n", *opts.jail_argv);
 343         execve(*opts.jail_argv, opts.jail_argv, envp);
 344         /* we get there only if execve fails */
 345         ERROR("failed to execve %s: %m\n", *opts.jail_argv);
 346         exit(EXIT_FAILURE);
 347 }
 348
 349 static int jail_running = 1;
 350 static int jail_return_code = 0;
 351
 352 static void jail_process_timeout_cb(struct uloop_timeout *t);
 353 static struct uloop_timeout jail_process_timeout = {
 354         .cb = jail_process_timeout_cb,
 355 };
 356
 357 static void jail_process_handler(struct uloop_process *c, int ret)
 358 {
 359         uloop_timeout_cancel(&jail_process_timeout);
 360         if (WIFEXITED(ret)) {
 361                 jail_return_code = WEXITSTATUS(ret);
 362                 INFO("jail (%d) exited with exit: %d\n", c->pid, jail_return_code);
 363         } else {
 364                 jail_return_code = WTERMSIG(ret);
 365                 INFO("jail (%d) exited with signal: %d\n", c->pid, jail_return_code);
 366         }
 367         jail_running = 0;
 368         uloop_end();
 369 }
 370
 371 static struct uloop_process jail_process = {
 372         .cb = jail_process_handler,
 373 };
 374
 375 static void jail_process_timeout_cb(struct uloop_timeout *t)
 376 {
 377         DEBUG("jail process failed to stop, sending SIGKILL\n");
 378         kill(jail_process.pid, SIGKILL);
 379 }
 380
 381 static void jail_handle_signal(int signo)
 382 {
 383         DEBUG("forwarding signal %d to the jailed process\n", signo);
 384         kill(jail_process.pid, signo);
 385 }
 386
 387 static void netns_updown(bool start)
 388 {
 389         struct ubus_context *ctx = ubus_connect(NULL);
 390         static struct blob_buf req;
 391         uint32_t id;
 392         pid_t pid = getpid();
 393
 394         if (!ctx)
 395                 return;
 396
 397         blob_buf_init(&req, 0);
 398         blobmsg_add_string(&req, "jail", opts.name);
 399         blobmsg_add_u32(&req, "pid", pid);
 400         blobmsg_add_u8(&req, "start", start);
 401
 402         if (ubus_lookup_id(ctx, "network", &id) ||
 403             ubus_invoke(ctx, id, "netns_updown", req.head, NULL, NULL, 3000))
 404                 INFO("ubus request failed\n");
 405
 406         blob_buf_free(&req);
 407         ubus_free(ctx);
 408 }
 409
 410 int main(int argc, char **argv)
 411 {
 412         sigset_t sigmask;
 413         uid_t uid = getuid();
 414         char log[] = "/dev/log";
 415         char ubus[] = "/var/run/ubus.sock";
 416         int ch, i;
 417
 418         if (uid) {
 419                 ERROR("not root, aborting: %m\n");
 420                 return EXIT_FAILURE;
 421         }
 422
 423         umask(022);
 424         mount_list_init();
 425         init_library_search();
 426
 427         while ((ch = getopt(argc, argv, OPT_ARGS)) != -1) {
 428                 switch (ch) {
 429                 case 'd':
 430                         debug = atoi(optarg);
 431                         break;
 432                 case 'p':
 433                         opts.namespace |= NAMESPACE_MOUNT;
 434                         opts.procfs = 1;
 435                         break;
 436                 case 'o':
 437                         opts.namespace |= NAMESPACE_MOUNT;
 438                         opts.ronly = 1;
 439                         break;
 440                 case 's':
 441                         opts.namespace |= NAMESPACE_MOUNT;
 442                         opts.sysfs = 1;
 443                         break;
 444                 case 'S':
 445                         opts.seccomp = optarg;
 446                         add_mount(optarg, 1, -1);
 447                         break;
 448                 case 'C':
 449                         opts.capabilities = optarg;
 450                         break;
 451                 case 'c':
 452                         opts.no_new_privs = 1;
 453                         break;
 454                 case 'n':
 455                         opts.name = optarg;
 456                         break;
 457                 case 'N':
 458                         opts.namespace |= NAMESPACE_NET;
 459                         break;
 460                 case 'h':
 461                         opts.hostname = optarg;
 462                         break;
 463                 case 'r':
 464                         opts.namespace |= NAMESPACE_MOUNT;
 465                         add_path_and_deps(optarg, 1, 0, 0);
 466                         break;
 467                 case 'w':
 468                         opts.namespace |= NAMESPACE_MOUNT;
 469                         add_path_and_deps(optarg, 0, 0, 0);
 470                         break;
 471                 case 'u':
 472                         opts.namespace |= NAMESPACE_MOUNT;
 473                         add_mount(ubus, 0, -1);
 474                         break;
 475                 case 'l':
 476                         opts.namespace |= NAMESPACE_MOUNT;
 477                         add_mount(log, 0, -1);
 478                         break;
 479                 case 'U':
 480                         opts.user = optarg;
 481                         break;
 482                 case 'G':
 483                         opts.group = optarg;
 484                         break;
 485                 }
 486         }
 487
 488         /* no <binary> param found */
 489         if (argc - optind < 1) {
 490                 usage();
 491                 return EXIT_FAILURE;
 492         }
 493         if (!(opts.namespace||opts.capabilities||opts.seccomp)) {
 494                 ERROR("Not using namespaces, capabilities or seccomp !!!\n\n");
 495                 usage();
 496                 return EXIT_FAILURE;
 497         }
 498         DEBUG("Using namespaces(%d), capabilities(%d), seccomp(%d)\n",
 499                 opts.namespace,
 500                 opts.capabilities != 0,
 501                 opts.seccomp != 0);
 502
 503         opts.jail_argv = &argv[optind];
 504
 505         if (opts.namespace && add_path_and_deps(*opts.jail_argv, 1, -1, 0)) {
 506                 ERROR("failed to load dependencies\n");
 507                 return -1;
 508         }
 509
 510         if (opts.namespace && opts.seccomp && add_path_and_deps("libpreload-seccomp.so", 1, -1, 1)) {
 511                 ERROR("failed to load libpreload-seccomp.so\n");
 512                 return -1;
 513         }
 514
 515         if (opts.name)
 516                 prctl(PR_SET_NAME, opts.name, NULL, NULL, NULL);
 517
 518         uloop_init();
 519
 520         sigfillset(&sigmask);
 521         for (i = 0; i < _NSIG; i++) {
 522                 struct sigaction s = { 0 };
 523
 524                 if (!sigismember(&sigmask, i))
 525                         continue;
 526                 if ((i == SIGCHLD) || (i == SIGPIPE))
 527                         continue;
 528
 529                 s.sa_handler = jail_handle_signal;
 530                 sigaction(i, &s, NULL);
 531         }
 532
 533         if (opts.namespace) {
 534                 int flags = SIGCHLD | CLONE_NEWPID | CLONE_NEWIPC;
 535
 536                 if (opts.namespace & NAMESPACE_MOUNT) {
 537                         flags |= CLONE_NEWNS;
 538                         add_mount("/dev/full", 0, -1);
 539                         add_mount("/dev/null", 0, -1);
 540                         add_mount("/dev/urandom", 0, -1);
 541                         add_mount("/dev/zero", 0, -1);
 542
 543                         if (opts.user || opts.group) {
 544                                 add_mount("/etc/passwd", 0, -1);
 545                                 add_mount("/etc/group", 0, -1);
 546                         }
 547
 548                         if (!(opts.namespace & CLONE_NEWNET)) {
 549                                 add_mount("/etc/resolv.conf", 0, -1);
 550                         }
 551                 }
 552
 553                 if (opts.hostname)
 554                         flags |= CLONE_NEWUTS;
 555
 556                 if (opts.namespace & NAMESPACE_NET) {
 557                         unshare(CLONE_NEWNET);
 558                         netns_updown(true);
 559                 };
 560
 561                 jail_process.pid = clone(exec_jail, child_stack + STACK_SIZE, flags, NULL);
 562         } else {
 563                 jail_process.pid = fork();
 564         }
 565
 566         if (jail_process.pid > 0) {
 567                 /* parent process */
 568                 uloop_process_add(&jail_process);
 569                 uloop_run();
 570                 if (jail_running) {
 571                         DEBUG("uloop interrupted, killing jail process\n");
 572                         kill(jail_process.pid, SIGTERM);
 573                         uloop_timeout_set(&jail_process_timeout, 1000);
 574                         uloop_run();
 575                 }
 576                 uloop_done();
 577                 if (opts.namespace & NAMESPACE_NET)
 578                         netns_updown(false);
 579
 580                 return jail_return_code;
 581         } else if (jail_process.pid == 0) {
 582                 /* fork child process */
 583                 return exec_jail(NULL);
 584         } else {
 585                 ERROR("failed to clone/fork: %m\n");
 586                 return EXIT_FAILURE;
 587         }
 588 }