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/syscall.h>
  16 #include <sys/mman.h>
  17 #include <sys/utsname.h>
  18 #include <sys/types.h>
  19 #include <sys/syscall.h>
  20 #include <sys/types.h>
  21 #include <sys/mount.h>
  22 #include <sys/prctl.h>
  23 #include <sys/wait.h>
  24
  25 #include <stdlib.h>
  26 #include <unistd.h>
  27 #include <values.h>
  28 #include <errno.h>
  29 #include <stdio.h>
  30 #include <string.h>
  31 #include <sys/stat.h>
  32 #include <fcntl.h>
  33 #include <syslog.h>
  34 #include <libgen.h>
  35 #include <glob.h>
  36 #include <elf.h>
  37 #include <sched.h>
  38
  39 #include "elf.h"
  40 #include "capabilities.h"
  41
  42 #include <libubox/utils.h>
  43 #include <libubox/list.h>
  44 #include <libubox/uloop.h>
  45
  46 #define STACK_SIZE      (1024 * 1024)
  47 #define OPT_ARGS        "P:S:C:n:r:w:d:psulo"
  48
  49 static struct {
  50         char *path;
  51         char *name;
  52         char **jail_argv;
  53         char *seccomp;
  54         char *capabilities;
  55         int procfs;
  56         int ronly;
  57         int sysfs;
  58 } opts;
  59
  60 struct extra {
  61         struct list_head list;
  62
  63         const char *path;
  64         const char *name;
  65         int readonly;
  66 };
  67
  68 static LIST_HEAD(extras);
  69
  70 extern int pivot_root(const char *new_root, const char *put_old);
  71
  72 int debug = 0;
  73
  74 static char child_stack[STACK_SIZE];
  75
  76 static int mkdir_p(char *dir, mode_t mask)
  77 {
  78         char *l = strrchr(dir, '/');
  79         int ret;
  80
  81         if (!l)
  82                 return 0;
  83
  84         *l = '\0';
  85
  86         if (mkdir_p(dir, mask))
  87                 return -1;
  88
  89         *l = '/';
  90
  91         ret = mkdir(dir, mask);
  92         if (ret && errno == EEXIST)
  93                 return 0;
  94
  95         if (ret)
  96                 ERROR("mkdir failed on %s: %s\n", dir, strerror(errno));
  97
  98         return ret;
  99 }
 100
 101 static int mount_bind(const char *root, const char *path, const char *name, int readonly, int error)
 102 {
 103         const char *p = path;
 104         struct stat s;
 105         char old[256];
 106         char new[256];
 107         int fd;
 108
 109         if (strstr(p, "local"))
 110                 p = "/lib";
 111
 112         snprintf(old, sizeof(old), "%s/%s", path, name);
 113         snprintf(new, sizeof(new), "%s%s", root, p);
 114
 115         mkdir_p(new, 0755);
 116
 117         snprintf(new, sizeof(new), "%s%s/%s", root, p, name);
 118
 119         if (stat(old, &s)) {
 120                 ERROR("%s does not exist\n", old);
 121                 return error;
 122         }
 123
 124         if (S_ISDIR(s.st_mode)) {
 125                 mkdir_p(new, 0755);
 126         } else {
 127                 fd = creat(new, 0644);
 128                 if (fd == -1) {
 129                         ERROR("failed to create %s: %s\n", new, strerror(errno));
 130                         return -1;
 131                 }
 132                 close(fd);
 133         }
 134
 135         if (mount(old, new, NULL, MS_BIND, NULL)) {
 136                 ERROR("failed to mount -B %s %s: %s\n", old, new, strerror(errno));
 137                 return -1;
 138         }
 139
 140         if (readonly && mount(NULL, new, NULL, MS_BIND | MS_REMOUNT | MS_RDONLY, NULL)) {
 141                 ERROR("failed to remount ro %s: %s\n", new, strerror(errno));
 142                 return -1;
 143         }
 144
 145         DEBUG("mount -B %s %s\n", old, new);
 146
 147         return 0;
 148 }
 149
 150 static int build_jail_fs()
 151 {
 152         struct library *l;
 153         struct extra *m;
 154
 155         if (mount("tmpfs", opts.path, "tmpfs", MS_NOATIME, "mode=0755")) {
 156                 ERROR("tmpfs mount failed %s\n", strerror(errno));
 157                 return -1;
 158         }
 159
 160         if (chdir(opts.path)) {
 161                 ERROR("failed to chdir() in the jail root\n");
 162                 return -1;
 163         }
 164
 165         avl_init(&libraries, avl_strcmp, false, NULL);
 166         alloc_library_path("/lib64");
 167         alloc_library_path("/lib");
 168         alloc_library_path("/usr/lib");
 169         load_ldso_conf("/etc/ld.so.conf");
 170
 171         if (elf_load_deps(*opts.jail_argv)) {
 172                 ERROR("failed to load dependencies\n");
 173                 return -1;
 174         }
 175
 176         if (opts.seccomp && elf_load_deps("libpreload-seccomp.so")) {
 177                 ERROR("failed to load libpreload-seccomp.so\n");
 178                 return -1;
 179         }
 180
 181         avl_for_each_element(&libraries, l, avl)
 182                 if (mount_bind(opts.path, l->path, l->name, 1, -1))
 183                         return -1;
 184
 185         list_for_each_entry(m, &extras, list)
 186                 if (mount_bind(opts.path, m->path, m->name, m->readonly, 0))
 187                         return -1;
 188
 189         char *mpoint;
 190         if (asprintf(&mpoint, "%s/old", opts.path) < 0) {
 191                 ERROR("failed to alloc pivot path: %s\n", strerror(errno));
 192                 return -1;
 193         }
 194         mkdir_p(mpoint, 0755);
 195         if (pivot_root(opts.path, mpoint) == -1) {
 196                 ERROR("pivot_root failed:%s\n", strerror(errno));
 197                 free(mpoint);
 198                 return -1;
 199         }
 200         free(mpoint);
 201         umount2("/old", MNT_DETACH);
 202         rmdir("/old");
 203         if (opts.procfs) {
 204                 mkdir("/proc", 0755);
 205                 mount("proc", "/proc", "proc", MS_NOATIME, 0);
 206         }
 207         if (opts.sysfs) {
 208                 mkdir("/sys", 0755);
 209                 mount("sysfs", "/sys", "sysfs", MS_NOATIME, 0);
 210         }
 211         if (opts.ronly)
 212                 mount(NULL, "/", NULL, MS_RDONLY | MS_REMOUNT, 0);
 213
 214         return 0;
 215 }
 216
 217 #define MAX_ENVP        8
 218 static char** build_envp(const char *seccomp)
 219 {
 220         static char *envp[MAX_ENVP];
 221         static char preload_var[64];
 222         static char seccomp_var[64];
 223         static char debug_var[] = "LD_DEBUG=all";
 224         char *preload_lib = find_lib("libpreload-seccomp.so");
 225         int count = 0;
 226
 227         if (seccomp && !preload_lib) {
 228                 ERROR("failed to add preload-lib to env\n");
 229                 return NULL;
 230         }
 231         if (seccomp) {
 232                 snprintf(seccomp_var, sizeof(seccomp_var), "SECCOMP_FILE=%s", seccomp);
 233                 envp[count++] = seccomp_var;
 234                 snprintf(preload_var, sizeof(preload_var), "LD_PRELOAD=%s", preload_lib);
 235                 envp[count++] = preload_var;
 236         }
 237         if (debug > 1)
 238                 envp[count++] = debug_var;
 239
 240         return envp;
 241 }
 242
 243 static void usage(void)
 244 {
 245         fprintf(stderr, "ujail <options> -- <binary> <params ...>\n");
 246         fprintf(stderr, "  -P <path>\tpath where the jail will be staged\n");
 247         fprintf(stderr, "  -S <file>\tseccomp filter\n");
 248         fprintf(stderr, "  -C <file>\tcapabilities drop config\n");
 249         fprintf(stderr, "  -n <name>\tthe name of the jail\n");
 250         fprintf(stderr, "  -r <file>\treadonly files that should be staged\n");
 251         fprintf(stderr, "  -w <file>\twriteable files that should be staged\n");
 252         fprintf(stderr, "  -d <num>\tshow debug log (increase num to increase verbosity)\n");
 253         fprintf(stderr, "  -p\t\tjail has /proc\n");
 254         fprintf(stderr, "  -s\t\tjail has /sys\n");
 255         fprintf(stderr, "  -l\t\tjail has /dev/log\n");
 256         fprintf(stderr, "  -u\t\tjail has a ubus socket\n");
 257         fprintf(stderr, "  -o\t\tremont jail root (/) read only\n");
 258         fprintf(stderr, "\nWarning: by default root inside the jail is the same\n\
 259 and he has the same powers as root outside the jail,\n\
 260 thus he can escape the jail and/or break stuff.\n\
 261 Please use an appropriate seccomp/capabilities filter (-S/-C) to restrict his powers\n");
 262 }
 263
 264 static int spawn_jail(void *arg)
 265 {
 266         if (opts.name && sethostname(opts.name, strlen(opts.name))) {
 267                 ERROR("failed to sethostname: %s\n", strerror(errno));
 268         }
 269
 270         if (build_jail_fs()) {
 271                 ERROR("failed to build jail fs");
 272                 exit(EXIT_FAILURE);
 273         }
 274
 275         char **envp = build_envp(opts.seccomp);
 276         if (!envp)
 277                 exit(EXIT_FAILURE);
 278
 279         if (opts.capabilities && drop_capabilities(opts.capabilities))
 280                 exit(EXIT_FAILURE);
 281
 282         INFO("exec-ing %s\n", *opts.jail_argv);
 283         execve(*opts.jail_argv, opts.jail_argv, envp);
 284         //we get there only if execve fails
 285         ERROR("failed to execve %s: %s\n", *opts.jail_argv, strerror(errno));
 286         exit(EXIT_FAILURE);
 287 }
 288
 289 static int jail_running = 1;
 290 static int jail_return_code = 0;
 291
 292 static void jail_process_handler(struct uloop_process *c, int ret)
 293 {
 294         if (WIFEXITED(ret)) {
 295                 jail_return_code = WEXITSTATUS(ret);
 296                 INFO("jail (%d) exited with exit: %d\n", c->pid, jail_return_code);
 297         } else {
 298                 jail_return_code = WTERMSIG(ret);
 299                 INFO("jail (%d) exited with signal: %d\n", c->pid, jail_return_code);
 300         }
 301         jail_running = 0;
 302         uloop_end();
 303 }
 304
 305 static struct uloop_process jail_process = {
 306         .cb = jail_process_handler,
 307 };
 308
 309 static void add_extra(char *name, int readonly)
 310 {
 311         struct extra *f;
 312
 313         if (*name != '/') {
 314                 ERROR("%s is not an absolute path\n", name);
 315                 return;
 316         }
 317
 318         f = calloc(1, sizeof(struct extra));
 319
 320         f->name = basename(name);
 321         f->path = dirname(strdup(name));
 322         f->readonly = readonly;
 323
 324         list_add_tail(&f->list, &extras);
 325 }
 326
 327 int main(int argc, char **argv)
 328 {
 329         uid_t uid = getuid();
 330         char log[] = "/dev/log";
 331         char ubus[] = "/var/run/ubus.sock";
 332         int ret = EXIT_SUCCESS;
 333         int ch;
 334
 335         if (uid) {
 336                 ERROR("not root, aborting: %s\n", strerror(errno));
 337                 return EXIT_FAILURE;
 338         }
 339
 340         umask(022);
 341
 342         while ((ch = getopt(argc, argv, OPT_ARGS)) != -1) {
 343                 switch (ch) {
 344                 case 'd':
 345                         debug = atoi(optarg);
 346                         break;
 347                 case 'p':
 348                         opts.procfs = 1;
 349                         break;
 350                 case 'o':
 351                         opts.ronly = 1;
 352                         break;
 353                 case 's':
 354                         opts.sysfs = 1;
 355                         break;
 356                 case 'S':
 357                         opts.seccomp = optarg;
 358                         add_extra(optarg, 1);
 359                         break;
 360                 case 'C':
 361                         opts.capabilities = optarg;
 362                         add_extra(optarg, 1);
 363                         break;
 364                 case 'P':
 365                         opts.path = optarg;
 366                         break;
 367                 case 'n':
 368                         opts.name = optarg;
 369                         break;
 370                 case 'r':
 371                         add_extra(optarg, 1);
 372                         break;
 373                 case 'w':
 374                         add_extra(optarg, 0);
 375                         break;
 376                 case 'u':
 377                         add_extra(ubus, 0);
 378                         break;
 379                 case 'l':
 380                         add_extra(log, 0);
 381                         break;
 382                 }
 383         }
 384
 385         //no <binary> param found
 386         if (argc - optind < 1) {
 387                 usage();
 388                 return EXIT_FAILURE;
 389         }
 390
 391         opts.jail_argv = &argv[optind];
 392
 393         if (opts.name)
 394                 prctl(PR_SET_NAME, opts.name, NULL, NULL, NULL);
 395
 396         if (!opts.path && asprintf(&opts.path, "/tmp/%s", basename(*opts.jail_argv)) == -1) {
 397                 ERROR("failed to asprintf root path: %s\n", strerror(errno));
 398                 return EXIT_FAILURE;
 399         }
 400
 401         if (mkdir(opts.path, 0755)) {
 402                 ERROR("unable to create root path: %s (%s)\n", opts.path, strerror(errno));
 403                 return EXIT_FAILURE;
 404         }
 405
 406         uloop_init();
 407         jail_process.pid = clone(spawn_jail,
 408                         child_stack + STACK_SIZE,
 409                         CLONE_NEWUTS | CLONE_NEWPID | CLONE_NEWNS | CLONE_NEWIPC | SIGCHLD, argv);
 410
 411         if (jail_process.pid != -1) {
 412                 uloop_process_add(&jail_process);
 413                 uloop_run();
 414                 uloop_done();
 415                 if (jail_running) {
 416                         kill(jail_process.pid, SIGTERM);
 417                         waitpid(jail_process.pid, NULL, 0);
 418                 }
 419         } else {
 420                 ERROR("failed to spawn namespace: %s\n", strerror(errno));
 421                 ret = EXIT_FAILURE;
 422         }
 423
 424         if (rmdir(opts.path)) {
 425                 ERROR("Unable to remove root path: %s (%s)\n", opts.path, strerror(errno));
 426                 ret = EXIT_FAILURE;
 427         }
 428
 429         if (ret)
 430                 return ret;
 431
 432         return jail_return_code;
 433 }