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 (mount_all(jail_root)) {
 143                 ERROR("mount_all() failed\n");
 144                 return -1;
 145         }
 146
 147         char dirbuf[sizeof(jail_root) + 4];
 148         snprintf(dirbuf, sizeof(dirbuf), "%s/old", jail_root);
 149         mkdir(dirbuf, 0755);
 150
 151         if (pivot_root(jail_root, dirbuf) == -1) {
 152                 ERROR("pivot_root failed: %s\n", strerror(errno));
 153                 return -1;
 154         }
 155         if (chdir("/")) {
 156                 ERROR("chdir(/) failed: %s\n", strerror(errno));
 157                 return -1;
 158         }
 159
 160         snprintf(dirbuf, sizeof(dirbuf), "/old%s", jail_root);
 161         rmdir(dirbuf);
 162         umount2("/old", MNT_DETACH);
 163         rmdir("/old");
 164
 165         if (opts.procfs) {
 166                 mkdir("/proc", 0755);
 167                 mount("proc", "/proc", "proc", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
 168         }
 169         if (opts.sysfs) {
 170                 mkdir("/sys", 0755);
 171                 mount("sysfs", "/sys", "sysfs", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
 172         }
 173         if (opts.ronly)
 174                 mount(NULL, "/", NULL, MS_RDONLY | MS_REMOUNT, 0);
 175
 176         return 0;
 177 }
 178
 179 #define MAX_ENVP        8
 180 static char** build_envp(const char *seccomp)
 181 {
 182         static char *envp[MAX_ENVP];
 183         static char preload_var[PATH_MAX];
 184         static char seccomp_var[PATH_MAX];
 185         static char debug_var[] = "LD_DEBUG=all";
 186         const char *preload_lib = find_lib("libpreload-seccomp.so");
 187         int count = 0;
 188
 189         if (seccomp && !preload_lib) {
 190                 ERROR("failed to add preload-lib to env\n");
 191                 return NULL;
 192         }
 193         if (seccomp) {
 194                 snprintf(seccomp_var, sizeof(seccomp_var), "SECCOMP_FILE=%s", seccomp);
 195                 envp[count++] = seccomp_var;
 196                 snprintf(preload_var, sizeof(preload_var), "LD_PRELOAD=%s", preload_lib);
 197                 envp[count++] = preload_var;
 198         }
 199         if (debug > 1)
 200                 envp[count++] = debug_var;
 201
 202         return envp;
 203 }
 204
 205 static void usage(void)
 206 {
 207         fprintf(stderr, "ujail <options> -- <binary> <params ...>\n");
 208         fprintf(stderr, "  -d <num>\tshow debug log (increase num to increase verbosity)\n");
 209         fprintf(stderr, "  -S <file>\tseccomp filter config\n");
 210         fprintf(stderr, "  -C <file>\tcapabilities drop config\n");
 211         fprintf(stderr, "  -c\t\tset PR_SET_NO_NEW_PRIVS\n");
 212         fprintf(stderr, "  -n <name>\tthe name of the jail\n");
 213         fprintf(stderr, "namespace jail options:\n");
 214         fprintf(stderr, "  -h <hostname>\tchange the hostname of the jail\n");
 215         fprintf(stderr, "  -r <file>\treadonly files that should be staged\n");
 216         fprintf(stderr, "  -w <file>\twriteable files that should be staged\n");
 217         fprintf(stderr, "  -p\t\tjail has /proc\n");
 218         fprintf(stderr, "  -s\t\tjail has /sys\n");
 219         fprintf(stderr, "  -l\t\tjail has /dev/log\n");
 220         fprintf(stderr, "  -u\t\tjail has a ubus socket\n");
 221         fprintf(stderr, "  -o\t\tremont jail root (/) read only\n");
 222         fprintf(stderr, "\nWarning: by default root inside the jail is the same\n\
 223 and he has the same powers as root outside the jail,\n\
 224 thus he can escape the jail and/or break stuff.\n\
 225 Please use seccomp/capabilities (-S/-C) to restrict his powers\n\n\
 226 If you use none of the namespace jail options,\n\
 227 ujail will not use namespace/build a jail,\n\
 228 and will only drop capabilities/apply seccomp filter.\n\n");
 229 }
 230
 231 static int exec_jail(void *_notused)
 232 {
 233         if (opts.capabilities && drop_capabilities(opts.capabilities))
 234                 exit(EXIT_FAILURE);
 235
 236         if (opts.no_new_privs && prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
 237                 ERROR("prctl(PR_SET_NO_NEW_PRIVS) failed: %s\n", strerror(errno));
 238                 exit(EXIT_FAILURE);
 239         }
 240
 241         if (opts.namespace && opts.hostname
 242                         && sethostname(opts.hostname, strlen(opts.hostname))) {
 243                 ERROR("sethostname(%s) failed: %s\n", opts.hostname, strerror(errno));
 244                 exit(EXIT_FAILURE);
 245         }
 246
 247         if (opts.namespace && build_jail_fs()) {
 248                 ERROR("failed to build jail fs\n");
 249                 exit(EXIT_FAILURE);
 250         }
 251
 252         char **envp = build_envp(opts.seccomp);
 253         if (!envp)
 254                 exit(EXIT_FAILURE);
 255
 256         INFO("exec-ing %s\n", *opts.jail_argv);
 257         execve(*opts.jail_argv, opts.jail_argv, envp);
 258         /* we get there only if execve fails */
 259         ERROR("failed to execve %s: %s\n", *opts.jail_argv, strerror(errno));
 260         exit(EXIT_FAILURE);
 261 }
 262
 263 static int jail_running = 1;
 264 static int jail_return_code = 0;
 265
 266 static void jail_process_handler(struct uloop_process *c, int ret)
 267 {
 268         if (WIFEXITED(ret)) {
 269                 jail_return_code = WEXITSTATUS(ret);
 270                 INFO("jail (%d) exited with exit: %d\n", c->pid, jail_return_code);
 271         } else {
 272                 jail_return_code = WTERMSIG(ret);
 273                 INFO("jail (%d) exited with signal: %d\n", c->pid, jail_return_code);
 274         }
 275         jail_running = 0;
 276         uloop_end();
 277 }
 278
 279 static struct uloop_process jail_process = {
 280         .cb = jail_process_handler,
 281 };
 282
 283 int main(int argc, char **argv)
 284 {
 285         uid_t uid = getuid();
 286         char log[] = "/dev/log";
 287         char ubus[] = "/var/run/ubus.sock";
 288         int ch;
 289
 290         if (uid) {
 291                 ERROR("not root, aborting: %s\n", strerror(errno));
 292                 return EXIT_FAILURE;
 293         }
 294
 295         umask(022);
 296         mount_list_init();
 297         init_library_search();
 298
 299         while ((ch = getopt(argc, argv, OPT_ARGS)) != -1) {
 300                 switch (ch) {
 301                 case 'd':
 302                         debug = atoi(optarg);
 303                         break;
 304                 case 'p':
 305                         opts.namespace = 1;
 306                         opts.procfs = 1;
 307                         break;
 308                 case 'o':
 309                         opts.namespace = 1;
 310                         opts.ronly = 1;
 311                         break;
 312                 case 's':
 313                         opts.namespace = 1;
 314                         opts.sysfs = 1;
 315                         break;
 316                 case 'S':
 317                         opts.seccomp = optarg;
 318                         add_mount(optarg, 1, -1);
 319                         break;
 320                 case 'C':
 321                         opts.capabilities = optarg;
 322                         break;
 323                 case 'c':
 324                         opts.no_new_privs = 1;
 325                         break;
 326                 case 'n':
 327                         opts.name = optarg;
 328                         break;
 329                 case 'h':
 330                         opts.hostname = optarg;
 331                         break;
 332                 case 'r':
 333                         opts.namespace = 1;
 334                         add_path_and_deps(optarg, 1, 0, 0);
 335                         break;
 336                 case 'w':
 337                         opts.namespace = 1;
 338                         add_path_and_deps(optarg, 0, 0, 0);
 339                         break;
 340                 case 'u':
 341                         opts.namespace = 1;
 342                         add_mount(ubus, 0, -1);
 343                         break;
 344                 case 'l':
 345                         opts.namespace = 1;
 346                         add_mount(log, 0, -1);
 347                         break;
 348                 }
 349         }
 350
 351         /* no <binary> param found */
 352         if (argc - optind < 1) {
 353                 usage();
 354                 return EXIT_FAILURE;
 355         }
 356         if (!(opts.namespace||opts.capabilities||opts.seccomp)) {
 357                 ERROR("Not using namespaces, capabilities or seccomp !!!\n\n");
 358                 usage();
 359                 return EXIT_FAILURE;
 360         }
 361         DEBUG("Using namespaces(%d), capabilities(%d), seccomp(%d)\n",
 362                 opts.namespace,
 363                 opts.capabilities != 0,
 364                 opts.seccomp != 0);
 365
 366         opts.jail_argv = &argv[optind];
 367
 368         if (opts.namespace && add_path_and_deps(*opts.jail_argv, 1, -1, 0)) {
 369                 ERROR("failed to load dependencies\n");
 370                 return -1;
 371         }
 372
 373         if (opts.namespace && opts.seccomp && add_path_and_deps("libpreload-seccomp.so", 1, -1, 1)) {
 374                 ERROR("failed to load libpreload-seccomp.so\n");
 375                 return -1;
 376         }
 377
 378         if (opts.name)
 379                 prctl(PR_SET_NAME, opts.name, NULL, NULL, NULL);
 380
 381         uloop_init();
 382         if (opts.namespace) {
 383                 jail_process.pid = clone(exec_jail,
 384                         child_stack + STACK_SIZE,
 385                         CLONE_NEWUTS | CLONE_NEWPID | CLONE_NEWNS | CLONE_NEWIPC | SIGCHLD, NULL);
 386         } else {
 387                 jail_process.pid = fork();
 388         }
 389
 390         if (jail_process.pid > 0) {
 391                 /* parent process */
 392                 uloop_process_add(&jail_process);
 393                 uloop_run();
 394                 uloop_done();
 395                 if (jail_running) {
 396                         DEBUG("uloop interrupted, killing jail process\n");
 397                         kill(jail_process.pid, SIGTERM);
 398                         waitpid(jail_process.pid, NULL, 0);
 399                 }
 400                 return jail_return_code;
 401         } else if (jail_process.pid == 0) {
 402                 /* fork child process */
 403                 return exec_jail(NULL);
 404         } else {
 405                 ERROR("failed to clone/fork: %s\n", strerror(errno));
 406                 return EXIT_FAILURE;
 407         }
 408 }