src/lxc/start.c

   1 /*
   2  * lxc: linux Container library
   3  *
   4  * (C) Copyright IBM Corp. 2007, 2008
   5  *
   6  * Authors:
   7  * Daniel Lezcano <daniel.lezcano at free.fr>
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23
  24 #define _GNU_SOURCE
  25 #include "config.h"
  26
  27 #include <alloca.h>
  28 #include <dirent.h>
  29 #include <errno.h>
  30 #include <fcntl.h>
  31 #include <grp.h>
  32 #include <poll.h>
  33 #include <signal.h>
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <string.h>
  37 #include <unistd.h>
  38 #include <sys/file.h>
  39 #include <sys/mount.h>
  40 #include <sys/param.h>
  41 #include <sys/prctl.h>
  42 #include <sys/socket.h>
  43 #include <sys/stat.h>
  44 #include <sys/syscall.h>
  45 #include <sys/types.h>
  46 #include <sys/un.h>
  47 #include <sys/wait.h>
  48
  49 #if HAVE_LIBCAP
  50 #include <sys/capability.h>
  51 #endif
  52
  53 #ifndef HAVE_DECL_PR_CAPBSET_DROP
  54 #define PR_CAPBSET_DROP 24
  55 #endif
  56
  57 #ifndef HAVE_DECL_PR_SET_NO_NEW_PRIVS
  58 #define PR_SET_NO_NEW_PRIVS 38
  59 #endif
  60
  61 #ifndef HAVE_DECL_PR_GET_NO_NEW_PRIVS
  62 #define PR_GET_NO_NEW_PRIVS 39
  63 #endif
  64
  65 #include "af_unix.h"
  66 #include "bdev.h"
  67 #include "caps.h"
  68 #include "cgroup.h"
  69 #include "commands.h"
  70 #include "conf.h"
  71 #include "console.h"
  72 #include "error.h"
  73 #include "log.h"
  74 #include "lxclock.h"
  75 #include "lxcseccomp.h"
  76 #include "lxcutmp.h"
  77 #include "mainloop.h"
  78 #include "monitor.h"
  79 #include "namespace.h"
  80 #include "start.h"
  81 #include "sync.h"
  82 #include "utils.h"
  83 #include "lsm/lsm.h"
  84
  85 lxc_log_define(lxc_start, lxc);
  86
  87 extern void mod_all_rdeps(struct lxc_container *c, bool inc);
  88 static bool do_destroy_container(struct lxc_conf *conf);
  89 static int lxc_rmdir_onedev_wrapper(void *data);
  90 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
  91                                             const char *name);
  92
  93 static void print_top_failing_dir(const char *path)
  94 {
  95         size_t len = strlen(path);
  96         char *copy = alloca(len + 1), *p, *e, saved;
  97         strcpy(copy, path);
  98
  99         p = copy;
 100         e = copy + len;
 101         while (p < e) {
 102                 while (p < e && *p == '/')
 103                         p++;
 104                 while (p < e && *p != '/')
 105                         p++;
 106                 saved = *p;
 107                 *p = '\0';
 108                 if (access(copy, X_OK)) {
 109                         SYSERROR("Could not access %s. Please grant it x "
 110                                  "access, or add an ACL for the container "
 111                                  "root.", copy);
 112                         return;
 113                 }
 114                 *p = saved;
 115         }
 116 }
 117
 118 static void close_ns(int ns_fd[LXC_NS_MAX])
 119 {
 120         int i;
 121
 122         for (i = 0; i < LXC_NS_MAX; i++) {
 123                 if (ns_fd[i] > -1) {
 124                         close(ns_fd[i]);
 125                         ns_fd[i] = -1;
 126                 }
 127         }
 128 }
 129
 130 /* preserve_ns: open /proc/@pid/ns/@ns for each namespace specified
 131  * in clone_flags.
 132  * Return true on success, false on failure.
 133  */
 134 static bool preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags, pid_t pid)
 135 {
 136         int i, ret;
 137
 138         for (i = 0; i < LXC_NS_MAX; i++)
 139                 ns_fd[i] = -1;
 140
 141         ret = lxc_preserve_ns(pid, "");
 142         if (ret < 0) {
 143                 SYSERROR("Kernel does not support attaching to namespaces.");
 144                 return false;
 145         } else {
 146                 close(ret);
 147         }
 148
 149         for (i = 0; i < LXC_NS_MAX; i++) {
 150                 if ((clone_flags & ns_info[i].clone_flag) == 0)
 151                         continue;
 152                 ns_fd[i] = lxc_preserve_ns(pid, ns_info[i].proc_name);
 153                 if (ns_fd[i] < 0)
 154                         goto error;
 155         }
 156
 157         return true;
 158
 159 error:
 160         if (errno == ENOENT)
 161                 SYSERROR("Kernel does not support attaching to %s namespaces.", ns_info[i].proc_name);
 162         else
 163                 SYSERROR("Failed to open file descriptor for %s namespace: %s.", ns_info[i].proc_name, strerror(errno));
 164         close_ns(ns_fd);
 165         return false;
 166 }
 167
 168 static int attach_ns(const int ns_fd[LXC_NS_MAX]) {
 169         int i;
 170
 171         for (i = 0; i < LXC_NS_MAX; i++) {
 172                 if (ns_fd[i] < 0)
 173                         continue;
 174
 175                 if (setns(ns_fd[i], 0) != 0)
 176                         goto error;
 177         }
 178         return 0;
 179
 180 error:
 181         SYSERROR("Failed to attach %s namespace.", ns_info[i].proc_name);
 182         return -1;
 183 }
 184
 185 static int match_fd(int fd)
 186 {
 187         return (fd == 0 || fd == 1 || fd == 2);
 188 }
 189
 190 /* Check for any fds we need to close.
 191  * - If fd_to_ignore != -1, then if we find that fd open we will ignore it.
 192  * - By default we warn about open fds we find.
 193  * - If closeall is true, we will close open fds.
 194  * - If lxc-start was passed "-C", then conf->close_all_fds will be true, in
 195  *   which case we also close all open fds.
 196  * - A daemonized container will always pass closeall=true.
 197  */
 198 int lxc_check_inherited(struct lxc_conf *conf, bool closeall, int fd_to_ignore)
 199 {
 200         struct dirent *direntp;
 201         int fd, fddir;
 202         DIR *dir;
 203
 204         if (conf && conf->close_all_fds)
 205                 closeall = true;
 206
 207 restart:
 208         dir = opendir("/proc/self/fd");
 209         if (!dir) {
 210                 WARN("Failed to open directory: %s.", strerror(errno));
 211                 return -1;
 212         }
 213
 214         fddir = dirfd(dir);
 215
 216         while ((direntp = readdir(dir))) {
 217                 if (!direntp)
 218                         break;
 219
 220                 if (!strcmp(direntp->d_name, "."))
 221                         continue;
 222
 223                 if (!strcmp(direntp->d_name, ".."))
 224                         continue;
 225
 226                 if (lxc_safe_int(direntp->d_name, &fd) < 0) {
 227                         INFO("Could not parse file descriptor for: %s", direntp->d_name);
 228                         continue;
 229                 }
 230
 231                 if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore)
 232                         continue;
 233
 234                 if (current_config && fd == current_config->logfd)
 235                         continue;
 236
 237                 if (match_fd(fd))
 238                         continue;
 239
 240                 if (closeall) {
 241                         close(fd);
 242                         closedir(dir);
 243                         INFO("Closed inherited fd: %d.", fd);
 244                         goto restart;
 245                 }
 246                 WARN("Inherited fd: %d.", fd);
 247         }
 248
 249         /* Only enable syslog at this point to avoid the above logging function
 250          * to open a new fd and make the check_inherited function enter an
 251          * infinite loop.
 252          */
 253         lxc_log_enable_syslog();
 254
 255         closedir(dir); /* cannot fail */
 256         return 0;
 257 }
 258
 259 static int setup_signal_fd(sigset_t *oldmask)
 260 {
 261         sigset_t mask;
 262         int fd;
 263
 264         /* Block everything except serious error signals. */
 265         if (sigfillset(&mask) ||
 266             sigdelset(&mask, SIGILL) ||
 267             sigdelset(&mask, SIGSEGV) ||
 268             sigdelset(&mask, SIGBUS) ||
 269             sigdelset(&mask, SIGWINCH) ||
 270             sigprocmask(SIG_BLOCK, &mask, oldmask)) {
 271                 SYSERROR("Failed to set signal mask.");
 272                 return -1;
 273         }
 274
 275         fd = signalfd(-1, &mask, 0);
 276         if (fd < 0) {
 277                 SYSERROR("Failed to create signal file descriptor.");
 278                 return -1;
 279         }
 280
 281         if (fcntl(fd, F_SETFD, FD_CLOEXEC)) {
 282                 SYSERROR("Failed to set FD_CLOEXEC on the signal file descriptor: %d.", fd);
 283                 close(fd);
 284                 return -1;
 285         }
 286
 287         DEBUG("Set SIGCHLD handler with file descriptor: %d.", fd);
 288
 289         return fd;
 290 }
 291
 292 static int signal_handler(int fd, uint32_t events, void *data,
 293                           struct lxc_epoll_descr *descr)
 294 {
 295         struct signalfd_siginfo siginfo;
 296         siginfo_t info;
 297         int ret;
 298         pid_t *pid = data;
 299         bool init_died = false;
 300
 301         ret = read(fd, &siginfo, sizeof(siginfo));
 302         if (ret < 0) {
 303                 ERROR("Failed to read signal info from signal file descriptor: %d.", fd);
 304                 return -1;
 305         }
 306
 307         if (ret != sizeof(siginfo)) {
 308                 ERROR("Unexpected size for siginfo struct.");
 309                 return -1;
 310         }
 311
 312         /* Check whether init is running. */
 313         info.si_pid = 0;
 314         ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG);
 315         if (ret == 0 && info.si_pid == *pid)
 316                 init_died = true;
 317
 318         if (siginfo.ssi_signo != SIGCHLD) {
 319                 kill(*pid, siginfo.ssi_signo);
 320                 INFO("Forwarded signal %d to pid %d.", siginfo.ssi_signo, *pid);
 321                 return init_died ? 1 : 0;
 322         }
 323
 324         if (siginfo.ssi_code == CLD_STOPPED) {
 325                 INFO("Container init process was stopped.");
 326                 return init_died ? 1 : 0;
 327         } else if (siginfo.ssi_code == CLD_CONTINUED) {
 328                 INFO("Container init process was continued.");
 329                 return init_died ? 1 : 0;
 330         }
 331
 332         /* More robustness, protect ourself from a SIGCHLD sent
 333          * by a process different from the container init.
 334          */
 335         if (siginfo.ssi_pid != *pid) {
 336                 NOTICE("Received SIGCHLD from pid %d instead of container init %d.", siginfo.ssi_pid, *pid);
 337                 return init_died ? 1 : 0;
 338         }
 339
 340         DEBUG("Container init process %d exited.", *pid);
 341         return 1;
 342 }
 343
 344 int lxc_set_state(const char *name, struct lxc_handler *handler,
 345                   lxc_state_t state)
 346 {
 347         ssize_t ret;
 348         struct lxc_list *cur, *next;
 349         struct state_client *client;
 350         struct lxc_msg msg = {.type = lxc_msg_state, .value = state};
 351
 352         process_lock();
 353         /* Only set state under process lock held so that we don't cause
 354          * lxc_cmd_state_server() to miss a state.
 355          */
 356         handler->state = state;
 357         TRACE("set container state to %s", lxc_state2str(state));
 358
 359         if (lxc_list_empty(&handler->state_clients)) {
 360                 TRACE("no state clients registered");
 361                 process_unlock();
 362                 return 0;
 363         }
 364
 365         strncpy(msg.name, name, sizeof(msg.name));
 366         msg.name[sizeof(msg.name) - 1] = 0;
 367
 368         lxc_list_for_each_safe(cur, &handler->state_clients, next) {
 369                 client = cur->elem;
 370
 371                 if (!client->states[state]) {
 372                         TRACE("state %s not registered for state client %d",
 373                               lxc_state2str(state), client->clientfd);
 374                         continue;
 375                 }
 376
 377                 TRACE("sending state %s to state client %d",
 378                       lxc_state2str(state), client->clientfd);
 379
 380         again:
 381                 ret = send(client->clientfd, &msg, sizeof(msg), 0);
 382                 if (ret < 0) {
 383                         if (errno == EINTR)
 384                                 goto again;
 385
 386                         ERROR("failed to send message to client");
 387                 }
 388
 389                 /* kick client from list */
 390                 close(client->clientfd);
 391                 lxc_list_del(cur);
 392                 free(cur->elem);
 393                 free(cur);
 394         }
 395         process_unlock();
 396
 397         /* This function will try to connect to the legacy lxc-monitord state
 398          * server and only exists for backwards compatibility.
 399          */
 400         lxc_monitor_send_state(name, state, handler->lxcpath);
 401
 402         return 0;
 403 }
 404
 405 int lxc_poll(const char *name, struct lxc_handler *handler)
 406 {
 407         int sigfd = handler->sigfd;
 408         int pid = handler->pid;
 409         struct lxc_epoll_descr descr;
 410
 411         if (lxc_mainloop_open(&descr)) {
 412                 ERROR("Failed to create LXC mainloop.");
 413                 goto out_sigfd;
 414         }
 415
 416         if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) {
 417                 ERROR("Failed to add signal handler with file descriptor %d to LXC mainloop.", sigfd);
 418                 goto out_mainloop_open;
 419         }
 420
 421         if (lxc_console_mainloop_add(&descr, handler->conf)) {
 422                 ERROR("Failed to add console handler to LXC mainloop.");
 423                 goto out_mainloop_open;
 424         }
 425
 426         if (lxc_cmd_mainloop_add(name, &descr, handler)) {
 427                 ERROR("Failed to add command handler to LXC mainloop.");
 428                 goto out_mainloop_open;
 429         }
 430
 431         if (handler->conf->need_utmp_watch) {
 432                 #if HAVE_LIBCAP
 433                 if (lxc_utmp_mainloop_add(&descr, handler)) {
 434                         ERROR("Failed to add utmp handler to LXC mainloop.");
 435                         goto out_mainloop_open;
 436                 }
 437                 #else
 438                         DEBUG("Not starting utmp handler as CAP_SYS_BOOT cannot be dropped without capabilities support.");
 439                 #endif
 440         }
 441         TRACE("lxc mainloop is ready");
 442
 443         return lxc_mainloop(&descr, -1);
 444
 445 out_mainloop_open:
 446         lxc_mainloop_close(&descr);
 447
 448 out_sigfd:
 449         close(sigfd);
 450
 451         return -1;
 452 }
 453
 454 void lxc_free_handler(struct lxc_handler *handler)
 455 {
 456         if (handler->conf && handler->conf->maincmd_fd)
 457                 close(handler->conf->maincmd_fd);
 458
 459         if (handler->name)
 460                 free(handler->name);
 461
 462         handler->conf = NULL;
 463         free(handler);
 464 }
 465
 466 struct lxc_handler *lxc_init_handler(const char *name, struct lxc_conf *conf,
 467                                      const char *lxcpath)
 468 {
 469         int i;
 470         struct lxc_handler *handler;
 471
 472         handler = malloc(sizeof(*handler));
 473         if (!handler) {
 474                 ERROR("failed to allocate memory");
 475                 return NULL;
 476         }
 477
 478         memset(handler, 0, sizeof(*handler));
 479
 480         handler->ttysock[0] = handler->ttysock[1] = -1;
 481         handler->conf = conf;
 482         handler->lxcpath = lxcpath;
 483         handler->pinfd = -1;
 484         lxc_list_init(&handler->state_clients);
 485
 486         for (i = 0; i < LXC_NS_MAX; i++)
 487                 handler->nsfd[i] = -1;
 488
 489         handler->name = strdup(name);
 490         if (!handler->name) {
 491                 ERROR("failed to allocate memory");
 492                 goto on_error;
 493         }
 494
 495         if (lxc_cmd_init(name, handler, lxcpath)) {
 496                 ERROR("failed to set up command socket");
 497                 goto on_error;
 498         }
 499
 500         TRACE("unix domain socket %d for command server is ready",
 501               handler->conf->maincmd_fd);
 502
 503         return handler;
 504
 505 on_error:
 506         lxc_free_handler(handler);
 507
 508         return NULL;
 509 }
 510
 511 int lxc_init(const char *name, struct lxc_handler *handler)
 512 {
 513         struct lxc_conf *conf = handler->conf;
 514
 515         lsm_init();
 516         TRACE("initialized LSM");
 517
 518         if (lxc_read_seccomp_config(conf) != 0) {
 519                 ERROR("Failed loading seccomp policy.");
 520                 goto out_close_maincmd_fd;
 521         }
 522         TRACE("read seccomp policy");
 523
 524         /* Begin by setting the state to STARTING. */
 525         if (lxc_set_state(name, handler, STARTING)) {
 526                 ERROR("Failed to set state for container \"%s\" to \"%s\".", name, lxc_state2str(STARTING));
 527                 goto out_close_maincmd_fd;
 528         }
 529         TRACE("set container state to \"STARTING\"");
 530
 531         /* Start of environment variable setup for hooks. */
 532         if (name && setenv("LXC_NAME", name, 1))
 533                 SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name);
 534
 535         if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1))
 536                 SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf->rcfile);
 537
 538         if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1))
 539                 SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf->rootfs.mount);
 540
 541         if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1))
 542                 SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf->rootfs.path);
 543
 544         if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1))
 545                 SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf->console.path);
 546
 547         if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1))
 548                 SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf->console.log_path);
 549
 550         if (setenv("LXC_CGNS_AWARE", "1", 1))
 551                 SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1.");
 552         /* End of environment variable setup for hooks. */
 553
 554         TRACE("set environment variables");
 555
 556         if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) {
 557                 ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name);
 558                 goto out_aborting;
 559         }
 560         TRACE("ran pre-start hooks");
 561
 562         /* The signal fd has to be created before forking otherwise if the child
 563          * process exits before we setup the signal fd, the event will be lost
 564          * and the command will be stuck.
 565          */
 566         handler->sigfd = setup_signal_fd(&handler->oldmask);
 567         if (handler->sigfd < 0) {
 568                 ERROR("Failed to setup SIGCHLD fd handler.");
 569                 goto out_delete_tty;
 570         }
 571         TRACE("set up signal fd");
 572
 573         /* Do this after setting up signals since it might unblock SIGWINCH. */
 574         if (lxc_console_create(conf)) {
 575                 ERROR("Failed to create console for container \"%s\".", name);
 576                 goto out_restore_sigmask;
 577         }
 578         TRACE("created console");
 579
 580         if (lxc_ttys_shift_ids(conf) < 0) {
 581                 ERROR("Failed to shift tty into container.");
 582                 goto out_restore_sigmask;
 583         }
 584         TRACE("shifted tty ids");
 585
 586         INFO("container \"%s\" is initialized", name);
 587         return 0;
 588
 589 out_restore_sigmask:
 590         sigprocmask(SIG_SETMASK, &handler->oldmask, NULL);
 591 out_delete_tty:
 592         lxc_delete_tty(&conf->tty_info);
 593 out_aborting:
 594         lxc_set_state(name, handler, ABORTING);
 595 out_close_maincmd_fd:
 596         close(conf->maincmd_fd);
 597         conf->maincmd_fd = -1;
 598         return -1;
 599 }
 600
 601 void lxc_fini(const char *name, struct lxc_handler *handler)
 602 {
 603         int i, rc;
 604         struct lxc_list *cur, *next;
 605         pid_t self = getpid();
 606         char *namespaces[LXC_NS_MAX + 1];
 607         size_t namespace_count = 0;
 608
 609         /* The STOPPING state is there for future cleanup code which can take
 610          * awhile.
 611          */
 612         lxc_set_state(name, handler, STOPPING);
 613
 614         for (i = 0; i < LXC_NS_MAX; i++) {
 615                 if (handler->nsfd[i] != -1) {
 616                         rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d",
 617                                       ns_info[i].proc_name, self, handler->nsfd[i]);
 618                         if (rc == -1) {
 619                                 SYSERROR("Failed to allocate memory.");
 620                                 break;
 621                         }
 622                         ++namespace_count;
 623                 }
 624         }
 625         namespaces[namespace_count] = NULL;
 626
 627         if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1))
 628                 SYSERROR("Failed to set environment variable: LXC_TARGET=reboot.");
 629
 630         if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1))
 631                 SYSERROR("Failed to set environment variable: LXC_TARGET=stop.");
 632
 633         if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces))
 634                 ERROR("Failed to run lxc.hook.stop for container \"%s\".", name);
 635
 636         while (namespace_count--)
 637                 free(namespaces[namespace_count]);
 638         for (i = 0; i < LXC_NS_MAX; i++) {
 639                 if (handler->nsfd[i] != -1) {
 640                         close(handler->nsfd[i]);
 641                         handler->nsfd[i] = -1;
 642                 }
 643         }
 644
 645         if (handler->netnsfd >= 0) {
 646                 close(handler->netnsfd);
 647                 handler->netnsfd = -1;
 648         }
 649
 650         lxc_set_state(name, handler, STOPPED);
 651
 652         if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) {
 653                 ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name);
 654                 if (handler->conf->reboot) {
 655                         WARN("Container will be stopped instead of rebooted.");
 656                         handler->conf->reboot = 0;
 657                         if (setenv("LXC_TARGET", "stop", 1))
 658                                 WARN("Failed to set environment variable: LXC_TARGET=stop.");
 659                 }
 660         }
 661
 662         /* Reset mask set by setup_signal_fd. */
 663         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL))
 664                 WARN("Failed to restore signal mask.");
 665
 666         lxc_console_delete(&handler->conf->console);
 667         lxc_delete_tty(&handler->conf->tty_info);
 668
 669         /* close the command socket */
 670         close(handler->conf->maincmd_fd);
 671         handler->conf->maincmd_fd = -1;
 672
 673         /* The command socket is now closed, no more state clients can register
 674          * themselves from now on. So free the list of state clients.
 675          */
 676         lxc_list_for_each_safe(cur, &handler->state_clients, next) {
 677                 struct state_client *client = cur->elem;
 678                 /* close state client socket */
 679                 close(client->clientfd);
 680                 lxc_list_del(cur);
 681                 free(cur->elem);
 682                 free(cur);
 683         }
 684
 685         free(handler->name);
 686         if (handler->ttysock[0] != -1) {
 687                 close(handler->ttysock[0]);
 688                 close(handler->ttysock[1]);
 689         }
 690
 691         if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1)
 692                 lxc_destroy_container_on_signal(handler, name);
 693
 694         cgroup_destroy(handler);
 695         free(handler);
 696 }
 697
 698 void lxc_abort(const char *name, struct lxc_handler *handler)
 699 {
 700         int ret, status;
 701
 702         lxc_set_state(name, handler, ABORTING);
 703         if (handler->pid > 0)
 704                 kill(handler->pid, SIGKILL);
 705         while ((ret = waitpid(-1, &status, 0)) > 0) {
 706                 ;
 707         }
 708 }
 709
 710 #include <sys/reboot.h>
 711 #include <linux/reboot.h>
 712
 713 /* reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL in a child pid namespace
 714  * if container reboot support exists.  Otherwise, it will either succeed or
 715  * return -EPERM.
 716  */
 717 static int container_reboot_supported(void *arg)
 718 {
 719         int *cmd = arg;
 720         int ret;
 721
 722         ret = reboot(*cmd);
 723         if (ret == -1 && errno == EINVAL)
 724                 return 1;
 725         return 0;
 726 }
 727
 728 static int must_drop_cap_sys_boot(struct lxc_conf *conf)
 729 {
 730         FILE *f;
 731         int ret, cmd, v, flags;
 732         long stack_size = 4096;
 733         void *stack = alloca(stack_size);
 734         int status;
 735         pid_t pid;
 736
 737         f = fopen("/proc/sys/kernel/ctrl-alt-del", "r");
 738         if (!f) {
 739                 DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del");
 740                 return 1;
 741         }
 742
 743         ret = fscanf(f, "%d", &v);
 744         fclose(f);
 745         if (ret != 1) {
 746                 DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del.");
 747                 return 1;
 748         }
 749         cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF;
 750
 751         flags = CLONE_NEWPID | SIGCHLD;
 752         if (!lxc_list_empty(&conf->id_map))
 753                 flags |= CLONE_NEWUSER;
 754
 755 #ifdef __ia64__
 756         pid = __clone2(container_reboot_supported, stack, stack_size, flags,  &cmd);
 757 #else
 758         stack += stack_size;
 759         pid = clone(container_reboot_supported, stack, flags, &cmd);
 760 #endif
 761         if (pid < 0) {
 762                 if (flags & CLONE_NEWUSER)
 763                         ERROR("Failed to clone (%#x): %s (includes CLONE_NEWUSER).", flags, strerror(errno));
 764                 else
 765                         ERROR("Failed to clone (%#x): %s.", flags, strerror(errno));
 766                 return -1;
 767         }
 768         if (wait(&status) < 0) {
 769                 SYSERROR("Unexpected wait error: %s.", strerror(errno));
 770                 return -1;
 771         }
 772
 773         if (WEXITSTATUS(status) != 1)
 774                 return 1;
 775
 776         return 0;
 777 }
 778
 779 /* netpipe is used in the unprivileged case to transfer the ifindexes from
 780  * parent to child
 781  */
 782 static int netpipe = -1;
 783
 784 static inline int count_veths(struct lxc_list *network)
 785 {
 786         struct lxc_list *iterator;
 787         struct lxc_netdev *netdev;
 788         int count = 0;
 789
 790         lxc_list_for_each(iterator, network) {
 791                 netdev = iterator->elem;
 792                 if (netdev->type != LXC_NET_VETH)
 793                         continue;
 794                 count++;
 795         }
 796         return count;
 797 }
 798
 799 static int read_unpriv_netifindex(struct lxc_list *network)
 800 {
 801         struct lxc_list *iterator;
 802         struct lxc_netdev *netdev;
 803
 804         if (netpipe == -1)
 805                 return 0;
 806         lxc_list_for_each(iterator, network) {
 807                 netdev = iterator->elem;
 808                 if (netdev->type != LXC_NET_VETH)
 809                         continue;
 810                 if (!(netdev->name = malloc(IFNAMSIZ))) {
 811                         ERROR("Out of memory.");
 812                         close(netpipe);
 813                         return -1;
 814                 }
 815                 if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) {
 816                         close(netpipe);
 817                         return -1;
 818                 }
 819         }
 820         close(netpipe);
 821         return 0;
 822 }
 823
 824 static int do_start(void *data)
 825 {
 826         struct lxc_list *iterator;
 827         struct lxc_handler *handler = data;
 828         int devnull_fd = -1, ret;
 829         char path[PATH_MAX];
 830
 831         if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) {
 832                 SYSERROR("Failed to set signal mask.");
 833                 return -1;
 834         }
 835
 836         /* This prctl must be before the synchro, so if the parent dies before
 837          * we set the parent death signal, we will detect its death with the
 838          * synchro right after, otherwise we have a window where the parent can
 839          * exit before we set the pdeath signal leading to a unsupervized
 840          * container.
 841          */
 842         if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) {
 843                 SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL.");
 844                 return -1;
 845         }
 846
 847         lxc_sync_fini_parent(handler);
 848
 849         /* Don't leak the pinfd to the container. */
 850         if (handler->pinfd >= 0) {
 851                 close(handler->pinfd);
 852         }
 853
 854         if (lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP))
 855                 return -1;
 856
 857         /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See
 858          * https://github.com/lxc/lxd/issues/1978.
 859          */
 860         if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) ==
 861             (CLONE_NEWNET | CLONE_NEWUSER)) {
 862                 ret = unshare(CLONE_NEWNET);
 863                 if (ret < 0) {
 864                         SYSERROR("Failed to unshare CLONE_NEWNET.");
 865                         goto out_warn_father;
 866                 }
 867                 INFO("Unshared CLONE_NEWNET.");
 868         }
 869
 870         /* Tell the parent task it can begin to configure the container and wait
 871          * for it to finish.
 872          */
 873         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE))
 874                 return -1;
 875
 876         if (read_unpriv_netifindex(&handler->conf->network) < 0)
 877                 goto out_warn_father;
 878
 879         /* If we are in a new user namespace, become root there to have
 880          * privilege over our namespace.
 881          */
 882         if (!lxc_list_empty(&handler->conf->id_map)) {
 883                 if (lxc_switch_uid_gid(0, 0) < 0)
 884                         goto out_warn_father;
 885
 886                 /* Drop groups only after we switched to a valid gid in the new
 887                  * user namespace.
 888                  */
 889                 if (lxc_setgroups(0, NULL) < 0)
 890                         goto out_warn_father;
 891         }
 892
 893         if (access(handler->lxcpath, X_OK)) {
 894                 print_top_failing_dir(handler->lxcpath);
 895                 goto out_warn_father;
 896         }
 897
 898         #if HAVE_LIBCAP
 899         if (handler->conf->need_utmp_watch) {
 900                 if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) {
 901                         SYSERROR("Failed to remove the CAP_SYS_BOOT capability.");
 902                         goto out_warn_father;
 903                 }
 904                 DEBUG("Dropped the CAP_SYS_BOOT capability.");
 905         }
 906         #endif
 907
 908         ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount);
 909         if (ret < 0 || ret >= sizeof(path))
 910                 goto out_warn_father;
 911
 912         /* In order to checkpoint restore, we need to have everything in the
 913          * same mount namespace. However, some containers may not have a
 914          * reasonable /dev (in particular, they may not have /dev/null), so we
 915          * can't set init's std fds to /dev/null by opening it from inside the
 916          * container.
 917          *
 918          * If that's the case, fall back to using the host's /dev/null. This
 919          * means that migration won't work, but at least we won't spew output
 920          * where it isn't wanted.
 921          */
 922         if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) {
 923                 devnull_fd = open_devnull();
 924
 925                 if (devnull_fd < 0)
 926                         goto out_warn_father;
 927                 WARN("Using /dev/null from the host for container init's "
 928                      "standard file descriptors. Migration will not work.");
 929         }
 930
 931         /* Setup the container, ip, names, utsname, ... */
 932         if (lxc_setup(handler)) {
 933                 ERROR("Failed to setup container \"%s\".", handler->name);
 934                 goto out_warn_father;
 935         }
 936
 937         /* Ask father to setup cgroups and wait for him to finish. */
 938         if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP))
 939                 goto out_error;
 940
 941         /* Unshare cgroup namespace after we have setup our cgroups. If we do it
 942          * earlier we end up with a wrong view of /proc/self/cgroup. For
 943          * example, assume we unshare(CLONE_NEWCGROUP) first, and then create
 944          * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then
 945          * /proc/self/cgroup would show us:
 946          *
 947          *      8:cpuset:/lxc/c
 948          *
 949          * whereas it should actually show
 950          *
 951          *      8:cpuset:/
 952          */
 953         if (cgns_supported()) {
 954                 if (unshare(CLONE_NEWCGROUP) < 0) {
 955                         INFO("Failed to unshare CLONE_NEWCGROUP.");
 956                         goto out_warn_father;
 957                 }
 958                 INFO("Unshared CLONE_NEWCGROUP.");
 959         }
 960
 961         /* Set the label to change to when we exec(2) the container's init. */
 962         if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0)
 963                 goto out_warn_father;
 964
 965         /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it
 966          * before we aren't allowed anymore.
 967          */
 968         if (handler->conf->no_new_privs) {
 969                 if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) {
 970                         SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
 971                         goto out_warn_father;
 972                 }
 973                 DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges.");
 974         }
 975
 976         /* Some init's such as busybox will set sane tty settings on stdin,
 977          * stdout, stderr which it thinks is the console. We already set them
 978          * the way we wanted on the real terminal, and we want init to do its
 979          * setup on its console ie. the pty allocated in lxc_console_create() so
 980          * make sure that that pty is stdin,stdout,stderr.
 981          */
 982         if (lxc_console_set_stdfds(handler->conf->console.slave) < 0)
 983                 goto out_warn_father;
 984
 985         /* If we mounted a temporary proc, then unmount it now. */
 986         tmp_proc_unmount(handler->conf);
 987
 988         if (lxc_seccomp_load(handler->conf) != 0)
 989                 goto out_warn_father;
 990
 991         if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) {
 992                 ERROR("Failed to run lxc.hook.start for container \"%s\".", handler->name);
 993                 goto out_warn_father;
 994         }
 995
 996         /* The container has been setup. We can now switch to an unprivileged
 997          * uid/gid.
 998          */
 999         if (handler->conf->is_execute) {
1000                 bool have_cap_setgid;
1001                 uid_t new_uid = handler->conf->init_uid;
1002                 gid_t new_gid = handler->conf->init_gid;
1003
1004                 /* If we are in a new user namespace we already dropped all
1005                  * groups when we switched to root in the new user namespace
1006                  * further above. Only drop groups if we can, so ensure that we
1007                  * have necessary privilege.
1008                  */
1009                 #if HAVE_LIBCAP
1010                 have_cap_setgid = lxc_proc_cap_is_set(CAP_SETGID, CAP_EFFECTIVE);
1011                 #else
1012                 have_cap_setgid = false;
1013                 #endif
1014                 if (lxc_list_empty(&handler->conf->id_map) && have_cap_setgid) {
1015                         if (lxc_setgroups(0, NULL) < 0)
1016                                 goto out_warn_father;
1017                 }
1018
1019                 if (lxc_switch_uid_gid(new_uid, new_gid) < 0)
1020                         goto out_warn_father;
1021         }
1022
1023         /* The clearenv() and putenv() calls have been moved here to allow us to
1024          * use environment variables passed to the various hooks, such as the
1025          * start hook above. Not all of the variables like CONFIG_PATH or ROOTFS
1026          * are valid in this context but others are.
1027          */
1028         if (clearenv()) {
1029                 SYSERROR("Failed to clear environment.");
1030                 /* Don't error out though. */
1031         }
1032
1033         lxc_list_for_each(iterator, &handler->conf->environment) {
1034                 if (putenv((char *)iterator->elem)) {
1035                         SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem);
1036                         goto out_warn_father;
1037                 }
1038         }
1039
1040         if (putenv("container=lxc")) {
1041                 SYSERROR("Failed to set environment variable: container=lxc.");
1042                 goto out_warn_father;
1043         }
1044
1045         if (handler->conf->pty_names) {
1046                 if (putenv(handler->conf->pty_names)) {
1047                         SYSERROR("Failed to set environment variable for container ptys.");
1048                         goto out_warn_father;
1049                 }
1050         }
1051
1052         close(handler->sigfd);
1053
1054         if (devnull_fd < 0) {
1055                 devnull_fd = open_devnull();
1056
1057                 if (devnull_fd < 0)
1058                         goto out_warn_father;
1059         }
1060
1061         if (handler->backgrounded && set_stdfds(devnull_fd))
1062                 goto out_warn_father;
1063
1064         if (devnull_fd >= 0) {
1065                 close(devnull_fd);
1066                 devnull_fd = -1;
1067         }
1068
1069         setsid();
1070
1071         /* After this call, we are in error because this ops should not return
1072          * as it execs.
1073          */
1074         handler->ops->start(handler, handler->data);
1075
1076 out_warn_father:
1077         /* We want the parent to know something went wrong, so we return a
1078          * special error code.
1079          */
1080         lxc_sync_wake_parent(handler, LXC_SYNC_ERROR);
1081
1082 out_error:
1083         if (devnull_fd >= 0)
1084                 close(devnull_fd);
1085
1086         return -1;
1087 }
1088
1089 static int save_phys_nics(struct lxc_conf *conf)
1090 {
1091         struct lxc_list *iterator;
1092         int am_root = (getuid() == 0);
1093
1094         if (!am_root)
1095                 return 0;
1096
1097         lxc_list_for_each(iterator, &conf->network) {
1098                 struct lxc_netdev *netdev = iterator->elem;
1099
1100                 if (netdev->type != LXC_NET_PHYS)
1101                         continue;
1102                 conf->saved_nics = realloc(conf->saved_nics,
1103                                 (conf->num_savednics+1)*sizeof(struct saved_nic));
1104                 if (!conf->saved_nics)
1105                         return -1;
1106                 conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex;
1107                 conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link);
1108                 if (!conf->saved_nics[conf->num_savednics].orig_name)
1109                         return -1;
1110                 INFO("Stored saved_nic #%d idx %d name %s.", conf->num_savednics,
1111                         conf->saved_nics[conf->num_savednics].ifindex,
1112                         conf->saved_nics[conf->num_savednics].orig_name);
1113                 conf->num_savednics++;
1114         }
1115
1116         return 0;
1117 }
1118
1119 static int lxc_recv_ttys_from_child(struct lxc_handler *handler)
1120 {
1121         int i;
1122         int *ttyfds;
1123         struct lxc_pty_info *pty_info;
1124         int ret = -1;
1125         int sock = handler->ttysock[1];
1126         struct lxc_conf *conf = handler->conf;
1127         struct lxc_tty_info *tty_info = &conf->tty_info;
1128         size_t num_ttyfds = (2 * conf->tty);
1129
1130         if (!conf->tty)
1131                 return 0;
1132
1133         tty_info->pty_info = malloc(sizeof(*tty_info->pty_info) * conf->tty);
1134         if (!tty_info->pty_info)
1135                 return -1;
1136
1137         ttyfds = malloc(num_ttyfds * sizeof(int));
1138         if (!ttyfds)
1139                 return -1;
1140
1141         ret = lxc_abstract_unix_recv_fds(sock, ttyfds, num_ttyfds, NULL, 0);
1142         for (i = 0; (ret >= 0 && *ttyfds != -1) && (i < num_ttyfds); i++) {
1143                 pty_info = &tty_info->pty_info[i / 2];
1144                 pty_info->busy = 0;
1145                 pty_info->slave = ttyfds[i++];
1146                 pty_info->master = ttyfds[i];
1147                 TRACE("received pty with master fd %d and slave fd %d from "
1148                       "parent", pty_info->master, pty_info->slave);
1149         }
1150
1151         tty_info->nbtty = conf->tty;
1152
1153         free(ttyfds);
1154
1155         if (ret < 0)
1156                 ERROR("failed to receive %d ttys from child: %s", conf->tty,
1157                       strerror(errno));
1158         else
1159                 TRACE("received %d ttys from child", conf->tty);
1160
1161         return ret;
1162 }
1163
1164 void resolve_clone_flags(struct lxc_handler *handler)
1165 {
1166         handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS;
1167
1168         if (!lxc_list_empty(&handler->conf->id_map))
1169                 handler->clone_flags |= CLONE_NEWUSER;
1170
1171         if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) {
1172                 if (!lxc_requests_empty_network(handler))
1173                         handler->clone_flags |= CLONE_NEWNET;
1174         } else {
1175                 INFO("Inheriting a NET namespace.");
1176         }
1177
1178         if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1)
1179                 handler->clone_flags |= CLONE_NEWIPC;
1180         else
1181                 INFO("Inheriting an IPC namespace.");
1182
1183         if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1)
1184                 handler->clone_flags |= CLONE_NEWUTS;
1185         else
1186                 INFO("Inheriting a UTS namespace.");
1187 }
1188
1189 /* lxc_spawn() performs crucial setup tasks and clone()s the new process which
1190  * exec()s the requested container binary.
1191  * Note that lxc_spawn() runs in the parent namespaces. Any operations performed
1192  * right here should be double checked if they'd pose a security risk. (For
1193  * example, any {u}mount() operations performed here will be reflected on the
1194  * host!)
1195  */
1196 static int lxc_spawn(struct lxc_handler *handler)
1197 {
1198         int failed_before_rename = 0;
1199         const char *name = handler->name;
1200         bool cgroups_connected = false;
1201         int saved_ns_fd[LXC_NS_MAX];
1202         int preserve_mask = 0, i, flags;
1203         int netpipepair[2], nveths;
1204         bool wants_to_map_ids;
1205         struct lxc_list *id_map;
1206
1207         netpipe = -1;
1208         id_map = &handler->conf->id_map;
1209         wants_to_map_ids = !lxc_list_empty(id_map);
1210
1211         for (i = 0; i < LXC_NS_MAX; i++)
1212                 if (handler->conf->inherit_ns_fd[i] != -1)
1213                         preserve_mask |= ns_info[i].clone_flag;
1214
1215         if (lxc_sync_init(handler))
1216                 return -1;
1217
1218         if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) {
1219                 lxc_sync_fini(handler);
1220                 return -1;
1221         }
1222
1223         resolve_clone_flags(handler);
1224
1225         if (handler->clone_flags & CLONE_NEWNET) {
1226                 if (!lxc_list_empty(&handler->conf->network)) {
1227
1228                         /* Find gateway addresses from the link device, which is
1229                          * no longer accessible inside the container. Do this
1230                          * before creating network interfaces, since goto
1231                          * out_delete_net does not work before lxc_clone.
1232                          */
1233                         if (lxc_find_gateway_addresses(handler)) {
1234                                 ERROR("Failed to find gateway addresses.");
1235                                 lxc_sync_fini(handler);
1236                                 return -1;
1237                         }
1238
1239                         /* That should be done before the clone because we will
1240                          * fill the netdev index and use them in the child.
1241                          */
1242                         if (lxc_setup_networks_in_parent_namespaces(handler)) {
1243                                 ERROR("Failed to create the network.");
1244                                 lxc_sync_fini(handler);
1245                                 return -1;
1246                         }
1247                 }
1248
1249                 if (save_phys_nics(handler->conf)) {
1250                         ERROR("Failed to save physical nic info.");
1251                         goto out_abort;
1252                 }
1253         }
1254
1255         if (!cgroup_init(handler)) {
1256                 ERROR("Failed initializing cgroup support.");
1257                 goto out_delete_net;
1258         }
1259
1260         cgroups_connected = true;
1261
1262         if (!cgroup_create(handler)) {
1263                 ERROR("Failed creating cgroups.");
1264                 goto out_delete_net;
1265         }
1266
1267         /* If the rootfs is not a blockdev, prevent the container from marking
1268          * it readonly.
1269          * If the container is unprivileged then skip rootfs pinning.
1270          */
1271         if (!wants_to_map_ids) {
1272                 handler->pinfd = pin_rootfs(handler->conf->rootfs.path);
1273                 if (handler->pinfd == -1)
1274                         INFO("Failed to pin the rootfs for container \"%s\".", handler->name);
1275         }
1276
1277         if (!preserve_ns(saved_ns_fd, preserve_mask, getpid()))
1278                 goto out_delete_net;
1279
1280         if (attach_ns(handler->conf->inherit_ns_fd) < 0)
1281                 goto out_delete_net;
1282
1283         if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) {
1284                 if (pipe(netpipepair) < 0) {
1285                         SYSERROR("Failed to create pipe.");
1286                         goto out_delete_net;
1287                 }
1288                 /* Store netpipe in the global var for do_start's use. */
1289                 netpipe = netpipepair[0];
1290         }
1291
1292         /* Create a process in a new set of namespaces. */
1293         flags = handler->clone_flags;
1294         if (handler->clone_flags & CLONE_NEWUSER) {
1295                 /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to
1296                  * clone a new user namespace first and only later unshare our
1297                  * network namespace to ensure that network devices ownership is
1298                  * set up correctly.
1299                  */
1300                 flags &= ~CLONE_NEWNET;
1301         }
1302         handler->pid = lxc_clone(do_start, handler, flags);
1303         if (handler->pid < 0) {
1304                 SYSERROR("Failed to clone a new set of namespaces.");
1305                 goto out_delete_net;
1306         }
1307         for (i = 0; i < LXC_NS_MAX; i++)
1308                 if (flags & ns_info[i].clone_flag)
1309                         INFO("Cloned %s.", ns_info[i].flag_name);
1310
1311         if (!preserve_ns(handler->nsfd, handler->clone_flags | preserve_mask, handler->pid))
1312                 INFO("Failed to preserve namespace for lxc.hook.stop.");
1313
1314         if (attach_ns(saved_ns_fd))
1315                 WARN("Failed to restore saved namespaces.");
1316
1317         lxc_sync_fini_child(handler);
1318
1319         /* Map the container uids. The container became an invalid userid the
1320          * moment it was cloned with CLONE_NEWUSER. This call doesn't change
1321          * anything immediately, but allows the container to setuid(0) (0 being
1322          * mapped to something else on the host.) later to become a valid uid
1323          * again.
1324          */
1325         if (wants_to_map_ids && lxc_map_ids(id_map, handler->pid)) {
1326                 ERROR("Failed to set up id mapping.");
1327                 goto out_delete_net;
1328         }
1329
1330         if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP)) {
1331                 failed_before_rename = 1;
1332                 goto out_delete_net;
1333         }
1334
1335         if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) {
1336                 failed_before_rename = 1;
1337                 goto out_delete_net;
1338         }
1339
1340         if (!cgroup_create_legacy(handler)) {
1341                 ERROR("Failed to setup legacy cgroups for container \"%s\".", name);
1342                 goto out_delete_net;
1343         }
1344         if (!cgroup_setup_limits(handler, false)) {
1345                 ERROR("Failed to setup cgroup limits for container \"%s\".", name);
1346                 goto out_delete_net;
1347         }
1348
1349         if (!cgroup_enter(handler))
1350                 goto out_delete_net;
1351
1352         if (!cgroup_chown(handler))
1353                 goto out_delete_net;
1354
1355         if (failed_before_rename)
1356                 goto out_delete_net;
1357
1358         /* Create the network configuration. */
1359         if (handler->clone_flags & CLONE_NEWNET) {
1360                 if (lxc_assign_network(handler->lxcpath, handler->name,
1361                                        &handler->conf->network, handler->pid)) {
1362                         ERROR("Failed to create the configured network.");
1363                         goto out_delete_net;
1364                 }
1365         }
1366
1367         if (netpipe != -1) {
1368                 struct lxc_list *iterator;
1369                 struct lxc_netdev *netdev;
1370
1371                 close(netpipe);
1372                 lxc_list_for_each(iterator, &handler->conf->network) {
1373                         netdev = iterator->elem;
1374                         if (netdev->type != LXC_NET_VETH)
1375                                 continue;
1376                         if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) {
1377                                 ERROR("Error writing veth name to container.");
1378                                 goto out_delete_net;
1379                         }
1380                 }
1381                 close(netpipepair[1]);
1382         }
1383
1384         /* Tell the child to continue its initialization. We'll get
1385          * LXC_SYNC_CGROUP when it is ready for us to setup cgroups.
1386          */
1387         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE))
1388                 goto out_delete_net;
1389
1390         if (!lxc_list_empty(&handler->conf->limits) && setup_resource_limits(&handler->conf->limits, handler->pid)) {
1391                 ERROR("failed to setup resource limits for '%s'", name);
1392                 goto out_delete_net;
1393         }
1394
1395         if (!cgroup_setup_limits(handler, true)) {
1396                 ERROR("Failed to setup the devices cgroup for container \"%s\".", name);
1397                 goto out_delete_net;
1398         }
1399
1400         cgroup_disconnect();
1401         cgroups_connected = false;
1402
1403         /* Read tty fds allocated by child. */
1404         if (lxc_recv_ttys_from_child(handler) < 0) {
1405                 ERROR("Failed to receive tty info from child process.");
1406                 goto out_delete_net;
1407         }
1408
1409         /* Tell the child to complete its initialization and wait for it to exec
1410          * or return an error. (The child will never return
1411          * LXC_SYNC_POST_CGROUP+1. It will either close the sync pipe, causing
1412          * lxc_sync_barrier_child to return success, or return a different
1413          * value, causing us to error out).
1414          */
1415         if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP))
1416                 return -1;
1417
1418         if (handler->ops->post_start(handler, handler->data))
1419                 goto out_abort;
1420
1421         if (lxc_set_state(name, handler, RUNNING)) {
1422                 ERROR("Failed to set state for container \"%s\" to \"%s\".", name,
1423                       lxc_state2str(RUNNING));
1424                 goto out_abort;
1425         }
1426
1427         lxc_sync_fini(handler);
1428         handler->netnsfd = lxc_preserve_ns(handler->pid, "net");
1429
1430         return 0;
1431
1432 out_delete_net:
1433         if (cgroups_connected)
1434                 cgroup_disconnect();
1435         if (handler->clone_flags & CLONE_NEWNET)
1436                 lxc_delete_network(handler);
1437 out_abort:
1438         lxc_abort(name, handler);
1439         lxc_sync_fini(handler);
1440         if (handler->pinfd >= 0) {
1441                 close(handler->pinfd);
1442                 handler->pinfd = -1;
1443         }
1444
1445         return -1;
1446 }
1447
1448 int __lxc_start(const char *name, struct lxc_handler *handler,
1449                 struct lxc_operations* ops, void *data, const char *lxcpath,
1450                 bool backgrounded)
1451 {
1452         int status;
1453         int err = -1;
1454         bool removed_all_netdevs = true;
1455         struct lxc_conf *conf = handler->conf;
1456
1457         if (lxc_init(name, handler) < 0) {
1458                 ERROR("Failed to initialize container \"%s\".", name);
1459                 return -1;
1460         }
1461         handler->ops = ops;
1462         handler->data = data;
1463         handler->backgrounded = backgrounded;
1464         handler->netnsfd = -1;
1465
1466         if (must_drop_cap_sys_boot(handler->conf)) {
1467                 #if HAVE_LIBCAP
1468                 DEBUG("Dropping CAP_SYS_BOOT capability.");
1469                 #else
1470                 DEBUG("Not dropping CAP_SYS_BOOT capability as capabilities aren't supported.");
1471                 #endif
1472         } else {
1473                 DEBUG("Not dropping CAP_SYS_BOOT or watching utmp.");
1474                 handler->conf->need_utmp_watch = 0;
1475         }
1476
1477         if (!attach_block_device(handler->conf)) {
1478                 ERROR("Failed to attach block device.");
1479                 goto out_fini_nonet;
1480         }
1481
1482         if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) {
1483                 /* If the backing store is a device, mount it here and now. */
1484                 if (rootfs_is_blockdev(conf)) {
1485                         if (unshare(CLONE_NEWNS) < 0) {
1486                                 ERROR("Failed to unshare CLONE_NEWNS.");
1487                                 goto out_fini_nonet;
1488                         }
1489                         INFO("Unshared CLONE_NEWNS.");
1490
1491                         remount_all_slave();
1492                         if (do_rootfs_setup(conf, name, lxcpath) < 0) {
1493                                 ERROR("Error setting up rootfs mount as root before spawn.");
1494                                 goto out_fini_nonet;
1495                         }
1496                         INFO("Set up container rootfs as host root.");
1497                 }
1498         }
1499
1500         err = lxc_spawn(handler);
1501         if (err) {
1502                 ERROR("Failed to spawn container \"%s\".", name);
1503                 goto out_detach_blockdev;
1504         }
1505
1506         handler->conf->reboot = 0;
1507
1508         err = lxc_poll(name, handler);
1509         if (err) {
1510                 ERROR("LXC mainloop exited with error: %d.", err);
1511                 if (handler->netnsfd >= 0) {
1512                         close(handler->netnsfd);
1513                         handler->netnsfd = -1;
1514                 }
1515                 goto out_abort;
1516         }
1517
1518         while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR)
1519                 continue;
1520
1521         /* If the child process exited but was not signaled, it didn't call
1522          * reboot. This should mean it was an lxc-execute which simply exited.
1523          * In any case, treat it as a 'halt'.
1524          */
1525         if (WIFSIGNALED(status)) {
1526                 switch(WTERMSIG(status)) {
1527                 case SIGINT: /* halt */
1528                         DEBUG("Container \"%s\" is halting.", name);
1529                         break;
1530                 case SIGHUP: /* reboot */
1531                         DEBUG("Container \"%s\" is rebooting.", name);
1532                         handler->conf->reboot = 1;
1533                         break;
1534                 case SIGSYS: /* seccomp */
1535                         DEBUG("Container \"%s\" violated its seccomp policy.", name);
1536                         break;
1537                 default:
1538                         DEBUG("Unknown exit status for container \"%s\" init %d.", name, WTERMSIG(status));
1539                         break;
1540                 }
1541         }
1542
1543         DEBUG("Pushing physical nics back to host namespace");
1544         lxc_restore_phys_nics_to_netns(handler->netnsfd, handler->conf);
1545
1546         DEBUG("Tearing down virtual network devices used by container \"%s\".", name);
1547         removed_all_netdevs = lxc_delete_network(handler);
1548
1549         if (handler->pinfd >= 0) {
1550                 close(handler->pinfd);
1551                 handler->pinfd = -1;
1552         }
1553
1554         lxc_monitor_send_exit_code(name, status, handler->lxcpath);
1555         err =  lxc_error_set_and_log(handler->pid, status);
1556 out_fini:
1557         if (!removed_all_netdevs) {
1558                 DEBUG("Failed tearing down network devices used by container. Trying again!");
1559                 removed_all_netdevs = lxc_delete_network(handler);
1560                 if (!removed_all_netdevs)
1561                         DEBUG("Failed tearing down network devices used by container. Not trying again!");
1562         }
1563
1564 out_detach_blockdev:
1565         detach_block_device(handler->conf);
1566
1567 out_fini_nonet:
1568         lxc_fini(name, handler);
1569         return err;
1570
1571 out_abort:
1572         lxc_abort(name, handler);
1573         goto out_fini;
1574 }
1575
1576 struct start_args {
1577         char *const *argv;
1578 };
1579
1580 static int start(struct lxc_handler *handler, void* data)
1581 {
1582         struct start_args *arg = data;
1583
1584         NOTICE("Exec'ing \"%s\".", arg->argv[0]);
1585
1586         execvp(arg->argv[0], arg->argv);
1587         SYSERROR("Failed to exec \"%s\".", arg->argv[0]);
1588         return 0;
1589 }
1590
1591 static int post_start(struct lxc_handler *handler, void* data)
1592 {
1593         struct start_args *arg = data;
1594
1595         NOTICE("Started \"%s\" with pid \"%d\".", arg->argv[0], handler->pid);
1596         return 0;
1597 }
1598
1599 static struct lxc_operations start_ops = {
1600         .start = start,
1601         .post_start = post_start
1602 };
1603
1604 int lxc_start(const char *name, char *const argv[], struct lxc_handler *handler,
1605               const char *lxcpath, bool backgrounded)
1606 {
1607         struct start_args start_arg = {
1608                 .argv = argv,
1609         };
1610
1611         handler->conf->need_utmp_watch = 1;
1612         return __lxc_start(name, handler, &start_ops, &start_arg, lxcpath, backgrounded);
1613 }
1614
1615 static void lxc_destroy_container_on_signal(struct lxc_handler *handler,
1616                                             const char *name)
1617 {
1618         char destroy[MAXPATHLEN];
1619         bool bret = true;
1620         int ret = 0;
1621         struct lxc_container *c;
1622         if (handler->conf->rootfs.path && handler->conf->rootfs.mount) {
1623                 bret = do_destroy_container(handler->conf);
1624                 if (!bret) {
1625                         ERROR("Error destroying rootfs for container \"%s\".", name);
1626                         return;
1627                 }
1628         }
1629         INFO("Destroyed rootfs for container \"%s\".", name);
1630
1631         ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name);
1632         if (ret < 0 || ret >= MAXPATHLEN) {
1633                 ERROR("Error destroying directory for container \"%s\".", name);
1634                 return;
1635         }
1636
1637         c = lxc_container_new(name, handler->lxcpath);
1638         if (c) {
1639                 if (container_disk_lock(c)) {
1640                         INFO("Could not update lxc_snapshots file.");
1641                         lxc_container_put(c);
1642                 } else {
1643                         mod_all_rdeps(c, false);
1644                         container_disk_unlock(c);
1645                         lxc_container_put(c);
1646                 }
1647         }
1648
1649         if (am_unpriv())
1650                 ret = userns_exec_1(handler->conf, lxc_rmdir_onedev_wrapper,
1651                                     destroy, "lxc_rmdir_onedev_wrapper");
1652         else
1653                 ret = lxc_rmdir_onedev(destroy, NULL);
1654
1655         if (ret < 0) {
1656                 ERROR("Error destroying directory for container \"%s\".", name);
1657                 return;
1658         }
1659         INFO("Destroyed directory for container \"%s\".", name);
1660 }
1661
1662 static int lxc_rmdir_onedev_wrapper(void *data)
1663 {
1664         char *arg = (char *) data;
1665         return lxc_rmdir_onedev(arg, NULL);
1666 }
1667
1668 static bool do_destroy_container(struct lxc_conf *conf) {
1669         if (am_unpriv()) {
1670                 if (userns_exec_1(conf, bdev_destroy_wrapper, conf,
1671                                   "bdev_destroy_wrapper") < 0)
1672                         return false;
1673                 return true;
1674         }
1675         return bdev_destroy(conf);
1676 }