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