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 <values.h>
  22 #include <errno.h>
  23 #include <string.h>
  24 #include <sys/stat.h>
  25 #include <fcntl.h>
  26 #include <libgen.h>
  27 #include <sched.h>
  28 #include <linux/limits.h>
  29
  30 #include "elf.h"
  31 #include "capabilities.h"
  32 #include "log.h"
  33
  34 #include <libubox/list.h>
  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 struct extra {
  53         struct list_head list;
  54
  55         const char *path;
  56         const char *name;
  57         int readonly;
  58 };
  59
  60 static LIST_HEAD(extras);
  61
  62 extern int pivot_root(const char *new_root, const char *put_old);
  63
  64 int debug = 0;
  65
  66 static char child_stack[STACK_SIZE];
  67
  68 static int mkdir_p(char *dir, mode_t mask)
  69 {
  70         char *l = strrchr(dir, '/');
  71         int ret;
  72
  73         if (!l)
  74                 return 0;
  75
  76         *l = '\0';
  77
  78         if (mkdir_p(dir, mask))
  79                 return -1;
  80
  81         *l = '/';
  82
  83         ret = mkdir(dir, mask);
  84         if (ret && errno == EEXIST)
  85                 return 0;
  86
  87         if (ret)
  88                 ERROR("mkdir failed on %s: %s\n", dir, strerror(errno));
  89
  90         return ret;
  91 }
  92
  93 static int mount_bind(const char *root, const char *path, const char *name, int readonly, int error)
  94 {
  95         const char *p = path;
  96         struct stat s;
  97         char old[PATH_MAX];
  98         char new[PATH_MAX];
  99         int fd;
 100
 101         if (strstr(p, "local"))
 102                 p = "/lib";
 103
 104         snprintf(old, sizeof(old), "%s/%s", path, name);
 105         snprintf(new, sizeof(new), "%s%s", root, p);
 106
 107         mkdir_p(new, 0755);
 108
 109         snprintf(new, sizeof(new), "%s%s/%s", root, p, name);
 110
 111         if (stat(old, &s)) {
 112                 ERROR("%s does not exist\n", old);
 113                 return error;
 114         }
 115
 116         if (S_ISDIR(s.st_mode)) {
 117                 mkdir_p(new, 0755);
 118         } else {
 119                 fd = creat(new, 0644);
 120                 if (fd == -1) {
 121                         ERROR("failed to create %s: %s\n", new, strerror(errno));
 122                         return -1;
 123                 }
 124                 close(fd);
 125         }
 126
 127         if (mount(old, new, NULL, MS_BIND, NULL)) {
 128                 ERROR("failed to mount -B %s %s: %s\n", old, new, strerror(errno));
 129                 return -1;
 130         }
 131
 132         if (readonly && mount(NULL, new, NULL, MS_BIND | MS_REMOUNT | MS_RDONLY, NULL)) {
 133                 ERROR("failed to remount ro %s: %s\n", new, strerror(errno));
 134                 return -1;
 135         }
 136
 137         DEBUG("mount -B %s %s\n", old, new);
 138
 139         return 0;
 140 }
 141
 142 static int build_jail_fs()
 143 {
 144         struct library *l;
 145         struct extra *m;
 146
 147         if (mount("tmpfs", opts.path, "tmpfs", MS_NOATIME, "mode=0755")) {
 148                 ERROR("tmpfs mount failed %s\n", strerror(errno));
 149                 return -1;
 150         }
 151
 152         if (chdir(opts.path)) {
 153                 ERROR("failed to chdir() in the jail root\n");
 154                 return -1;
 155         }
 156
 157         init_library_search();
 158
 159         if (elf_load_deps(*opts.jail_argv)) {
 160                 ERROR("failed to load dependencies\n");
 161                 return -1;
 162         }
 163
 164         if (opts.seccomp && elf_load_deps("libpreload-seccomp.so")) {
 165                 ERROR("failed to load libpreload-seccomp.so\n");
 166                 return -1;
 167         }
 168
 169         avl_for_each_element(&libraries, l, avl)
 170                 if (mount_bind(opts.path, l->path, l->name, 1, -1))
 171                         return -1;
 172
 173         list_for_each_entry(m, &extras, list)
 174                 if (mount_bind(opts.path, m->path, m->name, m->readonly, 0))
 175                         return -1;
 176
 177         char *mpoint;
 178         if (asprintf(&mpoint, "%s/old", opts.path) < 0) {
 179                 ERROR("failed to alloc pivot path: %s\n", strerror(errno));
 180                 return -1;
 181         }
 182         mkdir_p(mpoint, 0755);
 183         if (pivot_root(opts.path, mpoint) == -1) {
 184                 ERROR("pivot_root failed:%s\n", strerror(errno));
 185                 free(mpoint);
 186                 return -1;
 187         }
 188         free(mpoint);
 189         umount2("/old", MNT_DETACH);
 190         rmdir("/old");
 191         if (opts.procfs) {
 192                 mkdir("/proc", 0755);
 193                 mount("proc", "/proc", "proc", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
 194         }
 195         if (opts.sysfs) {
 196                 mkdir("/sys", 0755);
 197                 mount("sysfs", "/sys", "sysfs", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
 198         }
 199         if (opts.ronly)
 200                 mount(NULL, "/", NULL, MS_RDONLY | MS_REMOUNT, 0);
 201
 202         return 0;
 203 }
 204
 205 #define MAX_ENVP        8
 206 static char** build_envp(const char *seccomp)
 207 {
 208         static char *envp[MAX_ENVP];
 209         static char preload_var[PATH_MAX];
 210         static char seccomp_var[PATH_MAX];
 211         static char debug_var[] = "LD_DEBUG=all";
 212         const char *preload_lib = find_lib("libpreload-seccomp.so");
 213         int count = 0;
 214
 215         if (seccomp && !preload_lib) {
 216                 ERROR("failed to add preload-lib to env\n");
 217                 return NULL;
 218         }
 219         if (seccomp) {
 220                 snprintf(seccomp_var, sizeof(seccomp_var), "SECCOMP_FILE=%s", seccomp);
 221                 envp[count++] = seccomp_var;
 222                 snprintf(preload_var, sizeof(preload_var), "LD_PRELOAD=%s", preload_lib);
 223                 envp[count++] = preload_var;
 224         }
 225         if (debug > 1)
 226                 envp[count++] = debug_var;
 227
 228         return envp;
 229 }
 230
 231 static void usage(void)
 232 {
 233         fprintf(stderr, "ujail <options> -- <binary> <params ...>\n");
 234         fprintf(stderr, "  -d <num>\tshow debug log (increase num to increase verbosity)\n");
 235         fprintf(stderr, "  -S <file>\tseccomp filter config\n");
 236         fprintf(stderr, "  -C <file>\tcapabilities drop config\n");
 237         fprintf(stderr, "  -n <name>\tthe name of the jail\n");
 238         fprintf(stderr, "namespace jail options:\n");
 239         fprintf(stderr, "  -P <path>\tpath where the jail will be staged\n");
 240         fprintf(stderr, "  -r <file>\treadonly files that should be staged\n");
 241         fprintf(stderr, "  -w <file>\twriteable files that should be staged\n");
 242         fprintf(stderr, "  -p\t\tjail has /proc\n");
 243         fprintf(stderr, "  -s\t\tjail has /sys\n");
 244         fprintf(stderr, "  -l\t\tjail has /dev/log\n");
 245         fprintf(stderr, "  -u\t\tjail has a ubus socket\n");
 246         fprintf(stderr, "  -o\t\tremont jail root (/) read only\n");
 247         fprintf(stderr, "\nWarning: by default root inside the jail is the same\n\
 248 and he has the same powers as root outside the jail,\n\
 249 thus he can escape the jail and/or break stuff.\n\
 250 Please use seccomp/capabilities (-S/-C) to restrict his powers\n\n\
 251 If you use none of the namespace jail options,\n\
 252 ujail will not use namespace/build a jail,\n\
 253 and will only drop capabilities/apply seccomp filter.\n\n");
 254 }
 255
 256 static int exec_jail()
 257 {
 258         char **envp = build_envp(opts.seccomp);
 259         if (!envp)
 260                 exit(EXIT_FAILURE);
 261
 262         if (opts.capabilities && drop_capabilities(opts.capabilities))
 263                 exit(EXIT_FAILURE);
 264
 265         INFO("exec-ing %s\n", *opts.jail_argv);
 266         execve(*opts.jail_argv, opts.jail_argv, envp);
 267         //we get there only if execve fails
 268         ERROR("failed to execve %s: %s\n", *opts.jail_argv, strerror(errno));
 269         exit(EXIT_FAILURE);
 270 }
 271
 272 static int spawn_jail(void *_notused)
 273 {
 274         if (opts.name && sethostname(opts.name, strlen(opts.name))) {
 275                 ERROR("failed to sethostname: %s\n", strerror(errno));
 276         }
 277
 278         if (build_jail_fs()) {
 279                 ERROR("failed to build jail fs");
 280                 exit(EXIT_FAILURE);
 281         }
 282
 283         return exec_jail();
 284 }
 285
 286 static int jail_running = 1;
 287 static int jail_return_code = 0;
 288
 289 static void jail_process_handler(struct uloop_process *c, int ret)
 290 {
 291         if (WIFEXITED(ret)) {
 292                 jail_return_code = WEXITSTATUS(ret);
 293                 INFO("jail (%d) exited with exit: %d\n", c->pid, jail_return_code);
 294         } else {
 295                 jail_return_code = WTERMSIG(ret);
 296                 INFO("jail (%d) exited with signal: %d\n", c->pid, jail_return_code);
 297         }
 298         jail_running = 0;
 299         uloop_end();
 300 }
 301
 302 static struct uloop_process jail_process = {
 303         .cb = jail_process_handler,
 304 };
 305
 306 static void add_extra(char *name, int readonly)
 307 {
 308         struct extra *f;
 309
 310         if (*name != '/') {
 311                 ERROR("%s is not an absolute path\n", name);
 312                 return;
 313         }
 314
 315         f = calloc(1, sizeof(struct extra));
 316
 317         f->name = basename(name);
 318         f->path = dirname(strdup(name));
 319         f->readonly = readonly;
 320
 321         list_add_tail(&f->list, &extras);
 322 }
 323
 324 int main(int argc, char **argv)
 325 {
 326         uid_t uid = getuid();
 327         char log[] = "/dev/log";
 328         char ubus[] = "/var/run/ubus.sock";
 329         int ret = EXIT_SUCCESS;
 330         int ch;
 331
 332         if (uid) {
 333                 ERROR("not root, aborting: %s\n", strerror(errno));
 334                 return EXIT_FAILURE;
 335         }
 336
 337         umask(022);
 338
 339         while ((ch = getopt(argc, argv, OPT_ARGS)) != -1) {
 340                 switch (ch) {
 341                 case 'd':
 342                         debug = atoi(optarg);
 343                         break;
 344                 case 'p':
 345                         opts.namespace = 1;
 346                         opts.procfs = 1;
 347                         break;
 348                 case 'o':
 349                         opts.namespace = 1;
 350                         opts.ronly = 1;
 351                         break;
 352                 case 's':
 353                         opts.namespace = 1;
 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.namespace = 1;
 366                         opts.path = optarg;
 367                         break;
 368                 case 'n':
 369                         opts.name = optarg;
 370                         break;
 371                 case 'r':
 372                         opts.namespace = 1;
 373                         add_extra(optarg, 1);
 374                         break;
 375                 case 'w':
 376                         opts.namespace = 1;
 377                         add_extra(optarg, 0);
 378                         break;
 379                 case 'u':
 380                         opts.namespace = 1;
 381                         add_extra(ubus, 0);
 382                         break;
 383                 case 'l':
 384                         opts.namespace = 1;
 385                         add_extra(log, 0);
 386                         break;
 387                 }
 388         }
 389
 390         //no <binary> param found
 391         if (argc - optind < 1) {
 392                 usage();
 393                 return EXIT_FAILURE;
 394         }
 395         if (!(opts.namespace||opts.capabilities||opts.seccomp)) {
 396                 ERROR("Not using namespaces, capabilities or seccomp !!!\n\n");
 397                 usage();
 398                 return EXIT_FAILURE;
 399         }
 400         DEBUG("Using namespaces(%d), capabilities(%d), seccomp(%d)\n",
 401                 opts.namespace,
 402                 opts.capabilities != 0,
 403                 opts.seccomp != 0);
 404
 405         opts.jail_argv = &argv[optind];
 406
 407         if (opts.name)
 408                 prctl(PR_SET_NAME, opts.name, NULL, NULL, NULL);
 409
 410         if (opts.namespace && !opts.path && asprintf(&opts.path, "/tmp/%s", basename(*opts.jail_argv)) == -1) {
 411                 ERROR("failed to asprintf root path: %s\n", strerror(errno));
 412                 return EXIT_FAILURE;
 413         }
 414
 415         if (opts.namespace && mkdir(opts.path, 0755)) {
 416                 ERROR("unable to create root path: %s (%s)\n", opts.path, strerror(errno));
 417                 return EXIT_FAILURE;
 418         }
 419
 420         uloop_init();
 421         if (opts.namespace) {
 422                 jail_process.pid = clone(spawn_jail,
 423                         child_stack + STACK_SIZE,
 424                         CLONE_NEWUTS | CLONE_NEWPID | CLONE_NEWNS | CLONE_NEWIPC | SIGCHLD, NULL);
 425         } else {
 426                 jail_process.pid = fork();
 427         }
 428
 429         if (jail_process.pid > 0) {
 430                 //parent process
 431                 uloop_process_add(&jail_process);
 432                 uloop_run();
 433                 uloop_done();
 434                 if (jail_running) {
 435                         DEBUG("uloop interrupted, killing jail process\n");
 436                         kill(jail_process.pid, SIGTERM);
 437                         waitpid(jail_process.pid, NULL, 0);
 438                 }
 439         } else if (jail_process.pid == 0) {
 440                 //fork child process
 441                 return exec_jail();
 442         } else {
 443                 ERROR("failed to clone/fork: %s\n", strerror(errno));
 444                 ret = EXIT_FAILURE;
 445         }
 446
 447         if (opts.namespace && rmdir(opts.path)) {
 448                 ERROR("Unable to remove root path: %s (%s)\n", opts.path, strerror(errno));
 449                 ret = EXIT_FAILURE;
 450         }
 451
 452         if (ret)
 453                 return ret;
 454
 455         return jail_return_code;
 456 }