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