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
  41 #define STACK_SIZE      (1024 * 1024)
  42 #define OPT_ARGS        "S:C:n:h:r:w:d:psulocU:G:"
  43
  44 static struct {
  45         char *name;
  46         char *hostname;
  47         char **jail_argv;
  48         char *seccomp;
  49         char *capabilities;
  50         char *user;
  51         char *group;
  52         int no_new_privs;
  53         int namespace;
  54         int procfs;
  55         int ronly;
  56         int sysfs;
  57 } opts;
  58
  59 extern int pivot_root(const char *new_root, const char *put_old);
  60
  61 int debug = 0;
  62
  63 static char child_stack[STACK_SIZE];
  64
  65 static int mkdir_p(char *dir, mode_t mask)
  66 {
  67         char *l = strrchr(dir, '/');
  68         int ret;
  69
  70         if (!l)
  71                 return 0;
  72
  73         *l = '\0';
  74
  75         if (mkdir_p(dir, mask))
  76                 return -1;
  77
  78         *l = '/';
  79
  80         ret = mkdir(dir, mask);
  81         if (ret && errno == EEXIST)
  82                 return 0;
  83
  84         if (ret)
  85                 ERROR("mkdir(%s, %d) failed: %m\n", dir, mask);
  86
  87         return ret;
  88 }
  89
  90 int mount_bind(const char *root, const char *path, int readonly, int error)
  91 {
  92         struct stat s;
  93         char new[PATH_MAX];
  94         int fd;
  95
  96         if (stat(path, &s)) {
  97                 ERROR("stat(%s) failed: %m\n", path);
  98                 return error;
  99         }
 100
 101         snprintf(new, sizeof(new), "%s%s", root, path);
 102         if (S_ISDIR(s.st_mode)) {
 103                 mkdir_p(new, 0755);
 104         } else {
 105                 mkdir_p(dirname(new), 0755);
 106                 snprintf(new, sizeof(new), "%s%s", root, path);
 107                 fd = creat(new, 0644);
 108                 if (fd == -1) {
 109                         ERROR("creat(%s) failed: %m\n", new);
 110                         return -1;
 111                 }
 112                 close(fd);
 113         }
 114
 115         if (mount(path, new, NULL, MS_BIND, NULL)) {
 116                 ERROR("failed to mount -B %s %s: %m\n", path, new);
 117                 return -1;
 118         }
 119
 120         if (readonly && mount(NULL, new, NULL, MS_BIND | MS_REMOUNT | MS_RDONLY, NULL)) {
 121                 ERROR("failed to remount ro %s: %m\n", new);
 122                 return -1;
 123         }
 124
 125         DEBUG("mount -B %s %s (%s)\n", path, new, readonly?"ro":"rw");
 126
 127         return 0;
 128 }
 129
 130 static int build_jail_fs(void)
 131 {
 132         char jail_root[] = "/tmp/ujail-XXXXXX";
 133         if (mkdtemp(jail_root) == NULL) {
 134                 ERROR("mkdtemp(%s) failed: %m\n", jail_root);
 135                 return -1;
 136         }
 137
 138         /* oldroot can't be MS_SHARED else pivot_root() fails */
 139         if (mount("none", "/", NULL, MS_REC|MS_PRIVATE, NULL)) {
 140                 ERROR("private mount failed %m\n");
 141                 return -1;
 142         }
 143
 144         if (mount("tmpfs", jail_root, "tmpfs", MS_NOATIME, "mode=0755")) {
 145                 ERROR("tmpfs mount failed %m\n");
 146                 return -1;
 147         }
 148
 149         if (chdir(jail_root)) {
 150                 ERROR("chdir(%s) (jail_root) failed: %m\n", jail_root);
 151                 return -1;
 152         }
 153
 154         if (mount_all(jail_root)) {
 155                 ERROR("mount_all() failed\n");
 156                 return -1;
 157         }
 158
 159         char dirbuf[sizeof(jail_root) + 4];
 160         snprintf(dirbuf, sizeof(dirbuf), "%s/old", jail_root);
 161         mkdir(dirbuf, 0755);
 162
 163         if (pivot_root(jail_root, dirbuf) == -1) {
 164                 ERROR("pivot_root(%s, %s) failed: %m\n", jail_root, dirbuf);
 165                 return -1;
 166         }
 167         if (chdir("/")) {
 168                 ERROR("chdir(/) (after pivot_root) failed: %m\n");
 169                 return -1;
 170         }
 171
 172         snprintf(dirbuf, sizeof(dirbuf), "/old%s", jail_root);
 173         rmdir(dirbuf);
 174         umount2("/old", MNT_DETACH);
 175         rmdir("/old");
 176
 177         if (opts.procfs) {
 178                 mkdir("/proc", 0755);
 179                 mount("proc", "/proc", "proc", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
 180         }
 181         if (opts.sysfs) {
 182                 mkdir("/sys", 0755);
 183                 mount("sysfs", "/sys", "sysfs", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
 184         }
 185         if (opts.ronly)
 186                 mount(NULL, "/", NULL, MS_RDONLY | MS_REMOUNT, 0);
 187
 188         return 0;
 189 }
 190
 191 #define MAX_ENVP        8
 192 static char** build_envp(const char *seccomp)
 193 {
 194         static char *envp[MAX_ENVP];
 195         static char preload_var[PATH_MAX];
 196         static char seccomp_var[PATH_MAX];
 197         static char debug_var[] = "LD_DEBUG=all";
 198         const char *preload_lib = find_lib("libpreload-seccomp.so");
 199         int count = 0;
 200
 201         if (seccomp && !preload_lib) {
 202                 ERROR("failed to add preload-lib to env\n");
 203                 return NULL;
 204         }
 205         if (seccomp) {
 206                 snprintf(seccomp_var, sizeof(seccomp_var), "SECCOMP_FILE=%s", seccomp);
 207                 envp[count++] = seccomp_var;
 208                 snprintf(preload_var, sizeof(preload_var), "LD_PRELOAD=%s", preload_lib);
 209                 envp[count++] = preload_var;
 210         }
 211         if (debug > 1)
 212                 envp[count++] = debug_var;
 213
 214         return envp;
 215 }
 216
 217 static void usage(void)
 218 {
 219         fprintf(stderr, "ujail <options> -- <binary> <params ...>\n");
 220         fprintf(stderr, "  -d <num>\tshow debug log (increase num to increase verbosity)\n");
 221         fprintf(stderr, "  -S <file>\tseccomp filter config\n");
 222         fprintf(stderr, "  -C <file>\tcapabilities drop config\n");
 223         fprintf(stderr, "  -c\t\tset PR_SET_NO_NEW_PRIVS\n");
 224         fprintf(stderr, "  -n <name>\tthe name of the jail\n");
 225         fprintf(stderr, "namespace jail options:\n");
 226         fprintf(stderr, "  -h <hostname>\tchange the hostname of the jail\n");
 227         fprintf(stderr, "  -r <file>\treadonly files that should be staged\n");
 228         fprintf(stderr, "  -w <file>\twriteable files that should be staged\n");
 229         fprintf(stderr, "  -p\t\tjail has /proc\n");
 230         fprintf(stderr, "  -s\t\tjail has /sys\n");
 231         fprintf(stderr, "  -l\t\tjail has /dev/log\n");
 232         fprintf(stderr, "  -u\t\tjail has a ubus socket\n");
 233         fprintf(stderr, "  -U <name>\tuser to run jailed process\n");
 234         fprintf(stderr, "  -G <name>\tgroup to run jailed process\n");
 235         fprintf(stderr, "  -o\t\tremont jail root (/) read only\n");
 236         fprintf(stderr, "\nWarning: by default root inside the jail is the same\n\
 237 and he has the same powers as root outside the jail,\n\
 238 thus he can escape the jail and/or break stuff.\n\
 239 Please use seccomp/capabilities (-S/-C) to restrict his powers\n\n\
 240 If you use none of the namespace jail options,\n\
 241 ujail will not use namespace/build a jail,\n\
 242 and will only drop capabilities/apply seccomp filter.\n\n");
 243 }
 244
 245 static int exec_jail(void *_notused)
 246 {
 247         struct passwd *p = NULL;
 248         struct group *g = NULL;
 249
 250         if (opts.capabilities && drop_capabilities(opts.capabilities))
 251                 exit(EXIT_FAILURE);
 252
 253         if (opts.no_new_privs && prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
 254                 ERROR("prctl(PR_SET_NO_NEW_PRIVS) failed: %m\n");
 255                 exit(EXIT_FAILURE);
 256         }
 257
 258         if (opts.namespace && opts.hostname && strlen(opts.hostname) > 0
 259                         && sethostname(opts.hostname, strlen(opts.hostname))) {
 260                 ERROR("sethostname(%s) failed: %m\n", opts.hostname);
 261                 exit(EXIT_FAILURE);
 262         }
 263
 264         if (opts.namespace && build_jail_fs()) {
 265                 ERROR("failed to build jail fs\n");
 266                 exit(EXIT_FAILURE);
 267         }
 268
 269         if (opts.user) {
 270                 p = getpwnam(opts.user);
 271                 if (!p) {
 272                         ERROR("failed to get uid/gid for user %s: %d (%s)\n",
 273                               opts.user, errno, strerror(errno));
 274                         exit(EXIT_FAILURE);
 275                 }
 276         }
 277
 278         if (opts.group) {
 279                 g = getgrnam(opts.group);
 280                 if (!g) {
 281                         ERROR("failed to get gid for group %s: %m\n", opts.group);
 282                         exit(EXIT_FAILURE);
 283                 }
 284         }
 285
 286         if (p && p->pw_gid && initgroups(opts.user, p->pw_gid)) {
 287                 ERROR("failed to initgroups() for user %s: %m\n", opts.user);
 288                 exit(EXIT_FAILURE);
 289         }
 290
 291         if (g && g->gr_gid && setgid(g->gr_gid)) {
 292                 ERROR("failed to set group id %d: %m\n", g?g->gr_gid:p->pw_gid);
 293                 exit(EXIT_FAILURE);
 294         }
 295
 296         if (p && p->pw_uid && setuid(p->pw_uid)) {
 297                 ERROR("failed to set user id %d: %m\n", p->pw_uid);
 298                 exit(EXIT_FAILURE);
 299         }
 300
 301
 302         char **envp = build_envp(opts.seccomp);
 303         if (!envp)
 304                 exit(EXIT_FAILURE);
 305
 306         INFO("exec-ing %s\n", *opts.jail_argv);
 307         execve(*opts.jail_argv, opts.jail_argv, envp);
 308         /* we get there only if execve fails */
 309         ERROR("failed to execve %s: %m\n", *opts.jail_argv);
 310         exit(EXIT_FAILURE);
 311 }
 312
 313 static int jail_running = 1;
 314 static int jail_return_code = 0;
 315
 316 static void jail_process_timeout_cb(struct uloop_timeout *t);
 317 static struct uloop_timeout jail_process_timeout = {
 318         .cb = jail_process_timeout_cb,
 319 };
 320
 321 static void jail_process_handler(struct uloop_process *c, int ret)
 322 {
 323         uloop_timeout_cancel(&jail_process_timeout);
 324         if (WIFEXITED(ret)) {
 325                 jail_return_code = WEXITSTATUS(ret);
 326                 INFO("jail (%d) exited with exit: %d\n", c->pid, jail_return_code);
 327         } else {
 328                 jail_return_code = WTERMSIG(ret);
 329                 INFO("jail (%d) exited with signal: %d\n", c->pid, jail_return_code);
 330         }
 331         jail_running = 0;
 332         uloop_end();
 333 }
 334
 335 static struct uloop_process jail_process = {
 336         .cb = jail_process_handler,
 337 };
 338
 339 static void jail_process_timeout_cb(struct uloop_timeout *t)
 340 {
 341         DEBUG("jail process failed to stop, sending SIGKILL\n");
 342         kill(jail_process.pid, SIGKILL);
 343 }
 344
 345 static void jail_handle_signal(int signo)
 346 {
 347         DEBUG("forwarding signal %d to the jailed process\n", signo);
 348         kill(jail_process.pid, signo);
 349 }
 350
 351 int main(int argc, char **argv)
 352 {
 353         sigset_t sigmask;
 354         uid_t uid = getuid();
 355         char log[] = "/dev/log";
 356         char ubus[] = "/var/run/ubus.sock";
 357         int ch, i;
 358
 359         if (uid) {
 360                 ERROR("not root, aborting: %m\n");
 361                 return EXIT_FAILURE;
 362         }
 363
 364         umask(022);
 365         mount_list_init();
 366         init_library_search();
 367
 368         while ((ch = getopt(argc, argv, OPT_ARGS)) != -1) {
 369                 switch (ch) {
 370                 case 'd':
 371                         debug = atoi(optarg);
 372                         break;
 373                 case 'p':
 374                         opts.namespace = 1;
 375                         opts.procfs = 1;
 376                         break;
 377                 case 'o':
 378                         opts.namespace = 1;
 379                         opts.ronly = 1;
 380                         break;
 381                 case 's':
 382                         opts.namespace = 1;
 383                         opts.sysfs = 1;
 384                         break;
 385                 case 'S':
 386                         opts.seccomp = optarg;
 387                         add_mount(optarg, 1, -1);
 388                         break;
 389                 case 'C':
 390                         opts.capabilities = optarg;
 391                         break;
 392                 case 'c':
 393                         opts.no_new_privs = 1;
 394                         break;
 395                 case 'n':
 396                         opts.name = optarg;
 397                         break;
 398                 case 'h':
 399                         opts.hostname = optarg;
 400                         break;
 401                 case 'r':
 402                         opts.namespace = 1;
 403                         add_path_and_deps(optarg, 1, 0, 0);
 404                         break;
 405                 case 'w':
 406                         opts.namespace = 1;
 407                         add_path_and_deps(optarg, 0, 0, 0);
 408                         break;
 409                 case 'u':
 410                         opts.namespace = 1;
 411                         add_mount(ubus, 0, -1);
 412                         break;
 413                 case 'l':
 414                         opts.namespace = 1;
 415                         add_mount(log, 0, -1);
 416                         break;
 417                 case 'U':
 418                         opts.user = optarg;
 419                         break;
 420                 case 'G':
 421                         opts.group = optarg;
 422                         break;
 423                 }
 424         }
 425
 426         /* no <binary> param found */
 427         if (argc - optind < 1) {
 428                 usage();
 429                 return EXIT_FAILURE;
 430         }
 431         if (!(opts.namespace||opts.capabilities||opts.seccomp)) {
 432                 ERROR("Not using namespaces, capabilities or seccomp !!!\n\n");
 433                 usage();
 434                 return EXIT_FAILURE;
 435         }
 436         DEBUG("Using namespaces(%d), capabilities(%d), seccomp(%d)\n",
 437                 opts.namespace,
 438                 opts.capabilities != 0,
 439                 opts.seccomp != 0);
 440
 441         opts.jail_argv = &argv[optind];
 442
 443         if (opts.namespace && add_path_and_deps(*opts.jail_argv, 1, -1, 0)) {
 444                 ERROR("failed to load dependencies\n");
 445                 return -1;
 446         }
 447
 448         if (opts.namespace && opts.seccomp && add_path_and_deps("libpreload-seccomp.so", 1, -1, 1)) {
 449                 ERROR("failed to load libpreload-seccomp.so\n");
 450                 return -1;
 451         }
 452
 453         if (opts.name)
 454                 prctl(PR_SET_NAME, opts.name, NULL, NULL, NULL);
 455
 456         uloop_init();
 457
 458         sigfillset(&sigmask);
 459         for (i = 0; i < _NSIG; i++) {
 460                 struct sigaction s = { 0 };
 461
 462                 if (!sigismember(&sigmask, i))
 463                         continue;
 464                 if ((i == SIGCHLD) || (i == SIGPIPE))
 465                         continue;
 466
 467                 s.sa_handler = jail_handle_signal;
 468                 sigaction(i, &s, NULL);
 469         }
 470
 471         if (opts.namespace) {
 472                 add_mount("/dev/full", 0, -1);
 473                 add_mount("/dev/null", 0, -1);
 474                 add_mount("/dev/urandom", 0, -1);
 475                 add_mount("/dev/zero", 0, -1);
 476
 477                 if (opts.user || opts.group) {
 478                         add_mount("/etc/passwd", 0, -1);
 479                         add_mount("/etc/group", 0, -1);
 480                 }
 481
 482                 if (opts.namespace & NAMESPACE_IPC)
 483                         flags |= CLONE_NEWIPC;
 484
 485                 int flags = CLONE_NEWPID | CLONE_NEWNS | CLONE_NEWIPC | SIGCHLD;
 486                 if (opts.hostname)
 487                         flags |= CLONE_NEWUTS;
 488                 jail_process.pid = clone(exec_jail, child_stack + STACK_SIZE, flags, NULL);
 489         } else {
 490                 jail_process.pid = fork();
 491         }
 492
 493         if (jail_process.pid > 0) {
 494                 /* parent process */
 495                 uloop_process_add(&jail_process);
 496                 uloop_run();
 497                 if (jail_running) {
 498                         DEBUG("uloop interrupted, killing jail process\n");
 499                         kill(jail_process.pid, SIGTERM);
 500                         uloop_timeout_set(&jail_process_timeout, 1000);
 501                         uloop_run();
 502                 }
 503                 uloop_done();
 504                 return jail_return_code;
 505         } else if (jail_process.pid == 0) {
 506                 /* fork child process */
 507                 return exec_jail(NULL);
 508         } else {
 509                 ERROR("failed to clone/fork: %m\n");
 510                 return EXIT_FAILURE;
 511         }
 512 }