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