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 #include <sys/types.h>
  19
  20 #include <stdlib.h>
  21 #include <unistd.h>
  22 #include <errno.h>
  23 #include <pwd.h>
  24 #include <grp.h>
  25 #include <string.h>
  26 #include <sys/stat.h>
  27 #include <fcntl.h>
  28 #include <sched.h>
  29 #include <linux/limits.h>
  30 #include <linux/filter.h>
  31 #include <signal.h>
  32
  33 #include "capabilities.h"
  34 #include "elf.h"
  35 #include "fs.h"
  36 #include "jail.h"
  37 #include "log.h"
  38 #include "seccomp-oci.h"
  39
  40 #include <libubox/utils.h>
  41 #include <libubox/blobmsg.h>
  42 #include <libubox/blobmsg_json.h>
  43 #include <libubox/list.h>
  44 #include <libubox/vlist.h>
  45 #include <libubox/uloop.h>
  46 #include <libubus.h>
  47
  48 #ifndef CLONE_NEWCGROUP
  49 #define CLONE_NEWCGROUP 0x02000000
  50 #endif
  51
  52 #define STACK_SIZE      (1024 * 1024)
  53 #define OPT_ARGS        "S:C:n:h:r:w:d:psulocU:G:NR:fFO:T:EyJ:"
  54
  55 struct hook_execvpe {
  56         char *file;
  57         char **argv;
  58         char **envp;
  59         int timeout;
  60 };
  61
  62 static struct {
  63         char *name;
  64         char *hostname;
  65         char **jail_argv;
  66         char *cwd;
  67         char *seccomp;
  68         struct sock_fprog *ociseccomp;
  69         char *capabilities;
  70         struct jail_capset capset;
  71         char *user;
  72         char *group;
  73         char *extroot;
  74         char *overlaydir;
  75         char *tmpoverlaysize;
  76         char **envp;
  77         char *uidmap;
  78         char *gidmap;
  79         int no_new_privs;
  80         int namespace;
  81         int procfs;
  82         int ronly;
  83         int sysfs;
  84         int console;
  85         int pw_uid;
  86         int pw_gid;
  87         int gr_gid;
  88         int require_jail;
  89         struct {
  90                 struct hook_execvpe *createRuntime;
  91                 struct hook_execvpe *createContainer;
  92                 struct hook_execvpe *startContainer;
  93                 struct hook_execvpe *poststart;
  94                 struct hook_execvpe *poststop;
  95         } hooks;
  96 } opts;
  97
  98 static void free_hooklist(struct hook_execvpe *hooklist)
  99 {
 100         struct hook_execvpe *cur;
 101         char **tmp;
 102
 103         cur = hooklist;
 104         while (cur) {
 105                 free(cur->file);
 106                 tmp = cur->argv;
 107                 while (tmp)
 108                         free(*(tmp++));
 109
 110                 free(cur->argv);
 111
 112                 tmp = cur->envp;
 113                 while (tmp)
 114                         free(*(tmp++));
 115
 116                 free(cur->envp);
 117                 free(cur++);
 118         }
 119 }
 120
 121 static void free_opts(bool child) {
 122         char **tmp;
 123
 124         /* we need to keep argv, envp and seccomp filter in child */
 125         if (child) {
 126                 if (opts.ociseccomp) {
 127                         free(opts.ociseccomp->filter);
 128                         free(opts.ociseccomp);
 129                 }
 130
 131                 tmp = opts.jail_argv;
 132                 while(tmp)
 133                         free(*(tmp++));
 134
 135                 free(opts.jail_argv);
 136
 137                 tmp = opts.envp;
 138                 while (tmp)
 139                         free(*(tmp++));
 140
 141                 free(opts.envp);
 142         };
 143
 144         free(opts.hostname);
 145         free(opts.cwd);
 146         free(opts.extroot);
 147         free(opts.uidmap);
 148         free(opts.gidmap);
 149         free_hooklist(opts.hooks.createRuntime);
 150         free_hooklist(opts.hooks.createContainer);
 151         free_hooklist(opts.hooks.startContainer);
 152         free_hooklist(opts.hooks.poststart);
 153         free_hooklist(opts.hooks.poststop);
 154 }
 155
 156 static struct blob_buf ocibuf;
 157
 158 extern int pivot_root(const char *new_root, const char *put_old);
 159
 160 int debug = 0;
 161
 162 static char child_stack[STACK_SIZE];
 163
 164 int console_fd;
 165
 166 static int mount_overlay(char *jail_root, char *overlaydir) {
 167         char *upperdir, *workdir, *optsstr, *upperetc, *upperresolvconf;
 168         const char mountoptsformat[] = "lowerdir=%s,upperdir=%s,workdir=%s";
 169         int ret = -1, fd;
 170
 171         if (asprintf(&upperdir, "%s%s", overlaydir, "/upper") < 0)
 172                 goto out;
 173
 174         if (asprintf(&workdir, "%s%s", overlaydir, "/work") < 0)
 175                 goto upper_printf;
 176
 177         if (asprintf(&optsstr, mountoptsformat, jail_root, upperdir, workdir) < 0)
 178                 goto work_printf;
 179
 180         if (mkdir_p(upperdir, 0755) || mkdir_p(workdir, 0755))
 181                 goto opts_printf;
 182
 183 /*
 184  * make sure /etc/resolv.conf exists in overlay and is owned by jail userns root
 185  * this is to work-around a bug in overlayfs described in the overlayfs-userns
 186  * patch:
 187  * 3. modification of a file 'hithere' which is in l but not yet
 188  * in u, and which is not owned by T, is not allowed, even if
 189  * writes to u are allowed.  This may be a bug in overlayfs,
 190  * but it is safe behavior.
 191  */
 192         if (asprintf(&upperetc, "%s/etc", upperdir) < 0)
 193                 goto opts_printf;
 194
 195         if (mkdir_p(upperetc, 0755))
 196                 goto upper_etc_printf;
 197
 198         if (asprintf(&upperresolvconf, "%s/resolv.conf", upperetc) < 0)
 199                 goto upper_etc_printf;
 200
 201         fd = creat(upperresolvconf, 0644);
 202         if (fd == -1) {
 203                 ERROR("creat(%s) failed: %m\n", upperresolvconf);
 204                 goto upper_resolvconf_printf;
 205         }
 206         close(fd);
 207
 208         DEBUG("mount -t overlay %s %s (%s)\n", jail_root, jail_root, optsstr);
 209
 210         if (mount(jail_root, jail_root, "overlay", MS_NOATIME, optsstr))
 211                 goto opts_printf;
 212
 213         ret = 0;
 214
 215 upper_resolvconf_printf:
 216         free(upperresolvconf);
 217 upper_etc_printf:
 218         free(upperetc);
 219 opts_printf:
 220         free(optsstr);
 221 work_printf:
 222         free(workdir);
 223 upper_printf:
 224         free(upperdir);
 225 out:
 226         return ret;
 227 }
 228
 229 static void pass_console(int console_fd)
 230 {
 231         struct ubus_context *ctx = ubus_connect(NULL);
 232         static struct blob_buf req;
 233         uint32_t id;
 234
 235         if (!ctx)
 236                 return;
 237
 238         blob_buf_init(&req, 0);
 239         blobmsg_add_string(&req, "name", opts.name);
 240
 241         if (ubus_lookup_id(ctx, "container", &id) ||
 242             ubus_invoke_fd(ctx, id, "console_set", req.head, NULL, NULL, 3000, console_fd))
 243                 INFO("ubus request failed\n");
 244         else
 245                 close(console_fd);
 246
 247         blob_buf_free(&req);
 248         ubus_free(ctx);
 249 }
 250
 251 static int create_dev_console(const char *jail_root)
 252 {
 253         char *console_fname;
 254         char dev_console_path[PATH_MAX];
 255         int slave_console_fd;
 256
 257         /* Open UNIX/98 virtual console */
 258         console_fd = posix_openpt(O_RDWR | O_NOCTTY);
 259         if (console_fd == -1)
 260                 return -1;
 261
 262         console_fname = ptsname(console_fd);
 263         DEBUG("got console fd %d and PTS client name %s\n", console_fd, console_fname);
 264         if (!console_fname)
 265                 goto no_console;
 266
 267         grantpt(console_fd);
 268         unlockpt(console_fd);
 269
 270         /* pass PTY master to procd */
 271         pass_console(console_fd);
 272
 273         /* mount-bind PTY slave to /dev/console in jail */
 274         snprintf(dev_console_path, sizeof(dev_console_path), "%s/dev/console", jail_root);
 275         close(creat(dev_console_path, 0620));
 276
 277         if (mount(console_fname, dev_console_path, NULL, MS_BIND, NULL))
 278                 goto no_console;
 279
 280         /* use PTY slave for stdio */
 281         slave_console_fd = open(console_fname, O_RDWR); /* | O_NOCTTY */
 282         dup2(slave_console_fd, 0);
 283         dup2(slave_console_fd, 1);
 284         dup2(slave_console_fd, 2);
 285         close(slave_console_fd);
 286
 287         INFO("using guest console %s\n", console_fname);
 288
 289         return 0;
 290
 291 no_console:
 292         close(console_fd);
 293         return 1;
 294 }
 295
 296 static int hook_running = 0;
 297 static int hook_return_code = 0;
 298
 299 static void hook_process_timeout_cb(struct uloop_timeout *t);
 300 static struct uloop_timeout hook_process_timeout = {
 301         .cb = hook_process_timeout_cb,
 302 };
 303
 304 static void hook_process_handler(struct uloop_process *c, int ret)
 305 {
 306         uloop_timeout_cancel(&hook_process_timeout);
 307         if (WIFEXITED(ret)) {
 308                 hook_return_code = WEXITSTATUS(ret);
 309                 INFO("hook (%d) exited with exit: %d\n", c->pid, hook_return_code);
 310         } else {
 311                 hook_return_code = WTERMSIG(ret);
 312                 INFO("hook (%d) exited with signal: %d\n", c->pid, hook_return_code);
 313         }
 314         hook_running = 0;
 315         uloop_end();
 316 }
 317
 318 static struct uloop_process hook_process = {
 319         .cb = hook_process_handler,
 320 };
 321
 322 static void hook_process_timeout_cb(struct uloop_timeout *t)
 323 {
 324         DEBUG("hook process failed to stop, sending SIGKILL\n");
 325         kill(hook_process.pid, SIGKILL);
 326 }
 327
 328 static void hook_handle_signal(int signo)
 329 {
 330         DEBUG("forwarding signal %d to the jailed process\n", signo);
 331         kill(hook_process.pid, signo);
 332 }
 333
 334 static int run_hook(struct hook_execvpe *hook)
 335 {
 336         sigset_t sigmask;
 337         int i;
 338
 339         DEBUG("executing hook %s\n", hook->file);
 340         uloop_init();
 341         sigfillset(&sigmask);
 342         for (i = 0; i < _NSIG; i++) {
 343                 struct sigaction s = { 0 };
 344
 345                 if (!sigismember(&sigmask, i))
 346                         continue;
 347                 if ((i == SIGCHLD) || (i == SIGPIPE) || (i == SIGSEGV))
 348                         continue;
 349
 350                 s.sa_handler = hook_handle_signal;
 351                 sigaction(i, &s, NULL);
 352         }
 353         hook_running = 1;
 354         hook_process.pid = fork();
 355         if (hook_process.pid > 0) {
 356                 /* parent */
 357                 uloop_process_add(&hook_process);
 358                 if (hook->timeout > 0)
 359                         uloop_timeout_set(&hook_process_timeout, 1000 * hook->timeout);
 360
 361                 uloop_run();
 362                 if (hook_running) {
 363                         DEBUG("uloop interrupted, killing hook process\n");
 364                         kill(hook_process.pid, SIGTERM);
 365                         uloop_timeout_set(&hook_process_timeout, 1000);
 366                         uloop_run();
 367                 }
 368                 uloop_done();
 369
 370                 return 0;
 371         } else if (hook_process.pid == 0) {
 372                 /* child */
 373                 execvpe(hook->file, hook->argv, hook->envp);
 374                 return errno;
 375         } else {
 376                 /* fork error */
 377                 return errno;
 378         }
 379 }
 380
 381 static int run_hooks(struct hook_execvpe *hook)
 382 {
 383         struct hook_execvpe *cur;
 384         int res;
 385
 386         cur = hook;
 387         while (cur) {
 388                 res = run_hook(cur++);
 389                 if (res)
 390                         return res;
 391         }
 392
 393         return 0;
 394 }
 395
 396 static int build_jail_fs(void)
 397 {
 398         char jail_root[] = "/tmp/ujail-XXXXXX";
 399         char tmpovdir[] = "/tmp/ujail-overlay-XXXXXX";
 400         char tmpdevdir[] = "/tmp/ujail-XXXXXX/dev";
 401         char tmpdevptsdir[] = "/tmp/ujail-XXXXXX/dev/pts";
 402         char *overlaydir = NULL;
 403
 404         if (mkdtemp(jail_root) == NULL) {
 405                 ERROR("mkdtemp(%s) failed: %m\n", jail_root);
 406                 return -1;
 407         }
 408
 409         /* oldroot can't be MS_SHARED else pivot_root() fails */
 410         if (mount("none", "/", NULL, MS_REC|MS_PRIVATE, NULL)) {
 411                 ERROR("private mount failed %m\n");
 412                 return -1;
 413         }
 414
 415         if (opts.extroot) {
 416                 if (mount(opts.extroot, jail_root, NULL, MS_BIND, NULL)) {
 417                         ERROR("extroot mount failed %m\n");
 418                         return -1;
 419                 }
 420         } else {
 421                 if (mount("tmpfs", jail_root, "tmpfs", MS_NOATIME, "mode=0755")) {
 422                         ERROR("tmpfs mount failed %m\n");
 423                         return -1;
 424                 }
 425         }
 426
 427         if (opts.tmpoverlaysize) {
 428                 char mountoptsstr[] = "mode=0755,size=XXXXXXXX";
 429
 430                 snprintf(mountoptsstr, sizeof(mountoptsstr),
 431                          "mode=0755,size=%s", opts.tmpoverlaysize);
 432                 if (mkdtemp(tmpovdir) == NULL) {
 433                         ERROR("mkdtemp(%s) failed: %m\n", jail_root);
 434                         return -1;
 435                 }
 436                 if (mount("tmpfs", tmpovdir, "tmpfs", MS_NOATIME,
 437                           mountoptsstr)) {
 438                         ERROR("failed to mount tmpfs for overlay (size=%s)\n", opts.tmpoverlaysize);
 439                         return -1;
 440                 }
 441                 overlaydir = tmpovdir;
 442         }
 443
 444         if (opts.overlaydir)
 445                 overlaydir = opts.overlaydir;
 446
 447         if (overlaydir)
 448                 mount_overlay(jail_root, overlaydir);
 449
 450         if (chdir(jail_root)) {
 451                 ERROR("chdir(%s) (jail_root) failed: %m\n", jail_root);
 452                 return -1;
 453         }
 454
 455         snprintf(tmpdevdir, sizeof(tmpdevdir), "%s/dev", jail_root);
 456         mkdir_p(tmpdevdir, 0755);
 457         if (mount(NULL, tmpdevdir, "tmpfs", MS_NOATIME | MS_NOEXEC | MS_NOSUID, "size=1M"))
 458                 return -1;
 459
 460         snprintf(tmpdevptsdir, sizeof(tmpdevptsdir), "%s/dev/pts", jail_root);
 461         mkdir_p(tmpdevptsdir, 0755);
 462         if (mount(NULL, tmpdevptsdir, "devpts", MS_NOATIME | MS_NOEXEC | MS_NOSUID, NULL))
 463                 return -1;
 464
 465         if (opts.console)
 466                 create_dev_console(jail_root);
 467
 468         if (mount_all(jail_root)) {
 469                 ERROR("mount_all() failed\n");
 470                 return -1;
 471         }
 472
 473         /* make sure /etc/resolv.conf exists if in new network namespace */
 474         if (opts.namespace & CLONE_NEWNET) {
 475                 char jailetc[PATH_MAX], jaillink[PATH_MAX];
 476
 477                 snprintf(jailetc, PATH_MAX, "%s/etc", jail_root);
 478                 mkdir_p(jailetc, 0755);
 479                 snprintf(jaillink, PATH_MAX, "%s/etc/resolv.conf", jail_root);
 480                 if (overlaydir)
 481                         unlink(jaillink);
 482
 483                 symlink("../tmp/resolv.conf.d/resolv.conf.auto", jaillink);
 484         }
 485
 486         run_hooks(opts.hooks.createContainer);
 487
 488         char dirbuf[sizeof(jail_root) + 4];
 489         snprintf(dirbuf, sizeof(dirbuf), "%s/old", jail_root);
 490         mkdir(dirbuf, 0755);
 491
 492         if (pivot_root(jail_root, dirbuf) == -1) {
 493                 ERROR("pivot_root(%s, %s) failed: %m\n", jail_root, dirbuf);
 494                 return -1;
 495         }
 496         if (chdir("/")) {
 497                 ERROR("chdir(/) (after pivot_root) failed: %m\n");
 498                 return -1;
 499         }
 500
 501         snprintf(dirbuf, sizeof(dirbuf), "/old%s", jail_root);
 502         umount2(dirbuf, MNT_DETACH);
 503         rmdir(dirbuf);
 504         if (opts.tmpoverlaysize) {
 505                 char tmpdirbuf[sizeof(tmpovdir) + 4];
 506                 snprintf(tmpdirbuf, sizeof(tmpdirbuf), "/old%s", tmpovdir);
 507                 umount2(tmpdirbuf, MNT_DETACH);
 508                 rmdir(tmpdirbuf);
 509         }
 510
 511         umount2("/old", MNT_DETACH);
 512         rmdir("/old");
 513
 514         if (opts.procfs) {
 515                 mkdir("/proc", 0755);
 516                 mount("proc", "/proc", "proc", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID, 0);
 517                 /*
 518                  * make /proc/sys read-only while keeping read-write to
 519                  * /proc/sys/net if CLONE_NEWNET is set.
 520                  */
 521                 if (opts.namespace & CLONE_NEWNET)
 522                         mount("/proc/sys/net", "/proc/self/net", NULL, MS_BIND, 0);
 523
 524                 mount("/proc/sys", "/proc/sys", NULL, MS_BIND, 0);
 525                 mount(NULL, "/proc/sys", NULL, MS_REMOUNT | MS_RDONLY, 0);
 526                 mount(NULL, "/proc", NULL, MS_REMOUNT, 0);
 527
 528                 if (opts.namespace & CLONE_NEWNET)
 529                         mount("/proc/self/net", "/proc/sys/net", NULL, MS_MOVE, 0);
 530         }
 531         if (opts.sysfs) {
 532                 mkdir("/sys", 0755);
 533                 mount("sysfs", "/sys", "sysfs", MS_NOATIME | MS_NODEV | MS_NOEXEC | MS_NOSUID | MS_RDONLY, 0);
 534         }
 535         if (opts.ronly)
 536                 mount(NULL, "/", NULL, MS_RDONLY | MS_REMOUNT, 0);
 537
 538         return 0;
 539 }
 540
 541 static int write_uid_gid_map(pid_t child_pid, bool gidmap, char *mapstr)
 542 {
 543         int map_file;
 544         char map_path[64];
 545
 546         if (snprintf(map_path, sizeof(map_path), "/proc/%d/%s",
 547                 child_pid, gidmap?"gid_map":"uid_map") < 0)
 548                 return -1;
 549
 550         if ((map_file = open(map_path, O_WRONLY)) == -1)
 551                 return -1;
 552
 553         if (dprintf(map_file, "%s", mapstr)) {
 554                 close(map_file);
 555                 return -1;
 556         }
 557
 558         close(map_file);
 559         free(mapstr);
 560         return 0;
 561 }
 562
 563 static int write_single_uid_gid_map(pid_t child_pid, bool gidmap, int id)
 564 {
 565         int map_file;
 566         char map_path[64];
 567         const char *map_format = "%d %d %d\n";
 568         if (snprintf(map_path, sizeof(map_path), "/proc/%d/%s",
 569                 child_pid, gidmap?"gid_map":"uid_map") < 0)
 570                 return -1;
 571
 572         if ((map_file = open(map_path, O_WRONLY)) == -1)
 573                 return -1;
 574
 575         if (dprintf(map_file, map_format, 0, id, 1) == -1) {
 576                 close(map_file);
 577                 return -1;
 578         }
 579
 580         close(map_file);
 581         return 0;
 582 }
 583
 584 static int write_setgroups(pid_t child_pid, bool allow)
 585 {
 586         int setgroups_file;
 587         char setgroups_path[64];
 588
 589         if (snprintf(setgroups_path, sizeof(setgroups_path), "/proc/%d/setgroups",
 590                 child_pid) < 0) {
 591                 return -1;
 592         }
 593
 594         if ((setgroups_file = open(setgroups_path, O_WRONLY)) == -1) {
 595                 return -1;
 596         }
 597
 598         if (dprintf(setgroups_file, "%s", allow?"allow":"deny") == -1) {
 599                 close(setgroups_file);
 600                 return -1;
 601         }
 602
 603         close(setgroups_file);
 604         return 0;
 605 }
 606
 607 static void get_jail_user(int *user, int *user_gid, int *gr_gid)
 608 {
 609         struct passwd *p = NULL;
 610         struct group *g = NULL;
 611
 612         if (opts.user) {
 613                 p = getpwnam(opts.user);
 614                 if (!p) {
 615                         ERROR("failed to get uid/gid for user %s: %d (%s)\n",
 616                               opts.user, errno, strerror(errno));
 617                         exit(EXIT_FAILURE);
 618                 }
 619                 *user = p->pw_uid;
 620                 *user_gid = p->pw_gid;
 621         } else {
 622                 *user = -1;
 623                 *user_gid = -1;
 624         }
 625
 626         if (opts.group) {
 627                 g = getgrnam(opts.group);
 628                 if (!g) {
 629                         ERROR("failed to get gid for group %s: %m\n", opts.group);
 630                         exit(EXIT_FAILURE);
 631                 }
 632                 *gr_gid = g->gr_gid;
 633         } else {
 634                 *gr_gid = -1;
 635         }
 636 };
 637
 638 static void set_jail_user(int pw_uid, int user_gid, int gr_gid)
 639 {
 640         if (opts.user && (user_gid != -1) && initgroups(opts.user, user_gid)) {
 641                 ERROR("failed to initgroups() for user %s: %m\n", opts.user);
 642                 exit(EXIT_FAILURE);
 643         }
 644
 645         if ((gr_gid != -1) && setregid(gr_gid, gr_gid)) {
 646                 ERROR("failed to set group id %d: %m\n", gr_gid);
 647                 exit(EXIT_FAILURE);
 648         }
 649
 650         if ((pw_uid != -1) && setreuid(pw_uid, pw_uid)) {
 651                 ERROR("failed to set user id %d: %m\n", pw_uid);
 652                 exit(EXIT_FAILURE);
 653         }
 654 }
 655
 656 #define MAX_ENVP        8
 657 static char** build_envp(const char *seccomp, char **ocienvp)
 658 {
 659         static char *envp[MAX_ENVP];
 660         static char preload_var[PATH_MAX];
 661         static char seccomp_var[PATH_MAX];
 662         static char debug_var[] = "LD_DEBUG=all";
 663         static char container_var[] = "container=ujail";
 664         const char *preload_lib = find_lib("libpreload-seccomp.so");
 665         char **addenv;
 666
 667         int count = 0;
 668
 669         if (seccomp && !preload_lib) {
 670                 ERROR("failed to add preload-lib to env\n");
 671                 return NULL;
 672         }
 673         if (seccomp) {
 674                 snprintf(seccomp_var, sizeof(seccomp_var), "SECCOMP_FILE=%s", seccomp);
 675                 envp[count++] = seccomp_var;
 676                 snprintf(preload_var, sizeof(preload_var), "LD_PRELOAD=%s", preload_lib);
 677                 envp[count++] = preload_var;
 678         }
 679
 680         envp[count++] = container_var;
 681
 682         if (debug > 1)
 683                 envp[count++] = debug_var;
 684
 685         addenv = ocienvp;
 686         while (addenv && *addenv) {
 687                 envp[count++] = *(addenv++);
 688                 if (count >= MAX_ENVP) {
 689                         ERROR("environment limited to %d extra records, truncating\n", MAX_ENVP);
 690                         break;
 691                 }
 692         }
 693         return envp;
 694 }
 695
 696 static void usage(void)
 697 {
 698         fprintf(stderr, "ujail <options> -- <binary> <params ...>\n");
 699         fprintf(stderr, "  -d <num>\tshow debug log (increase num to increase verbosity)\n");
 700         fprintf(stderr, "  -S <file>\tseccomp filter config\n");
 701         fprintf(stderr, "  -C <file>\tcapabilities drop config\n");
 702         fprintf(stderr, "  -c\t\tset PR_SET_NO_NEW_PRIVS\n");
 703         fprintf(stderr, "  -n <name>\tthe name of the jail\n");
 704         fprintf(stderr, "namespace jail options:\n");
 705         fprintf(stderr, "  -h <hostname>\tchange the hostname of the jail\n");
 706         fprintf(stderr, "  -N\t\tjail has network namespace\n");
 707         fprintf(stderr, "  -f\t\tjail has user namespace\n");
 708         fprintf(stderr, "  -F\t\tjail has cgroups namespace\n");
 709         fprintf(stderr, "  -r <file>\treadonly files that should be staged\n");
 710         fprintf(stderr, "  -w <file>\twriteable files that should be staged\n");
 711         fprintf(stderr, "  -p\t\tjail has /proc\n");
 712         fprintf(stderr, "  -s\t\tjail has /sys\n");
 713         fprintf(stderr, "  -l\t\tjail has /dev/log\n");
 714         fprintf(stderr, "  -u\t\tjail has a ubus socket\n");
 715         fprintf(stderr, "  -U <name>\tuser to run jailed process\n");
 716         fprintf(stderr, "  -G <name>\tgroup to run jailed process\n");
 717         fprintf(stderr, "  -o\t\tremont jail root (/) read only\n");
 718         fprintf(stderr, "  -R <dir>\texternal jail rootfs (system container)\n");
 719         fprintf(stderr, "  -O <dir>\tdirectory for r/w overlayfs\n");
 720         fprintf(stderr, "  -T <size>\tuse tmpfs r/w overlayfs with <size>\n");
 721         fprintf(stderr, "  -E\t\tfail if jail cannot be setup\n");
 722         fprintf(stderr, "  -y\t\tprovide jail console\n");
 723         fprintf(stderr, "  -J <dir>\tstart OCI bundle\n");
 724         fprintf(stderr, "\nWarning: by default root inside the jail is the same\n\
 725 and he has the same powers as root outside the jail,\n\
 726 thus he can escape the jail and/or break stuff.\n\
 727 Please use seccomp/capabilities (-S/-C) to restrict his powers\n\n\
 728 If you use none of the namespace jail options,\n\
 729 ujail will not use namespace/build a jail,\n\
 730 and will only drop capabilities/apply seccomp filter.\n\n");
 731 }
 732
 733 static int exec_jail(void *pipes_ptr)
 734 {
 735         int *pipes = (int*)pipes_ptr;
 736         char buf[1];
 737         int pw_uid, pw_gid, gr_gid;
 738
 739         close(pipes[0]);
 740         close(pipes[3]);
 741
 742         buf[0] = 'i';
 743         if (write(pipes[1], buf, 1) < 1) {
 744                 ERROR("can't write to parent\n");
 745                 exit(EXIT_FAILURE);
 746         }
 747         if (read(pipes[2], buf, 1) < 1) {
 748                 ERROR("can't read from parent\n");
 749                 exit(EXIT_FAILURE);
 750         }
 751         if (buf[0] != 'O') {
 752                 ERROR("parent had an error, child exiting\n");
 753                 exit(EXIT_FAILURE);
 754         }
 755
 756         close(pipes[1]);
 757         close(pipes[2]);
 758
 759         if (opts.namespace & CLONE_NEWUSER) {
 760                 if (setregid(0, 0) < 0) {
 761                         ERROR("setgid\n");
 762                         exit(EXIT_FAILURE);
 763                 }
 764                 if (setreuid(0, 0) < 0) {
 765                         ERROR("setuid\n");
 766                         exit(EXIT_FAILURE);
 767                 }
 768                 if (setgroups(0, NULL) < 0) {
 769                         ERROR("setgroups\n");
 770                         exit(EXIT_FAILURE);
 771                 }
 772         }
 773
 774         if (opts.namespace && opts.hostname && strlen(opts.hostname) > 0
 775                         && sethostname(opts.hostname, strlen(opts.hostname))) {
 776                 ERROR("sethostname(%s) failed: %m\n", opts.hostname);
 777                 exit(EXIT_FAILURE);
 778         }
 779
 780         if ((opts.namespace & CLONE_NEWNS) && build_jail_fs()) {
 781                 ERROR("failed to build jail fs\n");
 782                 exit(EXIT_FAILURE);
 783         }
 784
 785         run_hooks(opts.hooks.startContainer);
 786
 787         if (applyOCIcapabilities(opts.capset))
 788                 exit(EXIT_FAILURE);
 789
 790         if (opts.capabilities && drop_capabilities(opts.capabilities))
 791                 exit(EXIT_FAILURE);
 792
 793         if (opts.no_new_privs && prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
 794                 ERROR("prctl(PR_SET_NO_NEW_PRIVS) failed: %m\n");
 795                 exit(EXIT_FAILURE);
 796         }
 797
 798         if (!(opts.namespace & CLONE_NEWUSER)) {
 799                 get_jail_user(&pw_uid, &pw_gid, &gr_gid);
 800
 801                 set_jail_user(opts.pw_uid?:pw_uid, opts.pw_gid?:pw_gid, opts.gr_gid?:gr_gid);
 802         }
 803
 804         char **envp = build_envp(opts.seccomp, opts.envp);
 805         if (!envp)
 806                 exit(EXIT_FAILURE);
 807
 808         if (opts.cwd && chdir(opts.cwd))
 809                 exit(EXIT_FAILURE);
 810
 811         if (opts.ociseccomp && applyOCIlinuxseccomp(opts.ociseccomp))
 812                 exit(EXIT_FAILURE);
 813
 814         free_opts(false);
 815         INFO("exec-ing %s\n", *opts.jail_argv);
 816         if (opts.envp) /* respect PATH if potentially set in ENV */
 817                 execvpe(*opts.jail_argv, opts.jail_argv, envp);
 818         else
 819                 execve(*opts.jail_argv, opts.jail_argv, envp);
 820
 821         /* we get there only if execve fails */
 822         ERROR("failed to execve %s: %m\n", *opts.jail_argv);
 823         exit(EXIT_FAILURE);
 824 }
 825
 826 static int jail_running = 1;
 827 static int jail_return_code = 0;
 828
 829 static void jail_process_timeout_cb(struct uloop_timeout *t);
 830 static struct uloop_timeout jail_process_timeout = {
 831         .cb = jail_process_timeout_cb,
 832 };
 833
 834 static void jail_process_handler(struct uloop_process *c, int ret)
 835 {
 836         uloop_timeout_cancel(&jail_process_timeout);
 837         if (WIFEXITED(ret)) {
 838                 jail_return_code = WEXITSTATUS(ret);
 839                 INFO("jail (%d) exited with exit: %d\n", c->pid, jail_return_code);
 840         } else {
 841                 jail_return_code = WTERMSIG(ret);
 842                 INFO("jail (%d) exited with signal: %d\n", c->pid, jail_return_code);
 843         }
 844         jail_running = 0;
 845         uloop_end();
 846 }
 847
 848 static struct uloop_process jail_process = {
 849         .cb = jail_process_handler,
 850 };
 851
 852 static void jail_process_timeout_cb(struct uloop_timeout *t)
 853 {
 854         DEBUG("jail process failed to stop, sending SIGKILL\n");
 855         kill(jail_process.pid, SIGKILL);
 856 }
 857
 858 static void jail_handle_signal(int signo)
 859 {
 860         DEBUG("forwarding signal %d to the jailed process\n", signo);
 861         kill(jail_process.pid, signo);
 862 }
 863
 864 static int netns_open_pid(const pid_t target_ns)
 865 {
 866         char pid_net_path[PATH_MAX];
 867
 868         snprintf(pid_net_path, sizeof(pid_net_path), "/proc/%u/ns/net", target_ns);
 869
 870         return open(pid_net_path, O_RDONLY);
 871 }
 872
 873 static void netns_updown(pid_t pid, bool start)
 874 {
 875         struct ubus_context *ctx = ubus_connect(NULL);
 876         static struct blob_buf req;
 877         uint32_t id;
 878
 879         if (!ctx)
 880                 return;
 881
 882         blob_buf_init(&req, 0);
 883         blobmsg_add_string(&req, "jail", opts.name);
 884         blobmsg_add_u32(&req, "pid", pid);
 885         blobmsg_add_u8(&req, "start", start);
 886
 887         if (ubus_lookup_id(ctx, "network", &id) ||
 888             ubus_invoke(ctx, id, "netns_updown", req.head, NULL, NULL, 3000))
 889                 INFO("ubus request failed\n");
 890
 891         blob_buf_free(&req);
 892         ubus_free(ctx);
 893 }
 894
 895 static int parseOCIenvarray(struct blob_attr *msg, char ***envp)
 896 {
 897         struct blob_attr *cur;
 898         int sz = 0, rem;
 899
 900         blobmsg_for_each_attr(cur, msg, rem)
 901                 ++sz;
 902
 903         if (sz > 0) {
 904                 *envp = calloc(1 + sz, sizeof(char*));
 905                 if (!(*envp))
 906                         return ENOMEM;
 907         } else {
 908                 *envp = NULL;
 909                 return 0;
 910         }
 911
 912         sz = 0;
 913         blobmsg_for_each_attr(cur, msg, rem)
 914                 (*envp)[sz++] = strdup(blobmsg_get_string(cur));
 915
 916         if (sz)
 917                 (*envp)[sz] = NULL;
 918
 919         return 0;
 920 }
 921
 922 enum {
 923         OCI_ROOT_PATH,
 924         OCI_ROOT_READONLY,
 925         __OCI_ROOT_MAX,
 926 };
 927
 928 static const struct blobmsg_policy oci_root_policy[] = {
 929         [OCI_ROOT_PATH] = { "path", BLOBMSG_TYPE_STRING },
 930         [OCI_ROOT_READONLY] = { "readonly", BLOBMSG_TYPE_BOOL },
 931 };
 932
 933 static int parseOCIroot(const char *jsonfile, struct blob_attr *msg)
 934 {
 935         static char rootpath[PATH_MAX] = { 0 };
 936         struct blob_attr *tb[__OCI_ROOT_MAX];
 937         char *cur;
 938
 939         blobmsg_parse(oci_root_policy, __OCI_ROOT_MAX, tb, blobmsg_data(msg), blobmsg_len(msg));
 940
 941         if (!tb[OCI_ROOT_PATH])
 942                 return ENODATA;
 943
 944         strncpy(rootpath, jsonfile, PATH_MAX);
 945         cur = strrchr(rootpath, '/');
 946
 947         if (!cur)
 948                 return ENOTDIR;
 949
 950         *(++cur) = '\0';
 951         strncat(rootpath, blobmsg_get_string(tb[OCI_ROOT_PATH]), PATH_MAX - (strlen(rootpath) + 1));
 952
 953         opts.extroot = rootpath;
 954
 955         opts.ronly = blobmsg_get_bool(tb[OCI_ROOT_READONLY]);
 956
 957         return 0;
 958 }
 959
 960
 961 enum {
 962         OCI_HOOK_PATH,
 963         OCI_HOOK_ARGS,
 964         OCI_HOOK_ENV,
 965         OCI_HOOK_TIMEOUT,
 966         __OCI_HOOK_MAX,
 967 };
 968
 969 static const struct blobmsg_policy oci_hook_policy[] = {
 970         [OCI_HOOK_PATH] = { "path", BLOBMSG_TYPE_STRING },
 971         [OCI_HOOK_ARGS] = { "args", BLOBMSG_TYPE_ARRAY },
 972         [OCI_HOOK_ENV] = { "env", BLOBMSG_TYPE_ARRAY },
 973         [OCI_HOOK_TIMEOUT] = { "timeout", BLOBMSG_TYPE_INT32 },
 974 };
 975
 976
 977 static int parseOCIhook(struct hook_execvpe **hooklist, struct blob_attr *msg)
 978 {
 979         struct blob_attr *tb[__OCI_HOOK_MAX];
 980         struct blob_attr *cur;
 981         int rem, ret = 0;
 982         int idx = 0;
 983
 984         blobmsg_for_each_attr(cur, msg, rem)
 985                 ++idx;
 986
 987         if (!idx)
 988                 goto errout;
 989
 990         *hooklist = calloc(idx + 1, sizeof(struct hook_execvpe *));
 991         idx = 0;
 992
 993         blobmsg_for_each_attr(cur, msg, rem) {
 994                 blobmsg_parse(oci_hook_policy, __OCI_HOOK_MAX, tb, blobmsg_data(cur), blobmsg_len(cur));
 995
 996                 if (!tb[OCI_HOOK_PATH]) {
 997                         ret = EINVAL;
 998                         goto errout;
 999                 }
1000
1001                 hooklist[idx] = malloc(sizeof(struct hook_execvpe));
1002                 if (tb[OCI_HOOK_ARGS]) {
1003                         ret = parseOCIenvarray(tb[OCI_HOOK_ARGS], &(hooklist[idx]->argv));
1004                         if (ret)
1005                                 goto errout;
1006                 }
1007
1008                 if (tb[OCI_HOOK_ENV]) {
1009                         ret = parseOCIenvarray(tb[OCI_HOOK_ENV], &(hooklist[idx]->envp));
1010                         if (ret)
1011                                 goto errout;
1012                 }
1013
1014                 if (tb[OCI_HOOK_TIMEOUT])
1015                         hooklist[idx]->timeout = blobmsg_get_u32(tb[OCI_HOOK_TIMEOUT]);
1016
1017                 hooklist[idx]->file = strdup(blobmsg_get_string(tb[OCI_HOOK_PATH]));
1018
1019                 ++idx;
1020         }
1021
1022         hooklist[idx] = NULL;
1023         return 0;
1024
1025 errout:
1026         free_hooklist(*hooklist);
1027         *hooklist = NULL;
1028
1029         return ret;
1030 };
1031
1032
1033 enum {
1034         OCI_HOOKS_PRESTART,
1035         OCI_HOOKS_CREATERUNTIME,
1036         OCI_HOOKS_CREATECONTAINER,
1037         OCI_HOOKS_STARTCONTAINER,
1038         OCI_HOOKS_POSTSTART,
1039         OCI_HOOKS_POSTSTOP,
1040         __OCI_HOOKS_MAX,
1041 };
1042
1043 static const struct blobmsg_policy oci_hooks_policy[] = {
1044         [OCI_HOOKS_PRESTART] = { "prestart", BLOBMSG_TYPE_ARRAY },
1045         [OCI_HOOKS_CREATERUNTIME] = { "createRuntime", BLOBMSG_TYPE_ARRAY },
1046         [OCI_HOOKS_CREATECONTAINER] = { "createContainer", BLOBMSG_TYPE_ARRAY },
1047         [OCI_HOOKS_STARTCONTAINER] = { "startContainer", BLOBMSG_TYPE_ARRAY },
1048         [OCI_HOOKS_POSTSTART] = { "poststart", BLOBMSG_TYPE_ARRAY },
1049         [OCI_HOOKS_POSTSTOP] = { "poststop", BLOBMSG_TYPE_ARRAY },
1050 };
1051
1052 static int parseOCIhooks(struct blob_attr *msg)
1053 {
1054         struct blob_attr *tb[__OCI_HOOKS_MAX];
1055         int ret;
1056
1057         blobmsg_parse(oci_hooks_policy, __OCI_HOOKS_MAX, tb, blobmsg_data(msg), blobmsg_len(msg));
1058
1059         if (tb[OCI_HOOKS_PRESTART])
1060                 INFO("warning: ignoring deprecated prestart hook");
1061
1062         if (tb[OCI_HOOKS_CREATERUNTIME]) {
1063                 ret = parseOCIhook(&opts.hooks.createRuntime, tb[OCI_HOOKS_CREATERUNTIME]);
1064                 if (ret)
1065                         return ret;
1066         }
1067
1068         if (tb[OCI_HOOKS_CREATECONTAINER]) {
1069                 ret = parseOCIhook(&opts.hooks.createContainer, tb[OCI_HOOKS_CREATECONTAINER]);
1070                 if (ret)
1071                         goto out_createruntime;
1072         }
1073
1074         if (tb[OCI_HOOKS_STARTCONTAINER]) {
1075                 ret = parseOCIhook(&opts.hooks.startContainer, tb[OCI_HOOKS_STARTCONTAINER]);
1076                 if (ret)
1077                         goto out_createcontainer;
1078         }
1079
1080         if (tb[OCI_HOOKS_POSTSTART]) {
1081                 ret = parseOCIhook(&opts.hooks.poststart, tb[OCI_HOOKS_POSTSTART]);
1082                 if (ret)
1083                         goto out_startcontainer;
1084         }
1085
1086         if (tb[OCI_HOOKS_POSTSTOP]) {
1087                 ret = parseOCIhook(&opts.hooks.poststop, tb[OCI_HOOKS_POSTSTOP]);
1088                 if (ret)
1089                         goto out_poststart;
1090         }
1091
1092         return 0;
1093
1094 out_poststart:
1095         free_hooklist(opts.hooks.poststart);
1096 out_startcontainer:
1097         free_hooklist(opts.hooks.startContainer);
1098 out_createcontainer:
1099         free_hooklist(opts.hooks.createContainer);
1100 out_createruntime:
1101         free_hooklist(opts.hooks.createRuntime);
1102
1103         return ret;
1104 };
1105
1106
1107 enum {
1108         OCI_PROCESS_USER_UID,
1109         OCI_PROCESS_USER_GID,
1110         OCI_PROCESS_USER_UMASK,
1111         OCI_PROCESS_USER_ADDITIONALGIDS,
1112         __OCI_PROCESS_USER_MAX,
1113 };
1114
1115 static const struct blobmsg_policy oci_process_user_policy[] = {
1116         [OCI_PROCESS_USER_UID] = { "uid", BLOBMSG_TYPE_INT32 },
1117         [OCI_PROCESS_USER_GID] = { "gid", BLOBMSG_TYPE_INT32 },
1118         [OCI_PROCESS_USER_UMASK] = { "umask", BLOBMSG_TYPE_INT32 },
1119         [OCI_PROCESS_USER_ADDITIONALGIDS] = { "additionalGids", BLOBMSG_TYPE_ARRAY },
1120 };
1121
1122 static int parseOCIprocessuser(struct blob_attr *msg) {
1123         struct blob_attr *tb[__OCI_PROCESS_USER_MAX];
1124
1125         blobmsg_parse(oci_process_user_policy, __OCI_PROCESS_USER_MAX, tb, blobmsg_data(msg), blobmsg_len(msg));
1126
1127         if (tb[OCI_PROCESS_USER_UID])
1128                 opts.pw_uid = blobmsg_get_u32(tb[OCI_PROCESS_USER_UID]);
1129
1130         if (tb[OCI_PROCESS_USER_GID]) {
1131                 opts.pw_gid = blobmsg_get_u32(tb[OCI_PROCESS_USER_GID]);
1132                 opts.gr_gid = blobmsg_get_u32(tb[OCI_PROCESS_USER_GID]);
1133         }
1134
1135         /* ToDo: umask, additional GIDs */
1136
1137         return 0;
1138 }
1139
1140 enum {
1141         OCI_PROCESS_ARGS,
1142         OCI_PROCESS_CAPABILITIES,
1143         OCI_PROCESS_CWD,
1144         OCI_PROCESS_ENV,
1145         OCI_PROCESS_NONEWPRIVILEGES,
1146         OCI_PROCESS_RLIMITS,
1147         OCI_PROCESS_TERMINAL,
1148         OCI_PROCESS_USER,
1149         __OCI_PROCESS_MAX,
1150 };
1151
1152 static const struct blobmsg_policy oci_process_policy[] = {
1153         [OCI_PROCESS_ARGS] = { "args", BLOBMSG_TYPE_ARRAY },
1154         [OCI_PROCESS_CAPABILITIES] = { "capabilities", BLOBMSG_TYPE_TABLE },
1155         [OCI_PROCESS_CWD] = { "cwd", BLOBMSG_TYPE_STRING },
1156         [OCI_PROCESS_ENV] = { "env", BLOBMSG_TYPE_ARRAY },
1157         [OCI_PROCESS_NONEWPRIVILEGES] = { "noNewPrivileges", BLOBMSG_TYPE_BOOL },
1158         [OCI_PROCESS_RLIMITS] = { "rlimits", BLOBMSG_TYPE_ARRAY },
1159         [OCI_PROCESS_TERMINAL] = { "terminal", BLOBMSG_TYPE_BOOL },
1160         [OCI_PROCESS_USER] = { "user", BLOBMSG_TYPE_TABLE },
1161 };
1162
1163
1164 static int parseOCIprocess(struct blob_attr *msg)
1165 {
1166         struct blob_attr *tb[__OCI_PROCESS_MAX];
1167         int res;
1168
1169         blobmsg_parse(oci_process_policy, __OCI_PROCESS_MAX, tb, blobmsg_data(msg), blobmsg_len(msg));
1170
1171         if (!tb[OCI_PROCESS_ARGS])
1172                 return ENOENT;
1173
1174         res = parseOCIenvarray(tb[OCI_PROCESS_ARGS], &opts.jail_argv);
1175         if (res)
1176                 return res;
1177
1178         opts.console = blobmsg_get_bool(tb[OCI_PROCESS_TERMINAL]);
1179         opts.no_new_privs = blobmsg_get_bool(tb[OCI_PROCESS_NONEWPRIVILEGES]);
1180
1181         if (tb[OCI_PROCESS_CWD])
1182                 opts.cwd = strdup(blobmsg_get_string(tb[OCI_PROCESS_CWD]));
1183
1184         if (tb[OCI_PROCESS_ENV]) {
1185                 res = parseOCIenvarray(tb[OCI_PROCESS_ENV], &opts.envp);
1186                 if (res)
1187                         return res;
1188         }
1189
1190         if (tb[OCI_PROCESS_USER] && (res = parseOCIprocessuser(tb[OCI_PROCESS_USER])))
1191                 return res;
1192
1193         if (tb[OCI_PROCESS_CAPABILITIES] &&
1194             (res = parseOCIcapabilities(&opts.capset, tb[OCI_PROCESS_CAPABILITIES])))
1195                 return res;
1196
1197         /* ToDo: rlimits, capabilities */
1198
1199         return 0;
1200 }
1201
1202 enum {
1203         OCI_LINUX_NAMESPACE_TYPE,
1204         OCI_LINUX_NAMESPACE_PATH,
1205         __OCI_LINUX_NAMESPACE_MAX,
1206 };
1207
1208 static const struct blobmsg_policy oci_linux_namespace_policy[] = {
1209         [OCI_LINUX_NAMESPACE_TYPE] = { "type", BLOBMSG_TYPE_STRING },
1210         [OCI_LINUX_NAMESPACE_PATH] = { "path", BLOBMSG_TYPE_STRING },
1211 };
1212
1213 static unsigned int resolve_nstype(char *type) {
1214         if (!strcmp("pid", type))
1215                 return CLONE_NEWPID;
1216         else if (!strcmp("network", type))
1217                 return CLONE_NEWNET;
1218         else if (!strcmp("mount", type))
1219                 return CLONE_NEWNS;
1220         else if (!strcmp("ipc", type))
1221                 return CLONE_NEWIPC;
1222         else if (!strcmp("uts", type))
1223                 return CLONE_NEWUTS;
1224         else if (!strcmp("user", type))
1225                 return CLONE_NEWUSER;
1226         else if (!strcmp("cgroup", type))
1227                 return CLONE_NEWCGROUP;
1228         else
1229                 return 0;
1230 }
1231
1232 static int parseOCIlinuxns(struct blob_attr *msg)
1233 {
1234         struct blob_attr *tb[__OCI_LINUX_NAMESPACE_MAX];
1235
1236
1237         blobmsg_parse(oci_linux_namespace_policy, __OCI_LINUX_NAMESPACE_MAX, tb, blobmsg_data(msg), blobmsg_len(msg));
1238
1239         if (!tb[OCI_LINUX_NAMESPACE_TYPE])
1240                 return EINVAL;
1241
1242         if (tb[OCI_LINUX_NAMESPACE_PATH])
1243                 return ENOTSUP; /* ToDo */
1244
1245         opts.namespace |= resolve_nstype(blobmsg_get_string(tb[OCI_LINUX_NAMESPACE_TYPE]));
1246
1247         return 0;
1248 };
1249
1250
1251 enum {
1252         OCI_LINUX_UIDGIDMAP_CONTAINERID,
1253         OCI_LINUX_UIDGIDMAP_HOSTID,
1254         OCI_LINUX_UIDGIDMAP_SIZE,
1255         __OCI_LINUX_UIDGIDMAP_MAX,
1256 };
1257
1258 static const struct blobmsg_policy oci_linux_uidgidmap_policy[] = {
1259         [OCI_LINUX_UIDGIDMAP_CONTAINERID] = { "containerID", BLOBMSG_TYPE_INT32 },
1260         [OCI_LINUX_UIDGIDMAP_HOSTID] = { "hostID", BLOBMSG_TYPE_INT32 },
1261         [OCI_LINUX_UIDGIDMAP_SIZE] = { "size", BLOBMSG_TYPE_INT32 },
1262 };
1263
1264 static int parseOCIuidgidmappings(struct blob_attr *msg, bool is_gidmap)
1265 {
1266         const char *map_format = "%d %d %d\n";
1267         struct blob_attr *tb[__OCI_LINUX_UIDGIDMAP_MAX];
1268         struct blob_attr *cur;
1269         int rem, len;
1270         char **mappings;
1271         char *map, *curstr;
1272         unsigned int cnt = 0;
1273         size_t totallen = 0;
1274
1275         /* count number of mappings */
1276         blobmsg_for_each_attr(cur, msg, rem)
1277                 cnt++;
1278
1279         if (!cnt)
1280                 return 0;
1281
1282         /* allocate array for mappings */
1283         mappings = calloc(1 + cnt, sizeof(char*));
1284         if (!mappings)
1285                 return ENOMEM;
1286
1287         mappings[cnt] = NULL;
1288
1289         cnt = 0;
1290         blobmsg_for_each_attr(cur, msg, rem) {
1291                 blobmsg_parse(oci_linux_uidgidmap_policy, __OCI_LINUX_UIDGIDMAP_MAX, tb, blobmsg_data(cur), blobmsg_len(cur));
1292
1293                 if (!tb[OCI_LINUX_UIDGIDMAP_CONTAINERID] ||
1294                     !tb[OCI_LINUX_UIDGIDMAP_HOSTID] ||
1295                     !tb[OCI_LINUX_UIDGIDMAP_SIZE])
1296                         return EINVAL;
1297
1298                 /* write mapping line into allocated string */
1299                 len = asprintf(&mappings[cnt++], map_format,
1300                          blobmsg_get_u32(tb[OCI_LINUX_UIDGIDMAP_CONTAINERID]),
1301                          blobmsg_get_u32(tb[OCI_LINUX_UIDGIDMAP_HOSTID]),
1302                          blobmsg_get_u32(tb[OCI_LINUX_UIDGIDMAP_SIZE]));
1303
1304                 if (len < 0)
1305                         return ENOMEM;
1306
1307                 totallen += len;
1308         }
1309
1310         /* allocate combined mapping string */
1311         map = calloc(1 + totallen, sizeof(char));
1312         if (!map)
1313                 return ENOMEM;
1314
1315         map[0] = '\0';
1316
1317         /* concatenate mapping strings into combined string */
1318         curstr = mappings[0];
1319         while (curstr) {
1320                 strcat(map, curstr);
1321                 free(curstr++);
1322         }
1323         free(mappings);
1324
1325         if (is_gidmap)
1326                 opts.gidmap = map;
1327         else
1328                 opts.uidmap = map;
1329
1330         return 0;
1331 }
1332
1333 enum {
1334         OCI_LINUX_RESOURCES,
1335         OCI_LINUX_SECCOMP,
1336         OCI_LINUX_SYSCTL,
1337         OCI_LINUX_NAMESPACES,
1338         OCI_LINUX_UIDMAPPINGS,
1339         OCI_LINUX_GIDMAPPINGS,
1340         OCI_LINUX_MASKEDPATHS,
1341         OCI_LINUX_READONLYPATHS,
1342         OCI_LINUX_ROOTFSPROPAGATION,
1343         __OCI_LINUX_MAX,
1344 };
1345
1346 static const struct blobmsg_policy oci_linux_policy[] = {
1347         [OCI_LINUX_RESOURCES] = { "resources", BLOBMSG_TYPE_TABLE },
1348         [OCI_LINUX_SECCOMP] = { "seccomp", BLOBMSG_TYPE_TABLE },
1349         [OCI_LINUX_SYSCTL] = { "sysctl", BLOBMSG_TYPE_TABLE },
1350         [OCI_LINUX_NAMESPACES] = { "namespaces", BLOBMSG_TYPE_ARRAY },
1351         [OCI_LINUX_UIDMAPPINGS] = { "uidMappings", BLOBMSG_TYPE_ARRAY },
1352         [OCI_LINUX_GIDMAPPINGS] = { "gidMappings", BLOBMSG_TYPE_ARRAY },
1353         [OCI_LINUX_MASKEDPATHS] = { "maskedPaths", BLOBMSG_TYPE_ARRAY },
1354         [OCI_LINUX_READONLYPATHS] = { "readonlyPaths", BLOBMSG_TYPE_ARRAY },
1355         [OCI_LINUX_ROOTFSPROPAGATION] = { "rootfsPropagation", BLOBMSG_TYPE_STRING },
1356 };
1357
1358 static int parseOCIlinux(struct blob_attr *msg)
1359 {
1360         struct blob_attr *tb[__OCI_LINUX_MAX];
1361         struct blob_attr *cur;
1362         int rem;
1363         int res = 0;
1364
1365         blobmsg_parse(oci_linux_policy, __OCI_LINUX_MAX, tb, blobmsg_data(msg), blobmsg_len(msg));
1366
1367         if (tb[OCI_LINUX_NAMESPACES]) {
1368                 blobmsg_for_each_attr(cur, tb[OCI_LINUX_NAMESPACES], rem) {
1369                         res = parseOCIlinuxns(cur);
1370                         if (res)
1371                                 return res;
1372                 }
1373         }
1374
1375         if (tb[OCI_LINUX_UIDMAPPINGS]) {
1376                 res = parseOCIuidgidmappings(tb[OCI_LINUX_GIDMAPPINGS], 0);
1377                 if (res)
1378                         return res;
1379         }
1380
1381         if (tb[OCI_LINUX_GIDMAPPINGS]) {
1382                 res = parseOCIuidgidmappings(tb[OCI_LINUX_GIDMAPPINGS], 1);
1383                 if (res)
1384                         return res;
1385         }
1386
1387         if (tb[OCI_LINUX_SECCOMP]) {
1388                 opts.ociseccomp = parseOCIlinuxseccomp(tb[OCI_LINUX_SECCOMP]);
1389                 if (!opts.ociseccomp)
1390                         return EINVAL;
1391         }
1392
1393         return 0;
1394 }
1395
1396 enum {
1397         OCI_VERSION,
1398         OCI_HOSTNAME,
1399         OCI_PROCESS,
1400         OCI_ROOT,
1401         OCI_MOUNTS,
1402         OCI_HOOKS,
1403         OCI_LINUX,
1404         __OCI_MAX,
1405 };
1406
1407 static const struct blobmsg_policy oci_policy[] = {
1408         [OCI_VERSION] = { "ociVersion", BLOBMSG_TYPE_STRING },
1409         [OCI_HOSTNAME] = { "hostname", BLOBMSG_TYPE_STRING },
1410         [OCI_PROCESS] = { "process", BLOBMSG_TYPE_TABLE },
1411         [OCI_ROOT] = { "root", BLOBMSG_TYPE_TABLE },
1412         [OCI_MOUNTS] = { "mounts", BLOBMSG_TYPE_ARRAY },
1413         [OCI_HOOKS] = { "hooks", BLOBMSG_TYPE_TABLE },
1414         [OCI_LINUX] = { "linux", BLOBMSG_TYPE_TABLE },
1415 };
1416
1417 static int parseOCI(const char *jsonfile)
1418 {
1419         struct blob_attr *tb[__OCI_MAX];
1420         struct blob_attr *cur;
1421         int rem;
1422         int res;
1423
1424         blob_buf_init(&ocibuf, 0);
1425         if (!blobmsg_add_json_from_file(&ocibuf, jsonfile))
1426                 return ENOENT;
1427
1428         blobmsg_parse(oci_policy, __OCI_MAX, tb, blob_data(ocibuf.head), blob_len(ocibuf.head));
1429
1430         if (!tb[OCI_VERSION])
1431                 return ENOMSG;
1432
1433         if (strncmp("1.0", blobmsg_get_string(tb[OCI_VERSION]), 3)) {
1434                 ERROR("unsupported ociVersion %s\n", blobmsg_get_string(tb[OCI_VERSION]));
1435                 return ENOTSUP;
1436         }
1437
1438         if (tb[OCI_HOSTNAME])
1439                 opts.hostname = strdup(blobmsg_get_string(tb[OCI_HOSTNAME]));
1440
1441         if (!tb[OCI_PROCESS])
1442                 return ENODATA;
1443
1444         if ((res = parseOCIprocess(tb[OCI_PROCESS])))
1445                 return res;
1446
1447         if (!tb[OCI_ROOT])
1448                 return ENODATA;
1449
1450         if ((res = parseOCIroot(jsonfile, tb[OCI_ROOT])))
1451                 return res;
1452
1453         if (!tb[OCI_MOUNTS])
1454                 return ENODATA;
1455
1456         blobmsg_for_each_attr(cur, tb[OCI_MOUNTS], rem)
1457                 if ((res = parseOCImount(cur)))
1458                         return res;
1459
1460         if (tb[OCI_LINUX] && (res = parseOCIlinux(tb[OCI_LINUX])))
1461                 return res;
1462
1463         if (tb[OCI_HOOKS] && (res = parseOCIhooks(tb[OCI_HOOKS])))
1464                 return res;
1465
1466         blob_buf_free(&ocibuf);
1467
1468         return 0;
1469 }
1470
1471 int main(int argc, char **argv)
1472 {
1473         sigset_t sigmask;
1474         uid_t uid = getuid();
1475         const char log[] = "/dev/log";
1476         const char ubus[] = "/var/run/ubus.sock";
1477         char *jsonfile = NULL;
1478         int ch, i;
1479         int pipes[4];
1480         char sig_buf[1];
1481         int netns_fd;
1482
1483         if (uid) {
1484                 ERROR("not root, aborting: %m\n");
1485                 return EXIT_FAILURE;
1486         }
1487
1488         umask(022);
1489         mount_list_init();
1490         init_library_search();
1491
1492         while ((ch = getopt(argc, argv, OPT_ARGS)) != -1) {
1493                 switch (ch) {
1494                 case 'd':
1495                         debug = atoi(optarg);
1496                         break;
1497                 case 'p':
1498                         opts.namespace |= CLONE_NEWNS;
1499                         opts.procfs = 1;
1500                         break;
1501                 case 'o':
1502                         opts.namespace |= CLONE_NEWNS;
1503                         opts.ronly = 1;
1504                         break;
1505                 case 'f':
1506                         opts.namespace |= CLONE_NEWUSER;
1507                         break;
1508                 case 'F':
1509                         opts.namespace |= CLONE_NEWCGROUP;
1510                         break;
1511                 case 'R':
1512                         opts.extroot = strdup(optarg);
1513                         break;
1514                 case 's':
1515                         opts.namespace |= CLONE_NEWNS;
1516                         opts.sysfs = 1;
1517                         break;
1518                 case 'S':
1519                         opts.seccomp = optarg;
1520                         add_mount_bind(optarg, 1, -1);
1521                         break;
1522                 case 'C':
1523                         opts.capabilities = optarg;
1524                         break;
1525                 case 'c':
1526                         opts.no_new_privs = 1;
1527                         break;
1528                 case 'n':
1529                         opts.name = optarg;
1530                         break;
1531                 case 'N':
1532                         opts.namespace |= CLONE_NEWNET;
1533                         break;
1534                 case 'h':
1535                         opts.namespace |= CLONE_NEWUTS;
1536                         opts.hostname = strdup(optarg);
1537                         break;
1538                 case 'r':
1539                         opts.namespace |= CLONE_NEWNS;
1540                         add_path_and_deps(optarg, 1, 0, 0);
1541                         break;
1542                 case 'w':
1543                         opts.namespace |= CLONE_NEWNS;
1544                         add_path_and_deps(optarg, 0, 0, 0);
1545                         break;
1546                 case 'u':
1547                         opts.namespace |= CLONE_NEWNS;
1548                         add_mount_bind(ubus, 0, -1);
1549                         break;
1550                 case 'l':
1551                         opts.namespace |= CLONE_NEWNS;
1552                         add_mount_bind(log, 0, -1);
1553                         break;
1554                 case 'U':
1555                         opts.user = optarg;
1556                         break;
1557                 case 'G':
1558                         opts.group = optarg;
1559                         break;
1560                 case 'O':
1561                         opts.overlaydir = optarg;
1562                         break;
1563                 case 'T':
1564                         opts.tmpoverlaysize = optarg;
1565                         break;
1566                 case 'E':
1567                         opts.require_jail = 1;
1568                         break;
1569                 case 'y':
1570                         opts.console = 1;
1571                         break;
1572                 case 'J':
1573                         asprintf(&jsonfile, "%s/config.json", optarg);
1574                         break;
1575                 }
1576         }
1577
1578         if (opts.namespace)
1579                 opts.namespace |= CLONE_NEWIPC | CLONE_NEWPID;
1580
1581         if (jsonfile) {
1582                 int ocires;
1583                 ocires = parseOCI(jsonfile);
1584                 free(jsonfile);
1585                 if (ocires) {
1586                         ERROR("parsing of OCI JSON spec has failed: %s (%d)\n", strerror(ocires), ocires);
1587                         return ocires;
1588                 }
1589         }
1590
1591         if (opts.tmpoverlaysize && strlen(opts.tmpoverlaysize) > 8) {
1592                 ERROR("size parameter too long: \"%s\"\n", opts.tmpoverlaysize);
1593                 return -1;
1594         }
1595
1596         /* no <binary> param found */
1597         if (!jsonfile && (argc - optind < 1)) {
1598                 usage();
1599                 return EXIT_FAILURE;
1600         }
1601         if (!(opts.namespace||opts.capabilities||opts.seccomp)) {
1602                 ERROR("Not using namespaces, capabilities or seccomp !!!\n\n");
1603                 usage();
1604                 return EXIT_FAILURE;
1605         }
1606         DEBUG("Using namespaces(0x%08x), capabilities(%d), seccomp(%d)\n",
1607                 opts.namespace,
1608                 opts.capabilities != 0 || opts.capset.apply,
1609                 opts.seccomp != 0 || opts.ociseccomp != 0);
1610
1611         if (!jsonfile) {
1612                 /* allocate NULL-terminated array for argv */
1613                 opts.jail_argv = calloc(1 + argc - optind, sizeof(char**));
1614                 if (!opts.jail_argv)
1615                         return EXIT_FAILURE;
1616
1617                 for (size_t s = optind; s < argc; s++)
1618                         opts.jail_argv[s - optind] = strdup(argv[s]);
1619
1620                 if (opts.namespace & CLONE_NEWUSER)
1621                         get_jail_user(&opts.pw_uid, &opts.pw_gid, &opts.gr_gid);
1622         }
1623
1624         if (!opts.extroot) {
1625                 if (opts.namespace && add_path_and_deps(*opts.jail_argv, 1, -1, 0)) {
1626                         ERROR("failed to load dependencies\n");
1627                         return -1;
1628                 }
1629         }
1630
1631         if (opts.namespace && opts.seccomp && add_path_and_deps("libpreload-seccomp.so", 1, -1, 1)) {
1632                 ERROR("failed to load libpreload-seccomp.so\n");
1633                 opts.seccomp = 0;
1634                 if (opts.require_jail)
1635                         return -1;
1636         }
1637
1638         if (opts.name)
1639                 prctl(PR_SET_NAME, opts.name, NULL, NULL, NULL);
1640
1641         uloop_init();
1642
1643         sigfillset(&sigmask);
1644         for (i = 0; i < _NSIG; i++) {
1645                 struct sigaction s = { 0 };
1646
1647                 if (!sigismember(&sigmask, i))
1648                         continue;
1649                 if ((i == SIGCHLD) || (i == SIGPIPE) || (i == SIGSEGV))
1650                         continue;
1651
1652                 s.sa_handler = jail_handle_signal;
1653                 sigaction(i, &s, NULL);
1654         }
1655
1656         if (opts.namespace) {
1657                 if (opts.namespace & CLONE_NEWNS) {
1658                         add_mount_bind("/dev/full", 0, -1);
1659                         add_mount_bind("/dev/null", 0, -1);
1660                         add_mount_bind("/dev/random", 0, -1);
1661                         add_mount_bind("/dev/urandom", 0, -1);
1662                         add_mount_bind("/dev/zero", 0, -1);
1663                         add_mount_bind("/dev/ptmx", 0, -1);
1664                         add_mount_bind("/dev/tty", 0, -1);
1665
1666                         if (!opts.extroot && (opts.user || opts.group)) {
1667                                 add_mount_bind("/etc/passwd", 0, -1);
1668                                 add_mount_bind("/etc/group", 0, -1);
1669                         }
1670
1671 #if defined(__GLIBC__)
1672                         if (!opts.extroot)
1673                                 add_mount_bind("/etc/nsswitch.conf", 0, -1);
1674 #endif
1675
1676                         if (!(opts.namespace & CLONE_NEWNET)) {
1677                                 add_mount_bind("/etc/resolv.conf", 0, -1);
1678                         } else {
1679                                 char hostdir[PATH_MAX];
1680
1681                                 snprintf(hostdir, PATH_MAX, "/tmp/resolv.conf-%s.d", opts.name);
1682                                 mkdir_p(hostdir, 0755);
1683                                 add_mount(hostdir, "/tmp/resolv.conf.d", NULL, MS_BIND | MS_NOEXEC | MS_NOATIME | MS_NOSUID | MS_NODEV | MS_RDONLY, NULL, -1);
1684                         }
1685                 }
1686
1687                 if (pipe(&pipes[0]) < 0 || pipe(&pipes[2]) < 0)
1688                         return -1;
1689
1690                 jail_process.pid = clone(exec_jail, child_stack + STACK_SIZE, SIGCHLD | opts.namespace, &pipes);
1691         } else {
1692                 jail_process.pid = fork();
1693         }
1694
1695         if (jail_process.pid > 0) {
1696                 seteuid(0);
1697                 /* parent process */
1698                 close(pipes[1]);
1699                 close(pipes[2]);
1700                 run_hooks(opts.hooks.createRuntime);
1701                 if (read(pipes[0], sig_buf, 1) < 1) {
1702                         ERROR("can't read from child\n");
1703                         return -1;
1704                 }
1705                 close(pipes[0]);
1706                 if (opts.namespace & CLONE_NEWUSER) {
1707                         if (write_setgroups(jail_process.pid, true)) {
1708                                 ERROR("can't write setgroups\n");
1709                                 return -1;
1710                         }
1711                         if (!opts.uidmap) {
1712                                 bool has_gr = (opts.gr_gid != -1);
1713                                 if (opts.pw_uid != -1) {
1714                                         write_single_uid_gid_map(jail_process.pid, 0, opts.pw_uid);
1715                                         write_single_uid_gid_map(jail_process.pid, 1, has_gr?opts.gr_gid:opts.pw_gid);
1716                                 } else {
1717                                         write_single_uid_gid_map(jail_process.pid, 0, 65534);
1718                                         write_single_uid_gid_map(jail_process.pid, 1, has_gr?opts.gr_gid:65534);
1719                                 }
1720                         } else {
1721                                 write_uid_gid_map(jail_process.pid, 0, opts.uidmap);
1722                                 if (opts.gidmap)
1723                                         write_uid_gid_map(jail_process.pid, 1, opts.gidmap);
1724                         }
1725                 }
1726
1727                 if (opts.namespace & CLONE_NEWNET) {
1728                         if (!opts.name) {
1729                                 ERROR("netns needs a named jail\n");
1730                                 return -1;
1731                         }
1732                         netns_fd = netns_open_pid(jail_process.pid);
1733                         netns_updown(jail_process.pid, true);
1734                 }
1735
1736                 sig_buf[0] = 'O';
1737                 if (write(pipes[3], sig_buf, 1) < 0) {
1738                         ERROR("can't write to child\n");
1739                         return -1;
1740                 }
1741                 close(pipes[3]);
1742                 run_hooks(opts.hooks.poststart);
1743                 uloop_process_add(&jail_process);
1744                 uloop_run();
1745                 if (jail_running) {
1746                         DEBUG("uloop interrupted, killing jail process\n");
1747                         kill(jail_process.pid, SIGTERM);
1748                         uloop_timeout_set(&jail_process_timeout, 1000);
1749                         uloop_run();
1750                 }
1751                 uloop_done();
1752                 if (opts.namespace & CLONE_NEWNET) {
1753                         setns(netns_fd, CLONE_NEWNET);
1754                         netns_updown(getpid(), false);
1755                         close(netns_fd);
1756                 }
1757                 run_hooks(opts.hooks.poststop);
1758                 free_opts(true);
1759                 return jail_return_code;
1760         } else if (jail_process.pid == 0) {
1761                 /* fork child process */
1762                 return exec_jail(NULL);
1763         } else {
1764                 ERROR("failed to clone/fork: %m\n");
1765                 return EXIT_FAILURE;
1766         }
1767 }