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
  19 #include <stdlib.h>
  20 #include <unistd.h>
  21 #include <errno.h>
  22 #include <string.h>
  23 #include <sys/stat.h>
  24 #include <fcntl.h>
  25 #include <libgen.h>
  26 #include <sched.h>
  27 #include <linux/limits.h>
  28 #include <signal.h>
  29
  30 #include "capabilities.h"
  31 #include "elf.h"
  32 #include "fs.h"
  33 #include "jail.h"
  34 #include "log.h"
  35
  36 #include <libubox/uloop.h>
  37
  38 #define STACK_SIZE      (1024 * 1024)
  39 #define OPT_ARGS        "S:C:n:h:r:w:d:psuloc"
  40
  41 static struct {
  42         char *name;
  43         char *hostname;
  44         char **jail_argv;
  45         char *seccomp;
  46         char *capabilities;
  47         int no_new_privs;
  48         int namespace;
  49         int procfs;
  50         int ronly;
  51         int sysfs;
  52 } opts;
  53
  54 extern int pivot_root(const char *new_root, const char *put_old);
  55
  56 int debug = 0;
  57
  58 static char child_stack[STACK_SIZE];
  59
  60 static int mkdir_p(char *dir, mode_t mask)
  61 {
  62         char *l = strrchr(dir, '/');
  63         int ret;
  64
  65         if (!l)
  66                 return 0;
  67
  68         *l = '\0';
  69
  70         if (mkdir_p(dir, mask))
  71                 return -1;
  72
  73         *l = '/';
  74
  75         ret = mkdir(dir, mask);
  76         if (ret && errno == EEXIST)
  77                 return 0;
  78
  79         if (ret)
  80                 ERROR("mkdir(%s, %d) failed: %s\n", dir, mask, strerror(errno));
  81
  82         return ret;
  83 }
  84
  85 int mount_bind(const char *root, const char *path, int readonly, int error)
  86 {
  87         struct stat s;
  88         char new[PATH_MAX];
  89         int fd;
  90
  91         if (stat(path, &s)) {
  92                 ERROR("stat(%s) failed: %s\n", path, strerror(errno));
  93                 return error;
  94         }
  95
  96         snprintf(new, sizeof(new), "%s%s", root, path);
  97         if (S_ISDIR(s.st_mode)) {
  98                 mkdir_p(new, 0755);
  99         } else {
 100                 mkdir_p(dirname(new), 0755);
 101                 snprintf(new, sizeof(new), "%s%s", root, path);
 102                 fd = creat(new, 0644);
 103                 if (fd == -1) {
 104                         ERROR("creat(%s) failed: %s\n", new, strerror(errno));
 105                         return -1;
 106                 }
 107                 close(fd);
 108         }
 109
 110         if (mount(path, new, NULL, MS_BIND, NULL)) {
 111                 ERROR("failed to mount -B %s %s: %s\n", path, new, strerror(errno));
 112                 return -1;
 113         }
 114
 115         if (readonly && mount(NULL, new, NULL, MS_BIND | MS_REMOUNT | MS_RDONLY, NULL)) {
 116                 ERROR("failed to remount ro %s: %s\n", new, strerror(errno));
 117                 return -1;
 118         }
 119
 120         DEBUG("mount -B %s %s (%s)\n", path, new, readonly?"ro":"rw");
 121
 122         return 0;
 123 }
 124
 125 static int build_jail_fs(void)
 126 {
 127         char jail_root[] = "/tmp/ujail-XXXXXX";
 128         if (mkdtemp(jail_root) == NULL) {
 129                 ERROR("mkdtemp(%s) failed: %s\n", jail_root, strerror(errno));
 130                 return -1;
 131         }
 132
 133         /* oldroot can't be MS_SHARED else pivot_root() fails */
 134         if (mount("none", "/", NULL, MS_REC|MS_PRIVATE, NULL)) {
 135                 ERROR("private mount failed %s\n", strerror(errno));
 136                 return -1;
 137         }
 138
 139         if (mount("tmpfs", jail_root, "tmpfs", MS_NOATIME, "mode=0755")) {
 140                 ERROR("tmpfs mount failed %s\n", strerror(errno));
 141                 return -1;
 142         }
 143
 144         if (chdir(jail_root)) {
 145                 ERROR("chdir(%s) (jail_root) failed: %s\n", jail_root, strerror(errno));
 146                 return -1;
 147         }
 148
 149         if (mount_all(jail_root)) {
 150                 ERROR("mount_all() failed\n");
 151                 return -1;
 152         }
 153
 154         char dirbuf[sizeof(jail_root) + 4];
 155         snprintf(dirbuf, sizeof(dirbuf), "%s/old", jail_root);
 156         mkdir(dirbuf, 0755);
 157
 158         if (pivot_root(jail_root, dirbuf) == -1) {
 159                 ERROR("pivot_root(%s, %s) failed: %s\n", jail_root, dirbuf, strerror(errno));
 160                 return -1;
 161         }
 162         if (chdir("/")) {
 163                 ERROR("chdir(/) (after pivot_root) failed: %s\n", strerror(errno));
 164                 return -1;
 165         }
 166
 167         snprintf(dirbuf, sizeof(dirbuf), "/old%s", jail_root);
 168         rmdir(dirbuf);
 169         umount2("/old", MNT_DETACH);
 170         rmdir("/old");
 171
 172         if (opts.procfs) {
 173                 mkdir("/proc", 0755);
 174                 mount("proc", "/proc", "proc", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
 175         }
 176         if (opts.sysfs) {
 177                 mkdir("/sys", 0755);
 178                 mount("sysfs", "/sys", "sysfs", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
 179         }
 180         if (opts.ronly)
 181                 mount(NULL, "/", NULL, MS_RDONLY | MS_REMOUNT, 0);
 182
 183         return 0;
 184 }
 185
 186 #define MAX_ENVP        8
 187 static char** build_envp(const char *seccomp)
 188 {
 189         static char *envp[MAX_ENVP];
 190         static char preload_var[PATH_MAX];
 191         static char seccomp_var[PATH_MAX];
 192         static char debug_var[] = "LD_DEBUG=all";
 193         const char *preload_lib = find_lib("libpreload-seccomp.so");
 194         int count = 0;
 195
 196         if (seccomp && !preload_lib) {
 197                 ERROR("failed to add preload-lib to env\n");
 198                 return NULL;
 199         }
 200         if (seccomp) {
 201                 snprintf(seccomp_var, sizeof(seccomp_var), "SECCOMP_FILE=%s", seccomp);
 202                 envp[count++] = seccomp_var;
 203                 snprintf(preload_var, sizeof(preload_var), "LD_PRELOAD=%s", preload_lib);
 204                 envp[count++] = preload_var;
 205         }
 206         if (debug > 1)
 207                 envp[count++] = debug_var;
 208
 209         return envp;
 210 }
 211
 212 static void usage(void)
 213 {
 214         fprintf(stderr, "ujail <options> -- <binary> <params ...>\n");
 215         fprintf(stderr, "  -d <num>\tshow debug log (increase num to increase verbosity)\n");
 216         fprintf(stderr, "  -S <file>\tseccomp filter config\n");
 217         fprintf(stderr, "  -C <file>\tcapabilities drop config\n");
 218         fprintf(stderr, "  -c\t\tset PR_SET_NO_NEW_PRIVS\n");
 219         fprintf(stderr, "  -n <name>\tthe name of the jail\n");
 220         fprintf(stderr, "namespace jail options:\n");
 221         fprintf(stderr, "  -h <hostname>\tchange the hostname of the jail\n");
 222         fprintf(stderr, "  -r <file>\treadonly files that should be staged\n");
 223         fprintf(stderr, "  -w <file>\twriteable files that should be staged\n");
 224         fprintf(stderr, "  -p\t\tjail has /proc\n");
 225         fprintf(stderr, "  -s\t\tjail has /sys\n");
 226         fprintf(stderr, "  -l\t\tjail has /dev/log\n");
 227         fprintf(stderr, "  -u\t\tjail has a ubus socket\n");
 228         fprintf(stderr, "  -o\t\tremont jail root (/) read only\n");
 229         fprintf(stderr, "\nWarning: by default root inside the jail is the same\n\
 230 and he has the same powers as root outside the jail,\n\
 231 thus he can escape the jail and/or break stuff.\n\
 232 Please use seccomp/capabilities (-S/-C) to restrict his powers\n\n\
 233 If you use none of the namespace jail options,\n\
 234 ujail will not use namespace/build a jail,\n\
 235 and will only drop capabilities/apply seccomp filter.\n\n");
 236 }
 237
 238 static int exec_jail(void *_notused)
 239 {
 240         if (opts.capabilities && drop_capabilities(opts.capabilities))
 241                 exit(EXIT_FAILURE);
 242
 243         if (opts.no_new_privs && prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
 244                 ERROR("prctl(PR_SET_NO_NEW_PRIVS) failed: %s\n", strerror(errno));
 245                 exit(EXIT_FAILURE);
 246         }
 247
 248         if (opts.namespace && opts.hostname && strlen(opts.hostname) > 0
 249                         && sethostname(opts.hostname, strlen(opts.hostname))) {
 250                 ERROR("sethostname(%s) failed: %s\n", opts.hostname, strerror(errno));
 251                 exit(EXIT_FAILURE);
 252         }
 253
 254         if (opts.namespace && build_jail_fs()) {
 255                 ERROR("failed to build jail fs\n");
 256                 exit(EXIT_FAILURE);
 257         }
 258
 259         char **envp = build_envp(opts.seccomp);
 260         if (!envp)
 261                 exit(EXIT_FAILURE);
 262
 263         INFO("exec-ing %s\n", *opts.jail_argv);
 264         execve(*opts.jail_argv, opts.jail_argv, envp);
 265         /* we get there only if execve fails */
 266         ERROR("failed to execve %s: %s\n", *opts.jail_argv, strerror(errno));
 267         exit(EXIT_FAILURE);
 268 }
 269
 270 static int jail_running = 1;
 271 static int jail_return_code = 0;
 272
 273 static void jail_process_timeout_cb(struct uloop_timeout *t);
 274 static struct uloop_timeout jail_process_timeout = {
 275         .cb = jail_process_timeout_cb,
 276 };
 277
 278 static void jail_process_handler(struct uloop_process *c, int ret)
 279 {
 280         uloop_timeout_cancel(&jail_process_timeout);
 281         if (WIFEXITED(ret)) {
 282                 jail_return_code = WEXITSTATUS(ret);
 283                 INFO("jail (%d) exited with exit: %d\n", c->pid, jail_return_code);
 284         } else {
 285                 jail_return_code = WTERMSIG(ret);
 286                 INFO("jail (%d) exited with signal: %d\n", c->pid, jail_return_code);
 287         }
 288         jail_running = 0;
 289         uloop_end();
 290 }
 291
 292 static struct uloop_process jail_process = {
 293         .cb = jail_process_handler,
 294 };
 295
 296 static void jail_process_timeout_cb(struct uloop_timeout *t)
 297 {
 298         DEBUG("jail process failed to stop, sending SIGKILL\n");
 299         kill(jail_process.pid, SIGKILL);
 300 }
 301
 302 static void jail_handle_signal(int signo)
 303 {
 304         DEBUG("forwarding signal %d to the jailed process\n", signo);
 305         kill(jail_process.pid, signo);
 306 }
 307
 308 int main(int argc, char **argv)
 309 {
 310         sigset_t sigmask;
 311         uid_t uid = getuid();
 312         char log[] = "/dev/log";
 313         char ubus[] = "/var/run/ubus.sock";
 314         int ch, i;
 315
 316         if (uid) {
 317                 ERROR("not root, aborting: %s\n", strerror(errno));
 318                 return EXIT_FAILURE;
 319         }
 320
 321         umask(022);
 322         mount_list_init();
 323         init_library_search();
 324
 325         while ((ch = getopt(argc, argv, OPT_ARGS)) != -1) {
 326                 switch (ch) {
 327                 case 'd':
 328                         debug = atoi(optarg);
 329                         break;
 330                 case 'p':
 331                         opts.namespace = 1;
 332                         opts.procfs = 1;
 333                         break;
 334                 case 'o':
 335                         opts.namespace = 1;
 336                         opts.ronly = 1;
 337                         break;
 338                 case 's':
 339                         opts.namespace = 1;
 340                         opts.sysfs = 1;
 341                         break;
 342                 case 'S':
 343                         opts.seccomp = optarg;
 344                         add_mount(optarg, 1, -1);
 345                         break;
 346                 case 'C':
 347                         opts.capabilities = optarg;
 348                         break;
 349                 case 'c':
 350                         opts.no_new_privs = 1;
 351                         break;
 352                 case 'n':
 353                         opts.name = optarg;
 354                         break;
 355                 case 'h':
 356                         opts.hostname = optarg;
 357                         break;
 358                 case 'r':
 359                         opts.namespace = 1;
 360                         add_path_and_deps(optarg, 1, 0, 0);
 361                         break;
 362                 case 'w':
 363                         opts.namespace = 1;
 364                         add_path_and_deps(optarg, 0, 0, 0);
 365                         break;
 366                 case 'u':
 367                         opts.namespace = 1;
 368                         add_mount(ubus, 0, -1);
 369                         break;
 370                 case 'l':
 371                         opts.namespace = 1;
 372                         add_mount(log, 0, -1);
 373                         break;
 374                 }
 375         }
 376
 377         /* no <binary> param found */
 378         if (argc - optind < 1) {
 379                 usage();
 380                 return EXIT_FAILURE;
 381         }
 382         if (!(opts.namespace||opts.capabilities||opts.seccomp)) {
 383                 ERROR("Not using namespaces, capabilities or seccomp !!!\n\n");
 384                 usage();
 385                 return EXIT_FAILURE;
 386         }
 387         DEBUG("Using namespaces(%d), capabilities(%d), seccomp(%d)\n",
 388                 opts.namespace,
 389                 opts.capabilities != 0,
 390                 opts.seccomp != 0);
 391
 392         opts.jail_argv = &argv[optind];
 393
 394         if (opts.namespace && add_path_and_deps(*opts.jail_argv, 1, -1, 0)) {
 395                 ERROR("failed to load dependencies\n");
 396                 return -1;
 397         }
 398
 399         if (opts.namespace && opts.seccomp && add_path_and_deps("libpreload-seccomp.so", 1, -1, 1)) {
 400                 ERROR("failed to load libpreload-seccomp.so\n");
 401                 return -1;
 402         }
 403
 404         if (opts.name)
 405                 prctl(PR_SET_NAME, opts.name, NULL, NULL, NULL);
 406
 407         uloop_init();
 408
 409         sigfillset(&sigmask);
 410         for (i = 0; i < _NSIG; i++) {
 411                 struct sigaction s = { 0 };
 412
 413                 if (!sigismember(&sigmask, i))
 414                         continue;
 415                 if ((i == SIGCHLD) || (i == SIGPIPE))
 416                         continue;
 417
 418                 s.sa_handler = jail_handle_signal;
 419                 sigaction(i, &s, NULL);
 420         }
 421
 422         if (opts.namespace) {
 423                 add_mount("/dev/full", 0, -1);
 424                 add_mount("/dev/null", 0, -1);
 425                 add_mount("/dev/urandom", 0, -1);
 426                 add_mount("/dev/zero", 0, -1);
 427
 428                 int flags = CLONE_NEWPID | CLONE_NEWNS | CLONE_NEWIPC | SIGCHLD;
 429                 if (opts.hostname)
 430                         flags |= CLONE_NEWUTS;
 431                 jail_process.pid = clone(exec_jail, child_stack + STACK_SIZE, flags, NULL);
 432         } else {
 433                 jail_process.pid = fork();
 434         }
 435
 436         if (jail_process.pid > 0) {
 437                 /* parent process */
 438                 uloop_process_add(&jail_process);
 439                 uloop_run();
 440                 if (jail_running) {
 441                         DEBUG("uloop interrupted, killing jail process\n");
 442                         kill(jail_process.pid, SIGTERM);
 443                         uloop_timeout_set(&jail_process_timeout, 1000);
 444                         uloop_run();
 445                 }
 446                 uloop_done();
 447                 return jail_return_code;
 448         } else if (jail_process.pid == 0) {
 449                 /* fork child process */
 450                 return exec_jail(NULL);
 451         } else {
 452                 ERROR("failed to clone/fork: %s\n", strerror(errno));
 453                 return EXIT_FAILURE;
 454         }
 455 }